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