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
86 547 : bdev_name_cmp(struct spdk_bdev_name *name1, struct spdk_bdev_name *name2)
87 : {
88 547 : 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 : struct spdk_thread *thread;
339 : struct {
340 : spdk_bdev_event_cb_t event_fn;
341 : void *ctx;
342 : } callback;
343 : bool closed;
344 : bool write;
345 : bool memory_domains_supported;
346 : bool accel_sequence_supported[SPDK_BDEV_NUM_IO_TYPES];
347 : struct spdk_spinlock spinlock;
348 : uint32_t refs;
349 : struct spdk_bdev_open_opts opts;
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 : BDEV_IO_RETRY_STATE_GET_ACCEL_BUF,
394 : };
395 :
396 : #define __bdev_to_io_dev(bdev) (((char *)bdev) + 1)
397 : #define __bdev_from_io_dev(io_dev) ((struct spdk_bdev *)(((char *)io_dev) - 1))
398 : #define __io_ch_to_bdev_ch(io_ch) ((struct spdk_bdev_channel *)spdk_io_channel_get_ctx(io_ch))
399 : #define __io_ch_to_bdev_mgmt_ch(io_ch) ((struct spdk_bdev_mgmt_channel *)spdk_io_channel_get_ctx(io_ch))
400 :
401 : static inline void bdev_io_complete(void *ctx);
402 : static inline void bdev_io_complete_unsubmitted(struct spdk_bdev_io *bdev_io);
403 : static void bdev_io_push_bounce_md_buf(struct spdk_bdev_io *bdev_io);
404 : static void bdev_io_push_bounce_data(struct spdk_bdev_io *bdev_io);
405 : static void _bdev_io_get_accel_buf(struct spdk_bdev_io *bdev_io);
406 :
407 : static void bdev_write_zero_buffer_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg);
408 : static int bdev_write_zero_buffer(struct spdk_bdev_io *bdev_io);
409 :
410 : static void bdev_enable_qos_msg(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
411 : struct spdk_io_channel *ch, void *_ctx);
412 : static void bdev_enable_qos_done(struct spdk_bdev *bdev, void *_ctx, int status);
413 :
414 : static int bdev_readv_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
415 : struct iovec *iov, int iovcnt, void *md_buf, uint64_t offset_blocks,
416 : uint64_t num_blocks,
417 : struct spdk_memory_domain *domain, void *domain_ctx,
418 : struct spdk_accel_sequence *seq, uint32_t dif_check_flags,
419 : spdk_bdev_io_completion_cb cb, void *cb_arg);
420 : static int bdev_writev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
421 : struct iovec *iov, int iovcnt, void *md_buf,
422 : uint64_t offset_blocks, uint64_t num_blocks,
423 : struct spdk_memory_domain *domain, void *domain_ctx,
424 : struct spdk_accel_sequence *seq, uint32_t dif_check_flags,
425 : uint32_t nvme_cdw12_raw, uint32_t nvme_cdw13_raw,
426 : spdk_bdev_io_completion_cb cb, void *cb_arg);
427 :
428 : static int bdev_lock_lba_range(struct spdk_bdev_desc *desc, struct spdk_io_channel *_ch,
429 : uint64_t offset, uint64_t length,
430 : lock_range_cb cb_fn, void *cb_arg);
431 :
432 : static int bdev_unlock_lba_range(struct spdk_bdev_desc *desc, struct spdk_io_channel *_ch,
433 : uint64_t offset, uint64_t length,
434 : lock_range_cb cb_fn, void *cb_arg);
435 :
436 : static bool bdev_abort_queued_io(bdev_io_tailq_t *queue, struct spdk_bdev_io *bio_to_abort);
437 : static bool bdev_abort_buf_io(struct spdk_bdev_mgmt_channel *ch, struct spdk_bdev_io *bio_to_abort);
438 :
439 : static bool claim_type_is_v2(enum spdk_bdev_claim_type type);
440 : static void bdev_desc_release_claims(struct spdk_bdev_desc *desc);
441 : static void claim_reset(struct spdk_bdev *bdev);
442 :
443 : static void bdev_ch_retry_io(struct spdk_bdev_channel *bdev_ch);
444 :
445 : static bool bdev_io_should_split(struct spdk_bdev_io *bdev_io);
446 :
447 : #define bdev_get_ext_io_opt(opts, field, defval) \
448 : ((opts) != NULL ? SPDK_GET_FIELD(opts, field, defval) : (defval))
449 :
450 : static inline void
451 669 : bdev_ch_add_to_io_submitted(struct spdk_bdev_io *bdev_io)
452 : {
453 669 : TAILQ_INSERT_TAIL(&bdev_io->internal.ch->io_submitted, bdev_io, internal.ch_link);
454 669 : bdev_io->internal.ch->queue_depth++;
455 669 : }
456 :
457 : static inline void
458 669 : bdev_ch_remove_from_io_submitted(struct spdk_bdev_io *bdev_io)
459 : {
460 669 : TAILQ_REMOVE(&bdev_io->internal.ch->io_submitted, bdev_io, internal.ch_link);
461 669 : bdev_io->internal.ch->queue_depth--;
462 669 : }
463 :
464 : void
465 14 : spdk_bdev_get_opts(struct spdk_bdev_opts *opts, size_t opts_size)
466 : {
467 14 : if (!opts) {
468 0 : SPDK_ERRLOG("opts should not be NULL\n");
469 0 : return;
470 : }
471 :
472 14 : if (!opts_size) {
473 0 : SPDK_ERRLOG("opts_size should not be zero value\n");
474 0 : return;
475 : }
476 :
477 14 : opts->opts_size = opts_size;
478 :
479 : #define SET_FIELD(field) \
480 : if (offsetof(struct spdk_bdev_opts, field) + sizeof(opts->field) <= opts_size) { \
481 : opts->field = g_bdev_opts.field; \
482 : } \
483 :
484 14 : SET_FIELD(bdev_io_pool_size);
485 14 : SET_FIELD(bdev_io_cache_size);
486 14 : SET_FIELD(bdev_auto_examine);
487 14 : SET_FIELD(iobuf_small_cache_size);
488 14 : SET_FIELD(iobuf_large_cache_size);
489 :
490 : /* Do not remove this statement, you should always update this statement when you adding a new field,
491 : * and do not forget to add the SET_FIELD statement for your added field. */
492 : SPDK_STATIC_ASSERT(sizeof(struct spdk_bdev_opts) == 32, "Incorrect size");
493 :
494 : #undef SET_FIELD
495 14 : }
496 :
497 : int
498 15 : spdk_bdev_set_opts(struct spdk_bdev_opts *opts)
499 : {
500 : uint32_t min_pool_size;
501 :
502 15 : if (!opts) {
503 0 : SPDK_ERRLOG("opts cannot be NULL\n");
504 0 : return -1;
505 : }
506 :
507 15 : if (!opts->opts_size) {
508 1 : SPDK_ERRLOG("opts_size inside opts cannot be zero value\n");
509 1 : return -1;
510 : }
511 :
512 : /*
513 : * Add 1 to the thread count to account for the extra mgmt_ch that gets created during subsystem
514 : * initialization. A second mgmt_ch will be created on the same thread when the application starts
515 : * but before the deferred put_io_channel event is executed for the first mgmt_ch.
516 : */
517 14 : min_pool_size = opts->bdev_io_cache_size * (spdk_thread_get_count() + 1);
518 14 : if (opts->bdev_io_pool_size < min_pool_size) {
519 0 : SPDK_ERRLOG("bdev_io_pool_size %" PRIu32 " is not compatible with bdev_io_cache_size %" PRIu32
520 : " and %" PRIu32 " threads\n", opts->bdev_io_pool_size, opts->bdev_io_cache_size,
521 : spdk_thread_get_count());
522 0 : SPDK_ERRLOG("bdev_io_pool_size must be at least %" PRIu32 "\n", min_pool_size);
523 0 : return -1;
524 : }
525 :
526 : #define SET_FIELD(field) \
527 : if (offsetof(struct spdk_bdev_opts, field) + sizeof(opts->field) <= opts->opts_size) { \
528 : g_bdev_opts.field = opts->field; \
529 : } \
530 :
531 14 : SET_FIELD(bdev_io_pool_size);
532 14 : SET_FIELD(bdev_io_cache_size);
533 14 : SET_FIELD(bdev_auto_examine);
534 14 : SET_FIELD(iobuf_small_cache_size);
535 14 : SET_FIELD(iobuf_large_cache_size);
536 :
537 14 : g_bdev_opts.opts_size = opts->opts_size;
538 :
539 : #undef SET_FIELD
540 :
541 14 : return 0;
542 15 : }
543 :
544 : static struct spdk_bdev *
545 146 : bdev_get_by_name(const char *bdev_name)
546 : {
547 : struct spdk_bdev_name find;
548 : struct spdk_bdev_name *res;
549 :
550 146 : find.name = (char *)bdev_name;
551 146 : res = RB_FIND(bdev_name_tree, &g_bdev_mgr.bdev_names, &find);
552 146 : if (res != NULL) {
553 139 : return res->bdev;
554 : }
555 :
556 7 : return NULL;
557 146 : }
558 :
559 : struct spdk_bdev *
560 16 : spdk_bdev_get_by_name(const char *bdev_name)
561 : {
562 : struct spdk_bdev *bdev;
563 :
564 16 : spdk_spin_lock(&g_bdev_mgr.spinlock);
565 16 : bdev = bdev_get_by_name(bdev_name);
566 16 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
567 :
568 16 : return bdev;
569 : }
570 :
571 : struct bdev_io_status_string {
572 : enum spdk_bdev_io_status status;
573 : const char *str;
574 : };
575 :
576 : static const struct bdev_io_status_string bdev_io_status_strings[] = {
577 : { SPDK_BDEV_IO_STATUS_AIO_ERROR, "aio_error" },
578 : { SPDK_BDEV_IO_STATUS_ABORTED, "aborted" },
579 : { SPDK_BDEV_IO_STATUS_FIRST_FUSED_FAILED, "first_fused_failed" },
580 : { SPDK_BDEV_IO_STATUS_MISCOMPARE, "miscompare" },
581 : { SPDK_BDEV_IO_STATUS_NOMEM, "nomem" },
582 : { SPDK_BDEV_IO_STATUS_SCSI_ERROR, "scsi_error" },
583 : { SPDK_BDEV_IO_STATUS_NVME_ERROR, "nvme_error" },
584 : { SPDK_BDEV_IO_STATUS_FAILED, "failed" },
585 : { SPDK_BDEV_IO_STATUS_PENDING, "pending" },
586 : { SPDK_BDEV_IO_STATUS_SUCCESS, "success" },
587 : };
588 :
589 : static const char *
590 0 : bdev_io_status_get_string(enum spdk_bdev_io_status status)
591 : {
592 : uint32_t i;
593 :
594 0 : for (i = 0; i < SPDK_COUNTOF(bdev_io_status_strings); i++) {
595 0 : if (bdev_io_status_strings[i].status == status) {
596 0 : return bdev_io_status_strings[i].str;
597 : }
598 0 : }
599 :
600 0 : return "reserved";
601 0 : }
602 :
603 : struct spdk_bdev_wait_for_examine_ctx {
604 : struct spdk_poller *poller;
605 : spdk_bdev_wait_for_examine_cb cb_fn;
606 : void *cb_arg;
607 : };
608 :
609 : static bool bdev_module_all_actions_completed(void);
610 :
611 : static int
612 193 : bdev_wait_for_examine_cb(void *arg)
613 : {
614 193 : struct spdk_bdev_wait_for_examine_ctx *ctx = arg;
615 :
616 193 : if (!bdev_module_all_actions_completed()) {
617 0 : return SPDK_POLLER_IDLE;
618 : }
619 :
620 193 : spdk_poller_unregister(&ctx->poller);
621 193 : ctx->cb_fn(ctx->cb_arg);
622 193 : free(ctx);
623 :
624 193 : return SPDK_POLLER_BUSY;
625 193 : }
626 :
627 : int
628 193 : spdk_bdev_wait_for_examine(spdk_bdev_wait_for_examine_cb cb_fn, void *cb_arg)
629 : {
630 : struct spdk_bdev_wait_for_examine_ctx *ctx;
631 :
632 193 : ctx = calloc(1, sizeof(*ctx));
633 193 : if (ctx == NULL) {
634 0 : return -ENOMEM;
635 : }
636 193 : ctx->cb_fn = cb_fn;
637 193 : ctx->cb_arg = cb_arg;
638 193 : ctx->poller = SPDK_POLLER_REGISTER(bdev_wait_for_examine_cb, ctx, 0);
639 :
640 193 : return 0;
641 193 : }
642 :
643 : struct spdk_bdev_examine_item {
644 : char *name;
645 : TAILQ_ENTRY(spdk_bdev_examine_item) link;
646 : };
647 :
648 : TAILQ_HEAD(spdk_bdev_examine_allowlist, spdk_bdev_examine_item);
649 :
650 : struct spdk_bdev_examine_allowlist g_bdev_examine_allowlist = TAILQ_HEAD_INITIALIZER(
651 : g_bdev_examine_allowlist);
652 :
653 : static inline bool
654 6 : bdev_examine_allowlist_check(const char *name)
655 : {
656 : struct spdk_bdev_examine_item *item;
657 6 : TAILQ_FOREACH(item, &g_bdev_examine_allowlist, link) {
658 0 : if (strcmp(name, item->name) == 0) {
659 0 : return true;
660 : }
661 0 : }
662 6 : return false;
663 6 : }
664 :
665 : static inline void
666 65 : bdev_examine_allowlist_free(void)
667 : {
668 : struct spdk_bdev_examine_item *item;
669 65 : while (!TAILQ_EMPTY(&g_bdev_examine_allowlist)) {
670 0 : item = TAILQ_FIRST(&g_bdev_examine_allowlist);
671 0 : TAILQ_REMOVE(&g_bdev_examine_allowlist, item, link);
672 0 : free(item->name);
673 0 : free(item);
674 : }
675 65 : }
676 :
677 : static inline bool
678 3 : bdev_in_examine_allowlist(struct spdk_bdev *bdev)
679 : {
680 : struct spdk_bdev_alias *tmp;
681 3 : if (bdev_examine_allowlist_check(bdev->name)) {
682 0 : return true;
683 : }
684 6 : TAILQ_FOREACH(tmp, &bdev->aliases, tailq) {
685 3 : if (bdev_examine_allowlist_check(tmp->alias.name)) {
686 0 : return true;
687 : }
688 3 : }
689 3 : return false;
690 3 : }
691 :
692 : static inline bool
693 126 : bdev_ok_to_examine(struct spdk_bdev *bdev)
694 : {
695 : /* Some bdevs may not support the READ command.
696 : * Do not try to examine them.
697 : */
698 126 : if (!spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_READ)) {
699 0 : return false;
700 : }
701 :
702 126 : if (g_bdev_opts.bdev_auto_examine) {
703 123 : return true;
704 : } else {
705 3 : return bdev_in_examine_allowlist(bdev);
706 : }
707 126 : }
708 :
709 : static void
710 126 : bdev_examine(struct spdk_bdev *bdev)
711 : {
712 : struct spdk_bdev_module *module;
713 : struct spdk_bdev_module_claim *claim, *tmpclaim;
714 : uint32_t action;
715 :
716 126 : if (!bdev_ok_to_examine(bdev)) {
717 3 : return;
718 : }
719 :
720 504 : TAILQ_FOREACH(module, &g_bdev_mgr.bdev_modules, internal.tailq) {
721 381 : if (module->examine_config) {
722 258 : spdk_spin_lock(&module->internal.spinlock);
723 258 : action = module->internal.action_in_progress;
724 258 : module->internal.action_in_progress++;
725 258 : spdk_spin_unlock(&module->internal.spinlock);
726 258 : module->examine_config(bdev);
727 258 : if (action != module->internal.action_in_progress) {
728 0 : SPDK_ERRLOG("examine_config for module %s did not call "
729 : "spdk_bdev_module_examine_done()\n", module->name);
730 0 : }
731 258 : }
732 381 : }
733 :
734 123 : spdk_spin_lock(&bdev->internal.spinlock);
735 :
736 123 : switch (bdev->internal.claim_type) {
737 : case SPDK_BDEV_CLAIM_NONE:
738 : /* Examine by all bdev modules */
739 479 : TAILQ_FOREACH(module, &g_bdev_mgr.bdev_modules, internal.tailq) {
740 361 : if (module->examine_disk) {
741 234 : spdk_spin_lock(&module->internal.spinlock);
742 234 : module->internal.action_in_progress++;
743 234 : spdk_spin_unlock(&module->internal.spinlock);
744 234 : spdk_spin_unlock(&bdev->internal.spinlock);
745 234 : module->examine_disk(bdev);
746 234 : spdk_spin_lock(&bdev->internal.spinlock);
747 234 : }
748 361 : }
749 118 : break;
750 : case SPDK_BDEV_CLAIM_EXCL_WRITE:
751 : /* Examine by the one bdev module with a v1 claim */
752 1 : module = bdev->internal.claim.v1.module;
753 1 : if (module->examine_disk) {
754 1 : spdk_spin_lock(&module->internal.spinlock);
755 1 : module->internal.action_in_progress++;
756 1 : spdk_spin_unlock(&module->internal.spinlock);
757 1 : spdk_spin_unlock(&bdev->internal.spinlock);
758 1 : module->examine_disk(bdev);
759 1 : return;
760 : }
761 0 : break;
762 : default:
763 : /* Examine by all bdev modules with a v2 claim */
764 4 : assert(claim_type_is_v2(bdev->internal.claim_type));
765 : /*
766 : * Removal of tailq nodes while iterating can cause the iteration to jump out of the
767 : * list, perhaps accessing freed memory. Without protection, this could happen
768 : * while the lock is dropped during the examine callback.
769 : */
770 4 : bdev->internal.examine_in_progress++;
771 :
772 9 : TAILQ_FOREACH(claim, &bdev->internal.claim.v2.claims, link) {
773 5 : module = claim->module;
774 :
775 5 : if (module == NULL) {
776 : /* This is a vestigial claim, held by examine_count */
777 0 : continue;
778 : }
779 :
780 5 : if (module->examine_disk == NULL) {
781 0 : continue;
782 : }
783 :
784 5 : spdk_spin_lock(&module->internal.spinlock);
785 5 : module->internal.action_in_progress++;
786 5 : spdk_spin_unlock(&module->internal.spinlock);
787 :
788 : /* Call examine_disk without holding internal.spinlock. */
789 5 : spdk_spin_unlock(&bdev->internal.spinlock);
790 5 : module->examine_disk(bdev);
791 5 : spdk_spin_lock(&bdev->internal.spinlock);
792 5 : }
793 :
794 4 : assert(bdev->internal.examine_in_progress > 0);
795 4 : bdev->internal.examine_in_progress--;
796 4 : if (bdev->internal.examine_in_progress == 0) {
797 : /* Remove any claims that were released during examine_disk */
798 9 : TAILQ_FOREACH_SAFE(claim, &bdev->internal.claim.v2.claims, link, tmpclaim) {
799 5 : if (claim->desc != NULL) {
800 5 : continue;
801 : }
802 :
803 0 : TAILQ_REMOVE(&bdev->internal.claim.v2.claims, claim, link);
804 0 : free(claim);
805 0 : }
806 4 : if (TAILQ_EMPTY(&bdev->internal.claim.v2.claims)) {
807 0 : claim_reset(bdev);
808 0 : }
809 4 : }
810 4 : }
811 :
812 122 : spdk_spin_unlock(&bdev->internal.spinlock);
813 126 : }
814 :
815 : int
816 1 : spdk_bdev_examine(const char *name)
817 : {
818 : struct spdk_bdev *bdev;
819 : struct spdk_bdev_examine_item *item;
820 1 : struct spdk_thread *thread = spdk_get_thread();
821 :
822 1 : if (spdk_unlikely(!spdk_thread_is_app_thread(thread))) {
823 1 : SPDK_ERRLOG("Cannot examine bdev %s on thread %p (%s)\n", name, thread,
824 : thread ? spdk_thread_get_name(thread) : "null");
825 1 : return -EINVAL;
826 : }
827 :
828 0 : if (g_bdev_opts.bdev_auto_examine) {
829 0 : SPDK_ERRLOG("Manual examine is not allowed if auto examine is enabled\n");
830 0 : return -EINVAL;
831 : }
832 :
833 0 : if (bdev_examine_allowlist_check(name)) {
834 0 : SPDK_ERRLOG("Duplicate bdev name for manual examine: %s\n", name);
835 0 : return -EEXIST;
836 : }
837 :
838 0 : item = calloc(1, sizeof(*item));
839 0 : if (!item) {
840 0 : return -ENOMEM;
841 : }
842 0 : item->name = strdup(name);
843 0 : if (!item->name) {
844 0 : free(item);
845 0 : return -ENOMEM;
846 : }
847 0 : TAILQ_INSERT_TAIL(&g_bdev_examine_allowlist, item, link);
848 :
849 0 : bdev = spdk_bdev_get_by_name(name);
850 0 : if (bdev) {
851 0 : bdev_examine(bdev);
852 0 : }
853 0 : return 0;
854 1 : }
855 :
856 : static inline void
857 0 : bdev_examine_allowlist_config_json(struct spdk_json_write_ctx *w)
858 : {
859 : struct spdk_bdev_examine_item *item;
860 0 : TAILQ_FOREACH(item, &g_bdev_examine_allowlist, link) {
861 0 : spdk_json_write_object_begin(w);
862 0 : spdk_json_write_named_string(w, "method", "bdev_examine");
863 0 : spdk_json_write_named_object_begin(w, "params");
864 0 : spdk_json_write_named_string(w, "name", item->name);
865 0 : spdk_json_write_object_end(w);
866 0 : spdk_json_write_object_end(w);
867 0 : }
868 0 : }
869 :
870 : struct spdk_bdev *
871 1 : spdk_bdev_first(void)
872 : {
873 : struct spdk_bdev *bdev;
874 :
875 1 : bdev = TAILQ_FIRST(&g_bdev_mgr.bdevs);
876 1 : if (bdev) {
877 1 : SPDK_DEBUGLOG(bdev, "Starting bdev iteration at %s\n", bdev->name);
878 1 : }
879 :
880 1 : return bdev;
881 : }
882 :
883 : struct spdk_bdev *
884 8 : spdk_bdev_next(struct spdk_bdev *prev)
885 : {
886 : struct spdk_bdev *bdev;
887 :
888 8 : bdev = TAILQ_NEXT(prev, internal.link);
889 8 : if (bdev) {
890 7 : SPDK_DEBUGLOG(bdev, "Continuing bdev iteration at %s\n", bdev->name);
891 7 : }
892 :
893 8 : return bdev;
894 : }
895 :
896 : static struct spdk_bdev *
897 6 : _bdev_next_leaf(struct spdk_bdev *bdev)
898 : {
899 9 : while (bdev != NULL) {
900 8 : if (bdev->internal.claim_type == SPDK_BDEV_CLAIM_NONE) {
901 5 : return bdev;
902 : } else {
903 3 : bdev = TAILQ_NEXT(bdev, internal.link);
904 : }
905 : }
906 :
907 1 : return bdev;
908 6 : }
909 :
910 : struct spdk_bdev *
911 1 : spdk_bdev_first_leaf(void)
912 : {
913 : struct spdk_bdev *bdev;
914 :
915 1 : bdev = _bdev_next_leaf(TAILQ_FIRST(&g_bdev_mgr.bdevs));
916 :
917 1 : if (bdev) {
918 1 : SPDK_DEBUGLOG(bdev, "Starting bdev iteration at %s\n", bdev->name);
919 1 : }
920 :
921 1 : return bdev;
922 : }
923 :
924 : struct spdk_bdev *
925 5 : spdk_bdev_next_leaf(struct spdk_bdev *prev)
926 : {
927 : struct spdk_bdev *bdev;
928 :
929 5 : bdev = _bdev_next_leaf(TAILQ_NEXT(prev, internal.link));
930 :
931 5 : if (bdev) {
932 4 : SPDK_DEBUGLOG(bdev, "Continuing bdev iteration at %s\n", bdev->name);
933 4 : }
934 :
935 5 : return bdev;
936 : }
937 :
938 : static inline bool
939 820 : bdev_io_use_memory_domain(struct spdk_bdev_io *bdev_io)
940 : {
941 820 : return bdev_io->internal.f.has_memory_domain;
942 : }
943 :
944 : static inline bool
945 1549 : bdev_io_use_accel_sequence(struct spdk_bdev_io *bdev_io)
946 : {
947 1549 : return bdev_io->internal.f.has_accel_sequence;
948 : }
949 :
950 : static inline uint32_t
951 373 : bdev_desc_get_block_size(struct spdk_bdev_desc *desc)
952 : {
953 373 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
954 :
955 373 : if (spdk_unlikely(desc->opts.no_metadata)) {
956 0 : return bdev->blocklen - bdev->md_len;
957 : } else {
958 373 : return bdev->blocklen;
959 : }
960 373 : }
961 :
962 : static inline uint32_t
963 110 : bdev_io_get_block_size(struct spdk_bdev_io *bdev_io)
964 : {
965 110 : struct spdk_bdev *bdev = bdev_io->bdev;
966 :
967 110 : if (bdev_io->u.bdev.dif_check_flags & SPDK_DIF_FLAGS_NVME_PRACT) {
968 0 : if (bdev->md_len == spdk_dif_pi_format_get_pi_size(bdev->dif_pi_format)) {
969 0 : return bdev->blocklen - bdev->md_len;
970 : } else {
971 0 : return bdev->blocklen;
972 : }
973 : }
974 :
975 110 : return bdev_desc_get_block_size(bdev_io->internal.desc);
976 110 : }
977 :
978 : static inline void
979 7 : bdev_queue_nomem_io_head(struct spdk_bdev_shared_resource *shared_resource,
980 : struct spdk_bdev_io *bdev_io, enum bdev_io_retry_state state)
981 : {
982 : /* Wait for some of the outstanding I/O to complete before we retry any of the nomem_io.
983 : * Normally we will wait for NOMEM_THRESHOLD_COUNT I/O to complete but for low queue depth
984 : * channels we will instead wait for half to complete.
985 : */
986 7 : shared_resource->nomem_threshold = spdk_max((int64_t)shared_resource->io_outstanding / 2,
987 : (int64_t)shared_resource->io_outstanding - NOMEM_THRESHOLD_COUNT);
988 :
989 7 : assert(state != BDEV_IO_RETRY_STATE_INVALID);
990 7 : bdev_io->internal.retry_state = state;
991 7 : TAILQ_INSERT_HEAD(&shared_resource->nomem_io, bdev_io, internal.link);
992 7 : }
993 :
994 : static inline void
995 43 : bdev_queue_nomem_io_tail(struct spdk_bdev_shared_resource *shared_resource,
996 : struct spdk_bdev_io *bdev_io, enum bdev_io_retry_state state)
997 : {
998 : /* We only queue IOs at the end of the nomem_io queue if they're submitted by the user while
999 : * the queue isn't empty, so we don't need to update the nomem_threshold here */
1000 43 : assert(!TAILQ_EMPTY(&shared_resource->nomem_io));
1001 :
1002 43 : assert(state != BDEV_IO_RETRY_STATE_INVALID);
1003 43 : bdev_io->internal.retry_state = state;
1004 43 : TAILQ_INSERT_TAIL(&shared_resource->nomem_io, bdev_io, internal.link);
1005 43 : }
1006 :
1007 : void
1008 16 : spdk_bdev_io_set_buf(struct spdk_bdev_io *bdev_io, void *buf, size_t len)
1009 : {
1010 : struct iovec *iovs;
1011 :
1012 16 : if (bdev_io->u.bdev.iovs == NULL) {
1013 3 : bdev_io->u.bdev.iovs = &bdev_io->iov;
1014 3 : bdev_io->u.bdev.iovcnt = 1;
1015 3 : }
1016 :
1017 16 : iovs = bdev_io->u.bdev.iovs;
1018 :
1019 16 : assert(iovs != NULL);
1020 16 : assert(bdev_io->u.bdev.iovcnt >= 1);
1021 :
1022 16 : iovs[0].iov_base = buf;
1023 16 : iovs[0].iov_len = len;
1024 16 : }
1025 :
1026 : void
1027 3 : spdk_bdev_io_set_md_buf(struct spdk_bdev_io *bdev_io, void *md_buf, size_t len)
1028 : {
1029 3 : assert((len / spdk_bdev_get_md_size(bdev_io->bdev)) >= bdev_io->u.bdev.num_blocks);
1030 3 : bdev_io->u.bdev.md_buf = md_buf;
1031 3 : }
1032 :
1033 : static bool
1034 167 : _is_buf_allocated(const struct iovec *iovs)
1035 : {
1036 167 : if (iovs == NULL) {
1037 6 : return false;
1038 : }
1039 :
1040 161 : return iovs[0].iov_base != NULL;
1041 167 : }
1042 :
1043 : static bool
1044 50 : _are_iovs_aligned(struct iovec *iovs, int iovcnt, uint32_t alignment)
1045 : {
1046 : int i;
1047 : uintptr_t iov_base;
1048 :
1049 50 : if (spdk_likely(alignment == 1)) {
1050 21 : return true;
1051 : }
1052 :
1053 36 : for (i = 0; i < iovcnt; i++) {
1054 29 : iov_base = (uintptr_t)iovs[i].iov_base;
1055 29 : if ((iov_base & (alignment - 1)) != 0) {
1056 22 : return false;
1057 : }
1058 7 : }
1059 :
1060 7 : return true;
1061 50 : }
1062 :
1063 : static inline bool
1064 895 : bdev_io_needs_metadata(struct spdk_bdev_desc *desc, struct spdk_bdev_io *bdev_io)
1065 : {
1066 1048 : return (bdev_io->bdev->md_len != 0) &&
1067 153 : (desc->opts.no_metadata ||
1068 153 : (bdev_io->u.bdev.dif_check_flags & SPDK_DIF_FLAGS_NVME_PRACT));
1069 : }
1070 :
1071 : static inline bool
1072 852 : bdev_io_needs_sequence_exec(struct spdk_bdev_desc *desc, struct spdk_bdev_io *bdev_io)
1073 : {
1074 852 : if (!bdev_io_use_accel_sequence(bdev_io)) {
1075 852 : return false;
1076 : }
1077 :
1078 : /* For now, we don't allow splitting IOs with an accel sequence and will treat them as if
1079 : * bdev module didn't support accel sequences */
1080 0 : return !desc->accel_sequence_supported[bdev_io->type] || bdev_io->internal.f.split;
1081 852 : }
1082 :
1083 : static inline void
1084 592 : bdev_io_increment_outstanding(struct spdk_bdev_channel *bdev_ch,
1085 : struct spdk_bdev_shared_resource *shared_resource)
1086 : {
1087 592 : bdev_ch->io_outstanding++;
1088 592 : shared_resource->io_outstanding++;
1089 592 : }
1090 :
1091 : static inline void
1092 592 : bdev_io_decrement_outstanding(struct spdk_bdev_channel *bdev_ch,
1093 : struct spdk_bdev_shared_resource *shared_resource)
1094 : {
1095 592 : assert(bdev_ch->io_outstanding > 0);
1096 592 : assert(shared_resource->io_outstanding > 0);
1097 592 : bdev_ch->io_outstanding--;
1098 592 : shared_resource->io_outstanding--;
1099 592 : }
1100 :
1101 : static void
1102 0 : bdev_io_submit_sequence_cb(void *ctx, int status)
1103 : {
1104 0 : struct spdk_bdev_io *bdev_io = ctx;
1105 :
1106 0 : assert(bdev_io_use_accel_sequence(bdev_io));
1107 :
1108 0 : bdev_io->u.bdev.accel_sequence = NULL;
1109 0 : bdev_io->internal.f.has_accel_sequence = false;
1110 :
1111 0 : if (spdk_unlikely(status != 0)) {
1112 0 : SPDK_ERRLOG("Failed to execute accel sequence, status=%d\n", status);
1113 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
1114 0 : bdev_io_complete_unsubmitted(bdev_io);
1115 0 : return;
1116 : }
1117 :
1118 0 : bdev_io_submit(bdev_io);
1119 0 : }
1120 :
1121 : static void
1122 0 : bdev_io_exec_sequence_cb(void *ctx, int status)
1123 : {
1124 0 : struct spdk_bdev_io *bdev_io = ctx;
1125 0 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1126 :
1127 0 : TAILQ_REMOVE(&bdev_io->internal.ch->io_accel_exec, bdev_io, internal.link);
1128 0 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1129 :
1130 0 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1131 0 : bdev_ch_retry_io(ch);
1132 0 : }
1133 :
1134 0 : bdev_io->internal.data_transfer_cpl(bdev_io, status);
1135 0 : }
1136 :
1137 : static void
1138 0 : bdev_io_exec_sequence(struct spdk_bdev_io *bdev_io, void (*cb_fn)(void *ctx, int status))
1139 : {
1140 0 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1141 :
1142 0 : assert(bdev_io_needs_sequence_exec(bdev_io->internal.desc, bdev_io));
1143 0 : assert(bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE || bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
1144 0 : assert(bdev_io_use_accel_sequence(bdev_io));
1145 :
1146 : /* Since the operations are appended during submission, they're in the opposite order than
1147 : * how we want to execute them for reads (i.e. we need to execute the most recently added
1148 : * operation first), so reverse the sequence before executing it.
1149 : */
1150 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_READ) {
1151 0 : spdk_accel_sequence_reverse(bdev_io->internal.accel_sequence);
1152 0 : }
1153 :
1154 0 : TAILQ_INSERT_TAIL(&bdev_io->internal.ch->io_accel_exec, bdev_io, internal.link);
1155 0 : bdev_io_increment_outstanding(ch, ch->shared_resource);
1156 0 : bdev_io->internal.data_transfer_cpl = cb_fn;
1157 :
1158 0 : spdk_accel_sequence_finish(bdev_io->internal.accel_sequence,
1159 0 : bdev_io_exec_sequence_cb, bdev_io);
1160 0 : }
1161 :
1162 : static void
1163 42 : bdev_io_get_buf_complete(struct spdk_bdev_io *bdev_io, bool status)
1164 : {
1165 42 : struct spdk_io_channel *ch = spdk_bdev_io_get_io_channel(bdev_io);
1166 : void *buf;
1167 :
1168 42 : if (spdk_unlikely(bdev_io->internal.get_aux_buf_cb != NULL)) {
1169 0 : buf = bdev_io->internal.buf.ptr;
1170 0 : bdev_io->internal.buf.ptr = NULL;
1171 0 : bdev_io->internal.f.has_buf = false;
1172 0 : bdev_io->internal.get_aux_buf_cb(ch, bdev_io, buf);
1173 0 : bdev_io->internal.get_aux_buf_cb = NULL;
1174 0 : } else {
1175 42 : assert(bdev_io->internal.get_buf_cb != NULL);
1176 42 : bdev_io->internal.get_buf_cb(ch, bdev_io, status);
1177 42 : bdev_io->internal.get_buf_cb = NULL;
1178 : }
1179 42 : }
1180 :
1181 : static void
1182 4 : _bdev_io_pull_buffer_cpl(void *ctx, int rc)
1183 : {
1184 4 : struct spdk_bdev_io *bdev_io = ctx;
1185 :
1186 4 : if (rc) {
1187 0 : SPDK_ERRLOG("Set bounce buffer failed with rc %d\n", rc);
1188 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
1189 0 : }
1190 4 : bdev_io_get_buf_complete(bdev_io, !rc);
1191 4 : }
1192 :
1193 : static void
1194 2 : bdev_io_pull_md_buf_done(void *ctx, int status)
1195 : {
1196 2 : struct spdk_bdev_io *bdev_io = ctx;
1197 2 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1198 :
1199 2 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1200 2 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1201 :
1202 2 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1203 0 : bdev_ch_retry_io(ch);
1204 0 : }
1205 :
1206 2 : assert(bdev_io->internal.data_transfer_cpl);
1207 2 : bdev_io->internal.data_transfer_cpl(bdev_io, status);
1208 2 : }
1209 :
1210 : static void
1211 4 : bdev_io_pull_md_buf(struct spdk_bdev_io *bdev_io)
1212 : {
1213 4 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1214 4 : int rc = 0;
1215 :
1216 4 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
1217 2 : assert(bdev_io->internal.f.has_bounce_buf);
1218 2 : if (bdev_io_use_memory_domain(bdev_io)) {
1219 2 : TAILQ_INSERT_TAIL(&ch->io_memory_domain, bdev_io, internal.link);
1220 2 : bdev_io_increment_outstanding(ch, ch->shared_resource);
1221 4 : rc = spdk_memory_domain_pull_data(bdev_io->internal.memory_domain,
1222 2 : bdev_io->internal.memory_domain_ctx,
1223 2 : &bdev_io->internal.bounce_buf.orig_md_iov, 1,
1224 2 : &bdev_io->internal.bounce_buf.md_iov, 1,
1225 2 : bdev_io_pull_md_buf_done, bdev_io);
1226 2 : if (rc == 0) {
1227 : /* Continue to submit IO in completion callback */
1228 2 : return;
1229 : }
1230 0 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1231 0 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1232 0 : if (rc != -ENOMEM) {
1233 0 : SPDK_ERRLOG("Failed to pull data from memory domain %s, rc %d\n",
1234 : spdk_memory_domain_get_dma_device_id(
1235 : bdev_io->internal.memory_domain), rc);
1236 0 : }
1237 0 : } else {
1238 0 : memcpy(bdev_io->internal.bounce_buf.md_iov.iov_base,
1239 0 : bdev_io->internal.bounce_buf.orig_md_iov.iov_base,
1240 0 : bdev_io->internal.bounce_buf.orig_md_iov.iov_len);
1241 : }
1242 0 : }
1243 :
1244 2 : if (spdk_unlikely(rc == -ENOMEM)) {
1245 0 : bdev_queue_nomem_io_head(ch->shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PULL_MD);
1246 0 : } else {
1247 2 : assert(bdev_io->internal.data_transfer_cpl);
1248 2 : bdev_io->internal.data_transfer_cpl(bdev_io, rc);
1249 : }
1250 4 : }
1251 :
1252 : static void
1253 4 : _bdev_io_pull_bounce_md_buf(struct spdk_bdev_io *bdev_io, void *md_buf, size_t len)
1254 : {
1255 4 : assert(bdev_io->internal.f.has_bounce_buf);
1256 :
1257 : /* save original md_buf */
1258 4 : bdev_io->internal.bounce_buf.orig_md_iov.iov_base = bdev_io->u.bdev.md_buf;
1259 4 : bdev_io->internal.bounce_buf.orig_md_iov.iov_len = len;
1260 4 : bdev_io->internal.bounce_buf.md_iov.iov_base = md_buf;
1261 4 : bdev_io->internal.bounce_buf.md_iov.iov_len = len;
1262 : /* set bounce md_buf */
1263 4 : bdev_io->u.bdev.md_buf = md_buf;
1264 :
1265 4 : bdev_io_pull_md_buf(bdev_io);
1266 4 : }
1267 :
1268 : static void
1269 42 : _bdev_io_set_md_buf(struct spdk_bdev_io *bdev_io)
1270 : {
1271 42 : struct spdk_bdev *bdev = bdev_io->bdev;
1272 : uint64_t md_len;
1273 : void *buf;
1274 :
1275 42 : if (spdk_bdev_is_md_separate(bdev)) {
1276 7 : assert(!bdev_io_use_accel_sequence(bdev_io));
1277 :
1278 7 : buf = (char *)bdev_io->u.bdev.iovs[0].iov_base + bdev_io->u.bdev.iovs[0].iov_len;
1279 7 : md_len = bdev_io->u.bdev.num_blocks * bdev->md_len;
1280 :
1281 7 : assert(((uintptr_t)buf & (spdk_bdev_get_buf_align(bdev) - 1)) == 0);
1282 :
1283 7 : if (bdev_io->u.bdev.md_buf != NULL) {
1284 4 : _bdev_io_pull_bounce_md_buf(bdev_io, buf, md_len);
1285 4 : return;
1286 : } else {
1287 3 : spdk_bdev_io_set_md_buf(bdev_io, buf, md_len);
1288 : }
1289 3 : }
1290 :
1291 38 : bdev_io_get_buf_complete(bdev_io, true);
1292 42 : }
1293 :
1294 : static inline void
1295 26 : bdev_io_pull_data_done(struct spdk_bdev_io *bdev_io, int rc)
1296 : {
1297 26 : if (rc) {
1298 0 : SPDK_ERRLOG("Failed to get data buffer\n");
1299 0 : assert(bdev_io->internal.data_transfer_cpl);
1300 0 : bdev_io->internal.data_transfer_cpl(bdev_io, rc);
1301 0 : return;
1302 : }
1303 :
1304 26 : _bdev_io_set_md_buf(bdev_io);
1305 26 : }
1306 :
1307 : static void
1308 2 : bdev_io_pull_data_done_and_track(void *ctx, int status)
1309 : {
1310 2 : struct spdk_bdev_io *bdev_io = ctx;
1311 2 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1312 :
1313 2 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1314 2 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1315 :
1316 2 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1317 0 : bdev_ch_retry_io(ch);
1318 0 : }
1319 :
1320 2 : bdev_io_pull_data_done(bdev_io, status);
1321 2 : }
1322 :
1323 : static void
1324 27 : bdev_io_pull_data(struct spdk_bdev_io *bdev_io)
1325 : {
1326 27 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1327 27 : struct spdk_bdev_desc *desc = bdev_io->internal.desc;
1328 27 : int rc = 0;
1329 :
1330 27 : assert(bdev_io->internal.f.has_bounce_buf);
1331 :
1332 27 : if (bdev_io_needs_metadata(desc, bdev_io)) {
1333 0 : assert(bdev_io->bdev->md_interleave);
1334 :
1335 0 : bdev_io->u.bdev.dif_check_flags &= ~SPDK_DIF_FLAGS_NVME_PRACT;
1336 :
1337 0 : if (!bdev_io_use_accel_sequence(bdev_io)) {
1338 0 : bdev_io->internal.accel_sequence = NULL;
1339 0 : }
1340 :
1341 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
1342 0 : rc = spdk_accel_append_dif_generate_copy(&bdev_io->internal.accel_sequence, ch->accel_channel,
1343 0 : bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt,
1344 0 : bdev_io->u.bdev.memory_domain,
1345 0 : bdev_io->u.bdev.memory_domain_ctx,
1346 0 : bdev_io->internal.bounce_buf.orig_iovs,
1347 0 : bdev_io->internal.bounce_buf.orig_iovcnt,
1348 0 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain : NULL,
1349 0 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain_ctx : NULL,
1350 0 : bdev_io->u.bdev.num_blocks,
1351 0 : &bdev_io->u.bdev.dif_ctx,
1352 : NULL, NULL);
1353 0 : } else {
1354 0 : assert(bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
1355 0 : rc = spdk_accel_append_dif_verify_copy(&bdev_io->internal.accel_sequence, ch->accel_channel,
1356 0 : bdev_io->internal.bounce_buf.orig_iovs,
1357 0 : bdev_io->internal.bounce_buf.orig_iovcnt,
1358 0 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain : NULL,
1359 0 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain_ctx : NULL,
1360 0 : bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt,
1361 0 : bdev_io->u.bdev.memory_domain,
1362 0 : bdev_io->u.bdev.memory_domain_ctx,
1363 0 : bdev_io->u.bdev.num_blocks,
1364 0 : &bdev_io->u.bdev.dif_ctx,
1365 0 : &bdev_io->u.bdev.dif_err,
1366 : NULL, NULL);
1367 : }
1368 :
1369 0 : if (spdk_likely(rc == 0)) {
1370 0 : bdev_io->internal.f.has_accel_sequence = true;
1371 0 : bdev_io->u.bdev.accel_sequence = bdev_io->internal.accel_sequence;
1372 0 : } else if (rc != -ENOMEM) {
1373 0 : SPDK_ERRLOG("Failed to append generate/verify_copy to accel sequence: %p\n",
1374 : bdev_io->internal.accel_sequence);
1375 0 : }
1376 27 : } else if (bdev_io_needs_sequence_exec(desc, bdev_io) ||
1377 27 : (bdev_io_use_accel_sequence(bdev_io) && bdev_io_use_memory_domain(bdev_io))) {
1378 : /* If we need to exec an accel sequence or the IO uses a memory domain buffer and has a
1379 : * sequence, append a copy operation making accel change the src/dst buffers of the previous
1380 : * operation */
1381 0 : assert(bdev_io_use_accel_sequence(bdev_io));
1382 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
1383 0 : rc = spdk_accel_append_copy(&bdev_io->internal.accel_sequence, ch->accel_channel,
1384 0 : bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt,
1385 : NULL, NULL,
1386 0 : bdev_io->internal.bounce_buf.orig_iovs,
1387 0 : bdev_io->internal.bounce_buf.orig_iovcnt,
1388 0 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain : NULL,
1389 0 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain_ctx : NULL,
1390 : NULL, NULL);
1391 0 : } else {
1392 : /* We need to reverse the src/dst for reads */
1393 0 : assert(bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
1394 0 : rc = spdk_accel_append_copy(&bdev_io->internal.accel_sequence, ch->accel_channel,
1395 0 : bdev_io->internal.bounce_buf.orig_iovs,
1396 0 : bdev_io->internal.bounce_buf.orig_iovcnt,
1397 0 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain : NULL,
1398 0 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain_ctx : NULL,
1399 0 : bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt,
1400 : NULL, NULL, NULL, NULL);
1401 : }
1402 :
1403 0 : if (spdk_unlikely(rc != 0 && rc != -ENOMEM)) {
1404 0 : SPDK_ERRLOG("Failed to append copy to accel sequence: %p\n",
1405 : bdev_io->internal.accel_sequence);
1406 0 : }
1407 27 : } else if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
1408 : /* if this is write path, copy data from original buffer to bounce buffer */
1409 17 : if (bdev_io_use_memory_domain(bdev_io)) {
1410 3 : TAILQ_INSERT_TAIL(&ch->io_memory_domain, bdev_io, internal.link);
1411 3 : bdev_io_increment_outstanding(ch, ch->shared_resource);
1412 6 : rc = spdk_memory_domain_pull_data(bdev_io->internal.memory_domain,
1413 3 : bdev_io->internal.memory_domain_ctx,
1414 3 : bdev_io->internal.bounce_buf.orig_iovs,
1415 3 : (uint32_t)bdev_io->internal.bounce_buf.orig_iovcnt,
1416 3 : bdev_io->u.bdev.iovs, 1,
1417 : bdev_io_pull_data_done_and_track,
1418 3 : bdev_io);
1419 3 : if (rc == 0) {
1420 : /* Continue to submit IO in completion callback */
1421 2 : return;
1422 : }
1423 1 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1424 1 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1425 1 : if (rc != -ENOMEM) {
1426 0 : SPDK_ERRLOG("Failed to pull data from memory domain %s\n",
1427 : spdk_memory_domain_get_dma_device_id(
1428 : bdev_io->internal.memory_domain));
1429 0 : }
1430 1 : } else {
1431 14 : assert(bdev_io->u.bdev.iovcnt == 1);
1432 28 : spdk_copy_iovs_to_buf(bdev_io->u.bdev.iovs[0].iov_base,
1433 14 : bdev_io->u.bdev.iovs[0].iov_len,
1434 14 : bdev_io->internal.bounce_buf.orig_iovs,
1435 14 : bdev_io->internal.bounce_buf.orig_iovcnt);
1436 : }
1437 15 : }
1438 :
1439 25 : if (spdk_unlikely(rc == -ENOMEM)) {
1440 1 : bdev_queue_nomem_io_head(ch->shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PULL);
1441 1 : } else {
1442 24 : bdev_io_pull_data_done(bdev_io, rc);
1443 : }
1444 27 : }
1445 :
1446 : static void
1447 26 : _bdev_io_pull_bounce_data_buf(struct spdk_bdev_io *bdev_io, void *buf, size_t len,
1448 : bdev_copy_bounce_buffer_cpl cpl_cb)
1449 : {
1450 26 : struct spdk_bdev_shared_resource *shared_resource = bdev_io->internal.ch->shared_resource;
1451 :
1452 26 : assert(bdev_io->internal.f.has_bounce_buf == false);
1453 :
1454 26 : bdev_io->internal.data_transfer_cpl = cpl_cb;
1455 26 : bdev_io->internal.f.has_bounce_buf = true;
1456 : /* save original iovec */
1457 26 : bdev_io->internal.bounce_buf.orig_iovs = bdev_io->u.bdev.iovs;
1458 26 : bdev_io->internal.bounce_buf.orig_iovcnt = bdev_io->u.bdev.iovcnt;
1459 : /* zero the other data members */
1460 26 : bdev_io->internal.bounce_buf.iov.iov_base = NULL;
1461 26 : bdev_io->internal.bounce_buf.md_iov.iov_base = NULL;
1462 26 : bdev_io->internal.bounce_buf.orig_md_iov.iov_base = NULL;
1463 : /* set bounce iov */
1464 26 : bdev_io->u.bdev.iovs = &bdev_io->internal.bounce_buf.iov;
1465 26 : bdev_io->u.bdev.iovcnt = 1;
1466 : /* set bounce buffer for this operation */
1467 26 : bdev_io->u.bdev.iovs[0].iov_base = buf;
1468 26 : bdev_io->u.bdev.iovs[0].iov_len = len;
1469 : /* Now we use 1 iov, the split condition could have been changed */
1470 26 : bdev_io->internal.f.split = bdev_io_should_split(bdev_io);
1471 :
1472 26 : if (spdk_unlikely(!TAILQ_EMPTY(&shared_resource->nomem_io))) {
1473 0 : bdev_queue_nomem_io_tail(shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PULL);
1474 0 : } else {
1475 26 : bdev_io_pull_data(bdev_io);
1476 : }
1477 26 : }
1478 :
1479 : static void
1480 42 : _bdev_io_set_buf(struct spdk_bdev_io *bdev_io, void *buf, uint64_t len)
1481 : {
1482 42 : struct spdk_bdev *bdev = bdev_io->bdev;
1483 : bool buf_allocated;
1484 : uint64_t alignment;
1485 : void *aligned_buf;
1486 :
1487 42 : bdev_io->internal.buf.ptr = buf;
1488 42 : bdev_io->internal.f.has_buf = true;
1489 :
1490 42 : if (spdk_unlikely(bdev_io->internal.get_aux_buf_cb != NULL)) {
1491 0 : bdev_io_get_buf_complete(bdev_io, true);
1492 0 : return;
1493 : }
1494 :
1495 42 : alignment = spdk_bdev_get_buf_align(bdev);
1496 42 : buf_allocated = _is_buf_allocated(bdev_io->u.bdev.iovs);
1497 42 : aligned_buf = (void *)(((uintptr_t)buf + (alignment - 1)) & ~(alignment - 1));
1498 :
1499 42 : if (buf_allocated) {
1500 26 : _bdev_io_pull_bounce_data_buf(bdev_io, aligned_buf, len, _bdev_io_pull_buffer_cpl);
1501 : /* Continue in completion callback */
1502 26 : return;
1503 : } else {
1504 16 : spdk_bdev_io_set_buf(bdev_io, aligned_buf, len);
1505 : }
1506 :
1507 16 : _bdev_io_set_md_buf(bdev_io);
1508 42 : }
1509 :
1510 : static inline uint64_t
1511 42 : bdev_io_get_max_buf_len(struct spdk_bdev_io *bdev_io, uint64_t len)
1512 : {
1513 42 : struct spdk_bdev *bdev = bdev_io->bdev;
1514 : uint64_t md_len, alignment;
1515 :
1516 42 : md_len = spdk_bdev_is_md_separate(bdev) ? bdev_io->u.bdev.num_blocks * bdev->md_len : 0;
1517 :
1518 : /* 1 byte alignment needs 0 byte of extra space, 64 bytes alignment needs 63 bytes of extra space, etc. */
1519 42 : alignment = spdk_bdev_get_buf_align(bdev) - 1;
1520 :
1521 42 : return len + alignment + md_len;
1522 : }
1523 :
1524 : static void
1525 42 : bdev_io_put_accel_buf(struct spdk_bdev_io *bdev_io)
1526 : {
1527 42 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1528 :
1529 84 : spdk_accel_put_buf(ch->accel_channel,
1530 42 : bdev_io->internal.buf.ptr,
1531 42 : bdev_io->u.bdev.memory_domain,
1532 42 : bdev_io->u.bdev.memory_domain_ctx);
1533 42 : }
1534 :
1535 : static void
1536 0 : _bdev_io_put_buf(struct spdk_bdev_io *bdev_io, void *buf, uint64_t buf_len)
1537 : {
1538 : struct spdk_bdev_mgmt_channel *ch;
1539 :
1540 0 : ch = bdev_io->internal.ch->shared_resource->mgmt_ch;
1541 0 : spdk_iobuf_put(&ch->iobuf, buf, bdev_io_get_max_buf_len(bdev_io, buf_len));
1542 0 : }
1543 :
1544 : static void
1545 42 : bdev_io_put_buf(struct spdk_bdev_io *bdev_io)
1546 : {
1547 42 : assert(bdev_io->internal.f.has_buf);
1548 :
1549 42 : if (bdev_io->u.bdev.memory_domain == spdk_accel_get_memory_domain()) {
1550 42 : bdev_io_put_accel_buf(bdev_io);
1551 42 : } else {
1552 0 : assert(bdev_io->u.bdev.memory_domain == NULL);
1553 0 : _bdev_io_put_buf(bdev_io, bdev_io->internal.buf.ptr,
1554 0 : bdev_io->internal.buf.len);
1555 : }
1556 42 : bdev_io->internal.buf.ptr = NULL;
1557 42 : bdev_io->internal.f.has_buf = false;
1558 42 : }
1559 :
1560 3 : SPDK_LOG_DEPRECATION_REGISTER(spdk_bdev_io_put_aux_buf,
1561 : "spdk_bdev_io_put_aux_buf is deprecated", "v25.01", 0);
1562 :
1563 : void
1564 0 : spdk_bdev_io_put_aux_buf(struct spdk_bdev_io *bdev_io, void *buf)
1565 : {
1566 0 : uint64_t len = bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen;
1567 :
1568 0 : SPDK_LOG_DEPRECATED(spdk_bdev_io_put_aux_buf);
1569 :
1570 0 : assert(buf != NULL);
1571 0 : _bdev_io_put_buf(bdev_io, buf, len);
1572 0 : }
1573 :
1574 : static inline bool
1575 548 : bdev_io_support_dif(struct spdk_bdev_io *bdev_io)
1576 : {
1577 548 : switch (bdev_io->type) {
1578 : case SPDK_BDEV_IO_TYPE_WRITE:
1579 : case SPDK_BDEV_IO_TYPE_READ:
1580 442 : return true;
1581 : default:
1582 106 : return false;
1583 : }
1584 548 : }
1585 :
1586 : static inline void
1587 548 : bdev_submit_request(struct spdk_bdev *bdev, struct spdk_io_channel *ioch,
1588 : struct spdk_bdev_io *bdev_io)
1589 : {
1590 : /* After a request is submitted to a bdev module, the ownership of an accel sequence
1591 : * associated with that bdev_io is transferred to the bdev module. So, clear the internal
1592 : * sequence pointer to make sure we won't touch it anymore. */
1593 1014 : if ((bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE ||
1594 548 : bdev_io->type == SPDK_BDEV_IO_TYPE_READ) && bdev_io->u.bdev.accel_sequence != NULL) {
1595 0 : assert(!bdev_io_needs_sequence_exec(bdev_io->internal.desc, bdev_io));
1596 0 : bdev_io->internal.f.has_accel_sequence = false;
1597 0 : }
1598 :
1599 : /* The generic bdev layer should not pass an I/O with a dif_check_flags set that
1600 : * the underlying bdev does not support. Add an assert to check this.
1601 : */
1602 548 : assert(!bdev_io_support_dif(bdev_io) ||
1603 : ((bdev_io->u.bdev.dif_check_flags & bdev->dif_check_flags) ==
1604 : bdev_io->u.bdev.dif_check_flags));
1605 :
1606 548 : bdev->fn_table->submit_request(ioch, bdev_io);
1607 548 : }
1608 :
1609 : static inline void
1610 10 : bdev_ch_resubmit_io(struct spdk_bdev_shared_resource *shared_resource, struct spdk_bdev_io *bdev_io)
1611 : {
1612 10 : struct spdk_bdev *bdev = bdev_io->bdev;
1613 :
1614 10 : bdev_io_increment_outstanding(bdev_io->internal.ch, shared_resource);
1615 10 : bdev_io->internal.error.nvme.cdw0 = 0;
1616 10 : bdev_io->num_retries++;
1617 10 : bdev_submit_request(bdev, spdk_bdev_io_get_io_channel(bdev_io), bdev_io);
1618 10 : }
1619 :
1620 : static void
1621 63 : bdev_shared_ch_retry_io(struct spdk_bdev_shared_resource *shared_resource)
1622 : {
1623 : struct spdk_bdev_io *bdev_io;
1624 :
1625 63 : if (shared_resource->io_outstanding > shared_resource->nomem_threshold) {
1626 : /*
1627 : * Allow some more I/O to complete before retrying the nomem_io queue.
1628 : * Some drivers (such as nvme) cannot immediately take a new I/O in
1629 : * the context of a completion, because the resources for the I/O are
1630 : * not released until control returns to the bdev poller. Also, we
1631 : * may require several small I/O to complete before a larger I/O
1632 : * (that requires splitting) can be submitted.
1633 : */
1634 58 : return;
1635 : }
1636 :
1637 16 : while (!TAILQ_EMPTY(&shared_resource->nomem_io)) {
1638 12 : bdev_io = TAILQ_FIRST(&shared_resource->nomem_io);
1639 12 : TAILQ_REMOVE(&shared_resource->nomem_io, bdev_io, internal.link);
1640 :
1641 12 : switch (bdev_io->internal.retry_state) {
1642 : case BDEV_IO_RETRY_STATE_SUBMIT:
1643 10 : bdev_ch_resubmit_io(shared_resource, bdev_io);
1644 10 : break;
1645 : case BDEV_IO_RETRY_STATE_PULL:
1646 1 : bdev_io_pull_data(bdev_io);
1647 1 : break;
1648 : case BDEV_IO_RETRY_STATE_PULL_MD:
1649 0 : bdev_io_pull_md_buf(bdev_io);
1650 0 : break;
1651 : case BDEV_IO_RETRY_STATE_PUSH:
1652 1 : bdev_io_push_bounce_data(bdev_io);
1653 1 : break;
1654 : case BDEV_IO_RETRY_STATE_PUSH_MD:
1655 0 : bdev_io_push_bounce_md_buf(bdev_io);
1656 0 : break;
1657 : case BDEV_IO_RETRY_STATE_GET_ACCEL_BUF:
1658 0 : _bdev_io_get_accel_buf(bdev_io);
1659 0 : break;
1660 : default:
1661 0 : assert(0 && "invalid retry state");
1662 : break;
1663 : }
1664 :
1665 12 : if (bdev_io == TAILQ_FIRST(&shared_resource->nomem_io)) {
1666 : /* This IO completed again with NOMEM status, so break the loop and
1667 : * don't try anymore. Note that a bdev_io that fails with NOMEM
1668 : * always gets requeued at the front of the list, to maintain
1669 : * ordering.
1670 : */
1671 1 : break;
1672 : }
1673 : }
1674 63 : }
1675 :
1676 : static void
1677 63 : bdev_ch_retry_io(struct spdk_bdev_channel *bdev_ch)
1678 : {
1679 63 : bdev_shared_ch_retry_io(bdev_ch->shared_resource);
1680 63 : }
1681 :
1682 : static int
1683 0 : bdev_no_mem_poller(void *ctx)
1684 : {
1685 0 : struct spdk_bdev_shared_resource *shared_resource = ctx;
1686 :
1687 0 : spdk_poller_unregister(&shared_resource->nomem_poller);
1688 :
1689 0 : if (!TAILQ_EMPTY(&shared_resource->nomem_io)) {
1690 0 : bdev_shared_ch_retry_io(shared_resource);
1691 0 : }
1692 : /* the retry cb may re-register the poller so double check */
1693 0 : if (!TAILQ_EMPTY(&shared_resource->nomem_io) &&
1694 0 : shared_resource->io_outstanding == 0 && shared_resource->nomem_poller == NULL) {
1695 : /* No IOs were submitted, try again */
1696 0 : shared_resource->nomem_poller = SPDK_POLLER_REGISTER(bdev_no_mem_poller, shared_resource,
1697 : SPDK_BDEV_IO_POLL_INTERVAL_IN_MSEC * 10);
1698 0 : }
1699 :
1700 0 : return SPDK_POLLER_BUSY;
1701 : }
1702 :
1703 : static inline bool
1704 556 : _bdev_io_handle_no_mem(struct spdk_bdev_io *bdev_io, enum bdev_io_retry_state state)
1705 : {
1706 556 : struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
1707 556 : struct spdk_bdev_shared_resource *shared_resource = bdev_ch->shared_resource;
1708 :
1709 556 : if (spdk_unlikely(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_NOMEM)) {
1710 5 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_PENDING;
1711 5 : bdev_queue_nomem_io_head(shared_resource, bdev_io, state);
1712 :
1713 5 : if (shared_resource->io_outstanding == 0 && !shared_resource->nomem_poller) {
1714 : /* Special case when we have nomem IOs and no outstanding IOs which completions
1715 : * could trigger retry of queued IOs
1716 : * Any IOs submitted may trigger retry of queued IOs. This poller handles a case when no
1717 : * new IOs submitted, e.g. qd==1 */
1718 0 : shared_resource->nomem_poller = SPDK_POLLER_REGISTER(bdev_no_mem_poller, shared_resource,
1719 : SPDK_BDEV_IO_POLL_INTERVAL_IN_MSEC * 10);
1720 0 : }
1721 : /* If bdev module completed an I/O that has an accel sequence with NOMEM status, the
1722 : * ownership of that sequence is transferred back to the bdev layer, so we need to
1723 : * restore internal.accel_sequence to make sure that the sequence is handled
1724 : * correctly in case the I/O is later aborted. */
1725 5 : if ((bdev_io->type == SPDK_BDEV_IO_TYPE_READ ||
1726 5 : bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) && bdev_io->u.bdev.accel_sequence) {
1727 0 : assert(!bdev_io_use_accel_sequence(bdev_io));
1728 0 : bdev_io->internal.f.has_accel_sequence = true;
1729 0 : bdev_io->internal.accel_sequence = bdev_io->u.bdev.accel_sequence;
1730 0 : }
1731 :
1732 5 : return true;
1733 : }
1734 :
1735 551 : if (spdk_unlikely(!TAILQ_EMPTY(&shared_resource->nomem_io))) {
1736 63 : bdev_ch_retry_io(bdev_ch);
1737 63 : }
1738 :
1739 551 : return false;
1740 556 : }
1741 :
1742 : static void
1743 26 : _bdev_io_complete_push_bounce_done(void *ctx, int rc)
1744 : {
1745 26 : struct spdk_bdev_io *bdev_io = ctx;
1746 26 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1747 :
1748 26 : if (rc) {
1749 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
1750 0 : }
1751 : /* We want to free the bounce buffer here since we know we're done with it (as opposed
1752 : * to waiting for the conditional free of internal.buf.ptr in spdk_bdev_free_io()).
1753 : */
1754 26 : bdev_io_put_buf(bdev_io);
1755 :
1756 26 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1757 0 : bdev_ch_retry_io(ch);
1758 0 : }
1759 :
1760 : /* Continue with IO completion flow */
1761 26 : bdev_io_complete(bdev_io);
1762 26 : }
1763 :
1764 : static void
1765 2 : bdev_io_push_bounce_md_buf_done(void *ctx, int rc)
1766 : {
1767 2 : struct spdk_bdev_io *bdev_io = ctx;
1768 2 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1769 :
1770 2 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1771 2 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1772 2 : bdev_io->internal.f.has_bounce_buf = false;
1773 :
1774 2 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1775 0 : bdev_ch_retry_io(ch);
1776 0 : }
1777 :
1778 2 : bdev_io->internal.data_transfer_cpl(bdev_io, rc);
1779 2 : }
1780 :
1781 : static inline void
1782 26 : bdev_io_push_bounce_md_buf(struct spdk_bdev_io *bdev_io)
1783 : {
1784 26 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1785 26 : int rc = 0;
1786 :
1787 26 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS);
1788 26 : assert(bdev_io->internal.f.has_bounce_buf);
1789 :
1790 : /* do the same for metadata buffer */
1791 26 : if (spdk_unlikely(bdev_io->internal.bounce_buf.orig_md_iov.iov_base != NULL)) {
1792 4 : assert(spdk_bdev_is_md_separate(bdev_io->bdev));
1793 :
1794 4 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_READ) {
1795 2 : if (bdev_io_use_memory_domain(bdev_io)) {
1796 2 : TAILQ_INSERT_TAIL(&ch->io_memory_domain, bdev_io, internal.link);
1797 2 : bdev_io_increment_outstanding(ch, ch->shared_resource);
1798 : /* If memory domain is used then we need to call async push function */
1799 4 : rc = spdk_memory_domain_push_data(bdev_io->internal.memory_domain,
1800 2 : bdev_io->internal.memory_domain_ctx,
1801 2 : &bdev_io->internal.bounce_buf.orig_md_iov,
1802 2 : (uint32_t)bdev_io->internal.bounce_buf.orig_iovcnt,
1803 2 : &bdev_io->internal.bounce_buf.md_iov, 1,
1804 : bdev_io_push_bounce_md_buf_done,
1805 2 : bdev_io);
1806 2 : if (rc == 0) {
1807 : /* Continue IO completion in async callback */
1808 2 : return;
1809 : }
1810 0 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1811 0 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1812 0 : if (rc != -ENOMEM) {
1813 0 : SPDK_ERRLOG("Failed to push md to memory domain %s\n",
1814 : spdk_memory_domain_get_dma_device_id(
1815 : bdev_io->internal.memory_domain));
1816 0 : }
1817 0 : } else {
1818 0 : memcpy(bdev_io->internal.bounce_buf.orig_md_iov.iov_base, bdev_io->u.bdev.md_buf,
1819 0 : bdev_io->internal.bounce_buf.orig_md_iov.iov_len);
1820 : }
1821 0 : }
1822 2 : }
1823 :
1824 24 : if (spdk_unlikely(rc == -ENOMEM)) {
1825 0 : bdev_queue_nomem_io_head(ch->shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PUSH_MD);
1826 0 : } else {
1827 24 : assert(bdev_io->internal.data_transfer_cpl);
1828 24 : bdev_io->internal.f.has_bounce_buf = false;
1829 24 : bdev_io->internal.data_transfer_cpl(bdev_io, rc);
1830 : }
1831 26 : }
1832 :
1833 : static inline void
1834 26 : bdev_io_push_bounce_data_done(struct spdk_bdev_io *bdev_io, int rc)
1835 : {
1836 26 : assert(bdev_io->internal.data_transfer_cpl);
1837 26 : if (rc) {
1838 0 : bdev_io->internal.data_transfer_cpl(bdev_io, rc);
1839 0 : return;
1840 : }
1841 :
1842 : /* set original buffer for this io */
1843 26 : bdev_io->u.bdev.iovcnt = bdev_io->internal.bounce_buf.orig_iovcnt;
1844 26 : bdev_io->u.bdev.iovs = bdev_io->internal.bounce_buf.orig_iovs;
1845 :
1846 : /* We don't set bdev_io->internal.f.has_bounce_buf to false here because
1847 : * we still need to clear the md buf */
1848 :
1849 26 : bdev_io_push_bounce_md_buf(bdev_io);
1850 26 : }
1851 :
1852 : static void
1853 2 : bdev_io_push_bounce_data_done_and_track(void *ctx, int status)
1854 : {
1855 2 : struct spdk_bdev_io *bdev_io = ctx;
1856 2 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1857 :
1858 2 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1859 2 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1860 :
1861 2 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1862 0 : bdev_ch_retry_io(ch);
1863 0 : }
1864 :
1865 2 : bdev_io_push_bounce_data_done(bdev_io, status);
1866 2 : }
1867 :
1868 : static inline void
1869 27 : bdev_io_push_bounce_data(struct spdk_bdev_io *bdev_io)
1870 : {
1871 27 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1872 27 : int rc = 0;
1873 :
1874 27 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS);
1875 27 : assert(!bdev_io_use_accel_sequence(bdev_io));
1876 27 : assert(bdev_io->internal.f.has_bounce_buf);
1877 :
1878 : /* if this is read path, copy data from bounce buffer to original buffer */
1879 27 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_READ) {
1880 11 : if (bdev_io_use_memory_domain(bdev_io)) {
1881 3 : TAILQ_INSERT_TAIL(&ch->io_memory_domain, bdev_io, internal.link);
1882 3 : bdev_io_increment_outstanding(ch, ch->shared_resource);
1883 : /* If memory domain is used then we need to call async push function */
1884 6 : rc = spdk_memory_domain_push_data(bdev_io->internal.memory_domain,
1885 3 : bdev_io->internal.memory_domain_ctx,
1886 3 : bdev_io->internal.bounce_buf.orig_iovs,
1887 3 : (uint32_t)bdev_io->internal.bounce_buf.orig_iovcnt,
1888 3 : &bdev_io->internal.bounce_buf.iov, 1,
1889 : bdev_io_push_bounce_data_done_and_track,
1890 3 : bdev_io);
1891 3 : if (rc == 0) {
1892 : /* Continue IO completion in async callback */
1893 2 : return;
1894 : }
1895 :
1896 1 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1897 1 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1898 1 : if (rc != -ENOMEM) {
1899 0 : SPDK_ERRLOG("Failed to push data to memory domain %s\n",
1900 : spdk_memory_domain_get_dma_device_id(
1901 : bdev_io->internal.memory_domain));
1902 0 : }
1903 1 : } else {
1904 16 : spdk_copy_buf_to_iovs(bdev_io->internal.bounce_buf.orig_iovs,
1905 8 : bdev_io->internal.bounce_buf.orig_iovcnt,
1906 8 : bdev_io->internal.bounce_buf.iov.iov_base,
1907 8 : bdev_io->internal.bounce_buf.iov.iov_len);
1908 : }
1909 9 : }
1910 :
1911 25 : if (spdk_unlikely(rc == -ENOMEM)) {
1912 1 : bdev_queue_nomem_io_head(ch->shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PUSH);
1913 1 : } else {
1914 24 : bdev_io_push_bounce_data_done(bdev_io, rc);
1915 : }
1916 27 : }
1917 :
1918 : static inline void
1919 26 : _bdev_io_push_bounce_data_buffer(struct spdk_bdev_io *bdev_io, bdev_copy_bounce_buffer_cpl cpl_cb)
1920 : {
1921 26 : bdev_io->internal.data_transfer_cpl = cpl_cb;
1922 26 : bdev_io_push_bounce_data(bdev_io);
1923 26 : }
1924 :
1925 : static void
1926 0 : bdev_io_get_iobuf_cb(struct spdk_iobuf_entry *iobuf, void *buf)
1927 : {
1928 : struct spdk_bdev_io *bdev_io;
1929 :
1930 0 : bdev_io = SPDK_CONTAINEROF(iobuf, struct spdk_bdev_io, internal.iobuf);
1931 0 : _bdev_io_set_buf(bdev_io, buf, bdev_io->internal.buf.len);
1932 0 : }
1933 :
1934 : static void
1935 42 : bdev_io_get_buf(struct spdk_bdev_io *bdev_io, uint64_t len)
1936 : {
1937 : struct spdk_bdev_mgmt_channel *mgmt_ch;
1938 : uint64_t max_len;
1939 : void *buf;
1940 :
1941 42 : assert(spdk_bdev_io_get_thread(bdev_io) == spdk_get_thread());
1942 42 : mgmt_ch = bdev_io->internal.ch->shared_resource->mgmt_ch;
1943 42 : max_len = bdev_io_get_max_buf_len(bdev_io, len);
1944 :
1945 42 : if (spdk_unlikely(max_len > mgmt_ch->iobuf.cache[0].large.bufsize)) {
1946 0 : SPDK_ERRLOG("Length %" PRIu64 " is larger than allowed\n", max_len);
1947 0 : bdev_io_get_buf_complete(bdev_io, false);
1948 0 : return;
1949 : }
1950 :
1951 42 : bdev_io->internal.buf.len = len;
1952 42 : buf = spdk_iobuf_get(&mgmt_ch->iobuf, max_len, &bdev_io->internal.iobuf,
1953 : bdev_io_get_iobuf_cb);
1954 42 : if (buf != NULL) {
1955 42 : _bdev_io_set_buf(bdev_io, buf, len);
1956 42 : }
1957 42 : }
1958 :
1959 : void
1960 56 : spdk_bdev_io_get_buf(struct spdk_bdev_io *bdev_io, spdk_bdev_io_get_buf_cb cb, uint64_t len)
1961 : {
1962 56 : struct spdk_bdev *bdev = bdev_io->bdev;
1963 : uint64_t alignment;
1964 :
1965 56 : assert(cb != NULL);
1966 56 : bdev_io->internal.get_buf_cb = cb;
1967 :
1968 56 : alignment = spdk_bdev_get_buf_align(bdev);
1969 :
1970 56 : if (_is_buf_allocated(bdev_io->u.bdev.iovs) &&
1971 40 : _are_iovs_aligned(bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt, alignment)) {
1972 : /* Buffer already present and aligned */
1973 18 : cb(spdk_bdev_io_get_io_channel(bdev_io), bdev_io, true);
1974 18 : return;
1975 : }
1976 :
1977 38 : bdev_io_get_buf(bdev_io, len);
1978 56 : }
1979 :
1980 : static void
1981 4 : _bdev_io_get_bounce_buf(struct spdk_bdev_io *bdev_io, spdk_bdev_io_get_buf_cb cb,
1982 : uint64_t len)
1983 : {
1984 4 : assert(cb != NULL);
1985 4 : bdev_io->internal.get_buf_cb = cb;
1986 :
1987 4 : bdev_io_get_buf(bdev_io, len);
1988 4 : }
1989 :
1990 : static void
1991 0 : _bdev_io_get_accel_buf(struct spdk_bdev_io *bdev_io)
1992 : {
1993 0 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1994 : void *buf;
1995 : int rc;
1996 :
1997 0 : rc = spdk_accel_get_buf(ch->accel_channel,
1998 0 : bdev_io->internal.buf.len,
1999 : &buf,
2000 0 : &bdev_io->u.bdev.memory_domain,
2001 0 : &bdev_io->u.bdev.memory_domain_ctx);
2002 0 : if (rc != 0) {
2003 0 : bdev_queue_nomem_io_tail(ch->shared_resource, bdev_io,
2004 : BDEV_IO_RETRY_STATE_GET_ACCEL_BUF);
2005 0 : return;
2006 : }
2007 :
2008 0 : _bdev_io_set_buf(bdev_io, buf, bdev_io->internal.buf.len);
2009 0 : }
2010 :
2011 : static inline void
2012 0 : bdev_io_get_accel_buf(struct spdk_bdev_io *bdev_io, spdk_bdev_io_get_buf_cb cb,
2013 : uint64_t len)
2014 : {
2015 0 : bdev_io->internal.buf.len = len;
2016 0 : bdev_io->internal.get_buf_cb = cb;
2017 :
2018 0 : _bdev_io_get_accel_buf(bdev_io);
2019 0 : }
2020 :
2021 3 : SPDK_LOG_DEPRECATION_REGISTER(spdk_bdev_io_get_aux_buf,
2022 : "spdk_bdev_io_get_aux_buf is deprecated", "v25.01", 0);
2023 :
2024 : void
2025 0 : spdk_bdev_io_get_aux_buf(struct spdk_bdev_io *bdev_io, spdk_bdev_io_get_aux_buf_cb cb)
2026 : {
2027 0 : uint64_t len = bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen;
2028 :
2029 0 : SPDK_LOG_DEPRECATED(spdk_bdev_io_get_aux_buf);
2030 :
2031 0 : assert(cb != NULL);
2032 0 : assert(bdev_io->internal.get_aux_buf_cb == NULL);
2033 0 : bdev_io->internal.get_aux_buf_cb = cb;
2034 0 : bdev_io_get_buf(bdev_io, len);
2035 0 : }
2036 :
2037 : static int
2038 65 : bdev_module_get_max_ctx_size(void)
2039 : {
2040 : struct spdk_bdev_module *bdev_module;
2041 65 : int max_bdev_module_size = 0;
2042 :
2043 254 : TAILQ_FOREACH(bdev_module, &g_bdev_mgr.bdev_modules, internal.tailq) {
2044 189 : if (bdev_module->get_ctx_size && bdev_module->get_ctx_size() > max_bdev_module_size) {
2045 64 : max_bdev_module_size = bdev_module->get_ctx_size();
2046 64 : }
2047 189 : }
2048 :
2049 65 : return max_bdev_module_size;
2050 : }
2051 :
2052 : static void
2053 0 : bdev_enable_histogram_config_json(struct spdk_bdev *bdev, struct spdk_json_write_ctx *w)
2054 : {
2055 0 : if (!bdev->internal.histogram_enabled) {
2056 0 : return;
2057 : }
2058 :
2059 0 : spdk_json_write_object_begin(w);
2060 0 : spdk_json_write_named_string(w, "method", "bdev_enable_histogram");
2061 :
2062 0 : spdk_json_write_named_object_begin(w, "params");
2063 0 : spdk_json_write_named_string(w, "name", bdev->name);
2064 :
2065 0 : spdk_json_write_named_bool(w, "enable", bdev->internal.histogram_enabled);
2066 :
2067 0 : if (bdev->internal.histogram_io_type) {
2068 0 : spdk_json_write_named_string(w, "opc",
2069 0 : spdk_bdev_get_io_type_name(bdev->internal.histogram_io_type));
2070 0 : }
2071 :
2072 0 : spdk_json_write_object_end(w);
2073 :
2074 0 : spdk_json_write_object_end(w);
2075 0 : }
2076 :
2077 : static void
2078 0 : bdev_qos_config_json(struct spdk_bdev *bdev, struct spdk_json_write_ctx *w)
2079 : {
2080 : int i;
2081 0 : struct spdk_bdev_qos *qos = bdev->internal.qos;
2082 : uint64_t limits[SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES];
2083 :
2084 0 : if (!qos) {
2085 0 : return;
2086 : }
2087 :
2088 0 : spdk_bdev_get_qos_rate_limits(bdev, limits);
2089 :
2090 0 : spdk_json_write_object_begin(w);
2091 0 : spdk_json_write_named_string(w, "method", "bdev_set_qos_limit");
2092 :
2093 0 : spdk_json_write_named_object_begin(w, "params");
2094 0 : spdk_json_write_named_string(w, "name", bdev->name);
2095 0 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
2096 0 : if (limits[i] > 0) {
2097 0 : spdk_json_write_named_uint64(w, qos_rpc_type[i], limits[i]);
2098 0 : }
2099 0 : }
2100 0 : spdk_json_write_object_end(w);
2101 :
2102 0 : spdk_json_write_object_end(w);
2103 0 : }
2104 :
2105 : void
2106 0 : spdk_bdev_subsystem_config_json(struct spdk_json_write_ctx *w)
2107 : {
2108 : struct spdk_bdev_module *bdev_module;
2109 : struct spdk_bdev *bdev;
2110 :
2111 0 : assert(w != NULL);
2112 :
2113 0 : spdk_json_write_array_begin(w);
2114 :
2115 0 : spdk_json_write_object_begin(w);
2116 0 : spdk_json_write_named_string(w, "method", "bdev_set_options");
2117 0 : spdk_json_write_named_object_begin(w, "params");
2118 0 : spdk_json_write_named_uint32(w, "bdev_io_pool_size", g_bdev_opts.bdev_io_pool_size);
2119 0 : spdk_json_write_named_uint32(w, "bdev_io_cache_size", g_bdev_opts.bdev_io_cache_size);
2120 0 : spdk_json_write_named_bool(w, "bdev_auto_examine", g_bdev_opts.bdev_auto_examine);
2121 0 : spdk_json_write_named_uint32(w, "iobuf_small_cache_size", g_bdev_opts.iobuf_small_cache_size);
2122 0 : spdk_json_write_named_uint32(w, "iobuf_large_cache_size", g_bdev_opts.iobuf_large_cache_size);
2123 0 : spdk_json_write_object_end(w);
2124 0 : spdk_json_write_object_end(w);
2125 :
2126 0 : bdev_examine_allowlist_config_json(w);
2127 :
2128 0 : TAILQ_FOREACH(bdev_module, &g_bdev_mgr.bdev_modules, internal.tailq) {
2129 0 : if (bdev_module->config_json) {
2130 0 : bdev_module->config_json(w);
2131 0 : }
2132 0 : }
2133 :
2134 0 : spdk_spin_lock(&g_bdev_mgr.spinlock);
2135 :
2136 0 : TAILQ_FOREACH(bdev, &g_bdev_mgr.bdevs, internal.link) {
2137 0 : if (bdev->fn_table->write_config_json) {
2138 0 : bdev->fn_table->write_config_json(bdev, w);
2139 0 : }
2140 :
2141 0 : bdev_qos_config_json(bdev, w);
2142 0 : bdev_enable_histogram_config_json(bdev, w);
2143 0 : }
2144 :
2145 0 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
2146 :
2147 : /* This has to be last RPC in array to make sure all bdevs finished examine */
2148 0 : spdk_json_write_object_begin(w);
2149 0 : spdk_json_write_named_string(w, "method", "bdev_wait_for_examine");
2150 0 : spdk_json_write_object_end(w);
2151 :
2152 0 : spdk_json_write_array_end(w);
2153 0 : }
2154 :
2155 : static void
2156 71 : bdev_mgmt_channel_destroy(void *io_device, void *ctx_buf)
2157 : {
2158 71 : struct spdk_bdev_mgmt_channel *ch = ctx_buf;
2159 : struct spdk_bdev_io *bdev_io;
2160 :
2161 71 : spdk_iobuf_channel_fini(&ch->iobuf);
2162 :
2163 9969 : while (!STAILQ_EMPTY(&ch->per_thread_cache)) {
2164 9898 : bdev_io = STAILQ_FIRST(&ch->per_thread_cache);
2165 9898 : STAILQ_REMOVE_HEAD(&ch->per_thread_cache, internal.buf_link);
2166 9898 : ch->per_thread_cache_count--;
2167 9898 : spdk_mempool_put(g_bdev_mgr.bdev_io_pool, (void *)bdev_io);
2168 : }
2169 :
2170 71 : assert(ch->per_thread_cache_count == 0);
2171 71 : }
2172 :
2173 : static int
2174 71 : bdev_mgmt_channel_create(void *io_device, void *ctx_buf)
2175 : {
2176 71 : struct spdk_bdev_mgmt_channel *ch = ctx_buf;
2177 : struct spdk_bdev_io *bdev_io;
2178 : uint32_t i;
2179 : int rc;
2180 :
2181 142 : rc = spdk_iobuf_channel_init(&ch->iobuf, "bdev",
2182 71 : g_bdev_opts.iobuf_small_cache_size,
2183 71 : g_bdev_opts.iobuf_large_cache_size);
2184 71 : if (rc != 0) {
2185 0 : SPDK_ERRLOG("Failed to create iobuf channel: %s\n", spdk_strerror(-rc));
2186 0 : return -1;
2187 : }
2188 :
2189 71 : STAILQ_INIT(&ch->per_thread_cache);
2190 71 : ch->bdev_io_cache_size = g_bdev_opts.bdev_io_cache_size;
2191 :
2192 : /* Pre-populate bdev_io cache to ensure this thread cannot be starved. */
2193 71 : ch->per_thread_cache_count = 0;
2194 9969 : for (i = 0; i < ch->bdev_io_cache_size; i++) {
2195 9898 : bdev_io = spdk_mempool_get(g_bdev_mgr.bdev_io_pool);
2196 9898 : if (bdev_io == NULL) {
2197 0 : SPDK_ERRLOG("You need to increase bdev_io_pool_size using bdev_set_options RPC.\n");
2198 0 : assert(false);
2199 : bdev_mgmt_channel_destroy(io_device, ctx_buf);
2200 : return -1;
2201 : }
2202 9898 : ch->per_thread_cache_count++;
2203 9898 : STAILQ_INSERT_HEAD(&ch->per_thread_cache, bdev_io, internal.buf_link);
2204 9898 : }
2205 :
2206 71 : TAILQ_INIT(&ch->shared_resources);
2207 71 : TAILQ_INIT(&ch->io_wait_queue);
2208 :
2209 71 : return 0;
2210 71 : }
2211 :
2212 : static void
2213 65 : bdev_init_complete(int rc)
2214 : {
2215 65 : spdk_bdev_init_cb cb_fn = g_init_cb_fn;
2216 65 : void *cb_arg = g_init_cb_arg;
2217 : struct spdk_bdev_module *m;
2218 :
2219 65 : g_bdev_mgr.init_complete = true;
2220 65 : g_init_cb_fn = NULL;
2221 65 : g_init_cb_arg = NULL;
2222 :
2223 : /*
2224 : * For modules that need to know when subsystem init is complete,
2225 : * inform them now.
2226 : */
2227 65 : if (rc == 0) {
2228 254 : TAILQ_FOREACH(m, &g_bdev_mgr.bdev_modules, internal.tailq) {
2229 189 : if (m->init_complete) {
2230 23 : m->init_complete();
2231 23 : }
2232 189 : }
2233 65 : }
2234 :
2235 65 : cb_fn(cb_arg, rc);
2236 65 : }
2237 :
2238 : static bool
2239 258 : bdev_module_all_actions_completed(void)
2240 : {
2241 : struct spdk_bdev_module *m;
2242 :
2243 1022 : TAILQ_FOREACH(m, &g_bdev_mgr.bdev_modules, internal.tailq) {
2244 764 : if (m->internal.action_in_progress > 0) {
2245 0 : return false;
2246 : }
2247 764 : }
2248 258 : return true;
2249 258 : }
2250 :
2251 : static void
2252 628 : bdev_module_action_complete(void)
2253 : {
2254 : /*
2255 : * Don't finish bdev subsystem initialization if
2256 : * module pre-initialization is still in progress, or
2257 : * the subsystem been already initialized.
2258 : */
2259 628 : if (!g_bdev_mgr.module_init_complete || g_bdev_mgr.init_complete) {
2260 563 : return;
2261 : }
2262 :
2263 : /*
2264 : * Check all bdev modules for inits/examinations in progress. If any
2265 : * exist, return immediately since we cannot finish bdev subsystem
2266 : * initialization until all are completed.
2267 : */
2268 65 : if (!bdev_module_all_actions_completed()) {
2269 0 : return;
2270 : }
2271 :
2272 : /*
2273 : * Modules already finished initialization - now that all
2274 : * the bdev modules have finished their asynchronous I/O
2275 : * processing, the entire bdev layer can be marked as complete.
2276 : */
2277 65 : bdev_init_complete(0);
2278 628 : }
2279 :
2280 : static void
2281 563 : bdev_module_action_done(struct spdk_bdev_module *module)
2282 : {
2283 563 : spdk_spin_lock(&module->internal.spinlock);
2284 563 : assert(module->internal.action_in_progress > 0);
2285 563 : module->internal.action_in_progress--;
2286 563 : spdk_spin_unlock(&module->internal.spinlock);
2287 563 : bdev_module_action_complete();
2288 563 : }
2289 :
2290 : void
2291 65 : spdk_bdev_module_init_done(struct spdk_bdev_module *module)
2292 : {
2293 65 : assert(module->async_init);
2294 65 : bdev_module_action_done(module);
2295 65 : }
2296 :
2297 : void
2298 498 : spdk_bdev_module_examine_done(struct spdk_bdev_module *module)
2299 : {
2300 498 : bdev_module_action_done(module);
2301 498 : }
2302 :
2303 : /** The last initialized bdev module */
2304 : static struct spdk_bdev_module *g_resume_bdev_module = NULL;
2305 :
2306 : static void
2307 0 : bdev_init_failed(void *cb_arg)
2308 : {
2309 0 : struct spdk_bdev_module *module = cb_arg;
2310 :
2311 0 : spdk_spin_lock(&module->internal.spinlock);
2312 0 : assert(module->internal.action_in_progress > 0);
2313 0 : module->internal.action_in_progress--;
2314 0 : spdk_spin_unlock(&module->internal.spinlock);
2315 0 : bdev_init_complete(-1);
2316 0 : }
2317 :
2318 : static int
2319 65 : bdev_modules_init(void)
2320 : {
2321 : struct spdk_bdev_module *module;
2322 65 : int rc = 0;
2323 :
2324 254 : TAILQ_FOREACH(module, &g_bdev_mgr.bdev_modules, internal.tailq) {
2325 189 : g_resume_bdev_module = module;
2326 189 : if (module->async_init) {
2327 65 : spdk_spin_lock(&module->internal.spinlock);
2328 65 : module->internal.action_in_progress = 1;
2329 65 : spdk_spin_unlock(&module->internal.spinlock);
2330 65 : }
2331 189 : rc = module->module_init();
2332 189 : if (rc != 0) {
2333 : /* Bump action_in_progress to prevent other modules from completion of modules_init
2334 : * Send message to defer application shutdown until resources are cleaned up */
2335 0 : spdk_spin_lock(&module->internal.spinlock);
2336 0 : module->internal.action_in_progress = 1;
2337 0 : spdk_spin_unlock(&module->internal.spinlock);
2338 0 : spdk_thread_send_msg(spdk_get_thread(), bdev_init_failed, module);
2339 0 : return rc;
2340 : }
2341 189 : }
2342 :
2343 65 : g_resume_bdev_module = NULL;
2344 65 : return 0;
2345 65 : }
2346 :
2347 : void
2348 65 : spdk_bdev_initialize(spdk_bdev_init_cb cb_fn, void *cb_arg)
2349 : {
2350 65 : int rc = 0;
2351 : char mempool_name[32];
2352 :
2353 65 : assert(cb_fn != NULL);
2354 :
2355 65 : g_init_cb_fn = cb_fn;
2356 65 : g_init_cb_arg = cb_arg;
2357 :
2358 65 : spdk_notify_type_register("bdev_register");
2359 65 : spdk_notify_type_register("bdev_unregister");
2360 :
2361 65 : snprintf(mempool_name, sizeof(mempool_name), "bdev_io_%d", getpid());
2362 :
2363 65 : rc = spdk_iobuf_register_module("bdev");
2364 65 : if (rc != 0) {
2365 0 : SPDK_ERRLOG("could not register bdev iobuf module: %s\n", spdk_strerror(-rc));
2366 0 : bdev_init_complete(-1);
2367 0 : return;
2368 : }
2369 :
2370 130 : g_bdev_mgr.bdev_io_pool = spdk_mempool_create(mempool_name,
2371 65 : g_bdev_opts.bdev_io_pool_size,
2372 65 : sizeof(struct spdk_bdev_io) +
2373 65 : bdev_module_get_max_ctx_size(),
2374 : 0,
2375 : SPDK_ENV_NUMA_ID_ANY);
2376 :
2377 65 : if (g_bdev_mgr.bdev_io_pool == NULL) {
2378 0 : SPDK_ERRLOG("could not allocate spdk_bdev_io pool\n");
2379 0 : bdev_init_complete(-1);
2380 0 : return;
2381 : }
2382 :
2383 65 : g_bdev_mgr.zero_buffer = spdk_zmalloc(ZERO_BUFFER_SIZE, ZERO_BUFFER_SIZE,
2384 : NULL, SPDK_ENV_LCORE_ID_ANY, SPDK_MALLOC_DMA);
2385 65 : if (!g_bdev_mgr.zero_buffer) {
2386 0 : SPDK_ERRLOG("create bdev zero buffer failed\n");
2387 0 : bdev_init_complete(-1);
2388 0 : return;
2389 : }
2390 :
2391 : #ifdef SPDK_CONFIG_VTUNE
2392 : g_bdev_mgr.domain = __itt_domain_create("spdk_bdev");
2393 : #endif
2394 :
2395 65 : spdk_io_device_register(&g_bdev_mgr, bdev_mgmt_channel_create,
2396 : bdev_mgmt_channel_destroy,
2397 : sizeof(struct spdk_bdev_mgmt_channel),
2398 : "bdev_mgr");
2399 :
2400 65 : rc = bdev_modules_init();
2401 65 : g_bdev_mgr.module_init_complete = true;
2402 65 : if (rc != 0) {
2403 0 : SPDK_ERRLOG("bdev modules init failed\n");
2404 0 : return;
2405 : }
2406 :
2407 65 : bdev_module_action_complete();
2408 65 : }
2409 :
2410 : static void
2411 65 : bdev_mgr_unregister_cb(void *io_device)
2412 : {
2413 65 : spdk_bdev_fini_cb cb_fn = g_fini_cb_fn;
2414 :
2415 65 : if (g_bdev_mgr.bdev_io_pool) {
2416 65 : if (spdk_mempool_count(g_bdev_mgr.bdev_io_pool) != g_bdev_opts.bdev_io_pool_size) {
2417 0 : SPDK_ERRLOG("bdev IO pool count is %zu but should be %u\n",
2418 : spdk_mempool_count(g_bdev_mgr.bdev_io_pool),
2419 : g_bdev_opts.bdev_io_pool_size);
2420 0 : }
2421 :
2422 65 : spdk_mempool_free(g_bdev_mgr.bdev_io_pool);
2423 65 : }
2424 :
2425 65 : spdk_free(g_bdev_mgr.zero_buffer);
2426 :
2427 65 : bdev_examine_allowlist_free();
2428 :
2429 65 : cb_fn(g_fini_cb_arg);
2430 65 : g_fini_cb_fn = NULL;
2431 65 : g_fini_cb_arg = NULL;
2432 65 : g_bdev_mgr.init_complete = false;
2433 65 : g_bdev_mgr.module_init_complete = false;
2434 65 : }
2435 :
2436 : static void
2437 65 : bdev_module_fini_iter(void *arg)
2438 : {
2439 : struct spdk_bdev_module *bdev_module;
2440 :
2441 : /* FIXME: Handling initialization failures is broken now,
2442 : * so we won't even try cleaning up after successfully
2443 : * initialized modules. if module_init_complete is false,
2444 : * just call spdk_bdev_mgr_unregister_cb
2445 : */
2446 65 : if (!g_bdev_mgr.module_init_complete) {
2447 0 : bdev_mgr_unregister_cb(NULL);
2448 0 : return;
2449 : }
2450 :
2451 : /* Start iterating from the last touched module */
2452 65 : if (!g_resume_bdev_module) {
2453 65 : bdev_module = TAILQ_LAST(&g_bdev_mgr.bdev_modules, bdev_module_list);
2454 65 : } else {
2455 0 : bdev_module = TAILQ_PREV(g_resume_bdev_module, bdev_module_list,
2456 : internal.tailq);
2457 : }
2458 :
2459 254 : while (bdev_module) {
2460 189 : if (bdev_module->async_fini) {
2461 : /* Save our place so we can resume later. We must
2462 : * save the variable here, before calling module_fini()
2463 : * below, because in some cases the module may immediately
2464 : * call spdk_bdev_module_fini_done() and re-enter
2465 : * this function to continue iterating. */
2466 0 : g_resume_bdev_module = bdev_module;
2467 0 : }
2468 :
2469 189 : if (bdev_module->module_fini) {
2470 189 : bdev_module->module_fini();
2471 189 : }
2472 :
2473 189 : if (bdev_module->async_fini) {
2474 0 : return;
2475 : }
2476 :
2477 189 : bdev_module = TAILQ_PREV(bdev_module, bdev_module_list,
2478 : internal.tailq);
2479 : }
2480 :
2481 65 : g_resume_bdev_module = NULL;
2482 65 : spdk_io_device_unregister(&g_bdev_mgr, bdev_mgr_unregister_cb);
2483 65 : }
2484 :
2485 : void
2486 0 : spdk_bdev_module_fini_done(void)
2487 : {
2488 0 : if (spdk_get_thread() != g_fini_thread) {
2489 0 : spdk_thread_send_msg(g_fini_thread, bdev_module_fini_iter, NULL);
2490 0 : } else {
2491 0 : bdev_module_fini_iter(NULL);
2492 : }
2493 0 : }
2494 :
2495 : static void
2496 65 : bdev_finish_unregister_bdevs_iter(void *cb_arg, int bdeverrno)
2497 : {
2498 65 : struct spdk_bdev *bdev = cb_arg;
2499 :
2500 65 : if (bdeverrno && bdev) {
2501 0 : SPDK_WARNLOG("Unable to unregister bdev '%s' during spdk_bdev_finish()\n",
2502 : bdev->name);
2503 :
2504 : /*
2505 : * Since the call to spdk_bdev_unregister() failed, we have no way to free this
2506 : * bdev; try to continue by manually removing this bdev from the list and continue
2507 : * with the next bdev in the list.
2508 : */
2509 0 : TAILQ_REMOVE(&g_bdev_mgr.bdevs, bdev, internal.link);
2510 0 : }
2511 :
2512 65 : if (TAILQ_EMPTY(&g_bdev_mgr.bdevs)) {
2513 65 : SPDK_DEBUGLOG(bdev, "Done unregistering bdevs\n");
2514 : /*
2515 : * Bdev module finish need to be deferred as we might be in the middle of some context
2516 : * (like bdev part free) that will use this bdev (or private bdev driver ctx data)
2517 : * after returning.
2518 : */
2519 65 : spdk_thread_send_msg(spdk_get_thread(), bdev_module_fini_iter, NULL);
2520 65 : return;
2521 : }
2522 :
2523 : /*
2524 : * Unregister last unclaimed bdev in the list, to ensure that bdev subsystem
2525 : * shutdown proceeds top-down. The goal is to give virtual bdevs an opportunity
2526 : * to detect clean shutdown as opposed to run-time hot removal of the underlying
2527 : * base bdevs.
2528 : *
2529 : * Also, walk the list in the reverse order.
2530 : */
2531 0 : for (bdev = TAILQ_LAST(&g_bdev_mgr.bdevs, spdk_bdev_list);
2532 0 : bdev; bdev = TAILQ_PREV(bdev, spdk_bdev_list, internal.link)) {
2533 0 : spdk_spin_lock(&bdev->internal.spinlock);
2534 0 : if (bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE) {
2535 0 : LOG_ALREADY_CLAIMED_DEBUG("claimed, skipping", bdev);
2536 0 : spdk_spin_unlock(&bdev->internal.spinlock);
2537 0 : continue;
2538 : }
2539 0 : spdk_spin_unlock(&bdev->internal.spinlock);
2540 :
2541 0 : SPDK_DEBUGLOG(bdev, "Unregistering bdev '%s'\n", bdev->name);
2542 0 : spdk_bdev_unregister(bdev, bdev_finish_unregister_bdevs_iter, bdev);
2543 0 : return;
2544 : }
2545 :
2546 : /*
2547 : * If any bdev fails to unclaim underlying bdev properly, we may face the
2548 : * case of bdev list consisting of claimed bdevs only (if claims are managed
2549 : * correctly, this would mean there's a loop in the claims graph which is
2550 : * clearly impossible). Warn and unregister last bdev on the list then.
2551 : */
2552 0 : for (bdev = TAILQ_LAST(&g_bdev_mgr.bdevs, spdk_bdev_list);
2553 0 : bdev; bdev = TAILQ_PREV(bdev, spdk_bdev_list, internal.link)) {
2554 0 : SPDK_WARNLOG("Unregistering claimed bdev '%s'!\n", bdev->name);
2555 0 : spdk_bdev_unregister(bdev, bdev_finish_unregister_bdevs_iter, bdev);
2556 0 : return;
2557 : }
2558 65 : }
2559 :
2560 : static void
2561 65 : bdev_module_fini_start_iter(void *arg)
2562 : {
2563 : struct spdk_bdev_module *bdev_module;
2564 :
2565 65 : if (!g_resume_bdev_module) {
2566 65 : bdev_module = TAILQ_LAST(&g_bdev_mgr.bdev_modules, bdev_module_list);
2567 65 : } else {
2568 0 : bdev_module = TAILQ_PREV(g_resume_bdev_module, bdev_module_list, internal.tailq);
2569 : }
2570 :
2571 254 : while (bdev_module) {
2572 189 : if (bdev_module->async_fini_start) {
2573 : /* Save our place so we can resume later. We must
2574 : * save the variable here, before calling fini_start()
2575 : * below, because in some cases the module may immediately
2576 : * call spdk_bdev_module_fini_start_done() and re-enter
2577 : * this function to continue iterating. */
2578 0 : g_resume_bdev_module = bdev_module;
2579 0 : }
2580 :
2581 189 : if (bdev_module->fini_start) {
2582 23 : bdev_module->fini_start();
2583 23 : }
2584 :
2585 189 : if (bdev_module->async_fini_start) {
2586 0 : return;
2587 : }
2588 :
2589 189 : bdev_module = TAILQ_PREV(bdev_module, bdev_module_list, internal.tailq);
2590 : }
2591 :
2592 65 : g_resume_bdev_module = NULL;
2593 :
2594 65 : bdev_finish_unregister_bdevs_iter(NULL, 0);
2595 65 : }
2596 :
2597 : void
2598 0 : spdk_bdev_module_fini_start_done(void)
2599 : {
2600 0 : if (spdk_get_thread() != g_fini_thread) {
2601 0 : spdk_thread_send_msg(g_fini_thread, bdev_module_fini_start_iter, NULL);
2602 0 : } else {
2603 0 : bdev_module_fini_start_iter(NULL);
2604 : }
2605 0 : }
2606 :
2607 : static void
2608 65 : bdev_finish_wait_for_examine_done(void *cb_arg)
2609 : {
2610 65 : bdev_module_fini_start_iter(NULL);
2611 65 : }
2612 :
2613 : static void bdev_open_async_fini(void);
2614 :
2615 : void
2616 65 : spdk_bdev_finish(spdk_bdev_fini_cb cb_fn, void *cb_arg)
2617 : {
2618 : int rc;
2619 :
2620 65 : assert(cb_fn != NULL);
2621 :
2622 65 : g_fini_thread = spdk_get_thread();
2623 :
2624 65 : g_fini_cb_fn = cb_fn;
2625 65 : g_fini_cb_arg = cb_arg;
2626 :
2627 65 : bdev_open_async_fini();
2628 :
2629 65 : rc = spdk_bdev_wait_for_examine(bdev_finish_wait_for_examine_done, NULL);
2630 65 : if (rc != 0) {
2631 0 : SPDK_ERRLOG("wait_for_examine failed: %s\n", spdk_strerror(-rc));
2632 0 : bdev_finish_wait_for_examine_done(NULL);
2633 0 : }
2634 65 : }
2635 :
2636 : struct spdk_bdev_io *
2637 697 : bdev_channel_get_io(struct spdk_bdev_channel *channel)
2638 : {
2639 697 : struct spdk_bdev_mgmt_channel *ch = channel->shared_resource->mgmt_ch;
2640 : struct spdk_bdev_io *bdev_io;
2641 :
2642 697 : if (ch->per_thread_cache_count > 0) {
2643 637 : bdev_io = STAILQ_FIRST(&ch->per_thread_cache);
2644 637 : STAILQ_REMOVE_HEAD(&ch->per_thread_cache, internal.buf_link);
2645 637 : ch->per_thread_cache_count--;
2646 697 : } else if (spdk_unlikely(!TAILQ_EMPTY(&ch->io_wait_queue))) {
2647 : /*
2648 : * Don't try to look for bdev_ios in the global pool if there are
2649 : * waiters on bdev_ios - we don't want this caller to jump the line.
2650 : */
2651 0 : bdev_io = NULL;
2652 0 : } else {
2653 60 : bdev_io = spdk_mempool_get(g_bdev_mgr.bdev_io_pool);
2654 : }
2655 :
2656 697 : return bdev_io;
2657 : }
2658 :
2659 : void
2660 691 : spdk_bdev_free_io(struct spdk_bdev_io *bdev_io)
2661 : {
2662 : struct spdk_bdev_mgmt_channel *ch;
2663 :
2664 691 : assert(bdev_io != NULL);
2665 691 : assert(bdev_io->internal.status != SPDK_BDEV_IO_STATUS_PENDING);
2666 :
2667 691 : ch = bdev_io->internal.ch->shared_resource->mgmt_ch;
2668 :
2669 691 : if (bdev_io->internal.f.has_buf) {
2670 16 : bdev_io_put_buf(bdev_io);
2671 16 : }
2672 :
2673 691 : if (ch->per_thread_cache_count < ch->bdev_io_cache_size) {
2674 637 : ch->per_thread_cache_count++;
2675 637 : STAILQ_INSERT_HEAD(&ch->per_thread_cache, bdev_io, internal.buf_link);
2676 641 : while (ch->per_thread_cache_count > 0 && !TAILQ_EMPTY(&ch->io_wait_queue)) {
2677 : struct spdk_bdev_io_wait_entry *entry;
2678 :
2679 4 : entry = TAILQ_FIRST(&ch->io_wait_queue);
2680 4 : TAILQ_REMOVE(&ch->io_wait_queue, entry, link);
2681 4 : entry->cb_fn(entry->cb_arg);
2682 : }
2683 637 : } else {
2684 : /* We should never have a full cache with entries on the io wait queue. */
2685 54 : assert(TAILQ_EMPTY(&ch->io_wait_queue));
2686 54 : spdk_mempool_put(g_bdev_mgr.bdev_io_pool, (void *)bdev_io);
2687 : }
2688 691 : }
2689 :
2690 : static bool
2691 72 : bdev_qos_is_iops_rate_limit(enum spdk_bdev_qos_rate_limit_type limit)
2692 : {
2693 72 : assert(limit != SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES);
2694 :
2695 72 : switch (limit) {
2696 : case SPDK_BDEV_QOS_RW_IOPS_RATE_LIMIT:
2697 18 : return true;
2698 : case SPDK_BDEV_QOS_RW_BPS_RATE_LIMIT:
2699 : case SPDK_BDEV_QOS_R_BPS_RATE_LIMIT:
2700 : case SPDK_BDEV_QOS_W_BPS_RATE_LIMIT:
2701 54 : return false;
2702 0 : case SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES:
2703 : default:
2704 0 : return false;
2705 : }
2706 72 : }
2707 :
2708 : static bool
2709 25 : bdev_qos_io_to_limit(struct spdk_bdev_io *bdev_io)
2710 : {
2711 25 : switch (bdev_io->type) {
2712 : case SPDK_BDEV_IO_TYPE_NVME_IO:
2713 : case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
2714 : case SPDK_BDEV_IO_TYPE_READ:
2715 : case SPDK_BDEV_IO_TYPE_WRITE:
2716 23 : return true;
2717 : case SPDK_BDEV_IO_TYPE_ZCOPY:
2718 0 : if (bdev_io->u.bdev.zcopy.start) {
2719 0 : return true;
2720 : } else {
2721 0 : return false;
2722 : }
2723 : default:
2724 2 : return false;
2725 : }
2726 25 : }
2727 :
2728 : static bool
2729 33 : bdev_is_read_io(struct spdk_bdev_io *bdev_io)
2730 : {
2731 33 : switch (bdev_io->type) {
2732 : case SPDK_BDEV_IO_TYPE_NVME_IO:
2733 : case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
2734 : /* Bit 1 (0x2) set for read operation */
2735 0 : if (bdev_io->u.nvme_passthru.cmd.opc & SPDK_NVME_OPC_READ) {
2736 0 : return true;
2737 : } else {
2738 0 : return false;
2739 : }
2740 : case SPDK_BDEV_IO_TYPE_READ:
2741 30 : return true;
2742 : case SPDK_BDEV_IO_TYPE_ZCOPY:
2743 : /* Populate to read from disk */
2744 0 : if (bdev_io->u.bdev.zcopy.populate) {
2745 0 : return true;
2746 : } else {
2747 0 : return false;
2748 : }
2749 : default:
2750 3 : return false;
2751 : }
2752 33 : }
2753 :
2754 : static uint64_t
2755 43 : bdev_get_io_size_in_byte(struct spdk_bdev_io *bdev_io)
2756 : {
2757 43 : uint32_t blocklen = bdev_io_get_block_size(bdev_io);
2758 :
2759 43 : switch (bdev_io->type) {
2760 : case SPDK_BDEV_IO_TYPE_NVME_IO:
2761 : case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
2762 0 : return bdev_io->u.nvme_passthru.nbytes;
2763 : case SPDK_BDEV_IO_TYPE_READ:
2764 : case SPDK_BDEV_IO_TYPE_WRITE:
2765 43 : return bdev_io->u.bdev.num_blocks * blocklen;
2766 : case SPDK_BDEV_IO_TYPE_ZCOPY:
2767 : /* Track the data in the start phase only */
2768 0 : if (bdev_io->u.bdev.zcopy.start) {
2769 0 : return bdev_io->u.bdev.num_blocks * blocklen;
2770 : } else {
2771 0 : return 0;
2772 : }
2773 : default:
2774 0 : return 0;
2775 : }
2776 43 : }
2777 :
2778 : static inline bool
2779 64 : bdev_qos_rw_queue_io(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io, uint64_t delta)
2780 : {
2781 : int64_t remaining_this_timeslice;
2782 :
2783 64 : if (!limit->max_per_timeslice) {
2784 : /* The QoS is disabled */
2785 0 : return false;
2786 : }
2787 :
2788 64 : remaining_this_timeslice = __atomic_sub_fetch(&limit->remaining_this_timeslice, delta,
2789 : __ATOMIC_RELAXED);
2790 64 : if (remaining_this_timeslice + (int64_t)delta > 0) {
2791 : /* There was still a quota for this delta -> the IO shouldn't be queued
2792 : *
2793 : * We allow a slight quota overrun here so an IO bigger than the per-timeslice
2794 : * quota can be allowed once a while. Such overrun then taken into account in
2795 : * the QoS poller, where the next timeslice quota is calculated.
2796 : */
2797 59 : return false;
2798 : }
2799 :
2800 : /* There was no quota for this delta -> the IO should be queued
2801 : * The remaining_this_timeslice must be rewinded so it reflects the real
2802 : * amount of IOs or bytes allowed.
2803 : */
2804 5 : __atomic_add_fetch(
2805 5 : &limit->remaining_this_timeslice, delta, __ATOMIC_RELAXED);
2806 5 : return true;
2807 64 : }
2808 :
2809 : static inline void
2810 5 : bdev_qos_rw_rewind_io(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io, uint64_t delta)
2811 : {
2812 5 : __atomic_add_fetch(&limit->remaining_this_timeslice, delta, __ATOMIC_RELAXED);
2813 5 : }
2814 :
2815 : static bool
2816 23 : bdev_qos_rw_iops_queue(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2817 : {
2818 23 : return bdev_qos_rw_queue_io(limit, io, 1);
2819 : }
2820 :
2821 : static void
2822 3 : bdev_qos_rw_iops_rewind_quota(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2823 : {
2824 3 : bdev_qos_rw_rewind_io(limit, io, 1);
2825 3 : }
2826 :
2827 : static bool
2828 41 : bdev_qos_rw_bps_queue(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2829 : {
2830 41 : return bdev_qos_rw_queue_io(limit, io, bdev_get_io_size_in_byte(io));
2831 : }
2832 :
2833 : static void
2834 2 : bdev_qos_rw_bps_rewind_quota(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2835 : {
2836 2 : bdev_qos_rw_rewind_io(limit, io, bdev_get_io_size_in_byte(io));
2837 2 : }
2838 :
2839 : static bool
2840 19 : bdev_qos_r_bps_queue(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2841 : {
2842 19 : if (bdev_is_read_io(io) == false) {
2843 1 : return false;
2844 : }
2845 :
2846 18 : return bdev_qos_rw_bps_queue(limit, io);
2847 19 : }
2848 :
2849 : static void
2850 0 : bdev_qos_r_bps_rewind_quota(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2851 : {
2852 0 : if (bdev_is_read_io(io) != false) {
2853 0 : bdev_qos_rw_rewind_io(limit, io, bdev_get_io_size_in_byte(io));
2854 0 : }
2855 0 : }
2856 :
2857 : static bool
2858 14 : bdev_qos_w_bps_queue(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2859 : {
2860 14 : if (bdev_is_read_io(io) == true) {
2861 12 : return false;
2862 : }
2863 :
2864 2 : return bdev_qos_rw_bps_queue(limit, io);
2865 14 : }
2866 :
2867 : static void
2868 0 : bdev_qos_w_bps_rewind_quota(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2869 : {
2870 0 : if (bdev_is_read_io(io) != true) {
2871 0 : bdev_qos_rw_rewind_io(limit, io, bdev_get_io_size_in_byte(io));
2872 0 : }
2873 0 : }
2874 :
2875 : static void
2876 10 : bdev_qos_set_ops(struct spdk_bdev_qos *qos)
2877 : {
2878 : int i;
2879 :
2880 50 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
2881 40 : if (qos->rate_limits[i].limit == SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
2882 15 : qos->rate_limits[i].queue_io = NULL;
2883 15 : continue;
2884 : }
2885 :
2886 25 : switch (i) {
2887 : case SPDK_BDEV_QOS_RW_IOPS_RATE_LIMIT:
2888 9 : qos->rate_limits[i].queue_io = bdev_qos_rw_iops_queue;
2889 9 : qos->rate_limits[i].rewind_quota = bdev_qos_rw_iops_rewind_quota;
2890 9 : break;
2891 : case SPDK_BDEV_QOS_RW_BPS_RATE_LIMIT:
2892 7 : qos->rate_limits[i].queue_io = bdev_qos_rw_bps_queue;
2893 7 : qos->rate_limits[i].rewind_quota = bdev_qos_rw_bps_rewind_quota;
2894 7 : break;
2895 : case SPDK_BDEV_QOS_R_BPS_RATE_LIMIT:
2896 5 : qos->rate_limits[i].queue_io = bdev_qos_r_bps_queue;
2897 5 : qos->rate_limits[i].rewind_quota = bdev_qos_r_bps_rewind_quota;
2898 5 : break;
2899 : case SPDK_BDEV_QOS_W_BPS_RATE_LIMIT:
2900 4 : qos->rate_limits[i].queue_io = bdev_qos_w_bps_queue;
2901 4 : qos->rate_limits[i].rewind_quota = bdev_qos_w_bps_rewind_quota;
2902 4 : break;
2903 : default:
2904 0 : break;
2905 : }
2906 25 : }
2907 10 : }
2908 :
2909 : static void
2910 6 : _bdev_io_complete_in_submit(struct spdk_bdev_channel *bdev_ch,
2911 : struct spdk_bdev_io *bdev_io,
2912 : enum spdk_bdev_io_status status)
2913 : {
2914 6 : bdev_io->internal.f.in_submit_request = true;
2915 6 : bdev_io_increment_outstanding(bdev_ch, bdev_ch->shared_resource);
2916 6 : spdk_bdev_io_complete(bdev_io, status);
2917 6 : bdev_io->internal.f.in_submit_request = false;
2918 6 : }
2919 :
2920 : static inline void
2921 574 : bdev_io_do_submit(struct spdk_bdev_channel *bdev_ch, struct spdk_bdev_io *bdev_io)
2922 : {
2923 574 : struct spdk_bdev *bdev = bdev_io->bdev;
2924 574 : struct spdk_io_channel *ch = bdev_ch->channel;
2925 574 : struct spdk_bdev_shared_resource *shared_resource = bdev_ch->shared_resource;
2926 :
2927 574 : if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_ABORT)) {
2928 16 : struct spdk_bdev_mgmt_channel *mgmt_channel = shared_resource->mgmt_ch;
2929 16 : struct spdk_bdev_io *bio_to_abort = bdev_io->u.abort.bio_to_abort;
2930 :
2931 16 : if (bdev_abort_queued_io(&shared_resource->nomem_io, bio_to_abort) ||
2932 16 : bdev_abort_buf_io(mgmt_channel, bio_to_abort)) {
2933 0 : _bdev_io_complete_in_submit(bdev_ch, bdev_io,
2934 : SPDK_BDEV_IO_STATUS_SUCCESS);
2935 0 : return;
2936 : }
2937 16 : }
2938 :
2939 574 : if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE &&
2940 : bdev_io->bdev->split_on_write_unit &&
2941 : bdev_io->u.bdev.num_blocks < bdev_io->bdev->write_unit_size)) {
2942 4 : SPDK_ERRLOG("IO num_blocks %lu does not match the write_unit_size %u\n",
2943 : bdev_io->u.bdev.num_blocks, bdev_io->bdev->write_unit_size);
2944 4 : _bdev_io_complete_in_submit(bdev_ch, bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
2945 4 : return;
2946 : }
2947 :
2948 570 : if (spdk_likely(TAILQ_EMPTY(&shared_resource->nomem_io))) {
2949 527 : bdev_io_increment_outstanding(bdev_ch, shared_resource);
2950 527 : bdev_io->internal.f.in_submit_request = true;
2951 527 : bdev_submit_request(bdev, ch, bdev_io);
2952 527 : bdev_io->internal.f.in_submit_request = false;
2953 527 : } else {
2954 43 : bdev_queue_nomem_io_tail(shared_resource, bdev_io, BDEV_IO_RETRY_STATE_SUBMIT);
2955 43 : if (shared_resource->nomem_threshold == 0 && shared_resource->io_outstanding == 0) {
2956 : /* Special case when we have nomem IOs and no outstanding IOs which completions
2957 : * could trigger retry of queued IOs */
2958 0 : bdev_shared_ch_retry_io(shared_resource);
2959 0 : }
2960 : }
2961 574 : }
2962 :
2963 : static bool
2964 25 : bdev_qos_queue_io(struct spdk_bdev_qos *qos, struct spdk_bdev_io *bdev_io)
2965 : {
2966 : int i;
2967 :
2968 25 : if (bdev_qos_io_to_limit(bdev_io) == true) {
2969 100 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
2970 82 : if (!qos->rate_limits[i].queue_io) {
2971 5 : continue;
2972 : }
2973 :
2974 231 : if (qos->rate_limits[i].queue_io(&qos->rate_limits[i],
2975 154 : bdev_io) == true) {
2976 10 : for (i -= 1; i >= 0 ; i--) {
2977 5 : if (!qos->rate_limits[i].queue_io) {
2978 0 : continue;
2979 : }
2980 :
2981 5 : qos->rate_limits[i].rewind_quota(&qos->rate_limits[i], bdev_io);
2982 5 : }
2983 5 : return true;
2984 : }
2985 72 : }
2986 18 : }
2987 :
2988 20 : return false;
2989 25 : }
2990 :
2991 : static int
2992 27 : bdev_qos_io_submit(struct spdk_bdev_channel *ch, struct spdk_bdev_qos *qos)
2993 : {
2994 27 : struct spdk_bdev_io *bdev_io = NULL, *tmp = NULL;
2995 27 : int submitted_ios = 0;
2996 :
2997 52 : TAILQ_FOREACH_SAFE(bdev_io, &ch->qos_queued_io, internal.link, tmp) {
2998 25 : if (!bdev_qos_queue_io(qos, bdev_io)) {
2999 20 : TAILQ_REMOVE(&ch->qos_queued_io, bdev_io, internal.link);
3000 20 : bdev_io_do_submit(ch, bdev_io);
3001 :
3002 20 : submitted_ios++;
3003 20 : }
3004 25 : }
3005 :
3006 27 : return submitted_ios;
3007 : }
3008 :
3009 : static void
3010 2 : bdev_queue_io_wait_with_cb(struct spdk_bdev_io *bdev_io, spdk_bdev_io_wait_cb cb_fn)
3011 : {
3012 : int rc;
3013 :
3014 2 : bdev_io->internal.waitq_entry.bdev = bdev_io->bdev;
3015 2 : bdev_io->internal.waitq_entry.cb_fn = cb_fn;
3016 2 : bdev_io->internal.waitq_entry.cb_arg = bdev_io;
3017 4 : rc = spdk_bdev_queue_io_wait(bdev_io->bdev, spdk_io_channel_from_ctx(bdev_io->internal.ch),
3018 2 : &bdev_io->internal.waitq_entry);
3019 2 : if (rc != 0) {
3020 0 : SPDK_ERRLOG("Queue IO failed, rc=%d\n", rc);
3021 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
3022 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
3023 0 : }
3024 2 : }
3025 :
3026 : static bool
3027 621 : bdev_rw_should_split(struct spdk_bdev_io *bdev_io)
3028 : {
3029 : uint32_t io_boundary;
3030 621 : struct spdk_bdev *bdev = bdev_io->bdev;
3031 621 : uint32_t max_segment_size = bdev->max_segment_size;
3032 621 : uint32_t max_size = bdev->max_rw_size;
3033 621 : int max_segs = bdev->max_num_segments;
3034 :
3035 621 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE && bdev->split_on_write_unit) {
3036 24 : io_boundary = bdev->write_unit_size;
3037 621 : } else if (bdev->split_on_optimal_io_boundary) {
3038 168 : io_boundary = bdev->optimal_io_boundary;
3039 168 : } else {
3040 429 : io_boundary = 0;
3041 : }
3042 :
3043 621 : if (spdk_likely(!io_boundary && !max_segs && !max_segment_size && !max_size)) {
3044 243 : return false;
3045 : }
3046 :
3047 378 : if (io_boundary) {
3048 : uint64_t start_stripe, end_stripe;
3049 :
3050 192 : start_stripe = bdev_io->u.bdev.offset_blocks;
3051 192 : end_stripe = start_stripe + bdev_io->u.bdev.num_blocks - 1;
3052 : /* Avoid expensive div operations if possible. These spdk_u32 functions are very cheap. */
3053 192 : if (spdk_likely(spdk_u32_is_pow2(io_boundary))) {
3054 192 : start_stripe >>= spdk_u32log2(io_boundary);
3055 192 : end_stripe >>= spdk_u32log2(io_boundary);
3056 192 : } else {
3057 0 : start_stripe /= io_boundary;
3058 0 : end_stripe /= io_boundary;
3059 : }
3060 :
3061 192 : if (start_stripe != end_stripe) {
3062 75 : return true;
3063 : }
3064 117 : }
3065 :
3066 303 : if (max_segs) {
3067 150 : if (bdev_io->u.bdev.iovcnt > max_segs) {
3068 15 : return true;
3069 : }
3070 135 : }
3071 :
3072 288 : if (max_segment_size) {
3073 470 : for (int i = 0; i < bdev_io->u.bdev.iovcnt; i++) {
3074 346 : if (bdev_io->u.bdev.iovs[i].iov_len > max_segment_size) {
3075 12 : return true;
3076 : }
3077 334 : }
3078 124 : }
3079 :
3080 276 : if (max_size) {
3081 52 : if (bdev_io->u.bdev.num_blocks > max_size) {
3082 7 : return true;
3083 : }
3084 45 : }
3085 :
3086 269 : return false;
3087 621 : }
3088 :
3089 : static bool
3090 24 : bdev_unmap_should_split(struct spdk_bdev_io *bdev_io)
3091 : {
3092 : uint32_t num_unmap_segments;
3093 :
3094 24 : if (!bdev_io->bdev->max_unmap || !bdev_io->bdev->max_unmap_segments) {
3095 3 : return false;
3096 : }
3097 21 : num_unmap_segments = spdk_divide_round_up(bdev_io->u.bdev.num_blocks, bdev_io->bdev->max_unmap);
3098 21 : if (num_unmap_segments > bdev_io->bdev->max_unmap_segments) {
3099 4 : return true;
3100 : }
3101 :
3102 17 : return false;
3103 24 : }
3104 :
3105 : static bool
3106 37 : bdev_write_zeroes_should_split(struct spdk_bdev_io *bdev_io)
3107 : {
3108 37 : if (!bdev_io->bdev->max_write_zeroes) {
3109 4 : return false;
3110 : }
3111 :
3112 33 : if (bdev_io->u.bdev.num_blocks > bdev_io->bdev->max_write_zeroes) {
3113 10 : return true;
3114 : }
3115 :
3116 23 : return false;
3117 37 : }
3118 :
3119 : static bool
3120 30 : bdev_copy_should_split(struct spdk_bdev_io *bdev_io)
3121 : {
3122 30 : if (bdev_io->bdev->max_copy != 0 &&
3123 25 : bdev_io->u.bdev.num_blocks > bdev_io->bdev->max_copy) {
3124 6 : return true;
3125 : }
3126 :
3127 24 : return false;
3128 30 : }
3129 :
3130 : static bool
3131 792 : bdev_io_should_split(struct spdk_bdev_io *bdev_io)
3132 : {
3133 792 : switch (bdev_io->type) {
3134 : case SPDK_BDEV_IO_TYPE_READ:
3135 : case SPDK_BDEV_IO_TYPE_WRITE:
3136 621 : return bdev_rw_should_split(bdev_io);
3137 : case SPDK_BDEV_IO_TYPE_UNMAP:
3138 24 : return bdev_unmap_should_split(bdev_io);
3139 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3140 37 : return bdev_write_zeroes_should_split(bdev_io);
3141 : case SPDK_BDEV_IO_TYPE_COPY:
3142 30 : return bdev_copy_should_split(bdev_io);
3143 : default:
3144 80 : return false;
3145 : }
3146 792 : }
3147 :
3148 : static uint32_t
3149 249 : _to_next_boundary(uint64_t offset, uint32_t boundary)
3150 : {
3151 249 : return (boundary - (offset % boundary));
3152 : }
3153 :
3154 : static void bdev_io_split_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg);
3155 :
3156 : static void _bdev_rw_split(void *_bdev_io);
3157 :
3158 : static void bdev_unmap_split(struct spdk_bdev_io *bdev_io);
3159 :
3160 : static void
3161 0 : _bdev_unmap_split(void *_bdev_io)
3162 : {
3163 0 : return bdev_unmap_split((struct spdk_bdev_io *)_bdev_io);
3164 : }
3165 :
3166 : static void bdev_write_zeroes_split(struct spdk_bdev_io *bdev_io);
3167 :
3168 : static void
3169 0 : _bdev_write_zeroes_split(void *_bdev_io)
3170 : {
3171 0 : return bdev_write_zeroes_split((struct spdk_bdev_io *)_bdev_io);
3172 : }
3173 :
3174 : static void bdev_copy_split(struct spdk_bdev_io *bdev_io);
3175 :
3176 : static void
3177 0 : _bdev_copy_split(void *_bdev_io)
3178 : {
3179 0 : return bdev_copy_split((struct spdk_bdev_io *)_bdev_io);
3180 : }
3181 :
3182 : static int
3183 305 : bdev_io_split_submit(struct spdk_bdev_io *bdev_io, struct iovec *iov, int iovcnt, void *md_buf,
3184 : uint64_t num_blocks, uint64_t *offset, uint64_t *remaining)
3185 : {
3186 : int rc;
3187 : uint64_t current_offset, current_remaining, current_src_offset;
3188 : spdk_bdev_io_wait_cb io_wait_fn;
3189 :
3190 305 : current_offset = *offset;
3191 305 : current_remaining = *remaining;
3192 :
3193 305 : assert(bdev_io->internal.f.split);
3194 :
3195 305 : bdev_io->internal.split.outstanding++;
3196 :
3197 305 : io_wait_fn = _bdev_rw_split;
3198 305 : switch (bdev_io->type) {
3199 : case SPDK_BDEV_IO_TYPE_READ:
3200 196 : assert(bdev_io->u.bdev.accel_sequence == NULL);
3201 392 : rc = bdev_readv_blocks_with_md(bdev_io->internal.desc,
3202 196 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
3203 196 : iov, iovcnt, md_buf, current_offset,
3204 196 : num_blocks,
3205 196 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain : NULL,
3206 196 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain_ctx : NULL,
3207 : NULL,
3208 196 : bdev_io->u.bdev.dif_check_flags,
3209 196 : bdev_io_split_done, bdev_io);
3210 196 : break;
3211 : case SPDK_BDEV_IO_TYPE_WRITE:
3212 50 : assert(bdev_io->u.bdev.accel_sequence == NULL);
3213 100 : rc = bdev_writev_blocks_with_md(bdev_io->internal.desc,
3214 50 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
3215 50 : iov, iovcnt, md_buf, current_offset,
3216 50 : num_blocks,
3217 50 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain : NULL,
3218 50 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain_ctx : NULL,
3219 : NULL,
3220 50 : bdev_io->u.bdev.dif_check_flags,
3221 50 : bdev_io->u.bdev.nvme_cdw12.raw,
3222 50 : bdev_io->u.bdev.nvme_cdw13.raw,
3223 50 : bdev_io_split_done, bdev_io);
3224 50 : break;
3225 : case SPDK_BDEV_IO_TYPE_UNMAP:
3226 17 : io_wait_fn = _bdev_unmap_split;
3227 34 : rc = spdk_bdev_unmap_blocks(bdev_io->internal.desc,
3228 17 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
3229 17 : current_offset, num_blocks,
3230 17 : bdev_io_split_done, bdev_io);
3231 17 : break;
3232 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3233 23 : io_wait_fn = _bdev_write_zeroes_split;
3234 46 : rc = spdk_bdev_write_zeroes_blocks(bdev_io->internal.desc,
3235 23 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
3236 23 : current_offset, num_blocks,
3237 23 : bdev_io_split_done, bdev_io);
3238 23 : break;
3239 : case SPDK_BDEV_IO_TYPE_COPY:
3240 19 : io_wait_fn = _bdev_copy_split;
3241 38 : current_src_offset = bdev_io->u.bdev.copy.src_offset_blocks +
3242 19 : (current_offset - bdev_io->u.bdev.offset_blocks);
3243 38 : rc = spdk_bdev_copy_blocks(bdev_io->internal.desc,
3244 19 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
3245 19 : current_offset, current_src_offset, num_blocks,
3246 19 : bdev_io_split_done, bdev_io);
3247 19 : break;
3248 : default:
3249 0 : assert(false);
3250 : rc = -EINVAL;
3251 : break;
3252 : }
3253 :
3254 305 : if (rc == 0) {
3255 301 : current_offset += num_blocks;
3256 301 : current_remaining -= num_blocks;
3257 301 : bdev_io->internal.split.current_offset_blocks = current_offset;
3258 301 : bdev_io->internal.split.remaining_num_blocks = current_remaining;
3259 301 : *offset = current_offset;
3260 301 : *remaining = current_remaining;
3261 301 : } else {
3262 4 : bdev_io->internal.split.outstanding--;
3263 4 : if (rc == -ENOMEM) {
3264 4 : if (bdev_io->internal.split.outstanding == 0) {
3265 : /* No I/O is outstanding. Hence we should wait here. */
3266 1 : bdev_queue_io_wait_with_cb(bdev_io, io_wait_fn);
3267 1 : }
3268 4 : } else {
3269 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
3270 0 : if (bdev_io->internal.split.outstanding == 0) {
3271 0 : bdev_ch_remove_from_io_submitted(bdev_io);
3272 0 : spdk_trace_record(TRACE_BDEV_IO_DONE, bdev_io->internal.ch->trace_id,
3273 : 0, (uintptr_t)bdev_io, bdev_io->internal.caller_ctx,
3274 : bdev_io->internal.ch->queue_depth);
3275 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
3276 0 : }
3277 : }
3278 : }
3279 :
3280 305 : return rc;
3281 : }
3282 :
3283 : static void
3284 67 : _bdev_rw_split(void *_bdev_io)
3285 : {
3286 : struct iovec *parent_iov, *iov;
3287 67 : struct spdk_bdev_io *bdev_io = _bdev_io;
3288 67 : struct spdk_bdev *bdev = bdev_io->bdev;
3289 : uint64_t parent_offset, current_offset, remaining;
3290 : uint32_t parent_iov_offset, parent_iovcnt, parent_iovpos, child_iovcnt;
3291 : uint32_t to_next_boundary, to_next_boundary_bytes, to_last_block_bytes;
3292 : uint32_t iovcnt, iov_len, child_iovsize;
3293 : uint32_t blocklen;
3294 : uint32_t io_boundary;
3295 67 : uint32_t max_segment_size = bdev->max_segment_size;
3296 67 : uint32_t max_child_iovcnt = bdev->max_num_segments;
3297 67 : uint32_t max_size = bdev->max_rw_size;
3298 67 : void *md_buf = NULL;
3299 : int rc;
3300 :
3301 67 : blocklen = bdev_io_get_block_size(bdev_io);
3302 :
3303 67 : max_size = max_size ? max_size : UINT32_MAX;
3304 67 : max_segment_size = max_segment_size ? max_segment_size : UINT32_MAX;
3305 67 : max_child_iovcnt = max_child_iovcnt ? spdk_min(max_child_iovcnt, SPDK_BDEV_IO_NUM_CHILD_IOV) :
3306 : SPDK_BDEV_IO_NUM_CHILD_IOV;
3307 :
3308 67 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE && bdev->split_on_write_unit) {
3309 5 : io_boundary = bdev->write_unit_size;
3310 67 : } else if (bdev->split_on_optimal_io_boundary) {
3311 40 : io_boundary = bdev->optimal_io_boundary;
3312 40 : } else {
3313 22 : io_boundary = UINT32_MAX;
3314 : }
3315 :
3316 67 : assert(bdev_io->internal.f.split);
3317 :
3318 67 : remaining = bdev_io->internal.split.remaining_num_blocks;
3319 67 : current_offset = bdev_io->internal.split.current_offset_blocks;
3320 67 : parent_offset = bdev_io->u.bdev.offset_blocks;
3321 67 : parent_iov_offset = (current_offset - parent_offset) * blocklen;
3322 67 : parent_iovcnt = bdev_io->u.bdev.iovcnt;
3323 :
3324 420 : for (parent_iovpos = 0; parent_iovpos < parent_iovcnt; parent_iovpos++) {
3325 420 : parent_iov = &bdev_io->u.bdev.iovs[parent_iovpos];
3326 420 : if (parent_iov_offset < parent_iov->iov_len) {
3327 67 : break;
3328 : }
3329 353 : parent_iov_offset -= parent_iov->iov_len;
3330 353 : }
3331 :
3332 67 : child_iovcnt = 0;
3333 573 : while (remaining > 0 && parent_iovpos < parent_iovcnt &&
3334 264 : child_iovcnt < SPDK_BDEV_IO_NUM_CHILD_IOV) {
3335 249 : to_next_boundary = _to_next_boundary(current_offset, io_boundary);
3336 249 : to_next_boundary = spdk_min(remaining, to_next_boundary);
3337 249 : to_next_boundary = spdk_min(max_size, to_next_boundary);
3338 249 : to_next_boundary_bytes = to_next_boundary * blocklen;
3339 :
3340 249 : iov = &bdev_io->child_iov[child_iovcnt];
3341 249 : iovcnt = 0;
3342 :
3343 249 : if (bdev_io->u.bdev.md_buf) {
3344 48 : md_buf = (char *)bdev_io->u.bdev.md_buf +
3345 24 : (current_offset - parent_offset) * spdk_bdev_get_md_size(bdev);
3346 24 : }
3347 :
3348 249 : child_iovsize = spdk_min(SPDK_BDEV_IO_NUM_CHILD_IOV - child_iovcnt, max_child_iovcnt);
3349 1810 : while (to_next_boundary_bytes > 0 && parent_iovpos < parent_iovcnt &&
3350 836 : iovcnt < child_iovsize) {
3351 725 : parent_iov = &bdev_io->u.bdev.iovs[parent_iovpos];
3352 725 : iov_len = parent_iov->iov_len - parent_iov_offset;
3353 :
3354 725 : iov_len = spdk_min(iov_len, max_segment_size);
3355 725 : iov_len = spdk_min(iov_len, to_next_boundary_bytes);
3356 725 : to_next_boundary_bytes -= iov_len;
3357 :
3358 725 : bdev_io->child_iov[child_iovcnt].iov_base = parent_iov->iov_base + parent_iov_offset;
3359 725 : bdev_io->child_iov[child_iovcnt].iov_len = iov_len;
3360 :
3361 725 : if (iov_len < parent_iov->iov_len - parent_iov_offset) {
3362 183 : parent_iov_offset += iov_len;
3363 183 : } else {
3364 542 : parent_iovpos++;
3365 542 : parent_iov_offset = 0;
3366 : }
3367 725 : child_iovcnt++;
3368 725 : iovcnt++;
3369 : }
3370 :
3371 249 : if (to_next_boundary_bytes > 0) {
3372 : /* We had to stop this child I/O early because we ran out of
3373 : * child_iov space or were limited by max_num_segments.
3374 : * Ensure the iovs to be aligned with block size and
3375 : * then adjust to_next_boundary before starting the
3376 : * child I/O.
3377 : */
3378 111 : assert(child_iovcnt == SPDK_BDEV_IO_NUM_CHILD_IOV ||
3379 : iovcnt == child_iovsize);
3380 111 : to_last_block_bytes = to_next_boundary_bytes % blocklen;
3381 111 : if (to_last_block_bytes != 0) {
3382 24 : uint32_t child_iovpos = child_iovcnt - 1;
3383 : /* don't decrease child_iovcnt when it equals to SPDK_BDEV_IO_NUM_CHILD_IOV
3384 : * so the loop will naturally end
3385 : */
3386 :
3387 24 : to_last_block_bytes = blocklen - to_last_block_bytes;
3388 24 : to_next_boundary_bytes += to_last_block_bytes;
3389 53 : while (to_last_block_bytes > 0 && iovcnt > 0) {
3390 32 : iov_len = spdk_min(to_last_block_bytes,
3391 : bdev_io->child_iov[child_iovpos].iov_len);
3392 32 : bdev_io->child_iov[child_iovpos].iov_len -= iov_len;
3393 32 : if (bdev_io->child_iov[child_iovpos].iov_len == 0) {
3394 15 : child_iovpos--;
3395 15 : if (--iovcnt == 0) {
3396 : /* If the child IO is less than a block size just return.
3397 : * If the first child IO of any split round is less than
3398 : * a block size, an error exit.
3399 : */
3400 3 : if (bdev_io->internal.split.outstanding == 0) {
3401 1 : SPDK_ERRLOG("The first child io was less than a block size\n");
3402 1 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
3403 1 : bdev_ch_remove_from_io_submitted(bdev_io);
3404 1 : spdk_trace_record(TRACE_BDEV_IO_DONE, bdev_io->internal.ch->trace_id,
3405 : 0, (uintptr_t)bdev_io, bdev_io->internal.caller_ctx,
3406 : bdev_io->internal.ch->queue_depth);
3407 1 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
3408 1 : }
3409 :
3410 3 : return;
3411 : }
3412 12 : }
3413 :
3414 29 : to_last_block_bytes -= iov_len;
3415 :
3416 29 : if (parent_iov_offset == 0) {
3417 14 : parent_iovpos--;
3418 14 : parent_iov_offset = bdev_io->u.bdev.iovs[parent_iovpos].iov_len;
3419 14 : }
3420 29 : parent_iov_offset -= iov_len;
3421 : }
3422 :
3423 21 : assert(to_last_block_bytes == 0);
3424 21 : }
3425 108 : to_next_boundary -= to_next_boundary_bytes / blocklen;
3426 108 : }
3427 :
3428 246 : rc = bdev_io_split_submit(bdev_io, iov, iovcnt, md_buf, to_next_boundary,
3429 : ¤t_offset, &remaining);
3430 246 : if (spdk_unlikely(rc)) {
3431 4 : return;
3432 : }
3433 : }
3434 67 : }
3435 :
3436 : static void
3437 3 : bdev_unmap_split(struct spdk_bdev_io *bdev_io)
3438 : {
3439 : uint64_t offset, unmap_blocks, remaining, max_unmap_blocks;
3440 3 : uint32_t num_children_reqs = 0;
3441 : int rc;
3442 :
3443 3 : assert(bdev_io->internal.f.split);
3444 :
3445 3 : offset = bdev_io->internal.split.current_offset_blocks;
3446 3 : remaining = bdev_io->internal.split.remaining_num_blocks;
3447 3 : max_unmap_blocks = bdev_io->bdev->max_unmap * bdev_io->bdev->max_unmap_segments;
3448 :
3449 20 : while (remaining && (num_children_reqs < SPDK_BDEV_MAX_CHILDREN_UNMAP_WRITE_ZEROES_REQS)) {
3450 17 : unmap_blocks = spdk_min(remaining, max_unmap_blocks);
3451 :
3452 17 : rc = bdev_io_split_submit(bdev_io, NULL, 0, NULL, unmap_blocks,
3453 : &offset, &remaining);
3454 17 : if (spdk_likely(rc == 0)) {
3455 17 : num_children_reqs++;
3456 17 : } else {
3457 0 : return;
3458 : }
3459 : }
3460 3 : }
3461 :
3462 : static void
3463 6 : bdev_write_zeroes_split(struct spdk_bdev_io *bdev_io)
3464 : {
3465 : uint64_t offset, write_zeroes_blocks, remaining;
3466 6 : uint32_t num_children_reqs = 0;
3467 : int rc;
3468 :
3469 6 : assert(bdev_io->internal.f.split);
3470 :
3471 6 : offset = bdev_io->internal.split.current_offset_blocks;
3472 6 : remaining = bdev_io->internal.split.remaining_num_blocks;
3473 :
3474 29 : while (remaining && (num_children_reqs < SPDK_BDEV_MAX_CHILDREN_UNMAP_WRITE_ZEROES_REQS)) {
3475 23 : write_zeroes_blocks = spdk_min(remaining, bdev_io->bdev->max_write_zeroes);
3476 :
3477 23 : rc = bdev_io_split_submit(bdev_io, NULL, 0, NULL, write_zeroes_blocks,
3478 : &offset, &remaining);
3479 23 : if (spdk_likely(rc == 0)) {
3480 23 : num_children_reqs++;
3481 23 : } else {
3482 0 : return;
3483 : }
3484 : }
3485 6 : }
3486 :
3487 : static void
3488 4 : bdev_copy_split(struct spdk_bdev_io *bdev_io)
3489 : {
3490 : uint64_t offset, copy_blocks, remaining;
3491 4 : uint32_t num_children_reqs = 0;
3492 : int rc;
3493 :
3494 4 : assert(bdev_io->internal.f.split);
3495 :
3496 4 : offset = bdev_io->internal.split.current_offset_blocks;
3497 4 : remaining = bdev_io->internal.split.remaining_num_blocks;
3498 :
3499 4 : assert(bdev_io->bdev->max_copy != 0);
3500 23 : while (remaining && (num_children_reqs < SPDK_BDEV_MAX_CHILDREN_COPY_REQS)) {
3501 19 : copy_blocks = spdk_min(remaining, bdev_io->bdev->max_copy);
3502 :
3503 19 : rc = bdev_io_split_submit(bdev_io, NULL, 0, NULL, copy_blocks,
3504 : &offset, &remaining);
3505 19 : if (spdk_likely(rc == 0)) {
3506 19 : num_children_reqs++;
3507 19 : } else {
3508 0 : return;
3509 : }
3510 : }
3511 4 : }
3512 :
3513 : static void
3514 58 : parent_bdev_io_complete(void *ctx, int rc)
3515 : {
3516 58 : struct spdk_bdev_io *parent_io = ctx;
3517 :
3518 58 : if (rc) {
3519 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
3520 0 : }
3521 :
3522 116 : parent_io->internal.cb(parent_io, parent_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS,
3523 58 : parent_io->internal.caller_ctx);
3524 58 : }
3525 :
3526 : static void
3527 0 : bdev_io_complete_parent_sequence_cb(void *ctx, int status)
3528 : {
3529 0 : struct spdk_bdev_io *bdev_io = ctx;
3530 :
3531 : /* u.bdev.accel_sequence should have already been cleared at this point */
3532 0 : assert(bdev_io->u.bdev.accel_sequence == NULL);
3533 0 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS);
3534 0 : bdev_io->internal.f.has_accel_sequence = false;
3535 :
3536 0 : if (spdk_unlikely(status != 0)) {
3537 0 : SPDK_ERRLOG("Failed to execute accel sequence, status=%d\n", status);
3538 0 : }
3539 :
3540 0 : parent_bdev_io_complete(bdev_io, status);
3541 0 : }
3542 :
3543 : static void
3544 301 : bdev_io_split_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
3545 : {
3546 301 : struct spdk_bdev_io *parent_io = cb_arg;
3547 :
3548 301 : spdk_bdev_free_io(bdev_io);
3549 :
3550 301 : assert(parent_io->internal.f.split);
3551 :
3552 301 : if (!success) {
3553 21 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
3554 : /* If any child I/O failed, stop further splitting process. */
3555 21 : parent_io->internal.split.current_offset_blocks += parent_io->internal.split.remaining_num_blocks;
3556 21 : parent_io->internal.split.remaining_num_blocks = 0;
3557 21 : }
3558 301 : parent_io->internal.split.outstanding--;
3559 301 : if (parent_io->internal.split.outstanding != 0) {
3560 223 : return;
3561 : }
3562 :
3563 : /*
3564 : * Parent I/O finishes when all blocks are consumed.
3565 : */
3566 78 : if (parent_io->internal.split.remaining_num_blocks == 0) {
3567 58 : assert(parent_io->internal.cb != bdev_io_split_done);
3568 58 : bdev_ch_remove_from_io_submitted(parent_io);
3569 58 : spdk_trace_record(TRACE_BDEV_IO_DONE, parent_io->internal.ch->trace_id,
3570 : 0, (uintptr_t)parent_io, bdev_io->internal.caller_ctx,
3571 : parent_io->internal.ch->queue_depth);
3572 :
3573 58 : if (spdk_likely(parent_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS)) {
3574 48 : if (bdev_io_needs_sequence_exec(parent_io->internal.desc, parent_io)) {
3575 0 : bdev_io_exec_sequence(parent_io, bdev_io_complete_parent_sequence_cb);
3576 0 : return;
3577 48 : } else if (parent_io->internal.f.has_bounce_buf &&
3578 0 : !bdev_io_use_accel_sequence(bdev_io)) {
3579 : /* bdev IO will be completed in the callback */
3580 0 : _bdev_io_push_bounce_data_buffer(parent_io, parent_bdev_io_complete);
3581 0 : return;
3582 : }
3583 48 : }
3584 :
3585 58 : parent_bdev_io_complete(parent_io, 0);
3586 58 : return;
3587 : }
3588 :
3589 : /*
3590 : * Continue with the splitting process. This function will complete the parent I/O if the
3591 : * splitting is done.
3592 : */
3593 20 : switch (parent_io->type) {
3594 : case SPDK_BDEV_IO_TYPE_READ:
3595 : case SPDK_BDEV_IO_TYPE_WRITE:
3596 17 : _bdev_rw_split(parent_io);
3597 17 : break;
3598 : case SPDK_BDEV_IO_TYPE_UNMAP:
3599 1 : bdev_unmap_split(parent_io);
3600 1 : break;
3601 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3602 1 : bdev_write_zeroes_split(parent_io);
3603 1 : break;
3604 : case SPDK_BDEV_IO_TYPE_COPY:
3605 1 : bdev_copy_split(parent_io);
3606 1 : break;
3607 : default:
3608 0 : assert(false);
3609 : break;
3610 : }
3611 301 : }
3612 :
3613 : static void bdev_rw_split_get_buf_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io,
3614 : bool success);
3615 :
3616 : static void
3617 59 : bdev_io_split(struct spdk_bdev_io *bdev_io)
3618 : {
3619 59 : assert(bdev_io_should_split(bdev_io));
3620 59 : assert(bdev_io->internal.f.split);
3621 :
3622 59 : bdev_io->internal.split.current_offset_blocks = bdev_io->u.bdev.offset_blocks;
3623 59 : bdev_io->internal.split.remaining_num_blocks = bdev_io->u.bdev.num_blocks;
3624 59 : bdev_io->internal.split.outstanding = 0;
3625 59 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
3626 :
3627 59 : switch (bdev_io->type) {
3628 : case SPDK_BDEV_IO_TYPE_READ:
3629 : case SPDK_BDEV_IO_TYPE_WRITE:
3630 49 : if (_is_buf_allocated(bdev_io->u.bdev.iovs)) {
3631 49 : _bdev_rw_split(bdev_io);
3632 49 : } else {
3633 0 : assert(bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
3634 0 : spdk_bdev_io_get_buf(bdev_io, bdev_rw_split_get_buf_cb,
3635 0 : bdev_io->u.bdev.num_blocks * bdev_io_get_block_size(bdev_io));
3636 : }
3637 49 : break;
3638 : case SPDK_BDEV_IO_TYPE_UNMAP:
3639 2 : bdev_unmap_split(bdev_io);
3640 2 : break;
3641 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3642 5 : bdev_write_zeroes_split(bdev_io);
3643 5 : break;
3644 : case SPDK_BDEV_IO_TYPE_COPY:
3645 3 : bdev_copy_split(bdev_io);
3646 3 : break;
3647 : default:
3648 0 : assert(false);
3649 : break;
3650 : }
3651 59 : }
3652 :
3653 : static void
3654 0 : bdev_rw_split_get_buf_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io, bool success)
3655 : {
3656 0 : if (!success) {
3657 0 : spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
3658 0 : return;
3659 : }
3660 :
3661 0 : _bdev_rw_split(bdev_io);
3662 0 : }
3663 :
3664 : static inline void
3665 579 : _bdev_io_submit(struct spdk_bdev_io *bdev_io)
3666 : {
3667 579 : struct spdk_bdev *bdev = bdev_io->bdev;
3668 579 : struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
3669 :
3670 579 : if (spdk_likely(bdev_ch->flags == 0)) {
3671 554 : bdev_io_do_submit(bdev_ch, bdev_io);
3672 554 : return;
3673 : }
3674 :
3675 25 : if (bdev_ch->flags & BDEV_CH_RESET_IN_PROGRESS) {
3676 2 : _bdev_io_complete_in_submit(bdev_ch, bdev_io, SPDK_BDEV_IO_STATUS_ABORTED);
3677 25 : } else if (bdev_ch->flags & BDEV_CH_QOS_ENABLED) {
3678 23 : if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_ABORT) &&
3679 2 : bdev_abort_queued_io(&bdev_ch->qos_queued_io, bdev_io->u.abort.bio_to_abort)) {
3680 0 : _bdev_io_complete_in_submit(bdev_ch, bdev_io, SPDK_BDEV_IO_STATUS_SUCCESS);
3681 0 : } else {
3682 23 : TAILQ_INSERT_TAIL(&bdev_ch->qos_queued_io, bdev_io, internal.link);
3683 23 : bdev_qos_io_submit(bdev_ch, bdev->internal.qos);
3684 : }
3685 23 : } else {
3686 0 : SPDK_ERRLOG("unknown bdev_ch flag %x found\n", bdev_ch->flags);
3687 0 : _bdev_io_complete_in_submit(bdev_ch, bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
3688 : }
3689 579 : }
3690 :
3691 : bool bdev_lba_range_overlapped(struct lba_range *range1, struct lba_range *range2);
3692 :
3693 : bool
3694 23 : bdev_lba_range_overlapped(struct lba_range *range1, struct lba_range *range2)
3695 : {
3696 23 : if (range1->length == 0 || range2->length == 0) {
3697 1 : return false;
3698 : }
3699 :
3700 22 : if (range1->offset + range1->length <= range2->offset) {
3701 1 : return false;
3702 : }
3703 :
3704 21 : if (range2->offset + range2->length <= range1->offset) {
3705 3 : return false;
3706 : }
3707 :
3708 18 : return true;
3709 23 : }
3710 :
3711 : static bool
3712 11 : bdev_io_range_is_locked(struct spdk_bdev_io *bdev_io, struct lba_range *range)
3713 : {
3714 11 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
3715 : struct lba_range r;
3716 :
3717 11 : switch (bdev_io->type) {
3718 : case SPDK_BDEV_IO_TYPE_NVME_IO:
3719 : case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
3720 : /* Don't try to decode the NVMe command - just assume worst-case and that
3721 : * it overlaps a locked range.
3722 : */
3723 0 : return true;
3724 : case SPDK_BDEV_IO_TYPE_READ:
3725 6 : if (!range->quiesce) {
3726 4 : return false;
3727 : }
3728 : /* fallthrough */
3729 : case SPDK_BDEV_IO_TYPE_WRITE:
3730 : case SPDK_BDEV_IO_TYPE_UNMAP:
3731 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3732 : case SPDK_BDEV_IO_TYPE_ZCOPY:
3733 : case SPDK_BDEV_IO_TYPE_COPY:
3734 7 : r.offset = bdev_io->u.bdev.offset_blocks;
3735 7 : r.length = bdev_io->u.bdev.num_blocks;
3736 7 : if (!bdev_lba_range_overlapped(range, &r)) {
3737 : /* This I/O doesn't overlap the specified LBA range. */
3738 0 : return false;
3739 7 : } else if (range->owner_ch == ch && range->locked_ctx == bdev_io->internal.caller_ctx) {
3740 : /* This I/O overlaps, but the I/O is on the same channel that locked this
3741 : * range, and the caller_ctx is the same as the locked_ctx. This means
3742 : * that this I/O is associated with the lock, and is allowed to execute.
3743 : */
3744 2 : return false;
3745 : } else {
3746 5 : return true;
3747 : }
3748 : default:
3749 0 : return false;
3750 : }
3751 11 : }
3752 :
3753 : void
3754 639 : bdev_io_submit(struct spdk_bdev_io *bdev_io)
3755 : {
3756 639 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
3757 :
3758 639 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_PENDING);
3759 :
3760 639 : if (!TAILQ_EMPTY(&ch->locked_ranges)) {
3761 : struct lba_range *range;
3762 :
3763 13 : TAILQ_FOREACH(range, &ch->locked_ranges, tailq) {
3764 8 : if (bdev_io_range_is_locked(bdev_io, range)) {
3765 3 : TAILQ_INSERT_TAIL(&ch->io_locked, bdev_io, internal.ch_link);
3766 3 : return;
3767 : }
3768 5 : }
3769 5 : }
3770 :
3771 636 : bdev_ch_add_to_io_submitted(bdev_io);
3772 :
3773 636 : bdev_io->internal.submit_tsc = spdk_get_ticks();
3774 636 : spdk_trace_record_tsc(bdev_io->internal.submit_tsc, TRACE_BDEV_IO_START,
3775 : ch->trace_id, bdev_io->u.bdev.num_blocks,
3776 : (uintptr_t)bdev_io, (uint64_t)bdev_io->type, bdev_io->internal.caller_ctx,
3777 : bdev_io->u.bdev.offset_blocks, ch->queue_depth);
3778 :
3779 636 : if (bdev_io->internal.f.split) {
3780 59 : bdev_io_split(bdev_io);
3781 59 : return;
3782 : }
3783 :
3784 577 : _bdev_io_submit(bdev_io);
3785 639 : }
3786 :
3787 : static inline int
3788 2 : bdev_io_init_dif_ctx(struct spdk_bdev_io *bdev_io)
3789 : {
3790 2 : struct spdk_bdev *bdev = bdev_io->bdev;
3791 : struct spdk_dif_ctx_init_ext_opts dif_opts;
3792 :
3793 2 : memset(&bdev_io->u.bdev.dif_err, 0, sizeof(struct spdk_dif_error));
3794 :
3795 2 : dif_opts.size = SPDK_SIZEOF(&dif_opts, dif_pi_format);
3796 2 : dif_opts.dif_pi_format = bdev->dif_pi_format;
3797 :
3798 4 : return spdk_dif_ctx_init(&bdev_io->u.bdev.dif_ctx,
3799 2 : bdev->blocklen,
3800 2 : bdev->md_len,
3801 2 : bdev->md_interleave,
3802 2 : bdev->dif_is_head_of_md,
3803 2 : bdev->dif_type,
3804 2 : bdev_io->u.bdev.dif_check_flags,
3805 2 : bdev_io->u.bdev.offset_blocks & 0xFFFFFFFF,
3806 : 0xFFFF, 0, 0, 0, &dif_opts);
3807 : }
3808 :
3809 : static void
3810 4 : _bdev_memory_domain_get_io_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io,
3811 : bool success)
3812 : {
3813 4 : if (!success) {
3814 0 : SPDK_ERRLOG("Failed to get data buffer, completing IO\n");
3815 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
3816 0 : bdev_io_complete_unsubmitted(bdev_io);
3817 0 : return;
3818 : }
3819 :
3820 4 : if (bdev_io_needs_sequence_exec(bdev_io->internal.desc, bdev_io)) {
3821 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
3822 0 : bdev_io_exec_sequence(bdev_io, bdev_io_submit_sequence_cb);
3823 0 : return;
3824 : }
3825 : /* For reads we'll execute the sequence after the data is read, so, for now, only
3826 : * clear out accel_sequence pointer and submit the IO */
3827 0 : assert(bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
3828 0 : bdev_io->u.bdev.accel_sequence = NULL;
3829 0 : }
3830 :
3831 4 : bdev_io_submit(bdev_io);
3832 4 : }
3833 :
3834 : static inline void
3835 4 : _bdev_io_ext_use_bounce_buffer(struct spdk_bdev_io *bdev_io)
3836 : {
3837 : /* bdev doesn't support memory domains, thereby buffers in this IO request can't
3838 : * be accessed directly. It is needed to allocate buffers before issuing IO operation.
3839 : * For write operation we need to pull buffers from memory domain before submitting IO.
3840 : * Once read operation completes, we need to use memory_domain push functionality to
3841 : * update data in original memory domain IO buffer.
3842 : *
3843 : * If this I/O request is not aware of metadata, buffers in thsi IO request can't be
3844 : * accessed directly too. It is needed to allocate buffers before issuing IO operation.
3845 : * For write operation we need to insert metadata before submitting IO. Once read
3846 : * operation completes, we need to strip metadata in original IO buffer.
3847 : *
3848 : * This IO request will go through a regular IO flow, so clear memory domains pointers */
3849 4 : assert(bdev_io_use_memory_domain(bdev_io) ||
3850 : bdev_io_needs_metadata(bdev_io->internal.desc, bdev_io));
3851 :
3852 4 : bdev_io->u.bdev.memory_domain = NULL;
3853 4 : bdev_io->u.bdev.memory_domain_ctx = NULL;
3854 8 : _bdev_io_get_bounce_buf(bdev_io, _bdev_memory_domain_get_io_cb,
3855 4 : bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen);
3856 4 : }
3857 :
3858 : static inline void
3859 0 : _bdev_io_ext_use_accel_buffer(struct spdk_bdev_io *bdev_io)
3860 : {
3861 0 : assert(bdev_io_use_memory_domain(bdev_io));
3862 0 : assert(bdev_io_needs_metadata(bdev_io->internal.desc, bdev_io));
3863 :
3864 0 : bdev_io->u.bdev.memory_domain = NULL;
3865 0 : bdev_io->u.bdev.memory_domain_ctx = NULL;
3866 0 : bdev_io_get_accel_buf(bdev_io, _bdev_memory_domain_get_io_cb,
3867 0 : bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen);
3868 0 : }
3869 :
3870 : /* We need to allocate bounce buffer
3871 : * - if bdev doesn't support memory domains,
3872 : * - if it does support them, but we need to execute an accel sequence and the data buffer is
3873 : * from accel memory domain (to avoid doing a push/pull from that domain), or
3874 : * - if IO is not aware of metadata.
3875 : */
3876 : static inline bool
3877 292 : bdev_io_needs_bounce_buffer(struct spdk_bdev_desc *desc, struct spdk_bdev_io *bdev_io)
3878 : {
3879 292 : if (bdev_io_use_memory_domain(bdev_io)) {
3880 4 : if (!desc->memory_domains_supported ||
3881 0 : (bdev_io_needs_sequence_exec(desc, bdev_io) &&
3882 0 : (bdev_io->internal.memory_domain == spdk_accel_get_memory_domain() ||
3883 0 : bdev_io_needs_metadata(desc, bdev_io)))) {
3884 4 : return true;
3885 : }
3886 :
3887 0 : return false;
3888 : }
3889 :
3890 288 : if (bdev_io_needs_metadata(desc, bdev_io)) {
3891 0 : return true;
3892 : }
3893 :
3894 288 : return false;
3895 292 : }
3896 :
3897 : /* We need to allocate fake accel buffer if bdev supports memory domains but IO is not
3898 : * aware of metadata.
3899 : */
3900 : static inline bool
3901 288 : bdev_io_needs_accel_buffer(struct spdk_bdev_desc *desc, struct spdk_bdev_io *bdev_io)
3902 : {
3903 288 : if (bdev_io_needs_metadata(desc, bdev_io)) {
3904 0 : assert(bdev_io_use_memory_domain(bdev_io));
3905 0 : return true;
3906 : }
3907 :
3908 288 : return false;
3909 288 : }
3910 :
3911 : static inline void
3912 292 : _bdev_io_submit_ext(struct spdk_bdev_desc *desc, struct spdk_bdev_io *bdev_io)
3913 : {
3914 292 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
3915 : int rc;
3916 :
3917 292 : if (spdk_unlikely(ch->flags & BDEV_CH_RESET_IN_PROGRESS)) {
3918 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_ABORTED;
3919 0 : bdev_io_complete_unsubmitted(bdev_io);
3920 0 : return;
3921 : }
3922 :
3923 292 : if (bdev_io_needs_metadata(desc, bdev_io)) {
3924 0 : rc = bdev_io_init_dif_ctx(bdev_io);
3925 0 : if (spdk_unlikely(rc != 0)) {
3926 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
3927 0 : bdev_io_complete_unsubmitted(bdev_io);
3928 0 : return;
3929 : }
3930 0 : }
3931 :
3932 292 : if (bdev_io_needs_bounce_buffer(desc, bdev_io)) {
3933 4 : _bdev_io_ext_use_bounce_buffer(bdev_io);
3934 4 : return;
3935 : }
3936 :
3937 288 : if (bdev_io_needs_accel_buffer(desc, bdev_io)) {
3938 0 : _bdev_io_ext_use_accel_buffer(bdev_io);
3939 0 : return;
3940 : }
3941 :
3942 288 : if (bdev_io_needs_sequence_exec(desc, bdev_io)) {
3943 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
3944 0 : bdev_io_exec_sequence(bdev_io, bdev_io_submit_sequence_cb);
3945 0 : return;
3946 : }
3947 : /* For reads we'll execute the sequence after the data is read, so, for now, only
3948 : * clear out accel_sequence pointer and submit the IO */
3949 0 : assert(bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
3950 0 : bdev_io->u.bdev.accel_sequence = NULL;
3951 0 : }
3952 :
3953 288 : bdev_io_submit(bdev_io);
3954 292 : }
3955 :
3956 : static void
3957 11 : bdev_io_submit_reset(struct spdk_bdev_io *bdev_io)
3958 : {
3959 11 : struct spdk_bdev *bdev = bdev_io->bdev;
3960 11 : struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
3961 11 : struct spdk_io_channel *ch = bdev_ch->channel;
3962 :
3963 11 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_PENDING);
3964 :
3965 11 : bdev_io->internal.f.in_submit_request = true;
3966 11 : bdev_submit_request(bdev, ch, bdev_io);
3967 11 : bdev_io->internal.f.in_submit_request = false;
3968 11 : }
3969 :
3970 : void
3971 691 : bdev_io_init(struct spdk_bdev_io *bdev_io,
3972 : struct spdk_bdev *bdev, void *cb_arg,
3973 : spdk_bdev_io_completion_cb cb)
3974 : {
3975 691 : bdev_io->bdev = bdev;
3976 691 : bdev_io->internal.f.raw = 0;
3977 691 : bdev_io->internal.caller_ctx = cb_arg;
3978 691 : bdev_io->internal.cb = cb;
3979 691 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_PENDING;
3980 691 : bdev_io->internal.f.in_submit_request = false;
3981 691 : bdev_io->internal.error.nvme.cdw0 = 0;
3982 691 : bdev_io->num_retries = 0;
3983 691 : bdev_io->internal.get_buf_cb = NULL;
3984 691 : bdev_io->internal.get_aux_buf_cb = NULL;
3985 691 : bdev_io->internal.data_transfer_cpl = NULL;
3986 691 : bdev_io->internal.f.split = bdev_io_should_split(bdev_io);
3987 691 : }
3988 :
3989 : static bool
3990 523 : bdev_io_type_supported(struct spdk_bdev *bdev, enum spdk_bdev_io_type io_type)
3991 : {
3992 523 : return bdev->fn_table->io_type_supported(bdev->ctxt, io_type);
3993 : }
3994 :
3995 : bool
3996 171 : spdk_bdev_io_type_supported(struct spdk_bdev *bdev, enum spdk_bdev_io_type io_type)
3997 : {
3998 : bool supported;
3999 :
4000 171 : supported = bdev_io_type_supported(bdev, io_type);
4001 :
4002 171 : if (!supported) {
4003 7 : switch (io_type) {
4004 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
4005 : /* The bdev layer will emulate write zeroes as long as write is supported. */
4006 0 : supported = bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE);
4007 0 : break;
4008 : default:
4009 7 : break;
4010 : }
4011 7 : }
4012 :
4013 171 : return supported;
4014 : }
4015 :
4016 : static const char *g_io_type_strings[] = {
4017 : [SPDK_BDEV_IO_TYPE_READ] = "read",
4018 : [SPDK_BDEV_IO_TYPE_WRITE] = "write",
4019 : [SPDK_BDEV_IO_TYPE_UNMAP] = "unmap",
4020 : [SPDK_BDEV_IO_TYPE_FLUSH] = "flush",
4021 : [SPDK_BDEV_IO_TYPE_RESET] = "reset",
4022 : [SPDK_BDEV_IO_TYPE_NVME_ADMIN] = "nvme_admin",
4023 : [SPDK_BDEV_IO_TYPE_NVME_IO] = "nvme_io",
4024 : [SPDK_BDEV_IO_TYPE_NVME_IO_MD] = "nvme_io_md",
4025 : [SPDK_BDEV_IO_TYPE_WRITE_ZEROES] = "write_zeroes",
4026 : [SPDK_BDEV_IO_TYPE_ZCOPY] = "zcopy",
4027 : [SPDK_BDEV_IO_TYPE_GET_ZONE_INFO] = "get_zone_info",
4028 : [SPDK_BDEV_IO_TYPE_ZONE_MANAGEMENT] = "zone_management",
4029 : [SPDK_BDEV_IO_TYPE_ZONE_APPEND] = "zone_append",
4030 : [SPDK_BDEV_IO_TYPE_COMPARE] = "compare",
4031 : [SPDK_BDEV_IO_TYPE_COMPARE_AND_WRITE] = "compare_and_write",
4032 : [SPDK_BDEV_IO_TYPE_ABORT] = "abort",
4033 : [SPDK_BDEV_IO_TYPE_SEEK_HOLE] = "seek_hole",
4034 : [SPDK_BDEV_IO_TYPE_SEEK_DATA] = "seek_data",
4035 : [SPDK_BDEV_IO_TYPE_COPY] = "copy",
4036 : [SPDK_BDEV_IO_TYPE_NVME_IOV_MD] = "nvme_iov_md",
4037 : };
4038 :
4039 : const char *
4040 0 : spdk_bdev_get_io_type_name(enum spdk_bdev_io_type io_type)
4041 : {
4042 0 : if (io_type <= SPDK_BDEV_IO_TYPE_INVALID || io_type >= SPDK_BDEV_NUM_IO_TYPES) {
4043 0 : return NULL;
4044 : }
4045 :
4046 0 : return g_io_type_strings[io_type];
4047 0 : }
4048 :
4049 : int
4050 0 : spdk_bdev_get_io_type(const char *io_type_string)
4051 : {
4052 : int i;
4053 :
4054 0 : for (i = SPDK_BDEV_IO_TYPE_READ; i < SPDK_BDEV_NUM_IO_TYPES; ++i) {
4055 0 : if (!strcmp(io_type_string, g_io_type_strings[i])) {
4056 0 : return i;
4057 : }
4058 0 : }
4059 :
4060 0 : return -1;
4061 0 : }
4062 :
4063 : uint64_t
4064 0 : spdk_bdev_io_get_submit_tsc(struct spdk_bdev_io *bdev_io)
4065 : {
4066 0 : return bdev_io->internal.submit_tsc;
4067 : }
4068 :
4069 : bool
4070 0 : spdk_bdev_io_has_no_metadata(struct spdk_bdev_io *bdev_io)
4071 : {
4072 0 : return bdev_io->internal.desc->opts.no_metadata;
4073 : }
4074 :
4075 : int
4076 0 : spdk_bdev_dump_info_json(struct spdk_bdev *bdev, struct spdk_json_write_ctx *w)
4077 : {
4078 0 : if (bdev->fn_table->dump_info_json) {
4079 0 : return bdev->fn_table->dump_info_json(bdev->ctxt, w);
4080 : }
4081 :
4082 0 : return 0;
4083 0 : }
4084 :
4085 : static void
4086 10 : bdev_qos_update_max_quota_per_timeslice(struct spdk_bdev_qos *qos)
4087 : {
4088 10 : uint32_t max_per_timeslice = 0;
4089 : int i;
4090 :
4091 50 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
4092 40 : if (qos->rate_limits[i].limit == SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
4093 15 : qos->rate_limits[i].max_per_timeslice = 0;
4094 15 : continue;
4095 : }
4096 :
4097 50 : max_per_timeslice = qos->rate_limits[i].limit *
4098 25 : SPDK_BDEV_QOS_TIMESLICE_IN_USEC / SPDK_SEC_TO_USEC;
4099 :
4100 25 : qos->rate_limits[i].max_per_timeslice = spdk_max(max_per_timeslice,
4101 : qos->rate_limits[i].min_per_timeslice);
4102 :
4103 50 : __atomic_store_n(&qos->rate_limits[i].remaining_this_timeslice,
4104 25 : qos->rate_limits[i].max_per_timeslice, __ATOMIC_RELEASE);
4105 25 : }
4106 :
4107 10 : bdev_qos_set_ops(qos);
4108 10 : }
4109 :
4110 : static void
4111 4 : bdev_channel_submit_qos_io(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
4112 : struct spdk_io_channel *io_ch, void *ctx)
4113 : {
4114 4 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
4115 : int status;
4116 :
4117 4 : bdev_qos_io_submit(bdev_ch, bdev->internal.qos);
4118 :
4119 : /* if all IOs were sent then continue the iteration, otherwise - stop it */
4120 : /* TODO: channels round robing */
4121 4 : status = TAILQ_EMPTY(&bdev_ch->qos_queued_io) ? 0 : 1;
4122 :
4123 4 : spdk_bdev_for_each_channel_continue(i, status);
4124 4 : }
4125 :
4126 :
4127 : static void
4128 2 : bdev_channel_submit_qos_io_done(struct spdk_bdev *bdev, void *ctx, int status)
4129 : {
4130 :
4131 2 : }
4132 :
4133 : static int
4134 3 : bdev_channel_poll_qos(void *arg)
4135 : {
4136 3 : struct spdk_bdev *bdev = arg;
4137 3 : struct spdk_bdev_qos *qos = bdev->internal.qos;
4138 3 : uint64_t now = spdk_get_ticks();
4139 : int i;
4140 : int64_t remaining_last_timeslice;
4141 :
4142 3 : if (spdk_unlikely(qos->thread == NULL)) {
4143 : /* Old QoS was unbound to remove and new QoS is not enabled yet. */
4144 1 : return SPDK_POLLER_IDLE;
4145 : }
4146 :
4147 2 : if (now < (qos->last_timeslice + qos->timeslice_size)) {
4148 : /* We received our callback earlier than expected - return
4149 : * immediately and wait to do accounting until at least one
4150 : * timeslice has actually expired. This should never happen
4151 : * with a well-behaved timer implementation.
4152 : */
4153 0 : return SPDK_POLLER_IDLE;
4154 : }
4155 :
4156 : /* Reset for next round of rate limiting */
4157 10 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
4158 : /* We may have allowed the IOs or bytes to slightly overrun in the last
4159 : * timeslice. remaining_this_timeslice is signed, so if it's negative
4160 : * here, we'll account for the overrun so that the next timeslice will
4161 : * be appropriately reduced.
4162 : */
4163 8 : remaining_last_timeslice = __atomic_exchange_n(&qos->rate_limits[i].remaining_this_timeslice,
4164 : 0, __ATOMIC_RELAXED);
4165 8 : if (remaining_last_timeslice < 0) {
4166 : /* There could be a race condition here as both bdev_qos_rw_queue_io() and bdev_channel_poll_qos()
4167 : * potentially use 2 atomic ops each, so they can intertwine.
4168 : * This race can potentially cause the limits to be a little fuzzy but won't cause any real damage.
4169 : */
4170 0 : __atomic_store_n(&qos->rate_limits[i].remaining_this_timeslice,
4171 0 : remaining_last_timeslice, __ATOMIC_RELAXED);
4172 0 : }
4173 8 : }
4174 :
4175 4 : while (now >= (qos->last_timeslice + qos->timeslice_size)) {
4176 2 : qos->last_timeslice += qos->timeslice_size;
4177 10 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
4178 16 : __atomic_add_fetch(&qos->rate_limits[i].remaining_this_timeslice,
4179 8 : qos->rate_limits[i].max_per_timeslice, __ATOMIC_RELAXED);
4180 8 : }
4181 : }
4182 :
4183 2 : spdk_bdev_for_each_channel(bdev, bdev_channel_submit_qos_io, qos,
4184 : bdev_channel_submit_qos_io_done);
4185 :
4186 2 : return SPDK_POLLER_BUSY;
4187 3 : }
4188 :
4189 : static void
4190 74 : bdev_channel_destroy_resource(struct spdk_bdev_channel *ch)
4191 : {
4192 : struct spdk_bdev_shared_resource *shared_resource;
4193 : struct lba_range *range;
4194 :
4195 74 : bdev_free_io_stat(ch->stat);
4196 : #ifdef SPDK_CONFIG_VTUNE
4197 : bdev_free_io_stat(ch->prev_stat);
4198 : #endif
4199 :
4200 74 : while (!TAILQ_EMPTY(&ch->locked_ranges)) {
4201 0 : range = TAILQ_FIRST(&ch->locked_ranges);
4202 0 : TAILQ_REMOVE(&ch->locked_ranges, range, tailq);
4203 0 : free(range);
4204 : }
4205 :
4206 74 : spdk_put_io_channel(ch->channel);
4207 74 : spdk_put_io_channel(ch->accel_channel);
4208 :
4209 74 : shared_resource = ch->shared_resource;
4210 :
4211 74 : assert(TAILQ_EMPTY(&ch->io_locked));
4212 74 : assert(TAILQ_EMPTY(&ch->io_submitted));
4213 74 : assert(TAILQ_EMPTY(&ch->io_accel_exec));
4214 74 : assert(TAILQ_EMPTY(&ch->io_memory_domain));
4215 74 : assert(ch->io_outstanding == 0);
4216 74 : assert(shared_resource->ref > 0);
4217 74 : shared_resource->ref--;
4218 74 : if (shared_resource->ref == 0) {
4219 73 : assert(shared_resource->io_outstanding == 0);
4220 73 : TAILQ_REMOVE(&shared_resource->mgmt_ch->shared_resources, shared_resource, link);
4221 73 : spdk_put_io_channel(spdk_io_channel_from_ctx(shared_resource->mgmt_ch));
4222 73 : spdk_poller_unregister(&shared_resource->nomem_poller);
4223 73 : free(shared_resource);
4224 73 : }
4225 74 : }
4226 :
4227 : static void
4228 83 : bdev_enable_qos(struct spdk_bdev *bdev, struct spdk_bdev_channel *ch)
4229 : {
4230 83 : struct spdk_bdev_qos *qos = bdev->internal.qos;
4231 : int i;
4232 :
4233 83 : assert(spdk_spin_held(&bdev->internal.spinlock));
4234 :
4235 : /* Rate limiting on this bdev enabled */
4236 83 : if (qos) {
4237 17 : if (qos->ch == NULL) {
4238 : struct spdk_io_channel *io_ch;
4239 :
4240 9 : SPDK_DEBUGLOG(bdev, "Selecting channel %p as QoS channel for bdev %s on thread %p\n", ch,
4241 : bdev->name, spdk_get_thread());
4242 :
4243 : /* No qos channel has been selected, so set one up */
4244 :
4245 : /* Take another reference to ch */
4246 9 : io_ch = spdk_get_io_channel(__bdev_to_io_dev(bdev));
4247 9 : assert(io_ch != NULL);
4248 9 : qos->ch = ch;
4249 :
4250 9 : qos->thread = spdk_io_channel_get_thread(io_ch);
4251 :
4252 45 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
4253 36 : if (bdev_qos_is_iops_rate_limit(i) == true) {
4254 9 : qos->rate_limits[i].min_per_timeslice =
4255 : SPDK_BDEV_QOS_MIN_IO_PER_TIMESLICE;
4256 9 : } else {
4257 27 : qos->rate_limits[i].min_per_timeslice =
4258 : SPDK_BDEV_QOS_MIN_BYTE_PER_TIMESLICE;
4259 : }
4260 :
4261 36 : if (qos->rate_limits[i].limit == 0) {
4262 2 : qos->rate_limits[i].limit = SPDK_BDEV_QOS_LIMIT_NOT_DEFINED;
4263 2 : }
4264 36 : }
4265 9 : bdev_qos_update_max_quota_per_timeslice(qos);
4266 9 : qos->timeslice_size =
4267 9 : SPDK_BDEV_QOS_TIMESLICE_IN_USEC * spdk_get_ticks_hz() / SPDK_SEC_TO_USEC;
4268 9 : qos->last_timeslice = spdk_get_ticks();
4269 9 : qos->poller = SPDK_POLLER_REGISTER(bdev_channel_poll_qos,
4270 : bdev,
4271 : SPDK_BDEV_QOS_TIMESLICE_IN_USEC);
4272 9 : }
4273 :
4274 17 : ch->flags |= BDEV_CH_QOS_ENABLED;
4275 17 : }
4276 83 : }
4277 :
4278 : struct poll_timeout_ctx {
4279 : struct spdk_bdev_desc *desc;
4280 : uint64_t timeout_in_sec;
4281 : spdk_bdev_io_timeout_cb cb_fn;
4282 : void *cb_arg;
4283 : };
4284 :
4285 : static void
4286 268 : bdev_desc_free(struct spdk_bdev_desc *desc)
4287 : {
4288 268 : spdk_spin_destroy(&desc->spinlock);
4289 268 : free(desc->media_events_buffer);
4290 268 : free(desc);
4291 268 : }
4292 :
4293 : static void
4294 8 : bdev_channel_poll_timeout_io_done(struct spdk_bdev *bdev, void *_ctx, int status)
4295 : {
4296 8 : struct poll_timeout_ctx *ctx = _ctx;
4297 8 : struct spdk_bdev_desc *desc = ctx->desc;
4298 :
4299 8 : free(ctx);
4300 :
4301 8 : spdk_spin_lock(&desc->spinlock);
4302 8 : desc->refs--;
4303 8 : if (desc->closed == true && desc->refs == 0) {
4304 1 : spdk_spin_unlock(&desc->spinlock);
4305 1 : bdev_desc_free(desc);
4306 1 : return;
4307 : }
4308 7 : spdk_spin_unlock(&desc->spinlock);
4309 8 : }
4310 :
4311 : static void
4312 13 : bdev_channel_poll_timeout_io(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
4313 : struct spdk_io_channel *io_ch, void *_ctx)
4314 : {
4315 13 : struct poll_timeout_ctx *ctx = _ctx;
4316 13 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
4317 13 : struct spdk_bdev_desc *desc = ctx->desc;
4318 : struct spdk_bdev_io *bdev_io;
4319 : uint64_t now;
4320 :
4321 13 : spdk_spin_lock(&desc->spinlock);
4322 13 : if (desc->closed == true) {
4323 1 : spdk_spin_unlock(&desc->spinlock);
4324 1 : spdk_bdev_for_each_channel_continue(i, -1);
4325 1 : return;
4326 : }
4327 12 : spdk_spin_unlock(&desc->spinlock);
4328 :
4329 12 : now = spdk_get_ticks();
4330 22 : TAILQ_FOREACH(bdev_io, &bdev_ch->io_submitted, internal.ch_link) {
4331 : /* Exclude any I/O that are generated via splitting. */
4332 15 : if (bdev_io->internal.cb == bdev_io_split_done) {
4333 3 : continue;
4334 : }
4335 :
4336 : /* Once we find an I/O that has not timed out, we can immediately
4337 : * exit the loop.
4338 : */
4339 24 : if (now < (bdev_io->internal.submit_tsc +
4340 12 : ctx->timeout_in_sec * spdk_get_ticks_hz())) {
4341 5 : goto end;
4342 : }
4343 :
4344 7 : if (bdev_io->internal.desc == desc) {
4345 7 : ctx->cb_fn(ctx->cb_arg, bdev_io);
4346 7 : }
4347 14 : }
4348 :
4349 : end:
4350 12 : spdk_bdev_for_each_channel_continue(i, 0);
4351 13 : }
4352 :
4353 : static int
4354 8 : bdev_poll_timeout_io(void *arg)
4355 : {
4356 8 : struct spdk_bdev_desc *desc = arg;
4357 8 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
4358 : struct poll_timeout_ctx *ctx;
4359 :
4360 8 : ctx = calloc(1, sizeof(struct poll_timeout_ctx));
4361 8 : if (!ctx) {
4362 0 : SPDK_ERRLOG("failed to allocate memory\n");
4363 0 : return SPDK_POLLER_BUSY;
4364 : }
4365 8 : ctx->desc = desc;
4366 8 : ctx->cb_arg = desc->cb_arg;
4367 8 : ctx->cb_fn = desc->cb_fn;
4368 8 : ctx->timeout_in_sec = desc->timeout_in_sec;
4369 :
4370 : /* Take a ref on the descriptor in case it gets closed while we are checking
4371 : * all of the channels.
4372 : */
4373 8 : spdk_spin_lock(&desc->spinlock);
4374 8 : desc->refs++;
4375 8 : spdk_spin_unlock(&desc->spinlock);
4376 :
4377 8 : spdk_bdev_for_each_channel(bdev, bdev_channel_poll_timeout_io, ctx,
4378 : bdev_channel_poll_timeout_io_done);
4379 :
4380 8 : return SPDK_POLLER_BUSY;
4381 8 : }
4382 :
4383 : int
4384 5 : spdk_bdev_set_timeout(struct spdk_bdev_desc *desc, uint64_t timeout_in_sec,
4385 : spdk_bdev_io_timeout_cb cb_fn, void *cb_arg)
4386 : {
4387 5 : assert(desc->thread == spdk_get_thread());
4388 :
4389 5 : spdk_poller_unregister(&desc->io_timeout_poller);
4390 :
4391 5 : if (timeout_in_sec) {
4392 4 : assert(cb_fn != NULL);
4393 4 : desc->io_timeout_poller = SPDK_POLLER_REGISTER(bdev_poll_timeout_io,
4394 : desc,
4395 : SPDK_BDEV_IO_POLL_INTERVAL_IN_MSEC * SPDK_SEC_TO_USEC /
4396 : 1000);
4397 4 : if (desc->io_timeout_poller == NULL) {
4398 0 : SPDK_ERRLOG("can not register the desc timeout IO poller\n");
4399 0 : return -1;
4400 : }
4401 4 : }
4402 :
4403 5 : desc->cb_fn = cb_fn;
4404 5 : desc->cb_arg = cb_arg;
4405 5 : desc->timeout_in_sec = timeout_in_sec;
4406 :
4407 5 : return 0;
4408 5 : }
4409 :
4410 : static int
4411 76 : bdev_channel_create(void *io_device, void *ctx_buf)
4412 : {
4413 76 : struct spdk_bdev *bdev = __bdev_from_io_dev(io_device);
4414 76 : struct spdk_bdev_channel *ch = ctx_buf;
4415 : struct spdk_io_channel *mgmt_io_ch;
4416 : struct spdk_bdev_mgmt_channel *mgmt_ch;
4417 : struct spdk_bdev_shared_resource *shared_resource;
4418 : struct lba_range *range;
4419 :
4420 76 : ch->bdev = bdev;
4421 76 : ch->channel = bdev->fn_table->get_io_channel(bdev->ctxt);
4422 76 : if (!ch->channel) {
4423 2 : return -1;
4424 : }
4425 :
4426 74 : ch->accel_channel = spdk_accel_get_io_channel();
4427 74 : if (!ch->accel_channel) {
4428 0 : spdk_put_io_channel(ch->channel);
4429 0 : return -1;
4430 : }
4431 :
4432 74 : spdk_trace_record(TRACE_BDEV_IOCH_CREATE, bdev->internal.trace_id, 0, 0,
4433 : spdk_thread_get_id(spdk_io_channel_get_thread(ch->channel)));
4434 :
4435 74 : assert(ch->histogram == NULL);
4436 74 : if (bdev->internal.histogram_enabled) {
4437 0 : ch->histogram = spdk_histogram_data_alloc();
4438 0 : if (ch->histogram == NULL) {
4439 0 : SPDK_ERRLOG("Could not allocate histogram\n");
4440 0 : }
4441 0 : }
4442 :
4443 74 : mgmt_io_ch = spdk_get_io_channel(&g_bdev_mgr);
4444 74 : if (!mgmt_io_ch) {
4445 0 : spdk_put_io_channel(ch->channel);
4446 0 : spdk_put_io_channel(ch->accel_channel);
4447 0 : return -1;
4448 : }
4449 :
4450 74 : mgmt_ch = __io_ch_to_bdev_mgmt_ch(mgmt_io_ch);
4451 76 : TAILQ_FOREACH(shared_resource, &mgmt_ch->shared_resources, link) {
4452 3 : if (shared_resource->shared_ch == ch->channel) {
4453 1 : spdk_put_io_channel(mgmt_io_ch);
4454 1 : shared_resource->ref++;
4455 1 : break;
4456 : }
4457 2 : }
4458 :
4459 74 : if (shared_resource == NULL) {
4460 73 : shared_resource = calloc(1, sizeof(*shared_resource));
4461 73 : if (shared_resource == NULL) {
4462 0 : spdk_put_io_channel(ch->channel);
4463 0 : spdk_put_io_channel(ch->accel_channel);
4464 0 : spdk_put_io_channel(mgmt_io_ch);
4465 0 : return -1;
4466 : }
4467 :
4468 73 : shared_resource->mgmt_ch = mgmt_ch;
4469 73 : shared_resource->io_outstanding = 0;
4470 73 : TAILQ_INIT(&shared_resource->nomem_io);
4471 73 : shared_resource->nomem_threshold = 0;
4472 73 : shared_resource->shared_ch = ch->channel;
4473 73 : shared_resource->ref = 1;
4474 73 : TAILQ_INSERT_TAIL(&mgmt_ch->shared_resources, shared_resource, link);
4475 73 : }
4476 :
4477 74 : ch->io_outstanding = 0;
4478 74 : TAILQ_INIT(&ch->queued_resets);
4479 74 : TAILQ_INIT(&ch->locked_ranges);
4480 74 : TAILQ_INIT(&ch->qos_queued_io);
4481 74 : ch->flags = 0;
4482 74 : ch->trace_id = bdev->internal.trace_id;
4483 74 : ch->shared_resource = shared_resource;
4484 :
4485 74 : TAILQ_INIT(&ch->io_submitted);
4486 74 : TAILQ_INIT(&ch->io_locked);
4487 74 : TAILQ_INIT(&ch->io_accel_exec);
4488 74 : TAILQ_INIT(&ch->io_memory_domain);
4489 :
4490 74 : ch->stat = bdev_alloc_io_stat(false);
4491 74 : if (ch->stat == NULL) {
4492 0 : bdev_channel_destroy_resource(ch);
4493 0 : return -1;
4494 : }
4495 :
4496 74 : ch->stat->ticks_rate = spdk_get_ticks_hz();
4497 :
4498 : #ifdef SPDK_CONFIG_VTUNE
4499 : {
4500 : char *name;
4501 : __itt_init_ittlib(NULL, 0);
4502 : name = spdk_sprintf_alloc("spdk_bdev_%s_%p", ch->bdev->name, ch);
4503 : if (!name) {
4504 : bdev_channel_destroy_resource(ch);
4505 : return -1;
4506 : }
4507 : ch->handle = __itt_string_handle_create(name);
4508 : free(name);
4509 : ch->start_tsc = spdk_get_ticks();
4510 : ch->interval_tsc = spdk_get_ticks_hz() / 100;
4511 : ch->prev_stat = bdev_alloc_io_stat(false);
4512 : if (ch->prev_stat == NULL) {
4513 : bdev_channel_destroy_resource(ch);
4514 : return -1;
4515 : }
4516 : }
4517 : #endif
4518 :
4519 74 : spdk_spin_lock(&bdev->internal.spinlock);
4520 74 : bdev_enable_qos(bdev, ch);
4521 :
4522 75 : TAILQ_FOREACH(range, &bdev->internal.locked_ranges, tailq) {
4523 : struct lba_range *new_range;
4524 :
4525 1 : new_range = calloc(1, sizeof(*new_range));
4526 1 : if (new_range == NULL) {
4527 0 : spdk_spin_unlock(&bdev->internal.spinlock);
4528 0 : bdev_channel_destroy_resource(ch);
4529 0 : return -1;
4530 : }
4531 1 : new_range->length = range->length;
4532 1 : new_range->offset = range->offset;
4533 1 : new_range->locked_ctx = range->locked_ctx;
4534 1 : TAILQ_INSERT_TAIL(&ch->locked_ranges, new_range, tailq);
4535 1 : }
4536 :
4537 74 : spdk_spin_unlock(&bdev->internal.spinlock);
4538 :
4539 74 : return 0;
4540 76 : }
4541 :
4542 : static int
4543 0 : bdev_abort_all_buf_io_cb(struct spdk_iobuf_channel *ch, struct spdk_iobuf_entry *entry,
4544 : void *cb_ctx)
4545 : {
4546 0 : struct spdk_bdev_channel *bdev_ch = cb_ctx;
4547 : struct spdk_bdev_io *bdev_io;
4548 : uint64_t buf_len;
4549 :
4550 0 : bdev_io = SPDK_CONTAINEROF(entry, struct spdk_bdev_io, internal.iobuf);
4551 0 : if (bdev_io->internal.ch == bdev_ch) {
4552 0 : buf_len = bdev_io_get_max_buf_len(bdev_io, bdev_io->internal.buf.len);
4553 0 : spdk_iobuf_entry_abort(ch, entry, buf_len);
4554 0 : spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_ABORTED);
4555 0 : }
4556 :
4557 0 : return 0;
4558 : }
4559 :
4560 : /*
4561 : * Abort I/O that are waiting on a data buffer.
4562 : */
4563 : static void
4564 96 : bdev_abort_all_buf_io(struct spdk_bdev_mgmt_channel *mgmt_ch, struct spdk_bdev_channel *ch)
4565 : {
4566 96 : spdk_iobuf_for_each_entry(&mgmt_ch->iobuf, bdev_abort_all_buf_io_cb, ch);
4567 96 : }
4568 :
4569 : /*
4570 : * Abort I/O that are queued waiting for submission. These types of I/O are
4571 : * linked using the spdk_bdev_io link TAILQ_ENTRY.
4572 : */
4573 : static void
4574 188 : bdev_abort_all_queued_io(bdev_io_tailq_t *queue, struct spdk_bdev_channel *ch)
4575 : {
4576 : struct spdk_bdev_io *bdev_io, *tmp;
4577 :
4578 228 : TAILQ_FOREACH_SAFE(bdev_io, queue, internal.link, tmp) {
4579 40 : if (bdev_io->internal.ch == ch) {
4580 40 : TAILQ_REMOVE(queue, bdev_io, internal.link);
4581 : /*
4582 : * spdk_bdev_io_complete() assumes that the completed I/O had
4583 : * been submitted to the bdev module. Since in this case it
4584 : * hadn't, bump io_outstanding to account for the decrement
4585 : * that spdk_bdev_io_complete() will do.
4586 : */
4587 40 : if (bdev_io->type != SPDK_BDEV_IO_TYPE_RESET) {
4588 39 : bdev_io_increment_outstanding(ch, ch->shared_resource);
4589 39 : }
4590 40 : spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_ABORTED);
4591 40 : }
4592 40 : }
4593 188 : }
4594 :
4595 : static bool
4596 18 : bdev_abort_queued_io(bdev_io_tailq_t *queue, struct spdk_bdev_io *bio_to_abort)
4597 : {
4598 : struct spdk_bdev_io *bdev_io;
4599 :
4600 18 : TAILQ_FOREACH(bdev_io, queue, internal.link) {
4601 0 : if (bdev_io == bio_to_abort) {
4602 0 : TAILQ_REMOVE(queue, bio_to_abort, internal.link);
4603 0 : spdk_bdev_io_complete(bio_to_abort, SPDK_BDEV_IO_STATUS_ABORTED);
4604 0 : return true;
4605 : }
4606 0 : }
4607 :
4608 18 : return false;
4609 18 : }
4610 :
4611 : static int
4612 0 : bdev_abort_buf_io_cb(struct spdk_iobuf_channel *ch, struct spdk_iobuf_entry *entry, void *cb_ctx)
4613 : {
4614 0 : struct spdk_bdev_io *bdev_io, *bio_to_abort = cb_ctx;
4615 : uint64_t buf_len;
4616 :
4617 0 : bdev_io = SPDK_CONTAINEROF(entry, struct spdk_bdev_io, internal.iobuf);
4618 0 : if (bdev_io == bio_to_abort) {
4619 0 : buf_len = bdev_io_get_max_buf_len(bdev_io, bdev_io->internal.buf.len);
4620 0 : spdk_iobuf_entry_abort(ch, entry, buf_len);
4621 0 : spdk_bdev_io_complete(bio_to_abort, SPDK_BDEV_IO_STATUS_ABORTED);
4622 0 : return 1;
4623 : }
4624 :
4625 0 : return 0;
4626 0 : }
4627 :
4628 : static bool
4629 16 : bdev_abort_buf_io(struct spdk_bdev_mgmt_channel *mgmt_ch, struct spdk_bdev_io *bio_to_abort)
4630 : {
4631 : int rc;
4632 :
4633 16 : rc = spdk_iobuf_for_each_entry(&mgmt_ch->iobuf, bdev_abort_buf_io_cb, bio_to_abort);
4634 16 : return rc == 1;
4635 : }
4636 :
4637 : static void
4638 7 : bdev_qos_channel_destroy(void *cb_arg)
4639 : {
4640 7 : struct spdk_bdev_qos *qos = cb_arg;
4641 :
4642 7 : spdk_put_io_channel(spdk_io_channel_from_ctx(qos->ch));
4643 7 : spdk_poller_unregister(&qos->poller);
4644 :
4645 7 : SPDK_DEBUGLOG(bdev, "Free QoS %p.\n", qos);
4646 :
4647 7 : free(qos);
4648 7 : }
4649 :
4650 : static int
4651 7 : bdev_qos_destroy(struct spdk_bdev *bdev)
4652 : {
4653 : int i;
4654 :
4655 : /*
4656 : * Cleanly shutting down the QoS poller is tricky, because
4657 : * during the asynchronous operation the user could open
4658 : * a new descriptor and create a new channel, spawning
4659 : * a new QoS poller.
4660 : *
4661 : * The strategy is to create a new QoS structure here and swap it
4662 : * in. The shutdown path then continues to refer to the old one
4663 : * until it completes and then releases it.
4664 : */
4665 : struct spdk_bdev_qos *new_qos, *old_qos;
4666 :
4667 7 : old_qos = bdev->internal.qos;
4668 :
4669 7 : new_qos = calloc(1, sizeof(*new_qos));
4670 7 : if (!new_qos) {
4671 0 : SPDK_ERRLOG("Unable to allocate memory to shut down QoS.\n");
4672 0 : return -ENOMEM;
4673 : }
4674 :
4675 : /* Copy the old QoS data into the newly allocated structure */
4676 7 : memcpy(new_qos, old_qos, sizeof(*new_qos));
4677 :
4678 : /* Zero out the key parts of the QoS structure */
4679 7 : new_qos->ch = NULL;
4680 7 : new_qos->thread = NULL;
4681 7 : new_qos->poller = NULL;
4682 : /*
4683 : * The limit member of spdk_bdev_qos_limit structure is not zeroed.
4684 : * It will be used later for the new QoS structure.
4685 : */
4686 35 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
4687 28 : new_qos->rate_limits[i].remaining_this_timeslice = 0;
4688 28 : new_qos->rate_limits[i].min_per_timeslice = 0;
4689 28 : new_qos->rate_limits[i].max_per_timeslice = 0;
4690 28 : }
4691 :
4692 7 : bdev->internal.qos = new_qos;
4693 :
4694 7 : if (old_qos->thread == NULL) {
4695 0 : free(old_qos);
4696 0 : } else {
4697 7 : spdk_thread_send_msg(old_qos->thread, bdev_qos_channel_destroy, old_qos);
4698 : }
4699 :
4700 : /* It is safe to continue with destroying the bdev even though the QoS channel hasn't
4701 : * been destroyed yet. The destruction path will end up waiting for the final
4702 : * channel to be put before it releases resources. */
4703 :
4704 7 : return 0;
4705 7 : }
4706 :
4707 : void
4708 78 : spdk_bdev_add_io_stat(struct spdk_bdev_io_stat *total, struct spdk_bdev_io_stat *add)
4709 : {
4710 78 : total->bytes_read += add->bytes_read;
4711 78 : total->num_read_ops += add->num_read_ops;
4712 78 : total->bytes_written += add->bytes_written;
4713 78 : total->num_write_ops += add->num_write_ops;
4714 78 : total->bytes_unmapped += add->bytes_unmapped;
4715 78 : total->num_unmap_ops += add->num_unmap_ops;
4716 78 : total->bytes_copied += add->bytes_copied;
4717 78 : total->num_copy_ops += add->num_copy_ops;
4718 78 : total->read_latency_ticks += add->read_latency_ticks;
4719 78 : total->write_latency_ticks += add->write_latency_ticks;
4720 78 : total->unmap_latency_ticks += add->unmap_latency_ticks;
4721 78 : total->copy_latency_ticks += add->copy_latency_ticks;
4722 78 : if (total->max_read_latency_ticks < add->max_read_latency_ticks) {
4723 7 : total->max_read_latency_ticks = add->max_read_latency_ticks;
4724 7 : }
4725 78 : if (total->min_read_latency_ticks > add->min_read_latency_ticks) {
4726 39 : total->min_read_latency_ticks = add->min_read_latency_ticks;
4727 39 : }
4728 78 : if (total->max_write_latency_ticks < add->max_write_latency_ticks) {
4729 4 : total->max_write_latency_ticks = add->max_write_latency_ticks;
4730 4 : }
4731 78 : if (total->min_write_latency_ticks > add->min_write_latency_ticks) {
4732 24 : total->min_write_latency_ticks = add->min_write_latency_ticks;
4733 24 : }
4734 78 : if (total->max_unmap_latency_ticks < add->max_unmap_latency_ticks) {
4735 0 : total->max_unmap_latency_ticks = add->max_unmap_latency_ticks;
4736 0 : }
4737 78 : if (total->min_unmap_latency_ticks > add->min_unmap_latency_ticks) {
4738 3 : total->min_unmap_latency_ticks = add->min_unmap_latency_ticks;
4739 3 : }
4740 78 : if (total->max_copy_latency_ticks < add->max_copy_latency_ticks) {
4741 0 : total->max_copy_latency_ticks = add->max_copy_latency_ticks;
4742 0 : }
4743 78 : if (total->min_copy_latency_ticks > add->min_copy_latency_ticks) {
4744 4 : total->min_copy_latency_ticks = add->min_copy_latency_ticks;
4745 4 : }
4746 78 : }
4747 :
4748 : static void
4749 5 : bdev_get_io_stat(struct spdk_bdev_io_stat *to_stat, struct spdk_bdev_io_stat *from_stat)
4750 : {
4751 5 : memcpy(to_stat, from_stat, offsetof(struct spdk_bdev_io_stat, io_error));
4752 :
4753 5 : if (to_stat->io_error != NULL && from_stat->io_error != NULL) {
4754 0 : memcpy(to_stat->io_error, from_stat->io_error,
4755 : sizeof(struct spdk_bdev_io_error_stat));
4756 0 : }
4757 5 : }
4758 :
4759 : void
4760 211 : spdk_bdev_reset_io_stat(struct spdk_bdev_io_stat *stat, enum spdk_bdev_reset_stat_mode mode)
4761 : {
4762 211 : if (mode == SPDK_BDEV_RESET_STAT_NONE) {
4763 5 : return;
4764 : }
4765 :
4766 206 : stat->max_read_latency_ticks = 0;
4767 206 : stat->min_read_latency_ticks = UINT64_MAX;
4768 206 : stat->max_write_latency_ticks = 0;
4769 206 : stat->min_write_latency_ticks = UINT64_MAX;
4770 206 : stat->max_unmap_latency_ticks = 0;
4771 206 : stat->min_unmap_latency_ticks = UINT64_MAX;
4772 206 : stat->max_copy_latency_ticks = 0;
4773 206 : stat->min_copy_latency_ticks = UINT64_MAX;
4774 :
4775 206 : if (mode != SPDK_BDEV_RESET_STAT_ALL) {
4776 2 : return;
4777 : }
4778 :
4779 204 : stat->bytes_read = 0;
4780 204 : stat->num_read_ops = 0;
4781 204 : stat->bytes_written = 0;
4782 204 : stat->num_write_ops = 0;
4783 204 : stat->bytes_unmapped = 0;
4784 204 : stat->num_unmap_ops = 0;
4785 204 : stat->bytes_copied = 0;
4786 204 : stat->num_copy_ops = 0;
4787 204 : stat->read_latency_ticks = 0;
4788 204 : stat->write_latency_ticks = 0;
4789 204 : stat->unmap_latency_ticks = 0;
4790 204 : stat->copy_latency_ticks = 0;
4791 :
4792 204 : if (stat->io_error != NULL) {
4793 129 : memset(stat->io_error, 0, sizeof(struct spdk_bdev_io_error_stat));
4794 129 : }
4795 211 : }
4796 :
4797 : struct spdk_bdev_io_stat *
4798 202 : bdev_alloc_io_stat(bool io_error_stat)
4799 : {
4800 : struct spdk_bdev_io_stat *stat;
4801 :
4802 202 : stat = malloc(sizeof(struct spdk_bdev_io_stat));
4803 202 : if (stat == NULL) {
4804 0 : return NULL;
4805 : }
4806 :
4807 202 : if (io_error_stat) {
4808 128 : stat->io_error = malloc(sizeof(struct spdk_bdev_io_error_stat));
4809 128 : if (stat->io_error == NULL) {
4810 0 : free(stat);
4811 0 : return NULL;
4812 : }
4813 128 : } else {
4814 74 : stat->io_error = NULL;
4815 : }
4816 :
4817 202 : spdk_bdev_reset_io_stat(stat, SPDK_BDEV_RESET_STAT_ALL);
4818 :
4819 202 : return stat;
4820 202 : }
4821 :
4822 : void
4823 202 : bdev_free_io_stat(struct spdk_bdev_io_stat *stat)
4824 : {
4825 202 : if (stat != NULL) {
4826 202 : free(stat->io_error);
4827 202 : free(stat);
4828 202 : }
4829 202 : }
4830 :
4831 : void
4832 0 : spdk_bdev_dump_io_stat_json(struct spdk_bdev_io_stat *stat, struct spdk_json_write_ctx *w)
4833 : {
4834 : int i;
4835 :
4836 0 : spdk_json_write_named_uint64(w, "bytes_read", stat->bytes_read);
4837 0 : spdk_json_write_named_uint64(w, "num_read_ops", stat->num_read_ops);
4838 0 : spdk_json_write_named_uint64(w, "bytes_written", stat->bytes_written);
4839 0 : spdk_json_write_named_uint64(w, "num_write_ops", stat->num_write_ops);
4840 0 : spdk_json_write_named_uint64(w, "bytes_unmapped", stat->bytes_unmapped);
4841 0 : spdk_json_write_named_uint64(w, "num_unmap_ops", stat->num_unmap_ops);
4842 0 : spdk_json_write_named_uint64(w, "bytes_copied", stat->bytes_copied);
4843 0 : spdk_json_write_named_uint64(w, "num_copy_ops", stat->num_copy_ops);
4844 0 : spdk_json_write_named_uint64(w, "read_latency_ticks", stat->read_latency_ticks);
4845 0 : spdk_json_write_named_uint64(w, "max_read_latency_ticks", stat->max_read_latency_ticks);
4846 0 : spdk_json_write_named_uint64(w, "min_read_latency_ticks",
4847 0 : stat->min_read_latency_ticks != UINT64_MAX ?
4848 0 : stat->min_read_latency_ticks : 0);
4849 0 : spdk_json_write_named_uint64(w, "write_latency_ticks", stat->write_latency_ticks);
4850 0 : spdk_json_write_named_uint64(w, "max_write_latency_ticks", stat->max_write_latency_ticks);
4851 0 : spdk_json_write_named_uint64(w, "min_write_latency_ticks",
4852 0 : stat->min_write_latency_ticks != UINT64_MAX ?
4853 0 : stat->min_write_latency_ticks : 0);
4854 0 : spdk_json_write_named_uint64(w, "unmap_latency_ticks", stat->unmap_latency_ticks);
4855 0 : spdk_json_write_named_uint64(w, "max_unmap_latency_ticks", stat->max_unmap_latency_ticks);
4856 0 : spdk_json_write_named_uint64(w, "min_unmap_latency_ticks",
4857 0 : stat->min_unmap_latency_ticks != UINT64_MAX ?
4858 0 : stat->min_unmap_latency_ticks : 0);
4859 0 : spdk_json_write_named_uint64(w, "copy_latency_ticks", stat->copy_latency_ticks);
4860 0 : spdk_json_write_named_uint64(w, "max_copy_latency_ticks", stat->max_copy_latency_ticks);
4861 0 : spdk_json_write_named_uint64(w, "min_copy_latency_ticks",
4862 0 : stat->min_copy_latency_ticks != UINT64_MAX ?
4863 0 : stat->min_copy_latency_ticks : 0);
4864 :
4865 0 : if (stat->io_error != NULL) {
4866 0 : spdk_json_write_named_object_begin(w, "io_error");
4867 0 : for (i = 0; i < -SPDK_MIN_BDEV_IO_STATUS; i++) {
4868 0 : if (stat->io_error->error_status[i] != 0) {
4869 0 : spdk_json_write_named_uint32(w, bdev_io_status_get_string(-(i + 1)),
4870 0 : stat->io_error->error_status[i]);
4871 0 : }
4872 0 : }
4873 0 : spdk_json_write_object_end(w);
4874 0 : }
4875 0 : }
4876 :
4877 : static void
4878 78 : bdev_channel_abort_queued_ios(struct spdk_bdev_channel *ch)
4879 : {
4880 78 : struct spdk_bdev_shared_resource *shared_resource = ch->shared_resource;
4881 78 : struct spdk_bdev_mgmt_channel *mgmt_ch = shared_resource->mgmt_ch;
4882 :
4883 78 : bdev_abort_all_queued_io(&shared_resource->nomem_io, ch);
4884 78 : bdev_abort_all_buf_io(mgmt_ch, ch);
4885 78 : }
4886 :
4887 : static void
4888 74 : bdev_channel_destroy(void *io_device, void *ctx_buf)
4889 : {
4890 74 : struct spdk_bdev_channel *ch = ctx_buf;
4891 :
4892 74 : SPDK_DEBUGLOG(bdev, "Destroying channel %p for bdev %s on thread %p\n", ch, ch->bdev->name,
4893 : spdk_get_thread());
4894 :
4895 74 : spdk_trace_record(TRACE_BDEV_IOCH_DESTROY, ch->bdev->internal.trace_id, 0, 0,
4896 : spdk_thread_get_id(spdk_io_channel_get_thread(ch->channel)));
4897 :
4898 : /* This channel is going away, so add its statistics into the bdev so that they don't get lost. */
4899 74 : spdk_spin_lock(&ch->bdev->internal.spinlock);
4900 74 : spdk_bdev_add_io_stat(ch->bdev->internal.stat, ch->stat);
4901 74 : spdk_spin_unlock(&ch->bdev->internal.spinlock);
4902 :
4903 74 : bdev_abort_all_queued_io(&ch->queued_resets, ch);
4904 :
4905 74 : bdev_channel_abort_queued_ios(ch);
4906 :
4907 74 : if (ch->histogram) {
4908 0 : spdk_histogram_data_free(ch->histogram);
4909 0 : }
4910 :
4911 74 : bdev_channel_destroy_resource(ch);
4912 74 : }
4913 :
4914 : /*
4915 : * If the name already exists in the global bdev name tree, RB_INSERT() returns a pointer
4916 : * to it. Hence we do not have to call bdev_get_by_name() when using this function.
4917 : */
4918 : static int
4919 259 : bdev_name_add(struct spdk_bdev_name *bdev_name, struct spdk_bdev *bdev, const char *name)
4920 : {
4921 : struct spdk_bdev_name *tmp;
4922 :
4923 259 : bdev_name->name = strdup(name);
4924 259 : if (bdev_name->name == NULL) {
4925 0 : SPDK_ERRLOG("Unable to allocate bdev name\n");
4926 0 : return -ENOMEM;
4927 : }
4928 :
4929 259 : bdev_name->bdev = bdev;
4930 :
4931 259 : spdk_spin_lock(&g_bdev_mgr.spinlock);
4932 259 : tmp = RB_INSERT(bdev_name_tree, &g_bdev_mgr.bdev_names, bdev_name);
4933 259 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
4934 :
4935 259 : if (tmp != NULL) {
4936 4 : SPDK_ERRLOG("Bdev name %s already exists\n", name);
4937 4 : free(bdev_name->name);
4938 4 : return -EEXIST;
4939 : }
4940 :
4941 255 : return 0;
4942 259 : }
4943 :
4944 : static void
4945 255 : bdev_name_del_unsafe(struct spdk_bdev_name *bdev_name)
4946 : {
4947 255 : RB_REMOVE(bdev_name_tree, &g_bdev_mgr.bdev_names, bdev_name);
4948 255 : free(bdev_name->name);
4949 255 : }
4950 :
4951 : static void
4952 5 : bdev_name_del(struct spdk_bdev_name *bdev_name)
4953 : {
4954 5 : spdk_spin_lock(&g_bdev_mgr.spinlock);
4955 5 : bdev_name_del_unsafe(bdev_name);
4956 5 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
4957 5 : }
4958 :
4959 : int
4960 134 : spdk_bdev_alias_add(struct spdk_bdev *bdev, const char *alias)
4961 : {
4962 : struct spdk_bdev_alias *tmp;
4963 : int ret;
4964 :
4965 134 : if (alias == NULL) {
4966 1 : SPDK_ERRLOG("Empty alias passed\n");
4967 1 : return -EINVAL;
4968 : }
4969 :
4970 133 : tmp = calloc(1, sizeof(*tmp));
4971 133 : if (tmp == NULL) {
4972 0 : SPDK_ERRLOG("Unable to allocate alias\n");
4973 0 : return -ENOMEM;
4974 : }
4975 :
4976 133 : ret = bdev_name_add(&tmp->alias, bdev, alias);
4977 133 : if (ret != 0) {
4978 4 : free(tmp);
4979 4 : return ret;
4980 : }
4981 :
4982 129 : TAILQ_INSERT_TAIL(&bdev->aliases, tmp, tailq);
4983 :
4984 129 : return 0;
4985 134 : }
4986 :
4987 : static int
4988 130 : bdev_alias_del(struct spdk_bdev *bdev, const char *alias,
4989 : void (*alias_del_fn)(struct spdk_bdev_name *n))
4990 : {
4991 : struct spdk_bdev_alias *tmp;
4992 :
4993 135 : TAILQ_FOREACH(tmp, &bdev->aliases, tailq) {
4994 131 : if (strcmp(alias, tmp->alias.name) == 0) {
4995 126 : TAILQ_REMOVE(&bdev->aliases, tmp, tailq);
4996 126 : alias_del_fn(&tmp->alias);
4997 126 : free(tmp);
4998 126 : return 0;
4999 : }
5000 5 : }
5001 :
5002 4 : return -ENOENT;
5003 130 : }
5004 :
5005 : int
5006 4 : spdk_bdev_alias_del(struct spdk_bdev *bdev, const char *alias)
5007 : {
5008 : int rc;
5009 :
5010 4 : rc = bdev_alias_del(bdev, alias, bdev_name_del);
5011 4 : if (rc == -ENOENT) {
5012 2 : SPDK_INFOLOG(bdev, "Alias %s does not exist\n", alias);
5013 2 : }
5014 :
5015 4 : return rc;
5016 : }
5017 :
5018 : void
5019 2 : spdk_bdev_alias_del_all(struct spdk_bdev *bdev)
5020 : {
5021 : struct spdk_bdev_alias *p, *tmp;
5022 :
5023 5 : TAILQ_FOREACH_SAFE(p, &bdev->aliases, tailq, tmp) {
5024 3 : TAILQ_REMOVE(&bdev->aliases, p, tailq);
5025 3 : bdev_name_del(&p->alias);
5026 3 : free(p);
5027 3 : }
5028 2 : }
5029 :
5030 : struct spdk_io_channel *
5031 76 : spdk_bdev_get_io_channel(struct spdk_bdev_desc *desc)
5032 : {
5033 76 : return spdk_get_io_channel(__bdev_to_io_dev(spdk_bdev_desc_get_bdev(desc)));
5034 : }
5035 :
5036 : void *
5037 0 : spdk_bdev_get_module_ctx(struct spdk_bdev_desc *desc)
5038 : {
5039 0 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5040 0 : void *ctx = NULL;
5041 :
5042 0 : if (bdev->fn_table->get_module_ctx) {
5043 0 : ctx = bdev->fn_table->get_module_ctx(bdev->ctxt);
5044 0 : }
5045 :
5046 0 : return ctx;
5047 : }
5048 :
5049 : const char *
5050 0 : spdk_bdev_get_module_name(const struct spdk_bdev *bdev)
5051 : {
5052 0 : return bdev->module->name;
5053 : }
5054 :
5055 : const char *
5056 255 : spdk_bdev_get_name(const struct spdk_bdev *bdev)
5057 : {
5058 255 : return bdev->name;
5059 : }
5060 :
5061 : const char *
5062 0 : spdk_bdev_get_product_name(const struct spdk_bdev *bdev)
5063 : {
5064 0 : return bdev->product_name;
5065 : }
5066 :
5067 : const struct spdk_bdev_aliases_list *
5068 0 : spdk_bdev_get_aliases(const struct spdk_bdev *bdev)
5069 : {
5070 0 : return &bdev->aliases;
5071 : }
5072 :
5073 : uint32_t
5074 5 : spdk_bdev_get_block_size(const struct spdk_bdev *bdev)
5075 : {
5076 5 : return bdev->blocklen;
5077 : }
5078 :
5079 : uint32_t
5080 0 : spdk_bdev_get_write_unit_size(const struct spdk_bdev *bdev)
5081 : {
5082 0 : return bdev->write_unit_size;
5083 : }
5084 :
5085 : uint64_t
5086 0 : spdk_bdev_get_num_blocks(const struct spdk_bdev *bdev)
5087 : {
5088 0 : return bdev->blockcnt;
5089 : }
5090 :
5091 : const char *
5092 0 : spdk_bdev_get_qos_rpc_type(enum spdk_bdev_qos_rate_limit_type type)
5093 : {
5094 0 : return qos_rpc_type[type];
5095 : }
5096 :
5097 : void
5098 0 : spdk_bdev_get_qos_rate_limits(struct spdk_bdev *bdev, uint64_t *limits)
5099 : {
5100 : int i;
5101 :
5102 0 : memset(limits, 0, sizeof(*limits) * SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES);
5103 :
5104 0 : spdk_spin_lock(&bdev->internal.spinlock);
5105 0 : if (bdev->internal.qos) {
5106 0 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
5107 0 : if (bdev->internal.qos->rate_limits[i].limit !=
5108 : SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
5109 0 : limits[i] = bdev->internal.qos->rate_limits[i].limit;
5110 0 : if (bdev_qos_is_iops_rate_limit(i) == false) {
5111 : /* Change from Byte to Megabyte which is user visible. */
5112 0 : limits[i] = limits[i] / 1024 / 1024;
5113 0 : }
5114 0 : }
5115 0 : }
5116 0 : }
5117 0 : spdk_spin_unlock(&bdev->internal.spinlock);
5118 0 : }
5119 :
5120 : size_t
5121 276 : spdk_bdev_get_buf_align(const struct spdk_bdev *bdev)
5122 : {
5123 276 : return 1 << bdev->required_alignment;
5124 : }
5125 :
5126 : uint32_t
5127 0 : spdk_bdev_get_optimal_io_boundary(const struct spdk_bdev *bdev)
5128 : {
5129 0 : return bdev->optimal_io_boundary;
5130 : }
5131 :
5132 : bool
5133 0 : spdk_bdev_has_write_cache(const struct spdk_bdev *bdev)
5134 : {
5135 0 : return bdev->write_cache;
5136 : }
5137 :
5138 : const struct spdk_uuid *
5139 0 : spdk_bdev_get_uuid(const struct spdk_bdev *bdev)
5140 : {
5141 0 : return &bdev->uuid;
5142 : }
5143 :
5144 : uint16_t
5145 0 : spdk_bdev_get_acwu(const struct spdk_bdev *bdev)
5146 : {
5147 0 : return bdev->acwu;
5148 : }
5149 :
5150 : uint32_t
5151 29 : spdk_bdev_get_md_size(const struct spdk_bdev *bdev)
5152 : {
5153 29 : return bdev->md_len;
5154 : }
5155 :
5156 : bool
5157 131 : spdk_bdev_is_md_interleaved(const struct spdk_bdev *bdev)
5158 : {
5159 131 : return (bdev->md_len != 0) && bdev->md_interleave;
5160 : }
5161 :
5162 : bool
5163 117 : spdk_bdev_is_md_separate(const struct spdk_bdev *bdev)
5164 : {
5165 117 : return (bdev->md_len != 0) && !bdev->md_interleave;
5166 : }
5167 :
5168 : bool
5169 0 : spdk_bdev_is_zoned(const struct spdk_bdev *bdev)
5170 : {
5171 0 : return bdev->zoned;
5172 : }
5173 :
5174 : uint32_t
5175 122 : spdk_bdev_get_data_block_size(const struct spdk_bdev *bdev)
5176 : {
5177 122 : if (spdk_bdev_is_md_interleaved(bdev)) {
5178 0 : return bdev->blocklen - bdev->md_len;
5179 : } else {
5180 122 : return bdev->blocklen;
5181 : }
5182 122 : }
5183 :
5184 : uint32_t
5185 0 : spdk_bdev_get_physical_block_size(const struct spdk_bdev *bdev)
5186 : {
5187 0 : return bdev->phys_blocklen;
5188 : }
5189 :
5190 : static uint32_t
5191 9 : _bdev_get_block_size_with_md(const struct spdk_bdev *bdev)
5192 : {
5193 9 : if (!spdk_bdev_is_md_interleaved(bdev)) {
5194 6 : return bdev->blocklen + bdev->md_len;
5195 : } else {
5196 3 : return bdev->blocklen;
5197 : }
5198 9 : }
5199 :
5200 : /* We have to use the typedef in the function declaration to appease astyle. */
5201 : typedef enum spdk_dif_type spdk_dif_type_t;
5202 : typedef enum spdk_dif_pi_format spdk_dif_pi_format_t;
5203 :
5204 : spdk_dif_type_t
5205 0 : spdk_bdev_get_dif_type(const struct spdk_bdev *bdev)
5206 : {
5207 0 : if (bdev->md_len != 0) {
5208 0 : return bdev->dif_type;
5209 : } else {
5210 0 : return SPDK_DIF_DISABLE;
5211 : }
5212 0 : }
5213 :
5214 : spdk_dif_pi_format_t
5215 0 : spdk_bdev_get_dif_pi_format(const struct spdk_bdev *bdev)
5216 : {
5217 0 : return bdev->dif_pi_format;
5218 : }
5219 :
5220 : bool
5221 0 : spdk_bdev_is_dif_head_of_md(const struct spdk_bdev *bdev)
5222 : {
5223 0 : if (spdk_bdev_get_dif_type(bdev) != SPDK_DIF_DISABLE) {
5224 0 : return bdev->dif_is_head_of_md;
5225 : } else {
5226 0 : return false;
5227 : }
5228 0 : }
5229 :
5230 : bool
5231 0 : spdk_bdev_is_dif_check_enabled(const struct spdk_bdev *bdev,
5232 : enum spdk_dif_check_type check_type)
5233 : {
5234 0 : if (spdk_bdev_get_dif_type(bdev) == SPDK_DIF_DISABLE) {
5235 0 : return false;
5236 : }
5237 :
5238 0 : switch (check_type) {
5239 : case SPDK_DIF_CHECK_TYPE_REFTAG:
5240 0 : return (bdev->dif_check_flags & SPDK_DIF_FLAGS_REFTAG_CHECK) != 0;
5241 : case SPDK_DIF_CHECK_TYPE_APPTAG:
5242 0 : return (bdev->dif_check_flags & SPDK_DIF_FLAGS_APPTAG_CHECK) != 0;
5243 : case SPDK_DIF_CHECK_TYPE_GUARD:
5244 0 : return (bdev->dif_check_flags & SPDK_DIF_FLAGS_GUARD_CHECK) != 0;
5245 : default:
5246 0 : return false;
5247 : }
5248 0 : }
5249 :
5250 : static uint32_t
5251 3 : bdev_get_max_write(const struct spdk_bdev *bdev, uint64_t num_bytes)
5252 : {
5253 : uint64_t aligned_length, max_write_blocks;
5254 :
5255 3 : aligned_length = num_bytes - (spdk_bdev_get_buf_align(bdev) - 1);
5256 3 : max_write_blocks = aligned_length / _bdev_get_block_size_with_md(bdev);
5257 3 : max_write_blocks -= max_write_blocks % bdev->write_unit_size;
5258 :
5259 3 : return max_write_blocks;
5260 : }
5261 :
5262 : uint32_t
5263 1 : spdk_bdev_get_max_copy(const struct spdk_bdev *bdev)
5264 : {
5265 1 : return bdev->max_copy;
5266 : }
5267 :
5268 : uint64_t
5269 0 : spdk_bdev_get_qd(const struct spdk_bdev *bdev)
5270 : {
5271 0 : return bdev->internal.measured_queue_depth;
5272 : }
5273 :
5274 : uint64_t
5275 0 : spdk_bdev_get_qd_sampling_period(const struct spdk_bdev *bdev)
5276 : {
5277 0 : return bdev->internal.period;
5278 : }
5279 :
5280 : uint64_t
5281 0 : spdk_bdev_get_weighted_io_time(const struct spdk_bdev *bdev)
5282 : {
5283 0 : return bdev->internal.weighted_io_time;
5284 : }
5285 :
5286 : uint64_t
5287 0 : spdk_bdev_get_io_time(const struct spdk_bdev *bdev)
5288 : {
5289 0 : return bdev->internal.io_time;
5290 : }
5291 :
5292 0 : union spdk_bdev_nvme_ctratt spdk_bdev_get_nvme_ctratt(struct spdk_bdev *bdev)
5293 : {
5294 0 : return bdev->ctratt;
5295 : }
5296 :
5297 : uint32_t
5298 0 : spdk_bdev_get_nvme_nsid(struct spdk_bdev *bdev)
5299 : {
5300 0 : return bdev->nsid;
5301 : }
5302 :
5303 : uint32_t
5304 0 : spdk_bdev_desc_get_block_size(struct spdk_bdev_desc *desc)
5305 : {
5306 0 : struct spdk_bdev *bdev = desc->bdev;
5307 :
5308 0 : return desc->opts.no_metadata ? bdev->blocklen - bdev->md_len : bdev->blocklen;
5309 : }
5310 :
5311 : uint32_t
5312 0 : spdk_bdev_desc_get_md_size(struct spdk_bdev_desc *desc)
5313 : {
5314 0 : struct spdk_bdev *bdev = desc->bdev;
5315 :
5316 0 : return desc->opts.no_metadata ? 0 : bdev->md_len;
5317 : }
5318 :
5319 : bool
5320 0 : spdk_bdev_desc_is_md_interleaved(struct spdk_bdev_desc *desc)
5321 : {
5322 0 : struct spdk_bdev *bdev = desc->bdev;
5323 :
5324 0 : return desc->opts.no_metadata ? false : spdk_bdev_is_md_interleaved(bdev);
5325 : }
5326 :
5327 : bool
5328 0 : spdk_bdev_desc_is_md_separate(struct spdk_bdev_desc *desc)
5329 : {
5330 0 : struct spdk_bdev *bdev = desc->bdev;
5331 :
5332 0 : return desc->opts.no_metadata ? false : spdk_bdev_is_md_separate(bdev);
5333 : }
5334 :
5335 : spdk_dif_type_t
5336 0 : spdk_bdev_desc_get_dif_type(struct spdk_bdev_desc *desc)
5337 : {
5338 0 : struct spdk_bdev *bdev = desc->bdev;
5339 :
5340 0 : return desc->opts.no_metadata ? SPDK_DIF_DISABLE : spdk_bdev_get_dif_type(bdev);
5341 : }
5342 :
5343 : spdk_dif_pi_format_t
5344 0 : spdk_bdev_desc_get_dif_pi_format(struct spdk_bdev_desc *desc)
5345 : {
5346 0 : struct spdk_bdev *bdev = desc->bdev;
5347 :
5348 0 : return desc->opts.no_metadata ? SPDK_DIF_PI_FORMAT_16 : spdk_bdev_get_dif_pi_format(bdev);
5349 : }
5350 :
5351 : bool
5352 0 : spdk_bdev_desc_is_dif_head_of_md(struct spdk_bdev_desc *desc)
5353 : {
5354 0 : struct spdk_bdev *bdev = desc->bdev;
5355 :
5356 0 : return desc->opts.no_metadata ? false : spdk_bdev_is_dif_head_of_md(bdev);
5357 : }
5358 :
5359 : bool
5360 0 : spdk_bdev_desc_is_dif_check_enabled(struct spdk_bdev_desc *desc,
5361 : enum spdk_dif_check_type check_type)
5362 : {
5363 0 : struct spdk_bdev *bdev = desc->bdev;
5364 :
5365 0 : return desc->opts.no_metadata ? false : spdk_bdev_is_dif_check_enabled(bdev, check_type);
5366 : }
5367 :
5368 : static void bdev_update_qd_sampling_period(void *ctx);
5369 :
5370 : static void
5371 1 : _calculate_measured_qd_cpl(struct spdk_bdev *bdev, void *_ctx, int status)
5372 : {
5373 1 : bdev->internal.measured_queue_depth = bdev->internal.temporary_queue_depth;
5374 :
5375 1 : if (bdev->internal.measured_queue_depth) {
5376 0 : bdev->internal.io_time += bdev->internal.period;
5377 0 : bdev->internal.weighted_io_time += bdev->internal.period * bdev->internal.measured_queue_depth;
5378 0 : }
5379 :
5380 1 : bdev->internal.qd_poll_in_progress = false;
5381 :
5382 1 : bdev_update_qd_sampling_period(bdev);
5383 1 : }
5384 :
5385 : static void
5386 1 : _calculate_measured_qd(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
5387 : struct spdk_io_channel *io_ch, void *_ctx)
5388 : {
5389 1 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(io_ch);
5390 :
5391 1 : bdev->internal.temporary_queue_depth += ch->io_outstanding;
5392 1 : spdk_bdev_for_each_channel_continue(i, 0);
5393 1 : }
5394 :
5395 : static int
5396 1 : bdev_calculate_measured_queue_depth(void *ctx)
5397 : {
5398 1 : struct spdk_bdev *bdev = ctx;
5399 :
5400 1 : bdev->internal.qd_poll_in_progress = true;
5401 1 : bdev->internal.temporary_queue_depth = 0;
5402 1 : spdk_bdev_for_each_channel(bdev, _calculate_measured_qd, bdev, _calculate_measured_qd_cpl);
5403 1 : return SPDK_POLLER_BUSY;
5404 : }
5405 :
5406 : static void
5407 5 : bdev_update_qd_sampling_period(void *ctx)
5408 : {
5409 5 : struct spdk_bdev *bdev = ctx;
5410 :
5411 5 : if (bdev->internal.period == bdev->internal.new_period) {
5412 0 : return;
5413 : }
5414 :
5415 5 : if (bdev->internal.qd_poll_in_progress) {
5416 1 : return;
5417 : }
5418 :
5419 4 : bdev->internal.period = bdev->internal.new_period;
5420 :
5421 4 : spdk_poller_unregister(&bdev->internal.qd_poller);
5422 4 : if (bdev->internal.period != 0) {
5423 2 : bdev->internal.qd_poller = SPDK_POLLER_REGISTER(bdev_calculate_measured_queue_depth,
5424 : bdev, bdev->internal.period);
5425 2 : } else {
5426 2 : spdk_bdev_close(bdev->internal.qd_desc);
5427 2 : bdev->internal.qd_desc = NULL;
5428 : }
5429 5 : }
5430 :
5431 : static void
5432 0 : _tmp_bdev_event_cb(enum spdk_bdev_event_type type, struct spdk_bdev *bdev, void *ctx)
5433 : {
5434 0 : SPDK_NOTICELOG("Unexpected event type: %d\n", type);
5435 0 : }
5436 :
5437 : void
5438 131 : spdk_bdev_set_qd_sampling_period(struct spdk_bdev *bdev, uint64_t period)
5439 : {
5440 : int rc;
5441 :
5442 131 : if (bdev->internal.new_period == period) {
5443 125 : return;
5444 : }
5445 :
5446 6 : bdev->internal.new_period = period;
5447 :
5448 6 : if (bdev->internal.qd_desc != NULL) {
5449 4 : assert(bdev->internal.period != 0);
5450 :
5451 8 : spdk_thread_send_msg(bdev->internal.qd_desc->thread,
5452 4 : bdev_update_qd_sampling_period, bdev);
5453 4 : return;
5454 : }
5455 :
5456 2 : assert(bdev->internal.period == 0);
5457 :
5458 4 : rc = spdk_bdev_open_ext(spdk_bdev_get_name(bdev), false, _tmp_bdev_event_cb,
5459 2 : NULL, &bdev->internal.qd_desc);
5460 2 : if (rc != 0) {
5461 0 : return;
5462 : }
5463 :
5464 2 : bdev->internal.period = period;
5465 2 : bdev->internal.qd_poller = SPDK_POLLER_REGISTER(bdev_calculate_measured_queue_depth,
5466 : bdev, period);
5467 131 : }
5468 :
5469 : struct bdev_get_current_qd_ctx {
5470 : uint64_t current_qd;
5471 : spdk_bdev_get_current_qd_cb cb_fn;
5472 : void *cb_arg;
5473 : };
5474 :
5475 : static void
5476 0 : bdev_get_current_qd_done(struct spdk_bdev *bdev, void *_ctx, int status)
5477 : {
5478 0 : struct bdev_get_current_qd_ctx *ctx = _ctx;
5479 :
5480 0 : ctx->cb_fn(bdev, ctx->current_qd, ctx->cb_arg, 0);
5481 :
5482 0 : free(ctx);
5483 0 : }
5484 :
5485 : static void
5486 0 : bdev_get_current_qd(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
5487 : struct spdk_io_channel *io_ch, void *_ctx)
5488 : {
5489 0 : struct bdev_get_current_qd_ctx *ctx = _ctx;
5490 0 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
5491 :
5492 0 : ctx->current_qd += bdev_ch->io_outstanding;
5493 :
5494 0 : spdk_bdev_for_each_channel_continue(i, 0);
5495 0 : }
5496 :
5497 : void
5498 0 : spdk_bdev_get_current_qd(struct spdk_bdev *bdev, spdk_bdev_get_current_qd_cb cb_fn,
5499 : void *cb_arg)
5500 : {
5501 : struct bdev_get_current_qd_ctx *ctx;
5502 :
5503 0 : assert(cb_fn != NULL);
5504 :
5505 0 : ctx = calloc(1, sizeof(*ctx));
5506 0 : if (ctx == NULL) {
5507 0 : cb_fn(bdev, 0, cb_arg, -ENOMEM);
5508 0 : return;
5509 : }
5510 :
5511 0 : ctx->cb_fn = cb_fn;
5512 0 : ctx->cb_arg = cb_arg;
5513 :
5514 0 : spdk_bdev_for_each_channel(bdev, bdev_get_current_qd, ctx, bdev_get_current_qd_done);
5515 0 : }
5516 :
5517 : static void
5518 25 : _event_notify(struct spdk_bdev_desc *desc, enum spdk_bdev_event_type type)
5519 : {
5520 25 : assert(desc->thread == spdk_get_thread());
5521 :
5522 25 : spdk_spin_lock(&desc->spinlock);
5523 25 : desc->refs--;
5524 25 : if (!desc->closed) {
5525 14 : spdk_spin_unlock(&desc->spinlock);
5526 28 : desc->callback.event_fn(type,
5527 14 : desc->bdev,
5528 14 : desc->callback.ctx);
5529 14 : return;
5530 11 : } else if (desc->refs == 0) {
5531 : /* This descriptor was closed after this event_notify message was sent.
5532 : * spdk_bdev_close() could not free the descriptor since this message was
5533 : * in flight, so we free it now using bdev_desc_free().
5534 : */
5535 10 : spdk_spin_unlock(&desc->spinlock);
5536 10 : bdev_desc_free(desc);
5537 10 : return;
5538 : }
5539 1 : spdk_spin_unlock(&desc->spinlock);
5540 25 : }
5541 :
5542 : static void
5543 25 : event_notify(struct spdk_bdev_desc *desc, spdk_msg_fn event_notify_fn)
5544 : {
5545 25 : spdk_spin_lock(&desc->spinlock);
5546 25 : desc->refs++;
5547 25 : spdk_thread_send_msg(desc->thread, event_notify_fn, desc);
5548 25 : spdk_spin_unlock(&desc->spinlock);
5549 25 : }
5550 :
5551 : static void
5552 6 : _resize_notify(void *ctx)
5553 : {
5554 6 : struct spdk_bdev_desc *desc = ctx;
5555 :
5556 6 : _event_notify(desc, SPDK_BDEV_EVENT_RESIZE);
5557 6 : }
5558 :
5559 : int
5560 11 : spdk_bdev_notify_blockcnt_change(struct spdk_bdev *bdev, uint64_t size)
5561 : {
5562 : struct spdk_bdev_desc *desc;
5563 : int ret;
5564 :
5565 11 : if (size == bdev->blockcnt) {
5566 0 : return 0;
5567 : }
5568 :
5569 11 : spdk_spin_lock(&bdev->internal.spinlock);
5570 :
5571 : /* bdev has open descriptors */
5572 11 : if (!TAILQ_EMPTY(&bdev->internal.open_descs) &&
5573 7 : bdev->blockcnt > size) {
5574 1 : ret = -EBUSY;
5575 1 : } else {
5576 10 : bdev->blockcnt = size;
5577 16 : TAILQ_FOREACH(desc, &bdev->internal.open_descs, link) {
5578 6 : event_notify(desc, _resize_notify);
5579 6 : }
5580 10 : ret = 0;
5581 : }
5582 :
5583 11 : spdk_spin_unlock(&bdev->internal.spinlock);
5584 :
5585 11 : return ret;
5586 11 : }
5587 :
5588 : /*
5589 : * Convert I/O offset and length from bytes to blocks.
5590 : *
5591 : * Returns zero on success or non-zero if the byte parameters aren't divisible by the block size.
5592 : */
5593 : static uint64_t
5594 20 : bdev_bytes_to_blocks(struct spdk_bdev_desc *desc, uint64_t offset_bytes,
5595 : uint64_t *offset_blocks, uint64_t num_bytes, uint64_t *num_blocks)
5596 : {
5597 20 : uint32_t block_size = bdev_desc_get_block_size(desc);
5598 : uint8_t shift_cnt;
5599 :
5600 : /* Avoid expensive div operations if possible. These spdk_u32 functions are very cheap. */
5601 20 : if (spdk_likely(spdk_u32_is_pow2(block_size))) {
5602 17 : shift_cnt = spdk_u32log2(block_size);
5603 17 : *offset_blocks = offset_bytes >> shift_cnt;
5604 17 : *num_blocks = num_bytes >> shift_cnt;
5605 34 : return (offset_bytes - (*offset_blocks << shift_cnt)) |
5606 17 : (num_bytes - (*num_blocks << shift_cnt));
5607 : } else {
5608 3 : *offset_blocks = offset_bytes / block_size;
5609 3 : *num_blocks = num_bytes / block_size;
5610 3 : return (offset_bytes % block_size) | (num_bytes % block_size);
5611 : }
5612 20 : }
5613 :
5614 : static bool
5615 689 : bdev_io_valid_blocks(struct spdk_bdev *bdev, uint64_t offset_blocks, uint64_t num_blocks)
5616 : {
5617 : /* Return failure if offset_blocks + num_blocks is less than offset_blocks; indicates there
5618 : * has been an overflow and hence the offset has been wrapped around */
5619 689 : if (offset_blocks + num_blocks < offset_blocks) {
5620 1 : return false;
5621 : }
5622 :
5623 : /* Return failure if offset_blocks + num_blocks exceeds the size of the bdev */
5624 688 : if (offset_blocks + num_blocks > bdev->blockcnt) {
5625 2 : return false;
5626 : }
5627 :
5628 686 : return true;
5629 689 : }
5630 :
5631 : static void
5632 2 : bdev_seek_complete_cb(void *ctx)
5633 : {
5634 2 : struct spdk_bdev_io *bdev_io = ctx;
5635 :
5636 2 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
5637 2 : bdev_io->internal.cb(bdev_io, true, bdev_io->internal.caller_ctx);
5638 2 : }
5639 :
5640 : static int
5641 4 : bdev_seek(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5642 : uint64_t offset_blocks, enum spdk_bdev_io_type io_type,
5643 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5644 : {
5645 4 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5646 : struct spdk_bdev_io *bdev_io;
5647 4 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5648 :
5649 4 : assert(io_type == SPDK_BDEV_IO_TYPE_SEEK_DATA || io_type == SPDK_BDEV_IO_TYPE_SEEK_HOLE);
5650 :
5651 : /* Check if offset_blocks is valid looking at the validity of one block */
5652 4 : if (!bdev_io_valid_blocks(bdev, offset_blocks, 1)) {
5653 0 : return -EINVAL;
5654 : }
5655 :
5656 4 : bdev_io = bdev_channel_get_io(channel);
5657 4 : if (!bdev_io) {
5658 0 : return -ENOMEM;
5659 : }
5660 :
5661 4 : bdev_io->internal.ch = channel;
5662 4 : bdev_io->internal.desc = desc;
5663 4 : bdev_io->type = io_type;
5664 4 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5665 4 : bdev_io->u.bdev.memory_domain = NULL;
5666 4 : bdev_io->u.bdev.memory_domain_ctx = NULL;
5667 4 : bdev_io->u.bdev.accel_sequence = NULL;
5668 4 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5669 :
5670 4 : if (!spdk_bdev_io_type_supported(bdev, io_type)) {
5671 : /* In case bdev doesn't support seek to next data/hole offset,
5672 : * it is assumed that only data and no holes are present */
5673 2 : if (io_type == SPDK_BDEV_IO_TYPE_SEEK_DATA) {
5674 1 : bdev_io->u.bdev.seek.offset = offset_blocks;
5675 1 : } else {
5676 1 : bdev_io->u.bdev.seek.offset = UINT64_MAX;
5677 : }
5678 :
5679 2 : spdk_thread_send_msg(spdk_get_thread(), bdev_seek_complete_cb, bdev_io);
5680 2 : return 0;
5681 : }
5682 :
5683 2 : bdev_io_submit(bdev_io);
5684 2 : return 0;
5685 4 : }
5686 :
5687 : int
5688 2 : spdk_bdev_seek_data(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5689 : uint64_t offset_blocks,
5690 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5691 : {
5692 2 : return bdev_seek(desc, ch, offset_blocks, SPDK_BDEV_IO_TYPE_SEEK_DATA, cb, cb_arg);
5693 : }
5694 :
5695 : int
5696 2 : spdk_bdev_seek_hole(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5697 : uint64_t offset_blocks,
5698 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5699 : {
5700 2 : return bdev_seek(desc, ch, offset_blocks, SPDK_BDEV_IO_TYPE_SEEK_HOLE, cb, cb_arg);
5701 : }
5702 :
5703 : uint64_t
5704 4 : spdk_bdev_io_get_seek_offset(const struct spdk_bdev_io *bdev_io)
5705 : {
5706 4 : return bdev_io->u.bdev.seek.offset;
5707 : }
5708 :
5709 : static int
5710 204 : bdev_read_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch, void *buf,
5711 : void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
5712 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5713 : {
5714 204 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5715 : struct spdk_bdev_io *bdev_io;
5716 204 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5717 :
5718 204 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
5719 0 : return -EINVAL;
5720 : }
5721 :
5722 204 : bdev_io = bdev_channel_get_io(channel);
5723 204 : if (!bdev_io) {
5724 1 : return -ENOMEM;
5725 : }
5726 :
5727 203 : bdev_io->internal.ch = channel;
5728 203 : bdev_io->internal.desc = desc;
5729 203 : bdev_io->type = SPDK_BDEV_IO_TYPE_READ;
5730 203 : bdev_io->u.bdev.iovs = &bdev_io->iov;
5731 203 : bdev_io->u.bdev.iovs[0].iov_base = buf;
5732 203 : bdev_io->u.bdev.iovs[0].iov_len = num_blocks * bdev_desc_get_block_size(desc);
5733 203 : bdev_io->u.bdev.iovcnt = 1;
5734 203 : bdev_io->u.bdev.md_buf = md_buf;
5735 203 : bdev_io->u.bdev.num_blocks = num_blocks;
5736 203 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5737 203 : bdev_io->u.bdev.memory_domain = NULL;
5738 203 : bdev_io->u.bdev.memory_domain_ctx = NULL;
5739 203 : bdev_io->u.bdev.accel_sequence = NULL;
5740 203 : bdev_io->u.bdev.dif_check_flags = bdev->dif_check_flags;
5741 203 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5742 :
5743 203 : bdev_io_submit(bdev_io);
5744 203 : return 0;
5745 204 : }
5746 :
5747 : int
5748 3 : spdk_bdev_read(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5749 : void *buf, uint64_t offset, uint64_t nbytes,
5750 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5751 : {
5752 : uint64_t offset_blocks, num_blocks;
5753 :
5754 3 : if (bdev_bytes_to_blocks(desc, offset, &offset_blocks, nbytes, &num_blocks) != 0) {
5755 0 : return -EINVAL;
5756 : }
5757 :
5758 3 : return spdk_bdev_read_blocks(desc, ch, buf, offset_blocks, num_blocks, cb, cb_arg);
5759 3 : }
5760 :
5761 : int
5762 200 : spdk_bdev_read_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5763 : void *buf, uint64_t offset_blocks, uint64_t num_blocks,
5764 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5765 : {
5766 200 : return bdev_read_blocks_with_md(desc, ch, buf, NULL, offset_blocks, num_blocks, cb, cb_arg);
5767 : }
5768 :
5769 : int
5770 4 : spdk_bdev_read_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5771 : void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
5772 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5773 : {
5774 8 : struct iovec iov = {
5775 4 : .iov_base = buf,
5776 : };
5777 :
5778 4 : if (md_buf && !spdk_bdev_is_md_separate(spdk_bdev_desc_get_bdev(desc))) {
5779 0 : return -EINVAL;
5780 : }
5781 :
5782 4 : if ((md_buf || desc->opts.no_metadata) && !_is_buf_allocated(&iov)) {
5783 0 : return -EINVAL;
5784 : }
5785 :
5786 8 : return bdev_read_blocks_with_md(desc, ch, buf, md_buf, offset_blocks, num_blocks,
5787 4 : cb, cb_arg);
5788 4 : }
5789 :
5790 : int
5791 5 : spdk_bdev_readv(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5792 : struct iovec *iov, int iovcnt,
5793 : uint64_t offset, uint64_t nbytes,
5794 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5795 : {
5796 : uint64_t offset_blocks, num_blocks;
5797 :
5798 5 : if (bdev_bytes_to_blocks(desc, offset, &offset_blocks, nbytes, &num_blocks) != 0) {
5799 0 : return -EINVAL;
5800 : }
5801 :
5802 5 : return spdk_bdev_readv_blocks(desc, ch, iov, iovcnt, offset_blocks, num_blocks, cb, cb_arg);
5803 5 : }
5804 :
5805 : static int
5806 226 : bdev_readv_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5807 : struct iovec *iov, int iovcnt, void *md_buf, uint64_t offset_blocks,
5808 : uint64_t num_blocks, struct spdk_memory_domain *domain, void *domain_ctx,
5809 : struct spdk_accel_sequence *seq, uint32_t dif_check_flags,
5810 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5811 : {
5812 226 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5813 : struct spdk_bdev_io *bdev_io;
5814 226 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5815 :
5816 226 : if (spdk_unlikely(!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks))) {
5817 0 : return -EINVAL;
5818 : }
5819 :
5820 226 : bdev_io = bdev_channel_get_io(channel);
5821 226 : if (spdk_unlikely(!bdev_io)) {
5822 2 : return -ENOMEM;
5823 : }
5824 :
5825 224 : bdev_io->internal.ch = channel;
5826 224 : bdev_io->internal.desc = desc;
5827 224 : bdev_io->type = SPDK_BDEV_IO_TYPE_READ;
5828 224 : bdev_io->u.bdev.iovs = iov;
5829 224 : bdev_io->u.bdev.iovcnt = iovcnt;
5830 224 : bdev_io->u.bdev.md_buf = md_buf;
5831 224 : bdev_io->u.bdev.num_blocks = num_blocks;
5832 224 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5833 224 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5834 :
5835 224 : if (seq != NULL) {
5836 0 : bdev_io->internal.f.has_accel_sequence = true;
5837 0 : bdev_io->internal.accel_sequence = seq;
5838 0 : }
5839 :
5840 224 : if (domain != NULL) {
5841 2 : bdev_io->internal.f.has_memory_domain = true;
5842 2 : bdev_io->internal.memory_domain = domain;
5843 2 : bdev_io->internal.memory_domain_ctx = domain_ctx;
5844 2 : }
5845 :
5846 224 : bdev_io->u.bdev.memory_domain = domain;
5847 224 : bdev_io->u.bdev.memory_domain_ctx = domain_ctx;
5848 224 : bdev_io->u.bdev.accel_sequence = seq;
5849 224 : bdev_io->u.bdev.dif_check_flags = dif_check_flags;
5850 :
5851 224 : _bdev_io_submit_ext(desc, bdev_io);
5852 :
5853 224 : return 0;
5854 226 : }
5855 :
5856 : int
5857 21 : spdk_bdev_readv_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5858 : struct iovec *iov, int iovcnt,
5859 : uint64_t offset_blocks, uint64_t num_blocks,
5860 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5861 : {
5862 21 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5863 :
5864 42 : return bdev_readv_blocks_with_md(desc, ch, iov, iovcnt, NULL, offset_blocks,
5865 21 : num_blocks, NULL, NULL, NULL, bdev->dif_check_flags, cb, cb_arg);
5866 : }
5867 :
5868 : int
5869 4 : spdk_bdev_readv_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5870 : struct iovec *iov, int iovcnt, void *md_buf,
5871 : uint64_t offset_blocks, uint64_t num_blocks,
5872 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5873 : {
5874 4 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5875 :
5876 4 : if (md_buf && !spdk_bdev_is_md_separate(bdev)) {
5877 0 : return -EINVAL;
5878 : }
5879 :
5880 4 : if (md_buf && !_is_buf_allocated(iov)) {
5881 0 : return -EINVAL;
5882 : }
5883 :
5884 8 : return bdev_readv_blocks_with_md(desc, ch, iov, iovcnt, md_buf, offset_blocks,
5885 4 : num_blocks, NULL, NULL, NULL, bdev->dif_check_flags, cb, cb_arg);
5886 4 : }
5887 :
5888 : static inline bool
5889 14 : _bdev_io_check_opts(struct spdk_bdev_ext_io_opts *opts, struct iovec *iov)
5890 : {
5891 : /*
5892 : * We check if opts size is at least of size when we first introduced
5893 : * spdk_bdev_ext_io_opts (ac6f2bdd8d) since access to those members
5894 : * are not checked internal.
5895 : */
5896 24 : return opts->size >= offsetof(struct spdk_bdev_ext_io_opts, metadata) +
5897 14 : sizeof(opts->metadata) &&
5898 10 : opts->size <= sizeof(*opts) &&
5899 : /* When memory domain is used, the user must provide data buffers */
5900 8 : (!opts->memory_domain || (iov && iov[0].iov_base));
5901 : }
5902 :
5903 : int
5904 8 : spdk_bdev_readv_blocks_ext(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5905 : struct iovec *iov, int iovcnt,
5906 : uint64_t offset_blocks, uint64_t num_blocks,
5907 : spdk_bdev_io_completion_cb cb, void *cb_arg,
5908 : struct spdk_bdev_ext_io_opts *opts)
5909 : {
5910 8 : struct spdk_memory_domain *domain = NULL;
5911 8 : struct spdk_accel_sequence *seq = NULL;
5912 8 : void *domain_ctx = NULL, *md = NULL;
5913 8 : uint32_t dif_check_flags = 0;
5914 : uint32_t nvme_cdw12_raw;
5915 8 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5916 :
5917 8 : if (opts) {
5918 7 : if (spdk_unlikely(!_bdev_io_check_opts(opts, iov))) {
5919 3 : return -EINVAL;
5920 : }
5921 :
5922 4 : md = opts->metadata;
5923 4 : domain = bdev_get_ext_io_opt(opts, memory_domain, NULL);
5924 4 : domain_ctx = bdev_get_ext_io_opt(opts, memory_domain_ctx, NULL);
5925 4 : seq = bdev_get_ext_io_opt(opts, accel_sequence, NULL);
5926 4 : nvme_cdw12_raw = bdev_get_ext_io_opt(opts, nvme_cdw12.raw, 0);
5927 4 : if (md) {
5928 4 : if (spdk_unlikely(!spdk_bdev_is_md_separate(bdev))) {
5929 0 : return -EINVAL;
5930 : }
5931 :
5932 4 : if (spdk_unlikely(!_is_buf_allocated(iov))) {
5933 0 : return -EINVAL;
5934 : }
5935 :
5936 4 : if (spdk_unlikely(seq != NULL)) {
5937 0 : return -EINVAL;
5938 : }
5939 :
5940 4 : if (nvme_cdw12_raw & SPDK_DIF_FLAGS_NVME_PRACT) {
5941 0 : SPDK_ERRLOG("Separate metadata with NVMe PRACT is not supported.\n");
5942 0 : return -ENOTSUP;
5943 : }
5944 4 : }
5945 :
5946 4 : if (nvme_cdw12_raw & SPDK_DIF_FLAGS_NVME_PRACT) {
5947 0 : dif_check_flags |= SPDK_DIF_FLAGS_NVME_PRACT;
5948 0 : }
5949 4 : }
5950 :
5951 10 : dif_check_flags |= bdev->dif_check_flags &
5952 5 : ~(bdev_get_ext_io_opt(opts, dif_check_flags_exclude_mask, 0));
5953 :
5954 10 : return bdev_readv_blocks_with_md(desc, ch, iov, iovcnt, md, offset_blocks,
5955 5 : num_blocks, domain, domain_ctx, seq, dif_check_flags, cb, cb_arg);
5956 8 : }
5957 :
5958 : static int
5959 36 : bdev_write_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5960 : void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
5961 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5962 : {
5963 36 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5964 : struct spdk_bdev_io *bdev_io;
5965 36 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5966 :
5967 36 : if (!desc->write) {
5968 0 : return -EBADF;
5969 : }
5970 :
5971 36 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
5972 0 : return -EINVAL;
5973 : }
5974 :
5975 36 : bdev_io = bdev_channel_get_io(channel);
5976 36 : if (!bdev_io) {
5977 0 : return -ENOMEM;
5978 : }
5979 :
5980 36 : bdev_io->internal.ch = channel;
5981 36 : bdev_io->internal.desc = desc;
5982 36 : bdev_io->type = SPDK_BDEV_IO_TYPE_WRITE;
5983 36 : bdev_io->u.bdev.iovs = &bdev_io->iov;
5984 36 : bdev_io->u.bdev.iovs[0].iov_base = buf;
5985 36 : bdev_io->u.bdev.iovs[0].iov_len = num_blocks * bdev_desc_get_block_size(desc);
5986 36 : bdev_io->u.bdev.iovcnt = 1;
5987 36 : bdev_io->u.bdev.md_buf = md_buf;
5988 36 : bdev_io->u.bdev.num_blocks = num_blocks;
5989 36 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5990 36 : bdev_io->u.bdev.memory_domain = NULL;
5991 36 : bdev_io->u.bdev.memory_domain_ctx = NULL;
5992 36 : bdev_io->u.bdev.accel_sequence = NULL;
5993 36 : bdev_io->u.bdev.dif_check_flags = bdev->dif_check_flags;
5994 36 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5995 :
5996 36 : bdev_io_submit(bdev_io);
5997 36 : return 0;
5998 36 : }
5999 :
6000 : int
6001 3 : spdk_bdev_write(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6002 : void *buf, uint64_t offset, uint64_t nbytes,
6003 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6004 : {
6005 : uint64_t offset_blocks, num_blocks;
6006 :
6007 3 : if (bdev_bytes_to_blocks(desc, offset, &offset_blocks, nbytes, &num_blocks) != 0) {
6008 0 : return -EINVAL;
6009 : }
6010 :
6011 3 : return spdk_bdev_write_blocks(desc, ch, buf, offset_blocks, num_blocks, cb, cb_arg);
6012 3 : }
6013 :
6014 : int
6015 27 : spdk_bdev_write_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6016 : void *buf, uint64_t offset_blocks, uint64_t num_blocks,
6017 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6018 : {
6019 54 : return bdev_write_blocks_with_md(desc, ch, buf, NULL, offset_blocks, num_blocks,
6020 27 : cb, cb_arg);
6021 : }
6022 :
6023 : int
6024 3 : spdk_bdev_write_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6025 : void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
6026 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6027 : {
6028 6 : struct iovec iov = {
6029 3 : .iov_base = buf,
6030 : };
6031 :
6032 3 : if (md_buf && !spdk_bdev_is_md_separate(spdk_bdev_desc_get_bdev(desc))) {
6033 0 : return -EINVAL;
6034 : }
6035 :
6036 3 : if (md_buf && !_is_buf_allocated(&iov)) {
6037 0 : return -EINVAL;
6038 : }
6039 :
6040 6 : return bdev_write_blocks_with_md(desc, ch, buf, md_buf, offset_blocks, num_blocks,
6041 3 : cb, cb_arg);
6042 3 : }
6043 :
6044 : static int
6045 70 : bdev_writev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6046 : struct iovec *iov, int iovcnt, void *md_buf,
6047 : uint64_t offset_blocks, uint64_t num_blocks,
6048 : struct spdk_memory_domain *domain, void *domain_ctx,
6049 : struct spdk_accel_sequence *seq, uint32_t dif_check_flags,
6050 : uint32_t nvme_cdw12_raw, uint32_t nvme_cdw13_raw,
6051 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6052 : {
6053 70 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6054 : struct spdk_bdev_io *bdev_io;
6055 70 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6056 :
6057 70 : if (spdk_unlikely(!desc->write)) {
6058 0 : return -EBADF;
6059 : }
6060 :
6061 70 : if (spdk_unlikely(!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks))) {
6062 0 : return -EINVAL;
6063 : }
6064 :
6065 70 : bdev_io = bdev_channel_get_io(channel);
6066 70 : if (spdk_unlikely(!bdev_io)) {
6067 2 : return -ENOMEM;
6068 : }
6069 :
6070 68 : bdev_io->internal.ch = channel;
6071 68 : bdev_io->internal.desc = desc;
6072 68 : bdev_io->type = SPDK_BDEV_IO_TYPE_WRITE;
6073 68 : bdev_io->u.bdev.iovs = iov;
6074 68 : bdev_io->u.bdev.iovcnt = iovcnt;
6075 68 : bdev_io->u.bdev.md_buf = md_buf;
6076 68 : bdev_io->u.bdev.num_blocks = num_blocks;
6077 68 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6078 68 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6079 68 : if (seq != NULL) {
6080 0 : bdev_io->internal.f.has_accel_sequence = true;
6081 0 : bdev_io->internal.accel_sequence = seq;
6082 0 : }
6083 :
6084 68 : if (domain != NULL) {
6085 2 : bdev_io->internal.f.has_memory_domain = true;
6086 2 : bdev_io->internal.memory_domain = domain;
6087 2 : bdev_io->internal.memory_domain_ctx = domain_ctx;
6088 2 : }
6089 :
6090 68 : bdev_io->u.bdev.memory_domain = domain;
6091 68 : bdev_io->u.bdev.memory_domain_ctx = domain_ctx;
6092 68 : bdev_io->u.bdev.accel_sequence = seq;
6093 68 : bdev_io->u.bdev.dif_check_flags = dif_check_flags;
6094 68 : bdev_io->u.bdev.nvme_cdw12.raw = nvme_cdw12_raw;
6095 68 : bdev_io->u.bdev.nvme_cdw13.raw = nvme_cdw13_raw;
6096 :
6097 68 : _bdev_io_submit_ext(desc, bdev_io);
6098 :
6099 68 : return 0;
6100 70 : }
6101 :
6102 : int
6103 3 : spdk_bdev_writev(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6104 : struct iovec *iov, int iovcnt,
6105 : uint64_t offset, uint64_t len,
6106 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6107 : {
6108 : uint64_t offset_blocks, num_blocks;
6109 :
6110 3 : if (bdev_bytes_to_blocks(desc, offset, &offset_blocks, len, &num_blocks) != 0) {
6111 0 : return -EINVAL;
6112 : }
6113 :
6114 3 : return spdk_bdev_writev_blocks(desc, ch, iov, iovcnt, offset_blocks, num_blocks, cb, cb_arg);
6115 3 : }
6116 :
6117 : int
6118 14 : spdk_bdev_writev_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6119 : struct iovec *iov, int iovcnt,
6120 : uint64_t offset_blocks, uint64_t num_blocks,
6121 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6122 : {
6123 14 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6124 :
6125 28 : return bdev_writev_blocks_with_md(desc, ch, iov, iovcnt, NULL, offset_blocks,
6126 14 : num_blocks, NULL, NULL, NULL, bdev->dif_check_flags, 0, 0,
6127 14 : cb, cb_arg);
6128 : }
6129 :
6130 : int
6131 1 : spdk_bdev_writev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6132 : struct iovec *iov, int iovcnt, void *md_buf,
6133 : uint64_t offset_blocks, uint64_t num_blocks,
6134 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6135 : {
6136 1 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6137 :
6138 1 : if (md_buf && !spdk_bdev_is_md_separate(bdev)) {
6139 0 : return -EINVAL;
6140 : }
6141 :
6142 1 : if (md_buf && !_is_buf_allocated(iov)) {
6143 0 : return -EINVAL;
6144 : }
6145 :
6146 2 : return bdev_writev_blocks_with_md(desc, ch, iov, iovcnt, md_buf, offset_blocks,
6147 1 : num_blocks, NULL, NULL, NULL, bdev->dif_check_flags, 0, 0,
6148 1 : cb, cb_arg);
6149 1 : }
6150 :
6151 : int
6152 8 : spdk_bdev_writev_blocks_ext(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6153 : struct iovec *iov, int iovcnt,
6154 : uint64_t offset_blocks, uint64_t num_blocks,
6155 : spdk_bdev_io_completion_cb cb, void *cb_arg,
6156 : struct spdk_bdev_ext_io_opts *opts)
6157 : {
6158 8 : struct spdk_memory_domain *domain = NULL;
6159 8 : struct spdk_accel_sequence *seq = NULL;
6160 8 : void *domain_ctx = NULL, *md = NULL;
6161 8 : uint32_t dif_check_flags = 0;
6162 8 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6163 8 : uint32_t nvme_cdw12_raw = 0;
6164 8 : uint32_t nvme_cdw13_raw = 0;
6165 :
6166 8 : if (opts) {
6167 7 : if (spdk_unlikely(!_bdev_io_check_opts(opts, iov))) {
6168 3 : return -EINVAL;
6169 : }
6170 4 : md = opts->metadata;
6171 4 : domain = bdev_get_ext_io_opt(opts, memory_domain, NULL);
6172 4 : domain_ctx = bdev_get_ext_io_opt(opts, memory_domain_ctx, NULL);
6173 4 : seq = bdev_get_ext_io_opt(opts, accel_sequence, NULL);
6174 4 : nvme_cdw12_raw = bdev_get_ext_io_opt(opts, nvme_cdw12.raw, 0);
6175 4 : nvme_cdw13_raw = bdev_get_ext_io_opt(opts, nvme_cdw13.raw, 0);
6176 4 : if (md) {
6177 4 : if (spdk_unlikely(!spdk_bdev_is_md_separate(bdev))) {
6178 0 : return -EINVAL;
6179 : }
6180 :
6181 4 : if (spdk_unlikely(!_is_buf_allocated(iov))) {
6182 0 : return -EINVAL;
6183 : }
6184 :
6185 4 : if (spdk_unlikely(seq != NULL)) {
6186 0 : return -EINVAL;
6187 : }
6188 :
6189 4 : if (nvme_cdw12_raw & SPDK_DIF_FLAGS_NVME_PRACT) {
6190 0 : SPDK_ERRLOG("Separate metadata with NVMe PRACT is not supported.\n");
6191 0 : return -ENOTSUP;
6192 : }
6193 4 : }
6194 :
6195 4 : if (nvme_cdw12_raw & SPDK_DIF_FLAGS_NVME_PRACT) {
6196 0 : dif_check_flags |= SPDK_DIF_FLAGS_NVME_PRACT;
6197 0 : }
6198 4 : }
6199 :
6200 10 : dif_check_flags |= bdev->dif_check_flags &
6201 5 : ~(bdev_get_ext_io_opt(opts, dif_check_flags_exclude_mask, 0));
6202 :
6203 10 : return bdev_writev_blocks_with_md(desc, ch, iov, iovcnt, md, offset_blocks, num_blocks,
6204 5 : domain, domain_ctx, seq, dif_check_flags,
6205 5 : nvme_cdw12_raw, nvme_cdw13_raw, cb, cb_arg);
6206 8 : }
6207 :
6208 : static void
6209 11 : bdev_compare_do_read_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
6210 : {
6211 11 : struct spdk_bdev_io *parent_io = cb_arg;
6212 11 : struct spdk_bdev *bdev = parent_io->bdev;
6213 11 : uint8_t *read_buf = bdev_io->u.bdev.iovs[0].iov_base;
6214 11 : int i, rc = 0;
6215 :
6216 11 : if (!success) {
6217 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
6218 0 : parent_io->internal.cb(parent_io, false, parent_io->internal.caller_ctx);
6219 0 : spdk_bdev_free_io(bdev_io);
6220 0 : return;
6221 : }
6222 :
6223 17 : for (i = 0; i < parent_io->u.bdev.iovcnt; i++) {
6224 22 : rc = memcmp(read_buf,
6225 11 : parent_io->u.bdev.iovs[i].iov_base,
6226 11 : parent_io->u.bdev.iovs[i].iov_len);
6227 11 : if (rc) {
6228 5 : break;
6229 : }
6230 6 : read_buf += parent_io->u.bdev.iovs[i].iov_len;
6231 6 : }
6232 :
6233 11 : if (rc == 0 && parent_io->u.bdev.md_buf && spdk_bdev_is_md_separate(bdev)) {
6234 4 : rc = memcmp(bdev_io->u.bdev.md_buf,
6235 2 : parent_io->u.bdev.md_buf,
6236 2 : spdk_bdev_get_md_size(bdev));
6237 2 : }
6238 :
6239 11 : spdk_bdev_free_io(bdev_io);
6240 :
6241 11 : if (rc == 0) {
6242 5 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
6243 5 : parent_io->internal.cb(parent_io, true, parent_io->internal.caller_ctx);
6244 5 : } else {
6245 6 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_MISCOMPARE;
6246 6 : parent_io->internal.cb(parent_io, false, parent_io->internal.caller_ctx);
6247 : }
6248 11 : }
6249 :
6250 : static void
6251 11 : bdev_compare_do_read(void *_bdev_io)
6252 : {
6253 11 : struct spdk_bdev_io *bdev_io = _bdev_io;
6254 : int rc;
6255 :
6256 22 : rc = spdk_bdev_read_blocks(bdev_io->internal.desc,
6257 11 : spdk_io_channel_from_ctx(bdev_io->internal.ch), NULL,
6258 11 : bdev_io->u.bdev.offset_blocks, bdev_io->u.bdev.num_blocks,
6259 11 : bdev_compare_do_read_done, bdev_io);
6260 :
6261 11 : if (rc == -ENOMEM) {
6262 0 : bdev_queue_io_wait_with_cb(bdev_io, bdev_compare_do_read);
6263 11 : } else if (rc != 0) {
6264 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
6265 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
6266 0 : }
6267 11 : }
6268 :
6269 : static int
6270 16 : bdev_comparev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6271 : struct iovec *iov, int iovcnt, void *md_buf,
6272 : uint64_t offset_blocks, uint64_t num_blocks,
6273 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6274 : {
6275 16 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6276 : struct spdk_bdev_io *bdev_io;
6277 16 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6278 :
6279 16 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6280 0 : return -EINVAL;
6281 : }
6282 :
6283 16 : bdev_io = bdev_channel_get_io(channel);
6284 16 : if (!bdev_io) {
6285 0 : return -ENOMEM;
6286 : }
6287 :
6288 16 : bdev_io->internal.ch = channel;
6289 16 : bdev_io->internal.desc = desc;
6290 16 : bdev_io->type = SPDK_BDEV_IO_TYPE_COMPARE;
6291 16 : bdev_io->u.bdev.iovs = iov;
6292 16 : bdev_io->u.bdev.iovcnt = iovcnt;
6293 16 : bdev_io->u.bdev.md_buf = md_buf;
6294 16 : bdev_io->u.bdev.num_blocks = num_blocks;
6295 16 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6296 16 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6297 16 : bdev_io->u.bdev.memory_domain = NULL;
6298 16 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6299 16 : bdev_io->u.bdev.accel_sequence = NULL;
6300 :
6301 16 : if (bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COMPARE)) {
6302 7 : bdev_io_submit(bdev_io);
6303 7 : return 0;
6304 : }
6305 :
6306 9 : bdev_compare_do_read(bdev_io);
6307 :
6308 9 : return 0;
6309 16 : }
6310 :
6311 : int
6312 10 : spdk_bdev_comparev_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6313 : struct iovec *iov, int iovcnt,
6314 : uint64_t offset_blocks, uint64_t num_blocks,
6315 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6316 : {
6317 20 : return bdev_comparev_blocks_with_md(desc, ch, iov, iovcnt, NULL, offset_blocks,
6318 10 : num_blocks, cb, cb_arg);
6319 : }
6320 :
6321 : int
6322 6 : spdk_bdev_comparev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6323 : struct iovec *iov, int iovcnt, void *md_buf,
6324 : uint64_t offset_blocks, uint64_t num_blocks,
6325 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6326 : {
6327 6 : if (md_buf && !spdk_bdev_is_md_separate(spdk_bdev_desc_get_bdev(desc))) {
6328 0 : return -EINVAL;
6329 : }
6330 :
6331 6 : if (md_buf && !_is_buf_allocated(iov)) {
6332 0 : return -EINVAL;
6333 : }
6334 :
6335 12 : return bdev_comparev_blocks_with_md(desc, ch, iov, iovcnt, md_buf, offset_blocks,
6336 6 : num_blocks, cb, cb_arg);
6337 6 : }
6338 :
6339 : static int
6340 4 : bdev_compare_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6341 : void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
6342 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6343 : {
6344 4 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6345 : struct spdk_bdev_io *bdev_io;
6346 4 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6347 :
6348 4 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6349 0 : return -EINVAL;
6350 : }
6351 :
6352 4 : bdev_io = bdev_channel_get_io(channel);
6353 4 : if (!bdev_io) {
6354 0 : return -ENOMEM;
6355 : }
6356 :
6357 4 : bdev_io->internal.ch = channel;
6358 4 : bdev_io->internal.desc = desc;
6359 4 : bdev_io->type = SPDK_BDEV_IO_TYPE_COMPARE;
6360 4 : bdev_io->u.bdev.iovs = &bdev_io->iov;
6361 4 : bdev_io->u.bdev.iovs[0].iov_base = buf;
6362 4 : bdev_io->u.bdev.iovs[0].iov_len = num_blocks * bdev_desc_get_block_size(desc);
6363 4 : bdev_io->u.bdev.iovcnt = 1;
6364 4 : bdev_io->u.bdev.md_buf = md_buf;
6365 4 : bdev_io->u.bdev.num_blocks = num_blocks;
6366 4 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6367 4 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6368 4 : bdev_io->u.bdev.memory_domain = NULL;
6369 4 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6370 4 : bdev_io->u.bdev.accel_sequence = NULL;
6371 :
6372 4 : if (bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COMPARE)) {
6373 2 : bdev_io_submit(bdev_io);
6374 2 : return 0;
6375 : }
6376 :
6377 2 : bdev_compare_do_read(bdev_io);
6378 :
6379 2 : return 0;
6380 4 : }
6381 :
6382 : int
6383 4 : spdk_bdev_compare_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6384 : void *buf, uint64_t offset_blocks, uint64_t num_blocks,
6385 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6386 : {
6387 8 : return bdev_compare_blocks_with_md(desc, ch, buf, NULL, offset_blocks, num_blocks,
6388 4 : cb, cb_arg);
6389 : }
6390 :
6391 : int
6392 0 : spdk_bdev_compare_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6393 : void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
6394 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6395 : {
6396 0 : struct iovec iov = {
6397 0 : .iov_base = buf,
6398 : };
6399 :
6400 0 : if (md_buf && !spdk_bdev_is_md_separate(spdk_bdev_desc_get_bdev(desc))) {
6401 0 : return -EINVAL;
6402 : }
6403 :
6404 0 : if (md_buf && !_is_buf_allocated(&iov)) {
6405 0 : return -EINVAL;
6406 : }
6407 :
6408 0 : return bdev_compare_blocks_with_md(desc, ch, buf, md_buf, offset_blocks, num_blocks,
6409 0 : cb, cb_arg);
6410 0 : }
6411 :
6412 : static void
6413 2 : bdev_comparev_and_writev_blocks_unlocked(struct lba_range *range, void *ctx, int unlock_status)
6414 : {
6415 2 : struct spdk_bdev_io *bdev_io = ctx;
6416 :
6417 2 : if (unlock_status) {
6418 0 : SPDK_ERRLOG("LBA range unlock failed\n");
6419 0 : }
6420 :
6421 4 : bdev_io->internal.cb(bdev_io, bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS ? true :
6422 2 : false, bdev_io->internal.caller_ctx);
6423 2 : }
6424 :
6425 : static void
6426 2 : bdev_comparev_and_writev_blocks_unlock(struct spdk_bdev_io *bdev_io, int status)
6427 : {
6428 2 : bdev_io->internal.status = status;
6429 :
6430 4 : bdev_unlock_lba_range(bdev_io->internal.desc, spdk_io_channel_from_ctx(bdev_io->internal.ch),
6431 2 : bdev_io->u.bdev.offset_blocks, bdev_io->u.bdev.num_blocks,
6432 2 : bdev_comparev_and_writev_blocks_unlocked, bdev_io);
6433 2 : }
6434 :
6435 : static void
6436 1 : bdev_compare_and_write_do_write_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
6437 : {
6438 1 : struct spdk_bdev_io *parent_io = cb_arg;
6439 :
6440 1 : if (!success) {
6441 0 : SPDK_ERRLOG("Compare and write operation failed\n");
6442 0 : }
6443 :
6444 1 : spdk_bdev_free_io(bdev_io);
6445 :
6446 2 : bdev_comparev_and_writev_blocks_unlock(parent_io,
6447 1 : success ? SPDK_BDEV_IO_STATUS_SUCCESS : SPDK_BDEV_IO_STATUS_FAILED);
6448 1 : }
6449 :
6450 : static void
6451 1 : bdev_compare_and_write_do_write(void *_bdev_io)
6452 : {
6453 1 : struct spdk_bdev_io *bdev_io = _bdev_io;
6454 : int rc;
6455 :
6456 2 : rc = spdk_bdev_writev_blocks(bdev_io->internal.desc,
6457 1 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
6458 1 : bdev_io->u.bdev.fused_iovs, bdev_io->u.bdev.fused_iovcnt,
6459 1 : bdev_io->u.bdev.offset_blocks, bdev_io->u.bdev.num_blocks,
6460 1 : bdev_compare_and_write_do_write_done, bdev_io);
6461 :
6462 :
6463 1 : if (rc == -ENOMEM) {
6464 0 : bdev_queue_io_wait_with_cb(bdev_io, bdev_compare_and_write_do_write);
6465 1 : } else if (rc != 0) {
6466 0 : bdev_comparev_and_writev_blocks_unlock(bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
6467 0 : }
6468 1 : }
6469 :
6470 : static void
6471 2 : bdev_compare_and_write_do_compare_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
6472 : {
6473 2 : struct spdk_bdev_io *parent_io = cb_arg;
6474 :
6475 2 : spdk_bdev_free_io(bdev_io);
6476 :
6477 2 : if (!success) {
6478 1 : bdev_comparev_and_writev_blocks_unlock(parent_io, SPDK_BDEV_IO_STATUS_MISCOMPARE);
6479 1 : return;
6480 : }
6481 :
6482 1 : bdev_compare_and_write_do_write(parent_io);
6483 2 : }
6484 :
6485 : static void
6486 2 : bdev_compare_and_write_do_compare(void *_bdev_io)
6487 : {
6488 2 : struct spdk_bdev_io *bdev_io = _bdev_io;
6489 : int rc;
6490 :
6491 4 : rc = spdk_bdev_comparev_blocks(bdev_io->internal.desc,
6492 2 : spdk_io_channel_from_ctx(bdev_io->internal.ch), bdev_io->u.bdev.iovs,
6493 2 : bdev_io->u.bdev.iovcnt, bdev_io->u.bdev.offset_blocks, bdev_io->u.bdev.num_blocks,
6494 2 : bdev_compare_and_write_do_compare_done, bdev_io);
6495 :
6496 2 : if (rc == -ENOMEM) {
6497 0 : bdev_queue_io_wait_with_cb(bdev_io, bdev_compare_and_write_do_compare);
6498 2 : } else if (rc != 0) {
6499 0 : bdev_comparev_and_writev_blocks_unlock(bdev_io, SPDK_BDEV_IO_STATUS_FIRST_FUSED_FAILED);
6500 0 : }
6501 2 : }
6502 :
6503 : static void
6504 2 : bdev_comparev_and_writev_blocks_locked(struct lba_range *range, void *ctx, int status)
6505 : {
6506 2 : struct spdk_bdev_io *bdev_io = ctx;
6507 :
6508 2 : if (status) {
6509 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FIRST_FUSED_FAILED;
6510 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
6511 0 : return;
6512 : }
6513 :
6514 2 : bdev_compare_and_write_do_compare(bdev_io);
6515 2 : }
6516 :
6517 : int
6518 2 : spdk_bdev_comparev_and_writev_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6519 : struct iovec *compare_iov, int compare_iovcnt,
6520 : struct iovec *write_iov, int write_iovcnt,
6521 : uint64_t offset_blocks, uint64_t num_blocks,
6522 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6523 : {
6524 2 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6525 : struct spdk_bdev_io *bdev_io;
6526 2 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6527 :
6528 2 : if (!desc->write) {
6529 0 : return -EBADF;
6530 : }
6531 :
6532 2 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6533 0 : return -EINVAL;
6534 : }
6535 :
6536 2 : if (num_blocks > bdev->acwu) {
6537 0 : return -EINVAL;
6538 : }
6539 :
6540 2 : bdev_io = bdev_channel_get_io(channel);
6541 2 : if (!bdev_io) {
6542 0 : return -ENOMEM;
6543 : }
6544 :
6545 2 : bdev_io->internal.ch = channel;
6546 2 : bdev_io->internal.desc = desc;
6547 2 : bdev_io->type = SPDK_BDEV_IO_TYPE_COMPARE_AND_WRITE;
6548 2 : bdev_io->u.bdev.iovs = compare_iov;
6549 2 : bdev_io->u.bdev.iovcnt = compare_iovcnt;
6550 2 : bdev_io->u.bdev.fused_iovs = write_iov;
6551 2 : bdev_io->u.bdev.fused_iovcnt = write_iovcnt;
6552 2 : bdev_io->u.bdev.md_buf = NULL;
6553 2 : bdev_io->u.bdev.num_blocks = num_blocks;
6554 2 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6555 2 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6556 2 : bdev_io->u.bdev.memory_domain = NULL;
6557 2 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6558 2 : bdev_io->u.bdev.accel_sequence = NULL;
6559 :
6560 2 : if (bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COMPARE_AND_WRITE)) {
6561 0 : bdev_io_submit(bdev_io);
6562 0 : return 0;
6563 : }
6564 :
6565 4 : return bdev_lock_lba_range(desc, ch, offset_blocks, num_blocks,
6566 2 : bdev_comparev_and_writev_blocks_locked, bdev_io);
6567 2 : }
6568 :
6569 : int
6570 2 : spdk_bdev_zcopy_start(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6571 : struct iovec *iov, int iovcnt,
6572 : uint64_t offset_blocks, uint64_t num_blocks,
6573 : bool populate,
6574 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6575 : {
6576 2 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6577 : struct spdk_bdev_io *bdev_io;
6578 2 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6579 :
6580 2 : if (!desc->write) {
6581 0 : return -EBADF;
6582 : }
6583 :
6584 2 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6585 0 : return -EINVAL;
6586 : }
6587 :
6588 2 : if (!spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_ZCOPY)) {
6589 0 : return -ENOTSUP;
6590 : }
6591 :
6592 2 : bdev_io = bdev_channel_get_io(channel);
6593 2 : if (!bdev_io) {
6594 0 : return -ENOMEM;
6595 : }
6596 :
6597 2 : bdev_io->internal.ch = channel;
6598 2 : bdev_io->internal.desc = desc;
6599 2 : bdev_io->type = SPDK_BDEV_IO_TYPE_ZCOPY;
6600 2 : bdev_io->u.bdev.num_blocks = num_blocks;
6601 2 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6602 2 : bdev_io->u.bdev.iovs = iov;
6603 2 : bdev_io->u.bdev.iovcnt = iovcnt;
6604 2 : bdev_io->u.bdev.md_buf = NULL;
6605 2 : bdev_io->u.bdev.zcopy.populate = populate ? 1 : 0;
6606 2 : bdev_io->u.bdev.zcopy.commit = 0;
6607 2 : bdev_io->u.bdev.zcopy.start = 1;
6608 2 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6609 2 : bdev_io->u.bdev.memory_domain = NULL;
6610 2 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6611 2 : bdev_io->u.bdev.accel_sequence = NULL;
6612 :
6613 2 : bdev_io_submit(bdev_io);
6614 :
6615 2 : return 0;
6616 2 : }
6617 :
6618 : int
6619 2 : spdk_bdev_zcopy_end(struct spdk_bdev_io *bdev_io, bool commit,
6620 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6621 : {
6622 2 : if (bdev_io->type != SPDK_BDEV_IO_TYPE_ZCOPY) {
6623 0 : return -EINVAL;
6624 : }
6625 :
6626 2 : bdev_io->u.bdev.zcopy.commit = commit ? 1 : 0;
6627 2 : bdev_io->u.bdev.zcopy.start = 0;
6628 2 : bdev_io->internal.caller_ctx = cb_arg;
6629 2 : bdev_io->internal.cb = cb;
6630 2 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_PENDING;
6631 :
6632 2 : bdev_io_submit(bdev_io);
6633 :
6634 2 : return 0;
6635 2 : }
6636 :
6637 : int
6638 0 : spdk_bdev_write_zeroes(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6639 : uint64_t offset, uint64_t len,
6640 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6641 : {
6642 : uint64_t offset_blocks, num_blocks;
6643 :
6644 0 : if (bdev_bytes_to_blocks(desc, offset, &offset_blocks, len, &num_blocks) != 0) {
6645 0 : return -EINVAL;
6646 : }
6647 :
6648 0 : return spdk_bdev_write_zeroes_blocks(desc, ch, offset_blocks, num_blocks, cb, cb_arg);
6649 0 : }
6650 :
6651 : int
6652 33 : spdk_bdev_write_zeroes_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6653 : uint64_t offset_blocks, uint64_t num_blocks,
6654 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6655 : {
6656 33 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6657 : struct spdk_bdev_io *bdev_io;
6658 33 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6659 :
6660 33 : if (!desc->write) {
6661 0 : return -EBADF;
6662 : }
6663 :
6664 33 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6665 0 : return -EINVAL;
6666 : }
6667 :
6668 33 : if (!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE_ZEROES) &&
6669 10 : !bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE)) {
6670 1 : return -ENOTSUP;
6671 : }
6672 :
6673 32 : bdev_io = bdev_channel_get_io(channel);
6674 :
6675 32 : if (!bdev_io) {
6676 0 : return -ENOMEM;
6677 : }
6678 :
6679 32 : bdev_io->type = SPDK_BDEV_IO_TYPE_WRITE_ZEROES;
6680 32 : bdev_io->internal.ch = channel;
6681 32 : bdev_io->internal.desc = desc;
6682 32 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6683 32 : bdev_io->u.bdev.num_blocks = num_blocks;
6684 32 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6685 32 : bdev_io->u.bdev.memory_domain = NULL;
6686 32 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6687 32 : bdev_io->u.bdev.accel_sequence = NULL;
6688 :
6689 : /* If the write_zeroes size is large and should be split, use the generic split
6690 : * logic regardless of whether SPDK_BDEV_IO_TYPE_WRITE_ZEREOS is supported or not.
6691 : *
6692 : * Then, send the write_zeroes request if SPDK_BDEV_IO_TYPE_WRITE_ZEROES is supported
6693 : * or emulate it using regular write request otherwise.
6694 : */
6695 32 : if (bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE_ZEROES) ||
6696 9 : bdev_io->internal.f.split) {
6697 26 : bdev_io_submit(bdev_io);
6698 26 : return 0;
6699 : }
6700 :
6701 6 : assert(_bdev_get_block_size_with_md(bdev) <= ZERO_BUFFER_SIZE);
6702 :
6703 6 : return bdev_write_zero_buffer(bdev_io);
6704 33 : }
6705 :
6706 : int
6707 0 : spdk_bdev_unmap(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6708 : uint64_t offset, uint64_t nbytes,
6709 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6710 : {
6711 : uint64_t offset_blocks, num_blocks;
6712 :
6713 0 : if (bdev_bytes_to_blocks(desc, offset, &offset_blocks, nbytes, &num_blocks) != 0) {
6714 0 : return -EINVAL;
6715 : }
6716 :
6717 0 : return spdk_bdev_unmap_blocks(desc, ch, offset_blocks, num_blocks, cb, cb_arg);
6718 0 : }
6719 :
6720 : static void
6721 0 : bdev_io_complete_cb(void *ctx)
6722 : {
6723 0 : struct spdk_bdev_io *bdev_io = ctx;
6724 :
6725 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
6726 0 : bdev_io->internal.cb(bdev_io, true, bdev_io->internal.caller_ctx);
6727 0 : }
6728 :
6729 : int
6730 22 : spdk_bdev_unmap_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6731 : uint64_t offset_blocks, uint64_t num_blocks,
6732 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6733 : {
6734 22 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6735 : struct spdk_bdev_io *bdev_io;
6736 22 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6737 :
6738 22 : if (!desc->write) {
6739 0 : return -EBADF;
6740 : }
6741 :
6742 22 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6743 0 : return -EINVAL;
6744 : }
6745 :
6746 22 : bdev_io = bdev_channel_get_io(channel);
6747 22 : if (!bdev_io) {
6748 0 : return -ENOMEM;
6749 : }
6750 :
6751 22 : bdev_io->internal.ch = channel;
6752 22 : bdev_io->internal.desc = desc;
6753 22 : bdev_io->type = SPDK_BDEV_IO_TYPE_UNMAP;
6754 :
6755 22 : bdev_io->u.bdev.iovs = &bdev_io->iov;
6756 22 : bdev_io->u.bdev.iovs[0].iov_base = NULL;
6757 22 : bdev_io->u.bdev.iovs[0].iov_len = 0;
6758 22 : bdev_io->u.bdev.iovcnt = 1;
6759 :
6760 22 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6761 22 : bdev_io->u.bdev.num_blocks = num_blocks;
6762 22 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6763 22 : bdev_io->u.bdev.memory_domain = NULL;
6764 22 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6765 22 : bdev_io->u.bdev.accel_sequence = NULL;
6766 :
6767 22 : if (num_blocks == 0) {
6768 0 : spdk_thread_send_msg(spdk_get_thread(), bdev_io_complete_cb, bdev_io);
6769 0 : return 0;
6770 : }
6771 :
6772 22 : bdev_io_submit(bdev_io);
6773 22 : return 0;
6774 22 : }
6775 :
6776 : int
6777 0 : spdk_bdev_flush(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6778 : uint64_t offset, uint64_t length,
6779 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6780 : {
6781 : uint64_t offset_blocks, num_blocks;
6782 :
6783 0 : if (bdev_bytes_to_blocks(desc, offset, &offset_blocks, length, &num_blocks) != 0) {
6784 0 : return -EINVAL;
6785 : }
6786 :
6787 0 : return spdk_bdev_flush_blocks(desc, ch, offset_blocks, num_blocks, cb, cb_arg);
6788 0 : }
6789 :
6790 : int
6791 2 : spdk_bdev_flush_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6792 : uint64_t offset_blocks, uint64_t num_blocks,
6793 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6794 : {
6795 2 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6796 : struct spdk_bdev_io *bdev_io;
6797 2 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6798 :
6799 2 : if (!desc->write) {
6800 0 : return -EBADF;
6801 : }
6802 :
6803 2 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6804 0 : return -EINVAL;
6805 : }
6806 :
6807 2 : bdev_io = bdev_channel_get_io(channel);
6808 2 : if (!bdev_io) {
6809 0 : return -ENOMEM;
6810 : }
6811 :
6812 2 : bdev_io->internal.ch = channel;
6813 2 : bdev_io->internal.desc = desc;
6814 2 : bdev_io->type = SPDK_BDEV_IO_TYPE_FLUSH;
6815 2 : bdev_io->u.bdev.iovs = NULL;
6816 2 : bdev_io->u.bdev.iovcnt = 0;
6817 2 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6818 2 : bdev_io->u.bdev.num_blocks = num_blocks;
6819 2 : bdev_io->u.bdev.memory_domain = NULL;
6820 2 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6821 2 : bdev_io->u.bdev.accel_sequence = NULL;
6822 2 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6823 :
6824 2 : bdev_io_submit(bdev_io);
6825 2 : return 0;
6826 2 : }
6827 :
6828 : static int bdev_reset_poll_for_outstanding_io(void *ctx);
6829 :
6830 : static void
6831 13 : bdev_reset_check_outstanding_io_done(struct spdk_bdev *bdev, void *_ctx, int status)
6832 : {
6833 13 : struct spdk_bdev_channel *ch = _ctx;
6834 : struct spdk_bdev_io *bdev_io;
6835 :
6836 13 : bdev_io = TAILQ_FIRST(&ch->queued_resets);
6837 :
6838 13 : if (status == -EBUSY) {
6839 9 : if (spdk_get_ticks() < bdev_io->u.reset.wait_poller.stop_time_tsc) {
6840 8 : bdev_io->u.reset.wait_poller.poller = SPDK_POLLER_REGISTER(bdev_reset_poll_for_outstanding_io,
6841 : ch, BDEV_RESET_CHECK_OUTSTANDING_IO_PERIOD);
6842 8 : } else {
6843 1 : TAILQ_REMOVE(&ch->queued_resets, bdev_io, internal.link);
6844 :
6845 1 : if (TAILQ_EMPTY(&ch->io_memory_domain) && TAILQ_EMPTY(&ch->io_accel_exec)) {
6846 : /* If outstanding IOs are still present and reset_io_drain_timeout
6847 : * seconds passed, start the reset. */
6848 1 : bdev_io_submit_reset(bdev_io);
6849 1 : } else {
6850 : /* We still have in progress memory domain pull/push or we're
6851 : * executing accel sequence. Since we cannot abort either of those
6852 : * operations, fail the reset request. */
6853 0 : spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
6854 : }
6855 : }
6856 9 : } else {
6857 4 : TAILQ_REMOVE(&ch->queued_resets, bdev_io, internal.link);
6858 4 : SPDK_DEBUGLOG(bdev,
6859 : "Skipping reset for underlying device of bdev: %s - no outstanding I/O.\n",
6860 : ch->bdev->name);
6861 : /* Mark the completion status as a SUCCESS and complete the reset. */
6862 4 : spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_SUCCESS);
6863 : }
6864 13 : }
6865 :
6866 : static void
6867 13 : bdev_reset_check_outstanding_io(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
6868 : struct spdk_io_channel *io_ch, void *_ctx)
6869 : {
6870 13 : struct spdk_bdev_channel *cur_ch = __io_ch_to_bdev_ch(io_ch);
6871 13 : int status = 0;
6872 :
6873 17 : if (cur_ch->io_outstanding > 0 ||
6874 4 : !TAILQ_EMPTY(&cur_ch->io_memory_domain) ||
6875 4 : !TAILQ_EMPTY(&cur_ch->io_accel_exec)) {
6876 : /* If a channel has outstanding IO, set status to -EBUSY code. This will stop
6877 : * further iteration over the rest of the channels and pass non-zero status
6878 : * to the callback function. */
6879 9 : status = -EBUSY;
6880 9 : }
6881 13 : spdk_bdev_for_each_channel_continue(i, status);
6882 13 : }
6883 :
6884 : static int
6885 8 : bdev_reset_poll_for_outstanding_io(void *ctx)
6886 : {
6887 8 : struct spdk_bdev_channel *ch = ctx;
6888 : struct spdk_bdev_io *bdev_io;
6889 :
6890 8 : bdev_io = TAILQ_FIRST(&ch->queued_resets);
6891 :
6892 8 : spdk_poller_unregister(&bdev_io->u.reset.wait_poller.poller);
6893 8 : spdk_bdev_for_each_channel(ch->bdev, bdev_reset_check_outstanding_io, ch,
6894 : bdev_reset_check_outstanding_io_done);
6895 :
6896 8 : return SPDK_POLLER_BUSY;
6897 : }
6898 :
6899 : static void
6900 15 : bdev_reset_freeze_channel_done(struct spdk_bdev *bdev, void *_ctx, int status)
6901 : {
6902 15 : struct spdk_bdev_channel *ch = _ctx;
6903 : struct spdk_bdev_io *bdev_io;
6904 :
6905 15 : bdev_io = TAILQ_FIRST(&ch->queued_resets);
6906 :
6907 15 : if (bdev->reset_io_drain_timeout == 0) {
6908 10 : TAILQ_REMOVE(&ch->queued_resets, bdev_io, internal.link);
6909 :
6910 10 : bdev_io_submit_reset(bdev_io);
6911 10 : return;
6912 : }
6913 :
6914 10 : bdev_io->u.reset.wait_poller.stop_time_tsc = spdk_get_ticks() +
6915 5 : (ch->bdev->reset_io_drain_timeout * spdk_get_ticks_hz());
6916 :
6917 : /* In case bdev->reset_io_drain_timeout is not equal to zero,
6918 : * submit the reset to the underlying module only if outstanding I/O
6919 : * remain after reset_io_drain_timeout seconds have passed. */
6920 5 : spdk_bdev_for_each_channel(ch->bdev, bdev_reset_check_outstanding_io, ch,
6921 : bdev_reset_check_outstanding_io_done);
6922 15 : }
6923 :
6924 : static void
6925 18 : bdev_reset_freeze_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
6926 : struct spdk_io_channel *ch, void *_ctx)
6927 : {
6928 : struct spdk_bdev_channel *channel;
6929 : struct spdk_bdev_mgmt_channel *mgmt_channel;
6930 : struct spdk_bdev_shared_resource *shared_resource;
6931 : bdev_io_tailq_t tmp_queued;
6932 :
6933 18 : TAILQ_INIT(&tmp_queued);
6934 :
6935 18 : channel = __io_ch_to_bdev_ch(ch);
6936 18 : shared_resource = channel->shared_resource;
6937 18 : mgmt_channel = shared_resource->mgmt_ch;
6938 :
6939 18 : channel->flags |= BDEV_CH_RESET_IN_PROGRESS;
6940 :
6941 18 : if ((channel->flags & BDEV_CH_QOS_ENABLED) != 0) {
6942 2 : TAILQ_SWAP(&channel->qos_queued_io, &tmp_queued, spdk_bdev_io, internal.link);
6943 2 : }
6944 :
6945 18 : bdev_abort_all_queued_io(&shared_resource->nomem_io, channel);
6946 18 : bdev_abort_all_buf_io(mgmt_channel, channel);
6947 18 : bdev_abort_all_queued_io(&tmp_queued, channel);
6948 :
6949 18 : spdk_bdev_for_each_channel_continue(i, 0);
6950 18 : }
6951 :
6952 : static void
6953 15 : bdev_start_reset(void *ctx)
6954 : {
6955 15 : struct spdk_bdev_channel *ch = ctx;
6956 :
6957 15 : spdk_bdev_for_each_channel(ch->bdev, bdev_reset_freeze_channel, ch,
6958 : bdev_reset_freeze_channel_done);
6959 15 : }
6960 :
6961 : static void
6962 16 : bdev_channel_start_reset(struct spdk_bdev_channel *ch)
6963 : {
6964 16 : struct spdk_bdev *bdev = ch->bdev;
6965 :
6966 16 : assert(!TAILQ_EMPTY(&ch->queued_resets));
6967 :
6968 16 : spdk_spin_lock(&bdev->internal.spinlock);
6969 16 : if (bdev->internal.reset_in_progress == NULL) {
6970 15 : bdev->internal.reset_in_progress = TAILQ_FIRST(&ch->queued_resets);
6971 : /*
6972 : * Take a channel reference for the target bdev for the life of this
6973 : * reset. This guards against the channel getting destroyed while
6974 : * spdk_bdev_for_each_channel() calls related to this reset IO are in
6975 : * progress. We will release the reference when this reset is
6976 : * completed.
6977 : */
6978 15 : bdev->internal.reset_in_progress->u.reset.ch_ref = spdk_get_io_channel(__bdev_to_io_dev(bdev));
6979 15 : bdev_start_reset(ch);
6980 15 : }
6981 16 : spdk_spin_unlock(&bdev->internal.spinlock);
6982 16 : }
6983 :
6984 : int
6985 16 : spdk_bdev_reset(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6986 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6987 : {
6988 16 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6989 : struct spdk_bdev_io *bdev_io;
6990 16 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6991 :
6992 16 : bdev_io = bdev_channel_get_io(channel);
6993 16 : if (!bdev_io) {
6994 0 : return -ENOMEM;
6995 : }
6996 :
6997 16 : bdev_io->internal.ch = channel;
6998 16 : bdev_io->internal.desc = desc;
6999 16 : bdev_io->internal.submit_tsc = spdk_get_ticks();
7000 16 : bdev_io->type = SPDK_BDEV_IO_TYPE_RESET;
7001 16 : bdev_io->u.reset.ch_ref = NULL;
7002 16 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
7003 :
7004 16 : spdk_spin_lock(&bdev->internal.spinlock);
7005 16 : TAILQ_INSERT_TAIL(&channel->queued_resets, bdev_io, internal.link);
7006 16 : spdk_spin_unlock(&bdev->internal.spinlock);
7007 :
7008 16 : bdev_ch_add_to_io_submitted(bdev_io);
7009 :
7010 16 : bdev_channel_start_reset(channel);
7011 :
7012 16 : return 0;
7013 16 : }
7014 :
7015 : void
7016 0 : spdk_bdev_get_io_stat(struct spdk_bdev *bdev, struct spdk_io_channel *ch,
7017 : struct spdk_bdev_io_stat *stat, enum spdk_bdev_reset_stat_mode reset_mode)
7018 : {
7019 0 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
7020 :
7021 0 : bdev_get_io_stat(stat, channel->stat);
7022 0 : spdk_bdev_reset_io_stat(stat, reset_mode);
7023 0 : }
7024 :
7025 : static void
7026 5 : bdev_get_device_stat_done(struct spdk_bdev *bdev, void *_ctx, int status)
7027 : {
7028 5 : struct spdk_bdev_iostat_ctx *bdev_iostat_ctx = _ctx;
7029 :
7030 10 : bdev_iostat_ctx->cb(bdev, bdev_iostat_ctx->stat,
7031 5 : bdev_iostat_ctx->cb_arg, 0);
7032 5 : free(bdev_iostat_ctx);
7033 5 : }
7034 :
7035 : static void
7036 4 : bdev_get_each_channel_stat(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
7037 : struct spdk_io_channel *ch, void *_ctx)
7038 : {
7039 4 : struct spdk_bdev_iostat_ctx *bdev_iostat_ctx = _ctx;
7040 4 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
7041 :
7042 4 : spdk_bdev_add_io_stat(bdev_iostat_ctx->stat, channel->stat);
7043 4 : spdk_bdev_reset_io_stat(channel->stat, bdev_iostat_ctx->reset_mode);
7044 4 : spdk_bdev_for_each_channel_continue(i, 0);
7045 4 : }
7046 :
7047 : void
7048 5 : spdk_bdev_get_device_stat(struct spdk_bdev *bdev, struct spdk_bdev_io_stat *stat,
7049 : enum spdk_bdev_reset_stat_mode reset_mode, spdk_bdev_get_device_stat_cb cb, void *cb_arg)
7050 : {
7051 : struct spdk_bdev_iostat_ctx *bdev_iostat_ctx;
7052 :
7053 5 : assert(bdev != NULL);
7054 5 : assert(stat != NULL);
7055 5 : assert(cb != NULL);
7056 :
7057 5 : bdev_iostat_ctx = calloc(1, sizeof(struct spdk_bdev_iostat_ctx));
7058 5 : if (bdev_iostat_ctx == NULL) {
7059 0 : SPDK_ERRLOG("Unable to allocate memory for spdk_bdev_iostat_ctx\n");
7060 0 : cb(bdev, stat, cb_arg, -ENOMEM);
7061 0 : return;
7062 : }
7063 :
7064 5 : bdev_iostat_ctx->stat = stat;
7065 5 : bdev_iostat_ctx->cb = cb;
7066 5 : bdev_iostat_ctx->cb_arg = cb_arg;
7067 5 : bdev_iostat_ctx->reset_mode = reset_mode;
7068 :
7069 : /* Start with the statistics from previously deleted channels. */
7070 5 : spdk_spin_lock(&bdev->internal.spinlock);
7071 5 : bdev_get_io_stat(bdev_iostat_ctx->stat, bdev->internal.stat);
7072 5 : spdk_bdev_reset_io_stat(bdev->internal.stat, reset_mode);
7073 5 : spdk_spin_unlock(&bdev->internal.spinlock);
7074 :
7075 : /* Then iterate and add the statistics from each existing channel. */
7076 5 : spdk_bdev_for_each_channel(bdev, bdev_get_each_channel_stat, bdev_iostat_ctx,
7077 : bdev_get_device_stat_done);
7078 5 : }
7079 :
7080 : struct bdev_iostat_reset_ctx {
7081 : enum spdk_bdev_reset_stat_mode mode;
7082 : bdev_reset_device_stat_cb cb;
7083 : void *cb_arg;
7084 : };
7085 :
7086 : static void
7087 0 : bdev_reset_device_stat_done(struct spdk_bdev *bdev, void *_ctx, int status)
7088 : {
7089 0 : struct bdev_iostat_reset_ctx *ctx = _ctx;
7090 :
7091 0 : ctx->cb(bdev, ctx->cb_arg, 0);
7092 :
7093 0 : free(ctx);
7094 0 : }
7095 :
7096 : static void
7097 0 : bdev_reset_each_channel_stat(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
7098 : struct spdk_io_channel *ch, void *_ctx)
7099 : {
7100 0 : struct bdev_iostat_reset_ctx *ctx = _ctx;
7101 0 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
7102 :
7103 0 : spdk_bdev_reset_io_stat(channel->stat, ctx->mode);
7104 :
7105 0 : spdk_bdev_for_each_channel_continue(i, 0);
7106 0 : }
7107 :
7108 : void
7109 0 : bdev_reset_device_stat(struct spdk_bdev *bdev, enum spdk_bdev_reset_stat_mode mode,
7110 : bdev_reset_device_stat_cb cb, void *cb_arg)
7111 : {
7112 : struct bdev_iostat_reset_ctx *ctx;
7113 :
7114 0 : assert(bdev != NULL);
7115 0 : assert(cb != NULL);
7116 :
7117 0 : ctx = calloc(1, sizeof(*ctx));
7118 0 : if (ctx == NULL) {
7119 0 : SPDK_ERRLOG("Unable to allocate bdev_iostat_reset_ctx.\n");
7120 0 : cb(bdev, cb_arg, -ENOMEM);
7121 0 : return;
7122 : }
7123 :
7124 0 : ctx->mode = mode;
7125 0 : ctx->cb = cb;
7126 0 : ctx->cb_arg = cb_arg;
7127 :
7128 0 : spdk_spin_lock(&bdev->internal.spinlock);
7129 0 : spdk_bdev_reset_io_stat(bdev->internal.stat, mode);
7130 0 : spdk_spin_unlock(&bdev->internal.spinlock);
7131 :
7132 0 : spdk_bdev_for_each_channel(bdev,
7133 : bdev_reset_each_channel_stat,
7134 0 : ctx,
7135 : bdev_reset_device_stat_done);
7136 0 : }
7137 :
7138 : int
7139 1 : spdk_bdev_nvme_admin_passthru(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
7140 : const struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes,
7141 : spdk_bdev_io_completion_cb cb, void *cb_arg)
7142 : {
7143 1 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
7144 : struct spdk_bdev_io *bdev_io;
7145 1 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
7146 :
7147 1 : if (!desc->write) {
7148 0 : return -EBADF;
7149 : }
7150 :
7151 1 : if (spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_ADMIN))) {
7152 1 : return -ENOTSUP;
7153 : }
7154 :
7155 0 : bdev_io = bdev_channel_get_io(channel);
7156 0 : if (!bdev_io) {
7157 0 : return -ENOMEM;
7158 : }
7159 :
7160 0 : bdev_io->internal.ch = channel;
7161 0 : bdev_io->internal.desc = desc;
7162 0 : bdev_io->type = SPDK_BDEV_IO_TYPE_NVME_ADMIN;
7163 0 : bdev_io->u.nvme_passthru.cmd = *cmd;
7164 0 : bdev_io->u.nvme_passthru.buf = buf;
7165 0 : bdev_io->u.nvme_passthru.nbytes = nbytes;
7166 0 : bdev_io->u.nvme_passthru.md_buf = NULL;
7167 0 : bdev_io->u.nvme_passthru.md_len = 0;
7168 :
7169 0 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
7170 :
7171 0 : bdev_io_submit(bdev_io);
7172 0 : return 0;
7173 1 : }
7174 :
7175 : int
7176 1 : spdk_bdev_nvme_io_passthru(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
7177 : const struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes,
7178 : spdk_bdev_io_completion_cb cb, void *cb_arg)
7179 : {
7180 1 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
7181 : struct spdk_bdev_io *bdev_io;
7182 1 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
7183 :
7184 1 : if (!desc->write) {
7185 : /*
7186 : * Do not try to parse the NVMe command - we could maybe use bits in the opcode
7187 : * to easily determine if the command is a read or write, but for now just
7188 : * do not allow io_passthru with a read-only descriptor.
7189 : */
7190 0 : return -EBADF;
7191 : }
7192 :
7193 1 : if (spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_IO))) {
7194 1 : return -ENOTSUP;
7195 : }
7196 :
7197 0 : bdev_io = bdev_channel_get_io(channel);
7198 0 : if (!bdev_io) {
7199 0 : return -ENOMEM;
7200 : }
7201 :
7202 0 : bdev_io->internal.ch = channel;
7203 0 : bdev_io->internal.desc = desc;
7204 0 : bdev_io->type = SPDK_BDEV_IO_TYPE_NVME_IO;
7205 0 : bdev_io->u.nvme_passthru.cmd = *cmd;
7206 0 : bdev_io->u.nvme_passthru.buf = buf;
7207 0 : bdev_io->u.nvme_passthru.nbytes = nbytes;
7208 0 : bdev_io->u.nvme_passthru.md_buf = NULL;
7209 0 : bdev_io->u.nvme_passthru.md_len = 0;
7210 :
7211 0 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
7212 :
7213 0 : bdev_io_submit(bdev_io);
7214 0 : return 0;
7215 1 : }
7216 :
7217 : int
7218 1 : spdk_bdev_nvme_io_passthru_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
7219 : const struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes, void *md_buf, size_t md_len,
7220 : spdk_bdev_io_completion_cb cb, void *cb_arg)
7221 : {
7222 1 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
7223 : struct spdk_bdev_io *bdev_io;
7224 1 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
7225 :
7226 1 : if (!desc->write) {
7227 : /*
7228 : * Do not try to parse the NVMe command - we could maybe use bits in the opcode
7229 : * to easily determine if the command is a read or write, but for now just
7230 : * do not allow io_passthru with a read-only descriptor.
7231 : */
7232 0 : return -EBADF;
7233 : }
7234 :
7235 1 : if (spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_IO_MD))) {
7236 1 : return -ENOTSUP;
7237 : }
7238 :
7239 0 : bdev_io = bdev_channel_get_io(channel);
7240 0 : if (!bdev_io) {
7241 0 : return -ENOMEM;
7242 : }
7243 :
7244 0 : bdev_io->internal.ch = channel;
7245 0 : bdev_io->internal.desc = desc;
7246 0 : bdev_io->type = SPDK_BDEV_IO_TYPE_NVME_IO_MD;
7247 0 : bdev_io->u.nvme_passthru.cmd = *cmd;
7248 0 : bdev_io->u.nvme_passthru.buf = buf;
7249 0 : bdev_io->u.nvme_passthru.nbytes = nbytes;
7250 0 : bdev_io->u.nvme_passthru.md_buf = md_buf;
7251 0 : bdev_io->u.nvme_passthru.md_len = md_len;
7252 :
7253 0 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
7254 :
7255 0 : bdev_io_submit(bdev_io);
7256 0 : return 0;
7257 1 : }
7258 :
7259 : int
7260 0 : spdk_bdev_nvme_iov_passthru_md(struct spdk_bdev_desc *desc,
7261 : struct spdk_io_channel *ch,
7262 : const struct spdk_nvme_cmd *cmd,
7263 : struct iovec *iov, int iovcnt, size_t nbytes,
7264 : void *md_buf, size_t md_len,
7265 : spdk_bdev_io_completion_cb cb, void *cb_arg)
7266 : {
7267 0 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
7268 : struct spdk_bdev_io *bdev_io;
7269 0 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
7270 :
7271 0 : if (!desc->write) {
7272 : /*
7273 : * Do not try to parse the NVMe command - we could maybe use bits in the opcode
7274 : * to easily determine if the command is a read or write, but for now just
7275 : * do not allow io_passthru with a read-only descriptor.
7276 : */
7277 0 : return -EBADF;
7278 : }
7279 :
7280 0 : if (md_buf && spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_IO_MD))) {
7281 0 : return -ENOTSUP;
7282 0 : } else if (spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_IO))) {
7283 0 : return -ENOTSUP;
7284 : }
7285 :
7286 0 : bdev_io = bdev_channel_get_io(channel);
7287 0 : if (!bdev_io) {
7288 0 : return -ENOMEM;
7289 : }
7290 :
7291 0 : bdev_io->internal.ch = channel;
7292 0 : bdev_io->internal.desc = desc;
7293 0 : bdev_io->type = SPDK_BDEV_IO_TYPE_NVME_IOV_MD;
7294 0 : bdev_io->u.nvme_passthru.cmd = *cmd;
7295 0 : bdev_io->u.nvme_passthru.iovs = iov;
7296 0 : bdev_io->u.nvme_passthru.iovcnt = iovcnt;
7297 0 : bdev_io->u.nvme_passthru.nbytes = nbytes;
7298 0 : bdev_io->u.nvme_passthru.md_buf = md_buf;
7299 0 : bdev_io->u.nvme_passthru.md_len = md_len;
7300 :
7301 0 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
7302 :
7303 0 : bdev_io_submit(bdev_io);
7304 0 : return 0;
7305 0 : }
7306 :
7307 : static void bdev_abort_retry(void *ctx);
7308 : static void bdev_abort(struct spdk_bdev_io *parent_io);
7309 :
7310 : static void
7311 22 : bdev_abort_io_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
7312 : {
7313 22 : struct spdk_bdev_channel *channel = bdev_io->internal.ch;
7314 22 : struct spdk_bdev_io *parent_io = cb_arg;
7315 : struct spdk_bdev_io *bio_to_abort, *tmp_io;
7316 :
7317 22 : bio_to_abort = bdev_io->u.abort.bio_to_abort;
7318 :
7319 22 : spdk_bdev_free_io(bdev_io);
7320 :
7321 22 : if (!success) {
7322 : /* Check if the target I/O completed in the meantime. */
7323 2 : TAILQ_FOREACH(tmp_io, &channel->io_submitted, internal.ch_link) {
7324 1 : if (tmp_io == bio_to_abort) {
7325 0 : break;
7326 : }
7327 1 : }
7328 :
7329 : /* If the target I/O still exists, set the parent to failed. */
7330 1 : if (tmp_io != NULL) {
7331 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
7332 0 : }
7333 1 : }
7334 :
7335 22 : assert(parent_io->internal.f.split);
7336 :
7337 22 : parent_io->internal.split.outstanding--;
7338 22 : if (parent_io->internal.split.outstanding == 0) {
7339 16 : if (parent_io->internal.status == SPDK_BDEV_IO_STATUS_NOMEM) {
7340 0 : bdev_abort_retry(parent_io);
7341 0 : } else {
7342 16 : bdev_io_complete(parent_io);
7343 : }
7344 16 : }
7345 22 : }
7346 :
7347 : static int
7348 23 : bdev_abort_io(struct spdk_bdev_desc *desc, struct spdk_bdev_channel *channel,
7349 : struct spdk_bdev_io *bio_to_abort,
7350 : spdk_bdev_io_completion_cb cb, void *cb_arg)
7351 : {
7352 23 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
7353 : struct spdk_bdev_io *bdev_io;
7354 :
7355 23 : if (bio_to_abort->type == SPDK_BDEV_IO_TYPE_ABORT ||
7356 23 : bio_to_abort->type == SPDK_BDEV_IO_TYPE_RESET) {
7357 : /* TODO: Abort reset or abort request. */
7358 0 : return -ENOTSUP;
7359 : }
7360 :
7361 23 : bdev_io = bdev_channel_get_io(channel);
7362 23 : if (bdev_io == NULL) {
7363 1 : return -ENOMEM;
7364 : }
7365 :
7366 22 : bdev_io->internal.ch = channel;
7367 22 : bdev_io->internal.desc = desc;
7368 22 : bdev_io->type = SPDK_BDEV_IO_TYPE_ABORT;
7369 22 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
7370 :
7371 22 : if (bio_to_abort->internal.f.split) {
7372 6 : assert(bdev_io_should_split(bio_to_abort));
7373 6 : bdev_io->u.bdev.abort.bio_cb_arg = bio_to_abort;
7374 :
7375 : /* Parent abort request is not submitted directly, but to manage its
7376 : * execution add it to the submitted list here.
7377 : */
7378 6 : bdev_io->internal.submit_tsc = spdk_get_ticks();
7379 6 : bdev_ch_add_to_io_submitted(bdev_io);
7380 :
7381 6 : bdev_abort(bdev_io);
7382 :
7383 6 : return 0;
7384 : }
7385 :
7386 16 : bdev_io->u.abort.bio_to_abort = bio_to_abort;
7387 :
7388 : /* Submit the abort request to the underlying bdev module. */
7389 16 : bdev_io_submit(bdev_io);
7390 :
7391 16 : return 0;
7392 23 : }
7393 :
7394 : static bool
7395 46 : bdev_io_on_tailq(struct spdk_bdev_io *bdev_io, bdev_io_tailq_t *tailq)
7396 : {
7397 : struct spdk_bdev_io *iter;
7398 :
7399 46 : TAILQ_FOREACH(iter, tailq, internal.link) {
7400 0 : if (iter == bdev_io) {
7401 0 : return true;
7402 : }
7403 0 : }
7404 :
7405 46 : return false;
7406 46 : }
7407 :
7408 : static uint32_t
7409 18 : _bdev_abort(struct spdk_bdev_io *parent_io)
7410 : {
7411 18 : struct spdk_bdev_desc *desc = parent_io->internal.desc;
7412 18 : struct spdk_bdev_channel *channel = parent_io->internal.ch;
7413 : void *bio_cb_arg;
7414 : struct spdk_bdev_io *bio_to_abort;
7415 : uint32_t matched_ios;
7416 : int rc;
7417 :
7418 18 : bio_cb_arg = parent_io->u.bdev.abort.bio_cb_arg;
7419 :
7420 : /* matched_ios is returned and will be kept by the caller.
7421 : *
7422 : * This function will be used for two cases, 1) the same cb_arg is used for
7423 : * multiple I/Os, 2) a single large I/O is split into smaller ones.
7424 : * Incrementing split_outstanding directly here may confuse readers especially
7425 : * for the 1st case.
7426 : *
7427 : * Completion of I/O abort is processed after stack unwinding. Hence this trick
7428 : * works as expected.
7429 : */
7430 18 : matched_ios = 0;
7431 18 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
7432 :
7433 105 : TAILQ_FOREACH(bio_to_abort, &channel->io_submitted, internal.ch_link) {
7434 88 : if (bio_to_abort->internal.caller_ctx != bio_cb_arg) {
7435 65 : continue;
7436 : }
7437 :
7438 23 : if (bio_to_abort->internal.submit_tsc > parent_io->internal.submit_tsc) {
7439 : /* Any I/O which was submitted after this abort command should be excluded. */
7440 0 : continue;
7441 : }
7442 :
7443 : /* We can't abort a request that's being pushed/pulled or executed by accel */
7444 23 : if (bdev_io_on_tailq(bio_to_abort, &channel->io_accel_exec) ||
7445 23 : bdev_io_on_tailq(bio_to_abort, &channel->io_memory_domain)) {
7446 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
7447 0 : break;
7448 : }
7449 :
7450 23 : rc = bdev_abort_io(desc, channel, bio_to_abort, bdev_abort_io_done, parent_io);
7451 23 : if (rc != 0) {
7452 1 : if (rc == -ENOMEM) {
7453 1 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_NOMEM;
7454 1 : } else {
7455 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
7456 : }
7457 1 : break;
7458 : }
7459 22 : matched_ios++;
7460 22 : }
7461 :
7462 18 : return matched_ios;
7463 : }
7464 :
7465 : static void
7466 1 : bdev_abort_retry(void *ctx)
7467 : {
7468 1 : struct spdk_bdev_io *parent_io = ctx;
7469 : uint32_t matched_ios;
7470 :
7471 1 : matched_ios = _bdev_abort(parent_io);
7472 :
7473 1 : if (matched_ios == 0) {
7474 0 : if (parent_io->internal.status == SPDK_BDEV_IO_STATUS_NOMEM) {
7475 0 : bdev_queue_io_wait_with_cb(parent_io, bdev_abort_retry);
7476 0 : } else {
7477 : /* For retry, the case that no target I/O was found is success
7478 : * because it means target I/Os completed in the meantime.
7479 : */
7480 0 : bdev_io_complete(parent_io);
7481 : }
7482 0 : return;
7483 : }
7484 :
7485 : /* Use split_outstanding to manage the progress of aborting I/Os. */
7486 1 : parent_io->internal.f.split = true;
7487 1 : parent_io->internal.split.outstanding = matched_ios;
7488 1 : }
7489 :
7490 : static void
7491 17 : bdev_abort(struct spdk_bdev_io *parent_io)
7492 : {
7493 : uint32_t matched_ios;
7494 :
7495 17 : matched_ios = _bdev_abort(parent_io);
7496 :
7497 17 : if (matched_ios == 0) {
7498 2 : if (parent_io->internal.status == SPDK_BDEV_IO_STATUS_NOMEM) {
7499 1 : bdev_queue_io_wait_with_cb(parent_io, bdev_abort_retry);
7500 1 : } else {
7501 : /* The case the no target I/O was found is failure. */
7502 1 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
7503 1 : bdev_io_complete(parent_io);
7504 : }
7505 2 : return;
7506 : }
7507 :
7508 : /* Use split_outstanding to manage the progress of aborting I/Os. */
7509 15 : parent_io->internal.f.split = true;
7510 15 : parent_io->internal.split.outstanding = matched_ios;
7511 17 : }
7512 :
7513 : int
7514 12 : spdk_bdev_abort(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
7515 : void *bio_cb_arg,
7516 : spdk_bdev_io_completion_cb cb, void *cb_arg)
7517 : {
7518 12 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
7519 12 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
7520 : struct spdk_bdev_io *bdev_io;
7521 :
7522 12 : if (bio_cb_arg == NULL) {
7523 0 : return -EINVAL;
7524 : }
7525 :
7526 12 : if (!spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_ABORT)) {
7527 1 : return -ENOTSUP;
7528 : }
7529 :
7530 11 : bdev_io = bdev_channel_get_io(channel);
7531 11 : if (bdev_io == NULL) {
7532 0 : return -ENOMEM;
7533 : }
7534 :
7535 11 : bdev_io->internal.ch = channel;
7536 11 : bdev_io->internal.desc = desc;
7537 11 : bdev_io->internal.submit_tsc = spdk_get_ticks();
7538 11 : bdev_io->type = SPDK_BDEV_IO_TYPE_ABORT;
7539 11 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
7540 :
7541 11 : bdev_io->u.bdev.abort.bio_cb_arg = bio_cb_arg;
7542 :
7543 : /* Parent abort request is not submitted directly, but to manage its execution,
7544 : * add it to the submitted list here.
7545 : */
7546 11 : bdev_ch_add_to_io_submitted(bdev_io);
7547 :
7548 11 : bdev_abort(bdev_io);
7549 :
7550 11 : return 0;
7551 12 : }
7552 :
7553 : int
7554 4 : spdk_bdev_queue_io_wait(struct spdk_bdev *bdev, struct spdk_io_channel *ch,
7555 : struct spdk_bdev_io_wait_entry *entry)
7556 : {
7557 4 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
7558 4 : struct spdk_bdev_mgmt_channel *mgmt_ch = channel->shared_resource->mgmt_ch;
7559 :
7560 4 : if (bdev != entry->bdev) {
7561 0 : SPDK_ERRLOG("bdevs do not match\n");
7562 0 : return -EINVAL;
7563 : }
7564 :
7565 4 : if (mgmt_ch->per_thread_cache_count > 0) {
7566 0 : SPDK_ERRLOG("Cannot queue io_wait if spdk_bdev_io available in per-thread cache\n");
7567 0 : return -EINVAL;
7568 : }
7569 :
7570 4 : TAILQ_INSERT_TAIL(&mgmt_ch->io_wait_queue, entry, link);
7571 4 : return 0;
7572 4 : }
7573 :
7574 : static inline void
7575 610 : bdev_io_update_io_stat(struct spdk_bdev_io *bdev_io, uint64_t tsc_diff)
7576 : {
7577 610 : enum spdk_bdev_io_status io_status = bdev_io->internal.status;
7578 610 : struct spdk_bdev_io_stat *io_stat = bdev_io->internal.ch->stat;
7579 610 : uint64_t num_blocks = bdev_io->u.bdev.num_blocks;
7580 610 : uint32_t blocklen = bdev_io->bdev->blocklen;
7581 :
7582 610 : if (spdk_likely(io_status == SPDK_BDEV_IO_STATUS_SUCCESS)) {
7583 516 : switch (bdev_io->type) {
7584 : case SPDK_BDEV_IO_TYPE_READ:
7585 321 : io_stat->bytes_read += num_blocks * blocklen;
7586 321 : io_stat->num_read_ops++;
7587 321 : io_stat->read_latency_ticks += tsc_diff;
7588 321 : if (io_stat->max_read_latency_ticks < tsc_diff) {
7589 7 : io_stat->max_read_latency_ticks = tsc_diff;
7590 7 : }
7591 321 : if (io_stat->min_read_latency_ticks > tsc_diff) {
7592 42 : io_stat->min_read_latency_ticks = tsc_diff;
7593 42 : }
7594 321 : break;
7595 : case SPDK_BDEV_IO_TYPE_WRITE:
7596 75 : io_stat->bytes_written += num_blocks * blocklen;
7597 75 : io_stat->num_write_ops++;
7598 75 : io_stat->write_latency_ticks += tsc_diff;
7599 75 : if (io_stat->max_write_latency_ticks < tsc_diff) {
7600 4 : io_stat->max_write_latency_ticks = tsc_diff;
7601 4 : }
7602 75 : if (io_stat->min_write_latency_ticks > tsc_diff) {
7603 25 : io_stat->min_write_latency_ticks = tsc_diff;
7604 25 : }
7605 75 : break;
7606 : case SPDK_BDEV_IO_TYPE_UNMAP:
7607 20 : io_stat->bytes_unmapped += num_blocks * blocklen;
7608 20 : io_stat->num_unmap_ops++;
7609 20 : io_stat->unmap_latency_ticks += tsc_diff;
7610 20 : if (io_stat->max_unmap_latency_ticks < tsc_diff) {
7611 0 : io_stat->max_unmap_latency_ticks = tsc_diff;
7612 0 : }
7613 20 : if (io_stat->min_unmap_latency_ticks > tsc_diff) {
7614 3 : io_stat->min_unmap_latency_ticks = tsc_diff;
7615 3 : }
7616 20 : break;
7617 : case SPDK_BDEV_IO_TYPE_ZCOPY:
7618 : /* Track the data in the start phase only */
7619 4 : if (bdev_io->u.bdev.zcopy.start) {
7620 2 : if (bdev_io->u.bdev.zcopy.populate) {
7621 1 : io_stat->bytes_read += num_blocks * blocklen;
7622 1 : io_stat->num_read_ops++;
7623 1 : io_stat->read_latency_ticks += tsc_diff;
7624 1 : if (io_stat->max_read_latency_ticks < tsc_diff) {
7625 0 : io_stat->max_read_latency_ticks = tsc_diff;
7626 0 : }
7627 1 : if (io_stat->min_read_latency_ticks > tsc_diff) {
7628 1 : io_stat->min_read_latency_ticks = tsc_diff;
7629 1 : }
7630 1 : } else {
7631 1 : io_stat->bytes_written += num_blocks * blocklen;
7632 1 : io_stat->num_write_ops++;
7633 1 : io_stat->write_latency_ticks += tsc_diff;
7634 1 : if (io_stat->max_write_latency_ticks < tsc_diff) {
7635 0 : io_stat->max_write_latency_ticks = tsc_diff;
7636 0 : }
7637 1 : if (io_stat->min_write_latency_ticks > tsc_diff) {
7638 1 : io_stat->min_write_latency_ticks = tsc_diff;
7639 1 : }
7640 : }
7641 2 : }
7642 4 : break;
7643 : case SPDK_BDEV_IO_TYPE_COPY:
7644 21 : io_stat->bytes_copied += num_blocks * blocklen;
7645 21 : io_stat->num_copy_ops++;
7646 21 : bdev_io->internal.ch->stat->copy_latency_ticks += tsc_diff;
7647 21 : if (io_stat->max_copy_latency_ticks < tsc_diff) {
7648 0 : io_stat->max_copy_latency_ticks = tsc_diff;
7649 0 : }
7650 21 : if (io_stat->min_copy_latency_ticks > tsc_diff) {
7651 4 : io_stat->min_copy_latency_ticks = tsc_diff;
7652 4 : }
7653 21 : break;
7654 : default:
7655 75 : break;
7656 : }
7657 610 : } else if (io_status <= SPDK_BDEV_IO_STATUS_FAILED && io_status >= SPDK_MIN_BDEV_IO_STATUS) {
7658 94 : io_stat = bdev_io->bdev->internal.stat;
7659 94 : assert(io_stat->io_error != NULL);
7660 :
7661 94 : spdk_spin_lock(&bdev_io->bdev->internal.spinlock);
7662 94 : io_stat->io_error->error_status[-io_status - 1]++;
7663 94 : spdk_spin_unlock(&bdev_io->bdev->internal.spinlock);
7664 94 : }
7665 :
7666 : #ifdef SPDK_CONFIG_VTUNE
7667 : uint64_t now_tsc = spdk_get_ticks();
7668 : if (now_tsc > (bdev_io->internal.ch->start_tsc + bdev_io->internal.ch->interval_tsc)) {
7669 : uint64_t data[5];
7670 : struct spdk_bdev_io_stat *prev_stat = bdev_io->internal.ch->prev_stat;
7671 :
7672 : data[0] = io_stat->num_read_ops - prev_stat->num_read_ops;
7673 : data[1] = io_stat->bytes_read - prev_stat->bytes_read;
7674 : data[2] = io_stat->num_write_ops - prev_stat->num_write_ops;
7675 : data[3] = io_stat->bytes_written - prev_stat->bytes_written;
7676 : data[4] = bdev_io->bdev->fn_table->get_spin_time ?
7677 : bdev_io->bdev->fn_table->get_spin_time(spdk_bdev_io_get_io_channel(bdev_io)) : 0;
7678 :
7679 : __itt_metadata_add(g_bdev_mgr.domain, __itt_null, bdev_io->internal.ch->handle,
7680 : __itt_metadata_u64, 5, data);
7681 :
7682 : memcpy(prev_stat, io_stat, sizeof(struct spdk_bdev_io_stat));
7683 : bdev_io->internal.ch->start_tsc = now_tsc;
7684 : }
7685 : #endif
7686 610 : }
7687 :
7688 : static inline void
7689 610 : _bdev_io_complete(void *ctx)
7690 : {
7691 610 : struct spdk_bdev_io *bdev_io = ctx;
7692 :
7693 610 : if (spdk_unlikely(bdev_io_use_accel_sequence(bdev_io))) {
7694 0 : assert(bdev_io->internal.status != SPDK_BDEV_IO_STATUS_SUCCESS);
7695 0 : spdk_accel_sequence_abort(bdev_io->internal.accel_sequence);
7696 0 : }
7697 :
7698 610 : assert(bdev_io->internal.cb != NULL);
7699 610 : assert(spdk_get_thread() == spdk_bdev_io_get_thread(bdev_io));
7700 :
7701 1220 : bdev_io->internal.cb(bdev_io, bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS,
7702 610 : bdev_io->internal.caller_ctx);
7703 610 : }
7704 :
7705 : static inline void
7706 618 : bdev_io_complete(void *ctx)
7707 : {
7708 618 : struct spdk_bdev_io *bdev_io = ctx;
7709 618 : struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
7710 : uint64_t tsc, tsc_diff;
7711 :
7712 618 : if (spdk_unlikely(bdev_io->internal.f.in_submit_request)) {
7713 : /*
7714 : * Defer completion to avoid potential infinite recursion if the
7715 : * user's completion callback issues a new I/O.
7716 : */
7717 16 : spdk_thread_send_msg(spdk_bdev_io_get_thread(bdev_io),
7718 8 : bdev_io_complete, bdev_io);
7719 8 : return;
7720 : }
7721 :
7722 610 : tsc = spdk_get_ticks();
7723 610 : tsc_diff = tsc - bdev_io->internal.submit_tsc;
7724 :
7725 610 : bdev_ch_remove_from_io_submitted(bdev_io);
7726 610 : spdk_trace_record_tsc(tsc, TRACE_BDEV_IO_DONE, bdev_ch->trace_id, 0, (uintptr_t)bdev_io,
7727 : bdev_io->internal.caller_ctx, bdev_ch->queue_depth);
7728 :
7729 610 : if (bdev_ch->histogram) {
7730 4 : if (bdev_io->bdev->internal.histogram_io_type == 0 ||
7731 0 : bdev_io->bdev->internal.histogram_io_type == bdev_io->type) {
7732 : /*
7733 : * Tally all I/O types if the histogram_io_type is set to 0.
7734 : */
7735 4 : spdk_histogram_data_tally(bdev_ch->histogram, tsc_diff);
7736 4 : }
7737 4 : }
7738 :
7739 610 : bdev_io_update_io_stat(bdev_io, tsc_diff);
7740 610 : _bdev_io_complete(bdev_io);
7741 618 : }
7742 :
7743 : /* The difference between this function and bdev_io_complete() is that this should be called to
7744 : * complete IOs that haven't been submitted via bdev_io_submit(), as they weren't added onto the
7745 : * io_submitted list and don't have submit_tsc updated.
7746 : */
7747 : static inline void
7748 0 : bdev_io_complete_unsubmitted(struct spdk_bdev_io *bdev_io)
7749 : {
7750 : /* Since the IO hasn't been submitted it's bound to be failed */
7751 0 : assert(bdev_io->internal.status != SPDK_BDEV_IO_STATUS_SUCCESS);
7752 :
7753 : /* At this point we don't know if the IO is completed from submission context or not, but,
7754 : * since this is an error path, we can always do an spdk_thread_send_msg(). */
7755 0 : spdk_thread_send_msg(spdk_bdev_io_get_thread(bdev_io),
7756 0 : _bdev_io_complete, bdev_io);
7757 0 : }
7758 :
7759 : static void bdev_destroy_cb(void *io_device);
7760 :
7761 : static void
7762 15 : bdev_reset_complete(struct spdk_bdev *bdev, void *_ctx, int status)
7763 : {
7764 15 : struct spdk_bdev_io *bdev_io = _ctx;
7765 :
7766 15 : if (bdev_io->u.reset.ch_ref != NULL) {
7767 15 : spdk_put_io_channel(bdev_io->u.reset.ch_ref);
7768 15 : bdev_io->u.reset.ch_ref = NULL;
7769 15 : }
7770 :
7771 15 : bdev_io_complete(bdev_io);
7772 :
7773 15 : if (bdev->internal.status == SPDK_BDEV_STATUS_REMOVING &&
7774 1 : TAILQ_EMPTY(&bdev->internal.open_descs)) {
7775 1 : spdk_io_device_unregister(__bdev_to_io_dev(bdev), bdev_destroy_cb);
7776 1 : }
7777 15 : }
7778 :
7779 : static void
7780 18 : bdev_unfreeze_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
7781 : struct spdk_io_channel *_ch, void *_ctx)
7782 : {
7783 18 : struct spdk_bdev_io *bdev_io = _ctx;
7784 18 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
7785 : struct spdk_bdev_io *queued_reset;
7786 :
7787 18 : ch->flags &= ~BDEV_CH_RESET_IN_PROGRESS;
7788 18 : while (!TAILQ_EMPTY(&ch->queued_resets)) {
7789 0 : queued_reset = TAILQ_FIRST(&ch->queued_resets);
7790 0 : TAILQ_REMOVE(&ch->queued_resets, queued_reset, internal.link);
7791 0 : spdk_bdev_io_complete(queued_reset, bdev_io->internal.status);
7792 : }
7793 :
7794 18 : spdk_bdev_for_each_channel_continue(i, 0);
7795 18 : }
7796 :
7797 : static void
7798 0 : bdev_io_complete_sequence_cb(void *ctx, int status)
7799 : {
7800 0 : struct spdk_bdev_io *bdev_io = ctx;
7801 :
7802 : /* u.bdev.accel_sequence should have already been cleared at this point */
7803 0 : assert(bdev_io->u.bdev.accel_sequence == NULL);
7804 0 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS);
7805 0 : bdev_io->internal.f.has_accel_sequence = false;
7806 :
7807 0 : if (spdk_unlikely(status != 0)) {
7808 0 : SPDK_ERRLOG("Failed to execute accel sequence, status=%d\n", status);
7809 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
7810 0 : }
7811 :
7812 0 : bdev_io_complete(bdev_io);
7813 0 : }
7814 :
7815 : void
7816 598 : spdk_bdev_io_complete(struct spdk_bdev_io *bdev_io, enum spdk_bdev_io_status status)
7817 : {
7818 598 : struct spdk_bdev *bdev = bdev_io->bdev;
7819 598 : struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
7820 598 : struct spdk_bdev_shared_resource *shared_resource = bdev_ch->shared_resource;
7821 :
7822 598 : if (spdk_unlikely(bdev_io->internal.status != SPDK_BDEV_IO_STATUS_PENDING)) {
7823 0 : SPDK_ERRLOG("Unexpected completion on IO from %s module, status was %s\n",
7824 : spdk_bdev_get_module_name(bdev),
7825 : bdev_io_status_get_string(bdev_io->internal.status));
7826 0 : assert(false);
7827 : }
7828 598 : bdev_io->internal.status = status;
7829 :
7830 598 : if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_RESET)) {
7831 16 : bool unlock_channels = false;
7832 :
7833 16 : if (status == SPDK_BDEV_IO_STATUS_NOMEM) {
7834 0 : SPDK_ERRLOG("NOMEM returned for reset\n");
7835 0 : }
7836 16 : spdk_spin_lock(&bdev->internal.spinlock);
7837 16 : if (bdev_io == bdev->internal.reset_in_progress) {
7838 15 : bdev->internal.reset_in_progress = NULL;
7839 15 : unlock_channels = true;
7840 15 : }
7841 16 : spdk_spin_unlock(&bdev->internal.spinlock);
7842 :
7843 16 : if (unlock_channels) {
7844 15 : spdk_bdev_for_each_channel(bdev, bdev_unfreeze_channel, bdev_io,
7845 : bdev_reset_complete);
7846 15 : return;
7847 : }
7848 1 : } else {
7849 582 : bdev_io_decrement_outstanding(bdev_ch, shared_resource);
7850 582 : if (spdk_likely(status == SPDK_BDEV_IO_STATUS_SUCCESS)) {
7851 485 : if (bdev_io_needs_sequence_exec(bdev_io->internal.desc, bdev_io)) {
7852 0 : bdev_io_exec_sequence(bdev_io, bdev_io_complete_sequence_cb);
7853 0 : return;
7854 485 : } else if (spdk_unlikely(bdev_io->internal.f.has_bounce_buf &&
7855 : !bdev_io_use_accel_sequence(bdev_io))) {
7856 26 : _bdev_io_push_bounce_data_buffer(bdev_io,
7857 : _bdev_io_complete_push_bounce_done);
7858 : /* bdev IO will be completed in the callback */
7859 26 : return;
7860 : }
7861 459 : }
7862 :
7863 556 : if (spdk_unlikely(_bdev_io_handle_no_mem(bdev_io, BDEV_IO_RETRY_STATE_SUBMIT))) {
7864 5 : return;
7865 : }
7866 : }
7867 :
7868 552 : bdev_io_complete(bdev_io);
7869 598 : }
7870 :
7871 : void
7872 0 : spdk_bdev_io_complete_scsi_status(struct spdk_bdev_io *bdev_io, enum spdk_scsi_status sc,
7873 : enum spdk_scsi_sense sk, uint8_t asc, uint8_t ascq)
7874 : {
7875 : enum spdk_bdev_io_status status;
7876 :
7877 0 : if (sc == SPDK_SCSI_STATUS_GOOD) {
7878 0 : status = SPDK_BDEV_IO_STATUS_SUCCESS;
7879 0 : } else {
7880 0 : status = SPDK_BDEV_IO_STATUS_SCSI_ERROR;
7881 0 : bdev_io->internal.error.scsi.sc = sc;
7882 0 : bdev_io->internal.error.scsi.sk = sk;
7883 0 : bdev_io->internal.error.scsi.asc = asc;
7884 0 : bdev_io->internal.error.scsi.ascq = ascq;
7885 : }
7886 :
7887 0 : spdk_bdev_io_complete(bdev_io, status);
7888 0 : }
7889 :
7890 : void
7891 0 : spdk_bdev_io_get_scsi_status(const struct spdk_bdev_io *bdev_io,
7892 : int *sc, int *sk, int *asc, int *ascq)
7893 : {
7894 0 : assert(sc != NULL);
7895 0 : assert(sk != NULL);
7896 0 : assert(asc != NULL);
7897 0 : assert(ascq != NULL);
7898 :
7899 0 : switch (bdev_io->internal.status) {
7900 : case SPDK_BDEV_IO_STATUS_SUCCESS:
7901 0 : *sc = SPDK_SCSI_STATUS_GOOD;
7902 0 : *sk = SPDK_SCSI_SENSE_NO_SENSE;
7903 0 : *asc = SPDK_SCSI_ASC_NO_ADDITIONAL_SENSE;
7904 0 : *ascq = SPDK_SCSI_ASCQ_CAUSE_NOT_REPORTABLE;
7905 0 : break;
7906 : case SPDK_BDEV_IO_STATUS_NVME_ERROR:
7907 0 : spdk_scsi_nvme_translate(bdev_io, sc, sk, asc, ascq);
7908 0 : break;
7909 : case SPDK_BDEV_IO_STATUS_MISCOMPARE:
7910 0 : *sc = SPDK_SCSI_STATUS_CHECK_CONDITION;
7911 0 : *sk = SPDK_SCSI_SENSE_MISCOMPARE;
7912 0 : *asc = SPDK_SCSI_ASC_MISCOMPARE_DURING_VERIFY_OPERATION;
7913 0 : *ascq = bdev_io->internal.error.scsi.ascq;
7914 0 : break;
7915 : case SPDK_BDEV_IO_STATUS_SCSI_ERROR:
7916 0 : *sc = bdev_io->internal.error.scsi.sc;
7917 0 : *sk = bdev_io->internal.error.scsi.sk;
7918 0 : *asc = bdev_io->internal.error.scsi.asc;
7919 0 : *ascq = bdev_io->internal.error.scsi.ascq;
7920 0 : break;
7921 : default:
7922 0 : *sc = SPDK_SCSI_STATUS_CHECK_CONDITION;
7923 0 : *sk = SPDK_SCSI_SENSE_ABORTED_COMMAND;
7924 0 : *asc = SPDK_SCSI_ASC_NO_ADDITIONAL_SENSE;
7925 0 : *ascq = SPDK_SCSI_ASCQ_CAUSE_NOT_REPORTABLE;
7926 0 : break;
7927 : }
7928 0 : }
7929 :
7930 : void
7931 0 : spdk_bdev_io_complete_aio_status(struct spdk_bdev_io *bdev_io, int aio_result)
7932 : {
7933 : enum spdk_bdev_io_status status;
7934 :
7935 0 : if (aio_result == 0) {
7936 0 : status = SPDK_BDEV_IO_STATUS_SUCCESS;
7937 0 : } else {
7938 0 : status = SPDK_BDEV_IO_STATUS_AIO_ERROR;
7939 : }
7940 :
7941 0 : bdev_io->internal.error.aio_result = aio_result;
7942 :
7943 0 : spdk_bdev_io_complete(bdev_io, status);
7944 0 : }
7945 :
7946 : void
7947 0 : spdk_bdev_io_get_aio_status(const struct spdk_bdev_io *bdev_io, int *aio_result)
7948 : {
7949 0 : assert(aio_result != NULL);
7950 :
7951 0 : if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_AIO_ERROR) {
7952 0 : *aio_result = bdev_io->internal.error.aio_result;
7953 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS) {
7954 0 : *aio_result = 0;
7955 0 : } else {
7956 0 : *aio_result = -EIO;
7957 : }
7958 0 : }
7959 :
7960 : void
7961 0 : spdk_bdev_io_complete_nvme_status(struct spdk_bdev_io *bdev_io, uint32_t cdw0, int sct, int sc)
7962 : {
7963 : enum spdk_bdev_io_status status;
7964 :
7965 0 : if (spdk_likely(sct == SPDK_NVME_SCT_GENERIC && sc == SPDK_NVME_SC_SUCCESS)) {
7966 0 : status = SPDK_BDEV_IO_STATUS_SUCCESS;
7967 0 : } else if (sct == SPDK_NVME_SCT_GENERIC && sc == SPDK_NVME_SC_ABORTED_BY_REQUEST) {
7968 0 : status = SPDK_BDEV_IO_STATUS_ABORTED;
7969 0 : } else {
7970 0 : status = SPDK_BDEV_IO_STATUS_NVME_ERROR;
7971 : }
7972 :
7973 0 : bdev_io->internal.error.nvme.cdw0 = cdw0;
7974 0 : bdev_io->internal.error.nvme.sct = sct;
7975 0 : bdev_io->internal.error.nvme.sc = sc;
7976 :
7977 0 : spdk_bdev_io_complete(bdev_io, status);
7978 0 : }
7979 :
7980 : void
7981 0 : spdk_bdev_io_get_nvme_status(const struct spdk_bdev_io *bdev_io, uint32_t *cdw0, int *sct, int *sc)
7982 : {
7983 0 : assert(sct != NULL);
7984 0 : assert(sc != NULL);
7985 0 : assert(cdw0 != NULL);
7986 :
7987 0 : if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_ABORT)) {
7988 0 : *sct = SPDK_NVME_SCT_GENERIC;
7989 0 : *sc = SPDK_NVME_SC_SUCCESS;
7990 0 : if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS) {
7991 0 : *cdw0 = 0;
7992 0 : } else {
7993 0 : *cdw0 = 1U;
7994 : }
7995 0 : return;
7996 : }
7997 :
7998 0 : if (spdk_likely(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS)) {
7999 0 : *sct = SPDK_NVME_SCT_GENERIC;
8000 0 : *sc = SPDK_NVME_SC_SUCCESS;
8001 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_NVME_ERROR) {
8002 0 : *sct = bdev_io->internal.error.nvme.sct;
8003 0 : *sc = bdev_io->internal.error.nvme.sc;
8004 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_ABORTED) {
8005 0 : *sct = SPDK_NVME_SCT_GENERIC;
8006 0 : *sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
8007 0 : } else {
8008 0 : *sct = SPDK_NVME_SCT_GENERIC;
8009 0 : *sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
8010 : }
8011 :
8012 0 : *cdw0 = bdev_io->internal.error.nvme.cdw0;
8013 0 : }
8014 :
8015 : void
8016 0 : spdk_bdev_io_get_nvme_fused_status(const struct spdk_bdev_io *bdev_io, uint32_t *cdw0,
8017 : int *first_sct, int *first_sc, int *second_sct, int *second_sc)
8018 : {
8019 0 : assert(first_sct != NULL);
8020 0 : assert(first_sc != NULL);
8021 0 : assert(second_sct != NULL);
8022 0 : assert(second_sc != NULL);
8023 0 : assert(cdw0 != NULL);
8024 :
8025 0 : if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_NVME_ERROR) {
8026 0 : if (bdev_io->internal.error.nvme.sct == SPDK_NVME_SCT_MEDIA_ERROR &&
8027 0 : bdev_io->internal.error.nvme.sc == SPDK_NVME_SC_COMPARE_FAILURE) {
8028 0 : *first_sct = bdev_io->internal.error.nvme.sct;
8029 0 : *first_sc = bdev_io->internal.error.nvme.sc;
8030 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
8031 0 : *second_sc = SPDK_NVME_SC_ABORTED_FAILED_FUSED;
8032 0 : } else {
8033 0 : *first_sct = SPDK_NVME_SCT_GENERIC;
8034 0 : *first_sc = SPDK_NVME_SC_SUCCESS;
8035 0 : *second_sct = bdev_io->internal.error.nvme.sct;
8036 0 : *second_sc = bdev_io->internal.error.nvme.sc;
8037 : }
8038 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_ABORTED) {
8039 0 : *first_sct = SPDK_NVME_SCT_GENERIC;
8040 0 : *first_sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
8041 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
8042 0 : *second_sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
8043 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS) {
8044 0 : *first_sct = SPDK_NVME_SCT_GENERIC;
8045 0 : *first_sc = SPDK_NVME_SC_SUCCESS;
8046 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
8047 0 : *second_sc = SPDK_NVME_SC_SUCCESS;
8048 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_FIRST_FUSED_FAILED) {
8049 0 : *first_sct = SPDK_NVME_SCT_GENERIC;
8050 0 : *first_sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
8051 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
8052 0 : *second_sc = SPDK_NVME_SC_ABORTED_FAILED_FUSED;
8053 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_MISCOMPARE) {
8054 0 : *first_sct = SPDK_NVME_SCT_MEDIA_ERROR;
8055 0 : *first_sc = SPDK_NVME_SC_COMPARE_FAILURE;
8056 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
8057 0 : *second_sc = SPDK_NVME_SC_ABORTED_FAILED_FUSED;
8058 0 : } else {
8059 0 : *first_sct = SPDK_NVME_SCT_GENERIC;
8060 0 : *first_sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
8061 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
8062 0 : *second_sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
8063 : }
8064 :
8065 0 : *cdw0 = bdev_io->internal.error.nvme.cdw0;
8066 0 : }
8067 :
8068 : void
8069 0 : spdk_bdev_io_complete_base_io_status(struct spdk_bdev_io *bdev_io,
8070 : const struct spdk_bdev_io *base_io)
8071 : {
8072 0 : switch (base_io->internal.status) {
8073 : case SPDK_BDEV_IO_STATUS_NVME_ERROR:
8074 0 : spdk_bdev_io_complete_nvme_status(bdev_io,
8075 0 : base_io->internal.error.nvme.cdw0,
8076 0 : base_io->internal.error.nvme.sct,
8077 0 : base_io->internal.error.nvme.sc);
8078 0 : break;
8079 : case SPDK_BDEV_IO_STATUS_SCSI_ERROR:
8080 0 : spdk_bdev_io_complete_scsi_status(bdev_io,
8081 0 : base_io->internal.error.scsi.sc,
8082 0 : base_io->internal.error.scsi.sk,
8083 0 : base_io->internal.error.scsi.asc,
8084 0 : base_io->internal.error.scsi.ascq);
8085 0 : break;
8086 : case SPDK_BDEV_IO_STATUS_AIO_ERROR:
8087 0 : spdk_bdev_io_complete_aio_status(bdev_io, base_io->internal.error.aio_result);
8088 0 : break;
8089 : default:
8090 0 : spdk_bdev_io_complete(bdev_io, base_io->internal.status);
8091 0 : break;
8092 : }
8093 0 : }
8094 :
8095 : struct spdk_thread *
8096 660 : spdk_bdev_io_get_thread(struct spdk_bdev_io *bdev_io)
8097 : {
8098 660 : return spdk_io_channel_get_thread(bdev_io->internal.ch->channel);
8099 : }
8100 :
8101 : struct spdk_io_channel *
8102 70 : spdk_bdev_io_get_io_channel(struct spdk_bdev_io *bdev_io)
8103 : {
8104 70 : return bdev_io->internal.ch->channel;
8105 : }
8106 :
8107 : static int
8108 128 : bdev_register(struct spdk_bdev *bdev)
8109 : {
8110 : char *bdev_name;
8111 : char uuid[SPDK_UUID_STRING_LEN];
8112 : struct spdk_iobuf_opts iobuf_opts;
8113 : int ret;
8114 :
8115 128 : assert(bdev->module != NULL);
8116 :
8117 128 : if (!bdev->name) {
8118 0 : SPDK_ERRLOG("Bdev name is NULL\n");
8119 0 : return -EINVAL;
8120 : }
8121 :
8122 128 : if (!strlen(bdev->name)) {
8123 0 : SPDK_ERRLOG("Bdev name must not be an empty string\n");
8124 0 : return -EINVAL;
8125 : }
8126 :
8127 : /* Users often register their own I/O devices using the bdev name. In
8128 : * order to avoid conflicts, prepend bdev_. */
8129 128 : bdev_name = spdk_sprintf_alloc("bdev_%s", bdev->name);
8130 128 : if (!bdev_name) {
8131 0 : SPDK_ERRLOG("Unable to allocate memory for internal bdev name.\n");
8132 0 : return -ENOMEM;
8133 : }
8134 :
8135 128 : bdev->internal.stat = bdev_alloc_io_stat(true);
8136 128 : if (!bdev->internal.stat) {
8137 0 : SPDK_ERRLOG("Unable to allocate I/O statistics structure.\n");
8138 0 : free(bdev_name);
8139 0 : return -ENOMEM;
8140 : }
8141 :
8142 128 : bdev->internal.status = SPDK_BDEV_STATUS_READY;
8143 128 : bdev->internal.measured_queue_depth = UINT64_MAX;
8144 128 : bdev->internal.claim_type = SPDK_BDEV_CLAIM_NONE;
8145 128 : memset(&bdev->internal.claim, 0, sizeof(bdev->internal.claim));
8146 128 : bdev->internal.qd_poller = NULL;
8147 128 : bdev->internal.qos = NULL;
8148 :
8149 128 : TAILQ_INIT(&bdev->internal.open_descs);
8150 128 : TAILQ_INIT(&bdev->internal.locked_ranges);
8151 128 : TAILQ_INIT(&bdev->internal.pending_locked_ranges);
8152 128 : TAILQ_INIT(&bdev->aliases);
8153 :
8154 : /* UUID may be specified by the user or defined by bdev itself.
8155 : * Otherwise it will be generated here, so this field will never be empty. */
8156 128 : if (spdk_uuid_is_null(&bdev->uuid)) {
8157 42 : spdk_uuid_generate(&bdev->uuid);
8158 42 : }
8159 :
8160 : /* Add the UUID alias only if it's different than the name */
8161 128 : spdk_uuid_fmt_lower(uuid, sizeof(uuid), &bdev->uuid);
8162 128 : if (strcmp(bdev->name, uuid) != 0) {
8163 127 : ret = spdk_bdev_alias_add(bdev, uuid);
8164 127 : if (ret != 0) {
8165 2 : SPDK_ERRLOG("Unable to add uuid:%s alias for bdev %s\n", uuid, bdev->name);
8166 2 : bdev_free_io_stat(bdev->internal.stat);
8167 2 : free(bdev_name);
8168 2 : return ret;
8169 : }
8170 125 : }
8171 :
8172 126 : spdk_iobuf_get_opts(&iobuf_opts, sizeof(iobuf_opts));
8173 126 : if (spdk_bdev_get_buf_align(bdev) > 1) {
8174 0 : bdev->max_rw_size = spdk_min(bdev->max_rw_size ? bdev->max_rw_size : UINT32_MAX,
8175 : iobuf_opts.large_bufsize / bdev->blocklen);
8176 0 : }
8177 :
8178 : /* If the user didn't specify a write unit size, set it to one. */
8179 126 : if (bdev->write_unit_size == 0) {
8180 122 : bdev->write_unit_size = 1;
8181 122 : }
8182 :
8183 : /* Set ACWU value to the write unit size if bdev module did not set it (does not support it natively) */
8184 126 : if (bdev->acwu == 0) {
8185 122 : bdev->acwu = bdev->write_unit_size;
8186 122 : }
8187 :
8188 126 : if (bdev->phys_blocklen == 0) {
8189 122 : bdev->phys_blocklen = spdk_bdev_get_data_block_size(bdev);
8190 122 : }
8191 :
8192 126 : if (!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COPY)) {
8193 0 : bdev->max_copy = bdev_get_max_write(bdev, iobuf_opts.large_bufsize);
8194 0 : }
8195 :
8196 126 : if (!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE_ZEROES)) {
8197 0 : bdev->max_write_zeroes = bdev_get_max_write(bdev, ZERO_BUFFER_SIZE);
8198 0 : }
8199 :
8200 126 : bdev->internal.reset_in_progress = NULL;
8201 126 : bdev->internal.qd_poll_in_progress = false;
8202 126 : bdev->internal.period = 0;
8203 126 : bdev->internal.new_period = 0;
8204 126 : bdev->internal.trace_id = spdk_trace_register_owner(OWNER_TYPE_BDEV, bdev_name);
8205 :
8206 : /*
8207 : * Initialize spinlock before registering IO device because spinlock is used in
8208 : * bdev_channel_create
8209 : */
8210 126 : spdk_spin_init(&bdev->internal.spinlock);
8211 :
8212 252 : spdk_io_device_register(__bdev_to_io_dev(bdev),
8213 : bdev_channel_create, bdev_channel_destroy,
8214 : sizeof(struct spdk_bdev_channel),
8215 126 : bdev_name);
8216 :
8217 : /*
8218 : * Register bdev name only after the bdev object is ready.
8219 : * After bdev_name_add returns, it is possible for other threads to start using the bdev,
8220 : * create IO channels...
8221 : */
8222 126 : ret = bdev_name_add(&bdev->internal.bdev_name, bdev, bdev->name);
8223 126 : if (ret != 0) {
8224 0 : spdk_io_device_unregister(__bdev_to_io_dev(bdev), NULL);
8225 0 : bdev_free_io_stat(bdev->internal.stat);
8226 0 : spdk_spin_destroy(&bdev->internal.spinlock);
8227 0 : free(bdev_name);
8228 0 : return ret;
8229 : }
8230 :
8231 126 : free(bdev_name);
8232 :
8233 126 : SPDK_DEBUGLOG(bdev, "Inserting bdev %s into list\n", bdev->name);
8234 126 : TAILQ_INSERT_TAIL(&g_bdev_mgr.bdevs, bdev, internal.link);
8235 :
8236 126 : return 0;
8237 128 : }
8238 :
8239 : static void
8240 127 : bdev_destroy_cb(void *io_device)
8241 : {
8242 : int rc;
8243 : struct spdk_bdev *bdev;
8244 : spdk_bdev_unregister_cb cb_fn;
8245 : void *cb_arg;
8246 :
8247 127 : bdev = __bdev_from_io_dev(io_device);
8248 :
8249 127 : if (bdev->internal.unregister_td != spdk_get_thread()) {
8250 1 : spdk_thread_send_msg(bdev->internal.unregister_td, bdev_destroy_cb, io_device);
8251 1 : return;
8252 : }
8253 :
8254 126 : cb_fn = bdev->internal.unregister_cb;
8255 126 : cb_arg = bdev->internal.unregister_ctx;
8256 :
8257 126 : spdk_spin_destroy(&bdev->internal.spinlock);
8258 126 : free(bdev->internal.qos);
8259 126 : bdev_free_io_stat(bdev->internal.stat);
8260 126 : spdk_trace_unregister_owner(bdev->internal.trace_id);
8261 :
8262 126 : rc = bdev->fn_table->destruct(bdev->ctxt);
8263 126 : if (rc < 0) {
8264 0 : SPDK_ERRLOG("destruct failed\n");
8265 0 : }
8266 126 : if (rc <= 0 && cb_fn != NULL) {
8267 10 : cb_fn(cb_arg, rc);
8268 10 : }
8269 127 : }
8270 :
8271 : void
8272 2 : spdk_bdev_destruct_done(struct spdk_bdev *bdev, int bdeverrno)
8273 : {
8274 2 : if (bdev->internal.unregister_cb != NULL) {
8275 0 : bdev->internal.unregister_cb(bdev->internal.unregister_ctx, bdeverrno);
8276 0 : }
8277 2 : }
8278 :
8279 : static void
8280 19 : _remove_notify(void *arg)
8281 : {
8282 19 : struct spdk_bdev_desc *desc = arg;
8283 :
8284 19 : _event_notify(desc, SPDK_BDEV_EVENT_REMOVE);
8285 19 : }
8286 :
8287 : /* returns: 0 - bdev removed and ready to be destructed.
8288 : * -EBUSY - bdev can't be destructed yet. */
8289 : static int
8290 141 : bdev_unregister_unsafe(struct spdk_bdev *bdev)
8291 : {
8292 : struct spdk_bdev_desc *desc, *tmp;
8293 141 : int rc = 0;
8294 : char uuid[SPDK_UUID_STRING_LEN];
8295 :
8296 141 : assert(spdk_spin_held(&g_bdev_mgr.spinlock));
8297 141 : assert(spdk_spin_held(&bdev->internal.spinlock));
8298 :
8299 : /* Notify each descriptor about hotremoval */
8300 160 : TAILQ_FOREACH_SAFE(desc, &bdev->internal.open_descs, link, tmp) {
8301 19 : rc = -EBUSY;
8302 : /*
8303 : * Defer invocation of the event_cb to a separate message that will
8304 : * run later on its thread. This ensures this context unwinds and
8305 : * we don't recursively unregister this bdev again if the event_cb
8306 : * immediately closes its descriptor.
8307 : */
8308 19 : event_notify(desc, _remove_notify);
8309 19 : }
8310 :
8311 : /* If there are no descriptors, proceed removing the bdev */
8312 141 : if (rc == 0) {
8313 126 : TAILQ_REMOVE(&g_bdev_mgr.bdevs, bdev, internal.link);
8314 126 : SPDK_DEBUGLOG(bdev, "Removing bdev %s from list done\n", bdev->name);
8315 :
8316 : /* Delete the name and the UUID alias */
8317 126 : spdk_uuid_fmt_lower(uuid, sizeof(uuid), &bdev->uuid);
8318 126 : bdev_name_del_unsafe(&bdev->internal.bdev_name);
8319 126 : bdev_alias_del(bdev, uuid, bdev_name_del_unsafe);
8320 :
8321 126 : spdk_notify_send("bdev_unregister", spdk_bdev_get_name(bdev));
8322 :
8323 126 : if (bdev->internal.reset_in_progress != NULL) {
8324 : /* If reset is in progress, let the completion callback for reset
8325 : * unregister the bdev.
8326 : */
8327 1 : rc = -EBUSY;
8328 1 : }
8329 126 : }
8330 :
8331 141 : return rc;
8332 : }
8333 :
8334 : static void
8335 4 : bdev_unregister_abort_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
8336 : struct spdk_io_channel *io_ch, void *_ctx)
8337 : {
8338 4 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
8339 :
8340 4 : bdev_channel_abort_queued_ios(bdev_ch);
8341 4 : spdk_bdev_for_each_channel_continue(i, 0);
8342 4 : }
8343 :
8344 : static void
8345 126 : bdev_unregister(struct spdk_bdev *bdev, void *_ctx, int status)
8346 : {
8347 : int rc;
8348 :
8349 126 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8350 126 : spdk_spin_lock(&bdev->internal.spinlock);
8351 : /*
8352 : * Set the status to REMOVING after completing to abort channels. Otherwise,
8353 : * the last spdk_bdev_close() may call spdk_io_device_unregister() while
8354 : * spdk_bdev_for_each_channel() is executed and spdk_io_device_unregister()
8355 : * may fail.
8356 : */
8357 126 : bdev->internal.status = SPDK_BDEV_STATUS_REMOVING;
8358 126 : rc = bdev_unregister_unsafe(bdev);
8359 126 : spdk_spin_unlock(&bdev->internal.spinlock);
8360 126 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8361 :
8362 126 : if (rc == 0) {
8363 110 : spdk_io_device_unregister(__bdev_to_io_dev(bdev), bdev_destroy_cb);
8364 110 : }
8365 126 : }
8366 :
8367 : void
8368 133 : spdk_bdev_unregister(struct spdk_bdev *bdev, spdk_bdev_unregister_cb cb_fn, void *cb_arg)
8369 : {
8370 : struct spdk_thread *thread;
8371 :
8372 133 : SPDK_DEBUGLOG(bdev, "Removing bdev %s from list\n", bdev->name);
8373 :
8374 133 : thread = spdk_get_thread();
8375 133 : if (!thread) {
8376 : /* The user called this from a non-SPDK thread. */
8377 0 : if (cb_fn != NULL) {
8378 0 : cb_fn(cb_arg, -ENOTSUP);
8379 0 : }
8380 0 : return;
8381 : }
8382 :
8383 133 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8384 133 : if (bdev->internal.status == SPDK_BDEV_STATUS_UNREGISTERING ||
8385 133 : bdev->internal.status == SPDK_BDEV_STATUS_REMOVING) {
8386 7 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8387 7 : if (cb_fn) {
8388 0 : cb_fn(cb_arg, -EBUSY);
8389 0 : }
8390 7 : return;
8391 : }
8392 :
8393 126 : spdk_spin_lock(&bdev->internal.spinlock);
8394 126 : bdev->internal.status = SPDK_BDEV_STATUS_UNREGISTERING;
8395 126 : bdev->internal.unregister_cb = cb_fn;
8396 126 : bdev->internal.unregister_ctx = cb_arg;
8397 126 : bdev->internal.unregister_td = thread;
8398 126 : spdk_spin_unlock(&bdev->internal.spinlock);
8399 126 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8400 :
8401 126 : spdk_bdev_set_qd_sampling_period(bdev, 0);
8402 :
8403 126 : spdk_bdev_for_each_channel(bdev, bdev_unregister_abort_channel, bdev,
8404 : bdev_unregister);
8405 133 : }
8406 :
8407 : int
8408 4 : spdk_bdev_unregister_by_name(const char *bdev_name, struct spdk_bdev_module *module,
8409 : spdk_bdev_unregister_cb cb_fn, void *cb_arg)
8410 : {
8411 : struct spdk_bdev_desc *desc;
8412 : struct spdk_bdev *bdev;
8413 : int rc;
8414 :
8415 4 : rc = spdk_bdev_open_ext(bdev_name, false, _tmp_bdev_event_cb, NULL, &desc);
8416 4 : if (rc != 0) {
8417 1 : SPDK_ERRLOG("Failed to open bdev with name: %s\n", bdev_name);
8418 1 : return rc;
8419 : }
8420 :
8421 3 : bdev = spdk_bdev_desc_get_bdev(desc);
8422 :
8423 3 : if (bdev->module != module) {
8424 1 : spdk_bdev_close(desc);
8425 1 : SPDK_ERRLOG("Bdev %s was not registered by the specified module.\n",
8426 : bdev_name);
8427 1 : return -ENODEV;
8428 : }
8429 :
8430 2 : spdk_bdev_unregister(bdev, cb_fn, cb_arg);
8431 :
8432 2 : spdk_bdev_close(desc);
8433 :
8434 2 : return 0;
8435 4 : }
8436 :
8437 : static int
8438 259 : bdev_start_qos(struct spdk_bdev *bdev)
8439 : {
8440 : struct set_qos_limit_ctx *ctx;
8441 :
8442 : /* Enable QoS */
8443 259 : if (bdev->internal.qos && bdev->internal.qos->thread == NULL) {
8444 2 : ctx = calloc(1, sizeof(*ctx));
8445 2 : if (ctx == NULL) {
8446 0 : SPDK_ERRLOG("Failed to allocate memory for QoS context\n");
8447 0 : return -ENOMEM;
8448 : }
8449 2 : ctx->bdev = bdev;
8450 2 : spdk_bdev_for_each_channel(bdev, bdev_enable_qos_msg, ctx, bdev_enable_qos_done);
8451 2 : }
8452 :
8453 259 : return 0;
8454 259 : }
8455 :
8456 : static void
8457 24 : log_already_claimed(enum spdk_log_level level, const int line, const char *func, const char *detail,
8458 : struct spdk_bdev *bdev)
8459 : {
8460 : enum spdk_bdev_claim_type type;
8461 : const char *typename, *modname;
8462 : extern struct spdk_log_flag SPDK_LOG_bdev;
8463 :
8464 24 : assert(spdk_spin_held(&bdev->internal.spinlock));
8465 :
8466 24 : if (level >= SPDK_LOG_INFO && !SPDK_LOG_bdev.enabled) {
8467 0 : return;
8468 : }
8469 :
8470 24 : type = bdev->internal.claim_type;
8471 24 : typename = spdk_bdev_claim_get_name(type);
8472 :
8473 24 : if (type == SPDK_BDEV_CLAIM_EXCL_WRITE) {
8474 6 : modname = bdev->internal.claim.v1.module->name;
8475 12 : spdk_log(level, __FILE__, line, func, "bdev %s %s: type %s by module %s\n",
8476 6 : bdev->name, detail, typename, modname);
8477 6 : return;
8478 : }
8479 :
8480 18 : if (claim_type_is_v2(type)) {
8481 : struct spdk_bdev_module_claim *claim;
8482 :
8483 36 : TAILQ_FOREACH(claim, &bdev->internal.claim.v2.claims, link) {
8484 18 : modname = claim->module->name;
8485 36 : spdk_log(level, __FILE__, line, func, "bdev %s %s: type %s by module %s\n",
8486 18 : bdev->name, detail, typename, modname);
8487 18 : }
8488 18 : return;
8489 : }
8490 :
8491 0 : assert(false);
8492 24 : }
8493 :
8494 : static int
8495 268 : bdev_open(struct spdk_bdev *bdev, bool write, struct spdk_bdev_desc *desc)
8496 : {
8497 : struct spdk_thread *thread;
8498 268 : int rc = 0;
8499 :
8500 268 : thread = spdk_get_thread();
8501 268 : if (!thread) {
8502 0 : SPDK_ERRLOG("Cannot open bdev from non-SPDK thread.\n");
8503 0 : return -ENOTSUP;
8504 : }
8505 :
8506 268 : SPDK_DEBUGLOG(bdev, "Opening descriptor %p for bdev %s on thread %p\n", desc, bdev->name,
8507 : spdk_get_thread());
8508 :
8509 268 : desc->bdev = bdev;
8510 268 : desc->thread = thread;
8511 268 : desc->write = write;
8512 :
8513 268 : spdk_spin_lock(&bdev->internal.spinlock);
8514 268 : if (bdev->internal.status == SPDK_BDEV_STATUS_UNREGISTERING ||
8515 268 : bdev->internal.status == SPDK_BDEV_STATUS_REMOVING) {
8516 3 : spdk_spin_unlock(&bdev->internal.spinlock);
8517 3 : return -ENODEV;
8518 : }
8519 :
8520 265 : if (write && bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE) {
8521 6 : LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
8522 6 : spdk_spin_unlock(&bdev->internal.spinlock);
8523 6 : return -EPERM;
8524 : }
8525 :
8526 259 : rc = bdev_start_qos(bdev);
8527 259 : if (rc != 0) {
8528 0 : SPDK_ERRLOG("Failed to start QoS on bdev %s\n", bdev->name);
8529 0 : spdk_spin_unlock(&bdev->internal.spinlock);
8530 0 : return rc;
8531 : }
8532 :
8533 259 : TAILQ_INSERT_TAIL(&bdev->internal.open_descs, desc, link);
8534 :
8535 259 : spdk_spin_unlock(&bdev->internal.spinlock);
8536 :
8537 259 : return 0;
8538 268 : }
8539 :
8540 : static void
8541 268 : bdev_open_opts_get_defaults(struct spdk_bdev_open_opts *opts, size_t opts_size)
8542 : {
8543 268 : if (!opts) {
8544 0 : SPDK_ERRLOG("opts should not be NULL.\n");
8545 0 : return;
8546 : }
8547 :
8548 268 : if (!opts_size) {
8549 0 : SPDK_ERRLOG("opts_size should not be zero.\n");
8550 0 : return;
8551 : }
8552 :
8553 268 : memset(opts, 0, opts_size);
8554 268 : opts->size = opts_size;
8555 :
8556 : #define FIELD_OK(field) \
8557 : offsetof(struct spdk_bdev_open_opts, field) + sizeof(opts->field) <= opts_size
8558 :
8559 : #define SET_FIELD(field, value) \
8560 : if (FIELD_OK(field)) { \
8561 : opts->field = value; \
8562 : } \
8563 :
8564 268 : SET_FIELD(no_metadata, false);
8565 :
8566 : #undef FIELD_OK
8567 : #undef SET_FIELD
8568 268 : }
8569 :
8570 : static void
8571 1 : bdev_open_opts_copy(struct spdk_bdev_open_opts *opts,
8572 : const struct spdk_bdev_open_opts *user_opts)
8573 : {
8574 : #define FIELD_OK(field) \
8575 : offsetof(struct spdk_bdev_open_opts, field) + sizeof(opts->field) <= user_opts->size
8576 :
8577 : #define SET_FIELD(field) \
8578 : if (FIELD_OK(field)) { \
8579 : opts->field = user_opts->field; \
8580 : } \
8581 :
8582 1 : SET_FIELD(no_metadata);
8583 :
8584 1 : opts->size = user_opts->size;
8585 :
8586 : /* We should not remove this statement, but need to update the assert statement
8587 : * if we add a new field, and also add a corresponding SET_FIELD statement.
8588 : */
8589 : SPDK_STATIC_ASSERT(sizeof(struct spdk_bdev_open_opts) == 16, "Incorrect size");
8590 :
8591 : #undef FIELD_OK
8592 : #undef SET_FIELD
8593 1 : }
8594 :
8595 : static int
8596 268 : bdev_desc_alloc(struct spdk_bdev *bdev, spdk_bdev_event_cb_t event_cb, void *event_ctx,
8597 : struct spdk_bdev_open_opts *user_opts, struct spdk_bdev_desc **_desc)
8598 : {
8599 : struct spdk_bdev_desc *desc;
8600 : struct spdk_bdev_open_opts opts;
8601 : unsigned int i;
8602 :
8603 268 : bdev_open_opts_get_defaults(&opts, sizeof(opts));
8604 268 : if (user_opts != NULL) {
8605 1 : bdev_open_opts_copy(&opts, user_opts);
8606 1 : }
8607 :
8608 268 : desc = calloc(1, sizeof(*desc));
8609 268 : if (desc == NULL) {
8610 0 : SPDK_ERRLOG("Failed to allocate memory for bdev descriptor\n");
8611 0 : return -ENOMEM;
8612 : }
8613 :
8614 268 : desc->opts = opts;
8615 :
8616 268 : TAILQ_INIT(&desc->pending_media_events);
8617 268 : TAILQ_INIT(&desc->free_media_events);
8618 :
8619 268 : desc->memory_domains_supported = spdk_bdev_get_memory_domains(bdev, NULL, 0) > 0;
8620 268 : desc->callback.event_fn = event_cb;
8621 268 : desc->callback.ctx = event_ctx;
8622 268 : spdk_spin_init(&desc->spinlock);
8623 :
8624 268 : if (desc->opts.no_metadata) {
8625 1 : if (spdk_bdev_is_md_separate(bdev)) {
8626 0 : SPDK_ERRLOG("no_metadata option is not supported with separate metadata.\n");
8627 0 : bdev_desc_free(desc);
8628 0 : return -EINVAL;
8629 : }
8630 1 : }
8631 :
8632 268 : if (bdev->media_events) {
8633 0 : desc->media_events_buffer = calloc(MEDIA_EVENT_POOL_SIZE,
8634 : sizeof(*desc->media_events_buffer));
8635 0 : if (desc->media_events_buffer == NULL) {
8636 0 : SPDK_ERRLOG("Failed to initialize media event pool\n");
8637 0 : bdev_desc_free(desc);
8638 0 : return -ENOMEM;
8639 : }
8640 :
8641 0 : for (i = 0; i < MEDIA_EVENT_POOL_SIZE; ++i) {
8642 0 : TAILQ_INSERT_TAIL(&desc->free_media_events,
8643 : &desc->media_events_buffer[i], tailq);
8644 0 : }
8645 0 : }
8646 :
8647 268 : if (bdev->fn_table->accel_sequence_supported != NULL) {
8648 0 : for (i = 0; i < SPDK_BDEV_NUM_IO_TYPES; ++i) {
8649 0 : desc->accel_sequence_supported[i] =
8650 0 : bdev->fn_table->accel_sequence_supported(bdev->ctxt,
8651 0 : (enum spdk_bdev_io_type)i);
8652 0 : }
8653 0 : }
8654 :
8655 268 : *_desc = desc;
8656 :
8657 268 : return 0;
8658 268 : }
8659 :
8660 : static int
8661 130 : bdev_open_ext(const char *bdev_name, bool write, spdk_bdev_event_cb_t event_cb,
8662 : void *event_ctx, struct spdk_bdev_open_opts *opts,
8663 : struct spdk_bdev_desc **_desc)
8664 : {
8665 : struct spdk_bdev_desc *desc;
8666 : struct spdk_bdev *bdev;
8667 : int rc;
8668 :
8669 130 : bdev = bdev_get_by_name(bdev_name);
8670 :
8671 130 : if (bdev == NULL) {
8672 1 : SPDK_NOTICELOG("Currently unable to find bdev with name: %s\n", bdev_name);
8673 1 : return -ENODEV;
8674 : }
8675 :
8676 129 : rc = bdev_desc_alloc(bdev, event_cb, event_ctx, opts, &desc);
8677 129 : if (rc != 0) {
8678 0 : return rc;
8679 : }
8680 :
8681 129 : rc = bdev_open(bdev, write, desc);
8682 129 : if (rc != 0) {
8683 7 : bdev_desc_free(desc);
8684 7 : desc = NULL;
8685 7 : }
8686 :
8687 129 : *_desc = desc;
8688 :
8689 129 : return rc;
8690 130 : }
8691 :
8692 : int
8693 132 : spdk_bdev_open_ext_v2(const char *bdev_name, bool write, spdk_bdev_event_cb_t event_cb,
8694 : void *event_ctx, struct spdk_bdev_open_opts *opts,
8695 : struct spdk_bdev_desc **_desc)
8696 : {
8697 : int rc;
8698 :
8699 132 : if (event_cb == NULL) {
8700 2 : SPDK_ERRLOG("Missing event callback function\n");
8701 2 : return -EINVAL;
8702 : }
8703 :
8704 130 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8705 130 : rc = bdev_open_ext(bdev_name, write, event_cb, event_ctx, opts, _desc);
8706 130 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8707 :
8708 130 : return rc;
8709 132 : }
8710 :
8711 : int
8712 130 : spdk_bdev_open_ext(const char *bdev_name, bool write, spdk_bdev_event_cb_t event_cb,
8713 : void *event_ctx, struct spdk_bdev_desc **_desc)
8714 : {
8715 130 : return spdk_bdev_open_ext_v2(bdev_name, write, event_cb, event_ctx, NULL, _desc);
8716 : }
8717 :
8718 : struct spdk_bdev_open_async_ctx {
8719 : char *bdev_name;
8720 : spdk_bdev_event_cb_t event_cb;
8721 : void *event_ctx;
8722 : bool write;
8723 : int rc;
8724 : spdk_bdev_open_async_cb_t cb_fn;
8725 : void *cb_arg;
8726 : struct spdk_bdev_desc *desc;
8727 : struct spdk_bdev_open_async_opts opts;
8728 : uint64_t start_ticks;
8729 : struct spdk_thread *orig_thread;
8730 : struct spdk_poller *poller;
8731 : TAILQ_ENTRY(spdk_bdev_open_async_ctx) tailq;
8732 : };
8733 :
8734 : static void
8735 0 : bdev_open_async_done(void *arg)
8736 : {
8737 0 : struct spdk_bdev_open_async_ctx *ctx = arg;
8738 :
8739 0 : ctx->cb_fn(ctx->desc, ctx->rc, ctx->cb_arg);
8740 :
8741 0 : free(ctx->bdev_name);
8742 0 : free(ctx);
8743 0 : }
8744 :
8745 : static void
8746 0 : bdev_open_async_cancel(void *arg)
8747 : {
8748 0 : struct spdk_bdev_open_async_ctx *ctx = arg;
8749 :
8750 0 : assert(ctx->rc == -ESHUTDOWN);
8751 :
8752 0 : spdk_poller_unregister(&ctx->poller);
8753 :
8754 0 : bdev_open_async_done(ctx);
8755 0 : }
8756 :
8757 : /* This is called when the bdev library finishes at shutdown. */
8758 : static void
8759 65 : bdev_open_async_fini(void)
8760 : {
8761 : struct spdk_bdev_open_async_ctx *ctx, *tmp_ctx;
8762 :
8763 65 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8764 65 : TAILQ_FOREACH_SAFE(ctx, &g_bdev_mgr.async_bdev_opens, tailq, tmp_ctx) {
8765 0 : TAILQ_REMOVE(&g_bdev_mgr.async_bdev_opens, ctx, tailq);
8766 : /*
8767 : * We have to move to ctx->orig_thread to unregister ctx->poller.
8768 : * However, there is a chance that ctx->poller is executed before
8769 : * message is executed, which could result in bdev_open_async_done()
8770 : * being called twice. To avoid such race condition, set ctx->rc to
8771 : * -ESHUTDOWN.
8772 : */
8773 0 : ctx->rc = -ESHUTDOWN;
8774 0 : spdk_thread_send_msg(ctx->orig_thread, bdev_open_async_cancel, ctx);
8775 0 : }
8776 65 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8777 65 : }
8778 :
8779 : static int bdev_open_async(void *arg);
8780 :
8781 : static void
8782 0 : _bdev_open_async(struct spdk_bdev_open_async_ctx *ctx)
8783 : {
8784 : uint64_t timeout_ticks;
8785 :
8786 0 : if (ctx->rc == -ESHUTDOWN) {
8787 : /* This context is being canceled. Do nothing. */
8788 0 : return;
8789 : }
8790 :
8791 0 : ctx->rc = bdev_open_ext(ctx->bdev_name, ctx->write, ctx->event_cb, ctx->event_ctx,
8792 0 : NULL, &ctx->desc);
8793 0 : if (ctx->rc == 0 || ctx->opts.timeout_ms == 0) {
8794 0 : goto exit;
8795 : }
8796 :
8797 0 : timeout_ticks = ctx->start_ticks + ctx->opts.timeout_ms * spdk_get_ticks_hz() / 1000ull;
8798 0 : if (spdk_get_ticks() >= timeout_ticks) {
8799 0 : SPDK_ERRLOG("Timed out while waiting for bdev '%s' to appear\n", ctx->bdev_name);
8800 0 : ctx->rc = -ETIMEDOUT;
8801 0 : goto exit;
8802 : }
8803 :
8804 0 : return;
8805 :
8806 : exit:
8807 0 : spdk_poller_unregister(&ctx->poller);
8808 0 : TAILQ_REMOVE(&g_bdev_mgr.async_bdev_opens, ctx, tailq);
8809 :
8810 : /* Completion callback is processed after stack unwinding. */
8811 0 : spdk_thread_send_msg(ctx->orig_thread, bdev_open_async_done, ctx);
8812 0 : }
8813 :
8814 : static int
8815 0 : bdev_open_async(void *arg)
8816 : {
8817 0 : struct spdk_bdev_open_async_ctx *ctx = arg;
8818 :
8819 0 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8820 :
8821 0 : _bdev_open_async(ctx);
8822 :
8823 0 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8824 :
8825 0 : return SPDK_POLLER_BUSY;
8826 : }
8827 :
8828 : static void
8829 0 : bdev_open_async_opts_copy(struct spdk_bdev_open_async_opts *opts,
8830 : struct spdk_bdev_open_async_opts *opts_src,
8831 : size_t size)
8832 : {
8833 0 : assert(opts);
8834 0 : assert(opts_src);
8835 :
8836 0 : opts->size = size;
8837 :
8838 : #define SET_FIELD(field) \
8839 : if (offsetof(struct spdk_bdev_open_async_opts, field) + sizeof(opts->field) <= size) { \
8840 : opts->field = opts_src->field; \
8841 : } \
8842 :
8843 0 : SET_FIELD(timeout_ms);
8844 :
8845 : /* Do not remove this statement, you should always update this statement when you adding a new field,
8846 : * and do not forget to add the SET_FIELD statement for your added field. */
8847 : SPDK_STATIC_ASSERT(sizeof(struct spdk_bdev_open_async_opts) == 16, "Incorrect size");
8848 :
8849 : #undef SET_FIELD
8850 0 : }
8851 :
8852 : static void
8853 0 : bdev_open_async_opts_get_default(struct spdk_bdev_open_async_opts *opts, size_t size)
8854 : {
8855 0 : assert(opts);
8856 :
8857 0 : opts->size = size;
8858 :
8859 : #define SET_FIELD(field, value) \
8860 : if (offsetof(struct spdk_bdev_open_async_opts, field) + sizeof(opts->field) <= size) { \
8861 : opts->field = value; \
8862 : } \
8863 :
8864 0 : SET_FIELD(timeout_ms, 0);
8865 :
8866 : #undef SET_FIELD
8867 0 : }
8868 :
8869 : int
8870 0 : spdk_bdev_open_async(const char *bdev_name, bool write, spdk_bdev_event_cb_t event_cb,
8871 : void *event_ctx, struct spdk_bdev_open_async_opts *opts,
8872 : spdk_bdev_open_async_cb_t open_cb, void *open_cb_arg)
8873 : {
8874 : struct spdk_bdev_open_async_ctx *ctx;
8875 :
8876 0 : if (event_cb == NULL) {
8877 0 : SPDK_ERRLOG("Missing event callback function\n");
8878 0 : return -EINVAL;
8879 : }
8880 :
8881 0 : if (open_cb == NULL) {
8882 0 : SPDK_ERRLOG("Missing open callback function\n");
8883 0 : return -EINVAL;
8884 : }
8885 :
8886 0 : if (opts != NULL && opts->size == 0) {
8887 0 : SPDK_ERRLOG("size in the options structure should not be zero\n");
8888 0 : return -EINVAL;
8889 : }
8890 :
8891 0 : ctx = calloc(1, sizeof(*ctx));
8892 0 : if (ctx == NULL) {
8893 0 : SPDK_ERRLOG("Failed to allocate open context\n");
8894 0 : return -ENOMEM;
8895 : }
8896 :
8897 0 : ctx->bdev_name = strdup(bdev_name);
8898 0 : if (ctx->bdev_name == NULL) {
8899 0 : SPDK_ERRLOG("Failed to duplicate bdev_name\n");
8900 0 : free(ctx);
8901 0 : return -ENOMEM;
8902 : }
8903 :
8904 0 : ctx->poller = SPDK_POLLER_REGISTER(bdev_open_async, ctx, 100 * 1000);
8905 0 : if (ctx->poller == NULL) {
8906 0 : SPDK_ERRLOG("Failed to register bdev_open_async poller\n");
8907 0 : free(ctx->bdev_name);
8908 0 : free(ctx);
8909 0 : return -ENOMEM;
8910 : }
8911 :
8912 0 : ctx->cb_fn = open_cb;
8913 0 : ctx->cb_arg = open_cb_arg;
8914 0 : ctx->write = write;
8915 0 : ctx->event_cb = event_cb;
8916 0 : ctx->event_ctx = event_ctx;
8917 0 : ctx->orig_thread = spdk_get_thread();
8918 0 : ctx->start_ticks = spdk_get_ticks();
8919 :
8920 0 : bdev_open_async_opts_get_default(&ctx->opts, sizeof(ctx->opts));
8921 0 : if (opts != NULL) {
8922 0 : bdev_open_async_opts_copy(&ctx->opts, opts, opts->size);
8923 0 : }
8924 :
8925 0 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8926 :
8927 0 : TAILQ_INSERT_TAIL(&g_bdev_mgr.async_bdev_opens, ctx, tailq);
8928 0 : _bdev_open_async(ctx);
8929 :
8930 0 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8931 :
8932 0 : return 0;
8933 0 : }
8934 :
8935 : static void
8936 259 : bdev_close(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc)
8937 : {
8938 : int rc;
8939 :
8940 259 : spdk_spin_lock(&bdev->internal.spinlock);
8941 259 : spdk_spin_lock(&desc->spinlock);
8942 :
8943 259 : TAILQ_REMOVE(&bdev->internal.open_descs, desc, link);
8944 :
8945 259 : desc->closed = true;
8946 :
8947 259 : if (desc->claim != NULL) {
8948 16 : bdev_desc_release_claims(desc);
8949 16 : }
8950 :
8951 259 : if (0 == desc->refs) {
8952 248 : spdk_spin_unlock(&desc->spinlock);
8953 248 : bdev_desc_free(desc);
8954 248 : } else {
8955 11 : spdk_spin_unlock(&desc->spinlock);
8956 : }
8957 :
8958 : /* If no more descriptors, kill QoS channel */
8959 259 : if (bdev->internal.qos && TAILQ_EMPTY(&bdev->internal.open_descs)) {
8960 7 : SPDK_DEBUGLOG(bdev, "Closed last descriptor for bdev %s on thread %p. Stopping QoS.\n",
8961 : bdev->name, spdk_get_thread());
8962 :
8963 7 : if (bdev_qos_destroy(bdev)) {
8964 : /* There isn't anything we can do to recover here. Just let the
8965 : * old QoS poller keep running. The QoS handling won't change
8966 : * cores when the user allocates a new channel, but it won't break. */
8967 0 : SPDK_ERRLOG("Unable to shut down QoS poller. It will continue running on the current thread.\n");
8968 0 : }
8969 7 : }
8970 :
8971 259 : if (bdev->internal.status == SPDK_BDEV_STATUS_REMOVING && TAILQ_EMPTY(&bdev->internal.open_descs)) {
8972 15 : rc = bdev_unregister_unsafe(bdev);
8973 15 : spdk_spin_unlock(&bdev->internal.spinlock);
8974 :
8975 15 : if (rc == 0) {
8976 15 : spdk_io_device_unregister(__bdev_to_io_dev(bdev), bdev_destroy_cb);
8977 15 : }
8978 15 : } else {
8979 244 : spdk_spin_unlock(&bdev->internal.spinlock);
8980 : }
8981 259 : }
8982 :
8983 : void
8984 122 : spdk_bdev_close(struct spdk_bdev_desc *desc)
8985 : {
8986 122 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
8987 :
8988 122 : SPDK_DEBUGLOG(bdev, "Closing descriptor %p for bdev %s on thread %p\n", desc, bdev->name,
8989 : spdk_get_thread());
8990 :
8991 122 : assert(desc->thread == spdk_get_thread());
8992 :
8993 122 : spdk_poller_unregister(&desc->io_timeout_poller);
8994 :
8995 122 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8996 :
8997 122 : bdev_close(bdev, desc);
8998 :
8999 122 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
9000 122 : }
9001 :
9002 : int32_t
9003 3 : spdk_bdev_get_numa_id(struct spdk_bdev *bdev)
9004 : {
9005 3 : if (bdev->numa.id_valid) {
9006 2 : return bdev->numa.id;
9007 : } else {
9008 1 : return SPDK_ENV_NUMA_ID_ANY;
9009 : }
9010 3 : }
9011 :
9012 : static void
9013 126 : bdev_register_finished(void *arg)
9014 : {
9015 126 : struct spdk_bdev_desc *desc = arg;
9016 126 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
9017 :
9018 126 : spdk_notify_send("bdev_register", spdk_bdev_get_name(bdev));
9019 :
9020 126 : spdk_spin_lock(&g_bdev_mgr.spinlock);
9021 :
9022 126 : bdev_close(bdev, desc);
9023 :
9024 126 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
9025 126 : }
9026 :
9027 : int
9028 129 : spdk_bdev_register(struct spdk_bdev *bdev)
9029 : {
9030 : struct spdk_bdev_desc *desc;
9031 129 : struct spdk_thread *thread = spdk_get_thread();
9032 : int rc;
9033 :
9034 129 : if (spdk_unlikely(!spdk_thread_is_app_thread(NULL))) {
9035 1 : SPDK_ERRLOG("Cannot register bdev %s on thread %p (%s)\n", bdev->name, thread,
9036 : thread ? spdk_thread_get_name(thread) : "null");
9037 1 : return -EINVAL;
9038 : }
9039 :
9040 128 : rc = bdev_register(bdev);
9041 128 : if (rc != 0) {
9042 2 : return rc;
9043 : }
9044 :
9045 : /* A descriptor is opened to prevent bdev deletion during examination */
9046 126 : rc = bdev_desc_alloc(bdev, _tmp_bdev_event_cb, NULL, NULL, &desc);
9047 126 : if (rc != 0) {
9048 0 : spdk_bdev_unregister(bdev, NULL, NULL);
9049 0 : return rc;
9050 : }
9051 :
9052 126 : rc = bdev_open(bdev, false, desc);
9053 126 : if (rc != 0) {
9054 0 : bdev_desc_free(desc);
9055 0 : spdk_bdev_unregister(bdev, NULL, NULL);
9056 0 : return rc;
9057 : }
9058 :
9059 : /* Examine configuration before initializing I/O */
9060 126 : bdev_examine(bdev);
9061 :
9062 126 : rc = spdk_bdev_wait_for_examine(bdev_register_finished, desc);
9063 126 : if (rc != 0) {
9064 0 : bdev_close(bdev, desc);
9065 0 : spdk_bdev_unregister(bdev, NULL, NULL);
9066 0 : }
9067 :
9068 126 : return rc;
9069 129 : }
9070 :
9071 : int
9072 26 : spdk_bdev_module_claim_bdev(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc,
9073 : struct spdk_bdev_module *module)
9074 : {
9075 26 : spdk_spin_lock(&bdev->internal.spinlock);
9076 :
9077 26 : if (bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE) {
9078 6 : LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
9079 6 : spdk_spin_unlock(&bdev->internal.spinlock);
9080 6 : return -EPERM;
9081 : }
9082 :
9083 20 : if (desc && !desc->write) {
9084 5 : desc->write = true;
9085 5 : }
9086 :
9087 20 : bdev->internal.claim_type = SPDK_BDEV_CLAIM_EXCL_WRITE;
9088 20 : bdev->internal.claim.v1.module = module;
9089 :
9090 20 : spdk_spin_unlock(&bdev->internal.spinlock);
9091 20 : return 0;
9092 26 : }
9093 :
9094 : void
9095 8 : spdk_bdev_module_release_bdev(struct spdk_bdev *bdev)
9096 : {
9097 8 : spdk_spin_lock(&bdev->internal.spinlock);
9098 :
9099 8 : assert(bdev->internal.claim.v1.module != NULL);
9100 8 : assert(bdev->internal.claim_type == SPDK_BDEV_CLAIM_EXCL_WRITE);
9101 8 : bdev->internal.claim_type = SPDK_BDEV_CLAIM_NONE;
9102 8 : bdev->internal.claim.v1.module = NULL;
9103 :
9104 8 : spdk_spin_unlock(&bdev->internal.spinlock);
9105 8 : }
9106 :
9107 : /*
9108 : * Start claims v2
9109 : */
9110 :
9111 : const char *
9112 24 : spdk_bdev_claim_get_name(enum spdk_bdev_claim_type type)
9113 : {
9114 24 : switch (type) {
9115 : case SPDK_BDEV_CLAIM_NONE:
9116 0 : return "not_claimed";
9117 : case SPDK_BDEV_CLAIM_EXCL_WRITE:
9118 6 : return "exclusive_write";
9119 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE:
9120 7 : return "read_many_write_one";
9121 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_NONE:
9122 5 : return "read_many_write_none";
9123 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
9124 6 : return "read_many_write_many";
9125 : default:
9126 0 : break;
9127 : }
9128 0 : return "invalid_claim";
9129 24 : }
9130 :
9131 : static bool
9132 96 : claim_type_is_v2(enum spdk_bdev_claim_type type)
9133 : {
9134 96 : switch (type) {
9135 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE:
9136 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_NONE:
9137 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
9138 96 : return true;
9139 : default:
9140 0 : break;
9141 : }
9142 0 : return false;
9143 96 : }
9144 :
9145 : /* Returns true if taking a claim with desc->write == false should make the descriptor writable. */
9146 : static bool
9147 13 : claim_type_promotes_to_write(enum spdk_bdev_claim_type type)
9148 : {
9149 13 : switch (type) {
9150 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE:
9151 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
9152 5 : return true;
9153 : default:
9154 8 : break;
9155 : }
9156 8 : return false;
9157 13 : }
9158 :
9159 : void
9160 44 : spdk_bdev_claim_opts_init(struct spdk_bdev_claim_opts *opts, size_t size)
9161 : {
9162 44 : if (opts == NULL) {
9163 0 : SPDK_ERRLOG("opts should not be NULL\n");
9164 0 : assert(opts != NULL);
9165 0 : return;
9166 : }
9167 44 : if (size == 0) {
9168 0 : SPDK_ERRLOG("size should not be zero\n");
9169 0 : assert(size != 0);
9170 0 : return;
9171 : }
9172 :
9173 44 : memset(opts, 0, size);
9174 44 : opts->opts_size = size;
9175 :
9176 : #define FIELD_OK(field) \
9177 : offsetof(struct spdk_bdev_claim_opts, field) + sizeof(opts->field) <= size
9178 :
9179 : #define SET_FIELD(field, value) \
9180 : if (FIELD_OK(field)) { \
9181 : opts->field = value; \
9182 : } \
9183 :
9184 44 : SET_FIELD(shared_claim_key, 0);
9185 :
9186 : #undef FIELD_OK
9187 : #undef SET_FIELD
9188 44 : }
9189 :
9190 : static int
9191 22 : claim_opts_copy(struct spdk_bdev_claim_opts *src, struct spdk_bdev_claim_opts *dst)
9192 : {
9193 22 : if (src->opts_size == 0) {
9194 0 : SPDK_ERRLOG("size should not be zero\n");
9195 0 : return -1;
9196 : }
9197 :
9198 22 : memset(dst, 0, sizeof(*dst));
9199 22 : dst->opts_size = src->opts_size;
9200 :
9201 : #define FIELD_OK(field) \
9202 : offsetof(struct spdk_bdev_claim_opts, field) + sizeof(src->field) <= src->opts_size
9203 :
9204 : #define SET_FIELD(field) \
9205 : if (FIELD_OK(field)) { \
9206 : dst->field = src->field; \
9207 : } \
9208 :
9209 22 : if (FIELD_OK(name)) {
9210 22 : snprintf(dst->name, sizeof(dst->name), "%s", src->name);
9211 22 : }
9212 :
9213 22 : SET_FIELD(shared_claim_key);
9214 :
9215 : /* You should not remove this statement, but need to update the assert statement
9216 : * if you add a new field, and also add a corresponding SET_FIELD statement */
9217 : SPDK_STATIC_ASSERT(sizeof(struct spdk_bdev_claim_opts) == 48, "Incorrect size");
9218 :
9219 : #undef FIELD_OK
9220 : #undef SET_FIELD
9221 22 : return 0;
9222 22 : }
9223 :
9224 : /* Returns 0 if a read-write-once claim can be taken. */
9225 : static int
9226 9 : claim_verify_rwo(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
9227 : struct spdk_bdev_claim_opts *opts, struct spdk_bdev_module *module)
9228 : {
9229 9 : struct spdk_bdev *bdev = desc->bdev;
9230 : struct spdk_bdev_desc *open_desc;
9231 :
9232 9 : assert(spdk_spin_held(&bdev->internal.spinlock));
9233 9 : assert(type == SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE);
9234 :
9235 9 : if (opts->shared_claim_key != 0) {
9236 1 : SPDK_ERRLOG("%s: key option not supported with read-write-once claims\n",
9237 : bdev->name);
9238 1 : return -EINVAL;
9239 : }
9240 8 : if (bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE) {
9241 1 : LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
9242 1 : return -EPERM;
9243 : }
9244 7 : if (desc->claim != NULL) {
9245 0 : SPDK_NOTICELOG("%s: descriptor already claimed bdev with module %s\n",
9246 : bdev->name, desc->claim->module->name);
9247 0 : return -EPERM;
9248 : }
9249 14 : TAILQ_FOREACH(open_desc, &bdev->internal.open_descs, link) {
9250 9 : if (desc != open_desc && open_desc->write) {
9251 2 : SPDK_NOTICELOG("%s: Cannot obtain read-write-once claim while "
9252 : "another descriptor is open for writing\n",
9253 : bdev->name);
9254 2 : return -EPERM;
9255 : }
9256 7 : }
9257 :
9258 5 : return 0;
9259 9 : }
9260 :
9261 : /* Returns 0 if a read-only-many claim can be taken. */
9262 : static int
9263 12 : claim_verify_rom(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
9264 : struct spdk_bdev_claim_opts *opts, struct spdk_bdev_module *module)
9265 : {
9266 12 : struct spdk_bdev *bdev = desc->bdev;
9267 : struct spdk_bdev_desc *open_desc;
9268 :
9269 12 : assert(spdk_spin_held(&bdev->internal.spinlock));
9270 12 : assert(type == SPDK_BDEV_CLAIM_READ_MANY_WRITE_NONE);
9271 12 : assert(desc->claim == NULL);
9272 :
9273 12 : if (desc->write) {
9274 3 : SPDK_ERRLOG("%s: Cannot obtain read-only-many claim with writable descriptor\n",
9275 : bdev->name);
9276 3 : return -EINVAL;
9277 : }
9278 9 : if (opts->shared_claim_key != 0) {
9279 1 : SPDK_ERRLOG("%s: key option not supported with read-only-may claims\n", bdev->name);
9280 1 : return -EINVAL;
9281 : }
9282 8 : if (bdev->internal.claim_type == SPDK_BDEV_CLAIM_NONE) {
9283 15 : TAILQ_FOREACH(open_desc, &bdev->internal.open_descs, link) {
9284 9 : if (open_desc->write) {
9285 0 : SPDK_NOTICELOG("%s: Cannot obtain read-only-many claim while "
9286 : "another descriptor is open for writing\n",
9287 : bdev->name);
9288 0 : return -EPERM;
9289 : }
9290 9 : }
9291 6 : }
9292 :
9293 8 : return 0;
9294 12 : }
9295 :
9296 : /* Returns 0 if a read-write-many claim can be taken. */
9297 : static int
9298 8 : claim_verify_rwm(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
9299 : struct spdk_bdev_claim_opts *opts, struct spdk_bdev_module *module)
9300 : {
9301 8 : struct spdk_bdev *bdev = desc->bdev;
9302 : struct spdk_bdev_desc *open_desc;
9303 :
9304 8 : assert(spdk_spin_held(&bdev->internal.spinlock));
9305 8 : assert(type == SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED);
9306 8 : assert(desc->claim == NULL);
9307 :
9308 8 : if (opts->shared_claim_key == 0) {
9309 2 : SPDK_ERRLOG("%s: shared_claim_key option required with read-write-may claims\n",
9310 : bdev->name);
9311 2 : return -EINVAL;
9312 : }
9313 6 : switch (bdev->internal.claim_type) {
9314 : case SPDK_BDEV_CLAIM_NONE:
9315 7 : TAILQ_FOREACH(open_desc, &bdev->internal.open_descs, link) {
9316 5 : if (open_desc == desc) {
9317 3 : continue;
9318 : }
9319 2 : if (open_desc->write) {
9320 2 : SPDK_NOTICELOG("%s: Cannot obtain read-write-many claim while "
9321 : "another descriptor is open for writing without a "
9322 : "claim\n", bdev->name);
9323 2 : return -EPERM;
9324 : }
9325 0 : }
9326 2 : break;
9327 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
9328 2 : if (opts->shared_claim_key != bdev->internal.claim.v2.key) {
9329 1 : LOG_ALREADY_CLAIMED_ERROR("already claimed with another key", bdev);
9330 1 : return -EPERM;
9331 : }
9332 1 : break;
9333 : default:
9334 0 : LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
9335 0 : return -EBUSY;
9336 : }
9337 :
9338 3 : return 0;
9339 8 : }
9340 :
9341 : /* Updates desc and its bdev with a v2 claim. */
9342 : static int
9343 16 : claim_bdev(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
9344 : struct spdk_bdev_claim_opts *opts, struct spdk_bdev_module *module)
9345 : {
9346 16 : struct spdk_bdev *bdev = desc->bdev;
9347 : struct spdk_bdev_module_claim *claim;
9348 :
9349 16 : assert(spdk_spin_held(&bdev->internal.spinlock));
9350 16 : assert(claim_type_is_v2(type));
9351 16 : assert(desc->claim == NULL);
9352 :
9353 16 : claim = calloc(1, sizeof(*desc->claim));
9354 16 : if (claim == NULL) {
9355 0 : SPDK_ERRLOG("%s: out of memory while allocating claim\n", bdev->name);
9356 0 : return -ENOMEM;
9357 : }
9358 16 : claim->module = module;
9359 16 : claim->desc = desc;
9360 : SPDK_STATIC_ASSERT(sizeof(claim->name) == sizeof(opts->name), "sizes must match");
9361 16 : memcpy(claim->name, opts->name, sizeof(claim->name));
9362 16 : desc->claim = claim;
9363 :
9364 16 : if (bdev->internal.claim_type == SPDK_BDEV_CLAIM_NONE) {
9365 13 : bdev->internal.claim_type = type;
9366 13 : TAILQ_INIT(&bdev->internal.claim.v2.claims);
9367 13 : bdev->internal.claim.v2.key = opts->shared_claim_key;
9368 13 : }
9369 16 : assert(type == bdev->internal.claim_type);
9370 :
9371 16 : TAILQ_INSERT_TAIL(&bdev->internal.claim.v2.claims, claim, link);
9372 :
9373 16 : if (!desc->write && claim_type_promotes_to_write(type)) {
9374 5 : desc->write = true;
9375 5 : }
9376 :
9377 16 : return 0;
9378 16 : }
9379 :
9380 : int
9381 39 : spdk_bdev_module_claim_bdev_desc(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
9382 : struct spdk_bdev_claim_opts *_opts,
9383 : struct spdk_bdev_module *module)
9384 : {
9385 : struct spdk_bdev *bdev;
9386 : struct spdk_bdev_claim_opts opts;
9387 39 : int rc = 0;
9388 :
9389 39 : if (desc == NULL) {
9390 0 : SPDK_ERRLOG("descriptor must not be NULL\n");
9391 0 : return -EINVAL;
9392 : }
9393 :
9394 39 : bdev = desc->bdev;
9395 :
9396 39 : if (_opts == NULL) {
9397 17 : spdk_bdev_claim_opts_init(&opts, sizeof(opts));
9398 39 : } else if (claim_opts_copy(_opts, &opts) != 0) {
9399 0 : return -EINVAL;
9400 : }
9401 :
9402 39 : spdk_spin_lock(&bdev->internal.spinlock);
9403 :
9404 39 : if (bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE &&
9405 15 : bdev->internal.claim_type != type) {
9406 10 : LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
9407 10 : spdk_spin_unlock(&bdev->internal.spinlock);
9408 10 : return -EPERM;
9409 : }
9410 :
9411 29 : if (claim_type_is_v2(type) && desc->claim != NULL) {
9412 0 : SPDK_ERRLOG("%s: descriptor already has %s claim with name '%s'\n",
9413 : bdev->name, spdk_bdev_claim_get_name(type), desc->claim->name);
9414 0 : spdk_spin_unlock(&bdev->internal.spinlock);
9415 0 : return -EPERM;
9416 : }
9417 :
9418 29 : switch (type) {
9419 : case SPDK_BDEV_CLAIM_EXCL_WRITE:
9420 0 : spdk_spin_unlock(&bdev->internal.spinlock);
9421 0 : return spdk_bdev_module_claim_bdev(bdev, desc, module);
9422 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE:
9423 9 : rc = claim_verify_rwo(desc, type, &opts, module);
9424 9 : break;
9425 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_NONE:
9426 12 : rc = claim_verify_rom(desc, type, &opts, module);
9427 12 : break;
9428 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
9429 8 : rc = claim_verify_rwm(desc, type, &opts, module);
9430 8 : break;
9431 : default:
9432 0 : SPDK_ERRLOG("%s: claim type %d not supported\n", bdev->name, type);
9433 0 : rc = -ENOTSUP;
9434 0 : }
9435 :
9436 29 : if (rc == 0) {
9437 16 : rc = claim_bdev(desc, type, &opts, module);
9438 16 : }
9439 :
9440 29 : spdk_spin_unlock(&bdev->internal.spinlock);
9441 29 : return rc;
9442 39 : }
9443 :
9444 : static void
9445 13 : claim_reset(struct spdk_bdev *bdev)
9446 : {
9447 13 : assert(spdk_spin_held(&bdev->internal.spinlock));
9448 13 : assert(claim_type_is_v2(bdev->internal.claim_type));
9449 13 : assert(TAILQ_EMPTY(&bdev->internal.claim.v2.claims));
9450 :
9451 13 : memset(&bdev->internal.claim, 0, sizeof(bdev->internal.claim));
9452 13 : bdev->internal.claim_type = SPDK_BDEV_CLAIM_NONE;
9453 13 : }
9454 :
9455 : static void
9456 16 : bdev_desc_release_claims(struct spdk_bdev_desc *desc)
9457 : {
9458 16 : struct spdk_bdev *bdev = desc->bdev;
9459 :
9460 16 : assert(spdk_spin_held(&bdev->internal.spinlock));
9461 16 : assert(claim_type_is_v2(bdev->internal.claim_type));
9462 :
9463 16 : if (bdev->internal.examine_in_progress == 0) {
9464 16 : TAILQ_REMOVE(&bdev->internal.claim.v2.claims, desc->claim, link);
9465 16 : free(desc->claim);
9466 16 : if (TAILQ_EMPTY(&bdev->internal.claim.v2.claims)) {
9467 13 : claim_reset(bdev);
9468 13 : }
9469 16 : } else {
9470 : /* This is a dead claim that will be cleaned up when bdev_examine() is done. */
9471 0 : desc->claim->module = NULL;
9472 0 : desc->claim->desc = NULL;
9473 : }
9474 16 : desc->claim = NULL;
9475 16 : }
9476 :
9477 : /*
9478 : * End claims v2
9479 : */
9480 :
9481 : struct spdk_bdev *
9482 1541 : spdk_bdev_desc_get_bdev(struct spdk_bdev_desc *desc)
9483 : {
9484 1541 : assert(desc != NULL);
9485 1541 : return desc->bdev;
9486 : }
9487 :
9488 : int
9489 1 : spdk_for_each_bdev(void *ctx, spdk_for_each_bdev_fn fn)
9490 : {
9491 : struct spdk_bdev *bdev, *tmp;
9492 : struct spdk_bdev_desc *desc;
9493 1 : int rc = 0;
9494 :
9495 1 : assert(fn != NULL);
9496 :
9497 1 : spdk_spin_lock(&g_bdev_mgr.spinlock);
9498 1 : bdev = spdk_bdev_first();
9499 9 : while (bdev != NULL) {
9500 8 : rc = bdev_desc_alloc(bdev, _tmp_bdev_event_cb, NULL, NULL, &desc);
9501 8 : if (rc != 0) {
9502 0 : break;
9503 : }
9504 8 : rc = bdev_open(bdev, false, desc);
9505 8 : if (rc != 0) {
9506 1 : bdev_desc_free(desc);
9507 1 : if (rc == -ENODEV) {
9508 : /* Ignore the error and move to the next bdev. */
9509 1 : rc = 0;
9510 1 : bdev = spdk_bdev_next(bdev);
9511 1 : continue;
9512 : }
9513 0 : break;
9514 : }
9515 7 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
9516 :
9517 7 : rc = fn(ctx, bdev);
9518 :
9519 7 : spdk_spin_lock(&g_bdev_mgr.spinlock);
9520 7 : tmp = spdk_bdev_next(bdev);
9521 7 : bdev_close(bdev, desc);
9522 7 : if (rc != 0) {
9523 0 : break;
9524 : }
9525 7 : bdev = tmp;
9526 : }
9527 1 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
9528 :
9529 1 : return rc;
9530 : }
9531 :
9532 : int
9533 1 : spdk_for_each_bdev_leaf(void *ctx, spdk_for_each_bdev_fn fn)
9534 : {
9535 : struct spdk_bdev *bdev, *tmp;
9536 : struct spdk_bdev_desc *desc;
9537 1 : int rc = 0;
9538 :
9539 1 : assert(fn != NULL);
9540 :
9541 1 : spdk_spin_lock(&g_bdev_mgr.spinlock);
9542 1 : bdev = spdk_bdev_first_leaf();
9543 6 : while (bdev != NULL) {
9544 5 : rc = bdev_desc_alloc(bdev, _tmp_bdev_event_cb, NULL, NULL, &desc);
9545 5 : if (rc != 0) {
9546 0 : break;
9547 : }
9548 5 : rc = bdev_open(bdev, false, desc);
9549 5 : if (rc != 0) {
9550 1 : bdev_desc_free(desc);
9551 1 : if (rc == -ENODEV) {
9552 : /* Ignore the error and move to the next bdev. */
9553 1 : rc = 0;
9554 1 : bdev = spdk_bdev_next_leaf(bdev);
9555 1 : continue;
9556 : }
9557 0 : break;
9558 : }
9559 4 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
9560 :
9561 4 : rc = fn(ctx, bdev);
9562 :
9563 4 : spdk_spin_lock(&g_bdev_mgr.spinlock);
9564 4 : tmp = spdk_bdev_next_leaf(bdev);
9565 4 : bdev_close(bdev, desc);
9566 4 : if (rc != 0) {
9567 0 : break;
9568 : }
9569 4 : bdev = tmp;
9570 : }
9571 1 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
9572 :
9573 1 : return rc;
9574 : }
9575 :
9576 : void
9577 0 : spdk_bdev_io_get_iovec(struct spdk_bdev_io *bdev_io, struct iovec **iovp, int *iovcntp)
9578 : {
9579 : struct iovec *iovs;
9580 : int iovcnt;
9581 :
9582 0 : if (bdev_io == NULL) {
9583 0 : return;
9584 : }
9585 :
9586 0 : switch (bdev_io->type) {
9587 : case SPDK_BDEV_IO_TYPE_READ:
9588 : case SPDK_BDEV_IO_TYPE_WRITE:
9589 : case SPDK_BDEV_IO_TYPE_ZCOPY:
9590 0 : iovs = bdev_io->u.bdev.iovs;
9591 0 : iovcnt = bdev_io->u.bdev.iovcnt;
9592 0 : break;
9593 : default:
9594 0 : iovs = NULL;
9595 0 : iovcnt = 0;
9596 0 : break;
9597 : }
9598 :
9599 0 : if (iovp) {
9600 0 : *iovp = iovs;
9601 0 : }
9602 0 : if (iovcntp) {
9603 0 : *iovcntp = iovcnt;
9604 0 : }
9605 0 : }
9606 :
9607 : void *
9608 0 : spdk_bdev_io_get_md_buf(struct spdk_bdev_io *bdev_io)
9609 : {
9610 0 : if (bdev_io == NULL) {
9611 0 : return NULL;
9612 : }
9613 :
9614 0 : if (!spdk_bdev_is_md_separate(bdev_io->bdev)) {
9615 0 : return NULL;
9616 : }
9617 :
9618 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_READ ||
9619 0 : bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
9620 0 : return bdev_io->u.bdev.md_buf;
9621 : }
9622 :
9623 0 : return NULL;
9624 0 : }
9625 :
9626 : void *
9627 0 : spdk_bdev_io_get_cb_arg(struct spdk_bdev_io *bdev_io)
9628 : {
9629 0 : if (bdev_io == NULL) {
9630 0 : assert(false);
9631 : return NULL;
9632 : }
9633 :
9634 0 : return bdev_io->internal.caller_ctx;
9635 : }
9636 :
9637 : void
9638 7 : spdk_bdev_module_list_add(struct spdk_bdev_module *bdev_module)
9639 : {
9640 :
9641 7 : if (spdk_bdev_module_list_find(bdev_module->name)) {
9642 0 : SPDK_ERRLOG("ERROR: module '%s' already registered.\n", bdev_module->name);
9643 0 : assert(false);
9644 : }
9645 :
9646 7 : spdk_spin_init(&bdev_module->internal.spinlock);
9647 7 : TAILQ_INIT(&bdev_module->internal.quiesced_ranges);
9648 :
9649 : /*
9650 : * Modules with examine callbacks must be initialized first, so they are
9651 : * ready to handle examine callbacks from later modules that will
9652 : * register physical bdevs.
9653 : */
9654 7 : if (bdev_module->examine_config != NULL || bdev_module->examine_disk != NULL) {
9655 4 : TAILQ_INSERT_HEAD(&g_bdev_mgr.bdev_modules, bdev_module, internal.tailq);
9656 4 : } else {
9657 3 : TAILQ_INSERT_TAIL(&g_bdev_mgr.bdev_modules, bdev_module, internal.tailq);
9658 : }
9659 7 : }
9660 :
9661 : struct spdk_bdev_module *
9662 7 : spdk_bdev_module_list_find(const char *name)
9663 : {
9664 : struct spdk_bdev_module *bdev_module;
9665 :
9666 14 : TAILQ_FOREACH(bdev_module, &g_bdev_mgr.bdev_modules, internal.tailq) {
9667 7 : if (strcmp(name, bdev_module->name) == 0) {
9668 0 : break;
9669 : }
9670 7 : }
9671 :
9672 7 : return bdev_module;
9673 : }
9674 :
9675 : static int
9676 6 : bdev_write_zero_buffer(struct spdk_bdev_io *bdev_io)
9677 : {
9678 : uint64_t num_blocks;
9679 6 : void *md_buf = NULL;
9680 :
9681 6 : num_blocks = bdev_io->u.bdev.num_blocks;
9682 :
9683 6 : if (spdk_bdev_is_md_separate(bdev_io->bdev)) {
9684 4 : md_buf = (char *)g_bdev_mgr.zero_buffer +
9685 2 : spdk_bdev_get_block_size(bdev_io->bdev) * num_blocks;
9686 2 : }
9687 :
9688 12 : return bdev_write_blocks_with_md(bdev_io->internal.desc,
9689 6 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
9690 6 : g_bdev_mgr.zero_buffer, md_buf,
9691 6 : bdev_io->u.bdev.offset_blocks, num_blocks,
9692 6 : bdev_write_zero_buffer_done, bdev_io);
9693 : }
9694 :
9695 : static void
9696 6 : bdev_write_zero_buffer_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
9697 : {
9698 6 : struct spdk_bdev_io *parent_io = cb_arg;
9699 :
9700 6 : spdk_bdev_free_io(bdev_io);
9701 :
9702 6 : parent_io->internal.status = success ? SPDK_BDEV_IO_STATUS_SUCCESS : SPDK_BDEV_IO_STATUS_FAILED;
9703 6 : parent_io->internal.cb(parent_io, success, parent_io->internal.caller_ctx);
9704 6 : }
9705 :
9706 : static void
9707 10 : bdev_set_qos_limit_done(struct set_qos_limit_ctx *ctx, int status)
9708 : {
9709 10 : spdk_spin_lock(&ctx->bdev->internal.spinlock);
9710 10 : ctx->bdev->internal.qos_mod_in_progress = false;
9711 10 : spdk_spin_unlock(&ctx->bdev->internal.spinlock);
9712 :
9713 10 : if (ctx->cb_fn) {
9714 8 : ctx->cb_fn(ctx->cb_arg, status);
9715 8 : }
9716 10 : free(ctx);
9717 10 : }
9718 :
9719 : static void
9720 2 : bdev_disable_qos_done(void *cb_arg)
9721 : {
9722 2 : struct set_qos_limit_ctx *ctx = cb_arg;
9723 2 : struct spdk_bdev *bdev = ctx->bdev;
9724 : struct spdk_bdev_qos *qos;
9725 :
9726 2 : spdk_spin_lock(&bdev->internal.spinlock);
9727 2 : qos = bdev->internal.qos;
9728 2 : bdev->internal.qos = NULL;
9729 2 : spdk_spin_unlock(&bdev->internal.spinlock);
9730 :
9731 2 : if (qos->thread != NULL) {
9732 2 : spdk_put_io_channel(spdk_io_channel_from_ctx(qos->ch));
9733 2 : spdk_poller_unregister(&qos->poller);
9734 2 : }
9735 :
9736 2 : free(qos);
9737 :
9738 2 : bdev_set_qos_limit_done(ctx, 0);
9739 2 : }
9740 :
9741 : static void
9742 2 : bdev_disable_qos_msg_done(struct spdk_bdev *bdev, void *_ctx, int status)
9743 : {
9744 2 : struct set_qos_limit_ctx *ctx = _ctx;
9745 : struct spdk_thread *thread;
9746 :
9747 2 : spdk_spin_lock(&bdev->internal.spinlock);
9748 2 : thread = bdev->internal.qos->thread;
9749 2 : spdk_spin_unlock(&bdev->internal.spinlock);
9750 :
9751 2 : if (thread != NULL) {
9752 2 : spdk_thread_send_msg(thread, bdev_disable_qos_done, ctx);
9753 2 : } else {
9754 0 : bdev_disable_qos_done(ctx);
9755 : }
9756 2 : }
9757 :
9758 : static void
9759 4 : bdev_disable_qos_msg(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
9760 : struct spdk_io_channel *ch, void *_ctx)
9761 : {
9762 4 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(ch);
9763 : struct spdk_bdev_io *bdev_io;
9764 :
9765 4 : bdev_ch->flags &= ~BDEV_CH_QOS_ENABLED;
9766 :
9767 6 : while (!TAILQ_EMPTY(&bdev_ch->qos_queued_io)) {
9768 : /* Re-submit the queued I/O. */
9769 2 : bdev_io = TAILQ_FIRST(&bdev_ch->qos_queued_io);
9770 2 : TAILQ_REMOVE(&bdev_ch->qos_queued_io, bdev_io, internal.link);
9771 2 : _bdev_io_submit(bdev_io);
9772 : }
9773 :
9774 4 : spdk_bdev_for_each_channel_continue(i, 0);
9775 4 : }
9776 :
9777 : static void
9778 1 : bdev_update_qos_rate_limit_msg(void *cb_arg)
9779 : {
9780 1 : struct set_qos_limit_ctx *ctx = cb_arg;
9781 1 : struct spdk_bdev *bdev = ctx->bdev;
9782 :
9783 1 : spdk_spin_lock(&bdev->internal.spinlock);
9784 1 : bdev_qos_update_max_quota_per_timeslice(bdev->internal.qos);
9785 1 : spdk_spin_unlock(&bdev->internal.spinlock);
9786 :
9787 1 : bdev_set_qos_limit_done(ctx, 0);
9788 1 : }
9789 :
9790 : static void
9791 9 : bdev_enable_qos_msg(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
9792 : struct spdk_io_channel *ch, void *_ctx)
9793 : {
9794 9 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(ch);
9795 :
9796 9 : spdk_spin_lock(&bdev->internal.spinlock);
9797 9 : bdev_enable_qos(bdev, bdev_ch);
9798 9 : spdk_spin_unlock(&bdev->internal.spinlock);
9799 9 : spdk_bdev_for_each_channel_continue(i, 0);
9800 9 : }
9801 :
9802 : static void
9803 6 : bdev_enable_qos_done(struct spdk_bdev *bdev, void *_ctx, int status)
9804 : {
9805 6 : struct set_qos_limit_ctx *ctx = _ctx;
9806 :
9807 6 : bdev_set_qos_limit_done(ctx, status);
9808 6 : }
9809 :
9810 : static void
9811 7 : bdev_set_qos_rate_limits(struct spdk_bdev *bdev, uint64_t *limits)
9812 : {
9813 : int i;
9814 :
9815 7 : assert(bdev->internal.qos != NULL);
9816 :
9817 35 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
9818 28 : if (limits[i] != SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
9819 28 : bdev->internal.qos->rate_limits[i].limit = limits[i];
9820 :
9821 28 : if (limits[i] == 0) {
9822 19 : bdev->internal.qos->rate_limits[i].limit =
9823 : SPDK_BDEV_QOS_LIMIT_NOT_DEFINED;
9824 19 : }
9825 28 : }
9826 28 : }
9827 7 : }
9828 :
9829 : void
9830 9 : spdk_bdev_set_qos_rate_limits(struct spdk_bdev *bdev, uint64_t *limits,
9831 : void (*cb_fn)(void *cb_arg, int status), void *cb_arg)
9832 : {
9833 : struct set_qos_limit_ctx *ctx;
9834 : uint32_t limit_set_complement;
9835 : uint64_t min_limit_per_sec;
9836 : int i;
9837 9 : bool disable_rate_limit = true;
9838 :
9839 45 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
9840 36 : if (limits[i] == SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
9841 0 : continue;
9842 : }
9843 :
9844 36 : if (limits[i] > 0) {
9845 10 : disable_rate_limit = false;
9846 10 : }
9847 :
9848 36 : if (bdev_qos_is_iops_rate_limit(i) == true) {
9849 9 : min_limit_per_sec = SPDK_BDEV_QOS_MIN_IOS_PER_SEC;
9850 9 : } else {
9851 27 : if (limits[i] > SPDK_BDEV_QOS_MAX_MBYTES_PER_SEC) {
9852 0 : SPDK_WARNLOG("Requested rate limit %" PRIu64 " will result in uint64_t overflow, "
9853 : "reset to %" PRIu64 "\n", limits[i], SPDK_BDEV_QOS_MAX_MBYTES_PER_SEC);
9854 0 : limits[i] = SPDK_BDEV_QOS_MAX_MBYTES_PER_SEC;
9855 0 : }
9856 : /* Change from megabyte to byte rate limit */
9857 27 : limits[i] = limits[i] * 1024 * 1024;
9858 27 : min_limit_per_sec = SPDK_BDEV_QOS_MIN_BYTES_PER_SEC;
9859 : }
9860 :
9861 36 : limit_set_complement = limits[i] % min_limit_per_sec;
9862 36 : if (limit_set_complement) {
9863 0 : SPDK_ERRLOG("Requested rate limit %" PRIu64 " is not a multiple of %" PRIu64 "\n",
9864 : limits[i], min_limit_per_sec);
9865 0 : limits[i] += min_limit_per_sec - limit_set_complement;
9866 0 : SPDK_ERRLOG("Round up the rate limit to %" PRIu64 "\n", limits[i]);
9867 0 : }
9868 36 : }
9869 :
9870 9 : ctx = calloc(1, sizeof(*ctx));
9871 9 : if (ctx == NULL) {
9872 0 : cb_fn(cb_arg, -ENOMEM);
9873 0 : return;
9874 : }
9875 :
9876 9 : ctx->cb_fn = cb_fn;
9877 9 : ctx->cb_arg = cb_arg;
9878 9 : ctx->bdev = bdev;
9879 :
9880 9 : spdk_spin_lock(&bdev->internal.spinlock);
9881 9 : if (bdev->internal.qos_mod_in_progress) {
9882 1 : spdk_spin_unlock(&bdev->internal.spinlock);
9883 1 : free(ctx);
9884 1 : cb_fn(cb_arg, -EAGAIN);
9885 1 : return;
9886 : }
9887 8 : bdev->internal.qos_mod_in_progress = true;
9888 :
9889 8 : if (disable_rate_limit == true && bdev->internal.qos) {
9890 10 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
9891 8 : if (limits[i] == SPDK_BDEV_QOS_LIMIT_NOT_DEFINED &&
9892 0 : (bdev->internal.qos->rate_limits[i].limit > 0 &&
9893 0 : bdev->internal.qos->rate_limits[i].limit !=
9894 : SPDK_BDEV_QOS_LIMIT_NOT_DEFINED)) {
9895 0 : disable_rate_limit = false;
9896 0 : break;
9897 : }
9898 8 : }
9899 2 : }
9900 :
9901 8 : if (disable_rate_limit == false) {
9902 5 : if (bdev->internal.qos == NULL) {
9903 4 : bdev->internal.qos = calloc(1, sizeof(*bdev->internal.qos));
9904 4 : if (!bdev->internal.qos) {
9905 0 : spdk_spin_unlock(&bdev->internal.spinlock);
9906 0 : SPDK_ERRLOG("Unable to allocate memory for QoS tracking\n");
9907 0 : bdev_set_qos_limit_done(ctx, -ENOMEM);
9908 0 : return;
9909 : }
9910 4 : }
9911 :
9912 5 : if (bdev->internal.qos->thread == NULL) {
9913 : /* Enabling */
9914 4 : bdev_set_qos_rate_limits(bdev, limits);
9915 :
9916 4 : spdk_bdev_for_each_channel(bdev, bdev_enable_qos_msg, ctx,
9917 : bdev_enable_qos_done);
9918 4 : } else {
9919 : /* Updating */
9920 1 : bdev_set_qos_rate_limits(bdev, limits);
9921 :
9922 2 : spdk_thread_send_msg(bdev->internal.qos->thread,
9923 1 : bdev_update_qos_rate_limit_msg, ctx);
9924 : }
9925 5 : } else {
9926 3 : if (bdev->internal.qos != NULL) {
9927 2 : bdev_set_qos_rate_limits(bdev, limits);
9928 :
9929 : /* Disabling */
9930 2 : spdk_bdev_for_each_channel(bdev, bdev_disable_qos_msg, ctx,
9931 : bdev_disable_qos_msg_done);
9932 2 : } else {
9933 1 : spdk_spin_unlock(&bdev->internal.spinlock);
9934 1 : bdev_set_qos_limit_done(ctx, 0);
9935 1 : return;
9936 : }
9937 : }
9938 :
9939 7 : spdk_spin_unlock(&bdev->internal.spinlock);
9940 9 : }
9941 :
9942 : struct spdk_bdev_histogram_ctx {
9943 : spdk_bdev_histogram_status_cb cb_fn;
9944 : void *cb_arg;
9945 : struct spdk_bdev *bdev;
9946 : int status;
9947 : };
9948 :
9949 : static void
9950 2 : bdev_histogram_disable_channel_cb(struct spdk_bdev *bdev, void *_ctx, int status)
9951 : {
9952 2 : struct spdk_bdev_histogram_ctx *ctx = _ctx;
9953 :
9954 2 : spdk_spin_lock(&ctx->bdev->internal.spinlock);
9955 2 : ctx->bdev->internal.histogram_in_progress = false;
9956 2 : spdk_spin_unlock(&ctx->bdev->internal.spinlock);
9957 2 : ctx->cb_fn(ctx->cb_arg, ctx->status);
9958 2 : free(ctx);
9959 2 : }
9960 :
9961 : static void
9962 3 : bdev_histogram_disable_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
9963 : struct spdk_io_channel *_ch, void *_ctx)
9964 : {
9965 3 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
9966 :
9967 3 : if (ch->histogram != NULL) {
9968 3 : spdk_histogram_data_free(ch->histogram);
9969 3 : ch->histogram = NULL;
9970 3 : }
9971 3 : spdk_bdev_for_each_channel_continue(i, 0);
9972 3 : }
9973 :
9974 : static void
9975 2 : bdev_histogram_enable_channel_cb(struct spdk_bdev *bdev, void *_ctx, int status)
9976 : {
9977 2 : struct spdk_bdev_histogram_ctx *ctx = _ctx;
9978 :
9979 2 : if (status != 0) {
9980 0 : ctx->status = status;
9981 0 : ctx->bdev->internal.histogram_enabled = false;
9982 0 : spdk_bdev_for_each_channel(ctx->bdev, bdev_histogram_disable_channel, ctx,
9983 : bdev_histogram_disable_channel_cb);
9984 0 : } else {
9985 2 : spdk_spin_lock(&ctx->bdev->internal.spinlock);
9986 2 : ctx->bdev->internal.histogram_in_progress = false;
9987 2 : spdk_spin_unlock(&ctx->bdev->internal.spinlock);
9988 2 : ctx->cb_fn(ctx->cb_arg, ctx->status);
9989 2 : free(ctx);
9990 : }
9991 2 : }
9992 :
9993 : static void
9994 3 : bdev_histogram_enable_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
9995 : struct spdk_io_channel *_ch, void *_ctx)
9996 : {
9997 3 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
9998 3 : int status = 0;
9999 :
10000 3 : if (ch->histogram == NULL) {
10001 3 : ch->histogram = spdk_histogram_data_alloc();
10002 3 : if (ch->histogram == NULL) {
10003 0 : status = -ENOMEM;
10004 0 : }
10005 3 : }
10006 :
10007 3 : spdk_bdev_for_each_channel_continue(i, status);
10008 3 : }
10009 :
10010 : void
10011 4 : spdk_bdev_histogram_enable_ext(struct spdk_bdev *bdev, spdk_bdev_histogram_status_cb cb_fn,
10012 : void *cb_arg, bool enable, struct spdk_bdev_enable_histogram_opts *opts)
10013 : {
10014 : struct spdk_bdev_histogram_ctx *ctx;
10015 :
10016 4 : ctx = calloc(1, sizeof(struct spdk_bdev_histogram_ctx));
10017 4 : if (ctx == NULL) {
10018 0 : cb_fn(cb_arg, -ENOMEM);
10019 0 : return;
10020 : }
10021 :
10022 4 : ctx->bdev = bdev;
10023 4 : ctx->status = 0;
10024 4 : ctx->cb_fn = cb_fn;
10025 4 : ctx->cb_arg = cb_arg;
10026 :
10027 4 : spdk_spin_lock(&bdev->internal.spinlock);
10028 4 : if (bdev->internal.histogram_in_progress) {
10029 0 : spdk_spin_unlock(&bdev->internal.spinlock);
10030 0 : free(ctx);
10031 0 : cb_fn(cb_arg, -EAGAIN);
10032 0 : return;
10033 : }
10034 :
10035 4 : bdev->internal.histogram_in_progress = true;
10036 4 : spdk_spin_unlock(&bdev->internal.spinlock);
10037 :
10038 4 : bdev->internal.histogram_enabled = enable;
10039 4 : bdev->internal.histogram_io_type = opts->io_type;
10040 :
10041 4 : if (enable) {
10042 : /* Allocate histogram for each channel */
10043 2 : spdk_bdev_for_each_channel(bdev, bdev_histogram_enable_channel, ctx,
10044 : bdev_histogram_enable_channel_cb);
10045 2 : } else {
10046 2 : spdk_bdev_for_each_channel(bdev, bdev_histogram_disable_channel, ctx,
10047 : bdev_histogram_disable_channel_cb);
10048 : }
10049 4 : }
10050 :
10051 : void
10052 4 : spdk_bdev_enable_histogram_opts_init(struct spdk_bdev_enable_histogram_opts *opts, size_t size)
10053 : {
10054 4 : if (opts == NULL) {
10055 0 : SPDK_ERRLOG("opts should not be NULL\n");
10056 0 : assert(opts != NULL);
10057 0 : return;
10058 : }
10059 4 : if (size == 0) {
10060 0 : SPDK_ERRLOG("size should not be zero\n");
10061 0 : assert(size != 0);
10062 0 : return;
10063 : }
10064 :
10065 4 : memset(opts, 0, size);
10066 4 : opts->size = size;
10067 :
10068 : #define FIELD_OK(field) \
10069 : offsetof(struct spdk_bdev_enable_histogram_opts, field) + sizeof(opts->field) <= size
10070 :
10071 : #define SET_FIELD(field, value) \
10072 : if (FIELD_OK(field)) { \
10073 : opts->field = value; \
10074 : } \
10075 :
10076 4 : SET_FIELD(io_type, 0);
10077 :
10078 : /* You should not remove this statement, but need to update the assert statement
10079 : * if you add a new field, and also add a corresponding SET_FIELD statement */
10080 : SPDK_STATIC_ASSERT(sizeof(struct spdk_bdev_enable_histogram_opts) == 9, "Incorrect size");
10081 :
10082 : #undef FIELD_OK
10083 : #undef SET_FIELD
10084 4 : }
10085 :
10086 : void
10087 4 : spdk_bdev_histogram_enable(struct spdk_bdev *bdev, spdk_bdev_histogram_status_cb cb_fn,
10088 : void *cb_arg, bool enable)
10089 : {
10090 : struct spdk_bdev_enable_histogram_opts opts;
10091 :
10092 4 : spdk_bdev_enable_histogram_opts_init(&opts, sizeof(opts));
10093 4 : spdk_bdev_histogram_enable_ext(bdev, cb_fn, cb_arg, enable, &opts);
10094 4 : }
10095 :
10096 : struct spdk_bdev_histogram_data_ctx {
10097 : spdk_bdev_histogram_data_cb cb_fn;
10098 : void *cb_arg;
10099 : struct spdk_bdev *bdev;
10100 : /** merged histogram data from all channels */
10101 : struct spdk_histogram_data *histogram;
10102 : };
10103 :
10104 : static void
10105 5 : bdev_histogram_get_channel_cb(struct spdk_bdev *bdev, void *_ctx, int status)
10106 : {
10107 5 : struct spdk_bdev_histogram_data_ctx *ctx = _ctx;
10108 :
10109 5 : ctx->cb_fn(ctx->cb_arg, status, ctx->histogram);
10110 5 : free(ctx);
10111 5 : }
10112 :
10113 : static void
10114 7 : bdev_histogram_get_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
10115 : struct spdk_io_channel *_ch, void *_ctx)
10116 : {
10117 7 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
10118 7 : struct spdk_bdev_histogram_data_ctx *ctx = _ctx;
10119 7 : int status = 0;
10120 :
10121 7 : if (ch->histogram == NULL) {
10122 1 : status = -EFAULT;
10123 1 : } else {
10124 6 : spdk_histogram_data_merge(ctx->histogram, ch->histogram);
10125 : }
10126 :
10127 7 : spdk_bdev_for_each_channel_continue(i, status);
10128 7 : }
10129 :
10130 : void
10131 5 : spdk_bdev_histogram_get(struct spdk_bdev *bdev, struct spdk_histogram_data *histogram,
10132 : spdk_bdev_histogram_data_cb cb_fn,
10133 : void *cb_arg)
10134 : {
10135 : struct spdk_bdev_histogram_data_ctx *ctx;
10136 :
10137 5 : ctx = calloc(1, sizeof(struct spdk_bdev_histogram_data_ctx));
10138 5 : if (ctx == NULL) {
10139 0 : cb_fn(cb_arg, -ENOMEM, NULL);
10140 0 : return;
10141 : }
10142 :
10143 5 : ctx->bdev = bdev;
10144 5 : ctx->cb_fn = cb_fn;
10145 5 : ctx->cb_arg = cb_arg;
10146 :
10147 5 : ctx->histogram = histogram;
10148 :
10149 5 : spdk_bdev_for_each_channel(bdev, bdev_histogram_get_channel, ctx,
10150 : bdev_histogram_get_channel_cb);
10151 5 : }
10152 :
10153 : void
10154 2 : spdk_bdev_channel_get_histogram(struct spdk_io_channel *ch, spdk_bdev_histogram_data_cb cb_fn,
10155 : void *cb_arg)
10156 : {
10157 2 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(ch);
10158 2 : int status = 0;
10159 :
10160 2 : assert(cb_fn != NULL);
10161 :
10162 2 : if (bdev_ch->histogram == NULL) {
10163 1 : status = -EFAULT;
10164 1 : }
10165 2 : cb_fn(cb_arg, status, bdev_ch->histogram);
10166 2 : }
10167 :
10168 : size_t
10169 0 : spdk_bdev_get_media_events(struct spdk_bdev_desc *desc, struct spdk_bdev_media_event *events,
10170 : size_t max_events)
10171 : {
10172 : struct media_event_entry *entry;
10173 0 : size_t num_events = 0;
10174 :
10175 0 : for (; num_events < max_events; ++num_events) {
10176 0 : entry = TAILQ_FIRST(&desc->pending_media_events);
10177 0 : if (entry == NULL) {
10178 0 : break;
10179 : }
10180 :
10181 0 : events[num_events] = entry->event;
10182 0 : TAILQ_REMOVE(&desc->pending_media_events, entry, tailq);
10183 0 : TAILQ_INSERT_TAIL(&desc->free_media_events, entry, tailq);
10184 0 : }
10185 :
10186 0 : return num_events;
10187 : }
10188 :
10189 : int
10190 0 : spdk_bdev_push_media_events(struct spdk_bdev *bdev, const struct spdk_bdev_media_event *events,
10191 : size_t num_events)
10192 : {
10193 : struct spdk_bdev_desc *desc;
10194 : struct media_event_entry *entry;
10195 : size_t event_id;
10196 0 : int rc = 0;
10197 :
10198 0 : assert(bdev->media_events);
10199 :
10200 0 : spdk_spin_lock(&bdev->internal.spinlock);
10201 0 : TAILQ_FOREACH(desc, &bdev->internal.open_descs, link) {
10202 0 : if (desc->write) {
10203 0 : break;
10204 : }
10205 0 : }
10206 :
10207 0 : if (desc == NULL || desc->media_events_buffer == NULL) {
10208 0 : rc = -ENODEV;
10209 0 : goto out;
10210 : }
10211 :
10212 0 : for (event_id = 0; event_id < num_events; ++event_id) {
10213 0 : entry = TAILQ_FIRST(&desc->free_media_events);
10214 0 : if (entry == NULL) {
10215 0 : break;
10216 : }
10217 :
10218 0 : TAILQ_REMOVE(&desc->free_media_events, entry, tailq);
10219 0 : TAILQ_INSERT_TAIL(&desc->pending_media_events, entry, tailq);
10220 0 : entry->event = events[event_id];
10221 0 : }
10222 :
10223 0 : rc = event_id;
10224 : out:
10225 0 : spdk_spin_unlock(&bdev->internal.spinlock);
10226 0 : return rc;
10227 : }
10228 :
10229 : static void
10230 0 : _media_management_notify(void *arg)
10231 : {
10232 0 : struct spdk_bdev_desc *desc = arg;
10233 :
10234 0 : _event_notify(desc, SPDK_BDEV_EVENT_MEDIA_MANAGEMENT);
10235 0 : }
10236 :
10237 : void
10238 0 : spdk_bdev_notify_media_management(struct spdk_bdev *bdev)
10239 : {
10240 : struct spdk_bdev_desc *desc;
10241 :
10242 0 : spdk_spin_lock(&bdev->internal.spinlock);
10243 0 : TAILQ_FOREACH(desc, &bdev->internal.open_descs, link) {
10244 0 : if (!TAILQ_EMPTY(&desc->pending_media_events)) {
10245 0 : event_notify(desc, _media_management_notify);
10246 0 : }
10247 0 : }
10248 0 : spdk_spin_unlock(&bdev->internal.spinlock);
10249 0 : }
10250 :
10251 : struct locked_lba_range_ctx {
10252 : struct lba_range range;
10253 : struct lba_range *current_range;
10254 : struct lba_range *owner_range;
10255 : struct spdk_poller *poller;
10256 : lock_range_cb cb_fn;
10257 : void *cb_arg;
10258 : };
10259 :
10260 : static void
10261 0 : bdev_lock_error_cleanup_cb(struct spdk_bdev *bdev, void *_ctx, int status)
10262 : {
10263 0 : struct locked_lba_range_ctx *ctx = _ctx;
10264 :
10265 0 : ctx->cb_fn(&ctx->range, ctx->cb_arg, -ENOMEM);
10266 0 : free(ctx);
10267 0 : }
10268 :
10269 : static void bdev_unlock_lba_range_get_channel(struct spdk_bdev_channel_iter *i,
10270 : struct spdk_bdev *bdev, struct spdk_io_channel *ch, void *_ctx);
10271 :
10272 : static void
10273 14 : bdev_lock_lba_range_cb(struct spdk_bdev *bdev, void *_ctx, int status)
10274 : {
10275 14 : struct locked_lba_range_ctx *ctx = _ctx;
10276 :
10277 14 : if (status == -ENOMEM) {
10278 : /* One of the channels could not allocate a range object.
10279 : * So we have to go back and clean up any ranges that were
10280 : * allocated successfully before we return error status to
10281 : * the caller. We can reuse the unlock function to do that
10282 : * clean up.
10283 : */
10284 0 : spdk_bdev_for_each_channel(bdev, bdev_unlock_lba_range_get_channel, ctx,
10285 : bdev_lock_error_cleanup_cb);
10286 0 : return;
10287 : }
10288 :
10289 : /* All channels have locked this range and no I/O overlapping the range
10290 : * are outstanding! Set the owner_ch for the range object for the
10291 : * locking channel, so that this channel will know that it is allowed
10292 : * to write to this range.
10293 : */
10294 14 : if (ctx->owner_range != NULL) {
10295 10 : ctx->owner_range->owner_ch = ctx->range.owner_ch;
10296 10 : }
10297 :
10298 14 : ctx->cb_fn(&ctx->range, ctx->cb_arg, status);
10299 :
10300 : /* Don't free the ctx here. Its range is in the bdev's global list of
10301 : * locked ranges still, and will be removed and freed when this range
10302 : * is later unlocked.
10303 : */
10304 14 : }
10305 :
10306 : static int
10307 17 : bdev_lock_lba_range_check_io(void *_i)
10308 : {
10309 17 : struct spdk_bdev_channel_iter *i = _i;
10310 17 : struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i->i);
10311 17 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
10312 17 : struct locked_lba_range_ctx *ctx = i->ctx;
10313 17 : struct lba_range *range = ctx->current_range;
10314 : struct spdk_bdev_io *bdev_io;
10315 :
10316 17 : spdk_poller_unregister(&ctx->poller);
10317 :
10318 : /* The range is now in the locked_ranges, so no new IO can be submitted to this
10319 : * range. But we need to wait until any outstanding IO overlapping with this range
10320 : * are completed.
10321 : */
10322 18 : TAILQ_FOREACH(bdev_io, &ch->io_submitted, internal.ch_link) {
10323 3 : if (bdev_io_range_is_locked(bdev_io, range)) {
10324 2 : ctx->poller = SPDK_POLLER_REGISTER(bdev_lock_lba_range_check_io, i, 100);
10325 2 : return SPDK_POLLER_BUSY;
10326 : }
10327 1 : }
10328 :
10329 15 : spdk_bdev_for_each_channel_continue(i, 0);
10330 15 : return SPDK_POLLER_BUSY;
10331 17 : }
10332 :
10333 : static void
10334 15 : bdev_lock_lba_range_get_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
10335 : struct spdk_io_channel *_ch, void *_ctx)
10336 : {
10337 15 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
10338 15 : struct locked_lba_range_ctx *ctx = _ctx;
10339 : struct lba_range *range;
10340 :
10341 16 : TAILQ_FOREACH(range, &ch->locked_ranges, tailq) {
10342 1 : if (range->length == ctx->range.length &&
10343 0 : range->offset == ctx->range.offset &&
10344 0 : range->locked_ctx == ctx->range.locked_ctx) {
10345 : /* This range already exists on this channel, so don't add
10346 : * it again. This can happen when a new channel is created
10347 : * while the for_each_channel operation is in progress.
10348 : * Do not check for outstanding I/O in that case, since the
10349 : * range was locked before any I/O could be submitted to the
10350 : * new channel.
10351 : */
10352 0 : spdk_bdev_for_each_channel_continue(i, 0);
10353 0 : return;
10354 : }
10355 1 : }
10356 :
10357 15 : range = calloc(1, sizeof(*range));
10358 15 : if (range == NULL) {
10359 0 : spdk_bdev_for_each_channel_continue(i, -ENOMEM);
10360 0 : return;
10361 : }
10362 :
10363 15 : range->length = ctx->range.length;
10364 15 : range->offset = ctx->range.offset;
10365 15 : range->locked_ctx = ctx->range.locked_ctx;
10366 15 : range->quiesce = ctx->range.quiesce;
10367 15 : ctx->current_range = range;
10368 15 : if (ctx->range.owner_ch == ch) {
10369 : /* This is the range object for the channel that will hold
10370 : * the lock. Store it in the ctx object so that we can easily
10371 : * set its owner_ch after the lock is finally acquired.
10372 : */
10373 10 : ctx->owner_range = range;
10374 10 : }
10375 15 : TAILQ_INSERT_TAIL(&ch->locked_ranges, range, tailq);
10376 15 : bdev_lock_lba_range_check_io(i);
10377 15 : }
10378 :
10379 : static void
10380 14 : bdev_lock_lba_range_ctx(struct spdk_bdev *bdev, struct locked_lba_range_ctx *ctx)
10381 : {
10382 14 : assert(spdk_get_thread() == ctx->range.owner_thread);
10383 14 : assert(ctx->range.owner_ch == NULL ||
10384 : spdk_io_channel_get_thread(ctx->range.owner_ch->channel) == ctx->range.owner_thread);
10385 :
10386 : /* We will add a copy of this range to each channel now. */
10387 14 : spdk_bdev_for_each_channel(bdev, bdev_lock_lba_range_get_channel, ctx,
10388 : bdev_lock_lba_range_cb);
10389 14 : }
10390 :
10391 : static bool
10392 17 : bdev_lba_range_overlaps_tailq(struct lba_range *range, lba_range_tailq_t *tailq)
10393 : {
10394 : struct lba_range *r;
10395 :
10396 18 : TAILQ_FOREACH(r, tailq, tailq) {
10397 4 : if (bdev_lba_range_overlapped(range, r)) {
10398 3 : return true;
10399 : }
10400 1 : }
10401 14 : return false;
10402 17 : }
10403 :
10404 : static void bdev_quiesce_range_locked(struct lba_range *range, void *ctx, int status);
10405 :
10406 : static int
10407 14 : _bdev_lock_lba_range(struct spdk_bdev *bdev, struct spdk_bdev_channel *ch,
10408 : uint64_t offset, uint64_t length,
10409 : lock_range_cb cb_fn, void *cb_arg)
10410 : {
10411 : struct locked_lba_range_ctx *ctx;
10412 :
10413 14 : ctx = calloc(1, sizeof(*ctx));
10414 14 : if (ctx == NULL) {
10415 0 : return -ENOMEM;
10416 : }
10417 :
10418 14 : ctx->range.offset = offset;
10419 14 : ctx->range.length = length;
10420 14 : ctx->range.owner_thread = spdk_get_thread();
10421 14 : ctx->range.owner_ch = ch;
10422 14 : ctx->range.locked_ctx = cb_arg;
10423 14 : ctx->range.bdev = bdev;
10424 14 : ctx->range.quiesce = (cb_fn == bdev_quiesce_range_locked);
10425 14 : ctx->cb_fn = cb_fn;
10426 14 : ctx->cb_arg = cb_arg;
10427 :
10428 14 : spdk_spin_lock(&bdev->internal.spinlock);
10429 14 : if (bdev_lba_range_overlaps_tailq(&ctx->range, &bdev->internal.locked_ranges)) {
10430 : /* There is an active lock overlapping with this range.
10431 : * Put it on the pending list until this range no
10432 : * longer overlaps with another.
10433 : */
10434 2 : TAILQ_INSERT_TAIL(&bdev->internal.pending_locked_ranges, &ctx->range, tailq);
10435 2 : } else {
10436 12 : TAILQ_INSERT_TAIL(&bdev->internal.locked_ranges, &ctx->range, tailq);
10437 12 : bdev_lock_lba_range_ctx(bdev, ctx);
10438 : }
10439 14 : spdk_spin_unlock(&bdev->internal.spinlock);
10440 14 : return 0;
10441 14 : }
10442 :
10443 : static int
10444 10 : bdev_lock_lba_range(struct spdk_bdev_desc *desc, struct spdk_io_channel *_ch,
10445 : uint64_t offset, uint64_t length,
10446 : lock_range_cb cb_fn, void *cb_arg)
10447 : {
10448 10 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
10449 10 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
10450 :
10451 10 : if (cb_arg == NULL) {
10452 0 : SPDK_ERRLOG("cb_arg must not be NULL\n");
10453 0 : return -EINVAL;
10454 : }
10455 :
10456 10 : return _bdev_lock_lba_range(bdev, ch, offset, length, cb_fn, cb_arg);
10457 10 : }
10458 :
10459 : static void
10460 2 : bdev_lock_lba_range_ctx_msg(void *_ctx)
10461 : {
10462 2 : struct locked_lba_range_ctx *ctx = _ctx;
10463 :
10464 2 : bdev_lock_lba_range_ctx(ctx->range.bdev, ctx);
10465 2 : }
10466 :
10467 : static void
10468 14 : bdev_unlock_lba_range_cb(struct spdk_bdev *bdev, void *_ctx, int status)
10469 : {
10470 14 : struct locked_lba_range_ctx *ctx = _ctx;
10471 : struct locked_lba_range_ctx *pending_ctx;
10472 : struct lba_range *range, *tmp;
10473 :
10474 14 : spdk_spin_lock(&bdev->internal.spinlock);
10475 : /* Check if there are any pending locked ranges that overlap with this range
10476 : * that was just unlocked. If there are, check that it doesn't overlap with any
10477 : * other locked ranges before calling bdev_lock_lba_range_ctx which will start
10478 : * the lock process.
10479 : */
10480 17 : TAILQ_FOREACH_SAFE(range, &bdev->internal.pending_locked_ranges, tailq, tmp) {
10481 3 : if (bdev_lba_range_overlapped(range, &ctx->range) &&
10482 3 : !bdev_lba_range_overlaps_tailq(range, &bdev->internal.locked_ranges)) {
10483 2 : TAILQ_REMOVE(&bdev->internal.pending_locked_ranges, range, tailq);
10484 2 : pending_ctx = SPDK_CONTAINEROF(range, struct locked_lba_range_ctx, range);
10485 2 : TAILQ_INSERT_TAIL(&bdev->internal.locked_ranges, range, tailq);
10486 4 : spdk_thread_send_msg(pending_ctx->range.owner_thread,
10487 2 : bdev_lock_lba_range_ctx_msg, pending_ctx);
10488 2 : }
10489 3 : }
10490 14 : spdk_spin_unlock(&bdev->internal.spinlock);
10491 :
10492 14 : ctx->cb_fn(&ctx->range, ctx->cb_arg, status);
10493 14 : free(ctx);
10494 14 : }
10495 :
10496 : static void
10497 16 : bdev_unlock_lba_range_get_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
10498 : struct spdk_io_channel *_ch, void *_ctx)
10499 : {
10500 16 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
10501 16 : struct locked_lba_range_ctx *ctx = _ctx;
10502 : TAILQ_HEAD(, spdk_bdev_io) io_locked;
10503 : struct spdk_bdev_io *bdev_io;
10504 : struct lba_range *range;
10505 :
10506 16 : TAILQ_FOREACH(range, &ch->locked_ranges, tailq) {
10507 32 : if (ctx->range.offset == range->offset &&
10508 16 : ctx->range.length == range->length &&
10509 16 : ctx->range.locked_ctx == range->locked_ctx) {
10510 16 : TAILQ_REMOVE(&ch->locked_ranges, range, tailq);
10511 16 : free(range);
10512 16 : break;
10513 : }
10514 0 : }
10515 :
10516 : /* Note: we should almost always be able to assert that the range specified
10517 : * was found. But there are some very rare corner cases where a new channel
10518 : * gets created simultaneously with a range unlock, where this function
10519 : * would execute on that new channel and wouldn't have the range.
10520 : * We also use this to clean up range allocations when a later allocation
10521 : * fails in the locking path.
10522 : * So we can't actually assert() here.
10523 : */
10524 :
10525 : /* Swap the locked IO into a temporary list, and then try to submit them again.
10526 : * We could hyper-optimize this to only resubmit locked I/O that overlap
10527 : * with the range that was just unlocked, but this isn't a performance path so
10528 : * we go for simplicity here.
10529 : */
10530 16 : TAILQ_INIT(&io_locked);
10531 16 : TAILQ_SWAP(&ch->io_locked, &io_locked, spdk_bdev_io, internal.ch_link);
10532 19 : while (!TAILQ_EMPTY(&io_locked)) {
10533 3 : bdev_io = TAILQ_FIRST(&io_locked);
10534 3 : TAILQ_REMOVE(&io_locked, bdev_io, internal.ch_link);
10535 3 : bdev_io_submit(bdev_io);
10536 : }
10537 :
10538 16 : spdk_bdev_for_each_channel_continue(i, 0);
10539 16 : }
10540 :
10541 : static int
10542 14 : _bdev_unlock_lba_range(struct spdk_bdev *bdev, uint64_t offset, uint64_t length,
10543 : lock_range_cb cb_fn, void *cb_arg)
10544 : {
10545 : struct locked_lba_range_ctx *ctx;
10546 : struct lba_range *range;
10547 :
10548 14 : spdk_spin_lock(&bdev->internal.spinlock);
10549 : /* To start the unlock the process, we find the range in the bdev's locked_ranges
10550 : * and remove it. This ensures new channels don't inherit the locked range.
10551 : * Then we will send a message to each channel to remove the range from its
10552 : * per-channel list.
10553 : */
10554 14 : TAILQ_FOREACH(range, &bdev->internal.locked_ranges, tailq) {
10555 24 : if (range->offset == offset && range->length == length &&
10556 14 : (range->owner_ch == NULL || range->locked_ctx == cb_arg)) {
10557 14 : break;
10558 : }
10559 0 : }
10560 14 : if (range == NULL) {
10561 0 : assert(false);
10562 : spdk_spin_unlock(&bdev->internal.spinlock);
10563 : return -EINVAL;
10564 : }
10565 14 : TAILQ_REMOVE(&bdev->internal.locked_ranges, range, tailq);
10566 14 : ctx = SPDK_CONTAINEROF(range, struct locked_lba_range_ctx, range);
10567 14 : spdk_spin_unlock(&bdev->internal.spinlock);
10568 :
10569 14 : ctx->cb_fn = cb_fn;
10570 14 : ctx->cb_arg = cb_arg;
10571 :
10572 14 : spdk_bdev_for_each_channel(bdev, bdev_unlock_lba_range_get_channel, ctx,
10573 : bdev_unlock_lba_range_cb);
10574 14 : return 0;
10575 : }
10576 :
10577 : static int
10578 12 : bdev_unlock_lba_range(struct spdk_bdev_desc *desc, struct spdk_io_channel *_ch,
10579 : uint64_t offset, uint64_t length,
10580 : lock_range_cb cb_fn, void *cb_arg)
10581 : {
10582 12 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
10583 12 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
10584 : struct lba_range *range;
10585 12 : bool range_found = false;
10586 :
10587 : /* Let's make sure the specified channel actually has a lock on
10588 : * the specified range. Note that the range must match exactly.
10589 : */
10590 14 : TAILQ_FOREACH(range, &ch->locked_ranges, tailq) {
10591 22 : if (range->offset == offset && range->length == length &&
10592 11 : range->owner_ch == ch && range->locked_ctx == cb_arg) {
10593 10 : range_found = true;
10594 10 : break;
10595 : }
10596 2 : }
10597 :
10598 12 : if (!range_found) {
10599 2 : return -EINVAL;
10600 : }
10601 :
10602 10 : return _bdev_unlock_lba_range(bdev, offset, length, cb_fn, cb_arg);
10603 12 : }
10604 :
10605 : struct bdev_quiesce_ctx {
10606 : spdk_bdev_quiesce_cb cb_fn;
10607 : void *cb_arg;
10608 : };
10609 :
10610 : static void
10611 4 : bdev_unquiesce_range_unlocked(struct lba_range *range, void *ctx, int status)
10612 : {
10613 4 : struct bdev_quiesce_ctx *quiesce_ctx = ctx;
10614 :
10615 4 : if (quiesce_ctx->cb_fn != NULL) {
10616 4 : quiesce_ctx->cb_fn(quiesce_ctx->cb_arg, status);
10617 4 : }
10618 :
10619 4 : free(quiesce_ctx);
10620 4 : }
10621 :
10622 : static void
10623 4 : bdev_quiesce_range_locked(struct lba_range *range, void *ctx, int status)
10624 : {
10625 4 : struct bdev_quiesce_ctx *quiesce_ctx = ctx;
10626 4 : struct spdk_bdev_module *module = range->bdev->module;
10627 :
10628 4 : if (status != 0) {
10629 0 : if (quiesce_ctx->cb_fn != NULL) {
10630 0 : quiesce_ctx->cb_fn(quiesce_ctx->cb_arg, status);
10631 0 : }
10632 0 : free(quiesce_ctx);
10633 0 : return;
10634 : }
10635 :
10636 4 : spdk_spin_lock(&module->internal.spinlock);
10637 4 : TAILQ_INSERT_TAIL(&module->internal.quiesced_ranges, range, tailq_module);
10638 4 : spdk_spin_unlock(&module->internal.spinlock);
10639 :
10640 4 : if (quiesce_ctx->cb_fn != NULL) {
10641 : /* copy the context in case the range is unlocked by the callback */
10642 4 : struct bdev_quiesce_ctx tmp = *quiesce_ctx;
10643 :
10644 4 : quiesce_ctx->cb_fn = NULL;
10645 4 : quiesce_ctx->cb_arg = NULL;
10646 :
10647 4 : tmp.cb_fn(tmp.cb_arg, status);
10648 4 : }
10649 : /* quiesce_ctx will be freed on unquiesce */
10650 4 : }
10651 :
10652 : static int
10653 9 : _spdk_bdev_quiesce(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
10654 : uint64_t offset, uint64_t length,
10655 : spdk_bdev_quiesce_cb cb_fn, void *cb_arg,
10656 : bool unquiesce)
10657 : {
10658 : struct bdev_quiesce_ctx *quiesce_ctx;
10659 : int rc;
10660 :
10661 9 : if (module != bdev->module) {
10662 0 : SPDK_ERRLOG("Bdev does not belong to specified module.\n");
10663 0 : return -EINVAL;
10664 : }
10665 :
10666 9 : if (!bdev_io_valid_blocks(bdev, offset, length)) {
10667 0 : return -EINVAL;
10668 : }
10669 :
10670 9 : if (unquiesce) {
10671 : struct lba_range *range;
10672 :
10673 : /* Make sure the specified range is actually quiesced in the specified module and
10674 : * then remove it from the list. Note that the range must match exactly.
10675 : */
10676 5 : spdk_spin_lock(&module->internal.spinlock);
10677 6 : TAILQ_FOREACH(range, &module->internal.quiesced_ranges, tailq_module) {
10678 5 : if (range->bdev == bdev && range->offset == offset && range->length == length) {
10679 4 : TAILQ_REMOVE(&module->internal.quiesced_ranges, range, tailq_module);
10680 4 : break;
10681 : }
10682 1 : }
10683 5 : spdk_spin_unlock(&module->internal.spinlock);
10684 :
10685 5 : if (range == NULL) {
10686 1 : SPDK_ERRLOG("The range to unquiesce was not found.\n");
10687 1 : return -EINVAL;
10688 : }
10689 :
10690 4 : quiesce_ctx = range->locked_ctx;
10691 4 : quiesce_ctx->cb_fn = cb_fn;
10692 4 : quiesce_ctx->cb_arg = cb_arg;
10693 :
10694 4 : rc = _bdev_unlock_lba_range(bdev, offset, length, bdev_unquiesce_range_unlocked, quiesce_ctx);
10695 4 : } else {
10696 4 : quiesce_ctx = malloc(sizeof(*quiesce_ctx));
10697 4 : if (quiesce_ctx == NULL) {
10698 0 : return -ENOMEM;
10699 : }
10700 :
10701 4 : quiesce_ctx->cb_fn = cb_fn;
10702 4 : quiesce_ctx->cb_arg = cb_arg;
10703 :
10704 4 : rc = _bdev_lock_lba_range(bdev, NULL, offset, length, bdev_quiesce_range_locked, quiesce_ctx);
10705 4 : if (rc != 0) {
10706 0 : free(quiesce_ctx);
10707 0 : }
10708 : }
10709 :
10710 8 : return rc;
10711 9 : }
10712 :
10713 : int
10714 3 : spdk_bdev_quiesce(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
10715 : spdk_bdev_quiesce_cb cb_fn, void *cb_arg)
10716 : {
10717 3 : return _spdk_bdev_quiesce(bdev, module, 0, bdev->blockcnt, cb_fn, cb_arg, false);
10718 : }
10719 :
10720 : int
10721 3 : spdk_bdev_unquiesce(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
10722 : spdk_bdev_quiesce_cb cb_fn, void *cb_arg)
10723 : {
10724 3 : return _spdk_bdev_quiesce(bdev, module, 0, bdev->blockcnt, cb_fn, cb_arg, true);
10725 : }
10726 :
10727 : int
10728 1 : spdk_bdev_quiesce_range(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
10729 : uint64_t offset, uint64_t length,
10730 : spdk_bdev_quiesce_cb cb_fn, void *cb_arg)
10731 : {
10732 1 : return _spdk_bdev_quiesce(bdev, module, offset, length, cb_fn, cb_arg, false);
10733 : }
10734 :
10735 : int
10736 2 : spdk_bdev_unquiesce_range(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
10737 : uint64_t offset, uint64_t length,
10738 : spdk_bdev_quiesce_cb cb_fn, void *cb_arg)
10739 : {
10740 2 : return _spdk_bdev_quiesce(bdev, module, offset, length, cb_fn, cb_arg, true);
10741 : }
10742 :
10743 : int
10744 273 : spdk_bdev_get_memory_domains(struct spdk_bdev *bdev, struct spdk_memory_domain **domains,
10745 : int array_size)
10746 : {
10747 273 : if (!bdev) {
10748 1 : return -EINVAL;
10749 : }
10750 :
10751 272 : if (bdev->fn_table->get_memory_domains) {
10752 3 : return bdev->fn_table->get_memory_domains(bdev->ctxt, domains, array_size);
10753 : }
10754 :
10755 269 : return 0;
10756 273 : }
10757 :
10758 : struct spdk_bdev_for_each_io_ctx {
10759 : void *ctx;
10760 : spdk_bdev_io_fn fn;
10761 : spdk_bdev_for_each_io_cb cb;
10762 : };
10763 :
10764 : static void
10765 0 : bdev_channel_for_each_io(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
10766 : struct spdk_io_channel *io_ch, void *_ctx)
10767 : {
10768 0 : struct spdk_bdev_for_each_io_ctx *ctx = _ctx;
10769 0 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
10770 : struct spdk_bdev_io *bdev_io;
10771 0 : int rc = 0;
10772 :
10773 0 : TAILQ_FOREACH(bdev_io, &bdev_ch->io_submitted, internal.ch_link) {
10774 0 : rc = ctx->fn(ctx->ctx, bdev_io);
10775 0 : if (rc != 0) {
10776 0 : break;
10777 : }
10778 0 : }
10779 :
10780 0 : spdk_bdev_for_each_channel_continue(i, rc);
10781 0 : }
10782 :
10783 : static void
10784 0 : bdev_for_each_io_done(struct spdk_bdev *bdev, void *_ctx, int status)
10785 : {
10786 0 : struct spdk_bdev_for_each_io_ctx *ctx = _ctx;
10787 :
10788 0 : ctx->cb(ctx->ctx, status);
10789 :
10790 0 : free(ctx);
10791 0 : }
10792 :
10793 : void
10794 0 : spdk_bdev_for_each_bdev_io(struct spdk_bdev *bdev, void *_ctx, spdk_bdev_io_fn fn,
10795 : spdk_bdev_for_each_io_cb cb)
10796 : {
10797 : struct spdk_bdev_for_each_io_ctx *ctx;
10798 :
10799 0 : assert(fn != NULL && cb != NULL);
10800 :
10801 0 : ctx = calloc(1, sizeof(*ctx));
10802 0 : if (ctx == NULL) {
10803 0 : SPDK_ERRLOG("Failed to allocate context.\n");
10804 0 : cb(_ctx, -ENOMEM);
10805 0 : return;
10806 : }
10807 :
10808 0 : ctx->ctx = _ctx;
10809 0 : ctx->fn = fn;
10810 0 : ctx->cb = cb;
10811 :
10812 0 : spdk_bdev_for_each_channel(bdev, bdev_channel_for_each_io, ctx,
10813 : bdev_for_each_io_done);
10814 0 : }
10815 :
10816 : void
10817 132 : spdk_bdev_for_each_channel_continue(struct spdk_bdev_channel_iter *iter, int status)
10818 : {
10819 132 : spdk_for_each_channel_continue(iter->i, status);
10820 132 : }
10821 :
10822 : static struct spdk_bdev *
10823 362 : io_channel_iter_get_bdev(struct spdk_io_channel_iter *i)
10824 : {
10825 362 : void *io_device = spdk_io_channel_iter_get_io_device(i);
10826 :
10827 362 : return __bdev_from_io_dev(io_device);
10828 : }
10829 :
10830 : static void
10831 132 : bdev_each_channel_msg(struct spdk_io_channel_iter *i)
10832 : {
10833 132 : struct spdk_bdev_channel_iter *iter = spdk_io_channel_iter_get_ctx(i);
10834 132 : struct spdk_bdev *bdev = io_channel_iter_get_bdev(i);
10835 132 : struct spdk_io_channel *ch = spdk_io_channel_iter_get_channel(i);
10836 :
10837 132 : iter->i = i;
10838 132 : iter->fn(iter, bdev, ch, iter->ctx);
10839 132 : }
10840 :
10841 : static void
10842 230 : bdev_each_channel_cpl(struct spdk_io_channel_iter *i, int status)
10843 : {
10844 230 : struct spdk_bdev_channel_iter *iter = spdk_io_channel_iter_get_ctx(i);
10845 230 : struct spdk_bdev *bdev = io_channel_iter_get_bdev(i);
10846 :
10847 230 : iter->i = i;
10848 230 : iter->cpl(bdev, iter->ctx, status);
10849 :
10850 230 : free(iter);
10851 230 : }
10852 :
10853 : void
10854 230 : spdk_bdev_for_each_channel(struct spdk_bdev *bdev, spdk_bdev_for_each_channel_msg fn,
10855 : void *ctx, spdk_bdev_for_each_channel_done cpl)
10856 : {
10857 : struct spdk_bdev_channel_iter *iter;
10858 :
10859 230 : assert(bdev != NULL && fn != NULL && ctx != NULL);
10860 :
10861 230 : iter = calloc(1, sizeof(struct spdk_bdev_channel_iter));
10862 230 : if (iter == NULL) {
10863 0 : SPDK_ERRLOG("Unable to allocate iterator\n");
10864 0 : assert(false);
10865 : return;
10866 : }
10867 :
10868 230 : iter->fn = fn;
10869 230 : iter->cpl = cpl;
10870 230 : iter->ctx = ctx;
10871 :
10872 460 : spdk_for_each_channel(__bdev_to_io_dev(bdev), bdev_each_channel_msg,
10873 230 : iter, bdev_each_channel_cpl);
10874 230 : }
10875 :
10876 : static void
10877 3 : bdev_copy_do_write_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
10878 : {
10879 3 : struct spdk_bdev_io *parent_io = cb_arg;
10880 :
10881 3 : spdk_bdev_free_io(bdev_io);
10882 :
10883 : /* Check return status of write */
10884 3 : parent_io->internal.status = success ? SPDK_BDEV_IO_STATUS_SUCCESS : SPDK_BDEV_IO_STATUS_FAILED;
10885 3 : parent_io->internal.cb(parent_io, success, parent_io->internal.caller_ctx);
10886 3 : }
10887 :
10888 : static void
10889 3 : bdev_copy_do_write(void *_bdev_io)
10890 : {
10891 3 : struct spdk_bdev_io *bdev_io = _bdev_io;
10892 : int rc;
10893 :
10894 : /* Write blocks */
10895 6 : rc = spdk_bdev_write_blocks_with_md(bdev_io->internal.desc,
10896 3 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
10897 3 : bdev_io->u.bdev.iovs[0].iov_base,
10898 3 : bdev_io->u.bdev.md_buf, bdev_io->u.bdev.offset_blocks,
10899 3 : bdev_io->u.bdev.num_blocks, bdev_copy_do_write_done, bdev_io);
10900 :
10901 3 : if (rc == -ENOMEM) {
10902 0 : bdev_queue_io_wait_with_cb(bdev_io, bdev_copy_do_write);
10903 3 : } else if (rc != 0) {
10904 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
10905 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
10906 0 : }
10907 3 : }
10908 :
10909 : static void
10910 3 : bdev_copy_do_read_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
10911 : {
10912 3 : struct spdk_bdev_io *parent_io = cb_arg;
10913 :
10914 3 : spdk_bdev_free_io(bdev_io);
10915 :
10916 : /* Check return status of read */
10917 3 : if (!success) {
10918 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
10919 0 : parent_io->internal.cb(parent_io, false, parent_io->internal.caller_ctx);
10920 0 : return;
10921 : }
10922 :
10923 : /* Do write */
10924 3 : bdev_copy_do_write(parent_io);
10925 3 : }
10926 :
10927 : static void
10928 3 : bdev_copy_do_read(void *_bdev_io)
10929 : {
10930 3 : struct spdk_bdev_io *bdev_io = _bdev_io;
10931 : int rc;
10932 :
10933 : /* Read blocks */
10934 6 : rc = spdk_bdev_read_blocks_with_md(bdev_io->internal.desc,
10935 3 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
10936 3 : bdev_io->u.bdev.iovs[0].iov_base,
10937 3 : bdev_io->u.bdev.md_buf, bdev_io->u.bdev.copy.src_offset_blocks,
10938 3 : bdev_io->u.bdev.num_blocks, bdev_copy_do_read_done, bdev_io);
10939 :
10940 3 : if (rc == -ENOMEM) {
10941 0 : bdev_queue_io_wait_with_cb(bdev_io, bdev_copy_do_read);
10942 3 : } else if (rc != 0) {
10943 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
10944 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
10945 0 : }
10946 3 : }
10947 :
10948 : static void
10949 3 : bdev_copy_get_buf_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io, bool success)
10950 : {
10951 3 : if (!success) {
10952 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
10953 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
10954 0 : return;
10955 : }
10956 :
10957 3 : bdev_copy_do_read(bdev_io);
10958 3 : }
10959 :
10960 : int
10961 27 : spdk_bdev_copy_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
10962 : uint64_t dst_offset_blocks, uint64_t src_offset_blocks, uint64_t num_blocks,
10963 : spdk_bdev_io_completion_cb cb, void *cb_arg)
10964 : {
10965 27 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
10966 : struct spdk_bdev_io *bdev_io;
10967 27 : struct spdk_bdev_channel *channel = spdk_io_channel_get_ctx(ch);
10968 :
10969 27 : if (!desc->write) {
10970 0 : return -EBADF;
10971 : }
10972 :
10973 27 : if (!bdev_io_valid_blocks(bdev, dst_offset_blocks, num_blocks) ||
10974 27 : !bdev_io_valid_blocks(bdev, src_offset_blocks, num_blocks)) {
10975 0 : SPDK_DEBUGLOG(bdev,
10976 : "Invalid offset or number of blocks: dst %lu, src %lu, count %lu\n",
10977 : dst_offset_blocks, src_offset_blocks, num_blocks);
10978 0 : return -EINVAL;
10979 : }
10980 :
10981 27 : bdev_io = bdev_channel_get_io(channel);
10982 27 : if (!bdev_io) {
10983 0 : return -ENOMEM;
10984 : }
10985 :
10986 27 : bdev_io->internal.ch = channel;
10987 27 : bdev_io->internal.desc = desc;
10988 27 : bdev_io->type = SPDK_BDEV_IO_TYPE_COPY;
10989 :
10990 27 : bdev_io->u.bdev.offset_blocks = dst_offset_blocks;
10991 27 : bdev_io->u.bdev.copy.src_offset_blocks = src_offset_blocks;
10992 27 : bdev_io->u.bdev.num_blocks = num_blocks;
10993 27 : bdev_io->u.bdev.memory_domain = NULL;
10994 27 : bdev_io->u.bdev.memory_domain_ctx = NULL;
10995 27 : bdev_io->u.bdev.iovs = NULL;
10996 27 : bdev_io->u.bdev.iovcnt = 0;
10997 27 : bdev_io->u.bdev.md_buf = NULL;
10998 27 : bdev_io->u.bdev.accel_sequence = NULL;
10999 27 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
11000 :
11001 27 : if (dst_offset_blocks == src_offset_blocks || num_blocks == 0) {
11002 0 : spdk_thread_send_msg(spdk_get_thread(), bdev_io_complete_cb, bdev_io);
11003 0 : return 0;
11004 : }
11005 :
11006 :
11007 : /* If the copy size is large and should be split, use the generic split logic
11008 : * regardless of whether SPDK_BDEV_IO_TYPE_COPY is supported or not.
11009 : *
11010 : * Then, send the copy request if SPDK_BDEV_IO_TYPE_COPY is supported or
11011 : * emulate it using regular read and write requests otherwise.
11012 : */
11013 27 : if (spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COPY) ||
11014 4 : bdev_io->internal.f.split) {
11015 24 : bdev_io_submit(bdev_io);
11016 24 : return 0;
11017 : }
11018 :
11019 3 : spdk_bdev_io_get_buf(bdev_io, bdev_copy_get_buf_cb, num_blocks * spdk_bdev_get_block_size(bdev));
11020 :
11021 3 : return 0;
11022 27 : }
11023 :
11024 3 : SPDK_LOG_REGISTER_COMPONENT(bdev)
11025 :
11026 : static void
11027 0 : bdev_trace(void)
11028 : {
11029 0 : struct spdk_trace_tpoint_opts opts[] = {
11030 : {
11031 : "BDEV_IO_START", TRACE_BDEV_IO_START,
11032 : OWNER_TYPE_BDEV, OBJECT_BDEV_IO, 1,
11033 : {
11034 : { "type", SPDK_TRACE_ARG_TYPE_INT, 8 },
11035 : { "ctx", SPDK_TRACE_ARG_TYPE_PTR, 8 },
11036 : { "offset", SPDK_TRACE_ARG_TYPE_INT, 8 },
11037 : { "qd", SPDK_TRACE_ARG_TYPE_INT, 4 }
11038 : }
11039 : },
11040 : {
11041 : "BDEV_IO_DONE", TRACE_BDEV_IO_DONE,
11042 : OWNER_TYPE_BDEV, OBJECT_BDEV_IO, 0,
11043 : {
11044 : { "ctx", SPDK_TRACE_ARG_TYPE_PTR, 8 },
11045 : { "qd", SPDK_TRACE_ARG_TYPE_INT, 4 }
11046 : }
11047 : },
11048 : {
11049 : "BDEV_IOCH_CREATE", TRACE_BDEV_IOCH_CREATE,
11050 : OWNER_TYPE_BDEV, OBJECT_NONE, 0,
11051 : {
11052 : { "tid", SPDK_TRACE_ARG_TYPE_INT, 8 }
11053 : }
11054 : },
11055 : {
11056 : "BDEV_IOCH_DESTROY", TRACE_BDEV_IOCH_DESTROY,
11057 : OWNER_TYPE_BDEV, OBJECT_NONE, 0,
11058 : {
11059 : { "tid", SPDK_TRACE_ARG_TYPE_INT, 8 }
11060 : }
11061 : },
11062 : };
11063 :
11064 :
11065 0 : spdk_trace_register_owner_type(OWNER_TYPE_BDEV, 'b');
11066 0 : spdk_trace_register_object(OBJECT_BDEV_IO, 'i');
11067 0 : spdk_trace_register_description_ext(opts, SPDK_COUNTOF(opts));
11068 0 : spdk_trace_tpoint_register_relation(TRACE_BDEV_NVME_IO_START, OBJECT_BDEV_IO, 0);
11069 0 : spdk_trace_tpoint_register_relation(TRACE_BDEV_NVME_IO_DONE, OBJECT_BDEV_IO, 0);
11070 0 : spdk_trace_tpoint_register_relation(TRACE_BLOB_REQ_SET_START, OBJECT_BDEV_IO, 0);
11071 0 : spdk_trace_tpoint_register_relation(TRACE_BLOB_REQ_SET_COMPLETE, OBJECT_BDEV_IO, 0);
11072 0 : spdk_trace_tpoint_register_relation(TRACE_BDEV_RAID_IO_START, OBJECT_BDEV_IO, 0);
11073 0 : spdk_trace_tpoint_register_relation(TRACE_BDEV_RAID_IO_DONE, OBJECT_BDEV_IO, 0);
11074 0 : }
11075 3 : SPDK_TRACE_REGISTER_FN(bdev_trace, "bdev", TRACE_GROUP_BDEV)
|