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