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
86 578 : bdev_name_cmp(struct spdk_bdev_name *name1, struct spdk_bdev_name *name2)
87 : {
88 578 : return strcmp(name1->name, name2->name);
89 : }
90 :
91 1732 : RB_GENERATE_STATIC(bdev_name_tree, spdk_bdev_name, node, bdev_name_cmp);
92 :
93 : struct spdk_bdev_mgr {
94 : struct spdk_mempool *bdev_io_pool;
95 :
96 : void *zero_buffer;
97 :
98 : TAILQ_HEAD(bdev_module_list, spdk_bdev_module) bdev_modules;
99 :
100 : struct spdk_bdev_list bdevs;
101 : struct bdev_name_tree bdev_names;
102 :
103 : bool init_complete;
104 : bool module_init_complete;
105 :
106 : struct spdk_spinlock spinlock;
107 :
108 : TAILQ_HEAD(, spdk_bdev_open_async_ctx) async_bdev_opens;
109 :
110 : #ifdef SPDK_CONFIG_VTUNE
111 : __itt_domain *domain;
112 : #endif
113 : };
114 :
115 : static struct spdk_bdev_mgr g_bdev_mgr = {
116 : .bdev_modules = TAILQ_HEAD_INITIALIZER(g_bdev_mgr.bdev_modules),
117 : .bdevs = TAILQ_HEAD_INITIALIZER(g_bdev_mgr.bdevs),
118 : .bdev_names = RB_INITIALIZER(g_bdev_mgr.bdev_names),
119 : .init_complete = false,
120 : .module_init_complete = false,
121 : .async_bdev_opens = TAILQ_HEAD_INITIALIZER(g_bdev_mgr.async_bdev_opens),
122 : };
123 :
124 : static void
125 : __attribute__((constructor))
126 3 : _bdev_init(void)
127 : {
128 3 : spdk_spin_init(&g_bdev_mgr.spinlock);
129 3 : }
130 :
131 : typedef void (*lock_range_cb)(struct lba_range *range, void *ctx, int status);
132 :
133 : typedef void (*bdev_copy_bounce_buffer_cpl)(void *ctx, int rc);
134 :
135 : struct lba_range {
136 : struct spdk_bdev *bdev;
137 : uint64_t offset;
138 : uint64_t length;
139 : bool quiesce;
140 : void *locked_ctx;
141 : struct spdk_thread *owner_thread;
142 : struct spdk_bdev_channel *owner_ch;
143 : TAILQ_ENTRY(lba_range) tailq;
144 : TAILQ_ENTRY(lba_range) tailq_module;
145 : };
146 :
147 : static struct spdk_bdev_opts g_bdev_opts = {
148 : .bdev_io_pool_size = SPDK_BDEV_IO_POOL_SIZE,
149 : .bdev_io_cache_size = SPDK_BDEV_IO_CACHE_SIZE,
150 : .bdev_auto_examine = SPDK_BDEV_AUTO_EXAMINE,
151 : .iobuf_small_cache_size = BUF_SMALL_CACHE_SIZE,
152 : .iobuf_large_cache_size = BUF_LARGE_CACHE_SIZE,
153 : };
154 :
155 : static spdk_bdev_init_cb g_init_cb_fn = NULL;
156 : static void *g_init_cb_arg = NULL;
157 :
158 : static spdk_bdev_fini_cb g_fini_cb_fn = NULL;
159 : static void *g_fini_cb_arg = NULL;
160 : static struct spdk_thread *g_fini_thread = NULL;
161 :
162 : struct spdk_bdev_qos_limit {
163 : /** IOs or bytes allowed per second (i.e., 1s). */
164 : uint64_t limit;
165 :
166 : /** Remaining IOs or bytes allowed in current timeslice (e.g., 1ms).
167 : * For remaining bytes, allowed to run negative if an I/O is submitted when
168 : * some bytes are remaining, but the I/O is bigger than that amount. The
169 : * excess will be deducted from the next timeslice.
170 : */
171 : int64_t remaining_this_timeslice;
172 :
173 : /** Minimum allowed IOs or bytes to be issued in one timeslice (e.g., 1ms). */
174 : uint32_t min_per_timeslice;
175 :
176 : /** Maximum allowed IOs or bytes to be issued in one timeslice (e.g., 1ms). */
177 : uint32_t max_per_timeslice;
178 :
179 : /** Function to check whether to queue the IO.
180 : * If The IO is allowed to pass, the quota will be reduced correspondingly.
181 : */
182 : bool (*queue_io)(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io);
183 :
184 : /** Function to rewind the quota once the IO was allowed to be sent by this
185 : * limit but queued due to one of the further limits.
186 : */
187 : void (*rewind_quota)(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io);
188 : };
189 :
190 : struct spdk_bdev_qos {
191 : /** Types of structure of rate limits. */
192 : struct spdk_bdev_qos_limit rate_limits[SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES];
193 :
194 : /** The channel that all I/O are funneled through. */
195 : struct spdk_bdev_channel *ch;
196 :
197 : /** The thread on which the poller is running. */
198 : struct spdk_thread *thread;
199 :
200 : /** Size of a timeslice in tsc ticks. */
201 : uint64_t timeslice_size;
202 :
203 : /** Timestamp of start of last timeslice. */
204 : uint64_t last_timeslice;
205 :
206 : /** Poller that processes queued I/O commands each time slice. */
207 : struct spdk_poller *poller;
208 : };
209 :
210 : struct spdk_bdev_mgmt_channel {
211 : /*
212 : * Each thread keeps a cache of bdev_io - this allows
213 : * bdev threads which are *not* DPDK threads to still
214 : * benefit from a per-thread bdev_io cache. Without
215 : * this, non-DPDK threads fetching from the mempool
216 : * incur a cmpxchg on get and put.
217 : */
218 : bdev_io_stailq_t per_thread_cache;
219 : uint32_t per_thread_cache_count;
220 : uint32_t bdev_io_cache_size;
221 :
222 : struct spdk_iobuf_channel iobuf;
223 :
224 : TAILQ_HEAD(, spdk_bdev_shared_resource) shared_resources;
225 : TAILQ_HEAD(, spdk_bdev_io_wait_entry) io_wait_queue;
226 : };
227 :
228 : /*
229 : * Per-module (or per-io_device) data. Multiple bdevs built on the same io_device
230 : * will queue here their IO that awaits retry. It makes it possible to retry sending
231 : * IO to one bdev after IO from other bdev completes.
232 : */
233 : struct spdk_bdev_shared_resource {
234 : /* The bdev management channel */
235 : struct spdk_bdev_mgmt_channel *mgmt_ch;
236 :
237 : /*
238 : * Count of I/O submitted to bdev module and waiting for completion.
239 : * Incremented before submit_request() is called on an spdk_bdev_io.
240 : */
241 : uint64_t io_outstanding;
242 :
243 : /*
244 : * Queue of IO awaiting retry because of a previous NOMEM status returned
245 : * on this channel.
246 : */
247 : bdev_io_tailq_t nomem_io;
248 :
249 : /*
250 : * Threshold which io_outstanding must drop to before retrying nomem_io.
251 : */
252 : uint64_t nomem_threshold;
253 :
254 : /* I/O channel allocated by a bdev module */
255 : struct spdk_io_channel *shared_ch;
256 :
257 : struct spdk_poller *nomem_poller;
258 :
259 : /* Refcount of bdev channels using this resource */
260 : uint32_t ref;
261 :
262 : TAILQ_ENTRY(spdk_bdev_shared_resource) link;
263 : };
264 :
265 : #define BDEV_CH_RESET_IN_PROGRESS (1 << 0)
266 : #define BDEV_CH_QOS_ENABLED (1 << 1)
267 :
268 : struct spdk_bdev_channel {
269 : struct spdk_bdev *bdev;
270 :
271 : /* The channel for the underlying device */
272 : struct spdk_io_channel *channel;
273 :
274 : /* Accel channel */
275 : struct spdk_io_channel *accel_channel;
276 :
277 : /* Per io_device per thread data */
278 : struct spdk_bdev_shared_resource *shared_resource;
279 :
280 : struct spdk_bdev_io_stat *stat;
281 :
282 : /*
283 : * Count of I/O submitted to the underlying dev module through this channel
284 : * and waiting for completion.
285 : */
286 : uint64_t io_outstanding;
287 :
288 : /*
289 : * List of all submitted I/Os including I/O that are generated via splitting.
290 : */
291 : bdev_io_tailq_t io_submitted;
292 :
293 : /*
294 : * List of spdk_bdev_io that are currently queued because they write to a locked
295 : * LBA range.
296 : */
297 : bdev_io_tailq_t io_locked;
298 :
299 : /* List of I/Os with accel sequence being currently executed */
300 : bdev_io_tailq_t io_accel_exec;
301 :
302 : /* List of I/Os doing memory domain pull/push */
303 : bdev_io_tailq_t io_memory_domain;
304 :
305 : uint32_t flags;
306 :
307 : /* Counts number of bdev_io in the io_submitted TAILQ */
308 : uint16_t queue_depth;
309 :
310 : uint16_t trace_id;
311 :
312 : struct spdk_histogram_data *histogram;
313 :
314 : #ifdef SPDK_CONFIG_VTUNE
315 : uint64_t start_tsc;
316 : uint64_t interval_tsc;
317 : __itt_string_handle *handle;
318 : struct spdk_bdev_io_stat *prev_stat;
319 : #endif
320 :
321 : lba_range_tailq_t locked_ranges;
322 :
323 : /** List of I/Os queued by QoS. */
324 : bdev_io_tailq_t qos_queued_io;
325 : };
326 :
327 : struct media_event_entry {
328 : struct spdk_bdev_media_event event;
329 : TAILQ_ENTRY(media_event_entry) tailq;
330 : };
331 :
332 : #define MEDIA_EVENT_POOL_SIZE 64
333 :
334 : struct spdk_bdev_desc {
335 : struct spdk_bdev *bdev;
336 : bool write;
337 : bool memory_domains_supported;
338 : bool accel_sequence_supported[SPDK_BDEV_NUM_IO_TYPES];
339 : struct spdk_bdev_open_opts opts;
340 : struct spdk_thread *thread;
341 : struct {
342 : spdk_bdev_event_cb_t event_fn;
343 : void *ctx;
344 : } callback;
345 : bool closed;
346 : struct spdk_spinlock spinlock;
347 : uint32_t refs;
348 : TAILQ_HEAD(, media_event_entry) pending_media_events;
349 : TAILQ_HEAD(, media_event_entry) free_media_events;
350 : struct media_event_entry *media_events_buffer;
351 : TAILQ_ENTRY(spdk_bdev_desc) link;
352 :
353 : uint64_t timeout_in_sec;
354 : spdk_bdev_io_timeout_cb cb_fn;
355 : void *cb_arg;
356 : struct spdk_poller *io_timeout_poller;
357 : struct spdk_bdev_module_claim *claim;
358 : };
359 :
360 : struct spdk_bdev_iostat_ctx {
361 : struct spdk_bdev_io_stat *stat;
362 : enum spdk_bdev_reset_stat_mode reset_mode;
363 : spdk_bdev_get_device_stat_cb cb;
364 : void *cb_arg;
365 : };
366 :
367 : struct set_qos_limit_ctx {
368 : void (*cb_fn)(void *cb_arg, int status);
369 : void *cb_arg;
370 : struct spdk_bdev *bdev;
371 : };
372 :
373 : struct spdk_bdev_channel_iter {
374 : spdk_bdev_for_each_channel_msg fn;
375 : spdk_bdev_for_each_channel_done cpl;
376 : struct spdk_io_channel_iter *i;
377 : void *ctx;
378 : };
379 :
380 : struct spdk_bdev_io_error_stat {
381 : uint32_t error_status[-SPDK_MIN_BDEV_IO_STATUS];
382 : };
383 :
384 : enum bdev_io_retry_state {
385 : BDEV_IO_RETRY_STATE_INVALID,
386 : BDEV_IO_RETRY_STATE_PULL,
387 : BDEV_IO_RETRY_STATE_PULL_MD,
388 : BDEV_IO_RETRY_STATE_SUBMIT,
389 : BDEV_IO_RETRY_STATE_PUSH,
390 : BDEV_IO_RETRY_STATE_PUSH_MD,
391 : BDEV_IO_RETRY_STATE_GET_ACCEL_BUF,
392 : };
393 :
394 : #define __bdev_to_io_dev(bdev) (((char *)bdev) + 1)
395 : #define __bdev_from_io_dev(io_dev) ((struct spdk_bdev *)(((char *)io_dev) - 1))
396 : #define __io_ch_to_bdev_ch(io_ch) ((struct spdk_bdev_channel *)spdk_io_channel_get_ctx(io_ch))
397 : #define __io_ch_to_bdev_mgmt_ch(io_ch) ((struct spdk_bdev_mgmt_channel *)spdk_io_channel_get_ctx(io_ch))
398 :
399 : static inline void bdev_io_complete(void *ctx);
400 : static inline void bdev_io_complete_unsubmitted(struct spdk_bdev_io *bdev_io);
401 : static void bdev_io_push_bounce_md_buf(struct spdk_bdev_io *bdev_io);
402 : static void bdev_io_push_bounce_data(struct spdk_bdev_io *bdev_io);
403 : static void _bdev_io_get_accel_buf(struct spdk_bdev_io *bdev_io);
404 :
405 : static void bdev_write_zero_buffer_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg);
406 : static int bdev_write_zero_buffer(struct spdk_bdev_io *bdev_io);
407 :
408 : static void bdev_enable_qos_msg(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
409 : struct spdk_io_channel *ch, void *_ctx);
410 : static void bdev_enable_qos_done(struct spdk_bdev *bdev, void *_ctx, int status);
411 :
412 : static int bdev_readv_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
413 : struct iovec *iov, int iovcnt, void *md_buf, uint64_t offset_blocks,
414 : uint64_t num_blocks,
415 : struct spdk_memory_domain *domain, void *domain_ctx,
416 : struct spdk_accel_sequence *seq, uint32_t dif_check_flags,
417 : spdk_bdev_io_completion_cb cb, void *cb_arg);
418 : static int bdev_writev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
419 : struct iovec *iov, int iovcnt, void *md_buf,
420 : uint64_t offset_blocks, uint64_t num_blocks,
421 : struct spdk_memory_domain *domain, void *domain_ctx,
422 : struct spdk_accel_sequence *seq, uint32_t dif_check_flags,
423 : uint32_t nvme_cdw12_raw, uint32_t nvme_cdw13_raw,
424 : spdk_bdev_io_completion_cb cb, void *cb_arg);
425 :
426 : static int bdev_lock_lba_range(struct spdk_bdev_desc *desc, struct spdk_io_channel *_ch,
427 : uint64_t offset, uint64_t length,
428 : lock_range_cb cb_fn, void *cb_arg);
429 :
430 : static int bdev_unlock_lba_range(struct spdk_bdev_desc *desc, struct spdk_io_channel *_ch,
431 : uint64_t offset, uint64_t length,
432 : lock_range_cb cb_fn, void *cb_arg);
433 :
434 : static bool bdev_abort_queued_io(bdev_io_tailq_t *queue, struct spdk_bdev_io *bio_to_abort);
435 : static bool bdev_abort_buf_io(struct spdk_bdev_mgmt_channel *ch, struct spdk_bdev_io *bio_to_abort);
436 :
437 : static bool claim_type_is_v2(enum spdk_bdev_claim_type type);
438 : static void bdev_desc_release_claims(struct spdk_bdev_desc *desc);
439 : static void claim_reset(struct spdk_bdev *bdev);
440 :
441 : static void bdev_ch_retry_io(struct spdk_bdev_channel *bdev_ch);
442 :
443 : static bool bdev_io_should_split(struct spdk_bdev_io *bdev_io);
444 :
445 : #define bdev_get_ext_io_opt(opts, field, defval) \
446 : ((opts) != NULL ? SPDK_GET_FIELD(opts, field, defval) : (defval))
447 :
448 : static inline void
449 671 : bdev_ch_add_to_io_submitted(struct spdk_bdev_io *bdev_io)
450 : {
451 671 : TAILQ_INSERT_TAIL(&bdev_io->internal.ch->io_submitted, bdev_io, internal.ch_link);
452 671 : bdev_io->internal.ch->queue_depth++;
453 671 : }
454 :
455 : static inline void
456 671 : bdev_ch_remove_from_io_submitted(struct spdk_bdev_io *bdev_io)
457 : {
458 671 : TAILQ_REMOVE(&bdev_io->internal.ch->io_submitted, bdev_io, internal.ch_link);
459 671 : bdev_io->internal.ch->queue_depth--;
460 671 : }
461 :
462 : void
463 16 : spdk_bdev_get_opts(struct spdk_bdev_opts *opts, size_t opts_size)
464 : {
465 16 : if (!opts) {
466 0 : SPDK_ERRLOG("opts should not be NULL\n");
467 0 : return;
468 : }
469 :
470 16 : if (!opts_size) {
471 0 : SPDK_ERRLOG("opts_size should not be zero value\n");
472 0 : return;
473 : }
474 :
475 16 : opts->opts_size = opts_size;
476 :
477 : #define SET_FIELD(field) \
478 : if (offsetof(struct spdk_bdev_opts, field) + sizeof(opts->field) <= opts_size) { \
479 : opts->field = g_bdev_opts.field; \
480 : } \
481 :
482 16 : SET_FIELD(bdev_io_pool_size);
483 16 : SET_FIELD(bdev_io_cache_size);
484 16 : SET_FIELD(bdev_auto_examine);
485 16 : SET_FIELD(iobuf_small_cache_size);
486 16 : SET_FIELD(iobuf_large_cache_size);
487 :
488 : /* Do not remove this statement, you should always update this statement when you adding a new field,
489 : * and do not forget to add the SET_FIELD statement for your added field. */
490 : SPDK_STATIC_ASSERT(sizeof(struct spdk_bdev_opts) == 32, "Incorrect size");
491 :
492 : #undef SET_FIELD
493 : }
494 :
495 : int
496 17 : spdk_bdev_set_opts(struct spdk_bdev_opts *opts)
497 : {
498 : uint32_t min_pool_size;
499 :
500 17 : if (!opts) {
501 0 : SPDK_ERRLOG("opts cannot be NULL\n");
502 0 : return -1;
503 : }
504 :
505 17 : if (!opts->opts_size) {
506 1 : SPDK_ERRLOG("opts_size inside opts cannot be zero value\n");
507 1 : return -1;
508 : }
509 :
510 : /*
511 : * Add 1 to the thread count to account for the extra mgmt_ch that gets created during subsystem
512 : * initialization. A second mgmt_ch will be created on the same thread when the application starts
513 : * but before the deferred put_io_channel event is executed for the first mgmt_ch.
514 : */
515 16 : min_pool_size = opts->bdev_io_cache_size * (spdk_thread_get_count() + 1);
516 16 : if (opts->bdev_io_pool_size < min_pool_size) {
517 0 : SPDK_ERRLOG("bdev_io_pool_size %" PRIu32 " is not compatible with bdev_io_cache_size %" PRIu32
518 : " and %" PRIu32 " threads\n", opts->bdev_io_pool_size, opts->bdev_io_cache_size,
519 : spdk_thread_get_count());
520 0 : SPDK_ERRLOG("bdev_io_pool_size must be at least %" PRIu32 "\n", min_pool_size);
521 0 : return -1;
522 : }
523 :
524 : #define SET_FIELD(field) \
525 : if (offsetof(struct spdk_bdev_opts, field) + sizeof(opts->field) <= opts->opts_size) { \
526 : g_bdev_opts.field = opts->field; \
527 : } \
528 :
529 16 : SET_FIELD(bdev_io_pool_size);
530 16 : SET_FIELD(bdev_io_cache_size);
531 16 : SET_FIELD(bdev_auto_examine);
532 16 : SET_FIELD(iobuf_small_cache_size);
533 16 : SET_FIELD(iobuf_large_cache_size);
534 :
535 16 : g_bdev_opts.opts_size = opts->opts_size;
536 :
537 : #undef SET_FIELD
538 :
539 16 : return 0;
540 : }
541 :
542 : static struct spdk_bdev *
543 155 : bdev_get_by_name(const char *bdev_name)
544 : {
545 155 : struct spdk_bdev_name find;
546 : struct spdk_bdev_name *res;
547 :
548 155 : find.name = (char *)bdev_name;
549 155 : res = RB_FIND(bdev_name_tree, &g_bdev_mgr.bdev_names, &find);
550 155 : if (res != NULL) {
551 148 : return res->bdev;
552 : }
553 :
554 7 : return NULL;
555 : }
556 :
557 : struct spdk_bdev *
558 19 : spdk_bdev_get_by_name(const char *bdev_name)
559 : {
560 : struct spdk_bdev *bdev;
561 :
562 19 : spdk_spin_lock(&g_bdev_mgr.spinlock);
563 19 : bdev = bdev_get_by_name(bdev_name);
564 19 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
565 :
566 19 : return bdev;
567 : }
568 :
569 : struct bdev_io_status_string {
570 : enum spdk_bdev_io_status status;
571 : const char *str;
572 : };
573 :
574 : static const struct bdev_io_status_string bdev_io_status_strings[] = {
575 : { SPDK_BDEV_IO_STATUS_AIO_ERROR, "aio_error" },
576 : { SPDK_BDEV_IO_STATUS_ABORTED, "aborted" },
577 : { SPDK_BDEV_IO_STATUS_FIRST_FUSED_FAILED, "first_fused_failed" },
578 : { SPDK_BDEV_IO_STATUS_MISCOMPARE, "miscompare" },
579 : { SPDK_BDEV_IO_STATUS_NOMEM, "nomem" },
580 : { SPDK_BDEV_IO_STATUS_SCSI_ERROR, "scsi_error" },
581 : { SPDK_BDEV_IO_STATUS_NVME_ERROR, "nvme_error" },
582 : { SPDK_BDEV_IO_STATUS_FAILED, "failed" },
583 : { SPDK_BDEV_IO_STATUS_PENDING, "pending" },
584 : { SPDK_BDEV_IO_STATUS_SUCCESS, "success" },
585 : };
586 :
587 : static const char *
588 0 : bdev_io_status_get_string(enum spdk_bdev_io_status status)
589 : {
590 : uint32_t i;
591 :
592 0 : for (i = 0; i < SPDK_COUNTOF(bdev_io_status_strings); i++) {
593 0 : if (bdev_io_status_strings[i].status == status) {
594 0 : return bdev_io_status_strings[i].str;
595 : }
596 : }
597 :
598 0 : return "reserved";
599 : }
600 :
601 : struct spdk_bdev_wait_for_examine_ctx {
602 : struct spdk_poller *poller;
603 : spdk_bdev_wait_for_examine_cb cb_fn;
604 : void *cb_arg;
605 : };
606 :
607 : static bool bdev_module_all_actions_completed(void);
608 :
609 : static int
610 203 : bdev_wait_for_examine_cb(void *arg)
611 : {
612 203 : struct spdk_bdev_wait_for_examine_ctx *ctx = arg;
613 :
614 203 : if (!bdev_module_all_actions_completed()) {
615 0 : return SPDK_POLLER_IDLE;
616 : }
617 :
618 203 : spdk_poller_unregister(&ctx->poller);
619 203 : ctx->cb_fn(ctx->cb_arg);
620 203 : free(ctx);
621 :
622 203 : return SPDK_POLLER_BUSY;
623 : }
624 :
625 : int
626 203 : spdk_bdev_wait_for_examine(spdk_bdev_wait_for_examine_cb cb_fn, void *cb_arg)
627 : {
628 : struct spdk_bdev_wait_for_examine_ctx *ctx;
629 :
630 203 : ctx = calloc(1, sizeof(*ctx));
631 203 : if (ctx == NULL) {
632 0 : return -ENOMEM;
633 : }
634 203 : ctx->cb_fn = cb_fn;
635 203 : ctx->cb_arg = cb_arg;
636 203 : ctx->poller = SPDK_POLLER_REGISTER(bdev_wait_for_examine_cb, ctx, 0);
637 :
638 203 : return 0;
639 : }
640 :
641 : struct spdk_bdev_examine_item {
642 : char *name;
643 : TAILQ_ENTRY(spdk_bdev_examine_item) link;
644 : };
645 :
646 : TAILQ_HEAD(spdk_bdev_examine_allowlist, spdk_bdev_examine_item);
647 :
648 : struct spdk_bdev_examine_allowlist g_bdev_examine_allowlist = TAILQ_HEAD_INITIALIZER(
649 : g_bdev_examine_allowlist);
650 :
651 : static inline bool
652 24 : bdev_examine_allowlist_check(const char *name)
653 : {
654 : struct spdk_bdev_examine_item *item;
655 24 : TAILQ_FOREACH(item, &g_bdev_examine_allowlist, link) {
656 3 : if (strcmp(name, item->name) == 0) {
657 3 : return true;
658 : }
659 : }
660 21 : return false;
661 : }
662 :
663 : static inline void
664 258 : bdev_examine_allowlist_remove(const char *name)
665 : {
666 : struct spdk_bdev_examine_item *item;
667 258 : TAILQ_FOREACH(item, &g_bdev_examine_allowlist, link) {
668 3 : if (strcmp(name, item->name) == 0) {
669 3 : TAILQ_REMOVE(&g_bdev_examine_allowlist, item, link);
670 3 : free(item->name);
671 3 : free(item);
672 3 : break;
673 : }
674 : }
675 258 : }
676 :
677 : static inline void
678 68 : bdev_examine_allowlist_free(void)
679 : {
680 : struct spdk_bdev_examine_item *item;
681 68 : while (!TAILQ_EMPTY(&g_bdev_examine_allowlist)) {
682 0 : item = TAILQ_FIRST(&g_bdev_examine_allowlist);
683 0 : TAILQ_REMOVE(&g_bdev_examine_allowlist, item, link);
684 0 : free(item->name);
685 0 : free(item);
686 : }
687 68 : }
688 :
689 : static inline bool
690 12 : bdev_in_examine_allowlist(struct spdk_bdev *bdev)
691 : {
692 : struct spdk_bdev_alias *tmp;
693 12 : if (bdev_examine_allowlist_check(bdev->name)) {
694 3 : return true;
695 : }
696 18 : TAILQ_FOREACH(tmp, &bdev->aliases, tailq) {
697 9 : if (bdev_examine_allowlist_check(tmp->alias.name)) {
698 0 : return true;
699 : }
700 : }
701 9 : return false;
702 : }
703 :
704 : static inline bool
705 133 : bdev_ok_to_examine(struct spdk_bdev *bdev)
706 : {
707 : /* Some bdevs may not support the READ command.
708 : * Do not try to examine them.
709 : */
710 133 : if (!spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_READ)) {
711 0 : return false;
712 : }
713 :
714 133 : if (g_bdev_opts.bdev_auto_examine) {
715 121 : return true;
716 : } else {
717 12 : return bdev_in_examine_allowlist(bdev);
718 : }
719 : }
720 :
721 : static void
722 133 : bdev_examine(struct spdk_bdev *bdev)
723 : {
724 : struct spdk_bdev_module *module;
725 : struct spdk_bdev_module_claim *claim, *tmpclaim;
726 : uint32_t action;
727 :
728 133 : if (!bdev_ok_to_examine(bdev)) {
729 9 : return;
730 : }
731 :
732 506 : TAILQ_FOREACH(module, &g_bdev_mgr.bdev_modules, internal.tailq) {
733 382 : if (module->examine_config) {
734 258 : spdk_spin_lock(&module->internal.spinlock);
735 258 : action = module->internal.action_in_progress;
736 258 : module->internal.action_in_progress++;
737 258 : spdk_spin_unlock(&module->internal.spinlock);
738 258 : module->examine_config(bdev);
739 258 : if (action != module->internal.action_in_progress) {
740 0 : SPDK_ERRLOG("examine_config for module %s did not call "
741 : "spdk_bdev_module_examine_done()\n", module->name);
742 : }
743 : }
744 : }
745 :
746 124 : spdk_spin_lock(&bdev->internal.spinlock);
747 :
748 124 : switch (bdev->internal.claim_type) {
749 116 : case SPDK_BDEV_CLAIM_NONE:
750 : /* Examine by all bdev modules */
751 466 : TAILQ_FOREACH(module, &g_bdev_mgr.bdev_modules, internal.tailq) {
752 350 : if (module->examine_disk) {
753 225 : spdk_spin_lock(&module->internal.spinlock);
754 225 : module->internal.action_in_progress++;
755 225 : spdk_spin_unlock(&module->internal.spinlock);
756 225 : spdk_spin_unlock(&bdev->internal.spinlock);
757 225 : module->examine_disk(bdev);
758 225 : spdk_spin_lock(&bdev->internal.spinlock);
759 : }
760 : }
761 116 : break;
762 1 : case SPDK_BDEV_CLAIM_EXCL_WRITE:
763 : /* Examine by the one bdev module with a v1 claim */
764 1 : module = bdev->internal.claim.v1.module;
765 1 : if (module->examine_disk) {
766 1 : spdk_spin_lock(&module->internal.spinlock);
767 1 : module->internal.action_in_progress++;
768 1 : spdk_spin_unlock(&module->internal.spinlock);
769 1 : spdk_spin_unlock(&bdev->internal.spinlock);
770 1 : module->examine_disk(bdev);
771 1 : return;
772 : }
773 0 : break;
774 7 : default:
775 : /* Examine by all bdev modules with a v2 claim */
776 7 : assert(claim_type_is_v2(bdev->internal.claim_type));
777 : /*
778 : * Removal of tailq nodes while iterating can cause the iteration to jump out of the
779 : * list, perhaps accessing freed memory. Without protection, this could happen
780 : * while the lock is dropped during the examine callback.
781 : */
782 7 : bdev->internal.examine_in_progress++;
783 :
784 16 : TAILQ_FOREACH(claim, &bdev->internal.claim.v2.claims, link) {
785 9 : module = claim->module;
786 :
787 9 : if (module == NULL) {
788 : /* This is a vestigial claim, held by examine_count */
789 0 : continue;
790 : }
791 :
792 9 : if (module->examine_disk == NULL) {
793 0 : continue;
794 : }
795 :
796 9 : spdk_spin_lock(&module->internal.spinlock);
797 9 : module->internal.action_in_progress++;
798 9 : spdk_spin_unlock(&module->internal.spinlock);
799 :
800 : /* Call examine_disk without holding internal.spinlock. */
801 9 : spdk_spin_unlock(&bdev->internal.spinlock);
802 9 : module->examine_disk(bdev);
803 9 : spdk_spin_lock(&bdev->internal.spinlock);
804 : }
805 :
806 7 : assert(bdev->internal.examine_in_progress > 0);
807 7 : bdev->internal.examine_in_progress--;
808 7 : if (bdev->internal.examine_in_progress == 0) {
809 : /* Remove any claims that were released during examine_disk */
810 16 : TAILQ_FOREACH_SAFE(claim, &bdev->internal.claim.v2.claims, link, tmpclaim) {
811 9 : if (claim->desc != NULL) {
812 9 : continue;
813 : }
814 :
815 0 : TAILQ_REMOVE(&bdev->internal.claim.v2.claims, claim, link);
816 0 : free(claim);
817 : }
818 7 : if (TAILQ_EMPTY(&bdev->internal.claim.v2.claims)) {
819 0 : claim_reset(bdev);
820 : }
821 : }
822 : }
823 :
824 123 : spdk_spin_unlock(&bdev->internal.spinlock);
825 : }
826 :
827 : int
828 4 : spdk_bdev_examine(const char *name)
829 : {
830 : struct spdk_bdev *bdev;
831 : struct spdk_bdev_examine_item *item;
832 4 : struct spdk_thread *thread = spdk_get_thread();
833 :
834 4 : if (spdk_unlikely(!spdk_thread_is_app_thread(thread))) {
835 1 : SPDK_ERRLOG("Cannot examine bdev %s on thread %p (%s)\n", name, thread,
836 : thread ? spdk_thread_get_name(thread) : "null");
837 1 : return -EINVAL;
838 : }
839 :
840 3 : if (g_bdev_opts.bdev_auto_examine) {
841 0 : SPDK_ERRLOG("Manual examine is not allowed if auto examine is enabled\n");
842 0 : return -EINVAL;
843 : }
844 :
845 3 : if (bdev_examine_allowlist_check(name)) {
846 0 : SPDK_ERRLOG("Duplicate bdev name for manual examine: %s\n", name);
847 0 : return -EEXIST;
848 : }
849 :
850 3 : item = calloc(1, sizeof(*item));
851 3 : if (!item) {
852 0 : return -ENOMEM;
853 : }
854 3 : item->name = strdup(name);
855 3 : if (!item->name) {
856 0 : free(item);
857 0 : return -ENOMEM;
858 : }
859 3 : TAILQ_INSERT_TAIL(&g_bdev_examine_allowlist, item, link);
860 :
861 3 : bdev = spdk_bdev_get_by_name(name);
862 3 : if (bdev) {
863 3 : bdev_examine(bdev);
864 : }
865 3 : return 0;
866 : }
867 :
868 : static inline void
869 0 : bdev_examine_allowlist_config_json(struct spdk_json_write_ctx *w)
870 : {
871 : struct spdk_bdev_examine_item *item;
872 0 : TAILQ_FOREACH(item, &g_bdev_examine_allowlist, link) {
873 0 : spdk_json_write_object_begin(w);
874 0 : spdk_json_write_named_string(w, "method", "bdev_examine");
875 0 : spdk_json_write_named_object_begin(w, "params");
876 0 : spdk_json_write_named_string(w, "name", item->name);
877 0 : spdk_json_write_object_end(w);
878 0 : spdk_json_write_object_end(w);
879 : }
880 0 : }
881 :
882 : struct spdk_bdev *
883 1 : spdk_bdev_first(void)
884 : {
885 : struct spdk_bdev *bdev;
886 :
887 1 : bdev = TAILQ_FIRST(&g_bdev_mgr.bdevs);
888 1 : if (bdev) {
889 1 : SPDK_DEBUGLOG(bdev, "Starting bdev iteration at %s\n", bdev->name);
890 : }
891 :
892 1 : return bdev;
893 : }
894 :
895 : struct spdk_bdev *
896 8 : spdk_bdev_next(struct spdk_bdev *prev)
897 : {
898 : struct spdk_bdev *bdev;
899 :
900 8 : bdev = TAILQ_NEXT(prev, internal.link);
901 8 : if (bdev) {
902 7 : SPDK_DEBUGLOG(bdev, "Continuing bdev iteration at %s\n", bdev->name);
903 : }
904 :
905 8 : return bdev;
906 : }
907 :
908 : static struct spdk_bdev *
909 6 : _bdev_next_leaf(struct spdk_bdev *bdev)
910 : {
911 9 : while (bdev != NULL) {
912 8 : if (bdev->internal.claim_type == SPDK_BDEV_CLAIM_NONE) {
913 5 : return bdev;
914 : } else {
915 3 : bdev = TAILQ_NEXT(bdev, internal.link);
916 : }
917 : }
918 :
919 1 : return bdev;
920 : }
921 :
922 : struct spdk_bdev *
923 1 : spdk_bdev_first_leaf(void)
924 : {
925 : struct spdk_bdev *bdev;
926 :
927 1 : bdev = _bdev_next_leaf(TAILQ_FIRST(&g_bdev_mgr.bdevs));
928 :
929 1 : if (bdev) {
930 1 : SPDK_DEBUGLOG(bdev, "Starting bdev iteration at %s\n", bdev->name);
931 : }
932 :
933 1 : return bdev;
934 : }
935 :
936 : struct spdk_bdev *
937 5 : spdk_bdev_next_leaf(struct spdk_bdev *prev)
938 : {
939 : struct spdk_bdev *bdev;
940 :
941 5 : bdev = _bdev_next_leaf(TAILQ_NEXT(prev, internal.link));
942 :
943 5 : if (bdev) {
944 4 : SPDK_DEBUGLOG(bdev, "Continuing bdev iteration at %s\n", bdev->name);
945 : }
946 :
947 5 : return bdev;
948 : }
949 :
950 : static inline bool
951 820 : bdev_io_use_memory_domain(struct spdk_bdev_io *bdev_io)
952 : {
953 820 : return bdev_io->internal.f.has_memory_domain;
954 : }
955 :
956 : static inline bool
957 1551 : bdev_io_use_accel_sequence(struct spdk_bdev_io *bdev_io)
958 : {
959 1551 : return bdev_io->internal.f.has_accel_sequence;
960 : }
961 :
962 : static inline uint32_t
963 373 : bdev_desc_get_block_size(struct spdk_bdev_desc *desc)
964 : {
965 373 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
966 :
967 373 : if (spdk_unlikely(desc->opts.hide_metadata)) {
968 0 : return bdev->blocklen - bdev->md_len;
969 : } else {
970 373 : return bdev->blocklen;
971 : }
972 : }
973 :
974 : static inline uint32_t
975 110 : bdev_io_get_block_size(struct spdk_bdev_io *bdev_io)
976 : {
977 110 : struct spdk_bdev *bdev = bdev_io->bdev;
978 :
979 110 : if (bdev_io->u.bdev.dif_check_flags & SPDK_DIF_FLAGS_NVME_PRACT) {
980 0 : if (bdev->md_len == spdk_dif_pi_format_get_size(bdev->dif_pi_format)) {
981 0 : return bdev->blocklen - bdev->md_len;
982 : } else {
983 0 : return bdev->blocklen;
984 : }
985 : }
986 :
987 110 : return bdev_desc_get_block_size(bdev_io->internal.desc);
988 : }
989 :
990 : static inline void
991 7 : bdev_queue_nomem_io_head(struct spdk_bdev_shared_resource *shared_resource,
992 : struct spdk_bdev_io *bdev_io, enum bdev_io_retry_state state)
993 : {
994 : /* Wait for some of the outstanding I/O to complete before we retry any of the nomem_io.
995 : * Normally we will wait for NOMEM_THRESHOLD_COUNT I/O to complete but for low queue depth
996 : * channels we will instead wait for half to complete.
997 : */
998 7 : shared_resource->nomem_threshold = spdk_max((int64_t)shared_resource->io_outstanding / 2,
999 : (int64_t)shared_resource->io_outstanding - NOMEM_THRESHOLD_COUNT);
1000 :
1001 7 : assert(state != BDEV_IO_RETRY_STATE_INVALID);
1002 7 : bdev_io->internal.retry_state = state;
1003 7 : TAILQ_INSERT_HEAD(&shared_resource->nomem_io, bdev_io, internal.link);
1004 7 : }
1005 :
1006 : static inline void
1007 43 : bdev_queue_nomem_io_tail(struct spdk_bdev_shared_resource *shared_resource,
1008 : struct spdk_bdev_io *bdev_io, enum bdev_io_retry_state state)
1009 : {
1010 : /* We only queue IOs at the end of the nomem_io queue if they're submitted by the user while
1011 : * the queue isn't empty, so we don't need to update the nomem_threshold here */
1012 43 : assert(!TAILQ_EMPTY(&shared_resource->nomem_io));
1013 :
1014 43 : assert(state != BDEV_IO_RETRY_STATE_INVALID);
1015 43 : bdev_io->internal.retry_state = state;
1016 43 : TAILQ_INSERT_TAIL(&shared_resource->nomem_io, bdev_io, internal.link);
1017 43 : }
1018 :
1019 : void
1020 16 : spdk_bdev_io_set_buf(struct spdk_bdev_io *bdev_io, void *buf, size_t len)
1021 : {
1022 : struct iovec *iovs;
1023 :
1024 16 : if (bdev_io->u.bdev.iovs == NULL) {
1025 3 : bdev_io->u.bdev.iovs = &bdev_io->iov;
1026 3 : bdev_io->u.bdev.iovcnt = 1;
1027 : }
1028 :
1029 16 : iovs = bdev_io->u.bdev.iovs;
1030 :
1031 16 : assert(iovs != NULL);
1032 16 : assert(bdev_io->u.bdev.iovcnt >= 1);
1033 :
1034 16 : iovs[0].iov_base = buf;
1035 16 : iovs[0].iov_len = len;
1036 16 : }
1037 :
1038 : void
1039 3 : spdk_bdev_io_set_md_buf(struct spdk_bdev_io *bdev_io, void *md_buf, size_t len)
1040 : {
1041 3 : assert((len / spdk_bdev_get_md_size(bdev_io->bdev)) >= bdev_io->u.bdev.num_blocks);
1042 3 : bdev_io->u.bdev.md_buf = md_buf;
1043 3 : }
1044 :
1045 : static bool
1046 167 : _is_buf_allocated(const struct iovec *iovs)
1047 : {
1048 167 : if (iovs == NULL) {
1049 6 : return false;
1050 : }
1051 :
1052 161 : return iovs[0].iov_base != NULL;
1053 : }
1054 :
1055 : static bool
1056 50 : _are_iovs_aligned(struct iovec *iovs, int iovcnt, uint32_t alignment)
1057 : {
1058 : int i;
1059 : uintptr_t iov_base;
1060 :
1061 50 : if (spdk_likely(alignment == 1)) {
1062 21 : return true;
1063 : }
1064 :
1065 36 : for (i = 0; i < iovcnt; i++) {
1066 29 : iov_base = (uintptr_t)iovs[i].iov_base;
1067 29 : if ((iov_base & (alignment - 1)) != 0) {
1068 22 : return false;
1069 : }
1070 : }
1071 :
1072 7 : return true;
1073 : }
1074 :
1075 : static inline bool
1076 895 : bdev_io_needs_metadata(struct spdk_bdev_desc *desc, struct spdk_bdev_io *bdev_io)
1077 : {
1078 1048 : return (bdev_io->bdev->md_len != 0) &&
1079 153 : (desc->opts.hide_metadata ||
1080 153 : (bdev_io->u.bdev.dif_check_flags & SPDK_DIF_FLAGS_NVME_PRACT));
1081 : }
1082 :
1083 : static inline bool
1084 852 : bdev_io_needs_sequence_exec(struct spdk_bdev_desc *desc, struct spdk_bdev_io *bdev_io)
1085 : {
1086 852 : if (!bdev_io_use_accel_sequence(bdev_io)) {
1087 852 : return false;
1088 : }
1089 :
1090 : /* For now, we don't allow splitting IOs with an accel sequence and will treat them as if
1091 : * bdev module didn't support accel sequences */
1092 0 : return !desc->accel_sequence_supported[bdev_io->type] || bdev_io->internal.f.split;
1093 : }
1094 :
1095 : static inline void
1096 592 : bdev_io_increment_outstanding(struct spdk_bdev_channel *bdev_ch,
1097 : struct spdk_bdev_shared_resource *shared_resource)
1098 : {
1099 592 : bdev_ch->io_outstanding++;
1100 592 : shared_resource->io_outstanding++;
1101 592 : }
1102 :
1103 : static inline void
1104 592 : bdev_io_decrement_outstanding(struct spdk_bdev_channel *bdev_ch,
1105 : struct spdk_bdev_shared_resource *shared_resource)
1106 : {
1107 592 : assert(bdev_ch->io_outstanding > 0);
1108 592 : assert(shared_resource->io_outstanding > 0);
1109 592 : bdev_ch->io_outstanding--;
1110 592 : shared_resource->io_outstanding--;
1111 592 : }
1112 :
1113 : static void
1114 0 : bdev_io_submit_sequence_cb(void *ctx, int status)
1115 : {
1116 0 : struct spdk_bdev_io *bdev_io = ctx;
1117 :
1118 0 : assert(bdev_io_use_accel_sequence(bdev_io));
1119 :
1120 0 : bdev_io->u.bdev.accel_sequence = NULL;
1121 0 : bdev_io->internal.f.has_accel_sequence = false;
1122 :
1123 0 : if (spdk_unlikely(status != 0)) {
1124 0 : SPDK_ERRLOG("Failed to execute accel sequence, status=%d\n", status);
1125 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
1126 0 : bdev_io_complete_unsubmitted(bdev_io);
1127 0 : return;
1128 : }
1129 :
1130 0 : bdev_io_submit(bdev_io);
1131 : }
1132 :
1133 : static void
1134 0 : bdev_io_exec_sequence_cb(void *ctx, int status)
1135 : {
1136 0 : struct spdk_bdev_io *bdev_io = ctx;
1137 0 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1138 :
1139 0 : TAILQ_REMOVE(&bdev_io->internal.ch->io_accel_exec, bdev_io, internal.link);
1140 0 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1141 :
1142 0 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1143 0 : bdev_ch_retry_io(ch);
1144 : }
1145 :
1146 0 : bdev_io->internal.data_transfer_cpl(bdev_io, status);
1147 0 : }
1148 :
1149 : static void
1150 0 : bdev_io_exec_sequence(struct spdk_bdev_io *bdev_io, void (*cb_fn)(void *ctx, int status))
1151 : {
1152 0 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1153 :
1154 0 : assert(bdev_io_needs_sequence_exec(bdev_io->internal.desc, bdev_io));
1155 0 : assert(bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE || bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
1156 0 : assert(bdev_io_use_accel_sequence(bdev_io));
1157 :
1158 : /* Since the operations are appended during submission, they're in the opposite order than
1159 : * how we want to execute them for reads (i.e. we need to execute the most recently added
1160 : * operation first), so reverse the sequence before executing it.
1161 : */
1162 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_READ) {
1163 0 : spdk_accel_sequence_reverse(bdev_io->internal.accel_sequence);
1164 : }
1165 :
1166 0 : TAILQ_INSERT_TAIL(&bdev_io->internal.ch->io_accel_exec, bdev_io, internal.link);
1167 0 : bdev_io_increment_outstanding(ch, ch->shared_resource);
1168 0 : bdev_io->internal.data_transfer_cpl = cb_fn;
1169 :
1170 0 : spdk_accel_sequence_finish(bdev_io->internal.accel_sequence,
1171 : bdev_io_exec_sequence_cb, bdev_io);
1172 0 : }
1173 :
1174 : static void
1175 42 : bdev_io_get_buf_complete(struct spdk_bdev_io *bdev_io, bool status)
1176 : {
1177 42 : struct spdk_io_channel *ch = spdk_bdev_io_get_io_channel(bdev_io);
1178 : void *buf;
1179 :
1180 42 : if (spdk_unlikely(bdev_io->internal.get_aux_buf_cb != NULL)) {
1181 0 : buf = bdev_io->internal.buf.ptr;
1182 0 : bdev_io->internal.buf.ptr = NULL;
1183 0 : bdev_io->internal.f.has_buf = false;
1184 0 : bdev_io->internal.get_aux_buf_cb(ch, bdev_io, buf);
1185 0 : bdev_io->internal.get_aux_buf_cb = NULL;
1186 : } else {
1187 42 : assert(bdev_io->internal.get_buf_cb != NULL);
1188 42 : bdev_io->internal.get_buf_cb(ch, bdev_io, status);
1189 42 : bdev_io->internal.get_buf_cb = NULL;
1190 : }
1191 42 : }
1192 :
1193 : static void
1194 4 : _bdev_io_pull_buffer_cpl(void *ctx, int rc)
1195 : {
1196 4 : struct spdk_bdev_io *bdev_io = ctx;
1197 :
1198 4 : if (rc) {
1199 0 : SPDK_ERRLOG("Set bounce buffer failed with rc %d\n", rc);
1200 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
1201 : }
1202 4 : bdev_io_get_buf_complete(bdev_io, !rc);
1203 4 : }
1204 :
1205 : static void
1206 2 : bdev_io_pull_md_buf_done(void *ctx, int status)
1207 : {
1208 2 : struct spdk_bdev_io *bdev_io = ctx;
1209 2 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1210 :
1211 2 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1212 2 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1213 :
1214 2 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1215 0 : bdev_ch_retry_io(ch);
1216 : }
1217 :
1218 2 : assert(bdev_io->internal.data_transfer_cpl);
1219 2 : bdev_io->internal.data_transfer_cpl(bdev_io, status);
1220 2 : }
1221 :
1222 : static void
1223 4 : bdev_io_pull_md_buf(struct spdk_bdev_io *bdev_io)
1224 : {
1225 4 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1226 4 : int rc = 0;
1227 :
1228 4 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
1229 2 : assert(bdev_io->internal.f.has_bounce_buf);
1230 2 : if (bdev_io_use_memory_domain(bdev_io)) {
1231 2 : TAILQ_INSERT_TAIL(&ch->io_memory_domain, bdev_io, internal.link);
1232 2 : bdev_io_increment_outstanding(ch, ch->shared_resource);
1233 2 : rc = spdk_memory_domain_pull_data(bdev_io->internal.memory_domain,
1234 : bdev_io->internal.memory_domain_ctx,
1235 : &bdev_io->internal.bounce_buf.orig_md_iov, 1,
1236 : &bdev_io->internal.bounce_buf.md_iov, 1,
1237 : bdev_io_pull_md_buf_done, bdev_io);
1238 2 : if (rc == 0) {
1239 : /* Continue to submit IO in completion callback */
1240 2 : return;
1241 : }
1242 0 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1243 0 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1244 0 : if (rc != -ENOMEM) {
1245 0 : SPDK_ERRLOG("Failed to pull data from memory domain %s, rc %d\n",
1246 : spdk_memory_domain_get_dma_device_id(
1247 : bdev_io->internal.memory_domain), rc);
1248 : }
1249 : } else {
1250 0 : memcpy(bdev_io->internal.bounce_buf.md_iov.iov_base,
1251 0 : bdev_io->internal.bounce_buf.orig_md_iov.iov_base,
1252 : bdev_io->internal.bounce_buf.orig_md_iov.iov_len);
1253 : }
1254 : }
1255 :
1256 2 : if (spdk_unlikely(rc == -ENOMEM)) {
1257 0 : bdev_queue_nomem_io_head(ch->shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PULL_MD);
1258 : } else {
1259 2 : assert(bdev_io->internal.data_transfer_cpl);
1260 2 : bdev_io->internal.data_transfer_cpl(bdev_io, rc);
1261 : }
1262 : }
1263 :
1264 : static void
1265 4 : _bdev_io_pull_bounce_md_buf(struct spdk_bdev_io *bdev_io, void *md_buf, size_t len)
1266 : {
1267 4 : assert(bdev_io->internal.f.has_bounce_buf);
1268 :
1269 : /* save original md_buf */
1270 4 : bdev_io->internal.bounce_buf.orig_md_iov.iov_base = bdev_io->u.bdev.md_buf;
1271 4 : bdev_io->internal.bounce_buf.orig_md_iov.iov_len = len;
1272 4 : bdev_io->internal.bounce_buf.md_iov.iov_base = md_buf;
1273 4 : bdev_io->internal.bounce_buf.md_iov.iov_len = len;
1274 : /* set bounce md_buf */
1275 4 : bdev_io->u.bdev.md_buf = md_buf;
1276 :
1277 4 : bdev_io_pull_md_buf(bdev_io);
1278 4 : }
1279 :
1280 : static void
1281 42 : _bdev_io_set_md_buf(struct spdk_bdev_io *bdev_io)
1282 : {
1283 42 : struct spdk_bdev *bdev = bdev_io->bdev;
1284 : uint64_t md_len;
1285 : void *buf;
1286 :
1287 42 : if (spdk_bdev_is_md_separate(bdev)) {
1288 7 : assert(!bdev_io_use_accel_sequence(bdev_io));
1289 :
1290 7 : buf = (char *)bdev_io->u.bdev.iovs[0].iov_base + bdev_io->u.bdev.iovs[0].iov_len;
1291 7 : md_len = bdev_io->u.bdev.num_blocks * bdev->md_len;
1292 :
1293 7 : assert(((uintptr_t)buf & (spdk_bdev_get_buf_align(bdev) - 1)) == 0);
1294 :
1295 7 : if (bdev_io->u.bdev.md_buf != NULL) {
1296 4 : _bdev_io_pull_bounce_md_buf(bdev_io, buf, md_len);
1297 4 : return;
1298 : } else {
1299 3 : spdk_bdev_io_set_md_buf(bdev_io, buf, md_len);
1300 : }
1301 : }
1302 :
1303 38 : bdev_io_get_buf_complete(bdev_io, true);
1304 : }
1305 :
1306 : static inline void
1307 26 : bdev_io_pull_data_done(struct spdk_bdev_io *bdev_io, int rc)
1308 : {
1309 26 : if (rc) {
1310 0 : SPDK_ERRLOG("Failed to get data buffer\n");
1311 0 : assert(bdev_io->internal.data_transfer_cpl);
1312 0 : bdev_io->internal.data_transfer_cpl(bdev_io, rc);
1313 0 : return;
1314 : }
1315 :
1316 26 : _bdev_io_set_md_buf(bdev_io);
1317 : }
1318 :
1319 : static void
1320 2 : bdev_io_pull_data_done_and_track(void *ctx, int status)
1321 : {
1322 2 : struct spdk_bdev_io *bdev_io = ctx;
1323 2 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1324 :
1325 2 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1326 2 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1327 :
1328 2 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1329 0 : bdev_ch_retry_io(ch);
1330 : }
1331 :
1332 2 : bdev_io_pull_data_done(bdev_io, status);
1333 2 : }
1334 :
1335 : static void
1336 27 : bdev_io_pull_data(struct spdk_bdev_io *bdev_io)
1337 : {
1338 27 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1339 27 : struct spdk_bdev_desc *desc = bdev_io->internal.desc;
1340 27 : int rc = 0;
1341 :
1342 27 : assert(bdev_io->internal.f.has_bounce_buf);
1343 :
1344 27 : if (bdev_io_needs_metadata(desc, bdev_io)) {
1345 0 : assert(bdev_io->bdev->md_interleave);
1346 :
1347 0 : bdev_io->u.bdev.dif_check_flags &= ~SPDK_DIF_FLAGS_NVME_PRACT;
1348 :
1349 0 : if (!bdev_io_use_accel_sequence(bdev_io)) {
1350 0 : bdev_io->internal.accel_sequence = NULL;
1351 : }
1352 :
1353 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
1354 0 : rc = spdk_accel_append_dif_generate_copy(&bdev_io->internal.accel_sequence, ch->accel_channel,
1355 0 : bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt,
1356 : bdev_io->u.bdev.memory_domain,
1357 : bdev_io->u.bdev.memory_domain_ctx,
1358 : bdev_io->internal.bounce_buf.orig_iovs,
1359 0 : bdev_io->internal.bounce_buf.orig_iovcnt,
1360 0 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain : NULL,
1361 0 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain_ctx : NULL,
1362 0 : bdev_io->u.bdev.num_blocks,
1363 0 : &bdev_io->u.bdev.dif_ctx,
1364 : NULL, NULL);
1365 : } else {
1366 0 : assert(bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
1367 0 : rc = spdk_accel_append_dif_verify_copy(&bdev_io->internal.accel_sequence, ch->accel_channel,
1368 : bdev_io->internal.bounce_buf.orig_iovs,
1369 0 : bdev_io->internal.bounce_buf.orig_iovcnt,
1370 0 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain : NULL,
1371 0 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain_ctx : NULL,
1372 0 : bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt,
1373 : bdev_io->u.bdev.memory_domain,
1374 : bdev_io->u.bdev.memory_domain_ctx,
1375 0 : bdev_io->u.bdev.num_blocks,
1376 0 : &bdev_io->u.bdev.dif_ctx,
1377 : &bdev_io->u.bdev.dif_err,
1378 : NULL, NULL);
1379 : }
1380 :
1381 0 : if (spdk_likely(rc == 0)) {
1382 0 : bdev_io->internal.f.has_accel_sequence = true;
1383 0 : bdev_io->u.bdev.accel_sequence = bdev_io->internal.accel_sequence;
1384 0 : } else if (rc != -ENOMEM) {
1385 0 : SPDK_ERRLOG("Failed to append generate/verify_copy to accel sequence: %p\n",
1386 : bdev_io->internal.accel_sequence);
1387 : }
1388 54 : } else if (bdev_io_needs_sequence_exec(desc, bdev_io) ||
1389 27 : (bdev_io_use_accel_sequence(bdev_io) && bdev_io_use_memory_domain(bdev_io))) {
1390 : /* If we need to exec an accel sequence or the IO uses a memory domain buffer and has a
1391 : * sequence, append a copy operation making accel change the src/dst buffers of the previous
1392 : * operation */
1393 0 : assert(bdev_io_use_accel_sequence(bdev_io));
1394 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
1395 0 : rc = spdk_accel_append_copy(&bdev_io->internal.accel_sequence, ch->accel_channel,
1396 0 : bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt,
1397 : NULL, NULL,
1398 : bdev_io->internal.bounce_buf.orig_iovs,
1399 0 : bdev_io->internal.bounce_buf.orig_iovcnt,
1400 0 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain : NULL,
1401 0 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain_ctx : NULL,
1402 : NULL, NULL);
1403 : } else {
1404 : /* We need to reverse the src/dst for reads */
1405 0 : assert(bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
1406 0 : rc = spdk_accel_append_copy(&bdev_io->internal.accel_sequence, ch->accel_channel,
1407 : bdev_io->internal.bounce_buf.orig_iovs,
1408 0 : bdev_io->internal.bounce_buf.orig_iovcnt,
1409 0 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain : NULL,
1410 0 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain_ctx : NULL,
1411 0 : bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt,
1412 : NULL, NULL, NULL, NULL);
1413 : }
1414 :
1415 0 : if (spdk_unlikely(rc != 0 && rc != -ENOMEM)) {
1416 0 : SPDK_ERRLOG("Failed to append copy to accel sequence: %p\n",
1417 : bdev_io->internal.accel_sequence);
1418 : }
1419 27 : } else if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
1420 : /* if this is write path, copy data from original buffer to bounce buffer */
1421 17 : if (bdev_io_use_memory_domain(bdev_io)) {
1422 3 : TAILQ_INSERT_TAIL(&ch->io_memory_domain, bdev_io, internal.link);
1423 3 : bdev_io_increment_outstanding(ch, ch->shared_resource);
1424 6 : rc = spdk_memory_domain_pull_data(bdev_io->internal.memory_domain,
1425 : bdev_io->internal.memory_domain_ctx,
1426 : bdev_io->internal.bounce_buf.orig_iovs,
1427 3 : (uint32_t)bdev_io->internal.bounce_buf.orig_iovcnt,
1428 : bdev_io->u.bdev.iovs, 1,
1429 : bdev_io_pull_data_done_and_track,
1430 : bdev_io);
1431 3 : if (rc == 0) {
1432 : /* Continue to submit IO in completion callback */
1433 2 : return;
1434 : }
1435 1 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1436 1 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1437 1 : if (rc != -ENOMEM) {
1438 0 : SPDK_ERRLOG("Failed to pull data from memory domain %s\n",
1439 : spdk_memory_domain_get_dma_device_id(
1440 : bdev_io->internal.memory_domain));
1441 : }
1442 : } else {
1443 14 : assert(bdev_io->u.bdev.iovcnt == 1);
1444 28 : spdk_copy_iovs_to_buf(bdev_io->u.bdev.iovs[0].iov_base,
1445 14 : bdev_io->u.bdev.iovs[0].iov_len,
1446 : bdev_io->internal.bounce_buf.orig_iovs,
1447 : bdev_io->internal.bounce_buf.orig_iovcnt);
1448 : }
1449 : }
1450 :
1451 25 : if (spdk_unlikely(rc == -ENOMEM)) {
1452 1 : bdev_queue_nomem_io_head(ch->shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PULL);
1453 : } else {
1454 24 : bdev_io_pull_data_done(bdev_io, rc);
1455 : }
1456 : }
1457 :
1458 : static void
1459 26 : _bdev_io_pull_bounce_data_buf(struct spdk_bdev_io *bdev_io, void *buf, size_t len,
1460 : bdev_copy_bounce_buffer_cpl cpl_cb)
1461 : {
1462 26 : struct spdk_bdev_shared_resource *shared_resource = bdev_io->internal.ch->shared_resource;
1463 :
1464 26 : assert(bdev_io->internal.f.has_bounce_buf == false);
1465 :
1466 26 : bdev_io->internal.data_transfer_cpl = cpl_cb;
1467 26 : bdev_io->internal.f.has_bounce_buf = true;
1468 : /* save original iovec */
1469 26 : bdev_io->internal.bounce_buf.orig_iovs = bdev_io->u.bdev.iovs;
1470 26 : bdev_io->internal.bounce_buf.orig_iovcnt = bdev_io->u.bdev.iovcnt;
1471 : /* zero the other data members */
1472 26 : bdev_io->internal.bounce_buf.iov.iov_base = NULL;
1473 26 : bdev_io->internal.bounce_buf.md_iov.iov_base = NULL;
1474 26 : bdev_io->internal.bounce_buf.orig_md_iov.iov_base = NULL;
1475 : /* set bounce iov */
1476 26 : bdev_io->u.bdev.iovs = &bdev_io->internal.bounce_buf.iov;
1477 26 : bdev_io->u.bdev.iovcnt = 1;
1478 : /* set bounce buffer for this operation */
1479 26 : bdev_io->u.bdev.iovs[0].iov_base = buf;
1480 26 : bdev_io->u.bdev.iovs[0].iov_len = len;
1481 : /* Now we use 1 iov, the split condition could have been changed */
1482 26 : bdev_io->internal.f.split = bdev_io_should_split(bdev_io);
1483 :
1484 26 : if (spdk_unlikely(!TAILQ_EMPTY(&shared_resource->nomem_io))) {
1485 0 : bdev_queue_nomem_io_tail(shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PULL);
1486 : } else {
1487 26 : bdev_io_pull_data(bdev_io);
1488 : }
1489 26 : }
1490 :
1491 : static void
1492 42 : _bdev_io_set_buf(struct spdk_bdev_io *bdev_io, void *buf, uint64_t len)
1493 : {
1494 42 : struct spdk_bdev *bdev = bdev_io->bdev;
1495 : bool buf_allocated;
1496 : uint64_t alignment;
1497 : void *aligned_buf;
1498 :
1499 42 : bdev_io->internal.buf.ptr = buf;
1500 42 : bdev_io->internal.f.has_buf = true;
1501 :
1502 42 : if (spdk_unlikely(bdev_io->internal.get_aux_buf_cb != NULL)) {
1503 0 : bdev_io_get_buf_complete(bdev_io, true);
1504 0 : return;
1505 : }
1506 :
1507 42 : alignment = spdk_bdev_get_buf_align(bdev);
1508 42 : buf_allocated = _is_buf_allocated(bdev_io->u.bdev.iovs);
1509 42 : aligned_buf = (void *)(((uintptr_t)buf + (alignment - 1)) & ~(alignment - 1));
1510 :
1511 42 : if (buf_allocated) {
1512 26 : _bdev_io_pull_bounce_data_buf(bdev_io, aligned_buf, len, _bdev_io_pull_buffer_cpl);
1513 : /* Continue in completion callback */
1514 26 : return;
1515 : } else {
1516 16 : spdk_bdev_io_set_buf(bdev_io, aligned_buf, len);
1517 : }
1518 :
1519 16 : _bdev_io_set_md_buf(bdev_io);
1520 : }
1521 :
1522 : static inline uint64_t
1523 42 : bdev_io_get_max_buf_len(struct spdk_bdev_io *bdev_io, uint64_t len)
1524 : {
1525 42 : struct spdk_bdev *bdev = bdev_io->bdev;
1526 : uint64_t md_len, alignment;
1527 :
1528 42 : md_len = spdk_bdev_is_md_separate(bdev) ? bdev_io->u.bdev.num_blocks * bdev->md_len : 0;
1529 :
1530 : /* 1 byte alignment needs 0 byte of extra space, 64 bytes alignment needs 63 bytes of extra space, etc. */
1531 42 : alignment = spdk_bdev_get_buf_align(bdev) - 1;
1532 :
1533 42 : return len + alignment + md_len;
1534 : }
1535 :
1536 : static void
1537 42 : bdev_io_put_accel_buf(struct spdk_bdev_io *bdev_io)
1538 : {
1539 42 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1540 :
1541 42 : spdk_accel_put_buf(ch->accel_channel,
1542 : bdev_io->internal.buf.ptr,
1543 : bdev_io->u.bdev.memory_domain,
1544 : bdev_io->u.bdev.memory_domain_ctx);
1545 42 : }
1546 :
1547 : static void
1548 0 : _bdev_io_put_buf(struct spdk_bdev_io *bdev_io, void *buf, uint64_t buf_len)
1549 : {
1550 : struct spdk_bdev_mgmt_channel *ch;
1551 :
1552 0 : ch = bdev_io->internal.ch->shared_resource->mgmt_ch;
1553 0 : spdk_iobuf_put(&ch->iobuf, buf, bdev_io_get_max_buf_len(bdev_io, buf_len));
1554 0 : }
1555 :
1556 : static void
1557 42 : bdev_io_put_buf(struct spdk_bdev_io *bdev_io)
1558 : {
1559 42 : assert(bdev_io->internal.f.has_buf);
1560 :
1561 42 : if (bdev_io->u.bdev.memory_domain == spdk_accel_get_memory_domain()) {
1562 42 : bdev_io_put_accel_buf(bdev_io);
1563 : } else {
1564 0 : assert(bdev_io->u.bdev.memory_domain == NULL);
1565 0 : _bdev_io_put_buf(bdev_io, bdev_io->internal.buf.ptr,
1566 : bdev_io->internal.buf.len);
1567 : }
1568 42 : bdev_io->internal.buf.ptr = NULL;
1569 42 : bdev_io->internal.f.has_buf = false;
1570 42 : }
1571 :
1572 3 : SPDK_LOG_DEPRECATION_REGISTER(spdk_bdev_io_put_aux_buf,
1573 : "spdk_bdev_io_put_aux_buf is deprecated", "v25.01", 0);
1574 :
1575 : void
1576 0 : spdk_bdev_io_put_aux_buf(struct spdk_bdev_io *bdev_io, void *buf)
1577 : {
1578 0 : uint64_t len = bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen;
1579 :
1580 0 : SPDK_LOG_DEPRECATED(spdk_bdev_io_put_aux_buf);
1581 :
1582 0 : assert(buf != NULL);
1583 0 : _bdev_io_put_buf(bdev_io, buf, len);
1584 0 : }
1585 :
1586 : static inline void
1587 549 : 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 549 : if ((bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE ||
1594 549 : 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 : }
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 549 : assert((bdev_io->type != SPDK_BDEV_IO_TYPE_WRITE &&
1603 : bdev_io->type != SPDK_BDEV_IO_TYPE_READ) ||
1604 : ((bdev_io->u.bdev.dif_check_flags & bdev->dif_check_flags) ==
1605 : bdev_io->u.bdev.dif_check_flags));
1606 :
1607 549 : bdev->fn_table->submit_request(ioch, bdev_io);
1608 549 : }
1609 :
1610 : static inline void
1611 10 : bdev_ch_resubmit_io(struct spdk_bdev_shared_resource *shared_resource, struct spdk_bdev_io *bdev_io)
1612 : {
1613 10 : struct spdk_bdev *bdev = bdev_io->bdev;
1614 :
1615 10 : bdev_io_increment_outstanding(bdev_io->internal.ch, shared_resource);
1616 10 : bdev_io->internal.error.nvme.cdw0 = 0;
1617 10 : bdev_io->num_retries++;
1618 10 : bdev_submit_request(bdev, spdk_bdev_io_get_io_channel(bdev_io), bdev_io);
1619 10 : }
1620 :
1621 : static void
1622 63 : bdev_shared_ch_retry_io(struct spdk_bdev_shared_resource *shared_resource)
1623 : {
1624 : struct spdk_bdev_io *bdev_io;
1625 :
1626 63 : if (shared_resource->io_outstanding > shared_resource->nomem_threshold) {
1627 : /*
1628 : * Allow some more I/O to complete before retrying the nomem_io queue.
1629 : * Some drivers (such as nvme) cannot immediately take a new I/O in
1630 : * the context of a completion, because the resources for the I/O are
1631 : * not released until control returns to the bdev poller. Also, we
1632 : * may require several small I/O to complete before a larger I/O
1633 : * (that requires splitting) can be submitted.
1634 : */
1635 58 : return;
1636 : }
1637 :
1638 16 : while (!TAILQ_EMPTY(&shared_resource->nomem_io)) {
1639 12 : bdev_io = TAILQ_FIRST(&shared_resource->nomem_io);
1640 12 : TAILQ_REMOVE(&shared_resource->nomem_io, bdev_io, internal.link);
1641 :
1642 12 : switch (bdev_io->internal.retry_state) {
1643 10 : case BDEV_IO_RETRY_STATE_SUBMIT:
1644 10 : bdev_ch_resubmit_io(shared_resource, bdev_io);
1645 10 : break;
1646 1 : case BDEV_IO_RETRY_STATE_PULL:
1647 1 : bdev_io_pull_data(bdev_io);
1648 1 : break;
1649 0 : case BDEV_IO_RETRY_STATE_PULL_MD:
1650 0 : bdev_io_pull_md_buf(bdev_io);
1651 0 : break;
1652 1 : case BDEV_IO_RETRY_STATE_PUSH:
1653 1 : bdev_io_push_bounce_data(bdev_io);
1654 1 : break;
1655 0 : case BDEV_IO_RETRY_STATE_PUSH_MD:
1656 0 : bdev_io_push_bounce_md_buf(bdev_io);
1657 0 : break;
1658 0 : case BDEV_IO_RETRY_STATE_GET_ACCEL_BUF:
1659 0 : _bdev_io_get_accel_buf(bdev_io);
1660 0 : break;
1661 0 : default:
1662 0 : assert(0 && "invalid retry state");
1663 : break;
1664 : }
1665 :
1666 12 : if (bdev_io == TAILQ_FIRST(&shared_resource->nomem_io)) {
1667 : /* This IO completed again with NOMEM status, so break the loop and
1668 : * don't try anymore. Note that a bdev_io that fails with NOMEM
1669 : * always gets requeued at the front of the list, to maintain
1670 : * ordering.
1671 : */
1672 1 : break;
1673 : }
1674 : }
1675 : }
1676 :
1677 : static void
1678 63 : bdev_ch_retry_io(struct spdk_bdev_channel *bdev_ch)
1679 : {
1680 63 : bdev_shared_ch_retry_io(bdev_ch->shared_resource);
1681 63 : }
1682 :
1683 : static int
1684 0 : bdev_no_mem_poller(void *ctx)
1685 : {
1686 0 : struct spdk_bdev_shared_resource *shared_resource = ctx;
1687 :
1688 0 : spdk_poller_unregister(&shared_resource->nomem_poller);
1689 :
1690 0 : if (!TAILQ_EMPTY(&shared_resource->nomem_io)) {
1691 0 : bdev_shared_ch_retry_io(shared_resource);
1692 : }
1693 : /* the retry cb may re-register the poller so double check */
1694 0 : if (!TAILQ_EMPTY(&shared_resource->nomem_io) &&
1695 0 : shared_resource->io_outstanding == 0 && shared_resource->nomem_poller == NULL) {
1696 : /* No IOs were submitted, try again */
1697 0 : shared_resource->nomem_poller = SPDK_POLLER_REGISTER(bdev_no_mem_poller, shared_resource,
1698 : SPDK_BDEV_IO_POLL_INTERVAL_IN_MSEC * 10);
1699 : }
1700 :
1701 0 : return SPDK_POLLER_BUSY;
1702 : }
1703 :
1704 : static inline bool
1705 556 : _bdev_io_handle_no_mem(struct spdk_bdev_io *bdev_io, enum bdev_io_retry_state state)
1706 : {
1707 556 : struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
1708 556 : struct spdk_bdev_shared_resource *shared_resource = bdev_ch->shared_resource;
1709 :
1710 556 : if (spdk_unlikely(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_NOMEM)) {
1711 5 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_PENDING;
1712 5 : bdev_queue_nomem_io_head(shared_resource, bdev_io, state);
1713 :
1714 5 : if (shared_resource->io_outstanding == 0 && !shared_resource->nomem_poller) {
1715 : /* Special case when we have nomem IOs and no outstanding IOs which completions
1716 : * could trigger retry of queued IOs
1717 : * Any IOs submitted may trigger retry of queued IOs. This poller handles a case when no
1718 : * new IOs submitted, e.g. qd==1 */
1719 0 : shared_resource->nomem_poller = SPDK_POLLER_REGISTER(bdev_no_mem_poller, shared_resource,
1720 : SPDK_BDEV_IO_POLL_INTERVAL_IN_MSEC * 10);
1721 : }
1722 : /* If bdev module completed an I/O that has an accel sequence with NOMEM status, the
1723 : * ownership of that sequence is transferred back to the bdev layer, so we need to
1724 : * restore internal.accel_sequence to make sure that the sequence is handled
1725 : * correctly in case the I/O is later aborted. */
1726 5 : if ((bdev_io->type == SPDK_BDEV_IO_TYPE_READ ||
1727 5 : bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) && bdev_io->u.bdev.accel_sequence) {
1728 0 : assert(!bdev_io_use_accel_sequence(bdev_io));
1729 0 : bdev_io->internal.f.has_accel_sequence = true;
1730 0 : bdev_io->internal.accel_sequence = bdev_io->u.bdev.accel_sequence;
1731 : }
1732 :
1733 5 : return true;
1734 : }
1735 :
1736 551 : if (spdk_unlikely(!TAILQ_EMPTY(&shared_resource->nomem_io))) {
1737 63 : bdev_ch_retry_io(bdev_ch);
1738 : }
1739 :
1740 551 : return false;
1741 : }
1742 :
1743 : static void
1744 26 : _bdev_io_complete_push_bounce_done(void *ctx, int rc)
1745 : {
1746 26 : struct spdk_bdev_io *bdev_io = ctx;
1747 26 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1748 :
1749 26 : if (rc) {
1750 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
1751 : }
1752 : /* We want to free the bounce buffer here since we know we're done with it (as opposed
1753 : * to waiting for the conditional free of internal.buf.ptr in spdk_bdev_free_io()).
1754 : */
1755 26 : bdev_io_put_buf(bdev_io);
1756 :
1757 26 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1758 0 : bdev_ch_retry_io(ch);
1759 : }
1760 :
1761 : /* Continue with IO completion flow */
1762 26 : bdev_io_complete(bdev_io);
1763 26 : }
1764 :
1765 : static void
1766 2 : bdev_io_push_bounce_md_buf_done(void *ctx, int rc)
1767 : {
1768 2 : struct spdk_bdev_io *bdev_io = ctx;
1769 2 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1770 :
1771 2 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1772 2 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1773 2 : bdev_io->internal.f.has_bounce_buf = false;
1774 :
1775 2 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1776 0 : bdev_ch_retry_io(ch);
1777 : }
1778 :
1779 2 : bdev_io->internal.data_transfer_cpl(bdev_io, rc);
1780 2 : }
1781 :
1782 : static inline void
1783 26 : bdev_io_push_bounce_md_buf(struct spdk_bdev_io *bdev_io)
1784 : {
1785 26 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1786 26 : int rc = 0;
1787 :
1788 26 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS);
1789 26 : assert(bdev_io->internal.f.has_bounce_buf);
1790 :
1791 : /* do the same for metadata buffer */
1792 26 : if (spdk_unlikely(bdev_io->internal.bounce_buf.orig_md_iov.iov_base != NULL)) {
1793 4 : assert(spdk_bdev_is_md_separate(bdev_io->bdev));
1794 :
1795 4 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_READ) {
1796 2 : if (bdev_io_use_memory_domain(bdev_io)) {
1797 2 : TAILQ_INSERT_TAIL(&ch->io_memory_domain, bdev_io, internal.link);
1798 2 : bdev_io_increment_outstanding(ch, ch->shared_resource);
1799 : /* If memory domain is used then we need to call async push function */
1800 4 : rc = spdk_memory_domain_push_data(bdev_io->internal.memory_domain,
1801 : bdev_io->internal.memory_domain_ctx,
1802 : &bdev_io->internal.bounce_buf.orig_md_iov,
1803 2 : (uint32_t)bdev_io->internal.bounce_buf.orig_iovcnt,
1804 : &bdev_io->internal.bounce_buf.md_iov, 1,
1805 : bdev_io_push_bounce_md_buf_done,
1806 : bdev_io);
1807 2 : if (rc == 0) {
1808 : /* Continue IO completion in async callback */
1809 2 : return;
1810 : }
1811 0 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1812 0 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1813 0 : if (rc != -ENOMEM) {
1814 0 : SPDK_ERRLOG("Failed to push md to memory domain %s\n",
1815 : spdk_memory_domain_get_dma_device_id(
1816 : bdev_io->internal.memory_domain));
1817 : }
1818 : } else {
1819 0 : memcpy(bdev_io->internal.bounce_buf.orig_md_iov.iov_base, bdev_io->u.bdev.md_buf,
1820 : bdev_io->internal.bounce_buf.orig_md_iov.iov_len);
1821 : }
1822 : }
1823 : }
1824 :
1825 24 : if (spdk_unlikely(rc == -ENOMEM)) {
1826 0 : bdev_queue_nomem_io_head(ch->shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PUSH_MD);
1827 : } else {
1828 24 : assert(bdev_io->internal.data_transfer_cpl);
1829 24 : bdev_io->internal.f.has_bounce_buf = false;
1830 24 : bdev_io->internal.data_transfer_cpl(bdev_io, rc);
1831 : }
1832 : }
1833 :
1834 : static inline void
1835 26 : bdev_io_push_bounce_data_done(struct spdk_bdev_io *bdev_io, int rc)
1836 : {
1837 26 : assert(bdev_io->internal.data_transfer_cpl);
1838 26 : if (rc) {
1839 0 : bdev_io->internal.data_transfer_cpl(bdev_io, rc);
1840 0 : return;
1841 : }
1842 :
1843 : /* set original buffer for this io */
1844 26 : bdev_io->u.bdev.iovcnt = bdev_io->internal.bounce_buf.orig_iovcnt;
1845 26 : bdev_io->u.bdev.iovs = bdev_io->internal.bounce_buf.orig_iovs;
1846 :
1847 : /* We don't set bdev_io->internal.f.has_bounce_buf to false here because
1848 : * we still need to clear the md buf */
1849 :
1850 26 : bdev_io_push_bounce_md_buf(bdev_io);
1851 : }
1852 :
1853 : static void
1854 2 : bdev_io_push_bounce_data_done_and_track(void *ctx, int status)
1855 : {
1856 2 : struct spdk_bdev_io *bdev_io = ctx;
1857 2 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1858 :
1859 2 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1860 2 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1861 :
1862 2 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1863 0 : bdev_ch_retry_io(ch);
1864 : }
1865 :
1866 2 : bdev_io_push_bounce_data_done(bdev_io, status);
1867 2 : }
1868 :
1869 : static inline void
1870 27 : bdev_io_push_bounce_data(struct spdk_bdev_io *bdev_io)
1871 : {
1872 27 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1873 27 : int rc = 0;
1874 :
1875 27 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS);
1876 27 : assert(!bdev_io_use_accel_sequence(bdev_io));
1877 27 : assert(bdev_io->internal.f.has_bounce_buf);
1878 :
1879 : /* if this is read path, copy data from bounce buffer to original buffer */
1880 27 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_READ) {
1881 11 : if (bdev_io_use_memory_domain(bdev_io)) {
1882 3 : TAILQ_INSERT_TAIL(&ch->io_memory_domain, bdev_io, internal.link);
1883 3 : bdev_io_increment_outstanding(ch, ch->shared_resource);
1884 : /* If memory domain is used then we need to call async push function */
1885 6 : rc = spdk_memory_domain_push_data(bdev_io->internal.memory_domain,
1886 : bdev_io->internal.memory_domain_ctx,
1887 : bdev_io->internal.bounce_buf.orig_iovs,
1888 3 : (uint32_t)bdev_io->internal.bounce_buf.orig_iovcnt,
1889 : &bdev_io->internal.bounce_buf.iov, 1,
1890 : bdev_io_push_bounce_data_done_and_track,
1891 : bdev_io);
1892 3 : if (rc == 0) {
1893 : /* Continue IO completion in async callback */
1894 2 : return;
1895 : }
1896 :
1897 1 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1898 1 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1899 1 : if (rc != -ENOMEM) {
1900 0 : SPDK_ERRLOG("Failed to push data to memory domain %s\n",
1901 : spdk_memory_domain_get_dma_device_id(
1902 : bdev_io->internal.memory_domain));
1903 : }
1904 : } else {
1905 8 : spdk_copy_buf_to_iovs(bdev_io->internal.bounce_buf.orig_iovs,
1906 : bdev_io->internal.bounce_buf.orig_iovcnt,
1907 : bdev_io->internal.bounce_buf.iov.iov_base,
1908 : bdev_io->internal.bounce_buf.iov.iov_len);
1909 : }
1910 : }
1911 :
1912 25 : if (spdk_unlikely(rc == -ENOMEM)) {
1913 1 : bdev_queue_nomem_io_head(ch->shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PUSH);
1914 : } else {
1915 24 : bdev_io_push_bounce_data_done(bdev_io, rc);
1916 : }
1917 : }
1918 :
1919 : static inline void
1920 26 : _bdev_io_push_bounce_data_buffer(struct spdk_bdev_io *bdev_io, bdev_copy_bounce_buffer_cpl cpl_cb)
1921 : {
1922 26 : bdev_io->internal.data_transfer_cpl = cpl_cb;
1923 26 : bdev_io_push_bounce_data(bdev_io);
1924 26 : }
1925 :
1926 : static void
1927 0 : bdev_io_get_iobuf_cb(struct spdk_iobuf_entry *iobuf, void *buf)
1928 : {
1929 : struct spdk_bdev_io *bdev_io;
1930 :
1931 0 : bdev_io = SPDK_CONTAINEROF(iobuf, struct spdk_bdev_io, internal.iobuf);
1932 0 : _bdev_io_set_buf(bdev_io, buf, bdev_io->internal.buf.len);
1933 0 : }
1934 :
1935 : static void
1936 42 : bdev_io_get_buf(struct spdk_bdev_io *bdev_io, uint64_t len)
1937 : {
1938 : struct spdk_bdev_mgmt_channel *mgmt_ch;
1939 : uint64_t max_len;
1940 : void *buf;
1941 :
1942 42 : assert(spdk_bdev_io_get_thread(bdev_io) == spdk_get_thread());
1943 42 : mgmt_ch = bdev_io->internal.ch->shared_resource->mgmt_ch;
1944 42 : max_len = bdev_io_get_max_buf_len(bdev_io, len);
1945 :
1946 42 : if (spdk_unlikely(max_len > mgmt_ch->iobuf.cache[0].large.bufsize)) {
1947 0 : SPDK_ERRLOG("Length %" PRIu64 " is larger than allowed\n", max_len);
1948 0 : bdev_io_get_buf_complete(bdev_io, false);
1949 0 : return;
1950 : }
1951 :
1952 42 : bdev_io->internal.buf.len = len;
1953 42 : buf = spdk_iobuf_get(&mgmt_ch->iobuf, max_len, &bdev_io->internal.iobuf,
1954 : bdev_io_get_iobuf_cb);
1955 42 : if (buf != NULL) {
1956 42 : _bdev_io_set_buf(bdev_io, buf, len);
1957 : }
1958 : }
1959 :
1960 : void
1961 56 : spdk_bdev_io_get_buf(struct spdk_bdev_io *bdev_io, spdk_bdev_io_get_buf_cb cb, uint64_t len)
1962 : {
1963 56 : struct spdk_bdev *bdev = bdev_io->bdev;
1964 : uint64_t alignment;
1965 :
1966 56 : assert(cb != NULL);
1967 56 : bdev_io->internal.get_buf_cb = cb;
1968 :
1969 56 : alignment = spdk_bdev_get_buf_align(bdev);
1970 :
1971 96 : if (_is_buf_allocated(bdev_io->u.bdev.iovs) &&
1972 40 : _are_iovs_aligned(bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt, alignment)) {
1973 : /* Buffer already present and aligned */
1974 18 : cb(spdk_bdev_io_get_io_channel(bdev_io), bdev_io, true);
1975 18 : return;
1976 : }
1977 :
1978 38 : bdev_io_get_buf(bdev_io, len);
1979 : }
1980 :
1981 : static void
1982 4 : _bdev_io_get_bounce_buf(struct spdk_bdev_io *bdev_io, spdk_bdev_io_get_buf_cb cb,
1983 : uint64_t len)
1984 : {
1985 4 : assert(cb != NULL);
1986 4 : bdev_io->internal.get_buf_cb = cb;
1987 :
1988 4 : bdev_io_get_buf(bdev_io, len);
1989 4 : }
1990 :
1991 : static void
1992 0 : _bdev_io_get_accel_buf(struct spdk_bdev_io *bdev_io)
1993 : {
1994 0 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1995 0 : void *buf;
1996 : int rc;
1997 :
1998 0 : rc = spdk_accel_get_buf(ch->accel_channel,
1999 : bdev_io->internal.buf.len,
2000 : &buf,
2001 : &bdev_io->u.bdev.memory_domain,
2002 : &bdev_io->u.bdev.memory_domain_ctx);
2003 0 : if (rc != 0) {
2004 0 : bdev_queue_nomem_io_tail(ch->shared_resource, bdev_io,
2005 : BDEV_IO_RETRY_STATE_GET_ACCEL_BUF);
2006 0 : return;
2007 : }
2008 :
2009 0 : _bdev_io_set_buf(bdev_io, buf, bdev_io->internal.buf.len);
2010 : }
2011 :
2012 : static inline void
2013 0 : bdev_io_get_accel_buf(struct spdk_bdev_io *bdev_io, spdk_bdev_io_get_buf_cb cb,
2014 : uint64_t len)
2015 : {
2016 0 : bdev_io->internal.buf.len = len;
2017 0 : bdev_io->internal.get_buf_cb = cb;
2018 :
2019 0 : _bdev_io_get_accel_buf(bdev_io);
2020 0 : }
2021 :
2022 3 : SPDK_LOG_DEPRECATION_REGISTER(spdk_bdev_io_get_aux_buf,
2023 : "spdk_bdev_io_get_aux_buf is deprecated", "v25.01", 0);
2024 :
2025 : void
2026 0 : spdk_bdev_io_get_aux_buf(struct spdk_bdev_io *bdev_io, spdk_bdev_io_get_aux_buf_cb cb)
2027 : {
2028 0 : uint64_t len = bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen;
2029 :
2030 0 : SPDK_LOG_DEPRECATED(spdk_bdev_io_get_aux_buf);
2031 :
2032 0 : assert(cb != NULL);
2033 0 : assert(bdev_io->internal.get_aux_buf_cb == NULL);
2034 0 : bdev_io->internal.get_aux_buf_cb = cb;
2035 0 : bdev_io_get_buf(bdev_io, len);
2036 0 : }
2037 :
2038 : static int
2039 68 : bdev_module_get_max_ctx_size(void)
2040 : {
2041 : struct spdk_bdev_module *bdev_module;
2042 68 : int max_bdev_module_size = 0;
2043 :
2044 266 : TAILQ_FOREACH(bdev_module, &g_bdev_mgr.bdev_modules, internal.tailq) {
2045 198 : if (bdev_module->get_ctx_size && bdev_module->get_ctx_size() > max_bdev_module_size) {
2046 67 : max_bdev_module_size = bdev_module->get_ctx_size();
2047 : }
2048 : }
2049 :
2050 68 : return max_bdev_module_size;
2051 : }
2052 :
2053 : static void
2054 0 : bdev_enable_histogram_config_json(struct spdk_bdev *bdev, struct spdk_json_write_ctx *w)
2055 : {
2056 0 : if (!bdev->internal.histogram_enabled) {
2057 0 : return;
2058 : }
2059 :
2060 0 : spdk_json_write_object_begin(w);
2061 0 : spdk_json_write_named_string(w, "method", "bdev_enable_histogram");
2062 :
2063 0 : spdk_json_write_named_object_begin(w, "params");
2064 0 : spdk_json_write_named_string(w, "name", bdev->name);
2065 :
2066 0 : spdk_json_write_named_bool(w, "enable", bdev->internal.histogram_enabled);
2067 :
2068 0 : if (bdev->internal.histogram_io_type) {
2069 0 : spdk_json_write_named_string(w, "opc",
2070 0 : spdk_bdev_get_io_type_name(bdev->internal.histogram_io_type));
2071 : }
2072 :
2073 0 : spdk_json_write_object_end(w);
2074 :
2075 0 : spdk_json_write_object_end(w);
2076 : }
2077 :
2078 : static void
2079 0 : bdev_qos_config_json(struct spdk_bdev *bdev, struct spdk_json_write_ctx *w)
2080 : {
2081 : int i;
2082 0 : struct spdk_bdev_qos *qos = bdev->internal.qos;
2083 0 : uint64_t limits[SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES];
2084 :
2085 0 : if (!qos) {
2086 0 : return;
2087 : }
2088 :
2089 0 : spdk_bdev_get_qos_rate_limits(bdev, limits);
2090 :
2091 0 : spdk_json_write_object_begin(w);
2092 0 : spdk_json_write_named_string(w, "method", "bdev_set_qos_limit");
2093 :
2094 0 : spdk_json_write_named_object_begin(w, "params");
2095 0 : spdk_json_write_named_string(w, "name", bdev->name);
2096 0 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
2097 0 : if (limits[i] > 0) {
2098 0 : spdk_json_write_named_uint64(w, qos_rpc_type[i], limits[i]);
2099 : }
2100 : }
2101 0 : spdk_json_write_object_end(w);
2102 :
2103 0 : spdk_json_write_object_end(w);
2104 : }
2105 :
2106 : void
2107 0 : spdk_bdev_subsystem_config_json(struct spdk_json_write_ctx *w)
2108 : {
2109 : struct spdk_bdev_module *bdev_module;
2110 : struct spdk_bdev *bdev;
2111 :
2112 0 : assert(w != NULL);
2113 :
2114 0 : spdk_json_write_array_begin(w);
2115 :
2116 0 : spdk_json_write_object_begin(w);
2117 0 : spdk_json_write_named_string(w, "method", "bdev_set_options");
2118 0 : spdk_json_write_named_object_begin(w, "params");
2119 0 : spdk_json_write_named_uint32(w, "bdev_io_pool_size", g_bdev_opts.bdev_io_pool_size);
2120 0 : spdk_json_write_named_uint32(w, "bdev_io_cache_size", g_bdev_opts.bdev_io_cache_size);
2121 0 : spdk_json_write_named_bool(w, "bdev_auto_examine", g_bdev_opts.bdev_auto_examine);
2122 0 : spdk_json_write_named_uint32(w, "iobuf_small_cache_size", g_bdev_opts.iobuf_small_cache_size);
2123 0 : spdk_json_write_named_uint32(w, "iobuf_large_cache_size", g_bdev_opts.iobuf_large_cache_size);
2124 0 : spdk_json_write_object_end(w);
2125 0 : spdk_json_write_object_end(w);
2126 :
2127 0 : bdev_examine_allowlist_config_json(w);
2128 :
2129 0 : TAILQ_FOREACH(bdev_module, &g_bdev_mgr.bdev_modules, internal.tailq) {
2130 0 : if (bdev_module->config_json) {
2131 0 : bdev_module->config_json(w);
2132 : }
2133 : }
2134 :
2135 0 : spdk_spin_lock(&g_bdev_mgr.spinlock);
2136 :
2137 0 : TAILQ_FOREACH(bdev, &g_bdev_mgr.bdevs, internal.link) {
2138 0 : if (bdev->fn_table->write_config_json) {
2139 0 : bdev->fn_table->write_config_json(bdev, w);
2140 : }
2141 :
2142 0 : bdev_qos_config_json(bdev, w);
2143 0 : bdev_enable_histogram_config_json(bdev, w);
2144 : }
2145 :
2146 0 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
2147 :
2148 : /* This has to be last RPC in array to make sure all bdevs finished examine */
2149 0 : spdk_json_write_object_begin(w);
2150 0 : spdk_json_write_named_string(w, "method", "bdev_wait_for_examine");
2151 0 : spdk_json_write_object_end(w);
2152 :
2153 0 : spdk_json_write_array_end(w);
2154 0 : }
2155 :
2156 : static void
2157 72 : bdev_mgmt_channel_destroy(void *io_device, void *ctx_buf)
2158 : {
2159 72 : struct spdk_bdev_mgmt_channel *ch = ctx_buf;
2160 : struct spdk_bdev_io *bdev_io;
2161 :
2162 72 : spdk_iobuf_channel_fini(&ch->iobuf);
2163 :
2164 10226 : while (!STAILQ_EMPTY(&ch->per_thread_cache)) {
2165 10154 : bdev_io = STAILQ_FIRST(&ch->per_thread_cache);
2166 10154 : STAILQ_REMOVE_HEAD(&ch->per_thread_cache, internal.buf_link);
2167 10154 : ch->per_thread_cache_count--;
2168 10154 : spdk_mempool_put(g_bdev_mgr.bdev_io_pool, (void *)bdev_io);
2169 : }
2170 :
2171 72 : assert(ch->per_thread_cache_count == 0);
2172 72 : }
2173 :
2174 : static int
2175 72 : bdev_mgmt_channel_create(void *io_device, void *ctx_buf)
2176 : {
2177 72 : struct spdk_bdev_mgmt_channel *ch = ctx_buf;
2178 : struct spdk_bdev_io *bdev_io;
2179 : uint32_t i;
2180 : int rc;
2181 :
2182 72 : rc = spdk_iobuf_channel_init(&ch->iobuf, "bdev",
2183 : g_bdev_opts.iobuf_small_cache_size,
2184 : g_bdev_opts.iobuf_large_cache_size);
2185 72 : if (rc != 0) {
2186 0 : SPDK_ERRLOG("Failed to create iobuf channel: %s\n", spdk_strerror(-rc));
2187 0 : return -1;
2188 : }
2189 :
2190 72 : STAILQ_INIT(&ch->per_thread_cache);
2191 72 : ch->bdev_io_cache_size = g_bdev_opts.bdev_io_cache_size;
2192 :
2193 : /* Pre-populate bdev_io cache to ensure this thread cannot be starved. */
2194 72 : ch->per_thread_cache_count = 0;
2195 10226 : for (i = 0; i < ch->bdev_io_cache_size; i++) {
2196 10154 : bdev_io = spdk_mempool_get(g_bdev_mgr.bdev_io_pool);
2197 10154 : if (bdev_io == NULL) {
2198 0 : SPDK_ERRLOG("You need to increase bdev_io_pool_size using bdev_set_options RPC.\n");
2199 0 : assert(false);
2200 : bdev_mgmt_channel_destroy(io_device, ctx_buf);
2201 : return -1;
2202 : }
2203 10154 : ch->per_thread_cache_count++;
2204 10154 : STAILQ_INSERT_HEAD(&ch->per_thread_cache, bdev_io, internal.buf_link);
2205 : }
2206 :
2207 72 : TAILQ_INIT(&ch->shared_resources);
2208 72 : TAILQ_INIT(&ch->io_wait_queue);
2209 :
2210 72 : return 0;
2211 : }
2212 :
2213 : static void
2214 68 : bdev_init_complete(int rc)
2215 : {
2216 68 : spdk_bdev_init_cb cb_fn = g_init_cb_fn;
2217 68 : void *cb_arg = g_init_cb_arg;
2218 : struct spdk_bdev_module *m;
2219 :
2220 68 : g_bdev_mgr.init_complete = true;
2221 68 : g_init_cb_fn = NULL;
2222 68 : g_init_cb_arg = NULL;
2223 :
2224 : /*
2225 : * For modules that need to know when subsystem init is complete,
2226 : * inform them now.
2227 : */
2228 68 : if (rc == 0) {
2229 266 : TAILQ_FOREACH(m, &g_bdev_mgr.bdev_modules, internal.tailq) {
2230 198 : if (m->init_complete) {
2231 24 : m->init_complete();
2232 : }
2233 : }
2234 : }
2235 :
2236 68 : cb_fn(cb_arg, rc);
2237 68 : }
2238 :
2239 : static bool
2240 271 : bdev_module_all_actions_completed(void)
2241 : {
2242 : struct spdk_bdev_module *m;
2243 :
2244 1078 : TAILQ_FOREACH(m, &g_bdev_mgr.bdev_modules, internal.tailq) {
2245 807 : if (m->internal.action_in_progress > 0) {
2246 0 : return false;
2247 : }
2248 : }
2249 271 : return true;
2250 : }
2251 :
2252 : static void
2253 629 : bdev_module_action_complete(void)
2254 : {
2255 : /*
2256 : * Don't finish bdev subsystem initialization if
2257 : * module pre-initialization is still in progress, or
2258 : * the subsystem been already initialized.
2259 : */
2260 629 : if (!g_bdev_mgr.module_init_complete || g_bdev_mgr.init_complete) {
2261 561 : return;
2262 : }
2263 :
2264 : /*
2265 : * Check all bdev modules for inits/examinations in progress. If any
2266 : * exist, return immediately since we cannot finish bdev subsystem
2267 : * initialization until all are completed.
2268 : */
2269 68 : if (!bdev_module_all_actions_completed()) {
2270 0 : return;
2271 : }
2272 :
2273 : /*
2274 : * Modules already finished initialization - now that all
2275 : * the bdev modules have finished their asynchronous I/O
2276 : * processing, the entire bdev layer can be marked as complete.
2277 : */
2278 68 : bdev_init_complete(0);
2279 : }
2280 :
2281 : static void
2282 561 : bdev_module_action_done(struct spdk_bdev_module *module)
2283 : {
2284 561 : spdk_spin_lock(&module->internal.spinlock);
2285 561 : assert(module->internal.action_in_progress > 0);
2286 561 : module->internal.action_in_progress--;
2287 561 : spdk_spin_unlock(&module->internal.spinlock);
2288 561 : bdev_module_action_complete();
2289 561 : }
2290 :
2291 : void
2292 68 : spdk_bdev_module_init_done(struct spdk_bdev_module *module)
2293 : {
2294 68 : assert(module->async_init);
2295 68 : bdev_module_action_done(module);
2296 68 : }
2297 :
2298 : void
2299 493 : spdk_bdev_module_examine_done(struct spdk_bdev_module *module)
2300 : {
2301 493 : bdev_module_action_done(module);
2302 493 : }
2303 :
2304 : /** The last initialized bdev module */
2305 : static struct spdk_bdev_module *g_resume_bdev_module = NULL;
2306 :
2307 : static void
2308 0 : bdev_init_failed(void *cb_arg)
2309 : {
2310 0 : struct spdk_bdev_module *module = cb_arg;
2311 :
2312 0 : spdk_spin_lock(&module->internal.spinlock);
2313 0 : assert(module->internal.action_in_progress > 0);
2314 0 : module->internal.action_in_progress--;
2315 0 : spdk_spin_unlock(&module->internal.spinlock);
2316 0 : bdev_init_complete(-1);
2317 0 : }
2318 :
2319 : static int
2320 68 : bdev_modules_init(void)
2321 : {
2322 : struct spdk_bdev_module *module;
2323 68 : int rc = 0;
2324 :
2325 266 : TAILQ_FOREACH(module, &g_bdev_mgr.bdev_modules, internal.tailq) {
2326 198 : g_resume_bdev_module = module;
2327 198 : if (module->async_init) {
2328 68 : spdk_spin_lock(&module->internal.spinlock);
2329 68 : module->internal.action_in_progress = 1;
2330 68 : spdk_spin_unlock(&module->internal.spinlock);
2331 : }
2332 198 : rc = module->module_init();
2333 198 : if (rc != 0) {
2334 : /* Bump action_in_progress to prevent other modules from completion of modules_init
2335 : * Send message to defer application shutdown until resources are cleaned up */
2336 0 : spdk_spin_lock(&module->internal.spinlock);
2337 0 : module->internal.action_in_progress = 1;
2338 0 : spdk_spin_unlock(&module->internal.spinlock);
2339 0 : spdk_thread_send_msg(spdk_get_thread(), bdev_init_failed, module);
2340 0 : return rc;
2341 : }
2342 : }
2343 :
2344 68 : g_resume_bdev_module = NULL;
2345 68 : return 0;
2346 : }
2347 :
2348 : void
2349 68 : spdk_bdev_initialize(spdk_bdev_init_cb cb_fn, void *cb_arg)
2350 : {
2351 68 : int rc = 0;
2352 68 : char mempool_name[32];
2353 :
2354 68 : assert(cb_fn != NULL);
2355 :
2356 68 : g_init_cb_fn = cb_fn;
2357 68 : g_init_cb_arg = cb_arg;
2358 :
2359 68 : spdk_notify_type_register("bdev_register");
2360 68 : spdk_notify_type_register("bdev_unregister");
2361 :
2362 68 : snprintf(mempool_name, sizeof(mempool_name), "bdev_io_%d", getpid());
2363 :
2364 68 : rc = spdk_iobuf_register_module("bdev");
2365 68 : if (rc != 0) {
2366 0 : SPDK_ERRLOG("could not register bdev iobuf module: %s\n", spdk_strerror(-rc));
2367 0 : bdev_init_complete(-1);
2368 0 : return;
2369 : }
2370 :
2371 136 : g_bdev_mgr.bdev_io_pool = spdk_mempool_create(mempool_name,
2372 68 : g_bdev_opts.bdev_io_pool_size,
2373 : sizeof(struct spdk_bdev_io) +
2374 68 : bdev_module_get_max_ctx_size(),
2375 : 0,
2376 : SPDK_ENV_NUMA_ID_ANY);
2377 :
2378 68 : if (g_bdev_mgr.bdev_io_pool == NULL) {
2379 0 : SPDK_ERRLOG("could not allocate spdk_bdev_io pool\n");
2380 0 : bdev_init_complete(-1);
2381 0 : return;
2382 : }
2383 :
2384 68 : g_bdev_mgr.zero_buffer = spdk_zmalloc(ZERO_BUFFER_SIZE, ZERO_BUFFER_SIZE,
2385 : NULL, SPDK_ENV_LCORE_ID_ANY, SPDK_MALLOC_DMA);
2386 68 : if (!g_bdev_mgr.zero_buffer) {
2387 0 : SPDK_ERRLOG("create bdev zero buffer failed\n");
2388 0 : bdev_init_complete(-1);
2389 0 : return;
2390 : }
2391 :
2392 : #ifdef SPDK_CONFIG_VTUNE
2393 : g_bdev_mgr.domain = __itt_domain_create("spdk_bdev");
2394 : #endif
2395 :
2396 68 : spdk_io_device_register(&g_bdev_mgr, bdev_mgmt_channel_create,
2397 : bdev_mgmt_channel_destroy,
2398 : sizeof(struct spdk_bdev_mgmt_channel),
2399 : "bdev_mgr");
2400 :
2401 68 : rc = bdev_modules_init();
2402 68 : g_bdev_mgr.module_init_complete = true;
2403 68 : if (rc != 0) {
2404 0 : SPDK_ERRLOG("bdev modules init failed\n");
2405 0 : return;
2406 : }
2407 :
2408 68 : bdev_module_action_complete();
2409 : }
2410 :
2411 : static void
2412 68 : bdev_mgr_unregister_cb(void *io_device)
2413 : {
2414 68 : spdk_bdev_fini_cb cb_fn = g_fini_cb_fn;
2415 :
2416 68 : if (g_bdev_mgr.bdev_io_pool) {
2417 68 : if (spdk_mempool_count(g_bdev_mgr.bdev_io_pool) != g_bdev_opts.bdev_io_pool_size) {
2418 0 : SPDK_ERRLOG("bdev IO pool count is %zu but should be %u\n",
2419 : spdk_mempool_count(g_bdev_mgr.bdev_io_pool),
2420 : g_bdev_opts.bdev_io_pool_size);
2421 : }
2422 :
2423 68 : spdk_mempool_free(g_bdev_mgr.bdev_io_pool);
2424 : }
2425 :
2426 68 : spdk_free(g_bdev_mgr.zero_buffer);
2427 :
2428 68 : bdev_examine_allowlist_free();
2429 :
2430 68 : cb_fn(g_fini_cb_arg);
2431 68 : g_fini_cb_fn = NULL;
2432 68 : g_fini_cb_arg = NULL;
2433 68 : g_bdev_mgr.init_complete = false;
2434 68 : g_bdev_mgr.module_init_complete = false;
2435 68 : }
2436 :
2437 : static void
2438 68 : bdev_module_fini_iter(void *arg)
2439 : {
2440 : struct spdk_bdev_module *bdev_module;
2441 :
2442 : /* FIXME: Handling initialization failures is broken now,
2443 : * so we won't even try cleaning up after successfully
2444 : * initialized modules. if module_init_complete is false,
2445 : * just call spdk_bdev_mgr_unregister_cb
2446 : */
2447 68 : if (!g_bdev_mgr.module_init_complete) {
2448 0 : bdev_mgr_unregister_cb(NULL);
2449 0 : return;
2450 : }
2451 :
2452 : /* Start iterating from the last touched module */
2453 68 : if (!g_resume_bdev_module) {
2454 68 : bdev_module = TAILQ_LAST(&g_bdev_mgr.bdev_modules, bdev_module_list);
2455 : } else {
2456 0 : bdev_module = TAILQ_PREV(g_resume_bdev_module, bdev_module_list,
2457 : internal.tailq);
2458 : }
2459 :
2460 266 : while (bdev_module) {
2461 198 : if (bdev_module->async_fini) {
2462 : /* Save our place so we can resume later. We must
2463 : * save the variable here, before calling module_fini()
2464 : * below, because in some cases the module may immediately
2465 : * call spdk_bdev_module_fini_done() and re-enter
2466 : * this function to continue iterating. */
2467 0 : g_resume_bdev_module = bdev_module;
2468 : }
2469 :
2470 198 : if (bdev_module->module_fini) {
2471 198 : bdev_module->module_fini();
2472 : }
2473 :
2474 198 : if (bdev_module->async_fini) {
2475 0 : return;
2476 : }
2477 :
2478 198 : bdev_module = TAILQ_PREV(bdev_module, bdev_module_list,
2479 : internal.tailq);
2480 : }
2481 :
2482 68 : g_resume_bdev_module = NULL;
2483 68 : spdk_io_device_unregister(&g_bdev_mgr, bdev_mgr_unregister_cb);
2484 : }
2485 :
2486 : void
2487 0 : spdk_bdev_module_fini_done(void)
2488 : {
2489 0 : if (spdk_get_thread() != g_fini_thread) {
2490 0 : spdk_thread_send_msg(g_fini_thread, bdev_module_fini_iter, NULL);
2491 : } else {
2492 0 : bdev_module_fini_iter(NULL);
2493 : }
2494 0 : }
2495 :
2496 : static void
2497 68 : bdev_finish_unregister_bdevs_iter(void *cb_arg, int bdeverrno)
2498 : {
2499 68 : struct spdk_bdev *bdev = cb_arg;
2500 :
2501 68 : if (bdeverrno && bdev) {
2502 0 : SPDK_WARNLOG("Unable to unregister bdev '%s' during spdk_bdev_finish()\n",
2503 : bdev->name);
2504 :
2505 : /*
2506 : * Since the call to spdk_bdev_unregister() failed, we have no way to free this
2507 : * bdev; try to continue by manually removing this bdev from the list and continue
2508 : * with the next bdev in the list.
2509 : */
2510 0 : TAILQ_REMOVE(&g_bdev_mgr.bdevs, bdev, internal.link);
2511 : }
2512 :
2513 68 : if (TAILQ_EMPTY(&g_bdev_mgr.bdevs)) {
2514 68 : SPDK_DEBUGLOG(bdev, "Done unregistering bdevs\n");
2515 : /*
2516 : * Bdev module finish need to be deferred as we might be in the middle of some context
2517 : * (like bdev part free) that will use this bdev (or private bdev driver ctx data)
2518 : * after returning.
2519 : */
2520 68 : spdk_thread_send_msg(spdk_get_thread(), bdev_module_fini_iter, NULL);
2521 68 : return;
2522 : }
2523 :
2524 : /*
2525 : * Unregister last unclaimed bdev in the list, to ensure that bdev subsystem
2526 : * shutdown proceeds top-down. The goal is to give virtual bdevs an opportunity
2527 : * to detect clean shutdown as opposed to run-time hot removal of the underlying
2528 : * base bdevs.
2529 : *
2530 : * Also, walk the list in the reverse order.
2531 : */
2532 0 : for (bdev = TAILQ_LAST(&g_bdev_mgr.bdevs, spdk_bdev_list);
2533 0 : bdev; bdev = TAILQ_PREV(bdev, spdk_bdev_list, internal.link)) {
2534 0 : spdk_spin_lock(&bdev->internal.spinlock);
2535 0 : if (bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE) {
2536 0 : LOG_ALREADY_CLAIMED_DEBUG("claimed, skipping", bdev);
2537 0 : spdk_spin_unlock(&bdev->internal.spinlock);
2538 0 : continue;
2539 : }
2540 0 : spdk_spin_unlock(&bdev->internal.spinlock);
2541 :
2542 0 : SPDK_DEBUGLOG(bdev, "Unregistering bdev '%s'\n", bdev->name);
2543 0 : spdk_bdev_unregister(bdev, bdev_finish_unregister_bdevs_iter, bdev);
2544 0 : return;
2545 : }
2546 :
2547 : /*
2548 : * If any bdev fails to unclaim underlying bdev properly, we may face the
2549 : * case of bdev list consisting of claimed bdevs only (if claims are managed
2550 : * correctly, this would mean there's a loop in the claims graph which is
2551 : * clearly impossible). Warn and unregister last bdev on the list then.
2552 : */
2553 0 : for (bdev = TAILQ_LAST(&g_bdev_mgr.bdevs, spdk_bdev_list);
2554 0 : bdev; bdev = TAILQ_PREV(bdev, spdk_bdev_list, internal.link)) {
2555 0 : SPDK_WARNLOG("Unregistering claimed bdev '%s'!\n", bdev->name);
2556 0 : spdk_bdev_unregister(bdev, bdev_finish_unregister_bdevs_iter, bdev);
2557 0 : return;
2558 : }
2559 : }
2560 :
2561 : static void
2562 68 : bdev_module_fini_start_iter(void *arg)
2563 : {
2564 : struct spdk_bdev_module *bdev_module;
2565 :
2566 68 : if (!g_resume_bdev_module) {
2567 68 : bdev_module = TAILQ_LAST(&g_bdev_mgr.bdev_modules, bdev_module_list);
2568 : } else {
2569 0 : bdev_module = TAILQ_PREV(g_resume_bdev_module, bdev_module_list, internal.tailq);
2570 : }
2571 :
2572 266 : while (bdev_module) {
2573 198 : if (bdev_module->async_fini_start) {
2574 : /* Save our place so we can resume later. We must
2575 : * save the variable here, before calling fini_start()
2576 : * below, because in some cases the module may immediately
2577 : * call spdk_bdev_module_fini_start_done() and re-enter
2578 : * this function to continue iterating. */
2579 0 : g_resume_bdev_module = bdev_module;
2580 : }
2581 :
2582 198 : if (bdev_module->fini_start) {
2583 24 : bdev_module->fini_start();
2584 : }
2585 :
2586 198 : if (bdev_module->async_fini_start) {
2587 0 : return;
2588 : }
2589 :
2590 198 : bdev_module = TAILQ_PREV(bdev_module, bdev_module_list, internal.tailq);
2591 : }
2592 :
2593 68 : g_resume_bdev_module = NULL;
2594 :
2595 68 : bdev_finish_unregister_bdevs_iter(NULL, 0);
2596 : }
2597 :
2598 : void
2599 0 : spdk_bdev_module_fini_start_done(void)
2600 : {
2601 0 : if (spdk_get_thread() != g_fini_thread) {
2602 0 : spdk_thread_send_msg(g_fini_thread, bdev_module_fini_start_iter, NULL);
2603 : } else {
2604 0 : bdev_module_fini_start_iter(NULL);
2605 : }
2606 0 : }
2607 :
2608 : static void
2609 68 : bdev_finish_wait_for_examine_done(void *cb_arg)
2610 : {
2611 68 : bdev_module_fini_start_iter(NULL);
2612 68 : }
2613 :
2614 : static void bdev_open_async_fini(void);
2615 :
2616 : void
2617 68 : spdk_bdev_finish(spdk_bdev_fini_cb cb_fn, void *cb_arg)
2618 : {
2619 : int rc;
2620 :
2621 68 : assert(cb_fn != NULL);
2622 :
2623 68 : g_fini_thread = spdk_get_thread();
2624 :
2625 68 : g_fini_cb_fn = cb_fn;
2626 68 : g_fini_cb_arg = cb_arg;
2627 :
2628 68 : bdev_open_async_fini();
2629 :
2630 68 : rc = spdk_bdev_wait_for_examine(bdev_finish_wait_for_examine_done, NULL);
2631 68 : if (rc != 0) {
2632 0 : SPDK_ERRLOG("wait_for_examine failed: %s\n", spdk_strerror(-rc));
2633 0 : bdev_finish_wait_for_examine_done(NULL);
2634 : }
2635 68 : }
2636 :
2637 : struct spdk_bdev_io *
2638 699 : bdev_channel_get_io(struct spdk_bdev_channel *channel)
2639 : {
2640 699 : struct spdk_bdev_mgmt_channel *ch = channel->shared_resource->mgmt_ch;
2641 : struct spdk_bdev_io *bdev_io;
2642 :
2643 699 : if (ch->per_thread_cache_count > 0) {
2644 639 : bdev_io = STAILQ_FIRST(&ch->per_thread_cache);
2645 639 : STAILQ_REMOVE_HEAD(&ch->per_thread_cache, internal.buf_link);
2646 639 : ch->per_thread_cache_count--;
2647 60 : } else if (spdk_unlikely(!TAILQ_EMPTY(&ch->io_wait_queue))) {
2648 : /*
2649 : * Don't try to look for bdev_ios in the global pool if there are
2650 : * waiters on bdev_ios - we don't want this caller to jump the line.
2651 : */
2652 0 : bdev_io = NULL;
2653 : } else {
2654 60 : bdev_io = spdk_mempool_get(g_bdev_mgr.bdev_io_pool);
2655 : }
2656 :
2657 699 : return bdev_io;
2658 : }
2659 :
2660 : void
2661 693 : spdk_bdev_free_io(struct spdk_bdev_io *bdev_io)
2662 : {
2663 : struct spdk_bdev_mgmt_channel *ch;
2664 :
2665 693 : assert(bdev_io != NULL);
2666 693 : assert(bdev_io->internal.status != SPDK_BDEV_IO_STATUS_PENDING);
2667 :
2668 693 : ch = bdev_io->internal.ch->shared_resource->mgmt_ch;
2669 :
2670 693 : if (bdev_io->internal.f.has_buf) {
2671 16 : bdev_io_put_buf(bdev_io);
2672 : }
2673 :
2674 693 : if (ch->per_thread_cache_count < ch->bdev_io_cache_size) {
2675 639 : ch->per_thread_cache_count++;
2676 639 : STAILQ_INSERT_HEAD(&ch->per_thread_cache, bdev_io, internal.buf_link);
2677 643 : while (ch->per_thread_cache_count > 0 && !TAILQ_EMPTY(&ch->io_wait_queue)) {
2678 : struct spdk_bdev_io_wait_entry *entry;
2679 :
2680 4 : entry = TAILQ_FIRST(&ch->io_wait_queue);
2681 4 : TAILQ_REMOVE(&ch->io_wait_queue, entry, link);
2682 4 : entry->cb_fn(entry->cb_arg);
2683 : }
2684 : } else {
2685 : /* We should never have a full cache with entries on the io wait queue. */
2686 54 : assert(TAILQ_EMPTY(&ch->io_wait_queue));
2687 54 : spdk_mempool_put(g_bdev_mgr.bdev_io_pool, (void *)bdev_io);
2688 : }
2689 693 : }
2690 :
2691 : static bool
2692 72 : bdev_qos_is_iops_rate_limit(enum spdk_bdev_qos_rate_limit_type limit)
2693 : {
2694 72 : assert(limit != SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES);
2695 :
2696 72 : switch (limit) {
2697 18 : case SPDK_BDEV_QOS_RW_IOPS_RATE_LIMIT:
2698 18 : return true;
2699 54 : case SPDK_BDEV_QOS_RW_BPS_RATE_LIMIT:
2700 : case SPDK_BDEV_QOS_R_BPS_RATE_LIMIT:
2701 : case SPDK_BDEV_QOS_W_BPS_RATE_LIMIT:
2702 54 : return false;
2703 0 : case SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES:
2704 : default:
2705 0 : return false;
2706 : }
2707 : }
2708 :
2709 : static bool
2710 25 : bdev_qos_io_to_limit(struct spdk_bdev_io *bdev_io)
2711 : {
2712 25 : switch (bdev_io->type) {
2713 23 : case SPDK_BDEV_IO_TYPE_NVME_IO:
2714 : case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
2715 : case SPDK_BDEV_IO_TYPE_READ:
2716 : case SPDK_BDEV_IO_TYPE_WRITE:
2717 23 : return true;
2718 0 : case SPDK_BDEV_IO_TYPE_ZCOPY:
2719 0 : if (bdev_io->u.bdev.zcopy.start) {
2720 0 : return true;
2721 : } else {
2722 0 : return false;
2723 : }
2724 2 : default:
2725 2 : return false;
2726 : }
2727 : }
2728 :
2729 : static bool
2730 33 : bdev_is_read_io(struct spdk_bdev_io *bdev_io)
2731 : {
2732 33 : switch (bdev_io->type) {
2733 0 : case SPDK_BDEV_IO_TYPE_NVME_IO:
2734 : case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
2735 : /* Bit 1 (0x2) set for read operation */
2736 0 : if (bdev_io->u.nvme_passthru.cmd.opc & SPDK_NVME_OPC_READ) {
2737 0 : return true;
2738 : } else {
2739 0 : return false;
2740 : }
2741 30 : case SPDK_BDEV_IO_TYPE_READ:
2742 30 : return true;
2743 0 : case SPDK_BDEV_IO_TYPE_ZCOPY:
2744 : /* Populate to read from disk */
2745 0 : if (bdev_io->u.bdev.zcopy.populate) {
2746 0 : return true;
2747 : } else {
2748 0 : return false;
2749 : }
2750 3 : default:
2751 3 : return false;
2752 : }
2753 : }
2754 :
2755 : static uint64_t
2756 43 : bdev_get_io_size_in_byte(struct spdk_bdev_io *bdev_io)
2757 : {
2758 43 : uint32_t blocklen = bdev_io_get_block_size(bdev_io);
2759 :
2760 43 : switch (bdev_io->type) {
2761 0 : case SPDK_BDEV_IO_TYPE_NVME_IO:
2762 : case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
2763 0 : return bdev_io->u.nvme_passthru.nbytes;
2764 43 : case SPDK_BDEV_IO_TYPE_READ:
2765 : case SPDK_BDEV_IO_TYPE_WRITE:
2766 43 : return bdev_io->u.bdev.num_blocks * blocklen;
2767 0 : case SPDK_BDEV_IO_TYPE_ZCOPY:
2768 : /* Track the data in the start phase only */
2769 0 : if (bdev_io->u.bdev.zcopy.start) {
2770 0 : return bdev_io->u.bdev.num_blocks * blocklen;
2771 : } else {
2772 0 : return 0;
2773 : }
2774 0 : default:
2775 0 : return 0;
2776 : }
2777 : }
2778 :
2779 : static inline bool
2780 64 : bdev_qos_rw_queue_io(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io, uint64_t delta)
2781 : {
2782 : int64_t remaining_this_timeslice;
2783 :
2784 64 : if (!limit->max_per_timeslice) {
2785 : /* The QoS is disabled */
2786 0 : return false;
2787 : }
2788 :
2789 64 : remaining_this_timeslice = __atomic_sub_fetch(&limit->remaining_this_timeslice, delta,
2790 : __ATOMIC_RELAXED);
2791 64 : if (remaining_this_timeslice + (int64_t)delta > 0) {
2792 : /* There was still a quota for this delta -> the IO shouldn't be queued
2793 : *
2794 : * We allow a slight quota overrun here so an IO bigger than the per-timeslice
2795 : * quota can be allowed once a while. Such overrun then taken into account in
2796 : * the QoS poller, where the next timeslice quota is calculated.
2797 : */
2798 59 : return false;
2799 : }
2800 :
2801 : /* There was no quota for this delta -> the IO should be queued
2802 : * The remaining_this_timeslice must be rewinded so it reflects the real
2803 : * amount of IOs or bytes allowed.
2804 : */
2805 5 : __atomic_add_fetch(
2806 5 : &limit->remaining_this_timeslice, delta, __ATOMIC_RELAXED);
2807 5 : return true;
2808 : }
2809 :
2810 : static inline void
2811 5 : bdev_qos_rw_rewind_io(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io, uint64_t delta)
2812 : {
2813 5 : __atomic_add_fetch(&limit->remaining_this_timeslice, delta, __ATOMIC_RELAXED);
2814 5 : }
2815 :
2816 : static bool
2817 23 : bdev_qos_rw_iops_queue(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2818 : {
2819 23 : return bdev_qos_rw_queue_io(limit, io, 1);
2820 : }
2821 :
2822 : static void
2823 3 : bdev_qos_rw_iops_rewind_quota(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2824 : {
2825 3 : bdev_qos_rw_rewind_io(limit, io, 1);
2826 3 : }
2827 :
2828 : static bool
2829 41 : bdev_qos_rw_bps_queue(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2830 : {
2831 41 : return bdev_qos_rw_queue_io(limit, io, bdev_get_io_size_in_byte(io));
2832 : }
2833 :
2834 : static void
2835 2 : bdev_qos_rw_bps_rewind_quota(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2836 : {
2837 2 : bdev_qos_rw_rewind_io(limit, io, bdev_get_io_size_in_byte(io));
2838 2 : }
2839 :
2840 : static bool
2841 19 : bdev_qos_r_bps_queue(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2842 : {
2843 19 : if (bdev_is_read_io(io) == false) {
2844 1 : return false;
2845 : }
2846 :
2847 18 : return bdev_qos_rw_bps_queue(limit, io);
2848 : }
2849 :
2850 : static void
2851 0 : bdev_qos_r_bps_rewind_quota(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2852 : {
2853 0 : if (bdev_is_read_io(io) != false) {
2854 0 : bdev_qos_rw_rewind_io(limit, io, bdev_get_io_size_in_byte(io));
2855 : }
2856 0 : }
2857 :
2858 : static bool
2859 14 : bdev_qos_w_bps_queue(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2860 : {
2861 14 : if (bdev_is_read_io(io) == true) {
2862 12 : return false;
2863 : }
2864 :
2865 2 : return bdev_qos_rw_bps_queue(limit, io);
2866 : }
2867 :
2868 : static void
2869 0 : bdev_qos_w_bps_rewind_quota(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2870 : {
2871 0 : if (bdev_is_read_io(io) != true) {
2872 0 : bdev_qos_rw_rewind_io(limit, io, bdev_get_io_size_in_byte(io));
2873 : }
2874 0 : }
2875 :
2876 : static void
2877 10 : bdev_qos_set_ops(struct spdk_bdev_qos *qos)
2878 : {
2879 : int i;
2880 :
2881 50 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
2882 40 : if (qos->rate_limits[i].limit == SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
2883 15 : qos->rate_limits[i].queue_io = NULL;
2884 15 : continue;
2885 : }
2886 :
2887 25 : switch (i) {
2888 9 : case SPDK_BDEV_QOS_RW_IOPS_RATE_LIMIT:
2889 9 : qos->rate_limits[i].queue_io = bdev_qos_rw_iops_queue;
2890 9 : qos->rate_limits[i].rewind_quota = bdev_qos_rw_iops_rewind_quota;
2891 9 : break;
2892 7 : case SPDK_BDEV_QOS_RW_BPS_RATE_LIMIT:
2893 7 : qos->rate_limits[i].queue_io = bdev_qos_rw_bps_queue;
2894 7 : qos->rate_limits[i].rewind_quota = bdev_qos_rw_bps_rewind_quota;
2895 7 : break;
2896 5 : case SPDK_BDEV_QOS_R_BPS_RATE_LIMIT:
2897 5 : qos->rate_limits[i].queue_io = bdev_qos_r_bps_queue;
2898 5 : qos->rate_limits[i].rewind_quota = bdev_qos_r_bps_rewind_quota;
2899 5 : break;
2900 4 : case SPDK_BDEV_QOS_W_BPS_RATE_LIMIT:
2901 4 : qos->rate_limits[i].queue_io = bdev_qos_w_bps_queue;
2902 4 : qos->rate_limits[i].rewind_quota = bdev_qos_w_bps_rewind_quota;
2903 4 : break;
2904 0 : default:
2905 0 : break;
2906 : }
2907 : }
2908 10 : }
2909 :
2910 : static void
2911 6 : _bdev_io_complete_in_submit(struct spdk_bdev_channel *bdev_ch,
2912 : struct spdk_bdev_io *bdev_io,
2913 : enum spdk_bdev_io_status status)
2914 : {
2915 6 : bdev_io->internal.f.in_submit_request = true;
2916 6 : bdev_io_increment_outstanding(bdev_ch, bdev_ch->shared_resource);
2917 6 : spdk_bdev_io_complete(bdev_io, status);
2918 6 : bdev_io->internal.f.in_submit_request = false;
2919 6 : }
2920 :
2921 : static inline void
2922 574 : bdev_io_do_submit(struct spdk_bdev_channel *bdev_ch, struct spdk_bdev_io *bdev_io)
2923 : {
2924 574 : struct spdk_bdev *bdev = bdev_io->bdev;
2925 574 : struct spdk_io_channel *ch = bdev_ch->channel;
2926 574 : struct spdk_bdev_shared_resource *shared_resource = bdev_ch->shared_resource;
2927 :
2928 574 : if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_ABORT)) {
2929 16 : struct spdk_bdev_mgmt_channel *mgmt_channel = shared_resource->mgmt_ch;
2930 16 : struct spdk_bdev_io *bio_to_abort = bdev_io->u.abort.bio_to_abort;
2931 :
2932 32 : if (bdev_abort_queued_io(&shared_resource->nomem_io, bio_to_abort) ||
2933 16 : bdev_abort_buf_io(mgmt_channel, bio_to_abort)) {
2934 0 : _bdev_io_complete_in_submit(bdev_ch, bdev_io,
2935 : SPDK_BDEV_IO_STATUS_SUCCESS);
2936 0 : return;
2937 : }
2938 : }
2939 :
2940 574 : if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE &&
2941 : bdev_io->bdev->split_on_write_unit &&
2942 : bdev_io->u.bdev.num_blocks < bdev_io->bdev->write_unit_size)) {
2943 4 : SPDK_ERRLOG("IO num_blocks %lu does not match the write_unit_size %u\n",
2944 : bdev_io->u.bdev.num_blocks, bdev_io->bdev->write_unit_size);
2945 4 : _bdev_io_complete_in_submit(bdev_ch, bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
2946 4 : return;
2947 : }
2948 :
2949 570 : if (spdk_likely(TAILQ_EMPTY(&shared_resource->nomem_io))) {
2950 527 : bdev_io_increment_outstanding(bdev_ch, shared_resource);
2951 527 : bdev_io->internal.f.in_submit_request = true;
2952 527 : bdev_submit_request(bdev, ch, bdev_io);
2953 527 : bdev_io->internal.f.in_submit_request = false;
2954 : } else {
2955 43 : bdev_queue_nomem_io_tail(shared_resource, bdev_io, BDEV_IO_RETRY_STATE_SUBMIT);
2956 43 : if (shared_resource->nomem_threshold == 0 && shared_resource->io_outstanding == 0) {
2957 : /* Special case when we have nomem IOs and no outstanding IOs which completions
2958 : * could trigger retry of queued IOs */
2959 0 : bdev_shared_ch_retry_io(shared_resource);
2960 : }
2961 : }
2962 : }
2963 :
2964 : static bool
2965 25 : bdev_qos_queue_io(struct spdk_bdev_qos *qos, struct spdk_bdev_io *bdev_io)
2966 : {
2967 : int i;
2968 :
2969 25 : if (bdev_qos_io_to_limit(bdev_io) == true) {
2970 100 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
2971 82 : if (!qos->rate_limits[i].queue_io) {
2972 5 : continue;
2973 : }
2974 :
2975 77 : if (qos->rate_limits[i].queue_io(&qos->rate_limits[i],
2976 : bdev_io) == true) {
2977 10 : for (i -= 1; i >= 0 ; i--) {
2978 5 : if (!qos->rate_limits[i].queue_io) {
2979 0 : continue;
2980 : }
2981 :
2982 5 : qos->rate_limits[i].rewind_quota(&qos->rate_limits[i], bdev_io);
2983 : }
2984 5 : return true;
2985 : }
2986 : }
2987 : }
2988 :
2989 20 : return false;
2990 : }
2991 :
2992 : static int
2993 27 : bdev_qos_io_submit(struct spdk_bdev_channel *ch, struct spdk_bdev_qos *qos)
2994 : {
2995 27 : struct spdk_bdev_io *bdev_io = NULL, *tmp = NULL;
2996 27 : int submitted_ios = 0;
2997 :
2998 52 : TAILQ_FOREACH_SAFE(bdev_io, &ch->qos_queued_io, internal.link, tmp) {
2999 25 : if (!bdev_qos_queue_io(qos, bdev_io)) {
3000 20 : TAILQ_REMOVE(&ch->qos_queued_io, bdev_io, internal.link);
3001 20 : bdev_io_do_submit(ch, bdev_io);
3002 :
3003 20 : submitted_ios++;
3004 : }
3005 : }
3006 :
3007 27 : return submitted_ios;
3008 : }
3009 :
3010 : static void
3011 2 : bdev_queue_io_wait_with_cb(struct spdk_bdev_io *bdev_io, spdk_bdev_io_wait_cb cb_fn)
3012 : {
3013 : int rc;
3014 :
3015 2 : bdev_io->internal.waitq_entry.bdev = bdev_io->bdev;
3016 2 : bdev_io->internal.waitq_entry.cb_fn = cb_fn;
3017 2 : bdev_io->internal.waitq_entry.cb_arg = bdev_io;
3018 2 : rc = spdk_bdev_queue_io_wait(bdev_io->bdev, spdk_io_channel_from_ctx(bdev_io->internal.ch),
3019 : &bdev_io->internal.waitq_entry);
3020 2 : if (rc != 0) {
3021 0 : SPDK_ERRLOG("Queue IO failed, rc=%d\n", rc);
3022 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
3023 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
3024 : }
3025 2 : }
3026 :
3027 : static bool
3028 621 : bdev_rw_should_split(struct spdk_bdev_io *bdev_io)
3029 : {
3030 : uint32_t io_boundary;
3031 621 : struct spdk_bdev *bdev = bdev_io->bdev;
3032 621 : uint32_t max_segment_size = bdev->max_segment_size;
3033 621 : uint32_t max_size = bdev->max_rw_size;
3034 621 : int max_segs = bdev->max_num_segments;
3035 :
3036 621 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE && bdev->split_on_write_unit) {
3037 24 : io_boundary = bdev->write_unit_size;
3038 597 : } else if (bdev->split_on_optimal_io_boundary) {
3039 168 : io_boundary = bdev->optimal_io_boundary;
3040 : } else {
3041 429 : io_boundary = 0;
3042 : }
3043 :
3044 621 : if (spdk_likely(!io_boundary && !max_segs && !max_segment_size && !max_size)) {
3045 243 : return false;
3046 : }
3047 :
3048 378 : if (io_boundary) {
3049 : uint64_t start_stripe, end_stripe;
3050 :
3051 192 : start_stripe = bdev_io->u.bdev.offset_blocks;
3052 192 : end_stripe = start_stripe + bdev_io->u.bdev.num_blocks - 1;
3053 : /* Avoid expensive div operations if possible. These spdk_u32 functions are very cheap. */
3054 192 : if (spdk_likely(spdk_u32_is_pow2(io_boundary))) {
3055 192 : start_stripe >>= spdk_u32log2(io_boundary);
3056 192 : end_stripe >>= spdk_u32log2(io_boundary);
3057 : } else {
3058 0 : start_stripe /= io_boundary;
3059 0 : end_stripe /= io_boundary;
3060 : }
3061 :
3062 192 : if (start_stripe != end_stripe) {
3063 75 : return true;
3064 : }
3065 : }
3066 :
3067 303 : if (max_segs) {
3068 150 : if (bdev_io->u.bdev.iovcnt > max_segs) {
3069 15 : return true;
3070 : }
3071 : }
3072 :
3073 288 : if (max_segment_size) {
3074 470 : for (int i = 0; i < bdev_io->u.bdev.iovcnt; i++) {
3075 346 : if (bdev_io->u.bdev.iovs[i].iov_len > max_segment_size) {
3076 12 : return true;
3077 : }
3078 : }
3079 : }
3080 :
3081 276 : if (max_size) {
3082 52 : if (bdev_io->u.bdev.num_blocks > max_size) {
3083 7 : return true;
3084 : }
3085 : }
3086 :
3087 269 : return false;
3088 : }
3089 :
3090 : static bool
3091 24 : bdev_unmap_should_split(struct spdk_bdev_io *bdev_io)
3092 : {
3093 : uint32_t num_unmap_segments;
3094 :
3095 24 : if (!bdev_io->bdev->max_unmap || !bdev_io->bdev->max_unmap_segments) {
3096 3 : return false;
3097 : }
3098 21 : num_unmap_segments = spdk_divide_round_up(bdev_io->u.bdev.num_blocks, bdev_io->bdev->max_unmap);
3099 21 : if (num_unmap_segments > bdev_io->bdev->max_unmap_segments) {
3100 4 : return true;
3101 : }
3102 :
3103 17 : return false;
3104 : }
3105 :
3106 : static bool
3107 37 : bdev_write_zeroes_should_split(struct spdk_bdev_io *bdev_io)
3108 : {
3109 37 : if (!bdev_io->bdev->max_write_zeroes) {
3110 4 : return false;
3111 : }
3112 :
3113 33 : if (bdev_io->u.bdev.num_blocks > bdev_io->bdev->max_write_zeroes) {
3114 10 : return true;
3115 : }
3116 :
3117 23 : return false;
3118 : }
3119 :
3120 : static bool
3121 30 : bdev_copy_should_split(struct spdk_bdev_io *bdev_io)
3122 : {
3123 30 : if (bdev_io->bdev->max_copy != 0 &&
3124 25 : bdev_io->u.bdev.num_blocks > bdev_io->bdev->max_copy) {
3125 6 : return true;
3126 : }
3127 :
3128 24 : return false;
3129 : }
3130 :
3131 : static bool
3132 794 : bdev_io_should_split(struct spdk_bdev_io *bdev_io)
3133 : {
3134 794 : switch (bdev_io->type) {
3135 621 : case SPDK_BDEV_IO_TYPE_READ:
3136 : case SPDK_BDEV_IO_TYPE_WRITE:
3137 621 : return bdev_rw_should_split(bdev_io);
3138 24 : case SPDK_BDEV_IO_TYPE_UNMAP:
3139 24 : return bdev_unmap_should_split(bdev_io);
3140 37 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3141 37 : return bdev_write_zeroes_should_split(bdev_io);
3142 30 : case SPDK_BDEV_IO_TYPE_COPY:
3143 30 : return bdev_copy_should_split(bdev_io);
3144 82 : default:
3145 82 : return false;
3146 : }
3147 : }
3148 :
3149 : static uint32_t
3150 249 : _to_next_boundary(uint64_t offset, uint32_t boundary)
3151 : {
3152 249 : return (boundary - (offset % boundary));
3153 : }
3154 :
3155 : static void bdev_io_split_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg);
3156 :
3157 : static void _bdev_rw_split(void *_bdev_io);
3158 :
3159 : static void bdev_unmap_split(struct spdk_bdev_io *bdev_io);
3160 :
3161 : static void
3162 0 : _bdev_unmap_split(void *_bdev_io)
3163 : {
3164 0 : return bdev_unmap_split((struct spdk_bdev_io *)_bdev_io);
3165 : }
3166 :
3167 : static void bdev_write_zeroes_split(struct spdk_bdev_io *bdev_io);
3168 :
3169 : static void
3170 0 : _bdev_write_zeroes_split(void *_bdev_io)
3171 : {
3172 0 : return bdev_write_zeroes_split((struct spdk_bdev_io *)_bdev_io);
3173 : }
3174 :
3175 : static void bdev_copy_split(struct spdk_bdev_io *bdev_io);
3176 :
3177 : static void
3178 0 : _bdev_copy_split(void *_bdev_io)
3179 : {
3180 0 : return bdev_copy_split((struct spdk_bdev_io *)_bdev_io);
3181 : }
3182 :
3183 : static int
3184 305 : bdev_io_split_submit(struct spdk_bdev_io *bdev_io, struct iovec *iov, int iovcnt, void *md_buf,
3185 : uint64_t num_blocks, uint64_t *offset, uint64_t *remaining)
3186 : {
3187 : int rc;
3188 : uint64_t current_offset, current_remaining, current_src_offset;
3189 : spdk_bdev_io_wait_cb io_wait_fn;
3190 :
3191 305 : current_offset = *offset;
3192 305 : current_remaining = *remaining;
3193 :
3194 305 : assert(bdev_io->internal.f.split);
3195 :
3196 305 : bdev_io->internal.split.outstanding++;
3197 :
3198 305 : io_wait_fn = _bdev_rw_split;
3199 305 : switch (bdev_io->type) {
3200 196 : case SPDK_BDEV_IO_TYPE_READ:
3201 196 : assert(bdev_io->u.bdev.accel_sequence == NULL);
3202 784 : rc = bdev_readv_blocks_with_md(bdev_io->internal.desc,
3203 196 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
3204 : iov, iovcnt, md_buf, current_offset,
3205 : num_blocks,
3206 196 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain : NULL,
3207 196 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain_ctx : NULL,
3208 : NULL,
3209 : bdev_io->u.bdev.dif_check_flags,
3210 : bdev_io_split_done, bdev_io);
3211 196 : break;
3212 50 : case SPDK_BDEV_IO_TYPE_WRITE:
3213 50 : assert(bdev_io->u.bdev.accel_sequence == NULL);
3214 200 : rc = bdev_writev_blocks_with_md(bdev_io->internal.desc,
3215 50 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
3216 : iov, iovcnt, md_buf, current_offset,
3217 : num_blocks,
3218 50 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain : NULL,
3219 50 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain_ctx : NULL,
3220 : NULL,
3221 : bdev_io->u.bdev.dif_check_flags,
3222 : bdev_io->u.bdev.nvme_cdw12.raw,
3223 : bdev_io->u.bdev.nvme_cdw13.raw,
3224 : bdev_io_split_done, bdev_io);
3225 50 : break;
3226 17 : case SPDK_BDEV_IO_TYPE_UNMAP:
3227 17 : io_wait_fn = _bdev_unmap_split;
3228 17 : rc = spdk_bdev_unmap_blocks(bdev_io->internal.desc,
3229 17 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
3230 : current_offset, num_blocks,
3231 : bdev_io_split_done, bdev_io);
3232 17 : break;
3233 23 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3234 23 : io_wait_fn = _bdev_write_zeroes_split;
3235 23 : rc = spdk_bdev_write_zeroes_blocks(bdev_io->internal.desc,
3236 23 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
3237 : current_offset, num_blocks,
3238 : bdev_io_split_done, bdev_io);
3239 23 : break;
3240 19 : case SPDK_BDEV_IO_TYPE_COPY:
3241 19 : io_wait_fn = _bdev_copy_split;
3242 19 : current_src_offset = bdev_io->u.bdev.copy.src_offset_blocks +
3243 19 : (current_offset - bdev_io->u.bdev.offset_blocks);
3244 19 : rc = spdk_bdev_copy_blocks(bdev_io->internal.desc,
3245 19 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
3246 : current_offset, current_src_offset, num_blocks,
3247 : bdev_io_split_done, bdev_io);
3248 19 : break;
3249 0 : default:
3250 0 : assert(false);
3251 : rc = -EINVAL;
3252 : break;
3253 : }
3254 :
3255 305 : if (rc == 0) {
3256 301 : current_offset += num_blocks;
3257 301 : current_remaining -= num_blocks;
3258 301 : bdev_io->internal.split.current_offset_blocks = current_offset;
3259 301 : bdev_io->internal.split.remaining_num_blocks = current_remaining;
3260 301 : *offset = current_offset;
3261 301 : *remaining = current_remaining;
3262 : } else {
3263 4 : bdev_io->internal.split.outstanding--;
3264 4 : if (rc == -ENOMEM) {
3265 4 : if (bdev_io->internal.split.outstanding == 0) {
3266 : /* No I/O is outstanding. Hence we should wait here. */
3267 1 : bdev_queue_io_wait_with_cb(bdev_io, io_wait_fn);
3268 : }
3269 : } else {
3270 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
3271 0 : if (bdev_io->internal.split.outstanding == 0) {
3272 0 : bdev_ch_remove_from_io_submitted(bdev_io);
3273 0 : spdk_trace_record(TRACE_BDEV_IO_DONE, bdev_io->internal.ch->trace_id,
3274 : 0, (uintptr_t)bdev_io, bdev_io->internal.caller_ctx,
3275 : bdev_io->internal.ch->queue_depth);
3276 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
3277 : }
3278 : }
3279 : }
3280 :
3281 305 : return rc;
3282 : }
3283 :
3284 : static void
3285 67 : _bdev_rw_split(void *_bdev_io)
3286 : {
3287 : struct iovec *parent_iov, *iov;
3288 67 : struct spdk_bdev_io *bdev_io = _bdev_io;
3289 67 : struct spdk_bdev *bdev = bdev_io->bdev;
3290 67 : uint64_t parent_offset, current_offset, remaining;
3291 : uint32_t parent_iov_offset, parent_iovcnt, parent_iovpos, child_iovcnt;
3292 : uint32_t to_next_boundary, to_next_boundary_bytes, to_last_block_bytes;
3293 : uint32_t iovcnt, iov_len, child_iovsize;
3294 : uint32_t blocklen;
3295 : uint32_t io_boundary;
3296 67 : uint32_t max_segment_size = bdev->max_segment_size;
3297 67 : uint32_t max_child_iovcnt = bdev->max_num_segments;
3298 67 : uint32_t max_size = bdev->max_rw_size;
3299 67 : void *md_buf = NULL;
3300 : int rc;
3301 :
3302 67 : blocklen = bdev_io_get_block_size(bdev_io);
3303 :
3304 67 : max_size = max_size ? max_size : UINT32_MAX;
3305 67 : max_segment_size = max_segment_size ? max_segment_size : UINT32_MAX;
3306 67 : max_child_iovcnt = max_child_iovcnt ? spdk_min(max_child_iovcnt, SPDK_BDEV_IO_NUM_CHILD_IOV) :
3307 : SPDK_BDEV_IO_NUM_CHILD_IOV;
3308 :
3309 67 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE && bdev->split_on_write_unit) {
3310 5 : io_boundary = bdev->write_unit_size;
3311 62 : } else if (bdev->split_on_optimal_io_boundary) {
3312 40 : io_boundary = bdev->optimal_io_boundary;
3313 : } else {
3314 22 : io_boundary = UINT32_MAX;
3315 : }
3316 :
3317 67 : assert(bdev_io->internal.f.split);
3318 :
3319 67 : remaining = bdev_io->internal.split.remaining_num_blocks;
3320 67 : current_offset = bdev_io->internal.split.current_offset_blocks;
3321 67 : parent_offset = bdev_io->u.bdev.offset_blocks;
3322 67 : parent_iov_offset = (current_offset - parent_offset) * blocklen;
3323 67 : parent_iovcnt = bdev_io->u.bdev.iovcnt;
3324 :
3325 420 : for (parent_iovpos = 0; parent_iovpos < parent_iovcnt; parent_iovpos++) {
3326 420 : parent_iov = &bdev_io->u.bdev.iovs[parent_iovpos];
3327 420 : if (parent_iov_offset < parent_iov->iov_len) {
3328 67 : break;
3329 : }
3330 353 : parent_iov_offset -= parent_iov->iov_len;
3331 : }
3332 :
3333 67 : child_iovcnt = 0;
3334 309 : while (remaining > 0 && parent_iovpos < parent_iovcnt &&
3335 : child_iovcnt < SPDK_BDEV_IO_NUM_CHILD_IOV) {
3336 249 : to_next_boundary = _to_next_boundary(current_offset, io_boundary);
3337 249 : to_next_boundary = spdk_min(remaining, to_next_boundary);
3338 249 : to_next_boundary = spdk_min(max_size, to_next_boundary);
3339 249 : to_next_boundary_bytes = to_next_boundary * blocklen;
3340 :
3341 249 : iov = &bdev_io->child_iov[child_iovcnt];
3342 249 : iovcnt = 0;
3343 :
3344 249 : if (bdev_io->u.bdev.md_buf) {
3345 48 : md_buf = (char *)bdev_io->u.bdev.md_buf +
3346 24 : (current_offset - parent_offset) * spdk_bdev_get_md_size(bdev);
3347 : }
3348 :
3349 249 : child_iovsize = spdk_min(SPDK_BDEV_IO_NUM_CHILD_IOV - child_iovcnt, max_child_iovcnt);
3350 974 : while (to_next_boundary_bytes > 0 && parent_iovpos < parent_iovcnt &&
3351 : iovcnt < child_iovsize) {
3352 725 : parent_iov = &bdev_io->u.bdev.iovs[parent_iovpos];
3353 725 : iov_len = parent_iov->iov_len - parent_iov_offset;
3354 :
3355 725 : iov_len = spdk_min(iov_len, max_segment_size);
3356 725 : iov_len = spdk_min(iov_len, to_next_boundary_bytes);
3357 725 : to_next_boundary_bytes -= iov_len;
3358 :
3359 725 : bdev_io->child_iov[child_iovcnt].iov_base = parent_iov->iov_base + parent_iov_offset;
3360 725 : bdev_io->child_iov[child_iovcnt].iov_len = iov_len;
3361 :
3362 725 : if (iov_len < parent_iov->iov_len - parent_iov_offset) {
3363 183 : parent_iov_offset += iov_len;
3364 : } else {
3365 542 : parent_iovpos++;
3366 542 : parent_iov_offset = 0;
3367 : }
3368 725 : child_iovcnt++;
3369 725 : iovcnt++;
3370 : }
3371 :
3372 249 : if (to_next_boundary_bytes > 0) {
3373 : /* We had to stop this child I/O early because we ran out of
3374 : * child_iov space or were limited by max_num_segments.
3375 : * Ensure the iovs to be aligned with block size and
3376 : * then adjust to_next_boundary before starting the
3377 : * child I/O.
3378 : */
3379 111 : assert(child_iovcnt == SPDK_BDEV_IO_NUM_CHILD_IOV ||
3380 : iovcnt == child_iovsize);
3381 111 : to_last_block_bytes = to_next_boundary_bytes % blocklen;
3382 111 : if (to_last_block_bytes != 0) {
3383 24 : uint32_t child_iovpos = child_iovcnt - 1;
3384 : /* don't decrease child_iovcnt when it equals to SPDK_BDEV_IO_NUM_CHILD_IOV
3385 : * so the loop will naturally end
3386 : */
3387 :
3388 24 : to_last_block_bytes = blocklen - to_last_block_bytes;
3389 24 : to_next_boundary_bytes += to_last_block_bytes;
3390 53 : while (to_last_block_bytes > 0 && iovcnt > 0) {
3391 32 : iov_len = spdk_min(to_last_block_bytes,
3392 : bdev_io->child_iov[child_iovpos].iov_len);
3393 32 : bdev_io->child_iov[child_iovpos].iov_len -= iov_len;
3394 32 : if (bdev_io->child_iov[child_iovpos].iov_len == 0) {
3395 15 : child_iovpos--;
3396 15 : if (--iovcnt == 0) {
3397 : /* If the child IO is less than a block size just return.
3398 : * If the first child IO of any split round is less than
3399 : * a block size, an error exit.
3400 : */
3401 3 : if (bdev_io->internal.split.outstanding == 0) {
3402 1 : SPDK_ERRLOG("The first child io was less than a block size\n");
3403 1 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
3404 1 : bdev_ch_remove_from_io_submitted(bdev_io);
3405 1 : spdk_trace_record(TRACE_BDEV_IO_DONE, bdev_io->internal.ch->trace_id,
3406 : 0, (uintptr_t)bdev_io, bdev_io->internal.caller_ctx,
3407 : bdev_io->internal.ch->queue_depth);
3408 1 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
3409 : }
3410 :
3411 3 : return;
3412 : }
3413 : }
3414 :
3415 29 : to_last_block_bytes -= iov_len;
3416 :
3417 29 : if (parent_iov_offset == 0) {
3418 14 : parent_iovpos--;
3419 14 : parent_iov_offset = bdev_io->u.bdev.iovs[parent_iovpos].iov_len;
3420 : }
3421 29 : parent_iov_offset -= iov_len;
3422 : }
3423 :
3424 21 : assert(to_last_block_bytes == 0);
3425 : }
3426 108 : to_next_boundary -= to_next_boundary_bytes / blocklen;
3427 : }
3428 :
3429 246 : rc = bdev_io_split_submit(bdev_io, iov, iovcnt, md_buf, to_next_boundary,
3430 : ¤t_offset, &remaining);
3431 246 : if (spdk_unlikely(rc)) {
3432 4 : return;
3433 : }
3434 : }
3435 : }
3436 :
3437 : static void
3438 3 : bdev_unmap_split(struct spdk_bdev_io *bdev_io)
3439 : {
3440 3 : uint64_t offset, unmap_blocks, remaining, max_unmap_blocks;
3441 3 : uint32_t num_children_reqs = 0;
3442 : int rc;
3443 :
3444 3 : assert(bdev_io->internal.f.split);
3445 :
3446 3 : offset = bdev_io->internal.split.current_offset_blocks;
3447 3 : remaining = bdev_io->internal.split.remaining_num_blocks;
3448 3 : max_unmap_blocks = bdev_io->bdev->max_unmap * bdev_io->bdev->max_unmap_segments;
3449 :
3450 20 : while (remaining && (num_children_reqs < SPDK_BDEV_MAX_CHILDREN_UNMAP_WRITE_ZEROES_REQS)) {
3451 17 : unmap_blocks = spdk_min(remaining, max_unmap_blocks);
3452 :
3453 17 : rc = bdev_io_split_submit(bdev_io, NULL, 0, NULL, unmap_blocks,
3454 : &offset, &remaining);
3455 17 : if (spdk_likely(rc == 0)) {
3456 17 : num_children_reqs++;
3457 : } else {
3458 0 : return;
3459 : }
3460 : }
3461 : }
3462 :
3463 : static void
3464 6 : bdev_write_zeroes_split(struct spdk_bdev_io *bdev_io)
3465 : {
3466 6 : uint64_t offset, write_zeroes_blocks, remaining;
3467 6 : uint32_t num_children_reqs = 0;
3468 : int rc;
3469 :
3470 6 : assert(bdev_io->internal.f.split);
3471 :
3472 6 : offset = bdev_io->internal.split.current_offset_blocks;
3473 6 : remaining = bdev_io->internal.split.remaining_num_blocks;
3474 :
3475 29 : while (remaining && (num_children_reqs < SPDK_BDEV_MAX_CHILDREN_UNMAP_WRITE_ZEROES_REQS)) {
3476 23 : write_zeroes_blocks = spdk_min(remaining, bdev_io->bdev->max_write_zeroes);
3477 :
3478 23 : rc = bdev_io_split_submit(bdev_io, NULL, 0, NULL, write_zeroes_blocks,
3479 : &offset, &remaining);
3480 23 : if (spdk_likely(rc == 0)) {
3481 23 : num_children_reqs++;
3482 : } else {
3483 0 : return;
3484 : }
3485 : }
3486 : }
3487 :
3488 : static void
3489 4 : bdev_copy_split(struct spdk_bdev_io *bdev_io)
3490 : {
3491 4 : uint64_t offset, copy_blocks, remaining;
3492 4 : uint32_t num_children_reqs = 0;
3493 : int rc;
3494 :
3495 4 : assert(bdev_io->internal.f.split);
3496 :
3497 4 : offset = bdev_io->internal.split.current_offset_blocks;
3498 4 : remaining = bdev_io->internal.split.remaining_num_blocks;
3499 :
3500 4 : assert(bdev_io->bdev->max_copy != 0);
3501 23 : while (remaining && (num_children_reqs < SPDK_BDEV_MAX_CHILDREN_COPY_REQS)) {
3502 19 : copy_blocks = spdk_min(remaining, bdev_io->bdev->max_copy);
3503 :
3504 19 : rc = bdev_io_split_submit(bdev_io, NULL, 0, NULL, copy_blocks,
3505 : &offset, &remaining);
3506 19 : if (spdk_likely(rc == 0)) {
3507 19 : num_children_reqs++;
3508 : } else {
3509 0 : return;
3510 : }
3511 : }
3512 : }
3513 :
3514 : static void
3515 58 : parent_bdev_io_complete(void *ctx, int rc)
3516 : {
3517 58 : struct spdk_bdev_io *parent_io = ctx;
3518 :
3519 58 : if (rc) {
3520 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
3521 : }
3522 :
3523 58 : parent_io->internal.cb(parent_io, parent_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS,
3524 : parent_io->internal.caller_ctx);
3525 58 : }
3526 :
3527 : static void
3528 0 : bdev_io_complete_parent_sequence_cb(void *ctx, int status)
3529 : {
3530 0 : struct spdk_bdev_io *bdev_io = ctx;
3531 :
3532 : /* u.bdev.accel_sequence should have already been cleared at this point */
3533 0 : assert(bdev_io->u.bdev.accel_sequence == NULL);
3534 0 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS);
3535 0 : bdev_io->internal.f.has_accel_sequence = false;
3536 :
3537 0 : if (spdk_unlikely(status != 0)) {
3538 0 : SPDK_ERRLOG("Failed to execute accel sequence, status=%d\n", status);
3539 : }
3540 :
3541 0 : parent_bdev_io_complete(bdev_io, status);
3542 0 : }
3543 :
3544 : static void
3545 301 : bdev_io_split_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
3546 : {
3547 301 : struct spdk_bdev_io *parent_io = cb_arg;
3548 :
3549 301 : spdk_bdev_free_io(bdev_io);
3550 :
3551 301 : assert(parent_io->internal.f.split);
3552 :
3553 301 : if (!success) {
3554 21 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
3555 : /* If any child I/O failed, stop further splitting process. */
3556 21 : parent_io->internal.split.current_offset_blocks += parent_io->internal.split.remaining_num_blocks;
3557 21 : parent_io->internal.split.remaining_num_blocks = 0;
3558 : }
3559 301 : parent_io->internal.split.outstanding--;
3560 301 : if (parent_io->internal.split.outstanding != 0) {
3561 223 : return;
3562 : }
3563 :
3564 : /*
3565 : * Parent I/O finishes when all blocks are consumed.
3566 : */
3567 78 : if (parent_io->internal.split.remaining_num_blocks == 0) {
3568 58 : assert(parent_io->internal.cb != bdev_io_split_done);
3569 58 : bdev_ch_remove_from_io_submitted(parent_io);
3570 58 : spdk_trace_record(TRACE_BDEV_IO_DONE, parent_io->internal.ch->trace_id,
3571 : 0, (uintptr_t)parent_io, bdev_io->internal.caller_ctx,
3572 : parent_io->internal.ch->queue_depth);
3573 :
3574 58 : if (spdk_likely(parent_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS)) {
3575 48 : if (bdev_io_needs_sequence_exec(parent_io->internal.desc, parent_io)) {
3576 0 : bdev_io_exec_sequence(parent_io, bdev_io_complete_parent_sequence_cb);
3577 0 : return;
3578 48 : } else if (parent_io->internal.f.has_bounce_buf &&
3579 0 : !bdev_io_use_accel_sequence(bdev_io)) {
3580 : /* bdev IO will be completed in the callback */
3581 0 : _bdev_io_push_bounce_data_buffer(parent_io, parent_bdev_io_complete);
3582 0 : return;
3583 : }
3584 : }
3585 :
3586 58 : parent_bdev_io_complete(parent_io, 0);
3587 58 : return;
3588 : }
3589 :
3590 : /*
3591 : * Continue with the splitting process. This function will complete the parent I/O if the
3592 : * splitting is done.
3593 : */
3594 20 : switch (parent_io->type) {
3595 17 : case SPDK_BDEV_IO_TYPE_READ:
3596 : case SPDK_BDEV_IO_TYPE_WRITE:
3597 17 : _bdev_rw_split(parent_io);
3598 17 : break;
3599 1 : case SPDK_BDEV_IO_TYPE_UNMAP:
3600 1 : bdev_unmap_split(parent_io);
3601 1 : break;
3602 1 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3603 1 : bdev_write_zeroes_split(parent_io);
3604 1 : break;
3605 1 : case SPDK_BDEV_IO_TYPE_COPY:
3606 1 : bdev_copy_split(parent_io);
3607 1 : break;
3608 0 : default:
3609 0 : assert(false);
3610 : break;
3611 : }
3612 : }
3613 :
3614 : static void bdev_rw_split_get_buf_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io,
3615 : bool success);
3616 :
3617 : static void
3618 59 : bdev_io_split(struct spdk_bdev_io *bdev_io)
3619 : {
3620 59 : assert(bdev_io_should_split(bdev_io));
3621 59 : assert(bdev_io->internal.f.split);
3622 :
3623 59 : bdev_io->internal.split.current_offset_blocks = bdev_io->u.bdev.offset_blocks;
3624 59 : bdev_io->internal.split.remaining_num_blocks = bdev_io->u.bdev.num_blocks;
3625 59 : bdev_io->internal.split.outstanding = 0;
3626 59 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
3627 :
3628 59 : switch (bdev_io->type) {
3629 49 : case SPDK_BDEV_IO_TYPE_READ:
3630 : case SPDK_BDEV_IO_TYPE_WRITE:
3631 49 : if (_is_buf_allocated(bdev_io->u.bdev.iovs)) {
3632 49 : _bdev_rw_split(bdev_io);
3633 : } else {
3634 0 : assert(bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
3635 0 : spdk_bdev_io_get_buf(bdev_io, bdev_rw_split_get_buf_cb,
3636 0 : bdev_io->u.bdev.num_blocks * bdev_io_get_block_size(bdev_io));
3637 : }
3638 49 : break;
3639 2 : case SPDK_BDEV_IO_TYPE_UNMAP:
3640 2 : bdev_unmap_split(bdev_io);
3641 2 : break;
3642 5 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3643 5 : bdev_write_zeroes_split(bdev_io);
3644 5 : break;
3645 3 : case SPDK_BDEV_IO_TYPE_COPY:
3646 3 : bdev_copy_split(bdev_io);
3647 3 : break;
3648 0 : default:
3649 0 : assert(false);
3650 : break;
3651 : }
3652 59 : }
3653 :
3654 : static void
3655 0 : bdev_rw_split_get_buf_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io, bool success)
3656 : {
3657 0 : if (!success) {
3658 0 : spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
3659 0 : return;
3660 : }
3661 :
3662 0 : _bdev_rw_split(bdev_io);
3663 : }
3664 :
3665 : static inline void
3666 579 : _bdev_io_submit(struct spdk_bdev_io *bdev_io)
3667 : {
3668 579 : struct spdk_bdev *bdev = bdev_io->bdev;
3669 579 : struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
3670 :
3671 579 : if (spdk_likely(bdev_ch->flags == 0)) {
3672 554 : bdev_io_do_submit(bdev_ch, bdev_io);
3673 554 : return;
3674 : }
3675 :
3676 25 : if (bdev_ch->flags & BDEV_CH_RESET_IN_PROGRESS) {
3677 2 : _bdev_io_complete_in_submit(bdev_ch, bdev_io, SPDK_BDEV_IO_STATUS_ABORTED);
3678 23 : } else if (bdev_ch->flags & BDEV_CH_QOS_ENABLED) {
3679 25 : if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_ABORT) &&
3680 2 : bdev_abort_queued_io(&bdev_ch->qos_queued_io, bdev_io->u.abort.bio_to_abort)) {
3681 0 : _bdev_io_complete_in_submit(bdev_ch, bdev_io, SPDK_BDEV_IO_STATUS_SUCCESS);
3682 : } else {
3683 23 : TAILQ_INSERT_TAIL(&bdev_ch->qos_queued_io, bdev_io, internal.link);
3684 23 : bdev_qos_io_submit(bdev_ch, bdev->internal.qos);
3685 : }
3686 : } else {
3687 0 : SPDK_ERRLOG("unknown bdev_ch flag %x found\n", bdev_ch->flags);
3688 0 : _bdev_io_complete_in_submit(bdev_ch, bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
3689 : }
3690 : }
3691 :
3692 : bool bdev_lba_range_overlapped(struct lba_range *range1, struct lba_range *range2);
3693 :
3694 : bool
3695 23 : bdev_lba_range_overlapped(struct lba_range *range1, struct lba_range *range2)
3696 : {
3697 23 : if (range1->length == 0 || range2->length == 0) {
3698 1 : return false;
3699 : }
3700 :
3701 22 : if (range1->offset + range1->length <= range2->offset) {
3702 1 : return false;
3703 : }
3704 :
3705 21 : if (range2->offset + range2->length <= range1->offset) {
3706 3 : return false;
3707 : }
3708 :
3709 18 : return true;
3710 : }
3711 :
3712 : static bool
3713 11 : bdev_io_range_is_locked(struct spdk_bdev_io *bdev_io, struct lba_range *range)
3714 : {
3715 11 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
3716 11 : struct lba_range r;
3717 :
3718 11 : switch (bdev_io->type) {
3719 0 : case SPDK_BDEV_IO_TYPE_NVME_IO:
3720 : case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
3721 : /* Don't try to decode the NVMe command - just assume worst-case and that
3722 : * it overlaps a locked range.
3723 : */
3724 0 : return true;
3725 6 : case SPDK_BDEV_IO_TYPE_READ:
3726 6 : if (!range->quiesce) {
3727 4 : return false;
3728 : }
3729 : /* fallthrough */
3730 : case SPDK_BDEV_IO_TYPE_WRITE:
3731 : case SPDK_BDEV_IO_TYPE_UNMAP:
3732 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3733 : case SPDK_BDEV_IO_TYPE_ZCOPY:
3734 : case SPDK_BDEV_IO_TYPE_COPY:
3735 7 : r.offset = bdev_io->u.bdev.offset_blocks;
3736 7 : r.length = bdev_io->u.bdev.num_blocks;
3737 7 : if (!bdev_lba_range_overlapped(range, &r)) {
3738 : /* This I/O doesn't overlap the specified LBA range. */
3739 0 : return false;
3740 7 : } else if (range->owner_ch == ch && range->locked_ctx == bdev_io->internal.caller_ctx) {
3741 : /* This I/O overlaps, but the I/O is on the same channel that locked this
3742 : * range, and the caller_ctx is the same as the locked_ctx. This means
3743 : * that this I/O is associated with the lock, and is allowed to execute.
3744 : */
3745 2 : return false;
3746 : } else {
3747 5 : return true;
3748 : }
3749 0 : default:
3750 0 : return false;
3751 : }
3752 : }
3753 :
3754 : void
3755 639 : bdev_io_submit(struct spdk_bdev_io *bdev_io)
3756 : {
3757 639 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
3758 :
3759 639 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_PENDING);
3760 :
3761 639 : if (!TAILQ_EMPTY(&ch->locked_ranges)) {
3762 : struct lba_range *range;
3763 :
3764 13 : TAILQ_FOREACH(range, &ch->locked_ranges, tailq) {
3765 8 : if (bdev_io_range_is_locked(bdev_io, range)) {
3766 3 : TAILQ_INSERT_TAIL(&ch->io_locked, bdev_io, internal.ch_link);
3767 3 : return;
3768 : }
3769 : }
3770 : }
3771 :
3772 636 : bdev_ch_add_to_io_submitted(bdev_io);
3773 :
3774 636 : bdev_io->internal.submit_tsc = spdk_get_ticks();
3775 636 : spdk_trace_record_tsc(bdev_io->internal.submit_tsc, TRACE_BDEV_IO_START,
3776 : ch->trace_id, bdev_io->u.bdev.num_blocks,
3777 : (uintptr_t)bdev_io, (uint64_t)bdev_io->type, bdev_io->internal.caller_ctx,
3778 : bdev_io->u.bdev.offset_blocks, ch->queue_depth);
3779 :
3780 636 : if (bdev_io->internal.f.split) {
3781 59 : bdev_io_split(bdev_io);
3782 59 : return;
3783 : }
3784 :
3785 577 : _bdev_io_submit(bdev_io);
3786 : }
3787 :
3788 : static inline int
3789 2 : bdev_io_init_dif_ctx(struct spdk_bdev_io *bdev_io)
3790 : {
3791 2 : struct spdk_bdev *bdev = bdev_io->bdev;
3792 2 : struct spdk_dif_ctx_init_ext_opts dif_opts;
3793 :
3794 2 : memset(&bdev_io->u.bdev.dif_err, 0, sizeof(struct spdk_dif_error));
3795 :
3796 2 : dif_opts.size = SPDK_SIZEOF(&dif_opts, dif_pi_format);
3797 2 : dif_opts.dif_pi_format = bdev->dif_pi_format;
3798 :
3799 4 : return spdk_dif_ctx_init(&bdev_io->u.bdev.dif_ctx,
3800 : bdev->blocklen,
3801 : bdev->md_len,
3802 2 : bdev->md_interleave,
3803 2 : bdev->dif_is_head_of_md,
3804 : bdev->dif_type,
3805 : bdev_io->u.bdev.dif_check_flags,
3806 2 : bdev_io->u.bdev.offset_blocks & 0xFFFFFFFF,
3807 : 0xFFFF, 0, 0, 0, &dif_opts);
3808 : }
3809 :
3810 : static void
3811 4 : _bdev_memory_domain_get_io_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io,
3812 : bool success)
3813 : {
3814 4 : if (!success) {
3815 0 : SPDK_ERRLOG("Failed to get data buffer, completing IO\n");
3816 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
3817 0 : bdev_io_complete_unsubmitted(bdev_io);
3818 0 : return;
3819 : }
3820 :
3821 4 : if (bdev_io_needs_sequence_exec(bdev_io->internal.desc, bdev_io)) {
3822 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
3823 0 : bdev_io_exec_sequence(bdev_io, bdev_io_submit_sequence_cb);
3824 0 : return;
3825 : }
3826 : /* For reads we'll execute the sequence after the data is read, so, for now, only
3827 : * clear out accel_sequence pointer and submit the IO */
3828 0 : assert(bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
3829 0 : bdev_io->u.bdev.accel_sequence = NULL;
3830 : }
3831 :
3832 4 : bdev_io_submit(bdev_io);
3833 : }
3834 :
3835 : static inline void
3836 4 : _bdev_io_ext_use_bounce_buffer(struct spdk_bdev_io *bdev_io)
3837 : {
3838 : /* bdev doesn't support memory domains, thereby buffers in this IO request can't
3839 : * be accessed directly. It is needed to allocate buffers before issuing IO operation.
3840 : * For write operation we need to pull buffers from memory domain before submitting IO.
3841 : * Once read operation completes, we need to use memory_domain push functionality to
3842 : * update data in original memory domain IO buffer.
3843 : *
3844 : * If this I/O request is not aware of metadata, buffers in thsi IO request can't be
3845 : * accessed directly too. It is needed to allocate buffers before issuing IO operation.
3846 : * For write operation we need to insert metadata before submitting IO. Once read
3847 : * operation completes, we need to strip metadata in original IO buffer.
3848 : *
3849 : * This IO request will go through a regular IO flow, so clear memory domains pointers */
3850 4 : assert(bdev_io_use_memory_domain(bdev_io) ||
3851 : bdev_io_needs_metadata(bdev_io->internal.desc, bdev_io));
3852 :
3853 4 : bdev_io->u.bdev.memory_domain = NULL;
3854 4 : bdev_io->u.bdev.memory_domain_ctx = NULL;
3855 4 : _bdev_io_get_bounce_buf(bdev_io, _bdev_memory_domain_get_io_cb,
3856 4 : bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen);
3857 4 : }
3858 :
3859 : static inline void
3860 0 : _bdev_io_ext_use_accel_buffer(struct spdk_bdev_io *bdev_io)
3861 : {
3862 0 : assert(bdev_io_use_memory_domain(bdev_io));
3863 0 : assert(bdev_io_needs_metadata(bdev_io->internal.desc, bdev_io));
3864 :
3865 0 : bdev_io->u.bdev.memory_domain = NULL;
3866 0 : bdev_io->u.bdev.memory_domain_ctx = NULL;
3867 0 : bdev_io_get_accel_buf(bdev_io, _bdev_memory_domain_get_io_cb,
3868 0 : bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen);
3869 0 : }
3870 :
3871 : /* We need to allocate bounce buffer
3872 : * - if bdev doesn't support memory domains,
3873 : * - if it does support them, but we need to execute an accel sequence and the data buffer is
3874 : * from accel memory domain (to avoid doing a push/pull from that domain), or
3875 : * - if IO is not aware of metadata.
3876 : */
3877 : static inline bool
3878 292 : bdev_io_needs_bounce_buffer(struct spdk_bdev_desc *desc, struct spdk_bdev_io *bdev_io)
3879 : {
3880 292 : if (bdev_io_use_memory_domain(bdev_io)) {
3881 4 : if (!desc->memory_domains_supported ||
3882 0 : (bdev_io_needs_sequence_exec(desc, bdev_io) &&
3883 0 : (bdev_io->internal.memory_domain == spdk_accel_get_memory_domain() ||
3884 0 : bdev_io_needs_metadata(desc, bdev_io)))) {
3885 4 : return true;
3886 : }
3887 :
3888 0 : return false;
3889 : }
3890 :
3891 288 : if (bdev_io_needs_metadata(desc, bdev_io)) {
3892 0 : return true;
3893 : }
3894 :
3895 288 : return false;
3896 : }
3897 :
3898 : /* We need to allocate fake accel buffer if bdev supports memory domains but IO is not
3899 : * aware of metadata.
3900 : */
3901 : static inline bool
3902 288 : bdev_io_needs_accel_buffer(struct spdk_bdev_desc *desc, struct spdk_bdev_io *bdev_io)
3903 : {
3904 288 : if (bdev_io_needs_metadata(desc, bdev_io)) {
3905 0 : assert(bdev_io_use_memory_domain(bdev_io));
3906 0 : return true;
3907 : }
3908 :
3909 288 : return false;
3910 : }
3911 :
3912 : static inline void
3913 292 : _bdev_io_submit_ext(struct spdk_bdev_desc *desc, struct spdk_bdev_io *bdev_io)
3914 : {
3915 292 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
3916 : int rc;
3917 :
3918 292 : if (spdk_unlikely(ch->flags & BDEV_CH_RESET_IN_PROGRESS)) {
3919 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_ABORTED;
3920 0 : bdev_io_complete_unsubmitted(bdev_io);
3921 0 : return;
3922 : }
3923 :
3924 292 : if (bdev_io_needs_metadata(desc, bdev_io)) {
3925 0 : rc = bdev_io_init_dif_ctx(bdev_io);
3926 0 : if (spdk_unlikely(rc != 0)) {
3927 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
3928 0 : bdev_io_complete_unsubmitted(bdev_io);
3929 0 : return;
3930 : }
3931 : }
3932 :
3933 292 : if (bdev_io_needs_bounce_buffer(desc, bdev_io)) {
3934 4 : _bdev_io_ext_use_bounce_buffer(bdev_io);
3935 4 : return;
3936 : }
3937 :
3938 288 : if (bdev_io_needs_accel_buffer(desc, bdev_io)) {
3939 0 : _bdev_io_ext_use_accel_buffer(bdev_io);
3940 0 : return;
3941 : }
3942 :
3943 288 : if (bdev_io_needs_sequence_exec(desc, bdev_io)) {
3944 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
3945 0 : bdev_io_exec_sequence(bdev_io, bdev_io_submit_sequence_cb);
3946 0 : return;
3947 : }
3948 : /* For reads we'll execute the sequence after the data is read, so, for now, only
3949 : * clear out accel_sequence pointer and submit the IO */
3950 0 : assert(bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
3951 0 : bdev_io->u.bdev.accel_sequence = NULL;
3952 : }
3953 :
3954 288 : bdev_io_submit(bdev_io);
3955 : }
3956 :
3957 : static void
3958 12 : bdev_io_submit_reset(struct spdk_bdev_io *bdev_io)
3959 : {
3960 12 : struct spdk_bdev *bdev = bdev_io->bdev;
3961 12 : struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
3962 12 : struct spdk_io_channel *ch = bdev_ch->channel;
3963 :
3964 12 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_PENDING);
3965 :
3966 12 : bdev_io->internal.f.in_submit_request = true;
3967 12 : bdev_submit_request(bdev, ch, bdev_io);
3968 12 : bdev_io->internal.f.in_submit_request = false;
3969 12 : }
3970 :
3971 : void
3972 693 : bdev_io_init(struct spdk_bdev_io *bdev_io,
3973 : struct spdk_bdev *bdev, void *cb_arg,
3974 : spdk_bdev_io_completion_cb cb)
3975 : {
3976 693 : bdev_io->bdev = bdev;
3977 693 : bdev_io->internal.f.raw = 0;
3978 693 : bdev_io->internal.caller_ctx = cb_arg;
3979 693 : bdev_io->internal.cb = cb;
3980 693 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_PENDING;
3981 693 : bdev_io->internal.f.in_submit_request = false;
3982 693 : bdev_io->internal.error.nvme.cdw0 = 0;
3983 693 : bdev_io->num_retries = 0;
3984 693 : bdev_io->internal.get_buf_cb = NULL;
3985 693 : bdev_io->internal.get_aux_buf_cb = NULL;
3986 693 : bdev_io->internal.data_transfer_cpl = NULL;
3987 693 : bdev_io->internal.f.split = bdev_io_should_split(bdev_io);
3988 693 : }
3989 :
3990 : static bool
3991 540 : bdev_io_type_supported(struct spdk_bdev *bdev, enum spdk_bdev_io_type io_type)
3992 : {
3993 540 : return bdev->fn_table->io_type_supported(bdev->ctxt, io_type);
3994 : }
3995 :
3996 : bool
3997 178 : spdk_bdev_io_type_supported(struct spdk_bdev *bdev, enum spdk_bdev_io_type io_type)
3998 : {
3999 : bool supported;
4000 :
4001 178 : supported = bdev_io_type_supported(bdev, io_type);
4002 :
4003 178 : if (!supported) {
4004 7 : switch (io_type) {
4005 0 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
4006 : /* The bdev layer will emulate write zeroes as long as write is supported. */
4007 0 : supported = bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE);
4008 0 : break;
4009 7 : default:
4010 7 : break;
4011 : }
4012 : }
4013 :
4014 178 : return supported;
4015 : }
4016 :
4017 : static const char *g_io_type_strings[] = {
4018 : [SPDK_BDEV_IO_TYPE_READ] = "read",
4019 : [SPDK_BDEV_IO_TYPE_WRITE] = "write",
4020 : [SPDK_BDEV_IO_TYPE_UNMAP] = "unmap",
4021 : [SPDK_BDEV_IO_TYPE_FLUSH] = "flush",
4022 : [SPDK_BDEV_IO_TYPE_RESET] = "reset",
4023 : [SPDK_BDEV_IO_TYPE_NVME_ADMIN] = "nvme_admin",
4024 : [SPDK_BDEV_IO_TYPE_NVME_IO] = "nvme_io",
4025 : [SPDK_BDEV_IO_TYPE_NVME_IO_MD] = "nvme_io_md",
4026 : [SPDK_BDEV_IO_TYPE_WRITE_ZEROES] = "write_zeroes",
4027 : [SPDK_BDEV_IO_TYPE_ZCOPY] = "zcopy",
4028 : [SPDK_BDEV_IO_TYPE_GET_ZONE_INFO] = "get_zone_info",
4029 : [SPDK_BDEV_IO_TYPE_ZONE_MANAGEMENT] = "zone_management",
4030 : [SPDK_BDEV_IO_TYPE_ZONE_APPEND] = "zone_append",
4031 : [SPDK_BDEV_IO_TYPE_COMPARE] = "compare",
4032 : [SPDK_BDEV_IO_TYPE_COMPARE_AND_WRITE] = "compare_and_write",
4033 : [SPDK_BDEV_IO_TYPE_ABORT] = "abort",
4034 : [SPDK_BDEV_IO_TYPE_SEEK_HOLE] = "seek_hole",
4035 : [SPDK_BDEV_IO_TYPE_SEEK_DATA] = "seek_data",
4036 : [SPDK_BDEV_IO_TYPE_COPY] = "copy",
4037 : [SPDK_BDEV_IO_TYPE_NVME_IOV_MD] = "nvme_iov_md",
4038 : };
4039 :
4040 : const char *
4041 0 : spdk_bdev_get_io_type_name(enum spdk_bdev_io_type io_type)
4042 : {
4043 0 : if (io_type <= SPDK_BDEV_IO_TYPE_INVALID || io_type >= SPDK_BDEV_NUM_IO_TYPES) {
4044 0 : return NULL;
4045 : }
4046 :
4047 0 : return g_io_type_strings[io_type];
4048 : }
4049 :
4050 : int
4051 0 : spdk_bdev_get_io_type(const char *io_type_string)
4052 : {
4053 : int i;
4054 :
4055 0 : for (i = SPDK_BDEV_IO_TYPE_READ; i < SPDK_BDEV_NUM_IO_TYPES; ++i) {
4056 0 : if (!strcmp(io_type_string, g_io_type_strings[i])) {
4057 0 : return i;
4058 : }
4059 : }
4060 :
4061 0 : return -1;
4062 : }
4063 :
4064 : uint64_t
4065 0 : spdk_bdev_io_get_submit_tsc(struct spdk_bdev_io *bdev_io)
4066 : {
4067 0 : return bdev_io->internal.submit_tsc;
4068 : }
4069 :
4070 : bool
4071 0 : spdk_bdev_io_hide_metadata(struct spdk_bdev_io *bdev_io)
4072 : {
4073 0 : return bdev_io->internal.desc->opts.hide_metadata;
4074 : }
4075 :
4076 : int
4077 0 : spdk_bdev_dump_info_json(struct spdk_bdev *bdev, struct spdk_json_write_ctx *w)
4078 : {
4079 0 : if (bdev->fn_table->dump_info_json) {
4080 0 : return bdev->fn_table->dump_info_json(bdev->ctxt, w);
4081 : }
4082 :
4083 0 : return 0;
4084 : }
4085 :
4086 : static void
4087 10 : bdev_qos_update_max_quota_per_timeslice(struct spdk_bdev_qos *qos)
4088 : {
4089 10 : uint32_t max_per_timeslice = 0;
4090 : int i;
4091 :
4092 50 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
4093 40 : if (qos->rate_limits[i].limit == SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
4094 15 : qos->rate_limits[i].max_per_timeslice = 0;
4095 15 : continue;
4096 : }
4097 :
4098 25 : max_per_timeslice = qos->rate_limits[i].limit *
4099 25 : SPDK_BDEV_QOS_TIMESLICE_IN_USEC / SPDK_SEC_TO_USEC;
4100 :
4101 25 : qos->rate_limits[i].max_per_timeslice = spdk_max(max_per_timeslice,
4102 : qos->rate_limits[i].min_per_timeslice);
4103 :
4104 25 : __atomic_store_n(&qos->rate_limits[i].remaining_this_timeslice,
4105 25 : qos->rate_limits[i].max_per_timeslice, __ATOMIC_RELEASE);
4106 : }
4107 :
4108 10 : bdev_qos_set_ops(qos);
4109 10 : }
4110 :
4111 : static void
4112 4 : bdev_channel_submit_qos_io(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
4113 : struct spdk_io_channel *io_ch, void *ctx)
4114 : {
4115 4 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
4116 : int status;
4117 :
4118 4 : bdev_qos_io_submit(bdev_ch, bdev->internal.qos);
4119 :
4120 : /* if all IOs were sent then continue the iteration, otherwise - stop it */
4121 : /* TODO: channels round robing */
4122 4 : status = TAILQ_EMPTY(&bdev_ch->qos_queued_io) ? 0 : 1;
4123 :
4124 4 : spdk_bdev_for_each_channel_continue(i, status);
4125 4 : }
4126 :
4127 :
4128 : static void
4129 2 : bdev_channel_submit_qos_io_done(struct spdk_bdev *bdev, void *ctx, int status)
4130 : {
4131 :
4132 2 : }
4133 :
4134 : static int
4135 3 : bdev_channel_poll_qos(void *arg)
4136 : {
4137 3 : struct spdk_bdev *bdev = arg;
4138 3 : struct spdk_bdev_qos *qos = bdev->internal.qos;
4139 3 : uint64_t now = spdk_get_ticks();
4140 : int i;
4141 : int64_t remaining_last_timeslice;
4142 :
4143 3 : if (spdk_unlikely(qos->thread == NULL)) {
4144 : /* Old QoS was unbound to remove and new QoS is not enabled yet. */
4145 1 : return SPDK_POLLER_IDLE;
4146 : }
4147 :
4148 2 : if (now < (qos->last_timeslice + qos->timeslice_size)) {
4149 : /* We received our callback earlier than expected - return
4150 : * immediately and wait to do accounting until at least one
4151 : * timeslice has actually expired. This should never happen
4152 : * with a well-behaved timer implementation.
4153 : */
4154 0 : return SPDK_POLLER_IDLE;
4155 : }
4156 :
4157 : /* Reset for next round of rate limiting */
4158 10 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
4159 : /* We may have allowed the IOs or bytes to slightly overrun in the last
4160 : * timeslice. remaining_this_timeslice is signed, so if it's negative
4161 : * here, we'll account for the overrun so that the next timeslice will
4162 : * be appropriately reduced.
4163 : */
4164 8 : remaining_last_timeslice = __atomic_exchange_n(&qos->rate_limits[i].remaining_this_timeslice,
4165 : 0, __ATOMIC_RELAXED);
4166 8 : if (remaining_last_timeslice < 0) {
4167 : /* There could be a race condition here as both bdev_qos_rw_queue_io() and bdev_channel_poll_qos()
4168 : * potentially use 2 atomic ops each, so they can intertwine.
4169 : * This race can potentially cause the limits to be a little fuzzy but won't cause any real damage.
4170 : */
4171 0 : __atomic_store_n(&qos->rate_limits[i].remaining_this_timeslice,
4172 : remaining_last_timeslice, __ATOMIC_RELAXED);
4173 : }
4174 : }
4175 :
4176 4 : while (now >= (qos->last_timeslice + qos->timeslice_size)) {
4177 2 : qos->last_timeslice += qos->timeslice_size;
4178 10 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
4179 8 : __atomic_add_fetch(&qos->rate_limits[i].remaining_this_timeslice,
4180 8 : qos->rate_limits[i].max_per_timeslice, __ATOMIC_RELAXED);
4181 : }
4182 : }
4183 :
4184 2 : spdk_bdev_for_each_channel(bdev, bdev_channel_submit_qos_io, qos,
4185 : bdev_channel_submit_qos_io_done);
4186 :
4187 2 : return SPDK_POLLER_BUSY;
4188 : }
4189 :
4190 : static void
4191 75 : bdev_channel_destroy_resource(struct spdk_bdev_channel *ch)
4192 : {
4193 : struct spdk_bdev_shared_resource *shared_resource;
4194 : struct lba_range *range;
4195 :
4196 75 : bdev_free_io_stat(ch->stat);
4197 : #ifdef SPDK_CONFIG_VTUNE
4198 : bdev_free_io_stat(ch->prev_stat);
4199 : #endif
4200 :
4201 75 : while (!TAILQ_EMPTY(&ch->locked_ranges)) {
4202 0 : range = TAILQ_FIRST(&ch->locked_ranges);
4203 0 : TAILQ_REMOVE(&ch->locked_ranges, range, tailq);
4204 0 : free(range);
4205 : }
4206 :
4207 75 : spdk_put_io_channel(ch->channel);
4208 75 : spdk_put_io_channel(ch->accel_channel);
4209 :
4210 75 : shared_resource = ch->shared_resource;
4211 :
4212 75 : assert(TAILQ_EMPTY(&ch->io_locked));
4213 75 : assert(TAILQ_EMPTY(&ch->io_submitted));
4214 75 : assert(TAILQ_EMPTY(&ch->io_accel_exec));
4215 75 : assert(TAILQ_EMPTY(&ch->io_memory_domain));
4216 75 : assert(ch->io_outstanding == 0);
4217 75 : assert(shared_resource->ref > 0);
4218 75 : shared_resource->ref--;
4219 75 : if (shared_resource->ref == 0) {
4220 74 : assert(shared_resource->io_outstanding == 0);
4221 74 : TAILQ_REMOVE(&shared_resource->mgmt_ch->shared_resources, shared_resource, link);
4222 74 : spdk_put_io_channel(spdk_io_channel_from_ctx(shared_resource->mgmt_ch));
4223 74 : spdk_poller_unregister(&shared_resource->nomem_poller);
4224 74 : free(shared_resource);
4225 : }
4226 75 : }
4227 :
4228 : static void
4229 84 : bdev_enable_qos(struct spdk_bdev *bdev, struct spdk_bdev_channel *ch)
4230 : {
4231 84 : struct spdk_bdev_qos *qos = bdev->internal.qos;
4232 : int i;
4233 :
4234 84 : assert(spdk_spin_held(&bdev->internal.spinlock));
4235 :
4236 : /* Rate limiting on this bdev enabled */
4237 84 : if (qos) {
4238 17 : if (qos->ch == NULL) {
4239 : struct spdk_io_channel *io_ch;
4240 :
4241 9 : SPDK_DEBUGLOG(bdev, "Selecting channel %p as QoS channel for bdev %s on thread %p\n", ch,
4242 : bdev->name, spdk_get_thread());
4243 :
4244 : /* No qos channel has been selected, so set one up */
4245 :
4246 : /* Take another reference to ch */
4247 9 : io_ch = spdk_get_io_channel(__bdev_to_io_dev(bdev));
4248 9 : assert(io_ch != NULL);
4249 9 : qos->ch = ch;
4250 :
4251 9 : qos->thread = spdk_io_channel_get_thread(io_ch);
4252 :
4253 45 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
4254 36 : if (bdev_qos_is_iops_rate_limit(i) == true) {
4255 9 : qos->rate_limits[i].min_per_timeslice =
4256 : SPDK_BDEV_QOS_MIN_IO_PER_TIMESLICE;
4257 : } else {
4258 27 : qos->rate_limits[i].min_per_timeslice =
4259 : SPDK_BDEV_QOS_MIN_BYTE_PER_TIMESLICE;
4260 : }
4261 :
4262 36 : if (qos->rate_limits[i].limit == 0) {
4263 2 : qos->rate_limits[i].limit = SPDK_BDEV_QOS_LIMIT_NOT_DEFINED;
4264 : }
4265 : }
4266 9 : bdev_qos_update_max_quota_per_timeslice(qos);
4267 9 : qos->timeslice_size =
4268 9 : SPDK_BDEV_QOS_TIMESLICE_IN_USEC * spdk_get_ticks_hz() / SPDK_SEC_TO_USEC;
4269 9 : qos->last_timeslice = spdk_get_ticks();
4270 9 : qos->poller = SPDK_POLLER_REGISTER(bdev_channel_poll_qos,
4271 : bdev,
4272 : SPDK_BDEV_QOS_TIMESLICE_IN_USEC);
4273 : }
4274 :
4275 17 : ch->flags |= BDEV_CH_QOS_ENABLED;
4276 : }
4277 84 : }
4278 :
4279 : struct poll_timeout_ctx {
4280 : struct spdk_bdev_desc *desc;
4281 : uint64_t timeout_in_sec;
4282 : spdk_bdev_io_timeout_cb cb_fn;
4283 : void *cb_arg;
4284 : };
4285 :
4286 : static void
4287 278 : bdev_desc_free(struct spdk_bdev_desc *desc)
4288 : {
4289 278 : spdk_spin_destroy(&desc->spinlock);
4290 278 : free(desc->media_events_buffer);
4291 278 : free(desc);
4292 278 : }
4293 :
4294 : static void
4295 8 : bdev_channel_poll_timeout_io_done(struct spdk_bdev *bdev, void *_ctx, int status)
4296 : {
4297 8 : struct poll_timeout_ctx *ctx = _ctx;
4298 8 : struct spdk_bdev_desc *desc = ctx->desc;
4299 :
4300 8 : free(ctx);
4301 :
4302 8 : spdk_spin_lock(&desc->spinlock);
4303 8 : desc->refs--;
4304 8 : if (desc->closed == true && desc->refs == 0) {
4305 1 : spdk_spin_unlock(&desc->spinlock);
4306 1 : bdev_desc_free(desc);
4307 1 : return;
4308 : }
4309 7 : spdk_spin_unlock(&desc->spinlock);
4310 : }
4311 :
4312 : static void
4313 13 : bdev_channel_poll_timeout_io(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
4314 : struct spdk_io_channel *io_ch, void *_ctx)
4315 : {
4316 13 : struct poll_timeout_ctx *ctx = _ctx;
4317 13 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
4318 13 : struct spdk_bdev_desc *desc = ctx->desc;
4319 : struct spdk_bdev_io *bdev_io;
4320 : uint64_t now;
4321 :
4322 13 : spdk_spin_lock(&desc->spinlock);
4323 13 : if (desc->closed == true) {
4324 1 : spdk_spin_unlock(&desc->spinlock);
4325 1 : spdk_bdev_for_each_channel_continue(i, -1);
4326 1 : return;
4327 : }
4328 12 : spdk_spin_unlock(&desc->spinlock);
4329 :
4330 12 : now = spdk_get_ticks();
4331 22 : TAILQ_FOREACH(bdev_io, &bdev_ch->io_submitted, internal.ch_link) {
4332 : /* Exclude any I/O that are generated via splitting. */
4333 15 : if (bdev_io->internal.cb == bdev_io_split_done) {
4334 3 : continue;
4335 : }
4336 :
4337 : /* Once we find an I/O that has not timed out, we can immediately
4338 : * exit the loop.
4339 : */
4340 12 : if (now < (bdev_io->internal.submit_tsc +
4341 12 : ctx->timeout_in_sec * spdk_get_ticks_hz())) {
4342 5 : goto end;
4343 : }
4344 :
4345 7 : if (bdev_io->internal.desc == desc) {
4346 7 : ctx->cb_fn(ctx->cb_arg, bdev_io);
4347 : }
4348 : }
4349 :
4350 7 : end:
4351 12 : spdk_bdev_for_each_channel_continue(i, 0);
4352 : }
4353 :
4354 : static int
4355 8 : bdev_poll_timeout_io(void *arg)
4356 : {
4357 8 : struct spdk_bdev_desc *desc = arg;
4358 8 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
4359 : struct poll_timeout_ctx *ctx;
4360 :
4361 8 : ctx = calloc(1, sizeof(struct poll_timeout_ctx));
4362 8 : if (!ctx) {
4363 0 : SPDK_ERRLOG("failed to allocate memory\n");
4364 0 : return SPDK_POLLER_BUSY;
4365 : }
4366 8 : ctx->desc = desc;
4367 8 : ctx->cb_arg = desc->cb_arg;
4368 8 : ctx->cb_fn = desc->cb_fn;
4369 8 : ctx->timeout_in_sec = desc->timeout_in_sec;
4370 :
4371 : /* Take a ref on the descriptor in case it gets closed while we are checking
4372 : * all of the channels.
4373 : */
4374 8 : spdk_spin_lock(&desc->spinlock);
4375 8 : desc->refs++;
4376 8 : spdk_spin_unlock(&desc->spinlock);
4377 :
4378 8 : spdk_bdev_for_each_channel(bdev, bdev_channel_poll_timeout_io, ctx,
4379 : bdev_channel_poll_timeout_io_done);
4380 :
4381 8 : return SPDK_POLLER_BUSY;
4382 : }
4383 :
4384 : int
4385 5 : spdk_bdev_set_timeout(struct spdk_bdev_desc *desc, uint64_t timeout_in_sec,
4386 : spdk_bdev_io_timeout_cb cb_fn, void *cb_arg)
4387 : {
4388 5 : assert(desc->thread == spdk_get_thread());
4389 :
4390 5 : spdk_poller_unregister(&desc->io_timeout_poller);
4391 :
4392 5 : if (timeout_in_sec) {
4393 4 : assert(cb_fn != NULL);
4394 4 : desc->io_timeout_poller = SPDK_POLLER_REGISTER(bdev_poll_timeout_io,
4395 : desc,
4396 : SPDK_BDEV_IO_POLL_INTERVAL_IN_MSEC * SPDK_SEC_TO_USEC /
4397 : 1000);
4398 4 : if (desc->io_timeout_poller == NULL) {
4399 0 : SPDK_ERRLOG("can not register the desc timeout IO poller\n");
4400 0 : return -1;
4401 : }
4402 : }
4403 :
4404 5 : desc->cb_fn = cb_fn;
4405 5 : desc->cb_arg = cb_arg;
4406 5 : desc->timeout_in_sec = timeout_in_sec;
4407 :
4408 5 : return 0;
4409 : }
4410 :
4411 : static int
4412 77 : bdev_channel_create(void *io_device, void *ctx_buf)
4413 : {
4414 77 : struct spdk_bdev *bdev = __bdev_from_io_dev(io_device);
4415 77 : struct spdk_bdev_channel *ch = ctx_buf;
4416 : struct spdk_io_channel *mgmt_io_ch;
4417 : struct spdk_bdev_mgmt_channel *mgmt_ch;
4418 : struct spdk_bdev_shared_resource *shared_resource;
4419 : struct lba_range *range;
4420 :
4421 77 : ch->bdev = bdev;
4422 77 : ch->channel = bdev->fn_table->get_io_channel(bdev->ctxt);
4423 77 : if (!ch->channel) {
4424 2 : return -1;
4425 : }
4426 :
4427 75 : ch->accel_channel = spdk_accel_get_io_channel();
4428 75 : if (!ch->accel_channel) {
4429 0 : spdk_put_io_channel(ch->channel);
4430 0 : return -1;
4431 : }
4432 :
4433 75 : spdk_trace_record(TRACE_BDEV_IOCH_CREATE, bdev->internal.trace_id, 0, 0,
4434 : spdk_thread_get_id(spdk_io_channel_get_thread(ch->channel)));
4435 :
4436 75 : assert(ch->histogram == NULL);
4437 75 : if (bdev->internal.histogram_enabled) {
4438 0 : ch->histogram = spdk_histogram_data_alloc();
4439 0 : if (ch->histogram == NULL) {
4440 0 : SPDK_ERRLOG("Could not allocate histogram\n");
4441 : }
4442 : }
4443 :
4444 75 : mgmt_io_ch = spdk_get_io_channel(&g_bdev_mgr);
4445 75 : if (!mgmt_io_ch) {
4446 0 : spdk_put_io_channel(ch->channel);
4447 0 : spdk_put_io_channel(ch->accel_channel);
4448 0 : return -1;
4449 : }
4450 :
4451 75 : mgmt_ch = __io_ch_to_bdev_mgmt_ch(mgmt_io_ch);
4452 77 : TAILQ_FOREACH(shared_resource, &mgmt_ch->shared_resources, link) {
4453 3 : if (shared_resource->shared_ch == ch->channel) {
4454 1 : spdk_put_io_channel(mgmt_io_ch);
4455 1 : shared_resource->ref++;
4456 1 : break;
4457 : }
4458 : }
4459 :
4460 75 : if (shared_resource == NULL) {
4461 74 : shared_resource = calloc(1, sizeof(*shared_resource));
4462 74 : if (shared_resource == NULL) {
4463 0 : spdk_put_io_channel(ch->channel);
4464 0 : spdk_put_io_channel(ch->accel_channel);
4465 0 : spdk_put_io_channel(mgmt_io_ch);
4466 0 : return -1;
4467 : }
4468 :
4469 74 : shared_resource->mgmt_ch = mgmt_ch;
4470 74 : shared_resource->io_outstanding = 0;
4471 74 : TAILQ_INIT(&shared_resource->nomem_io);
4472 74 : shared_resource->nomem_threshold = 0;
4473 74 : shared_resource->shared_ch = ch->channel;
4474 74 : shared_resource->ref = 1;
4475 74 : TAILQ_INSERT_TAIL(&mgmt_ch->shared_resources, shared_resource, link);
4476 : }
4477 :
4478 75 : ch->io_outstanding = 0;
4479 75 : TAILQ_INIT(&ch->locked_ranges);
4480 75 : TAILQ_INIT(&ch->qos_queued_io);
4481 75 : ch->flags = 0;
4482 75 : ch->trace_id = bdev->internal.trace_id;
4483 75 : ch->shared_resource = shared_resource;
4484 :
4485 75 : TAILQ_INIT(&ch->io_submitted);
4486 75 : TAILQ_INIT(&ch->io_locked);
4487 75 : TAILQ_INIT(&ch->io_accel_exec);
4488 75 : TAILQ_INIT(&ch->io_memory_domain);
4489 :
4490 75 : ch->stat = bdev_alloc_io_stat(false);
4491 75 : if (ch->stat == NULL) {
4492 0 : bdev_channel_destroy_resource(ch);
4493 0 : return -1;
4494 : }
4495 :
4496 75 : 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 75 : spdk_spin_lock(&bdev->internal.spinlock);
4520 75 : bdev_enable_qos(bdev, ch);
4521 :
4522 76 : 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 : }
4536 :
4537 75 : spdk_spin_unlock(&bdev->internal.spinlock);
4538 :
4539 75 : return 0;
4540 : }
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 : }
4556 :
4557 0 : return 0;
4558 : }
4559 :
4560 : /*
4561 : * Abort I/O that are waiting on a data buffer.
4562 : */
4563 : static void
4564 98 : bdev_abort_all_buf_io(struct spdk_bdev_mgmt_channel *mgmt_ch, struct spdk_bdev_channel *ch)
4565 : {
4566 98 : spdk_iobuf_for_each_entry(&mgmt_ch->iobuf, bdev_abort_all_buf_io_cb, ch);
4567 98 : }
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 117 : 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 156 : TAILQ_FOREACH_SAFE(bdev_io, queue, internal.link, tmp) {
4579 39 : if (bdev_io->internal.ch == ch) {
4580 39 : 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 39 : if (bdev_io->type != SPDK_BDEV_IO_TYPE_RESET) {
4588 39 : bdev_io_increment_outstanding(ch, ch->shared_resource);
4589 : }
4590 39 : spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_ABORTED);
4591 : }
4592 : }
4593 117 : }
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 : }
4607 :
4608 18 : return false;
4609 : }
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 : }
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 : }
4691 :
4692 7 : bdev->internal.qos = new_qos;
4693 :
4694 7 : if (old_qos->thread == NULL) {
4695 0 : free(old_qos);
4696 : } 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 : }
4706 :
4707 : void
4708 79 : spdk_bdev_add_io_stat(struct spdk_bdev_io_stat *total, struct spdk_bdev_io_stat *add)
4709 : {
4710 79 : total->bytes_read += add->bytes_read;
4711 79 : total->num_read_ops += add->num_read_ops;
4712 79 : total->bytes_written += add->bytes_written;
4713 79 : total->num_write_ops += add->num_write_ops;
4714 79 : total->bytes_unmapped += add->bytes_unmapped;
4715 79 : total->num_unmap_ops += add->num_unmap_ops;
4716 79 : total->bytes_copied += add->bytes_copied;
4717 79 : total->num_copy_ops += add->num_copy_ops;
4718 79 : total->read_latency_ticks += add->read_latency_ticks;
4719 79 : total->write_latency_ticks += add->write_latency_ticks;
4720 79 : total->unmap_latency_ticks += add->unmap_latency_ticks;
4721 79 : total->copy_latency_ticks += add->copy_latency_ticks;
4722 79 : if (total->max_read_latency_ticks < add->max_read_latency_ticks) {
4723 7 : total->max_read_latency_ticks = add->max_read_latency_ticks;
4724 : }
4725 79 : if (total->min_read_latency_ticks > add->min_read_latency_ticks) {
4726 39 : total->min_read_latency_ticks = add->min_read_latency_ticks;
4727 : }
4728 79 : if (total->max_write_latency_ticks < add->max_write_latency_ticks) {
4729 4 : total->max_write_latency_ticks = add->max_write_latency_ticks;
4730 : }
4731 79 : if (total->min_write_latency_ticks > add->min_write_latency_ticks) {
4732 24 : total->min_write_latency_ticks = add->min_write_latency_ticks;
4733 : }
4734 79 : if (total->max_unmap_latency_ticks < add->max_unmap_latency_ticks) {
4735 0 : total->max_unmap_latency_ticks = add->max_unmap_latency_ticks;
4736 : }
4737 79 : if (total->min_unmap_latency_ticks > add->min_unmap_latency_ticks) {
4738 3 : total->min_unmap_latency_ticks = add->min_unmap_latency_ticks;
4739 : }
4740 79 : if (total->max_copy_latency_ticks < add->max_copy_latency_ticks) {
4741 0 : total->max_copy_latency_ticks = add->max_copy_latency_ticks;
4742 : }
4743 79 : if (total->min_copy_latency_ticks > add->min_copy_latency_ticks) {
4744 4 : total->min_copy_latency_ticks = add->min_copy_latency_ticks;
4745 : }
4746 79 : }
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 : }
4757 5 : }
4758 :
4759 : void
4760 216 : spdk_bdev_reset_io_stat(struct spdk_bdev_io_stat *stat, enum spdk_bdev_reset_stat_mode mode)
4761 : {
4762 216 : if (mode == SPDK_BDEV_RESET_STAT_NONE) {
4763 5 : return;
4764 : }
4765 :
4766 211 : stat->max_read_latency_ticks = 0;
4767 211 : stat->min_read_latency_ticks = UINT64_MAX;
4768 211 : stat->max_write_latency_ticks = 0;
4769 211 : stat->min_write_latency_ticks = UINT64_MAX;
4770 211 : stat->max_unmap_latency_ticks = 0;
4771 211 : stat->min_unmap_latency_ticks = UINT64_MAX;
4772 211 : stat->max_copy_latency_ticks = 0;
4773 211 : stat->min_copy_latency_ticks = UINT64_MAX;
4774 :
4775 211 : if (mode != SPDK_BDEV_RESET_STAT_ALL) {
4776 2 : return;
4777 : }
4778 :
4779 209 : stat->bytes_read = 0;
4780 209 : stat->num_read_ops = 0;
4781 209 : stat->bytes_written = 0;
4782 209 : stat->num_write_ops = 0;
4783 209 : stat->bytes_unmapped = 0;
4784 209 : stat->num_unmap_ops = 0;
4785 209 : stat->bytes_copied = 0;
4786 209 : stat->num_copy_ops = 0;
4787 209 : stat->read_latency_ticks = 0;
4788 209 : stat->write_latency_ticks = 0;
4789 209 : stat->unmap_latency_ticks = 0;
4790 209 : stat->copy_latency_ticks = 0;
4791 :
4792 209 : if (stat->io_error != NULL) {
4793 133 : memset(stat->io_error, 0, sizeof(struct spdk_bdev_io_error_stat));
4794 : }
4795 : }
4796 :
4797 : struct spdk_bdev_io_stat *
4798 207 : bdev_alloc_io_stat(bool io_error_stat)
4799 : {
4800 : struct spdk_bdev_io_stat *stat;
4801 :
4802 207 : stat = malloc(sizeof(struct spdk_bdev_io_stat));
4803 207 : if (stat == NULL) {
4804 0 : return NULL;
4805 : }
4806 :
4807 207 : if (io_error_stat) {
4808 132 : stat->io_error = malloc(sizeof(struct spdk_bdev_io_error_stat));
4809 132 : if (stat->io_error == NULL) {
4810 0 : free(stat);
4811 0 : return NULL;
4812 : }
4813 : } else {
4814 75 : stat->io_error = NULL;
4815 : }
4816 :
4817 207 : spdk_bdev_reset_io_stat(stat, SPDK_BDEV_RESET_STAT_ALL);
4818 :
4819 207 : return stat;
4820 : }
4821 :
4822 : void
4823 207 : bdev_free_io_stat(struct spdk_bdev_io_stat *stat)
4824 : {
4825 207 : if (stat != NULL) {
4826 207 : free(stat->io_error);
4827 207 : free(stat);
4828 : }
4829 207 : }
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 : 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 : 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 : 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 : 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 : }
4872 : }
4873 0 : spdk_json_write_object_end(w);
4874 : }
4875 0 : }
4876 :
4877 : static void
4878 79 : bdev_channel_abort_queued_ios(struct spdk_bdev_channel *ch)
4879 : {
4880 79 : struct spdk_bdev_shared_resource *shared_resource = ch->shared_resource;
4881 79 : struct spdk_bdev_mgmt_channel *mgmt_ch = shared_resource->mgmt_ch;
4882 :
4883 79 : bdev_abort_all_queued_io(&shared_resource->nomem_io, ch);
4884 79 : bdev_abort_all_buf_io(mgmt_ch, ch);
4885 79 : }
4886 :
4887 : static void
4888 75 : bdev_channel_destroy(void *io_device, void *ctx_buf)
4889 : {
4890 75 : struct spdk_bdev_channel *ch = ctx_buf;
4891 :
4892 75 : SPDK_DEBUGLOG(bdev, "Destroying channel %p for bdev %s on thread %p\n", ch, ch->bdev->name,
4893 : spdk_get_thread());
4894 :
4895 75 : 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 75 : spdk_spin_lock(&ch->bdev->internal.spinlock);
4900 75 : spdk_bdev_add_io_stat(ch->bdev->internal.stat, ch->stat);
4901 75 : spdk_spin_unlock(&ch->bdev->internal.spinlock);
4902 :
4903 75 : bdev_channel_abort_queued_ios(ch);
4904 :
4905 75 : if (ch->histogram) {
4906 0 : spdk_histogram_data_free(ch->histogram);
4907 : }
4908 :
4909 75 : bdev_channel_destroy_resource(ch);
4910 75 : }
4911 :
4912 : /*
4913 : * If the name already exists in the global bdev name tree, RB_INSERT() returns a pointer
4914 : * to it. Hence we do not have to call bdev_get_by_name() when using this function.
4915 : */
4916 : static int
4917 267 : bdev_name_add(struct spdk_bdev_name *bdev_name, struct spdk_bdev *bdev, const char *name)
4918 : {
4919 : struct spdk_bdev_name *tmp;
4920 :
4921 267 : bdev_name->name = strdup(name);
4922 267 : if (bdev_name->name == NULL) {
4923 0 : SPDK_ERRLOG("Unable to allocate bdev name\n");
4924 0 : return -ENOMEM;
4925 : }
4926 :
4927 267 : bdev_name->bdev = bdev;
4928 :
4929 267 : spdk_spin_lock(&g_bdev_mgr.spinlock);
4930 267 : tmp = RB_INSERT(bdev_name_tree, &g_bdev_mgr.bdev_names, bdev_name);
4931 267 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
4932 :
4933 267 : if (tmp != NULL) {
4934 4 : SPDK_ERRLOG("Bdev name %s already exists\n", name);
4935 4 : free(bdev_name->name);
4936 4 : return -EEXIST;
4937 : }
4938 :
4939 263 : return 0;
4940 : }
4941 :
4942 : static void
4943 263 : bdev_name_del_unsafe(struct spdk_bdev_name *bdev_name)
4944 : {
4945 263 : RB_REMOVE(bdev_name_tree, &g_bdev_mgr.bdev_names, bdev_name);
4946 263 : free(bdev_name->name);
4947 263 : }
4948 :
4949 : static void
4950 5 : bdev_name_del(struct spdk_bdev_name *bdev_name)
4951 : {
4952 5 : spdk_spin_lock(&g_bdev_mgr.spinlock);
4953 5 : bdev_name_del_unsafe(bdev_name);
4954 5 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
4955 5 : }
4956 :
4957 : int
4958 138 : spdk_bdev_alias_add(struct spdk_bdev *bdev, const char *alias)
4959 : {
4960 : struct spdk_bdev_alias *tmp;
4961 : int ret;
4962 :
4963 138 : if (alias == NULL) {
4964 1 : SPDK_ERRLOG("Empty alias passed\n");
4965 1 : return -EINVAL;
4966 : }
4967 :
4968 137 : tmp = calloc(1, sizeof(*tmp));
4969 137 : if (tmp == NULL) {
4970 0 : SPDK_ERRLOG("Unable to allocate alias\n");
4971 0 : return -ENOMEM;
4972 : }
4973 :
4974 137 : ret = bdev_name_add(&tmp->alias, bdev, alias);
4975 137 : if (ret != 0) {
4976 4 : free(tmp);
4977 4 : return ret;
4978 : }
4979 :
4980 133 : TAILQ_INSERT_TAIL(&bdev->aliases, tmp, tailq);
4981 :
4982 133 : return 0;
4983 : }
4984 :
4985 : static int
4986 134 : bdev_alias_del(struct spdk_bdev *bdev, const char *alias,
4987 : void (*alias_del_fn)(struct spdk_bdev_name *n))
4988 : {
4989 : struct spdk_bdev_alias *tmp;
4990 :
4991 139 : TAILQ_FOREACH(tmp, &bdev->aliases, tailq) {
4992 135 : if (strcmp(alias, tmp->alias.name) == 0) {
4993 130 : TAILQ_REMOVE(&bdev->aliases, tmp, tailq);
4994 130 : alias_del_fn(&tmp->alias);
4995 130 : free(tmp);
4996 130 : return 0;
4997 : }
4998 : }
4999 :
5000 4 : return -ENOENT;
5001 : }
5002 :
5003 : int
5004 4 : spdk_bdev_alias_del(struct spdk_bdev *bdev, const char *alias)
5005 : {
5006 : int rc;
5007 :
5008 4 : rc = bdev_alias_del(bdev, alias, bdev_name_del);
5009 4 : if (rc == -ENOENT) {
5010 2 : SPDK_INFOLOG(bdev, "Alias %s does not exist\n", alias);
5011 : }
5012 :
5013 4 : return rc;
5014 : }
5015 :
5016 : void
5017 2 : spdk_bdev_alias_del_all(struct spdk_bdev *bdev)
5018 : {
5019 : struct spdk_bdev_alias *p, *tmp;
5020 :
5021 5 : TAILQ_FOREACH_SAFE(p, &bdev->aliases, tailq, tmp) {
5022 3 : TAILQ_REMOVE(&bdev->aliases, p, tailq);
5023 3 : bdev_name_del(&p->alias);
5024 3 : free(p);
5025 : }
5026 2 : }
5027 :
5028 : struct spdk_io_channel *
5029 77 : spdk_bdev_get_io_channel(struct spdk_bdev_desc *desc)
5030 : {
5031 77 : return spdk_get_io_channel(__bdev_to_io_dev(spdk_bdev_desc_get_bdev(desc)));
5032 : }
5033 :
5034 : void *
5035 0 : spdk_bdev_get_module_ctx(struct spdk_bdev_desc *desc)
5036 : {
5037 0 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5038 0 : void *ctx = NULL;
5039 :
5040 0 : if (bdev->fn_table->get_module_ctx) {
5041 0 : ctx = bdev->fn_table->get_module_ctx(bdev->ctxt);
5042 : }
5043 :
5044 0 : return ctx;
5045 : }
5046 :
5047 : const char *
5048 0 : spdk_bdev_get_module_name(const struct spdk_bdev *bdev)
5049 : {
5050 0 : return bdev->module->name;
5051 : }
5052 :
5053 : const char *
5054 263 : spdk_bdev_get_name(const struct spdk_bdev *bdev)
5055 : {
5056 263 : return bdev->name;
5057 : }
5058 :
5059 : const char *
5060 0 : spdk_bdev_get_product_name(const struct spdk_bdev *bdev)
5061 : {
5062 0 : return bdev->product_name;
5063 : }
5064 :
5065 : const struct spdk_bdev_aliases_list *
5066 0 : spdk_bdev_get_aliases(const struct spdk_bdev *bdev)
5067 : {
5068 0 : return &bdev->aliases;
5069 : }
5070 :
5071 : uint32_t
5072 5 : spdk_bdev_get_block_size(const struct spdk_bdev *bdev)
5073 : {
5074 5 : return bdev->blocklen;
5075 : }
5076 :
5077 : uint32_t
5078 0 : spdk_bdev_get_write_unit_size(const struct spdk_bdev *bdev)
5079 : {
5080 0 : return bdev->write_unit_size;
5081 : }
5082 :
5083 : uint64_t
5084 0 : spdk_bdev_get_num_blocks(const struct spdk_bdev *bdev)
5085 : {
5086 0 : return bdev->blockcnt;
5087 : }
5088 :
5089 : const char *
5090 0 : spdk_bdev_get_qos_rpc_type(enum spdk_bdev_qos_rate_limit_type type)
5091 : {
5092 0 : return qos_rpc_type[type];
5093 : }
5094 :
5095 : void
5096 0 : spdk_bdev_get_qos_rate_limits(struct spdk_bdev *bdev, uint64_t *limits)
5097 : {
5098 : int i;
5099 :
5100 0 : memset(limits, 0, sizeof(*limits) * SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES);
5101 :
5102 0 : spdk_spin_lock(&bdev->internal.spinlock);
5103 0 : if (bdev->internal.qos) {
5104 0 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
5105 0 : if (bdev->internal.qos->rate_limits[i].limit !=
5106 : SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
5107 0 : limits[i] = bdev->internal.qos->rate_limits[i].limit;
5108 0 : if (bdev_qos_is_iops_rate_limit(i) == false) {
5109 : /* Change from Byte to Megabyte which is user visible. */
5110 0 : limits[i] = limits[i] / 1024 / 1024;
5111 : }
5112 : }
5113 : }
5114 : }
5115 0 : spdk_spin_unlock(&bdev->internal.spinlock);
5116 0 : }
5117 :
5118 : size_t
5119 280 : spdk_bdev_get_buf_align(const struct spdk_bdev *bdev)
5120 : {
5121 280 : return 1 << bdev->required_alignment;
5122 : }
5123 :
5124 : uint32_t
5125 0 : spdk_bdev_get_optimal_io_boundary(const struct spdk_bdev *bdev)
5126 : {
5127 0 : return bdev->optimal_io_boundary;
5128 : }
5129 :
5130 : bool
5131 0 : spdk_bdev_has_write_cache(const struct spdk_bdev *bdev)
5132 : {
5133 0 : return bdev->write_cache;
5134 : }
5135 :
5136 : const struct spdk_uuid *
5137 0 : spdk_bdev_get_uuid(const struct spdk_bdev *bdev)
5138 : {
5139 0 : return &bdev->uuid;
5140 : }
5141 :
5142 : uint16_t
5143 0 : spdk_bdev_get_acwu(const struct spdk_bdev *bdev)
5144 : {
5145 0 : return bdev->acwu;
5146 : }
5147 :
5148 : uint32_t
5149 29 : spdk_bdev_get_md_size(const struct spdk_bdev *bdev)
5150 : {
5151 29 : return bdev->md_len;
5152 : }
5153 :
5154 : bool
5155 135 : spdk_bdev_is_md_interleaved(const struct spdk_bdev *bdev)
5156 : {
5157 135 : return (bdev->md_len != 0) && bdev->md_interleave;
5158 : }
5159 :
5160 : bool
5161 117 : spdk_bdev_is_md_separate(const struct spdk_bdev *bdev)
5162 : {
5163 117 : return (bdev->md_len != 0) && !bdev->md_interleave;
5164 : }
5165 :
5166 : bool
5167 0 : spdk_bdev_is_zoned(const struct spdk_bdev *bdev)
5168 : {
5169 0 : return bdev->zoned;
5170 : }
5171 :
5172 : uint32_t
5173 126 : spdk_bdev_get_data_block_size(const struct spdk_bdev *bdev)
5174 : {
5175 126 : if (spdk_bdev_is_md_interleaved(bdev)) {
5176 0 : return bdev->blocklen - bdev->md_len;
5177 : } else {
5178 126 : return bdev->blocklen;
5179 : }
5180 : }
5181 :
5182 : uint32_t
5183 0 : spdk_bdev_get_physical_block_size(const struct spdk_bdev *bdev)
5184 : {
5185 0 : return bdev->phys_blocklen;
5186 : }
5187 :
5188 : static uint32_t
5189 9 : _bdev_get_block_size_with_md(const struct spdk_bdev *bdev)
5190 : {
5191 9 : if (!spdk_bdev_is_md_interleaved(bdev)) {
5192 6 : return bdev->blocklen + bdev->md_len;
5193 : } else {
5194 3 : return bdev->blocklen;
5195 : }
5196 : }
5197 :
5198 : /* We have to use the typedef in the function declaration to appease astyle. */
5199 : typedef enum spdk_dif_type spdk_dif_type_t;
5200 : typedef enum spdk_dif_pi_format spdk_dif_pi_format_t;
5201 :
5202 : spdk_dif_type_t
5203 0 : spdk_bdev_get_dif_type(const struct spdk_bdev *bdev)
5204 : {
5205 0 : if (bdev->md_len != 0) {
5206 0 : return bdev->dif_type;
5207 : } else {
5208 0 : return SPDK_DIF_DISABLE;
5209 : }
5210 : }
5211 :
5212 : spdk_dif_pi_format_t
5213 0 : spdk_bdev_get_dif_pi_format(const struct spdk_bdev *bdev)
5214 : {
5215 0 : return bdev->dif_pi_format;
5216 : }
5217 :
5218 : bool
5219 0 : spdk_bdev_is_dif_head_of_md(const struct spdk_bdev *bdev)
5220 : {
5221 0 : if (spdk_bdev_get_dif_type(bdev) != SPDK_DIF_DISABLE) {
5222 0 : return bdev->dif_is_head_of_md;
5223 : } else {
5224 0 : return false;
5225 : }
5226 : }
5227 :
5228 : bool
5229 0 : spdk_bdev_is_dif_check_enabled(const struct spdk_bdev *bdev,
5230 : enum spdk_dif_check_type check_type)
5231 : {
5232 0 : if (spdk_bdev_get_dif_type(bdev) == SPDK_DIF_DISABLE) {
5233 0 : return false;
5234 : }
5235 :
5236 0 : switch (check_type) {
5237 0 : case SPDK_DIF_CHECK_TYPE_REFTAG:
5238 0 : return (bdev->dif_check_flags & SPDK_DIF_FLAGS_REFTAG_CHECK) != 0;
5239 0 : case SPDK_DIF_CHECK_TYPE_APPTAG:
5240 0 : return (bdev->dif_check_flags & SPDK_DIF_FLAGS_APPTAG_CHECK) != 0;
5241 0 : case SPDK_DIF_CHECK_TYPE_GUARD:
5242 0 : return (bdev->dif_check_flags & SPDK_DIF_FLAGS_GUARD_CHECK) != 0;
5243 0 : default:
5244 0 : return false;
5245 : }
5246 : }
5247 :
5248 : static uint32_t
5249 3 : bdev_get_max_write(const struct spdk_bdev *bdev, uint64_t num_bytes)
5250 : {
5251 : uint64_t aligned_length, max_write_blocks;
5252 :
5253 3 : aligned_length = num_bytes - (spdk_bdev_get_buf_align(bdev) - 1);
5254 3 : max_write_blocks = aligned_length / _bdev_get_block_size_with_md(bdev);
5255 3 : max_write_blocks -= max_write_blocks % bdev->write_unit_size;
5256 :
5257 3 : return max_write_blocks;
5258 : }
5259 :
5260 : uint32_t
5261 1 : spdk_bdev_get_max_copy(const struct spdk_bdev *bdev)
5262 : {
5263 1 : return bdev->max_copy;
5264 : }
5265 :
5266 : uint64_t
5267 0 : spdk_bdev_get_qd(const struct spdk_bdev *bdev)
5268 : {
5269 0 : return bdev->internal.measured_queue_depth;
5270 : }
5271 :
5272 : uint64_t
5273 0 : spdk_bdev_get_qd_sampling_period(const struct spdk_bdev *bdev)
5274 : {
5275 0 : return bdev->internal.period;
5276 : }
5277 :
5278 : uint64_t
5279 0 : spdk_bdev_get_weighted_io_time(const struct spdk_bdev *bdev)
5280 : {
5281 0 : return bdev->internal.weighted_io_time;
5282 : }
5283 :
5284 : uint64_t
5285 0 : spdk_bdev_get_io_time(const struct spdk_bdev *bdev)
5286 : {
5287 0 : return bdev->internal.io_time;
5288 : }
5289 :
5290 0 : union spdk_bdev_nvme_ctratt spdk_bdev_get_nvme_ctratt(struct spdk_bdev *bdev)
5291 : {
5292 0 : return bdev->ctratt;
5293 : }
5294 :
5295 : uint32_t
5296 0 : spdk_bdev_get_nvme_nsid(struct spdk_bdev *bdev)
5297 : {
5298 0 : return bdev->nsid;
5299 : }
5300 :
5301 : uint32_t
5302 0 : spdk_bdev_desc_get_block_size(struct spdk_bdev_desc *desc)
5303 : {
5304 0 : struct spdk_bdev *bdev = desc->bdev;
5305 :
5306 0 : return desc->opts.hide_metadata ? bdev->blocklen - bdev->md_len : bdev->blocklen;
5307 : }
5308 :
5309 : uint32_t
5310 0 : spdk_bdev_desc_get_md_size(struct spdk_bdev_desc *desc)
5311 : {
5312 0 : struct spdk_bdev *bdev = desc->bdev;
5313 :
5314 0 : return desc->opts.hide_metadata ? 0 : bdev->md_len;
5315 : }
5316 :
5317 : bool
5318 0 : spdk_bdev_desc_is_md_interleaved(struct spdk_bdev_desc *desc)
5319 : {
5320 0 : struct spdk_bdev *bdev = desc->bdev;
5321 :
5322 0 : return desc->opts.hide_metadata ? false : spdk_bdev_is_md_interleaved(bdev);
5323 : }
5324 :
5325 : bool
5326 0 : spdk_bdev_desc_is_md_separate(struct spdk_bdev_desc *desc)
5327 : {
5328 0 : struct spdk_bdev *bdev = desc->bdev;
5329 :
5330 0 : return desc->opts.hide_metadata ? false : spdk_bdev_is_md_separate(bdev);
5331 : }
5332 :
5333 : spdk_dif_type_t
5334 0 : spdk_bdev_desc_get_dif_type(struct spdk_bdev_desc *desc)
5335 : {
5336 0 : struct spdk_bdev *bdev = desc->bdev;
5337 :
5338 0 : return desc->opts.hide_metadata ? SPDK_DIF_DISABLE : spdk_bdev_get_dif_type(bdev);
5339 : }
5340 :
5341 : spdk_dif_pi_format_t
5342 0 : spdk_bdev_desc_get_dif_pi_format(struct spdk_bdev_desc *desc)
5343 : {
5344 0 : struct spdk_bdev *bdev = desc->bdev;
5345 :
5346 0 : return desc->opts.hide_metadata ? SPDK_DIF_PI_FORMAT_16 : spdk_bdev_get_dif_pi_format(bdev);
5347 : }
5348 :
5349 : bool
5350 0 : spdk_bdev_desc_is_dif_head_of_md(struct spdk_bdev_desc *desc)
5351 : {
5352 0 : struct spdk_bdev *bdev = desc->bdev;
5353 :
5354 0 : return desc->opts.hide_metadata ? false : spdk_bdev_is_dif_head_of_md(bdev);
5355 : }
5356 :
5357 : bool
5358 0 : spdk_bdev_desc_is_dif_check_enabled(struct spdk_bdev_desc *desc,
5359 : enum spdk_dif_check_type check_type)
5360 : {
5361 0 : struct spdk_bdev *bdev = desc->bdev;
5362 :
5363 0 : return desc->opts.hide_metadata ? false : spdk_bdev_is_dif_check_enabled(bdev, check_type);
5364 : }
5365 :
5366 : static void bdev_update_qd_sampling_period(void *ctx);
5367 :
5368 : static void
5369 1 : _calculate_measured_qd_cpl(struct spdk_bdev *bdev, void *_ctx, int status)
5370 : {
5371 1 : bdev->internal.measured_queue_depth = bdev->internal.temporary_queue_depth;
5372 :
5373 1 : if (bdev->internal.measured_queue_depth) {
5374 0 : bdev->internal.io_time += bdev->internal.period;
5375 0 : bdev->internal.weighted_io_time += bdev->internal.period * bdev->internal.measured_queue_depth;
5376 : }
5377 :
5378 1 : bdev->internal.qd_poll_in_progress = false;
5379 :
5380 1 : bdev_update_qd_sampling_period(bdev);
5381 1 : }
5382 :
5383 : static void
5384 1 : _calculate_measured_qd(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
5385 : struct spdk_io_channel *io_ch, void *_ctx)
5386 : {
5387 1 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(io_ch);
5388 :
5389 1 : bdev->internal.temporary_queue_depth += ch->io_outstanding;
5390 1 : spdk_bdev_for_each_channel_continue(i, 0);
5391 1 : }
5392 :
5393 : static int
5394 1 : bdev_calculate_measured_queue_depth(void *ctx)
5395 : {
5396 1 : struct spdk_bdev *bdev = ctx;
5397 :
5398 1 : bdev->internal.qd_poll_in_progress = true;
5399 1 : bdev->internal.temporary_queue_depth = 0;
5400 1 : spdk_bdev_for_each_channel(bdev, _calculate_measured_qd, bdev, _calculate_measured_qd_cpl);
5401 1 : return SPDK_POLLER_BUSY;
5402 : }
5403 :
5404 : static void
5405 5 : bdev_update_qd_sampling_period(void *ctx)
5406 : {
5407 5 : struct spdk_bdev *bdev = ctx;
5408 :
5409 5 : if (bdev->internal.period == bdev->internal.new_period) {
5410 0 : return;
5411 : }
5412 :
5413 5 : if (bdev->internal.qd_poll_in_progress) {
5414 1 : return;
5415 : }
5416 :
5417 4 : bdev->internal.period = bdev->internal.new_period;
5418 :
5419 4 : spdk_poller_unregister(&bdev->internal.qd_poller);
5420 4 : if (bdev->internal.period != 0) {
5421 2 : bdev->internal.qd_poller = SPDK_POLLER_REGISTER(bdev_calculate_measured_queue_depth,
5422 : bdev, bdev->internal.period);
5423 : } else {
5424 2 : spdk_bdev_close(bdev->internal.qd_desc);
5425 2 : bdev->internal.qd_desc = NULL;
5426 : }
5427 : }
5428 :
5429 : static void
5430 0 : _tmp_bdev_event_cb(enum spdk_bdev_event_type type, struct spdk_bdev *bdev, void *ctx)
5431 : {
5432 0 : SPDK_NOTICELOG("Unexpected event type: %d\n", type);
5433 0 : }
5434 :
5435 : void
5436 135 : spdk_bdev_set_qd_sampling_period(struct spdk_bdev *bdev, uint64_t period)
5437 : {
5438 : int rc;
5439 :
5440 135 : if (bdev->internal.new_period == period) {
5441 129 : return;
5442 : }
5443 :
5444 6 : bdev->internal.new_period = period;
5445 :
5446 6 : if (bdev->internal.qd_desc != NULL) {
5447 4 : assert(bdev->internal.period != 0);
5448 :
5449 4 : spdk_thread_send_msg(bdev->internal.qd_desc->thread,
5450 : bdev_update_qd_sampling_period, bdev);
5451 4 : return;
5452 : }
5453 :
5454 2 : assert(bdev->internal.period == 0);
5455 :
5456 2 : rc = spdk_bdev_open_ext(spdk_bdev_get_name(bdev), false, _tmp_bdev_event_cb,
5457 : NULL, &bdev->internal.qd_desc);
5458 2 : if (rc != 0) {
5459 0 : return;
5460 : }
5461 :
5462 2 : bdev->internal.period = period;
5463 2 : bdev->internal.qd_poller = SPDK_POLLER_REGISTER(bdev_calculate_measured_queue_depth,
5464 : bdev, period);
5465 : }
5466 :
5467 : struct bdev_get_current_qd_ctx {
5468 : uint64_t current_qd;
5469 : spdk_bdev_get_current_qd_cb cb_fn;
5470 : void *cb_arg;
5471 : };
5472 :
5473 : static void
5474 0 : bdev_get_current_qd_done(struct spdk_bdev *bdev, void *_ctx, int status)
5475 : {
5476 0 : struct bdev_get_current_qd_ctx *ctx = _ctx;
5477 :
5478 0 : ctx->cb_fn(bdev, ctx->current_qd, ctx->cb_arg, 0);
5479 :
5480 0 : free(ctx);
5481 0 : }
5482 :
5483 : static void
5484 0 : bdev_get_current_qd(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
5485 : struct spdk_io_channel *io_ch, void *_ctx)
5486 : {
5487 0 : struct bdev_get_current_qd_ctx *ctx = _ctx;
5488 0 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
5489 :
5490 0 : ctx->current_qd += bdev_ch->io_outstanding;
5491 :
5492 0 : spdk_bdev_for_each_channel_continue(i, 0);
5493 0 : }
5494 :
5495 : void
5496 0 : spdk_bdev_get_current_qd(struct spdk_bdev *bdev, spdk_bdev_get_current_qd_cb cb_fn,
5497 : void *cb_arg)
5498 : {
5499 : struct bdev_get_current_qd_ctx *ctx;
5500 :
5501 0 : assert(cb_fn != NULL);
5502 :
5503 0 : ctx = calloc(1, sizeof(*ctx));
5504 0 : if (ctx == NULL) {
5505 0 : cb_fn(bdev, 0, cb_arg, -ENOMEM);
5506 0 : return;
5507 : }
5508 :
5509 0 : ctx->cb_fn = cb_fn;
5510 0 : ctx->cb_arg = cb_arg;
5511 :
5512 0 : spdk_bdev_for_each_channel(bdev, bdev_get_current_qd, ctx, bdev_get_current_qd_done);
5513 : }
5514 :
5515 : static void
5516 25 : _event_notify(struct spdk_bdev_desc *desc, enum spdk_bdev_event_type type)
5517 : {
5518 25 : assert(desc->thread == spdk_get_thread());
5519 :
5520 25 : spdk_spin_lock(&desc->spinlock);
5521 25 : desc->refs--;
5522 25 : if (!desc->closed) {
5523 14 : spdk_spin_unlock(&desc->spinlock);
5524 14 : desc->callback.event_fn(type,
5525 : desc->bdev,
5526 : desc->callback.ctx);
5527 14 : return;
5528 11 : } else if (desc->refs == 0) {
5529 : /* This descriptor was closed after this event_notify message was sent.
5530 : * spdk_bdev_close() could not free the descriptor since this message was
5531 : * in flight, so we free it now using bdev_desc_free().
5532 : */
5533 10 : spdk_spin_unlock(&desc->spinlock);
5534 10 : bdev_desc_free(desc);
5535 10 : return;
5536 : }
5537 1 : spdk_spin_unlock(&desc->spinlock);
5538 : }
5539 :
5540 : static void
5541 25 : event_notify(struct spdk_bdev_desc *desc, spdk_msg_fn event_notify_fn)
5542 : {
5543 25 : spdk_spin_lock(&desc->spinlock);
5544 25 : desc->refs++;
5545 25 : spdk_thread_send_msg(desc->thread, event_notify_fn, desc);
5546 25 : spdk_spin_unlock(&desc->spinlock);
5547 25 : }
5548 :
5549 : static void
5550 6 : _resize_notify(void *ctx)
5551 : {
5552 6 : struct spdk_bdev_desc *desc = ctx;
5553 :
5554 6 : _event_notify(desc, SPDK_BDEV_EVENT_RESIZE);
5555 6 : }
5556 :
5557 : int
5558 11 : spdk_bdev_notify_blockcnt_change(struct spdk_bdev *bdev, uint64_t size)
5559 : {
5560 : struct spdk_bdev_desc *desc;
5561 : int ret;
5562 :
5563 11 : if (size == bdev->blockcnt) {
5564 0 : return 0;
5565 : }
5566 :
5567 11 : spdk_spin_lock(&bdev->internal.spinlock);
5568 :
5569 : /* bdev has open descriptors */
5570 11 : if (!TAILQ_EMPTY(&bdev->internal.open_descs) &&
5571 7 : bdev->blockcnt > size) {
5572 1 : ret = -EBUSY;
5573 : } else {
5574 10 : bdev->blockcnt = size;
5575 16 : TAILQ_FOREACH(desc, &bdev->internal.open_descs, link) {
5576 6 : event_notify(desc, _resize_notify);
5577 : }
5578 10 : ret = 0;
5579 : }
5580 :
5581 11 : spdk_spin_unlock(&bdev->internal.spinlock);
5582 :
5583 11 : return ret;
5584 : }
5585 :
5586 : /*
5587 : * Convert I/O offset and length from bytes to blocks.
5588 : *
5589 : * Returns zero on success or non-zero if the byte parameters aren't divisible by the block size.
5590 : */
5591 : static uint64_t
5592 20 : bdev_bytes_to_blocks(struct spdk_bdev_desc *desc, uint64_t offset_bytes,
5593 : uint64_t *offset_blocks, uint64_t num_bytes, uint64_t *num_blocks)
5594 : {
5595 20 : uint32_t block_size = bdev_desc_get_block_size(desc);
5596 : uint8_t shift_cnt;
5597 :
5598 : /* Avoid expensive div operations if possible. These spdk_u32 functions are very cheap. */
5599 20 : if (spdk_likely(spdk_u32_is_pow2(block_size))) {
5600 17 : shift_cnt = spdk_u32log2(block_size);
5601 17 : *offset_blocks = offset_bytes >> shift_cnt;
5602 17 : *num_blocks = num_bytes >> shift_cnt;
5603 17 : return (offset_bytes - (*offset_blocks << shift_cnt)) |
5604 17 : (num_bytes - (*num_blocks << shift_cnt));
5605 : } else {
5606 3 : *offset_blocks = offset_bytes / block_size;
5607 3 : *num_blocks = num_bytes / block_size;
5608 3 : return (offset_bytes % block_size) | (num_bytes % block_size);
5609 : }
5610 : }
5611 :
5612 : static bool
5613 689 : bdev_io_valid_blocks(struct spdk_bdev *bdev, uint64_t offset_blocks, uint64_t num_blocks)
5614 : {
5615 : /* Return failure if offset_blocks + num_blocks is less than offset_blocks; indicates there
5616 : * has been an overflow and hence the offset has been wrapped around */
5617 689 : if (offset_blocks + num_blocks < offset_blocks) {
5618 1 : return false;
5619 : }
5620 :
5621 : /* Return failure if offset_blocks + num_blocks exceeds the size of the bdev */
5622 688 : if (offset_blocks + num_blocks > bdev->blockcnt) {
5623 2 : return false;
5624 : }
5625 :
5626 686 : return true;
5627 : }
5628 :
5629 : static void
5630 2 : bdev_seek_complete_cb(void *ctx)
5631 : {
5632 2 : struct spdk_bdev_io *bdev_io = ctx;
5633 :
5634 2 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
5635 2 : bdev_io->internal.cb(bdev_io, true, bdev_io->internal.caller_ctx);
5636 2 : }
5637 :
5638 : static int
5639 4 : bdev_seek(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5640 : uint64_t offset_blocks, enum spdk_bdev_io_type io_type,
5641 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5642 : {
5643 4 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5644 : struct spdk_bdev_io *bdev_io;
5645 4 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5646 :
5647 4 : assert(io_type == SPDK_BDEV_IO_TYPE_SEEK_DATA || io_type == SPDK_BDEV_IO_TYPE_SEEK_HOLE);
5648 :
5649 : /* Check if offset_blocks is valid looking at the validity of one block */
5650 4 : if (!bdev_io_valid_blocks(bdev, offset_blocks, 1)) {
5651 0 : return -EINVAL;
5652 : }
5653 :
5654 4 : bdev_io = bdev_channel_get_io(channel);
5655 4 : if (!bdev_io) {
5656 0 : return -ENOMEM;
5657 : }
5658 :
5659 4 : bdev_io->internal.ch = channel;
5660 4 : bdev_io->internal.desc = desc;
5661 4 : bdev_io->type = io_type;
5662 4 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5663 4 : bdev_io->u.bdev.memory_domain = NULL;
5664 4 : bdev_io->u.bdev.memory_domain_ctx = NULL;
5665 4 : bdev_io->u.bdev.accel_sequence = NULL;
5666 4 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5667 :
5668 4 : if (!spdk_bdev_io_type_supported(bdev, io_type)) {
5669 : /* In case bdev doesn't support seek to next data/hole offset,
5670 : * it is assumed that only data and no holes are present */
5671 2 : if (io_type == SPDK_BDEV_IO_TYPE_SEEK_DATA) {
5672 1 : bdev_io->u.bdev.seek.offset = offset_blocks;
5673 : } else {
5674 1 : bdev_io->u.bdev.seek.offset = UINT64_MAX;
5675 : }
5676 :
5677 2 : spdk_thread_send_msg(spdk_get_thread(), bdev_seek_complete_cb, bdev_io);
5678 2 : return 0;
5679 : }
5680 :
5681 2 : bdev_io_submit(bdev_io);
5682 2 : return 0;
5683 : }
5684 :
5685 : int
5686 2 : spdk_bdev_seek_data(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5687 : uint64_t offset_blocks,
5688 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5689 : {
5690 2 : return bdev_seek(desc, ch, offset_blocks, SPDK_BDEV_IO_TYPE_SEEK_DATA, cb, cb_arg);
5691 : }
5692 :
5693 : int
5694 2 : spdk_bdev_seek_hole(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5695 : uint64_t offset_blocks,
5696 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5697 : {
5698 2 : return bdev_seek(desc, ch, offset_blocks, SPDK_BDEV_IO_TYPE_SEEK_HOLE, cb, cb_arg);
5699 : }
5700 :
5701 : uint64_t
5702 4 : spdk_bdev_io_get_seek_offset(const struct spdk_bdev_io *bdev_io)
5703 : {
5704 4 : return bdev_io->u.bdev.seek.offset;
5705 : }
5706 :
5707 : static int
5708 204 : bdev_read_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch, void *buf,
5709 : void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
5710 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5711 : {
5712 204 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5713 : struct spdk_bdev_io *bdev_io;
5714 204 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5715 :
5716 204 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
5717 0 : return -EINVAL;
5718 : }
5719 :
5720 204 : bdev_io = bdev_channel_get_io(channel);
5721 204 : if (!bdev_io) {
5722 1 : return -ENOMEM;
5723 : }
5724 :
5725 203 : bdev_io->internal.ch = channel;
5726 203 : bdev_io->internal.desc = desc;
5727 203 : bdev_io->type = SPDK_BDEV_IO_TYPE_READ;
5728 203 : bdev_io->u.bdev.iovs = &bdev_io->iov;
5729 203 : bdev_io->u.bdev.iovs[0].iov_base = buf;
5730 203 : bdev_io->u.bdev.iovs[0].iov_len = num_blocks * bdev_desc_get_block_size(desc);
5731 203 : bdev_io->u.bdev.iovcnt = 1;
5732 203 : bdev_io->u.bdev.md_buf = md_buf;
5733 203 : bdev_io->u.bdev.num_blocks = num_blocks;
5734 203 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5735 203 : bdev_io->u.bdev.memory_domain = NULL;
5736 203 : bdev_io->u.bdev.memory_domain_ctx = NULL;
5737 203 : bdev_io->u.bdev.accel_sequence = NULL;
5738 203 : bdev_io->u.bdev.dif_check_flags = bdev->dif_check_flags;
5739 203 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5740 :
5741 203 : bdev_io_submit(bdev_io);
5742 203 : return 0;
5743 : }
5744 :
5745 : int
5746 3 : spdk_bdev_read(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5747 : void *buf, uint64_t offset, uint64_t nbytes,
5748 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5749 : {
5750 3 : uint64_t offset_blocks, num_blocks;
5751 :
5752 3 : if (bdev_bytes_to_blocks(desc, offset, &offset_blocks, nbytes, &num_blocks) != 0) {
5753 0 : return -EINVAL;
5754 : }
5755 :
5756 3 : return spdk_bdev_read_blocks(desc, ch, buf, offset_blocks, num_blocks, cb, cb_arg);
5757 : }
5758 :
5759 : int
5760 200 : spdk_bdev_read_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5761 : void *buf, uint64_t offset_blocks, uint64_t num_blocks,
5762 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5763 : {
5764 200 : return bdev_read_blocks_with_md(desc, ch, buf, NULL, offset_blocks, num_blocks, cb, cb_arg);
5765 : }
5766 :
5767 : int
5768 4 : spdk_bdev_read_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5769 : void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
5770 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5771 : {
5772 4 : struct iovec iov = {
5773 : .iov_base = buf,
5774 : };
5775 :
5776 4 : if (md_buf && !spdk_bdev_is_md_separate(spdk_bdev_desc_get_bdev(desc))) {
5777 0 : return -EINVAL;
5778 : }
5779 :
5780 4 : if ((md_buf || desc->opts.hide_metadata) && !_is_buf_allocated(&iov)) {
5781 0 : return -EINVAL;
5782 : }
5783 :
5784 4 : return bdev_read_blocks_with_md(desc, ch, buf, md_buf, offset_blocks, num_blocks,
5785 : cb, cb_arg);
5786 : }
5787 :
5788 : int
5789 5 : spdk_bdev_readv(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5790 : struct iovec *iov, int iovcnt,
5791 : uint64_t offset, uint64_t nbytes,
5792 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5793 : {
5794 5 : uint64_t offset_blocks, num_blocks;
5795 :
5796 5 : if (bdev_bytes_to_blocks(desc, offset, &offset_blocks, nbytes, &num_blocks) != 0) {
5797 0 : return -EINVAL;
5798 : }
5799 :
5800 5 : return spdk_bdev_readv_blocks(desc, ch, iov, iovcnt, offset_blocks, num_blocks, cb, cb_arg);
5801 : }
5802 :
5803 : static int
5804 226 : bdev_readv_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5805 : struct iovec *iov, int iovcnt, void *md_buf, uint64_t offset_blocks,
5806 : uint64_t num_blocks, struct spdk_memory_domain *domain, void *domain_ctx,
5807 : struct spdk_accel_sequence *seq, uint32_t dif_check_flags,
5808 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5809 : {
5810 226 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5811 : struct spdk_bdev_io *bdev_io;
5812 226 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5813 :
5814 226 : if (spdk_unlikely(!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks))) {
5815 0 : return -EINVAL;
5816 : }
5817 :
5818 226 : bdev_io = bdev_channel_get_io(channel);
5819 226 : if (spdk_unlikely(!bdev_io)) {
5820 2 : return -ENOMEM;
5821 : }
5822 :
5823 224 : bdev_io->internal.ch = channel;
5824 224 : bdev_io->internal.desc = desc;
5825 224 : bdev_io->type = SPDK_BDEV_IO_TYPE_READ;
5826 224 : bdev_io->u.bdev.iovs = iov;
5827 224 : bdev_io->u.bdev.iovcnt = iovcnt;
5828 224 : bdev_io->u.bdev.md_buf = md_buf;
5829 224 : bdev_io->u.bdev.num_blocks = num_blocks;
5830 224 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5831 224 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5832 :
5833 224 : if (seq != NULL) {
5834 0 : bdev_io->internal.f.has_accel_sequence = true;
5835 0 : bdev_io->internal.accel_sequence = seq;
5836 : }
5837 :
5838 224 : if (domain != NULL) {
5839 2 : bdev_io->internal.f.has_memory_domain = true;
5840 2 : bdev_io->internal.memory_domain = domain;
5841 2 : bdev_io->internal.memory_domain_ctx = domain_ctx;
5842 : }
5843 :
5844 224 : bdev_io->u.bdev.memory_domain = domain;
5845 224 : bdev_io->u.bdev.memory_domain_ctx = domain_ctx;
5846 224 : bdev_io->u.bdev.accel_sequence = seq;
5847 224 : bdev_io->u.bdev.dif_check_flags = dif_check_flags;
5848 :
5849 224 : _bdev_io_submit_ext(desc, bdev_io);
5850 :
5851 224 : return 0;
5852 : }
5853 :
5854 : int
5855 21 : spdk_bdev_readv_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5856 : struct iovec *iov, int iovcnt,
5857 : uint64_t offset_blocks, uint64_t num_blocks,
5858 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5859 : {
5860 21 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5861 :
5862 21 : return bdev_readv_blocks_with_md(desc, ch, iov, iovcnt, NULL, offset_blocks,
5863 : num_blocks, NULL, NULL, NULL, bdev->dif_check_flags, cb, cb_arg);
5864 : }
5865 :
5866 : int
5867 4 : spdk_bdev_readv_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5868 : struct iovec *iov, int iovcnt, void *md_buf,
5869 : uint64_t offset_blocks, uint64_t num_blocks,
5870 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5871 : {
5872 4 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5873 :
5874 4 : if (md_buf && !spdk_bdev_is_md_separate(bdev)) {
5875 0 : return -EINVAL;
5876 : }
5877 :
5878 4 : if (md_buf && !_is_buf_allocated(iov)) {
5879 0 : return -EINVAL;
5880 : }
5881 :
5882 4 : return bdev_readv_blocks_with_md(desc, ch, iov, iovcnt, md_buf, offset_blocks,
5883 : num_blocks, NULL, NULL, NULL, bdev->dif_check_flags, cb, cb_arg);
5884 : }
5885 :
5886 : static inline bool
5887 14 : _bdev_io_check_opts(struct spdk_bdev_ext_io_opts *opts, struct iovec *iov)
5888 : {
5889 : /*
5890 : * We check if opts size is at least of size when we first introduced
5891 : * spdk_bdev_ext_io_opts (ac6f2bdd8d) since access to those members
5892 : * are not checked internal.
5893 : */
5894 14 : return opts->size >= offsetof(struct spdk_bdev_ext_io_opts, metadata) +
5895 10 : sizeof(opts->metadata) &&
5896 22 : opts->size <= sizeof(*opts) &&
5897 : /* When memory domain is used, the user must provide data buffers */
5898 8 : (!opts->memory_domain || (iov && iov[0].iov_base));
5899 : }
5900 :
5901 : int
5902 8 : spdk_bdev_readv_blocks_ext(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5903 : struct iovec *iov, int iovcnt,
5904 : uint64_t offset_blocks, uint64_t num_blocks,
5905 : spdk_bdev_io_completion_cb cb, void *cb_arg,
5906 : struct spdk_bdev_ext_io_opts *opts)
5907 : {
5908 8 : struct spdk_memory_domain *domain = NULL;
5909 8 : struct spdk_accel_sequence *seq = NULL;
5910 8 : void *domain_ctx = NULL, *md = NULL;
5911 8 : uint32_t dif_check_flags = 0;
5912 : uint32_t nvme_cdw12_raw;
5913 8 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5914 :
5915 8 : if (opts) {
5916 7 : if (spdk_unlikely(!_bdev_io_check_opts(opts, iov))) {
5917 3 : return -EINVAL;
5918 : }
5919 :
5920 4 : md = opts->metadata;
5921 4 : domain = bdev_get_ext_io_opt(opts, memory_domain, NULL);
5922 4 : domain_ctx = bdev_get_ext_io_opt(opts, memory_domain_ctx, NULL);
5923 4 : seq = bdev_get_ext_io_opt(opts, accel_sequence, NULL);
5924 4 : nvme_cdw12_raw = bdev_get_ext_io_opt(opts, nvme_cdw12.raw, 0);
5925 4 : if (md) {
5926 4 : if (spdk_unlikely(!spdk_bdev_is_md_separate(bdev))) {
5927 0 : return -EINVAL;
5928 : }
5929 :
5930 4 : if (spdk_unlikely(!_is_buf_allocated(iov))) {
5931 0 : return -EINVAL;
5932 : }
5933 :
5934 4 : if (spdk_unlikely(seq != NULL)) {
5935 0 : return -EINVAL;
5936 : }
5937 :
5938 4 : if (nvme_cdw12_raw & SPDK_DIF_FLAGS_NVME_PRACT) {
5939 0 : SPDK_ERRLOG("Separate metadata with NVMe PRACT is not supported.\n");
5940 0 : return -ENOTSUP;
5941 : }
5942 : }
5943 :
5944 4 : if (nvme_cdw12_raw & SPDK_DIF_FLAGS_NVME_PRACT) {
5945 0 : dif_check_flags |= SPDK_DIF_FLAGS_NVME_PRACT;
5946 : }
5947 : }
5948 :
5949 5 : dif_check_flags |= bdev->dif_check_flags &
5950 5 : ~(bdev_get_ext_io_opt(opts, dif_check_flags_exclude_mask, 0));
5951 :
5952 5 : return bdev_readv_blocks_with_md(desc, ch, iov, iovcnt, md, offset_blocks,
5953 : num_blocks, domain, domain_ctx, seq, dif_check_flags, cb, cb_arg);
5954 : }
5955 :
5956 : static int
5957 36 : bdev_write_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5958 : void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
5959 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5960 : {
5961 36 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5962 : struct spdk_bdev_io *bdev_io;
5963 36 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5964 :
5965 36 : if (!desc->write) {
5966 0 : return -EBADF;
5967 : }
5968 :
5969 36 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
5970 0 : return -EINVAL;
5971 : }
5972 :
5973 36 : bdev_io = bdev_channel_get_io(channel);
5974 36 : if (!bdev_io) {
5975 0 : return -ENOMEM;
5976 : }
5977 :
5978 36 : bdev_io->internal.ch = channel;
5979 36 : bdev_io->internal.desc = desc;
5980 36 : bdev_io->type = SPDK_BDEV_IO_TYPE_WRITE;
5981 36 : bdev_io->u.bdev.iovs = &bdev_io->iov;
5982 36 : bdev_io->u.bdev.iovs[0].iov_base = buf;
5983 36 : bdev_io->u.bdev.iovs[0].iov_len = num_blocks * bdev_desc_get_block_size(desc);
5984 36 : bdev_io->u.bdev.iovcnt = 1;
5985 36 : bdev_io->u.bdev.md_buf = md_buf;
5986 36 : bdev_io->u.bdev.num_blocks = num_blocks;
5987 36 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5988 36 : bdev_io->u.bdev.memory_domain = NULL;
5989 36 : bdev_io->u.bdev.memory_domain_ctx = NULL;
5990 36 : bdev_io->u.bdev.accel_sequence = NULL;
5991 36 : bdev_io->u.bdev.dif_check_flags = bdev->dif_check_flags;
5992 36 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5993 :
5994 36 : bdev_io_submit(bdev_io);
5995 36 : return 0;
5996 : }
5997 :
5998 : int
5999 3 : spdk_bdev_write(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6000 : void *buf, uint64_t offset, uint64_t nbytes,
6001 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6002 : {
6003 3 : uint64_t offset_blocks, num_blocks;
6004 :
6005 3 : if (bdev_bytes_to_blocks(desc, offset, &offset_blocks, nbytes, &num_blocks) != 0) {
6006 0 : return -EINVAL;
6007 : }
6008 :
6009 3 : return spdk_bdev_write_blocks(desc, ch, buf, offset_blocks, num_blocks, cb, cb_arg);
6010 : }
6011 :
6012 : int
6013 27 : spdk_bdev_write_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6014 : void *buf, uint64_t offset_blocks, uint64_t num_blocks,
6015 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6016 : {
6017 27 : return bdev_write_blocks_with_md(desc, ch, buf, NULL, offset_blocks, num_blocks,
6018 : cb, cb_arg);
6019 : }
6020 :
6021 : int
6022 3 : spdk_bdev_write_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6023 : void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
6024 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6025 : {
6026 3 : struct iovec iov = {
6027 : .iov_base = buf,
6028 : };
6029 :
6030 3 : if (md_buf && !spdk_bdev_is_md_separate(spdk_bdev_desc_get_bdev(desc))) {
6031 0 : return -EINVAL;
6032 : }
6033 :
6034 3 : if (md_buf && !_is_buf_allocated(&iov)) {
6035 0 : return -EINVAL;
6036 : }
6037 :
6038 3 : return bdev_write_blocks_with_md(desc, ch, buf, md_buf, offset_blocks, num_blocks,
6039 : cb, cb_arg);
6040 : }
6041 :
6042 : static int
6043 70 : bdev_writev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6044 : struct iovec *iov, int iovcnt, void *md_buf,
6045 : uint64_t offset_blocks, uint64_t num_blocks,
6046 : struct spdk_memory_domain *domain, void *domain_ctx,
6047 : struct spdk_accel_sequence *seq, uint32_t dif_check_flags,
6048 : uint32_t nvme_cdw12_raw, uint32_t nvme_cdw13_raw,
6049 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6050 : {
6051 70 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6052 : struct spdk_bdev_io *bdev_io;
6053 70 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6054 :
6055 70 : if (spdk_unlikely(!desc->write)) {
6056 0 : return -EBADF;
6057 : }
6058 :
6059 70 : if (spdk_unlikely(!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks))) {
6060 0 : return -EINVAL;
6061 : }
6062 :
6063 70 : bdev_io = bdev_channel_get_io(channel);
6064 70 : if (spdk_unlikely(!bdev_io)) {
6065 2 : return -ENOMEM;
6066 : }
6067 :
6068 68 : bdev_io->internal.ch = channel;
6069 68 : bdev_io->internal.desc = desc;
6070 68 : bdev_io->type = SPDK_BDEV_IO_TYPE_WRITE;
6071 68 : bdev_io->u.bdev.iovs = iov;
6072 68 : bdev_io->u.bdev.iovcnt = iovcnt;
6073 68 : bdev_io->u.bdev.md_buf = md_buf;
6074 68 : bdev_io->u.bdev.num_blocks = num_blocks;
6075 68 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6076 68 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6077 68 : if (seq != NULL) {
6078 0 : bdev_io->internal.f.has_accel_sequence = true;
6079 0 : bdev_io->internal.accel_sequence = seq;
6080 : }
6081 :
6082 68 : if (domain != NULL) {
6083 2 : bdev_io->internal.f.has_memory_domain = true;
6084 2 : bdev_io->internal.memory_domain = domain;
6085 2 : bdev_io->internal.memory_domain_ctx = domain_ctx;
6086 : }
6087 :
6088 68 : bdev_io->u.bdev.memory_domain = domain;
6089 68 : bdev_io->u.bdev.memory_domain_ctx = domain_ctx;
6090 68 : bdev_io->u.bdev.accel_sequence = seq;
6091 68 : bdev_io->u.bdev.dif_check_flags = dif_check_flags;
6092 68 : bdev_io->u.bdev.nvme_cdw12.raw = nvme_cdw12_raw;
6093 68 : bdev_io->u.bdev.nvme_cdw13.raw = nvme_cdw13_raw;
6094 :
6095 68 : _bdev_io_submit_ext(desc, bdev_io);
6096 :
6097 68 : return 0;
6098 : }
6099 :
6100 : int
6101 3 : spdk_bdev_writev(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6102 : struct iovec *iov, int iovcnt,
6103 : uint64_t offset, uint64_t len,
6104 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6105 : {
6106 3 : uint64_t offset_blocks, num_blocks;
6107 :
6108 3 : if (bdev_bytes_to_blocks(desc, offset, &offset_blocks, len, &num_blocks) != 0) {
6109 0 : return -EINVAL;
6110 : }
6111 :
6112 3 : return spdk_bdev_writev_blocks(desc, ch, iov, iovcnt, offset_blocks, num_blocks, cb, cb_arg);
6113 : }
6114 :
6115 : int
6116 14 : spdk_bdev_writev_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6117 : struct iovec *iov, int iovcnt,
6118 : uint64_t offset_blocks, uint64_t num_blocks,
6119 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6120 : {
6121 14 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6122 :
6123 14 : return bdev_writev_blocks_with_md(desc, ch, iov, iovcnt, NULL, offset_blocks,
6124 : num_blocks, NULL, NULL, NULL, bdev->dif_check_flags, 0, 0,
6125 : cb, cb_arg);
6126 : }
6127 :
6128 : int
6129 1 : spdk_bdev_writev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6130 : struct iovec *iov, int iovcnt, void *md_buf,
6131 : uint64_t offset_blocks, uint64_t num_blocks,
6132 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6133 : {
6134 1 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6135 :
6136 1 : if (md_buf && !spdk_bdev_is_md_separate(bdev)) {
6137 0 : return -EINVAL;
6138 : }
6139 :
6140 1 : if (md_buf && !_is_buf_allocated(iov)) {
6141 0 : return -EINVAL;
6142 : }
6143 :
6144 1 : return bdev_writev_blocks_with_md(desc, ch, iov, iovcnt, md_buf, offset_blocks,
6145 : num_blocks, NULL, NULL, NULL, bdev->dif_check_flags, 0, 0,
6146 : cb, cb_arg);
6147 : }
6148 :
6149 : int
6150 8 : spdk_bdev_writev_blocks_ext(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6151 : struct iovec *iov, int iovcnt,
6152 : uint64_t offset_blocks, uint64_t num_blocks,
6153 : spdk_bdev_io_completion_cb cb, void *cb_arg,
6154 : struct spdk_bdev_ext_io_opts *opts)
6155 : {
6156 8 : struct spdk_memory_domain *domain = NULL;
6157 8 : struct spdk_accel_sequence *seq = NULL;
6158 8 : void *domain_ctx = NULL, *md = NULL;
6159 8 : uint32_t dif_check_flags = 0;
6160 8 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6161 8 : uint32_t nvme_cdw12_raw = 0;
6162 8 : uint32_t nvme_cdw13_raw = 0;
6163 :
6164 8 : if (opts) {
6165 7 : if (spdk_unlikely(!_bdev_io_check_opts(opts, iov))) {
6166 3 : return -EINVAL;
6167 : }
6168 4 : md = opts->metadata;
6169 4 : domain = bdev_get_ext_io_opt(opts, memory_domain, NULL);
6170 4 : domain_ctx = bdev_get_ext_io_opt(opts, memory_domain_ctx, NULL);
6171 4 : seq = bdev_get_ext_io_opt(opts, accel_sequence, NULL);
6172 4 : nvme_cdw12_raw = bdev_get_ext_io_opt(opts, nvme_cdw12.raw, 0);
6173 4 : nvme_cdw13_raw = bdev_get_ext_io_opt(opts, nvme_cdw13.raw, 0);
6174 4 : if (md) {
6175 4 : if (spdk_unlikely(!spdk_bdev_is_md_separate(bdev))) {
6176 0 : return -EINVAL;
6177 : }
6178 :
6179 4 : if (spdk_unlikely(!_is_buf_allocated(iov))) {
6180 0 : return -EINVAL;
6181 : }
6182 :
6183 4 : if (spdk_unlikely(seq != NULL)) {
6184 0 : return -EINVAL;
6185 : }
6186 :
6187 4 : if (nvme_cdw12_raw & SPDK_DIF_FLAGS_NVME_PRACT) {
6188 0 : SPDK_ERRLOG("Separate metadata with NVMe PRACT is not supported.\n");
6189 0 : return -ENOTSUP;
6190 : }
6191 : }
6192 :
6193 4 : if (nvme_cdw12_raw & SPDK_DIF_FLAGS_NVME_PRACT) {
6194 0 : dif_check_flags |= SPDK_DIF_FLAGS_NVME_PRACT;
6195 : }
6196 : }
6197 :
6198 5 : dif_check_flags |= bdev->dif_check_flags &
6199 5 : ~(bdev_get_ext_io_opt(opts, dif_check_flags_exclude_mask, 0));
6200 :
6201 5 : return bdev_writev_blocks_with_md(desc, ch, iov, iovcnt, md, offset_blocks, num_blocks,
6202 : domain, domain_ctx, seq, dif_check_flags,
6203 : nvme_cdw12_raw, nvme_cdw13_raw, cb, cb_arg);
6204 : }
6205 :
6206 : static void
6207 11 : bdev_compare_do_read_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
6208 : {
6209 11 : struct spdk_bdev_io *parent_io = cb_arg;
6210 11 : struct spdk_bdev *bdev = parent_io->bdev;
6211 11 : uint8_t *read_buf = bdev_io->u.bdev.iovs[0].iov_base;
6212 11 : int i, rc = 0;
6213 :
6214 11 : if (!success) {
6215 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
6216 0 : parent_io->internal.cb(parent_io, false, parent_io->internal.caller_ctx);
6217 0 : spdk_bdev_free_io(bdev_io);
6218 0 : return;
6219 : }
6220 :
6221 17 : for (i = 0; i < parent_io->u.bdev.iovcnt; i++) {
6222 22 : rc = memcmp(read_buf,
6223 11 : parent_io->u.bdev.iovs[i].iov_base,
6224 11 : parent_io->u.bdev.iovs[i].iov_len);
6225 11 : if (rc) {
6226 5 : break;
6227 : }
6228 6 : read_buf += parent_io->u.bdev.iovs[i].iov_len;
6229 : }
6230 :
6231 11 : if (rc == 0 && parent_io->u.bdev.md_buf && spdk_bdev_is_md_separate(bdev)) {
6232 2 : rc = memcmp(bdev_io->u.bdev.md_buf,
6233 2 : parent_io->u.bdev.md_buf,
6234 2 : spdk_bdev_get_md_size(bdev));
6235 : }
6236 :
6237 11 : spdk_bdev_free_io(bdev_io);
6238 :
6239 11 : if (rc == 0) {
6240 5 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
6241 5 : parent_io->internal.cb(parent_io, true, parent_io->internal.caller_ctx);
6242 : } else {
6243 6 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_MISCOMPARE;
6244 6 : parent_io->internal.cb(parent_io, false, parent_io->internal.caller_ctx);
6245 : }
6246 : }
6247 :
6248 : static void
6249 11 : bdev_compare_do_read(void *_bdev_io)
6250 : {
6251 11 : struct spdk_bdev_io *bdev_io = _bdev_io;
6252 : int rc;
6253 :
6254 22 : rc = spdk_bdev_read_blocks(bdev_io->internal.desc,
6255 11 : spdk_io_channel_from_ctx(bdev_io->internal.ch), NULL,
6256 : bdev_io->u.bdev.offset_blocks, bdev_io->u.bdev.num_blocks,
6257 : bdev_compare_do_read_done, bdev_io);
6258 :
6259 11 : if (rc == -ENOMEM) {
6260 0 : bdev_queue_io_wait_with_cb(bdev_io, bdev_compare_do_read);
6261 11 : } else if (rc != 0) {
6262 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
6263 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
6264 : }
6265 11 : }
6266 :
6267 : static int
6268 16 : bdev_comparev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6269 : struct iovec *iov, int iovcnt, void *md_buf,
6270 : uint64_t offset_blocks, uint64_t num_blocks,
6271 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6272 : {
6273 16 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6274 : struct spdk_bdev_io *bdev_io;
6275 16 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6276 :
6277 16 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6278 0 : return -EINVAL;
6279 : }
6280 :
6281 16 : bdev_io = bdev_channel_get_io(channel);
6282 16 : if (!bdev_io) {
6283 0 : return -ENOMEM;
6284 : }
6285 :
6286 16 : bdev_io->internal.ch = channel;
6287 16 : bdev_io->internal.desc = desc;
6288 16 : bdev_io->type = SPDK_BDEV_IO_TYPE_COMPARE;
6289 16 : bdev_io->u.bdev.iovs = iov;
6290 16 : bdev_io->u.bdev.iovcnt = iovcnt;
6291 16 : bdev_io->u.bdev.md_buf = md_buf;
6292 16 : bdev_io->u.bdev.num_blocks = num_blocks;
6293 16 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6294 16 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6295 16 : bdev_io->u.bdev.memory_domain = NULL;
6296 16 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6297 16 : bdev_io->u.bdev.accel_sequence = NULL;
6298 :
6299 16 : if (bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COMPARE)) {
6300 7 : bdev_io_submit(bdev_io);
6301 7 : return 0;
6302 : }
6303 :
6304 9 : bdev_compare_do_read(bdev_io);
6305 :
6306 9 : return 0;
6307 : }
6308 :
6309 : int
6310 10 : spdk_bdev_comparev_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6311 : struct iovec *iov, int iovcnt,
6312 : uint64_t offset_blocks, uint64_t num_blocks,
6313 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6314 : {
6315 10 : return bdev_comparev_blocks_with_md(desc, ch, iov, iovcnt, NULL, offset_blocks,
6316 : num_blocks, cb, cb_arg);
6317 : }
6318 :
6319 : int
6320 6 : spdk_bdev_comparev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6321 : struct iovec *iov, int iovcnt, void *md_buf,
6322 : uint64_t offset_blocks, uint64_t num_blocks,
6323 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6324 : {
6325 6 : if (md_buf && !spdk_bdev_is_md_separate(spdk_bdev_desc_get_bdev(desc))) {
6326 0 : return -EINVAL;
6327 : }
6328 :
6329 6 : if (md_buf && !_is_buf_allocated(iov)) {
6330 0 : return -EINVAL;
6331 : }
6332 :
6333 6 : return bdev_comparev_blocks_with_md(desc, ch, iov, iovcnt, md_buf, offset_blocks,
6334 : num_blocks, cb, cb_arg);
6335 : }
6336 :
6337 : static int
6338 4 : bdev_compare_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6339 : void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
6340 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6341 : {
6342 4 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6343 : struct spdk_bdev_io *bdev_io;
6344 4 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6345 :
6346 4 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6347 0 : return -EINVAL;
6348 : }
6349 :
6350 4 : bdev_io = bdev_channel_get_io(channel);
6351 4 : if (!bdev_io) {
6352 0 : return -ENOMEM;
6353 : }
6354 :
6355 4 : bdev_io->internal.ch = channel;
6356 4 : bdev_io->internal.desc = desc;
6357 4 : bdev_io->type = SPDK_BDEV_IO_TYPE_COMPARE;
6358 4 : bdev_io->u.bdev.iovs = &bdev_io->iov;
6359 4 : bdev_io->u.bdev.iovs[0].iov_base = buf;
6360 4 : bdev_io->u.bdev.iovs[0].iov_len = num_blocks * bdev_desc_get_block_size(desc);
6361 4 : bdev_io->u.bdev.iovcnt = 1;
6362 4 : bdev_io->u.bdev.md_buf = md_buf;
6363 4 : bdev_io->u.bdev.num_blocks = num_blocks;
6364 4 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6365 4 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6366 4 : bdev_io->u.bdev.memory_domain = NULL;
6367 4 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6368 4 : bdev_io->u.bdev.accel_sequence = NULL;
6369 :
6370 4 : if (bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COMPARE)) {
6371 2 : bdev_io_submit(bdev_io);
6372 2 : return 0;
6373 : }
6374 :
6375 2 : bdev_compare_do_read(bdev_io);
6376 :
6377 2 : return 0;
6378 : }
6379 :
6380 : int
6381 4 : spdk_bdev_compare_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6382 : void *buf, uint64_t offset_blocks, uint64_t num_blocks,
6383 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6384 : {
6385 4 : return bdev_compare_blocks_with_md(desc, ch, buf, NULL, offset_blocks, num_blocks,
6386 : cb, cb_arg);
6387 : }
6388 :
6389 : int
6390 0 : spdk_bdev_compare_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6391 : void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
6392 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6393 : {
6394 0 : struct iovec iov = {
6395 : .iov_base = buf,
6396 : };
6397 :
6398 0 : if (md_buf && !spdk_bdev_is_md_separate(spdk_bdev_desc_get_bdev(desc))) {
6399 0 : return -EINVAL;
6400 : }
6401 :
6402 0 : if (md_buf && !_is_buf_allocated(&iov)) {
6403 0 : return -EINVAL;
6404 : }
6405 :
6406 0 : return bdev_compare_blocks_with_md(desc, ch, buf, md_buf, offset_blocks, num_blocks,
6407 : cb, cb_arg);
6408 : }
6409 :
6410 : static void
6411 2 : bdev_comparev_and_writev_blocks_unlocked(struct lba_range *range, void *ctx, int unlock_status)
6412 : {
6413 2 : struct spdk_bdev_io *bdev_io = ctx;
6414 :
6415 2 : if (unlock_status) {
6416 0 : SPDK_ERRLOG("LBA range unlock failed\n");
6417 : }
6418 :
6419 2 : bdev_io->internal.cb(bdev_io, bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS ? true :
6420 : false, bdev_io->internal.caller_ctx);
6421 2 : }
6422 :
6423 : static void
6424 2 : bdev_comparev_and_writev_blocks_unlock(struct spdk_bdev_io *bdev_io, int status)
6425 : {
6426 2 : bdev_io->internal.status = status;
6427 :
6428 2 : bdev_unlock_lba_range(bdev_io->internal.desc, spdk_io_channel_from_ctx(bdev_io->internal.ch),
6429 : bdev_io->u.bdev.offset_blocks, bdev_io->u.bdev.num_blocks,
6430 : bdev_comparev_and_writev_blocks_unlocked, bdev_io);
6431 2 : }
6432 :
6433 : static void
6434 1 : bdev_compare_and_write_do_write_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
6435 : {
6436 1 : struct spdk_bdev_io *parent_io = cb_arg;
6437 :
6438 1 : if (!success) {
6439 0 : SPDK_ERRLOG("Compare and write operation failed\n");
6440 : }
6441 :
6442 1 : spdk_bdev_free_io(bdev_io);
6443 :
6444 1 : bdev_comparev_and_writev_blocks_unlock(parent_io,
6445 : success ? SPDK_BDEV_IO_STATUS_SUCCESS : SPDK_BDEV_IO_STATUS_FAILED);
6446 1 : }
6447 :
6448 : static void
6449 1 : bdev_compare_and_write_do_write(void *_bdev_io)
6450 : {
6451 1 : struct spdk_bdev_io *bdev_io = _bdev_io;
6452 : int rc;
6453 :
6454 2 : rc = spdk_bdev_writev_blocks(bdev_io->internal.desc,
6455 1 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
6456 : bdev_io->u.bdev.fused_iovs, bdev_io->u.bdev.fused_iovcnt,
6457 : bdev_io->u.bdev.offset_blocks, bdev_io->u.bdev.num_blocks,
6458 : bdev_compare_and_write_do_write_done, bdev_io);
6459 :
6460 :
6461 1 : if (rc == -ENOMEM) {
6462 0 : bdev_queue_io_wait_with_cb(bdev_io, bdev_compare_and_write_do_write);
6463 1 : } else if (rc != 0) {
6464 0 : bdev_comparev_and_writev_blocks_unlock(bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
6465 : }
6466 1 : }
6467 :
6468 : static void
6469 2 : bdev_compare_and_write_do_compare_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
6470 : {
6471 2 : struct spdk_bdev_io *parent_io = cb_arg;
6472 :
6473 2 : spdk_bdev_free_io(bdev_io);
6474 :
6475 2 : if (!success) {
6476 1 : bdev_comparev_and_writev_blocks_unlock(parent_io, SPDK_BDEV_IO_STATUS_MISCOMPARE);
6477 1 : return;
6478 : }
6479 :
6480 1 : bdev_compare_and_write_do_write(parent_io);
6481 : }
6482 :
6483 : static void
6484 2 : bdev_compare_and_write_do_compare(void *_bdev_io)
6485 : {
6486 2 : struct spdk_bdev_io *bdev_io = _bdev_io;
6487 : int rc;
6488 :
6489 4 : rc = spdk_bdev_comparev_blocks(bdev_io->internal.desc,
6490 2 : spdk_io_channel_from_ctx(bdev_io->internal.ch), bdev_io->u.bdev.iovs,
6491 : bdev_io->u.bdev.iovcnt, bdev_io->u.bdev.offset_blocks, bdev_io->u.bdev.num_blocks,
6492 : bdev_compare_and_write_do_compare_done, bdev_io);
6493 :
6494 2 : if (rc == -ENOMEM) {
6495 0 : bdev_queue_io_wait_with_cb(bdev_io, bdev_compare_and_write_do_compare);
6496 2 : } else if (rc != 0) {
6497 0 : bdev_comparev_and_writev_blocks_unlock(bdev_io, SPDK_BDEV_IO_STATUS_FIRST_FUSED_FAILED);
6498 : }
6499 2 : }
6500 :
6501 : static void
6502 2 : bdev_comparev_and_writev_blocks_locked(struct lba_range *range, void *ctx, int status)
6503 : {
6504 2 : struct spdk_bdev_io *bdev_io = ctx;
6505 :
6506 2 : if (status) {
6507 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FIRST_FUSED_FAILED;
6508 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
6509 0 : return;
6510 : }
6511 :
6512 2 : bdev_compare_and_write_do_compare(bdev_io);
6513 : }
6514 :
6515 : int
6516 2 : spdk_bdev_comparev_and_writev_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6517 : struct iovec *compare_iov, int compare_iovcnt,
6518 : struct iovec *write_iov, int write_iovcnt,
6519 : uint64_t offset_blocks, uint64_t num_blocks,
6520 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6521 : {
6522 2 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6523 : struct spdk_bdev_io *bdev_io;
6524 2 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6525 :
6526 2 : if (!desc->write) {
6527 0 : return -EBADF;
6528 : }
6529 :
6530 2 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6531 0 : return -EINVAL;
6532 : }
6533 :
6534 2 : if (num_blocks > bdev->acwu) {
6535 0 : return -EINVAL;
6536 : }
6537 :
6538 2 : bdev_io = bdev_channel_get_io(channel);
6539 2 : if (!bdev_io) {
6540 0 : return -ENOMEM;
6541 : }
6542 :
6543 2 : bdev_io->internal.ch = channel;
6544 2 : bdev_io->internal.desc = desc;
6545 2 : bdev_io->type = SPDK_BDEV_IO_TYPE_COMPARE_AND_WRITE;
6546 2 : bdev_io->u.bdev.iovs = compare_iov;
6547 2 : bdev_io->u.bdev.iovcnt = compare_iovcnt;
6548 2 : bdev_io->u.bdev.fused_iovs = write_iov;
6549 2 : bdev_io->u.bdev.fused_iovcnt = write_iovcnt;
6550 2 : bdev_io->u.bdev.md_buf = NULL;
6551 2 : bdev_io->u.bdev.num_blocks = num_blocks;
6552 2 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6553 2 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6554 2 : bdev_io->u.bdev.memory_domain = NULL;
6555 2 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6556 2 : bdev_io->u.bdev.accel_sequence = NULL;
6557 :
6558 2 : if (bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COMPARE_AND_WRITE)) {
6559 0 : bdev_io_submit(bdev_io);
6560 0 : return 0;
6561 : }
6562 :
6563 2 : return bdev_lock_lba_range(desc, ch, offset_blocks, num_blocks,
6564 : bdev_comparev_and_writev_blocks_locked, bdev_io);
6565 : }
6566 :
6567 : int
6568 2 : spdk_bdev_zcopy_start(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6569 : struct iovec *iov, int iovcnt,
6570 : uint64_t offset_blocks, uint64_t num_blocks,
6571 : bool populate,
6572 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6573 : {
6574 2 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6575 : struct spdk_bdev_io *bdev_io;
6576 2 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6577 :
6578 2 : if (!desc->write) {
6579 0 : return -EBADF;
6580 : }
6581 :
6582 2 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6583 0 : return -EINVAL;
6584 : }
6585 :
6586 2 : if (!spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_ZCOPY)) {
6587 0 : return -ENOTSUP;
6588 : }
6589 :
6590 2 : bdev_io = bdev_channel_get_io(channel);
6591 2 : if (!bdev_io) {
6592 0 : return -ENOMEM;
6593 : }
6594 :
6595 2 : bdev_io->internal.ch = channel;
6596 2 : bdev_io->internal.desc = desc;
6597 2 : bdev_io->type = SPDK_BDEV_IO_TYPE_ZCOPY;
6598 2 : bdev_io->u.bdev.num_blocks = num_blocks;
6599 2 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6600 2 : bdev_io->u.bdev.iovs = iov;
6601 2 : bdev_io->u.bdev.iovcnt = iovcnt;
6602 2 : bdev_io->u.bdev.md_buf = NULL;
6603 2 : bdev_io->u.bdev.zcopy.populate = populate ? 1 : 0;
6604 2 : bdev_io->u.bdev.zcopy.commit = 0;
6605 2 : bdev_io->u.bdev.zcopy.start = 1;
6606 2 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6607 2 : bdev_io->u.bdev.memory_domain = NULL;
6608 2 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6609 2 : bdev_io->u.bdev.accel_sequence = NULL;
6610 :
6611 2 : bdev_io_submit(bdev_io);
6612 :
6613 2 : return 0;
6614 : }
6615 :
6616 : int
6617 2 : spdk_bdev_zcopy_end(struct spdk_bdev_io *bdev_io, bool commit,
6618 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6619 : {
6620 2 : if (bdev_io->type != SPDK_BDEV_IO_TYPE_ZCOPY) {
6621 0 : return -EINVAL;
6622 : }
6623 :
6624 2 : bdev_io->u.bdev.zcopy.commit = commit ? 1 : 0;
6625 2 : bdev_io->u.bdev.zcopy.start = 0;
6626 2 : bdev_io->internal.caller_ctx = cb_arg;
6627 2 : bdev_io->internal.cb = cb;
6628 2 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_PENDING;
6629 :
6630 2 : bdev_io_submit(bdev_io);
6631 :
6632 2 : return 0;
6633 : }
6634 :
6635 : int
6636 0 : spdk_bdev_write_zeroes(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6637 : uint64_t offset, uint64_t len,
6638 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6639 : {
6640 0 : uint64_t offset_blocks, num_blocks;
6641 :
6642 0 : if (bdev_bytes_to_blocks(desc, offset, &offset_blocks, len, &num_blocks) != 0) {
6643 0 : return -EINVAL;
6644 : }
6645 :
6646 0 : return spdk_bdev_write_zeroes_blocks(desc, ch, offset_blocks, num_blocks, cb, cb_arg);
6647 : }
6648 :
6649 : int
6650 33 : spdk_bdev_write_zeroes_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6651 : uint64_t offset_blocks, uint64_t num_blocks,
6652 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6653 : {
6654 33 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6655 : struct spdk_bdev_io *bdev_io;
6656 33 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6657 :
6658 33 : if (!desc->write) {
6659 0 : return -EBADF;
6660 : }
6661 :
6662 33 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6663 0 : return -EINVAL;
6664 : }
6665 :
6666 33 : if (!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE_ZEROES) &&
6667 10 : !bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE)) {
6668 1 : return -ENOTSUP;
6669 : }
6670 :
6671 32 : bdev_io = bdev_channel_get_io(channel);
6672 :
6673 32 : if (!bdev_io) {
6674 0 : return -ENOMEM;
6675 : }
6676 :
6677 32 : bdev_io->type = SPDK_BDEV_IO_TYPE_WRITE_ZEROES;
6678 32 : bdev_io->internal.ch = channel;
6679 32 : bdev_io->internal.desc = desc;
6680 32 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6681 32 : bdev_io->u.bdev.num_blocks = num_blocks;
6682 32 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6683 32 : bdev_io->u.bdev.memory_domain = NULL;
6684 32 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6685 32 : bdev_io->u.bdev.accel_sequence = NULL;
6686 :
6687 : /* If the write_zeroes size is large and should be split, use the generic split
6688 : * logic regardless of whether SPDK_BDEV_IO_TYPE_WRITE_ZEREOS is supported or not.
6689 : *
6690 : * Then, send the write_zeroes request if SPDK_BDEV_IO_TYPE_WRITE_ZEROES is supported
6691 : * or emulate it using regular write request otherwise.
6692 : */
6693 32 : if (bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE_ZEROES) ||
6694 : bdev_io->internal.f.split) {
6695 26 : bdev_io_submit(bdev_io);
6696 26 : return 0;
6697 : }
6698 :
6699 6 : assert(_bdev_get_block_size_with_md(bdev) <= ZERO_BUFFER_SIZE);
6700 :
6701 6 : return bdev_write_zero_buffer(bdev_io);
6702 : }
6703 :
6704 : int
6705 0 : spdk_bdev_unmap(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6706 : uint64_t offset, uint64_t nbytes,
6707 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6708 : {
6709 0 : uint64_t offset_blocks, num_blocks;
6710 :
6711 0 : if (bdev_bytes_to_blocks(desc, offset, &offset_blocks, nbytes, &num_blocks) != 0) {
6712 0 : return -EINVAL;
6713 : }
6714 :
6715 0 : return spdk_bdev_unmap_blocks(desc, ch, offset_blocks, num_blocks, cb, cb_arg);
6716 : }
6717 :
6718 : static void
6719 0 : bdev_io_complete_cb(void *ctx)
6720 : {
6721 0 : struct spdk_bdev_io *bdev_io = ctx;
6722 :
6723 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
6724 0 : bdev_io->internal.cb(bdev_io, true, bdev_io->internal.caller_ctx);
6725 0 : }
6726 :
6727 : int
6728 22 : spdk_bdev_unmap_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6729 : uint64_t offset_blocks, uint64_t num_blocks,
6730 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6731 : {
6732 22 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6733 : struct spdk_bdev_io *bdev_io;
6734 22 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6735 :
6736 22 : if (!desc->write) {
6737 0 : return -EBADF;
6738 : }
6739 :
6740 22 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6741 0 : return -EINVAL;
6742 : }
6743 :
6744 22 : bdev_io = bdev_channel_get_io(channel);
6745 22 : if (!bdev_io) {
6746 0 : return -ENOMEM;
6747 : }
6748 :
6749 22 : bdev_io->internal.ch = channel;
6750 22 : bdev_io->internal.desc = desc;
6751 22 : bdev_io->type = SPDK_BDEV_IO_TYPE_UNMAP;
6752 :
6753 22 : bdev_io->u.bdev.iovs = &bdev_io->iov;
6754 22 : bdev_io->u.bdev.iovs[0].iov_base = NULL;
6755 22 : bdev_io->u.bdev.iovs[0].iov_len = 0;
6756 22 : bdev_io->u.bdev.iovcnt = 1;
6757 :
6758 22 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6759 22 : bdev_io->u.bdev.num_blocks = num_blocks;
6760 22 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6761 22 : bdev_io->u.bdev.memory_domain = NULL;
6762 22 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6763 22 : bdev_io->u.bdev.accel_sequence = NULL;
6764 :
6765 22 : if (num_blocks == 0) {
6766 0 : spdk_thread_send_msg(spdk_get_thread(), bdev_io_complete_cb, bdev_io);
6767 0 : return 0;
6768 : }
6769 :
6770 22 : bdev_io_submit(bdev_io);
6771 22 : return 0;
6772 : }
6773 :
6774 : int
6775 0 : spdk_bdev_flush(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6776 : uint64_t offset, uint64_t length,
6777 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6778 : {
6779 0 : uint64_t offset_blocks, num_blocks;
6780 :
6781 0 : if (bdev_bytes_to_blocks(desc, offset, &offset_blocks, length, &num_blocks) != 0) {
6782 0 : return -EINVAL;
6783 : }
6784 :
6785 0 : return spdk_bdev_flush_blocks(desc, ch, offset_blocks, num_blocks, cb, cb_arg);
6786 : }
6787 :
6788 : int
6789 2 : spdk_bdev_flush_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6790 : uint64_t offset_blocks, uint64_t num_blocks,
6791 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6792 : {
6793 2 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6794 : struct spdk_bdev_io *bdev_io;
6795 2 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6796 :
6797 2 : if (!desc->write) {
6798 0 : return -EBADF;
6799 : }
6800 :
6801 2 : if (spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_FLUSH))) {
6802 0 : return -ENOTSUP;
6803 : }
6804 :
6805 2 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6806 0 : return -EINVAL;
6807 : }
6808 :
6809 2 : bdev_io = bdev_channel_get_io(channel);
6810 2 : if (!bdev_io) {
6811 0 : return -ENOMEM;
6812 : }
6813 :
6814 2 : bdev_io->internal.ch = channel;
6815 2 : bdev_io->internal.desc = desc;
6816 2 : bdev_io->type = SPDK_BDEV_IO_TYPE_FLUSH;
6817 2 : bdev_io->u.bdev.iovs = NULL;
6818 2 : bdev_io->u.bdev.iovcnt = 0;
6819 2 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6820 2 : bdev_io->u.bdev.num_blocks = num_blocks;
6821 2 : bdev_io->u.bdev.memory_domain = NULL;
6822 2 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6823 2 : bdev_io->u.bdev.accel_sequence = NULL;
6824 2 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6825 :
6826 2 : bdev_io_submit(bdev_io);
6827 2 : return 0;
6828 : }
6829 :
6830 : static int bdev_reset_poll_for_outstanding_io(void *ctx);
6831 :
6832 : static void
6833 13 : bdev_reset_check_outstanding_io_done(struct spdk_bdev *bdev, void *_ctx, int status)
6834 : {
6835 13 : struct spdk_bdev_io *bdev_io = _ctx;
6836 13 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
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 : bdev_io, BDEV_RESET_CHECK_OUTSTANDING_IO_PERIOD);
6842 : } else {
6843 1 : if (TAILQ_EMPTY(&ch->io_memory_domain) && TAILQ_EMPTY(&ch->io_accel_exec)) {
6844 : /* If outstanding IOs are still present and reset_io_drain_timeout
6845 : * seconds passed, start the reset. */
6846 1 : bdev_io_submit_reset(bdev_io);
6847 : } else {
6848 : /* We still have in progress memory domain pull/push or we're
6849 : * executing accel sequence. Since we cannot abort either of those
6850 : * operations, fail the reset request. */
6851 0 : spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
6852 : }
6853 : }
6854 : } else {
6855 4 : SPDK_DEBUGLOG(bdev,
6856 : "Skipping reset for underlying device of bdev: %s - no outstanding I/O.\n",
6857 : ch->bdev->name);
6858 : /* Mark the completion status as a SUCCESS and complete the reset. */
6859 4 : spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_SUCCESS);
6860 : }
6861 13 : }
6862 :
6863 : static void
6864 13 : bdev_reset_check_outstanding_io(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
6865 : struct spdk_io_channel *io_ch, void *_ctx)
6866 : {
6867 13 : struct spdk_bdev_channel *cur_ch = __io_ch_to_bdev_ch(io_ch);
6868 13 : int status = 0;
6869 :
6870 13 : if (cur_ch->io_outstanding > 0 ||
6871 4 : !TAILQ_EMPTY(&cur_ch->io_memory_domain) ||
6872 4 : !TAILQ_EMPTY(&cur_ch->io_accel_exec)) {
6873 : /* If a channel has outstanding IO, set status to -EBUSY code. This will stop
6874 : * further iteration over the rest of the channels and pass non-zero status
6875 : * to the callback function. */
6876 9 : status = -EBUSY;
6877 : }
6878 13 : spdk_bdev_for_each_channel_continue(i, status);
6879 13 : }
6880 :
6881 : static int
6882 8 : bdev_reset_poll_for_outstanding_io(void *ctx)
6883 : {
6884 8 : struct spdk_bdev_io *bdev_io = ctx;
6885 :
6886 8 : spdk_poller_unregister(&bdev_io->u.reset.wait_poller.poller);
6887 8 : spdk_bdev_for_each_channel(bdev_io->bdev, bdev_reset_check_outstanding_io, bdev_io,
6888 : bdev_reset_check_outstanding_io_done);
6889 :
6890 8 : return SPDK_POLLER_BUSY;
6891 : }
6892 :
6893 : static void
6894 16 : bdev_reset_freeze_channel_done(struct spdk_bdev *bdev, void *_ctx, int status)
6895 : {
6896 16 : struct spdk_bdev_io *bdev_io = _ctx;
6897 :
6898 16 : if (bdev->reset_io_drain_timeout == 0) {
6899 11 : bdev_io_submit_reset(bdev_io);
6900 11 : return;
6901 : }
6902 :
6903 10 : bdev_io->u.reset.wait_poller.stop_time_tsc = spdk_get_ticks() +
6904 5 : (bdev->reset_io_drain_timeout * spdk_get_ticks_hz());
6905 :
6906 : /* In case bdev->reset_io_drain_timeout is not equal to zero,
6907 : * submit the reset to the underlying module only if outstanding I/O
6908 : * remain after reset_io_drain_timeout seconds have passed. */
6909 5 : spdk_bdev_for_each_channel(bdev, bdev_reset_check_outstanding_io, bdev_io,
6910 : bdev_reset_check_outstanding_io_done);
6911 : }
6912 :
6913 : static void
6914 19 : bdev_reset_freeze_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
6915 : struct spdk_io_channel *ch, void *_ctx)
6916 : {
6917 : struct spdk_bdev_channel *channel;
6918 : struct spdk_bdev_mgmt_channel *mgmt_channel;
6919 : struct spdk_bdev_shared_resource *shared_resource;
6920 19 : bdev_io_tailq_t tmp_queued;
6921 :
6922 19 : TAILQ_INIT(&tmp_queued);
6923 :
6924 19 : channel = __io_ch_to_bdev_ch(ch);
6925 19 : shared_resource = channel->shared_resource;
6926 19 : mgmt_channel = shared_resource->mgmt_ch;
6927 :
6928 19 : channel->flags |= BDEV_CH_RESET_IN_PROGRESS;
6929 :
6930 19 : if ((channel->flags & BDEV_CH_QOS_ENABLED) != 0) {
6931 2 : TAILQ_SWAP(&channel->qos_queued_io, &tmp_queued, spdk_bdev_io, internal.link);
6932 : }
6933 :
6934 19 : bdev_abort_all_queued_io(&shared_resource->nomem_io, channel);
6935 19 : bdev_abort_all_buf_io(mgmt_channel, channel);
6936 19 : bdev_abort_all_queued_io(&tmp_queued, channel);
6937 :
6938 19 : spdk_bdev_for_each_channel_continue(i, 0);
6939 19 : }
6940 :
6941 : static void
6942 18 : bdev_start_reset(struct spdk_bdev_io *bdev_io)
6943 : {
6944 18 : struct spdk_bdev *bdev = bdev_io->bdev;
6945 18 : bool freeze_channel = false;
6946 :
6947 18 : bdev_ch_add_to_io_submitted(bdev_io);
6948 :
6949 : /**
6950 : * Take a channel reference for the target bdev for the life of this
6951 : * reset. This guards against the channel getting destroyed before
6952 : * the reset is completed. We will release the reference when this
6953 : * reset is completed.
6954 : */
6955 18 : bdev_io->u.reset.ch_ref = spdk_get_io_channel(__bdev_to_io_dev(bdev));
6956 :
6957 18 : spdk_spin_lock(&bdev->internal.spinlock);
6958 18 : if (bdev->internal.reset_in_progress == NULL) {
6959 16 : bdev->internal.reset_in_progress = bdev_io;
6960 16 : freeze_channel = true;
6961 : } else {
6962 2 : TAILQ_INSERT_TAIL(&bdev->internal.queued_resets, bdev_io, internal.link);
6963 : }
6964 18 : spdk_spin_unlock(&bdev->internal.spinlock);
6965 :
6966 18 : if (freeze_channel) {
6967 16 : spdk_bdev_for_each_channel(bdev, bdev_reset_freeze_channel, bdev_io,
6968 : bdev_reset_freeze_channel_done);
6969 : }
6970 18 : }
6971 :
6972 : int
6973 18 : spdk_bdev_reset(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6974 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6975 : {
6976 18 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6977 : struct spdk_bdev_io *bdev_io;
6978 18 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6979 :
6980 18 : bdev_io = bdev_channel_get_io(channel);
6981 18 : if (!bdev_io) {
6982 0 : return -ENOMEM;
6983 : }
6984 :
6985 18 : bdev_io->internal.ch = channel;
6986 18 : bdev_io->internal.desc = desc;
6987 18 : bdev_io->internal.submit_tsc = spdk_get_ticks();
6988 18 : bdev_io->type = SPDK_BDEV_IO_TYPE_RESET;
6989 18 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6990 :
6991 18 : bdev_start_reset(bdev_io);
6992 18 : return 0;
6993 : }
6994 :
6995 : void
6996 0 : spdk_bdev_get_io_stat(struct spdk_bdev *bdev, struct spdk_io_channel *ch,
6997 : struct spdk_bdev_io_stat *stat, enum spdk_bdev_reset_stat_mode reset_mode)
6998 : {
6999 0 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
7000 :
7001 0 : bdev_get_io_stat(stat, channel->stat);
7002 0 : spdk_bdev_reset_io_stat(channel->stat, reset_mode);
7003 0 : }
7004 :
7005 : static void
7006 5 : bdev_get_device_stat_done(struct spdk_bdev *bdev, void *_ctx, int status)
7007 : {
7008 5 : struct spdk_bdev_iostat_ctx *bdev_iostat_ctx = _ctx;
7009 :
7010 5 : bdev_iostat_ctx->cb(bdev, bdev_iostat_ctx->stat,
7011 : bdev_iostat_ctx->cb_arg, 0);
7012 5 : free(bdev_iostat_ctx);
7013 5 : }
7014 :
7015 : static void
7016 4 : bdev_get_each_channel_stat(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
7017 : struct spdk_io_channel *ch, void *_ctx)
7018 : {
7019 4 : struct spdk_bdev_iostat_ctx *bdev_iostat_ctx = _ctx;
7020 4 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
7021 :
7022 4 : spdk_bdev_add_io_stat(bdev_iostat_ctx->stat, channel->stat);
7023 4 : spdk_bdev_reset_io_stat(channel->stat, bdev_iostat_ctx->reset_mode);
7024 4 : spdk_bdev_for_each_channel_continue(i, 0);
7025 4 : }
7026 :
7027 : void
7028 5 : spdk_bdev_get_device_stat(struct spdk_bdev *bdev, struct spdk_bdev_io_stat *stat,
7029 : enum spdk_bdev_reset_stat_mode reset_mode, spdk_bdev_get_device_stat_cb cb, void *cb_arg)
7030 : {
7031 : struct spdk_bdev_iostat_ctx *bdev_iostat_ctx;
7032 :
7033 5 : assert(bdev != NULL);
7034 5 : assert(stat != NULL);
7035 5 : assert(cb != NULL);
7036 :
7037 5 : bdev_iostat_ctx = calloc(1, sizeof(struct spdk_bdev_iostat_ctx));
7038 5 : if (bdev_iostat_ctx == NULL) {
7039 0 : SPDK_ERRLOG("Unable to allocate memory for spdk_bdev_iostat_ctx\n");
7040 0 : cb(bdev, stat, cb_arg, -ENOMEM);
7041 0 : return;
7042 : }
7043 :
7044 5 : bdev_iostat_ctx->stat = stat;
7045 5 : bdev_iostat_ctx->cb = cb;
7046 5 : bdev_iostat_ctx->cb_arg = cb_arg;
7047 5 : bdev_iostat_ctx->reset_mode = reset_mode;
7048 :
7049 : /* Start with the statistics from previously deleted channels. */
7050 5 : spdk_spin_lock(&bdev->internal.spinlock);
7051 5 : bdev_get_io_stat(bdev_iostat_ctx->stat, bdev->internal.stat);
7052 5 : spdk_bdev_reset_io_stat(bdev->internal.stat, reset_mode);
7053 5 : spdk_spin_unlock(&bdev->internal.spinlock);
7054 :
7055 : /* Then iterate and add the statistics from each existing channel. */
7056 5 : spdk_bdev_for_each_channel(bdev, bdev_get_each_channel_stat, bdev_iostat_ctx,
7057 : bdev_get_device_stat_done);
7058 : }
7059 :
7060 : struct bdev_iostat_reset_ctx {
7061 : enum spdk_bdev_reset_stat_mode mode;
7062 : bdev_reset_device_stat_cb cb;
7063 : void *cb_arg;
7064 : };
7065 :
7066 : static void
7067 0 : bdev_reset_device_stat_done(struct spdk_bdev *bdev, void *_ctx, int status)
7068 : {
7069 0 : struct bdev_iostat_reset_ctx *ctx = _ctx;
7070 :
7071 0 : ctx->cb(bdev, ctx->cb_arg, 0);
7072 :
7073 0 : free(ctx);
7074 0 : }
7075 :
7076 : static void
7077 0 : bdev_reset_each_channel_stat(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
7078 : struct spdk_io_channel *ch, void *_ctx)
7079 : {
7080 0 : struct bdev_iostat_reset_ctx *ctx = _ctx;
7081 0 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
7082 :
7083 0 : spdk_bdev_reset_io_stat(channel->stat, ctx->mode);
7084 :
7085 0 : spdk_bdev_for_each_channel_continue(i, 0);
7086 0 : }
7087 :
7088 : void
7089 0 : bdev_reset_device_stat(struct spdk_bdev *bdev, enum spdk_bdev_reset_stat_mode mode,
7090 : bdev_reset_device_stat_cb cb, void *cb_arg)
7091 : {
7092 : struct bdev_iostat_reset_ctx *ctx;
7093 :
7094 0 : assert(bdev != NULL);
7095 0 : assert(cb != NULL);
7096 :
7097 0 : ctx = calloc(1, sizeof(*ctx));
7098 0 : if (ctx == NULL) {
7099 0 : SPDK_ERRLOG("Unable to allocate bdev_iostat_reset_ctx.\n");
7100 0 : cb(bdev, cb_arg, -ENOMEM);
7101 0 : return;
7102 : }
7103 :
7104 0 : ctx->mode = mode;
7105 0 : ctx->cb = cb;
7106 0 : ctx->cb_arg = cb_arg;
7107 :
7108 0 : spdk_spin_lock(&bdev->internal.spinlock);
7109 0 : spdk_bdev_reset_io_stat(bdev->internal.stat, mode);
7110 0 : spdk_spin_unlock(&bdev->internal.spinlock);
7111 :
7112 0 : spdk_bdev_for_each_channel(bdev,
7113 : bdev_reset_each_channel_stat,
7114 : ctx,
7115 : bdev_reset_device_stat_done);
7116 : }
7117 :
7118 : int
7119 1 : spdk_bdev_nvme_admin_passthru(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
7120 : const struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes,
7121 : spdk_bdev_io_completion_cb cb, void *cb_arg)
7122 : {
7123 1 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
7124 : struct spdk_bdev_io *bdev_io;
7125 1 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
7126 :
7127 1 : if (!desc->write) {
7128 0 : return -EBADF;
7129 : }
7130 :
7131 1 : if (spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_ADMIN))) {
7132 1 : return -ENOTSUP;
7133 : }
7134 :
7135 0 : bdev_io = bdev_channel_get_io(channel);
7136 0 : if (!bdev_io) {
7137 0 : return -ENOMEM;
7138 : }
7139 :
7140 0 : bdev_io->internal.ch = channel;
7141 0 : bdev_io->internal.desc = desc;
7142 0 : bdev_io->type = SPDK_BDEV_IO_TYPE_NVME_ADMIN;
7143 0 : bdev_io->u.nvme_passthru.cmd = *cmd;
7144 0 : bdev_io->u.nvme_passthru.buf = buf;
7145 0 : bdev_io->u.nvme_passthru.nbytes = nbytes;
7146 0 : bdev_io->u.nvme_passthru.md_buf = NULL;
7147 0 : bdev_io->u.nvme_passthru.md_len = 0;
7148 :
7149 0 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
7150 :
7151 0 : bdev_io_submit(bdev_io);
7152 0 : return 0;
7153 : }
7154 :
7155 : int
7156 1 : spdk_bdev_nvme_io_passthru(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
7157 : const struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes,
7158 : spdk_bdev_io_completion_cb cb, void *cb_arg)
7159 : {
7160 1 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
7161 : struct spdk_bdev_io *bdev_io;
7162 1 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
7163 :
7164 1 : if (!desc->write) {
7165 : /*
7166 : * Do not try to parse the NVMe command - we could maybe use bits in the opcode
7167 : * to easily determine if the command is a read or write, but for now just
7168 : * do not allow io_passthru with a read-only descriptor.
7169 : */
7170 0 : return -EBADF;
7171 : }
7172 :
7173 1 : if (spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_IO))) {
7174 1 : return -ENOTSUP;
7175 : }
7176 :
7177 0 : bdev_io = bdev_channel_get_io(channel);
7178 0 : if (!bdev_io) {
7179 0 : return -ENOMEM;
7180 : }
7181 :
7182 0 : bdev_io->internal.ch = channel;
7183 0 : bdev_io->internal.desc = desc;
7184 0 : bdev_io->type = SPDK_BDEV_IO_TYPE_NVME_IO;
7185 0 : bdev_io->u.nvme_passthru.cmd = *cmd;
7186 0 : bdev_io->u.nvme_passthru.buf = buf;
7187 0 : bdev_io->u.nvme_passthru.nbytes = nbytes;
7188 0 : bdev_io->u.nvme_passthru.md_buf = NULL;
7189 0 : bdev_io->u.nvme_passthru.md_len = 0;
7190 :
7191 0 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
7192 :
7193 0 : bdev_io_submit(bdev_io);
7194 0 : return 0;
7195 : }
7196 :
7197 : int
7198 1 : spdk_bdev_nvme_io_passthru_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
7199 : const struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes, void *md_buf, size_t md_len,
7200 : spdk_bdev_io_completion_cb cb, void *cb_arg)
7201 : {
7202 1 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
7203 : struct spdk_bdev_io *bdev_io;
7204 1 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
7205 :
7206 1 : if (!desc->write) {
7207 : /*
7208 : * Do not try to parse the NVMe command - we could maybe use bits in the opcode
7209 : * to easily determine if the command is a read or write, but for now just
7210 : * do not allow io_passthru with a read-only descriptor.
7211 : */
7212 0 : return -EBADF;
7213 : }
7214 :
7215 1 : if (spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_IO_MD))) {
7216 1 : return -ENOTSUP;
7217 : }
7218 :
7219 0 : bdev_io = bdev_channel_get_io(channel);
7220 0 : if (!bdev_io) {
7221 0 : return -ENOMEM;
7222 : }
7223 :
7224 0 : bdev_io->internal.ch = channel;
7225 0 : bdev_io->internal.desc = desc;
7226 0 : bdev_io->type = SPDK_BDEV_IO_TYPE_NVME_IO_MD;
7227 0 : bdev_io->u.nvme_passthru.cmd = *cmd;
7228 0 : bdev_io->u.nvme_passthru.buf = buf;
7229 0 : bdev_io->u.nvme_passthru.nbytes = nbytes;
7230 0 : bdev_io->u.nvme_passthru.md_buf = md_buf;
7231 0 : bdev_io->u.nvme_passthru.md_len = md_len;
7232 :
7233 0 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
7234 :
7235 0 : bdev_io_submit(bdev_io);
7236 0 : return 0;
7237 : }
7238 :
7239 : int
7240 0 : spdk_bdev_nvme_iov_passthru_md(struct spdk_bdev_desc *desc,
7241 : struct spdk_io_channel *ch,
7242 : const struct spdk_nvme_cmd *cmd,
7243 : struct iovec *iov, int iovcnt, size_t nbytes,
7244 : void *md_buf, size_t md_len,
7245 : spdk_bdev_io_completion_cb cb, void *cb_arg)
7246 : {
7247 0 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
7248 : struct spdk_bdev_io *bdev_io;
7249 0 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
7250 :
7251 0 : if (!desc->write) {
7252 : /*
7253 : * Do not try to parse the NVMe command - we could maybe use bits in the opcode
7254 : * to easily determine if the command is a read or write, but for now just
7255 : * do not allow io_passthru with a read-only descriptor.
7256 : */
7257 0 : return -EBADF;
7258 : }
7259 :
7260 0 : if (md_buf && spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_IO_MD))) {
7261 0 : return -ENOTSUP;
7262 0 : } else if (spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_IO))) {
7263 0 : return -ENOTSUP;
7264 : }
7265 :
7266 0 : bdev_io = bdev_channel_get_io(channel);
7267 0 : if (!bdev_io) {
7268 0 : return -ENOMEM;
7269 : }
7270 :
7271 0 : bdev_io->internal.ch = channel;
7272 0 : bdev_io->internal.desc = desc;
7273 0 : bdev_io->type = SPDK_BDEV_IO_TYPE_NVME_IOV_MD;
7274 0 : bdev_io->u.nvme_passthru.cmd = *cmd;
7275 0 : bdev_io->u.nvme_passthru.iovs = iov;
7276 0 : bdev_io->u.nvme_passthru.iovcnt = iovcnt;
7277 0 : bdev_io->u.nvme_passthru.nbytes = nbytes;
7278 0 : bdev_io->u.nvme_passthru.md_buf = md_buf;
7279 0 : bdev_io->u.nvme_passthru.md_len = md_len;
7280 :
7281 0 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
7282 :
7283 0 : bdev_io_submit(bdev_io);
7284 0 : return 0;
7285 : }
7286 :
7287 : static void bdev_abort_retry(void *ctx);
7288 : static void bdev_abort(struct spdk_bdev_io *parent_io);
7289 :
7290 : static void
7291 22 : bdev_abort_io_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
7292 : {
7293 22 : struct spdk_bdev_channel *channel = bdev_io->internal.ch;
7294 22 : struct spdk_bdev_io *parent_io = cb_arg;
7295 : struct spdk_bdev_io *bio_to_abort, *tmp_io;
7296 :
7297 22 : bio_to_abort = bdev_io->u.abort.bio_to_abort;
7298 :
7299 22 : spdk_bdev_free_io(bdev_io);
7300 :
7301 22 : if (!success) {
7302 : /* Check if the target I/O completed in the meantime. */
7303 2 : TAILQ_FOREACH(tmp_io, &channel->io_submitted, internal.ch_link) {
7304 1 : if (tmp_io == bio_to_abort) {
7305 0 : break;
7306 : }
7307 : }
7308 :
7309 : /* If the target I/O still exists, set the parent to failed. */
7310 1 : if (tmp_io != NULL) {
7311 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
7312 : }
7313 : }
7314 :
7315 22 : assert(parent_io->internal.f.split);
7316 :
7317 22 : parent_io->internal.split.outstanding--;
7318 22 : if (parent_io->internal.split.outstanding == 0) {
7319 16 : if (parent_io->internal.status == SPDK_BDEV_IO_STATUS_NOMEM) {
7320 0 : bdev_abort_retry(parent_io);
7321 : } else {
7322 16 : bdev_io_complete(parent_io);
7323 : }
7324 : }
7325 22 : }
7326 :
7327 : static int
7328 23 : bdev_abort_io(struct spdk_bdev_desc *desc, struct spdk_bdev_channel *channel,
7329 : struct spdk_bdev_io *bio_to_abort,
7330 : spdk_bdev_io_completion_cb cb, void *cb_arg)
7331 : {
7332 23 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
7333 : struct spdk_bdev_io *bdev_io;
7334 :
7335 23 : if (bio_to_abort->type == SPDK_BDEV_IO_TYPE_ABORT ||
7336 23 : bio_to_abort->type == SPDK_BDEV_IO_TYPE_RESET) {
7337 : /* TODO: Abort reset or abort request. */
7338 0 : return -ENOTSUP;
7339 : }
7340 :
7341 23 : bdev_io = bdev_channel_get_io(channel);
7342 23 : if (bdev_io == NULL) {
7343 1 : return -ENOMEM;
7344 : }
7345 :
7346 22 : bdev_io->internal.ch = channel;
7347 22 : bdev_io->internal.desc = desc;
7348 22 : bdev_io->type = SPDK_BDEV_IO_TYPE_ABORT;
7349 22 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
7350 :
7351 22 : if (bio_to_abort->internal.f.split) {
7352 6 : assert(bdev_io_should_split(bio_to_abort));
7353 6 : bdev_io->u.bdev.abort.bio_cb_arg = bio_to_abort;
7354 :
7355 : /* Parent abort request is not submitted directly, but to manage its
7356 : * execution add it to the submitted list here.
7357 : */
7358 6 : bdev_io->internal.submit_tsc = spdk_get_ticks();
7359 6 : bdev_ch_add_to_io_submitted(bdev_io);
7360 :
7361 6 : bdev_abort(bdev_io);
7362 :
7363 6 : return 0;
7364 : }
7365 :
7366 16 : bdev_io->u.abort.bio_to_abort = bio_to_abort;
7367 :
7368 : /* Submit the abort request to the underlying bdev module. */
7369 16 : bdev_io_submit(bdev_io);
7370 :
7371 16 : return 0;
7372 : }
7373 :
7374 : static bool
7375 46 : bdev_io_on_tailq(struct spdk_bdev_io *bdev_io, bdev_io_tailq_t *tailq)
7376 : {
7377 : struct spdk_bdev_io *iter;
7378 :
7379 46 : TAILQ_FOREACH(iter, tailq, internal.link) {
7380 0 : if (iter == bdev_io) {
7381 0 : return true;
7382 : }
7383 : }
7384 :
7385 46 : return false;
7386 : }
7387 :
7388 : static uint32_t
7389 18 : _bdev_abort(struct spdk_bdev_io *parent_io)
7390 : {
7391 18 : struct spdk_bdev_desc *desc = parent_io->internal.desc;
7392 18 : struct spdk_bdev_channel *channel = parent_io->internal.ch;
7393 : void *bio_cb_arg;
7394 : struct spdk_bdev_io *bio_to_abort;
7395 : uint32_t matched_ios;
7396 : int rc;
7397 :
7398 18 : bio_cb_arg = parent_io->u.bdev.abort.bio_cb_arg;
7399 :
7400 : /* matched_ios is returned and will be kept by the caller.
7401 : *
7402 : * This function will be used for two cases, 1) the same cb_arg is used for
7403 : * multiple I/Os, 2) a single large I/O is split into smaller ones.
7404 : * Incrementing split_outstanding directly here may confuse readers especially
7405 : * for the 1st case.
7406 : *
7407 : * Completion of I/O abort is processed after stack unwinding. Hence this trick
7408 : * works as expected.
7409 : */
7410 18 : matched_ios = 0;
7411 18 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
7412 :
7413 105 : TAILQ_FOREACH(bio_to_abort, &channel->io_submitted, internal.ch_link) {
7414 88 : if (bio_to_abort->internal.caller_ctx != bio_cb_arg) {
7415 65 : continue;
7416 : }
7417 :
7418 23 : if (bio_to_abort->internal.submit_tsc > parent_io->internal.submit_tsc) {
7419 : /* Any I/O which was submitted after this abort command should be excluded. */
7420 0 : continue;
7421 : }
7422 :
7423 : /* We can't abort a request that's being pushed/pulled or executed by accel */
7424 46 : if (bdev_io_on_tailq(bio_to_abort, &channel->io_accel_exec) ||
7425 23 : bdev_io_on_tailq(bio_to_abort, &channel->io_memory_domain)) {
7426 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
7427 0 : break;
7428 : }
7429 :
7430 23 : rc = bdev_abort_io(desc, channel, bio_to_abort, bdev_abort_io_done, parent_io);
7431 23 : if (rc != 0) {
7432 1 : if (rc == -ENOMEM) {
7433 1 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_NOMEM;
7434 : } else {
7435 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
7436 : }
7437 1 : break;
7438 : }
7439 22 : matched_ios++;
7440 : }
7441 :
7442 18 : return matched_ios;
7443 : }
7444 :
7445 : static void
7446 1 : bdev_abort_retry(void *ctx)
7447 : {
7448 1 : struct spdk_bdev_io *parent_io = ctx;
7449 : uint32_t matched_ios;
7450 :
7451 1 : matched_ios = _bdev_abort(parent_io);
7452 :
7453 1 : if (matched_ios == 0) {
7454 0 : if (parent_io->internal.status == SPDK_BDEV_IO_STATUS_NOMEM) {
7455 0 : bdev_queue_io_wait_with_cb(parent_io, bdev_abort_retry);
7456 : } else {
7457 : /* For retry, the case that no target I/O was found is success
7458 : * because it means target I/Os completed in the meantime.
7459 : */
7460 0 : bdev_io_complete(parent_io);
7461 : }
7462 0 : return;
7463 : }
7464 :
7465 : /* Use split_outstanding to manage the progress of aborting I/Os. */
7466 1 : parent_io->internal.f.split = true;
7467 1 : parent_io->internal.split.outstanding = matched_ios;
7468 : }
7469 :
7470 : static void
7471 17 : bdev_abort(struct spdk_bdev_io *parent_io)
7472 : {
7473 : uint32_t matched_ios;
7474 :
7475 17 : matched_ios = _bdev_abort(parent_io);
7476 :
7477 17 : if (matched_ios == 0) {
7478 2 : if (parent_io->internal.status == SPDK_BDEV_IO_STATUS_NOMEM) {
7479 1 : bdev_queue_io_wait_with_cb(parent_io, bdev_abort_retry);
7480 : } else {
7481 : /* The case the no target I/O was found is failure. */
7482 1 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
7483 1 : bdev_io_complete(parent_io);
7484 : }
7485 2 : return;
7486 : }
7487 :
7488 : /* Use split_outstanding to manage the progress of aborting I/Os. */
7489 15 : parent_io->internal.f.split = true;
7490 15 : parent_io->internal.split.outstanding = matched_ios;
7491 : }
7492 :
7493 : int
7494 12 : spdk_bdev_abort(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
7495 : void *bio_cb_arg,
7496 : spdk_bdev_io_completion_cb cb, void *cb_arg)
7497 : {
7498 12 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
7499 12 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
7500 : struct spdk_bdev_io *bdev_io;
7501 :
7502 12 : if (bio_cb_arg == NULL) {
7503 0 : return -EINVAL;
7504 : }
7505 :
7506 12 : if (!spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_ABORT)) {
7507 1 : return -ENOTSUP;
7508 : }
7509 :
7510 11 : bdev_io = bdev_channel_get_io(channel);
7511 11 : if (bdev_io == NULL) {
7512 0 : return -ENOMEM;
7513 : }
7514 :
7515 11 : bdev_io->internal.ch = channel;
7516 11 : bdev_io->internal.desc = desc;
7517 11 : bdev_io->internal.submit_tsc = spdk_get_ticks();
7518 11 : bdev_io->type = SPDK_BDEV_IO_TYPE_ABORT;
7519 11 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
7520 :
7521 11 : bdev_io->u.bdev.abort.bio_cb_arg = bio_cb_arg;
7522 :
7523 : /* Parent abort request is not submitted directly, but to manage its execution,
7524 : * add it to the submitted list here.
7525 : */
7526 11 : bdev_ch_add_to_io_submitted(bdev_io);
7527 :
7528 11 : bdev_abort(bdev_io);
7529 :
7530 11 : return 0;
7531 : }
7532 :
7533 : int
7534 4 : spdk_bdev_queue_io_wait(struct spdk_bdev *bdev, struct spdk_io_channel *ch,
7535 : struct spdk_bdev_io_wait_entry *entry)
7536 : {
7537 4 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
7538 4 : struct spdk_bdev_mgmt_channel *mgmt_ch = channel->shared_resource->mgmt_ch;
7539 :
7540 4 : if (bdev != entry->bdev) {
7541 0 : SPDK_ERRLOG("bdevs do not match\n");
7542 0 : return -EINVAL;
7543 : }
7544 :
7545 4 : if (mgmt_ch->per_thread_cache_count > 0) {
7546 0 : SPDK_ERRLOG("Cannot queue io_wait if spdk_bdev_io available in per-thread cache\n");
7547 0 : return -EINVAL;
7548 : }
7549 :
7550 4 : TAILQ_INSERT_TAIL(&mgmt_ch->io_wait_queue, entry, link);
7551 4 : return 0;
7552 : }
7553 :
7554 : static inline void
7555 612 : bdev_io_update_io_stat(struct spdk_bdev_io *bdev_io, uint64_t tsc_diff)
7556 : {
7557 612 : enum spdk_bdev_io_status io_status = bdev_io->internal.status;
7558 612 : struct spdk_bdev_io_stat *io_stat = bdev_io->internal.ch->stat;
7559 612 : uint64_t num_blocks = bdev_io->u.bdev.num_blocks;
7560 612 : uint32_t blocklen = bdev_io->bdev->blocklen;
7561 :
7562 612 : if (spdk_likely(io_status == SPDK_BDEV_IO_STATUS_SUCCESS)) {
7563 519 : switch (bdev_io->type) {
7564 321 : case SPDK_BDEV_IO_TYPE_READ:
7565 321 : io_stat->bytes_read += num_blocks * blocklen;
7566 321 : io_stat->num_read_ops++;
7567 321 : io_stat->read_latency_ticks += tsc_diff;
7568 321 : if (io_stat->max_read_latency_ticks < tsc_diff) {
7569 7 : io_stat->max_read_latency_ticks = tsc_diff;
7570 : }
7571 321 : if (io_stat->min_read_latency_ticks > tsc_diff) {
7572 42 : io_stat->min_read_latency_ticks = tsc_diff;
7573 : }
7574 321 : break;
7575 75 : case SPDK_BDEV_IO_TYPE_WRITE:
7576 75 : io_stat->bytes_written += num_blocks * blocklen;
7577 75 : io_stat->num_write_ops++;
7578 75 : io_stat->write_latency_ticks += tsc_diff;
7579 75 : if (io_stat->max_write_latency_ticks < tsc_diff) {
7580 4 : io_stat->max_write_latency_ticks = tsc_diff;
7581 : }
7582 75 : if (io_stat->min_write_latency_ticks > tsc_diff) {
7583 25 : io_stat->min_write_latency_ticks = tsc_diff;
7584 : }
7585 75 : break;
7586 20 : case SPDK_BDEV_IO_TYPE_UNMAP:
7587 20 : io_stat->bytes_unmapped += num_blocks * blocklen;
7588 20 : io_stat->num_unmap_ops++;
7589 20 : io_stat->unmap_latency_ticks += tsc_diff;
7590 20 : if (io_stat->max_unmap_latency_ticks < tsc_diff) {
7591 0 : io_stat->max_unmap_latency_ticks = tsc_diff;
7592 : }
7593 20 : if (io_stat->min_unmap_latency_ticks > tsc_diff) {
7594 3 : io_stat->min_unmap_latency_ticks = tsc_diff;
7595 : }
7596 20 : break;
7597 4 : case SPDK_BDEV_IO_TYPE_ZCOPY:
7598 : /* Track the data in the start phase only */
7599 4 : if (bdev_io->u.bdev.zcopy.start) {
7600 2 : if (bdev_io->u.bdev.zcopy.populate) {
7601 1 : io_stat->bytes_read += num_blocks * blocklen;
7602 1 : io_stat->num_read_ops++;
7603 1 : io_stat->read_latency_ticks += tsc_diff;
7604 1 : if (io_stat->max_read_latency_ticks < tsc_diff) {
7605 0 : io_stat->max_read_latency_ticks = tsc_diff;
7606 : }
7607 1 : if (io_stat->min_read_latency_ticks > tsc_diff) {
7608 1 : io_stat->min_read_latency_ticks = tsc_diff;
7609 : }
7610 : } else {
7611 1 : io_stat->bytes_written += num_blocks * blocklen;
7612 1 : io_stat->num_write_ops++;
7613 1 : io_stat->write_latency_ticks += tsc_diff;
7614 1 : if (io_stat->max_write_latency_ticks < tsc_diff) {
7615 0 : io_stat->max_write_latency_ticks = tsc_diff;
7616 : }
7617 1 : if (io_stat->min_write_latency_ticks > tsc_diff) {
7618 1 : io_stat->min_write_latency_ticks = tsc_diff;
7619 : }
7620 : }
7621 : }
7622 4 : break;
7623 21 : case SPDK_BDEV_IO_TYPE_COPY:
7624 21 : io_stat->bytes_copied += num_blocks * blocklen;
7625 21 : io_stat->num_copy_ops++;
7626 21 : bdev_io->internal.ch->stat->copy_latency_ticks += tsc_diff;
7627 21 : if (io_stat->max_copy_latency_ticks < tsc_diff) {
7628 0 : io_stat->max_copy_latency_ticks = tsc_diff;
7629 : }
7630 21 : if (io_stat->min_copy_latency_ticks > tsc_diff) {
7631 4 : io_stat->min_copy_latency_ticks = tsc_diff;
7632 : }
7633 21 : break;
7634 78 : default:
7635 78 : break;
7636 : }
7637 93 : } else if (io_status <= SPDK_BDEV_IO_STATUS_FAILED && io_status >= SPDK_MIN_BDEV_IO_STATUS) {
7638 93 : io_stat = bdev_io->bdev->internal.stat;
7639 93 : assert(io_stat->io_error != NULL);
7640 :
7641 93 : spdk_spin_lock(&bdev_io->bdev->internal.spinlock);
7642 93 : io_stat->io_error->error_status[-io_status - 1]++;
7643 93 : spdk_spin_unlock(&bdev_io->bdev->internal.spinlock);
7644 : }
7645 :
7646 : #ifdef SPDK_CONFIG_VTUNE
7647 : uint64_t now_tsc = spdk_get_ticks();
7648 : if (now_tsc > (bdev_io->internal.ch->start_tsc + bdev_io->internal.ch->interval_tsc)) {
7649 : uint64_t data[5];
7650 : struct spdk_bdev_io_stat *prev_stat = bdev_io->internal.ch->prev_stat;
7651 :
7652 : data[0] = io_stat->num_read_ops - prev_stat->num_read_ops;
7653 : data[1] = io_stat->bytes_read - prev_stat->bytes_read;
7654 : data[2] = io_stat->num_write_ops - prev_stat->num_write_ops;
7655 : data[3] = io_stat->bytes_written - prev_stat->bytes_written;
7656 : data[4] = bdev_io->bdev->fn_table->get_spin_time ?
7657 : bdev_io->bdev->fn_table->get_spin_time(spdk_bdev_io_get_io_channel(bdev_io)) : 0;
7658 :
7659 : __itt_metadata_add(g_bdev_mgr.domain, __itt_null, bdev_io->internal.ch->handle,
7660 : __itt_metadata_u64, 5, data);
7661 :
7662 : memcpy(prev_stat, io_stat, sizeof(struct spdk_bdev_io_stat));
7663 : bdev_io->internal.ch->start_tsc = now_tsc;
7664 : }
7665 : #endif
7666 612 : }
7667 :
7668 : static inline void
7669 612 : _bdev_io_complete(void *ctx)
7670 : {
7671 612 : struct spdk_bdev_io *bdev_io = ctx;
7672 :
7673 612 : if (spdk_unlikely(bdev_io_use_accel_sequence(bdev_io))) {
7674 0 : assert(bdev_io->internal.status != SPDK_BDEV_IO_STATUS_SUCCESS);
7675 0 : spdk_accel_sequence_abort(bdev_io->internal.accel_sequence);
7676 : }
7677 :
7678 612 : assert(bdev_io->internal.cb != NULL);
7679 612 : assert(spdk_get_thread() == spdk_bdev_io_get_thread(bdev_io));
7680 :
7681 612 : bdev_io->internal.cb(bdev_io, bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS,
7682 : bdev_io->internal.caller_ctx);
7683 612 : }
7684 :
7685 : static inline void
7686 620 : bdev_io_complete(void *ctx)
7687 : {
7688 620 : struct spdk_bdev_io *bdev_io = ctx;
7689 620 : struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
7690 : uint64_t tsc, tsc_diff;
7691 :
7692 620 : if (spdk_unlikely(bdev_io->internal.f.in_submit_request)) {
7693 : /*
7694 : * Defer completion to avoid potential infinite recursion if the
7695 : * user's completion callback issues a new I/O.
7696 : */
7697 8 : spdk_thread_send_msg(spdk_bdev_io_get_thread(bdev_io),
7698 : bdev_io_complete, bdev_io);
7699 8 : return;
7700 : }
7701 :
7702 612 : tsc = spdk_get_ticks();
7703 612 : tsc_diff = tsc - bdev_io->internal.submit_tsc;
7704 :
7705 612 : bdev_ch_remove_from_io_submitted(bdev_io);
7706 612 : spdk_trace_record_tsc(tsc, TRACE_BDEV_IO_DONE, bdev_ch->trace_id, 0, (uintptr_t)bdev_io,
7707 : bdev_io->internal.caller_ctx, bdev_ch->queue_depth);
7708 :
7709 612 : if (bdev_ch->histogram) {
7710 4 : if (bdev_io->bdev->internal.histogram_io_type == 0 ||
7711 0 : bdev_io->bdev->internal.histogram_io_type == bdev_io->type) {
7712 : /*
7713 : * Tally all I/O types if the histogram_io_type is set to 0.
7714 : */
7715 4 : spdk_histogram_data_tally(bdev_ch->histogram, tsc_diff);
7716 : }
7717 : }
7718 :
7719 612 : bdev_io_update_io_stat(bdev_io, tsc_diff);
7720 612 : _bdev_io_complete(bdev_io);
7721 : }
7722 :
7723 : /* The difference between this function and bdev_io_complete() is that this should be called to
7724 : * complete IOs that haven't been submitted via bdev_io_submit(), as they weren't added onto the
7725 : * io_submitted list and don't have submit_tsc updated.
7726 : */
7727 : static inline void
7728 0 : bdev_io_complete_unsubmitted(struct spdk_bdev_io *bdev_io)
7729 : {
7730 : /* Since the IO hasn't been submitted it's bound to be failed */
7731 0 : assert(bdev_io->internal.status != SPDK_BDEV_IO_STATUS_SUCCESS);
7732 :
7733 : /* At this point we don't know if the IO is completed from submission context or not, but,
7734 : * since this is an error path, we can always do an spdk_thread_send_msg(). */
7735 0 : spdk_thread_send_msg(spdk_bdev_io_get_thread(bdev_io),
7736 : _bdev_io_complete, bdev_io);
7737 0 : }
7738 :
7739 : static void bdev_destroy_cb(void *io_device);
7740 :
7741 : static inline void
7742 18 : _bdev_reset_complete(void *ctx)
7743 : {
7744 18 : struct spdk_bdev_io *bdev_io = ctx;
7745 :
7746 : /* Put the channel reference we got in submission. */
7747 18 : assert(bdev_io->u.reset.ch_ref != NULL);
7748 18 : spdk_put_io_channel(bdev_io->u.reset.ch_ref);
7749 18 : bdev_io->u.reset.ch_ref = NULL;
7750 :
7751 18 : bdev_io_complete(bdev_io);
7752 18 : }
7753 :
7754 : static void
7755 16 : bdev_reset_complete(struct spdk_bdev *bdev, void *_ctx, int status)
7756 : {
7757 16 : struct spdk_bdev_io *bdev_io = _ctx;
7758 16 : bdev_io_tailq_t queued_resets;
7759 : struct spdk_bdev_io *queued_reset;
7760 :
7761 16 : assert(bdev_io == bdev->internal.reset_in_progress);
7762 :
7763 16 : TAILQ_INIT(&queued_resets);
7764 :
7765 16 : spdk_spin_lock(&bdev->internal.spinlock);
7766 16 : TAILQ_SWAP(&bdev->internal.queued_resets, &queued_resets,
7767 : spdk_bdev_io, internal.link);
7768 16 : bdev->internal.reset_in_progress = NULL;
7769 16 : spdk_spin_unlock(&bdev->internal.spinlock);
7770 :
7771 18 : while (!TAILQ_EMPTY(&queued_resets)) {
7772 2 : queued_reset = TAILQ_FIRST(&queued_resets);
7773 2 : TAILQ_REMOVE(&queued_resets, queued_reset, internal.link);
7774 2 : queued_reset->internal.status = bdev_io->internal.status;
7775 2 : spdk_thread_send_msg(spdk_bdev_io_get_thread(queued_reset),
7776 : _bdev_reset_complete, queued_reset);
7777 : }
7778 :
7779 16 : _bdev_reset_complete(bdev_io);
7780 :
7781 16 : if (bdev->internal.status == SPDK_BDEV_STATUS_REMOVING &&
7782 1 : TAILQ_EMPTY(&bdev->internal.open_descs)) {
7783 1 : spdk_io_device_unregister(__bdev_to_io_dev(bdev), bdev_destroy_cb);
7784 : }
7785 16 : }
7786 :
7787 : static void
7788 20 : bdev_unfreeze_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
7789 : struct spdk_io_channel *_ch, void *_ctx)
7790 : {
7791 20 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
7792 :
7793 20 : ch->flags &= ~BDEV_CH_RESET_IN_PROGRESS;
7794 :
7795 20 : spdk_bdev_for_each_channel_continue(i, 0);
7796 20 : }
7797 :
7798 : static void
7799 0 : bdev_io_complete_sequence_cb(void *ctx, int status)
7800 : {
7801 0 : struct spdk_bdev_io *bdev_io = ctx;
7802 :
7803 : /* u.bdev.accel_sequence should have already been cleared at this point */
7804 0 : assert(bdev_io->u.bdev.accel_sequence == NULL);
7805 0 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS);
7806 0 : bdev_io->internal.f.has_accel_sequence = false;
7807 :
7808 0 : if (spdk_unlikely(status != 0)) {
7809 0 : SPDK_ERRLOG("Failed to execute accel sequence, status=%d\n", status);
7810 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
7811 : }
7812 :
7813 0 : bdev_io_complete(bdev_io);
7814 0 : }
7815 :
7816 : void
7817 598 : spdk_bdev_io_complete(struct spdk_bdev_io *bdev_io, enum spdk_bdev_io_status status)
7818 : {
7819 598 : struct spdk_bdev *bdev = bdev_io->bdev;
7820 598 : struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
7821 598 : struct spdk_bdev_shared_resource *shared_resource = bdev_ch->shared_resource;
7822 :
7823 598 : if (spdk_unlikely(bdev_io->internal.status != SPDK_BDEV_IO_STATUS_PENDING)) {
7824 0 : SPDK_ERRLOG("Unexpected completion on IO from %s module, status was %s\n",
7825 : spdk_bdev_get_module_name(bdev),
7826 : bdev_io_status_get_string(bdev_io->internal.status));
7827 0 : assert(false);
7828 : }
7829 598 : bdev_io->internal.status = status;
7830 :
7831 598 : if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_RESET)) {
7832 16 : assert(bdev_io == bdev->internal.reset_in_progress);
7833 16 : spdk_bdev_for_each_channel(bdev, bdev_unfreeze_channel, bdev_io,
7834 : bdev_reset_complete);
7835 16 : return;
7836 : } else {
7837 582 : bdev_io_decrement_outstanding(bdev_ch, shared_resource);
7838 582 : if (spdk_likely(status == SPDK_BDEV_IO_STATUS_SUCCESS)) {
7839 485 : if (bdev_io_needs_sequence_exec(bdev_io->internal.desc, bdev_io)) {
7840 0 : bdev_io_exec_sequence(bdev_io, bdev_io_complete_sequence_cb);
7841 0 : return;
7842 485 : } else if (spdk_unlikely(bdev_io->internal.f.has_bounce_buf &&
7843 : !bdev_io_use_accel_sequence(bdev_io))) {
7844 26 : _bdev_io_push_bounce_data_buffer(bdev_io,
7845 : _bdev_io_complete_push_bounce_done);
7846 : /* bdev IO will be completed in the callback */
7847 26 : return;
7848 : }
7849 : }
7850 :
7851 556 : if (spdk_unlikely(_bdev_io_handle_no_mem(bdev_io, BDEV_IO_RETRY_STATE_SUBMIT))) {
7852 5 : return;
7853 : }
7854 : }
7855 :
7856 551 : bdev_io_complete(bdev_io);
7857 : }
7858 :
7859 : void
7860 0 : spdk_bdev_io_complete_scsi_status(struct spdk_bdev_io *bdev_io, enum spdk_scsi_status sc,
7861 : enum spdk_scsi_sense sk, uint8_t asc, uint8_t ascq)
7862 : {
7863 : enum spdk_bdev_io_status status;
7864 :
7865 0 : if (sc == SPDK_SCSI_STATUS_GOOD) {
7866 0 : status = SPDK_BDEV_IO_STATUS_SUCCESS;
7867 : } else {
7868 0 : status = SPDK_BDEV_IO_STATUS_SCSI_ERROR;
7869 0 : bdev_io->internal.error.scsi.sc = sc;
7870 0 : bdev_io->internal.error.scsi.sk = sk;
7871 0 : bdev_io->internal.error.scsi.asc = asc;
7872 0 : bdev_io->internal.error.scsi.ascq = ascq;
7873 : }
7874 :
7875 0 : spdk_bdev_io_complete(bdev_io, status);
7876 0 : }
7877 :
7878 : void
7879 0 : spdk_bdev_io_get_scsi_status(const struct spdk_bdev_io *bdev_io,
7880 : int *sc, int *sk, int *asc, int *ascq)
7881 : {
7882 0 : assert(sc != NULL);
7883 0 : assert(sk != NULL);
7884 0 : assert(asc != NULL);
7885 0 : assert(ascq != NULL);
7886 :
7887 0 : switch (bdev_io->internal.status) {
7888 0 : case SPDK_BDEV_IO_STATUS_SUCCESS:
7889 0 : *sc = SPDK_SCSI_STATUS_GOOD;
7890 0 : *sk = SPDK_SCSI_SENSE_NO_SENSE;
7891 0 : *asc = SPDK_SCSI_ASC_NO_ADDITIONAL_SENSE;
7892 0 : *ascq = SPDK_SCSI_ASCQ_CAUSE_NOT_REPORTABLE;
7893 0 : break;
7894 0 : case SPDK_BDEV_IO_STATUS_NVME_ERROR:
7895 0 : spdk_scsi_nvme_translate(bdev_io, sc, sk, asc, ascq);
7896 0 : break;
7897 0 : case SPDK_BDEV_IO_STATUS_MISCOMPARE:
7898 0 : *sc = SPDK_SCSI_STATUS_CHECK_CONDITION;
7899 0 : *sk = SPDK_SCSI_SENSE_MISCOMPARE;
7900 0 : *asc = SPDK_SCSI_ASC_MISCOMPARE_DURING_VERIFY_OPERATION;
7901 0 : *ascq = bdev_io->internal.error.scsi.ascq;
7902 0 : break;
7903 0 : case SPDK_BDEV_IO_STATUS_SCSI_ERROR:
7904 0 : *sc = bdev_io->internal.error.scsi.sc;
7905 0 : *sk = bdev_io->internal.error.scsi.sk;
7906 0 : *asc = bdev_io->internal.error.scsi.asc;
7907 0 : *ascq = bdev_io->internal.error.scsi.ascq;
7908 0 : break;
7909 0 : default:
7910 0 : *sc = SPDK_SCSI_STATUS_CHECK_CONDITION;
7911 0 : *sk = SPDK_SCSI_SENSE_ABORTED_COMMAND;
7912 0 : *asc = SPDK_SCSI_ASC_NO_ADDITIONAL_SENSE;
7913 0 : *ascq = SPDK_SCSI_ASCQ_CAUSE_NOT_REPORTABLE;
7914 0 : break;
7915 : }
7916 0 : }
7917 :
7918 : void
7919 0 : spdk_bdev_io_complete_aio_status(struct spdk_bdev_io *bdev_io, int aio_result)
7920 : {
7921 : enum spdk_bdev_io_status status;
7922 :
7923 0 : if (aio_result == 0) {
7924 0 : status = SPDK_BDEV_IO_STATUS_SUCCESS;
7925 : } else {
7926 0 : status = SPDK_BDEV_IO_STATUS_AIO_ERROR;
7927 : }
7928 :
7929 0 : bdev_io->internal.error.aio_result = aio_result;
7930 :
7931 0 : spdk_bdev_io_complete(bdev_io, status);
7932 0 : }
7933 :
7934 : void
7935 0 : spdk_bdev_io_get_aio_status(const struct spdk_bdev_io *bdev_io, int *aio_result)
7936 : {
7937 0 : assert(aio_result != NULL);
7938 :
7939 0 : if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_AIO_ERROR) {
7940 0 : *aio_result = bdev_io->internal.error.aio_result;
7941 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS) {
7942 0 : *aio_result = 0;
7943 : } else {
7944 0 : *aio_result = -EIO;
7945 : }
7946 0 : }
7947 :
7948 : void
7949 0 : spdk_bdev_io_complete_nvme_status(struct spdk_bdev_io *bdev_io, uint32_t cdw0, int sct, int sc)
7950 : {
7951 : enum spdk_bdev_io_status status;
7952 :
7953 0 : if (spdk_likely(sct == SPDK_NVME_SCT_GENERIC && sc == SPDK_NVME_SC_SUCCESS)) {
7954 0 : status = SPDK_BDEV_IO_STATUS_SUCCESS;
7955 0 : } else if (sct == SPDK_NVME_SCT_GENERIC && sc == SPDK_NVME_SC_ABORTED_BY_REQUEST) {
7956 0 : status = SPDK_BDEV_IO_STATUS_ABORTED;
7957 : } else {
7958 0 : status = SPDK_BDEV_IO_STATUS_NVME_ERROR;
7959 : }
7960 :
7961 0 : bdev_io->internal.error.nvme.cdw0 = cdw0;
7962 0 : bdev_io->internal.error.nvme.sct = sct;
7963 0 : bdev_io->internal.error.nvme.sc = sc;
7964 :
7965 0 : spdk_bdev_io_complete(bdev_io, status);
7966 0 : }
7967 :
7968 : void
7969 0 : spdk_bdev_io_get_nvme_status(const struct spdk_bdev_io *bdev_io, uint32_t *cdw0, int *sct, int *sc)
7970 : {
7971 0 : assert(sct != NULL);
7972 0 : assert(sc != NULL);
7973 0 : assert(cdw0 != NULL);
7974 :
7975 0 : if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_ABORT)) {
7976 0 : *sct = SPDK_NVME_SCT_GENERIC;
7977 0 : *sc = SPDK_NVME_SC_SUCCESS;
7978 0 : if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS) {
7979 0 : *cdw0 = 0;
7980 : } else {
7981 0 : *cdw0 = 1U;
7982 : }
7983 0 : return;
7984 : }
7985 :
7986 0 : if (spdk_likely(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS)) {
7987 0 : *sct = SPDK_NVME_SCT_GENERIC;
7988 0 : *sc = SPDK_NVME_SC_SUCCESS;
7989 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_NVME_ERROR) {
7990 0 : *sct = bdev_io->internal.error.nvme.sct;
7991 0 : *sc = bdev_io->internal.error.nvme.sc;
7992 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_ABORTED) {
7993 0 : *sct = SPDK_NVME_SCT_GENERIC;
7994 0 : *sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
7995 : } else {
7996 0 : *sct = SPDK_NVME_SCT_GENERIC;
7997 0 : *sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
7998 : }
7999 :
8000 0 : *cdw0 = bdev_io->internal.error.nvme.cdw0;
8001 : }
8002 :
8003 : void
8004 0 : spdk_bdev_io_get_nvme_fused_status(const struct spdk_bdev_io *bdev_io, uint32_t *cdw0,
8005 : int *first_sct, int *first_sc, int *second_sct, int *second_sc)
8006 : {
8007 0 : assert(first_sct != NULL);
8008 0 : assert(first_sc != NULL);
8009 0 : assert(second_sct != NULL);
8010 0 : assert(second_sc != NULL);
8011 0 : assert(cdw0 != NULL);
8012 :
8013 0 : if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_NVME_ERROR) {
8014 0 : if (bdev_io->internal.error.nvme.sct == SPDK_NVME_SCT_MEDIA_ERROR &&
8015 0 : bdev_io->internal.error.nvme.sc == SPDK_NVME_SC_COMPARE_FAILURE) {
8016 0 : *first_sct = bdev_io->internal.error.nvme.sct;
8017 0 : *first_sc = bdev_io->internal.error.nvme.sc;
8018 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
8019 0 : *second_sc = SPDK_NVME_SC_ABORTED_FAILED_FUSED;
8020 : } else {
8021 0 : *first_sct = SPDK_NVME_SCT_GENERIC;
8022 0 : *first_sc = SPDK_NVME_SC_SUCCESS;
8023 0 : *second_sct = bdev_io->internal.error.nvme.sct;
8024 0 : *second_sc = bdev_io->internal.error.nvme.sc;
8025 : }
8026 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_ABORTED) {
8027 0 : *first_sct = SPDK_NVME_SCT_GENERIC;
8028 0 : *first_sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
8029 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
8030 0 : *second_sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
8031 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS) {
8032 0 : *first_sct = SPDK_NVME_SCT_GENERIC;
8033 0 : *first_sc = SPDK_NVME_SC_SUCCESS;
8034 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
8035 0 : *second_sc = SPDK_NVME_SC_SUCCESS;
8036 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_FIRST_FUSED_FAILED) {
8037 0 : *first_sct = SPDK_NVME_SCT_GENERIC;
8038 0 : *first_sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
8039 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
8040 0 : *second_sc = SPDK_NVME_SC_ABORTED_FAILED_FUSED;
8041 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_MISCOMPARE) {
8042 0 : *first_sct = SPDK_NVME_SCT_MEDIA_ERROR;
8043 0 : *first_sc = SPDK_NVME_SC_COMPARE_FAILURE;
8044 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
8045 0 : *second_sc = SPDK_NVME_SC_ABORTED_FAILED_FUSED;
8046 : } else {
8047 0 : *first_sct = SPDK_NVME_SCT_GENERIC;
8048 0 : *first_sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
8049 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
8050 0 : *second_sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
8051 : }
8052 :
8053 0 : *cdw0 = bdev_io->internal.error.nvme.cdw0;
8054 0 : }
8055 :
8056 : void
8057 0 : spdk_bdev_io_complete_base_io_status(struct spdk_bdev_io *bdev_io,
8058 : const struct spdk_bdev_io *base_io)
8059 : {
8060 0 : switch (base_io->internal.status) {
8061 0 : case SPDK_BDEV_IO_STATUS_NVME_ERROR:
8062 0 : spdk_bdev_io_complete_nvme_status(bdev_io,
8063 0 : base_io->internal.error.nvme.cdw0,
8064 0 : base_io->internal.error.nvme.sct,
8065 0 : base_io->internal.error.nvme.sc);
8066 0 : break;
8067 0 : case SPDK_BDEV_IO_STATUS_SCSI_ERROR:
8068 0 : spdk_bdev_io_complete_scsi_status(bdev_io,
8069 0 : base_io->internal.error.scsi.sc,
8070 0 : base_io->internal.error.scsi.sk,
8071 0 : base_io->internal.error.scsi.asc,
8072 0 : base_io->internal.error.scsi.ascq);
8073 0 : break;
8074 0 : case SPDK_BDEV_IO_STATUS_AIO_ERROR:
8075 0 : spdk_bdev_io_complete_aio_status(bdev_io, base_io->internal.error.aio_result);
8076 0 : break;
8077 0 : default:
8078 0 : spdk_bdev_io_complete(bdev_io, base_io->internal.status);
8079 0 : break;
8080 : }
8081 0 : }
8082 :
8083 : struct spdk_thread *
8084 664 : spdk_bdev_io_get_thread(struct spdk_bdev_io *bdev_io)
8085 : {
8086 664 : return spdk_io_channel_get_thread(bdev_io->internal.ch->channel);
8087 : }
8088 :
8089 : struct spdk_io_channel *
8090 70 : spdk_bdev_io_get_io_channel(struct spdk_bdev_io *bdev_io)
8091 : {
8092 70 : return bdev_io->internal.ch->channel;
8093 : }
8094 :
8095 : static int
8096 132 : bdev_register(struct spdk_bdev *bdev)
8097 : {
8098 : char *bdev_name;
8099 132 : char uuid[SPDK_UUID_STRING_LEN];
8100 132 : struct spdk_iobuf_opts iobuf_opts;
8101 : int ret;
8102 :
8103 132 : assert(bdev->module != NULL);
8104 :
8105 132 : if (!bdev->name) {
8106 0 : SPDK_ERRLOG("Bdev name is NULL\n");
8107 0 : return -EINVAL;
8108 : }
8109 :
8110 132 : if (!strlen(bdev->name)) {
8111 0 : SPDK_ERRLOG("Bdev name must not be an empty string\n");
8112 0 : return -EINVAL;
8113 : }
8114 :
8115 : /* Users often register their own I/O devices using the bdev name. In
8116 : * order to avoid conflicts, prepend bdev_. */
8117 132 : bdev_name = spdk_sprintf_alloc("bdev_%s", bdev->name);
8118 132 : if (!bdev_name) {
8119 0 : SPDK_ERRLOG("Unable to allocate memory for internal bdev name.\n");
8120 0 : return -ENOMEM;
8121 : }
8122 :
8123 132 : bdev->internal.stat = bdev_alloc_io_stat(true);
8124 132 : if (!bdev->internal.stat) {
8125 0 : SPDK_ERRLOG("Unable to allocate I/O statistics structure.\n");
8126 0 : free(bdev_name);
8127 0 : return -ENOMEM;
8128 : }
8129 :
8130 132 : bdev->internal.status = SPDK_BDEV_STATUS_READY;
8131 132 : bdev->internal.measured_queue_depth = UINT64_MAX;
8132 132 : bdev->internal.claim_type = SPDK_BDEV_CLAIM_NONE;
8133 132 : memset(&bdev->internal.claim, 0, sizeof(bdev->internal.claim));
8134 132 : bdev->internal.qd_poller = NULL;
8135 132 : bdev->internal.qos = NULL;
8136 :
8137 132 : TAILQ_INIT(&bdev->internal.open_descs);
8138 132 : TAILQ_INIT(&bdev->internal.locked_ranges);
8139 132 : TAILQ_INIT(&bdev->internal.pending_locked_ranges);
8140 132 : TAILQ_INIT(&bdev->internal.queued_resets);
8141 132 : TAILQ_INIT(&bdev->aliases);
8142 :
8143 : /* UUID may be specified by the user or defined by bdev itself.
8144 : * Otherwise it will be generated here, so this field will never be empty. */
8145 132 : if (spdk_uuid_is_null(&bdev->uuid)) {
8146 43 : spdk_uuid_generate(&bdev->uuid);
8147 : }
8148 :
8149 : /* Add the UUID alias only if it's different than the name */
8150 132 : spdk_uuid_fmt_lower(uuid, sizeof(uuid), &bdev->uuid);
8151 132 : if (strcmp(bdev->name, uuid) != 0) {
8152 131 : ret = spdk_bdev_alias_add(bdev, uuid);
8153 131 : if (ret != 0) {
8154 2 : SPDK_ERRLOG("Unable to add uuid:%s alias for bdev %s\n", uuid, bdev->name);
8155 2 : bdev_free_io_stat(bdev->internal.stat);
8156 2 : free(bdev_name);
8157 2 : return ret;
8158 : }
8159 : }
8160 :
8161 130 : spdk_iobuf_get_opts(&iobuf_opts, sizeof(iobuf_opts));
8162 130 : if (spdk_bdev_get_buf_align(bdev) > 1) {
8163 0 : bdev->max_rw_size = spdk_min(bdev->max_rw_size ? bdev->max_rw_size : UINT32_MAX,
8164 : iobuf_opts.large_bufsize / bdev->blocklen);
8165 : }
8166 :
8167 : /* If the user didn't specify a write unit size, set it to one. */
8168 130 : if (bdev->write_unit_size == 0) {
8169 126 : bdev->write_unit_size = 1;
8170 : }
8171 :
8172 : /* Set ACWU value to the write unit size if bdev module did not set it (does not support it natively) */
8173 130 : if (bdev->acwu == 0) {
8174 126 : bdev->acwu = bdev->write_unit_size;
8175 : }
8176 :
8177 130 : if (bdev->phys_blocklen == 0) {
8178 126 : bdev->phys_blocklen = spdk_bdev_get_data_block_size(bdev);
8179 : }
8180 :
8181 130 : if (!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COPY)) {
8182 0 : bdev->max_copy = bdev_get_max_write(bdev, iobuf_opts.large_bufsize);
8183 : }
8184 :
8185 130 : if (!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE_ZEROES)) {
8186 0 : bdev->max_write_zeroes = bdev_get_max_write(bdev, ZERO_BUFFER_SIZE);
8187 : }
8188 :
8189 130 : bdev->internal.reset_in_progress = NULL;
8190 130 : bdev->internal.qd_poll_in_progress = false;
8191 130 : bdev->internal.period = 0;
8192 130 : bdev->internal.new_period = 0;
8193 130 : bdev->internal.trace_id = spdk_trace_register_owner(OWNER_TYPE_BDEV, bdev_name);
8194 :
8195 : /*
8196 : * Initialize spinlock before registering IO device because spinlock is used in
8197 : * bdev_channel_create
8198 : */
8199 130 : spdk_spin_init(&bdev->internal.spinlock);
8200 :
8201 130 : spdk_io_device_register(__bdev_to_io_dev(bdev),
8202 : bdev_channel_create, bdev_channel_destroy,
8203 : sizeof(struct spdk_bdev_channel),
8204 : bdev_name);
8205 :
8206 : /*
8207 : * Register bdev name only after the bdev object is ready.
8208 : * After bdev_name_add returns, it is possible for other threads to start using the bdev,
8209 : * create IO channels...
8210 : */
8211 130 : ret = bdev_name_add(&bdev->internal.bdev_name, bdev, bdev->name);
8212 130 : if (ret != 0) {
8213 0 : spdk_io_device_unregister(__bdev_to_io_dev(bdev), NULL);
8214 0 : bdev_free_io_stat(bdev->internal.stat);
8215 0 : spdk_spin_destroy(&bdev->internal.spinlock);
8216 0 : free(bdev_name);
8217 0 : return ret;
8218 : }
8219 :
8220 130 : free(bdev_name);
8221 :
8222 130 : SPDK_DEBUGLOG(bdev, "Inserting bdev %s into list\n", bdev->name);
8223 130 : TAILQ_INSERT_TAIL(&g_bdev_mgr.bdevs, bdev, internal.link);
8224 :
8225 130 : return 0;
8226 : }
8227 :
8228 : static void
8229 131 : bdev_destroy_cb(void *io_device)
8230 : {
8231 : int rc;
8232 : struct spdk_bdev *bdev;
8233 : spdk_bdev_unregister_cb cb_fn;
8234 : void *cb_arg;
8235 :
8236 131 : bdev = __bdev_from_io_dev(io_device);
8237 :
8238 131 : if (bdev->internal.unregister_td != spdk_get_thread()) {
8239 1 : spdk_thread_send_msg(bdev->internal.unregister_td, bdev_destroy_cb, io_device);
8240 1 : return;
8241 : }
8242 :
8243 130 : cb_fn = bdev->internal.unregister_cb;
8244 130 : cb_arg = bdev->internal.unregister_ctx;
8245 :
8246 130 : spdk_spin_destroy(&bdev->internal.spinlock);
8247 130 : free(bdev->internal.qos);
8248 130 : bdev_free_io_stat(bdev->internal.stat);
8249 130 : spdk_trace_unregister_owner(bdev->internal.trace_id);
8250 :
8251 130 : rc = bdev->fn_table->destruct(bdev->ctxt);
8252 130 : if (rc < 0) {
8253 0 : SPDK_ERRLOG("destruct failed\n");
8254 : }
8255 130 : if (rc <= 0 && cb_fn != NULL) {
8256 10 : cb_fn(cb_arg, rc);
8257 : }
8258 : }
8259 :
8260 : void
8261 2 : spdk_bdev_destruct_done(struct spdk_bdev *bdev, int bdeverrno)
8262 : {
8263 2 : if (bdev->internal.unregister_cb != NULL) {
8264 0 : bdev->internal.unregister_cb(bdev->internal.unregister_ctx, bdeverrno);
8265 : }
8266 2 : }
8267 :
8268 : static void
8269 19 : _remove_notify(void *arg)
8270 : {
8271 19 : struct spdk_bdev_desc *desc = arg;
8272 :
8273 19 : _event_notify(desc, SPDK_BDEV_EVENT_REMOVE);
8274 19 : }
8275 :
8276 : /* returns: 0 - bdev removed and ready to be destructed.
8277 : * -EBUSY - bdev can't be destructed yet. */
8278 : static int
8279 145 : bdev_unregister_unsafe(struct spdk_bdev *bdev)
8280 : {
8281 : struct spdk_bdev_desc *desc, *tmp;
8282 : struct spdk_bdev_alias *alias;
8283 145 : int rc = 0;
8284 145 : char uuid[SPDK_UUID_STRING_LEN];
8285 :
8286 145 : assert(spdk_spin_held(&g_bdev_mgr.spinlock));
8287 145 : assert(spdk_spin_held(&bdev->internal.spinlock));
8288 :
8289 : /* Notify each descriptor about hotremoval */
8290 164 : TAILQ_FOREACH_SAFE(desc, &bdev->internal.open_descs, link, tmp) {
8291 19 : rc = -EBUSY;
8292 : /*
8293 : * Defer invocation of the event_cb to a separate message that will
8294 : * run later on its thread. This ensures this context unwinds and
8295 : * we don't recursively unregister this bdev again if the event_cb
8296 : * immediately closes its descriptor.
8297 : */
8298 19 : event_notify(desc, _remove_notify);
8299 : }
8300 :
8301 : /* If there are no descriptors, proceed removing the bdev */
8302 145 : if (rc == 0) {
8303 130 : bdev_examine_allowlist_remove(bdev->name);
8304 258 : TAILQ_FOREACH(alias, &bdev->aliases, tailq) {
8305 128 : bdev_examine_allowlist_remove(alias->alias.name);
8306 : }
8307 130 : TAILQ_REMOVE(&g_bdev_mgr.bdevs, bdev, internal.link);
8308 130 : SPDK_DEBUGLOG(bdev, "Removing bdev %s from list done\n", bdev->name);
8309 :
8310 : /* Delete the name and the UUID alias */
8311 130 : spdk_uuid_fmt_lower(uuid, sizeof(uuid), &bdev->uuid);
8312 130 : bdev_name_del_unsafe(&bdev->internal.bdev_name);
8313 130 : bdev_alias_del(bdev, uuid, bdev_name_del_unsafe);
8314 :
8315 130 : spdk_notify_send("bdev_unregister", spdk_bdev_get_name(bdev));
8316 :
8317 130 : if (bdev->internal.reset_in_progress != NULL) {
8318 : /* If reset is in progress, let the completion callback for reset
8319 : * unregister the bdev.
8320 : */
8321 1 : rc = -EBUSY;
8322 : }
8323 : }
8324 :
8325 145 : return rc;
8326 : }
8327 :
8328 : static void
8329 4 : bdev_unregister_abort_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
8330 : struct spdk_io_channel *io_ch, void *_ctx)
8331 : {
8332 4 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
8333 :
8334 4 : bdev_channel_abort_queued_ios(bdev_ch);
8335 4 : spdk_bdev_for_each_channel_continue(i, 0);
8336 4 : }
8337 :
8338 : static void
8339 130 : bdev_unregister(struct spdk_bdev *bdev, void *_ctx, int status)
8340 : {
8341 : int rc;
8342 :
8343 130 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8344 130 : spdk_spin_lock(&bdev->internal.spinlock);
8345 : /*
8346 : * Set the status to REMOVING after completing to abort channels. Otherwise,
8347 : * the last spdk_bdev_close() may call spdk_io_device_unregister() while
8348 : * spdk_bdev_for_each_channel() is executed and spdk_io_device_unregister()
8349 : * may fail.
8350 : */
8351 130 : bdev->internal.status = SPDK_BDEV_STATUS_REMOVING;
8352 130 : rc = bdev_unregister_unsafe(bdev);
8353 130 : spdk_spin_unlock(&bdev->internal.spinlock);
8354 130 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8355 :
8356 130 : if (rc == 0) {
8357 114 : spdk_io_device_unregister(__bdev_to_io_dev(bdev), bdev_destroy_cb);
8358 : }
8359 130 : }
8360 :
8361 : void
8362 137 : spdk_bdev_unregister(struct spdk_bdev *bdev, spdk_bdev_unregister_cb cb_fn, void *cb_arg)
8363 : {
8364 : struct spdk_thread *thread;
8365 :
8366 137 : SPDK_DEBUGLOG(bdev, "Removing bdev %s from list\n", bdev->name);
8367 :
8368 137 : thread = spdk_get_thread();
8369 137 : if (!thread) {
8370 : /* The user called this from a non-SPDK thread. */
8371 0 : if (cb_fn != NULL) {
8372 0 : cb_fn(cb_arg, -ENOTSUP);
8373 : }
8374 0 : return;
8375 : }
8376 :
8377 137 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8378 137 : if (bdev->internal.status == SPDK_BDEV_STATUS_UNREGISTERING ||
8379 137 : bdev->internal.status == SPDK_BDEV_STATUS_REMOVING) {
8380 7 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8381 7 : if (cb_fn) {
8382 0 : cb_fn(cb_arg, -EBUSY);
8383 : }
8384 7 : return;
8385 : }
8386 :
8387 130 : spdk_spin_lock(&bdev->internal.spinlock);
8388 130 : bdev->internal.status = SPDK_BDEV_STATUS_UNREGISTERING;
8389 130 : bdev->internal.unregister_cb = cb_fn;
8390 130 : bdev->internal.unregister_ctx = cb_arg;
8391 130 : bdev->internal.unregister_td = thread;
8392 130 : spdk_spin_unlock(&bdev->internal.spinlock);
8393 130 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8394 :
8395 130 : spdk_bdev_set_qd_sampling_period(bdev, 0);
8396 :
8397 130 : spdk_bdev_for_each_channel(bdev, bdev_unregister_abort_channel, bdev,
8398 : bdev_unregister);
8399 : }
8400 :
8401 : int
8402 4 : spdk_bdev_unregister_by_name(const char *bdev_name, struct spdk_bdev_module *module,
8403 : spdk_bdev_unregister_cb cb_fn, void *cb_arg)
8404 : {
8405 4 : struct spdk_bdev_desc *desc;
8406 : struct spdk_bdev *bdev;
8407 : int rc;
8408 :
8409 4 : rc = spdk_bdev_open_ext(bdev_name, false, _tmp_bdev_event_cb, NULL, &desc);
8410 4 : if (rc != 0) {
8411 1 : SPDK_ERRLOG("Failed to open bdev with name: %s\n", bdev_name);
8412 1 : return rc;
8413 : }
8414 :
8415 3 : bdev = spdk_bdev_desc_get_bdev(desc);
8416 :
8417 3 : if (bdev->module != module) {
8418 1 : spdk_bdev_close(desc);
8419 1 : SPDK_ERRLOG("Bdev %s was not registered by the specified module.\n",
8420 : bdev_name);
8421 1 : return -ENODEV;
8422 : }
8423 :
8424 2 : spdk_bdev_unregister(bdev, cb_fn, cb_arg);
8425 :
8426 2 : spdk_bdev_close(desc);
8427 :
8428 2 : return 0;
8429 : }
8430 :
8431 : static int
8432 269 : bdev_start_qos(struct spdk_bdev *bdev)
8433 : {
8434 : struct set_qos_limit_ctx *ctx;
8435 :
8436 : /* Enable QoS */
8437 269 : if (bdev->internal.qos && bdev->internal.qos->thread == NULL) {
8438 2 : ctx = calloc(1, sizeof(*ctx));
8439 2 : if (ctx == NULL) {
8440 0 : SPDK_ERRLOG("Failed to allocate memory for QoS context\n");
8441 0 : return -ENOMEM;
8442 : }
8443 2 : ctx->bdev = bdev;
8444 2 : spdk_bdev_for_each_channel(bdev, bdev_enable_qos_msg, ctx, bdev_enable_qos_done);
8445 : }
8446 :
8447 269 : return 0;
8448 : }
8449 :
8450 : static void
8451 25 : log_already_claimed(enum spdk_log_level level, const int line, const char *func, const char *detail,
8452 : struct spdk_bdev *bdev)
8453 : {
8454 : enum spdk_bdev_claim_type type;
8455 : const char *typename, *modname;
8456 : extern struct spdk_log_flag SPDK_LOG_bdev;
8457 :
8458 25 : assert(spdk_spin_held(&bdev->internal.spinlock));
8459 :
8460 25 : if (level >= SPDK_LOG_INFO && !SPDK_LOG_bdev.enabled) {
8461 0 : return;
8462 : }
8463 :
8464 25 : type = bdev->internal.claim_type;
8465 25 : typename = spdk_bdev_claim_get_name(type);
8466 :
8467 25 : if (type == SPDK_BDEV_CLAIM_EXCL_WRITE) {
8468 6 : modname = bdev->internal.claim.v1.module->name;
8469 6 : spdk_log(level, __FILE__, line, func, "bdev %s %s: type %s by module %s\n",
8470 : bdev->name, detail, typename, modname);
8471 6 : return;
8472 : }
8473 :
8474 19 : if (claim_type_is_v2(type)) {
8475 : struct spdk_bdev_module_claim *claim;
8476 :
8477 38 : TAILQ_FOREACH(claim, &bdev->internal.claim.v2.claims, link) {
8478 19 : modname = claim->module->name;
8479 19 : spdk_log(level, __FILE__, line, func, "bdev %s %s: type %s by module %s\n",
8480 : bdev->name, detail, typename, modname);
8481 : }
8482 19 : return;
8483 : }
8484 :
8485 0 : assert(false);
8486 : }
8487 :
8488 : static int
8489 278 : bdev_open(struct spdk_bdev *bdev, bool write, struct spdk_bdev_desc *desc)
8490 : {
8491 : struct spdk_thread *thread;
8492 278 : int rc = 0;
8493 :
8494 278 : thread = spdk_get_thread();
8495 278 : if (!thread) {
8496 0 : SPDK_ERRLOG("Cannot open bdev from non-SPDK thread.\n");
8497 0 : return -ENOTSUP;
8498 : }
8499 :
8500 278 : SPDK_DEBUGLOG(bdev, "Opening descriptor %p for bdev %s on thread %p\n", desc, bdev->name,
8501 : spdk_get_thread());
8502 :
8503 278 : desc->bdev = bdev;
8504 278 : desc->thread = thread;
8505 278 : desc->write = write;
8506 :
8507 278 : spdk_spin_lock(&bdev->internal.spinlock);
8508 278 : if (bdev->internal.status == SPDK_BDEV_STATUS_UNREGISTERING ||
8509 278 : bdev->internal.status == SPDK_BDEV_STATUS_REMOVING) {
8510 3 : spdk_spin_unlock(&bdev->internal.spinlock);
8511 3 : return -ENODEV;
8512 : }
8513 :
8514 275 : if (write && bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE) {
8515 6 : LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
8516 6 : spdk_spin_unlock(&bdev->internal.spinlock);
8517 6 : return -EPERM;
8518 : }
8519 :
8520 269 : rc = bdev_start_qos(bdev);
8521 269 : if (rc != 0) {
8522 0 : SPDK_ERRLOG("Failed to start QoS on bdev %s\n", bdev->name);
8523 0 : spdk_spin_unlock(&bdev->internal.spinlock);
8524 0 : return rc;
8525 : }
8526 :
8527 269 : TAILQ_INSERT_TAIL(&bdev->internal.open_descs, desc, link);
8528 :
8529 269 : spdk_spin_unlock(&bdev->internal.spinlock);
8530 :
8531 269 : return 0;
8532 : }
8533 :
8534 : static void
8535 279 : bdev_open_opts_get_defaults(struct spdk_bdev_open_opts *opts, size_t opts_size)
8536 : {
8537 279 : if (!opts) {
8538 0 : SPDK_ERRLOG("opts should not be NULL.\n");
8539 0 : return;
8540 : }
8541 :
8542 279 : if (!opts_size) {
8543 0 : SPDK_ERRLOG("opts_size should not be zero.\n");
8544 0 : return;
8545 : }
8546 :
8547 279 : memset(opts, 0, opts_size);
8548 279 : opts->size = opts_size;
8549 :
8550 : #define FIELD_OK(field) \
8551 : offsetof(struct spdk_bdev_open_opts, field) + sizeof(opts->field) <= opts_size
8552 :
8553 : #define SET_FIELD(field, value) \
8554 : if (FIELD_OK(field)) { \
8555 : opts->field = value; \
8556 : } \
8557 :
8558 279 : SET_FIELD(hide_metadata, false);
8559 :
8560 : #undef FIELD_OK
8561 : #undef SET_FIELD
8562 : }
8563 :
8564 : static void
8565 2 : bdev_open_opts_copy(struct spdk_bdev_open_opts *opts,
8566 : const struct spdk_bdev_open_opts *opts_src, size_t opts_size)
8567 : {
8568 2 : assert(opts);
8569 2 : assert(opts_src);
8570 :
8571 : #define SET_FIELD(field) \
8572 : if (offsetof(struct spdk_bdev_open_opts, field) + sizeof(opts->field) <= opts_size) { \
8573 : opts->field = opts_src->field; \
8574 : } \
8575 :
8576 2 : SET_FIELD(hide_metadata);
8577 :
8578 2 : opts->size = opts_src->size;
8579 :
8580 : /* We should not remove this statement, but need to update the assert statement
8581 : * if we add a new field, and also add a corresponding SET_FIELD statement.
8582 : */
8583 : SPDK_STATIC_ASSERT(sizeof(struct spdk_bdev_open_opts) == 16, "Incorrect size");
8584 :
8585 : #undef SET_FIELD
8586 2 : }
8587 :
8588 : void
8589 1 : spdk_bdev_open_opts_init(struct spdk_bdev_open_opts *opts, size_t opts_size)
8590 : {
8591 1 : struct spdk_bdev_open_opts opts_local;
8592 :
8593 1 : bdev_open_opts_get_defaults(&opts_local, sizeof(opts_local));
8594 1 : bdev_open_opts_copy(opts, &opts_local, opts_size);
8595 1 : }
8596 :
8597 : static int
8598 278 : bdev_desc_alloc(struct spdk_bdev *bdev, spdk_bdev_event_cb_t event_cb, void *event_ctx,
8599 : struct spdk_bdev_open_opts *user_opts, struct spdk_bdev_desc **_desc)
8600 : {
8601 : struct spdk_bdev_desc *desc;
8602 278 : struct spdk_bdev_open_opts opts;
8603 : unsigned int i;
8604 :
8605 278 : bdev_open_opts_get_defaults(&opts, sizeof(opts));
8606 278 : if (user_opts != NULL) {
8607 1 : bdev_open_opts_copy(&opts, user_opts, user_opts->size);
8608 : }
8609 :
8610 278 : desc = calloc(1, sizeof(*desc));
8611 278 : if (desc == NULL) {
8612 0 : SPDK_ERRLOG("Failed to allocate memory for bdev descriptor\n");
8613 0 : return -ENOMEM;
8614 : }
8615 :
8616 278 : desc->opts = opts;
8617 :
8618 278 : TAILQ_INIT(&desc->pending_media_events);
8619 278 : TAILQ_INIT(&desc->free_media_events);
8620 :
8621 278 : desc->memory_domains_supported = spdk_bdev_get_memory_domains(bdev, NULL, 0) > 0;
8622 278 : desc->callback.event_fn = event_cb;
8623 278 : desc->callback.ctx = event_ctx;
8624 278 : spdk_spin_init(&desc->spinlock);
8625 :
8626 278 : if (desc->opts.hide_metadata) {
8627 1 : if (spdk_bdev_is_md_separate(bdev)) {
8628 0 : SPDK_ERRLOG("hide_metadata option is not supported with separate metadata.\n");
8629 0 : bdev_desc_free(desc);
8630 0 : return -EINVAL;
8631 : }
8632 : }
8633 :
8634 278 : if (bdev->media_events) {
8635 0 : desc->media_events_buffer = calloc(MEDIA_EVENT_POOL_SIZE,
8636 : sizeof(*desc->media_events_buffer));
8637 0 : if (desc->media_events_buffer == NULL) {
8638 0 : SPDK_ERRLOG("Failed to initialize media event pool\n");
8639 0 : bdev_desc_free(desc);
8640 0 : return -ENOMEM;
8641 : }
8642 :
8643 0 : for (i = 0; i < MEDIA_EVENT_POOL_SIZE; ++i) {
8644 0 : TAILQ_INSERT_TAIL(&desc->free_media_events,
8645 : &desc->media_events_buffer[i], tailq);
8646 : }
8647 : }
8648 :
8649 278 : if (bdev->fn_table->accel_sequence_supported != NULL) {
8650 0 : for (i = 0; i < SPDK_BDEV_NUM_IO_TYPES; ++i) {
8651 0 : desc->accel_sequence_supported[i] =
8652 0 : bdev->fn_table->accel_sequence_supported(bdev->ctxt,
8653 : (enum spdk_bdev_io_type)i);
8654 : }
8655 : }
8656 :
8657 278 : *_desc = desc;
8658 :
8659 278 : return 0;
8660 : }
8661 :
8662 : static int
8663 136 : bdev_open_ext(const char *bdev_name, bool write, spdk_bdev_event_cb_t event_cb,
8664 : void *event_ctx, struct spdk_bdev_open_opts *opts,
8665 : struct spdk_bdev_desc **_desc)
8666 : {
8667 136 : struct spdk_bdev_desc *desc;
8668 : struct spdk_bdev *bdev;
8669 : int rc;
8670 :
8671 136 : bdev = bdev_get_by_name(bdev_name);
8672 :
8673 136 : if (bdev == NULL) {
8674 1 : SPDK_NOTICELOG("Currently unable to find bdev with name: %s\n", bdev_name);
8675 1 : return -ENODEV;
8676 : }
8677 :
8678 135 : rc = bdev_desc_alloc(bdev, event_cb, event_ctx, opts, &desc);
8679 135 : if (rc != 0) {
8680 0 : return rc;
8681 : }
8682 :
8683 135 : rc = bdev_open(bdev, write, desc);
8684 135 : if (rc != 0) {
8685 7 : bdev_desc_free(desc);
8686 7 : desc = NULL;
8687 : }
8688 :
8689 135 : *_desc = desc;
8690 :
8691 135 : return rc;
8692 : }
8693 :
8694 : int
8695 138 : spdk_bdev_open_ext_v2(const char *bdev_name, bool write, spdk_bdev_event_cb_t event_cb,
8696 : void *event_ctx, struct spdk_bdev_open_opts *opts,
8697 : struct spdk_bdev_desc **_desc)
8698 : {
8699 : int rc;
8700 :
8701 138 : if (event_cb == NULL) {
8702 2 : SPDK_ERRLOG("Missing event callback function\n");
8703 2 : return -EINVAL;
8704 : }
8705 :
8706 136 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8707 136 : rc = bdev_open_ext(bdev_name, write, event_cb, event_ctx, opts, _desc);
8708 136 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8709 :
8710 136 : return rc;
8711 : }
8712 :
8713 : int
8714 136 : spdk_bdev_open_ext(const char *bdev_name, bool write, spdk_bdev_event_cb_t event_cb,
8715 : void *event_ctx, struct spdk_bdev_desc **_desc)
8716 : {
8717 136 : return spdk_bdev_open_ext_v2(bdev_name, write, event_cb, event_ctx, NULL, _desc);
8718 : }
8719 :
8720 : struct spdk_bdev_open_async_ctx {
8721 : char *bdev_name;
8722 : spdk_bdev_event_cb_t event_cb;
8723 : void *event_ctx;
8724 : bool write;
8725 : int rc;
8726 : spdk_bdev_open_async_cb_t cb_fn;
8727 : void *cb_arg;
8728 : struct spdk_bdev_desc *desc;
8729 : struct spdk_bdev_open_async_opts opts;
8730 : uint64_t start_ticks;
8731 : struct spdk_thread *orig_thread;
8732 : struct spdk_poller *poller;
8733 : TAILQ_ENTRY(spdk_bdev_open_async_ctx) tailq;
8734 : };
8735 :
8736 : static void
8737 0 : bdev_open_async_done(void *arg)
8738 : {
8739 0 : struct spdk_bdev_open_async_ctx *ctx = arg;
8740 :
8741 0 : ctx->cb_fn(ctx->desc, ctx->rc, ctx->cb_arg);
8742 :
8743 0 : free(ctx->bdev_name);
8744 0 : free(ctx);
8745 0 : }
8746 :
8747 : static void
8748 0 : bdev_open_async_cancel(void *arg)
8749 : {
8750 0 : struct spdk_bdev_open_async_ctx *ctx = arg;
8751 :
8752 0 : assert(ctx->rc == -ESHUTDOWN);
8753 :
8754 0 : spdk_poller_unregister(&ctx->poller);
8755 :
8756 0 : bdev_open_async_done(ctx);
8757 0 : }
8758 :
8759 : /* This is called when the bdev library finishes at shutdown. */
8760 : static void
8761 68 : bdev_open_async_fini(void)
8762 : {
8763 : struct spdk_bdev_open_async_ctx *ctx, *tmp_ctx;
8764 :
8765 68 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8766 68 : TAILQ_FOREACH_SAFE(ctx, &g_bdev_mgr.async_bdev_opens, tailq, tmp_ctx) {
8767 0 : TAILQ_REMOVE(&g_bdev_mgr.async_bdev_opens, ctx, tailq);
8768 : /*
8769 : * We have to move to ctx->orig_thread to unregister ctx->poller.
8770 : * However, there is a chance that ctx->poller is executed before
8771 : * message is executed, which could result in bdev_open_async_done()
8772 : * being called twice. To avoid such race condition, set ctx->rc to
8773 : * -ESHUTDOWN.
8774 : */
8775 0 : ctx->rc = -ESHUTDOWN;
8776 0 : spdk_thread_send_msg(ctx->orig_thread, bdev_open_async_cancel, ctx);
8777 : }
8778 68 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8779 68 : }
8780 :
8781 : static int bdev_open_async(void *arg);
8782 :
8783 : static void
8784 0 : _bdev_open_async(struct spdk_bdev_open_async_ctx *ctx)
8785 : {
8786 : uint64_t timeout_ticks;
8787 :
8788 0 : if (ctx->rc == -ESHUTDOWN) {
8789 : /* This context is being canceled. Do nothing. */
8790 0 : return;
8791 : }
8792 :
8793 0 : ctx->rc = bdev_open_ext(ctx->bdev_name, ctx->write, ctx->event_cb, ctx->event_ctx,
8794 : NULL, &ctx->desc);
8795 0 : if (ctx->rc == 0 || ctx->opts.timeout_ms == 0) {
8796 0 : goto exit;
8797 : }
8798 :
8799 0 : timeout_ticks = ctx->start_ticks + ctx->opts.timeout_ms * spdk_get_ticks_hz() / 1000ull;
8800 0 : if (spdk_get_ticks() >= timeout_ticks) {
8801 0 : SPDK_ERRLOG("Timed out while waiting for bdev '%s' to appear\n", ctx->bdev_name);
8802 0 : ctx->rc = -ETIMEDOUT;
8803 0 : goto exit;
8804 : }
8805 :
8806 0 : return;
8807 :
8808 0 : exit:
8809 0 : spdk_poller_unregister(&ctx->poller);
8810 0 : TAILQ_REMOVE(&g_bdev_mgr.async_bdev_opens, ctx, tailq);
8811 :
8812 : /* Completion callback is processed after stack unwinding. */
8813 0 : spdk_thread_send_msg(ctx->orig_thread, bdev_open_async_done, ctx);
8814 : }
8815 :
8816 : static int
8817 0 : bdev_open_async(void *arg)
8818 : {
8819 0 : struct spdk_bdev_open_async_ctx *ctx = arg;
8820 :
8821 0 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8822 :
8823 0 : _bdev_open_async(ctx);
8824 :
8825 0 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8826 :
8827 0 : return SPDK_POLLER_BUSY;
8828 : }
8829 :
8830 : static void
8831 0 : bdev_open_async_opts_copy(struct spdk_bdev_open_async_opts *opts,
8832 : struct spdk_bdev_open_async_opts *opts_src,
8833 : size_t size)
8834 : {
8835 0 : assert(opts);
8836 0 : assert(opts_src);
8837 :
8838 0 : opts->size = size;
8839 :
8840 : #define SET_FIELD(field) \
8841 : if (offsetof(struct spdk_bdev_open_async_opts, field) + sizeof(opts->field) <= size) { \
8842 : opts->field = opts_src->field; \
8843 : } \
8844 :
8845 0 : SET_FIELD(timeout_ms);
8846 :
8847 : /* Do not remove this statement, you should always update this statement when you adding a new field,
8848 : * and do not forget to add the SET_FIELD statement for your added field. */
8849 : SPDK_STATIC_ASSERT(sizeof(struct spdk_bdev_open_async_opts) == 16, "Incorrect size");
8850 :
8851 : #undef SET_FIELD
8852 0 : }
8853 :
8854 : static void
8855 0 : bdev_open_async_opts_get_default(struct spdk_bdev_open_async_opts *opts, size_t size)
8856 : {
8857 0 : assert(opts);
8858 :
8859 0 : opts->size = size;
8860 :
8861 : #define SET_FIELD(field, value) \
8862 : if (offsetof(struct spdk_bdev_open_async_opts, field) + sizeof(opts->field) <= size) { \
8863 : opts->field = value; \
8864 : } \
8865 :
8866 0 : SET_FIELD(timeout_ms, 0);
8867 :
8868 : #undef SET_FIELD
8869 0 : }
8870 :
8871 : int
8872 0 : spdk_bdev_open_async(const char *bdev_name, bool write, spdk_bdev_event_cb_t event_cb,
8873 : void *event_ctx, struct spdk_bdev_open_async_opts *opts,
8874 : spdk_bdev_open_async_cb_t open_cb, void *open_cb_arg)
8875 : {
8876 : struct spdk_bdev_open_async_ctx *ctx;
8877 :
8878 0 : if (event_cb == NULL) {
8879 0 : SPDK_ERRLOG("Missing event callback function\n");
8880 0 : return -EINVAL;
8881 : }
8882 :
8883 0 : if (open_cb == NULL) {
8884 0 : SPDK_ERRLOG("Missing open callback function\n");
8885 0 : return -EINVAL;
8886 : }
8887 :
8888 0 : if (opts != NULL && opts->size == 0) {
8889 0 : SPDK_ERRLOG("size in the options structure should not be zero\n");
8890 0 : return -EINVAL;
8891 : }
8892 :
8893 0 : ctx = calloc(1, sizeof(*ctx));
8894 0 : if (ctx == NULL) {
8895 0 : SPDK_ERRLOG("Failed to allocate open context\n");
8896 0 : return -ENOMEM;
8897 : }
8898 :
8899 0 : ctx->bdev_name = strdup(bdev_name);
8900 0 : if (ctx->bdev_name == NULL) {
8901 0 : SPDK_ERRLOG("Failed to duplicate bdev_name\n");
8902 0 : free(ctx);
8903 0 : return -ENOMEM;
8904 : }
8905 :
8906 0 : ctx->poller = SPDK_POLLER_REGISTER(bdev_open_async, ctx, 100 * 1000);
8907 0 : if (ctx->poller == NULL) {
8908 0 : SPDK_ERRLOG("Failed to register bdev_open_async poller\n");
8909 0 : free(ctx->bdev_name);
8910 0 : free(ctx);
8911 0 : return -ENOMEM;
8912 : }
8913 :
8914 0 : ctx->cb_fn = open_cb;
8915 0 : ctx->cb_arg = open_cb_arg;
8916 0 : ctx->write = write;
8917 0 : ctx->event_cb = event_cb;
8918 0 : ctx->event_ctx = event_ctx;
8919 0 : ctx->orig_thread = spdk_get_thread();
8920 0 : ctx->start_ticks = spdk_get_ticks();
8921 :
8922 0 : bdev_open_async_opts_get_default(&ctx->opts, sizeof(ctx->opts));
8923 0 : if (opts != NULL) {
8924 0 : bdev_open_async_opts_copy(&ctx->opts, opts, opts->size);
8925 : }
8926 :
8927 0 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8928 :
8929 0 : TAILQ_INSERT_TAIL(&g_bdev_mgr.async_bdev_opens, ctx, tailq);
8930 0 : _bdev_open_async(ctx);
8931 :
8932 0 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8933 :
8934 0 : return 0;
8935 : }
8936 :
8937 : static void
8938 269 : bdev_close(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc)
8939 : {
8940 : int rc;
8941 :
8942 269 : spdk_spin_lock(&bdev->internal.spinlock);
8943 269 : spdk_spin_lock(&desc->spinlock);
8944 :
8945 269 : TAILQ_REMOVE(&bdev->internal.open_descs, desc, link);
8946 :
8947 269 : desc->closed = true;
8948 :
8949 269 : if (desc->claim != NULL) {
8950 20 : bdev_desc_release_claims(desc);
8951 : }
8952 :
8953 269 : if (0 == desc->refs) {
8954 258 : spdk_spin_unlock(&desc->spinlock);
8955 258 : bdev_desc_free(desc);
8956 : } else {
8957 11 : spdk_spin_unlock(&desc->spinlock);
8958 : }
8959 :
8960 : /* If no more descriptors, kill QoS channel */
8961 269 : if (bdev->internal.qos && TAILQ_EMPTY(&bdev->internal.open_descs)) {
8962 7 : SPDK_DEBUGLOG(bdev, "Closed last descriptor for bdev %s on thread %p. Stopping QoS.\n",
8963 : bdev->name, spdk_get_thread());
8964 :
8965 7 : if (bdev_qos_destroy(bdev)) {
8966 : /* There isn't anything we can do to recover here. Just let the
8967 : * old QoS poller keep running. The QoS handling won't change
8968 : * cores when the user allocates a new channel, but it won't break. */
8969 0 : SPDK_ERRLOG("Unable to shut down QoS poller. It will continue running on the current thread.\n");
8970 : }
8971 : }
8972 :
8973 269 : if (bdev->internal.status == SPDK_BDEV_STATUS_REMOVING && TAILQ_EMPTY(&bdev->internal.open_descs)) {
8974 15 : rc = bdev_unregister_unsafe(bdev);
8975 15 : spdk_spin_unlock(&bdev->internal.spinlock);
8976 :
8977 15 : if (rc == 0) {
8978 15 : spdk_io_device_unregister(__bdev_to_io_dev(bdev), bdev_destroy_cb);
8979 : }
8980 : } else {
8981 254 : spdk_spin_unlock(&bdev->internal.spinlock);
8982 : }
8983 269 : }
8984 :
8985 : void
8986 128 : spdk_bdev_close(struct spdk_bdev_desc *desc)
8987 : {
8988 128 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
8989 :
8990 128 : SPDK_DEBUGLOG(bdev, "Closing descriptor %p for bdev %s on thread %p\n", desc, bdev->name,
8991 : spdk_get_thread());
8992 :
8993 128 : assert(desc->thread == spdk_get_thread());
8994 :
8995 128 : spdk_poller_unregister(&desc->io_timeout_poller);
8996 :
8997 128 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8998 :
8999 128 : bdev_close(bdev, desc);
9000 :
9001 128 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
9002 128 : }
9003 :
9004 : int32_t
9005 3 : spdk_bdev_get_numa_id(struct spdk_bdev *bdev)
9006 : {
9007 3 : if (bdev->numa.id_valid) {
9008 2 : return bdev->numa.id;
9009 : } else {
9010 1 : return SPDK_ENV_NUMA_ID_ANY;
9011 : }
9012 : }
9013 :
9014 : static void
9015 130 : bdev_register_finished(void *arg)
9016 : {
9017 130 : struct spdk_bdev_desc *desc = arg;
9018 130 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
9019 :
9020 130 : spdk_notify_send("bdev_register", spdk_bdev_get_name(bdev));
9021 :
9022 130 : spdk_spin_lock(&g_bdev_mgr.spinlock);
9023 :
9024 130 : bdev_close(bdev, desc);
9025 :
9026 130 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
9027 130 : }
9028 :
9029 : int
9030 133 : spdk_bdev_register(struct spdk_bdev *bdev)
9031 : {
9032 133 : struct spdk_bdev_desc *desc;
9033 133 : struct spdk_thread *thread = spdk_get_thread();
9034 : int rc;
9035 :
9036 133 : if (spdk_unlikely(!spdk_thread_is_app_thread(NULL))) {
9037 1 : SPDK_ERRLOG("Cannot register bdev %s on thread %p (%s)\n", bdev->name, thread,
9038 : thread ? spdk_thread_get_name(thread) : "null");
9039 1 : return -EINVAL;
9040 : }
9041 :
9042 132 : rc = bdev_register(bdev);
9043 132 : if (rc != 0) {
9044 2 : return rc;
9045 : }
9046 :
9047 : /* A descriptor is opened to prevent bdev deletion during examination */
9048 130 : rc = bdev_desc_alloc(bdev, _tmp_bdev_event_cb, NULL, NULL, &desc);
9049 130 : if (rc != 0) {
9050 0 : spdk_bdev_unregister(bdev, NULL, NULL);
9051 0 : return rc;
9052 : }
9053 :
9054 130 : rc = bdev_open(bdev, false, desc);
9055 130 : if (rc != 0) {
9056 0 : bdev_desc_free(desc);
9057 0 : spdk_bdev_unregister(bdev, NULL, NULL);
9058 0 : return rc;
9059 : }
9060 :
9061 : /* Examine configuration before initializing I/O */
9062 130 : bdev_examine(bdev);
9063 :
9064 130 : rc = spdk_bdev_wait_for_examine(bdev_register_finished, desc);
9065 130 : if (rc != 0) {
9066 0 : bdev_close(bdev, desc);
9067 0 : spdk_bdev_unregister(bdev, NULL, NULL);
9068 : }
9069 :
9070 130 : return rc;
9071 : }
9072 :
9073 : int
9074 26 : spdk_bdev_module_claim_bdev(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc,
9075 : struct spdk_bdev_module *module)
9076 : {
9077 26 : spdk_spin_lock(&bdev->internal.spinlock);
9078 :
9079 26 : if (bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE) {
9080 6 : LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
9081 6 : spdk_spin_unlock(&bdev->internal.spinlock);
9082 6 : return -EPERM;
9083 : }
9084 :
9085 20 : if (desc && !desc->write) {
9086 5 : desc->write = true;
9087 : }
9088 :
9089 20 : bdev->internal.claim_type = SPDK_BDEV_CLAIM_EXCL_WRITE;
9090 20 : bdev->internal.claim.v1.module = module;
9091 :
9092 20 : spdk_spin_unlock(&bdev->internal.spinlock);
9093 20 : return 0;
9094 : }
9095 :
9096 : void
9097 8 : spdk_bdev_module_release_bdev(struct spdk_bdev *bdev)
9098 : {
9099 8 : spdk_spin_lock(&bdev->internal.spinlock);
9100 :
9101 8 : assert(bdev->internal.claim.v1.module != NULL);
9102 8 : assert(bdev->internal.claim_type == SPDK_BDEV_CLAIM_EXCL_WRITE);
9103 8 : bdev->internal.claim_type = SPDK_BDEV_CLAIM_NONE;
9104 8 : bdev->internal.claim.v1.module = NULL;
9105 :
9106 8 : spdk_spin_unlock(&bdev->internal.spinlock);
9107 8 : }
9108 :
9109 : /*
9110 : * Start claims v2
9111 : */
9112 :
9113 : const char *
9114 25 : spdk_bdev_claim_get_name(enum spdk_bdev_claim_type type)
9115 : {
9116 25 : switch (type) {
9117 0 : case SPDK_BDEV_CLAIM_NONE:
9118 0 : return "not_claimed";
9119 6 : case SPDK_BDEV_CLAIM_EXCL_WRITE:
9120 6 : return "exclusive_write";
9121 8 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE:
9122 8 : return "read_many_write_one";
9123 5 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_NONE:
9124 5 : return "read_many_write_none";
9125 6 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
9126 6 : return "read_many_write_many";
9127 0 : default:
9128 0 : break;
9129 : }
9130 0 : return "invalid_claim";
9131 : }
9132 :
9133 : static bool
9134 115 : claim_type_is_v2(enum spdk_bdev_claim_type type)
9135 : {
9136 115 : switch (type) {
9137 115 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE:
9138 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_NONE:
9139 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
9140 115 : return true;
9141 0 : default:
9142 0 : break;
9143 : }
9144 0 : return false;
9145 : }
9146 :
9147 : /* Returns true if taking a claim with desc->write == false should make the descriptor writable. */
9148 : static bool
9149 17 : claim_type_promotes_to_write(enum spdk_bdev_claim_type type)
9150 : {
9151 17 : switch (type) {
9152 6 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE:
9153 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
9154 6 : return true;
9155 11 : default:
9156 11 : break;
9157 : }
9158 11 : return false;
9159 : }
9160 :
9161 : void
9162 57 : spdk_bdev_claim_opts_init(struct spdk_bdev_claim_opts *opts, size_t size)
9163 : {
9164 57 : if (opts == NULL) {
9165 0 : SPDK_ERRLOG("opts should not be NULL\n");
9166 0 : assert(opts != NULL);
9167 0 : return;
9168 : }
9169 57 : if (size == 0) {
9170 0 : SPDK_ERRLOG("size should not be zero\n");
9171 0 : assert(size != 0);
9172 0 : return;
9173 : }
9174 :
9175 57 : memset(opts, 0, size);
9176 57 : opts->opts_size = size;
9177 :
9178 : #define FIELD_OK(field) \
9179 : offsetof(struct spdk_bdev_claim_opts, field) + sizeof(opts->field) <= size
9180 :
9181 : #define SET_FIELD(field, value) \
9182 : if (FIELD_OK(field)) { \
9183 : opts->field = value; \
9184 : } \
9185 :
9186 57 : SET_FIELD(shared_claim_key, 0);
9187 :
9188 : #undef FIELD_OK
9189 : #undef SET_FIELD
9190 : }
9191 :
9192 : static int
9193 22 : claim_opts_copy(struct spdk_bdev_claim_opts *src, struct spdk_bdev_claim_opts *dst)
9194 : {
9195 22 : if (src->opts_size == 0) {
9196 0 : SPDK_ERRLOG("size should not be zero\n");
9197 0 : return -1;
9198 : }
9199 :
9200 22 : memset(dst, 0, sizeof(*dst));
9201 22 : dst->opts_size = src->opts_size;
9202 :
9203 : #define FIELD_OK(field) \
9204 : offsetof(struct spdk_bdev_claim_opts, field) + sizeof(src->field) <= src->opts_size
9205 :
9206 : #define SET_FIELD(field) \
9207 : if (FIELD_OK(field)) { \
9208 : dst->field = src->field; \
9209 : } \
9210 :
9211 22 : if (FIELD_OK(name)) {
9212 22 : snprintf(dst->name, sizeof(dst->name), "%s", src->name);
9213 : }
9214 :
9215 22 : SET_FIELD(shared_claim_key);
9216 :
9217 : /* You should not remove this statement, but need to update the assert statement
9218 : * if you add a new field, and also add a corresponding SET_FIELD statement */
9219 : SPDK_STATIC_ASSERT(sizeof(struct spdk_bdev_claim_opts) == 48, "Incorrect size");
9220 :
9221 : #undef FIELD_OK
9222 : #undef SET_FIELD
9223 22 : return 0;
9224 : }
9225 :
9226 : /* Returns 0 if a read-write-once claim can be taken. */
9227 : static int
9228 10 : claim_verify_rwo(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
9229 : struct spdk_bdev_claim_opts *opts, struct spdk_bdev_module *module)
9230 : {
9231 10 : struct spdk_bdev *bdev = desc->bdev;
9232 : struct spdk_bdev_desc *open_desc;
9233 :
9234 10 : assert(spdk_spin_held(&bdev->internal.spinlock));
9235 10 : assert(type == SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE);
9236 :
9237 10 : if (opts->shared_claim_key != 0) {
9238 1 : SPDK_ERRLOG("%s: key option not supported with read-write-once claims\n",
9239 : bdev->name);
9240 1 : return -EINVAL;
9241 : }
9242 9 : if (bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE) {
9243 1 : LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
9244 1 : return -EPERM;
9245 : }
9246 8 : if (desc->claim != NULL) {
9247 0 : SPDK_NOTICELOG("%s: descriptor already claimed bdev with module %s\n",
9248 : bdev->name, desc->claim->module->name);
9249 0 : return -EPERM;
9250 : }
9251 16 : TAILQ_FOREACH(open_desc, &bdev->internal.open_descs, link) {
9252 10 : if (desc != open_desc && open_desc->write) {
9253 2 : SPDK_NOTICELOG("%s: Cannot obtain read-write-once claim while "
9254 : "another descriptor is open for writing\n",
9255 : bdev->name);
9256 2 : return -EPERM;
9257 : }
9258 : }
9259 :
9260 6 : return 0;
9261 : }
9262 :
9263 : /* Returns 0 if a read-only-many claim can be taken. */
9264 : static int
9265 15 : claim_verify_rom(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
9266 : struct spdk_bdev_claim_opts *opts, struct spdk_bdev_module *module)
9267 : {
9268 15 : struct spdk_bdev *bdev = desc->bdev;
9269 : struct spdk_bdev_desc *open_desc;
9270 :
9271 15 : assert(spdk_spin_held(&bdev->internal.spinlock));
9272 15 : assert(type == SPDK_BDEV_CLAIM_READ_MANY_WRITE_NONE);
9273 15 : assert(desc->claim == NULL);
9274 :
9275 15 : if (desc->write) {
9276 3 : SPDK_ERRLOG("%s: Cannot obtain read-only-many claim with writable descriptor\n",
9277 : bdev->name);
9278 3 : return -EINVAL;
9279 : }
9280 12 : if (opts->shared_claim_key != 0) {
9281 1 : SPDK_ERRLOG("%s: key option not supported with read-only-may claims\n", bdev->name);
9282 1 : return -EINVAL;
9283 : }
9284 11 : if (bdev->internal.claim_type == SPDK_BDEV_CLAIM_NONE) {
9285 19 : TAILQ_FOREACH(open_desc, &bdev->internal.open_descs, link) {
9286 11 : if (open_desc->write) {
9287 0 : SPDK_NOTICELOG("%s: Cannot obtain read-only-many claim while "
9288 : "another descriptor is open for writing\n",
9289 : bdev->name);
9290 0 : return -EPERM;
9291 : }
9292 : }
9293 : }
9294 :
9295 11 : return 0;
9296 : }
9297 :
9298 : /* Returns 0 if a read-write-many claim can be taken. */
9299 : static int
9300 8 : claim_verify_rwm(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
9301 : struct spdk_bdev_claim_opts *opts, struct spdk_bdev_module *module)
9302 : {
9303 8 : struct spdk_bdev *bdev = desc->bdev;
9304 : struct spdk_bdev_desc *open_desc;
9305 :
9306 8 : assert(spdk_spin_held(&bdev->internal.spinlock));
9307 8 : assert(type == SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED);
9308 8 : assert(desc->claim == NULL);
9309 :
9310 8 : if (opts->shared_claim_key == 0) {
9311 2 : SPDK_ERRLOG("%s: shared_claim_key option required with read-write-may claims\n",
9312 : bdev->name);
9313 2 : return -EINVAL;
9314 : }
9315 6 : switch (bdev->internal.claim_type) {
9316 4 : case SPDK_BDEV_CLAIM_NONE:
9317 7 : TAILQ_FOREACH(open_desc, &bdev->internal.open_descs, link) {
9318 5 : if (open_desc == desc) {
9319 3 : continue;
9320 : }
9321 2 : if (open_desc->write) {
9322 2 : SPDK_NOTICELOG("%s: Cannot obtain read-write-many claim while "
9323 : "another descriptor is open for writing without a "
9324 : "claim\n", bdev->name);
9325 2 : return -EPERM;
9326 : }
9327 : }
9328 2 : break;
9329 2 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
9330 2 : if (opts->shared_claim_key != bdev->internal.claim.v2.key) {
9331 1 : LOG_ALREADY_CLAIMED_ERROR("already claimed with another key", bdev);
9332 1 : return -EPERM;
9333 : }
9334 1 : break;
9335 0 : default:
9336 0 : LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
9337 0 : return -EBUSY;
9338 : }
9339 :
9340 3 : return 0;
9341 : }
9342 :
9343 : /* Updates desc and its bdev with a v2 claim. */
9344 : static int
9345 20 : claim_bdev(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
9346 : struct spdk_bdev_claim_opts *opts, struct spdk_bdev_module *module)
9347 : {
9348 20 : struct spdk_bdev *bdev = desc->bdev;
9349 : struct spdk_bdev_module_claim *claim;
9350 :
9351 20 : assert(spdk_spin_held(&bdev->internal.spinlock));
9352 20 : assert(claim_type_is_v2(type));
9353 20 : assert(desc->claim == NULL);
9354 :
9355 20 : claim = calloc(1, sizeof(*desc->claim));
9356 20 : if (claim == NULL) {
9357 0 : SPDK_ERRLOG("%s: out of memory while allocating claim\n", bdev->name);
9358 0 : return -ENOMEM;
9359 : }
9360 20 : claim->module = module;
9361 20 : claim->desc = desc;
9362 : SPDK_STATIC_ASSERT(sizeof(claim->name) == sizeof(opts->name), "sizes must match");
9363 20 : memcpy(claim->name, opts->name, sizeof(claim->name));
9364 20 : desc->claim = claim;
9365 :
9366 20 : if (bdev->internal.claim_type == SPDK_BDEV_CLAIM_NONE) {
9367 16 : bdev->internal.claim_type = type;
9368 16 : TAILQ_INIT(&bdev->internal.claim.v2.claims);
9369 16 : bdev->internal.claim.v2.key = opts->shared_claim_key;
9370 : }
9371 20 : assert(type == bdev->internal.claim_type);
9372 :
9373 20 : TAILQ_INSERT_TAIL(&bdev->internal.claim.v2.claims, claim, link);
9374 :
9375 20 : if (!desc->write && claim_type_promotes_to_write(type)) {
9376 6 : desc->write = true;
9377 : }
9378 :
9379 20 : return 0;
9380 : }
9381 :
9382 : int
9383 44 : spdk_bdev_module_claim_bdev_desc(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
9384 : struct spdk_bdev_claim_opts *_opts,
9385 : struct spdk_bdev_module *module)
9386 : {
9387 : struct spdk_bdev *bdev;
9388 44 : struct spdk_bdev_claim_opts opts;
9389 44 : int rc = 0;
9390 :
9391 44 : if (desc == NULL) {
9392 0 : SPDK_ERRLOG("descriptor must not be NULL\n");
9393 0 : return -EINVAL;
9394 : }
9395 :
9396 44 : bdev = desc->bdev;
9397 :
9398 44 : if (_opts == NULL) {
9399 22 : spdk_bdev_claim_opts_init(&opts, sizeof(opts));
9400 22 : } else if (claim_opts_copy(_opts, &opts) != 0) {
9401 0 : return -EINVAL;
9402 : }
9403 :
9404 44 : spdk_spin_lock(&bdev->internal.spinlock);
9405 :
9406 44 : if (bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE &&
9407 17 : bdev->internal.claim_type != type) {
9408 11 : LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
9409 11 : spdk_spin_unlock(&bdev->internal.spinlock);
9410 11 : return -EPERM;
9411 : }
9412 :
9413 33 : if (claim_type_is_v2(type) && desc->claim != NULL) {
9414 0 : SPDK_ERRLOG("%s: descriptor already has %s claim with name '%s'\n",
9415 : bdev->name, spdk_bdev_claim_get_name(type), desc->claim->name);
9416 0 : spdk_spin_unlock(&bdev->internal.spinlock);
9417 0 : return -EPERM;
9418 : }
9419 :
9420 33 : switch (type) {
9421 0 : case SPDK_BDEV_CLAIM_EXCL_WRITE:
9422 0 : spdk_spin_unlock(&bdev->internal.spinlock);
9423 0 : return spdk_bdev_module_claim_bdev(bdev, desc, module);
9424 10 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE:
9425 10 : rc = claim_verify_rwo(desc, type, &opts, module);
9426 10 : break;
9427 15 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_NONE:
9428 15 : rc = claim_verify_rom(desc, type, &opts, module);
9429 15 : break;
9430 8 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
9431 8 : rc = claim_verify_rwm(desc, type, &opts, module);
9432 8 : break;
9433 0 : default:
9434 0 : SPDK_ERRLOG("%s: claim type %d not supported\n", bdev->name, type);
9435 0 : rc = -ENOTSUP;
9436 : }
9437 :
9438 33 : if (rc == 0) {
9439 20 : rc = claim_bdev(desc, type, &opts, module);
9440 : }
9441 :
9442 33 : spdk_spin_unlock(&bdev->internal.spinlock);
9443 33 : return rc;
9444 : }
9445 :
9446 : static void
9447 16 : claim_reset(struct spdk_bdev *bdev)
9448 : {
9449 16 : assert(spdk_spin_held(&bdev->internal.spinlock));
9450 16 : assert(claim_type_is_v2(bdev->internal.claim_type));
9451 16 : assert(TAILQ_EMPTY(&bdev->internal.claim.v2.claims));
9452 :
9453 16 : memset(&bdev->internal.claim, 0, sizeof(bdev->internal.claim));
9454 16 : bdev->internal.claim_type = SPDK_BDEV_CLAIM_NONE;
9455 16 : }
9456 :
9457 : static void
9458 20 : bdev_desc_release_claims(struct spdk_bdev_desc *desc)
9459 : {
9460 20 : struct spdk_bdev *bdev = desc->bdev;
9461 :
9462 20 : assert(spdk_spin_held(&bdev->internal.spinlock));
9463 20 : assert(claim_type_is_v2(bdev->internal.claim_type));
9464 :
9465 20 : if (bdev->internal.examine_in_progress == 0) {
9466 20 : TAILQ_REMOVE(&bdev->internal.claim.v2.claims, desc->claim, link);
9467 20 : free(desc->claim);
9468 20 : if (TAILQ_EMPTY(&bdev->internal.claim.v2.claims)) {
9469 16 : claim_reset(bdev);
9470 : }
9471 : } else {
9472 : /* This is a dead claim that will be cleaned up when bdev_examine() is done. */
9473 0 : desc->claim->module = NULL;
9474 0 : desc->claim->desc = NULL;
9475 : }
9476 20 : desc->claim = NULL;
9477 20 : }
9478 :
9479 : /*
9480 : * End claims v2
9481 : */
9482 :
9483 : struct spdk_bdev *
9484 1554 : spdk_bdev_desc_get_bdev(struct spdk_bdev_desc *desc)
9485 : {
9486 1554 : assert(desc != NULL);
9487 1554 : return desc->bdev;
9488 : }
9489 :
9490 : int
9491 1 : spdk_for_each_bdev(void *ctx, spdk_for_each_bdev_fn fn)
9492 : {
9493 : struct spdk_bdev *bdev, *tmp;
9494 1 : struct spdk_bdev_desc *desc;
9495 1 : int rc = 0;
9496 :
9497 1 : assert(fn != NULL);
9498 :
9499 1 : spdk_spin_lock(&g_bdev_mgr.spinlock);
9500 1 : bdev = spdk_bdev_first();
9501 9 : while (bdev != NULL) {
9502 8 : rc = bdev_desc_alloc(bdev, _tmp_bdev_event_cb, NULL, NULL, &desc);
9503 8 : if (rc != 0) {
9504 0 : break;
9505 : }
9506 8 : rc = bdev_open(bdev, false, desc);
9507 8 : if (rc != 0) {
9508 1 : bdev_desc_free(desc);
9509 1 : if (rc == -ENODEV) {
9510 : /* Ignore the error and move to the next bdev. */
9511 1 : rc = 0;
9512 1 : bdev = spdk_bdev_next(bdev);
9513 1 : continue;
9514 : }
9515 0 : break;
9516 : }
9517 7 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
9518 :
9519 7 : rc = fn(ctx, bdev);
9520 :
9521 7 : spdk_spin_lock(&g_bdev_mgr.spinlock);
9522 7 : tmp = spdk_bdev_next(bdev);
9523 7 : bdev_close(bdev, desc);
9524 7 : if (rc != 0) {
9525 0 : break;
9526 : }
9527 7 : bdev = tmp;
9528 : }
9529 1 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
9530 :
9531 1 : return rc;
9532 : }
9533 :
9534 : int
9535 1 : spdk_for_each_bdev_leaf(void *ctx, spdk_for_each_bdev_fn fn)
9536 : {
9537 : struct spdk_bdev *bdev, *tmp;
9538 1 : struct spdk_bdev_desc *desc;
9539 1 : int rc = 0;
9540 :
9541 1 : assert(fn != NULL);
9542 :
9543 1 : spdk_spin_lock(&g_bdev_mgr.spinlock);
9544 1 : bdev = spdk_bdev_first_leaf();
9545 6 : while (bdev != NULL) {
9546 5 : rc = bdev_desc_alloc(bdev, _tmp_bdev_event_cb, NULL, NULL, &desc);
9547 5 : if (rc != 0) {
9548 0 : break;
9549 : }
9550 5 : rc = bdev_open(bdev, false, desc);
9551 5 : if (rc != 0) {
9552 1 : bdev_desc_free(desc);
9553 1 : if (rc == -ENODEV) {
9554 : /* Ignore the error and move to the next bdev. */
9555 1 : rc = 0;
9556 1 : bdev = spdk_bdev_next_leaf(bdev);
9557 1 : continue;
9558 : }
9559 0 : break;
9560 : }
9561 4 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
9562 :
9563 4 : rc = fn(ctx, bdev);
9564 :
9565 4 : spdk_spin_lock(&g_bdev_mgr.spinlock);
9566 4 : tmp = spdk_bdev_next_leaf(bdev);
9567 4 : bdev_close(bdev, desc);
9568 4 : if (rc != 0) {
9569 0 : break;
9570 : }
9571 4 : bdev = tmp;
9572 : }
9573 1 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
9574 :
9575 1 : return rc;
9576 : }
9577 :
9578 : void
9579 0 : spdk_bdev_io_get_iovec(struct spdk_bdev_io *bdev_io, struct iovec **iovp, int *iovcntp)
9580 : {
9581 : struct iovec *iovs;
9582 : int iovcnt;
9583 :
9584 0 : if (bdev_io == NULL) {
9585 0 : return;
9586 : }
9587 :
9588 0 : switch (bdev_io->type) {
9589 0 : case SPDK_BDEV_IO_TYPE_READ:
9590 : case SPDK_BDEV_IO_TYPE_WRITE:
9591 : case SPDK_BDEV_IO_TYPE_ZCOPY:
9592 0 : iovs = bdev_io->u.bdev.iovs;
9593 0 : iovcnt = bdev_io->u.bdev.iovcnt;
9594 0 : break;
9595 0 : default:
9596 0 : iovs = NULL;
9597 0 : iovcnt = 0;
9598 0 : break;
9599 : }
9600 :
9601 0 : if (iovp) {
9602 0 : *iovp = iovs;
9603 : }
9604 0 : if (iovcntp) {
9605 0 : *iovcntp = iovcnt;
9606 : }
9607 : }
9608 :
9609 : void *
9610 0 : spdk_bdev_io_get_md_buf(struct spdk_bdev_io *bdev_io)
9611 : {
9612 0 : if (bdev_io == NULL) {
9613 0 : return NULL;
9614 : }
9615 :
9616 0 : if (!spdk_bdev_is_md_separate(bdev_io->bdev)) {
9617 0 : return NULL;
9618 : }
9619 :
9620 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_READ ||
9621 0 : bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
9622 0 : return bdev_io->u.bdev.md_buf;
9623 : }
9624 :
9625 0 : return NULL;
9626 : }
9627 :
9628 : void *
9629 0 : spdk_bdev_io_get_cb_arg(struct spdk_bdev_io *bdev_io)
9630 : {
9631 0 : if (bdev_io == NULL) {
9632 0 : assert(false);
9633 : return NULL;
9634 : }
9635 :
9636 0 : return bdev_io->internal.caller_ctx;
9637 : }
9638 :
9639 : void
9640 7 : spdk_bdev_module_list_add(struct spdk_bdev_module *bdev_module)
9641 : {
9642 :
9643 7 : if (spdk_bdev_module_list_find(bdev_module->name)) {
9644 0 : SPDK_ERRLOG("ERROR: module '%s' already registered.\n", bdev_module->name);
9645 0 : assert(false);
9646 : }
9647 :
9648 7 : spdk_spin_init(&bdev_module->internal.spinlock);
9649 7 : TAILQ_INIT(&bdev_module->internal.quiesced_ranges);
9650 :
9651 : /*
9652 : * Modules with examine callbacks must be initialized first, so they are
9653 : * ready to handle examine callbacks from later modules that will
9654 : * register physical bdevs.
9655 : */
9656 7 : if (bdev_module->examine_config != NULL || bdev_module->examine_disk != NULL) {
9657 4 : TAILQ_INSERT_HEAD(&g_bdev_mgr.bdev_modules, bdev_module, internal.tailq);
9658 : } else {
9659 3 : TAILQ_INSERT_TAIL(&g_bdev_mgr.bdev_modules, bdev_module, internal.tailq);
9660 : }
9661 7 : }
9662 :
9663 : struct spdk_bdev_module *
9664 7 : spdk_bdev_module_list_find(const char *name)
9665 : {
9666 : struct spdk_bdev_module *bdev_module;
9667 :
9668 14 : TAILQ_FOREACH(bdev_module, &g_bdev_mgr.bdev_modules, internal.tailq) {
9669 7 : if (strcmp(name, bdev_module->name) == 0) {
9670 0 : break;
9671 : }
9672 : }
9673 :
9674 7 : return bdev_module;
9675 : }
9676 :
9677 : static int
9678 6 : bdev_write_zero_buffer(struct spdk_bdev_io *bdev_io)
9679 : {
9680 : uint64_t num_blocks;
9681 6 : void *md_buf = NULL;
9682 :
9683 6 : num_blocks = bdev_io->u.bdev.num_blocks;
9684 :
9685 6 : if (spdk_bdev_is_md_separate(bdev_io->bdev)) {
9686 2 : md_buf = (char *)g_bdev_mgr.zero_buffer +
9687 2 : spdk_bdev_get_block_size(bdev_io->bdev) * num_blocks;
9688 : }
9689 :
9690 12 : return bdev_write_blocks_with_md(bdev_io->internal.desc,
9691 6 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
9692 : g_bdev_mgr.zero_buffer, md_buf,
9693 : bdev_io->u.bdev.offset_blocks, num_blocks,
9694 : bdev_write_zero_buffer_done, bdev_io);
9695 : }
9696 :
9697 : static void
9698 6 : bdev_write_zero_buffer_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
9699 : {
9700 6 : struct spdk_bdev_io *parent_io = cb_arg;
9701 :
9702 6 : spdk_bdev_free_io(bdev_io);
9703 :
9704 6 : parent_io->internal.status = success ? SPDK_BDEV_IO_STATUS_SUCCESS : SPDK_BDEV_IO_STATUS_FAILED;
9705 6 : parent_io->internal.cb(parent_io, success, parent_io->internal.caller_ctx);
9706 6 : }
9707 :
9708 : static void
9709 10 : bdev_set_qos_limit_done(struct set_qos_limit_ctx *ctx, int status)
9710 : {
9711 10 : spdk_spin_lock(&ctx->bdev->internal.spinlock);
9712 10 : ctx->bdev->internal.qos_mod_in_progress = false;
9713 10 : spdk_spin_unlock(&ctx->bdev->internal.spinlock);
9714 :
9715 10 : if (ctx->cb_fn) {
9716 8 : ctx->cb_fn(ctx->cb_arg, status);
9717 : }
9718 10 : free(ctx);
9719 10 : }
9720 :
9721 : static void
9722 2 : bdev_disable_qos_done(void *cb_arg)
9723 : {
9724 2 : struct set_qos_limit_ctx *ctx = cb_arg;
9725 2 : struct spdk_bdev *bdev = ctx->bdev;
9726 : struct spdk_bdev_qos *qos;
9727 :
9728 2 : spdk_spin_lock(&bdev->internal.spinlock);
9729 2 : qos = bdev->internal.qos;
9730 2 : bdev->internal.qos = NULL;
9731 2 : spdk_spin_unlock(&bdev->internal.spinlock);
9732 :
9733 2 : if (qos->thread != NULL) {
9734 2 : spdk_put_io_channel(spdk_io_channel_from_ctx(qos->ch));
9735 2 : spdk_poller_unregister(&qos->poller);
9736 : }
9737 :
9738 2 : free(qos);
9739 :
9740 2 : bdev_set_qos_limit_done(ctx, 0);
9741 2 : }
9742 :
9743 : static void
9744 2 : bdev_disable_qos_msg_done(struct spdk_bdev *bdev, void *_ctx, int status)
9745 : {
9746 2 : struct set_qos_limit_ctx *ctx = _ctx;
9747 : struct spdk_thread *thread;
9748 :
9749 2 : spdk_spin_lock(&bdev->internal.spinlock);
9750 2 : thread = bdev->internal.qos->thread;
9751 2 : spdk_spin_unlock(&bdev->internal.spinlock);
9752 :
9753 2 : if (thread != NULL) {
9754 2 : spdk_thread_send_msg(thread, bdev_disable_qos_done, ctx);
9755 : } else {
9756 0 : bdev_disable_qos_done(ctx);
9757 : }
9758 2 : }
9759 :
9760 : static void
9761 4 : bdev_disable_qos_msg(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
9762 : struct spdk_io_channel *ch, void *_ctx)
9763 : {
9764 4 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(ch);
9765 : struct spdk_bdev_io *bdev_io;
9766 :
9767 4 : bdev_ch->flags &= ~BDEV_CH_QOS_ENABLED;
9768 :
9769 6 : while (!TAILQ_EMPTY(&bdev_ch->qos_queued_io)) {
9770 : /* Re-submit the queued I/O. */
9771 2 : bdev_io = TAILQ_FIRST(&bdev_ch->qos_queued_io);
9772 2 : TAILQ_REMOVE(&bdev_ch->qos_queued_io, bdev_io, internal.link);
9773 2 : _bdev_io_submit(bdev_io);
9774 : }
9775 :
9776 4 : spdk_bdev_for_each_channel_continue(i, 0);
9777 4 : }
9778 :
9779 : static void
9780 1 : bdev_update_qos_rate_limit_msg(void *cb_arg)
9781 : {
9782 1 : struct set_qos_limit_ctx *ctx = cb_arg;
9783 1 : struct spdk_bdev *bdev = ctx->bdev;
9784 :
9785 1 : spdk_spin_lock(&bdev->internal.spinlock);
9786 1 : bdev_qos_update_max_quota_per_timeslice(bdev->internal.qos);
9787 1 : spdk_spin_unlock(&bdev->internal.spinlock);
9788 :
9789 1 : bdev_set_qos_limit_done(ctx, 0);
9790 1 : }
9791 :
9792 : static void
9793 9 : bdev_enable_qos_msg(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
9794 : struct spdk_io_channel *ch, void *_ctx)
9795 : {
9796 9 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(ch);
9797 :
9798 9 : spdk_spin_lock(&bdev->internal.spinlock);
9799 9 : bdev_enable_qos(bdev, bdev_ch);
9800 9 : spdk_spin_unlock(&bdev->internal.spinlock);
9801 9 : spdk_bdev_for_each_channel_continue(i, 0);
9802 9 : }
9803 :
9804 : static void
9805 6 : bdev_enable_qos_done(struct spdk_bdev *bdev, void *_ctx, int status)
9806 : {
9807 6 : struct set_qos_limit_ctx *ctx = _ctx;
9808 :
9809 6 : bdev_set_qos_limit_done(ctx, status);
9810 6 : }
9811 :
9812 : static void
9813 7 : bdev_set_qos_rate_limits(struct spdk_bdev *bdev, uint64_t *limits)
9814 : {
9815 : int i;
9816 :
9817 7 : assert(bdev->internal.qos != NULL);
9818 :
9819 35 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
9820 28 : if (limits[i] != SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
9821 28 : bdev->internal.qos->rate_limits[i].limit = limits[i];
9822 :
9823 28 : if (limits[i] == 0) {
9824 19 : bdev->internal.qos->rate_limits[i].limit =
9825 : SPDK_BDEV_QOS_LIMIT_NOT_DEFINED;
9826 : }
9827 : }
9828 : }
9829 7 : }
9830 :
9831 : void
9832 9 : spdk_bdev_set_qos_rate_limits(struct spdk_bdev *bdev, uint64_t *limits,
9833 : void (*cb_fn)(void *cb_arg, int status), void *cb_arg)
9834 : {
9835 : struct set_qos_limit_ctx *ctx;
9836 : uint32_t limit_set_complement;
9837 : uint64_t min_limit_per_sec;
9838 : int i;
9839 9 : bool disable_rate_limit = true;
9840 :
9841 45 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
9842 36 : if (limits[i] == SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
9843 0 : continue;
9844 : }
9845 :
9846 36 : if (limits[i] > 0) {
9847 10 : disable_rate_limit = false;
9848 : }
9849 :
9850 36 : if (bdev_qos_is_iops_rate_limit(i) == true) {
9851 9 : min_limit_per_sec = SPDK_BDEV_QOS_MIN_IOS_PER_SEC;
9852 : } else {
9853 27 : if (limits[i] > SPDK_BDEV_QOS_MAX_MBYTES_PER_SEC) {
9854 0 : SPDK_WARNLOG("Requested rate limit %" PRIu64 " will result in uint64_t overflow, "
9855 : "reset to %" PRIu64 "\n", limits[i], SPDK_BDEV_QOS_MAX_MBYTES_PER_SEC);
9856 0 : limits[i] = SPDK_BDEV_QOS_MAX_MBYTES_PER_SEC;
9857 : }
9858 : /* Change from megabyte to byte rate limit */
9859 27 : limits[i] = limits[i] * 1024 * 1024;
9860 27 : min_limit_per_sec = SPDK_BDEV_QOS_MIN_BYTES_PER_SEC;
9861 : }
9862 :
9863 36 : limit_set_complement = limits[i] % min_limit_per_sec;
9864 36 : if (limit_set_complement) {
9865 0 : SPDK_ERRLOG("Requested rate limit %" PRIu64 " is not a multiple of %" PRIu64 "\n",
9866 : limits[i], min_limit_per_sec);
9867 0 : limits[i] += min_limit_per_sec - limit_set_complement;
9868 0 : SPDK_ERRLOG("Round up the rate limit to %" PRIu64 "\n", limits[i]);
9869 : }
9870 : }
9871 :
9872 9 : ctx = calloc(1, sizeof(*ctx));
9873 9 : if (ctx == NULL) {
9874 0 : cb_fn(cb_arg, -ENOMEM);
9875 0 : return;
9876 : }
9877 :
9878 9 : ctx->cb_fn = cb_fn;
9879 9 : ctx->cb_arg = cb_arg;
9880 9 : ctx->bdev = bdev;
9881 :
9882 9 : spdk_spin_lock(&bdev->internal.spinlock);
9883 9 : if (bdev->internal.qos_mod_in_progress) {
9884 1 : spdk_spin_unlock(&bdev->internal.spinlock);
9885 1 : free(ctx);
9886 1 : cb_fn(cb_arg, -EAGAIN);
9887 1 : return;
9888 : }
9889 8 : bdev->internal.qos_mod_in_progress = true;
9890 :
9891 8 : if (disable_rate_limit == true && bdev->internal.qos) {
9892 10 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
9893 8 : if (limits[i] == SPDK_BDEV_QOS_LIMIT_NOT_DEFINED &&
9894 0 : (bdev->internal.qos->rate_limits[i].limit > 0 &&
9895 0 : bdev->internal.qos->rate_limits[i].limit !=
9896 : SPDK_BDEV_QOS_LIMIT_NOT_DEFINED)) {
9897 0 : disable_rate_limit = false;
9898 0 : break;
9899 : }
9900 : }
9901 : }
9902 :
9903 8 : if (disable_rate_limit == false) {
9904 5 : if (bdev->internal.qos == NULL) {
9905 4 : bdev->internal.qos = calloc(1, sizeof(*bdev->internal.qos));
9906 4 : if (!bdev->internal.qos) {
9907 0 : spdk_spin_unlock(&bdev->internal.spinlock);
9908 0 : SPDK_ERRLOG("Unable to allocate memory for QoS tracking\n");
9909 0 : bdev_set_qos_limit_done(ctx, -ENOMEM);
9910 0 : return;
9911 : }
9912 : }
9913 :
9914 5 : if (bdev->internal.qos->thread == NULL) {
9915 : /* Enabling */
9916 4 : bdev_set_qos_rate_limits(bdev, limits);
9917 :
9918 4 : spdk_bdev_for_each_channel(bdev, bdev_enable_qos_msg, ctx,
9919 : bdev_enable_qos_done);
9920 : } else {
9921 : /* Updating */
9922 1 : bdev_set_qos_rate_limits(bdev, limits);
9923 :
9924 1 : spdk_thread_send_msg(bdev->internal.qos->thread,
9925 : bdev_update_qos_rate_limit_msg, ctx);
9926 : }
9927 : } else {
9928 3 : if (bdev->internal.qos != NULL) {
9929 2 : bdev_set_qos_rate_limits(bdev, limits);
9930 :
9931 : /* Disabling */
9932 2 : spdk_bdev_for_each_channel(bdev, bdev_disable_qos_msg, ctx,
9933 : bdev_disable_qos_msg_done);
9934 : } else {
9935 1 : spdk_spin_unlock(&bdev->internal.spinlock);
9936 1 : bdev_set_qos_limit_done(ctx, 0);
9937 1 : return;
9938 : }
9939 : }
9940 :
9941 7 : spdk_spin_unlock(&bdev->internal.spinlock);
9942 : }
9943 :
9944 : struct spdk_bdev_histogram_ctx {
9945 : spdk_bdev_histogram_status_cb cb_fn;
9946 : void *cb_arg;
9947 : struct spdk_bdev *bdev;
9948 : int status;
9949 : };
9950 :
9951 : static void
9952 2 : bdev_histogram_disable_channel_cb(struct spdk_bdev *bdev, void *_ctx, int status)
9953 : {
9954 2 : struct spdk_bdev_histogram_ctx *ctx = _ctx;
9955 :
9956 2 : spdk_spin_lock(&ctx->bdev->internal.spinlock);
9957 2 : ctx->bdev->internal.histogram_in_progress = false;
9958 2 : spdk_spin_unlock(&ctx->bdev->internal.spinlock);
9959 2 : ctx->cb_fn(ctx->cb_arg, ctx->status);
9960 2 : free(ctx);
9961 2 : }
9962 :
9963 : static void
9964 3 : bdev_histogram_disable_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
9965 : struct spdk_io_channel *_ch, void *_ctx)
9966 : {
9967 3 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
9968 :
9969 3 : if (ch->histogram != NULL) {
9970 3 : spdk_histogram_data_free(ch->histogram);
9971 3 : ch->histogram = NULL;
9972 : }
9973 3 : spdk_bdev_for_each_channel_continue(i, 0);
9974 3 : }
9975 :
9976 : static void
9977 2 : bdev_histogram_enable_channel_cb(struct spdk_bdev *bdev, void *_ctx, int status)
9978 : {
9979 2 : struct spdk_bdev_histogram_ctx *ctx = _ctx;
9980 :
9981 2 : if (status != 0) {
9982 0 : ctx->status = status;
9983 0 : ctx->bdev->internal.histogram_enabled = false;
9984 0 : spdk_bdev_for_each_channel(ctx->bdev, bdev_histogram_disable_channel, ctx,
9985 : bdev_histogram_disable_channel_cb);
9986 : } else {
9987 2 : spdk_spin_lock(&ctx->bdev->internal.spinlock);
9988 2 : ctx->bdev->internal.histogram_in_progress = false;
9989 2 : spdk_spin_unlock(&ctx->bdev->internal.spinlock);
9990 2 : ctx->cb_fn(ctx->cb_arg, ctx->status);
9991 2 : free(ctx);
9992 : }
9993 2 : }
9994 :
9995 : static void
9996 3 : bdev_histogram_enable_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
9997 : struct spdk_io_channel *_ch, void *_ctx)
9998 : {
9999 3 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
10000 3 : int status = 0;
10001 :
10002 3 : if (ch->histogram == NULL) {
10003 3 : ch->histogram = spdk_histogram_data_alloc();
10004 3 : if (ch->histogram == NULL) {
10005 0 : status = -ENOMEM;
10006 : }
10007 : }
10008 :
10009 3 : spdk_bdev_for_each_channel_continue(i, status);
10010 3 : }
10011 :
10012 : void
10013 4 : spdk_bdev_histogram_enable_ext(struct spdk_bdev *bdev, spdk_bdev_histogram_status_cb cb_fn,
10014 : void *cb_arg, bool enable, struct spdk_bdev_enable_histogram_opts *opts)
10015 : {
10016 : struct spdk_bdev_histogram_ctx *ctx;
10017 :
10018 4 : ctx = calloc(1, sizeof(struct spdk_bdev_histogram_ctx));
10019 4 : if (ctx == NULL) {
10020 0 : cb_fn(cb_arg, -ENOMEM);
10021 0 : return;
10022 : }
10023 :
10024 4 : ctx->bdev = bdev;
10025 4 : ctx->status = 0;
10026 4 : ctx->cb_fn = cb_fn;
10027 4 : ctx->cb_arg = cb_arg;
10028 :
10029 4 : spdk_spin_lock(&bdev->internal.spinlock);
10030 4 : if (bdev->internal.histogram_in_progress) {
10031 0 : spdk_spin_unlock(&bdev->internal.spinlock);
10032 0 : free(ctx);
10033 0 : cb_fn(cb_arg, -EAGAIN);
10034 0 : return;
10035 : }
10036 :
10037 4 : bdev->internal.histogram_in_progress = true;
10038 4 : spdk_spin_unlock(&bdev->internal.spinlock);
10039 :
10040 4 : bdev->internal.histogram_enabled = enable;
10041 4 : bdev->internal.histogram_io_type = opts->io_type;
10042 :
10043 4 : if (enable) {
10044 : /* Allocate histogram for each channel */
10045 2 : spdk_bdev_for_each_channel(bdev, bdev_histogram_enable_channel, ctx,
10046 : bdev_histogram_enable_channel_cb);
10047 : } else {
10048 2 : spdk_bdev_for_each_channel(bdev, bdev_histogram_disable_channel, ctx,
10049 : bdev_histogram_disable_channel_cb);
10050 : }
10051 : }
10052 :
10053 : void
10054 4 : spdk_bdev_enable_histogram_opts_init(struct spdk_bdev_enable_histogram_opts *opts, size_t size)
10055 : {
10056 4 : if (opts == NULL) {
10057 0 : SPDK_ERRLOG("opts should not be NULL\n");
10058 0 : assert(opts != NULL);
10059 0 : return;
10060 : }
10061 4 : if (size == 0) {
10062 0 : SPDK_ERRLOG("size should not be zero\n");
10063 0 : assert(size != 0);
10064 0 : return;
10065 : }
10066 :
10067 4 : memset(opts, 0, size);
10068 4 : opts->size = size;
10069 :
10070 : #define FIELD_OK(field) \
10071 : offsetof(struct spdk_bdev_enable_histogram_opts, field) + sizeof(opts->field) <= size
10072 :
10073 : #define SET_FIELD(field, value) \
10074 : if (FIELD_OK(field)) { \
10075 : opts->field = value; \
10076 : } \
10077 :
10078 4 : SET_FIELD(io_type, 0);
10079 :
10080 : /* You should not remove this statement, but need to update the assert statement
10081 : * if you add a new field, and also add a corresponding SET_FIELD statement */
10082 : SPDK_STATIC_ASSERT(sizeof(struct spdk_bdev_enable_histogram_opts) == 9, "Incorrect size");
10083 :
10084 : #undef FIELD_OK
10085 : #undef SET_FIELD
10086 : }
10087 :
10088 : void
10089 4 : spdk_bdev_histogram_enable(struct spdk_bdev *bdev, spdk_bdev_histogram_status_cb cb_fn,
10090 : void *cb_arg, bool enable)
10091 : {
10092 4 : struct spdk_bdev_enable_histogram_opts opts;
10093 :
10094 4 : spdk_bdev_enable_histogram_opts_init(&opts, sizeof(opts));
10095 4 : spdk_bdev_histogram_enable_ext(bdev, cb_fn, cb_arg, enable, &opts);
10096 4 : }
10097 :
10098 : struct spdk_bdev_histogram_data_ctx {
10099 : spdk_bdev_histogram_data_cb cb_fn;
10100 : void *cb_arg;
10101 : struct spdk_bdev *bdev;
10102 : /** merged histogram data from all channels */
10103 : struct spdk_histogram_data *histogram;
10104 : };
10105 :
10106 : static void
10107 5 : bdev_histogram_get_channel_cb(struct spdk_bdev *bdev, void *_ctx, int status)
10108 : {
10109 5 : struct spdk_bdev_histogram_data_ctx *ctx = _ctx;
10110 :
10111 5 : ctx->cb_fn(ctx->cb_arg, status, ctx->histogram);
10112 5 : free(ctx);
10113 5 : }
10114 :
10115 : static void
10116 7 : bdev_histogram_get_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
10117 : struct spdk_io_channel *_ch, void *_ctx)
10118 : {
10119 7 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
10120 7 : struct spdk_bdev_histogram_data_ctx *ctx = _ctx;
10121 7 : int status = 0;
10122 :
10123 7 : if (ch->histogram == NULL) {
10124 1 : status = -EFAULT;
10125 : } else {
10126 6 : spdk_histogram_data_merge(ctx->histogram, ch->histogram);
10127 : }
10128 :
10129 7 : spdk_bdev_for_each_channel_continue(i, status);
10130 7 : }
10131 :
10132 : void
10133 5 : spdk_bdev_histogram_get(struct spdk_bdev *bdev, struct spdk_histogram_data *histogram,
10134 : spdk_bdev_histogram_data_cb cb_fn,
10135 : void *cb_arg)
10136 : {
10137 : struct spdk_bdev_histogram_data_ctx *ctx;
10138 :
10139 5 : ctx = calloc(1, sizeof(struct spdk_bdev_histogram_data_ctx));
10140 5 : if (ctx == NULL) {
10141 0 : cb_fn(cb_arg, -ENOMEM, NULL);
10142 0 : return;
10143 : }
10144 :
10145 5 : ctx->bdev = bdev;
10146 5 : ctx->cb_fn = cb_fn;
10147 5 : ctx->cb_arg = cb_arg;
10148 :
10149 5 : ctx->histogram = histogram;
10150 :
10151 5 : spdk_bdev_for_each_channel(bdev, bdev_histogram_get_channel, ctx,
10152 : bdev_histogram_get_channel_cb);
10153 : }
10154 :
10155 : void
10156 2 : spdk_bdev_channel_get_histogram(struct spdk_io_channel *ch, spdk_bdev_histogram_data_cb cb_fn,
10157 : void *cb_arg)
10158 : {
10159 2 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(ch);
10160 2 : int status = 0;
10161 :
10162 2 : assert(cb_fn != NULL);
10163 :
10164 2 : if (bdev_ch->histogram == NULL) {
10165 1 : status = -EFAULT;
10166 : }
10167 2 : cb_fn(cb_arg, status, bdev_ch->histogram);
10168 2 : }
10169 :
10170 : size_t
10171 0 : spdk_bdev_get_media_events(struct spdk_bdev_desc *desc, struct spdk_bdev_media_event *events,
10172 : size_t max_events)
10173 : {
10174 : struct media_event_entry *entry;
10175 0 : size_t num_events = 0;
10176 :
10177 0 : for (; num_events < max_events; ++num_events) {
10178 0 : entry = TAILQ_FIRST(&desc->pending_media_events);
10179 0 : if (entry == NULL) {
10180 0 : break;
10181 : }
10182 :
10183 0 : events[num_events] = entry->event;
10184 0 : TAILQ_REMOVE(&desc->pending_media_events, entry, tailq);
10185 0 : TAILQ_INSERT_TAIL(&desc->free_media_events, entry, tailq);
10186 : }
10187 :
10188 0 : return num_events;
10189 : }
10190 :
10191 : int
10192 0 : spdk_bdev_push_media_events(struct spdk_bdev *bdev, const struct spdk_bdev_media_event *events,
10193 : size_t num_events)
10194 : {
10195 : struct spdk_bdev_desc *desc;
10196 : struct media_event_entry *entry;
10197 : size_t event_id;
10198 0 : int rc = 0;
10199 :
10200 0 : assert(bdev->media_events);
10201 :
10202 0 : spdk_spin_lock(&bdev->internal.spinlock);
10203 0 : TAILQ_FOREACH(desc, &bdev->internal.open_descs, link) {
10204 0 : if (desc->write) {
10205 0 : break;
10206 : }
10207 : }
10208 :
10209 0 : if (desc == NULL || desc->media_events_buffer == NULL) {
10210 0 : rc = -ENODEV;
10211 0 : goto out;
10212 : }
10213 :
10214 0 : for (event_id = 0; event_id < num_events; ++event_id) {
10215 0 : entry = TAILQ_FIRST(&desc->free_media_events);
10216 0 : if (entry == NULL) {
10217 0 : break;
10218 : }
10219 :
10220 0 : TAILQ_REMOVE(&desc->free_media_events, entry, tailq);
10221 0 : TAILQ_INSERT_TAIL(&desc->pending_media_events, entry, tailq);
10222 0 : entry->event = events[event_id];
10223 : }
10224 :
10225 0 : rc = event_id;
10226 0 : out:
10227 0 : spdk_spin_unlock(&bdev->internal.spinlock);
10228 0 : return rc;
10229 : }
10230 :
10231 : static void
10232 0 : _media_management_notify(void *arg)
10233 : {
10234 0 : struct spdk_bdev_desc *desc = arg;
10235 :
10236 0 : _event_notify(desc, SPDK_BDEV_EVENT_MEDIA_MANAGEMENT);
10237 0 : }
10238 :
10239 : void
10240 0 : spdk_bdev_notify_media_management(struct spdk_bdev *bdev)
10241 : {
10242 : struct spdk_bdev_desc *desc;
10243 :
10244 0 : spdk_spin_lock(&bdev->internal.spinlock);
10245 0 : TAILQ_FOREACH(desc, &bdev->internal.open_descs, link) {
10246 0 : if (!TAILQ_EMPTY(&desc->pending_media_events)) {
10247 0 : event_notify(desc, _media_management_notify);
10248 : }
10249 : }
10250 0 : spdk_spin_unlock(&bdev->internal.spinlock);
10251 0 : }
10252 :
10253 : struct locked_lba_range_ctx {
10254 : struct lba_range range;
10255 : struct lba_range *current_range;
10256 : struct lba_range *owner_range;
10257 : struct spdk_poller *poller;
10258 : lock_range_cb cb_fn;
10259 : void *cb_arg;
10260 : };
10261 :
10262 : static void
10263 0 : bdev_lock_error_cleanup_cb(struct spdk_bdev *bdev, void *_ctx, int status)
10264 : {
10265 0 : struct locked_lba_range_ctx *ctx = _ctx;
10266 :
10267 0 : ctx->cb_fn(&ctx->range, ctx->cb_arg, -ENOMEM);
10268 0 : free(ctx);
10269 0 : }
10270 :
10271 : static void bdev_unlock_lba_range_get_channel(struct spdk_bdev_channel_iter *i,
10272 : struct spdk_bdev *bdev, struct spdk_io_channel *ch, void *_ctx);
10273 :
10274 : static void
10275 14 : bdev_lock_lba_range_cb(struct spdk_bdev *bdev, void *_ctx, int status)
10276 : {
10277 14 : struct locked_lba_range_ctx *ctx = _ctx;
10278 :
10279 14 : if (status == -ENOMEM) {
10280 : /* One of the channels could not allocate a range object.
10281 : * So we have to go back and clean up any ranges that were
10282 : * allocated successfully before we return error status to
10283 : * the caller. We can reuse the unlock function to do that
10284 : * clean up.
10285 : */
10286 0 : spdk_bdev_for_each_channel(bdev, bdev_unlock_lba_range_get_channel, ctx,
10287 : bdev_lock_error_cleanup_cb);
10288 0 : return;
10289 : }
10290 :
10291 : /* All channels have locked this range and no I/O overlapping the range
10292 : * are outstanding! Set the owner_ch for the range object for the
10293 : * locking channel, so that this channel will know that it is allowed
10294 : * to write to this range.
10295 : */
10296 14 : if (ctx->owner_range != NULL) {
10297 10 : ctx->owner_range->owner_ch = ctx->range.owner_ch;
10298 : }
10299 :
10300 14 : ctx->cb_fn(&ctx->range, ctx->cb_arg, status);
10301 :
10302 : /* Don't free the ctx here. Its range is in the bdev's global list of
10303 : * locked ranges still, and will be removed and freed when this range
10304 : * is later unlocked.
10305 : */
10306 : }
10307 :
10308 : static int
10309 17 : bdev_lock_lba_range_check_io(void *_i)
10310 : {
10311 17 : struct spdk_bdev_channel_iter *i = _i;
10312 17 : struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i->i);
10313 17 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
10314 17 : struct locked_lba_range_ctx *ctx = i->ctx;
10315 17 : struct lba_range *range = ctx->current_range;
10316 : struct spdk_bdev_io *bdev_io;
10317 :
10318 17 : spdk_poller_unregister(&ctx->poller);
10319 :
10320 : /* The range is now in the locked_ranges, so no new IO can be submitted to this
10321 : * range. But we need to wait until any outstanding IO overlapping with this range
10322 : * are completed.
10323 : */
10324 18 : TAILQ_FOREACH(bdev_io, &ch->io_submitted, internal.ch_link) {
10325 3 : if (bdev_io_range_is_locked(bdev_io, range)) {
10326 2 : ctx->poller = SPDK_POLLER_REGISTER(bdev_lock_lba_range_check_io, i, 100);
10327 2 : return SPDK_POLLER_BUSY;
10328 : }
10329 : }
10330 :
10331 15 : spdk_bdev_for_each_channel_continue(i, 0);
10332 15 : return SPDK_POLLER_BUSY;
10333 : }
10334 :
10335 : static void
10336 15 : bdev_lock_lba_range_get_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
10337 : struct spdk_io_channel *_ch, void *_ctx)
10338 : {
10339 15 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
10340 15 : struct locked_lba_range_ctx *ctx = _ctx;
10341 : struct lba_range *range;
10342 :
10343 16 : TAILQ_FOREACH(range, &ch->locked_ranges, tailq) {
10344 1 : if (range->length == ctx->range.length &&
10345 0 : range->offset == ctx->range.offset &&
10346 0 : range->locked_ctx == ctx->range.locked_ctx) {
10347 : /* This range already exists on this channel, so don't add
10348 : * it again. This can happen when a new channel is created
10349 : * while the for_each_channel operation is in progress.
10350 : * Do not check for outstanding I/O in that case, since the
10351 : * range was locked before any I/O could be submitted to the
10352 : * new channel.
10353 : */
10354 0 : spdk_bdev_for_each_channel_continue(i, 0);
10355 0 : return;
10356 : }
10357 : }
10358 :
10359 15 : range = calloc(1, sizeof(*range));
10360 15 : if (range == NULL) {
10361 0 : spdk_bdev_for_each_channel_continue(i, -ENOMEM);
10362 0 : return;
10363 : }
10364 :
10365 15 : range->length = ctx->range.length;
10366 15 : range->offset = ctx->range.offset;
10367 15 : range->locked_ctx = ctx->range.locked_ctx;
10368 15 : range->quiesce = ctx->range.quiesce;
10369 15 : ctx->current_range = range;
10370 15 : if (ctx->range.owner_ch == ch) {
10371 : /* This is the range object for the channel that will hold
10372 : * the lock. Store it in the ctx object so that we can easily
10373 : * set its owner_ch after the lock is finally acquired.
10374 : */
10375 10 : ctx->owner_range = range;
10376 : }
10377 15 : TAILQ_INSERT_TAIL(&ch->locked_ranges, range, tailq);
10378 15 : bdev_lock_lba_range_check_io(i);
10379 : }
10380 :
10381 : static void
10382 14 : bdev_lock_lba_range_ctx(struct spdk_bdev *bdev, struct locked_lba_range_ctx *ctx)
10383 : {
10384 14 : assert(spdk_get_thread() == ctx->range.owner_thread);
10385 14 : assert(ctx->range.owner_ch == NULL ||
10386 : spdk_io_channel_get_thread(ctx->range.owner_ch->channel) == ctx->range.owner_thread);
10387 :
10388 : /* We will add a copy of this range to each channel now. */
10389 14 : spdk_bdev_for_each_channel(bdev, bdev_lock_lba_range_get_channel, ctx,
10390 : bdev_lock_lba_range_cb);
10391 14 : }
10392 :
10393 : static bool
10394 17 : bdev_lba_range_overlaps_tailq(struct lba_range *range, lba_range_tailq_t *tailq)
10395 : {
10396 : struct lba_range *r;
10397 :
10398 18 : TAILQ_FOREACH(r, tailq, tailq) {
10399 4 : if (bdev_lba_range_overlapped(range, r)) {
10400 3 : return true;
10401 : }
10402 : }
10403 14 : return false;
10404 : }
10405 :
10406 : static void bdev_quiesce_range_locked(struct lba_range *range, void *ctx, int status);
10407 :
10408 : static int
10409 14 : _bdev_lock_lba_range(struct spdk_bdev *bdev, struct spdk_bdev_channel *ch,
10410 : uint64_t offset, uint64_t length,
10411 : lock_range_cb cb_fn, void *cb_arg)
10412 : {
10413 : struct locked_lba_range_ctx *ctx;
10414 :
10415 14 : ctx = calloc(1, sizeof(*ctx));
10416 14 : if (ctx == NULL) {
10417 0 : return -ENOMEM;
10418 : }
10419 :
10420 14 : ctx->range.offset = offset;
10421 14 : ctx->range.length = length;
10422 14 : ctx->range.owner_thread = spdk_get_thread();
10423 14 : ctx->range.owner_ch = ch;
10424 14 : ctx->range.locked_ctx = cb_arg;
10425 14 : ctx->range.bdev = bdev;
10426 14 : ctx->range.quiesce = (cb_fn == bdev_quiesce_range_locked);
10427 14 : ctx->cb_fn = cb_fn;
10428 14 : ctx->cb_arg = cb_arg;
10429 :
10430 14 : spdk_spin_lock(&bdev->internal.spinlock);
10431 14 : if (bdev_lba_range_overlaps_tailq(&ctx->range, &bdev->internal.locked_ranges)) {
10432 : /* There is an active lock overlapping with this range.
10433 : * Put it on the pending list until this range no
10434 : * longer overlaps with another.
10435 : */
10436 2 : TAILQ_INSERT_TAIL(&bdev->internal.pending_locked_ranges, &ctx->range, tailq);
10437 : } else {
10438 12 : TAILQ_INSERT_TAIL(&bdev->internal.locked_ranges, &ctx->range, tailq);
10439 12 : bdev_lock_lba_range_ctx(bdev, ctx);
10440 : }
10441 14 : spdk_spin_unlock(&bdev->internal.spinlock);
10442 14 : return 0;
10443 : }
10444 :
10445 : static int
10446 10 : bdev_lock_lba_range(struct spdk_bdev_desc *desc, struct spdk_io_channel *_ch,
10447 : uint64_t offset, uint64_t length,
10448 : lock_range_cb cb_fn, void *cb_arg)
10449 : {
10450 10 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
10451 10 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
10452 :
10453 10 : if (cb_arg == NULL) {
10454 0 : SPDK_ERRLOG("cb_arg must not be NULL\n");
10455 0 : return -EINVAL;
10456 : }
10457 :
10458 10 : return _bdev_lock_lba_range(bdev, ch, offset, length, cb_fn, cb_arg);
10459 : }
10460 :
10461 : static void
10462 2 : bdev_lock_lba_range_ctx_msg(void *_ctx)
10463 : {
10464 2 : struct locked_lba_range_ctx *ctx = _ctx;
10465 :
10466 2 : bdev_lock_lba_range_ctx(ctx->range.bdev, ctx);
10467 2 : }
10468 :
10469 : static void
10470 14 : bdev_unlock_lba_range_cb(struct spdk_bdev *bdev, void *_ctx, int status)
10471 : {
10472 14 : struct locked_lba_range_ctx *ctx = _ctx;
10473 : struct locked_lba_range_ctx *pending_ctx;
10474 : struct lba_range *range, *tmp;
10475 :
10476 14 : spdk_spin_lock(&bdev->internal.spinlock);
10477 : /* Check if there are any pending locked ranges that overlap with this range
10478 : * that was just unlocked. If there are, check that it doesn't overlap with any
10479 : * other locked ranges before calling bdev_lock_lba_range_ctx which will start
10480 : * the lock process.
10481 : */
10482 17 : TAILQ_FOREACH_SAFE(range, &bdev->internal.pending_locked_ranges, tailq, tmp) {
10483 3 : if (bdev_lba_range_overlapped(range, &ctx->range) &&
10484 3 : !bdev_lba_range_overlaps_tailq(range, &bdev->internal.locked_ranges)) {
10485 2 : TAILQ_REMOVE(&bdev->internal.pending_locked_ranges, range, tailq);
10486 2 : pending_ctx = SPDK_CONTAINEROF(range, struct locked_lba_range_ctx, range);
10487 2 : TAILQ_INSERT_TAIL(&bdev->internal.locked_ranges, range, tailq);
10488 2 : spdk_thread_send_msg(pending_ctx->range.owner_thread,
10489 : bdev_lock_lba_range_ctx_msg, pending_ctx);
10490 : }
10491 : }
10492 14 : spdk_spin_unlock(&bdev->internal.spinlock);
10493 :
10494 14 : ctx->cb_fn(&ctx->range, ctx->cb_arg, status);
10495 14 : free(ctx);
10496 14 : }
10497 :
10498 : static void
10499 16 : bdev_unlock_lba_range_get_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
10500 : struct spdk_io_channel *_ch, void *_ctx)
10501 : {
10502 16 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
10503 16 : struct locked_lba_range_ctx *ctx = _ctx;
10504 16 : TAILQ_HEAD(, spdk_bdev_io) io_locked;
10505 : struct spdk_bdev_io *bdev_io;
10506 : struct lba_range *range;
10507 :
10508 16 : TAILQ_FOREACH(range, &ch->locked_ranges, tailq) {
10509 16 : if (ctx->range.offset == range->offset &&
10510 16 : ctx->range.length == range->length &&
10511 16 : ctx->range.locked_ctx == range->locked_ctx) {
10512 16 : TAILQ_REMOVE(&ch->locked_ranges, range, tailq);
10513 16 : free(range);
10514 16 : break;
10515 : }
10516 : }
10517 :
10518 : /* Note: we should almost always be able to assert that the range specified
10519 : * was found. But there are some very rare corner cases where a new channel
10520 : * gets created simultaneously with a range unlock, where this function
10521 : * would execute on that new channel and wouldn't have the range.
10522 : * We also use this to clean up range allocations when a later allocation
10523 : * fails in the locking path.
10524 : * So we can't actually assert() here.
10525 : */
10526 :
10527 : /* Swap the locked IO into a temporary list, and then try to submit them again.
10528 : * We could hyper-optimize this to only resubmit locked I/O that overlap
10529 : * with the range that was just unlocked, but this isn't a performance path so
10530 : * we go for simplicity here.
10531 : */
10532 16 : TAILQ_INIT(&io_locked);
10533 16 : TAILQ_SWAP(&ch->io_locked, &io_locked, spdk_bdev_io, internal.ch_link);
10534 19 : while (!TAILQ_EMPTY(&io_locked)) {
10535 3 : bdev_io = TAILQ_FIRST(&io_locked);
10536 3 : TAILQ_REMOVE(&io_locked, bdev_io, internal.ch_link);
10537 3 : bdev_io_submit(bdev_io);
10538 : }
10539 :
10540 16 : spdk_bdev_for_each_channel_continue(i, 0);
10541 16 : }
10542 :
10543 : static int
10544 14 : _bdev_unlock_lba_range(struct spdk_bdev *bdev, uint64_t offset, uint64_t length,
10545 : lock_range_cb cb_fn, void *cb_arg)
10546 : {
10547 : struct locked_lba_range_ctx *ctx;
10548 : struct lba_range *range;
10549 :
10550 14 : spdk_spin_lock(&bdev->internal.spinlock);
10551 : /* To start the unlock the process, we find the range in the bdev's locked_ranges
10552 : * and remove it. This ensures new channels don't inherit the locked range.
10553 : * Then we will send a message to each channel to remove the range from its
10554 : * per-channel list.
10555 : */
10556 14 : TAILQ_FOREACH(range, &bdev->internal.locked_ranges, tailq) {
10557 14 : if (range->offset == offset && range->length == length &&
10558 14 : (range->owner_ch == NULL || range->locked_ctx == cb_arg)) {
10559 : break;
10560 : }
10561 : }
10562 14 : if (range == NULL) {
10563 0 : assert(false);
10564 : spdk_spin_unlock(&bdev->internal.spinlock);
10565 : return -EINVAL;
10566 : }
10567 14 : TAILQ_REMOVE(&bdev->internal.locked_ranges, range, tailq);
10568 14 : ctx = SPDK_CONTAINEROF(range, struct locked_lba_range_ctx, range);
10569 14 : spdk_spin_unlock(&bdev->internal.spinlock);
10570 :
10571 14 : ctx->cb_fn = cb_fn;
10572 14 : ctx->cb_arg = cb_arg;
10573 :
10574 14 : spdk_bdev_for_each_channel(bdev, bdev_unlock_lba_range_get_channel, ctx,
10575 : bdev_unlock_lba_range_cb);
10576 14 : return 0;
10577 : }
10578 :
10579 : static int
10580 12 : bdev_unlock_lba_range(struct spdk_bdev_desc *desc, struct spdk_io_channel *_ch,
10581 : uint64_t offset, uint64_t length,
10582 : lock_range_cb cb_fn, void *cb_arg)
10583 : {
10584 12 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
10585 12 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
10586 : struct lba_range *range;
10587 12 : bool range_found = false;
10588 :
10589 : /* Let's make sure the specified channel actually has a lock on
10590 : * the specified range. Note that the range must match exactly.
10591 : */
10592 14 : TAILQ_FOREACH(range, &ch->locked_ranges, tailq) {
10593 12 : if (range->offset == offset && range->length == length &&
10594 11 : range->owner_ch == ch && range->locked_ctx == cb_arg) {
10595 10 : range_found = true;
10596 10 : break;
10597 : }
10598 : }
10599 :
10600 12 : if (!range_found) {
10601 2 : return -EINVAL;
10602 : }
10603 :
10604 10 : return _bdev_unlock_lba_range(bdev, offset, length, cb_fn, cb_arg);
10605 : }
10606 :
10607 : struct bdev_quiesce_ctx {
10608 : spdk_bdev_quiesce_cb cb_fn;
10609 : void *cb_arg;
10610 : };
10611 :
10612 : static void
10613 4 : bdev_unquiesce_range_unlocked(struct lba_range *range, void *ctx, int status)
10614 : {
10615 4 : struct bdev_quiesce_ctx *quiesce_ctx = ctx;
10616 :
10617 4 : if (quiesce_ctx->cb_fn != NULL) {
10618 4 : quiesce_ctx->cb_fn(quiesce_ctx->cb_arg, status);
10619 : }
10620 :
10621 4 : free(quiesce_ctx);
10622 4 : }
10623 :
10624 : static void
10625 4 : bdev_quiesce_range_locked(struct lba_range *range, void *ctx, int status)
10626 : {
10627 4 : struct bdev_quiesce_ctx *quiesce_ctx = ctx;
10628 4 : struct spdk_bdev_module *module = range->bdev->module;
10629 :
10630 4 : if (status != 0) {
10631 0 : if (quiesce_ctx->cb_fn != NULL) {
10632 0 : quiesce_ctx->cb_fn(quiesce_ctx->cb_arg, status);
10633 : }
10634 0 : free(quiesce_ctx);
10635 0 : return;
10636 : }
10637 :
10638 4 : spdk_spin_lock(&module->internal.spinlock);
10639 4 : TAILQ_INSERT_TAIL(&module->internal.quiesced_ranges, range, tailq_module);
10640 4 : spdk_spin_unlock(&module->internal.spinlock);
10641 :
10642 4 : if (quiesce_ctx->cb_fn != NULL) {
10643 : /* copy the context in case the range is unlocked by the callback */
10644 4 : struct bdev_quiesce_ctx tmp = *quiesce_ctx;
10645 :
10646 4 : quiesce_ctx->cb_fn = NULL;
10647 4 : quiesce_ctx->cb_arg = NULL;
10648 :
10649 4 : tmp.cb_fn(tmp.cb_arg, status);
10650 : }
10651 : /* quiesce_ctx will be freed on unquiesce */
10652 : }
10653 :
10654 : static int
10655 9 : _spdk_bdev_quiesce(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
10656 : uint64_t offset, uint64_t length,
10657 : spdk_bdev_quiesce_cb cb_fn, void *cb_arg,
10658 : bool unquiesce)
10659 : {
10660 : struct bdev_quiesce_ctx *quiesce_ctx;
10661 : int rc;
10662 :
10663 9 : if (module != bdev->module) {
10664 0 : SPDK_ERRLOG("Bdev does not belong to specified module.\n");
10665 0 : return -EINVAL;
10666 : }
10667 :
10668 9 : if (!bdev_io_valid_blocks(bdev, offset, length)) {
10669 0 : return -EINVAL;
10670 : }
10671 :
10672 9 : if (unquiesce) {
10673 : struct lba_range *range;
10674 :
10675 : /* Make sure the specified range is actually quiesced in the specified module and
10676 : * then remove it from the list. Note that the range must match exactly.
10677 : */
10678 5 : spdk_spin_lock(&module->internal.spinlock);
10679 6 : TAILQ_FOREACH(range, &module->internal.quiesced_ranges, tailq_module) {
10680 5 : if (range->bdev == bdev && range->offset == offset && range->length == length) {
10681 4 : TAILQ_REMOVE(&module->internal.quiesced_ranges, range, tailq_module);
10682 4 : break;
10683 : }
10684 : }
10685 5 : spdk_spin_unlock(&module->internal.spinlock);
10686 :
10687 5 : if (range == NULL) {
10688 1 : SPDK_ERRLOG("The range to unquiesce was not found.\n");
10689 1 : return -EINVAL;
10690 : }
10691 :
10692 4 : quiesce_ctx = range->locked_ctx;
10693 4 : quiesce_ctx->cb_fn = cb_fn;
10694 4 : quiesce_ctx->cb_arg = cb_arg;
10695 :
10696 4 : rc = _bdev_unlock_lba_range(bdev, offset, length, bdev_unquiesce_range_unlocked, quiesce_ctx);
10697 : } else {
10698 4 : quiesce_ctx = malloc(sizeof(*quiesce_ctx));
10699 4 : if (quiesce_ctx == NULL) {
10700 0 : return -ENOMEM;
10701 : }
10702 :
10703 4 : quiesce_ctx->cb_fn = cb_fn;
10704 4 : quiesce_ctx->cb_arg = cb_arg;
10705 :
10706 4 : rc = _bdev_lock_lba_range(bdev, NULL, offset, length, bdev_quiesce_range_locked, quiesce_ctx);
10707 4 : if (rc != 0) {
10708 0 : free(quiesce_ctx);
10709 : }
10710 : }
10711 :
10712 8 : return rc;
10713 : }
10714 :
10715 : int
10716 3 : spdk_bdev_quiesce(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
10717 : spdk_bdev_quiesce_cb cb_fn, void *cb_arg)
10718 : {
10719 3 : return _spdk_bdev_quiesce(bdev, module, 0, bdev->blockcnt, cb_fn, cb_arg, false);
10720 : }
10721 :
10722 : int
10723 3 : spdk_bdev_unquiesce(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
10724 : spdk_bdev_quiesce_cb cb_fn, void *cb_arg)
10725 : {
10726 3 : return _spdk_bdev_quiesce(bdev, module, 0, bdev->blockcnt, cb_fn, cb_arg, true);
10727 : }
10728 :
10729 : int
10730 1 : spdk_bdev_quiesce_range(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
10731 : uint64_t offset, uint64_t length,
10732 : spdk_bdev_quiesce_cb cb_fn, void *cb_arg)
10733 : {
10734 1 : return _spdk_bdev_quiesce(bdev, module, offset, length, cb_fn, cb_arg, false);
10735 : }
10736 :
10737 : int
10738 2 : spdk_bdev_unquiesce_range(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
10739 : uint64_t offset, uint64_t length,
10740 : spdk_bdev_quiesce_cb cb_fn, void *cb_arg)
10741 : {
10742 2 : return _spdk_bdev_quiesce(bdev, module, offset, length, cb_fn, cb_arg, true);
10743 : }
10744 :
10745 : int
10746 283 : spdk_bdev_get_memory_domains(struct spdk_bdev *bdev, struct spdk_memory_domain **domains,
10747 : int array_size)
10748 : {
10749 283 : if (!bdev) {
10750 1 : return -EINVAL;
10751 : }
10752 :
10753 282 : if (bdev->fn_table->get_memory_domains) {
10754 3 : return bdev->fn_table->get_memory_domains(bdev->ctxt, domains, array_size);
10755 : }
10756 :
10757 279 : return 0;
10758 : }
10759 :
10760 : struct spdk_bdev_for_each_io_ctx {
10761 : void *ctx;
10762 : spdk_bdev_io_fn fn;
10763 : spdk_bdev_for_each_io_cb cb;
10764 : };
10765 :
10766 : static void
10767 0 : bdev_channel_for_each_io(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
10768 : struct spdk_io_channel *io_ch, void *_ctx)
10769 : {
10770 0 : struct spdk_bdev_for_each_io_ctx *ctx = _ctx;
10771 0 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
10772 : struct spdk_bdev_io *bdev_io;
10773 0 : int rc = 0;
10774 :
10775 0 : TAILQ_FOREACH(bdev_io, &bdev_ch->io_submitted, internal.ch_link) {
10776 0 : rc = ctx->fn(ctx->ctx, bdev_io);
10777 0 : if (rc != 0) {
10778 0 : break;
10779 : }
10780 : }
10781 :
10782 0 : spdk_bdev_for_each_channel_continue(i, rc);
10783 0 : }
10784 :
10785 : static void
10786 0 : bdev_for_each_io_done(struct spdk_bdev *bdev, void *_ctx, int status)
10787 : {
10788 0 : struct spdk_bdev_for_each_io_ctx *ctx = _ctx;
10789 :
10790 0 : ctx->cb(ctx->ctx, status);
10791 :
10792 0 : free(ctx);
10793 0 : }
10794 :
10795 : void
10796 0 : spdk_bdev_for_each_bdev_io(struct spdk_bdev *bdev, void *_ctx, spdk_bdev_io_fn fn,
10797 : spdk_bdev_for_each_io_cb cb)
10798 : {
10799 : struct spdk_bdev_for_each_io_ctx *ctx;
10800 :
10801 0 : assert(fn != NULL && cb != NULL);
10802 :
10803 0 : ctx = calloc(1, sizeof(*ctx));
10804 0 : if (ctx == NULL) {
10805 0 : SPDK_ERRLOG("Failed to allocate context.\n");
10806 0 : cb(_ctx, -ENOMEM);
10807 0 : return;
10808 : }
10809 :
10810 0 : ctx->ctx = _ctx;
10811 0 : ctx->fn = fn;
10812 0 : ctx->cb = cb;
10813 :
10814 0 : spdk_bdev_for_each_channel(bdev, bdev_channel_for_each_io, ctx,
10815 : bdev_for_each_io_done);
10816 : }
10817 :
10818 : void
10819 135 : spdk_bdev_for_each_channel_continue(struct spdk_bdev_channel_iter *iter, int status)
10820 : {
10821 135 : spdk_for_each_channel_continue(iter->i, status);
10822 135 : }
10823 :
10824 : static struct spdk_bdev *
10825 371 : io_channel_iter_get_bdev(struct spdk_io_channel_iter *i)
10826 : {
10827 371 : void *io_device = spdk_io_channel_iter_get_io_device(i);
10828 :
10829 371 : return __bdev_from_io_dev(io_device);
10830 : }
10831 :
10832 : static void
10833 135 : bdev_each_channel_msg(struct spdk_io_channel_iter *i)
10834 : {
10835 135 : struct spdk_bdev_channel_iter *iter = spdk_io_channel_iter_get_ctx(i);
10836 135 : struct spdk_bdev *bdev = io_channel_iter_get_bdev(i);
10837 135 : struct spdk_io_channel *ch = spdk_io_channel_iter_get_channel(i);
10838 :
10839 135 : iter->i = i;
10840 135 : iter->fn(iter, bdev, ch, iter->ctx);
10841 135 : }
10842 :
10843 : static void
10844 236 : bdev_each_channel_cpl(struct spdk_io_channel_iter *i, int status)
10845 : {
10846 236 : struct spdk_bdev_channel_iter *iter = spdk_io_channel_iter_get_ctx(i);
10847 236 : struct spdk_bdev *bdev = io_channel_iter_get_bdev(i);
10848 :
10849 236 : iter->i = i;
10850 236 : iter->cpl(bdev, iter->ctx, status);
10851 :
10852 236 : free(iter);
10853 236 : }
10854 :
10855 : void
10856 236 : spdk_bdev_for_each_channel(struct spdk_bdev *bdev, spdk_bdev_for_each_channel_msg fn,
10857 : void *ctx, spdk_bdev_for_each_channel_done cpl)
10858 : {
10859 : struct spdk_bdev_channel_iter *iter;
10860 :
10861 236 : assert(bdev != NULL && fn != NULL && ctx != NULL);
10862 :
10863 236 : iter = calloc(1, sizeof(struct spdk_bdev_channel_iter));
10864 236 : if (iter == NULL) {
10865 0 : SPDK_ERRLOG("Unable to allocate iterator\n");
10866 0 : assert(false);
10867 : return;
10868 : }
10869 :
10870 236 : iter->fn = fn;
10871 236 : iter->cpl = cpl;
10872 236 : iter->ctx = ctx;
10873 :
10874 236 : spdk_for_each_channel(__bdev_to_io_dev(bdev), bdev_each_channel_msg,
10875 : iter, bdev_each_channel_cpl);
10876 : }
10877 :
10878 : static void
10879 3 : bdev_copy_do_write_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
10880 : {
10881 3 : struct spdk_bdev_io *parent_io = cb_arg;
10882 :
10883 3 : spdk_bdev_free_io(bdev_io);
10884 :
10885 : /* Check return status of write */
10886 3 : parent_io->internal.status = success ? SPDK_BDEV_IO_STATUS_SUCCESS : SPDK_BDEV_IO_STATUS_FAILED;
10887 3 : parent_io->internal.cb(parent_io, success, parent_io->internal.caller_ctx);
10888 3 : }
10889 :
10890 : static void
10891 3 : bdev_copy_do_write(void *_bdev_io)
10892 : {
10893 3 : struct spdk_bdev_io *bdev_io = _bdev_io;
10894 : int rc;
10895 :
10896 : /* Write blocks */
10897 6 : rc = spdk_bdev_write_blocks_with_md(bdev_io->internal.desc,
10898 3 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
10899 3 : bdev_io->u.bdev.iovs[0].iov_base,
10900 : bdev_io->u.bdev.md_buf, bdev_io->u.bdev.offset_blocks,
10901 : bdev_io->u.bdev.num_blocks, bdev_copy_do_write_done, bdev_io);
10902 :
10903 3 : if (rc == -ENOMEM) {
10904 0 : bdev_queue_io_wait_with_cb(bdev_io, bdev_copy_do_write);
10905 3 : } else if (rc != 0) {
10906 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
10907 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
10908 : }
10909 3 : }
10910 :
10911 : static void
10912 3 : bdev_copy_do_read_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
10913 : {
10914 3 : struct spdk_bdev_io *parent_io = cb_arg;
10915 :
10916 3 : spdk_bdev_free_io(bdev_io);
10917 :
10918 : /* Check return status of read */
10919 3 : if (!success) {
10920 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
10921 0 : parent_io->internal.cb(parent_io, false, parent_io->internal.caller_ctx);
10922 0 : return;
10923 : }
10924 :
10925 : /* Do write */
10926 3 : bdev_copy_do_write(parent_io);
10927 : }
10928 :
10929 : static void
10930 3 : bdev_copy_do_read(void *_bdev_io)
10931 : {
10932 3 : struct spdk_bdev_io *bdev_io = _bdev_io;
10933 : int rc;
10934 :
10935 : /* Read blocks */
10936 6 : rc = spdk_bdev_read_blocks_with_md(bdev_io->internal.desc,
10937 3 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
10938 3 : bdev_io->u.bdev.iovs[0].iov_base,
10939 : bdev_io->u.bdev.md_buf, bdev_io->u.bdev.copy.src_offset_blocks,
10940 : bdev_io->u.bdev.num_blocks, bdev_copy_do_read_done, bdev_io);
10941 :
10942 3 : if (rc == -ENOMEM) {
10943 0 : bdev_queue_io_wait_with_cb(bdev_io, bdev_copy_do_read);
10944 3 : } else if (rc != 0) {
10945 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
10946 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
10947 : }
10948 3 : }
10949 :
10950 : static void
10951 3 : bdev_copy_get_buf_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io, bool success)
10952 : {
10953 3 : if (!success) {
10954 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
10955 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
10956 0 : return;
10957 : }
10958 :
10959 3 : bdev_copy_do_read(bdev_io);
10960 : }
10961 :
10962 : int
10963 27 : spdk_bdev_copy_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
10964 : uint64_t dst_offset_blocks, uint64_t src_offset_blocks, uint64_t num_blocks,
10965 : spdk_bdev_io_completion_cb cb, void *cb_arg)
10966 : {
10967 27 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
10968 : struct spdk_bdev_io *bdev_io;
10969 27 : struct spdk_bdev_channel *channel = spdk_io_channel_get_ctx(ch);
10970 :
10971 27 : if (!desc->write) {
10972 0 : return -EBADF;
10973 : }
10974 :
10975 27 : if (!bdev_io_valid_blocks(bdev, dst_offset_blocks, num_blocks) ||
10976 27 : !bdev_io_valid_blocks(bdev, src_offset_blocks, num_blocks)) {
10977 0 : SPDK_DEBUGLOG(bdev,
10978 : "Invalid offset or number of blocks: dst %lu, src %lu, count %lu\n",
10979 : dst_offset_blocks, src_offset_blocks, num_blocks);
10980 0 : return -EINVAL;
10981 : }
10982 :
10983 27 : bdev_io = bdev_channel_get_io(channel);
10984 27 : if (!bdev_io) {
10985 0 : return -ENOMEM;
10986 : }
10987 :
10988 27 : bdev_io->internal.ch = channel;
10989 27 : bdev_io->internal.desc = desc;
10990 27 : bdev_io->type = SPDK_BDEV_IO_TYPE_COPY;
10991 :
10992 27 : bdev_io->u.bdev.offset_blocks = dst_offset_blocks;
10993 27 : bdev_io->u.bdev.copy.src_offset_blocks = src_offset_blocks;
10994 27 : bdev_io->u.bdev.num_blocks = num_blocks;
10995 27 : bdev_io->u.bdev.memory_domain = NULL;
10996 27 : bdev_io->u.bdev.memory_domain_ctx = NULL;
10997 27 : bdev_io->u.bdev.iovs = NULL;
10998 27 : bdev_io->u.bdev.iovcnt = 0;
10999 27 : bdev_io->u.bdev.md_buf = NULL;
11000 27 : bdev_io->u.bdev.accel_sequence = NULL;
11001 27 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
11002 :
11003 27 : if (dst_offset_blocks == src_offset_blocks || num_blocks == 0) {
11004 0 : spdk_thread_send_msg(spdk_get_thread(), bdev_io_complete_cb, bdev_io);
11005 0 : return 0;
11006 : }
11007 :
11008 :
11009 : /* If the copy size is large and should be split, use the generic split logic
11010 : * regardless of whether SPDK_BDEV_IO_TYPE_COPY is supported or not.
11011 : *
11012 : * Then, send the copy request if SPDK_BDEV_IO_TYPE_COPY is supported or
11013 : * emulate it using regular read and write requests otherwise.
11014 : */
11015 27 : if (spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COPY) ||
11016 : bdev_io->internal.f.split) {
11017 24 : bdev_io_submit(bdev_io);
11018 24 : return 0;
11019 : }
11020 :
11021 3 : spdk_bdev_io_get_buf(bdev_io, bdev_copy_get_buf_cb, num_blocks * spdk_bdev_get_block_size(bdev));
11022 :
11023 3 : return 0;
11024 : }
11025 :
11026 3 : SPDK_LOG_REGISTER_COMPONENT(bdev)
11027 :
11028 : static void
11029 0 : bdev_trace(void)
11030 : {
11031 0 : struct spdk_trace_tpoint_opts opts[] = {
11032 : {
11033 : "BDEV_IO_START", TRACE_BDEV_IO_START,
11034 : OWNER_TYPE_BDEV, OBJECT_BDEV_IO, 1,
11035 : {
11036 : { "type", SPDK_TRACE_ARG_TYPE_INT, 8 },
11037 : { "ctx", SPDK_TRACE_ARG_TYPE_PTR, 8 },
11038 : { "offset", SPDK_TRACE_ARG_TYPE_INT, 8 },
11039 : { "qd", SPDK_TRACE_ARG_TYPE_INT, 4 }
11040 : }
11041 : },
11042 : {
11043 : "BDEV_IO_DONE", TRACE_BDEV_IO_DONE,
11044 : OWNER_TYPE_BDEV, OBJECT_BDEV_IO, 0,
11045 : {
11046 : { "ctx", SPDK_TRACE_ARG_TYPE_PTR, 8 },
11047 : { "qd", SPDK_TRACE_ARG_TYPE_INT, 4 }
11048 : }
11049 : },
11050 : {
11051 : "BDEV_IOCH_CREATE", TRACE_BDEV_IOCH_CREATE,
11052 : OWNER_TYPE_BDEV, OBJECT_NONE, 0,
11053 : {
11054 : { "tid", SPDK_TRACE_ARG_TYPE_INT, 8 }
11055 : }
11056 : },
11057 : {
11058 : "BDEV_IOCH_DESTROY", TRACE_BDEV_IOCH_DESTROY,
11059 : OWNER_TYPE_BDEV, OBJECT_NONE, 0,
11060 : {
11061 : { "tid", SPDK_TRACE_ARG_TYPE_INT, 8 }
11062 : }
11063 : },
11064 : };
11065 :
11066 :
11067 0 : spdk_trace_register_owner_type(OWNER_TYPE_BDEV, 'b');
11068 0 : spdk_trace_register_object(OBJECT_BDEV_IO, 'i');
11069 0 : spdk_trace_register_description_ext(opts, SPDK_COUNTOF(opts));
11070 0 : spdk_trace_tpoint_register_relation(TRACE_BDEV_NVME_IO_START, OBJECT_BDEV_IO, 0);
11071 0 : spdk_trace_tpoint_register_relation(TRACE_BDEV_NVME_IO_DONE, OBJECT_BDEV_IO, 0);
11072 0 : spdk_trace_tpoint_register_relation(TRACE_BLOB_REQ_SET_START, OBJECT_BDEV_IO, 0);
11073 0 : spdk_trace_tpoint_register_relation(TRACE_BLOB_REQ_SET_COMPLETE, OBJECT_BDEV_IO, 0);
11074 0 : spdk_trace_tpoint_register_relation(TRACE_BDEV_RAID_IO_START, OBJECT_BDEV_IO, 0);
11075 0 : spdk_trace_tpoint_register_relation(TRACE_BDEV_RAID_IO_DONE, OBJECT_BDEV_IO, 0);
11076 0 : }
11077 3 : SPDK_TRACE_REGISTER_FN(bdev_trace, "bdev", TRACE_GROUP_BDEV)
|
