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