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