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
86 549 : bdev_name_cmp(struct spdk_bdev_name *name1, struct spdk_bdev_name *name2)
87 : {
88 549 : return strcmp(name1->name, name2->name);
89 : }
90 :
91 2065 : RB_GENERATE_STATIC(bdev_name_tree, spdk_bdev_name, node, bdev_name_cmp);
92 :
93 : struct spdk_bdev_mgr {
94 : struct spdk_mempool *bdev_io_pool;
95 :
96 : void *zero_buffer;
97 :
98 : TAILQ_HEAD(bdev_module_list, spdk_bdev_module) bdev_modules;
99 :
100 : struct spdk_bdev_list bdevs;
101 : struct bdev_name_tree bdev_names;
102 :
103 : bool init_complete;
104 : bool module_init_complete;
105 :
106 : struct spdk_spinlock spinlock;
107 :
108 : TAILQ_HEAD(, spdk_bdev_open_async_ctx) async_bdev_opens;
109 :
110 : #ifdef SPDK_CONFIG_VTUNE
111 : __itt_domain *domain;
112 : #endif
113 : };
114 :
115 : static struct spdk_bdev_mgr g_bdev_mgr = {
116 : .bdev_modules = TAILQ_HEAD_INITIALIZER(g_bdev_mgr.bdev_modules),
117 : .bdevs = TAILQ_HEAD_INITIALIZER(g_bdev_mgr.bdevs),
118 : .bdev_names = RB_INITIALIZER(g_bdev_mgr.bdev_names),
119 : .init_complete = false,
120 : .module_init_complete = false,
121 : .async_bdev_opens = TAILQ_HEAD_INITIALIZER(g_bdev_mgr.async_bdev_opens),
122 : };
123 :
124 : static void
125 : __attribute__((constructor))
126 3 : _bdev_init(void)
127 : {
128 3 : spdk_spin_init(&g_bdev_mgr.spinlock);
129 3 : }
130 :
131 : typedef void (*lock_range_cb)(struct lba_range *range, void *ctx, int status);
132 :
133 : typedef void (*bdev_copy_bounce_buffer_cpl)(void *ctx, int rc);
134 :
135 : struct lba_range {
136 : struct spdk_bdev *bdev;
137 : uint64_t offset;
138 : uint64_t length;
139 : bool quiesce;
140 : void *locked_ctx;
141 : struct spdk_thread *owner_thread;
142 : struct spdk_bdev_channel *owner_ch;
143 : TAILQ_ENTRY(lba_range) tailq;
144 : TAILQ_ENTRY(lba_range) tailq_module;
145 : };
146 :
147 : static struct spdk_bdev_opts g_bdev_opts = {
148 : .bdev_io_pool_size = SPDK_BDEV_IO_POOL_SIZE,
149 : .bdev_io_cache_size = SPDK_BDEV_IO_CACHE_SIZE,
150 : .bdev_auto_examine = SPDK_BDEV_AUTO_EXAMINE,
151 : .iobuf_small_cache_size = BUF_SMALL_CACHE_SIZE,
152 : .iobuf_large_cache_size = BUF_LARGE_CACHE_SIZE,
153 : };
154 :
155 : static spdk_bdev_init_cb g_init_cb_fn = NULL;
156 : static void *g_init_cb_arg = NULL;
157 :
158 : static spdk_bdev_fini_cb g_fini_cb_fn = NULL;
159 : static void *g_fini_cb_arg = NULL;
160 : static struct spdk_thread *g_fini_thread = NULL;
161 :
162 : struct spdk_bdev_qos_limit {
163 : /** IOs or bytes allowed per second (i.e., 1s). */
164 : uint64_t limit;
165 :
166 : /** Remaining IOs or bytes allowed in current timeslice (e.g., 1ms).
167 : * For remaining bytes, allowed to run negative if an I/O is submitted when
168 : * some bytes are remaining, but the I/O is bigger than that amount. The
169 : * excess will be deducted from the next timeslice.
170 : */
171 : int64_t remaining_this_timeslice;
172 :
173 : /** Minimum allowed IOs or bytes to be issued in one timeslice (e.g., 1ms). */
174 : uint32_t min_per_timeslice;
175 :
176 : /** Maximum allowed IOs or bytes to be issued in one timeslice (e.g., 1ms). */
177 : uint32_t max_per_timeslice;
178 :
179 : /** Function to check whether to queue the IO.
180 : * If The IO is allowed to pass, the quota will be reduced correspondingly.
181 : */
182 : bool (*queue_io)(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io);
183 :
184 : /** Function to rewind the quota once the IO was allowed to be sent by this
185 : * limit but queued due to one of the further limits.
186 : */
187 : void (*rewind_quota)(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io);
188 : };
189 :
190 : struct spdk_bdev_qos {
191 : /** Types of structure of rate limits. */
192 : struct spdk_bdev_qos_limit rate_limits[SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES];
193 :
194 : /** The channel that all I/O are funneled through. */
195 : struct spdk_bdev_channel *ch;
196 :
197 : /** The thread on which the poller is running. */
198 : struct spdk_thread *thread;
199 :
200 : /** Size of a timeslice in tsc ticks. */
201 : uint64_t timeslice_size;
202 :
203 : /** Timestamp of start of last timeslice. */
204 : uint64_t last_timeslice;
205 :
206 : /** Poller that processes queued I/O commands each time slice. */
207 : struct spdk_poller *poller;
208 : };
209 :
210 : struct spdk_bdev_mgmt_channel {
211 : /*
212 : * Each thread keeps a cache of bdev_io - this allows
213 : * bdev threads which are *not* DPDK threads to still
214 : * benefit from a per-thread bdev_io cache. Without
215 : * this, non-DPDK threads fetching from the mempool
216 : * incur a cmpxchg on get and put.
217 : */
218 : bdev_io_stailq_t per_thread_cache;
219 : uint32_t per_thread_cache_count;
220 : uint32_t bdev_io_cache_size;
221 :
222 : struct spdk_iobuf_channel iobuf;
223 :
224 : TAILQ_HEAD(, spdk_bdev_shared_resource) shared_resources;
225 : TAILQ_HEAD(, spdk_bdev_io_wait_entry) io_wait_queue;
226 : };
227 :
228 : /*
229 : * Per-module (or per-io_device) data. Multiple bdevs built on the same io_device
230 : * will queue here their IO that awaits retry. It makes it possible to retry sending
231 : * IO to one bdev after IO from other bdev completes.
232 : */
233 : struct spdk_bdev_shared_resource {
234 : /* The bdev management channel */
235 : struct spdk_bdev_mgmt_channel *mgmt_ch;
236 :
237 : /*
238 : * Count of I/O submitted to bdev module and waiting for completion.
239 : * Incremented before submit_request() is called on an spdk_bdev_io.
240 : */
241 : uint64_t io_outstanding;
242 :
243 : /*
244 : * Queue of IO awaiting retry because of a previous NOMEM status returned
245 : * on this channel.
246 : */
247 : bdev_io_tailq_t nomem_io;
248 :
249 : /*
250 : * Threshold which io_outstanding must drop to before retrying nomem_io.
251 : */
252 : uint64_t nomem_threshold;
253 :
254 : /* I/O channel allocated by a bdev module */
255 : struct spdk_io_channel *shared_ch;
256 :
257 : struct spdk_poller *nomem_poller;
258 :
259 : /* Refcount of bdev channels using this resource */
260 : uint32_t ref;
261 :
262 : TAILQ_ENTRY(spdk_bdev_shared_resource) link;
263 : };
264 :
265 : #define BDEV_CH_RESET_IN_PROGRESS (1 << 0)
266 : #define BDEV_CH_QOS_ENABLED (1 << 1)
267 :
268 : struct spdk_bdev_channel {
269 : struct spdk_bdev *bdev;
270 :
271 : /* The channel for the underlying device */
272 : struct spdk_io_channel *channel;
273 :
274 : /* Accel channel */
275 : struct spdk_io_channel *accel_channel;
276 :
277 : /* Per io_device per thread data */
278 : struct spdk_bdev_shared_resource *shared_resource;
279 :
280 : struct spdk_bdev_io_stat *stat;
281 :
282 : /*
283 : * Count of I/O submitted to the underlying dev module through this channel
284 : * and waiting for completion.
285 : */
286 : uint64_t io_outstanding;
287 :
288 : /*
289 : * List of all submitted I/Os including I/O that are generated via splitting.
290 : */
291 : bdev_io_tailq_t io_submitted;
292 :
293 : /*
294 : * List of spdk_bdev_io that are currently queued because they write to a locked
295 : * LBA range.
296 : */
297 : bdev_io_tailq_t io_locked;
298 :
299 : /* List of I/Os with accel sequence being currently executed */
300 : bdev_io_tailq_t io_accel_exec;
301 :
302 : /* List of I/Os doing memory domain pull/push */
303 : bdev_io_tailq_t io_memory_domain;
304 :
305 : uint32_t flags;
306 :
307 : /* Counts number of bdev_io in the io_submitted TAILQ */
308 : uint16_t queue_depth;
309 :
310 : uint16_t trace_id;
311 :
312 : struct spdk_histogram_data *histogram;
313 :
314 : #ifdef SPDK_CONFIG_VTUNE
315 : uint64_t start_tsc;
316 : uint64_t interval_tsc;
317 : __itt_string_handle *handle;
318 : struct spdk_bdev_io_stat *prev_stat;
319 : #endif
320 :
321 : bdev_io_tailq_t queued_resets;
322 :
323 : lba_range_tailq_t locked_ranges;
324 :
325 : /** List of I/Os queued by QoS. */
326 : bdev_io_tailq_t qos_queued_io;
327 : };
328 :
329 : struct media_event_entry {
330 : struct spdk_bdev_media_event event;
331 : TAILQ_ENTRY(media_event_entry) tailq;
332 : };
333 :
334 : #define MEDIA_EVENT_POOL_SIZE 64
335 :
336 : struct spdk_bdev_desc {
337 : struct spdk_bdev *bdev;
338 : bool write;
339 : bool memory_domains_supported;
340 : bool accel_sequence_supported[SPDK_BDEV_NUM_IO_TYPES];
341 : struct spdk_bdev_open_opts opts;
342 : struct spdk_thread *thread;
343 : struct {
344 : spdk_bdev_event_cb_t event_fn;
345 : void *ctx;
346 : } callback;
347 : bool closed;
348 : struct spdk_spinlock spinlock;
349 : uint32_t refs;
350 : TAILQ_HEAD(, media_event_entry) pending_media_events;
351 : TAILQ_HEAD(, media_event_entry) free_media_events;
352 : struct media_event_entry *media_events_buffer;
353 : TAILQ_ENTRY(spdk_bdev_desc) link;
354 :
355 : uint64_t timeout_in_sec;
356 : spdk_bdev_io_timeout_cb cb_fn;
357 : void *cb_arg;
358 : struct spdk_poller *io_timeout_poller;
359 : struct spdk_bdev_module_claim *claim;
360 : };
361 :
362 : struct spdk_bdev_iostat_ctx {
363 : struct spdk_bdev_io_stat *stat;
364 : enum spdk_bdev_reset_stat_mode reset_mode;
365 : spdk_bdev_get_device_stat_cb cb;
366 : void *cb_arg;
367 : };
368 :
369 : struct set_qos_limit_ctx {
370 : void (*cb_fn)(void *cb_arg, int status);
371 : void *cb_arg;
372 : struct spdk_bdev *bdev;
373 : };
374 :
375 : struct spdk_bdev_channel_iter {
376 : spdk_bdev_for_each_channel_msg fn;
377 : spdk_bdev_for_each_channel_done cpl;
378 : struct spdk_io_channel_iter *i;
379 : void *ctx;
380 : };
381 :
382 : struct spdk_bdev_io_error_stat {
383 : uint32_t error_status[-SPDK_MIN_BDEV_IO_STATUS];
384 : };
385 :
386 : enum bdev_io_retry_state {
387 : BDEV_IO_RETRY_STATE_INVALID,
388 : BDEV_IO_RETRY_STATE_PULL,
389 : BDEV_IO_RETRY_STATE_PULL_MD,
390 : BDEV_IO_RETRY_STATE_SUBMIT,
391 : BDEV_IO_RETRY_STATE_PUSH,
392 : BDEV_IO_RETRY_STATE_PUSH_MD,
393 : };
394 :
395 : #define __bdev_to_io_dev(bdev) (((char *)bdev) + 1)
396 : #define __bdev_from_io_dev(io_dev) ((struct spdk_bdev *)(((char *)io_dev) - 1))
397 : #define __io_ch_to_bdev_ch(io_ch) ((struct spdk_bdev_channel *)spdk_io_channel_get_ctx(io_ch))
398 : #define __io_ch_to_bdev_mgmt_ch(io_ch) ((struct spdk_bdev_mgmt_channel *)spdk_io_channel_get_ctx(io_ch))
399 :
400 : static inline void bdev_io_complete(void *ctx);
401 : static inline void bdev_io_complete_unsubmitted(struct spdk_bdev_io *bdev_io);
402 : static void bdev_io_push_bounce_md_buf(struct spdk_bdev_io *bdev_io);
403 : static void bdev_io_push_bounce_data(struct spdk_bdev_io *bdev_io);
404 :
405 : static void bdev_write_zero_buffer_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg);
406 : static int bdev_write_zero_buffer(struct spdk_bdev_io *bdev_io);
407 :
408 : static void bdev_enable_qos_msg(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
409 : struct spdk_io_channel *ch, void *_ctx);
410 : static void bdev_enable_qos_done(struct spdk_bdev *bdev, void *_ctx, int status);
411 :
412 : static int bdev_readv_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
413 : struct iovec *iov, int iovcnt, void *md_buf, uint64_t offset_blocks,
414 : uint64_t num_blocks,
415 : struct spdk_memory_domain *domain, void *domain_ctx,
416 : struct spdk_accel_sequence *seq, uint32_t dif_check_flags,
417 : spdk_bdev_io_completion_cb cb, void *cb_arg);
418 : static int bdev_writev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
419 : struct iovec *iov, int iovcnt, void *md_buf,
420 : uint64_t offset_blocks, uint64_t num_blocks,
421 : struct spdk_memory_domain *domain, void *domain_ctx,
422 : struct spdk_accel_sequence *seq, uint32_t dif_check_flags,
423 : uint32_t nvme_cdw12_raw, uint32_t nvme_cdw13_raw,
424 : spdk_bdev_io_completion_cb cb, void *cb_arg);
425 :
426 : static int bdev_lock_lba_range(struct spdk_bdev_desc *desc, struct spdk_io_channel *_ch,
427 : uint64_t offset, uint64_t length,
428 : lock_range_cb cb_fn, void *cb_arg);
429 :
430 : static int bdev_unlock_lba_range(struct spdk_bdev_desc *desc, struct spdk_io_channel *_ch,
431 : uint64_t offset, uint64_t length,
432 : lock_range_cb cb_fn, void *cb_arg);
433 :
434 : static bool bdev_abort_queued_io(bdev_io_tailq_t *queue, struct spdk_bdev_io *bio_to_abort);
435 : static bool bdev_abort_buf_io(struct spdk_bdev_mgmt_channel *ch, struct spdk_bdev_io *bio_to_abort);
436 :
437 : static bool claim_type_is_v2(enum spdk_bdev_claim_type type);
438 : static void bdev_desc_release_claims(struct spdk_bdev_desc *desc);
439 : static void claim_reset(struct spdk_bdev *bdev);
440 :
441 : static void bdev_ch_retry_io(struct spdk_bdev_channel *bdev_ch);
442 :
443 : static bool bdev_io_should_split(struct spdk_bdev_io *bdev_io);
444 :
445 : #define bdev_get_ext_io_opt(opts, field, defval) \
446 : ((opts) != NULL ? SPDK_GET_FIELD(opts, field, defval) : (defval))
447 :
448 : static inline void
449 669 : bdev_ch_add_to_io_submitted(struct spdk_bdev_io *bdev_io)
450 : {
451 669 : TAILQ_INSERT_TAIL(&bdev_io->internal.ch->io_submitted, bdev_io, internal.ch_link);
452 669 : bdev_io->internal.ch->queue_depth++;
453 669 : }
454 :
455 : static inline void
456 669 : bdev_ch_remove_from_io_submitted(struct spdk_bdev_io *bdev_io)
457 : {
458 669 : TAILQ_REMOVE(&bdev_io->internal.ch->io_submitted, bdev_io, internal.ch_link);
459 669 : bdev_io->internal.ch->queue_depth--;
460 669 : }
461 :
462 : void
463 14 : spdk_bdev_get_opts(struct spdk_bdev_opts *opts, size_t opts_size)
464 : {
465 14 : if (!opts) {
466 0 : SPDK_ERRLOG("opts should not be NULL\n");
467 0 : return;
468 : }
469 :
470 14 : if (!opts_size) {
471 0 : SPDK_ERRLOG("opts_size should not be zero value\n");
472 0 : return;
473 : }
474 :
475 14 : opts->opts_size = opts_size;
476 :
477 : #define SET_FIELD(field) \
478 : if (offsetof(struct spdk_bdev_opts, field) + sizeof(opts->field) <= opts_size) { \
479 : opts->field = g_bdev_opts.field; \
480 : } \
481 :
482 14 : SET_FIELD(bdev_io_pool_size);
483 14 : SET_FIELD(bdev_io_cache_size);
484 14 : SET_FIELD(bdev_auto_examine);
485 14 : SET_FIELD(iobuf_small_cache_size);
486 14 : SET_FIELD(iobuf_large_cache_size);
487 :
488 : /* Do not remove this statement, you should always update this statement when you adding a new field,
489 : * and do not forget to add the SET_FIELD statement for your added field. */
490 : SPDK_STATIC_ASSERT(sizeof(struct spdk_bdev_opts) == 32, "Incorrect size");
491 :
492 : #undef SET_FIELD
493 14 : }
494 :
495 : int
496 15 : spdk_bdev_set_opts(struct spdk_bdev_opts *opts)
497 : {
498 : uint32_t min_pool_size;
499 :
500 15 : if (!opts) {
501 0 : SPDK_ERRLOG("opts cannot be NULL\n");
502 0 : return -1;
503 : }
504 :
505 15 : if (!opts->opts_size) {
506 1 : SPDK_ERRLOG("opts_size inside opts cannot be zero value\n");
507 1 : return -1;
508 : }
509 :
510 : /*
511 : * Add 1 to the thread count to account for the extra mgmt_ch that gets created during subsystem
512 : * initialization. A second mgmt_ch will be created on the same thread when the application starts
513 : * but before the deferred put_io_channel event is executed for the first mgmt_ch.
514 : */
515 14 : min_pool_size = opts->bdev_io_cache_size * (spdk_thread_get_count() + 1);
516 14 : if (opts->bdev_io_pool_size < min_pool_size) {
517 0 : SPDK_ERRLOG("bdev_io_pool_size %" PRIu32 " is not compatible with bdev_io_cache_size %" PRIu32
518 : " and %" PRIu32 " threads\n", opts->bdev_io_pool_size, opts->bdev_io_cache_size,
519 : spdk_thread_get_count());
520 0 : SPDK_ERRLOG("bdev_io_pool_size must be at least %" PRIu32 "\n", min_pool_size);
521 0 : return -1;
522 : }
523 :
524 : #define SET_FIELD(field) \
525 : if (offsetof(struct spdk_bdev_opts, field) + sizeof(opts->field) <= opts->opts_size) { \
526 : g_bdev_opts.field = opts->field; \
527 : } \
528 :
529 14 : SET_FIELD(bdev_io_pool_size);
530 14 : SET_FIELD(bdev_io_cache_size);
531 14 : SET_FIELD(bdev_auto_examine);
532 14 : SET_FIELD(iobuf_small_cache_size);
533 14 : SET_FIELD(iobuf_large_cache_size);
534 :
535 14 : g_bdev_opts.opts_size = opts->opts_size;
536 :
537 : #undef SET_FIELD
538 :
539 14 : return 0;
540 15 : }
541 :
542 : static struct spdk_bdev *
543 146 : bdev_get_by_name(const char *bdev_name)
544 : {
545 : struct spdk_bdev_name find;
546 : struct spdk_bdev_name *res;
547 :
548 146 : find.name = (char *)bdev_name;
549 146 : res = RB_FIND(bdev_name_tree, &g_bdev_mgr.bdev_names, &find);
550 146 : if (res != NULL) {
551 139 : return res->bdev;
552 : }
553 :
554 7 : return NULL;
555 146 : }
556 :
557 : struct spdk_bdev *
558 16 : spdk_bdev_get_by_name(const char *bdev_name)
559 : {
560 : struct spdk_bdev *bdev;
561 :
562 16 : spdk_spin_lock(&g_bdev_mgr.spinlock);
563 16 : bdev = bdev_get_by_name(bdev_name);
564 16 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
565 :
566 16 : return bdev;
567 : }
568 :
569 : struct bdev_io_status_string {
570 : enum spdk_bdev_io_status status;
571 : const char *str;
572 : };
573 :
574 : static const struct bdev_io_status_string bdev_io_status_strings[] = {
575 : { SPDK_BDEV_IO_STATUS_AIO_ERROR, "aio_error" },
576 : { SPDK_BDEV_IO_STATUS_ABORTED, "aborted" },
577 : { SPDK_BDEV_IO_STATUS_FIRST_FUSED_FAILED, "first_fused_failed" },
578 : { SPDK_BDEV_IO_STATUS_MISCOMPARE, "miscompare" },
579 : { SPDK_BDEV_IO_STATUS_NOMEM, "nomem" },
580 : { SPDK_BDEV_IO_STATUS_SCSI_ERROR, "scsi_error" },
581 : { SPDK_BDEV_IO_STATUS_NVME_ERROR, "nvme_error" },
582 : { SPDK_BDEV_IO_STATUS_FAILED, "failed" },
583 : { SPDK_BDEV_IO_STATUS_PENDING, "pending" },
584 : { SPDK_BDEV_IO_STATUS_SUCCESS, "success" },
585 : };
586 :
587 : static const char *
588 0 : bdev_io_status_get_string(enum spdk_bdev_io_status status)
589 : {
590 : uint32_t i;
591 :
592 0 : for (i = 0; i < SPDK_COUNTOF(bdev_io_status_strings); i++) {
593 0 : if (bdev_io_status_strings[i].status == status) {
594 0 : return bdev_io_status_strings[i].str;
595 : }
596 0 : }
597 :
598 0 : return "reserved";
599 0 : }
600 :
601 : struct spdk_bdev_wait_for_examine_ctx {
602 : struct spdk_poller *poller;
603 : spdk_bdev_wait_for_examine_cb cb_fn;
604 : void *cb_arg;
605 : };
606 :
607 : static bool bdev_module_all_actions_completed(void);
608 :
609 : static int
610 192 : bdev_wait_for_examine_cb(void *arg)
611 : {
612 192 : struct spdk_bdev_wait_for_examine_ctx *ctx = arg;
613 :
614 192 : if (!bdev_module_all_actions_completed()) {
615 0 : return SPDK_POLLER_IDLE;
616 : }
617 :
618 192 : spdk_poller_unregister(&ctx->poller);
619 192 : ctx->cb_fn(ctx->cb_arg);
620 192 : free(ctx);
621 :
622 192 : return SPDK_POLLER_BUSY;
623 192 : }
624 :
625 : int
626 192 : spdk_bdev_wait_for_examine(spdk_bdev_wait_for_examine_cb cb_fn, void *cb_arg)
627 : {
628 : struct spdk_bdev_wait_for_examine_ctx *ctx;
629 :
630 192 : ctx = calloc(1, sizeof(*ctx));
631 192 : if (ctx == NULL) {
632 0 : return -ENOMEM;
633 : }
634 192 : ctx->cb_fn = cb_fn;
635 192 : ctx->cb_arg = cb_arg;
636 192 : ctx->poller = SPDK_POLLER_REGISTER(bdev_wait_for_examine_cb, ctx, 0);
637 :
638 192 : return 0;
639 192 : }
640 :
641 : struct spdk_bdev_examine_item {
642 : char *name;
643 : TAILQ_ENTRY(spdk_bdev_examine_item) link;
644 : };
645 :
646 : TAILQ_HEAD(spdk_bdev_examine_allowlist, spdk_bdev_examine_item);
647 :
648 : struct spdk_bdev_examine_allowlist g_bdev_examine_allowlist = TAILQ_HEAD_INITIALIZER(
649 : g_bdev_examine_allowlist);
650 :
651 : static inline bool
652 6 : bdev_examine_allowlist_check(const char *name)
653 : {
654 : struct spdk_bdev_examine_item *item;
655 6 : TAILQ_FOREACH(item, &g_bdev_examine_allowlist, link) {
656 0 : if (strcmp(name, item->name) == 0) {
657 0 : return true;
658 : }
659 0 : }
660 6 : return false;
661 6 : }
662 :
663 : static inline void
664 65 : bdev_examine_allowlist_free(void)
665 : {
666 : struct spdk_bdev_examine_item *item;
667 65 : while (!TAILQ_EMPTY(&g_bdev_examine_allowlist)) {
668 0 : item = TAILQ_FIRST(&g_bdev_examine_allowlist);
669 0 : TAILQ_REMOVE(&g_bdev_examine_allowlist, item, link);
670 0 : free(item->name);
671 0 : free(item);
672 : }
673 65 : }
674 :
675 : static inline bool
676 3 : bdev_in_examine_allowlist(struct spdk_bdev *bdev)
677 : {
678 : struct spdk_bdev_alias *tmp;
679 3 : if (bdev_examine_allowlist_check(bdev->name)) {
680 0 : return true;
681 : }
682 6 : TAILQ_FOREACH(tmp, &bdev->aliases, tailq) {
683 3 : if (bdev_examine_allowlist_check(tmp->alias.name)) {
684 0 : return true;
685 : }
686 3 : }
687 3 : return false;
688 3 : }
689 :
690 : static inline bool
691 125 : bdev_ok_to_examine(struct spdk_bdev *bdev)
692 : {
693 : /* Some bdevs may not support the READ command.
694 : * Do not try to examine them.
695 : */
696 125 : if (!spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_READ)) {
697 0 : return false;
698 : }
699 :
700 125 : if (g_bdev_opts.bdev_auto_examine) {
701 122 : return true;
702 : } else {
703 3 : return bdev_in_examine_allowlist(bdev);
704 : }
705 125 : }
706 :
707 : static void
708 125 : bdev_examine(struct spdk_bdev *bdev)
709 : {
710 : struct spdk_bdev_module *module;
711 : struct spdk_bdev_module_claim *claim, *tmpclaim;
712 : uint32_t action;
713 :
714 125 : if (!bdev_ok_to_examine(bdev)) {
715 3 : return;
716 : }
717 :
718 499 : TAILQ_FOREACH(module, &g_bdev_mgr.bdev_modules, internal.tailq) {
719 377 : if (module->examine_config) {
720 255 : spdk_spin_lock(&module->internal.spinlock);
721 255 : action = module->internal.action_in_progress;
722 255 : module->internal.action_in_progress++;
723 255 : spdk_spin_unlock(&module->internal.spinlock);
724 255 : module->examine_config(bdev);
725 255 : if (action != module->internal.action_in_progress) {
726 0 : SPDK_ERRLOG("examine_config for module %s did not call "
727 : "spdk_bdev_module_examine_done()\n", module->name);
728 0 : }
729 255 : }
730 377 : }
731 :
732 122 : spdk_spin_lock(&bdev->internal.spinlock);
733 :
734 122 : switch (bdev->internal.claim_type) {
735 : case SPDK_BDEV_CLAIM_NONE:
736 : /* Examine by all bdev modules */
737 474 : TAILQ_FOREACH(module, &g_bdev_mgr.bdev_modules, internal.tailq) {
738 357 : if (module->examine_disk) {
739 231 : spdk_spin_lock(&module->internal.spinlock);
740 231 : module->internal.action_in_progress++;
741 231 : spdk_spin_unlock(&module->internal.spinlock);
742 231 : spdk_spin_unlock(&bdev->internal.spinlock);
743 231 : module->examine_disk(bdev);
744 231 : spdk_spin_lock(&bdev->internal.spinlock);
745 231 : }
746 357 : }
747 117 : break;
748 : case SPDK_BDEV_CLAIM_EXCL_WRITE:
749 : /* Examine by the one bdev module with a v1 claim */
750 1 : module = bdev->internal.claim.v1.module;
751 1 : if (module->examine_disk) {
752 1 : spdk_spin_lock(&module->internal.spinlock);
753 1 : module->internal.action_in_progress++;
754 1 : spdk_spin_unlock(&module->internal.spinlock);
755 1 : spdk_spin_unlock(&bdev->internal.spinlock);
756 1 : module->examine_disk(bdev);
757 1 : return;
758 : }
759 0 : break;
760 : default:
761 : /* Examine by all bdev modules with a v2 claim */
762 4 : assert(claim_type_is_v2(bdev->internal.claim_type));
763 : /*
764 : * Removal of tailq nodes while iterating can cause the iteration to jump out of the
765 : * list, perhaps accessing freed memory. Without protection, this could happen
766 : * while the lock is dropped during the examine callback.
767 : */
768 4 : bdev->internal.examine_in_progress++;
769 :
770 9 : TAILQ_FOREACH(claim, &bdev->internal.claim.v2.claims, link) {
771 5 : module = claim->module;
772 :
773 5 : if (module == NULL) {
774 : /* This is a vestigial claim, held by examine_count */
775 0 : continue;
776 : }
777 :
778 5 : if (module->examine_disk == NULL) {
779 0 : continue;
780 : }
781 :
782 5 : spdk_spin_lock(&module->internal.spinlock);
783 5 : module->internal.action_in_progress++;
784 5 : spdk_spin_unlock(&module->internal.spinlock);
785 :
786 : /* Call examine_disk without holding internal.spinlock. */
787 5 : spdk_spin_unlock(&bdev->internal.spinlock);
788 5 : module->examine_disk(bdev);
789 5 : spdk_spin_lock(&bdev->internal.spinlock);
790 5 : }
791 :
792 4 : assert(bdev->internal.examine_in_progress > 0);
793 4 : bdev->internal.examine_in_progress--;
794 4 : if (bdev->internal.examine_in_progress == 0) {
795 : /* Remove any claims that were released during examine_disk */
796 9 : TAILQ_FOREACH_SAFE(claim, &bdev->internal.claim.v2.claims, link, tmpclaim) {
797 5 : if (claim->desc != NULL) {
798 5 : continue;
799 : }
800 :
801 0 : TAILQ_REMOVE(&bdev->internal.claim.v2.claims, claim, link);
802 0 : free(claim);
803 0 : }
804 4 : if (TAILQ_EMPTY(&bdev->internal.claim.v2.claims)) {
805 0 : claim_reset(bdev);
806 0 : }
807 4 : }
808 4 : }
809 :
810 121 : spdk_spin_unlock(&bdev->internal.spinlock);
811 125 : }
812 :
813 : int
814 1 : spdk_bdev_examine(const char *name)
815 : {
816 : struct spdk_bdev *bdev;
817 : struct spdk_bdev_examine_item *item;
818 1 : struct spdk_thread *thread = spdk_get_thread();
819 :
820 1 : if (spdk_unlikely(!spdk_thread_is_app_thread(thread))) {
821 1 : SPDK_ERRLOG("Cannot examine bdev %s on thread %p (%s)\n", name, thread,
822 : thread ? spdk_thread_get_name(thread) : "null");
823 1 : return -EINVAL;
824 : }
825 :
826 0 : if (g_bdev_opts.bdev_auto_examine) {
827 0 : SPDK_ERRLOG("Manual examine is not allowed if auto examine is enabled\n");
828 0 : return -EINVAL;
829 : }
830 :
831 0 : if (bdev_examine_allowlist_check(name)) {
832 0 : SPDK_ERRLOG("Duplicate bdev name for manual examine: %s\n", name);
833 0 : return -EEXIST;
834 : }
835 :
836 0 : item = calloc(1, sizeof(*item));
837 0 : if (!item) {
838 0 : return -ENOMEM;
839 : }
840 0 : item->name = strdup(name);
841 0 : if (!item->name) {
842 0 : free(item);
843 0 : return -ENOMEM;
844 : }
845 0 : TAILQ_INSERT_TAIL(&g_bdev_examine_allowlist, item, link);
846 :
847 0 : bdev = spdk_bdev_get_by_name(name);
848 0 : if (bdev) {
849 0 : bdev_examine(bdev);
850 0 : }
851 0 : return 0;
852 1 : }
853 :
854 : static inline void
855 0 : bdev_examine_allowlist_config_json(struct spdk_json_write_ctx *w)
856 : {
857 : struct spdk_bdev_examine_item *item;
858 0 : TAILQ_FOREACH(item, &g_bdev_examine_allowlist, link) {
859 0 : spdk_json_write_object_begin(w);
860 0 : spdk_json_write_named_string(w, "method", "bdev_examine");
861 0 : spdk_json_write_named_object_begin(w, "params");
862 0 : spdk_json_write_named_string(w, "name", item->name);
863 0 : spdk_json_write_object_end(w);
864 0 : spdk_json_write_object_end(w);
865 0 : }
866 0 : }
867 :
868 : struct spdk_bdev *
869 1 : spdk_bdev_first(void)
870 : {
871 : struct spdk_bdev *bdev;
872 :
873 1 : bdev = TAILQ_FIRST(&g_bdev_mgr.bdevs);
874 1 : if (bdev) {
875 1 : SPDK_DEBUGLOG(bdev, "Starting bdev iteration at %s\n", bdev->name);
876 1 : }
877 :
878 1 : return bdev;
879 : }
880 :
881 : struct spdk_bdev *
882 8 : spdk_bdev_next(struct spdk_bdev *prev)
883 : {
884 : struct spdk_bdev *bdev;
885 :
886 8 : bdev = TAILQ_NEXT(prev, internal.link);
887 8 : if (bdev) {
888 7 : SPDK_DEBUGLOG(bdev, "Continuing bdev iteration at %s\n", bdev->name);
889 7 : }
890 :
891 8 : return bdev;
892 : }
893 :
894 : static struct spdk_bdev *
895 6 : _bdev_next_leaf(struct spdk_bdev *bdev)
896 : {
897 9 : while (bdev != NULL) {
898 8 : if (bdev->internal.claim_type == SPDK_BDEV_CLAIM_NONE) {
899 5 : return bdev;
900 : } else {
901 3 : bdev = TAILQ_NEXT(bdev, internal.link);
902 : }
903 : }
904 :
905 1 : return bdev;
906 6 : }
907 :
908 : struct spdk_bdev *
909 1 : spdk_bdev_first_leaf(void)
910 : {
911 : struct spdk_bdev *bdev;
912 :
913 1 : bdev = _bdev_next_leaf(TAILQ_FIRST(&g_bdev_mgr.bdevs));
914 :
915 1 : if (bdev) {
916 1 : SPDK_DEBUGLOG(bdev, "Starting bdev iteration at %s\n", bdev->name);
917 1 : }
918 :
919 1 : return bdev;
920 : }
921 :
922 : struct spdk_bdev *
923 5 : spdk_bdev_next_leaf(struct spdk_bdev *prev)
924 : {
925 : struct spdk_bdev *bdev;
926 :
927 5 : bdev = _bdev_next_leaf(TAILQ_NEXT(prev, internal.link));
928 :
929 5 : if (bdev) {
930 4 : SPDK_DEBUGLOG(bdev, "Continuing bdev iteration at %s\n", bdev->name);
931 4 : }
932 :
933 5 : return bdev;
934 : }
935 :
936 : static inline bool
937 816 : bdev_io_use_memory_domain(struct spdk_bdev_io *bdev_io)
938 : {
939 816 : return bdev_io->internal.f.has_memory_domain;
940 : }
941 :
942 : static inline bool
943 1553 : bdev_io_use_accel_sequence(struct spdk_bdev_io *bdev_io)
944 : {
945 1553 : return bdev_io->internal.f.has_accel_sequence;
946 : }
947 :
948 : static inline void
949 7 : bdev_queue_nomem_io_head(struct spdk_bdev_shared_resource *shared_resource,
950 : struct spdk_bdev_io *bdev_io, enum bdev_io_retry_state state)
951 : {
952 : /* Wait for some of the outstanding I/O to complete before we retry any of the nomem_io.
953 : * Normally we will wait for NOMEM_THRESHOLD_COUNT I/O to complete but for low queue depth
954 : * channels we will instead wait for half to complete.
955 : */
956 7 : shared_resource->nomem_threshold = spdk_max((int64_t)shared_resource->io_outstanding / 2,
957 : (int64_t)shared_resource->io_outstanding - NOMEM_THRESHOLD_COUNT);
958 :
959 7 : assert(state != BDEV_IO_RETRY_STATE_INVALID);
960 7 : bdev_io->internal.retry_state = state;
961 7 : TAILQ_INSERT_HEAD(&shared_resource->nomem_io, bdev_io, internal.link);
962 7 : }
963 :
964 : static inline void
965 43 : bdev_queue_nomem_io_tail(struct spdk_bdev_shared_resource *shared_resource,
966 : struct spdk_bdev_io *bdev_io, enum bdev_io_retry_state state)
967 : {
968 : /* We only queue IOs at the end of the nomem_io queue if they're submitted by the user while
969 : * the queue isn't empty, so we don't need to update the nomem_threshold here */
970 43 : assert(!TAILQ_EMPTY(&shared_resource->nomem_io));
971 :
972 43 : assert(state != BDEV_IO_RETRY_STATE_INVALID);
973 43 : bdev_io->internal.retry_state = state;
974 43 : TAILQ_INSERT_TAIL(&shared_resource->nomem_io, bdev_io, internal.link);
975 43 : }
976 :
977 : void
978 16 : spdk_bdev_io_set_buf(struct spdk_bdev_io *bdev_io, void *buf, size_t len)
979 : {
980 : struct iovec *iovs;
981 :
982 16 : if (bdev_io->u.bdev.iovs == NULL) {
983 3 : bdev_io->u.bdev.iovs = &bdev_io->iov;
984 3 : bdev_io->u.bdev.iovcnt = 1;
985 3 : }
986 :
987 16 : iovs = bdev_io->u.bdev.iovs;
988 :
989 16 : assert(iovs != NULL);
990 16 : assert(bdev_io->u.bdev.iovcnt >= 1);
991 :
992 16 : iovs[0].iov_base = buf;
993 16 : iovs[0].iov_len = len;
994 16 : }
995 :
996 : void
997 3 : spdk_bdev_io_set_md_buf(struct spdk_bdev_io *bdev_io, void *md_buf, size_t len)
998 : {
999 3 : assert((len / spdk_bdev_get_md_size(bdev_io->bdev)) >= bdev_io->u.bdev.num_blocks);
1000 3 : bdev_io->u.bdev.md_buf = md_buf;
1001 3 : }
1002 :
1003 : static bool
1004 167 : _is_buf_allocated(const struct iovec *iovs)
1005 : {
1006 167 : if (iovs == NULL) {
1007 6 : return false;
1008 : }
1009 :
1010 161 : return iovs[0].iov_base != NULL;
1011 167 : }
1012 :
1013 : static bool
1014 50 : _are_iovs_aligned(struct iovec *iovs, int iovcnt, uint32_t alignment)
1015 : {
1016 : int i;
1017 : uintptr_t iov_base;
1018 :
1019 50 : if (spdk_likely(alignment == 1)) {
1020 21 : return true;
1021 : }
1022 :
1023 36 : for (i = 0; i < iovcnt; i++) {
1024 29 : iov_base = (uintptr_t)iovs[i].iov_base;
1025 29 : if ((iov_base & (alignment - 1)) != 0) {
1026 22 : return false;
1027 : }
1028 7 : }
1029 :
1030 7 : return true;
1031 50 : }
1032 :
1033 : static inline bool
1034 856 : bdev_io_needs_sequence_exec(struct spdk_bdev_desc *desc, struct spdk_bdev_io *bdev_io)
1035 : {
1036 856 : if (!bdev_io_use_accel_sequence(bdev_io)) {
1037 856 : return false;
1038 : }
1039 :
1040 : /* For now, we don't allow splitting IOs with an accel sequence and will treat them as if
1041 : * bdev module didn't support accel sequences */
1042 0 : return !desc->accel_sequence_supported[bdev_io->type] || bdev_io->internal.f.split;
1043 856 : }
1044 :
1045 : static inline void
1046 592 : bdev_io_increment_outstanding(struct spdk_bdev_channel *bdev_ch,
1047 : struct spdk_bdev_shared_resource *shared_resource)
1048 : {
1049 592 : bdev_ch->io_outstanding++;
1050 592 : shared_resource->io_outstanding++;
1051 592 : }
1052 :
1053 : static inline void
1054 592 : bdev_io_decrement_outstanding(struct spdk_bdev_channel *bdev_ch,
1055 : struct spdk_bdev_shared_resource *shared_resource)
1056 : {
1057 592 : assert(bdev_ch->io_outstanding > 0);
1058 592 : assert(shared_resource->io_outstanding > 0);
1059 592 : bdev_ch->io_outstanding--;
1060 592 : shared_resource->io_outstanding--;
1061 592 : }
1062 :
1063 : static void
1064 0 : bdev_io_submit_sequence_cb(void *ctx, int status)
1065 : {
1066 0 : struct spdk_bdev_io *bdev_io = ctx;
1067 :
1068 0 : assert(bdev_io_use_accel_sequence(bdev_io));
1069 :
1070 0 : bdev_io->u.bdev.accel_sequence = NULL;
1071 0 : bdev_io->internal.f.has_accel_sequence = false;
1072 :
1073 0 : if (spdk_unlikely(status != 0)) {
1074 0 : SPDK_ERRLOG("Failed to execute accel sequence, status=%d\n", status);
1075 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
1076 0 : bdev_io_complete_unsubmitted(bdev_io);
1077 0 : return;
1078 : }
1079 :
1080 0 : bdev_io_submit(bdev_io);
1081 0 : }
1082 :
1083 : static void
1084 0 : bdev_io_exec_sequence_cb(void *ctx, int status)
1085 : {
1086 0 : struct spdk_bdev_io *bdev_io = ctx;
1087 0 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1088 :
1089 0 : TAILQ_REMOVE(&bdev_io->internal.ch->io_accel_exec, bdev_io, internal.link);
1090 0 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1091 :
1092 0 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1093 0 : bdev_ch_retry_io(ch);
1094 0 : }
1095 :
1096 0 : bdev_io->internal.data_transfer_cpl(bdev_io, status);
1097 0 : }
1098 :
1099 : static void
1100 0 : bdev_io_exec_sequence(struct spdk_bdev_io *bdev_io, void (*cb_fn)(void *ctx, int status))
1101 : {
1102 0 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1103 :
1104 0 : assert(bdev_io_needs_sequence_exec(bdev_io->internal.desc, bdev_io));
1105 0 : assert(bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE || bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
1106 0 : assert(bdev_io_use_accel_sequence(bdev_io));
1107 :
1108 : /* Since the operations are appended during submission, they're in the opposite order than
1109 : * how we want to execute them for reads (i.e. we need to execute the most recently added
1110 : * operation first), so reverse the sequence before executing it.
1111 : */
1112 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_READ) {
1113 0 : spdk_accel_sequence_reverse(bdev_io->internal.accel_sequence);
1114 0 : }
1115 :
1116 0 : TAILQ_INSERT_TAIL(&bdev_io->internal.ch->io_accel_exec, bdev_io, internal.link);
1117 0 : bdev_io_increment_outstanding(ch, ch->shared_resource);
1118 0 : bdev_io->internal.data_transfer_cpl = cb_fn;
1119 :
1120 0 : spdk_accel_sequence_finish(bdev_io->internal.accel_sequence,
1121 0 : bdev_io_exec_sequence_cb, bdev_io);
1122 0 : }
1123 :
1124 : static void
1125 42 : bdev_io_get_buf_complete(struct spdk_bdev_io *bdev_io, bool status)
1126 : {
1127 42 : struct spdk_io_channel *ch = spdk_bdev_io_get_io_channel(bdev_io);
1128 : void *buf;
1129 :
1130 42 : if (spdk_unlikely(bdev_io->internal.get_aux_buf_cb != NULL)) {
1131 0 : buf = bdev_io->internal.buf.ptr;
1132 0 : bdev_io->internal.buf.ptr = NULL;
1133 0 : bdev_io->internal.f.has_buf = false;
1134 0 : bdev_io->internal.get_aux_buf_cb(ch, bdev_io, buf);
1135 0 : bdev_io->internal.get_aux_buf_cb = NULL;
1136 0 : } else {
1137 42 : assert(bdev_io->internal.get_buf_cb != NULL);
1138 42 : bdev_io->internal.get_buf_cb(ch, bdev_io, status);
1139 42 : bdev_io->internal.get_buf_cb = NULL;
1140 : }
1141 42 : }
1142 :
1143 : static void
1144 4 : _bdev_io_pull_buffer_cpl(void *ctx, int rc)
1145 : {
1146 4 : struct spdk_bdev_io *bdev_io = ctx;
1147 :
1148 4 : if (rc) {
1149 0 : SPDK_ERRLOG("Set bounce buffer failed with rc %d\n", rc);
1150 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
1151 0 : }
1152 4 : bdev_io_get_buf_complete(bdev_io, !rc);
1153 4 : }
1154 :
1155 : static void
1156 2 : bdev_io_pull_md_buf_done(void *ctx, int status)
1157 : {
1158 2 : struct spdk_bdev_io *bdev_io = ctx;
1159 2 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1160 :
1161 2 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1162 2 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1163 :
1164 2 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1165 0 : bdev_ch_retry_io(ch);
1166 0 : }
1167 :
1168 2 : assert(bdev_io->internal.data_transfer_cpl);
1169 2 : bdev_io->internal.data_transfer_cpl(bdev_io, status);
1170 2 : }
1171 :
1172 : static void
1173 4 : bdev_io_pull_md_buf(struct spdk_bdev_io *bdev_io)
1174 : {
1175 4 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1176 4 : int rc = 0;
1177 :
1178 4 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
1179 2 : assert(bdev_io->internal.f.has_bounce_buf);
1180 2 : if (bdev_io_use_memory_domain(bdev_io)) {
1181 2 : TAILQ_INSERT_TAIL(&ch->io_memory_domain, bdev_io, internal.link);
1182 2 : bdev_io_increment_outstanding(ch, ch->shared_resource);
1183 4 : rc = spdk_memory_domain_pull_data(bdev_io->internal.memory_domain,
1184 2 : bdev_io->internal.memory_domain_ctx,
1185 2 : &bdev_io->internal.bounce_buf.orig_md_iov, 1,
1186 2 : &bdev_io->internal.bounce_buf.md_iov, 1,
1187 2 : bdev_io_pull_md_buf_done, bdev_io);
1188 2 : if (rc == 0) {
1189 : /* Continue to submit IO in completion callback */
1190 2 : return;
1191 : }
1192 0 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1193 0 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1194 0 : if (rc != -ENOMEM) {
1195 0 : SPDK_ERRLOG("Failed to pull data from memory domain %s, rc %d\n",
1196 : spdk_memory_domain_get_dma_device_id(
1197 : bdev_io->internal.memory_domain), rc);
1198 0 : }
1199 0 : } else {
1200 0 : memcpy(bdev_io->internal.bounce_buf.md_iov.iov_base,
1201 0 : bdev_io->internal.bounce_buf.orig_md_iov.iov_base,
1202 0 : bdev_io->internal.bounce_buf.orig_md_iov.iov_len);
1203 : }
1204 0 : }
1205 :
1206 2 : if (spdk_unlikely(rc == -ENOMEM)) {
1207 0 : bdev_queue_nomem_io_head(ch->shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PULL_MD);
1208 0 : } else {
1209 2 : assert(bdev_io->internal.data_transfer_cpl);
1210 2 : bdev_io->internal.data_transfer_cpl(bdev_io, rc);
1211 : }
1212 4 : }
1213 :
1214 : static void
1215 4 : _bdev_io_pull_bounce_md_buf(struct spdk_bdev_io *bdev_io, void *md_buf, size_t len)
1216 : {
1217 4 : assert(bdev_io->internal.f.has_bounce_buf);
1218 :
1219 : /* save original md_buf */
1220 4 : bdev_io->internal.bounce_buf.orig_md_iov.iov_base = bdev_io->u.bdev.md_buf;
1221 4 : bdev_io->internal.bounce_buf.orig_md_iov.iov_len = len;
1222 4 : bdev_io->internal.bounce_buf.md_iov.iov_base = md_buf;
1223 4 : bdev_io->internal.bounce_buf.md_iov.iov_len = len;
1224 : /* set bounce md_buf */
1225 4 : bdev_io->u.bdev.md_buf = md_buf;
1226 :
1227 4 : bdev_io_pull_md_buf(bdev_io);
1228 4 : }
1229 :
1230 : static void
1231 42 : _bdev_io_set_md_buf(struct spdk_bdev_io *bdev_io)
1232 : {
1233 42 : struct spdk_bdev *bdev = bdev_io->bdev;
1234 : uint64_t md_len;
1235 : void *buf;
1236 :
1237 42 : if (spdk_bdev_is_md_separate(bdev)) {
1238 7 : assert(!bdev_io_use_accel_sequence(bdev_io));
1239 :
1240 7 : buf = (char *)bdev_io->u.bdev.iovs[0].iov_base + bdev_io->u.bdev.iovs[0].iov_len;
1241 7 : md_len = bdev_io->u.bdev.num_blocks * bdev->md_len;
1242 :
1243 7 : assert(((uintptr_t)buf & (spdk_bdev_get_buf_align(bdev) - 1)) == 0);
1244 :
1245 7 : if (bdev_io->u.bdev.md_buf != NULL) {
1246 4 : _bdev_io_pull_bounce_md_buf(bdev_io, buf, md_len);
1247 4 : return;
1248 : } else {
1249 3 : spdk_bdev_io_set_md_buf(bdev_io, buf, md_len);
1250 : }
1251 3 : }
1252 :
1253 38 : bdev_io_get_buf_complete(bdev_io, true);
1254 42 : }
1255 :
1256 : static inline void
1257 26 : bdev_io_pull_data_done(struct spdk_bdev_io *bdev_io, int rc)
1258 : {
1259 26 : if (rc) {
1260 0 : SPDK_ERRLOG("Failed to get data buffer\n");
1261 0 : assert(bdev_io->internal.data_transfer_cpl);
1262 0 : bdev_io->internal.data_transfer_cpl(bdev_io, rc);
1263 0 : return;
1264 : }
1265 :
1266 26 : _bdev_io_set_md_buf(bdev_io);
1267 26 : }
1268 :
1269 : static void
1270 2 : bdev_io_pull_data_done_and_track(void *ctx, int status)
1271 : {
1272 2 : struct spdk_bdev_io *bdev_io = ctx;
1273 2 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1274 :
1275 2 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1276 2 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1277 :
1278 2 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1279 0 : bdev_ch_retry_io(ch);
1280 0 : }
1281 :
1282 2 : bdev_io_pull_data_done(bdev_io, status);
1283 2 : }
1284 :
1285 : static void
1286 27 : bdev_io_pull_data(struct spdk_bdev_io *bdev_io)
1287 : {
1288 27 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1289 27 : int rc = 0;
1290 :
1291 : /* If we need to exec an accel sequence or the IO uses a memory domain buffer and has a
1292 : * sequence, append a copy operation making accel change the src/dst buffers of the previous
1293 : * operation */
1294 27 : if (bdev_io_needs_sequence_exec(bdev_io->internal.desc, bdev_io) ||
1295 27 : (bdev_io_use_accel_sequence(bdev_io) && bdev_io_use_memory_domain(bdev_io))) {
1296 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
1297 0 : assert(bdev_io_use_accel_sequence(bdev_io));
1298 0 : assert(bdev_io->internal.f.has_bounce_buf);
1299 0 : rc = spdk_accel_append_copy(&bdev_io->internal.accel_sequence, ch->accel_channel,
1300 0 : bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt,
1301 : NULL, NULL,
1302 0 : bdev_io->internal.bounce_buf.orig_iovs,
1303 0 : bdev_io->internal.bounce_buf.orig_iovcnt,
1304 0 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain : NULL,
1305 0 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain_ctx : NULL,
1306 : NULL, NULL);
1307 0 : } else {
1308 : /* We need to reverse the src/dst for reads */
1309 0 : assert(bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
1310 0 : assert(bdev_io_use_accel_sequence(bdev_io));
1311 0 : assert(bdev_io->internal.f.has_bounce_buf);
1312 0 : rc = spdk_accel_append_copy(&bdev_io->internal.accel_sequence, ch->accel_channel,
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 0 : bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt,
1318 : NULL, NULL, NULL, NULL);
1319 : }
1320 :
1321 0 : if (spdk_unlikely(rc != 0 && rc != -ENOMEM)) {
1322 0 : SPDK_ERRLOG("Failed to append copy to accel sequence: %p\n",
1323 : bdev_io->internal.accel_sequence);
1324 0 : }
1325 27 : } else if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
1326 : /* if this is write path, copy data from original buffer to bounce buffer */
1327 17 : if (bdev_io_use_memory_domain(bdev_io)) {
1328 3 : assert(bdev_io->internal.f.has_bounce_buf);
1329 3 : TAILQ_INSERT_TAIL(&ch->io_memory_domain, bdev_io, internal.link);
1330 3 : bdev_io_increment_outstanding(ch, ch->shared_resource);
1331 6 : rc = spdk_memory_domain_pull_data(bdev_io->internal.memory_domain,
1332 3 : bdev_io->internal.memory_domain_ctx,
1333 3 : bdev_io->internal.bounce_buf.orig_iovs,
1334 3 : (uint32_t)bdev_io->internal.bounce_buf.orig_iovcnt,
1335 3 : bdev_io->u.bdev.iovs, 1,
1336 : bdev_io_pull_data_done_and_track,
1337 3 : bdev_io);
1338 3 : if (rc == 0) {
1339 : /* Continue to submit IO in completion callback */
1340 2 : return;
1341 : }
1342 1 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1343 1 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1344 1 : if (rc != -ENOMEM) {
1345 0 : SPDK_ERRLOG("Failed to pull data from memory domain %s\n",
1346 : spdk_memory_domain_get_dma_device_id(
1347 : bdev_io->internal.memory_domain));
1348 0 : }
1349 1 : } else {
1350 14 : assert(bdev_io->u.bdev.iovcnt == 1);
1351 14 : assert(bdev_io->internal.f.has_bounce_buf);
1352 28 : spdk_copy_iovs_to_buf(bdev_io->u.bdev.iovs[0].iov_base,
1353 14 : bdev_io->u.bdev.iovs[0].iov_len,
1354 14 : bdev_io->internal.bounce_buf.orig_iovs,
1355 14 : bdev_io->internal.bounce_buf.orig_iovcnt);
1356 : }
1357 15 : }
1358 :
1359 25 : if (spdk_unlikely(rc == -ENOMEM)) {
1360 1 : bdev_queue_nomem_io_head(ch->shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PULL);
1361 1 : } else {
1362 24 : bdev_io_pull_data_done(bdev_io, rc);
1363 : }
1364 27 : }
1365 :
1366 : static void
1367 26 : _bdev_io_pull_bounce_data_buf(struct spdk_bdev_io *bdev_io, void *buf, size_t len,
1368 : bdev_copy_bounce_buffer_cpl cpl_cb)
1369 : {
1370 26 : struct spdk_bdev_shared_resource *shared_resource = bdev_io->internal.ch->shared_resource;
1371 :
1372 26 : assert(bdev_io->internal.f.has_bounce_buf == false);
1373 :
1374 26 : bdev_io->internal.data_transfer_cpl = cpl_cb;
1375 26 : bdev_io->internal.f.has_bounce_buf = true;
1376 : /* save original iovec */
1377 26 : bdev_io->internal.bounce_buf.orig_iovs = bdev_io->u.bdev.iovs;
1378 26 : bdev_io->internal.bounce_buf.orig_iovcnt = bdev_io->u.bdev.iovcnt;
1379 : /* zero the other data members */
1380 26 : bdev_io->internal.bounce_buf.iov.iov_base = NULL;
1381 26 : bdev_io->internal.bounce_buf.md_iov.iov_base = NULL;
1382 26 : bdev_io->internal.bounce_buf.orig_md_iov.iov_base = NULL;
1383 : /* set bounce iov */
1384 26 : bdev_io->u.bdev.iovs = &bdev_io->internal.bounce_buf.iov;
1385 26 : bdev_io->u.bdev.iovcnt = 1;
1386 : /* set bounce buffer for this operation */
1387 26 : bdev_io->u.bdev.iovs[0].iov_base = buf;
1388 26 : bdev_io->u.bdev.iovs[0].iov_len = len;
1389 : /* Now we use 1 iov, the split condition could have been changed */
1390 26 : bdev_io->internal.f.split = bdev_io_should_split(bdev_io);
1391 :
1392 26 : if (spdk_unlikely(!TAILQ_EMPTY(&shared_resource->nomem_io))) {
1393 0 : bdev_queue_nomem_io_tail(shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PULL);
1394 0 : } else {
1395 26 : bdev_io_pull_data(bdev_io);
1396 : }
1397 26 : }
1398 :
1399 : static void
1400 42 : _bdev_io_set_buf(struct spdk_bdev_io *bdev_io, void *buf, uint64_t len)
1401 : {
1402 42 : struct spdk_bdev *bdev = bdev_io->bdev;
1403 : bool buf_allocated;
1404 : uint64_t alignment;
1405 : void *aligned_buf;
1406 :
1407 42 : bdev_io->internal.buf.ptr = buf;
1408 42 : bdev_io->internal.f.has_buf = true;
1409 :
1410 42 : if (spdk_unlikely(bdev_io->internal.get_aux_buf_cb != NULL)) {
1411 0 : bdev_io_get_buf_complete(bdev_io, true);
1412 0 : return;
1413 : }
1414 :
1415 42 : alignment = spdk_bdev_get_buf_align(bdev);
1416 42 : buf_allocated = _is_buf_allocated(bdev_io->u.bdev.iovs);
1417 42 : aligned_buf = (void *)(((uintptr_t)buf + (alignment - 1)) & ~(alignment - 1));
1418 :
1419 42 : if (buf_allocated) {
1420 26 : _bdev_io_pull_bounce_data_buf(bdev_io, aligned_buf, len, _bdev_io_pull_buffer_cpl);
1421 : /* Continue in completion callback */
1422 26 : return;
1423 : } else {
1424 16 : spdk_bdev_io_set_buf(bdev_io, aligned_buf, len);
1425 : }
1426 :
1427 16 : _bdev_io_set_md_buf(bdev_io);
1428 42 : }
1429 :
1430 : static inline uint64_t
1431 84 : bdev_io_get_max_buf_len(struct spdk_bdev_io *bdev_io, uint64_t len)
1432 : {
1433 84 : struct spdk_bdev *bdev = bdev_io->bdev;
1434 : uint64_t md_len, alignment;
1435 :
1436 84 : md_len = spdk_bdev_is_md_separate(bdev) ? bdev_io->u.bdev.num_blocks * bdev->md_len : 0;
1437 :
1438 : /* 1 byte alignment needs 0 byte of extra space, 64 bytes alignment needs 63 bytes of extra space, etc. */
1439 84 : alignment = spdk_bdev_get_buf_align(bdev) - 1;
1440 :
1441 84 : return len + alignment + md_len;
1442 : }
1443 :
1444 : static void
1445 42 : _bdev_io_put_buf(struct spdk_bdev_io *bdev_io, void *buf, uint64_t buf_len)
1446 : {
1447 : struct spdk_bdev_mgmt_channel *ch;
1448 :
1449 42 : ch = bdev_io->internal.ch->shared_resource->mgmt_ch;
1450 42 : spdk_iobuf_put(&ch->iobuf, buf, bdev_io_get_max_buf_len(bdev_io, buf_len));
1451 42 : }
1452 :
1453 : static void
1454 42 : bdev_io_put_buf(struct spdk_bdev_io *bdev_io)
1455 : {
1456 42 : assert(bdev_io->internal.f.has_buf);
1457 42 : _bdev_io_put_buf(bdev_io, bdev_io->internal.buf.ptr, bdev_io->internal.buf.len);
1458 42 : bdev_io->internal.buf.ptr = NULL;
1459 42 : bdev_io->internal.f.has_buf = false;
1460 42 : }
1461 :
1462 3 : SPDK_LOG_DEPRECATION_REGISTER(spdk_bdev_io_put_aux_buf,
1463 : "spdk_bdev_io_put_aux_buf is deprecated", "v25.01", 0);
1464 :
1465 : void
1466 0 : spdk_bdev_io_put_aux_buf(struct spdk_bdev_io *bdev_io, void *buf)
1467 : {
1468 0 : uint64_t len = bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen;
1469 :
1470 0 : SPDK_LOG_DEPRECATED(spdk_bdev_io_put_aux_buf);
1471 :
1472 0 : assert(buf != NULL);
1473 0 : _bdev_io_put_buf(bdev_io, buf, len);
1474 0 : }
1475 :
1476 : static inline void
1477 548 : bdev_submit_request(struct spdk_bdev *bdev, struct spdk_io_channel *ioch,
1478 : struct spdk_bdev_io *bdev_io)
1479 : {
1480 : /* After a request is submitted to a bdev module, the ownership of an accel sequence
1481 : * associated with that bdev_io is transferred to the bdev module. So, clear the internal
1482 : * sequence pointer to make sure we won't touch it anymore. */
1483 1014 : if ((bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE ||
1484 548 : bdev_io->type == SPDK_BDEV_IO_TYPE_READ) && bdev_io->u.bdev.accel_sequence != NULL) {
1485 0 : assert(!bdev_io_needs_sequence_exec(bdev_io->internal.desc, bdev_io));
1486 0 : bdev_io->internal.f.has_accel_sequence = false;
1487 0 : }
1488 :
1489 548 : bdev->fn_table->submit_request(ioch, bdev_io);
1490 548 : }
1491 :
1492 : static inline void
1493 10 : bdev_ch_resubmit_io(struct spdk_bdev_shared_resource *shared_resource, struct spdk_bdev_io *bdev_io)
1494 : {
1495 10 : struct spdk_bdev *bdev = bdev_io->bdev;
1496 :
1497 10 : bdev_io_increment_outstanding(bdev_io->internal.ch, shared_resource);
1498 10 : bdev_io->internal.error.nvme.cdw0 = 0;
1499 10 : bdev_io->num_retries++;
1500 10 : bdev_submit_request(bdev, spdk_bdev_io_get_io_channel(bdev_io), bdev_io);
1501 10 : }
1502 :
1503 : static void
1504 63 : bdev_shared_ch_retry_io(struct spdk_bdev_shared_resource *shared_resource)
1505 : {
1506 : struct spdk_bdev_io *bdev_io;
1507 :
1508 63 : if (shared_resource->io_outstanding > shared_resource->nomem_threshold) {
1509 : /*
1510 : * Allow some more I/O to complete before retrying the nomem_io queue.
1511 : * Some drivers (such as nvme) cannot immediately take a new I/O in
1512 : * the context of a completion, because the resources for the I/O are
1513 : * not released until control returns to the bdev poller. Also, we
1514 : * may require several small I/O to complete before a larger I/O
1515 : * (that requires splitting) can be submitted.
1516 : */
1517 58 : return;
1518 : }
1519 :
1520 16 : while (!TAILQ_EMPTY(&shared_resource->nomem_io)) {
1521 12 : bdev_io = TAILQ_FIRST(&shared_resource->nomem_io);
1522 12 : TAILQ_REMOVE(&shared_resource->nomem_io, bdev_io, internal.link);
1523 :
1524 12 : switch (bdev_io->internal.retry_state) {
1525 : case BDEV_IO_RETRY_STATE_SUBMIT:
1526 10 : bdev_ch_resubmit_io(shared_resource, bdev_io);
1527 10 : break;
1528 : case BDEV_IO_RETRY_STATE_PULL:
1529 1 : bdev_io_pull_data(bdev_io);
1530 1 : break;
1531 : case BDEV_IO_RETRY_STATE_PULL_MD:
1532 0 : bdev_io_pull_md_buf(bdev_io);
1533 0 : break;
1534 : case BDEV_IO_RETRY_STATE_PUSH:
1535 1 : bdev_io_push_bounce_data(bdev_io);
1536 1 : break;
1537 : case BDEV_IO_RETRY_STATE_PUSH_MD:
1538 0 : bdev_io_push_bounce_md_buf(bdev_io);
1539 0 : break;
1540 : default:
1541 0 : assert(0 && "invalid retry state");
1542 : break;
1543 : }
1544 :
1545 12 : if (bdev_io == TAILQ_FIRST(&shared_resource->nomem_io)) {
1546 : /* This IO completed again with NOMEM status, so break the loop and
1547 : * don't try anymore. Note that a bdev_io that fails with NOMEM
1548 : * always gets requeued at the front of the list, to maintain
1549 : * ordering.
1550 : */
1551 1 : break;
1552 : }
1553 : }
1554 63 : }
1555 :
1556 : static void
1557 63 : bdev_ch_retry_io(struct spdk_bdev_channel *bdev_ch)
1558 : {
1559 63 : bdev_shared_ch_retry_io(bdev_ch->shared_resource);
1560 63 : }
1561 :
1562 : static int
1563 0 : bdev_no_mem_poller(void *ctx)
1564 : {
1565 0 : struct spdk_bdev_shared_resource *shared_resource = ctx;
1566 :
1567 0 : spdk_poller_unregister(&shared_resource->nomem_poller);
1568 :
1569 0 : if (!TAILQ_EMPTY(&shared_resource->nomem_io)) {
1570 0 : bdev_shared_ch_retry_io(shared_resource);
1571 0 : }
1572 : /* the retry cb may re-register the poller so double check */
1573 0 : if (!TAILQ_EMPTY(&shared_resource->nomem_io) &&
1574 0 : shared_resource->io_outstanding == 0 && shared_resource->nomem_poller == NULL) {
1575 : /* No IOs were submitted, try again */
1576 0 : shared_resource->nomem_poller = SPDK_POLLER_REGISTER(bdev_no_mem_poller, shared_resource,
1577 : SPDK_BDEV_IO_POLL_INTERVAL_IN_MSEC * 10);
1578 0 : }
1579 :
1580 0 : return SPDK_POLLER_BUSY;
1581 : }
1582 :
1583 : static inline bool
1584 556 : _bdev_io_handle_no_mem(struct spdk_bdev_io *bdev_io, enum bdev_io_retry_state state)
1585 : {
1586 556 : struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
1587 556 : struct spdk_bdev_shared_resource *shared_resource = bdev_ch->shared_resource;
1588 :
1589 556 : if (spdk_unlikely(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_NOMEM)) {
1590 5 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_PENDING;
1591 5 : bdev_queue_nomem_io_head(shared_resource, bdev_io, state);
1592 :
1593 5 : if (shared_resource->io_outstanding == 0 && !shared_resource->nomem_poller) {
1594 : /* Special case when we have nomem IOs and no outstanding IOs which completions
1595 : * could trigger retry of queued IOs
1596 : * Any IOs submitted may trigger retry of queued IOs. This poller handles a case when no
1597 : * new IOs submitted, e.g. qd==1 */
1598 0 : shared_resource->nomem_poller = SPDK_POLLER_REGISTER(bdev_no_mem_poller, shared_resource,
1599 : SPDK_BDEV_IO_POLL_INTERVAL_IN_MSEC * 10);
1600 0 : }
1601 : /* If bdev module completed an I/O that has an accel sequence with NOMEM status, the
1602 : * ownership of that sequence is transferred back to the bdev layer, so we need to
1603 : * restore internal.accel_sequence to make sure that the sequence is handled
1604 : * correctly in case the I/O is later aborted. */
1605 5 : if ((bdev_io->type == SPDK_BDEV_IO_TYPE_READ ||
1606 5 : bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) && bdev_io->u.bdev.accel_sequence) {
1607 0 : assert(!bdev_io_use_accel_sequence(bdev_io));
1608 0 : bdev_io->internal.f.has_accel_sequence = true;
1609 0 : bdev_io->internal.accel_sequence = bdev_io->u.bdev.accel_sequence;
1610 0 : }
1611 :
1612 5 : return true;
1613 : }
1614 :
1615 551 : if (spdk_unlikely(!TAILQ_EMPTY(&shared_resource->nomem_io))) {
1616 63 : bdev_ch_retry_io(bdev_ch);
1617 63 : }
1618 :
1619 551 : return false;
1620 556 : }
1621 :
1622 : static void
1623 26 : _bdev_io_complete_push_bounce_done(void *ctx, int rc)
1624 : {
1625 26 : struct spdk_bdev_io *bdev_io = ctx;
1626 26 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1627 :
1628 26 : if (rc) {
1629 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
1630 0 : }
1631 : /* We want to free the bounce buffer here since we know we're done with it (as opposed
1632 : * to waiting for the conditional free of internal.buf.ptr in spdk_bdev_free_io()).
1633 : */
1634 26 : bdev_io_put_buf(bdev_io);
1635 :
1636 26 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1637 0 : bdev_ch_retry_io(ch);
1638 0 : }
1639 :
1640 : /* Continue with IO completion flow */
1641 26 : bdev_io_complete(bdev_io);
1642 26 : }
1643 :
1644 : static void
1645 2 : bdev_io_push_bounce_md_buf_done(void *ctx, int rc)
1646 : {
1647 2 : struct spdk_bdev_io *bdev_io = ctx;
1648 2 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1649 :
1650 2 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1651 2 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1652 2 : bdev_io->internal.f.has_bounce_buf = false;
1653 :
1654 2 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1655 0 : bdev_ch_retry_io(ch);
1656 0 : }
1657 :
1658 2 : bdev_io->internal.data_transfer_cpl(bdev_io, rc);
1659 2 : }
1660 :
1661 : static inline void
1662 26 : bdev_io_push_bounce_md_buf(struct spdk_bdev_io *bdev_io)
1663 : {
1664 26 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1665 26 : int rc = 0;
1666 :
1667 26 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS);
1668 26 : assert(bdev_io->internal.f.has_bounce_buf);
1669 :
1670 : /* do the same for metadata buffer */
1671 26 : if (spdk_unlikely(bdev_io->internal.bounce_buf.orig_md_iov.iov_base != NULL)) {
1672 4 : assert(spdk_bdev_is_md_separate(bdev_io->bdev));
1673 :
1674 4 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_READ) {
1675 2 : if (bdev_io_use_memory_domain(bdev_io)) {
1676 2 : TAILQ_INSERT_TAIL(&ch->io_memory_domain, bdev_io, internal.link);
1677 2 : bdev_io_increment_outstanding(ch, ch->shared_resource);
1678 : /* If memory domain is used then we need to call async push function */
1679 4 : rc = spdk_memory_domain_push_data(bdev_io->internal.memory_domain,
1680 2 : bdev_io->internal.memory_domain_ctx,
1681 2 : &bdev_io->internal.bounce_buf.orig_md_iov,
1682 2 : (uint32_t)bdev_io->internal.bounce_buf.orig_iovcnt,
1683 2 : &bdev_io->internal.bounce_buf.md_iov, 1,
1684 : bdev_io_push_bounce_md_buf_done,
1685 2 : bdev_io);
1686 2 : if (rc == 0) {
1687 : /* Continue IO completion in async callback */
1688 2 : return;
1689 : }
1690 0 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1691 0 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1692 0 : if (rc != -ENOMEM) {
1693 0 : SPDK_ERRLOG("Failed to push md to memory domain %s\n",
1694 : spdk_memory_domain_get_dma_device_id(
1695 : bdev_io->internal.memory_domain));
1696 0 : }
1697 0 : } else {
1698 0 : memcpy(bdev_io->internal.bounce_buf.orig_md_iov.iov_base, bdev_io->u.bdev.md_buf,
1699 0 : bdev_io->internal.bounce_buf.orig_md_iov.iov_len);
1700 : }
1701 0 : }
1702 2 : }
1703 :
1704 24 : if (spdk_unlikely(rc == -ENOMEM)) {
1705 0 : bdev_queue_nomem_io_head(ch->shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PUSH_MD);
1706 0 : } else {
1707 24 : assert(bdev_io->internal.data_transfer_cpl);
1708 24 : bdev_io->internal.f.has_bounce_buf = false;
1709 24 : bdev_io->internal.data_transfer_cpl(bdev_io, rc);
1710 : }
1711 26 : }
1712 :
1713 : static inline void
1714 26 : bdev_io_push_bounce_data_done(struct spdk_bdev_io *bdev_io, int rc)
1715 : {
1716 26 : assert(bdev_io->internal.data_transfer_cpl);
1717 26 : if (rc) {
1718 0 : bdev_io->internal.data_transfer_cpl(bdev_io, rc);
1719 0 : return;
1720 : }
1721 :
1722 : /* set original buffer for this io */
1723 26 : bdev_io->u.bdev.iovcnt = bdev_io->internal.bounce_buf.orig_iovcnt;
1724 26 : bdev_io->u.bdev.iovs = bdev_io->internal.bounce_buf.orig_iovs;
1725 :
1726 : /* We don't set bdev_io->internal.f.has_bounce_buf to false here because
1727 : * we still need to clear the md buf */
1728 :
1729 26 : bdev_io_push_bounce_md_buf(bdev_io);
1730 26 : }
1731 :
1732 : static void
1733 2 : bdev_io_push_bounce_data_done_and_track(void *ctx, int status)
1734 : {
1735 2 : struct spdk_bdev_io *bdev_io = ctx;
1736 2 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1737 :
1738 2 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1739 2 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1740 :
1741 2 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1742 0 : bdev_ch_retry_io(ch);
1743 0 : }
1744 :
1745 2 : bdev_io_push_bounce_data_done(bdev_io, status);
1746 2 : }
1747 :
1748 : static inline void
1749 27 : bdev_io_push_bounce_data(struct spdk_bdev_io *bdev_io)
1750 : {
1751 27 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1752 27 : int rc = 0;
1753 :
1754 27 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS);
1755 27 : assert(!bdev_io_use_accel_sequence(bdev_io));
1756 27 : assert(bdev_io->internal.f.has_bounce_buf);
1757 :
1758 : /* if this is read path, copy data from bounce buffer to original buffer */
1759 27 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_READ) {
1760 11 : if (bdev_io_use_memory_domain(bdev_io)) {
1761 3 : TAILQ_INSERT_TAIL(&ch->io_memory_domain, bdev_io, internal.link);
1762 3 : bdev_io_increment_outstanding(ch, ch->shared_resource);
1763 : /* If memory domain is used then we need to call async push function */
1764 6 : rc = spdk_memory_domain_push_data(bdev_io->internal.memory_domain,
1765 3 : bdev_io->internal.memory_domain_ctx,
1766 3 : bdev_io->internal.bounce_buf.orig_iovs,
1767 3 : (uint32_t)bdev_io->internal.bounce_buf.orig_iovcnt,
1768 3 : &bdev_io->internal.bounce_buf.iov, 1,
1769 : bdev_io_push_bounce_data_done_and_track,
1770 3 : bdev_io);
1771 3 : if (rc == 0) {
1772 : /* Continue IO completion in async callback */
1773 2 : return;
1774 : }
1775 :
1776 1 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1777 1 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1778 1 : if (rc != -ENOMEM) {
1779 0 : SPDK_ERRLOG("Failed to push data to memory domain %s\n",
1780 : spdk_memory_domain_get_dma_device_id(
1781 : bdev_io->internal.memory_domain));
1782 0 : }
1783 1 : } else {
1784 16 : spdk_copy_buf_to_iovs(bdev_io->internal.bounce_buf.orig_iovs,
1785 8 : bdev_io->internal.bounce_buf.orig_iovcnt,
1786 8 : bdev_io->internal.bounce_buf.iov.iov_base,
1787 8 : bdev_io->internal.bounce_buf.iov.iov_len);
1788 : }
1789 9 : }
1790 :
1791 25 : if (spdk_unlikely(rc == -ENOMEM)) {
1792 1 : bdev_queue_nomem_io_head(ch->shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PUSH);
1793 1 : } else {
1794 24 : bdev_io_push_bounce_data_done(bdev_io, rc);
1795 : }
1796 27 : }
1797 :
1798 : static inline void
1799 26 : _bdev_io_push_bounce_data_buffer(struct spdk_bdev_io *bdev_io, bdev_copy_bounce_buffer_cpl cpl_cb)
1800 : {
1801 26 : bdev_io->internal.data_transfer_cpl = cpl_cb;
1802 26 : bdev_io_push_bounce_data(bdev_io);
1803 26 : }
1804 :
1805 : static void
1806 0 : bdev_io_get_iobuf_cb(struct spdk_iobuf_entry *iobuf, void *buf)
1807 : {
1808 : struct spdk_bdev_io *bdev_io;
1809 :
1810 0 : bdev_io = SPDK_CONTAINEROF(iobuf, struct spdk_bdev_io, internal.iobuf);
1811 0 : _bdev_io_set_buf(bdev_io, buf, bdev_io->internal.buf.len);
1812 0 : }
1813 :
1814 : static void
1815 42 : bdev_io_get_buf(struct spdk_bdev_io *bdev_io, uint64_t len)
1816 : {
1817 : struct spdk_bdev_mgmt_channel *mgmt_ch;
1818 : uint64_t max_len;
1819 : void *buf;
1820 :
1821 42 : assert(spdk_bdev_io_get_thread(bdev_io) == spdk_get_thread());
1822 42 : mgmt_ch = bdev_io->internal.ch->shared_resource->mgmt_ch;
1823 42 : max_len = bdev_io_get_max_buf_len(bdev_io, len);
1824 :
1825 42 : if (spdk_unlikely(max_len > mgmt_ch->iobuf.cache[0].large.bufsize)) {
1826 0 : SPDK_ERRLOG("Length %" PRIu64 " is larger than allowed\n", max_len);
1827 0 : bdev_io_get_buf_complete(bdev_io, false);
1828 0 : return;
1829 : }
1830 :
1831 42 : bdev_io->internal.buf.len = len;
1832 42 : buf = spdk_iobuf_get(&mgmt_ch->iobuf, max_len, &bdev_io->internal.iobuf,
1833 : bdev_io_get_iobuf_cb);
1834 42 : if (buf != NULL) {
1835 42 : _bdev_io_set_buf(bdev_io, buf, len);
1836 42 : }
1837 42 : }
1838 :
1839 : void
1840 56 : spdk_bdev_io_get_buf(struct spdk_bdev_io *bdev_io, spdk_bdev_io_get_buf_cb cb, uint64_t len)
1841 : {
1842 56 : struct spdk_bdev *bdev = bdev_io->bdev;
1843 : uint64_t alignment;
1844 :
1845 56 : assert(cb != NULL);
1846 56 : bdev_io->internal.get_buf_cb = cb;
1847 :
1848 56 : alignment = spdk_bdev_get_buf_align(bdev);
1849 :
1850 56 : if (_is_buf_allocated(bdev_io->u.bdev.iovs) &&
1851 40 : _are_iovs_aligned(bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt, alignment)) {
1852 : /* Buffer already present and aligned */
1853 18 : cb(spdk_bdev_io_get_io_channel(bdev_io), bdev_io, true);
1854 18 : return;
1855 : }
1856 :
1857 38 : bdev_io_get_buf(bdev_io, len);
1858 56 : }
1859 :
1860 : static void
1861 4 : _bdev_memory_domain_get_io_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io,
1862 : bool success)
1863 : {
1864 4 : if (!success) {
1865 0 : SPDK_ERRLOG("Failed to get data buffer, completing IO\n");
1866 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
1867 0 : bdev_io_complete_unsubmitted(bdev_io);
1868 0 : return;
1869 : }
1870 :
1871 4 : if (bdev_io_needs_sequence_exec(bdev_io->internal.desc, bdev_io)) {
1872 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
1873 0 : bdev_io_exec_sequence(bdev_io, bdev_io_submit_sequence_cb);
1874 0 : return;
1875 : }
1876 : /* For reads we'll execute the sequence after the data is read, so, for now, only
1877 : * clear out accel_sequence pointer and submit the IO */
1878 0 : assert(bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
1879 0 : bdev_io->u.bdev.accel_sequence = NULL;
1880 0 : }
1881 :
1882 4 : bdev_io_submit(bdev_io);
1883 4 : }
1884 :
1885 : static void
1886 4 : _bdev_memory_domain_io_get_buf(struct spdk_bdev_io *bdev_io, spdk_bdev_io_get_buf_cb cb,
1887 : uint64_t len)
1888 : {
1889 4 : assert(cb != NULL);
1890 4 : bdev_io->internal.get_buf_cb = cb;
1891 :
1892 4 : bdev_io_get_buf(bdev_io, len);
1893 4 : }
1894 :
1895 :
1896 3 : SPDK_LOG_DEPRECATION_REGISTER(spdk_bdev_io_get_aux_buf,
1897 : "spdk_bdev_io_get_aux_buf is deprecated", "v25.01", 0);
1898 :
1899 : void
1900 0 : spdk_bdev_io_get_aux_buf(struct spdk_bdev_io *bdev_io, spdk_bdev_io_get_aux_buf_cb cb)
1901 : {
1902 0 : uint64_t len = bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen;
1903 :
1904 0 : SPDK_LOG_DEPRECATED(spdk_bdev_io_get_aux_buf);
1905 :
1906 0 : assert(cb != NULL);
1907 0 : assert(bdev_io->internal.get_aux_buf_cb == NULL);
1908 0 : bdev_io->internal.get_aux_buf_cb = cb;
1909 0 : bdev_io_get_buf(bdev_io, len);
1910 0 : }
1911 :
1912 : static int
1913 65 : bdev_module_get_max_ctx_size(void)
1914 : {
1915 : struct spdk_bdev_module *bdev_module;
1916 65 : int max_bdev_module_size = 0;
1917 :
1918 254 : TAILQ_FOREACH(bdev_module, &g_bdev_mgr.bdev_modules, internal.tailq) {
1919 189 : if (bdev_module->get_ctx_size && bdev_module->get_ctx_size() > max_bdev_module_size) {
1920 64 : max_bdev_module_size = bdev_module->get_ctx_size();
1921 64 : }
1922 189 : }
1923 :
1924 65 : return max_bdev_module_size;
1925 : }
1926 :
1927 : static void
1928 0 : bdev_enable_histogram_config_json(struct spdk_bdev *bdev, struct spdk_json_write_ctx *w)
1929 : {
1930 0 : if (!bdev->internal.histogram_enabled) {
1931 0 : return;
1932 : }
1933 :
1934 0 : spdk_json_write_object_begin(w);
1935 0 : spdk_json_write_named_string(w, "method", "bdev_enable_histogram");
1936 :
1937 0 : spdk_json_write_named_object_begin(w, "params");
1938 0 : spdk_json_write_named_string(w, "name", bdev->name);
1939 :
1940 0 : spdk_json_write_named_bool(w, "enable", bdev->internal.histogram_enabled);
1941 :
1942 0 : if (bdev->internal.histogram_io_type) {
1943 0 : spdk_json_write_named_string(w, "opc",
1944 0 : spdk_bdev_get_io_type_name(bdev->internal.histogram_io_type));
1945 0 : }
1946 :
1947 0 : spdk_json_write_object_end(w);
1948 :
1949 0 : spdk_json_write_object_end(w);
1950 0 : }
1951 :
1952 : static void
1953 0 : bdev_qos_config_json(struct spdk_bdev *bdev, struct spdk_json_write_ctx *w)
1954 : {
1955 : int i;
1956 0 : struct spdk_bdev_qos *qos = bdev->internal.qos;
1957 : uint64_t limits[SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES];
1958 :
1959 0 : if (!qos) {
1960 0 : return;
1961 : }
1962 :
1963 0 : spdk_bdev_get_qos_rate_limits(bdev, limits);
1964 :
1965 0 : spdk_json_write_object_begin(w);
1966 0 : spdk_json_write_named_string(w, "method", "bdev_set_qos_limit");
1967 :
1968 0 : spdk_json_write_named_object_begin(w, "params");
1969 0 : spdk_json_write_named_string(w, "name", bdev->name);
1970 0 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
1971 0 : if (limits[i] > 0) {
1972 0 : spdk_json_write_named_uint64(w, qos_rpc_type[i], limits[i]);
1973 0 : }
1974 0 : }
1975 0 : spdk_json_write_object_end(w);
1976 :
1977 0 : spdk_json_write_object_end(w);
1978 0 : }
1979 :
1980 : void
1981 0 : spdk_bdev_subsystem_config_json(struct spdk_json_write_ctx *w)
1982 : {
1983 : struct spdk_bdev_module *bdev_module;
1984 : struct spdk_bdev *bdev;
1985 :
1986 0 : assert(w != NULL);
1987 :
1988 0 : spdk_json_write_array_begin(w);
1989 :
1990 0 : spdk_json_write_object_begin(w);
1991 0 : spdk_json_write_named_string(w, "method", "bdev_set_options");
1992 0 : spdk_json_write_named_object_begin(w, "params");
1993 0 : spdk_json_write_named_uint32(w, "bdev_io_pool_size", g_bdev_opts.bdev_io_pool_size);
1994 0 : spdk_json_write_named_uint32(w, "bdev_io_cache_size", g_bdev_opts.bdev_io_cache_size);
1995 0 : spdk_json_write_named_bool(w, "bdev_auto_examine", g_bdev_opts.bdev_auto_examine);
1996 0 : spdk_json_write_named_uint32(w, "iobuf_small_cache_size", g_bdev_opts.iobuf_small_cache_size);
1997 0 : spdk_json_write_named_uint32(w, "iobuf_large_cache_size", g_bdev_opts.iobuf_large_cache_size);
1998 0 : spdk_json_write_object_end(w);
1999 0 : spdk_json_write_object_end(w);
2000 :
2001 0 : bdev_examine_allowlist_config_json(w);
2002 :
2003 0 : TAILQ_FOREACH(bdev_module, &g_bdev_mgr.bdev_modules, internal.tailq) {
2004 0 : if (bdev_module->config_json) {
2005 0 : bdev_module->config_json(w);
2006 0 : }
2007 0 : }
2008 :
2009 0 : spdk_spin_lock(&g_bdev_mgr.spinlock);
2010 :
2011 0 : TAILQ_FOREACH(bdev, &g_bdev_mgr.bdevs, internal.link) {
2012 0 : if (bdev->fn_table->write_config_json) {
2013 0 : bdev->fn_table->write_config_json(bdev, w);
2014 0 : }
2015 :
2016 0 : bdev_qos_config_json(bdev, w);
2017 0 : bdev_enable_histogram_config_json(bdev, w);
2018 0 : }
2019 :
2020 0 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
2021 :
2022 : /* This has to be last RPC in array to make sure all bdevs finished examine */
2023 0 : spdk_json_write_object_begin(w);
2024 0 : spdk_json_write_named_string(w, "method", "bdev_wait_for_examine");
2025 0 : spdk_json_write_object_end(w);
2026 :
2027 0 : spdk_json_write_array_end(w);
2028 0 : }
2029 :
2030 : static void
2031 71 : bdev_mgmt_channel_destroy(void *io_device, void *ctx_buf)
2032 : {
2033 71 : struct spdk_bdev_mgmt_channel *ch = ctx_buf;
2034 : struct spdk_bdev_io *bdev_io;
2035 :
2036 71 : spdk_iobuf_channel_fini(&ch->iobuf);
2037 :
2038 9969 : while (!STAILQ_EMPTY(&ch->per_thread_cache)) {
2039 9898 : bdev_io = STAILQ_FIRST(&ch->per_thread_cache);
2040 9898 : STAILQ_REMOVE_HEAD(&ch->per_thread_cache, internal.buf_link);
2041 9898 : ch->per_thread_cache_count--;
2042 9898 : spdk_mempool_put(g_bdev_mgr.bdev_io_pool, (void *)bdev_io);
2043 : }
2044 :
2045 71 : assert(ch->per_thread_cache_count == 0);
2046 71 : }
2047 :
2048 : static int
2049 71 : bdev_mgmt_channel_create(void *io_device, void *ctx_buf)
2050 : {
2051 71 : struct spdk_bdev_mgmt_channel *ch = ctx_buf;
2052 : struct spdk_bdev_io *bdev_io;
2053 : uint32_t i;
2054 : int rc;
2055 :
2056 142 : rc = spdk_iobuf_channel_init(&ch->iobuf, "bdev",
2057 71 : g_bdev_opts.iobuf_small_cache_size,
2058 71 : g_bdev_opts.iobuf_large_cache_size);
2059 71 : if (rc != 0) {
2060 0 : SPDK_ERRLOG("Failed to create iobuf channel: %s\n", spdk_strerror(-rc));
2061 0 : return -1;
2062 : }
2063 :
2064 71 : STAILQ_INIT(&ch->per_thread_cache);
2065 71 : ch->bdev_io_cache_size = g_bdev_opts.bdev_io_cache_size;
2066 :
2067 : /* Pre-populate bdev_io cache to ensure this thread cannot be starved. */
2068 71 : ch->per_thread_cache_count = 0;
2069 9969 : for (i = 0; i < ch->bdev_io_cache_size; i++) {
2070 9898 : bdev_io = spdk_mempool_get(g_bdev_mgr.bdev_io_pool);
2071 9898 : if (bdev_io == NULL) {
2072 0 : SPDK_ERRLOG("You need to increase bdev_io_pool_size using bdev_set_options RPC.\n");
2073 0 : assert(false);
2074 : bdev_mgmt_channel_destroy(io_device, ctx_buf);
2075 : return -1;
2076 : }
2077 9898 : ch->per_thread_cache_count++;
2078 9898 : STAILQ_INSERT_HEAD(&ch->per_thread_cache, bdev_io, internal.buf_link);
2079 9898 : }
2080 :
2081 71 : TAILQ_INIT(&ch->shared_resources);
2082 71 : TAILQ_INIT(&ch->io_wait_queue);
2083 :
2084 71 : return 0;
2085 71 : }
2086 :
2087 : static void
2088 65 : bdev_init_complete(int rc)
2089 : {
2090 65 : spdk_bdev_init_cb cb_fn = g_init_cb_fn;
2091 65 : void *cb_arg = g_init_cb_arg;
2092 : struct spdk_bdev_module *m;
2093 :
2094 65 : g_bdev_mgr.init_complete = true;
2095 65 : g_init_cb_fn = NULL;
2096 65 : g_init_cb_arg = NULL;
2097 :
2098 : /*
2099 : * For modules that need to know when subsystem init is complete,
2100 : * inform them now.
2101 : */
2102 65 : if (rc == 0) {
2103 254 : TAILQ_FOREACH(m, &g_bdev_mgr.bdev_modules, internal.tailq) {
2104 189 : if (m->init_complete) {
2105 23 : m->init_complete();
2106 23 : }
2107 189 : }
2108 65 : }
2109 :
2110 65 : cb_fn(cb_arg, rc);
2111 65 : }
2112 :
2113 : static bool
2114 257 : bdev_module_all_actions_completed(void)
2115 : {
2116 : struct spdk_bdev_module *m;
2117 :
2118 1017 : TAILQ_FOREACH(m, &g_bdev_mgr.bdev_modules, internal.tailq) {
2119 760 : if (m->internal.action_in_progress > 0) {
2120 0 : return false;
2121 : }
2122 760 : }
2123 257 : return true;
2124 257 : }
2125 :
2126 : static void
2127 622 : bdev_module_action_complete(void)
2128 : {
2129 : /*
2130 : * Don't finish bdev subsystem initialization if
2131 : * module pre-initialization is still in progress, or
2132 : * the subsystem been already initialized.
2133 : */
2134 622 : if (!g_bdev_mgr.module_init_complete || g_bdev_mgr.init_complete) {
2135 557 : return;
2136 : }
2137 :
2138 : /*
2139 : * Check all bdev modules for inits/examinations in progress. If any
2140 : * exist, return immediately since we cannot finish bdev subsystem
2141 : * initialization until all are completed.
2142 : */
2143 65 : if (!bdev_module_all_actions_completed()) {
2144 0 : return;
2145 : }
2146 :
2147 : /*
2148 : * Modules already finished initialization - now that all
2149 : * the bdev modules have finished their asynchronous I/O
2150 : * processing, the entire bdev layer can be marked as complete.
2151 : */
2152 65 : bdev_init_complete(0);
2153 622 : }
2154 :
2155 : static void
2156 557 : bdev_module_action_done(struct spdk_bdev_module *module)
2157 : {
2158 557 : spdk_spin_lock(&module->internal.spinlock);
2159 557 : assert(module->internal.action_in_progress > 0);
2160 557 : module->internal.action_in_progress--;
2161 557 : spdk_spin_unlock(&module->internal.spinlock);
2162 557 : bdev_module_action_complete();
2163 557 : }
2164 :
2165 : void
2166 65 : spdk_bdev_module_init_done(struct spdk_bdev_module *module)
2167 : {
2168 65 : assert(module->async_init);
2169 65 : bdev_module_action_done(module);
2170 65 : }
2171 :
2172 : void
2173 492 : spdk_bdev_module_examine_done(struct spdk_bdev_module *module)
2174 : {
2175 492 : bdev_module_action_done(module);
2176 492 : }
2177 :
2178 : /** The last initialized bdev module */
2179 : static struct spdk_bdev_module *g_resume_bdev_module = NULL;
2180 :
2181 : static void
2182 0 : bdev_init_failed(void *cb_arg)
2183 : {
2184 0 : struct spdk_bdev_module *module = cb_arg;
2185 :
2186 0 : spdk_spin_lock(&module->internal.spinlock);
2187 0 : assert(module->internal.action_in_progress > 0);
2188 0 : module->internal.action_in_progress--;
2189 0 : spdk_spin_unlock(&module->internal.spinlock);
2190 0 : bdev_init_complete(-1);
2191 0 : }
2192 :
2193 : static int
2194 65 : bdev_modules_init(void)
2195 : {
2196 : struct spdk_bdev_module *module;
2197 65 : int rc = 0;
2198 :
2199 254 : TAILQ_FOREACH(module, &g_bdev_mgr.bdev_modules, internal.tailq) {
2200 189 : g_resume_bdev_module = module;
2201 189 : if (module->async_init) {
2202 65 : spdk_spin_lock(&module->internal.spinlock);
2203 65 : module->internal.action_in_progress = 1;
2204 65 : spdk_spin_unlock(&module->internal.spinlock);
2205 65 : }
2206 189 : rc = module->module_init();
2207 189 : if (rc != 0) {
2208 : /* Bump action_in_progress to prevent other modules from completion of modules_init
2209 : * Send message to defer application shutdown until resources are cleaned up */
2210 0 : spdk_spin_lock(&module->internal.spinlock);
2211 0 : module->internal.action_in_progress = 1;
2212 0 : spdk_spin_unlock(&module->internal.spinlock);
2213 0 : spdk_thread_send_msg(spdk_get_thread(), bdev_init_failed, module);
2214 0 : return rc;
2215 : }
2216 189 : }
2217 :
2218 65 : g_resume_bdev_module = NULL;
2219 65 : return 0;
2220 65 : }
2221 :
2222 : void
2223 65 : spdk_bdev_initialize(spdk_bdev_init_cb cb_fn, void *cb_arg)
2224 : {
2225 65 : int rc = 0;
2226 : char mempool_name[32];
2227 :
2228 65 : assert(cb_fn != NULL);
2229 :
2230 65 : g_init_cb_fn = cb_fn;
2231 65 : g_init_cb_arg = cb_arg;
2232 :
2233 65 : spdk_notify_type_register("bdev_register");
2234 65 : spdk_notify_type_register("bdev_unregister");
2235 :
2236 65 : snprintf(mempool_name, sizeof(mempool_name), "bdev_io_%d", getpid());
2237 :
2238 65 : rc = spdk_iobuf_register_module("bdev");
2239 65 : if (rc != 0) {
2240 0 : SPDK_ERRLOG("could not register bdev iobuf module: %s\n", spdk_strerror(-rc));
2241 0 : bdev_init_complete(-1);
2242 0 : return;
2243 : }
2244 :
2245 130 : g_bdev_mgr.bdev_io_pool = spdk_mempool_create(mempool_name,
2246 65 : g_bdev_opts.bdev_io_pool_size,
2247 65 : sizeof(struct spdk_bdev_io) +
2248 65 : bdev_module_get_max_ctx_size(),
2249 : 0,
2250 : SPDK_ENV_NUMA_ID_ANY);
2251 :
2252 65 : if (g_bdev_mgr.bdev_io_pool == NULL) {
2253 0 : SPDK_ERRLOG("could not allocate spdk_bdev_io pool\n");
2254 0 : bdev_init_complete(-1);
2255 0 : return;
2256 : }
2257 :
2258 65 : g_bdev_mgr.zero_buffer = spdk_zmalloc(ZERO_BUFFER_SIZE, ZERO_BUFFER_SIZE,
2259 : NULL, SPDK_ENV_LCORE_ID_ANY, SPDK_MALLOC_DMA);
2260 65 : if (!g_bdev_mgr.zero_buffer) {
2261 0 : SPDK_ERRLOG("create bdev zero buffer failed\n");
2262 0 : bdev_init_complete(-1);
2263 0 : return;
2264 : }
2265 :
2266 : #ifdef SPDK_CONFIG_VTUNE
2267 : g_bdev_mgr.domain = __itt_domain_create("spdk_bdev");
2268 : #endif
2269 :
2270 65 : spdk_io_device_register(&g_bdev_mgr, bdev_mgmt_channel_create,
2271 : bdev_mgmt_channel_destroy,
2272 : sizeof(struct spdk_bdev_mgmt_channel),
2273 : "bdev_mgr");
2274 :
2275 65 : rc = bdev_modules_init();
2276 65 : g_bdev_mgr.module_init_complete = true;
2277 65 : if (rc != 0) {
2278 0 : SPDK_ERRLOG("bdev modules init failed\n");
2279 0 : return;
2280 : }
2281 :
2282 65 : bdev_module_action_complete();
2283 65 : }
2284 :
2285 : static void
2286 65 : bdev_mgr_unregister_cb(void *io_device)
2287 : {
2288 65 : spdk_bdev_fini_cb cb_fn = g_fini_cb_fn;
2289 :
2290 65 : if (g_bdev_mgr.bdev_io_pool) {
2291 65 : if (spdk_mempool_count(g_bdev_mgr.bdev_io_pool) != g_bdev_opts.bdev_io_pool_size) {
2292 0 : SPDK_ERRLOG("bdev IO pool count is %zu but should be %u\n",
2293 : spdk_mempool_count(g_bdev_mgr.bdev_io_pool),
2294 : g_bdev_opts.bdev_io_pool_size);
2295 0 : }
2296 :
2297 65 : spdk_mempool_free(g_bdev_mgr.bdev_io_pool);
2298 65 : }
2299 :
2300 65 : spdk_free(g_bdev_mgr.zero_buffer);
2301 :
2302 65 : bdev_examine_allowlist_free();
2303 :
2304 65 : cb_fn(g_fini_cb_arg);
2305 65 : g_fini_cb_fn = NULL;
2306 65 : g_fini_cb_arg = NULL;
2307 65 : g_bdev_mgr.init_complete = false;
2308 65 : g_bdev_mgr.module_init_complete = false;
2309 65 : }
2310 :
2311 : static void
2312 65 : bdev_module_fini_iter(void *arg)
2313 : {
2314 : struct spdk_bdev_module *bdev_module;
2315 :
2316 : /* FIXME: Handling initialization failures is broken now,
2317 : * so we won't even try cleaning up after successfully
2318 : * initialized modules. if module_init_complete is false,
2319 : * just call spdk_bdev_mgr_unregister_cb
2320 : */
2321 65 : if (!g_bdev_mgr.module_init_complete) {
2322 0 : bdev_mgr_unregister_cb(NULL);
2323 0 : return;
2324 : }
2325 :
2326 : /* Start iterating from the last touched module */
2327 65 : if (!g_resume_bdev_module) {
2328 65 : bdev_module = TAILQ_LAST(&g_bdev_mgr.bdev_modules, bdev_module_list);
2329 65 : } else {
2330 0 : bdev_module = TAILQ_PREV(g_resume_bdev_module, bdev_module_list,
2331 : internal.tailq);
2332 : }
2333 :
2334 254 : while (bdev_module) {
2335 189 : if (bdev_module->async_fini) {
2336 : /* Save our place so we can resume later. We must
2337 : * save the variable here, before calling module_fini()
2338 : * below, because in some cases the module may immediately
2339 : * call spdk_bdev_module_fini_done() and re-enter
2340 : * this function to continue iterating. */
2341 0 : g_resume_bdev_module = bdev_module;
2342 0 : }
2343 :
2344 189 : if (bdev_module->module_fini) {
2345 189 : bdev_module->module_fini();
2346 189 : }
2347 :
2348 189 : if (bdev_module->async_fini) {
2349 0 : return;
2350 : }
2351 :
2352 189 : bdev_module = TAILQ_PREV(bdev_module, bdev_module_list,
2353 : internal.tailq);
2354 : }
2355 :
2356 65 : g_resume_bdev_module = NULL;
2357 65 : spdk_io_device_unregister(&g_bdev_mgr, bdev_mgr_unregister_cb);
2358 65 : }
2359 :
2360 : void
2361 0 : spdk_bdev_module_fini_done(void)
2362 : {
2363 0 : if (spdk_get_thread() != g_fini_thread) {
2364 0 : spdk_thread_send_msg(g_fini_thread, bdev_module_fini_iter, NULL);
2365 0 : } else {
2366 0 : bdev_module_fini_iter(NULL);
2367 : }
2368 0 : }
2369 :
2370 : static void
2371 65 : bdev_finish_unregister_bdevs_iter(void *cb_arg, int bdeverrno)
2372 : {
2373 65 : struct spdk_bdev *bdev = cb_arg;
2374 :
2375 65 : if (bdeverrno && bdev) {
2376 0 : SPDK_WARNLOG("Unable to unregister bdev '%s' during spdk_bdev_finish()\n",
2377 : bdev->name);
2378 :
2379 : /*
2380 : * Since the call to spdk_bdev_unregister() failed, we have no way to free this
2381 : * bdev; try to continue by manually removing this bdev from the list and continue
2382 : * with the next bdev in the list.
2383 : */
2384 0 : TAILQ_REMOVE(&g_bdev_mgr.bdevs, bdev, internal.link);
2385 0 : }
2386 :
2387 65 : if (TAILQ_EMPTY(&g_bdev_mgr.bdevs)) {
2388 65 : SPDK_DEBUGLOG(bdev, "Done unregistering bdevs\n");
2389 : /*
2390 : * Bdev module finish need to be deferred as we might be in the middle of some context
2391 : * (like bdev part free) that will use this bdev (or private bdev driver ctx data)
2392 : * after returning.
2393 : */
2394 65 : spdk_thread_send_msg(spdk_get_thread(), bdev_module_fini_iter, NULL);
2395 65 : return;
2396 : }
2397 :
2398 : /*
2399 : * Unregister last unclaimed bdev in the list, to ensure that bdev subsystem
2400 : * shutdown proceeds top-down. The goal is to give virtual bdevs an opportunity
2401 : * to detect clean shutdown as opposed to run-time hot removal of the underlying
2402 : * base bdevs.
2403 : *
2404 : * Also, walk the list in the reverse order.
2405 : */
2406 0 : for (bdev = TAILQ_LAST(&g_bdev_mgr.bdevs, spdk_bdev_list);
2407 0 : bdev; bdev = TAILQ_PREV(bdev, spdk_bdev_list, internal.link)) {
2408 0 : spdk_spin_lock(&bdev->internal.spinlock);
2409 0 : if (bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE) {
2410 0 : LOG_ALREADY_CLAIMED_DEBUG("claimed, skipping", bdev);
2411 0 : spdk_spin_unlock(&bdev->internal.spinlock);
2412 0 : continue;
2413 : }
2414 0 : spdk_spin_unlock(&bdev->internal.spinlock);
2415 :
2416 0 : SPDK_DEBUGLOG(bdev, "Unregistering bdev '%s'\n", bdev->name);
2417 0 : spdk_bdev_unregister(bdev, bdev_finish_unregister_bdevs_iter, bdev);
2418 0 : return;
2419 : }
2420 :
2421 : /*
2422 : * If any bdev fails to unclaim underlying bdev properly, we may face the
2423 : * case of bdev list consisting of claimed bdevs only (if claims are managed
2424 : * correctly, this would mean there's a loop in the claims graph which is
2425 : * clearly impossible). Warn and unregister last bdev on the list then.
2426 : */
2427 0 : for (bdev = TAILQ_LAST(&g_bdev_mgr.bdevs, spdk_bdev_list);
2428 0 : bdev; bdev = TAILQ_PREV(bdev, spdk_bdev_list, internal.link)) {
2429 0 : SPDK_WARNLOG("Unregistering claimed bdev '%s'!\n", bdev->name);
2430 0 : spdk_bdev_unregister(bdev, bdev_finish_unregister_bdevs_iter, bdev);
2431 0 : return;
2432 : }
2433 65 : }
2434 :
2435 : static void
2436 65 : bdev_module_fini_start_iter(void *arg)
2437 : {
2438 : struct spdk_bdev_module *bdev_module;
2439 :
2440 65 : if (!g_resume_bdev_module) {
2441 65 : bdev_module = TAILQ_LAST(&g_bdev_mgr.bdev_modules, bdev_module_list);
2442 65 : } else {
2443 0 : bdev_module = TAILQ_PREV(g_resume_bdev_module, bdev_module_list, internal.tailq);
2444 : }
2445 :
2446 254 : while (bdev_module) {
2447 189 : if (bdev_module->async_fini_start) {
2448 : /* Save our place so we can resume later. We must
2449 : * save the variable here, before calling fini_start()
2450 : * below, because in some cases the module may immediately
2451 : * call spdk_bdev_module_fini_start_done() and re-enter
2452 : * this function to continue iterating. */
2453 0 : g_resume_bdev_module = bdev_module;
2454 0 : }
2455 :
2456 189 : if (bdev_module->fini_start) {
2457 23 : bdev_module->fini_start();
2458 23 : }
2459 :
2460 189 : if (bdev_module->async_fini_start) {
2461 0 : return;
2462 : }
2463 :
2464 189 : bdev_module = TAILQ_PREV(bdev_module, bdev_module_list, internal.tailq);
2465 : }
2466 :
2467 65 : g_resume_bdev_module = NULL;
2468 :
2469 65 : bdev_finish_unregister_bdevs_iter(NULL, 0);
2470 65 : }
2471 :
2472 : void
2473 0 : spdk_bdev_module_fini_start_done(void)
2474 : {
2475 0 : if (spdk_get_thread() != g_fini_thread) {
2476 0 : spdk_thread_send_msg(g_fini_thread, bdev_module_fini_start_iter, NULL);
2477 0 : } else {
2478 0 : bdev_module_fini_start_iter(NULL);
2479 : }
2480 0 : }
2481 :
2482 : static void
2483 65 : bdev_finish_wait_for_examine_done(void *cb_arg)
2484 : {
2485 65 : bdev_module_fini_start_iter(NULL);
2486 65 : }
2487 :
2488 : static void bdev_open_async_fini(void);
2489 :
2490 : void
2491 65 : spdk_bdev_finish(spdk_bdev_fini_cb cb_fn, void *cb_arg)
2492 : {
2493 : int rc;
2494 :
2495 65 : assert(cb_fn != NULL);
2496 :
2497 65 : g_fini_thread = spdk_get_thread();
2498 :
2499 65 : g_fini_cb_fn = cb_fn;
2500 65 : g_fini_cb_arg = cb_arg;
2501 :
2502 65 : bdev_open_async_fini();
2503 :
2504 65 : rc = spdk_bdev_wait_for_examine(bdev_finish_wait_for_examine_done, NULL);
2505 65 : if (rc != 0) {
2506 0 : SPDK_ERRLOG("wait_for_examine failed: %s\n", spdk_strerror(-rc));
2507 0 : bdev_finish_wait_for_examine_done(NULL);
2508 0 : }
2509 65 : }
2510 :
2511 : struct spdk_bdev_io *
2512 697 : bdev_channel_get_io(struct spdk_bdev_channel *channel)
2513 : {
2514 697 : struct spdk_bdev_mgmt_channel *ch = channel->shared_resource->mgmt_ch;
2515 : struct spdk_bdev_io *bdev_io;
2516 :
2517 697 : if (ch->per_thread_cache_count > 0) {
2518 637 : bdev_io = STAILQ_FIRST(&ch->per_thread_cache);
2519 637 : STAILQ_REMOVE_HEAD(&ch->per_thread_cache, internal.buf_link);
2520 637 : ch->per_thread_cache_count--;
2521 697 : } else if (spdk_unlikely(!TAILQ_EMPTY(&ch->io_wait_queue))) {
2522 : /*
2523 : * Don't try to look for bdev_ios in the global pool if there are
2524 : * waiters on bdev_ios - we don't want this caller to jump the line.
2525 : */
2526 0 : bdev_io = NULL;
2527 0 : } else {
2528 60 : bdev_io = spdk_mempool_get(g_bdev_mgr.bdev_io_pool);
2529 : }
2530 :
2531 697 : return bdev_io;
2532 : }
2533 :
2534 : void
2535 691 : spdk_bdev_free_io(struct spdk_bdev_io *bdev_io)
2536 : {
2537 : struct spdk_bdev_mgmt_channel *ch;
2538 :
2539 691 : assert(bdev_io != NULL);
2540 691 : assert(bdev_io->internal.status != SPDK_BDEV_IO_STATUS_PENDING);
2541 :
2542 691 : ch = bdev_io->internal.ch->shared_resource->mgmt_ch;
2543 :
2544 691 : if (bdev_io->internal.f.has_buf) {
2545 16 : bdev_io_put_buf(bdev_io);
2546 16 : }
2547 :
2548 691 : if (ch->per_thread_cache_count < ch->bdev_io_cache_size) {
2549 637 : ch->per_thread_cache_count++;
2550 637 : STAILQ_INSERT_HEAD(&ch->per_thread_cache, bdev_io, internal.buf_link);
2551 641 : while (ch->per_thread_cache_count > 0 && !TAILQ_EMPTY(&ch->io_wait_queue)) {
2552 : struct spdk_bdev_io_wait_entry *entry;
2553 :
2554 4 : entry = TAILQ_FIRST(&ch->io_wait_queue);
2555 4 : TAILQ_REMOVE(&ch->io_wait_queue, entry, link);
2556 4 : entry->cb_fn(entry->cb_arg);
2557 : }
2558 637 : } else {
2559 : /* We should never have a full cache with entries on the io wait queue. */
2560 54 : assert(TAILQ_EMPTY(&ch->io_wait_queue));
2561 54 : spdk_mempool_put(g_bdev_mgr.bdev_io_pool, (void *)bdev_io);
2562 : }
2563 691 : }
2564 :
2565 : static bool
2566 72 : bdev_qos_is_iops_rate_limit(enum spdk_bdev_qos_rate_limit_type limit)
2567 : {
2568 72 : assert(limit != SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES);
2569 :
2570 72 : switch (limit) {
2571 : case SPDK_BDEV_QOS_RW_IOPS_RATE_LIMIT:
2572 18 : return true;
2573 : case SPDK_BDEV_QOS_RW_BPS_RATE_LIMIT:
2574 : case SPDK_BDEV_QOS_R_BPS_RATE_LIMIT:
2575 : case SPDK_BDEV_QOS_W_BPS_RATE_LIMIT:
2576 54 : return false;
2577 0 : case SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES:
2578 : default:
2579 0 : return false;
2580 : }
2581 72 : }
2582 :
2583 : static bool
2584 25 : bdev_qos_io_to_limit(struct spdk_bdev_io *bdev_io)
2585 : {
2586 25 : switch (bdev_io->type) {
2587 : case SPDK_BDEV_IO_TYPE_NVME_IO:
2588 : case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
2589 : case SPDK_BDEV_IO_TYPE_READ:
2590 : case SPDK_BDEV_IO_TYPE_WRITE:
2591 23 : return true;
2592 : case SPDK_BDEV_IO_TYPE_ZCOPY:
2593 0 : if (bdev_io->u.bdev.zcopy.start) {
2594 0 : return true;
2595 : } else {
2596 0 : return false;
2597 : }
2598 : default:
2599 2 : return false;
2600 : }
2601 25 : }
2602 :
2603 : static bool
2604 33 : bdev_is_read_io(struct spdk_bdev_io *bdev_io)
2605 : {
2606 33 : switch (bdev_io->type) {
2607 : case SPDK_BDEV_IO_TYPE_NVME_IO:
2608 : case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
2609 : /* Bit 1 (0x2) set for read operation */
2610 0 : if (bdev_io->u.nvme_passthru.cmd.opc & SPDK_NVME_OPC_READ) {
2611 0 : return true;
2612 : } else {
2613 0 : return false;
2614 : }
2615 : case SPDK_BDEV_IO_TYPE_READ:
2616 30 : return true;
2617 : case SPDK_BDEV_IO_TYPE_ZCOPY:
2618 : /* Populate to read from disk */
2619 0 : if (bdev_io->u.bdev.zcopy.populate) {
2620 0 : return true;
2621 : } else {
2622 0 : return false;
2623 : }
2624 : default:
2625 3 : return false;
2626 : }
2627 33 : }
2628 :
2629 : static uint64_t
2630 43 : bdev_get_io_size_in_byte(struct spdk_bdev_io *bdev_io)
2631 : {
2632 43 : struct spdk_bdev *bdev = bdev_io->bdev;
2633 :
2634 43 : switch (bdev_io->type) {
2635 : case SPDK_BDEV_IO_TYPE_NVME_IO:
2636 : case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
2637 0 : return bdev_io->u.nvme_passthru.nbytes;
2638 : case SPDK_BDEV_IO_TYPE_READ:
2639 : case SPDK_BDEV_IO_TYPE_WRITE:
2640 43 : return bdev_io->u.bdev.num_blocks * bdev->blocklen;
2641 : case SPDK_BDEV_IO_TYPE_ZCOPY:
2642 : /* Track the data in the start phase only */
2643 0 : if (bdev_io->u.bdev.zcopy.start) {
2644 0 : return bdev_io->u.bdev.num_blocks * bdev->blocklen;
2645 : } else {
2646 0 : return 0;
2647 : }
2648 : default:
2649 0 : return 0;
2650 : }
2651 43 : }
2652 :
2653 : static inline bool
2654 64 : bdev_qos_rw_queue_io(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io, uint64_t delta)
2655 : {
2656 : int64_t remaining_this_timeslice;
2657 :
2658 64 : if (!limit->max_per_timeslice) {
2659 : /* The QoS is disabled */
2660 0 : return false;
2661 : }
2662 :
2663 64 : remaining_this_timeslice = __atomic_sub_fetch(&limit->remaining_this_timeslice, delta,
2664 : __ATOMIC_RELAXED);
2665 64 : if (remaining_this_timeslice + (int64_t)delta > 0) {
2666 : /* There was still a quota for this delta -> the IO shouldn't be queued
2667 : *
2668 : * We allow a slight quota overrun here so an IO bigger than the per-timeslice
2669 : * quota can be allowed once a while. Such overrun then taken into account in
2670 : * the QoS poller, where the next timeslice quota is calculated.
2671 : */
2672 59 : return false;
2673 : }
2674 :
2675 : /* There was no quota for this delta -> the IO should be queued
2676 : * The remaining_this_timeslice must be rewinded so it reflects the real
2677 : * amount of IOs or bytes allowed.
2678 : */
2679 5 : __atomic_add_fetch(
2680 5 : &limit->remaining_this_timeslice, delta, __ATOMIC_RELAXED);
2681 5 : return true;
2682 64 : }
2683 :
2684 : static inline void
2685 5 : bdev_qos_rw_rewind_io(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io, uint64_t delta)
2686 : {
2687 5 : __atomic_add_fetch(&limit->remaining_this_timeslice, delta, __ATOMIC_RELAXED);
2688 5 : }
2689 :
2690 : static bool
2691 23 : bdev_qos_rw_iops_queue(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2692 : {
2693 23 : return bdev_qos_rw_queue_io(limit, io, 1);
2694 : }
2695 :
2696 : static void
2697 3 : bdev_qos_rw_iops_rewind_quota(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2698 : {
2699 3 : bdev_qos_rw_rewind_io(limit, io, 1);
2700 3 : }
2701 :
2702 : static bool
2703 41 : bdev_qos_rw_bps_queue(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2704 : {
2705 41 : return bdev_qos_rw_queue_io(limit, io, bdev_get_io_size_in_byte(io));
2706 : }
2707 :
2708 : static void
2709 2 : bdev_qos_rw_bps_rewind_quota(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2710 : {
2711 2 : bdev_qos_rw_rewind_io(limit, io, bdev_get_io_size_in_byte(io));
2712 2 : }
2713 :
2714 : static bool
2715 19 : bdev_qos_r_bps_queue(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2716 : {
2717 19 : if (bdev_is_read_io(io) == false) {
2718 1 : return false;
2719 : }
2720 :
2721 18 : return bdev_qos_rw_bps_queue(limit, io);
2722 19 : }
2723 :
2724 : static void
2725 0 : bdev_qos_r_bps_rewind_quota(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2726 : {
2727 0 : if (bdev_is_read_io(io) != false) {
2728 0 : bdev_qos_rw_rewind_io(limit, io, bdev_get_io_size_in_byte(io));
2729 0 : }
2730 0 : }
2731 :
2732 : static bool
2733 14 : bdev_qos_w_bps_queue(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2734 : {
2735 14 : if (bdev_is_read_io(io) == true) {
2736 12 : return false;
2737 : }
2738 :
2739 2 : return bdev_qos_rw_bps_queue(limit, io);
2740 14 : }
2741 :
2742 : static void
2743 0 : bdev_qos_w_bps_rewind_quota(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2744 : {
2745 0 : if (bdev_is_read_io(io) != true) {
2746 0 : bdev_qos_rw_rewind_io(limit, io, bdev_get_io_size_in_byte(io));
2747 0 : }
2748 0 : }
2749 :
2750 : static void
2751 10 : bdev_qos_set_ops(struct spdk_bdev_qos *qos)
2752 : {
2753 : int i;
2754 :
2755 50 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
2756 40 : if (qos->rate_limits[i].limit == SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
2757 15 : qos->rate_limits[i].queue_io = NULL;
2758 15 : continue;
2759 : }
2760 :
2761 25 : switch (i) {
2762 : case SPDK_BDEV_QOS_RW_IOPS_RATE_LIMIT:
2763 9 : qos->rate_limits[i].queue_io = bdev_qos_rw_iops_queue;
2764 9 : qos->rate_limits[i].rewind_quota = bdev_qos_rw_iops_rewind_quota;
2765 9 : break;
2766 : case SPDK_BDEV_QOS_RW_BPS_RATE_LIMIT:
2767 7 : qos->rate_limits[i].queue_io = bdev_qos_rw_bps_queue;
2768 7 : qos->rate_limits[i].rewind_quota = bdev_qos_rw_bps_rewind_quota;
2769 7 : break;
2770 : case SPDK_BDEV_QOS_R_BPS_RATE_LIMIT:
2771 5 : qos->rate_limits[i].queue_io = bdev_qos_r_bps_queue;
2772 5 : qos->rate_limits[i].rewind_quota = bdev_qos_r_bps_rewind_quota;
2773 5 : break;
2774 : case SPDK_BDEV_QOS_W_BPS_RATE_LIMIT:
2775 4 : qos->rate_limits[i].queue_io = bdev_qos_w_bps_queue;
2776 4 : qos->rate_limits[i].rewind_quota = bdev_qos_w_bps_rewind_quota;
2777 4 : break;
2778 : default:
2779 0 : break;
2780 : }
2781 25 : }
2782 10 : }
2783 :
2784 : static void
2785 6 : _bdev_io_complete_in_submit(struct spdk_bdev_channel *bdev_ch,
2786 : struct spdk_bdev_io *bdev_io,
2787 : enum spdk_bdev_io_status status)
2788 : {
2789 6 : bdev_io->internal.f.in_submit_request = true;
2790 6 : bdev_io_increment_outstanding(bdev_ch, bdev_ch->shared_resource);
2791 6 : spdk_bdev_io_complete(bdev_io, status);
2792 6 : bdev_io->internal.f.in_submit_request = false;
2793 6 : }
2794 :
2795 : static inline void
2796 574 : bdev_io_do_submit(struct spdk_bdev_channel *bdev_ch, struct spdk_bdev_io *bdev_io)
2797 : {
2798 574 : struct spdk_bdev *bdev = bdev_io->bdev;
2799 574 : struct spdk_io_channel *ch = bdev_ch->channel;
2800 574 : struct spdk_bdev_shared_resource *shared_resource = bdev_ch->shared_resource;
2801 :
2802 574 : if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_ABORT)) {
2803 16 : struct spdk_bdev_mgmt_channel *mgmt_channel = shared_resource->mgmt_ch;
2804 16 : struct spdk_bdev_io *bio_to_abort = bdev_io->u.abort.bio_to_abort;
2805 :
2806 16 : if (bdev_abort_queued_io(&shared_resource->nomem_io, bio_to_abort) ||
2807 16 : bdev_abort_buf_io(mgmt_channel, bio_to_abort)) {
2808 0 : _bdev_io_complete_in_submit(bdev_ch, bdev_io,
2809 : SPDK_BDEV_IO_STATUS_SUCCESS);
2810 0 : return;
2811 : }
2812 16 : }
2813 :
2814 574 : if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE &&
2815 : bdev_io->bdev->split_on_write_unit &&
2816 : bdev_io->u.bdev.num_blocks < bdev_io->bdev->write_unit_size)) {
2817 4 : SPDK_ERRLOG("IO num_blocks %lu does not match the write_unit_size %u\n",
2818 : bdev_io->u.bdev.num_blocks, bdev_io->bdev->write_unit_size);
2819 4 : _bdev_io_complete_in_submit(bdev_ch, bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
2820 4 : return;
2821 : }
2822 :
2823 570 : if (spdk_likely(TAILQ_EMPTY(&shared_resource->nomem_io))) {
2824 527 : bdev_io_increment_outstanding(bdev_ch, shared_resource);
2825 527 : bdev_io->internal.f.in_submit_request = true;
2826 527 : bdev_submit_request(bdev, ch, bdev_io);
2827 527 : bdev_io->internal.f.in_submit_request = false;
2828 527 : } else {
2829 43 : bdev_queue_nomem_io_tail(shared_resource, bdev_io, BDEV_IO_RETRY_STATE_SUBMIT);
2830 43 : if (shared_resource->nomem_threshold == 0 && shared_resource->io_outstanding == 0) {
2831 : /* Special case when we have nomem IOs and no outstanding IOs which completions
2832 : * could trigger retry of queued IOs */
2833 0 : bdev_shared_ch_retry_io(shared_resource);
2834 0 : }
2835 : }
2836 574 : }
2837 :
2838 : static bool
2839 25 : bdev_qos_queue_io(struct spdk_bdev_qos *qos, struct spdk_bdev_io *bdev_io)
2840 : {
2841 : int i;
2842 :
2843 25 : if (bdev_qos_io_to_limit(bdev_io) == true) {
2844 100 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
2845 82 : if (!qos->rate_limits[i].queue_io) {
2846 5 : continue;
2847 : }
2848 :
2849 231 : if (qos->rate_limits[i].queue_io(&qos->rate_limits[i],
2850 154 : bdev_io) == true) {
2851 10 : for (i -= 1; i >= 0 ; i--) {
2852 5 : if (!qos->rate_limits[i].queue_io) {
2853 0 : continue;
2854 : }
2855 :
2856 5 : qos->rate_limits[i].rewind_quota(&qos->rate_limits[i], bdev_io);
2857 5 : }
2858 5 : return true;
2859 : }
2860 72 : }
2861 18 : }
2862 :
2863 20 : return false;
2864 25 : }
2865 :
2866 : static int
2867 27 : bdev_qos_io_submit(struct spdk_bdev_channel *ch, struct spdk_bdev_qos *qos)
2868 : {
2869 27 : struct spdk_bdev_io *bdev_io = NULL, *tmp = NULL;
2870 27 : int submitted_ios = 0;
2871 :
2872 52 : TAILQ_FOREACH_SAFE(bdev_io, &ch->qos_queued_io, internal.link, tmp) {
2873 25 : if (!bdev_qos_queue_io(qos, bdev_io)) {
2874 20 : TAILQ_REMOVE(&ch->qos_queued_io, bdev_io, internal.link);
2875 20 : bdev_io_do_submit(ch, bdev_io);
2876 :
2877 20 : submitted_ios++;
2878 20 : }
2879 25 : }
2880 :
2881 27 : return submitted_ios;
2882 : }
2883 :
2884 : static void
2885 2 : bdev_queue_io_wait_with_cb(struct spdk_bdev_io *bdev_io, spdk_bdev_io_wait_cb cb_fn)
2886 : {
2887 : int rc;
2888 :
2889 2 : bdev_io->internal.waitq_entry.bdev = bdev_io->bdev;
2890 2 : bdev_io->internal.waitq_entry.cb_fn = cb_fn;
2891 2 : bdev_io->internal.waitq_entry.cb_arg = bdev_io;
2892 4 : rc = spdk_bdev_queue_io_wait(bdev_io->bdev, spdk_io_channel_from_ctx(bdev_io->internal.ch),
2893 2 : &bdev_io->internal.waitq_entry);
2894 2 : if (rc != 0) {
2895 0 : SPDK_ERRLOG("Queue IO failed, rc=%d\n", rc);
2896 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
2897 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
2898 0 : }
2899 2 : }
2900 :
2901 : static bool
2902 621 : bdev_rw_should_split(struct spdk_bdev_io *bdev_io)
2903 : {
2904 : uint32_t io_boundary;
2905 621 : struct spdk_bdev *bdev = bdev_io->bdev;
2906 621 : uint32_t max_segment_size = bdev->max_segment_size;
2907 621 : uint32_t max_size = bdev->max_rw_size;
2908 621 : int max_segs = bdev->max_num_segments;
2909 :
2910 621 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE && bdev->split_on_write_unit) {
2911 24 : io_boundary = bdev->write_unit_size;
2912 621 : } else if (bdev->split_on_optimal_io_boundary) {
2913 168 : io_boundary = bdev->optimal_io_boundary;
2914 168 : } else {
2915 429 : io_boundary = 0;
2916 : }
2917 :
2918 621 : if (spdk_likely(!io_boundary && !max_segs && !max_segment_size && !max_size)) {
2919 243 : return false;
2920 : }
2921 :
2922 378 : if (io_boundary) {
2923 : uint64_t start_stripe, end_stripe;
2924 :
2925 192 : start_stripe = bdev_io->u.bdev.offset_blocks;
2926 192 : end_stripe = start_stripe + bdev_io->u.bdev.num_blocks - 1;
2927 : /* Avoid expensive div operations if possible. These spdk_u32 functions are very cheap. */
2928 192 : if (spdk_likely(spdk_u32_is_pow2(io_boundary))) {
2929 192 : start_stripe >>= spdk_u32log2(io_boundary);
2930 192 : end_stripe >>= spdk_u32log2(io_boundary);
2931 192 : } else {
2932 0 : start_stripe /= io_boundary;
2933 0 : end_stripe /= io_boundary;
2934 : }
2935 :
2936 192 : if (start_stripe != end_stripe) {
2937 75 : return true;
2938 : }
2939 117 : }
2940 :
2941 303 : if (max_segs) {
2942 150 : if (bdev_io->u.bdev.iovcnt > max_segs) {
2943 15 : return true;
2944 : }
2945 135 : }
2946 :
2947 288 : if (max_segment_size) {
2948 470 : for (int i = 0; i < bdev_io->u.bdev.iovcnt; i++) {
2949 346 : if (bdev_io->u.bdev.iovs[i].iov_len > max_segment_size) {
2950 12 : return true;
2951 : }
2952 334 : }
2953 124 : }
2954 :
2955 276 : if (max_size) {
2956 52 : if (bdev_io->u.bdev.num_blocks > max_size) {
2957 7 : return true;
2958 : }
2959 45 : }
2960 :
2961 269 : return false;
2962 621 : }
2963 :
2964 : static bool
2965 24 : bdev_unmap_should_split(struct spdk_bdev_io *bdev_io)
2966 : {
2967 : uint32_t num_unmap_segments;
2968 :
2969 24 : if (!bdev_io->bdev->max_unmap || !bdev_io->bdev->max_unmap_segments) {
2970 3 : return false;
2971 : }
2972 21 : num_unmap_segments = spdk_divide_round_up(bdev_io->u.bdev.num_blocks, bdev_io->bdev->max_unmap);
2973 21 : if (num_unmap_segments > bdev_io->bdev->max_unmap_segments) {
2974 4 : return true;
2975 : }
2976 :
2977 17 : return false;
2978 24 : }
2979 :
2980 : static bool
2981 37 : bdev_write_zeroes_should_split(struct spdk_bdev_io *bdev_io)
2982 : {
2983 37 : if (!bdev_io->bdev->max_write_zeroes) {
2984 4 : return false;
2985 : }
2986 :
2987 33 : if (bdev_io->u.bdev.num_blocks > bdev_io->bdev->max_write_zeroes) {
2988 10 : return true;
2989 : }
2990 :
2991 23 : return false;
2992 37 : }
2993 :
2994 : static bool
2995 30 : bdev_copy_should_split(struct spdk_bdev_io *bdev_io)
2996 : {
2997 30 : if (bdev_io->bdev->max_copy != 0 &&
2998 25 : bdev_io->u.bdev.num_blocks > bdev_io->bdev->max_copy) {
2999 6 : return true;
3000 : }
3001 :
3002 24 : return false;
3003 30 : }
3004 :
3005 : static bool
3006 792 : bdev_io_should_split(struct spdk_bdev_io *bdev_io)
3007 : {
3008 792 : switch (bdev_io->type) {
3009 : case SPDK_BDEV_IO_TYPE_READ:
3010 : case SPDK_BDEV_IO_TYPE_WRITE:
3011 621 : return bdev_rw_should_split(bdev_io);
3012 : case SPDK_BDEV_IO_TYPE_UNMAP:
3013 24 : return bdev_unmap_should_split(bdev_io);
3014 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3015 37 : return bdev_write_zeroes_should_split(bdev_io);
3016 : case SPDK_BDEV_IO_TYPE_COPY:
3017 30 : return bdev_copy_should_split(bdev_io);
3018 : default:
3019 80 : return false;
3020 : }
3021 792 : }
3022 :
3023 : static uint32_t
3024 249 : _to_next_boundary(uint64_t offset, uint32_t boundary)
3025 : {
3026 249 : return (boundary - (offset % boundary));
3027 : }
3028 :
3029 : static void bdev_io_split_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg);
3030 :
3031 : static void _bdev_rw_split(void *_bdev_io);
3032 :
3033 : static void bdev_unmap_split(struct spdk_bdev_io *bdev_io);
3034 :
3035 : static void
3036 0 : _bdev_unmap_split(void *_bdev_io)
3037 : {
3038 0 : return bdev_unmap_split((struct spdk_bdev_io *)_bdev_io);
3039 : }
3040 :
3041 : static void bdev_write_zeroes_split(struct spdk_bdev_io *bdev_io);
3042 :
3043 : static void
3044 0 : _bdev_write_zeroes_split(void *_bdev_io)
3045 : {
3046 0 : return bdev_write_zeroes_split((struct spdk_bdev_io *)_bdev_io);
3047 : }
3048 :
3049 : static void bdev_copy_split(struct spdk_bdev_io *bdev_io);
3050 :
3051 : static void
3052 0 : _bdev_copy_split(void *_bdev_io)
3053 : {
3054 0 : return bdev_copy_split((struct spdk_bdev_io *)_bdev_io);
3055 : }
3056 :
3057 : static int
3058 305 : bdev_io_split_submit(struct spdk_bdev_io *bdev_io, struct iovec *iov, int iovcnt, void *md_buf,
3059 : uint64_t num_blocks, uint64_t *offset, uint64_t *remaining)
3060 : {
3061 : int rc;
3062 : uint64_t current_offset, current_remaining, current_src_offset;
3063 : spdk_bdev_io_wait_cb io_wait_fn;
3064 :
3065 305 : current_offset = *offset;
3066 305 : current_remaining = *remaining;
3067 :
3068 305 : assert(bdev_io->internal.f.split);
3069 :
3070 305 : bdev_io->internal.split.outstanding++;
3071 :
3072 305 : io_wait_fn = _bdev_rw_split;
3073 305 : switch (bdev_io->type) {
3074 : case SPDK_BDEV_IO_TYPE_READ:
3075 196 : assert(bdev_io->u.bdev.accel_sequence == NULL);
3076 392 : rc = bdev_readv_blocks_with_md(bdev_io->internal.desc,
3077 196 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
3078 196 : iov, iovcnt, md_buf, current_offset,
3079 196 : num_blocks,
3080 196 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain : NULL,
3081 196 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain_ctx : NULL,
3082 : NULL,
3083 196 : bdev_io->u.bdev.dif_check_flags,
3084 196 : bdev_io_split_done, bdev_io);
3085 196 : break;
3086 : case SPDK_BDEV_IO_TYPE_WRITE:
3087 50 : assert(bdev_io->u.bdev.accel_sequence == NULL);
3088 100 : rc = bdev_writev_blocks_with_md(bdev_io->internal.desc,
3089 50 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
3090 50 : iov, iovcnt, md_buf, current_offset,
3091 50 : num_blocks,
3092 50 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain : NULL,
3093 50 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain_ctx : NULL,
3094 : NULL,
3095 50 : bdev_io->u.bdev.dif_check_flags,
3096 50 : bdev_io->u.bdev.nvme_cdw12.raw,
3097 50 : bdev_io->u.bdev.nvme_cdw13.raw,
3098 50 : bdev_io_split_done, bdev_io);
3099 50 : break;
3100 : case SPDK_BDEV_IO_TYPE_UNMAP:
3101 17 : io_wait_fn = _bdev_unmap_split;
3102 34 : rc = spdk_bdev_unmap_blocks(bdev_io->internal.desc,
3103 17 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
3104 17 : current_offset, num_blocks,
3105 17 : bdev_io_split_done, bdev_io);
3106 17 : break;
3107 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3108 23 : io_wait_fn = _bdev_write_zeroes_split;
3109 46 : rc = spdk_bdev_write_zeroes_blocks(bdev_io->internal.desc,
3110 23 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
3111 23 : current_offset, num_blocks,
3112 23 : bdev_io_split_done, bdev_io);
3113 23 : break;
3114 : case SPDK_BDEV_IO_TYPE_COPY:
3115 19 : io_wait_fn = _bdev_copy_split;
3116 38 : current_src_offset = bdev_io->u.bdev.copy.src_offset_blocks +
3117 19 : (current_offset - bdev_io->u.bdev.offset_blocks);
3118 38 : rc = spdk_bdev_copy_blocks(bdev_io->internal.desc,
3119 19 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
3120 19 : current_offset, current_src_offset, num_blocks,
3121 19 : bdev_io_split_done, bdev_io);
3122 19 : break;
3123 : default:
3124 0 : assert(false);
3125 : rc = -EINVAL;
3126 : break;
3127 : }
3128 :
3129 305 : if (rc == 0) {
3130 301 : current_offset += num_blocks;
3131 301 : current_remaining -= num_blocks;
3132 301 : bdev_io->internal.split.current_offset_blocks = current_offset;
3133 301 : bdev_io->internal.split.remaining_num_blocks = current_remaining;
3134 301 : *offset = current_offset;
3135 301 : *remaining = current_remaining;
3136 301 : } else {
3137 4 : bdev_io->internal.split.outstanding--;
3138 4 : if (rc == -ENOMEM) {
3139 4 : if (bdev_io->internal.split.outstanding == 0) {
3140 : /* No I/O is outstanding. Hence we should wait here. */
3141 1 : bdev_queue_io_wait_with_cb(bdev_io, io_wait_fn);
3142 1 : }
3143 4 : } else {
3144 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
3145 0 : if (bdev_io->internal.split.outstanding == 0) {
3146 0 : bdev_ch_remove_from_io_submitted(bdev_io);
3147 0 : spdk_trace_record(TRACE_BDEV_IO_DONE, bdev_io->internal.ch->trace_id,
3148 : 0, (uintptr_t)bdev_io, bdev_io->internal.caller_ctx,
3149 : bdev_io->internal.ch->queue_depth);
3150 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
3151 0 : }
3152 : }
3153 : }
3154 :
3155 305 : return rc;
3156 : }
3157 :
3158 : static void
3159 67 : _bdev_rw_split(void *_bdev_io)
3160 : {
3161 : struct iovec *parent_iov, *iov;
3162 67 : struct spdk_bdev_io *bdev_io = _bdev_io;
3163 67 : struct spdk_bdev *bdev = bdev_io->bdev;
3164 : uint64_t parent_offset, current_offset, remaining;
3165 : uint32_t parent_iov_offset, parent_iovcnt, parent_iovpos, child_iovcnt;
3166 : uint32_t to_next_boundary, to_next_boundary_bytes, to_last_block_bytes;
3167 : uint32_t iovcnt, iov_len, child_iovsize;
3168 67 : uint32_t blocklen = bdev->blocklen;
3169 : uint32_t io_boundary;
3170 67 : uint32_t max_segment_size = bdev->max_segment_size;
3171 67 : uint32_t max_child_iovcnt = bdev->max_num_segments;
3172 67 : uint32_t max_size = bdev->max_rw_size;
3173 67 : void *md_buf = NULL;
3174 : int rc;
3175 :
3176 67 : max_size = max_size ? max_size : UINT32_MAX;
3177 67 : max_segment_size = max_segment_size ? max_segment_size : UINT32_MAX;
3178 67 : max_child_iovcnt = max_child_iovcnt ? spdk_min(max_child_iovcnt, SPDK_BDEV_IO_NUM_CHILD_IOV) :
3179 : SPDK_BDEV_IO_NUM_CHILD_IOV;
3180 :
3181 67 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE && bdev->split_on_write_unit) {
3182 5 : io_boundary = bdev->write_unit_size;
3183 67 : } else if (bdev->split_on_optimal_io_boundary) {
3184 40 : io_boundary = bdev->optimal_io_boundary;
3185 40 : } else {
3186 22 : io_boundary = UINT32_MAX;
3187 : }
3188 :
3189 67 : assert(bdev_io->internal.f.split);
3190 :
3191 67 : remaining = bdev_io->internal.split.remaining_num_blocks;
3192 67 : current_offset = bdev_io->internal.split.current_offset_blocks;
3193 67 : parent_offset = bdev_io->u.bdev.offset_blocks;
3194 67 : parent_iov_offset = (current_offset - parent_offset) * blocklen;
3195 67 : parent_iovcnt = bdev_io->u.bdev.iovcnt;
3196 :
3197 420 : for (parent_iovpos = 0; parent_iovpos < parent_iovcnt; parent_iovpos++) {
3198 420 : parent_iov = &bdev_io->u.bdev.iovs[parent_iovpos];
3199 420 : if (parent_iov_offset < parent_iov->iov_len) {
3200 67 : break;
3201 : }
3202 353 : parent_iov_offset -= parent_iov->iov_len;
3203 353 : }
3204 :
3205 67 : child_iovcnt = 0;
3206 573 : while (remaining > 0 && parent_iovpos < parent_iovcnt &&
3207 264 : child_iovcnt < SPDK_BDEV_IO_NUM_CHILD_IOV) {
3208 249 : to_next_boundary = _to_next_boundary(current_offset, io_boundary);
3209 249 : to_next_boundary = spdk_min(remaining, to_next_boundary);
3210 249 : to_next_boundary = spdk_min(max_size, to_next_boundary);
3211 249 : to_next_boundary_bytes = to_next_boundary * blocklen;
3212 :
3213 249 : iov = &bdev_io->child_iov[child_iovcnt];
3214 249 : iovcnt = 0;
3215 :
3216 249 : if (bdev_io->u.bdev.md_buf) {
3217 48 : md_buf = (char *)bdev_io->u.bdev.md_buf +
3218 24 : (current_offset - parent_offset) * spdk_bdev_get_md_size(bdev);
3219 24 : }
3220 :
3221 249 : child_iovsize = spdk_min(SPDK_BDEV_IO_NUM_CHILD_IOV - child_iovcnt, max_child_iovcnt);
3222 1810 : while (to_next_boundary_bytes > 0 && parent_iovpos < parent_iovcnt &&
3223 836 : iovcnt < child_iovsize) {
3224 725 : parent_iov = &bdev_io->u.bdev.iovs[parent_iovpos];
3225 725 : iov_len = parent_iov->iov_len - parent_iov_offset;
3226 :
3227 725 : iov_len = spdk_min(iov_len, max_segment_size);
3228 725 : iov_len = spdk_min(iov_len, to_next_boundary_bytes);
3229 725 : to_next_boundary_bytes -= iov_len;
3230 :
3231 725 : bdev_io->child_iov[child_iovcnt].iov_base = parent_iov->iov_base + parent_iov_offset;
3232 725 : bdev_io->child_iov[child_iovcnt].iov_len = iov_len;
3233 :
3234 725 : if (iov_len < parent_iov->iov_len - parent_iov_offset) {
3235 183 : parent_iov_offset += iov_len;
3236 183 : } else {
3237 542 : parent_iovpos++;
3238 542 : parent_iov_offset = 0;
3239 : }
3240 725 : child_iovcnt++;
3241 725 : iovcnt++;
3242 : }
3243 :
3244 249 : if (to_next_boundary_bytes > 0) {
3245 : /* We had to stop this child I/O early because we ran out of
3246 : * child_iov space or were limited by max_num_segments.
3247 : * Ensure the iovs to be aligned with block size and
3248 : * then adjust to_next_boundary before starting the
3249 : * child I/O.
3250 : */
3251 111 : assert(child_iovcnt == SPDK_BDEV_IO_NUM_CHILD_IOV ||
3252 : iovcnt == child_iovsize);
3253 111 : to_last_block_bytes = to_next_boundary_bytes % blocklen;
3254 111 : if (to_last_block_bytes != 0) {
3255 24 : uint32_t child_iovpos = child_iovcnt - 1;
3256 : /* don't decrease child_iovcnt when it equals to SPDK_BDEV_IO_NUM_CHILD_IOV
3257 : * so the loop will naturally end
3258 : */
3259 :
3260 24 : to_last_block_bytes = blocklen - to_last_block_bytes;
3261 24 : to_next_boundary_bytes += to_last_block_bytes;
3262 53 : while (to_last_block_bytes > 0 && iovcnt > 0) {
3263 32 : iov_len = spdk_min(to_last_block_bytes,
3264 : bdev_io->child_iov[child_iovpos].iov_len);
3265 32 : bdev_io->child_iov[child_iovpos].iov_len -= iov_len;
3266 32 : if (bdev_io->child_iov[child_iovpos].iov_len == 0) {
3267 15 : child_iovpos--;
3268 15 : if (--iovcnt == 0) {
3269 : /* If the child IO is less than a block size just return.
3270 : * If the first child IO of any split round is less than
3271 : * a block size, an error exit.
3272 : */
3273 3 : if (bdev_io->internal.split.outstanding == 0) {
3274 1 : SPDK_ERRLOG("The first child io was less than a block size\n");
3275 1 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
3276 1 : bdev_ch_remove_from_io_submitted(bdev_io);
3277 1 : spdk_trace_record(TRACE_BDEV_IO_DONE, bdev_io->internal.ch->trace_id,
3278 : 0, (uintptr_t)bdev_io, bdev_io->internal.caller_ctx,
3279 : bdev_io->internal.ch->queue_depth);
3280 1 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
3281 1 : }
3282 :
3283 3 : return;
3284 : }
3285 12 : }
3286 :
3287 29 : to_last_block_bytes -= iov_len;
3288 :
3289 29 : if (parent_iov_offset == 0) {
3290 14 : parent_iovpos--;
3291 14 : parent_iov_offset = bdev_io->u.bdev.iovs[parent_iovpos].iov_len;
3292 14 : }
3293 29 : parent_iov_offset -= iov_len;
3294 : }
3295 :
3296 21 : assert(to_last_block_bytes == 0);
3297 21 : }
3298 108 : to_next_boundary -= to_next_boundary_bytes / blocklen;
3299 108 : }
3300 :
3301 246 : rc = bdev_io_split_submit(bdev_io, iov, iovcnt, md_buf, to_next_boundary,
3302 : ¤t_offset, &remaining);
3303 246 : if (spdk_unlikely(rc)) {
3304 4 : return;
3305 : }
3306 : }
3307 67 : }
3308 :
3309 : static void
3310 3 : bdev_unmap_split(struct spdk_bdev_io *bdev_io)
3311 : {
3312 : uint64_t offset, unmap_blocks, remaining, max_unmap_blocks;
3313 3 : uint32_t num_children_reqs = 0;
3314 : int rc;
3315 :
3316 3 : assert(bdev_io->internal.f.split);
3317 :
3318 3 : offset = bdev_io->internal.split.current_offset_blocks;
3319 3 : remaining = bdev_io->internal.split.remaining_num_blocks;
3320 3 : max_unmap_blocks = bdev_io->bdev->max_unmap * bdev_io->bdev->max_unmap_segments;
3321 :
3322 20 : while (remaining && (num_children_reqs < SPDK_BDEV_MAX_CHILDREN_UNMAP_WRITE_ZEROES_REQS)) {
3323 17 : unmap_blocks = spdk_min(remaining, max_unmap_blocks);
3324 :
3325 17 : rc = bdev_io_split_submit(bdev_io, NULL, 0, NULL, unmap_blocks,
3326 : &offset, &remaining);
3327 17 : if (spdk_likely(rc == 0)) {
3328 17 : num_children_reqs++;
3329 17 : } else {
3330 0 : return;
3331 : }
3332 : }
3333 3 : }
3334 :
3335 : static void
3336 6 : bdev_write_zeroes_split(struct spdk_bdev_io *bdev_io)
3337 : {
3338 : uint64_t offset, write_zeroes_blocks, remaining;
3339 6 : uint32_t num_children_reqs = 0;
3340 : int rc;
3341 :
3342 6 : assert(bdev_io->internal.f.split);
3343 :
3344 6 : offset = bdev_io->internal.split.current_offset_blocks;
3345 6 : remaining = bdev_io->internal.split.remaining_num_blocks;
3346 :
3347 29 : while (remaining && (num_children_reqs < SPDK_BDEV_MAX_CHILDREN_UNMAP_WRITE_ZEROES_REQS)) {
3348 23 : write_zeroes_blocks = spdk_min(remaining, bdev_io->bdev->max_write_zeroes);
3349 :
3350 23 : rc = bdev_io_split_submit(bdev_io, NULL, 0, NULL, write_zeroes_blocks,
3351 : &offset, &remaining);
3352 23 : if (spdk_likely(rc == 0)) {
3353 23 : num_children_reqs++;
3354 23 : } else {
3355 0 : return;
3356 : }
3357 : }
3358 6 : }
3359 :
3360 : static void
3361 4 : bdev_copy_split(struct spdk_bdev_io *bdev_io)
3362 : {
3363 : uint64_t offset, copy_blocks, remaining;
3364 4 : uint32_t num_children_reqs = 0;
3365 : int rc;
3366 :
3367 4 : assert(bdev_io->internal.f.split);
3368 :
3369 4 : offset = bdev_io->internal.split.current_offset_blocks;
3370 4 : remaining = bdev_io->internal.split.remaining_num_blocks;
3371 :
3372 4 : assert(bdev_io->bdev->max_copy != 0);
3373 23 : while (remaining && (num_children_reqs < SPDK_BDEV_MAX_CHILDREN_COPY_REQS)) {
3374 19 : copy_blocks = spdk_min(remaining, bdev_io->bdev->max_copy);
3375 :
3376 19 : rc = bdev_io_split_submit(bdev_io, NULL, 0, NULL, copy_blocks,
3377 : &offset, &remaining);
3378 19 : if (spdk_likely(rc == 0)) {
3379 19 : num_children_reqs++;
3380 19 : } else {
3381 0 : return;
3382 : }
3383 : }
3384 4 : }
3385 :
3386 : static void
3387 58 : parent_bdev_io_complete(void *ctx, int rc)
3388 : {
3389 58 : struct spdk_bdev_io *parent_io = ctx;
3390 :
3391 58 : if (rc) {
3392 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
3393 0 : }
3394 :
3395 116 : parent_io->internal.cb(parent_io, parent_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS,
3396 58 : parent_io->internal.caller_ctx);
3397 58 : }
3398 :
3399 : static void
3400 0 : bdev_io_complete_parent_sequence_cb(void *ctx, int status)
3401 : {
3402 0 : struct spdk_bdev_io *bdev_io = ctx;
3403 :
3404 : /* u.bdev.accel_sequence should have already been cleared at this point */
3405 0 : assert(bdev_io->u.bdev.accel_sequence == NULL);
3406 0 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS);
3407 0 : bdev_io->internal.f.has_accel_sequence = false;
3408 :
3409 0 : if (spdk_unlikely(status != 0)) {
3410 0 : SPDK_ERRLOG("Failed to execute accel sequence, status=%d\n", status);
3411 0 : }
3412 :
3413 0 : parent_bdev_io_complete(bdev_io, status);
3414 0 : }
3415 :
3416 : static void
3417 301 : bdev_io_split_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
3418 : {
3419 301 : struct spdk_bdev_io *parent_io = cb_arg;
3420 :
3421 301 : spdk_bdev_free_io(bdev_io);
3422 :
3423 301 : assert(parent_io->internal.f.split);
3424 :
3425 301 : if (!success) {
3426 21 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
3427 : /* If any child I/O failed, stop further splitting process. */
3428 21 : parent_io->internal.split.current_offset_blocks += parent_io->internal.split.remaining_num_blocks;
3429 21 : parent_io->internal.split.remaining_num_blocks = 0;
3430 21 : }
3431 301 : parent_io->internal.split.outstanding--;
3432 301 : if (parent_io->internal.split.outstanding != 0) {
3433 223 : return;
3434 : }
3435 :
3436 : /*
3437 : * Parent I/O finishes when all blocks are consumed.
3438 : */
3439 78 : if (parent_io->internal.split.remaining_num_blocks == 0) {
3440 58 : assert(parent_io->internal.cb != bdev_io_split_done);
3441 58 : bdev_ch_remove_from_io_submitted(parent_io);
3442 58 : spdk_trace_record(TRACE_BDEV_IO_DONE, parent_io->internal.ch->trace_id,
3443 : 0, (uintptr_t)parent_io, bdev_io->internal.caller_ctx,
3444 : parent_io->internal.ch->queue_depth);
3445 :
3446 58 : if (spdk_likely(parent_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS)) {
3447 48 : if (bdev_io_needs_sequence_exec(parent_io->internal.desc, parent_io)) {
3448 0 : bdev_io_exec_sequence(parent_io, bdev_io_complete_parent_sequence_cb);
3449 0 : return;
3450 48 : } else if (parent_io->internal.f.has_bounce_buf &&
3451 0 : !bdev_io_use_accel_sequence(bdev_io)) {
3452 : /* bdev IO will be completed in the callback */
3453 0 : _bdev_io_push_bounce_data_buffer(parent_io, parent_bdev_io_complete);
3454 0 : return;
3455 : }
3456 48 : }
3457 :
3458 58 : parent_bdev_io_complete(parent_io, 0);
3459 58 : return;
3460 : }
3461 :
3462 : /*
3463 : * Continue with the splitting process. This function will complete the parent I/O if the
3464 : * splitting is done.
3465 : */
3466 20 : switch (parent_io->type) {
3467 : case SPDK_BDEV_IO_TYPE_READ:
3468 : case SPDK_BDEV_IO_TYPE_WRITE:
3469 17 : _bdev_rw_split(parent_io);
3470 17 : break;
3471 : case SPDK_BDEV_IO_TYPE_UNMAP:
3472 1 : bdev_unmap_split(parent_io);
3473 1 : break;
3474 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3475 1 : bdev_write_zeroes_split(parent_io);
3476 1 : break;
3477 : case SPDK_BDEV_IO_TYPE_COPY:
3478 1 : bdev_copy_split(parent_io);
3479 1 : break;
3480 : default:
3481 0 : assert(false);
3482 : break;
3483 : }
3484 301 : }
3485 :
3486 : static void bdev_rw_split_get_buf_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io,
3487 : bool success);
3488 :
3489 : static void
3490 59 : bdev_io_split(struct spdk_bdev_io *bdev_io)
3491 : {
3492 59 : assert(bdev_io_should_split(bdev_io));
3493 59 : assert(bdev_io->internal.f.split);
3494 :
3495 59 : bdev_io->internal.split.current_offset_blocks = bdev_io->u.bdev.offset_blocks;
3496 59 : bdev_io->internal.split.remaining_num_blocks = bdev_io->u.bdev.num_blocks;
3497 59 : bdev_io->internal.split.outstanding = 0;
3498 59 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
3499 :
3500 59 : switch (bdev_io->type) {
3501 : case SPDK_BDEV_IO_TYPE_READ:
3502 : case SPDK_BDEV_IO_TYPE_WRITE:
3503 49 : if (_is_buf_allocated(bdev_io->u.bdev.iovs)) {
3504 49 : _bdev_rw_split(bdev_io);
3505 49 : } else {
3506 0 : assert(bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
3507 0 : spdk_bdev_io_get_buf(bdev_io, bdev_rw_split_get_buf_cb,
3508 0 : bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen);
3509 : }
3510 49 : break;
3511 : case SPDK_BDEV_IO_TYPE_UNMAP:
3512 2 : bdev_unmap_split(bdev_io);
3513 2 : break;
3514 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3515 5 : bdev_write_zeroes_split(bdev_io);
3516 5 : break;
3517 : case SPDK_BDEV_IO_TYPE_COPY:
3518 3 : bdev_copy_split(bdev_io);
3519 3 : break;
3520 : default:
3521 0 : assert(false);
3522 : break;
3523 : }
3524 59 : }
3525 :
3526 : static void
3527 0 : bdev_rw_split_get_buf_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io, bool success)
3528 : {
3529 0 : if (!success) {
3530 0 : spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
3531 0 : return;
3532 : }
3533 :
3534 0 : _bdev_rw_split(bdev_io);
3535 0 : }
3536 :
3537 : static inline void
3538 579 : _bdev_io_submit(struct spdk_bdev_io *bdev_io)
3539 : {
3540 579 : struct spdk_bdev *bdev = bdev_io->bdev;
3541 579 : struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
3542 :
3543 579 : if (spdk_likely(bdev_ch->flags == 0)) {
3544 554 : bdev_io_do_submit(bdev_ch, bdev_io);
3545 554 : return;
3546 : }
3547 :
3548 25 : if (bdev_ch->flags & BDEV_CH_RESET_IN_PROGRESS) {
3549 2 : _bdev_io_complete_in_submit(bdev_ch, bdev_io, SPDK_BDEV_IO_STATUS_ABORTED);
3550 25 : } else if (bdev_ch->flags & BDEV_CH_QOS_ENABLED) {
3551 23 : if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_ABORT) &&
3552 2 : bdev_abort_queued_io(&bdev_ch->qos_queued_io, bdev_io->u.abort.bio_to_abort)) {
3553 0 : _bdev_io_complete_in_submit(bdev_ch, bdev_io, SPDK_BDEV_IO_STATUS_SUCCESS);
3554 0 : } else {
3555 23 : TAILQ_INSERT_TAIL(&bdev_ch->qos_queued_io, bdev_io, internal.link);
3556 23 : bdev_qos_io_submit(bdev_ch, bdev->internal.qos);
3557 : }
3558 23 : } else {
3559 0 : SPDK_ERRLOG("unknown bdev_ch flag %x found\n", bdev_ch->flags);
3560 0 : _bdev_io_complete_in_submit(bdev_ch, bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
3561 : }
3562 579 : }
3563 :
3564 : bool bdev_lba_range_overlapped(struct lba_range *range1, struct lba_range *range2);
3565 :
3566 : bool
3567 23 : bdev_lba_range_overlapped(struct lba_range *range1, struct lba_range *range2)
3568 : {
3569 23 : if (range1->length == 0 || range2->length == 0) {
3570 1 : return false;
3571 : }
3572 :
3573 22 : if (range1->offset + range1->length <= range2->offset) {
3574 1 : return false;
3575 : }
3576 :
3577 21 : if (range2->offset + range2->length <= range1->offset) {
3578 3 : return false;
3579 : }
3580 :
3581 18 : return true;
3582 23 : }
3583 :
3584 : static bool
3585 11 : bdev_io_range_is_locked(struct spdk_bdev_io *bdev_io, struct lba_range *range)
3586 : {
3587 11 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
3588 : struct lba_range r;
3589 :
3590 11 : switch (bdev_io->type) {
3591 : case SPDK_BDEV_IO_TYPE_NVME_IO:
3592 : case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
3593 : /* Don't try to decode the NVMe command - just assume worst-case and that
3594 : * it overlaps a locked range.
3595 : */
3596 0 : return true;
3597 : case SPDK_BDEV_IO_TYPE_READ:
3598 6 : if (!range->quiesce) {
3599 4 : return false;
3600 : }
3601 : /* fallthrough */
3602 : case SPDK_BDEV_IO_TYPE_WRITE:
3603 : case SPDK_BDEV_IO_TYPE_UNMAP:
3604 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3605 : case SPDK_BDEV_IO_TYPE_ZCOPY:
3606 : case SPDK_BDEV_IO_TYPE_COPY:
3607 7 : r.offset = bdev_io->u.bdev.offset_blocks;
3608 7 : r.length = bdev_io->u.bdev.num_blocks;
3609 7 : if (!bdev_lba_range_overlapped(range, &r)) {
3610 : /* This I/O doesn't overlap the specified LBA range. */
3611 0 : return false;
3612 7 : } else if (range->owner_ch == ch && range->locked_ctx == bdev_io->internal.caller_ctx) {
3613 : /* This I/O overlaps, but the I/O is on the same channel that locked this
3614 : * range, and the caller_ctx is the same as the locked_ctx. This means
3615 : * that this I/O is associated with the lock, and is allowed to execute.
3616 : */
3617 2 : return false;
3618 : } else {
3619 5 : return true;
3620 : }
3621 : default:
3622 0 : return false;
3623 : }
3624 11 : }
3625 :
3626 : void
3627 639 : bdev_io_submit(struct spdk_bdev_io *bdev_io)
3628 : {
3629 639 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
3630 :
3631 639 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_PENDING);
3632 :
3633 639 : if (!TAILQ_EMPTY(&ch->locked_ranges)) {
3634 : struct lba_range *range;
3635 :
3636 13 : TAILQ_FOREACH(range, &ch->locked_ranges, tailq) {
3637 8 : if (bdev_io_range_is_locked(bdev_io, range)) {
3638 3 : TAILQ_INSERT_TAIL(&ch->io_locked, bdev_io, internal.ch_link);
3639 3 : return;
3640 : }
3641 5 : }
3642 5 : }
3643 :
3644 636 : bdev_ch_add_to_io_submitted(bdev_io);
3645 :
3646 636 : bdev_io->internal.submit_tsc = spdk_get_ticks();
3647 636 : spdk_trace_record_tsc(bdev_io->internal.submit_tsc, TRACE_BDEV_IO_START,
3648 : ch->trace_id, bdev_io->u.bdev.num_blocks,
3649 : (uintptr_t)bdev_io, (uint64_t)bdev_io->type, bdev_io->internal.caller_ctx,
3650 : bdev_io->u.bdev.offset_blocks, ch->queue_depth);
3651 :
3652 636 : if (bdev_io->internal.f.split) {
3653 59 : bdev_io_split(bdev_io);
3654 59 : return;
3655 : }
3656 :
3657 577 : _bdev_io_submit(bdev_io);
3658 639 : }
3659 :
3660 : static inline void
3661 4 : _bdev_io_ext_use_bounce_buffer(struct spdk_bdev_io *bdev_io)
3662 : {
3663 : /* bdev doesn't support memory domains, thereby buffers in this IO request can't
3664 : * be accessed directly. It is needed to allocate buffers before issuing IO operation.
3665 : * For write operation we need to pull buffers from memory domain before submitting IO.
3666 : * Once read operation completes, we need to use memory_domain push functionality to
3667 : * update data in original memory domain IO buffer
3668 : * This IO request will go through a regular IO flow, so clear memory domains pointers */
3669 4 : assert(bdev_io->internal.f.has_memory_domain);
3670 4 : bdev_io->u.bdev.memory_domain = NULL;
3671 4 : bdev_io->u.bdev.memory_domain_ctx = NULL;
3672 8 : _bdev_memory_domain_io_get_buf(bdev_io, _bdev_memory_domain_get_io_cb,
3673 4 : bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen);
3674 4 : }
3675 :
3676 : static inline void
3677 292 : _bdev_io_submit_ext(struct spdk_bdev_desc *desc, struct spdk_bdev_io *bdev_io)
3678 : {
3679 292 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
3680 292 : bool needs_exec = bdev_io_needs_sequence_exec(desc, bdev_io);
3681 :
3682 292 : if (spdk_unlikely(ch->flags & BDEV_CH_RESET_IN_PROGRESS)) {
3683 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_ABORTED;
3684 0 : bdev_io_complete_unsubmitted(bdev_io);
3685 0 : return;
3686 : }
3687 :
3688 : /* We need to allocate bounce buffer if bdev doesn't support memory domains, or if it does
3689 : * support them, but we need to execute an accel sequence and the data buffer is from accel
3690 : * memory domain (to avoid doing a push/pull from that domain).
3691 : */
3692 292 : if (bdev_io_use_memory_domain(bdev_io)) {
3693 4 : if (!desc->memory_domains_supported ||
3694 0 : (needs_exec && bdev_io->internal.memory_domain == spdk_accel_get_memory_domain())) {
3695 4 : _bdev_io_ext_use_bounce_buffer(bdev_io);
3696 4 : return;
3697 : }
3698 0 : }
3699 :
3700 288 : if (needs_exec) {
3701 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
3702 0 : bdev_io_exec_sequence(bdev_io, bdev_io_submit_sequence_cb);
3703 0 : return;
3704 : }
3705 : /* For reads we'll execute the sequence after the data is read, so, for now, only
3706 : * clear out accel_sequence pointer and submit the IO */
3707 0 : assert(bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
3708 0 : bdev_io->u.bdev.accel_sequence = NULL;
3709 0 : }
3710 :
3711 288 : bdev_io_submit(bdev_io);
3712 292 : }
3713 :
3714 : static void
3715 11 : bdev_io_submit_reset(struct spdk_bdev_io *bdev_io)
3716 : {
3717 11 : struct spdk_bdev *bdev = bdev_io->bdev;
3718 11 : struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
3719 11 : struct spdk_io_channel *ch = bdev_ch->channel;
3720 :
3721 11 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_PENDING);
3722 :
3723 11 : bdev_io->internal.f.in_submit_request = true;
3724 11 : bdev_submit_request(bdev, ch, bdev_io);
3725 11 : bdev_io->internal.f.in_submit_request = false;
3726 11 : }
3727 :
3728 : void
3729 691 : bdev_io_init(struct spdk_bdev_io *bdev_io,
3730 : struct spdk_bdev *bdev, void *cb_arg,
3731 : spdk_bdev_io_completion_cb cb)
3732 : {
3733 691 : bdev_io->bdev = bdev;
3734 691 : bdev_io->internal.f.raw = 0;
3735 691 : bdev_io->internal.caller_ctx = cb_arg;
3736 691 : bdev_io->internal.cb = cb;
3737 691 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_PENDING;
3738 691 : bdev_io->internal.f.in_submit_request = false;
3739 691 : bdev_io->internal.error.nvme.cdw0 = 0;
3740 691 : bdev_io->num_retries = 0;
3741 691 : bdev_io->internal.get_buf_cb = NULL;
3742 691 : bdev_io->internal.get_aux_buf_cb = NULL;
3743 691 : bdev_io->internal.data_transfer_cpl = NULL;
3744 691 : bdev_io->internal.f.split = bdev_io_should_split(bdev_io);
3745 691 : }
3746 :
3747 : static bool
3748 520 : bdev_io_type_supported(struct spdk_bdev *bdev, enum spdk_bdev_io_type io_type)
3749 : {
3750 520 : return bdev->fn_table->io_type_supported(bdev->ctxt, io_type);
3751 : }
3752 :
3753 : bool
3754 170 : spdk_bdev_io_type_supported(struct spdk_bdev *bdev, enum spdk_bdev_io_type io_type)
3755 : {
3756 : bool supported;
3757 :
3758 170 : supported = bdev_io_type_supported(bdev, io_type);
3759 :
3760 170 : if (!supported) {
3761 7 : switch (io_type) {
3762 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3763 : /* The bdev layer will emulate write zeroes as long as write is supported. */
3764 0 : supported = bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE);
3765 0 : break;
3766 : default:
3767 7 : break;
3768 : }
3769 7 : }
3770 :
3771 170 : return supported;
3772 : }
3773 :
3774 : static const char *g_io_type_strings[] = {
3775 : [SPDK_BDEV_IO_TYPE_READ] = "read",
3776 : [SPDK_BDEV_IO_TYPE_WRITE] = "write",
3777 : [SPDK_BDEV_IO_TYPE_UNMAP] = "unmap",
3778 : [SPDK_BDEV_IO_TYPE_FLUSH] = "flush",
3779 : [SPDK_BDEV_IO_TYPE_RESET] = "reset",
3780 : [SPDK_BDEV_IO_TYPE_NVME_ADMIN] = "nvme_admin",
3781 : [SPDK_BDEV_IO_TYPE_NVME_IO] = "nvme_io",
3782 : [SPDK_BDEV_IO_TYPE_NVME_IO_MD] = "nvme_io_md",
3783 : [SPDK_BDEV_IO_TYPE_WRITE_ZEROES] = "write_zeroes",
3784 : [SPDK_BDEV_IO_TYPE_ZCOPY] = "zcopy",
3785 : [SPDK_BDEV_IO_TYPE_GET_ZONE_INFO] = "get_zone_info",
3786 : [SPDK_BDEV_IO_TYPE_ZONE_MANAGEMENT] = "zone_management",
3787 : [SPDK_BDEV_IO_TYPE_ZONE_APPEND] = "zone_append",
3788 : [SPDK_BDEV_IO_TYPE_COMPARE] = "compare",
3789 : [SPDK_BDEV_IO_TYPE_COMPARE_AND_WRITE] = "compare_and_write",
3790 : [SPDK_BDEV_IO_TYPE_ABORT] = "abort",
3791 : [SPDK_BDEV_IO_TYPE_SEEK_HOLE] = "seek_hole",
3792 : [SPDK_BDEV_IO_TYPE_SEEK_DATA] = "seek_data",
3793 : [SPDK_BDEV_IO_TYPE_COPY] = "copy",
3794 : [SPDK_BDEV_IO_TYPE_NVME_IOV_MD] = "nvme_iov_md",
3795 : };
3796 :
3797 : const char *
3798 0 : spdk_bdev_get_io_type_name(enum spdk_bdev_io_type io_type)
3799 : {
3800 0 : if (io_type <= SPDK_BDEV_IO_TYPE_INVALID || io_type >= SPDK_BDEV_NUM_IO_TYPES) {
3801 0 : return NULL;
3802 : }
3803 :
3804 0 : return g_io_type_strings[io_type];
3805 0 : }
3806 :
3807 : int
3808 0 : spdk_bdev_get_io_type(const char *io_type_string)
3809 : {
3810 : int i;
3811 :
3812 0 : for (i = SPDK_BDEV_IO_TYPE_READ; i < SPDK_BDEV_NUM_IO_TYPES; ++i) {
3813 0 : if (!strcmp(io_type_string, g_io_type_strings[i])) {
3814 0 : return i;
3815 : }
3816 0 : }
3817 :
3818 0 : return -1;
3819 0 : }
3820 :
3821 : uint64_t
3822 0 : spdk_bdev_io_get_submit_tsc(struct spdk_bdev_io *bdev_io)
3823 : {
3824 0 : return bdev_io->internal.submit_tsc;
3825 : }
3826 :
3827 : int
3828 0 : spdk_bdev_dump_info_json(struct spdk_bdev *bdev, struct spdk_json_write_ctx *w)
3829 : {
3830 0 : if (bdev->fn_table->dump_info_json) {
3831 0 : return bdev->fn_table->dump_info_json(bdev->ctxt, w);
3832 : }
3833 :
3834 0 : return 0;
3835 0 : }
3836 :
3837 : static void
3838 10 : bdev_qos_update_max_quota_per_timeslice(struct spdk_bdev_qos *qos)
3839 : {
3840 10 : uint32_t max_per_timeslice = 0;
3841 : int i;
3842 :
3843 50 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
3844 40 : if (qos->rate_limits[i].limit == SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
3845 15 : qos->rate_limits[i].max_per_timeslice = 0;
3846 15 : continue;
3847 : }
3848 :
3849 50 : max_per_timeslice = qos->rate_limits[i].limit *
3850 25 : SPDK_BDEV_QOS_TIMESLICE_IN_USEC / SPDK_SEC_TO_USEC;
3851 :
3852 25 : qos->rate_limits[i].max_per_timeslice = spdk_max(max_per_timeslice,
3853 : qos->rate_limits[i].min_per_timeslice);
3854 :
3855 50 : __atomic_store_n(&qos->rate_limits[i].remaining_this_timeslice,
3856 25 : qos->rate_limits[i].max_per_timeslice, __ATOMIC_RELEASE);
3857 25 : }
3858 :
3859 10 : bdev_qos_set_ops(qos);
3860 10 : }
3861 :
3862 : static void
3863 4 : bdev_channel_submit_qos_io(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
3864 : struct spdk_io_channel *io_ch, void *ctx)
3865 : {
3866 4 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
3867 : int status;
3868 :
3869 4 : bdev_qos_io_submit(bdev_ch, bdev->internal.qos);
3870 :
3871 : /* if all IOs were sent then continue the iteration, otherwise - stop it */
3872 : /* TODO: channels round robing */
3873 4 : status = TAILQ_EMPTY(&bdev_ch->qos_queued_io) ? 0 : 1;
3874 :
3875 4 : spdk_bdev_for_each_channel_continue(i, status);
3876 4 : }
3877 :
3878 :
3879 : static void
3880 2 : bdev_channel_submit_qos_io_done(struct spdk_bdev *bdev, void *ctx, int status)
3881 : {
3882 :
3883 2 : }
3884 :
3885 : static int
3886 3 : bdev_channel_poll_qos(void *arg)
3887 : {
3888 3 : struct spdk_bdev *bdev = arg;
3889 3 : struct spdk_bdev_qos *qos = bdev->internal.qos;
3890 3 : uint64_t now = spdk_get_ticks();
3891 : int i;
3892 : int64_t remaining_last_timeslice;
3893 :
3894 3 : if (spdk_unlikely(qos->thread == NULL)) {
3895 : /* Old QoS was unbound to remove and new QoS is not enabled yet. */
3896 1 : return SPDK_POLLER_IDLE;
3897 : }
3898 :
3899 2 : if (now < (qos->last_timeslice + qos->timeslice_size)) {
3900 : /* We received our callback earlier than expected - return
3901 : * immediately and wait to do accounting until at least one
3902 : * timeslice has actually expired. This should never happen
3903 : * with a well-behaved timer implementation.
3904 : */
3905 0 : return SPDK_POLLER_IDLE;
3906 : }
3907 :
3908 : /* Reset for next round of rate limiting */
3909 10 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
3910 : /* We may have allowed the IOs or bytes to slightly overrun in the last
3911 : * timeslice. remaining_this_timeslice is signed, so if it's negative
3912 : * here, we'll account for the overrun so that the next timeslice will
3913 : * be appropriately reduced.
3914 : */
3915 8 : remaining_last_timeslice = __atomic_exchange_n(&qos->rate_limits[i].remaining_this_timeslice,
3916 : 0, __ATOMIC_RELAXED);
3917 8 : if (remaining_last_timeslice < 0) {
3918 : /* There could be a race condition here as both bdev_qos_rw_queue_io() and bdev_channel_poll_qos()
3919 : * potentially use 2 atomic ops each, so they can intertwine.
3920 : * This race can potentially cause the limits to be a little fuzzy but won't cause any real damage.
3921 : */
3922 0 : __atomic_store_n(&qos->rate_limits[i].remaining_this_timeslice,
3923 0 : remaining_last_timeslice, __ATOMIC_RELAXED);
3924 0 : }
3925 8 : }
3926 :
3927 4 : while (now >= (qos->last_timeslice + qos->timeslice_size)) {
3928 2 : qos->last_timeslice += qos->timeslice_size;
3929 10 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
3930 16 : __atomic_add_fetch(&qos->rate_limits[i].remaining_this_timeslice,
3931 8 : qos->rate_limits[i].max_per_timeslice, __ATOMIC_RELAXED);
3932 8 : }
3933 : }
3934 :
3935 2 : spdk_bdev_for_each_channel(bdev, bdev_channel_submit_qos_io, qos,
3936 : bdev_channel_submit_qos_io_done);
3937 :
3938 2 : return SPDK_POLLER_BUSY;
3939 3 : }
3940 :
3941 : static void
3942 74 : bdev_channel_destroy_resource(struct spdk_bdev_channel *ch)
3943 : {
3944 : struct spdk_bdev_shared_resource *shared_resource;
3945 : struct lba_range *range;
3946 :
3947 74 : bdev_free_io_stat(ch->stat);
3948 : #ifdef SPDK_CONFIG_VTUNE
3949 : bdev_free_io_stat(ch->prev_stat);
3950 : #endif
3951 :
3952 74 : while (!TAILQ_EMPTY(&ch->locked_ranges)) {
3953 0 : range = TAILQ_FIRST(&ch->locked_ranges);
3954 0 : TAILQ_REMOVE(&ch->locked_ranges, range, tailq);
3955 0 : free(range);
3956 : }
3957 :
3958 74 : spdk_put_io_channel(ch->channel);
3959 74 : spdk_put_io_channel(ch->accel_channel);
3960 :
3961 74 : shared_resource = ch->shared_resource;
3962 :
3963 74 : assert(TAILQ_EMPTY(&ch->io_locked));
3964 74 : assert(TAILQ_EMPTY(&ch->io_submitted));
3965 74 : assert(TAILQ_EMPTY(&ch->io_accel_exec));
3966 74 : assert(TAILQ_EMPTY(&ch->io_memory_domain));
3967 74 : assert(ch->io_outstanding == 0);
3968 74 : assert(shared_resource->ref > 0);
3969 74 : shared_resource->ref--;
3970 74 : if (shared_resource->ref == 0) {
3971 73 : assert(shared_resource->io_outstanding == 0);
3972 73 : TAILQ_REMOVE(&shared_resource->mgmt_ch->shared_resources, shared_resource, link);
3973 73 : spdk_put_io_channel(spdk_io_channel_from_ctx(shared_resource->mgmt_ch));
3974 73 : spdk_poller_unregister(&shared_resource->nomem_poller);
3975 73 : free(shared_resource);
3976 73 : }
3977 74 : }
3978 :
3979 : static void
3980 83 : bdev_enable_qos(struct spdk_bdev *bdev, struct spdk_bdev_channel *ch)
3981 : {
3982 83 : struct spdk_bdev_qos *qos = bdev->internal.qos;
3983 : int i;
3984 :
3985 83 : assert(spdk_spin_held(&bdev->internal.spinlock));
3986 :
3987 : /* Rate limiting on this bdev enabled */
3988 83 : if (qos) {
3989 17 : if (qos->ch == NULL) {
3990 : struct spdk_io_channel *io_ch;
3991 :
3992 9 : SPDK_DEBUGLOG(bdev, "Selecting channel %p as QoS channel for bdev %s on thread %p\n", ch,
3993 : bdev->name, spdk_get_thread());
3994 :
3995 : /* No qos channel has been selected, so set one up */
3996 :
3997 : /* Take another reference to ch */
3998 9 : io_ch = spdk_get_io_channel(__bdev_to_io_dev(bdev));
3999 9 : assert(io_ch != NULL);
4000 9 : qos->ch = ch;
4001 :
4002 9 : qos->thread = spdk_io_channel_get_thread(io_ch);
4003 :
4004 45 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
4005 36 : if (bdev_qos_is_iops_rate_limit(i) == true) {
4006 9 : qos->rate_limits[i].min_per_timeslice =
4007 : SPDK_BDEV_QOS_MIN_IO_PER_TIMESLICE;
4008 9 : } else {
4009 27 : qos->rate_limits[i].min_per_timeslice =
4010 : SPDK_BDEV_QOS_MIN_BYTE_PER_TIMESLICE;
4011 : }
4012 :
4013 36 : if (qos->rate_limits[i].limit == 0) {
4014 2 : qos->rate_limits[i].limit = SPDK_BDEV_QOS_LIMIT_NOT_DEFINED;
4015 2 : }
4016 36 : }
4017 9 : bdev_qos_update_max_quota_per_timeslice(qos);
4018 9 : qos->timeslice_size =
4019 9 : SPDK_BDEV_QOS_TIMESLICE_IN_USEC * spdk_get_ticks_hz() / SPDK_SEC_TO_USEC;
4020 9 : qos->last_timeslice = spdk_get_ticks();
4021 9 : qos->poller = SPDK_POLLER_REGISTER(bdev_channel_poll_qos,
4022 : bdev,
4023 : SPDK_BDEV_QOS_TIMESLICE_IN_USEC);
4024 9 : }
4025 :
4026 17 : ch->flags |= BDEV_CH_QOS_ENABLED;
4027 17 : }
4028 83 : }
4029 :
4030 : struct poll_timeout_ctx {
4031 : struct spdk_bdev_desc *desc;
4032 : uint64_t timeout_in_sec;
4033 : spdk_bdev_io_timeout_cb cb_fn;
4034 : void *cb_arg;
4035 : };
4036 :
4037 : static void
4038 267 : bdev_desc_free(struct spdk_bdev_desc *desc)
4039 : {
4040 267 : spdk_spin_destroy(&desc->spinlock);
4041 267 : free(desc->media_events_buffer);
4042 267 : free(desc);
4043 267 : }
4044 :
4045 : static void
4046 8 : bdev_channel_poll_timeout_io_done(struct spdk_bdev *bdev, void *_ctx, int status)
4047 : {
4048 8 : struct poll_timeout_ctx *ctx = _ctx;
4049 8 : struct spdk_bdev_desc *desc = ctx->desc;
4050 :
4051 8 : free(ctx);
4052 :
4053 8 : spdk_spin_lock(&desc->spinlock);
4054 8 : desc->refs--;
4055 8 : if (desc->closed == true && desc->refs == 0) {
4056 1 : spdk_spin_unlock(&desc->spinlock);
4057 1 : bdev_desc_free(desc);
4058 1 : return;
4059 : }
4060 7 : spdk_spin_unlock(&desc->spinlock);
4061 8 : }
4062 :
4063 : static void
4064 13 : bdev_channel_poll_timeout_io(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
4065 : struct spdk_io_channel *io_ch, void *_ctx)
4066 : {
4067 13 : struct poll_timeout_ctx *ctx = _ctx;
4068 13 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
4069 13 : struct spdk_bdev_desc *desc = ctx->desc;
4070 : struct spdk_bdev_io *bdev_io;
4071 : uint64_t now;
4072 :
4073 13 : spdk_spin_lock(&desc->spinlock);
4074 13 : if (desc->closed == true) {
4075 1 : spdk_spin_unlock(&desc->spinlock);
4076 1 : spdk_bdev_for_each_channel_continue(i, -1);
4077 1 : return;
4078 : }
4079 12 : spdk_spin_unlock(&desc->spinlock);
4080 :
4081 12 : now = spdk_get_ticks();
4082 22 : TAILQ_FOREACH(bdev_io, &bdev_ch->io_submitted, internal.ch_link) {
4083 : /* Exclude any I/O that are generated via splitting. */
4084 15 : if (bdev_io->internal.cb == bdev_io_split_done) {
4085 3 : continue;
4086 : }
4087 :
4088 : /* Once we find an I/O that has not timed out, we can immediately
4089 : * exit the loop.
4090 : */
4091 24 : if (now < (bdev_io->internal.submit_tsc +
4092 12 : ctx->timeout_in_sec * spdk_get_ticks_hz())) {
4093 5 : goto end;
4094 : }
4095 :
4096 7 : if (bdev_io->internal.desc == desc) {
4097 7 : ctx->cb_fn(ctx->cb_arg, bdev_io);
4098 7 : }
4099 14 : }
4100 :
4101 : end:
4102 12 : spdk_bdev_for_each_channel_continue(i, 0);
4103 13 : }
4104 :
4105 : static int
4106 8 : bdev_poll_timeout_io(void *arg)
4107 : {
4108 8 : struct spdk_bdev_desc *desc = arg;
4109 8 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
4110 : struct poll_timeout_ctx *ctx;
4111 :
4112 8 : ctx = calloc(1, sizeof(struct poll_timeout_ctx));
4113 8 : if (!ctx) {
4114 0 : SPDK_ERRLOG("failed to allocate memory\n");
4115 0 : return SPDK_POLLER_BUSY;
4116 : }
4117 8 : ctx->desc = desc;
4118 8 : ctx->cb_arg = desc->cb_arg;
4119 8 : ctx->cb_fn = desc->cb_fn;
4120 8 : ctx->timeout_in_sec = desc->timeout_in_sec;
4121 :
4122 : /* Take a ref on the descriptor in case it gets closed while we are checking
4123 : * all of the channels.
4124 : */
4125 8 : spdk_spin_lock(&desc->spinlock);
4126 8 : desc->refs++;
4127 8 : spdk_spin_unlock(&desc->spinlock);
4128 :
4129 8 : spdk_bdev_for_each_channel(bdev, bdev_channel_poll_timeout_io, ctx,
4130 : bdev_channel_poll_timeout_io_done);
4131 :
4132 8 : return SPDK_POLLER_BUSY;
4133 8 : }
4134 :
4135 : int
4136 5 : spdk_bdev_set_timeout(struct spdk_bdev_desc *desc, uint64_t timeout_in_sec,
4137 : spdk_bdev_io_timeout_cb cb_fn, void *cb_arg)
4138 : {
4139 5 : assert(desc->thread == spdk_get_thread());
4140 :
4141 5 : spdk_poller_unregister(&desc->io_timeout_poller);
4142 :
4143 5 : if (timeout_in_sec) {
4144 4 : assert(cb_fn != NULL);
4145 4 : desc->io_timeout_poller = SPDK_POLLER_REGISTER(bdev_poll_timeout_io,
4146 : desc,
4147 : SPDK_BDEV_IO_POLL_INTERVAL_IN_MSEC * SPDK_SEC_TO_USEC /
4148 : 1000);
4149 4 : if (desc->io_timeout_poller == NULL) {
4150 0 : SPDK_ERRLOG("can not register the desc timeout IO poller\n");
4151 0 : return -1;
4152 : }
4153 4 : }
4154 :
4155 5 : desc->cb_fn = cb_fn;
4156 5 : desc->cb_arg = cb_arg;
4157 5 : desc->timeout_in_sec = timeout_in_sec;
4158 :
4159 5 : return 0;
4160 5 : }
4161 :
4162 : static int
4163 76 : bdev_channel_create(void *io_device, void *ctx_buf)
4164 : {
4165 76 : struct spdk_bdev *bdev = __bdev_from_io_dev(io_device);
4166 76 : struct spdk_bdev_channel *ch = ctx_buf;
4167 : struct spdk_io_channel *mgmt_io_ch;
4168 : struct spdk_bdev_mgmt_channel *mgmt_ch;
4169 : struct spdk_bdev_shared_resource *shared_resource;
4170 : struct lba_range *range;
4171 :
4172 76 : ch->bdev = bdev;
4173 76 : ch->channel = bdev->fn_table->get_io_channel(bdev->ctxt);
4174 76 : if (!ch->channel) {
4175 2 : return -1;
4176 : }
4177 :
4178 74 : ch->accel_channel = spdk_accel_get_io_channel();
4179 74 : if (!ch->accel_channel) {
4180 0 : spdk_put_io_channel(ch->channel);
4181 0 : return -1;
4182 : }
4183 :
4184 74 : spdk_trace_record(TRACE_BDEV_IOCH_CREATE, bdev->internal.trace_id, 0, 0,
4185 : spdk_thread_get_id(spdk_io_channel_get_thread(ch->channel)));
4186 :
4187 74 : assert(ch->histogram == NULL);
4188 74 : if (bdev->internal.histogram_enabled) {
4189 0 : ch->histogram = spdk_histogram_data_alloc();
4190 0 : if (ch->histogram == NULL) {
4191 0 : SPDK_ERRLOG("Could not allocate histogram\n");
4192 0 : }
4193 0 : }
4194 :
4195 74 : mgmt_io_ch = spdk_get_io_channel(&g_bdev_mgr);
4196 74 : if (!mgmt_io_ch) {
4197 0 : spdk_put_io_channel(ch->channel);
4198 0 : spdk_put_io_channel(ch->accel_channel);
4199 0 : return -1;
4200 : }
4201 :
4202 74 : mgmt_ch = __io_ch_to_bdev_mgmt_ch(mgmt_io_ch);
4203 76 : TAILQ_FOREACH(shared_resource, &mgmt_ch->shared_resources, link) {
4204 3 : if (shared_resource->shared_ch == ch->channel) {
4205 1 : spdk_put_io_channel(mgmt_io_ch);
4206 1 : shared_resource->ref++;
4207 1 : break;
4208 : }
4209 2 : }
4210 :
4211 74 : if (shared_resource == NULL) {
4212 73 : shared_resource = calloc(1, sizeof(*shared_resource));
4213 73 : if (shared_resource == NULL) {
4214 0 : spdk_put_io_channel(ch->channel);
4215 0 : spdk_put_io_channel(ch->accel_channel);
4216 0 : spdk_put_io_channel(mgmt_io_ch);
4217 0 : return -1;
4218 : }
4219 :
4220 73 : shared_resource->mgmt_ch = mgmt_ch;
4221 73 : shared_resource->io_outstanding = 0;
4222 73 : TAILQ_INIT(&shared_resource->nomem_io);
4223 73 : shared_resource->nomem_threshold = 0;
4224 73 : shared_resource->shared_ch = ch->channel;
4225 73 : shared_resource->ref = 1;
4226 73 : TAILQ_INSERT_TAIL(&mgmt_ch->shared_resources, shared_resource, link);
4227 73 : }
4228 :
4229 74 : ch->io_outstanding = 0;
4230 74 : TAILQ_INIT(&ch->queued_resets);
4231 74 : TAILQ_INIT(&ch->locked_ranges);
4232 74 : TAILQ_INIT(&ch->qos_queued_io);
4233 74 : ch->flags = 0;
4234 74 : ch->trace_id = bdev->internal.trace_id;
4235 74 : ch->shared_resource = shared_resource;
4236 :
4237 74 : TAILQ_INIT(&ch->io_submitted);
4238 74 : TAILQ_INIT(&ch->io_locked);
4239 74 : TAILQ_INIT(&ch->io_accel_exec);
4240 74 : TAILQ_INIT(&ch->io_memory_domain);
4241 :
4242 74 : ch->stat = bdev_alloc_io_stat(false);
4243 74 : if (ch->stat == NULL) {
4244 0 : bdev_channel_destroy_resource(ch);
4245 0 : return -1;
4246 : }
4247 :
4248 74 : ch->stat->ticks_rate = spdk_get_ticks_hz();
4249 :
4250 : #ifdef SPDK_CONFIG_VTUNE
4251 : {
4252 : char *name;
4253 : __itt_init_ittlib(NULL, 0);
4254 : name = spdk_sprintf_alloc("spdk_bdev_%s_%p", ch->bdev->name, ch);
4255 : if (!name) {
4256 : bdev_channel_destroy_resource(ch);
4257 : return -1;
4258 : }
4259 : ch->handle = __itt_string_handle_create(name);
4260 : free(name);
4261 : ch->start_tsc = spdk_get_ticks();
4262 : ch->interval_tsc = spdk_get_ticks_hz() / 100;
4263 : ch->prev_stat = bdev_alloc_io_stat(false);
4264 : if (ch->prev_stat == NULL) {
4265 : bdev_channel_destroy_resource(ch);
4266 : return -1;
4267 : }
4268 : }
4269 : #endif
4270 :
4271 74 : spdk_spin_lock(&bdev->internal.spinlock);
4272 74 : bdev_enable_qos(bdev, ch);
4273 :
4274 75 : TAILQ_FOREACH(range, &bdev->internal.locked_ranges, tailq) {
4275 : struct lba_range *new_range;
4276 :
4277 1 : new_range = calloc(1, sizeof(*new_range));
4278 1 : if (new_range == NULL) {
4279 0 : spdk_spin_unlock(&bdev->internal.spinlock);
4280 0 : bdev_channel_destroy_resource(ch);
4281 0 : return -1;
4282 : }
4283 1 : new_range->length = range->length;
4284 1 : new_range->offset = range->offset;
4285 1 : new_range->locked_ctx = range->locked_ctx;
4286 1 : TAILQ_INSERT_TAIL(&ch->locked_ranges, new_range, tailq);
4287 1 : }
4288 :
4289 74 : spdk_spin_unlock(&bdev->internal.spinlock);
4290 :
4291 74 : return 0;
4292 76 : }
4293 :
4294 : static int
4295 0 : bdev_abort_all_buf_io_cb(struct spdk_iobuf_channel *ch, struct spdk_iobuf_entry *entry,
4296 : void *cb_ctx)
4297 : {
4298 0 : struct spdk_bdev_channel *bdev_ch = cb_ctx;
4299 : struct spdk_bdev_io *bdev_io;
4300 : uint64_t buf_len;
4301 :
4302 0 : bdev_io = SPDK_CONTAINEROF(entry, struct spdk_bdev_io, internal.iobuf);
4303 0 : if (bdev_io->internal.ch == bdev_ch) {
4304 0 : buf_len = bdev_io_get_max_buf_len(bdev_io, bdev_io->internal.buf.len);
4305 0 : spdk_iobuf_entry_abort(ch, entry, buf_len);
4306 0 : spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_ABORTED);
4307 0 : }
4308 :
4309 0 : return 0;
4310 : }
4311 :
4312 : /*
4313 : * Abort I/O that are waiting on a data buffer.
4314 : */
4315 : static void
4316 96 : bdev_abort_all_buf_io(struct spdk_bdev_mgmt_channel *mgmt_ch, struct spdk_bdev_channel *ch)
4317 : {
4318 96 : spdk_iobuf_for_each_entry(&mgmt_ch->iobuf, bdev_abort_all_buf_io_cb, ch);
4319 96 : }
4320 :
4321 : /*
4322 : * Abort I/O that are queued waiting for submission. These types of I/O are
4323 : * linked using the spdk_bdev_io link TAILQ_ENTRY.
4324 : */
4325 : static void
4326 188 : bdev_abort_all_queued_io(bdev_io_tailq_t *queue, struct spdk_bdev_channel *ch)
4327 : {
4328 : struct spdk_bdev_io *bdev_io, *tmp;
4329 :
4330 228 : TAILQ_FOREACH_SAFE(bdev_io, queue, internal.link, tmp) {
4331 40 : if (bdev_io->internal.ch == ch) {
4332 40 : TAILQ_REMOVE(queue, bdev_io, internal.link);
4333 : /*
4334 : * spdk_bdev_io_complete() assumes that the completed I/O had
4335 : * been submitted to the bdev module. Since in this case it
4336 : * hadn't, bump io_outstanding to account for the decrement
4337 : * that spdk_bdev_io_complete() will do.
4338 : */
4339 40 : if (bdev_io->type != SPDK_BDEV_IO_TYPE_RESET) {
4340 39 : bdev_io_increment_outstanding(ch, ch->shared_resource);
4341 39 : }
4342 40 : spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_ABORTED);
4343 40 : }
4344 40 : }
4345 188 : }
4346 :
4347 : static bool
4348 18 : bdev_abort_queued_io(bdev_io_tailq_t *queue, struct spdk_bdev_io *bio_to_abort)
4349 : {
4350 : struct spdk_bdev_io *bdev_io;
4351 :
4352 18 : TAILQ_FOREACH(bdev_io, queue, internal.link) {
4353 0 : if (bdev_io == bio_to_abort) {
4354 0 : TAILQ_REMOVE(queue, bio_to_abort, internal.link);
4355 0 : spdk_bdev_io_complete(bio_to_abort, SPDK_BDEV_IO_STATUS_ABORTED);
4356 0 : return true;
4357 : }
4358 0 : }
4359 :
4360 18 : return false;
4361 18 : }
4362 :
4363 : static int
4364 0 : bdev_abort_buf_io_cb(struct spdk_iobuf_channel *ch, struct spdk_iobuf_entry *entry, void *cb_ctx)
4365 : {
4366 0 : struct spdk_bdev_io *bdev_io, *bio_to_abort = cb_ctx;
4367 : uint64_t buf_len;
4368 :
4369 0 : bdev_io = SPDK_CONTAINEROF(entry, struct spdk_bdev_io, internal.iobuf);
4370 0 : if (bdev_io == bio_to_abort) {
4371 0 : buf_len = bdev_io_get_max_buf_len(bdev_io, bdev_io->internal.buf.len);
4372 0 : spdk_iobuf_entry_abort(ch, entry, buf_len);
4373 0 : spdk_bdev_io_complete(bio_to_abort, SPDK_BDEV_IO_STATUS_ABORTED);
4374 0 : return 1;
4375 : }
4376 :
4377 0 : return 0;
4378 0 : }
4379 :
4380 : static bool
4381 16 : bdev_abort_buf_io(struct spdk_bdev_mgmt_channel *mgmt_ch, struct spdk_bdev_io *bio_to_abort)
4382 : {
4383 : int rc;
4384 :
4385 16 : rc = spdk_iobuf_for_each_entry(&mgmt_ch->iobuf, bdev_abort_buf_io_cb, bio_to_abort);
4386 16 : return rc == 1;
4387 : }
4388 :
4389 : static void
4390 7 : bdev_qos_channel_destroy(void *cb_arg)
4391 : {
4392 7 : struct spdk_bdev_qos *qos = cb_arg;
4393 :
4394 7 : spdk_put_io_channel(spdk_io_channel_from_ctx(qos->ch));
4395 7 : spdk_poller_unregister(&qos->poller);
4396 :
4397 7 : SPDK_DEBUGLOG(bdev, "Free QoS %p.\n", qos);
4398 :
4399 7 : free(qos);
4400 7 : }
4401 :
4402 : static int
4403 7 : bdev_qos_destroy(struct spdk_bdev *bdev)
4404 : {
4405 : int i;
4406 :
4407 : /*
4408 : * Cleanly shutting down the QoS poller is tricky, because
4409 : * during the asynchronous operation the user could open
4410 : * a new descriptor and create a new channel, spawning
4411 : * a new QoS poller.
4412 : *
4413 : * The strategy is to create a new QoS structure here and swap it
4414 : * in. The shutdown path then continues to refer to the old one
4415 : * until it completes and then releases it.
4416 : */
4417 : struct spdk_bdev_qos *new_qos, *old_qos;
4418 :
4419 7 : old_qos = bdev->internal.qos;
4420 :
4421 7 : new_qos = calloc(1, sizeof(*new_qos));
4422 7 : if (!new_qos) {
4423 0 : SPDK_ERRLOG("Unable to allocate memory to shut down QoS.\n");
4424 0 : return -ENOMEM;
4425 : }
4426 :
4427 : /* Copy the old QoS data into the newly allocated structure */
4428 7 : memcpy(new_qos, old_qos, sizeof(*new_qos));
4429 :
4430 : /* Zero out the key parts of the QoS structure */
4431 7 : new_qos->ch = NULL;
4432 7 : new_qos->thread = NULL;
4433 7 : new_qos->poller = NULL;
4434 : /*
4435 : * The limit member of spdk_bdev_qos_limit structure is not zeroed.
4436 : * It will be used later for the new QoS structure.
4437 : */
4438 35 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
4439 28 : new_qos->rate_limits[i].remaining_this_timeslice = 0;
4440 28 : new_qos->rate_limits[i].min_per_timeslice = 0;
4441 28 : new_qos->rate_limits[i].max_per_timeslice = 0;
4442 28 : }
4443 :
4444 7 : bdev->internal.qos = new_qos;
4445 :
4446 7 : if (old_qos->thread == NULL) {
4447 0 : free(old_qos);
4448 0 : } else {
4449 7 : spdk_thread_send_msg(old_qos->thread, bdev_qos_channel_destroy, old_qos);
4450 : }
4451 :
4452 : /* It is safe to continue with destroying the bdev even though the QoS channel hasn't
4453 : * been destroyed yet. The destruction path will end up waiting for the final
4454 : * channel to be put before it releases resources. */
4455 :
4456 7 : return 0;
4457 7 : }
4458 :
4459 : void
4460 78 : spdk_bdev_add_io_stat(struct spdk_bdev_io_stat *total, struct spdk_bdev_io_stat *add)
4461 : {
4462 78 : total->bytes_read += add->bytes_read;
4463 78 : total->num_read_ops += add->num_read_ops;
4464 78 : total->bytes_written += add->bytes_written;
4465 78 : total->num_write_ops += add->num_write_ops;
4466 78 : total->bytes_unmapped += add->bytes_unmapped;
4467 78 : total->num_unmap_ops += add->num_unmap_ops;
4468 78 : total->bytes_copied += add->bytes_copied;
4469 78 : total->num_copy_ops += add->num_copy_ops;
4470 78 : total->read_latency_ticks += add->read_latency_ticks;
4471 78 : total->write_latency_ticks += add->write_latency_ticks;
4472 78 : total->unmap_latency_ticks += add->unmap_latency_ticks;
4473 78 : total->copy_latency_ticks += add->copy_latency_ticks;
4474 78 : if (total->max_read_latency_ticks < add->max_read_latency_ticks) {
4475 7 : total->max_read_latency_ticks = add->max_read_latency_ticks;
4476 7 : }
4477 78 : if (total->min_read_latency_ticks > add->min_read_latency_ticks) {
4478 39 : total->min_read_latency_ticks = add->min_read_latency_ticks;
4479 39 : }
4480 78 : if (total->max_write_latency_ticks < add->max_write_latency_ticks) {
4481 4 : total->max_write_latency_ticks = add->max_write_latency_ticks;
4482 4 : }
4483 78 : if (total->min_write_latency_ticks > add->min_write_latency_ticks) {
4484 24 : total->min_write_latency_ticks = add->min_write_latency_ticks;
4485 24 : }
4486 78 : if (total->max_unmap_latency_ticks < add->max_unmap_latency_ticks) {
4487 0 : total->max_unmap_latency_ticks = add->max_unmap_latency_ticks;
4488 0 : }
4489 78 : if (total->min_unmap_latency_ticks > add->min_unmap_latency_ticks) {
4490 3 : total->min_unmap_latency_ticks = add->min_unmap_latency_ticks;
4491 3 : }
4492 78 : if (total->max_copy_latency_ticks < add->max_copy_latency_ticks) {
4493 0 : total->max_copy_latency_ticks = add->max_copy_latency_ticks;
4494 0 : }
4495 78 : if (total->min_copy_latency_ticks > add->min_copy_latency_ticks) {
4496 4 : total->min_copy_latency_ticks = add->min_copy_latency_ticks;
4497 4 : }
4498 78 : }
4499 :
4500 : static void
4501 5 : bdev_get_io_stat(struct spdk_bdev_io_stat *to_stat, struct spdk_bdev_io_stat *from_stat)
4502 : {
4503 5 : memcpy(to_stat, from_stat, offsetof(struct spdk_bdev_io_stat, io_error));
4504 :
4505 5 : if (to_stat->io_error != NULL && from_stat->io_error != NULL) {
4506 0 : memcpy(to_stat->io_error, from_stat->io_error,
4507 : sizeof(struct spdk_bdev_io_error_stat));
4508 0 : }
4509 5 : }
4510 :
4511 : void
4512 210 : spdk_bdev_reset_io_stat(struct spdk_bdev_io_stat *stat, enum spdk_bdev_reset_stat_mode mode)
4513 : {
4514 210 : if (mode == SPDK_BDEV_RESET_STAT_NONE) {
4515 5 : return;
4516 : }
4517 :
4518 205 : stat->max_read_latency_ticks = 0;
4519 205 : stat->min_read_latency_ticks = UINT64_MAX;
4520 205 : stat->max_write_latency_ticks = 0;
4521 205 : stat->min_write_latency_ticks = UINT64_MAX;
4522 205 : stat->max_unmap_latency_ticks = 0;
4523 205 : stat->min_unmap_latency_ticks = UINT64_MAX;
4524 205 : stat->max_copy_latency_ticks = 0;
4525 205 : stat->min_copy_latency_ticks = UINT64_MAX;
4526 :
4527 205 : if (mode != SPDK_BDEV_RESET_STAT_ALL) {
4528 2 : return;
4529 : }
4530 :
4531 203 : stat->bytes_read = 0;
4532 203 : stat->num_read_ops = 0;
4533 203 : stat->bytes_written = 0;
4534 203 : stat->num_write_ops = 0;
4535 203 : stat->bytes_unmapped = 0;
4536 203 : stat->num_unmap_ops = 0;
4537 203 : stat->bytes_copied = 0;
4538 203 : stat->num_copy_ops = 0;
4539 203 : stat->read_latency_ticks = 0;
4540 203 : stat->write_latency_ticks = 0;
4541 203 : stat->unmap_latency_ticks = 0;
4542 203 : stat->copy_latency_ticks = 0;
4543 :
4544 203 : if (stat->io_error != NULL) {
4545 128 : memset(stat->io_error, 0, sizeof(struct spdk_bdev_io_error_stat));
4546 128 : }
4547 210 : }
4548 :
4549 : struct spdk_bdev_io_stat *
4550 201 : bdev_alloc_io_stat(bool io_error_stat)
4551 : {
4552 : struct spdk_bdev_io_stat *stat;
4553 :
4554 201 : stat = malloc(sizeof(struct spdk_bdev_io_stat));
4555 201 : if (stat == NULL) {
4556 0 : return NULL;
4557 : }
4558 :
4559 201 : if (io_error_stat) {
4560 127 : stat->io_error = malloc(sizeof(struct spdk_bdev_io_error_stat));
4561 127 : if (stat->io_error == NULL) {
4562 0 : free(stat);
4563 0 : return NULL;
4564 : }
4565 127 : } else {
4566 74 : stat->io_error = NULL;
4567 : }
4568 :
4569 201 : spdk_bdev_reset_io_stat(stat, SPDK_BDEV_RESET_STAT_ALL);
4570 :
4571 201 : return stat;
4572 201 : }
4573 :
4574 : void
4575 201 : bdev_free_io_stat(struct spdk_bdev_io_stat *stat)
4576 : {
4577 201 : if (stat != NULL) {
4578 201 : free(stat->io_error);
4579 201 : free(stat);
4580 201 : }
4581 201 : }
4582 :
4583 : void
4584 0 : spdk_bdev_dump_io_stat_json(struct spdk_bdev_io_stat *stat, struct spdk_json_write_ctx *w)
4585 : {
4586 : int i;
4587 :
4588 0 : spdk_json_write_named_uint64(w, "bytes_read", stat->bytes_read);
4589 0 : spdk_json_write_named_uint64(w, "num_read_ops", stat->num_read_ops);
4590 0 : spdk_json_write_named_uint64(w, "bytes_written", stat->bytes_written);
4591 0 : spdk_json_write_named_uint64(w, "num_write_ops", stat->num_write_ops);
4592 0 : spdk_json_write_named_uint64(w, "bytes_unmapped", stat->bytes_unmapped);
4593 0 : spdk_json_write_named_uint64(w, "num_unmap_ops", stat->num_unmap_ops);
4594 0 : spdk_json_write_named_uint64(w, "bytes_copied", stat->bytes_copied);
4595 0 : spdk_json_write_named_uint64(w, "num_copy_ops", stat->num_copy_ops);
4596 0 : spdk_json_write_named_uint64(w, "read_latency_ticks", stat->read_latency_ticks);
4597 0 : spdk_json_write_named_uint64(w, "max_read_latency_ticks", stat->max_read_latency_ticks);
4598 0 : spdk_json_write_named_uint64(w, "min_read_latency_ticks",
4599 0 : stat->min_read_latency_ticks != UINT64_MAX ?
4600 0 : stat->min_read_latency_ticks : 0);
4601 0 : spdk_json_write_named_uint64(w, "write_latency_ticks", stat->write_latency_ticks);
4602 0 : spdk_json_write_named_uint64(w, "max_write_latency_ticks", stat->max_write_latency_ticks);
4603 0 : spdk_json_write_named_uint64(w, "min_write_latency_ticks",
4604 0 : stat->min_write_latency_ticks != UINT64_MAX ?
4605 0 : stat->min_write_latency_ticks : 0);
4606 0 : spdk_json_write_named_uint64(w, "unmap_latency_ticks", stat->unmap_latency_ticks);
4607 0 : spdk_json_write_named_uint64(w, "max_unmap_latency_ticks", stat->max_unmap_latency_ticks);
4608 0 : spdk_json_write_named_uint64(w, "min_unmap_latency_ticks",
4609 0 : stat->min_unmap_latency_ticks != UINT64_MAX ?
4610 0 : stat->min_unmap_latency_ticks : 0);
4611 0 : spdk_json_write_named_uint64(w, "copy_latency_ticks", stat->copy_latency_ticks);
4612 0 : spdk_json_write_named_uint64(w, "max_copy_latency_ticks", stat->max_copy_latency_ticks);
4613 0 : spdk_json_write_named_uint64(w, "min_copy_latency_ticks",
4614 0 : stat->min_copy_latency_ticks != UINT64_MAX ?
4615 0 : stat->min_copy_latency_ticks : 0);
4616 :
4617 0 : if (stat->io_error != NULL) {
4618 0 : spdk_json_write_named_object_begin(w, "io_error");
4619 0 : for (i = 0; i < -SPDK_MIN_BDEV_IO_STATUS; i++) {
4620 0 : if (stat->io_error->error_status[i] != 0) {
4621 0 : spdk_json_write_named_uint32(w, bdev_io_status_get_string(-(i + 1)),
4622 0 : stat->io_error->error_status[i]);
4623 0 : }
4624 0 : }
4625 0 : spdk_json_write_object_end(w);
4626 0 : }
4627 0 : }
4628 :
4629 : static void
4630 78 : bdev_channel_abort_queued_ios(struct spdk_bdev_channel *ch)
4631 : {
4632 78 : struct spdk_bdev_shared_resource *shared_resource = ch->shared_resource;
4633 78 : struct spdk_bdev_mgmt_channel *mgmt_ch = shared_resource->mgmt_ch;
4634 :
4635 78 : bdev_abort_all_queued_io(&shared_resource->nomem_io, ch);
4636 78 : bdev_abort_all_buf_io(mgmt_ch, ch);
4637 78 : }
4638 :
4639 : static void
4640 74 : bdev_channel_destroy(void *io_device, void *ctx_buf)
4641 : {
4642 74 : struct spdk_bdev_channel *ch = ctx_buf;
4643 :
4644 74 : SPDK_DEBUGLOG(bdev, "Destroying channel %p for bdev %s on thread %p\n", ch, ch->bdev->name,
4645 : spdk_get_thread());
4646 :
4647 74 : spdk_trace_record(TRACE_BDEV_IOCH_DESTROY, ch->bdev->internal.trace_id, 0, 0,
4648 : spdk_thread_get_id(spdk_io_channel_get_thread(ch->channel)));
4649 :
4650 : /* This channel is going away, so add its statistics into the bdev so that they don't get lost. */
4651 74 : spdk_spin_lock(&ch->bdev->internal.spinlock);
4652 74 : spdk_bdev_add_io_stat(ch->bdev->internal.stat, ch->stat);
4653 74 : spdk_spin_unlock(&ch->bdev->internal.spinlock);
4654 :
4655 74 : bdev_abort_all_queued_io(&ch->queued_resets, ch);
4656 :
4657 74 : bdev_channel_abort_queued_ios(ch);
4658 :
4659 74 : if (ch->histogram) {
4660 0 : spdk_histogram_data_free(ch->histogram);
4661 0 : }
4662 :
4663 74 : bdev_channel_destroy_resource(ch);
4664 74 : }
4665 :
4666 : /*
4667 : * If the name already exists in the global bdev name tree, RB_INSERT() returns a pointer
4668 : * to it. Hence we do not have to call bdev_get_by_name() when using this function.
4669 : */
4670 : static int
4671 257 : bdev_name_add(struct spdk_bdev_name *bdev_name, struct spdk_bdev *bdev, const char *name)
4672 : {
4673 : struct spdk_bdev_name *tmp;
4674 :
4675 257 : bdev_name->name = strdup(name);
4676 257 : if (bdev_name->name == NULL) {
4677 0 : SPDK_ERRLOG("Unable to allocate bdev name\n");
4678 0 : return -ENOMEM;
4679 : }
4680 :
4681 257 : bdev_name->bdev = bdev;
4682 :
4683 257 : spdk_spin_lock(&g_bdev_mgr.spinlock);
4684 257 : tmp = RB_INSERT(bdev_name_tree, &g_bdev_mgr.bdev_names, bdev_name);
4685 257 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
4686 :
4687 257 : if (tmp != NULL) {
4688 4 : SPDK_ERRLOG("Bdev name %s already exists\n", name);
4689 4 : free(bdev_name->name);
4690 4 : return -EEXIST;
4691 : }
4692 :
4693 253 : return 0;
4694 257 : }
4695 :
4696 : static void
4697 253 : bdev_name_del_unsafe(struct spdk_bdev_name *bdev_name)
4698 : {
4699 253 : RB_REMOVE(bdev_name_tree, &g_bdev_mgr.bdev_names, bdev_name);
4700 253 : free(bdev_name->name);
4701 253 : }
4702 :
4703 : static void
4704 5 : bdev_name_del(struct spdk_bdev_name *bdev_name)
4705 : {
4706 5 : spdk_spin_lock(&g_bdev_mgr.spinlock);
4707 5 : bdev_name_del_unsafe(bdev_name);
4708 5 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
4709 5 : }
4710 :
4711 : int
4712 133 : spdk_bdev_alias_add(struct spdk_bdev *bdev, const char *alias)
4713 : {
4714 : struct spdk_bdev_alias *tmp;
4715 : int ret;
4716 :
4717 133 : if (alias == NULL) {
4718 1 : SPDK_ERRLOG("Empty alias passed\n");
4719 1 : return -EINVAL;
4720 : }
4721 :
4722 132 : tmp = calloc(1, sizeof(*tmp));
4723 132 : if (tmp == NULL) {
4724 0 : SPDK_ERRLOG("Unable to allocate alias\n");
4725 0 : return -ENOMEM;
4726 : }
4727 :
4728 132 : ret = bdev_name_add(&tmp->alias, bdev, alias);
4729 132 : if (ret != 0) {
4730 4 : free(tmp);
4731 4 : return ret;
4732 : }
4733 :
4734 128 : TAILQ_INSERT_TAIL(&bdev->aliases, tmp, tailq);
4735 :
4736 128 : return 0;
4737 133 : }
4738 :
4739 : static int
4740 129 : bdev_alias_del(struct spdk_bdev *bdev, const char *alias,
4741 : void (*alias_del_fn)(struct spdk_bdev_name *n))
4742 : {
4743 : struct spdk_bdev_alias *tmp;
4744 :
4745 134 : TAILQ_FOREACH(tmp, &bdev->aliases, tailq) {
4746 130 : if (strcmp(alias, tmp->alias.name) == 0) {
4747 125 : TAILQ_REMOVE(&bdev->aliases, tmp, tailq);
4748 125 : alias_del_fn(&tmp->alias);
4749 125 : free(tmp);
4750 125 : return 0;
4751 : }
4752 5 : }
4753 :
4754 4 : return -ENOENT;
4755 129 : }
4756 :
4757 : int
4758 4 : spdk_bdev_alias_del(struct spdk_bdev *bdev, const char *alias)
4759 : {
4760 : int rc;
4761 :
4762 4 : rc = bdev_alias_del(bdev, alias, bdev_name_del);
4763 4 : if (rc == -ENOENT) {
4764 2 : SPDK_INFOLOG(bdev, "Alias %s does not exist\n", alias);
4765 2 : }
4766 :
4767 4 : return rc;
4768 : }
4769 :
4770 : void
4771 2 : spdk_bdev_alias_del_all(struct spdk_bdev *bdev)
4772 : {
4773 : struct spdk_bdev_alias *p, *tmp;
4774 :
4775 5 : TAILQ_FOREACH_SAFE(p, &bdev->aliases, tailq, tmp) {
4776 3 : TAILQ_REMOVE(&bdev->aliases, p, tailq);
4777 3 : bdev_name_del(&p->alias);
4778 3 : free(p);
4779 3 : }
4780 2 : }
4781 :
4782 : struct spdk_io_channel *
4783 76 : spdk_bdev_get_io_channel(struct spdk_bdev_desc *desc)
4784 : {
4785 76 : return spdk_get_io_channel(__bdev_to_io_dev(spdk_bdev_desc_get_bdev(desc)));
4786 : }
4787 :
4788 : void *
4789 0 : spdk_bdev_get_module_ctx(struct spdk_bdev_desc *desc)
4790 : {
4791 0 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
4792 0 : void *ctx = NULL;
4793 :
4794 0 : if (bdev->fn_table->get_module_ctx) {
4795 0 : ctx = bdev->fn_table->get_module_ctx(bdev->ctxt);
4796 0 : }
4797 :
4798 0 : return ctx;
4799 : }
4800 :
4801 : const char *
4802 0 : spdk_bdev_get_module_name(const struct spdk_bdev *bdev)
4803 : {
4804 0 : return bdev->module->name;
4805 : }
4806 :
4807 : const char *
4808 253 : spdk_bdev_get_name(const struct spdk_bdev *bdev)
4809 : {
4810 253 : return bdev->name;
4811 : }
4812 :
4813 : const char *
4814 0 : spdk_bdev_get_product_name(const struct spdk_bdev *bdev)
4815 : {
4816 0 : return bdev->product_name;
4817 : }
4818 :
4819 : const struct spdk_bdev_aliases_list *
4820 0 : spdk_bdev_get_aliases(const struct spdk_bdev *bdev)
4821 : {
4822 0 : return &bdev->aliases;
4823 : }
4824 :
4825 : uint32_t
4826 5 : spdk_bdev_get_block_size(const struct spdk_bdev *bdev)
4827 : {
4828 5 : return bdev->blocklen;
4829 : }
4830 :
4831 : uint32_t
4832 0 : spdk_bdev_get_write_unit_size(const struct spdk_bdev *bdev)
4833 : {
4834 0 : return bdev->write_unit_size;
4835 : }
4836 :
4837 : uint64_t
4838 0 : spdk_bdev_get_num_blocks(const struct spdk_bdev *bdev)
4839 : {
4840 0 : return bdev->blockcnt;
4841 : }
4842 :
4843 : const char *
4844 0 : spdk_bdev_get_qos_rpc_type(enum spdk_bdev_qos_rate_limit_type type)
4845 : {
4846 0 : return qos_rpc_type[type];
4847 : }
4848 :
4849 : void
4850 0 : spdk_bdev_get_qos_rate_limits(struct spdk_bdev *bdev, uint64_t *limits)
4851 : {
4852 : int i;
4853 :
4854 0 : memset(limits, 0, sizeof(*limits) * SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES);
4855 :
4856 0 : spdk_spin_lock(&bdev->internal.spinlock);
4857 0 : if (bdev->internal.qos) {
4858 0 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
4859 0 : if (bdev->internal.qos->rate_limits[i].limit !=
4860 : SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
4861 0 : limits[i] = bdev->internal.qos->rate_limits[i].limit;
4862 0 : if (bdev_qos_is_iops_rate_limit(i) == false) {
4863 : /* Change from Byte to Megabyte which is user visible. */
4864 0 : limits[i] = limits[i] / 1024 / 1024;
4865 0 : }
4866 0 : }
4867 0 : }
4868 0 : }
4869 0 : spdk_spin_unlock(&bdev->internal.spinlock);
4870 0 : }
4871 :
4872 : size_t
4873 317 : spdk_bdev_get_buf_align(const struct spdk_bdev *bdev)
4874 : {
4875 317 : return 1 << bdev->required_alignment;
4876 : }
4877 :
4878 : uint32_t
4879 0 : spdk_bdev_get_optimal_io_boundary(const struct spdk_bdev *bdev)
4880 : {
4881 0 : return bdev->optimal_io_boundary;
4882 : }
4883 :
4884 : bool
4885 0 : spdk_bdev_has_write_cache(const struct spdk_bdev *bdev)
4886 : {
4887 0 : return bdev->write_cache;
4888 : }
4889 :
4890 : const struct spdk_uuid *
4891 0 : spdk_bdev_get_uuid(const struct spdk_bdev *bdev)
4892 : {
4893 0 : return &bdev->uuid;
4894 : }
4895 :
4896 : uint16_t
4897 0 : spdk_bdev_get_acwu(const struct spdk_bdev *bdev)
4898 : {
4899 0 : return bdev->acwu;
4900 : }
4901 :
4902 : uint32_t
4903 29 : spdk_bdev_get_md_size(const struct spdk_bdev *bdev)
4904 : {
4905 29 : return bdev->md_len;
4906 : }
4907 :
4908 : bool
4909 130 : spdk_bdev_is_md_interleaved(const struct spdk_bdev *bdev)
4910 : {
4911 130 : return (bdev->md_len != 0) && bdev->md_interleave;
4912 : }
4913 :
4914 : bool
4915 159 : spdk_bdev_is_md_separate(const struct spdk_bdev *bdev)
4916 : {
4917 159 : return (bdev->md_len != 0) && !bdev->md_interleave;
4918 : }
4919 :
4920 : bool
4921 0 : spdk_bdev_is_zoned(const struct spdk_bdev *bdev)
4922 : {
4923 0 : return bdev->zoned;
4924 : }
4925 :
4926 : uint32_t
4927 121 : spdk_bdev_get_data_block_size(const struct spdk_bdev *bdev)
4928 : {
4929 121 : if (spdk_bdev_is_md_interleaved(bdev)) {
4930 0 : return bdev->blocklen - bdev->md_len;
4931 : } else {
4932 121 : return bdev->blocklen;
4933 : }
4934 121 : }
4935 :
4936 : uint32_t
4937 0 : spdk_bdev_get_physical_block_size(const struct spdk_bdev *bdev)
4938 : {
4939 0 : return bdev->phys_blocklen;
4940 : }
4941 :
4942 : static uint32_t
4943 9 : _bdev_get_block_size_with_md(const struct spdk_bdev *bdev)
4944 : {
4945 9 : if (!spdk_bdev_is_md_interleaved(bdev)) {
4946 6 : return bdev->blocklen + bdev->md_len;
4947 : } else {
4948 3 : return bdev->blocklen;
4949 : }
4950 9 : }
4951 :
4952 : /* We have to use the typedef in the function declaration to appease astyle. */
4953 : typedef enum spdk_dif_type spdk_dif_type_t;
4954 : typedef enum spdk_dif_pi_format spdk_dif_pi_format_t;
4955 :
4956 : spdk_dif_type_t
4957 0 : spdk_bdev_get_dif_type(const struct spdk_bdev *bdev)
4958 : {
4959 0 : if (bdev->md_len != 0) {
4960 0 : return bdev->dif_type;
4961 : } else {
4962 0 : return SPDK_DIF_DISABLE;
4963 : }
4964 0 : }
4965 :
4966 : spdk_dif_pi_format_t
4967 0 : spdk_bdev_get_dif_pi_format(const struct spdk_bdev *bdev)
4968 : {
4969 0 : return bdev->dif_pi_format;
4970 : }
4971 :
4972 : bool
4973 0 : spdk_bdev_is_dif_head_of_md(const struct spdk_bdev *bdev)
4974 : {
4975 0 : if (spdk_bdev_get_dif_type(bdev) != SPDK_DIF_DISABLE) {
4976 0 : return bdev->dif_is_head_of_md;
4977 : } else {
4978 0 : return false;
4979 : }
4980 0 : }
4981 :
4982 : bool
4983 0 : spdk_bdev_is_dif_check_enabled(const struct spdk_bdev *bdev,
4984 : enum spdk_dif_check_type check_type)
4985 : {
4986 0 : if (spdk_bdev_get_dif_type(bdev) == SPDK_DIF_DISABLE) {
4987 0 : return false;
4988 : }
4989 :
4990 0 : switch (check_type) {
4991 : case SPDK_DIF_CHECK_TYPE_REFTAG:
4992 0 : return (bdev->dif_check_flags & SPDK_DIF_FLAGS_REFTAG_CHECK) != 0;
4993 : case SPDK_DIF_CHECK_TYPE_APPTAG:
4994 0 : return (bdev->dif_check_flags & SPDK_DIF_FLAGS_APPTAG_CHECK) != 0;
4995 : case SPDK_DIF_CHECK_TYPE_GUARD:
4996 0 : return (bdev->dif_check_flags & SPDK_DIF_FLAGS_GUARD_CHECK) != 0;
4997 : default:
4998 0 : return false;
4999 : }
5000 0 : }
5001 :
5002 : static uint32_t
5003 3 : bdev_get_max_write(const struct spdk_bdev *bdev, uint64_t num_bytes)
5004 : {
5005 : uint64_t aligned_length, max_write_blocks;
5006 :
5007 3 : aligned_length = num_bytes - (spdk_bdev_get_buf_align(bdev) - 1);
5008 3 : max_write_blocks = aligned_length / _bdev_get_block_size_with_md(bdev);
5009 3 : max_write_blocks -= max_write_blocks % bdev->write_unit_size;
5010 :
5011 3 : return max_write_blocks;
5012 : }
5013 :
5014 : uint32_t
5015 1 : spdk_bdev_get_max_copy(const struct spdk_bdev *bdev)
5016 : {
5017 1 : return bdev->max_copy;
5018 : }
5019 :
5020 : uint64_t
5021 0 : spdk_bdev_get_qd(const struct spdk_bdev *bdev)
5022 : {
5023 0 : return bdev->internal.measured_queue_depth;
5024 : }
5025 :
5026 : uint64_t
5027 0 : spdk_bdev_get_qd_sampling_period(const struct spdk_bdev *bdev)
5028 : {
5029 0 : return bdev->internal.period;
5030 : }
5031 :
5032 : uint64_t
5033 0 : spdk_bdev_get_weighted_io_time(const struct spdk_bdev *bdev)
5034 : {
5035 0 : return bdev->internal.weighted_io_time;
5036 : }
5037 :
5038 : uint64_t
5039 0 : spdk_bdev_get_io_time(const struct spdk_bdev *bdev)
5040 : {
5041 0 : return bdev->internal.io_time;
5042 : }
5043 :
5044 0 : union spdk_bdev_nvme_ctratt spdk_bdev_get_nvme_ctratt(struct spdk_bdev *bdev)
5045 : {
5046 0 : return bdev->ctratt;
5047 : }
5048 :
5049 : uint32_t
5050 0 : spdk_bdev_get_nvme_nsid(struct spdk_bdev *bdev)
5051 : {
5052 0 : return bdev->nsid;
5053 : }
5054 :
5055 : static void bdev_update_qd_sampling_period(void *ctx);
5056 :
5057 : static void
5058 1 : _calculate_measured_qd_cpl(struct spdk_bdev *bdev, void *_ctx, int status)
5059 : {
5060 1 : bdev->internal.measured_queue_depth = bdev->internal.temporary_queue_depth;
5061 :
5062 1 : if (bdev->internal.measured_queue_depth) {
5063 0 : bdev->internal.io_time += bdev->internal.period;
5064 0 : bdev->internal.weighted_io_time += bdev->internal.period * bdev->internal.measured_queue_depth;
5065 0 : }
5066 :
5067 1 : bdev->internal.qd_poll_in_progress = false;
5068 :
5069 1 : bdev_update_qd_sampling_period(bdev);
5070 1 : }
5071 :
5072 : static void
5073 1 : _calculate_measured_qd(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
5074 : struct spdk_io_channel *io_ch, void *_ctx)
5075 : {
5076 1 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(io_ch);
5077 :
5078 1 : bdev->internal.temporary_queue_depth += ch->io_outstanding;
5079 1 : spdk_bdev_for_each_channel_continue(i, 0);
5080 1 : }
5081 :
5082 : static int
5083 1 : bdev_calculate_measured_queue_depth(void *ctx)
5084 : {
5085 1 : struct spdk_bdev *bdev = ctx;
5086 :
5087 1 : bdev->internal.qd_poll_in_progress = true;
5088 1 : bdev->internal.temporary_queue_depth = 0;
5089 1 : spdk_bdev_for_each_channel(bdev, _calculate_measured_qd, bdev, _calculate_measured_qd_cpl);
5090 1 : return SPDK_POLLER_BUSY;
5091 : }
5092 :
5093 : static void
5094 5 : bdev_update_qd_sampling_period(void *ctx)
5095 : {
5096 5 : struct spdk_bdev *bdev = ctx;
5097 :
5098 5 : if (bdev->internal.period == bdev->internal.new_period) {
5099 0 : return;
5100 : }
5101 :
5102 5 : if (bdev->internal.qd_poll_in_progress) {
5103 1 : return;
5104 : }
5105 :
5106 4 : bdev->internal.period = bdev->internal.new_period;
5107 :
5108 4 : spdk_poller_unregister(&bdev->internal.qd_poller);
5109 4 : if (bdev->internal.period != 0) {
5110 2 : bdev->internal.qd_poller = SPDK_POLLER_REGISTER(bdev_calculate_measured_queue_depth,
5111 : bdev, bdev->internal.period);
5112 2 : } else {
5113 2 : spdk_bdev_close(bdev->internal.qd_desc);
5114 2 : bdev->internal.qd_desc = NULL;
5115 : }
5116 5 : }
5117 :
5118 : static void
5119 0 : _tmp_bdev_event_cb(enum spdk_bdev_event_type type, struct spdk_bdev *bdev, void *ctx)
5120 : {
5121 0 : SPDK_NOTICELOG("Unexpected event type: %d\n", type);
5122 0 : }
5123 :
5124 : void
5125 130 : spdk_bdev_set_qd_sampling_period(struct spdk_bdev *bdev, uint64_t period)
5126 : {
5127 : int rc;
5128 :
5129 130 : if (bdev->internal.new_period == period) {
5130 124 : return;
5131 : }
5132 :
5133 6 : bdev->internal.new_period = period;
5134 :
5135 6 : if (bdev->internal.qd_desc != NULL) {
5136 4 : assert(bdev->internal.period != 0);
5137 :
5138 8 : spdk_thread_send_msg(bdev->internal.qd_desc->thread,
5139 4 : bdev_update_qd_sampling_period, bdev);
5140 4 : return;
5141 : }
5142 :
5143 2 : assert(bdev->internal.period == 0);
5144 :
5145 4 : rc = spdk_bdev_open_ext(spdk_bdev_get_name(bdev), false, _tmp_bdev_event_cb,
5146 2 : NULL, &bdev->internal.qd_desc);
5147 2 : if (rc != 0) {
5148 0 : return;
5149 : }
5150 :
5151 2 : bdev->internal.period = period;
5152 2 : bdev->internal.qd_poller = SPDK_POLLER_REGISTER(bdev_calculate_measured_queue_depth,
5153 : bdev, period);
5154 130 : }
5155 :
5156 : struct bdev_get_current_qd_ctx {
5157 : uint64_t current_qd;
5158 : spdk_bdev_get_current_qd_cb cb_fn;
5159 : void *cb_arg;
5160 : };
5161 :
5162 : static void
5163 0 : bdev_get_current_qd_done(struct spdk_bdev *bdev, void *_ctx, int status)
5164 : {
5165 0 : struct bdev_get_current_qd_ctx *ctx = _ctx;
5166 :
5167 0 : ctx->cb_fn(bdev, ctx->current_qd, ctx->cb_arg, 0);
5168 :
5169 0 : free(ctx);
5170 0 : }
5171 :
5172 : static void
5173 0 : bdev_get_current_qd(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
5174 : struct spdk_io_channel *io_ch, void *_ctx)
5175 : {
5176 0 : struct bdev_get_current_qd_ctx *ctx = _ctx;
5177 0 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
5178 :
5179 0 : ctx->current_qd += bdev_ch->io_outstanding;
5180 :
5181 0 : spdk_bdev_for_each_channel_continue(i, 0);
5182 0 : }
5183 :
5184 : void
5185 0 : spdk_bdev_get_current_qd(struct spdk_bdev *bdev, spdk_bdev_get_current_qd_cb cb_fn,
5186 : void *cb_arg)
5187 : {
5188 : struct bdev_get_current_qd_ctx *ctx;
5189 :
5190 0 : assert(cb_fn != NULL);
5191 :
5192 0 : ctx = calloc(1, sizeof(*ctx));
5193 0 : if (ctx == NULL) {
5194 0 : cb_fn(bdev, 0, cb_arg, -ENOMEM);
5195 0 : return;
5196 : }
5197 :
5198 0 : ctx->cb_fn = cb_fn;
5199 0 : ctx->cb_arg = cb_arg;
5200 :
5201 0 : spdk_bdev_for_each_channel(bdev, bdev_get_current_qd, ctx, bdev_get_current_qd_done);
5202 0 : }
5203 :
5204 : static void
5205 25 : _event_notify(struct spdk_bdev_desc *desc, enum spdk_bdev_event_type type)
5206 : {
5207 25 : assert(desc->thread == spdk_get_thread());
5208 :
5209 25 : spdk_spin_lock(&desc->spinlock);
5210 25 : desc->refs--;
5211 25 : if (!desc->closed) {
5212 14 : spdk_spin_unlock(&desc->spinlock);
5213 28 : desc->callback.event_fn(type,
5214 14 : desc->bdev,
5215 14 : desc->callback.ctx);
5216 14 : return;
5217 11 : } else if (desc->refs == 0) {
5218 : /* This descriptor was closed after this event_notify message was sent.
5219 : * spdk_bdev_close() could not free the descriptor since this message was
5220 : * in flight, so we free it now using bdev_desc_free().
5221 : */
5222 10 : spdk_spin_unlock(&desc->spinlock);
5223 10 : bdev_desc_free(desc);
5224 10 : return;
5225 : }
5226 1 : spdk_spin_unlock(&desc->spinlock);
5227 25 : }
5228 :
5229 : static void
5230 25 : event_notify(struct spdk_bdev_desc *desc, spdk_msg_fn event_notify_fn)
5231 : {
5232 25 : spdk_spin_lock(&desc->spinlock);
5233 25 : desc->refs++;
5234 25 : spdk_thread_send_msg(desc->thread, event_notify_fn, desc);
5235 25 : spdk_spin_unlock(&desc->spinlock);
5236 25 : }
5237 :
5238 : static void
5239 6 : _resize_notify(void *ctx)
5240 : {
5241 6 : struct spdk_bdev_desc *desc = ctx;
5242 :
5243 6 : _event_notify(desc, SPDK_BDEV_EVENT_RESIZE);
5244 6 : }
5245 :
5246 : int
5247 11 : spdk_bdev_notify_blockcnt_change(struct spdk_bdev *bdev, uint64_t size)
5248 : {
5249 : struct spdk_bdev_desc *desc;
5250 : int ret;
5251 :
5252 11 : if (size == bdev->blockcnt) {
5253 0 : return 0;
5254 : }
5255 :
5256 11 : spdk_spin_lock(&bdev->internal.spinlock);
5257 :
5258 : /* bdev has open descriptors */
5259 11 : if (!TAILQ_EMPTY(&bdev->internal.open_descs) &&
5260 7 : bdev->blockcnt > size) {
5261 1 : ret = -EBUSY;
5262 1 : } else {
5263 10 : bdev->blockcnt = size;
5264 16 : TAILQ_FOREACH(desc, &bdev->internal.open_descs, link) {
5265 6 : event_notify(desc, _resize_notify);
5266 6 : }
5267 10 : ret = 0;
5268 : }
5269 :
5270 11 : spdk_spin_unlock(&bdev->internal.spinlock);
5271 :
5272 11 : return ret;
5273 11 : }
5274 :
5275 : /*
5276 : * Convert I/O offset and length from bytes to blocks.
5277 : *
5278 : * Returns zero on success or non-zero if the byte parameters aren't divisible by the block size.
5279 : */
5280 : static uint64_t
5281 20 : bdev_bytes_to_blocks(struct spdk_bdev_desc *desc, uint64_t offset_bytes,
5282 : uint64_t *offset_blocks, uint64_t num_bytes, uint64_t *num_blocks)
5283 : {
5284 20 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5285 20 : uint32_t block_size = bdev->blocklen;
5286 : uint8_t shift_cnt;
5287 :
5288 : /* Avoid expensive div operations if possible. These spdk_u32 functions are very cheap. */
5289 20 : if (spdk_likely(spdk_u32_is_pow2(block_size))) {
5290 17 : shift_cnt = spdk_u32log2(block_size);
5291 17 : *offset_blocks = offset_bytes >> shift_cnt;
5292 17 : *num_blocks = num_bytes >> shift_cnt;
5293 34 : return (offset_bytes - (*offset_blocks << shift_cnt)) |
5294 17 : (num_bytes - (*num_blocks << shift_cnt));
5295 : } else {
5296 3 : *offset_blocks = offset_bytes / block_size;
5297 3 : *num_blocks = num_bytes / block_size;
5298 3 : return (offset_bytes % block_size) | (num_bytes % block_size);
5299 : }
5300 20 : }
5301 :
5302 : static bool
5303 689 : bdev_io_valid_blocks(struct spdk_bdev *bdev, uint64_t offset_blocks, uint64_t num_blocks)
5304 : {
5305 : /* Return failure if offset_blocks + num_blocks is less than offset_blocks; indicates there
5306 : * has been an overflow and hence the offset has been wrapped around */
5307 689 : if (offset_blocks + num_blocks < offset_blocks) {
5308 1 : return false;
5309 : }
5310 :
5311 : /* Return failure if offset_blocks + num_blocks exceeds the size of the bdev */
5312 688 : if (offset_blocks + num_blocks > bdev->blockcnt) {
5313 2 : return false;
5314 : }
5315 :
5316 686 : return true;
5317 689 : }
5318 :
5319 : static void
5320 2 : bdev_seek_complete_cb(void *ctx)
5321 : {
5322 2 : struct spdk_bdev_io *bdev_io = ctx;
5323 :
5324 2 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
5325 2 : bdev_io->internal.cb(bdev_io, true, bdev_io->internal.caller_ctx);
5326 2 : }
5327 :
5328 : static int
5329 4 : bdev_seek(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5330 : uint64_t offset_blocks, enum spdk_bdev_io_type io_type,
5331 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5332 : {
5333 4 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5334 : struct spdk_bdev_io *bdev_io;
5335 4 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5336 :
5337 4 : assert(io_type == SPDK_BDEV_IO_TYPE_SEEK_DATA || io_type == SPDK_BDEV_IO_TYPE_SEEK_HOLE);
5338 :
5339 : /* Check if offset_blocks is valid looking at the validity of one block */
5340 4 : if (!bdev_io_valid_blocks(bdev, offset_blocks, 1)) {
5341 0 : return -EINVAL;
5342 : }
5343 :
5344 4 : bdev_io = bdev_channel_get_io(channel);
5345 4 : if (!bdev_io) {
5346 0 : return -ENOMEM;
5347 : }
5348 :
5349 4 : bdev_io->internal.ch = channel;
5350 4 : bdev_io->internal.desc = desc;
5351 4 : bdev_io->type = io_type;
5352 4 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5353 4 : bdev_io->u.bdev.memory_domain = NULL;
5354 4 : bdev_io->u.bdev.memory_domain_ctx = NULL;
5355 4 : bdev_io->u.bdev.accel_sequence = NULL;
5356 4 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5357 :
5358 4 : if (!spdk_bdev_io_type_supported(bdev, io_type)) {
5359 : /* In case bdev doesn't support seek to next data/hole offset,
5360 : * it is assumed that only data and no holes are present */
5361 2 : if (io_type == SPDK_BDEV_IO_TYPE_SEEK_DATA) {
5362 1 : bdev_io->u.bdev.seek.offset = offset_blocks;
5363 1 : } else {
5364 1 : bdev_io->u.bdev.seek.offset = UINT64_MAX;
5365 : }
5366 :
5367 2 : spdk_thread_send_msg(spdk_get_thread(), bdev_seek_complete_cb, bdev_io);
5368 2 : return 0;
5369 : }
5370 :
5371 2 : bdev_io_submit(bdev_io);
5372 2 : return 0;
5373 4 : }
5374 :
5375 : int
5376 2 : spdk_bdev_seek_data(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5377 : uint64_t offset_blocks,
5378 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5379 : {
5380 2 : return bdev_seek(desc, ch, offset_blocks, SPDK_BDEV_IO_TYPE_SEEK_DATA, cb, cb_arg);
5381 : }
5382 :
5383 : int
5384 2 : spdk_bdev_seek_hole(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5385 : uint64_t offset_blocks,
5386 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5387 : {
5388 2 : return bdev_seek(desc, ch, offset_blocks, SPDK_BDEV_IO_TYPE_SEEK_HOLE, cb, cb_arg);
5389 : }
5390 :
5391 : uint64_t
5392 4 : spdk_bdev_io_get_seek_offset(const struct spdk_bdev_io *bdev_io)
5393 : {
5394 4 : return bdev_io->u.bdev.seek.offset;
5395 : }
5396 :
5397 : static int
5398 204 : bdev_read_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch, void *buf,
5399 : void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
5400 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5401 : {
5402 204 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5403 : struct spdk_bdev_io *bdev_io;
5404 204 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5405 :
5406 204 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
5407 0 : return -EINVAL;
5408 : }
5409 :
5410 204 : bdev_io = bdev_channel_get_io(channel);
5411 204 : if (!bdev_io) {
5412 1 : return -ENOMEM;
5413 : }
5414 :
5415 203 : bdev_io->internal.ch = channel;
5416 203 : bdev_io->internal.desc = desc;
5417 203 : bdev_io->type = SPDK_BDEV_IO_TYPE_READ;
5418 203 : bdev_io->u.bdev.iovs = &bdev_io->iov;
5419 203 : bdev_io->u.bdev.iovs[0].iov_base = buf;
5420 203 : bdev_io->u.bdev.iovs[0].iov_len = num_blocks * bdev->blocklen;
5421 203 : bdev_io->u.bdev.iovcnt = 1;
5422 203 : bdev_io->u.bdev.md_buf = md_buf;
5423 203 : bdev_io->u.bdev.num_blocks = num_blocks;
5424 203 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5425 203 : bdev_io->u.bdev.memory_domain = NULL;
5426 203 : bdev_io->u.bdev.memory_domain_ctx = NULL;
5427 203 : bdev_io->u.bdev.accel_sequence = NULL;
5428 203 : bdev_io->u.bdev.dif_check_flags = bdev->dif_check_flags;
5429 203 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5430 :
5431 203 : bdev_io_submit(bdev_io);
5432 203 : return 0;
5433 204 : }
5434 :
5435 : int
5436 3 : spdk_bdev_read(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5437 : void *buf, uint64_t offset, uint64_t nbytes,
5438 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5439 : {
5440 : uint64_t offset_blocks, num_blocks;
5441 :
5442 3 : if (bdev_bytes_to_blocks(desc, offset, &offset_blocks, nbytes, &num_blocks) != 0) {
5443 0 : return -EINVAL;
5444 : }
5445 :
5446 3 : return spdk_bdev_read_blocks(desc, ch, buf, offset_blocks, num_blocks, cb, cb_arg);
5447 3 : }
5448 :
5449 : int
5450 200 : spdk_bdev_read_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5451 : void *buf, uint64_t offset_blocks, uint64_t num_blocks,
5452 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5453 : {
5454 200 : return bdev_read_blocks_with_md(desc, ch, buf, NULL, offset_blocks, num_blocks, cb, cb_arg);
5455 : }
5456 :
5457 : int
5458 4 : spdk_bdev_read_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5459 : void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
5460 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5461 : {
5462 8 : struct iovec iov = {
5463 4 : .iov_base = buf,
5464 : };
5465 :
5466 4 : if (md_buf && !spdk_bdev_is_md_separate(spdk_bdev_desc_get_bdev(desc))) {
5467 0 : return -EINVAL;
5468 : }
5469 :
5470 4 : if (md_buf && !_is_buf_allocated(&iov)) {
5471 0 : return -EINVAL;
5472 : }
5473 :
5474 8 : return bdev_read_blocks_with_md(desc, ch, buf, md_buf, offset_blocks, num_blocks,
5475 4 : cb, cb_arg);
5476 4 : }
5477 :
5478 : int
5479 5 : spdk_bdev_readv(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5480 : struct iovec *iov, int iovcnt,
5481 : uint64_t offset, uint64_t nbytes,
5482 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5483 : {
5484 : uint64_t offset_blocks, num_blocks;
5485 :
5486 5 : if (bdev_bytes_to_blocks(desc, offset, &offset_blocks, nbytes, &num_blocks) != 0) {
5487 0 : return -EINVAL;
5488 : }
5489 :
5490 5 : return spdk_bdev_readv_blocks(desc, ch, iov, iovcnt, offset_blocks, num_blocks, cb, cb_arg);
5491 5 : }
5492 :
5493 : static int
5494 226 : bdev_readv_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5495 : struct iovec *iov, int iovcnt, void *md_buf, uint64_t offset_blocks,
5496 : uint64_t num_blocks, struct spdk_memory_domain *domain, void *domain_ctx,
5497 : struct spdk_accel_sequence *seq, uint32_t dif_check_flags,
5498 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5499 : {
5500 226 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5501 : struct spdk_bdev_io *bdev_io;
5502 226 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5503 :
5504 226 : if (spdk_unlikely(!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks))) {
5505 0 : return -EINVAL;
5506 : }
5507 :
5508 226 : bdev_io = bdev_channel_get_io(channel);
5509 226 : if (spdk_unlikely(!bdev_io)) {
5510 2 : return -ENOMEM;
5511 : }
5512 :
5513 224 : bdev_io->internal.ch = channel;
5514 224 : bdev_io->internal.desc = desc;
5515 224 : bdev_io->type = SPDK_BDEV_IO_TYPE_READ;
5516 224 : bdev_io->u.bdev.iovs = iov;
5517 224 : bdev_io->u.bdev.iovcnt = iovcnt;
5518 224 : bdev_io->u.bdev.md_buf = md_buf;
5519 224 : bdev_io->u.bdev.num_blocks = num_blocks;
5520 224 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5521 224 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5522 :
5523 224 : if (seq != NULL) {
5524 0 : bdev_io->internal.f.has_accel_sequence = true;
5525 0 : bdev_io->internal.accel_sequence = seq;
5526 0 : }
5527 :
5528 224 : if (domain != NULL) {
5529 2 : bdev_io->internal.f.has_memory_domain = true;
5530 2 : bdev_io->internal.memory_domain = domain;
5531 2 : bdev_io->internal.memory_domain_ctx = domain_ctx;
5532 2 : }
5533 :
5534 224 : bdev_io->u.bdev.memory_domain = domain;
5535 224 : bdev_io->u.bdev.memory_domain_ctx = domain_ctx;
5536 224 : bdev_io->u.bdev.accel_sequence = seq;
5537 224 : bdev_io->u.bdev.dif_check_flags = dif_check_flags;
5538 :
5539 224 : _bdev_io_submit_ext(desc, bdev_io);
5540 :
5541 224 : return 0;
5542 226 : }
5543 :
5544 : int
5545 21 : spdk_bdev_readv_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5546 : struct iovec *iov, int iovcnt,
5547 : uint64_t offset_blocks, uint64_t num_blocks,
5548 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5549 : {
5550 21 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5551 :
5552 42 : return bdev_readv_blocks_with_md(desc, ch, iov, iovcnt, NULL, offset_blocks,
5553 21 : num_blocks, NULL, NULL, NULL, bdev->dif_check_flags, cb, cb_arg);
5554 : }
5555 :
5556 : int
5557 4 : spdk_bdev_readv_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5558 : struct iovec *iov, int iovcnt, void *md_buf,
5559 : uint64_t offset_blocks, uint64_t num_blocks,
5560 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5561 : {
5562 4 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5563 :
5564 4 : if (md_buf && !spdk_bdev_is_md_separate(bdev)) {
5565 0 : return -EINVAL;
5566 : }
5567 :
5568 4 : if (md_buf && !_is_buf_allocated(iov)) {
5569 0 : return -EINVAL;
5570 : }
5571 :
5572 8 : return bdev_readv_blocks_with_md(desc, ch, iov, iovcnt, md_buf, offset_blocks,
5573 4 : num_blocks, NULL, NULL, NULL, bdev->dif_check_flags, cb, cb_arg);
5574 4 : }
5575 :
5576 : static inline bool
5577 14 : _bdev_io_check_opts(struct spdk_bdev_ext_io_opts *opts, struct iovec *iov)
5578 : {
5579 : /*
5580 : * We check if opts size is at least of size when we first introduced
5581 : * spdk_bdev_ext_io_opts (ac6f2bdd8d) since access to those members
5582 : * are not checked internal.
5583 : */
5584 24 : return opts->size >= offsetof(struct spdk_bdev_ext_io_opts, metadata) +
5585 14 : sizeof(opts->metadata) &&
5586 10 : opts->size <= sizeof(*opts) &&
5587 : /* When memory domain is used, the user must provide data buffers */
5588 8 : (!opts->memory_domain || (iov && iov[0].iov_base));
5589 : }
5590 :
5591 : int
5592 8 : spdk_bdev_readv_blocks_ext(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5593 : struct iovec *iov, int iovcnt,
5594 : uint64_t offset_blocks, uint64_t num_blocks,
5595 : spdk_bdev_io_completion_cb cb, void *cb_arg,
5596 : struct spdk_bdev_ext_io_opts *opts)
5597 : {
5598 8 : struct spdk_memory_domain *domain = NULL;
5599 8 : struct spdk_accel_sequence *seq = NULL;
5600 8 : void *domain_ctx = NULL, *md = NULL;
5601 8 : uint32_t dif_check_flags = 0;
5602 8 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5603 :
5604 8 : if (opts) {
5605 7 : if (spdk_unlikely(!_bdev_io_check_opts(opts, iov))) {
5606 3 : return -EINVAL;
5607 : }
5608 :
5609 4 : md = opts->metadata;
5610 4 : domain = bdev_get_ext_io_opt(opts, memory_domain, NULL);
5611 4 : domain_ctx = bdev_get_ext_io_opt(opts, memory_domain_ctx, NULL);
5612 4 : seq = bdev_get_ext_io_opt(opts, accel_sequence, NULL);
5613 4 : if (md) {
5614 4 : if (spdk_unlikely(!spdk_bdev_is_md_separate(bdev))) {
5615 0 : return -EINVAL;
5616 : }
5617 :
5618 4 : if (spdk_unlikely(!_is_buf_allocated(iov))) {
5619 0 : return -EINVAL;
5620 : }
5621 :
5622 4 : if (spdk_unlikely(seq != NULL)) {
5623 0 : return -EINVAL;
5624 : }
5625 4 : }
5626 4 : }
5627 :
5628 10 : dif_check_flags = bdev->dif_check_flags &
5629 5 : ~(bdev_get_ext_io_opt(opts, dif_check_flags_exclude_mask, 0));
5630 :
5631 10 : return bdev_readv_blocks_with_md(desc, ch, iov, iovcnt, md, offset_blocks,
5632 5 : num_blocks, domain, domain_ctx, seq, dif_check_flags, cb, cb_arg);
5633 8 : }
5634 :
5635 : static int
5636 36 : bdev_write_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5637 : void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
5638 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5639 : {
5640 36 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5641 : struct spdk_bdev_io *bdev_io;
5642 36 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5643 :
5644 36 : if (!desc->write) {
5645 0 : return -EBADF;
5646 : }
5647 :
5648 36 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
5649 0 : return -EINVAL;
5650 : }
5651 :
5652 36 : bdev_io = bdev_channel_get_io(channel);
5653 36 : if (!bdev_io) {
5654 0 : return -ENOMEM;
5655 : }
5656 :
5657 36 : bdev_io->internal.ch = channel;
5658 36 : bdev_io->internal.desc = desc;
5659 36 : bdev_io->type = SPDK_BDEV_IO_TYPE_WRITE;
5660 36 : bdev_io->u.bdev.iovs = &bdev_io->iov;
5661 36 : bdev_io->u.bdev.iovs[0].iov_base = buf;
5662 36 : bdev_io->u.bdev.iovs[0].iov_len = num_blocks * bdev->blocklen;
5663 36 : bdev_io->u.bdev.iovcnt = 1;
5664 36 : bdev_io->u.bdev.md_buf = md_buf;
5665 36 : bdev_io->u.bdev.num_blocks = num_blocks;
5666 36 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5667 36 : bdev_io->u.bdev.memory_domain = NULL;
5668 36 : bdev_io->u.bdev.memory_domain_ctx = NULL;
5669 36 : bdev_io->u.bdev.accel_sequence = NULL;
5670 36 : bdev_io->u.bdev.dif_check_flags = bdev->dif_check_flags;
5671 36 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5672 :
5673 36 : bdev_io_submit(bdev_io);
5674 36 : return 0;
5675 36 : }
5676 :
5677 : int
5678 3 : spdk_bdev_write(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5679 : void *buf, uint64_t offset, uint64_t nbytes,
5680 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5681 : {
5682 : uint64_t offset_blocks, num_blocks;
5683 :
5684 3 : if (bdev_bytes_to_blocks(desc, offset, &offset_blocks, nbytes, &num_blocks) != 0) {
5685 0 : return -EINVAL;
5686 : }
5687 :
5688 3 : return spdk_bdev_write_blocks(desc, ch, buf, offset_blocks, num_blocks, cb, cb_arg);
5689 3 : }
5690 :
5691 : int
5692 27 : spdk_bdev_write_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5693 : void *buf, uint64_t offset_blocks, uint64_t num_blocks,
5694 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5695 : {
5696 54 : return bdev_write_blocks_with_md(desc, ch, buf, NULL, offset_blocks, num_blocks,
5697 27 : cb, cb_arg);
5698 : }
5699 :
5700 : int
5701 3 : spdk_bdev_write_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5702 : void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
5703 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5704 : {
5705 6 : struct iovec iov = {
5706 3 : .iov_base = buf,
5707 : };
5708 :
5709 3 : if (md_buf && !spdk_bdev_is_md_separate(spdk_bdev_desc_get_bdev(desc))) {
5710 0 : return -EINVAL;
5711 : }
5712 :
5713 3 : if (md_buf && !_is_buf_allocated(&iov)) {
5714 0 : return -EINVAL;
5715 : }
5716 :
5717 6 : return bdev_write_blocks_with_md(desc, ch, buf, md_buf, offset_blocks, num_blocks,
5718 3 : cb, cb_arg);
5719 3 : }
5720 :
5721 : static int
5722 70 : bdev_writev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5723 : struct iovec *iov, int iovcnt, void *md_buf,
5724 : uint64_t offset_blocks, uint64_t num_blocks,
5725 : struct spdk_memory_domain *domain, void *domain_ctx,
5726 : struct spdk_accel_sequence *seq, uint32_t dif_check_flags,
5727 : uint32_t nvme_cdw12_raw, uint32_t nvme_cdw13_raw,
5728 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5729 : {
5730 70 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5731 : struct spdk_bdev_io *bdev_io;
5732 70 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5733 :
5734 70 : if (spdk_unlikely(!desc->write)) {
5735 0 : return -EBADF;
5736 : }
5737 :
5738 70 : if (spdk_unlikely(!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks))) {
5739 0 : return -EINVAL;
5740 : }
5741 :
5742 70 : bdev_io = bdev_channel_get_io(channel);
5743 70 : if (spdk_unlikely(!bdev_io)) {
5744 2 : return -ENOMEM;
5745 : }
5746 :
5747 68 : bdev_io->internal.ch = channel;
5748 68 : bdev_io->internal.desc = desc;
5749 68 : bdev_io->type = SPDK_BDEV_IO_TYPE_WRITE;
5750 68 : bdev_io->u.bdev.iovs = iov;
5751 68 : bdev_io->u.bdev.iovcnt = iovcnt;
5752 68 : bdev_io->u.bdev.md_buf = md_buf;
5753 68 : bdev_io->u.bdev.num_blocks = num_blocks;
5754 68 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5755 68 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5756 68 : if (seq != NULL) {
5757 0 : bdev_io->internal.f.has_accel_sequence = true;
5758 0 : bdev_io->internal.accel_sequence = seq;
5759 0 : }
5760 :
5761 68 : if (domain != NULL) {
5762 2 : bdev_io->internal.f.has_memory_domain = true;
5763 2 : bdev_io->internal.memory_domain = domain;
5764 2 : bdev_io->internal.memory_domain_ctx = domain_ctx;
5765 2 : }
5766 :
5767 68 : bdev_io->u.bdev.memory_domain = domain;
5768 68 : bdev_io->u.bdev.memory_domain_ctx = domain_ctx;
5769 68 : bdev_io->u.bdev.accel_sequence = seq;
5770 68 : bdev_io->u.bdev.dif_check_flags = dif_check_flags;
5771 68 : bdev_io->u.bdev.nvme_cdw12.raw = nvme_cdw12_raw;
5772 68 : bdev_io->u.bdev.nvme_cdw13.raw = nvme_cdw13_raw;
5773 :
5774 68 : _bdev_io_submit_ext(desc, bdev_io);
5775 :
5776 68 : return 0;
5777 70 : }
5778 :
5779 : int
5780 3 : spdk_bdev_writev(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5781 : struct iovec *iov, int iovcnt,
5782 : uint64_t offset, uint64_t len,
5783 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5784 : {
5785 : uint64_t offset_blocks, num_blocks;
5786 :
5787 3 : if (bdev_bytes_to_blocks(desc, offset, &offset_blocks, len, &num_blocks) != 0) {
5788 0 : return -EINVAL;
5789 : }
5790 :
5791 3 : return spdk_bdev_writev_blocks(desc, ch, iov, iovcnt, offset_blocks, num_blocks, cb, cb_arg);
5792 3 : }
5793 :
5794 : int
5795 14 : spdk_bdev_writev_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5796 : struct iovec *iov, int iovcnt,
5797 : uint64_t offset_blocks, uint64_t num_blocks,
5798 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5799 : {
5800 14 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5801 :
5802 28 : return bdev_writev_blocks_with_md(desc, ch, iov, iovcnt, NULL, offset_blocks,
5803 14 : num_blocks, NULL, NULL, NULL, bdev->dif_check_flags, 0, 0,
5804 14 : cb, cb_arg);
5805 : }
5806 :
5807 : int
5808 1 : spdk_bdev_writev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5809 : struct iovec *iov, int iovcnt, void *md_buf,
5810 : uint64_t offset_blocks, uint64_t num_blocks,
5811 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5812 : {
5813 1 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5814 :
5815 1 : if (md_buf && !spdk_bdev_is_md_separate(bdev)) {
5816 0 : return -EINVAL;
5817 : }
5818 :
5819 1 : if (md_buf && !_is_buf_allocated(iov)) {
5820 0 : return -EINVAL;
5821 : }
5822 :
5823 2 : return bdev_writev_blocks_with_md(desc, ch, iov, iovcnt, md_buf, offset_blocks,
5824 1 : num_blocks, NULL, NULL, NULL, bdev->dif_check_flags, 0, 0,
5825 1 : cb, cb_arg);
5826 1 : }
5827 :
5828 : int
5829 8 : spdk_bdev_writev_blocks_ext(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5830 : struct iovec *iov, int iovcnt,
5831 : uint64_t offset_blocks, uint64_t num_blocks,
5832 : spdk_bdev_io_completion_cb cb, void *cb_arg,
5833 : struct spdk_bdev_ext_io_opts *opts)
5834 : {
5835 8 : struct spdk_memory_domain *domain = NULL;
5836 8 : struct spdk_accel_sequence *seq = NULL;
5837 8 : void *domain_ctx = NULL, *md = NULL;
5838 8 : uint32_t dif_check_flags = 0;
5839 8 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5840 8 : uint32_t nvme_cdw12_raw = 0;
5841 8 : uint32_t nvme_cdw13_raw = 0;
5842 :
5843 8 : if (opts) {
5844 7 : if (spdk_unlikely(!_bdev_io_check_opts(opts, iov))) {
5845 3 : return -EINVAL;
5846 : }
5847 4 : md = opts->metadata;
5848 4 : domain = bdev_get_ext_io_opt(opts, memory_domain, NULL);
5849 4 : domain_ctx = bdev_get_ext_io_opt(opts, memory_domain_ctx, NULL);
5850 4 : seq = bdev_get_ext_io_opt(opts, accel_sequence, NULL);
5851 4 : nvme_cdw12_raw = bdev_get_ext_io_opt(opts, nvme_cdw12.raw, 0);
5852 4 : nvme_cdw13_raw = bdev_get_ext_io_opt(opts, nvme_cdw13.raw, 0);
5853 4 : if (md) {
5854 4 : if (spdk_unlikely(!spdk_bdev_is_md_separate(bdev))) {
5855 0 : return -EINVAL;
5856 : }
5857 :
5858 4 : if (spdk_unlikely(!_is_buf_allocated(iov))) {
5859 0 : return -EINVAL;
5860 : }
5861 :
5862 4 : if (spdk_unlikely(seq != NULL)) {
5863 0 : return -EINVAL;
5864 : }
5865 4 : }
5866 4 : }
5867 :
5868 10 : dif_check_flags = bdev->dif_check_flags &
5869 5 : ~(bdev_get_ext_io_opt(opts, dif_check_flags_exclude_mask, 0));
5870 :
5871 10 : return bdev_writev_blocks_with_md(desc, ch, iov, iovcnt, md, offset_blocks, num_blocks,
5872 5 : domain, domain_ctx, seq, dif_check_flags,
5873 5 : nvme_cdw12_raw, nvme_cdw13_raw, cb, cb_arg);
5874 8 : }
5875 :
5876 : static void
5877 11 : bdev_compare_do_read_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
5878 : {
5879 11 : struct spdk_bdev_io *parent_io = cb_arg;
5880 11 : struct spdk_bdev *bdev = parent_io->bdev;
5881 11 : uint8_t *read_buf = bdev_io->u.bdev.iovs[0].iov_base;
5882 11 : int i, rc = 0;
5883 :
5884 11 : if (!success) {
5885 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
5886 0 : parent_io->internal.cb(parent_io, false, parent_io->internal.caller_ctx);
5887 0 : spdk_bdev_free_io(bdev_io);
5888 0 : return;
5889 : }
5890 :
5891 17 : for (i = 0; i < parent_io->u.bdev.iovcnt; i++) {
5892 22 : rc = memcmp(read_buf,
5893 11 : parent_io->u.bdev.iovs[i].iov_base,
5894 11 : parent_io->u.bdev.iovs[i].iov_len);
5895 11 : if (rc) {
5896 5 : break;
5897 : }
5898 6 : read_buf += parent_io->u.bdev.iovs[i].iov_len;
5899 6 : }
5900 :
5901 11 : if (rc == 0 && parent_io->u.bdev.md_buf && spdk_bdev_is_md_separate(bdev)) {
5902 4 : rc = memcmp(bdev_io->u.bdev.md_buf,
5903 2 : parent_io->u.bdev.md_buf,
5904 2 : spdk_bdev_get_md_size(bdev));
5905 2 : }
5906 :
5907 11 : spdk_bdev_free_io(bdev_io);
5908 :
5909 11 : if (rc == 0) {
5910 5 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
5911 5 : parent_io->internal.cb(parent_io, true, parent_io->internal.caller_ctx);
5912 5 : } else {
5913 6 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_MISCOMPARE;
5914 6 : parent_io->internal.cb(parent_io, false, parent_io->internal.caller_ctx);
5915 : }
5916 11 : }
5917 :
5918 : static void
5919 11 : bdev_compare_do_read(void *_bdev_io)
5920 : {
5921 11 : struct spdk_bdev_io *bdev_io = _bdev_io;
5922 : int rc;
5923 :
5924 22 : rc = spdk_bdev_read_blocks(bdev_io->internal.desc,
5925 11 : spdk_io_channel_from_ctx(bdev_io->internal.ch), NULL,
5926 11 : bdev_io->u.bdev.offset_blocks, bdev_io->u.bdev.num_blocks,
5927 11 : bdev_compare_do_read_done, bdev_io);
5928 :
5929 11 : if (rc == -ENOMEM) {
5930 0 : bdev_queue_io_wait_with_cb(bdev_io, bdev_compare_do_read);
5931 11 : } else if (rc != 0) {
5932 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
5933 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
5934 0 : }
5935 11 : }
5936 :
5937 : static int
5938 16 : bdev_comparev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5939 : struct iovec *iov, int iovcnt, void *md_buf,
5940 : uint64_t offset_blocks, uint64_t num_blocks,
5941 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5942 : {
5943 16 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5944 : struct spdk_bdev_io *bdev_io;
5945 16 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5946 :
5947 16 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
5948 0 : return -EINVAL;
5949 : }
5950 :
5951 16 : bdev_io = bdev_channel_get_io(channel);
5952 16 : if (!bdev_io) {
5953 0 : return -ENOMEM;
5954 : }
5955 :
5956 16 : bdev_io->internal.ch = channel;
5957 16 : bdev_io->internal.desc = desc;
5958 16 : bdev_io->type = SPDK_BDEV_IO_TYPE_COMPARE;
5959 16 : bdev_io->u.bdev.iovs = iov;
5960 16 : bdev_io->u.bdev.iovcnt = iovcnt;
5961 16 : bdev_io->u.bdev.md_buf = md_buf;
5962 16 : bdev_io->u.bdev.num_blocks = num_blocks;
5963 16 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5964 16 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5965 16 : bdev_io->u.bdev.memory_domain = NULL;
5966 16 : bdev_io->u.bdev.memory_domain_ctx = NULL;
5967 16 : bdev_io->u.bdev.accel_sequence = NULL;
5968 :
5969 16 : if (bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COMPARE)) {
5970 7 : bdev_io_submit(bdev_io);
5971 7 : return 0;
5972 : }
5973 :
5974 9 : bdev_compare_do_read(bdev_io);
5975 :
5976 9 : return 0;
5977 16 : }
5978 :
5979 : int
5980 10 : spdk_bdev_comparev_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5981 : struct iovec *iov, int iovcnt,
5982 : uint64_t offset_blocks, uint64_t num_blocks,
5983 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5984 : {
5985 20 : return bdev_comparev_blocks_with_md(desc, ch, iov, iovcnt, NULL, offset_blocks,
5986 10 : num_blocks, cb, cb_arg);
5987 : }
5988 :
5989 : int
5990 6 : spdk_bdev_comparev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5991 : struct iovec *iov, int iovcnt, void *md_buf,
5992 : uint64_t offset_blocks, uint64_t num_blocks,
5993 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5994 : {
5995 6 : if (md_buf && !spdk_bdev_is_md_separate(spdk_bdev_desc_get_bdev(desc))) {
5996 0 : return -EINVAL;
5997 : }
5998 :
5999 6 : if (md_buf && !_is_buf_allocated(iov)) {
6000 0 : return -EINVAL;
6001 : }
6002 :
6003 12 : return bdev_comparev_blocks_with_md(desc, ch, iov, iovcnt, md_buf, offset_blocks,
6004 6 : num_blocks, cb, cb_arg);
6005 6 : }
6006 :
6007 : static int
6008 4 : bdev_compare_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6009 : void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
6010 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6011 : {
6012 4 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6013 : struct spdk_bdev_io *bdev_io;
6014 4 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6015 :
6016 4 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6017 0 : return -EINVAL;
6018 : }
6019 :
6020 4 : bdev_io = bdev_channel_get_io(channel);
6021 4 : if (!bdev_io) {
6022 0 : return -ENOMEM;
6023 : }
6024 :
6025 4 : bdev_io->internal.ch = channel;
6026 4 : bdev_io->internal.desc = desc;
6027 4 : bdev_io->type = SPDK_BDEV_IO_TYPE_COMPARE;
6028 4 : bdev_io->u.bdev.iovs = &bdev_io->iov;
6029 4 : bdev_io->u.bdev.iovs[0].iov_base = buf;
6030 4 : bdev_io->u.bdev.iovs[0].iov_len = num_blocks * bdev->blocklen;
6031 4 : bdev_io->u.bdev.iovcnt = 1;
6032 4 : bdev_io->u.bdev.md_buf = md_buf;
6033 4 : bdev_io->u.bdev.num_blocks = num_blocks;
6034 4 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6035 4 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6036 4 : bdev_io->u.bdev.memory_domain = NULL;
6037 4 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6038 4 : bdev_io->u.bdev.accel_sequence = NULL;
6039 :
6040 4 : if (bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COMPARE)) {
6041 2 : bdev_io_submit(bdev_io);
6042 2 : return 0;
6043 : }
6044 :
6045 2 : bdev_compare_do_read(bdev_io);
6046 :
6047 2 : return 0;
6048 4 : }
6049 :
6050 : int
6051 4 : spdk_bdev_compare_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6052 : void *buf, uint64_t offset_blocks, uint64_t num_blocks,
6053 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6054 : {
6055 8 : return bdev_compare_blocks_with_md(desc, ch, buf, NULL, offset_blocks, num_blocks,
6056 4 : cb, cb_arg);
6057 : }
6058 :
6059 : int
6060 0 : spdk_bdev_compare_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6061 : void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
6062 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6063 : {
6064 0 : struct iovec iov = {
6065 0 : .iov_base = buf,
6066 : };
6067 :
6068 0 : if (md_buf && !spdk_bdev_is_md_separate(spdk_bdev_desc_get_bdev(desc))) {
6069 0 : return -EINVAL;
6070 : }
6071 :
6072 0 : if (md_buf && !_is_buf_allocated(&iov)) {
6073 0 : return -EINVAL;
6074 : }
6075 :
6076 0 : return bdev_compare_blocks_with_md(desc, ch, buf, md_buf, offset_blocks, num_blocks,
6077 0 : cb, cb_arg);
6078 0 : }
6079 :
6080 : static void
6081 2 : bdev_comparev_and_writev_blocks_unlocked(struct lba_range *range, void *ctx, int unlock_status)
6082 : {
6083 2 : struct spdk_bdev_io *bdev_io = ctx;
6084 :
6085 2 : if (unlock_status) {
6086 0 : SPDK_ERRLOG("LBA range unlock failed\n");
6087 0 : }
6088 :
6089 4 : bdev_io->internal.cb(bdev_io, bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS ? true :
6090 2 : false, bdev_io->internal.caller_ctx);
6091 2 : }
6092 :
6093 : static void
6094 2 : bdev_comparev_and_writev_blocks_unlock(struct spdk_bdev_io *bdev_io, int status)
6095 : {
6096 2 : bdev_io->internal.status = status;
6097 :
6098 4 : bdev_unlock_lba_range(bdev_io->internal.desc, spdk_io_channel_from_ctx(bdev_io->internal.ch),
6099 2 : bdev_io->u.bdev.offset_blocks, bdev_io->u.bdev.num_blocks,
6100 2 : bdev_comparev_and_writev_blocks_unlocked, bdev_io);
6101 2 : }
6102 :
6103 : static void
6104 1 : bdev_compare_and_write_do_write_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
6105 : {
6106 1 : struct spdk_bdev_io *parent_io = cb_arg;
6107 :
6108 1 : if (!success) {
6109 0 : SPDK_ERRLOG("Compare and write operation failed\n");
6110 0 : }
6111 :
6112 1 : spdk_bdev_free_io(bdev_io);
6113 :
6114 2 : bdev_comparev_and_writev_blocks_unlock(parent_io,
6115 1 : success ? SPDK_BDEV_IO_STATUS_SUCCESS : SPDK_BDEV_IO_STATUS_FAILED);
6116 1 : }
6117 :
6118 : static void
6119 1 : bdev_compare_and_write_do_write(void *_bdev_io)
6120 : {
6121 1 : struct spdk_bdev_io *bdev_io = _bdev_io;
6122 : int rc;
6123 :
6124 2 : rc = spdk_bdev_writev_blocks(bdev_io->internal.desc,
6125 1 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
6126 1 : bdev_io->u.bdev.fused_iovs, bdev_io->u.bdev.fused_iovcnt,
6127 1 : bdev_io->u.bdev.offset_blocks, bdev_io->u.bdev.num_blocks,
6128 1 : bdev_compare_and_write_do_write_done, bdev_io);
6129 :
6130 :
6131 1 : if (rc == -ENOMEM) {
6132 0 : bdev_queue_io_wait_with_cb(bdev_io, bdev_compare_and_write_do_write);
6133 1 : } else if (rc != 0) {
6134 0 : bdev_comparev_and_writev_blocks_unlock(bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
6135 0 : }
6136 1 : }
6137 :
6138 : static void
6139 2 : bdev_compare_and_write_do_compare_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
6140 : {
6141 2 : struct spdk_bdev_io *parent_io = cb_arg;
6142 :
6143 2 : spdk_bdev_free_io(bdev_io);
6144 :
6145 2 : if (!success) {
6146 1 : bdev_comparev_and_writev_blocks_unlock(parent_io, SPDK_BDEV_IO_STATUS_MISCOMPARE);
6147 1 : return;
6148 : }
6149 :
6150 1 : bdev_compare_and_write_do_write(parent_io);
6151 2 : }
6152 :
6153 : static void
6154 2 : bdev_compare_and_write_do_compare(void *_bdev_io)
6155 : {
6156 2 : struct spdk_bdev_io *bdev_io = _bdev_io;
6157 : int rc;
6158 :
6159 4 : rc = spdk_bdev_comparev_blocks(bdev_io->internal.desc,
6160 2 : spdk_io_channel_from_ctx(bdev_io->internal.ch), bdev_io->u.bdev.iovs,
6161 2 : bdev_io->u.bdev.iovcnt, bdev_io->u.bdev.offset_blocks, bdev_io->u.bdev.num_blocks,
6162 2 : bdev_compare_and_write_do_compare_done, bdev_io);
6163 :
6164 2 : if (rc == -ENOMEM) {
6165 0 : bdev_queue_io_wait_with_cb(bdev_io, bdev_compare_and_write_do_compare);
6166 2 : } else if (rc != 0) {
6167 0 : bdev_comparev_and_writev_blocks_unlock(bdev_io, SPDK_BDEV_IO_STATUS_FIRST_FUSED_FAILED);
6168 0 : }
6169 2 : }
6170 :
6171 : static void
6172 2 : bdev_comparev_and_writev_blocks_locked(struct lba_range *range, void *ctx, int status)
6173 : {
6174 2 : struct spdk_bdev_io *bdev_io = ctx;
6175 :
6176 2 : if (status) {
6177 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FIRST_FUSED_FAILED;
6178 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
6179 0 : return;
6180 : }
6181 :
6182 2 : bdev_compare_and_write_do_compare(bdev_io);
6183 2 : }
6184 :
6185 : int
6186 2 : spdk_bdev_comparev_and_writev_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6187 : struct iovec *compare_iov, int compare_iovcnt,
6188 : struct iovec *write_iov, int write_iovcnt,
6189 : uint64_t offset_blocks, uint64_t num_blocks,
6190 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6191 : {
6192 2 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6193 : struct spdk_bdev_io *bdev_io;
6194 2 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6195 :
6196 2 : if (!desc->write) {
6197 0 : return -EBADF;
6198 : }
6199 :
6200 2 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6201 0 : return -EINVAL;
6202 : }
6203 :
6204 2 : if (num_blocks > bdev->acwu) {
6205 0 : return -EINVAL;
6206 : }
6207 :
6208 2 : bdev_io = bdev_channel_get_io(channel);
6209 2 : if (!bdev_io) {
6210 0 : return -ENOMEM;
6211 : }
6212 :
6213 2 : bdev_io->internal.ch = channel;
6214 2 : bdev_io->internal.desc = desc;
6215 2 : bdev_io->type = SPDK_BDEV_IO_TYPE_COMPARE_AND_WRITE;
6216 2 : bdev_io->u.bdev.iovs = compare_iov;
6217 2 : bdev_io->u.bdev.iovcnt = compare_iovcnt;
6218 2 : bdev_io->u.bdev.fused_iovs = write_iov;
6219 2 : bdev_io->u.bdev.fused_iovcnt = write_iovcnt;
6220 2 : bdev_io->u.bdev.md_buf = NULL;
6221 2 : bdev_io->u.bdev.num_blocks = num_blocks;
6222 2 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6223 2 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6224 2 : bdev_io->u.bdev.memory_domain = NULL;
6225 2 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6226 2 : bdev_io->u.bdev.accel_sequence = NULL;
6227 :
6228 2 : if (bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COMPARE_AND_WRITE)) {
6229 0 : bdev_io_submit(bdev_io);
6230 0 : return 0;
6231 : }
6232 :
6233 4 : return bdev_lock_lba_range(desc, ch, offset_blocks, num_blocks,
6234 2 : bdev_comparev_and_writev_blocks_locked, bdev_io);
6235 2 : }
6236 :
6237 : int
6238 2 : spdk_bdev_zcopy_start(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6239 : struct iovec *iov, int iovcnt,
6240 : uint64_t offset_blocks, uint64_t num_blocks,
6241 : bool populate,
6242 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6243 : {
6244 2 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6245 : struct spdk_bdev_io *bdev_io;
6246 2 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6247 :
6248 2 : if (!desc->write) {
6249 0 : return -EBADF;
6250 : }
6251 :
6252 2 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6253 0 : return -EINVAL;
6254 : }
6255 :
6256 2 : if (!spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_ZCOPY)) {
6257 0 : return -ENOTSUP;
6258 : }
6259 :
6260 2 : bdev_io = bdev_channel_get_io(channel);
6261 2 : if (!bdev_io) {
6262 0 : return -ENOMEM;
6263 : }
6264 :
6265 2 : bdev_io->internal.ch = channel;
6266 2 : bdev_io->internal.desc = desc;
6267 2 : bdev_io->type = SPDK_BDEV_IO_TYPE_ZCOPY;
6268 2 : bdev_io->u.bdev.num_blocks = num_blocks;
6269 2 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6270 2 : bdev_io->u.bdev.iovs = iov;
6271 2 : bdev_io->u.bdev.iovcnt = iovcnt;
6272 2 : bdev_io->u.bdev.md_buf = NULL;
6273 2 : bdev_io->u.bdev.zcopy.populate = populate ? 1 : 0;
6274 2 : bdev_io->u.bdev.zcopy.commit = 0;
6275 2 : bdev_io->u.bdev.zcopy.start = 1;
6276 2 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6277 2 : bdev_io->u.bdev.memory_domain = NULL;
6278 2 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6279 2 : bdev_io->u.bdev.accel_sequence = NULL;
6280 :
6281 2 : bdev_io_submit(bdev_io);
6282 :
6283 2 : return 0;
6284 2 : }
6285 :
6286 : int
6287 2 : spdk_bdev_zcopy_end(struct spdk_bdev_io *bdev_io, bool commit,
6288 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6289 : {
6290 2 : if (bdev_io->type != SPDK_BDEV_IO_TYPE_ZCOPY) {
6291 0 : return -EINVAL;
6292 : }
6293 :
6294 2 : bdev_io->u.bdev.zcopy.commit = commit ? 1 : 0;
6295 2 : bdev_io->u.bdev.zcopy.start = 0;
6296 2 : bdev_io->internal.caller_ctx = cb_arg;
6297 2 : bdev_io->internal.cb = cb;
6298 2 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_PENDING;
6299 :
6300 2 : bdev_io_submit(bdev_io);
6301 :
6302 2 : return 0;
6303 2 : }
6304 :
6305 : int
6306 0 : spdk_bdev_write_zeroes(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6307 : uint64_t offset, uint64_t len,
6308 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6309 : {
6310 : uint64_t offset_blocks, num_blocks;
6311 :
6312 0 : if (bdev_bytes_to_blocks(desc, offset, &offset_blocks, len, &num_blocks) != 0) {
6313 0 : return -EINVAL;
6314 : }
6315 :
6316 0 : return spdk_bdev_write_zeroes_blocks(desc, ch, offset_blocks, num_blocks, cb, cb_arg);
6317 0 : }
6318 :
6319 : int
6320 33 : spdk_bdev_write_zeroes_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6321 : uint64_t offset_blocks, uint64_t num_blocks,
6322 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6323 : {
6324 33 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6325 : struct spdk_bdev_io *bdev_io;
6326 33 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6327 :
6328 33 : if (!desc->write) {
6329 0 : return -EBADF;
6330 : }
6331 :
6332 33 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6333 0 : return -EINVAL;
6334 : }
6335 :
6336 33 : if (!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE_ZEROES) &&
6337 10 : !bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE)) {
6338 1 : return -ENOTSUP;
6339 : }
6340 :
6341 32 : bdev_io = bdev_channel_get_io(channel);
6342 :
6343 32 : if (!bdev_io) {
6344 0 : return -ENOMEM;
6345 : }
6346 :
6347 32 : bdev_io->type = SPDK_BDEV_IO_TYPE_WRITE_ZEROES;
6348 32 : bdev_io->internal.ch = channel;
6349 32 : bdev_io->internal.desc = desc;
6350 32 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6351 32 : bdev_io->u.bdev.num_blocks = num_blocks;
6352 32 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6353 32 : bdev_io->u.bdev.memory_domain = NULL;
6354 32 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6355 32 : bdev_io->u.bdev.accel_sequence = NULL;
6356 :
6357 : /* If the write_zeroes size is large and should be split, use the generic split
6358 : * logic regardless of whether SPDK_BDEV_IO_TYPE_WRITE_ZEREOS is supported or not.
6359 : *
6360 : * Then, send the write_zeroes request if SPDK_BDEV_IO_TYPE_WRITE_ZEROES is supported
6361 : * or emulate it using regular write request otherwise.
6362 : */
6363 32 : if (bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE_ZEROES) ||
6364 9 : bdev_io->internal.f.split) {
6365 26 : bdev_io_submit(bdev_io);
6366 26 : return 0;
6367 : }
6368 :
6369 6 : assert(_bdev_get_block_size_with_md(bdev) <= ZERO_BUFFER_SIZE);
6370 :
6371 6 : return bdev_write_zero_buffer(bdev_io);
6372 33 : }
6373 :
6374 : int
6375 0 : spdk_bdev_unmap(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6376 : uint64_t offset, uint64_t nbytes,
6377 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6378 : {
6379 : uint64_t offset_blocks, num_blocks;
6380 :
6381 0 : if (bdev_bytes_to_blocks(desc, offset, &offset_blocks, nbytes, &num_blocks) != 0) {
6382 0 : return -EINVAL;
6383 : }
6384 :
6385 0 : return spdk_bdev_unmap_blocks(desc, ch, offset_blocks, num_blocks, cb, cb_arg);
6386 0 : }
6387 :
6388 : static void
6389 0 : bdev_io_complete_cb(void *ctx)
6390 : {
6391 0 : struct spdk_bdev_io *bdev_io = ctx;
6392 :
6393 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
6394 0 : bdev_io->internal.cb(bdev_io, true, bdev_io->internal.caller_ctx);
6395 0 : }
6396 :
6397 : int
6398 22 : spdk_bdev_unmap_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6399 : uint64_t offset_blocks, uint64_t num_blocks,
6400 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6401 : {
6402 22 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6403 : struct spdk_bdev_io *bdev_io;
6404 22 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6405 :
6406 22 : if (!desc->write) {
6407 0 : return -EBADF;
6408 : }
6409 :
6410 22 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6411 0 : return -EINVAL;
6412 : }
6413 :
6414 22 : bdev_io = bdev_channel_get_io(channel);
6415 22 : if (!bdev_io) {
6416 0 : return -ENOMEM;
6417 : }
6418 :
6419 22 : bdev_io->internal.ch = channel;
6420 22 : bdev_io->internal.desc = desc;
6421 22 : bdev_io->type = SPDK_BDEV_IO_TYPE_UNMAP;
6422 :
6423 22 : bdev_io->u.bdev.iovs = &bdev_io->iov;
6424 22 : bdev_io->u.bdev.iovs[0].iov_base = NULL;
6425 22 : bdev_io->u.bdev.iovs[0].iov_len = 0;
6426 22 : bdev_io->u.bdev.iovcnt = 1;
6427 :
6428 22 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6429 22 : bdev_io->u.bdev.num_blocks = num_blocks;
6430 22 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6431 22 : bdev_io->u.bdev.memory_domain = NULL;
6432 22 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6433 22 : bdev_io->u.bdev.accel_sequence = NULL;
6434 :
6435 22 : if (num_blocks == 0) {
6436 0 : spdk_thread_send_msg(spdk_get_thread(), bdev_io_complete_cb, bdev_io);
6437 0 : return 0;
6438 : }
6439 :
6440 22 : bdev_io_submit(bdev_io);
6441 22 : return 0;
6442 22 : }
6443 :
6444 : int
6445 0 : spdk_bdev_flush(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6446 : uint64_t offset, uint64_t length,
6447 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6448 : {
6449 : uint64_t offset_blocks, num_blocks;
6450 :
6451 0 : if (bdev_bytes_to_blocks(desc, offset, &offset_blocks, length, &num_blocks) != 0) {
6452 0 : return -EINVAL;
6453 : }
6454 :
6455 0 : return spdk_bdev_flush_blocks(desc, ch, offset_blocks, num_blocks, cb, cb_arg);
6456 0 : }
6457 :
6458 : int
6459 2 : spdk_bdev_flush_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6460 : uint64_t offset_blocks, uint64_t num_blocks,
6461 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6462 : {
6463 2 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6464 : struct spdk_bdev_io *bdev_io;
6465 2 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6466 :
6467 2 : if (!desc->write) {
6468 0 : return -EBADF;
6469 : }
6470 :
6471 2 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6472 0 : return -EINVAL;
6473 : }
6474 :
6475 2 : bdev_io = bdev_channel_get_io(channel);
6476 2 : if (!bdev_io) {
6477 0 : return -ENOMEM;
6478 : }
6479 :
6480 2 : bdev_io->internal.ch = channel;
6481 2 : bdev_io->internal.desc = desc;
6482 2 : bdev_io->type = SPDK_BDEV_IO_TYPE_FLUSH;
6483 2 : bdev_io->u.bdev.iovs = NULL;
6484 2 : bdev_io->u.bdev.iovcnt = 0;
6485 2 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6486 2 : bdev_io->u.bdev.num_blocks = num_blocks;
6487 2 : bdev_io->u.bdev.memory_domain = NULL;
6488 2 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6489 2 : bdev_io->u.bdev.accel_sequence = NULL;
6490 2 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6491 :
6492 2 : bdev_io_submit(bdev_io);
6493 2 : return 0;
6494 2 : }
6495 :
6496 : static int bdev_reset_poll_for_outstanding_io(void *ctx);
6497 :
6498 : static void
6499 13 : bdev_reset_check_outstanding_io_done(struct spdk_bdev *bdev, void *_ctx, int status)
6500 : {
6501 13 : struct spdk_bdev_channel *ch = _ctx;
6502 : struct spdk_bdev_io *bdev_io;
6503 :
6504 13 : bdev_io = TAILQ_FIRST(&ch->queued_resets);
6505 :
6506 13 : if (status == -EBUSY) {
6507 9 : if (spdk_get_ticks() < bdev_io->u.reset.wait_poller.stop_time_tsc) {
6508 8 : bdev_io->u.reset.wait_poller.poller = SPDK_POLLER_REGISTER(bdev_reset_poll_for_outstanding_io,
6509 : ch, BDEV_RESET_CHECK_OUTSTANDING_IO_PERIOD);
6510 8 : } else {
6511 1 : TAILQ_REMOVE(&ch->queued_resets, bdev_io, internal.link);
6512 :
6513 1 : if (TAILQ_EMPTY(&ch->io_memory_domain) && TAILQ_EMPTY(&ch->io_accel_exec)) {
6514 : /* If outstanding IOs are still present and reset_io_drain_timeout
6515 : * seconds passed, start the reset. */
6516 1 : bdev_io_submit_reset(bdev_io);
6517 1 : } else {
6518 : /* We still have in progress memory domain pull/push or we're
6519 : * executing accel sequence. Since we cannot abort either of those
6520 : * operations, fail the reset request. */
6521 0 : spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
6522 : }
6523 : }
6524 9 : } else {
6525 4 : TAILQ_REMOVE(&ch->queued_resets, bdev_io, internal.link);
6526 4 : SPDK_DEBUGLOG(bdev,
6527 : "Skipping reset for underlying device of bdev: %s - no outstanding I/O.\n",
6528 : ch->bdev->name);
6529 : /* Mark the completion status as a SUCCESS and complete the reset. */
6530 4 : spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_SUCCESS);
6531 : }
6532 13 : }
6533 :
6534 : static void
6535 13 : bdev_reset_check_outstanding_io(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
6536 : struct spdk_io_channel *io_ch, void *_ctx)
6537 : {
6538 13 : struct spdk_bdev_channel *cur_ch = __io_ch_to_bdev_ch(io_ch);
6539 13 : int status = 0;
6540 :
6541 17 : if (cur_ch->io_outstanding > 0 ||
6542 4 : !TAILQ_EMPTY(&cur_ch->io_memory_domain) ||
6543 4 : !TAILQ_EMPTY(&cur_ch->io_accel_exec)) {
6544 : /* If a channel has outstanding IO, set status to -EBUSY code. This will stop
6545 : * further iteration over the rest of the channels and pass non-zero status
6546 : * to the callback function. */
6547 9 : status = -EBUSY;
6548 9 : }
6549 13 : spdk_bdev_for_each_channel_continue(i, status);
6550 13 : }
6551 :
6552 : static int
6553 8 : bdev_reset_poll_for_outstanding_io(void *ctx)
6554 : {
6555 8 : struct spdk_bdev_channel *ch = ctx;
6556 : struct spdk_bdev_io *bdev_io;
6557 :
6558 8 : bdev_io = TAILQ_FIRST(&ch->queued_resets);
6559 :
6560 8 : spdk_poller_unregister(&bdev_io->u.reset.wait_poller.poller);
6561 8 : spdk_bdev_for_each_channel(ch->bdev, bdev_reset_check_outstanding_io, ch,
6562 : bdev_reset_check_outstanding_io_done);
6563 :
6564 8 : return SPDK_POLLER_BUSY;
6565 : }
6566 :
6567 : static void
6568 15 : bdev_reset_freeze_channel_done(struct spdk_bdev *bdev, void *_ctx, int status)
6569 : {
6570 15 : struct spdk_bdev_channel *ch = _ctx;
6571 : struct spdk_bdev_io *bdev_io;
6572 :
6573 15 : bdev_io = TAILQ_FIRST(&ch->queued_resets);
6574 :
6575 15 : if (bdev->reset_io_drain_timeout == 0) {
6576 10 : TAILQ_REMOVE(&ch->queued_resets, bdev_io, internal.link);
6577 :
6578 10 : bdev_io_submit_reset(bdev_io);
6579 10 : return;
6580 : }
6581 :
6582 10 : bdev_io->u.reset.wait_poller.stop_time_tsc = spdk_get_ticks() +
6583 5 : (ch->bdev->reset_io_drain_timeout * spdk_get_ticks_hz());
6584 :
6585 : /* In case bdev->reset_io_drain_timeout is not equal to zero,
6586 : * submit the reset to the underlying module only if outstanding I/O
6587 : * remain after reset_io_drain_timeout seconds have passed. */
6588 5 : spdk_bdev_for_each_channel(ch->bdev, bdev_reset_check_outstanding_io, ch,
6589 : bdev_reset_check_outstanding_io_done);
6590 15 : }
6591 :
6592 : static void
6593 18 : bdev_reset_freeze_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
6594 : struct spdk_io_channel *ch, void *_ctx)
6595 : {
6596 : struct spdk_bdev_channel *channel;
6597 : struct spdk_bdev_mgmt_channel *mgmt_channel;
6598 : struct spdk_bdev_shared_resource *shared_resource;
6599 : bdev_io_tailq_t tmp_queued;
6600 :
6601 18 : TAILQ_INIT(&tmp_queued);
6602 :
6603 18 : channel = __io_ch_to_bdev_ch(ch);
6604 18 : shared_resource = channel->shared_resource;
6605 18 : mgmt_channel = shared_resource->mgmt_ch;
6606 :
6607 18 : channel->flags |= BDEV_CH_RESET_IN_PROGRESS;
6608 :
6609 18 : if ((channel->flags & BDEV_CH_QOS_ENABLED) != 0) {
6610 2 : TAILQ_SWAP(&channel->qos_queued_io, &tmp_queued, spdk_bdev_io, internal.link);
6611 2 : }
6612 :
6613 18 : bdev_abort_all_queued_io(&shared_resource->nomem_io, channel);
6614 18 : bdev_abort_all_buf_io(mgmt_channel, channel);
6615 18 : bdev_abort_all_queued_io(&tmp_queued, channel);
6616 :
6617 18 : spdk_bdev_for_each_channel_continue(i, 0);
6618 18 : }
6619 :
6620 : static void
6621 15 : bdev_start_reset(void *ctx)
6622 : {
6623 15 : struct spdk_bdev_channel *ch = ctx;
6624 :
6625 15 : spdk_bdev_for_each_channel(ch->bdev, bdev_reset_freeze_channel, ch,
6626 : bdev_reset_freeze_channel_done);
6627 15 : }
6628 :
6629 : static void
6630 16 : bdev_channel_start_reset(struct spdk_bdev_channel *ch)
6631 : {
6632 16 : struct spdk_bdev *bdev = ch->bdev;
6633 :
6634 16 : assert(!TAILQ_EMPTY(&ch->queued_resets));
6635 :
6636 16 : spdk_spin_lock(&bdev->internal.spinlock);
6637 16 : if (bdev->internal.reset_in_progress == NULL) {
6638 15 : bdev->internal.reset_in_progress = TAILQ_FIRST(&ch->queued_resets);
6639 : /*
6640 : * Take a channel reference for the target bdev for the life of this
6641 : * reset. This guards against the channel getting destroyed while
6642 : * spdk_bdev_for_each_channel() calls related to this reset IO are in
6643 : * progress. We will release the reference when this reset is
6644 : * completed.
6645 : */
6646 15 : bdev->internal.reset_in_progress->u.reset.ch_ref = spdk_get_io_channel(__bdev_to_io_dev(bdev));
6647 15 : bdev_start_reset(ch);
6648 15 : }
6649 16 : spdk_spin_unlock(&bdev->internal.spinlock);
6650 16 : }
6651 :
6652 : int
6653 16 : spdk_bdev_reset(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6654 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6655 : {
6656 16 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6657 : struct spdk_bdev_io *bdev_io;
6658 16 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6659 :
6660 16 : bdev_io = bdev_channel_get_io(channel);
6661 16 : if (!bdev_io) {
6662 0 : return -ENOMEM;
6663 : }
6664 :
6665 16 : bdev_io->internal.ch = channel;
6666 16 : bdev_io->internal.desc = desc;
6667 16 : bdev_io->internal.submit_tsc = spdk_get_ticks();
6668 16 : bdev_io->type = SPDK_BDEV_IO_TYPE_RESET;
6669 16 : bdev_io->u.reset.ch_ref = NULL;
6670 16 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6671 :
6672 16 : spdk_spin_lock(&bdev->internal.spinlock);
6673 16 : TAILQ_INSERT_TAIL(&channel->queued_resets, bdev_io, internal.link);
6674 16 : spdk_spin_unlock(&bdev->internal.spinlock);
6675 :
6676 16 : bdev_ch_add_to_io_submitted(bdev_io);
6677 :
6678 16 : bdev_channel_start_reset(channel);
6679 :
6680 16 : return 0;
6681 16 : }
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(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 610 : bdev_io_update_io_stat(struct spdk_bdev_io *bdev_io, uint64_t tsc_diff)
7244 : {
7245 610 : enum spdk_bdev_io_status io_status = bdev_io->internal.status;
7246 610 : struct spdk_bdev_io_stat *io_stat = bdev_io->internal.ch->stat;
7247 610 : uint64_t num_blocks = bdev_io->u.bdev.num_blocks;
7248 610 : uint32_t blocklen = bdev_io->bdev->blocklen;
7249 :
7250 610 : if (spdk_likely(io_status == SPDK_BDEV_IO_STATUS_SUCCESS)) {
7251 516 : 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 75 : break;
7324 : }
7325 610 : } else if (io_status <= SPDK_BDEV_IO_STATUS_FAILED && io_status >= SPDK_MIN_BDEV_IO_STATUS) {
7326 94 : io_stat = bdev_io->bdev->internal.stat;
7327 94 : assert(io_stat->io_error != NULL);
7328 :
7329 94 : spdk_spin_lock(&bdev_io->bdev->internal.spinlock);
7330 94 : io_stat->io_error->error_status[-io_status - 1]++;
7331 94 : spdk_spin_unlock(&bdev_io->bdev->internal.spinlock);
7332 94 : }
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 610 : }
7355 :
7356 : static inline void
7357 610 : _bdev_io_complete(void *ctx)
7358 : {
7359 610 : struct spdk_bdev_io *bdev_io = ctx;
7360 :
7361 610 : 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 610 : assert(bdev_io->internal.cb != NULL);
7367 610 : assert(spdk_get_thread() == spdk_bdev_io_get_thread(bdev_io));
7368 :
7369 1220 : bdev_io->internal.cb(bdev_io, bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS,
7370 610 : bdev_io->internal.caller_ctx);
7371 610 : }
7372 :
7373 : static inline void
7374 618 : bdev_io_complete(void *ctx)
7375 : {
7376 618 : struct spdk_bdev_io *bdev_io = ctx;
7377 618 : struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
7378 : uint64_t tsc, tsc_diff;
7379 :
7380 618 : 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 610 : tsc = spdk_get_ticks();
7391 610 : tsc_diff = tsc - bdev_io->internal.submit_tsc;
7392 :
7393 610 : bdev_ch_remove_from_io_submitted(bdev_io);
7394 610 : 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 610 : 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 610 : bdev_io_update_io_stat(bdev_io, tsc_diff);
7408 610 : _bdev_io_complete(bdev_io);
7409 618 : }
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 void
7430 15 : bdev_reset_complete(struct spdk_bdev *bdev, void *_ctx, int status)
7431 : {
7432 15 : struct spdk_bdev_io *bdev_io = _ctx;
7433 :
7434 15 : if (bdev_io->u.reset.ch_ref != NULL) {
7435 15 : spdk_put_io_channel(bdev_io->u.reset.ch_ref);
7436 15 : bdev_io->u.reset.ch_ref = NULL;
7437 15 : }
7438 :
7439 15 : bdev_io_complete(bdev_io);
7440 :
7441 15 : if (bdev->internal.status == SPDK_BDEV_STATUS_REMOVING &&
7442 1 : TAILQ_EMPTY(&bdev->internal.open_descs)) {
7443 1 : spdk_io_device_unregister(__bdev_to_io_dev(bdev), bdev_destroy_cb);
7444 1 : }
7445 15 : }
7446 :
7447 : static void
7448 18 : bdev_unfreeze_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
7449 : struct spdk_io_channel *_ch, void *_ctx)
7450 : {
7451 18 : struct spdk_bdev_io *bdev_io = _ctx;
7452 18 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
7453 : struct spdk_bdev_io *queued_reset;
7454 :
7455 18 : ch->flags &= ~BDEV_CH_RESET_IN_PROGRESS;
7456 18 : while (!TAILQ_EMPTY(&ch->queued_resets)) {
7457 0 : queued_reset = TAILQ_FIRST(&ch->queued_resets);
7458 0 : TAILQ_REMOVE(&ch->queued_resets, queued_reset, internal.link);
7459 0 : spdk_bdev_io_complete(queued_reset, bdev_io->internal.status);
7460 : }
7461 :
7462 18 : spdk_bdev_for_each_channel_continue(i, 0);
7463 18 : }
7464 :
7465 : static void
7466 0 : bdev_io_complete_sequence_cb(void *ctx, int status)
7467 : {
7468 0 : struct spdk_bdev_io *bdev_io = ctx;
7469 :
7470 : /* u.bdev.accel_sequence should have already been cleared at this point */
7471 0 : assert(bdev_io->u.bdev.accel_sequence == NULL);
7472 0 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS);
7473 0 : bdev_io->internal.f.has_accel_sequence = false;
7474 :
7475 0 : if (spdk_unlikely(status != 0)) {
7476 0 : SPDK_ERRLOG("Failed to execute accel sequence, status=%d\n", status);
7477 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
7478 0 : }
7479 :
7480 0 : bdev_io_complete(bdev_io);
7481 0 : }
7482 :
7483 : void
7484 598 : spdk_bdev_io_complete(struct spdk_bdev_io *bdev_io, enum spdk_bdev_io_status status)
7485 : {
7486 598 : struct spdk_bdev *bdev = bdev_io->bdev;
7487 598 : struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
7488 598 : struct spdk_bdev_shared_resource *shared_resource = bdev_ch->shared_resource;
7489 :
7490 598 : if (spdk_unlikely(bdev_io->internal.status != SPDK_BDEV_IO_STATUS_PENDING)) {
7491 0 : SPDK_ERRLOG("Unexpected completion on IO from %s module, status was %s\n",
7492 : spdk_bdev_get_module_name(bdev),
7493 : bdev_io_status_get_string(bdev_io->internal.status));
7494 0 : assert(false);
7495 : }
7496 598 : bdev_io->internal.status = status;
7497 :
7498 598 : if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_RESET)) {
7499 16 : bool unlock_channels = false;
7500 :
7501 16 : if (status == SPDK_BDEV_IO_STATUS_NOMEM) {
7502 0 : SPDK_ERRLOG("NOMEM returned for reset\n");
7503 0 : }
7504 16 : spdk_spin_lock(&bdev->internal.spinlock);
7505 16 : if (bdev_io == bdev->internal.reset_in_progress) {
7506 15 : bdev->internal.reset_in_progress = NULL;
7507 15 : unlock_channels = true;
7508 15 : }
7509 16 : spdk_spin_unlock(&bdev->internal.spinlock);
7510 :
7511 16 : if (unlock_channels) {
7512 15 : spdk_bdev_for_each_channel(bdev, bdev_unfreeze_channel, bdev_io,
7513 : bdev_reset_complete);
7514 15 : return;
7515 : }
7516 1 : } else {
7517 582 : bdev_io_decrement_outstanding(bdev_ch, shared_resource);
7518 582 : if (spdk_likely(status == SPDK_BDEV_IO_STATUS_SUCCESS)) {
7519 485 : if (bdev_io_needs_sequence_exec(bdev_io->internal.desc, bdev_io)) {
7520 0 : bdev_io_exec_sequence(bdev_io, bdev_io_complete_sequence_cb);
7521 0 : return;
7522 485 : } else if (spdk_unlikely(bdev_io->internal.f.has_bounce_buf &&
7523 : !bdev_io_use_accel_sequence(bdev_io))) {
7524 26 : _bdev_io_push_bounce_data_buffer(bdev_io,
7525 : _bdev_io_complete_push_bounce_done);
7526 : /* bdev IO will be completed in the callback */
7527 26 : return;
7528 : }
7529 459 : }
7530 :
7531 556 : if (spdk_unlikely(_bdev_io_handle_no_mem(bdev_io, BDEV_IO_RETRY_STATE_SUBMIT))) {
7532 5 : return;
7533 : }
7534 : }
7535 :
7536 552 : bdev_io_complete(bdev_io);
7537 598 : }
7538 :
7539 : void
7540 0 : spdk_bdev_io_complete_scsi_status(struct spdk_bdev_io *bdev_io, enum spdk_scsi_status sc,
7541 : enum spdk_scsi_sense sk, uint8_t asc, uint8_t ascq)
7542 : {
7543 : enum spdk_bdev_io_status status;
7544 :
7545 0 : if (sc == SPDK_SCSI_STATUS_GOOD) {
7546 0 : status = SPDK_BDEV_IO_STATUS_SUCCESS;
7547 0 : } else {
7548 0 : status = SPDK_BDEV_IO_STATUS_SCSI_ERROR;
7549 0 : bdev_io->internal.error.scsi.sc = sc;
7550 0 : bdev_io->internal.error.scsi.sk = sk;
7551 0 : bdev_io->internal.error.scsi.asc = asc;
7552 0 : bdev_io->internal.error.scsi.ascq = ascq;
7553 : }
7554 :
7555 0 : spdk_bdev_io_complete(bdev_io, status);
7556 0 : }
7557 :
7558 : void
7559 0 : spdk_bdev_io_get_scsi_status(const struct spdk_bdev_io *bdev_io,
7560 : int *sc, int *sk, int *asc, int *ascq)
7561 : {
7562 0 : assert(sc != NULL);
7563 0 : assert(sk != NULL);
7564 0 : assert(asc != NULL);
7565 0 : assert(ascq != NULL);
7566 :
7567 0 : switch (bdev_io->internal.status) {
7568 : case SPDK_BDEV_IO_STATUS_SUCCESS:
7569 0 : *sc = SPDK_SCSI_STATUS_GOOD;
7570 0 : *sk = SPDK_SCSI_SENSE_NO_SENSE;
7571 0 : *asc = SPDK_SCSI_ASC_NO_ADDITIONAL_SENSE;
7572 0 : *ascq = SPDK_SCSI_ASCQ_CAUSE_NOT_REPORTABLE;
7573 0 : break;
7574 : case SPDK_BDEV_IO_STATUS_NVME_ERROR:
7575 0 : spdk_scsi_nvme_translate(bdev_io, sc, sk, asc, ascq);
7576 0 : break;
7577 : case SPDK_BDEV_IO_STATUS_MISCOMPARE:
7578 0 : *sc = SPDK_SCSI_STATUS_CHECK_CONDITION;
7579 0 : *sk = SPDK_SCSI_SENSE_MISCOMPARE;
7580 0 : *asc = SPDK_SCSI_ASC_MISCOMPARE_DURING_VERIFY_OPERATION;
7581 0 : *ascq = bdev_io->internal.error.scsi.ascq;
7582 0 : break;
7583 : case SPDK_BDEV_IO_STATUS_SCSI_ERROR:
7584 0 : *sc = bdev_io->internal.error.scsi.sc;
7585 0 : *sk = bdev_io->internal.error.scsi.sk;
7586 0 : *asc = bdev_io->internal.error.scsi.asc;
7587 0 : *ascq = bdev_io->internal.error.scsi.ascq;
7588 0 : break;
7589 : default:
7590 0 : *sc = SPDK_SCSI_STATUS_CHECK_CONDITION;
7591 0 : *sk = SPDK_SCSI_SENSE_ABORTED_COMMAND;
7592 0 : *asc = SPDK_SCSI_ASC_NO_ADDITIONAL_SENSE;
7593 0 : *ascq = SPDK_SCSI_ASCQ_CAUSE_NOT_REPORTABLE;
7594 0 : break;
7595 : }
7596 0 : }
7597 :
7598 : void
7599 0 : spdk_bdev_io_complete_aio_status(struct spdk_bdev_io *bdev_io, int aio_result)
7600 : {
7601 : enum spdk_bdev_io_status status;
7602 :
7603 0 : if (aio_result == 0) {
7604 0 : status = SPDK_BDEV_IO_STATUS_SUCCESS;
7605 0 : } else {
7606 0 : status = SPDK_BDEV_IO_STATUS_AIO_ERROR;
7607 : }
7608 :
7609 0 : bdev_io->internal.error.aio_result = aio_result;
7610 :
7611 0 : spdk_bdev_io_complete(bdev_io, status);
7612 0 : }
7613 :
7614 : void
7615 0 : spdk_bdev_io_get_aio_status(const struct spdk_bdev_io *bdev_io, int *aio_result)
7616 : {
7617 0 : assert(aio_result != NULL);
7618 :
7619 0 : if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_AIO_ERROR) {
7620 0 : *aio_result = bdev_io->internal.error.aio_result;
7621 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS) {
7622 0 : *aio_result = 0;
7623 0 : } else {
7624 0 : *aio_result = -EIO;
7625 : }
7626 0 : }
7627 :
7628 : void
7629 0 : spdk_bdev_io_complete_nvme_status(struct spdk_bdev_io *bdev_io, uint32_t cdw0, int sct, int sc)
7630 : {
7631 : enum spdk_bdev_io_status status;
7632 :
7633 0 : if (spdk_likely(sct == SPDK_NVME_SCT_GENERIC && sc == SPDK_NVME_SC_SUCCESS)) {
7634 0 : status = SPDK_BDEV_IO_STATUS_SUCCESS;
7635 0 : } else if (sct == SPDK_NVME_SCT_GENERIC && sc == SPDK_NVME_SC_ABORTED_BY_REQUEST) {
7636 0 : status = SPDK_BDEV_IO_STATUS_ABORTED;
7637 0 : } else {
7638 0 : status = SPDK_BDEV_IO_STATUS_NVME_ERROR;
7639 : }
7640 :
7641 0 : bdev_io->internal.error.nvme.cdw0 = cdw0;
7642 0 : bdev_io->internal.error.nvme.sct = sct;
7643 0 : bdev_io->internal.error.nvme.sc = sc;
7644 :
7645 0 : spdk_bdev_io_complete(bdev_io, status);
7646 0 : }
7647 :
7648 : void
7649 0 : spdk_bdev_io_get_nvme_status(const struct spdk_bdev_io *bdev_io, uint32_t *cdw0, int *sct, int *sc)
7650 : {
7651 0 : assert(sct != NULL);
7652 0 : assert(sc != NULL);
7653 0 : assert(cdw0 != NULL);
7654 :
7655 0 : if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_ABORT)) {
7656 0 : *sct = SPDK_NVME_SCT_GENERIC;
7657 0 : *sc = SPDK_NVME_SC_SUCCESS;
7658 0 : if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS) {
7659 0 : *cdw0 = 0;
7660 0 : } else {
7661 0 : *cdw0 = 1U;
7662 : }
7663 0 : return;
7664 : }
7665 :
7666 0 : if (spdk_likely(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS)) {
7667 0 : *sct = SPDK_NVME_SCT_GENERIC;
7668 0 : *sc = SPDK_NVME_SC_SUCCESS;
7669 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_NVME_ERROR) {
7670 0 : *sct = bdev_io->internal.error.nvme.sct;
7671 0 : *sc = bdev_io->internal.error.nvme.sc;
7672 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_ABORTED) {
7673 0 : *sct = SPDK_NVME_SCT_GENERIC;
7674 0 : *sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
7675 0 : } else {
7676 0 : *sct = SPDK_NVME_SCT_GENERIC;
7677 0 : *sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
7678 : }
7679 :
7680 0 : *cdw0 = bdev_io->internal.error.nvme.cdw0;
7681 0 : }
7682 :
7683 : void
7684 0 : spdk_bdev_io_get_nvme_fused_status(const struct spdk_bdev_io *bdev_io, uint32_t *cdw0,
7685 : int *first_sct, int *first_sc, int *second_sct, int *second_sc)
7686 : {
7687 0 : assert(first_sct != NULL);
7688 0 : assert(first_sc != NULL);
7689 0 : assert(second_sct != NULL);
7690 0 : assert(second_sc != NULL);
7691 0 : assert(cdw0 != NULL);
7692 :
7693 0 : if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_NVME_ERROR) {
7694 0 : if (bdev_io->internal.error.nvme.sct == SPDK_NVME_SCT_MEDIA_ERROR &&
7695 0 : bdev_io->internal.error.nvme.sc == SPDK_NVME_SC_COMPARE_FAILURE) {
7696 0 : *first_sct = bdev_io->internal.error.nvme.sct;
7697 0 : *first_sc = bdev_io->internal.error.nvme.sc;
7698 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
7699 0 : *second_sc = SPDK_NVME_SC_ABORTED_FAILED_FUSED;
7700 0 : } else {
7701 0 : *first_sct = SPDK_NVME_SCT_GENERIC;
7702 0 : *first_sc = SPDK_NVME_SC_SUCCESS;
7703 0 : *second_sct = bdev_io->internal.error.nvme.sct;
7704 0 : *second_sc = bdev_io->internal.error.nvme.sc;
7705 : }
7706 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_ABORTED) {
7707 0 : *first_sct = SPDK_NVME_SCT_GENERIC;
7708 0 : *first_sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
7709 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
7710 0 : *second_sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
7711 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS) {
7712 0 : *first_sct = SPDK_NVME_SCT_GENERIC;
7713 0 : *first_sc = SPDK_NVME_SC_SUCCESS;
7714 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
7715 0 : *second_sc = SPDK_NVME_SC_SUCCESS;
7716 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_FIRST_FUSED_FAILED) {
7717 0 : *first_sct = SPDK_NVME_SCT_GENERIC;
7718 0 : *first_sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
7719 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
7720 0 : *second_sc = SPDK_NVME_SC_ABORTED_FAILED_FUSED;
7721 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_MISCOMPARE) {
7722 0 : *first_sct = SPDK_NVME_SCT_MEDIA_ERROR;
7723 0 : *first_sc = SPDK_NVME_SC_COMPARE_FAILURE;
7724 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
7725 0 : *second_sc = SPDK_NVME_SC_ABORTED_FAILED_FUSED;
7726 0 : } else {
7727 0 : *first_sct = SPDK_NVME_SCT_GENERIC;
7728 0 : *first_sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
7729 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
7730 0 : *second_sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
7731 : }
7732 :
7733 0 : *cdw0 = bdev_io->internal.error.nvme.cdw0;
7734 0 : }
7735 :
7736 : void
7737 0 : spdk_bdev_io_complete_base_io_status(struct spdk_bdev_io *bdev_io,
7738 : const struct spdk_bdev_io *base_io)
7739 : {
7740 0 : switch (base_io->internal.status) {
7741 : case SPDK_BDEV_IO_STATUS_NVME_ERROR:
7742 0 : spdk_bdev_io_complete_nvme_status(bdev_io,
7743 0 : base_io->internal.error.nvme.cdw0,
7744 0 : base_io->internal.error.nvme.sct,
7745 0 : base_io->internal.error.nvme.sc);
7746 0 : break;
7747 : case SPDK_BDEV_IO_STATUS_SCSI_ERROR:
7748 0 : spdk_bdev_io_complete_scsi_status(bdev_io,
7749 0 : base_io->internal.error.scsi.sc,
7750 0 : base_io->internal.error.scsi.sk,
7751 0 : base_io->internal.error.scsi.asc,
7752 0 : base_io->internal.error.scsi.ascq);
7753 0 : break;
7754 : case SPDK_BDEV_IO_STATUS_AIO_ERROR:
7755 0 : spdk_bdev_io_complete_aio_status(bdev_io, base_io->internal.error.aio_result);
7756 0 : break;
7757 : default:
7758 0 : spdk_bdev_io_complete(bdev_io, base_io->internal.status);
7759 0 : break;
7760 : }
7761 0 : }
7762 :
7763 : struct spdk_thread *
7764 660 : spdk_bdev_io_get_thread(struct spdk_bdev_io *bdev_io)
7765 : {
7766 660 : return spdk_io_channel_get_thread(bdev_io->internal.ch->channel);
7767 : }
7768 :
7769 : struct spdk_io_channel *
7770 70 : spdk_bdev_io_get_io_channel(struct spdk_bdev_io *bdev_io)
7771 : {
7772 70 : return bdev_io->internal.ch->channel;
7773 : }
7774 :
7775 : static int
7776 127 : bdev_register(struct spdk_bdev *bdev)
7777 : {
7778 : char *bdev_name;
7779 : char uuid[SPDK_UUID_STRING_LEN];
7780 : struct spdk_iobuf_opts iobuf_opts;
7781 : int ret;
7782 :
7783 127 : assert(bdev->module != NULL);
7784 :
7785 127 : if (!bdev->name) {
7786 0 : SPDK_ERRLOG("Bdev name is NULL\n");
7787 0 : return -EINVAL;
7788 : }
7789 :
7790 127 : if (!strlen(bdev->name)) {
7791 0 : SPDK_ERRLOG("Bdev name must not be an empty string\n");
7792 0 : return -EINVAL;
7793 : }
7794 :
7795 : /* Users often register their own I/O devices using the bdev name. In
7796 : * order to avoid conflicts, prepend bdev_. */
7797 127 : bdev_name = spdk_sprintf_alloc("bdev_%s", bdev->name);
7798 127 : if (!bdev_name) {
7799 0 : SPDK_ERRLOG("Unable to allocate memory for internal bdev name.\n");
7800 0 : return -ENOMEM;
7801 : }
7802 :
7803 127 : bdev->internal.stat = bdev_alloc_io_stat(true);
7804 127 : if (!bdev->internal.stat) {
7805 0 : SPDK_ERRLOG("Unable to allocate I/O statistics structure.\n");
7806 0 : free(bdev_name);
7807 0 : return -ENOMEM;
7808 : }
7809 :
7810 127 : bdev->internal.status = SPDK_BDEV_STATUS_READY;
7811 127 : bdev->internal.measured_queue_depth = UINT64_MAX;
7812 127 : bdev->internal.claim_type = SPDK_BDEV_CLAIM_NONE;
7813 127 : memset(&bdev->internal.claim, 0, sizeof(bdev->internal.claim));
7814 127 : bdev->internal.qd_poller = NULL;
7815 127 : bdev->internal.qos = NULL;
7816 :
7817 127 : TAILQ_INIT(&bdev->internal.open_descs);
7818 127 : TAILQ_INIT(&bdev->internal.locked_ranges);
7819 127 : TAILQ_INIT(&bdev->internal.pending_locked_ranges);
7820 127 : TAILQ_INIT(&bdev->aliases);
7821 :
7822 : /* UUID may be specified by the user or defined by bdev itself.
7823 : * Otherwise it will be generated here, so this field will never be empty. */
7824 127 : if (spdk_uuid_is_null(&bdev->uuid)) {
7825 42 : spdk_uuid_generate(&bdev->uuid);
7826 42 : }
7827 :
7828 : /* Add the UUID alias only if it's different than the name */
7829 127 : spdk_uuid_fmt_lower(uuid, sizeof(uuid), &bdev->uuid);
7830 127 : if (strcmp(bdev->name, uuid) != 0) {
7831 126 : ret = spdk_bdev_alias_add(bdev, uuid);
7832 126 : if (ret != 0) {
7833 2 : SPDK_ERRLOG("Unable to add uuid:%s alias for bdev %s\n", uuid, bdev->name);
7834 2 : bdev_free_io_stat(bdev->internal.stat);
7835 2 : free(bdev_name);
7836 2 : return ret;
7837 : }
7838 124 : }
7839 :
7840 125 : spdk_iobuf_get_opts(&iobuf_opts, sizeof(iobuf_opts));
7841 125 : if (spdk_bdev_get_buf_align(bdev) > 1) {
7842 0 : bdev->max_rw_size = spdk_min(bdev->max_rw_size ? bdev->max_rw_size : UINT32_MAX,
7843 : iobuf_opts.large_bufsize / bdev->blocklen);
7844 0 : }
7845 :
7846 : /* If the user didn't specify a write unit size, set it to one. */
7847 125 : if (bdev->write_unit_size == 0) {
7848 121 : bdev->write_unit_size = 1;
7849 121 : }
7850 :
7851 : /* Set ACWU value to the write unit size if bdev module did not set it (does not support it natively) */
7852 125 : if (bdev->acwu == 0) {
7853 121 : bdev->acwu = bdev->write_unit_size;
7854 121 : }
7855 :
7856 125 : if (bdev->phys_blocklen == 0) {
7857 121 : bdev->phys_blocklen = spdk_bdev_get_data_block_size(bdev);
7858 121 : }
7859 :
7860 125 : if (!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COPY)) {
7861 0 : bdev->max_copy = bdev_get_max_write(bdev, iobuf_opts.large_bufsize);
7862 0 : }
7863 :
7864 125 : if (!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE_ZEROES)) {
7865 0 : bdev->max_write_zeroes = bdev_get_max_write(bdev, ZERO_BUFFER_SIZE);
7866 0 : }
7867 :
7868 125 : bdev->internal.reset_in_progress = NULL;
7869 125 : bdev->internal.qd_poll_in_progress = false;
7870 125 : bdev->internal.period = 0;
7871 125 : bdev->internal.new_period = 0;
7872 125 : bdev->internal.trace_id = spdk_trace_register_owner(OWNER_TYPE_BDEV, bdev_name);
7873 :
7874 : /*
7875 : * Initialize spinlock before registering IO device because spinlock is used in
7876 : * bdev_channel_create
7877 : */
7878 125 : spdk_spin_init(&bdev->internal.spinlock);
7879 :
7880 250 : spdk_io_device_register(__bdev_to_io_dev(bdev),
7881 : bdev_channel_create, bdev_channel_destroy,
7882 : sizeof(struct spdk_bdev_channel),
7883 125 : bdev_name);
7884 :
7885 : /*
7886 : * Register bdev name only after the bdev object is ready.
7887 : * After bdev_name_add returns, it is possible for other threads to start using the bdev,
7888 : * create IO channels...
7889 : */
7890 125 : ret = bdev_name_add(&bdev->internal.bdev_name, bdev, bdev->name);
7891 125 : if (ret != 0) {
7892 0 : spdk_io_device_unregister(__bdev_to_io_dev(bdev), NULL);
7893 0 : bdev_free_io_stat(bdev->internal.stat);
7894 0 : spdk_spin_destroy(&bdev->internal.spinlock);
7895 0 : free(bdev_name);
7896 0 : return ret;
7897 : }
7898 :
7899 125 : free(bdev_name);
7900 :
7901 125 : SPDK_DEBUGLOG(bdev, "Inserting bdev %s into list\n", bdev->name);
7902 125 : TAILQ_INSERT_TAIL(&g_bdev_mgr.bdevs, bdev, internal.link);
7903 :
7904 125 : return 0;
7905 127 : }
7906 :
7907 : static void
7908 126 : bdev_destroy_cb(void *io_device)
7909 : {
7910 : int rc;
7911 : struct spdk_bdev *bdev;
7912 : spdk_bdev_unregister_cb cb_fn;
7913 : void *cb_arg;
7914 :
7915 126 : bdev = __bdev_from_io_dev(io_device);
7916 :
7917 126 : if (bdev->internal.unregister_td != spdk_get_thread()) {
7918 1 : spdk_thread_send_msg(bdev->internal.unregister_td, bdev_destroy_cb, io_device);
7919 1 : return;
7920 : }
7921 :
7922 125 : cb_fn = bdev->internal.unregister_cb;
7923 125 : cb_arg = bdev->internal.unregister_ctx;
7924 :
7925 125 : spdk_spin_destroy(&bdev->internal.spinlock);
7926 125 : free(bdev->internal.qos);
7927 125 : bdev_free_io_stat(bdev->internal.stat);
7928 125 : spdk_trace_unregister_owner(bdev->internal.trace_id);
7929 :
7930 125 : rc = bdev->fn_table->destruct(bdev->ctxt);
7931 125 : if (rc < 0) {
7932 0 : SPDK_ERRLOG("destruct failed\n");
7933 0 : }
7934 125 : if (rc <= 0 && cb_fn != NULL) {
7935 10 : cb_fn(cb_arg, rc);
7936 10 : }
7937 126 : }
7938 :
7939 : void
7940 2 : spdk_bdev_destruct_done(struct spdk_bdev *bdev, int bdeverrno)
7941 : {
7942 2 : if (bdev->internal.unregister_cb != NULL) {
7943 0 : bdev->internal.unregister_cb(bdev->internal.unregister_ctx, bdeverrno);
7944 0 : }
7945 2 : }
7946 :
7947 : static void
7948 19 : _remove_notify(void *arg)
7949 : {
7950 19 : struct spdk_bdev_desc *desc = arg;
7951 :
7952 19 : _event_notify(desc, SPDK_BDEV_EVENT_REMOVE);
7953 19 : }
7954 :
7955 : /* returns: 0 - bdev removed and ready to be destructed.
7956 : * -EBUSY - bdev can't be destructed yet. */
7957 : static int
7958 140 : bdev_unregister_unsafe(struct spdk_bdev *bdev)
7959 : {
7960 : struct spdk_bdev_desc *desc, *tmp;
7961 140 : int rc = 0;
7962 : char uuid[SPDK_UUID_STRING_LEN];
7963 :
7964 140 : assert(spdk_spin_held(&g_bdev_mgr.spinlock));
7965 140 : assert(spdk_spin_held(&bdev->internal.spinlock));
7966 :
7967 : /* Notify each descriptor about hotremoval */
7968 159 : TAILQ_FOREACH_SAFE(desc, &bdev->internal.open_descs, link, tmp) {
7969 19 : rc = -EBUSY;
7970 : /*
7971 : * Defer invocation of the event_cb to a separate message that will
7972 : * run later on its thread. This ensures this context unwinds and
7973 : * we don't recursively unregister this bdev again if the event_cb
7974 : * immediately closes its descriptor.
7975 : */
7976 19 : event_notify(desc, _remove_notify);
7977 19 : }
7978 :
7979 : /* If there are no descriptors, proceed removing the bdev */
7980 140 : if (rc == 0) {
7981 125 : TAILQ_REMOVE(&g_bdev_mgr.bdevs, bdev, internal.link);
7982 125 : SPDK_DEBUGLOG(bdev, "Removing bdev %s from list done\n", bdev->name);
7983 :
7984 : /* Delete the name and the UUID alias */
7985 125 : spdk_uuid_fmt_lower(uuid, sizeof(uuid), &bdev->uuid);
7986 125 : bdev_name_del_unsafe(&bdev->internal.bdev_name);
7987 125 : bdev_alias_del(bdev, uuid, bdev_name_del_unsafe);
7988 :
7989 125 : spdk_notify_send("bdev_unregister", spdk_bdev_get_name(bdev));
7990 :
7991 125 : if (bdev->internal.reset_in_progress != NULL) {
7992 : /* If reset is in progress, let the completion callback for reset
7993 : * unregister the bdev.
7994 : */
7995 1 : rc = -EBUSY;
7996 1 : }
7997 125 : }
7998 :
7999 140 : return rc;
8000 : }
8001 :
8002 : static void
8003 4 : bdev_unregister_abort_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
8004 : struct spdk_io_channel *io_ch, void *_ctx)
8005 : {
8006 4 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
8007 :
8008 4 : bdev_channel_abort_queued_ios(bdev_ch);
8009 4 : spdk_bdev_for_each_channel_continue(i, 0);
8010 4 : }
8011 :
8012 : static void
8013 125 : bdev_unregister(struct spdk_bdev *bdev, void *_ctx, int status)
8014 : {
8015 : int rc;
8016 :
8017 125 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8018 125 : spdk_spin_lock(&bdev->internal.spinlock);
8019 : /*
8020 : * Set the status to REMOVING after completing to abort channels. Otherwise,
8021 : * the last spdk_bdev_close() may call spdk_io_device_unregister() while
8022 : * spdk_bdev_for_each_channel() is executed and spdk_io_device_unregister()
8023 : * may fail.
8024 : */
8025 125 : bdev->internal.status = SPDK_BDEV_STATUS_REMOVING;
8026 125 : rc = bdev_unregister_unsafe(bdev);
8027 125 : spdk_spin_unlock(&bdev->internal.spinlock);
8028 125 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8029 :
8030 125 : if (rc == 0) {
8031 109 : spdk_io_device_unregister(__bdev_to_io_dev(bdev), bdev_destroy_cb);
8032 109 : }
8033 125 : }
8034 :
8035 : void
8036 132 : spdk_bdev_unregister(struct spdk_bdev *bdev, spdk_bdev_unregister_cb cb_fn, void *cb_arg)
8037 : {
8038 : struct spdk_thread *thread;
8039 :
8040 132 : SPDK_DEBUGLOG(bdev, "Removing bdev %s from list\n", bdev->name);
8041 :
8042 132 : thread = spdk_get_thread();
8043 132 : if (!thread) {
8044 : /* The user called this from a non-SPDK thread. */
8045 0 : if (cb_fn != NULL) {
8046 0 : cb_fn(cb_arg, -ENOTSUP);
8047 0 : }
8048 0 : return;
8049 : }
8050 :
8051 132 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8052 132 : if (bdev->internal.status == SPDK_BDEV_STATUS_UNREGISTERING ||
8053 132 : bdev->internal.status == SPDK_BDEV_STATUS_REMOVING) {
8054 7 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8055 7 : if (cb_fn) {
8056 0 : cb_fn(cb_arg, -EBUSY);
8057 0 : }
8058 7 : return;
8059 : }
8060 :
8061 125 : spdk_spin_lock(&bdev->internal.spinlock);
8062 125 : bdev->internal.status = SPDK_BDEV_STATUS_UNREGISTERING;
8063 125 : bdev->internal.unregister_cb = cb_fn;
8064 125 : bdev->internal.unregister_ctx = cb_arg;
8065 125 : bdev->internal.unregister_td = thread;
8066 125 : spdk_spin_unlock(&bdev->internal.spinlock);
8067 125 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8068 :
8069 125 : spdk_bdev_set_qd_sampling_period(bdev, 0);
8070 :
8071 125 : spdk_bdev_for_each_channel(bdev, bdev_unregister_abort_channel, bdev,
8072 : bdev_unregister);
8073 132 : }
8074 :
8075 : int
8076 4 : spdk_bdev_unregister_by_name(const char *bdev_name, struct spdk_bdev_module *module,
8077 : spdk_bdev_unregister_cb cb_fn, void *cb_arg)
8078 : {
8079 : struct spdk_bdev_desc *desc;
8080 : struct spdk_bdev *bdev;
8081 : int rc;
8082 :
8083 4 : rc = spdk_bdev_open_ext(bdev_name, false, _tmp_bdev_event_cb, NULL, &desc);
8084 4 : if (rc != 0) {
8085 1 : SPDK_ERRLOG("Failed to open bdev with name: %s\n", bdev_name);
8086 1 : return rc;
8087 : }
8088 :
8089 3 : bdev = spdk_bdev_desc_get_bdev(desc);
8090 :
8091 3 : if (bdev->module != module) {
8092 1 : spdk_bdev_close(desc);
8093 1 : SPDK_ERRLOG("Bdev %s was not registered by the specified module.\n",
8094 : bdev_name);
8095 1 : return -ENODEV;
8096 : }
8097 :
8098 2 : spdk_bdev_unregister(bdev, cb_fn, cb_arg);
8099 :
8100 2 : spdk_bdev_close(desc);
8101 :
8102 2 : return 0;
8103 4 : }
8104 :
8105 : static int
8106 258 : bdev_start_qos(struct spdk_bdev *bdev)
8107 : {
8108 : struct set_qos_limit_ctx *ctx;
8109 :
8110 : /* Enable QoS */
8111 258 : if (bdev->internal.qos && bdev->internal.qos->thread == NULL) {
8112 2 : ctx = calloc(1, sizeof(*ctx));
8113 2 : if (ctx == NULL) {
8114 0 : SPDK_ERRLOG("Failed to allocate memory for QoS context\n");
8115 0 : return -ENOMEM;
8116 : }
8117 2 : ctx->bdev = bdev;
8118 2 : spdk_bdev_for_each_channel(bdev, bdev_enable_qos_msg, ctx, bdev_enable_qos_done);
8119 2 : }
8120 :
8121 258 : return 0;
8122 258 : }
8123 :
8124 : static void
8125 24 : log_already_claimed(enum spdk_log_level level, const int line, const char *func, const char *detail,
8126 : struct spdk_bdev *bdev)
8127 : {
8128 : enum spdk_bdev_claim_type type;
8129 : const char *typename, *modname;
8130 : extern struct spdk_log_flag SPDK_LOG_bdev;
8131 :
8132 24 : assert(spdk_spin_held(&bdev->internal.spinlock));
8133 :
8134 24 : if (level >= SPDK_LOG_INFO && !SPDK_LOG_bdev.enabled) {
8135 0 : return;
8136 : }
8137 :
8138 24 : type = bdev->internal.claim_type;
8139 24 : typename = spdk_bdev_claim_get_name(type);
8140 :
8141 24 : if (type == SPDK_BDEV_CLAIM_EXCL_WRITE) {
8142 6 : modname = bdev->internal.claim.v1.module->name;
8143 12 : spdk_log(level, __FILE__, line, func, "bdev %s %s: type %s by module %s\n",
8144 6 : bdev->name, detail, typename, modname);
8145 6 : return;
8146 : }
8147 :
8148 18 : if (claim_type_is_v2(type)) {
8149 : struct spdk_bdev_module_claim *claim;
8150 :
8151 36 : TAILQ_FOREACH(claim, &bdev->internal.claim.v2.claims, link) {
8152 18 : modname = claim->module->name;
8153 36 : spdk_log(level, __FILE__, line, func, "bdev %s %s: type %s by module %s\n",
8154 18 : bdev->name, detail, typename, modname);
8155 18 : }
8156 18 : return;
8157 : }
8158 :
8159 0 : assert(false);
8160 24 : }
8161 :
8162 : static int
8163 267 : bdev_open(struct spdk_bdev *bdev, bool write, struct spdk_bdev_desc *desc)
8164 : {
8165 : struct spdk_thread *thread;
8166 267 : int rc = 0;
8167 :
8168 267 : thread = spdk_get_thread();
8169 267 : if (!thread) {
8170 0 : SPDK_ERRLOG("Cannot open bdev from non-SPDK thread.\n");
8171 0 : return -ENOTSUP;
8172 : }
8173 :
8174 267 : SPDK_DEBUGLOG(bdev, "Opening descriptor %p for bdev %s on thread %p\n", desc, bdev->name,
8175 : spdk_get_thread());
8176 :
8177 267 : desc->bdev = bdev;
8178 267 : desc->thread = thread;
8179 267 : desc->write = write;
8180 :
8181 267 : spdk_spin_lock(&bdev->internal.spinlock);
8182 267 : if (bdev->internal.status == SPDK_BDEV_STATUS_UNREGISTERING ||
8183 267 : bdev->internal.status == SPDK_BDEV_STATUS_REMOVING) {
8184 3 : spdk_spin_unlock(&bdev->internal.spinlock);
8185 3 : return -ENODEV;
8186 : }
8187 :
8188 264 : if (write && bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE) {
8189 6 : LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
8190 6 : spdk_spin_unlock(&bdev->internal.spinlock);
8191 6 : return -EPERM;
8192 : }
8193 :
8194 258 : rc = bdev_start_qos(bdev);
8195 258 : if (rc != 0) {
8196 0 : SPDK_ERRLOG("Failed to start QoS on bdev %s\n", bdev->name);
8197 0 : spdk_spin_unlock(&bdev->internal.spinlock);
8198 0 : return rc;
8199 : }
8200 :
8201 258 : TAILQ_INSERT_TAIL(&bdev->internal.open_descs, desc, link);
8202 :
8203 258 : spdk_spin_unlock(&bdev->internal.spinlock);
8204 :
8205 258 : return 0;
8206 267 : }
8207 :
8208 : static void
8209 268 : bdev_open_opts_get_defaults(struct spdk_bdev_open_opts *opts, size_t opts_size)
8210 : {
8211 268 : if (!opts) {
8212 0 : SPDK_ERRLOG("opts should not be NULL.\n");
8213 0 : return;
8214 : }
8215 :
8216 268 : if (!opts_size) {
8217 0 : SPDK_ERRLOG("opts_size should not be zero.\n");
8218 0 : return;
8219 : }
8220 :
8221 268 : memset(opts, 0, opts_size);
8222 268 : opts->size = opts_size;
8223 :
8224 : #define FIELD_OK(field) \
8225 : offsetof(struct spdk_bdev_open_opts, field) + sizeof(opts->field) <= opts_size
8226 :
8227 : #define SET_FIELD(field, value) \
8228 : if (FIELD_OK(field)) { \
8229 : opts->field = value; \
8230 : } \
8231 :
8232 268 : SET_FIELD(hide_metadata, false);
8233 :
8234 : #undef FIELD_OK
8235 : #undef SET_FIELD
8236 268 : }
8237 :
8238 : static void
8239 2 : bdev_open_opts_copy(struct spdk_bdev_open_opts *opts,
8240 : const struct spdk_bdev_open_opts *opts_src, size_t opts_size)
8241 : {
8242 2 : assert(opts);
8243 2 : assert(opts_src);
8244 :
8245 : #define SET_FIELD(field) \
8246 : if (offsetof(struct spdk_bdev_open_opts, field) + sizeof(opts->field) <= opts_size) { \
8247 : opts->field = opts_src->field; \
8248 : } \
8249 :
8250 2 : SET_FIELD(hide_metadata);
8251 :
8252 2 : opts->size = opts_src->size;
8253 :
8254 : /* We should not remove this statement, but need to update the assert statement
8255 : * if we add a new field, and also add a corresponding SET_FIELD statement.
8256 : */
8257 : SPDK_STATIC_ASSERT(sizeof(struct spdk_bdev_open_opts) == 16, "Incorrect size");
8258 :
8259 : #undef SET_FIELD
8260 2 : }
8261 :
8262 : void
8263 1 : spdk_bdev_open_opts_init(struct spdk_bdev_open_opts *opts, size_t opts_size)
8264 : {
8265 : struct spdk_bdev_open_opts opts_local;
8266 :
8267 1 : bdev_open_opts_get_defaults(&opts_local, sizeof(opts_local));
8268 1 : bdev_open_opts_copy(opts, &opts_local, opts_size);
8269 1 : }
8270 :
8271 : static int
8272 267 : bdev_desc_alloc(struct spdk_bdev *bdev, spdk_bdev_event_cb_t event_cb, void *event_ctx,
8273 : struct spdk_bdev_open_opts *user_opts, struct spdk_bdev_desc **_desc)
8274 : {
8275 : struct spdk_bdev_desc *desc;
8276 : struct spdk_bdev_open_opts opts;
8277 : unsigned int i;
8278 :
8279 267 : bdev_open_opts_get_defaults(&opts, sizeof(opts));
8280 267 : if (user_opts != NULL) {
8281 1 : bdev_open_opts_copy(&opts, user_opts, user_opts->size);
8282 1 : }
8283 :
8284 267 : desc = calloc(1, sizeof(*desc));
8285 267 : if (desc == NULL) {
8286 0 : SPDK_ERRLOG("Failed to allocate memory for bdev descriptor\n");
8287 0 : return -ENOMEM;
8288 : }
8289 :
8290 267 : desc->opts = opts;
8291 :
8292 267 : TAILQ_INIT(&desc->pending_media_events);
8293 267 : TAILQ_INIT(&desc->free_media_events);
8294 :
8295 267 : desc->memory_domains_supported = spdk_bdev_get_memory_domains(bdev, NULL, 0) > 0;
8296 267 : desc->callback.event_fn = event_cb;
8297 267 : desc->callback.ctx = event_ctx;
8298 267 : spdk_spin_init(&desc->spinlock);
8299 :
8300 267 : if (desc->opts.hide_metadata) {
8301 1 : if (spdk_bdev_is_md_separate(bdev)) {
8302 0 : SPDK_ERRLOG("hide_metadata option is not supported with separate metadata.\n");
8303 0 : bdev_desc_free(desc);
8304 0 : return -EINVAL;
8305 : }
8306 1 : }
8307 :
8308 267 : if (bdev->media_events) {
8309 0 : desc->media_events_buffer = calloc(MEDIA_EVENT_POOL_SIZE,
8310 : sizeof(*desc->media_events_buffer));
8311 0 : if (desc->media_events_buffer == NULL) {
8312 0 : SPDK_ERRLOG("Failed to initialize media event pool\n");
8313 0 : bdev_desc_free(desc);
8314 0 : return -ENOMEM;
8315 : }
8316 :
8317 0 : for (i = 0; i < MEDIA_EVENT_POOL_SIZE; ++i) {
8318 0 : TAILQ_INSERT_TAIL(&desc->free_media_events,
8319 : &desc->media_events_buffer[i], tailq);
8320 0 : }
8321 0 : }
8322 :
8323 267 : if (bdev->fn_table->accel_sequence_supported != NULL) {
8324 0 : for (i = 0; i < SPDK_BDEV_NUM_IO_TYPES; ++i) {
8325 0 : desc->accel_sequence_supported[i] =
8326 0 : bdev->fn_table->accel_sequence_supported(bdev->ctxt,
8327 0 : (enum spdk_bdev_io_type)i);
8328 0 : }
8329 0 : }
8330 :
8331 267 : *_desc = desc;
8332 :
8333 267 : return 0;
8334 267 : }
8335 :
8336 : static int
8337 130 : bdev_open_ext(const char *bdev_name, bool write, spdk_bdev_event_cb_t event_cb,
8338 : void *event_ctx, struct spdk_bdev_open_opts *opts,
8339 : struct spdk_bdev_desc **_desc)
8340 : {
8341 : struct spdk_bdev_desc *desc;
8342 : struct spdk_bdev *bdev;
8343 : int rc;
8344 :
8345 130 : bdev = bdev_get_by_name(bdev_name);
8346 :
8347 130 : if (bdev == NULL) {
8348 1 : SPDK_NOTICELOG("Currently unable to find bdev with name: %s\n", bdev_name);
8349 1 : return -ENODEV;
8350 : }
8351 :
8352 129 : rc = bdev_desc_alloc(bdev, event_cb, event_ctx, opts, &desc);
8353 129 : if (rc != 0) {
8354 0 : return rc;
8355 : }
8356 :
8357 129 : rc = bdev_open(bdev, write, desc);
8358 129 : if (rc != 0) {
8359 7 : bdev_desc_free(desc);
8360 7 : desc = NULL;
8361 7 : }
8362 :
8363 129 : *_desc = desc;
8364 :
8365 129 : return rc;
8366 130 : }
8367 :
8368 : int
8369 132 : spdk_bdev_open_ext_v2(const char *bdev_name, bool write, spdk_bdev_event_cb_t event_cb,
8370 : void *event_ctx, struct spdk_bdev_open_opts *opts,
8371 : struct spdk_bdev_desc **_desc)
8372 : {
8373 : int rc;
8374 :
8375 132 : if (event_cb == NULL) {
8376 2 : SPDK_ERRLOG("Missing event callback function\n");
8377 2 : return -EINVAL;
8378 : }
8379 :
8380 130 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8381 130 : rc = bdev_open_ext(bdev_name, write, event_cb, event_ctx, opts, _desc);
8382 130 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8383 :
8384 130 : return rc;
8385 132 : }
8386 :
8387 : int
8388 130 : spdk_bdev_open_ext(const char *bdev_name, bool write, spdk_bdev_event_cb_t event_cb,
8389 : void *event_ctx, struct spdk_bdev_desc **_desc)
8390 : {
8391 130 : return spdk_bdev_open_ext_v2(bdev_name, write, event_cb, event_ctx, NULL, _desc);
8392 : }
8393 :
8394 : struct spdk_bdev_open_async_ctx {
8395 : char *bdev_name;
8396 : spdk_bdev_event_cb_t event_cb;
8397 : void *event_ctx;
8398 : bool write;
8399 : int rc;
8400 : spdk_bdev_open_async_cb_t cb_fn;
8401 : void *cb_arg;
8402 : struct spdk_bdev_desc *desc;
8403 : struct spdk_bdev_open_async_opts opts;
8404 : uint64_t start_ticks;
8405 : struct spdk_thread *orig_thread;
8406 : struct spdk_poller *poller;
8407 : TAILQ_ENTRY(spdk_bdev_open_async_ctx) tailq;
8408 : };
8409 :
8410 : static void
8411 0 : bdev_open_async_done(void *arg)
8412 : {
8413 0 : struct spdk_bdev_open_async_ctx *ctx = arg;
8414 :
8415 0 : ctx->cb_fn(ctx->desc, ctx->rc, ctx->cb_arg);
8416 :
8417 0 : free(ctx->bdev_name);
8418 0 : free(ctx);
8419 0 : }
8420 :
8421 : static void
8422 0 : bdev_open_async_cancel(void *arg)
8423 : {
8424 0 : struct spdk_bdev_open_async_ctx *ctx = arg;
8425 :
8426 0 : assert(ctx->rc == -ESHUTDOWN);
8427 :
8428 0 : spdk_poller_unregister(&ctx->poller);
8429 :
8430 0 : bdev_open_async_done(ctx);
8431 0 : }
8432 :
8433 : /* This is called when the bdev library finishes at shutdown. */
8434 : static void
8435 65 : bdev_open_async_fini(void)
8436 : {
8437 : struct spdk_bdev_open_async_ctx *ctx, *tmp_ctx;
8438 :
8439 65 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8440 65 : TAILQ_FOREACH_SAFE(ctx, &g_bdev_mgr.async_bdev_opens, tailq, tmp_ctx) {
8441 0 : TAILQ_REMOVE(&g_bdev_mgr.async_bdev_opens, ctx, tailq);
8442 : /*
8443 : * We have to move to ctx->orig_thread to unregister ctx->poller.
8444 : * However, there is a chance that ctx->poller is executed before
8445 : * message is executed, which could result in bdev_open_async_done()
8446 : * being called twice. To avoid such race condition, set ctx->rc to
8447 : * -ESHUTDOWN.
8448 : */
8449 0 : ctx->rc = -ESHUTDOWN;
8450 0 : spdk_thread_send_msg(ctx->orig_thread, bdev_open_async_cancel, ctx);
8451 0 : }
8452 65 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8453 65 : }
8454 :
8455 : static int bdev_open_async(void *arg);
8456 :
8457 : static void
8458 0 : _bdev_open_async(struct spdk_bdev_open_async_ctx *ctx)
8459 : {
8460 : uint64_t timeout_ticks;
8461 :
8462 0 : if (ctx->rc == -ESHUTDOWN) {
8463 : /* This context is being canceled. Do nothing. */
8464 0 : return;
8465 : }
8466 :
8467 0 : ctx->rc = bdev_open_ext(ctx->bdev_name, ctx->write, ctx->event_cb, ctx->event_ctx,
8468 0 : NULL, &ctx->desc);
8469 0 : if (ctx->rc == 0 || ctx->opts.timeout_ms == 0) {
8470 0 : goto exit;
8471 : }
8472 :
8473 0 : timeout_ticks = ctx->start_ticks + ctx->opts.timeout_ms * spdk_get_ticks_hz() / 1000ull;
8474 0 : if (spdk_get_ticks() >= timeout_ticks) {
8475 0 : SPDK_ERRLOG("Timed out while waiting for bdev '%s' to appear\n", ctx->bdev_name);
8476 0 : ctx->rc = -ETIMEDOUT;
8477 0 : goto exit;
8478 : }
8479 :
8480 0 : return;
8481 :
8482 : exit:
8483 0 : spdk_poller_unregister(&ctx->poller);
8484 0 : TAILQ_REMOVE(&g_bdev_mgr.async_bdev_opens, ctx, tailq);
8485 :
8486 : /* Completion callback is processed after stack unwinding. */
8487 0 : spdk_thread_send_msg(ctx->orig_thread, bdev_open_async_done, ctx);
8488 0 : }
8489 :
8490 : static int
8491 0 : bdev_open_async(void *arg)
8492 : {
8493 0 : struct spdk_bdev_open_async_ctx *ctx = arg;
8494 :
8495 0 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8496 :
8497 0 : _bdev_open_async(ctx);
8498 :
8499 0 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8500 :
8501 0 : return SPDK_POLLER_BUSY;
8502 : }
8503 :
8504 : static void
8505 0 : bdev_open_async_opts_copy(struct spdk_bdev_open_async_opts *opts,
8506 : struct spdk_bdev_open_async_opts *opts_src,
8507 : size_t size)
8508 : {
8509 0 : assert(opts);
8510 0 : assert(opts_src);
8511 :
8512 0 : opts->size = size;
8513 :
8514 : #define SET_FIELD(field) \
8515 : if (offsetof(struct spdk_bdev_open_async_opts, field) + sizeof(opts->field) <= size) { \
8516 : opts->field = opts_src->field; \
8517 : } \
8518 :
8519 0 : SET_FIELD(timeout_ms);
8520 :
8521 : /* Do not remove this statement, you should always update this statement when you adding a new field,
8522 : * and do not forget to add the SET_FIELD statement for your added field. */
8523 : SPDK_STATIC_ASSERT(sizeof(struct spdk_bdev_open_async_opts) == 16, "Incorrect size");
8524 :
8525 : #undef SET_FIELD
8526 0 : }
8527 :
8528 : static void
8529 0 : bdev_open_async_opts_get_default(struct spdk_bdev_open_async_opts *opts, size_t size)
8530 : {
8531 0 : assert(opts);
8532 :
8533 0 : opts->size = size;
8534 :
8535 : #define SET_FIELD(field, value) \
8536 : if (offsetof(struct spdk_bdev_open_async_opts, field) + sizeof(opts->field) <= size) { \
8537 : opts->field = value; \
8538 : } \
8539 :
8540 0 : SET_FIELD(timeout_ms, 0);
8541 :
8542 : #undef SET_FIELD
8543 0 : }
8544 :
8545 : int
8546 0 : spdk_bdev_open_async(const char *bdev_name, bool write, spdk_bdev_event_cb_t event_cb,
8547 : void *event_ctx, struct spdk_bdev_open_async_opts *opts,
8548 : spdk_bdev_open_async_cb_t open_cb, void *open_cb_arg)
8549 : {
8550 : struct spdk_bdev_open_async_ctx *ctx;
8551 :
8552 0 : if (event_cb == NULL) {
8553 0 : SPDK_ERRLOG("Missing event callback function\n");
8554 0 : return -EINVAL;
8555 : }
8556 :
8557 0 : if (open_cb == NULL) {
8558 0 : SPDK_ERRLOG("Missing open callback function\n");
8559 0 : return -EINVAL;
8560 : }
8561 :
8562 0 : if (opts != NULL && opts->size == 0) {
8563 0 : SPDK_ERRLOG("size in the options structure should not be zero\n");
8564 0 : return -EINVAL;
8565 : }
8566 :
8567 0 : ctx = calloc(1, sizeof(*ctx));
8568 0 : if (ctx == NULL) {
8569 0 : SPDK_ERRLOG("Failed to allocate open context\n");
8570 0 : return -ENOMEM;
8571 : }
8572 :
8573 0 : ctx->bdev_name = strdup(bdev_name);
8574 0 : if (ctx->bdev_name == NULL) {
8575 0 : SPDK_ERRLOG("Failed to duplicate bdev_name\n");
8576 0 : free(ctx);
8577 0 : return -ENOMEM;
8578 : }
8579 :
8580 0 : ctx->poller = SPDK_POLLER_REGISTER(bdev_open_async, ctx, 100 * 1000);
8581 0 : if (ctx->poller == NULL) {
8582 0 : SPDK_ERRLOG("Failed to register bdev_open_async poller\n");
8583 0 : free(ctx->bdev_name);
8584 0 : free(ctx);
8585 0 : return -ENOMEM;
8586 : }
8587 :
8588 0 : ctx->cb_fn = open_cb;
8589 0 : ctx->cb_arg = open_cb_arg;
8590 0 : ctx->write = write;
8591 0 : ctx->event_cb = event_cb;
8592 0 : ctx->event_ctx = event_ctx;
8593 0 : ctx->orig_thread = spdk_get_thread();
8594 0 : ctx->start_ticks = spdk_get_ticks();
8595 :
8596 0 : bdev_open_async_opts_get_default(&ctx->opts, sizeof(ctx->opts));
8597 0 : if (opts != NULL) {
8598 0 : bdev_open_async_opts_copy(&ctx->opts, opts, opts->size);
8599 0 : }
8600 :
8601 0 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8602 :
8603 0 : TAILQ_INSERT_TAIL(&g_bdev_mgr.async_bdev_opens, ctx, tailq);
8604 0 : _bdev_open_async(ctx);
8605 :
8606 0 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8607 :
8608 0 : return 0;
8609 0 : }
8610 :
8611 : static void
8612 258 : bdev_close(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc)
8613 : {
8614 : int rc;
8615 :
8616 258 : spdk_spin_lock(&bdev->internal.spinlock);
8617 258 : spdk_spin_lock(&desc->spinlock);
8618 :
8619 258 : TAILQ_REMOVE(&bdev->internal.open_descs, desc, link);
8620 :
8621 258 : desc->closed = true;
8622 :
8623 258 : if (desc->claim != NULL) {
8624 16 : bdev_desc_release_claims(desc);
8625 16 : }
8626 :
8627 258 : if (0 == desc->refs) {
8628 247 : spdk_spin_unlock(&desc->spinlock);
8629 247 : bdev_desc_free(desc);
8630 247 : } else {
8631 11 : spdk_spin_unlock(&desc->spinlock);
8632 : }
8633 :
8634 : /* If no more descriptors, kill QoS channel */
8635 258 : if (bdev->internal.qos && TAILQ_EMPTY(&bdev->internal.open_descs)) {
8636 7 : SPDK_DEBUGLOG(bdev, "Closed last descriptor for bdev %s on thread %p. Stopping QoS.\n",
8637 : bdev->name, spdk_get_thread());
8638 :
8639 7 : if (bdev_qos_destroy(bdev)) {
8640 : /* There isn't anything we can do to recover here. Just let the
8641 : * old QoS poller keep running. The QoS handling won't change
8642 : * cores when the user allocates a new channel, but it won't break. */
8643 0 : SPDK_ERRLOG("Unable to shut down QoS poller. It will continue running on the current thread.\n");
8644 0 : }
8645 7 : }
8646 :
8647 258 : if (bdev->internal.status == SPDK_BDEV_STATUS_REMOVING && TAILQ_EMPTY(&bdev->internal.open_descs)) {
8648 15 : rc = bdev_unregister_unsafe(bdev);
8649 15 : spdk_spin_unlock(&bdev->internal.spinlock);
8650 :
8651 15 : if (rc == 0) {
8652 15 : spdk_io_device_unregister(__bdev_to_io_dev(bdev), bdev_destroy_cb);
8653 15 : }
8654 15 : } else {
8655 243 : spdk_spin_unlock(&bdev->internal.spinlock);
8656 : }
8657 258 : }
8658 :
8659 : void
8660 122 : spdk_bdev_close(struct spdk_bdev_desc *desc)
8661 : {
8662 122 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
8663 :
8664 122 : SPDK_DEBUGLOG(bdev, "Closing descriptor %p for bdev %s on thread %p\n", desc, bdev->name,
8665 : spdk_get_thread());
8666 :
8667 122 : assert(desc->thread == spdk_get_thread());
8668 :
8669 122 : spdk_poller_unregister(&desc->io_timeout_poller);
8670 :
8671 122 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8672 :
8673 122 : bdev_close(bdev, desc);
8674 :
8675 122 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8676 122 : }
8677 :
8678 : int32_t
8679 3 : spdk_bdev_get_numa_id(struct spdk_bdev *bdev)
8680 : {
8681 3 : if (bdev->numa.id_valid) {
8682 2 : return bdev->numa.id;
8683 : } else {
8684 1 : return SPDK_ENV_NUMA_ID_ANY;
8685 : }
8686 3 : }
8687 :
8688 : static void
8689 125 : bdev_register_finished(void *arg)
8690 : {
8691 125 : struct spdk_bdev_desc *desc = arg;
8692 125 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
8693 :
8694 125 : spdk_notify_send("bdev_register", spdk_bdev_get_name(bdev));
8695 :
8696 125 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8697 :
8698 125 : bdev_close(bdev, desc);
8699 :
8700 125 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8701 125 : }
8702 :
8703 : int
8704 128 : spdk_bdev_register(struct spdk_bdev *bdev)
8705 : {
8706 : struct spdk_bdev_desc *desc;
8707 128 : struct spdk_thread *thread = spdk_get_thread();
8708 : int rc;
8709 :
8710 128 : if (spdk_unlikely(!spdk_thread_is_app_thread(NULL))) {
8711 1 : SPDK_ERRLOG("Cannot register bdev %s on thread %p (%s)\n", bdev->name, thread,
8712 : thread ? spdk_thread_get_name(thread) : "null");
8713 1 : return -EINVAL;
8714 : }
8715 :
8716 127 : rc = bdev_register(bdev);
8717 127 : if (rc != 0) {
8718 2 : return rc;
8719 : }
8720 :
8721 : /* A descriptor is opened to prevent bdev deletion during examination */
8722 125 : rc = bdev_desc_alloc(bdev, _tmp_bdev_event_cb, NULL, NULL, &desc);
8723 125 : if (rc != 0) {
8724 0 : spdk_bdev_unregister(bdev, NULL, NULL);
8725 0 : return rc;
8726 : }
8727 :
8728 125 : rc = bdev_open(bdev, false, desc);
8729 125 : if (rc != 0) {
8730 0 : bdev_desc_free(desc);
8731 0 : spdk_bdev_unregister(bdev, NULL, NULL);
8732 0 : return rc;
8733 : }
8734 :
8735 : /* Examine configuration before initializing I/O */
8736 125 : bdev_examine(bdev);
8737 :
8738 125 : rc = spdk_bdev_wait_for_examine(bdev_register_finished, desc);
8739 125 : if (rc != 0) {
8740 0 : bdev_close(bdev, desc);
8741 0 : spdk_bdev_unregister(bdev, NULL, NULL);
8742 0 : }
8743 :
8744 125 : return rc;
8745 128 : }
8746 :
8747 : int
8748 26 : spdk_bdev_module_claim_bdev(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc,
8749 : struct spdk_bdev_module *module)
8750 : {
8751 26 : spdk_spin_lock(&bdev->internal.spinlock);
8752 :
8753 26 : if (bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE) {
8754 6 : LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
8755 6 : spdk_spin_unlock(&bdev->internal.spinlock);
8756 6 : return -EPERM;
8757 : }
8758 :
8759 20 : if (desc && !desc->write) {
8760 5 : desc->write = true;
8761 5 : }
8762 :
8763 20 : bdev->internal.claim_type = SPDK_BDEV_CLAIM_EXCL_WRITE;
8764 20 : bdev->internal.claim.v1.module = module;
8765 :
8766 20 : spdk_spin_unlock(&bdev->internal.spinlock);
8767 20 : return 0;
8768 26 : }
8769 :
8770 : void
8771 8 : spdk_bdev_module_release_bdev(struct spdk_bdev *bdev)
8772 : {
8773 8 : spdk_spin_lock(&bdev->internal.spinlock);
8774 :
8775 8 : assert(bdev->internal.claim.v1.module != NULL);
8776 8 : assert(bdev->internal.claim_type == SPDK_BDEV_CLAIM_EXCL_WRITE);
8777 8 : bdev->internal.claim_type = SPDK_BDEV_CLAIM_NONE;
8778 8 : bdev->internal.claim.v1.module = NULL;
8779 :
8780 8 : spdk_spin_unlock(&bdev->internal.spinlock);
8781 8 : }
8782 :
8783 : /*
8784 : * Start claims v2
8785 : */
8786 :
8787 : const char *
8788 24 : spdk_bdev_claim_get_name(enum spdk_bdev_claim_type type)
8789 : {
8790 24 : switch (type) {
8791 : case SPDK_BDEV_CLAIM_NONE:
8792 0 : return "not_claimed";
8793 : case SPDK_BDEV_CLAIM_EXCL_WRITE:
8794 6 : return "exclusive_write";
8795 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE:
8796 7 : return "read_many_write_one";
8797 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_NONE:
8798 5 : return "read_many_write_none";
8799 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
8800 6 : return "read_many_write_many";
8801 : default:
8802 0 : break;
8803 : }
8804 0 : return "invalid_claim";
8805 24 : }
8806 :
8807 : static bool
8808 96 : claim_type_is_v2(enum spdk_bdev_claim_type type)
8809 : {
8810 96 : switch (type) {
8811 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE:
8812 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_NONE:
8813 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
8814 96 : return true;
8815 : default:
8816 0 : break;
8817 : }
8818 0 : return false;
8819 96 : }
8820 :
8821 : /* Returns true if taking a claim with desc->write == false should make the descriptor writable. */
8822 : static bool
8823 13 : claim_type_promotes_to_write(enum spdk_bdev_claim_type type)
8824 : {
8825 13 : switch (type) {
8826 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE:
8827 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
8828 5 : return true;
8829 : default:
8830 8 : break;
8831 : }
8832 8 : return false;
8833 13 : }
8834 :
8835 : void
8836 44 : spdk_bdev_claim_opts_init(struct spdk_bdev_claim_opts *opts, size_t size)
8837 : {
8838 44 : if (opts == NULL) {
8839 0 : SPDK_ERRLOG("opts should not be NULL\n");
8840 0 : assert(opts != NULL);
8841 0 : return;
8842 : }
8843 44 : if (size == 0) {
8844 0 : SPDK_ERRLOG("size should not be zero\n");
8845 0 : assert(size != 0);
8846 0 : return;
8847 : }
8848 :
8849 44 : memset(opts, 0, size);
8850 44 : opts->opts_size = size;
8851 :
8852 : #define FIELD_OK(field) \
8853 : offsetof(struct spdk_bdev_claim_opts, field) + sizeof(opts->field) <= size
8854 :
8855 : #define SET_FIELD(field, value) \
8856 : if (FIELD_OK(field)) { \
8857 : opts->field = value; \
8858 : } \
8859 :
8860 44 : SET_FIELD(shared_claim_key, 0);
8861 :
8862 : #undef FIELD_OK
8863 : #undef SET_FIELD
8864 44 : }
8865 :
8866 : static int
8867 22 : claim_opts_copy(struct spdk_bdev_claim_opts *src, struct spdk_bdev_claim_opts *dst)
8868 : {
8869 22 : if (src->opts_size == 0) {
8870 0 : SPDK_ERRLOG("size should not be zero\n");
8871 0 : return -1;
8872 : }
8873 :
8874 22 : memset(dst, 0, sizeof(*dst));
8875 22 : dst->opts_size = src->opts_size;
8876 :
8877 : #define FIELD_OK(field) \
8878 : offsetof(struct spdk_bdev_claim_opts, field) + sizeof(src->field) <= src->opts_size
8879 :
8880 : #define SET_FIELD(field) \
8881 : if (FIELD_OK(field)) { \
8882 : dst->field = src->field; \
8883 : } \
8884 :
8885 22 : if (FIELD_OK(name)) {
8886 22 : snprintf(dst->name, sizeof(dst->name), "%s", src->name);
8887 22 : }
8888 :
8889 22 : SET_FIELD(shared_claim_key);
8890 :
8891 : /* You should not remove this statement, but need to update the assert statement
8892 : * if you add a new field, and also add a corresponding SET_FIELD statement */
8893 : SPDK_STATIC_ASSERT(sizeof(struct spdk_bdev_claim_opts) == 48, "Incorrect size");
8894 :
8895 : #undef FIELD_OK
8896 : #undef SET_FIELD
8897 22 : return 0;
8898 22 : }
8899 :
8900 : /* Returns 0 if a read-write-once claim can be taken. */
8901 : static int
8902 9 : claim_verify_rwo(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
8903 : struct spdk_bdev_claim_opts *opts, struct spdk_bdev_module *module)
8904 : {
8905 9 : struct spdk_bdev *bdev = desc->bdev;
8906 : struct spdk_bdev_desc *open_desc;
8907 :
8908 9 : assert(spdk_spin_held(&bdev->internal.spinlock));
8909 9 : assert(type == SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE);
8910 :
8911 9 : if (opts->shared_claim_key != 0) {
8912 1 : SPDK_ERRLOG("%s: key option not supported with read-write-once claims\n",
8913 : bdev->name);
8914 1 : return -EINVAL;
8915 : }
8916 8 : if (bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE) {
8917 1 : LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
8918 1 : return -EPERM;
8919 : }
8920 7 : if (desc->claim != NULL) {
8921 0 : SPDK_NOTICELOG("%s: descriptor already claimed bdev with module %s\n",
8922 : bdev->name, desc->claim->module->name);
8923 0 : return -EPERM;
8924 : }
8925 14 : TAILQ_FOREACH(open_desc, &bdev->internal.open_descs, link) {
8926 9 : if (desc != open_desc && open_desc->write) {
8927 2 : SPDK_NOTICELOG("%s: Cannot obtain read-write-once claim while "
8928 : "another descriptor is open for writing\n",
8929 : bdev->name);
8930 2 : return -EPERM;
8931 : }
8932 7 : }
8933 :
8934 5 : return 0;
8935 9 : }
8936 :
8937 : /* Returns 0 if a read-only-many claim can be taken. */
8938 : static int
8939 12 : claim_verify_rom(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
8940 : struct spdk_bdev_claim_opts *opts, struct spdk_bdev_module *module)
8941 : {
8942 12 : struct spdk_bdev *bdev = desc->bdev;
8943 : struct spdk_bdev_desc *open_desc;
8944 :
8945 12 : assert(spdk_spin_held(&bdev->internal.spinlock));
8946 12 : assert(type == SPDK_BDEV_CLAIM_READ_MANY_WRITE_NONE);
8947 12 : assert(desc->claim == NULL);
8948 :
8949 12 : if (desc->write) {
8950 3 : SPDK_ERRLOG("%s: Cannot obtain read-only-many claim with writable descriptor\n",
8951 : bdev->name);
8952 3 : return -EINVAL;
8953 : }
8954 9 : if (opts->shared_claim_key != 0) {
8955 1 : SPDK_ERRLOG("%s: key option not supported with read-only-may claims\n", bdev->name);
8956 1 : return -EINVAL;
8957 : }
8958 8 : if (bdev->internal.claim_type == SPDK_BDEV_CLAIM_NONE) {
8959 15 : TAILQ_FOREACH(open_desc, &bdev->internal.open_descs, link) {
8960 9 : if (open_desc->write) {
8961 0 : SPDK_NOTICELOG("%s: Cannot obtain read-only-many claim while "
8962 : "another descriptor is open for writing\n",
8963 : bdev->name);
8964 0 : return -EPERM;
8965 : }
8966 9 : }
8967 6 : }
8968 :
8969 8 : return 0;
8970 12 : }
8971 :
8972 : /* Returns 0 if a read-write-many claim can be taken. */
8973 : static int
8974 8 : claim_verify_rwm(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
8975 : struct spdk_bdev_claim_opts *opts, struct spdk_bdev_module *module)
8976 : {
8977 8 : struct spdk_bdev *bdev = desc->bdev;
8978 : struct spdk_bdev_desc *open_desc;
8979 :
8980 8 : assert(spdk_spin_held(&bdev->internal.spinlock));
8981 8 : assert(type == SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED);
8982 8 : assert(desc->claim == NULL);
8983 :
8984 8 : if (opts->shared_claim_key == 0) {
8985 2 : SPDK_ERRLOG("%s: shared_claim_key option required with read-write-may claims\n",
8986 : bdev->name);
8987 2 : return -EINVAL;
8988 : }
8989 6 : switch (bdev->internal.claim_type) {
8990 : case SPDK_BDEV_CLAIM_NONE:
8991 7 : TAILQ_FOREACH(open_desc, &bdev->internal.open_descs, link) {
8992 5 : if (open_desc == desc) {
8993 3 : continue;
8994 : }
8995 2 : if (open_desc->write) {
8996 2 : SPDK_NOTICELOG("%s: Cannot obtain read-write-many claim while "
8997 : "another descriptor is open for writing without a "
8998 : "claim\n", bdev->name);
8999 2 : return -EPERM;
9000 : }
9001 0 : }
9002 2 : break;
9003 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
9004 2 : if (opts->shared_claim_key != bdev->internal.claim.v2.key) {
9005 1 : LOG_ALREADY_CLAIMED_ERROR("already claimed with another key", bdev);
9006 1 : return -EPERM;
9007 : }
9008 1 : break;
9009 : default:
9010 0 : LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
9011 0 : return -EBUSY;
9012 : }
9013 :
9014 3 : return 0;
9015 8 : }
9016 :
9017 : /* Updates desc and its bdev with a v2 claim. */
9018 : static int
9019 16 : claim_bdev(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
9020 : struct spdk_bdev_claim_opts *opts, struct spdk_bdev_module *module)
9021 : {
9022 16 : struct spdk_bdev *bdev = desc->bdev;
9023 : struct spdk_bdev_module_claim *claim;
9024 :
9025 16 : assert(spdk_spin_held(&bdev->internal.spinlock));
9026 16 : assert(claim_type_is_v2(type));
9027 16 : assert(desc->claim == NULL);
9028 :
9029 16 : claim = calloc(1, sizeof(*desc->claim));
9030 16 : if (claim == NULL) {
9031 0 : SPDK_ERRLOG("%s: out of memory while allocating claim\n", bdev->name);
9032 0 : return -ENOMEM;
9033 : }
9034 16 : claim->module = module;
9035 16 : claim->desc = desc;
9036 : SPDK_STATIC_ASSERT(sizeof(claim->name) == sizeof(opts->name), "sizes must match");
9037 16 : memcpy(claim->name, opts->name, sizeof(claim->name));
9038 16 : desc->claim = claim;
9039 :
9040 16 : if (bdev->internal.claim_type == SPDK_BDEV_CLAIM_NONE) {
9041 13 : bdev->internal.claim_type = type;
9042 13 : TAILQ_INIT(&bdev->internal.claim.v2.claims);
9043 13 : bdev->internal.claim.v2.key = opts->shared_claim_key;
9044 13 : }
9045 16 : assert(type == bdev->internal.claim_type);
9046 :
9047 16 : TAILQ_INSERT_TAIL(&bdev->internal.claim.v2.claims, claim, link);
9048 :
9049 16 : if (!desc->write && claim_type_promotes_to_write(type)) {
9050 5 : desc->write = true;
9051 5 : }
9052 :
9053 16 : return 0;
9054 16 : }
9055 :
9056 : int
9057 39 : spdk_bdev_module_claim_bdev_desc(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
9058 : struct spdk_bdev_claim_opts *_opts,
9059 : struct spdk_bdev_module *module)
9060 : {
9061 : struct spdk_bdev *bdev;
9062 : struct spdk_bdev_claim_opts opts;
9063 39 : int rc = 0;
9064 :
9065 39 : if (desc == NULL) {
9066 0 : SPDK_ERRLOG("descriptor must not be NULL\n");
9067 0 : return -EINVAL;
9068 : }
9069 :
9070 39 : bdev = desc->bdev;
9071 :
9072 39 : if (_opts == NULL) {
9073 17 : spdk_bdev_claim_opts_init(&opts, sizeof(opts));
9074 39 : } else if (claim_opts_copy(_opts, &opts) != 0) {
9075 0 : return -EINVAL;
9076 : }
9077 :
9078 39 : spdk_spin_lock(&bdev->internal.spinlock);
9079 :
9080 39 : if (bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE &&
9081 15 : bdev->internal.claim_type != type) {
9082 10 : LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
9083 10 : spdk_spin_unlock(&bdev->internal.spinlock);
9084 10 : return -EPERM;
9085 : }
9086 :
9087 29 : if (claim_type_is_v2(type) && desc->claim != NULL) {
9088 0 : SPDK_ERRLOG("%s: descriptor already has %s claim with name '%s'\n",
9089 : bdev->name, spdk_bdev_claim_get_name(type), desc->claim->name);
9090 0 : spdk_spin_unlock(&bdev->internal.spinlock);
9091 0 : return -EPERM;
9092 : }
9093 :
9094 29 : switch (type) {
9095 : case SPDK_BDEV_CLAIM_EXCL_WRITE:
9096 0 : spdk_spin_unlock(&bdev->internal.spinlock);
9097 0 : return spdk_bdev_module_claim_bdev(bdev, desc, module);
9098 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE:
9099 9 : rc = claim_verify_rwo(desc, type, &opts, module);
9100 9 : break;
9101 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_NONE:
9102 12 : rc = claim_verify_rom(desc, type, &opts, module);
9103 12 : break;
9104 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
9105 8 : rc = claim_verify_rwm(desc, type, &opts, module);
9106 8 : break;
9107 : default:
9108 0 : SPDK_ERRLOG("%s: claim type %d not supported\n", bdev->name, type);
9109 0 : rc = -ENOTSUP;
9110 0 : }
9111 :
9112 29 : if (rc == 0) {
9113 16 : rc = claim_bdev(desc, type, &opts, module);
9114 16 : }
9115 :
9116 29 : spdk_spin_unlock(&bdev->internal.spinlock);
9117 29 : return rc;
9118 39 : }
9119 :
9120 : static void
9121 13 : claim_reset(struct spdk_bdev *bdev)
9122 : {
9123 13 : assert(spdk_spin_held(&bdev->internal.spinlock));
9124 13 : assert(claim_type_is_v2(bdev->internal.claim_type));
9125 13 : assert(TAILQ_EMPTY(&bdev->internal.claim.v2.claims));
9126 :
9127 13 : memset(&bdev->internal.claim, 0, sizeof(bdev->internal.claim));
9128 13 : bdev->internal.claim_type = SPDK_BDEV_CLAIM_NONE;
9129 13 : }
9130 :
9131 : static void
9132 16 : bdev_desc_release_claims(struct spdk_bdev_desc *desc)
9133 : {
9134 16 : struct spdk_bdev *bdev = desc->bdev;
9135 :
9136 16 : assert(spdk_spin_held(&bdev->internal.spinlock));
9137 16 : assert(claim_type_is_v2(bdev->internal.claim_type));
9138 :
9139 16 : if (bdev->internal.examine_in_progress == 0) {
9140 16 : TAILQ_REMOVE(&bdev->internal.claim.v2.claims, desc->claim, link);
9141 16 : free(desc->claim);
9142 16 : if (TAILQ_EMPTY(&bdev->internal.claim.v2.claims)) {
9143 13 : claim_reset(bdev);
9144 13 : }
9145 16 : } else {
9146 : /* This is a dead claim that will be cleaned up when bdev_examine() is done. */
9147 0 : desc->claim->module = NULL;
9148 0 : desc->claim->desc = NULL;
9149 : }
9150 16 : desc->claim = NULL;
9151 16 : }
9152 :
9153 : /*
9154 : * End claims v2
9155 : */
9156 :
9157 : struct spdk_bdev *
9158 1187 : spdk_bdev_desc_get_bdev(struct spdk_bdev_desc *desc)
9159 : {
9160 1187 : assert(desc != NULL);
9161 1187 : return desc->bdev;
9162 : }
9163 :
9164 : int
9165 1 : spdk_for_each_bdev(void *ctx, spdk_for_each_bdev_fn fn)
9166 : {
9167 : struct spdk_bdev *bdev, *tmp;
9168 : struct spdk_bdev_desc *desc;
9169 1 : int rc = 0;
9170 :
9171 1 : assert(fn != NULL);
9172 :
9173 1 : spdk_spin_lock(&g_bdev_mgr.spinlock);
9174 1 : bdev = spdk_bdev_first();
9175 9 : while (bdev != NULL) {
9176 8 : rc = bdev_desc_alloc(bdev, _tmp_bdev_event_cb, NULL, NULL, &desc);
9177 8 : if (rc != 0) {
9178 0 : break;
9179 : }
9180 8 : rc = bdev_open(bdev, false, desc);
9181 8 : if (rc != 0) {
9182 1 : bdev_desc_free(desc);
9183 1 : if (rc == -ENODEV) {
9184 : /* Ignore the error and move to the next bdev. */
9185 1 : rc = 0;
9186 1 : bdev = spdk_bdev_next(bdev);
9187 1 : continue;
9188 : }
9189 0 : break;
9190 : }
9191 7 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
9192 :
9193 7 : rc = fn(ctx, bdev);
9194 :
9195 7 : spdk_spin_lock(&g_bdev_mgr.spinlock);
9196 7 : tmp = spdk_bdev_next(bdev);
9197 7 : bdev_close(bdev, desc);
9198 7 : if (rc != 0) {
9199 0 : break;
9200 : }
9201 7 : bdev = tmp;
9202 : }
9203 1 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
9204 :
9205 1 : return rc;
9206 : }
9207 :
9208 : int
9209 1 : spdk_for_each_bdev_leaf(void *ctx, spdk_for_each_bdev_fn fn)
9210 : {
9211 : struct spdk_bdev *bdev, *tmp;
9212 : struct spdk_bdev_desc *desc;
9213 1 : int rc = 0;
9214 :
9215 1 : assert(fn != NULL);
9216 :
9217 1 : spdk_spin_lock(&g_bdev_mgr.spinlock);
9218 1 : bdev = spdk_bdev_first_leaf();
9219 6 : while (bdev != NULL) {
9220 5 : rc = bdev_desc_alloc(bdev, _tmp_bdev_event_cb, NULL, NULL, &desc);
9221 5 : if (rc != 0) {
9222 0 : break;
9223 : }
9224 5 : rc = bdev_open(bdev, false, desc);
9225 5 : if (rc != 0) {
9226 1 : bdev_desc_free(desc);
9227 1 : if (rc == -ENODEV) {
9228 : /* Ignore the error and move to the next bdev. */
9229 1 : rc = 0;
9230 1 : bdev = spdk_bdev_next_leaf(bdev);
9231 1 : continue;
9232 : }
9233 0 : break;
9234 : }
9235 4 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
9236 :
9237 4 : rc = fn(ctx, bdev);
9238 :
9239 4 : spdk_spin_lock(&g_bdev_mgr.spinlock);
9240 4 : tmp = spdk_bdev_next_leaf(bdev);
9241 4 : bdev_close(bdev, desc);
9242 4 : if (rc != 0) {
9243 0 : break;
9244 : }
9245 4 : bdev = tmp;
9246 : }
9247 1 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
9248 :
9249 1 : return rc;
9250 : }
9251 :
9252 : void
9253 0 : spdk_bdev_io_get_iovec(struct spdk_bdev_io *bdev_io, struct iovec **iovp, int *iovcntp)
9254 : {
9255 : struct iovec *iovs;
9256 : int iovcnt;
9257 :
9258 0 : if (bdev_io == NULL) {
9259 0 : return;
9260 : }
9261 :
9262 0 : switch (bdev_io->type) {
9263 : case SPDK_BDEV_IO_TYPE_READ:
9264 : case SPDK_BDEV_IO_TYPE_WRITE:
9265 : case SPDK_BDEV_IO_TYPE_ZCOPY:
9266 0 : iovs = bdev_io->u.bdev.iovs;
9267 0 : iovcnt = bdev_io->u.bdev.iovcnt;
9268 0 : break;
9269 : default:
9270 0 : iovs = NULL;
9271 0 : iovcnt = 0;
9272 0 : break;
9273 : }
9274 :
9275 0 : if (iovp) {
9276 0 : *iovp = iovs;
9277 0 : }
9278 0 : if (iovcntp) {
9279 0 : *iovcntp = iovcnt;
9280 0 : }
9281 0 : }
9282 :
9283 : void *
9284 0 : spdk_bdev_io_get_md_buf(struct spdk_bdev_io *bdev_io)
9285 : {
9286 0 : if (bdev_io == NULL) {
9287 0 : return NULL;
9288 : }
9289 :
9290 0 : if (!spdk_bdev_is_md_separate(bdev_io->bdev)) {
9291 0 : return NULL;
9292 : }
9293 :
9294 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_READ ||
9295 0 : bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
9296 0 : return bdev_io->u.bdev.md_buf;
9297 : }
9298 :
9299 0 : return NULL;
9300 0 : }
9301 :
9302 : void *
9303 0 : spdk_bdev_io_get_cb_arg(struct spdk_bdev_io *bdev_io)
9304 : {
9305 0 : if (bdev_io == NULL) {
9306 0 : assert(false);
9307 : return NULL;
9308 : }
9309 :
9310 0 : return bdev_io->internal.caller_ctx;
9311 : }
9312 :
9313 : void
9314 7 : spdk_bdev_module_list_add(struct spdk_bdev_module *bdev_module)
9315 : {
9316 :
9317 7 : if (spdk_bdev_module_list_find(bdev_module->name)) {
9318 0 : SPDK_ERRLOG("ERROR: module '%s' already registered.\n", bdev_module->name);
9319 0 : assert(false);
9320 : }
9321 :
9322 7 : spdk_spin_init(&bdev_module->internal.spinlock);
9323 7 : TAILQ_INIT(&bdev_module->internal.quiesced_ranges);
9324 :
9325 : /*
9326 : * Modules with examine callbacks must be initialized first, so they are
9327 : * ready to handle examine callbacks from later modules that will
9328 : * register physical bdevs.
9329 : */
9330 7 : if (bdev_module->examine_config != NULL || bdev_module->examine_disk != NULL) {
9331 4 : TAILQ_INSERT_HEAD(&g_bdev_mgr.bdev_modules, bdev_module, internal.tailq);
9332 4 : } else {
9333 3 : TAILQ_INSERT_TAIL(&g_bdev_mgr.bdev_modules, bdev_module, internal.tailq);
9334 : }
9335 7 : }
9336 :
9337 : struct spdk_bdev_module *
9338 7 : spdk_bdev_module_list_find(const char *name)
9339 : {
9340 : struct spdk_bdev_module *bdev_module;
9341 :
9342 14 : TAILQ_FOREACH(bdev_module, &g_bdev_mgr.bdev_modules, internal.tailq) {
9343 7 : if (strcmp(name, bdev_module->name) == 0) {
9344 0 : break;
9345 : }
9346 7 : }
9347 :
9348 7 : return bdev_module;
9349 : }
9350 :
9351 : static int
9352 6 : bdev_write_zero_buffer(struct spdk_bdev_io *bdev_io)
9353 : {
9354 : uint64_t num_blocks;
9355 6 : void *md_buf = NULL;
9356 :
9357 6 : num_blocks = bdev_io->u.bdev.num_blocks;
9358 :
9359 6 : if (spdk_bdev_is_md_separate(bdev_io->bdev)) {
9360 4 : md_buf = (char *)g_bdev_mgr.zero_buffer +
9361 2 : spdk_bdev_get_block_size(bdev_io->bdev) * num_blocks;
9362 2 : }
9363 :
9364 12 : return bdev_write_blocks_with_md(bdev_io->internal.desc,
9365 6 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
9366 6 : g_bdev_mgr.zero_buffer, md_buf,
9367 6 : bdev_io->u.bdev.offset_blocks, num_blocks,
9368 6 : bdev_write_zero_buffer_done, bdev_io);
9369 : }
9370 :
9371 : static void
9372 6 : bdev_write_zero_buffer_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
9373 : {
9374 6 : struct spdk_bdev_io *parent_io = cb_arg;
9375 :
9376 6 : spdk_bdev_free_io(bdev_io);
9377 :
9378 6 : parent_io->internal.status = success ? SPDK_BDEV_IO_STATUS_SUCCESS : SPDK_BDEV_IO_STATUS_FAILED;
9379 6 : parent_io->internal.cb(parent_io, success, parent_io->internal.caller_ctx);
9380 6 : }
9381 :
9382 : static void
9383 10 : bdev_set_qos_limit_done(struct set_qos_limit_ctx *ctx, int status)
9384 : {
9385 10 : spdk_spin_lock(&ctx->bdev->internal.spinlock);
9386 10 : ctx->bdev->internal.qos_mod_in_progress = false;
9387 10 : spdk_spin_unlock(&ctx->bdev->internal.spinlock);
9388 :
9389 10 : if (ctx->cb_fn) {
9390 8 : ctx->cb_fn(ctx->cb_arg, status);
9391 8 : }
9392 10 : free(ctx);
9393 10 : }
9394 :
9395 : static void
9396 2 : bdev_disable_qos_done(void *cb_arg)
9397 : {
9398 2 : struct set_qos_limit_ctx *ctx = cb_arg;
9399 2 : struct spdk_bdev *bdev = ctx->bdev;
9400 : struct spdk_bdev_qos *qos;
9401 :
9402 2 : spdk_spin_lock(&bdev->internal.spinlock);
9403 2 : qos = bdev->internal.qos;
9404 2 : bdev->internal.qos = NULL;
9405 2 : spdk_spin_unlock(&bdev->internal.spinlock);
9406 :
9407 2 : if (qos->thread != NULL) {
9408 2 : spdk_put_io_channel(spdk_io_channel_from_ctx(qos->ch));
9409 2 : spdk_poller_unregister(&qos->poller);
9410 2 : }
9411 :
9412 2 : free(qos);
9413 :
9414 2 : bdev_set_qos_limit_done(ctx, 0);
9415 2 : }
9416 :
9417 : static void
9418 2 : bdev_disable_qos_msg_done(struct spdk_bdev *bdev, void *_ctx, int status)
9419 : {
9420 2 : struct set_qos_limit_ctx *ctx = _ctx;
9421 : struct spdk_thread *thread;
9422 :
9423 2 : spdk_spin_lock(&bdev->internal.spinlock);
9424 2 : thread = bdev->internal.qos->thread;
9425 2 : spdk_spin_unlock(&bdev->internal.spinlock);
9426 :
9427 2 : if (thread != NULL) {
9428 2 : spdk_thread_send_msg(thread, bdev_disable_qos_done, ctx);
9429 2 : } else {
9430 0 : bdev_disable_qos_done(ctx);
9431 : }
9432 2 : }
9433 :
9434 : static void
9435 4 : bdev_disable_qos_msg(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
9436 : struct spdk_io_channel *ch, void *_ctx)
9437 : {
9438 4 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(ch);
9439 : struct spdk_bdev_io *bdev_io;
9440 :
9441 4 : bdev_ch->flags &= ~BDEV_CH_QOS_ENABLED;
9442 :
9443 6 : while (!TAILQ_EMPTY(&bdev_ch->qos_queued_io)) {
9444 : /* Re-submit the queued I/O. */
9445 2 : bdev_io = TAILQ_FIRST(&bdev_ch->qos_queued_io);
9446 2 : TAILQ_REMOVE(&bdev_ch->qos_queued_io, bdev_io, internal.link);
9447 2 : _bdev_io_submit(bdev_io);
9448 : }
9449 :
9450 4 : spdk_bdev_for_each_channel_continue(i, 0);
9451 4 : }
9452 :
9453 : static void
9454 1 : bdev_update_qos_rate_limit_msg(void *cb_arg)
9455 : {
9456 1 : struct set_qos_limit_ctx *ctx = cb_arg;
9457 1 : struct spdk_bdev *bdev = ctx->bdev;
9458 :
9459 1 : spdk_spin_lock(&bdev->internal.spinlock);
9460 1 : bdev_qos_update_max_quota_per_timeslice(bdev->internal.qos);
9461 1 : spdk_spin_unlock(&bdev->internal.spinlock);
9462 :
9463 1 : bdev_set_qos_limit_done(ctx, 0);
9464 1 : }
9465 :
9466 : static void
9467 9 : bdev_enable_qos_msg(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
9468 : struct spdk_io_channel *ch, void *_ctx)
9469 : {
9470 9 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(ch);
9471 :
9472 9 : spdk_spin_lock(&bdev->internal.spinlock);
9473 9 : bdev_enable_qos(bdev, bdev_ch);
9474 9 : spdk_spin_unlock(&bdev->internal.spinlock);
9475 9 : spdk_bdev_for_each_channel_continue(i, 0);
9476 9 : }
9477 :
9478 : static void
9479 6 : bdev_enable_qos_done(struct spdk_bdev *bdev, void *_ctx, int status)
9480 : {
9481 6 : struct set_qos_limit_ctx *ctx = _ctx;
9482 :
9483 6 : bdev_set_qos_limit_done(ctx, status);
9484 6 : }
9485 :
9486 : static void
9487 7 : bdev_set_qos_rate_limits(struct spdk_bdev *bdev, uint64_t *limits)
9488 : {
9489 : int i;
9490 :
9491 7 : assert(bdev->internal.qos != NULL);
9492 :
9493 35 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
9494 28 : if (limits[i] != SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
9495 28 : bdev->internal.qos->rate_limits[i].limit = limits[i];
9496 :
9497 28 : if (limits[i] == 0) {
9498 19 : bdev->internal.qos->rate_limits[i].limit =
9499 : SPDK_BDEV_QOS_LIMIT_NOT_DEFINED;
9500 19 : }
9501 28 : }
9502 28 : }
9503 7 : }
9504 :
9505 : void
9506 9 : spdk_bdev_set_qos_rate_limits(struct spdk_bdev *bdev, uint64_t *limits,
9507 : void (*cb_fn)(void *cb_arg, int status), void *cb_arg)
9508 : {
9509 : struct set_qos_limit_ctx *ctx;
9510 : uint32_t limit_set_complement;
9511 : uint64_t min_limit_per_sec;
9512 : int i;
9513 9 : bool disable_rate_limit = true;
9514 :
9515 45 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
9516 36 : if (limits[i] == SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
9517 0 : continue;
9518 : }
9519 :
9520 36 : if (limits[i] > 0) {
9521 10 : disable_rate_limit = false;
9522 10 : }
9523 :
9524 36 : if (bdev_qos_is_iops_rate_limit(i) == true) {
9525 9 : min_limit_per_sec = SPDK_BDEV_QOS_MIN_IOS_PER_SEC;
9526 9 : } else {
9527 27 : if (limits[i] > SPDK_BDEV_QOS_MAX_MBYTES_PER_SEC) {
9528 0 : SPDK_WARNLOG("Requested rate limit %" PRIu64 " will result in uint64_t overflow, "
9529 : "reset to %" PRIu64 "\n", limits[i], SPDK_BDEV_QOS_MAX_MBYTES_PER_SEC);
9530 0 : limits[i] = SPDK_BDEV_QOS_MAX_MBYTES_PER_SEC;
9531 0 : }
9532 : /* Change from megabyte to byte rate limit */
9533 27 : limits[i] = limits[i] * 1024 * 1024;
9534 27 : min_limit_per_sec = SPDK_BDEV_QOS_MIN_BYTES_PER_SEC;
9535 : }
9536 :
9537 36 : limit_set_complement = limits[i] % min_limit_per_sec;
9538 36 : if (limit_set_complement) {
9539 0 : SPDK_ERRLOG("Requested rate limit %" PRIu64 " is not a multiple of %" PRIu64 "\n",
9540 : limits[i], min_limit_per_sec);
9541 0 : limits[i] += min_limit_per_sec - limit_set_complement;
9542 0 : SPDK_ERRLOG("Round up the rate limit to %" PRIu64 "\n", limits[i]);
9543 0 : }
9544 36 : }
9545 :
9546 9 : ctx = calloc(1, sizeof(*ctx));
9547 9 : if (ctx == NULL) {
9548 0 : cb_fn(cb_arg, -ENOMEM);
9549 0 : return;
9550 : }
9551 :
9552 9 : ctx->cb_fn = cb_fn;
9553 9 : ctx->cb_arg = cb_arg;
9554 9 : ctx->bdev = bdev;
9555 :
9556 9 : spdk_spin_lock(&bdev->internal.spinlock);
9557 9 : if (bdev->internal.qos_mod_in_progress) {
9558 1 : spdk_spin_unlock(&bdev->internal.spinlock);
9559 1 : free(ctx);
9560 1 : cb_fn(cb_arg, -EAGAIN);
9561 1 : return;
9562 : }
9563 8 : bdev->internal.qos_mod_in_progress = true;
9564 :
9565 8 : if (disable_rate_limit == true && bdev->internal.qos) {
9566 10 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
9567 8 : if (limits[i] == SPDK_BDEV_QOS_LIMIT_NOT_DEFINED &&
9568 0 : (bdev->internal.qos->rate_limits[i].limit > 0 &&
9569 0 : bdev->internal.qos->rate_limits[i].limit !=
9570 : SPDK_BDEV_QOS_LIMIT_NOT_DEFINED)) {
9571 0 : disable_rate_limit = false;
9572 0 : break;
9573 : }
9574 8 : }
9575 2 : }
9576 :
9577 8 : if (disable_rate_limit == false) {
9578 5 : if (bdev->internal.qos == NULL) {
9579 4 : bdev->internal.qos = calloc(1, sizeof(*bdev->internal.qos));
9580 4 : if (!bdev->internal.qos) {
9581 0 : spdk_spin_unlock(&bdev->internal.spinlock);
9582 0 : SPDK_ERRLOG("Unable to allocate memory for QoS tracking\n");
9583 0 : bdev_set_qos_limit_done(ctx, -ENOMEM);
9584 0 : return;
9585 : }
9586 4 : }
9587 :
9588 5 : if (bdev->internal.qos->thread == NULL) {
9589 : /* Enabling */
9590 4 : bdev_set_qos_rate_limits(bdev, limits);
9591 :
9592 4 : spdk_bdev_for_each_channel(bdev, bdev_enable_qos_msg, ctx,
9593 : bdev_enable_qos_done);
9594 4 : } else {
9595 : /* Updating */
9596 1 : bdev_set_qos_rate_limits(bdev, limits);
9597 :
9598 2 : spdk_thread_send_msg(bdev->internal.qos->thread,
9599 1 : bdev_update_qos_rate_limit_msg, ctx);
9600 : }
9601 5 : } else {
9602 3 : if (bdev->internal.qos != NULL) {
9603 2 : bdev_set_qos_rate_limits(bdev, limits);
9604 :
9605 : /* Disabling */
9606 2 : spdk_bdev_for_each_channel(bdev, bdev_disable_qos_msg, ctx,
9607 : bdev_disable_qos_msg_done);
9608 2 : } else {
9609 1 : spdk_spin_unlock(&bdev->internal.spinlock);
9610 1 : bdev_set_qos_limit_done(ctx, 0);
9611 1 : return;
9612 : }
9613 : }
9614 :
9615 7 : spdk_spin_unlock(&bdev->internal.spinlock);
9616 9 : }
9617 :
9618 : struct spdk_bdev_histogram_ctx {
9619 : spdk_bdev_histogram_status_cb cb_fn;
9620 : void *cb_arg;
9621 : struct spdk_bdev *bdev;
9622 : int status;
9623 : };
9624 :
9625 : static void
9626 2 : bdev_histogram_disable_channel_cb(struct spdk_bdev *bdev, void *_ctx, int status)
9627 : {
9628 2 : struct spdk_bdev_histogram_ctx *ctx = _ctx;
9629 :
9630 2 : spdk_spin_lock(&ctx->bdev->internal.spinlock);
9631 2 : ctx->bdev->internal.histogram_in_progress = false;
9632 2 : spdk_spin_unlock(&ctx->bdev->internal.spinlock);
9633 2 : ctx->cb_fn(ctx->cb_arg, ctx->status);
9634 2 : free(ctx);
9635 2 : }
9636 :
9637 : static void
9638 3 : bdev_histogram_disable_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
9639 : struct spdk_io_channel *_ch, void *_ctx)
9640 : {
9641 3 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
9642 :
9643 3 : if (ch->histogram != NULL) {
9644 3 : spdk_histogram_data_free(ch->histogram);
9645 3 : ch->histogram = NULL;
9646 3 : }
9647 3 : spdk_bdev_for_each_channel_continue(i, 0);
9648 3 : }
9649 :
9650 : static void
9651 2 : bdev_histogram_enable_channel_cb(struct spdk_bdev *bdev, void *_ctx, int status)
9652 : {
9653 2 : struct spdk_bdev_histogram_ctx *ctx = _ctx;
9654 :
9655 2 : if (status != 0) {
9656 0 : ctx->status = status;
9657 0 : ctx->bdev->internal.histogram_enabled = false;
9658 0 : spdk_bdev_for_each_channel(ctx->bdev, bdev_histogram_disable_channel, ctx,
9659 : bdev_histogram_disable_channel_cb);
9660 0 : } else {
9661 2 : spdk_spin_lock(&ctx->bdev->internal.spinlock);
9662 2 : ctx->bdev->internal.histogram_in_progress = false;
9663 2 : spdk_spin_unlock(&ctx->bdev->internal.spinlock);
9664 2 : ctx->cb_fn(ctx->cb_arg, ctx->status);
9665 2 : free(ctx);
9666 : }
9667 2 : }
9668 :
9669 : static void
9670 3 : bdev_histogram_enable_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
9671 : struct spdk_io_channel *_ch, void *_ctx)
9672 : {
9673 3 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
9674 3 : int status = 0;
9675 :
9676 3 : if (ch->histogram == NULL) {
9677 3 : ch->histogram = spdk_histogram_data_alloc();
9678 3 : if (ch->histogram == NULL) {
9679 0 : status = -ENOMEM;
9680 0 : }
9681 3 : }
9682 :
9683 3 : spdk_bdev_for_each_channel_continue(i, status);
9684 3 : }
9685 :
9686 : void
9687 4 : spdk_bdev_histogram_enable_ext(struct spdk_bdev *bdev, spdk_bdev_histogram_status_cb cb_fn,
9688 : void *cb_arg, bool enable, struct spdk_bdev_enable_histogram_opts *opts)
9689 : {
9690 : struct spdk_bdev_histogram_ctx *ctx;
9691 :
9692 4 : ctx = calloc(1, sizeof(struct spdk_bdev_histogram_ctx));
9693 4 : if (ctx == NULL) {
9694 0 : cb_fn(cb_arg, -ENOMEM);
9695 0 : return;
9696 : }
9697 :
9698 4 : ctx->bdev = bdev;
9699 4 : ctx->status = 0;
9700 4 : ctx->cb_fn = cb_fn;
9701 4 : ctx->cb_arg = cb_arg;
9702 :
9703 4 : spdk_spin_lock(&bdev->internal.spinlock);
9704 4 : if (bdev->internal.histogram_in_progress) {
9705 0 : spdk_spin_unlock(&bdev->internal.spinlock);
9706 0 : free(ctx);
9707 0 : cb_fn(cb_arg, -EAGAIN);
9708 0 : return;
9709 : }
9710 :
9711 4 : bdev->internal.histogram_in_progress = true;
9712 4 : spdk_spin_unlock(&bdev->internal.spinlock);
9713 :
9714 4 : bdev->internal.histogram_enabled = enable;
9715 4 : bdev->internal.histogram_io_type = opts->io_type;
9716 :
9717 4 : if (enable) {
9718 : /* Allocate histogram for each channel */
9719 2 : spdk_bdev_for_each_channel(bdev, bdev_histogram_enable_channel, ctx,
9720 : bdev_histogram_enable_channel_cb);
9721 2 : } else {
9722 2 : spdk_bdev_for_each_channel(bdev, bdev_histogram_disable_channel, ctx,
9723 : bdev_histogram_disable_channel_cb);
9724 : }
9725 4 : }
9726 :
9727 : void
9728 4 : spdk_bdev_enable_histogram_opts_init(struct spdk_bdev_enable_histogram_opts *opts, size_t size)
9729 : {
9730 4 : if (opts == NULL) {
9731 0 : SPDK_ERRLOG("opts should not be NULL\n");
9732 0 : assert(opts != NULL);
9733 0 : return;
9734 : }
9735 4 : if (size == 0) {
9736 0 : SPDK_ERRLOG("size should not be zero\n");
9737 0 : assert(size != 0);
9738 0 : return;
9739 : }
9740 :
9741 4 : memset(opts, 0, size);
9742 4 : opts->size = size;
9743 :
9744 : #define FIELD_OK(field) \
9745 : offsetof(struct spdk_bdev_enable_histogram_opts, field) + sizeof(opts->field) <= size
9746 :
9747 : #define SET_FIELD(field, value) \
9748 : if (FIELD_OK(field)) { \
9749 : opts->field = value; \
9750 : } \
9751 :
9752 4 : SET_FIELD(io_type, 0);
9753 :
9754 : /* You should not remove this statement, but need to update the assert statement
9755 : * if you add a new field, and also add a corresponding SET_FIELD statement */
9756 : SPDK_STATIC_ASSERT(sizeof(struct spdk_bdev_enable_histogram_opts) == 9, "Incorrect size");
9757 :
9758 : #undef FIELD_OK
9759 : #undef SET_FIELD
9760 4 : }
9761 :
9762 : void
9763 4 : spdk_bdev_histogram_enable(struct spdk_bdev *bdev, spdk_bdev_histogram_status_cb cb_fn,
9764 : void *cb_arg, bool enable)
9765 : {
9766 : struct spdk_bdev_enable_histogram_opts opts;
9767 :
9768 4 : spdk_bdev_enable_histogram_opts_init(&opts, sizeof(opts));
9769 4 : spdk_bdev_histogram_enable_ext(bdev, cb_fn, cb_arg, enable, &opts);
9770 4 : }
9771 :
9772 : struct spdk_bdev_histogram_data_ctx {
9773 : spdk_bdev_histogram_data_cb cb_fn;
9774 : void *cb_arg;
9775 : struct spdk_bdev *bdev;
9776 : /** merged histogram data from all channels */
9777 : struct spdk_histogram_data *histogram;
9778 : };
9779 :
9780 : static void
9781 5 : bdev_histogram_get_channel_cb(struct spdk_bdev *bdev, void *_ctx, int status)
9782 : {
9783 5 : struct spdk_bdev_histogram_data_ctx *ctx = _ctx;
9784 :
9785 5 : ctx->cb_fn(ctx->cb_arg, status, ctx->histogram);
9786 5 : free(ctx);
9787 5 : }
9788 :
9789 : static void
9790 7 : bdev_histogram_get_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
9791 : struct spdk_io_channel *_ch, void *_ctx)
9792 : {
9793 7 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
9794 7 : struct spdk_bdev_histogram_data_ctx *ctx = _ctx;
9795 7 : int status = 0;
9796 :
9797 7 : if (ch->histogram == NULL) {
9798 1 : status = -EFAULT;
9799 1 : } else {
9800 6 : spdk_histogram_data_merge(ctx->histogram, ch->histogram);
9801 : }
9802 :
9803 7 : spdk_bdev_for_each_channel_continue(i, status);
9804 7 : }
9805 :
9806 : void
9807 5 : spdk_bdev_histogram_get(struct spdk_bdev *bdev, struct spdk_histogram_data *histogram,
9808 : spdk_bdev_histogram_data_cb cb_fn,
9809 : void *cb_arg)
9810 : {
9811 : struct spdk_bdev_histogram_data_ctx *ctx;
9812 :
9813 5 : ctx = calloc(1, sizeof(struct spdk_bdev_histogram_data_ctx));
9814 5 : if (ctx == NULL) {
9815 0 : cb_fn(cb_arg, -ENOMEM, NULL);
9816 0 : return;
9817 : }
9818 :
9819 5 : ctx->bdev = bdev;
9820 5 : ctx->cb_fn = cb_fn;
9821 5 : ctx->cb_arg = cb_arg;
9822 :
9823 5 : ctx->histogram = histogram;
9824 :
9825 5 : spdk_bdev_for_each_channel(bdev, bdev_histogram_get_channel, ctx,
9826 : bdev_histogram_get_channel_cb);
9827 5 : }
9828 :
9829 : void
9830 2 : spdk_bdev_channel_get_histogram(struct spdk_io_channel *ch, spdk_bdev_histogram_data_cb cb_fn,
9831 : void *cb_arg)
9832 : {
9833 2 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(ch);
9834 2 : int status = 0;
9835 :
9836 2 : assert(cb_fn != NULL);
9837 :
9838 2 : if (bdev_ch->histogram == NULL) {
9839 1 : status = -EFAULT;
9840 1 : }
9841 2 : cb_fn(cb_arg, status, bdev_ch->histogram);
9842 2 : }
9843 :
9844 : size_t
9845 0 : spdk_bdev_get_media_events(struct spdk_bdev_desc *desc, struct spdk_bdev_media_event *events,
9846 : size_t max_events)
9847 : {
9848 : struct media_event_entry *entry;
9849 0 : size_t num_events = 0;
9850 :
9851 0 : for (; num_events < max_events; ++num_events) {
9852 0 : entry = TAILQ_FIRST(&desc->pending_media_events);
9853 0 : if (entry == NULL) {
9854 0 : break;
9855 : }
9856 :
9857 0 : events[num_events] = entry->event;
9858 0 : TAILQ_REMOVE(&desc->pending_media_events, entry, tailq);
9859 0 : TAILQ_INSERT_TAIL(&desc->free_media_events, entry, tailq);
9860 0 : }
9861 :
9862 0 : return num_events;
9863 : }
9864 :
9865 : int
9866 0 : spdk_bdev_push_media_events(struct spdk_bdev *bdev, const struct spdk_bdev_media_event *events,
9867 : size_t num_events)
9868 : {
9869 : struct spdk_bdev_desc *desc;
9870 : struct media_event_entry *entry;
9871 : size_t event_id;
9872 0 : int rc = 0;
9873 :
9874 0 : assert(bdev->media_events);
9875 :
9876 0 : spdk_spin_lock(&bdev->internal.spinlock);
9877 0 : TAILQ_FOREACH(desc, &bdev->internal.open_descs, link) {
9878 0 : if (desc->write) {
9879 0 : break;
9880 : }
9881 0 : }
9882 :
9883 0 : if (desc == NULL || desc->media_events_buffer == NULL) {
9884 0 : rc = -ENODEV;
9885 0 : goto out;
9886 : }
9887 :
9888 0 : for (event_id = 0; event_id < num_events; ++event_id) {
9889 0 : entry = TAILQ_FIRST(&desc->free_media_events);
9890 0 : if (entry == NULL) {
9891 0 : break;
9892 : }
9893 :
9894 0 : TAILQ_REMOVE(&desc->free_media_events, entry, tailq);
9895 0 : TAILQ_INSERT_TAIL(&desc->pending_media_events, entry, tailq);
9896 0 : entry->event = events[event_id];
9897 0 : }
9898 :
9899 0 : rc = event_id;
9900 : out:
9901 0 : spdk_spin_unlock(&bdev->internal.spinlock);
9902 0 : return rc;
9903 : }
9904 :
9905 : static void
9906 0 : _media_management_notify(void *arg)
9907 : {
9908 0 : struct spdk_bdev_desc *desc = arg;
9909 :
9910 0 : _event_notify(desc, SPDK_BDEV_EVENT_MEDIA_MANAGEMENT);
9911 0 : }
9912 :
9913 : void
9914 0 : spdk_bdev_notify_media_management(struct spdk_bdev *bdev)
9915 : {
9916 : struct spdk_bdev_desc *desc;
9917 :
9918 0 : spdk_spin_lock(&bdev->internal.spinlock);
9919 0 : TAILQ_FOREACH(desc, &bdev->internal.open_descs, link) {
9920 0 : if (!TAILQ_EMPTY(&desc->pending_media_events)) {
9921 0 : event_notify(desc, _media_management_notify);
9922 0 : }
9923 0 : }
9924 0 : spdk_spin_unlock(&bdev->internal.spinlock);
9925 0 : }
9926 :
9927 : struct locked_lba_range_ctx {
9928 : struct lba_range range;
9929 : struct lba_range *current_range;
9930 : struct lba_range *owner_range;
9931 : struct spdk_poller *poller;
9932 : lock_range_cb cb_fn;
9933 : void *cb_arg;
9934 : };
9935 :
9936 : static void
9937 0 : bdev_lock_error_cleanup_cb(struct spdk_bdev *bdev, void *_ctx, int status)
9938 : {
9939 0 : struct locked_lba_range_ctx *ctx = _ctx;
9940 :
9941 0 : ctx->cb_fn(&ctx->range, ctx->cb_arg, -ENOMEM);
9942 0 : free(ctx);
9943 0 : }
9944 :
9945 : static void bdev_unlock_lba_range_get_channel(struct spdk_bdev_channel_iter *i,
9946 : struct spdk_bdev *bdev, struct spdk_io_channel *ch, void *_ctx);
9947 :
9948 : static void
9949 14 : bdev_lock_lba_range_cb(struct spdk_bdev *bdev, void *_ctx, int status)
9950 : {
9951 14 : struct locked_lba_range_ctx *ctx = _ctx;
9952 :
9953 14 : if (status == -ENOMEM) {
9954 : /* One of the channels could not allocate a range object.
9955 : * So we have to go back and clean up any ranges that were
9956 : * allocated successfully before we return error status to
9957 : * the caller. We can reuse the unlock function to do that
9958 : * clean up.
9959 : */
9960 0 : spdk_bdev_for_each_channel(bdev, bdev_unlock_lba_range_get_channel, ctx,
9961 : bdev_lock_error_cleanup_cb);
9962 0 : return;
9963 : }
9964 :
9965 : /* All channels have locked this range and no I/O overlapping the range
9966 : * are outstanding! Set the owner_ch for the range object for the
9967 : * locking channel, so that this channel will know that it is allowed
9968 : * to write to this range.
9969 : */
9970 14 : if (ctx->owner_range != NULL) {
9971 10 : ctx->owner_range->owner_ch = ctx->range.owner_ch;
9972 10 : }
9973 :
9974 14 : ctx->cb_fn(&ctx->range, ctx->cb_arg, status);
9975 :
9976 : /* Don't free the ctx here. Its range is in the bdev's global list of
9977 : * locked ranges still, and will be removed and freed when this range
9978 : * is later unlocked.
9979 : */
9980 14 : }
9981 :
9982 : static int
9983 17 : bdev_lock_lba_range_check_io(void *_i)
9984 : {
9985 17 : struct spdk_bdev_channel_iter *i = _i;
9986 17 : struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i->i);
9987 17 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
9988 17 : struct locked_lba_range_ctx *ctx = i->ctx;
9989 17 : struct lba_range *range = ctx->current_range;
9990 : struct spdk_bdev_io *bdev_io;
9991 :
9992 17 : spdk_poller_unregister(&ctx->poller);
9993 :
9994 : /* The range is now in the locked_ranges, so no new IO can be submitted to this
9995 : * range. But we need to wait until any outstanding IO overlapping with this range
9996 : * are completed.
9997 : */
9998 18 : TAILQ_FOREACH(bdev_io, &ch->io_submitted, internal.ch_link) {
9999 3 : if (bdev_io_range_is_locked(bdev_io, range)) {
10000 2 : ctx->poller = SPDK_POLLER_REGISTER(bdev_lock_lba_range_check_io, i, 100);
10001 2 : return SPDK_POLLER_BUSY;
10002 : }
10003 1 : }
10004 :
10005 15 : spdk_bdev_for_each_channel_continue(i, 0);
10006 15 : return SPDK_POLLER_BUSY;
10007 17 : }
10008 :
10009 : static void
10010 15 : bdev_lock_lba_range_get_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
10011 : struct spdk_io_channel *_ch, void *_ctx)
10012 : {
10013 15 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
10014 15 : struct locked_lba_range_ctx *ctx = _ctx;
10015 : struct lba_range *range;
10016 :
10017 16 : TAILQ_FOREACH(range, &ch->locked_ranges, tailq) {
10018 1 : if (range->length == ctx->range.length &&
10019 0 : range->offset == ctx->range.offset &&
10020 0 : range->locked_ctx == ctx->range.locked_ctx) {
10021 : /* This range already exists on this channel, so don't add
10022 : * it again. This can happen when a new channel is created
10023 : * while the for_each_channel operation is in progress.
10024 : * Do not check for outstanding I/O in that case, since the
10025 : * range was locked before any I/O could be submitted to the
10026 : * new channel.
10027 : */
10028 0 : spdk_bdev_for_each_channel_continue(i, 0);
10029 0 : return;
10030 : }
10031 1 : }
10032 :
10033 15 : range = calloc(1, sizeof(*range));
10034 15 : if (range == NULL) {
10035 0 : spdk_bdev_for_each_channel_continue(i, -ENOMEM);
10036 0 : return;
10037 : }
10038 :
10039 15 : range->length = ctx->range.length;
10040 15 : range->offset = ctx->range.offset;
10041 15 : range->locked_ctx = ctx->range.locked_ctx;
10042 15 : range->quiesce = ctx->range.quiesce;
10043 15 : ctx->current_range = range;
10044 15 : if (ctx->range.owner_ch == ch) {
10045 : /* This is the range object for the channel that will hold
10046 : * the lock. Store it in the ctx object so that we can easily
10047 : * set its owner_ch after the lock is finally acquired.
10048 : */
10049 10 : ctx->owner_range = range;
10050 10 : }
10051 15 : TAILQ_INSERT_TAIL(&ch->locked_ranges, range, tailq);
10052 15 : bdev_lock_lba_range_check_io(i);
10053 15 : }
10054 :
10055 : static void
10056 14 : bdev_lock_lba_range_ctx(struct spdk_bdev *bdev, struct locked_lba_range_ctx *ctx)
10057 : {
10058 14 : assert(spdk_get_thread() == ctx->range.owner_thread);
10059 14 : assert(ctx->range.owner_ch == NULL ||
10060 : spdk_io_channel_get_thread(ctx->range.owner_ch->channel) == ctx->range.owner_thread);
10061 :
10062 : /* We will add a copy of this range to each channel now. */
10063 14 : spdk_bdev_for_each_channel(bdev, bdev_lock_lba_range_get_channel, ctx,
10064 : bdev_lock_lba_range_cb);
10065 14 : }
10066 :
10067 : static bool
10068 17 : bdev_lba_range_overlaps_tailq(struct lba_range *range, lba_range_tailq_t *tailq)
10069 : {
10070 : struct lba_range *r;
10071 :
10072 18 : TAILQ_FOREACH(r, tailq, tailq) {
10073 4 : if (bdev_lba_range_overlapped(range, r)) {
10074 3 : return true;
10075 : }
10076 1 : }
10077 14 : return false;
10078 17 : }
10079 :
10080 : static void bdev_quiesce_range_locked(struct lba_range *range, void *ctx, int status);
10081 :
10082 : static int
10083 14 : _bdev_lock_lba_range(struct spdk_bdev *bdev, struct spdk_bdev_channel *ch,
10084 : uint64_t offset, uint64_t length,
10085 : lock_range_cb cb_fn, void *cb_arg)
10086 : {
10087 : struct locked_lba_range_ctx *ctx;
10088 :
10089 14 : ctx = calloc(1, sizeof(*ctx));
10090 14 : if (ctx == NULL) {
10091 0 : return -ENOMEM;
10092 : }
10093 :
10094 14 : ctx->range.offset = offset;
10095 14 : ctx->range.length = length;
10096 14 : ctx->range.owner_thread = spdk_get_thread();
10097 14 : ctx->range.owner_ch = ch;
10098 14 : ctx->range.locked_ctx = cb_arg;
10099 14 : ctx->range.bdev = bdev;
10100 14 : ctx->range.quiesce = (cb_fn == bdev_quiesce_range_locked);
10101 14 : ctx->cb_fn = cb_fn;
10102 14 : ctx->cb_arg = cb_arg;
10103 :
10104 14 : spdk_spin_lock(&bdev->internal.spinlock);
10105 14 : if (bdev_lba_range_overlaps_tailq(&ctx->range, &bdev->internal.locked_ranges)) {
10106 : /* There is an active lock overlapping with this range.
10107 : * Put it on the pending list until this range no
10108 : * longer overlaps with another.
10109 : */
10110 2 : TAILQ_INSERT_TAIL(&bdev->internal.pending_locked_ranges, &ctx->range, tailq);
10111 2 : } else {
10112 12 : TAILQ_INSERT_TAIL(&bdev->internal.locked_ranges, &ctx->range, tailq);
10113 12 : bdev_lock_lba_range_ctx(bdev, ctx);
10114 : }
10115 14 : spdk_spin_unlock(&bdev->internal.spinlock);
10116 14 : return 0;
10117 14 : }
10118 :
10119 : static int
10120 10 : bdev_lock_lba_range(struct spdk_bdev_desc *desc, struct spdk_io_channel *_ch,
10121 : uint64_t offset, uint64_t length,
10122 : lock_range_cb cb_fn, void *cb_arg)
10123 : {
10124 10 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
10125 10 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
10126 :
10127 10 : if (cb_arg == NULL) {
10128 0 : SPDK_ERRLOG("cb_arg must not be NULL\n");
10129 0 : return -EINVAL;
10130 : }
10131 :
10132 10 : return _bdev_lock_lba_range(bdev, ch, offset, length, cb_fn, cb_arg);
10133 10 : }
10134 :
10135 : static void
10136 2 : bdev_lock_lba_range_ctx_msg(void *_ctx)
10137 : {
10138 2 : struct locked_lba_range_ctx *ctx = _ctx;
10139 :
10140 2 : bdev_lock_lba_range_ctx(ctx->range.bdev, ctx);
10141 2 : }
10142 :
10143 : static void
10144 14 : bdev_unlock_lba_range_cb(struct spdk_bdev *bdev, void *_ctx, int status)
10145 : {
10146 14 : struct locked_lba_range_ctx *ctx = _ctx;
10147 : struct locked_lba_range_ctx *pending_ctx;
10148 : struct lba_range *range, *tmp;
10149 :
10150 14 : spdk_spin_lock(&bdev->internal.spinlock);
10151 : /* Check if there are any pending locked ranges that overlap with this range
10152 : * that was just unlocked. If there are, check that it doesn't overlap with any
10153 : * other locked ranges before calling bdev_lock_lba_range_ctx which will start
10154 : * the lock process.
10155 : */
10156 17 : TAILQ_FOREACH_SAFE(range, &bdev->internal.pending_locked_ranges, tailq, tmp) {
10157 3 : if (bdev_lba_range_overlapped(range, &ctx->range) &&
10158 3 : !bdev_lba_range_overlaps_tailq(range, &bdev->internal.locked_ranges)) {
10159 2 : TAILQ_REMOVE(&bdev->internal.pending_locked_ranges, range, tailq);
10160 2 : pending_ctx = SPDK_CONTAINEROF(range, struct locked_lba_range_ctx, range);
10161 2 : TAILQ_INSERT_TAIL(&bdev->internal.locked_ranges, range, tailq);
10162 4 : spdk_thread_send_msg(pending_ctx->range.owner_thread,
10163 2 : bdev_lock_lba_range_ctx_msg, pending_ctx);
10164 2 : }
10165 3 : }
10166 14 : spdk_spin_unlock(&bdev->internal.spinlock);
10167 :
10168 14 : ctx->cb_fn(&ctx->range, ctx->cb_arg, status);
10169 14 : free(ctx);
10170 14 : }
10171 :
10172 : static void
10173 16 : bdev_unlock_lba_range_get_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
10174 : struct spdk_io_channel *_ch, void *_ctx)
10175 : {
10176 16 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
10177 16 : struct locked_lba_range_ctx *ctx = _ctx;
10178 : TAILQ_HEAD(, spdk_bdev_io) io_locked;
10179 : struct spdk_bdev_io *bdev_io;
10180 : struct lba_range *range;
10181 :
10182 16 : TAILQ_FOREACH(range, &ch->locked_ranges, tailq) {
10183 32 : if (ctx->range.offset == range->offset &&
10184 16 : ctx->range.length == range->length &&
10185 16 : ctx->range.locked_ctx == range->locked_ctx) {
10186 16 : TAILQ_REMOVE(&ch->locked_ranges, range, tailq);
10187 16 : free(range);
10188 16 : break;
10189 : }
10190 0 : }
10191 :
10192 : /* Note: we should almost always be able to assert that the range specified
10193 : * was found. But there are some very rare corner cases where a new channel
10194 : * gets created simultaneously with a range unlock, where this function
10195 : * would execute on that new channel and wouldn't have the range.
10196 : * We also use this to clean up range allocations when a later allocation
10197 : * fails in the locking path.
10198 : * So we can't actually assert() here.
10199 : */
10200 :
10201 : /* Swap the locked IO into a temporary list, and then try to submit them again.
10202 : * We could hyper-optimize this to only resubmit locked I/O that overlap
10203 : * with the range that was just unlocked, but this isn't a performance path so
10204 : * we go for simplicity here.
10205 : */
10206 16 : TAILQ_INIT(&io_locked);
10207 16 : TAILQ_SWAP(&ch->io_locked, &io_locked, spdk_bdev_io, internal.ch_link);
10208 19 : while (!TAILQ_EMPTY(&io_locked)) {
10209 3 : bdev_io = TAILQ_FIRST(&io_locked);
10210 3 : TAILQ_REMOVE(&io_locked, bdev_io, internal.ch_link);
10211 3 : bdev_io_submit(bdev_io);
10212 : }
10213 :
10214 16 : spdk_bdev_for_each_channel_continue(i, 0);
10215 16 : }
10216 :
10217 : static int
10218 14 : _bdev_unlock_lba_range(struct spdk_bdev *bdev, uint64_t offset, uint64_t length,
10219 : lock_range_cb cb_fn, void *cb_arg)
10220 : {
10221 : struct locked_lba_range_ctx *ctx;
10222 : struct lba_range *range;
10223 :
10224 14 : spdk_spin_lock(&bdev->internal.spinlock);
10225 : /* To start the unlock the process, we find the range in the bdev's locked_ranges
10226 : * and remove it. This ensures new channels don't inherit the locked range.
10227 : * Then we will send a message to each channel to remove the range from its
10228 : * per-channel list.
10229 : */
10230 14 : TAILQ_FOREACH(range, &bdev->internal.locked_ranges, tailq) {
10231 24 : if (range->offset == offset && range->length == length &&
10232 14 : (range->owner_ch == NULL || range->locked_ctx == cb_arg)) {
10233 14 : break;
10234 : }
10235 0 : }
10236 14 : if (range == NULL) {
10237 0 : assert(false);
10238 : spdk_spin_unlock(&bdev->internal.spinlock);
10239 : return -EINVAL;
10240 : }
10241 14 : TAILQ_REMOVE(&bdev->internal.locked_ranges, range, tailq);
10242 14 : ctx = SPDK_CONTAINEROF(range, struct locked_lba_range_ctx, range);
10243 14 : spdk_spin_unlock(&bdev->internal.spinlock);
10244 :
10245 14 : ctx->cb_fn = cb_fn;
10246 14 : ctx->cb_arg = cb_arg;
10247 :
10248 14 : spdk_bdev_for_each_channel(bdev, bdev_unlock_lba_range_get_channel, ctx,
10249 : bdev_unlock_lba_range_cb);
10250 14 : return 0;
10251 : }
10252 :
10253 : static int
10254 12 : bdev_unlock_lba_range(struct spdk_bdev_desc *desc, struct spdk_io_channel *_ch,
10255 : uint64_t offset, uint64_t length,
10256 : lock_range_cb cb_fn, void *cb_arg)
10257 : {
10258 12 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
10259 12 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
10260 : struct lba_range *range;
10261 12 : bool range_found = false;
10262 :
10263 : /* Let's make sure the specified channel actually has a lock on
10264 : * the specified range. Note that the range must match exactly.
10265 : */
10266 14 : TAILQ_FOREACH(range, &ch->locked_ranges, tailq) {
10267 22 : if (range->offset == offset && range->length == length &&
10268 11 : range->owner_ch == ch && range->locked_ctx == cb_arg) {
10269 10 : range_found = true;
10270 10 : break;
10271 : }
10272 2 : }
10273 :
10274 12 : if (!range_found) {
10275 2 : return -EINVAL;
10276 : }
10277 :
10278 10 : return _bdev_unlock_lba_range(bdev, offset, length, cb_fn, cb_arg);
10279 12 : }
10280 :
10281 : struct bdev_quiesce_ctx {
10282 : spdk_bdev_quiesce_cb cb_fn;
10283 : void *cb_arg;
10284 : };
10285 :
10286 : static void
10287 4 : bdev_unquiesce_range_unlocked(struct lba_range *range, void *ctx, int status)
10288 : {
10289 4 : struct bdev_quiesce_ctx *quiesce_ctx = ctx;
10290 :
10291 4 : if (quiesce_ctx->cb_fn != NULL) {
10292 4 : quiesce_ctx->cb_fn(quiesce_ctx->cb_arg, status);
10293 4 : }
10294 :
10295 4 : free(quiesce_ctx);
10296 4 : }
10297 :
10298 : static void
10299 4 : bdev_quiesce_range_locked(struct lba_range *range, void *ctx, int status)
10300 : {
10301 4 : struct bdev_quiesce_ctx *quiesce_ctx = ctx;
10302 4 : struct spdk_bdev_module *module = range->bdev->module;
10303 :
10304 4 : if (status != 0) {
10305 0 : if (quiesce_ctx->cb_fn != NULL) {
10306 0 : quiesce_ctx->cb_fn(quiesce_ctx->cb_arg, status);
10307 0 : }
10308 0 : free(quiesce_ctx);
10309 0 : return;
10310 : }
10311 :
10312 4 : spdk_spin_lock(&module->internal.spinlock);
10313 4 : TAILQ_INSERT_TAIL(&module->internal.quiesced_ranges, range, tailq_module);
10314 4 : spdk_spin_unlock(&module->internal.spinlock);
10315 :
10316 4 : if (quiesce_ctx->cb_fn != NULL) {
10317 : /* copy the context in case the range is unlocked by the callback */
10318 4 : struct bdev_quiesce_ctx tmp = *quiesce_ctx;
10319 :
10320 4 : quiesce_ctx->cb_fn = NULL;
10321 4 : quiesce_ctx->cb_arg = NULL;
10322 :
10323 4 : tmp.cb_fn(tmp.cb_arg, status);
10324 4 : }
10325 : /* quiesce_ctx will be freed on unquiesce */
10326 4 : }
10327 :
10328 : static int
10329 9 : _spdk_bdev_quiesce(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
10330 : uint64_t offset, uint64_t length,
10331 : spdk_bdev_quiesce_cb cb_fn, void *cb_arg,
10332 : bool unquiesce)
10333 : {
10334 : struct bdev_quiesce_ctx *quiesce_ctx;
10335 : int rc;
10336 :
10337 9 : if (module != bdev->module) {
10338 0 : SPDK_ERRLOG("Bdev does not belong to specified module.\n");
10339 0 : return -EINVAL;
10340 : }
10341 :
10342 9 : if (!bdev_io_valid_blocks(bdev, offset, length)) {
10343 0 : return -EINVAL;
10344 : }
10345 :
10346 9 : if (unquiesce) {
10347 : struct lba_range *range;
10348 :
10349 : /* Make sure the specified range is actually quiesced in the specified module and
10350 : * then remove it from the list. Note that the range must match exactly.
10351 : */
10352 5 : spdk_spin_lock(&module->internal.spinlock);
10353 6 : TAILQ_FOREACH(range, &module->internal.quiesced_ranges, tailq_module) {
10354 5 : if (range->bdev == bdev && range->offset == offset && range->length == length) {
10355 4 : TAILQ_REMOVE(&module->internal.quiesced_ranges, range, tailq_module);
10356 4 : break;
10357 : }
10358 1 : }
10359 5 : spdk_spin_unlock(&module->internal.spinlock);
10360 :
10361 5 : if (range == NULL) {
10362 1 : SPDK_ERRLOG("The range to unquiesce was not found.\n");
10363 1 : return -EINVAL;
10364 : }
10365 :
10366 4 : quiesce_ctx = range->locked_ctx;
10367 4 : quiesce_ctx->cb_fn = cb_fn;
10368 4 : quiesce_ctx->cb_arg = cb_arg;
10369 :
10370 4 : rc = _bdev_unlock_lba_range(bdev, offset, length, bdev_unquiesce_range_unlocked, quiesce_ctx);
10371 4 : } else {
10372 4 : quiesce_ctx = malloc(sizeof(*quiesce_ctx));
10373 4 : if (quiesce_ctx == NULL) {
10374 0 : return -ENOMEM;
10375 : }
10376 :
10377 4 : quiesce_ctx->cb_fn = cb_fn;
10378 4 : quiesce_ctx->cb_arg = cb_arg;
10379 :
10380 4 : rc = _bdev_lock_lba_range(bdev, NULL, offset, length, bdev_quiesce_range_locked, quiesce_ctx);
10381 4 : if (rc != 0) {
10382 0 : free(quiesce_ctx);
10383 0 : }
10384 : }
10385 :
10386 8 : return rc;
10387 9 : }
10388 :
10389 : int
10390 3 : spdk_bdev_quiesce(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
10391 : spdk_bdev_quiesce_cb cb_fn, void *cb_arg)
10392 : {
10393 3 : return _spdk_bdev_quiesce(bdev, module, 0, bdev->blockcnt, cb_fn, cb_arg, false);
10394 : }
10395 :
10396 : int
10397 3 : spdk_bdev_unquiesce(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
10398 : spdk_bdev_quiesce_cb cb_fn, void *cb_arg)
10399 : {
10400 3 : return _spdk_bdev_quiesce(bdev, module, 0, bdev->blockcnt, cb_fn, cb_arg, true);
10401 : }
10402 :
10403 : int
10404 1 : spdk_bdev_quiesce_range(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
10405 : uint64_t offset, uint64_t length,
10406 : spdk_bdev_quiesce_cb cb_fn, void *cb_arg)
10407 : {
10408 1 : return _spdk_bdev_quiesce(bdev, module, offset, length, cb_fn, cb_arg, false);
10409 : }
10410 :
10411 : int
10412 2 : spdk_bdev_unquiesce_range(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
10413 : uint64_t offset, uint64_t length,
10414 : spdk_bdev_quiesce_cb cb_fn, void *cb_arg)
10415 : {
10416 2 : return _spdk_bdev_quiesce(bdev, module, offset, length, cb_fn, cb_arg, true);
10417 : }
10418 :
10419 : int
10420 272 : spdk_bdev_get_memory_domains(struct spdk_bdev *bdev, struct spdk_memory_domain **domains,
10421 : int array_size)
10422 : {
10423 272 : if (!bdev) {
10424 1 : return -EINVAL;
10425 : }
10426 :
10427 271 : if (bdev->fn_table->get_memory_domains) {
10428 3 : return bdev->fn_table->get_memory_domains(bdev->ctxt, domains, array_size);
10429 : }
10430 :
10431 268 : return 0;
10432 272 : }
10433 :
10434 : struct spdk_bdev_for_each_io_ctx {
10435 : void *ctx;
10436 : spdk_bdev_io_fn fn;
10437 : spdk_bdev_for_each_io_cb cb;
10438 : };
10439 :
10440 : static void
10441 0 : bdev_channel_for_each_io(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
10442 : struct spdk_io_channel *io_ch, void *_ctx)
10443 : {
10444 0 : struct spdk_bdev_for_each_io_ctx *ctx = _ctx;
10445 0 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
10446 : struct spdk_bdev_io *bdev_io;
10447 0 : int rc = 0;
10448 :
10449 0 : TAILQ_FOREACH(bdev_io, &bdev_ch->io_submitted, internal.ch_link) {
10450 0 : rc = ctx->fn(ctx->ctx, bdev_io);
10451 0 : if (rc != 0) {
10452 0 : break;
10453 : }
10454 0 : }
10455 :
10456 0 : spdk_bdev_for_each_channel_continue(i, rc);
10457 0 : }
10458 :
10459 : static void
10460 0 : bdev_for_each_io_done(struct spdk_bdev *bdev, void *_ctx, int status)
10461 : {
10462 0 : struct spdk_bdev_for_each_io_ctx *ctx = _ctx;
10463 :
10464 0 : ctx->cb(ctx->ctx, status);
10465 :
10466 0 : free(ctx);
10467 0 : }
10468 :
10469 : void
10470 0 : spdk_bdev_for_each_bdev_io(struct spdk_bdev *bdev, void *_ctx, spdk_bdev_io_fn fn,
10471 : spdk_bdev_for_each_io_cb cb)
10472 : {
10473 : struct spdk_bdev_for_each_io_ctx *ctx;
10474 :
10475 0 : assert(fn != NULL && cb != NULL);
10476 :
10477 0 : ctx = calloc(1, sizeof(*ctx));
10478 0 : if (ctx == NULL) {
10479 0 : SPDK_ERRLOG("Failed to allocate context.\n");
10480 0 : cb(_ctx, -ENOMEM);
10481 0 : return;
10482 : }
10483 :
10484 0 : ctx->ctx = _ctx;
10485 0 : ctx->fn = fn;
10486 0 : ctx->cb = cb;
10487 :
10488 0 : spdk_bdev_for_each_channel(bdev, bdev_channel_for_each_io, ctx,
10489 : bdev_for_each_io_done);
10490 0 : }
10491 :
10492 : void
10493 132 : spdk_bdev_for_each_channel_continue(struct spdk_bdev_channel_iter *iter, int status)
10494 : {
10495 132 : spdk_for_each_channel_continue(iter->i, status);
10496 132 : }
10497 :
10498 : static struct spdk_bdev *
10499 361 : io_channel_iter_get_bdev(struct spdk_io_channel_iter *i)
10500 : {
10501 361 : void *io_device = spdk_io_channel_iter_get_io_device(i);
10502 :
10503 361 : return __bdev_from_io_dev(io_device);
10504 : }
10505 :
10506 : static void
10507 132 : bdev_each_channel_msg(struct spdk_io_channel_iter *i)
10508 : {
10509 132 : struct spdk_bdev_channel_iter *iter = spdk_io_channel_iter_get_ctx(i);
10510 132 : struct spdk_bdev *bdev = io_channel_iter_get_bdev(i);
10511 132 : struct spdk_io_channel *ch = spdk_io_channel_iter_get_channel(i);
10512 :
10513 132 : iter->i = i;
10514 132 : iter->fn(iter, bdev, ch, iter->ctx);
10515 132 : }
10516 :
10517 : static void
10518 229 : bdev_each_channel_cpl(struct spdk_io_channel_iter *i, int status)
10519 : {
10520 229 : struct spdk_bdev_channel_iter *iter = spdk_io_channel_iter_get_ctx(i);
10521 229 : struct spdk_bdev *bdev = io_channel_iter_get_bdev(i);
10522 :
10523 229 : iter->i = i;
10524 229 : iter->cpl(bdev, iter->ctx, status);
10525 :
10526 229 : free(iter);
10527 229 : }
10528 :
10529 : void
10530 229 : spdk_bdev_for_each_channel(struct spdk_bdev *bdev, spdk_bdev_for_each_channel_msg fn,
10531 : void *ctx, spdk_bdev_for_each_channel_done cpl)
10532 : {
10533 : struct spdk_bdev_channel_iter *iter;
10534 :
10535 229 : assert(bdev != NULL && fn != NULL && ctx != NULL);
10536 :
10537 229 : iter = calloc(1, sizeof(struct spdk_bdev_channel_iter));
10538 229 : if (iter == NULL) {
10539 0 : SPDK_ERRLOG("Unable to allocate iterator\n");
10540 0 : assert(false);
10541 : return;
10542 : }
10543 :
10544 229 : iter->fn = fn;
10545 229 : iter->cpl = cpl;
10546 229 : iter->ctx = ctx;
10547 :
10548 458 : spdk_for_each_channel(__bdev_to_io_dev(bdev), bdev_each_channel_msg,
10549 229 : iter, bdev_each_channel_cpl);
10550 229 : }
10551 :
10552 : static void
10553 3 : bdev_copy_do_write_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
10554 : {
10555 3 : struct spdk_bdev_io *parent_io = cb_arg;
10556 :
10557 3 : spdk_bdev_free_io(bdev_io);
10558 :
10559 : /* Check return status of write */
10560 3 : parent_io->internal.status = success ? SPDK_BDEV_IO_STATUS_SUCCESS : SPDK_BDEV_IO_STATUS_FAILED;
10561 3 : parent_io->internal.cb(parent_io, success, parent_io->internal.caller_ctx);
10562 3 : }
10563 :
10564 : static void
10565 3 : bdev_copy_do_write(void *_bdev_io)
10566 : {
10567 3 : struct spdk_bdev_io *bdev_io = _bdev_io;
10568 : int rc;
10569 :
10570 : /* Write blocks */
10571 6 : rc = spdk_bdev_write_blocks_with_md(bdev_io->internal.desc,
10572 3 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
10573 3 : bdev_io->u.bdev.iovs[0].iov_base,
10574 3 : bdev_io->u.bdev.md_buf, bdev_io->u.bdev.offset_blocks,
10575 3 : bdev_io->u.bdev.num_blocks, bdev_copy_do_write_done, bdev_io);
10576 :
10577 3 : if (rc == -ENOMEM) {
10578 0 : bdev_queue_io_wait_with_cb(bdev_io, bdev_copy_do_write);
10579 3 : } else if (rc != 0) {
10580 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
10581 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
10582 0 : }
10583 3 : }
10584 :
10585 : static void
10586 3 : bdev_copy_do_read_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
10587 : {
10588 3 : struct spdk_bdev_io *parent_io = cb_arg;
10589 :
10590 3 : spdk_bdev_free_io(bdev_io);
10591 :
10592 : /* Check return status of read */
10593 3 : if (!success) {
10594 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
10595 0 : parent_io->internal.cb(parent_io, false, parent_io->internal.caller_ctx);
10596 0 : return;
10597 : }
10598 :
10599 : /* Do write */
10600 3 : bdev_copy_do_write(parent_io);
10601 3 : }
10602 :
10603 : static void
10604 3 : bdev_copy_do_read(void *_bdev_io)
10605 : {
10606 3 : struct spdk_bdev_io *bdev_io = _bdev_io;
10607 : int rc;
10608 :
10609 : /* Read blocks */
10610 6 : rc = spdk_bdev_read_blocks_with_md(bdev_io->internal.desc,
10611 3 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
10612 3 : bdev_io->u.bdev.iovs[0].iov_base,
10613 3 : bdev_io->u.bdev.md_buf, bdev_io->u.bdev.copy.src_offset_blocks,
10614 3 : bdev_io->u.bdev.num_blocks, bdev_copy_do_read_done, bdev_io);
10615 :
10616 3 : if (rc == -ENOMEM) {
10617 0 : bdev_queue_io_wait_with_cb(bdev_io, bdev_copy_do_read);
10618 3 : } else if (rc != 0) {
10619 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
10620 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
10621 0 : }
10622 3 : }
10623 :
10624 : static void
10625 3 : bdev_copy_get_buf_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io, bool success)
10626 : {
10627 3 : if (!success) {
10628 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
10629 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
10630 0 : return;
10631 : }
10632 :
10633 3 : bdev_copy_do_read(bdev_io);
10634 3 : }
10635 :
10636 : int
10637 27 : spdk_bdev_copy_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
10638 : uint64_t dst_offset_blocks, uint64_t src_offset_blocks, uint64_t num_blocks,
10639 : spdk_bdev_io_completion_cb cb, void *cb_arg)
10640 : {
10641 27 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
10642 : struct spdk_bdev_io *bdev_io;
10643 27 : struct spdk_bdev_channel *channel = spdk_io_channel_get_ctx(ch);
10644 :
10645 27 : if (!desc->write) {
10646 0 : return -EBADF;
10647 : }
10648 :
10649 27 : if (!bdev_io_valid_blocks(bdev, dst_offset_blocks, num_blocks) ||
10650 27 : !bdev_io_valid_blocks(bdev, src_offset_blocks, num_blocks)) {
10651 0 : SPDK_DEBUGLOG(bdev,
10652 : "Invalid offset or number of blocks: dst %lu, src %lu, count %lu\n",
10653 : dst_offset_blocks, src_offset_blocks, num_blocks);
10654 0 : return -EINVAL;
10655 : }
10656 :
10657 27 : bdev_io = bdev_channel_get_io(channel);
10658 27 : if (!bdev_io) {
10659 0 : return -ENOMEM;
10660 : }
10661 :
10662 27 : bdev_io->internal.ch = channel;
10663 27 : bdev_io->internal.desc = desc;
10664 27 : bdev_io->type = SPDK_BDEV_IO_TYPE_COPY;
10665 :
10666 27 : bdev_io->u.bdev.offset_blocks = dst_offset_blocks;
10667 27 : bdev_io->u.bdev.copy.src_offset_blocks = src_offset_blocks;
10668 27 : bdev_io->u.bdev.num_blocks = num_blocks;
10669 27 : bdev_io->u.bdev.memory_domain = NULL;
10670 27 : bdev_io->u.bdev.memory_domain_ctx = NULL;
10671 27 : bdev_io->u.bdev.iovs = NULL;
10672 27 : bdev_io->u.bdev.iovcnt = 0;
10673 27 : bdev_io->u.bdev.md_buf = NULL;
10674 27 : bdev_io->u.bdev.accel_sequence = NULL;
10675 27 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
10676 :
10677 27 : if (dst_offset_blocks == src_offset_blocks || num_blocks == 0) {
10678 0 : spdk_thread_send_msg(spdk_get_thread(), bdev_io_complete_cb, bdev_io);
10679 0 : return 0;
10680 : }
10681 :
10682 :
10683 : /* If the copy size is large and should be split, use the generic split logic
10684 : * regardless of whether SPDK_BDEV_IO_TYPE_COPY is supported or not.
10685 : *
10686 : * Then, send the copy request if SPDK_BDEV_IO_TYPE_COPY is supported or
10687 : * emulate it using regular read and write requests otherwise.
10688 : */
10689 27 : if (spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COPY) ||
10690 4 : bdev_io->internal.f.split) {
10691 24 : bdev_io_submit(bdev_io);
10692 24 : return 0;
10693 : }
10694 :
10695 3 : spdk_bdev_io_get_buf(bdev_io, bdev_copy_get_buf_cb, num_blocks * spdk_bdev_get_block_size(bdev));
10696 :
10697 3 : return 0;
10698 27 : }
10699 :
10700 3 : SPDK_LOG_REGISTER_COMPONENT(bdev)
10701 :
10702 : static void
10703 0 : bdev_trace(void)
10704 : {
10705 0 : struct spdk_trace_tpoint_opts opts[] = {
10706 : {
10707 : "BDEV_IO_START", TRACE_BDEV_IO_START,
10708 : OWNER_TYPE_BDEV, OBJECT_BDEV_IO, 1,
10709 : {
10710 : { "type", SPDK_TRACE_ARG_TYPE_INT, 8 },
10711 : { "ctx", SPDK_TRACE_ARG_TYPE_PTR, 8 },
10712 : { "offset", SPDK_TRACE_ARG_TYPE_INT, 8 },
10713 : { "qd", SPDK_TRACE_ARG_TYPE_INT, 4 }
10714 : }
10715 : },
10716 : {
10717 : "BDEV_IO_DONE", TRACE_BDEV_IO_DONE,
10718 : OWNER_TYPE_BDEV, OBJECT_BDEV_IO, 0,
10719 : {
10720 : { "ctx", SPDK_TRACE_ARG_TYPE_PTR, 8 },
10721 : { "qd", SPDK_TRACE_ARG_TYPE_INT, 4 }
10722 : }
10723 : },
10724 : {
10725 : "BDEV_IOCH_CREATE", TRACE_BDEV_IOCH_CREATE,
10726 : OWNER_TYPE_BDEV, OBJECT_NONE, 0,
10727 : {
10728 : { "tid", SPDK_TRACE_ARG_TYPE_INT, 8 }
10729 : }
10730 : },
10731 : {
10732 : "BDEV_IOCH_DESTROY", TRACE_BDEV_IOCH_DESTROY,
10733 : OWNER_TYPE_BDEV, OBJECT_NONE, 0,
10734 : {
10735 : { "tid", SPDK_TRACE_ARG_TYPE_INT, 8 }
10736 : }
10737 : },
10738 : };
10739 :
10740 :
10741 0 : spdk_trace_register_owner_type(OWNER_TYPE_BDEV, 'b');
10742 0 : spdk_trace_register_object(OBJECT_BDEV_IO, 'i');
10743 0 : spdk_trace_register_description_ext(opts, SPDK_COUNTOF(opts));
10744 0 : spdk_trace_tpoint_register_relation(TRACE_BDEV_NVME_IO_START, OBJECT_BDEV_IO, 0);
10745 0 : spdk_trace_tpoint_register_relation(TRACE_BDEV_NVME_IO_DONE, OBJECT_BDEV_IO, 0);
10746 0 : spdk_trace_tpoint_register_relation(TRACE_BLOB_REQ_SET_START, OBJECT_BDEV_IO, 0);
10747 0 : spdk_trace_tpoint_register_relation(TRACE_BLOB_REQ_SET_COMPLETE, OBJECT_BDEV_IO, 0);
10748 0 : spdk_trace_tpoint_register_relation(TRACE_BDEV_RAID_IO_START, OBJECT_BDEV_IO, 0);
10749 0 : spdk_trace_tpoint_register_relation(TRACE_BDEV_RAID_IO_DONE, OBJECT_BDEV_IO, 0);
10750 0 : }
10751 3 : SPDK_TRACE_REGISTER_FN(bdev_trace, "bdev", TRACE_GROUP_BDEV)
|