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