LCOV - code coverage report
Current view: top level - lib/bdev - bdev.c (source / functions) Hit Total Coverage
Test: ut_cov_unit.info Lines: 4018 5707 70.4 %
Date: 2024-11-15 10:22:08 Functions: 362 471 76.9 %

          Line data    Source code
       1             : /*   SPDX-License-Identifier: BSD-3-Clause
       2             :  *   Copyright (C) 2016 Intel Corporation. All rights reserved.
       3             :  *   Copyright (c) 2019 Mellanox Technologies LTD. All rights reserved.
       4             :  *   Copyright (c) 2021-2023 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
       5             :  */
       6             : 
       7             : #include "spdk/stdinc.h"
       8             : 
       9             : #include "spdk/bdev.h"
      10             : 
      11             : #include "spdk/accel.h"
      12             : #include "spdk/config.h"
      13             : #include "spdk/env.h"
      14             : #include "spdk/thread.h"
      15             : #include "spdk/likely.h"
      16             : #include "spdk/queue.h"
      17             : #include "spdk/nvme_spec.h"
      18             : #include "spdk/scsi_spec.h"
      19             : #include "spdk/notify.h"
      20             : #include "spdk/util.h"
      21             : #include "spdk/trace.h"
      22             : #include "spdk/dma.h"
      23             : 
      24             : #include "spdk/bdev_module.h"
      25             : #include "spdk/log.h"
      26             : #include "spdk/string.h"
      27             : 
      28             : #include "bdev_internal.h"
      29             : #include "spdk_internal/trace_defs.h"
      30             : #include "spdk_internal/assert.h"
      31             : 
      32             : #ifdef SPDK_CONFIG_VTUNE
      33             : #include "ittnotify.h"
      34             : #include "ittnotify_types.h"
      35             : int __itt_init_ittlib(const char *, __itt_group_id);
      36             : #endif
      37             : 
      38             : #define SPDK_BDEV_IO_POOL_SIZE                  (64 * 1024 - 1)
      39             : #define SPDK_BDEV_IO_CACHE_SIZE                 256
      40             : #define SPDK_BDEV_AUTO_EXAMINE                  true
      41             : #define BUF_SMALL_CACHE_SIZE                    128
      42             : #define BUF_LARGE_CACHE_SIZE                    16
      43             : #define NOMEM_THRESHOLD_COUNT                   8
      44             : 
      45             : #define SPDK_BDEV_QOS_TIMESLICE_IN_USEC         1000
      46             : #define SPDK_BDEV_QOS_MIN_IO_PER_TIMESLICE      1
      47             : #define SPDK_BDEV_QOS_MIN_BYTE_PER_TIMESLICE    512
      48             : #define SPDK_BDEV_QOS_MIN_IOS_PER_SEC           1000
      49             : #define SPDK_BDEV_QOS_MIN_BYTES_PER_SEC         (1024 * 1024)
      50             : #define SPDK_BDEV_QOS_MAX_MBYTES_PER_SEC        (UINT64_MAX / (1024 * 1024))
      51             : #define SPDK_BDEV_QOS_LIMIT_NOT_DEFINED         UINT64_MAX
      52             : #define SPDK_BDEV_IO_POLL_INTERVAL_IN_MSEC      1000
      53             : 
      54             : /* The maximum number of children requests for a UNMAP or WRITE ZEROES command
      55             :  * when splitting into children requests at a time.
      56             :  */
      57             : #define SPDK_BDEV_MAX_CHILDREN_UNMAP_WRITE_ZEROES_REQS (8)
      58             : #define BDEV_RESET_CHECK_OUTSTANDING_IO_PERIOD 1000000
      59             : 
      60             : /* The maximum number of children requests for a COPY command
      61             :  * when splitting into children requests at a time.
      62             :  */
      63             : #define SPDK_BDEV_MAX_CHILDREN_COPY_REQS (8)
      64             : 
      65             : #define LOG_ALREADY_CLAIMED_ERROR(detail, bdev) \
      66             :         log_already_claimed(SPDK_LOG_ERROR, __LINE__, __func__, detail, bdev)
      67             : #ifdef DEBUG
      68             : #define LOG_ALREADY_CLAIMED_DEBUG(detail, bdev) \
      69             :         log_already_claimed(SPDK_LOG_DEBUG, __LINE__, __func__, detail, bdev)
      70             : #else
      71             : #define LOG_ALREADY_CLAIMED_DEBUG(detail, bdev) do {} while(0)
      72             : #endif
      73             : 
      74             : static void log_already_claimed(enum spdk_log_level level, const int line, const char *func,
      75             :                                 const char *detail, struct spdk_bdev *bdev);
      76             : 
      77             : static const char *qos_rpc_type[] = {"rw_ios_per_sec",
      78             :                                      "rw_mbytes_per_sec", "r_mbytes_per_sec", "w_mbytes_per_sec"
      79             :                                     };
      80             : 
      81             : TAILQ_HEAD(spdk_bdev_list, spdk_bdev);
      82             : 
      83             : RB_HEAD(bdev_name_tree, spdk_bdev_name);
      84             : 
      85             : static int
      86         546 : bdev_name_cmp(struct spdk_bdev_name *name1, struct spdk_bdev_name *name2)
      87             : {
      88         546 :         return strcmp(name1->name, name2->name);
      89             : }
      90             : 
      91        2060 : RB_GENERATE_STATIC(bdev_name_tree, spdk_bdev_name, node, bdev_name_cmp);
      92             : 
      93             : struct spdk_bdev_mgr {
      94             :         struct spdk_mempool *bdev_io_pool;
      95             : 
      96             :         void *zero_buffer;
      97             : 
      98             :         TAILQ_HEAD(bdev_module_list, spdk_bdev_module) bdev_modules;
      99             : 
     100             :         struct spdk_bdev_list bdevs;
     101             :         struct bdev_name_tree bdev_names;
     102             : 
     103             :         bool init_complete;
     104             :         bool module_init_complete;
     105             : 
     106             :         struct spdk_spinlock spinlock;
     107             : 
     108             :         TAILQ_HEAD(, spdk_bdev_open_async_ctx) async_bdev_opens;
     109             : 
     110             : #ifdef SPDK_CONFIG_VTUNE
     111             :         __itt_domain    *domain;
     112             : #endif
     113             : };
     114             : 
     115             : static struct spdk_bdev_mgr g_bdev_mgr = {
     116             :         .bdev_modules = TAILQ_HEAD_INITIALIZER(g_bdev_mgr.bdev_modules),
     117             :         .bdevs = TAILQ_HEAD_INITIALIZER(g_bdev_mgr.bdevs),
     118             :         .bdev_names = RB_INITIALIZER(g_bdev_mgr.bdev_names),
     119             :         .init_complete = false,
     120             :         .module_init_complete = false,
     121             :         .async_bdev_opens = TAILQ_HEAD_INITIALIZER(g_bdev_mgr.async_bdev_opens),
     122             : };
     123             : 
     124             : static void
     125             : __attribute__((constructor))
     126           3 : _bdev_init(void)
     127             : {
     128           3 :         spdk_spin_init(&g_bdev_mgr.spinlock);
     129           3 : }
     130             : 
     131             : typedef void (*lock_range_cb)(struct lba_range *range, void *ctx, int status);
     132             : 
     133             : typedef void (*bdev_copy_bounce_buffer_cpl)(void *ctx, int rc);
     134             : 
     135             : struct lba_range {
     136             :         struct spdk_bdev                *bdev;
     137             :         uint64_t                        offset;
     138             :         uint64_t                        length;
     139             :         bool                            quiesce;
     140             :         void                            *locked_ctx;
     141             :         struct spdk_thread              *owner_thread;
     142             :         struct spdk_bdev_channel        *owner_ch;
     143             :         TAILQ_ENTRY(lba_range)          tailq;
     144             :         TAILQ_ENTRY(lba_range)          tailq_module;
     145             : };
     146             : 
     147             : static struct spdk_bdev_opts    g_bdev_opts = {
     148             :         .bdev_io_pool_size = SPDK_BDEV_IO_POOL_SIZE,
     149             :         .bdev_io_cache_size = SPDK_BDEV_IO_CACHE_SIZE,
     150             :         .bdev_auto_examine = SPDK_BDEV_AUTO_EXAMINE,
     151             :         .iobuf_small_cache_size = BUF_SMALL_CACHE_SIZE,
     152             :         .iobuf_large_cache_size = BUF_LARGE_CACHE_SIZE,
     153             : };
     154             : 
     155             : static spdk_bdev_init_cb        g_init_cb_fn = NULL;
     156             : static void                     *g_init_cb_arg = NULL;
     157             : 
     158             : static spdk_bdev_fini_cb        g_fini_cb_fn = NULL;
     159             : static void                     *g_fini_cb_arg = NULL;
     160             : static struct spdk_thread       *g_fini_thread = NULL;
     161             : 
     162             : struct spdk_bdev_qos_limit {
     163             :         /** IOs or bytes allowed per second (i.e., 1s). */
     164             :         uint64_t limit;
     165             : 
     166             :         /** Remaining IOs or bytes allowed in current timeslice (e.g., 1ms).
     167             :          *  For remaining bytes, allowed to run negative if an I/O is submitted when
     168             :          *  some bytes are remaining, but the I/O is bigger than that amount. The
     169             :          *  excess will be deducted from the next timeslice.
     170             :          */
     171             :         int64_t remaining_this_timeslice;
     172             : 
     173             :         /** Minimum allowed IOs or bytes to be issued in one timeslice (e.g., 1ms). */
     174             :         uint32_t min_per_timeslice;
     175             : 
     176             :         /** Maximum allowed IOs or bytes to be issued in one timeslice (e.g., 1ms). */
     177             :         uint32_t max_per_timeslice;
     178             : 
     179             :         /** Function to check whether to queue the IO.
     180             :          * If The IO is allowed to pass, the quota will be reduced correspondingly.
     181             :          */
     182             :         bool (*queue_io)(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io);
     183             : 
     184             :         /** Function to rewind the quota once the IO was allowed to be sent by this
     185             :          * limit but queued due to one of the further limits.
     186             :          */
     187             :         void (*rewind_quota)(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io);
     188             : };
     189             : 
     190             : struct spdk_bdev_qos {
     191             :         /** Types of structure of rate limits. */
     192             :         struct spdk_bdev_qos_limit rate_limits[SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES];
     193             : 
     194             :         /** The channel that all I/O are funneled through. */
     195             :         struct spdk_bdev_channel *ch;
     196             : 
     197             :         /** The thread on which the poller is running. */
     198             :         struct spdk_thread *thread;
     199             : 
     200             :         /** Size of a timeslice in tsc ticks. */
     201             :         uint64_t timeslice_size;
     202             : 
     203             :         /** Timestamp of start of last timeslice. */
     204             :         uint64_t last_timeslice;
     205             : 
     206             :         /** Poller that processes queued I/O commands each time slice. */
     207             :         struct spdk_poller *poller;
     208             : };
     209             : 
     210             : struct spdk_bdev_mgmt_channel {
     211             :         /*
     212             :          * Each thread keeps a cache of bdev_io - this allows
     213             :          *  bdev threads which are *not* DPDK threads to still
     214             :          *  benefit from a per-thread bdev_io cache.  Without
     215             :          *  this, non-DPDK threads fetching from the mempool
     216             :          *  incur a cmpxchg on get and put.
     217             :          */
     218             :         bdev_io_stailq_t per_thread_cache;
     219             :         uint32_t        per_thread_cache_count;
     220             :         uint32_t        bdev_io_cache_size;
     221             : 
     222             :         struct spdk_iobuf_channel iobuf;
     223             : 
     224             :         TAILQ_HEAD(, spdk_bdev_shared_resource) shared_resources;
     225             :         TAILQ_HEAD(, spdk_bdev_io_wait_entry)   io_wait_queue;
     226             : };
     227             : 
     228             : /*
     229             :  * Per-module (or per-io_device) data. Multiple bdevs built on the same io_device
     230             :  * will queue here their IO that awaits retry. It makes it possible to retry sending
     231             :  * IO to one bdev after IO from other bdev completes.
     232             :  */
     233             : struct spdk_bdev_shared_resource {
     234             :         /* The bdev management channel */
     235             :         struct spdk_bdev_mgmt_channel *mgmt_ch;
     236             : 
     237             :         /*
     238             :          * Count of I/O submitted to bdev module and waiting for completion.
     239             :          * Incremented before submit_request() is called on an spdk_bdev_io.
     240             :          */
     241             :         uint64_t                io_outstanding;
     242             : 
     243             :         /*
     244             :          * Queue of IO awaiting retry because of a previous NOMEM status returned
     245             :          *  on this channel.
     246             :          */
     247             :         bdev_io_tailq_t         nomem_io;
     248             : 
     249             :         /*
     250             :          * Threshold which io_outstanding must drop to before retrying nomem_io.
     251             :          */
     252             :         uint64_t                nomem_threshold;
     253             : 
     254             :         /* I/O channel allocated by a bdev module */
     255             :         struct spdk_io_channel  *shared_ch;
     256             : 
     257             :         struct spdk_poller      *nomem_poller;
     258             : 
     259             :         /* Refcount of bdev channels using this resource */
     260             :         uint32_t                ref;
     261             : 
     262             :         TAILQ_ENTRY(spdk_bdev_shared_resource) link;
     263             : };
     264             : 
     265             : #define BDEV_CH_RESET_IN_PROGRESS       (1 << 0)
     266             : #define BDEV_CH_QOS_ENABLED             (1 << 1)
     267             : 
     268             : struct spdk_bdev_channel {
     269             :         struct spdk_bdev        *bdev;
     270             : 
     271             :         /* The channel for the underlying device */
     272             :         struct spdk_io_channel  *channel;
     273             : 
     274             :         /* Accel channel */
     275             :         struct spdk_io_channel  *accel_channel;
     276             : 
     277             :         /* Per io_device per thread data */
     278             :         struct spdk_bdev_shared_resource *shared_resource;
     279             : 
     280             :         struct spdk_bdev_io_stat *stat;
     281             : 
     282             :         /*
     283             :          * Count of I/O submitted to the underlying dev module through this channel
     284             :          * and waiting for completion.
     285             :          */
     286             :         uint64_t                io_outstanding;
     287             : 
     288             :         /*
     289             :          * List of all submitted I/Os including I/O that are generated via splitting.
     290             :          */
     291             :         bdev_io_tailq_t         io_submitted;
     292             : 
     293             :         /*
     294             :          * List of spdk_bdev_io that are currently queued because they write to a locked
     295             :          * LBA range.
     296             :          */
     297             :         bdev_io_tailq_t         io_locked;
     298             : 
     299             :         /* List of I/Os with accel sequence being currently executed */
     300             :         bdev_io_tailq_t         io_accel_exec;
     301             : 
     302             :         /* List of I/Os doing memory domain pull/push */
     303             :         bdev_io_tailq_t         io_memory_domain;
     304             : 
     305             :         uint32_t                flags;
     306             : 
     307             :         /* Counts number of bdev_io in the io_submitted TAILQ */
     308             :         uint16_t                queue_depth;
     309             : 
     310             :         uint16_t                trace_id;
     311             : 
     312             :         struct spdk_histogram_data *histogram;
     313             : 
     314             : #ifdef SPDK_CONFIG_VTUNE
     315             :         uint64_t                start_tsc;
     316             :         uint64_t                interval_tsc;
     317             :         __itt_string_handle     *handle;
     318             :         struct spdk_bdev_io_stat *prev_stat;
     319             : #endif
     320             : 
     321             :         bdev_io_tailq_t         queued_resets;
     322             : 
     323             :         lba_range_tailq_t       locked_ranges;
     324             : 
     325             :         /** List of I/Os queued by QoS. */
     326             :         bdev_io_tailq_t         qos_queued_io;
     327             : };
     328             : 
     329             : struct media_event_entry {
     330             :         struct spdk_bdev_media_event    event;
     331             :         TAILQ_ENTRY(media_event_entry)  tailq;
     332             : };
     333             : 
     334             : #define MEDIA_EVENT_POOL_SIZE 64
     335             : 
     336             : struct spdk_bdev_desc {
     337             :         struct spdk_bdev                *bdev;
     338             :         struct spdk_thread              *thread;
     339             :         struct {
     340             :                 spdk_bdev_event_cb_t event_fn;
     341             :                 void *ctx;
     342             :         }                               callback;
     343             :         bool                            closed;
     344             :         bool                            write;
     345             :         bool                            memory_domains_supported;
     346             :         bool                            accel_sequence_supported[SPDK_BDEV_NUM_IO_TYPES];
     347             :         struct spdk_spinlock            spinlock;
     348             :         uint32_t                        refs;
     349             :         struct spdk_bdev_open_opts      opts;
     350             :         TAILQ_HEAD(, media_event_entry) pending_media_events;
     351             :         TAILQ_HEAD(, media_event_entry) free_media_events;
     352             :         struct media_event_entry        *media_events_buffer;
     353             :         TAILQ_ENTRY(spdk_bdev_desc)     link;
     354             : 
     355             :         uint64_t                timeout_in_sec;
     356             :         spdk_bdev_io_timeout_cb cb_fn;
     357             :         void                    *cb_arg;
     358             :         struct spdk_poller      *io_timeout_poller;
     359             :         struct spdk_bdev_module_claim   *claim;
     360             : };
     361             : 
     362             : struct spdk_bdev_iostat_ctx {
     363             :         struct spdk_bdev_io_stat *stat;
     364             :         enum spdk_bdev_reset_stat_mode reset_mode;
     365             :         spdk_bdev_get_device_stat_cb cb;
     366             :         void *cb_arg;
     367             : };
     368             : 
     369             : struct set_qos_limit_ctx {
     370             :         void (*cb_fn)(void *cb_arg, int status);
     371             :         void *cb_arg;
     372             :         struct spdk_bdev *bdev;
     373             : };
     374             : 
     375             : struct spdk_bdev_channel_iter {
     376             :         spdk_bdev_for_each_channel_msg fn;
     377             :         spdk_bdev_for_each_channel_done cpl;
     378             :         struct spdk_io_channel_iter *i;
     379             :         void *ctx;
     380             : };
     381             : 
     382             : struct spdk_bdev_io_error_stat {
     383             :         uint32_t error_status[-SPDK_MIN_BDEV_IO_STATUS];
     384             : };
     385             : 
     386             : enum bdev_io_retry_state {
     387             :         BDEV_IO_RETRY_STATE_INVALID,
     388             :         BDEV_IO_RETRY_STATE_PULL,
     389             :         BDEV_IO_RETRY_STATE_PULL_MD,
     390             :         BDEV_IO_RETRY_STATE_SUBMIT,
     391             :         BDEV_IO_RETRY_STATE_PUSH,
     392             :         BDEV_IO_RETRY_STATE_PUSH_MD,
     393             :         BDEV_IO_RETRY_STATE_GET_ACCEL_BUF,
     394             : };
     395             : 
     396             : #define __bdev_to_io_dev(bdev)          (((char *)bdev) + 1)
     397             : #define __bdev_from_io_dev(io_dev)      ((struct spdk_bdev *)(((char *)io_dev) - 1))
     398             : #define __io_ch_to_bdev_ch(io_ch)       ((struct spdk_bdev_channel *)spdk_io_channel_get_ctx(io_ch))
     399             : #define __io_ch_to_bdev_mgmt_ch(io_ch)  ((struct spdk_bdev_mgmt_channel *)spdk_io_channel_get_ctx(io_ch))
     400             : 
     401             : static inline void bdev_io_complete(void *ctx);
     402             : static inline void bdev_io_complete_unsubmitted(struct spdk_bdev_io *bdev_io);
     403             : static void bdev_io_push_bounce_md_buf(struct spdk_bdev_io *bdev_io);
     404             : static void bdev_io_push_bounce_data(struct spdk_bdev_io *bdev_io);
     405             : static void _bdev_io_get_accel_buf(struct spdk_bdev_io *bdev_io);
     406             : 
     407             : static void bdev_write_zero_buffer_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg);
     408             : static int bdev_write_zero_buffer(struct spdk_bdev_io *bdev_io);
     409             : 
     410             : static void bdev_enable_qos_msg(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
     411             :                                 struct spdk_io_channel *ch, void *_ctx);
     412             : static void bdev_enable_qos_done(struct spdk_bdev *bdev, void *_ctx, int status);
     413             : 
     414             : static int bdev_readv_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
     415             :                                      struct iovec *iov, int iovcnt, void *md_buf, uint64_t offset_blocks,
     416             :                                      uint64_t num_blocks,
     417             :                                      struct spdk_memory_domain *domain, void *domain_ctx,
     418             :                                      struct spdk_accel_sequence *seq, uint32_t dif_check_flags,
     419             :                                      spdk_bdev_io_completion_cb cb, void *cb_arg);
     420             : static int bdev_writev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
     421             :                                       struct iovec *iov, int iovcnt, void *md_buf,
     422             :                                       uint64_t offset_blocks, uint64_t num_blocks,
     423             :                                       struct spdk_memory_domain *domain, void *domain_ctx,
     424             :                                       struct spdk_accel_sequence *seq, uint32_t dif_check_flags,
     425             :                                       uint32_t nvme_cdw12_raw, uint32_t nvme_cdw13_raw,
     426             :                                       spdk_bdev_io_completion_cb cb, void *cb_arg);
     427             : 
     428             : static int bdev_lock_lba_range(struct spdk_bdev_desc *desc, struct spdk_io_channel *_ch,
     429             :                                uint64_t offset, uint64_t length,
     430             :                                lock_range_cb cb_fn, void *cb_arg);
     431             : 
     432             : static int bdev_unlock_lba_range(struct spdk_bdev_desc *desc, struct spdk_io_channel *_ch,
     433             :                                  uint64_t offset, uint64_t length,
     434             :                                  lock_range_cb cb_fn, void *cb_arg);
     435             : 
     436             : static bool bdev_abort_queued_io(bdev_io_tailq_t *queue, struct spdk_bdev_io *bio_to_abort);
     437             : static bool bdev_abort_buf_io(struct spdk_bdev_mgmt_channel *ch, struct spdk_bdev_io *bio_to_abort);
     438             : 
     439             : static bool claim_type_is_v2(enum spdk_bdev_claim_type type);
     440             : static void bdev_desc_release_claims(struct spdk_bdev_desc *desc);
     441             : static void claim_reset(struct spdk_bdev *bdev);
     442             : 
     443             : static void bdev_ch_retry_io(struct spdk_bdev_channel *bdev_ch);
     444             : 
     445             : static bool bdev_io_should_split(struct spdk_bdev_io *bdev_io);
     446             : 
     447             : #define bdev_get_ext_io_opt(opts, field, defval) \
     448             :         ((opts) != NULL ? SPDK_GET_FIELD(opts, field, defval) : (defval))
     449             : 
     450             : static inline void
     451         669 : bdev_ch_add_to_io_submitted(struct spdk_bdev_io *bdev_io)
     452             : {
     453         669 :         TAILQ_INSERT_TAIL(&bdev_io->internal.ch->io_submitted, bdev_io, internal.ch_link);
     454         669 :         bdev_io->internal.ch->queue_depth++;
     455         669 : }
     456             : 
     457             : static inline void
     458         669 : bdev_ch_remove_from_io_submitted(struct spdk_bdev_io *bdev_io)
     459             : {
     460         669 :         TAILQ_REMOVE(&bdev_io->internal.ch->io_submitted, bdev_io, internal.ch_link);
     461         669 :         bdev_io->internal.ch->queue_depth--;
     462         669 : }
     463             : 
     464             : void
     465          14 : spdk_bdev_get_opts(struct spdk_bdev_opts *opts, size_t opts_size)
     466             : {
     467          14 :         if (!opts) {
     468           0 :                 SPDK_ERRLOG("opts should not be NULL\n");
     469           0 :                 return;
     470             :         }
     471             : 
     472          14 :         if (!opts_size) {
     473           0 :                 SPDK_ERRLOG("opts_size should not be zero value\n");
     474           0 :                 return;
     475             :         }
     476             : 
     477          14 :         opts->opts_size = opts_size;
     478             : 
     479             : #define SET_FIELD(field) \
     480             :         if (offsetof(struct spdk_bdev_opts, field) + sizeof(opts->field) <= opts_size) { \
     481             :                 opts->field = g_bdev_opts.field; \
     482             :         } \
     483             : 
     484          14 :         SET_FIELD(bdev_io_pool_size);
     485          14 :         SET_FIELD(bdev_io_cache_size);
     486          14 :         SET_FIELD(bdev_auto_examine);
     487          14 :         SET_FIELD(iobuf_small_cache_size);
     488          14 :         SET_FIELD(iobuf_large_cache_size);
     489             : 
     490             :         /* Do not remove this statement, you should always update this statement when you adding a new field,
     491             :          * and do not forget to add the SET_FIELD statement for your added field. */
     492             :         SPDK_STATIC_ASSERT(sizeof(struct spdk_bdev_opts) == 32, "Incorrect size");
     493             : 
     494             : #undef SET_FIELD
     495          14 : }
     496             : 
     497             : int
     498          15 : spdk_bdev_set_opts(struct spdk_bdev_opts *opts)
     499             : {
     500             :         uint32_t min_pool_size;
     501             : 
     502          15 :         if (!opts) {
     503           0 :                 SPDK_ERRLOG("opts cannot be NULL\n");
     504           0 :                 return -1;
     505             :         }
     506             : 
     507          15 :         if (!opts->opts_size) {
     508           1 :                 SPDK_ERRLOG("opts_size inside opts cannot be zero value\n");
     509           1 :                 return -1;
     510             :         }
     511             : 
     512             :         /*
     513             :          * Add 1 to the thread count to account for the extra mgmt_ch that gets created during subsystem
     514             :          *  initialization.  A second mgmt_ch will be created on the same thread when the application starts
     515             :          *  but before the deferred put_io_channel event is executed for the first mgmt_ch.
     516             :          */
     517          14 :         min_pool_size = opts->bdev_io_cache_size * (spdk_thread_get_count() + 1);
     518          14 :         if (opts->bdev_io_pool_size < min_pool_size) {
     519           0 :                 SPDK_ERRLOG("bdev_io_pool_size %" PRIu32 " is not compatible with bdev_io_cache_size %" PRIu32
     520             :                             " and %" PRIu32 " threads\n", opts->bdev_io_pool_size, opts->bdev_io_cache_size,
     521             :                             spdk_thread_get_count());
     522           0 :                 SPDK_ERRLOG("bdev_io_pool_size must be at least %" PRIu32 "\n", min_pool_size);
     523           0 :                 return -1;
     524             :         }
     525             : 
     526             : #define SET_FIELD(field) \
     527             :         if (offsetof(struct spdk_bdev_opts, field) + sizeof(opts->field) <= opts->opts_size) { \
     528             :                 g_bdev_opts.field = opts->field; \
     529             :         } \
     530             : 
     531          14 :         SET_FIELD(bdev_io_pool_size);
     532          14 :         SET_FIELD(bdev_io_cache_size);
     533          14 :         SET_FIELD(bdev_auto_examine);
     534          14 :         SET_FIELD(iobuf_small_cache_size);
     535          14 :         SET_FIELD(iobuf_large_cache_size);
     536             : 
     537          14 :         g_bdev_opts.opts_size = opts->opts_size;
     538             : 
     539             : #undef SET_FIELD
     540             : 
     541          14 :         return 0;
     542          15 : }
     543             : 
     544             : static struct spdk_bdev *
     545         146 : bdev_get_by_name(const char *bdev_name)
     546             : {
     547             :         struct spdk_bdev_name find;
     548             :         struct spdk_bdev_name *res;
     549             : 
     550         146 :         find.name = (char *)bdev_name;
     551         146 :         res = RB_FIND(bdev_name_tree, &g_bdev_mgr.bdev_names, &find);
     552         146 :         if (res != NULL) {
     553         139 :                 return res->bdev;
     554             :         }
     555             : 
     556           7 :         return NULL;
     557         146 : }
     558             : 
     559             : struct spdk_bdev *
     560          16 : spdk_bdev_get_by_name(const char *bdev_name)
     561             : {
     562             :         struct spdk_bdev *bdev;
     563             : 
     564          16 :         spdk_spin_lock(&g_bdev_mgr.spinlock);
     565          16 :         bdev = bdev_get_by_name(bdev_name);
     566          16 :         spdk_spin_unlock(&g_bdev_mgr.spinlock);
     567             : 
     568          16 :         return bdev;
     569             : }
     570             : 
     571             : struct bdev_io_status_string {
     572             :         enum spdk_bdev_io_status status;
     573             :         const char *str;
     574             : };
     575             : 
     576             : static const struct bdev_io_status_string bdev_io_status_strings[] = {
     577             :         { SPDK_BDEV_IO_STATUS_AIO_ERROR, "aio_error" },
     578             :         { SPDK_BDEV_IO_STATUS_ABORTED, "aborted" },
     579             :         { SPDK_BDEV_IO_STATUS_FIRST_FUSED_FAILED, "first_fused_failed" },
     580             :         { SPDK_BDEV_IO_STATUS_MISCOMPARE, "miscompare" },
     581             :         { SPDK_BDEV_IO_STATUS_NOMEM, "nomem" },
     582             :         { SPDK_BDEV_IO_STATUS_SCSI_ERROR, "scsi_error" },
     583             :         { SPDK_BDEV_IO_STATUS_NVME_ERROR, "nvme_error" },
     584             :         { SPDK_BDEV_IO_STATUS_FAILED, "failed" },
     585             :         { SPDK_BDEV_IO_STATUS_PENDING, "pending" },
     586             :         { SPDK_BDEV_IO_STATUS_SUCCESS, "success" },
     587             : };
     588             : 
     589             : static const char *
     590           0 : bdev_io_status_get_string(enum spdk_bdev_io_status status)
     591             : {
     592             :         uint32_t i;
     593             : 
     594           0 :         for (i = 0; i < SPDK_COUNTOF(bdev_io_status_strings); i++) {
     595           0 :                 if (bdev_io_status_strings[i].status == status) {
     596           0 :                         return bdev_io_status_strings[i].str;
     597             :                 }
     598           0 :         }
     599             : 
     600           0 :         return "reserved";
     601           0 : }
     602             : 
     603             : struct spdk_bdev_wait_for_examine_ctx {
     604             :         struct spdk_poller              *poller;
     605             :         spdk_bdev_wait_for_examine_cb   cb_fn;
     606             :         void                            *cb_arg;
     607             : };
     608             : 
     609             : static bool bdev_module_all_actions_completed(void);
     610             : 
     611             : static int
     612         193 : bdev_wait_for_examine_cb(void *arg)
     613             : {
     614         193 :         struct spdk_bdev_wait_for_examine_ctx *ctx = arg;
     615             : 
     616         193 :         if (!bdev_module_all_actions_completed()) {
     617           0 :                 return SPDK_POLLER_IDLE;
     618             :         }
     619             : 
     620         193 :         spdk_poller_unregister(&ctx->poller);
     621         193 :         ctx->cb_fn(ctx->cb_arg);
     622         193 :         free(ctx);
     623             : 
     624         193 :         return SPDK_POLLER_BUSY;
     625         193 : }
     626             : 
     627             : int
     628         193 : spdk_bdev_wait_for_examine(spdk_bdev_wait_for_examine_cb cb_fn, void *cb_arg)
     629             : {
     630             :         struct spdk_bdev_wait_for_examine_ctx *ctx;
     631             : 
     632         193 :         ctx = calloc(1, sizeof(*ctx));
     633         193 :         if (ctx == NULL) {
     634           0 :                 return -ENOMEM;
     635             :         }
     636         193 :         ctx->cb_fn = cb_fn;
     637         193 :         ctx->cb_arg = cb_arg;
     638         193 :         ctx->poller = SPDK_POLLER_REGISTER(bdev_wait_for_examine_cb, ctx, 0);
     639             : 
     640         193 :         return 0;
     641         193 : }
     642             : 
     643             : struct spdk_bdev_examine_item {
     644             :         char *name;
     645             :         TAILQ_ENTRY(spdk_bdev_examine_item) link;
     646             : };
     647             : 
     648             : TAILQ_HEAD(spdk_bdev_examine_allowlist, spdk_bdev_examine_item);
     649             : 
     650             : struct spdk_bdev_examine_allowlist g_bdev_examine_allowlist = TAILQ_HEAD_INITIALIZER(
     651             :                         g_bdev_examine_allowlist);
     652             : 
     653             : static inline bool
     654           6 : bdev_examine_allowlist_check(const char *name)
     655             : {
     656             :         struct spdk_bdev_examine_item *item;
     657           6 :         TAILQ_FOREACH(item, &g_bdev_examine_allowlist, link) {
     658           0 :                 if (strcmp(name, item->name) == 0) {
     659           0 :                         return true;
     660             :                 }
     661           0 :         }
     662           6 :         return false;
     663           6 : }
     664             : 
     665             : static inline void
     666          65 : bdev_examine_allowlist_free(void)
     667             : {
     668             :         struct spdk_bdev_examine_item *item;
     669          65 :         while (!TAILQ_EMPTY(&g_bdev_examine_allowlist)) {
     670           0 :                 item = TAILQ_FIRST(&g_bdev_examine_allowlist);
     671           0 :                 TAILQ_REMOVE(&g_bdev_examine_allowlist, item, link);
     672           0 :                 free(item->name);
     673           0 :                 free(item);
     674             :         }
     675          65 : }
     676             : 
     677             : static inline bool
     678           3 : bdev_in_examine_allowlist(struct spdk_bdev *bdev)
     679             : {
     680             :         struct spdk_bdev_alias *tmp;
     681           3 :         if (bdev_examine_allowlist_check(bdev->name)) {
     682           0 :                 return true;
     683             :         }
     684           6 :         TAILQ_FOREACH(tmp, &bdev->aliases, tailq) {
     685           3 :                 if (bdev_examine_allowlist_check(tmp->alias.name)) {
     686           0 :                         return true;
     687             :                 }
     688           3 :         }
     689           3 :         return false;
     690           3 : }
     691             : 
     692             : static inline bool
     693         126 : bdev_ok_to_examine(struct spdk_bdev *bdev)
     694             : {
     695             :         /* Some bdevs may not support the READ command.
     696             :          * Do not try to examine them.
     697             :          */
     698         126 :         if (!spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_READ)) {
     699           0 :                 return false;
     700             :         }
     701             : 
     702         126 :         if (g_bdev_opts.bdev_auto_examine) {
     703         123 :                 return true;
     704             :         } else {
     705           3 :                 return bdev_in_examine_allowlist(bdev);
     706             :         }
     707         126 : }
     708             : 
     709             : static void
     710         126 : bdev_examine(struct spdk_bdev *bdev)
     711             : {
     712             :         struct spdk_bdev_module *module;
     713             :         struct spdk_bdev_module_claim *claim, *tmpclaim;
     714             :         uint32_t action;
     715             : 
     716         126 :         if (!bdev_ok_to_examine(bdev)) {
     717           3 :                 return;
     718             :         }
     719             : 
     720         504 :         TAILQ_FOREACH(module, &g_bdev_mgr.bdev_modules, internal.tailq) {
     721         381 :                 if (module->examine_config) {
     722         258 :                         spdk_spin_lock(&module->internal.spinlock);
     723         258 :                         action = module->internal.action_in_progress;
     724         258 :                         module->internal.action_in_progress++;
     725         258 :                         spdk_spin_unlock(&module->internal.spinlock);
     726         258 :                         module->examine_config(bdev);
     727         258 :                         if (action != module->internal.action_in_progress) {
     728           0 :                                 SPDK_ERRLOG("examine_config for module %s did not call "
     729             :                                             "spdk_bdev_module_examine_done()\n", module->name);
     730           0 :                         }
     731         258 :                 }
     732         381 :         }
     733             : 
     734         123 :         spdk_spin_lock(&bdev->internal.spinlock);
     735             : 
     736         123 :         switch (bdev->internal.claim_type) {
     737             :         case SPDK_BDEV_CLAIM_NONE:
     738             :                 /* Examine by all bdev modules */
     739         479 :                 TAILQ_FOREACH(module, &g_bdev_mgr.bdev_modules, internal.tailq) {
     740         361 :                         if (module->examine_disk) {
     741         234 :                                 spdk_spin_lock(&module->internal.spinlock);
     742         234 :                                 module->internal.action_in_progress++;
     743         234 :                                 spdk_spin_unlock(&module->internal.spinlock);
     744         234 :                                 spdk_spin_unlock(&bdev->internal.spinlock);
     745         234 :                                 module->examine_disk(bdev);
     746         234 :                                 spdk_spin_lock(&bdev->internal.spinlock);
     747         234 :                         }
     748         361 :                 }
     749         118 :                 break;
     750             :         case SPDK_BDEV_CLAIM_EXCL_WRITE:
     751             :                 /* Examine by the one bdev module with a v1 claim */
     752           1 :                 module = bdev->internal.claim.v1.module;
     753           1 :                 if (module->examine_disk) {
     754           1 :                         spdk_spin_lock(&module->internal.spinlock);
     755           1 :                         module->internal.action_in_progress++;
     756           1 :                         spdk_spin_unlock(&module->internal.spinlock);
     757           1 :                         spdk_spin_unlock(&bdev->internal.spinlock);
     758           1 :                         module->examine_disk(bdev);
     759           1 :                         return;
     760             :                 }
     761           0 :                 break;
     762             :         default:
     763             :                 /* Examine by all bdev modules with a v2 claim */
     764           4 :                 assert(claim_type_is_v2(bdev->internal.claim_type));
     765             :                 /*
     766             :                  * Removal of tailq nodes while iterating can cause the iteration to jump out of the
     767             :                  * list, perhaps accessing freed memory. Without protection, this could happen
     768             :                  * while the lock is dropped during the examine callback.
     769             :                  */
     770           4 :                 bdev->internal.examine_in_progress++;
     771             : 
     772           9 :                 TAILQ_FOREACH(claim, &bdev->internal.claim.v2.claims, link) {
     773           5 :                         module = claim->module;
     774             : 
     775           5 :                         if (module == NULL) {
     776             :                                 /* This is a vestigial claim, held by examine_count */
     777           0 :                                 continue;
     778             :                         }
     779             : 
     780           5 :                         if (module->examine_disk == NULL) {
     781           0 :                                 continue;
     782             :                         }
     783             : 
     784           5 :                         spdk_spin_lock(&module->internal.spinlock);
     785           5 :                         module->internal.action_in_progress++;
     786           5 :                         spdk_spin_unlock(&module->internal.spinlock);
     787             : 
     788             :                         /* Call examine_disk without holding internal.spinlock. */
     789           5 :                         spdk_spin_unlock(&bdev->internal.spinlock);
     790           5 :                         module->examine_disk(bdev);
     791           5 :                         spdk_spin_lock(&bdev->internal.spinlock);
     792           5 :                 }
     793             : 
     794           4 :                 assert(bdev->internal.examine_in_progress > 0);
     795           4 :                 bdev->internal.examine_in_progress--;
     796           4 :                 if (bdev->internal.examine_in_progress == 0) {
     797             :                         /* Remove any claims that were released during examine_disk */
     798           9 :                         TAILQ_FOREACH_SAFE(claim, &bdev->internal.claim.v2.claims, link, tmpclaim) {
     799           5 :                                 if (claim->desc != NULL) {
     800           5 :                                         continue;
     801             :                                 }
     802             : 
     803           0 :                                 TAILQ_REMOVE(&bdev->internal.claim.v2.claims, claim, link);
     804           0 :                                 free(claim);
     805           0 :                         }
     806           4 :                         if (TAILQ_EMPTY(&bdev->internal.claim.v2.claims)) {
     807           0 :                                 claim_reset(bdev);
     808           0 :                         }
     809           4 :                 }
     810           4 :         }
     811             : 
     812         122 :         spdk_spin_unlock(&bdev->internal.spinlock);
     813         126 : }
     814             : 
     815             : int
     816           1 : spdk_bdev_examine(const char *name)
     817             : {
     818             :         struct spdk_bdev *bdev;
     819             :         struct spdk_bdev_examine_item *item;
     820           1 :         struct spdk_thread *thread = spdk_get_thread();
     821             : 
     822           1 :         if (spdk_unlikely(!spdk_thread_is_app_thread(thread))) {
     823           1 :                 SPDK_ERRLOG("Cannot examine bdev %s on thread %p (%s)\n", name, thread,
     824             :                             thread ? spdk_thread_get_name(thread) : "null");
     825           1 :                 return -EINVAL;
     826             :         }
     827             : 
     828           0 :         if (g_bdev_opts.bdev_auto_examine) {
     829           0 :                 SPDK_ERRLOG("Manual examine is not allowed if auto examine is enabled\n");
     830           0 :                 return -EINVAL;
     831             :         }
     832             : 
     833           0 :         if (bdev_examine_allowlist_check(name)) {
     834           0 :                 SPDK_ERRLOG("Duplicate bdev name for manual examine: %s\n", name);
     835           0 :                 return -EEXIST;
     836             :         }
     837             : 
     838           0 :         item = calloc(1, sizeof(*item));
     839           0 :         if (!item) {
     840           0 :                 return -ENOMEM;
     841             :         }
     842           0 :         item->name = strdup(name);
     843           0 :         if (!item->name) {
     844           0 :                 free(item);
     845           0 :                 return -ENOMEM;
     846             :         }
     847           0 :         TAILQ_INSERT_TAIL(&g_bdev_examine_allowlist, item, link);
     848             : 
     849           0 :         bdev = spdk_bdev_get_by_name(name);
     850           0 :         if (bdev) {
     851           0 :                 bdev_examine(bdev);
     852           0 :         }
     853           0 :         return 0;
     854           1 : }
     855             : 
     856             : static inline void
     857           0 : bdev_examine_allowlist_config_json(struct spdk_json_write_ctx *w)
     858             : {
     859             :         struct spdk_bdev_examine_item *item;
     860           0 :         TAILQ_FOREACH(item, &g_bdev_examine_allowlist, link) {
     861           0 :                 spdk_json_write_object_begin(w);
     862           0 :                 spdk_json_write_named_string(w, "method", "bdev_examine");
     863           0 :                 spdk_json_write_named_object_begin(w, "params");
     864           0 :                 spdk_json_write_named_string(w, "name", item->name);
     865           0 :                 spdk_json_write_object_end(w);
     866           0 :                 spdk_json_write_object_end(w);
     867           0 :         }
     868           0 : }
     869             : 
     870             : struct spdk_bdev *
     871           1 : spdk_bdev_first(void)
     872             : {
     873             :         struct spdk_bdev *bdev;
     874             : 
     875           1 :         bdev = TAILQ_FIRST(&g_bdev_mgr.bdevs);
     876           1 :         if (bdev) {
     877           1 :                 SPDK_DEBUGLOG(bdev, "Starting bdev iteration at %s\n", bdev->name);
     878           1 :         }
     879             : 
     880           1 :         return bdev;
     881             : }
     882             : 
     883             : struct spdk_bdev *
     884           8 : spdk_bdev_next(struct spdk_bdev *prev)
     885             : {
     886             :         struct spdk_bdev *bdev;
     887             : 
     888           8 :         bdev = TAILQ_NEXT(prev, internal.link);
     889           8 :         if (bdev) {
     890           7 :                 SPDK_DEBUGLOG(bdev, "Continuing bdev iteration at %s\n", bdev->name);
     891           7 :         }
     892             : 
     893           8 :         return bdev;
     894             : }
     895             : 
     896             : static struct spdk_bdev *
     897           6 : _bdev_next_leaf(struct spdk_bdev *bdev)
     898             : {
     899           9 :         while (bdev != NULL) {
     900           8 :                 if (bdev->internal.claim_type == SPDK_BDEV_CLAIM_NONE) {
     901           5 :                         return bdev;
     902             :                 } else {
     903           3 :                         bdev = TAILQ_NEXT(bdev, internal.link);
     904             :                 }
     905             :         }
     906             : 
     907           1 :         return bdev;
     908           6 : }
     909             : 
     910             : struct spdk_bdev *
     911           1 : spdk_bdev_first_leaf(void)
     912             : {
     913             :         struct spdk_bdev *bdev;
     914             : 
     915           1 :         bdev = _bdev_next_leaf(TAILQ_FIRST(&g_bdev_mgr.bdevs));
     916             : 
     917           1 :         if (bdev) {
     918           1 :                 SPDK_DEBUGLOG(bdev, "Starting bdev iteration at %s\n", bdev->name);
     919           1 :         }
     920             : 
     921           1 :         return bdev;
     922             : }
     923             : 
     924             : struct spdk_bdev *
     925           5 : spdk_bdev_next_leaf(struct spdk_bdev *prev)
     926             : {
     927             :         struct spdk_bdev *bdev;
     928             : 
     929           5 :         bdev = _bdev_next_leaf(TAILQ_NEXT(prev, internal.link));
     930             : 
     931           5 :         if (bdev) {
     932           4 :                 SPDK_DEBUGLOG(bdev, "Continuing bdev iteration at %s\n", bdev->name);
     933           4 :         }
     934             : 
     935           5 :         return bdev;
     936             : }
     937             : 
     938             : static inline bool
     939         820 : bdev_io_use_memory_domain(struct spdk_bdev_io *bdev_io)
     940             : {
     941         820 :         return bdev_io->internal.f.has_memory_domain;
     942             : }
     943             : 
     944             : static inline bool
     945        1549 : bdev_io_use_accel_sequence(struct spdk_bdev_io *bdev_io)
     946             : {
     947        1549 :         return bdev_io->internal.f.has_accel_sequence;
     948             : }
     949             : 
     950             : static inline uint32_t
     951         373 : bdev_desc_get_block_size(struct spdk_bdev_desc *desc)
     952             : {
     953         373 :         struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
     954             : 
     955         373 :         if (spdk_unlikely(desc->opts.no_metadata)) {
     956           0 :                 return bdev->blocklen - bdev->md_len;
     957             :         } else {
     958         373 :                 return bdev->blocklen;
     959             :         }
     960         373 : }
     961             : 
     962             : static inline uint32_t
     963         110 : bdev_io_get_block_size(struct spdk_bdev_io *bdev_io)
     964             : {
     965         110 :         return bdev_desc_get_block_size(bdev_io->internal.desc);
     966             : }
     967             : 
     968             : static inline void
     969           7 : bdev_queue_nomem_io_head(struct spdk_bdev_shared_resource *shared_resource,
     970             :                          struct spdk_bdev_io *bdev_io, enum bdev_io_retry_state state)
     971             : {
     972             :         /* Wait for some of the outstanding I/O to complete before we retry any of the nomem_io.
     973             :          * Normally we will wait for NOMEM_THRESHOLD_COUNT I/O to complete but for low queue depth
     974             :          * channels we will instead wait for half to complete.
     975             :          */
     976           7 :         shared_resource->nomem_threshold = spdk_max((int64_t)shared_resource->io_outstanding / 2,
     977             :                                            (int64_t)shared_resource->io_outstanding - NOMEM_THRESHOLD_COUNT);
     978             : 
     979           7 :         assert(state != BDEV_IO_RETRY_STATE_INVALID);
     980           7 :         bdev_io->internal.retry_state = state;
     981           7 :         TAILQ_INSERT_HEAD(&shared_resource->nomem_io, bdev_io, internal.link);
     982           7 : }
     983             : 
     984             : static inline void
     985          43 : bdev_queue_nomem_io_tail(struct spdk_bdev_shared_resource *shared_resource,
     986             :                          struct spdk_bdev_io *bdev_io, enum bdev_io_retry_state state)
     987             : {
     988             :         /* We only queue IOs at the end of the nomem_io queue if they're submitted by the user while
     989             :          * the queue isn't empty, so we don't need to update the nomem_threshold here */
     990          43 :         assert(!TAILQ_EMPTY(&shared_resource->nomem_io));
     991             : 
     992          43 :         assert(state != BDEV_IO_RETRY_STATE_INVALID);
     993          43 :         bdev_io->internal.retry_state = state;
     994          43 :         TAILQ_INSERT_TAIL(&shared_resource->nomem_io, bdev_io, internal.link);
     995          43 : }
     996             : 
     997             : void
     998          16 : spdk_bdev_io_set_buf(struct spdk_bdev_io *bdev_io, void *buf, size_t len)
     999             : {
    1000             :         struct iovec *iovs;
    1001             : 
    1002          16 :         if (bdev_io->u.bdev.iovs == NULL) {
    1003           3 :                 bdev_io->u.bdev.iovs = &bdev_io->iov;
    1004           3 :                 bdev_io->u.bdev.iovcnt = 1;
    1005           3 :         }
    1006             : 
    1007          16 :         iovs = bdev_io->u.bdev.iovs;
    1008             : 
    1009          16 :         assert(iovs != NULL);
    1010          16 :         assert(bdev_io->u.bdev.iovcnt >= 1);
    1011             : 
    1012          16 :         iovs[0].iov_base = buf;
    1013          16 :         iovs[0].iov_len = len;
    1014          16 : }
    1015             : 
    1016             : void
    1017           3 : spdk_bdev_io_set_md_buf(struct spdk_bdev_io *bdev_io, void *md_buf, size_t len)
    1018             : {
    1019           3 :         assert((len / spdk_bdev_get_md_size(bdev_io->bdev)) >= bdev_io->u.bdev.num_blocks);
    1020           3 :         bdev_io->u.bdev.md_buf = md_buf;
    1021           3 : }
    1022             : 
    1023             : static bool
    1024         167 : _is_buf_allocated(const struct iovec *iovs)
    1025             : {
    1026         167 :         if (iovs == NULL) {
    1027           6 :                 return false;
    1028             :         }
    1029             : 
    1030         161 :         return iovs[0].iov_base != NULL;
    1031         167 : }
    1032             : 
    1033             : static bool
    1034          50 : _are_iovs_aligned(struct iovec *iovs, int iovcnt, uint32_t alignment)
    1035             : {
    1036             :         int i;
    1037             :         uintptr_t iov_base;
    1038             : 
    1039          50 :         if (spdk_likely(alignment == 1)) {
    1040          21 :                 return true;
    1041             :         }
    1042             : 
    1043          36 :         for (i = 0; i < iovcnt; i++) {
    1044          29 :                 iov_base = (uintptr_t)iovs[i].iov_base;
    1045          29 :                 if ((iov_base & (alignment - 1)) != 0) {
    1046          22 :                         return false;
    1047             :                 }
    1048           7 :         }
    1049             : 
    1050           7 :         return true;
    1051          50 : }
    1052             : 
    1053             : static inline bool
    1054         895 : bdev_io_needs_metadata(struct spdk_bdev_desc *desc, struct spdk_bdev_io *bdev_io)
    1055             : {
    1056         895 :         return desc->opts.no_metadata && bdev_io->bdev->md_len != 0;
    1057             : }
    1058             : 
    1059             : static inline bool
    1060         852 : bdev_io_needs_sequence_exec(struct spdk_bdev_desc *desc, struct spdk_bdev_io *bdev_io)
    1061             : {
    1062         852 :         if (!bdev_io_use_accel_sequence(bdev_io)) {
    1063         852 :                 return false;
    1064             :         }
    1065             : 
    1066             :         /* For now, we don't allow splitting IOs with an accel sequence and will treat them as if
    1067             :          * bdev module didn't support accel sequences */
    1068           0 :         return !desc->accel_sequence_supported[bdev_io->type] || bdev_io->internal.f.split;
    1069         852 : }
    1070             : 
    1071             : static inline void
    1072         592 : bdev_io_increment_outstanding(struct spdk_bdev_channel *bdev_ch,
    1073             :                               struct spdk_bdev_shared_resource *shared_resource)
    1074             : {
    1075         592 :         bdev_ch->io_outstanding++;
    1076         592 :         shared_resource->io_outstanding++;
    1077         592 : }
    1078             : 
    1079             : static inline void
    1080         592 : bdev_io_decrement_outstanding(struct spdk_bdev_channel *bdev_ch,
    1081             :                               struct spdk_bdev_shared_resource *shared_resource)
    1082             : {
    1083         592 :         assert(bdev_ch->io_outstanding > 0);
    1084         592 :         assert(shared_resource->io_outstanding > 0);
    1085         592 :         bdev_ch->io_outstanding--;
    1086         592 :         shared_resource->io_outstanding--;
    1087         592 : }
    1088             : 
    1089             : static void
    1090           0 : bdev_io_submit_sequence_cb(void *ctx, int status)
    1091             : {
    1092           0 :         struct spdk_bdev_io *bdev_io = ctx;
    1093             : 
    1094           0 :         assert(bdev_io_use_accel_sequence(bdev_io));
    1095             : 
    1096           0 :         bdev_io->u.bdev.accel_sequence = NULL;
    1097           0 :         bdev_io->internal.f.has_accel_sequence = false;
    1098             : 
    1099           0 :         if (spdk_unlikely(status != 0)) {
    1100           0 :                 SPDK_ERRLOG("Failed to execute accel sequence, status=%d\n", status);
    1101           0 :                 bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
    1102           0 :                 bdev_io_complete_unsubmitted(bdev_io);
    1103           0 :                 return;
    1104             :         }
    1105             : 
    1106           0 :         bdev_io_submit(bdev_io);
    1107           0 : }
    1108             : 
    1109             : static void
    1110           0 : bdev_io_exec_sequence_cb(void *ctx, int status)
    1111             : {
    1112           0 :         struct spdk_bdev_io *bdev_io = ctx;
    1113           0 :         struct spdk_bdev_channel *ch = bdev_io->internal.ch;
    1114             : 
    1115           0 :         TAILQ_REMOVE(&bdev_io->internal.ch->io_accel_exec, bdev_io, internal.link);
    1116           0 :         bdev_io_decrement_outstanding(ch, ch->shared_resource);
    1117             : 
    1118           0 :         if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
    1119           0 :                 bdev_ch_retry_io(ch);
    1120           0 :         }
    1121             : 
    1122           0 :         bdev_io->internal.data_transfer_cpl(bdev_io, status);
    1123           0 : }
    1124             : 
    1125             : static void
    1126           0 : bdev_io_exec_sequence(struct spdk_bdev_io *bdev_io, void (*cb_fn)(void *ctx, int status))
    1127             : {
    1128           0 :         struct spdk_bdev_channel *ch = bdev_io->internal.ch;
    1129             : 
    1130           0 :         assert(bdev_io_needs_sequence_exec(bdev_io->internal.desc, bdev_io));
    1131           0 :         assert(bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE || bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
    1132           0 :         assert(bdev_io_use_accel_sequence(bdev_io));
    1133             : 
    1134             :         /* Since the operations are appended during submission, they're in the opposite order than
    1135             :          * how we want to execute them for reads (i.e. we need to execute the most recently added
    1136             :          * operation first), so reverse the sequence before executing it.
    1137             :          */
    1138           0 :         if (bdev_io->type == SPDK_BDEV_IO_TYPE_READ) {
    1139           0 :                 spdk_accel_sequence_reverse(bdev_io->internal.accel_sequence);
    1140           0 :         }
    1141             : 
    1142           0 :         TAILQ_INSERT_TAIL(&bdev_io->internal.ch->io_accel_exec, bdev_io, internal.link);
    1143           0 :         bdev_io_increment_outstanding(ch, ch->shared_resource);
    1144           0 :         bdev_io->internal.data_transfer_cpl = cb_fn;
    1145             : 
    1146           0 :         spdk_accel_sequence_finish(bdev_io->internal.accel_sequence,
    1147           0 :                                    bdev_io_exec_sequence_cb, bdev_io);
    1148           0 : }
    1149             : 
    1150             : static void
    1151          42 : bdev_io_get_buf_complete(struct spdk_bdev_io *bdev_io, bool status)
    1152             : {
    1153          42 :         struct spdk_io_channel *ch = spdk_bdev_io_get_io_channel(bdev_io);
    1154             :         void *buf;
    1155             : 
    1156          42 :         if (spdk_unlikely(bdev_io->internal.get_aux_buf_cb != NULL)) {
    1157           0 :                 buf = bdev_io->internal.buf.ptr;
    1158           0 :                 bdev_io->internal.buf.ptr = NULL;
    1159           0 :                 bdev_io->internal.f.has_buf = false;
    1160           0 :                 bdev_io->internal.get_aux_buf_cb(ch, bdev_io, buf);
    1161           0 :                 bdev_io->internal.get_aux_buf_cb = NULL;
    1162           0 :         } else {
    1163          42 :                 assert(bdev_io->internal.get_buf_cb != NULL);
    1164          42 :                 bdev_io->internal.get_buf_cb(ch, bdev_io, status);
    1165          42 :                 bdev_io->internal.get_buf_cb = NULL;
    1166             :         }
    1167          42 : }
    1168             : 
    1169             : static void
    1170           4 : _bdev_io_pull_buffer_cpl(void *ctx, int rc)
    1171             : {
    1172           4 :         struct spdk_bdev_io *bdev_io = ctx;
    1173             : 
    1174           4 :         if (rc) {
    1175           0 :                 SPDK_ERRLOG("Set bounce buffer failed with rc %d\n", rc);
    1176           0 :                 bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
    1177           0 :         }
    1178           4 :         bdev_io_get_buf_complete(bdev_io, !rc);
    1179           4 : }
    1180             : 
    1181             : static void
    1182           2 : bdev_io_pull_md_buf_done(void *ctx, int status)
    1183             : {
    1184           2 :         struct spdk_bdev_io *bdev_io = ctx;
    1185           2 :         struct spdk_bdev_channel *ch = bdev_io->internal.ch;
    1186             : 
    1187           2 :         TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
    1188           2 :         bdev_io_decrement_outstanding(ch, ch->shared_resource);
    1189             : 
    1190           2 :         if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
    1191           0 :                 bdev_ch_retry_io(ch);
    1192           0 :         }
    1193             : 
    1194           2 :         assert(bdev_io->internal.data_transfer_cpl);
    1195           2 :         bdev_io->internal.data_transfer_cpl(bdev_io, status);
    1196           2 : }
    1197             : 
    1198             : static void
    1199           4 : bdev_io_pull_md_buf(struct spdk_bdev_io *bdev_io)
    1200             : {
    1201           4 :         struct spdk_bdev_channel *ch = bdev_io->internal.ch;
    1202           4 :         int rc = 0;
    1203             : 
    1204           4 :         if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
    1205           2 :                 assert(bdev_io->internal.f.has_bounce_buf);
    1206           2 :                 if (bdev_io_use_memory_domain(bdev_io)) {
    1207           2 :                         TAILQ_INSERT_TAIL(&ch->io_memory_domain, bdev_io, internal.link);
    1208           2 :                         bdev_io_increment_outstanding(ch, ch->shared_resource);
    1209           4 :                         rc = spdk_memory_domain_pull_data(bdev_io->internal.memory_domain,
    1210           2 :                                                           bdev_io->internal.memory_domain_ctx,
    1211           2 :                                                           &bdev_io->internal.bounce_buf.orig_md_iov, 1,
    1212           2 :                                                           &bdev_io->internal.bounce_buf.md_iov, 1,
    1213           2 :                                                           bdev_io_pull_md_buf_done, bdev_io);
    1214           2 :                         if (rc == 0) {
    1215             :                                 /* Continue to submit IO in completion callback */
    1216           2 :                                 return;
    1217             :                         }
    1218           0 :                         bdev_io_decrement_outstanding(ch, ch->shared_resource);
    1219           0 :                         TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
    1220           0 :                         if (rc != -ENOMEM) {
    1221           0 :                                 SPDK_ERRLOG("Failed to pull data from memory domain %s, rc %d\n",
    1222             :                                             spdk_memory_domain_get_dma_device_id(
    1223             :                                                     bdev_io->internal.memory_domain), rc);
    1224           0 :                         }
    1225           0 :                 } else {
    1226           0 :                         memcpy(bdev_io->internal.bounce_buf.md_iov.iov_base,
    1227           0 :                                bdev_io->internal.bounce_buf.orig_md_iov.iov_base,
    1228           0 :                                bdev_io->internal.bounce_buf.orig_md_iov.iov_len);
    1229             :                 }
    1230           0 :         }
    1231             : 
    1232           2 :         if (spdk_unlikely(rc == -ENOMEM)) {
    1233           0 :                 bdev_queue_nomem_io_head(ch->shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PULL_MD);
    1234           0 :         } else {
    1235           2 :                 assert(bdev_io->internal.data_transfer_cpl);
    1236           2 :                 bdev_io->internal.data_transfer_cpl(bdev_io, rc);
    1237             :         }
    1238           4 : }
    1239             : 
    1240             : static void
    1241           4 : _bdev_io_pull_bounce_md_buf(struct spdk_bdev_io *bdev_io, void *md_buf, size_t len)
    1242             : {
    1243           4 :         assert(bdev_io->internal.f.has_bounce_buf);
    1244             : 
    1245             :         /* save original md_buf */
    1246           4 :         bdev_io->internal.bounce_buf.orig_md_iov.iov_base = bdev_io->u.bdev.md_buf;
    1247           4 :         bdev_io->internal.bounce_buf.orig_md_iov.iov_len = len;
    1248           4 :         bdev_io->internal.bounce_buf.md_iov.iov_base = md_buf;
    1249           4 :         bdev_io->internal.bounce_buf.md_iov.iov_len = len;
    1250             :         /* set bounce md_buf */
    1251           4 :         bdev_io->u.bdev.md_buf = md_buf;
    1252             : 
    1253           4 :         bdev_io_pull_md_buf(bdev_io);
    1254           4 : }
    1255             : 
    1256             : static void
    1257          42 : _bdev_io_set_md_buf(struct spdk_bdev_io *bdev_io)
    1258             : {
    1259          42 :         struct spdk_bdev *bdev = bdev_io->bdev;
    1260             :         uint64_t md_len;
    1261             :         void *buf;
    1262             : 
    1263          42 :         if (spdk_bdev_is_md_separate(bdev)) {
    1264           7 :                 assert(!bdev_io_use_accel_sequence(bdev_io));
    1265             : 
    1266           7 :                 buf = (char *)bdev_io->u.bdev.iovs[0].iov_base + bdev_io->u.bdev.iovs[0].iov_len;
    1267           7 :                 md_len = bdev_io->u.bdev.num_blocks * bdev->md_len;
    1268             : 
    1269           7 :                 assert(((uintptr_t)buf & (spdk_bdev_get_buf_align(bdev) - 1)) == 0);
    1270             : 
    1271           7 :                 if (bdev_io->u.bdev.md_buf != NULL) {
    1272           4 :                         _bdev_io_pull_bounce_md_buf(bdev_io, buf, md_len);
    1273           4 :                         return;
    1274             :                 } else {
    1275           3 :                         spdk_bdev_io_set_md_buf(bdev_io, buf, md_len);
    1276             :                 }
    1277           3 :         }
    1278             : 
    1279          38 :         bdev_io_get_buf_complete(bdev_io, true);
    1280          42 : }
    1281             : 
    1282             : static inline void
    1283          26 : bdev_io_pull_data_done(struct spdk_bdev_io *bdev_io, int rc)
    1284             : {
    1285          26 :         if (rc) {
    1286           0 :                 SPDK_ERRLOG("Failed to get data buffer\n");
    1287           0 :                 assert(bdev_io->internal.data_transfer_cpl);
    1288           0 :                 bdev_io->internal.data_transfer_cpl(bdev_io, rc);
    1289           0 :                 return;
    1290             :         }
    1291             : 
    1292          26 :         _bdev_io_set_md_buf(bdev_io);
    1293          26 : }
    1294             : 
    1295             : static void
    1296           2 : bdev_io_pull_data_done_and_track(void *ctx, int status)
    1297             : {
    1298           2 :         struct spdk_bdev_io *bdev_io = ctx;
    1299           2 :         struct spdk_bdev_channel *ch = bdev_io->internal.ch;
    1300             : 
    1301           2 :         TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
    1302           2 :         bdev_io_decrement_outstanding(ch, ch->shared_resource);
    1303             : 
    1304           2 :         if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
    1305           0 :                 bdev_ch_retry_io(ch);
    1306           0 :         }
    1307             : 
    1308           2 :         bdev_io_pull_data_done(bdev_io, status);
    1309           2 : }
    1310             : 
    1311             : static void
    1312          27 : bdev_io_pull_data(struct spdk_bdev_io *bdev_io)
    1313             : {
    1314          27 :         struct spdk_bdev_channel *ch = bdev_io->internal.ch;
    1315          27 :         struct spdk_bdev_desc *desc = bdev_io->internal.desc;
    1316          27 :         int rc = 0;
    1317             : 
    1318          27 :         assert(bdev_io->internal.f.has_bounce_buf);
    1319             : 
    1320          27 :         if (bdev_io_needs_metadata(desc, bdev_io)) {
    1321           0 :                 assert(bdev_io->bdev->md_interleave);
    1322             : 
    1323           0 :                 if (!bdev_io_use_accel_sequence(bdev_io)) {
    1324           0 :                         bdev_io->internal.accel_sequence = NULL;
    1325           0 :                 }
    1326             : 
    1327           0 :                 if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
    1328           0 :                         rc = spdk_accel_append_dif_generate_copy(&bdev_io->internal.accel_sequence, ch->accel_channel,
    1329           0 :                                         bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt,
    1330           0 :                                         bdev_io->u.bdev.memory_domain,
    1331           0 :                                         bdev_io->u.bdev.memory_domain_ctx,
    1332           0 :                                         bdev_io->internal.bounce_buf.orig_iovs,
    1333           0 :                                         bdev_io->internal.bounce_buf.orig_iovcnt,
    1334           0 :                                         bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain : NULL,
    1335           0 :                                         bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain_ctx : NULL,
    1336           0 :                                         bdev_io->u.bdev.num_blocks,
    1337           0 :                                         &bdev_io->u.bdev.dif_ctx,
    1338             :                                         NULL, NULL);
    1339           0 :                 } else {
    1340           0 :                         assert(bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
    1341           0 :                         rc = spdk_accel_append_dif_verify_copy(&bdev_io->internal.accel_sequence, ch->accel_channel,
    1342           0 :                                                                bdev_io->internal.bounce_buf.orig_iovs,
    1343           0 :                                                                bdev_io->internal.bounce_buf.orig_iovcnt,
    1344           0 :                                                                bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain : NULL,
    1345           0 :                                                                bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain_ctx : NULL,
    1346           0 :                                                                bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt,
    1347           0 :                                                                bdev_io->u.bdev.memory_domain,
    1348           0 :                                                                bdev_io->u.bdev.memory_domain_ctx,
    1349           0 :                                                                bdev_io->u.bdev.num_blocks,
    1350           0 :                                                                &bdev_io->u.bdev.dif_ctx,
    1351           0 :                                                                &bdev_io->u.bdev.dif_err,
    1352             :                                                                NULL, NULL);
    1353             :                 }
    1354             : 
    1355           0 :                 if (spdk_likely(rc == 0)) {
    1356           0 :                         bdev_io->internal.f.has_accel_sequence = true;
    1357           0 :                         bdev_io->u.bdev.accel_sequence = bdev_io->internal.accel_sequence;
    1358           0 :                 } else if (rc != -ENOMEM) {
    1359           0 :                         SPDK_ERRLOG("Failed to append generate/verify_copy to accel sequence: %p\n",
    1360             :                                     bdev_io->internal.accel_sequence);
    1361           0 :                 }
    1362          27 :         } else if (bdev_io_needs_sequence_exec(desc, bdev_io) ||
    1363          27 :                    (bdev_io_use_accel_sequence(bdev_io) && bdev_io_use_memory_domain(bdev_io))) {
    1364             :                 /* If we need to exec an accel sequence or the IO uses a memory domain buffer and has a
    1365             :                  * sequence, append a copy operation making accel change the src/dst buffers of the previous
    1366             :                  * operation */
    1367           0 :                 assert(bdev_io_use_accel_sequence(bdev_io));
    1368           0 :                 if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
    1369           0 :                         rc = spdk_accel_append_copy(&bdev_io->internal.accel_sequence, ch->accel_channel,
    1370           0 :                                                     bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt,
    1371             :                                                     NULL, NULL,
    1372           0 :                                                     bdev_io->internal.bounce_buf.orig_iovs,
    1373           0 :                                                     bdev_io->internal.bounce_buf.orig_iovcnt,
    1374           0 :                                                     bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain : NULL,
    1375           0 :                                                     bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain_ctx : NULL,
    1376             :                                                     NULL, NULL);
    1377           0 :                 } else {
    1378             :                         /* We need to reverse the src/dst for reads */
    1379           0 :                         assert(bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
    1380           0 :                         rc = spdk_accel_append_copy(&bdev_io->internal.accel_sequence, ch->accel_channel,
    1381           0 :                                                     bdev_io->internal.bounce_buf.orig_iovs,
    1382           0 :                                                     bdev_io->internal.bounce_buf.orig_iovcnt,
    1383           0 :                                                     bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain : NULL,
    1384           0 :                                                     bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain_ctx : NULL,
    1385           0 :                                                     bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt,
    1386             :                                                     NULL, NULL, NULL, NULL);
    1387             :                 }
    1388             : 
    1389           0 :                 if (spdk_unlikely(rc != 0 && rc != -ENOMEM)) {
    1390           0 :                         SPDK_ERRLOG("Failed to append copy to accel sequence: %p\n",
    1391             :                                     bdev_io->internal.accel_sequence);
    1392           0 :                 }
    1393          27 :         } else if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
    1394             :                 /* if this is write path, copy data from original buffer to bounce buffer */
    1395          17 :                 if (bdev_io_use_memory_domain(bdev_io)) {
    1396           3 :                         TAILQ_INSERT_TAIL(&ch->io_memory_domain, bdev_io, internal.link);
    1397           3 :                         bdev_io_increment_outstanding(ch, ch->shared_resource);
    1398           6 :                         rc = spdk_memory_domain_pull_data(bdev_io->internal.memory_domain,
    1399           3 :                                                           bdev_io->internal.memory_domain_ctx,
    1400           3 :                                                           bdev_io->internal.bounce_buf.orig_iovs,
    1401           3 :                                                           (uint32_t)bdev_io->internal.bounce_buf.orig_iovcnt,
    1402           3 :                                                           bdev_io->u.bdev.iovs, 1,
    1403             :                                                           bdev_io_pull_data_done_and_track,
    1404           3 :                                                           bdev_io);
    1405           3 :                         if (rc == 0) {
    1406             :                                 /* Continue to submit IO in completion callback */
    1407           2 :                                 return;
    1408             :                         }
    1409           1 :                         TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
    1410           1 :                         bdev_io_decrement_outstanding(ch, ch->shared_resource);
    1411           1 :                         if (rc != -ENOMEM) {
    1412           0 :                                 SPDK_ERRLOG("Failed to pull data from memory domain %s\n",
    1413             :                                             spdk_memory_domain_get_dma_device_id(
    1414             :                                                     bdev_io->internal.memory_domain));
    1415           0 :                         }
    1416           1 :                 } else {
    1417          14 :                         assert(bdev_io->u.bdev.iovcnt == 1);
    1418          28 :                         spdk_copy_iovs_to_buf(bdev_io->u.bdev.iovs[0].iov_base,
    1419          14 :                                               bdev_io->u.bdev.iovs[0].iov_len,
    1420          14 :                                               bdev_io->internal.bounce_buf.orig_iovs,
    1421          14 :                                               bdev_io->internal.bounce_buf.orig_iovcnt);
    1422             :                 }
    1423          15 :         }
    1424             : 
    1425          25 :         if (spdk_unlikely(rc == -ENOMEM)) {
    1426           1 :                 bdev_queue_nomem_io_head(ch->shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PULL);
    1427           1 :         } else {
    1428          24 :                 bdev_io_pull_data_done(bdev_io, rc);
    1429             :         }
    1430          27 : }
    1431             : 
    1432             : static void
    1433          26 : _bdev_io_pull_bounce_data_buf(struct spdk_bdev_io *bdev_io, void *buf, size_t len,
    1434             :                               bdev_copy_bounce_buffer_cpl cpl_cb)
    1435             : {
    1436          26 :         struct spdk_bdev_shared_resource *shared_resource = bdev_io->internal.ch->shared_resource;
    1437             : 
    1438          26 :         assert(bdev_io->internal.f.has_bounce_buf == false);
    1439             : 
    1440          26 :         bdev_io->internal.data_transfer_cpl = cpl_cb;
    1441          26 :         bdev_io->internal.f.has_bounce_buf = true;
    1442             :         /* save original iovec */
    1443          26 :         bdev_io->internal.bounce_buf.orig_iovs = bdev_io->u.bdev.iovs;
    1444          26 :         bdev_io->internal.bounce_buf.orig_iovcnt = bdev_io->u.bdev.iovcnt;
    1445             :         /* zero the other data members */
    1446          26 :         bdev_io->internal.bounce_buf.iov.iov_base = NULL;
    1447          26 :         bdev_io->internal.bounce_buf.md_iov.iov_base = NULL;
    1448          26 :         bdev_io->internal.bounce_buf.orig_md_iov.iov_base = NULL;
    1449             :         /* set bounce iov */
    1450          26 :         bdev_io->u.bdev.iovs = &bdev_io->internal.bounce_buf.iov;
    1451          26 :         bdev_io->u.bdev.iovcnt = 1;
    1452             :         /* set bounce buffer for this operation */
    1453          26 :         bdev_io->u.bdev.iovs[0].iov_base = buf;
    1454          26 :         bdev_io->u.bdev.iovs[0].iov_len = len;
    1455             :         /* Now we use 1 iov, the split condition could have been changed */
    1456          26 :         bdev_io->internal.f.split = bdev_io_should_split(bdev_io);
    1457             : 
    1458          26 :         if (spdk_unlikely(!TAILQ_EMPTY(&shared_resource->nomem_io))) {
    1459           0 :                 bdev_queue_nomem_io_tail(shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PULL);
    1460           0 :         } else {
    1461          26 :                 bdev_io_pull_data(bdev_io);
    1462             :         }
    1463          26 : }
    1464             : 
    1465             : static void
    1466          42 : _bdev_io_set_buf(struct spdk_bdev_io *bdev_io, void *buf, uint64_t len)
    1467             : {
    1468          42 :         struct spdk_bdev *bdev = bdev_io->bdev;
    1469             :         bool buf_allocated;
    1470             :         uint64_t alignment;
    1471             :         void *aligned_buf;
    1472             : 
    1473          42 :         bdev_io->internal.buf.ptr = buf;
    1474          42 :         bdev_io->internal.f.has_buf = true;
    1475             : 
    1476          42 :         if (spdk_unlikely(bdev_io->internal.get_aux_buf_cb != NULL)) {
    1477           0 :                 bdev_io_get_buf_complete(bdev_io, true);
    1478           0 :                 return;
    1479             :         }
    1480             : 
    1481          42 :         alignment = spdk_bdev_get_buf_align(bdev);
    1482          42 :         buf_allocated = _is_buf_allocated(bdev_io->u.bdev.iovs);
    1483          42 :         aligned_buf = (void *)(((uintptr_t)buf + (alignment - 1)) & ~(alignment - 1));
    1484             : 
    1485          42 :         if (buf_allocated) {
    1486          26 :                 _bdev_io_pull_bounce_data_buf(bdev_io, aligned_buf, len, _bdev_io_pull_buffer_cpl);
    1487             :                 /* Continue in completion callback */
    1488          26 :                 return;
    1489             :         } else {
    1490          16 :                 spdk_bdev_io_set_buf(bdev_io, aligned_buf, len);
    1491             :         }
    1492             : 
    1493          16 :         _bdev_io_set_md_buf(bdev_io);
    1494          42 : }
    1495             : 
    1496             : static inline uint64_t
    1497          42 : bdev_io_get_max_buf_len(struct spdk_bdev_io *bdev_io, uint64_t len)
    1498             : {
    1499          42 :         struct spdk_bdev *bdev = bdev_io->bdev;
    1500             :         uint64_t md_len, alignment;
    1501             : 
    1502          42 :         md_len = spdk_bdev_is_md_separate(bdev) ? bdev_io->u.bdev.num_blocks * bdev->md_len : 0;
    1503             : 
    1504             :         /* 1 byte alignment needs 0 byte of extra space, 64 bytes alignment needs 63 bytes of extra space, etc. */
    1505          42 :         alignment = spdk_bdev_get_buf_align(bdev) - 1;
    1506             : 
    1507          42 :         return len + alignment + md_len;
    1508             : }
    1509             : 
    1510             : static void
    1511          42 : bdev_io_put_accel_buf(struct spdk_bdev_io *bdev_io)
    1512             : {
    1513          42 :         struct spdk_bdev_channel *ch = bdev_io->internal.ch;
    1514             : 
    1515          84 :         spdk_accel_put_buf(ch->accel_channel,
    1516          42 :                            bdev_io->internal.buf.ptr,
    1517          42 :                            bdev_io->u.bdev.memory_domain,
    1518          42 :                            bdev_io->u.bdev.memory_domain_ctx);
    1519          42 : }
    1520             : 
    1521             : static void
    1522           0 : _bdev_io_put_buf(struct spdk_bdev_io *bdev_io, void *buf, uint64_t buf_len)
    1523             : {
    1524             :         struct spdk_bdev_mgmt_channel *ch;
    1525             : 
    1526           0 :         ch = bdev_io->internal.ch->shared_resource->mgmt_ch;
    1527           0 :         spdk_iobuf_put(&ch->iobuf, buf, bdev_io_get_max_buf_len(bdev_io, buf_len));
    1528           0 : }
    1529             : 
    1530             : static void
    1531          42 : bdev_io_put_buf(struct spdk_bdev_io *bdev_io)
    1532             : {
    1533          42 :         assert(bdev_io->internal.f.has_buf);
    1534             : 
    1535          42 :         if (bdev_io->u.bdev.memory_domain == spdk_accel_get_memory_domain()) {
    1536          42 :                 bdev_io_put_accel_buf(bdev_io);
    1537          42 :         } else {
    1538           0 :                 assert(bdev_io->u.bdev.memory_domain == NULL);
    1539           0 :                 _bdev_io_put_buf(bdev_io, bdev_io->internal.buf.ptr,
    1540           0 :                                  bdev_io->internal.buf.len);
    1541             :         }
    1542          42 :         bdev_io->internal.buf.ptr = NULL;
    1543          42 :         bdev_io->internal.f.has_buf = false;
    1544          42 : }
    1545             : 
    1546           3 : SPDK_LOG_DEPRECATION_REGISTER(spdk_bdev_io_put_aux_buf,
    1547             :                               "spdk_bdev_io_put_aux_buf is deprecated", "v25.01", 0);
    1548             : 
    1549             : void
    1550           0 : spdk_bdev_io_put_aux_buf(struct spdk_bdev_io *bdev_io, void *buf)
    1551             : {
    1552           0 :         uint64_t len = bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen;
    1553             : 
    1554           0 :         SPDK_LOG_DEPRECATED(spdk_bdev_io_put_aux_buf);
    1555             : 
    1556           0 :         assert(buf != NULL);
    1557           0 :         _bdev_io_put_buf(bdev_io, buf, len);
    1558           0 : }
    1559             : 
    1560             : static inline void
    1561         548 : bdev_submit_request(struct spdk_bdev *bdev, struct spdk_io_channel *ioch,
    1562             :                     struct spdk_bdev_io *bdev_io)
    1563             : {
    1564             :         /* After a request is submitted to a bdev module, the ownership of an accel sequence
    1565             :          * associated with that bdev_io is transferred to the bdev module. So, clear the internal
    1566             :          * sequence pointer to make sure we won't touch it anymore. */
    1567        1014 :         if ((bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE ||
    1568         548 :              bdev_io->type == SPDK_BDEV_IO_TYPE_READ) && bdev_io->u.bdev.accel_sequence != NULL) {
    1569           0 :                 assert(!bdev_io_needs_sequence_exec(bdev_io->internal.desc, bdev_io));
    1570           0 :                 bdev_io->internal.f.has_accel_sequence = false;
    1571           0 :         }
    1572             : 
    1573         548 :         bdev->fn_table->submit_request(ioch, bdev_io);
    1574         548 : }
    1575             : 
    1576             : static inline void
    1577          10 : bdev_ch_resubmit_io(struct spdk_bdev_shared_resource *shared_resource, struct spdk_bdev_io *bdev_io)
    1578             : {
    1579          10 :         struct spdk_bdev *bdev = bdev_io->bdev;
    1580             : 
    1581          10 :         bdev_io_increment_outstanding(bdev_io->internal.ch, shared_resource);
    1582          10 :         bdev_io->internal.error.nvme.cdw0 = 0;
    1583          10 :         bdev_io->num_retries++;
    1584          10 :         bdev_submit_request(bdev, spdk_bdev_io_get_io_channel(bdev_io), bdev_io);
    1585          10 : }
    1586             : 
    1587             : static void
    1588          63 : bdev_shared_ch_retry_io(struct spdk_bdev_shared_resource *shared_resource)
    1589             : {
    1590             :         struct spdk_bdev_io *bdev_io;
    1591             : 
    1592          63 :         if (shared_resource->io_outstanding > shared_resource->nomem_threshold) {
    1593             :                 /*
    1594             :                  * Allow some more I/O to complete before retrying the nomem_io queue.
    1595             :                  *  Some drivers (such as nvme) cannot immediately take a new I/O in
    1596             :                  *  the context of a completion, because the resources for the I/O are
    1597             :                  *  not released until control returns to the bdev poller.  Also, we
    1598             :                  *  may require several small I/O to complete before a larger I/O
    1599             :                  *  (that requires splitting) can be submitted.
    1600             :                  */
    1601          58 :                 return;
    1602             :         }
    1603             : 
    1604          16 :         while (!TAILQ_EMPTY(&shared_resource->nomem_io)) {
    1605          12 :                 bdev_io = TAILQ_FIRST(&shared_resource->nomem_io);
    1606          12 :                 TAILQ_REMOVE(&shared_resource->nomem_io, bdev_io, internal.link);
    1607             : 
    1608          12 :                 switch (bdev_io->internal.retry_state) {
    1609             :                 case BDEV_IO_RETRY_STATE_SUBMIT:
    1610          10 :                         bdev_ch_resubmit_io(shared_resource, bdev_io);
    1611          10 :                         break;
    1612             :                 case BDEV_IO_RETRY_STATE_PULL:
    1613           1 :                         bdev_io_pull_data(bdev_io);
    1614           1 :                         break;
    1615             :                 case BDEV_IO_RETRY_STATE_PULL_MD:
    1616           0 :                         bdev_io_pull_md_buf(bdev_io);
    1617           0 :                         break;
    1618             :                 case BDEV_IO_RETRY_STATE_PUSH:
    1619           1 :                         bdev_io_push_bounce_data(bdev_io);
    1620           1 :                         break;
    1621             :                 case BDEV_IO_RETRY_STATE_PUSH_MD:
    1622           0 :                         bdev_io_push_bounce_md_buf(bdev_io);
    1623           0 :                         break;
    1624             :                 case BDEV_IO_RETRY_STATE_GET_ACCEL_BUF:
    1625           0 :                         _bdev_io_get_accel_buf(bdev_io);
    1626           0 :                         break;
    1627             :                 default:
    1628           0 :                         assert(0 && "invalid retry state");
    1629             :                         break;
    1630             :                 }
    1631             : 
    1632          12 :                 if (bdev_io == TAILQ_FIRST(&shared_resource->nomem_io)) {
    1633             :                         /* This IO completed again with NOMEM status, so break the loop and
    1634             :                          * don't try anymore.  Note that a bdev_io that fails with NOMEM
    1635             :                          * always gets requeued at the front of the list, to maintain
    1636             :                          * ordering.
    1637             :                          */
    1638           1 :                         break;
    1639             :                 }
    1640             :         }
    1641          63 : }
    1642             : 
    1643             : static void
    1644          63 : bdev_ch_retry_io(struct spdk_bdev_channel *bdev_ch)
    1645             : {
    1646          63 :         bdev_shared_ch_retry_io(bdev_ch->shared_resource);
    1647          63 : }
    1648             : 
    1649             : static int
    1650           0 : bdev_no_mem_poller(void *ctx)
    1651             : {
    1652           0 :         struct spdk_bdev_shared_resource *shared_resource = ctx;
    1653             : 
    1654           0 :         spdk_poller_unregister(&shared_resource->nomem_poller);
    1655             : 
    1656           0 :         if (!TAILQ_EMPTY(&shared_resource->nomem_io)) {
    1657           0 :                 bdev_shared_ch_retry_io(shared_resource);
    1658           0 :         }
    1659             :         /* the retry cb may re-register the poller so double check */
    1660           0 :         if (!TAILQ_EMPTY(&shared_resource->nomem_io) &&
    1661           0 :             shared_resource->io_outstanding == 0 && shared_resource->nomem_poller == NULL) {
    1662             :                 /* No IOs were submitted, try again */
    1663           0 :                 shared_resource->nomem_poller = SPDK_POLLER_REGISTER(bdev_no_mem_poller, shared_resource,
    1664             :                                                 SPDK_BDEV_IO_POLL_INTERVAL_IN_MSEC * 10);
    1665           0 :         }
    1666             : 
    1667           0 :         return SPDK_POLLER_BUSY;
    1668             : }
    1669             : 
    1670             : static inline bool
    1671         556 : _bdev_io_handle_no_mem(struct spdk_bdev_io *bdev_io, enum bdev_io_retry_state state)
    1672             : {
    1673         556 :         struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
    1674         556 :         struct spdk_bdev_shared_resource *shared_resource = bdev_ch->shared_resource;
    1675             : 
    1676         556 :         if (spdk_unlikely(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_NOMEM)) {
    1677           5 :                 bdev_io->internal.status = SPDK_BDEV_IO_STATUS_PENDING;
    1678           5 :                 bdev_queue_nomem_io_head(shared_resource, bdev_io, state);
    1679             : 
    1680           5 :                 if (shared_resource->io_outstanding == 0 && !shared_resource->nomem_poller) {
    1681             :                         /* Special case when we have nomem IOs and no outstanding IOs which completions
    1682             :                          * could trigger retry of queued IOs
    1683             :                          * Any IOs submitted may trigger retry of queued IOs. This poller handles a case when no
    1684             :                          * new IOs submitted, e.g. qd==1 */
    1685           0 :                         shared_resource->nomem_poller = SPDK_POLLER_REGISTER(bdev_no_mem_poller, shared_resource,
    1686             :                                                         SPDK_BDEV_IO_POLL_INTERVAL_IN_MSEC * 10);
    1687           0 :                 }
    1688             :                 /* If bdev module completed an I/O that has an accel sequence with NOMEM status, the
    1689             :                  * ownership of that sequence is transferred back to the bdev layer, so we need to
    1690             :                  * restore internal.accel_sequence to make sure that the sequence is handled
    1691             :                  * correctly in case the I/O is later aborted. */
    1692           5 :                 if ((bdev_io->type == SPDK_BDEV_IO_TYPE_READ ||
    1693           5 :                      bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) && bdev_io->u.bdev.accel_sequence) {
    1694           0 :                         assert(!bdev_io_use_accel_sequence(bdev_io));
    1695           0 :                         bdev_io->internal.f.has_accel_sequence = true;
    1696           0 :                         bdev_io->internal.accel_sequence = bdev_io->u.bdev.accel_sequence;
    1697           0 :                 }
    1698             : 
    1699           5 :                 return true;
    1700             :         }
    1701             : 
    1702         551 :         if (spdk_unlikely(!TAILQ_EMPTY(&shared_resource->nomem_io))) {
    1703          63 :                 bdev_ch_retry_io(bdev_ch);
    1704          63 :         }
    1705             : 
    1706         551 :         return false;
    1707         556 : }
    1708             : 
    1709             : static void
    1710          26 : _bdev_io_complete_push_bounce_done(void *ctx, int rc)
    1711             : {
    1712          26 :         struct spdk_bdev_io *bdev_io = ctx;
    1713          26 :         struct spdk_bdev_channel *ch = bdev_io->internal.ch;
    1714             : 
    1715          26 :         if (rc) {
    1716           0 :                 bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
    1717           0 :         }
    1718             :         /* We want to free the bounce buffer here since we know we're done with it (as opposed
    1719             :          * to waiting for the conditional free of internal.buf.ptr in spdk_bdev_free_io()).
    1720             :          */
    1721          26 :         bdev_io_put_buf(bdev_io);
    1722             : 
    1723          26 :         if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
    1724           0 :                 bdev_ch_retry_io(ch);
    1725           0 :         }
    1726             : 
    1727             :         /* Continue with IO completion flow */
    1728          26 :         bdev_io_complete(bdev_io);
    1729          26 : }
    1730             : 
    1731             : static void
    1732           2 : bdev_io_push_bounce_md_buf_done(void *ctx, int rc)
    1733             : {
    1734           2 :         struct spdk_bdev_io *bdev_io = ctx;
    1735           2 :         struct spdk_bdev_channel *ch = bdev_io->internal.ch;
    1736             : 
    1737           2 :         TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
    1738           2 :         bdev_io_decrement_outstanding(ch, ch->shared_resource);
    1739           2 :         bdev_io->internal.f.has_bounce_buf = false;
    1740             : 
    1741           2 :         if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
    1742           0 :                 bdev_ch_retry_io(ch);
    1743           0 :         }
    1744             : 
    1745           2 :         bdev_io->internal.data_transfer_cpl(bdev_io, rc);
    1746           2 : }
    1747             : 
    1748             : static inline void
    1749          26 : bdev_io_push_bounce_md_buf(struct spdk_bdev_io *bdev_io)
    1750             : {
    1751          26 :         struct spdk_bdev_channel *ch = bdev_io->internal.ch;
    1752          26 :         int rc = 0;
    1753             : 
    1754          26 :         assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS);
    1755          26 :         assert(bdev_io->internal.f.has_bounce_buf);
    1756             : 
    1757             :         /* do the same for metadata buffer */
    1758          26 :         if (spdk_unlikely(bdev_io->internal.bounce_buf.orig_md_iov.iov_base != NULL)) {
    1759           4 :                 assert(spdk_bdev_is_md_separate(bdev_io->bdev));
    1760             : 
    1761           4 :                 if (bdev_io->type == SPDK_BDEV_IO_TYPE_READ) {
    1762           2 :                         if (bdev_io_use_memory_domain(bdev_io)) {
    1763           2 :                                 TAILQ_INSERT_TAIL(&ch->io_memory_domain, bdev_io, internal.link);
    1764           2 :                                 bdev_io_increment_outstanding(ch, ch->shared_resource);
    1765             :                                 /* If memory domain is used then we need to call async push function */
    1766           4 :                                 rc = spdk_memory_domain_push_data(bdev_io->internal.memory_domain,
    1767           2 :                                                                   bdev_io->internal.memory_domain_ctx,
    1768           2 :                                                                   &bdev_io->internal.bounce_buf.orig_md_iov,
    1769           2 :                                                                   (uint32_t)bdev_io->internal.bounce_buf.orig_iovcnt,
    1770           2 :                                                                   &bdev_io->internal.bounce_buf.md_iov, 1,
    1771             :                                                                   bdev_io_push_bounce_md_buf_done,
    1772           2 :                                                                   bdev_io);
    1773           2 :                                 if (rc == 0) {
    1774             :                                         /* Continue IO completion in async callback */
    1775           2 :                                         return;
    1776             :                                 }
    1777           0 :                                 TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
    1778           0 :                                 bdev_io_decrement_outstanding(ch, ch->shared_resource);
    1779           0 :                                 if (rc != -ENOMEM) {
    1780           0 :                                         SPDK_ERRLOG("Failed to push md to memory domain %s\n",
    1781             :                                                     spdk_memory_domain_get_dma_device_id(
    1782             :                                                             bdev_io->internal.memory_domain));
    1783           0 :                                 }
    1784           0 :                         } else {
    1785           0 :                                 memcpy(bdev_io->internal.bounce_buf.orig_md_iov.iov_base, bdev_io->u.bdev.md_buf,
    1786           0 :                                        bdev_io->internal.bounce_buf.orig_md_iov.iov_len);
    1787             :                         }
    1788           0 :                 }
    1789           2 :         }
    1790             : 
    1791          24 :         if (spdk_unlikely(rc == -ENOMEM)) {
    1792           0 :                 bdev_queue_nomem_io_head(ch->shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PUSH_MD);
    1793           0 :         } else {
    1794          24 :                 assert(bdev_io->internal.data_transfer_cpl);
    1795          24 :                 bdev_io->internal.f.has_bounce_buf = false;
    1796          24 :                 bdev_io->internal.data_transfer_cpl(bdev_io, rc);
    1797             :         }
    1798          26 : }
    1799             : 
    1800             : static inline void
    1801          26 : bdev_io_push_bounce_data_done(struct spdk_bdev_io *bdev_io, int rc)
    1802             : {
    1803          26 :         assert(bdev_io->internal.data_transfer_cpl);
    1804          26 :         if (rc) {
    1805           0 :                 bdev_io->internal.data_transfer_cpl(bdev_io, rc);
    1806           0 :                 return;
    1807             :         }
    1808             : 
    1809             :         /* set original buffer for this io */
    1810          26 :         bdev_io->u.bdev.iovcnt = bdev_io->internal.bounce_buf.orig_iovcnt;
    1811          26 :         bdev_io->u.bdev.iovs = bdev_io->internal.bounce_buf.orig_iovs;
    1812             : 
    1813             :         /* We don't set bdev_io->internal.f.has_bounce_buf to false here because
    1814             :          * we still need to clear the md buf */
    1815             : 
    1816          26 :         bdev_io_push_bounce_md_buf(bdev_io);
    1817          26 : }
    1818             : 
    1819             : static void
    1820           2 : bdev_io_push_bounce_data_done_and_track(void *ctx, int status)
    1821             : {
    1822           2 :         struct spdk_bdev_io *bdev_io = ctx;
    1823           2 :         struct spdk_bdev_channel *ch = bdev_io->internal.ch;
    1824             : 
    1825           2 :         TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
    1826           2 :         bdev_io_decrement_outstanding(ch, ch->shared_resource);
    1827             : 
    1828           2 :         if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
    1829           0 :                 bdev_ch_retry_io(ch);
    1830           0 :         }
    1831             : 
    1832           2 :         bdev_io_push_bounce_data_done(bdev_io, status);
    1833           2 : }
    1834             : 
    1835             : static inline void
    1836          27 : bdev_io_push_bounce_data(struct spdk_bdev_io *bdev_io)
    1837             : {
    1838          27 :         struct spdk_bdev_channel *ch = bdev_io->internal.ch;
    1839          27 :         int rc = 0;
    1840             : 
    1841          27 :         assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS);
    1842          27 :         assert(!bdev_io_use_accel_sequence(bdev_io));
    1843          27 :         assert(bdev_io->internal.f.has_bounce_buf);
    1844             : 
    1845             :         /* if this is read path, copy data from bounce buffer to original buffer */
    1846          27 :         if (bdev_io->type == SPDK_BDEV_IO_TYPE_READ) {
    1847          11 :                 if (bdev_io_use_memory_domain(bdev_io)) {
    1848           3 :                         TAILQ_INSERT_TAIL(&ch->io_memory_domain, bdev_io, internal.link);
    1849           3 :                         bdev_io_increment_outstanding(ch, ch->shared_resource);
    1850             :                         /* If memory domain is used then we need to call async push function */
    1851           6 :                         rc = spdk_memory_domain_push_data(bdev_io->internal.memory_domain,
    1852           3 :                                                           bdev_io->internal.memory_domain_ctx,
    1853           3 :                                                           bdev_io->internal.bounce_buf.orig_iovs,
    1854           3 :                                                           (uint32_t)bdev_io->internal.bounce_buf.orig_iovcnt,
    1855           3 :                                                           &bdev_io->internal.bounce_buf.iov, 1,
    1856             :                                                           bdev_io_push_bounce_data_done_and_track,
    1857           3 :                                                           bdev_io);
    1858           3 :                         if (rc == 0) {
    1859             :                                 /* Continue IO completion in async callback */
    1860           2 :                                 return;
    1861             :                         }
    1862             : 
    1863           1 :                         TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
    1864           1 :                         bdev_io_decrement_outstanding(ch, ch->shared_resource);
    1865           1 :                         if (rc != -ENOMEM) {
    1866           0 :                                 SPDK_ERRLOG("Failed to push data to memory domain %s\n",
    1867             :                                             spdk_memory_domain_get_dma_device_id(
    1868             :                                                     bdev_io->internal.memory_domain));
    1869           0 :                         }
    1870           1 :                 } else {
    1871          16 :                         spdk_copy_buf_to_iovs(bdev_io->internal.bounce_buf.orig_iovs,
    1872           8 :                                               bdev_io->internal.bounce_buf.orig_iovcnt,
    1873           8 :                                               bdev_io->internal.bounce_buf.iov.iov_base,
    1874           8 :                                               bdev_io->internal.bounce_buf.iov.iov_len);
    1875             :                 }
    1876           9 :         }
    1877             : 
    1878          25 :         if (spdk_unlikely(rc == -ENOMEM)) {
    1879           1 :                 bdev_queue_nomem_io_head(ch->shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PUSH);
    1880           1 :         } else {
    1881          24 :                 bdev_io_push_bounce_data_done(bdev_io, rc);
    1882             :         }
    1883          27 : }
    1884             : 
    1885             : static inline void
    1886          26 : _bdev_io_push_bounce_data_buffer(struct spdk_bdev_io *bdev_io, bdev_copy_bounce_buffer_cpl cpl_cb)
    1887             : {
    1888          26 :         bdev_io->internal.data_transfer_cpl = cpl_cb;
    1889          26 :         bdev_io_push_bounce_data(bdev_io);
    1890          26 : }
    1891             : 
    1892             : static void
    1893           0 : bdev_io_get_iobuf_cb(struct spdk_iobuf_entry *iobuf, void *buf)
    1894             : {
    1895             :         struct spdk_bdev_io *bdev_io;
    1896             : 
    1897           0 :         bdev_io = SPDK_CONTAINEROF(iobuf, struct spdk_bdev_io, internal.iobuf);
    1898           0 :         _bdev_io_set_buf(bdev_io, buf, bdev_io->internal.buf.len);
    1899           0 : }
    1900             : 
    1901             : static void
    1902          42 : bdev_io_get_buf(struct spdk_bdev_io *bdev_io, uint64_t len)
    1903             : {
    1904             :         struct spdk_bdev_mgmt_channel *mgmt_ch;
    1905             :         uint64_t max_len;
    1906             :         void *buf;
    1907             : 
    1908          42 :         assert(spdk_bdev_io_get_thread(bdev_io) == spdk_get_thread());
    1909          42 :         mgmt_ch = bdev_io->internal.ch->shared_resource->mgmt_ch;
    1910          42 :         max_len = bdev_io_get_max_buf_len(bdev_io, len);
    1911             : 
    1912          42 :         if (spdk_unlikely(max_len > mgmt_ch->iobuf.cache[0].large.bufsize)) {
    1913           0 :                 SPDK_ERRLOG("Length %" PRIu64 " is larger than allowed\n", max_len);
    1914           0 :                 bdev_io_get_buf_complete(bdev_io, false);
    1915           0 :                 return;
    1916             :         }
    1917             : 
    1918          42 :         bdev_io->internal.buf.len = len;
    1919          42 :         buf = spdk_iobuf_get(&mgmt_ch->iobuf, max_len, &bdev_io->internal.iobuf,
    1920             :                              bdev_io_get_iobuf_cb);
    1921          42 :         if (buf != NULL) {
    1922          42 :                 _bdev_io_set_buf(bdev_io, buf, len);
    1923          42 :         }
    1924          42 : }
    1925             : 
    1926             : void
    1927          56 : spdk_bdev_io_get_buf(struct spdk_bdev_io *bdev_io, spdk_bdev_io_get_buf_cb cb, uint64_t len)
    1928             : {
    1929          56 :         struct spdk_bdev *bdev = bdev_io->bdev;
    1930             :         uint64_t alignment;
    1931             : 
    1932          56 :         assert(cb != NULL);
    1933          56 :         bdev_io->internal.get_buf_cb = cb;
    1934             : 
    1935          56 :         alignment = spdk_bdev_get_buf_align(bdev);
    1936             : 
    1937          56 :         if (_is_buf_allocated(bdev_io->u.bdev.iovs) &&
    1938          40 :             _are_iovs_aligned(bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt, alignment)) {
    1939             :                 /* Buffer already present and aligned */
    1940          18 :                 cb(spdk_bdev_io_get_io_channel(bdev_io), bdev_io, true);
    1941          18 :                 return;
    1942             :         }
    1943             : 
    1944          38 :         bdev_io_get_buf(bdev_io, len);
    1945          56 : }
    1946             : 
    1947             : static void
    1948           4 : _bdev_io_get_bounce_buf(struct spdk_bdev_io *bdev_io, spdk_bdev_io_get_buf_cb cb,
    1949             :                         uint64_t len)
    1950             : {
    1951           4 :         assert(cb != NULL);
    1952           4 :         bdev_io->internal.get_buf_cb = cb;
    1953             : 
    1954           4 :         bdev_io_get_buf(bdev_io, len);
    1955           4 : }
    1956             : 
    1957             : static void
    1958           0 : _bdev_io_get_accel_buf(struct spdk_bdev_io *bdev_io)
    1959             : {
    1960           0 :         struct spdk_bdev_channel *ch = bdev_io->internal.ch;
    1961             :         void *buf;
    1962             :         int rc;
    1963             : 
    1964           0 :         rc = spdk_accel_get_buf(ch->accel_channel,
    1965           0 :                                 bdev_io->internal.buf.len,
    1966             :                                 &buf,
    1967           0 :                                 &bdev_io->u.bdev.memory_domain,
    1968           0 :                                 &bdev_io->u.bdev.memory_domain_ctx);
    1969           0 :         if (rc != 0) {
    1970           0 :                 bdev_queue_nomem_io_tail(ch->shared_resource, bdev_io,
    1971             :                                          BDEV_IO_RETRY_STATE_GET_ACCEL_BUF);
    1972           0 :                 return;
    1973             :         }
    1974             : 
    1975           0 :         _bdev_io_set_buf(bdev_io, buf, bdev_io->internal.buf.len);
    1976           0 : }
    1977             : 
    1978             : static inline void
    1979           0 : bdev_io_get_accel_buf(struct spdk_bdev_io *bdev_io, spdk_bdev_io_get_buf_cb cb,
    1980             :                       uint64_t len)
    1981             : {
    1982           0 :         bdev_io->internal.buf.len = len;
    1983           0 :         bdev_io->internal.get_buf_cb = cb;
    1984             : 
    1985           0 :         _bdev_io_get_accel_buf(bdev_io);
    1986           0 : }
    1987             : 
    1988           3 : SPDK_LOG_DEPRECATION_REGISTER(spdk_bdev_io_get_aux_buf,
    1989             :                               "spdk_bdev_io_get_aux_buf is deprecated", "v25.01", 0);
    1990             : 
    1991             : void
    1992           0 : spdk_bdev_io_get_aux_buf(struct spdk_bdev_io *bdev_io, spdk_bdev_io_get_aux_buf_cb cb)
    1993             : {
    1994           0 :         uint64_t len = bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen;
    1995             : 
    1996           0 :         SPDK_LOG_DEPRECATED(spdk_bdev_io_get_aux_buf);
    1997             : 
    1998           0 :         assert(cb != NULL);
    1999           0 :         assert(bdev_io->internal.get_aux_buf_cb == NULL);
    2000           0 :         bdev_io->internal.get_aux_buf_cb = cb;
    2001           0 :         bdev_io_get_buf(bdev_io, len);
    2002           0 : }
    2003             : 
    2004             : static int
    2005          65 : bdev_module_get_max_ctx_size(void)
    2006             : {
    2007             :         struct spdk_bdev_module *bdev_module;
    2008          65 :         int max_bdev_module_size = 0;
    2009             : 
    2010         254 :         TAILQ_FOREACH(bdev_module, &g_bdev_mgr.bdev_modules, internal.tailq) {
    2011         189 :                 if (bdev_module->get_ctx_size && bdev_module->get_ctx_size() > max_bdev_module_size) {
    2012          64 :                         max_bdev_module_size = bdev_module->get_ctx_size();
    2013          64 :                 }
    2014         189 :         }
    2015             : 
    2016          65 :         return max_bdev_module_size;
    2017             : }
    2018             : 
    2019             : static void
    2020           0 : bdev_enable_histogram_config_json(struct spdk_bdev *bdev, struct spdk_json_write_ctx *w)
    2021             : {
    2022           0 :         if (!bdev->internal.histogram_enabled) {
    2023           0 :                 return;
    2024             :         }
    2025             : 
    2026           0 :         spdk_json_write_object_begin(w);
    2027           0 :         spdk_json_write_named_string(w, "method", "bdev_enable_histogram");
    2028             : 
    2029           0 :         spdk_json_write_named_object_begin(w, "params");
    2030           0 :         spdk_json_write_named_string(w, "name", bdev->name);
    2031             : 
    2032           0 :         spdk_json_write_named_bool(w, "enable", bdev->internal.histogram_enabled);
    2033             : 
    2034           0 :         if (bdev->internal.histogram_io_type) {
    2035           0 :                 spdk_json_write_named_string(w, "opc",
    2036           0 :                                              spdk_bdev_get_io_type_name(bdev->internal.histogram_io_type));
    2037           0 :         }
    2038             : 
    2039           0 :         spdk_json_write_object_end(w);
    2040             : 
    2041           0 :         spdk_json_write_object_end(w);
    2042           0 : }
    2043             : 
    2044             : static void
    2045           0 : bdev_qos_config_json(struct spdk_bdev *bdev, struct spdk_json_write_ctx *w)
    2046             : {
    2047             :         int i;
    2048           0 :         struct spdk_bdev_qos *qos = bdev->internal.qos;
    2049             :         uint64_t limits[SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES];
    2050             : 
    2051           0 :         if (!qos) {
    2052           0 :                 return;
    2053             :         }
    2054             : 
    2055           0 :         spdk_bdev_get_qos_rate_limits(bdev, limits);
    2056             : 
    2057           0 :         spdk_json_write_object_begin(w);
    2058           0 :         spdk_json_write_named_string(w, "method", "bdev_set_qos_limit");
    2059             : 
    2060           0 :         spdk_json_write_named_object_begin(w, "params");
    2061           0 :         spdk_json_write_named_string(w, "name", bdev->name);
    2062           0 :         for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
    2063           0 :                 if (limits[i] > 0) {
    2064           0 :                         spdk_json_write_named_uint64(w, qos_rpc_type[i], limits[i]);
    2065           0 :                 }
    2066           0 :         }
    2067           0 :         spdk_json_write_object_end(w);
    2068             : 
    2069           0 :         spdk_json_write_object_end(w);
    2070           0 : }
    2071             : 
    2072             : void
    2073           0 : spdk_bdev_subsystem_config_json(struct spdk_json_write_ctx *w)
    2074             : {
    2075             :         struct spdk_bdev_module *bdev_module;
    2076             :         struct spdk_bdev *bdev;
    2077             : 
    2078           0 :         assert(w != NULL);
    2079             : 
    2080           0 :         spdk_json_write_array_begin(w);
    2081             : 
    2082           0 :         spdk_json_write_object_begin(w);
    2083           0 :         spdk_json_write_named_string(w, "method", "bdev_set_options");
    2084           0 :         spdk_json_write_named_object_begin(w, "params");
    2085           0 :         spdk_json_write_named_uint32(w, "bdev_io_pool_size", g_bdev_opts.bdev_io_pool_size);
    2086           0 :         spdk_json_write_named_uint32(w, "bdev_io_cache_size", g_bdev_opts.bdev_io_cache_size);
    2087           0 :         spdk_json_write_named_bool(w, "bdev_auto_examine", g_bdev_opts.bdev_auto_examine);
    2088           0 :         spdk_json_write_named_uint32(w, "iobuf_small_cache_size", g_bdev_opts.iobuf_small_cache_size);
    2089           0 :         spdk_json_write_named_uint32(w, "iobuf_large_cache_size", g_bdev_opts.iobuf_large_cache_size);
    2090           0 :         spdk_json_write_object_end(w);
    2091           0 :         spdk_json_write_object_end(w);
    2092             : 
    2093           0 :         bdev_examine_allowlist_config_json(w);
    2094             : 
    2095           0 :         TAILQ_FOREACH(bdev_module, &g_bdev_mgr.bdev_modules, internal.tailq) {
    2096           0 :                 if (bdev_module->config_json) {
    2097           0 :                         bdev_module->config_json(w);
    2098           0 :                 }
    2099           0 :         }
    2100             : 
    2101           0 :         spdk_spin_lock(&g_bdev_mgr.spinlock);
    2102             : 
    2103           0 :         TAILQ_FOREACH(bdev, &g_bdev_mgr.bdevs, internal.link) {
    2104           0 :                 if (bdev->fn_table->write_config_json) {
    2105           0 :                         bdev->fn_table->write_config_json(bdev, w);
    2106           0 :                 }
    2107             : 
    2108           0 :                 bdev_qos_config_json(bdev, w);
    2109           0 :                 bdev_enable_histogram_config_json(bdev, w);
    2110           0 :         }
    2111             : 
    2112           0 :         spdk_spin_unlock(&g_bdev_mgr.spinlock);
    2113             : 
    2114             :         /* This has to be last RPC in array to make sure all bdevs finished examine */
    2115           0 :         spdk_json_write_object_begin(w);
    2116           0 :         spdk_json_write_named_string(w, "method", "bdev_wait_for_examine");
    2117           0 :         spdk_json_write_object_end(w);
    2118             : 
    2119           0 :         spdk_json_write_array_end(w);
    2120           0 : }
    2121             : 
    2122             : static void
    2123          71 : bdev_mgmt_channel_destroy(void *io_device, void *ctx_buf)
    2124             : {
    2125          71 :         struct spdk_bdev_mgmt_channel *ch = ctx_buf;
    2126             :         struct spdk_bdev_io *bdev_io;
    2127             : 
    2128          71 :         spdk_iobuf_channel_fini(&ch->iobuf);
    2129             : 
    2130        9969 :         while (!STAILQ_EMPTY(&ch->per_thread_cache)) {
    2131        9898 :                 bdev_io = STAILQ_FIRST(&ch->per_thread_cache);
    2132        9898 :                 STAILQ_REMOVE_HEAD(&ch->per_thread_cache, internal.buf_link);
    2133        9898 :                 ch->per_thread_cache_count--;
    2134        9898 :                 spdk_mempool_put(g_bdev_mgr.bdev_io_pool, (void *)bdev_io);
    2135             :         }
    2136             : 
    2137          71 :         assert(ch->per_thread_cache_count == 0);
    2138          71 : }
    2139             : 
    2140             : static int
    2141          71 : bdev_mgmt_channel_create(void *io_device, void *ctx_buf)
    2142             : {
    2143          71 :         struct spdk_bdev_mgmt_channel *ch = ctx_buf;
    2144             :         struct spdk_bdev_io *bdev_io;
    2145             :         uint32_t i;
    2146             :         int rc;
    2147             : 
    2148         142 :         rc = spdk_iobuf_channel_init(&ch->iobuf, "bdev",
    2149          71 :                                      g_bdev_opts.iobuf_small_cache_size,
    2150          71 :                                      g_bdev_opts.iobuf_large_cache_size);
    2151          71 :         if (rc != 0) {
    2152           0 :                 SPDK_ERRLOG("Failed to create iobuf channel: %s\n", spdk_strerror(-rc));
    2153           0 :                 return -1;
    2154             :         }
    2155             : 
    2156          71 :         STAILQ_INIT(&ch->per_thread_cache);
    2157          71 :         ch->bdev_io_cache_size = g_bdev_opts.bdev_io_cache_size;
    2158             : 
    2159             :         /* Pre-populate bdev_io cache to ensure this thread cannot be starved. */
    2160          71 :         ch->per_thread_cache_count = 0;
    2161        9969 :         for (i = 0; i < ch->bdev_io_cache_size; i++) {
    2162        9898 :                 bdev_io = spdk_mempool_get(g_bdev_mgr.bdev_io_pool);
    2163        9898 :                 if (bdev_io == NULL) {
    2164           0 :                         SPDK_ERRLOG("You need to increase bdev_io_pool_size using bdev_set_options RPC.\n");
    2165           0 :                         assert(false);
    2166             :                         bdev_mgmt_channel_destroy(io_device, ctx_buf);
    2167             :                         return -1;
    2168             :                 }
    2169        9898 :                 ch->per_thread_cache_count++;
    2170        9898 :                 STAILQ_INSERT_HEAD(&ch->per_thread_cache, bdev_io, internal.buf_link);
    2171        9898 :         }
    2172             : 
    2173          71 :         TAILQ_INIT(&ch->shared_resources);
    2174          71 :         TAILQ_INIT(&ch->io_wait_queue);
    2175             : 
    2176          71 :         return 0;
    2177          71 : }
    2178             : 
    2179             : static void
    2180          65 : bdev_init_complete(int rc)
    2181             : {
    2182          65 :         spdk_bdev_init_cb cb_fn = g_init_cb_fn;
    2183          65 :         void *cb_arg = g_init_cb_arg;
    2184             :         struct spdk_bdev_module *m;
    2185             : 
    2186          65 :         g_bdev_mgr.init_complete = true;
    2187          65 :         g_init_cb_fn = NULL;
    2188          65 :         g_init_cb_arg = NULL;
    2189             : 
    2190             :         /*
    2191             :          * For modules that need to know when subsystem init is complete,
    2192             :          * inform them now.
    2193             :          */
    2194          65 :         if (rc == 0) {
    2195         254 :                 TAILQ_FOREACH(m, &g_bdev_mgr.bdev_modules, internal.tailq) {
    2196         189 :                         if (m->init_complete) {
    2197          23 :                                 m->init_complete();
    2198          23 :                         }
    2199         189 :                 }
    2200          65 :         }
    2201             : 
    2202          65 :         cb_fn(cb_arg, rc);
    2203          65 : }
    2204             : 
    2205             : static bool
    2206         258 : bdev_module_all_actions_completed(void)
    2207             : {
    2208             :         struct spdk_bdev_module *m;
    2209             : 
    2210        1022 :         TAILQ_FOREACH(m, &g_bdev_mgr.bdev_modules, internal.tailq) {
    2211         764 :                 if (m->internal.action_in_progress > 0) {
    2212           0 :                         return false;
    2213             :                 }
    2214         764 :         }
    2215         258 :         return true;
    2216         258 : }
    2217             : 
    2218             : static void
    2219         628 : bdev_module_action_complete(void)
    2220             : {
    2221             :         /*
    2222             :          * Don't finish bdev subsystem initialization if
    2223             :          * module pre-initialization is still in progress, or
    2224             :          * the subsystem been already initialized.
    2225             :          */
    2226         628 :         if (!g_bdev_mgr.module_init_complete || g_bdev_mgr.init_complete) {
    2227         563 :                 return;
    2228             :         }
    2229             : 
    2230             :         /*
    2231             :          * Check all bdev modules for inits/examinations in progress. If any
    2232             :          * exist, return immediately since we cannot finish bdev subsystem
    2233             :          * initialization until all are completed.
    2234             :          */
    2235          65 :         if (!bdev_module_all_actions_completed()) {
    2236           0 :                 return;
    2237             :         }
    2238             : 
    2239             :         /*
    2240             :          * Modules already finished initialization - now that all
    2241             :          * the bdev modules have finished their asynchronous I/O
    2242             :          * processing, the entire bdev layer can be marked as complete.
    2243             :          */
    2244          65 :         bdev_init_complete(0);
    2245         628 : }
    2246             : 
    2247             : static void
    2248         563 : bdev_module_action_done(struct spdk_bdev_module *module)
    2249             : {
    2250         563 :         spdk_spin_lock(&module->internal.spinlock);
    2251         563 :         assert(module->internal.action_in_progress > 0);
    2252         563 :         module->internal.action_in_progress--;
    2253         563 :         spdk_spin_unlock(&module->internal.spinlock);
    2254         563 :         bdev_module_action_complete();
    2255         563 : }
    2256             : 
    2257             : void
    2258          65 : spdk_bdev_module_init_done(struct spdk_bdev_module *module)
    2259             : {
    2260          65 :         assert(module->async_init);
    2261          65 :         bdev_module_action_done(module);
    2262          65 : }
    2263             : 
    2264             : void
    2265         498 : spdk_bdev_module_examine_done(struct spdk_bdev_module *module)
    2266             : {
    2267         498 :         bdev_module_action_done(module);
    2268         498 : }
    2269             : 
    2270             : /** The last initialized bdev module */
    2271             : static struct spdk_bdev_module *g_resume_bdev_module = NULL;
    2272             : 
    2273             : static void
    2274           0 : bdev_init_failed(void *cb_arg)
    2275             : {
    2276           0 :         struct spdk_bdev_module *module = cb_arg;
    2277             : 
    2278           0 :         spdk_spin_lock(&module->internal.spinlock);
    2279           0 :         assert(module->internal.action_in_progress > 0);
    2280           0 :         module->internal.action_in_progress--;
    2281           0 :         spdk_spin_unlock(&module->internal.spinlock);
    2282           0 :         bdev_init_complete(-1);
    2283           0 : }
    2284             : 
    2285             : static int
    2286          65 : bdev_modules_init(void)
    2287             : {
    2288             :         struct spdk_bdev_module *module;
    2289          65 :         int rc = 0;
    2290             : 
    2291         254 :         TAILQ_FOREACH(module, &g_bdev_mgr.bdev_modules, internal.tailq) {
    2292         189 :                 g_resume_bdev_module = module;
    2293         189 :                 if (module->async_init) {
    2294          65 :                         spdk_spin_lock(&module->internal.spinlock);
    2295          65 :                         module->internal.action_in_progress = 1;
    2296          65 :                         spdk_spin_unlock(&module->internal.spinlock);
    2297          65 :                 }
    2298         189 :                 rc = module->module_init();
    2299         189 :                 if (rc != 0) {
    2300             :                         /* Bump action_in_progress to prevent other modules from completion of modules_init
    2301             :                          * Send message to defer application shutdown until resources are cleaned up */
    2302           0 :                         spdk_spin_lock(&module->internal.spinlock);
    2303           0 :                         module->internal.action_in_progress = 1;
    2304           0 :                         spdk_spin_unlock(&module->internal.spinlock);
    2305           0 :                         spdk_thread_send_msg(spdk_get_thread(), bdev_init_failed, module);
    2306           0 :                         return rc;
    2307             :                 }
    2308         189 :         }
    2309             : 
    2310          65 :         g_resume_bdev_module = NULL;
    2311          65 :         return 0;
    2312          65 : }
    2313             : 
    2314             : void
    2315          65 : spdk_bdev_initialize(spdk_bdev_init_cb cb_fn, void *cb_arg)
    2316             : {
    2317          65 :         int rc = 0;
    2318             :         char mempool_name[32];
    2319             : 
    2320          65 :         assert(cb_fn != NULL);
    2321             : 
    2322          65 :         g_init_cb_fn = cb_fn;
    2323          65 :         g_init_cb_arg = cb_arg;
    2324             : 
    2325          65 :         spdk_notify_type_register("bdev_register");
    2326          65 :         spdk_notify_type_register("bdev_unregister");
    2327             : 
    2328          65 :         snprintf(mempool_name, sizeof(mempool_name), "bdev_io_%d", getpid());
    2329             : 
    2330          65 :         rc = spdk_iobuf_register_module("bdev");
    2331          65 :         if (rc != 0) {
    2332           0 :                 SPDK_ERRLOG("could not register bdev iobuf module: %s\n", spdk_strerror(-rc));
    2333           0 :                 bdev_init_complete(-1);
    2334           0 :                 return;
    2335             :         }
    2336             : 
    2337         130 :         g_bdev_mgr.bdev_io_pool = spdk_mempool_create(mempool_name,
    2338          65 :                                   g_bdev_opts.bdev_io_pool_size,
    2339          65 :                                   sizeof(struct spdk_bdev_io) +
    2340          65 :                                   bdev_module_get_max_ctx_size(),
    2341             :                                   0,
    2342             :                                   SPDK_ENV_NUMA_ID_ANY);
    2343             : 
    2344          65 :         if (g_bdev_mgr.bdev_io_pool == NULL) {
    2345           0 :                 SPDK_ERRLOG("could not allocate spdk_bdev_io pool\n");
    2346           0 :                 bdev_init_complete(-1);
    2347           0 :                 return;
    2348             :         }
    2349             : 
    2350          65 :         g_bdev_mgr.zero_buffer = spdk_zmalloc(ZERO_BUFFER_SIZE, ZERO_BUFFER_SIZE,
    2351             :                                               NULL, SPDK_ENV_LCORE_ID_ANY, SPDK_MALLOC_DMA);
    2352          65 :         if (!g_bdev_mgr.zero_buffer) {
    2353           0 :                 SPDK_ERRLOG("create bdev zero buffer failed\n");
    2354           0 :                 bdev_init_complete(-1);
    2355           0 :                 return;
    2356             :         }
    2357             : 
    2358             : #ifdef SPDK_CONFIG_VTUNE
    2359             :         g_bdev_mgr.domain = __itt_domain_create("spdk_bdev");
    2360             : #endif
    2361             : 
    2362          65 :         spdk_io_device_register(&g_bdev_mgr, bdev_mgmt_channel_create,
    2363             :                                 bdev_mgmt_channel_destroy,
    2364             :                                 sizeof(struct spdk_bdev_mgmt_channel),
    2365             :                                 "bdev_mgr");
    2366             : 
    2367          65 :         rc = bdev_modules_init();
    2368          65 :         g_bdev_mgr.module_init_complete = true;
    2369          65 :         if (rc != 0) {
    2370           0 :                 SPDK_ERRLOG("bdev modules init failed\n");
    2371           0 :                 return;
    2372             :         }
    2373             : 
    2374          65 :         bdev_module_action_complete();
    2375          65 : }
    2376             : 
    2377             : static void
    2378          65 : bdev_mgr_unregister_cb(void *io_device)
    2379             : {
    2380          65 :         spdk_bdev_fini_cb cb_fn = g_fini_cb_fn;
    2381             : 
    2382          65 :         if (g_bdev_mgr.bdev_io_pool) {
    2383          65 :                 if (spdk_mempool_count(g_bdev_mgr.bdev_io_pool) != g_bdev_opts.bdev_io_pool_size) {
    2384           0 :                         SPDK_ERRLOG("bdev IO pool count is %zu but should be %u\n",
    2385             :                                     spdk_mempool_count(g_bdev_mgr.bdev_io_pool),
    2386             :                                     g_bdev_opts.bdev_io_pool_size);
    2387           0 :                 }
    2388             : 
    2389          65 :                 spdk_mempool_free(g_bdev_mgr.bdev_io_pool);
    2390          65 :         }
    2391             : 
    2392          65 :         spdk_free(g_bdev_mgr.zero_buffer);
    2393             : 
    2394          65 :         bdev_examine_allowlist_free();
    2395             : 
    2396          65 :         cb_fn(g_fini_cb_arg);
    2397          65 :         g_fini_cb_fn = NULL;
    2398          65 :         g_fini_cb_arg = NULL;
    2399          65 :         g_bdev_mgr.init_complete = false;
    2400          65 :         g_bdev_mgr.module_init_complete = false;
    2401          65 : }
    2402             : 
    2403             : static void
    2404          65 : bdev_module_fini_iter(void *arg)
    2405             : {
    2406             :         struct spdk_bdev_module *bdev_module;
    2407             : 
    2408             :         /* FIXME: Handling initialization failures is broken now,
    2409             :          * so we won't even try cleaning up after successfully
    2410             :          * initialized modules. if module_init_complete is false,
    2411             :          * just call spdk_bdev_mgr_unregister_cb
    2412             :          */
    2413          65 :         if (!g_bdev_mgr.module_init_complete) {
    2414           0 :                 bdev_mgr_unregister_cb(NULL);
    2415           0 :                 return;
    2416             :         }
    2417             : 
    2418             :         /* Start iterating from the last touched module */
    2419          65 :         if (!g_resume_bdev_module) {
    2420          65 :                 bdev_module = TAILQ_LAST(&g_bdev_mgr.bdev_modules, bdev_module_list);
    2421          65 :         } else {
    2422           0 :                 bdev_module = TAILQ_PREV(g_resume_bdev_module, bdev_module_list,
    2423             :                                          internal.tailq);
    2424             :         }
    2425             : 
    2426         254 :         while (bdev_module) {
    2427         189 :                 if (bdev_module->async_fini) {
    2428             :                         /* Save our place so we can resume later. We must
    2429             :                          * save the variable here, before calling module_fini()
    2430             :                          * below, because in some cases the module may immediately
    2431             :                          * call spdk_bdev_module_fini_done() and re-enter
    2432             :                          * this function to continue iterating. */
    2433           0 :                         g_resume_bdev_module = bdev_module;
    2434           0 :                 }
    2435             : 
    2436         189 :                 if (bdev_module->module_fini) {
    2437         189 :                         bdev_module->module_fini();
    2438         189 :                 }
    2439             : 
    2440         189 :                 if (bdev_module->async_fini) {
    2441           0 :                         return;
    2442             :                 }
    2443             : 
    2444         189 :                 bdev_module = TAILQ_PREV(bdev_module, bdev_module_list,
    2445             :                                          internal.tailq);
    2446             :         }
    2447             : 
    2448          65 :         g_resume_bdev_module = NULL;
    2449          65 :         spdk_io_device_unregister(&g_bdev_mgr, bdev_mgr_unregister_cb);
    2450          65 : }
    2451             : 
    2452             : void
    2453           0 : spdk_bdev_module_fini_done(void)
    2454             : {
    2455           0 :         if (spdk_get_thread() != g_fini_thread) {
    2456           0 :                 spdk_thread_send_msg(g_fini_thread, bdev_module_fini_iter, NULL);
    2457           0 :         } else {
    2458           0 :                 bdev_module_fini_iter(NULL);
    2459             :         }
    2460           0 : }
    2461             : 
    2462             : static void
    2463          65 : bdev_finish_unregister_bdevs_iter(void *cb_arg, int bdeverrno)
    2464             : {
    2465          65 :         struct spdk_bdev *bdev = cb_arg;
    2466             : 
    2467          65 :         if (bdeverrno && bdev) {
    2468           0 :                 SPDK_WARNLOG("Unable to unregister bdev '%s' during spdk_bdev_finish()\n",
    2469             :                              bdev->name);
    2470             : 
    2471             :                 /*
    2472             :                  * Since the call to spdk_bdev_unregister() failed, we have no way to free this
    2473             :                  *  bdev; try to continue by manually removing this bdev from the list and continue
    2474             :                  *  with the next bdev in the list.
    2475             :                  */
    2476           0 :                 TAILQ_REMOVE(&g_bdev_mgr.bdevs, bdev, internal.link);
    2477           0 :         }
    2478             : 
    2479          65 :         if (TAILQ_EMPTY(&g_bdev_mgr.bdevs)) {
    2480          65 :                 SPDK_DEBUGLOG(bdev, "Done unregistering bdevs\n");
    2481             :                 /*
    2482             :                  * Bdev module finish need to be deferred as we might be in the middle of some context
    2483             :                  * (like bdev part free) that will use this bdev (or private bdev driver ctx data)
    2484             :                  * after returning.
    2485             :                  */
    2486          65 :                 spdk_thread_send_msg(spdk_get_thread(), bdev_module_fini_iter, NULL);
    2487          65 :                 return;
    2488             :         }
    2489             : 
    2490             :         /*
    2491             :          * Unregister last unclaimed bdev in the list, to ensure that bdev subsystem
    2492             :          * shutdown proceeds top-down. The goal is to give virtual bdevs an opportunity
    2493             :          * to detect clean shutdown as opposed to run-time hot removal of the underlying
    2494             :          * base bdevs.
    2495             :          *
    2496             :          * Also, walk the list in the reverse order.
    2497             :          */
    2498           0 :         for (bdev = TAILQ_LAST(&g_bdev_mgr.bdevs, spdk_bdev_list);
    2499           0 :              bdev; bdev = TAILQ_PREV(bdev, spdk_bdev_list, internal.link)) {
    2500           0 :                 spdk_spin_lock(&bdev->internal.spinlock);
    2501           0 :                 if (bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE) {
    2502           0 :                         LOG_ALREADY_CLAIMED_DEBUG("claimed, skipping", bdev);
    2503           0 :                         spdk_spin_unlock(&bdev->internal.spinlock);
    2504           0 :                         continue;
    2505             :                 }
    2506           0 :                 spdk_spin_unlock(&bdev->internal.spinlock);
    2507             : 
    2508           0 :                 SPDK_DEBUGLOG(bdev, "Unregistering bdev '%s'\n", bdev->name);
    2509           0 :                 spdk_bdev_unregister(bdev, bdev_finish_unregister_bdevs_iter, bdev);
    2510           0 :                 return;
    2511             :         }
    2512             : 
    2513             :         /*
    2514             :          * If any bdev fails to unclaim underlying bdev properly, we may face the
    2515             :          * case of bdev list consisting of claimed bdevs only (if claims are managed
    2516             :          * correctly, this would mean there's a loop in the claims graph which is
    2517             :          * clearly impossible). Warn and unregister last bdev on the list then.
    2518             :          */
    2519           0 :         for (bdev = TAILQ_LAST(&g_bdev_mgr.bdevs, spdk_bdev_list);
    2520           0 :              bdev; bdev = TAILQ_PREV(bdev, spdk_bdev_list, internal.link)) {
    2521           0 :                 SPDK_WARNLOG("Unregistering claimed bdev '%s'!\n", bdev->name);
    2522           0 :                 spdk_bdev_unregister(bdev, bdev_finish_unregister_bdevs_iter, bdev);
    2523           0 :                 return;
    2524             :         }
    2525          65 : }
    2526             : 
    2527             : static void
    2528          65 : bdev_module_fini_start_iter(void *arg)
    2529             : {
    2530             :         struct spdk_bdev_module *bdev_module;
    2531             : 
    2532          65 :         if (!g_resume_bdev_module) {
    2533          65 :                 bdev_module = TAILQ_LAST(&g_bdev_mgr.bdev_modules, bdev_module_list);
    2534          65 :         } else {
    2535           0 :                 bdev_module = TAILQ_PREV(g_resume_bdev_module, bdev_module_list, internal.tailq);
    2536             :         }
    2537             : 
    2538         254 :         while (bdev_module) {
    2539         189 :                 if (bdev_module->async_fini_start) {
    2540             :                         /* Save our place so we can resume later. We must
    2541             :                          * save the variable here, before calling fini_start()
    2542             :                          * below, because in some cases the module may immediately
    2543             :                          * call spdk_bdev_module_fini_start_done() and re-enter
    2544             :                          * this function to continue iterating. */
    2545           0 :                         g_resume_bdev_module = bdev_module;
    2546           0 :                 }
    2547             : 
    2548         189 :                 if (bdev_module->fini_start) {
    2549          23 :                         bdev_module->fini_start();
    2550          23 :                 }
    2551             : 
    2552         189 :                 if (bdev_module->async_fini_start) {
    2553           0 :                         return;
    2554             :                 }
    2555             : 
    2556         189 :                 bdev_module = TAILQ_PREV(bdev_module, bdev_module_list, internal.tailq);
    2557             :         }
    2558             : 
    2559          65 :         g_resume_bdev_module = NULL;
    2560             : 
    2561          65 :         bdev_finish_unregister_bdevs_iter(NULL, 0);
    2562          65 : }
    2563             : 
    2564             : void
    2565           0 : spdk_bdev_module_fini_start_done(void)
    2566             : {
    2567           0 :         if (spdk_get_thread() != g_fini_thread) {
    2568           0 :                 spdk_thread_send_msg(g_fini_thread, bdev_module_fini_start_iter, NULL);
    2569           0 :         } else {
    2570           0 :                 bdev_module_fini_start_iter(NULL);
    2571             :         }
    2572           0 : }
    2573             : 
    2574             : static void
    2575          65 : bdev_finish_wait_for_examine_done(void *cb_arg)
    2576             : {
    2577          65 :         bdev_module_fini_start_iter(NULL);
    2578          65 : }
    2579             : 
    2580             : static void bdev_open_async_fini(void);
    2581             : 
    2582             : void
    2583          65 : spdk_bdev_finish(spdk_bdev_fini_cb cb_fn, void *cb_arg)
    2584             : {
    2585             :         int rc;
    2586             : 
    2587          65 :         assert(cb_fn != NULL);
    2588             : 
    2589          65 :         g_fini_thread = spdk_get_thread();
    2590             : 
    2591          65 :         g_fini_cb_fn = cb_fn;
    2592          65 :         g_fini_cb_arg = cb_arg;
    2593             : 
    2594          65 :         bdev_open_async_fini();
    2595             : 
    2596          65 :         rc = spdk_bdev_wait_for_examine(bdev_finish_wait_for_examine_done, NULL);
    2597          65 :         if (rc != 0) {
    2598           0 :                 SPDK_ERRLOG("wait_for_examine failed: %s\n", spdk_strerror(-rc));
    2599           0 :                 bdev_finish_wait_for_examine_done(NULL);
    2600           0 :         }
    2601          65 : }
    2602             : 
    2603             : struct spdk_bdev_io *
    2604         697 : bdev_channel_get_io(struct spdk_bdev_channel *channel)
    2605             : {
    2606         697 :         struct spdk_bdev_mgmt_channel *ch = channel->shared_resource->mgmt_ch;
    2607             :         struct spdk_bdev_io *bdev_io;
    2608             : 
    2609         697 :         if (ch->per_thread_cache_count > 0) {
    2610         637 :                 bdev_io = STAILQ_FIRST(&ch->per_thread_cache);
    2611         637 :                 STAILQ_REMOVE_HEAD(&ch->per_thread_cache, internal.buf_link);
    2612         637 :                 ch->per_thread_cache_count--;
    2613         697 :         } else if (spdk_unlikely(!TAILQ_EMPTY(&ch->io_wait_queue))) {
    2614             :                 /*
    2615             :                  * Don't try to look for bdev_ios in the global pool if there are
    2616             :                  * waiters on bdev_ios - we don't want this caller to jump the line.
    2617             :                  */
    2618           0 :                 bdev_io = NULL;
    2619           0 :         } else {
    2620          60 :                 bdev_io = spdk_mempool_get(g_bdev_mgr.bdev_io_pool);
    2621             :         }
    2622             : 
    2623         697 :         return bdev_io;
    2624             : }
    2625             : 
    2626             : void
    2627         691 : spdk_bdev_free_io(struct spdk_bdev_io *bdev_io)
    2628             : {
    2629             :         struct spdk_bdev_mgmt_channel *ch;
    2630             : 
    2631         691 :         assert(bdev_io != NULL);
    2632         691 :         assert(bdev_io->internal.status != SPDK_BDEV_IO_STATUS_PENDING);
    2633             : 
    2634         691 :         ch = bdev_io->internal.ch->shared_resource->mgmt_ch;
    2635             : 
    2636         691 :         if (bdev_io->internal.f.has_buf) {
    2637          16 :                 bdev_io_put_buf(bdev_io);
    2638          16 :         }
    2639             : 
    2640         691 :         if (ch->per_thread_cache_count < ch->bdev_io_cache_size) {
    2641         637 :                 ch->per_thread_cache_count++;
    2642         637 :                 STAILQ_INSERT_HEAD(&ch->per_thread_cache, bdev_io, internal.buf_link);
    2643         641 :                 while (ch->per_thread_cache_count > 0 && !TAILQ_EMPTY(&ch->io_wait_queue)) {
    2644             :                         struct spdk_bdev_io_wait_entry *entry;
    2645             : 
    2646           4 :                         entry = TAILQ_FIRST(&ch->io_wait_queue);
    2647           4 :                         TAILQ_REMOVE(&ch->io_wait_queue, entry, link);
    2648           4 :                         entry->cb_fn(entry->cb_arg);
    2649             :                 }
    2650         637 :         } else {
    2651             :                 /* We should never have a full cache with entries on the io wait queue. */
    2652          54 :                 assert(TAILQ_EMPTY(&ch->io_wait_queue));
    2653          54 :                 spdk_mempool_put(g_bdev_mgr.bdev_io_pool, (void *)bdev_io);
    2654             :         }
    2655         691 : }
    2656             : 
    2657             : static bool
    2658          72 : bdev_qos_is_iops_rate_limit(enum spdk_bdev_qos_rate_limit_type limit)
    2659             : {
    2660          72 :         assert(limit != SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES);
    2661             : 
    2662          72 :         switch (limit) {
    2663             :         case SPDK_BDEV_QOS_RW_IOPS_RATE_LIMIT:
    2664          18 :                 return true;
    2665             :         case SPDK_BDEV_QOS_RW_BPS_RATE_LIMIT:
    2666             :         case SPDK_BDEV_QOS_R_BPS_RATE_LIMIT:
    2667             :         case SPDK_BDEV_QOS_W_BPS_RATE_LIMIT:
    2668          54 :                 return false;
    2669           0 :         case SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES:
    2670             :         default:
    2671           0 :                 return false;
    2672             :         }
    2673          72 : }
    2674             : 
    2675             : static bool
    2676          25 : bdev_qos_io_to_limit(struct spdk_bdev_io *bdev_io)
    2677             : {
    2678          25 :         switch (bdev_io->type) {
    2679             :         case SPDK_BDEV_IO_TYPE_NVME_IO:
    2680             :         case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
    2681             :         case SPDK_BDEV_IO_TYPE_READ:
    2682             :         case SPDK_BDEV_IO_TYPE_WRITE:
    2683          23 :                 return true;
    2684             :         case SPDK_BDEV_IO_TYPE_ZCOPY:
    2685           0 :                 if (bdev_io->u.bdev.zcopy.start) {
    2686           0 :                         return true;
    2687             :                 } else {
    2688           0 :                         return false;
    2689             :                 }
    2690             :         default:
    2691           2 :                 return false;
    2692             :         }
    2693          25 : }
    2694             : 
    2695             : static bool
    2696          33 : bdev_is_read_io(struct spdk_bdev_io *bdev_io)
    2697             : {
    2698          33 :         switch (bdev_io->type) {
    2699             :         case SPDK_BDEV_IO_TYPE_NVME_IO:
    2700             :         case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
    2701             :                 /* Bit 1 (0x2) set for read operation */
    2702           0 :                 if (bdev_io->u.nvme_passthru.cmd.opc & SPDK_NVME_OPC_READ) {
    2703           0 :                         return true;
    2704             :                 } else {
    2705           0 :                         return false;
    2706             :                 }
    2707             :         case SPDK_BDEV_IO_TYPE_READ:
    2708          30 :                 return true;
    2709             :         case SPDK_BDEV_IO_TYPE_ZCOPY:
    2710             :                 /* Populate to read from disk */
    2711           0 :                 if (bdev_io->u.bdev.zcopy.populate) {
    2712           0 :                         return true;
    2713             :                 } else {
    2714           0 :                         return false;
    2715             :                 }
    2716             :         default:
    2717           3 :                 return false;
    2718             :         }
    2719          33 : }
    2720             : 
    2721             : static uint64_t
    2722          43 : bdev_get_io_size_in_byte(struct spdk_bdev_io *bdev_io)
    2723             : {
    2724          43 :         uint32_t blocklen = bdev_io_get_block_size(bdev_io);
    2725             : 
    2726          43 :         switch (bdev_io->type) {
    2727             :         case SPDK_BDEV_IO_TYPE_NVME_IO:
    2728             :         case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
    2729           0 :                 return bdev_io->u.nvme_passthru.nbytes;
    2730             :         case SPDK_BDEV_IO_TYPE_READ:
    2731             :         case SPDK_BDEV_IO_TYPE_WRITE:
    2732          43 :                 return bdev_io->u.bdev.num_blocks * blocklen;
    2733             :         case SPDK_BDEV_IO_TYPE_ZCOPY:
    2734             :                 /* Track the data in the start phase only */
    2735           0 :                 if (bdev_io->u.bdev.zcopy.start) {
    2736           0 :                         return bdev_io->u.bdev.num_blocks * blocklen;
    2737             :                 } else {
    2738           0 :                         return 0;
    2739             :                 }
    2740             :         default:
    2741           0 :                 return 0;
    2742             :         }
    2743          43 : }
    2744             : 
    2745             : static inline bool
    2746          64 : bdev_qos_rw_queue_io(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io, uint64_t delta)
    2747             : {
    2748             :         int64_t remaining_this_timeslice;
    2749             : 
    2750          64 :         if (!limit->max_per_timeslice) {
    2751             :                 /* The QoS is disabled */
    2752           0 :                 return false;
    2753             :         }
    2754             : 
    2755          64 :         remaining_this_timeslice = __atomic_sub_fetch(&limit->remaining_this_timeslice, delta,
    2756             :                                    __ATOMIC_RELAXED);
    2757          64 :         if (remaining_this_timeslice + (int64_t)delta > 0) {
    2758             :                 /* There was still a quota for this delta -> the IO shouldn't be queued
    2759             :                  *
    2760             :                  * We allow a slight quota overrun here so an IO bigger than the per-timeslice
    2761             :                  * quota can be allowed once a while. Such overrun then taken into account in
    2762             :                  * the QoS poller, where the next timeslice quota is calculated.
    2763             :                  */
    2764          59 :                 return false;
    2765             :         }
    2766             : 
    2767             :         /* There was no quota for this delta -> the IO should be queued
    2768             :          * The remaining_this_timeslice must be rewinded so it reflects the real
    2769             :          * amount of IOs or bytes allowed.
    2770             :          */
    2771           5 :         __atomic_add_fetch(
    2772           5 :                 &limit->remaining_this_timeslice, delta, __ATOMIC_RELAXED);
    2773           5 :         return true;
    2774          64 : }
    2775             : 
    2776             : static inline void
    2777           5 : bdev_qos_rw_rewind_io(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io, uint64_t delta)
    2778             : {
    2779           5 :         __atomic_add_fetch(&limit->remaining_this_timeslice, delta, __ATOMIC_RELAXED);
    2780           5 : }
    2781             : 
    2782             : static bool
    2783          23 : bdev_qos_rw_iops_queue(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
    2784             : {
    2785          23 :         return bdev_qos_rw_queue_io(limit, io, 1);
    2786             : }
    2787             : 
    2788             : static void
    2789           3 : bdev_qos_rw_iops_rewind_quota(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
    2790             : {
    2791           3 :         bdev_qos_rw_rewind_io(limit, io, 1);
    2792           3 : }
    2793             : 
    2794             : static bool
    2795          41 : bdev_qos_rw_bps_queue(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
    2796             : {
    2797          41 :         return bdev_qos_rw_queue_io(limit, io, bdev_get_io_size_in_byte(io));
    2798             : }
    2799             : 
    2800             : static void
    2801           2 : bdev_qos_rw_bps_rewind_quota(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
    2802             : {
    2803           2 :         bdev_qos_rw_rewind_io(limit, io, bdev_get_io_size_in_byte(io));
    2804           2 : }
    2805             : 
    2806             : static bool
    2807          19 : bdev_qos_r_bps_queue(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
    2808             : {
    2809          19 :         if (bdev_is_read_io(io) == false) {
    2810           1 :                 return false;
    2811             :         }
    2812             : 
    2813          18 :         return bdev_qos_rw_bps_queue(limit, io);
    2814          19 : }
    2815             : 
    2816             : static void
    2817           0 : bdev_qos_r_bps_rewind_quota(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
    2818             : {
    2819           0 :         if (bdev_is_read_io(io) != false) {
    2820           0 :                 bdev_qos_rw_rewind_io(limit, io, bdev_get_io_size_in_byte(io));
    2821           0 :         }
    2822           0 : }
    2823             : 
    2824             : static bool
    2825          14 : bdev_qos_w_bps_queue(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
    2826             : {
    2827          14 :         if (bdev_is_read_io(io) == true) {
    2828          12 :                 return false;
    2829             :         }
    2830             : 
    2831           2 :         return bdev_qos_rw_bps_queue(limit, io);
    2832          14 : }
    2833             : 
    2834             : static void
    2835           0 : bdev_qos_w_bps_rewind_quota(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
    2836             : {
    2837           0 :         if (bdev_is_read_io(io) != true) {
    2838           0 :                 bdev_qos_rw_rewind_io(limit, io, bdev_get_io_size_in_byte(io));
    2839           0 :         }
    2840           0 : }
    2841             : 
    2842             : static void
    2843          10 : bdev_qos_set_ops(struct spdk_bdev_qos *qos)
    2844             : {
    2845             :         int i;
    2846             : 
    2847          50 :         for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
    2848          40 :                 if (qos->rate_limits[i].limit == SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
    2849          15 :                         qos->rate_limits[i].queue_io = NULL;
    2850          15 :                         continue;
    2851             :                 }
    2852             : 
    2853          25 :                 switch (i) {
    2854             :                 case SPDK_BDEV_QOS_RW_IOPS_RATE_LIMIT:
    2855           9 :                         qos->rate_limits[i].queue_io = bdev_qos_rw_iops_queue;
    2856           9 :                         qos->rate_limits[i].rewind_quota = bdev_qos_rw_iops_rewind_quota;
    2857           9 :                         break;
    2858             :                 case SPDK_BDEV_QOS_RW_BPS_RATE_LIMIT:
    2859           7 :                         qos->rate_limits[i].queue_io = bdev_qos_rw_bps_queue;
    2860           7 :                         qos->rate_limits[i].rewind_quota = bdev_qos_rw_bps_rewind_quota;
    2861           7 :                         break;
    2862             :                 case SPDK_BDEV_QOS_R_BPS_RATE_LIMIT:
    2863           5 :                         qos->rate_limits[i].queue_io = bdev_qos_r_bps_queue;
    2864           5 :                         qos->rate_limits[i].rewind_quota = bdev_qos_r_bps_rewind_quota;
    2865           5 :                         break;
    2866             :                 case SPDK_BDEV_QOS_W_BPS_RATE_LIMIT:
    2867           4 :                         qos->rate_limits[i].queue_io = bdev_qos_w_bps_queue;
    2868           4 :                         qos->rate_limits[i].rewind_quota = bdev_qos_w_bps_rewind_quota;
    2869           4 :                         break;
    2870             :                 default:
    2871           0 :                         break;
    2872             :                 }
    2873          25 :         }
    2874          10 : }
    2875             : 
    2876             : static void
    2877           6 : _bdev_io_complete_in_submit(struct spdk_bdev_channel *bdev_ch,
    2878             :                             struct spdk_bdev_io *bdev_io,
    2879             :                             enum spdk_bdev_io_status status)
    2880             : {
    2881           6 :         bdev_io->internal.f.in_submit_request = true;
    2882           6 :         bdev_io_increment_outstanding(bdev_ch, bdev_ch->shared_resource);
    2883           6 :         spdk_bdev_io_complete(bdev_io, status);
    2884           6 :         bdev_io->internal.f.in_submit_request = false;
    2885           6 : }
    2886             : 
    2887             : static inline void
    2888         574 : bdev_io_do_submit(struct spdk_bdev_channel *bdev_ch, struct spdk_bdev_io *bdev_io)
    2889             : {
    2890         574 :         struct spdk_bdev *bdev = bdev_io->bdev;
    2891         574 :         struct spdk_io_channel *ch = bdev_ch->channel;
    2892         574 :         struct spdk_bdev_shared_resource *shared_resource = bdev_ch->shared_resource;
    2893             : 
    2894         574 :         if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_ABORT)) {
    2895          16 :                 struct spdk_bdev_mgmt_channel *mgmt_channel = shared_resource->mgmt_ch;
    2896          16 :                 struct spdk_bdev_io *bio_to_abort = bdev_io->u.abort.bio_to_abort;
    2897             : 
    2898          16 :                 if (bdev_abort_queued_io(&shared_resource->nomem_io, bio_to_abort) ||
    2899          16 :                     bdev_abort_buf_io(mgmt_channel, bio_to_abort)) {
    2900           0 :                         _bdev_io_complete_in_submit(bdev_ch, bdev_io,
    2901             :                                                     SPDK_BDEV_IO_STATUS_SUCCESS);
    2902           0 :                         return;
    2903             :                 }
    2904          16 :         }
    2905             : 
    2906         574 :         if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE &&
    2907             :                           bdev_io->bdev->split_on_write_unit &&
    2908             :                           bdev_io->u.bdev.num_blocks < bdev_io->bdev->write_unit_size)) {
    2909           4 :                 SPDK_ERRLOG("IO num_blocks %lu does not match the write_unit_size %u\n",
    2910             :                             bdev_io->u.bdev.num_blocks, bdev_io->bdev->write_unit_size);
    2911           4 :                 _bdev_io_complete_in_submit(bdev_ch, bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
    2912           4 :                 return;
    2913             :         }
    2914             : 
    2915         570 :         if (spdk_likely(TAILQ_EMPTY(&shared_resource->nomem_io))) {
    2916         527 :                 bdev_io_increment_outstanding(bdev_ch, shared_resource);
    2917         527 :                 bdev_io->internal.f.in_submit_request = true;
    2918         527 :                 bdev_submit_request(bdev, ch, bdev_io);
    2919         527 :                 bdev_io->internal.f.in_submit_request = false;
    2920         527 :         } else {
    2921          43 :                 bdev_queue_nomem_io_tail(shared_resource, bdev_io, BDEV_IO_RETRY_STATE_SUBMIT);
    2922          43 :                 if (shared_resource->nomem_threshold == 0 && shared_resource->io_outstanding == 0) {
    2923             :                         /* Special case when we have nomem IOs and no outstanding IOs which completions
    2924             :                          * could trigger retry of queued IOs */
    2925           0 :                         bdev_shared_ch_retry_io(shared_resource);
    2926           0 :                 }
    2927             :         }
    2928         574 : }
    2929             : 
    2930             : static bool
    2931          25 : bdev_qos_queue_io(struct spdk_bdev_qos *qos, struct spdk_bdev_io *bdev_io)
    2932             : {
    2933             :         int i;
    2934             : 
    2935          25 :         if (bdev_qos_io_to_limit(bdev_io) == true) {
    2936         100 :                 for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
    2937          82 :                         if (!qos->rate_limits[i].queue_io) {
    2938           5 :                                 continue;
    2939             :                         }
    2940             : 
    2941         231 :                         if (qos->rate_limits[i].queue_io(&qos->rate_limits[i],
    2942         154 :                                                          bdev_io) == true) {
    2943          10 :                                 for (i -= 1; i >= 0 ; i--) {
    2944           5 :                                         if (!qos->rate_limits[i].queue_io) {
    2945           0 :                                                 continue;
    2946             :                                         }
    2947             : 
    2948           5 :                                         qos->rate_limits[i].rewind_quota(&qos->rate_limits[i], bdev_io);
    2949           5 :                                 }
    2950           5 :                                 return true;
    2951             :                         }
    2952          72 :                 }
    2953          18 :         }
    2954             : 
    2955          20 :         return false;
    2956          25 : }
    2957             : 
    2958             : static int
    2959          27 : bdev_qos_io_submit(struct spdk_bdev_channel *ch, struct spdk_bdev_qos *qos)
    2960             : {
    2961          27 :         struct spdk_bdev_io             *bdev_io = NULL, *tmp = NULL;
    2962          27 :         int                             submitted_ios = 0;
    2963             : 
    2964          52 :         TAILQ_FOREACH_SAFE(bdev_io, &ch->qos_queued_io, internal.link, tmp) {
    2965          25 :                 if (!bdev_qos_queue_io(qos, bdev_io)) {
    2966          20 :                         TAILQ_REMOVE(&ch->qos_queued_io, bdev_io, internal.link);
    2967          20 :                         bdev_io_do_submit(ch, bdev_io);
    2968             : 
    2969          20 :                         submitted_ios++;
    2970          20 :                 }
    2971          25 :         }
    2972             : 
    2973          27 :         return submitted_ios;
    2974             : }
    2975             : 
    2976             : static void
    2977           2 : bdev_queue_io_wait_with_cb(struct spdk_bdev_io *bdev_io, spdk_bdev_io_wait_cb cb_fn)
    2978             : {
    2979             :         int rc;
    2980             : 
    2981           2 :         bdev_io->internal.waitq_entry.bdev = bdev_io->bdev;
    2982           2 :         bdev_io->internal.waitq_entry.cb_fn = cb_fn;
    2983           2 :         bdev_io->internal.waitq_entry.cb_arg = bdev_io;
    2984           4 :         rc = spdk_bdev_queue_io_wait(bdev_io->bdev, spdk_io_channel_from_ctx(bdev_io->internal.ch),
    2985           2 :                                      &bdev_io->internal.waitq_entry);
    2986           2 :         if (rc != 0) {
    2987           0 :                 SPDK_ERRLOG("Queue IO failed, rc=%d\n", rc);
    2988           0 :                 bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
    2989           0 :                 bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
    2990           0 :         }
    2991           2 : }
    2992             : 
    2993             : static bool
    2994         621 : bdev_rw_should_split(struct spdk_bdev_io *bdev_io)
    2995             : {
    2996             :         uint32_t io_boundary;
    2997         621 :         struct spdk_bdev *bdev = bdev_io->bdev;
    2998         621 :         uint32_t max_segment_size = bdev->max_segment_size;
    2999         621 :         uint32_t max_size = bdev->max_rw_size;
    3000         621 :         int max_segs = bdev->max_num_segments;
    3001             : 
    3002         621 :         if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE && bdev->split_on_write_unit) {
    3003          24 :                 io_boundary = bdev->write_unit_size;
    3004         621 :         } else if (bdev->split_on_optimal_io_boundary) {
    3005         168 :                 io_boundary = bdev->optimal_io_boundary;
    3006         168 :         } else {
    3007         429 :                 io_boundary = 0;
    3008             :         }
    3009             : 
    3010         621 :         if (spdk_likely(!io_boundary && !max_segs && !max_segment_size && !max_size)) {
    3011         243 :                 return false;
    3012             :         }
    3013             : 
    3014         378 :         if (io_boundary) {
    3015             :                 uint64_t start_stripe, end_stripe;
    3016             : 
    3017         192 :                 start_stripe = bdev_io->u.bdev.offset_blocks;
    3018         192 :                 end_stripe = start_stripe + bdev_io->u.bdev.num_blocks - 1;
    3019             :                 /* Avoid expensive div operations if possible.  These spdk_u32 functions are very cheap. */
    3020         192 :                 if (spdk_likely(spdk_u32_is_pow2(io_boundary))) {
    3021         192 :                         start_stripe >>= spdk_u32log2(io_boundary);
    3022         192 :                         end_stripe >>= spdk_u32log2(io_boundary);
    3023         192 :                 } else {
    3024           0 :                         start_stripe /= io_boundary;
    3025           0 :                         end_stripe /= io_boundary;
    3026             :                 }
    3027             : 
    3028         192 :                 if (start_stripe != end_stripe) {
    3029          75 :                         return true;
    3030             :                 }
    3031         117 :         }
    3032             : 
    3033         303 :         if (max_segs) {
    3034         150 :                 if (bdev_io->u.bdev.iovcnt > max_segs) {
    3035          15 :                         return true;
    3036             :                 }
    3037         135 :         }
    3038             : 
    3039         288 :         if (max_segment_size) {
    3040         470 :                 for (int i = 0; i < bdev_io->u.bdev.iovcnt; i++) {
    3041         346 :                         if (bdev_io->u.bdev.iovs[i].iov_len > max_segment_size) {
    3042          12 :                                 return true;
    3043             :                         }
    3044         334 :                 }
    3045         124 :         }
    3046             : 
    3047         276 :         if (max_size) {
    3048          52 :                 if (bdev_io->u.bdev.num_blocks > max_size) {
    3049           7 :                         return true;
    3050             :                 }
    3051          45 :         }
    3052             : 
    3053         269 :         return false;
    3054         621 : }
    3055             : 
    3056             : static bool
    3057          24 : bdev_unmap_should_split(struct spdk_bdev_io *bdev_io)
    3058             : {
    3059             :         uint32_t num_unmap_segments;
    3060             : 
    3061          24 :         if (!bdev_io->bdev->max_unmap || !bdev_io->bdev->max_unmap_segments) {
    3062           3 :                 return false;
    3063             :         }
    3064          21 :         num_unmap_segments = spdk_divide_round_up(bdev_io->u.bdev.num_blocks, bdev_io->bdev->max_unmap);
    3065          21 :         if (num_unmap_segments > bdev_io->bdev->max_unmap_segments) {
    3066           4 :                 return true;
    3067             :         }
    3068             : 
    3069          17 :         return false;
    3070          24 : }
    3071             : 
    3072             : static bool
    3073          37 : bdev_write_zeroes_should_split(struct spdk_bdev_io *bdev_io)
    3074             : {
    3075          37 :         if (!bdev_io->bdev->max_write_zeroes) {
    3076           4 :                 return false;
    3077             :         }
    3078             : 
    3079          33 :         if (bdev_io->u.bdev.num_blocks > bdev_io->bdev->max_write_zeroes) {
    3080          10 :                 return true;
    3081             :         }
    3082             : 
    3083          23 :         return false;
    3084          37 : }
    3085             : 
    3086             : static bool
    3087          30 : bdev_copy_should_split(struct spdk_bdev_io *bdev_io)
    3088             : {
    3089          30 :         if (bdev_io->bdev->max_copy != 0 &&
    3090          25 :             bdev_io->u.bdev.num_blocks > bdev_io->bdev->max_copy) {
    3091           6 :                 return true;
    3092             :         }
    3093             : 
    3094          24 :         return false;
    3095          30 : }
    3096             : 
    3097             : static bool
    3098         792 : bdev_io_should_split(struct spdk_bdev_io *bdev_io)
    3099             : {
    3100         792 :         switch (bdev_io->type) {
    3101             :         case SPDK_BDEV_IO_TYPE_READ:
    3102             :         case SPDK_BDEV_IO_TYPE_WRITE:
    3103         621 :                 return bdev_rw_should_split(bdev_io);
    3104             :         case SPDK_BDEV_IO_TYPE_UNMAP:
    3105          24 :                 return bdev_unmap_should_split(bdev_io);
    3106             :         case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
    3107          37 :                 return bdev_write_zeroes_should_split(bdev_io);
    3108             :         case SPDK_BDEV_IO_TYPE_COPY:
    3109          30 :                 return bdev_copy_should_split(bdev_io);
    3110             :         default:
    3111          80 :                 return false;
    3112             :         }
    3113         792 : }
    3114             : 
    3115             : static uint32_t
    3116         249 : _to_next_boundary(uint64_t offset, uint32_t boundary)
    3117             : {
    3118         249 :         return (boundary - (offset % boundary));
    3119             : }
    3120             : 
    3121             : static void bdev_io_split_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg);
    3122             : 
    3123             : static void _bdev_rw_split(void *_bdev_io);
    3124             : 
    3125             : static void bdev_unmap_split(struct spdk_bdev_io *bdev_io);
    3126             : 
    3127             : static void
    3128           0 : _bdev_unmap_split(void *_bdev_io)
    3129             : {
    3130           0 :         return bdev_unmap_split((struct spdk_bdev_io *)_bdev_io);
    3131             : }
    3132             : 
    3133             : static void bdev_write_zeroes_split(struct spdk_bdev_io *bdev_io);
    3134             : 
    3135             : static void
    3136           0 : _bdev_write_zeroes_split(void *_bdev_io)
    3137             : {
    3138           0 :         return bdev_write_zeroes_split((struct spdk_bdev_io *)_bdev_io);
    3139             : }
    3140             : 
    3141             : static void bdev_copy_split(struct spdk_bdev_io *bdev_io);
    3142             : 
    3143             : static void
    3144           0 : _bdev_copy_split(void *_bdev_io)
    3145             : {
    3146           0 :         return bdev_copy_split((struct spdk_bdev_io *)_bdev_io);
    3147             : }
    3148             : 
    3149             : static int
    3150         305 : bdev_io_split_submit(struct spdk_bdev_io *bdev_io, struct iovec *iov, int iovcnt, void *md_buf,
    3151             :                      uint64_t num_blocks, uint64_t *offset, uint64_t *remaining)
    3152             : {
    3153             :         int rc;
    3154             :         uint64_t current_offset, current_remaining, current_src_offset;
    3155             :         spdk_bdev_io_wait_cb io_wait_fn;
    3156             : 
    3157         305 :         current_offset = *offset;
    3158         305 :         current_remaining = *remaining;
    3159             : 
    3160         305 :         assert(bdev_io->internal.f.split);
    3161             : 
    3162         305 :         bdev_io->internal.split.outstanding++;
    3163             : 
    3164         305 :         io_wait_fn = _bdev_rw_split;
    3165         305 :         switch (bdev_io->type) {
    3166             :         case SPDK_BDEV_IO_TYPE_READ:
    3167         196 :                 assert(bdev_io->u.bdev.accel_sequence == NULL);
    3168         392 :                 rc = bdev_readv_blocks_with_md(bdev_io->internal.desc,
    3169         196 :                                                spdk_io_channel_from_ctx(bdev_io->internal.ch),
    3170         196 :                                                iov, iovcnt, md_buf, current_offset,
    3171         196 :                                                num_blocks,
    3172         196 :                                                bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain : NULL,
    3173         196 :                                                bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain_ctx : NULL,
    3174             :                                                NULL,
    3175         196 :                                                bdev_io->u.bdev.dif_check_flags,
    3176         196 :                                                bdev_io_split_done, bdev_io);
    3177         196 :                 break;
    3178             :         case SPDK_BDEV_IO_TYPE_WRITE:
    3179          50 :                 assert(bdev_io->u.bdev.accel_sequence == NULL);
    3180         100 :                 rc = bdev_writev_blocks_with_md(bdev_io->internal.desc,
    3181          50 :                                                 spdk_io_channel_from_ctx(bdev_io->internal.ch),
    3182          50 :                                                 iov, iovcnt, md_buf, current_offset,
    3183          50 :                                                 num_blocks,
    3184          50 :                                                 bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain : NULL,
    3185          50 :                                                 bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain_ctx : NULL,
    3186             :                                                 NULL,
    3187          50 :                                                 bdev_io->u.bdev.dif_check_flags,
    3188          50 :                                                 bdev_io->u.bdev.nvme_cdw12.raw,
    3189          50 :                                                 bdev_io->u.bdev.nvme_cdw13.raw,
    3190          50 :                                                 bdev_io_split_done, bdev_io);
    3191          50 :                 break;
    3192             :         case SPDK_BDEV_IO_TYPE_UNMAP:
    3193          17 :                 io_wait_fn = _bdev_unmap_split;
    3194          34 :                 rc = spdk_bdev_unmap_blocks(bdev_io->internal.desc,
    3195          17 :                                             spdk_io_channel_from_ctx(bdev_io->internal.ch),
    3196          17 :                                             current_offset, num_blocks,
    3197          17 :                                             bdev_io_split_done, bdev_io);
    3198          17 :                 break;
    3199             :         case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
    3200          23 :                 io_wait_fn = _bdev_write_zeroes_split;
    3201          46 :                 rc = spdk_bdev_write_zeroes_blocks(bdev_io->internal.desc,
    3202          23 :                                                    spdk_io_channel_from_ctx(bdev_io->internal.ch),
    3203          23 :                                                    current_offset, num_blocks,
    3204          23 :                                                    bdev_io_split_done, bdev_io);
    3205          23 :                 break;
    3206             :         case SPDK_BDEV_IO_TYPE_COPY:
    3207          19 :                 io_wait_fn = _bdev_copy_split;
    3208          38 :                 current_src_offset = bdev_io->u.bdev.copy.src_offset_blocks +
    3209          19 :                                      (current_offset - bdev_io->u.bdev.offset_blocks);
    3210          38 :                 rc = spdk_bdev_copy_blocks(bdev_io->internal.desc,
    3211          19 :                                            spdk_io_channel_from_ctx(bdev_io->internal.ch),
    3212          19 :                                            current_offset, current_src_offset, num_blocks,
    3213          19 :                                            bdev_io_split_done, bdev_io);
    3214          19 :                 break;
    3215             :         default:
    3216           0 :                 assert(false);
    3217             :                 rc = -EINVAL;
    3218             :                 break;
    3219             :         }
    3220             : 
    3221         305 :         if (rc == 0) {
    3222         301 :                 current_offset += num_blocks;
    3223         301 :                 current_remaining -= num_blocks;
    3224         301 :                 bdev_io->internal.split.current_offset_blocks = current_offset;
    3225         301 :                 bdev_io->internal.split.remaining_num_blocks = current_remaining;
    3226         301 :                 *offset = current_offset;
    3227         301 :                 *remaining = current_remaining;
    3228         301 :         } else {
    3229           4 :                 bdev_io->internal.split.outstanding--;
    3230           4 :                 if (rc == -ENOMEM) {
    3231           4 :                         if (bdev_io->internal.split.outstanding == 0) {
    3232             :                                 /* No I/O is outstanding. Hence we should wait here. */
    3233           1 :                                 bdev_queue_io_wait_with_cb(bdev_io, io_wait_fn);
    3234           1 :                         }
    3235           4 :                 } else {
    3236           0 :                         bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
    3237           0 :                         if (bdev_io->internal.split.outstanding == 0) {
    3238           0 :                                 bdev_ch_remove_from_io_submitted(bdev_io);
    3239           0 :                                 spdk_trace_record(TRACE_BDEV_IO_DONE, bdev_io->internal.ch->trace_id,
    3240             :                                                   0, (uintptr_t)bdev_io, bdev_io->internal.caller_ctx,
    3241             :                                                   bdev_io->internal.ch->queue_depth);
    3242           0 :                                 bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
    3243           0 :                         }
    3244             :                 }
    3245             :         }
    3246             : 
    3247         305 :         return rc;
    3248             : }
    3249             : 
    3250             : static void
    3251          67 : _bdev_rw_split(void *_bdev_io)
    3252             : {
    3253             :         struct iovec *parent_iov, *iov;
    3254          67 :         struct spdk_bdev_io *bdev_io = _bdev_io;
    3255          67 :         struct spdk_bdev *bdev = bdev_io->bdev;
    3256             :         uint64_t parent_offset, current_offset, remaining;
    3257             :         uint32_t parent_iov_offset, parent_iovcnt, parent_iovpos, child_iovcnt;
    3258             :         uint32_t to_next_boundary, to_next_boundary_bytes, to_last_block_bytes;
    3259             :         uint32_t iovcnt, iov_len, child_iovsize;
    3260             :         uint32_t blocklen;
    3261             :         uint32_t io_boundary;
    3262          67 :         uint32_t max_segment_size = bdev->max_segment_size;
    3263          67 :         uint32_t max_child_iovcnt = bdev->max_num_segments;
    3264          67 :         uint32_t max_size = bdev->max_rw_size;
    3265          67 :         void *md_buf = NULL;
    3266             :         int rc;
    3267             : 
    3268          67 :         blocklen = bdev_io_get_block_size(bdev_io);
    3269             : 
    3270          67 :         max_size = max_size ? max_size : UINT32_MAX;
    3271          67 :         max_segment_size = max_segment_size ? max_segment_size : UINT32_MAX;
    3272          67 :         max_child_iovcnt = max_child_iovcnt ? spdk_min(max_child_iovcnt, SPDK_BDEV_IO_NUM_CHILD_IOV) :
    3273             :                            SPDK_BDEV_IO_NUM_CHILD_IOV;
    3274             : 
    3275          67 :         if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE && bdev->split_on_write_unit) {
    3276           5 :                 io_boundary = bdev->write_unit_size;
    3277          67 :         } else if (bdev->split_on_optimal_io_boundary) {
    3278          40 :                 io_boundary = bdev->optimal_io_boundary;
    3279          40 :         } else {
    3280          22 :                 io_boundary = UINT32_MAX;
    3281             :         }
    3282             : 
    3283          67 :         assert(bdev_io->internal.f.split);
    3284             : 
    3285          67 :         remaining = bdev_io->internal.split.remaining_num_blocks;
    3286          67 :         current_offset = bdev_io->internal.split.current_offset_blocks;
    3287          67 :         parent_offset = bdev_io->u.bdev.offset_blocks;
    3288          67 :         parent_iov_offset = (current_offset - parent_offset) * blocklen;
    3289          67 :         parent_iovcnt = bdev_io->u.bdev.iovcnt;
    3290             : 
    3291         420 :         for (parent_iovpos = 0; parent_iovpos < parent_iovcnt; parent_iovpos++) {
    3292         420 :                 parent_iov = &bdev_io->u.bdev.iovs[parent_iovpos];
    3293         420 :                 if (parent_iov_offset < parent_iov->iov_len) {
    3294          67 :                         break;
    3295             :                 }
    3296         353 :                 parent_iov_offset -= parent_iov->iov_len;
    3297         353 :         }
    3298             : 
    3299          67 :         child_iovcnt = 0;
    3300         573 :         while (remaining > 0 && parent_iovpos < parent_iovcnt &&
    3301         264 :                child_iovcnt < SPDK_BDEV_IO_NUM_CHILD_IOV) {
    3302         249 :                 to_next_boundary = _to_next_boundary(current_offset, io_boundary);
    3303         249 :                 to_next_boundary = spdk_min(remaining, to_next_boundary);
    3304         249 :                 to_next_boundary = spdk_min(max_size, to_next_boundary);
    3305         249 :                 to_next_boundary_bytes = to_next_boundary * blocklen;
    3306             : 
    3307         249 :                 iov = &bdev_io->child_iov[child_iovcnt];
    3308         249 :                 iovcnt = 0;
    3309             : 
    3310         249 :                 if (bdev_io->u.bdev.md_buf) {
    3311          48 :                         md_buf = (char *)bdev_io->u.bdev.md_buf +
    3312          24 :                                  (current_offset - parent_offset) * spdk_bdev_get_md_size(bdev);
    3313          24 :                 }
    3314             : 
    3315         249 :                 child_iovsize = spdk_min(SPDK_BDEV_IO_NUM_CHILD_IOV - child_iovcnt, max_child_iovcnt);
    3316        1810 :                 while (to_next_boundary_bytes > 0 && parent_iovpos < parent_iovcnt &&
    3317         836 :                        iovcnt < child_iovsize) {
    3318         725 :                         parent_iov = &bdev_io->u.bdev.iovs[parent_iovpos];
    3319         725 :                         iov_len = parent_iov->iov_len - parent_iov_offset;
    3320             : 
    3321         725 :                         iov_len = spdk_min(iov_len, max_segment_size);
    3322         725 :                         iov_len = spdk_min(iov_len, to_next_boundary_bytes);
    3323         725 :                         to_next_boundary_bytes -= iov_len;
    3324             : 
    3325         725 :                         bdev_io->child_iov[child_iovcnt].iov_base = parent_iov->iov_base + parent_iov_offset;
    3326         725 :                         bdev_io->child_iov[child_iovcnt].iov_len = iov_len;
    3327             : 
    3328         725 :                         if (iov_len < parent_iov->iov_len - parent_iov_offset) {
    3329         183 :                                 parent_iov_offset += iov_len;
    3330         183 :                         } else {
    3331         542 :                                 parent_iovpos++;
    3332         542 :                                 parent_iov_offset = 0;
    3333             :                         }
    3334         725 :                         child_iovcnt++;
    3335         725 :                         iovcnt++;
    3336             :                 }
    3337             : 
    3338         249 :                 if (to_next_boundary_bytes > 0) {
    3339             :                         /* We had to stop this child I/O early because we ran out of
    3340             :                          * child_iov space or were limited by max_num_segments.
    3341             :                          * Ensure the iovs to be aligned with block size and
    3342             :                          * then adjust to_next_boundary before starting the
    3343             :                          * child I/O.
    3344             :                          */
    3345         111 :                         assert(child_iovcnt == SPDK_BDEV_IO_NUM_CHILD_IOV ||
    3346             :                                iovcnt == child_iovsize);
    3347         111 :                         to_last_block_bytes = to_next_boundary_bytes % blocklen;
    3348         111 :                         if (to_last_block_bytes != 0) {
    3349          24 :                                 uint32_t child_iovpos = child_iovcnt - 1;
    3350             :                                 /* don't decrease child_iovcnt when it equals to SPDK_BDEV_IO_NUM_CHILD_IOV
    3351             :                                  * so the loop will naturally end
    3352             :                                  */
    3353             : 
    3354          24 :                                 to_last_block_bytes = blocklen - to_last_block_bytes;
    3355          24 :                                 to_next_boundary_bytes += to_last_block_bytes;
    3356          53 :                                 while (to_last_block_bytes > 0 && iovcnt > 0) {
    3357          32 :                                         iov_len = spdk_min(to_last_block_bytes,
    3358             :                                                            bdev_io->child_iov[child_iovpos].iov_len);
    3359          32 :                                         bdev_io->child_iov[child_iovpos].iov_len -= iov_len;
    3360          32 :                                         if (bdev_io->child_iov[child_iovpos].iov_len == 0) {
    3361          15 :                                                 child_iovpos--;
    3362          15 :                                                 if (--iovcnt == 0) {
    3363             :                                                         /* If the child IO is less than a block size just return.
    3364             :                                                          * If the first child IO of any split round is less than
    3365             :                                                          * a block size, an error exit.
    3366             :                                                          */
    3367           3 :                                                         if (bdev_io->internal.split.outstanding == 0) {
    3368           1 :                                                                 SPDK_ERRLOG("The first child io was less than a block size\n");
    3369           1 :                                                                 bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
    3370           1 :                                                                 bdev_ch_remove_from_io_submitted(bdev_io);
    3371           1 :                                                                 spdk_trace_record(TRACE_BDEV_IO_DONE, bdev_io->internal.ch->trace_id,
    3372             :                                                                                   0, (uintptr_t)bdev_io, bdev_io->internal.caller_ctx,
    3373             :                                                                                   bdev_io->internal.ch->queue_depth);
    3374           1 :                                                                 bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
    3375           1 :                                                         }
    3376             : 
    3377           3 :                                                         return;
    3378             :                                                 }
    3379          12 :                                         }
    3380             : 
    3381          29 :                                         to_last_block_bytes -= iov_len;
    3382             : 
    3383          29 :                                         if (parent_iov_offset == 0) {
    3384          14 :                                                 parent_iovpos--;
    3385          14 :                                                 parent_iov_offset = bdev_io->u.bdev.iovs[parent_iovpos].iov_len;
    3386          14 :                                         }
    3387          29 :                                         parent_iov_offset -= iov_len;
    3388             :                                 }
    3389             : 
    3390          21 :                                 assert(to_last_block_bytes == 0);
    3391          21 :                         }
    3392         108 :                         to_next_boundary -= to_next_boundary_bytes / blocklen;
    3393         108 :                 }
    3394             : 
    3395         246 :                 rc = bdev_io_split_submit(bdev_io, iov, iovcnt, md_buf, to_next_boundary,
    3396             :                                           &current_offset, &remaining);
    3397         246 :                 if (spdk_unlikely(rc)) {
    3398           4 :                         return;
    3399             :                 }
    3400             :         }
    3401          67 : }
    3402             : 
    3403             : static void
    3404           3 : bdev_unmap_split(struct spdk_bdev_io *bdev_io)
    3405             : {
    3406             :         uint64_t offset, unmap_blocks, remaining, max_unmap_blocks;
    3407           3 :         uint32_t num_children_reqs = 0;
    3408             :         int rc;
    3409             : 
    3410           3 :         assert(bdev_io->internal.f.split);
    3411             : 
    3412           3 :         offset = bdev_io->internal.split.current_offset_blocks;
    3413           3 :         remaining = bdev_io->internal.split.remaining_num_blocks;
    3414           3 :         max_unmap_blocks = bdev_io->bdev->max_unmap * bdev_io->bdev->max_unmap_segments;
    3415             : 
    3416          20 :         while (remaining && (num_children_reqs < SPDK_BDEV_MAX_CHILDREN_UNMAP_WRITE_ZEROES_REQS)) {
    3417          17 :                 unmap_blocks = spdk_min(remaining, max_unmap_blocks);
    3418             : 
    3419          17 :                 rc = bdev_io_split_submit(bdev_io, NULL, 0, NULL, unmap_blocks,
    3420             :                                           &offset, &remaining);
    3421          17 :                 if (spdk_likely(rc == 0)) {
    3422          17 :                         num_children_reqs++;
    3423          17 :                 } else {
    3424           0 :                         return;
    3425             :                 }
    3426             :         }
    3427           3 : }
    3428             : 
    3429             : static void
    3430           6 : bdev_write_zeroes_split(struct spdk_bdev_io *bdev_io)
    3431             : {
    3432             :         uint64_t offset, write_zeroes_blocks, remaining;
    3433           6 :         uint32_t num_children_reqs = 0;
    3434             :         int rc;
    3435             : 
    3436           6 :         assert(bdev_io->internal.f.split);
    3437             : 
    3438           6 :         offset = bdev_io->internal.split.current_offset_blocks;
    3439           6 :         remaining = bdev_io->internal.split.remaining_num_blocks;
    3440             : 
    3441          29 :         while (remaining && (num_children_reqs < SPDK_BDEV_MAX_CHILDREN_UNMAP_WRITE_ZEROES_REQS)) {
    3442          23 :                 write_zeroes_blocks = spdk_min(remaining, bdev_io->bdev->max_write_zeroes);
    3443             : 
    3444          23 :                 rc = bdev_io_split_submit(bdev_io, NULL, 0, NULL, write_zeroes_blocks,
    3445             :                                           &offset, &remaining);
    3446          23 :                 if (spdk_likely(rc == 0)) {
    3447          23 :                         num_children_reqs++;
    3448          23 :                 } else {
    3449           0 :                         return;
    3450             :                 }
    3451             :         }
    3452           6 : }
    3453             : 
    3454             : static void
    3455           4 : bdev_copy_split(struct spdk_bdev_io *bdev_io)
    3456             : {
    3457             :         uint64_t offset, copy_blocks, remaining;
    3458           4 :         uint32_t num_children_reqs = 0;
    3459             :         int rc;
    3460             : 
    3461           4 :         assert(bdev_io->internal.f.split);
    3462             : 
    3463           4 :         offset = bdev_io->internal.split.current_offset_blocks;
    3464           4 :         remaining = bdev_io->internal.split.remaining_num_blocks;
    3465             : 
    3466           4 :         assert(bdev_io->bdev->max_copy != 0);
    3467          23 :         while (remaining && (num_children_reqs < SPDK_BDEV_MAX_CHILDREN_COPY_REQS)) {
    3468          19 :                 copy_blocks = spdk_min(remaining, bdev_io->bdev->max_copy);
    3469             : 
    3470          19 :                 rc = bdev_io_split_submit(bdev_io, NULL, 0, NULL, copy_blocks,
    3471             :                                           &offset, &remaining);
    3472          19 :                 if (spdk_likely(rc == 0)) {
    3473          19 :                         num_children_reqs++;
    3474          19 :                 } else {
    3475           0 :                         return;
    3476             :                 }
    3477             :         }
    3478           4 : }
    3479             : 
    3480             : static void
    3481          58 : parent_bdev_io_complete(void *ctx, int rc)
    3482             : {
    3483          58 :         struct spdk_bdev_io *parent_io = ctx;
    3484             : 
    3485          58 :         if (rc) {
    3486           0 :                 parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
    3487           0 :         }
    3488             : 
    3489         116 :         parent_io->internal.cb(parent_io, parent_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS,
    3490          58 :                                parent_io->internal.caller_ctx);
    3491          58 : }
    3492             : 
    3493             : static void
    3494           0 : bdev_io_complete_parent_sequence_cb(void *ctx, int status)
    3495             : {
    3496           0 :         struct spdk_bdev_io *bdev_io = ctx;
    3497             : 
    3498             :         /* u.bdev.accel_sequence should have already been cleared at this point */
    3499           0 :         assert(bdev_io->u.bdev.accel_sequence == NULL);
    3500           0 :         assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS);
    3501           0 :         bdev_io->internal.f.has_accel_sequence = false;
    3502             : 
    3503           0 :         if (spdk_unlikely(status != 0)) {
    3504           0 :                 SPDK_ERRLOG("Failed to execute accel sequence, status=%d\n", status);
    3505           0 :         }
    3506             : 
    3507           0 :         parent_bdev_io_complete(bdev_io, status);
    3508           0 : }
    3509             : 
    3510             : static void
    3511         301 : bdev_io_split_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
    3512             : {
    3513         301 :         struct spdk_bdev_io *parent_io = cb_arg;
    3514             : 
    3515         301 :         spdk_bdev_free_io(bdev_io);
    3516             : 
    3517         301 :         assert(parent_io->internal.f.split);
    3518             : 
    3519         301 :         if (!success) {
    3520          21 :                 parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
    3521             :                 /* If any child I/O failed, stop further splitting process. */
    3522          21 :                 parent_io->internal.split.current_offset_blocks += parent_io->internal.split.remaining_num_blocks;
    3523          21 :                 parent_io->internal.split.remaining_num_blocks = 0;
    3524          21 :         }
    3525         301 :         parent_io->internal.split.outstanding--;
    3526         301 :         if (parent_io->internal.split.outstanding != 0) {
    3527         223 :                 return;
    3528             :         }
    3529             : 
    3530             :         /*
    3531             :          * Parent I/O finishes when all blocks are consumed.
    3532             :          */
    3533          78 :         if (parent_io->internal.split.remaining_num_blocks == 0) {
    3534          58 :                 assert(parent_io->internal.cb != bdev_io_split_done);
    3535          58 :                 bdev_ch_remove_from_io_submitted(parent_io);
    3536          58 :                 spdk_trace_record(TRACE_BDEV_IO_DONE, parent_io->internal.ch->trace_id,
    3537             :                                   0, (uintptr_t)parent_io, bdev_io->internal.caller_ctx,
    3538             :                                   parent_io->internal.ch->queue_depth);
    3539             : 
    3540          58 :                 if (spdk_likely(parent_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS)) {
    3541          48 :                         if (bdev_io_needs_sequence_exec(parent_io->internal.desc, parent_io)) {
    3542           0 :                                 bdev_io_exec_sequence(parent_io, bdev_io_complete_parent_sequence_cb);
    3543           0 :                                 return;
    3544          48 :                         } else if (parent_io->internal.f.has_bounce_buf &&
    3545           0 :                                    !bdev_io_use_accel_sequence(bdev_io)) {
    3546             :                                 /* bdev IO will be completed in the callback */
    3547           0 :                                 _bdev_io_push_bounce_data_buffer(parent_io, parent_bdev_io_complete);
    3548           0 :                                 return;
    3549             :                         }
    3550          48 :                 }
    3551             : 
    3552          58 :                 parent_bdev_io_complete(parent_io, 0);
    3553          58 :                 return;
    3554             :         }
    3555             : 
    3556             :         /*
    3557             :          * Continue with the splitting process.  This function will complete the parent I/O if the
    3558             :          * splitting is done.
    3559             :          */
    3560          20 :         switch (parent_io->type) {
    3561             :         case SPDK_BDEV_IO_TYPE_READ:
    3562             :         case SPDK_BDEV_IO_TYPE_WRITE:
    3563          17 :                 _bdev_rw_split(parent_io);
    3564          17 :                 break;
    3565             :         case SPDK_BDEV_IO_TYPE_UNMAP:
    3566           1 :                 bdev_unmap_split(parent_io);
    3567           1 :                 break;
    3568             :         case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
    3569           1 :                 bdev_write_zeroes_split(parent_io);
    3570           1 :                 break;
    3571             :         case SPDK_BDEV_IO_TYPE_COPY:
    3572           1 :                 bdev_copy_split(parent_io);
    3573           1 :                 break;
    3574             :         default:
    3575           0 :                 assert(false);
    3576             :                 break;
    3577             :         }
    3578         301 : }
    3579             : 
    3580             : static void bdev_rw_split_get_buf_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io,
    3581             :                                      bool success);
    3582             : 
    3583             : static void
    3584          59 : bdev_io_split(struct spdk_bdev_io *bdev_io)
    3585             : {
    3586          59 :         assert(bdev_io_should_split(bdev_io));
    3587          59 :         assert(bdev_io->internal.f.split);
    3588             : 
    3589          59 :         bdev_io->internal.split.current_offset_blocks = bdev_io->u.bdev.offset_blocks;
    3590          59 :         bdev_io->internal.split.remaining_num_blocks = bdev_io->u.bdev.num_blocks;
    3591          59 :         bdev_io->internal.split.outstanding = 0;
    3592          59 :         bdev_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
    3593             : 
    3594          59 :         switch (bdev_io->type) {
    3595             :         case SPDK_BDEV_IO_TYPE_READ:
    3596             :         case SPDK_BDEV_IO_TYPE_WRITE:
    3597          49 :                 if (_is_buf_allocated(bdev_io->u.bdev.iovs)) {
    3598          49 :                         _bdev_rw_split(bdev_io);
    3599          49 :                 } else {
    3600           0 :                         assert(bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
    3601           0 :                         spdk_bdev_io_get_buf(bdev_io, bdev_rw_split_get_buf_cb,
    3602           0 :                                              bdev_io->u.bdev.num_blocks * bdev_io_get_block_size(bdev_io));
    3603             :                 }
    3604          49 :                 break;
    3605             :         case SPDK_BDEV_IO_TYPE_UNMAP:
    3606           2 :                 bdev_unmap_split(bdev_io);
    3607           2 :                 break;
    3608             :         case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
    3609           5 :                 bdev_write_zeroes_split(bdev_io);
    3610           5 :                 break;
    3611             :         case SPDK_BDEV_IO_TYPE_COPY:
    3612           3 :                 bdev_copy_split(bdev_io);
    3613           3 :                 break;
    3614             :         default:
    3615           0 :                 assert(false);
    3616             :                 break;
    3617             :         }
    3618          59 : }
    3619             : 
    3620             : static void
    3621           0 : bdev_rw_split_get_buf_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io, bool success)
    3622             : {
    3623           0 :         if (!success) {
    3624           0 :                 spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
    3625           0 :                 return;
    3626             :         }
    3627             : 
    3628           0 :         _bdev_rw_split(bdev_io);
    3629           0 : }
    3630             : 
    3631             : static inline void
    3632         579 : _bdev_io_submit(struct spdk_bdev_io *bdev_io)
    3633             : {
    3634         579 :         struct spdk_bdev *bdev = bdev_io->bdev;
    3635         579 :         struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
    3636             : 
    3637         579 :         if (spdk_likely(bdev_ch->flags == 0)) {
    3638         554 :                 bdev_io_do_submit(bdev_ch, bdev_io);
    3639         554 :                 return;
    3640             :         }
    3641             : 
    3642          25 :         if (bdev_ch->flags & BDEV_CH_RESET_IN_PROGRESS) {
    3643           2 :                 _bdev_io_complete_in_submit(bdev_ch, bdev_io, SPDK_BDEV_IO_STATUS_ABORTED);
    3644          25 :         } else if (bdev_ch->flags & BDEV_CH_QOS_ENABLED) {
    3645          23 :                 if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_ABORT) &&
    3646           2 :                     bdev_abort_queued_io(&bdev_ch->qos_queued_io, bdev_io->u.abort.bio_to_abort)) {
    3647           0 :                         _bdev_io_complete_in_submit(bdev_ch, bdev_io, SPDK_BDEV_IO_STATUS_SUCCESS);
    3648           0 :                 } else {
    3649          23 :                         TAILQ_INSERT_TAIL(&bdev_ch->qos_queued_io, bdev_io, internal.link);
    3650          23 :                         bdev_qos_io_submit(bdev_ch, bdev->internal.qos);
    3651             :                 }
    3652          23 :         } else {
    3653           0 :                 SPDK_ERRLOG("unknown bdev_ch flag %x found\n", bdev_ch->flags);
    3654           0 :                 _bdev_io_complete_in_submit(bdev_ch, bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
    3655             :         }
    3656         579 : }
    3657             : 
    3658             : bool bdev_lba_range_overlapped(struct lba_range *range1, struct lba_range *range2);
    3659             : 
    3660             : bool
    3661          23 : bdev_lba_range_overlapped(struct lba_range *range1, struct lba_range *range2)
    3662             : {
    3663          23 :         if (range1->length == 0 || range2->length == 0) {
    3664           1 :                 return false;
    3665             :         }
    3666             : 
    3667          22 :         if (range1->offset + range1->length <= range2->offset) {
    3668           1 :                 return false;
    3669             :         }
    3670             : 
    3671          21 :         if (range2->offset + range2->length <= range1->offset) {
    3672           3 :                 return false;
    3673             :         }
    3674             : 
    3675          18 :         return true;
    3676          23 : }
    3677             : 
    3678             : static bool
    3679          11 : bdev_io_range_is_locked(struct spdk_bdev_io *bdev_io, struct lba_range *range)
    3680             : {
    3681          11 :         struct spdk_bdev_channel *ch = bdev_io->internal.ch;
    3682             :         struct lba_range r;
    3683             : 
    3684          11 :         switch (bdev_io->type) {
    3685             :         case SPDK_BDEV_IO_TYPE_NVME_IO:
    3686             :         case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
    3687             :                 /* Don't try to decode the NVMe command - just assume worst-case and that
    3688             :                  * it overlaps a locked range.
    3689             :                  */
    3690           0 :                 return true;
    3691             :         case SPDK_BDEV_IO_TYPE_READ:
    3692           6 :                 if (!range->quiesce) {
    3693           4 :                         return false;
    3694             :                 }
    3695             :         /* fallthrough */
    3696             :         case SPDK_BDEV_IO_TYPE_WRITE:
    3697             :         case SPDK_BDEV_IO_TYPE_UNMAP:
    3698             :         case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
    3699             :         case SPDK_BDEV_IO_TYPE_ZCOPY:
    3700             :         case SPDK_BDEV_IO_TYPE_COPY:
    3701           7 :                 r.offset = bdev_io->u.bdev.offset_blocks;
    3702           7 :                 r.length = bdev_io->u.bdev.num_blocks;
    3703           7 :                 if (!bdev_lba_range_overlapped(range, &r)) {
    3704             :                         /* This I/O doesn't overlap the specified LBA range. */
    3705           0 :                         return false;
    3706           7 :                 } else if (range->owner_ch == ch && range->locked_ctx == bdev_io->internal.caller_ctx) {
    3707             :                         /* This I/O overlaps, but the I/O is on the same channel that locked this
    3708             :                          * range, and the caller_ctx is the same as the locked_ctx.  This means
    3709             :                          * that this I/O is associated with the lock, and is allowed to execute.
    3710             :                          */
    3711           2 :                         return false;
    3712             :                 } else {
    3713           5 :                         return true;
    3714             :                 }
    3715             :         default:
    3716           0 :                 return false;
    3717             :         }
    3718          11 : }
    3719             : 
    3720             : void
    3721         639 : bdev_io_submit(struct spdk_bdev_io *bdev_io)
    3722             : {
    3723         639 :         struct spdk_bdev_channel *ch = bdev_io->internal.ch;
    3724             : 
    3725         639 :         assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_PENDING);
    3726             : 
    3727         639 :         if (!TAILQ_EMPTY(&ch->locked_ranges)) {
    3728             :                 struct lba_range *range;
    3729             : 
    3730          13 :                 TAILQ_FOREACH(range, &ch->locked_ranges, tailq) {
    3731           8 :                         if (bdev_io_range_is_locked(bdev_io, range)) {
    3732           3 :                                 TAILQ_INSERT_TAIL(&ch->io_locked, bdev_io, internal.ch_link);
    3733           3 :                                 return;
    3734             :                         }
    3735           5 :                 }
    3736           5 :         }
    3737             : 
    3738         636 :         bdev_ch_add_to_io_submitted(bdev_io);
    3739             : 
    3740         636 :         bdev_io->internal.submit_tsc = spdk_get_ticks();
    3741         636 :         spdk_trace_record_tsc(bdev_io->internal.submit_tsc, TRACE_BDEV_IO_START,
    3742             :                               ch->trace_id, bdev_io->u.bdev.num_blocks,
    3743             :                               (uintptr_t)bdev_io, (uint64_t)bdev_io->type, bdev_io->internal.caller_ctx,
    3744             :                               bdev_io->u.bdev.offset_blocks, ch->queue_depth);
    3745             : 
    3746         636 :         if (bdev_io->internal.f.split) {
    3747          59 :                 bdev_io_split(bdev_io);
    3748          59 :                 return;
    3749             :         }
    3750             : 
    3751         577 :         _bdev_io_submit(bdev_io);
    3752         639 : }
    3753             : 
    3754             : static inline int
    3755           2 : bdev_io_init_dif_ctx(struct spdk_bdev_io *bdev_io)
    3756             : {
    3757           2 :         struct spdk_bdev *bdev = bdev_io->bdev;
    3758             :         struct spdk_dif_ctx_init_ext_opts dif_opts;
    3759             : 
    3760           2 :         memset(&bdev_io->u.bdev.dif_err, 0, sizeof(struct spdk_dif_error));
    3761             : 
    3762           2 :         dif_opts.size = SPDK_SIZEOF(&dif_opts, dif_pi_format);
    3763           2 :         dif_opts.dif_pi_format = bdev->dif_pi_format;
    3764             : 
    3765           4 :         return spdk_dif_ctx_init(&bdev_io->u.bdev.dif_ctx,
    3766           2 :                                  bdev->blocklen,
    3767           2 :                                  bdev->md_len,
    3768           2 :                                  bdev->md_interleave,
    3769           2 :                                  bdev->dif_is_head_of_md,
    3770           2 :                                  bdev->dif_type,
    3771           2 :                                  bdev_io->u.bdev.dif_check_flags,
    3772           2 :                                  bdev_io->u.bdev.offset_blocks & 0xFFFFFFFF,
    3773             :                                  0xFFFF, 0, 0, 0, &dif_opts);
    3774             : }
    3775             : 
    3776             : static void
    3777           4 : _bdev_memory_domain_get_io_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io,
    3778             :                               bool success)
    3779             : {
    3780           4 :         if (!success) {
    3781           0 :                 SPDK_ERRLOG("Failed to get data buffer, completing IO\n");
    3782           0 :                 bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
    3783           0 :                 bdev_io_complete_unsubmitted(bdev_io);
    3784           0 :                 return;
    3785             :         }
    3786             : 
    3787           4 :         if (bdev_io_needs_sequence_exec(bdev_io->internal.desc, bdev_io)) {
    3788           0 :                 if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
    3789           0 :                         bdev_io_exec_sequence(bdev_io, bdev_io_submit_sequence_cb);
    3790           0 :                         return;
    3791             :                 }
    3792             :                 /* For reads we'll execute the sequence after the data is read, so, for now, only
    3793             :                  * clear out accel_sequence pointer and submit the IO */
    3794           0 :                 assert(bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
    3795           0 :                 bdev_io->u.bdev.accel_sequence = NULL;
    3796           0 :         }
    3797             : 
    3798           4 :         bdev_io_submit(bdev_io);
    3799           4 : }
    3800             : 
    3801             : static inline void
    3802           4 : _bdev_io_ext_use_bounce_buffer(struct spdk_bdev_io *bdev_io)
    3803             : {
    3804             :         /* bdev doesn't support memory domains, thereby buffers in this IO request can't
    3805             :          * be accessed directly. It is needed to allocate buffers before issuing IO operation.
    3806             :          * For write operation we need to pull buffers from memory domain before submitting IO.
    3807             :          * Once read operation completes, we need to use memory_domain push functionality to
    3808             :          * update data in original memory domain IO buffer.
    3809             :          *
    3810             :          * If this I/O request is not aware of metadata, buffers in thsi IO request can't be
    3811             :          * accessed directly too. It is needed to allocate buffers before issuing IO operation.
    3812             :          * For write operation we need to insert metadata before submitting IO. Once read
    3813             :          * operation completes, we need to strip metadata in original IO buffer.
    3814             :          *
    3815             :          * This IO request will go through a regular IO flow, so clear memory domains pointers */
    3816           4 :         assert(bdev_io_use_memory_domain(bdev_io) ||
    3817             :                bdev_io_needs_metadata(bdev_io->internal.desc, bdev_io));
    3818             : 
    3819           4 :         bdev_io->u.bdev.memory_domain = NULL;
    3820           4 :         bdev_io->u.bdev.memory_domain_ctx = NULL;
    3821           8 :         _bdev_io_get_bounce_buf(bdev_io, _bdev_memory_domain_get_io_cb,
    3822           4 :                                 bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen);
    3823           4 : }
    3824             : 
    3825             : static inline void
    3826           0 : _bdev_io_ext_use_accel_buffer(struct spdk_bdev_io *bdev_io)
    3827             : {
    3828           0 :         assert(bdev_io_use_memory_domain(bdev_io));
    3829           0 :         assert(bdev_io_needs_metadata(bdev_io->internal.desc, bdev_io));
    3830             : 
    3831           0 :         bdev_io->u.bdev.memory_domain = NULL;
    3832           0 :         bdev_io->u.bdev.memory_domain_ctx = NULL;
    3833           0 :         bdev_io_get_accel_buf(bdev_io, _bdev_memory_domain_get_io_cb,
    3834           0 :                               bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen);
    3835           0 : }
    3836             : 
    3837             : /* We need to allocate bounce buffer
    3838             :  * - if bdev doesn't support memory domains,
    3839             :  * -  if it does support them, but we need to execute an accel sequence and the data buffer is
    3840             :  *    from accel memory domain (to avoid doing a push/pull from that domain), or
    3841             :  * - if IO is not aware of metadata.
    3842             :  */
    3843             : static inline bool
    3844         292 : bdev_io_needs_bounce_buffer(struct spdk_bdev_desc *desc, struct spdk_bdev_io *bdev_io)
    3845             : {
    3846         292 :         if (bdev_io_use_memory_domain(bdev_io)) {
    3847           4 :                 if (!desc->memory_domains_supported ||
    3848           0 :                     (bdev_io_needs_sequence_exec(desc, bdev_io) &&
    3849           0 :                      (bdev_io->internal.memory_domain == spdk_accel_get_memory_domain() ||
    3850           0 :                       bdev_io_needs_metadata(desc, bdev_io)))) {
    3851           4 :                         return true;
    3852             :                 }
    3853             : 
    3854           0 :                 return false;
    3855             :         }
    3856             : 
    3857         288 :         if (bdev_io_needs_metadata(desc, bdev_io)) {
    3858           0 :                 return true;
    3859             :         }
    3860             : 
    3861         288 :         return false;
    3862         292 : }
    3863             : 
    3864             : /* We need to allocate fake accel buffer if bdev supports memory domains but IO is not
    3865             :  * aware of metadata.
    3866             :  */
    3867             : static inline bool
    3868         288 : bdev_io_needs_accel_buffer(struct spdk_bdev_desc *desc, struct spdk_bdev_io *bdev_io)
    3869             : {
    3870         288 :         if (bdev_io_needs_metadata(desc, bdev_io)) {
    3871           0 :                 assert(bdev_io_use_memory_domain(bdev_io));
    3872           0 :                 return true;
    3873             :         }
    3874             : 
    3875         288 :         return false;
    3876         288 : }
    3877             : 
    3878             : static inline void
    3879         292 : _bdev_io_submit_ext(struct spdk_bdev_desc *desc, struct spdk_bdev_io *bdev_io)
    3880             : {
    3881         292 :         struct spdk_bdev_channel *ch = bdev_io->internal.ch;
    3882             :         int rc;
    3883             : 
    3884         292 :         if (spdk_unlikely(ch->flags & BDEV_CH_RESET_IN_PROGRESS)) {
    3885           0 :                 bdev_io->internal.status = SPDK_BDEV_IO_STATUS_ABORTED;
    3886           0 :                 bdev_io_complete_unsubmitted(bdev_io);
    3887           0 :                 return;
    3888             :         }
    3889             : 
    3890         292 :         if (bdev_io_needs_metadata(desc, bdev_io)) {
    3891           0 :                 rc = bdev_io_init_dif_ctx(bdev_io);
    3892           0 :                 if (spdk_unlikely(rc != 0)) {
    3893           0 :                         bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
    3894           0 :                         bdev_io_complete_unsubmitted(bdev_io);
    3895           0 :                         return;
    3896             :                 }
    3897           0 :         }
    3898             : 
    3899         292 :         if (bdev_io_needs_bounce_buffer(desc, bdev_io)) {
    3900           4 :                 _bdev_io_ext_use_bounce_buffer(bdev_io);
    3901           4 :                 return;
    3902             :         }
    3903             : 
    3904         288 :         if (bdev_io_needs_accel_buffer(desc, bdev_io)) {
    3905           0 :                 _bdev_io_ext_use_accel_buffer(bdev_io);
    3906           0 :                 return;
    3907             :         }
    3908             : 
    3909         288 :         if (bdev_io_needs_sequence_exec(desc, bdev_io)) {
    3910           0 :                 if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
    3911           0 :                         bdev_io_exec_sequence(bdev_io, bdev_io_submit_sequence_cb);
    3912           0 :                         return;
    3913             :                 }
    3914             :                 /* For reads we'll execute the sequence after the data is read, so, for now, only
    3915             :                  * clear out accel_sequence pointer and submit the IO */
    3916           0 :                 assert(bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
    3917           0 :                 bdev_io->u.bdev.accel_sequence = NULL;
    3918           0 :         }
    3919             : 
    3920         288 :         bdev_io_submit(bdev_io);
    3921         292 : }
    3922             : 
    3923             : static void
    3924          11 : bdev_io_submit_reset(struct spdk_bdev_io *bdev_io)
    3925             : {
    3926          11 :         struct spdk_bdev *bdev = bdev_io->bdev;
    3927          11 :         struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
    3928          11 :         struct spdk_io_channel *ch = bdev_ch->channel;
    3929             : 
    3930          11 :         assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_PENDING);
    3931             : 
    3932          11 :         bdev_io->internal.f.in_submit_request = true;
    3933          11 :         bdev_submit_request(bdev, ch, bdev_io);
    3934          11 :         bdev_io->internal.f.in_submit_request = false;
    3935          11 : }
    3936             : 
    3937             : void
    3938         691 : bdev_io_init(struct spdk_bdev_io *bdev_io,
    3939             :              struct spdk_bdev *bdev, void *cb_arg,
    3940             :              spdk_bdev_io_completion_cb cb)
    3941             : {
    3942         691 :         bdev_io->bdev = bdev;
    3943         691 :         bdev_io->internal.f.raw = 0;
    3944         691 :         bdev_io->internal.caller_ctx = cb_arg;
    3945         691 :         bdev_io->internal.cb = cb;
    3946         691 :         bdev_io->internal.status = SPDK_BDEV_IO_STATUS_PENDING;
    3947         691 :         bdev_io->internal.f.in_submit_request = false;
    3948         691 :         bdev_io->internal.error.nvme.cdw0 = 0;
    3949         691 :         bdev_io->num_retries = 0;
    3950         691 :         bdev_io->internal.get_buf_cb = NULL;
    3951         691 :         bdev_io->internal.get_aux_buf_cb = NULL;
    3952         691 :         bdev_io->internal.data_transfer_cpl = NULL;
    3953         691 :         bdev_io->internal.f.split = bdev_io_should_split(bdev_io);
    3954         691 : }
    3955             : 
    3956             : static bool
    3957         523 : bdev_io_type_supported(struct spdk_bdev *bdev, enum spdk_bdev_io_type io_type)
    3958             : {
    3959         523 :         return bdev->fn_table->io_type_supported(bdev->ctxt, io_type);
    3960             : }
    3961             : 
    3962             : bool
    3963         171 : spdk_bdev_io_type_supported(struct spdk_bdev *bdev, enum spdk_bdev_io_type io_type)
    3964             : {
    3965             :         bool supported;
    3966             : 
    3967         171 :         supported = bdev_io_type_supported(bdev, io_type);
    3968             : 
    3969         171 :         if (!supported) {
    3970           7 :                 switch (io_type) {
    3971             :                 case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
    3972             :                         /* The bdev layer will emulate write zeroes as long as write is supported. */
    3973           0 :                         supported = bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE);
    3974           0 :                         break;
    3975             :                 default:
    3976           7 :                         break;
    3977             :                 }
    3978           7 :         }
    3979             : 
    3980         171 :         return supported;
    3981             : }
    3982             : 
    3983             : static const char *g_io_type_strings[] = {
    3984             :         [SPDK_BDEV_IO_TYPE_READ] = "read",
    3985             :         [SPDK_BDEV_IO_TYPE_WRITE] = "write",
    3986             :         [SPDK_BDEV_IO_TYPE_UNMAP] = "unmap",
    3987             :         [SPDK_BDEV_IO_TYPE_FLUSH] = "flush",
    3988             :         [SPDK_BDEV_IO_TYPE_RESET] = "reset",
    3989             :         [SPDK_BDEV_IO_TYPE_NVME_ADMIN] = "nvme_admin",
    3990             :         [SPDK_BDEV_IO_TYPE_NVME_IO] = "nvme_io",
    3991             :         [SPDK_BDEV_IO_TYPE_NVME_IO_MD] = "nvme_io_md",
    3992             :         [SPDK_BDEV_IO_TYPE_WRITE_ZEROES] = "write_zeroes",
    3993             :         [SPDK_BDEV_IO_TYPE_ZCOPY] = "zcopy",
    3994             :         [SPDK_BDEV_IO_TYPE_GET_ZONE_INFO] = "get_zone_info",
    3995             :         [SPDK_BDEV_IO_TYPE_ZONE_MANAGEMENT] = "zone_management",
    3996             :         [SPDK_BDEV_IO_TYPE_ZONE_APPEND] = "zone_append",
    3997             :         [SPDK_BDEV_IO_TYPE_COMPARE] = "compare",
    3998             :         [SPDK_BDEV_IO_TYPE_COMPARE_AND_WRITE] = "compare_and_write",
    3999             :         [SPDK_BDEV_IO_TYPE_ABORT] = "abort",
    4000             :         [SPDK_BDEV_IO_TYPE_SEEK_HOLE] = "seek_hole",
    4001             :         [SPDK_BDEV_IO_TYPE_SEEK_DATA] = "seek_data",
    4002             :         [SPDK_BDEV_IO_TYPE_COPY] = "copy",
    4003             :         [SPDK_BDEV_IO_TYPE_NVME_IOV_MD] = "nvme_iov_md",
    4004             : };
    4005             : 
    4006             : const char *
    4007           0 : spdk_bdev_get_io_type_name(enum spdk_bdev_io_type io_type)
    4008             : {
    4009           0 :         if (io_type <= SPDK_BDEV_IO_TYPE_INVALID || io_type >= SPDK_BDEV_NUM_IO_TYPES) {
    4010           0 :                 return NULL;
    4011             :         }
    4012             : 
    4013           0 :         return g_io_type_strings[io_type];
    4014           0 : }
    4015             : 
    4016             : int
    4017           0 : spdk_bdev_get_io_type(const char *io_type_string)
    4018             : {
    4019             :         int i;
    4020             : 
    4021           0 :         for (i = SPDK_BDEV_IO_TYPE_READ; i < SPDK_BDEV_NUM_IO_TYPES; ++i) {
    4022           0 :                 if (!strcmp(io_type_string, g_io_type_strings[i])) {
    4023           0 :                         return i;
    4024             :                 }
    4025           0 :         }
    4026             : 
    4027           0 :         return -1;
    4028           0 : }
    4029             : 
    4030             : uint64_t
    4031           0 : spdk_bdev_io_get_submit_tsc(struct spdk_bdev_io *bdev_io)
    4032             : {
    4033           0 :         return bdev_io->internal.submit_tsc;
    4034             : }
    4035             : 
    4036             : bool
    4037           0 : spdk_bdev_io_has_no_metadata(struct spdk_bdev_io *bdev_io)
    4038             : {
    4039           0 :         return bdev_io->internal.desc->opts.no_metadata;
    4040             : }
    4041             : 
    4042             : int
    4043           0 : spdk_bdev_dump_info_json(struct spdk_bdev *bdev, struct spdk_json_write_ctx *w)
    4044             : {
    4045           0 :         if (bdev->fn_table->dump_info_json) {
    4046           0 :                 return bdev->fn_table->dump_info_json(bdev->ctxt, w);
    4047             :         }
    4048             : 
    4049           0 :         return 0;
    4050           0 : }
    4051             : 
    4052             : static void
    4053          10 : bdev_qos_update_max_quota_per_timeslice(struct spdk_bdev_qos *qos)
    4054             : {
    4055          10 :         uint32_t max_per_timeslice = 0;
    4056             :         int i;
    4057             : 
    4058          50 :         for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
    4059          40 :                 if (qos->rate_limits[i].limit == SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
    4060          15 :                         qos->rate_limits[i].max_per_timeslice = 0;
    4061          15 :                         continue;
    4062             :                 }
    4063             : 
    4064          50 :                 max_per_timeslice = qos->rate_limits[i].limit *
    4065          25 :                                     SPDK_BDEV_QOS_TIMESLICE_IN_USEC / SPDK_SEC_TO_USEC;
    4066             : 
    4067          25 :                 qos->rate_limits[i].max_per_timeslice = spdk_max(max_per_timeslice,
    4068             :                                                         qos->rate_limits[i].min_per_timeslice);
    4069             : 
    4070          50 :                 __atomic_store_n(&qos->rate_limits[i].remaining_this_timeslice,
    4071          25 :                                  qos->rate_limits[i].max_per_timeslice, __ATOMIC_RELEASE);
    4072          25 :         }
    4073             : 
    4074          10 :         bdev_qos_set_ops(qos);
    4075          10 : }
    4076             : 
    4077             : static void
    4078           4 : bdev_channel_submit_qos_io(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
    4079             :                            struct spdk_io_channel *io_ch, void *ctx)
    4080             : {
    4081           4 :         struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
    4082             :         int status;
    4083             : 
    4084           4 :         bdev_qos_io_submit(bdev_ch, bdev->internal.qos);
    4085             : 
    4086             :         /* if all IOs were sent then continue the iteration, otherwise - stop it */
    4087             :         /* TODO: channels round robing */
    4088           4 :         status = TAILQ_EMPTY(&bdev_ch->qos_queued_io) ? 0 : 1;
    4089             : 
    4090           4 :         spdk_bdev_for_each_channel_continue(i, status);
    4091           4 : }
    4092             : 
    4093             : 
    4094             : static void
    4095           2 : bdev_channel_submit_qos_io_done(struct spdk_bdev *bdev, void *ctx, int status)
    4096             : {
    4097             : 
    4098           2 : }
    4099             : 
    4100             : static int
    4101           3 : bdev_channel_poll_qos(void *arg)
    4102             : {
    4103           3 :         struct spdk_bdev *bdev = arg;
    4104           3 :         struct spdk_bdev_qos *qos = bdev->internal.qos;
    4105           3 :         uint64_t now = spdk_get_ticks();
    4106             :         int i;
    4107             :         int64_t remaining_last_timeslice;
    4108             : 
    4109           3 :         if (spdk_unlikely(qos->thread == NULL)) {
    4110             :                 /* Old QoS was unbound to remove and new QoS is not enabled yet. */
    4111           1 :                 return SPDK_POLLER_IDLE;
    4112             :         }
    4113             : 
    4114           2 :         if (now < (qos->last_timeslice + qos->timeslice_size)) {
    4115             :                 /* We received our callback earlier than expected - return
    4116             :                  *  immediately and wait to do accounting until at least one
    4117             :                  *  timeslice has actually expired.  This should never happen
    4118             :                  *  with a well-behaved timer implementation.
    4119             :                  */
    4120           0 :                 return SPDK_POLLER_IDLE;
    4121             :         }
    4122             : 
    4123             :         /* Reset for next round of rate limiting */
    4124          10 :         for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
    4125             :                 /* We may have allowed the IOs or bytes to slightly overrun in the last
    4126             :                  * timeslice. remaining_this_timeslice is signed, so if it's negative
    4127             :                  * here, we'll account for the overrun so that the next timeslice will
    4128             :                  * be appropriately reduced.
    4129             :                  */
    4130           8 :                 remaining_last_timeslice = __atomic_exchange_n(&qos->rate_limits[i].remaining_this_timeslice,
    4131             :                                            0, __ATOMIC_RELAXED);
    4132           8 :                 if (remaining_last_timeslice < 0) {
    4133             :                         /* There could be a race condition here as both bdev_qos_rw_queue_io() and bdev_channel_poll_qos()
    4134             :                          * potentially use 2 atomic ops each, so they can intertwine.
    4135             :                          * This race can potentially cause the limits to be a little fuzzy but won't cause any real damage.
    4136             :                          */
    4137           0 :                         __atomic_store_n(&qos->rate_limits[i].remaining_this_timeslice,
    4138           0 :                                          remaining_last_timeslice, __ATOMIC_RELAXED);
    4139           0 :                 }
    4140           8 :         }
    4141             : 
    4142           4 :         while (now >= (qos->last_timeslice + qos->timeslice_size)) {
    4143           2 :                 qos->last_timeslice += qos->timeslice_size;
    4144          10 :                 for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
    4145          16 :                         __atomic_add_fetch(&qos->rate_limits[i].remaining_this_timeslice,
    4146           8 :                                            qos->rate_limits[i].max_per_timeslice, __ATOMIC_RELAXED);
    4147           8 :                 }
    4148             :         }
    4149             : 
    4150           2 :         spdk_bdev_for_each_channel(bdev, bdev_channel_submit_qos_io, qos,
    4151             :                                    bdev_channel_submit_qos_io_done);
    4152             : 
    4153           2 :         return SPDK_POLLER_BUSY;
    4154           3 : }
    4155             : 
    4156             : static void
    4157          74 : bdev_channel_destroy_resource(struct spdk_bdev_channel *ch)
    4158             : {
    4159             :         struct spdk_bdev_shared_resource *shared_resource;
    4160             :         struct lba_range *range;
    4161             : 
    4162          74 :         bdev_free_io_stat(ch->stat);
    4163             : #ifdef SPDK_CONFIG_VTUNE
    4164             :         bdev_free_io_stat(ch->prev_stat);
    4165             : #endif
    4166             : 
    4167          74 :         while (!TAILQ_EMPTY(&ch->locked_ranges)) {
    4168           0 :                 range = TAILQ_FIRST(&ch->locked_ranges);
    4169           0 :                 TAILQ_REMOVE(&ch->locked_ranges, range, tailq);
    4170           0 :                 free(range);
    4171             :         }
    4172             : 
    4173          74 :         spdk_put_io_channel(ch->channel);
    4174          74 :         spdk_put_io_channel(ch->accel_channel);
    4175             : 
    4176          74 :         shared_resource = ch->shared_resource;
    4177             : 
    4178          74 :         assert(TAILQ_EMPTY(&ch->io_locked));
    4179          74 :         assert(TAILQ_EMPTY(&ch->io_submitted));
    4180          74 :         assert(TAILQ_EMPTY(&ch->io_accel_exec));
    4181          74 :         assert(TAILQ_EMPTY(&ch->io_memory_domain));
    4182          74 :         assert(ch->io_outstanding == 0);
    4183          74 :         assert(shared_resource->ref > 0);
    4184          74 :         shared_resource->ref--;
    4185          74 :         if (shared_resource->ref == 0) {
    4186          73 :                 assert(shared_resource->io_outstanding == 0);
    4187          73 :                 TAILQ_REMOVE(&shared_resource->mgmt_ch->shared_resources, shared_resource, link);
    4188          73 :                 spdk_put_io_channel(spdk_io_channel_from_ctx(shared_resource->mgmt_ch));
    4189          73 :                 spdk_poller_unregister(&shared_resource->nomem_poller);
    4190          73 :                 free(shared_resource);
    4191          73 :         }
    4192          74 : }
    4193             : 
    4194             : static void
    4195          83 : bdev_enable_qos(struct spdk_bdev *bdev, struct spdk_bdev_channel *ch)
    4196             : {
    4197          83 :         struct spdk_bdev_qos    *qos = bdev->internal.qos;
    4198             :         int                     i;
    4199             : 
    4200          83 :         assert(spdk_spin_held(&bdev->internal.spinlock));
    4201             : 
    4202             :         /* Rate limiting on this bdev enabled */
    4203          83 :         if (qos) {
    4204          17 :                 if (qos->ch == NULL) {
    4205             :                         struct spdk_io_channel *io_ch;
    4206             : 
    4207           9 :                         SPDK_DEBUGLOG(bdev, "Selecting channel %p as QoS channel for bdev %s on thread %p\n", ch,
    4208             :                                       bdev->name, spdk_get_thread());
    4209             : 
    4210             :                         /* No qos channel has been selected, so set one up */
    4211             : 
    4212             :                         /* Take another reference to ch */
    4213           9 :                         io_ch = spdk_get_io_channel(__bdev_to_io_dev(bdev));
    4214           9 :                         assert(io_ch != NULL);
    4215           9 :                         qos->ch = ch;
    4216             : 
    4217           9 :                         qos->thread = spdk_io_channel_get_thread(io_ch);
    4218             : 
    4219          45 :                         for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
    4220          36 :                                 if (bdev_qos_is_iops_rate_limit(i) == true) {
    4221           9 :                                         qos->rate_limits[i].min_per_timeslice =
    4222             :                                                 SPDK_BDEV_QOS_MIN_IO_PER_TIMESLICE;
    4223           9 :                                 } else {
    4224          27 :                                         qos->rate_limits[i].min_per_timeslice =
    4225             :                                                 SPDK_BDEV_QOS_MIN_BYTE_PER_TIMESLICE;
    4226             :                                 }
    4227             : 
    4228          36 :                                 if (qos->rate_limits[i].limit == 0) {
    4229           2 :                                         qos->rate_limits[i].limit = SPDK_BDEV_QOS_LIMIT_NOT_DEFINED;
    4230           2 :                                 }
    4231          36 :                         }
    4232           9 :                         bdev_qos_update_max_quota_per_timeslice(qos);
    4233           9 :                         qos->timeslice_size =
    4234           9 :                                 SPDK_BDEV_QOS_TIMESLICE_IN_USEC * spdk_get_ticks_hz() / SPDK_SEC_TO_USEC;
    4235           9 :                         qos->last_timeslice = spdk_get_ticks();
    4236           9 :                         qos->poller = SPDK_POLLER_REGISTER(bdev_channel_poll_qos,
    4237             :                                                            bdev,
    4238             :                                                            SPDK_BDEV_QOS_TIMESLICE_IN_USEC);
    4239           9 :                 }
    4240             : 
    4241          17 :                 ch->flags |= BDEV_CH_QOS_ENABLED;
    4242          17 :         }
    4243          83 : }
    4244             : 
    4245             : struct poll_timeout_ctx {
    4246             :         struct spdk_bdev_desc   *desc;
    4247             :         uint64_t                timeout_in_sec;
    4248             :         spdk_bdev_io_timeout_cb cb_fn;
    4249             :         void                    *cb_arg;
    4250             : };
    4251             : 
    4252             : static void
    4253         268 : bdev_desc_free(struct spdk_bdev_desc *desc)
    4254             : {
    4255         268 :         spdk_spin_destroy(&desc->spinlock);
    4256         268 :         free(desc->media_events_buffer);
    4257         268 :         free(desc);
    4258         268 : }
    4259             : 
    4260             : static void
    4261           8 : bdev_channel_poll_timeout_io_done(struct spdk_bdev *bdev, void *_ctx, int status)
    4262             : {
    4263           8 :         struct poll_timeout_ctx *ctx  = _ctx;
    4264           8 :         struct spdk_bdev_desc *desc = ctx->desc;
    4265             : 
    4266           8 :         free(ctx);
    4267             : 
    4268           8 :         spdk_spin_lock(&desc->spinlock);
    4269           8 :         desc->refs--;
    4270           8 :         if (desc->closed == true && desc->refs == 0) {
    4271           1 :                 spdk_spin_unlock(&desc->spinlock);
    4272           1 :                 bdev_desc_free(desc);
    4273           1 :                 return;
    4274             :         }
    4275           7 :         spdk_spin_unlock(&desc->spinlock);
    4276           8 : }
    4277             : 
    4278             : static void
    4279          13 : bdev_channel_poll_timeout_io(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
    4280             :                              struct spdk_io_channel *io_ch, void *_ctx)
    4281             : {
    4282          13 :         struct poll_timeout_ctx *ctx  = _ctx;
    4283          13 :         struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
    4284          13 :         struct spdk_bdev_desc *desc = ctx->desc;
    4285             :         struct spdk_bdev_io *bdev_io;
    4286             :         uint64_t now;
    4287             : 
    4288          13 :         spdk_spin_lock(&desc->spinlock);
    4289          13 :         if (desc->closed == true) {
    4290           1 :                 spdk_spin_unlock(&desc->spinlock);
    4291           1 :                 spdk_bdev_for_each_channel_continue(i, -1);
    4292           1 :                 return;
    4293             :         }
    4294          12 :         spdk_spin_unlock(&desc->spinlock);
    4295             : 
    4296          12 :         now = spdk_get_ticks();
    4297          22 :         TAILQ_FOREACH(bdev_io, &bdev_ch->io_submitted, internal.ch_link) {
    4298             :                 /* Exclude any I/O that are generated via splitting. */
    4299          15 :                 if (bdev_io->internal.cb == bdev_io_split_done) {
    4300           3 :                         continue;
    4301             :                 }
    4302             : 
    4303             :                 /* Once we find an I/O that has not timed out, we can immediately
    4304             :                  * exit the loop.
    4305             :                  */
    4306          24 :                 if (now < (bdev_io->internal.submit_tsc +
    4307          12 :                            ctx->timeout_in_sec * spdk_get_ticks_hz())) {
    4308           5 :                         goto end;
    4309             :                 }
    4310             : 
    4311           7 :                 if (bdev_io->internal.desc == desc) {
    4312           7 :                         ctx->cb_fn(ctx->cb_arg, bdev_io);
    4313           7 :                 }
    4314          14 :         }
    4315             : 
    4316             : end:
    4317          12 :         spdk_bdev_for_each_channel_continue(i, 0);
    4318          13 : }
    4319             : 
    4320             : static int
    4321           8 : bdev_poll_timeout_io(void *arg)
    4322             : {
    4323           8 :         struct spdk_bdev_desc *desc = arg;
    4324           8 :         struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
    4325             :         struct poll_timeout_ctx *ctx;
    4326             : 
    4327           8 :         ctx = calloc(1, sizeof(struct poll_timeout_ctx));
    4328           8 :         if (!ctx) {
    4329           0 :                 SPDK_ERRLOG("failed to allocate memory\n");
    4330           0 :                 return SPDK_POLLER_BUSY;
    4331             :         }
    4332           8 :         ctx->desc = desc;
    4333           8 :         ctx->cb_arg = desc->cb_arg;
    4334           8 :         ctx->cb_fn = desc->cb_fn;
    4335           8 :         ctx->timeout_in_sec = desc->timeout_in_sec;
    4336             : 
    4337             :         /* Take a ref on the descriptor in case it gets closed while we are checking
    4338             :          * all of the channels.
    4339             :          */
    4340           8 :         spdk_spin_lock(&desc->spinlock);
    4341           8 :         desc->refs++;
    4342           8 :         spdk_spin_unlock(&desc->spinlock);
    4343             : 
    4344           8 :         spdk_bdev_for_each_channel(bdev, bdev_channel_poll_timeout_io, ctx,
    4345             :                                    bdev_channel_poll_timeout_io_done);
    4346             : 
    4347           8 :         return SPDK_POLLER_BUSY;
    4348           8 : }
    4349             : 
    4350             : int
    4351           5 : spdk_bdev_set_timeout(struct spdk_bdev_desc *desc, uint64_t timeout_in_sec,
    4352             :                       spdk_bdev_io_timeout_cb cb_fn, void *cb_arg)
    4353             : {
    4354           5 :         assert(desc->thread == spdk_get_thread());
    4355             : 
    4356           5 :         spdk_poller_unregister(&desc->io_timeout_poller);
    4357             : 
    4358           5 :         if (timeout_in_sec) {
    4359           4 :                 assert(cb_fn != NULL);
    4360           4 :                 desc->io_timeout_poller = SPDK_POLLER_REGISTER(bdev_poll_timeout_io,
    4361             :                                           desc,
    4362             :                                           SPDK_BDEV_IO_POLL_INTERVAL_IN_MSEC * SPDK_SEC_TO_USEC /
    4363             :                                           1000);
    4364           4 :                 if (desc->io_timeout_poller == NULL) {
    4365           0 :                         SPDK_ERRLOG("can not register the desc timeout IO poller\n");
    4366           0 :                         return -1;
    4367             :                 }
    4368           4 :         }
    4369             : 
    4370           5 :         desc->cb_fn = cb_fn;
    4371           5 :         desc->cb_arg = cb_arg;
    4372           5 :         desc->timeout_in_sec = timeout_in_sec;
    4373             : 
    4374           5 :         return 0;
    4375           5 : }
    4376             : 
    4377             : static int
    4378          76 : bdev_channel_create(void *io_device, void *ctx_buf)
    4379             : {
    4380          76 :         struct spdk_bdev                *bdev = __bdev_from_io_dev(io_device);
    4381          76 :         struct spdk_bdev_channel        *ch = ctx_buf;
    4382             :         struct spdk_io_channel          *mgmt_io_ch;
    4383             :         struct spdk_bdev_mgmt_channel   *mgmt_ch;
    4384             :         struct spdk_bdev_shared_resource *shared_resource;
    4385             :         struct lba_range                *range;
    4386             : 
    4387          76 :         ch->bdev = bdev;
    4388          76 :         ch->channel = bdev->fn_table->get_io_channel(bdev->ctxt);
    4389          76 :         if (!ch->channel) {
    4390           2 :                 return -1;
    4391             :         }
    4392             : 
    4393          74 :         ch->accel_channel = spdk_accel_get_io_channel();
    4394          74 :         if (!ch->accel_channel) {
    4395           0 :                 spdk_put_io_channel(ch->channel);
    4396           0 :                 return -1;
    4397             :         }
    4398             : 
    4399          74 :         spdk_trace_record(TRACE_BDEV_IOCH_CREATE, bdev->internal.trace_id, 0, 0,
    4400             :                           spdk_thread_get_id(spdk_io_channel_get_thread(ch->channel)));
    4401             : 
    4402          74 :         assert(ch->histogram == NULL);
    4403          74 :         if (bdev->internal.histogram_enabled) {
    4404           0 :                 ch->histogram = spdk_histogram_data_alloc();
    4405           0 :                 if (ch->histogram == NULL) {
    4406           0 :                         SPDK_ERRLOG("Could not allocate histogram\n");
    4407           0 :                 }
    4408           0 :         }
    4409             : 
    4410          74 :         mgmt_io_ch = spdk_get_io_channel(&g_bdev_mgr);
    4411          74 :         if (!mgmt_io_ch) {
    4412           0 :                 spdk_put_io_channel(ch->channel);
    4413           0 :                 spdk_put_io_channel(ch->accel_channel);
    4414           0 :                 return -1;
    4415             :         }
    4416             : 
    4417          74 :         mgmt_ch = __io_ch_to_bdev_mgmt_ch(mgmt_io_ch);
    4418          76 :         TAILQ_FOREACH(shared_resource, &mgmt_ch->shared_resources, link) {
    4419           3 :                 if (shared_resource->shared_ch == ch->channel) {
    4420           1 :                         spdk_put_io_channel(mgmt_io_ch);
    4421           1 :                         shared_resource->ref++;
    4422           1 :                         break;
    4423             :                 }
    4424           2 :         }
    4425             : 
    4426          74 :         if (shared_resource == NULL) {
    4427          73 :                 shared_resource = calloc(1, sizeof(*shared_resource));
    4428          73 :                 if (shared_resource == NULL) {
    4429           0 :                         spdk_put_io_channel(ch->channel);
    4430           0 :                         spdk_put_io_channel(ch->accel_channel);
    4431           0 :                         spdk_put_io_channel(mgmt_io_ch);
    4432           0 :                         return -1;
    4433             :                 }
    4434             : 
    4435          73 :                 shared_resource->mgmt_ch = mgmt_ch;
    4436          73 :                 shared_resource->io_outstanding = 0;
    4437          73 :                 TAILQ_INIT(&shared_resource->nomem_io);
    4438          73 :                 shared_resource->nomem_threshold = 0;
    4439          73 :                 shared_resource->shared_ch = ch->channel;
    4440          73 :                 shared_resource->ref = 1;
    4441          73 :                 TAILQ_INSERT_TAIL(&mgmt_ch->shared_resources, shared_resource, link);
    4442          73 :         }
    4443             : 
    4444          74 :         ch->io_outstanding = 0;
    4445          74 :         TAILQ_INIT(&ch->queued_resets);
    4446          74 :         TAILQ_INIT(&ch->locked_ranges);
    4447          74 :         TAILQ_INIT(&ch->qos_queued_io);
    4448          74 :         ch->flags = 0;
    4449          74 :         ch->trace_id = bdev->internal.trace_id;
    4450          74 :         ch->shared_resource = shared_resource;
    4451             : 
    4452          74 :         TAILQ_INIT(&ch->io_submitted);
    4453          74 :         TAILQ_INIT(&ch->io_locked);
    4454          74 :         TAILQ_INIT(&ch->io_accel_exec);
    4455          74 :         TAILQ_INIT(&ch->io_memory_domain);
    4456             : 
    4457          74 :         ch->stat = bdev_alloc_io_stat(false);
    4458          74 :         if (ch->stat == NULL) {
    4459           0 :                 bdev_channel_destroy_resource(ch);
    4460           0 :                 return -1;
    4461             :         }
    4462             : 
    4463          74 :         ch->stat->ticks_rate = spdk_get_ticks_hz();
    4464             : 
    4465             : #ifdef SPDK_CONFIG_VTUNE
    4466             :         {
    4467             :                 char *name;
    4468             :                 __itt_init_ittlib(NULL, 0);
    4469             :                 name = spdk_sprintf_alloc("spdk_bdev_%s_%p", ch->bdev->name, ch);
    4470             :                 if (!name) {
    4471             :                         bdev_channel_destroy_resource(ch);
    4472             :                         return -1;
    4473             :                 }
    4474             :                 ch->handle = __itt_string_handle_create(name);
    4475             :                 free(name);
    4476             :                 ch->start_tsc = spdk_get_ticks();
    4477             :                 ch->interval_tsc = spdk_get_ticks_hz() / 100;
    4478             :                 ch->prev_stat = bdev_alloc_io_stat(false);
    4479             :                 if (ch->prev_stat == NULL) {
    4480             :                         bdev_channel_destroy_resource(ch);
    4481             :                         return -1;
    4482             :                 }
    4483             :         }
    4484             : #endif
    4485             : 
    4486          74 :         spdk_spin_lock(&bdev->internal.spinlock);
    4487          74 :         bdev_enable_qos(bdev, ch);
    4488             : 
    4489          75 :         TAILQ_FOREACH(range, &bdev->internal.locked_ranges, tailq) {
    4490             :                 struct lba_range *new_range;
    4491             : 
    4492           1 :                 new_range = calloc(1, sizeof(*new_range));
    4493           1 :                 if (new_range == NULL) {
    4494           0 :                         spdk_spin_unlock(&bdev->internal.spinlock);
    4495           0 :                         bdev_channel_destroy_resource(ch);
    4496           0 :                         return -1;
    4497             :                 }
    4498           1 :                 new_range->length = range->length;
    4499           1 :                 new_range->offset = range->offset;
    4500           1 :                 new_range->locked_ctx = range->locked_ctx;
    4501           1 :                 TAILQ_INSERT_TAIL(&ch->locked_ranges, new_range, tailq);
    4502           1 :         }
    4503             : 
    4504          74 :         spdk_spin_unlock(&bdev->internal.spinlock);
    4505             : 
    4506          74 :         return 0;
    4507          76 : }
    4508             : 
    4509             : static int
    4510           0 : bdev_abort_all_buf_io_cb(struct spdk_iobuf_channel *ch, struct spdk_iobuf_entry *entry,
    4511             :                          void *cb_ctx)
    4512             : {
    4513           0 :         struct spdk_bdev_channel *bdev_ch = cb_ctx;
    4514             :         struct spdk_bdev_io *bdev_io;
    4515             :         uint64_t buf_len;
    4516             : 
    4517           0 :         bdev_io = SPDK_CONTAINEROF(entry, struct spdk_bdev_io, internal.iobuf);
    4518           0 :         if (bdev_io->internal.ch == bdev_ch) {
    4519           0 :                 buf_len = bdev_io_get_max_buf_len(bdev_io, bdev_io->internal.buf.len);
    4520           0 :                 spdk_iobuf_entry_abort(ch, entry, buf_len);
    4521           0 :                 spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_ABORTED);
    4522           0 :         }
    4523             : 
    4524           0 :         return 0;
    4525             : }
    4526             : 
    4527             : /*
    4528             :  * Abort I/O that are waiting on a data buffer.
    4529             :  */
    4530             : static void
    4531          96 : bdev_abort_all_buf_io(struct spdk_bdev_mgmt_channel *mgmt_ch, struct spdk_bdev_channel *ch)
    4532             : {
    4533          96 :         spdk_iobuf_for_each_entry(&mgmt_ch->iobuf, bdev_abort_all_buf_io_cb, ch);
    4534          96 : }
    4535             : 
    4536             : /*
    4537             :  * Abort I/O that are queued waiting for submission.  These types of I/O are
    4538             :  *  linked using the spdk_bdev_io link TAILQ_ENTRY.
    4539             :  */
    4540             : static void
    4541         188 : bdev_abort_all_queued_io(bdev_io_tailq_t *queue, struct spdk_bdev_channel *ch)
    4542             : {
    4543             :         struct spdk_bdev_io *bdev_io, *tmp;
    4544             : 
    4545         228 :         TAILQ_FOREACH_SAFE(bdev_io, queue, internal.link, tmp) {
    4546          40 :                 if (bdev_io->internal.ch == ch) {
    4547          40 :                         TAILQ_REMOVE(queue, bdev_io, internal.link);
    4548             :                         /*
    4549             :                          * spdk_bdev_io_complete() assumes that the completed I/O had
    4550             :                          *  been submitted to the bdev module.  Since in this case it
    4551             :                          *  hadn't, bump io_outstanding to account for the decrement
    4552             :                          *  that spdk_bdev_io_complete() will do.
    4553             :                          */
    4554          40 :                         if (bdev_io->type != SPDK_BDEV_IO_TYPE_RESET) {
    4555          39 :                                 bdev_io_increment_outstanding(ch, ch->shared_resource);
    4556          39 :                         }
    4557          40 :                         spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_ABORTED);
    4558          40 :                 }
    4559          40 :         }
    4560         188 : }
    4561             : 
    4562             : static bool
    4563          18 : bdev_abort_queued_io(bdev_io_tailq_t *queue, struct spdk_bdev_io *bio_to_abort)
    4564             : {
    4565             :         struct spdk_bdev_io *bdev_io;
    4566             : 
    4567          18 :         TAILQ_FOREACH(bdev_io, queue, internal.link) {
    4568           0 :                 if (bdev_io == bio_to_abort) {
    4569           0 :                         TAILQ_REMOVE(queue, bio_to_abort, internal.link);
    4570           0 :                         spdk_bdev_io_complete(bio_to_abort, SPDK_BDEV_IO_STATUS_ABORTED);
    4571           0 :                         return true;
    4572             :                 }
    4573           0 :         }
    4574             : 
    4575          18 :         return false;
    4576          18 : }
    4577             : 
    4578             : static int
    4579           0 : bdev_abort_buf_io_cb(struct spdk_iobuf_channel *ch, struct spdk_iobuf_entry *entry, void *cb_ctx)
    4580             : {
    4581           0 :         struct spdk_bdev_io *bdev_io, *bio_to_abort = cb_ctx;
    4582             :         uint64_t buf_len;
    4583             : 
    4584           0 :         bdev_io = SPDK_CONTAINEROF(entry, struct spdk_bdev_io, internal.iobuf);
    4585           0 :         if (bdev_io == bio_to_abort) {
    4586           0 :                 buf_len = bdev_io_get_max_buf_len(bdev_io, bdev_io->internal.buf.len);
    4587           0 :                 spdk_iobuf_entry_abort(ch, entry, buf_len);
    4588           0 :                 spdk_bdev_io_complete(bio_to_abort, SPDK_BDEV_IO_STATUS_ABORTED);
    4589           0 :                 return 1;
    4590             :         }
    4591             : 
    4592           0 :         return 0;
    4593           0 : }
    4594             : 
    4595             : static bool
    4596          16 : bdev_abort_buf_io(struct spdk_bdev_mgmt_channel *mgmt_ch, struct spdk_bdev_io *bio_to_abort)
    4597             : {
    4598             :         int rc;
    4599             : 
    4600          16 :         rc = spdk_iobuf_for_each_entry(&mgmt_ch->iobuf, bdev_abort_buf_io_cb, bio_to_abort);
    4601          16 :         return rc == 1;
    4602             : }
    4603             : 
    4604             : static void
    4605           7 : bdev_qos_channel_destroy(void *cb_arg)
    4606             : {
    4607           7 :         struct spdk_bdev_qos *qos = cb_arg;
    4608             : 
    4609           7 :         spdk_put_io_channel(spdk_io_channel_from_ctx(qos->ch));
    4610           7 :         spdk_poller_unregister(&qos->poller);
    4611             : 
    4612           7 :         SPDK_DEBUGLOG(bdev, "Free QoS %p.\n", qos);
    4613             : 
    4614           7 :         free(qos);
    4615           7 : }
    4616             : 
    4617             : static int
    4618           7 : bdev_qos_destroy(struct spdk_bdev *bdev)
    4619             : {
    4620             :         int i;
    4621             : 
    4622             :         /*
    4623             :          * Cleanly shutting down the QoS poller is tricky, because
    4624             :          * during the asynchronous operation the user could open
    4625             :          * a new descriptor and create a new channel, spawning
    4626             :          * a new QoS poller.
    4627             :          *
    4628             :          * The strategy is to create a new QoS structure here and swap it
    4629             :          * in. The shutdown path then continues to refer to the old one
    4630             :          * until it completes and then releases it.
    4631             :          */
    4632             :         struct spdk_bdev_qos *new_qos, *old_qos;
    4633             : 
    4634           7 :         old_qos = bdev->internal.qos;
    4635             : 
    4636           7 :         new_qos = calloc(1, sizeof(*new_qos));
    4637           7 :         if (!new_qos) {
    4638           0 :                 SPDK_ERRLOG("Unable to allocate memory to shut down QoS.\n");
    4639           0 :                 return -ENOMEM;
    4640             :         }
    4641             : 
    4642             :         /* Copy the old QoS data into the newly allocated structure */
    4643           7 :         memcpy(new_qos, old_qos, sizeof(*new_qos));
    4644             : 
    4645             :         /* Zero out the key parts of the QoS structure */
    4646           7 :         new_qos->ch = NULL;
    4647           7 :         new_qos->thread = NULL;
    4648           7 :         new_qos->poller = NULL;
    4649             :         /*
    4650             :          * The limit member of spdk_bdev_qos_limit structure is not zeroed.
    4651             :          * It will be used later for the new QoS structure.
    4652             :          */
    4653          35 :         for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
    4654          28 :                 new_qos->rate_limits[i].remaining_this_timeslice = 0;
    4655          28 :                 new_qos->rate_limits[i].min_per_timeslice = 0;
    4656          28 :                 new_qos->rate_limits[i].max_per_timeslice = 0;
    4657          28 :         }
    4658             : 
    4659           7 :         bdev->internal.qos = new_qos;
    4660             : 
    4661           7 :         if (old_qos->thread == NULL) {
    4662           0 :                 free(old_qos);
    4663           0 :         } else {
    4664           7 :                 spdk_thread_send_msg(old_qos->thread, bdev_qos_channel_destroy, old_qos);
    4665             :         }
    4666             : 
    4667             :         /* It is safe to continue with destroying the bdev even though the QoS channel hasn't
    4668             :          * been destroyed yet. The destruction path will end up waiting for the final
    4669             :          * channel to be put before it releases resources. */
    4670             : 
    4671           7 :         return 0;
    4672           7 : }
    4673             : 
    4674             : void
    4675          78 : spdk_bdev_add_io_stat(struct spdk_bdev_io_stat *total, struct spdk_bdev_io_stat *add)
    4676             : {
    4677          78 :         total->bytes_read += add->bytes_read;
    4678          78 :         total->num_read_ops += add->num_read_ops;
    4679          78 :         total->bytes_written += add->bytes_written;
    4680          78 :         total->num_write_ops += add->num_write_ops;
    4681          78 :         total->bytes_unmapped += add->bytes_unmapped;
    4682          78 :         total->num_unmap_ops += add->num_unmap_ops;
    4683          78 :         total->bytes_copied += add->bytes_copied;
    4684          78 :         total->num_copy_ops += add->num_copy_ops;
    4685          78 :         total->read_latency_ticks += add->read_latency_ticks;
    4686          78 :         total->write_latency_ticks += add->write_latency_ticks;
    4687          78 :         total->unmap_latency_ticks += add->unmap_latency_ticks;
    4688          78 :         total->copy_latency_ticks += add->copy_latency_ticks;
    4689          78 :         if (total->max_read_latency_ticks < add->max_read_latency_ticks) {
    4690           7 :                 total->max_read_latency_ticks = add->max_read_latency_ticks;
    4691           7 :         }
    4692          78 :         if (total->min_read_latency_ticks > add->min_read_latency_ticks) {
    4693          39 :                 total->min_read_latency_ticks = add->min_read_latency_ticks;
    4694          39 :         }
    4695          78 :         if (total->max_write_latency_ticks < add->max_write_latency_ticks) {
    4696           4 :                 total->max_write_latency_ticks = add->max_write_latency_ticks;
    4697           4 :         }
    4698          78 :         if (total->min_write_latency_ticks > add->min_write_latency_ticks) {
    4699          24 :                 total->min_write_latency_ticks = add->min_write_latency_ticks;
    4700          24 :         }
    4701          78 :         if (total->max_unmap_latency_ticks < add->max_unmap_latency_ticks) {
    4702           0 :                 total->max_unmap_latency_ticks = add->max_unmap_latency_ticks;
    4703           0 :         }
    4704          78 :         if (total->min_unmap_latency_ticks > add->min_unmap_latency_ticks) {
    4705           3 :                 total->min_unmap_latency_ticks = add->min_unmap_latency_ticks;
    4706           3 :         }
    4707          78 :         if (total->max_copy_latency_ticks < add->max_copy_latency_ticks) {
    4708           0 :                 total->max_copy_latency_ticks = add->max_copy_latency_ticks;
    4709           0 :         }
    4710          78 :         if (total->min_copy_latency_ticks > add->min_copy_latency_ticks) {
    4711           4 :                 total->min_copy_latency_ticks = add->min_copy_latency_ticks;
    4712           4 :         }
    4713          78 : }
    4714             : 
    4715             : static void
    4716           5 : bdev_get_io_stat(struct spdk_bdev_io_stat *to_stat, struct spdk_bdev_io_stat *from_stat)
    4717             : {
    4718           5 :         memcpy(to_stat, from_stat, offsetof(struct spdk_bdev_io_stat, io_error));
    4719             : 
    4720           5 :         if (to_stat->io_error != NULL && from_stat->io_error != NULL) {
    4721           0 :                 memcpy(to_stat->io_error, from_stat->io_error,
    4722             :                        sizeof(struct spdk_bdev_io_error_stat));
    4723           0 :         }
    4724           5 : }
    4725             : 
    4726             : void
    4727         211 : spdk_bdev_reset_io_stat(struct spdk_bdev_io_stat *stat, enum spdk_bdev_reset_stat_mode mode)
    4728             : {
    4729         211 :         if (mode == SPDK_BDEV_RESET_STAT_NONE) {
    4730           5 :                 return;
    4731             :         }
    4732             : 
    4733         206 :         stat->max_read_latency_ticks = 0;
    4734         206 :         stat->min_read_latency_ticks = UINT64_MAX;
    4735         206 :         stat->max_write_latency_ticks = 0;
    4736         206 :         stat->min_write_latency_ticks = UINT64_MAX;
    4737         206 :         stat->max_unmap_latency_ticks = 0;
    4738         206 :         stat->min_unmap_latency_ticks = UINT64_MAX;
    4739         206 :         stat->max_copy_latency_ticks = 0;
    4740         206 :         stat->min_copy_latency_ticks = UINT64_MAX;
    4741             : 
    4742         206 :         if (mode != SPDK_BDEV_RESET_STAT_ALL) {
    4743           2 :                 return;
    4744             :         }
    4745             : 
    4746         204 :         stat->bytes_read = 0;
    4747         204 :         stat->num_read_ops = 0;
    4748         204 :         stat->bytes_written = 0;
    4749         204 :         stat->num_write_ops = 0;
    4750         204 :         stat->bytes_unmapped = 0;
    4751         204 :         stat->num_unmap_ops = 0;
    4752         204 :         stat->bytes_copied = 0;
    4753         204 :         stat->num_copy_ops = 0;
    4754         204 :         stat->read_latency_ticks = 0;
    4755         204 :         stat->write_latency_ticks = 0;
    4756         204 :         stat->unmap_latency_ticks = 0;
    4757         204 :         stat->copy_latency_ticks = 0;
    4758             : 
    4759         204 :         if (stat->io_error != NULL) {
    4760         129 :                 memset(stat->io_error, 0, sizeof(struct spdk_bdev_io_error_stat));
    4761         129 :         }
    4762         211 : }
    4763             : 
    4764             : struct spdk_bdev_io_stat *
    4765         202 : bdev_alloc_io_stat(bool io_error_stat)
    4766             : {
    4767             :         struct spdk_bdev_io_stat *stat;
    4768             : 
    4769         202 :         stat = malloc(sizeof(struct spdk_bdev_io_stat));
    4770         202 :         if (stat == NULL) {
    4771           0 :                 return NULL;
    4772             :         }
    4773             : 
    4774         202 :         if (io_error_stat) {
    4775         128 :                 stat->io_error = malloc(sizeof(struct spdk_bdev_io_error_stat));
    4776         128 :                 if (stat->io_error == NULL) {
    4777           0 :                         free(stat);
    4778           0 :                         return NULL;
    4779             :                 }
    4780         128 :         } else {
    4781          74 :                 stat->io_error = NULL;
    4782             :         }
    4783             : 
    4784         202 :         spdk_bdev_reset_io_stat(stat, SPDK_BDEV_RESET_STAT_ALL);
    4785             : 
    4786         202 :         return stat;
    4787         202 : }
    4788             : 
    4789             : void
    4790         202 : bdev_free_io_stat(struct spdk_bdev_io_stat *stat)
    4791             : {
    4792         202 :         if (stat != NULL) {
    4793         202 :                 free(stat->io_error);
    4794         202 :                 free(stat);
    4795         202 :         }
    4796         202 : }
    4797             : 
    4798             : void
    4799           0 : spdk_bdev_dump_io_stat_json(struct spdk_bdev_io_stat *stat, struct spdk_json_write_ctx *w)
    4800             : {
    4801             :         int i;
    4802             : 
    4803           0 :         spdk_json_write_named_uint64(w, "bytes_read", stat->bytes_read);
    4804           0 :         spdk_json_write_named_uint64(w, "num_read_ops", stat->num_read_ops);
    4805           0 :         spdk_json_write_named_uint64(w, "bytes_written", stat->bytes_written);
    4806           0 :         spdk_json_write_named_uint64(w, "num_write_ops", stat->num_write_ops);
    4807           0 :         spdk_json_write_named_uint64(w, "bytes_unmapped", stat->bytes_unmapped);
    4808           0 :         spdk_json_write_named_uint64(w, "num_unmap_ops", stat->num_unmap_ops);
    4809           0 :         spdk_json_write_named_uint64(w, "bytes_copied", stat->bytes_copied);
    4810           0 :         spdk_json_write_named_uint64(w, "num_copy_ops", stat->num_copy_ops);
    4811           0 :         spdk_json_write_named_uint64(w, "read_latency_ticks", stat->read_latency_ticks);
    4812           0 :         spdk_json_write_named_uint64(w, "max_read_latency_ticks", stat->max_read_latency_ticks);
    4813           0 :         spdk_json_write_named_uint64(w, "min_read_latency_ticks",
    4814           0 :                                      stat->min_read_latency_ticks != UINT64_MAX ?
    4815           0 :                                      stat->min_read_latency_ticks : 0);
    4816           0 :         spdk_json_write_named_uint64(w, "write_latency_ticks", stat->write_latency_ticks);
    4817           0 :         spdk_json_write_named_uint64(w, "max_write_latency_ticks", stat->max_write_latency_ticks);
    4818           0 :         spdk_json_write_named_uint64(w, "min_write_latency_ticks",
    4819           0 :                                      stat->min_write_latency_ticks != UINT64_MAX ?
    4820           0 :                                      stat->min_write_latency_ticks : 0);
    4821           0 :         spdk_json_write_named_uint64(w, "unmap_latency_ticks", stat->unmap_latency_ticks);
    4822           0 :         spdk_json_write_named_uint64(w, "max_unmap_latency_ticks", stat->max_unmap_latency_ticks);
    4823           0 :         spdk_json_write_named_uint64(w, "min_unmap_latency_ticks",
    4824           0 :                                      stat->min_unmap_latency_ticks != UINT64_MAX ?
    4825           0 :                                      stat->min_unmap_latency_ticks : 0);
    4826           0 :         spdk_json_write_named_uint64(w, "copy_latency_ticks", stat->copy_latency_ticks);
    4827           0 :         spdk_json_write_named_uint64(w, "max_copy_latency_ticks", stat->max_copy_latency_ticks);
    4828           0 :         spdk_json_write_named_uint64(w, "min_copy_latency_ticks",
    4829           0 :                                      stat->min_copy_latency_ticks != UINT64_MAX ?
    4830           0 :                                      stat->min_copy_latency_ticks : 0);
    4831             : 
    4832           0 :         if (stat->io_error != NULL) {
    4833           0 :                 spdk_json_write_named_object_begin(w, "io_error");
    4834           0 :                 for (i = 0; i < -SPDK_MIN_BDEV_IO_STATUS; i++) {
    4835           0 :                         if (stat->io_error->error_status[i] != 0) {
    4836           0 :                                 spdk_json_write_named_uint32(w, bdev_io_status_get_string(-(i + 1)),
    4837           0 :                                                              stat->io_error->error_status[i]);
    4838           0 :                         }
    4839           0 :                 }
    4840           0 :                 spdk_json_write_object_end(w);
    4841           0 :         }
    4842           0 : }
    4843             : 
    4844             : static void
    4845          78 : bdev_channel_abort_queued_ios(struct spdk_bdev_channel *ch)
    4846             : {
    4847          78 :         struct spdk_bdev_shared_resource *shared_resource = ch->shared_resource;
    4848          78 :         struct spdk_bdev_mgmt_channel *mgmt_ch = shared_resource->mgmt_ch;
    4849             : 
    4850          78 :         bdev_abort_all_queued_io(&shared_resource->nomem_io, ch);
    4851          78 :         bdev_abort_all_buf_io(mgmt_ch, ch);
    4852          78 : }
    4853             : 
    4854             : static void
    4855          74 : bdev_channel_destroy(void *io_device, void *ctx_buf)
    4856             : {
    4857          74 :         struct spdk_bdev_channel *ch = ctx_buf;
    4858             : 
    4859          74 :         SPDK_DEBUGLOG(bdev, "Destroying channel %p for bdev %s on thread %p\n", ch, ch->bdev->name,
    4860             :                       spdk_get_thread());
    4861             : 
    4862          74 :         spdk_trace_record(TRACE_BDEV_IOCH_DESTROY, ch->bdev->internal.trace_id, 0, 0,
    4863             :                           spdk_thread_get_id(spdk_io_channel_get_thread(ch->channel)));
    4864             : 
    4865             :         /* This channel is going away, so add its statistics into the bdev so that they don't get lost. */
    4866          74 :         spdk_spin_lock(&ch->bdev->internal.spinlock);
    4867          74 :         spdk_bdev_add_io_stat(ch->bdev->internal.stat, ch->stat);
    4868          74 :         spdk_spin_unlock(&ch->bdev->internal.spinlock);
    4869             : 
    4870          74 :         bdev_abort_all_queued_io(&ch->queued_resets, ch);
    4871             : 
    4872          74 :         bdev_channel_abort_queued_ios(ch);
    4873             : 
    4874          74 :         if (ch->histogram) {
    4875           0 :                 spdk_histogram_data_free(ch->histogram);
    4876           0 :         }
    4877             : 
    4878          74 :         bdev_channel_destroy_resource(ch);
    4879          74 : }
    4880             : 
    4881             : /*
    4882             :  * If the name already exists in the global bdev name tree, RB_INSERT() returns a pointer
    4883             :  * to it. Hence we do not have to call bdev_get_by_name() when using this function.
    4884             :  */
    4885             : static int
    4886         259 : bdev_name_add(struct spdk_bdev_name *bdev_name, struct spdk_bdev *bdev, const char *name)
    4887             : {
    4888             :         struct spdk_bdev_name *tmp;
    4889             : 
    4890         259 :         bdev_name->name = strdup(name);
    4891         259 :         if (bdev_name->name == NULL) {
    4892           0 :                 SPDK_ERRLOG("Unable to allocate bdev name\n");
    4893           0 :                 return -ENOMEM;
    4894             :         }
    4895             : 
    4896         259 :         bdev_name->bdev = bdev;
    4897             : 
    4898         259 :         spdk_spin_lock(&g_bdev_mgr.spinlock);
    4899         259 :         tmp = RB_INSERT(bdev_name_tree, &g_bdev_mgr.bdev_names, bdev_name);
    4900         259 :         spdk_spin_unlock(&g_bdev_mgr.spinlock);
    4901             : 
    4902         259 :         if (tmp != NULL) {
    4903           4 :                 SPDK_ERRLOG("Bdev name %s already exists\n", name);
    4904           4 :                 free(bdev_name->name);
    4905           4 :                 return -EEXIST;
    4906             :         }
    4907             : 
    4908         255 :         return 0;
    4909         259 : }
    4910             : 
    4911             : static void
    4912         255 : bdev_name_del_unsafe(struct spdk_bdev_name *bdev_name)
    4913             : {
    4914         255 :         RB_REMOVE(bdev_name_tree, &g_bdev_mgr.bdev_names, bdev_name);
    4915         255 :         free(bdev_name->name);
    4916         255 : }
    4917             : 
    4918             : static void
    4919           5 : bdev_name_del(struct spdk_bdev_name *bdev_name)
    4920             : {
    4921           5 :         spdk_spin_lock(&g_bdev_mgr.spinlock);
    4922           5 :         bdev_name_del_unsafe(bdev_name);
    4923           5 :         spdk_spin_unlock(&g_bdev_mgr.spinlock);
    4924           5 : }
    4925             : 
    4926             : int
    4927         134 : spdk_bdev_alias_add(struct spdk_bdev *bdev, const char *alias)
    4928             : {
    4929             :         struct spdk_bdev_alias *tmp;
    4930             :         int ret;
    4931             : 
    4932         134 :         if (alias == NULL) {
    4933           1 :                 SPDK_ERRLOG("Empty alias passed\n");
    4934           1 :                 return -EINVAL;
    4935             :         }
    4936             : 
    4937         133 :         tmp = calloc(1, sizeof(*tmp));
    4938         133 :         if (tmp == NULL) {
    4939           0 :                 SPDK_ERRLOG("Unable to allocate alias\n");
    4940           0 :                 return -ENOMEM;
    4941             :         }
    4942             : 
    4943         133 :         ret = bdev_name_add(&tmp->alias, bdev, alias);
    4944         133 :         if (ret != 0) {
    4945           4 :                 free(tmp);
    4946           4 :                 return ret;
    4947             :         }
    4948             : 
    4949         129 :         TAILQ_INSERT_TAIL(&bdev->aliases, tmp, tailq);
    4950             : 
    4951         129 :         return 0;
    4952         134 : }
    4953             : 
    4954             : static int
    4955         130 : bdev_alias_del(struct spdk_bdev *bdev, const char *alias,
    4956             :                void (*alias_del_fn)(struct spdk_bdev_name *n))
    4957             : {
    4958             :         struct spdk_bdev_alias *tmp;
    4959             : 
    4960         135 :         TAILQ_FOREACH(tmp, &bdev->aliases, tailq) {
    4961         131 :                 if (strcmp(alias, tmp->alias.name) == 0) {
    4962         126 :                         TAILQ_REMOVE(&bdev->aliases, tmp, tailq);
    4963         126 :                         alias_del_fn(&tmp->alias);
    4964         126 :                         free(tmp);
    4965         126 :                         return 0;
    4966             :                 }
    4967           5 :         }
    4968             : 
    4969           4 :         return -ENOENT;
    4970         130 : }
    4971             : 
    4972             : int
    4973           4 : spdk_bdev_alias_del(struct spdk_bdev *bdev, const char *alias)
    4974             : {
    4975             :         int rc;
    4976             : 
    4977           4 :         rc = bdev_alias_del(bdev, alias, bdev_name_del);
    4978           4 :         if (rc == -ENOENT) {
    4979           2 :                 SPDK_INFOLOG(bdev, "Alias %s does not exist\n", alias);
    4980           2 :         }
    4981             : 
    4982           4 :         return rc;
    4983             : }
    4984             : 
    4985             : void
    4986           2 : spdk_bdev_alias_del_all(struct spdk_bdev *bdev)
    4987             : {
    4988             :         struct spdk_bdev_alias *p, *tmp;
    4989             : 
    4990           5 :         TAILQ_FOREACH_SAFE(p, &bdev->aliases, tailq, tmp) {
    4991           3 :                 TAILQ_REMOVE(&bdev->aliases, p, tailq);
    4992           3 :                 bdev_name_del(&p->alias);
    4993           3 :                 free(p);
    4994           3 :         }
    4995           2 : }
    4996             : 
    4997             : struct spdk_io_channel *
    4998          76 : spdk_bdev_get_io_channel(struct spdk_bdev_desc *desc)
    4999             : {
    5000          76 :         return spdk_get_io_channel(__bdev_to_io_dev(spdk_bdev_desc_get_bdev(desc)));
    5001             : }
    5002             : 
    5003             : void *
    5004           0 : spdk_bdev_get_module_ctx(struct spdk_bdev_desc *desc)
    5005             : {
    5006           0 :         struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
    5007           0 :         void *ctx = NULL;
    5008             : 
    5009           0 :         if (bdev->fn_table->get_module_ctx) {
    5010           0 :                 ctx = bdev->fn_table->get_module_ctx(bdev->ctxt);
    5011           0 :         }
    5012             : 
    5013           0 :         return ctx;
    5014             : }
    5015             : 
    5016             : const char *
    5017           0 : spdk_bdev_get_module_name(const struct spdk_bdev *bdev)
    5018             : {
    5019           0 :         return bdev->module->name;
    5020             : }
    5021             : 
    5022             : const char *
    5023         255 : spdk_bdev_get_name(const struct spdk_bdev *bdev)
    5024             : {
    5025         255 :         return bdev->name;
    5026             : }
    5027             : 
    5028             : const char *
    5029           0 : spdk_bdev_get_product_name(const struct spdk_bdev *bdev)
    5030             : {
    5031           0 :         return bdev->product_name;
    5032             : }
    5033             : 
    5034             : const struct spdk_bdev_aliases_list *
    5035           0 : spdk_bdev_get_aliases(const struct spdk_bdev *bdev)
    5036             : {
    5037           0 :         return &bdev->aliases;
    5038             : }
    5039             : 
    5040             : uint32_t
    5041           5 : spdk_bdev_get_block_size(const struct spdk_bdev *bdev)
    5042             : {
    5043           5 :         return bdev->blocklen;
    5044             : }
    5045             : 
    5046             : uint32_t
    5047           0 : spdk_bdev_get_write_unit_size(const struct spdk_bdev *bdev)
    5048             : {
    5049           0 :         return bdev->write_unit_size;
    5050             : }
    5051             : 
    5052             : uint64_t
    5053           0 : spdk_bdev_get_num_blocks(const struct spdk_bdev *bdev)
    5054             : {
    5055           0 :         return bdev->blockcnt;
    5056             : }
    5057             : 
    5058             : const char *
    5059           0 : spdk_bdev_get_qos_rpc_type(enum spdk_bdev_qos_rate_limit_type type)
    5060             : {
    5061           0 :         return qos_rpc_type[type];
    5062             : }
    5063             : 
    5064             : void
    5065           0 : spdk_bdev_get_qos_rate_limits(struct spdk_bdev *bdev, uint64_t *limits)
    5066             : {
    5067             :         int i;
    5068             : 
    5069           0 :         memset(limits, 0, sizeof(*limits) * SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES);
    5070             : 
    5071           0 :         spdk_spin_lock(&bdev->internal.spinlock);
    5072           0 :         if (bdev->internal.qos) {
    5073           0 :                 for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
    5074           0 :                         if (bdev->internal.qos->rate_limits[i].limit !=
    5075             :                             SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
    5076           0 :                                 limits[i] = bdev->internal.qos->rate_limits[i].limit;
    5077           0 :                                 if (bdev_qos_is_iops_rate_limit(i) == false) {
    5078             :                                         /* Change from Byte to Megabyte which is user visible. */
    5079           0 :                                         limits[i] = limits[i] / 1024 / 1024;
    5080           0 :                                 }
    5081           0 :                         }
    5082           0 :                 }
    5083           0 :         }
    5084           0 :         spdk_spin_unlock(&bdev->internal.spinlock);
    5085           0 : }
    5086             : 
    5087             : size_t
    5088         276 : spdk_bdev_get_buf_align(const struct spdk_bdev *bdev)
    5089             : {
    5090         276 :         return 1 << bdev->required_alignment;
    5091             : }
    5092             : 
    5093             : uint32_t
    5094           0 : spdk_bdev_get_optimal_io_boundary(const struct spdk_bdev *bdev)
    5095             : {
    5096           0 :         return bdev->optimal_io_boundary;
    5097             : }
    5098             : 
    5099             : bool
    5100           0 : spdk_bdev_has_write_cache(const struct spdk_bdev *bdev)
    5101             : {
    5102           0 :         return bdev->write_cache;
    5103             : }
    5104             : 
    5105             : const struct spdk_uuid *
    5106           0 : spdk_bdev_get_uuid(const struct spdk_bdev *bdev)
    5107             : {
    5108           0 :         return &bdev->uuid;
    5109             : }
    5110             : 
    5111             : uint16_t
    5112           0 : spdk_bdev_get_acwu(const struct spdk_bdev *bdev)
    5113             : {
    5114           0 :         return bdev->acwu;
    5115             : }
    5116             : 
    5117             : uint32_t
    5118          29 : spdk_bdev_get_md_size(const struct spdk_bdev *bdev)
    5119             : {
    5120          29 :         return bdev->md_len;
    5121             : }
    5122             : 
    5123             : bool
    5124         131 : spdk_bdev_is_md_interleaved(const struct spdk_bdev *bdev)
    5125             : {
    5126         131 :         return (bdev->md_len != 0) && bdev->md_interleave;
    5127             : }
    5128             : 
    5129             : bool
    5130         117 : spdk_bdev_is_md_separate(const struct spdk_bdev *bdev)
    5131             : {
    5132         117 :         return (bdev->md_len != 0) && !bdev->md_interleave;
    5133             : }
    5134             : 
    5135             : bool
    5136           0 : spdk_bdev_is_zoned(const struct spdk_bdev *bdev)
    5137             : {
    5138           0 :         return bdev->zoned;
    5139             : }
    5140             : 
    5141             : uint32_t
    5142         122 : spdk_bdev_get_data_block_size(const struct spdk_bdev *bdev)
    5143             : {
    5144         122 :         if (spdk_bdev_is_md_interleaved(bdev)) {
    5145           0 :                 return bdev->blocklen - bdev->md_len;
    5146             :         } else {
    5147         122 :                 return bdev->blocklen;
    5148             :         }
    5149         122 : }
    5150             : 
    5151             : uint32_t
    5152           0 : spdk_bdev_get_physical_block_size(const struct spdk_bdev *bdev)
    5153             : {
    5154           0 :         return bdev->phys_blocklen;
    5155             : }
    5156             : 
    5157             : static uint32_t
    5158           9 : _bdev_get_block_size_with_md(const struct spdk_bdev *bdev)
    5159             : {
    5160           9 :         if (!spdk_bdev_is_md_interleaved(bdev)) {
    5161           6 :                 return bdev->blocklen + bdev->md_len;
    5162             :         } else {
    5163           3 :                 return bdev->blocklen;
    5164             :         }
    5165           9 : }
    5166             : 
    5167             : /* We have to use the typedef in the function declaration to appease astyle. */
    5168             : typedef enum spdk_dif_type spdk_dif_type_t;
    5169             : typedef enum spdk_dif_pi_format spdk_dif_pi_format_t;
    5170             : 
    5171             : spdk_dif_type_t
    5172           0 : spdk_bdev_get_dif_type(const struct spdk_bdev *bdev)
    5173             : {
    5174           0 :         if (bdev->md_len != 0) {
    5175           0 :                 return bdev->dif_type;
    5176             :         } else {
    5177           0 :                 return SPDK_DIF_DISABLE;
    5178             :         }
    5179           0 : }
    5180             : 
    5181             : spdk_dif_pi_format_t
    5182           0 : spdk_bdev_get_dif_pi_format(const struct spdk_bdev *bdev)
    5183             : {
    5184           0 :         return bdev->dif_pi_format;
    5185             : }
    5186             : 
    5187             : bool
    5188           0 : spdk_bdev_is_dif_head_of_md(const struct spdk_bdev *bdev)
    5189             : {
    5190           0 :         if (spdk_bdev_get_dif_type(bdev) != SPDK_DIF_DISABLE) {
    5191           0 :                 return bdev->dif_is_head_of_md;
    5192             :         } else {
    5193           0 :                 return false;
    5194             :         }
    5195           0 : }
    5196             : 
    5197             : bool
    5198           0 : spdk_bdev_is_dif_check_enabled(const struct spdk_bdev *bdev,
    5199             :                                enum spdk_dif_check_type check_type)
    5200             : {
    5201           0 :         if (spdk_bdev_get_dif_type(bdev) == SPDK_DIF_DISABLE) {
    5202           0 :                 return false;
    5203             :         }
    5204             : 
    5205           0 :         switch (check_type) {
    5206             :         case SPDK_DIF_CHECK_TYPE_REFTAG:
    5207           0 :                 return (bdev->dif_check_flags & SPDK_DIF_FLAGS_REFTAG_CHECK) != 0;
    5208             :         case SPDK_DIF_CHECK_TYPE_APPTAG:
    5209           0 :                 return (bdev->dif_check_flags & SPDK_DIF_FLAGS_APPTAG_CHECK) != 0;
    5210             :         case SPDK_DIF_CHECK_TYPE_GUARD:
    5211           0 :                 return (bdev->dif_check_flags & SPDK_DIF_FLAGS_GUARD_CHECK) != 0;
    5212             :         default:
    5213           0 :                 return false;
    5214             :         }
    5215           0 : }
    5216             : 
    5217             : static uint32_t
    5218           3 : bdev_get_max_write(const struct spdk_bdev *bdev, uint64_t num_bytes)
    5219             : {
    5220             :         uint64_t aligned_length, max_write_blocks;
    5221             : 
    5222           3 :         aligned_length = num_bytes - (spdk_bdev_get_buf_align(bdev) - 1);
    5223           3 :         max_write_blocks = aligned_length / _bdev_get_block_size_with_md(bdev);
    5224           3 :         max_write_blocks -= max_write_blocks % bdev->write_unit_size;
    5225             : 
    5226           3 :         return max_write_blocks;
    5227             : }
    5228             : 
    5229             : uint32_t
    5230           1 : spdk_bdev_get_max_copy(const struct spdk_bdev *bdev)
    5231             : {
    5232           1 :         return bdev->max_copy;
    5233             : }
    5234             : 
    5235             : uint64_t
    5236           0 : spdk_bdev_get_qd(const struct spdk_bdev *bdev)
    5237             : {
    5238           0 :         return bdev->internal.measured_queue_depth;
    5239             : }
    5240             : 
    5241             : uint64_t
    5242           0 : spdk_bdev_get_qd_sampling_period(const struct spdk_bdev *bdev)
    5243             : {
    5244           0 :         return bdev->internal.period;
    5245             : }
    5246             : 
    5247             : uint64_t
    5248           0 : spdk_bdev_get_weighted_io_time(const struct spdk_bdev *bdev)
    5249             : {
    5250           0 :         return bdev->internal.weighted_io_time;
    5251             : }
    5252             : 
    5253             : uint64_t
    5254           0 : spdk_bdev_get_io_time(const struct spdk_bdev *bdev)
    5255             : {
    5256           0 :         return bdev->internal.io_time;
    5257             : }
    5258             : 
    5259           0 : union spdk_bdev_nvme_ctratt spdk_bdev_get_nvme_ctratt(struct spdk_bdev *bdev)
    5260             : {
    5261           0 :         return bdev->ctratt;
    5262             : }
    5263             : 
    5264             : uint32_t
    5265           0 : spdk_bdev_get_nvme_nsid(struct spdk_bdev *bdev)
    5266             : {
    5267           0 :         return bdev->nsid;
    5268             : }
    5269             : 
    5270             : uint32_t
    5271           0 : spdk_bdev_desc_get_block_size(struct spdk_bdev_desc *desc)
    5272             : {
    5273           0 :         struct spdk_bdev *bdev = desc->bdev;
    5274             : 
    5275           0 :         return desc->opts.no_metadata ? bdev->blocklen - bdev->md_len : bdev->blocklen;
    5276             : }
    5277             : 
    5278             : uint32_t
    5279           0 : spdk_bdev_desc_get_md_size(struct spdk_bdev_desc *desc)
    5280             : {
    5281           0 :         struct spdk_bdev *bdev = desc->bdev;
    5282             : 
    5283           0 :         return desc->opts.no_metadata ? 0 : bdev->md_len;
    5284             : }
    5285             : 
    5286             : bool
    5287           0 : spdk_bdev_desc_is_md_interleaved(struct spdk_bdev_desc *desc)
    5288             : {
    5289           0 :         struct spdk_bdev *bdev = desc->bdev;
    5290             : 
    5291           0 :         return desc->opts.no_metadata ? false : spdk_bdev_is_md_interleaved(bdev);
    5292             : }
    5293             : 
    5294             : bool
    5295           0 : spdk_bdev_desc_is_md_separate(struct spdk_bdev_desc *desc)
    5296             : {
    5297           0 :         struct spdk_bdev *bdev = desc->bdev;
    5298             : 
    5299           0 :         return desc->opts.no_metadata ? false : spdk_bdev_is_md_separate(bdev);
    5300             : }
    5301             : 
    5302             : spdk_dif_type_t
    5303           0 : spdk_bdev_desc_get_dif_type(struct spdk_bdev_desc *desc)
    5304             : {
    5305           0 :         struct spdk_bdev *bdev = desc->bdev;
    5306             : 
    5307           0 :         return desc->opts.no_metadata ? SPDK_DIF_DISABLE : spdk_bdev_get_dif_type(bdev);
    5308             : }
    5309             : 
    5310             : spdk_dif_pi_format_t
    5311           0 : spdk_bdev_desc_get_dif_pi_format(struct spdk_bdev_desc *desc)
    5312             : {
    5313           0 :         struct spdk_bdev *bdev = desc->bdev;
    5314             : 
    5315           0 :         return desc->opts.no_metadata ? SPDK_DIF_PI_FORMAT_16 : spdk_bdev_get_dif_pi_format(bdev);
    5316             : }
    5317             : 
    5318             : bool
    5319           0 : spdk_bdev_desc_is_dif_head_of_md(struct spdk_bdev_desc *desc)
    5320             : {
    5321           0 :         struct spdk_bdev *bdev = desc->bdev;
    5322             : 
    5323           0 :         return desc->opts.no_metadata ? false : spdk_bdev_is_dif_head_of_md(bdev);
    5324             : }
    5325             : 
    5326             : bool
    5327           0 : spdk_bdev_desc_is_dif_check_enabled(struct spdk_bdev_desc *desc,
    5328             :                                     enum spdk_dif_check_type check_type)
    5329             : {
    5330           0 :         struct spdk_bdev *bdev = desc->bdev;
    5331             : 
    5332           0 :         return desc->opts.no_metadata ? false : spdk_bdev_is_dif_check_enabled(bdev, check_type);
    5333             : }
    5334             : 
    5335             : static void bdev_update_qd_sampling_period(void *ctx);
    5336             : 
    5337             : static void
    5338           1 : _calculate_measured_qd_cpl(struct spdk_bdev *bdev, void *_ctx, int status)
    5339             : {
    5340           1 :         bdev->internal.measured_queue_depth = bdev->internal.temporary_queue_depth;
    5341             : 
    5342           1 :         if (bdev->internal.measured_queue_depth) {
    5343           0 :                 bdev->internal.io_time += bdev->internal.period;
    5344           0 :                 bdev->internal.weighted_io_time += bdev->internal.period * bdev->internal.measured_queue_depth;
    5345           0 :         }
    5346             : 
    5347           1 :         bdev->internal.qd_poll_in_progress = false;
    5348             : 
    5349           1 :         bdev_update_qd_sampling_period(bdev);
    5350           1 : }
    5351             : 
    5352             : static void
    5353           1 : _calculate_measured_qd(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
    5354             :                        struct spdk_io_channel *io_ch, void *_ctx)
    5355             : {
    5356           1 :         struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(io_ch);
    5357             : 
    5358           1 :         bdev->internal.temporary_queue_depth += ch->io_outstanding;
    5359           1 :         spdk_bdev_for_each_channel_continue(i, 0);
    5360           1 : }
    5361             : 
    5362             : static int
    5363           1 : bdev_calculate_measured_queue_depth(void *ctx)
    5364             : {
    5365           1 :         struct spdk_bdev *bdev = ctx;
    5366             : 
    5367           1 :         bdev->internal.qd_poll_in_progress = true;
    5368           1 :         bdev->internal.temporary_queue_depth = 0;
    5369           1 :         spdk_bdev_for_each_channel(bdev, _calculate_measured_qd, bdev, _calculate_measured_qd_cpl);
    5370           1 :         return SPDK_POLLER_BUSY;
    5371             : }
    5372             : 
    5373             : static void
    5374           5 : bdev_update_qd_sampling_period(void *ctx)
    5375             : {
    5376           5 :         struct spdk_bdev *bdev = ctx;
    5377             : 
    5378           5 :         if (bdev->internal.period == bdev->internal.new_period) {
    5379           0 :                 return;
    5380             :         }
    5381             : 
    5382           5 :         if (bdev->internal.qd_poll_in_progress) {
    5383           1 :                 return;
    5384             :         }
    5385             : 
    5386           4 :         bdev->internal.period = bdev->internal.new_period;
    5387             : 
    5388           4 :         spdk_poller_unregister(&bdev->internal.qd_poller);
    5389           4 :         if (bdev->internal.period != 0) {
    5390           2 :                 bdev->internal.qd_poller = SPDK_POLLER_REGISTER(bdev_calculate_measured_queue_depth,
    5391             :                                            bdev, bdev->internal.period);
    5392           2 :         } else {
    5393           2 :                 spdk_bdev_close(bdev->internal.qd_desc);
    5394           2 :                 bdev->internal.qd_desc = NULL;
    5395             :         }
    5396           5 : }
    5397             : 
    5398             : static void
    5399           0 : _tmp_bdev_event_cb(enum spdk_bdev_event_type type, struct spdk_bdev *bdev, void *ctx)
    5400             : {
    5401           0 :         SPDK_NOTICELOG("Unexpected event type: %d\n", type);
    5402           0 : }
    5403             : 
    5404             : void
    5405         131 : spdk_bdev_set_qd_sampling_period(struct spdk_bdev *bdev, uint64_t period)
    5406             : {
    5407             :         int rc;
    5408             : 
    5409         131 :         if (bdev->internal.new_period == period) {
    5410         125 :                 return;
    5411             :         }
    5412             : 
    5413           6 :         bdev->internal.new_period = period;
    5414             : 
    5415           6 :         if (bdev->internal.qd_desc != NULL) {
    5416           4 :                 assert(bdev->internal.period != 0);
    5417             : 
    5418           8 :                 spdk_thread_send_msg(bdev->internal.qd_desc->thread,
    5419           4 :                                      bdev_update_qd_sampling_period, bdev);
    5420           4 :                 return;
    5421             :         }
    5422             : 
    5423           2 :         assert(bdev->internal.period == 0);
    5424             : 
    5425           4 :         rc = spdk_bdev_open_ext(spdk_bdev_get_name(bdev), false, _tmp_bdev_event_cb,
    5426           2 :                                 NULL, &bdev->internal.qd_desc);
    5427           2 :         if (rc != 0) {
    5428           0 :                 return;
    5429             :         }
    5430             : 
    5431           2 :         bdev->internal.period = period;
    5432           2 :         bdev->internal.qd_poller = SPDK_POLLER_REGISTER(bdev_calculate_measured_queue_depth,
    5433             :                                    bdev, period);
    5434         131 : }
    5435             : 
    5436             : struct bdev_get_current_qd_ctx {
    5437             :         uint64_t current_qd;
    5438             :         spdk_bdev_get_current_qd_cb cb_fn;
    5439             :         void *cb_arg;
    5440             : };
    5441             : 
    5442             : static void
    5443           0 : bdev_get_current_qd_done(struct spdk_bdev *bdev, void *_ctx, int status)
    5444             : {
    5445           0 :         struct bdev_get_current_qd_ctx *ctx = _ctx;
    5446             : 
    5447           0 :         ctx->cb_fn(bdev, ctx->current_qd, ctx->cb_arg, 0);
    5448             : 
    5449           0 :         free(ctx);
    5450           0 : }
    5451             : 
    5452             : static void
    5453           0 : bdev_get_current_qd(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
    5454             :                     struct spdk_io_channel *io_ch, void *_ctx)
    5455             : {
    5456           0 :         struct bdev_get_current_qd_ctx *ctx = _ctx;
    5457           0 :         struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
    5458             : 
    5459           0 :         ctx->current_qd += bdev_ch->io_outstanding;
    5460             : 
    5461           0 :         spdk_bdev_for_each_channel_continue(i, 0);
    5462           0 : }
    5463             : 
    5464             : void
    5465           0 : spdk_bdev_get_current_qd(struct spdk_bdev *bdev, spdk_bdev_get_current_qd_cb cb_fn,
    5466             :                          void *cb_arg)
    5467             : {
    5468             :         struct bdev_get_current_qd_ctx *ctx;
    5469             : 
    5470           0 :         assert(cb_fn != NULL);
    5471             : 
    5472           0 :         ctx = calloc(1, sizeof(*ctx));
    5473           0 :         if (ctx == NULL) {
    5474           0 :                 cb_fn(bdev, 0, cb_arg, -ENOMEM);
    5475           0 :                 return;
    5476             :         }
    5477             : 
    5478           0 :         ctx->cb_fn = cb_fn;
    5479           0 :         ctx->cb_arg = cb_arg;
    5480             : 
    5481           0 :         spdk_bdev_for_each_channel(bdev, bdev_get_current_qd, ctx, bdev_get_current_qd_done);
    5482           0 : }
    5483             : 
    5484             : static void
    5485          25 : _event_notify(struct spdk_bdev_desc *desc, enum spdk_bdev_event_type type)
    5486             : {
    5487          25 :         assert(desc->thread == spdk_get_thread());
    5488             : 
    5489          25 :         spdk_spin_lock(&desc->spinlock);
    5490          25 :         desc->refs--;
    5491          25 :         if (!desc->closed) {
    5492          14 :                 spdk_spin_unlock(&desc->spinlock);
    5493          28 :                 desc->callback.event_fn(type,
    5494          14 :                                         desc->bdev,
    5495          14 :                                         desc->callback.ctx);
    5496          14 :                 return;
    5497          11 :         } else if (desc->refs == 0) {
    5498             :                 /* This descriptor was closed after this event_notify message was sent.
    5499             :                  * spdk_bdev_close() could not free the descriptor since this message was
    5500             :                  * in flight, so we free it now using bdev_desc_free().
    5501             :                  */
    5502          10 :                 spdk_spin_unlock(&desc->spinlock);
    5503          10 :                 bdev_desc_free(desc);
    5504          10 :                 return;
    5505             :         }
    5506           1 :         spdk_spin_unlock(&desc->spinlock);
    5507          25 : }
    5508             : 
    5509             : static void
    5510          25 : event_notify(struct spdk_bdev_desc *desc, spdk_msg_fn event_notify_fn)
    5511             : {
    5512          25 :         spdk_spin_lock(&desc->spinlock);
    5513          25 :         desc->refs++;
    5514          25 :         spdk_thread_send_msg(desc->thread, event_notify_fn, desc);
    5515          25 :         spdk_spin_unlock(&desc->spinlock);
    5516          25 : }
    5517             : 
    5518             : static void
    5519           6 : _resize_notify(void *ctx)
    5520             : {
    5521           6 :         struct spdk_bdev_desc *desc = ctx;
    5522             : 
    5523           6 :         _event_notify(desc, SPDK_BDEV_EVENT_RESIZE);
    5524           6 : }
    5525             : 
    5526             : int
    5527          11 : spdk_bdev_notify_blockcnt_change(struct spdk_bdev *bdev, uint64_t size)
    5528             : {
    5529             :         struct spdk_bdev_desc *desc;
    5530             :         int ret;
    5531             : 
    5532          11 :         if (size == bdev->blockcnt) {
    5533           0 :                 return 0;
    5534             :         }
    5535             : 
    5536          11 :         spdk_spin_lock(&bdev->internal.spinlock);
    5537             : 
    5538             :         /* bdev has open descriptors */
    5539          11 :         if (!TAILQ_EMPTY(&bdev->internal.open_descs) &&
    5540           7 :             bdev->blockcnt > size) {
    5541           1 :                 ret = -EBUSY;
    5542           1 :         } else {
    5543          10 :                 bdev->blockcnt = size;
    5544          16 :                 TAILQ_FOREACH(desc, &bdev->internal.open_descs, link) {
    5545           6 :                         event_notify(desc, _resize_notify);
    5546           6 :                 }
    5547          10 :                 ret = 0;
    5548             :         }
    5549             : 
    5550          11 :         spdk_spin_unlock(&bdev->internal.spinlock);
    5551             : 
    5552          11 :         return ret;
    5553          11 : }
    5554             : 
    5555             : /*
    5556             :  * Convert I/O offset and length from bytes to blocks.
    5557             :  *
    5558             :  * Returns zero on success or non-zero if the byte parameters aren't divisible by the block size.
    5559             :  */
    5560             : static uint64_t
    5561          20 : bdev_bytes_to_blocks(struct spdk_bdev_desc *desc, uint64_t offset_bytes,
    5562             :                      uint64_t *offset_blocks, uint64_t num_bytes, uint64_t *num_blocks)
    5563             : {
    5564          20 :         uint32_t block_size = bdev_desc_get_block_size(desc);
    5565             :         uint8_t shift_cnt;
    5566             : 
    5567             :         /* Avoid expensive div operations if possible. These spdk_u32 functions are very cheap. */
    5568          20 :         if (spdk_likely(spdk_u32_is_pow2(block_size))) {
    5569          17 :                 shift_cnt = spdk_u32log2(block_size);
    5570          17 :                 *offset_blocks = offset_bytes >> shift_cnt;
    5571          17 :                 *num_blocks = num_bytes >> shift_cnt;
    5572          34 :                 return (offset_bytes - (*offset_blocks << shift_cnt)) |
    5573          17 :                        (num_bytes - (*num_blocks << shift_cnt));
    5574             :         } else {
    5575           3 :                 *offset_blocks = offset_bytes / block_size;
    5576           3 :                 *num_blocks = num_bytes / block_size;
    5577           3 :                 return (offset_bytes % block_size) | (num_bytes % block_size);
    5578             :         }
    5579          20 : }
    5580             : 
    5581             : static bool
    5582         689 : bdev_io_valid_blocks(struct spdk_bdev *bdev, uint64_t offset_blocks, uint64_t num_blocks)
    5583             : {
    5584             :         /* Return failure if offset_blocks + num_blocks is less than offset_blocks; indicates there
    5585             :          * has been an overflow and hence the offset has been wrapped around */
    5586         689 :         if (offset_blocks + num_blocks < offset_blocks) {
    5587           1 :                 return false;
    5588             :         }
    5589             : 
    5590             :         /* Return failure if offset_blocks + num_blocks exceeds the size of the bdev */
    5591         688 :         if (offset_blocks + num_blocks > bdev->blockcnt) {
    5592           2 :                 return false;
    5593             :         }
    5594             : 
    5595         686 :         return true;
    5596         689 : }
    5597             : 
    5598             : static void
    5599           2 : bdev_seek_complete_cb(void *ctx)
    5600             : {
    5601           2 :         struct spdk_bdev_io *bdev_io = ctx;
    5602             : 
    5603           2 :         bdev_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
    5604           2 :         bdev_io->internal.cb(bdev_io, true, bdev_io->internal.caller_ctx);
    5605           2 : }
    5606             : 
    5607             : static int
    5608           4 : bdev_seek(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
    5609             :           uint64_t offset_blocks, enum spdk_bdev_io_type io_type,
    5610             :           spdk_bdev_io_completion_cb cb, void *cb_arg)
    5611             : {
    5612           4 :         struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
    5613             :         struct spdk_bdev_io *bdev_io;
    5614           4 :         struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
    5615             : 
    5616           4 :         assert(io_type == SPDK_BDEV_IO_TYPE_SEEK_DATA || io_type == SPDK_BDEV_IO_TYPE_SEEK_HOLE);
    5617             : 
    5618             :         /* Check if offset_blocks is valid looking at the validity of one block */
    5619           4 :         if (!bdev_io_valid_blocks(bdev, offset_blocks, 1)) {
    5620           0 :                 return -EINVAL;
    5621             :         }
    5622             : 
    5623           4 :         bdev_io = bdev_channel_get_io(channel);
    5624           4 :         if (!bdev_io) {
    5625           0 :                 return -ENOMEM;
    5626             :         }
    5627             : 
    5628           4 :         bdev_io->internal.ch = channel;
    5629           4 :         bdev_io->internal.desc = desc;
    5630           4 :         bdev_io->type = io_type;
    5631           4 :         bdev_io->u.bdev.offset_blocks = offset_blocks;
    5632           4 :         bdev_io->u.bdev.memory_domain = NULL;
    5633           4 :         bdev_io->u.bdev.memory_domain_ctx = NULL;
    5634           4 :         bdev_io->u.bdev.accel_sequence = NULL;
    5635           4 :         bdev_io_init(bdev_io, bdev, cb_arg, cb);
    5636             : 
    5637           4 :         if (!spdk_bdev_io_type_supported(bdev, io_type)) {
    5638             :                 /* In case bdev doesn't support seek to next data/hole offset,
    5639             :                  * it is assumed that only data and no holes are present */
    5640           2 :                 if (io_type == SPDK_BDEV_IO_TYPE_SEEK_DATA) {
    5641           1 :                         bdev_io->u.bdev.seek.offset = offset_blocks;
    5642           1 :                 } else {
    5643           1 :                         bdev_io->u.bdev.seek.offset = UINT64_MAX;
    5644             :                 }
    5645             : 
    5646           2 :                 spdk_thread_send_msg(spdk_get_thread(), bdev_seek_complete_cb, bdev_io);
    5647           2 :                 return 0;
    5648             :         }
    5649             : 
    5650           2 :         bdev_io_submit(bdev_io);
    5651           2 :         return 0;
    5652           4 : }
    5653             : 
    5654             : int
    5655           2 : spdk_bdev_seek_data(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
    5656             :                     uint64_t offset_blocks,
    5657             :                     spdk_bdev_io_completion_cb cb, void *cb_arg)
    5658             : {
    5659           2 :         return bdev_seek(desc, ch, offset_blocks, SPDK_BDEV_IO_TYPE_SEEK_DATA, cb, cb_arg);
    5660             : }
    5661             : 
    5662             : int
    5663           2 : spdk_bdev_seek_hole(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
    5664             :                     uint64_t offset_blocks,
    5665             :                     spdk_bdev_io_completion_cb cb, void *cb_arg)
    5666             : {
    5667           2 :         return bdev_seek(desc, ch, offset_blocks, SPDK_BDEV_IO_TYPE_SEEK_HOLE, cb, cb_arg);
    5668             : }
    5669             : 
    5670             : uint64_t
    5671           4 : spdk_bdev_io_get_seek_offset(const struct spdk_bdev_io *bdev_io)
    5672             : {
    5673           4 :         return bdev_io->u.bdev.seek.offset;
    5674             : }
    5675             : 
    5676             : static int
    5677         204 : bdev_read_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch, void *buf,
    5678             :                          void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
    5679             :                          spdk_bdev_io_completion_cb cb, void *cb_arg)
    5680             : {
    5681         204 :         struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
    5682             :         struct spdk_bdev_io *bdev_io;
    5683         204 :         struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
    5684             : 
    5685         204 :         if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
    5686           0 :                 return -EINVAL;
    5687             :         }
    5688             : 
    5689         204 :         bdev_io = bdev_channel_get_io(channel);
    5690         204 :         if (!bdev_io) {
    5691           1 :                 return -ENOMEM;
    5692             :         }
    5693             : 
    5694         203 :         bdev_io->internal.ch = channel;
    5695         203 :         bdev_io->internal.desc = desc;
    5696         203 :         bdev_io->type = SPDK_BDEV_IO_TYPE_READ;
    5697         203 :         bdev_io->u.bdev.iovs = &bdev_io->iov;
    5698         203 :         bdev_io->u.bdev.iovs[0].iov_base = buf;
    5699         203 :         bdev_io->u.bdev.iovs[0].iov_len = num_blocks * bdev_desc_get_block_size(desc);
    5700         203 :         bdev_io->u.bdev.iovcnt = 1;
    5701         203 :         bdev_io->u.bdev.md_buf = md_buf;
    5702         203 :         bdev_io->u.bdev.num_blocks = num_blocks;
    5703         203 :         bdev_io->u.bdev.offset_blocks = offset_blocks;
    5704         203 :         bdev_io->u.bdev.memory_domain = NULL;
    5705         203 :         bdev_io->u.bdev.memory_domain_ctx = NULL;
    5706         203 :         bdev_io->u.bdev.accel_sequence = NULL;
    5707         203 :         bdev_io->u.bdev.dif_check_flags = bdev->dif_check_flags;
    5708         203 :         bdev_io_init(bdev_io, bdev, cb_arg, cb);
    5709             : 
    5710         203 :         bdev_io_submit(bdev_io);
    5711         203 :         return 0;
    5712         204 : }
    5713             : 
    5714             : int
    5715           3 : spdk_bdev_read(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
    5716             :                void *buf, uint64_t offset, uint64_t nbytes,
    5717             :                spdk_bdev_io_completion_cb cb, void *cb_arg)
    5718             : {
    5719             :         uint64_t offset_blocks, num_blocks;
    5720             : 
    5721           3 :         if (bdev_bytes_to_blocks(desc, offset, &offset_blocks, nbytes, &num_blocks) != 0) {
    5722           0 :                 return -EINVAL;
    5723             :         }
    5724             : 
    5725           3 :         return spdk_bdev_read_blocks(desc, ch, buf, offset_blocks, num_blocks, cb, cb_arg);
    5726           3 : }
    5727             : 
    5728             : int
    5729         200 : spdk_bdev_read_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
    5730             :                       void *buf, uint64_t offset_blocks, uint64_t num_blocks,
    5731             :                       spdk_bdev_io_completion_cb cb, void *cb_arg)
    5732             : {
    5733         200 :         return bdev_read_blocks_with_md(desc, ch, buf, NULL, offset_blocks, num_blocks, cb, cb_arg);
    5734             : }
    5735             : 
    5736             : int
    5737           4 : spdk_bdev_read_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
    5738             :                               void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
    5739             :                               spdk_bdev_io_completion_cb cb, void *cb_arg)
    5740             : {
    5741           8 :         struct iovec iov = {
    5742           4 :                 .iov_base = buf,
    5743             :         };
    5744             : 
    5745           4 :         if (md_buf && !spdk_bdev_is_md_separate(spdk_bdev_desc_get_bdev(desc))) {
    5746           0 :                 return -EINVAL;
    5747             :         }
    5748             : 
    5749           4 :         if ((md_buf || desc->opts.no_metadata) && !_is_buf_allocated(&iov)) {
    5750           0 :                 return -EINVAL;
    5751             :         }
    5752             : 
    5753           8 :         return bdev_read_blocks_with_md(desc, ch, buf, md_buf, offset_blocks, num_blocks,
    5754           4 :                                         cb, cb_arg);
    5755           4 : }
    5756             : 
    5757             : int
    5758           5 : spdk_bdev_readv(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
    5759             :                 struct iovec *iov, int iovcnt,
    5760             :                 uint64_t offset, uint64_t nbytes,
    5761             :                 spdk_bdev_io_completion_cb cb, void *cb_arg)
    5762             : {
    5763             :         uint64_t offset_blocks, num_blocks;
    5764             : 
    5765           5 :         if (bdev_bytes_to_blocks(desc, offset, &offset_blocks, nbytes, &num_blocks) != 0) {
    5766           0 :                 return -EINVAL;
    5767             :         }
    5768             : 
    5769           5 :         return spdk_bdev_readv_blocks(desc, ch, iov, iovcnt, offset_blocks, num_blocks, cb, cb_arg);
    5770           5 : }
    5771             : 
    5772             : static int
    5773         226 : bdev_readv_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
    5774             :                           struct iovec *iov, int iovcnt, void *md_buf, uint64_t offset_blocks,
    5775             :                           uint64_t num_blocks, struct spdk_memory_domain *domain, void *domain_ctx,
    5776             :                           struct spdk_accel_sequence *seq, uint32_t dif_check_flags,
    5777             :                           spdk_bdev_io_completion_cb cb, void *cb_arg)
    5778             : {
    5779         226 :         struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
    5780             :         struct spdk_bdev_io *bdev_io;
    5781         226 :         struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
    5782             : 
    5783         226 :         if (spdk_unlikely(!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks))) {
    5784           0 :                 return -EINVAL;
    5785             :         }
    5786             : 
    5787         226 :         bdev_io = bdev_channel_get_io(channel);
    5788         226 :         if (spdk_unlikely(!bdev_io)) {
    5789           2 :                 return -ENOMEM;
    5790             :         }
    5791             : 
    5792         224 :         bdev_io->internal.ch = channel;
    5793         224 :         bdev_io->internal.desc = desc;
    5794         224 :         bdev_io->type = SPDK_BDEV_IO_TYPE_READ;
    5795         224 :         bdev_io->u.bdev.iovs = iov;
    5796         224 :         bdev_io->u.bdev.iovcnt = iovcnt;
    5797         224 :         bdev_io->u.bdev.md_buf = md_buf;
    5798         224 :         bdev_io->u.bdev.num_blocks = num_blocks;
    5799         224 :         bdev_io->u.bdev.offset_blocks = offset_blocks;
    5800         224 :         bdev_io_init(bdev_io, bdev, cb_arg, cb);
    5801             : 
    5802         224 :         if (seq != NULL) {
    5803           0 :                 bdev_io->internal.f.has_accel_sequence = true;
    5804           0 :                 bdev_io->internal.accel_sequence = seq;
    5805           0 :         }
    5806             : 
    5807         224 :         if (domain != NULL) {
    5808           2 :                 bdev_io->internal.f.has_memory_domain = true;
    5809           2 :                 bdev_io->internal.memory_domain = domain;
    5810           2 :                 bdev_io->internal.memory_domain_ctx = domain_ctx;
    5811           2 :         }
    5812             : 
    5813         224 :         bdev_io->u.bdev.memory_domain = domain;
    5814         224 :         bdev_io->u.bdev.memory_domain_ctx = domain_ctx;
    5815         224 :         bdev_io->u.bdev.accel_sequence = seq;
    5816         224 :         bdev_io->u.bdev.dif_check_flags = dif_check_flags;
    5817             : 
    5818         224 :         _bdev_io_submit_ext(desc, bdev_io);
    5819             : 
    5820         224 :         return 0;
    5821         226 : }
    5822             : 
    5823             : int
    5824          21 : spdk_bdev_readv_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
    5825             :                        struct iovec *iov, int iovcnt,
    5826             :                        uint64_t offset_blocks, uint64_t num_blocks,
    5827             :                        spdk_bdev_io_completion_cb cb, void *cb_arg)
    5828             : {
    5829          21 :         struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
    5830             : 
    5831          42 :         return bdev_readv_blocks_with_md(desc, ch, iov, iovcnt, NULL, offset_blocks,
    5832          21 :                                          num_blocks, NULL, NULL, NULL, bdev->dif_check_flags, cb, cb_arg);
    5833             : }
    5834             : 
    5835             : int
    5836           4 : spdk_bdev_readv_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
    5837             :                                struct iovec *iov, int iovcnt, void *md_buf,
    5838             :                                uint64_t offset_blocks, uint64_t num_blocks,
    5839             :                                spdk_bdev_io_completion_cb cb, void *cb_arg)
    5840             : {
    5841           4 :         struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
    5842             : 
    5843           4 :         if (md_buf && !spdk_bdev_is_md_separate(bdev)) {
    5844           0 :                 return -EINVAL;
    5845             :         }
    5846             : 
    5847           4 :         if (md_buf && !_is_buf_allocated(iov)) {
    5848           0 :                 return -EINVAL;
    5849             :         }
    5850             : 
    5851           8 :         return bdev_readv_blocks_with_md(desc, ch, iov, iovcnt, md_buf, offset_blocks,
    5852           4 :                                          num_blocks, NULL, NULL, NULL, bdev->dif_check_flags, cb, cb_arg);
    5853           4 : }
    5854             : 
    5855             : static inline bool
    5856          14 : _bdev_io_check_opts(struct spdk_bdev_ext_io_opts *opts, struct iovec *iov)
    5857             : {
    5858             :         /*
    5859             :          * We check if opts size is at least of size when we first introduced
    5860             :          * spdk_bdev_ext_io_opts (ac6f2bdd8d) since access to those members
    5861             :          * are not checked internal.
    5862             :          */
    5863          24 :         return opts->size >= offsetof(struct spdk_bdev_ext_io_opts, metadata) +
    5864          14 :                sizeof(opts->metadata) &&
    5865          10 :                opts->size <= sizeof(*opts) &&
    5866             :                /* When memory domain is used, the user must provide data buffers */
    5867           8 :                (!opts->memory_domain || (iov && iov[0].iov_base));
    5868             : }
    5869             : 
    5870             : int
    5871           8 : spdk_bdev_readv_blocks_ext(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
    5872             :                            struct iovec *iov, int iovcnt,
    5873             :                            uint64_t offset_blocks, uint64_t num_blocks,
    5874             :                            spdk_bdev_io_completion_cb cb, void *cb_arg,
    5875             :                            struct spdk_bdev_ext_io_opts *opts)
    5876             : {
    5877           8 :         struct spdk_memory_domain *domain = NULL;
    5878           8 :         struct spdk_accel_sequence *seq = NULL;
    5879           8 :         void *domain_ctx = NULL, *md = NULL;
    5880           8 :         uint32_t dif_check_flags = 0;
    5881           8 :         struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
    5882             : 
    5883           8 :         if (opts) {
    5884           7 :                 if (spdk_unlikely(!_bdev_io_check_opts(opts, iov))) {
    5885           3 :                         return -EINVAL;
    5886             :                 }
    5887             : 
    5888           4 :                 md = opts->metadata;
    5889           4 :                 domain = bdev_get_ext_io_opt(opts, memory_domain, NULL);
    5890           4 :                 domain_ctx = bdev_get_ext_io_opt(opts, memory_domain_ctx, NULL);
    5891           4 :                 seq = bdev_get_ext_io_opt(opts, accel_sequence, NULL);
    5892           4 :                 if (md) {
    5893           4 :                         if (spdk_unlikely(!spdk_bdev_is_md_separate(bdev))) {
    5894           0 :                                 return -EINVAL;
    5895             :                         }
    5896             : 
    5897           4 :                         if (spdk_unlikely(!_is_buf_allocated(iov))) {
    5898           0 :                                 return -EINVAL;
    5899             :                         }
    5900             : 
    5901           4 :                         if (spdk_unlikely(seq != NULL)) {
    5902           0 :                                 return -EINVAL;
    5903             :                         }
    5904           4 :                 }
    5905           4 :         }
    5906             : 
    5907          10 :         dif_check_flags = bdev->dif_check_flags &
    5908           5 :                           ~(bdev_get_ext_io_opt(opts, dif_check_flags_exclude_mask, 0));
    5909             : 
    5910          10 :         return bdev_readv_blocks_with_md(desc, ch, iov, iovcnt, md, offset_blocks,
    5911           5 :                                          num_blocks, domain, domain_ctx, seq, dif_check_flags, cb, cb_arg);
    5912           8 : }
    5913             : 
    5914             : static int
    5915          36 : bdev_write_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
    5916             :                           void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
    5917             :                           spdk_bdev_io_completion_cb cb, void *cb_arg)
    5918             : {
    5919          36 :         struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
    5920             :         struct spdk_bdev_io *bdev_io;
    5921          36 :         struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
    5922             : 
    5923          36 :         if (!desc->write) {
    5924           0 :                 return -EBADF;
    5925             :         }
    5926             : 
    5927          36 :         if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
    5928           0 :                 return -EINVAL;
    5929             :         }
    5930             : 
    5931          36 :         bdev_io = bdev_channel_get_io(channel);
    5932          36 :         if (!bdev_io) {
    5933           0 :                 return -ENOMEM;
    5934             :         }
    5935             : 
    5936          36 :         bdev_io->internal.ch = channel;
    5937          36 :         bdev_io->internal.desc = desc;
    5938          36 :         bdev_io->type = SPDK_BDEV_IO_TYPE_WRITE;
    5939          36 :         bdev_io->u.bdev.iovs = &bdev_io->iov;
    5940          36 :         bdev_io->u.bdev.iovs[0].iov_base = buf;
    5941          36 :         bdev_io->u.bdev.iovs[0].iov_len = num_blocks * bdev_desc_get_block_size(desc);
    5942          36 :         bdev_io->u.bdev.iovcnt = 1;
    5943          36 :         bdev_io->u.bdev.md_buf = md_buf;
    5944          36 :         bdev_io->u.bdev.num_blocks = num_blocks;
    5945          36 :         bdev_io->u.bdev.offset_blocks = offset_blocks;
    5946          36 :         bdev_io->u.bdev.memory_domain = NULL;
    5947          36 :         bdev_io->u.bdev.memory_domain_ctx = NULL;
    5948          36 :         bdev_io->u.bdev.accel_sequence = NULL;
    5949          36 :         bdev_io->u.bdev.dif_check_flags = bdev->dif_check_flags;
    5950          36 :         bdev_io_init(bdev_io, bdev, cb_arg, cb);
    5951             : 
    5952          36 :         bdev_io_submit(bdev_io);
    5953          36 :         return 0;
    5954          36 : }
    5955             : 
    5956             : int
    5957           3 : spdk_bdev_write(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
    5958             :                 void *buf, uint64_t offset, uint64_t nbytes,
    5959             :                 spdk_bdev_io_completion_cb cb, void *cb_arg)
    5960             : {
    5961             :         uint64_t offset_blocks, num_blocks;
    5962             : 
    5963           3 :         if (bdev_bytes_to_blocks(desc, offset, &offset_blocks, nbytes, &num_blocks) != 0) {
    5964           0 :                 return -EINVAL;
    5965             :         }
    5966             : 
    5967           3 :         return spdk_bdev_write_blocks(desc, ch, buf, offset_blocks, num_blocks, cb, cb_arg);
    5968           3 : }
    5969             : 
    5970             : int
    5971          27 : spdk_bdev_write_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
    5972             :                        void *buf, uint64_t offset_blocks, uint64_t num_blocks,
    5973             :                        spdk_bdev_io_completion_cb cb, void *cb_arg)
    5974             : {
    5975          54 :         return bdev_write_blocks_with_md(desc, ch, buf, NULL, offset_blocks, num_blocks,
    5976          27 :                                          cb, cb_arg);
    5977             : }
    5978             : 
    5979             : int
    5980           3 : spdk_bdev_write_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
    5981             :                                void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
    5982             :                                spdk_bdev_io_completion_cb cb, void *cb_arg)
    5983             : {
    5984           6 :         struct iovec iov = {
    5985           3 :                 .iov_base = buf,
    5986             :         };
    5987             : 
    5988           3 :         if (md_buf && !spdk_bdev_is_md_separate(spdk_bdev_desc_get_bdev(desc))) {
    5989           0 :                 return -EINVAL;
    5990             :         }
    5991             : 
    5992           3 :         if (md_buf && !_is_buf_allocated(&iov)) {
    5993           0 :                 return -EINVAL;
    5994             :         }
    5995             : 
    5996           6 :         return bdev_write_blocks_with_md(desc, ch, buf, md_buf, offset_blocks, num_blocks,
    5997           3 :                                          cb, cb_arg);
    5998           3 : }
    5999             : 
    6000             : static int
    6001          70 : bdev_writev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
    6002             :                            struct iovec *iov, int iovcnt, void *md_buf,
    6003             :                            uint64_t offset_blocks, uint64_t num_blocks,
    6004             :                            struct spdk_memory_domain *domain, void *domain_ctx,
    6005             :                            struct spdk_accel_sequence *seq, uint32_t dif_check_flags,
    6006             :                            uint32_t nvme_cdw12_raw, uint32_t nvme_cdw13_raw,
    6007             :                            spdk_bdev_io_completion_cb cb, void *cb_arg)
    6008             : {
    6009          70 :         struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
    6010             :         struct spdk_bdev_io *bdev_io;
    6011          70 :         struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
    6012             : 
    6013          70 :         if (spdk_unlikely(!desc->write)) {
    6014           0 :                 return -EBADF;
    6015             :         }
    6016             : 
    6017          70 :         if (spdk_unlikely(!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks))) {
    6018           0 :                 return -EINVAL;
    6019             :         }
    6020             : 
    6021          70 :         bdev_io = bdev_channel_get_io(channel);
    6022          70 :         if (spdk_unlikely(!bdev_io)) {
    6023           2 :                 return -ENOMEM;
    6024             :         }
    6025             : 
    6026          68 :         bdev_io->internal.ch = channel;
    6027          68 :         bdev_io->internal.desc = desc;
    6028          68 :         bdev_io->type = SPDK_BDEV_IO_TYPE_WRITE;
    6029          68 :         bdev_io->u.bdev.iovs = iov;
    6030          68 :         bdev_io->u.bdev.iovcnt = iovcnt;
    6031          68 :         bdev_io->u.bdev.md_buf = md_buf;
    6032          68 :         bdev_io->u.bdev.num_blocks = num_blocks;
    6033          68 :         bdev_io->u.bdev.offset_blocks = offset_blocks;
    6034          68 :         bdev_io_init(bdev_io, bdev, cb_arg, cb);
    6035          68 :         if (seq != NULL) {
    6036           0 :                 bdev_io->internal.f.has_accel_sequence = true;
    6037           0 :                 bdev_io->internal.accel_sequence = seq;
    6038           0 :         }
    6039             : 
    6040          68 :         if (domain != NULL) {
    6041           2 :                 bdev_io->internal.f.has_memory_domain = true;
    6042           2 :                 bdev_io->internal.memory_domain = domain;
    6043           2 :                 bdev_io->internal.memory_domain_ctx = domain_ctx;
    6044           2 :         }
    6045             : 
    6046          68 :         bdev_io->u.bdev.memory_domain = domain;
    6047          68 :         bdev_io->u.bdev.memory_domain_ctx = domain_ctx;
    6048          68 :         bdev_io->u.bdev.accel_sequence = seq;
    6049          68 :         bdev_io->u.bdev.dif_check_flags = dif_check_flags;
    6050          68 :         bdev_io->u.bdev.nvme_cdw12.raw = nvme_cdw12_raw;
    6051          68 :         bdev_io->u.bdev.nvme_cdw13.raw = nvme_cdw13_raw;
    6052             : 
    6053          68 :         _bdev_io_submit_ext(desc, bdev_io);
    6054             : 
    6055          68 :         return 0;
    6056          70 : }
    6057             : 
    6058             : int
    6059           3 : spdk_bdev_writev(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
    6060             :                  struct iovec *iov, int iovcnt,
    6061             :                  uint64_t offset, uint64_t len,
    6062             :                  spdk_bdev_io_completion_cb cb, void *cb_arg)
    6063             : {
    6064             :         uint64_t offset_blocks, num_blocks;
    6065             : 
    6066           3 :         if (bdev_bytes_to_blocks(desc, offset, &offset_blocks, len, &num_blocks) != 0) {
    6067           0 :                 return -EINVAL;
    6068             :         }
    6069             : 
    6070           3 :         return spdk_bdev_writev_blocks(desc, ch, iov, iovcnt, offset_blocks, num_blocks, cb, cb_arg);
    6071           3 : }
    6072             : 
    6073             : int
    6074          14 : spdk_bdev_writev_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
    6075             :                         struct iovec *iov, int iovcnt,
    6076             :                         uint64_t offset_blocks, uint64_t num_blocks,
    6077             :                         spdk_bdev_io_completion_cb cb, void *cb_arg)
    6078             : {
    6079          14 :         struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
    6080             : 
    6081          28 :         return bdev_writev_blocks_with_md(desc, ch, iov, iovcnt, NULL, offset_blocks,
    6082          14 :                                           num_blocks, NULL, NULL, NULL, bdev->dif_check_flags, 0, 0,
    6083          14 :                                           cb, cb_arg);
    6084             : }
    6085             : 
    6086             : int
    6087           1 : spdk_bdev_writev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
    6088             :                                 struct iovec *iov, int iovcnt, void *md_buf,
    6089             :                                 uint64_t offset_blocks, uint64_t num_blocks,
    6090             :                                 spdk_bdev_io_completion_cb cb, void *cb_arg)
    6091             : {
    6092           1 :         struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
    6093             : 
    6094           1 :         if (md_buf && !spdk_bdev_is_md_separate(bdev)) {
    6095           0 :                 return -EINVAL;
    6096             :         }
    6097             : 
    6098           1 :         if (md_buf && !_is_buf_allocated(iov)) {
    6099           0 :                 return -EINVAL;
    6100             :         }
    6101             : 
    6102           2 :         return bdev_writev_blocks_with_md(desc, ch, iov, iovcnt, md_buf, offset_blocks,
    6103           1 :                                           num_blocks, NULL, NULL, NULL, bdev->dif_check_flags, 0, 0,
    6104           1 :                                           cb, cb_arg);
    6105           1 : }
    6106             : 
    6107             : int
    6108           8 : spdk_bdev_writev_blocks_ext(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
    6109             :                             struct iovec *iov, int iovcnt,
    6110             :                             uint64_t offset_blocks, uint64_t num_blocks,
    6111             :                             spdk_bdev_io_completion_cb cb, void *cb_arg,
    6112             :                             struct spdk_bdev_ext_io_opts *opts)
    6113             : {
    6114           8 :         struct spdk_memory_domain *domain = NULL;
    6115           8 :         struct spdk_accel_sequence *seq = NULL;
    6116           8 :         void *domain_ctx = NULL, *md = NULL;
    6117           8 :         uint32_t dif_check_flags = 0;
    6118           8 :         struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
    6119           8 :         uint32_t nvme_cdw12_raw = 0;
    6120           8 :         uint32_t nvme_cdw13_raw = 0;
    6121             : 
    6122           8 :         if (opts) {
    6123           7 :                 if (spdk_unlikely(!_bdev_io_check_opts(opts, iov))) {
    6124           3 :                         return -EINVAL;
    6125             :                 }
    6126           4 :                 md = opts->metadata;
    6127           4 :                 domain = bdev_get_ext_io_opt(opts, memory_domain, NULL);
    6128           4 :                 domain_ctx = bdev_get_ext_io_opt(opts, memory_domain_ctx, NULL);
    6129           4 :                 seq = bdev_get_ext_io_opt(opts, accel_sequence, NULL);
    6130           4 :                 nvme_cdw12_raw = bdev_get_ext_io_opt(opts, nvme_cdw12.raw, 0);
    6131           4 :                 nvme_cdw13_raw = bdev_get_ext_io_opt(opts, nvme_cdw13.raw, 0);
    6132           4 :                 if (md) {
    6133           4 :                         if (spdk_unlikely(!spdk_bdev_is_md_separate(bdev))) {
    6134           0 :                                 return -EINVAL;
    6135             :                         }
    6136             : 
    6137           4 :                         if (spdk_unlikely(!_is_buf_allocated(iov))) {
    6138           0 :                                 return -EINVAL;
    6139             :                         }
    6140             : 
    6141           4 :                         if (spdk_unlikely(seq != NULL)) {
    6142           0 :                                 return -EINVAL;
    6143             :                         }
    6144           4 :                 }
    6145           4 :         }
    6146             : 
    6147          10 :         dif_check_flags = bdev->dif_check_flags &
    6148           5 :                           ~(bdev_get_ext_io_opt(opts, dif_check_flags_exclude_mask, 0));
    6149             : 
    6150          10 :         return bdev_writev_blocks_with_md(desc, ch, iov, iovcnt, md, offset_blocks, num_blocks,
    6151           5 :                                           domain, domain_ctx, seq, dif_check_flags,
    6152           5 :                                           nvme_cdw12_raw, nvme_cdw13_raw, cb, cb_arg);
    6153           8 : }
    6154             : 
    6155             : static void
    6156          11 : bdev_compare_do_read_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
    6157             : {
    6158          11 :         struct spdk_bdev_io *parent_io = cb_arg;
    6159          11 :         struct spdk_bdev *bdev = parent_io->bdev;
    6160          11 :         uint8_t *read_buf = bdev_io->u.bdev.iovs[0].iov_base;
    6161          11 :         int i, rc = 0;
    6162             : 
    6163          11 :         if (!success) {
    6164           0 :                 parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
    6165           0 :                 parent_io->internal.cb(parent_io, false, parent_io->internal.caller_ctx);
    6166           0 :                 spdk_bdev_free_io(bdev_io);
    6167           0 :                 return;
    6168             :         }
    6169             : 
    6170          17 :         for (i = 0; i < parent_io->u.bdev.iovcnt; i++) {
    6171          22 :                 rc = memcmp(read_buf,
    6172          11 :                             parent_io->u.bdev.iovs[i].iov_base,
    6173          11 :                             parent_io->u.bdev.iovs[i].iov_len);
    6174          11 :                 if (rc) {
    6175           5 :                         break;
    6176             :                 }
    6177           6 :                 read_buf += parent_io->u.bdev.iovs[i].iov_len;
    6178           6 :         }
    6179             : 
    6180          11 :         if (rc == 0 && parent_io->u.bdev.md_buf && spdk_bdev_is_md_separate(bdev)) {
    6181           4 :                 rc = memcmp(bdev_io->u.bdev.md_buf,
    6182           2 :                             parent_io->u.bdev.md_buf,
    6183           2 :                             spdk_bdev_get_md_size(bdev));
    6184           2 :         }
    6185             : 
    6186          11 :         spdk_bdev_free_io(bdev_io);
    6187             : 
    6188          11 :         if (rc == 0) {
    6189           5 :                 parent_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
    6190           5 :                 parent_io->internal.cb(parent_io, true, parent_io->internal.caller_ctx);
    6191           5 :         } else {
    6192           6 :                 parent_io->internal.status = SPDK_BDEV_IO_STATUS_MISCOMPARE;
    6193           6 :                 parent_io->internal.cb(parent_io, false, parent_io->internal.caller_ctx);
    6194             :         }
    6195          11 : }
    6196             : 
    6197             : static void
    6198          11 : bdev_compare_do_read(void *_bdev_io)
    6199             : {
    6200          11 :         struct spdk_bdev_io *bdev_io = _bdev_io;
    6201             :         int rc;
    6202             : 
    6203          22 :         rc = spdk_bdev_read_blocks(bdev_io->internal.desc,
    6204          11 :                                    spdk_io_channel_from_ctx(bdev_io->internal.ch), NULL,
    6205          11 :                                    bdev_io->u.bdev.offset_blocks, bdev_io->u.bdev.num_blocks,
    6206          11 :                                    bdev_compare_do_read_done, bdev_io);
    6207             : 
    6208          11 :         if (rc == -ENOMEM) {
    6209           0 :                 bdev_queue_io_wait_with_cb(bdev_io, bdev_compare_do_read);
    6210          11 :         } else if (rc != 0) {
    6211           0 :                 bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
    6212           0 :                 bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
    6213           0 :         }
    6214          11 : }
    6215             : 
    6216             : static int
    6217          16 : bdev_comparev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
    6218             :                              struct iovec *iov, int iovcnt, void *md_buf,
    6219             :                              uint64_t offset_blocks, uint64_t num_blocks,
    6220             :                              spdk_bdev_io_completion_cb cb, void *cb_arg)
    6221             : {
    6222          16 :         struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
    6223             :         struct spdk_bdev_io *bdev_io;
    6224          16 :         struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
    6225             : 
    6226          16 :         if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
    6227           0 :                 return -EINVAL;
    6228             :         }
    6229             : 
    6230          16 :         bdev_io = bdev_channel_get_io(channel);
    6231          16 :         if (!bdev_io) {
    6232           0 :                 return -ENOMEM;
    6233             :         }
    6234             : 
    6235          16 :         bdev_io->internal.ch = channel;
    6236          16 :         bdev_io->internal.desc = desc;
    6237          16 :         bdev_io->type = SPDK_BDEV_IO_TYPE_COMPARE;
    6238          16 :         bdev_io->u.bdev.iovs = iov;
    6239          16 :         bdev_io->u.bdev.iovcnt = iovcnt;
    6240          16 :         bdev_io->u.bdev.md_buf = md_buf;
    6241          16 :         bdev_io->u.bdev.num_blocks = num_blocks;
    6242          16 :         bdev_io->u.bdev.offset_blocks = offset_blocks;
    6243          16 :         bdev_io_init(bdev_io, bdev, cb_arg, cb);
    6244          16 :         bdev_io->u.bdev.memory_domain = NULL;
    6245          16 :         bdev_io->u.bdev.memory_domain_ctx = NULL;
    6246          16 :         bdev_io->u.bdev.accel_sequence = NULL;
    6247             : 
    6248          16 :         if (bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COMPARE)) {
    6249           7 :                 bdev_io_submit(bdev_io);
    6250           7 :                 return 0;
    6251             :         }
    6252             : 
    6253           9 :         bdev_compare_do_read(bdev_io);
    6254             : 
    6255           9 :         return 0;
    6256          16 : }
    6257             : 
    6258             : int
    6259          10 : spdk_bdev_comparev_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
    6260             :                           struct iovec *iov, int iovcnt,
    6261             :                           uint64_t offset_blocks, uint64_t num_blocks,
    6262             :                           spdk_bdev_io_completion_cb cb, void *cb_arg)
    6263             : {
    6264          20 :         return bdev_comparev_blocks_with_md(desc, ch, iov, iovcnt, NULL, offset_blocks,
    6265          10 :                                             num_blocks, cb, cb_arg);
    6266             : }
    6267             : 
    6268             : int
    6269           6 : spdk_bdev_comparev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
    6270             :                                   struct iovec *iov, int iovcnt, void *md_buf,
    6271             :                                   uint64_t offset_blocks, uint64_t num_blocks,
    6272             :                                   spdk_bdev_io_completion_cb cb, void *cb_arg)
    6273             : {
    6274           6 :         if (md_buf && !spdk_bdev_is_md_separate(spdk_bdev_desc_get_bdev(desc))) {
    6275           0 :                 return -EINVAL;
    6276             :         }
    6277             : 
    6278           6 :         if (md_buf && !_is_buf_allocated(iov)) {
    6279           0 :                 return -EINVAL;
    6280             :         }
    6281             : 
    6282          12 :         return bdev_comparev_blocks_with_md(desc, ch, iov, iovcnt, md_buf, offset_blocks,
    6283           6 :                                             num_blocks, cb, cb_arg);
    6284           6 : }
    6285             : 
    6286             : static int
    6287           4 : bdev_compare_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
    6288             :                             void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
    6289             :                             spdk_bdev_io_completion_cb cb, void *cb_arg)
    6290             : {
    6291           4 :         struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
    6292             :         struct spdk_bdev_io *bdev_io;
    6293           4 :         struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
    6294             : 
    6295           4 :         if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
    6296           0 :                 return -EINVAL;
    6297             :         }
    6298             : 
    6299           4 :         bdev_io = bdev_channel_get_io(channel);
    6300           4 :         if (!bdev_io) {
    6301           0 :                 return -ENOMEM;
    6302             :         }
    6303             : 
    6304           4 :         bdev_io->internal.ch = channel;
    6305           4 :         bdev_io->internal.desc = desc;
    6306           4 :         bdev_io->type = SPDK_BDEV_IO_TYPE_COMPARE;
    6307           4 :         bdev_io->u.bdev.iovs = &bdev_io->iov;
    6308           4 :         bdev_io->u.bdev.iovs[0].iov_base = buf;
    6309           4 :         bdev_io->u.bdev.iovs[0].iov_len = num_blocks * bdev_desc_get_block_size(desc);
    6310           4 :         bdev_io->u.bdev.iovcnt = 1;
    6311           4 :         bdev_io->u.bdev.md_buf = md_buf;
    6312           4 :         bdev_io->u.bdev.num_blocks = num_blocks;
    6313           4 :         bdev_io->u.bdev.offset_blocks = offset_blocks;
    6314           4 :         bdev_io_init(bdev_io, bdev, cb_arg, cb);
    6315           4 :         bdev_io->u.bdev.memory_domain = NULL;
    6316           4 :         bdev_io->u.bdev.memory_domain_ctx = NULL;
    6317           4 :         bdev_io->u.bdev.accel_sequence = NULL;
    6318             : 
    6319           4 :         if (bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COMPARE)) {
    6320           2 :                 bdev_io_submit(bdev_io);
    6321           2 :                 return 0;
    6322             :         }
    6323             : 
    6324           2 :         bdev_compare_do_read(bdev_io);
    6325             : 
    6326           2 :         return 0;
    6327           4 : }
    6328             : 
    6329             : int
    6330           4 : spdk_bdev_compare_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
    6331             :                          void *buf, uint64_t offset_blocks, uint64_t num_blocks,
    6332             :                          spdk_bdev_io_completion_cb cb, void *cb_arg)
    6333             : {
    6334           8 :         return bdev_compare_blocks_with_md(desc, ch, buf, NULL, offset_blocks, num_blocks,
    6335           4 :                                            cb, cb_arg);
    6336             : }
    6337             : 
    6338             : int
    6339           0 : spdk_bdev_compare_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
    6340             :                                  void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
    6341             :                                  spdk_bdev_io_completion_cb cb, void *cb_arg)
    6342             : {
    6343           0 :         struct iovec iov = {
    6344           0 :                 .iov_base = buf,
    6345             :         };
    6346             : 
    6347           0 :         if (md_buf && !spdk_bdev_is_md_separate(spdk_bdev_desc_get_bdev(desc))) {
    6348           0 :                 return -EINVAL;
    6349             :         }
    6350             : 
    6351           0 :         if (md_buf && !_is_buf_allocated(&iov)) {
    6352           0 :                 return -EINVAL;
    6353             :         }
    6354             : 
    6355           0 :         return bdev_compare_blocks_with_md(desc, ch, buf, md_buf, offset_blocks, num_blocks,
    6356           0 :                                            cb, cb_arg);
    6357           0 : }
    6358             : 
    6359             : static void
    6360           2 : bdev_comparev_and_writev_blocks_unlocked(struct lba_range *range, void *ctx, int unlock_status)
    6361             : {
    6362           2 :         struct spdk_bdev_io *bdev_io = ctx;
    6363             : 
    6364           2 :         if (unlock_status) {
    6365           0 :                 SPDK_ERRLOG("LBA range unlock failed\n");
    6366           0 :         }
    6367             : 
    6368           4 :         bdev_io->internal.cb(bdev_io, bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS ? true :
    6369           2 :                              false, bdev_io->internal.caller_ctx);
    6370           2 : }
    6371             : 
    6372             : static void
    6373           2 : bdev_comparev_and_writev_blocks_unlock(struct spdk_bdev_io *bdev_io, int status)
    6374             : {
    6375           2 :         bdev_io->internal.status = status;
    6376             : 
    6377           4 :         bdev_unlock_lba_range(bdev_io->internal.desc, spdk_io_channel_from_ctx(bdev_io->internal.ch),
    6378           2 :                               bdev_io->u.bdev.offset_blocks, bdev_io->u.bdev.num_blocks,
    6379           2 :                               bdev_comparev_and_writev_blocks_unlocked, bdev_io);
    6380           2 : }
    6381             : 
    6382             : static void
    6383           1 : bdev_compare_and_write_do_write_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
    6384             : {
    6385           1 :         struct spdk_bdev_io *parent_io = cb_arg;
    6386             : 
    6387           1 :         if (!success) {
    6388           0 :                 SPDK_ERRLOG("Compare and write operation failed\n");
    6389           0 :         }
    6390             : 
    6391           1 :         spdk_bdev_free_io(bdev_io);
    6392             : 
    6393           2 :         bdev_comparev_and_writev_blocks_unlock(parent_io,
    6394           1 :                                                success ? SPDK_BDEV_IO_STATUS_SUCCESS : SPDK_BDEV_IO_STATUS_FAILED);
    6395           1 : }
    6396             : 
    6397             : static void
    6398           1 : bdev_compare_and_write_do_write(void *_bdev_io)
    6399             : {
    6400           1 :         struct spdk_bdev_io *bdev_io = _bdev_io;
    6401             :         int rc;
    6402             : 
    6403           2 :         rc = spdk_bdev_writev_blocks(bdev_io->internal.desc,
    6404           1 :                                      spdk_io_channel_from_ctx(bdev_io->internal.ch),
    6405           1 :                                      bdev_io->u.bdev.fused_iovs, bdev_io->u.bdev.fused_iovcnt,
    6406           1 :                                      bdev_io->u.bdev.offset_blocks, bdev_io->u.bdev.num_blocks,
    6407           1 :                                      bdev_compare_and_write_do_write_done, bdev_io);
    6408             : 
    6409             : 
    6410           1 :         if (rc == -ENOMEM) {
    6411           0 :                 bdev_queue_io_wait_with_cb(bdev_io, bdev_compare_and_write_do_write);
    6412           1 :         } else if (rc != 0) {
    6413           0 :                 bdev_comparev_and_writev_blocks_unlock(bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
    6414           0 :         }
    6415           1 : }
    6416             : 
    6417             : static void
    6418           2 : bdev_compare_and_write_do_compare_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
    6419             : {
    6420           2 :         struct spdk_bdev_io *parent_io = cb_arg;
    6421             : 
    6422           2 :         spdk_bdev_free_io(bdev_io);
    6423             : 
    6424           2 :         if (!success) {
    6425           1 :                 bdev_comparev_and_writev_blocks_unlock(parent_io, SPDK_BDEV_IO_STATUS_MISCOMPARE);
    6426           1 :                 return;
    6427             :         }
    6428             : 
    6429           1 :         bdev_compare_and_write_do_write(parent_io);
    6430           2 : }
    6431             : 
    6432             : static void
    6433           2 : bdev_compare_and_write_do_compare(void *_bdev_io)
    6434             : {
    6435           2 :         struct spdk_bdev_io *bdev_io = _bdev_io;
    6436             :         int rc;
    6437             : 
    6438           4 :         rc = spdk_bdev_comparev_blocks(bdev_io->internal.desc,
    6439           2 :                                        spdk_io_channel_from_ctx(bdev_io->internal.ch), bdev_io->u.bdev.iovs,
    6440           2 :                                        bdev_io->u.bdev.iovcnt, bdev_io->u.bdev.offset_blocks, bdev_io->u.bdev.num_blocks,
    6441           2 :                                        bdev_compare_and_write_do_compare_done, bdev_io);
    6442             : 
    6443           2 :         if (rc == -ENOMEM) {
    6444           0 :                 bdev_queue_io_wait_with_cb(bdev_io, bdev_compare_and_write_do_compare);
    6445           2 :         } else if (rc != 0) {
    6446           0 :                 bdev_comparev_and_writev_blocks_unlock(bdev_io, SPDK_BDEV_IO_STATUS_FIRST_FUSED_FAILED);
    6447           0 :         }
    6448           2 : }
    6449             : 
    6450             : static void
    6451           2 : bdev_comparev_and_writev_blocks_locked(struct lba_range *range, void *ctx, int status)
    6452             : {
    6453           2 :         struct spdk_bdev_io *bdev_io = ctx;
    6454             : 
    6455           2 :         if (status) {
    6456           0 :                 bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FIRST_FUSED_FAILED;
    6457           0 :                 bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
    6458           0 :                 return;
    6459             :         }
    6460             : 
    6461           2 :         bdev_compare_and_write_do_compare(bdev_io);
    6462           2 : }
    6463             : 
    6464             : int
    6465           2 : spdk_bdev_comparev_and_writev_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
    6466             :                                      struct iovec *compare_iov, int compare_iovcnt,
    6467             :                                      struct iovec *write_iov, int write_iovcnt,
    6468             :                                      uint64_t offset_blocks, uint64_t num_blocks,
    6469             :                                      spdk_bdev_io_completion_cb cb, void *cb_arg)
    6470             : {
    6471           2 :         struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
    6472             :         struct spdk_bdev_io *bdev_io;
    6473           2 :         struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
    6474             : 
    6475           2 :         if (!desc->write) {
    6476           0 :                 return -EBADF;
    6477             :         }
    6478             : 
    6479           2 :         if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
    6480           0 :                 return -EINVAL;
    6481             :         }
    6482             : 
    6483           2 :         if (num_blocks > bdev->acwu) {
    6484           0 :                 return -EINVAL;
    6485             :         }
    6486             : 
    6487           2 :         bdev_io = bdev_channel_get_io(channel);
    6488           2 :         if (!bdev_io) {
    6489           0 :                 return -ENOMEM;
    6490             :         }
    6491             : 
    6492           2 :         bdev_io->internal.ch = channel;
    6493           2 :         bdev_io->internal.desc = desc;
    6494           2 :         bdev_io->type = SPDK_BDEV_IO_TYPE_COMPARE_AND_WRITE;
    6495           2 :         bdev_io->u.bdev.iovs = compare_iov;
    6496           2 :         bdev_io->u.bdev.iovcnt = compare_iovcnt;
    6497           2 :         bdev_io->u.bdev.fused_iovs = write_iov;
    6498           2 :         bdev_io->u.bdev.fused_iovcnt = write_iovcnt;
    6499           2 :         bdev_io->u.bdev.md_buf = NULL;
    6500           2 :         bdev_io->u.bdev.num_blocks = num_blocks;
    6501           2 :         bdev_io->u.bdev.offset_blocks = offset_blocks;
    6502           2 :         bdev_io_init(bdev_io, bdev, cb_arg, cb);
    6503           2 :         bdev_io->u.bdev.memory_domain = NULL;
    6504           2 :         bdev_io->u.bdev.memory_domain_ctx = NULL;
    6505           2 :         bdev_io->u.bdev.accel_sequence = NULL;
    6506             : 
    6507           2 :         if (bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COMPARE_AND_WRITE)) {
    6508           0 :                 bdev_io_submit(bdev_io);
    6509           0 :                 return 0;
    6510             :         }
    6511             : 
    6512           4 :         return bdev_lock_lba_range(desc, ch, offset_blocks, num_blocks,
    6513           2 :                                    bdev_comparev_and_writev_blocks_locked, bdev_io);
    6514           2 : }
    6515             : 
    6516             : int
    6517           2 : spdk_bdev_zcopy_start(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
    6518             :                       struct iovec *iov, int iovcnt,
    6519             :                       uint64_t offset_blocks, uint64_t num_blocks,
    6520             :                       bool populate,
    6521             :                       spdk_bdev_io_completion_cb cb, void *cb_arg)
    6522             : {
    6523           2 :         struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
    6524             :         struct spdk_bdev_io *bdev_io;
    6525           2 :         struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
    6526             : 
    6527           2 :         if (!desc->write) {
    6528           0 :                 return -EBADF;
    6529             :         }
    6530             : 
    6531           2 :         if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
    6532           0 :                 return -EINVAL;
    6533             :         }
    6534             : 
    6535           2 :         if (!spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_ZCOPY)) {
    6536           0 :                 return -ENOTSUP;
    6537             :         }
    6538             : 
    6539           2 :         bdev_io = bdev_channel_get_io(channel);
    6540           2 :         if (!bdev_io) {
    6541           0 :                 return -ENOMEM;
    6542             :         }
    6543             : 
    6544           2 :         bdev_io->internal.ch = channel;
    6545           2 :         bdev_io->internal.desc = desc;
    6546           2 :         bdev_io->type = SPDK_BDEV_IO_TYPE_ZCOPY;
    6547           2 :         bdev_io->u.bdev.num_blocks = num_blocks;
    6548           2 :         bdev_io->u.bdev.offset_blocks = offset_blocks;
    6549           2 :         bdev_io->u.bdev.iovs = iov;
    6550           2 :         bdev_io->u.bdev.iovcnt = iovcnt;
    6551           2 :         bdev_io->u.bdev.md_buf = NULL;
    6552           2 :         bdev_io->u.bdev.zcopy.populate = populate ? 1 : 0;
    6553           2 :         bdev_io->u.bdev.zcopy.commit = 0;
    6554           2 :         bdev_io->u.bdev.zcopy.start = 1;
    6555           2 :         bdev_io_init(bdev_io, bdev, cb_arg, cb);
    6556           2 :         bdev_io->u.bdev.memory_domain = NULL;
    6557           2 :         bdev_io->u.bdev.memory_domain_ctx = NULL;
    6558           2 :         bdev_io->u.bdev.accel_sequence = NULL;
    6559             : 
    6560           2 :         bdev_io_submit(bdev_io);
    6561             : 
    6562           2 :         return 0;
    6563           2 : }
    6564             : 
    6565             : int
    6566           2 : spdk_bdev_zcopy_end(struct spdk_bdev_io *bdev_io, bool commit,
    6567             :                     spdk_bdev_io_completion_cb cb, void *cb_arg)
    6568             : {
    6569           2 :         if (bdev_io->type != SPDK_BDEV_IO_TYPE_ZCOPY) {
    6570           0 :                 return -EINVAL;
    6571             :         }
    6572             : 
    6573           2 :         bdev_io->u.bdev.zcopy.commit = commit ? 1 : 0;
    6574           2 :         bdev_io->u.bdev.zcopy.start = 0;
    6575           2 :         bdev_io->internal.caller_ctx = cb_arg;
    6576           2 :         bdev_io->internal.cb = cb;
    6577           2 :         bdev_io->internal.status = SPDK_BDEV_IO_STATUS_PENDING;
    6578             : 
    6579           2 :         bdev_io_submit(bdev_io);
    6580             : 
    6581           2 :         return 0;
    6582           2 : }
    6583             : 
    6584             : int
    6585           0 : spdk_bdev_write_zeroes(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
    6586             :                        uint64_t offset, uint64_t len,
    6587             :                        spdk_bdev_io_completion_cb cb, void *cb_arg)
    6588             : {
    6589             :         uint64_t offset_blocks, num_blocks;
    6590             : 
    6591           0 :         if (bdev_bytes_to_blocks(desc, offset, &offset_blocks, len, &num_blocks) != 0) {
    6592           0 :                 return -EINVAL;
    6593             :         }
    6594             : 
    6595           0 :         return spdk_bdev_write_zeroes_blocks(desc, ch, offset_blocks, num_blocks, cb, cb_arg);
    6596           0 : }
    6597             : 
    6598             : int
    6599          33 : spdk_bdev_write_zeroes_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
    6600             :                               uint64_t offset_blocks, uint64_t num_blocks,
    6601             :                               spdk_bdev_io_completion_cb cb, void *cb_arg)
    6602             : {
    6603          33 :         struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
    6604             :         struct spdk_bdev_io *bdev_io;
    6605          33 :         struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
    6606             : 
    6607          33 :         if (!desc->write) {
    6608           0 :                 return -EBADF;
    6609             :         }
    6610             : 
    6611          33 :         if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
    6612           0 :                 return -EINVAL;
    6613             :         }
    6614             : 
    6615          33 :         if (!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE_ZEROES) &&
    6616          10 :             !bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE)) {
    6617           1 :                 return -ENOTSUP;
    6618             :         }
    6619             : 
    6620          32 :         bdev_io = bdev_channel_get_io(channel);
    6621             : 
    6622          32 :         if (!bdev_io) {
    6623           0 :                 return -ENOMEM;
    6624             :         }
    6625             : 
    6626          32 :         bdev_io->type = SPDK_BDEV_IO_TYPE_WRITE_ZEROES;
    6627          32 :         bdev_io->internal.ch = channel;
    6628          32 :         bdev_io->internal.desc = desc;
    6629          32 :         bdev_io->u.bdev.offset_blocks = offset_blocks;
    6630          32 :         bdev_io->u.bdev.num_blocks = num_blocks;
    6631          32 :         bdev_io_init(bdev_io, bdev, cb_arg, cb);
    6632          32 :         bdev_io->u.bdev.memory_domain = NULL;
    6633          32 :         bdev_io->u.bdev.memory_domain_ctx = NULL;
    6634          32 :         bdev_io->u.bdev.accel_sequence = NULL;
    6635             : 
    6636             :         /* If the write_zeroes size is large and should be split, use the generic split
    6637             :          * logic regardless of whether SPDK_BDEV_IO_TYPE_WRITE_ZEREOS is supported or not.
    6638             :          *
    6639             :          * Then, send the write_zeroes request if SPDK_BDEV_IO_TYPE_WRITE_ZEROES is supported
    6640             :          * or emulate it using regular write request otherwise.
    6641             :          */
    6642          32 :         if (bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE_ZEROES) ||
    6643           9 :             bdev_io->internal.f.split) {
    6644          26 :                 bdev_io_submit(bdev_io);
    6645          26 :                 return 0;
    6646             :         }
    6647             : 
    6648           6 :         assert(_bdev_get_block_size_with_md(bdev) <= ZERO_BUFFER_SIZE);
    6649             : 
    6650           6 :         return bdev_write_zero_buffer(bdev_io);
    6651          33 : }
    6652             : 
    6653             : int
    6654           0 : spdk_bdev_unmap(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
    6655             :                 uint64_t offset, uint64_t nbytes,
    6656             :                 spdk_bdev_io_completion_cb cb, void *cb_arg)
    6657             : {
    6658             :         uint64_t offset_blocks, num_blocks;
    6659             : 
    6660           0 :         if (bdev_bytes_to_blocks(desc, offset, &offset_blocks, nbytes, &num_blocks) != 0) {
    6661           0 :                 return -EINVAL;
    6662             :         }
    6663             : 
    6664           0 :         return spdk_bdev_unmap_blocks(desc, ch, offset_blocks, num_blocks, cb, cb_arg);
    6665           0 : }
    6666             : 
    6667             : static void
    6668           0 : bdev_io_complete_cb(void *ctx)
    6669             : {
    6670           0 :         struct spdk_bdev_io *bdev_io = ctx;
    6671             : 
    6672           0 :         bdev_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
    6673           0 :         bdev_io->internal.cb(bdev_io, true, bdev_io->internal.caller_ctx);
    6674           0 : }
    6675             : 
    6676             : int
    6677          22 : spdk_bdev_unmap_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
    6678             :                        uint64_t offset_blocks, uint64_t num_blocks,
    6679             :                        spdk_bdev_io_completion_cb cb, void *cb_arg)
    6680             : {
    6681          22 :         struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
    6682             :         struct spdk_bdev_io *bdev_io;
    6683          22 :         struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
    6684             : 
    6685          22 :         if (!desc->write) {
    6686           0 :                 return -EBADF;
    6687             :         }
    6688             : 
    6689          22 :         if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
    6690           0 :                 return -EINVAL;
    6691             :         }
    6692             : 
    6693          22 :         bdev_io = bdev_channel_get_io(channel);
    6694          22 :         if (!bdev_io) {
    6695           0 :                 return -ENOMEM;
    6696             :         }
    6697             : 
    6698          22 :         bdev_io->internal.ch = channel;
    6699          22 :         bdev_io->internal.desc = desc;
    6700          22 :         bdev_io->type = SPDK_BDEV_IO_TYPE_UNMAP;
    6701             : 
    6702          22 :         bdev_io->u.bdev.iovs = &bdev_io->iov;
    6703          22 :         bdev_io->u.bdev.iovs[0].iov_base = NULL;
    6704          22 :         bdev_io->u.bdev.iovs[0].iov_len = 0;
    6705          22 :         bdev_io->u.bdev.iovcnt = 1;
    6706             : 
    6707          22 :         bdev_io->u.bdev.offset_blocks = offset_blocks;
    6708          22 :         bdev_io->u.bdev.num_blocks = num_blocks;
    6709          22 :         bdev_io_init(bdev_io, bdev, cb_arg, cb);
    6710          22 :         bdev_io->u.bdev.memory_domain = NULL;
    6711          22 :         bdev_io->u.bdev.memory_domain_ctx = NULL;
    6712          22 :         bdev_io->u.bdev.accel_sequence = NULL;
    6713             : 
    6714          22 :         if (num_blocks == 0) {
    6715           0 :                 spdk_thread_send_msg(spdk_get_thread(), bdev_io_complete_cb, bdev_io);
    6716           0 :                 return 0;
    6717             :         }
    6718             : 
    6719          22 :         bdev_io_submit(bdev_io);
    6720          22 :         return 0;
    6721          22 : }
    6722             : 
    6723             : int
    6724           0 : spdk_bdev_flush(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
    6725             :                 uint64_t offset, uint64_t length,
    6726             :                 spdk_bdev_io_completion_cb cb, void *cb_arg)
    6727             : {
    6728             :         uint64_t offset_blocks, num_blocks;
    6729             : 
    6730           0 :         if (bdev_bytes_to_blocks(desc, offset, &offset_blocks, length, &num_blocks) != 0) {
    6731           0 :                 return -EINVAL;
    6732             :         }
    6733             : 
    6734           0 :         return spdk_bdev_flush_blocks(desc, ch, offset_blocks, num_blocks, cb, cb_arg);
    6735           0 : }
    6736             : 
    6737             : int
    6738           2 : spdk_bdev_flush_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
    6739             :                        uint64_t offset_blocks, uint64_t num_blocks,
    6740             :                        spdk_bdev_io_completion_cb cb, void *cb_arg)
    6741             : {
    6742           2 :         struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
    6743             :         struct spdk_bdev_io *bdev_io;
    6744           2 :         struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
    6745             : 
    6746           2 :         if (!desc->write) {
    6747           0 :                 return -EBADF;
    6748             :         }
    6749             : 
    6750           2 :         if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
    6751           0 :                 return -EINVAL;
    6752             :         }
    6753             : 
    6754           2 :         bdev_io = bdev_channel_get_io(channel);
    6755           2 :         if (!bdev_io) {
    6756           0 :                 return -ENOMEM;
    6757             :         }
    6758             : 
    6759           2 :         bdev_io->internal.ch = channel;
    6760           2 :         bdev_io->internal.desc = desc;
    6761           2 :         bdev_io->type = SPDK_BDEV_IO_TYPE_FLUSH;
    6762           2 :         bdev_io->u.bdev.iovs = NULL;
    6763           2 :         bdev_io->u.bdev.iovcnt = 0;
    6764           2 :         bdev_io->u.bdev.offset_blocks = offset_blocks;
    6765           2 :         bdev_io->u.bdev.num_blocks = num_blocks;
    6766           2 :         bdev_io->u.bdev.memory_domain = NULL;
    6767           2 :         bdev_io->u.bdev.memory_domain_ctx = NULL;
    6768           2 :         bdev_io->u.bdev.accel_sequence = NULL;
    6769           2 :         bdev_io_init(bdev_io, bdev, cb_arg, cb);
    6770             : 
    6771           2 :         bdev_io_submit(bdev_io);
    6772           2 :         return 0;
    6773           2 : }
    6774             : 
    6775             : static int bdev_reset_poll_for_outstanding_io(void *ctx);
    6776             : 
    6777             : static void
    6778          13 : bdev_reset_check_outstanding_io_done(struct spdk_bdev *bdev, void *_ctx, int status)
    6779             : {
    6780          13 :         struct spdk_bdev_channel *ch = _ctx;
    6781             :         struct spdk_bdev_io *bdev_io;
    6782             : 
    6783          13 :         bdev_io = TAILQ_FIRST(&ch->queued_resets);
    6784             : 
    6785          13 :         if (status == -EBUSY) {
    6786           9 :                 if (spdk_get_ticks() < bdev_io->u.reset.wait_poller.stop_time_tsc) {
    6787           8 :                         bdev_io->u.reset.wait_poller.poller = SPDK_POLLER_REGISTER(bdev_reset_poll_for_outstanding_io,
    6788             :                                                               ch, BDEV_RESET_CHECK_OUTSTANDING_IO_PERIOD);
    6789           8 :                 } else {
    6790           1 :                         TAILQ_REMOVE(&ch->queued_resets, bdev_io, internal.link);
    6791             : 
    6792           1 :                         if (TAILQ_EMPTY(&ch->io_memory_domain) && TAILQ_EMPTY(&ch->io_accel_exec)) {
    6793             :                                 /* If outstanding IOs are still present and reset_io_drain_timeout
    6794             :                                  * seconds passed, start the reset. */
    6795           1 :                                 bdev_io_submit_reset(bdev_io);
    6796           1 :                         } else {
    6797             :                                 /* We still have in progress memory domain pull/push or we're
    6798             :                                  * executing accel sequence.  Since we cannot abort either of those
    6799             :                                  * operations, fail the reset request. */
    6800           0 :                                 spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
    6801             :                         }
    6802             :                 }
    6803           9 :         } else {
    6804           4 :                 TAILQ_REMOVE(&ch->queued_resets, bdev_io, internal.link);
    6805           4 :                 SPDK_DEBUGLOG(bdev,
    6806             :                               "Skipping reset for underlying device of bdev: %s - no outstanding I/O.\n",
    6807             :                               ch->bdev->name);
    6808             :                 /* Mark the completion status as a SUCCESS and complete the reset. */
    6809           4 :                 spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_SUCCESS);
    6810             :         }
    6811          13 : }
    6812             : 
    6813             : static void
    6814          13 : bdev_reset_check_outstanding_io(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
    6815             :                                 struct spdk_io_channel *io_ch, void *_ctx)
    6816             : {
    6817          13 :         struct spdk_bdev_channel *cur_ch = __io_ch_to_bdev_ch(io_ch);
    6818          13 :         int status = 0;
    6819             : 
    6820          17 :         if (cur_ch->io_outstanding > 0 ||
    6821           4 :             !TAILQ_EMPTY(&cur_ch->io_memory_domain) ||
    6822           4 :             !TAILQ_EMPTY(&cur_ch->io_accel_exec)) {
    6823             :                 /* If a channel has outstanding IO, set status to -EBUSY code. This will stop
    6824             :                  * further iteration over the rest of the channels and pass non-zero status
    6825             :                  * to the callback function. */
    6826           9 :                 status = -EBUSY;
    6827           9 :         }
    6828          13 :         spdk_bdev_for_each_channel_continue(i, status);
    6829          13 : }
    6830             : 
    6831             : static int
    6832           8 : bdev_reset_poll_for_outstanding_io(void *ctx)
    6833             : {
    6834           8 :         struct spdk_bdev_channel *ch = ctx;
    6835             :         struct spdk_bdev_io *bdev_io;
    6836             : 
    6837           8 :         bdev_io = TAILQ_FIRST(&ch->queued_resets);
    6838             : 
    6839           8 :         spdk_poller_unregister(&bdev_io->u.reset.wait_poller.poller);
    6840           8 :         spdk_bdev_for_each_channel(ch->bdev, bdev_reset_check_outstanding_io, ch,
    6841             :                                    bdev_reset_check_outstanding_io_done);
    6842             : 
    6843           8 :         return SPDK_POLLER_BUSY;
    6844             : }
    6845             : 
    6846             : static void
    6847          15 : bdev_reset_freeze_channel_done(struct spdk_bdev *bdev, void *_ctx, int status)
    6848             : {
    6849          15 :         struct spdk_bdev_channel *ch = _ctx;
    6850             :         struct spdk_bdev_io *bdev_io;
    6851             : 
    6852          15 :         bdev_io = TAILQ_FIRST(&ch->queued_resets);
    6853             : 
    6854          15 :         if (bdev->reset_io_drain_timeout == 0) {
    6855          10 :                 TAILQ_REMOVE(&ch->queued_resets, bdev_io, internal.link);
    6856             : 
    6857          10 :                 bdev_io_submit_reset(bdev_io);
    6858          10 :                 return;
    6859             :         }
    6860             : 
    6861          10 :         bdev_io->u.reset.wait_poller.stop_time_tsc = spdk_get_ticks() +
    6862           5 :                         (ch->bdev->reset_io_drain_timeout * spdk_get_ticks_hz());
    6863             : 
    6864             :         /* In case bdev->reset_io_drain_timeout is not equal to zero,
    6865             :          * submit the reset to the underlying module only if outstanding I/O
    6866             :          * remain after reset_io_drain_timeout seconds have passed. */
    6867           5 :         spdk_bdev_for_each_channel(ch->bdev, bdev_reset_check_outstanding_io, ch,
    6868             :                                    bdev_reset_check_outstanding_io_done);
    6869          15 : }
    6870             : 
    6871             : static void
    6872          18 : bdev_reset_freeze_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
    6873             :                           struct spdk_io_channel *ch, void *_ctx)
    6874             : {
    6875             :         struct spdk_bdev_channel        *channel;
    6876             :         struct spdk_bdev_mgmt_channel   *mgmt_channel;
    6877             :         struct spdk_bdev_shared_resource *shared_resource;
    6878             :         bdev_io_tailq_t                 tmp_queued;
    6879             : 
    6880          18 :         TAILQ_INIT(&tmp_queued);
    6881             : 
    6882          18 :         channel = __io_ch_to_bdev_ch(ch);
    6883          18 :         shared_resource = channel->shared_resource;
    6884          18 :         mgmt_channel = shared_resource->mgmt_ch;
    6885             : 
    6886          18 :         channel->flags |= BDEV_CH_RESET_IN_PROGRESS;
    6887             : 
    6888          18 :         if ((channel->flags & BDEV_CH_QOS_ENABLED) != 0) {
    6889           2 :                 TAILQ_SWAP(&channel->qos_queued_io, &tmp_queued, spdk_bdev_io, internal.link);
    6890           2 :         }
    6891             : 
    6892          18 :         bdev_abort_all_queued_io(&shared_resource->nomem_io, channel);
    6893          18 :         bdev_abort_all_buf_io(mgmt_channel, channel);
    6894          18 :         bdev_abort_all_queued_io(&tmp_queued, channel);
    6895             : 
    6896          18 :         spdk_bdev_for_each_channel_continue(i, 0);
    6897          18 : }
    6898             : 
    6899             : static void
    6900          15 : bdev_start_reset(void *ctx)
    6901             : {
    6902          15 :         struct spdk_bdev_channel *ch = ctx;
    6903             : 
    6904          15 :         spdk_bdev_for_each_channel(ch->bdev, bdev_reset_freeze_channel, ch,
    6905             :                                    bdev_reset_freeze_channel_done);
    6906          15 : }
    6907             : 
    6908             : static void
    6909          16 : bdev_channel_start_reset(struct spdk_bdev_channel *ch)
    6910             : {
    6911          16 :         struct spdk_bdev *bdev = ch->bdev;
    6912             : 
    6913          16 :         assert(!TAILQ_EMPTY(&ch->queued_resets));
    6914             : 
    6915          16 :         spdk_spin_lock(&bdev->internal.spinlock);
    6916          16 :         if (bdev->internal.reset_in_progress == NULL) {
    6917          15 :                 bdev->internal.reset_in_progress = TAILQ_FIRST(&ch->queued_resets);
    6918             :                 /*
    6919             :                  * Take a channel reference for the target bdev for the life of this
    6920             :                  *  reset.  This guards against the channel getting destroyed while
    6921             :                  *  spdk_bdev_for_each_channel() calls related to this reset IO are in
    6922             :                  *  progress.  We will release the reference when this reset is
    6923             :                  *  completed.
    6924             :                  */
    6925          15 :                 bdev->internal.reset_in_progress->u.reset.ch_ref = spdk_get_io_channel(__bdev_to_io_dev(bdev));
    6926          15 :                 bdev_start_reset(ch);
    6927          15 :         }
    6928          16 :         spdk_spin_unlock(&bdev->internal.spinlock);
    6929          16 : }
    6930             : 
    6931             : int
    6932          16 : spdk_bdev_reset(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
    6933             :                 spdk_bdev_io_completion_cb cb, void *cb_arg)
    6934             : {
    6935          16 :         struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
    6936             :         struct spdk_bdev_io *bdev_io;
    6937          16 :         struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
    6938             : 
    6939          16 :         bdev_io = bdev_channel_get_io(channel);
    6940          16 :         if (!bdev_io) {
    6941           0 :                 return -ENOMEM;
    6942             :         }
    6943             : 
    6944          16 :         bdev_io->internal.ch = channel;
    6945          16 :         bdev_io->internal.desc = desc;
    6946          16 :         bdev_io->internal.submit_tsc = spdk_get_ticks();
    6947          16 :         bdev_io->type = SPDK_BDEV_IO_TYPE_RESET;
    6948          16 :         bdev_io->u.reset.ch_ref = NULL;
    6949          16 :         bdev_io_init(bdev_io, bdev, cb_arg, cb);
    6950             : 
    6951          16 :         spdk_spin_lock(&bdev->internal.spinlock);
    6952          16 :         TAILQ_INSERT_TAIL(&channel->queued_resets, bdev_io, internal.link);
    6953          16 :         spdk_spin_unlock(&bdev->internal.spinlock);
    6954             : 
    6955          16 :         bdev_ch_add_to_io_submitted(bdev_io);
    6956             : 
    6957          16 :         bdev_channel_start_reset(channel);
    6958             : 
    6959          16 :         return 0;
    6960          16 : }
    6961             : 
    6962             : void
    6963           0 : spdk_bdev_get_io_stat(struct spdk_bdev *bdev, struct spdk_io_channel *ch,
    6964             :                       struct spdk_bdev_io_stat *stat, enum spdk_bdev_reset_stat_mode reset_mode)
    6965             : {
    6966           0 :         struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
    6967             : 
    6968           0 :         bdev_get_io_stat(stat, channel->stat);
    6969           0 :         spdk_bdev_reset_io_stat(stat, reset_mode);
    6970           0 : }
    6971             : 
    6972             : static void
    6973           5 : bdev_get_device_stat_done(struct spdk_bdev *bdev, void *_ctx, int status)
    6974             : {
    6975           5 :         struct spdk_bdev_iostat_ctx *bdev_iostat_ctx = _ctx;
    6976             : 
    6977          10 :         bdev_iostat_ctx->cb(bdev, bdev_iostat_ctx->stat,
    6978           5 :                             bdev_iostat_ctx->cb_arg, 0);
    6979           5 :         free(bdev_iostat_ctx);
    6980           5 : }
    6981             : 
    6982             : static void
    6983           4 : bdev_get_each_channel_stat(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
    6984             :                            struct spdk_io_channel *ch, void *_ctx)
    6985             : {
    6986           4 :         struct spdk_bdev_iostat_ctx *bdev_iostat_ctx = _ctx;
    6987           4 :         struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
    6988             : 
    6989           4 :         spdk_bdev_add_io_stat(bdev_iostat_ctx->stat, channel->stat);
    6990           4 :         spdk_bdev_reset_io_stat(channel->stat, bdev_iostat_ctx->reset_mode);
    6991           4 :         spdk_bdev_for_each_channel_continue(i, 0);
    6992           4 : }
    6993             : 
    6994             : void
    6995           5 : spdk_bdev_get_device_stat(struct spdk_bdev *bdev, struct spdk_bdev_io_stat *stat,
    6996             :                           enum spdk_bdev_reset_stat_mode reset_mode, spdk_bdev_get_device_stat_cb cb, void *cb_arg)
    6997             : {
    6998             :         struct spdk_bdev_iostat_ctx *bdev_iostat_ctx;
    6999             : 
    7000           5 :         assert(bdev != NULL);
    7001           5 :         assert(stat != NULL);
    7002           5 :         assert(cb != NULL);
    7003             : 
    7004           5 :         bdev_iostat_ctx = calloc(1, sizeof(struct spdk_bdev_iostat_ctx));
    7005           5 :         if (bdev_iostat_ctx == NULL) {
    7006           0 :                 SPDK_ERRLOG("Unable to allocate memory for spdk_bdev_iostat_ctx\n");
    7007           0 :                 cb(bdev, stat, cb_arg, -ENOMEM);
    7008           0 :                 return;
    7009             :         }
    7010             : 
    7011           5 :         bdev_iostat_ctx->stat = stat;
    7012           5 :         bdev_iostat_ctx->cb = cb;
    7013           5 :         bdev_iostat_ctx->cb_arg = cb_arg;
    7014           5 :         bdev_iostat_ctx->reset_mode = reset_mode;
    7015             : 
    7016             :         /* Start with the statistics from previously deleted channels. */
    7017           5 :         spdk_spin_lock(&bdev->internal.spinlock);
    7018           5 :         bdev_get_io_stat(bdev_iostat_ctx->stat, bdev->internal.stat);
    7019           5 :         spdk_bdev_reset_io_stat(bdev->internal.stat, reset_mode);
    7020           5 :         spdk_spin_unlock(&bdev->internal.spinlock);
    7021             : 
    7022             :         /* Then iterate and add the statistics from each existing channel. */
    7023           5 :         spdk_bdev_for_each_channel(bdev, bdev_get_each_channel_stat, bdev_iostat_ctx,
    7024             :                                    bdev_get_device_stat_done);
    7025           5 : }
    7026             : 
    7027             : struct bdev_iostat_reset_ctx {
    7028             :         enum spdk_bdev_reset_stat_mode mode;
    7029             :         bdev_reset_device_stat_cb cb;
    7030             :         void *cb_arg;
    7031             : };
    7032             : 
    7033             : static void
    7034           0 : bdev_reset_device_stat_done(struct spdk_bdev *bdev, void *_ctx, int status)
    7035             : {
    7036           0 :         struct bdev_iostat_reset_ctx *ctx = _ctx;
    7037             : 
    7038           0 :         ctx->cb(bdev, ctx->cb_arg, 0);
    7039             : 
    7040           0 :         free(ctx);
    7041           0 : }
    7042             : 
    7043             : static void
    7044           0 : bdev_reset_each_channel_stat(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
    7045             :                              struct spdk_io_channel *ch, void *_ctx)
    7046             : {
    7047           0 :         struct bdev_iostat_reset_ctx *ctx = _ctx;
    7048           0 :         struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
    7049             : 
    7050           0 :         spdk_bdev_reset_io_stat(channel->stat, ctx->mode);
    7051             : 
    7052           0 :         spdk_bdev_for_each_channel_continue(i, 0);
    7053           0 : }
    7054             : 
    7055             : void
    7056           0 : bdev_reset_device_stat(struct spdk_bdev *bdev, enum spdk_bdev_reset_stat_mode mode,
    7057             :                        bdev_reset_device_stat_cb cb, void *cb_arg)
    7058             : {
    7059             :         struct bdev_iostat_reset_ctx *ctx;
    7060             : 
    7061           0 :         assert(bdev != NULL);
    7062           0 :         assert(cb != NULL);
    7063             : 
    7064           0 :         ctx = calloc(1, sizeof(*ctx));
    7065           0 :         if (ctx == NULL) {
    7066           0 :                 SPDK_ERRLOG("Unable to allocate bdev_iostat_reset_ctx.\n");
    7067           0 :                 cb(bdev, cb_arg, -ENOMEM);
    7068           0 :                 return;
    7069             :         }
    7070             : 
    7071           0 :         ctx->mode = mode;
    7072           0 :         ctx->cb = cb;
    7073           0 :         ctx->cb_arg = cb_arg;
    7074             : 
    7075           0 :         spdk_spin_lock(&bdev->internal.spinlock);
    7076           0 :         spdk_bdev_reset_io_stat(bdev->internal.stat, mode);
    7077           0 :         spdk_spin_unlock(&bdev->internal.spinlock);
    7078             : 
    7079           0 :         spdk_bdev_for_each_channel(bdev,
    7080             :                                    bdev_reset_each_channel_stat,
    7081           0 :                                    ctx,
    7082             :                                    bdev_reset_device_stat_done);
    7083           0 : }
    7084             : 
    7085             : int
    7086           1 : spdk_bdev_nvme_admin_passthru(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
    7087             :                               const struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes,
    7088             :                               spdk_bdev_io_completion_cb cb, void *cb_arg)
    7089             : {
    7090           1 :         struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
    7091             :         struct spdk_bdev_io *bdev_io;
    7092           1 :         struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
    7093             : 
    7094           1 :         if (!desc->write) {
    7095           0 :                 return -EBADF;
    7096             :         }
    7097             : 
    7098           1 :         if (spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_ADMIN))) {
    7099           1 :                 return -ENOTSUP;
    7100             :         }
    7101             : 
    7102           0 :         bdev_io = bdev_channel_get_io(channel);
    7103           0 :         if (!bdev_io) {
    7104           0 :                 return -ENOMEM;
    7105             :         }
    7106             : 
    7107           0 :         bdev_io->internal.ch = channel;
    7108           0 :         bdev_io->internal.desc = desc;
    7109           0 :         bdev_io->type = SPDK_BDEV_IO_TYPE_NVME_ADMIN;
    7110           0 :         bdev_io->u.nvme_passthru.cmd = *cmd;
    7111           0 :         bdev_io->u.nvme_passthru.buf = buf;
    7112           0 :         bdev_io->u.nvme_passthru.nbytes = nbytes;
    7113           0 :         bdev_io->u.nvme_passthru.md_buf = NULL;
    7114           0 :         bdev_io->u.nvme_passthru.md_len = 0;
    7115             : 
    7116           0 :         bdev_io_init(bdev_io, bdev, cb_arg, cb);
    7117             : 
    7118           0 :         bdev_io_submit(bdev_io);
    7119           0 :         return 0;
    7120           1 : }
    7121             : 
    7122             : int
    7123           1 : spdk_bdev_nvme_io_passthru(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
    7124             :                            const struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes,
    7125             :                            spdk_bdev_io_completion_cb cb, void *cb_arg)
    7126             : {
    7127           1 :         struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
    7128             :         struct spdk_bdev_io *bdev_io;
    7129           1 :         struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
    7130             : 
    7131           1 :         if (!desc->write) {
    7132             :                 /*
    7133             :                  * Do not try to parse the NVMe command - we could maybe use bits in the opcode
    7134             :                  *  to easily determine if the command is a read or write, but for now just
    7135             :                  *  do not allow io_passthru with a read-only descriptor.
    7136             :                  */
    7137           0 :                 return -EBADF;
    7138             :         }
    7139             : 
    7140           1 :         if (spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_IO))) {
    7141           1 :                 return -ENOTSUP;
    7142             :         }
    7143             : 
    7144           0 :         bdev_io = bdev_channel_get_io(channel);
    7145           0 :         if (!bdev_io) {
    7146           0 :                 return -ENOMEM;
    7147             :         }
    7148             : 
    7149           0 :         bdev_io->internal.ch = channel;
    7150           0 :         bdev_io->internal.desc = desc;
    7151           0 :         bdev_io->type = SPDK_BDEV_IO_TYPE_NVME_IO;
    7152           0 :         bdev_io->u.nvme_passthru.cmd = *cmd;
    7153           0 :         bdev_io->u.nvme_passthru.buf = buf;
    7154           0 :         bdev_io->u.nvme_passthru.nbytes = nbytes;
    7155           0 :         bdev_io->u.nvme_passthru.md_buf = NULL;
    7156           0 :         bdev_io->u.nvme_passthru.md_len = 0;
    7157             : 
    7158           0 :         bdev_io_init(bdev_io, bdev, cb_arg, cb);
    7159             : 
    7160           0 :         bdev_io_submit(bdev_io);
    7161           0 :         return 0;
    7162           1 : }
    7163             : 
    7164             : int
    7165           1 : spdk_bdev_nvme_io_passthru_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
    7166             :                               const struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes, void *md_buf, size_t md_len,
    7167             :                               spdk_bdev_io_completion_cb cb, void *cb_arg)
    7168             : {
    7169           1 :         struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
    7170             :         struct spdk_bdev_io *bdev_io;
    7171           1 :         struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
    7172             : 
    7173           1 :         if (!desc->write) {
    7174             :                 /*
    7175             :                  * Do not try to parse the NVMe command - we could maybe use bits in the opcode
    7176             :                  *  to easily determine if the command is a read or write, but for now just
    7177             :                  *  do not allow io_passthru with a read-only descriptor.
    7178             :                  */
    7179           0 :                 return -EBADF;
    7180             :         }
    7181             : 
    7182           1 :         if (spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_IO_MD))) {
    7183           1 :                 return -ENOTSUP;
    7184             :         }
    7185             : 
    7186           0 :         bdev_io = bdev_channel_get_io(channel);
    7187           0 :         if (!bdev_io) {
    7188           0 :                 return -ENOMEM;
    7189             :         }
    7190             : 
    7191           0 :         bdev_io->internal.ch = channel;
    7192           0 :         bdev_io->internal.desc = desc;
    7193           0 :         bdev_io->type = SPDK_BDEV_IO_TYPE_NVME_IO_MD;
    7194           0 :         bdev_io->u.nvme_passthru.cmd = *cmd;
    7195           0 :         bdev_io->u.nvme_passthru.buf = buf;
    7196           0 :         bdev_io->u.nvme_passthru.nbytes = nbytes;
    7197           0 :         bdev_io->u.nvme_passthru.md_buf = md_buf;
    7198           0 :         bdev_io->u.nvme_passthru.md_len = md_len;
    7199             : 
    7200           0 :         bdev_io_init(bdev_io, bdev, cb_arg, cb);
    7201             : 
    7202           0 :         bdev_io_submit(bdev_io);
    7203           0 :         return 0;
    7204           1 : }
    7205             : 
    7206             : int
    7207           0 : spdk_bdev_nvme_iov_passthru_md(struct spdk_bdev_desc *desc,
    7208             :                                struct spdk_io_channel *ch,
    7209             :                                const struct spdk_nvme_cmd *cmd,
    7210             :                                struct iovec *iov, int iovcnt, size_t nbytes,
    7211             :                                void *md_buf, size_t md_len,
    7212             :                                spdk_bdev_io_completion_cb cb, void *cb_arg)
    7213             : {
    7214           0 :         struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
    7215             :         struct spdk_bdev_io *bdev_io;
    7216           0 :         struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
    7217             : 
    7218           0 :         if (!desc->write) {
    7219             :                 /*
    7220             :                  * Do not try to parse the NVMe command - we could maybe use bits in the opcode
    7221             :                  * to easily determine if the command is a read or write, but for now just
    7222             :                  * do not allow io_passthru with a read-only descriptor.
    7223             :                  */
    7224           0 :                 return -EBADF;
    7225             :         }
    7226             : 
    7227           0 :         if (md_buf && spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_IO_MD))) {
    7228           0 :                 return -ENOTSUP;
    7229           0 :         } else if (spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_IO))) {
    7230           0 :                 return -ENOTSUP;
    7231             :         }
    7232             : 
    7233           0 :         bdev_io = bdev_channel_get_io(channel);
    7234           0 :         if (!bdev_io) {
    7235           0 :                 return -ENOMEM;
    7236             :         }
    7237             : 
    7238           0 :         bdev_io->internal.ch = channel;
    7239           0 :         bdev_io->internal.desc = desc;
    7240           0 :         bdev_io->type = SPDK_BDEV_IO_TYPE_NVME_IOV_MD;
    7241           0 :         bdev_io->u.nvme_passthru.cmd = *cmd;
    7242           0 :         bdev_io->u.nvme_passthru.iovs = iov;
    7243           0 :         bdev_io->u.nvme_passthru.iovcnt = iovcnt;
    7244           0 :         bdev_io->u.nvme_passthru.nbytes = nbytes;
    7245           0 :         bdev_io->u.nvme_passthru.md_buf = md_buf;
    7246           0 :         bdev_io->u.nvme_passthru.md_len = md_len;
    7247             : 
    7248           0 :         bdev_io_init(bdev_io, bdev, cb_arg, cb);
    7249             : 
    7250           0 :         bdev_io_submit(bdev_io);
    7251           0 :         return 0;
    7252           0 : }
    7253             : 
    7254             : static void bdev_abort_retry(void *ctx);
    7255             : static void bdev_abort(struct spdk_bdev_io *parent_io);
    7256             : 
    7257             : static void
    7258          22 : bdev_abort_io_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
    7259             : {
    7260          22 :         struct spdk_bdev_channel *channel = bdev_io->internal.ch;
    7261          22 :         struct spdk_bdev_io *parent_io = cb_arg;
    7262             :         struct spdk_bdev_io *bio_to_abort, *tmp_io;
    7263             : 
    7264          22 :         bio_to_abort = bdev_io->u.abort.bio_to_abort;
    7265             : 
    7266          22 :         spdk_bdev_free_io(bdev_io);
    7267             : 
    7268          22 :         if (!success) {
    7269             :                 /* Check if the target I/O completed in the meantime. */
    7270           2 :                 TAILQ_FOREACH(tmp_io, &channel->io_submitted, internal.ch_link) {
    7271           1 :                         if (tmp_io == bio_to_abort) {
    7272           0 :                                 break;
    7273             :                         }
    7274           1 :                 }
    7275             : 
    7276             :                 /* If the target I/O still exists, set the parent to failed. */
    7277           1 :                 if (tmp_io != NULL) {
    7278           0 :                         parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
    7279           0 :                 }
    7280           1 :         }
    7281             : 
    7282          22 :         assert(parent_io->internal.f.split);
    7283             : 
    7284          22 :         parent_io->internal.split.outstanding--;
    7285          22 :         if (parent_io->internal.split.outstanding == 0) {
    7286          16 :                 if (parent_io->internal.status == SPDK_BDEV_IO_STATUS_NOMEM) {
    7287           0 :                         bdev_abort_retry(parent_io);
    7288           0 :                 } else {
    7289          16 :                         bdev_io_complete(parent_io);
    7290             :                 }
    7291          16 :         }
    7292          22 : }
    7293             : 
    7294             : static int
    7295          23 : bdev_abort_io(struct spdk_bdev_desc *desc, struct spdk_bdev_channel *channel,
    7296             :               struct spdk_bdev_io *bio_to_abort,
    7297             :               spdk_bdev_io_completion_cb cb, void *cb_arg)
    7298             : {
    7299          23 :         struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
    7300             :         struct spdk_bdev_io *bdev_io;
    7301             : 
    7302          23 :         if (bio_to_abort->type == SPDK_BDEV_IO_TYPE_ABORT ||
    7303          23 :             bio_to_abort->type == SPDK_BDEV_IO_TYPE_RESET) {
    7304             :                 /* TODO: Abort reset or abort request. */
    7305           0 :                 return -ENOTSUP;
    7306             :         }
    7307             : 
    7308          23 :         bdev_io = bdev_channel_get_io(channel);
    7309          23 :         if (bdev_io == NULL) {
    7310           1 :                 return -ENOMEM;
    7311             :         }
    7312             : 
    7313          22 :         bdev_io->internal.ch = channel;
    7314          22 :         bdev_io->internal.desc = desc;
    7315          22 :         bdev_io->type = SPDK_BDEV_IO_TYPE_ABORT;
    7316          22 :         bdev_io_init(bdev_io, bdev, cb_arg, cb);
    7317             : 
    7318          22 :         if (bio_to_abort->internal.f.split) {
    7319           6 :                 assert(bdev_io_should_split(bio_to_abort));
    7320           6 :                 bdev_io->u.bdev.abort.bio_cb_arg = bio_to_abort;
    7321             : 
    7322             :                 /* Parent abort request is not submitted directly, but to manage its
    7323             :                  * execution add it to the submitted list here.
    7324             :                  */
    7325           6 :                 bdev_io->internal.submit_tsc = spdk_get_ticks();
    7326           6 :                 bdev_ch_add_to_io_submitted(bdev_io);
    7327             : 
    7328           6 :                 bdev_abort(bdev_io);
    7329             : 
    7330           6 :                 return 0;
    7331             :         }
    7332             : 
    7333          16 :         bdev_io->u.abort.bio_to_abort = bio_to_abort;
    7334             : 
    7335             :         /* Submit the abort request to the underlying bdev module. */
    7336          16 :         bdev_io_submit(bdev_io);
    7337             : 
    7338          16 :         return 0;
    7339          23 : }
    7340             : 
    7341             : static bool
    7342          46 : bdev_io_on_tailq(struct spdk_bdev_io *bdev_io, bdev_io_tailq_t *tailq)
    7343             : {
    7344             :         struct spdk_bdev_io *iter;
    7345             : 
    7346          46 :         TAILQ_FOREACH(iter, tailq, internal.link) {
    7347           0 :                 if (iter == bdev_io) {
    7348           0 :                         return true;
    7349             :                 }
    7350           0 :         }
    7351             : 
    7352          46 :         return false;
    7353          46 : }
    7354             : 
    7355             : static uint32_t
    7356          18 : _bdev_abort(struct spdk_bdev_io *parent_io)
    7357             : {
    7358          18 :         struct spdk_bdev_desc *desc = parent_io->internal.desc;
    7359          18 :         struct spdk_bdev_channel *channel = parent_io->internal.ch;
    7360             :         void *bio_cb_arg;
    7361             :         struct spdk_bdev_io *bio_to_abort;
    7362             :         uint32_t matched_ios;
    7363             :         int rc;
    7364             : 
    7365          18 :         bio_cb_arg = parent_io->u.bdev.abort.bio_cb_arg;
    7366             : 
    7367             :         /* matched_ios is returned and will be kept by the caller.
    7368             :          *
    7369             :          * This function will be used for two cases, 1) the same cb_arg is used for
    7370             :          * multiple I/Os, 2) a single large I/O is split into smaller ones.
    7371             :          * Incrementing split_outstanding directly here may confuse readers especially
    7372             :          * for the 1st case.
    7373             :          *
    7374             :          * Completion of I/O abort is processed after stack unwinding. Hence this trick
    7375             :          * works as expected.
    7376             :          */
    7377          18 :         matched_ios = 0;
    7378          18 :         parent_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
    7379             : 
    7380         105 :         TAILQ_FOREACH(bio_to_abort, &channel->io_submitted, internal.ch_link) {
    7381          88 :                 if (bio_to_abort->internal.caller_ctx != bio_cb_arg) {
    7382          65 :                         continue;
    7383             :                 }
    7384             : 
    7385          23 :                 if (bio_to_abort->internal.submit_tsc > parent_io->internal.submit_tsc) {
    7386             :                         /* Any I/O which was submitted after this abort command should be excluded. */
    7387           0 :                         continue;
    7388             :                 }
    7389             : 
    7390             :                 /* We can't abort a request that's being pushed/pulled or executed by accel */
    7391          23 :                 if (bdev_io_on_tailq(bio_to_abort, &channel->io_accel_exec) ||
    7392          23 :                     bdev_io_on_tailq(bio_to_abort, &channel->io_memory_domain)) {
    7393           0 :                         parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
    7394           0 :                         break;
    7395             :                 }
    7396             : 
    7397          23 :                 rc = bdev_abort_io(desc, channel, bio_to_abort, bdev_abort_io_done, parent_io);
    7398          23 :                 if (rc != 0) {
    7399           1 :                         if (rc == -ENOMEM) {
    7400           1 :                                 parent_io->internal.status = SPDK_BDEV_IO_STATUS_NOMEM;
    7401           1 :                         } else {
    7402           0 :                                 parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
    7403             :                         }
    7404           1 :                         break;
    7405             :                 }
    7406          22 :                 matched_ios++;
    7407          22 :         }
    7408             : 
    7409          18 :         return matched_ios;
    7410             : }
    7411             : 
    7412             : static void
    7413           1 : bdev_abort_retry(void *ctx)
    7414             : {
    7415           1 :         struct spdk_bdev_io *parent_io = ctx;
    7416             :         uint32_t matched_ios;
    7417             : 
    7418           1 :         matched_ios = _bdev_abort(parent_io);
    7419             : 
    7420           1 :         if (matched_ios == 0) {
    7421           0 :                 if (parent_io->internal.status == SPDK_BDEV_IO_STATUS_NOMEM) {
    7422           0 :                         bdev_queue_io_wait_with_cb(parent_io, bdev_abort_retry);
    7423           0 :                 } else {
    7424             :                         /* For retry, the case that no target I/O was found is success
    7425             :                          * because it means target I/Os completed in the meantime.
    7426             :                          */
    7427           0 :                         bdev_io_complete(parent_io);
    7428             :                 }
    7429           0 :                 return;
    7430             :         }
    7431             : 
    7432             :         /* Use split_outstanding to manage the progress of aborting I/Os. */
    7433           1 :         parent_io->internal.f.split = true;
    7434           1 :         parent_io->internal.split.outstanding = matched_ios;
    7435           1 : }
    7436             : 
    7437             : static void
    7438          17 : bdev_abort(struct spdk_bdev_io *parent_io)
    7439             : {
    7440             :         uint32_t matched_ios;
    7441             : 
    7442          17 :         matched_ios = _bdev_abort(parent_io);
    7443             : 
    7444          17 :         if (matched_ios == 0) {
    7445           2 :                 if (parent_io->internal.status == SPDK_BDEV_IO_STATUS_NOMEM) {
    7446           1 :                         bdev_queue_io_wait_with_cb(parent_io, bdev_abort_retry);
    7447           1 :                 } else {
    7448             :                         /* The case the no target I/O was found is failure. */
    7449           1 :                         parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
    7450           1 :                         bdev_io_complete(parent_io);
    7451             :                 }
    7452           2 :                 return;
    7453             :         }
    7454             : 
    7455             :         /* Use split_outstanding to manage the progress of aborting I/Os. */
    7456          15 :         parent_io->internal.f.split = true;
    7457          15 :         parent_io->internal.split.outstanding = matched_ios;
    7458          17 : }
    7459             : 
    7460             : int
    7461          12 : spdk_bdev_abort(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
    7462             :                 void *bio_cb_arg,
    7463             :                 spdk_bdev_io_completion_cb cb, void *cb_arg)
    7464             : {
    7465          12 :         struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
    7466          12 :         struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
    7467             :         struct spdk_bdev_io *bdev_io;
    7468             : 
    7469          12 :         if (bio_cb_arg == NULL) {
    7470           0 :                 return -EINVAL;
    7471             :         }
    7472             : 
    7473          12 :         if (!spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_ABORT)) {
    7474           1 :                 return -ENOTSUP;
    7475             :         }
    7476             : 
    7477          11 :         bdev_io = bdev_channel_get_io(channel);
    7478          11 :         if (bdev_io == NULL) {
    7479           0 :                 return -ENOMEM;
    7480             :         }
    7481             : 
    7482          11 :         bdev_io->internal.ch = channel;
    7483          11 :         bdev_io->internal.desc = desc;
    7484          11 :         bdev_io->internal.submit_tsc = spdk_get_ticks();
    7485          11 :         bdev_io->type = SPDK_BDEV_IO_TYPE_ABORT;
    7486          11 :         bdev_io_init(bdev_io, bdev, cb_arg, cb);
    7487             : 
    7488          11 :         bdev_io->u.bdev.abort.bio_cb_arg = bio_cb_arg;
    7489             : 
    7490             :         /* Parent abort request is not submitted directly, but to manage its execution,
    7491             :          * add it to the submitted list here.
    7492             :          */
    7493          11 :         bdev_ch_add_to_io_submitted(bdev_io);
    7494             : 
    7495          11 :         bdev_abort(bdev_io);
    7496             : 
    7497          11 :         return 0;
    7498          12 : }
    7499             : 
    7500             : int
    7501           4 : spdk_bdev_queue_io_wait(struct spdk_bdev *bdev, struct spdk_io_channel *ch,
    7502             :                         struct spdk_bdev_io_wait_entry *entry)
    7503             : {
    7504           4 :         struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
    7505           4 :         struct spdk_bdev_mgmt_channel *mgmt_ch = channel->shared_resource->mgmt_ch;
    7506             : 
    7507           4 :         if (bdev != entry->bdev) {
    7508           0 :                 SPDK_ERRLOG("bdevs do not match\n");
    7509           0 :                 return -EINVAL;
    7510             :         }
    7511             : 
    7512           4 :         if (mgmt_ch->per_thread_cache_count > 0) {
    7513           0 :                 SPDK_ERRLOG("Cannot queue io_wait if spdk_bdev_io available in per-thread cache\n");
    7514           0 :                 return -EINVAL;
    7515             :         }
    7516             : 
    7517           4 :         TAILQ_INSERT_TAIL(&mgmt_ch->io_wait_queue, entry, link);
    7518           4 :         return 0;
    7519           4 : }
    7520             : 
    7521             : static inline void
    7522         610 : bdev_io_update_io_stat(struct spdk_bdev_io *bdev_io, uint64_t tsc_diff)
    7523             : {
    7524         610 :         enum spdk_bdev_io_status io_status = bdev_io->internal.status;
    7525         610 :         struct spdk_bdev_io_stat *io_stat = bdev_io->internal.ch->stat;
    7526         610 :         uint64_t num_blocks = bdev_io->u.bdev.num_blocks;
    7527         610 :         uint32_t blocklen = bdev_io->bdev->blocklen;
    7528             : 
    7529         610 :         if (spdk_likely(io_status == SPDK_BDEV_IO_STATUS_SUCCESS)) {
    7530         516 :                 switch (bdev_io->type) {
    7531             :                 case SPDK_BDEV_IO_TYPE_READ:
    7532         321 :                         io_stat->bytes_read += num_blocks * blocklen;
    7533         321 :                         io_stat->num_read_ops++;
    7534         321 :                         io_stat->read_latency_ticks += tsc_diff;
    7535         321 :                         if (io_stat->max_read_latency_ticks < tsc_diff) {
    7536           7 :                                 io_stat->max_read_latency_ticks = tsc_diff;
    7537           7 :                         }
    7538         321 :                         if (io_stat->min_read_latency_ticks > tsc_diff) {
    7539          42 :                                 io_stat->min_read_latency_ticks = tsc_diff;
    7540          42 :                         }
    7541         321 :                         break;
    7542             :                 case SPDK_BDEV_IO_TYPE_WRITE:
    7543          75 :                         io_stat->bytes_written += num_blocks * blocklen;
    7544          75 :                         io_stat->num_write_ops++;
    7545          75 :                         io_stat->write_latency_ticks += tsc_diff;
    7546          75 :                         if (io_stat->max_write_latency_ticks < tsc_diff) {
    7547           4 :                                 io_stat->max_write_latency_ticks = tsc_diff;
    7548           4 :                         }
    7549          75 :                         if (io_stat->min_write_latency_ticks > tsc_diff) {
    7550          25 :                                 io_stat->min_write_latency_ticks = tsc_diff;
    7551          25 :                         }
    7552          75 :                         break;
    7553             :                 case SPDK_BDEV_IO_TYPE_UNMAP:
    7554          20 :                         io_stat->bytes_unmapped += num_blocks * blocklen;
    7555          20 :                         io_stat->num_unmap_ops++;
    7556          20 :                         io_stat->unmap_latency_ticks += tsc_diff;
    7557          20 :                         if (io_stat->max_unmap_latency_ticks < tsc_diff) {
    7558           0 :                                 io_stat->max_unmap_latency_ticks = tsc_diff;
    7559           0 :                         }
    7560          20 :                         if (io_stat->min_unmap_latency_ticks > tsc_diff) {
    7561           3 :                                 io_stat->min_unmap_latency_ticks = tsc_diff;
    7562           3 :                         }
    7563          20 :                         break;
    7564             :                 case SPDK_BDEV_IO_TYPE_ZCOPY:
    7565             :                         /* Track the data in the start phase only */
    7566           4 :                         if (bdev_io->u.bdev.zcopy.start) {
    7567           2 :                                 if (bdev_io->u.bdev.zcopy.populate) {
    7568           1 :                                         io_stat->bytes_read += num_blocks * blocklen;
    7569           1 :                                         io_stat->num_read_ops++;
    7570           1 :                                         io_stat->read_latency_ticks += tsc_diff;
    7571           1 :                                         if (io_stat->max_read_latency_ticks < tsc_diff) {
    7572           0 :                                                 io_stat->max_read_latency_ticks = tsc_diff;
    7573           0 :                                         }
    7574           1 :                                         if (io_stat->min_read_latency_ticks > tsc_diff) {
    7575           1 :                                                 io_stat->min_read_latency_ticks = tsc_diff;
    7576           1 :                                         }
    7577           1 :                                 } else {
    7578           1 :                                         io_stat->bytes_written += num_blocks * blocklen;
    7579           1 :                                         io_stat->num_write_ops++;
    7580           1 :                                         io_stat->write_latency_ticks += tsc_diff;
    7581           1 :                                         if (io_stat->max_write_latency_ticks < tsc_diff) {
    7582           0 :                                                 io_stat->max_write_latency_ticks = tsc_diff;
    7583           0 :                                         }
    7584           1 :                                         if (io_stat->min_write_latency_ticks > tsc_diff) {
    7585           1 :                                                 io_stat->min_write_latency_ticks = tsc_diff;
    7586           1 :                                         }
    7587             :                                 }
    7588           2 :                         }
    7589           4 :                         break;
    7590             :                 case SPDK_BDEV_IO_TYPE_COPY:
    7591          21 :                         io_stat->bytes_copied += num_blocks * blocklen;
    7592          21 :                         io_stat->num_copy_ops++;
    7593          21 :                         bdev_io->internal.ch->stat->copy_latency_ticks += tsc_diff;
    7594          21 :                         if (io_stat->max_copy_latency_ticks < tsc_diff) {
    7595           0 :                                 io_stat->max_copy_latency_ticks = tsc_diff;
    7596           0 :                         }
    7597          21 :                         if (io_stat->min_copy_latency_ticks > tsc_diff) {
    7598           4 :                                 io_stat->min_copy_latency_ticks = tsc_diff;
    7599           4 :                         }
    7600          21 :                         break;
    7601             :                 default:
    7602          75 :                         break;
    7603             :                 }
    7604         610 :         } else if (io_status <= SPDK_BDEV_IO_STATUS_FAILED && io_status >= SPDK_MIN_BDEV_IO_STATUS) {
    7605          94 :                 io_stat = bdev_io->bdev->internal.stat;
    7606          94 :                 assert(io_stat->io_error != NULL);
    7607             : 
    7608          94 :                 spdk_spin_lock(&bdev_io->bdev->internal.spinlock);
    7609          94 :                 io_stat->io_error->error_status[-io_status - 1]++;
    7610          94 :                 spdk_spin_unlock(&bdev_io->bdev->internal.spinlock);
    7611          94 :         }
    7612             : 
    7613             : #ifdef SPDK_CONFIG_VTUNE
    7614             :         uint64_t now_tsc = spdk_get_ticks();
    7615             :         if (now_tsc > (bdev_io->internal.ch->start_tsc + bdev_io->internal.ch->interval_tsc)) {
    7616             :                 uint64_t data[5];
    7617             :                 struct spdk_bdev_io_stat *prev_stat = bdev_io->internal.ch->prev_stat;
    7618             : 
    7619             :                 data[0] = io_stat->num_read_ops - prev_stat->num_read_ops;
    7620             :                 data[1] = io_stat->bytes_read - prev_stat->bytes_read;
    7621             :                 data[2] = io_stat->num_write_ops - prev_stat->num_write_ops;
    7622             :                 data[3] = io_stat->bytes_written - prev_stat->bytes_written;
    7623             :                 data[4] = bdev_io->bdev->fn_table->get_spin_time ?
    7624             :                           bdev_io->bdev->fn_table->get_spin_time(spdk_bdev_io_get_io_channel(bdev_io)) : 0;
    7625             : 
    7626             :                 __itt_metadata_add(g_bdev_mgr.domain, __itt_null, bdev_io->internal.ch->handle,
    7627             :                                    __itt_metadata_u64, 5, data);
    7628             : 
    7629             :                 memcpy(prev_stat, io_stat, sizeof(struct spdk_bdev_io_stat));
    7630             :                 bdev_io->internal.ch->start_tsc = now_tsc;
    7631             :         }
    7632             : #endif
    7633         610 : }
    7634             : 
    7635             : static inline void
    7636         610 : _bdev_io_complete(void *ctx)
    7637             : {
    7638         610 :         struct spdk_bdev_io *bdev_io = ctx;
    7639             : 
    7640         610 :         if (spdk_unlikely(bdev_io_use_accel_sequence(bdev_io))) {
    7641           0 :                 assert(bdev_io->internal.status != SPDK_BDEV_IO_STATUS_SUCCESS);
    7642           0 :                 spdk_accel_sequence_abort(bdev_io->internal.accel_sequence);
    7643           0 :         }
    7644             : 
    7645         610 :         assert(bdev_io->internal.cb != NULL);
    7646         610 :         assert(spdk_get_thread() == spdk_bdev_io_get_thread(bdev_io));
    7647             : 
    7648        1220 :         bdev_io->internal.cb(bdev_io, bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS,
    7649         610 :                              bdev_io->internal.caller_ctx);
    7650         610 : }
    7651             : 
    7652             : static inline void
    7653         618 : bdev_io_complete(void *ctx)
    7654             : {
    7655         618 :         struct spdk_bdev_io *bdev_io = ctx;
    7656         618 :         struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
    7657             :         uint64_t tsc, tsc_diff;
    7658             : 
    7659         618 :         if (spdk_unlikely(bdev_io->internal.f.in_submit_request)) {
    7660             :                 /*
    7661             :                  * Defer completion to avoid potential infinite recursion if the
    7662             :                  * user's completion callback issues a new I/O.
    7663             :                  */
    7664          16 :                 spdk_thread_send_msg(spdk_bdev_io_get_thread(bdev_io),
    7665           8 :                                      bdev_io_complete, bdev_io);
    7666           8 :                 return;
    7667             :         }
    7668             : 
    7669         610 :         tsc = spdk_get_ticks();
    7670         610 :         tsc_diff = tsc - bdev_io->internal.submit_tsc;
    7671             : 
    7672         610 :         bdev_ch_remove_from_io_submitted(bdev_io);
    7673         610 :         spdk_trace_record_tsc(tsc, TRACE_BDEV_IO_DONE, bdev_ch->trace_id, 0, (uintptr_t)bdev_io,
    7674             :                               bdev_io->internal.caller_ctx, bdev_ch->queue_depth);
    7675             : 
    7676         610 :         if (bdev_ch->histogram) {
    7677           4 :                 if (bdev_io->bdev->internal.histogram_io_type == 0 ||
    7678           0 :                     bdev_io->bdev->internal.histogram_io_type == bdev_io->type) {
    7679             :                         /*
    7680             :                          * Tally all I/O types if the histogram_io_type is set to 0.
    7681             :                          */
    7682           4 :                         spdk_histogram_data_tally(bdev_ch->histogram, tsc_diff);
    7683           4 :                 }
    7684           4 :         }
    7685             : 
    7686         610 :         bdev_io_update_io_stat(bdev_io, tsc_diff);
    7687         610 :         _bdev_io_complete(bdev_io);
    7688         618 : }
    7689             : 
    7690             : /* The difference between this function and bdev_io_complete() is that this should be called to
    7691             :  * complete IOs that haven't been submitted via bdev_io_submit(), as they weren't added onto the
    7692             :  * io_submitted list and don't have submit_tsc updated.
    7693             :  */
    7694             : static inline void
    7695           0 : bdev_io_complete_unsubmitted(struct spdk_bdev_io *bdev_io)
    7696             : {
    7697             :         /* Since the IO hasn't been submitted it's bound to be failed */
    7698           0 :         assert(bdev_io->internal.status != SPDK_BDEV_IO_STATUS_SUCCESS);
    7699             : 
    7700             :         /* At this point we don't know if the IO is completed from submission context or not, but,
    7701             :          * since this is an error path, we can always do an spdk_thread_send_msg(). */
    7702           0 :         spdk_thread_send_msg(spdk_bdev_io_get_thread(bdev_io),
    7703           0 :                              _bdev_io_complete, bdev_io);
    7704           0 : }
    7705             : 
    7706             : static void bdev_destroy_cb(void *io_device);
    7707             : 
    7708             : static void
    7709          15 : bdev_reset_complete(struct spdk_bdev *bdev, void *_ctx, int status)
    7710             : {
    7711          15 :         struct spdk_bdev_io *bdev_io = _ctx;
    7712             : 
    7713          15 :         if (bdev_io->u.reset.ch_ref != NULL) {
    7714          15 :                 spdk_put_io_channel(bdev_io->u.reset.ch_ref);
    7715          15 :                 bdev_io->u.reset.ch_ref = NULL;
    7716          15 :         }
    7717             : 
    7718          15 :         bdev_io_complete(bdev_io);
    7719             : 
    7720          15 :         if (bdev->internal.status == SPDK_BDEV_STATUS_REMOVING &&
    7721           1 :             TAILQ_EMPTY(&bdev->internal.open_descs)) {
    7722           1 :                 spdk_io_device_unregister(__bdev_to_io_dev(bdev), bdev_destroy_cb);
    7723           1 :         }
    7724          15 : }
    7725             : 
    7726             : static void
    7727          18 : bdev_unfreeze_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
    7728             :                       struct spdk_io_channel *_ch, void *_ctx)
    7729             : {
    7730          18 :         struct spdk_bdev_io *bdev_io = _ctx;
    7731          18 :         struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
    7732             :         struct spdk_bdev_io *queued_reset;
    7733             : 
    7734          18 :         ch->flags &= ~BDEV_CH_RESET_IN_PROGRESS;
    7735          18 :         while (!TAILQ_EMPTY(&ch->queued_resets)) {
    7736           0 :                 queued_reset = TAILQ_FIRST(&ch->queued_resets);
    7737           0 :                 TAILQ_REMOVE(&ch->queued_resets, queued_reset, internal.link);
    7738           0 :                 spdk_bdev_io_complete(queued_reset, bdev_io->internal.status);
    7739             :         }
    7740             : 
    7741          18 :         spdk_bdev_for_each_channel_continue(i, 0);
    7742          18 : }
    7743             : 
    7744             : static void
    7745           0 : bdev_io_complete_sequence_cb(void *ctx, int status)
    7746             : {
    7747           0 :         struct spdk_bdev_io *bdev_io = ctx;
    7748             : 
    7749             :         /* u.bdev.accel_sequence should have already been cleared at this point */
    7750           0 :         assert(bdev_io->u.bdev.accel_sequence == NULL);
    7751           0 :         assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS);
    7752           0 :         bdev_io->internal.f.has_accel_sequence = false;
    7753             : 
    7754           0 :         if (spdk_unlikely(status != 0)) {
    7755           0 :                 SPDK_ERRLOG("Failed to execute accel sequence, status=%d\n", status);
    7756           0 :                 bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
    7757           0 :         }
    7758             : 
    7759           0 :         bdev_io_complete(bdev_io);
    7760           0 : }
    7761             : 
    7762             : void
    7763         598 : spdk_bdev_io_complete(struct spdk_bdev_io *bdev_io, enum spdk_bdev_io_status status)
    7764             : {
    7765         598 :         struct spdk_bdev *bdev = bdev_io->bdev;
    7766         598 :         struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
    7767         598 :         struct spdk_bdev_shared_resource *shared_resource = bdev_ch->shared_resource;
    7768             : 
    7769         598 :         if (spdk_unlikely(bdev_io->internal.status != SPDK_BDEV_IO_STATUS_PENDING)) {
    7770           0 :                 SPDK_ERRLOG("Unexpected completion on IO from %s module, status was %s\n",
    7771             :                             spdk_bdev_get_module_name(bdev),
    7772             :                             bdev_io_status_get_string(bdev_io->internal.status));
    7773           0 :                 assert(false);
    7774             :         }
    7775         598 :         bdev_io->internal.status = status;
    7776             : 
    7777         598 :         if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_RESET)) {
    7778          16 :                 bool unlock_channels = false;
    7779             : 
    7780          16 :                 if (status == SPDK_BDEV_IO_STATUS_NOMEM) {
    7781           0 :                         SPDK_ERRLOG("NOMEM returned for reset\n");
    7782           0 :                 }
    7783          16 :                 spdk_spin_lock(&bdev->internal.spinlock);
    7784          16 :                 if (bdev_io == bdev->internal.reset_in_progress) {
    7785          15 :                         bdev->internal.reset_in_progress = NULL;
    7786          15 :                         unlock_channels = true;
    7787          15 :                 }
    7788          16 :                 spdk_spin_unlock(&bdev->internal.spinlock);
    7789             : 
    7790          16 :                 if (unlock_channels) {
    7791          15 :                         spdk_bdev_for_each_channel(bdev, bdev_unfreeze_channel, bdev_io,
    7792             :                                                    bdev_reset_complete);
    7793          15 :                         return;
    7794             :                 }
    7795           1 :         } else {
    7796         582 :                 bdev_io_decrement_outstanding(bdev_ch, shared_resource);
    7797         582 :                 if (spdk_likely(status == SPDK_BDEV_IO_STATUS_SUCCESS)) {
    7798         485 :                         if (bdev_io_needs_sequence_exec(bdev_io->internal.desc, bdev_io)) {
    7799           0 :                                 bdev_io_exec_sequence(bdev_io, bdev_io_complete_sequence_cb);
    7800           0 :                                 return;
    7801         485 :                         } else if (spdk_unlikely(bdev_io->internal.f.has_bounce_buf &&
    7802             :                                                  !bdev_io_use_accel_sequence(bdev_io))) {
    7803          26 :                                 _bdev_io_push_bounce_data_buffer(bdev_io,
    7804             :                                                                  _bdev_io_complete_push_bounce_done);
    7805             :                                 /* bdev IO will be completed in the callback */
    7806          26 :                                 return;
    7807             :                         }
    7808         459 :                 }
    7809             : 
    7810         556 :                 if (spdk_unlikely(_bdev_io_handle_no_mem(bdev_io, BDEV_IO_RETRY_STATE_SUBMIT))) {
    7811           5 :                         return;
    7812             :                 }
    7813             :         }
    7814             : 
    7815         552 :         bdev_io_complete(bdev_io);
    7816         598 : }
    7817             : 
    7818             : void
    7819           0 : spdk_bdev_io_complete_scsi_status(struct spdk_bdev_io *bdev_io, enum spdk_scsi_status sc,
    7820             :                                   enum spdk_scsi_sense sk, uint8_t asc, uint8_t ascq)
    7821             : {
    7822             :         enum spdk_bdev_io_status status;
    7823             : 
    7824           0 :         if (sc == SPDK_SCSI_STATUS_GOOD) {
    7825           0 :                 status = SPDK_BDEV_IO_STATUS_SUCCESS;
    7826           0 :         } else {
    7827           0 :                 status = SPDK_BDEV_IO_STATUS_SCSI_ERROR;
    7828           0 :                 bdev_io->internal.error.scsi.sc = sc;
    7829           0 :                 bdev_io->internal.error.scsi.sk = sk;
    7830           0 :                 bdev_io->internal.error.scsi.asc = asc;
    7831           0 :                 bdev_io->internal.error.scsi.ascq = ascq;
    7832             :         }
    7833             : 
    7834           0 :         spdk_bdev_io_complete(bdev_io, status);
    7835           0 : }
    7836             : 
    7837             : void
    7838           0 : spdk_bdev_io_get_scsi_status(const struct spdk_bdev_io *bdev_io,
    7839             :                              int *sc, int *sk, int *asc, int *ascq)
    7840             : {
    7841           0 :         assert(sc != NULL);
    7842           0 :         assert(sk != NULL);
    7843           0 :         assert(asc != NULL);
    7844           0 :         assert(ascq != NULL);
    7845             : 
    7846           0 :         switch (bdev_io->internal.status) {
    7847             :         case SPDK_BDEV_IO_STATUS_SUCCESS:
    7848           0 :                 *sc = SPDK_SCSI_STATUS_GOOD;
    7849           0 :                 *sk = SPDK_SCSI_SENSE_NO_SENSE;
    7850           0 :                 *asc = SPDK_SCSI_ASC_NO_ADDITIONAL_SENSE;
    7851           0 :                 *ascq = SPDK_SCSI_ASCQ_CAUSE_NOT_REPORTABLE;
    7852           0 :                 break;
    7853             :         case SPDK_BDEV_IO_STATUS_NVME_ERROR:
    7854           0 :                 spdk_scsi_nvme_translate(bdev_io, sc, sk, asc, ascq);
    7855           0 :                 break;
    7856             :         case SPDK_BDEV_IO_STATUS_MISCOMPARE:
    7857           0 :                 *sc = SPDK_SCSI_STATUS_CHECK_CONDITION;
    7858           0 :                 *sk = SPDK_SCSI_SENSE_MISCOMPARE;
    7859           0 :                 *asc = SPDK_SCSI_ASC_MISCOMPARE_DURING_VERIFY_OPERATION;
    7860           0 :                 *ascq = bdev_io->internal.error.scsi.ascq;
    7861           0 :                 break;
    7862             :         case SPDK_BDEV_IO_STATUS_SCSI_ERROR:
    7863           0 :                 *sc = bdev_io->internal.error.scsi.sc;
    7864           0 :                 *sk = bdev_io->internal.error.scsi.sk;
    7865           0 :                 *asc = bdev_io->internal.error.scsi.asc;
    7866           0 :                 *ascq = bdev_io->internal.error.scsi.ascq;
    7867           0 :                 break;
    7868             :         default:
    7869           0 :                 *sc = SPDK_SCSI_STATUS_CHECK_CONDITION;
    7870           0 :                 *sk = SPDK_SCSI_SENSE_ABORTED_COMMAND;
    7871           0 :                 *asc = SPDK_SCSI_ASC_NO_ADDITIONAL_SENSE;
    7872           0 :                 *ascq = SPDK_SCSI_ASCQ_CAUSE_NOT_REPORTABLE;
    7873           0 :                 break;
    7874             :         }
    7875           0 : }
    7876             : 
    7877             : void
    7878           0 : spdk_bdev_io_complete_aio_status(struct spdk_bdev_io *bdev_io, int aio_result)
    7879             : {
    7880             :         enum spdk_bdev_io_status status;
    7881             : 
    7882           0 :         if (aio_result == 0) {
    7883           0 :                 status = SPDK_BDEV_IO_STATUS_SUCCESS;
    7884           0 :         } else {
    7885           0 :                 status = SPDK_BDEV_IO_STATUS_AIO_ERROR;
    7886             :         }
    7887             : 
    7888           0 :         bdev_io->internal.error.aio_result = aio_result;
    7889             : 
    7890           0 :         spdk_bdev_io_complete(bdev_io, status);
    7891           0 : }
    7892             : 
    7893             : void
    7894           0 : spdk_bdev_io_get_aio_status(const struct spdk_bdev_io *bdev_io, int *aio_result)
    7895             : {
    7896           0 :         assert(aio_result != NULL);
    7897             : 
    7898           0 :         if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_AIO_ERROR) {
    7899           0 :                 *aio_result = bdev_io->internal.error.aio_result;
    7900           0 :         } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS) {
    7901           0 :                 *aio_result = 0;
    7902           0 :         } else {
    7903           0 :                 *aio_result = -EIO;
    7904             :         }
    7905           0 : }
    7906             : 
    7907             : void
    7908           0 : spdk_bdev_io_complete_nvme_status(struct spdk_bdev_io *bdev_io, uint32_t cdw0, int sct, int sc)
    7909             : {
    7910             :         enum spdk_bdev_io_status status;
    7911             : 
    7912           0 :         if (spdk_likely(sct == SPDK_NVME_SCT_GENERIC && sc == SPDK_NVME_SC_SUCCESS)) {
    7913           0 :                 status = SPDK_BDEV_IO_STATUS_SUCCESS;
    7914           0 :         } else if (sct == SPDK_NVME_SCT_GENERIC && sc == SPDK_NVME_SC_ABORTED_BY_REQUEST) {
    7915           0 :                 status = SPDK_BDEV_IO_STATUS_ABORTED;
    7916           0 :         } else {
    7917           0 :                 status = SPDK_BDEV_IO_STATUS_NVME_ERROR;
    7918             :         }
    7919             : 
    7920           0 :         bdev_io->internal.error.nvme.cdw0 = cdw0;
    7921           0 :         bdev_io->internal.error.nvme.sct = sct;
    7922           0 :         bdev_io->internal.error.nvme.sc = sc;
    7923             : 
    7924           0 :         spdk_bdev_io_complete(bdev_io, status);
    7925           0 : }
    7926             : 
    7927             : void
    7928           0 : spdk_bdev_io_get_nvme_status(const struct spdk_bdev_io *bdev_io, uint32_t *cdw0, int *sct, int *sc)
    7929             : {
    7930           0 :         assert(sct != NULL);
    7931           0 :         assert(sc != NULL);
    7932           0 :         assert(cdw0 != NULL);
    7933             : 
    7934           0 :         if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_ABORT)) {
    7935           0 :                 *sct = SPDK_NVME_SCT_GENERIC;
    7936           0 :                 *sc = SPDK_NVME_SC_SUCCESS;
    7937           0 :                 if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS) {
    7938           0 :                         *cdw0 = 0;
    7939           0 :                 } else {
    7940           0 :                         *cdw0 = 1U;
    7941             :                 }
    7942           0 :                 return;
    7943             :         }
    7944             : 
    7945           0 :         if (spdk_likely(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS)) {
    7946           0 :                 *sct = SPDK_NVME_SCT_GENERIC;
    7947           0 :                 *sc = SPDK_NVME_SC_SUCCESS;
    7948           0 :         } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_NVME_ERROR) {
    7949           0 :                 *sct = bdev_io->internal.error.nvme.sct;
    7950           0 :                 *sc = bdev_io->internal.error.nvme.sc;
    7951           0 :         } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_ABORTED) {
    7952           0 :                 *sct = SPDK_NVME_SCT_GENERIC;
    7953           0 :                 *sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
    7954           0 :         } else {
    7955           0 :                 *sct = SPDK_NVME_SCT_GENERIC;
    7956           0 :                 *sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
    7957             :         }
    7958             : 
    7959           0 :         *cdw0 = bdev_io->internal.error.nvme.cdw0;
    7960           0 : }
    7961             : 
    7962             : void
    7963           0 : spdk_bdev_io_get_nvme_fused_status(const struct spdk_bdev_io *bdev_io, uint32_t *cdw0,
    7964             :                                    int *first_sct, int *first_sc, int *second_sct, int *second_sc)
    7965             : {
    7966           0 :         assert(first_sct != NULL);
    7967           0 :         assert(first_sc != NULL);
    7968           0 :         assert(second_sct != NULL);
    7969           0 :         assert(second_sc != NULL);
    7970           0 :         assert(cdw0 != NULL);
    7971             : 
    7972           0 :         if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_NVME_ERROR) {
    7973           0 :                 if (bdev_io->internal.error.nvme.sct == SPDK_NVME_SCT_MEDIA_ERROR &&
    7974           0 :                     bdev_io->internal.error.nvme.sc == SPDK_NVME_SC_COMPARE_FAILURE) {
    7975           0 :                         *first_sct = bdev_io->internal.error.nvme.sct;
    7976           0 :                         *first_sc = bdev_io->internal.error.nvme.sc;
    7977           0 :                         *second_sct = SPDK_NVME_SCT_GENERIC;
    7978           0 :                         *second_sc = SPDK_NVME_SC_ABORTED_FAILED_FUSED;
    7979           0 :                 } else {
    7980           0 :                         *first_sct = SPDK_NVME_SCT_GENERIC;
    7981           0 :                         *first_sc = SPDK_NVME_SC_SUCCESS;
    7982           0 :                         *second_sct = bdev_io->internal.error.nvme.sct;
    7983           0 :                         *second_sc = bdev_io->internal.error.nvme.sc;
    7984             :                 }
    7985           0 :         } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_ABORTED) {
    7986           0 :                 *first_sct = SPDK_NVME_SCT_GENERIC;
    7987           0 :                 *first_sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
    7988           0 :                 *second_sct = SPDK_NVME_SCT_GENERIC;
    7989           0 :                 *second_sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
    7990           0 :         } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS) {
    7991           0 :                 *first_sct = SPDK_NVME_SCT_GENERIC;
    7992           0 :                 *first_sc = SPDK_NVME_SC_SUCCESS;
    7993           0 :                 *second_sct = SPDK_NVME_SCT_GENERIC;
    7994           0 :                 *second_sc = SPDK_NVME_SC_SUCCESS;
    7995           0 :         } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_FIRST_FUSED_FAILED) {
    7996           0 :                 *first_sct = SPDK_NVME_SCT_GENERIC;
    7997           0 :                 *first_sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
    7998           0 :                 *second_sct = SPDK_NVME_SCT_GENERIC;
    7999           0 :                 *second_sc = SPDK_NVME_SC_ABORTED_FAILED_FUSED;
    8000           0 :         } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_MISCOMPARE) {
    8001           0 :                 *first_sct = SPDK_NVME_SCT_MEDIA_ERROR;
    8002           0 :                 *first_sc = SPDK_NVME_SC_COMPARE_FAILURE;
    8003           0 :                 *second_sct = SPDK_NVME_SCT_GENERIC;
    8004           0 :                 *second_sc = SPDK_NVME_SC_ABORTED_FAILED_FUSED;
    8005           0 :         } else {
    8006           0 :                 *first_sct = SPDK_NVME_SCT_GENERIC;
    8007           0 :                 *first_sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
    8008           0 :                 *second_sct = SPDK_NVME_SCT_GENERIC;
    8009           0 :                 *second_sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
    8010             :         }
    8011             : 
    8012           0 :         *cdw0 = bdev_io->internal.error.nvme.cdw0;
    8013           0 : }
    8014             : 
    8015             : void
    8016           0 : spdk_bdev_io_complete_base_io_status(struct spdk_bdev_io *bdev_io,
    8017             :                                      const struct spdk_bdev_io *base_io)
    8018             : {
    8019           0 :         switch (base_io->internal.status) {
    8020             :         case SPDK_BDEV_IO_STATUS_NVME_ERROR:
    8021           0 :                 spdk_bdev_io_complete_nvme_status(bdev_io,
    8022           0 :                                                   base_io->internal.error.nvme.cdw0,
    8023           0 :                                                   base_io->internal.error.nvme.sct,
    8024           0 :                                                   base_io->internal.error.nvme.sc);
    8025           0 :                 break;
    8026             :         case SPDK_BDEV_IO_STATUS_SCSI_ERROR:
    8027           0 :                 spdk_bdev_io_complete_scsi_status(bdev_io,
    8028           0 :                                                   base_io->internal.error.scsi.sc,
    8029           0 :                                                   base_io->internal.error.scsi.sk,
    8030           0 :                                                   base_io->internal.error.scsi.asc,
    8031           0 :                                                   base_io->internal.error.scsi.ascq);
    8032           0 :                 break;
    8033             :         case SPDK_BDEV_IO_STATUS_AIO_ERROR:
    8034           0 :                 spdk_bdev_io_complete_aio_status(bdev_io, base_io->internal.error.aio_result);
    8035           0 :                 break;
    8036             :         default:
    8037           0 :                 spdk_bdev_io_complete(bdev_io, base_io->internal.status);
    8038           0 :                 break;
    8039             :         }
    8040           0 : }
    8041             : 
    8042             : struct spdk_thread *
    8043         660 : spdk_bdev_io_get_thread(struct spdk_bdev_io *bdev_io)
    8044             : {
    8045         660 :         return spdk_io_channel_get_thread(bdev_io->internal.ch->channel);
    8046             : }
    8047             : 
    8048             : struct spdk_io_channel *
    8049          70 : spdk_bdev_io_get_io_channel(struct spdk_bdev_io *bdev_io)
    8050             : {
    8051          70 :         return bdev_io->internal.ch->channel;
    8052             : }
    8053             : 
    8054             : static int
    8055         128 : bdev_register(struct spdk_bdev *bdev)
    8056             : {
    8057             :         char *bdev_name;
    8058             :         char uuid[SPDK_UUID_STRING_LEN];
    8059             :         struct spdk_iobuf_opts iobuf_opts;
    8060             :         int ret;
    8061             : 
    8062         128 :         assert(bdev->module != NULL);
    8063             : 
    8064         128 :         if (!bdev->name) {
    8065           0 :                 SPDK_ERRLOG("Bdev name is NULL\n");
    8066           0 :                 return -EINVAL;
    8067             :         }
    8068             : 
    8069         128 :         if (!strlen(bdev->name)) {
    8070           0 :                 SPDK_ERRLOG("Bdev name must not be an empty string\n");
    8071           0 :                 return -EINVAL;
    8072             :         }
    8073             : 
    8074             :         /* Users often register their own I/O devices using the bdev name. In
    8075             :          * order to avoid conflicts, prepend bdev_. */
    8076         128 :         bdev_name = spdk_sprintf_alloc("bdev_%s", bdev->name);
    8077         128 :         if (!bdev_name) {
    8078           0 :                 SPDK_ERRLOG("Unable to allocate memory for internal bdev name.\n");
    8079           0 :                 return -ENOMEM;
    8080             :         }
    8081             : 
    8082         128 :         bdev->internal.stat = bdev_alloc_io_stat(true);
    8083         128 :         if (!bdev->internal.stat) {
    8084           0 :                 SPDK_ERRLOG("Unable to allocate I/O statistics structure.\n");
    8085           0 :                 free(bdev_name);
    8086           0 :                 return -ENOMEM;
    8087             :         }
    8088             : 
    8089         128 :         bdev->internal.status = SPDK_BDEV_STATUS_READY;
    8090         128 :         bdev->internal.measured_queue_depth = UINT64_MAX;
    8091         128 :         bdev->internal.claim_type = SPDK_BDEV_CLAIM_NONE;
    8092         128 :         memset(&bdev->internal.claim, 0, sizeof(bdev->internal.claim));
    8093         128 :         bdev->internal.qd_poller = NULL;
    8094         128 :         bdev->internal.qos = NULL;
    8095             : 
    8096         128 :         TAILQ_INIT(&bdev->internal.open_descs);
    8097         128 :         TAILQ_INIT(&bdev->internal.locked_ranges);
    8098         128 :         TAILQ_INIT(&bdev->internal.pending_locked_ranges);
    8099         128 :         TAILQ_INIT(&bdev->aliases);
    8100             : 
    8101             :         /* UUID may be specified by the user or defined by bdev itself.
    8102             :          * Otherwise it will be generated here, so this field will never be empty. */
    8103         128 :         if (spdk_uuid_is_null(&bdev->uuid)) {
    8104          42 :                 spdk_uuid_generate(&bdev->uuid);
    8105          42 :         }
    8106             : 
    8107             :         /* Add the UUID alias only if it's different than the name */
    8108         128 :         spdk_uuid_fmt_lower(uuid, sizeof(uuid), &bdev->uuid);
    8109         128 :         if (strcmp(bdev->name, uuid) != 0) {
    8110         127 :                 ret = spdk_bdev_alias_add(bdev, uuid);
    8111         127 :                 if (ret != 0) {
    8112           2 :                         SPDK_ERRLOG("Unable to add uuid:%s alias for bdev %s\n", uuid, bdev->name);
    8113           2 :                         bdev_free_io_stat(bdev->internal.stat);
    8114           2 :                         free(bdev_name);
    8115           2 :                         return ret;
    8116             :                 }
    8117         125 :         }
    8118             : 
    8119         126 :         spdk_iobuf_get_opts(&iobuf_opts, sizeof(iobuf_opts));
    8120         126 :         if (spdk_bdev_get_buf_align(bdev) > 1) {
    8121           0 :                 bdev->max_rw_size = spdk_min(bdev->max_rw_size ? bdev->max_rw_size : UINT32_MAX,
    8122             :                                              iobuf_opts.large_bufsize / bdev->blocklen);
    8123           0 :         }
    8124             : 
    8125             :         /* If the user didn't specify a write unit size, set it to one. */
    8126         126 :         if (bdev->write_unit_size == 0) {
    8127         122 :                 bdev->write_unit_size = 1;
    8128         122 :         }
    8129             : 
    8130             :         /* Set ACWU value to the write unit size if bdev module did not set it (does not support it natively) */
    8131         126 :         if (bdev->acwu == 0) {
    8132         122 :                 bdev->acwu = bdev->write_unit_size;
    8133         122 :         }
    8134             : 
    8135         126 :         if (bdev->phys_blocklen == 0) {
    8136         122 :                 bdev->phys_blocklen = spdk_bdev_get_data_block_size(bdev);
    8137         122 :         }
    8138             : 
    8139         126 :         if (!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COPY)) {
    8140           0 :                 bdev->max_copy = bdev_get_max_write(bdev, iobuf_opts.large_bufsize);
    8141           0 :         }
    8142             : 
    8143         126 :         if (!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE_ZEROES)) {
    8144           0 :                 bdev->max_write_zeroes = bdev_get_max_write(bdev, ZERO_BUFFER_SIZE);
    8145           0 :         }
    8146             : 
    8147         126 :         bdev->internal.reset_in_progress = NULL;
    8148         126 :         bdev->internal.qd_poll_in_progress = false;
    8149         126 :         bdev->internal.period = 0;
    8150         126 :         bdev->internal.new_period = 0;
    8151         126 :         bdev->internal.trace_id = spdk_trace_register_owner(OWNER_TYPE_BDEV, bdev_name);
    8152             : 
    8153             :         /*
    8154             :          * Initialize spinlock before registering IO device because spinlock is used in
    8155             :          * bdev_channel_create
    8156             :          */
    8157         126 :         spdk_spin_init(&bdev->internal.spinlock);
    8158             : 
    8159         252 :         spdk_io_device_register(__bdev_to_io_dev(bdev),
    8160             :                                 bdev_channel_create, bdev_channel_destroy,
    8161             :                                 sizeof(struct spdk_bdev_channel),
    8162         126 :                                 bdev_name);
    8163             : 
    8164             :         /*
    8165             :          * Register bdev name only after the bdev object is ready.
    8166             :          * After bdev_name_add returns, it is possible for other threads to start using the bdev,
    8167             :          * create IO channels...
    8168             :          */
    8169         126 :         ret = bdev_name_add(&bdev->internal.bdev_name, bdev, bdev->name);
    8170         126 :         if (ret != 0) {
    8171           0 :                 spdk_io_device_unregister(__bdev_to_io_dev(bdev), NULL);
    8172           0 :                 bdev_free_io_stat(bdev->internal.stat);
    8173           0 :                 spdk_spin_destroy(&bdev->internal.spinlock);
    8174           0 :                 free(bdev_name);
    8175           0 :                 return ret;
    8176             :         }
    8177             : 
    8178         126 :         free(bdev_name);
    8179             : 
    8180         126 :         SPDK_DEBUGLOG(bdev, "Inserting bdev %s into list\n", bdev->name);
    8181         126 :         TAILQ_INSERT_TAIL(&g_bdev_mgr.bdevs, bdev, internal.link);
    8182             : 
    8183         126 :         return 0;
    8184         128 : }
    8185             : 
    8186             : static void
    8187         127 : bdev_destroy_cb(void *io_device)
    8188             : {
    8189             :         int                     rc;
    8190             :         struct spdk_bdev        *bdev;
    8191             :         spdk_bdev_unregister_cb cb_fn;
    8192             :         void                    *cb_arg;
    8193             : 
    8194         127 :         bdev = __bdev_from_io_dev(io_device);
    8195             : 
    8196         127 :         if (bdev->internal.unregister_td != spdk_get_thread()) {
    8197           1 :                 spdk_thread_send_msg(bdev->internal.unregister_td, bdev_destroy_cb, io_device);
    8198           1 :                 return;
    8199             :         }
    8200             : 
    8201         126 :         cb_fn = bdev->internal.unregister_cb;
    8202         126 :         cb_arg = bdev->internal.unregister_ctx;
    8203             : 
    8204         126 :         spdk_spin_destroy(&bdev->internal.spinlock);
    8205         126 :         free(bdev->internal.qos);
    8206         126 :         bdev_free_io_stat(bdev->internal.stat);
    8207         126 :         spdk_trace_unregister_owner(bdev->internal.trace_id);
    8208             : 
    8209         126 :         rc = bdev->fn_table->destruct(bdev->ctxt);
    8210         126 :         if (rc < 0) {
    8211           0 :                 SPDK_ERRLOG("destruct failed\n");
    8212           0 :         }
    8213         126 :         if (rc <= 0 && cb_fn != NULL) {
    8214          10 :                 cb_fn(cb_arg, rc);
    8215          10 :         }
    8216         127 : }
    8217             : 
    8218             : void
    8219           2 : spdk_bdev_destruct_done(struct spdk_bdev *bdev, int bdeverrno)
    8220             : {
    8221           2 :         if (bdev->internal.unregister_cb != NULL) {
    8222           0 :                 bdev->internal.unregister_cb(bdev->internal.unregister_ctx, bdeverrno);
    8223           0 :         }
    8224           2 : }
    8225             : 
    8226             : static void
    8227          19 : _remove_notify(void *arg)
    8228             : {
    8229          19 :         struct spdk_bdev_desc *desc = arg;
    8230             : 
    8231          19 :         _event_notify(desc, SPDK_BDEV_EVENT_REMOVE);
    8232          19 : }
    8233             : 
    8234             : /* returns: 0 - bdev removed and ready to be destructed.
    8235             :  *          -EBUSY - bdev can't be destructed yet.  */
    8236             : static int
    8237         141 : bdev_unregister_unsafe(struct spdk_bdev *bdev)
    8238             : {
    8239             :         struct spdk_bdev_desc   *desc, *tmp;
    8240         141 :         int                     rc = 0;
    8241             :         char                    uuid[SPDK_UUID_STRING_LEN];
    8242             : 
    8243         141 :         assert(spdk_spin_held(&g_bdev_mgr.spinlock));
    8244         141 :         assert(spdk_spin_held(&bdev->internal.spinlock));
    8245             : 
    8246             :         /* Notify each descriptor about hotremoval */
    8247         160 :         TAILQ_FOREACH_SAFE(desc, &bdev->internal.open_descs, link, tmp) {
    8248          19 :                 rc = -EBUSY;
    8249             :                 /*
    8250             :                  * Defer invocation of the event_cb to a separate message that will
    8251             :                  *  run later on its thread.  This ensures this context unwinds and
    8252             :                  *  we don't recursively unregister this bdev again if the event_cb
    8253             :                  *  immediately closes its descriptor.
    8254             :                  */
    8255          19 :                 event_notify(desc, _remove_notify);
    8256          19 :         }
    8257             : 
    8258             :         /* If there are no descriptors, proceed removing the bdev */
    8259         141 :         if (rc == 0) {
    8260         126 :                 TAILQ_REMOVE(&g_bdev_mgr.bdevs, bdev, internal.link);
    8261         126 :                 SPDK_DEBUGLOG(bdev, "Removing bdev %s from list done\n", bdev->name);
    8262             : 
    8263             :                 /* Delete the name and the UUID alias */
    8264         126 :                 spdk_uuid_fmt_lower(uuid, sizeof(uuid), &bdev->uuid);
    8265         126 :                 bdev_name_del_unsafe(&bdev->internal.bdev_name);
    8266         126 :                 bdev_alias_del(bdev, uuid, bdev_name_del_unsafe);
    8267             : 
    8268         126 :                 spdk_notify_send("bdev_unregister", spdk_bdev_get_name(bdev));
    8269             : 
    8270         126 :                 if (bdev->internal.reset_in_progress != NULL) {
    8271             :                         /* If reset is in progress, let the completion callback for reset
    8272             :                          * unregister the bdev.
    8273             :                          */
    8274           1 :                         rc = -EBUSY;
    8275           1 :                 }
    8276         126 :         }
    8277             : 
    8278         141 :         return rc;
    8279             : }
    8280             : 
    8281             : static void
    8282           4 : bdev_unregister_abort_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
    8283             :                               struct spdk_io_channel *io_ch, void *_ctx)
    8284             : {
    8285           4 :         struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
    8286             : 
    8287           4 :         bdev_channel_abort_queued_ios(bdev_ch);
    8288           4 :         spdk_bdev_for_each_channel_continue(i, 0);
    8289           4 : }
    8290             : 
    8291             : static void
    8292         126 : bdev_unregister(struct spdk_bdev *bdev, void *_ctx, int status)
    8293             : {
    8294             :         int rc;
    8295             : 
    8296         126 :         spdk_spin_lock(&g_bdev_mgr.spinlock);
    8297         126 :         spdk_spin_lock(&bdev->internal.spinlock);
    8298             :         /*
    8299             :          * Set the status to REMOVING after completing to abort channels. Otherwise,
    8300             :          * the last spdk_bdev_close() may call spdk_io_device_unregister() while
    8301             :          * spdk_bdev_for_each_channel() is executed and spdk_io_device_unregister()
    8302             :          * may fail.
    8303             :          */
    8304         126 :         bdev->internal.status = SPDK_BDEV_STATUS_REMOVING;
    8305         126 :         rc = bdev_unregister_unsafe(bdev);
    8306         126 :         spdk_spin_unlock(&bdev->internal.spinlock);
    8307         126 :         spdk_spin_unlock(&g_bdev_mgr.spinlock);
    8308             : 
    8309         126 :         if (rc == 0) {
    8310         110 :                 spdk_io_device_unregister(__bdev_to_io_dev(bdev), bdev_destroy_cb);
    8311         110 :         }
    8312         126 : }
    8313             : 
    8314             : void
    8315         133 : spdk_bdev_unregister(struct spdk_bdev *bdev, spdk_bdev_unregister_cb cb_fn, void *cb_arg)
    8316             : {
    8317             :         struct spdk_thread      *thread;
    8318             : 
    8319         133 :         SPDK_DEBUGLOG(bdev, "Removing bdev %s from list\n", bdev->name);
    8320             : 
    8321         133 :         thread = spdk_get_thread();
    8322         133 :         if (!thread) {
    8323             :                 /* The user called this from a non-SPDK thread. */
    8324           0 :                 if (cb_fn != NULL) {
    8325           0 :                         cb_fn(cb_arg, -ENOTSUP);
    8326           0 :                 }
    8327           0 :                 return;
    8328             :         }
    8329             : 
    8330         133 :         spdk_spin_lock(&g_bdev_mgr.spinlock);
    8331         133 :         if (bdev->internal.status == SPDK_BDEV_STATUS_UNREGISTERING ||
    8332         133 :             bdev->internal.status == SPDK_BDEV_STATUS_REMOVING) {
    8333           7 :                 spdk_spin_unlock(&g_bdev_mgr.spinlock);
    8334           7 :                 if (cb_fn) {
    8335           0 :                         cb_fn(cb_arg, -EBUSY);
    8336           0 :                 }
    8337           7 :                 return;
    8338             :         }
    8339             : 
    8340         126 :         spdk_spin_lock(&bdev->internal.spinlock);
    8341         126 :         bdev->internal.status = SPDK_BDEV_STATUS_UNREGISTERING;
    8342         126 :         bdev->internal.unregister_cb = cb_fn;
    8343         126 :         bdev->internal.unregister_ctx = cb_arg;
    8344         126 :         bdev->internal.unregister_td = thread;
    8345         126 :         spdk_spin_unlock(&bdev->internal.spinlock);
    8346         126 :         spdk_spin_unlock(&g_bdev_mgr.spinlock);
    8347             : 
    8348         126 :         spdk_bdev_set_qd_sampling_period(bdev, 0);
    8349             : 
    8350         126 :         spdk_bdev_for_each_channel(bdev, bdev_unregister_abort_channel, bdev,
    8351             :                                    bdev_unregister);
    8352         133 : }
    8353             : 
    8354             : int
    8355           4 : spdk_bdev_unregister_by_name(const char *bdev_name, struct spdk_bdev_module *module,
    8356             :                              spdk_bdev_unregister_cb cb_fn, void *cb_arg)
    8357             : {
    8358             :         struct spdk_bdev_desc *desc;
    8359             :         struct spdk_bdev *bdev;
    8360             :         int rc;
    8361             : 
    8362           4 :         rc = spdk_bdev_open_ext(bdev_name, false, _tmp_bdev_event_cb, NULL, &desc);
    8363           4 :         if (rc != 0) {
    8364           1 :                 SPDK_ERRLOG("Failed to open bdev with name: %s\n", bdev_name);
    8365           1 :                 return rc;
    8366             :         }
    8367             : 
    8368           3 :         bdev = spdk_bdev_desc_get_bdev(desc);
    8369             : 
    8370           3 :         if (bdev->module != module) {
    8371           1 :                 spdk_bdev_close(desc);
    8372           1 :                 SPDK_ERRLOG("Bdev %s was not registered by the specified module.\n",
    8373             :                             bdev_name);
    8374           1 :                 return -ENODEV;
    8375             :         }
    8376             : 
    8377           2 :         spdk_bdev_unregister(bdev, cb_fn, cb_arg);
    8378             : 
    8379           2 :         spdk_bdev_close(desc);
    8380             : 
    8381           2 :         return 0;
    8382           4 : }
    8383             : 
    8384             : static int
    8385         259 : bdev_start_qos(struct spdk_bdev *bdev)
    8386             : {
    8387             :         struct set_qos_limit_ctx *ctx;
    8388             : 
    8389             :         /* Enable QoS */
    8390         259 :         if (bdev->internal.qos && bdev->internal.qos->thread == NULL) {
    8391           2 :                 ctx = calloc(1, sizeof(*ctx));
    8392           2 :                 if (ctx == NULL) {
    8393           0 :                         SPDK_ERRLOG("Failed to allocate memory for QoS context\n");
    8394           0 :                         return -ENOMEM;
    8395             :                 }
    8396           2 :                 ctx->bdev = bdev;
    8397           2 :                 spdk_bdev_for_each_channel(bdev, bdev_enable_qos_msg, ctx, bdev_enable_qos_done);
    8398           2 :         }
    8399             : 
    8400         259 :         return 0;
    8401         259 : }
    8402             : 
    8403             : static void
    8404          24 : log_already_claimed(enum spdk_log_level level, const int line, const char *func, const char *detail,
    8405             :                     struct spdk_bdev *bdev)
    8406             : {
    8407             :         enum spdk_bdev_claim_type type;
    8408             :         const char *typename, *modname;
    8409             :         extern struct spdk_log_flag SPDK_LOG_bdev;
    8410             : 
    8411          24 :         assert(spdk_spin_held(&bdev->internal.spinlock));
    8412             : 
    8413          24 :         if (level >= SPDK_LOG_INFO && !SPDK_LOG_bdev.enabled) {
    8414           0 :                 return;
    8415             :         }
    8416             : 
    8417          24 :         type = bdev->internal.claim_type;
    8418          24 :         typename = spdk_bdev_claim_get_name(type);
    8419             : 
    8420          24 :         if (type == SPDK_BDEV_CLAIM_EXCL_WRITE) {
    8421           6 :                 modname = bdev->internal.claim.v1.module->name;
    8422          12 :                 spdk_log(level, __FILE__, line, func, "bdev %s %s: type %s by module %s\n",
    8423           6 :                          bdev->name, detail, typename, modname);
    8424           6 :                 return;
    8425             :         }
    8426             : 
    8427          18 :         if (claim_type_is_v2(type)) {
    8428             :                 struct spdk_bdev_module_claim *claim;
    8429             : 
    8430          36 :                 TAILQ_FOREACH(claim, &bdev->internal.claim.v2.claims, link) {
    8431          18 :                         modname = claim->module->name;
    8432          36 :                         spdk_log(level, __FILE__, line, func, "bdev %s %s: type %s by module %s\n",
    8433          18 :                                  bdev->name, detail, typename, modname);
    8434          18 :                 }
    8435          18 :                 return;
    8436             :         }
    8437             : 
    8438           0 :         assert(false);
    8439          24 : }
    8440             : 
    8441             : static int
    8442         268 : bdev_open(struct spdk_bdev *bdev, bool write, struct spdk_bdev_desc *desc)
    8443             : {
    8444             :         struct spdk_thread *thread;
    8445         268 :         int rc = 0;
    8446             : 
    8447         268 :         thread = spdk_get_thread();
    8448         268 :         if (!thread) {
    8449           0 :                 SPDK_ERRLOG("Cannot open bdev from non-SPDK thread.\n");
    8450           0 :                 return -ENOTSUP;
    8451             :         }
    8452             : 
    8453         268 :         SPDK_DEBUGLOG(bdev, "Opening descriptor %p for bdev %s on thread %p\n", desc, bdev->name,
    8454             :                       spdk_get_thread());
    8455             : 
    8456         268 :         desc->bdev = bdev;
    8457         268 :         desc->thread = thread;
    8458         268 :         desc->write = write;
    8459             : 
    8460         268 :         spdk_spin_lock(&bdev->internal.spinlock);
    8461         268 :         if (bdev->internal.status == SPDK_BDEV_STATUS_UNREGISTERING ||
    8462         268 :             bdev->internal.status == SPDK_BDEV_STATUS_REMOVING) {
    8463           3 :                 spdk_spin_unlock(&bdev->internal.spinlock);
    8464           3 :                 return -ENODEV;
    8465             :         }
    8466             : 
    8467         265 :         if (write && bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE) {
    8468           6 :                 LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
    8469           6 :                 spdk_spin_unlock(&bdev->internal.spinlock);
    8470           6 :                 return -EPERM;
    8471             :         }
    8472             : 
    8473         259 :         rc = bdev_start_qos(bdev);
    8474         259 :         if (rc != 0) {
    8475           0 :                 SPDK_ERRLOG("Failed to start QoS on bdev %s\n", bdev->name);
    8476           0 :                 spdk_spin_unlock(&bdev->internal.spinlock);
    8477           0 :                 return rc;
    8478             :         }
    8479             : 
    8480         259 :         TAILQ_INSERT_TAIL(&bdev->internal.open_descs, desc, link);
    8481             : 
    8482         259 :         spdk_spin_unlock(&bdev->internal.spinlock);
    8483             : 
    8484         259 :         return 0;
    8485         268 : }
    8486             : 
    8487             : static void
    8488         268 : bdev_open_opts_get_defaults(struct spdk_bdev_open_opts *opts, size_t opts_size)
    8489             : {
    8490         268 :         if (!opts) {
    8491           0 :                 SPDK_ERRLOG("opts should not be NULL.\n");
    8492           0 :                 return;
    8493             :         }
    8494             : 
    8495         268 :         if (!opts_size) {
    8496           0 :                 SPDK_ERRLOG("opts_size should not be zero.\n");
    8497           0 :                 return;
    8498             :         }
    8499             : 
    8500         268 :         memset(opts, 0, opts_size);
    8501         268 :         opts->size = opts_size;
    8502             : 
    8503             : #define FIELD_OK(field) \
    8504             :         offsetof(struct spdk_bdev_open_opts, field) + sizeof(opts->field) <= opts_size
    8505             : 
    8506             : #define SET_FIELD(field, value) \
    8507             :         if (FIELD_OK(field)) { \
    8508             :                 opts->field = value; \
    8509             :         } \
    8510             : 
    8511         268 :         SET_FIELD(no_metadata, false);
    8512             : 
    8513             : #undef FIELD_OK
    8514             : #undef SET_FIELD
    8515         268 : }
    8516             : 
    8517             : static void
    8518           1 : bdev_open_opts_copy(struct spdk_bdev_open_opts *opts,
    8519             :                     const struct spdk_bdev_open_opts *user_opts)
    8520             : {
    8521             : #define FIELD_OK(field) \
    8522             :         offsetof(struct spdk_bdev_open_opts, field) + sizeof(opts->field) <= user_opts->size
    8523             : 
    8524             : #define SET_FIELD(field) \
    8525             :         if (FIELD_OK(field)) { \
    8526             :                 opts->field = user_opts->field; \
    8527             :         } \
    8528             : 
    8529           1 :         SET_FIELD(no_metadata);
    8530             : 
    8531           1 :         opts->size = user_opts->size;
    8532             : 
    8533             :         /* We should not remove this statement, but need to update the assert statement
    8534             :          * if we add a new field, and also add a corresponding SET_FIELD statement.
    8535             :          */
    8536             :         SPDK_STATIC_ASSERT(sizeof(struct spdk_bdev_open_opts) == 16, "Incorrect size");
    8537             : 
    8538             : #undef FIELD_OK
    8539             : #undef SET_FIELD
    8540           1 : }
    8541             : 
    8542             : static int
    8543         268 : bdev_desc_alloc(struct spdk_bdev *bdev, spdk_bdev_event_cb_t event_cb, void *event_ctx,
    8544             :                 struct spdk_bdev_open_opts *user_opts, struct spdk_bdev_desc **_desc)
    8545             : {
    8546             :         struct spdk_bdev_desc *desc;
    8547             :         struct spdk_bdev_open_opts opts;
    8548             :         unsigned int i;
    8549             : 
    8550         268 :         bdev_open_opts_get_defaults(&opts, sizeof(opts));
    8551         268 :         if (user_opts != NULL) {
    8552           1 :                 bdev_open_opts_copy(&opts, user_opts);
    8553           1 :         }
    8554             : 
    8555         268 :         desc = calloc(1, sizeof(*desc));
    8556         268 :         if (desc == NULL) {
    8557           0 :                 SPDK_ERRLOG("Failed to allocate memory for bdev descriptor\n");
    8558           0 :                 return -ENOMEM;
    8559             :         }
    8560             : 
    8561         268 :         desc->opts = opts;
    8562             : 
    8563         268 :         TAILQ_INIT(&desc->pending_media_events);
    8564         268 :         TAILQ_INIT(&desc->free_media_events);
    8565             : 
    8566         268 :         desc->memory_domains_supported = spdk_bdev_get_memory_domains(bdev, NULL, 0) > 0;
    8567         268 :         desc->callback.event_fn = event_cb;
    8568         268 :         desc->callback.ctx = event_ctx;
    8569         268 :         spdk_spin_init(&desc->spinlock);
    8570             : 
    8571         268 :         if (desc->opts.no_metadata) {
    8572           1 :                 if (spdk_bdev_is_md_separate(bdev)) {
    8573           0 :                         SPDK_ERRLOG("no_metadata option is not supported with separate metadata.\n");
    8574           0 :                         bdev_desc_free(desc);
    8575           0 :                         return -EINVAL;
    8576             :                 }
    8577           1 :         }
    8578             : 
    8579         268 :         if (bdev->media_events) {
    8580           0 :                 desc->media_events_buffer = calloc(MEDIA_EVENT_POOL_SIZE,
    8581             :                                                    sizeof(*desc->media_events_buffer));
    8582           0 :                 if (desc->media_events_buffer == NULL) {
    8583           0 :                         SPDK_ERRLOG("Failed to initialize media event pool\n");
    8584           0 :                         bdev_desc_free(desc);
    8585           0 :                         return -ENOMEM;
    8586             :                 }
    8587             : 
    8588           0 :                 for (i = 0; i < MEDIA_EVENT_POOL_SIZE; ++i) {
    8589           0 :                         TAILQ_INSERT_TAIL(&desc->free_media_events,
    8590             :                                           &desc->media_events_buffer[i], tailq);
    8591           0 :                 }
    8592           0 :         }
    8593             : 
    8594         268 :         if (bdev->fn_table->accel_sequence_supported != NULL) {
    8595           0 :                 for (i = 0; i < SPDK_BDEV_NUM_IO_TYPES; ++i) {
    8596           0 :                         desc->accel_sequence_supported[i] =
    8597           0 :                                 bdev->fn_table->accel_sequence_supported(bdev->ctxt,
    8598           0 :                                                 (enum spdk_bdev_io_type)i);
    8599           0 :                 }
    8600           0 :         }
    8601             : 
    8602         268 :         *_desc = desc;
    8603             : 
    8604         268 :         return 0;
    8605         268 : }
    8606             : 
    8607             : static int
    8608         130 : bdev_open_ext(const char *bdev_name, bool write, spdk_bdev_event_cb_t event_cb,
    8609             :               void *event_ctx, struct spdk_bdev_open_opts *opts,
    8610             :               struct spdk_bdev_desc **_desc)
    8611             : {
    8612             :         struct spdk_bdev_desc *desc;
    8613             :         struct spdk_bdev *bdev;
    8614             :         int rc;
    8615             : 
    8616         130 :         bdev = bdev_get_by_name(bdev_name);
    8617             : 
    8618         130 :         if (bdev == NULL) {
    8619           1 :                 SPDK_NOTICELOG("Currently unable to find bdev with name: %s\n", bdev_name);
    8620           1 :                 return -ENODEV;
    8621             :         }
    8622             : 
    8623         129 :         rc = bdev_desc_alloc(bdev, event_cb, event_ctx, opts, &desc);
    8624         129 :         if (rc != 0) {
    8625           0 :                 return rc;
    8626             :         }
    8627             : 
    8628         129 :         rc = bdev_open(bdev, write, desc);
    8629         129 :         if (rc != 0) {
    8630           7 :                 bdev_desc_free(desc);
    8631           7 :                 desc = NULL;
    8632           7 :         }
    8633             : 
    8634         129 :         *_desc = desc;
    8635             : 
    8636         129 :         return rc;
    8637         130 : }
    8638             : 
    8639             : int
    8640         132 : spdk_bdev_open_ext_v2(const char *bdev_name, bool write, spdk_bdev_event_cb_t event_cb,
    8641             :                       void *event_ctx, struct spdk_bdev_open_opts *opts,
    8642             :                       struct spdk_bdev_desc **_desc)
    8643             : {
    8644             :         int rc;
    8645             : 
    8646         132 :         if (event_cb == NULL) {
    8647           2 :                 SPDK_ERRLOG("Missing event callback function\n");
    8648           2 :                 return -EINVAL;
    8649             :         }
    8650             : 
    8651         130 :         spdk_spin_lock(&g_bdev_mgr.spinlock);
    8652         130 :         rc = bdev_open_ext(bdev_name, write, event_cb, event_ctx, opts, _desc);
    8653         130 :         spdk_spin_unlock(&g_bdev_mgr.spinlock);
    8654             : 
    8655         130 :         return rc;
    8656         132 : }
    8657             : 
    8658             : int
    8659         130 : spdk_bdev_open_ext(const char *bdev_name, bool write, spdk_bdev_event_cb_t event_cb,
    8660             :                    void *event_ctx, struct spdk_bdev_desc **_desc)
    8661             : {
    8662         130 :         return spdk_bdev_open_ext_v2(bdev_name, write, event_cb, event_ctx, NULL, _desc);
    8663             : }
    8664             : 
    8665             : struct spdk_bdev_open_async_ctx {
    8666             :         char                                    *bdev_name;
    8667             :         spdk_bdev_event_cb_t                    event_cb;
    8668             :         void                                    *event_ctx;
    8669             :         bool                                    write;
    8670             :         int                                     rc;
    8671             :         spdk_bdev_open_async_cb_t               cb_fn;
    8672             :         void                                    *cb_arg;
    8673             :         struct spdk_bdev_desc                   *desc;
    8674             :         struct spdk_bdev_open_async_opts        opts;
    8675             :         uint64_t                                start_ticks;
    8676             :         struct spdk_thread                      *orig_thread;
    8677             :         struct spdk_poller                      *poller;
    8678             :         TAILQ_ENTRY(spdk_bdev_open_async_ctx)   tailq;
    8679             : };
    8680             : 
    8681             : static void
    8682           0 : bdev_open_async_done(void *arg)
    8683             : {
    8684           0 :         struct spdk_bdev_open_async_ctx *ctx = arg;
    8685             : 
    8686           0 :         ctx->cb_fn(ctx->desc, ctx->rc, ctx->cb_arg);
    8687             : 
    8688           0 :         free(ctx->bdev_name);
    8689           0 :         free(ctx);
    8690           0 : }
    8691             : 
    8692             : static void
    8693           0 : bdev_open_async_cancel(void *arg)
    8694             : {
    8695           0 :         struct spdk_bdev_open_async_ctx *ctx = arg;
    8696             : 
    8697           0 :         assert(ctx->rc == -ESHUTDOWN);
    8698             : 
    8699           0 :         spdk_poller_unregister(&ctx->poller);
    8700             : 
    8701           0 :         bdev_open_async_done(ctx);
    8702           0 : }
    8703             : 
    8704             : /* This is called when the bdev library finishes at shutdown. */
    8705             : static void
    8706          65 : bdev_open_async_fini(void)
    8707             : {
    8708             :         struct spdk_bdev_open_async_ctx *ctx, *tmp_ctx;
    8709             : 
    8710          65 :         spdk_spin_lock(&g_bdev_mgr.spinlock);
    8711          65 :         TAILQ_FOREACH_SAFE(ctx, &g_bdev_mgr.async_bdev_opens, tailq, tmp_ctx) {
    8712           0 :                 TAILQ_REMOVE(&g_bdev_mgr.async_bdev_opens, ctx, tailq);
    8713             :                 /*
    8714             :                  * We have to move to ctx->orig_thread to unregister ctx->poller.
    8715             :                  * However, there is a chance that ctx->poller is executed before
    8716             :                  * message is executed, which could result in bdev_open_async_done()
    8717             :                  * being called twice. To avoid such race condition, set ctx->rc to
    8718             :                  * -ESHUTDOWN.
    8719             :                  */
    8720           0 :                 ctx->rc = -ESHUTDOWN;
    8721           0 :                 spdk_thread_send_msg(ctx->orig_thread, bdev_open_async_cancel, ctx);
    8722           0 :         }
    8723          65 :         spdk_spin_unlock(&g_bdev_mgr.spinlock);
    8724          65 : }
    8725             : 
    8726             : static int bdev_open_async(void *arg);
    8727             : 
    8728             : static void
    8729           0 : _bdev_open_async(struct spdk_bdev_open_async_ctx *ctx)
    8730             : {
    8731             :         uint64_t timeout_ticks;
    8732             : 
    8733           0 :         if (ctx->rc == -ESHUTDOWN) {
    8734             :                 /* This context is being canceled. Do nothing. */
    8735           0 :                 return;
    8736             :         }
    8737             : 
    8738           0 :         ctx->rc = bdev_open_ext(ctx->bdev_name, ctx->write, ctx->event_cb, ctx->event_ctx,
    8739           0 :                                 NULL, &ctx->desc);
    8740           0 :         if (ctx->rc == 0 || ctx->opts.timeout_ms == 0) {
    8741           0 :                 goto exit;
    8742             :         }
    8743             : 
    8744           0 :         timeout_ticks = ctx->start_ticks + ctx->opts.timeout_ms * spdk_get_ticks_hz() / 1000ull;
    8745           0 :         if (spdk_get_ticks() >= timeout_ticks) {
    8746           0 :                 SPDK_ERRLOG("Timed out while waiting for bdev '%s' to appear\n", ctx->bdev_name);
    8747           0 :                 ctx->rc = -ETIMEDOUT;
    8748           0 :                 goto exit;
    8749             :         }
    8750             : 
    8751           0 :         return;
    8752             : 
    8753             : exit:
    8754           0 :         spdk_poller_unregister(&ctx->poller);
    8755           0 :         TAILQ_REMOVE(&g_bdev_mgr.async_bdev_opens, ctx, tailq);
    8756             : 
    8757             :         /* Completion callback is processed after stack unwinding. */
    8758           0 :         spdk_thread_send_msg(ctx->orig_thread, bdev_open_async_done, ctx);
    8759           0 : }
    8760             : 
    8761             : static int
    8762           0 : bdev_open_async(void *arg)
    8763             : {
    8764           0 :         struct spdk_bdev_open_async_ctx *ctx = arg;
    8765             : 
    8766           0 :         spdk_spin_lock(&g_bdev_mgr.spinlock);
    8767             : 
    8768           0 :         _bdev_open_async(ctx);
    8769             : 
    8770           0 :         spdk_spin_unlock(&g_bdev_mgr.spinlock);
    8771             : 
    8772           0 :         return SPDK_POLLER_BUSY;
    8773             : }
    8774             : 
    8775             : static void
    8776           0 : bdev_open_async_opts_copy(struct spdk_bdev_open_async_opts *opts,
    8777             :                           struct spdk_bdev_open_async_opts *opts_src,
    8778             :                           size_t size)
    8779             : {
    8780           0 :         assert(opts);
    8781           0 :         assert(opts_src);
    8782             : 
    8783           0 :         opts->size = size;
    8784             : 
    8785             : #define SET_FIELD(field) \
    8786             :         if (offsetof(struct spdk_bdev_open_async_opts, field) + sizeof(opts->field) <= size) { \
    8787             :                 opts->field = opts_src->field; \
    8788             :         } \
    8789             : 
    8790           0 :         SET_FIELD(timeout_ms);
    8791             : 
    8792             :         /* Do not remove this statement, you should always update this statement when you adding a new field,
    8793             :          * and do not forget to add the SET_FIELD statement for your added field. */
    8794             :         SPDK_STATIC_ASSERT(sizeof(struct spdk_bdev_open_async_opts) == 16, "Incorrect size");
    8795             : 
    8796             : #undef SET_FIELD
    8797           0 : }
    8798             : 
    8799             : static void
    8800           0 : bdev_open_async_opts_get_default(struct spdk_bdev_open_async_opts *opts, size_t size)
    8801             : {
    8802           0 :         assert(opts);
    8803             : 
    8804           0 :         opts->size = size;
    8805             : 
    8806             : #define SET_FIELD(field, value) \
    8807             :         if (offsetof(struct spdk_bdev_open_async_opts, field) + sizeof(opts->field) <= size) { \
    8808             :                 opts->field = value; \
    8809             :         } \
    8810             : 
    8811           0 :         SET_FIELD(timeout_ms, 0);
    8812             : 
    8813             : #undef SET_FIELD
    8814           0 : }
    8815             : 
    8816             : int
    8817           0 : spdk_bdev_open_async(const char *bdev_name, bool write, spdk_bdev_event_cb_t event_cb,
    8818             :                      void *event_ctx, struct spdk_bdev_open_async_opts *opts,
    8819             :                      spdk_bdev_open_async_cb_t open_cb, void *open_cb_arg)
    8820             : {
    8821             :         struct spdk_bdev_open_async_ctx *ctx;
    8822             : 
    8823           0 :         if (event_cb == NULL) {
    8824           0 :                 SPDK_ERRLOG("Missing event callback function\n");
    8825           0 :                 return -EINVAL;
    8826             :         }
    8827             : 
    8828           0 :         if (open_cb == NULL) {
    8829           0 :                 SPDK_ERRLOG("Missing open callback function\n");
    8830           0 :                 return -EINVAL;
    8831             :         }
    8832             : 
    8833           0 :         if (opts != NULL && opts->size == 0) {
    8834           0 :                 SPDK_ERRLOG("size in the options structure should not be zero\n");
    8835           0 :                 return -EINVAL;
    8836             :         }
    8837             : 
    8838           0 :         ctx = calloc(1, sizeof(*ctx));
    8839           0 :         if (ctx == NULL) {
    8840           0 :                 SPDK_ERRLOG("Failed to allocate open context\n");
    8841           0 :                 return -ENOMEM;
    8842             :         }
    8843             : 
    8844           0 :         ctx->bdev_name = strdup(bdev_name);
    8845           0 :         if (ctx->bdev_name == NULL) {
    8846           0 :                 SPDK_ERRLOG("Failed to duplicate bdev_name\n");
    8847           0 :                 free(ctx);
    8848           0 :                 return -ENOMEM;
    8849             :         }
    8850             : 
    8851           0 :         ctx->poller = SPDK_POLLER_REGISTER(bdev_open_async, ctx, 100 * 1000);
    8852           0 :         if (ctx->poller == NULL) {
    8853           0 :                 SPDK_ERRLOG("Failed to register bdev_open_async poller\n");
    8854           0 :                 free(ctx->bdev_name);
    8855           0 :                 free(ctx);
    8856           0 :                 return -ENOMEM;
    8857             :         }
    8858             : 
    8859           0 :         ctx->cb_fn = open_cb;
    8860           0 :         ctx->cb_arg = open_cb_arg;
    8861           0 :         ctx->write = write;
    8862           0 :         ctx->event_cb = event_cb;
    8863           0 :         ctx->event_ctx = event_ctx;
    8864           0 :         ctx->orig_thread = spdk_get_thread();
    8865           0 :         ctx->start_ticks = spdk_get_ticks();
    8866             : 
    8867           0 :         bdev_open_async_opts_get_default(&ctx->opts, sizeof(ctx->opts));
    8868           0 :         if (opts != NULL) {
    8869           0 :                 bdev_open_async_opts_copy(&ctx->opts, opts, opts->size);
    8870           0 :         }
    8871             : 
    8872           0 :         spdk_spin_lock(&g_bdev_mgr.spinlock);
    8873             : 
    8874           0 :         TAILQ_INSERT_TAIL(&g_bdev_mgr.async_bdev_opens, ctx, tailq);
    8875           0 :         _bdev_open_async(ctx);
    8876             : 
    8877           0 :         spdk_spin_unlock(&g_bdev_mgr.spinlock);
    8878             : 
    8879           0 :         return 0;
    8880           0 : }
    8881             : 
    8882             : static void
    8883         259 : bdev_close(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc)
    8884             : {
    8885             :         int rc;
    8886             : 
    8887         259 :         spdk_spin_lock(&bdev->internal.spinlock);
    8888         259 :         spdk_spin_lock(&desc->spinlock);
    8889             : 
    8890         259 :         TAILQ_REMOVE(&bdev->internal.open_descs, desc, link);
    8891             : 
    8892         259 :         desc->closed = true;
    8893             : 
    8894         259 :         if (desc->claim != NULL) {
    8895          16 :                 bdev_desc_release_claims(desc);
    8896          16 :         }
    8897             : 
    8898         259 :         if (0 == desc->refs) {
    8899         248 :                 spdk_spin_unlock(&desc->spinlock);
    8900         248 :                 bdev_desc_free(desc);
    8901         248 :         } else {
    8902          11 :                 spdk_spin_unlock(&desc->spinlock);
    8903             :         }
    8904             : 
    8905             :         /* If no more descriptors, kill QoS channel */
    8906         259 :         if (bdev->internal.qos && TAILQ_EMPTY(&bdev->internal.open_descs)) {
    8907           7 :                 SPDK_DEBUGLOG(bdev, "Closed last descriptor for bdev %s on thread %p. Stopping QoS.\n",
    8908             :                               bdev->name, spdk_get_thread());
    8909             : 
    8910           7 :                 if (bdev_qos_destroy(bdev)) {
    8911             :                         /* There isn't anything we can do to recover here. Just let the
    8912             :                          * old QoS poller keep running. The QoS handling won't change
    8913             :                          * cores when the user allocates a new channel, but it won't break. */
    8914           0 :                         SPDK_ERRLOG("Unable to shut down QoS poller. It will continue running on the current thread.\n");
    8915           0 :                 }
    8916           7 :         }
    8917             : 
    8918         259 :         if (bdev->internal.status == SPDK_BDEV_STATUS_REMOVING && TAILQ_EMPTY(&bdev->internal.open_descs)) {
    8919          15 :                 rc = bdev_unregister_unsafe(bdev);
    8920          15 :                 spdk_spin_unlock(&bdev->internal.spinlock);
    8921             : 
    8922          15 :                 if (rc == 0) {
    8923          15 :                         spdk_io_device_unregister(__bdev_to_io_dev(bdev), bdev_destroy_cb);
    8924          15 :                 }
    8925          15 :         } else {
    8926         244 :                 spdk_spin_unlock(&bdev->internal.spinlock);
    8927             :         }
    8928         259 : }
    8929             : 
    8930             : void
    8931         122 : spdk_bdev_close(struct spdk_bdev_desc *desc)
    8932             : {
    8933         122 :         struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
    8934             : 
    8935         122 :         SPDK_DEBUGLOG(bdev, "Closing descriptor %p for bdev %s on thread %p\n", desc, bdev->name,
    8936             :                       spdk_get_thread());
    8937             : 
    8938         122 :         assert(desc->thread == spdk_get_thread());
    8939             : 
    8940         122 :         spdk_poller_unregister(&desc->io_timeout_poller);
    8941             : 
    8942         122 :         spdk_spin_lock(&g_bdev_mgr.spinlock);
    8943             : 
    8944         122 :         bdev_close(bdev, desc);
    8945             : 
    8946         122 :         spdk_spin_unlock(&g_bdev_mgr.spinlock);
    8947         122 : }
    8948             : 
    8949             : int32_t
    8950           3 : spdk_bdev_get_numa_id(struct spdk_bdev *bdev)
    8951             : {
    8952           3 :         if (bdev->numa.id_valid) {
    8953           2 :                 return bdev->numa.id;
    8954             :         } else {
    8955           1 :                 return SPDK_ENV_NUMA_ID_ANY;
    8956             :         }
    8957           3 : }
    8958             : 
    8959             : static void
    8960         126 : bdev_register_finished(void *arg)
    8961             : {
    8962         126 :         struct spdk_bdev_desc *desc = arg;
    8963         126 :         struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
    8964             : 
    8965         126 :         spdk_notify_send("bdev_register", spdk_bdev_get_name(bdev));
    8966             : 
    8967         126 :         spdk_spin_lock(&g_bdev_mgr.spinlock);
    8968             : 
    8969         126 :         bdev_close(bdev, desc);
    8970             : 
    8971         126 :         spdk_spin_unlock(&g_bdev_mgr.spinlock);
    8972         126 : }
    8973             : 
    8974             : int
    8975         129 : spdk_bdev_register(struct spdk_bdev *bdev)
    8976             : {
    8977             :         struct spdk_bdev_desc *desc;
    8978         129 :         struct spdk_thread *thread = spdk_get_thread();
    8979             :         int rc;
    8980             : 
    8981         129 :         if (spdk_unlikely(!spdk_thread_is_app_thread(NULL))) {
    8982           1 :                 SPDK_ERRLOG("Cannot register bdev %s on thread %p (%s)\n", bdev->name, thread,
    8983             :                             thread ? spdk_thread_get_name(thread) : "null");
    8984           1 :                 return -EINVAL;
    8985             :         }
    8986             : 
    8987         128 :         rc = bdev_register(bdev);
    8988         128 :         if (rc != 0) {
    8989           2 :                 return rc;
    8990             :         }
    8991             : 
    8992             :         /* A descriptor is opened to prevent bdev deletion during examination */
    8993         126 :         rc = bdev_desc_alloc(bdev, _tmp_bdev_event_cb, NULL, NULL, &desc);
    8994         126 :         if (rc != 0) {
    8995           0 :                 spdk_bdev_unregister(bdev, NULL, NULL);
    8996           0 :                 return rc;
    8997             :         }
    8998             : 
    8999         126 :         rc = bdev_open(bdev, false, desc);
    9000         126 :         if (rc != 0) {
    9001           0 :                 bdev_desc_free(desc);
    9002           0 :                 spdk_bdev_unregister(bdev, NULL, NULL);
    9003           0 :                 return rc;
    9004             :         }
    9005             : 
    9006             :         /* Examine configuration before initializing I/O */
    9007         126 :         bdev_examine(bdev);
    9008             : 
    9009         126 :         rc = spdk_bdev_wait_for_examine(bdev_register_finished, desc);
    9010         126 :         if (rc != 0) {
    9011           0 :                 bdev_close(bdev, desc);
    9012           0 :                 spdk_bdev_unregister(bdev, NULL, NULL);
    9013           0 :         }
    9014             : 
    9015         126 :         return rc;
    9016         129 : }
    9017             : 
    9018             : int
    9019          26 : spdk_bdev_module_claim_bdev(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc,
    9020             :                             struct spdk_bdev_module *module)
    9021             : {
    9022          26 :         spdk_spin_lock(&bdev->internal.spinlock);
    9023             : 
    9024          26 :         if (bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE) {
    9025           6 :                 LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
    9026           6 :                 spdk_spin_unlock(&bdev->internal.spinlock);
    9027           6 :                 return -EPERM;
    9028             :         }
    9029             : 
    9030          20 :         if (desc && !desc->write) {
    9031           5 :                 desc->write = true;
    9032           5 :         }
    9033             : 
    9034          20 :         bdev->internal.claim_type = SPDK_BDEV_CLAIM_EXCL_WRITE;
    9035          20 :         bdev->internal.claim.v1.module = module;
    9036             : 
    9037          20 :         spdk_spin_unlock(&bdev->internal.spinlock);
    9038          20 :         return 0;
    9039          26 : }
    9040             : 
    9041             : void
    9042           8 : spdk_bdev_module_release_bdev(struct spdk_bdev *bdev)
    9043             : {
    9044           8 :         spdk_spin_lock(&bdev->internal.spinlock);
    9045             : 
    9046           8 :         assert(bdev->internal.claim.v1.module != NULL);
    9047           8 :         assert(bdev->internal.claim_type == SPDK_BDEV_CLAIM_EXCL_WRITE);
    9048           8 :         bdev->internal.claim_type = SPDK_BDEV_CLAIM_NONE;
    9049           8 :         bdev->internal.claim.v1.module = NULL;
    9050             : 
    9051           8 :         spdk_spin_unlock(&bdev->internal.spinlock);
    9052           8 : }
    9053             : 
    9054             : /*
    9055             :  * Start claims v2
    9056             :  */
    9057             : 
    9058             : const char *
    9059          24 : spdk_bdev_claim_get_name(enum spdk_bdev_claim_type type)
    9060             : {
    9061          24 :         switch (type) {
    9062             :         case SPDK_BDEV_CLAIM_NONE:
    9063           0 :                 return "not_claimed";
    9064             :         case SPDK_BDEV_CLAIM_EXCL_WRITE:
    9065           6 :                 return "exclusive_write";
    9066             :         case SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE:
    9067           7 :                 return "read_many_write_one";
    9068             :         case SPDK_BDEV_CLAIM_READ_MANY_WRITE_NONE:
    9069           5 :                 return "read_many_write_none";
    9070             :         case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
    9071           6 :                 return "read_many_write_many";
    9072             :         default:
    9073           0 :                 break;
    9074             :         }
    9075           0 :         return "invalid_claim";
    9076          24 : }
    9077             : 
    9078             : static bool
    9079          96 : claim_type_is_v2(enum spdk_bdev_claim_type type)
    9080             : {
    9081          96 :         switch (type) {
    9082             :         case SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE:
    9083             :         case SPDK_BDEV_CLAIM_READ_MANY_WRITE_NONE:
    9084             :         case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
    9085          96 :                 return true;
    9086             :         default:
    9087           0 :                 break;
    9088             :         }
    9089           0 :         return false;
    9090          96 : }
    9091             : 
    9092             : /* Returns true if taking a claim with desc->write == false should make the descriptor writable. */
    9093             : static bool
    9094          13 : claim_type_promotes_to_write(enum spdk_bdev_claim_type type)
    9095             : {
    9096          13 :         switch (type) {
    9097             :         case SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE:
    9098             :         case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
    9099           5 :                 return true;
    9100             :         default:
    9101           8 :                 break;
    9102             :         }
    9103           8 :         return false;
    9104          13 : }
    9105             : 
    9106             : void
    9107          44 : spdk_bdev_claim_opts_init(struct spdk_bdev_claim_opts *opts, size_t size)
    9108             : {
    9109          44 :         if (opts == NULL) {
    9110           0 :                 SPDK_ERRLOG("opts should not be NULL\n");
    9111           0 :                 assert(opts != NULL);
    9112           0 :                 return;
    9113             :         }
    9114          44 :         if (size == 0) {
    9115           0 :                 SPDK_ERRLOG("size should not be zero\n");
    9116           0 :                 assert(size != 0);
    9117           0 :                 return;
    9118             :         }
    9119             : 
    9120          44 :         memset(opts, 0, size);
    9121          44 :         opts->opts_size = size;
    9122             : 
    9123             : #define FIELD_OK(field) \
    9124             :         offsetof(struct spdk_bdev_claim_opts, field) + sizeof(opts->field) <= size
    9125             : 
    9126             : #define SET_FIELD(field, value) \
    9127             :         if (FIELD_OK(field)) { \
    9128             :                 opts->field = value; \
    9129             :         } \
    9130             : 
    9131          44 :         SET_FIELD(shared_claim_key, 0);
    9132             : 
    9133             : #undef FIELD_OK
    9134             : #undef SET_FIELD
    9135          44 : }
    9136             : 
    9137             : static int
    9138          22 : claim_opts_copy(struct spdk_bdev_claim_opts *src, struct spdk_bdev_claim_opts *dst)
    9139             : {
    9140          22 :         if (src->opts_size == 0) {
    9141           0 :                 SPDK_ERRLOG("size should not be zero\n");
    9142           0 :                 return -1;
    9143             :         }
    9144             : 
    9145          22 :         memset(dst, 0, sizeof(*dst));
    9146          22 :         dst->opts_size = src->opts_size;
    9147             : 
    9148             : #define FIELD_OK(field) \
    9149             :         offsetof(struct spdk_bdev_claim_opts, field) + sizeof(src->field) <= src->opts_size
    9150             : 
    9151             : #define SET_FIELD(field) \
    9152             :         if (FIELD_OK(field)) { \
    9153             :                 dst->field = src->field; \
    9154             :         } \
    9155             : 
    9156          22 :         if (FIELD_OK(name)) {
    9157          22 :                 snprintf(dst->name, sizeof(dst->name), "%s", src->name);
    9158          22 :         }
    9159             : 
    9160          22 :         SET_FIELD(shared_claim_key);
    9161             : 
    9162             :         /* You should not remove this statement, but need to update the assert statement
    9163             :          * if you add a new field, and also add a corresponding SET_FIELD statement */
    9164             :         SPDK_STATIC_ASSERT(sizeof(struct spdk_bdev_claim_opts) == 48, "Incorrect size");
    9165             : 
    9166             : #undef FIELD_OK
    9167             : #undef SET_FIELD
    9168          22 :         return 0;
    9169          22 : }
    9170             : 
    9171             : /* Returns 0 if a read-write-once claim can be taken. */
    9172             : static int
    9173           9 : claim_verify_rwo(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
    9174             :                  struct spdk_bdev_claim_opts *opts, struct spdk_bdev_module *module)
    9175             : {
    9176           9 :         struct spdk_bdev *bdev = desc->bdev;
    9177             :         struct spdk_bdev_desc *open_desc;
    9178             : 
    9179           9 :         assert(spdk_spin_held(&bdev->internal.spinlock));
    9180           9 :         assert(type == SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE);
    9181             : 
    9182           9 :         if (opts->shared_claim_key != 0) {
    9183           1 :                 SPDK_ERRLOG("%s: key option not supported with read-write-once claims\n",
    9184             :                             bdev->name);
    9185           1 :                 return -EINVAL;
    9186             :         }
    9187           8 :         if (bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE) {
    9188           1 :                 LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
    9189           1 :                 return -EPERM;
    9190             :         }
    9191           7 :         if (desc->claim != NULL) {
    9192           0 :                 SPDK_NOTICELOG("%s: descriptor already claimed bdev with module %s\n",
    9193             :                                bdev->name, desc->claim->module->name);
    9194           0 :                 return -EPERM;
    9195             :         }
    9196          14 :         TAILQ_FOREACH(open_desc, &bdev->internal.open_descs, link) {
    9197           9 :                 if (desc != open_desc && open_desc->write) {
    9198           2 :                         SPDK_NOTICELOG("%s: Cannot obtain read-write-once claim while "
    9199             :                                        "another descriptor is open for writing\n",
    9200             :                                        bdev->name);
    9201           2 :                         return -EPERM;
    9202             :                 }
    9203           7 :         }
    9204             : 
    9205           5 :         return 0;
    9206           9 : }
    9207             : 
    9208             : /* Returns 0 if a read-only-many claim can be taken. */
    9209             : static int
    9210          12 : claim_verify_rom(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
    9211             :                  struct spdk_bdev_claim_opts *opts, struct spdk_bdev_module *module)
    9212             : {
    9213          12 :         struct spdk_bdev *bdev = desc->bdev;
    9214             :         struct spdk_bdev_desc *open_desc;
    9215             : 
    9216          12 :         assert(spdk_spin_held(&bdev->internal.spinlock));
    9217          12 :         assert(type == SPDK_BDEV_CLAIM_READ_MANY_WRITE_NONE);
    9218          12 :         assert(desc->claim == NULL);
    9219             : 
    9220          12 :         if (desc->write) {
    9221           3 :                 SPDK_ERRLOG("%s: Cannot obtain read-only-many claim with writable descriptor\n",
    9222             :                             bdev->name);
    9223           3 :                 return -EINVAL;
    9224             :         }
    9225           9 :         if (opts->shared_claim_key != 0) {
    9226           1 :                 SPDK_ERRLOG("%s: key option not supported with read-only-may claims\n", bdev->name);
    9227           1 :                 return -EINVAL;
    9228             :         }
    9229           8 :         if (bdev->internal.claim_type == SPDK_BDEV_CLAIM_NONE) {
    9230          15 :                 TAILQ_FOREACH(open_desc, &bdev->internal.open_descs, link) {
    9231           9 :                         if (open_desc->write) {
    9232           0 :                                 SPDK_NOTICELOG("%s: Cannot obtain read-only-many claim while "
    9233             :                                                "another descriptor is open for writing\n",
    9234             :                                                bdev->name);
    9235           0 :                                 return -EPERM;
    9236             :                         }
    9237           9 :                 }
    9238           6 :         }
    9239             : 
    9240           8 :         return 0;
    9241          12 : }
    9242             : 
    9243             : /* Returns 0 if a read-write-many claim can be taken. */
    9244             : static int
    9245           8 : claim_verify_rwm(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
    9246             :                  struct spdk_bdev_claim_opts *opts, struct spdk_bdev_module *module)
    9247             : {
    9248           8 :         struct spdk_bdev *bdev = desc->bdev;
    9249             :         struct spdk_bdev_desc *open_desc;
    9250             : 
    9251           8 :         assert(spdk_spin_held(&bdev->internal.spinlock));
    9252           8 :         assert(type == SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED);
    9253           8 :         assert(desc->claim == NULL);
    9254             : 
    9255           8 :         if (opts->shared_claim_key == 0) {
    9256           2 :                 SPDK_ERRLOG("%s: shared_claim_key option required with read-write-may claims\n",
    9257             :                             bdev->name);
    9258           2 :                 return -EINVAL;
    9259             :         }
    9260           6 :         switch (bdev->internal.claim_type) {
    9261             :         case SPDK_BDEV_CLAIM_NONE:
    9262           7 :                 TAILQ_FOREACH(open_desc, &bdev->internal.open_descs, link) {
    9263           5 :                         if (open_desc == desc) {
    9264           3 :                                 continue;
    9265             :                         }
    9266           2 :                         if (open_desc->write) {
    9267           2 :                                 SPDK_NOTICELOG("%s: Cannot obtain read-write-many claim while "
    9268             :                                                "another descriptor is open for writing without a "
    9269             :                                                "claim\n", bdev->name);
    9270           2 :                                 return -EPERM;
    9271             :                         }
    9272           0 :                 }
    9273           2 :                 break;
    9274             :         case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
    9275           2 :                 if (opts->shared_claim_key != bdev->internal.claim.v2.key) {
    9276           1 :                         LOG_ALREADY_CLAIMED_ERROR("already claimed with another key", bdev);
    9277           1 :                         return -EPERM;
    9278             :                 }
    9279           1 :                 break;
    9280             :         default:
    9281           0 :                 LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
    9282           0 :                 return -EBUSY;
    9283             :         }
    9284             : 
    9285           3 :         return 0;
    9286           8 : }
    9287             : 
    9288             : /* Updates desc and its bdev with a v2 claim. */
    9289             : static int
    9290          16 : claim_bdev(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
    9291             :            struct spdk_bdev_claim_opts *opts, struct spdk_bdev_module *module)
    9292             : {
    9293          16 :         struct spdk_bdev *bdev = desc->bdev;
    9294             :         struct spdk_bdev_module_claim *claim;
    9295             : 
    9296          16 :         assert(spdk_spin_held(&bdev->internal.spinlock));
    9297          16 :         assert(claim_type_is_v2(type));
    9298          16 :         assert(desc->claim == NULL);
    9299             : 
    9300          16 :         claim = calloc(1, sizeof(*desc->claim));
    9301          16 :         if (claim == NULL) {
    9302           0 :                 SPDK_ERRLOG("%s: out of memory while allocating claim\n", bdev->name);
    9303           0 :                 return -ENOMEM;
    9304             :         }
    9305          16 :         claim->module = module;
    9306          16 :         claim->desc = desc;
    9307             :         SPDK_STATIC_ASSERT(sizeof(claim->name) == sizeof(opts->name), "sizes must match");
    9308          16 :         memcpy(claim->name, opts->name, sizeof(claim->name));
    9309          16 :         desc->claim = claim;
    9310             : 
    9311          16 :         if (bdev->internal.claim_type == SPDK_BDEV_CLAIM_NONE) {
    9312          13 :                 bdev->internal.claim_type = type;
    9313          13 :                 TAILQ_INIT(&bdev->internal.claim.v2.claims);
    9314          13 :                 bdev->internal.claim.v2.key = opts->shared_claim_key;
    9315          13 :         }
    9316          16 :         assert(type == bdev->internal.claim_type);
    9317             : 
    9318          16 :         TAILQ_INSERT_TAIL(&bdev->internal.claim.v2.claims, claim, link);
    9319             : 
    9320          16 :         if (!desc->write && claim_type_promotes_to_write(type)) {
    9321           5 :                 desc->write = true;
    9322           5 :         }
    9323             : 
    9324          16 :         return 0;
    9325          16 : }
    9326             : 
    9327             : int
    9328          39 : spdk_bdev_module_claim_bdev_desc(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
    9329             :                                  struct spdk_bdev_claim_opts *_opts,
    9330             :                                  struct spdk_bdev_module *module)
    9331             : {
    9332             :         struct spdk_bdev *bdev;
    9333             :         struct spdk_bdev_claim_opts opts;
    9334          39 :         int rc = 0;
    9335             : 
    9336          39 :         if (desc == NULL) {
    9337           0 :                 SPDK_ERRLOG("descriptor must not be NULL\n");
    9338           0 :                 return -EINVAL;
    9339             :         }
    9340             : 
    9341          39 :         bdev = desc->bdev;
    9342             : 
    9343          39 :         if (_opts == NULL) {
    9344          17 :                 spdk_bdev_claim_opts_init(&opts, sizeof(opts));
    9345          39 :         } else if (claim_opts_copy(_opts, &opts) != 0) {
    9346           0 :                 return -EINVAL;
    9347             :         }
    9348             : 
    9349          39 :         spdk_spin_lock(&bdev->internal.spinlock);
    9350             : 
    9351          39 :         if (bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE &&
    9352          15 :             bdev->internal.claim_type != type) {
    9353          10 :                 LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
    9354          10 :                 spdk_spin_unlock(&bdev->internal.spinlock);
    9355          10 :                 return -EPERM;
    9356             :         }
    9357             : 
    9358          29 :         if (claim_type_is_v2(type) && desc->claim != NULL) {
    9359           0 :                 SPDK_ERRLOG("%s: descriptor already has %s claim with name '%s'\n",
    9360             :                             bdev->name, spdk_bdev_claim_get_name(type), desc->claim->name);
    9361           0 :                 spdk_spin_unlock(&bdev->internal.spinlock);
    9362           0 :                 return -EPERM;
    9363             :         }
    9364             : 
    9365          29 :         switch (type) {
    9366             :         case SPDK_BDEV_CLAIM_EXCL_WRITE:
    9367           0 :                 spdk_spin_unlock(&bdev->internal.spinlock);
    9368           0 :                 return spdk_bdev_module_claim_bdev(bdev, desc, module);
    9369             :         case SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE:
    9370           9 :                 rc = claim_verify_rwo(desc, type, &opts, module);
    9371           9 :                 break;
    9372             :         case SPDK_BDEV_CLAIM_READ_MANY_WRITE_NONE:
    9373          12 :                 rc = claim_verify_rom(desc, type, &opts, module);
    9374          12 :                 break;
    9375             :         case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
    9376           8 :                 rc = claim_verify_rwm(desc, type, &opts, module);
    9377           8 :                 break;
    9378             :         default:
    9379           0 :                 SPDK_ERRLOG("%s: claim type %d not supported\n", bdev->name, type);
    9380           0 :                 rc = -ENOTSUP;
    9381           0 :         }
    9382             : 
    9383          29 :         if (rc == 0) {
    9384          16 :                 rc = claim_bdev(desc, type, &opts, module);
    9385          16 :         }
    9386             : 
    9387          29 :         spdk_spin_unlock(&bdev->internal.spinlock);
    9388          29 :         return rc;
    9389          39 : }
    9390             : 
    9391             : static void
    9392          13 : claim_reset(struct spdk_bdev *bdev)
    9393             : {
    9394          13 :         assert(spdk_spin_held(&bdev->internal.spinlock));
    9395          13 :         assert(claim_type_is_v2(bdev->internal.claim_type));
    9396          13 :         assert(TAILQ_EMPTY(&bdev->internal.claim.v2.claims));
    9397             : 
    9398          13 :         memset(&bdev->internal.claim, 0, sizeof(bdev->internal.claim));
    9399          13 :         bdev->internal.claim_type = SPDK_BDEV_CLAIM_NONE;
    9400          13 : }
    9401             : 
    9402             : static void
    9403          16 : bdev_desc_release_claims(struct spdk_bdev_desc *desc)
    9404             : {
    9405          16 :         struct spdk_bdev *bdev = desc->bdev;
    9406             : 
    9407          16 :         assert(spdk_spin_held(&bdev->internal.spinlock));
    9408          16 :         assert(claim_type_is_v2(bdev->internal.claim_type));
    9409             : 
    9410          16 :         if (bdev->internal.examine_in_progress == 0) {
    9411          16 :                 TAILQ_REMOVE(&bdev->internal.claim.v2.claims, desc->claim, link);
    9412          16 :                 free(desc->claim);
    9413          16 :                 if (TAILQ_EMPTY(&bdev->internal.claim.v2.claims)) {
    9414          13 :                         claim_reset(bdev);
    9415          13 :                 }
    9416          16 :         } else {
    9417             :                 /* This is a dead claim that will be cleaned up when bdev_examine() is done. */
    9418           0 :                 desc->claim->module = NULL;
    9419           0 :                 desc->claim->desc = NULL;
    9420             :         }
    9421          16 :         desc->claim = NULL;
    9422          16 : }
    9423             : 
    9424             : /*
    9425             :  * End claims v2
    9426             :  */
    9427             : 
    9428             : struct spdk_bdev *
    9429        1541 : spdk_bdev_desc_get_bdev(struct spdk_bdev_desc *desc)
    9430             : {
    9431        1541 :         assert(desc != NULL);
    9432        1541 :         return desc->bdev;
    9433             : }
    9434             : 
    9435             : int
    9436           1 : spdk_for_each_bdev(void *ctx, spdk_for_each_bdev_fn fn)
    9437             : {
    9438             :         struct spdk_bdev *bdev, *tmp;
    9439             :         struct spdk_bdev_desc *desc;
    9440           1 :         int rc = 0;
    9441             : 
    9442           1 :         assert(fn != NULL);
    9443             : 
    9444           1 :         spdk_spin_lock(&g_bdev_mgr.spinlock);
    9445           1 :         bdev = spdk_bdev_first();
    9446           9 :         while (bdev != NULL) {
    9447           8 :                 rc = bdev_desc_alloc(bdev, _tmp_bdev_event_cb, NULL, NULL, &desc);
    9448           8 :                 if (rc != 0) {
    9449           0 :                         break;
    9450             :                 }
    9451           8 :                 rc = bdev_open(bdev, false, desc);
    9452           8 :                 if (rc != 0) {
    9453           1 :                         bdev_desc_free(desc);
    9454           1 :                         if (rc == -ENODEV) {
    9455             :                                 /* Ignore the error and move to the next bdev. */
    9456           1 :                                 rc = 0;
    9457           1 :                                 bdev = spdk_bdev_next(bdev);
    9458           1 :                                 continue;
    9459             :                         }
    9460           0 :                         break;
    9461             :                 }
    9462           7 :                 spdk_spin_unlock(&g_bdev_mgr.spinlock);
    9463             : 
    9464           7 :                 rc = fn(ctx, bdev);
    9465             : 
    9466           7 :                 spdk_spin_lock(&g_bdev_mgr.spinlock);
    9467           7 :                 tmp = spdk_bdev_next(bdev);
    9468           7 :                 bdev_close(bdev, desc);
    9469           7 :                 if (rc != 0) {
    9470           0 :                         break;
    9471             :                 }
    9472           7 :                 bdev = tmp;
    9473             :         }
    9474           1 :         spdk_spin_unlock(&g_bdev_mgr.spinlock);
    9475             : 
    9476           1 :         return rc;
    9477             : }
    9478             : 
    9479             : int
    9480           1 : spdk_for_each_bdev_leaf(void *ctx, spdk_for_each_bdev_fn fn)
    9481             : {
    9482             :         struct spdk_bdev *bdev, *tmp;
    9483             :         struct spdk_bdev_desc *desc;
    9484           1 :         int rc = 0;
    9485             : 
    9486           1 :         assert(fn != NULL);
    9487             : 
    9488           1 :         spdk_spin_lock(&g_bdev_mgr.spinlock);
    9489           1 :         bdev = spdk_bdev_first_leaf();
    9490           6 :         while (bdev != NULL) {
    9491           5 :                 rc = bdev_desc_alloc(bdev, _tmp_bdev_event_cb, NULL, NULL, &desc);
    9492           5 :                 if (rc != 0) {
    9493           0 :                         break;
    9494             :                 }
    9495           5 :                 rc = bdev_open(bdev, false, desc);
    9496           5 :                 if (rc != 0) {
    9497           1 :                         bdev_desc_free(desc);
    9498           1 :                         if (rc == -ENODEV) {
    9499             :                                 /* Ignore the error and move to the next bdev. */
    9500           1 :                                 rc = 0;
    9501           1 :                                 bdev = spdk_bdev_next_leaf(bdev);
    9502           1 :                                 continue;
    9503             :                         }
    9504           0 :                         break;
    9505             :                 }
    9506           4 :                 spdk_spin_unlock(&g_bdev_mgr.spinlock);
    9507             : 
    9508           4 :                 rc = fn(ctx, bdev);
    9509             : 
    9510           4 :                 spdk_spin_lock(&g_bdev_mgr.spinlock);
    9511           4 :                 tmp = spdk_bdev_next_leaf(bdev);
    9512           4 :                 bdev_close(bdev, desc);
    9513           4 :                 if (rc != 0) {
    9514           0 :                         break;
    9515             :                 }
    9516           4 :                 bdev = tmp;
    9517             :         }
    9518           1 :         spdk_spin_unlock(&g_bdev_mgr.spinlock);
    9519             : 
    9520           1 :         return rc;
    9521             : }
    9522             : 
    9523             : void
    9524           0 : spdk_bdev_io_get_iovec(struct spdk_bdev_io *bdev_io, struct iovec **iovp, int *iovcntp)
    9525             : {
    9526             :         struct iovec *iovs;
    9527             :         int iovcnt;
    9528             : 
    9529           0 :         if (bdev_io == NULL) {
    9530           0 :                 return;
    9531             :         }
    9532             : 
    9533           0 :         switch (bdev_io->type) {
    9534             :         case SPDK_BDEV_IO_TYPE_READ:
    9535             :         case SPDK_BDEV_IO_TYPE_WRITE:
    9536             :         case SPDK_BDEV_IO_TYPE_ZCOPY:
    9537           0 :                 iovs = bdev_io->u.bdev.iovs;
    9538           0 :                 iovcnt = bdev_io->u.bdev.iovcnt;
    9539           0 :                 break;
    9540             :         default:
    9541           0 :                 iovs = NULL;
    9542           0 :                 iovcnt = 0;
    9543           0 :                 break;
    9544             :         }
    9545             : 
    9546           0 :         if (iovp) {
    9547           0 :                 *iovp = iovs;
    9548           0 :         }
    9549           0 :         if (iovcntp) {
    9550           0 :                 *iovcntp = iovcnt;
    9551           0 :         }
    9552           0 : }
    9553             : 
    9554             : void *
    9555           0 : spdk_bdev_io_get_md_buf(struct spdk_bdev_io *bdev_io)
    9556             : {
    9557           0 :         if (bdev_io == NULL) {
    9558           0 :                 return NULL;
    9559             :         }
    9560             : 
    9561           0 :         if (!spdk_bdev_is_md_separate(bdev_io->bdev)) {
    9562           0 :                 return NULL;
    9563             :         }
    9564             : 
    9565           0 :         if (bdev_io->type == SPDK_BDEV_IO_TYPE_READ ||
    9566           0 :             bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
    9567           0 :                 return bdev_io->u.bdev.md_buf;
    9568             :         }
    9569             : 
    9570           0 :         return NULL;
    9571           0 : }
    9572             : 
    9573             : void *
    9574           0 : spdk_bdev_io_get_cb_arg(struct spdk_bdev_io *bdev_io)
    9575             : {
    9576           0 :         if (bdev_io == NULL) {
    9577           0 :                 assert(false);
    9578             :                 return NULL;
    9579             :         }
    9580             : 
    9581           0 :         return bdev_io->internal.caller_ctx;
    9582             : }
    9583             : 
    9584             : void
    9585           7 : spdk_bdev_module_list_add(struct spdk_bdev_module *bdev_module)
    9586             : {
    9587             : 
    9588           7 :         if (spdk_bdev_module_list_find(bdev_module->name)) {
    9589           0 :                 SPDK_ERRLOG("ERROR: module '%s' already registered.\n", bdev_module->name);
    9590           0 :                 assert(false);
    9591             :         }
    9592             : 
    9593           7 :         spdk_spin_init(&bdev_module->internal.spinlock);
    9594           7 :         TAILQ_INIT(&bdev_module->internal.quiesced_ranges);
    9595             : 
    9596             :         /*
    9597             :          * Modules with examine callbacks must be initialized first, so they are
    9598             :          *  ready to handle examine callbacks from later modules that will
    9599             :          *  register physical bdevs.
    9600             :          */
    9601           7 :         if (bdev_module->examine_config != NULL || bdev_module->examine_disk != NULL) {
    9602           4 :                 TAILQ_INSERT_HEAD(&g_bdev_mgr.bdev_modules, bdev_module, internal.tailq);
    9603           4 :         } else {
    9604           3 :                 TAILQ_INSERT_TAIL(&g_bdev_mgr.bdev_modules, bdev_module, internal.tailq);
    9605             :         }
    9606           7 : }
    9607             : 
    9608             : struct spdk_bdev_module *
    9609           7 : spdk_bdev_module_list_find(const char *name)
    9610             : {
    9611             :         struct spdk_bdev_module *bdev_module;
    9612             : 
    9613          14 :         TAILQ_FOREACH(bdev_module, &g_bdev_mgr.bdev_modules, internal.tailq) {
    9614           7 :                 if (strcmp(name, bdev_module->name) == 0) {
    9615           0 :                         break;
    9616             :                 }
    9617           7 :         }
    9618             : 
    9619           7 :         return bdev_module;
    9620             : }
    9621             : 
    9622             : static int
    9623           6 : bdev_write_zero_buffer(struct spdk_bdev_io *bdev_io)
    9624             : {
    9625             :         uint64_t num_blocks;
    9626           6 :         void *md_buf = NULL;
    9627             : 
    9628           6 :         num_blocks = bdev_io->u.bdev.num_blocks;
    9629             : 
    9630           6 :         if (spdk_bdev_is_md_separate(bdev_io->bdev)) {
    9631           4 :                 md_buf = (char *)g_bdev_mgr.zero_buffer +
    9632           2 :                          spdk_bdev_get_block_size(bdev_io->bdev) * num_blocks;
    9633           2 :         }
    9634             : 
    9635          12 :         return bdev_write_blocks_with_md(bdev_io->internal.desc,
    9636           6 :                                          spdk_io_channel_from_ctx(bdev_io->internal.ch),
    9637           6 :                                          g_bdev_mgr.zero_buffer, md_buf,
    9638           6 :                                          bdev_io->u.bdev.offset_blocks, num_blocks,
    9639           6 :                                          bdev_write_zero_buffer_done, bdev_io);
    9640             : }
    9641             : 
    9642             : static void
    9643           6 : bdev_write_zero_buffer_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
    9644             : {
    9645           6 :         struct spdk_bdev_io *parent_io = cb_arg;
    9646             : 
    9647           6 :         spdk_bdev_free_io(bdev_io);
    9648             : 
    9649           6 :         parent_io->internal.status = success ? SPDK_BDEV_IO_STATUS_SUCCESS : SPDK_BDEV_IO_STATUS_FAILED;
    9650           6 :         parent_io->internal.cb(parent_io, success, parent_io->internal.caller_ctx);
    9651           6 : }
    9652             : 
    9653             : static void
    9654          10 : bdev_set_qos_limit_done(struct set_qos_limit_ctx *ctx, int status)
    9655             : {
    9656          10 :         spdk_spin_lock(&ctx->bdev->internal.spinlock);
    9657          10 :         ctx->bdev->internal.qos_mod_in_progress = false;
    9658          10 :         spdk_spin_unlock(&ctx->bdev->internal.spinlock);
    9659             : 
    9660          10 :         if (ctx->cb_fn) {
    9661           8 :                 ctx->cb_fn(ctx->cb_arg, status);
    9662           8 :         }
    9663          10 :         free(ctx);
    9664          10 : }
    9665             : 
    9666             : static void
    9667           2 : bdev_disable_qos_done(void *cb_arg)
    9668             : {
    9669           2 :         struct set_qos_limit_ctx *ctx = cb_arg;
    9670           2 :         struct spdk_bdev *bdev = ctx->bdev;
    9671             :         struct spdk_bdev_qos *qos;
    9672             : 
    9673           2 :         spdk_spin_lock(&bdev->internal.spinlock);
    9674           2 :         qos = bdev->internal.qos;
    9675           2 :         bdev->internal.qos = NULL;
    9676           2 :         spdk_spin_unlock(&bdev->internal.spinlock);
    9677             : 
    9678           2 :         if (qos->thread != NULL) {
    9679           2 :                 spdk_put_io_channel(spdk_io_channel_from_ctx(qos->ch));
    9680           2 :                 spdk_poller_unregister(&qos->poller);
    9681           2 :         }
    9682             : 
    9683           2 :         free(qos);
    9684             : 
    9685           2 :         bdev_set_qos_limit_done(ctx, 0);
    9686           2 : }
    9687             : 
    9688             : static void
    9689           2 : bdev_disable_qos_msg_done(struct spdk_bdev *bdev, void *_ctx, int status)
    9690             : {
    9691           2 :         struct set_qos_limit_ctx *ctx = _ctx;
    9692             :         struct spdk_thread *thread;
    9693             : 
    9694           2 :         spdk_spin_lock(&bdev->internal.spinlock);
    9695           2 :         thread = bdev->internal.qos->thread;
    9696           2 :         spdk_spin_unlock(&bdev->internal.spinlock);
    9697             : 
    9698           2 :         if (thread != NULL) {
    9699           2 :                 spdk_thread_send_msg(thread, bdev_disable_qos_done, ctx);
    9700           2 :         } else {
    9701           0 :                 bdev_disable_qos_done(ctx);
    9702             :         }
    9703           2 : }
    9704             : 
    9705             : static void
    9706           4 : bdev_disable_qos_msg(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
    9707             :                      struct spdk_io_channel *ch, void *_ctx)
    9708             : {
    9709           4 :         struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(ch);
    9710             :         struct spdk_bdev_io *bdev_io;
    9711             : 
    9712           4 :         bdev_ch->flags &= ~BDEV_CH_QOS_ENABLED;
    9713             : 
    9714           6 :         while (!TAILQ_EMPTY(&bdev_ch->qos_queued_io)) {
    9715             :                 /* Re-submit the queued I/O. */
    9716           2 :                 bdev_io = TAILQ_FIRST(&bdev_ch->qos_queued_io);
    9717           2 :                 TAILQ_REMOVE(&bdev_ch->qos_queued_io, bdev_io, internal.link);
    9718           2 :                 _bdev_io_submit(bdev_io);
    9719             :         }
    9720             : 
    9721           4 :         spdk_bdev_for_each_channel_continue(i, 0);
    9722           4 : }
    9723             : 
    9724             : static void
    9725           1 : bdev_update_qos_rate_limit_msg(void *cb_arg)
    9726             : {
    9727           1 :         struct set_qos_limit_ctx *ctx = cb_arg;
    9728           1 :         struct spdk_bdev *bdev = ctx->bdev;
    9729             : 
    9730           1 :         spdk_spin_lock(&bdev->internal.spinlock);
    9731           1 :         bdev_qos_update_max_quota_per_timeslice(bdev->internal.qos);
    9732           1 :         spdk_spin_unlock(&bdev->internal.spinlock);
    9733             : 
    9734           1 :         bdev_set_qos_limit_done(ctx, 0);
    9735           1 : }
    9736             : 
    9737             : static void
    9738           9 : bdev_enable_qos_msg(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
    9739             :                     struct spdk_io_channel *ch, void *_ctx)
    9740             : {
    9741           9 :         struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(ch);
    9742             : 
    9743           9 :         spdk_spin_lock(&bdev->internal.spinlock);
    9744           9 :         bdev_enable_qos(bdev, bdev_ch);
    9745           9 :         spdk_spin_unlock(&bdev->internal.spinlock);
    9746           9 :         spdk_bdev_for_each_channel_continue(i, 0);
    9747           9 : }
    9748             : 
    9749             : static void
    9750           6 : bdev_enable_qos_done(struct spdk_bdev *bdev, void *_ctx, int status)
    9751             : {
    9752           6 :         struct set_qos_limit_ctx *ctx = _ctx;
    9753             : 
    9754           6 :         bdev_set_qos_limit_done(ctx, status);
    9755           6 : }
    9756             : 
    9757             : static void
    9758           7 : bdev_set_qos_rate_limits(struct spdk_bdev *bdev, uint64_t *limits)
    9759             : {
    9760             :         int i;
    9761             : 
    9762           7 :         assert(bdev->internal.qos != NULL);
    9763             : 
    9764          35 :         for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
    9765          28 :                 if (limits[i] != SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
    9766          28 :                         bdev->internal.qos->rate_limits[i].limit = limits[i];
    9767             : 
    9768          28 :                         if (limits[i] == 0) {
    9769          19 :                                 bdev->internal.qos->rate_limits[i].limit =
    9770             :                                         SPDK_BDEV_QOS_LIMIT_NOT_DEFINED;
    9771          19 :                         }
    9772          28 :                 }
    9773          28 :         }
    9774           7 : }
    9775             : 
    9776             : void
    9777           9 : spdk_bdev_set_qos_rate_limits(struct spdk_bdev *bdev, uint64_t *limits,
    9778             :                               void (*cb_fn)(void *cb_arg, int status), void *cb_arg)
    9779             : {
    9780             :         struct set_qos_limit_ctx        *ctx;
    9781             :         uint32_t                        limit_set_complement;
    9782             :         uint64_t                        min_limit_per_sec;
    9783             :         int                             i;
    9784           9 :         bool                            disable_rate_limit = true;
    9785             : 
    9786          45 :         for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
    9787          36 :                 if (limits[i] == SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
    9788           0 :                         continue;
    9789             :                 }
    9790             : 
    9791          36 :                 if (limits[i] > 0) {
    9792          10 :                         disable_rate_limit = false;
    9793          10 :                 }
    9794             : 
    9795          36 :                 if (bdev_qos_is_iops_rate_limit(i) == true) {
    9796           9 :                         min_limit_per_sec = SPDK_BDEV_QOS_MIN_IOS_PER_SEC;
    9797           9 :                 } else {
    9798          27 :                         if (limits[i] > SPDK_BDEV_QOS_MAX_MBYTES_PER_SEC) {
    9799           0 :                                 SPDK_WARNLOG("Requested rate limit %" PRIu64 " will result in uint64_t overflow, "
    9800             :                                              "reset to %" PRIu64 "\n", limits[i], SPDK_BDEV_QOS_MAX_MBYTES_PER_SEC);
    9801           0 :                                 limits[i] = SPDK_BDEV_QOS_MAX_MBYTES_PER_SEC;
    9802           0 :                         }
    9803             :                         /* Change from megabyte to byte rate limit */
    9804          27 :                         limits[i] = limits[i] * 1024 * 1024;
    9805          27 :                         min_limit_per_sec = SPDK_BDEV_QOS_MIN_BYTES_PER_SEC;
    9806             :                 }
    9807             : 
    9808          36 :                 limit_set_complement = limits[i] % min_limit_per_sec;
    9809          36 :                 if (limit_set_complement) {
    9810           0 :                         SPDK_ERRLOG("Requested rate limit %" PRIu64 " is not a multiple of %" PRIu64 "\n",
    9811             :                                     limits[i], min_limit_per_sec);
    9812           0 :                         limits[i] += min_limit_per_sec - limit_set_complement;
    9813           0 :                         SPDK_ERRLOG("Round up the rate limit to %" PRIu64 "\n", limits[i]);
    9814           0 :                 }
    9815          36 :         }
    9816             : 
    9817           9 :         ctx = calloc(1, sizeof(*ctx));
    9818           9 :         if (ctx == NULL) {
    9819           0 :                 cb_fn(cb_arg, -ENOMEM);
    9820           0 :                 return;
    9821             :         }
    9822             : 
    9823           9 :         ctx->cb_fn = cb_fn;
    9824           9 :         ctx->cb_arg = cb_arg;
    9825           9 :         ctx->bdev = bdev;
    9826             : 
    9827           9 :         spdk_spin_lock(&bdev->internal.spinlock);
    9828           9 :         if (bdev->internal.qos_mod_in_progress) {
    9829           1 :                 spdk_spin_unlock(&bdev->internal.spinlock);
    9830           1 :                 free(ctx);
    9831           1 :                 cb_fn(cb_arg, -EAGAIN);
    9832           1 :                 return;
    9833             :         }
    9834           8 :         bdev->internal.qos_mod_in_progress = true;
    9835             : 
    9836           8 :         if (disable_rate_limit == true && bdev->internal.qos) {
    9837          10 :                 for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
    9838           8 :                         if (limits[i] == SPDK_BDEV_QOS_LIMIT_NOT_DEFINED &&
    9839           0 :                             (bdev->internal.qos->rate_limits[i].limit > 0 &&
    9840           0 :                              bdev->internal.qos->rate_limits[i].limit !=
    9841             :                              SPDK_BDEV_QOS_LIMIT_NOT_DEFINED)) {
    9842           0 :                                 disable_rate_limit = false;
    9843           0 :                                 break;
    9844             :                         }
    9845           8 :                 }
    9846           2 :         }
    9847             : 
    9848           8 :         if (disable_rate_limit == false) {
    9849           5 :                 if (bdev->internal.qos == NULL) {
    9850           4 :                         bdev->internal.qos = calloc(1, sizeof(*bdev->internal.qos));
    9851           4 :                         if (!bdev->internal.qos) {
    9852           0 :                                 spdk_spin_unlock(&bdev->internal.spinlock);
    9853           0 :                                 SPDK_ERRLOG("Unable to allocate memory for QoS tracking\n");
    9854           0 :                                 bdev_set_qos_limit_done(ctx, -ENOMEM);
    9855           0 :                                 return;
    9856             :                         }
    9857           4 :                 }
    9858             : 
    9859           5 :                 if (bdev->internal.qos->thread == NULL) {
    9860             :                         /* Enabling */
    9861           4 :                         bdev_set_qos_rate_limits(bdev, limits);
    9862             : 
    9863           4 :                         spdk_bdev_for_each_channel(bdev, bdev_enable_qos_msg, ctx,
    9864             :                                                    bdev_enable_qos_done);
    9865           4 :                 } else {
    9866             :                         /* Updating */
    9867           1 :                         bdev_set_qos_rate_limits(bdev, limits);
    9868             : 
    9869           2 :                         spdk_thread_send_msg(bdev->internal.qos->thread,
    9870           1 :                                              bdev_update_qos_rate_limit_msg, ctx);
    9871             :                 }
    9872           5 :         } else {
    9873           3 :                 if (bdev->internal.qos != NULL) {
    9874           2 :                         bdev_set_qos_rate_limits(bdev, limits);
    9875             : 
    9876             :                         /* Disabling */
    9877           2 :                         spdk_bdev_for_each_channel(bdev, bdev_disable_qos_msg, ctx,
    9878             :                                                    bdev_disable_qos_msg_done);
    9879           2 :                 } else {
    9880           1 :                         spdk_spin_unlock(&bdev->internal.spinlock);
    9881           1 :                         bdev_set_qos_limit_done(ctx, 0);
    9882           1 :                         return;
    9883             :                 }
    9884             :         }
    9885             : 
    9886           7 :         spdk_spin_unlock(&bdev->internal.spinlock);
    9887           9 : }
    9888             : 
    9889             : struct spdk_bdev_histogram_ctx {
    9890             :         spdk_bdev_histogram_status_cb cb_fn;
    9891             :         void *cb_arg;
    9892             :         struct spdk_bdev *bdev;
    9893             :         int status;
    9894             : };
    9895             : 
    9896             : static void
    9897           2 : bdev_histogram_disable_channel_cb(struct spdk_bdev *bdev, void *_ctx, int status)
    9898             : {
    9899           2 :         struct spdk_bdev_histogram_ctx *ctx = _ctx;
    9900             : 
    9901           2 :         spdk_spin_lock(&ctx->bdev->internal.spinlock);
    9902           2 :         ctx->bdev->internal.histogram_in_progress = false;
    9903           2 :         spdk_spin_unlock(&ctx->bdev->internal.spinlock);
    9904           2 :         ctx->cb_fn(ctx->cb_arg, ctx->status);
    9905           2 :         free(ctx);
    9906           2 : }
    9907             : 
    9908             : static void
    9909           3 : bdev_histogram_disable_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
    9910             :                                struct spdk_io_channel *_ch, void *_ctx)
    9911             : {
    9912           3 :         struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
    9913             : 
    9914           3 :         if (ch->histogram != NULL) {
    9915           3 :                 spdk_histogram_data_free(ch->histogram);
    9916           3 :                 ch->histogram = NULL;
    9917           3 :         }
    9918           3 :         spdk_bdev_for_each_channel_continue(i, 0);
    9919           3 : }
    9920             : 
    9921             : static void
    9922           2 : bdev_histogram_enable_channel_cb(struct spdk_bdev *bdev, void *_ctx, int status)
    9923             : {
    9924           2 :         struct spdk_bdev_histogram_ctx *ctx = _ctx;
    9925             : 
    9926           2 :         if (status != 0) {
    9927           0 :                 ctx->status = status;
    9928           0 :                 ctx->bdev->internal.histogram_enabled = false;
    9929           0 :                 spdk_bdev_for_each_channel(ctx->bdev, bdev_histogram_disable_channel, ctx,
    9930             :                                            bdev_histogram_disable_channel_cb);
    9931           0 :         } else {
    9932           2 :                 spdk_spin_lock(&ctx->bdev->internal.spinlock);
    9933           2 :                 ctx->bdev->internal.histogram_in_progress = false;
    9934           2 :                 spdk_spin_unlock(&ctx->bdev->internal.spinlock);
    9935           2 :                 ctx->cb_fn(ctx->cb_arg, ctx->status);
    9936           2 :                 free(ctx);
    9937             :         }
    9938           2 : }
    9939             : 
    9940             : static void
    9941           3 : bdev_histogram_enable_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
    9942             :                               struct spdk_io_channel *_ch, void *_ctx)
    9943             : {
    9944           3 :         struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
    9945           3 :         int status = 0;
    9946             : 
    9947           3 :         if (ch->histogram == NULL) {
    9948           3 :                 ch->histogram = spdk_histogram_data_alloc();
    9949           3 :                 if (ch->histogram == NULL) {
    9950           0 :                         status = -ENOMEM;
    9951           0 :                 }
    9952           3 :         }
    9953             : 
    9954           3 :         spdk_bdev_for_each_channel_continue(i, status);
    9955           3 : }
    9956             : 
    9957             : void
    9958           4 : spdk_bdev_histogram_enable_ext(struct spdk_bdev *bdev, spdk_bdev_histogram_status_cb cb_fn,
    9959             :                                void *cb_arg, bool enable, struct spdk_bdev_enable_histogram_opts *opts)
    9960             : {
    9961             :         struct spdk_bdev_histogram_ctx *ctx;
    9962             : 
    9963           4 :         ctx = calloc(1, sizeof(struct spdk_bdev_histogram_ctx));
    9964           4 :         if (ctx == NULL) {
    9965           0 :                 cb_fn(cb_arg, -ENOMEM);
    9966           0 :                 return;
    9967             :         }
    9968             : 
    9969           4 :         ctx->bdev = bdev;
    9970           4 :         ctx->status = 0;
    9971           4 :         ctx->cb_fn = cb_fn;
    9972           4 :         ctx->cb_arg = cb_arg;
    9973             : 
    9974           4 :         spdk_spin_lock(&bdev->internal.spinlock);
    9975           4 :         if (bdev->internal.histogram_in_progress) {
    9976           0 :                 spdk_spin_unlock(&bdev->internal.spinlock);
    9977           0 :                 free(ctx);
    9978           0 :                 cb_fn(cb_arg, -EAGAIN);
    9979           0 :                 return;
    9980             :         }
    9981             : 
    9982           4 :         bdev->internal.histogram_in_progress = true;
    9983           4 :         spdk_spin_unlock(&bdev->internal.spinlock);
    9984             : 
    9985           4 :         bdev->internal.histogram_enabled = enable;
    9986           4 :         bdev->internal.histogram_io_type = opts->io_type;
    9987             : 
    9988           4 :         if (enable) {
    9989             :                 /* Allocate histogram for each channel */
    9990           2 :                 spdk_bdev_for_each_channel(bdev, bdev_histogram_enable_channel, ctx,
    9991             :                                            bdev_histogram_enable_channel_cb);
    9992           2 :         } else {
    9993           2 :                 spdk_bdev_for_each_channel(bdev, bdev_histogram_disable_channel, ctx,
    9994             :                                            bdev_histogram_disable_channel_cb);
    9995             :         }
    9996           4 : }
    9997             : 
    9998             : void
    9999           4 : spdk_bdev_enable_histogram_opts_init(struct spdk_bdev_enable_histogram_opts *opts, size_t size)
   10000             : {
   10001           4 :         if (opts == NULL) {
   10002           0 :                 SPDK_ERRLOG("opts should not be NULL\n");
   10003           0 :                 assert(opts != NULL);
   10004           0 :                 return;
   10005             :         }
   10006           4 :         if (size == 0) {
   10007           0 :                 SPDK_ERRLOG("size should not be zero\n");
   10008           0 :                 assert(size != 0);
   10009           0 :                 return;
   10010             :         }
   10011             : 
   10012           4 :         memset(opts, 0, size);
   10013           4 :         opts->size = size;
   10014             : 
   10015             : #define FIELD_OK(field) \
   10016             :         offsetof(struct spdk_bdev_enable_histogram_opts, field) + sizeof(opts->field) <= size
   10017             : 
   10018             : #define SET_FIELD(field, value) \
   10019             :         if (FIELD_OK(field)) { \
   10020             :                 opts->field = value; \
   10021             :         } \
   10022             : 
   10023           4 :         SET_FIELD(io_type, 0);
   10024             : 
   10025             :         /* You should not remove this statement, but need to update the assert statement
   10026             :          * if you add a new field, and also add a corresponding SET_FIELD statement */
   10027             :         SPDK_STATIC_ASSERT(sizeof(struct spdk_bdev_enable_histogram_opts) == 9, "Incorrect size");
   10028             : 
   10029             : #undef FIELD_OK
   10030             : #undef SET_FIELD
   10031           4 : }
   10032             : 
   10033             : void
   10034           4 : spdk_bdev_histogram_enable(struct spdk_bdev *bdev, spdk_bdev_histogram_status_cb cb_fn,
   10035             :                            void *cb_arg, bool enable)
   10036             : {
   10037             :         struct spdk_bdev_enable_histogram_opts opts;
   10038             : 
   10039           4 :         spdk_bdev_enable_histogram_opts_init(&opts, sizeof(opts));
   10040           4 :         spdk_bdev_histogram_enable_ext(bdev, cb_fn, cb_arg, enable, &opts);
   10041           4 : }
   10042             : 
   10043             : struct spdk_bdev_histogram_data_ctx {
   10044             :         spdk_bdev_histogram_data_cb cb_fn;
   10045             :         void *cb_arg;
   10046             :         struct spdk_bdev *bdev;
   10047             :         /** merged histogram data from all channels */
   10048             :         struct spdk_histogram_data      *histogram;
   10049             : };
   10050             : 
   10051             : static void
   10052           5 : bdev_histogram_get_channel_cb(struct spdk_bdev *bdev, void *_ctx, int status)
   10053             : {
   10054           5 :         struct spdk_bdev_histogram_data_ctx *ctx = _ctx;
   10055             : 
   10056           5 :         ctx->cb_fn(ctx->cb_arg, status, ctx->histogram);
   10057           5 :         free(ctx);
   10058           5 : }
   10059             : 
   10060             : static void
   10061           7 : bdev_histogram_get_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
   10062             :                            struct spdk_io_channel *_ch, void *_ctx)
   10063             : {
   10064           7 :         struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
   10065           7 :         struct spdk_bdev_histogram_data_ctx *ctx = _ctx;
   10066           7 :         int status = 0;
   10067             : 
   10068           7 :         if (ch->histogram == NULL) {
   10069           1 :                 status = -EFAULT;
   10070           1 :         } else {
   10071           6 :                 spdk_histogram_data_merge(ctx->histogram, ch->histogram);
   10072             :         }
   10073             : 
   10074           7 :         spdk_bdev_for_each_channel_continue(i, status);
   10075           7 : }
   10076             : 
   10077             : void
   10078           5 : spdk_bdev_histogram_get(struct spdk_bdev *bdev, struct spdk_histogram_data *histogram,
   10079             :                         spdk_bdev_histogram_data_cb cb_fn,
   10080             :                         void *cb_arg)
   10081             : {
   10082             :         struct spdk_bdev_histogram_data_ctx *ctx;
   10083             : 
   10084           5 :         ctx = calloc(1, sizeof(struct spdk_bdev_histogram_data_ctx));
   10085           5 :         if (ctx == NULL) {
   10086           0 :                 cb_fn(cb_arg, -ENOMEM, NULL);
   10087           0 :                 return;
   10088             :         }
   10089             : 
   10090           5 :         ctx->bdev = bdev;
   10091           5 :         ctx->cb_fn = cb_fn;
   10092           5 :         ctx->cb_arg = cb_arg;
   10093             : 
   10094           5 :         ctx->histogram = histogram;
   10095             : 
   10096           5 :         spdk_bdev_for_each_channel(bdev, bdev_histogram_get_channel, ctx,
   10097             :                                    bdev_histogram_get_channel_cb);
   10098           5 : }
   10099             : 
   10100             : void
   10101           2 : spdk_bdev_channel_get_histogram(struct spdk_io_channel *ch, spdk_bdev_histogram_data_cb cb_fn,
   10102             :                                 void *cb_arg)
   10103             : {
   10104           2 :         struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(ch);
   10105           2 :         int status = 0;
   10106             : 
   10107           2 :         assert(cb_fn != NULL);
   10108             : 
   10109           2 :         if (bdev_ch->histogram == NULL) {
   10110           1 :                 status = -EFAULT;
   10111           1 :         }
   10112           2 :         cb_fn(cb_arg, status, bdev_ch->histogram);
   10113           2 : }
   10114             : 
   10115             : size_t
   10116           0 : spdk_bdev_get_media_events(struct spdk_bdev_desc *desc, struct spdk_bdev_media_event *events,
   10117             :                            size_t max_events)
   10118             : {
   10119             :         struct media_event_entry *entry;
   10120           0 :         size_t num_events = 0;
   10121             : 
   10122           0 :         for (; num_events < max_events; ++num_events) {
   10123           0 :                 entry = TAILQ_FIRST(&desc->pending_media_events);
   10124           0 :                 if (entry == NULL) {
   10125           0 :                         break;
   10126             :                 }
   10127             : 
   10128           0 :                 events[num_events] = entry->event;
   10129           0 :                 TAILQ_REMOVE(&desc->pending_media_events, entry, tailq);
   10130           0 :                 TAILQ_INSERT_TAIL(&desc->free_media_events, entry, tailq);
   10131           0 :         }
   10132             : 
   10133           0 :         return num_events;
   10134             : }
   10135             : 
   10136             : int
   10137           0 : spdk_bdev_push_media_events(struct spdk_bdev *bdev, const struct spdk_bdev_media_event *events,
   10138             :                             size_t num_events)
   10139             : {
   10140             :         struct spdk_bdev_desc *desc;
   10141             :         struct media_event_entry *entry;
   10142             :         size_t event_id;
   10143           0 :         int rc = 0;
   10144             : 
   10145           0 :         assert(bdev->media_events);
   10146             : 
   10147           0 :         spdk_spin_lock(&bdev->internal.spinlock);
   10148           0 :         TAILQ_FOREACH(desc, &bdev->internal.open_descs, link) {
   10149           0 :                 if (desc->write) {
   10150           0 :                         break;
   10151             :                 }
   10152           0 :         }
   10153             : 
   10154           0 :         if (desc == NULL || desc->media_events_buffer == NULL) {
   10155           0 :                 rc = -ENODEV;
   10156           0 :                 goto out;
   10157             :         }
   10158             : 
   10159           0 :         for (event_id = 0; event_id < num_events; ++event_id) {
   10160           0 :                 entry = TAILQ_FIRST(&desc->free_media_events);
   10161           0 :                 if (entry == NULL) {
   10162           0 :                         break;
   10163             :                 }
   10164             : 
   10165           0 :                 TAILQ_REMOVE(&desc->free_media_events, entry, tailq);
   10166           0 :                 TAILQ_INSERT_TAIL(&desc->pending_media_events, entry, tailq);
   10167           0 :                 entry->event = events[event_id];
   10168           0 :         }
   10169             : 
   10170           0 :         rc = event_id;
   10171             : out:
   10172           0 :         spdk_spin_unlock(&bdev->internal.spinlock);
   10173           0 :         return rc;
   10174             : }
   10175             : 
   10176             : static void
   10177           0 : _media_management_notify(void *arg)
   10178             : {
   10179           0 :         struct spdk_bdev_desc *desc = arg;
   10180             : 
   10181           0 :         _event_notify(desc, SPDK_BDEV_EVENT_MEDIA_MANAGEMENT);
   10182           0 : }
   10183             : 
   10184             : void
   10185           0 : spdk_bdev_notify_media_management(struct spdk_bdev *bdev)
   10186             : {
   10187             :         struct spdk_bdev_desc *desc;
   10188             : 
   10189           0 :         spdk_spin_lock(&bdev->internal.spinlock);
   10190           0 :         TAILQ_FOREACH(desc, &bdev->internal.open_descs, link) {
   10191           0 :                 if (!TAILQ_EMPTY(&desc->pending_media_events)) {
   10192           0 :                         event_notify(desc, _media_management_notify);
   10193           0 :                 }
   10194           0 :         }
   10195           0 :         spdk_spin_unlock(&bdev->internal.spinlock);
   10196           0 : }
   10197             : 
   10198             : struct locked_lba_range_ctx {
   10199             :         struct lba_range                range;
   10200             :         struct lba_range                *current_range;
   10201             :         struct lba_range                *owner_range;
   10202             :         struct spdk_poller              *poller;
   10203             :         lock_range_cb                   cb_fn;
   10204             :         void                            *cb_arg;
   10205             : };
   10206             : 
   10207             : static void
   10208           0 : bdev_lock_error_cleanup_cb(struct spdk_bdev *bdev, void *_ctx, int status)
   10209             : {
   10210           0 :         struct locked_lba_range_ctx *ctx = _ctx;
   10211             : 
   10212           0 :         ctx->cb_fn(&ctx->range, ctx->cb_arg, -ENOMEM);
   10213           0 :         free(ctx);
   10214           0 : }
   10215             : 
   10216             : static void bdev_unlock_lba_range_get_channel(struct spdk_bdev_channel_iter *i,
   10217             :                 struct spdk_bdev *bdev, struct spdk_io_channel *ch, void *_ctx);
   10218             : 
   10219             : static void
   10220          14 : bdev_lock_lba_range_cb(struct spdk_bdev *bdev, void *_ctx, int status)
   10221             : {
   10222          14 :         struct locked_lba_range_ctx *ctx = _ctx;
   10223             : 
   10224          14 :         if (status == -ENOMEM) {
   10225             :                 /* One of the channels could not allocate a range object.
   10226             :                  * So we have to go back and clean up any ranges that were
   10227             :                  * allocated successfully before we return error status to
   10228             :                  * the caller.  We can reuse the unlock function to do that
   10229             :                  * clean up.
   10230             :                  */
   10231           0 :                 spdk_bdev_for_each_channel(bdev, bdev_unlock_lba_range_get_channel, ctx,
   10232             :                                            bdev_lock_error_cleanup_cb);
   10233           0 :                 return;
   10234             :         }
   10235             : 
   10236             :         /* All channels have locked this range and no I/O overlapping the range
   10237             :          * are outstanding!  Set the owner_ch for the range object for the
   10238             :          * locking channel, so that this channel will know that it is allowed
   10239             :          * to write to this range.
   10240             :          */
   10241          14 :         if (ctx->owner_range != NULL) {
   10242          10 :                 ctx->owner_range->owner_ch = ctx->range.owner_ch;
   10243          10 :         }
   10244             : 
   10245          14 :         ctx->cb_fn(&ctx->range, ctx->cb_arg, status);
   10246             : 
   10247             :         /* Don't free the ctx here.  Its range is in the bdev's global list of
   10248             :          * locked ranges still, and will be removed and freed when this range
   10249             :          * is later unlocked.
   10250             :          */
   10251          14 : }
   10252             : 
   10253             : static int
   10254          17 : bdev_lock_lba_range_check_io(void *_i)
   10255             : {
   10256          17 :         struct spdk_bdev_channel_iter *i = _i;
   10257          17 :         struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i->i);
   10258          17 :         struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
   10259          17 :         struct locked_lba_range_ctx *ctx = i->ctx;
   10260          17 :         struct lba_range *range = ctx->current_range;
   10261             :         struct spdk_bdev_io *bdev_io;
   10262             : 
   10263          17 :         spdk_poller_unregister(&ctx->poller);
   10264             : 
   10265             :         /* The range is now in the locked_ranges, so no new IO can be submitted to this
   10266             :          * range.  But we need to wait until any outstanding IO overlapping with this range
   10267             :          * are completed.
   10268             :          */
   10269          18 :         TAILQ_FOREACH(bdev_io, &ch->io_submitted, internal.ch_link) {
   10270           3 :                 if (bdev_io_range_is_locked(bdev_io, range)) {
   10271           2 :                         ctx->poller = SPDK_POLLER_REGISTER(bdev_lock_lba_range_check_io, i, 100);
   10272           2 :                         return SPDK_POLLER_BUSY;
   10273             :                 }
   10274           1 :         }
   10275             : 
   10276          15 :         spdk_bdev_for_each_channel_continue(i, 0);
   10277          15 :         return SPDK_POLLER_BUSY;
   10278          17 : }
   10279             : 
   10280             : static void
   10281          15 : bdev_lock_lba_range_get_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
   10282             :                                 struct spdk_io_channel *_ch, void *_ctx)
   10283             : {
   10284          15 :         struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
   10285          15 :         struct locked_lba_range_ctx *ctx = _ctx;
   10286             :         struct lba_range *range;
   10287             : 
   10288          16 :         TAILQ_FOREACH(range, &ch->locked_ranges, tailq) {
   10289           1 :                 if (range->length == ctx->range.length &&
   10290           0 :                     range->offset == ctx->range.offset &&
   10291           0 :                     range->locked_ctx == ctx->range.locked_ctx) {
   10292             :                         /* This range already exists on this channel, so don't add
   10293             :                          * it again.  This can happen when a new channel is created
   10294             :                          * while the for_each_channel operation is in progress.
   10295             :                          * Do not check for outstanding I/O in that case, since the
   10296             :                          * range was locked before any I/O could be submitted to the
   10297             :                          * new channel.
   10298             :                          */
   10299           0 :                         spdk_bdev_for_each_channel_continue(i, 0);
   10300           0 :                         return;
   10301             :                 }
   10302           1 :         }
   10303             : 
   10304          15 :         range = calloc(1, sizeof(*range));
   10305          15 :         if (range == NULL) {
   10306           0 :                 spdk_bdev_for_each_channel_continue(i, -ENOMEM);
   10307           0 :                 return;
   10308             :         }
   10309             : 
   10310          15 :         range->length = ctx->range.length;
   10311          15 :         range->offset = ctx->range.offset;
   10312          15 :         range->locked_ctx = ctx->range.locked_ctx;
   10313          15 :         range->quiesce = ctx->range.quiesce;
   10314          15 :         ctx->current_range = range;
   10315          15 :         if (ctx->range.owner_ch == ch) {
   10316             :                 /* This is the range object for the channel that will hold
   10317             :                  * the lock.  Store it in the ctx object so that we can easily
   10318             :                  * set its owner_ch after the lock is finally acquired.
   10319             :                  */
   10320          10 :                 ctx->owner_range = range;
   10321          10 :         }
   10322          15 :         TAILQ_INSERT_TAIL(&ch->locked_ranges, range, tailq);
   10323          15 :         bdev_lock_lba_range_check_io(i);
   10324          15 : }
   10325             : 
   10326             : static void
   10327          14 : bdev_lock_lba_range_ctx(struct spdk_bdev *bdev, struct locked_lba_range_ctx *ctx)
   10328             : {
   10329          14 :         assert(spdk_get_thread() == ctx->range.owner_thread);
   10330          14 :         assert(ctx->range.owner_ch == NULL ||
   10331             :                spdk_io_channel_get_thread(ctx->range.owner_ch->channel) == ctx->range.owner_thread);
   10332             : 
   10333             :         /* We will add a copy of this range to each channel now. */
   10334          14 :         spdk_bdev_for_each_channel(bdev, bdev_lock_lba_range_get_channel, ctx,
   10335             :                                    bdev_lock_lba_range_cb);
   10336          14 : }
   10337             : 
   10338             : static bool
   10339          17 : bdev_lba_range_overlaps_tailq(struct lba_range *range, lba_range_tailq_t *tailq)
   10340             : {
   10341             :         struct lba_range *r;
   10342             : 
   10343          18 :         TAILQ_FOREACH(r, tailq, tailq) {
   10344           4 :                 if (bdev_lba_range_overlapped(range, r)) {
   10345           3 :                         return true;
   10346             :                 }
   10347           1 :         }
   10348          14 :         return false;
   10349          17 : }
   10350             : 
   10351             : static void bdev_quiesce_range_locked(struct lba_range *range, void *ctx, int status);
   10352             : 
   10353             : static int
   10354          14 : _bdev_lock_lba_range(struct spdk_bdev *bdev, struct spdk_bdev_channel *ch,
   10355             :                      uint64_t offset, uint64_t length,
   10356             :                      lock_range_cb cb_fn, void *cb_arg)
   10357             : {
   10358             :         struct locked_lba_range_ctx *ctx;
   10359             : 
   10360          14 :         ctx = calloc(1, sizeof(*ctx));
   10361          14 :         if (ctx == NULL) {
   10362           0 :                 return -ENOMEM;
   10363             :         }
   10364             : 
   10365          14 :         ctx->range.offset = offset;
   10366          14 :         ctx->range.length = length;
   10367          14 :         ctx->range.owner_thread = spdk_get_thread();
   10368          14 :         ctx->range.owner_ch = ch;
   10369          14 :         ctx->range.locked_ctx = cb_arg;
   10370          14 :         ctx->range.bdev = bdev;
   10371          14 :         ctx->range.quiesce = (cb_fn == bdev_quiesce_range_locked);
   10372          14 :         ctx->cb_fn = cb_fn;
   10373          14 :         ctx->cb_arg = cb_arg;
   10374             : 
   10375          14 :         spdk_spin_lock(&bdev->internal.spinlock);
   10376          14 :         if (bdev_lba_range_overlaps_tailq(&ctx->range, &bdev->internal.locked_ranges)) {
   10377             :                 /* There is an active lock overlapping with this range.
   10378             :                  * Put it on the pending list until this range no
   10379             :                  * longer overlaps with another.
   10380             :                  */
   10381           2 :                 TAILQ_INSERT_TAIL(&bdev->internal.pending_locked_ranges, &ctx->range, tailq);
   10382           2 :         } else {
   10383          12 :                 TAILQ_INSERT_TAIL(&bdev->internal.locked_ranges, &ctx->range, tailq);
   10384          12 :                 bdev_lock_lba_range_ctx(bdev, ctx);
   10385             :         }
   10386          14 :         spdk_spin_unlock(&bdev->internal.spinlock);
   10387          14 :         return 0;
   10388          14 : }
   10389             : 
   10390             : static int
   10391          10 : bdev_lock_lba_range(struct spdk_bdev_desc *desc, struct spdk_io_channel *_ch,
   10392             :                     uint64_t offset, uint64_t length,
   10393             :                     lock_range_cb cb_fn, void *cb_arg)
   10394             : {
   10395          10 :         struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
   10396          10 :         struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
   10397             : 
   10398          10 :         if (cb_arg == NULL) {
   10399           0 :                 SPDK_ERRLOG("cb_arg must not be NULL\n");
   10400           0 :                 return -EINVAL;
   10401             :         }
   10402             : 
   10403          10 :         return _bdev_lock_lba_range(bdev, ch, offset, length, cb_fn, cb_arg);
   10404          10 : }
   10405             : 
   10406             : static void
   10407           2 : bdev_lock_lba_range_ctx_msg(void *_ctx)
   10408             : {
   10409           2 :         struct locked_lba_range_ctx *ctx = _ctx;
   10410             : 
   10411           2 :         bdev_lock_lba_range_ctx(ctx->range.bdev, ctx);
   10412           2 : }
   10413             : 
   10414             : static void
   10415          14 : bdev_unlock_lba_range_cb(struct spdk_bdev *bdev, void *_ctx, int status)
   10416             : {
   10417          14 :         struct locked_lba_range_ctx *ctx = _ctx;
   10418             :         struct locked_lba_range_ctx *pending_ctx;
   10419             :         struct lba_range *range, *tmp;
   10420             : 
   10421          14 :         spdk_spin_lock(&bdev->internal.spinlock);
   10422             :         /* Check if there are any pending locked ranges that overlap with this range
   10423             :          * that was just unlocked.  If there are, check that it doesn't overlap with any
   10424             :          * other locked ranges before calling bdev_lock_lba_range_ctx which will start
   10425             :          * the lock process.
   10426             :          */
   10427          17 :         TAILQ_FOREACH_SAFE(range, &bdev->internal.pending_locked_ranges, tailq, tmp) {
   10428           3 :                 if (bdev_lba_range_overlapped(range, &ctx->range) &&
   10429           3 :                     !bdev_lba_range_overlaps_tailq(range, &bdev->internal.locked_ranges)) {
   10430           2 :                         TAILQ_REMOVE(&bdev->internal.pending_locked_ranges, range, tailq);
   10431           2 :                         pending_ctx = SPDK_CONTAINEROF(range, struct locked_lba_range_ctx, range);
   10432           2 :                         TAILQ_INSERT_TAIL(&bdev->internal.locked_ranges, range, tailq);
   10433           4 :                         spdk_thread_send_msg(pending_ctx->range.owner_thread,
   10434           2 :                                              bdev_lock_lba_range_ctx_msg, pending_ctx);
   10435           2 :                 }
   10436           3 :         }
   10437          14 :         spdk_spin_unlock(&bdev->internal.spinlock);
   10438             : 
   10439          14 :         ctx->cb_fn(&ctx->range, ctx->cb_arg, status);
   10440          14 :         free(ctx);
   10441          14 : }
   10442             : 
   10443             : static void
   10444          16 : bdev_unlock_lba_range_get_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
   10445             :                                   struct spdk_io_channel *_ch, void *_ctx)
   10446             : {
   10447          16 :         struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
   10448          16 :         struct locked_lba_range_ctx *ctx = _ctx;
   10449             :         TAILQ_HEAD(, spdk_bdev_io) io_locked;
   10450             :         struct spdk_bdev_io *bdev_io;
   10451             :         struct lba_range *range;
   10452             : 
   10453          16 :         TAILQ_FOREACH(range, &ch->locked_ranges, tailq) {
   10454          32 :                 if (ctx->range.offset == range->offset &&
   10455          16 :                     ctx->range.length == range->length &&
   10456          16 :                     ctx->range.locked_ctx == range->locked_ctx) {
   10457          16 :                         TAILQ_REMOVE(&ch->locked_ranges, range, tailq);
   10458          16 :                         free(range);
   10459          16 :                         break;
   10460             :                 }
   10461           0 :         }
   10462             : 
   10463             :         /* Note: we should almost always be able to assert that the range specified
   10464             :          * was found.  But there are some very rare corner cases where a new channel
   10465             :          * gets created simultaneously with a range unlock, where this function
   10466             :          * would execute on that new channel and wouldn't have the range.
   10467             :          * We also use this to clean up range allocations when a later allocation
   10468             :          * fails in the locking path.
   10469             :          * So we can't actually assert() here.
   10470             :          */
   10471             : 
   10472             :         /* Swap the locked IO into a temporary list, and then try to submit them again.
   10473             :          * We could hyper-optimize this to only resubmit locked I/O that overlap
   10474             :          * with the range that was just unlocked, but this isn't a performance path so
   10475             :          * we go for simplicity here.
   10476             :          */
   10477          16 :         TAILQ_INIT(&io_locked);
   10478          16 :         TAILQ_SWAP(&ch->io_locked, &io_locked, spdk_bdev_io, internal.ch_link);
   10479          19 :         while (!TAILQ_EMPTY(&io_locked)) {
   10480           3 :                 bdev_io = TAILQ_FIRST(&io_locked);
   10481           3 :                 TAILQ_REMOVE(&io_locked, bdev_io, internal.ch_link);
   10482           3 :                 bdev_io_submit(bdev_io);
   10483             :         }
   10484             : 
   10485          16 :         spdk_bdev_for_each_channel_continue(i, 0);
   10486          16 : }
   10487             : 
   10488             : static int
   10489          14 : _bdev_unlock_lba_range(struct spdk_bdev *bdev, uint64_t offset, uint64_t length,
   10490             :                        lock_range_cb cb_fn, void *cb_arg)
   10491             : {
   10492             :         struct locked_lba_range_ctx *ctx;
   10493             :         struct lba_range *range;
   10494             : 
   10495          14 :         spdk_spin_lock(&bdev->internal.spinlock);
   10496             :         /* To start the unlock the process, we find the range in the bdev's locked_ranges
   10497             :          * and remove it. This ensures new channels don't inherit the locked range.
   10498             :          * Then we will send a message to each channel to remove the range from its
   10499             :          * per-channel list.
   10500             :          */
   10501          14 :         TAILQ_FOREACH(range, &bdev->internal.locked_ranges, tailq) {
   10502          24 :                 if (range->offset == offset && range->length == length &&
   10503          14 :                     (range->owner_ch == NULL || range->locked_ctx == cb_arg)) {
   10504          14 :                         break;
   10505             :                 }
   10506           0 :         }
   10507          14 :         if (range == NULL) {
   10508           0 :                 assert(false);
   10509             :                 spdk_spin_unlock(&bdev->internal.spinlock);
   10510             :                 return -EINVAL;
   10511             :         }
   10512          14 :         TAILQ_REMOVE(&bdev->internal.locked_ranges, range, tailq);
   10513          14 :         ctx = SPDK_CONTAINEROF(range, struct locked_lba_range_ctx, range);
   10514          14 :         spdk_spin_unlock(&bdev->internal.spinlock);
   10515             : 
   10516          14 :         ctx->cb_fn = cb_fn;
   10517          14 :         ctx->cb_arg = cb_arg;
   10518             : 
   10519          14 :         spdk_bdev_for_each_channel(bdev, bdev_unlock_lba_range_get_channel, ctx,
   10520             :                                    bdev_unlock_lba_range_cb);
   10521          14 :         return 0;
   10522             : }
   10523             : 
   10524             : static int
   10525          12 : bdev_unlock_lba_range(struct spdk_bdev_desc *desc, struct spdk_io_channel *_ch,
   10526             :                       uint64_t offset, uint64_t length,
   10527             :                       lock_range_cb cb_fn, void *cb_arg)
   10528             : {
   10529          12 :         struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
   10530          12 :         struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
   10531             :         struct lba_range *range;
   10532          12 :         bool range_found = false;
   10533             : 
   10534             :         /* Let's make sure the specified channel actually has a lock on
   10535             :          * the specified range.  Note that the range must match exactly.
   10536             :          */
   10537          14 :         TAILQ_FOREACH(range, &ch->locked_ranges, tailq) {
   10538          22 :                 if (range->offset == offset && range->length == length &&
   10539          11 :                     range->owner_ch == ch && range->locked_ctx == cb_arg) {
   10540          10 :                         range_found = true;
   10541          10 :                         break;
   10542             :                 }
   10543           2 :         }
   10544             : 
   10545          12 :         if (!range_found) {
   10546           2 :                 return -EINVAL;
   10547             :         }
   10548             : 
   10549          10 :         return _bdev_unlock_lba_range(bdev, offset, length, cb_fn, cb_arg);
   10550          12 : }
   10551             : 
   10552             : struct bdev_quiesce_ctx {
   10553             :         spdk_bdev_quiesce_cb cb_fn;
   10554             :         void *cb_arg;
   10555             : };
   10556             : 
   10557             : static void
   10558           4 : bdev_unquiesce_range_unlocked(struct lba_range *range, void *ctx, int status)
   10559             : {
   10560           4 :         struct bdev_quiesce_ctx *quiesce_ctx = ctx;
   10561             : 
   10562           4 :         if (quiesce_ctx->cb_fn != NULL) {
   10563           4 :                 quiesce_ctx->cb_fn(quiesce_ctx->cb_arg, status);
   10564           4 :         }
   10565             : 
   10566           4 :         free(quiesce_ctx);
   10567           4 : }
   10568             : 
   10569             : static void
   10570           4 : bdev_quiesce_range_locked(struct lba_range *range, void *ctx, int status)
   10571             : {
   10572           4 :         struct bdev_quiesce_ctx *quiesce_ctx = ctx;
   10573           4 :         struct spdk_bdev_module *module = range->bdev->module;
   10574             : 
   10575           4 :         if (status != 0) {
   10576           0 :                 if (quiesce_ctx->cb_fn != NULL) {
   10577           0 :                         quiesce_ctx->cb_fn(quiesce_ctx->cb_arg, status);
   10578           0 :                 }
   10579           0 :                 free(quiesce_ctx);
   10580           0 :                 return;
   10581             :         }
   10582             : 
   10583           4 :         spdk_spin_lock(&module->internal.spinlock);
   10584           4 :         TAILQ_INSERT_TAIL(&module->internal.quiesced_ranges, range, tailq_module);
   10585           4 :         spdk_spin_unlock(&module->internal.spinlock);
   10586             : 
   10587           4 :         if (quiesce_ctx->cb_fn != NULL) {
   10588             :                 /* copy the context in case the range is unlocked by the callback */
   10589           4 :                 struct bdev_quiesce_ctx tmp = *quiesce_ctx;
   10590             : 
   10591           4 :                 quiesce_ctx->cb_fn = NULL;
   10592           4 :                 quiesce_ctx->cb_arg = NULL;
   10593             : 
   10594           4 :                 tmp.cb_fn(tmp.cb_arg, status);
   10595           4 :         }
   10596             :         /* quiesce_ctx will be freed on unquiesce */
   10597           4 : }
   10598             : 
   10599             : static int
   10600           9 : _spdk_bdev_quiesce(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
   10601             :                    uint64_t offset, uint64_t length,
   10602             :                    spdk_bdev_quiesce_cb cb_fn, void *cb_arg,
   10603             :                    bool unquiesce)
   10604             : {
   10605             :         struct bdev_quiesce_ctx *quiesce_ctx;
   10606             :         int rc;
   10607             : 
   10608           9 :         if (module != bdev->module) {
   10609           0 :                 SPDK_ERRLOG("Bdev does not belong to specified module.\n");
   10610           0 :                 return -EINVAL;
   10611             :         }
   10612             : 
   10613           9 :         if (!bdev_io_valid_blocks(bdev, offset, length)) {
   10614           0 :                 return -EINVAL;
   10615             :         }
   10616             : 
   10617           9 :         if (unquiesce) {
   10618             :                 struct lba_range *range;
   10619             : 
   10620             :                 /* Make sure the specified range is actually quiesced in the specified module and
   10621             :                  * then remove it from the list. Note that the range must match exactly.
   10622             :                  */
   10623           5 :                 spdk_spin_lock(&module->internal.spinlock);
   10624           6 :                 TAILQ_FOREACH(range, &module->internal.quiesced_ranges, tailq_module) {
   10625           5 :                         if (range->bdev == bdev && range->offset == offset && range->length == length) {
   10626           4 :                                 TAILQ_REMOVE(&module->internal.quiesced_ranges, range, tailq_module);
   10627           4 :                                 break;
   10628             :                         }
   10629           1 :                 }
   10630           5 :                 spdk_spin_unlock(&module->internal.spinlock);
   10631             : 
   10632           5 :                 if (range == NULL) {
   10633           1 :                         SPDK_ERRLOG("The range to unquiesce was not found.\n");
   10634           1 :                         return -EINVAL;
   10635             :                 }
   10636             : 
   10637           4 :                 quiesce_ctx = range->locked_ctx;
   10638           4 :                 quiesce_ctx->cb_fn = cb_fn;
   10639           4 :                 quiesce_ctx->cb_arg = cb_arg;
   10640             : 
   10641           4 :                 rc = _bdev_unlock_lba_range(bdev, offset, length, bdev_unquiesce_range_unlocked, quiesce_ctx);
   10642           4 :         } else {
   10643           4 :                 quiesce_ctx = malloc(sizeof(*quiesce_ctx));
   10644           4 :                 if (quiesce_ctx == NULL) {
   10645           0 :                         return -ENOMEM;
   10646             :                 }
   10647             : 
   10648           4 :                 quiesce_ctx->cb_fn = cb_fn;
   10649           4 :                 quiesce_ctx->cb_arg = cb_arg;
   10650             : 
   10651           4 :                 rc = _bdev_lock_lba_range(bdev, NULL, offset, length, bdev_quiesce_range_locked, quiesce_ctx);
   10652           4 :                 if (rc != 0) {
   10653           0 :                         free(quiesce_ctx);
   10654           0 :                 }
   10655             :         }
   10656             : 
   10657           8 :         return rc;
   10658           9 : }
   10659             : 
   10660             : int
   10661           3 : spdk_bdev_quiesce(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
   10662             :                   spdk_bdev_quiesce_cb cb_fn, void *cb_arg)
   10663             : {
   10664           3 :         return _spdk_bdev_quiesce(bdev, module, 0, bdev->blockcnt, cb_fn, cb_arg, false);
   10665             : }
   10666             : 
   10667             : int
   10668           3 : spdk_bdev_unquiesce(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
   10669             :                     spdk_bdev_quiesce_cb cb_fn, void *cb_arg)
   10670             : {
   10671           3 :         return _spdk_bdev_quiesce(bdev, module, 0, bdev->blockcnt, cb_fn, cb_arg, true);
   10672             : }
   10673             : 
   10674             : int
   10675           1 : spdk_bdev_quiesce_range(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
   10676             :                         uint64_t offset, uint64_t length,
   10677             :                         spdk_bdev_quiesce_cb cb_fn, void *cb_arg)
   10678             : {
   10679           1 :         return _spdk_bdev_quiesce(bdev, module, offset, length, cb_fn, cb_arg, false);
   10680             : }
   10681             : 
   10682             : int
   10683           2 : spdk_bdev_unquiesce_range(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
   10684             :                           uint64_t offset, uint64_t length,
   10685             :                           spdk_bdev_quiesce_cb cb_fn, void *cb_arg)
   10686             : {
   10687           2 :         return _spdk_bdev_quiesce(bdev, module, offset, length, cb_fn, cb_arg, true);
   10688             : }
   10689             : 
   10690             : int
   10691         273 : spdk_bdev_get_memory_domains(struct spdk_bdev *bdev, struct spdk_memory_domain **domains,
   10692             :                              int array_size)
   10693             : {
   10694         273 :         if (!bdev) {
   10695           1 :                 return -EINVAL;
   10696             :         }
   10697             : 
   10698         272 :         if (bdev->fn_table->get_memory_domains) {
   10699           3 :                 return bdev->fn_table->get_memory_domains(bdev->ctxt, domains, array_size);
   10700             :         }
   10701             : 
   10702         269 :         return 0;
   10703         273 : }
   10704             : 
   10705             : struct spdk_bdev_for_each_io_ctx {
   10706             :         void *ctx;
   10707             :         spdk_bdev_io_fn fn;
   10708             :         spdk_bdev_for_each_io_cb cb;
   10709             : };
   10710             : 
   10711             : static void
   10712           0 : bdev_channel_for_each_io(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
   10713             :                          struct spdk_io_channel *io_ch, void *_ctx)
   10714             : {
   10715           0 :         struct spdk_bdev_for_each_io_ctx *ctx = _ctx;
   10716           0 :         struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
   10717             :         struct spdk_bdev_io *bdev_io;
   10718           0 :         int rc = 0;
   10719             : 
   10720           0 :         TAILQ_FOREACH(bdev_io, &bdev_ch->io_submitted, internal.ch_link) {
   10721           0 :                 rc = ctx->fn(ctx->ctx, bdev_io);
   10722           0 :                 if (rc != 0) {
   10723           0 :                         break;
   10724             :                 }
   10725           0 :         }
   10726             : 
   10727           0 :         spdk_bdev_for_each_channel_continue(i, rc);
   10728           0 : }
   10729             : 
   10730             : static void
   10731           0 : bdev_for_each_io_done(struct spdk_bdev *bdev, void *_ctx, int status)
   10732             : {
   10733           0 :         struct spdk_bdev_for_each_io_ctx *ctx = _ctx;
   10734             : 
   10735           0 :         ctx->cb(ctx->ctx, status);
   10736             : 
   10737           0 :         free(ctx);
   10738           0 : }
   10739             : 
   10740             : void
   10741           0 : spdk_bdev_for_each_bdev_io(struct spdk_bdev *bdev, void *_ctx, spdk_bdev_io_fn fn,
   10742             :                            spdk_bdev_for_each_io_cb cb)
   10743             : {
   10744             :         struct spdk_bdev_for_each_io_ctx *ctx;
   10745             : 
   10746           0 :         assert(fn != NULL && cb != NULL);
   10747             : 
   10748           0 :         ctx = calloc(1, sizeof(*ctx));
   10749           0 :         if (ctx == NULL) {
   10750           0 :                 SPDK_ERRLOG("Failed to allocate context.\n");
   10751           0 :                 cb(_ctx, -ENOMEM);
   10752           0 :                 return;
   10753             :         }
   10754             : 
   10755           0 :         ctx->ctx = _ctx;
   10756           0 :         ctx->fn = fn;
   10757           0 :         ctx->cb = cb;
   10758             : 
   10759           0 :         spdk_bdev_for_each_channel(bdev, bdev_channel_for_each_io, ctx,
   10760             :                                    bdev_for_each_io_done);
   10761           0 : }
   10762             : 
   10763             : void
   10764         132 : spdk_bdev_for_each_channel_continue(struct spdk_bdev_channel_iter *iter, int status)
   10765             : {
   10766         132 :         spdk_for_each_channel_continue(iter->i, status);
   10767         132 : }
   10768             : 
   10769             : static struct spdk_bdev *
   10770         362 : io_channel_iter_get_bdev(struct spdk_io_channel_iter *i)
   10771             : {
   10772         362 :         void *io_device = spdk_io_channel_iter_get_io_device(i);
   10773             : 
   10774         362 :         return __bdev_from_io_dev(io_device);
   10775             : }
   10776             : 
   10777             : static void
   10778         132 : bdev_each_channel_msg(struct spdk_io_channel_iter *i)
   10779             : {
   10780         132 :         struct spdk_bdev_channel_iter *iter = spdk_io_channel_iter_get_ctx(i);
   10781         132 :         struct spdk_bdev *bdev = io_channel_iter_get_bdev(i);
   10782         132 :         struct spdk_io_channel *ch = spdk_io_channel_iter_get_channel(i);
   10783             : 
   10784         132 :         iter->i = i;
   10785         132 :         iter->fn(iter, bdev, ch, iter->ctx);
   10786         132 : }
   10787             : 
   10788             : static void
   10789         230 : bdev_each_channel_cpl(struct spdk_io_channel_iter *i, int status)
   10790             : {
   10791         230 :         struct spdk_bdev_channel_iter *iter = spdk_io_channel_iter_get_ctx(i);
   10792         230 :         struct spdk_bdev *bdev = io_channel_iter_get_bdev(i);
   10793             : 
   10794         230 :         iter->i = i;
   10795         230 :         iter->cpl(bdev, iter->ctx, status);
   10796             : 
   10797         230 :         free(iter);
   10798         230 : }
   10799             : 
   10800             : void
   10801         230 : spdk_bdev_for_each_channel(struct spdk_bdev *bdev, spdk_bdev_for_each_channel_msg fn,
   10802             :                            void *ctx, spdk_bdev_for_each_channel_done cpl)
   10803             : {
   10804             :         struct spdk_bdev_channel_iter *iter;
   10805             : 
   10806         230 :         assert(bdev != NULL && fn != NULL && ctx != NULL);
   10807             : 
   10808         230 :         iter = calloc(1, sizeof(struct spdk_bdev_channel_iter));
   10809         230 :         if (iter == NULL) {
   10810           0 :                 SPDK_ERRLOG("Unable to allocate iterator\n");
   10811           0 :                 assert(false);
   10812             :                 return;
   10813             :         }
   10814             : 
   10815         230 :         iter->fn = fn;
   10816         230 :         iter->cpl = cpl;
   10817         230 :         iter->ctx = ctx;
   10818             : 
   10819         460 :         spdk_for_each_channel(__bdev_to_io_dev(bdev), bdev_each_channel_msg,
   10820         230 :                               iter, bdev_each_channel_cpl);
   10821         230 : }
   10822             : 
   10823             : static void
   10824           3 : bdev_copy_do_write_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
   10825             : {
   10826           3 :         struct spdk_bdev_io *parent_io = cb_arg;
   10827             : 
   10828           3 :         spdk_bdev_free_io(bdev_io);
   10829             : 
   10830             :         /* Check return status of write */
   10831           3 :         parent_io->internal.status = success ? SPDK_BDEV_IO_STATUS_SUCCESS : SPDK_BDEV_IO_STATUS_FAILED;
   10832           3 :         parent_io->internal.cb(parent_io, success, parent_io->internal.caller_ctx);
   10833           3 : }
   10834             : 
   10835             : static void
   10836           3 : bdev_copy_do_write(void *_bdev_io)
   10837             : {
   10838           3 :         struct spdk_bdev_io *bdev_io = _bdev_io;
   10839             :         int rc;
   10840             : 
   10841             :         /* Write blocks */
   10842           6 :         rc = spdk_bdev_write_blocks_with_md(bdev_io->internal.desc,
   10843           3 :                                             spdk_io_channel_from_ctx(bdev_io->internal.ch),
   10844           3 :                                             bdev_io->u.bdev.iovs[0].iov_base,
   10845           3 :                                             bdev_io->u.bdev.md_buf, bdev_io->u.bdev.offset_blocks,
   10846           3 :                                             bdev_io->u.bdev.num_blocks, bdev_copy_do_write_done, bdev_io);
   10847             : 
   10848           3 :         if (rc == -ENOMEM) {
   10849           0 :                 bdev_queue_io_wait_with_cb(bdev_io, bdev_copy_do_write);
   10850           3 :         } else if (rc != 0) {
   10851           0 :                 bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
   10852           0 :                 bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
   10853           0 :         }
   10854           3 : }
   10855             : 
   10856             : static void
   10857           3 : bdev_copy_do_read_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
   10858             : {
   10859           3 :         struct spdk_bdev_io *parent_io = cb_arg;
   10860             : 
   10861           3 :         spdk_bdev_free_io(bdev_io);
   10862             : 
   10863             :         /* Check return status of read */
   10864           3 :         if (!success) {
   10865           0 :                 parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
   10866           0 :                 parent_io->internal.cb(parent_io, false, parent_io->internal.caller_ctx);
   10867           0 :                 return;
   10868             :         }
   10869             : 
   10870             :         /* Do write */
   10871           3 :         bdev_copy_do_write(parent_io);
   10872           3 : }
   10873             : 
   10874             : static void
   10875           3 : bdev_copy_do_read(void *_bdev_io)
   10876             : {
   10877           3 :         struct spdk_bdev_io *bdev_io = _bdev_io;
   10878             :         int rc;
   10879             : 
   10880             :         /* Read blocks */
   10881           6 :         rc = spdk_bdev_read_blocks_with_md(bdev_io->internal.desc,
   10882           3 :                                            spdk_io_channel_from_ctx(bdev_io->internal.ch),
   10883           3 :                                            bdev_io->u.bdev.iovs[0].iov_base,
   10884           3 :                                            bdev_io->u.bdev.md_buf, bdev_io->u.bdev.copy.src_offset_blocks,
   10885           3 :                                            bdev_io->u.bdev.num_blocks, bdev_copy_do_read_done, bdev_io);
   10886             : 
   10887           3 :         if (rc == -ENOMEM) {
   10888           0 :                 bdev_queue_io_wait_with_cb(bdev_io, bdev_copy_do_read);
   10889           3 :         } else if (rc != 0) {
   10890           0 :                 bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
   10891           0 :                 bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
   10892           0 :         }
   10893           3 : }
   10894             : 
   10895             : static void
   10896           3 : bdev_copy_get_buf_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io, bool success)
   10897             : {
   10898           3 :         if (!success) {
   10899           0 :                 bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
   10900           0 :                 bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
   10901           0 :                 return;
   10902             :         }
   10903             : 
   10904           3 :         bdev_copy_do_read(bdev_io);
   10905           3 : }
   10906             : 
   10907             : int
   10908          27 : spdk_bdev_copy_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
   10909             :                       uint64_t dst_offset_blocks, uint64_t src_offset_blocks, uint64_t num_blocks,
   10910             :                       spdk_bdev_io_completion_cb cb, void *cb_arg)
   10911             : {
   10912          27 :         struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
   10913             :         struct spdk_bdev_io *bdev_io;
   10914          27 :         struct spdk_bdev_channel *channel = spdk_io_channel_get_ctx(ch);
   10915             : 
   10916          27 :         if (!desc->write) {
   10917           0 :                 return -EBADF;
   10918             :         }
   10919             : 
   10920          27 :         if (!bdev_io_valid_blocks(bdev, dst_offset_blocks, num_blocks) ||
   10921          27 :             !bdev_io_valid_blocks(bdev, src_offset_blocks, num_blocks)) {
   10922           0 :                 SPDK_DEBUGLOG(bdev,
   10923             :                               "Invalid offset or number of blocks: dst %lu, src %lu, count %lu\n",
   10924             :                               dst_offset_blocks, src_offset_blocks, num_blocks);
   10925           0 :                 return -EINVAL;
   10926             :         }
   10927             : 
   10928          27 :         bdev_io = bdev_channel_get_io(channel);
   10929          27 :         if (!bdev_io) {
   10930           0 :                 return -ENOMEM;
   10931             :         }
   10932             : 
   10933          27 :         bdev_io->internal.ch = channel;
   10934          27 :         bdev_io->internal.desc = desc;
   10935          27 :         bdev_io->type = SPDK_BDEV_IO_TYPE_COPY;
   10936             : 
   10937          27 :         bdev_io->u.bdev.offset_blocks = dst_offset_blocks;
   10938          27 :         bdev_io->u.bdev.copy.src_offset_blocks = src_offset_blocks;
   10939          27 :         bdev_io->u.bdev.num_blocks = num_blocks;
   10940          27 :         bdev_io->u.bdev.memory_domain = NULL;
   10941          27 :         bdev_io->u.bdev.memory_domain_ctx = NULL;
   10942          27 :         bdev_io->u.bdev.iovs = NULL;
   10943          27 :         bdev_io->u.bdev.iovcnt = 0;
   10944          27 :         bdev_io->u.bdev.md_buf = NULL;
   10945          27 :         bdev_io->u.bdev.accel_sequence = NULL;
   10946          27 :         bdev_io_init(bdev_io, bdev, cb_arg, cb);
   10947             : 
   10948          27 :         if (dst_offset_blocks == src_offset_blocks || num_blocks == 0) {
   10949           0 :                 spdk_thread_send_msg(spdk_get_thread(), bdev_io_complete_cb, bdev_io);
   10950           0 :                 return 0;
   10951             :         }
   10952             : 
   10953             : 
   10954             :         /* If the copy size is large and should be split, use the generic split logic
   10955             :          * regardless of whether SPDK_BDEV_IO_TYPE_COPY is supported or not.
   10956             :          *
   10957             :          * Then, send the copy request if SPDK_BDEV_IO_TYPE_COPY is supported or
   10958             :          * emulate it using regular read and write requests otherwise.
   10959             :          */
   10960          27 :         if (spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COPY) ||
   10961           4 :             bdev_io->internal.f.split) {
   10962          24 :                 bdev_io_submit(bdev_io);
   10963          24 :                 return 0;
   10964             :         }
   10965             : 
   10966           3 :         spdk_bdev_io_get_buf(bdev_io, bdev_copy_get_buf_cb, num_blocks * spdk_bdev_get_block_size(bdev));
   10967             : 
   10968           3 :         return 0;
   10969          27 : }
   10970             : 
   10971           3 : SPDK_LOG_REGISTER_COMPONENT(bdev)
   10972             : 
   10973             : static void
   10974           0 : bdev_trace(void)
   10975             : {
   10976           0 :         struct spdk_trace_tpoint_opts opts[] = {
   10977             :                 {
   10978             :                         "BDEV_IO_START", TRACE_BDEV_IO_START,
   10979             :                         OWNER_TYPE_BDEV, OBJECT_BDEV_IO, 1,
   10980             :                         {
   10981             :                                 { "type", SPDK_TRACE_ARG_TYPE_INT, 8 },
   10982             :                                 { "ctx", SPDK_TRACE_ARG_TYPE_PTR, 8 },
   10983             :                                 { "offset", SPDK_TRACE_ARG_TYPE_INT, 8 },
   10984             :                                 { "qd", SPDK_TRACE_ARG_TYPE_INT, 4 }
   10985             :                         }
   10986             :                 },
   10987             :                 {
   10988             :                         "BDEV_IO_DONE", TRACE_BDEV_IO_DONE,
   10989             :                         OWNER_TYPE_BDEV, OBJECT_BDEV_IO, 0,
   10990             :                         {
   10991             :                                 { "ctx", SPDK_TRACE_ARG_TYPE_PTR, 8 },
   10992             :                                 { "qd", SPDK_TRACE_ARG_TYPE_INT, 4 }
   10993             :                         }
   10994             :                 },
   10995             :                 {
   10996             :                         "BDEV_IOCH_CREATE", TRACE_BDEV_IOCH_CREATE,
   10997             :                         OWNER_TYPE_BDEV, OBJECT_NONE, 0,
   10998             :                         {
   10999             :                                 { "tid", SPDK_TRACE_ARG_TYPE_INT, 8 }
   11000             :                         }
   11001             :                 },
   11002             :                 {
   11003             :                         "BDEV_IOCH_DESTROY", TRACE_BDEV_IOCH_DESTROY,
   11004             :                         OWNER_TYPE_BDEV, OBJECT_NONE, 0,
   11005             :                         {
   11006             :                                 { "tid", SPDK_TRACE_ARG_TYPE_INT, 8 }
   11007             :                         }
   11008             :                 },
   11009             :         };
   11010             : 
   11011             : 
   11012           0 :         spdk_trace_register_owner_type(OWNER_TYPE_BDEV, 'b');
   11013           0 :         spdk_trace_register_object(OBJECT_BDEV_IO, 'i');
   11014           0 :         spdk_trace_register_description_ext(opts, SPDK_COUNTOF(opts));
   11015           0 :         spdk_trace_tpoint_register_relation(TRACE_BDEV_NVME_IO_START, OBJECT_BDEV_IO, 0);
   11016           0 :         spdk_trace_tpoint_register_relation(TRACE_BDEV_NVME_IO_DONE, OBJECT_BDEV_IO, 0);
   11017           0 :         spdk_trace_tpoint_register_relation(TRACE_BLOB_REQ_SET_START, OBJECT_BDEV_IO, 0);
   11018           0 :         spdk_trace_tpoint_register_relation(TRACE_BLOB_REQ_SET_COMPLETE, OBJECT_BDEV_IO, 0);
   11019           0 :         spdk_trace_tpoint_register_relation(TRACE_BDEV_RAID_IO_START, OBJECT_BDEV_IO, 0);
   11020           0 :         spdk_trace_tpoint_register_relation(TRACE_BDEV_RAID_IO_DONE, OBJECT_BDEV_IO, 0);
   11021           0 : }
   11022           3 : SPDK_TRACE_REGISTER_FN(bdev_trace, "bdev", TRACE_GROUP_BDEV)

Generated by: LCOV version 1.15