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
86 546 : bdev_name_cmp(struct spdk_bdev_name *name1, struct spdk_bdev_name *name2)
87 : {
88 546 : return strcmp(name1->name, name2->name);
89 : }
90 :
91 1724 : 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 : }
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 : }
540 :
541 : static struct spdk_bdev *
542 144 : bdev_get_by_name(const char *bdev_name)
543 : {
544 144 : 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 : }
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 : }
596 :
597 0 : return "reserved";
598 : }
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 : }
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 : }
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 : }
659 6 : return false;
660 : }
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 : }
686 3 : return false;
687 : }
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 : }
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 : }
728 : }
729 : }
730 :
731 121 : spdk_spin_lock(&bdev->internal.spinlock);
732 :
733 121 : switch (bdev->internal.claim_type) {
734 116 : 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 : }
745 : }
746 116 : break;
747 1 : 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 4 : 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 : }
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 : }
803 4 : if (TAILQ_EMPTY(&bdev->internal.claim.v2.claims)) {
804 0 : claim_reset(bdev);
805 : }
806 : }
807 : }
808 :
809 120 : spdk_spin_unlock(&bdev->internal.spinlock);
810 : }
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 : }
850 0 : return 0;
851 : }
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 : }
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 : }
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 : }
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 : }
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 : }
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 : }
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 : }
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 : }
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 : }
1028 :
1029 7 : return true;
1030 : }
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 : }
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 : }
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 : }
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 : }
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 : 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 : } 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 : }
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 : }
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 2 : rc = spdk_memory_domain_pull_data(bdev_io->internal.memory_domain,
1183 : bdev_io->internal.memory_domain_ctx,
1184 : &bdev_io->internal.bounce_buf.orig_md_iov, 1,
1185 : &bdev_io->internal.bounce_buf.md_iov, 1,
1186 : 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 : }
1198 : } 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 : bdev_io->internal.bounce_buf.orig_md_iov.iov_len);
1202 : }
1203 : }
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 : } else {
1208 2 : assert(bdev_io->internal.data_transfer_cpl);
1209 2 : bdev_io->internal.data_transfer_cpl(bdev_io, rc);
1210 : }
1211 : }
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 : }
1251 :
1252 38 : bdev_io_get_buf_complete(bdev_io, true);
1253 : }
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 : }
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 : }
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 54 : 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 : 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 : } 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 : 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 : }
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 : bdev_io->internal.memory_domain_ctx,
1332 : bdev_io->internal.bounce_buf.orig_iovs,
1333 3 : (uint32_t)bdev_io->internal.bounce_buf.orig_iovcnt,
1334 : bdev_io->u.bdev.iovs, 1,
1335 : bdev_io_pull_data_done_and_track,
1336 : 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 : }
1348 : } 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 : bdev_io->internal.bounce_buf.orig_iovs,
1354 : bdev_io->internal.bounce_buf.orig_iovcnt);
1355 : }
1356 : }
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 : } else {
1361 24 : bdev_io_pull_data_done(bdev_io, rc);
1362 : }
1363 : }
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 : } 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 : }
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 548 : 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 : }
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 10 : case BDEV_IO_RETRY_STATE_SUBMIT:
1525 10 : bdev_ch_resubmit_io(shared_resource, bdev_io);
1526 10 : break;
1527 1 : case BDEV_IO_RETRY_STATE_PULL:
1528 1 : bdev_io_pull_data(bdev_io);
1529 1 : break;
1530 0 : case BDEV_IO_RETRY_STATE_PULL_MD:
1531 0 : bdev_io_pull_md_buf(bdev_io);
1532 0 : break;
1533 1 : case BDEV_IO_RETRY_STATE_PUSH:
1534 1 : bdev_io_push_bounce_data(bdev_io);
1535 1 : break;
1536 0 : case BDEV_IO_RETRY_STATE_PUSH_MD:
1537 0 : bdev_io_push_bounce_md_buf(bdev_io);
1538 0 : break;
1539 0 : 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 : }
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 : }
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 : }
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 : }
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 : }
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 : }
1617 :
1618 551 : return false;
1619 : }
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 : }
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 : }
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 : }
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 : bdev_io->internal.memory_domain_ctx,
1680 : &bdev_io->internal.bounce_buf.orig_md_iov,
1681 2 : (uint32_t)bdev_io->internal.bounce_buf.orig_iovcnt,
1682 : &bdev_io->internal.bounce_buf.md_iov, 1,
1683 : bdev_io_push_bounce_md_buf_done,
1684 : 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 : }
1696 : } else {
1697 0 : memcpy(bdev_io->internal.bounce_buf.orig_md_iov.iov_base, bdev_io->u.bdev.md_buf,
1698 : bdev_io->internal.bounce_buf.orig_md_iov.iov_len);
1699 : }
1700 : }
1701 : }
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 : } 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 : }
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 : }
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 : }
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 : bdev_io->internal.memory_domain_ctx,
1765 : bdev_io->internal.bounce_buf.orig_iovs,
1766 3 : (uint32_t)bdev_io->internal.bounce_buf.orig_iovcnt,
1767 : &bdev_io->internal.bounce_buf.iov, 1,
1768 : bdev_io_push_bounce_data_done_and_track,
1769 : 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 : }
1782 : } else {
1783 8 : spdk_copy_buf_to_iovs(bdev_io->internal.bounce_buf.orig_iovs,
1784 : bdev_io->internal.bounce_buf.orig_iovcnt,
1785 : bdev_io->internal.bounce_buf.iov.iov_base,
1786 : bdev_io->internal.bounce_buf.iov.iov_len);
1787 : }
1788 : }
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 : } else {
1793 24 : bdev_io_push_bounce_data_done(bdev_io, rc);
1794 : }
1795 : }
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 : }
1836 : }
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 96 : 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 : }
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 : }
1880 :
1881 4 : bdev_io_submit(bdev_io);
1882 : }
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 : }
1921 : }
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 : }
1945 :
1946 0 : spdk_json_write_object_end(w);
1947 :
1948 0 : spdk_json_write_object_end(w);
1949 : }
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 0 : 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 : }
1973 : }
1974 0 : spdk_json_write_object_end(w);
1975 :
1976 0 : spdk_json_write_object_end(w);
1977 : }
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 : }
2006 : }
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 : }
2014 :
2015 0 : bdev_qos_config_json(bdev, w);
2016 0 : bdev_enable_histogram_config_json(bdev, w);
2017 : }
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 71 : rc = spdk_iobuf_channel_init(&ch->iobuf, "bdev",
2056 : g_bdev_opts.iobuf_small_cache_size,
2057 : 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 : }
2079 :
2080 71 : TAILQ_INIT(&ch->shared_resources);
2081 71 : TAILQ_INIT(&ch->io_wait_queue);
2082 :
2083 71 : return 0;
2084 : }
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 : }
2106 : }
2107 : }
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 : }
2122 256 : return true;
2123 : }
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 : }
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 : }
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 : }
2216 :
2217 65 : g_resume_bdev_module = NULL;
2218 65 : return 0;
2219 : }
2220 :
2221 : void
2222 65 : spdk_bdev_initialize(spdk_bdev_init_cb cb_fn, void *cb_arg)
2223 : {
2224 65 : int rc = 0;
2225 65 : 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 : 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 : }
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 : }
2295 :
2296 65 : spdk_mempool_free(g_bdev_mgr.bdev_io_pool);
2297 : }
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 : } 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 : }
2342 :
2343 189 : if (bdev_module->module_fini) {
2344 189 : bdev_module->module_fini();
2345 : }
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 : }
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 : } 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 : }
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 : }
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 : } 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 : }
2454 :
2455 189 : if (bdev_module->fini_start) {
2456 23 : bdev_module->fini_start();
2457 : }
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 : }
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 : } 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 : }
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 60 : } 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 : } 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 : }
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 : } 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 18 : case SPDK_BDEV_QOS_RW_IOPS_RATE_LIMIT:
2571 18 : return true;
2572 54 : 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 : }
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 23 : 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 0 : 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 2 : default:
2598 2 : return false;
2599 : }
2600 : }
2601 :
2602 : static bool
2603 33 : bdev_is_read_io(struct spdk_bdev_io *bdev_io)
2604 : {
2605 33 : switch (bdev_io->type) {
2606 0 : 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 30 : case SPDK_BDEV_IO_TYPE_READ:
2615 30 : return true;
2616 0 : 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 3 : default:
2624 3 : return false;
2625 : }
2626 : }
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 0 : 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 43 : 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 0 : 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 0 : default:
2648 0 : return 0;
2649 : }
2650 : }
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 : }
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 : }
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 : }
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 : }
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 : }
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 9 : 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 7 : 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 5 : 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 4 : 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 0 : default:
2778 0 : break;
2779 : }
2780 : }
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 32 : 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 : }
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 : } 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 : }
2834 : }
2835 : }
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 77 : if (qos->rate_limits[i].queue_io(&qos->rate_limits[i],
2849 : 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 : }
2857 5 : return true;
2858 : }
2859 : }
2860 : }
2861 :
2862 20 : return false;
2863 : }
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 : }
2878 : }
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 2 : rc = spdk_bdev_queue_io_wait(bdev_io->bdev, spdk_io_channel_from_ctx(bdev_io->internal.ch),
2892 : &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 : }
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 597 : } else if (bdev->split_on_optimal_io_boundary) {
2912 168 : io_boundary = bdev->optimal_io_boundary;
2913 : } 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 : } 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 : }
2939 :
2940 303 : if (max_segs) {
2941 150 : if (bdev_io->u.bdev.iovcnt > max_segs) {
2942 15 : return true;
2943 : }
2944 : }
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 : }
2952 : }
2953 :
2954 276 : if (max_size) {
2955 52 : if (bdev_io->u.bdev.num_blocks > max_size) {
2956 7 : return true;
2957 : }
2958 : }
2959 :
2960 269 : return false;
2961 : }
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 : }
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 : }
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 : }
3003 :
3004 : static bool
3005 792 : bdev_io_should_split(struct spdk_bdev_io *bdev_io)
3006 : {
3007 792 : switch (bdev_io->type) {
3008 621 : case SPDK_BDEV_IO_TYPE_READ:
3009 : case SPDK_BDEV_IO_TYPE_WRITE:
3010 621 : return bdev_rw_should_split(bdev_io);
3011 24 : case SPDK_BDEV_IO_TYPE_UNMAP:
3012 24 : return bdev_unmap_should_split(bdev_io);
3013 37 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3014 37 : return bdev_write_zeroes_should_split(bdev_io);
3015 30 : case SPDK_BDEV_IO_TYPE_COPY:
3016 30 : return bdev_copy_should_split(bdev_io);
3017 80 : default:
3018 80 : return false;
3019 : }
3020 : }
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 196 : case SPDK_BDEV_IO_TYPE_READ:
3074 196 : assert(bdev_io->u.bdev.accel_sequence == NULL);
3075 784 : rc = bdev_readv_blocks_with_md(bdev_io->internal.desc,
3076 196 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
3077 : iov, iovcnt, md_buf, current_offset,
3078 : 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 : bdev_io->u.bdev.dif_check_flags,
3083 : bdev_io_split_done, bdev_io);
3084 196 : break;
3085 50 : case SPDK_BDEV_IO_TYPE_WRITE:
3086 50 : assert(bdev_io->u.bdev.accel_sequence == NULL);
3087 200 : rc = bdev_writev_blocks_with_md(bdev_io->internal.desc,
3088 50 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
3089 : iov, iovcnt, md_buf, current_offset,
3090 : 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 : bdev_io->u.bdev.dif_check_flags,
3095 : bdev_io->u.bdev.nvme_cdw12.raw,
3096 : bdev_io->u.bdev.nvme_cdw13.raw,
3097 : bdev_io_split_done, bdev_io);
3098 50 : break;
3099 17 : case SPDK_BDEV_IO_TYPE_UNMAP:
3100 17 : io_wait_fn = _bdev_unmap_split;
3101 17 : rc = spdk_bdev_unmap_blocks(bdev_io->internal.desc,
3102 17 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
3103 : current_offset, num_blocks,
3104 : bdev_io_split_done, bdev_io);
3105 17 : break;
3106 23 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3107 23 : io_wait_fn = _bdev_write_zeroes_split;
3108 23 : rc = spdk_bdev_write_zeroes_blocks(bdev_io->internal.desc,
3109 23 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
3110 : current_offset, num_blocks,
3111 : bdev_io_split_done, bdev_io);
3112 23 : break;
3113 19 : case SPDK_BDEV_IO_TYPE_COPY:
3114 19 : io_wait_fn = _bdev_copy_split;
3115 19 : current_src_offset = bdev_io->u.bdev.copy.src_offset_blocks +
3116 19 : (current_offset - bdev_io->u.bdev.offset_blocks);
3117 19 : rc = spdk_bdev_copy_blocks(bdev_io->internal.desc,
3118 19 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
3119 : current_offset, current_src_offset, num_blocks,
3120 : bdev_io_split_done, bdev_io);
3121 19 : break;
3122 0 : 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 : } 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 : }
3142 : } 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 : }
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 67 : 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 62 : } else if (bdev->split_on_optimal_io_boundary) {
3183 40 : io_boundary = bdev->optimal_io_boundary;
3184 : } 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 : }
3203 :
3204 67 : child_iovcnt = 0;
3205 309 : while (remaining > 0 && parent_iovpos < parent_iovcnt &&
3206 : 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 : }
3219 :
3220 249 : child_iovsize = spdk_min(SPDK_BDEV_IO_NUM_CHILD_IOV - child_iovcnt, max_child_iovcnt);
3221 974 : while (to_next_boundary_bytes > 0 && parent_iovpos < parent_iovcnt &&
3222 : 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 : } 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 : }
3281 :
3282 3 : return;
3283 : }
3284 : }
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 : }
3292 29 : parent_iov_offset -= iov_len;
3293 : }
3294 :
3295 21 : assert(to_last_block_bytes == 0);
3296 : }
3297 108 : to_next_boundary -= to_next_boundary_bytes / blocklen;
3298 : }
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 : }
3307 :
3308 : static void
3309 3 : bdev_unmap_split(struct spdk_bdev_io *bdev_io)
3310 : {
3311 3 : 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 : } else {
3329 0 : return;
3330 : }
3331 : }
3332 : }
3333 :
3334 : static void
3335 6 : bdev_write_zeroes_split(struct spdk_bdev_io *bdev_io)
3336 : {
3337 6 : 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 : } else {
3354 0 : return;
3355 : }
3356 : }
3357 : }
3358 :
3359 : static void
3360 4 : bdev_copy_split(struct spdk_bdev_io *bdev_io)
3361 : {
3362 4 : 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 : } else {
3380 0 : return;
3381 : }
3382 : }
3383 : }
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 : }
3393 :
3394 58 : parent_io->internal.cb(parent_io, parent_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS,
3395 : 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 : }
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 : }
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 : }
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 17 : case SPDK_BDEV_IO_TYPE_READ:
3467 : case SPDK_BDEV_IO_TYPE_WRITE:
3468 17 : _bdev_rw_split(parent_io);
3469 17 : break;
3470 1 : case SPDK_BDEV_IO_TYPE_UNMAP:
3471 1 : bdev_unmap_split(parent_io);
3472 1 : break;
3473 1 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3474 1 : bdev_write_zeroes_split(parent_io);
3475 1 : break;
3476 1 : case SPDK_BDEV_IO_TYPE_COPY:
3477 1 : bdev_copy_split(parent_io);
3478 1 : break;
3479 0 : default:
3480 0 : assert(false);
3481 : break;
3482 : }
3483 : }
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 49 : 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 : } 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 2 : case SPDK_BDEV_IO_TYPE_UNMAP:
3511 2 : bdev_unmap_split(bdev_io);
3512 2 : break;
3513 5 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3514 5 : bdev_write_zeroes_split(bdev_io);
3515 5 : break;
3516 3 : case SPDK_BDEV_IO_TYPE_COPY:
3517 3 : bdev_copy_split(bdev_io);
3518 3 : break;
3519 0 : 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 : }
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 23 : } else if (bdev_ch->flags & BDEV_CH_QOS_ENABLED) {
3554 25 : 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 : } 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 : } 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 : }
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 : }
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 11 : struct lba_range r;
3592 :
3593 11 : switch (bdev_io->type) {
3594 0 : 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 6 : 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 0 : default:
3625 0 : return false;
3626 : }
3627 : }
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 : }
3645 : }
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 : }
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 4 : _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 : }
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 : }
3713 :
3714 288 : bdev_io_submit(bdev_io);
3715 : }
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 0 : 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 7 : default:
3770 7 : break;
3771 : }
3772 : }
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 : }
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 : }
3820 :
3821 0 : return -1;
3822 : }
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 : }
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 25 : 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 25 : __atomic_store_n(&qos->rate_limits[i].remaining_this_timeslice,
3859 25 : qos->rate_limits[i].max_per_timeslice, __ATOMIC_RELEASE);
3860 : }
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 : remaining_last_timeslice, __ATOMIC_RELAXED);
3927 : }
3928 : }
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 8 : __atomic_add_fetch(&qos->rate_limits[i].remaining_this_timeslice,
3934 8 : qos->rate_limits[i].max_per_timeslice, __ATOMIC_RELAXED);
3935 : }
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 : }
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 : }
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 : } 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 : }
4019 : }
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 : }
4028 :
4029 17 : ch->flags |= BDEV_CH_QOS_ENABLED;
4030 : }
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 : }
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 12 : 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 : }
4102 : }
4103 :
4104 7 : end:
4105 12 : spdk_bdev_for_each_channel_continue(i, 0);
4106 : }
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 : }
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 : }
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 : }
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 : }
4196 : }
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 : }
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 : }
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 : }
4291 :
4292 74 : spdk_spin_unlock(&bdev->internal.spinlock);
4293 :
4294 74 : return 0;
4295 : }
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 : }
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 : }
4345 40 : spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_ABORTED);
4346 : }
4347 : }
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 : }
4362 :
4363 18 : return false;
4364 : }
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 : }
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 : }
4446 :
4447 7 : bdev->internal.qos = new_qos;
4448 :
4449 7 : if (old_qos->thread == NULL) {
4450 0 : free(old_qos);
4451 : } 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 : }
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 : }
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 : }
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 : }
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 : }
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 : }
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 : }
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 : }
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 : }
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 : }
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 : }
4550 : }
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 : } 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 : }
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 : }
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 : 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 : 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 : 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 : 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 : }
4627 : }
4628 0 : spdk_json_write_object_end(w);
4629 : }
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 : }
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 : }
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 : }
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 : }
4756 :
4757 4 : return -ENOENT;
4758 : }
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 : }
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 : }
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 : }
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 : }
4869 : }
4870 : }
4871 : }
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 : }
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 : }
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 : }
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 : }
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 0 : case SPDK_DIF_CHECK_TYPE_REFTAG:
4995 0 : return (bdev->dif_check_flags & SPDK_DIF_FLAGS_REFTAG_CHECK) != 0;
4996 0 : case SPDK_DIF_CHECK_TYPE_APPTAG:
4997 0 : return (bdev->dif_check_flags & SPDK_DIF_FLAGS_APPTAG_CHECK) != 0;
4998 0 : case SPDK_DIF_CHECK_TYPE_GUARD:
4999 0 : return (bdev->dif_check_flags & SPDK_DIF_FLAGS_GUARD_CHECK) != 0;
5000 0 : default:
5001 0 : return false;
5002 : }
5003 : }
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 : uint32_t
5053 0 : spdk_bdev_get_nvme_nsid(struct spdk_bdev *bdev)
5054 : {
5055 0 : return bdev->nsid;
5056 : }
5057 :
5058 : static void bdev_update_qd_sampling_period(void *ctx);
5059 :
5060 : static void
5061 1 : _calculate_measured_qd_cpl(struct spdk_bdev *bdev, void *_ctx, int status)
5062 : {
5063 1 : bdev->internal.measured_queue_depth = bdev->internal.temporary_queue_depth;
5064 :
5065 1 : if (bdev->internal.measured_queue_depth) {
5066 0 : bdev->internal.io_time += bdev->internal.period;
5067 0 : bdev->internal.weighted_io_time += bdev->internal.period * bdev->internal.measured_queue_depth;
5068 : }
5069 :
5070 1 : bdev->internal.qd_poll_in_progress = false;
5071 :
5072 1 : bdev_update_qd_sampling_period(bdev);
5073 1 : }
5074 :
5075 : static void
5076 1 : _calculate_measured_qd(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
5077 : struct spdk_io_channel *io_ch, void *_ctx)
5078 : {
5079 1 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(io_ch);
5080 :
5081 1 : bdev->internal.temporary_queue_depth += ch->io_outstanding;
5082 1 : spdk_bdev_for_each_channel_continue(i, 0);
5083 1 : }
5084 :
5085 : static int
5086 1 : bdev_calculate_measured_queue_depth(void *ctx)
5087 : {
5088 1 : struct spdk_bdev *bdev = ctx;
5089 :
5090 1 : bdev->internal.qd_poll_in_progress = true;
5091 1 : bdev->internal.temporary_queue_depth = 0;
5092 1 : spdk_bdev_for_each_channel(bdev, _calculate_measured_qd, bdev, _calculate_measured_qd_cpl);
5093 1 : return SPDK_POLLER_BUSY;
5094 : }
5095 :
5096 : static void
5097 5 : bdev_update_qd_sampling_period(void *ctx)
5098 : {
5099 5 : struct spdk_bdev *bdev = ctx;
5100 :
5101 5 : if (bdev->internal.period == bdev->internal.new_period) {
5102 0 : return;
5103 : }
5104 :
5105 5 : if (bdev->internal.qd_poll_in_progress) {
5106 1 : return;
5107 : }
5108 :
5109 4 : bdev->internal.period = bdev->internal.new_period;
5110 :
5111 4 : spdk_poller_unregister(&bdev->internal.qd_poller);
5112 4 : if (bdev->internal.period != 0) {
5113 2 : bdev->internal.qd_poller = SPDK_POLLER_REGISTER(bdev_calculate_measured_queue_depth,
5114 : bdev, bdev->internal.period);
5115 : } else {
5116 2 : spdk_bdev_close(bdev->internal.qd_desc);
5117 2 : bdev->internal.qd_desc = NULL;
5118 : }
5119 : }
5120 :
5121 : static void
5122 0 : _tmp_bdev_event_cb(enum spdk_bdev_event_type type, struct spdk_bdev *bdev, void *ctx)
5123 : {
5124 0 : SPDK_NOTICELOG("Unexpected event type: %d\n", type);
5125 0 : }
5126 :
5127 : void
5128 129 : spdk_bdev_set_qd_sampling_period(struct spdk_bdev *bdev, uint64_t period)
5129 : {
5130 : int rc;
5131 :
5132 129 : if (bdev->internal.new_period == period) {
5133 123 : return;
5134 : }
5135 :
5136 6 : bdev->internal.new_period = period;
5137 :
5138 6 : if (bdev->internal.qd_desc != NULL) {
5139 4 : assert(bdev->internal.period != 0);
5140 :
5141 4 : spdk_thread_send_msg(bdev->internal.qd_desc->thread,
5142 : bdev_update_qd_sampling_period, bdev);
5143 4 : return;
5144 : }
5145 :
5146 2 : assert(bdev->internal.period == 0);
5147 :
5148 2 : rc = spdk_bdev_open_ext(spdk_bdev_get_name(bdev), false, _tmp_bdev_event_cb,
5149 : NULL, &bdev->internal.qd_desc);
5150 2 : if (rc != 0) {
5151 0 : return;
5152 : }
5153 :
5154 2 : bdev->internal.period = period;
5155 2 : bdev->internal.qd_poller = SPDK_POLLER_REGISTER(bdev_calculate_measured_queue_depth,
5156 : bdev, period);
5157 : }
5158 :
5159 : struct bdev_get_current_qd_ctx {
5160 : uint64_t current_qd;
5161 : spdk_bdev_get_current_qd_cb cb_fn;
5162 : void *cb_arg;
5163 : };
5164 :
5165 : static void
5166 0 : bdev_get_current_qd_done(struct spdk_bdev *bdev, void *_ctx, int status)
5167 : {
5168 0 : struct bdev_get_current_qd_ctx *ctx = _ctx;
5169 :
5170 0 : ctx->cb_fn(bdev, ctx->current_qd, ctx->cb_arg, 0);
5171 :
5172 0 : free(ctx);
5173 0 : }
5174 :
5175 : static void
5176 0 : bdev_get_current_qd(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
5177 : struct spdk_io_channel *io_ch, void *_ctx)
5178 : {
5179 0 : struct bdev_get_current_qd_ctx *ctx = _ctx;
5180 0 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
5181 :
5182 0 : ctx->current_qd += bdev_ch->io_outstanding;
5183 :
5184 0 : spdk_bdev_for_each_channel_continue(i, 0);
5185 0 : }
5186 :
5187 : void
5188 0 : spdk_bdev_get_current_qd(struct spdk_bdev *bdev, spdk_bdev_get_current_qd_cb cb_fn,
5189 : void *cb_arg)
5190 : {
5191 : struct bdev_get_current_qd_ctx *ctx;
5192 :
5193 0 : assert(cb_fn != NULL);
5194 :
5195 0 : ctx = calloc(1, sizeof(*ctx));
5196 0 : if (ctx == NULL) {
5197 0 : cb_fn(bdev, 0, cb_arg, -ENOMEM);
5198 0 : return;
5199 : }
5200 :
5201 0 : ctx->cb_fn = cb_fn;
5202 0 : ctx->cb_arg = cb_arg;
5203 :
5204 0 : spdk_bdev_for_each_channel(bdev, bdev_get_current_qd, ctx, bdev_get_current_qd_done);
5205 : }
5206 :
5207 : static void
5208 25 : _event_notify(struct spdk_bdev_desc *desc, enum spdk_bdev_event_type type)
5209 : {
5210 25 : assert(desc->thread == spdk_get_thread());
5211 :
5212 25 : spdk_spin_lock(&desc->spinlock);
5213 25 : desc->refs--;
5214 25 : if (!desc->closed) {
5215 14 : spdk_spin_unlock(&desc->spinlock);
5216 14 : desc->callback.event_fn(type,
5217 : desc->bdev,
5218 : desc->callback.ctx);
5219 14 : return;
5220 11 : } else if (desc->refs == 0) {
5221 : /* This descriptor was closed after this event_notify message was sent.
5222 : * spdk_bdev_close() could not free the descriptor since this message was
5223 : * in flight, so we free it now using bdev_desc_free().
5224 : */
5225 10 : spdk_spin_unlock(&desc->spinlock);
5226 10 : bdev_desc_free(desc);
5227 10 : return;
5228 : }
5229 1 : spdk_spin_unlock(&desc->spinlock);
5230 : }
5231 :
5232 : static void
5233 25 : event_notify(struct spdk_bdev_desc *desc, spdk_msg_fn event_notify_fn)
5234 : {
5235 25 : spdk_spin_lock(&desc->spinlock);
5236 25 : desc->refs++;
5237 25 : spdk_thread_send_msg(desc->thread, event_notify_fn, desc);
5238 25 : spdk_spin_unlock(&desc->spinlock);
5239 25 : }
5240 :
5241 : static void
5242 6 : _resize_notify(void *ctx)
5243 : {
5244 6 : struct spdk_bdev_desc *desc = ctx;
5245 :
5246 6 : _event_notify(desc, SPDK_BDEV_EVENT_RESIZE);
5247 6 : }
5248 :
5249 : int
5250 11 : spdk_bdev_notify_blockcnt_change(struct spdk_bdev *bdev, uint64_t size)
5251 : {
5252 : struct spdk_bdev_desc *desc;
5253 : int ret;
5254 :
5255 11 : if (size == bdev->blockcnt) {
5256 0 : return 0;
5257 : }
5258 :
5259 11 : spdk_spin_lock(&bdev->internal.spinlock);
5260 :
5261 : /* bdev has open descriptors */
5262 11 : if (!TAILQ_EMPTY(&bdev->internal.open_descs) &&
5263 7 : bdev->blockcnt > size) {
5264 1 : ret = -EBUSY;
5265 : } else {
5266 10 : bdev->blockcnt = size;
5267 16 : TAILQ_FOREACH(desc, &bdev->internal.open_descs, link) {
5268 6 : event_notify(desc, _resize_notify);
5269 : }
5270 10 : ret = 0;
5271 : }
5272 :
5273 11 : spdk_spin_unlock(&bdev->internal.spinlock);
5274 :
5275 11 : return ret;
5276 : }
5277 :
5278 : /*
5279 : * Convert I/O offset and length from bytes to blocks.
5280 : *
5281 : * Returns zero on success or non-zero if the byte parameters aren't divisible by the block size.
5282 : */
5283 : static uint64_t
5284 20 : bdev_bytes_to_blocks(struct spdk_bdev *bdev, uint64_t offset_bytes, uint64_t *offset_blocks,
5285 : uint64_t num_bytes, uint64_t *num_blocks)
5286 : {
5287 20 : uint32_t block_size = bdev->blocklen;
5288 : uint8_t shift_cnt;
5289 :
5290 : /* Avoid expensive div operations if possible. These spdk_u32 functions are very cheap. */
5291 20 : if (spdk_likely(spdk_u32_is_pow2(block_size))) {
5292 17 : shift_cnt = spdk_u32log2(block_size);
5293 17 : *offset_blocks = offset_bytes >> shift_cnt;
5294 17 : *num_blocks = num_bytes >> shift_cnt;
5295 17 : return (offset_bytes - (*offset_blocks << shift_cnt)) |
5296 17 : (num_bytes - (*num_blocks << shift_cnt));
5297 : } else {
5298 3 : *offset_blocks = offset_bytes / block_size;
5299 3 : *num_blocks = num_bytes / block_size;
5300 3 : return (offset_bytes % block_size) | (num_bytes % block_size);
5301 : }
5302 : }
5303 :
5304 : static bool
5305 689 : bdev_io_valid_blocks(struct spdk_bdev *bdev, uint64_t offset_blocks, uint64_t num_blocks)
5306 : {
5307 : /* Return failure if offset_blocks + num_blocks is less than offset_blocks; indicates there
5308 : * has been an overflow and hence the offset has been wrapped around */
5309 689 : if (offset_blocks + num_blocks < offset_blocks) {
5310 1 : return false;
5311 : }
5312 :
5313 : /* Return failure if offset_blocks + num_blocks exceeds the size of the bdev */
5314 688 : if (offset_blocks + num_blocks > bdev->blockcnt) {
5315 2 : return false;
5316 : }
5317 :
5318 686 : return true;
5319 : }
5320 :
5321 : static void
5322 2 : bdev_seek_complete_cb(void *ctx)
5323 : {
5324 2 : struct spdk_bdev_io *bdev_io = ctx;
5325 :
5326 2 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
5327 2 : bdev_io->internal.cb(bdev_io, true, bdev_io->internal.caller_ctx);
5328 2 : }
5329 :
5330 : static int
5331 4 : bdev_seek(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5332 : uint64_t offset_blocks, enum spdk_bdev_io_type io_type,
5333 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5334 : {
5335 4 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5336 : struct spdk_bdev_io *bdev_io;
5337 4 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5338 :
5339 4 : assert(io_type == SPDK_BDEV_IO_TYPE_SEEK_DATA || io_type == SPDK_BDEV_IO_TYPE_SEEK_HOLE);
5340 :
5341 : /* Check if offset_blocks is valid looking at the validity of one block */
5342 4 : if (!bdev_io_valid_blocks(bdev, offset_blocks, 1)) {
5343 0 : return -EINVAL;
5344 : }
5345 :
5346 4 : bdev_io = bdev_channel_get_io(channel);
5347 4 : if (!bdev_io) {
5348 0 : return -ENOMEM;
5349 : }
5350 :
5351 4 : bdev_io->internal.ch = channel;
5352 4 : bdev_io->internal.desc = desc;
5353 4 : bdev_io->type = io_type;
5354 4 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5355 4 : bdev_io->u.bdev.memory_domain = NULL;
5356 4 : bdev_io->u.bdev.memory_domain_ctx = NULL;
5357 4 : bdev_io->u.bdev.accel_sequence = NULL;
5358 4 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5359 :
5360 4 : if (!spdk_bdev_io_type_supported(bdev, io_type)) {
5361 : /* In case bdev doesn't support seek to next data/hole offset,
5362 : * it is assumed that only data and no holes are present */
5363 2 : if (io_type == SPDK_BDEV_IO_TYPE_SEEK_DATA) {
5364 1 : bdev_io->u.bdev.seek.offset = offset_blocks;
5365 : } else {
5366 1 : bdev_io->u.bdev.seek.offset = UINT64_MAX;
5367 : }
5368 :
5369 2 : spdk_thread_send_msg(spdk_get_thread(), bdev_seek_complete_cb, bdev_io);
5370 2 : return 0;
5371 : }
5372 :
5373 2 : bdev_io_submit(bdev_io);
5374 2 : return 0;
5375 : }
5376 :
5377 : int
5378 2 : spdk_bdev_seek_data(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5379 : uint64_t offset_blocks,
5380 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5381 : {
5382 2 : return bdev_seek(desc, ch, offset_blocks, SPDK_BDEV_IO_TYPE_SEEK_DATA, cb, cb_arg);
5383 : }
5384 :
5385 : int
5386 2 : spdk_bdev_seek_hole(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5387 : uint64_t offset_blocks,
5388 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5389 : {
5390 2 : return bdev_seek(desc, ch, offset_blocks, SPDK_BDEV_IO_TYPE_SEEK_HOLE, cb, cb_arg);
5391 : }
5392 :
5393 : uint64_t
5394 4 : spdk_bdev_io_get_seek_offset(const struct spdk_bdev_io *bdev_io)
5395 : {
5396 4 : return bdev_io->u.bdev.seek.offset;
5397 : }
5398 :
5399 : static int
5400 204 : bdev_read_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch, void *buf,
5401 : void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
5402 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5403 : {
5404 204 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5405 : struct spdk_bdev_io *bdev_io;
5406 204 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5407 :
5408 204 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
5409 0 : return -EINVAL;
5410 : }
5411 :
5412 204 : bdev_io = bdev_channel_get_io(channel);
5413 204 : if (!bdev_io) {
5414 1 : return -ENOMEM;
5415 : }
5416 :
5417 203 : bdev_io->internal.ch = channel;
5418 203 : bdev_io->internal.desc = desc;
5419 203 : bdev_io->type = SPDK_BDEV_IO_TYPE_READ;
5420 203 : bdev_io->u.bdev.iovs = &bdev_io->iov;
5421 203 : bdev_io->u.bdev.iovs[0].iov_base = buf;
5422 203 : bdev_io->u.bdev.iovs[0].iov_len = num_blocks * bdev->blocklen;
5423 203 : bdev_io->u.bdev.iovcnt = 1;
5424 203 : bdev_io->u.bdev.md_buf = md_buf;
5425 203 : bdev_io->u.bdev.num_blocks = num_blocks;
5426 203 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5427 203 : bdev_io->u.bdev.memory_domain = NULL;
5428 203 : bdev_io->u.bdev.memory_domain_ctx = NULL;
5429 203 : bdev_io->u.bdev.accel_sequence = NULL;
5430 203 : bdev_io->u.bdev.dif_check_flags = bdev->dif_check_flags;
5431 203 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5432 :
5433 203 : bdev_io_submit(bdev_io);
5434 203 : return 0;
5435 : }
5436 :
5437 : int
5438 3 : spdk_bdev_read(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5439 : void *buf, uint64_t offset, uint64_t nbytes,
5440 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5441 : {
5442 3 : uint64_t offset_blocks, num_blocks;
5443 :
5444 3 : if (bdev_bytes_to_blocks(spdk_bdev_desc_get_bdev(desc), offset, &offset_blocks,
5445 : nbytes, &num_blocks) != 0) {
5446 0 : return -EINVAL;
5447 : }
5448 :
5449 3 : return spdk_bdev_read_blocks(desc, ch, buf, offset_blocks, num_blocks, cb, cb_arg);
5450 : }
5451 :
5452 : int
5453 200 : spdk_bdev_read_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5454 : void *buf, uint64_t offset_blocks, uint64_t num_blocks,
5455 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5456 : {
5457 200 : return bdev_read_blocks_with_md(desc, ch, buf, NULL, offset_blocks, num_blocks, cb, cb_arg);
5458 : }
5459 :
5460 : int
5461 4 : spdk_bdev_read_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5462 : void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
5463 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5464 : {
5465 4 : struct iovec iov = {
5466 : .iov_base = buf,
5467 : };
5468 :
5469 4 : if (md_buf && !spdk_bdev_is_md_separate(spdk_bdev_desc_get_bdev(desc))) {
5470 0 : return -EINVAL;
5471 : }
5472 :
5473 4 : if (md_buf && !_is_buf_allocated(&iov)) {
5474 0 : return -EINVAL;
5475 : }
5476 :
5477 4 : return bdev_read_blocks_with_md(desc, ch, buf, md_buf, offset_blocks, num_blocks,
5478 : cb, cb_arg);
5479 : }
5480 :
5481 : int
5482 5 : spdk_bdev_readv(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5483 : struct iovec *iov, int iovcnt,
5484 : uint64_t offset, uint64_t nbytes,
5485 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5486 : {
5487 5 : uint64_t offset_blocks, num_blocks;
5488 :
5489 5 : if (bdev_bytes_to_blocks(spdk_bdev_desc_get_bdev(desc), offset, &offset_blocks,
5490 : nbytes, &num_blocks) != 0) {
5491 0 : return -EINVAL;
5492 : }
5493 :
5494 5 : return spdk_bdev_readv_blocks(desc, ch, iov, iovcnt, offset_blocks, num_blocks, cb, cb_arg);
5495 : }
5496 :
5497 : static int
5498 226 : bdev_readv_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5499 : struct iovec *iov, int iovcnt, void *md_buf, uint64_t offset_blocks,
5500 : uint64_t num_blocks, struct spdk_memory_domain *domain, void *domain_ctx,
5501 : struct spdk_accel_sequence *seq, uint32_t dif_check_flags,
5502 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5503 : {
5504 226 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5505 : struct spdk_bdev_io *bdev_io;
5506 226 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5507 :
5508 226 : if (spdk_unlikely(!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks))) {
5509 0 : return -EINVAL;
5510 : }
5511 :
5512 226 : bdev_io = bdev_channel_get_io(channel);
5513 226 : if (spdk_unlikely(!bdev_io)) {
5514 2 : return -ENOMEM;
5515 : }
5516 :
5517 224 : bdev_io->internal.ch = channel;
5518 224 : bdev_io->internal.desc = desc;
5519 224 : bdev_io->type = SPDK_BDEV_IO_TYPE_READ;
5520 224 : bdev_io->u.bdev.iovs = iov;
5521 224 : bdev_io->u.bdev.iovcnt = iovcnt;
5522 224 : bdev_io->u.bdev.md_buf = md_buf;
5523 224 : bdev_io->u.bdev.num_blocks = num_blocks;
5524 224 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5525 224 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5526 :
5527 224 : if (seq != NULL) {
5528 0 : bdev_io->internal.f.has_accel_sequence = true;
5529 0 : bdev_io->internal.accel_sequence = seq;
5530 : }
5531 :
5532 224 : if (domain != NULL) {
5533 2 : bdev_io->internal.f.has_memory_domain = true;
5534 2 : bdev_io->internal.memory_domain = domain;
5535 2 : bdev_io->internal.memory_domain_ctx = domain_ctx;
5536 : }
5537 :
5538 224 : bdev_io->u.bdev.memory_domain = domain;
5539 224 : bdev_io->u.bdev.memory_domain_ctx = domain_ctx;
5540 224 : bdev_io->u.bdev.accel_sequence = seq;
5541 224 : bdev_io->u.bdev.dif_check_flags = dif_check_flags;
5542 :
5543 224 : _bdev_io_submit_ext(desc, bdev_io);
5544 :
5545 224 : return 0;
5546 : }
5547 :
5548 : int
5549 21 : spdk_bdev_readv_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5550 : struct iovec *iov, int iovcnt,
5551 : uint64_t offset_blocks, uint64_t num_blocks,
5552 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5553 : {
5554 21 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5555 :
5556 21 : return bdev_readv_blocks_with_md(desc, ch, iov, iovcnt, NULL, offset_blocks,
5557 : num_blocks, NULL, NULL, NULL, bdev->dif_check_flags, cb, cb_arg);
5558 : }
5559 :
5560 : int
5561 4 : spdk_bdev_readv_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5562 : struct iovec *iov, int iovcnt, void *md_buf,
5563 : uint64_t offset_blocks, uint64_t num_blocks,
5564 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5565 : {
5566 4 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5567 :
5568 4 : if (md_buf && !spdk_bdev_is_md_separate(bdev)) {
5569 0 : return -EINVAL;
5570 : }
5571 :
5572 4 : if (md_buf && !_is_buf_allocated(iov)) {
5573 0 : return -EINVAL;
5574 : }
5575 :
5576 4 : return bdev_readv_blocks_with_md(desc, ch, iov, iovcnt, md_buf, offset_blocks,
5577 : num_blocks, NULL, NULL, NULL, bdev->dif_check_flags, cb, cb_arg);
5578 : }
5579 :
5580 : static inline bool
5581 14 : _bdev_io_check_opts(struct spdk_bdev_ext_io_opts *opts, struct iovec *iov)
5582 : {
5583 : /*
5584 : * We check if opts size is at least of size when we first introduced
5585 : * spdk_bdev_ext_io_opts (ac6f2bdd8d) since access to those members
5586 : * are not checked internal.
5587 : */
5588 14 : return opts->size >= offsetof(struct spdk_bdev_ext_io_opts, metadata) +
5589 10 : sizeof(opts->metadata) &&
5590 22 : opts->size <= sizeof(*opts) &&
5591 : /* When memory domain is used, the user must provide data buffers */
5592 8 : (!opts->memory_domain || (iov && iov[0].iov_base));
5593 : }
5594 :
5595 : int
5596 8 : spdk_bdev_readv_blocks_ext(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5597 : struct iovec *iov, int iovcnt,
5598 : uint64_t offset_blocks, uint64_t num_blocks,
5599 : spdk_bdev_io_completion_cb cb, void *cb_arg,
5600 : struct spdk_bdev_ext_io_opts *opts)
5601 : {
5602 8 : struct spdk_memory_domain *domain = NULL;
5603 8 : struct spdk_accel_sequence *seq = NULL;
5604 8 : void *domain_ctx = NULL, *md = NULL;
5605 8 : uint32_t dif_check_flags = 0;
5606 8 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5607 :
5608 8 : if (opts) {
5609 7 : if (spdk_unlikely(!_bdev_io_check_opts(opts, iov))) {
5610 3 : return -EINVAL;
5611 : }
5612 :
5613 4 : md = opts->metadata;
5614 4 : domain = bdev_get_ext_io_opt(opts, memory_domain, NULL);
5615 4 : domain_ctx = bdev_get_ext_io_opt(opts, memory_domain_ctx, NULL);
5616 4 : seq = bdev_get_ext_io_opt(opts, accel_sequence, NULL);
5617 4 : if (md) {
5618 4 : if (spdk_unlikely(!spdk_bdev_is_md_separate(bdev))) {
5619 0 : return -EINVAL;
5620 : }
5621 :
5622 4 : if (spdk_unlikely(!_is_buf_allocated(iov))) {
5623 0 : return -EINVAL;
5624 : }
5625 :
5626 4 : if (spdk_unlikely(seq != NULL)) {
5627 0 : return -EINVAL;
5628 : }
5629 : }
5630 : }
5631 :
5632 5 : dif_check_flags = bdev->dif_check_flags &
5633 5 : ~(bdev_get_ext_io_opt(opts, dif_check_flags_exclude_mask, 0));
5634 :
5635 5 : return bdev_readv_blocks_with_md(desc, ch, iov, iovcnt, md, offset_blocks,
5636 : num_blocks, domain, domain_ctx, seq, dif_check_flags, cb, cb_arg);
5637 : }
5638 :
5639 : static int
5640 36 : bdev_write_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5641 : void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
5642 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5643 : {
5644 36 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5645 : struct spdk_bdev_io *bdev_io;
5646 36 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5647 :
5648 36 : if (!desc->write) {
5649 0 : return -EBADF;
5650 : }
5651 :
5652 36 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
5653 0 : return -EINVAL;
5654 : }
5655 :
5656 36 : bdev_io = bdev_channel_get_io(channel);
5657 36 : if (!bdev_io) {
5658 0 : return -ENOMEM;
5659 : }
5660 :
5661 36 : bdev_io->internal.ch = channel;
5662 36 : bdev_io->internal.desc = desc;
5663 36 : bdev_io->type = SPDK_BDEV_IO_TYPE_WRITE;
5664 36 : bdev_io->u.bdev.iovs = &bdev_io->iov;
5665 36 : bdev_io->u.bdev.iovs[0].iov_base = buf;
5666 36 : bdev_io->u.bdev.iovs[0].iov_len = num_blocks * bdev->blocklen;
5667 36 : bdev_io->u.bdev.iovcnt = 1;
5668 36 : bdev_io->u.bdev.md_buf = md_buf;
5669 36 : bdev_io->u.bdev.num_blocks = num_blocks;
5670 36 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5671 36 : bdev_io->u.bdev.memory_domain = NULL;
5672 36 : bdev_io->u.bdev.memory_domain_ctx = NULL;
5673 36 : bdev_io->u.bdev.accel_sequence = NULL;
5674 36 : bdev_io->u.bdev.dif_check_flags = bdev->dif_check_flags;
5675 36 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5676 :
5677 36 : bdev_io_submit(bdev_io);
5678 36 : return 0;
5679 : }
5680 :
5681 : int
5682 3 : spdk_bdev_write(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5683 : void *buf, uint64_t offset, uint64_t nbytes,
5684 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5685 : {
5686 3 : uint64_t offset_blocks, num_blocks;
5687 :
5688 3 : if (bdev_bytes_to_blocks(spdk_bdev_desc_get_bdev(desc), offset, &offset_blocks,
5689 : nbytes, &num_blocks) != 0) {
5690 0 : return -EINVAL;
5691 : }
5692 :
5693 3 : return spdk_bdev_write_blocks(desc, ch, buf, offset_blocks, num_blocks, cb, cb_arg);
5694 : }
5695 :
5696 : int
5697 27 : spdk_bdev_write_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5698 : void *buf, uint64_t offset_blocks, uint64_t num_blocks,
5699 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5700 : {
5701 27 : return bdev_write_blocks_with_md(desc, ch, buf, NULL, offset_blocks, num_blocks,
5702 : cb, cb_arg);
5703 : }
5704 :
5705 : int
5706 3 : spdk_bdev_write_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5707 : void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
5708 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5709 : {
5710 3 : struct iovec iov = {
5711 : .iov_base = buf,
5712 : };
5713 :
5714 3 : if (md_buf && !spdk_bdev_is_md_separate(spdk_bdev_desc_get_bdev(desc))) {
5715 0 : return -EINVAL;
5716 : }
5717 :
5718 3 : if (md_buf && !_is_buf_allocated(&iov)) {
5719 0 : return -EINVAL;
5720 : }
5721 :
5722 3 : return bdev_write_blocks_with_md(desc, ch, buf, md_buf, offset_blocks, num_blocks,
5723 : cb, cb_arg);
5724 : }
5725 :
5726 : static int
5727 70 : bdev_writev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5728 : struct iovec *iov, int iovcnt, void *md_buf,
5729 : uint64_t offset_blocks, uint64_t num_blocks,
5730 : struct spdk_memory_domain *domain, void *domain_ctx,
5731 : struct spdk_accel_sequence *seq, uint32_t dif_check_flags,
5732 : uint32_t nvme_cdw12_raw, uint32_t nvme_cdw13_raw,
5733 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5734 : {
5735 70 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5736 : struct spdk_bdev_io *bdev_io;
5737 70 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5738 :
5739 70 : if (spdk_unlikely(!desc->write)) {
5740 0 : return -EBADF;
5741 : }
5742 :
5743 70 : if (spdk_unlikely(!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks))) {
5744 0 : return -EINVAL;
5745 : }
5746 :
5747 70 : bdev_io = bdev_channel_get_io(channel);
5748 70 : if (spdk_unlikely(!bdev_io)) {
5749 2 : return -ENOMEM;
5750 : }
5751 :
5752 68 : bdev_io->internal.ch = channel;
5753 68 : bdev_io->internal.desc = desc;
5754 68 : bdev_io->type = SPDK_BDEV_IO_TYPE_WRITE;
5755 68 : bdev_io->u.bdev.iovs = iov;
5756 68 : bdev_io->u.bdev.iovcnt = iovcnt;
5757 68 : bdev_io->u.bdev.md_buf = md_buf;
5758 68 : bdev_io->u.bdev.num_blocks = num_blocks;
5759 68 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5760 68 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5761 68 : if (seq != NULL) {
5762 0 : bdev_io->internal.f.has_accel_sequence = true;
5763 0 : bdev_io->internal.accel_sequence = seq;
5764 : }
5765 :
5766 68 : if (domain != NULL) {
5767 2 : bdev_io->internal.f.has_memory_domain = true;
5768 2 : bdev_io->internal.memory_domain = domain;
5769 2 : bdev_io->internal.memory_domain_ctx = domain_ctx;
5770 : }
5771 :
5772 68 : bdev_io->u.bdev.memory_domain = domain;
5773 68 : bdev_io->u.bdev.memory_domain_ctx = domain_ctx;
5774 68 : bdev_io->u.bdev.accel_sequence = seq;
5775 68 : bdev_io->u.bdev.dif_check_flags = dif_check_flags;
5776 68 : bdev_io->u.bdev.nvme_cdw12.raw = nvme_cdw12_raw;
5777 68 : bdev_io->u.bdev.nvme_cdw13.raw = nvme_cdw13_raw;
5778 :
5779 68 : _bdev_io_submit_ext(desc, bdev_io);
5780 :
5781 68 : return 0;
5782 : }
5783 :
5784 : int
5785 3 : spdk_bdev_writev(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5786 : struct iovec *iov, int iovcnt,
5787 : uint64_t offset, uint64_t len,
5788 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5789 : {
5790 3 : uint64_t offset_blocks, num_blocks;
5791 :
5792 3 : if (bdev_bytes_to_blocks(spdk_bdev_desc_get_bdev(desc), offset, &offset_blocks,
5793 : len, &num_blocks) != 0) {
5794 0 : return -EINVAL;
5795 : }
5796 :
5797 3 : return spdk_bdev_writev_blocks(desc, ch, iov, iovcnt, offset_blocks, num_blocks, cb, cb_arg);
5798 : }
5799 :
5800 : int
5801 14 : spdk_bdev_writev_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5802 : struct iovec *iov, int iovcnt,
5803 : uint64_t offset_blocks, uint64_t num_blocks,
5804 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5805 : {
5806 14 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5807 :
5808 14 : return bdev_writev_blocks_with_md(desc, ch, iov, iovcnt, NULL, offset_blocks,
5809 : num_blocks, NULL, NULL, NULL, bdev->dif_check_flags, 0, 0,
5810 : cb, cb_arg);
5811 : }
5812 :
5813 : int
5814 1 : spdk_bdev_writev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5815 : struct iovec *iov, int iovcnt, void *md_buf,
5816 : uint64_t offset_blocks, uint64_t num_blocks,
5817 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5818 : {
5819 1 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5820 :
5821 1 : if (md_buf && !spdk_bdev_is_md_separate(bdev)) {
5822 0 : return -EINVAL;
5823 : }
5824 :
5825 1 : if (md_buf && !_is_buf_allocated(iov)) {
5826 0 : return -EINVAL;
5827 : }
5828 :
5829 1 : return bdev_writev_blocks_with_md(desc, ch, iov, iovcnt, md_buf, offset_blocks,
5830 : num_blocks, NULL, NULL, NULL, bdev->dif_check_flags, 0, 0,
5831 : cb, cb_arg);
5832 : }
5833 :
5834 : int
5835 8 : spdk_bdev_writev_blocks_ext(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5836 : struct iovec *iov, int iovcnt,
5837 : uint64_t offset_blocks, uint64_t num_blocks,
5838 : spdk_bdev_io_completion_cb cb, void *cb_arg,
5839 : struct spdk_bdev_ext_io_opts *opts)
5840 : {
5841 8 : struct spdk_memory_domain *domain = NULL;
5842 8 : struct spdk_accel_sequence *seq = NULL;
5843 8 : void *domain_ctx = NULL, *md = NULL;
5844 8 : uint32_t dif_check_flags = 0;
5845 8 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5846 8 : uint32_t nvme_cdw12_raw = 0;
5847 8 : uint32_t nvme_cdw13_raw = 0;
5848 :
5849 8 : if (opts) {
5850 7 : if (spdk_unlikely(!_bdev_io_check_opts(opts, iov))) {
5851 3 : return -EINVAL;
5852 : }
5853 4 : md = opts->metadata;
5854 4 : domain = bdev_get_ext_io_opt(opts, memory_domain, NULL);
5855 4 : domain_ctx = bdev_get_ext_io_opt(opts, memory_domain_ctx, NULL);
5856 4 : seq = bdev_get_ext_io_opt(opts, accel_sequence, NULL);
5857 4 : nvme_cdw12_raw = bdev_get_ext_io_opt(opts, nvme_cdw12.raw, 0);
5858 4 : nvme_cdw13_raw = bdev_get_ext_io_opt(opts, nvme_cdw13.raw, 0);
5859 4 : if (md) {
5860 4 : if (spdk_unlikely(!spdk_bdev_is_md_separate(bdev))) {
5861 0 : return -EINVAL;
5862 : }
5863 :
5864 4 : if (spdk_unlikely(!_is_buf_allocated(iov))) {
5865 0 : return -EINVAL;
5866 : }
5867 :
5868 4 : if (spdk_unlikely(seq != NULL)) {
5869 0 : return -EINVAL;
5870 : }
5871 : }
5872 : }
5873 :
5874 5 : dif_check_flags = bdev->dif_check_flags &
5875 5 : ~(bdev_get_ext_io_opt(opts, dif_check_flags_exclude_mask, 0));
5876 :
5877 5 : return bdev_writev_blocks_with_md(desc, ch, iov, iovcnt, md, offset_blocks, num_blocks,
5878 : domain, domain_ctx, seq, dif_check_flags,
5879 : nvme_cdw12_raw, nvme_cdw13_raw, cb, cb_arg);
5880 : }
5881 :
5882 : static void
5883 11 : bdev_compare_do_read_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
5884 : {
5885 11 : struct spdk_bdev_io *parent_io = cb_arg;
5886 11 : struct spdk_bdev *bdev = parent_io->bdev;
5887 11 : uint8_t *read_buf = bdev_io->u.bdev.iovs[0].iov_base;
5888 11 : int i, rc = 0;
5889 :
5890 11 : if (!success) {
5891 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
5892 0 : parent_io->internal.cb(parent_io, false, parent_io->internal.caller_ctx);
5893 0 : spdk_bdev_free_io(bdev_io);
5894 0 : return;
5895 : }
5896 :
5897 17 : for (i = 0; i < parent_io->u.bdev.iovcnt; i++) {
5898 22 : rc = memcmp(read_buf,
5899 11 : parent_io->u.bdev.iovs[i].iov_base,
5900 11 : parent_io->u.bdev.iovs[i].iov_len);
5901 11 : if (rc) {
5902 5 : break;
5903 : }
5904 6 : read_buf += parent_io->u.bdev.iovs[i].iov_len;
5905 : }
5906 :
5907 11 : if (rc == 0 && parent_io->u.bdev.md_buf && spdk_bdev_is_md_separate(bdev)) {
5908 2 : rc = memcmp(bdev_io->u.bdev.md_buf,
5909 2 : parent_io->u.bdev.md_buf,
5910 2 : spdk_bdev_get_md_size(bdev));
5911 : }
5912 :
5913 11 : spdk_bdev_free_io(bdev_io);
5914 :
5915 11 : if (rc == 0) {
5916 5 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
5917 5 : parent_io->internal.cb(parent_io, true, parent_io->internal.caller_ctx);
5918 : } else {
5919 6 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_MISCOMPARE;
5920 6 : parent_io->internal.cb(parent_io, false, parent_io->internal.caller_ctx);
5921 : }
5922 : }
5923 :
5924 : static void
5925 11 : bdev_compare_do_read(void *_bdev_io)
5926 : {
5927 11 : struct spdk_bdev_io *bdev_io = _bdev_io;
5928 : int rc;
5929 :
5930 22 : rc = spdk_bdev_read_blocks(bdev_io->internal.desc,
5931 11 : spdk_io_channel_from_ctx(bdev_io->internal.ch), NULL,
5932 : bdev_io->u.bdev.offset_blocks, bdev_io->u.bdev.num_blocks,
5933 : bdev_compare_do_read_done, bdev_io);
5934 :
5935 11 : if (rc == -ENOMEM) {
5936 0 : bdev_queue_io_wait_with_cb(bdev_io, bdev_compare_do_read);
5937 11 : } else if (rc != 0) {
5938 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
5939 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
5940 : }
5941 11 : }
5942 :
5943 : static int
5944 16 : bdev_comparev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5945 : struct iovec *iov, int iovcnt, void *md_buf,
5946 : uint64_t offset_blocks, uint64_t num_blocks,
5947 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5948 : {
5949 16 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5950 : struct spdk_bdev_io *bdev_io;
5951 16 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5952 :
5953 16 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
5954 0 : return -EINVAL;
5955 : }
5956 :
5957 16 : bdev_io = bdev_channel_get_io(channel);
5958 16 : if (!bdev_io) {
5959 0 : return -ENOMEM;
5960 : }
5961 :
5962 16 : bdev_io->internal.ch = channel;
5963 16 : bdev_io->internal.desc = desc;
5964 16 : bdev_io->type = SPDK_BDEV_IO_TYPE_COMPARE;
5965 16 : bdev_io->u.bdev.iovs = iov;
5966 16 : bdev_io->u.bdev.iovcnt = iovcnt;
5967 16 : bdev_io->u.bdev.md_buf = md_buf;
5968 16 : bdev_io->u.bdev.num_blocks = num_blocks;
5969 16 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5970 16 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5971 16 : bdev_io->u.bdev.memory_domain = NULL;
5972 16 : bdev_io->u.bdev.memory_domain_ctx = NULL;
5973 16 : bdev_io->u.bdev.accel_sequence = NULL;
5974 :
5975 16 : if (bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COMPARE)) {
5976 7 : bdev_io_submit(bdev_io);
5977 7 : return 0;
5978 : }
5979 :
5980 9 : bdev_compare_do_read(bdev_io);
5981 :
5982 9 : return 0;
5983 : }
5984 :
5985 : int
5986 10 : spdk_bdev_comparev_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5987 : struct iovec *iov, int iovcnt,
5988 : uint64_t offset_blocks, uint64_t num_blocks,
5989 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5990 : {
5991 10 : return bdev_comparev_blocks_with_md(desc, ch, iov, iovcnt, NULL, offset_blocks,
5992 : num_blocks, cb, cb_arg);
5993 : }
5994 :
5995 : int
5996 6 : spdk_bdev_comparev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5997 : struct iovec *iov, int iovcnt, void *md_buf,
5998 : uint64_t offset_blocks, uint64_t num_blocks,
5999 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6000 : {
6001 6 : if (md_buf && !spdk_bdev_is_md_separate(spdk_bdev_desc_get_bdev(desc))) {
6002 0 : return -EINVAL;
6003 : }
6004 :
6005 6 : if (md_buf && !_is_buf_allocated(iov)) {
6006 0 : return -EINVAL;
6007 : }
6008 :
6009 6 : return bdev_comparev_blocks_with_md(desc, ch, iov, iovcnt, md_buf, offset_blocks,
6010 : num_blocks, cb, cb_arg);
6011 : }
6012 :
6013 : static int
6014 4 : bdev_compare_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6015 : void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
6016 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6017 : {
6018 4 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6019 : struct spdk_bdev_io *bdev_io;
6020 4 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6021 :
6022 4 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6023 0 : return -EINVAL;
6024 : }
6025 :
6026 4 : bdev_io = bdev_channel_get_io(channel);
6027 4 : if (!bdev_io) {
6028 0 : return -ENOMEM;
6029 : }
6030 :
6031 4 : bdev_io->internal.ch = channel;
6032 4 : bdev_io->internal.desc = desc;
6033 4 : bdev_io->type = SPDK_BDEV_IO_TYPE_COMPARE;
6034 4 : bdev_io->u.bdev.iovs = &bdev_io->iov;
6035 4 : bdev_io->u.bdev.iovs[0].iov_base = buf;
6036 4 : bdev_io->u.bdev.iovs[0].iov_len = num_blocks * bdev->blocklen;
6037 4 : bdev_io->u.bdev.iovcnt = 1;
6038 4 : bdev_io->u.bdev.md_buf = md_buf;
6039 4 : bdev_io->u.bdev.num_blocks = num_blocks;
6040 4 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6041 4 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6042 4 : bdev_io->u.bdev.memory_domain = NULL;
6043 4 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6044 4 : bdev_io->u.bdev.accel_sequence = NULL;
6045 :
6046 4 : if (bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COMPARE)) {
6047 2 : bdev_io_submit(bdev_io);
6048 2 : return 0;
6049 : }
6050 :
6051 2 : bdev_compare_do_read(bdev_io);
6052 :
6053 2 : return 0;
6054 : }
6055 :
6056 : int
6057 4 : spdk_bdev_compare_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6058 : void *buf, uint64_t offset_blocks, uint64_t num_blocks,
6059 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6060 : {
6061 4 : return bdev_compare_blocks_with_md(desc, ch, buf, NULL, offset_blocks, num_blocks,
6062 : cb, cb_arg);
6063 : }
6064 :
6065 : int
6066 0 : spdk_bdev_compare_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6067 : void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
6068 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6069 : {
6070 0 : struct iovec iov = {
6071 : .iov_base = buf,
6072 : };
6073 :
6074 0 : if (md_buf && !spdk_bdev_is_md_separate(spdk_bdev_desc_get_bdev(desc))) {
6075 0 : return -EINVAL;
6076 : }
6077 :
6078 0 : if (md_buf && !_is_buf_allocated(&iov)) {
6079 0 : return -EINVAL;
6080 : }
6081 :
6082 0 : return bdev_compare_blocks_with_md(desc, ch, buf, md_buf, offset_blocks, num_blocks,
6083 : cb, cb_arg);
6084 : }
6085 :
6086 : static void
6087 2 : bdev_comparev_and_writev_blocks_unlocked(struct lba_range *range, void *ctx, int unlock_status)
6088 : {
6089 2 : struct spdk_bdev_io *bdev_io = ctx;
6090 :
6091 2 : if (unlock_status) {
6092 0 : SPDK_ERRLOG("LBA range unlock failed\n");
6093 : }
6094 :
6095 2 : bdev_io->internal.cb(bdev_io, bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS ? true :
6096 : false, bdev_io->internal.caller_ctx);
6097 2 : }
6098 :
6099 : static void
6100 2 : bdev_comparev_and_writev_blocks_unlock(struct spdk_bdev_io *bdev_io, int status)
6101 : {
6102 2 : bdev_io->internal.status = status;
6103 :
6104 2 : bdev_unlock_lba_range(bdev_io->internal.desc, spdk_io_channel_from_ctx(bdev_io->internal.ch),
6105 : bdev_io->u.bdev.offset_blocks, bdev_io->u.bdev.num_blocks,
6106 : bdev_comparev_and_writev_blocks_unlocked, bdev_io);
6107 2 : }
6108 :
6109 : static void
6110 1 : bdev_compare_and_write_do_write_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
6111 : {
6112 1 : struct spdk_bdev_io *parent_io = cb_arg;
6113 :
6114 1 : if (!success) {
6115 0 : SPDK_ERRLOG("Compare and write operation failed\n");
6116 : }
6117 :
6118 1 : spdk_bdev_free_io(bdev_io);
6119 :
6120 1 : bdev_comparev_and_writev_blocks_unlock(parent_io,
6121 : success ? SPDK_BDEV_IO_STATUS_SUCCESS : SPDK_BDEV_IO_STATUS_FAILED);
6122 1 : }
6123 :
6124 : static void
6125 1 : bdev_compare_and_write_do_write(void *_bdev_io)
6126 : {
6127 1 : struct spdk_bdev_io *bdev_io = _bdev_io;
6128 : int rc;
6129 :
6130 2 : rc = spdk_bdev_writev_blocks(bdev_io->internal.desc,
6131 1 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
6132 : bdev_io->u.bdev.fused_iovs, bdev_io->u.bdev.fused_iovcnt,
6133 : bdev_io->u.bdev.offset_blocks, bdev_io->u.bdev.num_blocks,
6134 : bdev_compare_and_write_do_write_done, bdev_io);
6135 :
6136 :
6137 1 : if (rc == -ENOMEM) {
6138 0 : bdev_queue_io_wait_with_cb(bdev_io, bdev_compare_and_write_do_write);
6139 1 : } else if (rc != 0) {
6140 0 : bdev_comparev_and_writev_blocks_unlock(bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
6141 : }
6142 1 : }
6143 :
6144 : static void
6145 2 : bdev_compare_and_write_do_compare_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
6146 : {
6147 2 : struct spdk_bdev_io *parent_io = cb_arg;
6148 :
6149 2 : spdk_bdev_free_io(bdev_io);
6150 :
6151 2 : if (!success) {
6152 1 : bdev_comparev_and_writev_blocks_unlock(parent_io, SPDK_BDEV_IO_STATUS_MISCOMPARE);
6153 1 : return;
6154 : }
6155 :
6156 1 : bdev_compare_and_write_do_write(parent_io);
6157 : }
6158 :
6159 : static void
6160 2 : bdev_compare_and_write_do_compare(void *_bdev_io)
6161 : {
6162 2 : struct spdk_bdev_io *bdev_io = _bdev_io;
6163 : int rc;
6164 :
6165 4 : rc = spdk_bdev_comparev_blocks(bdev_io->internal.desc,
6166 2 : spdk_io_channel_from_ctx(bdev_io->internal.ch), bdev_io->u.bdev.iovs,
6167 : bdev_io->u.bdev.iovcnt, bdev_io->u.bdev.offset_blocks, bdev_io->u.bdev.num_blocks,
6168 : bdev_compare_and_write_do_compare_done, bdev_io);
6169 :
6170 2 : if (rc == -ENOMEM) {
6171 0 : bdev_queue_io_wait_with_cb(bdev_io, bdev_compare_and_write_do_compare);
6172 2 : } else if (rc != 0) {
6173 0 : bdev_comparev_and_writev_blocks_unlock(bdev_io, SPDK_BDEV_IO_STATUS_FIRST_FUSED_FAILED);
6174 : }
6175 2 : }
6176 :
6177 : static void
6178 2 : bdev_comparev_and_writev_blocks_locked(struct lba_range *range, void *ctx, int status)
6179 : {
6180 2 : struct spdk_bdev_io *bdev_io = ctx;
6181 :
6182 2 : if (status) {
6183 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FIRST_FUSED_FAILED;
6184 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
6185 0 : return;
6186 : }
6187 :
6188 2 : bdev_compare_and_write_do_compare(bdev_io);
6189 : }
6190 :
6191 : int
6192 2 : spdk_bdev_comparev_and_writev_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6193 : struct iovec *compare_iov, int compare_iovcnt,
6194 : struct iovec *write_iov, int write_iovcnt,
6195 : uint64_t offset_blocks, uint64_t num_blocks,
6196 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6197 : {
6198 2 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6199 : struct spdk_bdev_io *bdev_io;
6200 2 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6201 :
6202 2 : if (!desc->write) {
6203 0 : return -EBADF;
6204 : }
6205 :
6206 2 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6207 0 : return -EINVAL;
6208 : }
6209 :
6210 2 : if (num_blocks > bdev->acwu) {
6211 0 : return -EINVAL;
6212 : }
6213 :
6214 2 : bdev_io = bdev_channel_get_io(channel);
6215 2 : if (!bdev_io) {
6216 0 : return -ENOMEM;
6217 : }
6218 :
6219 2 : bdev_io->internal.ch = channel;
6220 2 : bdev_io->internal.desc = desc;
6221 2 : bdev_io->type = SPDK_BDEV_IO_TYPE_COMPARE_AND_WRITE;
6222 2 : bdev_io->u.bdev.iovs = compare_iov;
6223 2 : bdev_io->u.bdev.iovcnt = compare_iovcnt;
6224 2 : bdev_io->u.bdev.fused_iovs = write_iov;
6225 2 : bdev_io->u.bdev.fused_iovcnt = write_iovcnt;
6226 2 : bdev_io->u.bdev.md_buf = NULL;
6227 2 : bdev_io->u.bdev.num_blocks = num_blocks;
6228 2 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6229 2 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6230 2 : bdev_io->u.bdev.memory_domain = NULL;
6231 2 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6232 2 : bdev_io->u.bdev.accel_sequence = NULL;
6233 :
6234 2 : if (bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COMPARE_AND_WRITE)) {
6235 0 : bdev_io_submit(bdev_io);
6236 0 : return 0;
6237 : }
6238 :
6239 2 : return bdev_lock_lba_range(desc, ch, offset_blocks, num_blocks,
6240 : bdev_comparev_and_writev_blocks_locked, bdev_io);
6241 : }
6242 :
6243 : int
6244 2 : spdk_bdev_zcopy_start(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6245 : struct iovec *iov, int iovcnt,
6246 : uint64_t offset_blocks, uint64_t num_blocks,
6247 : bool populate,
6248 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6249 : {
6250 2 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6251 : struct spdk_bdev_io *bdev_io;
6252 2 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6253 :
6254 2 : if (!desc->write) {
6255 0 : return -EBADF;
6256 : }
6257 :
6258 2 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6259 0 : return -EINVAL;
6260 : }
6261 :
6262 2 : if (!spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_ZCOPY)) {
6263 0 : return -ENOTSUP;
6264 : }
6265 :
6266 2 : bdev_io = bdev_channel_get_io(channel);
6267 2 : if (!bdev_io) {
6268 0 : return -ENOMEM;
6269 : }
6270 :
6271 2 : bdev_io->internal.ch = channel;
6272 2 : bdev_io->internal.desc = desc;
6273 2 : bdev_io->type = SPDK_BDEV_IO_TYPE_ZCOPY;
6274 2 : bdev_io->u.bdev.num_blocks = num_blocks;
6275 2 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6276 2 : bdev_io->u.bdev.iovs = iov;
6277 2 : bdev_io->u.bdev.iovcnt = iovcnt;
6278 2 : bdev_io->u.bdev.md_buf = NULL;
6279 2 : bdev_io->u.bdev.zcopy.populate = populate ? 1 : 0;
6280 2 : bdev_io->u.bdev.zcopy.commit = 0;
6281 2 : bdev_io->u.bdev.zcopy.start = 1;
6282 2 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6283 2 : bdev_io->u.bdev.memory_domain = NULL;
6284 2 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6285 2 : bdev_io->u.bdev.accel_sequence = NULL;
6286 :
6287 2 : bdev_io_submit(bdev_io);
6288 :
6289 2 : return 0;
6290 : }
6291 :
6292 : int
6293 2 : spdk_bdev_zcopy_end(struct spdk_bdev_io *bdev_io, bool commit,
6294 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6295 : {
6296 2 : if (bdev_io->type != SPDK_BDEV_IO_TYPE_ZCOPY) {
6297 0 : return -EINVAL;
6298 : }
6299 :
6300 2 : bdev_io->u.bdev.zcopy.commit = commit ? 1 : 0;
6301 2 : bdev_io->u.bdev.zcopy.start = 0;
6302 2 : bdev_io->internal.caller_ctx = cb_arg;
6303 2 : bdev_io->internal.cb = cb;
6304 2 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_PENDING;
6305 :
6306 2 : bdev_io_submit(bdev_io);
6307 :
6308 2 : return 0;
6309 : }
6310 :
6311 : int
6312 0 : spdk_bdev_write_zeroes(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6313 : uint64_t offset, uint64_t len,
6314 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6315 : {
6316 0 : uint64_t offset_blocks, num_blocks;
6317 :
6318 0 : if (bdev_bytes_to_blocks(spdk_bdev_desc_get_bdev(desc), offset, &offset_blocks,
6319 : len, &num_blocks) != 0) {
6320 0 : return -EINVAL;
6321 : }
6322 :
6323 0 : return spdk_bdev_write_zeroes_blocks(desc, ch, offset_blocks, num_blocks, cb, cb_arg);
6324 : }
6325 :
6326 : int
6327 33 : spdk_bdev_write_zeroes_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6328 : uint64_t offset_blocks, uint64_t num_blocks,
6329 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6330 : {
6331 33 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6332 : struct spdk_bdev_io *bdev_io;
6333 33 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6334 :
6335 33 : if (!desc->write) {
6336 0 : return -EBADF;
6337 : }
6338 :
6339 33 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6340 0 : return -EINVAL;
6341 : }
6342 :
6343 33 : if (!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE_ZEROES) &&
6344 10 : !bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE)) {
6345 1 : return -ENOTSUP;
6346 : }
6347 :
6348 32 : bdev_io = bdev_channel_get_io(channel);
6349 :
6350 32 : if (!bdev_io) {
6351 0 : return -ENOMEM;
6352 : }
6353 :
6354 32 : bdev_io->type = SPDK_BDEV_IO_TYPE_WRITE_ZEROES;
6355 32 : bdev_io->internal.ch = channel;
6356 32 : bdev_io->internal.desc = desc;
6357 32 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6358 32 : bdev_io->u.bdev.num_blocks = num_blocks;
6359 32 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6360 32 : bdev_io->u.bdev.memory_domain = NULL;
6361 32 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6362 32 : bdev_io->u.bdev.accel_sequence = NULL;
6363 :
6364 : /* If the write_zeroes size is large and should be split, use the generic split
6365 : * logic regardless of whether SPDK_BDEV_IO_TYPE_WRITE_ZEREOS is supported or not.
6366 : *
6367 : * Then, send the write_zeroes request if SPDK_BDEV_IO_TYPE_WRITE_ZEROES is supported
6368 : * or emulate it using regular write request otherwise.
6369 : */
6370 32 : if (bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE_ZEROES) ||
6371 : bdev_io->internal.f.split) {
6372 26 : bdev_io_submit(bdev_io);
6373 26 : return 0;
6374 : }
6375 :
6376 6 : assert(_bdev_get_block_size_with_md(bdev) <= ZERO_BUFFER_SIZE);
6377 :
6378 6 : return bdev_write_zero_buffer(bdev_io);
6379 : }
6380 :
6381 : int
6382 0 : spdk_bdev_unmap(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6383 : uint64_t offset, uint64_t nbytes,
6384 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6385 : {
6386 0 : uint64_t offset_blocks, num_blocks;
6387 :
6388 0 : if (bdev_bytes_to_blocks(spdk_bdev_desc_get_bdev(desc), offset, &offset_blocks,
6389 : nbytes, &num_blocks) != 0) {
6390 0 : return -EINVAL;
6391 : }
6392 :
6393 0 : return spdk_bdev_unmap_blocks(desc, ch, offset_blocks, num_blocks, cb, cb_arg);
6394 : }
6395 :
6396 : static void
6397 0 : bdev_io_complete_cb(void *ctx)
6398 : {
6399 0 : struct spdk_bdev_io *bdev_io = ctx;
6400 :
6401 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
6402 0 : bdev_io->internal.cb(bdev_io, true, bdev_io->internal.caller_ctx);
6403 0 : }
6404 :
6405 : int
6406 22 : spdk_bdev_unmap_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6407 : uint64_t offset_blocks, uint64_t num_blocks,
6408 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6409 : {
6410 22 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6411 : struct spdk_bdev_io *bdev_io;
6412 22 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6413 :
6414 22 : if (!desc->write) {
6415 0 : return -EBADF;
6416 : }
6417 :
6418 22 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6419 0 : return -EINVAL;
6420 : }
6421 :
6422 22 : bdev_io = bdev_channel_get_io(channel);
6423 22 : if (!bdev_io) {
6424 0 : return -ENOMEM;
6425 : }
6426 :
6427 22 : bdev_io->internal.ch = channel;
6428 22 : bdev_io->internal.desc = desc;
6429 22 : bdev_io->type = SPDK_BDEV_IO_TYPE_UNMAP;
6430 :
6431 22 : bdev_io->u.bdev.iovs = &bdev_io->iov;
6432 22 : bdev_io->u.bdev.iovs[0].iov_base = NULL;
6433 22 : bdev_io->u.bdev.iovs[0].iov_len = 0;
6434 22 : bdev_io->u.bdev.iovcnt = 1;
6435 :
6436 22 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6437 22 : bdev_io->u.bdev.num_blocks = num_blocks;
6438 22 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6439 22 : bdev_io->u.bdev.memory_domain = NULL;
6440 22 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6441 22 : bdev_io->u.bdev.accel_sequence = NULL;
6442 :
6443 22 : if (num_blocks == 0) {
6444 0 : spdk_thread_send_msg(spdk_get_thread(), bdev_io_complete_cb, bdev_io);
6445 0 : return 0;
6446 : }
6447 :
6448 22 : bdev_io_submit(bdev_io);
6449 22 : return 0;
6450 : }
6451 :
6452 : int
6453 0 : spdk_bdev_flush(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6454 : uint64_t offset, uint64_t length,
6455 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6456 : {
6457 0 : uint64_t offset_blocks, num_blocks;
6458 :
6459 0 : if (bdev_bytes_to_blocks(spdk_bdev_desc_get_bdev(desc), offset, &offset_blocks,
6460 : length, &num_blocks) != 0) {
6461 0 : return -EINVAL;
6462 : }
6463 :
6464 0 : return spdk_bdev_flush_blocks(desc, ch, offset_blocks, num_blocks, cb, cb_arg);
6465 : }
6466 :
6467 : int
6468 2 : spdk_bdev_flush_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6469 : uint64_t offset_blocks, uint64_t num_blocks,
6470 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6471 : {
6472 2 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6473 : struct spdk_bdev_io *bdev_io;
6474 2 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6475 :
6476 2 : if (!desc->write) {
6477 0 : return -EBADF;
6478 : }
6479 :
6480 2 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6481 0 : return -EINVAL;
6482 : }
6483 :
6484 2 : bdev_io = bdev_channel_get_io(channel);
6485 2 : if (!bdev_io) {
6486 0 : return -ENOMEM;
6487 : }
6488 :
6489 2 : bdev_io->internal.ch = channel;
6490 2 : bdev_io->internal.desc = desc;
6491 2 : bdev_io->type = SPDK_BDEV_IO_TYPE_FLUSH;
6492 2 : bdev_io->u.bdev.iovs = NULL;
6493 2 : bdev_io->u.bdev.iovcnt = 0;
6494 2 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6495 2 : bdev_io->u.bdev.num_blocks = num_blocks;
6496 2 : bdev_io->u.bdev.memory_domain = NULL;
6497 2 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6498 2 : bdev_io->u.bdev.accel_sequence = NULL;
6499 2 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6500 :
6501 2 : bdev_io_submit(bdev_io);
6502 2 : return 0;
6503 : }
6504 :
6505 : static int bdev_reset_poll_for_outstanding_io(void *ctx);
6506 :
6507 : static void
6508 13 : bdev_reset_check_outstanding_io_done(struct spdk_bdev *bdev, void *_ctx, int status)
6509 : {
6510 13 : struct spdk_bdev_channel *ch = _ctx;
6511 : struct spdk_bdev_io *bdev_io;
6512 :
6513 13 : bdev_io = TAILQ_FIRST(&ch->queued_resets);
6514 :
6515 13 : if (status == -EBUSY) {
6516 9 : if (spdk_get_ticks() < bdev_io->u.reset.wait_poller.stop_time_tsc) {
6517 8 : bdev_io->u.reset.wait_poller.poller = SPDK_POLLER_REGISTER(bdev_reset_poll_for_outstanding_io,
6518 : ch, BDEV_RESET_CHECK_OUTSTANDING_IO_PERIOD);
6519 : } else {
6520 1 : TAILQ_REMOVE(&ch->queued_resets, bdev_io, internal.link);
6521 :
6522 1 : if (TAILQ_EMPTY(&ch->io_memory_domain) && TAILQ_EMPTY(&ch->io_accel_exec)) {
6523 : /* If outstanding IOs are still present and reset_io_drain_timeout
6524 : * seconds passed, start the reset. */
6525 1 : bdev_io_submit_reset(bdev_io);
6526 : } else {
6527 : /* We still have in progress memory domain pull/push or we're
6528 : * executing accel sequence. Since we cannot abort either of those
6529 : * operations, fail the reset request. */
6530 0 : spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
6531 : }
6532 : }
6533 : } else {
6534 4 : TAILQ_REMOVE(&ch->queued_resets, bdev_io, internal.link);
6535 4 : SPDK_DEBUGLOG(bdev,
6536 : "Skipping reset for underlying device of bdev: %s - no outstanding I/O.\n",
6537 : ch->bdev->name);
6538 : /* Mark the completion status as a SUCCESS and complete the reset. */
6539 4 : spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_SUCCESS);
6540 : }
6541 13 : }
6542 :
6543 : static void
6544 13 : bdev_reset_check_outstanding_io(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
6545 : struct spdk_io_channel *io_ch, void *_ctx)
6546 : {
6547 13 : struct spdk_bdev_channel *cur_ch = __io_ch_to_bdev_ch(io_ch);
6548 13 : int status = 0;
6549 :
6550 13 : if (cur_ch->io_outstanding > 0 ||
6551 4 : !TAILQ_EMPTY(&cur_ch->io_memory_domain) ||
6552 4 : !TAILQ_EMPTY(&cur_ch->io_accel_exec)) {
6553 : /* If a channel has outstanding IO, set status to -EBUSY code. This will stop
6554 : * further iteration over the rest of the channels and pass non-zero status
6555 : * to the callback function. */
6556 9 : status = -EBUSY;
6557 : }
6558 13 : spdk_bdev_for_each_channel_continue(i, status);
6559 13 : }
6560 :
6561 : static int
6562 8 : bdev_reset_poll_for_outstanding_io(void *ctx)
6563 : {
6564 8 : struct spdk_bdev_channel *ch = ctx;
6565 : struct spdk_bdev_io *bdev_io;
6566 :
6567 8 : bdev_io = TAILQ_FIRST(&ch->queued_resets);
6568 :
6569 8 : spdk_poller_unregister(&bdev_io->u.reset.wait_poller.poller);
6570 8 : spdk_bdev_for_each_channel(ch->bdev, bdev_reset_check_outstanding_io, ch,
6571 : bdev_reset_check_outstanding_io_done);
6572 :
6573 8 : return SPDK_POLLER_BUSY;
6574 : }
6575 :
6576 : static void
6577 15 : bdev_reset_freeze_channel_done(struct spdk_bdev *bdev, void *_ctx, int status)
6578 : {
6579 15 : struct spdk_bdev_channel *ch = _ctx;
6580 : struct spdk_bdev_io *bdev_io;
6581 :
6582 15 : bdev_io = TAILQ_FIRST(&ch->queued_resets);
6583 :
6584 15 : if (bdev->reset_io_drain_timeout == 0) {
6585 10 : TAILQ_REMOVE(&ch->queued_resets, bdev_io, internal.link);
6586 :
6587 10 : bdev_io_submit_reset(bdev_io);
6588 10 : return;
6589 : }
6590 :
6591 10 : bdev_io->u.reset.wait_poller.stop_time_tsc = spdk_get_ticks() +
6592 5 : (ch->bdev->reset_io_drain_timeout * spdk_get_ticks_hz());
6593 :
6594 : /* In case bdev->reset_io_drain_timeout is not equal to zero,
6595 : * submit the reset to the underlying module only if outstanding I/O
6596 : * remain after reset_io_drain_timeout seconds have passed. */
6597 5 : spdk_bdev_for_each_channel(ch->bdev, bdev_reset_check_outstanding_io, ch,
6598 : bdev_reset_check_outstanding_io_done);
6599 : }
6600 :
6601 : static void
6602 18 : bdev_reset_freeze_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
6603 : struct spdk_io_channel *ch, void *_ctx)
6604 : {
6605 : struct spdk_bdev_channel *channel;
6606 : struct spdk_bdev_mgmt_channel *mgmt_channel;
6607 : struct spdk_bdev_shared_resource *shared_resource;
6608 18 : bdev_io_tailq_t tmp_queued;
6609 :
6610 18 : TAILQ_INIT(&tmp_queued);
6611 :
6612 18 : channel = __io_ch_to_bdev_ch(ch);
6613 18 : shared_resource = channel->shared_resource;
6614 18 : mgmt_channel = shared_resource->mgmt_ch;
6615 :
6616 18 : channel->flags |= BDEV_CH_RESET_IN_PROGRESS;
6617 :
6618 18 : if ((channel->flags & BDEV_CH_QOS_ENABLED) != 0) {
6619 2 : TAILQ_SWAP(&channel->qos_queued_io, &tmp_queued, spdk_bdev_io, internal.link);
6620 : }
6621 :
6622 18 : bdev_abort_all_queued_io(&shared_resource->nomem_io, channel);
6623 18 : bdev_abort_all_buf_io(mgmt_channel, channel);
6624 18 : bdev_abort_all_queued_io(&tmp_queued, channel);
6625 :
6626 18 : spdk_bdev_for_each_channel_continue(i, 0);
6627 18 : }
6628 :
6629 : static void
6630 15 : bdev_start_reset(void *ctx)
6631 : {
6632 15 : struct spdk_bdev_channel *ch = ctx;
6633 :
6634 15 : spdk_bdev_for_each_channel(ch->bdev, bdev_reset_freeze_channel, ch,
6635 : bdev_reset_freeze_channel_done);
6636 15 : }
6637 :
6638 : static void
6639 16 : bdev_channel_start_reset(struct spdk_bdev_channel *ch)
6640 : {
6641 16 : struct spdk_bdev *bdev = ch->bdev;
6642 :
6643 16 : assert(!TAILQ_EMPTY(&ch->queued_resets));
6644 :
6645 16 : spdk_spin_lock(&bdev->internal.spinlock);
6646 16 : if (bdev->internal.reset_in_progress == NULL) {
6647 15 : bdev->internal.reset_in_progress = TAILQ_FIRST(&ch->queued_resets);
6648 : /*
6649 : * Take a channel reference for the target bdev for the life of this
6650 : * reset. This guards against the channel getting destroyed while
6651 : * spdk_bdev_for_each_channel() calls related to this reset IO are in
6652 : * progress. We will release the reference when this reset is
6653 : * completed.
6654 : */
6655 15 : bdev->internal.reset_in_progress->u.reset.ch_ref = spdk_get_io_channel(__bdev_to_io_dev(bdev));
6656 15 : bdev_start_reset(ch);
6657 : }
6658 16 : spdk_spin_unlock(&bdev->internal.spinlock);
6659 16 : }
6660 :
6661 : int
6662 16 : spdk_bdev_reset(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6663 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6664 : {
6665 16 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6666 : struct spdk_bdev_io *bdev_io;
6667 16 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6668 :
6669 16 : bdev_io = bdev_channel_get_io(channel);
6670 16 : if (!bdev_io) {
6671 0 : return -ENOMEM;
6672 : }
6673 :
6674 16 : bdev_io->internal.ch = channel;
6675 16 : bdev_io->internal.desc = desc;
6676 16 : bdev_io->internal.submit_tsc = spdk_get_ticks();
6677 16 : bdev_io->type = SPDK_BDEV_IO_TYPE_RESET;
6678 16 : bdev_io->u.reset.ch_ref = NULL;
6679 16 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6680 :
6681 16 : spdk_spin_lock(&bdev->internal.spinlock);
6682 16 : TAILQ_INSERT_TAIL(&channel->queued_resets, bdev_io, internal.link);
6683 16 : spdk_spin_unlock(&bdev->internal.spinlock);
6684 :
6685 16 : bdev_ch_add_to_io_submitted(bdev_io);
6686 :
6687 16 : bdev_channel_start_reset(channel);
6688 :
6689 16 : return 0;
6690 : }
6691 :
6692 : void
6693 0 : spdk_bdev_get_io_stat(struct spdk_bdev *bdev, struct spdk_io_channel *ch,
6694 : struct spdk_bdev_io_stat *stat, enum spdk_bdev_reset_stat_mode reset_mode)
6695 : {
6696 0 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6697 :
6698 0 : bdev_get_io_stat(stat, channel->stat);
6699 0 : spdk_bdev_reset_io_stat(stat, reset_mode);
6700 0 : }
6701 :
6702 : static void
6703 5 : bdev_get_device_stat_done(struct spdk_bdev *bdev, void *_ctx, int status)
6704 : {
6705 5 : struct spdk_bdev_iostat_ctx *bdev_iostat_ctx = _ctx;
6706 :
6707 5 : bdev_iostat_ctx->cb(bdev, bdev_iostat_ctx->stat,
6708 : bdev_iostat_ctx->cb_arg, 0);
6709 5 : free(bdev_iostat_ctx);
6710 5 : }
6711 :
6712 : static void
6713 4 : bdev_get_each_channel_stat(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
6714 : struct spdk_io_channel *ch, void *_ctx)
6715 : {
6716 4 : struct spdk_bdev_iostat_ctx *bdev_iostat_ctx = _ctx;
6717 4 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6718 :
6719 4 : spdk_bdev_add_io_stat(bdev_iostat_ctx->stat, channel->stat);
6720 4 : spdk_bdev_reset_io_stat(channel->stat, bdev_iostat_ctx->reset_mode);
6721 4 : spdk_bdev_for_each_channel_continue(i, 0);
6722 4 : }
6723 :
6724 : void
6725 5 : spdk_bdev_get_device_stat(struct spdk_bdev *bdev, struct spdk_bdev_io_stat *stat,
6726 : enum spdk_bdev_reset_stat_mode reset_mode, spdk_bdev_get_device_stat_cb cb, void *cb_arg)
6727 : {
6728 : struct spdk_bdev_iostat_ctx *bdev_iostat_ctx;
6729 :
6730 5 : assert(bdev != NULL);
6731 5 : assert(stat != NULL);
6732 5 : assert(cb != NULL);
6733 :
6734 5 : bdev_iostat_ctx = calloc(1, sizeof(struct spdk_bdev_iostat_ctx));
6735 5 : if (bdev_iostat_ctx == NULL) {
6736 0 : SPDK_ERRLOG("Unable to allocate memory for spdk_bdev_iostat_ctx\n");
6737 0 : cb(bdev, stat, cb_arg, -ENOMEM);
6738 0 : return;
6739 : }
6740 :
6741 5 : bdev_iostat_ctx->stat = stat;
6742 5 : bdev_iostat_ctx->cb = cb;
6743 5 : bdev_iostat_ctx->cb_arg = cb_arg;
6744 5 : bdev_iostat_ctx->reset_mode = reset_mode;
6745 :
6746 : /* Start with the statistics from previously deleted channels. */
6747 5 : spdk_spin_lock(&bdev->internal.spinlock);
6748 5 : bdev_get_io_stat(bdev_iostat_ctx->stat, bdev->internal.stat);
6749 5 : spdk_bdev_reset_io_stat(bdev->internal.stat, reset_mode);
6750 5 : spdk_spin_unlock(&bdev->internal.spinlock);
6751 :
6752 : /* Then iterate and add the statistics from each existing channel. */
6753 5 : spdk_bdev_for_each_channel(bdev, bdev_get_each_channel_stat, bdev_iostat_ctx,
6754 : bdev_get_device_stat_done);
6755 : }
6756 :
6757 : struct bdev_iostat_reset_ctx {
6758 : enum spdk_bdev_reset_stat_mode mode;
6759 : bdev_reset_device_stat_cb cb;
6760 : void *cb_arg;
6761 : };
6762 :
6763 : static void
6764 0 : bdev_reset_device_stat_done(struct spdk_bdev *bdev, void *_ctx, int status)
6765 : {
6766 0 : struct bdev_iostat_reset_ctx *ctx = _ctx;
6767 :
6768 0 : ctx->cb(bdev, ctx->cb_arg, 0);
6769 :
6770 0 : free(ctx);
6771 0 : }
6772 :
6773 : static void
6774 0 : bdev_reset_each_channel_stat(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
6775 : struct spdk_io_channel *ch, void *_ctx)
6776 : {
6777 0 : struct bdev_iostat_reset_ctx *ctx = _ctx;
6778 0 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6779 :
6780 0 : spdk_bdev_reset_io_stat(channel->stat, ctx->mode);
6781 :
6782 0 : spdk_bdev_for_each_channel_continue(i, 0);
6783 0 : }
6784 :
6785 : void
6786 0 : bdev_reset_device_stat(struct spdk_bdev *bdev, enum spdk_bdev_reset_stat_mode mode,
6787 : bdev_reset_device_stat_cb cb, void *cb_arg)
6788 : {
6789 : struct bdev_iostat_reset_ctx *ctx;
6790 :
6791 0 : assert(bdev != NULL);
6792 0 : assert(cb != NULL);
6793 :
6794 0 : ctx = calloc(1, sizeof(*ctx));
6795 0 : if (ctx == NULL) {
6796 0 : SPDK_ERRLOG("Unable to allocate bdev_iostat_reset_ctx.\n");
6797 0 : cb(bdev, cb_arg, -ENOMEM);
6798 0 : return;
6799 : }
6800 :
6801 0 : ctx->mode = mode;
6802 0 : ctx->cb = cb;
6803 0 : ctx->cb_arg = cb_arg;
6804 :
6805 0 : spdk_spin_lock(&bdev->internal.spinlock);
6806 0 : spdk_bdev_reset_io_stat(bdev->internal.stat, mode);
6807 0 : spdk_spin_unlock(&bdev->internal.spinlock);
6808 :
6809 0 : spdk_bdev_for_each_channel(bdev,
6810 : bdev_reset_each_channel_stat,
6811 : ctx,
6812 : bdev_reset_device_stat_done);
6813 : }
6814 :
6815 : int
6816 1 : spdk_bdev_nvme_admin_passthru(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6817 : const struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes,
6818 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6819 : {
6820 1 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6821 : struct spdk_bdev_io *bdev_io;
6822 1 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6823 :
6824 1 : if (!desc->write) {
6825 0 : return -EBADF;
6826 : }
6827 :
6828 1 : if (spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_ADMIN))) {
6829 1 : return -ENOTSUP;
6830 : }
6831 :
6832 0 : bdev_io = bdev_channel_get_io(channel);
6833 0 : if (!bdev_io) {
6834 0 : return -ENOMEM;
6835 : }
6836 :
6837 0 : bdev_io->internal.ch = channel;
6838 0 : bdev_io->internal.desc = desc;
6839 0 : bdev_io->type = SPDK_BDEV_IO_TYPE_NVME_ADMIN;
6840 0 : bdev_io->u.nvme_passthru.cmd = *cmd;
6841 0 : bdev_io->u.nvme_passthru.buf = buf;
6842 0 : bdev_io->u.nvme_passthru.nbytes = nbytes;
6843 0 : bdev_io->u.nvme_passthru.md_buf = NULL;
6844 0 : bdev_io->u.nvme_passthru.md_len = 0;
6845 :
6846 0 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6847 :
6848 0 : bdev_io_submit(bdev_io);
6849 0 : return 0;
6850 : }
6851 :
6852 : int
6853 1 : spdk_bdev_nvme_io_passthru(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6854 : const struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes,
6855 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6856 : {
6857 1 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6858 : struct spdk_bdev_io *bdev_io;
6859 1 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6860 :
6861 1 : if (!desc->write) {
6862 : /*
6863 : * Do not try to parse the NVMe command - we could maybe use bits in the opcode
6864 : * to easily determine if the command is a read or write, but for now just
6865 : * do not allow io_passthru with a read-only descriptor.
6866 : */
6867 0 : return -EBADF;
6868 : }
6869 :
6870 1 : if (spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_IO))) {
6871 1 : return -ENOTSUP;
6872 : }
6873 :
6874 0 : bdev_io = bdev_channel_get_io(channel);
6875 0 : if (!bdev_io) {
6876 0 : return -ENOMEM;
6877 : }
6878 :
6879 0 : bdev_io->internal.ch = channel;
6880 0 : bdev_io->internal.desc = desc;
6881 0 : bdev_io->type = SPDK_BDEV_IO_TYPE_NVME_IO;
6882 0 : bdev_io->u.nvme_passthru.cmd = *cmd;
6883 0 : bdev_io->u.nvme_passthru.buf = buf;
6884 0 : bdev_io->u.nvme_passthru.nbytes = nbytes;
6885 0 : bdev_io->u.nvme_passthru.md_buf = NULL;
6886 0 : bdev_io->u.nvme_passthru.md_len = 0;
6887 :
6888 0 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6889 :
6890 0 : bdev_io_submit(bdev_io);
6891 0 : return 0;
6892 : }
6893 :
6894 : int
6895 1 : spdk_bdev_nvme_io_passthru_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6896 : const struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes, void *md_buf, size_t md_len,
6897 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6898 : {
6899 1 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6900 : struct spdk_bdev_io *bdev_io;
6901 1 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6902 :
6903 1 : if (!desc->write) {
6904 : /*
6905 : * Do not try to parse the NVMe command - we could maybe use bits in the opcode
6906 : * to easily determine if the command is a read or write, but for now just
6907 : * do not allow io_passthru with a read-only descriptor.
6908 : */
6909 0 : return -EBADF;
6910 : }
6911 :
6912 1 : if (spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_IO_MD))) {
6913 1 : return -ENOTSUP;
6914 : }
6915 :
6916 0 : bdev_io = bdev_channel_get_io(channel);
6917 0 : if (!bdev_io) {
6918 0 : return -ENOMEM;
6919 : }
6920 :
6921 0 : bdev_io->internal.ch = channel;
6922 0 : bdev_io->internal.desc = desc;
6923 0 : bdev_io->type = SPDK_BDEV_IO_TYPE_NVME_IO_MD;
6924 0 : bdev_io->u.nvme_passthru.cmd = *cmd;
6925 0 : bdev_io->u.nvme_passthru.buf = buf;
6926 0 : bdev_io->u.nvme_passthru.nbytes = nbytes;
6927 0 : bdev_io->u.nvme_passthru.md_buf = md_buf;
6928 0 : bdev_io->u.nvme_passthru.md_len = md_len;
6929 :
6930 0 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6931 :
6932 0 : bdev_io_submit(bdev_io);
6933 0 : return 0;
6934 : }
6935 :
6936 : int
6937 0 : spdk_bdev_nvme_iov_passthru_md(struct spdk_bdev_desc *desc,
6938 : struct spdk_io_channel *ch,
6939 : const struct spdk_nvme_cmd *cmd,
6940 : struct iovec *iov, int iovcnt, size_t nbytes,
6941 : void *md_buf, size_t md_len,
6942 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6943 : {
6944 0 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6945 : struct spdk_bdev_io *bdev_io;
6946 0 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6947 :
6948 0 : if (!desc->write) {
6949 : /*
6950 : * Do not try to parse the NVMe command - we could maybe use bits in the opcode
6951 : * to easily determine if the command is a read or write, but for now just
6952 : * do not allow io_passthru with a read-only descriptor.
6953 : */
6954 0 : return -EBADF;
6955 : }
6956 :
6957 0 : if (md_buf && spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_IO_MD))) {
6958 0 : return -ENOTSUP;
6959 0 : } else if (spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_IO))) {
6960 0 : return -ENOTSUP;
6961 : }
6962 :
6963 0 : bdev_io = bdev_channel_get_io(channel);
6964 0 : if (!bdev_io) {
6965 0 : return -ENOMEM;
6966 : }
6967 :
6968 0 : bdev_io->internal.ch = channel;
6969 0 : bdev_io->internal.desc = desc;
6970 0 : bdev_io->type = SPDK_BDEV_IO_TYPE_NVME_IOV_MD;
6971 0 : bdev_io->u.nvme_passthru.cmd = *cmd;
6972 0 : bdev_io->u.nvme_passthru.iovs = iov;
6973 0 : bdev_io->u.nvme_passthru.iovcnt = iovcnt;
6974 0 : bdev_io->u.nvme_passthru.nbytes = nbytes;
6975 0 : bdev_io->u.nvme_passthru.md_buf = md_buf;
6976 0 : bdev_io->u.nvme_passthru.md_len = md_len;
6977 :
6978 0 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6979 :
6980 0 : bdev_io_submit(bdev_io);
6981 0 : return 0;
6982 : }
6983 :
6984 : static void bdev_abort_retry(void *ctx);
6985 : static void bdev_abort(struct spdk_bdev_io *parent_io);
6986 :
6987 : static void
6988 22 : bdev_abort_io_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
6989 : {
6990 22 : struct spdk_bdev_channel *channel = bdev_io->internal.ch;
6991 22 : struct spdk_bdev_io *parent_io = cb_arg;
6992 : struct spdk_bdev_io *bio_to_abort, *tmp_io;
6993 :
6994 22 : bio_to_abort = bdev_io->u.abort.bio_to_abort;
6995 :
6996 22 : spdk_bdev_free_io(bdev_io);
6997 :
6998 22 : if (!success) {
6999 : /* Check if the target I/O completed in the meantime. */
7000 2 : TAILQ_FOREACH(tmp_io, &channel->io_submitted, internal.ch_link) {
7001 1 : if (tmp_io == bio_to_abort) {
7002 0 : break;
7003 : }
7004 : }
7005 :
7006 : /* If the target I/O still exists, set the parent to failed. */
7007 1 : if (tmp_io != NULL) {
7008 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
7009 : }
7010 : }
7011 :
7012 22 : assert(parent_io->internal.f.split);
7013 :
7014 22 : parent_io->internal.split.outstanding--;
7015 22 : if (parent_io->internal.split.outstanding == 0) {
7016 16 : if (parent_io->internal.status == SPDK_BDEV_IO_STATUS_NOMEM) {
7017 0 : bdev_abort_retry(parent_io);
7018 : } else {
7019 16 : bdev_io_complete(parent_io);
7020 : }
7021 : }
7022 22 : }
7023 :
7024 : static int
7025 23 : bdev_abort_io(struct spdk_bdev_desc *desc, struct spdk_bdev_channel *channel,
7026 : struct spdk_bdev_io *bio_to_abort,
7027 : spdk_bdev_io_completion_cb cb, void *cb_arg)
7028 : {
7029 23 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
7030 : struct spdk_bdev_io *bdev_io;
7031 :
7032 23 : if (bio_to_abort->type == SPDK_BDEV_IO_TYPE_ABORT ||
7033 23 : bio_to_abort->type == SPDK_BDEV_IO_TYPE_RESET) {
7034 : /* TODO: Abort reset or abort request. */
7035 0 : return -ENOTSUP;
7036 : }
7037 :
7038 23 : bdev_io = bdev_channel_get_io(channel);
7039 23 : if (bdev_io == NULL) {
7040 1 : return -ENOMEM;
7041 : }
7042 :
7043 22 : bdev_io->internal.ch = channel;
7044 22 : bdev_io->internal.desc = desc;
7045 22 : bdev_io->type = SPDK_BDEV_IO_TYPE_ABORT;
7046 22 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
7047 :
7048 22 : if (bio_to_abort->internal.f.split) {
7049 6 : assert(bdev_io_should_split(bio_to_abort));
7050 6 : bdev_io->u.bdev.abort.bio_cb_arg = bio_to_abort;
7051 :
7052 : /* Parent abort request is not submitted directly, but to manage its
7053 : * execution add it to the submitted list here.
7054 : */
7055 6 : bdev_io->internal.submit_tsc = spdk_get_ticks();
7056 6 : bdev_ch_add_to_io_submitted(bdev_io);
7057 :
7058 6 : bdev_abort(bdev_io);
7059 :
7060 6 : return 0;
7061 : }
7062 :
7063 16 : bdev_io->u.abort.bio_to_abort = bio_to_abort;
7064 :
7065 : /* Submit the abort request to the underlying bdev module. */
7066 16 : bdev_io_submit(bdev_io);
7067 :
7068 16 : return 0;
7069 : }
7070 :
7071 : static bool
7072 46 : bdev_io_on_tailq(struct spdk_bdev_io *bdev_io, bdev_io_tailq_t *tailq)
7073 : {
7074 : struct spdk_bdev_io *iter;
7075 :
7076 46 : TAILQ_FOREACH(iter, tailq, internal.link) {
7077 0 : if (iter == bdev_io) {
7078 0 : return true;
7079 : }
7080 : }
7081 :
7082 46 : return false;
7083 : }
7084 :
7085 : static uint32_t
7086 18 : _bdev_abort(struct spdk_bdev_io *parent_io)
7087 : {
7088 18 : struct spdk_bdev_desc *desc = parent_io->internal.desc;
7089 18 : struct spdk_bdev_channel *channel = parent_io->internal.ch;
7090 : void *bio_cb_arg;
7091 : struct spdk_bdev_io *bio_to_abort;
7092 : uint32_t matched_ios;
7093 : int rc;
7094 :
7095 18 : bio_cb_arg = parent_io->u.bdev.abort.bio_cb_arg;
7096 :
7097 : /* matched_ios is returned and will be kept by the caller.
7098 : *
7099 : * This function will be used for two cases, 1) the same cb_arg is used for
7100 : * multiple I/Os, 2) a single large I/O is split into smaller ones.
7101 : * Incrementing split_outstanding directly here may confuse readers especially
7102 : * for the 1st case.
7103 : *
7104 : * Completion of I/O abort is processed after stack unwinding. Hence this trick
7105 : * works as expected.
7106 : */
7107 18 : matched_ios = 0;
7108 18 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
7109 :
7110 105 : TAILQ_FOREACH(bio_to_abort, &channel->io_submitted, internal.ch_link) {
7111 88 : if (bio_to_abort->internal.caller_ctx != bio_cb_arg) {
7112 65 : continue;
7113 : }
7114 :
7115 23 : if (bio_to_abort->internal.submit_tsc > parent_io->internal.submit_tsc) {
7116 : /* Any I/O which was submitted after this abort command should be excluded. */
7117 0 : continue;
7118 : }
7119 :
7120 : /* We can't abort a request that's being pushed/pulled or executed by accel */
7121 46 : if (bdev_io_on_tailq(bio_to_abort, &channel->io_accel_exec) ||
7122 23 : bdev_io_on_tailq(bio_to_abort, &channel->io_memory_domain)) {
7123 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
7124 0 : break;
7125 : }
7126 :
7127 23 : rc = bdev_abort_io(desc, channel, bio_to_abort, bdev_abort_io_done, parent_io);
7128 23 : if (rc != 0) {
7129 1 : if (rc == -ENOMEM) {
7130 1 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_NOMEM;
7131 : } else {
7132 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
7133 : }
7134 1 : break;
7135 : }
7136 22 : matched_ios++;
7137 : }
7138 :
7139 18 : return matched_ios;
7140 : }
7141 :
7142 : static void
7143 1 : bdev_abort_retry(void *ctx)
7144 : {
7145 1 : struct spdk_bdev_io *parent_io = ctx;
7146 : uint32_t matched_ios;
7147 :
7148 1 : matched_ios = _bdev_abort(parent_io);
7149 :
7150 1 : if (matched_ios == 0) {
7151 0 : if (parent_io->internal.status == SPDK_BDEV_IO_STATUS_NOMEM) {
7152 0 : bdev_queue_io_wait_with_cb(parent_io, bdev_abort_retry);
7153 : } else {
7154 : /* For retry, the case that no target I/O was found is success
7155 : * because it means target I/Os completed in the meantime.
7156 : */
7157 0 : bdev_io_complete(parent_io);
7158 : }
7159 0 : return;
7160 : }
7161 :
7162 : /* Use split_outstanding to manage the progress of aborting I/Os. */
7163 1 : parent_io->internal.f.split = true;
7164 1 : parent_io->internal.split.outstanding = matched_ios;
7165 : }
7166 :
7167 : static void
7168 17 : bdev_abort(struct spdk_bdev_io *parent_io)
7169 : {
7170 : uint32_t matched_ios;
7171 :
7172 17 : matched_ios = _bdev_abort(parent_io);
7173 :
7174 17 : if (matched_ios == 0) {
7175 2 : if (parent_io->internal.status == SPDK_BDEV_IO_STATUS_NOMEM) {
7176 1 : bdev_queue_io_wait_with_cb(parent_io, bdev_abort_retry);
7177 : } else {
7178 : /* The case the no target I/O was found is failure. */
7179 1 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
7180 1 : bdev_io_complete(parent_io);
7181 : }
7182 2 : return;
7183 : }
7184 :
7185 : /* Use split_outstanding to manage the progress of aborting I/Os. */
7186 15 : parent_io->internal.f.split = true;
7187 15 : parent_io->internal.split.outstanding = matched_ios;
7188 : }
7189 :
7190 : int
7191 12 : spdk_bdev_abort(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
7192 : void *bio_cb_arg,
7193 : spdk_bdev_io_completion_cb cb, void *cb_arg)
7194 : {
7195 12 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
7196 12 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
7197 : struct spdk_bdev_io *bdev_io;
7198 :
7199 12 : if (bio_cb_arg == NULL) {
7200 0 : return -EINVAL;
7201 : }
7202 :
7203 12 : if (!spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_ABORT)) {
7204 1 : return -ENOTSUP;
7205 : }
7206 :
7207 11 : bdev_io = bdev_channel_get_io(channel);
7208 11 : if (bdev_io == NULL) {
7209 0 : return -ENOMEM;
7210 : }
7211 :
7212 11 : bdev_io->internal.ch = channel;
7213 11 : bdev_io->internal.desc = desc;
7214 11 : bdev_io->internal.submit_tsc = spdk_get_ticks();
7215 11 : bdev_io->type = SPDK_BDEV_IO_TYPE_ABORT;
7216 11 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
7217 :
7218 11 : bdev_io->u.bdev.abort.bio_cb_arg = bio_cb_arg;
7219 :
7220 : /* Parent abort request is not submitted directly, but to manage its execution,
7221 : * add it to the submitted list here.
7222 : */
7223 11 : bdev_ch_add_to_io_submitted(bdev_io);
7224 :
7225 11 : bdev_abort(bdev_io);
7226 :
7227 11 : return 0;
7228 : }
7229 :
7230 : int
7231 4 : spdk_bdev_queue_io_wait(struct spdk_bdev *bdev, struct spdk_io_channel *ch,
7232 : struct spdk_bdev_io_wait_entry *entry)
7233 : {
7234 4 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
7235 4 : struct spdk_bdev_mgmt_channel *mgmt_ch = channel->shared_resource->mgmt_ch;
7236 :
7237 4 : if (bdev != entry->bdev) {
7238 0 : SPDK_ERRLOG("bdevs do not match\n");
7239 0 : return -EINVAL;
7240 : }
7241 :
7242 4 : if (mgmt_ch->per_thread_cache_count > 0) {
7243 0 : SPDK_ERRLOG("Cannot queue io_wait if spdk_bdev_io available in per-thread cache\n");
7244 0 : return -EINVAL;
7245 : }
7246 :
7247 4 : TAILQ_INSERT_TAIL(&mgmt_ch->io_wait_queue, entry, link);
7248 4 : return 0;
7249 : }
7250 :
7251 : static inline void
7252 610 : bdev_io_update_io_stat(struct spdk_bdev_io *bdev_io, uint64_t tsc_diff)
7253 : {
7254 610 : enum spdk_bdev_io_status io_status = bdev_io->internal.status;
7255 610 : struct spdk_bdev_io_stat *io_stat = bdev_io->internal.ch->stat;
7256 610 : uint64_t num_blocks = bdev_io->u.bdev.num_blocks;
7257 610 : uint32_t blocklen = bdev_io->bdev->blocklen;
7258 :
7259 610 : if (spdk_likely(io_status == SPDK_BDEV_IO_STATUS_SUCCESS)) {
7260 516 : switch (bdev_io->type) {
7261 321 : case SPDK_BDEV_IO_TYPE_READ:
7262 321 : io_stat->bytes_read += num_blocks * blocklen;
7263 321 : io_stat->num_read_ops++;
7264 321 : io_stat->read_latency_ticks += tsc_diff;
7265 321 : if (io_stat->max_read_latency_ticks < tsc_diff) {
7266 7 : io_stat->max_read_latency_ticks = tsc_diff;
7267 : }
7268 321 : if (io_stat->min_read_latency_ticks > tsc_diff) {
7269 42 : io_stat->min_read_latency_ticks = tsc_diff;
7270 : }
7271 321 : break;
7272 75 : case SPDK_BDEV_IO_TYPE_WRITE:
7273 75 : io_stat->bytes_written += num_blocks * blocklen;
7274 75 : io_stat->num_write_ops++;
7275 75 : io_stat->write_latency_ticks += tsc_diff;
7276 75 : if (io_stat->max_write_latency_ticks < tsc_diff) {
7277 4 : io_stat->max_write_latency_ticks = tsc_diff;
7278 : }
7279 75 : if (io_stat->min_write_latency_ticks > tsc_diff) {
7280 25 : io_stat->min_write_latency_ticks = tsc_diff;
7281 : }
7282 75 : break;
7283 20 : case SPDK_BDEV_IO_TYPE_UNMAP:
7284 20 : io_stat->bytes_unmapped += num_blocks * blocklen;
7285 20 : io_stat->num_unmap_ops++;
7286 20 : io_stat->unmap_latency_ticks += tsc_diff;
7287 20 : if (io_stat->max_unmap_latency_ticks < tsc_diff) {
7288 0 : io_stat->max_unmap_latency_ticks = tsc_diff;
7289 : }
7290 20 : if (io_stat->min_unmap_latency_ticks > tsc_diff) {
7291 3 : io_stat->min_unmap_latency_ticks = tsc_diff;
7292 : }
7293 20 : break;
7294 4 : case SPDK_BDEV_IO_TYPE_ZCOPY:
7295 : /* Track the data in the start phase only */
7296 4 : if (bdev_io->u.bdev.zcopy.start) {
7297 2 : if (bdev_io->u.bdev.zcopy.populate) {
7298 1 : io_stat->bytes_read += num_blocks * blocklen;
7299 1 : io_stat->num_read_ops++;
7300 1 : io_stat->read_latency_ticks += tsc_diff;
7301 1 : if (io_stat->max_read_latency_ticks < tsc_diff) {
7302 0 : io_stat->max_read_latency_ticks = tsc_diff;
7303 : }
7304 1 : if (io_stat->min_read_latency_ticks > tsc_diff) {
7305 1 : io_stat->min_read_latency_ticks = tsc_diff;
7306 : }
7307 : } else {
7308 1 : io_stat->bytes_written += num_blocks * blocklen;
7309 1 : io_stat->num_write_ops++;
7310 1 : io_stat->write_latency_ticks += tsc_diff;
7311 1 : if (io_stat->max_write_latency_ticks < tsc_diff) {
7312 0 : io_stat->max_write_latency_ticks = tsc_diff;
7313 : }
7314 1 : if (io_stat->min_write_latency_ticks > tsc_diff) {
7315 1 : io_stat->min_write_latency_ticks = tsc_diff;
7316 : }
7317 : }
7318 : }
7319 4 : break;
7320 21 : case SPDK_BDEV_IO_TYPE_COPY:
7321 21 : io_stat->bytes_copied += num_blocks * blocklen;
7322 21 : io_stat->num_copy_ops++;
7323 21 : bdev_io->internal.ch->stat->copy_latency_ticks += tsc_diff;
7324 21 : if (io_stat->max_copy_latency_ticks < tsc_diff) {
7325 0 : io_stat->max_copy_latency_ticks = tsc_diff;
7326 : }
7327 21 : if (io_stat->min_copy_latency_ticks > tsc_diff) {
7328 4 : io_stat->min_copy_latency_ticks = tsc_diff;
7329 : }
7330 21 : break;
7331 75 : default:
7332 75 : break;
7333 : }
7334 94 : } else if (io_status <= SPDK_BDEV_IO_STATUS_FAILED && io_status >= SPDK_MIN_BDEV_IO_STATUS) {
7335 94 : io_stat = bdev_io->bdev->internal.stat;
7336 94 : assert(io_stat->io_error != NULL);
7337 :
7338 94 : spdk_spin_lock(&bdev_io->bdev->internal.spinlock);
7339 94 : io_stat->io_error->error_status[-io_status - 1]++;
7340 94 : spdk_spin_unlock(&bdev_io->bdev->internal.spinlock);
7341 : }
7342 :
7343 : #ifdef SPDK_CONFIG_VTUNE
7344 : uint64_t now_tsc = spdk_get_ticks();
7345 : if (now_tsc > (bdev_io->internal.ch->start_tsc + bdev_io->internal.ch->interval_tsc)) {
7346 : uint64_t data[5];
7347 : struct spdk_bdev_io_stat *prev_stat = bdev_io->internal.ch->prev_stat;
7348 :
7349 : data[0] = io_stat->num_read_ops - prev_stat->num_read_ops;
7350 : data[1] = io_stat->bytes_read - prev_stat->bytes_read;
7351 : data[2] = io_stat->num_write_ops - prev_stat->num_write_ops;
7352 : data[3] = io_stat->bytes_written - prev_stat->bytes_written;
7353 : data[4] = bdev_io->bdev->fn_table->get_spin_time ?
7354 : bdev_io->bdev->fn_table->get_spin_time(spdk_bdev_io_get_io_channel(bdev_io)) : 0;
7355 :
7356 : __itt_metadata_add(g_bdev_mgr.domain, __itt_null, bdev_io->internal.ch->handle,
7357 : __itt_metadata_u64, 5, data);
7358 :
7359 : memcpy(prev_stat, io_stat, sizeof(struct spdk_bdev_io_stat));
7360 : bdev_io->internal.ch->start_tsc = now_tsc;
7361 : }
7362 : #endif
7363 610 : }
7364 :
7365 : static inline void
7366 610 : _bdev_io_complete(void *ctx)
7367 : {
7368 610 : struct spdk_bdev_io *bdev_io = ctx;
7369 :
7370 610 : if (spdk_unlikely(bdev_io_use_accel_sequence(bdev_io))) {
7371 0 : assert(bdev_io->internal.status != SPDK_BDEV_IO_STATUS_SUCCESS);
7372 0 : spdk_accel_sequence_abort(bdev_io->internal.accel_sequence);
7373 : }
7374 :
7375 610 : assert(bdev_io->internal.cb != NULL);
7376 610 : assert(spdk_get_thread() == spdk_bdev_io_get_thread(bdev_io));
7377 :
7378 610 : bdev_io->internal.cb(bdev_io, bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS,
7379 : bdev_io->internal.caller_ctx);
7380 610 : }
7381 :
7382 : static inline void
7383 618 : bdev_io_complete(void *ctx)
7384 : {
7385 618 : struct spdk_bdev_io *bdev_io = ctx;
7386 618 : struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
7387 : uint64_t tsc, tsc_diff;
7388 :
7389 618 : if (spdk_unlikely(bdev_io->internal.f.in_submit_request)) {
7390 : /*
7391 : * Defer completion to avoid potential infinite recursion if the
7392 : * user's completion callback issues a new I/O.
7393 : */
7394 8 : spdk_thread_send_msg(spdk_bdev_io_get_thread(bdev_io),
7395 : bdev_io_complete, bdev_io);
7396 8 : return;
7397 : }
7398 :
7399 610 : tsc = spdk_get_ticks();
7400 610 : tsc_diff = tsc - bdev_io->internal.submit_tsc;
7401 :
7402 610 : bdev_ch_remove_from_io_submitted(bdev_io);
7403 610 : spdk_trace_record_tsc(tsc, TRACE_BDEV_IO_DONE, bdev_ch->trace_id, 0, (uintptr_t)bdev_io,
7404 : bdev_io->internal.caller_ctx, bdev_ch->queue_depth);
7405 :
7406 610 : if (bdev_ch->histogram) {
7407 4 : if (bdev_io->bdev->internal.histogram_io_type == 0 ||
7408 0 : bdev_io->bdev->internal.histogram_io_type == bdev_io->type) {
7409 : /*
7410 : * Tally all I/O types if the histogram_io_type is set to 0.
7411 : */
7412 4 : spdk_histogram_data_tally(bdev_ch->histogram, tsc_diff);
7413 : }
7414 : }
7415 :
7416 610 : bdev_io_update_io_stat(bdev_io, tsc_diff);
7417 610 : _bdev_io_complete(bdev_io);
7418 : }
7419 :
7420 : /* The difference between this function and bdev_io_complete() is that this should be called to
7421 : * complete IOs that haven't been submitted via bdev_io_submit(), as they weren't added onto the
7422 : * io_submitted list and don't have submit_tsc updated.
7423 : */
7424 : static inline void
7425 0 : bdev_io_complete_unsubmitted(struct spdk_bdev_io *bdev_io)
7426 : {
7427 : /* Since the IO hasn't been submitted it's bound to be failed */
7428 0 : assert(bdev_io->internal.status != SPDK_BDEV_IO_STATUS_SUCCESS);
7429 :
7430 : /* At this point we don't know if the IO is completed from submission context or not, but,
7431 : * since this is an error path, we can always do an spdk_thread_send_msg(). */
7432 0 : spdk_thread_send_msg(spdk_bdev_io_get_thread(bdev_io),
7433 : _bdev_io_complete, bdev_io);
7434 0 : }
7435 :
7436 : static void bdev_destroy_cb(void *io_device);
7437 :
7438 : static void
7439 15 : bdev_reset_complete(struct spdk_bdev *bdev, void *_ctx, int status)
7440 : {
7441 15 : struct spdk_bdev_io *bdev_io = _ctx;
7442 :
7443 15 : if (bdev_io->u.reset.ch_ref != NULL) {
7444 15 : spdk_put_io_channel(bdev_io->u.reset.ch_ref);
7445 15 : bdev_io->u.reset.ch_ref = NULL;
7446 : }
7447 :
7448 15 : bdev_io_complete(bdev_io);
7449 :
7450 15 : if (bdev->internal.status == SPDK_BDEV_STATUS_REMOVING &&
7451 1 : TAILQ_EMPTY(&bdev->internal.open_descs)) {
7452 1 : spdk_io_device_unregister(__bdev_to_io_dev(bdev), bdev_destroy_cb);
7453 : }
7454 15 : }
7455 :
7456 : static void
7457 18 : bdev_unfreeze_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
7458 : struct spdk_io_channel *_ch, void *_ctx)
7459 : {
7460 18 : struct spdk_bdev_io *bdev_io = _ctx;
7461 18 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
7462 : struct spdk_bdev_io *queued_reset;
7463 :
7464 18 : ch->flags &= ~BDEV_CH_RESET_IN_PROGRESS;
7465 18 : while (!TAILQ_EMPTY(&ch->queued_resets)) {
7466 0 : queued_reset = TAILQ_FIRST(&ch->queued_resets);
7467 0 : TAILQ_REMOVE(&ch->queued_resets, queued_reset, internal.link);
7468 0 : spdk_bdev_io_complete(queued_reset, bdev_io->internal.status);
7469 : }
7470 :
7471 18 : spdk_bdev_for_each_channel_continue(i, 0);
7472 18 : }
7473 :
7474 : static void
7475 0 : bdev_io_complete_sequence_cb(void *ctx, int status)
7476 : {
7477 0 : struct spdk_bdev_io *bdev_io = ctx;
7478 :
7479 : /* u.bdev.accel_sequence should have already been cleared at this point */
7480 0 : assert(bdev_io->u.bdev.accel_sequence == NULL);
7481 0 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS);
7482 0 : bdev_io->internal.f.has_accel_sequence = false;
7483 :
7484 0 : if (spdk_unlikely(status != 0)) {
7485 0 : SPDK_ERRLOG("Failed to execute accel sequence, status=%d\n", status);
7486 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
7487 : }
7488 :
7489 0 : bdev_io_complete(bdev_io);
7490 0 : }
7491 :
7492 : void
7493 598 : spdk_bdev_io_complete(struct spdk_bdev_io *bdev_io, enum spdk_bdev_io_status status)
7494 : {
7495 598 : struct spdk_bdev *bdev = bdev_io->bdev;
7496 598 : struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
7497 598 : struct spdk_bdev_shared_resource *shared_resource = bdev_ch->shared_resource;
7498 :
7499 598 : if (spdk_unlikely(bdev_io->internal.status != SPDK_BDEV_IO_STATUS_PENDING)) {
7500 0 : SPDK_ERRLOG("Unexpected completion on IO from %s module, status was %s\n",
7501 : spdk_bdev_get_module_name(bdev),
7502 : bdev_io_status_get_string(bdev_io->internal.status));
7503 0 : assert(false);
7504 : }
7505 598 : bdev_io->internal.status = status;
7506 :
7507 598 : if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_RESET)) {
7508 16 : bool unlock_channels = false;
7509 :
7510 16 : if (status == SPDK_BDEV_IO_STATUS_NOMEM) {
7511 0 : SPDK_ERRLOG("NOMEM returned for reset\n");
7512 : }
7513 16 : spdk_spin_lock(&bdev->internal.spinlock);
7514 16 : if (bdev_io == bdev->internal.reset_in_progress) {
7515 15 : bdev->internal.reset_in_progress = NULL;
7516 15 : unlock_channels = true;
7517 : }
7518 16 : spdk_spin_unlock(&bdev->internal.spinlock);
7519 :
7520 16 : if (unlock_channels) {
7521 15 : spdk_bdev_for_each_channel(bdev, bdev_unfreeze_channel, bdev_io,
7522 : bdev_reset_complete);
7523 15 : return;
7524 : }
7525 : } else {
7526 582 : bdev_io_decrement_outstanding(bdev_ch, shared_resource);
7527 582 : if (spdk_likely(status == SPDK_BDEV_IO_STATUS_SUCCESS)) {
7528 485 : if (bdev_io_needs_sequence_exec(bdev_io->internal.desc, bdev_io)) {
7529 0 : bdev_io_exec_sequence(bdev_io, bdev_io_complete_sequence_cb);
7530 0 : return;
7531 485 : } else if (spdk_unlikely(bdev_io->internal.f.has_bounce_buf &&
7532 : !bdev_io_use_accel_sequence(bdev_io))) {
7533 26 : _bdev_io_push_bounce_data_buffer(bdev_io,
7534 : _bdev_io_complete_push_bounce_done);
7535 : /* bdev IO will be completed in the callback */
7536 26 : return;
7537 : }
7538 : }
7539 :
7540 556 : if (spdk_unlikely(_bdev_io_handle_no_mem(bdev_io, BDEV_IO_RETRY_STATE_SUBMIT))) {
7541 5 : return;
7542 : }
7543 : }
7544 :
7545 552 : bdev_io_complete(bdev_io);
7546 : }
7547 :
7548 : void
7549 0 : spdk_bdev_io_complete_scsi_status(struct spdk_bdev_io *bdev_io, enum spdk_scsi_status sc,
7550 : enum spdk_scsi_sense sk, uint8_t asc, uint8_t ascq)
7551 : {
7552 : enum spdk_bdev_io_status status;
7553 :
7554 0 : if (sc == SPDK_SCSI_STATUS_GOOD) {
7555 0 : status = SPDK_BDEV_IO_STATUS_SUCCESS;
7556 : } else {
7557 0 : status = SPDK_BDEV_IO_STATUS_SCSI_ERROR;
7558 0 : bdev_io->internal.error.scsi.sc = sc;
7559 0 : bdev_io->internal.error.scsi.sk = sk;
7560 0 : bdev_io->internal.error.scsi.asc = asc;
7561 0 : bdev_io->internal.error.scsi.ascq = ascq;
7562 : }
7563 :
7564 0 : spdk_bdev_io_complete(bdev_io, status);
7565 0 : }
7566 :
7567 : void
7568 0 : spdk_bdev_io_get_scsi_status(const struct spdk_bdev_io *bdev_io,
7569 : int *sc, int *sk, int *asc, int *ascq)
7570 : {
7571 0 : assert(sc != NULL);
7572 0 : assert(sk != NULL);
7573 0 : assert(asc != NULL);
7574 0 : assert(ascq != NULL);
7575 :
7576 0 : switch (bdev_io->internal.status) {
7577 0 : case SPDK_BDEV_IO_STATUS_SUCCESS:
7578 0 : *sc = SPDK_SCSI_STATUS_GOOD;
7579 0 : *sk = SPDK_SCSI_SENSE_NO_SENSE;
7580 0 : *asc = SPDK_SCSI_ASC_NO_ADDITIONAL_SENSE;
7581 0 : *ascq = SPDK_SCSI_ASCQ_CAUSE_NOT_REPORTABLE;
7582 0 : break;
7583 0 : case SPDK_BDEV_IO_STATUS_NVME_ERROR:
7584 0 : spdk_scsi_nvme_translate(bdev_io, sc, sk, asc, ascq);
7585 0 : break;
7586 0 : case SPDK_BDEV_IO_STATUS_MISCOMPARE:
7587 0 : *sc = SPDK_SCSI_STATUS_CHECK_CONDITION;
7588 0 : *sk = SPDK_SCSI_SENSE_MISCOMPARE;
7589 0 : *asc = SPDK_SCSI_ASC_MISCOMPARE_DURING_VERIFY_OPERATION;
7590 0 : *ascq = bdev_io->internal.error.scsi.ascq;
7591 0 : break;
7592 0 : case SPDK_BDEV_IO_STATUS_SCSI_ERROR:
7593 0 : *sc = bdev_io->internal.error.scsi.sc;
7594 0 : *sk = bdev_io->internal.error.scsi.sk;
7595 0 : *asc = bdev_io->internal.error.scsi.asc;
7596 0 : *ascq = bdev_io->internal.error.scsi.ascq;
7597 0 : break;
7598 0 : default:
7599 0 : *sc = SPDK_SCSI_STATUS_CHECK_CONDITION;
7600 0 : *sk = SPDK_SCSI_SENSE_ABORTED_COMMAND;
7601 0 : *asc = SPDK_SCSI_ASC_NO_ADDITIONAL_SENSE;
7602 0 : *ascq = SPDK_SCSI_ASCQ_CAUSE_NOT_REPORTABLE;
7603 0 : break;
7604 : }
7605 0 : }
7606 :
7607 : void
7608 0 : spdk_bdev_io_complete_aio_status(struct spdk_bdev_io *bdev_io, int aio_result)
7609 : {
7610 : enum spdk_bdev_io_status status;
7611 :
7612 0 : if (aio_result == 0) {
7613 0 : status = SPDK_BDEV_IO_STATUS_SUCCESS;
7614 : } else {
7615 0 : status = SPDK_BDEV_IO_STATUS_AIO_ERROR;
7616 : }
7617 :
7618 0 : bdev_io->internal.error.aio_result = aio_result;
7619 :
7620 0 : spdk_bdev_io_complete(bdev_io, status);
7621 0 : }
7622 :
7623 : void
7624 0 : spdk_bdev_io_get_aio_status(const struct spdk_bdev_io *bdev_io, int *aio_result)
7625 : {
7626 0 : assert(aio_result != NULL);
7627 :
7628 0 : if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_AIO_ERROR) {
7629 0 : *aio_result = bdev_io->internal.error.aio_result;
7630 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS) {
7631 0 : *aio_result = 0;
7632 : } else {
7633 0 : *aio_result = -EIO;
7634 : }
7635 0 : }
7636 :
7637 : void
7638 0 : spdk_bdev_io_complete_nvme_status(struct spdk_bdev_io *bdev_io, uint32_t cdw0, int sct, int sc)
7639 : {
7640 : enum spdk_bdev_io_status status;
7641 :
7642 0 : if (spdk_likely(sct == SPDK_NVME_SCT_GENERIC && sc == SPDK_NVME_SC_SUCCESS)) {
7643 0 : status = SPDK_BDEV_IO_STATUS_SUCCESS;
7644 0 : } else if (sct == SPDK_NVME_SCT_GENERIC && sc == SPDK_NVME_SC_ABORTED_BY_REQUEST) {
7645 0 : status = SPDK_BDEV_IO_STATUS_ABORTED;
7646 : } else {
7647 0 : status = SPDK_BDEV_IO_STATUS_NVME_ERROR;
7648 : }
7649 :
7650 0 : bdev_io->internal.error.nvme.cdw0 = cdw0;
7651 0 : bdev_io->internal.error.nvme.sct = sct;
7652 0 : bdev_io->internal.error.nvme.sc = sc;
7653 :
7654 0 : spdk_bdev_io_complete(bdev_io, status);
7655 0 : }
7656 :
7657 : void
7658 0 : spdk_bdev_io_get_nvme_status(const struct spdk_bdev_io *bdev_io, uint32_t *cdw0, int *sct, int *sc)
7659 : {
7660 0 : assert(sct != NULL);
7661 0 : assert(sc != NULL);
7662 0 : assert(cdw0 != NULL);
7663 :
7664 0 : if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_ABORT)) {
7665 0 : *sct = SPDK_NVME_SCT_GENERIC;
7666 0 : *sc = SPDK_NVME_SC_SUCCESS;
7667 0 : if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS) {
7668 0 : *cdw0 = 0;
7669 : } else {
7670 0 : *cdw0 = 1U;
7671 : }
7672 0 : return;
7673 : }
7674 :
7675 0 : if (spdk_likely(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS)) {
7676 0 : *sct = SPDK_NVME_SCT_GENERIC;
7677 0 : *sc = SPDK_NVME_SC_SUCCESS;
7678 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_NVME_ERROR) {
7679 0 : *sct = bdev_io->internal.error.nvme.sct;
7680 0 : *sc = bdev_io->internal.error.nvme.sc;
7681 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_ABORTED) {
7682 0 : *sct = SPDK_NVME_SCT_GENERIC;
7683 0 : *sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
7684 : } else {
7685 0 : *sct = SPDK_NVME_SCT_GENERIC;
7686 0 : *sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
7687 : }
7688 :
7689 0 : *cdw0 = bdev_io->internal.error.nvme.cdw0;
7690 : }
7691 :
7692 : void
7693 0 : spdk_bdev_io_get_nvme_fused_status(const struct spdk_bdev_io *bdev_io, uint32_t *cdw0,
7694 : int *first_sct, int *first_sc, int *second_sct, int *second_sc)
7695 : {
7696 0 : assert(first_sct != NULL);
7697 0 : assert(first_sc != NULL);
7698 0 : assert(second_sct != NULL);
7699 0 : assert(second_sc != NULL);
7700 0 : assert(cdw0 != NULL);
7701 :
7702 0 : if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_NVME_ERROR) {
7703 0 : if (bdev_io->internal.error.nvme.sct == SPDK_NVME_SCT_MEDIA_ERROR &&
7704 0 : bdev_io->internal.error.nvme.sc == SPDK_NVME_SC_COMPARE_FAILURE) {
7705 0 : *first_sct = bdev_io->internal.error.nvme.sct;
7706 0 : *first_sc = bdev_io->internal.error.nvme.sc;
7707 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
7708 0 : *second_sc = SPDK_NVME_SC_ABORTED_FAILED_FUSED;
7709 : } else {
7710 0 : *first_sct = SPDK_NVME_SCT_GENERIC;
7711 0 : *first_sc = SPDK_NVME_SC_SUCCESS;
7712 0 : *second_sct = bdev_io->internal.error.nvme.sct;
7713 0 : *second_sc = bdev_io->internal.error.nvme.sc;
7714 : }
7715 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_ABORTED) {
7716 0 : *first_sct = SPDK_NVME_SCT_GENERIC;
7717 0 : *first_sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
7718 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
7719 0 : *second_sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
7720 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS) {
7721 0 : *first_sct = SPDK_NVME_SCT_GENERIC;
7722 0 : *first_sc = SPDK_NVME_SC_SUCCESS;
7723 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
7724 0 : *second_sc = SPDK_NVME_SC_SUCCESS;
7725 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_FIRST_FUSED_FAILED) {
7726 0 : *first_sct = SPDK_NVME_SCT_GENERIC;
7727 0 : *first_sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
7728 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
7729 0 : *second_sc = SPDK_NVME_SC_ABORTED_FAILED_FUSED;
7730 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_MISCOMPARE) {
7731 0 : *first_sct = SPDK_NVME_SCT_MEDIA_ERROR;
7732 0 : *first_sc = SPDK_NVME_SC_COMPARE_FAILURE;
7733 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
7734 0 : *second_sc = SPDK_NVME_SC_ABORTED_FAILED_FUSED;
7735 : } else {
7736 0 : *first_sct = SPDK_NVME_SCT_GENERIC;
7737 0 : *first_sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
7738 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
7739 0 : *second_sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
7740 : }
7741 :
7742 0 : *cdw0 = bdev_io->internal.error.nvme.cdw0;
7743 0 : }
7744 :
7745 : void
7746 0 : spdk_bdev_io_complete_base_io_status(struct spdk_bdev_io *bdev_io,
7747 : const struct spdk_bdev_io *base_io)
7748 : {
7749 0 : switch (base_io->internal.status) {
7750 0 : case SPDK_BDEV_IO_STATUS_NVME_ERROR:
7751 0 : spdk_bdev_io_complete_nvme_status(bdev_io,
7752 0 : base_io->internal.error.nvme.cdw0,
7753 0 : base_io->internal.error.nvme.sct,
7754 0 : base_io->internal.error.nvme.sc);
7755 0 : break;
7756 0 : case SPDK_BDEV_IO_STATUS_SCSI_ERROR:
7757 0 : spdk_bdev_io_complete_scsi_status(bdev_io,
7758 0 : base_io->internal.error.scsi.sc,
7759 0 : base_io->internal.error.scsi.sk,
7760 0 : base_io->internal.error.scsi.asc,
7761 0 : base_io->internal.error.scsi.ascq);
7762 0 : break;
7763 0 : case SPDK_BDEV_IO_STATUS_AIO_ERROR:
7764 0 : spdk_bdev_io_complete_aio_status(bdev_io, base_io->internal.error.aio_result);
7765 0 : break;
7766 0 : default:
7767 0 : spdk_bdev_io_complete(bdev_io, base_io->internal.status);
7768 0 : break;
7769 : }
7770 0 : }
7771 :
7772 : struct spdk_thread *
7773 660 : spdk_bdev_io_get_thread(struct spdk_bdev_io *bdev_io)
7774 : {
7775 660 : return spdk_io_channel_get_thread(bdev_io->internal.ch->channel);
7776 : }
7777 :
7778 : struct spdk_io_channel *
7779 70 : spdk_bdev_io_get_io_channel(struct spdk_bdev_io *bdev_io)
7780 : {
7781 70 : return bdev_io->internal.ch->channel;
7782 : }
7783 :
7784 : static int
7785 126 : bdev_register(struct spdk_bdev *bdev)
7786 : {
7787 : char *bdev_name;
7788 126 : char uuid[SPDK_UUID_STRING_LEN];
7789 126 : struct spdk_iobuf_opts iobuf_opts;
7790 : int ret;
7791 :
7792 126 : assert(bdev->module != NULL);
7793 :
7794 126 : if (!bdev->name) {
7795 0 : SPDK_ERRLOG("Bdev name is NULL\n");
7796 0 : return -EINVAL;
7797 : }
7798 :
7799 126 : if (!strlen(bdev->name)) {
7800 0 : SPDK_ERRLOG("Bdev name must not be an empty string\n");
7801 0 : return -EINVAL;
7802 : }
7803 :
7804 : /* Users often register their own I/O devices using the bdev name. In
7805 : * order to avoid conflicts, prepend bdev_. */
7806 126 : bdev_name = spdk_sprintf_alloc("bdev_%s", bdev->name);
7807 126 : if (!bdev_name) {
7808 0 : SPDK_ERRLOG("Unable to allocate memory for internal bdev name.\n");
7809 0 : return -ENOMEM;
7810 : }
7811 :
7812 126 : bdev->internal.stat = bdev_alloc_io_stat(true);
7813 126 : if (!bdev->internal.stat) {
7814 0 : SPDK_ERRLOG("Unable to allocate I/O statistics structure.\n");
7815 0 : free(bdev_name);
7816 0 : return -ENOMEM;
7817 : }
7818 :
7819 126 : bdev->internal.status = SPDK_BDEV_STATUS_READY;
7820 126 : bdev->internal.measured_queue_depth = UINT64_MAX;
7821 126 : bdev->internal.claim_type = SPDK_BDEV_CLAIM_NONE;
7822 126 : memset(&bdev->internal.claim, 0, sizeof(bdev->internal.claim));
7823 126 : bdev->internal.qd_poller = NULL;
7824 126 : bdev->internal.qos = NULL;
7825 :
7826 126 : TAILQ_INIT(&bdev->internal.open_descs);
7827 126 : TAILQ_INIT(&bdev->internal.locked_ranges);
7828 126 : TAILQ_INIT(&bdev->internal.pending_locked_ranges);
7829 126 : TAILQ_INIT(&bdev->aliases);
7830 :
7831 : /* UUID may be specified by the user or defined by bdev itself.
7832 : * Otherwise it will be generated here, so this field will never be empty. */
7833 126 : if (spdk_uuid_is_null(&bdev->uuid)) {
7834 42 : spdk_uuid_generate(&bdev->uuid);
7835 : }
7836 :
7837 : /* Add the UUID alias only if it's different than the name */
7838 126 : spdk_uuid_fmt_lower(uuid, sizeof(uuid), &bdev->uuid);
7839 126 : if (strcmp(bdev->name, uuid) != 0) {
7840 125 : ret = spdk_bdev_alias_add(bdev, uuid);
7841 125 : if (ret != 0) {
7842 2 : SPDK_ERRLOG("Unable to add uuid:%s alias for bdev %s\n", uuid, bdev->name);
7843 2 : bdev_free_io_stat(bdev->internal.stat);
7844 2 : free(bdev_name);
7845 2 : return ret;
7846 : }
7847 : }
7848 :
7849 124 : spdk_iobuf_get_opts(&iobuf_opts, sizeof(iobuf_opts));
7850 124 : if (spdk_bdev_get_buf_align(bdev) > 1) {
7851 0 : bdev->max_rw_size = spdk_min(bdev->max_rw_size ? bdev->max_rw_size : UINT32_MAX,
7852 : iobuf_opts.large_bufsize / bdev->blocklen);
7853 : }
7854 :
7855 : /* If the user didn't specify a write unit size, set it to one. */
7856 124 : if (bdev->write_unit_size == 0) {
7857 120 : bdev->write_unit_size = 1;
7858 : }
7859 :
7860 : /* Set ACWU value to the write unit size if bdev module did not set it (does not support it natively) */
7861 124 : if (bdev->acwu == 0) {
7862 120 : bdev->acwu = bdev->write_unit_size;
7863 : }
7864 :
7865 124 : if (bdev->phys_blocklen == 0) {
7866 120 : bdev->phys_blocklen = spdk_bdev_get_data_block_size(bdev);
7867 : }
7868 :
7869 124 : if (!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COPY)) {
7870 0 : bdev->max_copy = bdev_get_max_write(bdev, iobuf_opts.large_bufsize);
7871 : }
7872 :
7873 124 : if (!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE_ZEROES)) {
7874 0 : bdev->max_write_zeroes = bdev_get_max_write(bdev, ZERO_BUFFER_SIZE);
7875 : }
7876 :
7877 124 : bdev->internal.reset_in_progress = NULL;
7878 124 : bdev->internal.qd_poll_in_progress = false;
7879 124 : bdev->internal.period = 0;
7880 124 : bdev->internal.new_period = 0;
7881 124 : bdev->internal.trace_id = spdk_trace_register_owner(OWNER_TYPE_BDEV, bdev_name);
7882 :
7883 : /*
7884 : * Initialize spinlock before registering IO device because spinlock is used in
7885 : * bdev_channel_create
7886 : */
7887 124 : spdk_spin_init(&bdev->internal.spinlock);
7888 :
7889 124 : spdk_io_device_register(__bdev_to_io_dev(bdev),
7890 : bdev_channel_create, bdev_channel_destroy,
7891 : sizeof(struct spdk_bdev_channel),
7892 : bdev_name);
7893 :
7894 : /*
7895 : * Register bdev name only after the bdev object is ready.
7896 : * After bdev_name_add returns, it is possible for other threads to start using the bdev,
7897 : * create IO channels...
7898 : */
7899 124 : ret = bdev_name_add(&bdev->internal.bdev_name, bdev, bdev->name);
7900 124 : if (ret != 0) {
7901 0 : spdk_io_device_unregister(__bdev_to_io_dev(bdev), NULL);
7902 0 : bdev_free_io_stat(bdev->internal.stat);
7903 0 : spdk_spin_destroy(&bdev->internal.spinlock);
7904 0 : free(bdev_name);
7905 0 : return ret;
7906 : }
7907 :
7908 124 : free(bdev_name);
7909 :
7910 124 : SPDK_DEBUGLOG(bdev, "Inserting bdev %s into list\n", bdev->name);
7911 124 : TAILQ_INSERT_TAIL(&g_bdev_mgr.bdevs, bdev, internal.link);
7912 :
7913 124 : return 0;
7914 : }
7915 :
7916 : static void
7917 125 : bdev_destroy_cb(void *io_device)
7918 : {
7919 : int rc;
7920 : struct spdk_bdev *bdev;
7921 : spdk_bdev_unregister_cb cb_fn;
7922 : void *cb_arg;
7923 :
7924 125 : bdev = __bdev_from_io_dev(io_device);
7925 :
7926 125 : if (bdev->internal.unregister_td != spdk_get_thread()) {
7927 1 : spdk_thread_send_msg(bdev->internal.unregister_td, bdev_destroy_cb, io_device);
7928 1 : return;
7929 : }
7930 :
7931 124 : cb_fn = bdev->internal.unregister_cb;
7932 124 : cb_arg = bdev->internal.unregister_ctx;
7933 :
7934 124 : spdk_spin_destroy(&bdev->internal.spinlock);
7935 124 : free(bdev->internal.qos);
7936 124 : bdev_free_io_stat(bdev->internal.stat);
7937 124 : spdk_trace_unregister_owner(bdev->internal.trace_id);
7938 :
7939 124 : rc = bdev->fn_table->destruct(bdev->ctxt);
7940 124 : if (rc < 0) {
7941 0 : SPDK_ERRLOG("destruct failed\n");
7942 : }
7943 124 : if (rc <= 0 && cb_fn != NULL) {
7944 10 : cb_fn(cb_arg, rc);
7945 : }
7946 : }
7947 :
7948 : void
7949 2 : spdk_bdev_destruct_done(struct spdk_bdev *bdev, int bdeverrno)
7950 : {
7951 2 : if (bdev->internal.unregister_cb != NULL) {
7952 0 : bdev->internal.unregister_cb(bdev->internal.unregister_ctx, bdeverrno);
7953 : }
7954 2 : }
7955 :
7956 : static void
7957 19 : _remove_notify(void *arg)
7958 : {
7959 19 : struct spdk_bdev_desc *desc = arg;
7960 :
7961 19 : _event_notify(desc, SPDK_BDEV_EVENT_REMOVE);
7962 19 : }
7963 :
7964 : /* returns: 0 - bdev removed and ready to be destructed.
7965 : * -EBUSY - bdev can't be destructed yet. */
7966 : static int
7967 139 : bdev_unregister_unsafe(struct spdk_bdev *bdev)
7968 : {
7969 : struct spdk_bdev_desc *desc, *tmp;
7970 139 : int rc = 0;
7971 139 : char uuid[SPDK_UUID_STRING_LEN];
7972 :
7973 139 : assert(spdk_spin_held(&g_bdev_mgr.spinlock));
7974 139 : assert(spdk_spin_held(&bdev->internal.spinlock));
7975 :
7976 : /* Notify each descriptor about hotremoval */
7977 158 : TAILQ_FOREACH_SAFE(desc, &bdev->internal.open_descs, link, tmp) {
7978 19 : rc = -EBUSY;
7979 : /*
7980 : * Defer invocation of the event_cb to a separate message that will
7981 : * run later on its thread. This ensures this context unwinds and
7982 : * we don't recursively unregister this bdev again if the event_cb
7983 : * immediately closes its descriptor.
7984 : */
7985 19 : event_notify(desc, _remove_notify);
7986 : }
7987 :
7988 : /* If there are no descriptors, proceed removing the bdev */
7989 139 : if (rc == 0) {
7990 124 : TAILQ_REMOVE(&g_bdev_mgr.bdevs, bdev, internal.link);
7991 124 : SPDK_DEBUGLOG(bdev, "Removing bdev %s from list done\n", bdev->name);
7992 :
7993 : /* Delete the name and the UUID alias */
7994 124 : spdk_uuid_fmt_lower(uuid, sizeof(uuid), &bdev->uuid);
7995 124 : bdev_name_del_unsafe(&bdev->internal.bdev_name);
7996 124 : bdev_alias_del(bdev, uuid, bdev_name_del_unsafe);
7997 :
7998 124 : spdk_notify_send("bdev_unregister", spdk_bdev_get_name(bdev));
7999 :
8000 124 : if (bdev->internal.reset_in_progress != NULL) {
8001 : /* If reset is in progress, let the completion callback for reset
8002 : * unregister the bdev.
8003 : */
8004 1 : rc = -EBUSY;
8005 : }
8006 : }
8007 :
8008 139 : return rc;
8009 : }
8010 :
8011 : static void
8012 4 : bdev_unregister_abort_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
8013 : struct spdk_io_channel *io_ch, void *_ctx)
8014 : {
8015 4 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
8016 :
8017 4 : bdev_channel_abort_queued_ios(bdev_ch);
8018 4 : spdk_bdev_for_each_channel_continue(i, 0);
8019 4 : }
8020 :
8021 : static void
8022 124 : bdev_unregister(struct spdk_bdev *bdev, void *_ctx, int status)
8023 : {
8024 : int rc;
8025 :
8026 124 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8027 124 : spdk_spin_lock(&bdev->internal.spinlock);
8028 : /*
8029 : * Set the status to REMOVING after completing to abort channels. Otherwise,
8030 : * the last spdk_bdev_close() may call spdk_io_device_unregister() while
8031 : * spdk_bdev_for_each_channel() is executed and spdk_io_device_unregister()
8032 : * may fail.
8033 : */
8034 124 : bdev->internal.status = SPDK_BDEV_STATUS_REMOVING;
8035 124 : rc = bdev_unregister_unsafe(bdev);
8036 124 : spdk_spin_unlock(&bdev->internal.spinlock);
8037 124 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8038 :
8039 124 : if (rc == 0) {
8040 108 : spdk_io_device_unregister(__bdev_to_io_dev(bdev), bdev_destroy_cb);
8041 : }
8042 124 : }
8043 :
8044 : void
8045 131 : spdk_bdev_unregister(struct spdk_bdev *bdev, spdk_bdev_unregister_cb cb_fn, void *cb_arg)
8046 : {
8047 : struct spdk_thread *thread;
8048 :
8049 131 : SPDK_DEBUGLOG(bdev, "Removing bdev %s from list\n", bdev->name);
8050 :
8051 131 : thread = spdk_get_thread();
8052 131 : if (!thread) {
8053 : /* The user called this from a non-SPDK thread. */
8054 0 : if (cb_fn != NULL) {
8055 0 : cb_fn(cb_arg, -ENOTSUP);
8056 : }
8057 0 : return;
8058 : }
8059 :
8060 131 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8061 131 : if (bdev->internal.status == SPDK_BDEV_STATUS_UNREGISTERING ||
8062 131 : bdev->internal.status == SPDK_BDEV_STATUS_REMOVING) {
8063 7 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8064 7 : if (cb_fn) {
8065 0 : cb_fn(cb_arg, -EBUSY);
8066 : }
8067 7 : return;
8068 : }
8069 :
8070 124 : spdk_spin_lock(&bdev->internal.spinlock);
8071 124 : bdev->internal.status = SPDK_BDEV_STATUS_UNREGISTERING;
8072 124 : bdev->internal.unregister_cb = cb_fn;
8073 124 : bdev->internal.unregister_ctx = cb_arg;
8074 124 : bdev->internal.unregister_td = thread;
8075 124 : spdk_spin_unlock(&bdev->internal.spinlock);
8076 124 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8077 :
8078 124 : spdk_bdev_set_qd_sampling_period(bdev, 0);
8079 :
8080 124 : spdk_bdev_for_each_channel(bdev, bdev_unregister_abort_channel, bdev,
8081 : bdev_unregister);
8082 : }
8083 :
8084 : int
8085 4 : spdk_bdev_unregister_by_name(const char *bdev_name, struct spdk_bdev_module *module,
8086 : spdk_bdev_unregister_cb cb_fn, void *cb_arg)
8087 : {
8088 4 : struct spdk_bdev_desc *desc;
8089 : struct spdk_bdev *bdev;
8090 : int rc;
8091 :
8092 4 : rc = spdk_bdev_open_ext(bdev_name, false, _tmp_bdev_event_cb, NULL, &desc);
8093 4 : if (rc != 0) {
8094 1 : SPDK_ERRLOG("Failed to open bdev with name: %s\n", bdev_name);
8095 1 : return rc;
8096 : }
8097 :
8098 3 : bdev = spdk_bdev_desc_get_bdev(desc);
8099 :
8100 3 : if (bdev->module != module) {
8101 1 : spdk_bdev_close(desc);
8102 1 : SPDK_ERRLOG("Bdev %s was not registered by the specified module.\n",
8103 : bdev_name);
8104 1 : return -ENODEV;
8105 : }
8106 :
8107 2 : spdk_bdev_unregister(bdev, cb_fn, cb_arg);
8108 :
8109 2 : spdk_bdev_close(desc);
8110 :
8111 2 : return 0;
8112 : }
8113 :
8114 : static int
8115 255 : bdev_start_qos(struct spdk_bdev *bdev)
8116 : {
8117 : struct set_qos_limit_ctx *ctx;
8118 :
8119 : /* Enable QoS */
8120 255 : if (bdev->internal.qos && bdev->internal.qos->thread == NULL) {
8121 2 : ctx = calloc(1, sizeof(*ctx));
8122 2 : if (ctx == NULL) {
8123 0 : SPDK_ERRLOG("Failed to allocate memory for QoS context\n");
8124 0 : return -ENOMEM;
8125 : }
8126 2 : ctx->bdev = bdev;
8127 2 : spdk_bdev_for_each_channel(bdev, bdev_enable_qos_msg, ctx, bdev_enable_qos_done);
8128 : }
8129 :
8130 255 : return 0;
8131 : }
8132 :
8133 : static void
8134 24 : log_already_claimed(enum spdk_log_level level, const int line, const char *func, const char *detail,
8135 : struct spdk_bdev *bdev)
8136 : {
8137 : enum spdk_bdev_claim_type type;
8138 : const char *typename, *modname;
8139 : extern struct spdk_log_flag SPDK_LOG_bdev;
8140 :
8141 24 : assert(spdk_spin_held(&bdev->internal.spinlock));
8142 :
8143 24 : if (level >= SPDK_LOG_INFO && !SPDK_LOG_bdev.enabled) {
8144 0 : return;
8145 : }
8146 :
8147 24 : type = bdev->internal.claim_type;
8148 24 : typename = spdk_bdev_claim_get_name(type);
8149 :
8150 24 : if (type == SPDK_BDEV_CLAIM_EXCL_WRITE) {
8151 6 : modname = bdev->internal.claim.v1.module->name;
8152 6 : spdk_log(level, __FILE__, line, func, "bdev %s %s: type %s by module %s\n",
8153 : bdev->name, detail, typename, modname);
8154 6 : return;
8155 : }
8156 :
8157 18 : if (claim_type_is_v2(type)) {
8158 : struct spdk_bdev_module_claim *claim;
8159 :
8160 36 : TAILQ_FOREACH(claim, &bdev->internal.claim.v2.claims, link) {
8161 18 : modname = claim->module->name;
8162 18 : spdk_log(level, __FILE__, line, func, "bdev %s %s: type %s by module %s\n",
8163 : bdev->name, detail, typename, modname);
8164 : }
8165 18 : return;
8166 : }
8167 :
8168 0 : assert(false);
8169 : }
8170 :
8171 : static int
8172 264 : bdev_open(struct spdk_bdev *bdev, bool write, struct spdk_bdev_desc *desc)
8173 : {
8174 : struct spdk_thread *thread;
8175 264 : int rc = 0;
8176 :
8177 264 : thread = spdk_get_thread();
8178 264 : if (!thread) {
8179 0 : SPDK_ERRLOG("Cannot open bdev from non-SPDK thread.\n");
8180 0 : return -ENOTSUP;
8181 : }
8182 :
8183 264 : SPDK_DEBUGLOG(bdev, "Opening descriptor %p for bdev %s on thread %p\n", desc, bdev->name,
8184 : spdk_get_thread());
8185 :
8186 264 : desc->bdev = bdev;
8187 264 : desc->thread = thread;
8188 264 : desc->write = write;
8189 :
8190 264 : spdk_spin_lock(&bdev->internal.spinlock);
8191 264 : if (bdev->internal.status == SPDK_BDEV_STATUS_UNREGISTERING ||
8192 264 : bdev->internal.status == SPDK_BDEV_STATUS_REMOVING) {
8193 3 : spdk_spin_unlock(&bdev->internal.spinlock);
8194 3 : return -ENODEV;
8195 : }
8196 :
8197 261 : if (write && bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE) {
8198 6 : LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
8199 6 : spdk_spin_unlock(&bdev->internal.spinlock);
8200 6 : return -EPERM;
8201 : }
8202 :
8203 255 : rc = bdev_start_qos(bdev);
8204 255 : if (rc != 0) {
8205 0 : SPDK_ERRLOG("Failed to start QoS on bdev %s\n", bdev->name);
8206 0 : spdk_spin_unlock(&bdev->internal.spinlock);
8207 0 : return rc;
8208 : }
8209 :
8210 255 : TAILQ_INSERT_TAIL(&bdev->internal.open_descs, desc, link);
8211 :
8212 255 : spdk_spin_unlock(&bdev->internal.spinlock);
8213 :
8214 255 : return 0;
8215 : }
8216 :
8217 : static int
8218 264 : bdev_desc_alloc(struct spdk_bdev *bdev, spdk_bdev_event_cb_t event_cb, void *event_ctx,
8219 : struct spdk_bdev_desc **_desc)
8220 : {
8221 : struct spdk_bdev_desc *desc;
8222 : unsigned int i;
8223 :
8224 264 : desc = calloc(1, sizeof(*desc));
8225 264 : if (desc == NULL) {
8226 0 : SPDK_ERRLOG("Failed to allocate memory for bdev descriptor\n");
8227 0 : return -ENOMEM;
8228 : }
8229 :
8230 264 : TAILQ_INIT(&desc->pending_media_events);
8231 264 : TAILQ_INIT(&desc->free_media_events);
8232 :
8233 264 : desc->memory_domains_supported = spdk_bdev_get_memory_domains(bdev, NULL, 0) > 0;
8234 264 : desc->callback.event_fn = event_cb;
8235 264 : desc->callback.ctx = event_ctx;
8236 264 : spdk_spin_init(&desc->spinlock);
8237 :
8238 264 : if (bdev->media_events) {
8239 0 : desc->media_events_buffer = calloc(MEDIA_EVENT_POOL_SIZE,
8240 : sizeof(*desc->media_events_buffer));
8241 0 : if (desc->media_events_buffer == NULL) {
8242 0 : SPDK_ERRLOG("Failed to initialize media event pool\n");
8243 0 : bdev_desc_free(desc);
8244 0 : return -ENOMEM;
8245 : }
8246 :
8247 0 : for (i = 0; i < MEDIA_EVENT_POOL_SIZE; ++i) {
8248 0 : TAILQ_INSERT_TAIL(&desc->free_media_events,
8249 : &desc->media_events_buffer[i], tailq);
8250 : }
8251 : }
8252 :
8253 264 : if (bdev->fn_table->accel_sequence_supported != NULL) {
8254 0 : for (i = 0; i < SPDK_BDEV_NUM_IO_TYPES; ++i) {
8255 0 : desc->accel_sequence_supported[i] =
8256 0 : bdev->fn_table->accel_sequence_supported(bdev->ctxt,
8257 : (enum spdk_bdev_io_type)i);
8258 : }
8259 : }
8260 :
8261 264 : *_desc = desc;
8262 :
8263 264 : return 0;
8264 : }
8265 :
8266 : static int
8267 128 : bdev_open_ext(const char *bdev_name, bool write, spdk_bdev_event_cb_t event_cb,
8268 : void *event_ctx, struct spdk_bdev_desc **_desc)
8269 : {
8270 128 : struct spdk_bdev_desc *desc;
8271 : struct spdk_bdev *bdev;
8272 : int rc;
8273 :
8274 128 : bdev = bdev_get_by_name(bdev_name);
8275 :
8276 128 : if (bdev == NULL) {
8277 1 : SPDK_NOTICELOG("Currently unable to find bdev with name: %s\n", bdev_name);
8278 1 : return -ENODEV;
8279 : }
8280 :
8281 127 : rc = bdev_desc_alloc(bdev, event_cb, event_ctx, &desc);
8282 127 : if (rc != 0) {
8283 0 : return rc;
8284 : }
8285 :
8286 127 : rc = bdev_open(bdev, write, desc);
8287 127 : if (rc != 0) {
8288 7 : bdev_desc_free(desc);
8289 7 : desc = NULL;
8290 : }
8291 :
8292 127 : *_desc = desc;
8293 :
8294 127 : return rc;
8295 : }
8296 :
8297 : int
8298 130 : spdk_bdev_open_ext(const char *bdev_name, bool write, spdk_bdev_event_cb_t event_cb,
8299 : void *event_ctx, struct spdk_bdev_desc **_desc)
8300 : {
8301 : int rc;
8302 :
8303 130 : if (event_cb == NULL) {
8304 2 : SPDK_ERRLOG("Missing event callback function\n");
8305 2 : return -EINVAL;
8306 : }
8307 :
8308 128 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8309 128 : rc = bdev_open_ext(bdev_name, write, event_cb, event_ctx, _desc);
8310 128 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8311 :
8312 128 : return rc;
8313 : }
8314 :
8315 : struct spdk_bdev_open_async_ctx {
8316 : char *bdev_name;
8317 : spdk_bdev_event_cb_t event_cb;
8318 : void *event_ctx;
8319 : bool write;
8320 : int rc;
8321 : spdk_bdev_open_async_cb_t cb_fn;
8322 : void *cb_arg;
8323 : struct spdk_bdev_desc *desc;
8324 : struct spdk_bdev_open_async_opts opts;
8325 : uint64_t start_ticks;
8326 : struct spdk_thread *orig_thread;
8327 : struct spdk_poller *poller;
8328 : TAILQ_ENTRY(spdk_bdev_open_async_ctx) tailq;
8329 : };
8330 :
8331 : static void
8332 0 : bdev_open_async_done(void *arg)
8333 : {
8334 0 : struct spdk_bdev_open_async_ctx *ctx = arg;
8335 :
8336 0 : ctx->cb_fn(ctx->desc, ctx->rc, ctx->cb_arg);
8337 :
8338 0 : free(ctx->bdev_name);
8339 0 : free(ctx);
8340 0 : }
8341 :
8342 : static void
8343 0 : bdev_open_async_cancel(void *arg)
8344 : {
8345 0 : struct spdk_bdev_open_async_ctx *ctx = arg;
8346 :
8347 0 : assert(ctx->rc == -ESHUTDOWN);
8348 :
8349 0 : spdk_poller_unregister(&ctx->poller);
8350 :
8351 0 : bdev_open_async_done(ctx);
8352 0 : }
8353 :
8354 : /* This is called when the bdev library finishes at shutdown. */
8355 : static void
8356 65 : bdev_open_async_fini(void)
8357 : {
8358 : struct spdk_bdev_open_async_ctx *ctx, *tmp_ctx;
8359 :
8360 65 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8361 65 : TAILQ_FOREACH_SAFE(ctx, &g_bdev_mgr.async_bdev_opens, tailq, tmp_ctx) {
8362 0 : TAILQ_REMOVE(&g_bdev_mgr.async_bdev_opens, ctx, tailq);
8363 : /*
8364 : * We have to move to ctx->orig_thread to unregister ctx->poller.
8365 : * However, there is a chance that ctx->poller is executed before
8366 : * message is executed, which could result in bdev_open_async_done()
8367 : * being called twice. To avoid such race condition, set ctx->rc to
8368 : * -ESHUTDOWN.
8369 : */
8370 0 : ctx->rc = -ESHUTDOWN;
8371 0 : spdk_thread_send_msg(ctx->orig_thread, bdev_open_async_cancel, ctx);
8372 : }
8373 65 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8374 65 : }
8375 :
8376 : static int bdev_open_async(void *arg);
8377 :
8378 : static void
8379 0 : _bdev_open_async(struct spdk_bdev_open_async_ctx *ctx)
8380 : {
8381 : uint64_t timeout_ticks;
8382 :
8383 0 : if (ctx->rc == -ESHUTDOWN) {
8384 : /* This context is being canceled. Do nothing. */
8385 0 : return;
8386 : }
8387 :
8388 0 : ctx->rc = bdev_open_ext(ctx->bdev_name, ctx->write, ctx->event_cb, ctx->event_ctx,
8389 : &ctx->desc);
8390 0 : if (ctx->rc == 0 || ctx->opts.timeout_ms == 0) {
8391 0 : goto exit;
8392 : }
8393 :
8394 0 : timeout_ticks = ctx->start_ticks + ctx->opts.timeout_ms * spdk_get_ticks_hz() / 1000ull;
8395 0 : if (spdk_get_ticks() >= timeout_ticks) {
8396 0 : SPDK_ERRLOG("Timed out while waiting for bdev '%s' to appear\n", ctx->bdev_name);
8397 0 : ctx->rc = -ETIMEDOUT;
8398 0 : goto exit;
8399 : }
8400 :
8401 0 : return;
8402 :
8403 0 : exit:
8404 0 : spdk_poller_unregister(&ctx->poller);
8405 0 : TAILQ_REMOVE(&g_bdev_mgr.async_bdev_opens, ctx, tailq);
8406 :
8407 : /* Completion callback is processed after stack unwinding. */
8408 0 : spdk_thread_send_msg(ctx->orig_thread, bdev_open_async_done, ctx);
8409 : }
8410 :
8411 : static int
8412 0 : bdev_open_async(void *arg)
8413 : {
8414 0 : struct spdk_bdev_open_async_ctx *ctx = arg;
8415 :
8416 0 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8417 :
8418 0 : _bdev_open_async(ctx);
8419 :
8420 0 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8421 :
8422 0 : return SPDK_POLLER_BUSY;
8423 : }
8424 :
8425 : static void
8426 0 : bdev_open_async_opts_copy(struct spdk_bdev_open_async_opts *opts,
8427 : struct spdk_bdev_open_async_opts *opts_src,
8428 : size_t size)
8429 : {
8430 0 : assert(opts);
8431 0 : assert(opts_src);
8432 :
8433 0 : opts->size = size;
8434 :
8435 : #define SET_FIELD(field) \
8436 : if (offsetof(struct spdk_bdev_open_async_opts, field) + sizeof(opts->field) <= size) { \
8437 : opts->field = opts_src->field; \
8438 : } \
8439 :
8440 0 : SET_FIELD(timeout_ms);
8441 :
8442 : /* Do not remove this statement, you should always update this statement when you adding a new field,
8443 : * and do not forget to add the SET_FIELD statement for your added field. */
8444 : SPDK_STATIC_ASSERT(sizeof(struct spdk_bdev_open_async_opts) == 16, "Incorrect size");
8445 :
8446 : #undef SET_FIELD
8447 0 : }
8448 :
8449 : static void
8450 0 : bdev_open_async_opts_get_default(struct spdk_bdev_open_async_opts *opts, size_t size)
8451 : {
8452 0 : assert(opts);
8453 :
8454 0 : opts->size = size;
8455 :
8456 : #define SET_FIELD(field, value) \
8457 : if (offsetof(struct spdk_bdev_open_async_opts, field) + sizeof(opts->field) <= size) { \
8458 : opts->field = value; \
8459 : } \
8460 :
8461 0 : SET_FIELD(timeout_ms, 0);
8462 :
8463 : #undef SET_FIELD
8464 0 : }
8465 :
8466 : int
8467 0 : spdk_bdev_open_async(const char *bdev_name, bool write, spdk_bdev_event_cb_t event_cb,
8468 : void *event_ctx, struct spdk_bdev_open_async_opts *opts,
8469 : spdk_bdev_open_async_cb_t open_cb, void *open_cb_arg)
8470 : {
8471 : struct spdk_bdev_open_async_ctx *ctx;
8472 :
8473 0 : if (event_cb == NULL) {
8474 0 : SPDK_ERRLOG("Missing event callback function\n");
8475 0 : return -EINVAL;
8476 : }
8477 :
8478 0 : if (open_cb == NULL) {
8479 0 : SPDK_ERRLOG("Missing open callback function\n");
8480 0 : return -EINVAL;
8481 : }
8482 :
8483 0 : if (opts != NULL && opts->size == 0) {
8484 0 : SPDK_ERRLOG("size in the options structure should not be zero\n");
8485 0 : return -EINVAL;
8486 : }
8487 :
8488 0 : ctx = calloc(1, sizeof(*ctx));
8489 0 : if (ctx == NULL) {
8490 0 : SPDK_ERRLOG("Failed to allocate open context\n");
8491 0 : return -ENOMEM;
8492 : }
8493 :
8494 0 : ctx->bdev_name = strdup(bdev_name);
8495 0 : if (ctx->bdev_name == NULL) {
8496 0 : SPDK_ERRLOG("Failed to duplicate bdev_name\n");
8497 0 : free(ctx);
8498 0 : return -ENOMEM;
8499 : }
8500 :
8501 0 : ctx->poller = SPDK_POLLER_REGISTER(bdev_open_async, ctx, 100 * 1000);
8502 0 : if (ctx->poller == NULL) {
8503 0 : SPDK_ERRLOG("Failed to register bdev_open_async poller\n");
8504 0 : free(ctx->bdev_name);
8505 0 : free(ctx);
8506 0 : return -ENOMEM;
8507 : }
8508 :
8509 0 : ctx->cb_fn = open_cb;
8510 0 : ctx->cb_arg = open_cb_arg;
8511 0 : ctx->write = write;
8512 0 : ctx->event_cb = event_cb;
8513 0 : ctx->event_ctx = event_ctx;
8514 0 : ctx->orig_thread = spdk_get_thread();
8515 0 : ctx->start_ticks = spdk_get_ticks();
8516 :
8517 0 : bdev_open_async_opts_get_default(&ctx->opts, sizeof(ctx->opts));
8518 0 : if (opts != NULL) {
8519 0 : bdev_open_async_opts_copy(&ctx->opts, opts, opts->size);
8520 : }
8521 :
8522 0 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8523 :
8524 0 : TAILQ_INSERT_TAIL(&g_bdev_mgr.async_bdev_opens, ctx, tailq);
8525 0 : _bdev_open_async(ctx);
8526 :
8527 0 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8528 :
8529 0 : return 0;
8530 : }
8531 :
8532 : static void
8533 255 : bdev_close(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc)
8534 : {
8535 : int rc;
8536 :
8537 255 : spdk_spin_lock(&bdev->internal.spinlock);
8538 255 : spdk_spin_lock(&desc->spinlock);
8539 :
8540 255 : TAILQ_REMOVE(&bdev->internal.open_descs, desc, link);
8541 :
8542 255 : desc->closed = true;
8543 :
8544 255 : if (desc->claim != NULL) {
8545 16 : bdev_desc_release_claims(desc);
8546 : }
8547 :
8548 255 : if (0 == desc->refs) {
8549 244 : spdk_spin_unlock(&desc->spinlock);
8550 244 : bdev_desc_free(desc);
8551 : } else {
8552 11 : spdk_spin_unlock(&desc->spinlock);
8553 : }
8554 :
8555 : /* If no more descriptors, kill QoS channel */
8556 255 : if (bdev->internal.qos && TAILQ_EMPTY(&bdev->internal.open_descs)) {
8557 7 : SPDK_DEBUGLOG(bdev, "Closed last descriptor for bdev %s on thread %p. Stopping QoS.\n",
8558 : bdev->name, spdk_get_thread());
8559 :
8560 7 : if (bdev_qos_destroy(bdev)) {
8561 : /* There isn't anything we can do to recover here. Just let the
8562 : * old QoS poller keep running. The QoS handling won't change
8563 : * cores when the user allocates a new channel, but it won't break. */
8564 0 : SPDK_ERRLOG("Unable to shut down QoS poller. It will continue running on the current thread.\n");
8565 : }
8566 : }
8567 :
8568 255 : if (bdev->internal.status == SPDK_BDEV_STATUS_REMOVING && TAILQ_EMPTY(&bdev->internal.open_descs)) {
8569 15 : rc = bdev_unregister_unsafe(bdev);
8570 15 : spdk_spin_unlock(&bdev->internal.spinlock);
8571 :
8572 15 : if (rc == 0) {
8573 15 : spdk_io_device_unregister(__bdev_to_io_dev(bdev), bdev_destroy_cb);
8574 : }
8575 : } else {
8576 240 : spdk_spin_unlock(&bdev->internal.spinlock);
8577 : }
8578 255 : }
8579 :
8580 : void
8581 120 : spdk_bdev_close(struct spdk_bdev_desc *desc)
8582 : {
8583 120 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
8584 :
8585 120 : SPDK_DEBUGLOG(bdev, "Closing descriptor %p for bdev %s on thread %p\n", desc, bdev->name,
8586 : spdk_get_thread());
8587 :
8588 120 : assert(desc->thread == spdk_get_thread());
8589 :
8590 120 : spdk_poller_unregister(&desc->io_timeout_poller);
8591 :
8592 120 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8593 :
8594 120 : bdev_close(bdev, desc);
8595 :
8596 120 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8597 120 : }
8598 :
8599 : int32_t
8600 3 : spdk_bdev_get_numa_id(struct spdk_bdev *bdev)
8601 : {
8602 3 : if (bdev->numa.id_valid) {
8603 2 : return bdev->numa.id;
8604 : } else {
8605 1 : return SPDK_ENV_NUMA_ID_ANY;
8606 : }
8607 : }
8608 :
8609 : static void
8610 124 : bdev_register_finished(void *arg)
8611 : {
8612 124 : struct spdk_bdev_desc *desc = arg;
8613 124 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
8614 :
8615 124 : spdk_notify_send("bdev_register", spdk_bdev_get_name(bdev));
8616 :
8617 124 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8618 :
8619 124 : bdev_close(bdev, desc);
8620 :
8621 124 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8622 124 : }
8623 :
8624 : int
8625 127 : spdk_bdev_register(struct spdk_bdev *bdev)
8626 : {
8627 127 : struct spdk_bdev_desc *desc;
8628 127 : struct spdk_thread *thread = spdk_get_thread();
8629 : int rc;
8630 :
8631 127 : if (spdk_unlikely(!spdk_thread_is_app_thread(NULL))) {
8632 1 : SPDK_ERRLOG("Cannot register bdev %s on thread %p (%s)\n", bdev->name, thread,
8633 : thread ? spdk_thread_get_name(thread) : "null");
8634 1 : return -EINVAL;
8635 : }
8636 :
8637 126 : rc = bdev_register(bdev);
8638 126 : if (rc != 0) {
8639 2 : return rc;
8640 : }
8641 :
8642 : /* A descriptor is opened to prevent bdev deletion during examination */
8643 124 : rc = bdev_desc_alloc(bdev, _tmp_bdev_event_cb, NULL, &desc);
8644 124 : if (rc != 0) {
8645 0 : spdk_bdev_unregister(bdev, NULL, NULL);
8646 0 : return rc;
8647 : }
8648 :
8649 124 : rc = bdev_open(bdev, false, desc);
8650 124 : if (rc != 0) {
8651 0 : bdev_desc_free(desc);
8652 0 : spdk_bdev_unregister(bdev, NULL, NULL);
8653 0 : return rc;
8654 : }
8655 :
8656 : /* Examine configuration before initializing I/O */
8657 124 : bdev_examine(bdev);
8658 :
8659 124 : rc = spdk_bdev_wait_for_examine(bdev_register_finished, desc);
8660 124 : if (rc != 0) {
8661 0 : bdev_close(bdev, desc);
8662 0 : spdk_bdev_unregister(bdev, NULL, NULL);
8663 : }
8664 :
8665 124 : return rc;
8666 : }
8667 :
8668 : int
8669 26 : spdk_bdev_module_claim_bdev(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc,
8670 : struct spdk_bdev_module *module)
8671 : {
8672 26 : spdk_spin_lock(&bdev->internal.spinlock);
8673 :
8674 26 : if (bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE) {
8675 6 : LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
8676 6 : spdk_spin_unlock(&bdev->internal.spinlock);
8677 6 : return -EPERM;
8678 : }
8679 :
8680 20 : if (desc && !desc->write) {
8681 5 : desc->write = true;
8682 : }
8683 :
8684 20 : bdev->internal.claim_type = SPDK_BDEV_CLAIM_EXCL_WRITE;
8685 20 : bdev->internal.claim.v1.module = module;
8686 :
8687 20 : spdk_spin_unlock(&bdev->internal.spinlock);
8688 20 : return 0;
8689 : }
8690 :
8691 : void
8692 8 : spdk_bdev_module_release_bdev(struct spdk_bdev *bdev)
8693 : {
8694 8 : spdk_spin_lock(&bdev->internal.spinlock);
8695 :
8696 8 : assert(bdev->internal.claim.v1.module != NULL);
8697 8 : assert(bdev->internal.claim_type == SPDK_BDEV_CLAIM_EXCL_WRITE);
8698 8 : bdev->internal.claim_type = SPDK_BDEV_CLAIM_NONE;
8699 8 : bdev->internal.claim.v1.module = NULL;
8700 :
8701 8 : spdk_spin_unlock(&bdev->internal.spinlock);
8702 8 : }
8703 :
8704 : /*
8705 : * Start claims v2
8706 : */
8707 :
8708 : const char *
8709 24 : spdk_bdev_claim_get_name(enum spdk_bdev_claim_type type)
8710 : {
8711 24 : switch (type) {
8712 0 : case SPDK_BDEV_CLAIM_NONE:
8713 0 : return "not_claimed";
8714 6 : case SPDK_BDEV_CLAIM_EXCL_WRITE:
8715 6 : return "exclusive_write";
8716 7 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE:
8717 7 : return "read_many_write_one";
8718 5 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_NONE:
8719 5 : return "read_many_write_none";
8720 6 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
8721 6 : return "read_many_write_many";
8722 0 : default:
8723 0 : break;
8724 : }
8725 0 : return "invalid_claim";
8726 : }
8727 :
8728 : static bool
8729 96 : claim_type_is_v2(enum spdk_bdev_claim_type type)
8730 : {
8731 96 : switch (type) {
8732 96 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE:
8733 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_NONE:
8734 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
8735 96 : return true;
8736 0 : default:
8737 0 : break;
8738 : }
8739 0 : return false;
8740 : }
8741 :
8742 : /* Returns true if taking a claim with desc->write == false should make the descriptor writable. */
8743 : static bool
8744 13 : claim_type_promotes_to_write(enum spdk_bdev_claim_type type)
8745 : {
8746 13 : switch (type) {
8747 5 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE:
8748 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
8749 5 : return true;
8750 8 : default:
8751 8 : break;
8752 : }
8753 8 : return false;
8754 : }
8755 :
8756 : void
8757 44 : spdk_bdev_claim_opts_init(struct spdk_bdev_claim_opts *opts, size_t size)
8758 : {
8759 44 : if (opts == NULL) {
8760 0 : SPDK_ERRLOG("opts should not be NULL\n");
8761 0 : assert(opts != NULL);
8762 0 : return;
8763 : }
8764 44 : if (size == 0) {
8765 0 : SPDK_ERRLOG("size should not be zero\n");
8766 0 : assert(size != 0);
8767 0 : return;
8768 : }
8769 :
8770 44 : memset(opts, 0, size);
8771 44 : opts->opts_size = size;
8772 :
8773 : #define FIELD_OK(field) \
8774 : offsetof(struct spdk_bdev_claim_opts, field) + sizeof(opts->field) <= size
8775 :
8776 : #define SET_FIELD(field, value) \
8777 : if (FIELD_OK(field)) { \
8778 : opts->field = value; \
8779 : } \
8780 :
8781 44 : SET_FIELD(shared_claim_key, 0);
8782 :
8783 : #undef FIELD_OK
8784 : #undef SET_FIELD
8785 : }
8786 :
8787 : static int
8788 22 : claim_opts_copy(struct spdk_bdev_claim_opts *src, struct spdk_bdev_claim_opts *dst)
8789 : {
8790 22 : if (src->opts_size == 0) {
8791 0 : SPDK_ERRLOG("size should not be zero\n");
8792 0 : return -1;
8793 : }
8794 :
8795 22 : memset(dst, 0, sizeof(*dst));
8796 22 : dst->opts_size = src->opts_size;
8797 :
8798 : #define FIELD_OK(field) \
8799 : offsetof(struct spdk_bdev_claim_opts, field) + sizeof(src->field) <= src->opts_size
8800 :
8801 : #define SET_FIELD(field) \
8802 : if (FIELD_OK(field)) { \
8803 : dst->field = src->field; \
8804 : } \
8805 :
8806 22 : if (FIELD_OK(name)) {
8807 22 : snprintf(dst->name, sizeof(dst->name), "%s", src->name);
8808 : }
8809 :
8810 22 : SET_FIELD(shared_claim_key);
8811 :
8812 : /* You should not remove this statement, but need to update the assert statement
8813 : * if you add a new field, and also add a corresponding SET_FIELD statement */
8814 : SPDK_STATIC_ASSERT(sizeof(struct spdk_bdev_claim_opts) == 48, "Incorrect size");
8815 :
8816 : #undef FIELD_OK
8817 : #undef SET_FIELD
8818 22 : return 0;
8819 : }
8820 :
8821 : /* Returns 0 if a read-write-once claim can be taken. */
8822 : static int
8823 9 : claim_verify_rwo(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
8824 : struct spdk_bdev_claim_opts *opts, struct spdk_bdev_module *module)
8825 : {
8826 9 : struct spdk_bdev *bdev = desc->bdev;
8827 : struct spdk_bdev_desc *open_desc;
8828 :
8829 9 : assert(spdk_spin_held(&bdev->internal.spinlock));
8830 9 : assert(type == SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE);
8831 :
8832 9 : if (opts->shared_claim_key != 0) {
8833 1 : SPDK_ERRLOG("%s: key option not supported with read-write-once claims\n",
8834 : bdev->name);
8835 1 : return -EINVAL;
8836 : }
8837 8 : if (bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE) {
8838 1 : LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
8839 1 : return -EPERM;
8840 : }
8841 7 : if (desc->claim != NULL) {
8842 0 : SPDK_NOTICELOG("%s: descriptor already claimed bdev with module %s\n",
8843 : bdev->name, desc->claim->module->name);
8844 0 : return -EPERM;
8845 : }
8846 14 : TAILQ_FOREACH(open_desc, &bdev->internal.open_descs, link) {
8847 9 : if (desc != open_desc && open_desc->write) {
8848 2 : SPDK_NOTICELOG("%s: Cannot obtain read-write-once claim while "
8849 : "another descriptor is open for writing\n",
8850 : bdev->name);
8851 2 : return -EPERM;
8852 : }
8853 : }
8854 :
8855 5 : return 0;
8856 : }
8857 :
8858 : /* Returns 0 if a read-only-many claim can be taken. */
8859 : static int
8860 12 : claim_verify_rom(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
8861 : struct spdk_bdev_claim_opts *opts, struct spdk_bdev_module *module)
8862 : {
8863 12 : struct spdk_bdev *bdev = desc->bdev;
8864 : struct spdk_bdev_desc *open_desc;
8865 :
8866 12 : assert(spdk_spin_held(&bdev->internal.spinlock));
8867 12 : assert(type == SPDK_BDEV_CLAIM_READ_MANY_WRITE_NONE);
8868 12 : assert(desc->claim == NULL);
8869 :
8870 12 : if (desc->write) {
8871 3 : SPDK_ERRLOG("%s: Cannot obtain read-only-many claim with writable descriptor\n",
8872 : bdev->name);
8873 3 : return -EINVAL;
8874 : }
8875 9 : if (opts->shared_claim_key != 0) {
8876 1 : SPDK_ERRLOG("%s: key option not supported with read-only-may claims\n", bdev->name);
8877 1 : return -EINVAL;
8878 : }
8879 8 : if (bdev->internal.claim_type == SPDK_BDEV_CLAIM_NONE) {
8880 15 : TAILQ_FOREACH(open_desc, &bdev->internal.open_descs, link) {
8881 9 : if (open_desc->write) {
8882 0 : SPDK_NOTICELOG("%s: Cannot obtain read-only-many claim while "
8883 : "another descriptor is open for writing\n",
8884 : bdev->name);
8885 0 : return -EPERM;
8886 : }
8887 : }
8888 : }
8889 :
8890 8 : return 0;
8891 : }
8892 :
8893 : /* Returns 0 if a read-write-many claim can be taken. */
8894 : static int
8895 8 : claim_verify_rwm(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
8896 : struct spdk_bdev_claim_opts *opts, struct spdk_bdev_module *module)
8897 : {
8898 8 : struct spdk_bdev *bdev = desc->bdev;
8899 : struct spdk_bdev_desc *open_desc;
8900 :
8901 8 : assert(spdk_spin_held(&bdev->internal.spinlock));
8902 8 : assert(type == SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED);
8903 8 : assert(desc->claim == NULL);
8904 :
8905 8 : if (opts->shared_claim_key == 0) {
8906 2 : SPDK_ERRLOG("%s: shared_claim_key option required with read-write-may claims\n",
8907 : bdev->name);
8908 2 : return -EINVAL;
8909 : }
8910 6 : switch (bdev->internal.claim_type) {
8911 4 : case SPDK_BDEV_CLAIM_NONE:
8912 7 : TAILQ_FOREACH(open_desc, &bdev->internal.open_descs, link) {
8913 5 : if (open_desc == desc) {
8914 3 : continue;
8915 : }
8916 2 : if (open_desc->write) {
8917 2 : SPDK_NOTICELOG("%s: Cannot obtain read-write-many claim while "
8918 : "another descriptor is open for writing without a "
8919 : "claim\n", bdev->name);
8920 2 : return -EPERM;
8921 : }
8922 : }
8923 2 : break;
8924 2 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
8925 2 : if (opts->shared_claim_key != bdev->internal.claim.v2.key) {
8926 1 : LOG_ALREADY_CLAIMED_ERROR("already claimed with another key", bdev);
8927 1 : return -EPERM;
8928 : }
8929 1 : break;
8930 0 : default:
8931 0 : LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
8932 0 : return -EBUSY;
8933 : }
8934 :
8935 3 : return 0;
8936 : }
8937 :
8938 : /* Updates desc and its bdev with a v2 claim. */
8939 : static int
8940 16 : claim_bdev(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
8941 : struct spdk_bdev_claim_opts *opts, struct spdk_bdev_module *module)
8942 : {
8943 16 : struct spdk_bdev *bdev = desc->bdev;
8944 : struct spdk_bdev_module_claim *claim;
8945 :
8946 16 : assert(spdk_spin_held(&bdev->internal.spinlock));
8947 16 : assert(claim_type_is_v2(type));
8948 16 : assert(desc->claim == NULL);
8949 :
8950 16 : claim = calloc(1, sizeof(*desc->claim));
8951 16 : if (claim == NULL) {
8952 0 : SPDK_ERRLOG("%s: out of memory while allocating claim\n", bdev->name);
8953 0 : return -ENOMEM;
8954 : }
8955 16 : claim->module = module;
8956 16 : claim->desc = desc;
8957 : SPDK_STATIC_ASSERT(sizeof(claim->name) == sizeof(opts->name), "sizes must match");
8958 16 : memcpy(claim->name, opts->name, sizeof(claim->name));
8959 16 : desc->claim = claim;
8960 :
8961 16 : if (bdev->internal.claim_type == SPDK_BDEV_CLAIM_NONE) {
8962 13 : bdev->internal.claim_type = type;
8963 13 : TAILQ_INIT(&bdev->internal.claim.v2.claims);
8964 13 : bdev->internal.claim.v2.key = opts->shared_claim_key;
8965 : }
8966 16 : assert(type == bdev->internal.claim_type);
8967 :
8968 16 : TAILQ_INSERT_TAIL(&bdev->internal.claim.v2.claims, claim, link);
8969 :
8970 16 : if (!desc->write && claim_type_promotes_to_write(type)) {
8971 5 : desc->write = true;
8972 : }
8973 :
8974 16 : return 0;
8975 : }
8976 :
8977 : int
8978 39 : spdk_bdev_module_claim_bdev_desc(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
8979 : struct spdk_bdev_claim_opts *_opts,
8980 : struct spdk_bdev_module *module)
8981 : {
8982 : struct spdk_bdev *bdev;
8983 39 : struct spdk_bdev_claim_opts opts;
8984 39 : int rc = 0;
8985 :
8986 39 : if (desc == NULL) {
8987 0 : SPDK_ERRLOG("descriptor must not be NULL\n");
8988 0 : return -EINVAL;
8989 : }
8990 :
8991 39 : bdev = desc->bdev;
8992 :
8993 39 : if (_opts == NULL) {
8994 17 : spdk_bdev_claim_opts_init(&opts, sizeof(opts));
8995 22 : } else if (claim_opts_copy(_opts, &opts) != 0) {
8996 0 : return -EINVAL;
8997 : }
8998 :
8999 39 : spdk_spin_lock(&bdev->internal.spinlock);
9000 :
9001 39 : if (bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE &&
9002 15 : bdev->internal.claim_type != type) {
9003 10 : LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
9004 10 : spdk_spin_unlock(&bdev->internal.spinlock);
9005 10 : return -EPERM;
9006 : }
9007 :
9008 29 : if (claim_type_is_v2(type) && desc->claim != NULL) {
9009 0 : SPDK_ERRLOG("%s: descriptor already has %s claim with name '%s'\n",
9010 : bdev->name, spdk_bdev_claim_get_name(type), desc->claim->name);
9011 0 : spdk_spin_unlock(&bdev->internal.spinlock);
9012 0 : return -EPERM;
9013 : }
9014 :
9015 29 : switch (type) {
9016 0 : case SPDK_BDEV_CLAIM_EXCL_WRITE:
9017 0 : spdk_spin_unlock(&bdev->internal.spinlock);
9018 0 : return spdk_bdev_module_claim_bdev(bdev, desc, module);
9019 9 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE:
9020 9 : rc = claim_verify_rwo(desc, type, &opts, module);
9021 9 : break;
9022 12 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_NONE:
9023 12 : rc = claim_verify_rom(desc, type, &opts, module);
9024 12 : break;
9025 8 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
9026 8 : rc = claim_verify_rwm(desc, type, &opts, module);
9027 8 : break;
9028 0 : default:
9029 0 : SPDK_ERRLOG("%s: claim type %d not supported\n", bdev->name, type);
9030 0 : rc = -ENOTSUP;
9031 : }
9032 :
9033 29 : if (rc == 0) {
9034 16 : rc = claim_bdev(desc, type, &opts, module);
9035 : }
9036 :
9037 29 : spdk_spin_unlock(&bdev->internal.spinlock);
9038 29 : return rc;
9039 : }
9040 :
9041 : static void
9042 13 : claim_reset(struct spdk_bdev *bdev)
9043 : {
9044 13 : assert(spdk_spin_held(&bdev->internal.spinlock));
9045 13 : assert(claim_type_is_v2(bdev->internal.claim_type));
9046 13 : assert(TAILQ_EMPTY(&bdev->internal.claim.v2.claims));
9047 :
9048 13 : memset(&bdev->internal.claim, 0, sizeof(bdev->internal.claim));
9049 13 : bdev->internal.claim_type = SPDK_BDEV_CLAIM_NONE;
9050 13 : }
9051 :
9052 : static void
9053 16 : bdev_desc_release_claims(struct spdk_bdev_desc *desc)
9054 : {
9055 16 : struct spdk_bdev *bdev = desc->bdev;
9056 :
9057 16 : assert(spdk_spin_held(&bdev->internal.spinlock));
9058 16 : assert(claim_type_is_v2(bdev->internal.claim_type));
9059 :
9060 16 : if (bdev->internal.examine_in_progress == 0) {
9061 16 : TAILQ_REMOVE(&bdev->internal.claim.v2.claims, desc->claim, link);
9062 16 : free(desc->claim);
9063 16 : if (TAILQ_EMPTY(&bdev->internal.claim.v2.claims)) {
9064 13 : claim_reset(bdev);
9065 : }
9066 : } else {
9067 : /* This is a dead claim that will be cleaned up when bdev_examine() is done. */
9068 0 : desc->claim->module = NULL;
9069 0 : desc->claim->desc = NULL;
9070 : }
9071 16 : desc->claim = NULL;
9072 16 : }
9073 :
9074 : /*
9075 : * End claims v2
9076 : */
9077 :
9078 : struct spdk_bdev *
9079 1178 : spdk_bdev_desc_get_bdev(struct spdk_bdev_desc *desc)
9080 : {
9081 1178 : assert(desc != NULL);
9082 1178 : return desc->bdev;
9083 : }
9084 :
9085 : int
9086 1 : spdk_for_each_bdev(void *ctx, spdk_for_each_bdev_fn fn)
9087 : {
9088 : struct spdk_bdev *bdev, *tmp;
9089 1 : struct spdk_bdev_desc *desc;
9090 1 : int rc = 0;
9091 :
9092 1 : assert(fn != NULL);
9093 :
9094 1 : spdk_spin_lock(&g_bdev_mgr.spinlock);
9095 1 : bdev = spdk_bdev_first();
9096 9 : while (bdev != NULL) {
9097 8 : rc = bdev_desc_alloc(bdev, _tmp_bdev_event_cb, NULL, &desc);
9098 8 : if (rc != 0) {
9099 0 : break;
9100 : }
9101 8 : rc = bdev_open(bdev, false, desc);
9102 8 : if (rc != 0) {
9103 1 : bdev_desc_free(desc);
9104 1 : if (rc == -ENODEV) {
9105 : /* Ignore the error and move to the next bdev. */
9106 1 : rc = 0;
9107 1 : bdev = spdk_bdev_next(bdev);
9108 1 : continue;
9109 : }
9110 0 : break;
9111 : }
9112 7 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
9113 :
9114 7 : rc = fn(ctx, bdev);
9115 :
9116 7 : spdk_spin_lock(&g_bdev_mgr.spinlock);
9117 7 : tmp = spdk_bdev_next(bdev);
9118 7 : bdev_close(bdev, desc);
9119 7 : if (rc != 0) {
9120 0 : break;
9121 : }
9122 7 : bdev = tmp;
9123 : }
9124 1 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
9125 :
9126 1 : return rc;
9127 : }
9128 :
9129 : int
9130 1 : spdk_for_each_bdev_leaf(void *ctx, spdk_for_each_bdev_fn fn)
9131 : {
9132 : struct spdk_bdev *bdev, *tmp;
9133 1 : struct spdk_bdev_desc *desc;
9134 1 : int rc = 0;
9135 :
9136 1 : assert(fn != NULL);
9137 :
9138 1 : spdk_spin_lock(&g_bdev_mgr.spinlock);
9139 1 : bdev = spdk_bdev_first_leaf();
9140 6 : while (bdev != NULL) {
9141 5 : rc = bdev_desc_alloc(bdev, _tmp_bdev_event_cb, NULL, &desc);
9142 5 : if (rc != 0) {
9143 0 : break;
9144 : }
9145 5 : rc = bdev_open(bdev, false, desc);
9146 5 : if (rc != 0) {
9147 1 : bdev_desc_free(desc);
9148 1 : if (rc == -ENODEV) {
9149 : /* Ignore the error and move to the next bdev. */
9150 1 : rc = 0;
9151 1 : bdev = spdk_bdev_next_leaf(bdev);
9152 1 : continue;
9153 : }
9154 0 : break;
9155 : }
9156 4 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
9157 :
9158 4 : rc = fn(ctx, bdev);
9159 :
9160 4 : spdk_spin_lock(&g_bdev_mgr.spinlock);
9161 4 : tmp = spdk_bdev_next_leaf(bdev);
9162 4 : bdev_close(bdev, desc);
9163 4 : if (rc != 0) {
9164 0 : break;
9165 : }
9166 4 : bdev = tmp;
9167 : }
9168 1 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
9169 :
9170 1 : return rc;
9171 : }
9172 :
9173 : void
9174 0 : spdk_bdev_io_get_iovec(struct spdk_bdev_io *bdev_io, struct iovec **iovp, int *iovcntp)
9175 : {
9176 : struct iovec *iovs;
9177 : int iovcnt;
9178 :
9179 0 : if (bdev_io == NULL) {
9180 0 : return;
9181 : }
9182 :
9183 0 : switch (bdev_io->type) {
9184 0 : case SPDK_BDEV_IO_TYPE_READ:
9185 : case SPDK_BDEV_IO_TYPE_WRITE:
9186 : case SPDK_BDEV_IO_TYPE_ZCOPY:
9187 0 : iovs = bdev_io->u.bdev.iovs;
9188 0 : iovcnt = bdev_io->u.bdev.iovcnt;
9189 0 : break;
9190 0 : default:
9191 0 : iovs = NULL;
9192 0 : iovcnt = 0;
9193 0 : break;
9194 : }
9195 :
9196 0 : if (iovp) {
9197 0 : *iovp = iovs;
9198 : }
9199 0 : if (iovcntp) {
9200 0 : *iovcntp = iovcnt;
9201 : }
9202 : }
9203 :
9204 : void *
9205 0 : spdk_bdev_io_get_md_buf(struct spdk_bdev_io *bdev_io)
9206 : {
9207 0 : if (bdev_io == NULL) {
9208 0 : return NULL;
9209 : }
9210 :
9211 0 : if (!spdk_bdev_is_md_separate(bdev_io->bdev)) {
9212 0 : return NULL;
9213 : }
9214 :
9215 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_READ ||
9216 0 : bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
9217 0 : return bdev_io->u.bdev.md_buf;
9218 : }
9219 :
9220 0 : return NULL;
9221 : }
9222 :
9223 : void *
9224 0 : spdk_bdev_io_get_cb_arg(struct spdk_bdev_io *bdev_io)
9225 : {
9226 0 : if (bdev_io == NULL) {
9227 0 : assert(false);
9228 : return NULL;
9229 : }
9230 :
9231 0 : return bdev_io->internal.caller_ctx;
9232 : }
9233 :
9234 : void
9235 7 : spdk_bdev_module_list_add(struct spdk_bdev_module *bdev_module)
9236 : {
9237 :
9238 7 : if (spdk_bdev_module_list_find(bdev_module->name)) {
9239 0 : SPDK_ERRLOG("ERROR: module '%s' already registered.\n", bdev_module->name);
9240 0 : assert(false);
9241 : }
9242 :
9243 7 : spdk_spin_init(&bdev_module->internal.spinlock);
9244 7 : TAILQ_INIT(&bdev_module->internal.quiesced_ranges);
9245 :
9246 : /*
9247 : * Modules with examine callbacks must be initialized first, so they are
9248 : * ready to handle examine callbacks from later modules that will
9249 : * register physical bdevs.
9250 : */
9251 7 : if (bdev_module->examine_config != NULL || bdev_module->examine_disk != NULL) {
9252 4 : TAILQ_INSERT_HEAD(&g_bdev_mgr.bdev_modules, bdev_module, internal.tailq);
9253 : } else {
9254 3 : TAILQ_INSERT_TAIL(&g_bdev_mgr.bdev_modules, bdev_module, internal.tailq);
9255 : }
9256 7 : }
9257 :
9258 : struct spdk_bdev_module *
9259 7 : spdk_bdev_module_list_find(const char *name)
9260 : {
9261 : struct spdk_bdev_module *bdev_module;
9262 :
9263 14 : TAILQ_FOREACH(bdev_module, &g_bdev_mgr.bdev_modules, internal.tailq) {
9264 7 : if (strcmp(name, bdev_module->name) == 0) {
9265 0 : break;
9266 : }
9267 : }
9268 :
9269 7 : return bdev_module;
9270 : }
9271 :
9272 : static int
9273 6 : bdev_write_zero_buffer(struct spdk_bdev_io *bdev_io)
9274 : {
9275 : uint64_t num_blocks;
9276 6 : void *md_buf = NULL;
9277 :
9278 6 : num_blocks = bdev_io->u.bdev.num_blocks;
9279 :
9280 6 : if (spdk_bdev_is_md_separate(bdev_io->bdev)) {
9281 2 : md_buf = (char *)g_bdev_mgr.zero_buffer +
9282 2 : spdk_bdev_get_block_size(bdev_io->bdev) * num_blocks;
9283 : }
9284 :
9285 12 : return bdev_write_blocks_with_md(bdev_io->internal.desc,
9286 6 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
9287 : g_bdev_mgr.zero_buffer, md_buf,
9288 : bdev_io->u.bdev.offset_blocks, num_blocks,
9289 : bdev_write_zero_buffer_done, bdev_io);
9290 : }
9291 :
9292 : static void
9293 6 : bdev_write_zero_buffer_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
9294 : {
9295 6 : struct spdk_bdev_io *parent_io = cb_arg;
9296 :
9297 6 : spdk_bdev_free_io(bdev_io);
9298 :
9299 6 : parent_io->internal.status = success ? SPDK_BDEV_IO_STATUS_SUCCESS : SPDK_BDEV_IO_STATUS_FAILED;
9300 6 : parent_io->internal.cb(parent_io, success, parent_io->internal.caller_ctx);
9301 6 : }
9302 :
9303 : static void
9304 10 : bdev_set_qos_limit_done(struct set_qos_limit_ctx *ctx, int status)
9305 : {
9306 10 : spdk_spin_lock(&ctx->bdev->internal.spinlock);
9307 10 : ctx->bdev->internal.qos_mod_in_progress = false;
9308 10 : spdk_spin_unlock(&ctx->bdev->internal.spinlock);
9309 :
9310 10 : if (ctx->cb_fn) {
9311 8 : ctx->cb_fn(ctx->cb_arg, status);
9312 : }
9313 10 : free(ctx);
9314 10 : }
9315 :
9316 : static void
9317 2 : bdev_disable_qos_done(void *cb_arg)
9318 : {
9319 2 : struct set_qos_limit_ctx *ctx = cb_arg;
9320 2 : struct spdk_bdev *bdev = ctx->bdev;
9321 : struct spdk_bdev_qos *qos;
9322 :
9323 2 : spdk_spin_lock(&bdev->internal.spinlock);
9324 2 : qos = bdev->internal.qos;
9325 2 : bdev->internal.qos = NULL;
9326 2 : spdk_spin_unlock(&bdev->internal.spinlock);
9327 :
9328 2 : if (qos->thread != NULL) {
9329 2 : spdk_put_io_channel(spdk_io_channel_from_ctx(qos->ch));
9330 2 : spdk_poller_unregister(&qos->poller);
9331 : }
9332 :
9333 2 : free(qos);
9334 :
9335 2 : bdev_set_qos_limit_done(ctx, 0);
9336 2 : }
9337 :
9338 : static void
9339 2 : bdev_disable_qos_msg_done(struct spdk_bdev *bdev, void *_ctx, int status)
9340 : {
9341 2 : struct set_qos_limit_ctx *ctx = _ctx;
9342 : struct spdk_thread *thread;
9343 :
9344 2 : spdk_spin_lock(&bdev->internal.spinlock);
9345 2 : thread = bdev->internal.qos->thread;
9346 2 : spdk_spin_unlock(&bdev->internal.spinlock);
9347 :
9348 2 : if (thread != NULL) {
9349 2 : spdk_thread_send_msg(thread, bdev_disable_qos_done, ctx);
9350 : } else {
9351 0 : bdev_disable_qos_done(ctx);
9352 : }
9353 2 : }
9354 :
9355 : static void
9356 4 : bdev_disable_qos_msg(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
9357 : struct spdk_io_channel *ch, void *_ctx)
9358 : {
9359 4 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(ch);
9360 : struct spdk_bdev_io *bdev_io;
9361 :
9362 4 : bdev_ch->flags &= ~BDEV_CH_QOS_ENABLED;
9363 :
9364 6 : while (!TAILQ_EMPTY(&bdev_ch->qos_queued_io)) {
9365 : /* Re-submit the queued I/O. */
9366 2 : bdev_io = TAILQ_FIRST(&bdev_ch->qos_queued_io);
9367 2 : TAILQ_REMOVE(&bdev_ch->qos_queued_io, bdev_io, internal.link);
9368 2 : _bdev_io_submit(bdev_io);
9369 : }
9370 :
9371 4 : spdk_bdev_for_each_channel_continue(i, 0);
9372 4 : }
9373 :
9374 : static void
9375 1 : bdev_update_qos_rate_limit_msg(void *cb_arg)
9376 : {
9377 1 : struct set_qos_limit_ctx *ctx = cb_arg;
9378 1 : struct spdk_bdev *bdev = ctx->bdev;
9379 :
9380 1 : spdk_spin_lock(&bdev->internal.spinlock);
9381 1 : bdev_qos_update_max_quota_per_timeslice(bdev->internal.qos);
9382 1 : spdk_spin_unlock(&bdev->internal.spinlock);
9383 :
9384 1 : bdev_set_qos_limit_done(ctx, 0);
9385 1 : }
9386 :
9387 : static void
9388 9 : bdev_enable_qos_msg(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
9389 : struct spdk_io_channel *ch, void *_ctx)
9390 : {
9391 9 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(ch);
9392 :
9393 9 : spdk_spin_lock(&bdev->internal.spinlock);
9394 9 : bdev_enable_qos(bdev, bdev_ch);
9395 9 : spdk_spin_unlock(&bdev->internal.spinlock);
9396 9 : spdk_bdev_for_each_channel_continue(i, 0);
9397 9 : }
9398 :
9399 : static void
9400 6 : bdev_enable_qos_done(struct spdk_bdev *bdev, void *_ctx, int status)
9401 : {
9402 6 : struct set_qos_limit_ctx *ctx = _ctx;
9403 :
9404 6 : bdev_set_qos_limit_done(ctx, status);
9405 6 : }
9406 :
9407 : static void
9408 7 : bdev_set_qos_rate_limits(struct spdk_bdev *bdev, uint64_t *limits)
9409 : {
9410 : int i;
9411 :
9412 7 : assert(bdev->internal.qos != NULL);
9413 :
9414 35 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
9415 28 : if (limits[i] != SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
9416 28 : bdev->internal.qos->rate_limits[i].limit = limits[i];
9417 :
9418 28 : if (limits[i] == 0) {
9419 19 : bdev->internal.qos->rate_limits[i].limit =
9420 : SPDK_BDEV_QOS_LIMIT_NOT_DEFINED;
9421 : }
9422 : }
9423 : }
9424 7 : }
9425 :
9426 : void
9427 9 : spdk_bdev_set_qos_rate_limits(struct spdk_bdev *bdev, uint64_t *limits,
9428 : void (*cb_fn)(void *cb_arg, int status), void *cb_arg)
9429 : {
9430 : struct set_qos_limit_ctx *ctx;
9431 : uint32_t limit_set_complement;
9432 : uint64_t min_limit_per_sec;
9433 : int i;
9434 9 : bool disable_rate_limit = true;
9435 :
9436 45 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
9437 36 : if (limits[i] == SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
9438 0 : continue;
9439 : }
9440 :
9441 36 : if (limits[i] > 0) {
9442 10 : disable_rate_limit = false;
9443 : }
9444 :
9445 36 : if (bdev_qos_is_iops_rate_limit(i) == true) {
9446 9 : min_limit_per_sec = SPDK_BDEV_QOS_MIN_IOS_PER_SEC;
9447 : } else {
9448 27 : if (limits[i] > SPDK_BDEV_QOS_MAX_MBYTES_PER_SEC) {
9449 0 : SPDK_WARNLOG("Requested rate limit %" PRIu64 " will result in uint64_t overflow, "
9450 : "reset to %" PRIu64 "\n", limits[i], SPDK_BDEV_QOS_MAX_MBYTES_PER_SEC);
9451 0 : limits[i] = SPDK_BDEV_QOS_MAX_MBYTES_PER_SEC;
9452 : }
9453 : /* Change from megabyte to byte rate limit */
9454 27 : limits[i] = limits[i] * 1024 * 1024;
9455 27 : min_limit_per_sec = SPDK_BDEV_QOS_MIN_BYTES_PER_SEC;
9456 : }
9457 :
9458 36 : limit_set_complement = limits[i] % min_limit_per_sec;
9459 36 : if (limit_set_complement) {
9460 0 : SPDK_ERRLOG("Requested rate limit %" PRIu64 " is not a multiple of %" PRIu64 "\n",
9461 : limits[i], min_limit_per_sec);
9462 0 : limits[i] += min_limit_per_sec - limit_set_complement;
9463 0 : SPDK_ERRLOG("Round up the rate limit to %" PRIu64 "\n", limits[i]);
9464 : }
9465 : }
9466 :
9467 9 : ctx = calloc(1, sizeof(*ctx));
9468 9 : if (ctx == NULL) {
9469 0 : cb_fn(cb_arg, -ENOMEM);
9470 0 : return;
9471 : }
9472 :
9473 9 : ctx->cb_fn = cb_fn;
9474 9 : ctx->cb_arg = cb_arg;
9475 9 : ctx->bdev = bdev;
9476 :
9477 9 : spdk_spin_lock(&bdev->internal.spinlock);
9478 9 : if (bdev->internal.qos_mod_in_progress) {
9479 1 : spdk_spin_unlock(&bdev->internal.spinlock);
9480 1 : free(ctx);
9481 1 : cb_fn(cb_arg, -EAGAIN);
9482 1 : return;
9483 : }
9484 8 : bdev->internal.qos_mod_in_progress = true;
9485 :
9486 8 : if (disable_rate_limit == true && bdev->internal.qos) {
9487 10 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
9488 8 : if (limits[i] == SPDK_BDEV_QOS_LIMIT_NOT_DEFINED &&
9489 0 : (bdev->internal.qos->rate_limits[i].limit > 0 &&
9490 0 : bdev->internal.qos->rate_limits[i].limit !=
9491 : SPDK_BDEV_QOS_LIMIT_NOT_DEFINED)) {
9492 0 : disable_rate_limit = false;
9493 0 : break;
9494 : }
9495 : }
9496 : }
9497 :
9498 8 : if (disable_rate_limit == false) {
9499 5 : if (bdev->internal.qos == NULL) {
9500 4 : bdev->internal.qos = calloc(1, sizeof(*bdev->internal.qos));
9501 4 : if (!bdev->internal.qos) {
9502 0 : spdk_spin_unlock(&bdev->internal.spinlock);
9503 0 : SPDK_ERRLOG("Unable to allocate memory for QoS tracking\n");
9504 0 : bdev_set_qos_limit_done(ctx, -ENOMEM);
9505 0 : return;
9506 : }
9507 : }
9508 :
9509 5 : if (bdev->internal.qos->thread == NULL) {
9510 : /* Enabling */
9511 4 : bdev_set_qos_rate_limits(bdev, limits);
9512 :
9513 4 : spdk_bdev_for_each_channel(bdev, bdev_enable_qos_msg, ctx,
9514 : bdev_enable_qos_done);
9515 : } else {
9516 : /* Updating */
9517 1 : bdev_set_qos_rate_limits(bdev, limits);
9518 :
9519 1 : spdk_thread_send_msg(bdev->internal.qos->thread,
9520 : bdev_update_qos_rate_limit_msg, ctx);
9521 : }
9522 : } else {
9523 3 : if (bdev->internal.qos != NULL) {
9524 2 : bdev_set_qos_rate_limits(bdev, limits);
9525 :
9526 : /* Disabling */
9527 2 : spdk_bdev_for_each_channel(bdev, bdev_disable_qos_msg, ctx,
9528 : bdev_disable_qos_msg_done);
9529 : } else {
9530 1 : spdk_spin_unlock(&bdev->internal.spinlock);
9531 1 : bdev_set_qos_limit_done(ctx, 0);
9532 1 : return;
9533 : }
9534 : }
9535 :
9536 7 : spdk_spin_unlock(&bdev->internal.spinlock);
9537 : }
9538 :
9539 : struct spdk_bdev_histogram_ctx {
9540 : spdk_bdev_histogram_status_cb cb_fn;
9541 : void *cb_arg;
9542 : struct spdk_bdev *bdev;
9543 : int status;
9544 : };
9545 :
9546 : static void
9547 2 : bdev_histogram_disable_channel_cb(struct spdk_bdev *bdev, void *_ctx, int status)
9548 : {
9549 2 : struct spdk_bdev_histogram_ctx *ctx = _ctx;
9550 :
9551 2 : spdk_spin_lock(&ctx->bdev->internal.spinlock);
9552 2 : ctx->bdev->internal.histogram_in_progress = false;
9553 2 : spdk_spin_unlock(&ctx->bdev->internal.spinlock);
9554 2 : ctx->cb_fn(ctx->cb_arg, ctx->status);
9555 2 : free(ctx);
9556 2 : }
9557 :
9558 : static void
9559 3 : bdev_histogram_disable_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
9560 : struct spdk_io_channel *_ch, void *_ctx)
9561 : {
9562 3 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
9563 :
9564 3 : if (ch->histogram != NULL) {
9565 3 : spdk_histogram_data_free(ch->histogram);
9566 3 : ch->histogram = NULL;
9567 : }
9568 3 : spdk_bdev_for_each_channel_continue(i, 0);
9569 3 : }
9570 :
9571 : static void
9572 2 : bdev_histogram_enable_channel_cb(struct spdk_bdev *bdev, void *_ctx, int status)
9573 : {
9574 2 : struct spdk_bdev_histogram_ctx *ctx = _ctx;
9575 :
9576 2 : if (status != 0) {
9577 0 : ctx->status = status;
9578 0 : ctx->bdev->internal.histogram_enabled = false;
9579 0 : spdk_bdev_for_each_channel(ctx->bdev, bdev_histogram_disable_channel, ctx,
9580 : bdev_histogram_disable_channel_cb);
9581 : } else {
9582 2 : spdk_spin_lock(&ctx->bdev->internal.spinlock);
9583 2 : ctx->bdev->internal.histogram_in_progress = false;
9584 2 : spdk_spin_unlock(&ctx->bdev->internal.spinlock);
9585 2 : ctx->cb_fn(ctx->cb_arg, ctx->status);
9586 2 : free(ctx);
9587 : }
9588 2 : }
9589 :
9590 : static void
9591 3 : bdev_histogram_enable_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
9592 : struct spdk_io_channel *_ch, void *_ctx)
9593 : {
9594 3 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
9595 3 : int status = 0;
9596 :
9597 3 : if (ch->histogram == NULL) {
9598 3 : ch->histogram = spdk_histogram_data_alloc();
9599 3 : if (ch->histogram == NULL) {
9600 0 : status = -ENOMEM;
9601 : }
9602 : }
9603 :
9604 3 : spdk_bdev_for_each_channel_continue(i, status);
9605 3 : }
9606 :
9607 : void
9608 4 : spdk_bdev_histogram_enable_ext(struct spdk_bdev *bdev, spdk_bdev_histogram_status_cb cb_fn,
9609 : void *cb_arg, bool enable, struct spdk_bdev_enable_histogram_opts *opts)
9610 : {
9611 : struct spdk_bdev_histogram_ctx *ctx;
9612 :
9613 4 : ctx = calloc(1, sizeof(struct spdk_bdev_histogram_ctx));
9614 4 : if (ctx == NULL) {
9615 0 : cb_fn(cb_arg, -ENOMEM);
9616 0 : return;
9617 : }
9618 :
9619 4 : ctx->bdev = bdev;
9620 4 : ctx->status = 0;
9621 4 : ctx->cb_fn = cb_fn;
9622 4 : ctx->cb_arg = cb_arg;
9623 :
9624 4 : spdk_spin_lock(&bdev->internal.spinlock);
9625 4 : if (bdev->internal.histogram_in_progress) {
9626 0 : spdk_spin_unlock(&bdev->internal.spinlock);
9627 0 : free(ctx);
9628 0 : cb_fn(cb_arg, -EAGAIN);
9629 0 : return;
9630 : }
9631 :
9632 4 : bdev->internal.histogram_in_progress = true;
9633 4 : spdk_spin_unlock(&bdev->internal.spinlock);
9634 :
9635 4 : bdev->internal.histogram_enabled = enable;
9636 4 : bdev->internal.histogram_io_type = opts->io_type;
9637 :
9638 4 : if (enable) {
9639 : /* Allocate histogram for each channel */
9640 2 : spdk_bdev_for_each_channel(bdev, bdev_histogram_enable_channel, ctx,
9641 : bdev_histogram_enable_channel_cb);
9642 : } else {
9643 2 : spdk_bdev_for_each_channel(bdev, bdev_histogram_disable_channel, ctx,
9644 : bdev_histogram_disable_channel_cb);
9645 : }
9646 : }
9647 :
9648 : void
9649 4 : spdk_bdev_enable_histogram_opts_init(struct spdk_bdev_enable_histogram_opts *opts, size_t size)
9650 : {
9651 4 : if (opts == NULL) {
9652 0 : SPDK_ERRLOG("opts should not be NULL\n");
9653 0 : assert(opts != NULL);
9654 0 : return;
9655 : }
9656 4 : if (size == 0) {
9657 0 : SPDK_ERRLOG("size should not be zero\n");
9658 0 : assert(size != 0);
9659 0 : return;
9660 : }
9661 :
9662 4 : memset(opts, 0, size);
9663 4 : opts->size = size;
9664 :
9665 : #define FIELD_OK(field) \
9666 : offsetof(struct spdk_bdev_enable_histogram_opts, field) + sizeof(opts->field) <= size
9667 :
9668 : #define SET_FIELD(field, value) \
9669 : if (FIELD_OK(field)) { \
9670 : opts->field = value; \
9671 : } \
9672 :
9673 4 : SET_FIELD(io_type, 0);
9674 :
9675 : /* You should not remove this statement, but need to update the assert statement
9676 : * if you add a new field, and also add a corresponding SET_FIELD statement */
9677 : SPDK_STATIC_ASSERT(sizeof(struct spdk_bdev_enable_histogram_opts) == 9, "Incorrect size");
9678 :
9679 : #undef FIELD_OK
9680 : #undef SET_FIELD
9681 : }
9682 :
9683 : void
9684 4 : spdk_bdev_histogram_enable(struct spdk_bdev *bdev, spdk_bdev_histogram_status_cb cb_fn,
9685 : void *cb_arg, bool enable)
9686 : {
9687 4 : struct spdk_bdev_enable_histogram_opts opts;
9688 :
9689 4 : spdk_bdev_enable_histogram_opts_init(&opts, sizeof(opts));
9690 4 : spdk_bdev_histogram_enable_ext(bdev, cb_fn, cb_arg, enable, &opts);
9691 4 : }
9692 :
9693 : struct spdk_bdev_histogram_data_ctx {
9694 : spdk_bdev_histogram_data_cb cb_fn;
9695 : void *cb_arg;
9696 : struct spdk_bdev *bdev;
9697 : /** merged histogram data from all channels */
9698 : struct spdk_histogram_data *histogram;
9699 : };
9700 :
9701 : static void
9702 5 : bdev_histogram_get_channel_cb(struct spdk_bdev *bdev, void *_ctx, int status)
9703 : {
9704 5 : struct spdk_bdev_histogram_data_ctx *ctx = _ctx;
9705 :
9706 5 : ctx->cb_fn(ctx->cb_arg, status, ctx->histogram);
9707 5 : free(ctx);
9708 5 : }
9709 :
9710 : static void
9711 7 : bdev_histogram_get_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
9712 : struct spdk_io_channel *_ch, void *_ctx)
9713 : {
9714 7 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
9715 7 : struct spdk_bdev_histogram_data_ctx *ctx = _ctx;
9716 7 : int status = 0;
9717 :
9718 7 : if (ch->histogram == NULL) {
9719 1 : status = -EFAULT;
9720 : } else {
9721 6 : spdk_histogram_data_merge(ctx->histogram, ch->histogram);
9722 : }
9723 :
9724 7 : spdk_bdev_for_each_channel_continue(i, status);
9725 7 : }
9726 :
9727 : void
9728 5 : spdk_bdev_histogram_get(struct spdk_bdev *bdev, struct spdk_histogram_data *histogram,
9729 : spdk_bdev_histogram_data_cb cb_fn,
9730 : void *cb_arg)
9731 : {
9732 : struct spdk_bdev_histogram_data_ctx *ctx;
9733 :
9734 5 : ctx = calloc(1, sizeof(struct spdk_bdev_histogram_data_ctx));
9735 5 : if (ctx == NULL) {
9736 0 : cb_fn(cb_arg, -ENOMEM, NULL);
9737 0 : return;
9738 : }
9739 :
9740 5 : ctx->bdev = bdev;
9741 5 : ctx->cb_fn = cb_fn;
9742 5 : ctx->cb_arg = cb_arg;
9743 :
9744 5 : ctx->histogram = histogram;
9745 :
9746 5 : spdk_bdev_for_each_channel(bdev, bdev_histogram_get_channel, ctx,
9747 : bdev_histogram_get_channel_cb);
9748 : }
9749 :
9750 : void
9751 2 : spdk_bdev_channel_get_histogram(struct spdk_io_channel *ch, spdk_bdev_histogram_data_cb cb_fn,
9752 : void *cb_arg)
9753 : {
9754 2 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(ch);
9755 2 : int status = 0;
9756 :
9757 2 : assert(cb_fn != NULL);
9758 :
9759 2 : if (bdev_ch->histogram == NULL) {
9760 1 : status = -EFAULT;
9761 : }
9762 2 : cb_fn(cb_arg, status, bdev_ch->histogram);
9763 2 : }
9764 :
9765 : size_t
9766 0 : spdk_bdev_get_media_events(struct spdk_bdev_desc *desc, struct spdk_bdev_media_event *events,
9767 : size_t max_events)
9768 : {
9769 : struct media_event_entry *entry;
9770 0 : size_t num_events = 0;
9771 :
9772 0 : for (; num_events < max_events; ++num_events) {
9773 0 : entry = TAILQ_FIRST(&desc->pending_media_events);
9774 0 : if (entry == NULL) {
9775 0 : break;
9776 : }
9777 :
9778 0 : events[num_events] = entry->event;
9779 0 : TAILQ_REMOVE(&desc->pending_media_events, entry, tailq);
9780 0 : TAILQ_INSERT_TAIL(&desc->free_media_events, entry, tailq);
9781 : }
9782 :
9783 0 : return num_events;
9784 : }
9785 :
9786 : int
9787 0 : spdk_bdev_push_media_events(struct spdk_bdev *bdev, const struct spdk_bdev_media_event *events,
9788 : size_t num_events)
9789 : {
9790 : struct spdk_bdev_desc *desc;
9791 : struct media_event_entry *entry;
9792 : size_t event_id;
9793 0 : int rc = 0;
9794 :
9795 0 : assert(bdev->media_events);
9796 :
9797 0 : spdk_spin_lock(&bdev->internal.spinlock);
9798 0 : TAILQ_FOREACH(desc, &bdev->internal.open_descs, link) {
9799 0 : if (desc->write) {
9800 0 : break;
9801 : }
9802 : }
9803 :
9804 0 : if (desc == NULL || desc->media_events_buffer == NULL) {
9805 0 : rc = -ENODEV;
9806 0 : goto out;
9807 : }
9808 :
9809 0 : for (event_id = 0; event_id < num_events; ++event_id) {
9810 0 : entry = TAILQ_FIRST(&desc->free_media_events);
9811 0 : if (entry == NULL) {
9812 0 : break;
9813 : }
9814 :
9815 0 : TAILQ_REMOVE(&desc->free_media_events, entry, tailq);
9816 0 : TAILQ_INSERT_TAIL(&desc->pending_media_events, entry, tailq);
9817 0 : entry->event = events[event_id];
9818 : }
9819 :
9820 0 : rc = event_id;
9821 0 : out:
9822 0 : spdk_spin_unlock(&bdev->internal.spinlock);
9823 0 : return rc;
9824 : }
9825 :
9826 : static void
9827 0 : _media_management_notify(void *arg)
9828 : {
9829 0 : struct spdk_bdev_desc *desc = arg;
9830 :
9831 0 : _event_notify(desc, SPDK_BDEV_EVENT_MEDIA_MANAGEMENT);
9832 0 : }
9833 :
9834 : void
9835 0 : spdk_bdev_notify_media_management(struct spdk_bdev *bdev)
9836 : {
9837 : struct spdk_bdev_desc *desc;
9838 :
9839 0 : spdk_spin_lock(&bdev->internal.spinlock);
9840 0 : TAILQ_FOREACH(desc, &bdev->internal.open_descs, link) {
9841 0 : if (!TAILQ_EMPTY(&desc->pending_media_events)) {
9842 0 : event_notify(desc, _media_management_notify);
9843 : }
9844 : }
9845 0 : spdk_spin_unlock(&bdev->internal.spinlock);
9846 0 : }
9847 :
9848 : struct locked_lba_range_ctx {
9849 : struct lba_range range;
9850 : struct lba_range *current_range;
9851 : struct lba_range *owner_range;
9852 : struct spdk_poller *poller;
9853 : lock_range_cb cb_fn;
9854 : void *cb_arg;
9855 : };
9856 :
9857 : static void
9858 0 : bdev_lock_error_cleanup_cb(struct spdk_bdev *bdev, void *_ctx, int status)
9859 : {
9860 0 : struct locked_lba_range_ctx *ctx = _ctx;
9861 :
9862 0 : ctx->cb_fn(&ctx->range, ctx->cb_arg, -ENOMEM);
9863 0 : free(ctx);
9864 0 : }
9865 :
9866 : static void bdev_unlock_lba_range_get_channel(struct spdk_bdev_channel_iter *i,
9867 : struct spdk_bdev *bdev, struct spdk_io_channel *ch, void *_ctx);
9868 :
9869 : static void
9870 14 : bdev_lock_lba_range_cb(struct spdk_bdev *bdev, void *_ctx, int status)
9871 : {
9872 14 : struct locked_lba_range_ctx *ctx = _ctx;
9873 :
9874 14 : if (status == -ENOMEM) {
9875 : /* One of the channels could not allocate a range object.
9876 : * So we have to go back and clean up any ranges that were
9877 : * allocated successfully before we return error status to
9878 : * the caller. We can reuse the unlock function to do that
9879 : * clean up.
9880 : */
9881 0 : spdk_bdev_for_each_channel(bdev, bdev_unlock_lba_range_get_channel, ctx,
9882 : bdev_lock_error_cleanup_cb);
9883 0 : return;
9884 : }
9885 :
9886 : /* All channels have locked this range and no I/O overlapping the range
9887 : * are outstanding! Set the owner_ch for the range object for the
9888 : * locking channel, so that this channel will know that it is allowed
9889 : * to write to this range.
9890 : */
9891 14 : if (ctx->owner_range != NULL) {
9892 10 : ctx->owner_range->owner_ch = ctx->range.owner_ch;
9893 : }
9894 :
9895 14 : ctx->cb_fn(&ctx->range, ctx->cb_arg, status);
9896 :
9897 : /* Don't free the ctx here. Its range is in the bdev's global list of
9898 : * locked ranges still, and will be removed and freed when this range
9899 : * is later unlocked.
9900 : */
9901 : }
9902 :
9903 : static int
9904 17 : bdev_lock_lba_range_check_io(void *_i)
9905 : {
9906 17 : struct spdk_bdev_channel_iter *i = _i;
9907 17 : struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i->i);
9908 17 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
9909 17 : struct locked_lba_range_ctx *ctx = i->ctx;
9910 17 : struct lba_range *range = ctx->current_range;
9911 : struct spdk_bdev_io *bdev_io;
9912 :
9913 17 : spdk_poller_unregister(&ctx->poller);
9914 :
9915 : /* The range is now in the locked_ranges, so no new IO can be submitted to this
9916 : * range. But we need to wait until any outstanding IO overlapping with this range
9917 : * are completed.
9918 : */
9919 18 : TAILQ_FOREACH(bdev_io, &ch->io_submitted, internal.ch_link) {
9920 3 : if (bdev_io_range_is_locked(bdev_io, range)) {
9921 2 : ctx->poller = SPDK_POLLER_REGISTER(bdev_lock_lba_range_check_io, i, 100);
9922 2 : return SPDK_POLLER_BUSY;
9923 : }
9924 : }
9925 :
9926 15 : spdk_bdev_for_each_channel_continue(i, 0);
9927 15 : return SPDK_POLLER_BUSY;
9928 : }
9929 :
9930 : static void
9931 15 : bdev_lock_lba_range_get_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
9932 : struct spdk_io_channel *_ch, void *_ctx)
9933 : {
9934 15 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
9935 15 : struct locked_lba_range_ctx *ctx = _ctx;
9936 : struct lba_range *range;
9937 :
9938 16 : TAILQ_FOREACH(range, &ch->locked_ranges, tailq) {
9939 1 : if (range->length == ctx->range.length &&
9940 0 : range->offset == ctx->range.offset &&
9941 0 : range->locked_ctx == ctx->range.locked_ctx) {
9942 : /* This range already exists on this channel, so don't add
9943 : * it again. This can happen when a new channel is created
9944 : * while the for_each_channel operation is in progress.
9945 : * Do not check for outstanding I/O in that case, since the
9946 : * range was locked before any I/O could be submitted to the
9947 : * new channel.
9948 : */
9949 0 : spdk_bdev_for_each_channel_continue(i, 0);
9950 0 : return;
9951 : }
9952 : }
9953 :
9954 15 : range = calloc(1, sizeof(*range));
9955 15 : if (range == NULL) {
9956 0 : spdk_bdev_for_each_channel_continue(i, -ENOMEM);
9957 0 : return;
9958 : }
9959 :
9960 15 : range->length = ctx->range.length;
9961 15 : range->offset = ctx->range.offset;
9962 15 : range->locked_ctx = ctx->range.locked_ctx;
9963 15 : range->quiesce = ctx->range.quiesce;
9964 15 : ctx->current_range = range;
9965 15 : if (ctx->range.owner_ch == ch) {
9966 : /* This is the range object for the channel that will hold
9967 : * the lock. Store it in the ctx object so that we can easily
9968 : * set its owner_ch after the lock is finally acquired.
9969 : */
9970 10 : ctx->owner_range = range;
9971 : }
9972 15 : TAILQ_INSERT_TAIL(&ch->locked_ranges, range, tailq);
9973 15 : bdev_lock_lba_range_check_io(i);
9974 : }
9975 :
9976 : static void
9977 14 : bdev_lock_lba_range_ctx(struct spdk_bdev *bdev, struct locked_lba_range_ctx *ctx)
9978 : {
9979 14 : assert(spdk_get_thread() == ctx->range.owner_thread);
9980 14 : assert(ctx->range.owner_ch == NULL ||
9981 : spdk_io_channel_get_thread(ctx->range.owner_ch->channel) == ctx->range.owner_thread);
9982 :
9983 : /* We will add a copy of this range to each channel now. */
9984 14 : spdk_bdev_for_each_channel(bdev, bdev_lock_lba_range_get_channel, ctx,
9985 : bdev_lock_lba_range_cb);
9986 14 : }
9987 :
9988 : static bool
9989 17 : bdev_lba_range_overlaps_tailq(struct lba_range *range, lba_range_tailq_t *tailq)
9990 : {
9991 : struct lba_range *r;
9992 :
9993 18 : TAILQ_FOREACH(r, tailq, tailq) {
9994 4 : if (bdev_lba_range_overlapped(range, r)) {
9995 3 : return true;
9996 : }
9997 : }
9998 14 : return false;
9999 : }
10000 :
10001 : static void bdev_quiesce_range_locked(struct lba_range *range, void *ctx, int status);
10002 :
10003 : static int
10004 14 : _bdev_lock_lba_range(struct spdk_bdev *bdev, struct spdk_bdev_channel *ch,
10005 : uint64_t offset, uint64_t length,
10006 : lock_range_cb cb_fn, void *cb_arg)
10007 : {
10008 : struct locked_lba_range_ctx *ctx;
10009 :
10010 14 : ctx = calloc(1, sizeof(*ctx));
10011 14 : if (ctx == NULL) {
10012 0 : return -ENOMEM;
10013 : }
10014 :
10015 14 : ctx->range.offset = offset;
10016 14 : ctx->range.length = length;
10017 14 : ctx->range.owner_thread = spdk_get_thread();
10018 14 : ctx->range.owner_ch = ch;
10019 14 : ctx->range.locked_ctx = cb_arg;
10020 14 : ctx->range.bdev = bdev;
10021 14 : ctx->range.quiesce = (cb_fn == bdev_quiesce_range_locked);
10022 14 : ctx->cb_fn = cb_fn;
10023 14 : ctx->cb_arg = cb_arg;
10024 :
10025 14 : spdk_spin_lock(&bdev->internal.spinlock);
10026 14 : if (bdev_lba_range_overlaps_tailq(&ctx->range, &bdev->internal.locked_ranges)) {
10027 : /* There is an active lock overlapping with this range.
10028 : * Put it on the pending list until this range no
10029 : * longer overlaps with another.
10030 : */
10031 2 : TAILQ_INSERT_TAIL(&bdev->internal.pending_locked_ranges, &ctx->range, tailq);
10032 : } else {
10033 12 : TAILQ_INSERT_TAIL(&bdev->internal.locked_ranges, &ctx->range, tailq);
10034 12 : bdev_lock_lba_range_ctx(bdev, ctx);
10035 : }
10036 14 : spdk_spin_unlock(&bdev->internal.spinlock);
10037 14 : return 0;
10038 : }
10039 :
10040 : static int
10041 10 : bdev_lock_lba_range(struct spdk_bdev_desc *desc, struct spdk_io_channel *_ch,
10042 : uint64_t offset, uint64_t length,
10043 : lock_range_cb cb_fn, void *cb_arg)
10044 : {
10045 10 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
10046 10 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
10047 :
10048 10 : if (cb_arg == NULL) {
10049 0 : SPDK_ERRLOG("cb_arg must not be NULL\n");
10050 0 : return -EINVAL;
10051 : }
10052 :
10053 10 : return _bdev_lock_lba_range(bdev, ch, offset, length, cb_fn, cb_arg);
10054 : }
10055 :
10056 : static void
10057 2 : bdev_lock_lba_range_ctx_msg(void *_ctx)
10058 : {
10059 2 : struct locked_lba_range_ctx *ctx = _ctx;
10060 :
10061 2 : bdev_lock_lba_range_ctx(ctx->range.bdev, ctx);
10062 2 : }
10063 :
10064 : static void
10065 14 : bdev_unlock_lba_range_cb(struct spdk_bdev *bdev, void *_ctx, int status)
10066 : {
10067 14 : struct locked_lba_range_ctx *ctx = _ctx;
10068 : struct locked_lba_range_ctx *pending_ctx;
10069 : struct lba_range *range, *tmp;
10070 :
10071 14 : spdk_spin_lock(&bdev->internal.spinlock);
10072 : /* Check if there are any pending locked ranges that overlap with this range
10073 : * that was just unlocked. If there are, check that it doesn't overlap with any
10074 : * other locked ranges before calling bdev_lock_lba_range_ctx which will start
10075 : * the lock process.
10076 : */
10077 17 : TAILQ_FOREACH_SAFE(range, &bdev->internal.pending_locked_ranges, tailq, tmp) {
10078 3 : if (bdev_lba_range_overlapped(range, &ctx->range) &&
10079 3 : !bdev_lba_range_overlaps_tailq(range, &bdev->internal.locked_ranges)) {
10080 2 : TAILQ_REMOVE(&bdev->internal.pending_locked_ranges, range, tailq);
10081 2 : pending_ctx = SPDK_CONTAINEROF(range, struct locked_lba_range_ctx, range);
10082 2 : TAILQ_INSERT_TAIL(&bdev->internal.locked_ranges, range, tailq);
10083 2 : spdk_thread_send_msg(pending_ctx->range.owner_thread,
10084 : bdev_lock_lba_range_ctx_msg, pending_ctx);
10085 : }
10086 : }
10087 14 : spdk_spin_unlock(&bdev->internal.spinlock);
10088 :
10089 14 : ctx->cb_fn(&ctx->range, ctx->cb_arg, status);
10090 14 : free(ctx);
10091 14 : }
10092 :
10093 : static void
10094 16 : bdev_unlock_lba_range_get_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
10095 : struct spdk_io_channel *_ch, void *_ctx)
10096 : {
10097 16 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
10098 16 : struct locked_lba_range_ctx *ctx = _ctx;
10099 16 : TAILQ_HEAD(, spdk_bdev_io) io_locked;
10100 : struct spdk_bdev_io *bdev_io;
10101 : struct lba_range *range;
10102 :
10103 16 : TAILQ_FOREACH(range, &ch->locked_ranges, tailq) {
10104 16 : if (ctx->range.offset == range->offset &&
10105 16 : ctx->range.length == range->length &&
10106 16 : ctx->range.locked_ctx == range->locked_ctx) {
10107 16 : TAILQ_REMOVE(&ch->locked_ranges, range, tailq);
10108 16 : free(range);
10109 16 : break;
10110 : }
10111 : }
10112 :
10113 : /* Note: we should almost always be able to assert that the range specified
10114 : * was found. But there are some very rare corner cases where a new channel
10115 : * gets created simultaneously with a range unlock, where this function
10116 : * would execute on that new channel and wouldn't have the range.
10117 : * We also use this to clean up range allocations when a later allocation
10118 : * fails in the locking path.
10119 : * So we can't actually assert() here.
10120 : */
10121 :
10122 : /* Swap the locked IO into a temporary list, and then try to submit them again.
10123 : * We could hyper-optimize this to only resubmit locked I/O that overlap
10124 : * with the range that was just unlocked, but this isn't a performance path so
10125 : * we go for simplicity here.
10126 : */
10127 16 : TAILQ_INIT(&io_locked);
10128 16 : TAILQ_SWAP(&ch->io_locked, &io_locked, spdk_bdev_io, internal.ch_link);
10129 19 : while (!TAILQ_EMPTY(&io_locked)) {
10130 3 : bdev_io = TAILQ_FIRST(&io_locked);
10131 3 : TAILQ_REMOVE(&io_locked, bdev_io, internal.ch_link);
10132 3 : bdev_io_submit(bdev_io);
10133 : }
10134 :
10135 16 : spdk_bdev_for_each_channel_continue(i, 0);
10136 16 : }
10137 :
10138 : static int
10139 14 : _bdev_unlock_lba_range(struct spdk_bdev *bdev, uint64_t offset, uint64_t length,
10140 : lock_range_cb cb_fn, void *cb_arg)
10141 : {
10142 : struct locked_lba_range_ctx *ctx;
10143 : struct lba_range *range;
10144 :
10145 14 : spdk_spin_lock(&bdev->internal.spinlock);
10146 : /* To start the unlock the process, we find the range in the bdev's locked_ranges
10147 : * and remove it. This ensures new channels don't inherit the locked range.
10148 : * Then we will send a message to each channel to remove the range from its
10149 : * per-channel list.
10150 : */
10151 14 : TAILQ_FOREACH(range, &bdev->internal.locked_ranges, tailq) {
10152 14 : if (range->offset == offset && range->length == length &&
10153 14 : (range->owner_ch == NULL || range->locked_ctx == cb_arg)) {
10154 : break;
10155 : }
10156 : }
10157 14 : if (range == NULL) {
10158 0 : assert(false);
10159 : spdk_spin_unlock(&bdev->internal.spinlock);
10160 : return -EINVAL;
10161 : }
10162 14 : TAILQ_REMOVE(&bdev->internal.locked_ranges, range, tailq);
10163 14 : ctx = SPDK_CONTAINEROF(range, struct locked_lba_range_ctx, range);
10164 14 : spdk_spin_unlock(&bdev->internal.spinlock);
10165 :
10166 14 : ctx->cb_fn = cb_fn;
10167 14 : ctx->cb_arg = cb_arg;
10168 :
10169 14 : spdk_bdev_for_each_channel(bdev, bdev_unlock_lba_range_get_channel, ctx,
10170 : bdev_unlock_lba_range_cb);
10171 14 : return 0;
10172 : }
10173 :
10174 : static int
10175 12 : bdev_unlock_lba_range(struct spdk_bdev_desc *desc, struct spdk_io_channel *_ch,
10176 : uint64_t offset, uint64_t length,
10177 : lock_range_cb cb_fn, void *cb_arg)
10178 : {
10179 12 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
10180 12 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
10181 : struct lba_range *range;
10182 12 : bool range_found = false;
10183 :
10184 : /* Let's make sure the specified channel actually has a lock on
10185 : * the specified range. Note that the range must match exactly.
10186 : */
10187 14 : TAILQ_FOREACH(range, &ch->locked_ranges, tailq) {
10188 12 : if (range->offset == offset && range->length == length &&
10189 11 : range->owner_ch == ch && range->locked_ctx == cb_arg) {
10190 10 : range_found = true;
10191 10 : break;
10192 : }
10193 : }
10194 :
10195 12 : if (!range_found) {
10196 2 : return -EINVAL;
10197 : }
10198 :
10199 10 : return _bdev_unlock_lba_range(bdev, offset, length, cb_fn, cb_arg);
10200 : }
10201 :
10202 : struct bdev_quiesce_ctx {
10203 : spdk_bdev_quiesce_cb cb_fn;
10204 : void *cb_arg;
10205 : };
10206 :
10207 : static void
10208 4 : bdev_unquiesce_range_unlocked(struct lba_range *range, void *ctx, int status)
10209 : {
10210 4 : struct bdev_quiesce_ctx *quiesce_ctx = ctx;
10211 :
10212 4 : if (quiesce_ctx->cb_fn != NULL) {
10213 4 : quiesce_ctx->cb_fn(quiesce_ctx->cb_arg, status);
10214 : }
10215 :
10216 4 : free(quiesce_ctx);
10217 4 : }
10218 :
10219 : static void
10220 4 : bdev_quiesce_range_locked(struct lba_range *range, void *ctx, int status)
10221 : {
10222 4 : struct bdev_quiesce_ctx *quiesce_ctx = ctx;
10223 4 : struct spdk_bdev_module *module = range->bdev->module;
10224 :
10225 4 : if (status != 0) {
10226 0 : if (quiesce_ctx->cb_fn != NULL) {
10227 0 : quiesce_ctx->cb_fn(quiesce_ctx->cb_arg, status);
10228 : }
10229 0 : free(quiesce_ctx);
10230 0 : return;
10231 : }
10232 :
10233 4 : spdk_spin_lock(&module->internal.spinlock);
10234 4 : TAILQ_INSERT_TAIL(&module->internal.quiesced_ranges, range, tailq_module);
10235 4 : spdk_spin_unlock(&module->internal.spinlock);
10236 :
10237 4 : if (quiesce_ctx->cb_fn != NULL) {
10238 : /* copy the context in case the range is unlocked by the callback */
10239 4 : struct bdev_quiesce_ctx tmp = *quiesce_ctx;
10240 :
10241 4 : quiesce_ctx->cb_fn = NULL;
10242 4 : quiesce_ctx->cb_arg = NULL;
10243 :
10244 4 : tmp.cb_fn(tmp.cb_arg, status);
10245 : }
10246 : /* quiesce_ctx will be freed on unquiesce */
10247 : }
10248 :
10249 : static int
10250 9 : _spdk_bdev_quiesce(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
10251 : uint64_t offset, uint64_t length,
10252 : spdk_bdev_quiesce_cb cb_fn, void *cb_arg,
10253 : bool unquiesce)
10254 : {
10255 : struct bdev_quiesce_ctx *quiesce_ctx;
10256 : int rc;
10257 :
10258 9 : if (module != bdev->module) {
10259 0 : SPDK_ERRLOG("Bdev does not belong to specified module.\n");
10260 0 : return -EINVAL;
10261 : }
10262 :
10263 9 : if (!bdev_io_valid_blocks(bdev, offset, length)) {
10264 0 : return -EINVAL;
10265 : }
10266 :
10267 9 : if (unquiesce) {
10268 : struct lba_range *range;
10269 :
10270 : /* Make sure the specified range is actually quiesced in the specified module and
10271 : * then remove it from the list. Note that the range must match exactly.
10272 : */
10273 5 : spdk_spin_lock(&module->internal.spinlock);
10274 6 : TAILQ_FOREACH(range, &module->internal.quiesced_ranges, tailq_module) {
10275 5 : if (range->bdev == bdev && range->offset == offset && range->length == length) {
10276 4 : TAILQ_REMOVE(&module->internal.quiesced_ranges, range, tailq_module);
10277 4 : break;
10278 : }
10279 : }
10280 5 : spdk_spin_unlock(&module->internal.spinlock);
10281 :
10282 5 : if (range == NULL) {
10283 1 : SPDK_ERRLOG("The range to unquiesce was not found.\n");
10284 1 : return -EINVAL;
10285 : }
10286 :
10287 4 : quiesce_ctx = range->locked_ctx;
10288 4 : quiesce_ctx->cb_fn = cb_fn;
10289 4 : quiesce_ctx->cb_arg = cb_arg;
10290 :
10291 4 : rc = _bdev_unlock_lba_range(bdev, offset, length, bdev_unquiesce_range_unlocked, quiesce_ctx);
10292 : } else {
10293 4 : quiesce_ctx = malloc(sizeof(*quiesce_ctx));
10294 4 : if (quiesce_ctx == NULL) {
10295 0 : return -ENOMEM;
10296 : }
10297 :
10298 4 : quiesce_ctx->cb_fn = cb_fn;
10299 4 : quiesce_ctx->cb_arg = cb_arg;
10300 :
10301 4 : rc = _bdev_lock_lba_range(bdev, NULL, offset, length, bdev_quiesce_range_locked, quiesce_ctx);
10302 4 : if (rc != 0) {
10303 0 : free(quiesce_ctx);
10304 : }
10305 : }
10306 :
10307 8 : return rc;
10308 : }
10309 :
10310 : int
10311 3 : spdk_bdev_quiesce(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
10312 : spdk_bdev_quiesce_cb cb_fn, void *cb_arg)
10313 : {
10314 3 : return _spdk_bdev_quiesce(bdev, module, 0, bdev->blockcnt, cb_fn, cb_arg, false);
10315 : }
10316 :
10317 : int
10318 3 : spdk_bdev_unquiesce(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
10319 : spdk_bdev_quiesce_cb cb_fn, void *cb_arg)
10320 : {
10321 3 : return _spdk_bdev_quiesce(bdev, module, 0, bdev->blockcnt, cb_fn, cb_arg, true);
10322 : }
10323 :
10324 : int
10325 1 : spdk_bdev_quiesce_range(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
10326 : uint64_t offset, uint64_t length,
10327 : spdk_bdev_quiesce_cb cb_fn, void *cb_arg)
10328 : {
10329 1 : return _spdk_bdev_quiesce(bdev, module, offset, length, cb_fn, cb_arg, false);
10330 : }
10331 :
10332 : int
10333 2 : spdk_bdev_unquiesce_range(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
10334 : uint64_t offset, uint64_t length,
10335 : spdk_bdev_quiesce_cb cb_fn, void *cb_arg)
10336 : {
10337 2 : return _spdk_bdev_quiesce(bdev, module, offset, length, cb_fn, cb_arg, true);
10338 : }
10339 :
10340 : int
10341 269 : spdk_bdev_get_memory_domains(struct spdk_bdev *bdev, struct spdk_memory_domain **domains,
10342 : int array_size)
10343 : {
10344 269 : if (!bdev) {
10345 1 : return -EINVAL;
10346 : }
10347 :
10348 268 : if (bdev->fn_table->get_memory_domains) {
10349 3 : return bdev->fn_table->get_memory_domains(bdev->ctxt, domains, array_size);
10350 : }
10351 :
10352 265 : return 0;
10353 : }
10354 :
10355 : struct spdk_bdev_for_each_io_ctx {
10356 : void *ctx;
10357 : spdk_bdev_io_fn fn;
10358 : spdk_bdev_for_each_io_cb cb;
10359 : };
10360 :
10361 : static void
10362 0 : bdev_channel_for_each_io(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
10363 : struct spdk_io_channel *io_ch, void *_ctx)
10364 : {
10365 0 : struct spdk_bdev_for_each_io_ctx *ctx = _ctx;
10366 0 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
10367 : struct spdk_bdev_io *bdev_io;
10368 0 : int rc = 0;
10369 :
10370 0 : TAILQ_FOREACH(bdev_io, &bdev_ch->io_submitted, internal.ch_link) {
10371 0 : rc = ctx->fn(ctx->ctx, bdev_io);
10372 0 : if (rc != 0) {
10373 0 : break;
10374 : }
10375 : }
10376 :
10377 0 : spdk_bdev_for_each_channel_continue(i, rc);
10378 0 : }
10379 :
10380 : static void
10381 0 : bdev_for_each_io_done(struct spdk_bdev *bdev, void *_ctx, int status)
10382 : {
10383 0 : struct spdk_bdev_for_each_io_ctx *ctx = _ctx;
10384 :
10385 0 : ctx->cb(ctx->ctx, status);
10386 :
10387 0 : free(ctx);
10388 0 : }
10389 :
10390 : void
10391 0 : spdk_bdev_for_each_bdev_io(struct spdk_bdev *bdev, void *_ctx, spdk_bdev_io_fn fn,
10392 : spdk_bdev_for_each_io_cb cb)
10393 : {
10394 : struct spdk_bdev_for_each_io_ctx *ctx;
10395 :
10396 0 : assert(fn != NULL && cb != NULL);
10397 :
10398 0 : ctx = calloc(1, sizeof(*ctx));
10399 0 : if (ctx == NULL) {
10400 0 : SPDK_ERRLOG("Failed to allocate context.\n");
10401 0 : cb(_ctx, -ENOMEM);
10402 0 : return;
10403 : }
10404 :
10405 0 : ctx->ctx = _ctx;
10406 0 : ctx->fn = fn;
10407 0 : ctx->cb = cb;
10408 :
10409 0 : spdk_bdev_for_each_channel(bdev, bdev_channel_for_each_io, ctx,
10410 : bdev_for_each_io_done);
10411 : }
10412 :
10413 : void
10414 132 : spdk_bdev_for_each_channel_continue(struct spdk_bdev_channel_iter *iter, int status)
10415 : {
10416 132 : spdk_for_each_channel_continue(iter->i, status);
10417 132 : }
10418 :
10419 : static struct spdk_bdev *
10420 360 : io_channel_iter_get_bdev(struct spdk_io_channel_iter *i)
10421 : {
10422 360 : void *io_device = spdk_io_channel_iter_get_io_device(i);
10423 :
10424 360 : return __bdev_from_io_dev(io_device);
10425 : }
10426 :
10427 : static void
10428 132 : bdev_each_channel_msg(struct spdk_io_channel_iter *i)
10429 : {
10430 132 : struct spdk_bdev_channel_iter *iter = spdk_io_channel_iter_get_ctx(i);
10431 132 : struct spdk_bdev *bdev = io_channel_iter_get_bdev(i);
10432 132 : struct spdk_io_channel *ch = spdk_io_channel_iter_get_channel(i);
10433 :
10434 132 : iter->i = i;
10435 132 : iter->fn(iter, bdev, ch, iter->ctx);
10436 132 : }
10437 :
10438 : static void
10439 228 : bdev_each_channel_cpl(struct spdk_io_channel_iter *i, int status)
10440 : {
10441 228 : struct spdk_bdev_channel_iter *iter = spdk_io_channel_iter_get_ctx(i);
10442 228 : struct spdk_bdev *bdev = io_channel_iter_get_bdev(i);
10443 :
10444 228 : iter->i = i;
10445 228 : iter->cpl(bdev, iter->ctx, status);
10446 :
10447 228 : free(iter);
10448 228 : }
10449 :
10450 : void
10451 228 : spdk_bdev_for_each_channel(struct spdk_bdev *bdev, spdk_bdev_for_each_channel_msg fn,
10452 : void *ctx, spdk_bdev_for_each_channel_done cpl)
10453 : {
10454 : struct spdk_bdev_channel_iter *iter;
10455 :
10456 228 : assert(bdev != NULL && fn != NULL && ctx != NULL);
10457 :
10458 228 : iter = calloc(1, sizeof(struct spdk_bdev_channel_iter));
10459 228 : if (iter == NULL) {
10460 0 : SPDK_ERRLOG("Unable to allocate iterator\n");
10461 0 : assert(false);
10462 : return;
10463 : }
10464 :
10465 228 : iter->fn = fn;
10466 228 : iter->cpl = cpl;
10467 228 : iter->ctx = ctx;
10468 :
10469 228 : spdk_for_each_channel(__bdev_to_io_dev(bdev), bdev_each_channel_msg,
10470 : iter, bdev_each_channel_cpl);
10471 : }
10472 :
10473 : static void
10474 3 : bdev_copy_do_write_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
10475 : {
10476 3 : struct spdk_bdev_io *parent_io = cb_arg;
10477 :
10478 3 : spdk_bdev_free_io(bdev_io);
10479 :
10480 : /* Check return status of write */
10481 3 : parent_io->internal.status = success ? SPDK_BDEV_IO_STATUS_SUCCESS : SPDK_BDEV_IO_STATUS_FAILED;
10482 3 : parent_io->internal.cb(parent_io, success, parent_io->internal.caller_ctx);
10483 3 : }
10484 :
10485 : static void
10486 3 : bdev_copy_do_write(void *_bdev_io)
10487 : {
10488 3 : struct spdk_bdev_io *bdev_io = _bdev_io;
10489 : int rc;
10490 :
10491 : /* Write blocks */
10492 6 : rc = spdk_bdev_write_blocks_with_md(bdev_io->internal.desc,
10493 3 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
10494 3 : bdev_io->u.bdev.iovs[0].iov_base,
10495 : bdev_io->u.bdev.md_buf, bdev_io->u.bdev.offset_blocks,
10496 : bdev_io->u.bdev.num_blocks, bdev_copy_do_write_done, bdev_io);
10497 :
10498 3 : if (rc == -ENOMEM) {
10499 0 : bdev_queue_io_wait_with_cb(bdev_io, bdev_copy_do_write);
10500 3 : } else if (rc != 0) {
10501 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
10502 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
10503 : }
10504 3 : }
10505 :
10506 : static void
10507 3 : bdev_copy_do_read_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
10508 : {
10509 3 : struct spdk_bdev_io *parent_io = cb_arg;
10510 :
10511 3 : spdk_bdev_free_io(bdev_io);
10512 :
10513 : /* Check return status of read */
10514 3 : if (!success) {
10515 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
10516 0 : parent_io->internal.cb(parent_io, false, parent_io->internal.caller_ctx);
10517 0 : return;
10518 : }
10519 :
10520 : /* Do write */
10521 3 : bdev_copy_do_write(parent_io);
10522 : }
10523 :
10524 : static void
10525 3 : bdev_copy_do_read(void *_bdev_io)
10526 : {
10527 3 : struct spdk_bdev_io *bdev_io = _bdev_io;
10528 : int rc;
10529 :
10530 : /* Read blocks */
10531 6 : rc = spdk_bdev_read_blocks_with_md(bdev_io->internal.desc,
10532 3 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
10533 3 : bdev_io->u.bdev.iovs[0].iov_base,
10534 : bdev_io->u.bdev.md_buf, bdev_io->u.bdev.copy.src_offset_blocks,
10535 : bdev_io->u.bdev.num_blocks, bdev_copy_do_read_done, bdev_io);
10536 :
10537 3 : if (rc == -ENOMEM) {
10538 0 : bdev_queue_io_wait_with_cb(bdev_io, bdev_copy_do_read);
10539 3 : } else if (rc != 0) {
10540 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
10541 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
10542 : }
10543 3 : }
10544 :
10545 : static void
10546 3 : bdev_copy_get_buf_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io, bool success)
10547 : {
10548 3 : if (!success) {
10549 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
10550 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
10551 0 : return;
10552 : }
10553 :
10554 3 : bdev_copy_do_read(bdev_io);
10555 : }
10556 :
10557 : int
10558 27 : spdk_bdev_copy_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
10559 : uint64_t dst_offset_blocks, uint64_t src_offset_blocks, uint64_t num_blocks,
10560 : spdk_bdev_io_completion_cb cb, void *cb_arg)
10561 : {
10562 27 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
10563 : struct spdk_bdev_io *bdev_io;
10564 27 : struct spdk_bdev_channel *channel = spdk_io_channel_get_ctx(ch);
10565 :
10566 27 : if (!desc->write) {
10567 0 : return -EBADF;
10568 : }
10569 :
10570 27 : if (!bdev_io_valid_blocks(bdev, dst_offset_blocks, num_blocks) ||
10571 27 : !bdev_io_valid_blocks(bdev, src_offset_blocks, num_blocks)) {
10572 0 : SPDK_DEBUGLOG(bdev,
10573 : "Invalid offset or number of blocks: dst %lu, src %lu, count %lu\n",
10574 : dst_offset_blocks, src_offset_blocks, num_blocks);
10575 0 : return -EINVAL;
10576 : }
10577 :
10578 27 : bdev_io = bdev_channel_get_io(channel);
10579 27 : if (!bdev_io) {
10580 0 : return -ENOMEM;
10581 : }
10582 :
10583 27 : bdev_io->internal.ch = channel;
10584 27 : bdev_io->internal.desc = desc;
10585 27 : bdev_io->type = SPDK_BDEV_IO_TYPE_COPY;
10586 :
10587 27 : bdev_io->u.bdev.offset_blocks = dst_offset_blocks;
10588 27 : bdev_io->u.bdev.copy.src_offset_blocks = src_offset_blocks;
10589 27 : bdev_io->u.bdev.num_blocks = num_blocks;
10590 27 : bdev_io->u.bdev.memory_domain = NULL;
10591 27 : bdev_io->u.bdev.memory_domain_ctx = NULL;
10592 27 : bdev_io->u.bdev.iovs = NULL;
10593 27 : bdev_io->u.bdev.iovcnt = 0;
10594 27 : bdev_io->u.bdev.md_buf = NULL;
10595 27 : bdev_io->u.bdev.accel_sequence = NULL;
10596 27 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
10597 :
10598 27 : if (dst_offset_blocks == src_offset_blocks || num_blocks == 0) {
10599 0 : spdk_thread_send_msg(spdk_get_thread(), bdev_io_complete_cb, bdev_io);
10600 0 : return 0;
10601 : }
10602 :
10603 :
10604 : /* If the copy size is large and should be split, use the generic split logic
10605 : * regardless of whether SPDK_BDEV_IO_TYPE_COPY is supported or not.
10606 : *
10607 : * Then, send the copy request if SPDK_BDEV_IO_TYPE_COPY is supported or
10608 : * emulate it using regular read and write requests otherwise.
10609 : */
10610 27 : if (spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COPY) ||
10611 : bdev_io->internal.f.split) {
10612 24 : bdev_io_submit(bdev_io);
10613 24 : return 0;
10614 : }
10615 :
10616 3 : spdk_bdev_io_get_buf(bdev_io, bdev_copy_get_buf_cb, num_blocks * spdk_bdev_get_block_size(bdev));
10617 :
10618 3 : return 0;
10619 : }
10620 :
10621 3 : SPDK_LOG_REGISTER_COMPONENT(bdev)
10622 :
10623 : static void
10624 0 : bdev_trace(void)
10625 : {
10626 0 : struct spdk_trace_tpoint_opts opts[] = {
10627 : {
10628 : "BDEV_IO_START", TRACE_BDEV_IO_START,
10629 : OWNER_TYPE_BDEV, OBJECT_BDEV_IO, 1,
10630 : {
10631 : { "type", SPDK_TRACE_ARG_TYPE_INT, 8 },
10632 : { "ctx", SPDK_TRACE_ARG_TYPE_PTR, 8 },
10633 : { "offset", SPDK_TRACE_ARG_TYPE_INT, 8 },
10634 : { "qd", SPDK_TRACE_ARG_TYPE_INT, 4 }
10635 : }
10636 : },
10637 : {
10638 : "BDEV_IO_DONE", TRACE_BDEV_IO_DONE,
10639 : OWNER_TYPE_BDEV, OBJECT_BDEV_IO, 0,
10640 : {
10641 : { "ctx", SPDK_TRACE_ARG_TYPE_PTR, 8 },
10642 : { "qd", SPDK_TRACE_ARG_TYPE_INT, 4 }
10643 : }
10644 : },
10645 : {
10646 : "BDEV_IOCH_CREATE", TRACE_BDEV_IOCH_CREATE,
10647 : OWNER_TYPE_BDEV, OBJECT_NONE, 0,
10648 : {
10649 : { "tid", SPDK_TRACE_ARG_TYPE_INT, 8 }
10650 : }
10651 : },
10652 : {
10653 : "BDEV_IOCH_DESTROY", TRACE_BDEV_IOCH_DESTROY,
10654 : OWNER_TYPE_BDEV, OBJECT_NONE, 0,
10655 : {
10656 : { "tid", SPDK_TRACE_ARG_TYPE_INT, 8 }
10657 : }
10658 : },
10659 : };
10660 :
10661 :
10662 0 : spdk_trace_register_owner_type(OWNER_TYPE_BDEV, 'b');
10663 0 : spdk_trace_register_object(OBJECT_BDEV_IO, 'i');
10664 0 : spdk_trace_register_description_ext(opts, SPDK_COUNTOF(opts));
10665 0 : spdk_trace_tpoint_register_relation(TRACE_BDEV_NVME_IO_START, OBJECT_BDEV_IO, 0);
10666 0 : spdk_trace_tpoint_register_relation(TRACE_BDEV_NVME_IO_DONE, OBJECT_BDEV_IO, 0);
10667 0 : spdk_trace_tpoint_register_relation(TRACE_BLOB_REQ_SET_START, OBJECT_BDEV_IO, 0);
10668 0 : spdk_trace_tpoint_register_relation(TRACE_BLOB_REQ_SET_COMPLETE, OBJECT_BDEV_IO, 0);
10669 0 : spdk_trace_tpoint_register_relation(TRACE_BDEV_RAID_IO_START, OBJECT_BDEV_IO, 0);
10670 0 : spdk_trace_tpoint_register_relation(TRACE_BDEV_RAID_IO_DONE, OBJECT_BDEV_IO, 0);
10671 0 : }
10672 3 : SPDK_TRACE_REGISTER_FN(bdev_trace, "bdev", TRACE_GROUP_BDEV)
|
