Line data Source code
1 : /* SPDX-License-Identifier: BSD-3-Clause
2 : * Copyright (C) 2016 Intel Corporation. All rights reserved.
3 : * Copyright (c) 2019 Mellanox Technologies LTD. All rights reserved.
4 : * Copyright (c) 2021-2023 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
5 : */
6 :
7 : #include "spdk/stdinc.h"
8 :
9 : #include "spdk/bdev.h"
10 :
11 : #include "spdk/accel.h"
12 : #include "spdk/config.h"
13 : #include "spdk/env.h"
14 : #include "spdk/thread.h"
15 : #include "spdk/likely.h"
16 : #include "spdk/queue.h"
17 : #include "spdk/nvme_spec.h"
18 : #include "spdk/scsi_spec.h"
19 : #include "spdk/notify.h"
20 : #include "spdk/util.h"
21 : #include "spdk/trace.h"
22 : #include "spdk/dma.h"
23 :
24 : #include "spdk/bdev_module.h"
25 : #include "spdk/log.h"
26 : #include "spdk/string.h"
27 :
28 : #include "bdev_internal.h"
29 : #include "spdk_internal/trace_defs.h"
30 : #include "spdk_internal/assert.h"
31 :
32 : #ifdef SPDK_CONFIG_VTUNE
33 : #include "ittnotify.h"
34 : #include "ittnotify_types.h"
35 : int __itt_init_ittlib(const char *, __itt_group_id);
36 : #endif
37 :
38 : #define SPDK_BDEV_IO_POOL_SIZE (64 * 1024 - 1)
39 : #define SPDK_BDEV_IO_CACHE_SIZE 256
40 : #define SPDK_BDEV_AUTO_EXAMINE true
41 : #define BUF_SMALL_CACHE_SIZE 128
42 : #define BUF_LARGE_CACHE_SIZE 16
43 : #define NOMEM_THRESHOLD_COUNT 8
44 :
45 : #define SPDK_BDEV_QOS_TIMESLICE_IN_USEC 1000
46 : #define SPDK_BDEV_QOS_MIN_IO_PER_TIMESLICE 1
47 : #define SPDK_BDEV_QOS_MIN_BYTE_PER_TIMESLICE 512
48 : #define SPDK_BDEV_QOS_MIN_IOS_PER_SEC 1000
49 : #define SPDK_BDEV_QOS_MIN_BYTES_PER_SEC (1024 * 1024)
50 : #define SPDK_BDEV_QOS_MAX_MBYTES_PER_SEC (UINT64_MAX / (1024 * 1024))
51 : #define SPDK_BDEV_QOS_LIMIT_NOT_DEFINED UINT64_MAX
52 : #define SPDK_BDEV_IO_POLL_INTERVAL_IN_MSEC 1000
53 :
54 : /* The maximum number of children requests for a UNMAP or WRITE ZEROES command
55 : * when splitting into children requests at a time.
56 : */
57 : #define SPDK_BDEV_MAX_CHILDREN_UNMAP_WRITE_ZEROES_REQS (8)
58 : #define BDEV_RESET_CHECK_OUTSTANDING_IO_PERIOD 1000000
59 :
60 : /* The maximum number of children requests for a COPY command
61 : * when splitting into children requests at a time.
62 : */
63 : #define SPDK_BDEV_MAX_CHILDREN_COPY_REQS (8)
64 :
65 : #define LOG_ALREADY_CLAIMED_ERROR(detail, bdev) \
66 : log_already_claimed(SPDK_LOG_ERROR, __LINE__, __func__, detail, bdev)
67 : #ifdef DEBUG
68 : #define LOG_ALREADY_CLAIMED_DEBUG(detail, bdev) \
69 : log_already_claimed(SPDK_LOG_DEBUG, __LINE__, __func__, detail, bdev)
70 : #else
71 : #define LOG_ALREADY_CLAIMED_DEBUG(detail, bdev) do {} while(0)
72 : #endif
73 :
74 : static void log_already_claimed(enum spdk_log_level level, const int line, const char *func,
75 : const char *detail, struct spdk_bdev *bdev);
76 :
77 : static const char *qos_rpc_type[] = {"rw_ios_per_sec",
78 : "rw_mbytes_per_sec", "r_mbytes_per_sec", "w_mbytes_per_sec"
79 : };
80 :
81 : TAILQ_HEAD(spdk_bdev_list, spdk_bdev);
82 :
83 : RB_HEAD(bdev_name_tree, spdk_bdev_name);
84 :
85 : static int
86 541 : bdev_name_cmp(struct spdk_bdev_name *name1, struct spdk_bdev_name *name2)
87 : {
88 541 : return strcmp(name1->name, name2->name);
89 : }
90 :
91 2034 : RB_GENERATE_STATIC(bdev_name_tree, spdk_bdev_name, node, bdev_name_cmp);
92 :
93 : struct spdk_bdev_mgr {
94 : struct spdk_mempool *bdev_io_pool;
95 :
96 : void *zero_buffer;
97 :
98 : TAILQ_HEAD(bdev_module_list, spdk_bdev_module) bdev_modules;
99 :
100 : struct spdk_bdev_list bdevs;
101 : struct bdev_name_tree bdev_names;
102 :
103 : bool init_complete;
104 : bool module_init_complete;
105 :
106 : struct spdk_spinlock spinlock;
107 :
108 : TAILQ_HEAD(, spdk_bdev_open_async_ctx) async_bdev_opens;
109 :
110 : #ifdef SPDK_CONFIG_VTUNE
111 : __itt_domain *domain;
112 : #endif
113 : };
114 :
115 : static struct spdk_bdev_mgr g_bdev_mgr = {
116 : .bdev_modules = TAILQ_HEAD_INITIALIZER(g_bdev_mgr.bdev_modules),
117 : .bdevs = TAILQ_HEAD_INITIALIZER(g_bdev_mgr.bdevs),
118 : .bdev_names = RB_INITIALIZER(g_bdev_mgr.bdev_names),
119 : .init_complete = false,
120 : .module_init_complete = false,
121 : .async_bdev_opens = TAILQ_HEAD_INITIALIZER(g_bdev_mgr.async_bdev_opens),
122 : };
123 :
124 : static void
125 : __attribute__((constructor))
126 3 : _bdev_init(void)
127 : {
128 3 : spdk_spin_init(&g_bdev_mgr.spinlock);
129 3 : }
130 :
131 : typedef void (*lock_range_cb)(struct lba_range *range, void *ctx, int status);
132 :
133 : typedef void (*bdev_copy_bounce_buffer_cpl)(void *ctx, int rc);
134 :
135 : struct lba_range {
136 : struct spdk_bdev *bdev;
137 : uint64_t offset;
138 : uint64_t length;
139 : bool quiesce;
140 : void *locked_ctx;
141 : struct spdk_thread *owner_thread;
142 : struct spdk_bdev_channel *owner_ch;
143 : TAILQ_ENTRY(lba_range) tailq;
144 : TAILQ_ENTRY(lba_range) tailq_module;
145 : };
146 :
147 : static struct spdk_bdev_opts g_bdev_opts = {
148 : .bdev_io_pool_size = SPDK_BDEV_IO_POOL_SIZE,
149 : .bdev_io_cache_size = SPDK_BDEV_IO_CACHE_SIZE,
150 : .bdev_auto_examine = SPDK_BDEV_AUTO_EXAMINE,
151 : .iobuf_small_cache_size = BUF_SMALL_CACHE_SIZE,
152 : .iobuf_large_cache_size = BUF_LARGE_CACHE_SIZE,
153 : };
154 :
155 : static spdk_bdev_init_cb g_init_cb_fn = NULL;
156 : static void *g_init_cb_arg = NULL;
157 :
158 : static spdk_bdev_fini_cb g_fini_cb_fn = NULL;
159 : static void *g_fini_cb_arg = NULL;
160 : static struct spdk_thread *g_fini_thread = NULL;
161 :
162 : struct spdk_bdev_qos_limit {
163 : /** IOs or bytes allowed per second (i.e., 1s). */
164 : uint64_t limit;
165 :
166 : /** Remaining IOs or bytes allowed in current timeslice (e.g., 1ms).
167 : * For remaining bytes, allowed to run negative if an I/O is submitted when
168 : * some bytes are remaining, but the I/O is bigger than that amount. The
169 : * excess will be deducted from the next timeslice.
170 : */
171 : int64_t remaining_this_timeslice;
172 :
173 : /** Minimum allowed IOs or bytes to be issued in one timeslice (e.g., 1ms). */
174 : uint32_t min_per_timeslice;
175 :
176 : /** Maximum allowed IOs or bytes to be issued in one timeslice (e.g., 1ms). */
177 : uint32_t max_per_timeslice;
178 :
179 : /** Function to check whether to queue the IO.
180 : * If The IO is allowed to pass, the quota will be reduced correspondingly.
181 : */
182 : bool (*queue_io)(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io);
183 :
184 : /** Function to rewind the quota once the IO was allowed to be sent by this
185 : * limit but queued due to one of the further limits.
186 : */
187 : void (*rewind_quota)(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io);
188 : };
189 :
190 : struct spdk_bdev_qos {
191 : /** Types of structure of rate limits. */
192 : struct spdk_bdev_qos_limit rate_limits[SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES];
193 :
194 : /** The channel that all I/O are funneled through. */
195 : struct spdk_bdev_channel *ch;
196 :
197 : /** The thread on which the poller is running. */
198 : struct spdk_thread *thread;
199 :
200 : /** Size of a timeslice in tsc ticks. */
201 : uint64_t timeslice_size;
202 :
203 : /** Timestamp of start of last timeslice. */
204 : uint64_t last_timeslice;
205 :
206 : /** Poller that processes queued I/O commands each time slice. */
207 : struct spdk_poller *poller;
208 : };
209 :
210 : struct spdk_bdev_mgmt_channel {
211 : /*
212 : * Each thread keeps a cache of bdev_io - this allows
213 : * bdev threads which are *not* DPDK threads to still
214 : * benefit from a per-thread bdev_io cache. Without
215 : * this, non-DPDK threads fetching from the mempool
216 : * incur a cmpxchg on get and put.
217 : */
218 : bdev_io_stailq_t per_thread_cache;
219 : uint32_t per_thread_cache_count;
220 : uint32_t bdev_io_cache_size;
221 :
222 : struct spdk_iobuf_channel iobuf;
223 :
224 : TAILQ_HEAD(, spdk_bdev_shared_resource) shared_resources;
225 : TAILQ_HEAD(, spdk_bdev_io_wait_entry) io_wait_queue;
226 : };
227 :
228 : /*
229 : * Per-module (or per-io_device) data. Multiple bdevs built on the same io_device
230 : * will queue here their IO that awaits retry. It makes it possible to retry sending
231 : * IO to one bdev after IO from other bdev completes.
232 : */
233 : struct spdk_bdev_shared_resource {
234 : /* The bdev management channel */
235 : struct spdk_bdev_mgmt_channel *mgmt_ch;
236 :
237 : /*
238 : * Count of I/O submitted to bdev module and waiting for completion.
239 : * Incremented before submit_request() is called on an spdk_bdev_io.
240 : */
241 : uint64_t io_outstanding;
242 :
243 : /*
244 : * Queue of IO awaiting retry because of a previous NOMEM status returned
245 : * on this channel.
246 : */
247 : bdev_io_tailq_t nomem_io;
248 :
249 : /*
250 : * Threshold which io_outstanding must drop to before retrying nomem_io.
251 : */
252 : uint64_t nomem_threshold;
253 :
254 : /* I/O channel allocated by a bdev module */
255 : struct spdk_io_channel *shared_ch;
256 :
257 : struct spdk_poller *nomem_poller;
258 :
259 : /* Refcount of bdev channels using this resource */
260 : uint32_t ref;
261 :
262 : TAILQ_ENTRY(spdk_bdev_shared_resource) link;
263 : };
264 :
265 : #define BDEV_CH_RESET_IN_PROGRESS (1 << 0)
266 : #define BDEV_CH_QOS_ENABLED (1 << 1)
267 :
268 : struct spdk_bdev_channel {
269 : struct spdk_bdev *bdev;
270 :
271 : /* The channel for the underlying device */
272 : struct spdk_io_channel *channel;
273 :
274 : /* Accel channel */
275 : struct spdk_io_channel *accel_channel;
276 :
277 : /* Per io_device per thread data */
278 : struct spdk_bdev_shared_resource *shared_resource;
279 :
280 : struct spdk_bdev_io_stat *stat;
281 :
282 : /*
283 : * Count of I/O submitted to the underlying dev module through this channel
284 : * and waiting for completion.
285 : */
286 : uint64_t io_outstanding;
287 :
288 : /*
289 : * List of all submitted I/Os including I/O that are generated via splitting.
290 : */
291 : bdev_io_tailq_t io_submitted;
292 :
293 : /*
294 : * List of spdk_bdev_io that are currently queued because they write to a locked
295 : * LBA range.
296 : */
297 : bdev_io_tailq_t io_locked;
298 :
299 : /* List of I/Os with accel sequence being currently executed */
300 : bdev_io_tailq_t io_accel_exec;
301 :
302 : /* List of I/Os doing memory domain pull/push */
303 : bdev_io_tailq_t io_memory_domain;
304 :
305 : uint32_t flags;
306 :
307 : /* Counts number of bdev_io in the io_submitted TAILQ */
308 : uint16_t queue_depth;
309 :
310 : uint16_t trace_id;
311 :
312 : struct spdk_histogram_data *histogram;
313 :
314 : #ifdef SPDK_CONFIG_VTUNE
315 : uint64_t start_tsc;
316 : uint64_t interval_tsc;
317 : __itt_string_handle *handle;
318 : struct spdk_bdev_io_stat *prev_stat;
319 : #endif
320 :
321 : bdev_io_tailq_t queued_resets;
322 :
323 : lba_range_tailq_t locked_ranges;
324 :
325 : /** List of I/Os queued by QoS. */
326 : bdev_io_tailq_t qos_queued_io;
327 : };
328 :
329 : struct media_event_entry {
330 : struct spdk_bdev_media_event event;
331 : TAILQ_ENTRY(media_event_entry) tailq;
332 : };
333 :
334 : #define MEDIA_EVENT_POOL_SIZE 64
335 :
336 : struct spdk_bdev_desc {
337 : struct spdk_bdev *bdev;
338 : 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 : enum spdk_bdev_reset_stat_mode reset_mode;
364 : spdk_bdev_get_device_stat_cb cb;
365 : void *cb_arg;
366 : };
367 :
368 : struct set_qos_limit_ctx {
369 : void (*cb_fn)(void *cb_arg, int status);
370 : void *cb_arg;
371 : struct spdk_bdev *bdev;
372 : };
373 :
374 : struct spdk_bdev_channel_iter {
375 : spdk_bdev_for_each_channel_msg fn;
376 : spdk_bdev_for_each_channel_done cpl;
377 : struct spdk_io_channel_iter *i;
378 : void *ctx;
379 : };
380 :
381 : struct spdk_bdev_io_error_stat {
382 : uint32_t error_status[-SPDK_MIN_BDEV_IO_STATUS];
383 : };
384 :
385 : enum bdev_io_retry_state {
386 : BDEV_IO_RETRY_STATE_INVALID,
387 : BDEV_IO_RETRY_STATE_PULL,
388 : BDEV_IO_RETRY_STATE_PULL_MD,
389 : BDEV_IO_RETRY_STATE_SUBMIT,
390 : BDEV_IO_RETRY_STATE_PUSH,
391 : BDEV_IO_RETRY_STATE_PUSH_MD,
392 : };
393 :
394 : #define __bdev_to_io_dev(bdev) (((char *)bdev) + 1)
395 : #define __bdev_from_io_dev(io_dev) ((struct spdk_bdev *)(((char *)io_dev) - 1))
396 : #define __io_ch_to_bdev_ch(io_ch) ((struct spdk_bdev_channel *)spdk_io_channel_get_ctx(io_ch))
397 : #define __io_ch_to_bdev_mgmt_ch(io_ch) ((struct spdk_bdev_mgmt_channel *)spdk_io_channel_get_ctx(io_ch))
398 :
399 : static inline void bdev_io_complete(void *ctx);
400 : static inline void bdev_io_complete_unsubmitted(struct spdk_bdev_io *bdev_io);
401 : static void bdev_io_push_bounce_md_buf(struct spdk_bdev_io *bdev_io);
402 : static void bdev_io_push_bounce_data(struct spdk_bdev_io *bdev_io);
403 :
404 : static void bdev_write_zero_buffer_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg);
405 : static int bdev_write_zero_buffer(struct spdk_bdev_io *bdev_io);
406 :
407 : static void bdev_enable_qos_msg(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
408 : struct spdk_io_channel *ch, void *_ctx);
409 : static void bdev_enable_qos_done(struct spdk_bdev *bdev, void *_ctx, int status);
410 :
411 : static int bdev_readv_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
412 : struct iovec *iov, int iovcnt, void *md_buf, uint64_t offset_blocks,
413 : uint64_t num_blocks,
414 : struct spdk_memory_domain *domain, void *domain_ctx,
415 : struct spdk_accel_sequence *seq, uint32_t dif_check_flags,
416 : spdk_bdev_io_completion_cb cb, void *cb_arg);
417 : static int bdev_writev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
418 : struct iovec *iov, int iovcnt, void *md_buf,
419 : uint64_t offset_blocks, uint64_t num_blocks,
420 : struct spdk_memory_domain *domain, void *domain_ctx,
421 : struct spdk_accel_sequence *seq, uint32_t dif_check_flags,
422 : uint32_t nvme_cdw12_raw, uint32_t nvme_cdw13_raw,
423 : spdk_bdev_io_completion_cb cb, void *cb_arg);
424 :
425 : static int bdev_lock_lba_range(struct spdk_bdev_desc *desc, struct spdk_io_channel *_ch,
426 : uint64_t offset, uint64_t length,
427 : lock_range_cb cb_fn, void *cb_arg);
428 :
429 : static int bdev_unlock_lba_range(struct spdk_bdev_desc *desc, struct spdk_io_channel *_ch,
430 : uint64_t offset, uint64_t length,
431 : lock_range_cb cb_fn, void *cb_arg);
432 :
433 : static bool bdev_abort_queued_io(bdev_io_tailq_t *queue, struct spdk_bdev_io *bio_to_abort);
434 : static bool bdev_abort_buf_io(struct spdk_bdev_mgmt_channel *ch, struct spdk_bdev_io *bio_to_abort);
435 :
436 : static bool claim_type_is_v2(enum spdk_bdev_claim_type type);
437 : static void bdev_desc_release_claims(struct spdk_bdev_desc *desc);
438 : static void claim_reset(struct spdk_bdev *bdev);
439 :
440 : static void bdev_ch_retry_io(struct spdk_bdev_channel *bdev_ch);
441 :
442 : static bool bdev_io_should_split(struct spdk_bdev_io *bdev_io);
443 :
444 : #define bdev_get_ext_io_opt(opts, field, defval) \
445 : ((opts) != NULL ? SPDK_GET_FIELD(opts, field, defval) : (defval))
446 :
447 : static inline void
448 669 : bdev_ch_add_to_io_submitted(struct spdk_bdev_io *bdev_io)
449 : {
450 669 : TAILQ_INSERT_TAIL(&bdev_io->internal.ch->io_submitted, bdev_io, internal.ch_link);
451 669 : bdev_io->internal.ch->queue_depth++;
452 669 : }
453 :
454 : static inline void
455 669 : bdev_ch_remove_from_io_submitted(struct spdk_bdev_io *bdev_io)
456 : {
457 669 : TAILQ_REMOVE(&bdev_io->internal.ch->io_submitted, bdev_io, internal.ch_link);
458 669 : bdev_io->internal.ch->queue_depth--;
459 669 : }
460 :
461 : void
462 14 : spdk_bdev_get_opts(struct spdk_bdev_opts *opts, size_t opts_size)
463 : {
464 14 : if (!opts) {
465 0 : SPDK_ERRLOG("opts should not be NULL\n");
466 0 : return;
467 : }
468 :
469 14 : if (!opts_size) {
470 0 : SPDK_ERRLOG("opts_size should not be zero value\n");
471 0 : return;
472 : }
473 :
474 14 : opts->opts_size = opts_size;
475 :
476 : #define SET_FIELD(field) \
477 : if (offsetof(struct spdk_bdev_opts, field) + sizeof(opts->field) <= opts_size) { \
478 : opts->field = g_bdev_opts.field; \
479 : } \
480 :
481 14 : SET_FIELD(bdev_io_pool_size);
482 14 : SET_FIELD(bdev_io_cache_size);
483 14 : SET_FIELD(bdev_auto_examine);
484 14 : SET_FIELD(iobuf_small_cache_size);
485 14 : SET_FIELD(iobuf_large_cache_size);
486 :
487 : /* Do not remove this statement, you should always update this statement when you adding a new field,
488 : * and do not forget to add the SET_FIELD statement for your added field. */
489 : SPDK_STATIC_ASSERT(sizeof(struct spdk_bdev_opts) == 32, "Incorrect size");
490 :
491 : #undef SET_FIELD
492 14 : }
493 :
494 : int
495 15 : spdk_bdev_set_opts(struct spdk_bdev_opts *opts)
496 : {
497 : uint32_t min_pool_size;
498 :
499 15 : if (!opts) {
500 0 : SPDK_ERRLOG("opts cannot be NULL\n");
501 0 : return -1;
502 : }
503 :
504 15 : if (!opts->opts_size) {
505 1 : SPDK_ERRLOG("opts_size inside opts cannot be zero value\n");
506 1 : return -1;
507 : }
508 :
509 : /*
510 : * Add 1 to the thread count to account for the extra mgmt_ch that gets created during subsystem
511 : * initialization. A second mgmt_ch will be created on the same thread when the application starts
512 : * but before the deferred put_io_channel event is executed for the first mgmt_ch.
513 : */
514 14 : min_pool_size = opts->bdev_io_cache_size * (spdk_thread_get_count() + 1);
515 14 : if (opts->bdev_io_pool_size < min_pool_size) {
516 0 : SPDK_ERRLOG("bdev_io_pool_size %" PRIu32 " is not compatible with bdev_io_cache_size %" PRIu32
517 : " and %" PRIu32 " threads\n", opts->bdev_io_pool_size, opts->bdev_io_cache_size,
518 : spdk_thread_get_count());
519 0 : SPDK_ERRLOG("bdev_io_pool_size must be at least %" PRIu32 "\n", min_pool_size);
520 0 : return -1;
521 : }
522 :
523 : #define SET_FIELD(field) \
524 : if (offsetof(struct spdk_bdev_opts, field) + sizeof(opts->field) <= opts->opts_size) { \
525 : g_bdev_opts.field = opts->field; \
526 : } \
527 :
528 14 : SET_FIELD(bdev_io_pool_size);
529 14 : SET_FIELD(bdev_io_cache_size);
530 14 : SET_FIELD(bdev_auto_examine);
531 14 : SET_FIELD(iobuf_small_cache_size);
532 14 : SET_FIELD(iobuf_large_cache_size);
533 :
534 14 : g_bdev_opts.opts_size = opts->opts_size;
535 :
536 : #undef SET_FIELD
537 :
538 14 : return 0;
539 15 : }
540 :
541 : static struct spdk_bdev *
542 144 : bdev_get_by_name(const char *bdev_name)
543 : {
544 : struct spdk_bdev_name find;
545 : struct spdk_bdev_name *res;
546 :
547 144 : find.name = (char *)bdev_name;
548 144 : res = RB_FIND(bdev_name_tree, &g_bdev_mgr.bdev_names, &find);
549 144 : if (res != NULL) {
550 137 : return res->bdev;
551 : }
552 :
553 7 : return NULL;
554 144 : }
555 :
556 : struct spdk_bdev *
557 16 : spdk_bdev_get_by_name(const char *bdev_name)
558 : {
559 : struct spdk_bdev *bdev;
560 :
561 16 : spdk_spin_lock(&g_bdev_mgr.spinlock);
562 16 : bdev = bdev_get_by_name(bdev_name);
563 16 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
564 :
565 16 : return bdev;
566 : }
567 :
568 : struct bdev_io_status_string {
569 : enum spdk_bdev_io_status status;
570 : const char *str;
571 : };
572 :
573 : static const struct bdev_io_status_string bdev_io_status_strings[] = {
574 : { SPDK_BDEV_IO_STATUS_AIO_ERROR, "aio_error" },
575 : { SPDK_BDEV_IO_STATUS_ABORTED, "aborted" },
576 : { SPDK_BDEV_IO_STATUS_FIRST_FUSED_FAILED, "first_fused_failed" },
577 : { SPDK_BDEV_IO_STATUS_MISCOMPARE, "miscompare" },
578 : { SPDK_BDEV_IO_STATUS_NOMEM, "nomem" },
579 : { SPDK_BDEV_IO_STATUS_SCSI_ERROR, "scsi_error" },
580 : { SPDK_BDEV_IO_STATUS_NVME_ERROR, "nvme_error" },
581 : { SPDK_BDEV_IO_STATUS_FAILED, "failed" },
582 : { SPDK_BDEV_IO_STATUS_PENDING, "pending" },
583 : { SPDK_BDEV_IO_STATUS_SUCCESS, "success" },
584 : };
585 :
586 : static const char *
587 0 : bdev_io_status_get_string(enum spdk_bdev_io_status status)
588 : {
589 : uint32_t i;
590 :
591 0 : for (i = 0; i < SPDK_COUNTOF(bdev_io_status_strings); i++) {
592 0 : if (bdev_io_status_strings[i].status == status) {
593 0 : return bdev_io_status_strings[i].str;
594 : }
595 0 : }
596 :
597 0 : return "reserved";
598 0 : }
599 :
600 : struct spdk_bdev_wait_for_examine_ctx {
601 : struct spdk_poller *poller;
602 : spdk_bdev_wait_for_examine_cb cb_fn;
603 : void *cb_arg;
604 : };
605 :
606 : static bool bdev_module_all_actions_completed(void);
607 :
608 : static int
609 191 : bdev_wait_for_examine_cb(void *arg)
610 : {
611 191 : struct spdk_bdev_wait_for_examine_ctx *ctx = arg;
612 :
613 191 : if (!bdev_module_all_actions_completed()) {
614 0 : return SPDK_POLLER_IDLE;
615 : }
616 :
617 191 : spdk_poller_unregister(&ctx->poller);
618 191 : ctx->cb_fn(ctx->cb_arg);
619 191 : free(ctx);
620 :
621 191 : return SPDK_POLLER_BUSY;
622 191 : }
623 :
624 : int
625 191 : spdk_bdev_wait_for_examine(spdk_bdev_wait_for_examine_cb cb_fn, void *cb_arg)
626 : {
627 : struct spdk_bdev_wait_for_examine_ctx *ctx;
628 :
629 191 : ctx = calloc(1, sizeof(*ctx));
630 191 : if (ctx == NULL) {
631 0 : return -ENOMEM;
632 : }
633 191 : ctx->cb_fn = cb_fn;
634 191 : ctx->cb_arg = cb_arg;
635 191 : ctx->poller = SPDK_POLLER_REGISTER(bdev_wait_for_examine_cb, ctx, 0);
636 :
637 191 : return 0;
638 191 : }
639 :
640 : struct spdk_bdev_examine_item {
641 : char *name;
642 : TAILQ_ENTRY(spdk_bdev_examine_item) link;
643 : };
644 :
645 : TAILQ_HEAD(spdk_bdev_examine_allowlist, spdk_bdev_examine_item);
646 :
647 : struct spdk_bdev_examine_allowlist g_bdev_examine_allowlist = TAILQ_HEAD_INITIALIZER(
648 : g_bdev_examine_allowlist);
649 :
650 : static inline bool
651 6 : bdev_examine_allowlist_check(const char *name)
652 : {
653 : struct spdk_bdev_examine_item *item;
654 6 : TAILQ_FOREACH(item, &g_bdev_examine_allowlist, link) {
655 0 : if (strcmp(name, item->name) == 0) {
656 0 : return true;
657 : }
658 0 : }
659 6 : return false;
660 6 : }
661 :
662 : static inline void
663 65 : bdev_examine_allowlist_free(void)
664 : {
665 : struct spdk_bdev_examine_item *item;
666 65 : while (!TAILQ_EMPTY(&g_bdev_examine_allowlist)) {
667 0 : item = TAILQ_FIRST(&g_bdev_examine_allowlist);
668 0 : TAILQ_REMOVE(&g_bdev_examine_allowlist, item, link);
669 0 : free(item->name);
670 0 : free(item);
671 : }
672 65 : }
673 :
674 : static inline bool
675 3 : bdev_in_examine_allowlist(struct spdk_bdev *bdev)
676 : {
677 : struct spdk_bdev_alias *tmp;
678 3 : if (bdev_examine_allowlist_check(bdev->name)) {
679 0 : return true;
680 : }
681 6 : TAILQ_FOREACH(tmp, &bdev->aliases, tailq) {
682 3 : if (bdev_examine_allowlist_check(tmp->alias.name)) {
683 0 : return true;
684 : }
685 3 : }
686 3 : return false;
687 3 : }
688 :
689 : static inline bool
690 124 : bdev_ok_to_examine(struct spdk_bdev *bdev)
691 : {
692 : /* Some bdevs may not support the READ command.
693 : * Do not try to examine them.
694 : */
695 124 : if (!spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_READ)) {
696 0 : return false;
697 : }
698 :
699 124 : if (g_bdev_opts.bdev_auto_examine) {
700 121 : return true;
701 : } else {
702 3 : return bdev_in_examine_allowlist(bdev);
703 : }
704 124 : }
705 :
706 : static void
707 124 : bdev_examine(struct spdk_bdev *bdev)
708 : {
709 : struct spdk_bdev_module *module;
710 : struct spdk_bdev_module_claim *claim, *tmpclaim;
711 : uint32_t action;
712 :
713 124 : if (!bdev_ok_to_examine(bdev)) {
714 3 : return;
715 : }
716 :
717 494 : TAILQ_FOREACH(module, &g_bdev_mgr.bdev_modules, internal.tailq) {
718 373 : if (module->examine_config) {
719 252 : spdk_spin_lock(&module->internal.spinlock);
720 252 : action = module->internal.action_in_progress;
721 252 : module->internal.action_in_progress++;
722 252 : spdk_spin_unlock(&module->internal.spinlock);
723 252 : module->examine_config(bdev);
724 252 : if (action != module->internal.action_in_progress) {
725 0 : SPDK_ERRLOG("examine_config for module %s did not call "
726 : "spdk_bdev_module_examine_done()\n", module->name);
727 0 : }
728 252 : }
729 373 : }
730 :
731 121 : spdk_spin_lock(&bdev->internal.spinlock);
732 :
733 121 : switch (bdev->internal.claim_type) {
734 : case SPDK_BDEV_CLAIM_NONE:
735 : /* Examine by all bdev modules */
736 469 : TAILQ_FOREACH(module, &g_bdev_mgr.bdev_modules, internal.tailq) {
737 353 : if (module->examine_disk) {
738 228 : spdk_spin_lock(&module->internal.spinlock);
739 228 : module->internal.action_in_progress++;
740 228 : spdk_spin_unlock(&module->internal.spinlock);
741 228 : spdk_spin_unlock(&bdev->internal.spinlock);
742 228 : module->examine_disk(bdev);
743 228 : spdk_spin_lock(&bdev->internal.spinlock);
744 228 : }
745 353 : }
746 116 : break;
747 : case SPDK_BDEV_CLAIM_EXCL_WRITE:
748 : /* Examine by the one bdev module with a v1 claim */
749 1 : module = bdev->internal.claim.v1.module;
750 1 : if (module->examine_disk) {
751 1 : spdk_spin_lock(&module->internal.spinlock);
752 1 : module->internal.action_in_progress++;
753 1 : spdk_spin_unlock(&module->internal.spinlock);
754 1 : spdk_spin_unlock(&bdev->internal.spinlock);
755 1 : module->examine_disk(bdev);
756 1 : return;
757 : }
758 0 : break;
759 : default:
760 : /* Examine by all bdev modules with a v2 claim */
761 4 : assert(claim_type_is_v2(bdev->internal.claim_type));
762 : /*
763 : * Removal of tailq nodes while iterating can cause the iteration to jump out of the
764 : * list, perhaps accessing freed memory. Without protection, this could happen
765 : * while the lock is dropped during the examine callback.
766 : */
767 4 : bdev->internal.examine_in_progress++;
768 :
769 9 : TAILQ_FOREACH(claim, &bdev->internal.claim.v2.claims, link) {
770 5 : module = claim->module;
771 :
772 5 : if (module == NULL) {
773 : /* This is a vestigial claim, held by examine_count */
774 0 : continue;
775 : }
776 :
777 5 : if (module->examine_disk == NULL) {
778 0 : continue;
779 : }
780 :
781 5 : spdk_spin_lock(&module->internal.spinlock);
782 5 : module->internal.action_in_progress++;
783 5 : spdk_spin_unlock(&module->internal.spinlock);
784 :
785 : /* Call examine_disk without holding internal.spinlock. */
786 5 : spdk_spin_unlock(&bdev->internal.spinlock);
787 5 : module->examine_disk(bdev);
788 5 : spdk_spin_lock(&bdev->internal.spinlock);
789 5 : }
790 :
791 4 : assert(bdev->internal.examine_in_progress > 0);
792 4 : bdev->internal.examine_in_progress--;
793 4 : if (bdev->internal.examine_in_progress == 0) {
794 : /* Remove any claims that were released during examine_disk */
795 9 : TAILQ_FOREACH_SAFE(claim, &bdev->internal.claim.v2.claims, link, tmpclaim) {
796 5 : if (claim->desc != NULL) {
797 5 : continue;
798 : }
799 :
800 0 : TAILQ_REMOVE(&bdev->internal.claim.v2.claims, claim, link);
801 0 : free(claim);
802 0 : }
803 4 : if (TAILQ_EMPTY(&bdev->internal.claim.v2.claims)) {
804 0 : claim_reset(bdev);
805 0 : }
806 4 : }
807 4 : }
808 :
809 120 : spdk_spin_unlock(&bdev->internal.spinlock);
810 124 : }
811 :
812 : int
813 1 : spdk_bdev_examine(const char *name)
814 : {
815 : struct spdk_bdev *bdev;
816 : struct spdk_bdev_examine_item *item;
817 1 : struct spdk_thread *thread = spdk_get_thread();
818 :
819 1 : if (spdk_unlikely(!spdk_thread_is_app_thread(thread))) {
820 1 : SPDK_ERRLOG("Cannot examine bdev %s on thread %p (%s)\n", name, thread,
821 : thread ? spdk_thread_get_name(thread) : "null");
822 1 : return -EINVAL;
823 : }
824 :
825 0 : if (g_bdev_opts.bdev_auto_examine) {
826 0 : SPDK_ERRLOG("Manual examine is not allowed if auto examine is enabled\n");
827 0 : return -EINVAL;
828 : }
829 :
830 0 : if (bdev_examine_allowlist_check(name)) {
831 0 : SPDK_ERRLOG("Duplicate bdev name for manual examine: %s\n", name);
832 0 : return -EEXIST;
833 : }
834 :
835 0 : item = calloc(1, sizeof(*item));
836 0 : if (!item) {
837 0 : return -ENOMEM;
838 : }
839 0 : item->name = strdup(name);
840 0 : if (!item->name) {
841 0 : free(item);
842 0 : return -ENOMEM;
843 : }
844 0 : TAILQ_INSERT_TAIL(&g_bdev_examine_allowlist, item, link);
845 :
846 0 : bdev = spdk_bdev_get_by_name(name);
847 0 : if (bdev) {
848 0 : bdev_examine(bdev);
849 0 : }
850 0 : return 0;
851 1 : }
852 :
853 : static inline void
854 0 : bdev_examine_allowlist_config_json(struct spdk_json_write_ctx *w)
855 : {
856 : struct spdk_bdev_examine_item *item;
857 0 : TAILQ_FOREACH(item, &g_bdev_examine_allowlist, link) {
858 0 : spdk_json_write_object_begin(w);
859 0 : spdk_json_write_named_string(w, "method", "bdev_examine");
860 0 : spdk_json_write_named_object_begin(w, "params");
861 0 : spdk_json_write_named_string(w, "name", item->name);
862 0 : spdk_json_write_object_end(w);
863 0 : spdk_json_write_object_end(w);
864 0 : }
865 0 : }
866 :
867 : struct spdk_bdev *
868 1 : spdk_bdev_first(void)
869 : {
870 : struct spdk_bdev *bdev;
871 :
872 1 : bdev = TAILQ_FIRST(&g_bdev_mgr.bdevs);
873 1 : if (bdev) {
874 1 : SPDK_DEBUGLOG(bdev, "Starting bdev iteration at %s\n", bdev->name);
875 1 : }
876 :
877 1 : return bdev;
878 : }
879 :
880 : struct spdk_bdev *
881 8 : spdk_bdev_next(struct spdk_bdev *prev)
882 : {
883 : struct spdk_bdev *bdev;
884 :
885 8 : bdev = TAILQ_NEXT(prev, internal.link);
886 8 : if (bdev) {
887 7 : SPDK_DEBUGLOG(bdev, "Continuing bdev iteration at %s\n", bdev->name);
888 7 : }
889 :
890 8 : return bdev;
891 : }
892 :
893 : static struct spdk_bdev *
894 6 : _bdev_next_leaf(struct spdk_bdev *bdev)
895 : {
896 9 : while (bdev != NULL) {
897 8 : if (bdev->internal.claim_type == SPDK_BDEV_CLAIM_NONE) {
898 5 : return bdev;
899 : } else {
900 3 : bdev = TAILQ_NEXT(bdev, internal.link);
901 : }
902 : }
903 :
904 1 : return bdev;
905 6 : }
906 :
907 : struct spdk_bdev *
908 1 : spdk_bdev_first_leaf(void)
909 : {
910 : struct spdk_bdev *bdev;
911 :
912 1 : bdev = _bdev_next_leaf(TAILQ_FIRST(&g_bdev_mgr.bdevs));
913 :
914 1 : if (bdev) {
915 1 : SPDK_DEBUGLOG(bdev, "Starting bdev iteration at %s\n", bdev->name);
916 1 : }
917 :
918 1 : return bdev;
919 : }
920 :
921 : struct spdk_bdev *
922 5 : spdk_bdev_next_leaf(struct spdk_bdev *prev)
923 : {
924 : struct spdk_bdev *bdev;
925 :
926 5 : bdev = _bdev_next_leaf(TAILQ_NEXT(prev, internal.link));
927 :
928 5 : if (bdev) {
929 4 : SPDK_DEBUGLOG(bdev, "Continuing bdev iteration at %s\n", bdev->name);
930 4 : }
931 :
932 5 : return bdev;
933 : }
934 :
935 : static inline bool
936 816 : bdev_io_use_memory_domain(struct spdk_bdev_io *bdev_io)
937 : {
938 816 : return bdev_io->internal.f.has_memory_domain;
939 : }
940 :
941 : static inline bool
942 1553 : bdev_io_use_accel_sequence(struct spdk_bdev_io *bdev_io)
943 : {
944 1553 : return bdev_io->internal.f.has_accel_sequence;
945 : }
946 :
947 : static inline void
948 7 : bdev_queue_nomem_io_head(struct spdk_bdev_shared_resource *shared_resource,
949 : struct spdk_bdev_io *bdev_io, enum bdev_io_retry_state state)
950 : {
951 : /* Wait for some of the outstanding I/O to complete before we retry any of the nomem_io.
952 : * Normally we will wait for NOMEM_THRESHOLD_COUNT I/O to complete but for low queue depth
953 : * channels we will instead wait for half to complete.
954 : */
955 7 : shared_resource->nomem_threshold = spdk_max((int64_t)shared_resource->io_outstanding / 2,
956 : (int64_t)shared_resource->io_outstanding - NOMEM_THRESHOLD_COUNT);
957 :
958 7 : assert(state != BDEV_IO_RETRY_STATE_INVALID);
959 7 : bdev_io->internal.retry_state = state;
960 7 : TAILQ_INSERT_HEAD(&shared_resource->nomem_io, bdev_io, internal.link);
961 7 : }
962 :
963 : static inline void
964 43 : bdev_queue_nomem_io_tail(struct spdk_bdev_shared_resource *shared_resource,
965 : struct spdk_bdev_io *bdev_io, enum bdev_io_retry_state state)
966 : {
967 : /* We only queue IOs at the end of the nomem_io queue if they're submitted by the user while
968 : * the queue isn't empty, so we don't need to update the nomem_threshold here */
969 43 : assert(!TAILQ_EMPTY(&shared_resource->nomem_io));
970 :
971 43 : assert(state != BDEV_IO_RETRY_STATE_INVALID);
972 43 : bdev_io->internal.retry_state = state;
973 43 : TAILQ_INSERT_TAIL(&shared_resource->nomem_io, bdev_io, internal.link);
974 43 : }
975 :
976 : void
977 16 : spdk_bdev_io_set_buf(struct spdk_bdev_io *bdev_io, void *buf, size_t len)
978 : {
979 : struct iovec *iovs;
980 :
981 16 : if (bdev_io->u.bdev.iovs == NULL) {
982 3 : bdev_io->u.bdev.iovs = &bdev_io->iov;
983 3 : bdev_io->u.bdev.iovcnt = 1;
984 3 : }
985 :
986 16 : iovs = bdev_io->u.bdev.iovs;
987 :
988 16 : assert(iovs != NULL);
989 16 : assert(bdev_io->u.bdev.iovcnt >= 1);
990 :
991 16 : iovs[0].iov_base = buf;
992 16 : iovs[0].iov_len = len;
993 16 : }
994 :
995 : void
996 3 : spdk_bdev_io_set_md_buf(struct spdk_bdev_io *bdev_io, void *md_buf, size_t len)
997 : {
998 3 : assert((len / spdk_bdev_get_md_size(bdev_io->bdev)) >= bdev_io->u.bdev.num_blocks);
999 3 : bdev_io->u.bdev.md_buf = md_buf;
1000 3 : }
1001 :
1002 : static bool
1003 167 : _is_buf_allocated(const struct iovec *iovs)
1004 : {
1005 167 : if (iovs == NULL) {
1006 6 : return false;
1007 : }
1008 :
1009 161 : return iovs[0].iov_base != NULL;
1010 167 : }
1011 :
1012 : static bool
1013 50 : _are_iovs_aligned(struct iovec *iovs, int iovcnt, uint32_t alignment)
1014 : {
1015 : int i;
1016 : uintptr_t iov_base;
1017 :
1018 50 : if (spdk_likely(alignment == 1)) {
1019 21 : return true;
1020 : }
1021 :
1022 36 : for (i = 0; i < iovcnt; i++) {
1023 29 : iov_base = (uintptr_t)iovs[i].iov_base;
1024 29 : if ((iov_base & (alignment - 1)) != 0) {
1025 22 : return false;
1026 : }
1027 7 : }
1028 :
1029 7 : return true;
1030 50 : }
1031 :
1032 : static inline bool
1033 856 : bdev_io_needs_sequence_exec(struct spdk_bdev_desc *desc, struct spdk_bdev_io *bdev_io)
1034 : {
1035 856 : if (!bdev_io_use_accel_sequence(bdev_io)) {
1036 856 : return false;
1037 : }
1038 :
1039 : /* For now, we don't allow splitting IOs with an accel sequence and will treat them as if
1040 : * bdev module didn't support accel sequences */
1041 0 : return !desc->accel_sequence_supported[bdev_io->type] || bdev_io->internal.f.split;
1042 856 : }
1043 :
1044 : static inline void
1045 592 : bdev_io_increment_outstanding(struct spdk_bdev_channel *bdev_ch,
1046 : struct spdk_bdev_shared_resource *shared_resource)
1047 : {
1048 592 : bdev_ch->io_outstanding++;
1049 592 : shared_resource->io_outstanding++;
1050 592 : }
1051 :
1052 : static inline void
1053 592 : bdev_io_decrement_outstanding(struct spdk_bdev_channel *bdev_ch,
1054 : struct spdk_bdev_shared_resource *shared_resource)
1055 : {
1056 592 : assert(bdev_ch->io_outstanding > 0);
1057 592 : assert(shared_resource->io_outstanding > 0);
1058 592 : bdev_ch->io_outstanding--;
1059 592 : shared_resource->io_outstanding--;
1060 592 : }
1061 :
1062 : static void
1063 0 : bdev_io_submit_sequence_cb(void *ctx, int status)
1064 : {
1065 0 : struct spdk_bdev_io *bdev_io = ctx;
1066 :
1067 0 : assert(bdev_io_use_accel_sequence(bdev_io));
1068 :
1069 0 : bdev_io->u.bdev.accel_sequence = NULL;
1070 0 : bdev_io->internal.f.has_accel_sequence = false;
1071 :
1072 0 : if (spdk_unlikely(status != 0)) {
1073 0 : SPDK_ERRLOG("Failed to execute accel sequence, status=%d\n", status);
1074 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
1075 0 : bdev_io_complete_unsubmitted(bdev_io);
1076 0 : return;
1077 : }
1078 :
1079 0 : bdev_io_submit(bdev_io);
1080 0 : }
1081 :
1082 : static void
1083 0 : bdev_io_exec_sequence_cb(void *ctx, int status)
1084 : {
1085 0 : struct spdk_bdev_io *bdev_io = ctx;
1086 0 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1087 :
1088 0 : TAILQ_REMOVE(&bdev_io->internal.ch->io_accel_exec, bdev_io, internal.link);
1089 0 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1090 :
1091 0 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1092 0 : bdev_ch_retry_io(ch);
1093 0 : }
1094 :
1095 0 : bdev_io->internal.data_transfer_cpl(bdev_io, status);
1096 0 : }
1097 :
1098 : static void
1099 0 : bdev_io_exec_sequence(struct spdk_bdev_io *bdev_io, void (*cb_fn)(void *ctx, int status))
1100 : {
1101 0 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1102 :
1103 0 : assert(bdev_io_needs_sequence_exec(bdev_io->internal.desc, bdev_io));
1104 0 : assert(bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE || bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
1105 0 : assert(bdev_io_use_accel_sequence(bdev_io));
1106 :
1107 : /* Since the operations are appended during submission, they're in the opposite order than
1108 : * how we want to execute them for reads (i.e. we need to execute the most recently added
1109 : * operation first), so reverse the sequence before executing it.
1110 : */
1111 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_READ) {
1112 0 : spdk_accel_sequence_reverse(bdev_io->internal.accel_sequence);
1113 0 : }
1114 :
1115 0 : TAILQ_INSERT_TAIL(&bdev_io->internal.ch->io_accel_exec, bdev_io, internal.link);
1116 0 : bdev_io_increment_outstanding(ch, ch->shared_resource);
1117 0 : bdev_io->internal.data_transfer_cpl = cb_fn;
1118 :
1119 0 : spdk_accel_sequence_finish(bdev_io->internal.accel_sequence,
1120 0 : bdev_io_exec_sequence_cb, bdev_io);
1121 0 : }
1122 :
1123 : static void
1124 42 : bdev_io_get_buf_complete(struct spdk_bdev_io *bdev_io, bool status)
1125 : {
1126 42 : struct spdk_io_channel *ch = spdk_bdev_io_get_io_channel(bdev_io);
1127 : void *buf;
1128 :
1129 42 : if (spdk_unlikely(bdev_io->internal.get_aux_buf_cb != NULL)) {
1130 0 : buf = bdev_io->internal.buf.ptr;
1131 0 : bdev_io->internal.buf.ptr = NULL;
1132 0 : bdev_io->internal.f.has_buf = false;
1133 0 : bdev_io->internal.get_aux_buf_cb(ch, bdev_io, buf);
1134 0 : bdev_io->internal.get_aux_buf_cb = NULL;
1135 0 : } else {
1136 42 : assert(bdev_io->internal.get_buf_cb != NULL);
1137 42 : bdev_io->internal.get_buf_cb(ch, bdev_io, status);
1138 42 : bdev_io->internal.get_buf_cb = NULL;
1139 : }
1140 42 : }
1141 :
1142 : static void
1143 4 : _bdev_io_pull_buffer_cpl(void *ctx, int rc)
1144 : {
1145 4 : struct spdk_bdev_io *bdev_io = ctx;
1146 :
1147 4 : if (rc) {
1148 0 : SPDK_ERRLOG("Set bounce buffer failed with rc %d\n", rc);
1149 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
1150 0 : }
1151 4 : bdev_io_get_buf_complete(bdev_io, !rc);
1152 4 : }
1153 :
1154 : static void
1155 2 : bdev_io_pull_md_buf_done(void *ctx, int status)
1156 : {
1157 2 : struct spdk_bdev_io *bdev_io = ctx;
1158 2 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1159 :
1160 2 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1161 2 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1162 :
1163 2 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1164 0 : bdev_ch_retry_io(ch);
1165 0 : }
1166 :
1167 2 : assert(bdev_io->internal.data_transfer_cpl);
1168 2 : bdev_io->internal.data_transfer_cpl(bdev_io, status);
1169 2 : }
1170 :
1171 : static void
1172 4 : bdev_io_pull_md_buf(struct spdk_bdev_io *bdev_io)
1173 : {
1174 4 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1175 4 : int rc = 0;
1176 :
1177 4 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
1178 2 : assert(bdev_io->internal.f.has_bounce_buf);
1179 2 : if (bdev_io_use_memory_domain(bdev_io)) {
1180 2 : TAILQ_INSERT_TAIL(&ch->io_memory_domain, bdev_io, internal.link);
1181 2 : bdev_io_increment_outstanding(ch, ch->shared_resource);
1182 4 : rc = spdk_memory_domain_pull_data(bdev_io->internal.memory_domain,
1183 2 : bdev_io->internal.memory_domain_ctx,
1184 2 : &bdev_io->internal.bounce_buf.orig_md_iov, 1,
1185 2 : &bdev_io->internal.bounce_buf.md_iov, 1,
1186 2 : bdev_io_pull_md_buf_done, bdev_io);
1187 2 : if (rc == 0) {
1188 : /* Continue to submit IO in completion callback */
1189 2 : return;
1190 : }
1191 0 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1192 0 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1193 0 : if (rc != -ENOMEM) {
1194 0 : SPDK_ERRLOG("Failed to pull data from memory domain %s, rc %d\n",
1195 : spdk_memory_domain_get_dma_device_id(
1196 : bdev_io->internal.memory_domain), rc);
1197 0 : }
1198 0 : } else {
1199 0 : memcpy(bdev_io->internal.bounce_buf.md_iov.iov_base,
1200 0 : bdev_io->internal.bounce_buf.orig_md_iov.iov_base,
1201 0 : bdev_io->internal.bounce_buf.orig_md_iov.iov_len);
1202 : }
1203 0 : }
1204 :
1205 2 : if (spdk_unlikely(rc == -ENOMEM)) {
1206 0 : bdev_queue_nomem_io_head(ch->shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PULL_MD);
1207 0 : } else {
1208 2 : assert(bdev_io->internal.data_transfer_cpl);
1209 2 : bdev_io->internal.data_transfer_cpl(bdev_io, rc);
1210 : }
1211 4 : }
1212 :
1213 : static void
1214 4 : _bdev_io_pull_bounce_md_buf(struct spdk_bdev_io *bdev_io, void *md_buf, size_t len)
1215 : {
1216 4 : assert(bdev_io->internal.f.has_bounce_buf);
1217 :
1218 : /* save original md_buf */
1219 4 : bdev_io->internal.bounce_buf.orig_md_iov.iov_base = bdev_io->u.bdev.md_buf;
1220 4 : bdev_io->internal.bounce_buf.orig_md_iov.iov_len = len;
1221 4 : bdev_io->internal.bounce_buf.md_iov.iov_base = md_buf;
1222 4 : bdev_io->internal.bounce_buf.md_iov.iov_len = len;
1223 : /* set bounce md_buf */
1224 4 : bdev_io->u.bdev.md_buf = md_buf;
1225 :
1226 4 : bdev_io_pull_md_buf(bdev_io);
1227 4 : }
1228 :
1229 : static void
1230 42 : _bdev_io_set_md_buf(struct spdk_bdev_io *bdev_io)
1231 : {
1232 42 : struct spdk_bdev *bdev = bdev_io->bdev;
1233 : uint64_t md_len;
1234 : void *buf;
1235 :
1236 42 : if (spdk_bdev_is_md_separate(bdev)) {
1237 7 : assert(!bdev_io_use_accel_sequence(bdev_io));
1238 :
1239 7 : buf = (char *)bdev_io->u.bdev.iovs[0].iov_base + bdev_io->u.bdev.iovs[0].iov_len;
1240 7 : md_len = bdev_io->u.bdev.num_blocks * bdev->md_len;
1241 :
1242 7 : assert(((uintptr_t)buf & (spdk_bdev_get_buf_align(bdev) - 1)) == 0);
1243 :
1244 7 : if (bdev_io->u.bdev.md_buf != NULL) {
1245 4 : _bdev_io_pull_bounce_md_buf(bdev_io, buf, md_len);
1246 4 : return;
1247 : } else {
1248 3 : spdk_bdev_io_set_md_buf(bdev_io, buf, md_len);
1249 : }
1250 3 : }
1251 :
1252 38 : bdev_io_get_buf_complete(bdev_io, true);
1253 42 : }
1254 :
1255 : static inline void
1256 26 : bdev_io_pull_data_done(struct spdk_bdev_io *bdev_io, int rc)
1257 : {
1258 26 : if (rc) {
1259 0 : SPDK_ERRLOG("Failed to get data buffer\n");
1260 0 : assert(bdev_io->internal.data_transfer_cpl);
1261 0 : bdev_io->internal.data_transfer_cpl(bdev_io, rc);
1262 0 : return;
1263 : }
1264 :
1265 26 : _bdev_io_set_md_buf(bdev_io);
1266 26 : }
1267 :
1268 : static void
1269 2 : bdev_io_pull_data_done_and_track(void *ctx, int status)
1270 : {
1271 2 : struct spdk_bdev_io *bdev_io = ctx;
1272 2 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1273 :
1274 2 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1275 2 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1276 :
1277 2 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1278 0 : bdev_ch_retry_io(ch);
1279 0 : }
1280 :
1281 2 : bdev_io_pull_data_done(bdev_io, status);
1282 2 : }
1283 :
1284 : static void
1285 27 : bdev_io_pull_data(struct spdk_bdev_io *bdev_io)
1286 : {
1287 27 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1288 27 : int rc = 0;
1289 :
1290 : /* If we need to exec an accel sequence or the IO uses a memory domain buffer and has a
1291 : * sequence, append a copy operation making accel change the src/dst buffers of the previous
1292 : * operation */
1293 27 : if (bdev_io_needs_sequence_exec(bdev_io->internal.desc, bdev_io) ||
1294 27 : (bdev_io_use_accel_sequence(bdev_io) && bdev_io_use_memory_domain(bdev_io))) {
1295 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
1296 0 : assert(bdev_io_use_accel_sequence(bdev_io));
1297 0 : assert(bdev_io->internal.f.has_bounce_buf);
1298 0 : rc = spdk_accel_append_copy(&bdev_io->internal.accel_sequence, ch->accel_channel,
1299 0 : bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt,
1300 : NULL, NULL,
1301 0 : bdev_io->internal.bounce_buf.orig_iovs,
1302 0 : bdev_io->internal.bounce_buf.orig_iovcnt,
1303 0 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain : NULL,
1304 0 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain_ctx : NULL,
1305 : NULL, NULL);
1306 0 : } else {
1307 : /* We need to reverse the src/dst for reads */
1308 0 : assert(bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
1309 0 : assert(bdev_io_use_accel_sequence(bdev_io));
1310 0 : assert(bdev_io->internal.f.has_bounce_buf);
1311 0 : rc = spdk_accel_append_copy(&bdev_io->internal.accel_sequence, ch->accel_channel,
1312 0 : bdev_io->internal.bounce_buf.orig_iovs,
1313 0 : bdev_io->internal.bounce_buf.orig_iovcnt,
1314 0 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain : NULL,
1315 0 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain_ctx : NULL,
1316 0 : bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt,
1317 : NULL, NULL, NULL, NULL);
1318 : }
1319 :
1320 0 : if (spdk_unlikely(rc != 0 && rc != -ENOMEM)) {
1321 0 : SPDK_ERRLOG("Failed to append copy to accel sequence: %p\n",
1322 : bdev_io->internal.accel_sequence);
1323 0 : }
1324 27 : } else if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
1325 : /* if this is write path, copy data from original buffer to bounce buffer */
1326 17 : if (bdev_io_use_memory_domain(bdev_io)) {
1327 3 : assert(bdev_io->internal.f.has_bounce_buf);
1328 3 : TAILQ_INSERT_TAIL(&ch->io_memory_domain, bdev_io, internal.link);
1329 3 : bdev_io_increment_outstanding(ch, ch->shared_resource);
1330 6 : rc = spdk_memory_domain_pull_data(bdev_io->internal.memory_domain,
1331 3 : bdev_io->internal.memory_domain_ctx,
1332 3 : bdev_io->internal.bounce_buf.orig_iovs,
1333 3 : (uint32_t)bdev_io->internal.bounce_buf.orig_iovcnt,
1334 3 : bdev_io->u.bdev.iovs, 1,
1335 : bdev_io_pull_data_done_and_track,
1336 3 : bdev_io);
1337 3 : if (rc == 0) {
1338 : /* Continue to submit IO in completion callback */
1339 2 : return;
1340 : }
1341 1 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1342 1 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1343 1 : if (rc != -ENOMEM) {
1344 0 : SPDK_ERRLOG("Failed to pull data from memory domain %s\n",
1345 : spdk_memory_domain_get_dma_device_id(
1346 : bdev_io->internal.memory_domain));
1347 0 : }
1348 1 : } else {
1349 14 : assert(bdev_io->u.bdev.iovcnt == 1);
1350 14 : assert(bdev_io->internal.f.has_bounce_buf);
1351 28 : spdk_copy_iovs_to_buf(bdev_io->u.bdev.iovs[0].iov_base,
1352 14 : bdev_io->u.bdev.iovs[0].iov_len,
1353 14 : bdev_io->internal.bounce_buf.orig_iovs,
1354 14 : bdev_io->internal.bounce_buf.orig_iovcnt);
1355 : }
1356 15 : }
1357 :
1358 25 : if (spdk_unlikely(rc == -ENOMEM)) {
1359 1 : bdev_queue_nomem_io_head(ch->shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PULL);
1360 1 : } else {
1361 24 : bdev_io_pull_data_done(bdev_io, rc);
1362 : }
1363 27 : }
1364 :
1365 : static void
1366 26 : _bdev_io_pull_bounce_data_buf(struct spdk_bdev_io *bdev_io, void *buf, size_t len,
1367 : bdev_copy_bounce_buffer_cpl cpl_cb)
1368 : {
1369 26 : struct spdk_bdev_shared_resource *shared_resource = bdev_io->internal.ch->shared_resource;
1370 :
1371 26 : assert(bdev_io->internal.f.has_bounce_buf == false);
1372 :
1373 26 : bdev_io->internal.data_transfer_cpl = cpl_cb;
1374 26 : bdev_io->internal.f.has_bounce_buf = true;
1375 : /* save original iovec */
1376 26 : bdev_io->internal.bounce_buf.orig_iovs = bdev_io->u.bdev.iovs;
1377 26 : bdev_io->internal.bounce_buf.orig_iovcnt = bdev_io->u.bdev.iovcnt;
1378 : /* zero the other data members */
1379 26 : bdev_io->internal.bounce_buf.iov.iov_base = NULL;
1380 26 : bdev_io->internal.bounce_buf.md_iov.iov_base = NULL;
1381 26 : bdev_io->internal.bounce_buf.orig_md_iov.iov_base = NULL;
1382 : /* set bounce iov */
1383 26 : bdev_io->u.bdev.iovs = &bdev_io->internal.bounce_buf.iov;
1384 26 : bdev_io->u.bdev.iovcnt = 1;
1385 : /* set bounce buffer for this operation */
1386 26 : bdev_io->u.bdev.iovs[0].iov_base = buf;
1387 26 : bdev_io->u.bdev.iovs[0].iov_len = len;
1388 : /* Now we use 1 iov, the split condition could have been changed */
1389 26 : bdev_io->internal.f.split = bdev_io_should_split(bdev_io);
1390 :
1391 26 : if (spdk_unlikely(!TAILQ_EMPTY(&shared_resource->nomem_io))) {
1392 0 : bdev_queue_nomem_io_tail(shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PULL);
1393 0 : } else {
1394 26 : bdev_io_pull_data(bdev_io);
1395 : }
1396 26 : }
1397 :
1398 : static void
1399 42 : _bdev_io_set_buf(struct spdk_bdev_io *bdev_io, void *buf, uint64_t len)
1400 : {
1401 42 : struct spdk_bdev *bdev = bdev_io->bdev;
1402 : bool buf_allocated;
1403 : uint64_t alignment;
1404 : void *aligned_buf;
1405 :
1406 42 : bdev_io->internal.buf.ptr = buf;
1407 42 : bdev_io->internal.f.has_buf = true;
1408 :
1409 42 : if (spdk_unlikely(bdev_io->internal.get_aux_buf_cb != NULL)) {
1410 0 : bdev_io_get_buf_complete(bdev_io, true);
1411 0 : return;
1412 : }
1413 :
1414 42 : alignment = spdk_bdev_get_buf_align(bdev);
1415 42 : buf_allocated = _is_buf_allocated(bdev_io->u.bdev.iovs);
1416 42 : aligned_buf = (void *)(((uintptr_t)buf + (alignment - 1)) & ~(alignment - 1));
1417 :
1418 42 : if (buf_allocated) {
1419 26 : _bdev_io_pull_bounce_data_buf(bdev_io, aligned_buf, len, _bdev_io_pull_buffer_cpl);
1420 : /* Continue in completion callback */
1421 26 : return;
1422 : } else {
1423 16 : spdk_bdev_io_set_buf(bdev_io, aligned_buf, len);
1424 : }
1425 :
1426 16 : _bdev_io_set_md_buf(bdev_io);
1427 42 : }
1428 :
1429 : static inline uint64_t
1430 84 : bdev_io_get_max_buf_len(struct spdk_bdev_io *bdev_io, uint64_t len)
1431 : {
1432 84 : struct spdk_bdev *bdev = bdev_io->bdev;
1433 : uint64_t md_len, alignment;
1434 :
1435 84 : md_len = spdk_bdev_is_md_separate(bdev) ? bdev_io->u.bdev.num_blocks * bdev->md_len : 0;
1436 :
1437 : /* 1 byte alignment needs 0 byte of extra space, 64 bytes alignment needs 63 bytes of extra space, etc. */
1438 84 : alignment = spdk_bdev_get_buf_align(bdev) - 1;
1439 :
1440 84 : return len + alignment + md_len;
1441 : }
1442 :
1443 : static void
1444 42 : _bdev_io_put_buf(struct spdk_bdev_io *bdev_io, void *buf, uint64_t buf_len)
1445 : {
1446 : struct spdk_bdev_mgmt_channel *ch;
1447 :
1448 42 : ch = bdev_io->internal.ch->shared_resource->mgmt_ch;
1449 42 : spdk_iobuf_put(&ch->iobuf, buf, bdev_io_get_max_buf_len(bdev_io, buf_len));
1450 42 : }
1451 :
1452 : static void
1453 42 : bdev_io_put_buf(struct spdk_bdev_io *bdev_io)
1454 : {
1455 42 : assert(bdev_io->internal.f.has_buf);
1456 42 : _bdev_io_put_buf(bdev_io, bdev_io->internal.buf.ptr, bdev_io->internal.buf.len);
1457 42 : bdev_io->internal.buf.ptr = NULL;
1458 42 : bdev_io->internal.f.has_buf = false;
1459 42 : }
1460 :
1461 3 : SPDK_LOG_DEPRECATION_REGISTER(spdk_bdev_io_put_aux_buf,
1462 : "spdk_bdev_io_put_aux_buf is deprecated", "v25.01", 0);
1463 :
1464 : void
1465 0 : spdk_bdev_io_put_aux_buf(struct spdk_bdev_io *bdev_io, void *buf)
1466 : {
1467 0 : uint64_t len = bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen;
1468 :
1469 0 : SPDK_LOG_DEPRECATED(spdk_bdev_io_put_aux_buf);
1470 :
1471 0 : assert(buf != NULL);
1472 0 : _bdev_io_put_buf(bdev_io, buf, len);
1473 0 : }
1474 :
1475 : static inline void
1476 548 : bdev_submit_request(struct spdk_bdev *bdev, struct spdk_io_channel *ioch,
1477 : struct spdk_bdev_io *bdev_io)
1478 : {
1479 : /* After a request is submitted to a bdev module, the ownership of an accel sequence
1480 : * associated with that bdev_io is transferred to the bdev module. So, clear the internal
1481 : * sequence pointer to make sure we won't touch it anymore. */
1482 1014 : if ((bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE ||
1483 548 : bdev_io->type == SPDK_BDEV_IO_TYPE_READ) && bdev_io->u.bdev.accel_sequence != NULL) {
1484 0 : assert(!bdev_io_needs_sequence_exec(bdev_io->internal.desc, bdev_io));
1485 0 : bdev_io->internal.f.has_accel_sequence = false;
1486 0 : }
1487 :
1488 548 : bdev->fn_table->submit_request(ioch, bdev_io);
1489 548 : }
1490 :
1491 : static inline void
1492 10 : bdev_ch_resubmit_io(struct spdk_bdev_shared_resource *shared_resource, struct spdk_bdev_io *bdev_io)
1493 : {
1494 10 : struct spdk_bdev *bdev = bdev_io->bdev;
1495 :
1496 10 : bdev_io_increment_outstanding(bdev_io->internal.ch, shared_resource);
1497 10 : bdev_io->internal.error.nvme.cdw0 = 0;
1498 10 : bdev_io->num_retries++;
1499 10 : bdev_submit_request(bdev, spdk_bdev_io_get_io_channel(bdev_io), bdev_io);
1500 10 : }
1501 :
1502 : static void
1503 63 : bdev_shared_ch_retry_io(struct spdk_bdev_shared_resource *shared_resource)
1504 : {
1505 : struct spdk_bdev_io *bdev_io;
1506 :
1507 63 : if (shared_resource->io_outstanding > shared_resource->nomem_threshold) {
1508 : /*
1509 : * Allow some more I/O to complete before retrying the nomem_io queue.
1510 : * Some drivers (such as nvme) cannot immediately take a new I/O in
1511 : * the context of a completion, because the resources for the I/O are
1512 : * not released until control returns to the bdev poller. Also, we
1513 : * may require several small I/O to complete before a larger I/O
1514 : * (that requires splitting) can be submitted.
1515 : */
1516 58 : return;
1517 : }
1518 :
1519 16 : while (!TAILQ_EMPTY(&shared_resource->nomem_io)) {
1520 12 : bdev_io = TAILQ_FIRST(&shared_resource->nomem_io);
1521 12 : TAILQ_REMOVE(&shared_resource->nomem_io, bdev_io, internal.link);
1522 :
1523 12 : switch (bdev_io->internal.retry_state) {
1524 : case BDEV_IO_RETRY_STATE_SUBMIT:
1525 10 : bdev_ch_resubmit_io(shared_resource, bdev_io);
1526 10 : break;
1527 : case BDEV_IO_RETRY_STATE_PULL:
1528 1 : bdev_io_pull_data(bdev_io);
1529 1 : break;
1530 : case BDEV_IO_RETRY_STATE_PULL_MD:
1531 0 : bdev_io_pull_md_buf(bdev_io);
1532 0 : break;
1533 : case BDEV_IO_RETRY_STATE_PUSH:
1534 1 : bdev_io_push_bounce_data(bdev_io);
1535 1 : break;
1536 : case BDEV_IO_RETRY_STATE_PUSH_MD:
1537 0 : bdev_io_push_bounce_md_buf(bdev_io);
1538 0 : break;
1539 : default:
1540 0 : assert(0 && "invalid retry state");
1541 : break;
1542 : }
1543 :
1544 12 : if (bdev_io == TAILQ_FIRST(&shared_resource->nomem_io)) {
1545 : /* This IO completed again with NOMEM status, so break the loop and
1546 : * don't try anymore. Note that a bdev_io that fails with NOMEM
1547 : * always gets requeued at the front of the list, to maintain
1548 : * ordering.
1549 : */
1550 1 : break;
1551 : }
1552 : }
1553 63 : }
1554 :
1555 : static void
1556 63 : bdev_ch_retry_io(struct spdk_bdev_channel *bdev_ch)
1557 : {
1558 63 : bdev_shared_ch_retry_io(bdev_ch->shared_resource);
1559 63 : }
1560 :
1561 : static int
1562 0 : bdev_no_mem_poller(void *ctx)
1563 : {
1564 0 : struct spdk_bdev_shared_resource *shared_resource = ctx;
1565 :
1566 0 : spdk_poller_unregister(&shared_resource->nomem_poller);
1567 :
1568 0 : if (!TAILQ_EMPTY(&shared_resource->nomem_io)) {
1569 0 : bdev_shared_ch_retry_io(shared_resource);
1570 0 : }
1571 : /* the retry cb may re-register the poller so double check */
1572 0 : if (!TAILQ_EMPTY(&shared_resource->nomem_io) &&
1573 0 : shared_resource->io_outstanding == 0 && shared_resource->nomem_poller == NULL) {
1574 : /* No IOs were submitted, try again */
1575 0 : shared_resource->nomem_poller = SPDK_POLLER_REGISTER(bdev_no_mem_poller, shared_resource,
1576 : SPDK_BDEV_IO_POLL_INTERVAL_IN_MSEC * 10);
1577 0 : }
1578 :
1579 0 : return SPDK_POLLER_BUSY;
1580 : }
1581 :
1582 : static inline bool
1583 556 : _bdev_io_handle_no_mem(struct spdk_bdev_io *bdev_io, enum bdev_io_retry_state state)
1584 : {
1585 556 : struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
1586 556 : struct spdk_bdev_shared_resource *shared_resource = bdev_ch->shared_resource;
1587 :
1588 556 : if (spdk_unlikely(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_NOMEM)) {
1589 5 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_PENDING;
1590 5 : bdev_queue_nomem_io_head(shared_resource, bdev_io, state);
1591 :
1592 5 : if (shared_resource->io_outstanding == 0 && !shared_resource->nomem_poller) {
1593 : /* Special case when we have nomem IOs and no outstanding IOs which completions
1594 : * could trigger retry of queued IOs
1595 : * Any IOs submitted may trigger retry of queued IOs. This poller handles a case when no
1596 : * new IOs submitted, e.g. qd==1 */
1597 0 : shared_resource->nomem_poller = SPDK_POLLER_REGISTER(bdev_no_mem_poller, shared_resource,
1598 : SPDK_BDEV_IO_POLL_INTERVAL_IN_MSEC * 10);
1599 0 : }
1600 : /* If bdev module completed an I/O that has an accel sequence with NOMEM status, the
1601 : * ownership of that sequence is transferred back to the bdev layer, so we need to
1602 : * restore internal.accel_sequence to make sure that the sequence is handled
1603 : * correctly in case the I/O is later aborted. */
1604 5 : if ((bdev_io->type == SPDK_BDEV_IO_TYPE_READ ||
1605 5 : bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) && bdev_io->u.bdev.accel_sequence) {
1606 0 : assert(!bdev_io_use_accel_sequence(bdev_io));
1607 0 : bdev_io->internal.f.has_accel_sequence = true;
1608 0 : bdev_io->internal.accel_sequence = bdev_io->u.bdev.accel_sequence;
1609 0 : }
1610 :
1611 5 : return true;
1612 : }
1613 :
1614 551 : if (spdk_unlikely(!TAILQ_EMPTY(&shared_resource->nomem_io))) {
1615 63 : bdev_ch_retry_io(bdev_ch);
1616 63 : }
1617 :
1618 551 : return false;
1619 556 : }
1620 :
1621 : static void
1622 26 : _bdev_io_complete_push_bounce_done(void *ctx, int rc)
1623 : {
1624 26 : struct spdk_bdev_io *bdev_io = ctx;
1625 26 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1626 :
1627 26 : if (rc) {
1628 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
1629 0 : }
1630 : /* We want to free the bounce buffer here since we know we're done with it (as opposed
1631 : * to waiting for the conditional free of internal.buf.ptr in spdk_bdev_free_io()).
1632 : */
1633 26 : bdev_io_put_buf(bdev_io);
1634 :
1635 26 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1636 0 : bdev_ch_retry_io(ch);
1637 0 : }
1638 :
1639 : /* Continue with IO completion flow */
1640 26 : bdev_io_complete(bdev_io);
1641 26 : }
1642 :
1643 : static void
1644 2 : bdev_io_push_bounce_md_buf_done(void *ctx, int rc)
1645 : {
1646 2 : struct spdk_bdev_io *bdev_io = ctx;
1647 2 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1648 :
1649 2 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1650 2 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1651 2 : bdev_io->internal.f.has_bounce_buf = false;
1652 :
1653 2 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1654 0 : bdev_ch_retry_io(ch);
1655 0 : }
1656 :
1657 2 : bdev_io->internal.data_transfer_cpl(bdev_io, rc);
1658 2 : }
1659 :
1660 : static inline void
1661 26 : bdev_io_push_bounce_md_buf(struct spdk_bdev_io *bdev_io)
1662 : {
1663 26 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1664 26 : int rc = 0;
1665 :
1666 26 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS);
1667 26 : assert(bdev_io->internal.f.has_bounce_buf);
1668 :
1669 : /* do the same for metadata buffer */
1670 26 : if (spdk_unlikely(bdev_io->internal.bounce_buf.orig_md_iov.iov_base != NULL)) {
1671 4 : assert(spdk_bdev_is_md_separate(bdev_io->bdev));
1672 :
1673 4 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_READ) {
1674 2 : if (bdev_io_use_memory_domain(bdev_io)) {
1675 2 : TAILQ_INSERT_TAIL(&ch->io_memory_domain, bdev_io, internal.link);
1676 2 : bdev_io_increment_outstanding(ch, ch->shared_resource);
1677 : /* If memory domain is used then we need to call async push function */
1678 4 : rc = spdk_memory_domain_push_data(bdev_io->internal.memory_domain,
1679 2 : bdev_io->internal.memory_domain_ctx,
1680 2 : &bdev_io->internal.bounce_buf.orig_md_iov,
1681 2 : (uint32_t)bdev_io->internal.bounce_buf.orig_iovcnt,
1682 2 : &bdev_io->internal.bounce_buf.md_iov, 1,
1683 : bdev_io_push_bounce_md_buf_done,
1684 2 : bdev_io);
1685 2 : if (rc == 0) {
1686 : /* Continue IO completion in async callback */
1687 2 : return;
1688 : }
1689 0 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1690 0 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1691 0 : if (rc != -ENOMEM) {
1692 0 : SPDK_ERRLOG("Failed to push md to memory domain %s\n",
1693 : spdk_memory_domain_get_dma_device_id(
1694 : bdev_io->internal.memory_domain));
1695 0 : }
1696 0 : } else {
1697 0 : memcpy(bdev_io->internal.bounce_buf.orig_md_iov.iov_base, bdev_io->u.bdev.md_buf,
1698 0 : bdev_io->internal.bounce_buf.orig_md_iov.iov_len);
1699 : }
1700 0 : }
1701 2 : }
1702 :
1703 24 : if (spdk_unlikely(rc == -ENOMEM)) {
1704 0 : bdev_queue_nomem_io_head(ch->shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PUSH_MD);
1705 0 : } else {
1706 24 : assert(bdev_io->internal.data_transfer_cpl);
1707 24 : bdev_io->internal.f.has_bounce_buf = false;
1708 24 : bdev_io->internal.data_transfer_cpl(bdev_io, rc);
1709 : }
1710 26 : }
1711 :
1712 : static inline void
1713 26 : bdev_io_push_bounce_data_done(struct spdk_bdev_io *bdev_io, int rc)
1714 : {
1715 26 : assert(bdev_io->internal.data_transfer_cpl);
1716 26 : if (rc) {
1717 0 : bdev_io->internal.data_transfer_cpl(bdev_io, rc);
1718 0 : return;
1719 : }
1720 :
1721 : /* set original buffer for this io */
1722 26 : bdev_io->u.bdev.iovcnt = bdev_io->internal.bounce_buf.orig_iovcnt;
1723 26 : bdev_io->u.bdev.iovs = bdev_io->internal.bounce_buf.orig_iovs;
1724 :
1725 : /* We don't set bdev_io->internal.f.has_bounce_buf to false here because
1726 : * we still need to clear the md buf */
1727 :
1728 26 : bdev_io_push_bounce_md_buf(bdev_io);
1729 26 : }
1730 :
1731 : static void
1732 2 : bdev_io_push_bounce_data_done_and_track(void *ctx, int status)
1733 : {
1734 2 : struct spdk_bdev_io *bdev_io = ctx;
1735 2 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1736 :
1737 2 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1738 2 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1739 :
1740 2 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1741 0 : bdev_ch_retry_io(ch);
1742 0 : }
1743 :
1744 2 : bdev_io_push_bounce_data_done(bdev_io, status);
1745 2 : }
1746 :
1747 : static inline void
1748 27 : bdev_io_push_bounce_data(struct spdk_bdev_io *bdev_io)
1749 : {
1750 27 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1751 27 : int rc = 0;
1752 :
1753 27 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS);
1754 27 : assert(!bdev_io_use_accel_sequence(bdev_io));
1755 27 : assert(bdev_io->internal.f.has_bounce_buf);
1756 :
1757 : /* if this is read path, copy data from bounce buffer to original buffer */
1758 27 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_READ) {
1759 11 : if (bdev_io_use_memory_domain(bdev_io)) {
1760 3 : TAILQ_INSERT_TAIL(&ch->io_memory_domain, bdev_io, internal.link);
1761 3 : bdev_io_increment_outstanding(ch, ch->shared_resource);
1762 : /* If memory domain is used then we need to call async push function */
1763 6 : rc = spdk_memory_domain_push_data(bdev_io->internal.memory_domain,
1764 3 : bdev_io->internal.memory_domain_ctx,
1765 3 : bdev_io->internal.bounce_buf.orig_iovs,
1766 3 : (uint32_t)bdev_io->internal.bounce_buf.orig_iovcnt,
1767 3 : &bdev_io->internal.bounce_buf.iov, 1,
1768 : bdev_io_push_bounce_data_done_and_track,
1769 3 : bdev_io);
1770 3 : if (rc == 0) {
1771 : /* Continue IO completion in async callback */
1772 2 : return;
1773 : }
1774 :
1775 1 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1776 1 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1777 1 : if (rc != -ENOMEM) {
1778 0 : SPDK_ERRLOG("Failed to push data to memory domain %s\n",
1779 : spdk_memory_domain_get_dma_device_id(
1780 : bdev_io->internal.memory_domain));
1781 0 : }
1782 1 : } else {
1783 16 : spdk_copy_buf_to_iovs(bdev_io->internal.bounce_buf.orig_iovs,
1784 8 : bdev_io->internal.bounce_buf.orig_iovcnt,
1785 8 : bdev_io->internal.bounce_buf.iov.iov_base,
1786 8 : bdev_io->internal.bounce_buf.iov.iov_len);
1787 : }
1788 9 : }
1789 :
1790 25 : if (spdk_unlikely(rc == -ENOMEM)) {
1791 1 : bdev_queue_nomem_io_head(ch->shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PUSH);
1792 1 : } else {
1793 24 : bdev_io_push_bounce_data_done(bdev_io, rc);
1794 : }
1795 27 : }
1796 :
1797 : static inline void
1798 26 : _bdev_io_push_bounce_data_buffer(struct spdk_bdev_io *bdev_io, bdev_copy_bounce_buffer_cpl cpl_cb)
1799 : {
1800 26 : bdev_io->internal.data_transfer_cpl = cpl_cb;
1801 26 : bdev_io_push_bounce_data(bdev_io);
1802 26 : }
1803 :
1804 : static void
1805 0 : bdev_io_get_iobuf_cb(struct spdk_iobuf_entry *iobuf, void *buf)
1806 : {
1807 : struct spdk_bdev_io *bdev_io;
1808 :
1809 0 : bdev_io = SPDK_CONTAINEROF(iobuf, struct spdk_bdev_io, internal.iobuf);
1810 0 : _bdev_io_set_buf(bdev_io, buf, bdev_io->internal.buf.len);
1811 0 : }
1812 :
1813 : static void
1814 42 : bdev_io_get_buf(struct spdk_bdev_io *bdev_io, uint64_t len)
1815 : {
1816 : struct spdk_bdev_mgmt_channel *mgmt_ch;
1817 : uint64_t max_len;
1818 : void *buf;
1819 :
1820 42 : assert(spdk_bdev_io_get_thread(bdev_io) == spdk_get_thread());
1821 42 : mgmt_ch = bdev_io->internal.ch->shared_resource->mgmt_ch;
1822 42 : max_len = bdev_io_get_max_buf_len(bdev_io, len);
1823 :
1824 42 : if (spdk_unlikely(max_len > mgmt_ch->iobuf.cache[0].large.bufsize)) {
1825 0 : SPDK_ERRLOG("Length %" PRIu64 " is larger than allowed\n", max_len);
1826 0 : bdev_io_get_buf_complete(bdev_io, false);
1827 0 : return;
1828 : }
1829 :
1830 42 : bdev_io->internal.buf.len = len;
1831 42 : buf = spdk_iobuf_get(&mgmt_ch->iobuf, max_len, &bdev_io->internal.iobuf,
1832 : bdev_io_get_iobuf_cb);
1833 42 : if (buf != NULL) {
1834 42 : _bdev_io_set_buf(bdev_io, buf, len);
1835 42 : }
1836 42 : }
1837 :
1838 : void
1839 56 : spdk_bdev_io_get_buf(struct spdk_bdev_io *bdev_io, spdk_bdev_io_get_buf_cb cb, uint64_t len)
1840 : {
1841 56 : struct spdk_bdev *bdev = bdev_io->bdev;
1842 : uint64_t alignment;
1843 :
1844 56 : assert(cb != NULL);
1845 56 : bdev_io->internal.get_buf_cb = cb;
1846 :
1847 56 : alignment = spdk_bdev_get_buf_align(bdev);
1848 :
1849 56 : if (_is_buf_allocated(bdev_io->u.bdev.iovs) &&
1850 40 : _are_iovs_aligned(bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt, alignment)) {
1851 : /* Buffer already present and aligned */
1852 18 : cb(spdk_bdev_io_get_io_channel(bdev_io), bdev_io, true);
1853 18 : return;
1854 : }
1855 :
1856 38 : bdev_io_get_buf(bdev_io, len);
1857 56 : }
1858 :
1859 : static void
1860 4 : _bdev_memory_domain_get_io_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io,
1861 : bool success)
1862 : {
1863 4 : if (!success) {
1864 0 : SPDK_ERRLOG("Failed to get data buffer, completing IO\n");
1865 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
1866 0 : bdev_io_complete_unsubmitted(bdev_io);
1867 0 : return;
1868 : }
1869 :
1870 4 : if (bdev_io_needs_sequence_exec(bdev_io->internal.desc, bdev_io)) {
1871 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
1872 0 : bdev_io_exec_sequence(bdev_io, bdev_io_submit_sequence_cb);
1873 0 : return;
1874 : }
1875 : /* For reads we'll execute the sequence after the data is read, so, for now, only
1876 : * clear out accel_sequence pointer and submit the IO */
1877 0 : assert(bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
1878 0 : bdev_io->u.bdev.accel_sequence = NULL;
1879 0 : }
1880 :
1881 4 : bdev_io_submit(bdev_io);
1882 4 : }
1883 :
1884 : static void
1885 4 : _bdev_memory_domain_io_get_buf(struct spdk_bdev_io *bdev_io, spdk_bdev_io_get_buf_cb cb,
1886 : uint64_t len)
1887 : {
1888 4 : assert(cb != NULL);
1889 4 : bdev_io->internal.get_buf_cb = cb;
1890 :
1891 4 : bdev_io_get_buf(bdev_io, len);
1892 4 : }
1893 :
1894 :
1895 3 : SPDK_LOG_DEPRECATION_REGISTER(spdk_bdev_io_get_aux_buf,
1896 : "spdk_bdev_io_get_aux_buf is deprecated", "v25.01", 0);
1897 :
1898 : void
1899 0 : spdk_bdev_io_get_aux_buf(struct spdk_bdev_io *bdev_io, spdk_bdev_io_get_aux_buf_cb cb)
1900 : {
1901 0 : uint64_t len = bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen;
1902 :
1903 0 : SPDK_LOG_DEPRECATED(spdk_bdev_io_get_aux_buf);
1904 :
1905 0 : assert(cb != NULL);
1906 0 : assert(bdev_io->internal.get_aux_buf_cb == NULL);
1907 0 : bdev_io->internal.get_aux_buf_cb = cb;
1908 0 : bdev_io_get_buf(bdev_io, len);
1909 0 : }
1910 :
1911 : static int
1912 65 : bdev_module_get_max_ctx_size(void)
1913 : {
1914 : struct spdk_bdev_module *bdev_module;
1915 65 : int max_bdev_module_size = 0;
1916 :
1917 254 : TAILQ_FOREACH(bdev_module, &g_bdev_mgr.bdev_modules, internal.tailq) {
1918 189 : if (bdev_module->get_ctx_size && bdev_module->get_ctx_size() > max_bdev_module_size) {
1919 64 : max_bdev_module_size = bdev_module->get_ctx_size();
1920 64 : }
1921 189 : }
1922 :
1923 65 : return max_bdev_module_size;
1924 : }
1925 :
1926 : static void
1927 0 : bdev_enable_histogram_config_json(struct spdk_bdev *bdev, struct spdk_json_write_ctx *w)
1928 : {
1929 0 : if (!bdev->internal.histogram_enabled) {
1930 0 : return;
1931 : }
1932 :
1933 0 : spdk_json_write_object_begin(w);
1934 0 : spdk_json_write_named_string(w, "method", "bdev_enable_histogram");
1935 :
1936 0 : spdk_json_write_named_object_begin(w, "params");
1937 0 : spdk_json_write_named_string(w, "name", bdev->name);
1938 :
1939 0 : spdk_json_write_named_bool(w, "enable", bdev->internal.histogram_enabled);
1940 :
1941 0 : if (bdev->internal.histogram_io_type) {
1942 0 : spdk_json_write_named_string(w, "opc",
1943 0 : spdk_bdev_get_io_type_name(bdev->internal.histogram_io_type));
1944 0 : }
1945 :
1946 0 : spdk_json_write_object_end(w);
1947 :
1948 0 : spdk_json_write_object_end(w);
1949 0 : }
1950 :
1951 : static void
1952 0 : bdev_qos_config_json(struct spdk_bdev *bdev, struct spdk_json_write_ctx *w)
1953 : {
1954 : int i;
1955 0 : struct spdk_bdev_qos *qos = bdev->internal.qos;
1956 : uint64_t limits[SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES];
1957 :
1958 0 : if (!qos) {
1959 0 : return;
1960 : }
1961 :
1962 0 : spdk_bdev_get_qos_rate_limits(bdev, limits);
1963 :
1964 0 : spdk_json_write_object_begin(w);
1965 0 : spdk_json_write_named_string(w, "method", "bdev_set_qos_limit");
1966 :
1967 0 : spdk_json_write_named_object_begin(w, "params");
1968 0 : spdk_json_write_named_string(w, "name", bdev->name);
1969 0 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
1970 0 : if (limits[i] > 0) {
1971 0 : spdk_json_write_named_uint64(w, qos_rpc_type[i], limits[i]);
1972 0 : }
1973 0 : }
1974 0 : spdk_json_write_object_end(w);
1975 :
1976 0 : spdk_json_write_object_end(w);
1977 0 : }
1978 :
1979 : void
1980 0 : spdk_bdev_subsystem_config_json(struct spdk_json_write_ctx *w)
1981 : {
1982 : struct spdk_bdev_module *bdev_module;
1983 : struct spdk_bdev *bdev;
1984 :
1985 0 : assert(w != NULL);
1986 :
1987 0 : spdk_json_write_array_begin(w);
1988 :
1989 0 : spdk_json_write_object_begin(w);
1990 0 : spdk_json_write_named_string(w, "method", "bdev_set_options");
1991 0 : spdk_json_write_named_object_begin(w, "params");
1992 0 : spdk_json_write_named_uint32(w, "bdev_io_pool_size", g_bdev_opts.bdev_io_pool_size);
1993 0 : spdk_json_write_named_uint32(w, "bdev_io_cache_size", g_bdev_opts.bdev_io_cache_size);
1994 0 : spdk_json_write_named_bool(w, "bdev_auto_examine", g_bdev_opts.bdev_auto_examine);
1995 0 : spdk_json_write_named_uint32(w, "iobuf_small_cache_size", g_bdev_opts.iobuf_small_cache_size);
1996 0 : spdk_json_write_named_uint32(w, "iobuf_large_cache_size", g_bdev_opts.iobuf_large_cache_size);
1997 0 : spdk_json_write_object_end(w);
1998 0 : spdk_json_write_object_end(w);
1999 :
2000 0 : bdev_examine_allowlist_config_json(w);
2001 :
2002 0 : TAILQ_FOREACH(bdev_module, &g_bdev_mgr.bdev_modules, internal.tailq) {
2003 0 : if (bdev_module->config_json) {
2004 0 : bdev_module->config_json(w);
2005 0 : }
2006 0 : }
2007 :
2008 0 : spdk_spin_lock(&g_bdev_mgr.spinlock);
2009 :
2010 0 : TAILQ_FOREACH(bdev, &g_bdev_mgr.bdevs, internal.link) {
2011 0 : if (bdev->fn_table->write_config_json) {
2012 0 : bdev->fn_table->write_config_json(bdev, w);
2013 0 : }
2014 :
2015 0 : bdev_qos_config_json(bdev, w);
2016 0 : bdev_enable_histogram_config_json(bdev, w);
2017 0 : }
2018 :
2019 0 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
2020 :
2021 : /* This has to be last RPC in array to make sure all bdevs finished examine */
2022 0 : spdk_json_write_object_begin(w);
2023 0 : spdk_json_write_named_string(w, "method", "bdev_wait_for_examine");
2024 0 : spdk_json_write_object_end(w);
2025 :
2026 0 : spdk_json_write_array_end(w);
2027 0 : }
2028 :
2029 : static void
2030 71 : bdev_mgmt_channel_destroy(void *io_device, void *ctx_buf)
2031 : {
2032 71 : struct spdk_bdev_mgmt_channel *ch = ctx_buf;
2033 : struct spdk_bdev_io *bdev_io;
2034 :
2035 71 : spdk_iobuf_channel_fini(&ch->iobuf);
2036 :
2037 9969 : while (!STAILQ_EMPTY(&ch->per_thread_cache)) {
2038 9898 : bdev_io = STAILQ_FIRST(&ch->per_thread_cache);
2039 9898 : STAILQ_REMOVE_HEAD(&ch->per_thread_cache, internal.buf_link);
2040 9898 : ch->per_thread_cache_count--;
2041 9898 : spdk_mempool_put(g_bdev_mgr.bdev_io_pool, (void *)bdev_io);
2042 : }
2043 :
2044 71 : assert(ch->per_thread_cache_count == 0);
2045 71 : }
2046 :
2047 : static int
2048 71 : bdev_mgmt_channel_create(void *io_device, void *ctx_buf)
2049 : {
2050 71 : struct spdk_bdev_mgmt_channel *ch = ctx_buf;
2051 : struct spdk_bdev_io *bdev_io;
2052 : uint32_t i;
2053 : int rc;
2054 :
2055 142 : rc = spdk_iobuf_channel_init(&ch->iobuf, "bdev",
2056 71 : g_bdev_opts.iobuf_small_cache_size,
2057 71 : g_bdev_opts.iobuf_large_cache_size);
2058 71 : if (rc != 0) {
2059 0 : SPDK_ERRLOG("Failed to create iobuf channel: %s\n", spdk_strerror(-rc));
2060 0 : return -1;
2061 : }
2062 :
2063 71 : STAILQ_INIT(&ch->per_thread_cache);
2064 71 : ch->bdev_io_cache_size = g_bdev_opts.bdev_io_cache_size;
2065 :
2066 : /* Pre-populate bdev_io cache to ensure this thread cannot be starved. */
2067 71 : ch->per_thread_cache_count = 0;
2068 9969 : for (i = 0; i < ch->bdev_io_cache_size; i++) {
2069 9898 : bdev_io = spdk_mempool_get(g_bdev_mgr.bdev_io_pool);
2070 9898 : if (bdev_io == NULL) {
2071 0 : SPDK_ERRLOG("You need to increase bdev_io_pool_size using bdev_set_options RPC.\n");
2072 0 : assert(false);
2073 : bdev_mgmt_channel_destroy(io_device, ctx_buf);
2074 : return -1;
2075 : }
2076 9898 : ch->per_thread_cache_count++;
2077 9898 : STAILQ_INSERT_HEAD(&ch->per_thread_cache, bdev_io, internal.buf_link);
2078 9898 : }
2079 :
2080 71 : TAILQ_INIT(&ch->shared_resources);
2081 71 : TAILQ_INIT(&ch->io_wait_queue);
2082 :
2083 71 : return 0;
2084 71 : }
2085 :
2086 : static void
2087 65 : bdev_init_complete(int rc)
2088 : {
2089 65 : spdk_bdev_init_cb cb_fn = g_init_cb_fn;
2090 65 : void *cb_arg = g_init_cb_arg;
2091 : struct spdk_bdev_module *m;
2092 :
2093 65 : g_bdev_mgr.init_complete = true;
2094 65 : g_init_cb_fn = NULL;
2095 65 : g_init_cb_arg = NULL;
2096 :
2097 : /*
2098 : * For modules that need to know when subsystem init is complete,
2099 : * inform them now.
2100 : */
2101 65 : if (rc == 0) {
2102 254 : TAILQ_FOREACH(m, &g_bdev_mgr.bdev_modules, internal.tailq) {
2103 189 : if (m->init_complete) {
2104 23 : m->init_complete();
2105 23 : }
2106 189 : }
2107 65 : }
2108 :
2109 65 : cb_fn(cb_arg, rc);
2110 65 : }
2111 :
2112 : static bool
2113 256 : bdev_module_all_actions_completed(void)
2114 : {
2115 : struct spdk_bdev_module *m;
2116 :
2117 1012 : TAILQ_FOREACH(m, &g_bdev_mgr.bdev_modules, internal.tailq) {
2118 756 : if (m->internal.action_in_progress > 0) {
2119 0 : return false;
2120 : }
2121 756 : }
2122 256 : return true;
2123 256 : }
2124 :
2125 : static void
2126 616 : bdev_module_action_complete(void)
2127 : {
2128 : /*
2129 : * Don't finish bdev subsystem initialization if
2130 : * module pre-initialization is still in progress, or
2131 : * the subsystem been already initialized.
2132 : */
2133 616 : if (!g_bdev_mgr.module_init_complete || g_bdev_mgr.init_complete) {
2134 551 : return;
2135 : }
2136 :
2137 : /*
2138 : * Check all bdev modules for inits/examinations in progress. If any
2139 : * exist, return immediately since we cannot finish bdev subsystem
2140 : * initialization until all are completed.
2141 : */
2142 65 : if (!bdev_module_all_actions_completed()) {
2143 0 : return;
2144 : }
2145 :
2146 : /*
2147 : * Modules already finished initialization - now that all
2148 : * the bdev modules have finished their asynchronous I/O
2149 : * processing, the entire bdev layer can be marked as complete.
2150 : */
2151 65 : bdev_init_complete(0);
2152 616 : }
2153 :
2154 : static void
2155 551 : bdev_module_action_done(struct spdk_bdev_module *module)
2156 : {
2157 551 : spdk_spin_lock(&module->internal.spinlock);
2158 551 : assert(module->internal.action_in_progress > 0);
2159 551 : module->internal.action_in_progress--;
2160 551 : spdk_spin_unlock(&module->internal.spinlock);
2161 551 : bdev_module_action_complete();
2162 551 : }
2163 :
2164 : void
2165 65 : spdk_bdev_module_init_done(struct spdk_bdev_module *module)
2166 : {
2167 65 : assert(module->async_init);
2168 65 : bdev_module_action_done(module);
2169 65 : }
2170 :
2171 : void
2172 486 : spdk_bdev_module_examine_done(struct spdk_bdev_module *module)
2173 : {
2174 486 : bdev_module_action_done(module);
2175 486 : }
2176 :
2177 : /** The last initialized bdev module */
2178 : static struct spdk_bdev_module *g_resume_bdev_module = NULL;
2179 :
2180 : static void
2181 0 : bdev_init_failed(void *cb_arg)
2182 : {
2183 0 : struct spdk_bdev_module *module = cb_arg;
2184 :
2185 0 : spdk_spin_lock(&module->internal.spinlock);
2186 0 : assert(module->internal.action_in_progress > 0);
2187 0 : module->internal.action_in_progress--;
2188 0 : spdk_spin_unlock(&module->internal.spinlock);
2189 0 : bdev_init_complete(-1);
2190 0 : }
2191 :
2192 : static int
2193 65 : bdev_modules_init(void)
2194 : {
2195 : struct spdk_bdev_module *module;
2196 65 : int rc = 0;
2197 :
2198 254 : TAILQ_FOREACH(module, &g_bdev_mgr.bdev_modules, internal.tailq) {
2199 189 : g_resume_bdev_module = module;
2200 189 : if (module->async_init) {
2201 65 : spdk_spin_lock(&module->internal.spinlock);
2202 65 : module->internal.action_in_progress = 1;
2203 65 : spdk_spin_unlock(&module->internal.spinlock);
2204 65 : }
2205 189 : rc = module->module_init();
2206 189 : if (rc != 0) {
2207 : /* Bump action_in_progress to prevent other modules from completion of modules_init
2208 : * Send message to defer application shutdown until resources are cleaned up */
2209 0 : spdk_spin_lock(&module->internal.spinlock);
2210 0 : module->internal.action_in_progress = 1;
2211 0 : spdk_spin_unlock(&module->internal.spinlock);
2212 0 : spdk_thread_send_msg(spdk_get_thread(), bdev_init_failed, module);
2213 0 : return rc;
2214 : }
2215 189 : }
2216 :
2217 65 : g_resume_bdev_module = NULL;
2218 65 : return 0;
2219 65 : }
2220 :
2221 : void
2222 65 : spdk_bdev_initialize(spdk_bdev_init_cb cb_fn, void *cb_arg)
2223 : {
2224 65 : int rc = 0;
2225 : char mempool_name[32];
2226 :
2227 65 : assert(cb_fn != NULL);
2228 :
2229 65 : g_init_cb_fn = cb_fn;
2230 65 : g_init_cb_arg = cb_arg;
2231 :
2232 65 : spdk_notify_type_register("bdev_register");
2233 65 : spdk_notify_type_register("bdev_unregister");
2234 :
2235 65 : snprintf(mempool_name, sizeof(mempool_name), "bdev_io_%d", getpid());
2236 :
2237 65 : rc = spdk_iobuf_register_module("bdev");
2238 65 : if (rc != 0) {
2239 0 : SPDK_ERRLOG("could not register bdev iobuf module: %s\n", spdk_strerror(-rc));
2240 0 : bdev_init_complete(-1);
2241 0 : return;
2242 : }
2243 :
2244 130 : g_bdev_mgr.bdev_io_pool = spdk_mempool_create(mempool_name,
2245 65 : g_bdev_opts.bdev_io_pool_size,
2246 65 : sizeof(struct spdk_bdev_io) +
2247 65 : bdev_module_get_max_ctx_size(),
2248 : 0,
2249 : SPDK_ENV_NUMA_ID_ANY);
2250 :
2251 65 : if (g_bdev_mgr.bdev_io_pool == NULL) {
2252 0 : SPDK_ERRLOG("could not allocate spdk_bdev_io pool\n");
2253 0 : bdev_init_complete(-1);
2254 0 : return;
2255 : }
2256 :
2257 65 : g_bdev_mgr.zero_buffer = spdk_zmalloc(ZERO_BUFFER_SIZE, ZERO_BUFFER_SIZE,
2258 : NULL, SPDK_ENV_LCORE_ID_ANY, SPDK_MALLOC_DMA);
2259 65 : if (!g_bdev_mgr.zero_buffer) {
2260 0 : SPDK_ERRLOG("create bdev zero buffer failed\n");
2261 0 : bdev_init_complete(-1);
2262 0 : return;
2263 : }
2264 :
2265 : #ifdef SPDK_CONFIG_VTUNE
2266 : g_bdev_mgr.domain = __itt_domain_create("spdk_bdev");
2267 : #endif
2268 :
2269 65 : spdk_io_device_register(&g_bdev_mgr, bdev_mgmt_channel_create,
2270 : bdev_mgmt_channel_destroy,
2271 : sizeof(struct spdk_bdev_mgmt_channel),
2272 : "bdev_mgr");
2273 :
2274 65 : rc = bdev_modules_init();
2275 65 : g_bdev_mgr.module_init_complete = true;
2276 65 : if (rc != 0) {
2277 0 : SPDK_ERRLOG("bdev modules init failed\n");
2278 0 : return;
2279 : }
2280 :
2281 65 : bdev_module_action_complete();
2282 65 : }
2283 :
2284 : static void
2285 65 : bdev_mgr_unregister_cb(void *io_device)
2286 : {
2287 65 : spdk_bdev_fini_cb cb_fn = g_fini_cb_fn;
2288 :
2289 65 : if (g_bdev_mgr.bdev_io_pool) {
2290 65 : if (spdk_mempool_count(g_bdev_mgr.bdev_io_pool) != g_bdev_opts.bdev_io_pool_size) {
2291 0 : SPDK_ERRLOG("bdev IO pool count is %zu but should be %u\n",
2292 : spdk_mempool_count(g_bdev_mgr.bdev_io_pool),
2293 : g_bdev_opts.bdev_io_pool_size);
2294 0 : }
2295 :
2296 65 : spdk_mempool_free(g_bdev_mgr.bdev_io_pool);
2297 65 : }
2298 :
2299 65 : spdk_free(g_bdev_mgr.zero_buffer);
2300 :
2301 65 : bdev_examine_allowlist_free();
2302 :
2303 65 : cb_fn(g_fini_cb_arg);
2304 65 : g_fini_cb_fn = NULL;
2305 65 : g_fini_cb_arg = NULL;
2306 65 : g_bdev_mgr.init_complete = false;
2307 65 : g_bdev_mgr.module_init_complete = false;
2308 65 : }
2309 :
2310 : static void
2311 65 : bdev_module_fini_iter(void *arg)
2312 : {
2313 : struct spdk_bdev_module *bdev_module;
2314 :
2315 : /* FIXME: Handling initialization failures is broken now,
2316 : * so we won't even try cleaning up after successfully
2317 : * initialized modules. if module_init_complete is false,
2318 : * just call spdk_bdev_mgr_unregister_cb
2319 : */
2320 65 : if (!g_bdev_mgr.module_init_complete) {
2321 0 : bdev_mgr_unregister_cb(NULL);
2322 0 : return;
2323 : }
2324 :
2325 : /* Start iterating from the last touched module */
2326 65 : if (!g_resume_bdev_module) {
2327 65 : bdev_module = TAILQ_LAST(&g_bdev_mgr.bdev_modules, bdev_module_list);
2328 65 : } else {
2329 0 : bdev_module = TAILQ_PREV(g_resume_bdev_module, bdev_module_list,
2330 : internal.tailq);
2331 : }
2332 :
2333 254 : while (bdev_module) {
2334 189 : if (bdev_module->async_fini) {
2335 : /* Save our place so we can resume later. We must
2336 : * save the variable here, before calling module_fini()
2337 : * below, because in some cases the module may immediately
2338 : * call spdk_bdev_module_fini_done() and re-enter
2339 : * this function to continue iterating. */
2340 0 : g_resume_bdev_module = bdev_module;
2341 0 : }
2342 :
2343 189 : if (bdev_module->module_fini) {
2344 189 : bdev_module->module_fini();
2345 189 : }
2346 :
2347 189 : if (bdev_module->async_fini) {
2348 0 : return;
2349 : }
2350 :
2351 189 : bdev_module = TAILQ_PREV(bdev_module, bdev_module_list,
2352 : internal.tailq);
2353 : }
2354 :
2355 65 : g_resume_bdev_module = NULL;
2356 65 : spdk_io_device_unregister(&g_bdev_mgr, bdev_mgr_unregister_cb);
2357 65 : }
2358 :
2359 : void
2360 0 : spdk_bdev_module_fini_done(void)
2361 : {
2362 0 : if (spdk_get_thread() != g_fini_thread) {
2363 0 : spdk_thread_send_msg(g_fini_thread, bdev_module_fini_iter, NULL);
2364 0 : } else {
2365 0 : bdev_module_fini_iter(NULL);
2366 : }
2367 0 : }
2368 :
2369 : static void
2370 65 : bdev_finish_unregister_bdevs_iter(void *cb_arg, int bdeverrno)
2371 : {
2372 65 : struct spdk_bdev *bdev = cb_arg;
2373 :
2374 65 : if (bdeverrno && bdev) {
2375 0 : SPDK_WARNLOG("Unable to unregister bdev '%s' during spdk_bdev_finish()\n",
2376 : bdev->name);
2377 :
2378 : /*
2379 : * Since the call to spdk_bdev_unregister() failed, we have no way to free this
2380 : * bdev; try to continue by manually removing this bdev from the list and continue
2381 : * with the next bdev in the list.
2382 : */
2383 0 : TAILQ_REMOVE(&g_bdev_mgr.bdevs, bdev, internal.link);
2384 0 : }
2385 :
2386 65 : if (TAILQ_EMPTY(&g_bdev_mgr.bdevs)) {
2387 65 : SPDK_DEBUGLOG(bdev, "Done unregistering bdevs\n");
2388 : /*
2389 : * Bdev module finish need to be deferred as we might be in the middle of some context
2390 : * (like bdev part free) that will use this bdev (or private bdev driver ctx data)
2391 : * after returning.
2392 : */
2393 65 : spdk_thread_send_msg(spdk_get_thread(), bdev_module_fini_iter, NULL);
2394 65 : return;
2395 : }
2396 :
2397 : /*
2398 : * Unregister last unclaimed bdev in the list, to ensure that bdev subsystem
2399 : * shutdown proceeds top-down. The goal is to give virtual bdevs an opportunity
2400 : * to detect clean shutdown as opposed to run-time hot removal of the underlying
2401 : * base bdevs.
2402 : *
2403 : * Also, walk the list in the reverse order.
2404 : */
2405 0 : for (bdev = TAILQ_LAST(&g_bdev_mgr.bdevs, spdk_bdev_list);
2406 0 : bdev; bdev = TAILQ_PREV(bdev, spdk_bdev_list, internal.link)) {
2407 0 : spdk_spin_lock(&bdev->internal.spinlock);
2408 0 : if (bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE) {
2409 0 : LOG_ALREADY_CLAIMED_DEBUG("claimed, skipping", bdev);
2410 0 : spdk_spin_unlock(&bdev->internal.spinlock);
2411 0 : continue;
2412 : }
2413 0 : spdk_spin_unlock(&bdev->internal.spinlock);
2414 :
2415 0 : SPDK_DEBUGLOG(bdev, "Unregistering bdev '%s'\n", bdev->name);
2416 0 : spdk_bdev_unregister(bdev, bdev_finish_unregister_bdevs_iter, bdev);
2417 0 : return;
2418 : }
2419 :
2420 : /*
2421 : * If any bdev fails to unclaim underlying bdev properly, we may face the
2422 : * case of bdev list consisting of claimed bdevs only (if claims are managed
2423 : * correctly, this would mean there's a loop in the claims graph which is
2424 : * clearly impossible). Warn and unregister last bdev on the list then.
2425 : */
2426 0 : for (bdev = TAILQ_LAST(&g_bdev_mgr.bdevs, spdk_bdev_list);
2427 0 : bdev; bdev = TAILQ_PREV(bdev, spdk_bdev_list, internal.link)) {
2428 0 : SPDK_WARNLOG("Unregistering claimed bdev '%s'!\n", bdev->name);
2429 0 : spdk_bdev_unregister(bdev, bdev_finish_unregister_bdevs_iter, bdev);
2430 0 : return;
2431 : }
2432 65 : }
2433 :
2434 : static void
2435 65 : bdev_module_fini_start_iter(void *arg)
2436 : {
2437 : struct spdk_bdev_module *bdev_module;
2438 :
2439 65 : if (!g_resume_bdev_module) {
2440 65 : bdev_module = TAILQ_LAST(&g_bdev_mgr.bdev_modules, bdev_module_list);
2441 65 : } else {
2442 0 : bdev_module = TAILQ_PREV(g_resume_bdev_module, bdev_module_list, internal.tailq);
2443 : }
2444 :
2445 254 : while (bdev_module) {
2446 189 : if (bdev_module->async_fini_start) {
2447 : /* Save our place so we can resume later. We must
2448 : * save the variable here, before calling fini_start()
2449 : * below, because in some cases the module may immediately
2450 : * call spdk_bdev_module_fini_start_done() and re-enter
2451 : * this function to continue iterating. */
2452 0 : g_resume_bdev_module = bdev_module;
2453 0 : }
2454 :
2455 189 : if (bdev_module->fini_start) {
2456 23 : bdev_module->fini_start();
2457 23 : }
2458 :
2459 189 : if (bdev_module->async_fini_start) {
2460 0 : return;
2461 : }
2462 :
2463 189 : bdev_module = TAILQ_PREV(bdev_module, bdev_module_list, internal.tailq);
2464 : }
2465 :
2466 65 : g_resume_bdev_module = NULL;
2467 :
2468 65 : bdev_finish_unregister_bdevs_iter(NULL, 0);
2469 65 : }
2470 :
2471 : void
2472 0 : spdk_bdev_module_fini_start_done(void)
2473 : {
2474 0 : if (spdk_get_thread() != g_fini_thread) {
2475 0 : spdk_thread_send_msg(g_fini_thread, bdev_module_fini_start_iter, NULL);
2476 0 : } else {
2477 0 : bdev_module_fini_start_iter(NULL);
2478 : }
2479 0 : }
2480 :
2481 : static void
2482 65 : bdev_finish_wait_for_examine_done(void *cb_arg)
2483 : {
2484 65 : bdev_module_fini_start_iter(NULL);
2485 65 : }
2486 :
2487 : static void bdev_open_async_fini(void);
2488 :
2489 : void
2490 65 : spdk_bdev_finish(spdk_bdev_fini_cb cb_fn, void *cb_arg)
2491 : {
2492 : int rc;
2493 :
2494 65 : assert(cb_fn != NULL);
2495 :
2496 65 : g_fini_thread = spdk_get_thread();
2497 :
2498 65 : g_fini_cb_fn = cb_fn;
2499 65 : g_fini_cb_arg = cb_arg;
2500 :
2501 65 : bdev_open_async_fini();
2502 :
2503 65 : rc = spdk_bdev_wait_for_examine(bdev_finish_wait_for_examine_done, NULL);
2504 65 : if (rc != 0) {
2505 0 : SPDK_ERRLOG("wait_for_examine failed: %s\n", spdk_strerror(-rc));
2506 0 : bdev_finish_wait_for_examine_done(NULL);
2507 0 : }
2508 65 : }
2509 :
2510 : struct spdk_bdev_io *
2511 697 : bdev_channel_get_io(struct spdk_bdev_channel *channel)
2512 : {
2513 697 : struct spdk_bdev_mgmt_channel *ch = channel->shared_resource->mgmt_ch;
2514 : struct spdk_bdev_io *bdev_io;
2515 :
2516 697 : if (ch->per_thread_cache_count > 0) {
2517 637 : bdev_io = STAILQ_FIRST(&ch->per_thread_cache);
2518 637 : STAILQ_REMOVE_HEAD(&ch->per_thread_cache, internal.buf_link);
2519 637 : ch->per_thread_cache_count--;
2520 697 : } else if (spdk_unlikely(!TAILQ_EMPTY(&ch->io_wait_queue))) {
2521 : /*
2522 : * Don't try to look for bdev_ios in the global pool if there are
2523 : * waiters on bdev_ios - we don't want this caller to jump the line.
2524 : */
2525 0 : bdev_io = NULL;
2526 0 : } else {
2527 60 : bdev_io = spdk_mempool_get(g_bdev_mgr.bdev_io_pool);
2528 : }
2529 :
2530 697 : return bdev_io;
2531 : }
2532 :
2533 : void
2534 691 : spdk_bdev_free_io(struct spdk_bdev_io *bdev_io)
2535 : {
2536 : struct spdk_bdev_mgmt_channel *ch;
2537 :
2538 691 : assert(bdev_io != NULL);
2539 691 : assert(bdev_io->internal.status != SPDK_BDEV_IO_STATUS_PENDING);
2540 :
2541 691 : ch = bdev_io->internal.ch->shared_resource->mgmt_ch;
2542 :
2543 691 : if (bdev_io->internal.f.has_buf) {
2544 16 : bdev_io_put_buf(bdev_io);
2545 16 : }
2546 :
2547 691 : if (ch->per_thread_cache_count < ch->bdev_io_cache_size) {
2548 637 : ch->per_thread_cache_count++;
2549 637 : STAILQ_INSERT_HEAD(&ch->per_thread_cache, bdev_io, internal.buf_link);
2550 641 : while (ch->per_thread_cache_count > 0 && !TAILQ_EMPTY(&ch->io_wait_queue)) {
2551 : struct spdk_bdev_io_wait_entry *entry;
2552 :
2553 4 : entry = TAILQ_FIRST(&ch->io_wait_queue);
2554 4 : TAILQ_REMOVE(&ch->io_wait_queue, entry, link);
2555 4 : entry->cb_fn(entry->cb_arg);
2556 : }
2557 637 : } else {
2558 : /* We should never have a full cache with entries on the io wait queue. */
2559 54 : assert(TAILQ_EMPTY(&ch->io_wait_queue));
2560 54 : spdk_mempool_put(g_bdev_mgr.bdev_io_pool, (void *)bdev_io);
2561 : }
2562 691 : }
2563 :
2564 : static bool
2565 72 : bdev_qos_is_iops_rate_limit(enum spdk_bdev_qos_rate_limit_type limit)
2566 : {
2567 72 : assert(limit != SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES);
2568 :
2569 72 : switch (limit) {
2570 : case SPDK_BDEV_QOS_RW_IOPS_RATE_LIMIT:
2571 18 : return true;
2572 : case SPDK_BDEV_QOS_RW_BPS_RATE_LIMIT:
2573 : case SPDK_BDEV_QOS_R_BPS_RATE_LIMIT:
2574 : case SPDK_BDEV_QOS_W_BPS_RATE_LIMIT:
2575 54 : return false;
2576 0 : case SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES:
2577 : default:
2578 0 : return false;
2579 : }
2580 72 : }
2581 :
2582 : static bool
2583 25 : bdev_qos_io_to_limit(struct spdk_bdev_io *bdev_io)
2584 : {
2585 25 : switch (bdev_io->type) {
2586 : case SPDK_BDEV_IO_TYPE_NVME_IO:
2587 : case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
2588 : case SPDK_BDEV_IO_TYPE_READ:
2589 : case SPDK_BDEV_IO_TYPE_WRITE:
2590 23 : return true;
2591 : case SPDK_BDEV_IO_TYPE_ZCOPY:
2592 0 : if (bdev_io->u.bdev.zcopy.start) {
2593 0 : return true;
2594 : } else {
2595 0 : return false;
2596 : }
2597 : default:
2598 2 : return false;
2599 : }
2600 25 : }
2601 :
2602 : static bool
2603 33 : bdev_is_read_io(struct spdk_bdev_io *bdev_io)
2604 : {
2605 33 : switch (bdev_io->type) {
2606 : case SPDK_BDEV_IO_TYPE_NVME_IO:
2607 : case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
2608 : /* Bit 1 (0x2) set for read operation */
2609 0 : if (bdev_io->u.nvme_passthru.cmd.opc & SPDK_NVME_OPC_READ) {
2610 0 : return true;
2611 : } else {
2612 0 : return false;
2613 : }
2614 : case SPDK_BDEV_IO_TYPE_READ:
2615 30 : return true;
2616 : case SPDK_BDEV_IO_TYPE_ZCOPY:
2617 : /* Populate to read from disk */
2618 0 : if (bdev_io->u.bdev.zcopy.populate) {
2619 0 : return true;
2620 : } else {
2621 0 : return false;
2622 : }
2623 : default:
2624 3 : return false;
2625 : }
2626 33 : }
2627 :
2628 : static uint64_t
2629 43 : bdev_get_io_size_in_byte(struct spdk_bdev_io *bdev_io)
2630 : {
2631 43 : struct spdk_bdev *bdev = bdev_io->bdev;
2632 :
2633 43 : switch (bdev_io->type) {
2634 : case SPDK_BDEV_IO_TYPE_NVME_IO:
2635 : case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
2636 0 : return bdev_io->u.nvme_passthru.nbytes;
2637 : case SPDK_BDEV_IO_TYPE_READ:
2638 : case SPDK_BDEV_IO_TYPE_WRITE:
2639 43 : return bdev_io->u.bdev.num_blocks * bdev->blocklen;
2640 : case SPDK_BDEV_IO_TYPE_ZCOPY:
2641 : /* Track the data in the start phase only */
2642 0 : if (bdev_io->u.bdev.zcopy.start) {
2643 0 : return bdev_io->u.bdev.num_blocks * bdev->blocklen;
2644 : } else {
2645 0 : return 0;
2646 : }
2647 : default:
2648 0 : return 0;
2649 : }
2650 43 : }
2651 :
2652 : static inline bool
2653 64 : bdev_qos_rw_queue_io(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io, uint64_t delta)
2654 : {
2655 : int64_t remaining_this_timeslice;
2656 :
2657 64 : if (!limit->max_per_timeslice) {
2658 : /* The QoS is disabled */
2659 0 : return false;
2660 : }
2661 :
2662 64 : remaining_this_timeslice = __atomic_sub_fetch(&limit->remaining_this_timeslice, delta,
2663 : __ATOMIC_RELAXED);
2664 64 : if (remaining_this_timeslice + (int64_t)delta > 0) {
2665 : /* There was still a quota for this delta -> the IO shouldn't be queued
2666 : *
2667 : * We allow a slight quota overrun here so an IO bigger than the per-timeslice
2668 : * quota can be allowed once a while. Such overrun then taken into account in
2669 : * the QoS poller, where the next timeslice quota is calculated.
2670 : */
2671 59 : return false;
2672 : }
2673 :
2674 : /* There was no quota for this delta -> the IO should be queued
2675 : * The remaining_this_timeslice must be rewinded so it reflects the real
2676 : * amount of IOs or bytes allowed.
2677 : */
2678 5 : __atomic_add_fetch(
2679 5 : &limit->remaining_this_timeslice, delta, __ATOMIC_RELAXED);
2680 5 : return true;
2681 64 : }
2682 :
2683 : static inline void
2684 5 : bdev_qos_rw_rewind_io(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io, uint64_t delta)
2685 : {
2686 5 : __atomic_add_fetch(&limit->remaining_this_timeslice, delta, __ATOMIC_RELAXED);
2687 5 : }
2688 :
2689 : static bool
2690 23 : bdev_qos_rw_iops_queue(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2691 : {
2692 23 : return bdev_qos_rw_queue_io(limit, io, 1);
2693 : }
2694 :
2695 : static void
2696 3 : bdev_qos_rw_iops_rewind_quota(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2697 : {
2698 3 : bdev_qos_rw_rewind_io(limit, io, 1);
2699 3 : }
2700 :
2701 : static bool
2702 41 : bdev_qos_rw_bps_queue(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2703 : {
2704 41 : return bdev_qos_rw_queue_io(limit, io, bdev_get_io_size_in_byte(io));
2705 : }
2706 :
2707 : static void
2708 2 : bdev_qos_rw_bps_rewind_quota(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2709 : {
2710 2 : bdev_qos_rw_rewind_io(limit, io, bdev_get_io_size_in_byte(io));
2711 2 : }
2712 :
2713 : static bool
2714 19 : bdev_qos_r_bps_queue(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2715 : {
2716 19 : if (bdev_is_read_io(io) == false) {
2717 1 : return false;
2718 : }
2719 :
2720 18 : return bdev_qos_rw_bps_queue(limit, io);
2721 19 : }
2722 :
2723 : static void
2724 0 : bdev_qos_r_bps_rewind_quota(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2725 : {
2726 0 : if (bdev_is_read_io(io) != false) {
2727 0 : bdev_qos_rw_rewind_io(limit, io, bdev_get_io_size_in_byte(io));
2728 0 : }
2729 0 : }
2730 :
2731 : static bool
2732 14 : bdev_qos_w_bps_queue(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2733 : {
2734 14 : if (bdev_is_read_io(io) == true) {
2735 12 : return false;
2736 : }
2737 :
2738 2 : return bdev_qos_rw_bps_queue(limit, io);
2739 14 : }
2740 :
2741 : static void
2742 0 : bdev_qos_w_bps_rewind_quota(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2743 : {
2744 0 : if (bdev_is_read_io(io) != true) {
2745 0 : bdev_qos_rw_rewind_io(limit, io, bdev_get_io_size_in_byte(io));
2746 0 : }
2747 0 : }
2748 :
2749 : static void
2750 10 : bdev_qos_set_ops(struct spdk_bdev_qos *qos)
2751 : {
2752 : int i;
2753 :
2754 50 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
2755 40 : if (qos->rate_limits[i].limit == SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
2756 15 : qos->rate_limits[i].queue_io = NULL;
2757 15 : continue;
2758 : }
2759 :
2760 25 : switch (i) {
2761 : case SPDK_BDEV_QOS_RW_IOPS_RATE_LIMIT:
2762 9 : qos->rate_limits[i].queue_io = bdev_qos_rw_iops_queue;
2763 9 : qos->rate_limits[i].rewind_quota = bdev_qos_rw_iops_rewind_quota;
2764 9 : break;
2765 : case SPDK_BDEV_QOS_RW_BPS_RATE_LIMIT:
2766 7 : qos->rate_limits[i].queue_io = bdev_qos_rw_bps_queue;
2767 7 : qos->rate_limits[i].rewind_quota = bdev_qos_rw_bps_rewind_quota;
2768 7 : break;
2769 : case SPDK_BDEV_QOS_R_BPS_RATE_LIMIT:
2770 5 : qos->rate_limits[i].queue_io = bdev_qos_r_bps_queue;
2771 5 : qos->rate_limits[i].rewind_quota = bdev_qos_r_bps_rewind_quota;
2772 5 : break;
2773 : case SPDK_BDEV_QOS_W_BPS_RATE_LIMIT:
2774 4 : qos->rate_limits[i].queue_io = bdev_qos_w_bps_queue;
2775 4 : qos->rate_limits[i].rewind_quota = bdev_qos_w_bps_rewind_quota;
2776 4 : break;
2777 : default:
2778 0 : break;
2779 : }
2780 25 : }
2781 10 : }
2782 :
2783 : static void
2784 6 : _bdev_io_complete_in_submit(struct spdk_bdev_channel *bdev_ch,
2785 : struct spdk_bdev_io *bdev_io,
2786 : enum spdk_bdev_io_status status)
2787 : {
2788 6 : bdev_io->internal.f.in_submit_request = true;
2789 6 : bdev_io_increment_outstanding(bdev_ch, bdev_ch->shared_resource);
2790 6 : spdk_bdev_io_complete(bdev_io, status);
2791 6 : bdev_io->internal.f.in_submit_request = false;
2792 6 : }
2793 :
2794 : static inline void
2795 574 : bdev_io_do_submit(struct spdk_bdev_channel *bdev_ch, struct spdk_bdev_io *bdev_io)
2796 : {
2797 574 : struct spdk_bdev *bdev = bdev_io->bdev;
2798 574 : struct spdk_io_channel *ch = bdev_ch->channel;
2799 574 : struct spdk_bdev_shared_resource *shared_resource = bdev_ch->shared_resource;
2800 :
2801 574 : if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_ABORT)) {
2802 16 : struct spdk_bdev_mgmt_channel *mgmt_channel = shared_resource->mgmt_ch;
2803 16 : struct spdk_bdev_io *bio_to_abort = bdev_io->u.abort.bio_to_abort;
2804 :
2805 16 : if (bdev_abort_queued_io(&shared_resource->nomem_io, bio_to_abort) ||
2806 16 : bdev_abort_buf_io(mgmt_channel, bio_to_abort)) {
2807 0 : _bdev_io_complete_in_submit(bdev_ch, bdev_io,
2808 : SPDK_BDEV_IO_STATUS_SUCCESS);
2809 0 : return;
2810 : }
2811 16 : }
2812 :
2813 574 : if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE &&
2814 : bdev_io->bdev->split_on_write_unit &&
2815 : bdev_io->u.bdev.num_blocks < bdev_io->bdev->write_unit_size)) {
2816 4 : SPDK_ERRLOG("IO num_blocks %lu does not match the write_unit_size %u\n",
2817 : bdev_io->u.bdev.num_blocks, bdev_io->bdev->write_unit_size);
2818 4 : _bdev_io_complete_in_submit(bdev_ch, bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
2819 4 : return;
2820 : }
2821 :
2822 570 : if (spdk_likely(TAILQ_EMPTY(&shared_resource->nomem_io))) {
2823 527 : bdev_io_increment_outstanding(bdev_ch, shared_resource);
2824 527 : bdev_io->internal.f.in_submit_request = true;
2825 527 : bdev_submit_request(bdev, ch, bdev_io);
2826 527 : bdev_io->internal.f.in_submit_request = false;
2827 527 : } else {
2828 43 : bdev_queue_nomem_io_tail(shared_resource, bdev_io, BDEV_IO_RETRY_STATE_SUBMIT);
2829 43 : if (shared_resource->nomem_threshold == 0 && shared_resource->io_outstanding == 0) {
2830 : /* Special case when we have nomem IOs and no outstanding IOs which completions
2831 : * could trigger retry of queued IOs */
2832 0 : bdev_shared_ch_retry_io(shared_resource);
2833 0 : }
2834 : }
2835 574 : }
2836 :
2837 : static bool
2838 25 : bdev_qos_queue_io(struct spdk_bdev_qos *qos, struct spdk_bdev_io *bdev_io)
2839 : {
2840 : int i;
2841 :
2842 25 : if (bdev_qos_io_to_limit(bdev_io) == true) {
2843 100 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
2844 82 : if (!qos->rate_limits[i].queue_io) {
2845 5 : continue;
2846 : }
2847 :
2848 231 : if (qos->rate_limits[i].queue_io(&qos->rate_limits[i],
2849 154 : bdev_io) == true) {
2850 10 : for (i -= 1; i >= 0 ; i--) {
2851 5 : if (!qos->rate_limits[i].queue_io) {
2852 0 : continue;
2853 : }
2854 :
2855 5 : qos->rate_limits[i].rewind_quota(&qos->rate_limits[i], bdev_io);
2856 5 : }
2857 5 : return true;
2858 : }
2859 72 : }
2860 18 : }
2861 :
2862 20 : return false;
2863 25 : }
2864 :
2865 : static int
2866 27 : bdev_qos_io_submit(struct spdk_bdev_channel *ch, struct spdk_bdev_qos *qos)
2867 : {
2868 27 : struct spdk_bdev_io *bdev_io = NULL, *tmp = NULL;
2869 27 : int submitted_ios = 0;
2870 :
2871 52 : TAILQ_FOREACH_SAFE(bdev_io, &ch->qos_queued_io, internal.link, tmp) {
2872 25 : if (!bdev_qos_queue_io(qos, bdev_io)) {
2873 20 : TAILQ_REMOVE(&ch->qos_queued_io, bdev_io, internal.link);
2874 20 : bdev_io_do_submit(ch, bdev_io);
2875 :
2876 20 : submitted_ios++;
2877 20 : }
2878 25 : }
2879 :
2880 27 : return submitted_ios;
2881 : }
2882 :
2883 : static void
2884 2 : bdev_queue_io_wait_with_cb(struct spdk_bdev_io *bdev_io, spdk_bdev_io_wait_cb cb_fn)
2885 : {
2886 : int rc;
2887 :
2888 2 : bdev_io->internal.waitq_entry.bdev = bdev_io->bdev;
2889 2 : bdev_io->internal.waitq_entry.cb_fn = cb_fn;
2890 2 : bdev_io->internal.waitq_entry.cb_arg = bdev_io;
2891 4 : rc = spdk_bdev_queue_io_wait(bdev_io->bdev, spdk_io_channel_from_ctx(bdev_io->internal.ch),
2892 2 : &bdev_io->internal.waitq_entry);
2893 2 : if (rc != 0) {
2894 0 : SPDK_ERRLOG("Queue IO failed, rc=%d\n", rc);
2895 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
2896 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
2897 0 : }
2898 2 : }
2899 :
2900 : static bool
2901 621 : bdev_rw_should_split(struct spdk_bdev_io *bdev_io)
2902 : {
2903 : uint32_t io_boundary;
2904 621 : struct spdk_bdev *bdev = bdev_io->bdev;
2905 621 : uint32_t max_segment_size = bdev->max_segment_size;
2906 621 : uint32_t max_size = bdev->max_rw_size;
2907 621 : int max_segs = bdev->max_num_segments;
2908 :
2909 621 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE && bdev->split_on_write_unit) {
2910 24 : io_boundary = bdev->write_unit_size;
2911 621 : } else if (bdev->split_on_optimal_io_boundary) {
2912 168 : io_boundary = bdev->optimal_io_boundary;
2913 168 : } else {
2914 429 : io_boundary = 0;
2915 : }
2916 :
2917 621 : if (spdk_likely(!io_boundary && !max_segs && !max_segment_size && !max_size)) {
2918 243 : return false;
2919 : }
2920 :
2921 378 : if (io_boundary) {
2922 : uint64_t start_stripe, end_stripe;
2923 :
2924 192 : start_stripe = bdev_io->u.bdev.offset_blocks;
2925 192 : end_stripe = start_stripe + bdev_io->u.bdev.num_blocks - 1;
2926 : /* Avoid expensive div operations if possible. These spdk_u32 functions are very cheap. */
2927 192 : if (spdk_likely(spdk_u32_is_pow2(io_boundary))) {
2928 192 : start_stripe >>= spdk_u32log2(io_boundary);
2929 192 : end_stripe >>= spdk_u32log2(io_boundary);
2930 192 : } else {
2931 0 : start_stripe /= io_boundary;
2932 0 : end_stripe /= io_boundary;
2933 : }
2934 :
2935 192 : if (start_stripe != end_stripe) {
2936 75 : return true;
2937 : }
2938 117 : }
2939 :
2940 303 : if (max_segs) {
2941 150 : if (bdev_io->u.bdev.iovcnt > max_segs) {
2942 15 : return true;
2943 : }
2944 135 : }
2945 :
2946 288 : if (max_segment_size) {
2947 470 : for (int i = 0; i < bdev_io->u.bdev.iovcnt; i++) {
2948 346 : if (bdev_io->u.bdev.iovs[i].iov_len > max_segment_size) {
2949 12 : return true;
2950 : }
2951 334 : }
2952 124 : }
2953 :
2954 276 : if (max_size) {
2955 52 : if (bdev_io->u.bdev.num_blocks > max_size) {
2956 7 : return true;
2957 : }
2958 45 : }
2959 :
2960 269 : return false;
2961 621 : }
2962 :
2963 : static bool
2964 24 : bdev_unmap_should_split(struct spdk_bdev_io *bdev_io)
2965 : {
2966 : uint32_t num_unmap_segments;
2967 :
2968 24 : if (!bdev_io->bdev->max_unmap || !bdev_io->bdev->max_unmap_segments) {
2969 3 : return false;
2970 : }
2971 21 : num_unmap_segments = spdk_divide_round_up(bdev_io->u.bdev.num_blocks, bdev_io->bdev->max_unmap);
2972 21 : if (num_unmap_segments > bdev_io->bdev->max_unmap_segments) {
2973 4 : return true;
2974 : }
2975 :
2976 17 : return false;
2977 24 : }
2978 :
2979 : static bool
2980 37 : bdev_write_zeroes_should_split(struct spdk_bdev_io *bdev_io)
2981 : {
2982 37 : if (!bdev_io->bdev->max_write_zeroes) {
2983 4 : return false;
2984 : }
2985 :
2986 33 : if (bdev_io->u.bdev.num_blocks > bdev_io->bdev->max_write_zeroes) {
2987 10 : return true;
2988 : }
2989 :
2990 23 : return false;
2991 37 : }
2992 :
2993 : static bool
2994 30 : bdev_copy_should_split(struct spdk_bdev_io *bdev_io)
2995 : {
2996 30 : if (bdev_io->bdev->max_copy != 0 &&
2997 25 : bdev_io->u.bdev.num_blocks > bdev_io->bdev->max_copy) {
2998 6 : return true;
2999 : }
3000 :
3001 24 : return false;
3002 30 : }
3003 :
3004 : static bool
3005 792 : bdev_io_should_split(struct spdk_bdev_io *bdev_io)
3006 : {
3007 792 : switch (bdev_io->type) {
3008 : case SPDK_BDEV_IO_TYPE_READ:
3009 : case SPDK_BDEV_IO_TYPE_WRITE:
3010 621 : return bdev_rw_should_split(bdev_io);
3011 : case SPDK_BDEV_IO_TYPE_UNMAP:
3012 24 : return bdev_unmap_should_split(bdev_io);
3013 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3014 37 : return bdev_write_zeroes_should_split(bdev_io);
3015 : case SPDK_BDEV_IO_TYPE_COPY:
3016 30 : return bdev_copy_should_split(bdev_io);
3017 : default:
3018 80 : return false;
3019 : }
3020 792 : }
3021 :
3022 : static uint32_t
3023 249 : _to_next_boundary(uint64_t offset, uint32_t boundary)
3024 : {
3025 249 : return (boundary - (offset % boundary));
3026 : }
3027 :
3028 : static void bdev_io_split_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg);
3029 :
3030 : static void _bdev_rw_split(void *_bdev_io);
3031 :
3032 : static void bdev_unmap_split(struct spdk_bdev_io *bdev_io);
3033 :
3034 : static void
3035 0 : _bdev_unmap_split(void *_bdev_io)
3036 : {
3037 0 : return bdev_unmap_split((struct spdk_bdev_io *)_bdev_io);
3038 : }
3039 :
3040 : static void bdev_write_zeroes_split(struct spdk_bdev_io *bdev_io);
3041 :
3042 : static void
3043 0 : _bdev_write_zeroes_split(void *_bdev_io)
3044 : {
3045 0 : return bdev_write_zeroes_split((struct spdk_bdev_io *)_bdev_io);
3046 : }
3047 :
3048 : static void bdev_copy_split(struct spdk_bdev_io *bdev_io);
3049 :
3050 : static void
3051 0 : _bdev_copy_split(void *_bdev_io)
3052 : {
3053 0 : return bdev_copy_split((struct spdk_bdev_io *)_bdev_io);
3054 : }
3055 :
3056 : static int
3057 305 : bdev_io_split_submit(struct spdk_bdev_io *bdev_io, struct iovec *iov, int iovcnt, void *md_buf,
3058 : uint64_t num_blocks, uint64_t *offset, uint64_t *remaining)
3059 : {
3060 : int rc;
3061 : uint64_t current_offset, current_remaining, current_src_offset;
3062 : spdk_bdev_io_wait_cb io_wait_fn;
3063 :
3064 305 : current_offset = *offset;
3065 305 : current_remaining = *remaining;
3066 :
3067 305 : assert(bdev_io->internal.f.split);
3068 :
3069 305 : bdev_io->internal.split.outstanding++;
3070 :
3071 305 : io_wait_fn = _bdev_rw_split;
3072 305 : switch (bdev_io->type) {
3073 : case SPDK_BDEV_IO_TYPE_READ:
3074 196 : assert(bdev_io->u.bdev.accel_sequence == NULL);
3075 392 : rc = bdev_readv_blocks_with_md(bdev_io->internal.desc,
3076 196 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
3077 196 : iov, iovcnt, md_buf, current_offset,
3078 196 : num_blocks,
3079 196 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain : NULL,
3080 196 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain_ctx : NULL,
3081 : NULL,
3082 196 : bdev_io->u.bdev.dif_check_flags,
3083 196 : bdev_io_split_done, bdev_io);
3084 196 : break;
3085 : case SPDK_BDEV_IO_TYPE_WRITE:
3086 50 : assert(bdev_io->u.bdev.accel_sequence == NULL);
3087 100 : rc = bdev_writev_blocks_with_md(bdev_io->internal.desc,
3088 50 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
3089 50 : iov, iovcnt, md_buf, current_offset,
3090 50 : num_blocks,
3091 50 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain : NULL,
3092 50 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain_ctx : NULL,
3093 : NULL,
3094 50 : bdev_io->u.bdev.dif_check_flags,
3095 50 : bdev_io->u.bdev.nvme_cdw12.raw,
3096 50 : bdev_io->u.bdev.nvme_cdw13.raw,
3097 50 : bdev_io_split_done, bdev_io);
3098 50 : break;
3099 : case SPDK_BDEV_IO_TYPE_UNMAP:
3100 17 : io_wait_fn = _bdev_unmap_split;
3101 34 : rc = spdk_bdev_unmap_blocks(bdev_io->internal.desc,
3102 17 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
3103 17 : current_offset, num_blocks,
3104 17 : bdev_io_split_done, bdev_io);
3105 17 : break;
3106 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3107 23 : io_wait_fn = _bdev_write_zeroes_split;
3108 46 : rc = spdk_bdev_write_zeroes_blocks(bdev_io->internal.desc,
3109 23 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
3110 23 : current_offset, num_blocks,
3111 23 : bdev_io_split_done, bdev_io);
3112 23 : break;
3113 : case SPDK_BDEV_IO_TYPE_COPY:
3114 19 : io_wait_fn = _bdev_copy_split;
3115 38 : current_src_offset = bdev_io->u.bdev.copy.src_offset_blocks +
3116 19 : (current_offset - bdev_io->u.bdev.offset_blocks);
3117 38 : rc = spdk_bdev_copy_blocks(bdev_io->internal.desc,
3118 19 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
3119 19 : current_offset, current_src_offset, num_blocks,
3120 19 : bdev_io_split_done, bdev_io);
3121 19 : break;
3122 : default:
3123 0 : assert(false);
3124 : rc = -EINVAL;
3125 : break;
3126 : }
3127 :
3128 305 : if (rc == 0) {
3129 301 : current_offset += num_blocks;
3130 301 : current_remaining -= num_blocks;
3131 301 : bdev_io->internal.split.current_offset_blocks = current_offset;
3132 301 : bdev_io->internal.split.remaining_num_blocks = current_remaining;
3133 301 : *offset = current_offset;
3134 301 : *remaining = current_remaining;
3135 301 : } else {
3136 4 : bdev_io->internal.split.outstanding--;
3137 4 : if (rc == -ENOMEM) {
3138 4 : if (bdev_io->internal.split.outstanding == 0) {
3139 : /* No I/O is outstanding. Hence we should wait here. */
3140 1 : bdev_queue_io_wait_with_cb(bdev_io, io_wait_fn);
3141 1 : }
3142 4 : } else {
3143 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
3144 0 : if (bdev_io->internal.split.outstanding == 0) {
3145 0 : bdev_ch_remove_from_io_submitted(bdev_io);
3146 0 : spdk_trace_record(TRACE_BDEV_IO_DONE, bdev_io->internal.ch->trace_id,
3147 : 0, (uintptr_t)bdev_io, bdev_io->internal.caller_ctx,
3148 : bdev_io->internal.ch->queue_depth);
3149 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
3150 0 : }
3151 : }
3152 : }
3153 :
3154 305 : return rc;
3155 : }
3156 :
3157 : static void
3158 67 : _bdev_rw_split(void *_bdev_io)
3159 : {
3160 : struct iovec *parent_iov, *iov;
3161 67 : struct spdk_bdev_io *bdev_io = _bdev_io;
3162 67 : struct spdk_bdev *bdev = bdev_io->bdev;
3163 : uint64_t parent_offset, current_offset, remaining;
3164 : uint32_t parent_iov_offset, parent_iovcnt, parent_iovpos, child_iovcnt;
3165 : uint32_t to_next_boundary, to_next_boundary_bytes, to_last_block_bytes;
3166 : uint32_t iovcnt, iov_len, child_iovsize;
3167 67 : uint32_t blocklen = bdev->blocklen;
3168 : uint32_t io_boundary;
3169 67 : uint32_t max_segment_size = bdev->max_segment_size;
3170 67 : uint32_t max_child_iovcnt = bdev->max_num_segments;
3171 67 : uint32_t max_size = bdev->max_rw_size;
3172 67 : void *md_buf = NULL;
3173 : int rc;
3174 :
3175 67 : max_size = max_size ? max_size : UINT32_MAX;
3176 67 : max_segment_size = max_segment_size ? max_segment_size : UINT32_MAX;
3177 67 : max_child_iovcnt = max_child_iovcnt ? spdk_min(max_child_iovcnt, SPDK_BDEV_IO_NUM_CHILD_IOV) :
3178 : SPDK_BDEV_IO_NUM_CHILD_IOV;
3179 :
3180 67 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE && bdev->split_on_write_unit) {
3181 5 : io_boundary = bdev->write_unit_size;
3182 67 : } else if (bdev->split_on_optimal_io_boundary) {
3183 40 : io_boundary = bdev->optimal_io_boundary;
3184 40 : } else {
3185 22 : io_boundary = UINT32_MAX;
3186 : }
3187 :
3188 67 : assert(bdev_io->internal.f.split);
3189 :
3190 67 : remaining = bdev_io->internal.split.remaining_num_blocks;
3191 67 : current_offset = bdev_io->internal.split.current_offset_blocks;
3192 67 : parent_offset = bdev_io->u.bdev.offset_blocks;
3193 67 : parent_iov_offset = (current_offset - parent_offset) * blocklen;
3194 67 : parent_iovcnt = bdev_io->u.bdev.iovcnt;
3195 :
3196 420 : for (parent_iovpos = 0; parent_iovpos < parent_iovcnt; parent_iovpos++) {
3197 420 : parent_iov = &bdev_io->u.bdev.iovs[parent_iovpos];
3198 420 : if (parent_iov_offset < parent_iov->iov_len) {
3199 67 : break;
3200 : }
3201 353 : parent_iov_offset -= parent_iov->iov_len;
3202 353 : }
3203 :
3204 67 : child_iovcnt = 0;
3205 573 : while (remaining > 0 && parent_iovpos < parent_iovcnt &&
3206 264 : child_iovcnt < SPDK_BDEV_IO_NUM_CHILD_IOV) {
3207 249 : to_next_boundary = _to_next_boundary(current_offset, io_boundary);
3208 249 : to_next_boundary = spdk_min(remaining, to_next_boundary);
3209 249 : to_next_boundary = spdk_min(max_size, to_next_boundary);
3210 249 : to_next_boundary_bytes = to_next_boundary * blocklen;
3211 :
3212 249 : iov = &bdev_io->child_iov[child_iovcnt];
3213 249 : iovcnt = 0;
3214 :
3215 249 : if (bdev_io->u.bdev.md_buf) {
3216 48 : md_buf = (char *)bdev_io->u.bdev.md_buf +
3217 24 : (current_offset - parent_offset) * spdk_bdev_get_md_size(bdev);
3218 24 : }
3219 :
3220 249 : child_iovsize = spdk_min(SPDK_BDEV_IO_NUM_CHILD_IOV - child_iovcnt, max_child_iovcnt);
3221 1810 : while (to_next_boundary_bytes > 0 && parent_iovpos < parent_iovcnt &&
3222 836 : iovcnt < child_iovsize) {
3223 725 : parent_iov = &bdev_io->u.bdev.iovs[parent_iovpos];
3224 725 : iov_len = parent_iov->iov_len - parent_iov_offset;
3225 :
3226 725 : iov_len = spdk_min(iov_len, max_segment_size);
3227 725 : iov_len = spdk_min(iov_len, to_next_boundary_bytes);
3228 725 : to_next_boundary_bytes -= iov_len;
3229 :
3230 725 : bdev_io->child_iov[child_iovcnt].iov_base = parent_iov->iov_base + parent_iov_offset;
3231 725 : bdev_io->child_iov[child_iovcnt].iov_len = iov_len;
3232 :
3233 725 : if (iov_len < parent_iov->iov_len - parent_iov_offset) {
3234 183 : parent_iov_offset += iov_len;
3235 183 : } else {
3236 542 : parent_iovpos++;
3237 542 : parent_iov_offset = 0;
3238 : }
3239 725 : child_iovcnt++;
3240 725 : iovcnt++;
3241 : }
3242 :
3243 249 : if (to_next_boundary_bytes > 0) {
3244 : /* We had to stop this child I/O early because we ran out of
3245 : * child_iov space or were limited by max_num_segments.
3246 : * Ensure the iovs to be aligned with block size and
3247 : * then adjust to_next_boundary before starting the
3248 : * child I/O.
3249 : */
3250 111 : assert(child_iovcnt == SPDK_BDEV_IO_NUM_CHILD_IOV ||
3251 : iovcnt == child_iovsize);
3252 111 : to_last_block_bytes = to_next_boundary_bytes % blocklen;
3253 111 : if (to_last_block_bytes != 0) {
3254 24 : uint32_t child_iovpos = child_iovcnt - 1;
3255 : /* don't decrease child_iovcnt when it equals to SPDK_BDEV_IO_NUM_CHILD_IOV
3256 : * so the loop will naturally end
3257 : */
3258 :
3259 24 : to_last_block_bytes = blocklen - to_last_block_bytes;
3260 24 : to_next_boundary_bytes += to_last_block_bytes;
3261 53 : while (to_last_block_bytes > 0 && iovcnt > 0) {
3262 32 : iov_len = spdk_min(to_last_block_bytes,
3263 : bdev_io->child_iov[child_iovpos].iov_len);
3264 32 : bdev_io->child_iov[child_iovpos].iov_len -= iov_len;
3265 32 : if (bdev_io->child_iov[child_iovpos].iov_len == 0) {
3266 15 : child_iovpos--;
3267 15 : if (--iovcnt == 0) {
3268 : /* If the child IO is less than a block size just return.
3269 : * If the first child IO of any split round is less than
3270 : * a block size, an error exit.
3271 : */
3272 3 : if (bdev_io->internal.split.outstanding == 0) {
3273 1 : SPDK_ERRLOG("The first child io was less than a block size\n");
3274 1 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
3275 1 : bdev_ch_remove_from_io_submitted(bdev_io);
3276 1 : spdk_trace_record(TRACE_BDEV_IO_DONE, bdev_io->internal.ch->trace_id,
3277 : 0, (uintptr_t)bdev_io, bdev_io->internal.caller_ctx,
3278 : bdev_io->internal.ch->queue_depth);
3279 1 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
3280 1 : }
3281 :
3282 3 : return;
3283 : }
3284 12 : }
3285 :
3286 29 : to_last_block_bytes -= iov_len;
3287 :
3288 29 : if (parent_iov_offset == 0) {
3289 14 : parent_iovpos--;
3290 14 : parent_iov_offset = bdev_io->u.bdev.iovs[parent_iovpos].iov_len;
3291 14 : }
3292 29 : parent_iov_offset -= iov_len;
3293 : }
3294 :
3295 21 : assert(to_last_block_bytes == 0);
3296 21 : }
3297 108 : to_next_boundary -= to_next_boundary_bytes / blocklen;
3298 108 : }
3299 :
3300 246 : rc = bdev_io_split_submit(bdev_io, iov, iovcnt, md_buf, to_next_boundary,
3301 : ¤t_offset, &remaining);
3302 246 : if (spdk_unlikely(rc)) {
3303 4 : return;
3304 : }
3305 : }
3306 67 : }
3307 :
3308 : static void
3309 3 : bdev_unmap_split(struct spdk_bdev_io *bdev_io)
3310 : {
3311 : uint64_t offset, unmap_blocks, remaining, max_unmap_blocks;
3312 3 : uint32_t num_children_reqs = 0;
3313 : int rc;
3314 :
3315 3 : assert(bdev_io->internal.f.split);
3316 :
3317 3 : offset = bdev_io->internal.split.current_offset_blocks;
3318 3 : remaining = bdev_io->internal.split.remaining_num_blocks;
3319 3 : max_unmap_blocks = bdev_io->bdev->max_unmap * bdev_io->bdev->max_unmap_segments;
3320 :
3321 20 : while (remaining && (num_children_reqs < SPDK_BDEV_MAX_CHILDREN_UNMAP_WRITE_ZEROES_REQS)) {
3322 17 : unmap_blocks = spdk_min(remaining, max_unmap_blocks);
3323 :
3324 17 : rc = bdev_io_split_submit(bdev_io, NULL, 0, NULL, unmap_blocks,
3325 : &offset, &remaining);
3326 17 : if (spdk_likely(rc == 0)) {
3327 17 : num_children_reqs++;
3328 17 : } else {
3329 0 : return;
3330 : }
3331 : }
3332 3 : }
3333 :
3334 : static void
3335 6 : bdev_write_zeroes_split(struct spdk_bdev_io *bdev_io)
3336 : {
3337 : uint64_t offset, write_zeroes_blocks, remaining;
3338 6 : uint32_t num_children_reqs = 0;
3339 : int rc;
3340 :
3341 6 : assert(bdev_io->internal.f.split);
3342 :
3343 6 : offset = bdev_io->internal.split.current_offset_blocks;
3344 6 : remaining = bdev_io->internal.split.remaining_num_blocks;
3345 :
3346 29 : while (remaining && (num_children_reqs < SPDK_BDEV_MAX_CHILDREN_UNMAP_WRITE_ZEROES_REQS)) {
3347 23 : write_zeroes_blocks = spdk_min(remaining, bdev_io->bdev->max_write_zeroes);
3348 :
3349 23 : rc = bdev_io_split_submit(bdev_io, NULL, 0, NULL, write_zeroes_blocks,
3350 : &offset, &remaining);
3351 23 : if (spdk_likely(rc == 0)) {
3352 23 : num_children_reqs++;
3353 23 : } else {
3354 0 : return;
3355 : }
3356 : }
3357 6 : }
3358 :
3359 : static void
3360 4 : bdev_copy_split(struct spdk_bdev_io *bdev_io)
3361 : {
3362 : uint64_t offset, copy_blocks, remaining;
3363 4 : uint32_t num_children_reqs = 0;
3364 : int rc;
3365 :
3366 4 : assert(bdev_io->internal.f.split);
3367 :
3368 4 : offset = bdev_io->internal.split.current_offset_blocks;
3369 4 : remaining = bdev_io->internal.split.remaining_num_blocks;
3370 :
3371 4 : assert(bdev_io->bdev->max_copy != 0);
3372 23 : while (remaining && (num_children_reqs < SPDK_BDEV_MAX_CHILDREN_COPY_REQS)) {
3373 19 : copy_blocks = spdk_min(remaining, bdev_io->bdev->max_copy);
3374 :
3375 19 : rc = bdev_io_split_submit(bdev_io, NULL, 0, NULL, copy_blocks,
3376 : &offset, &remaining);
3377 19 : if (spdk_likely(rc == 0)) {
3378 19 : num_children_reqs++;
3379 19 : } else {
3380 0 : return;
3381 : }
3382 : }
3383 4 : }
3384 :
3385 : static void
3386 58 : parent_bdev_io_complete(void *ctx, int rc)
3387 : {
3388 58 : struct spdk_bdev_io *parent_io = ctx;
3389 :
3390 58 : if (rc) {
3391 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
3392 0 : }
3393 :
3394 116 : parent_io->internal.cb(parent_io, parent_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS,
3395 58 : parent_io->internal.caller_ctx);
3396 58 : }
3397 :
3398 : static void
3399 0 : bdev_io_complete_parent_sequence_cb(void *ctx, int status)
3400 : {
3401 0 : struct spdk_bdev_io *bdev_io = ctx;
3402 :
3403 : /* u.bdev.accel_sequence should have already been cleared at this point */
3404 0 : assert(bdev_io->u.bdev.accel_sequence == NULL);
3405 0 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS);
3406 0 : bdev_io->internal.f.has_accel_sequence = false;
3407 :
3408 0 : if (spdk_unlikely(status != 0)) {
3409 0 : SPDK_ERRLOG("Failed to execute accel sequence, status=%d\n", status);
3410 0 : }
3411 :
3412 0 : parent_bdev_io_complete(bdev_io, status);
3413 0 : }
3414 :
3415 : static void
3416 301 : bdev_io_split_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
3417 : {
3418 301 : struct spdk_bdev_io *parent_io = cb_arg;
3419 :
3420 301 : spdk_bdev_free_io(bdev_io);
3421 :
3422 301 : assert(parent_io->internal.f.split);
3423 :
3424 301 : if (!success) {
3425 21 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
3426 : /* If any child I/O failed, stop further splitting process. */
3427 21 : parent_io->internal.split.current_offset_blocks += parent_io->internal.split.remaining_num_blocks;
3428 21 : parent_io->internal.split.remaining_num_blocks = 0;
3429 21 : }
3430 301 : parent_io->internal.split.outstanding--;
3431 301 : if (parent_io->internal.split.outstanding != 0) {
3432 223 : return;
3433 : }
3434 :
3435 : /*
3436 : * Parent I/O finishes when all blocks are consumed.
3437 : */
3438 78 : if (parent_io->internal.split.remaining_num_blocks == 0) {
3439 58 : assert(parent_io->internal.cb != bdev_io_split_done);
3440 58 : bdev_ch_remove_from_io_submitted(parent_io);
3441 58 : spdk_trace_record(TRACE_BDEV_IO_DONE, parent_io->internal.ch->trace_id,
3442 : 0, (uintptr_t)parent_io, bdev_io->internal.caller_ctx,
3443 : parent_io->internal.ch->queue_depth);
3444 :
3445 58 : if (spdk_likely(parent_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS)) {
3446 48 : if (bdev_io_needs_sequence_exec(parent_io->internal.desc, parent_io)) {
3447 0 : bdev_io_exec_sequence(parent_io, bdev_io_complete_parent_sequence_cb);
3448 0 : return;
3449 48 : } else if (parent_io->internal.f.has_bounce_buf &&
3450 0 : !bdev_io_use_accel_sequence(bdev_io)) {
3451 : /* bdev IO will be completed in the callback */
3452 0 : _bdev_io_push_bounce_data_buffer(parent_io, parent_bdev_io_complete);
3453 0 : return;
3454 : }
3455 48 : }
3456 :
3457 58 : parent_bdev_io_complete(parent_io, 0);
3458 58 : return;
3459 : }
3460 :
3461 : /*
3462 : * Continue with the splitting process. This function will complete the parent I/O if the
3463 : * splitting is done.
3464 : */
3465 20 : switch (parent_io->type) {
3466 : case SPDK_BDEV_IO_TYPE_READ:
3467 : case SPDK_BDEV_IO_TYPE_WRITE:
3468 17 : _bdev_rw_split(parent_io);
3469 17 : break;
3470 : case SPDK_BDEV_IO_TYPE_UNMAP:
3471 1 : bdev_unmap_split(parent_io);
3472 1 : break;
3473 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3474 1 : bdev_write_zeroes_split(parent_io);
3475 1 : break;
3476 : case SPDK_BDEV_IO_TYPE_COPY:
3477 1 : bdev_copy_split(parent_io);
3478 1 : break;
3479 : default:
3480 0 : assert(false);
3481 : break;
3482 : }
3483 301 : }
3484 :
3485 : static void bdev_rw_split_get_buf_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io,
3486 : bool success);
3487 :
3488 : static void
3489 59 : bdev_io_split(struct spdk_bdev_io *bdev_io)
3490 : {
3491 59 : assert(bdev_io_should_split(bdev_io));
3492 59 : assert(bdev_io->internal.f.split);
3493 :
3494 59 : bdev_io->internal.split.current_offset_blocks = bdev_io->u.bdev.offset_blocks;
3495 59 : bdev_io->internal.split.remaining_num_blocks = bdev_io->u.bdev.num_blocks;
3496 59 : bdev_io->internal.split.outstanding = 0;
3497 59 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
3498 :
3499 59 : switch (bdev_io->type) {
3500 : case SPDK_BDEV_IO_TYPE_READ:
3501 : case SPDK_BDEV_IO_TYPE_WRITE:
3502 49 : if (_is_buf_allocated(bdev_io->u.bdev.iovs)) {
3503 49 : _bdev_rw_split(bdev_io);
3504 49 : } else {
3505 0 : assert(bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
3506 0 : spdk_bdev_io_get_buf(bdev_io, bdev_rw_split_get_buf_cb,
3507 0 : bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen);
3508 : }
3509 49 : break;
3510 : case SPDK_BDEV_IO_TYPE_UNMAP:
3511 2 : bdev_unmap_split(bdev_io);
3512 2 : break;
3513 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3514 5 : bdev_write_zeroes_split(bdev_io);
3515 5 : break;
3516 : case SPDK_BDEV_IO_TYPE_COPY:
3517 3 : bdev_copy_split(bdev_io);
3518 3 : break;
3519 : default:
3520 0 : assert(false);
3521 : break;
3522 : }
3523 59 : }
3524 :
3525 : static void
3526 0 : bdev_rw_split_get_buf_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io, bool success)
3527 : {
3528 0 : if (!success) {
3529 0 : spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
3530 0 : return;
3531 : }
3532 :
3533 0 : _bdev_rw_split(bdev_io);
3534 0 : }
3535 :
3536 : /* Explicitly mark this inline, since it's used as a function pointer and otherwise won't
3537 : * be inlined, at least on some compilers.
3538 : */
3539 : static inline void
3540 579 : _bdev_io_submit(void *ctx)
3541 : {
3542 579 : struct spdk_bdev_io *bdev_io = ctx;
3543 579 : struct spdk_bdev *bdev = bdev_io->bdev;
3544 579 : struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
3545 :
3546 579 : if (spdk_likely(bdev_ch->flags == 0)) {
3547 554 : bdev_io_do_submit(bdev_ch, bdev_io);
3548 554 : return;
3549 : }
3550 :
3551 25 : if (bdev_ch->flags & BDEV_CH_RESET_IN_PROGRESS) {
3552 2 : _bdev_io_complete_in_submit(bdev_ch, bdev_io, SPDK_BDEV_IO_STATUS_ABORTED);
3553 25 : } else if (bdev_ch->flags & BDEV_CH_QOS_ENABLED) {
3554 23 : if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_ABORT) &&
3555 2 : bdev_abort_queued_io(&bdev_ch->qos_queued_io, bdev_io->u.abort.bio_to_abort)) {
3556 0 : _bdev_io_complete_in_submit(bdev_ch, bdev_io, SPDK_BDEV_IO_STATUS_SUCCESS);
3557 0 : } else {
3558 23 : TAILQ_INSERT_TAIL(&bdev_ch->qos_queued_io, bdev_io, internal.link);
3559 23 : bdev_qos_io_submit(bdev_ch, bdev->internal.qos);
3560 : }
3561 23 : } else {
3562 0 : SPDK_ERRLOG("unknown bdev_ch flag %x found\n", bdev_ch->flags);
3563 0 : _bdev_io_complete_in_submit(bdev_ch, bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
3564 : }
3565 579 : }
3566 :
3567 : bool bdev_lba_range_overlapped(struct lba_range *range1, struct lba_range *range2);
3568 :
3569 : bool
3570 23 : bdev_lba_range_overlapped(struct lba_range *range1, struct lba_range *range2)
3571 : {
3572 23 : if (range1->length == 0 || range2->length == 0) {
3573 1 : return false;
3574 : }
3575 :
3576 22 : if (range1->offset + range1->length <= range2->offset) {
3577 1 : return false;
3578 : }
3579 :
3580 21 : if (range2->offset + range2->length <= range1->offset) {
3581 3 : return false;
3582 : }
3583 :
3584 18 : return true;
3585 23 : }
3586 :
3587 : static bool
3588 11 : bdev_io_range_is_locked(struct spdk_bdev_io *bdev_io, struct lba_range *range)
3589 : {
3590 11 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
3591 : struct lba_range r;
3592 :
3593 11 : switch (bdev_io->type) {
3594 : case SPDK_BDEV_IO_TYPE_NVME_IO:
3595 : case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
3596 : /* Don't try to decode the NVMe command - just assume worst-case and that
3597 : * it overlaps a locked range.
3598 : */
3599 0 : return true;
3600 : case SPDK_BDEV_IO_TYPE_READ:
3601 6 : if (!range->quiesce) {
3602 4 : return false;
3603 : }
3604 : /* fallthrough */
3605 : case SPDK_BDEV_IO_TYPE_WRITE:
3606 : case SPDK_BDEV_IO_TYPE_UNMAP:
3607 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3608 : case SPDK_BDEV_IO_TYPE_ZCOPY:
3609 : case SPDK_BDEV_IO_TYPE_COPY:
3610 7 : r.offset = bdev_io->u.bdev.offset_blocks;
3611 7 : r.length = bdev_io->u.bdev.num_blocks;
3612 7 : if (!bdev_lba_range_overlapped(range, &r)) {
3613 : /* This I/O doesn't overlap the specified LBA range. */
3614 0 : return false;
3615 7 : } else if (range->owner_ch == ch && range->locked_ctx == bdev_io->internal.caller_ctx) {
3616 : /* This I/O overlaps, but the I/O is on the same channel that locked this
3617 : * range, and the caller_ctx is the same as the locked_ctx. This means
3618 : * that this I/O is associated with the lock, and is allowed to execute.
3619 : */
3620 2 : return false;
3621 : } else {
3622 5 : return true;
3623 : }
3624 : default:
3625 0 : return false;
3626 : }
3627 11 : }
3628 :
3629 : void
3630 639 : bdev_io_submit(struct spdk_bdev_io *bdev_io)
3631 : {
3632 639 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
3633 :
3634 639 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_PENDING);
3635 :
3636 639 : if (!TAILQ_EMPTY(&ch->locked_ranges)) {
3637 : struct lba_range *range;
3638 :
3639 13 : TAILQ_FOREACH(range, &ch->locked_ranges, tailq) {
3640 8 : if (bdev_io_range_is_locked(bdev_io, range)) {
3641 3 : TAILQ_INSERT_TAIL(&ch->io_locked, bdev_io, internal.ch_link);
3642 3 : return;
3643 : }
3644 5 : }
3645 5 : }
3646 :
3647 636 : bdev_ch_add_to_io_submitted(bdev_io);
3648 :
3649 636 : bdev_io->internal.submit_tsc = spdk_get_ticks();
3650 636 : spdk_trace_record_tsc(bdev_io->internal.submit_tsc, TRACE_BDEV_IO_START,
3651 : ch->trace_id, bdev_io->u.bdev.num_blocks,
3652 : (uintptr_t)bdev_io, (uint64_t)bdev_io->type, bdev_io->internal.caller_ctx,
3653 : bdev_io->u.bdev.offset_blocks, ch->queue_depth);
3654 :
3655 636 : if (bdev_io->internal.f.split) {
3656 59 : bdev_io_split(bdev_io);
3657 59 : return;
3658 : }
3659 :
3660 577 : _bdev_io_submit(bdev_io);
3661 639 : }
3662 :
3663 : static inline void
3664 4 : _bdev_io_ext_use_bounce_buffer(struct spdk_bdev_io *bdev_io)
3665 : {
3666 : /* bdev doesn't support memory domains, thereby buffers in this IO request can't
3667 : * be accessed directly. It is needed to allocate buffers before issuing IO operation.
3668 : * For write operation we need to pull buffers from memory domain before submitting IO.
3669 : * Once read operation completes, we need to use memory_domain push functionality to
3670 : * update data in original memory domain IO buffer
3671 : * This IO request will go through a regular IO flow, so clear memory domains pointers */
3672 4 : assert(bdev_io->internal.f.has_memory_domain);
3673 4 : bdev_io->u.bdev.memory_domain = NULL;
3674 4 : bdev_io->u.bdev.memory_domain_ctx = NULL;
3675 8 : _bdev_memory_domain_io_get_buf(bdev_io, _bdev_memory_domain_get_io_cb,
3676 4 : bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen);
3677 4 : }
3678 :
3679 : static inline void
3680 292 : _bdev_io_submit_ext(struct spdk_bdev_desc *desc, struct spdk_bdev_io *bdev_io)
3681 : {
3682 292 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
3683 292 : bool needs_exec = bdev_io_needs_sequence_exec(desc, bdev_io);
3684 :
3685 292 : if (spdk_unlikely(ch->flags & BDEV_CH_RESET_IN_PROGRESS)) {
3686 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_ABORTED;
3687 0 : bdev_io_complete_unsubmitted(bdev_io);
3688 0 : return;
3689 : }
3690 :
3691 : /* We need to allocate bounce buffer if bdev doesn't support memory domains, or if it does
3692 : * support them, but we need to execute an accel sequence and the data buffer is from accel
3693 : * memory domain (to avoid doing a push/pull from that domain).
3694 : */
3695 292 : if (bdev_io_use_memory_domain(bdev_io)) {
3696 4 : if (!desc->memory_domains_supported ||
3697 0 : (needs_exec && bdev_io->internal.memory_domain == spdk_accel_get_memory_domain())) {
3698 4 : _bdev_io_ext_use_bounce_buffer(bdev_io);
3699 4 : return;
3700 : }
3701 0 : }
3702 :
3703 288 : if (needs_exec) {
3704 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
3705 0 : bdev_io_exec_sequence(bdev_io, bdev_io_submit_sequence_cb);
3706 0 : return;
3707 : }
3708 : /* For reads we'll execute the sequence after the data is read, so, for now, only
3709 : * clear out accel_sequence pointer and submit the IO */
3710 0 : assert(bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
3711 0 : bdev_io->u.bdev.accel_sequence = NULL;
3712 0 : }
3713 :
3714 288 : bdev_io_submit(bdev_io);
3715 292 : }
3716 :
3717 : static void
3718 11 : bdev_io_submit_reset(struct spdk_bdev_io *bdev_io)
3719 : {
3720 11 : struct spdk_bdev *bdev = bdev_io->bdev;
3721 11 : struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
3722 11 : struct spdk_io_channel *ch = bdev_ch->channel;
3723 :
3724 11 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_PENDING);
3725 :
3726 11 : bdev_io->internal.f.in_submit_request = true;
3727 11 : bdev_submit_request(bdev, ch, bdev_io);
3728 11 : bdev_io->internal.f.in_submit_request = false;
3729 11 : }
3730 :
3731 : void
3732 691 : bdev_io_init(struct spdk_bdev_io *bdev_io,
3733 : struct spdk_bdev *bdev, void *cb_arg,
3734 : spdk_bdev_io_completion_cb cb)
3735 : {
3736 691 : bdev_io->bdev = bdev;
3737 691 : bdev_io->internal.f.raw = 0;
3738 691 : bdev_io->internal.caller_ctx = cb_arg;
3739 691 : bdev_io->internal.cb = cb;
3740 691 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_PENDING;
3741 691 : bdev_io->internal.f.in_submit_request = false;
3742 691 : bdev_io->internal.error.nvme.cdw0 = 0;
3743 691 : bdev_io->num_retries = 0;
3744 691 : bdev_io->internal.get_buf_cb = NULL;
3745 691 : bdev_io->internal.get_aux_buf_cb = NULL;
3746 691 : bdev_io->internal.data_transfer_cpl = NULL;
3747 691 : bdev_io->internal.f.split = bdev_io_should_split(bdev_io);
3748 691 : }
3749 :
3750 : static bool
3751 517 : bdev_io_type_supported(struct spdk_bdev *bdev, enum spdk_bdev_io_type io_type)
3752 : {
3753 517 : return bdev->fn_table->io_type_supported(bdev->ctxt, io_type);
3754 : }
3755 :
3756 : bool
3757 169 : spdk_bdev_io_type_supported(struct spdk_bdev *bdev, enum spdk_bdev_io_type io_type)
3758 : {
3759 : bool supported;
3760 :
3761 169 : supported = bdev_io_type_supported(bdev, io_type);
3762 :
3763 169 : if (!supported) {
3764 7 : switch (io_type) {
3765 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3766 : /* The bdev layer will emulate write zeroes as long as write is supported. */
3767 0 : supported = bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE);
3768 0 : break;
3769 : default:
3770 7 : break;
3771 : }
3772 7 : }
3773 :
3774 169 : return supported;
3775 : }
3776 :
3777 : static const char *g_io_type_strings[] = {
3778 : [SPDK_BDEV_IO_TYPE_READ] = "read",
3779 : [SPDK_BDEV_IO_TYPE_WRITE] = "write",
3780 : [SPDK_BDEV_IO_TYPE_UNMAP] = "unmap",
3781 : [SPDK_BDEV_IO_TYPE_FLUSH] = "flush",
3782 : [SPDK_BDEV_IO_TYPE_RESET] = "reset",
3783 : [SPDK_BDEV_IO_TYPE_NVME_ADMIN] = "nvme_admin",
3784 : [SPDK_BDEV_IO_TYPE_NVME_IO] = "nvme_io",
3785 : [SPDK_BDEV_IO_TYPE_NVME_IO_MD] = "nvme_io_md",
3786 : [SPDK_BDEV_IO_TYPE_WRITE_ZEROES] = "write_zeroes",
3787 : [SPDK_BDEV_IO_TYPE_ZCOPY] = "zcopy",
3788 : [SPDK_BDEV_IO_TYPE_GET_ZONE_INFO] = "get_zone_info",
3789 : [SPDK_BDEV_IO_TYPE_ZONE_MANAGEMENT] = "zone_management",
3790 : [SPDK_BDEV_IO_TYPE_ZONE_APPEND] = "zone_append",
3791 : [SPDK_BDEV_IO_TYPE_COMPARE] = "compare",
3792 : [SPDK_BDEV_IO_TYPE_COMPARE_AND_WRITE] = "compare_and_write",
3793 : [SPDK_BDEV_IO_TYPE_ABORT] = "abort",
3794 : [SPDK_BDEV_IO_TYPE_SEEK_HOLE] = "seek_hole",
3795 : [SPDK_BDEV_IO_TYPE_SEEK_DATA] = "seek_data",
3796 : [SPDK_BDEV_IO_TYPE_COPY] = "copy",
3797 : [SPDK_BDEV_IO_TYPE_NVME_IOV_MD] = "nvme_iov_md",
3798 : };
3799 :
3800 : const char *
3801 0 : spdk_bdev_get_io_type_name(enum spdk_bdev_io_type io_type)
3802 : {
3803 0 : if (io_type <= SPDK_BDEV_IO_TYPE_INVALID || io_type >= SPDK_BDEV_NUM_IO_TYPES) {
3804 0 : return NULL;
3805 : }
3806 :
3807 0 : return g_io_type_strings[io_type];
3808 0 : }
3809 :
3810 : int
3811 0 : spdk_bdev_get_io_type(const char *io_type_string)
3812 : {
3813 : int i;
3814 :
3815 0 : for (i = SPDK_BDEV_IO_TYPE_READ; i < SPDK_BDEV_NUM_IO_TYPES; ++i) {
3816 0 : if (!strcmp(io_type_string, g_io_type_strings[i])) {
3817 0 : return i;
3818 : }
3819 0 : }
3820 :
3821 0 : return -1;
3822 0 : }
3823 :
3824 : uint64_t
3825 0 : spdk_bdev_io_get_submit_tsc(struct spdk_bdev_io *bdev_io)
3826 : {
3827 0 : return bdev_io->internal.submit_tsc;
3828 : }
3829 :
3830 : int
3831 0 : spdk_bdev_dump_info_json(struct spdk_bdev *bdev, struct spdk_json_write_ctx *w)
3832 : {
3833 0 : if (bdev->fn_table->dump_info_json) {
3834 0 : return bdev->fn_table->dump_info_json(bdev->ctxt, w);
3835 : }
3836 :
3837 0 : return 0;
3838 0 : }
3839 :
3840 : static void
3841 10 : bdev_qos_update_max_quota_per_timeslice(struct spdk_bdev_qos *qos)
3842 : {
3843 10 : uint32_t max_per_timeslice = 0;
3844 : int i;
3845 :
3846 50 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
3847 40 : if (qos->rate_limits[i].limit == SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
3848 15 : qos->rate_limits[i].max_per_timeslice = 0;
3849 15 : continue;
3850 : }
3851 :
3852 50 : max_per_timeslice = qos->rate_limits[i].limit *
3853 25 : SPDK_BDEV_QOS_TIMESLICE_IN_USEC / SPDK_SEC_TO_USEC;
3854 :
3855 25 : qos->rate_limits[i].max_per_timeslice = spdk_max(max_per_timeslice,
3856 : qos->rate_limits[i].min_per_timeslice);
3857 :
3858 50 : __atomic_store_n(&qos->rate_limits[i].remaining_this_timeslice,
3859 25 : qos->rate_limits[i].max_per_timeslice, __ATOMIC_RELEASE);
3860 25 : }
3861 :
3862 10 : bdev_qos_set_ops(qos);
3863 10 : }
3864 :
3865 : static void
3866 4 : bdev_channel_submit_qos_io(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
3867 : struct spdk_io_channel *io_ch, void *ctx)
3868 : {
3869 4 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
3870 : int status;
3871 :
3872 4 : bdev_qos_io_submit(bdev_ch, bdev->internal.qos);
3873 :
3874 : /* if all IOs were sent then continue the iteration, otherwise - stop it */
3875 : /* TODO: channels round robing */
3876 4 : status = TAILQ_EMPTY(&bdev_ch->qos_queued_io) ? 0 : 1;
3877 :
3878 4 : spdk_bdev_for_each_channel_continue(i, status);
3879 4 : }
3880 :
3881 :
3882 : static void
3883 2 : bdev_channel_submit_qos_io_done(struct spdk_bdev *bdev, void *ctx, int status)
3884 : {
3885 :
3886 2 : }
3887 :
3888 : static int
3889 3 : bdev_channel_poll_qos(void *arg)
3890 : {
3891 3 : struct spdk_bdev *bdev = arg;
3892 3 : struct spdk_bdev_qos *qos = bdev->internal.qos;
3893 3 : uint64_t now = spdk_get_ticks();
3894 : int i;
3895 : int64_t remaining_last_timeslice;
3896 :
3897 3 : if (spdk_unlikely(qos->thread == NULL)) {
3898 : /* Old QoS was unbound to remove and new QoS is not enabled yet. */
3899 1 : return SPDK_POLLER_IDLE;
3900 : }
3901 :
3902 2 : if (now < (qos->last_timeslice + qos->timeslice_size)) {
3903 : /* We received our callback earlier than expected - return
3904 : * immediately and wait to do accounting until at least one
3905 : * timeslice has actually expired. This should never happen
3906 : * with a well-behaved timer implementation.
3907 : */
3908 0 : return SPDK_POLLER_IDLE;
3909 : }
3910 :
3911 : /* Reset for next round of rate limiting */
3912 10 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
3913 : /* We may have allowed the IOs or bytes to slightly overrun in the last
3914 : * timeslice. remaining_this_timeslice is signed, so if it's negative
3915 : * here, we'll account for the overrun so that the next timeslice will
3916 : * be appropriately reduced.
3917 : */
3918 8 : remaining_last_timeslice = __atomic_exchange_n(&qos->rate_limits[i].remaining_this_timeslice,
3919 : 0, __ATOMIC_RELAXED);
3920 8 : if (remaining_last_timeslice < 0) {
3921 : /* There could be a race condition here as both bdev_qos_rw_queue_io() and bdev_channel_poll_qos()
3922 : * potentially use 2 atomic ops each, so they can intertwine.
3923 : * This race can potentially cause the limits to be a little fuzzy but won't cause any real damage.
3924 : */
3925 0 : __atomic_store_n(&qos->rate_limits[i].remaining_this_timeslice,
3926 0 : remaining_last_timeslice, __ATOMIC_RELAXED);
3927 0 : }
3928 8 : }
3929 :
3930 4 : while (now >= (qos->last_timeslice + qos->timeslice_size)) {
3931 2 : qos->last_timeslice += qos->timeslice_size;
3932 10 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
3933 16 : __atomic_add_fetch(&qos->rate_limits[i].remaining_this_timeslice,
3934 8 : qos->rate_limits[i].max_per_timeslice, __ATOMIC_RELAXED);
3935 8 : }
3936 : }
3937 :
3938 2 : spdk_bdev_for_each_channel(bdev, bdev_channel_submit_qos_io, qos,
3939 : bdev_channel_submit_qos_io_done);
3940 :
3941 2 : return SPDK_POLLER_BUSY;
3942 3 : }
3943 :
3944 : static void
3945 74 : bdev_channel_destroy_resource(struct spdk_bdev_channel *ch)
3946 : {
3947 : struct spdk_bdev_shared_resource *shared_resource;
3948 : struct lba_range *range;
3949 :
3950 74 : bdev_free_io_stat(ch->stat);
3951 : #ifdef SPDK_CONFIG_VTUNE
3952 : bdev_free_io_stat(ch->prev_stat);
3953 : #endif
3954 :
3955 74 : while (!TAILQ_EMPTY(&ch->locked_ranges)) {
3956 0 : range = TAILQ_FIRST(&ch->locked_ranges);
3957 0 : TAILQ_REMOVE(&ch->locked_ranges, range, tailq);
3958 0 : free(range);
3959 : }
3960 :
3961 74 : spdk_put_io_channel(ch->channel);
3962 74 : spdk_put_io_channel(ch->accel_channel);
3963 :
3964 74 : shared_resource = ch->shared_resource;
3965 :
3966 74 : assert(TAILQ_EMPTY(&ch->io_locked));
3967 74 : assert(TAILQ_EMPTY(&ch->io_submitted));
3968 74 : assert(TAILQ_EMPTY(&ch->io_accel_exec));
3969 74 : assert(TAILQ_EMPTY(&ch->io_memory_domain));
3970 74 : assert(ch->io_outstanding == 0);
3971 74 : assert(shared_resource->ref > 0);
3972 74 : shared_resource->ref--;
3973 74 : if (shared_resource->ref == 0) {
3974 73 : assert(shared_resource->io_outstanding == 0);
3975 73 : TAILQ_REMOVE(&shared_resource->mgmt_ch->shared_resources, shared_resource, link);
3976 73 : spdk_put_io_channel(spdk_io_channel_from_ctx(shared_resource->mgmt_ch));
3977 73 : spdk_poller_unregister(&shared_resource->nomem_poller);
3978 73 : free(shared_resource);
3979 73 : }
3980 74 : }
3981 :
3982 : static void
3983 83 : bdev_enable_qos(struct spdk_bdev *bdev, struct spdk_bdev_channel *ch)
3984 : {
3985 83 : struct spdk_bdev_qos *qos = bdev->internal.qos;
3986 : int i;
3987 :
3988 83 : assert(spdk_spin_held(&bdev->internal.spinlock));
3989 :
3990 : /* Rate limiting on this bdev enabled */
3991 83 : if (qos) {
3992 17 : if (qos->ch == NULL) {
3993 : struct spdk_io_channel *io_ch;
3994 :
3995 9 : SPDK_DEBUGLOG(bdev, "Selecting channel %p as QoS channel for bdev %s on thread %p\n", ch,
3996 : bdev->name, spdk_get_thread());
3997 :
3998 : /* No qos channel has been selected, so set one up */
3999 :
4000 : /* Take another reference to ch */
4001 9 : io_ch = spdk_get_io_channel(__bdev_to_io_dev(bdev));
4002 9 : assert(io_ch != NULL);
4003 9 : qos->ch = ch;
4004 :
4005 9 : qos->thread = spdk_io_channel_get_thread(io_ch);
4006 :
4007 45 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
4008 36 : if (bdev_qos_is_iops_rate_limit(i) == true) {
4009 9 : qos->rate_limits[i].min_per_timeslice =
4010 : SPDK_BDEV_QOS_MIN_IO_PER_TIMESLICE;
4011 9 : } else {
4012 27 : qos->rate_limits[i].min_per_timeslice =
4013 : SPDK_BDEV_QOS_MIN_BYTE_PER_TIMESLICE;
4014 : }
4015 :
4016 36 : if (qos->rate_limits[i].limit == 0) {
4017 2 : qos->rate_limits[i].limit = SPDK_BDEV_QOS_LIMIT_NOT_DEFINED;
4018 2 : }
4019 36 : }
4020 9 : bdev_qos_update_max_quota_per_timeslice(qos);
4021 9 : qos->timeslice_size =
4022 9 : SPDK_BDEV_QOS_TIMESLICE_IN_USEC * spdk_get_ticks_hz() / SPDK_SEC_TO_USEC;
4023 9 : qos->last_timeslice = spdk_get_ticks();
4024 9 : qos->poller = SPDK_POLLER_REGISTER(bdev_channel_poll_qos,
4025 : bdev,
4026 : SPDK_BDEV_QOS_TIMESLICE_IN_USEC);
4027 9 : }
4028 :
4029 17 : ch->flags |= BDEV_CH_QOS_ENABLED;
4030 17 : }
4031 83 : }
4032 :
4033 : struct poll_timeout_ctx {
4034 : struct spdk_bdev_desc *desc;
4035 : uint64_t timeout_in_sec;
4036 : spdk_bdev_io_timeout_cb cb_fn;
4037 : void *cb_arg;
4038 : };
4039 :
4040 : static void
4041 264 : bdev_desc_free(struct spdk_bdev_desc *desc)
4042 : {
4043 264 : spdk_spin_destroy(&desc->spinlock);
4044 264 : free(desc->media_events_buffer);
4045 264 : free(desc);
4046 264 : }
4047 :
4048 : static void
4049 8 : bdev_channel_poll_timeout_io_done(struct spdk_bdev *bdev, void *_ctx, int status)
4050 : {
4051 8 : struct poll_timeout_ctx *ctx = _ctx;
4052 8 : struct spdk_bdev_desc *desc = ctx->desc;
4053 :
4054 8 : free(ctx);
4055 :
4056 8 : spdk_spin_lock(&desc->spinlock);
4057 8 : desc->refs--;
4058 8 : if (desc->closed == true && desc->refs == 0) {
4059 1 : spdk_spin_unlock(&desc->spinlock);
4060 1 : bdev_desc_free(desc);
4061 1 : return;
4062 : }
4063 7 : spdk_spin_unlock(&desc->spinlock);
4064 8 : }
4065 :
4066 : static void
4067 13 : bdev_channel_poll_timeout_io(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
4068 : struct spdk_io_channel *io_ch, void *_ctx)
4069 : {
4070 13 : struct poll_timeout_ctx *ctx = _ctx;
4071 13 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
4072 13 : struct spdk_bdev_desc *desc = ctx->desc;
4073 : struct spdk_bdev_io *bdev_io;
4074 : uint64_t now;
4075 :
4076 13 : spdk_spin_lock(&desc->spinlock);
4077 13 : if (desc->closed == true) {
4078 1 : spdk_spin_unlock(&desc->spinlock);
4079 1 : spdk_bdev_for_each_channel_continue(i, -1);
4080 1 : return;
4081 : }
4082 12 : spdk_spin_unlock(&desc->spinlock);
4083 :
4084 12 : now = spdk_get_ticks();
4085 22 : TAILQ_FOREACH(bdev_io, &bdev_ch->io_submitted, internal.ch_link) {
4086 : /* Exclude any I/O that are generated via splitting. */
4087 15 : if (bdev_io->internal.cb == bdev_io_split_done) {
4088 3 : continue;
4089 : }
4090 :
4091 : /* Once we find an I/O that has not timed out, we can immediately
4092 : * exit the loop.
4093 : */
4094 24 : if (now < (bdev_io->internal.submit_tsc +
4095 12 : ctx->timeout_in_sec * spdk_get_ticks_hz())) {
4096 5 : goto end;
4097 : }
4098 :
4099 7 : if (bdev_io->internal.desc == desc) {
4100 7 : ctx->cb_fn(ctx->cb_arg, bdev_io);
4101 7 : }
4102 14 : }
4103 :
4104 : end:
4105 12 : spdk_bdev_for_each_channel_continue(i, 0);
4106 13 : }
4107 :
4108 : static int
4109 8 : bdev_poll_timeout_io(void *arg)
4110 : {
4111 8 : struct spdk_bdev_desc *desc = arg;
4112 8 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
4113 : struct poll_timeout_ctx *ctx;
4114 :
4115 8 : ctx = calloc(1, sizeof(struct poll_timeout_ctx));
4116 8 : if (!ctx) {
4117 0 : SPDK_ERRLOG("failed to allocate memory\n");
4118 0 : return SPDK_POLLER_BUSY;
4119 : }
4120 8 : ctx->desc = desc;
4121 8 : ctx->cb_arg = desc->cb_arg;
4122 8 : ctx->cb_fn = desc->cb_fn;
4123 8 : ctx->timeout_in_sec = desc->timeout_in_sec;
4124 :
4125 : /* Take a ref on the descriptor in case it gets closed while we are checking
4126 : * all of the channels.
4127 : */
4128 8 : spdk_spin_lock(&desc->spinlock);
4129 8 : desc->refs++;
4130 8 : spdk_spin_unlock(&desc->spinlock);
4131 :
4132 8 : spdk_bdev_for_each_channel(bdev, bdev_channel_poll_timeout_io, ctx,
4133 : bdev_channel_poll_timeout_io_done);
4134 :
4135 8 : return SPDK_POLLER_BUSY;
4136 8 : }
4137 :
4138 : int
4139 5 : spdk_bdev_set_timeout(struct spdk_bdev_desc *desc, uint64_t timeout_in_sec,
4140 : spdk_bdev_io_timeout_cb cb_fn, void *cb_arg)
4141 : {
4142 5 : assert(desc->thread == spdk_get_thread());
4143 :
4144 5 : spdk_poller_unregister(&desc->io_timeout_poller);
4145 :
4146 5 : if (timeout_in_sec) {
4147 4 : assert(cb_fn != NULL);
4148 4 : desc->io_timeout_poller = SPDK_POLLER_REGISTER(bdev_poll_timeout_io,
4149 : desc,
4150 : SPDK_BDEV_IO_POLL_INTERVAL_IN_MSEC * SPDK_SEC_TO_USEC /
4151 : 1000);
4152 4 : if (desc->io_timeout_poller == NULL) {
4153 0 : SPDK_ERRLOG("can not register the desc timeout IO poller\n");
4154 0 : return -1;
4155 : }
4156 4 : }
4157 :
4158 5 : desc->cb_fn = cb_fn;
4159 5 : desc->cb_arg = cb_arg;
4160 5 : desc->timeout_in_sec = timeout_in_sec;
4161 :
4162 5 : return 0;
4163 5 : }
4164 :
4165 : static int
4166 76 : bdev_channel_create(void *io_device, void *ctx_buf)
4167 : {
4168 76 : struct spdk_bdev *bdev = __bdev_from_io_dev(io_device);
4169 76 : struct spdk_bdev_channel *ch = ctx_buf;
4170 : struct spdk_io_channel *mgmt_io_ch;
4171 : struct spdk_bdev_mgmt_channel *mgmt_ch;
4172 : struct spdk_bdev_shared_resource *shared_resource;
4173 : struct lba_range *range;
4174 :
4175 76 : ch->bdev = bdev;
4176 76 : ch->channel = bdev->fn_table->get_io_channel(bdev->ctxt);
4177 76 : if (!ch->channel) {
4178 2 : return -1;
4179 : }
4180 :
4181 74 : ch->accel_channel = spdk_accel_get_io_channel();
4182 74 : if (!ch->accel_channel) {
4183 0 : spdk_put_io_channel(ch->channel);
4184 0 : return -1;
4185 : }
4186 :
4187 74 : spdk_trace_record(TRACE_BDEV_IOCH_CREATE, bdev->internal.trace_id, 0, 0,
4188 : spdk_thread_get_id(spdk_io_channel_get_thread(ch->channel)));
4189 :
4190 74 : assert(ch->histogram == NULL);
4191 74 : if (bdev->internal.histogram_enabled) {
4192 0 : ch->histogram = spdk_histogram_data_alloc();
4193 0 : if (ch->histogram == NULL) {
4194 0 : SPDK_ERRLOG("Could not allocate histogram\n");
4195 0 : }
4196 0 : }
4197 :
4198 74 : mgmt_io_ch = spdk_get_io_channel(&g_bdev_mgr);
4199 74 : if (!mgmt_io_ch) {
4200 0 : spdk_put_io_channel(ch->channel);
4201 0 : spdk_put_io_channel(ch->accel_channel);
4202 0 : return -1;
4203 : }
4204 :
4205 74 : mgmt_ch = __io_ch_to_bdev_mgmt_ch(mgmt_io_ch);
4206 76 : TAILQ_FOREACH(shared_resource, &mgmt_ch->shared_resources, link) {
4207 3 : if (shared_resource->shared_ch == ch->channel) {
4208 1 : spdk_put_io_channel(mgmt_io_ch);
4209 1 : shared_resource->ref++;
4210 1 : break;
4211 : }
4212 2 : }
4213 :
4214 74 : if (shared_resource == NULL) {
4215 73 : shared_resource = calloc(1, sizeof(*shared_resource));
4216 73 : if (shared_resource == NULL) {
4217 0 : spdk_put_io_channel(ch->channel);
4218 0 : spdk_put_io_channel(ch->accel_channel);
4219 0 : spdk_put_io_channel(mgmt_io_ch);
4220 0 : return -1;
4221 : }
4222 :
4223 73 : shared_resource->mgmt_ch = mgmt_ch;
4224 73 : shared_resource->io_outstanding = 0;
4225 73 : TAILQ_INIT(&shared_resource->nomem_io);
4226 73 : shared_resource->nomem_threshold = 0;
4227 73 : shared_resource->shared_ch = ch->channel;
4228 73 : shared_resource->ref = 1;
4229 73 : TAILQ_INSERT_TAIL(&mgmt_ch->shared_resources, shared_resource, link);
4230 73 : }
4231 :
4232 74 : ch->io_outstanding = 0;
4233 74 : TAILQ_INIT(&ch->queued_resets);
4234 74 : TAILQ_INIT(&ch->locked_ranges);
4235 74 : TAILQ_INIT(&ch->qos_queued_io);
4236 74 : ch->flags = 0;
4237 74 : ch->trace_id = bdev->internal.trace_id;
4238 74 : ch->shared_resource = shared_resource;
4239 :
4240 74 : TAILQ_INIT(&ch->io_submitted);
4241 74 : TAILQ_INIT(&ch->io_locked);
4242 74 : TAILQ_INIT(&ch->io_accel_exec);
4243 74 : TAILQ_INIT(&ch->io_memory_domain);
4244 :
4245 74 : ch->stat = bdev_alloc_io_stat(false);
4246 74 : if (ch->stat == NULL) {
4247 0 : bdev_channel_destroy_resource(ch);
4248 0 : return -1;
4249 : }
4250 :
4251 74 : ch->stat->ticks_rate = spdk_get_ticks_hz();
4252 :
4253 : #ifdef SPDK_CONFIG_VTUNE
4254 : {
4255 : char *name;
4256 : __itt_init_ittlib(NULL, 0);
4257 : name = spdk_sprintf_alloc("spdk_bdev_%s_%p", ch->bdev->name, ch);
4258 : if (!name) {
4259 : bdev_channel_destroy_resource(ch);
4260 : return -1;
4261 : }
4262 : ch->handle = __itt_string_handle_create(name);
4263 : free(name);
4264 : ch->start_tsc = spdk_get_ticks();
4265 : ch->interval_tsc = spdk_get_ticks_hz() / 100;
4266 : ch->prev_stat = bdev_alloc_io_stat(false);
4267 : if (ch->prev_stat == NULL) {
4268 : bdev_channel_destroy_resource(ch);
4269 : return -1;
4270 : }
4271 : }
4272 : #endif
4273 :
4274 74 : spdk_spin_lock(&bdev->internal.spinlock);
4275 74 : bdev_enable_qos(bdev, ch);
4276 :
4277 75 : TAILQ_FOREACH(range, &bdev->internal.locked_ranges, tailq) {
4278 : struct lba_range *new_range;
4279 :
4280 1 : new_range = calloc(1, sizeof(*new_range));
4281 1 : if (new_range == NULL) {
4282 0 : spdk_spin_unlock(&bdev->internal.spinlock);
4283 0 : bdev_channel_destroy_resource(ch);
4284 0 : return -1;
4285 : }
4286 1 : new_range->length = range->length;
4287 1 : new_range->offset = range->offset;
4288 1 : new_range->locked_ctx = range->locked_ctx;
4289 1 : TAILQ_INSERT_TAIL(&ch->locked_ranges, new_range, tailq);
4290 1 : }
4291 :
4292 74 : spdk_spin_unlock(&bdev->internal.spinlock);
4293 :
4294 74 : return 0;
4295 76 : }
4296 :
4297 : static int
4298 0 : bdev_abort_all_buf_io_cb(struct spdk_iobuf_channel *ch, struct spdk_iobuf_entry *entry,
4299 : void *cb_ctx)
4300 : {
4301 0 : struct spdk_bdev_channel *bdev_ch = cb_ctx;
4302 : struct spdk_bdev_io *bdev_io;
4303 : uint64_t buf_len;
4304 :
4305 0 : bdev_io = SPDK_CONTAINEROF(entry, struct spdk_bdev_io, internal.iobuf);
4306 0 : if (bdev_io->internal.ch == bdev_ch) {
4307 0 : buf_len = bdev_io_get_max_buf_len(bdev_io, bdev_io->internal.buf.len);
4308 0 : spdk_iobuf_entry_abort(ch, entry, buf_len);
4309 0 : spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_ABORTED);
4310 0 : }
4311 :
4312 0 : return 0;
4313 : }
4314 :
4315 : /*
4316 : * Abort I/O that are waiting on a data buffer.
4317 : */
4318 : static void
4319 96 : bdev_abort_all_buf_io(struct spdk_bdev_mgmt_channel *mgmt_ch, struct spdk_bdev_channel *ch)
4320 : {
4321 96 : spdk_iobuf_for_each_entry(&mgmt_ch->iobuf, bdev_abort_all_buf_io_cb, ch);
4322 96 : }
4323 :
4324 : /*
4325 : * Abort I/O that are queued waiting for submission. These types of I/O are
4326 : * linked using the spdk_bdev_io link TAILQ_ENTRY.
4327 : */
4328 : static void
4329 188 : bdev_abort_all_queued_io(bdev_io_tailq_t *queue, struct spdk_bdev_channel *ch)
4330 : {
4331 : struct spdk_bdev_io *bdev_io, *tmp;
4332 :
4333 228 : TAILQ_FOREACH_SAFE(bdev_io, queue, internal.link, tmp) {
4334 40 : if (bdev_io->internal.ch == ch) {
4335 40 : TAILQ_REMOVE(queue, bdev_io, internal.link);
4336 : /*
4337 : * spdk_bdev_io_complete() assumes that the completed I/O had
4338 : * been submitted to the bdev module. Since in this case it
4339 : * hadn't, bump io_outstanding to account for the decrement
4340 : * that spdk_bdev_io_complete() will do.
4341 : */
4342 40 : if (bdev_io->type != SPDK_BDEV_IO_TYPE_RESET) {
4343 39 : bdev_io_increment_outstanding(ch, ch->shared_resource);
4344 39 : }
4345 40 : spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_ABORTED);
4346 40 : }
4347 40 : }
4348 188 : }
4349 :
4350 : static bool
4351 18 : bdev_abort_queued_io(bdev_io_tailq_t *queue, struct spdk_bdev_io *bio_to_abort)
4352 : {
4353 : struct spdk_bdev_io *bdev_io;
4354 :
4355 18 : TAILQ_FOREACH(bdev_io, queue, internal.link) {
4356 0 : if (bdev_io == bio_to_abort) {
4357 0 : TAILQ_REMOVE(queue, bio_to_abort, internal.link);
4358 0 : spdk_bdev_io_complete(bio_to_abort, SPDK_BDEV_IO_STATUS_ABORTED);
4359 0 : return true;
4360 : }
4361 0 : }
4362 :
4363 18 : return false;
4364 18 : }
4365 :
4366 : static int
4367 0 : bdev_abort_buf_io_cb(struct spdk_iobuf_channel *ch, struct spdk_iobuf_entry *entry, void *cb_ctx)
4368 : {
4369 0 : struct spdk_bdev_io *bdev_io, *bio_to_abort = cb_ctx;
4370 : uint64_t buf_len;
4371 :
4372 0 : bdev_io = SPDK_CONTAINEROF(entry, struct spdk_bdev_io, internal.iobuf);
4373 0 : if (bdev_io == bio_to_abort) {
4374 0 : buf_len = bdev_io_get_max_buf_len(bdev_io, bdev_io->internal.buf.len);
4375 0 : spdk_iobuf_entry_abort(ch, entry, buf_len);
4376 0 : spdk_bdev_io_complete(bio_to_abort, SPDK_BDEV_IO_STATUS_ABORTED);
4377 0 : return 1;
4378 : }
4379 :
4380 0 : return 0;
4381 0 : }
4382 :
4383 : static bool
4384 16 : bdev_abort_buf_io(struct spdk_bdev_mgmt_channel *mgmt_ch, struct spdk_bdev_io *bio_to_abort)
4385 : {
4386 : int rc;
4387 :
4388 16 : rc = spdk_iobuf_for_each_entry(&mgmt_ch->iobuf, bdev_abort_buf_io_cb, bio_to_abort);
4389 16 : return rc == 1;
4390 : }
4391 :
4392 : static void
4393 7 : bdev_qos_channel_destroy(void *cb_arg)
4394 : {
4395 7 : struct spdk_bdev_qos *qos = cb_arg;
4396 :
4397 7 : spdk_put_io_channel(spdk_io_channel_from_ctx(qos->ch));
4398 7 : spdk_poller_unregister(&qos->poller);
4399 :
4400 7 : SPDK_DEBUGLOG(bdev, "Free QoS %p.\n", qos);
4401 :
4402 7 : free(qos);
4403 7 : }
4404 :
4405 : static int
4406 7 : bdev_qos_destroy(struct spdk_bdev *bdev)
4407 : {
4408 : int i;
4409 :
4410 : /*
4411 : * Cleanly shutting down the QoS poller is tricky, because
4412 : * during the asynchronous operation the user could open
4413 : * a new descriptor and create a new channel, spawning
4414 : * a new QoS poller.
4415 : *
4416 : * The strategy is to create a new QoS structure here and swap it
4417 : * in. The shutdown path then continues to refer to the old one
4418 : * until it completes and then releases it.
4419 : */
4420 : struct spdk_bdev_qos *new_qos, *old_qos;
4421 :
4422 7 : old_qos = bdev->internal.qos;
4423 :
4424 7 : new_qos = calloc(1, sizeof(*new_qos));
4425 7 : if (!new_qos) {
4426 0 : SPDK_ERRLOG("Unable to allocate memory to shut down QoS.\n");
4427 0 : return -ENOMEM;
4428 : }
4429 :
4430 : /* Copy the old QoS data into the newly allocated structure */
4431 7 : memcpy(new_qos, old_qos, sizeof(*new_qos));
4432 :
4433 : /* Zero out the key parts of the QoS structure */
4434 7 : new_qos->ch = NULL;
4435 7 : new_qos->thread = NULL;
4436 7 : new_qos->poller = NULL;
4437 : /*
4438 : * The limit member of spdk_bdev_qos_limit structure is not zeroed.
4439 : * It will be used later for the new QoS structure.
4440 : */
4441 35 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
4442 28 : new_qos->rate_limits[i].remaining_this_timeslice = 0;
4443 28 : new_qos->rate_limits[i].min_per_timeslice = 0;
4444 28 : new_qos->rate_limits[i].max_per_timeslice = 0;
4445 28 : }
4446 :
4447 7 : bdev->internal.qos = new_qos;
4448 :
4449 7 : if (old_qos->thread == NULL) {
4450 0 : free(old_qos);
4451 0 : } else {
4452 7 : spdk_thread_send_msg(old_qos->thread, bdev_qos_channel_destroy, old_qos);
4453 : }
4454 :
4455 : /* It is safe to continue with destroying the bdev even though the QoS channel hasn't
4456 : * been destroyed yet. The destruction path will end up waiting for the final
4457 : * channel to be put before it releases resources. */
4458 :
4459 7 : return 0;
4460 7 : }
4461 :
4462 : void
4463 78 : spdk_bdev_add_io_stat(struct spdk_bdev_io_stat *total, struct spdk_bdev_io_stat *add)
4464 : {
4465 78 : total->bytes_read += add->bytes_read;
4466 78 : total->num_read_ops += add->num_read_ops;
4467 78 : total->bytes_written += add->bytes_written;
4468 78 : total->num_write_ops += add->num_write_ops;
4469 78 : total->bytes_unmapped += add->bytes_unmapped;
4470 78 : total->num_unmap_ops += add->num_unmap_ops;
4471 78 : total->bytes_copied += add->bytes_copied;
4472 78 : total->num_copy_ops += add->num_copy_ops;
4473 78 : total->read_latency_ticks += add->read_latency_ticks;
4474 78 : total->write_latency_ticks += add->write_latency_ticks;
4475 78 : total->unmap_latency_ticks += add->unmap_latency_ticks;
4476 78 : total->copy_latency_ticks += add->copy_latency_ticks;
4477 78 : if (total->max_read_latency_ticks < add->max_read_latency_ticks) {
4478 7 : total->max_read_latency_ticks = add->max_read_latency_ticks;
4479 7 : }
4480 78 : if (total->min_read_latency_ticks > add->min_read_latency_ticks) {
4481 39 : total->min_read_latency_ticks = add->min_read_latency_ticks;
4482 39 : }
4483 78 : if (total->max_write_latency_ticks < add->max_write_latency_ticks) {
4484 4 : total->max_write_latency_ticks = add->max_write_latency_ticks;
4485 4 : }
4486 78 : if (total->min_write_latency_ticks > add->min_write_latency_ticks) {
4487 24 : total->min_write_latency_ticks = add->min_write_latency_ticks;
4488 24 : }
4489 78 : if (total->max_unmap_latency_ticks < add->max_unmap_latency_ticks) {
4490 0 : total->max_unmap_latency_ticks = add->max_unmap_latency_ticks;
4491 0 : }
4492 78 : if (total->min_unmap_latency_ticks > add->min_unmap_latency_ticks) {
4493 3 : total->min_unmap_latency_ticks = add->min_unmap_latency_ticks;
4494 3 : }
4495 78 : if (total->max_copy_latency_ticks < add->max_copy_latency_ticks) {
4496 0 : total->max_copy_latency_ticks = add->max_copy_latency_ticks;
4497 0 : }
4498 78 : if (total->min_copy_latency_ticks > add->min_copy_latency_ticks) {
4499 4 : total->min_copy_latency_ticks = add->min_copy_latency_ticks;
4500 4 : }
4501 78 : }
4502 :
4503 : static void
4504 5 : bdev_get_io_stat(struct spdk_bdev_io_stat *to_stat, struct spdk_bdev_io_stat *from_stat)
4505 : {
4506 5 : memcpy(to_stat, from_stat, offsetof(struct spdk_bdev_io_stat, io_error));
4507 :
4508 5 : if (to_stat->io_error != NULL && from_stat->io_error != NULL) {
4509 0 : memcpy(to_stat->io_error, from_stat->io_error,
4510 : sizeof(struct spdk_bdev_io_error_stat));
4511 0 : }
4512 5 : }
4513 :
4514 : void
4515 209 : spdk_bdev_reset_io_stat(struct spdk_bdev_io_stat *stat, enum spdk_bdev_reset_stat_mode mode)
4516 : {
4517 209 : if (mode == SPDK_BDEV_RESET_STAT_NONE) {
4518 5 : return;
4519 : }
4520 :
4521 204 : stat->max_read_latency_ticks = 0;
4522 204 : stat->min_read_latency_ticks = UINT64_MAX;
4523 204 : stat->max_write_latency_ticks = 0;
4524 204 : stat->min_write_latency_ticks = UINT64_MAX;
4525 204 : stat->max_unmap_latency_ticks = 0;
4526 204 : stat->min_unmap_latency_ticks = UINT64_MAX;
4527 204 : stat->max_copy_latency_ticks = 0;
4528 204 : stat->min_copy_latency_ticks = UINT64_MAX;
4529 :
4530 204 : if (mode != SPDK_BDEV_RESET_STAT_ALL) {
4531 2 : return;
4532 : }
4533 :
4534 202 : stat->bytes_read = 0;
4535 202 : stat->num_read_ops = 0;
4536 202 : stat->bytes_written = 0;
4537 202 : stat->num_write_ops = 0;
4538 202 : stat->bytes_unmapped = 0;
4539 202 : stat->num_unmap_ops = 0;
4540 202 : stat->bytes_copied = 0;
4541 202 : stat->num_copy_ops = 0;
4542 202 : stat->read_latency_ticks = 0;
4543 202 : stat->write_latency_ticks = 0;
4544 202 : stat->unmap_latency_ticks = 0;
4545 202 : stat->copy_latency_ticks = 0;
4546 :
4547 202 : if (stat->io_error != NULL) {
4548 127 : memset(stat->io_error, 0, sizeof(struct spdk_bdev_io_error_stat));
4549 127 : }
4550 209 : }
4551 :
4552 : struct spdk_bdev_io_stat *
4553 200 : bdev_alloc_io_stat(bool io_error_stat)
4554 : {
4555 : struct spdk_bdev_io_stat *stat;
4556 :
4557 200 : stat = malloc(sizeof(struct spdk_bdev_io_stat));
4558 200 : if (stat == NULL) {
4559 0 : return NULL;
4560 : }
4561 :
4562 200 : if (io_error_stat) {
4563 126 : stat->io_error = malloc(sizeof(struct spdk_bdev_io_error_stat));
4564 126 : if (stat->io_error == NULL) {
4565 0 : free(stat);
4566 0 : return NULL;
4567 : }
4568 126 : } else {
4569 74 : stat->io_error = NULL;
4570 : }
4571 :
4572 200 : spdk_bdev_reset_io_stat(stat, SPDK_BDEV_RESET_STAT_ALL);
4573 :
4574 200 : return stat;
4575 200 : }
4576 :
4577 : void
4578 200 : bdev_free_io_stat(struct spdk_bdev_io_stat *stat)
4579 : {
4580 200 : if (stat != NULL) {
4581 200 : free(stat->io_error);
4582 200 : free(stat);
4583 200 : }
4584 200 : }
4585 :
4586 : void
4587 0 : spdk_bdev_dump_io_stat_json(struct spdk_bdev_io_stat *stat, struct spdk_json_write_ctx *w)
4588 : {
4589 : int i;
4590 :
4591 0 : spdk_json_write_named_uint64(w, "bytes_read", stat->bytes_read);
4592 0 : spdk_json_write_named_uint64(w, "num_read_ops", stat->num_read_ops);
4593 0 : spdk_json_write_named_uint64(w, "bytes_written", stat->bytes_written);
4594 0 : spdk_json_write_named_uint64(w, "num_write_ops", stat->num_write_ops);
4595 0 : spdk_json_write_named_uint64(w, "bytes_unmapped", stat->bytes_unmapped);
4596 0 : spdk_json_write_named_uint64(w, "num_unmap_ops", stat->num_unmap_ops);
4597 0 : spdk_json_write_named_uint64(w, "bytes_copied", stat->bytes_copied);
4598 0 : spdk_json_write_named_uint64(w, "num_copy_ops", stat->num_copy_ops);
4599 0 : spdk_json_write_named_uint64(w, "read_latency_ticks", stat->read_latency_ticks);
4600 0 : spdk_json_write_named_uint64(w, "max_read_latency_ticks", stat->max_read_latency_ticks);
4601 0 : spdk_json_write_named_uint64(w, "min_read_latency_ticks",
4602 0 : stat->min_read_latency_ticks != UINT64_MAX ?
4603 0 : stat->min_read_latency_ticks : 0);
4604 0 : spdk_json_write_named_uint64(w, "write_latency_ticks", stat->write_latency_ticks);
4605 0 : spdk_json_write_named_uint64(w, "max_write_latency_ticks", stat->max_write_latency_ticks);
4606 0 : spdk_json_write_named_uint64(w, "min_write_latency_ticks",
4607 0 : stat->min_write_latency_ticks != UINT64_MAX ?
4608 0 : stat->min_write_latency_ticks : 0);
4609 0 : spdk_json_write_named_uint64(w, "unmap_latency_ticks", stat->unmap_latency_ticks);
4610 0 : spdk_json_write_named_uint64(w, "max_unmap_latency_ticks", stat->max_unmap_latency_ticks);
4611 0 : spdk_json_write_named_uint64(w, "min_unmap_latency_ticks",
4612 0 : stat->min_unmap_latency_ticks != UINT64_MAX ?
4613 0 : stat->min_unmap_latency_ticks : 0);
4614 0 : spdk_json_write_named_uint64(w, "copy_latency_ticks", stat->copy_latency_ticks);
4615 0 : spdk_json_write_named_uint64(w, "max_copy_latency_ticks", stat->max_copy_latency_ticks);
4616 0 : spdk_json_write_named_uint64(w, "min_copy_latency_ticks",
4617 0 : stat->min_copy_latency_ticks != UINT64_MAX ?
4618 0 : stat->min_copy_latency_ticks : 0);
4619 :
4620 0 : if (stat->io_error != NULL) {
4621 0 : spdk_json_write_named_object_begin(w, "io_error");
4622 0 : for (i = 0; i < -SPDK_MIN_BDEV_IO_STATUS; i++) {
4623 0 : if (stat->io_error->error_status[i] != 0) {
4624 0 : spdk_json_write_named_uint32(w, bdev_io_status_get_string(-(i + 1)),
4625 0 : stat->io_error->error_status[i]);
4626 0 : }
4627 0 : }
4628 0 : spdk_json_write_object_end(w);
4629 0 : }
4630 0 : }
4631 :
4632 : static void
4633 78 : bdev_channel_abort_queued_ios(struct spdk_bdev_channel *ch)
4634 : {
4635 78 : struct spdk_bdev_shared_resource *shared_resource = ch->shared_resource;
4636 78 : struct spdk_bdev_mgmt_channel *mgmt_ch = shared_resource->mgmt_ch;
4637 :
4638 78 : bdev_abort_all_queued_io(&shared_resource->nomem_io, ch);
4639 78 : bdev_abort_all_buf_io(mgmt_ch, ch);
4640 78 : }
4641 :
4642 : static void
4643 74 : bdev_channel_destroy(void *io_device, void *ctx_buf)
4644 : {
4645 74 : struct spdk_bdev_channel *ch = ctx_buf;
4646 :
4647 74 : SPDK_DEBUGLOG(bdev, "Destroying channel %p for bdev %s on thread %p\n", ch, ch->bdev->name,
4648 : spdk_get_thread());
4649 :
4650 74 : spdk_trace_record(TRACE_BDEV_IOCH_DESTROY, ch->bdev->internal.trace_id, 0, 0,
4651 : spdk_thread_get_id(spdk_io_channel_get_thread(ch->channel)));
4652 :
4653 : /* This channel is going away, so add its statistics into the bdev so that they don't get lost. */
4654 74 : spdk_spin_lock(&ch->bdev->internal.spinlock);
4655 74 : spdk_bdev_add_io_stat(ch->bdev->internal.stat, ch->stat);
4656 74 : spdk_spin_unlock(&ch->bdev->internal.spinlock);
4657 :
4658 74 : bdev_abort_all_queued_io(&ch->queued_resets, ch);
4659 :
4660 74 : bdev_channel_abort_queued_ios(ch);
4661 :
4662 74 : if (ch->histogram) {
4663 0 : spdk_histogram_data_free(ch->histogram);
4664 0 : }
4665 :
4666 74 : bdev_channel_destroy_resource(ch);
4667 74 : }
4668 :
4669 : /*
4670 : * If the name already exists in the global bdev name tree, RB_INSERT() returns a pointer
4671 : * to it. Hence we do not have to call bdev_get_by_name() when using this function.
4672 : */
4673 : static int
4674 255 : bdev_name_add(struct spdk_bdev_name *bdev_name, struct spdk_bdev *bdev, const char *name)
4675 : {
4676 : struct spdk_bdev_name *tmp;
4677 :
4678 255 : bdev_name->name = strdup(name);
4679 255 : if (bdev_name->name == NULL) {
4680 0 : SPDK_ERRLOG("Unable to allocate bdev name\n");
4681 0 : return -ENOMEM;
4682 : }
4683 :
4684 255 : bdev_name->bdev = bdev;
4685 :
4686 255 : spdk_spin_lock(&g_bdev_mgr.spinlock);
4687 255 : tmp = RB_INSERT(bdev_name_tree, &g_bdev_mgr.bdev_names, bdev_name);
4688 255 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
4689 :
4690 255 : if (tmp != NULL) {
4691 4 : SPDK_ERRLOG("Bdev name %s already exists\n", name);
4692 4 : free(bdev_name->name);
4693 4 : return -EEXIST;
4694 : }
4695 :
4696 251 : return 0;
4697 255 : }
4698 :
4699 : static void
4700 251 : bdev_name_del_unsafe(struct spdk_bdev_name *bdev_name)
4701 : {
4702 251 : RB_REMOVE(bdev_name_tree, &g_bdev_mgr.bdev_names, bdev_name);
4703 251 : free(bdev_name->name);
4704 251 : }
4705 :
4706 : static void
4707 5 : bdev_name_del(struct spdk_bdev_name *bdev_name)
4708 : {
4709 5 : spdk_spin_lock(&g_bdev_mgr.spinlock);
4710 5 : bdev_name_del_unsafe(bdev_name);
4711 5 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
4712 5 : }
4713 :
4714 : int
4715 132 : spdk_bdev_alias_add(struct spdk_bdev *bdev, const char *alias)
4716 : {
4717 : struct spdk_bdev_alias *tmp;
4718 : int ret;
4719 :
4720 132 : if (alias == NULL) {
4721 1 : SPDK_ERRLOG("Empty alias passed\n");
4722 1 : return -EINVAL;
4723 : }
4724 :
4725 131 : tmp = calloc(1, sizeof(*tmp));
4726 131 : if (tmp == NULL) {
4727 0 : SPDK_ERRLOG("Unable to allocate alias\n");
4728 0 : return -ENOMEM;
4729 : }
4730 :
4731 131 : ret = bdev_name_add(&tmp->alias, bdev, alias);
4732 131 : if (ret != 0) {
4733 4 : free(tmp);
4734 4 : return ret;
4735 : }
4736 :
4737 127 : TAILQ_INSERT_TAIL(&bdev->aliases, tmp, tailq);
4738 :
4739 127 : return 0;
4740 132 : }
4741 :
4742 : static int
4743 128 : bdev_alias_del(struct spdk_bdev *bdev, const char *alias,
4744 : void (*alias_del_fn)(struct spdk_bdev_name *n))
4745 : {
4746 : struct spdk_bdev_alias *tmp;
4747 :
4748 133 : TAILQ_FOREACH(tmp, &bdev->aliases, tailq) {
4749 129 : if (strcmp(alias, tmp->alias.name) == 0) {
4750 124 : TAILQ_REMOVE(&bdev->aliases, tmp, tailq);
4751 124 : alias_del_fn(&tmp->alias);
4752 124 : free(tmp);
4753 124 : return 0;
4754 : }
4755 5 : }
4756 :
4757 4 : return -ENOENT;
4758 128 : }
4759 :
4760 : int
4761 4 : spdk_bdev_alias_del(struct spdk_bdev *bdev, const char *alias)
4762 : {
4763 : int rc;
4764 :
4765 4 : rc = bdev_alias_del(bdev, alias, bdev_name_del);
4766 4 : if (rc == -ENOENT) {
4767 2 : SPDK_INFOLOG(bdev, "Alias %s does not exist\n", alias);
4768 2 : }
4769 :
4770 4 : return rc;
4771 : }
4772 :
4773 : void
4774 2 : spdk_bdev_alias_del_all(struct spdk_bdev *bdev)
4775 : {
4776 : struct spdk_bdev_alias *p, *tmp;
4777 :
4778 5 : TAILQ_FOREACH_SAFE(p, &bdev->aliases, tailq, tmp) {
4779 3 : TAILQ_REMOVE(&bdev->aliases, p, tailq);
4780 3 : bdev_name_del(&p->alias);
4781 3 : free(p);
4782 3 : }
4783 2 : }
4784 :
4785 : struct spdk_io_channel *
4786 76 : spdk_bdev_get_io_channel(struct spdk_bdev_desc *desc)
4787 : {
4788 76 : return spdk_get_io_channel(__bdev_to_io_dev(spdk_bdev_desc_get_bdev(desc)));
4789 : }
4790 :
4791 : void *
4792 0 : spdk_bdev_get_module_ctx(struct spdk_bdev_desc *desc)
4793 : {
4794 0 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
4795 0 : void *ctx = NULL;
4796 :
4797 0 : if (bdev->fn_table->get_module_ctx) {
4798 0 : ctx = bdev->fn_table->get_module_ctx(bdev->ctxt);
4799 0 : }
4800 :
4801 0 : return ctx;
4802 : }
4803 :
4804 : const char *
4805 0 : spdk_bdev_get_module_name(const struct spdk_bdev *bdev)
4806 : {
4807 0 : return bdev->module->name;
4808 : }
4809 :
4810 : const char *
4811 251 : spdk_bdev_get_name(const struct spdk_bdev *bdev)
4812 : {
4813 251 : return bdev->name;
4814 : }
4815 :
4816 : const char *
4817 0 : spdk_bdev_get_product_name(const struct spdk_bdev *bdev)
4818 : {
4819 0 : return bdev->product_name;
4820 : }
4821 :
4822 : const struct spdk_bdev_aliases_list *
4823 0 : spdk_bdev_get_aliases(const struct spdk_bdev *bdev)
4824 : {
4825 0 : return &bdev->aliases;
4826 : }
4827 :
4828 : uint32_t
4829 5 : spdk_bdev_get_block_size(const struct spdk_bdev *bdev)
4830 : {
4831 5 : return bdev->blocklen;
4832 : }
4833 :
4834 : uint32_t
4835 0 : spdk_bdev_get_write_unit_size(const struct spdk_bdev *bdev)
4836 : {
4837 0 : return bdev->write_unit_size;
4838 : }
4839 :
4840 : uint64_t
4841 0 : spdk_bdev_get_num_blocks(const struct spdk_bdev *bdev)
4842 : {
4843 0 : return bdev->blockcnt;
4844 : }
4845 :
4846 : const char *
4847 0 : spdk_bdev_get_qos_rpc_type(enum spdk_bdev_qos_rate_limit_type type)
4848 : {
4849 0 : return qos_rpc_type[type];
4850 : }
4851 :
4852 : void
4853 0 : spdk_bdev_get_qos_rate_limits(struct spdk_bdev *bdev, uint64_t *limits)
4854 : {
4855 : int i;
4856 :
4857 0 : memset(limits, 0, sizeof(*limits) * SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES);
4858 :
4859 0 : spdk_spin_lock(&bdev->internal.spinlock);
4860 0 : if (bdev->internal.qos) {
4861 0 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
4862 0 : if (bdev->internal.qos->rate_limits[i].limit !=
4863 : SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
4864 0 : limits[i] = bdev->internal.qos->rate_limits[i].limit;
4865 0 : if (bdev_qos_is_iops_rate_limit(i) == false) {
4866 : /* Change from Byte to Megabyte which is user visible. */
4867 0 : limits[i] = limits[i] / 1024 / 1024;
4868 0 : }
4869 0 : }
4870 0 : }
4871 0 : }
4872 0 : spdk_spin_unlock(&bdev->internal.spinlock);
4873 0 : }
4874 :
4875 : size_t
4876 316 : spdk_bdev_get_buf_align(const struct spdk_bdev *bdev)
4877 : {
4878 316 : return 1 << bdev->required_alignment;
4879 : }
4880 :
4881 : uint32_t
4882 0 : spdk_bdev_get_optimal_io_boundary(const struct spdk_bdev *bdev)
4883 : {
4884 0 : return bdev->optimal_io_boundary;
4885 : }
4886 :
4887 : bool
4888 0 : spdk_bdev_has_write_cache(const struct spdk_bdev *bdev)
4889 : {
4890 0 : return bdev->write_cache;
4891 : }
4892 :
4893 : const struct spdk_uuid *
4894 0 : spdk_bdev_get_uuid(const struct spdk_bdev *bdev)
4895 : {
4896 0 : return &bdev->uuid;
4897 : }
4898 :
4899 : uint16_t
4900 0 : spdk_bdev_get_acwu(const struct spdk_bdev *bdev)
4901 : {
4902 0 : return bdev->acwu;
4903 : }
4904 :
4905 : uint32_t
4906 29 : spdk_bdev_get_md_size(const struct spdk_bdev *bdev)
4907 : {
4908 29 : return bdev->md_len;
4909 : }
4910 :
4911 : bool
4912 129 : spdk_bdev_is_md_interleaved(const struct spdk_bdev *bdev)
4913 : {
4914 129 : return (bdev->md_len != 0) && bdev->md_interleave;
4915 : }
4916 :
4917 : bool
4918 158 : spdk_bdev_is_md_separate(const struct spdk_bdev *bdev)
4919 : {
4920 158 : return (bdev->md_len != 0) && !bdev->md_interleave;
4921 : }
4922 :
4923 : bool
4924 0 : spdk_bdev_is_zoned(const struct spdk_bdev *bdev)
4925 : {
4926 0 : return bdev->zoned;
4927 : }
4928 :
4929 : uint32_t
4930 120 : spdk_bdev_get_data_block_size(const struct spdk_bdev *bdev)
4931 : {
4932 120 : if (spdk_bdev_is_md_interleaved(bdev)) {
4933 0 : return bdev->blocklen - bdev->md_len;
4934 : } else {
4935 120 : return bdev->blocklen;
4936 : }
4937 120 : }
4938 :
4939 : uint32_t
4940 0 : spdk_bdev_get_physical_block_size(const struct spdk_bdev *bdev)
4941 : {
4942 0 : return bdev->phys_blocklen;
4943 : }
4944 :
4945 : static uint32_t
4946 9 : _bdev_get_block_size_with_md(const struct spdk_bdev *bdev)
4947 : {
4948 9 : if (!spdk_bdev_is_md_interleaved(bdev)) {
4949 6 : return bdev->blocklen + bdev->md_len;
4950 : } else {
4951 3 : return bdev->blocklen;
4952 : }
4953 9 : }
4954 :
4955 : /* We have to use the typedef in the function declaration to appease astyle. */
4956 : typedef enum spdk_dif_type spdk_dif_type_t;
4957 : typedef enum spdk_dif_pi_format spdk_dif_pi_format_t;
4958 :
4959 : spdk_dif_type_t
4960 0 : spdk_bdev_get_dif_type(const struct spdk_bdev *bdev)
4961 : {
4962 0 : if (bdev->md_len != 0) {
4963 0 : return bdev->dif_type;
4964 : } else {
4965 0 : return SPDK_DIF_DISABLE;
4966 : }
4967 0 : }
4968 :
4969 : spdk_dif_pi_format_t
4970 0 : spdk_bdev_get_dif_pi_format(const struct spdk_bdev *bdev)
4971 : {
4972 0 : return bdev->dif_pi_format;
4973 : }
4974 :
4975 : bool
4976 0 : spdk_bdev_is_dif_head_of_md(const struct spdk_bdev *bdev)
4977 : {
4978 0 : if (spdk_bdev_get_dif_type(bdev) != SPDK_DIF_DISABLE) {
4979 0 : return bdev->dif_is_head_of_md;
4980 : } else {
4981 0 : return false;
4982 : }
4983 0 : }
4984 :
4985 : bool
4986 0 : spdk_bdev_is_dif_check_enabled(const struct spdk_bdev *bdev,
4987 : enum spdk_dif_check_type check_type)
4988 : {
4989 0 : if (spdk_bdev_get_dif_type(bdev) == SPDK_DIF_DISABLE) {
4990 0 : return false;
4991 : }
4992 :
4993 0 : switch (check_type) {
4994 : case SPDK_DIF_CHECK_TYPE_REFTAG:
4995 0 : return (bdev->dif_check_flags & SPDK_DIF_FLAGS_REFTAG_CHECK) != 0;
4996 : case SPDK_DIF_CHECK_TYPE_APPTAG:
4997 0 : return (bdev->dif_check_flags & SPDK_DIF_FLAGS_APPTAG_CHECK) != 0;
4998 : case SPDK_DIF_CHECK_TYPE_GUARD:
4999 0 : return (bdev->dif_check_flags & SPDK_DIF_FLAGS_GUARD_CHECK) != 0;
5000 : default:
5001 0 : return false;
5002 : }
5003 0 : }
5004 :
5005 : static uint32_t
5006 3 : bdev_get_max_write(const struct spdk_bdev *bdev, uint64_t num_bytes)
5007 : {
5008 : uint64_t aligned_length, max_write_blocks;
5009 :
5010 3 : aligned_length = num_bytes - (spdk_bdev_get_buf_align(bdev) - 1);
5011 3 : max_write_blocks = aligned_length / _bdev_get_block_size_with_md(bdev);
5012 3 : max_write_blocks -= max_write_blocks % bdev->write_unit_size;
5013 :
5014 3 : return max_write_blocks;
5015 : }
5016 :
5017 : uint32_t
5018 1 : spdk_bdev_get_max_copy(const struct spdk_bdev *bdev)
5019 : {
5020 1 : return bdev->max_copy;
5021 : }
5022 :
5023 : uint64_t
5024 0 : spdk_bdev_get_qd(const struct spdk_bdev *bdev)
5025 : {
5026 0 : return bdev->internal.measured_queue_depth;
5027 : }
5028 :
5029 : uint64_t
5030 0 : spdk_bdev_get_qd_sampling_period(const struct spdk_bdev *bdev)
5031 : {
5032 0 : return bdev->internal.period;
5033 : }
5034 :
5035 : uint64_t
5036 0 : spdk_bdev_get_weighted_io_time(const struct spdk_bdev *bdev)
5037 : {
5038 0 : return bdev->internal.weighted_io_time;
5039 : }
5040 :
5041 : uint64_t
5042 0 : spdk_bdev_get_io_time(const struct spdk_bdev *bdev)
5043 : {
5044 0 : return bdev->internal.io_time;
5045 : }
5046 :
5047 0 : union spdk_bdev_nvme_ctratt spdk_bdev_get_nvme_ctratt(struct spdk_bdev *bdev)
5048 : {
5049 0 : return bdev->ctratt;
5050 : }
5051 :
5052 : static void bdev_update_qd_sampling_period(void *ctx);
5053 :
5054 : static void
5055 1 : _calculate_measured_qd_cpl(struct spdk_bdev *bdev, void *_ctx, int status)
5056 : {
5057 1 : bdev->internal.measured_queue_depth = bdev->internal.temporary_queue_depth;
5058 :
5059 1 : if (bdev->internal.measured_queue_depth) {
5060 0 : bdev->internal.io_time += bdev->internal.period;
5061 0 : bdev->internal.weighted_io_time += bdev->internal.period * bdev->internal.measured_queue_depth;
5062 0 : }
5063 :
5064 1 : bdev->internal.qd_poll_in_progress = false;
5065 :
5066 1 : bdev_update_qd_sampling_period(bdev);
5067 1 : }
5068 :
5069 : static void
5070 1 : _calculate_measured_qd(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
5071 : struct spdk_io_channel *io_ch, void *_ctx)
5072 : {
5073 1 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(io_ch);
5074 :
5075 1 : bdev->internal.temporary_queue_depth += ch->io_outstanding;
5076 1 : spdk_bdev_for_each_channel_continue(i, 0);
5077 1 : }
5078 :
5079 : static int
5080 1 : bdev_calculate_measured_queue_depth(void *ctx)
5081 : {
5082 1 : struct spdk_bdev *bdev = ctx;
5083 :
5084 1 : bdev->internal.qd_poll_in_progress = true;
5085 1 : bdev->internal.temporary_queue_depth = 0;
5086 1 : spdk_bdev_for_each_channel(bdev, _calculate_measured_qd, bdev, _calculate_measured_qd_cpl);
5087 1 : return SPDK_POLLER_BUSY;
5088 : }
5089 :
5090 : static void
5091 5 : bdev_update_qd_sampling_period(void *ctx)
5092 : {
5093 5 : struct spdk_bdev *bdev = ctx;
5094 :
5095 5 : if (bdev->internal.period == bdev->internal.new_period) {
5096 0 : return;
5097 : }
5098 :
5099 5 : if (bdev->internal.qd_poll_in_progress) {
5100 1 : return;
5101 : }
5102 :
5103 4 : bdev->internal.period = bdev->internal.new_period;
5104 :
5105 4 : spdk_poller_unregister(&bdev->internal.qd_poller);
5106 4 : if (bdev->internal.period != 0) {
5107 2 : bdev->internal.qd_poller = SPDK_POLLER_REGISTER(bdev_calculate_measured_queue_depth,
5108 : bdev, bdev->internal.period);
5109 2 : } else {
5110 2 : spdk_bdev_close(bdev->internal.qd_desc);
5111 2 : bdev->internal.qd_desc = NULL;
5112 : }
5113 5 : }
5114 :
5115 : static void
5116 0 : _tmp_bdev_event_cb(enum spdk_bdev_event_type type, struct spdk_bdev *bdev, void *ctx)
5117 : {
5118 0 : SPDK_NOTICELOG("Unexpected event type: %d\n", type);
5119 0 : }
5120 :
5121 : void
5122 129 : spdk_bdev_set_qd_sampling_period(struct spdk_bdev *bdev, uint64_t period)
5123 : {
5124 : int rc;
5125 :
5126 129 : if (bdev->internal.new_period == period) {
5127 123 : return;
5128 : }
5129 :
5130 6 : bdev->internal.new_period = period;
5131 :
5132 6 : if (bdev->internal.qd_desc != NULL) {
5133 4 : assert(bdev->internal.period != 0);
5134 :
5135 8 : spdk_thread_send_msg(bdev->internal.qd_desc->thread,
5136 4 : bdev_update_qd_sampling_period, bdev);
5137 4 : return;
5138 : }
5139 :
5140 2 : assert(bdev->internal.period == 0);
5141 :
5142 4 : rc = spdk_bdev_open_ext(spdk_bdev_get_name(bdev), false, _tmp_bdev_event_cb,
5143 2 : NULL, &bdev->internal.qd_desc);
5144 2 : if (rc != 0) {
5145 0 : return;
5146 : }
5147 :
5148 2 : bdev->internal.period = period;
5149 2 : bdev->internal.qd_poller = SPDK_POLLER_REGISTER(bdev_calculate_measured_queue_depth,
5150 : bdev, period);
5151 129 : }
5152 :
5153 : struct bdev_get_current_qd_ctx {
5154 : uint64_t current_qd;
5155 : spdk_bdev_get_current_qd_cb cb_fn;
5156 : void *cb_arg;
5157 : };
5158 :
5159 : static void
5160 0 : bdev_get_current_qd_done(struct spdk_bdev *bdev, void *_ctx, int status)
5161 : {
5162 0 : struct bdev_get_current_qd_ctx *ctx = _ctx;
5163 :
5164 0 : ctx->cb_fn(bdev, ctx->current_qd, ctx->cb_arg, 0);
5165 :
5166 0 : free(ctx);
5167 0 : }
5168 :
5169 : static void
5170 0 : bdev_get_current_qd(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
5171 : struct spdk_io_channel *io_ch, void *_ctx)
5172 : {
5173 0 : struct bdev_get_current_qd_ctx *ctx = _ctx;
5174 0 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
5175 :
5176 0 : ctx->current_qd += bdev_ch->io_outstanding;
5177 :
5178 0 : spdk_bdev_for_each_channel_continue(i, 0);
5179 0 : }
5180 :
5181 : void
5182 0 : spdk_bdev_get_current_qd(struct spdk_bdev *bdev, spdk_bdev_get_current_qd_cb cb_fn,
5183 : void *cb_arg)
5184 : {
5185 : struct bdev_get_current_qd_ctx *ctx;
5186 :
5187 0 : assert(cb_fn != NULL);
5188 :
5189 0 : ctx = calloc(1, sizeof(*ctx));
5190 0 : if (ctx == NULL) {
5191 0 : cb_fn(bdev, 0, cb_arg, -ENOMEM);
5192 0 : return;
5193 : }
5194 :
5195 0 : ctx->cb_fn = cb_fn;
5196 0 : ctx->cb_arg = cb_arg;
5197 :
5198 0 : spdk_bdev_for_each_channel(bdev, bdev_get_current_qd, ctx, bdev_get_current_qd_done);
5199 0 : }
5200 :
5201 : static void
5202 25 : _event_notify(struct spdk_bdev_desc *desc, enum spdk_bdev_event_type type)
5203 : {
5204 25 : assert(desc->thread == spdk_get_thread());
5205 :
5206 25 : spdk_spin_lock(&desc->spinlock);
5207 25 : desc->refs--;
5208 25 : if (!desc->closed) {
5209 14 : spdk_spin_unlock(&desc->spinlock);
5210 28 : desc->callback.event_fn(type,
5211 14 : desc->bdev,
5212 14 : desc->callback.ctx);
5213 14 : return;
5214 11 : } else if (desc->refs == 0) {
5215 : /* This descriptor was closed after this event_notify message was sent.
5216 : * spdk_bdev_close() could not free the descriptor since this message was
5217 : * in flight, so we free it now using bdev_desc_free().
5218 : */
5219 10 : spdk_spin_unlock(&desc->spinlock);
5220 10 : bdev_desc_free(desc);
5221 10 : return;
5222 : }
5223 1 : spdk_spin_unlock(&desc->spinlock);
5224 25 : }
5225 :
5226 : static void
5227 25 : event_notify(struct spdk_bdev_desc *desc, spdk_msg_fn event_notify_fn)
5228 : {
5229 25 : spdk_spin_lock(&desc->spinlock);
5230 25 : desc->refs++;
5231 25 : spdk_thread_send_msg(desc->thread, event_notify_fn, desc);
5232 25 : spdk_spin_unlock(&desc->spinlock);
5233 25 : }
5234 :
5235 : static void
5236 6 : _resize_notify(void *ctx)
5237 : {
5238 6 : struct spdk_bdev_desc *desc = ctx;
5239 :
5240 6 : _event_notify(desc, SPDK_BDEV_EVENT_RESIZE);
5241 6 : }
5242 :
5243 : int
5244 11 : spdk_bdev_notify_blockcnt_change(struct spdk_bdev *bdev, uint64_t size)
5245 : {
5246 : struct spdk_bdev_desc *desc;
5247 : int ret;
5248 :
5249 11 : if (size == bdev->blockcnt) {
5250 0 : return 0;
5251 : }
5252 :
5253 11 : spdk_spin_lock(&bdev->internal.spinlock);
5254 :
5255 : /* bdev has open descriptors */
5256 11 : if (!TAILQ_EMPTY(&bdev->internal.open_descs) &&
5257 7 : bdev->blockcnt > size) {
5258 1 : ret = -EBUSY;
5259 1 : } else {
5260 10 : bdev->blockcnt = size;
5261 16 : TAILQ_FOREACH(desc, &bdev->internal.open_descs, link) {
5262 6 : event_notify(desc, _resize_notify);
5263 6 : }
5264 10 : ret = 0;
5265 : }
5266 :
5267 11 : spdk_spin_unlock(&bdev->internal.spinlock);
5268 :
5269 11 : return ret;
5270 11 : }
5271 :
5272 : /*
5273 : * Convert I/O offset and length from bytes to blocks.
5274 : *
5275 : * Returns zero on success or non-zero if the byte parameters aren't divisible by the block size.
5276 : */
5277 : static uint64_t
5278 20 : bdev_bytes_to_blocks(struct spdk_bdev *bdev, uint64_t offset_bytes, uint64_t *offset_blocks,
5279 : uint64_t num_bytes, uint64_t *num_blocks)
5280 : {
5281 20 : uint32_t block_size = bdev->blocklen;
5282 : uint8_t shift_cnt;
5283 :
5284 : /* Avoid expensive div operations if possible. These spdk_u32 functions are very cheap. */
5285 20 : if (spdk_likely(spdk_u32_is_pow2(block_size))) {
5286 17 : shift_cnt = spdk_u32log2(block_size);
5287 17 : *offset_blocks = offset_bytes >> shift_cnt;
5288 17 : *num_blocks = num_bytes >> shift_cnt;
5289 34 : return (offset_bytes - (*offset_blocks << shift_cnt)) |
5290 17 : (num_bytes - (*num_blocks << shift_cnt));
5291 : } else {
5292 3 : *offset_blocks = offset_bytes / block_size;
5293 3 : *num_blocks = num_bytes / block_size;
5294 3 : return (offset_bytes % block_size) | (num_bytes % block_size);
5295 : }
5296 20 : }
5297 :
5298 : static bool
5299 689 : bdev_io_valid_blocks(struct spdk_bdev *bdev, uint64_t offset_blocks, uint64_t num_blocks)
5300 : {
5301 : /* Return failure if offset_blocks + num_blocks is less than offset_blocks; indicates there
5302 : * has been an overflow and hence the offset has been wrapped around */
5303 689 : if (offset_blocks + num_blocks < offset_blocks) {
5304 1 : return false;
5305 : }
5306 :
5307 : /* Return failure if offset_blocks + num_blocks exceeds the size of the bdev */
5308 688 : if (offset_blocks + num_blocks > bdev->blockcnt) {
5309 2 : return false;
5310 : }
5311 :
5312 686 : return true;
5313 689 : }
5314 :
5315 : static void
5316 2 : bdev_seek_complete_cb(void *ctx)
5317 : {
5318 2 : struct spdk_bdev_io *bdev_io = ctx;
5319 :
5320 2 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
5321 2 : bdev_io->internal.cb(bdev_io, true, bdev_io->internal.caller_ctx);
5322 2 : }
5323 :
5324 : static int
5325 4 : bdev_seek(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5326 : uint64_t offset_blocks, enum spdk_bdev_io_type io_type,
5327 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5328 : {
5329 4 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5330 : struct spdk_bdev_io *bdev_io;
5331 4 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5332 :
5333 4 : assert(io_type == SPDK_BDEV_IO_TYPE_SEEK_DATA || io_type == SPDK_BDEV_IO_TYPE_SEEK_HOLE);
5334 :
5335 : /* Check if offset_blocks is valid looking at the validity of one block */
5336 4 : if (!bdev_io_valid_blocks(bdev, offset_blocks, 1)) {
5337 0 : return -EINVAL;
5338 : }
5339 :
5340 4 : bdev_io = bdev_channel_get_io(channel);
5341 4 : if (!bdev_io) {
5342 0 : return -ENOMEM;
5343 : }
5344 :
5345 4 : bdev_io->internal.ch = channel;
5346 4 : bdev_io->internal.desc = desc;
5347 4 : bdev_io->type = io_type;
5348 4 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5349 4 : bdev_io->u.bdev.memory_domain = NULL;
5350 4 : bdev_io->u.bdev.memory_domain_ctx = NULL;
5351 4 : bdev_io->u.bdev.accel_sequence = NULL;
5352 4 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5353 :
5354 4 : if (!spdk_bdev_io_type_supported(bdev, io_type)) {
5355 : /* In case bdev doesn't support seek to next data/hole offset,
5356 : * it is assumed that only data and no holes are present */
5357 2 : if (io_type == SPDK_BDEV_IO_TYPE_SEEK_DATA) {
5358 1 : bdev_io->u.bdev.seek.offset = offset_blocks;
5359 1 : } else {
5360 1 : bdev_io->u.bdev.seek.offset = UINT64_MAX;
5361 : }
5362 :
5363 2 : spdk_thread_send_msg(spdk_get_thread(), bdev_seek_complete_cb, bdev_io);
5364 2 : return 0;
5365 : }
5366 :
5367 2 : bdev_io_submit(bdev_io);
5368 2 : return 0;
5369 4 : }
5370 :
5371 : int
5372 2 : spdk_bdev_seek_data(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5373 : uint64_t offset_blocks,
5374 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5375 : {
5376 2 : return bdev_seek(desc, ch, offset_blocks, SPDK_BDEV_IO_TYPE_SEEK_DATA, cb, cb_arg);
5377 : }
5378 :
5379 : int
5380 2 : spdk_bdev_seek_hole(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5381 : uint64_t offset_blocks,
5382 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5383 : {
5384 2 : return bdev_seek(desc, ch, offset_blocks, SPDK_BDEV_IO_TYPE_SEEK_HOLE, cb, cb_arg);
5385 : }
5386 :
5387 : uint64_t
5388 4 : spdk_bdev_io_get_seek_offset(const struct spdk_bdev_io *bdev_io)
5389 : {
5390 4 : return bdev_io->u.bdev.seek.offset;
5391 : }
5392 :
5393 : static int
5394 204 : bdev_read_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch, void *buf,
5395 : void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
5396 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5397 : {
5398 204 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5399 : struct spdk_bdev_io *bdev_io;
5400 204 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5401 :
5402 204 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
5403 0 : return -EINVAL;
5404 : }
5405 :
5406 204 : bdev_io = bdev_channel_get_io(channel);
5407 204 : if (!bdev_io) {
5408 1 : return -ENOMEM;
5409 : }
5410 :
5411 203 : bdev_io->internal.ch = channel;
5412 203 : bdev_io->internal.desc = desc;
5413 203 : bdev_io->type = SPDK_BDEV_IO_TYPE_READ;
5414 203 : bdev_io->u.bdev.iovs = &bdev_io->iov;
5415 203 : bdev_io->u.bdev.iovs[0].iov_base = buf;
5416 203 : bdev_io->u.bdev.iovs[0].iov_len = num_blocks * bdev->blocklen;
5417 203 : bdev_io->u.bdev.iovcnt = 1;
5418 203 : bdev_io->u.bdev.md_buf = md_buf;
5419 203 : bdev_io->u.bdev.num_blocks = num_blocks;
5420 203 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5421 203 : bdev_io->u.bdev.memory_domain = NULL;
5422 203 : bdev_io->u.bdev.memory_domain_ctx = NULL;
5423 203 : bdev_io->u.bdev.accel_sequence = NULL;
5424 203 : bdev_io->u.bdev.dif_check_flags = bdev->dif_check_flags;
5425 203 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5426 :
5427 203 : bdev_io_submit(bdev_io);
5428 203 : return 0;
5429 204 : }
5430 :
5431 : int
5432 3 : spdk_bdev_read(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5433 : void *buf, uint64_t offset, uint64_t nbytes,
5434 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5435 : {
5436 : uint64_t offset_blocks, num_blocks;
5437 :
5438 9 : if (bdev_bytes_to_blocks(spdk_bdev_desc_get_bdev(desc), offset, &offset_blocks,
5439 6 : nbytes, &num_blocks) != 0) {
5440 0 : return -EINVAL;
5441 : }
5442 :
5443 3 : return spdk_bdev_read_blocks(desc, ch, buf, offset_blocks, num_blocks, cb, cb_arg);
5444 3 : }
5445 :
5446 : int
5447 200 : spdk_bdev_read_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5448 : void *buf, uint64_t offset_blocks, uint64_t num_blocks,
5449 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5450 : {
5451 200 : return bdev_read_blocks_with_md(desc, ch, buf, NULL, offset_blocks, num_blocks, cb, cb_arg);
5452 : }
5453 :
5454 : int
5455 4 : spdk_bdev_read_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5456 : void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
5457 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5458 : {
5459 8 : struct iovec iov = {
5460 4 : .iov_base = buf,
5461 : };
5462 :
5463 4 : if (md_buf && !spdk_bdev_is_md_separate(spdk_bdev_desc_get_bdev(desc))) {
5464 0 : return -EINVAL;
5465 : }
5466 :
5467 4 : if (md_buf && !_is_buf_allocated(&iov)) {
5468 0 : return -EINVAL;
5469 : }
5470 :
5471 8 : return bdev_read_blocks_with_md(desc, ch, buf, md_buf, offset_blocks, num_blocks,
5472 4 : cb, cb_arg);
5473 4 : }
5474 :
5475 : int
5476 5 : spdk_bdev_readv(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5477 : struct iovec *iov, int iovcnt,
5478 : uint64_t offset, uint64_t nbytes,
5479 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5480 : {
5481 : uint64_t offset_blocks, num_blocks;
5482 :
5483 15 : if (bdev_bytes_to_blocks(spdk_bdev_desc_get_bdev(desc), offset, &offset_blocks,
5484 10 : nbytes, &num_blocks) != 0) {
5485 0 : return -EINVAL;
5486 : }
5487 :
5488 5 : return spdk_bdev_readv_blocks(desc, ch, iov, iovcnt, offset_blocks, num_blocks, cb, cb_arg);
5489 5 : }
5490 :
5491 : static int
5492 226 : bdev_readv_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5493 : struct iovec *iov, int iovcnt, void *md_buf, uint64_t offset_blocks,
5494 : uint64_t num_blocks, struct spdk_memory_domain *domain, void *domain_ctx,
5495 : struct spdk_accel_sequence *seq, uint32_t dif_check_flags,
5496 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5497 : {
5498 226 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5499 : struct spdk_bdev_io *bdev_io;
5500 226 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5501 :
5502 226 : if (spdk_unlikely(!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks))) {
5503 0 : return -EINVAL;
5504 : }
5505 :
5506 226 : bdev_io = bdev_channel_get_io(channel);
5507 226 : if (spdk_unlikely(!bdev_io)) {
5508 2 : return -ENOMEM;
5509 : }
5510 :
5511 224 : bdev_io->internal.ch = channel;
5512 224 : bdev_io->internal.desc = desc;
5513 224 : bdev_io->type = SPDK_BDEV_IO_TYPE_READ;
5514 224 : bdev_io->u.bdev.iovs = iov;
5515 224 : bdev_io->u.bdev.iovcnt = iovcnt;
5516 224 : bdev_io->u.bdev.md_buf = md_buf;
5517 224 : bdev_io->u.bdev.num_blocks = num_blocks;
5518 224 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5519 224 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5520 :
5521 224 : if (seq != NULL) {
5522 0 : bdev_io->internal.f.has_accel_sequence = true;
5523 0 : bdev_io->internal.accel_sequence = seq;
5524 0 : }
5525 :
5526 224 : if (domain != NULL) {
5527 2 : bdev_io->internal.f.has_memory_domain = true;
5528 2 : bdev_io->internal.memory_domain = domain;
5529 2 : bdev_io->internal.memory_domain_ctx = domain_ctx;
5530 2 : }
5531 :
5532 224 : bdev_io->u.bdev.memory_domain = domain;
5533 224 : bdev_io->u.bdev.memory_domain_ctx = domain_ctx;
5534 224 : bdev_io->u.bdev.accel_sequence = seq;
5535 224 : bdev_io->u.bdev.dif_check_flags = dif_check_flags;
5536 :
5537 224 : _bdev_io_submit_ext(desc, bdev_io);
5538 :
5539 224 : return 0;
5540 226 : }
5541 :
5542 : int
5543 21 : spdk_bdev_readv_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5544 : struct iovec *iov, int iovcnt,
5545 : uint64_t offset_blocks, uint64_t num_blocks,
5546 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5547 : {
5548 21 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5549 :
5550 42 : return bdev_readv_blocks_with_md(desc, ch, iov, iovcnt, NULL, offset_blocks,
5551 21 : num_blocks, NULL, NULL, NULL, bdev->dif_check_flags, cb, cb_arg);
5552 : }
5553 :
5554 : int
5555 4 : spdk_bdev_readv_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5556 : struct iovec *iov, int iovcnt, void *md_buf,
5557 : uint64_t offset_blocks, uint64_t num_blocks,
5558 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5559 : {
5560 4 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5561 :
5562 4 : if (md_buf && !spdk_bdev_is_md_separate(bdev)) {
5563 0 : return -EINVAL;
5564 : }
5565 :
5566 4 : if (md_buf && !_is_buf_allocated(iov)) {
5567 0 : return -EINVAL;
5568 : }
5569 :
5570 8 : return bdev_readv_blocks_with_md(desc, ch, iov, iovcnt, md_buf, offset_blocks,
5571 4 : num_blocks, NULL, NULL, NULL, bdev->dif_check_flags, cb, cb_arg);
5572 4 : }
5573 :
5574 : static inline bool
5575 14 : _bdev_io_check_opts(struct spdk_bdev_ext_io_opts *opts, struct iovec *iov)
5576 : {
5577 : /*
5578 : * We check if opts size is at least of size when we first introduced
5579 : * spdk_bdev_ext_io_opts (ac6f2bdd8d) since access to those members
5580 : * are not checked internal.
5581 : */
5582 24 : return opts->size >= offsetof(struct spdk_bdev_ext_io_opts, metadata) +
5583 14 : sizeof(opts->metadata) &&
5584 10 : opts->size <= sizeof(*opts) &&
5585 : /* When memory domain is used, the user must provide data buffers */
5586 8 : (!opts->memory_domain || (iov && iov[0].iov_base));
5587 : }
5588 :
5589 : int
5590 8 : spdk_bdev_readv_blocks_ext(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5591 : struct iovec *iov, int iovcnt,
5592 : uint64_t offset_blocks, uint64_t num_blocks,
5593 : spdk_bdev_io_completion_cb cb, void *cb_arg,
5594 : struct spdk_bdev_ext_io_opts *opts)
5595 : {
5596 8 : struct spdk_memory_domain *domain = NULL;
5597 8 : struct spdk_accel_sequence *seq = NULL;
5598 8 : void *domain_ctx = NULL, *md = NULL;
5599 8 : uint32_t dif_check_flags = 0;
5600 8 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5601 :
5602 8 : if (opts) {
5603 7 : if (spdk_unlikely(!_bdev_io_check_opts(opts, iov))) {
5604 3 : return -EINVAL;
5605 : }
5606 :
5607 4 : md = opts->metadata;
5608 4 : domain = bdev_get_ext_io_opt(opts, memory_domain, NULL);
5609 4 : domain_ctx = bdev_get_ext_io_opt(opts, memory_domain_ctx, NULL);
5610 4 : seq = bdev_get_ext_io_opt(opts, accel_sequence, NULL);
5611 4 : if (md) {
5612 4 : if (spdk_unlikely(!spdk_bdev_is_md_separate(bdev))) {
5613 0 : return -EINVAL;
5614 : }
5615 :
5616 4 : if (spdk_unlikely(!_is_buf_allocated(iov))) {
5617 0 : return -EINVAL;
5618 : }
5619 :
5620 4 : if (spdk_unlikely(seq != NULL)) {
5621 0 : return -EINVAL;
5622 : }
5623 4 : }
5624 4 : }
5625 :
5626 10 : dif_check_flags = bdev->dif_check_flags &
5627 5 : ~(bdev_get_ext_io_opt(opts, dif_check_flags_exclude_mask, 0));
5628 :
5629 10 : return bdev_readv_blocks_with_md(desc, ch, iov, iovcnt, md, offset_blocks,
5630 5 : num_blocks, domain, domain_ctx, seq, dif_check_flags, cb, cb_arg);
5631 8 : }
5632 :
5633 : static int
5634 36 : bdev_write_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5635 : void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
5636 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5637 : {
5638 36 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5639 : struct spdk_bdev_io *bdev_io;
5640 36 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5641 :
5642 36 : if (!desc->write) {
5643 0 : return -EBADF;
5644 : }
5645 :
5646 36 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
5647 0 : return -EINVAL;
5648 : }
5649 :
5650 36 : bdev_io = bdev_channel_get_io(channel);
5651 36 : if (!bdev_io) {
5652 0 : return -ENOMEM;
5653 : }
5654 :
5655 36 : bdev_io->internal.ch = channel;
5656 36 : bdev_io->internal.desc = desc;
5657 36 : bdev_io->type = SPDK_BDEV_IO_TYPE_WRITE;
5658 36 : bdev_io->u.bdev.iovs = &bdev_io->iov;
5659 36 : bdev_io->u.bdev.iovs[0].iov_base = buf;
5660 36 : bdev_io->u.bdev.iovs[0].iov_len = num_blocks * bdev->blocklen;
5661 36 : bdev_io->u.bdev.iovcnt = 1;
5662 36 : bdev_io->u.bdev.md_buf = md_buf;
5663 36 : bdev_io->u.bdev.num_blocks = num_blocks;
5664 36 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5665 36 : bdev_io->u.bdev.memory_domain = NULL;
5666 36 : bdev_io->u.bdev.memory_domain_ctx = NULL;
5667 36 : bdev_io->u.bdev.accel_sequence = NULL;
5668 36 : bdev_io->u.bdev.dif_check_flags = bdev->dif_check_flags;
5669 36 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5670 :
5671 36 : bdev_io_submit(bdev_io);
5672 36 : return 0;
5673 36 : }
5674 :
5675 : int
5676 3 : spdk_bdev_write(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5677 : void *buf, uint64_t offset, uint64_t nbytes,
5678 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5679 : {
5680 : uint64_t offset_blocks, num_blocks;
5681 :
5682 9 : if (bdev_bytes_to_blocks(spdk_bdev_desc_get_bdev(desc), offset, &offset_blocks,
5683 6 : nbytes, &num_blocks) != 0) {
5684 0 : return -EINVAL;
5685 : }
5686 :
5687 3 : return spdk_bdev_write_blocks(desc, ch, buf, offset_blocks, num_blocks, cb, cb_arg);
5688 3 : }
5689 :
5690 : int
5691 27 : spdk_bdev_write_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5692 : void *buf, uint64_t offset_blocks, uint64_t num_blocks,
5693 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5694 : {
5695 54 : return bdev_write_blocks_with_md(desc, ch, buf, NULL, offset_blocks, num_blocks,
5696 27 : cb, cb_arg);
5697 : }
5698 :
5699 : int
5700 3 : spdk_bdev_write_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5701 : void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
5702 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5703 : {
5704 6 : struct iovec iov = {
5705 3 : .iov_base = buf,
5706 : };
5707 :
5708 3 : if (md_buf && !spdk_bdev_is_md_separate(spdk_bdev_desc_get_bdev(desc))) {
5709 0 : return -EINVAL;
5710 : }
5711 :
5712 3 : if (md_buf && !_is_buf_allocated(&iov)) {
5713 0 : return -EINVAL;
5714 : }
5715 :
5716 6 : return bdev_write_blocks_with_md(desc, ch, buf, md_buf, offset_blocks, num_blocks,
5717 3 : cb, cb_arg);
5718 3 : }
5719 :
5720 : static int
5721 70 : bdev_writev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5722 : struct iovec *iov, int iovcnt, void *md_buf,
5723 : uint64_t offset_blocks, uint64_t num_blocks,
5724 : struct spdk_memory_domain *domain, void *domain_ctx,
5725 : struct spdk_accel_sequence *seq, uint32_t dif_check_flags,
5726 : uint32_t nvme_cdw12_raw, uint32_t nvme_cdw13_raw,
5727 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5728 : {
5729 70 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5730 : struct spdk_bdev_io *bdev_io;
5731 70 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5732 :
5733 70 : if (spdk_unlikely(!desc->write)) {
5734 0 : return -EBADF;
5735 : }
5736 :
5737 70 : if (spdk_unlikely(!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks))) {
5738 0 : return -EINVAL;
5739 : }
5740 :
5741 70 : bdev_io = bdev_channel_get_io(channel);
5742 70 : if (spdk_unlikely(!bdev_io)) {
5743 2 : return -ENOMEM;
5744 : }
5745 :
5746 68 : bdev_io->internal.ch = channel;
5747 68 : bdev_io->internal.desc = desc;
5748 68 : bdev_io->type = SPDK_BDEV_IO_TYPE_WRITE;
5749 68 : bdev_io->u.bdev.iovs = iov;
5750 68 : bdev_io->u.bdev.iovcnt = iovcnt;
5751 68 : bdev_io->u.bdev.md_buf = md_buf;
5752 68 : bdev_io->u.bdev.num_blocks = num_blocks;
5753 68 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5754 68 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5755 68 : if (seq != NULL) {
5756 0 : bdev_io->internal.f.has_accel_sequence = true;
5757 0 : bdev_io->internal.accel_sequence = seq;
5758 0 : }
5759 :
5760 68 : if (domain != NULL) {
5761 2 : bdev_io->internal.f.has_memory_domain = true;
5762 2 : bdev_io->internal.memory_domain = domain;
5763 2 : bdev_io->internal.memory_domain_ctx = domain_ctx;
5764 2 : }
5765 :
5766 68 : bdev_io->u.bdev.memory_domain = domain;
5767 68 : bdev_io->u.bdev.memory_domain_ctx = domain_ctx;
5768 68 : bdev_io->u.bdev.accel_sequence = seq;
5769 68 : bdev_io->u.bdev.dif_check_flags = dif_check_flags;
5770 68 : bdev_io->u.bdev.nvme_cdw12.raw = nvme_cdw12_raw;
5771 68 : bdev_io->u.bdev.nvme_cdw13.raw = nvme_cdw13_raw;
5772 :
5773 68 : _bdev_io_submit_ext(desc, bdev_io);
5774 :
5775 68 : return 0;
5776 70 : }
5777 :
5778 : int
5779 3 : spdk_bdev_writev(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5780 : struct iovec *iov, int iovcnt,
5781 : uint64_t offset, uint64_t len,
5782 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5783 : {
5784 : uint64_t offset_blocks, num_blocks;
5785 :
5786 9 : if (bdev_bytes_to_blocks(spdk_bdev_desc_get_bdev(desc), offset, &offset_blocks,
5787 6 : len, &num_blocks) != 0) {
5788 0 : return -EINVAL;
5789 : }
5790 :
5791 3 : return spdk_bdev_writev_blocks(desc, ch, iov, iovcnt, offset_blocks, num_blocks, cb, cb_arg);
5792 3 : }
5793 :
5794 : int
5795 14 : spdk_bdev_writev_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5796 : struct iovec *iov, int iovcnt,
5797 : uint64_t offset_blocks, uint64_t num_blocks,
5798 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5799 : {
5800 14 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5801 :
5802 28 : return bdev_writev_blocks_with_md(desc, ch, iov, iovcnt, NULL, offset_blocks,
5803 14 : num_blocks, NULL, NULL, NULL, bdev->dif_check_flags, 0, 0,
5804 14 : cb, cb_arg);
5805 : }
5806 :
5807 : int
5808 1 : spdk_bdev_writev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5809 : struct iovec *iov, int iovcnt, void *md_buf,
5810 : uint64_t offset_blocks, uint64_t num_blocks,
5811 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5812 : {
5813 1 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5814 :
5815 1 : if (md_buf && !spdk_bdev_is_md_separate(bdev)) {
5816 0 : return -EINVAL;
5817 : }
5818 :
5819 1 : if (md_buf && !_is_buf_allocated(iov)) {
5820 0 : return -EINVAL;
5821 : }
5822 :
5823 2 : return bdev_writev_blocks_with_md(desc, ch, iov, iovcnt, md_buf, offset_blocks,
5824 1 : num_blocks, NULL, NULL, NULL, bdev->dif_check_flags, 0, 0,
5825 1 : cb, cb_arg);
5826 1 : }
5827 :
5828 : int
5829 8 : spdk_bdev_writev_blocks_ext(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5830 : struct iovec *iov, int iovcnt,
5831 : uint64_t offset_blocks, uint64_t num_blocks,
5832 : spdk_bdev_io_completion_cb cb, void *cb_arg,
5833 : struct spdk_bdev_ext_io_opts *opts)
5834 : {
5835 8 : struct spdk_memory_domain *domain = NULL;
5836 8 : struct spdk_accel_sequence *seq = NULL;
5837 8 : void *domain_ctx = NULL, *md = NULL;
5838 8 : uint32_t dif_check_flags = 0;
5839 8 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5840 8 : uint32_t nvme_cdw12_raw = 0;
5841 8 : uint32_t nvme_cdw13_raw = 0;
5842 :
5843 8 : if (opts) {
5844 7 : if (spdk_unlikely(!_bdev_io_check_opts(opts, iov))) {
5845 3 : return -EINVAL;
5846 : }
5847 4 : md = opts->metadata;
5848 4 : domain = bdev_get_ext_io_opt(opts, memory_domain, NULL);
5849 4 : domain_ctx = bdev_get_ext_io_opt(opts, memory_domain_ctx, NULL);
5850 4 : seq = bdev_get_ext_io_opt(opts, accel_sequence, NULL);
5851 4 : nvme_cdw12_raw = bdev_get_ext_io_opt(opts, nvme_cdw12.raw, 0);
5852 4 : nvme_cdw13_raw = bdev_get_ext_io_opt(opts, nvme_cdw13.raw, 0);
5853 4 : if (md) {
5854 4 : if (spdk_unlikely(!spdk_bdev_is_md_separate(bdev))) {
5855 0 : return -EINVAL;
5856 : }
5857 :
5858 4 : if (spdk_unlikely(!_is_buf_allocated(iov))) {
5859 0 : return -EINVAL;
5860 : }
5861 :
5862 4 : if (spdk_unlikely(seq != NULL)) {
5863 0 : return -EINVAL;
5864 : }
5865 4 : }
5866 4 : }
5867 :
5868 10 : dif_check_flags = bdev->dif_check_flags &
5869 5 : ~(bdev_get_ext_io_opt(opts, dif_check_flags_exclude_mask, 0));
5870 :
5871 10 : return bdev_writev_blocks_with_md(desc, ch, iov, iovcnt, md, offset_blocks, num_blocks,
5872 5 : domain, domain_ctx, seq, dif_check_flags,
5873 5 : nvme_cdw12_raw, nvme_cdw13_raw, cb, cb_arg);
5874 8 : }
5875 :
5876 : static void
5877 11 : bdev_compare_do_read_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
5878 : {
5879 11 : struct spdk_bdev_io *parent_io = cb_arg;
5880 11 : struct spdk_bdev *bdev = parent_io->bdev;
5881 11 : uint8_t *read_buf = bdev_io->u.bdev.iovs[0].iov_base;
5882 11 : int i, rc = 0;
5883 :
5884 11 : if (!success) {
5885 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
5886 0 : parent_io->internal.cb(parent_io, false, parent_io->internal.caller_ctx);
5887 0 : spdk_bdev_free_io(bdev_io);
5888 0 : return;
5889 : }
5890 :
5891 17 : for (i = 0; i < parent_io->u.bdev.iovcnt; i++) {
5892 22 : rc = memcmp(read_buf,
5893 11 : parent_io->u.bdev.iovs[i].iov_base,
5894 11 : parent_io->u.bdev.iovs[i].iov_len);
5895 11 : if (rc) {
5896 5 : break;
5897 : }
5898 6 : read_buf += parent_io->u.bdev.iovs[i].iov_len;
5899 6 : }
5900 :
5901 11 : if (rc == 0 && parent_io->u.bdev.md_buf && spdk_bdev_is_md_separate(bdev)) {
5902 4 : rc = memcmp(bdev_io->u.bdev.md_buf,
5903 2 : parent_io->u.bdev.md_buf,
5904 2 : spdk_bdev_get_md_size(bdev));
5905 2 : }
5906 :
5907 11 : spdk_bdev_free_io(bdev_io);
5908 :
5909 11 : if (rc == 0) {
5910 5 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
5911 5 : parent_io->internal.cb(parent_io, true, parent_io->internal.caller_ctx);
5912 5 : } else {
5913 6 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_MISCOMPARE;
5914 6 : parent_io->internal.cb(parent_io, false, parent_io->internal.caller_ctx);
5915 : }
5916 11 : }
5917 :
5918 : static void
5919 11 : bdev_compare_do_read(void *_bdev_io)
5920 : {
5921 11 : struct spdk_bdev_io *bdev_io = _bdev_io;
5922 : int rc;
5923 :
5924 22 : rc = spdk_bdev_read_blocks(bdev_io->internal.desc,
5925 11 : spdk_io_channel_from_ctx(bdev_io->internal.ch), NULL,
5926 11 : bdev_io->u.bdev.offset_blocks, bdev_io->u.bdev.num_blocks,
5927 11 : bdev_compare_do_read_done, bdev_io);
5928 :
5929 11 : if (rc == -ENOMEM) {
5930 0 : bdev_queue_io_wait_with_cb(bdev_io, bdev_compare_do_read);
5931 11 : } else if (rc != 0) {
5932 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
5933 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
5934 0 : }
5935 11 : }
5936 :
5937 : static int
5938 16 : bdev_comparev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5939 : struct iovec *iov, int iovcnt, void *md_buf,
5940 : uint64_t offset_blocks, uint64_t num_blocks,
5941 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5942 : {
5943 16 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5944 : struct spdk_bdev_io *bdev_io;
5945 16 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5946 :
5947 16 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
5948 0 : return -EINVAL;
5949 : }
5950 :
5951 16 : bdev_io = bdev_channel_get_io(channel);
5952 16 : if (!bdev_io) {
5953 0 : return -ENOMEM;
5954 : }
5955 :
5956 16 : bdev_io->internal.ch = channel;
5957 16 : bdev_io->internal.desc = desc;
5958 16 : bdev_io->type = SPDK_BDEV_IO_TYPE_COMPARE;
5959 16 : bdev_io->u.bdev.iovs = iov;
5960 16 : bdev_io->u.bdev.iovcnt = iovcnt;
5961 16 : bdev_io->u.bdev.md_buf = md_buf;
5962 16 : bdev_io->u.bdev.num_blocks = num_blocks;
5963 16 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5964 16 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5965 16 : bdev_io->u.bdev.memory_domain = NULL;
5966 16 : bdev_io->u.bdev.memory_domain_ctx = NULL;
5967 16 : bdev_io->u.bdev.accel_sequence = NULL;
5968 :
5969 16 : if (bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COMPARE)) {
5970 7 : bdev_io_submit(bdev_io);
5971 7 : return 0;
5972 : }
5973 :
5974 9 : bdev_compare_do_read(bdev_io);
5975 :
5976 9 : return 0;
5977 16 : }
5978 :
5979 : int
5980 10 : spdk_bdev_comparev_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5981 : struct iovec *iov, int iovcnt,
5982 : uint64_t offset_blocks, uint64_t num_blocks,
5983 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5984 : {
5985 20 : return bdev_comparev_blocks_with_md(desc, ch, iov, iovcnt, NULL, offset_blocks,
5986 10 : num_blocks, cb, cb_arg);
5987 : }
5988 :
5989 : int
5990 6 : spdk_bdev_comparev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5991 : struct iovec *iov, int iovcnt, void *md_buf,
5992 : uint64_t offset_blocks, uint64_t num_blocks,
5993 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5994 : {
5995 6 : if (md_buf && !spdk_bdev_is_md_separate(spdk_bdev_desc_get_bdev(desc))) {
5996 0 : return -EINVAL;
5997 : }
5998 :
5999 6 : if (md_buf && !_is_buf_allocated(iov)) {
6000 0 : return -EINVAL;
6001 : }
6002 :
6003 12 : return bdev_comparev_blocks_with_md(desc, ch, iov, iovcnt, md_buf, offset_blocks,
6004 6 : num_blocks, cb, cb_arg);
6005 6 : }
6006 :
6007 : static int
6008 4 : bdev_compare_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6009 : void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
6010 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6011 : {
6012 4 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6013 : struct spdk_bdev_io *bdev_io;
6014 4 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6015 :
6016 4 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6017 0 : return -EINVAL;
6018 : }
6019 :
6020 4 : bdev_io = bdev_channel_get_io(channel);
6021 4 : if (!bdev_io) {
6022 0 : return -ENOMEM;
6023 : }
6024 :
6025 4 : bdev_io->internal.ch = channel;
6026 4 : bdev_io->internal.desc = desc;
6027 4 : bdev_io->type = SPDK_BDEV_IO_TYPE_COMPARE;
6028 4 : bdev_io->u.bdev.iovs = &bdev_io->iov;
6029 4 : bdev_io->u.bdev.iovs[0].iov_base = buf;
6030 4 : bdev_io->u.bdev.iovs[0].iov_len = num_blocks * bdev->blocklen;
6031 4 : bdev_io->u.bdev.iovcnt = 1;
6032 4 : bdev_io->u.bdev.md_buf = md_buf;
6033 4 : bdev_io->u.bdev.num_blocks = num_blocks;
6034 4 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6035 4 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6036 4 : bdev_io->u.bdev.memory_domain = NULL;
6037 4 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6038 4 : bdev_io->u.bdev.accel_sequence = NULL;
6039 :
6040 4 : if (bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COMPARE)) {
6041 2 : bdev_io_submit(bdev_io);
6042 2 : return 0;
6043 : }
6044 :
6045 2 : bdev_compare_do_read(bdev_io);
6046 :
6047 2 : return 0;
6048 4 : }
6049 :
6050 : int
6051 4 : spdk_bdev_compare_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6052 : void *buf, uint64_t offset_blocks, uint64_t num_blocks,
6053 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6054 : {
6055 8 : return bdev_compare_blocks_with_md(desc, ch, buf, NULL, offset_blocks, num_blocks,
6056 4 : cb, cb_arg);
6057 : }
6058 :
6059 : int
6060 0 : spdk_bdev_compare_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6061 : void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
6062 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6063 : {
6064 0 : struct iovec iov = {
6065 0 : .iov_base = buf,
6066 : };
6067 :
6068 0 : if (md_buf && !spdk_bdev_is_md_separate(spdk_bdev_desc_get_bdev(desc))) {
6069 0 : return -EINVAL;
6070 : }
6071 :
6072 0 : if (md_buf && !_is_buf_allocated(&iov)) {
6073 0 : return -EINVAL;
6074 : }
6075 :
6076 0 : return bdev_compare_blocks_with_md(desc, ch, buf, md_buf, offset_blocks, num_blocks,
6077 0 : cb, cb_arg);
6078 0 : }
6079 :
6080 : static void
6081 2 : bdev_comparev_and_writev_blocks_unlocked(struct lba_range *range, void *ctx, int unlock_status)
6082 : {
6083 2 : struct spdk_bdev_io *bdev_io = ctx;
6084 :
6085 2 : if (unlock_status) {
6086 0 : SPDK_ERRLOG("LBA range unlock failed\n");
6087 0 : }
6088 :
6089 4 : bdev_io->internal.cb(bdev_io, bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS ? true :
6090 2 : false, bdev_io->internal.caller_ctx);
6091 2 : }
6092 :
6093 : static void
6094 2 : bdev_comparev_and_writev_blocks_unlock(struct spdk_bdev_io *bdev_io, int status)
6095 : {
6096 2 : bdev_io->internal.status = status;
6097 :
6098 4 : bdev_unlock_lba_range(bdev_io->internal.desc, spdk_io_channel_from_ctx(bdev_io->internal.ch),
6099 2 : bdev_io->u.bdev.offset_blocks, bdev_io->u.bdev.num_blocks,
6100 2 : bdev_comparev_and_writev_blocks_unlocked, bdev_io);
6101 2 : }
6102 :
6103 : static void
6104 1 : bdev_compare_and_write_do_write_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
6105 : {
6106 1 : struct spdk_bdev_io *parent_io = cb_arg;
6107 :
6108 1 : if (!success) {
6109 0 : SPDK_ERRLOG("Compare and write operation failed\n");
6110 0 : }
6111 :
6112 1 : spdk_bdev_free_io(bdev_io);
6113 :
6114 2 : bdev_comparev_and_writev_blocks_unlock(parent_io,
6115 1 : success ? SPDK_BDEV_IO_STATUS_SUCCESS : SPDK_BDEV_IO_STATUS_FAILED);
6116 1 : }
6117 :
6118 : static void
6119 1 : bdev_compare_and_write_do_write(void *_bdev_io)
6120 : {
6121 1 : struct spdk_bdev_io *bdev_io = _bdev_io;
6122 : int rc;
6123 :
6124 2 : rc = spdk_bdev_writev_blocks(bdev_io->internal.desc,
6125 1 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
6126 1 : bdev_io->u.bdev.fused_iovs, bdev_io->u.bdev.fused_iovcnt,
6127 1 : bdev_io->u.bdev.offset_blocks, bdev_io->u.bdev.num_blocks,
6128 1 : bdev_compare_and_write_do_write_done, bdev_io);
6129 :
6130 :
6131 1 : if (rc == -ENOMEM) {
6132 0 : bdev_queue_io_wait_with_cb(bdev_io, bdev_compare_and_write_do_write);
6133 1 : } else if (rc != 0) {
6134 0 : bdev_comparev_and_writev_blocks_unlock(bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
6135 0 : }
6136 1 : }
6137 :
6138 : static void
6139 2 : bdev_compare_and_write_do_compare_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
6140 : {
6141 2 : struct spdk_bdev_io *parent_io = cb_arg;
6142 :
6143 2 : spdk_bdev_free_io(bdev_io);
6144 :
6145 2 : if (!success) {
6146 1 : bdev_comparev_and_writev_blocks_unlock(parent_io, SPDK_BDEV_IO_STATUS_MISCOMPARE);
6147 1 : return;
6148 : }
6149 :
6150 1 : bdev_compare_and_write_do_write(parent_io);
6151 2 : }
6152 :
6153 : static void
6154 2 : bdev_compare_and_write_do_compare(void *_bdev_io)
6155 : {
6156 2 : struct spdk_bdev_io *bdev_io = _bdev_io;
6157 : int rc;
6158 :
6159 4 : rc = spdk_bdev_comparev_blocks(bdev_io->internal.desc,
6160 2 : spdk_io_channel_from_ctx(bdev_io->internal.ch), bdev_io->u.bdev.iovs,
6161 2 : bdev_io->u.bdev.iovcnt, bdev_io->u.bdev.offset_blocks, bdev_io->u.bdev.num_blocks,
6162 2 : bdev_compare_and_write_do_compare_done, bdev_io);
6163 :
6164 2 : if (rc == -ENOMEM) {
6165 0 : bdev_queue_io_wait_with_cb(bdev_io, bdev_compare_and_write_do_compare);
6166 2 : } else if (rc != 0) {
6167 0 : bdev_comparev_and_writev_blocks_unlock(bdev_io, SPDK_BDEV_IO_STATUS_FIRST_FUSED_FAILED);
6168 0 : }
6169 2 : }
6170 :
6171 : static void
6172 2 : bdev_comparev_and_writev_blocks_locked(struct lba_range *range, void *ctx, int status)
6173 : {
6174 2 : struct spdk_bdev_io *bdev_io = ctx;
6175 :
6176 2 : if (status) {
6177 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FIRST_FUSED_FAILED;
6178 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
6179 0 : return;
6180 : }
6181 :
6182 2 : bdev_compare_and_write_do_compare(bdev_io);
6183 2 : }
6184 :
6185 : int
6186 2 : spdk_bdev_comparev_and_writev_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6187 : struct iovec *compare_iov, int compare_iovcnt,
6188 : struct iovec *write_iov, int write_iovcnt,
6189 : uint64_t offset_blocks, uint64_t num_blocks,
6190 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6191 : {
6192 2 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6193 : struct spdk_bdev_io *bdev_io;
6194 2 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6195 :
6196 2 : if (!desc->write) {
6197 0 : return -EBADF;
6198 : }
6199 :
6200 2 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6201 0 : return -EINVAL;
6202 : }
6203 :
6204 2 : if (num_blocks > bdev->acwu) {
6205 0 : return -EINVAL;
6206 : }
6207 :
6208 2 : bdev_io = bdev_channel_get_io(channel);
6209 2 : if (!bdev_io) {
6210 0 : return -ENOMEM;
6211 : }
6212 :
6213 2 : bdev_io->internal.ch = channel;
6214 2 : bdev_io->internal.desc = desc;
6215 2 : bdev_io->type = SPDK_BDEV_IO_TYPE_COMPARE_AND_WRITE;
6216 2 : bdev_io->u.bdev.iovs = compare_iov;
6217 2 : bdev_io->u.bdev.iovcnt = compare_iovcnt;
6218 2 : bdev_io->u.bdev.fused_iovs = write_iov;
6219 2 : bdev_io->u.bdev.fused_iovcnt = write_iovcnt;
6220 2 : bdev_io->u.bdev.md_buf = NULL;
6221 2 : bdev_io->u.bdev.num_blocks = num_blocks;
6222 2 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6223 2 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6224 2 : bdev_io->u.bdev.memory_domain = NULL;
6225 2 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6226 2 : bdev_io->u.bdev.accel_sequence = NULL;
6227 :
6228 2 : if (bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COMPARE_AND_WRITE)) {
6229 0 : bdev_io_submit(bdev_io);
6230 0 : return 0;
6231 : }
6232 :
6233 4 : return bdev_lock_lba_range(desc, ch, offset_blocks, num_blocks,
6234 2 : bdev_comparev_and_writev_blocks_locked, bdev_io);
6235 2 : }
6236 :
6237 : int
6238 2 : spdk_bdev_zcopy_start(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6239 : struct iovec *iov, int iovcnt,
6240 : uint64_t offset_blocks, uint64_t num_blocks,
6241 : bool populate,
6242 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6243 : {
6244 2 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6245 : struct spdk_bdev_io *bdev_io;
6246 2 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6247 :
6248 2 : if (!desc->write) {
6249 0 : return -EBADF;
6250 : }
6251 :
6252 2 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6253 0 : return -EINVAL;
6254 : }
6255 :
6256 2 : if (!spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_ZCOPY)) {
6257 0 : return -ENOTSUP;
6258 : }
6259 :
6260 2 : bdev_io = bdev_channel_get_io(channel);
6261 2 : if (!bdev_io) {
6262 0 : return -ENOMEM;
6263 : }
6264 :
6265 2 : bdev_io->internal.ch = channel;
6266 2 : bdev_io->internal.desc = desc;
6267 2 : bdev_io->type = SPDK_BDEV_IO_TYPE_ZCOPY;
6268 2 : bdev_io->u.bdev.num_blocks = num_blocks;
6269 2 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6270 2 : bdev_io->u.bdev.iovs = iov;
6271 2 : bdev_io->u.bdev.iovcnt = iovcnt;
6272 2 : bdev_io->u.bdev.md_buf = NULL;
6273 2 : bdev_io->u.bdev.zcopy.populate = populate ? 1 : 0;
6274 2 : bdev_io->u.bdev.zcopy.commit = 0;
6275 2 : bdev_io->u.bdev.zcopy.start = 1;
6276 2 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6277 2 : bdev_io->u.bdev.memory_domain = NULL;
6278 2 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6279 2 : bdev_io->u.bdev.accel_sequence = NULL;
6280 :
6281 2 : bdev_io_submit(bdev_io);
6282 :
6283 2 : return 0;
6284 2 : }
6285 :
6286 : int
6287 2 : spdk_bdev_zcopy_end(struct spdk_bdev_io *bdev_io, bool commit,
6288 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6289 : {
6290 2 : if (bdev_io->type != SPDK_BDEV_IO_TYPE_ZCOPY) {
6291 0 : return -EINVAL;
6292 : }
6293 :
6294 2 : bdev_io->u.bdev.zcopy.commit = commit ? 1 : 0;
6295 2 : bdev_io->u.bdev.zcopy.start = 0;
6296 2 : bdev_io->internal.caller_ctx = cb_arg;
6297 2 : bdev_io->internal.cb = cb;
6298 2 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_PENDING;
6299 :
6300 2 : bdev_io_submit(bdev_io);
6301 :
6302 2 : return 0;
6303 2 : }
6304 :
6305 : int
6306 0 : spdk_bdev_write_zeroes(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6307 : uint64_t offset, uint64_t len,
6308 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6309 : {
6310 : uint64_t offset_blocks, num_blocks;
6311 :
6312 0 : if (bdev_bytes_to_blocks(spdk_bdev_desc_get_bdev(desc), offset, &offset_blocks,
6313 0 : len, &num_blocks) != 0) {
6314 0 : return -EINVAL;
6315 : }
6316 :
6317 0 : return spdk_bdev_write_zeroes_blocks(desc, ch, offset_blocks, num_blocks, cb, cb_arg);
6318 0 : }
6319 :
6320 : int
6321 33 : spdk_bdev_write_zeroes_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6322 : uint64_t offset_blocks, uint64_t num_blocks,
6323 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6324 : {
6325 33 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6326 : struct spdk_bdev_io *bdev_io;
6327 33 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6328 :
6329 33 : if (!desc->write) {
6330 0 : return -EBADF;
6331 : }
6332 :
6333 33 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6334 0 : return -EINVAL;
6335 : }
6336 :
6337 33 : if (!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE_ZEROES) &&
6338 10 : !bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE)) {
6339 1 : return -ENOTSUP;
6340 : }
6341 :
6342 32 : bdev_io = bdev_channel_get_io(channel);
6343 :
6344 32 : if (!bdev_io) {
6345 0 : return -ENOMEM;
6346 : }
6347 :
6348 32 : bdev_io->type = SPDK_BDEV_IO_TYPE_WRITE_ZEROES;
6349 32 : bdev_io->internal.ch = channel;
6350 32 : bdev_io->internal.desc = desc;
6351 32 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6352 32 : bdev_io->u.bdev.num_blocks = num_blocks;
6353 32 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6354 32 : bdev_io->u.bdev.memory_domain = NULL;
6355 32 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6356 32 : bdev_io->u.bdev.accel_sequence = NULL;
6357 :
6358 : /* If the write_zeroes size is large and should be split, use the generic split
6359 : * logic regardless of whether SPDK_BDEV_IO_TYPE_WRITE_ZEREOS is supported or not.
6360 : *
6361 : * Then, send the write_zeroes request if SPDK_BDEV_IO_TYPE_WRITE_ZEROES is supported
6362 : * or emulate it using regular write request otherwise.
6363 : */
6364 32 : if (bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE_ZEROES) ||
6365 9 : bdev_io->internal.f.split) {
6366 26 : bdev_io_submit(bdev_io);
6367 26 : return 0;
6368 : }
6369 :
6370 6 : assert(_bdev_get_block_size_with_md(bdev) <= ZERO_BUFFER_SIZE);
6371 :
6372 6 : return bdev_write_zero_buffer(bdev_io);
6373 33 : }
6374 :
6375 : int
6376 0 : spdk_bdev_unmap(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6377 : uint64_t offset, uint64_t nbytes,
6378 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6379 : {
6380 : uint64_t offset_blocks, num_blocks;
6381 :
6382 0 : if (bdev_bytes_to_blocks(spdk_bdev_desc_get_bdev(desc), offset, &offset_blocks,
6383 0 : nbytes, &num_blocks) != 0) {
6384 0 : return -EINVAL;
6385 : }
6386 :
6387 0 : return spdk_bdev_unmap_blocks(desc, ch, offset_blocks, num_blocks, cb, cb_arg);
6388 0 : }
6389 :
6390 : static void
6391 0 : bdev_io_complete_cb(void *ctx)
6392 : {
6393 0 : struct spdk_bdev_io *bdev_io = ctx;
6394 :
6395 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
6396 0 : bdev_io->internal.cb(bdev_io, true, bdev_io->internal.caller_ctx);
6397 0 : }
6398 :
6399 : int
6400 22 : spdk_bdev_unmap_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6401 : uint64_t offset_blocks, uint64_t num_blocks,
6402 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6403 : {
6404 22 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6405 : struct spdk_bdev_io *bdev_io;
6406 22 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6407 :
6408 22 : if (!desc->write) {
6409 0 : return -EBADF;
6410 : }
6411 :
6412 22 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6413 0 : return -EINVAL;
6414 : }
6415 :
6416 22 : bdev_io = bdev_channel_get_io(channel);
6417 22 : if (!bdev_io) {
6418 0 : return -ENOMEM;
6419 : }
6420 :
6421 22 : bdev_io->internal.ch = channel;
6422 22 : bdev_io->internal.desc = desc;
6423 22 : bdev_io->type = SPDK_BDEV_IO_TYPE_UNMAP;
6424 :
6425 22 : bdev_io->u.bdev.iovs = &bdev_io->iov;
6426 22 : bdev_io->u.bdev.iovs[0].iov_base = NULL;
6427 22 : bdev_io->u.bdev.iovs[0].iov_len = 0;
6428 22 : bdev_io->u.bdev.iovcnt = 1;
6429 :
6430 22 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6431 22 : bdev_io->u.bdev.num_blocks = num_blocks;
6432 22 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6433 22 : bdev_io->u.bdev.memory_domain = NULL;
6434 22 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6435 22 : bdev_io->u.bdev.accel_sequence = NULL;
6436 :
6437 22 : if (num_blocks == 0) {
6438 0 : spdk_thread_send_msg(spdk_get_thread(), bdev_io_complete_cb, bdev_io);
6439 0 : return 0;
6440 : }
6441 :
6442 22 : bdev_io_submit(bdev_io);
6443 22 : return 0;
6444 22 : }
6445 :
6446 : int
6447 0 : spdk_bdev_flush(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6448 : uint64_t offset, uint64_t length,
6449 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6450 : {
6451 : uint64_t offset_blocks, num_blocks;
6452 :
6453 0 : if (bdev_bytes_to_blocks(spdk_bdev_desc_get_bdev(desc), offset, &offset_blocks,
6454 0 : length, &num_blocks) != 0) {
6455 0 : return -EINVAL;
6456 : }
6457 :
6458 0 : return spdk_bdev_flush_blocks(desc, ch, offset_blocks, num_blocks, cb, cb_arg);
6459 0 : }
6460 :
6461 : int
6462 2 : spdk_bdev_flush_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6463 : uint64_t offset_blocks, uint64_t num_blocks,
6464 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6465 : {
6466 2 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6467 : struct spdk_bdev_io *bdev_io;
6468 2 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6469 :
6470 2 : if (!desc->write) {
6471 0 : return -EBADF;
6472 : }
6473 :
6474 2 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6475 0 : return -EINVAL;
6476 : }
6477 :
6478 2 : bdev_io = bdev_channel_get_io(channel);
6479 2 : if (!bdev_io) {
6480 0 : return -ENOMEM;
6481 : }
6482 :
6483 2 : bdev_io->internal.ch = channel;
6484 2 : bdev_io->internal.desc = desc;
6485 2 : bdev_io->type = SPDK_BDEV_IO_TYPE_FLUSH;
6486 2 : bdev_io->u.bdev.iovs = NULL;
6487 2 : bdev_io->u.bdev.iovcnt = 0;
6488 2 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6489 2 : bdev_io->u.bdev.num_blocks = num_blocks;
6490 2 : bdev_io->u.bdev.memory_domain = NULL;
6491 2 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6492 2 : bdev_io->u.bdev.accel_sequence = NULL;
6493 2 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6494 :
6495 2 : bdev_io_submit(bdev_io);
6496 2 : return 0;
6497 2 : }
6498 :
6499 : static int bdev_reset_poll_for_outstanding_io(void *ctx);
6500 :
6501 : static void
6502 13 : bdev_reset_check_outstanding_io_done(struct spdk_bdev *bdev, void *_ctx, int status)
6503 : {
6504 13 : struct spdk_bdev_channel *ch = _ctx;
6505 : struct spdk_bdev_io *bdev_io;
6506 :
6507 13 : bdev_io = TAILQ_FIRST(&ch->queued_resets);
6508 :
6509 13 : if (status == -EBUSY) {
6510 9 : if (spdk_get_ticks() < bdev_io->u.reset.wait_poller.stop_time_tsc) {
6511 8 : bdev_io->u.reset.wait_poller.poller = SPDK_POLLER_REGISTER(bdev_reset_poll_for_outstanding_io,
6512 : ch, BDEV_RESET_CHECK_OUTSTANDING_IO_PERIOD);
6513 8 : } else {
6514 1 : TAILQ_REMOVE(&ch->queued_resets, bdev_io, internal.link);
6515 :
6516 1 : if (TAILQ_EMPTY(&ch->io_memory_domain) && TAILQ_EMPTY(&ch->io_accel_exec)) {
6517 : /* If outstanding IOs are still present and reset_io_drain_timeout
6518 : * seconds passed, start the reset. */
6519 1 : bdev_io_submit_reset(bdev_io);
6520 1 : } else {
6521 : /* We still have in progress memory domain pull/push or we're
6522 : * executing accel sequence. Since we cannot abort either of those
6523 : * operations, fail the reset request. */
6524 0 : spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
6525 : }
6526 : }
6527 9 : } else {
6528 4 : TAILQ_REMOVE(&ch->queued_resets, bdev_io, internal.link);
6529 4 : SPDK_DEBUGLOG(bdev,
6530 : "Skipping reset for underlying device of bdev: %s - no outstanding I/O.\n",
6531 : ch->bdev->name);
6532 : /* Mark the completion status as a SUCCESS and complete the reset. */
6533 4 : spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_SUCCESS);
6534 : }
6535 13 : }
6536 :
6537 : static void
6538 13 : bdev_reset_check_outstanding_io(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
6539 : struct spdk_io_channel *io_ch, void *_ctx)
6540 : {
6541 13 : struct spdk_bdev_channel *cur_ch = __io_ch_to_bdev_ch(io_ch);
6542 13 : int status = 0;
6543 :
6544 17 : if (cur_ch->io_outstanding > 0 ||
6545 4 : !TAILQ_EMPTY(&cur_ch->io_memory_domain) ||
6546 4 : !TAILQ_EMPTY(&cur_ch->io_accel_exec)) {
6547 : /* If a channel has outstanding IO, set status to -EBUSY code. This will stop
6548 : * further iteration over the rest of the channels and pass non-zero status
6549 : * to the callback function. */
6550 9 : status = -EBUSY;
6551 9 : }
6552 13 : spdk_bdev_for_each_channel_continue(i, status);
6553 13 : }
6554 :
6555 : static int
6556 8 : bdev_reset_poll_for_outstanding_io(void *ctx)
6557 : {
6558 8 : struct spdk_bdev_channel *ch = ctx;
6559 : struct spdk_bdev_io *bdev_io;
6560 :
6561 8 : bdev_io = TAILQ_FIRST(&ch->queued_resets);
6562 :
6563 8 : spdk_poller_unregister(&bdev_io->u.reset.wait_poller.poller);
6564 8 : spdk_bdev_for_each_channel(ch->bdev, bdev_reset_check_outstanding_io, ch,
6565 : bdev_reset_check_outstanding_io_done);
6566 :
6567 8 : return SPDK_POLLER_BUSY;
6568 : }
6569 :
6570 : static void
6571 15 : bdev_reset_freeze_channel_done(struct spdk_bdev *bdev, void *_ctx, int status)
6572 : {
6573 15 : struct spdk_bdev_channel *ch = _ctx;
6574 : struct spdk_bdev_io *bdev_io;
6575 :
6576 15 : bdev_io = TAILQ_FIRST(&ch->queued_resets);
6577 :
6578 15 : if (bdev->reset_io_drain_timeout == 0) {
6579 10 : TAILQ_REMOVE(&ch->queued_resets, bdev_io, internal.link);
6580 :
6581 10 : bdev_io_submit_reset(bdev_io);
6582 10 : return;
6583 : }
6584 :
6585 10 : bdev_io->u.reset.wait_poller.stop_time_tsc = spdk_get_ticks() +
6586 5 : (ch->bdev->reset_io_drain_timeout * spdk_get_ticks_hz());
6587 :
6588 : /* In case bdev->reset_io_drain_timeout is not equal to zero,
6589 : * submit the reset to the underlying module only if outstanding I/O
6590 : * remain after reset_io_drain_timeout seconds have passed. */
6591 5 : spdk_bdev_for_each_channel(ch->bdev, bdev_reset_check_outstanding_io, ch,
6592 : bdev_reset_check_outstanding_io_done);
6593 15 : }
6594 :
6595 : static void
6596 18 : bdev_reset_freeze_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
6597 : struct spdk_io_channel *ch, void *_ctx)
6598 : {
6599 : struct spdk_bdev_channel *channel;
6600 : struct spdk_bdev_mgmt_channel *mgmt_channel;
6601 : struct spdk_bdev_shared_resource *shared_resource;
6602 : bdev_io_tailq_t tmp_queued;
6603 :
6604 18 : TAILQ_INIT(&tmp_queued);
6605 :
6606 18 : channel = __io_ch_to_bdev_ch(ch);
6607 18 : shared_resource = channel->shared_resource;
6608 18 : mgmt_channel = shared_resource->mgmt_ch;
6609 :
6610 18 : channel->flags |= BDEV_CH_RESET_IN_PROGRESS;
6611 :
6612 18 : if ((channel->flags & BDEV_CH_QOS_ENABLED) != 0) {
6613 2 : TAILQ_SWAP(&channel->qos_queued_io, &tmp_queued, spdk_bdev_io, internal.link);
6614 2 : }
6615 :
6616 18 : bdev_abort_all_queued_io(&shared_resource->nomem_io, channel);
6617 18 : bdev_abort_all_buf_io(mgmt_channel, channel);
6618 18 : bdev_abort_all_queued_io(&tmp_queued, channel);
6619 :
6620 18 : spdk_bdev_for_each_channel_continue(i, 0);
6621 18 : }
6622 :
6623 : static void
6624 15 : bdev_start_reset(void *ctx)
6625 : {
6626 15 : struct spdk_bdev_channel *ch = ctx;
6627 :
6628 15 : spdk_bdev_for_each_channel(ch->bdev, bdev_reset_freeze_channel, ch,
6629 : bdev_reset_freeze_channel_done);
6630 15 : }
6631 :
6632 : static void
6633 16 : bdev_channel_start_reset(struct spdk_bdev_channel *ch)
6634 : {
6635 16 : struct spdk_bdev *bdev = ch->bdev;
6636 :
6637 16 : assert(!TAILQ_EMPTY(&ch->queued_resets));
6638 :
6639 16 : spdk_spin_lock(&bdev->internal.spinlock);
6640 16 : if (bdev->internal.reset_in_progress == NULL) {
6641 15 : bdev->internal.reset_in_progress = TAILQ_FIRST(&ch->queued_resets);
6642 : /*
6643 : * Take a channel reference for the target bdev for the life of this
6644 : * reset. This guards against the channel getting destroyed while
6645 : * spdk_bdev_for_each_channel() calls related to this reset IO are in
6646 : * progress. We will release the reference when this reset is
6647 : * completed.
6648 : */
6649 15 : bdev->internal.reset_in_progress->u.reset.ch_ref = spdk_get_io_channel(__bdev_to_io_dev(bdev));
6650 15 : bdev_start_reset(ch);
6651 15 : }
6652 16 : spdk_spin_unlock(&bdev->internal.spinlock);
6653 16 : }
6654 :
6655 : int
6656 16 : spdk_bdev_reset(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6657 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6658 : {
6659 16 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6660 : struct spdk_bdev_io *bdev_io;
6661 16 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6662 :
6663 16 : bdev_io = bdev_channel_get_io(channel);
6664 16 : if (!bdev_io) {
6665 0 : return -ENOMEM;
6666 : }
6667 :
6668 16 : bdev_io->internal.ch = channel;
6669 16 : bdev_io->internal.desc = desc;
6670 16 : bdev_io->internal.submit_tsc = spdk_get_ticks();
6671 16 : bdev_io->type = SPDK_BDEV_IO_TYPE_RESET;
6672 16 : bdev_io->u.reset.ch_ref = NULL;
6673 16 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6674 :
6675 16 : spdk_spin_lock(&bdev->internal.spinlock);
6676 16 : TAILQ_INSERT_TAIL(&channel->queued_resets, bdev_io, internal.link);
6677 16 : spdk_spin_unlock(&bdev->internal.spinlock);
6678 :
6679 16 : bdev_ch_add_to_io_submitted(bdev_io);
6680 :
6681 16 : bdev_channel_start_reset(channel);
6682 :
6683 16 : return 0;
6684 16 : }
6685 :
6686 : void
6687 0 : spdk_bdev_get_io_stat(struct spdk_bdev *bdev, struct spdk_io_channel *ch,
6688 : struct spdk_bdev_io_stat *stat, enum spdk_bdev_reset_stat_mode reset_mode)
6689 : {
6690 0 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6691 :
6692 0 : bdev_get_io_stat(stat, channel->stat);
6693 0 : spdk_bdev_reset_io_stat(stat, reset_mode);
6694 0 : }
6695 :
6696 : static void
6697 5 : bdev_get_device_stat_done(struct spdk_bdev *bdev, void *_ctx, int status)
6698 : {
6699 5 : struct spdk_bdev_iostat_ctx *bdev_iostat_ctx = _ctx;
6700 :
6701 10 : bdev_iostat_ctx->cb(bdev, bdev_iostat_ctx->stat,
6702 5 : bdev_iostat_ctx->cb_arg, 0);
6703 5 : free(bdev_iostat_ctx);
6704 5 : }
6705 :
6706 : static void
6707 4 : bdev_get_each_channel_stat(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
6708 : struct spdk_io_channel *ch, void *_ctx)
6709 : {
6710 4 : struct spdk_bdev_iostat_ctx *bdev_iostat_ctx = _ctx;
6711 4 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6712 :
6713 4 : spdk_bdev_add_io_stat(bdev_iostat_ctx->stat, channel->stat);
6714 4 : spdk_bdev_reset_io_stat(channel->stat, bdev_iostat_ctx->reset_mode);
6715 4 : spdk_bdev_for_each_channel_continue(i, 0);
6716 4 : }
6717 :
6718 : void
6719 5 : spdk_bdev_get_device_stat(struct spdk_bdev *bdev, struct spdk_bdev_io_stat *stat,
6720 : enum spdk_bdev_reset_stat_mode reset_mode, spdk_bdev_get_device_stat_cb cb, void *cb_arg)
6721 : {
6722 : struct spdk_bdev_iostat_ctx *bdev_iostat_ctx;
6723 :
6724 5 : assert(bdev != NULL);
6725 5 : assert(stat != NULL);
6726 5 : assert(cb != NULL);
6727 :
6728 5 : bdev_iostat_ctx = calloc(1, sizeof(struct spdk_bdev_iostat_ctx));
6729 5 : if (bdev_iostat_ctx == NULL) {
6730 0 : SPDK_ERRLOG("Unable to allocate memory for spdk_bdev_iostat_ctx\n");
6731 0 : cb(bdev, stat, cb_arg, -ENOMEM);
6732 0 : return;
6733 : }
6734 :
6735 5 : bdev_iostat_ctx->stat = stat;
6736 5 : bdev_iostat_ctx->cb = cb;
6737 5 : bdev_iostat_ctx->cb_arg = cb_arg;
6738 5 : bdev_iostat_ctx->reset_mode = reset_mode;
6739 :
6740 : /* Start with the statistics from previously deleted channels. */
6741 5 : spdk_spin_lock(&bdev->internal.spinlock);
6742 5 : bdev_get_io_stat(bdev_iostat_ctx->stat, bdev->internal.stat);
6743 5 : spdk_bdev_reset_io_stat(bdev->internal.stat, reset_mode);
6744 5 : spdk_spin_unlock(&bdev->internal.spinlock);
6745 :
6746 : /* Then iterate and add the statistics from each existing channel. */
6747 5 : spdk_bdev_for_each_channel(bdev, bdev_get_each_channel_stat, bdev_iostat_ctx,
6748 : bdev_get_device_stat_done);
6749 5 : }
6750 :
6751 : struct bdev_iostat_reset_ctx {
6752 : enum spdk_bdev_reset_stat_mode mode;
6753 : bdev_reset_device_stat_cb cb;
6754 : void *cb_arg;
6755 : };
6756 :
6757 : static void
6758 0 : bdev_reset_device_stat_done(struct spdk_bdev *bdev, void *_ctx, int status)
6759 : {
6760 0 : struct bdev_iostat_reset_ctx *ctx = _ctx;
6761 :
6762 0 : ctx->cb(bdev, ctx->cb_arg, 0);
6763 :
6764 0 : free(ctx);
6765 0 : }
6766 :
6767 : static void
6768 0 : bdev_reset_each_channel_stat(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
6769 : struct spdk_io_channel *ch, void *_ctx)
6770 : {
6771 0 : struct bdev_iostat_reset_ctx *ctx = _ctx;
6772 0 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6773 :
6774 0 : spdk_bdev_reset_io_stat(channel->stat, ctx->mode);
6775 :
6776 0 : spdk_bdev_for_each_channel_continue(i, 0);
6777 0 : }
6778 :
6779 : void
6780 0 : bdev_reset_device_stat(struct spdk_bdev *bdev, enum spdk_bdev_reset_stat_mode mode,
6781 : bdev_reset_device_stat_cb cb, void *cb_arg)
6782 : {
6783 : struct bdev_iostat_reset_ctx *ctx;
6784 :
6785 0 : assert(bdev != NULL);
6786 0 : assert(cb != NULL);
6787 :
6788 0 : ctx = calloc(1, sizeof(*ctx));
6789 0 : if (ctx == NULL) {
6790 0 : SPDK_ERRLOG("Unable to allocate bdev_iostat_reset_ctx.\n");
6791 0 : cb(bdev, cb_arg, -ENOMEM);
6792 0 : return;
6793 : }
6794 :
6795 0 : ctx->mode = mode;
6796 0 : ctx->cb = cb;
6797 0 : ctx->cb_arg = cb_arg;
6798 :
6799 0 : spdk_spin_lock(&bdev->internal.spinlock);
6800 0 : spdk_bdev_reset_io_stat(bdev->internal.stat, mode);
6801 0 : spdk_spin_unlock(&bdev->internal.spinlock);
6802 :
6803 0 : spdk_bdev_for_each_channel(bdev,
6804 : bdev_reset_each_channel_stat,
6805 0 : ctx,
6806 : bdev_reset_device_stat_done);
6807 0 : }
6808 :
6809 : int
6810 1 : spdk_bdev_nvme_admin_passthru(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6811 : const struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes,
6812 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6813 : {
6814 1 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6815 : struct spdk_bdev_io *bdev_io;
6816 1 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6817 :
6818 1 : if (!desc->write) {
6819 0 : return -EBADF;
6820 : }
6821 :
6822 1 : if (spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_ADMIN))) {
6823 1 : return -ENOTSUP;
6824 : }
6825 :
6826 0 : bdev_io = bdev_channel_get_io(channel);
6827 0 : if (!bdev_io) {
6828 0 : return -ENOMEM;
6829 : }
6830 :
6831 0 : bdev_io->internal.ch = channel;
6832 0 : bdev_io->internal.desc = desc;
6833 0 : bdev_io->type = SPDK_BDEV_IO_TYPE_NVME_ADMIN;
6834 0 : bdev_io->u.nvme_passthru.cmd = *cmd;
6835 0 : bdev_io->u.nvme_passthru.buf = buf;
6836 0 : bdev_io->u.nvme_passthru.nbytes = nbytes;
6837 0 : bdev_io->u.nvme_passthru.md_buf = NULL;
6838 0 : bdev_io->u.nvme_passthru.md_len = 0;
6839 :
6840 0 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6841 :
6842 0 : bdev_io_submit(bdev_io);
6843 0 : return 0;
6844 1 : }
6845 :
6846 : int
6847 1 : spdk_bdev_nvme_io_passthru(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6848 : const struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes,
6849 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6850 : {
6851 1 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6852 : struct spdk_bdev_io *bdev_io;
6853 1 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6854 :
6855 1 : if (!desc->write) {
6856 : /*
6857 : * Do not try to parse the NVMe command - we could maybe use bits in the opcode
6858 : * to easily determine if the command is a read or write, but for now just
6859 : * do not allow io_passthru with a read-only descriptor.
6860 : */
6861 0 : return -EBADF;
6862 : }
6863 :
6864 1 : if (spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_IO))) {
6865 1 : return -ENOTSUP;
6866 : }
6867 :
6868 0 : bdev_io = bdev_channel_get_io(channel);
6869 0 : if (!bdev_io) {
6870 0 : return -ENOMEM;
6871 : }
6872 :
6873 0 : bdev_io->internal.ch = channel;
6874 0 : bdev_io->internal.desc = desc;
6875 0 : bdev_io->type = SPDK_BDEV_IO_TYPE_NVME_IO;
6876 0 : bdev_io->u.nvme_passthru.cmd = *cmd;
6877 0 : bdev_io->u.nvme_passthru.buf = buf;
6878 0 : bdev_io->u.nvme_passthru.nbytes = nbytes;
6879 0 : bdev_io->u.nvme_passthru.md_buf = NULL;
6880 0 : bdev_io->u.nvme_passthru.md_len = 0;
6881 :
6882 0 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6883 :
6884 0 : bdev_io_submit(bdev_io);
6885 0 : return 0;
6886 1 : }
6887 :
6888 : int
6889 1 : spdk_bdev_nvme_io_passthru_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6890 : const struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes, void *md_buf, size_t md_len,
6891 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6892 : {
6893 1 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6894 : struct spdk_bdev_io *bdev_io;
6895 1 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6896 :
6897 1 : if (!desc->write) {
6898 : /*
6899 : * Do not try to parse the NVMe command - we could maybe use bits in the opcode
6900 : * to easily determine if the command is a read or write, but for now just
6901 : * do not allow io_passthru with a read-only descriptor.
6902 : */
6903 0 : return -EBADF;
6904 : }
6905 :
6906 1 : if (spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_IO_MD))) {
6907 1 : return -ENOTSUP;
6908 : }
6909 :
6910 0 : bdev_io = bdev_channel_get_io(channel);
6911 0 : if (!bdev_io) {
6912 0 : return -ENOMEM;
6913 : }
6914 :
6915 0 : bdev_io->internal.ch = channel;
6916 0 : bdev_io->internal.desc = desc;
6917 0 : bdev_io->type = SPDK_BDEV_IO_TYPE_NVME_IO_MD;
6918 0 : bdev_io->u.nvme_passthru.cmd = *cmd;
6919 0 : bdev_io->u.nvme_passthru.buf = buf;
6920 0 : bdev_io->u.nvme_passthru.nbytes = nbytes;
6921 0 : bdev_io->u.nvme_passthru.md_buf = md_buf;
6922 0 : bdev_io->u.nvme_passthru.md_len = md_len;
6923 :
6924 0 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6925 :
6926 0 : bdev_io_submit(bdev_io);
6927 0 : return 0;
6928 1 : }
6929 :
6930 : int
6931 0 : spdk_bdev_nvme_iov_passthru_md(struct spdk_bdev_desc *desc,
6932 : struct spdk_io_channel *ch,
6933 : const struct spdk_nvme_cmd *cmd,
6934 : struct iovec *iov, int iovcnt, size_t nbytes,
6935 : void *md_buf, size_t md_len,
6936 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6937 : {
6938 0 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6939 : struct spdk_bdev_io *bdev_io;
6940 0 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6941 :
6942 0 : if (!desc->write) {
6943 : /*
6944 : * Do not try to parse the NVMe command - we could maybe use bits in the opcode
6945 : * to easily determine if the command is a read or write, but for now just
6946 : * do not allow io_passthru with a read-only descriptor.
6947 : */
6948 0 : return -EBADF;
6949 : }
6950 :
6951 0 : if (md_buf && spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_IO_MD))) {
6952 0 : return -ENOTSUP;
6953 0 : } else if (spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_IO))) {
6954 0 : return -ENOTSUP;
6955 : }
6956 :
6957 0 : bdev_io = bdev_channel_get_io(channel);
6958 0 : if (!bdev_io) {
6959 0 : return -ENOMEM;
6960 : }
6961 :
6962 0 : bdev_io->internal.ch = channel;
6963 0 : bdev_io->internal.desc = desc;
6964 0 : bdev_io->type = SPDK_BDEV_IO_TYPE_NVME_IOV_MD;
6965 0 : bdev_io->u.nvme_passthru.cmd = *cmd;
6966 0 : bdev_io->u.nvme_passthru.iovs = iov;
6967 0 : bdev_io->u.nvme_passthru.iovcnt = iovcnt;
6968 0 : bdev_io->u.nvme_passthru.nbytes = nbytes;
6969 0 : bdev_io->u.nvme_passthru.md_buf = md_buf;
6970 0 : bdev_io->u.nvme_passthru.md_len = md_len;
6971 :
6972 0 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6973 :
6974 0 : bdev_io_submit(bdev_io);
6975 0 : return 0;
6976 0 : }
6977 :
6978 : static void bdev_abort_retry(void *ctx);
6979 : static void bdev_abort(struct spdk_bdev_io *parent_io);
6980 :
6981 : static void
6982 22 : bdev_abort_io_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
6983 : {
6984 22 : struct spdk_bdev_channel *channel = bdev_io->internal.ch;
6985 22 : struct spdk_bdev_io *parent_io = cb_arg;
6986 : struct spdk_bdev_io *bio_to_abort, *tmp_io;
6987 :
6988 22 : bio_to_abort = bdev_io->u.abort.bio_to_abort;
6989 :
6990 22 : spdk_bdev_free_io(bdev_io);
6991 :
6992 22 : if (!success) {
6993 : /* Check if the target I/O completed in the meantime. */
6994 2 : TAILQ_FOREACH(tmp_io, &channel->io_submitted, internal.ch_link) {
6995 1 : if (tmp_io == bio_to_abort) {
6996 0 : break;
6997 : }
6998 1 : }
6999 :
7000 : /* If the target I/O still exists, set the parent to failed. */
7001 1 : if (tmp_io != NULL) {
7002 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
7003 0 : }
7004 1 : }
7005 :
7006 22 : assert(parent_io->internal.f.split);
7007 :
7008 22 : parent_io->internal.split.outstanding--;
7009 22 : if (parent_io->internal.split.outstanding == 0) {
7010 16 : if (parent_io->internal.status == SPDK_BDEV_IO_STATUS_NOMEM) {
7011 0 : bdev_abort_retry(parent_io);
7012 0 : } else {
7013 16 : bdev_io_complete(parent_io);
7014 : }
7015 16 : }
7016 22 : }
7017 :
7018 : static int
7019 23 : bdev_abort_io(struct spdk_bdev_desc *desc, struct spdk_bdev_channel *channel,
7020 : struct spdk_bdev_io *bio_to_abort,
7021 : spdk_bdev_io_completion_cb cb, void *cb_arg)
7022 : {
7023 23 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
7024 : struct spdk_bdev_io *bdev_io;
7025 :
7026 23 : if (bio_to_abort->type == SPDK_BDEV_IO_TYPE_ABORT ||
7027 23 : bio_to_abort->type == SPDK_BDEV_IO_TYPE_RESET) {
7028 : /* TODO: Abort reset or abort request. */
7029 0 : return -ENOTSUP;
7030 : }
7031 :
7032 23 : bdev_io = bdev_channel_get_io(channel);
7033 23 : if (bdev_io == NULL) {
7034 1 : return -ENOMEM;
7035 : }
7036 :
7037 22 : bdev_io->internal.ch = channel;
7038 22 : bdev_io->internal.desc = desc;
7039 22 : bdev_io->type = SPDK_BDEV_IO_TYPE_ABORT;
7040 22 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
7041 :
7042 22 : if (bio_to_abort->internal.f.split) {
7043 6 : assert(bdev_io_should_split(bio_to_abort));
7044 6 : bdev_io->u.bdev.abort.bio_cb_arg = bio_to_abort;
7045 :
7046 : /* Parent abort request is not submitted directly, but to manage its
7047 : * execution add it to the submitted list here.
7048 : */
7049 6 : bdev_io->internal.submit_tsc = spdk_get_ticks();
7050 6 : bdev_ch_add_to_io_submitted(bdev_io);
7051 :
7052 6 : bdev_abort(bdev_io);
7053 :
7054 6 : return 0;
7055 : }
7056 :
7057 16 : bdev_io->u.abort.bio_to_abort = bio_to_abort;
7058 :
7059 : /* Submit the abort request to the underlying bdev module. */
7060 16 : bdev_io_submit(bdev_io);
7061 :
7062 16 : return 0;
7063 23 : }
7064 :
7065 : static bool
7066 46 : bdev_io_on_tailq(struct spdk_bdev_io *bdev_io, bdev_io_tailq_t *tailq)
7067 : {
7068 : struct spdk_bdev_io *iter;
7069 :
7070 46 : TAILQ_FOREACH(iter, tailq, internal.link) {
7071 0 : if (iter == bdev_io) {
7072 0 : return true;
7073 : }
7074 0 : }
7075 :
7076 46 : return false;
7077 46 : }
7078 :
7079 : static uint32_t
7080 18 : _bdev_abort(struct spdk_bdev_io *parent_io)
7081 : {
7082 18 : struct spdk_bdev_desc *desc = parent_io->internal.desc;
7083 18 : struct spdk_bdev_channel *channel = parent_io->internal.ch;
7084 : void *bio_cb_arg;
7085 : struct spdk_bdev_io *bio_to_abort;
7086 : uint32_t matched_ios;
7087 : int rc;
7088 :
7089 18 : bio_cb_arg = parent_io->u.bdev.abort.bio_cb_arg;
7090 :
7091 : /* matched_ios is returned and will be kept by the caller.
7092 : *
7093 : * This function will be used for two cases, 1) the same cb_arg is used for
7094 : * multiple I/Os, 2) a single large I/O is split into smaller ones.
7095 : * Incrementing split_outstanding directly here may confuse readers especially
7096 : * for the 1st case.
7097 : *
7098 : * Completion of I/O abort is processed after stack unwinding. Hence this trick
7099 : * works as expected.
7100 : */
7101 18 : matched_ios = 0;
7102 18 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
7103 :
7104 105 : TAILQ_FOREACH(bio_to_abort, &channel->io_submitted, internal.ch_link) {
7105 88 : if (bio_to_abort->internal.caller_ctx != bio_cb_arg) {
7106 65 : continue;
7107 : }
7108 :
7109 23 : if (bio_to_abort->internal.submit_tsc > parent_io->internal.submit_tsc) {
7110 : /* Any I/O which was submitted after this abort command should be excluded. */
7111 0 : continue;
7112 : }
7113 :
7114 : /* We can't abort a request that's being pushed/pulled or executed by accel */
7115 23 : if (bdev_io_on_tailq(bio_to_abort, &channel->io_accel_exec) ||
7116 23 : bdev_io_on_tailq(bio_to_abort, &channel->io_memory_domain)) {
7117 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
7118 0 : break;
7119 : }
7120 :
7121 23 : rc = bdev_abort_io(desc, channel, bio_to_abort, bdev_abort_io_done, parent_io);
7122 23 : if (rc != 0) {
7123 1 : if (rc == -ENOMEM) {
7124 1 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_NOMEM;
7125 1 : } else {
7126 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
7127 : }
7128 1 : break;
7129 : }
7130 22 : matched_ios++;
7131 22 : }
7132 :
7133 18 : return matched_ios;
7134 : }
7135 :
7136 : static void
7137 1 : bdev_abort_retry(void *ctx)
7138 : {
7139 1 : struct spdk_bdev_io *parent_io = ctx;
7140 : uint32_t matched_ios;
7141 :
7142 1 : matched_ios = _bdev_abort(parent_io);
7143 :
7144 1 : if (matched_ios == 0) {
7145 0 : if (parent_io->internal.status == SPDK_BDEV_IO_STATUS_NOMEM) {
7146 0 : bdev_queue_io_wait_with_cb(parent_io, bdev_abort_retry);
7147 0 : } else {
7148 : /* For retry, the case that no target I/O was found is success
7149 : * because it means target I/Os completed in the meantime.
7150 : */
7151 0 : bdev_io_complete(parent_io);
7152 : }
7153 0 : return;
7154 : }
7155 :
7156 : /* Use split_outstanding to manage the progress of aborting I/Os. */
7157 1 : parent_io->internal.f.split = true;
7158 1 : parent_io->internal.split.outstanding = matched_ios;
7159 1 : }
7160 :
7161 : static void
7162 17 : bdev_abort(struct spdk_bdev_io *parent_io)
7163 : {
7164 : uint32_t matched_ios;
7165 :
7166 17 : matched_ios = _bdev_abort(parent_io);
7167 :
7168 17 : if (matched_ios == 0) {
7169 2 : if (parent_io->internal.status == SPDK_BDEV_IO_STATUS_NOMEM) {
7170 1 : bdev_queue_io_wait_with_cb(parent_io, bdev_abort_retry);
7171 1 : } else {
7172 : /* The case the no target I/O was found is failure. */
7173 1 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
7174 1 : bdev_io_complete(parent_io);
7175 : }
7176 2 : return;
7177 : }
7178 :
7179 : /* Use split_outstanding to manage the progress of aborting I/Os. */
7180 15 : parent_io->internal.f.split = true;
7181 15 : parent_io->internal.split.outstanding = matched_ios;
7182 17 : }
7183 :
7184 : int
7185 12 : spdk_bdev_abort(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
7186 : void *bio_cb_arg,
7187 : spdk_bdev_io_completion_cb cb, void *cb_arg)
7188 : {
7189 12 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
7190 12 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
7191 : struct spdk_bdev_io *bdev_io;
7192 :
7193 12 : if (bio_cb_arg == NULL) {
7194 0 : return -EINVAL;
7195 : }
7196 :
7197 12 : if (!spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_ABORT)) {
7198 1 : return -ENOTSUP;
7199 : }
7200 :
7201 11 : bdev_io = bdev_channel_get_io(channel);
7202 11 : if (bdev_io == NULL) {
7203 0 : return -ENOMEM;
7204 : }
7205 :
7206 11 : bdev_io->internal.ch = channel;
7207 11 : bdev_io->internal.desc = desc;
7208 11 : bdev_io->internal.submit_tsc = spdk_get_ticks();
7209 11 : bdev_io->type = SPDK_BDEV_IO_TYPE_ABORT;
7210 11 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
7211 :
7212 11 : bdev_io->u.bdev.abort.bio_cb_arg = bio_cb_arg;
7213 :
7214 : /* Parent abort request is not submitted directly, but to manage its execution,
7215 : * add it to the submitted list here.
7216 : */
7217 11 : bdev_ch_add_to_io_submitted(bdev_io);
7218 :
7219 11 : bdev_abort(bdev_io);
7220 :
7221 11 : return 0;
7222 12 : }
7223 :
7224 : int
7225 4 : spdk_bdev_queue_io_wait(struct spdk_bdev *bdev, struct spdk_io_channel *ch,
7226 : struct spdk_bdev_io_wait_entry *entry)
7227 : {
7228 4 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
7229 4 : struct spdk_bdev_mgmt_channel *mgmt_ch = channel->shared_resource->mgmt_ch;
7230 :
7231 4 : if (bdev != entry->bdev) {
7232 0 : SPDK_ERRLOG("bdevs do not match\n");
7233 0 : return -EINVAL;
7234 : }
7235 :
7236 4 : if (mgmt_ch->per_thread_cache_count > 0) {
7237 0 : SPDK_ERRLOG("Cannot queue io_wait if spdk_bdev_io available in per-thread cache\n");
7238 0 : return -EINVAL;
7239 : }
7240 :
7241 4 : TAILQ_INSERT_TAIL(&mgmt_ch->io_wait_queue, entry, link);
7242 4 : return 0;
7243 4 : }
7244 :
7245 : static inline void
7246 610 : bdev_io_update_io_stat(struct spdk_bdev_io *bdev_io, uint64_t tsc_diff)
7247 : {
7248 610 : enum spdk_bdev_io_status io_status = bdev_io->internal.status;
7249 610 : struct spdk_bdev_io_stat *io_stat = bdev_io->internal.ch->stat;
7250 610 : uint64_t num_blocks = bdev_io->u.bdev.num_blocks;
7251 610 : uint32_t blocklen = bdev_io->bdev->blocklen;
7252 :
7253 610 : if (spdk_likely(io_status == SPDK_BDEV_IO_STATUS_SUCCESS)) {
7254 516 : switch (bdev_io->type) {
7255 : case SPDK_BDEV_IO_TYPE_READ:
7256 321 : io_stat->bytes_read += num_blocks * blocklen;
7257 321 : io_stat->num_read_ops++;
7258 321 : io_stat->read_latency_ticks += tsc_diff;
7259 321 : if (io_stat->max_read_latency_ticks < tsc_diff) {
7260 7 : io_stat->max_read_latency_ticks = tsc_diff;
7261 7 : }
7262 321 : if (io_stat->min_read_latency_ticks > tsc_diff) {
7263 42 : io_stat->min_read_latency_ticks = tsc_diff;
7264 42 : }
7265 321 : break;
7266 : case SPDK_BDEV_IO_TYPE_WRITE:
7267 75 : io_stat->bytes_written += num_blocks * blocklen;
7268 75 : io_stat->num_write_ops++;
7269 75 : io_stat->write_latency_ticks += tsc_diff;
7270 75 : if (io_stat->max_write_latency_ticks < tsc_diff) {
7271 4 : io_stat->max_write_latency_ticks = tsc_diff;
7272 4 : }
7273 75 : if (io_stat->min_write_latency_ticks > tsc_diff) {
7274 25 : io_stat->min_write_latency_ticks = tsc_diff;
7275 25 : }
7276 75 : break;
7277 : case SPDK_BDEV_IO_TYPE_UNMAP:
7278 20 : io_stat->bytes_unmapped += num_blocks * blocklen;
7279 20 : io_stat->num_unmap_ops++;
7280 20 : io_stat->unmap_latency_ticks += tsc_diff;
7281 20 : if (io_stat->max_unmap_latency_ticks < tsc_diff) {
7282 0 : io_stat->max_unmap_latency_ticks = tsc_diff;
7283 0 : }
7284 20 : if (io_stat->min_unmap_latency_ticks > tsc_diff) {
7285 3 : io_stat->min_unmap_latency_ticks = tsc_diff;
7286 3 : }
7287 20 : break;
7288 : case SPDK_BDEV_IO_TYPE_ZCOPY:
7289 : /* Track the data in the start phase only */
7290 4 : if (bdev_io->u.bdev.zcopy.start) {
7291 2 : if (bdev_io->u.bdev.zcopy.populate) {
7292 1 : io_stat->bytes_read += num_blocks * blocklen;
7293 1 : io_stat->num_read_ops++;
7294 1 : io_stat->read_latency_ticks += tsc_diff;
7295 1 : if (io_stat->max_read_latency_ticks < tsc_diff) {
7296 0 : io_stat->max_read_latency_ticks = tsc_diff;
7297 0 : }
7298 1 : if (io_stat->min_read_latency_ticks > tsc_diff) {
7299 1 : io_stat->min_read_latency_ticks = tsc_diff;
7300 1 : }
7301 1 : } else {
7302 1 : io_stat->bytes_written += num_blocks * blocklen;
7303 1 : io_stat->num_write_ops++;
7304 1 : io_stat->write_latency_ticks += tsc_diff;
7305 1 : if (io_stat->max_write_latency_ticks < tsc_diff) {
7306 0 : io_stat->max_write_latency_ticks = tsc_diff;
7307 0 : }
7308 1 : if (io_stat->min_write_latency_ticks > tsc_diff) {
7309 1 : io_stat->min_write_latency_ticks = tsc_diff;
7310 1 : }
7311 : }
7312 2 : }
7313 4 : break;
7314 : case SPDK_BDEV_IO_TYPE_COPY:
7315 21 : io_stat->bytes_copied += num_blocks * blocklen;
7316 21 : io_stat->num_copy_ops++;
7317 21 : bdev_io->internal.ch->stat->copy_latency_ticks += tsc_diff;
7318 21 : if (io_stat->max_copy_latency_ticks < tsc_diff) {
7319 0 : io_stat->max_copy_latency_ticks = tsc_diff;
7320 0 : }
7321 21 : if (io_stat->min_copy_latency_ticks > tsc_diff) {
7322 4 : io_stat->min_copy_latency_ticks = tsc_diff;
7323 4 : }
7324 21 : break;
7325 : default:
7326 75 : break;
7327 : }
7328 610 : } else if (io_status <= SPDK_BDEV_IO_STATUS_FAILED && io_status >= SPDK_MIN_BDEV_IO_STATUS) {
7329 94 : io_stat = bdev_io->bdev->internal.stat;
7330 94 : assert(io_stat->io_error != NULL);
7331 :
7332 94 : spdk_spin_lock(&bdev_io->bdev->internal.spinlock);
7333 94 : io_stat->io_error->error_status[-io_status - 1]++;
7334 94 : spdk_spin_unlock(&bdev_io->bdev->internal.spinlock);
7335 94 : }
7336 :
7337 : #ifdef SPDK_CONFIG_VTUNE
7338 : uint64_t now_tsc = spdk_get_ticks();
7339 : if (now_tsc > (bdev_io->internal.ch->start_tsc + bdev_io->internal.ch->interval_tsc)) {
7340 : uint64_t data[5];
7341 : struct spdk_bdev_io_stat *prev_stat = bdev_io->internal.ch->prev_stat;
7342 :
7343 : data[0] = io_stat->num_read_ops - prev_stat->num_read_ops;
7344 : data[1] = io_stat->bytes_read - prev_stat->bytes_read;
7345 : data[2] = io_stat->num_write_ops - prev_stat->num_write_ops;
7346 : data[3] = io_stat->bytes_written - prev_stat->bytes_written;
7347 : data[4] = bdev_io->bdev->fn_table->get_spin_time ?
7348 : bdev_io->bdev->fn_table->get_spin_time(spdk_bdev_io_get_io_channel(bdev_io)) : 0;
7349 :
7350 : __itt_metadata_add(g_bdev_mgr.domain, __itt_null, bdev_io->internal.ch->handle,
7351 : __itt_metadata_u64, 5, data);
7352 :
7353 : memcpy(prev_stat, io_stat, sizeof(struct spdk_bdev_io_stat));
7354 : bdev_io->internal.ch->start_tsc = now_tsc;
7355 : }
7356 : #endif
7357 610 : }
7358 :
7359 : static inline void
7360 610 : _bdev_io_complete(void *ctx)
7361 : {
7362 610 : struct spdk_bdev_io *bdev_io = ctx;
7363 :
7364 610 : if (spdk_unlikely(bdev_io_use_accel_sequence(bdev_io))) {
7365 0 : assert(bdev_io->internal.status != SPDK_BDEV_IO_STATUS_SUCCESS);
7366 0 : spdk_accel_sequence_abort(bdev_io->internal.accel_sequence);
7367 0 : }
7368 :
7369 610 : assert(bdev_io->internal.cb != NULL);
7370 610 : assert(spdk_get_thread() == spdk_bdev_io_get_thread(bdev_io));
7371 :
7372 1220 : bdev_io->internal.cb(bdev_io, bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS,
7373 610 : bdev_io->internal.caller_ctx);
7374 610 : }
7375 :
7376 : static inline void
7377 618 : bdev_io_complete(void *ctx)
7378 : {
7379 618 : struct spdk_bdev_io *bdev_io = ctx;
7380 618 : struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
7381 : uint64_t tsc, tsc_diff;
7382 :
7383 618 : if (spdk_unlikely(bdev_io->internal.f.in_submit_request)) {
7384 : /*
7385 : * Defer completion to avoid potential infinite recursion if the
7386 : * user's completion callback issues a new I/O.
7387 : */
7388 16 : spdk_thread_send_msg(spdk_bdev_io_get_thread(bdev_io),
7389 8 : bdev_io_complete, bdev_io);
7390 8 : return;
7391 : }
7392 :
7393 610 : tsc = spdk_get_ticks();
7394 610 : tsc_diff = tsc - bdev_io->internal.submit_tsc;
7395 :
7396 610 : bdev_ch_remove_from_io_submitted(bdev_io);
7397 610 : spdk_trace_record_tsc(tsc, TRACE_BDEV_IO_DONE, bdev_ch->trace_id, 0, (uintptr_t)bdev_io,
7398 : bdev_io->internal.caller_ctx, bdev_ch->queue_depth);
7399 :
7400 610 : if (bdev_ch->histogram) {
7401 4 : if (bdev_io->bdev->internal.histogram_io_type == 0 ||
7402 0 : bdev_io->bdev->internal.histogram_io_type == bdev_io->type) {
7403 : /*
7404 : * Tally all I/O types if the histogram_io_type is set to 0.
7405 : */
7406 4 : spdk_histogram_data_tally(bdev_ch->histogram, tsc_diff);
7407 4 : }
7408 4 : }
7409 :
7410 610 : bdev_io_update_io_stat(bdev_io, tsc_diff);
7411 610 : _bdev_io_complete(bdev_io);
7412 618 : }
7413 :
7414 : /* The difference between this function and bdev_io_complete() is that this should be called to
7415 : * complete IOs that haven't been submitted via bdev_io_submit(), as they weren't added onto the
7416 : * io_submitted list and don't have submit_tsc updated.
7417 : */
7418 : static inline void
7419 0 : bdev_io_complete_unsubmitted(struct spdk_bdev_io *bdev_io)
7420 : {
7421 : /* Since the IO hasn't been submitted it's bound to be failed */
7422 0 : assert(bdev_io->internal.status != SPDK_BDEV_IO_STATUS_SUCCESS);
7423 :
7424 : /* At this point we don't know if the IO is completed from submission context or not, but,
7425 : * since this is an error path, we can always do an spdk_thread_send_msg(). */
7426 0 : spdk_thread_send_msg(spdk_bdev_io_get_thread(bdev_io),
7427 0 : _bdev_io_complete, bdev_io);
7428 0 : }
7429 :
7430 : static void bdev_destroy_cb(void *io_device);
7431 :
7432 : static void
7433 15 : bdev_reset_complete(struct spdk_bdev *bdev, void *_ctx, int status)
7434 : {
7435 15 : struct spdk_bdev_io *bdev_io = _ctx;
7436 :
7437 15 : if (bdev_io->u.reset.ch_ref != NULL) {
7438 15 : spdk_put_io_channel(bdev_io->u.reset.ch_ref);
7439 15 : bdev_io->u.reset.ch_ref = NULL;
7440 15 : }
7441 :
7442 15 : bdev_io_complete(bdev_io);
7443 :
7444 15 : if (bdev->internal.status == SPDK_BDEV_STATUS_REMOVING &&
7445 1 : TAILQ_EMPTY(&bdev->internal.open_descs)) {
7446 1 : spdk_io_device_unregister(__bdev_to_io_dev(bdev), bdev_destroy_cb);
7447 1 : }
7448 15 : }
7449 :
7450 : static void
7451 18 : bdev_unfreeze_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
7452 : struct spdk_io_channel *_ch, void *_ctx)
7453 : {
7454 18 : struct spdk_bdev_io *bdev_io = _ctx;
7455 18 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
7456 : struct spdk_bdev_io *queued_reset;
7457 :
7458 18 : ch->flags &= ~BDEV_CH_RESET_IN_PROGRESS;
7459 18 : while (!TAILQ_EMPTY(&ch->queued_resets)) {
7460 0 : queued_reset = TAILQ_FIRST(&ch->queued_resets);
7461 0 : TAILQ_REMOVE(&ch->queued_resets, queued_reset, internal.link);
7462 0 : spdk_bdev_io_complete(queued_reset, bdev_io->internal.status);
7463 : }
7464 :
7465 18 : spdk_bdev_for_each_channel_continue(i, 0);
7466 18 : }
7467 :
7468 : static void
7469 0 : bdev_io_complete_sequence_cb(void *ctx, int status)
7470 : {
7471 0 : struct spdk_bdev_io *bdev_io = ctx;
7472 :
7473 : /* u.bdev.accel_sequence should have already been cleared at this point */
7474 0 : assert(bdev_io->u.bdev.accel_sequence == NULL);
7475 0 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS);
7476 0 : bdev_io->internal.f.has_accel_sequence = false;
7477 :
7478 0 : if (spdk_unlikely(status != 0)) {
7479 0 : SPDK_ERRLOG("Failed to execute accel sequence, status=%d\n", status);
7480 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
7481 0 : }
7482 :
7483 0 : bdev_io_complete(bdev_io);
7484 0 : }
7485 :
7486 : void
7487 598 : spdk_bdev_io_complete(struct spdk_bdev_io *bdev_io, enum spdk_bdev_io_status status)
7488 : {
7489 598 : struct spdk_bdev *bdev = bdev_io->bdev;
7490 598 : struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
7491 598 : struct spdk_bdev_shared_resource *shared_resource = bdev_ch->shared_resource;
7492 :
7493 598 : if (spdk_unlikely(bdev_io->internal.status != SPDK_BDEV_IO_STATUS_PENDING)) {
7494 0 : SPDK_ERRLOG("Unexpected completion on IO from %s module, status was %s\n",
7495 : spdk_bdev_get_module_name(bdev),
7496 : bdev_io_status_get_string(bdev_io->internal.status));
7497 0 : assert(false);
7498 : }
7499 598 : bdev_io->internal.status = status;
7500 :
7501 598 : if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_RESET)) {
7502 16 : bool unlock_channels = false;
7503 :
7504 16 : if (status == SPDK_BDEV_IO_STATUS_NOMEM) {
7505 0 : SPDK_ERRLOG("NOMEM returned for reset\n");
7506 0 : }
7507 16 : spdk_spin_lock(&bdev->internal.spinlock);
7508 16 : if (bdev_io == bdev->internal.reset_in_progress) {
7509 15 : bdev->internal.reset_in_progress = NULL;
7510 15 : unlock_channels = true;
7511 15 : }
7512 16 : spdk_spin_unlock(&bdev->internal.spinlock);
7513 :
7514 16 : if (unlock_channels) {
7515 15 : spdk_bdev_for_each_channel(bdev, bdev_unfreeze_channel, bdev_io,
7516 : bdev_reset_complete);
7517 15 : return;
7518 : }
7519 1 : } else {
7520 582 : bdev_io_decrement_outstanding(bdev_ch, shared_resource);
7521 582 : if (spdk_likely(status == SPDK_BDEV_IO_STATUS_SUCCESS)) {
7522 485 : if (bdev_io_needs_sequence_exec(bdev_io->internal.desc, bdev_io)) {
7523 0 : bdev_io_exec_sequence(bdev_io, bdev_io_complete_sequence_cb);
7524 0 : return;
7525 485 : } else if (spdk_unlikely(bdev_io->internal.f.has_bounce_buf &&
7526 : !bdev_io_use_accel_sequence(bdev_io))) {
7527 26 : _bdev_io_push_bounce_data_buffer(bdev_io,
7528 : _bdev_io_complete_push_bounce_done);
7529 : /* bdev IO will be completed in the callback */
7530 26 : return;
7531 : }
7532 459 : }
7533 :
7534 556 : if (spdk_unlikely(_bdev_io_handle_no_mem(bdev_io, BDEV_IO_RETRY_STATE_SUBMIT))) {
7535 5 : return;
7536 : }
7537 : }
7538 :
7539 552 : bdev_io_complete(bdev_io);
7540 598 : }
7541 :
7542 : void
7543 0 : spdk_bdev_io_complete_scsi_status(struct spdk_bdev_io *bdev_io, enum spdk_scsi_status sc,
7544 : enum spdk_scsi_sense sk, uint8_t asc, uint8_t ascq)
7545 : {
7546 : enum spdk_bdev_io_status status;
7547 :
7548 0 : if (sc == SPDK_SCSI_STATUS_GOOD) {
7549 0 : status = SPDK_BDEV_IO_STATUS_SUCCESS;
7550 0 : } else {
7551 0 : status = SPDK_BDEV_IO_STATUS_SCSI_ERROR;
7552 0 : bdev_io->internal.error.scsi.sc = sc;
7553 0 : bdev_io->internal.error.scsi.sk = sk;
7554 0 : bdev_io->internal.error.scsi.asc = asc;
7555 0 : bdev_io->internal.error.scsi.ascq = ascq;
7556 : }
7557 :
7558 0 : spdk_bdev_io_complete(bdev_io, status);
7559 0 : }
7560 :
7561 : void
7562 0 : spdk_bdev_io_get_scsi_status(const struct spdk_bdev_io *bdev_io,
7563 : int *sc, int *sk, int *asc, int *ascq)
7564 : {
7565 0 : assert(sc != NULL);
7566 0 : assert(sk != NULL);
7567 0 : assert(asc != NULL);
7568 0 : assert(ascq != NULL);
7569 :
7570 0 : switch (bdev_io->internal.status) {
7571 : case SPDK_BDEV_IO_STATUS_SUCCESS:
7572 0 : *sc = SPDK_SCSI_STATUS_GOOD;
7573 0 : *sk = SPDK_SCSI_SENSE_NO_SENSE;
7574 0 : *asc = SPDK_SCSI_ASC_NO_ADDITIONAL_SENSE;
7575 0 : *ascq = SPDK_SCSI_ASCQ_CAUSE_NOT_REPORTABLE;
7576 0 : break;
7577 : case SPDK_BDEV_IO_STATUS_NVME_ERROR:
7578 0 : spdk_scsi_nvme_translate(bdev_io, sc, sk, asc, ascq);
7579 0 : break;
7580 : case SPDK_BDEV_IO_STATUS_MISCOMPARE:
7581 0 : *sc = SPDK_SCSI_STATUS_CHECK_CONDITION;
7582 0 : *sk = SPDK_SCSI_SENSE_MISCOMPARE;
7583 0 : *asc = SPDK_SCSI_ASC_MISCOMPARE_DURING_VERIFY_OPERATION;
7584 0 : *ascq = bdev_io->internal.error.scsi.ascq;
7585 0 : break;
7586 : case SPDK_BDEV_IO_STATUS_SCSI_ERROR:
7587 0 : *sc = bdev_io->internal.error.scsi.sc;
7588 0 : *sk = bdev_io->internal.error.scsi.sk;
7589 0 : *asc = bdev_io->internal.error.scsi.asc;
7590 0 : *ascq = bdev_io->internal.error.scsi.ascq;
7591 0 : break;
7592 : default:
7593 0 : *sc = SPDK_SCSI_STATUS_CHECK_CONDITION;
7594 0 : *sk = SPDK_SCSI_SENSE_ABORTED_COMMAND;
7595 0 : *asc = SPDK_SCSI_ASC_NO_ADDITIONAL_SENSE;
7596 0 : *ascq = SPDK_SCSI_ASCQ_CAUSE_NOT_REPORTABLE;
7597 0 : break;
7598 : }
7599 0 : }
7600 :
7601 : void
7602 0 : spdk_bdev_io_complete_aio_status(struct spdk_bdev_io *bdev_io, int aio_result)
7603 : {
7604 : enum spdk_bdev_io_status status;
7605 :
7606 0 : if (aio_result == 0) {
7607 0 : status = SPDK_BDEV_IO_STATUS_SUCCESS;
7608 0 : } else {
7609 0 : status = SPDK_BDEV_IO_STATUS_AIO_ERROR;
7610 : }
7611 :
7612 0 : bdev_io->internal.error.aio_result = aio_result;
7613 :
7614 0 : spdk_bdev_io_complete(bdev_io, status);
7615 0 : }
7616 :
7617 : void
7618 0 : spdk_bdev_io_get_aio_status(const struct spdk_bdev_io *bdev_io, int *aio_result)
7619 : {
7620 0 : assert(aio_result != NULL);
7621 :
7622 0 : if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_AIO_ERROR) {
7623 0 : *aio_result = bdev_io->internal.error.aio_result;
7624 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS) {
7625 0 : *aio_result = 0;
7626 0 : } else {
7627 0 : *aio_result = -EIO;
7628 : }
7629 0 : }
7630 :
7631 : void
7632 0 : spdk_bdev_io_complete_nvme_status(struct spdk_bdev_io *bdev_io, uint32_t cdw0, int sct, int sc)
7633 : {
7634 : enum spdk_bdev_io_status status;
7635 :
7636 0 : if (spdk_likely(sct == SPDK_NVME_SCT_GENERIC && sc == SPDK_NVME_SC_SUCCESS)) {
7637 0 : status = SPDK_BDEV_IO_STATUS_SUCCESS;
7638 0 : } else if (sct == SPDK_NVME_SCT_GENERIC && sc == SPDK_NVME_SC_ABORTED_BY_REQUEST) {
7639 0 : status = SPDK_BDEV_IO_STATUS_ABORTED;
7640 0 : } else {
7641 0 : status = SPDK_BDEV_IO_STATUS_NVME_ERROR;
7642 : }
7643 :
7644 0 : bdev_io->internal.error.nvme.cdw0 = cdw0;
7645 0 : bdev_io->internal.error.nvme.sct = sct;
7646 0 : bdev_io->internal.error.nvme.sc = sc;
7647 :
7648 0 : spdk_bdev_io_complete(bdev_io, status);
7649 0 : }
7650 :
7651 : void
7652 0 : spdk_bdev_io_get_nvme_status(const struct spdk_bdev_io *bdev_io, uint32_t *cdw0, int *sct, int *sc)
7653 : {
7654 0 : assert(sct != NULL);
7655 0 : assert(sc != NULL);
7656 0 : assert(cdw0 != NULL);
7657 :
7658 0 : if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_ABORT)) {
7659 0 : *sct = SPDK_NVME_SCT_GENERIC;
7660 0 : *sc = SPDK_NVME_SC_SUCCESS;
7661 0 : if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS) {
7662 0 : *cdw0 = 0;
7663 0 : } else {
7664 0 : *cdw0 = 1U;
7665 : }
7666 0 : return;
7667 : }
7668 :
7669 0 : if (spdk_likely(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS)) {
7670 0 : *sct = SPDK_NVME_SCT_GENERIC;
7671 0 : *sc = SPDK_NVME_SC_SUCCESS;
7672 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_NVME_ERROR) {
7673 0 : *sct = bdev_io->internal.error.nvme.sct;
7674 0 : *sc = bdev_io->internal.error.nvme.sc;
7675 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_ABORTED) {
7676 0 : *sct = SPDK_NVME_SCT_GENERIC;
7677 0 : *sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
7678 0 : } else {
7679 0 : *sct = SPDK_NVME_SCT_GENERIC;
7680 0 : *sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
7681 : }
7682 :
7683 0 : *cdw0 = bdev_io->internal.error.nvme.cdw0;
7684 0 : }
7685 :
7686 : void
7687 0 : spdk_bdev_io_get_nvme_fused_status(const struct spdk_bdev_io *bdev_io, uint32_t *cdw0,
7688 : int *first_sct, int *first_sc, int *second_sct, int *second_sc)
7689 : {
7690 0 : assert(first_sct != NULL);
7691 0 : assert(first_sc != NULL);
7692 0 : assert(second_sct != NULL);
7693 0 : assert(second_sc != NULL);
7694 0 : assert(cdw0 != NULL);
7695 :
7696 0 : if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_NVME_ERROR) {
7697 0 : if (bdev_io->internal.error.nvme.sct == SPDK_NVME_SCT_MEDIA_ERROR &&
7698 0 : bdev_io->internal.error.nvme.sc == SPDK_NVME_SC_COMPARE_FAILURE) {
7699 0 : *first_sct = bdev_io->internal.error.nvme.sct;
7700 0 : *first_sc = bdev_io->internal.error.nvme.sc;
7701 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
7702 0 : *second_sc = SPDK_NVME_SC_ABORTED_FAILED_FUSED;
7703 0 : } else {
7704 0 : *first_sct = SPDK_NVME_SCT_GENERIC;
7705 0 : *first_sc = SPDK_NVME_SC_SUCCESS;
7706 0 : *second_sct = bdev_io->internal.error.nvme.sct;
7707 0 : *second_sc = bdev_io->internal.error.nvme.sc;
7708 : }
7709 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_ABORTED) {
7710 0 : *first_sct = SPDK_NVME_SCT_GENERIC;
7711 0 : *first_sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
7712 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
7713 0 : *second_sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
7714 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS) {
7715 0 : *first_sct = SPDK_NVME_SCT_GENERIC;
7716 0 : *first_sc = SPDK_NVME_SC_SUCCESS;
7717 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
7718 0 : *second_sc = SPDK_NVME_SC_SUCCESS;
7719 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_FIRST_FUSED_FAILED) {
7720 0 : *first_sct = SPDK_NVME_SCT_GENERIC;
7721 0 : *first_sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
7722 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
7723 0 : *second_sc = SPDK_NVME_SC_ABORTED_FAILED_FUSED;
7724 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_MISCOMPARE) {
7725 0 : *first_sct = SPDK_NVME_SCT_MEDIA_ERROR;
7726 0 : *first_sc = SPDK_NVME_SC_COMPARE_FAILURE;
7727 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
7728 0 : *second_sc = SPDK_NVME_SC_ABORTED_FAILED_FUSED;
7729 0 : } else {
7730 0 : *first_sct = SPDK_NVME_SCT_GENERIC;
7731 0 : *first_sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
7732 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
7733 0 : *second_sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
7734 : }
7735 :
7736 0 : *cdw0 = bdev_io->internal.error.nvme.cdw0;
7737 0 : }
7738 :
7739 : void
7740 0 : spdk_bdev_io_complete_base_io_status(struct spdk_bdev_io *bdev_io,
7741 : const struct spdk_bdev_io *base_io)
7742 : {
7743 0 : switch (base_io->internal.status) {
7744 : case SPDK_BDEV_IO_STATUS_NVME_ERROR:
7745 0 : spdk_bdev_io_complete_nvme_status(bdev_io,
7746 0 : base_io->internal.error.nvme.cdw0,
7747 0 : base_io->internal.error.nvme.sct,
7748 0 : base_io->internal.error.nvme.sc);
7749 0 : break;
7750 : case SPDK_BDEV_IO_STATUS_SCSI_ERROR:
7751 0 : spdk_bdev_io_complete_scsi_status(bdev_io,
7752 0 : base_io->internal.error.scsi.sc,
7753 0 : base_io->internal.error.scsi.sk,
7754 0 : base_io->internal.error.scsi.asc,
7755 0 : base_io->internal.error.scsi.ascq);
7756 0 : break;
7757 : case SPDK_BDEV_IO_STATUS_AIO_ERROR:
7758 0 : spdk_bdev_io_complete_aio_status(bdev_io, base_io->internal.error.aio_result);
7759 0 : break;
7760 : default:
7761 0 : spdk_bdev_io_complete(bdev_io, base_io->internal.status);
7762 0 : break;
7763 : }
7764 0 : }
7765 :
7766 : struct spdk_thread *
7767 660 : spdk_bdev_io_get_thread(struct spdk_bdev_io *bdev_io)
7768 : {
7769 660 : return spdk_io_channel_get_thread(bdev_io->internal.ch->channel);
7770 : }
7771 :
7772 : struct spdk_io_channel *
7773 70 : spdk_bdev_io_get_io_channel(struct spdk_bdev_io *bdev_io)
7774 : {
7775 70 : return bdev_io->internal.ch->channel;
7776 : }
7777 :
7778 : static int
7779 126 : bdev_register(struct spdk_bdev *bdev)
7780 : {
7781 : char *bdev_name;
7782 : char uuid[SPDK_UUID_STRING_LEN];
7783 : struct spdk_iobuf_opts iobuf_opts;
7784 : int ret;
7785 :
7786 126 : assert(bdev->module != NULL);
7787 :
7788 126 : if (!bdev->name) {
7789 0 : SPDK_ERRLOG("Bdev name is NULL\n");
7790 0 : return -EINVAL;
7791 : }
7792 :
7793 126 : if (!strlen(bdev->name)) {
7794 0 : SPDK_ERRLOG("Bdev name must not be an empty string\n");
7795 0 : return -EINVAL;
7796 : }
7797 :
7798 : /* Users often register their own I/O devices using the bdev name. In
7799 : * order to avoid conflicts, prepend bdev_. */
7800 126 : bdev_name = spdk_sprintf_alloc("bdev_%s", bdev->name);
7801 126 : if (!bdev_name) {
7802 0 : SPDK_ERRLOG("Unable to allocate memory for internal bdev name.\n");
7803 0 : return -ENOMEM;
7804 : }
7805 :
7806 126 : bdev->internal.stat = bdev_alloc_io_stat(true);
7807 126 : if (!bdev->internal.stat) {
7808 0 : SPDK_ERRLOG("Unable to allocate I/O statistics structure.\n");
7809 0 : free(bdev_name);
7810 0 : return -ENOMEM;
7811 : }
7812 :
7813 126 : bdev->internal.status = SPDK_BDEV_STATUS_READY;
7814 126 : bdev->internal.measured_queue_depth = UINT64_MAX;
7815 126 : bdev->internal.claim_type = SPDK_BDEV_CLAIM_NONE;
7816 126 : memset(&bdev->internal.claim, 0, sizeof(bdev->internal.claim));
7817 126 : bdev->internal.qd_poller = NULL;
7818 126 : bdev->internal.qos = NULL;
7819 :
7820 126 : TAILQ_INIT(&bdev->internal.open_descs);
7821 126 : TAILQ_INIT(&bdev->internal.locked_ranges);
7822 126 : TAILQ_INIT(&bdev->internal.pending_locked_ranges);
7823 126 : TAILQ_INIT(&bdev->aliases);
7824 :
7825 : /* UUID may be specified by the user or defined by bdev itself.
7826 : * Otherwise it will be generated here, so this field will never be empty. */
7827 126 : if (spdk_uuid_is_null(&bdev->uuid)) {
7828 42 : spdk_uuid_generate(&bdev->uuid);
7829 42 : }
7830 :
7831 : /* Add the UUID alias only if it's different than the name */
7832 126 : spdk_uuid_fmt_lower(uuid, sizeof(uuid), &bdev->uuid);
7833 126 : if (strcmp(bdev->name, uuid) != 0) {
7834 125 : ret = spdk_bdev_alias_add(bdev, uuid);
7835 125 : if (ret != 0) {
7836 2 : SPDK_ERRLOG("Unable to add uuid:%s alias for bdev %s\n", uuid, bdev->name);
7837 2 : bdev_free_io_stat(bdev->internal.stat);
7838 2 : free(bdev_name);
7839 2 : return ret;
7840 : }
7841 123 : }
7842 :
7843 124 : spdk_iobuf_get_opts(&iobuf_opts, sizeof(iobuf_opts));
7844 124 : if (spdk_bdev_get_buf_align(bdev) > 1) {
7845 0 : bdev->max_rw_size = spdk_min(bdev->max_rw_size ? bdev->max_rw_size : UINT32_MAX,
7846 : iobuf_opts.large_bufsize / bdev->blocklen);
7847 0 : }
7848 :
7849 : /* If the user didn't specify a write unit size, set it to one. */
7850 124 : if (bdev->write_unit_size == 0) {
7851 120 : bdev->write_unit_size = 1;
7852 120 : }
7853 :
7854 : /* Set ACWU value to the write unit size if bdev module did not set it (does not support it natively) */
7855 124 : if (bdev->acwu == 0) {
7856 120 : bdev->acwu = bdev->write_unit_size;
7857 120 : }
7858 :
7859 124 : if (bdev->phys_blocklen == 0) {
7860 120 : bdev->phys_blocklen = spdk_bdev_get_data_block_size(bdev);
7861 120 : }
7862 :
7863 124 : if (!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COPY)) {
7864 0 : bdev->max_copy = bdev_get_max_write(bdev, iobuf_opts.large_bufsize);
7865 0 : }
7866 :
7867 124 : if (!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE_ZEROES)) {
7868 0 : bdev->max_write_zeroes = bdev_get_max_write(bdev, ZERO_BUFFER_SIZE);
7869 0 : }
7870 :
7871 124 : bdev->internal.reset_in_progress = NULL;
7872 124 : bdev->internal.qd_poll_in_progress = false;
7873 124 : bdev->internal.period = 0;
7874 124 : bdev->internal.new_period = 0;
7875 124 : bdev->internal.trace_id = spdk_trace_register_owner(OWNER_TYPE_BDEV, bdev_name);
7876 :
7877 : /*
7878 : * Initialize spinlock before registering IO device because spinlock is used in
7879 : * bdev_channel_create
7880 : */
7881 124 : spdk_spin_init(&bdev->internal.spinlock);
7882 :
7883 248 : spdk_io_device_register(__bdev_to_io_dev(bdev),
7884 : bdev_channel_create, bdev_channel_destroy,
7885 : sizeof(struct spdk_bdev_channel),
7886 124 : bdev_name);
7887 :
7888 : /*
7889 : * Register bdev name only after the bdev object is ready.
7890 : * After bdev_name_add returns, it is possible for other threads to start using the bdev,
7891 : * create IO channels...
7892 : */
7893 124 : ret = bdev_name_add(&bdev->internal.bdev_name, bdev, bdev->name);
7894 124 : if (ret != 0) {
7895 0 : spdk_io_device_unregister(__bdev_to_io_dev(bdev), NULL);
7896 0 : bdev_free_io_stat(bdev->internal.stat);
7897 0 : spdk_spin_destroy(&bdev->internal.spinlock);
7898 0 : free(bdev_name);
7899 0 : return ret;
7900 : }
7901 :
7902 124 : free(bdev_name);
7903 :
7904 124 : SPDK_DEBUGLOG(bdev, "Inserting bdev %s into list\n", bdev->name);
7905 124 : TAILQ_INSERT_TAIL(&g_bdev_mgr.bdevs, bdev, internal.link);
7906 :
7907 124 : return 0;
7908 126 : }
7909 :
7910 : static void
7911 125 : bdev_destroy_cb(void *io_device)
7912 : {
7913 : int rc;
7914 : struct spdk_bdev *bdev;
7915 : spdk_bdev_unregister_cb cb_fn;
7916 : void *cb_arg;
7917 :
7918 125 : bdev = __bdev_from_io_dev(io_device);
7919 :
7920 125 : if (bdev->internal.unregister_td != spdk_get_thread()) {
7921 1 : spdk_thread_send_msg(bdev->internal.unregister_td, bdev_destroy_cb, io_device);
7922 1 : return;
7923 : }
7924 :
7925 124 : cb_fn = bdev->internal.unregister_cb;
7926 124 : cb_arg = bdev->internal.unregister_ctx;
7927 :
7928 124 : spdk_spin_destroy(&bdev->internal.spinlock);
7929 124 : free(bdev->internal.qos);
7930 124 : bdev_free_io_stat(bdev->internal.stat);
7931 124 : spdk_trace_unregister_owner(bdev->internal.trace_id);
7932 :
7933 124 : rc = bdev->fn_table->destruct(bdev->ctxt);
7934 124 : if (rc < 0) {
7935 0 : SPDK_ERRLOG("destruct failed\n");
7936 0 : }
7937 124 : if (rc <= 0 && cb_fn != NULL) {
7938 10 : cb_fn(cb_arg, rc);
7939 10 : }
7940 125 : }
7941 :
7942 : void
7943 2 : spdk_bdev_destruct_done(struct spdk_bdev *bdev, int bdeverrno)
7944 : {
7945 2 : if (bdev->internal.unregister_cb != NULL) {
7946 0 : bdev->internal.unregister_cb(bdev->internal.unregister_ctx, bdeverrno);
7947 0 : }
7948 2 : }
7949 :
7950 : static void
7951 19 : _remove_notify(void *arg)
7952 : {
7953 19 : struct spdk_bdev_desc *desc = arg;
7954 :
7955 19 : _event_notify(desc, SPDK_BDEV_EVENT_REMOVE);
7956 19 : }
7957 :
7958 : /* returns: 0 - bdev removed and ready to be destructed.
7959 : * -EBUSY - bdev can't be destructed yet. */
7960 : static int
7961 139 : bdev_unregister_unsafe(struct spdk_bdev *bdev)
7962 : {
7963 : struct spdk_bdev_desc *desc, *tmp;
7964 139 : int rc = 0;
7965 : char uuid[SPDK_UUID_STRING_LEN];
7966 :
7967 139 : assert(spdk_spin_held(&g_bdev_mgr.spinlock));
7968 139 : assert(spdk_spin_held(&bdev->internal.spinlock));
7969 :
7970 : /* Notify each descriptor about hotremoval */
7971 158 : TAILQ_FOREACH_SAFE(desc, &bdev->internal.open_descs, link, tmp) {
7972 19 : rc = -EBUSY;
7973 : /*
7974 : * Defer invocation of the event_cb to a separate message that will
7975 : * run later on its thread. This ensures this context unwinds and
7976 : * we don't recursively unregister this bdev again if the event_cb
7977 : * immediately closes its descriptor.
7978 : */
7979 19 : event_notify(desc, _remove_notify);
7980 19 : }
7981 :
7982 : /* If there are no descriptors, proceed removing the bdev */
7983 139 : if (rc == 0) {
7984 124 : TAILQ_REMOVE(&g_bdev_mgr.bdevs, bdev, internal.link);
7985 124 : SPDK_DEBUGLOG(bdev, "Removing bdev %s from list done\n", bdev->name);
7986 :
7987 : /* Delete the name and the UUID alias */
7988 124 : spdk_uuid_fmt_lower(uuid, sizeof(uuid), &bdev->uuid);
7989 124 : bdev_name_del_unsafe(&bdev->internal.bdev_name);
7990 124 : bdev_alias_del(bdev, uuid, bdev_name_del_unsafe);
7991 :
7992 124 : spdk_notify_send("bdev_unregister", spdk_bdev_get_name(bdev));
7993 :
7994 124 : if (bdev->internal.reset_in_progress != NULL) {
7995 : /* If reset is in progress, let the completion callback for reset
7996 : * unregister the bdev.
7997 : */
7998 1 : rc = -EBUSY;
7999 1 : }
8000 124 : }
8001 :
8002 139 : return rc;
8003 : }
8004 :
8005 : static void
8006 4 : bdev_unregister_abort_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
8007 : struct spdk_io_channel *io_ch, void *_ctx)
8008 : {
8009 4 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
8010 :
8011 4 : bdev_channel_abort_queued_ios(bdev_ch);
8012 4 : spdk_bdev_for_each_channel_continue(i, 0);
8013 4 : }
8014 :
8015 : static void
8016 124 : bdev_unregister(struct spdk_bdev *bdev, void *_ctx, int status)
8017 : {
8018 : int rc;
8019 :
8020 124 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8021 124 : spdk_spin_lock(&bdev->internal.spinlock);
8022 : /*
8023 : * Set the status to REMOVING after completing to abort channels. Otherwise,
8024 : * the last spdk_bdev_close() may call spdk_io_device_unregister() while
8025 : * spdk_bdev_for_each_channel() is executed and spdk_io_device_unregister()
8026 : * may fail.
8027 : */
8028 124 : bdev->internal.status = SPDK_BDEV_STATUS_REMOVING;
8029 124 : rc = bdev_unregister_unsafe(bdev);
8030 124 : spdk_spin_unlock(&bdev->internal.spinlock);
8031 124 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8032 :
8033 124 : if (rc == 0) {
8034 108 : spdk_io_device_unregister(__bdev_to_io_dev(bdev), bdev_destroy_cb);
8035 108 : }
8036 124 : }
8037 :
8038 : void
8039 131 : spdk_bdev_unregister(struct spdk_bdev *bdev, spdk_bdev_unregister_cb cb_fn, void *cb_arg)
8040 : {
8041 : struct spdk_thread *thread;
8042 :
8043 131 : SPDK_DEBUGLOG(bdev, "Removing bdev %s from list\n", bdev->name);
8044 :
8045 131 : thread = spdk_get_thread();
8046 131 : if (!thread) {
8047 : /* The user called this from a non-SPDK thread. */
8048 0 : if (cb_fn != NULL) {
8049 0 : cb_fn(cb_arg, -ENOTSUP);
8050 0 : }
8051 0 : return;
8052 : }
8053 :
8054 131 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8055 131 : if (bdev->internal.status == SPDK_BDEV_STATUS_UNREGISTERING ||
8056 131 : bdev->internal.status == SPDK_BDEV_STATUS_REMOVING) {
8057 7 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8058 7 : if (cb_fn) {
8059 0 : cb_fn(cb_arg, -EBUSY);
8060 0 : }
8061 7 : return;
8062 : }
8063 :
8064 124 : spdk_spin_lock(&bdev->internal.spinlock);
8065 124 : bdev->internal.status = SPDK_BDEV_STATUS_UNREGISTERING;
8066 124 : bdev->internal.unregister_cb = cb_fn;
8067 124 : bdev->internal.unregister_ctx = cb_arg;
8068 124 : bdev->internal.unregister_td = thread;
8069 124 : spdk_spin_unlock(&bdev->internal.spinlock);
8070 124 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8071 :
8072 124 : spdk_bdev_set_qd_sampling_period(bdev, 0);
8073 :
8074 124 : spdk_bdev_for_each_channel(bdev, bdev_unregister_abort_channel, bdev,
8075 : bdev_unregister);
8076 131 : }
8077 :
8078 : int
8079 4 : spdk_bdev_unregister_by_name(const char *bdev_name, struct spdk_bdev_module *module,
8080 : spdk_bdev_unregister_cb cb_fn, void *cb_arg)
8081 : {
8082 : struct spdk_bdev_desc *desc;
8083 : struct spdk_bdev *bdev;
8084 : int rc;
8085 :
8086 4 : rc = spdk_bdev_open_ext(bdev_name, false, _tmp_bdev_event_cb, NULL, &desc);
8087 4 : if (rc != 0) {
8088 1 : SPDK_ERRLOG("Failed to open bdev with name: %s\n", bdev_name);
8089 1 : return rc;
8090 : }
8091 :
8092 3 : bdev = spdk_bdev_desc_get_bdev(desc);
8093 :
8094 3 : if (bdev->module != module) {
8095 1 : spdk_bdev_close(desc);
8096 1 : SPDK_ERRLOG("Bdev %s was not registered by the specified module.\n",
8097 : bdev_name);
8098 1 : return -ENODEV;
8099 : }
8100 :
8101 2 : spdk_bdev_unregister(bdev, cb_fn, cb_arg);
8102 :
8103 2 : spdk_bdev_close(desc);
8104 :
8105 2 : return 0;
8106 4 : }
8107 :
8108 : static int
8109 255 : bdev_start_qos(struct spdk_bdev *bdev)
8110 : {
8111 : struct set_qos_limit_ctx *ctx;
8112 :
8113 : /* Enable QoS */
8114 255 : if (bdev->internal.qos && bdev->internal.qos->thread == NULL) {
8115 2 : ctx = calloc(1, sizeof(*ctx));
8116 2 : if (ctx == NULL) {
8117 0 : SPDK_ERRLOG("Failed to allocate memory for QoS context\n");
8118 0 : return -ENOMEM;
8119 : }
8120 2 : ctx->bdev = bdev;
8121 2 : spdk_bdev_for_each_channel(bdev, bdev_enable_qos_msg, ctx, bdev_enable_qos_done);
8122 2 : }
8123 :
8124 255 : return 0;
8125 255 : }
8126 :
8127 : static void
8128 24 : log_already_claimed(enum spdk_log_level level, const int line, const char *func, const char *detail,
8129 : struct spdk_bdev *bdev)
8130 : {
8131 : enum spdk_bdev_claim_type type;
8132 : const char *typename, *modname;
8133 : extern struct spdk_log_flag SPDK_LOG_bdev;
8134 :
8135 24 : assert(spdk_spin_held(&bdev->internal.spinlock));
8136 :
8137 24 : if (level >= SPDK_LOG_INFO && !SPDK_LOG_bdev.enabled) {
8138 0 : return;
8139 : }
8140 :
8141 24 : type = bdev->internal.claim_type;
8142 24 : typename = spdk_bdev_claim_get_name(type);
8143 :
8144 24 : if (type == SPDK_BDEV_CLAIM_EXCL_WRITE) {
8145 6 : modname = bdev->internal.claim.v1.module->name;
8146 12 : spdk_log(level, __FILE__, line, func, "bdev %s %s: type %s by module %s\n",
8147 6 : bdev->name, detail, typename, modname);
8148 6 : return;
8149 : }
8150 :
8151 18 : if (claim_type_is_v2(type)) {
8152 : struct spdk_bdev_module_claim *claim;
8153 :
8154 36 : TAILQ_FOREACH(claim, &bdev->internal.claim.v2.claims, link) {
8155 18 : modname = claim->module->name;
8156 36 : spdk_log(level, __FILE__, line, func, "bdev %s %s: type %s by module %s\n",
8157 18 : bdev->name, detail, typename, modname);
8158 18 : }
8159 18 : return;
8160 : }
8161 :
8162 0 : assert(false);
8163 24 : }
8164 :
8165 : static int
8166 264 : bdev_open(struct spdk_bdev *bdev, bool write, struct spdk_bdev_desc *desc)
8167 : {
8168 : struct spdk_thread *thread;
8169 264 : int rc = 0;
8170 :
8171 264 : thread = spdk_get_thread();
8172 264 : if (!thread) {
8173 0 : SPDK_ERRLOG("Cannot open bdev from non-SPDK thread.\n");
8174 0 : return -ENOTSUP;
8175 : }
8176 :
8177 264 : SPDK_DEBUGLOG(bdev, "Opening descriptor %p for bdev %s on thread %p\n", desc, bdev->name,
8178 : spdk_get_thread());
8179 :
8180 264 : desc->bdev = bdev;
8181 264 : desc->thread = thread;
8182 264 : desc->write = write;
8183 :
8184 264 : spdk_spin_lock(&bdev->internal.spinlock);
8185 264 : if (bdev->internal.status == SPDK_BDEV_STATUS_UNREGISTERING ||
8186 264 : bdev->internal.status == SPDK_BDEV_STATUS_REMOVING) {
8187 3 : spdk_spin_unlock(&bdev->internal.spinlock);
8188 3 : return -ENODEV;
8189 : }
8190 :
8191 261 : if (write && bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE) {
8192 6 : LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
8193 6 : spdk_spin_unlock(&bdev->internal.spinlock);
8194 6 : return -EPERM;
8195 : }
8196 :
8197 255 : rc = bdev_start_qos(bdev);
8198 255 : if (rc != 0) {
8199 0 : SPDK_ERRLOG("Failed to start QoS on bdev %s\n", bdev->name);
8200 0 : spdk_spin_unlock(&bdev->internal.spinlock);
8201 0 : return rc;
8202 : }
8203 :
8204 255 : TAILQ_INSERT_TAIL(&bdev->internal.open_descs, desc, link);
8205 :
8206 255 : spdk_spin_unlock(&bdev->internal.spinlock);
8207 :
8208 255 : return 0;
8209 264 : }
8210 :
8211 : static int
8212 264 : bdev_desc_alloc(struct spdk_bdev *bdev, spdk_bdev_event_cb_t event_cb, void *event_ctx,
8213 : struct spdk_bdev_desc **_desc)
8214 : {
8215 : struct spdk_bdev_desc *desc;
8216 : unsigned int i;
8217 :
8218 264 : desc = calloc(1, sizeof(*desc));
8219 264 : if (desc == NULL) {
8220 0 : SPDK_ERRLOG("Failed to allocate memory for bdev descriptor\n");
8221 0 : return -ENOMEM;
8222 : }
8223 :
8224 264 : TAILQ_INIT(&desc->pending_media_events);
8225 264 : TAILQ_INIT(&desc->free_media_events);
8226 :
8227 264 : desc->memory_domains_supported = spdk_bdev_get_memory_domains(bdev, NULL, 0) > 0;
8228 264 : desc->callback.event_fn = event_cb;
8229 264 : desc->callback.ctx = event_ctx;
8230 264 : spdk_spin_init(&desc->spinlock);
8231 :
8232 264 : if (bdev->media_events) {
8233 0 : desc->media_events_buffer = calloc(MEDIA_EVENT_POOL_SIZE,
8234 : sizeof(*desc->media_events_buffer));
8235 0 : if (desc->media_events_buffer == NULL) {
8236 0 : SPDK_ERRLOG("Failed to initialize media event pool\n");
8237 0 : bdev_desc_free(desc);
8238 0 : return -ENOMEM;
8239 : }
8240 :
8241 0 : for (i = 0; i < MEDIA_EVENT_POOL_SIZE; ++i) {
8242 0 : TAILQ_INSERT_TAIL(&desc->free_media_events,
8243 : &desc->media_events_buffer[i], tailq);
8244 0 : }
8245 0 : }
8246 :
8247 264 : if (bdev->fn_table->accel_sequence_supported != NULL) {
8248 0 : for (i = 0; i < SPDK_BDEV_NUM_IO_TYPES; ++i) {
8249 0 : desc->accel_sequence_supported[i] =
8250 0 : bdev->fn_table->accel_sequence_supported(bdev->ctxt,
8251 0 : (enum spdk_bdev_io_type)i);
8252 0 : }
8253 0 : }
8254 :
8255 264 : *_desc = desc;
8256 :
8257 264 : return 0;
8258 264 : }
8259 :
8260 : static int
8261 128 : bdev_open_ext(const char *bdev_name, bool write, spdk_bdev_event_cb_t event_cb,
8262 : void *event_ctx, struct spdk_bdev_desc **_desc)
8263 : {
8264 : struct spdk_bdev_desc *desc;
8265 : struct spdk_bdev *bdev;
8266 : int rc;
8267 :
8268 128 : bdev = bdev_get_by_name(bdev_name);
8269 :
8270 128 : if (bdev == NULL) {
8271 1 : SPDK_NOTICELOG("Currently unable to find bdev with name: %s\n", bdev_name);
8272 1 : return -ENODEV;
8273 : }
8274 :
8275 127 : rc = bdev_desc_alloc(bdev, event_cb, event_ctx, &desc);
8276 127 : if (rc != 0) {
8277 0 : return rc;
8278 : }
8279 :
8280 127 : rc = bdev_open(bdev, write, desc);
8281 127 : if (rc != 0) {
8282 7 : bdev_desc_free(desc);
8283 7 : desc = NULL;
8284 7 : }
8285 :
8286 127 : *_desc = desc;
8287 :
8288 127 : return rc;
8289 128 : }
8290 :
8291 : int
8292 130 : spdk_bdev_open_ext(const char *bdev_name, bool write, spdk_bdev_event_cb_t event_cb,
8293 : void *event_ctx, struct spdk_bdev_desc **_desc)
8294 : {
8295 : int rc;
8296 :
8297 130 : if (event_cb == NULL) {
8298 2 : SPDK_ERRLOG("Missing event callback function\n");
8299 2 : return -EINVAL;
8300 : }
8301 :
8302 128 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8303 128 : rc = bdev_open_ext(bdev_name, write, event_cb, event_ctx, _desc);
8304 128 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8305 :
8306 128 : return rc;
8307 130 : }
8308 :
8309 : struct spdk_bdev_open_async_ctx {
8310 : char *bdev_name;
8311 : spdk_bdev_event_cb_t event_cb;
8312 : void *event_ctx;
8313 : bool write;
8314 : int rc;
8315 : spdk_bdev_open_async_cb_t cb_fn;
8316 : void *cb_arg;
8317 : struct spdk_bdev_desc *desc;
8318 : struct spdk_bdev_open_async_opts opts;
8319 : uint64_t start_ticks;
8320 : struct spdk_thread *orig_thread;
8321 : struct spdk_poller *poller;
8322 : TAILQ_ENTRY(spdk_bdev_open_async_ctx) tailq;
8323 : };
8324 :
8325 : static void
8326 0 : bdev_open_async_done(void *arg)
8327 : {
8328 0 : struct spdk_bdev_open_async_ctx *ctx = arg;
8329 :
8330 0 : ctx->cb_fn(ctx->desc, ctx->rc, ctx->cb_arg);
8331 :
8332 0 : free(ctx->bdev_name);
8333 0 : free(ctx);
8334 0 : }
8335 :
8336 : static void
8337 0 : bdev_open_async_cancel(void *arg)
8338 : {
8339 0 : struct spdk_bdev_open_async_ctx *ctx = arg;
8340 :
8341 0 : assert(ctx->rc == -ESHUTDOWN);
8342 :
8343 0 : spdk_poller_unregister(&ctx->poller);
8344 :
8345 0 : bdev_open_async_done(ctx);
8346 0 : }
8347 :
8348 : /* This is called when the bdev library finishes at shutdown. */
8349 : static void
8350 65 : bdev_open_async_fini(void)
8351 : {
8352 : struct spdk_bdev_open_async_ctx *ctx, *tmp_ctx;
8353 :
8354 65 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8355 65 : TAILQ_FOREACH_SAFE(ctx, &g_bdev_mgr.async_bdev_opens, tailq, tmp_ctx) {
8356 0 : TAILQ_REMOVE(&g_bdev_mgr.async_bdev_opens, ctx, tailq);
8357 : /*
8358 : * We have to move to ctx->orig_thread to unregister ctx->poller.
8359 : * However, there is a chance that ctx->poller is executed before
8360 : * message is executed, which could result in bdev_open_async_done()
8361 : * being called twice. To avoid such race condition, set ctx->rc to
8362 : * -ESHUTDOWN.
8363 : */
8364 0 : ctx->rc = -ESHUTDOWN;
8365 0 : spdk_thread_send_msg(ctx->orig_thread, bdev_open_async_cancel, ctx);
8366 0 : }
8367 65 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8368 65 : }
8369 :
8370 : static int bdev_open_async(void *arg);
8371 :
8372 : static void
8373 0 : _bdev_open_async(struct spdk_bdev_open_async_ctx *ctx)
8374 : {
8375 : uint64_t timeout_ticks;
8376 :
8377 0 : if (ctx->rc == -ESHUTDOWN) {
8378 : /* This context is being canceled. Do nothing. */
8379 0 : return;
8380 : }
8381 :
8382 0 : ctx->rc = bdev_open_ext(ctx->bdev_name, ctx->write, ctx->event_cb, ctx->event_ctx,
8383 0 : &ctx->desc);
8384 0 : if (ctx->rc == 0 || ctx->opts.timeout_ms == 0) {
8385 0 : goto exit;
8386 : }
8387 :
8388 0 : timeout_ticks = ctx->start_ticks + ctx->opts.timeout_ms * spdk_get_ticks_hz() / 1000ull;
8389 0 : if (spdk_get_ticks() >= timeout_ticks) {
8390 0 : SPDK_ERRLOG("Timed out while waiting for bdev '%s' to appear\n", ctx->bdev_name);
8391 0 : ctx->rc = -ETIMEDOUT;
8392 0 : goto exit;
8393 : }
8394 :
8395 0 : return;
8396 :
8397 : exit:
8398 0 : spdk_poller_unregister(&ctx->poller);
8399 0 : TAILQ_REMOVE(&g_bdev_mgr.async_bdev_opens, ctx, tailq);
8400 :
8401 : /* Completion callback is processed after stack unwinding. */
8402 0 : spdk_thread_send_msg(ctx->orig_thread, bdev_open_async_done, ctx);
8403 0 : }
8404 :
8405 : static int
8406 0 : bdev_open_async(void *arg)
8407 : {
8408 0 : struct spdk_bdev_open_async_ctx *ctx = arg;
8409 :
8410 0 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8411 :
8412 0 : _bdev_open_async(ctx);
8413 :
8414 0 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8415 :
8416 0 : return SPDK_POLLER_BUSY;
8417 : }
8418 :
8419 : static void
8420 0 : bdev_open_async_opts_copy(struct spdk_bdev_open_async_opts *opts,
8421 : struct spdk_bdev_open_async_opts *opts_src,
8422 : size_t size)
8423 : {
8424 0 : assert(opts);
8425 0 : assert(opts_src);
8426 :
8427 0 : opts->size = size;
8428 :
8429 : #define SET_FIELD(field) \
8430 : if (offsetof(struct spdk_bdev_open_async_opts, field) + sizeof(opts->field) <= size) { \
8431 : opts->field = opts_src->field; \
8432 : } \
8433 :
8434 0 : SET_FIELD(timeout_ms);
8435 :
8436 : /* Do not remove this statement, you should always update this statement when you adding a new field,
8437 : * and do not forget to add the SET_FIELD statement for your added field. */
8438 : SPDK_STATIC_ASSERT(sizeof(struct spdk_bdev_open_async_opts) == 16, "Incorrect size");
8439 :
8440 : #undef SET_FIELD
8441 0 : }
8442 :
8443 : static void
8444 0 : bdev_open_async_opts_get_default(struct spdk_bdev_open_async_opts *opts, size_t size)
8445 : {
8446 0 : assert(opts);
8447 :
8448 0 : opts->size = size;
8449 :
8450 : #define SET_FIELD(field, value) \
8451 : if (offsetof(struct spdk_bdev_open_async_opts, field) + sizeof(opts->field) <= size) { \
8452 : opts->field = value; \
8453 : } \
8454 :
8455 0 : SET_FIELD(timeout_ms, 0);
8456 :
8457 : #undef SET_FIELD
8458 0 : }
8459 :
8460 : int
8461 0 : spdk_bdev_open_async(const char *bdev_name, bool write, spdk_bdev_event_cb_t event_cb,
8462 : void *event_ctx, struct spdk_bdev_open_async_opts *opts,
8463 : spdk_bdev_open_async_cb_t open_cb, void *open_cb_arg)
8464 : {
8465 : struct spdk_bdev_open_async_ctx *ctx;
8466 :
8467 0 : if (event_cb == NULL) {
8468 0 : SPDK_ERRLOG("Missing event callback function\n");
8469 0 : return -EINVAL;
8470 : }
8471 :
8472 0 : if (open_cb == NULL) {
8473 0 : SPDK_ERRLOG("Missing open callback function\n");
8474 0 : return -EINVAL;
8475 : }
8476 :
8477 0 : if (opts != NULL && opts->size == 0) {
8478 0 : SPDK_ERRLOG("size in the options structure should not be zero\n");
8479 0 : return -EINVAL;
8480 : }
8481 :
8482 0 : ctx = calloc(1, sizeof(*ctx));
8483 0 : if (ctx == NULL) {
8484 0 : SPDK_ERRLOG("Failed to allocate open context\n");
8485 0 : return -ENOMEM;
8486 : }
8487 :
8488 0 : ctx->bdev_name = strdup(bdev_name);
8489 0 : if (ctx->bdev_name == NULL) {
8490 0 : SPDK_ERRLOG("Failed to duplicate bdev_name\n");
8491 0 : free(ctx);
8492 0 : return -ENOMEM;
8493 : }
8494 :
8495 0 : ctx->poller = SPDK_POLLER_REGISTER(bdev_open_async, ctx, 100 * 1000);
8496 0 : if (ctx->poller == NULL) {
8497 0 : SPDK_ERRLOG("Failed to register bdev_open_async poller\n");
8498 0 : free(ctx->bdev_name);
8499 0 : free(ctx);
8500 0 : return -ENOMEM;
8501 : }
8502 :
8503 0 : ctx->cb_fn = open_cb;
8504 0 : ctx->cb_arg = open_cb_arg;
8505 0 : ctx->write = write;
8506 0 : ctx->event_cb = event_cb;
8507 0 : ctx->event_ctx = event_ctx;
8508 0 : ctx->orig_thread = spdk_get_thread();
8509 0 : ctx->start_ticks = spdk_get_ticks();
8510 :
8511 0 : bdev_open_async_opts_get_default(&ctx->opts, sizeof(ctx->opts));
8512 0 : if (opts != NULL) {
8513 0 : bdev_open_async_opts_copy(&ctx->opts, opts, opts->size);
8514 0 : }
8515 :
8516 0 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8517 :
8518 0 : TAILQ_INSERT_TAIL(&g_bdev_mgr.async_bdev_opens, ctx, tailq);
8519 0 : _bdev_open_async(ctx);
8520 :
8521 0 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8522 :
8523 0 : return 0;
8524 0 : }
8525 :
8526 : static void
8527 255 : bdev_close(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc)
8528 : {
8529 : int rc;
8530 :
8531 255 : spdk_spin_lock(&bdev->internal.spinlock);
8532 255 : spdk_spin_lock(&desc->spinlock);
8533 :
8534 255 : TAILQ_REMOVE(&bdev->internal.open_descs, desc, link);
8535 :
8536 255 : desc->closed = true;
8537 :
8538 255 : if (desc->claim != NULL) {
8539 16 : bdev_desc_release_claims(desc);
8540 16 : }
8541 :
8542 255 : if (0 == desc->refs) {
8543 244 : spdk_spin_unlock(&desc->spinlock);
8544 244 : bdev_desc_free(desc);
8545 244 : } else {
8546 11 : spdk_spin_unlock(&desc->spinlock);
8547 : }
8548 :
8549 : /* If no more descriptors, kill QoS channel */
8550 255 : if (bdev->internal.qos && TAILQ_EMPTY(&bdev->internal.open_descs)) {
8551 7 : SPDK_DEBUGLOG(bdev, "Closed last descriptor for bdev %s on thread %p. Stopping QoS.\n",
8552 : bdev->name, spdk_get_thread());
8553 :
8554 7 : if (bdev_qos_destroy(bdev)) {
8555 : /* There isn't anything we can do to recover here. Just let the
8556 : * old QoS poller keep running. The QoS handling won't change
8557 : * cores when the user allocates a new channel, but it won't break. */
8558 0 : SPDK_ERRLOG("Unable to shut down QoS poller. It will continue running on the current thread.\n");
8559 0 : }
8560 7 : }
8561 :
8562 255 : if (bdev->internal.status == SPDK_BDEV_STATUS_REMOVING && TAILQ_EMPTY(&bdev->internal.open_descs)) {
8563 15 : rc = bdev_unregister_unsafe(bdev);
8564 15 : spdk_spin_unlock(&bdev->internal.spinlock);
8565 :
8566 15 : if (rc == 0) {
8567 15 : spdk_io_device_unregister(__bdev_to_io_dev(bdev), bdev_destroy_cb);
8568 15 : }
8569 15 : } else {
8570 240 : spdk_spin_unlock(&bdev->internal.spinlock);
8571 : }
8572 255 : }
8573 :
8574 : void
8575 120 : spdk_bdev_close(struct spdk_bdev_desc *desc)
8576 : {
8577 120 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
8578 :
8579 120 : SPDK_DEBUGLOG(bdev, "Closing descriptor %p for bdev %s on thread %p\n", desc, bdev->name,
8580 : spdk_get_thread());
8581 :
8582 120 : assert(desc->thread == spdk_get_thread());
8583 :
8584 120 : spdk_poller_unregister(&desc->io_timeout_poller);
8585 :
8586 120 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8587 :
8588 120 : bdev_close(bdev, desc);
8589 :
8590 120 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8591 120 : }
8592 :
8593 : int32_t
8594 3 : spdk_bdev_get_numa_id(struct spdk_bdev *bdev)
8595 : {
8596 3 : if (bdev->numa.id_valid) {
8597 2 : return bdev->numa.id;
8598 : } else {
8599 1 : return SPDK_ENV_NUMA_ID_ANY;
8600 : }
8601 3 : }
8602 :
8603 : static void
8604 124 : bdev_register_finished(void *arg)
8605 : {
8606 124 : struct spdk_bdev_desc *desc = arg;
8607 124 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
8608 :
8609 124 : spdk_notify_send("bdev_register", spdk_bdev_get_name(bdev));
8610 :
8611 124 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8612 :
8613 124 : bdev_close(bdev, desc);
8614 :
8615 124 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8616 124 : }
8617 :
8618 : int
8619 127 : spdk_bdev_register(struct spdk_bdev *bdev)
8620 : {
8621 : struct spdk_bdev_desc *desc;
8622 127 : struct spdk_thread *thread = spdk_get_thread();
8623 : int rc;
8624 :
8625 127 : if (spdk_unlikely(!spdk_thread_is_app_thread(NULL))) {
8626 1 : SPDK_ERRLOG("Cannot register bdev %s on thread %p (%s)\n", bdev->name, thread,
8627 : thread ? spdk_thread_get_name(thread) : "null");
8628 1 : return -EINVAL;
8629 : }
8630 :
8631 126 : rc = bdev_register(bdev);
8632 126 : if (rc != 0) {
8633 2 : return rc;
8634 : }
8635 :
8636 : /* A descriptor is opened to prevent bdev deletion during examination */
8637 124 : rc = bdev_desc_alloc(bdev, _tmp_bdev_event_cb, NULL, &desc);
8638 124 : if (rc != 0) {
8639 0 : spdk_bdev_unregister(bdev, NULL, NULL);
8640 0 : return rc;
8641 : }
8642 :
8643 124 : rc = bdev_open(bdev, false, desc);
8644 124 : if (rc != 0) {
8645 0 : bdev_desc_free(desc);
8646 0 : spdk_bdev_unregister(bdev, NULL, NULL);
8647 0 : return rc;
8648 : }
8649 :
8650 : /* Examine configuration before initializing I/O */
8651 124 : bdev_examine(bdev);
8652 :
8653 124 : rc = spdk_bdev_wait_for_examine(bdev_register_finished, desc);
8654 124 : if (rc != 0) {
8655 0 : bdev_close(bdev, desc);
8656 0 : spdk_bdev_unregister(bdev, NULL, NULL);
8657 0 : }
8658 :
8659 124 : return rc;
8660 127 : }
8661 :
8662 : int
8663 26 : spdk_bdev_module_claim_bdev(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc,
8664 : struct spdk_bdev_module *module)
8665 : {
8666 26 : spdk_spin_lock(&bdev->internal.spinlock);
8667 :
8668 26 : if (bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE) {
8669 6 : LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
8670 6 : spdk_spin_unlock(&bdev->internal.spinlock);
8671 6 : return -EPERM;
8672 : }
8673 :
8674 20 : if (desc && !desc->write) {
8675 5 : desc->write = true;
8676 5 : }
8677 :
8678 20 : bdev->internal.claim_type = SPDK_BDEV_CLAIM_EXCL_WRITE;
8679 20 : bdev->internal.claim.v1.module = module;
8680 :
8681 20 : spdk_spin_unlock(&bdev->internal.spinlock);
8682 20 : return 0;
8683 26 : }
8684 :
8685 : void
8686 8 : spdk_bdev_module_release_bdev(struct spdk_bdev *bdev)
8687 : {
8688 8 : spdk_spin_lock(&bdev->internal.spinlock);
8689 :
8690 8 : assert(bdev->internal.claim.v1.module != NULL);
8691 8 : assert(bdev->internal.claim_type == SPDK_BDEV_CLAIM_EXCL_WRITE);
8692 8 : bdev->internal.claim_type = SPDK_BDEV_CLAIM_NONE;
8693 8 : bdev->internal.claim.v1.module = NULL;
8694 :
8695 8 : spdk_spin_unlock(&bdev->internal.spinlock);
8696 8 : }
8697 :
8698 : /*
8699 : * Start claims v2
8700 : */
8701 :
8702 : const char *
8703 24 : spdk_bdev_claim_get_name(enum spdk_bdev_claim_type type)
8704 : {
8705 24 : switch (type) {
8706 : case SPDK_BDEV_CLAIM_NONE:
8707 0 : return "not_claimed";
8708 : case SPDK_BDEV_CLAIM_EXCL_WRITE:
8709 6 : return "exclusive_write";
8710 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE:
8711 7 : return "read_many_write_one";
8712 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_NONE:
8713 5 : return "read_many_write_none";
8714 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
8715 6 : return "read_many_write_many";
8716 : default:
8717 0 : break;
8718 : }
8719 0 : return "invalid_claim";
8720 24 : }
8721 :
8722 : static bool
8723 96 : claim_type_is_v2(enum spdk_bdev_claim_type type)
8724 : {
8725 96 : switch (type) {
8726 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE:
8727 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_NONE:
8728 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
8729 96 : return true;
8730 : default:
8731 0 : break;
8732 : }
8733 0 : return false;
8734 96 : }
8735 :
8736 : /* Returns true if taking a claim with desc->write == false should make the descriptor writable. */
8737 : static bool
8738 13 : claim_type_promotes_to_write(enum spdk_bdev_claim_type type)
8739 : {
8740 13 : switch (type) {
8741 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE:
8742 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
8743 5 : return true;
8744 : default:
8745 8 : break;
8746 : }
8747 8 : return false;
8748 13 : }
8749 :
8750 : void
8751 44 : spdk_bdev_claim_opts_init(struct spdk_bdev_claim_opts *opts, size_t size)
8752 : {
8753 44 : if (opts == NULL) {
8754 0 : SPDK_ERRLOG("opts should not be NULL\n");
8755 0 : assert(opts != NULL);
8756 0 : return;
8757 : }
8758 44 : if (size == 0) {
8759 0 : SPDK_ERRLOG("size should not be zero\n");
8760 0 : assert(size != 0);
8761 0 : return;
8762 : }
8763 :
8764 44 : memset(opts, 0, size);
8765 44 : opts->opts_size = size;
8766 :
8767 : #define FIELD_OK(field) \
8768 : offsetof(struct spdk_bdev_claim_opts, field) + sizeof(opts->field) <= size
8769 :
8770 : #define SET_FIELD(field, value) \
8771 : if (FIELD_OK(field)) { \
8772 : opts->field = value; \
8773 : } \
8774 :
8775 44 : SET_FIELD(shared_claim_key, 0);
8776 :
8777 : #undef FIELD_OK
8778 : #undef SET_FIELD
8779 44 : }
8780 :
8781 : static int
8782 22 : claim_opts_copy(struct spdk_bdev_claim_opts *src, struct spdk_bdev_claim_opts *dst)
8783 : {
8784 22 : if (src->opts_size == 0) {
8785 0 : SPDK_ERRLOG("size should not be zero\n");
8786 0 : return -1;
8787 : }
8788 :
8789 22 : memset(dst, 0, sizeof(*dst));
8790 22 : dst->opts_size = src->opts_size;
8791 :
8792 : #define FIELD_OK(field) \
8793 : offsetof(struct spdk_bdev_claim_opts, field) + sizeof(src->field) <= src->opts_size
8794 :
8795 : #define SET_FIELD(field) \
8796 : if (FIELD_OK(field)) { \
8797 : dst->field = src->field; \
8798 : } \
8799 :
8800 22 : if (FIELD_OK(name)) {
8801 22 : snprintf(dst->name, sizeof(dst->name), "%s", src->name);
8802 22 : }
8803 :
8804 22 : SET_FIELD(shared_claim_key);
8805 :
8806 : /* You should not remove this statement, but need to update the assert statement
8807 : * if you add a new field, and also add a corresponding SET_FIELD statement */
8808 : SPDK_STATIC_ASSERT(sizeof(struct spdk_bdev_claim_opts) == 48, "Incorrect size");
8809 :
8810 : #undef FIELD_OK
8811 : #undef SET_FIELD
8812 22 : return 0;
8813 22 : }
8814 :
8815 : /* Returns 0 if a read-write-once claim can be taken. */
8816 : static int
8817 9 : claim_verify_rwo(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
8818 : struct spdk_bdev_claim_opts *opts, struct spdk_bdev_module *module)
8819 : {
8820 9 : struct spdk_bdev *bdev = desc->bdev;
8821 : struct spdk_bdev_desc *open_desc;
8822 :
8823 9 : assert(spdk_spin_held(&bdev->internal.spinlock));
8824 9 : assert(type == SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE);
8825 :
8826 9 : if (opts->shared_claim_key != 0) {
8827 1 : SPDK_ERRLOG("%s: key option not supported with read-write-once claims\n",
8828 : bdev->name);
8829 1 : return -EINVAL;
8830 : }
8831 8 : if (bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE) {
8832 1 : LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
8833 1 : return -EPERM;
8834 : }
8835 7 : if (desc->claim != NULL) {
8836 0 : SPDK_NOTICELOG("%s: descriptor already claimed bdev with module %s\n",
8837 : bdev->name, desc->claim->module->name);
8838 0 : return -EPERM;
8839 : }
8840 14 : TAILQ_FOREACH(open_desc, &bdev->internal.open_descs, link) {
8841 9 : if (desc != open_desc && open_desc->write) {
8842 2 : SPDK_NOTICELOG("%s: Cannot obtain read-write-once claim while "
8843 : "another descriptor is open for writing\n",
8844 : bdev->name);
8845 2 : return -EPERM;
8846 : }
8847 7 : }
8848 :
8849 5 : return 0;
8850 9 : }
8851 :
8852 : /* Returns 0 if a read-only-many claim can be taken. */
8853 : static int
8854 12 : claim_verify_rom(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
8855 : struct spdk_bdev_claim_opts *opts, struct spdk_bdev_module *module)
8856 : {
8857 12 : struct spdk_bdev *bdev = desc->bdev;
8858 : struct spdk_bdev_desc *open_desc;
8859 :
8860 12 : assert(spdk_spin_held(&bdev->internal.spinlock));
8861 12 : assert(type == SPDK_BDEV_CLAIM_READ_MANY_WRITE_NONE);
8862 12 : assert(desc->claim == NULL);
8863 :
8864 12 : if (desc->write) {
8865 3 : SPDK_ERRLOG("%s: Cannot obtain read-only-many claim with writable descriptor\n",
8866 : bdev->name);
8867 3 : return -EINVAL;
8868 : }
8869 9 : if (opts->shared_claim_key != 0) {
8870 1 : SPDK_ERRLOG("%s: key option not supported with read-only-may claims\n", bdev->name);
8871 1 : return -EINVAL;
8872 : }
8873 8 : if (bdev->internal.claim_type == SPDK_BDEV_CLAIM_NONE) {
8874 15 : TAILQ_FOREACH(open_desc, &bdev->internal.open_descs, link) {
8875 9 : if (open_desc->write) {
8876 0 : SPDK_NOTICELOG("%s: Cannot obtain read-only-many claim while "
8877 : "another descriptor is open for writing\n",
8878 : bdev->name);
8879 0 : return -EPERM;
8880 : }
8881 9 : }
8882 6 : }
8883 :
8884 8 : return 0;
8885 12 : }
8886 :
8887 : /* Returns 0 if a read-write-many claim can be taken. */
8888 : static int
8889 8 : claim_verify_rwm(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
8890 : struct spdk_bdev_claim_opts *opts, struct spdk_bdev_module *module)
8891 : {
8892 8 : struct spdk_bdev *bdev = desc->bdev;
8893 : struct spdk_bdev_desc *open_desc;
8894 :
8895 8 : assert(spdk_spin_held(&bdev->internal.spinlock));
8896 8 : assert(type == SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED);
8897 8 : assert(desc->claim == NULL);
8898 :
8899 8 : if (opts->shared_claim_key == 0) {
8900 2 : SPDK_ERRLOG("%s: shared_claim_key option required with read-write-may claims\n",
8901 : bdev->name);
8902 2 : return -EINVAL;
8903 : }
8904 6 : switch (bdev->internal.claim_type) {
8905 : case SPDK_BDEV_CLAIM_NONE:
8906 7 : TAILQ_FOREACH(open_desc, &bdev->internal.open_descs, link) {
8907 5 : if (open_desc == desc) {
8908 3 : continue;
8909 : }
8910 2 : if (open_desc->write) {
8911 2 : SPDK_NOTICELOG("%s: Cannot obtain read-write-many claim while "
8912 : "another descriptor is open for writing without a "
8913 : "claim\n", bdev->name);
8914 2 : return -EPERM;
8915 : }
8916 0 : }
8917 2 : break;
8918 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
8919 2 : if (opts->shared_claim_key != bdev->internal.claim.v2.key) {
8920 1 : LOG_ALREADY_CLAIMED_ERROR("already claimed with another key", bdev);
8921 1 : return -EPERM;
8922 : }
8923 1 : break;
8924 : default:
8925 0 : LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
8926 0 : return -EBUSY;
8927 : }
8928 :
8929 3 : return 0;
8930 8 : }
8931 :
8932 : /* Updates desc and its bdev with a v2 claim. */
8933 : static int
8934 16 : claim_bdev(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
8935 : struct spdk_bdev_claim_opts *opts, struct spdk_bdev_module *module)
8936 : {
8937 16 : struct spdk_bdev *bdev = desc->bdev;
8938 : struct spdk_bdev_module_claim *claim;
8939 :
8940 16 : assert(spdk_spin_held(&bdev->internal.spinlock));
8941 16 : assert(claim_type_is_v2(type));
8942 16 : assert(desc->claim == NULL);
8943 :
8944 16 : claim = calloc(1, sizeof(*desc->claim));
8945 16 : if (claim == NULL) {
8946 0 : SPDK_ERRLOG("%s: out of memory while allocating claim\n", bdev->name);
8947 0 : return -ENOMEM;
8948 : }
8949 16 : claim->module = module;
8950 16 : claim->desc = desc;
8951 : SPDK_STATIC_ASSERT(sizeof(claim->name) == sizeof(opts->name), "sizes must match");
8952 16 : memcpy(claim->name, opts->name, sizeof(claim->name));
8953 16 : desc->claim = claim;
8954 :
8955 16 : if (bdev->internal.claim_type == SPDK_BDEV_CLAIM_NONE) {
8956 13 : bdev->internal.claim_type = type;
8957 13 : TAILQ_INIT(&bdev->internal.claim.v2.claims);
8958 13 : bdev->internal.claim.v2.key = opts->shared_claim_key;
8959 13 : }
8960 16 : assert(type == bdev->internal.claim_type);
8961 :
8962 16 : TAILQ_INSERT_TAIL(&bdev->internal.claim.v2.claims, claim, link);
8963 :
8964 16 : if (!desc->write && claim_type_promotes_to_write(type)) {
8965 5 : desc->write = true;
8966 5 : }
8967 :
8968 16 : return 0;
8969 16 : }
8970 :
8971 : int
8972 39 : spdk_bdev_module_claim_bdev_desc(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
8973 : struct spdk_bdev_claim_opts *_opts,
8974 : struct spdk_bdev_module *module)
8975 : {
8976 : struct spdk_bdev *bdev;
8977 : struct spdk_bdev_claim_opts opts;
8978 39 : int rc = 0;
8979 :
8980 39 : if (desc == NULL) {
8981 0 : SPDK_ERRLOG("descriptor must not be NULL\n");
8982 0 : return -EINVAL;
8983 : }
8984 :
8985 39 : bdev = desc->bdev;
8986 :
8987 39 : if (_opts == NULL) {
8988 17 : spdk_bdev_claim_opts_init(&opts, sizeof(opts));
8989 39 : } else if (claim_opts_copy(_opts, &opts) != 0) {
8990 0 : return -EINVAL;
8991 : }
8992 :
8993 39 : spdk_spin_lock(&bdev->internal.spinlock);
8994 :
8995 39 : if (bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE &&
8996 15 : bdev->internal.claim_type != type) {
8997 10 : LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
8998 10 : spdk_spin_unlock(&bdev->internal.spinlock);
8999 10 : return -EPERM;
9000 : }
9001 :
9002 29 : if (claim_type_is_v2(type) && desc->claim != NULL) {
9003 0 : SPDK_ERRLOG("%s: descriptor already has %s claim with name '%s'\n",
9004 : bdev->name, spdk_bdev_claim_get_name(type), desc->claim->name);
9005 0 : spdk_spin_unlock(&bdev->internal.spinlock);
9006 0 : return -EPERM;
9007 : }
9008 :
9009 29 : switch (type) {
9010 : case SPDK_BDEV_CLAIM_EXCL_WRITE:
9011 0 : spdk_spin_unlock(&bdev->internal.spinlock);
9012 0 : return spdk_bdev_module_claim_bdev(bdev, desc, module);
9013 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE:
9014 9 : rc = claim_verify_rwo(desc, type, &opts, module);
9015 9 : break;
9016 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_NONE:
9017 12 : rc = claim_verify_rom(desc, type, &opts, module);
9018 12 : break;
9019 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
9020 8 : rc = claim_verify_rwm(desc, type, &opts, module);
9021 8 : break;
9022 : default:
9023 0 : SPDK_ERRLOG("%s: claim type %d not supported\n", bdev->name, type);
9024 0 : rc = -ENOTSUP;
9025 0 : }
9026 :
9027 29 : if (rc == 0) {
9028 16 : rc = claim_bdev(desc, type, &opts, module);
9029 16 : }
9030 :
9031 29 : spdk_spin_unlock(&bdev->internal.spinlock);
9032 29 : return rc;
9033 39 : }
9034 :
9035 : static void
9036 13 : claim_reset(struct spdk_bdev *bdev)
9037 : {
9038 13 : assert(spdk_spin_held(&bdev->internal.spinlock));
9039 13 : assert(claim_type_is_v2(bdev->internal.claim_type));
9040 13 : assert(TAILQ_EMPTY(&bdev->internal.claim.v2.claims));
9041 :
9042 13 : memset(&bdev->internal.claim, 0, sizeof(bdev->internal.claim));
9043 13 : bdev->internal.claim_type = SPDK_BDEV_CLAIM_NONE;
9044 13 : }
9045 :
9046 : static void
9047 16 : bdev_desc_release_claims(struct spdk_bdev_desc *desc)
9048 : {
9049 16 : struct spdk_bdev *bdev = desc->bdev;
9050 :
9051 16 : assert(spdk_spin_held(&bdev->internal.spinlock));
9052 16 : assert(claim_type_is_v2(bdev->internal.claim_type));
9053 :
9054 16 : if (bdev->internal.examine_in_progress == 0) {
9055 16 : TAILQ_REMOVE(&bdev->internal.claim.v2.claims, desc->claim, link);
9056 16 : free(desc->claim);
9057 16 : if (TAILQ_EMPTY(&bdev->internal.claim.v2.claims)) {
9058 13 : claim_reset(bdev);
9059 13 : }
9060 16 : } else {
9061 : /* This is a dead claim that will be cleaned up when bdev_examine() is done. */
9062 0 : desc->claim->module = NULL;
9063 0 : desc->claim->desc = NULL;
9064 : }
9065 16 : desc->claim = NULL;
9066 16 : }
9067 :
9068 : /*
9069 : * End claims v2
9070 : */
9071 :
9072 : struct spdk_bdev *
9073 1178 : spdk_bdev_desc_get_bdev(struct spdk_bdev_desc *desc)
9074 : {
9075 1178 : assert(desc != NULL);
9076 1178 : return desc->bdev;
9077 : }
9078 :
9079 : int
9080 1 : spdk_for_each_bdev(void *ctx, spdk_for_each_bdev_fn fn)
9081 : {
9082 : struct spdk_bdev *bdev, *tmp;
9083 : struct spdk_bdev_desc *desc;
9084 1 : int rc = 0;
9085 :
9086 1 : assert(fn != NULL);
9087 :
9088 1 : spdk_spin_lock(&g_bdev_mgr.spinlock);
9089 1 : bdev = spdk_bdev_first();
9090 9 : while (bdev != NULL) {
9091 8 : rc = bdev_desc_alloc(bdev, _tmp_bdev_event_cb, NULL, &desc);
9092 8 : if (rc != 0) {
9093 0 : break;
9094 : }
9095 8 : rc = bdev_open(bdev, false, desc);
9096 8 : if (rc != 0) {
9097 1 : bdev_desc_free(desc);
9098 1 : if (rc == -ENODEV) {
9099 : /* Ignore the error and move to the next bdev. */
9100 1 : rc = 0;
9101 1 : bdev = spdk_bdev_next(bdev);
9102 1 : continue;
9103 : }
9104 0 : break;
9105 : }
9106 7 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
9107 :
9108 7 : rc = fn(ctx, bdev);
9109 :
9110 7 : spdk_spin_lock(&g_bdev_mgr.spinlock);
9111 7 : tmp = spdk_bdev_next(bdev);
9112 7 : bdev_close(bdev, desc);
9113 7 : if (rc != 0) {
9114 0 : break;
9115 : }
9116 7 : bdev = tmp;
9117 : }
9118 1 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
9119 :
9120 1 : return rc;
9121 : }
9122 :
9123 : int
9124 1 : spdk_for_each_bdev_leaf(void *ctx, spdk_for_each_bdev_fn fn)
9125 : {
9126 : struct spdk_bdev *bdev, *tmp;
9127 : struct spdk_bdev_desc *desc;
9128 1 : int rc = 0;
9129 :
9130 1 : assert(fn != NULL);
9131 :
9132 1 : spdk_spin_lock(&g_bdev_mgr.spinlock);
9133 1 : bdev = spdk_bdev_first_leaf();
9134 6 : while (bdev != NULL) {
9135 5 : rc = bdev_desc_alloc(bdev, _tmp_bdev_event_cb, NULL, &desc);
9136 5 : if (rc != 0) {
9137 0 : break;
9138 : }
9139 5 : rc = bdev_open(bdev, false, desc);
9140 5 : if (rc != 0) {
9141 1 : bdev_desc_free(desc);
9142 1 : if (rc == -ENODEV) {
9143 : /* Ignore the error and move to the next bdev. */
9144 1 : rc = 0;
9145 1 : bdev = spdk_bdev_next_leaf(bdev);
9146 1 : continue;
9147 : }
9148 0 : break;
9149 : }
9150 4 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
9151 :
9152 4 : rc = fn(ctx, bdev);
9153 :
9154 4 : spdk_spin_lock(&g_bdev_mgr.spinlock);
9155 4 : tmp = spdk_bdev_next_leaf(bdev);
9156 4 : bdev_close(bdev, desc);
9157 4 : if (rc != 0) {
9158 0 : break;
9159 : }
9160 4 : bdev = tmp;
9161 : }
9162 1 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
9163 :
9164 1 : return rc;
9165 : }
9166 :
9167 : void
9168 0 : spdk_bdev_io_get_iovec(struct spdk_bdev_io *bdev_io, struct iovec **iovp, int *iovcntp)
9169 : {
9170 : struct iovec *iovs;
9171 : int iovcnt;
9172 :
9173 0 : if (bdev_io == NULL) {
9174 0 : return;
9175 : }
9176 :
9177 0 : switch (bdev_io->type) {
9178 : case SPDK_BDEV_IO_TYPE_READ:
9179 : case SPDK_BDEV_IO_TYPE_WRITE:
9180 : case SPDK_BDEV_IO_TYPE_ZCOPY:
9181 0 : iovs = bdev_io->u.bdev.iovs;
9182 0 : iovcnt = bdev_io->u.bdev.iovcnt;
9183 0 : break;
9184 : default:
9185 0 : iovs = NULL;
9186 0 : iovcnt = 0;
9187 0 : break;
9188 : }
9189 :
9190 0 : if (iovp) {
9191 0 : *iovp = iovs;
9192 0 : }
9193 0 : if (iovcntp) {
9194 0 : *iovcntp = iovcnt;
9195 0 : }
9196 0 : }
9197 :
9198 : void *
9199 0 : spdk_bdev_io_get_md_buf(struct spdk_bdev_io *bdev_io)
9200 : {
9201 0 : if (bdev_io == NULL) {
9202 0 : return NULL;
9203 : }
9204 :
9205 0 : if (!spdk_bdev_is_md_separate(bdev_io->bdev)) {
9206 0 : return NULL;
9207 : }
9208 :
9209 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_READ ||
9210 0 : bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
9211 0 : return bdev_io->u.bdev.md_buf;
9212 : }
9213 :
9214 0 : return NULL;
9215 0 : }
9216 :
9217 : void *
9218 0 : spdk_bdev_io_get_cb_arg(struct spdk_bdev_io *bdev_io)
9219 : {
9220 0 : if (bdev_io == NULL) {
9221 0 : assert(false);
9222 : return NULL;
9223 : }
9224 :
9225 0 : return bdev_io->internal.caller_ctx;
9226 : }
9227 :
9228 : void
9229 7 : spdk_bdev_module_list_add(struct spdk_bdev_module *bdev_module)
9230 : {
9231 :
9232 7 : if (spdk_bdev_module_list_find(bdev_module->name)) {
9233 0 : SPDK_ERRLOG("ERROR: module '%s' already registered.\n", bdev_module->name);
9234 0 : assert(false);
9235 : }
9236 :
9237 7 : spdk_spin_init(&bdev_module->internal.spinlock);
9238 7 : TAILQ_INIT(&bdev_module->internal.quiesced_ranges);
9239 :
9240 : /*
9241 : * Modules with examine callbacks must be initialized first, so they are
9242 : * ready to handle examine callbacks from later modules that will
9243 : * register physical bdevs.
9244 : */
9245 7 : if (bdev_module->examine_config != NULL || bdev_module->examine_disk != NULL) {
9246 4 : TAILQ_INSERT_HEAD(&g_bdev_mgr.bdev_modules, bdev_module, internal.tailq);
9247 4 : } else {
9248 3 : TAILQ_INSERT_TAIL(&g_bdev_mgr.bdev_modules, bdev_module, internal.tailq);
9249 : }
9250 7 : }
9251 :
9252 : struct spdk_bdev_module *
9253 7 : spdk_bdev_module_list_find(const char *name)
9254 : {
9255 : struct spdk_bdev_module *bdev_module;
9256 :
9257 14 : TAILQ_FOREACH(bdev_module, &g_bdev_mgr.bdev_modules, internal.tailq) {
9258 7 : if (strcmp(name, bdev_module->name) == 0) {
9259 0 : break;
9260 : }
9261 7 : }
9262 :
9263 7 : return bdev_module;
9264 : }
9265 :
9266 : static int
9267 6 : bdev_write_zero_buffer(struct spdk_bdev_io *bdev_io)
9268 : {
9269 : uint64_t num_blocks;
9270 6 : void *md_buf = NULL;
9271 :
9272 6 : num_blocks = bdev_io->u.bdev.num_blocks;
9273 :
9274 6 : if (spdk_bdev_is_md_separate(bdev_io->bdev)) {
9275 4 : md_buf = (char *)g_bdev_mgr.zero_buffer +
9276 2 : spdk_bdev_get_block_size(bdev_io->bdev) * num_blocks;
9277 2 : }
9278 :
9279 12 : return bdev_write_blocks_with_md(bdev_io->internal.desc,
9280 6 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
9281 6 : g_bdev_mgr.zero_buffer, md_buf,
9282 6 : bdev_io->u.bdev.offset_blocks, num_blocks,
9283 6 : bdev_write_zero_buffer_done, bdev_io);
9284 : }
9285 :
9286 : static void
9287 6 : bdev_write_zero_buffer_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
9288 : {
9289 6 : struct spdk_bdev_io *parent_io = cb_arg;
9290 :
9291 6 : spdk_bdev_free_io(bdev_io);
9292 :
9293 6 : parent_io->internal.status = success ? SPDK_BDEV_IO_STATUS_SUCCESS : SPDK_BDEV_IO_STATUS_FAILED;
9294 6 : parent_io->internal.cb(parent_io, success, parent_io->internal.caller_ctx);
9295 6 : }
9296 :
9297 : static void
9298 10 : bdev_set_qos_limit_done(struct set_qos_limit_ctx *ctx, int status)
9299 : {
9300 10 : spdk_spin_lock(&ctx->bdev->internal.spinlock);
9301 10 : ctx->bdev->internal.qos_mod_in_progress = false;
9302 10 : spdk_spin_unlock(&ctx->bdev->internal.spinlock);
9303 :
9304 10 : if (ctx->cb_fn) {
9305 8 : ctx->cb_fn(ctx->cb_arg, status);
9306 8 : }
9307 10 : free(ctx);
9308 10 : }
9309 :
9310 : static void
9311 2 : bdev_disable_qos_done(void *cb_arg)
9312 : {
9313 2 : struct set_qos_limit_ctx *ctx = cb_arg;
9314 2 : struct spdk_bdev *bdev = ctx->bdev;
9315 : struct spdk_bdev_qos *qos;
9316 :
9317 2 : spdk_spin_lock(&bdev->internal.spinlock);
9318 2 : qos = bdev->internal.qos;
9319 2 : bdev->internal.qos = NULL;
9320 2 : spdk_spin_unlock(&bdev->internal.spinlock);
9321 :
9322 2 : if (qos->thread != NULL) {
9323 2 : spdk_put_io_channel(spdk_io_channel_from_ctx(qos->ch));
9324 2 : spdk_poller_unregister(&qos->poller);
9325 2 : }
9326 :
9327 2 : free(qos);
9328 :
9329 2 : bdev_set_qos_limit_done(ctx, 0);
9330 2 : }
9331 :
9332 : static void
9333 2 : bdev_disable_qos_msg_done(struct spdk_bdev *bdev, void *_ctx, int status)
9334 : {
9335 2 : struct set_qos_limit_ctx *ctx = _ctx;
9336 : struct spdk_thread *thread;
9337 :
9338 2 : spdk_spin_lock(&bdev->internal.spinlock);
9339 2 : thread = bdev->internal.qos->thread;
9340 2 : spdk_spin_unlock(&bdev->internal.spinlock);
9341 :
9342 2 : if (thread != NULL) {
9343 2 : spdk_thread_send_msg(thread, bdev_disable_qos_done, ctx);
9344 2 : } else {
9345 0 : bdev_disable_qos_done(ctx);
9346 : }
9347 2 : }
9348 :
9349 : static void
9350 4 : bdev_disable_qos_msg(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
9351 : struct spdk_io_channel *ch, void *_ctx)
9352 : {
9353 4 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(ch);
9354 : struct spdk_bdev_io *bdev_io;
9355 :
9356 4 : bdev_ch->flags &= ~BDEV_CH_QOS_ENABLED;
9357 :
9358 6 : while (!TAILQ_EMPTY(&bdev_ch->qos_queued_io)) {
9359 : /* Re-submit the queued I/O. */
9360 2 : bdev_io = TAILQ_FIRST(&bdev_ch->qos_queued_io);
9361 2 : TAILQ_REMOVE(&bdev_ch->qos_queued_io, bdev_io, internal.link);
9362 2 : _bdev_io_submit(bdev_io);
9363 : }
9364 :
9365 4 : spdk_bdev_for_each_channel_continue(i, 0);
9366 4 : }
9367 :
9368 : static void
9369 1 : bdev_update_qos_rate_limit_msg(void *cb_arg)
9370 : {
9371 1 : struct set_qos_limit_ctx *ctx = cb_arg;
9372 1 : struct spdk_bdev *bdev = ctx->bdev;
9373 :
9374 1 : spdk_spin_lock(&bdev->internal.spinlock);
9375 1 : bdev_qos_update_max_quota_per_timeslice(bdev->internal.qos);
9376 1 : spdk_spin_unlock(&bdev->internal.spinlock);
9377 :
9378 1 : bdev_set_qos_limit_done(ctx, 0);
9379 1 : }
9380 :
9381 : static void
9382 9 : bdev_enable_qos_msg(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
9383 : struct spdk_io_channel *ch, void *_ctx)
9384 : {
9385 9 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(ch);
9386 :
9387 9 : spdk_spin_lock(&bdev->internal.spinlock);
9388 9 : bdev_enable_qos(bdev, bdev_ch);
9389 9 : spdk_spin_unlock(&bdev->internal.spinlock);
9390 9 : spdk_bdev_for_each_channel_continue(i, 0);
9391 9 : }
9392 :
9393 : static void
9394 6 : bdev_enable_qos_done(struct spdk_bdev *bdev, void *_ctx, int status)
9395 : {
9396 6 : struct set_qos_limit_ctx *ctx = _ctx;
9397 :
9398 6 : bdev_set_qos_limit_done(ctx, status);
9399 6 : }
9400 :
9401 : static void
9402 7 : bdev_set_qos_rate_limits(struct spdk_bdev *bdev, uint64_t *limits)
9403 : {
9404 : int i;
9405 :
9406 7 : assert(bdev->internal.qos != NULL);
9407 :
9408 35 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
9409 28 : if (limits[i] != SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
9410 28 : bdev->internal.qos->rate_limits[i].limit = limits[i];
9411 :
9412 28 : if (limits[i] == 0) {
9413 19 : bdev->internal.qos->rate_limits[i].limit =
9414 : SPDK_BDEV_QOS_LIMIT_NOT_DEFINED;
9415 19 : }
9416 28 : }
9417 28 : }
9418 7 : }
9419 :
9420 : void
9421 9 : spdk_bdev_set_qos_rate_limits(struct spdk_bdev *bdev, uint64_t *limits,
9422 : void (*cb_fn)(void *cb_arg, int status), void *cb_arg)
9423 : {
9424 : struct set_qos_limit_ctx *ctx;
9425 : uint32_t limit_set_complement;
9426 : uint64_t min_limit_per_sec;
9427 : int i;
9428 9 : bool disable_rate_limit = true;
9429 :
9430 45 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
9431 36 : if (limits[i] == SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
9432 0 : continue;
9433 : }
9434 :
9435 36 : if (limits[i] > 0) {
9436 10 : disable_rate_limit = false;
9437 10 : }
9438 :
9439 36 : if (bdev_qos_is_iops_rate_limit(i) == true) {
9440 9 : min_limit_per_sec = SPDK_BDEV_QOS_MIN_IOS_PER_SEC;
9441 9 : } else {
9442 27 : if (limits[i] > SPDK_BDEV_QOS_MAX_MBYTES_PER_SEC) {
9443 0 : SPDK_WARNLOG("Requested rate limit %" PRIu64 " will result in uint64_t overflow, "
9444 : "reset to %" PRIu64 "\n", limits[i], SPDK_BDEV_QOS_MAX_MBYTES_PER_SEC);
9445 0 : limits[i] = SPDK_BDEV_QOS_MAX_MBYTES_PER_SEC;
9446 0 : }
9447 : /* Change from megabyte to byte rate limit */
9448 27 : limits[i] = limits[i] * 1024 * 1024;
9449 27 : min_limit_per_sec = SPDK_BDEV_QOS_MIN_BYTES_PER_SEC;
9450 : }
9451 :
9452 36 : limit_set_complement = limits[i] % min_limit_per_sec;
9453 36 : if (limit_set_complement) {
9454 0 : SPDK_ERRLOG("Requested rate limit %" PRIu64 " is not a multiple of %" PRIu64 "\n",
9455 : limits[i], min_limit_per_sec);
9456 0 : limits[i] += min_limit_per_sec - limit_set_complement;
9457 0 : SPDK_ERRLOG("Round up the rate limit to %" PRIu64 "\n", limits[i]);
9458 0 : }
9459 36 : }
9460 :
9461 9 : ctx = calloc(1, sizeof(*ctx));
9462 9 : if (ctx == NULL) {
9463 0 : cb_fn(cb_arg, -ENOMEM);
9464 0 : return;
9465 : }
9466 :
9467 9 : ctx->cb_fn = cb_fn;
9468 9 : ctx->cb_arg = cb_arg;
9469 9 : ctx->bdev = bdev;
9470 :
9471 9 : spdk_spin_lock(&bdev->internal.spinlock);
9472 9 : if (bdev->internal.qos_mod_in_progress) {
9473 1 : spdk_spin_unlock(&bdev->internal.spinlock);
9474 1 : free(ctx);
9475 1 : cb_fn(cb_arg, -EAGAIN);
9476 1 : return;
9477 : }
9478 8 : bdev->internal.qos_mod_in_progress = true;
9479 :
9480 8 : if (disable_rate_limit == true && bdev->internal.qos) {
9481 10 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
9482 8 : if (limits[i] == SPDK_BDEV_QOS_LIMIT_NOT_DEFINED &&
9483 0 : (bdev->internal.qos->rate_limits[i].limit > 0 &&
9484 0 : bdev->internal.qos->rate_limits[i].limit !=
9485 : SPDK_BDEV_QOS_LIMIT_NOT_DEFINED)) {
9486 0 : disable_rate_limit = false;
9487 0 : break;
9488 : }
9489 8 : }
9490 2 : }
9491 :
9492 8 : if (disable_rate_limit == false) {
9493 5 : if (bdev->internal.qos == NULL) {
9494 4 : bdev->internal.qos = calloc(1, sizeof(*bdev->internal.qos));
9495 4 : if (!bdev->internal.qos) {
9496 0 : spdk_spin_unlock(&bdev->internal.spinlock);
9497 0 : SPDK_ERRLOG("Unable to allocate memory for QoS tracking\n");
9498 0 : bdev_set_qos_limit_done(ctx, -ENOMEM);
9499 0 : return;
9500 : }
9501 4 : }
9502 :
9503 5 : if (bdev->internal.qos->thread == NULL) {
9504 : /* Enabling */
9505 4 : bdev_set_qos_rate_limits(bdev, limits);
9506 :
9507 4 : spdk_bdev_for_each_channel(bdev, bdev_enable_qos_msg, ctx,
9508 : bdev_enable_qos_done);
9509 4 : } else {
9510 : /* Updating */
9511 1 : bdev_set_qos_rate_limits(bdev, limits);
9512 :
9513 2 : spdk_thread_send_msg(bdev->internal.qos->thread,
9514 1 : bdev_update_qos_rate_limit_msg, ctx);
9515 : }
9516 5 : } else {
9517 3 : if (bdev->internal.qos != NULL) {
9518 2 : bdev_set_qos_rate_limits(bdev, limits);
9519 :
9520 : /* Disabling */
9521 2 : spdk_bdev_for_each_channel(bdev, bdev_disable_qos_msg, ctx,
9522 : bdev_disable_qos_msg_done);
9523 2 : } else {
9524 1 : spdk_spin_unlock(&bdev->internal.spinlock);
9525 1 : bdev_set_qos_limit_done(ctx, 0);
9526 1 : return;
9527 : }
9528 : }
9529 :
9530 7 : spdk_spin_unlock(&bdev->internal.spinlock);
9531 9 : }
9532 :
9533 : struct spdk_bdev_histogram_ctx {
9534 : spdk_bdev_histogram_status_cb cb_fn;
9535 : void *cb_arg;
9536 : struct spdk_bdev *bdev;
9537 : int status;
9538 : };
9539 :
9540 : static void
9541 2 : bdev_histogram_disable_channel_cb(struct spdk_bdev *bdev, void *_ctx, int status)
9542 : {
9543 2 : struct spdk_bdev_histogram_ctx *ctx = _ctx;
9544 :
9545 2 : spdk_spin_lock(&ctx->bdev->internal.spinlock);
9546 2 : ctx->bdev->internal.histogram_in_progress = false;
9547 2 : spdk_spin_unlock(&ctx->bdev->internal.spinlock);
9548 2 : ctx->cb_fn(ctx->cb_arg, ctx->status);
9549 2 : free(ctx);
9550 2 : }
9551 :
9552 : static void
9553 3 : bdev_histogram_disable_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
9554 : struct spdk_io_channel *_ch, void *_ctx)
9555 : {
9556 3 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
9557 :
9558 3 : if (ch->histogram != NULL) {
9559 3 : spdk_histogram_data_free(ch->histogram);
9560 3 : ch->histogram = NULL;
9561 3 : }
9562 3 : spdk_bdev_for_each_channel_continue(i, 0);
9563 3 : }
9564 :
9565 : static void
9566 2 : bdev_histogram_enable_channel_cb(struct spdk_bdev *bdev, void *_ctx, int status)
9567 : {
9568 2 : struct spdk_bdev_histogram_ctx *ctx = _ctx;
9569 :
9570 2 : if (status != 0) {
9571 0 : ctx->status = status;
9572 0 : ctx->bdev->internal.histogram_enabled = false;
9573 0 : spdk_bdev_for_each_channel(ctx->bdev, bdev_histogram_disable_channel, ctx,
9574 : bdev_histogram_disable_channel_cb);
9575 0 : } else {
9576 2 : spdk_spin_lock(&ctx->bdev->internal.spinlock);
9577 2 : ctx->bdev->internal.histogram_in_progress = false;
9578 2 : spdk_spin_unlock(&ctx->bdev->internal.spinlock);
9579 2 : ctx->cb_fn(ctx->cb_arg, ctx->status);
9580 2 : free(ctx);
9581 : }
9582 2 : }
9583 :
9584 : static void
9585 3 : bdev_histogram_enable_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
9586 : struct spdk_io_channel *_ch, void *_ctx)
9587 : {
9588 3 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
9589 3 : int status = 0;
9590 :
9591 3 : if (ch->histogram == NULL) {
9592 3 : ch->histogram = spdk_histogram_data_alloc();
9593 3 : if (ch->histogram == NULL) {
9594 0 : status = -ENOMEM;
9595 0 : }
9596 3 : }
9597 :
9598 3 : spdk_bdev_for_each_channel_continue(i, status);
9599 3 : }
9600 :
9601 : void
9602 4 : spdk_bdev_histogram_enable_ext(struct spdk_bdev *bdev, spdk_bdev_histogram_status_cb cb_fn,
9603 : void *cb_arg, bool enable, struct spdk_bdev_enable_histogram_opts *opts)
9604 : {
9605 : struct spdk_bdev_histogram_ctx *ctx;
9606 :
9607 4 : ctx = calloc(1, sizeof(struct spdk_bdev_histogram_ctx));
9608 4 : if (ctx == NULL) {
9609 0 : cb_fn(cb_arg, -ENOMEM);
9610 0 : return;
9611 : }
9612 :
9613 4 : ctx->bdev = bdev;
9614 4 : ctx->status = 0;
9615 4 : ctx->cb_fn = cb_fn;
9616 4 : ctx->cb_arg = cb_arg;
9617 :
9618 4 : spdk_spin_lock(&bdev->internal.spinlock);
9619 4 : if (bdev->internal.histogram_in_progress) {
9620 0 : spdk_spin_unlock(&bdev->internal.spinlock);
9621 0 : free(ctx);
9622 0 : cb_fn(cb_arg, -EAGAIN);
9623 0 : return;
9624 : }
9625 :
9626 4 : bdev->internal.histogram_in_progress = true;
9627 4 : spdk_spin_unlock(&bdev->internal.spinlock);
9628 :
9629 4 : bdev->internal.histogram_enabled = enable;
9630 4 : bdev->internal.histogram_io_type = opts->io_type;
9631 :
9632 4 : if (enable) {
9633 : /* Allocate histogram for each channel */
9634 2 : spdk_bdev_for_each_channel(bdev, bdev_histogram_enable_channel, ctx,
9635 : bdev_histogram_enable_channel_cb);
9636 2 : } else {
9637 2 : spdk_bdev_for_each_channel(bdev, bdev_histogram_disable_channel, ctx,
9638 : bdev_histogram_disable_channel_cb);
9639 : }
9640 4 : }
9641 :
9642 : void
9643 4 : spdk_bdev_enable_histogram_opts_init(struct spdk_bdev_enable_histogram_opts *opts, size_t size)
9644 : {
9645 4 : if (opts == NULL) {
9646 0 : SPDK_ERRLOG("opts should not be NULL\n");
9647 0 : assert(opts != NULL);
9648 0 : return;
9649 : }
9650 4 : if (size == 0) {
9651 0 : SPDK_ERRLOG("size should not be zero\n");
9652 0 : assert(size != 0);
9653 0 : return;
9654 : }
9655 :
9656 4 : memset(opts, 0, size);
9657 4 : opts->size = size;
9658 :
9659 : #define FIELD_OK(field) \
9660 : offsetof(struct spdk_bdev_enable_histogram_opts, field) + sizeof(opts->field) <= size
9661 :
9662 : #define SET_FIELD(field, value) \
9663 : if (FIELD_OK(field)) { \
9664 : opts->field = value; \
9665 : } \
9666 :
9667 4 : SET_FIELD(io_type, 0);
9668 :
9669 : /* You should not remove this statement, but need to update the assert statement
9670 : * if you add a new field, and also add a corresponding SET_FIELD statement */
9671 : SPDK_STATIC_ASSERT(sizeof(struct spdk_bdev_enable_histogram_opts) == 9, "Incorrect size");
9672 :
9673 : #undef FIELD_OK
9674 : #undef SET_FIELD
9675 4 : }
9676 :
9677 : void
9678 4 : spdk_bdev_histogram_enable(struct spdk_bdev *bdev, spdk_bdev_histogram_status_cb cb_fn,
9679 : void *cb_arg, bool enable)
9680 : {
9681 : struct spdk_bdev_enable_histogram_opts opts;
9682 :
9683 4 : spdk_bdev_enable_histogram_opts_init(&opts, sizeof(opts));
9684 4 : spdk_bdev_histogram_enable_ext(bdev, cb_fn, cb_arg, enable, &opts);
9685 4 : }
9686 :
9687 : struct spdk_bdev_histogram_data_ctx {
9688 : spdk_bdev_histogram_data_cb cb_fn;
9689 : void *cb_arg;
9690 : struct spdk_bdev *bdev;
9691 : /** merged histogram data from all channels */
9692 : struct spdk_histogram_data *histogram;
9693 : };
9694 :
9695 : static void
9696 5 : bdev_histogram_get_channel_cb(struct spdk_bdev *bdev, void *_ctx, int status)
9697 : {
9698 5 : struct spdk_bdev_histogram_data_ctx *ctx = _ctx;
9699 :
9700 5 : ctx->cb_fn(ctx->cb_arg, status, ctx->histogram);
9701 5 : free(ctx);
9702 5 : }
9703 :
9704 : static void
9705 7 : bdev_histogram_get_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
9706 : struct spdk_io_channel *_ch, void *_ctx)
9707 : {
9708 7 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
9709 7 : struct spdk_bdev_histogram_data_ctx *ctx = _ctx;
9710 7 : int status = 0;
9711 :
9712 7 : if (ch->histogram == NULL) {
9713 1 : status = -EFAULT;
9714 1 : } else {
9715 6 : spdk_histogram_data_merge(ctx->histogram, ch->histogram);
9716 : }
9717 :
9718 7 : spdk_bdev_for_each_channel_continue(i, status);
9719 7 : }
9720 :
9721 : void
9722 5 : spdk_bdev_histogram_get(struct spdk_bdev *bdev, struct spdk_histogram_data *histogram,
9723 : spdk_bdev_histogram_data_cb cb_fn,
9724 : void *cb_arg)
9725 : {
9726 : struct spdk_bdev_histogram_data_ctx *ctx;
9727 :
9728 5 : ctx = calloc(1, sizeof(struct spdk_bdev_histogram_data_ctx));
9729 5 : if (ctx == NULL) {
9730 0 : cb_fn(cb_arg, -ENOMEM, NULL);
9731 0 : return;
9732 : }
9733 :
9734 5 : ctx->bdev = bdev;
9735 5 : ctx->cb_fn = cb_fn;
9736 5 : ctx->cb_arg = cb_arg;
9737 :
9738 5 : ctx->histogram = histogram;
9739 :
9740 5 : spdk_bdev_for_each_channel(bdev, bdev_histogram_get_channel, ctx,
9741 : bdev_histogram_get_channel_cb);
9742 5 : }
9743 :
9744 : void
9745 2 : spdk_bdev_channel_get_histogram(struct spdk_io_channel *ch, spdk_bdev_histogram_data_cb cb_fn,
9746 : void *cb_arg)
9747 : {
9748 2 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(ch);
9749 2 : int status = 0;
9750 :
9751 2 : assert(cb_fn != NULL);
9752 :
9753 2 : if (bdev_ch->histogram == NULL) {
9754 1 : status = -EFAULT;
9755 1 : }
9756 2 : cb_fn(cb_arg, status, bdev_ch->histogram);
9757 2 : }
9758 :
9759 : size_t
9760 0 : spdk_bdev_get_media_events(struct spdk_bdev_desc *desc, struct spdk_bdev_media_event *events,
9761 : size_t max_events)
9762 : {
9763 : struct media_event_entry *entry;
9764 0 : size_t num_events = 0;
9765 :
9766 0 : for (; num_events < max_events; ++num_events) {
9767 0 : entry = TAILQ_FIRST(&desc->pending_media_events);
9768 0 : if (entry == NULL) {
9769 0 : break;
9770 : }
9771 :
9772 0 : events[num_events] = entry->event;
9773 0 : TAILQ_REMOVE(&desc->pending_media_events, entry, tailq);
9774 0 : TAILQ_INSERT_TAIL(&desc->free_media_events, entry, tailq);
9775 0 : }
9776 :
9777 0 : return num_events;
9778 : }
9779 :
9780 : int
9781 0 : spdk_bdev_push_media_events(struct spdk_bdev *bdev, const struct spdk_bdev_media_event *events,
9782 : size_t num_events)
9783 : {
9784 : struct spdk_bdev_desc *desc;
9785 : struct media_event_entry *entry;
9786 : size_t event_id;
9787 0 : int rc = 0;
9788 :
9789 0 : assert(bdev->media_events);
9790 :
9791 0 : spdk_spin_lock(&bdev->internal.spinlock);
9792 0 : TAILQ_FOREACH(desc, &bdev->internal.open_descs, link) {
9793 0 : if (desc->write) {
9794 0 : break;
9795 : }
9796 0 : }
9797 :
9798 0 : if (desc == NULL || desc->media_events_buffer == NULL) {
9799 0 : rc = -ENODEV;
9800 0 : goto out;
9801 : }
9802 :
9803 0 : for (event_id = 0; event_id < num_events; ++event_id) {
9804 0 : entry = TAILQ_FIRST(&desc->free_media_events);
9805 0 : if (entry == NULL) {
9806 0 : break;
9807 : }
9808 :
9809 0 : TAILQ_REMOVE(&desc->free_media_events, entry, tailq);
9810 0 : TAILQ_INSERT_TAIL(&desc->pending_media_events, entry, tailq);
9811 0 : entry->event = events[event_id];
9812 0 : }
9813 :
9814 0 : rc = event_id;
9815 : out:
9816 0 : spdk_spin_unlock(&bdev->internal.spinlock);
9817 0 : return rc;
9818 : }
9819 :
9820 : static void
9821 0 : _media_management_notify(void *arg)
9822 : {
9823 0 : struct spdk_bdev_desc *desc = arg;
9824 :
9825 0 : _event_notify(desc, SPDK_BDEV_EVENT_MEDIA_MANAGEMENT);
9826 0 : }
9827 :
9828 : void
9829 0 : spdk_bdev_notify_media_management(struct spdk_bdev *bdev)
9830 : {
9831 : struct spdk_bdev_desc *desc;
9832 :
9833 0 : spdk_spin_lock(&bdev->internal.spinlock);
9834 0 : TAILQ_FOREACH(desc, &bdev->internal.open_descs, link) {
9835 0 : if (!TAILQ_EMPTY(&desc->pending_media_events)) {
9836 0 : event_notify(desc, _media_management_notify);
9837 0 : }
9838 0 : }
9839 0 : spdk_spin_unlock(&bdev->internal.spinlock);
9840 0 : }
9841 :
9842 : struct locked_lba_range_ctx {
9843 : struct lba_range range;
9844 : struct lba_range *current_range;
9845 : struct lba_range *owner_range;
9846 : struct spdk_poller *poller;
9847 : lock_range_cb cb_fn;
9848 : void *cb_arg;
9849 : };
9850 :
9851 : static void
9852 0 : bdev_lock_error_cleanup_cb(struct spdk_bdev *bdev, void *_ctx, int status)
9853 : {
9854 0 : struct locked_lba_range_ctx *ctx = _ctx;
9855 :
9856 0 : ctx->cb_fn(&ctx->range, ctx->cb_arg, -ENOMEM);
9857 0 : free(ctx);
9858 0 : }
9859 :
9860 : static void bdev_unlock_lba_range_get_channel(struct spdk_bdev_channel_iter *i,
9861 : struct spdk_bdev *bdev, struct spdk_io_channel *ch, void *_ctx);
9862 :
9863 : static void
9864 14 : bdev_lock_lba_range_cb(struct spdk_bdev *bdev, void *_ctx, int status)
9865 : {
9866 14 : struct locked_lba_range_ctx *ctx = _ctx;
9867 :
9868 14 : if (status == -ENOMEM) {
9869 : /* One of the channels could not allocate a range object.
9870 : * So we have to go back and clean up any ranges that were
9871 : * allocated successfully before we return error status to
9872 : * the caller. We can reuse the unlock function to do that
9873 : * clean up.
9874 : */
9875 0 : spdk_bdev_for_each_channel(bdev, bdev_unlock_lba_range_get_channel, ctx,
9876 : bdev_lock_error_cleanup_cb);
9877 0 : return;
9878 : }
9879 :
9880 : /* All channels have locked this range and no I/O overlapping the range
9881 : * are outstanding! Set the owner_ch for the range object for the
9882 : * locking channel, so that this channel will know that it is allowed
9883 : * to write to this range.
9884 : */
9885 14 : if (ctx->owner_range != NULL) {
9886 10 : ctx->owner_range->owner_ch = ctx->range.owner_ch;
9887 10 : }
9888 :
9889 14 : ctx->cb_fn(&ctx->range, ctx->cb_arg, status);
9890 :
9891 : /* Don't free the ctx here. Its range is in the bdev's global list of
9892 : * locked ranges still, and will be removed and freed when this range
9893 : * is later unlocked.
9894 : */
9895 14 : }
9896 :
9897 : static int
9898 17 : bdev_lock_lba_range_check_io(void *_i)
9899 : {
9900 17 : struct spdk_bdev_channel_iter *i = _i;
9901 17 : struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i->i);
9902 17 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
9903 17 : struct locked_lba_range_ctx *ctx = i->ctx;
9904 17 : struct lba_range *range = ctx->current_range;
9905 : struct spdk_bdev_io *bdev_io;
9906 :
9907 17 : spdk_poller_unregister(&ctx->poller);
9908 :
9909 : /* The range is now in the locked_ranges, so no new IO can be submitted to this
9910 : * range. But we need to wait until any outstanding IO overlapping with this range
9911 : * are completed.
9912 : */
9913 18 : TAILQ_FOREACH(bdev_io, &ch->io_submitted, internal.ch_link) {
9914 3 : if (bdev_io_range_is_locked(bdev_io, range)) {
9915 2 : ctx->poller = SPDK_POLLER_REGISTER(bdev_lock_lba_range_check_io, i, 100);
9916 2 : return SPDK_POLLER_BUSY;
9917 : }
9918 1 : }
9919 :
9920 15 : spdk_bdev_for_each_channel_continue(i, 0);
9921 15 : return SPDK_POLLER_BUSY;
9922 17 : }
9923 :
9924 : static void
9925 15 : bdev_lock_lba_range_get_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
9926 : struct spdk_io_channel *_ch, void *_ctx)
9927 : {
9928 15 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
9929 15 : struct locked_lba_range_ctx *ctx = _ctx;
9930 : struct lba_range *range;
9931 :
9932 16 : TAILQ_FOREACH(range, &ch->locked_ranges, tailq) {
9933 1 : if (range->length == ctx->range.length &&
9934 0 : range->offset == ctx->range.offset &&
9935 0 : range->locked_ctx == ctx->range.locked_ctx) {
9936 : /* This range already exists on this channel, so don't add
9937 : * it again. This can happen when a new channel is created
9938 : * while the for_each_channel operation is in progress.
9939 : * Do not check for outstanding I/O in that case, since the
9940 : * range was locked before any I/O could be submitted to the
9941 : * new channel.
9942 : */
9943 0 : spdk_bdev_for_each_channel_continue(i, 0);
9944 0 : return;
9945 : }
9946 1 : }
9947 :
9948 15 : range = calloc(1, sizeof(*range));
9949 15 : if (range == NULL) {
9950 0 : spdk_bdev_for_each_channel_continue(i, -ENOMEM);
9951 0 : return;
9952 : }
9953 :
9954 15 : range->length = ctx->range.length;
9955 15 : range->offset = ctx->range.offset;
9956 15 : range->locked_ctx = ctx->range.locked_ctx;
9957 15 : range->quiesce = ctx->range.quiesce;
9958 15 : ctx->current_range = range;
9959 15 : if (ctx->range.owner_ch == ch) {
9960 : /* This is the range object for the channel that will hold
9961 : * the lock. Store it in the ctx object so that we can easily
9962 : * set its owner_ch after the lock is finally acquired.
9963 : */
9964 10 : ctx->owner_range = range;
9965 10 : }
9966 15 : TAILQ_INSERT_TAIL(&ch->locked_ranges, range, tailq);
9967 15 : bdev_lock_lba_range_check_io(i);
9968 15 : }
9969 :
9970 : static void
9971 14 : bdev_lock_lba_range_ctx(struct spdk_bdev *bdev, struct locked_lba_range_ctx *ctx)
9972 : {
9973 14 : assert(spdk_get_thread() == ctx->range.owner_thread);
9974 14 : assert(ctx->range.owner_ch == NULL ||
9975 : spdk_io_channel_get_thread(ctx->range.owner_ch->channel) == ctx->range.owner_thread);
9976 :
9977 : /* We will add a copy of this range to each channel now. */
9978 14 : spdk_bdev_for_each_channel(bdev, bdev_lock_lba_range_get_channel, ctx,
9979 : bdev_lock_lba_range_cb);
9980 14 : }
9981 :
9982 : static bool
9983 17 : bdev_lba_range_overlaps_tailq(struct lba_range *range, lba_range_tailq_t *tailq)
9984 : {
9985 : struct lba_range *r;
9986 :
9987 18 : TAILQ_FOREACH(r, tailq, tailq) {
9988 4 : if (bdev_lba_range_overlapped(range, r)) {
9989 3 : return true;
9990 : }
9991 1 : }
9992 14 : return false;
9993 17 : }
9994 :
9995 : static void bdev_quiesce_range_locked(struct lba_range *range, void *ctx, int status);
9996 :
9997 : static int
9998 14 : _bdev_lock_lba_range(struct spdk_bdev *bdev, struct spdk_bdev_channel *ch,
9999 : uint64_t offset, uint64_t length,
10000 : lock_range_cb cb_fn, void *cb_arg)
10001 : {
10002 : struct locked_lba_range_ctx *ctx;
10003 :
10004 14 : ctx = calloc(1, sizeof(*ctx));
10005 14 : if (ctx == NULL) {
10006 0 : return -ENOMEM;
10007 : }
10008 :
10009 14 : ctx->range.offset = offset;
10010 14 : ctx->range.length = length;
10011 14 : ctx->range.owner_thread = spdk_get_thread();
10012 14 : ctx->range.owner_ch = ch;
10013 14 : ctx->range.locked_ctx = cb_arg;
10014 14 : ctx->range.bdev = bdev;
10015 14 : ctx->range.quiesce = (cb_fn == bdev_quiesce_range_locked);
10016 14 : ctx->cb_fn = cb_fn;
10017 14 : ctx->cb_arg = cb_arg;
10018 :
10019 14 : spdk_spin_lock(&bdev->internal.spinlock);
10020 14 : if (bdev_lba_range_overlaps_tailq(&ctx->range, &bdev->internal.locked_ranges)) {
10021 : /* There is an active lock overlapping with this range.
10022 : * Put it on the pending list until this range no
10023 : * longer overlaps with another.
10024 : */
10025 2 : TAILQ_INSERT_TAIL(&bdev->internal.pending_locked_ranges, &ctx->range, tailq);
10026 2 : } else {
10027 12 : TAILQ_INSERT_TAIL(&bdev->internal.locked_ranges, &ctx->range, tailq);
10028 12 : bdev_lock_lba_range_ctx(bdev, ctx);
10029 : }
10030 14 : spdk_spin_unlock(&bdev->internal.spinlock);
10031 14 : return 0;
10032 14 : }
10033 :
10034 : static int
10035 10 : bdev_lock_lba_range(struct spdk_bdev_desc *desc, struct spdk_io_channel *_ch,
10036 : uint64_t offset, uint64_t length,
10037 : lock_range_cb cb_fn, void *cb_arg)
10038 : {
10039 10 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
10040 10 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
10041 :
10042 10 : if (cb_arg == NULL) {
10043 0 : SPDK_ERRLOG("cb_arg must not be NULL\n");
10044 0 : return -EINVAL;
10045 : }
10046 :
10047 10 : return _bdev_lock_lba_range(bdev, ch, offset, length, cb_fn, cb_arg);
10048 10 : }
10049 :
10050 : static void
10051 2 : bdev_lock_lba_range_ctx_msg(void *_ctx)
10052 : {
10053 2 : struct locked_lba_range_ctx *ctx = _ctx;
10054 :
10055 2 : bdev_lock_lba_range_ctx(ctx->range.bdev, ctx);
10056 2 : }
10057 :
10058 : static void
10059 14 : bdev_unlock_lba_range_cb(struct spdk_bdev *bdev, void *_ctx, int status)
10060 : {
10061 14 : struct locked_lba_range_ctx *ctx = _ctx;
10062 : struct locked_lba_range_ctx *pending_ctx;
10063 : struct lba_range *range, *tmp;
10064 :
10065 14 : spdk_spin_lock(&bdev->internal.spinlock);
10066 : /* Check if there are any pending locked ranges that overlap with this range
10067 : * that was just unlocked. If there are, check that it doesn't overlap with any
10068 : * other locked ranges before calling bdev_lock_lba_range_ctx which will start
10069 : * the lock process.
10070 : */
10071 17 : TAILQ_FOREACH_SAFE(range, &bdev->internal.pending_locked_ranges, tailq, tmp) {
10072 3 : if (bdev_lba_range_overlapped(range, &ctx->range) &&
10073 3 : !bdev_lba_range_overlaps_tailq(range, &bdev->internal.locked_ranges)) {
10074 2 : TAILQ_REMOVE(&bdev->internal.pending_locked_ranges, range, tailq);
10075 2 : pending_ctx = SPDK_CONTAINEROF(range, struct locked_lba_range_ctx, range);
10076 2 : TAILQ_INSERT_TAIL(&bdev->internal.locked_ranges, range, tailq);
10077 4 : spdk_thread_send_msg(pending_ctx->range.owner_thread,
10078 2 : bdev_lock_lba_range_ctx_msg, pending_ctx);
10079 2 : }
10080 3 : }
10081 14 : spdk_spin_unlock(&bdev->internal.spinlock);
10082 :
10083 14 : ctx->cb_fn(&ctx->range, ctx->cb_arg, status);
10084 14 : free(ctx);
10085 14 : }
10086 :
10087 : static void
10088 16 : bdev_unlock_lba_range_get_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
10089 : struct spdk_io_channel *_ch, void *_ctx)
10090 : {
10091 16 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
10092 16 : struct locked_lba_range_ctx *ctx = _ctx;
10093 : TAILQ_HEAD(, spdk_bdev_io) io_locked;
10094 : struct spdk_bdev_io *bdev_io;
10095 : struct lba_range *range;
10096 :
10097 16 : TAILQ_FOREACH(range, &ch->locked_ranges, tailq) {
10098 32 : if (ctx->range.offset == range->offset &&
10099 16 : ctx->range.length == range->length &&
10100 16 : ctx->range.locked_ctx == range->locked_ctx) {
10101 16 : TAILQ_REMOVE(&ch->locked_ranges, range, tailq);
10102 16 : free(range);
10103 16 : break;
10104 : }
10105 0 : }
10106 :
10107 : /* Note: we should almost always be able to assert that the range specified
10108 : * was found. But there are some very rare corner cases where a new channel
10109 : * gets created simultaneously with a range unlock, where this function
10110 : * would execute on that new channel and wouldn't have the range.
10111 : * We also use this to clean up range allocations when a later allocation
10112 : * fails in the locking path.
10113 : * So we can't actually assert() here.
10114 : */
10115 :
10116 : /* Swap the locked IO into a temporary list, and then try to submit them again.
10117 : * We could hyper-optimize this to only resubmit locked I/O that overlap
10118 : * with the range that was just unlocked, but this isn't a performance path so
10119 : * we go for simplicity here.
10120 : */
10121 16 : TAILQ_INIT(&io_locked);
10122 16 : TAILQ_SWAP(&ch->io_locked, &io_locked, spdk_bdev_io, internal.ch_link);
10123 19 : while (!TAILQ_EMPTY(&io_locked)) {
10124 3 : bdev_io = TAILQ_FIRST(&io_locked);
10125 3 : TAILQ_REMOVE(&io_locked, bdev_io, internal.ch_link);
10126 3 : bdev_io_submit(bdev_io);
10127 : }
10128 :
10129 16 : spdk_bdev_for_each_channel_continue(i, 0);
10130 16 : }
10131 :
10132 : static int
10133 14 : _bdev_unlock_lba_range(struct spdk_bdev *bdev, uint64_t offset, uint64_t length,
10134 : lock_range_cb cb_fn, void *cb_arg)
10135 : {
10136 : struct locked_lba_range_ctx *ctx;
10137 : struct lba_range *range;
10138 :
10139 14 : spdk_spin_lock(&bdev->internal.spinlock);
10140 : /* To start the unlock the process, we find the range in the bdev's locked_ranges
10141 : * and remove it. This ensures new channels don't inherit the locked range.
10142 : * Then we will send a message to each channel to remove the range from its
10143 : * per-channel list.
10144 : */
10145 14 : TAILQ_FOREACH(range, &bdev->internal.locked_ranges, tailq) {
10146 24 : if (range->offset == offset && range->length == length &&
10147 14 : (range->owner_ch == NULL || range->locked_ctx == cb_arg)) {
10148 14 : break;
10149 : }
10150 0 : }
10151 14 : if (range == NULL) {
10152 0 : assert(false);
10153 : spdk_spin_unlock(&bdev->internal.spinlock);
10154 : return -EINVAL;
10155 : }
10156 14 : TAILQ_REMOVE(&bdev->internal.locked_ranges, range, tailq);
10157 14 : ctx = SPDK_CONTAINEROF(range, struct locked_lba_range_ctx, range);
10158 14 : spdk_spin_unlock(&bdev->internal.spinlock);
10159 :
10160 14 : ctx->cb_fn = cb_fn;
10161 14 : ctx->cb_arg = cb_arg;
10162 :
10163 14 : spdk_bdev_for_each_channel(bdev, bdev_unlock_lba_range_get_channel, ctx,
10164 : bdev_unlock_lba_range_cb);
10165 14 : return 0;
10166 : }
10167 :
10168 : static int
10169 12 : bdev_unlock_lba_range(struct spdk_bdev_desc *desc, struct spdk_io_channel *_ch,
10170 : uint64_t offset, uint64_t length,
10171 : lock_range_cb cb_fn, void *cb_arg)
10172 : {
10173 12 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
10174 12 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
10175 : struct lba_range *range;
10176 12 : bool range_found = false;
10177 :
10178 : /* Let's make sure the specified channel actually has a lock on
10179 : * the specified range. Note that the range must match exactly.
10180 : */
10181 14 : TAILQ_FOREACH(range, &ch->locked_ranges, tailq) {
10182 22 : if (range->offset == offset && range->length == length &&
10183 11 : range->owner_ch == ch && range->locked_ctx == cb_arg) {
10184 10 : range_found = true;
10185 10 : break;
10186 : }
10187 2 : }
10188 :
10189 12 : if (!range_found) {
10190 2 : return -EINVAL;
10191 : }
10192 :
10193 10 : return _bdev_unlock_lba_range(bdev, offset, length, cb_fn, cb_arg);
10194 12 : }
10195 :
10196 : struct bdev_quiesce_ctx {
10197 : spdk_bdev_quiesce_cb cb_fn;
10198 : void *cb_arg;
10199 : };
10200 :
10201 : static void
10202 4 : bdev_unquiesce_range_unlocked(struct lba_range *range, void *ctx, int status)
10203 : {
10204 4 : struct bdev_quiesce_ctx *quiesce_ctx = ctx;
10205 :
10206 4 : if (quiesce_ctx->cb_fn != NULL) {
10207 4 : quiesce_ctx->cb_fn(quiesce_ctx->cb_arg, status);
10208 4 : }
10209 :
10210 4 : free(quiesce_ctx);
10211 4 : }
10212 :
10213 : static void
10214 4 : bdev_quiesce_range_locked(struct lba_range *range, void *ctx, int status)
10215 : {
10216 4 : struct bdev_quiesce_ctx *quiesce_ctx = ctx;
10217 4 : struct spdk_bdev_module *module = range->bdev->module;
10218 :
10219 4 : if (status != 0) {
10220 0 : if (quiesce_ctx->cb_fn != NULL) {
10221 0 : quiesce_ctx->cb_fn(quiesce_ctx->cb_arg, status);
10222 0 : }
10223 0 : free(quiesce_ctx);
10224 0 : return;
10225 : }
10226 :
10227 4 : spdk_spin_lock(&module->internal.spinlock);
10228 4 : TAILQ_INSERT_TAIL(&module->internal.quiesced_ranges, range, tailq_module);
10229 4 : spdk_spin_unlock(&module->internal.spinlock);
10230 :
10231 4 : if (quiesce_ctx->cb_fn != NULL) {
10232 : /* copy the context in case the range is unlocked by the callback */
10233 4 : struct bdev_quiesce_ctx tmp = *quiesce_ctx;
10234 :
10235 4 : quiesce_ctx->cb_fn = NULL;
10236 4 : quiesce_ctx->cb_arg = NULL;
10237 :
10238 4 : tmp.cb_fn(tmp.cb_arg, status);
10239 4 : }
10240 : /* quiesce_ctx will be freed on unquiesce */
10241 4 : }
10242 :
10243 : static int
10244 9 : _spdk_bdev_quiesce(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
10245 : uint64_t offset, uint64_t length,
10246 : spdk_bdev_quiesce_cb cb_fn, void *cb_arg,
10247 : bool unquiesce)
10248 : {
10249 : struct bdev_quiesce_ctx *quiesce_ctx;
10250 : int rc;
10251 :
10252 9 : if (module != bdev->module) {
10253 0 : SPDK_ERRLOG("Bdev does not belong to specified module.\n");
10254 0 : return -EINVAL;
10255 : }
10256 :
10257 9 : if (!bdev_io_valid_blocks(bdev, offset, length)) {
10258 0 : return -EINVAL;
10259 : }
10260 :
10261 9 : if (unquiesce) {
10262 : struct lba_range *range;
10263 :
10264 : /* Make sure the specified range is actually quiesced in the specified module and
10265 : * then remove it from the list. Note that the range must match exactly.
10266 : */
10267 5 : spdk_spin_lock(&module->internal.spinlock);
10268 6 : TAILQ_FOREACH(range, &module->internal.quiesced_ranges, tailq_module) {
10269 5 : if (range->bdev == bdev && range->offset == offset && range->length == length) {
10270 4 : TAILQ_REMOVE(&module->internal.quiesced_ranges, range, tailq_module);
10271 4 : break;
10272 : }
10273 1 : }
10274 5 : spdk_spin_unlock(&module->internal.spinlock);
10275 :
10276 5 : if (range == NULL) {
10277 1 : SPDK_ERRLOG("The range to unquiesce was not found.\n");
10278 1 : return -EINVAL;
10279 : }
10280 :
10281 4 : quiesce_ctx = range->locked_ctx;
10282 4 : quiesce_ctx->cb_fn = cb_fn;
10283 4 : quiesce_ctx->cb_arg = cb_arg;
10284 :
10285 4 : rc = _bdev_unlock_lba_range(bdev, offset, length, bdev_unquiesce_range_unlocked, quiesce_ctx);
10286 4 : } else {
10287 4 : quiesce_ctx = malloc(sizeof(*quiesce_ctx));
10288 4 : if (quiesce_ctx == NULL) {
10289 0 : return -ENOMEM;
10290 : }
10291 :
10292 4 : quiesce_ctx->cb_fn = cb_fn;
10293 4 : quiesce_ctx->cb_arg = cb_arg;
10294 :
10295 4 : rc = _bdev_lock_lba_range(bdev, NULL, offset, length, bdev_quiesce_range_locked, quiesce_ctx);
10296 4 : if (rc != 0) {
10297 0 : free(quiesce_ctx);
10298 0 : }
10299 : }
10300 :
10301 8 : return rc;
10302 9 : }
10303 :
10304 : int
10305 3 : spdk_bdev_quiesce(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
10306 : spdk_bdev_quiesce_cb cb_fn, void *cb_arg)
10307 : {
10308 3 : return _spdk_bdev_quiesce(bdev, module, 0, bdev->blockcnt, cb_fn, cb_arg, false);
10309 : }
10310 :
10311 : int
10312 3 : spdk_bdev_unquiesce(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
10313 : spdk_bdev_quiesce_cb cb_fn, void *cb_arg)
10314 : {
10315 3 : return _spdk_bdev_quiesce(bdev, module, 0, bdev->blockcnt, cb_fn, cb_arg, true);
10316 : }
10317 :
10318 : int
10319 1 : spdk_bdev_quiesce_range(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
10320 : uint64_t offset, uint64_t length,
10321 : spdk_bdev_quiesce_cb cb_fn, void *cb_arg)
10322 : {
10323 1 : return _spdk_bdev_quiesce(bdev, module, offset, length, cb_fn, cb_arg, false);
10324 : }
10325 :
10326 : int
10327 2 : spdk_bdev_unquiesce_range(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
10328 : uint64_t offset, uint64_t length,
10329 : spdk_bdev_quiesce_cb cb_fn, void *cb_arg)
10330 : {
10331 2 : return _spdk_bdev_quiesce(bdev, module, offset, length, cb_fn, cb_arg, true);
10332 : }
10333 :
10334 : int
10335 269 : spdk_bdev_get_memory_domains(struct spdk_bdev *bdev, struct spdk_memory_domain **domains,
10336 : int array_size)
10337 : {
10338 269 : if (!bdev) {
10339 1 : return -EINVAL;
10340 : }
10341 :
10342 268 : if (bdev->fn_table->get_memory_domains) {
10343 3 : return bdev->fn_table->get_memory_domains(bdev->ctxt, domains, array_size);
10344 : }
10345 :
10346 265 : return 0;
10347 269 : }
10348 :
10349 : struct spdk_bdev_for_each_io_ctx {
10350 : void *ctx;
10351 : spdk_bdev_io_fn fn;
10352 : spdk_bdev_for_each_io_cb cb;
10353 : };
10354 :
10355 : static void
10356 0 : bdev_channel_for_each_io(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
10357 : struct spdk_io_channel *io_ch, void *_ctx)
10358 : {
10359 0 : struct spdk_bdev_for_each_io_ctx *ctx = _ctx;
10360 0 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
10361 : struct spdk_bdev_io *bdev_io;
10362 0 : int rc = 0;
10363 :
10364 0 : TAILQ_FOREACH(bdev_io, &bdev_ch->io_submitted, internal.ch_link) {
10365 0 : rc = ctx->fn(ctx->ctx, bdev_io);
10366 0 : if (rc != 0) {
10367 0 : break;
10368 : }
10369 0 : }
10370 :
10371 0 : spdk_bdev_for_each_channel_continue(i, rc);
10372 0 : }
10373 :
10374 : static void
10375 0 : bdev_for_each_io_done(struct spdk_bdev *bdev, void *_ctx, int status)
10376 : {
10377 0 : struct spdk_bdev_for_each_io_ctx *ctx = _ctx;
10378 :
10379 0 : ctx->cb(ctx->ctx, status);
10380 :
10381 0 : free(ctx);
10382 0 : }
10383 :
10384 : void
10385 0 : spdk_bdev_for_each_bdev_io(struct spdk_bdev *bdev, void *_ctx, spdk_bdev_io_fn fn,
10386 : spdk_bdev_for_each_io_cb cb)
10387 : {
10388 : struct spdk_bdev_for_each_io_ctx *ctx;
10389 :
10390 0 : assert(fn != NULL && cb != NULL);
10391 :
10392 0 : ctx = calloc(1, sizeof(*ctx));
10393 0 : if (ctx == NULL) {
10394 0 : SPDK_ERRLOG("Failed to allocate context.\n");
10395 0 : cb(_ctx, -ENOMEM);
10396 0 : return;
10397 : }
10398 :
10399 0 : ctx->ctx = _ctx;
10400 0 : ctx->fn = fn;
10401 0 : ctx->cb = cb;
10402 :
10403 0 : spdk_bdev_for_each_channel(bdev, bdev_channel_for_each_io, ctx,
10404 : bdev_for_each_io_done);
10405 0 : }
10406 :
10407 : void
10408 132 : spdk_bdev_for_each_channel_continue(struct spdk_bdev_channel_iter *iter, int status)
10409 : {
10410 132 : spdk_for_each_channel_continue(iter->i, status);
10411 132 : }
10412 :
10413 : static struct spdk_bdev *
10414 360 : io_channel_iter_get_bdev(struct spdk_io_channel_iter *i)
10415 : {
10416 360 : void *io_device = spdk_io_channel_iter_get_io_device(i);
10417 :
10418 360 : return __bdev_from_io_dev(io_device);
10419 : }
10420 :
10421 : static void
10422 132 : bdev_each_channel_msg(struct spdk_io_channel_iter *i)
10423 : {
10424 132 : struct spdk_bdev_channel_iter *iter = spdk_io_channel_iter_get_ctx(i);
10425 132 : struct spdk_bdev *bdev = io_channel_iter_get_bdev(i);
10426 132 : struct spdk_io_channel *ch = spdk_io_channel_iter_get_channel(i);
10427 :
10428 132 : iter->i = i;
10429 132 : iter->fn(iter, bdev, ch, iter->ctx);
10430 132 : }
10431 :
10432 : static void
10433 228 : bdev_each_channel_cpl(struct spdk_io_channel_iter *i, int status)
10434 : {
10435 228 : struct spdk_bdev_channel_iter *iter = spdk_io_channel_iter_get_ctx(i);
10436 228 : struct spdk_bdev *bdev = io_channel_iter_get_bdev(i);
10437 :
10438 228 : iter->i = i;
10439 228 : iter->cpl(bdev, iter->ctx, status);
10440 :
10441 228 : free(iter);
10442 228 : }
10443 :
10444 : void
10445 228 : spdk_bdev_for_each_channel(struct spdk_bdev *bdev, spdk_bdev_for_each_channel_msg fn,
10446 : void *ctx, spdk_bdev_for_each_channel_done cpl)
10447 : {
10448 : struct spdk_bdev_channel_iter *iter;
10449 :
10450 228 : assert(bdev != NULL && fn != NULL && ctx != NULL);
10451 :
10452 228 : iter = calloc(1, sizeof(struct spdk_bdev_channel_iter));
10453 228 : if (iter == NULL) {
10454 0 : SPDK_ERRLOG("Unable to allocate iterator\n");
10455 0 : assert(false);
10456 : return;
10457 : }
10458 :
10459 228 : iter->fn = fn;
10460 228 : iter->cpl = cpl;
10461 228 : iter->ctx = ctx;
10462 :
10463 456 : spdk_for_each_channel(__bdev_to_io_dev(bdev), bdev_each_channel_msg,
10464 228 : iter, bdev_each_channel_cpl);
10465 228 : }
10466 :
10467 : static void
10468 3 : bdev_copy_do_write_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
10469 : {
10470 3 : struct spdk_bdev_io *parent_io = cb_arg;
10471 :
10472 3 : spdk_bdev_free_io(bdev_io);
10473 :
10474 : /* Check return status of write */
10475 3 : parent_io->internal.status = success ? SPDK_BDEV_IO_STATUS_SUCCESS : SPDK_BDEV_IO_STATUS_FAILED;
10476 3 : parent_io->internal.cb(parent_io, success, parent_io->internal.caller_ctx);
10477 3 : }
10478 :
10479 : static void
10480 3 : bdev_copy_do_write(void *_bdev_io)
10481 : {
10482 3 : struct spdk_bdev_io *bdev_io = _bdev_io;
10483 : int rc;
10484 :
10485 : /* Write blocks */
10486 6 : rc = spdk_bdev_write_blocks_with_md(bdev_io->internal.desc,
10487 3 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
10488 3 : bdev_io->u.bdev.iovs[0].iov_base,
10489 3 : bdev_io->u.bdev.md_buf, bdev_io->u.bdev.offset_blocks,
10490 3 : bdev_io->u.bdev.num_blocks, bdev_copy_do_write_done, bdev_io);
10491 :
10492 3 : if (rc == -ENOMEM) {
10493 0 : bdev_queue_io_wait_with_cb(bdev_io, bdev_copy_do_write);
10494 3 : } else if (rc != 0) {
10495 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
10496 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
10497 0 : }
10498 3 : }
10499 :
10500 : static void
10501 3 : bdev_copy_do_read_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
10502 : {
10503 3 : struct spdk_bdev_io *parent_io = cb_arg;
10504 :
10505 3 : spdk_bdev_free_io(bdev_io);
10506 :
10507 : /* Check return status of read */
10508 3 : if (!success) {
10509 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
10510 0 : parent_io->internal.cb(parent_io, false, parent_io->internal.caller_ctx);
10511 0 : return;
10512 : }
10513 :
10514 : /* Do write */
10515 3 : bdev_copy_do_write(parent_io);
10516 3 : }
10517 :
10518 : static void
10519 3 : bdev_copy_do_read(void *_bdev_io)
10520 : {
10521 3 : struct spdk_bdev_io *bdev_io = _bdev_io;
10522 : int rc;
10523 :
10524 : /* Read blocks */
10525 6 : rc = spdk_bdev_read_blocks_with_md(bdev_io->internal.desc,
10526 3 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
10527 3 : bdev_io->u.bdev.iovs[0].iov_base,
10528 3 : bdev_io->u.bdev.md_buf, bdev_io->u.bdev.copy.src_offset_blocks,
10529 3 : bdev_io->u.bdev.num_blocks, bdev_copy_do_read_done, bdev_io);
10530 :
10531 3 : if (rc == -ENOMEM) {
10532 0 : bdev_queue_io_wait_with_cb(bdev_io, bdev_copy_do_read);
10533 3 : } else if (rc != 0) {
10534 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
10535 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
10536 0 : }
10537 3 : }
10538 :
10539 : static void
10540 3 : bdev_copy_get_buf_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io, bool success)
10541 : {
10542 3 : if (!success) {
10543 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
10544 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
10545 0 : return;
10546 : }
10547 :
10548 3 : bdev_copy_do_read(bdev_io);
10549 3 : }
10550 :
10551 : int
10552 27 : spdk_bdev_copy_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
10553 : uint64_t dst_offset_blocks, uint64_t src_offset_blocks, uint64_t num_blocks,
10554 : spdk_bdev_io_completion_cb cb, void *cb_arg)
10555 : {
10556 27 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
10557 : struct spdk_bdev_io *bdev_io;
10558 27 : struct spdk_bdev_channel *channel = spdk_io_channel_get_ctx(ch);
10559 :
10560 27 : if (!desc->write) {
10561 0 : return -EBADF;
10562 : }
10563 :
10564 27 : if (!bdev_io_valid_blocks(bdev, dst_offset_blocks, num_blocks) ||
10565 27 : !bdev_io_valid_blocks(bdev, src_offset_blocks, num_blocks)) {
10566 0 : SPDK_DEBUGLOG(bdev,
10567 : "Invalid offset or number of blocks: dst %lu, src %lu, count %lu\n",
10568 : dst_offset_blocks, src_offset_blocks, num_blocks);
10569 0 : return -EINVAL;
10570 : }
10571 :
10572 27 : bdev_io = bdev_channel_get_io(channel);
10573 27 : if (!bdev_io) {
10574 0 : return -ENOMEM;
10575 : }
10576 :
10577 27 : bdev_io->internal.ch = channel;
10578 27 : bdev_io->internal.desc = desc;
10579 27 : bdev_io->type = SPDK_BDEV_IO_TYPE_COPY;
10580 :
10581 27 : bdev_io->u.bdev.offset_blocks = dst_offset_blocks;
10582 27 : bdev_io->u.bdev.copy.src_offset_blocks = src_offset_blocks;
10583 27 : bdev_io->u.bdev.num_blocks = num_blocks;
10584 27 : bdev_io->u.bdev.memory_domain = NULL;
10585 27 : bdev_io->u.bdev.memory_domain_ctx = NULL;
10586 27 : bdev_io->u.bdev.iovs = NULL;
10587 27 : bdev_io->u.bdev.iovcnt = 0;
10588 27 : bdev_io->u.bdev.md_buf = NULL;
10589 27 : bdev_io->u.bdev.accel_sequence = NULL;
10590 27 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
10591 :
10592 27 : if (dst_offset_blocks == src_offset_blocks || num_blocks == 0) {
10593 0 : spdk_thread_send_msg(spdk_get_thread(), bdev_io_complete_cb, bdev_io);
10594 0 : return 0;
10595 : }
10596 :
10597 :
10598 : /* If the copy size is large and should be split, use the generic split logic
10599 : * regardless of whether SPDK_BDEV_IO_TYPE_COPY is supported or not.
10600 : *
10601 : * Then, send the copy request if SPDK_BDEV_IO_TYPE_COPY is supported or
10602 : * emulate it using regular read and write requests otherwise.
10603 : */
10604 27 : if (spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COPY) ||
10605 4 : bdev_io->internal.f.split) {
10606 24 : bdev_io_submit(bdev_io);
10607 24 : return 0;
10608 : }
10609 :
10610 3 : spdk_bdev_io_get_buf(bdev_io, bdev_copy_get_buf_cb, num_blocks * spdk_bdev_get_block_size(bdev));
10611 :
10612 3 : return 0;
10613 27 : }
10614 :
10615 3 : SPDK_LOG_REGISTER_COMPONENT(bdev)
10616 :
10617 : static void
10618 0 : bdev_trace(void)
10619 : {
10620 0 : struct spdk_trace_tpoint_opts opts[] = {
10621 : {
10622 : "BDEV_IO_START", TRACE_BDEV_IO_START,
10623 : OWNER_TYPE_BDEV, OBJECT_BDEV_IO, 1,
10624 : {
10625 : { "type", SPDK_TRACE_ARG_TYPE_INT, 8 },
10626 : { "ctx", SPDK_TRACE_ARG_TYPE_PTR, 8 },
10627 : { "offset", SPDK_TRACE_ARG_TYPE_INT, 8 },
10628 : { "qd", SPDK_TRACE_ARG_TYPE_INT, 4 }
10629 : }
10630 : },
10631 : {
10632 : "BDEV_IO_DONE", TRACE_BDEV_IO_DONE,
10633 : OWNER_TYPE_BDEV, OBJECT_BDEV_IO, 0,
10634 : {
10635 : { "ctx", SPDK_TRACE_ARG_TYPE_PTR, 8 },
10636 : { "qd", SPDK_TRACE_ARG_TYPE_INT, 4 }
10637 : }
10638 : },
10639 : {
10640 : "BDEV_IOCH_CREATE", TRACE_BDEV_IOCH_CREATE,
10641 : OWNER_TYPE_BDEV, OBJECT_NONE, 0,
10642 : {
10643 : { "tid", SPDK_TRACE_ARG_TYPE_INT, 8 }
10644 : }
10645 : },
10646 : {
10647 : "BDEV_IOCH_DESTROY", TRACE_BDEV_IOCH_DESTROY,
10648 : OWNER_TYPE_BDEV, OBJECT_NONE, 0,
10649 : {
10650 : { "tid", SPDK_TRACE_ARG_TYPE_INT, 8 }
10651 : }
10652 : },
10653 : };
10654 :
10655 :
10656 0 : spdk_trace_register_owner_type(OWNER_TYPE_BDEV, 'b');
10657 0 : spdk_trace_register_object(OBJECT_BDEV_IO, 'i');
10658 0 : spdk_trace_register_description_ext(opts, SPDK_COUNTOF(opts));
10659 0 : spdk_trace_tpoint_register_relation(TRACE_BDEV_NVME_IO_START, OBJECT_BDEV_IO, 0);
10660 0 : spdk_trace_tpoint_register_relation(TRACE_BDEV_NVME_IO_DONE, OBJECT_BDEV_IO, 0);
10661 0 : spdk_trace_tpoint_register_relation(TRACE_BLOB_REQ_SET_START, OBJECT_BDEV_IO, 0);
10662 0 : spdk_trace_tpoint_register_relation(TRACE_BLOB_REQ_SET_COMPLETE, OBJECT_BDEV_IO, 0);
10663 0 : spdk_trace_tpoint_register_relation(TRACE_BDEV_RAID_IO_START, OBJECT_BDEV_IO, 0);
10664 0 : spdk_trace_tpoint_register_relation(TRACE_BDEV_RAID_IO_DONE, OBJECT_BDEV_IO, 0);
10665 0 : }
10666 3 : SPDK_TRACE_REGISTER_FN(bdev_trace, "bdev", TRACE_GROUP_BDEV)
|