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