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
86 540 : bdev_name_cmp(struct spdk_bdev_name *name1, struct spdk_bdev_name *name2)
87 : {
88 540 : return strcmp(name1->name, name2->name);
89 : }
90 :
91 1651 : RB_GENERATE_STATIC(bdev_name_tree, spdk_bdev_name, node, bdev_name_cmp);
92 :
93 : struct spdk_bdev_mgr {
94 : struct spdk_mempool *bdev_io_pool;
95 :
96 : void *zero_buffer;
97 :
98 : TAILQ_HEAD(bdev_module_list, spdk_bdev_module) bdev_modules;
99 :
100 : struct spdk_bdev_list bdevs;
101 : struct bdev_name_tree bdev_names;
102 :
103 : bool init_complete;
104 : bool module_init_complete;
105 :
106 : struct spdk_spinlock spinlock;
107 :
108 : TAILQ_HEAD(, spdk_bdev_open_async_ctx) async_bdev_opens;
109 :
110 : #ifdef SPDK_CONFIG_VTUNE
111 : __itt_domain *domain;
112 : #endif
113 : };
114 :
115 : static struct spdk_bdev_mgr g_bdev_mgr = {
116 : .bdev_modules = TAILQ_HEAD_INITIALIZER(g_bdev_mgr.bdev_modules),
117 : .bdevs = TAILQ_HEAD_INITIALIZER(g_bdev_mgr.bdevs),
118 : .bdev_names = RB_INITIALIZER(g_bdev_mgr.bdev_names),
119 : .init_complete = false,
120 : .module_init_complete = false,
121 : .async_bdev_opens = TAILQ_HEAD_INITIALIZER(g_bdev_mgr.async_bdev_opens),
122 : };
123 :
124 : static void
125 : __attribute__((constructor))
126 3 : _bdev_init(void)
127 : {
128 3 : spdk_spin_init(&g_bdev_mgr.spinlock);
129 3 : }
130 :
131 : typedef void (*lock_range_cb)(struct lba_range *range, void *ctx, int status);
132 :
133 : typedef void (*bdev_copy_bounce_buffer_cpl)(void *ctx, int rc);
134 :
135 : struct lba_range {
136 : struct spdk_bdev *bdev;
137 : uint64_t offset;
138 : uint64_t length;
139 : bool quiesce;
140 : void *locked_ctx;
141 : struct spdk_thread *owner_thread;
142 : struct spdk_bdev_channel *owner_ch;
143 : TAILQ_ENTRY(lba_range) tailq;
144 : TAILQ_ENTRY(lba_range) tailq_module;
145 : };
146 :
147 : static struct spdk_bdev_opts g_bdev_opts = {
148 : .bdev_io_pool_size = SPDK_BDEV_IO_POOL_SIZE,
149 : .bdev_io_cache_size = SPDK_BDEV_IO_CACHE_SIZE,
150 : .bdev_auto_examine = SPDK_BDEV_AUTO_EXAMINE,
151 : .iobuf_small_cache_size = BUF_SMALL_CACHE_SIZE,
152 : .iobuf_large_cache_size = BUF_LARGE_CACHE_SIZE,
153 : };
154 :
155 : static spdk_bdev_init_cb g_init_cb_fn = NULL;
156 : static void *g_init_cb_arg = NULL;
157 :
158 : static spdk_bdev_fini_cb g_fini_cb_fn = NULL;
159 : static void *g_fini_cb_arg = NULL;
160 : static struct spdk_thread *g_fini_thread = NULL;
161 :
162 : struct spdk_bdev_qos_limit {
163 : /** IOs or bytes allowed per second (i.e., 1s). */
164 : uint64_t limit;
165 :
166 : /** Remaining IOs or bytes allowed in current timeslice (e.g., 1ms).
167 : * For remaining bytes, allowed to run negative if an I/O is submitted when
168 : * some bytes are remaining, but the I/O is bigger than that amount. The
169 : * excess will be deducted from the next timeslice.
170 : */
171 : int64_t remaining_this_timeslice;
172 :
173 : /** Minimum allowed IOs or bytes to be issued in one timeslice (e.g., 1ms). */
174 : uint32_t min_per_timeslice;
175 :
176 : /** Maximum allowed IOs or bytes to be issued in one timeslice (e.g., 1ms). */
177 : uint32_t max_per_timeslice;
178 :
179 : /** Function to check whether to queue the IO.
180 : * If The IO is allowed to pass, the quota will be reduced correspondingly.
181 : */
182 : bool (*queue_io)(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io);
183 :
184 : /** Function to rewind the quota once the IO was allowed to be sent by this
185 : * limit but queued due to one of the further limits.
186 : */
187 : void (*rewind_quota)(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io);
188 : };
189 :
190 : struct spdk_bdev_qos {
191 : /** Types of structure of rate limits. */
192 : struct spdk_bdev_qos_limit rate_limits[SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES];
193 :
194 : /** The channel that all I/O are funneled through. */
195 : struct spdk_bdev_channel *ch;
196 :
197 : /** The thread on which the poller is running. */
198 : struct spdk_thread *thread;
199 :
200 : /** Size of a timeslice in tsc ticks. */
201 : uint64_t timeslice_size;
202 :
203 : /** Timestamp of start of last timeslice. */
204 : uint64_t last_timeslice;
205 :
206 : /** Poller that processes queued I/O commands each time slice. */
207 : struct spdk_poller *poller;
208 : };
209 :
210 : struct spdk_bdev_mgmt_channel {
211 : /*
212 : * Each thread keeps a cache of bdev_io - this allows
213 : * bdev threads which are *not* DPDK threads to still
214 : * benefit from a per-thread bdev_io cache. Without
215 : * this, non-DPDK threads fetching from the mempool
216 : * incur a cmpxchg on get and put.
217 : */
218 : bdev_io_stailq_t per_thread_cache;
219 : uint32_t per_thread_cache_count;
220 : uint32_t bdev_io_cache_size;
221 :
222 : struct spdk_iobuf_channel iobuf;
223 :
224 : TAILQ_HEAD(, spdk_bdev_shared_resource) shared_resources;
225 : TAILQ_HEAD(, spdk_bdev_io_wait_entry) io_wait_queue;
226 : };
227 :
228 : /*
229 : * Per-module (or per-io_device) data. Multiple bdevs built on the same io_device
230 : * will queue here their IO that awaits retry. It makes it possible to retry sending
231 : * IO to one bdev after IO from other bdev completes.
232 : */
233 : struct spdk_bdev_shared_resource {
234 : /* The bdev management channel */
235 : struct spdk_bdev_mgmt_channel *mgmt_ch;
236 :
237 : /*
238 : * Count of I/O submitted to bdev module and waiting for completion.
239 : * Incremented before submit_request() is called on an spdk_bdev_io.
240 : */
241 : uint64_t io_outstanding;
242 :
243 : /*
244 : * Queue of IO awaiting retry because of a previous NOMEM status returned
245 : * on this channel.
246 : */
247 : bdev_io_tailq_t nomem_io;
248 :
249 : /*
250 : * Threshold which io_outstanding must drop to before retrying nomem_io.
251 : */
252 : uint64_t nomem_threshold;
253 :
254 : /* I/O channel allocated by a bdev module */
255 : struct spdk_io_channel *shared_ch;
256 :
257 : struct spdk_poller *nomem_poller;
258 :
259 : /* Refcount of bdev channels using this resource */
260 : uint32_t ref;
261 :
262 : TAILQ_ENTRY(spdk_bdev_shared_resource) link;
263 : };
264 :
265 : #define BDEV_CH_RESET_IN_PROGRESS (1 << 0)
266 : #define BDEV_CH_QOS_ENABLED (1 << 1)
267 :
268 : struct spdk_bdev_channel {
269 : struct spdk_bdev *bdev;
270 :
271 : /* The channel for the underlying device */
272 : struct spdk_io_channel *channel;
273 :
274 : /* Accel channel */
275 : struct spdk_io_channel *accel_channel;
276 :
277 : /* Per io_device per thread data */
278 : struct spdk_bdev_shared_resource *shared_resource;
279 :
280 : struct spdk_bdev_io_stat *stat;
281 :
282 : /*
283 : * Count of I/O submitted to the underlying dev module through this channel
284 : * and waiting for completion.
285 : */
286 : uint64_t io_outstanding;
287 :
288 : /*
289 : * List of all submitted I/Os including I/O that are generated via splitting.
290 : */
291 : bdev_io_tailq_t io_submitted;
292 :
293 : /*
294 : * List of spdk_bdev_io that are currently queued because they write to a locked
295 : * LBA range.
296 : */
297 : bdev_io_tailq_t io_locked;
298 :
299 : /* List of I/Os with accel sequence being currently executed */
300 : bdev_io_tailq_t io_accel_exec;
301 :
302 : /* List of I/Os doing memory domain pull/push */
303 : bdev_io_tailq_t io_memory_domain;
304 :
305 : uint32_t flags;
306 :
307 : /* Counts number of bdev_io in the io_submitted TAILQ */
308 : uint16_t queue_depth;
309 :
310 : uint16_t trace_id;
311 :
312 : struct spdk_histogram_data *histogram;
313 :
314 : #ifdef SPDK_CONFIG_VTUNE
315 : uint64_t start_tsc;
316 : uint64_t interval_tsc;
317 : __itt_string_handle *handle;
318 : struct spdk_bdev_io_stat *prev_stat;
319 : #endif
320 :
321 : bdev_io_tailq_t queued_resets;
322 :
323 : lba_range_tailq_t locked_ranges;
324 :
325 : /** List of I/Os queued by QoS. */
326 : bdev_io_tailq_t qos_queued_io;
327 : };
328 :
329 : struct media_event_entry {
330 : struct spdk_bdev_media_event event;
331 : TAILQ_ENTRY(media_event_entry) tailq;
332 : };
333 :
334 : #define MEDIA_EVENT_POOL_SIZE 64
335 :
336 : struct spdk_bdev_desc {
337 : struct spdk_bdev *bdev;
338 : struct spdk_thread *thread;
339 : struct {
340 : spdk_bdev_event_cb_t event_fn;
341 : void *ctx;
342 : } callback;
343 : bool closed;
344 : bool write;
345 : bool memory_domains_supported;
346 : bool accel_sequence_supported[SPDK_BDEV_NUM_IO_TYPES];
347 : struct spdk_spinlock spinlock;
348 : uint32_t refs;
349 : TAILQ_HEAD(, media_event_entry) pending_media_events;
350 : TAILQ_HEAD(, media_event_entry) free_media_events;
351 : struct media_event_entry *media_events_buffer;
352 : TAILQ_ENTRY(spdk_bdev_desc) link;
353 :
354 : uint64_t timeout_in_sec;
355 : spdk_bdev_io_timeout_cb cb_fn;
356 : void *cb_arg;
357 : struct spdk_poller *io_timeout_poller;
358 : struct spdk_bdev_module_claim *claim;
359 : };
360 :
361 : struct spdk_bdev_iostat_ctx {
362 : struct spdk_bdev_io_stat *stat;
363 : 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 : };
392 :
393 : #define __bdev_to_io_dev(bdev) (((char *)bdev) + 1)
394 : #define __bdev_from_io_dev(io_dev) ((struct spdk_bdev *)(((char *)io_dev) - 1))
395 : #define __io_ch_to_bdev_ch(io_ch) ((struct spdk_bdev_channel *)spdk_io_channel_get_ctx(io_ch))
396 : #define __io_ch_to_bdev_mgmt_ch(io_ch) ((struct spdk_bdev_mgmt_channel *)spdk_io_channel_get_ctx(io_ch))
397 :
398 : static inline void bdev_io_complete(void *ctx);
399 : static inline void bdev_io_complete_unsubmitted(struct spdk_bdev_io *bdev_io);
400 : static void bdev_io_push_bounce_md_buf(struct spdk_bdev_io *bdev_io);
401 : static void bdev_io_push_bounce_data(struct spdk_bdev_io *bdev_io);
402 :
403 : static void bdev_write_zero_buffer_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg);
404 : static int bdev_write_zero_buffer(struct spdk_bdev_io *bdev_io);
405 :
406 : static void bdev_enable_qos_msg(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
407 : struct spdk_io_channel *ch, void *_ctx);
408 : static void bdev_enable_qos_done(struct spdk_bdev *bdev, void *_ctx, int status);
409 :
410 : static int bdev_readv_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
411 : struct iovec *iov, int iovcnt, void *md_buf, uint64_t offset_blocks,
412 : uint64_t num_blocks,
413 : struct spdk_memory_domain *domain, void *domain_ctx,
414 : struct spdk_accel_sequence *seq, uint32_t dif_check_flags,
415 : spdk_bdev_io_completion_cb cb, void *cb_arg);
416 : static int bdev_writev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
417 : struct iovec *iov, int iovcnt, void *md_buf,
418 : uint64_t offset_blocks, uint64_t num_blocks,
419 : struct spdk_memory_domain *domain, void *domain_ctx,
420 : struct spdk_accel_sequence *seq, uint32_t dif_check_flags,
421 : uint32_t nvme_cdw12_raw, uint32_t nvme_cdw13_raw,
422 : spdk_bdev_io_completion_cb cb, void *cb_arg);
423 :
424 : static int bdev_lock_lba_range(struct spdk_bdev_desc *desc, struct spdk_io_channel *_ch,
425 : uint64_t offset, uint64_t length,
426 : lock_range_cb cb_fn, void *cb_arg);
427 :
428 : static int bdev_unlock_lba_range(struct spdk_bdev_desc *desc, struct spdk_io_channel *_ch,
429 : uint64_t offset, uint64_t length,
430 : lock_range_cb cb_fn, void *cb_arg);
431 :
432 : static bool bdev_abort_queued_io(bdev_io_tailq_t *queue, struct spdk_bdev_io *bio_to_abort);
433 : static bool bdev_abort_buf_io(struct spdk_bdev_mgmt_channel *ch, struct spdk_bdev_io *bio_to_abort);
434 :
435 : static bool claim_type_is_v2(enum spdk_bdev_claim_type type);
436 : static void bdev_desc_release_claims(struct spdk_bdev_desc *desc);
437 : static void claim_reset(struct spdk_bdev *bdev);
438 :
439 : static void bdev_ch_retry_io(struct spdk_bdev_channel *bdev_ch);
440 :
441 : #define bdev_get_ext_io_opt(opts, field, defval) \
442 : ((opts) != NULL ? SPDK_GET_FIELD(opts, field, defval) : (defval))
443 :
444 : static inline void
445 656 : bdev_ch_add_to_io_submitted(struct spdk_bdev_io *bdev_io)
446 : {
447 656 : TAILQ_INSERT_TAIL(&bdev_io->internal.ch->io_submitted, bdev_io, internal.ch_link);
448 656 : bdev_io->internal.ch->queue_depth++;
449 656 : }
450 :
451 : static inline void
452 656 : bdev_ch_remove_from_io_submitted(struct spdk_bdev_io *bdev_io)
453 : {
454 656 : TAILQ_REMOVE(&bdev_io->internal.ch->io_submitted, bdev_io, internal.ch_link);
455 656 : bdev_io->internal.ch->queue_depth--;
456 656 : }
457 :
458 : void
459 13 : spdk_bdev_get_opts(struct spdk_bdev_opts *opts, size_t opts_size)
460 : {
461 13 : if (!opts) {
462 0 : SPDK_ERRLOG("opts should not be NULL\n");
463 0 : return;
464 : }
465 :
466 13 : if (!opts_size) {
467 0 : SPDK_ERRLOG("opts_size should not be zero value\n");
468 0 : return;
469 : }
470 :
471 13 : opts->opts_size = opts_size;
472 :
473 : #define SET_FIELD(field) \
474 : if (offsetof(struct spdk_bdev_opts, field) + sizeof(opts->field) <= opts_size) { \
475 : opts->field = g_bdev_opts.field; \
476 : } \
477 :
478 13 : SET_FIELD(bdev_io_pool_size);
479 13 : SET_FIELD(bdev_io_cache_size);
480 13 : SET_FIELD(bdev_auto_examine);
481 13 : SET_FIELD(iobuf_small_cache_size);
482 13 : SET_FIELD(iobuf_large_cache_size);
483 :
484 : /* Do not remove this statement, you should always update this statement when you adding a new field,
485 : * and do not forget to add the SET_FIELD statement for your added field. */
486 : SPDK_STATIC_ASSERT(sizeof(struct spdk_bdev_opts) == 32, "Incorrect size");
487 :
488 : #undef SET_FIELD
489 : }
490 :
491 : int
492 14 : spdk_bdev_set_opts(struct spdk_bdev_opts *opts)
493 : {
494 : uint32_t min_pool_size;
495 :
496 14 : if (!opts) {
497 0 : SPDK_ERRLOG("opts cannot be NULL\n");
498 0 : return -1;
499 : }
500 :
501 14 : if (!opts->opts_size) {
502 1 : SPDK_ERRLOG("opts_size inside opts cannot be zero value\n");
503 1 : return -1;
504 : }
505 :
506 : /*
507 : * Add 1 to the thread count to account for the extra mgmt_ch that gets created during subsystem
508 : * initialization. A second mgmt_ch will be created on the same thread when the application starts
509 : * but before the deferred put_io_channel event is executed for the first mgmt_ch.
510 : */
511 13 : min_pool_size = opts->bdev_io_cache_size * (spdk_thread_get_count() + 1);
512 13 : if (opts->bdev_io_pool_size < min_pool_size) {
513 0 : SPDK_ERRLOG("bdev_io_pool_size %" PRIu32 " is not compatible with bdev_io_cache_size %" PRIu32
514 : " and %" PRIu32 " threads\n", opts->bdev_io_pool_size, opts->bdev_io_cache_size,
515 : spdk_thread_get_count());
516 0 : SPDK_ERRLOG("bdev_io_pool_size must be at least %" PRIu32 "\n", min_pool_size);
517 0 : return -1;
518 : }
519 :
520 : #define SET_FIELD(field) \
521 : if (offsetof(struct spdk_bdev_opts, field) + sizeof(opts->field) <= opts->opts_size) { \
522 : g_bdev_opts.field = opts->field; \
523 : } \
524 :
525 13 : SET_FIELD(bdev_io_pool_size);
526 13 : SET_FIELD(bdev_io_cache_size);
527 13 : SET_FIELD(bdev_auto_examine);
528 13 : SET_FIELD(iobuf_small_cache_size);
529 13 : SET_FIELD(iobuf_large_cache_size);
530 :
531 13 : g_bdev_opts.opts_size = opts->opts_size;
532 :
533 : #undef SET_FIELD
534 :
535 13 : return 0;
536 : }
537 :
538 : static struct spdk_bdev *
539 143 : bdev_get_by_name(const char *bdev_name)
540 : {
541 143 : struct spdk_bdev_name find;
542 : struct spdk_bdev_name *res;
543 :
544 143 : find.name = (char *)bdev_name;
545 143 : res = RB_FIND(bdev_name_tree, &g_bdev_mgr.bdev_names, &find);
546 143 : if (res != NULL) {
547 136 : return res->bdev;
548 : }
549 :
550 7 : return NULL;
551 : }
552 :
553 : struct spdk_bdev *
554 16 : spdk_bdev_get_by_name(const char *bdev_name)
555 : {
556 : struct spdk_bdev *bdev;
557 :
558 16 : spdk_spin_lock(&g_bdev_mgr.spinlock);
559 16 : bdev = bdev_get_by_name(bdev_name);
560 16 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
561 :
562 16 : return bdev;
563 : }
564 :
565 : struct bdev_io_status_string {
566 : enum spdk_bdev_io_status status;
567 : const char *str;
568 : };
569 :
570 : static const struct bdev_io_status_string bdev_io_status_strings[] = {
571 : { SPDK_BDEV_IO_STATUS_AIO_ERROR, "aio_error" },
572 : { SPDK_BDEV_IO_STATUS_ABORTED, "aborted" },
573 : { SPDK_BDEV_IO_STATUS_FIRST_FUSED_FAILED, "first_fused_failed" },
574 : { SPDK_BDEV_IO_STATUS_MISCOMPARE, "miscompare" },
575 : { SPDK_BDEV_IO_STATUS_NOMEM, "nomem" },
576 : { SPDK_BDEV_IO_STATUS_SCSI_ERROR, "scsi_error" },
577 : { SPDK_BDEV_IO_STATUS_NVME_ERROR, "nvme_error" },
578 : { SPDK_BDEV_IO_STATUS_FAILED, "failed" },
579 : { SPDK_BDEV_IO_STATUS_PENDING, "pending" },
580 : { SPDK_BDEV_IO_STATUS_SUCCESS, "success" },
581 : };
582 :
583 : static const char *
584 0 : bdev_io_status_get_string(enum spdk_bdev_io_status status)
585 : {
586 : uint32_t i;
587 :
588 0 : for (i = 0; i < SPDK_COUNTOF(bdev_io_status_strings); i++) {
589 0 : if (bdev_io_status_strings[i].status == status) {
590 0 : return bdev_io_status_strings[i].str;
591 : }
592 : }
593 :
594 0 : return "reserved";
595 : }
596 :
597 : struct spdk_bdev_wait_for_examine_ctx {
598 : struct spdk_poller *poller;
599 : spdk_bdev_wait_for_examine_cb cb_fn;
600 : void *cb_arg;
601 : };
602 :
603 : static bool bdev_module_all_actions_completed(void);
604 :
605 : static int
606 189 : bdev_wait_for_examine_cb(void *arg)
607 : {
608 189 : struct spdk_bdev_wait_for_examine_ctx *ctx = arg;
609 :
610 189 : if (!bdev_module_all_actions_completed()) {
611 0 : return SPDK_POLLER_IDLE;
612 : }
613 :
614 189 : spdk_poller_unregister(&ctx->poller);
615 189 : ctx->cb_fn(ctx->cb_arg);
616 189 : free(ctx);
617 :
618 189 : return SPDK_POLLER_BUSY;
619 : }
620 :
621 : int
622 189 : spdk_bdev_wait_for_examine(spdk_bdev_wait_for_examine_cb cb_fn, void *cb_arg)
623 : {
624 : struct spdk_bdev_wait_for_examine_ctx *ctx;
625 :
626 189 : ctx = calloc(1, sizeof(*ctx));
627 189 : if (ctx == NULL) {
628 0 : return -ENOMEM;
629 : }
630 189 : ctx->cb_fn = cb_fn;
631 189 : ctx->cb_arg = cb_arg;
632 189 : ctx->poller = SPDK_POLLER_REGISTER(bdev_wait_for_examine_cb, ctx, 0);
633 :
634 189 : return 0;
635 : }
636 :
637 : struct spdk_bdev_examine_item {
638 : char *name;
639 : TAILQ_ENTRY(spdk_bdev_examine_item) link;
640 : };
641 :
642 : TAILQ_HEAD(spdk_bdev_examine_allowlist, spdk_bdev_examine_item);
643 :
644 : struct spdk_bdev_examine_allowlist g_bdev_examine_allowlist = TAILQ_HEAD_INITIALIZER(
645 : g_bdev_examine_allowlist);
646 :
647 : static inline bool
648 6 : bdev_examine_allowlist_check(const char *name)
649 : {
650 : struct spdk_bdev_examine_item *item;
651 6 : TAILQ_FOREACH(item, &g_bdev_examine_allowlist, link) {
652 0 : if (strcmp(name, item->name) == 0) {
653 0 : return true;
654 : }
655 : }
656 6 : return false;
657 : }
658 :
659 : static inline void
660 64 : bdev_examine_allowlist_free(void)
661 : {
662 : struct spdk_bdev_examine_item *item;
663 64 : while (!TAILQ_EMPTY(&g_bdev_examine_allowlist)) {
664 0 : item = TAILQ_FIRST(&g_bdev_examine_allowlist);
665 0 : TAILQ_REMOVE(&g_bdev_examine_allowlist, item, link);
666 0 : free(item->name);
667 0 : free(item);
668 : }
669 64 : }
670 :
671 : static inline bool
672 3 : bdev_in_examine_allowlist(struct spdk_bdev *bdev)
673 : {
674 : struct spdk_bdev_alias *tmp;
675 3 : if (bdev_examine_allowlist_check(bdev->name)) {
676 0 : return true;
677 : }
678 6 : TAILQ_FOREACH(tmp, &bdev->aliases, tailq) {
679 3 : if (bdev_examine_allowlist_check(tmp->alias.name)) {
680 0 : return true;
681 : }
682 : }
683 3 : return false;
684 : }
685 :
686 : static inline bool
687 123 : bdev_ok_to_examine(struct spdk_bdev *bdev)
688 : {
689 123 : if (g_bdev_opts.bdev_auto_examine) {
690 120 : return true;
691 : } else {
692 3 : return bdev_in_examine_allowlist(bdev);
693 : }
694 : }
695 :
696 : static void
697 123 : bdev_examine(struct spdk_bdev *bdev)
698 : {
699 : struct spdk_bdev_module *module;
700 : struct spdk_bdev_module_claim *claim, *tmpclaim;
701 : uint32_t action;
702 :
703 123 : if (!bdev_ok_to_examine(bdev)) {
704 3 : return;
705 : }
706 :
707 489 : TAILQ_FOREACH(module, &g_bdev_mgr.bdev_modules, internal.tailq) {
708 369 : if (module->examine_config) {
709 249 : spdk_spin_lock(&module->internal.spinlock);
710 249 : action = module->internal.action_in_progress;
711 249 : module->internal.action_in_progress++;
712 249 : spdk_spin_unlock(&module->internal.spinlock);
713 249 : module->examine_config(bdev);
714 249 : if (action != module->internal.action_in_progress) {
715 0 : SPDK_ERRLOG("examine_config for module %s did not call "
716 : "spdk_bdev_module_examine_done()\n", module->name);
717 : }
718 : }
719 : }
720 :
721 120 : spdk_spin_lock(&bdev->internal.spinlock);
722 :
723 120 : switch (bdev->internal.claim_type) {
724 115 : case SPDK_BDEV_CLAIM_NONE:
725 : /* Examine by all bdev modules */
726 464 : TAILQ_FOREACH(module, &g_bdev_mgr.bdev_modules, internal.tailq) {
727 349 : if (module->examine_disk) {
728 225 : spdk_spin_lock(&module->internal.spinlock);
729 225 : module->internal.action_in_progress++;
730 225 : spdk_spin_unlock(&module->internal.spinlock);
731 225 : spdk_spin_unlock(&bdev->internal.spinlock);
732 225 : module->examine_disk(bdev);
733 225 : spdk_spin_lock(&bdev->internal.spinlock);
734 : }
735 : }
736 115 : break;
737 1 : case SPDK_BDEV_CLAIM_EXCL_WRITE:
738 : /* Examine by the one bdev module with a v1 claim */
739 1 : module = bdev->internal.claim.v1.module;
740 1 : if (module->examine_disk) {
741 1 : spdk_spin_lock(&module->internal.spinlock);
742 1 : module->internal.action_in_progress++;
743 1 : spdk_spin_unlock(&module->internal.spinlock);
744 1 : spdk_spin_unlock(&bdev->internal.spinlock);
745 1 : module->examine_disk(bdev);
746 1 : return;
747 : }
748 0 : break;
749 4 : default:
750 : /* Examine by all bdev modules with a v2 claim */
751 4 : assert(claim_type_is_v2(bdev->internal.claim_type));
752 : /*
753 : * Removal of tailq nodes while iterating can cause the iteration to jump out of the
754 : * list, perhaps accessing freed memory. Without protection, this could happen
755 : * while the lock is dropped during the examine callback.
756 : */
757 4 : bdev->internal.examine_in_progress++;
758 :
759 9 : TAILQ_FOREACH(claim, &bdev->internal.claim.v2.claims, link) {
760 5 : module = claim->module;
761 :
762 5 : if (module == NULL) {
763 : /* This is a vestigial claim, held by examine_count */
764 0 : continue;
765 : }
766 :
767 5 : if (module->examine_disk == NULL) {
768 0 : continue;
769 : }
770 :
771 5 : spdk_spin_lock(&module->internal.spinlock);
772 5 : module->internal.action_in_progress++;
773 5 : spdk_spin_unlock(&module->internal.spinlock);
774 :
775 : /* Call examine_disk without holding internal.spinlock. */
776 5 : spdk_spin_unlock(&bdev->internal.spinlock);
777 5 : module->examine_disk(bdev);
778 5 : spdk_spin_lock(&bdev->internal.spinlock);
779 : }
780 :
781 4 : assert(bdev->internal.examine_in_progress > 0);
782 4 : bdev->internal.examine_in_progress--;
783 4 : if (bdev->internal.examine_in_progress == 0) {
784 : /* Remove any claims that were released during examine_disk */
785 9 : TAILQ_FOREACH_SAFE(claim, &bdev->internal.claim.v2.claims, link, tmpclaim) {
786 5 : if (claim->desc != NULL) {
787 5 : continue;
788 : }
789 :
790 0 : TAILQ_REMOVE(&bdev->internal.claim.v2.claims, claim, link);
791 0 : free(claim);
792 : }
793 4 : if (TAILQ_EMPTY(&bdev->internal.claim.v2.claims)) {
794 0 : claim_reset(bdev);
795 : }
796 : }
797 : }
798 :
799 119 : spdk_spin_unlock(&bdev->internal.spinlock);
800 : }
801 :
802 : int
803 1 : spdk_bdev_examine(const char *name)
804 : {
805 : struct spdk_bdev *bdev;
806 : struct spdk_bdev_examine_item *item;
807 1 : struct spdk_thread *thread = spdk_get_thread();
808 :
809 1 : if (spdk_unlikely(!spdk_thread_is_app_thread(thread))) {
810 1 : SPDK_ERRLOG("Cannot examine bdev %s on thread %p (%s)\n", name, thread,
811 : thread ? spdk_thread_get_name(thread) : "null");
812 1 : return -EINVAL;
813 : }
814 :
815 0 : if (g_bdev_opts.bdev_auto_examine) {
816 0 : SPDK_ERRLOG("Manual examine is not allowed if auto examine is enabled\n");
817 0 : return -EINVAL;
818 : }
819 :
820 0 : if (bdev_examine_allowlist_check(name)) {
821 0 : SPDK_ERRLOG("Duplicate bdev name for manual examine: %s\n", name);
822 0 : return -EEXIST;
823 : }
824 :
825 0 : item = calloc(1, sizeof(*item));
826 0 : if (!item) {
827 0 : return -ENOMEM;
828 : }
829 0 : item->name = strdup(name);
830 0 : if (!item->name) {
831 0 : free(item);
832 0 : return -ENOMEM;
833 : }
834 0 : TAILQ_INSERT_TAIL(&g_bdev_examine_allowlist, item, link);
835 :
836 0 : bdev = spdk_bdev_get_by_name(name);
837 0 : if (bdev) {
838 0 : bdev_examine(bdev);
839 : }
840 0 : return 0;
841 : }
842 :
843 : static inline void
844 0 : bdev_examine_allowlist_config_json(struct spdk_json_write_ctx *w)
845 : {
846 : struct spdk_bdev_examine_item *item;
847 0 : TAILQ_FOREACH(item, &g_bdev_examine_allowlist, link) {
848 0 : spdk_json_write_object_begin(w);
849 0 : spdk_json_write_named_string(w, "method", "bdev_examine");
850 0 : spdk_json_write_named_object_begin(w, "params");
851 0 : spdk_json_write_named_string(w, "name", item->name);
852 0 : spdk_json_write_object_end(w);
853 0 : spdk_json_write_object_end(w);
854 : }
855 0 : }
856 :
857 : struct spdk_bdev *
858 1 : spdk_bdev_first(void)
859 : {
860 : struct spdk_bdev *bdev;
861 :
862 1 : bdev = TAILQ_FIRST(&g_bdev_mgr.bdevs);
863 1 : if (bdev) {
864 1 : SPDK_DEBUGLOG(bdev, "Starting bdev iteration at %s\n", bdev->name);
865 : }
866 :
867 1 : return bdev;
868 : }
869 :
870 : struct spdk_bdev *
871 8 : spdk_bdev_next(struct spdk_bdev *prev)
872 : {
873 : struct spdk_bdev *bdev;
874 :
875 8 : bdev = TAILQ_NEXT(prev, internal.link);
876 8 : if (bdev) {
877 7 : SPDK_DEBUGLOG(bdev, "Continuing bdev iteration at %s\n", bdev->name);
878 : }
879 :
880 8 : return bdev;
881 : }
882 :
883 : static struct spdk_bdev *
884 6 : _bdev_next_leaf(struct spdk_bdev *bdev)
885 : {
886 9 : while (bdev != NULL) {
887 8 : if (bdev->internal.claim_type == SPDK_BDEV_CLAIM_NONE) {
888 5 : return bdev;
889 : } else {
890 3 : bdev = TAILQ_NEXT(bdev, internal.link);
891 : }
892 : }
893 :
894 1 : return bdev;
895 : }
896 :
897 : struct spdk_bdev *
898 1 : spdk_bdev_first_leaf(void)
899 : {
900 : struct spdk_bdev *bdev;
901 :
902 1 : bdev = _bdev_next_leaf(TAILQ_FIRST(&g_bdev_mgr.bdevs));
903 :
904 1 : if (bdev) {
905 1 : SPDK_DEBUGLOG(bdev, "Starting bdev iteration at %s\n", bdev->name);
906 : }
907 :
908 1 : return bdev;
909 : }
910 :
911 : struct spdk_bdev *
912 5 : spdk_bdev_next_leaf(struct spdk_bdev *prev)
913 : {
914 : struct spdk_bdev *bdev;
915 :
916 5 : bdev = _bdev_next_leaf(TAILQ_NEXT(prev, internal.link));
917 :
918 5 : if (bdev) {
919 4 : SPDK_DEBUGLOG(bdev, "Continuing bdev iteration at %s\n", bdev->name);
920 : }
921 :
922 5 : return bdev;
923 : }
924 :
925 : static inline bool
926 821 : bdev_io_use_memory_domain(struct spdk_bdev_io *bdev_io)
927 : {
928 821 : return bdev_io->internal.f.has_memory_domain;
929 : }
930 :
931 : static inline bool
932 1553 : bdev_io_use_accel_sequence(struct spdk_bdev_io *bdev_io)
933 : {
934 1553 : return bdev_io->internal.f.has_accel_sequence;
935 : }
936 :
937 : static inline void
938 7 : bdev_queue_nomem_io_head(struct spdk_bdev_shared_resource *shared_resource,
939 : struct spdk_bdev_io *bdev_io, enum bdev_io_retry_state state)
940 : {
941 : /* Wait for some of the outstanding I/O to complete before we retry any of the nomem_io.
942 : * Normally we will wait for NOMEM_THRESHOLD_COUNT I/O to complete but for low queue depth
943 : * channels we will instead wait for half to complete.
944 : */
945 7 : shared_resource->nomem_threshold = spdk_max((int64_t)shared_resource->io_outstanding / 2,
946 : (int64_t)shared_resource->io_outstanding - NOMEM_THRESHOLD_COUNT);
947 :
948 7 : assert(state != BDEV_IO_RETRY_STATE_INVALID);
949 7 : bdev_io->internal.retry_state = state;
950 7 : TAILQ_INSERT_HEAD(&shared_resource->nomem_io, bdev_io, internal.link);
951 7 : }
952 :
953 : static inline void
954 43 : bdev_queue_nomem_io_tail(struct spdk_bdev_shared_resource *shared_resource,
955 : struct spdk_bdev_io *bdev_io, enum bdev_io_retry_state state)
956 : {
957 : /* We only queue IOs at the end of the nomem_io queue if they're submitted by the user while
958 : * the queue isn't empty, so we don't need to update the nomem_threshold here */
959 43 : assert(!TAILQ_EMPTY(&shared_resource->nomem_io));
960 :
961 43 : assert(state != BDEV_IO_RETRY_STATE_INVALID);
962 43 : bdev_io->internal.retry_state = state;
963 43 : TAILQ_INSERT_TAIL(&shared_resource->nomem_io, bdev_io, internal.link);
964 43 : }
965 :
966 : void
967 18 : spdk_bdev_io_set_buf(struct spdk_bdev_io *bdev_io, void *buf, size_t len)
968 : {
969 : struct iovec *iovs;
970 :
971 18 : if (bdev_io->u.bdev.iovs == NULL) {
972 3 : bdev_io->u.bdev.iovs = &bdev_io->iov;
973 3 : bdev_io->u.bdev.iovcnt = 1;
974 : }
975 :
976 18 : iovs = bdev_io->u.bdev.iovs;
977 :
978 18 : assert(iovs != NULL);
979 18 : assert(bdev_io->u.bdev.iovcnt >= 1);
980 :
981 18 : iovs[0].iov_base = buf;
982 18 : iovs[0].iov_len = len;
983 18 : }
984 :
985 : void
986 3 : spdk_bdev_io_set_md_buf(struct spdk_bdev_io *bdev_io, void *md_buf, size_t len)
987 : {
988 3 : assert((len / spdk_bdev_get_md_size(bdev_io->bdev)) >= bdev_io->u.bdev.num_blocks);
989 3 : bdev_io->u.bdev.md_buf = md_buf;
990 3 : }
991 :
992 : static bool
993 179 : _is_buf_allocated(const struct iovec *iovs)
994 : {
995 179 : if (iovs == NULL) {
996 6 : return false;
997 : }
998 :
999 173 : return iovs[0].iov_base != NULL;
1000 : }
1001 :
1002 : static bool
1003 56 : _are_iovs_aligned(struct iovec *iovs, int iovcnt, uint32_t alignment)
1004 : {
1005 : int i;
1006 : uintptr_t iov_base;
1007 :
1008 56 : if (spdk_likely(alignment == 1)) {
1009 21 : return true;
1010 : }
1011 :
1012 46 : for (i = 0; i < iovcnt; i++) {
1013 35 : iov_base = (uintptr_t)iovs[i].iov_base;
1014 35 : if ((iov_base & (alignment - 1)) != 0) {
1015 24 : return false;
1016 : }
1017 : }
1018 :
1019 11 : return true;
1020 : }
1021 :
1022 : static inline bool
1023 863 : bdev_io_needs_sequence_exec(struct spdk_bdev_desc *desc, struct spdk_bdev_io *bdev_io)
1024 : {
1025 863 : if (!bdev_io_use_accel_sequence(bdev_io)) {
1026 863 : return false;
1027 : }
1028 :
1029 : /* For now, we don't allow splitting IOs with an accel sequence and will treat them as if
1030 : * bdev module didn't support accel sequences */
1031 0 : return !desc->accel_sequence_supported[bdev_io->type] || bdev_io->internal.f.split;
1032 : }
1033 :
1034 : static inline void
1035 585 : bdev_io_increment_outstanding(struct spdk_bdev_channel *bdev_ch,
1036 : struct spdk_bdev_shared_resource *shared_resource)
1037 : {
1038 585 : bdev_ch->io_outstanding++;
1039 585 : shared_resource->io_outstanding++;
1040 585 : }
1041 :
1042 : static inline void
1043 585 : bdev_io_decrement_outstanding(struct spdk_bdev_channel *bdev_ch,
1044 : struct spdk_bdev_shared_resource *shared_resource)
1045 : {
1046 585 : assert(bdev_ch->io_outstanding > 0);
1047 585 : assert(shared_resource->io_outstanding > 0);
1048 585 : bdev_ch->io_outstanding--;
1049 585 : shared_resource->io_outstanding--;
1050 585 : }
1051 :
1052 : static void
1053 0 : bdev_io_submit_sequence_cb(void *ctx, int status)
1054 : {
1055 0 : struct spdk_bdev_io *bdev_io = ctx;
1056 :
1057 0 : assert(bdev_io_use_accel_sequence(bdev_io));
1058 :
1059 0 : bdev_io->u.bdev.accel_sequence = NULL;
1060 0 : bdev_io->internal.f.has_accel_sequence = false;
1061 :
1062 0 : if (spdk_unlikely(status != 0)) {
1063 0 : SPDK_ERRLOG("Failed to execute accel sequence, status=%d\n", status);
1064 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
1065 0 : bdev_io_complete_unsubmitted(bdev_io);
1066 0 : return;
1067 : }
1068 :
1069 0 : bdev_io_submit(bdev_io);
1070 : }
1071 :
1072 : static void
1073 0 : bdev_io_exec_sequence_cb(void *ctx, int status)
1074 : {
1075 0 : struct spdk_bdev_io *bdev_io = ctx;
1076 0 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1077 :
1078 0 : TAILQ_REMOVE(&bdev_io->internal.ch->io_accel_exec, bdev_io, internal.link);
1079 0 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1080 :
1081 0 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1082 0 : bdev_ch_retry_io(ch);
1083 : }
1084 :
1085 0 : bdev_io->internal.data_transfer_cpl(bdev_io, status);
1086 0 : }
1087 :
1088 : static void
1089 0 : bdev_io_exec_sequence(struct spdk_bdev_io *bdev_io, void (*cb_fn)(void *ctx, int status))
1090 : {
1091 0 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1092 :
1093 0 : assert(bdev_io_needs_sequence_exec(bdev_io->internal.desc, bdev_io));
1094 0 : assert(bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE || bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
1095 0 : assert(bdev_io_use_accel_sequence(bdev_io));
1096 :
1097 : /* Since the operations are appended during submission, they're in the opposite order than
1098 : * how we want to execute them for reads (i.e. we need to execute the most recently added
1099 : * operation first), so reverse the sequence before executing it.
1100 : */
1101 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_READ) {
1102 0 : spdk_accel_sequence_reverse(bdev_io->internal.accel_sequence);
1103 : }
1104 :
1105 0 : TAILQ_INSERT_TAIL(&bdev_io->internal.ch->io_accel_exec, bdev_io, internal.link);
1106 0 : bdev_io_increment_outstanding(ch, ch->shared_resource);
1107 0 : bdev_io->internal.data_transfer_cpl = cb_fn;
1108 :
1109 0 : spdk_accel_sequence_finish(bdev_io->internal.accel_sequence,
1110 : bdev_io_exec_sequence_cb, bdev_io);
1111 0 : }
1112 :
1113 : static void
1114 46 : bdev_io_get_buf_complete(struct spdk_bdev_io *bdev_io, bool status)
1115 : {
1116 46 : struct spdk_io_channel *ch = spdk_bdev_io_get_io_channel(bdev_io);
1117 : void *buf;
1118 :
1119 46 : if (spdk_unlikely(bdev_io->internal.get_aux_buf_cb != NULL)) {
1120 0 : buf = bdev_io->internal.buf.ptr;
1121 0 : bdev_io->internal.buf.ptr = NULL;
1122 0 : bdev_io->internal.f.has_buf = false;
1123 0 : bdev_io->internal.get_aux_buf_cb(ch, bdev_io, buf);
1124 0 : bdev_io->internal.get_aux_buf_cb = NULL;
1125 : } else {
1126 46 : assert(bdev_io->internal.get_buf_cb != NULL);
1127 46 : bdev_io->internal.get_buf_cb(ch, bdev_io, status);
1128 46 : bdev_io->internal.get_buf_cb = NULL;
1129 : }
1130 46 : }
1131 :
1132 : static void
1133 4 : _bdev_io_pull_buffer_cpl(void *ctx, int rc)
1134 : {
1135 4 : struct spdk_bdev_io *bdev_io = ctx;
1136 :
1137 4 : if (rc) {
1138 0 : SPDK_ERRLOG("Set bounce buffer failed with rc %d\n", rc);
1139 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
1140 : }
1141 4 : bdev_io_get_buf_complete(bdev_io, !rc);
1142 4 : }
1143 :
1144 : static void
1145 2 : bdev_io_pull_md_buf_done(void *ctx, int status)
1146 : {
1147 2 : struct spdk_bdev_io *bdev_io = ctx;
1148 2 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1149 :
1150 2 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1151 2 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1152 :
1153 2 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1154 0 : bdev_ch_retry_io(ch);
1155 : }
1156 :
1157 2 : assert(bdev_io->internal.data_transfer_cpl);
1158 2 : bdev_io->internal.data_transfer_cpl(bdev_io, status);
1159 2 : }
1160 :
1161 : static void
1162 4 : bdev_io_pull_md_buf(struct spdk_bdev_io *bdev_io)
1163 : {
1164 4 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1165 4 : int rc = 0;
1166 :
1167 4 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
1168 2 : if (bdev_io_use_memory_domain(bdev_io)) {
1169 2 : TAILQ_INSERT_TAIL(&ch->io_memory_domain, bdev_io, internal.link);
1170 2 : bdev_io_increment_outstanding(ch, ch->shared_resource);
1171 2 : rc = spdk_memory_domain_pull_data(bdev_io->internal.memory_domain,
1172 : bdev_io->internal.memory_domain_ctx,
1173 : &bdev_io->internal.orig_md_iov, 1,
1174 : &bdev_io->internal.bounce_md_iov, 1,
1175 : bdev_io_pull_md_buf_done, bdev_io);
1176 2 : if (rc == 0) {
1177 : /* Continue to submit IO in completion callback */
1178 2 : return;
1179 : }
1180 0 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1181 0 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1182 0 : if (rc != -ENOMEM) {
1183 0 : SPDK_ERRLOG("Failed to pull data from memory domain %s, rc %d\n",
1184 : spdk_memory_domain_get_dma_device_id(
1185 : bdev_io->internal.memory_domain), rc);
1186 : }
1187 : } else {
1188 0 : memcpy(bdev_io->internal.bounce_md_iov.iov_base,
1189 0 : bdev_io->internal.orig_md_iov.iov_base,
1190 : bdev_io->internal.orig_md_iov.iov_len);
1191 : }
1192 : }
1193 :
1194 2 : if (spdk_unlikely(rc == -ENOMEM)) {
1195 0 : bdev_queue_nomem_io_head(ch->shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PULL_MD);
1196 : } else {
1197 2 : assert(bdev_io->internal.data_transfer_cpl);
1198 2 : bdev_io->internal.data_transfer_cpl(bdev_io, rc);
1199 : }
1200 : }
1201 :
1202 : static void
1203 4 : _bdev_io_pull_bounce_md_buf(struct spdk_bdev_io *bdev_io, void *md_buf, size_t len)
1204 : {
1205 : /* save original md_buf */
1206 4 : bdev_io->internal.orig_md_iov.iov_base = bdev_io->u.bdev.md_buf;
1207 4 : bdev_io->internal.orig_md_iov.iov_len = len;
1208 4 : bdev_io->internal.bounce_md_iov.iov_base = md_buf;
1209 4 : bdev_io->internal.bounce_md_iov.iov_len = len;
1210 : /* set bounce md_buf */
1211 4 : bdev_io->u.bdev.md_buf = md_buf;
1212 :
1213 4 : bdev_io_pull_md_buf(bdev_io);
1214 4 : }
1215 :
1216 : static void
1217 46 : _bdev_io_set_md_buf(struct spdk_bdev_io *bdev_io)
1218 : {
1219 46 : struct spdk_bdev *bdev = bdev_io->bdev;
1220 : uint64_t md_len;
1221 : void *buf;
1222 :
1223 46 : if (spdk_bdev_is_md_separate(bdev)) {
1224 7 : assert(!bdev_io_use_accel_sequence(bdev_io));
1225 :
1226 7 : buf = (char *)bdev_io->u.bdev.iovs[0].iov_base + bdev_io->u.bdev.iovs[0].iov_len;
1227 7 : md_len = bdev_io->u.bdev.num_blocks * bdev->md_len;
1228 :
1229 7 : assert(((uintptr_t)buf & (spdk_bdev_get_buf_align(bdev) - 1)) == 0);
1230 :
1231 7 : if (bdev_io->u.bdev.md_buf != NULL) {
1232 4 : _bdev_io_pull_bounce_md_buf(bdev_io, buf, md_len);
1233 4 : return;
1234 : } else {
1235 3 : spdk_bdev_io_set_md_buf(bdev_io, buf, md_len);
1236 : }
1237 : }
1238 :
1239 42 : bdev_io_get_buf_complete(bdev_io, true);
1240 : }
1241 :
1242 : static inline void
1243 28 : bdev_io_pull_data_done(struct spdk_bdev_io *bdev_io, int rc)
1244 : {
1245 28 : if (rc) {
1246 0 : SPDK_ERRLOG("Failed to get data buffer\n");
1247 0 : assert(bdev_io->internal.data_transfer_cpl);
1248 0 : bdev_io->internal.data_transfer_cpl(bdev_io, rc);
1249 0 : return;
1250 : }
1251 :
1252 28 : _bdev_io_set_md_buf(bdev_io);
1253 : }
1254 :
1255 : static void
1256 2 : bdev_io_pull_data_done_and_track(void *ctx, int status)
1257 : {
1258 2 : struct spdk_bdev_io *bdev_io = ctx;
1259 2 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1260 :
1261 2 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1262 2 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1263 :
1264 2 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1265 0 : bdev_ch_retry_io(ch);
1266 : }
1267 :
1268 2 : bdev_io_pull_data_done(bdev_io, status);
1269 2 : }
1270 :
1271 : static void
1272 29 : bdev_io_pull_data(struct spdk_bdev_io *bdev_io)
1273 : {
1274 29 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1275 29 : int rc = 0;
1276 :
1277 : /* If we need to exec an accel sequence or the IO uses a memory domain buffer and has a
1278 : * sequence, append a copy operation making accel change the src/dst buffers of the previous
1279 : * operation */
1280 58 : if (bdev_io_needs_sequence_exec(bdev_io->internal.desc, bdev_io) ||
1281 29 : (bdev_io_use_accel_sequence(bdev_io) && bdev_io_use_memory_domain(bdev_io))) {
1282 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
1283 0 : assert(bdev_io_use_accel_sequence(bdev_io));
1284 0 : rc = spdk_accel_append_copy(&bdev_io->internal.accel_sequence, ch->accel_channel,
1285 0 : bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt,
1286 : NULL, NULL,
1287 : bdev_io->internal.orig_iovs,
1288 0 : bdev_io->internal.orig_iovcnt,
1289 0 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain : NULL,
1290 0 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain_ctx : NULL,
1291 : NULL, NULL);
1292 : } else {
1293 : /* We need to reverse the src/dst for reads */
1294 0 : assert(bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
1295 0 : assert(bdev_io_use_accel_sequence(bdev_io));
1296 0 : rc = spdk_accel_append_copy(&bdev_io->internal.accel_sequence, ch->accel_channel,
1297 : bdev_io->internal.orig_iovs,
1298 0 : bdev_io->internal.orig_iovcnt,
1299 0 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain : NULL,
1300 0 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain_ctx : NULL,
1301 0 : bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt,
1302 : NULL, NULL, NULL, NULL);
1303 : }
1304 :
1305 0 : if (spdk_unlikely(rc != 0 && rc != -ENOMEM)) {
1306 0 : SPDK_ERRLOG("Failed to append copy to accel sequence: %p\n",
1307 : bdev_io->internal.accel_sequence);
1308 : }
1309 29 : } else if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
1310 : /* if this is write path, copy data from original buffer to bounce buffer */
1311 17 : if (bdev_io_use_memory_domain(bdev_io)) {
1312 3 : TAILQ_INSERT_TAIL(&ch->io_memory_domain, bdev_io, internal.link);
1313 3 : bdev_io_increment_outstanding(ch, ch->shared_resource);
1314 3 : rc = spdk_memory_domain_pull_data(bdev_io->internal.memory_domain,
1315 : bdev_io->internal.memory_domain_ctx,
1316 : bdev_io->internal.orig_iovs,
1317 3 : (uint32_t) bdev_io->internal.orig_iovcnt,
1318 : bdev_io->u.bdev.iovs, 1,
1319 : bdev_io_pull_data_done_and_track,
1320 : bdev_io);
1321 3 : if (rc == 0) {
1322 : /* Continue to submit IO in completion callback */
1323 2 : return;
1324 : }
1325 1 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1326 1 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1327 1 : if (rc != -ENOMEM) {
1328 0 : SPDK_ERRLOG("Failed to pull data from memory domain %s\n",
1329 : spdk_memory_domain_get_dma_device_id(
1330 : bdev_io->internal.memory_domain));
1331 : }
1332 : } else {
1333 14 : assert(bdev_io->u.bdev.iovcnt == 1);
1334 14 : spdk_copy_iovs_to_buf(bdev_io->u.bdev.iovs[0].iov_base,
1335 14 : bdev_io->u.bdev.iovs[0].iov_len,
1336 : bdev_io->internal.orig_iovs,
1337 : bdev_io->internal.orig_iovcnt);
1338 : }
1339 : }
1340 :
1341 27 : if (spdk_unlikely(rc == -ENOMEM)) {
1342 1 : bdev_queue_nomem_io_head(ch->shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PULL);
1343 : } else {
1344 26 : bdev_io_pull_data_done(bdev_io, rc);
1345 : }
1346 : }
1347 :
1348 : static void
1349 28 : _bdev_io_pull_bounce_data_buf(struct spdk_bdev_io *bdev_io, void *buf, size_t len,
1350 : bdev_copy_bounce_buffer_cpl cpl_cb)
1351 : {
1352 28 : struct spdk_bdev_shared_resource *shared_resource = bdev_io->internal.ch->shared_resource;
1353 :
1354 28 : bdev_io->internal.data_transfer_cpl = cpl_cb;
1355 : /* save original iovec */
1356 28 : bdev_io->internal.orig_iovs = bdev_io->u.bdev.iovs;
1357 28 : bdev_io->internal.orig_iovcnt = bdev_io->u.bdev.iovcnt;
1358 : /* set bounce iov */
1359 28 : bdev_io->u.bdev.iovs = &bdev_io->internal.bounce_iov;
1360 28 : bdev_io->u.bdev.iovcnt = 1;
1361 : /* set bounce buffer for this operation */
1362 28 : bdev_io->u.bdev.iovs[0].iov_base = buf;
1363 28 : bdev_io->u.bdev.iovs[0].iov_len = len;
1364 :
1365 28 : if (spdk_unlikely(!TAILQ_EMPTY(&shared_resource->nomem_io))) {
1366 0 : bdev_queue_nomem_io_tail(shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PULL);
1367 : } else {
1368 28 : bdev_io_pull_data(bdev_io);
1369 : }
1370 28 : }
1371 :
1372 : static void
1373 46 : _bdev_io_set_buf(struct spdk_bdev_io *bdev_io, void *buf, uint64_t len)
1374 : {
1375 46 : struct spdk_bdev *bdev = bdev_io->bdev;
1376 : bool buf_allocated;
1377 : uint64_t alignment;
1378 : void *aligned_buf;
1379 :
1380 46 : bdev_io->internal.buf.ptr = buf;
1381 46 : bdev_io->internal.f.has_buf = true;
1382 :
1383 46 : if (spdk_unlikely(bdev_io->internal.get_aux_buf_cb != NULL)) {
1384 0 : bdev_io_get_buf_complete(bdev_io, true);
1385 0 : return;
1386 : }
1387 :
1388 46 : alignment = spdk_bdev_get_buf_align(bdev);
1389 46 : buf_allocated = _is_buf_allocated(bdev_io->u.bdev.iovs);
1390 46 : aligned_buf = (void *)(((uintptr_t)buf + (alignment - 1)) & ~(alignment - 1));
1391 :
1392 46 : if (buf_allocated) {
1393 28 : _bdev_io_pull_bounce_data_buf(bdev_io, aligned_buf, len, _bdev_io_pull_buffer_cpl);
1394 : /* Continue in completion callback */
1395 28 : return;
1396 : } else {
1397 18 : spdk_bdev_io_set_buf(bdev_io, aligned_buf, len);
1398 : }
1399 :
1400 18 : _bdev_io_set_md_buf(bdev_io);
1401 : }
1402 :
1403 : static inline uint64_t
1404 92 : bdev_io_get_max_buf_len(struct spdk_bdev_io *bdev_io, uint64_t len)
1405 : {
1406 92 : struct spdk_bdev *bdev = bdev_io->bdev;
1407 : uint64_t md_len, alignment;
1408 :
1409 92 : md_len = spdk_bdev_is_md_separate(bdev) ? bdev_io->u.bdev.num_blocks * bdev->md_len : 0;
1410 :
1411 : /* 1 byte alignment needs 0 byte of extra space, 64 bytes alignment needs 63 bytes of extra space, etc. */
1412 92 : alignment = spdk_bdev_get_buf_align(bdev) - 1;
1413 :
1414 92 : return len + alignment + md_len;
1415 : }
1416 :
1417 : static void
1418 46 : _bdev_io_put_buf(struct spdk_bdev_io *bdev_io, void *buf, uint64_t buf_len)
1419 : {
1420 : struct spdk_bdev_mgmt_channel *ch;
1421 :
1422 46 : ch = bdev_io->internal.ch->shared_resource->mgmt_ch;
1423 46 : spdk_iobuf_put(&ch->iobuf, buf, bdev_io_get_max_buf_len(bdev_io, buf_len));
1424 46 : }
1425 :
1426 : static void
1427 46 : bdev_io_put_buf(struct spdk_bdev_io *bdev_io)
1428 : {
1429 46 : assert(bdev_io->internal.f.has_buf);
1430 46 : _bdev_io_put_buf(bdev_io, bdev_io->internal.buf.ptr, bdev_io->internal.buf.len);
1431 46 : bdev_io->internal.buf.ptr = NULL;
1432 46 : bdev_io->internal.f.has_buf = false;
1433 46 : }
1434 :
1435 : void
1436 0 : spdk_bdev_io_put_aux_buf(struct spdk_bdev_io *bdev_io, void *buf)
1437 : {
1438 0 : uint64_t len = bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen;
1439 :
1440 0 : assert(buf != NULL);
1441 0 : _bdev_io_put_buf(bdev_io, buf, len);
1442 0 : }
1443 :
1444 : static inline void
1445 541 : bdev_submit_request(struct spdk_bdev *bdev, struct spdk_io_channel *ioch,
1446 : struct spdk_bdev_io *bdev_io)
1447 : {
1448 : /* After a request is submitted to a bdev module, the ownership of an accel sequence
1449 : * associated with that bdev_io is transferred to the bdev module. So, clear the internal
1450 : * sequence pointer to make sure we won't touch it anymore. */
1451 541 : if ((bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE ||
1452 541 : bdev_io->type == SPDK_BDEV_IO_TYPE_READ) && bdev_io->u.bdev.accel_sequence != NULL) {
1453 0 : assert(!bdev_io_needs_sequence_exec(bdev_io->internal.desc, bdev_io));
1454 0 : bdev_io->internal.f.has_accel_sequence = false;
1455 : }
1456 :
1457 541 : bdev->fn_table->submit_request(ioch, bdev_io);
1458 541 : }
1459 :
1460 : static inline void
1461 10 : bdev_ch_resubmit_io(struct spdk_bdev_shared_resource *shared_resource, struct spdk_bdev_io *bdev_io)
1462 : {
1463 10 : struct spdk_bdev *bdev = bdev_io->bdev;
1464 :
1465 10 : bdev_io_increment_outstanding(bdev_io->internal.ch, shared_resource);
1466 10 : bdev_io->internal.error.nvme.cdw0 = 0;
1467 10 : bdev_io->num_retries++;
1468 10 : bdev_submit_request(bdev, spdk_bdev_io_get_io_channel(bdev_io), bdev_io);
1469 10 : }
1470 :
1471 : static void
1472 63 : bdev_shared_ch_retry_io(struct spdk_bdev_shared_resource *shared_resource)
1473 : {
1474 : struct spdk_bdev_io *bdev_io;
1475 :
1476 63 : if (shared_resource->io_outstanding > shared_resource->nomem_threshold) {
1477 : /*
1478 : * Allow some more I/O to complete before retrying the nomem_io queue.
1479 : * Some drivers (such as nvme) cannot immediately take a new I/O in
1480 : * the context of a completion, because the resources for the I/O are
1481 : * not released until control returns to the bdev poller. Also, we
1482 : * may require several small I/O to complete before a larger I/O
1483 : * (that requires splitting) can be submitted.
1484 : */
1485 58 : return;
1486 : }
1487 :
1488 16 : while (!TAILQ_EMPTY(&shared_resource->nomem_io)) {
1489 12 : bdev_io = TAILQ_FIRST(&shared_resource->nomem_io);
1490 12 : TAILQ_REMOVE(&shared_resource->nomem_io, bdev_io, internal.link);
1491 :
1492 12 : switch (bdev_io->internal.retry_state) {
1493 10 : case BDEV_IO_RETRY_STATE_SUBMIT:
1494 10 : bdev_ch_resubmit_io(shared_resource, bdev_io);
1495 10 : break;
1496 1 : case BDEV_IO_RETRY_STATE_PULL:
1497 1 : bdev_io_pull_data(bdev_io);
1498 1 : break;
1499 0 : case BDEV_IO_RETRY_STATE_PULL_MD:
1500 0 : bdev_io_pull_md_buf(bdev_io);
1501 0 : break;
1502 1 : case BDEV_IO_RETRY_STATE_PUSH:
1503 1 : bdev_io_push_bounce_data(bdev_io);
1504 1 : break;
1505 0 : case BDEV_IO_RETRY_STATE_PUSH_MD:
1506 0 : bdev_io_push_bounce_md_buf(bdev_io);
1507 0 : break;
1508 0 : default:
1509 0 : assert(0 && "invalid retry state");
1510 : break;
1511 : }
1512 :
1513 12 : if (bdev_io == TAILQ_FIRST(&shared_resource->nomem_io)) {
1514 : /* This IO completed again with NOMEM status, so break the loop and
1515 : * don't try anymore. Note that a bdev_io that fails with NOMEM
1516 : * always gets requeued at the front of the list, to maintain
1517 : * ordering.
1518 : */
1519 1 : break;
1520 : }
1521 : }
1522 : }
1523 :
1524 : static void
1525 63 : bdev_ch_retry_io(struct spdk_bdev_channel *bdev_ch)
1526 : {
1527 63 : bdev_shared_ch_retry_io(bdev_ch->shared_resource);
1528 63 : }
1529 :
1530 : static int
1531 0 : bdev_no_mem_poller(void *ctx)
1532 : {
1533 0 : struct spdk_bdev_shared_resource *shared_resource = ctx;
1534 :
1535 0 : spdk_poller_unregister(&shared_resource->nomem_poller);
1536 :
1537 0 : if (!TAILQ_EMPTY(&shared_resource->nomem_io)) {
1538 0 : bdev_shared_ch_retry_io(shared_resource);
1539 : }
1540 : /* the retry cb may re-register the poller so double check */
1541 0 : if (!TAILQ_EMPTY(&shared_resource->nomem_io) &&
1542 0 : shared_resource->io_outstanding == 0 && shared_resource->nomem_poller == NULL) {
1543 : /* No IOs were submitted, try again */
1544 0 : shared_resource->nomem_poller = SPDK_POLLER_REGISTER(bdev_no_mem_poller, shared_resource,
1545 : SPDK_BDEV_IO_POLL_INTERVAL_IN_MSEC * 10);
1546 : }
1547 :
1548 0 : return SPDK_POLLER_BUSY;
1549 : }
1550 :
1551 : static inline bool
1552 547 : _bdev_io_handle_no_mem(struct spdk_bdev_io *bdev_io, enum bdev_io_retry_state state)
1553 : {
1554 547 : struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
1555 547 : struct spdk_bdev_shared_resource *shared_resource = bdev_ch->shared_resource;
1556 :
1557 547 : if (spdk_unlikely(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_NOMEM)) {
1558 5 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_PENDING;
1559 5 : bdev_queue_nomem_io_head(shared_resource, bdev_io, state);
1560 :
1561 5 : if (shared_resource->io_outstanding == 0 && !shared_resource->nomem_poller) {
1562 : /* Special case when we have nomem IOs and no outstanding IOs which completions
1563 : * could trigger retry of queued IOs
1564 : * Any IOs submitted may trigger retry of queued IOs. This poller handles a case when no
1565 : * new IOs submitted, e.g. qd==1 */
1566 0 : shared_resource->nomem_poller = SPDK_POLLER_REGISTER(bdev_no_mem_poller, shared_resource,
1567 : SPDK_BDEV_IO_POLL_INTERVAL_IN_MSEC * 10);
1568 : }
1569 : /* If bdev module completed an I/O that has an accel sequence with NOMEM status, the
1570 : * ownership of that sequence is transferred back to the bdev layer, so we need to
1571 : * restore internal.accel_sequence to make sure that the sequence is handled
1572 : * correctly in case the I/O is later aborted. */
1573 5 : if ((bdev_io->type == SPDK_BDEV_IO_TYPE_READ ||
1574 5 : bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) && bdev_io->u.bdev.accel_sequence) {
1575 0 : assert(!bdev_io_use_accel_sequence(bdev_io));
1576 0 : bdev_io->internal.f.has_accel_sequence = true;
1577 0 : bdev_io->internal.accel_sequence = bdev_io->u.bdev.accel_sequence;
1578 : }
1579 :
1580 5 : return true;
1581 : }
1582 :
1583 542 : if (spdk_unlikely(!TAILQ_EMPTY(&shared_resource->nomem_io))) {
1584 63 : bdev_ch_retry_io(bdev_ch);
1585 : }
1586 :
1587 542 : return false;
1588 : }
1589 :
1590 : static void
1591 28 : _bdev_io_complete_push_bounce_done(void *ctx, int rc)
1592 : {
1593 28 : struct spdk_bdev_io *bdev_io = ctx;
1594 28 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1595 :
1596 28 : if (rc) {
1597 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
1598 : }
1599 : /* We want to free the bounce buffer here since we know we're done with it (as opposed
1600 : * to waiting for the conditional free of internal.buf.ptr in spdk_bdev_free_io()).
1601 : */
1602 28 : bdev_io_put_buf(bdev_io);
1603 :
1604 28 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1605 0 : bdev_ch_retry_io(ch);
1606 : }
1607 :
1608 : /* Continue with IO completion flow */
1609 28 : bdev_io_complete(bdev_io);
1610 28 : }
1611 :
1612 : static void
1613 2 : bdev_io_push_bounce_md_buf_done(void *ctx, int rc)
1614 : {
1615 2 : struct spdk_bdev_io *bdev_io = ctx;
1616 2 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1617 :
1618 2 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1619 2 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1620 :
1621 2 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1622 0 : bdev_ch_retry_io(ch);
1623 : }
1624 :
1625 2 : bdev_io->internal.data_transfer_cpl(bdev_io, rc);
1626 2 : }
1627 :
1628 : static inline void
1629 28 : bdev_io_push_bounce_md_buf(struct spdk_bdev_io *bdev_io)
1630 : {
1631 28 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1632 28 : int rc = 0;
1633 :
1634 28 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS);
1635 : /* do the same for metadata buffer */
1636 28 : if (spdk_unlikely(bdev_io->internal.orig_md_iov.iov_base != NULL)) {
1637 4 : assert(spdk_bdev_is_md_separate(bdev_io->bdev));
1638 :
1639 4 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_READ) {
1640 2 : if (bdev_io_use_memory_domain(bdev_io)) {
1641 2 : TAILQ_INSERT_TAIL(&ch->io_memory_domain, bdev_io, internal.link);
1642 2 : bdev_io_increment_outstanding(ch, ch->shared_resource);
1643 : /* If memory domain is used then we need to call async push function */
1644 2 : rc = spdk_memory_domain_push_data(bdev_io->internal.memory_domain,
1645 : bdev_io->internal.memory_domain_ctx,
1646 : &bdev_io->internal.orig_md_iov,
1647 2 : (uint32_t)bdev_io->internal.orig_iovcnt,
1648 : &bdev_io->internal.bounce_md_iov, 1,
1649 : bdev_io_push_bounce_md_buf_done,
1650 : bdev_io);
1651 2 : if (rc == 0) {
1652 : /* Continue IO completion in async callback */
1653 2 : return;
1654 : }
1655 0 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1656 0 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1657 0 : if (rc != -ENOMEM) {
1658 0 : SPDK_ERRLOG("Failed to push md to memory domain %s\n",
1659 : spdk_memory_domain_get_dma_device_id(
1660 : bdev_io->internal.memory_domain));
1661 : }
1662 : } else {
1663 0 : memcpy(bdev_io->internal.orig_md_iov.iov_base, bdev_io->u.bdev.md_buf,
1664 : bdev_io->internal.orig_md_iov.iov_len);
1665 : }
1666 : }
1667 : }
1668 :
1669 26 : if (spdk_unlikely(rc == -ENOMEM)) {
1670 0 : bdev_queue_nomem_io_head(ch->shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PUSH_MD);
1671 : } else {
1672 26 : assert(bdev_io->internal.data_transfer_cpl);
1673 26 : bdev_io->internal.data_transfer_cpl(bdev_io, rc);
1674 : }
1675 : }
1676 :
1677 : static inline void
1678 28 : bdev_io_push_bounce_data_done(struct spdk_bdev_io *bdev_io, int rc)
1679 : {
1680 28 : assert(bdev_io->internal.data_transfer_cpl);
1681 28 : if (rc) {
1682 0 : bdev_io->internal.data_transfer_cpl(bdev_io, rc);
1683 0 : return;
1684 : }
1685 :
1686 : /* set original buffer for this io */
1687 28 : bdev_io->u.bdev.iovcnt = bdev_io->internal.orig_iovcnt;
1688 28 : bdev_io->u.bdev.iovs = bdev_io->internal.orig_iovs;
1689 : /* disable bouncing buffer for this io */
1690 28 : bdev_io->internal.orig_iovcnt = 0;
1691 28 : bdev_io->internal.orig_iovs = NULL;
1692 :
1693 28 : bdev_io_push_bounce_md_buf(bdev_io);
1694 : }
1695 :
1696 : static void
1697 2 : bdev_io_push_bounce_data_done_and_track(void *ctx, int status)
1698 : {
1699 2 : struct spdk_bdev_io *bdev_io = ctx;
1700 2 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1701 :
1702 2 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1703 2 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1704 :
1705 2 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1706 0 : bdev_ch_retry_io(ch);
1707 : }
1708 :
1709 2 : bdev_io_push_bounce_data_done(bdev_io, status);
1710 2 : }
1711 :
1712 : static inline void
1713 29 : bdev_io_push_bounce_data(struct spdk_bdev_io *bdev_io)
1714 : {
1715 29 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1716 29 : int rc = 0;
1717 :
1718 29 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS);
1719 29 : assert(!bdev_io_use_accel_sequence(bdev_io));
1720 :
1721 : /* if this is read path, copy data from bounce buffer to original buffer */
1722 29 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_READ) {
1723 13 : if (bdev_io_use_memory_domain(bdev_io)) {
1724 3 : TAILQ_INSERT_TAIL(&ch->io_memory_domain, bdev_io, internal.link);
1725 3 : bdev_io_increment_outstanding(ch, ch->shared_resource);
1726 : /* If memory domain is used then we need to call async push function */
1727 3 : rc = spdk_memory_domain_push_data(bdev_io->internal.memory_domain,
1728 : bdev_io->internal.memory_domain_ctx,
1729 : bdev_io->internal.orig_iovs,
1730 3 : (uint32_t)bdev_io->internal.orig_iovcnt,
1731 : &bdev_io->internal.bounce_iov, 1,
1732 : bdev_io_push_bounce_data_done_and_track,
1733 : bdev_io);
1734 3 : if (rc == 0) {
1735 : /* Continue IO completion in async callback */
1736 2 : return;
1737 : }
1738 :
1739 1 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1740 1 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1741 1 : if (rc != -ENOMEM) {
1742 0 : SPDK_ERRLOG("Failed to push data to memory domain %s\n",
1743 : spdk_memory_domain_get_dma_device_id(
1744 : bdev_io->internal.memory_domain));
1745 : }
1746 : } else {
1747 10 : spdk_copy_buf_to_iovs(bdev_io->internal.orig_iovs,
1748 : bdev_io->internal.orig_iovcnt,
1749 : bdev_io->internal.bounce_iov.iov_base,
1750 : bdev_io->internal.bounce_iov.iov_len);
1751 : }
1752 : }
1753 :
1754 27 : if (spdk_unlikely(rc == -ENOMEM)) {
1755 1 : bdev_queue_nomem_io_head(ch->shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PUSH);
1756 : } else {
1757 26 : bdev_io_push_bounce_data_done(bdev_io, rc);
1758 : }
1759 : }
1760 :
1761 : static inline void
1762 28 : _bdev_io_push_bounce_data_buffer(struct spdk_bdev_io *bdev_io, bdev_copy_bounce_buffer_cpl cpl_cb)
1763 : {
1764 28 : bdev_io->internal.data_transfer_cpl = cpl_cb;
1765 28 : bdev_io_push_bounce_data(bdev_io);
1766 28 : }
1767 :
1768 : static void
1769 0 : bdev_io_get_iobuf_cb(struct spdk_iobuf_entry *iobuf, void *buf)
1770 : {
1771 : struct spdk_bdev_io *bdev_io;
1772 :
1773 0 : bdev_io = SPDK_CONTAINEROF(iobuf, struct spdk_bdev_io, internal.iobuf);
1774 0 : _bdev_io_set_buf(bdev_io, buf, bdev_io->internal.buf.len);
1775 0 : }
1776 :
1777 : static void
1778 46 : bdev_io_get_buf(struct spdk_bdev_io *bdev_io, uint64_t len)
1779 : {
1780 : struct spdk_bdev_mgmt_channel *mgmt_ch;
1781 : uint64_t max_len;
1782 : void *buf;
1783 :
1784 46 : assert(spdk_bdev_io_get_thread(bdev_io) == spdk_get_thread());
1785 46 : mgmt_ch = bdev_io->internal.ch->shared_resource->mgmt_ch;
1786 46 : max_len = bdev_io_get_max_buf_len(bdev_io, len);
1787 :
1788 46 : if (spdk_unlikely(max_len > mgmt_ch->iobuf.large.bufsize)) {
1789 0 : SPDK_ERRLOG("Length %" PRIu64 " is larger than allowed\n", max_len);
1790 0 : bdev_io_get_buf_complete(bdev_io, false);
1791 0 : return;
1792 : }
1793 :
1794 46 : bdev_io->internal.buf.len = len;
1795 46 : buf = spdk_iobuf_get(&mgmt_ch->iobuf, max_len, &bdev_io->internal.iobuf,
1796 : bdev_io_get_iobuf_cb);
1797 46 : if (buf != NULL) {
1798 46 : _bdev_io_set_buf(bdev_io, buf, len);
1799 : }
1800 : }
1801 :
1802 : void
1803 64 : spdk_bdev_io_get_buf(struct spdk_bdev_io *bdev_io, spdk_bdev_io_get_buf_cb cb, uint64_t len)
1804 : {
1805 64 : struct spdk_bdev *bdev = bdev_io->bdev;
1806 : uint64_t alignment;
1807 :
1808 64 : assert(cb != NULL);
1809 64 : bdev_io->internal.get_buf_cb = cb;
1810 :
1811 64 : alignment = spdk_bdev_get_buf_align(bdev);
1812 :
1813 110 : if (_is_buf_allocated(bdev_io->u.bdev.iovs) &&
1814 46 : _are_iovs_aligned(bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt, alignment)) {
1815 : /* Buffer already present and aligned */
1816 22 : cb(spdk_bdev_io_get_io_channel(bdev_io), bdev_io, true);
1817 22 : return;
1818 : }
1819 :
1820 42 : bdev_io_get_buf(bdev_io, len);
1821 : }
1822 :
1823 : static void
1824 4 : _bdev_memory_domain_get_io_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io,
1825 : bool success)
1826 : {
1827 4 : if (!success) {
1828 0 : SPDK_ERRLOG("Failed to get data buffer, completing IO\n");
1829 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
1830 0 : bdev_io_complete_unsubmitted(bdev_io);
1831 0 : return;
1832 : }
1833 :
1834 4 : if (bdev_io_needs_sequence_exec(bdev_io->internal.desc, bdev_io)) {
1835 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
1836 0 : bdev_io_exec_sequence(bdev_io, bdev_io_submit_sequence_cb);
1837 0 : return;
1838 : }
1839 : /* For reads we'll execute the sequence after the data is read, so, for now, only
1840 : * clear out accel_sequence pointer and submit the IO */
1841 0 : assert(bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
1842 0 : bdev_io->u.bdev.accel_sequence = NULL;
1843 : }
1844 :
1845 4 : bdev_io_submit(bdev_io);
1846 : }
1847 :
1848 : static void
1849 4 : _bdev_memory_domain_io_get_buf(struct spdk_bdev_io *bdev_io, spdk_bdev_io_get_buf_cb cb,
1850 : uint64_t len)
1851 : {
1852 4 : assert(cb != NULL);
1853 4 : bdev_io->internal.get_buf_cb = cb;
1854 :
1855 4 : bdev_io_get_buf(bdev_io, len);
1856 4 : }
1857 :
1858 : void
1859 0 : spdk_bdev_io_get_aux_buf(struct spdk_bdev_io *bdev_io, spdk_bdev_io_get_aux_buf_cb cb)
1860 : {
1861 0 : uint64_t len = bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen;
1862 :
1863 0 : assert(cb != NULL);
1864 0 : assert(bdev_io->internal.get_aux_buf_cb == NULL);
1865 0 : bdev_io->internal.get_aux_buf_cb = cb;
1866 0 : bdev_io_get_buf(bdev_io, len);
1867 0 : }
1868 :
1869 : static int
1870 64 : bdev_module_get_max_ctx_size(void)
1871 : {
1872 : struct spdk_bdev_module *bdev_module;
1873 64 : int max_bdev_module_size = 0;
1874 :
1875 249 : TAILQ_FOREACH(bdev_module, &g_bdev_mgr.bdev_modules, internal.tailq) {
1876 185 : if (bdev_module->get_ctx_size && bdev_module->get_ctx_size() > max_bdev_module_size) {
1877 63 : max_bdev_module_size = bdev_module->get_ctx_size();
1878 : }
1879 : }
1880 :
1881 64 : return max_bdev_module_size;
1882 : }
1883 :
1884 : static void
1885 0 : bdev_enable_histogram_config_json(struct spdk_bdev *bdev, struct spdk_json_write_ctx *w)
1886 : {
1887 0 : if (!bdev->internal.histogram_enabled) {
1888 0 : return;
1889 : }
1890 :
1891 0 : spdk_json_write_object_begin(w);
1892 0 : spdk_json_write_named_string(w, "method", "bdev_enable_histogram");
1893 :
1894 0 : spdk_json_write_named_object_begin(w, "params");
1895 0 : spdk_json_write_named_string(w, "name", bdev->name);
1896 :
1897 0 : spdk_json_write_named_bool(w, "enable", bdev->internal.histogram_enabled);
1898 :
1899 0 : if (bdev->internal.histogram_io_type) {
1900 0 : spdk_json_write_named_string(w, "opc",
1901 0 : spdk_bdev_get_io_type_name(bdev->internal.histogram_io_type));
1902 : }
1903 :
1904 0 : spdk_json_write_object_end(w);
1905 :
1906 0 : spdk_json_write_object_end(w);
1907 : }
1908 :
1909 : static void
1910 0 : bdev_qos_config_json(struct spdk_bdev *bdev, struct spdk_json_write_ctx *w)
1911 : {
1912 : int i;
1913 0 : struct spdk_bdev_qos *qos = bdev->internal.qos;
1914 0 : uint64_t limits[SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES];
1915 :
1916 0 : if (!qos) {
1917 0 : return;
1918 : }
1919 :
1920 0 : spdk_bdev_get_qos_rate_limits(bdev, limits);
1921 :
1922 0 : spdk_json_write_object_begin(w);
1923 0 : spdk_json_write_named_string(w, "method", "bdev_set_qos_limit");
1924 :
1925 0 : spdk_json_write_named_object_begin(w, "params");
1926 0 : spdk_json_write_named_string(w, "name", bdev->name);
1927 0 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
1928 0 : if (limits[i] > 0) {
1929 0 : spdk_json_write_named_uint64(w, qos_rpc_type[i], limits[i]);
1930 : }
1931 : }
1932 0 : spdk_json_write_object_end(w);
1933 :
1934 0 : spdk_json_write_object_end(w);
1935 : }
1936 :
1937 : void
1938 0 : spdk_bdev_subsystem_config_json(struct spdk_json_write_ctx *w)
1939 : {
1940 : struct spdk_bdev_module *bdev_module;
1941 : struct spdk_bdev *bdev;
1942 :
1943 0 : assert(w != NULL);
1944 :
1945 0 : spdk_json_write_array_begin(w);
1946 :
1947 0 : spdk_json_write_object_begin(w);
1948 0 : spdk_json_write_named_string(w, "method", "bdev_set_options");
1949 0 : spdk_json_write_named_object_begin(w, "params");
1950 0 : spdk_json_write_named_uint32(w, "bdev_io_pool_size", g_bdev_opts.bdev_io_pool_size);
1951 0 : spdk_json_write_named_uint32(w, "bdev_io_cache_size", g_bdev_opts.bdev_io_cache_size);
1952 0 : spdk_json_write_named_bool(w, "bdev_auto_examine", g_bdev_opts.bdev_auto_examine);
1953 0 : spdk_json_write_named_uint32(w, "iobuf_small_cache_size", g_bdev_opts.iobuf_small_cache_size);
1954 0 : spdk_json_write_named_uint32(w, "iobuf_large_cache_size", g_bdev_opts.iobuf_large_cache_size);
1955 0 : spdk_json_write_object_end(w);
1956 0 : spdk_json_write_object_end(w);
1957 :
1958 0 : bdev_examine_allowlist_config_json(w);
1959 :
1960 0 : TAILQ_FOREACH(bdev_module, &g_bdev_mgr.bdev_modules, internal.tailq) {
1961 0 : if (bdev_module->config_json) {
1962 0 : bdev_module->config_json(w);
1963 : }
1964 : }
1965 :
1966 0 : spdk_spin_lock(&g_bdev_mgr.spinlock);
1967 :
1968 0 : TAILQ_FOREACH(bdev, &g_bdev_mgr.bdevs, internal.link) {
1969 0 : if (bdev->fn_table->write_config_json) {
1970 0 : bdev->fn_table->write_config_json(bdev, w);
1971 : }
1972 :
1973 0 : bdev_qos_config_json(bdev, w);
1974 0 : bdev_enable_histogram_config_json(bdev, w);
1975 : }
1976 :
1977 0 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
1978 :
1979 : /* This has to be last RPC in array to make sure all bdevs finished examine */
1980 0 : spdk_json_write_object_begin(w);
1981 0 : spdk_json_write_named_string(w, "method", "bdev_wait_for_examine");
1982 0 : spdk_json_write_object_end(w);
1983 :
1984 0 : spdk_json_write_array_end(w);
1985 0 : }
1986 :
1987 : static void
1988 70 : bdev_mgmt_channel_destroy(void *io_device, void *ctx_buf)
1989 : {
1990 70 : struct spdk_bdev_mgmt_channel *ch = ctx_buf;
1991 : struct spdk_bdev_io *bdev_io;
1992 :
1993 70 : spdk_iobuf_channel_fini(&ch->iobuf);
1994 :
1995 9967 : while (!STAILQ_EMPTY(&ch->per_thread_cache)) {
1996 9897 : bdev_io = STAILQ_FIRST(&ch->per_thread_cache);
1997 9897 : STAILQ_REMOVE_HEAD(&ch->per_thread_cache, internal.buf_link);
1998 9897 : ch->per_thread_cache_count--;
1999 9897 : spdk_mempool_put(g_bdev_mgr.bdev_io_pool, (void *)bdev_io);
2000 : }
2001 :
2002 70 : assert(ch->per_thread_cache_count == 0);
2003 70 : }
2004 :
2005 : static int
2006 70 : bdev_mgmt_channel_create(void *io_device, void *ctx_buf)
2007 : {
2008 70 : struct spdk_bdev_mgmt_channel *ch = ctx_buf;
2009 : struct spdk_bdev_io *bdev_io;
2010 : uint32_t i;
2011 : int rc;
2012 :
2013 70 : rc = spdk_iobuf_channel_init(&ch->iobuf, "bdev",
2014 : g_bdev_opts.iobuf_small_cache_size,
2015 : g_bdev_opts.iobuf_large_cache_size);
2016 70 : if (rc != 0) {
2017 0 : SPDK_ERRLOG("Failed to create iobuf channel: %s\n", spdk_strerror(-rc));
2018 0 : return -1;
2019 : }
2020 :
2021 70 : STAILQ_INIT(&ch->per_thread_cache);
2022 70 : ch->bdev_io_cache_size = g_bdev_opts.bdev_io_cache_size;
2023 :
2024 : /* Pre-populate bdev_io cache to ensure this thread cannot be starved. */
2025 70 : ch->per_thread_cache_count = 0;
2026 9967 : for (i = 0; i < ch->bdev_io_cache_size; i++) {
2027 9897 : bdev_io = spdk_mempool_get(g_bdev_mgr.bdev_io_pool);
2028 9897 : if (bdev_io == NULL) {
2029 0 : SPDK_ERRLOG("You need to increase bdev_io_pool_size using bdev_set_options RPC.\n");
2030 0 : assert(false);
2031 : bdev_mgmt_channel_destroy(io_device, ctx_buf);
2032 : return -1;
2033 : }
2034 9897 : ch->per_thread_cache_count++;
2035 9897 : STAILQ_INSERT_HEAD(&ch->per_thread_cache, bdev_io, internal.buf_link);
2036 : }
2037 :
2038 70 : TAILQ_INIT(&ch->shared_resources);
2039 70 : TAILQ_INIT(&ch->io_wait_queue);
2040 :
2041 70 : return 0;
2042 : }
2043 :
2044 : static void
2045 64 : bdev_init_complete(int rc)
2046 : {
2047 64 : spdk_bdev_init_cb cb_fn = g_init_cb_fn;
2048 64 : void *cb_arg = g_init_cb_arg;
2049 : struct spdk_bdev_module *m;
2050 :
2051 64 : g_bdev_mgr.init_complete = true;
2052 64 : g_init_cb_fn = NULL;
2053 64 : g_init_cb_arg = NULL;
2054 :
2055 : /*
2056 : * For modules that need to know when subsystem init is complete,
2057 : * inform them now.
2058 : */
2059 64 : if (rc == 0) {
2060 249 : TAILQ_FOREACH(m, &g_bdev_mgr.bdev_modules, internal.tailq) {
2061 185 : if (m->init_complete) {
2062 23 : m->init_complete();
2063 : }
2064 : }
2065 : }
2066 :
2067 64 : cb_fn(cb_arg, rc);
2068 64 : }
2069 :
2070 : static bool
2071 253 : bdev_module_all_actions_completed(void)
2072 : {
2073 : struct spdk_bdev_module *m;
2074 :
2075 997 : TAILQ_FOREACH(m, &g_bdev_mgr.bdev_modules, internal.tailq) {
2076 744 : if (m->internal.action_in_progress > 0) {
2077 0 : return false;
2078 : }
2079 : }
2080 253 : return true;
2081 : }
2082 :
2083 : static void
2084 608 : bdev_module_action_complete(void)
2085 : {
2086 : /*
2087 : * Don't finish bdev subsystem initialization if
2088 : * module pre-initialization is still in progress, or
2089 : * the subsystem been already initialized.
2090 : */
2091 608 : if (!g_bdev_mgr.module_init_complete || g_bdev_mgr.init_complete) {
2092 544 : return;
2093 : }
2094 :
2095 : /*
2096 : * Check all bdev modules for inits/examinations in progress. If any
2097 : * exist, return immediately since we cannot finish bdev subsystem
2098 : * initialization until all are completed.
2099 : */
2100 64 : if (!bdev_module_all_actions_completed()) {
2101 0 : return;
2102 : }
2103 :
2104 : /*
2105 : * Modules already finished initialization - now that all
2106 : * the bdev modules have finished their asynchronous I/O
2107 : * processing, the entire bdev layer can be marked as complete.
2108 : */
2109 64 : bdev_init_complete(0);
2110 : }
2111 :
2112 : static void
2113 544 : bdev_module_action_done(struct spdk_bdev_module *module)
2114 : {
2115 544 : spdk_spin_lock(&module->internal.spinlock);
2116 544 : assert(module->internal.action_in_progress > 0);
2117 544 : module->internal.action_in_progress--;
2118 544 : spdk_spin_unlock(&module->internal.spinlock);
2119 544 : bdev_module_action_complete();
2120 544 : }
2121 :
2122 : void
2123 64 : spdk_bdev_module_init_done(struct spdk_bdev_module *module)
2124 : {
2125 64 : assert(module->async_init);
2126 64 : bdev_module_action_done(module);
2127 64 : }
2128 :
2129 : void
2130 480 : spdk_bdev_module_examine_done(struct spdk_bdev_module *module)
2131 : {
2132 480 : bdev_module_action_done(module);
2133 480 : }
2134 :
2135 : /** The last initialized bdev module */
2136 : static struct spdk_bdev_module *g_resume_bdev_module = NULL;
2137 :
2138 : static void
2139 0 : bdev_init_failed(void *cb_arg)
2140 : {
2141 0 : struct spdk_bdev_module *module = cb_arg;
2142 :
2143 0 : spdk_spin_lock(&module->internal.spinlock);
2144 0 : assert(module->internal.action_in_progress > 0);
2145 0 : module->internal.action_in_progress--;
2146 0 : spdk_spin_unlock(&module->internal.spinlock);
2147 0 : bdev_init_complete(-1);
2148 0 : }
2149 :
2150 : static int
2151 64 : bdev_modules_init(void)
2152 : {
2153 : struct spdk_bdev_module *module;
2154 64 : int rc = 0;
2155 :
2156 249 : TAILQ_FOREACH(module, &g_bdev_mgr.bdev_modules, internal.tailq) {
2157 185 : g_resume_bdev_module = module;
2158 185 : if (module->async_init) {
2159 64 : spdk_spin_lock(&module->internal.spinlock);
2160 64 : module->internal.action_in_progress = 1;
2161 64 : spdk_spin_unlock(&module->internal.spinlock);
2162 : }
2163 185 : rc = module->module_init();
2164 185 : if (rc != 0) {
2165 : /* Bump action_in_progress to prevent other modules from completion of modules_init
2166 : * Send message to defer application shutdown until resources are cleaned up */
2167 0 : spdk_spin_lock(&module->internal.spinlock);
2168 0 : module->internal.action_in_progress = 1;
2169 0 : spdk_spin_unlock(&module->internal.spinlock);
2170 0 : spdk_thread_send_msg(spdk_get_thread(), bdev_init_failed, module);
2171 0 : return rc;
2172 : }
2173 : }
2174 :
2175 64 : g_resume_bdev_module = NULL;
2176 64 : return 0;
2177 : }
2178 :
2179 : void
2180 64 : spdk_bdev_initialize(spdk_bdev_init_cb cb_fn, void *cb_arg)
2181 : {
2182 64 : int rc = 0;
2183 64 : char mempool_name[32];
2184 :
2185 64 : assert(cb_fn != NULL);
2186 :
2187 64 : g_init_cb_fn = cb_fn;
2188 64 : g_init_cb_arg = cb_arg;
2189 :
2190 64 : spdk_notify_type_register("bdev_register");
2191 64 : spdk_notify_type_register("bdev_unregister");
2192 :
2193 64 : snprintf(mempool_name, sizeof(mempool_name), "bdev_io_%d", getpid());
2194 :
2195 64 : rc = spdk_iobuf_register_module("bdev");
2196 64 : if (rc != 0) {
2197 0 : SPDK_ERRLOG("could not register bdev iobuf module: %s\n", spdk_strerror(-rc));
2198 0 : bdev_init_complete(-1);
2199 0 : return;
2200 : }
2201 :
2202 128 : g_bdev_mgr.bdev_io_pool = spdk_mempool_create(mempool_name,
2203 64 : g_bdev_opts.bdev_io_pool_size,
2204 : sizeof(struct spdk_bdev_io) +
2205 64 : bdev_module_get_max_ctx_size(),
2206 : 0,
2207 : SPDK_ENV_SOCKET_ID_ANY);
2208 :
2209 64 : if (g_bdev_mgr.bdev_io_pool == NULL) {
2210 0 : SPDK_ERRLOG("could not allocate spdk_bdev_io pool\n");
2211 0 : bdev_init_complete(-1);
2212 0 : return;
2213 : }
2214 :
2215 64 : g_bdev_mgr.zero_buffer = spdk_zmalloc(ZERO_BUFFER_SIZE, ZERO_BUFFER_SIZE,
2216 : NULL, SPDK_ENV_LCORE_ID_ANY, SPDK_MALLOC_DMA);
2217 64 : if (!g_bdev_mgr.zero_buffer) {
2218 0 : SPDK_ERRLOG("create bdev zero buffer failed\n");
2219 0 : bdev_init_complete(-1);
2220 0 : return;
2221 : }
2222 :
2223 : #ifdef SPDK_CONFIG_VTUNE
2224 : g_bdev_mgr.domain = __itt_domain_create("spdk_bdev");
2225 : #endif
2226 :
2227 64 : spdk_io_device_register(&g_bdev_mgr, bdev_mgmt_channel_create,
2228 : bdev_mgmt_channel_destroy,
2229 : sizeof(struct spdk_bdev_mgmt_channel),
2230 : "bdev_mgr");
2231 :
2232 64 : rc = bdev_modules_init();
2233 64 : g_bdev_mgr.module_init_complete = true;
2234 64 : if (rc != 0) {
2235 0 : SPDK_ERRLOG("bdev modules init failed\n");
2236 0 : return;
2237 : }
2238 :
2239 64 : bdev_module_action_complete();
2240 : }
2241 :
2242 : static void
2243 64 : bdev_mgr_unregister_cb(void *io_device)
2244 : {
2245 64 : spdk_bdev_fini_cb cb_fn = g_fini_cb_fn;
2246 :
2247 64 : if (g_bdev_mgr.bdev_io_pool) {
2248 64 : if (spdk_mempool_count(g_bdev_mgr.bdev_io_pool) != g_bdev_opts.bdev_io_pool_size) {
2249 0 : SPDK_ERRLOG("bdev IO pool count is %zu but should be %u\n",
2250 : spdk_mempool_count(g_bdev_mgr.bdev_io_pool),
2251 : g_bdev_opts.bdev_io_pool_size);
2252 : }
2253 :
2254 64 : spdk_mempool_free(g_bdev_mgr.bdev_io_pool);
2255 : }
2256 :
2257 64 : spdk_free(g_bdev_mgr.zero_buffer);
2258 :
2259 64 : bdev_examine_allowlist_free();
2260 :
2261 64 : cb_fn(g_fini_cb_arg);
2262 64 : g_fini_cb_fn = NULL;
2263 64 : g_fini_cb_arg = NULL;
2264 64 : g_bdev_mgr.init_complete = false;
2265 64 : g_bdev_mgr.module_init_complete = false;
2266 64 : }
2267 :
2268 : static void
2269 64 : bdev_module_fini_iter(void *arg)
2270 : {
2271 : struct spdk_bdev_module *bdev_module;
2272 :
2273 : /* FIXME: Handling initialization failures is broken now,
2274 : * so we won't even try cleaning up after successfully
2275 : * initialized modules. if module_init_complete is false,
2276 : * just call spdk_bdev_mgr_unregister_cb
2277 : */
2278 64 : if (!g_bdev_mgr.module_init_complete) {
2279 0 : bdev_mgr_unregister_cb(NULL);
2280 0 : return;
2281 : }
2282 :
2283 : /* Start iterating from the last touched module */
2284 64 : if (!g_resume_bdev_module) {
2285 64 : bdev_module = TAILQ_LAST(&g_bdev_mgr.bdev_modules, bdev_module_list);
2286 : } else {
2287 0 : bdev_module = TAILQ_PREV(g_resume_bdev_module, bdev_module_list,
2288 : internal.tailq);
2289 : }
2290 :
2291 249 : while (bdev_module) {
2292 185 : if (bdev_module->async_fini) {
2293 : /* Save our place so we can resume later. We must
2294 : * save the variable here, before calling module_fini()
2295 : * below, because in some cases the module may immediately
2296 : * call spdk_bdev_module_fini_done() and re-enter
2297 : * this function to continue iterating. */
2298 0 : g_resume_bdev_module = bdev_module;
2299 : }
2300 :
2301 185 : if (bdev_module->module_fini) {
2302 185 : bdev_module->module_fini();
2303 : }
2304 :
2305 185 : if (bdev_module->async_fini) {
2306 0 : return;
2307 : }
2308 :
2309 185 : bdev_module = TAILQ_PREV(bdev_module, bdev_module_list,
2310 : internal.tailq);
2311 : }
2312 :
2313 64 : g_resume_bdev_module = NULL;
2314 64 : spdk_io_device_unregister(&g_bdev_mgr, bdev_mgr_unregister_cb);
2315 : }
2316 :
2317 : void
2318 0 : spdk_bdev_module_fini_done(void)
2319 : {
2320 0 : if (spdk_get_thread() != g_fini_thread) {
2321 0 : spdk_thread_send_msg(g_fini_thread, bdev_module_fini_iter, NULL);
2322 : } else {
2323 0 : bdev_module_fini_iter(NULL);
2324 : }
2325 0 : }
2326 :
2327 : static void
2328 64 : bdev_finish_unregister_bdevs_iter(void *cb_arg, int bdeverrno)
2329 : {
2330 64 : struct spdk_bdev *bdev = cb_arg;
2331 :
2332 64 : if (bdeverrno && bdev) {
2333 0 : SPDK_WARNLOG("Unable to unregister bdev '%s' during spdk_bdev_finish()\n",
2334 : bdev->name);
2335 :
2336 : /*
2337 : * Since the call to spdk_bdev_unregister() failed, we have no way to free this
2338 : * bdev; try to continue by manually removing this bdev from the list and continue
2339 : * with the next bdev in the list.
2340 : */
2341 0 : TAILQ_REMOVE(&g_bdev_mgr.bdevs, bdev, internal.link);
2342 : }
2343 :
2344 64 : if (TAILQ_EMPTY(&g_bdev_mgr.bdevs)) {
2345 64 : SPDK_DEBUGLOG(bdev, "Done unregistering bdevs\n");
2346 : /*
2347 : * Bdev module finish need to be deferred as we might be in the middle of some context
2348 : * (like bdev part free) that will use this bdev (or private bdev driver ctx data)
2349 : * after returning.
2350 : */
2351 64 : spdk_thread_send_msg(spdk_get_thread(), bdev_module_fini_iter, NULL);
2352 64 : return;
2353 : }
2354 :
2355 : /*
2356 : * Unregister last unclaimed bdev in the list, to ensure that bdev subsystem
2357 : * shutdown proceeds top-down. The goal is to give virtual bdevs an opportunity
2358 : * to detect clean shutdown as opposed to run-time hot removal of the underlying
2359 : * base bdevs.
2360 : *
2361 : * Also, walk the list in the reverse order.
2362 : */
2363 0 : for (bdev = TAILQ_LAST(&g_bdev_mgr.bdevs, spdk_bdev_list);
2364 0 : bdev; bdev = TAILQ_PREV(bdev, spdk_bdev_list, internal.link)) {
2365 0 : spdk_spin_lock(&bdev->internal.spinlock);
2366 0 : if (bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE) {
2367 0 : LOG_ALREADY_CLAIMED_DEBUG("claimed, skipping", bdev);
2368 0 : spdk_spin_unlock(&bdev->internal.spinlock);
2369 0 : continue;
2370 : }
2371 0 : spdk_spin_unlock(&bdev->internal.spinlock);
2372 :
2373 0 : SPDK_DEBUGLOG(bdev, "Unregistering bdev '%s'\n", bdev->name);
2374 0 : spdk_bdev_unregister(bdev, bdev_finish_unregister_bdevs_iter, bdev);
2375 0 : return;
2376 : }
2377 :
2378 : /*
2379 : * If any bdev fails to unclaim underlying bdev properly, we may face the
2380 : * case of bdev list consisting of claimed bdevs only (if claims are managed
2381 : * correctly, this would mean there's a loop in the claims graph which is
2382 : * clearly impossible). Warn and unregister last bdev on the list then.
2383 : */
2384 0 : for (bdev = TAILQ_LAST(&g_bdev_mgr.bdevs, spdk_bdev_list);
2385 0 : bdev; bdev = TAILQ_PREV(bdev, spdk_bdev_list, internal.link)) {
2386 0 : SPDK_WARNLOG("Unregistering claimed bdev '%s'!\n", bdev->name);
2387 0 : spdk_bdev_unregister(bdev, bdev_finish_unregister_bdevs_iter, bdev);
2388 0 : return;
2389 : }
2390 : }
2391 :
2392 : static void
2393 64 : bdev_module_fini_start_iter(void *arg)
2394 : {
2395 : struct spdk_bdev_module *bdev_module;
2396 :
2397 64 : if (!g_resume_bdev_module) {
2398 64 : bdev_module = TAILQ_LAST(&g_bdev_mgr.bdev_modules, bdev_module_list);
2399 : } else {
2400 0 : bdev_module = TAILQ_PREV(g_resume_bdev_module, bdev_module_list, internal.tailq);
2401 : }
2402 :
2403 249 : while (bdev_module) {
2404 185 : if (bdev_module->async_fini_start) {
2405 : /* Save our place so we can resume later. We must
2406 : * save the variable here, before calling fini_start()
2407 : * below, because in some cases the module may immediately
2408 : * call spdk_bdev_module_fini_start_done() and re-enter
2409 : * this function to continue iterating. */
2410 0 : g_resume_bdev_module = bdev_module;
2411 : }
2412 :
2413 185 : if (bdev_module->fini_start) {
2414 23 : bdev_module->fini_start();
2415 : }
2416 :
2417 185 : if (bdev_module->async_fini_start) {
2418 0 : return;
2419 : }
2420 :
2421 185 : bdev_module = TAILQ_PREV(bdev_module, bdev_module_list, internal.tailq);
2422 : }
2423 :
2424 64 : g_resume_bdev_module = NULL;
2425 :
2426 64 : bdev_finish_unregister_bdevs_iter(NULL, 0);
2427 : }
2428 :
2429 : void
2430 0 : spdk_bdev_module_fini_start_done(void)
2431 : {
2432 0 : if (spdk_get_thread() != g_fini_thread) {
2433 0 : spdk_thread_send_msg(g_fini_thread, bdev_module_fini_start_iter, NULL);
2434 : } else {
2435 0 : bdev_module_fini_start_iter(NULL);
2436 : }
2437 0 : }
2438 :
2439 : static void
2440 64 : bdev_finish_wait_for_examine_done(void *cb_arg)
2441 : {
2442 64 : bdev_module_fini_start_iter(NULL);
2443 64 : }
2444 :
2445 : static void bdev_open_async_fini(void);
2446 :
2447 : void
2448 64 : spdk_bdev_finish(spdk_bdev_fini_cb cb_fn, void *cb_arg)
2449 : {
2450 : int rc;
2451 :
2452 64 : assert(cb_fn != NULL);
2453 :
2454 64 : g_fini_thread = spdk_get_thread();
2455 :
2456 64 : g_fini_cb_fn = cb_fn;
2457 64 : g_fini_cb_arg = cb_arg;
2458 :
2459 64 : bdev_open_async_fini();
2460 :
2461 64 : rc = spdk_bdev_wait_for_examine(bdev_finish_wait_for_examine_done, NULL);
2462 64 : if (rc != 0) {
2463 0 : SPDK_ERRLOG("wait_for_examine failed: %s\n", spdk_strerror(-rc));
2464 0 : bdev_finish_wait_for_examine_done(NULL);
2465 : }
2466 64 : }
2467 :
2468 : struct spdk_bdev_io *
2469 684 : bdev_channel_get_io(struct spdk_bdev_channel *channel)
2470 : {
2471 684 : struct spdk_bdev_mgmt_channel *ch = channel->shared_resource->mgmt_ch;
2472 : struct spdk_bdev_io *bdev_io;
2473 :
2474 684 : if (ch->per_thread_cache_count > 0) {
2475 633 : bdev_io = STAILQ_FIRST(&ch->per_thread_cache);
2476 633 : STAILQ_REMOVE_HEAD(&ch->per_thread_cache, internal.buf_link);
2477 633 : ch->per_thread_cache_count--;
2478 51 : } else if (spdk_unlikely(!TAILQ_EMPTY(&ch->io_wait_queue))) {
2479 : /*
2480 : * Don't try to look for bdev_ios in the global pool if there are
2481 : * waiters on bdev_ios - we don't want this caller to jump the line.
2482 : */
2483 0 : bdev_io = NULL;
2484 : } else {
2485 51 : bdev_io = spdk_mempool_get(g_bdev_mgr.bdev_io_pool);
2486 : }
2487 :
2488 684 : return bdev_io;
2489 : }
2490 :
2491 : void
2492 678 : spdk_bdev_free_io(struct spdk_bdev_io *bdev_io)
2493 : {
2494 : struct spdk_bdev_mgmt_channel *ch;
2495 :
2496 678 : assert(bdev_io != NULL);
2497 678 : assert(bdev_io->internal.status != SPDK_BDEV_IO_STATUS_PENDING);
2498 :
2499 678 : ch = bdev_io->internal.ch->shared_resource->mgmt_ch;
2500 :
2501 678 : if (bdev_io->internal.f.has_buf) {
2502 18 : bdev_io_put_buf(bdev_io);
2503 : }
2504 :
2505 678 : if (ch->per_thread_cache_count < ch->bdev_io_cache_size) {
2506 633 : ch->per_thread_cache_count++;
2507 633 : STAILQ_INSERT_HEAD(&ch->per_thread_cache, bdev_io, internal.buf_link);
2508 637 : while (ch->per_thread_cache_count > 0 && !TAILQ_EMPTY(&ch->io_wait_queue)) {
2509 : struct spdk_bdev_io_wait_entry *entry;
2510 :
2511 4 : entry = TAILQ_FIRST(&ch->io_wait_queue);
2512 4 : TAILQ_REMOVE(&ch->io_wait_queue, entry, link);
2513 4 : entry->cb_fn(entry->cb_arg);
2514 : }
2515 : } else {
2516 : /* We should never have a full cache with entries on the io wait queue. */
2517 45 : assert(TAILQ_EMPTY(&ch->io_wait_queue));
2518 45 : spdk_mempool_put(g_bdev_mgr.bdev_io_pool, (void *)bdev_io);
2519 : }
2520 678 : }
2521 :
2522 : static bool
2523 72 : bdev_qos_is_iops_rate_limit(enum spdk_bdev_qos_rate_limit_type limit)
2524 : {
2525 72 : assert(limit != SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES);
2526 :
2527 72 : switch (limit) {
2528 18 : case SPDK_BDEV_QOS_RW_IOPS_RATE_LIMIT:
2529 18 : return true;
2530 54 : case SPDK_BDEV_QOS_RW_BPS_RATE_LIMIT:
2531 : case SPDK_BDEV_QOS_R_BPS_RATE_LIMIT:
2532 : case SPDK_BDEV_QOS_W_BPS_RATE_LIMIT:
2533 54 : return false;
2534 0 : case SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES:
2535 : default:
2536 0 : return false;
2537 : }
2538 : }
2539 :
2540 : static bool
2541 25 : bdev_qos_io_to_limit(struct spdk_bdev_io *bdev_io)
2542 : {
2543 25 : switch (bdev_io->type) {
2544 23 : case SPDK_BDEV_IO_TYPE_NVME_IO:
2545 : case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
2546 : case SPDK_BDEV_IO_TYPE_READ:
2547 : case SPDK_BDEV_IO_TYPE_WRITE:
2548 23 : return true;
2549 0 : case SPDK_BDEV_IO_TYPE_ZCOPY:
2550 0 : if (bdev_io->u.bdev.zcopy.start) {
2551 0 : return true;
2552 : } else {
2553 0 : return false;
2554 : }
2555 2 : default:
2556 2 : return false;
2557 : }
2558 : }
2559 :
2560 : static bool
2561 33 : bdev_is_read_io(struct spdk_bdev_io *bdev_io)
2562 : {
2563 33 : switch (bdev_io->type) {
2564 0 : case SPDK_BDEV_IO_TYPE_NVME_IO:
2565 : case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
2566 : /* Bit 1 (0x2) set for read operation */
2567 0 : if (bdev_io->u.nvme_passthru.cmd.opc & SPDK_NVME_OPC_READ) {
2568 0 : return true;
2569 : } else {
2570 0 : return false;
2571 : }
2572 30 : case SPDK_BDEV_IO_TYPE_READ:
2573 30 : return true;
2574 0 : case SPDK_BDEV_IO_TYPE_ZCOPY:
2575 : /* Populate to read from disk */
2576 0 : if (bdev_io->u.bdev.zcopy.populate) {
2577 0 : return true;
2578 : } else {
2579 0 : return false;
2580 : }
2581 3 : default:
2582 3 : return false;
2583 : }
2584 : }
2585 :
2586 : static uint64_t
2587 43 : bdev_get_io_size_in_byte(struct spdk_bdev_io *bdev_io)
2588 : {
2589 43 : struct spdk_bdev *bdev = bdev_io->bdev;
2590 :
2591 43 : switch (bdev_io->type) {
2592 0 : case SPDK_BDEV_IO_TYPE_NVME_IO:
2593 : case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
2594 0 : return bdev_io->u.nvme_passthru.nbytes;
2595 43 : case SPDK_BDEV_IO_TYPE_READ:
2596 : case SPDK_BDEV_IO_TYPE_WRITE:
2597 43 : return bdev_io->u.bdev.num_blocks * bdev->blocklen;
2598 0 : case SPDK_BDEV_IO_TYPE_ZCOPY:
2599 : /* Track the data in the start phase only */
2600 0 : if (bdev_io->u.bdev.zcopy.start) {
2601 0 : return bdev_io->u.bdev.num_blocks * bdev->blocklen;
2602 : } else {
2603 0 : return 0;
2604 : }
2605 0 : default:
2606 0 : return 0;
2607 : }
2608 : }
2609 :
2610 : static inline bool
2611 64 : bdev_qos_rw_queue_io(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io, uint64_t delta)
2612 : {
2613 : int64_t remaining_this_timeslice;
2614 :
2615 64 : if (!limit->max_per_timeslice) {
2616 : /* The QoS is disabled */
2617 0 : return false;
2618 : }
2619 :
2620 64 : remaining_this_timeslice = __atomic_sub_fetch(&limit->remaining_this_timeslice, delta,
2621 : __ATOMIC_RELAXED);
2622 64 : if (remaining_this_timeslice + (int64_t)delta > 0) {
2623 : /* There was still a quota for this delta -> the IO shouldn't be queued
2624 : *
2625 : * We allow a slight quota overrun here so an IO bigger than the per-timeslice
2626 : * quota can be allowed once a while. Such overrun then taken into account in
2627 : * the QoS poller, where the next timeslice quota is calculated.
2628 : */
2629 59 : return false;
2630 : }
2631 :
2632 : /* There was no quota for this delta -> the IO should be queued
2633 : * The remaining_this_timeslice must be rewinded so it reflects the real
2634 : * amount of IOs or bytes allowed.
2635 : */
2636 5 : __atomic_add_fetch(
2637 5 : &limit->remaining_this_timeslice, delta, __ATOMIC_RELAXED);
2638 5 : return true;
2639 : }
2640 :
2641 : static inline void
2642 5 : bdev_qos_rw_rewind_io(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io, uint64_t delta)
2643 : {
2644 5 : __atomic_add_fetch(&limit->remaining_this_timeslice, delta, __ATOMIC_RELAXED);
2645 5 : }
2646 :
2647 : static bool
2648 23 : bdev_qos_rw_iops_queue(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2649 : {
2650 23 : return bdev_qos_rw_queue_io(limit, io, 1);
2651 : }
2652 :
2653 : static void
2654 3 : bdev_qos_rw_iops_rewind_quota(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2655 : {
2656 3 : bdev_qos_rw_rewind_io(limit, io, 1);
2657 3 : }
2658 :
2659 : static bool
2660 41 : bdev_qos_rw_bps_queue(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2661 : {
2662 41 : return bdev_qos_rw_queue_io(limit, io, bdev_get_io_size_in_byte(io));
2663 : }
2664 :
2665 : static void
2666 2 : bdev_qos_rw_bps_rewind_quota(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2667 : {
2668 2 : bdev_qos_rw_rewind_io(limit, io, bdev_get_io_size_in_byte(io));
2669 2 : }
2670 :
2671 : static bool
2672 19 : bdev_qos_r_bps_queue(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2673 : {
2674 19 : if (bdev_is_read_io(io) == false) {
2675 1 : return false;
2676 : }
2677 :
2678 18 : return bdev_qos_rw_bps_queue(limit, io);
2679 : }
2680 :
2681 : static void
2682 0 : bdev_qos_r_bps_rewind_quota(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2683 : {
2684 0 : if (bdev_is_read_io(io) != false) {
2685 0 : bdev_qos_rw_rewind_io(limit, io, bdev_get_io_size_in_byte(io));
2686 : }
2687 0 : }
2688 :
2689 : static bool
2690 14 : bdev_qos_w_bps_queue(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2691 : {
2692 14 : if (bdev_is_read_io(io) == true) {
2693 12 : return false;
2694 : }
2695 :
2696 2 : return bdev_qos_rw_bps_queue(limit, io);
2697 : }
2698 :
2699 : static void
2700 0 : bdev_qos_w_bps_rewind_quota(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2701 : {
2702 0 : if (bdev_is_read_io(io) != true) {
2703 0 : bdev_qos_rw_rewind_io(limit, io, bdev_get_io_size_in_byte(io));
2704 : }
2705 0 : }
2706 :
2707 : static void
2708 10 : bdev_qos_set_ops(struct spdk_bdev_qos *qos)
2709 : {
2710 : int i;
2711 :
2712 50 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
2713 40 : if (qos->rate_limits[i].limit == SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
2714 15 : qos->rate_limits[i].queue_io = NULL;
2715 15 : continue;
2716 : }
2717 :
2718 25 : switch (i) {
2719 9 : case SPDK_BDEV_QOS_RW_IOPS_RATE_LIMIT:
2720 9 : qos->rate_limits[i].queue_io = bdev_qos_rw_iops_queue;
2721 9 : qos->rate_limits[i].rewind_quota = bdev_qos_rw_iops_rewind_quota;
2722 9 : break;
2723 7 : case SPDK_BDEV_QOS_RW_BPS_RATE_LIMIT:
2724 7 : qos->rate_limits[i].queue_io = bdev_qos_rw_bps_queue;
2725 7 : qos->rate_limits[i].rewind_quota = bdev_qos_rw_bps_rewind_quota;
2726 7 : break;
2727 5 : case SPDK_BDEV_QOS_R_BPS_RATE_LIMIT:
2728 5 : qos->rate_limits[i].queue_io = bdev_qos_r_bps_queue;
2729 5 : qos->rate_limits[i].rewind_quota = bdev_qos_r_bps_rewind_quota;
2730 5 : break;
2731 4 : case SPDK_BDEV_QOS_W_BPS_RATE_LIMIT:
2732 4 : qos->rate_limits[i].queue_io = bdev_qos_w_bps_queue;
2733 4 : qos->rate_limits[i].rewind_quota = bdev_qos_w_bps_rewind_quota;
2734 4 : break;
2735 0 : default:
2736 0 : break;
2737 : }
2738 : }
2739 10 : }
2740 :
2741 : static void
2742 6 : _bdev_io_complete_in_submit(struct spdk_bdev_channel *bdev_ch,
2743 : struct spdk_bdev_io *bdev_io,
2744 : enum spdk_bdev_io_status status)
2745 : {
2746 6 : bdev_io->internal.in_submit_request = true;
2747 6 : bdev_io_increment_outstanding(bdev_ch, bdev_ch->shared_resource);
2748 6 : spdk_bdev_io_complete(bdev_io, status);
2749 6 : bdev_io->internal.in_submit_request = false;
2750 6 : }
2751 :
2752 : static inline void
2753 567 : bdev_io_do_submit(struct spdk_bdev_channel *bdev_ch, struct spdk_bdev_io *bdev_io)
2754 : {
2755 567 : struct spdk_bdev *bdev = bdev_io->bdev;
2756 567 : struct spdk_io_channel *ch = bdev_ch->channel;
2757 567 : struct spdk_bdev_shared_resource *shared_resource = bdev_ch->shared_resource;
2758 :
2759 567 : if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_ABORT)) {
2760 10 : struct spdk_bdev_mgmt_channel *mgmt_channel = shared_resource->mgmt_ch;
2761 10 : struct spdk_bdev_io *bio_to_abort = bdev_io->u.abort.bio_to_abort;
2762 :
2763 20 : if (bdev_abort_queued_io(&shared_resource->nomem_io, bio_to_abort) ||
2764 10 : bdev_abort_buf_io(mgmt_channel, bio_to_abort)) {
2765 0 : _bdev_io_complete_in_submit(bdev_ch, bdev_io,
2766 : SPDK_BDEV_IO_STATUS_SUCCESS);
2767 0 : return;
2768 : }
2769 : }
2770 :
2771 567 : if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE &&
2772 : bdev_io->bdev->split_on_write_unit &&
2773 : bdev_io->u.bdev.num_blocks < bdev_io->bdev->write_unit_size)) {
2774 4 : SPDK_ERRLOG("IO num_blocks %lu does not match the write_unit_size %u\n",
2775 : bdev_io->u.bdev.num_blocks, bdev_io->bdev->write_unit_size);
2776 4 : _bdev_io_complete_in_submit(bdev_ch, bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
2777 4 : return;
2778 : }
2779 :
2780 563 : if (spdk_likely(TAILQ_EMPTY(&shared_resource->nomem_io))) {
2781 520 : bdev_io_increment_outstanding(bdev_ch, shared_resource);
2782 520 : bdev_io->internal.in_submit_request = true;
2783 520 : bdev_submit_request(bdev, ch, bdev_io);
2784 520 : bdev_io->internal.in_submit_request = false;
2785 : } else {
2786 43 : bdev_queue_nomem_io_tail(shared_resource, bdev_io, BDEV_IO_RETRY_STATE_SUBMIT);
2787 43 : if (shared_resource->nomem_threshold == 0 && shared_resource->io_outstanding == 0) {
2788 : /* Special case when we have nomem IOs and no outstanding IOs which completions
2789 : * could trigger retry of queued IOs */
2790 0 : bdev_shared_ch_retry_io(shared_resource);
2791 : }
2792 : }
2793 : }
2794 :
2795 : static bool
2796 25 : bdev_qos_queue_io(struct spdk_bdev_qos *qos, struct spdk_bdev_io *bdev_io)
2797 : {
2798 : int i;
2799 :
2800 25 : if (bdev_qos_io_to_limit(bdev_io) == true) {
2801 100 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
2802 82 : if (!qos->rate_limits[i].queue_io) {
2803 5 : continue;
2804 : }
2805 :
2806 77 : if (qos->rate_limits[i].queue_io(&qos->rate_limits[i],
2807 : bdev_io) == true) {
2808 10 : for (i -= 1; i >= 0 ; i--) {
2809 5 : if (!qos->rate_limits[i].queue_io) {
2810 0 : continue;
2811 : }
2812 :
2813 5 : qos->rate_limits[i].rewind_quota(&qos->rate_limits[i], bdev_io);
2814 : }
2815 5 : return true;
2816 : }
2817 : }
2818 : }
2819 :
2820 20 : return false;
2821 : }
2822 :
2823 : static int
2824 27 : bdev_qos_io_submit(struct spdk_bdev_channel *ch, struct spdk_bdev_qos *qos)
2825 : {
2826 27 : struct spdk_bdev_io *bdev_io = NULL, *tmp = NULL;
2827 27 : int submitted_ios = 0;
2828 :
2829 52 : TAILQ_FOREACH_SAFE(bdev_io, &ch->qos_queued_io, internal.link, tmp) {
2830 25 : if (!bdev_qos_queue_io(qos, bdev_io)) {
2831 20 : TAILQ_REMOVE(&ch->qos_queued_io, bdev_io, internal.link);
2832 20 : bdev_io_do_submit(ch, bdev_io);
2833 :
2834 20 : submitted_ios++;
2835 : }
2836 : }
2837 :
2838 27 : return submitted_ios;
2839 : }
2840 :
2841 : static void
2842 2 : bdev_queue_io_wait_with_cb(struct spdk_bdev_io *bdev_io, spdk_bdev_io_wait_cb cb_fn)
2843 : {
2844 : int rc;
2845 :
2846 2 : bdev_io->internal.waitq_entry.bdev = bdev_io->bdev;
2847 2 : bdev_io->internal.waitq_entry.cb_fn = cb_fn;
2848 2 : bdev_io->internal.waitq_entry.cb_arg = bdev_io;
2849 2 : rc = spdk_bdev_queue_io_wait(bdev_io->bdev, spdk_io_channel_from_ctx(bdev_io->internal.ch),
2850 : &bdev_io->internal.waitq_entry);
2851 2 : if (rc != 0) {
2852 0 : SPDK_ERRLOG("Queue IO failed, rc=%d\n", rc);
2853 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
2854 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
2855 : }
2856 2 : }
2857 :
2858 : static bool
2859 591 : bdev_rw_should_split(struct spdk_bdev_io *bdev_io)
2860 : {
2861 : uint32_t io_boundary;
2862 591 : struct spdk_bdev *bdev = bdev_io->bdev;
2863 591 : uint32_t max_segment_size = bdev->max_segment_size;
2864 591 : uint32_t max_size = bdev->max_rw_size;
2865 591 : int max_segs = bdev->max_num_segments;
2866 :
2867 591 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE && bdev->split_on_write_unit) {
2868 19 : io_boundary = bdev->write_unit_size;
2869 572 : } else if (bdev->split_on_optimal_io_boundary) {
2870 164 : io_boundary = bdev->optimal_io_boundary;
2871 : } else {
2872 408 : io_boundary = 0;
2873 : }
2874 :
2875 591 : if (spdk_likely(!io_boundary && !max_segs && !max_segment_size && !max_size)) {
2876 232 : return false;
2877 : }
2878 :
2879 359 : if (io_boundary) {
2880 : uint64_t start_stripe, end_stripe;
2881 :
2882 183 : start_stripe = bdev_io->u.bdev.offset_blocks;
2883 183 : end_stripe = start_stripe + bdev_io->u.bdev.num_blocks - 1;
2884 : /* Avoid expensive div operations if possible. These spdk_u32 functions are very cheap. */
2885 183 : if (spdk_likely(spdk_u32_is_pow2(io_boundary))) {
2886 183 : start_stripe >>= spdk_u32log2(io_boundary);
2887 183 : end_stripe >>= spdk_u32log2(io_boundary);
2888 : } else {
2889 0 : start_stripe /= io_boundary;
2890 0 : end_stripe /= io_boundary;
2891 : }
2892 :
2893 183 : if (start_stripe != end_stripe) {
2894 78 : return true;
2895 : }
2896 : }
2897 :
2898 281 : if (max_segs) {
2899 145 : if (bdev_io->u.bdev.iovcnt > max_segs) {
2900 15 : return true;
2901 : }
2902 : }
2903 :
2904 266 : if (max_segment_size) {
2905 463 : for (int i = 0; i < bdev_io->u.bdev.iovcnt; i++) {
2906 341 : if (bdev_io->u.bdev.iovs[i].iov_len > max_segment_size) {
2907 9 : return true;
2908 : }
2909 : }
2910 : }
2911 :
2912 257 : if (max_size) {
2913 47 : if (bdev_io->u.bdev.num_blocks > max_size) {
2914 4 : return true;
2915 : }
2916 : }
2917 :
2918 253 : return false;
2919 : }
2920 :
2921 : static bool
2922 24 : bdev_unmap_should_split(struct spdk_bdev_io *bdev_io)
2923 : {
2924 : uint32_t num_unmap_segments;
2925 :
2926 24 : if (!bdev_io->bdev->max_unmap || !bdev_io->bdev->max_unmap_segments) {
2927 3 : return false;
2928 : }
2929 21 : num_unmap_segments = spdk_divide_round_up(bdev_io->u.bdev.num_blocks, bdev_io->bdev->max_unmap);
2930 21 : if (num_unmap_segments > bdev_io->bdev->max_unmap_segments) {
2931 4 : return true;
2932 : }
2933 :
2934 17 : return false;
2935 : }
2936 :
2937 : static bool
2938 37 : bdev_write_zeroes_should_split(struct spdk_bdev_io *bdev_io)
2939 : {
2940 37 : if (!bdev_io->bdev->max_write_zeroes) {
2941 4 : return false;
2942 : }
2943 :
2944 33 : if (bdev_io->u.bdev.num_blocks > bdev_io->bdev->max_write_zeroes) {
2945 10 : return true;
2946 : }
2947 :
2948 23 : return false;
2949 : }
2950 :
2951 : static bool
2952 30 : bdev_copy_should_split(struct spdk_bdev_io *bdev_io)
2953 : {
2954 30 : if (bdev_io->bdev->max_copy != 0 &&
2955 25 : bdev_io->u.bdev.num_blocks > bdev_io->bdev->max_copy) {
2956 6 : return true;
2957 : }
2958 :
2959 24 : return false;
2960 : }
2961 :
2962 : static bool
2963 750 : bdev_io_should_split(struct spdk_bdev_io *bdev_io)
2964 : {
2965 750 : switch (bdev_io->type) {
2966 591 : case SPDK_BDEV_IO_TYPE_READ:
2967 : case SPDK_BDEV_IO_TYPE_WRITE:
2968 591 : return bdev_rw_should_split(bdev_io);
2969 24 : case SPDK_BDEV_IO_TYPE_UNMAP:
2970 24 : return bdev_unmap_should_split(bdev_io);
2971 37 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
2972 37 : return bdev_write_zeroes_should_split(bdev_io);
2973 30 : case SPDK_BDEV_IO_TYPE_COPY:
2974 30 : return bdev_copy_should_split(bdev_io);
2975 68 : default:
2976 68 : return false;
2977 : }
2978 : }
2979 :
2980 : static uint32_t
2981 249 : _to_next_boundary(uint64_t offset, uint32_t boundary)
2982 : {
2983 249 : return (boundary - (offset % boundary));
2984 : }
2985 :
2986 : static void bdev_io_split_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg);
2987 :
2988 : static void _bdev_rw_split(void *_bdev_io);
2989 :
2990 : static void bdev_unmap_split(struct spdk_bdev_io *bdev_io);
2991 :
2992 : static void
2993 0 : _bdev_unmap_split(void *_bdev_io)
2994 : {
2995 0 : return bdev_unmap_split((struct spdk_bdev_io *)_bdev_io);
2996 : }
2997 :
2998 : static void bdev_write_zeroes_split(struct spdk_bdev_io *bdev_io);
2999 :
3000 : static void
3001 0 : _bdev_write_zeroes_split(void *_bdev_io)
3002 : {
3003 0 : return bdev_write_zeroes_split((struct spdk_bdev_io *)_bdev_io);
3004 : }
3005 :
3006 : static void bdev_copy_split(struct spdk_bdev_io *bdev_io);
3007 :
3008 : static void
3009 0 : _bdev_copy_split(void *_bdev_io)
3010 : {
3011 0 : return bdev_copy_split((struct spdk_bdev_io *)_bdev_io);
3012 : }
3013 :
3014 : static int
3015 305 : bdev_io_split_submit(struct spdk_bdev_io *bdev_io, struct iovec *iov, int iovcnt, void *md_buf,
3016 : uint64_t num_blocks, uint64_t *offset, uint64_t *remaining)
3017 : {
3018 : int rc;
3019 : uint64_t current_offset, current_remaining, current_src_offset;
3020 : spdk_bdev_io_wait_cb io_wait_fn;
3021 :
3022 305 : current_offset = *offset;
3023 305 : current_remaining = *remaining;
3024 :
3025 305 : assert(bdev_io->internal.f.split);
3026 :
3027 305 : bdev_io->internal.split.outstanding++;
3028 :
3029 305 : io_wait_fn = _bdev_rw_split;
3030 305 : switch (bdev_io->type) {
3031 202 : case SPDK_BDEV_IO_TYPE_READ:
3032 202 : assert(bdev_io->u.bdev.accel_sequence == NULL);
3033 606 : rc = bdev_readv_blocks_with_md(bdev_io->internal.desc,
3034 202 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
3035 : iov, iovcnt, md_buf, current_offset,
3036 : num_blocks,
3037 202 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain : NULL,
3038 202 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain_ctx : NULL,
3039 : NULL,
3040 : bdev_io->u.bdev.dif_check_flags,
3041 : bdev_io_split_done, bdev_io);
3042 202 : break;
3043 44 : case SPDK_BDEV_IO_TYPE_WRITE:
3044 44 : assert(bdev_io->u.bdev.accel_sequence == NULL);
3045 132 : rc = bdev_writev_blocks_with_md(bdev_io->internal.desc,
3046 44 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
3047 : iov, iovcnt, md_buf, current_offset,
3048 : num_blocks,
3049 44 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain : NULL,
3050 44 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain_ctx : NULL,
3051 : NULL,
3052 : bdev_io->u.bdev.dif_check_flags,
3053 : bdev_io->u.bdev.nvme_cdw12.raw,
3054 : bdev_io->u.bdev.nvme_cdw13.raw,
3055 : bdev_io_split_done, bdev_io);
3056 44 : break;
3057 17 : case SPDK_BDEV_IO_TYPE_UNMAP:
3058 17 : io_wait_fn = _bdev_unmap_split;
3059 17 : rc = spdk_bdev_unmap_blocks(bdev_io->internal.desc,
3060 17 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
3061 : current_offset, num_blocks,
3062 : bdev_io_split_done, bdev_io);
3063 17 : break;
3064 23 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3065 23 : io_wait_fn = _bdev_write_zeroes_split;
3066 23 : rc = spdk_bdev_write_zeroes_blocks(bdev_io->internal.desc,
3067 23 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
3068 : current_offset, num_blocks,
3069 : bdev_io_split_done, bdev_io);
3070 23 : break;
3071 19 : case SPDK_BDEV_IO_TYPE_COPY:
3072 19 : io_wait_fn = _bdev_copy_split;
3073 19 : current_src_offset = bdev_io->u.bdev.copy.src_offset_blocks +
3074 19 : (current_offset - bdev_io->u.bdev.offset_blocks);
3075 19 : rc = spdk_bdev_copy_blocks(bdev_io->internal.desc,
3076 19 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
3077 : current_offset, current_src_offset, num_blocks,
3078 : bdev_io_split_done, bdev_io);
3079 19 : break;
3080 0 : default:
3081 0 : assert(false);
3082 : rc = -EINVAL;
3083 : break;
3084 : }
3085 :
3086 305 : if (rc == 0) {
3087 301 : current_offset += num_blocks;
3088 301 : current_remaining -= num_blocks;
3089 301 : bdev_io->internal.split.current_offset_blocks = current_offset;
3090 301 : bdev_io->internal.split.remaining_num_blocks = current_remaining;
3091 301 : *offset = current_offset;
3092 301 : *remaining = current_remaining;
3093 : } else {
3094 4 : bdev_io->internal.split.outstanding--;
3095 4 : if (rc == -ENOMEM) {
3096 4 : if (bdev_io->internal.split.outstanding == 0) {
3097 : /* No I/O is outstanding. Hence we should wait here. */
3098 1 : bdev_queue_io_wait_with_cb(bdev_io, io_wait_fn);
3099 : }
3100 : } else {
3101 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
3102 0 : if (bdev_io->internal.split.outstanding == 0) {
3103 0 : bdev_ch_remove_from_io_submitted(bdev_io);
3104 0 : spdk_trace_record(TRACE_BDEV_IO_DONE, bdev_io->internal.ch->trace_id,
3105 : 0, (uintptr_t)bdev_io, bdev_io->internal.caller_ctx,
3106 : bdev_io->internal.ch->queue_depth);
3107 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
3108 : }
3109 : }
3110 : }
3111 :
3112 305 : return rc;
3113 : }
3114 :
3115 : static void
3116 67 : _bdev_rw_split(void *_bdev_io)
3117 : {
3118 : struct iovec *parent_iov, *iov;
3119 67 : struct spdk_bdev_io *bdev_io = _bdev_io;
3120 67 : struct spdk_bdev *bdev = bdev_io->bdev;
3121 67 : uint64_t parent_offset, current_offset, remaining;
3122 : uint32_t parent_iov_offset, parent_iovcnt, parent_iovpos, child_iovcnt;
3123 : uint32_t to_next_boundary, to_next_boundary_bytes, to_last_block_bytes;
3124 : uint32_t iovcnt, iov_len, child_iovsize;
3125 67 : uint32_t blocklen = bdev->blocklen;
3126 : uint32_t io_boundary;
3127 67 : uint32_t max_segment_size = bdev->max_segment_size;
3128 67 : uint32_t max_child_iovcnt = bdev->max_num_segments;
3129 67 : uint32_t max_size = bdev->max_rw_size;
3130 67 : void *md_buf = NULL;
3131 : int rc;
3132 :
3133 67 : max_size = max_size ? max_size : UINT32_MAX;
3134 67 : max_segment_size = max_segment_size ? max_segment_size : UINT32_MAX;
3135 67 : max_child_iovcnt = max_child_iovcnt ? spdk_min(max_child_iovcnt, SPDK_BDEV_IO_NUM_CHILD_IOV) :
3136 : SPDK_BDEV_IO_NUM_CHILD_IOV;
3137 :
3138 67 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE && bdev->split_on_write_unit) {
3139 4 : io_boundary = bdev->write_unit_size;
3140 63 : } else if (bdev->split_on_optimal_io_boundary) {
3141 43 : io_boundary = bdev->optimal_io_boundary;
3142 : } else {
3143 20 : io_boundary = UINT32_MAX;
3144 : }
3145 :
3146 67 : assert(bdev_io->internal.f.split);
3147 :
3148 67 : remaining = bdev_io->internal.split.remaining_num_blocks;
3149 67 : current_offset = bdev_io->internal.split.current_offset_blocks;
3150 67 : parent_offset = bdev_io->u.bdev.offset_blocks;
3151 67 : parent_iov_offset = (current_offset - parent_offset) * blocklen;
3152 67 : parent_iovcnt = bdev_io->u.bdev.iovcnt;
3153 :
3154 420 : for (parent_iovpos = 0; parent_iovpos < parent_iovcnt; parent_iovpos++) {
3155 420 : parent_iov = &bdev_io->u.bdev.iovs[parent_iovpos];
3156 420 : if (parent_iov_offset < parent_iov->iov_len) {
3157 67 : break;
3158 : }
3159 353 : parent_iov_offset -= parent_iov->iov_len;
3160 : }
3161 :
3162 67 : child_iovcnt = 0;
3163 309 : while (remaining > 0 && parent_iovpos < parent_iovcnt &&
3164 : child_iovcnt < SPDK_BDEV_IO_NUM_CHILD_IOV) {
3165 249 : to_next_boundary = _to_next_boundary(current_offset, io_boundary);
3166 249 : to_next_boundary = spdk_min(remaining, to_next_boundary);
3167 249 : to_next_boundary = spdk_min(max_size, to_next_boundary);
3168 249 : to_next_boundary_bytes = to_next_boundary * blocklen;
3169 :
3170 249 : iov = &bdev_io->child_iov[child_iovcnt];
3171 249 : iovcnt = 0;
3172 :
3173 249 : if (bdev_io->u.bdev.md_buf) {
3174 48 : md_buf = (char *)bdev_io->u.bdev.md_buf +
3175 24 : (current_offset - parent_offset) * spdk_bdev_get_md_size(bdev);
3176 : }
3177 :
3178 249 : child_iovsize = spdk_min(SPDK_BDEV_IO_NUM_CHILD_IOV - child_iovcnt, max_child_iovcnt);
3179 974 : while (to_next_boundary_bytes > 0 && parent_iovpos < parent_iovcnt &&
3180 : iovcnt < child_iovsize) {
3181 725 : parent_iov = &bdev_io->u.bdev.iovs[parent_iovpos];
3182 725 : iov_len = parent_iov->iov_len - parent_iov_offset;
3183 :
3184 725 : iov_len = spdk_min(iov_len, max_segment_size);
3185 725 : iov_len = spdk_min(iov_len, to_next_boundary_bytes);
3186 725 : to_next_boundary_bytes -= iov_len;
3187 :
3188 725 : bdev_io->child_iov[child_iovcnt].iov_base = parent_iov->iov_base + parent_iov_offset;
3189 725 : bdev_io->child_iov[child_iovcnt].iov_len = iov_len;
3190 :
3191 725 : if (iov_len < parent_iov->iov_len - parent_iov_offset) {
3192 183 : parent_iov_offset += iov_len;
3193 : } else {
3194 542 : parent_iovpos++;
3195 542 : parent_iov_offset = 0;
3196 : }
3197 725 : child_iovcnt++;
3198 725 : iovcnt++;
3199 : }
3200 :
3201 249 : if (to_next_boundary_bytes > 0) {
3202 : /* We had to stop this child I/O early because we ran out of
3203 : * child_iov space or were limited by max_num_segments.
3204 : * Ensure the iovs to be aligned with block size and
3205 : * then adjust to_next_boundary before starting the
3206 : * child I/O.
3207 : */
3208 110 : assert(child_iovcnt == SPDK_BDEV_IO_NUM_CHILD_IOV ||
3209 : iovcnt == child_iovsize);
3210 110 : to_last_block_bytes = to_next_boundary_bytes % blocklen;
3211 110 : if (to_last_block_bytes != 0) {
3212 24 : uint32_t child_iovpos = child_iovcnt - 1;
3213 : /* don't decrease child_iovcnt when it equals to SPDK_BDEV_IO_NUM_CHILD_IOV
3214 : * so the loop will naturally end
3215 : */
3216 :
3217 24 : to_last_block_bytes = blocklen - to_last_block_bytes;
3218 24 : to_next_boundary_bytes += to_last_block_bytes;
3219 53 : while (to_last_block_bytes > 0 && iovcnt > 0) {
3220 32 : iov_len = spdk_min(to_last_block_bytes,
3221 : bdev_io->child_iov[child_iovpos].iov_len);
3222 32 : bdev_io->child_iov[child_iovpos].iov_len -= iov_len;
3223 32 : if (bdev_io->child_iov[child_iovpos].iov_len == 0) {
3224 15 : child_iovpos--;
3225 15 : if (--iovcnt == 0) {
3226 : /* If the child IO is less than a block size just return.
3227 : * If the first child IO of any split round is less than
3228 : * a block size, an error exit.
3229 : */
3230 3 : if (bdev_io->internal.split.outstanding == 0) {
3231 1 : SPDK_ERRLOG("The first child io was less than a block size\n");
3232 1 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
3233 1 : bdev_ch_remove_from_io_submitted(bdev_io);
3234 1 : spdk_trace_record(TRACE_BDEV_IO_DONE, bdev_io->internal.ch->trace_id,
3235 : 0, (uintptr_t)bdev_io, bdev_io->internal.caller_ctx,
3236 : bdev_io->internal.ch->queue_depth);
3237 1 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
3238 : }
3239 :
3240 3 : return;
3241 : }
3242 : }
3243 :
3244 29 : to_last_block_bytes -= iov_len;
3245 :
3246 29 : if (parent_iov_offset == 0) {
3247 14 : parent_iovpos--;
3248 14 : parent_iov_offset = bdev_io->u.bdev.iovs[parent_iovpos].iov_len;
3249 : }
3250 29 : parent_iov_offset -= iov_len;
3251 : }
3252 :
3253 21 : assert(to_last_block_bytes == 0);
3254 : }
3255 107 : to_next_boundary -= to_next_boundary_bytes / blocklen;
3256 : }
3257 :
3258 246 : rc = bdev_io_split_submit(bdev_io, iov, iovcnt, md_buf, to_next_boundary,
3259 : ¤t_offset, &remaining);
3260 246 : if (spdk_unlikely(rc)) {
3261 4 : return;
3262 : }
3263 : }
3264 : }
3265 :
3266 : static void
3267 3 : bdev_unmap_split(struct spdk_bdev_io *bdev_io)
3268 : {
3269 3 : uint64_t offset, unmap_blocks, remaining, max_unmap_blocks;
3270 3 : uint32_t num_children_reqs = 0;
3271 : int rc;
3272 :
3273 3 : assert(bdev_io->internal.f.split);
3274 :
3275 3 : offset = bdev_io->internal.split.current_offset_blocks;
3276 3 : remaining = bdev_io->internal.split.remaining_num_blocks;
3277 3 : max_unmap_blocks = bdev_io->bdev->max_unmap * bdev_io->bdev->max_unmap_segments;
3278 :
3279 20 : while (remaining && (num_children_reqs < SPDK_BDEV_MAX_CHILDREN_UNMAP_WRITE_ZEROES_REQS)) {
3280 17 : unmap_blocks = spdk_min(remaining, max_unmap_blocks);
3281 :
3282 17 : rc = bdev_io_split_submit(bdev_io, NULL, 0, NULL, unmap_blocks,
3283 : &offset, &remaining);
3284 17 : if (spdk_likely(rc == 0)) {
3285 17 : num_children_reqs++;
3286 : } else {
3287 0 : return;
3288 : }
3289 : }
3290 : }
3291 :
3292 : static void
3293 6 : bdev_write_zeroes_split(struct spdk_bdev_io *bdev_io)
3294 : {
3295 6 : uint64_t offset, write_zeroes_blocks, remaining;
3296 6 : uint32_t num_children_reqs = 0;
3297 : int rc;
3298 :
3299 6 : assert(bdev_io->internal.f.split);
3300 :
3301 6 : offset = bdev_io->internal.split.current_offset_blocks;
3302 6 : remaining = bdev_io->internal.split.remaining_num_blocks;
3303 :
3304 29 : while (remaining && (num_children_reqs < SPDK_BDEV_MAX_CHILDREN_UNMAP_WRITE_ZEROES_REQS)) {
3305 23 : write_zeroes_blocks = spdk_min(remaining, bdev_io->bdev->max_write_zeroes);
3306 :
3307 23 : rc = bdev_io_split_submit(bdev_io, NULL, 0, NULL, write_zeroes_blocks,
3308 : &offset, &remaining);
3309 23 : if (spdk_likely(rc == 0)) {
3310 23 : num_children_reqs++;
3311 : } else {
3312 0 : return;
3313 : }
3314 : }
3315 : }
3316 :
3317 : static void
3318 4 : bdev_copy_split(struct spdk_bdev_io *bdev_io)
3319 : {
3320 4 : uint64_t offset, copy_blocks, remaining;
3321 4 : uint32_t num_children_reqs = 0;
3322 : int rc;
3323 :
3324 4 : assert(bdev_io->internal.f.split);
3325 :
3326 4 : offset = bdev_io->internal.split.current_offset_blocks;
3327 4 : remaining = bdev_io->internal.split.remaining_num_blocks;
3328 :
3329 4 : assert(bdev_io->bdev->max_copy != 0);
3330 23 : while (remaining && (num_children_reqs < SPDK_BDEV_MAX_CHILDREN_COPY_REQS)) {
3331 19 : copy_blocks = spdk_min(remaining, bdev_io->bdev->max_copy);
3332 :
3333 19 : rc = bdev_io_split_submit(bdev_io, NULL, 0, NULL, copy_blocks,
3334 : &offset, &remaining);
3335 19 : if (spdk_likely(rc == 0)) {
3336 19 : num_children_reqs++;
3337 : } else {
3338 0 : return;
3339 : }
3340 : }
3341 : }
3342 :
3343 : static void
3344 58 : parent_bdev_io_complete(void *ctx, int rc)
3345 : {
3346 58 : struct spdk_bdev_io *parent_io = ctx;
3347 :
3348 58 : if (rc) {
3349 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
3350 : }
3351 :
3352 58 : parent_io->internal.cb(parent_io, parent_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS,
3353 : parent_io->internal.caller_ctx);
3354 58 : }
3355 :
3356 : static void
3357 0 : bdev_io_complete_parent_sequence_cb(void *ctx, int status)
3358 : {
3359 0 : struct spdk_bdev_io *bdev_io = ctx;
3360 :
3361 : /* u.bdev.accel_sequence should have already been cleared at this point */
3362 0 : assert(bdev_io->u.bdev.accel_sequence == NULL);
3363 0 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS);
3364 0 : bdev_io->internal.f.has_accel_sequence = false;
3365 :
3366 0 : if (spdk_unlikely(status != 0)) {
3367 0 : SPDK_ERRLOG("Failed to execute accel sequence, status=%d\n", status);
3368 : }
3369 :
3370 0 : parent_bdev_io_complete(bdev_io, status);
3371 0 : }
3372 :
3373 : static void
3374 301 : bdev_io_split_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
3375 : {
3376 301 : struct spdk_bdev_io *parent_io = cb_arg;
3377 :
3378 301 : spdk_bdev_free_io(bdev_io);
3379 :
3380 301 : assert(parent_io->internal.f.split);
3381 :
3382 301 : if (!success) {
3383 15 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
3384 : /* If any child I/O failed, stop further splitting process. */
3385 15 : parent_io->internal.split.current_offset_blocks += parent_io->internal.split.remaining_num_blocks;
3386 15 : parent_io->internal.split.remaining_num_blocks = 0;
3387 : }
3388 301 : parent_io->internal.split.outstanding--;
3389 301 : if (parent_io->internal.split.outstanding != 0) {
3390 223 : return;
3391 : }
3392 :
3393 : /*
3394 : * Parent I/O finishes when all blocks are consumed.
3395 : */
3396 78 : if (parent_io->internal.split.remaining_num_blocks == 0) {
3397 58 : assert(parent_io->internal.cb != bdev_io_split_done);
3398 58 : bdev_ch_remove_from_io_submitted(parent_io);
3399 58 : spdk_trace_record(TRACE_BDEV_IO_DONE, parent_io->internal.ch->trace_id,
3400 : 0, (uintptr_t)parent_io, bdev_io->internal.caller_ctx,
3401 : parent_io->internal.ch->queue_depth);
3402 :
3403 58 : if (spdk_likely(parent_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS)) {
3404 51 : if (bdev_io_needs_sequence_exec(parent_io->internal.desc, parent_io)) {
3405 0 : bdev_io_exec_sequence(parent_io, bdev_io_complete_parent_sequence_cb);
3406 0 : return;
3407 51 : } else if (parent_io->internal.orig_iovcnt != 0 &&
3408 0 : !bdev_io_use_accel_sequence(bdev_io)) {
3409 : /* bdev IO will be completed in the callback */
3410 0 : _bdev_io_push_bounce_data_buffer(parent_io, parent_bdev_io_complete);
3411 0 : return;
3412 : }
3413 : }
3414 :
3415 58 : parent_bdev_io_complete(parent_io, 0);
3416 58 : return;
3417 : }
3418 :
3419 : /*
3420 : * Continue with the splitting process. This function will complete the parent I/O if the
3421 : * splitting is done.
3422 : */
3423 20 : switch (parent_io->type) {
3424 17 : case SPDK_BDEV_IO_TYPE_READ:
3425 : case SPDK_BDEV_IO_TYPE_WRITE:
3426 17 : _bdev_rw_split(parent_io);
3427 17 : break;
3428 1 : case SPDK_BDEV_IO_TYPE_UNMAP:
3429 1 : bdev_unmap_split(parent_io);
3430 1 : break;
3431 1 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3432 1 : bdev_write_zeroes_split(parent_io);
3433 1 : break;
3434 1 : case SPDK_BDEV_IO_TYPE_COPY:
3435 1 : bdev_copy_split(parent_io);
3436 1 : break;
3437 0 : default:
3438 0 : assert(false);
3439 : break;
3440 : }
3441 : }
3442 :
3443 : static void bdev_rw_split_get_buf_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io,
3444 : bool success);
3445 :
3446 : static void
3447 59 : bdev_io_split(struct spdk_bdev_io *bdev_io)
3448 : {
3449 59 : assert(bdev_io_should_split(bdev_io));
3450 59 : assert(bdev_io->internal.f.split);
3451 :
3452 59 : bdev_io->internal.split.current_offset_blocks = bdev_io->u.bdev.offset_blocks;
3453 59 : bdev_io->internal.split.remaining_num_blocks = bdev_io->u.bdev.num_blocks;
3454 59 : bdev_io->internal.split.outstanding = 0;
3455 59 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
3456 :
3457 59 : switch (bdev_io->type) {
3458 49 : case SPDK_BDEV_IO_TYPE_READ:
3459 : case SPDK_BDEV_IO_TYPE_WRITE:
3460 49 : if (_is_buf_allocated(bdev_io->u.bdev.iovs)) {
3461 47 : _bdev_rw_split(bdev_io);
3462 : } else {
3463 2 : assert(bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
3464 2 : spdk_bdev_io_get_buf(bdev_io, bdev_rw_split_get_buf_cb,
3465 2 : bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen);
3466 : }
3467 49 : break;
3468 2 : case SPDK_BDEV_IO_TYPE_UNMAP:
3469 2 : bdev_unmap_split(bdev_io);
3470 2 : break;
3471 5 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3472 5 : bdev_write_zeroes_split(bdev_io);
3473 5 : break;
3474 3 : case SPDK_BDEV_IO_TYPE_COPY:
3475 3 : bdev_copy_split(bdev_io);
3476 3 : break;
3477 0 : default:
3478 0 : assert(false);
3479 : break;
3480 : }
3481 59 : }
3482 :
3483 : static void
3484 2 : bdev_rw_split_get_buf_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io, bool success)
3485 : {
3486 2 : if (!success) {
3487 0 : spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
3488 0 : return;
3489 : }
3490 :
3491 2 : _bdev_rw_split(bdev_io);
3492 : }
3493 :
3494 : /* Explicitly mark this inline, since it's used as a function pointer and otherwise won't
3495 : * be inlined, at least on some compilers.
3496 : */
3497 : static inline void
3498 572 : _bdev_io_submit(void *ctx)
3499 : {
3500 572 : struct spdk_bdev_io *bdev_io = ctx;
3501 572 : struct spdk_bdev *bdev = bdev_io->bdev;
3502 572 : struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
3503 :
3504 572 : if (spdk_likely(bdev_ch->flags == 0)) {
3505 547 : bdev_io_do_submit(bdev_ch, bdev_io);
3506 547 : return;
3507 : }
3508 :
3509 25 : if (bdev_ch->flags & BDEV_CH_RESET_IN_PROGRESS) {
3510 2 : _bdev_io_complete_in_submit(bdev_ch, bdev_io, SPDK_BDEV_IO_STATUS_ABORTED);
3511 23 : } else if (bdev_ch->flags & BDEV_CH_QOS_ENABLED) {
3512 25 : if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_ABORT) &&
3513 2 : bdev_abort_queued_io(&bdev_ch->qos_queued_io, bdev_io->u.abort.bio_to_abort)) {
3514 0 : _bdev_io_complete_in_submit(bdev_ch, bdev_io, SPDK_BDEV_IO_STATUS_SUCCESS);
3515 : } else {
3516 23 : TAILQ_INSERT_TAIL(&bdev_ch->qos_queued_io, bdev_io, internal.link);
3517 23 : bdev_qos_io_submit(bdev_ch, bdev->internal.qos);
3518 : }
3519 : } else {
3520 0 : SPDK_ERRLOG("unknown bdev_ch flag %x found\n", bdev_ch->flags);
3521 0 : _bdev_io_complete_in_submit(bdev_ch, bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
3522 : }
3523 : }
3524 :
3525 : bool bdev_lba_range_overlapped(struct lba_range *range1, struct lba_range *range2);
3526 :
3527 : bool
3528 23 : bdev_lba_range_overlapped(struct lba_range *range1, struct lba_range *range2)
3529 : {
3530 23 : if (range1->length == 0 || range2->length == 0) {
3531 1 : return false;
3532 : }
3533 :
3534 22 : if (range1->offset + range1->length <= range2->offset) {
3535 1 : return false;
3536 : }
3537 :
3538 21 : if (range2->offset + range2->length <= range1->offset) {
3539 3 : return false;
3540 : }
3541 :
3542 18 : return true;
3543 : }
3544 :
3545 : static bool
3546 11 : bdev_io_range_is_locked(struct spdk_bdev_io *bdev_io, struct lba_range *range)
3547 : {
3548 11 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
3549 11 : struct lba_range r;
3550 :
3551 11 : switch (bdev_io->type) {
3552 0 : case SPDK_BDEV_IO_TYPE_NVME_IO:
3553 : case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
3554 : /* Don't try to decode the NVMe command - just assume worst-case and that
3555 : * it overlaps a locked range.
3556 : */
3557 0 : return true;
3558 6 : case SPDK_BDEV_IO_TYPE_READ:
3559 6 : if (!range->quiesce) {
3560 4 : return false;
3561 : }
3562 : /* fallthrough */
3563 : case SPDK_BDEV_IO_TYPE_WRITE:
3564 : case SPDK_BDEV_IO_TYPE_UNMAP:
3565 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3566 : case SPDK_BDEV_IO_TYPE_ZCOPY:
3567 : case SPDK_BDEV_IO_TYPE_COPY:
3568 7 : r.offset = bdev_io->u.bdev.offset_blocks;
3569 7 : r.length = bdev_io->u.bdev.num_blocks;
3570 7 : if (!bdev_lba_range_overlapped(range, &r)) {
3571 : /* This I/O doesn't overlap the specified LBA range. */
3572 0 : return false;
3573 7 : } else if (range->owner_ch == ch && range->locked_ctx == bdev_io->internal.caller_ctx) {
3574 : /* This I/O overlaps, but the I/O is on the same channel that locked this
3575 : * range, and the caller_ctx is the same as the locked_ctx. This means
3576 : * that this I/O is associated with the lock, and is allowed to execute.
3577 : */
3578 2 : return false;
3579 : } else {
3580 5 : return true;
3581 : }
3582 0 : default:
3583 0 : return false;
3584 : }
3585 : }
3586 :
3587 : void
3588 632 : bdev_io_submit(struct spdk_bdev_io *bdev_io)
3589 : {
3590 632 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
3591 :
3592 632 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_PENDING);
3593 :
3594 632 : if (!TAILQ_EMPTY(&ch->locked_ranges)) {
3595 : struct lba_range *range;
3596 :
3597 13 : TAILQ_FOREACH(range, &ch->locked_ranges, tailq) {
3598 8 : if (bdev_io_range_is_locked(bdev_io, range)) {
3599 3 : TAILQ_INSERT_TAIL(&ch->io_locked, bdev_io, internal.ch_link);
3600 3 : return;
3601 : }
3602 : }
3603 : }
3604 :
3605 629 : bdev_ch_add_to_io_submitted(bdev_io);
3606 :
3607 629 : bdev_io->internal.submit_tsc = spdk_get_ticks();
3608 629 : spdk_trace_record_tsc(bdev_io->internal.submit_tsc, TRACE_BDEV_IO_START,
3609 : ch->trace_id, bdev_io->u.bdev.num_blocks,
3610 : (uintptr_t)bdev_io, (uint64_t)bdev_io->type, bdev_io->internal.caller_ctx,
3611 : bdev_io->u.bdev.offset_blocks, ch->queue_depth);
3612 :
3613 629 : if (bdev_io->internal.f.split) {
3614 59 : bdev_io_split(bdev_io);
3615 59 : return;
3616 : }
3617 :
3618 570 : _bdev_io_submit(bdev_io);
3619 : }
3620 :
3621 : static inline void
3622 4 : _bdev_io_ext_use_bounce_buffer(struct spdk_bdev_io *bdev_io)
3623 : {
3624 : /* bdev doesn't support memory domains, thereby buffers in this IO request can't
3625 : * be accessed directly. It is needed to allocate buffers before issuing IO operation.
3626 : * For write operation we need to pull buffers from memory domain before submitting IO.
3627 : * Once read operation completes, we need to use memory_domain push functionality to
3628 : * update data in original memory domain IO buffer
3629 : * This IO request will go through a regular IO flow, so clear memory domains pointers */
3630 4 : assert(bdev_io->internal.f.has_memory_domain);
3631 4 : bdev_io->u.bdev.memory_domain = NULL;
3632 4 : bdev_io->u.bdev.memory_domain_ctx = NULL;
3633 4 : _bdev_memory_domain_io_get_buf(bdev_io, _bdev_memory_domain_get_io_cb,
3634 4 : bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen);
3635 4 : }
3636 :
3637 : static inline void
3638 295 : _bdev_io_submit_ext(struct spdk_bdev_desc *desc, struct spdk_bdev_io *bdev_io)
3639 : {
3640 295 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
3641 295 : bool needs_exec = bdev_io_needs_sequence_exec(desc, bdev_io);
3642 :
3643 295 : if (spdk_unlikely(ch->flags & BDEV_CH_RESET_IN_PROGRESS)) {
3644 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_ABORTED;
3645 0 : bdev_io_complete_unsubmitted(bdev_io);
3646 0 : return;
3647 : }
3648 :
3649 : /* We need to allocate bounce buffer if bdev doesn't support memory domains, or if it does
3650 : * support them, but we need to execute an accel sequence and the data buffer is from accel
3651 : * memory domain (to avoid doing a push/pull from that domain).
3652 : */
3653 295 : if (bdev_io_use_memory_domain(bdev_io)) {
3654 4 : if (!desc->memory_domains_supported ||
3655 0 : (needs_exec && bdev_io->internal.memory_domain == spdk_accel_get_memory_domain())) {
3656 4 : _bdev_io_ext_use_bounce_buffer(bdev_io);
3657 4 : return;
3658 : }
3659 : }
3660 :
3661 291 : if (needs_exec) {
3662 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
3663 0 : bdev_io_exec_sequence(bdev_io, bdev_io_submit_sequence_cb);
3664 0 : return;
3665 : }
3666 : /* For reads we'll execute the sequence after the data is read, so, for now, only
3667 : * clear out accel_sequence pointer and submit the IO */
3668 0 : assert(bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
3669 0 : bdev_io->u.bdev.accel_sequence = NULL;
3670 : }
3671 :
3672 291 : bdev_io_submit(bdev_io);
3673 : }
3674 :
3675 : static void
3676 11 : bdev_io_submit_reset(struct spdk_bdev_io *bdev_io)
3677 : {
3678 11 : struct spdk_bdev *bdev = bdev_io->bdev;
3679 11 : struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
3680 11 : struct spdk_io_channel *ch = bdev_ch->channel;
3681 :
3682 11 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_PENDING);
3683 :
3684 11 : bdev_io->internal.in_submit_request = true;
3685 11 : bdev_submit_request(bdev, ch, bdev_io);
3686 11 : bdev_io->internal.in_submit_request = false;
3687 11 : }
3688 :
3689 : void
3690 678 : bdev_io_init(struct spdk_bdev_io *bdev_io,
3691 : struct spdk_bdev *bdev, void *cb_arg,
3692 : spdk_bdev_io_completion_cb cb)
3693 : {
3694 678 : bdev_io->bdev = bdev;
3695 678 : bdev_io->internal.f.raw = 0;
3696 678 : bdev_io->internal.caller_ctx = cb_arg;
3697 678 : bdev_io->internal.cb = cb;
3698 678 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_PENDING;
3699 678 : bdev_io->internal.in_submit_request = false;
3700 678 : bdev_io->internal.orig_iovs = NULL;
3701 678 : bdev_io->internal.orig_iovcnt = 0;
3702 678 : bdev_io->internal.orig_md_iov.iov_base = NULL;
3703 678 : bdev_io->internal.error.nvme.cdw0 = 0;
3704 678 : bdev_io->num_retries = 0;
3705 678 : bdev_io->internal.get_buf_cb = NULL;
3706 678 : bdev_io->internal.get_aux_buf_cb = NULL;
3707 678 : bdev_io->internal.data_transfer_cpl = NULL;
3708 678 : bdev_io->internal.f.split = bdev_io_should_split(bdev_io);
3709 678 : }
3710 :
3711 : static bool
3712 388 : bdev_io_type_supported(struct spdk_bdev *bdev, enum spdk_bdev_io_type io_type)
3713 : {
3714 388 : return bdev->fn_table->io_type_supported(bdev->ctxt, io_type);
3715 : }
3716 :
3717 : bool
3718 42 : spdk_bdev_io_type_supported(struct spdk_bdev *bdev, enum spdk_bdev_io_type io_type)
3719 : {
3720 : bool supported;
3721 :
3722 42 : supported = bdev_io_type_supported(bdev, io_type);
3723 :
3724 42 : if (!supported) {
3725 7 : switch (io_type) {
3726 0 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3727 : /* The bdev layer will emulate write zeroes as long as write is supported. */
3728 0 : supported = bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE);
3729 0 : break;
3730 7 : default:
3731 7 : break;
3732 : }
3733 : }
3734 :
3735 42 : return supported;
3736 : }
3737 :
3738 : static const char *g_io_type_strings[] = {
3739 : [SPDK_BDEV_IO_TYPE_READ] = "read",
3740 : [SPDK_BDEV_IO_TYPE_WRITE] = "write",
3741 : [SPDK_BDEV_IO_TYPE_UNMAP] = "unmap",
3742 : [SPDK_BDEV_IO_TYPE_FLUSH] = "flush",
3743 : [SPDK_BDEV_IO_TYPE_RESET] = "reset",
3744 : [SPDK_BDEV_IO_TYPE_NVME_ADMIN] = "nvme_admin",
3745 : [SPDK_BDEV_IO_TYPE_NVME_IO] = "nvme_io",
3746 : [SPDK_BDEV_IO_TYPE_NVME_IO_MD] = "nvme_io_md",
3747 : [SPDK_BDEV_IO_TYPE_WRITE_ZEROES] = "write_zeroes",
3748 : [SPDK_BDEV_IO_TYPE_ZCOPY] = "zcopy",
3749 : [SPDK_BDEV_IO_TYPE_GET_ZONE_INFO] = "get_zone_info",
3750 : [SPDK_BDEV_IO_TYPE_ZONE_MANAGEMENT] = "zone_management",
3751 : [SPDK_BDEV_IO_TYPE_ZONE_APPEND] = "zone_append",
3752 : [SPDK_BDEV_IO_TYPE_COMPARE] = "compare",
3753 : [SPDK_BDEV_IO_TYPE_COMPARE_AND_WRITE] = "compare_and_write",
3754 : [SPDK_BDEV_IO_TYPE_ABORT] = "abort",
3755 : [SPDK_BDEV_IO_TYPE_SEEK_HOLE] = "seek_hole",
3756 : [SPDK_BDEV_IO_TYPE_SEEK_DATA] = "seek_data",
3757 : [SPDK_BDEV_IO_TYPE_COPY] = "copy",
3758 : [SPDK_BDEV_IO_TYPE_NVME_IOV_MD] = "nvme_iov_md",
3759 : };
3760 :
3761 : const char *
3762 0 : spdk_bdev_get_io_type_name(enum spdk_bdev_io_type io_type)
3763 : {
3764 0 : if (io_type <= SPDK_BDEV_IO_TYPE_INVALID || io_type >= SPDK_BDEV_NUM_IO_TYPES) {
3765 0 : return NULL;
3766 : }
3767 :
3768 0 : return g_io_type_strings[io_type];
3769 : }
3770 :
3771 : int
3772 0 : spdk_bdev_get_io_type(const char *io_type_string)
3773 : {
3774 : int i;
3775 :
3776 0 : for (i = SPDK_BDEV_IO_TYPE_READ; i < SPDK_BDEV_NUM_IO_TYPES; ++i) {
3777 0 : if (!strcmp(io_type_string, g_io_type_strings[i])) {
3778 0 : return i;
3779 : }
3780 : }
3781 :
3782 0 : return -1;
3783 : }
3784 :
3785 : uint64_t
3786 0 : spdk_bdev_io_get_submit_tsc(struct spdk_bdev_io *bdev_io)
3787 : {
3788 0 : return bdev_io->internal.submit_tsc;
3789 : }
3790 :
3791 : int
3792 0 : spdk_bdev_dump_info_json(struct spdk_bdev *bdev, struct spdk_json_write_ctx *w)
3793 : {
3794 0 : if (bdev->fn_table->dump_info_json) {
3795 0 : return bdev->fn_table->dump_info_json(bdev->ctxt, w);
3796 : }
3797 :
3798 0 : return 0;
3799 : }
3800 :
3801 : static void
3802 10 : bdev_qos_update_max_quota_per_timeslice(struct spdk_bdev_qos *qos)
3803 : {
3804 10 : uint32_t max_per_timeslice = 0;
3805 : int i;
3806 :
3807 50 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
3808 40 : if (qos->rate_limits[i].limit == SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
3809 15 : qos->rate_limits[i].max_per_timeslice = 0;
3810 15 : continue;
3811 : }
3812 :
3813 25 : max_per_timeslice = qos->rate_limits[i].limit *
3814 25 : SPDK_BDEV_QOS_TIMESLICE_IN_USEC / SPDK_SEC_TO_USEC;
3815 :
3816 25 : qos->rate_limits[i].max_per_timeslice = spdk_max(max_per_timeslice,
3817 : qos->rate_limits[i].min_per_timeslice);
3818 :
3819 25 : __atomic_store_n(&qos->rate_limits[i].remaining_this_timeslice,
3820 25 : qos->rate_limits[i].max_per_timeslice, __ATOMIC_RELEASE);
3821 : }
3822 :
3823 10 : bdev_qos_set_ops(qos);
3824 10 : }
3825 :
3826 : static void
3827 4 : bdev_channel_submit_qos_io(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
3828 : struct spdk_io_channel *io_ch, void *ctx)
3829 : {
3830 4 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
3831 : int status;
3832 :
3833 4 : bdev_qos_io_submit(bdev_ch, bdev->internal.qos);
3834 :
3835 : /* if all IOs were sent then continue the iteration, otherwise - stop it */
3836 : /* TODO: channels round robing */
3837 4 : status = TAILQ_EMPTY(&bdev_ch->qos_queued_io) ? 0 : 1;
3838 :
3839 4 : spdk_bdev_for_each_channel_continue(i, status);
3840 4 : }
3841 :
3842 :
3843 : static void
3844 2 : bdev_channel_submit_qos_io_done(struct spdk_bdev *bdev, void *ctx, int status)
3845 : {
3846 :
3847 2 : }
3848 :
3849 : static int
3850 3 : bdev_channel_poll_qos(void *arg)
3851 : {
3852 3 : struct spdk_bdev *bdev = arg;
3853 3 : struct spdk_bdev_qos *qos = bdev->internal.qos;
3854 3 : uint64_t now = spdk_get_ticks();
3855 : int i;
3856 : int64_t remaining_last_timeslice;
3857 :
3858 3 : if (spdk_unlikely(qos->thread == NULL)) {
3859 : /* Old QoS was unbound to remove and new QoS is not enabled yet. */
3860 1 : return SPDK_POLLER_IDLE;
3861 : }
3862 :
3863 2 : if (now < (qos->last_timeslice + qos->timeslice_size)) {
3864 : /* We received our callback earlier than expected - return
3865 : * immediately and wait to do accounting until at least one
3866 : * timeslice has actually expired. This should never happen
3867 : * with a well-behaved timer implementation.
3868 : */
3869 0 : return SPDK_POLLER_IDLE;
3870 : }
3871 :
3872 : /* Reset for next round of rate limiting */
3873 10 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
3874 : /* We may have allowed the IOs or bytes to slightly overrun in the last
3875 : * timeslice. remaining_this_timeslice is signed, so if it's negative
3876 : * here, we'll account for the overrun so that the next timeslice will
3877 : * be appropriately reduced.
3878 : */
3879 8 : remaining_last_timeslice = __atomic_exchange_n(&qos->rate_limits[i].remaining_this_timeslice,
3880 : 0, __ATOMIC_RELAXED);
3881 8 : if (remaining_last_timeslice < 0) {
3882 : /* There could be a race condition here as both bdev_qos_rw_queue_io() and bdev_channel_poll_qos()
3883 : * potentially use 2 atomic ops each, so they can intertwine.
3884 : * This race can potentially cause the limits to be a little fuzzy but won't cause any real damage.
3885 : */
3886 0 : __atomic_store_n(&qos->rate_limits[i].remaining_this_timeslice,
3887 : remaining_last_timeslice, __ATOMIC_RELAXED);
3888 : }
3889 : }
3890 :
3891 4 : while (now >= (qos->last_timeslice + qos->timeslice_size)) {
3892 2 : qos->last_timeslice += qos->timeslice_size;
3893 10 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
3894 8 : __atomic_add_fetch(&qos->rate_limits[i].remaining_this_timeslice,
3895 8 : qos->rate_limits[i].max_per_timeslice, __ATOMIC_RELAXED);
3896 : }
3897 : }
3898 :
3899 2 : spdk_bdev_for_each_channel(bdev, bdev_channel_submit_qos_io, qos,
3900 : bdev_channel_submit_qos_io_done);
3901 :
3902 2 : return SPDK_POLLER_BUSY;
3903 : }
3904 :
3905 : static void
3906 73 : bdev_channel_destroy_resource(struct spdk_bdev_channel *ch)
3907 : {
3908 : struct spdk_bdev_shared_resource *shared_resource;
3909 : struct lba_range *range;
3910 :
3911 73 : bdev_free_io_stat(ch->stat);
3912 : #ifdef SPDK_CONFIG_VTUNE
3913 : bdev_free_io_stat(ch->prev_stat);
3914 : #endif
3915 :
3916 73 : while (!TAILQ_EMPTY(&ch->locked_ranges)) {
3917 0 : range = TAILQ_FIRST(&ch->locked_ranges);
3918 0 : TAILQ_REMOVE(&ch->locked_ranges, range, tailq);
3919 0 : free(range);
3920 : }
3921 :
3922 73 : spdk_put_io_channel(ch->channel);
3923 73 : spdk_put_io_channel(ch->accel_channel);
3924 :
3925 73 : shared_resource = ch->shared_resource;
3926 :
3927 73 : assert(TAILQ_EMPTY(&ch->io_locked));
3928 73 : assert(TAILQ_EMPTY(&ch->io_submitted));
3929 73 : assert(TAILQ_EMPTY(&ch->io_accel_exec));
3930 73 : assert(TAILQ_EMPTY(&ch->io_memory_domain));
3931 73 : assert(ch->io_outstanding == 0);
3932 73 : assert(shared_resource->ref > 0);
3933 73 : shared_resource->ref--;
3934 73 : if (shared_resource->ref == 0) {
3935 72 : assert(shared_resource->io_outstanding == 0);
3936 72 : TAILQ_REMOVE(&shared_resource->mgmt_ch->shared_resources, shared_resource, link);
3937 72 : spdk_put_io_channel(spdk_io_channel_from_ctx(shared_resource->mgmt_ch));
3938 72 : spdk_poller_unregister(&shared_resource->nomem_poller);
3939 72 : free(shared_resource);
3940 : }
3941 73 : }
3942 :
3943 : static void
3944 82 : bdev_enable_qos(struct spdk_bdev *bdev, struct spdk_bdev_channel *ch)
3945 : {
3946 82 : struct spdk_bdev_qos *qos = bdev->internal.qos;
3947 : int i;
3948 :
3949 82 : assert(spdk_spin_held(&bdev->internal.spinlock));
3950 :
3951 : /* Rate limiting on this bdev enabled */
3952 82 : if (qos) {
3953 17 : if (qos->ch == NULL) {
3954 : struct spdk_io_channel *io_ch;
3955 :
3956 9 : SPDK_DEBUGLOG(bdev, "Selecting channel %p as QoS channel for bdev %s on thread %p\n", ch,
3957 : bdev->name, spdk_get_thread());
3958 :
3959 : /* No qos channel has been selected, so set one up */
3960 :
3961 : /* Take another reference to ch */
3962 9 : io_ch = spdk_get_io_channel(__bdev_to_io_dev(bdev));
3963 9 : assert(io_ch != NULL);
3964 9 : qos->ch = ch;
3965 :
3966 9 : qos->thread = spdk_io_channel_get_thread(io_ch);
3967 :
3968 45 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
3969 36 : if (bdev_qos_is_iops_rate_limit(i) == true) {
3970 9 : qos->rate_limits[i].min_per_timeslice =
3971 : SPDK_BDEV_QOS_MIN_IO_PER_TIMESLICE;
3972 : } else {
3973 27 : qos->rate_limits[i].min_per_timeslice =
3974 : SPDK_BDEV_QOS_MIN_BYTE_PER_TIMESLICE;
3975 : }
3976 :
3977 36 : if (qos->rate_limits[i].limit == 0) {
3978 2 : qos->rate_limits[i].limit = SPDK_BDEV_QOS_LIMIT_NOT_DEFINED;
3979 : }
3980 : }
3981 9 : bdev_qos_update_max_quota_per_timeslice(qos);
3982 9 : qos->timeslice_size =
3983 9 : SPDK_BDEV_QOS_TIMESLICE_IN_USEC * spdk_get_ticks_hz() / SPDK_SEC_TO_USEC;
3984 9 : qos->last_timeslice = spdk_get_ticks();
3985 9 : qos->poller = SPDK_POLLER_REGISTER(bdev_channel_poll_qos,
3986 : bdev,
3987 : SPDK_BDEV_QOS_TIMESLICE_IN_USEC);
3988 : }
3989 :
3990 17 : ch->flags |= BDEV_CH_QOS_ENABLED;
3991 : }
3992 82 : }
3993 :
3994 : struct poll_timeout_ctx {
3995 : struct spdk_bdev_desc *desc;
3996 : uint64_t timeout_in_sec;
3997 : spdk_bdev_io_timeout_cb cb_fn;
3998 : void *cb_arg;
3999 : };
4000 :
4001 : static void
4002 262 : bdev_desc_free(struct spdk_bdev_desc *desc)
4003 : {
4004 262 : spdk_spin_destroy(&desc->spinlock);
4005 262 : free(desc->media_events_buffer);
4006 262 : free(desc);
4007 262 : }
4008 :
4009 : static void
4010 8 : bdev_channel_poll_timeout_io_done(struct spdk_bdev *bdev, void *_ctx, int status)
4011 : {
4012 8 : struct poll_timeout_ctx *ctx = _ctx;
4013 8 : struct spdk_bdev_desc *desc = ctx->desc;
4014 :
4015 8 : free(ctx);
4016 :
4017 8 : spdk_spin_lock(&desc->spinlock);
4018 8 : desc->refs--;
4019 8 : if (desc->closed == true && desc->refs == 0) {
4020 1 : spdk_spin_unlock(&desc->spinlock);
4021 1 : bdev_desc_free(desc);
4022 1 : return;
4023 : }
4024 7 : spdk_spin_unlock(&desc->spinlock);
4025 : }
4026 :
4027 : static void
4028 13 : bdev_channel_poll_timeout_io(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
4029 : struct spdk_io_channel *io_ch, void *_ctx)
4030 : {
4031 13 : struct poll_timeout_ctx *ctx = _ctx;
4032 13 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
4033 13 : struct spdk_bdev_desc *desc = ctx->desc;
4034 : struct spdk_bdev_io *bdev_io;
4035 : uint64_t now;
4036 :
4037 13 : spdk_spin_lock(&desc->spinlock);
4038 13 : if (desc->closed == true) {
4039 1 : spdk_spin_unlock(&desc->spinlock);
4040 1 : spdk_bdev_for_each_channel_continue(i, -1);
4041 1 : return;
4042 : }
4043 12 : spdk_spin_unlock(&desc->spinlock);
4044 :
4045 12 : now = spdk_get_ticks();
4046 22 : TAILQ_FOREACH(bdev_io, &bdev_ch->io_submitted, internal.ch_link) {
4047 : /* Exclude any I/O that are generated via splitting. */
4048 15 : if (bdev_io->internal.cb == bdev_io_split_done) {
4049 3 : continue;
4050 : }
4051 :
4052 : /* Once we find an I/O that has not timed out, we can immediately
4053 : * exit the loop.
4054 : */
4055 24 : if (now < (bdev_io->internal.submit_tsc +
4056 12 : ctx->timeout_in_sec * spdk_get_ticks_hz())) {
4057 5 : goto end;
4058 : }
4059 :
4060 7 : if (bdev_io->internal.desc == desc) {
4061 7 : ctx->cb_fn(ctx->cb_arg, bdev_io);
4062 : }
4063 : }
4064 :
4065 7 : end:
4066 12 : spdk_bdev_for_each_channel_continue(i, 0);
4067 : }
4068 :
4069 : static int
4070 8 : bdev_poll_timeout_io(void *arg)
4071 : {
4072 8 : struct spdk_bdev_desc *desc = arg;
4073 8 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
4074 : struct poll_timeout_ctx *ctx;
4075 :
4076 8 : ctx = calloc(1, sizeof(struct poll_timeout_ctx));
4077 8 : if (!ctx) {
4078 0 : SPDK_ERRLOG("failed to allocate memory\n");
4079 0 : return SPDK_POLLER_BUSY;
4080 : }
4081 8 : ctx->desc = desc;
4082 8 : ctx->cb_arg = desc->cb_arg;
4083 8 : ctx->cb_fn = desc->cb_fn;
4084 8 : ctx->timeout_in_sec = desc->timeout_in_sec;
4085 :
4086 : /* Take a ref on the descriptor in case it gets closed while we are checking
4087 : * all of the channels.
4088 : */
4089 8 : spdk_spin_lock(&desc->spinlock);
4090 8 : desc->refs++;
4091 8 : spdk_spin_unlock(&desc->spinlock);
4092 :
4093 8 : spdk_bdev_for_each_channel(bdev, bdev_channel_poll_timeout_io, ctx,
4094 : bdev_channel_poll_timeout_io_done);
4095 :
4096 8 : return SPDK_POLLER_BUSY;
4097 : }
4098 :
4099 : int
4100 5 : spdk_bdev_set_timeout(struct spdk_bdev_desc *desc, uint64_t timeout_in_sec,
4101 : spdk_bdev_io_timeout_cb cb_fn, void *cb_arg)
4102 : {
4103 5 : assert(desc->thread == spdk_get_thread());
4104 :
4105 5 : spdk_poller_unregister(&desc->io_timeout_poller);
4106 :
4107 5 : if (timeout_in_sec) {
4108 4 : assert(cb_fn != NULL);
4109 4 : desc->io_timeout_poller = SPDK_POLLER_REGISTER(bdev_poll_timeout_io,
4110 : desc,
4111 : SPDK_BDEV_IO_POLL_INTERVAL_IN_MSEC * SPDK_SEC_TO_USEC /
4112 : 1000);
4113 4 : if (desc->io_timeout_poller == NULL) {
4114 0 : SPDK_ERRLOG("can not register the desc timeout IO poller\n");
4115 0 : return -1;
4116 : }
4117 : }
4118 :
4119 5 : desc->cb_fn = cb_fn;
4120 5 : desc->cb_arg = cb_arg;
4121 5 : desc->timeout_in_sec = timeout_in_sec;
4122 :
4123 5 : return 0;
4124 : }
4125 :
4126 : static int
4127 75 : bdev_channel_create(void *io_device, void *ctx_buf)
4128 : {
4129 75 : struct spdk_bdev *bdev = __bdev_from_io_dev(io_device);
4130 75 : struct spdk_bdev_channel *ch = ctx_buf;
4131 : struct spdk_io_channel *mgmt_io_ch;
4132 : struct spdk_bdev_mgmt_channel *mgmt_ch;
4133 : struct spdk_bdev_shared_resource *shared_resource;
4134 : struct lba_range *range;
4135 :
4136 75 : ch->bdev = bdev;
4137 75 : ch->channel = bdev->fn_table->get_io_channel(bdev->ctxt);
4138 75 : if (!ch->channel) {
4139 2 : return -1;
4140 : }
4141 :
4142 73 : ch->accel_channel = spdk_accel_get_io_channel();
4143 73 : if (!ch->accel_channel) {
4144 0 : spdk_put_io_channel(ch->channel);
4145 0 : return -1;
4146 : }
4147 :
4148 73 : spdk_trace_record(TRACE_BDEV_IOCH_CREATE, bdev->internal.trace_id, 0, 0,
4149 : spdk_thread_get_id(spdk_io_channel_get_thread(ch->channel)));
4150 :
4151 73 : assert(ch->histogram == NULL);
4152 73 : if (bdev->internal.histogram_enabled) {
4153 0 : ch->histogram = spdk_histogram_data_alloc();
4154 0 : if (ch->histogram == NULL) {
4155 0 : SPDK_ERRLOG("Could not allocate histogram\n");
4156 : }
4157 : }
4158 :
4159 73 : mgmt_io_ch = spdk_get_io_channel(&g_bdev_mgr);
4160 73 : if (!mgmt_io_ch) {
4161 0 : spdk_put_io_channel(ch->channel);
4162 0 : spdk_put_io_channel(ch->accel_channel);
4163 0 : return -1;
4164 : }
4165 :
4166 73 : mgmt_ch = __io_ch_to_bdev_mgmt_ch(mgmt_io_ch);
4167 75 : TAILQ_FOREACH(shared_resource, &mgmt_ch->shared_resources, link) {
4168 3 : if (shared_resource->shared_ch == ch->channel) {
4169 1 : spdk_put_io_channel(mgmt_io_ch);
4170 1 : shared_resource->ref++;
4171 1 : break;
4172 : }
4173 : }
4174 :
4175 73 : if (shared_resource == NULL) {
4176 72 : shared_resource = calloc(1, sizeof(*shared_resource));
4177 72 : if (shared_resource == NULL) {
4178 0 : spdk_put_io_channel(ch->channel);
4179 0 : spdk_put_io_channel(ch->accel_channel);
4180 0 : spdk_put_io_channel(mgmt_io_ch);
4181 0 : return -1;
4182 : }
4183 :
4184 72 : shared_resource->mgmt_ch = mgmt_ch;
4185 72 : shared_resource->io_outstanding = 0;
4186 72 : TAILQ_INIT(&shared_resource->nomem_io);
4187 72 : shared_resource->nomem_threshold = 0;
4188 72 : shared_resource->shared_ch = ch->channel;
4189 72 : shared_resource->ref = 1;
4190 72 : TAILQ_INSERT_TAIL(&mgmt_ch->shared_resources, shared_resource, link);
4191 : }
4192 :
4193 73 : ch->io_outstanding = 0;
4194 73 : TAILQ_INIT(&ch->queued_resets);
4195 73 : TAILQ_INIT(&ch->locked_ranges);
4196 73 : TAILQ_INIT(&ch->qos_queued_io);
4197 73 : ch->flags = 0;
4198 73 : ch->trace_id = bdev->internal.trace_id;
4199 73 : ch->shared_resource = shared_resource;
4200 :
4201 73 : TAILQ_INIT(&ch->io_submitted);
4202 73 : TAILQ_INIT(&ch->io_locked);
4203 73 : TAILQ_INIT(&ch->io_accel_exec);
4204 73 : TAILQ_INIT(&ch->io_memory_domain);
4205 :
4206 73 : ch->stat = bdev_alloc_io_stat(false);
4207 73 : if (ch->stat == NULL) {
4208 0 : bdev_channel_destroy_resource(ch);
4209 0 : return -1;
4210 : }
4211 :
4212 73 : ch->stat->ticks_rate = spdk_get_ticks_hz();
4213 :
4214 : #ifdef SPDK_CONFIG_VTUNE
4215 : {
4216 : char *name;
4217 : __itt_init_ittlib(NULL, 0);
4218 : name = spdk_sprintf_alloc("spdk_bdev_%s_%p", ch->bdev->name, ch);
4219 : if (!name) {
4220 : bdev_channel_destroy_resource(ch);
4221 : return -1;
4222 : }
4223 : ch->handle = __itt_string_handle_create(name);
4224 : free(name);
4225 : ch->start_tsc = spdk_get_ticks();
4226 : ch->interval_tsc = spdk_get_ticks_hz() / 100;
4227 : ch->prev_stat = bdev_alloc_io_stat(false);
4228 : if (ch->prev_stat == NULL) {
4229 : bdev_channel_destroy_resource(ch);
4230 : return -1;
4231 : }
4232 : }
4233 : #endif
4234 :
4235 73 : spdk_spin_lock(&bdev->internal.spinlock);
4236 73 : bdev_enable_qos(bdev, ch);
4237 :
4238 74 : TAILQ_FOREACH(range, &bdev->internal.locked_ranges, tailq) {
4239 : struct lba_range *new_range;
4240 :
4241 1 : new_range = calloc(1, sizeof(*new_range));
4242 1 : if (new_range == NULL) {
4243 0 : spdk_spin_unlock(&bdev->internal.spinlock);
4244 0 : bdev_channel_destroy_resource(ch);
4245 0 : return -1;
4246 : }
4247 1 : new_range->length = range->length;
4248 1 : new_range->offset = range->offset;
4249 1 : new_range->locked_ctx = range->locked_ctx;
4250 1 : TAILQ_INSERT_TAIL(&ch->locked_ranges, new_range, tailq);
4251 : }
4252 :
4253 73 : spdk_spin_unlock(&bdev->internal.spinlock);
4254 :
4255 73 : return 0;
4256 : }
4257 :
4258 : static int
4259 0 : bdev_abort_all_buf_io_cb(struct spdk_iobuf_channel *ch, struct spdk_iobuf_entry *entry,
4260 : void *cb_ctx)
4261 : {
4262 0 : struct spdk_bdev_channel *bdev_ch = cb_ctx;
4263 : struct spdk_bdev_io *bdev_io;
4264 : uint64_t buf_len;
4265 :
4266 0 : bdev_io = SPDK_CONTAINEROF(entry, struct spdk_bdev_io, internal.iobuf);
4267 0 : if (bdev_io->internal.ch == bdev_ch) {
4268 0 : buf_len = bdev_io_get_max_buf_len(bdev_io, bdev_io->internal.buf.len);
4269 0 : spdk_iobuf_entry_abort(ch, entry, buf_len);
4270 0 : spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_ABORTED);
4271 : }
4272 :
4273 0 : return 0;
4274 : }
4275 :
4276 : /*
4277 : * Abort I/O that are waiting on a data buffer.
4278 : */
4279 : static void
4280 95 : bdev_abort_all_buf_io(struct spdk_bdev_mgmt_channel *mgmt_ch, struct spdk_bdev_channel *ch)
4281 : {
4282 95 : spdk_iobuf_for_each_entry(&mgmt_ch->iobuf, &mgmt_ch->iobuf.small,
4283 : bdev_abort_all_buf_io_cb, ch);
4284 95 : spdk_iobuf_for_each_entry(&mgmt_ch->iobuf, &mgmt_ch->iobuf.large,
4285 : bdev_abort_all_buf_io_cb, ch);
4286 95 : }
4287 :
4288 : /*
4289 : * Abort I/O that are queued waiting for submission. These types of I/O are
4290 : * linked using the spdk_bdev_io link TAILQ_ENTRY.
4291 : */
4292 : static void
4293 186 : bdev_abort_all_queued_io(bdev_io_tailq_t *queue, struct spdk_bdev_channel *ch)
4294 : {
4295 : struct spdk_bdev_io *bdev_io, *tmp;
4296 :
4297 226 : TAILQ_FOREACH_SAFE(bdev_io, queue, internal.link, tmp) {
4298 40 : if (bdev_io->internal.ch == ch) {
4299 40 : TAILQ_REMOVE(queue, bdev_io, internal.link);
4300 : /*
4301 : * spdk_bdev_io_complete() assumes that the completed I/O had
4302 : * been submitted to the bdev module. Since in this case it
4303 : * hadn't, bump io_outstanding to account for the decrement
4304 : * that spdk_bdev_io_complete() will do.
4305 : */
4306 40 : if (bdev_io->type != SPDK_BDEV_IO_TYPE_RESET) {
4307 39 : bdev_io_increment_outstanding(ch, ch->shared_resource);
4308 : }
4309 40 : spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_ABORTED);
4310 : }
4311 : }
4312 186 : }
4313 :
4314 : static bool
4315 12 : bdev_abort_queued_io(bdev_io_tailq_t *queue, struct spdk_bdev_io *bio_to_abort)
4316 : {
4317 : struct spdk_bdev_io *bdev_io;
4318 :
4319 12 : TAILQ_FOREACH(bdev_io, queue, internal.link) {
4320 0 : if (bdev_io == bio_to_abort) {
4321 0 : TAILQ_REMOVE(queue, bio_to_abort, internal.link);
4322 0 : spdk_bdev_io_complete(bio_to_abort, SPDK_BDEV_IO_STATUS_ABORTED);
4323 0 : return true;
4324 : }
4325 : }
4326 :
4327 12 : return false;
4328 : }
4329 :
4330 : static int
4331 0 : bdev_abort_buf_io_cb(struct spdk_iobuf_channel *ch, struct spdk_iobuf_entry *entry, void *cb_ctx)
4332 : {
4333 0 : struct spdk_bdev_io *bdev_io, *bio_to_abort = cb_ctx;
4334 : uint64_t buf_len;
4335 :
4336 0 : bdev_io = SPDK_CONTAINEROF(entry, struct spdk_bdev_io, internal.iobuf);
4337 0 : if (bdev_io == bio_to_abort) {
4338 0 : buf_len = bdev_io_get_max_buf_len(bdev_io, bdev_io->internal.buf.len);
4339 0 : spdk_iobuf_entry_abort(ch, entry, buf_len);
4340 0 : spdk_bdev_io_complete(bio_to_abort, SPDK_BDEV_IO_STATUS_ABORTED);
4341 0 : return 1;
4342 : }
4343 :
4344 0 : return 0;
4345 : }
4346 :
4347 : static bool
4348 10 : bdev_abort_buf_io(struct spdk_bdev_mgmt_channel *mgmt_ch, struct spdk_bdev_io *bio_to_abort)
4349 : {
4350 : int rc;
4351 :
4352 10 : rc = spdk_iobuf_for_each_entry(&mgmt_ch->iobuf, &mgmt_ch->iobuf.small,
4353 : bdev_abort_buf_io_cb, bio_to_abort);
4354 10 : if (rc == 1) {
4355 0 : return true;
4356 : }
4357 :
4358 10 : rc = spdk_iobuf_for_each_entry(&mgmt_ch->iobuf, &mgmt_ch->iobuf.large,
4359 : bdev_abort_buf_io_cb, bio_to_abort);
4360 10 : return rc == 1;
4361 : }
4362 :
4363 : static void
4364 7 : bdev_qos_channel_destroy(void *cb_arg)
4365 : {
4366 7 : struct spdk_bdev_qos *qos = cb_arg;
4367 :
4368 7 : spdk_put_io_channel(spdk_io_channel_from_ctx(qos->ch));
4369 7 : spdk_poller_unregister(&qos->poller);
4370 :
4371 7 : SPDK_DEBUGLOG(bdev, "Free QoS %p.\n", qos);
4372 :
4373 7 : free(qos);
4374 7 : }
4375 :
4376 : static int
4377 7 : bdev_qos_destroy(struct spdk_bdev *bdev)
4378 : {
4379 : int i;
4380 :
4381 : /*
4382 : * Cleanly shutting down the QoS poller is tricky, because
4383 : * during the asynchronous operation the user could open
4384 : * a new descriptor and create a new channel, spawning
4385 : * a new QoS poller.
4386 : *
4387 : * The strategy is to create a new QoS structure here and swap it
4388 : * in. The shutdown path then continues to refer to the old one
4389 : * until it completes and then releases it.
4390 : */
4391 : struct spdk_bdev_qos *new_qos, *old_qos;
4392 :
4393 7 : old_qos = bdev->internal.qos;
4394 :
4395 7 : new_qos = calloc(1, sizeof(*new_qos));
4396 7 : if (!new_qos) {
4397 0 : SPDK_ERRLOG("Unable to allocate memory to shut down QoS.\n");
4398 0 : return -ENOMEM;
4399 : }
4400 :
4401 : /* Copy the old QoS data into the newly allocated structure */
4402 7 : memcpy(new_qos, old_qos, sizeof(*new_qos));
4403 :
4404 : /* Zero out the key parts of the QoS structure */
4405 7 : new_qos->ch = NULL;
4406 7 : new_qos->thread = NULL;
4407 7 : new_qos->poller = NULL;
4408 : /*
4409 : * The limit member of spdk_bdev_qos_limit structure is not zeroed.
4410 : * It will be used later for the new QoS structure.
4411 : */
4412 35 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
4413 28 : new_qos->rate_limits[i].remaining_this_timeslice = 0;
4414 28 : new_qos->rate_limits[i].min_per_timeslice = 0;
4415 28 : new_qos->rate_limits[i].max_per_timeslice = 0;
4416 : }
4417 :
4418 7 : bdev->internal.qos = new_qos;
4419 :
4420 7 : if (old_qos->thread == NULL) {
4421 0 : free(old_qos);
4422 : } else {
4423 7 : spdk_thread_send_msg(old_qos->thread, bdev_qos_channel_destroy, old_qos);
4424 : }
4425 :
4426 : /* It is safe to continue with destroying the bdev even though the QoS channel hasn't
4427 : * been destroyed yet. The destruction path will end up waiting for the final
4428 : * channel to be put before it releases resources. */
4429 :
4430 7 : return 0;
4431 : }
4432 :
4433 : void
4434 73 : spdk_bdev_add_io_stat(struct spdk_bdev_io_stat *total, struct spdk_bdev_io_stat *add)
4435 : {
4436 73 : total->bytes_read += add->bytes_read;
4437 73 : total->num_read_ops += add->num_read_ops;
4438 73 : total->bytes_written += add->bytes_written;
4439 73 : total->num_write_ops += add->num_write_ops;
4440 73 : total->bytes_unmapped += add->bytes_unmapped;
4441 73 : total->num_unmap_ops += add->num_unmap_ops;
4442 73 : total->bytes_copied += add->bytes_copied;
4443 73 : total->num_copy_ops += add->num_copy_ops;
4444 73 : total->read_latency_ticks += add->read_latency_ticks;
4445 73 : total->write_latency_ticks += add->write_latency_ticks;
4446 73 : total->unmap_latency_ticks += add->unmap_latency_ticks;
4447 73 : total->copy_latency_ticks += add->copy_latency_ticks;
4448 73 : if (total->max_read_latency_ticks < add->max_read_latency_ticks) {
4449 5 : total->max_read_latency_ticks = add->max_read_latency_ticks;
4450 : }
4451 73 : if (total->min_read_latency_ticks > add->min_read_latency_ticks) {
4452 37 : total->min_read_latency_ticks = add->min_read_latency_ticks;
4453 : }
4454 73 : if (total->max_write_latency_ticks < add->max_write_latency_ticks) {
4455 4 : total->max_write_latency_ticks = add->max_write_latency_ticks;
4456 : }
4457 73 : if (total->min_write_latency_ticks > add->min_write_latency_ticks) {
4458 24 : total->min_write_latency_ticks = add->min_write_latency_ticks;
4459 : }
4460 73 : if (total->max_unmap_latency_ticks < add->max_unmap_latency_ticks) {
4461 0 : total->max_unmap_latency_ticks = add->max_unmap_latency_ticks;
4462 : }
4463 73 : if (total->min_unmap_latency_ticks > add->min_unmap_latency_ticks) {
4464 3 : total->min_unmap_latency_ticks = add->min_unmap_latency_ticks;
4465 : }
4466 73 : if (total->max_copy_latency_ticks < add->max_copy_latency_ticks) {
4467 0 : total->max_copy_latency_ticks = add->max_copy_latency_ticks;
4468 : }
4469 73 : if (total->min_copy_latency_ticks > add->min_copy_latency_ticks) {
4470 4 : total->min_copy_latency_ticks = add->min_copy_latency_ticks;
4471 : }
4472 73 : }
4473 :
4474 : static void
4475 1 : bdev_get_io_stat(struct spdk_bdev_io_stat *to_stat, struct spdk_bdev_io_stat *from_stat)
4476 : {
4477 1 : memcpy(to_stat, from_stat, offsetof(struct spdk_bdev_io_stat, io_error));
4478 :
4479 1 : if (to_stat->io_error != NULL && from_stat->io_error != NULL) {
4480 0 : memcpy(to_stat->io_error, from_stat->io_error,
4481 : sizeof(struct spdk_bdev_io_error_stat));
4482 : }
4483 1 : }
4484 :
4485 : void
4486 198 : spdk_bdev_reset_io_stat(struct spdk_bdev_io_stat *stat, enum spdk_bdev_reset_stat_mode mode)
4487 : {
4488 198 : stat->max_read_latency_ticks = 0;
4489 198 : stat->min_read_latency_ticks = UINT64_MAX;
4490 198 : stat->max_write_latency_ticks = 0;
4491 198 : stat->min_write_latency_ticks = UINT64_MAX;
4492 198 : stat->max_unmap_latency_ticks = 0;
4493 198 : stat->min_unmap_latency_ticks = UINT64_MAX;
4494 198 : stat->max_copy_latency_ticks = 0;
4495 198 : stat->min_copy_latency_ticks = UINT64_MAX;
4496 :
4497 198 : if (mode != SPDK_BDEV_RESET_STAT_ALL) {
4498 0 : return;
4499 : }
4500 :
4501 198 : stat->bytes_read = 0;
4502 198 : stat->num_read_ops = 0;
4503 198 : stat->bytes_written = 0;
4504 198 : stat->num_write_ops = 0;
4505 198 : stat->bytes_unmapped = 0;
4506 198 : stat->num_unmap_ops = 0;
4507 198 : stat->bytes_copied = 0;
4508 198 : stat->num_copy_ops = 0;
4509 198 : stat->read_latency_ticks = 0;
4510 198 : stat->write_latency_ticks = 0;
4511 198 : stat->unmap_latency_ticks = 0;
4512 198 : stat->copy_latency_ticks = 0;
4513 :
4514 198 : if (stat->io_error != NULL) {
4515 125 : memset(stat->io_error, 0, sizeof(struct spdk_bdev_io_error_stat));
4516 : }
4517 : }
4518 :
4519 : struct spdk_bdev_io_stat *
4520 198 : bdev_alloc_io_stat(bool io_error_stat)
4521 : {
4522 : struct spdk_bdev_io_stat *stat;
4523 :
4524 198 : stat = malloc(sizeof(struct spdk_bdev_io_stat));
4525 198 : if (stat == NULL) {
4526 0 : return NULL;
4527 : }
4528 :
4529 198 : if (io_error_stat) {
4530 125 : stat->io_error = malloc(sizeof(struct spdk_bdev_io_error_stat));
4531 125 : if (stat->io_error == NULL) {
4532 0 : free(stat);
4533 0 : return NULL;
4534 : }
4535 : } else {
4536 73 : stat->io_error = NULL;
4537 : }
4538 :
4539 198 : spdk_bdev_reset_io_stat(stat, SPDK_BDEV_RESET_STAT_ALL);
4540 :
4541 198 : return stat;
4542 : }
4543 :
4544 : void
4545 198 : bdev_free_io_stat(struct spdk_bdev_io_stat *stat)
4546 : {
4547 198 : if (stat != NULL) {
4548 198 : free(stat->io_error);
4549 198 : free(stat);
4550 : }
4551 198 : }
4552 :
4553 : void
4554 0 : spdk_bdev_dump_io_stat_json(struct spdk_bdev_io_stat *stat, struct spdk_json_write_ctx *w)
4555 : {
4556 : int i;
4557 :
4558 0 : spdk_json_write_named_uint64(w, "bytes_read", stat->bytes_read);
4559 0 : spdk_json_write_named_uint64(w, "num_read_ops", stat->num_read_ops);
4560 0 : spdk_json_write_named_uint64(w, "bytes_written", stat->bytes_written);
4561 0 : spdk_json_write_named_uint64(w, "num_write_ops", stat->num_write_ops);
4562 0 : spdk_json_write_named_uint64(w, "bytes_unmapped", stat->bytes_unmapped);
4563 0 : spdk_json_write_named_uint64(w, "num_unmap_ops", stat->num_unmap_ops);
4564 0 : spdk_json_write_named_uint64(w, "bytes_copied", stat->bytes_copied);
4565 0 : spdk_json_write_named_uint64(w, "num_copy_ops", stat->num_copy_ops);
4566 0 : spdk_json_write_named_uint64(w, "read_latency_ticks", stat->read_latency_ticks);
4567 0 : spdk_json_write_named_uint64(w, "max_read_latency_ticks", stat->max_read_latency_ticks);
4568 0 : spdk_json_write_named_uint64(w, "min_read_latency_ticks",
4569 0 : stat->min_read_latency_ticks != UINT64_MAX ?
4570 : stat->min_read_latency_ticks : 0);
4571 0 : spdk_json_write_named_uint64(w, "write_latency_ticks", stat->write_latency_ticks);
4572 0 : spdk_json_write_named_uint64(w, "max_write_latency_ticks", stat->max_write_latency_ticks);
4573 0 : spdk_json_write_named_uint64(w, "min_write_latency_ticks",
4574 0 : stat->min_write_latency_ticks != UINT64_MAX ?
4575 : stat->min_write_latency_ticks : 0);
4576 0 : spdk_json_write_named_uint64(w, "unmap_latency_ticks", stat->unmap_latency_ticks);
4577 0 : spdk_json_write_named_uint64(w, "max_unmap_latency_ticks", stat->max_unmap_latency_ticks);
4578 0 : spdk_json_write_named_uint64(w, "min_unmap_latency_ticks",
4579 0 : stat->min_unmap_latency_ticks != UINT64_MAX ?
4580 : stat->min_unmap_latency_ticks : 0);
4581 0 : spdk_json_write_named_uint64(w, "copy_latency_ticks", stat->copy_latency_ticks);
4582 0 : spdk_json_write_named_uint64(w, "max_copy_latency_ticks", stat->max_copy_latency_ticks);
4583 0 : spdk_json_write_named_uint64(w, "min_copy_latency_ticks",
4584 0 : stat->min_copy_latency_ticks != UINT64_MAX ?
4585 : stat->min_copy_latency_ticks : 0);
4586 :
4587 0 : if (stat->io_error != NULL) {
4588 0 : spdk_json_write_named_object_begin(w, "io_error");
4589 0 : for (i = 0; i < -SPDK_MIN_BDEV_IO_STATUS; i++) {
4590 0 : if (stat->io_error->error_status[i] != 0) {
4591 0 : spdk_json_write_named_uint32(w, bdev_io_status_get_string(-(i + 1)),
4592 0 : stat->io_error->error_status[i]);
4593 : }
4594 : }
4595 0 : spdk_json_write_object_end(w);
4596 : }
4597 0 : }
4598 :
4599 : static void
4600 77 : bdev_channel_abort_queued_ios(struct spdk_bdev_channel *ch)
4601 : {
4602 77 : struct spdk_bdev_shared_resource *shared_resource = ch->shared_resource;
4603 77 : struct spdk_bdev_mgmt_channel *mgmt_ch = shared_resource->mgmt_ch;
4604 :
4605 77 : bdev_abort_all_queued_io(&shared_resource->nomem_io, ch);
4606 77 : bdev_abort_all_buf_io(mgmt_ch, ch);
4607 77 : }
4608 :
4609 : static void
4610 73 : bdev_channel_destroy(void *io_device, void *ctx_buf)
4611 : {
4612 73 : struct spdk_bdev_channel *ch = ctx_buf;
4613 :
4614 73 : SPDK_DEBUGLOG(bdev, "Destroying channel %p for bdev %s on thread %p\n", ch, ch->bdev->name,
4615 : spdk_get_thread());
4616 :
4617 73 : spdk_trace_record(TRACE_BDEV_IOCH_DESTROY, ch->bdev->internal.trace_id, 0, 0,
4618 : spdk_thread_get_id(spdk_io_channel_get_thread(ch->channel)));
4619 :
4620 : /* This channel is going away, so add its statistics into the bdev so that they don't get lost. */
4621 73 : spdk_spin_lock(&ch->bdev->internal.spinlock);
4622 73 : spdk_bdev_add_io_stat(ch->bdev->internal.stat, ch->stat);
4623 73 : spdk_spin_unlock(&ch->bdev->internal.spinlock);
4624 :
4625 73 : bdev_abort_all_queued_io(&ch->queued_resets, ch);
4626 :
4627 73 : bdev_channel_abort_queued_ios(ch);
4628 :
4629 73 : if (ch->histogram) {
4630 0 : spdk_histogram_data_free(ch->histogram);
4631 : }
4632 :
4633 73 : bdev_channel_destroy_resource(ch);
4634 73 : }
4635 :
4636 : /*
4637 : * If the name already exists in the global bdev name tree, RB_INSERT() returns a pointer
4638 : * to it. Hence we do not have to call bdev_get_by_name() when using this function.
4639 : */
4640 : static int
4641 253 : bdev_name_add(struct spdk_bdev_name *bdev_name, struct spdk_bdev *bdev, const char *name)
4642 : {
4643 : struct spdk_bdev_name *tmp;
4644 :
4645 253 : bdev_name->name = strdup(name);
4646 253 : if (bdev_name->name == NULL) {
4647 0 : SPDK_ERRLOG("Unable to allocate bdev name\n");
4648 0 : return -ENOMEM;
4649 : }
4650 :
4651 253 : bdev_name->bdev = bdev;
4652 :
4653 253 : spdk_spin_lock(&g_bdev_mgr.spinlock);
4654 253 : tmp = RB_INSERT(bdev_name_tree, &g_bdev_mgr.bdev_names, bdev_name);
4655 253 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
4656 :
4657 253 : if (tmp != NULL) {
4658 4 : SPDK_ERRLOG("Bdev name %s already exists\n", name);
4659 4 : free(bdev_name->name);
4660 4 : return -EEXIST;
4661 : }
4662 :
4663 249 : return 0;
4664 : }
4665 :
4666 : static void
4667 249 : bdev_name_del_unsafe(struct spdk_bdev_name *bdev_name)
4668 : {
4669 249 : RB_REMOVE(bdev_name_tree, &g_bdev_mgr.bdev_names, bdev_name);
4670 249 : free(bdev_name->name);
4671 249 : }
4672 :
4673 : static void
4674 5 : bdev_name_del(struct spdk_bdev_name *bdev_name)
4675 : {
4676 5 : spdk_spin_lock(&g_bdev_mgr.spinlock);
4677 5 : bdev_name_del_unsafe(bdev_name);
4678 5 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
4679 5 : }
4680 :
4681 : int
4682 131 : spdk_bdev_alias_add(struct spdk_bdev *bdev, const char *alias)
4683 : {
4684 : struct spdk_bdev_alias *tmp;
4685 : int ret;
4686 :
4687 131 : if (alias == NULL) {
4688 1 : SPDK_ERRLOG("Empty alias passed\n");
4689 1 : return -EINVAL;
4690 : }
4691 :
4692 130 : tmp = calloc(1, sizeof(*tmp));
4693 130 : if (tmp == NULL) {
4694 0 : SPDK_ERRLOG("Unable to allocate alias\n");
4695 0 : return -ENOMEM;
4696 : }
4697 :
4698 130 : ret = bdev_name_add(&tmp->alias, bdev, alias);
4699 130 : if (ret != 0) {
4700 4 : free(tmp);
4701 4 : return ret;
4702 : }
4703 :
4704 126 : TAILQ_INSERT_TAIL(&bdev->aliases, tmp, tailq);
4705 :
4706 126 : return 0;
4707 : }
4708 :
4709 : static int
4710 127 : bdev_alias_del(struct spdk_bdev *bdev, const char *alias,
4711 : void (*alias_del_fn)(struct spdk_bdev_name *n))
4712 : {
4713 : struct spdk_bdev_alias *tmp;
4714 :
4715 132 : TAILQ_FOREACH(tmp, &bdev->aliases, tailq) {
4716 128 : if (strcmp(alias, tmp->alias.name) == 0) {
4717 123 : TAILQ_REMOVE(&bdev->aliases, tmp, tailq);
4718 123 : alias_del_fn(&tmp->alias);
4719 123 : free(tmp);
4720 123 : return 0;
4721 : }
4722 : }
4723 :
4724 4 : return -ENOENT;
4725 : }
4726 :
4727 : int
4728 4 : spdk_bdev_alias_del(struct spdk_bdev *bdev, const char *alias)
4729 : {
4730 : int rc;
4731 :
4732 4 : rc = bdev_alias_del(bdev, alias, bdev_name_del);
4733 4 : if (rc == -ENOENT) {
4734 2 : SPDK_INFOLOG(bdev, "Alias %s does not exist\n", alias);
4735 : }
4736 :
4737 4 : return rc;
4738 : }
4739 :
4740 : void
4741 2 : spdk_bdev_alias_del_all(struct spdk_bdev *bdev)
4742 : {
4743 : struct spdk_bdev_alias *p, *tmp;
4744 :
4745 5 : TAILQ_FOREACH_SAFE(p, &bdev->aliases, tailq, tmp) {
4746 3 : TAILQ_REMOVE(&bdev->aliases, p, tailq);
4747 3 : bdev_name_del(&p->alias);
4748 3 : free(p);
4749 : }
4750 2 : }
4751 :
4752 : struct spdk_io_channel *
4753 75 : spdk_bdev_get_io_channel(struct spdk_bdev_desc *desc)
4754 : {
4755 75 : return spdk_get_io_channel(__bdev_to_io_dev(spdk_bdev_desc_get_bdev(desc)));
4756 : }
4757 :
4758 : void *
4759 0 : spdk_bdev_get_module_ctx(struct spdk_bdev_desc *desc)
4760 : {
4761 0 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
4762 0 : void *ctx = NULL;
4763 :
4764 0 : if (bdev->fn_table->get_module_ctx) {
4765 0 : ctx = bdev->fn_table->get_module_ctx(bdev->ctxt);
4766 : }
4767 :
4768 0 : return ctx;
4769 : }
4770 :
4771 : const char *
4772 0 : spdk_bdev_get_module_name(const struct spdk_bdev *bdev)
4773 : {
4774 0 : return bdev->module->name;
4775 : }
4776 :
4777 : const char *
4778 249 : spdk_bdev_get_name(const struct spdk_bdev *bdev)
4779 : {
4780 249 : return bdev->name;
4781 : }
4782 :
4783 : const char *
4784 0 : spdk_bdev_get_product_name(const struct spdk_bdev *bdev)
4785 : {
4786 0 : return bdev->product_name;
4787 : }
4788 :
4789 : const struct spdk_bdev_aliases_list *
4790 0 : spdk_bdev_get_aliases(const struct spdk_bdev *bdev)
4791 : {
4792 0 : return &bdev->aliases;
4793 : }
4794 :
4795 : uint32_t
4796 5 : spdk_bdev_get_block_size(const struct spdk_bdev *bdev)
4797 : {
4798 5 : return bdev->blocklen;
4799 : }
4800 :
4801 : uint32_t
4802 0 : spdk_bdev_get_write_unit_size(const struct spdk_bdev *bdev)
4803 : {
4804 0 : return bdev->write_unit_size;
4805 : }
4806 :
4807 : uint64_t
4808 0 : spdk_bdev_get_num_blocks(const struct spdk_bdev *bdev)
4809 : {
4810 0 : return bdev->blockcnt;
4811 : }
4812 :
4813 : const char *
4814 0 : spdk_bdev_get_qos_rpc_type(enum spdk_bdev_qos_rate_limit_type type)
4815 : {
4816 0 : return qos_rpc_type[type];
4817 : }
4818 :
4819 : void
4820 0 : spdk_bdev_get_qos_rate_limits(struct spdk_bdev *bdev, uint64_t *limits)
4821 : {
4822 : int i;
4823 :
4824 0 : memset(limits, 0, sizeof(*limits) * SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES);
4825 :
4826 0 : spdk_spin_lock(&bdev->internal.spinlock);
4827 0 : if (bdev->internal.qos) {
4828 0 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
4829 0 : if (bdev->internal.qos->rate_limits[i].limit !=
4830 : SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
4831 0 : limits[i] = bdev->internal.qos->rate_limits[i].limit;
4832 0 : if (bdev_qos_is_iops_rate_limit(i) == false) {
4833 : /* Change from Byte to Megabyte which is user visible. */
4834 0 : limits[i] = limits[i] / 1024 / 1024;
4835 : }
4836 : }
4837 : }
4838 : }
4839 0 : spdk_spin_unlock(&bdev->internal.spinlock);
4840 0 : }
4841 :
4842 : size_t
4843 335 : spdk_bdev_get_buf_align(const struct spdk_bdev *bdev)
4844 : {
4845 335 : return 1 << bdev->required_alignment;
4846 : }
4847 :
4848 : uint32_t
4849 0 : spdk_bdev_get_optimal_io_boundary(const struct spdk_bdev *bdev)
4850 : {
4851 0 : return bdev->optimal_io_boundary;
4852 : }
4853 :
4854 : bool
4855 0 : spdk_bdev_has_write_cache(const struct spdk_bdev *bdev)
4856 : {
4857 0 : return bdev->write_cache;
4858 : }
4859 :
4860 : const struct spdk_uuid *
4861 0 : spdk_bdev_get_uuid(const struct spdk_bdev *bdev)
4862 : {
4863 0 : return &bdev->uuid;
4864 : }
4865 :
4866 : uint16_t
4867 0 : spdk_bdev_get_acwu(const struct spdk_bdev *bdev)
4868 : {
4869 0 : return bdev->acwu;
4870 : }
4871 :
4872 : uint32_t
4873 29 : spdk_bdev_get_md_size(const struct spdk_bdev *bdev)
4874 : {
4875 29 : return bdev->md_len;
4876 : }
4877 :
4878 : bool
4879 128 : spdk_bdev_is_md_interleaved(const struct spdk_bdev *bdev)
4880 : {
4881 128 : return (bdev->md_len != 0) && bdev->md_interleave;
4882 : }
4883 :
4884 : bool
4885 170 : spdk_bdev_is_md_separate(const struct spdk_bdev *bdev)
4886 : {
4887 170 : return (bdev->md_len != 0) && !bdev->md_interleave;
4888 : }
4889 :
4890 : bool
4891 0 : spdk_bdev_is_zoned(const struct spdk_bdev *bdev)
4892 : {
4893 0 : return bdev->zoned;
4894 : }
4895 :
4896 : uint32_t
4897 119 : spdk_bdev_get_data_block_size(const struct spdk_bdev *bdev)
4898 : {
4899 119 : if (spdk_bdev_is_md_interleaved(bdev)) {
4900 0 : return bdev->blocklen - bdev->md_len;
4901 : } else {
4902 119 : return bdev->blocklen;
4903 : }
4904 : }
4905 :
4906 : uint32_t
4907 0 : spdk_bdev_get_physical_block_size(const struct spdk_bdev *bdev)
4908 : {
4909 0 : return bdev->phys_blocklen;
4910 : }
4911 :
4912 : static uint32_t
4913 9 : _bdev_get_block_size_with_md(const struct spdk_bdev *bdev)
4914 : {
4915 9 : if (!spdk_bdev_is_md_interleaved(bdev)) {
4916 6 : return bdev->blocklen + bdev->md_len;
4917 : } else {
4918 3 : return bdev->blocklen;
4919 : }
4920 : }
4921 :
4922 : /* We have to use the typedef in the function declaration to appease astyle. */
4923 : typedef enum spdk_dif_type spdk_dif_type_t;
4924 : typedef enum spdk_dif_pi_format spdk_dif_pi_format_t;
4925 :
4926 : spdk_dif_type_t
4927 0 : spdk_bdev_get_dif_type(const struct spdk_bdev *bdev)
4928 : {
4929 0 : if (bdev->md_len != 0) {
4930 0 : return bdev->dif_type;
4931 : } else {
4932 0 : return SPDK_DIF_DISABLE;
4933 : }
4934 : }
4935 :
4936 : spdk_dif_pi_format_t
4937 0 : spdk_bdev_get_dif_pi_format(const struct spdk_bdev *bdev)
4938 : {
4939 0 : return bdev->dif_pi_format;
4940 : }
4941 :
4942 : bool
4943 0 : spdk_bdev_is_dif_head_of_md(const struct spdk_bdev *bdev)
4944 : {
4945 0 : if (spdk_bdev_get_dif_type(bdev) != SPDK_DIF_DISABLE) {
4946 0 : return bdev->dif_is_head_of_md;
4947 : } else {
4948 0 : return false;
4949 : }
4950 : }
4951 :
4952 : bool
4953 0 : spdk_bdev_is_dif_check_enabled(const struct spdk_bdev *bdev,
4954 : enum spdk_dif_check_type check_type)
4955 : {
4956 0 : if (spdk_bdev_get_dif_type(bdev) == SPDK_DIF_DISABLE) {
4957 0 : return false;
4958 : }
4959 :
4960 0 : switch (check_type) {
4961 0 : case SPDK_DIF_CHECK_TYPE_REFTAG:
4962 0 : return (bdev->dif_check_flags & SPDK_DIF_FLAGS_REFTAG_CHECK) != 0;
4963 0 : case SPDK_DIF_CHECK_TYPE_APPTAG:
4964 0 : return (bdev->dif_check_flags & SPDK_DIF_FLAGS_APPTAG_CHECK) != 0;
4965 0 : case SPDK_DIF_CHECK_TYPE_GUARD:
4966 0 : return (bdev->dif_check_flags & SPDK_DIF_FLAGS_GUARD_CHECK) != 0;
4967 0 : default:
4968 0 : return false;
4969 : }
4970 : }
4971 :
4972 : static uint32_t
4973 3 : bdev_get_max_write(const struct spdk_bdev *bdev, uint64_t num_bytes)
4974 : {
4975 : uint64_t aligned_length, max_write_blocks;
4976 :
4977 3 : aligned_length = num_bytes - (spdk_bdev_get_buf_align(bdev) - 1);
4978 3 : max_write_blocks = aligned_length / _bdev_get_block_size_with_md(bdev);
4979 3 : max_write_blocks -= max_write_blocks % bdev->write_unit_size;
4980 :
4981 3 : return max_write_blocks;
4982 : }
4983 :
4984 : uint32_t
4985 1 : spdk_bdev_get_max_copy(const struct spdk_bdev *bdev)
4986 : {
4987 1 : return bdev->max_copy;
4988 : }
4989 :
4990 : uint64_t
4991 0 : spdk_bdev_get_qd(const struct spdk_bdev *bdev)
4992 : {
4993 0 : return bdev->internal.measured_queue_depth;
4994 : }
4995 :
4996 : uint64_t
4997 0 : spdk_bdev_get_qd_sampling_period(const struct spdk_bdev *bdev)
4998 : {
4999 0 : return bdev->internal.period;
5000 : }
5001 :
5002 : uint64_t
5003 0 : spdk_bdev_get_weighted_io_time(const struct spdk_bdev *bdev)
5004 : {
5005 0 : return bdev->internal.weighted_io_time;
5006 : }
5007 :
5008 : uint64_t
5009 0 : spdk_bdev_get_io_time(const struct spdk_bdev *bdev)
5010 : {
5011 0 : return bdev->internal.io_time;
5012 : }
5013 :
5014 0 : union spdk_bdev_nvme_ctratt spdk_bdev_get_nvme_ctratt(struct spdk_bdev *bdev)
5015 : {
5016 0 : return bdev->ctratt;
5017 : }
5018 :
5019 : static void bdev_update_qd_sampling_period(void *ctx);
5020 :
5021 : static void
5022 1 : _calculate_measured_qd_cpl(struct spdk_bdev *bdev, void *_ctx, int status)
5023 : {
5024 1 : bdev->internal.measured_queue_depth = bdev->internal.temporary_queue_depth;
5025 :
5026 1 : if (bdev->internal.measured_queue_depth) {
5027 0 : bdev->internal.io_time += bdev->internal.period;
5028 0 : bdev->internal.weighted_io_time += bdev->internal.period * bdev->internal.measured_queue_depth;
5029 : }
5030 :
5031 1 : bdev->internal.qd_poll_in_progress = false;
5032 :
5033 1 : bdev_update_qd_sampling_period(bdev);
5034 1 : }
5035 :
5036 : static void
5037 1 : _calculate_measured_qd(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
5038 : struct spdk_io_channel *io_ch, void *_ctx)
5039 : {
5040 1 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(io_ch);
5041 :
5042 1 : bdev->internal.temporary_queue_depth += ch->io_outstanding;
5043 1 : spdk_bdev_for_each_channel_continue(i, 0);
5044 1 : }
5045 :
5046 : static int
5047 1 : bdev_calculate_measured_queue_depth(void *ctx)
5048 : {
5049 1 : struct spdk_bdev *bdev = ctx;
5050 :
5051 1 : bdev->internal.qd_poll_in_progress = true;
5052 1 : bdev->internal.temporary_queue_depth = 0;
5053 1 : spdk_bdev_for_each_channel(bdev, _calculate_measured_qd, bdev, _calculate_measured_qd_cpl);
5054 1 : return SPDK_POLLER_BUSY;
5055 : }
5056 :
5057 : static void
5058 5 : bdev_update_qd_sampling_period(void *ctx)
5059 : {
5060 5 : struct spdk_bdev *bdev = ctx;
5061 :
5062 5 : if (bdev->internal.period == bdev->internal.new_period) {
5063 0 : return;
5064 : }
5065 :
5066 5 : if (bdev->internal.qd_poll_in_progress) {
5067 1 : return;
5068 : }
5069 :
5070 4 : bdev->internal.period = bdev->internal.new_period;
5071 :
5072 4 : spdk_poller_unregister(&bdev->internal.qd_poller);
5073 4 : if (bdev->internal.period != 0) {
5074 2 : bdev->internal.qd_poller = SPDK_POLLER_REGISTER(bdev_calculate_measured_queue_depth,
5075 : bdev, bdev->internal.period);
5076 : } else {
5077 2 : spdk_bdev_close(bdev->internal.qd_desc);
5078 2 : bdev->internal.qd_desc = NULL;
5079 : }
5080 : }
5081 :
5082 : static void
5083 0 : _tmp_bdev_event_cb(enum spdk_bdev_event_type type, struct spdk_bdev *bdev, void *ctx)
5084 : {
5085 0 : SPDK_NOTICELOG("Unexpected event type: %d\n", type);
5086 0 : }
5087 :
5088 : void
5089 128 : spdk_bdev_set_qd_sampling_period(struct spdk_bdev *bdev, uint64_t period)
5090 : {
5091 : int rc;
5092 :
5093 128 : if (bdev->internal.new_period == period) {
5094 122 : return;
5095 : }
5096 :
5097 6 : bdev->internal.new_period = period;
5098 :
5099 6 : if (bdev->internal.qd_desc != NULL) {
5100 4 : assert(bdev->internal.period != 0);
5101 :
5102 4 : spdk_thread_send_msg(bdev->internal.qd_desc->thread,
5103 : bdev_update_qd_sampling_period, bdev);
5104 4 : return;
5105 : }
5106 :
5107 2 : assert(bdev->internal.period == 0);
5108 :
5109 2 : rc = spdk_bdev_open_ext(spdk_bdev_get_name(bdev), false, _tmp_bdev_event_cb,
5110 : NULL, &bdev->internal.qd_desc);
5111 2 : if (rc != 0) {
5112 0 : return;
5113 : }
5114 :
5115 2 : bdev->internal.period = period;
5116 2 : bdev->internal.qd_poller = SPDK_POLLER_REGISTER(bdev_calculate_measured_queue_depth,
5117 : bdev, period);
5118 : }
5119 :
5120 : struct bdev_get_current_qd_ctx {
5121 : uint64_t current_qd;
5122 : spdk_bdev_get_current_qd_cb cb_fn;
5123 : void *cb_arg;
5124 : };
5125 :
5126 : static void
5127 0 : bdev_get_current_qd_done(struct spdk_bdev *bdev, void *_ctx, int status)
5128 : {
5129 0 : struct bdev_get_current_qd_ctx *ctx = _ctx;
5130 :
5131 0 : ctx->cb_fn(bdev, ctx->current_qd, ctx->cb_arg, 0);
5132 :
5133 0 : free(ctx);
5134 0 : }
5135 :
5136 : static void
5137 0 : bdev_get_current_qd(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
5138 : struct spdk_io_channel *io_ch, void *_ctx)
5139 : {
5140 0 : struct bdev_get_current_qd_ctx *ctx = _ctx;
5141 0 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
5142 :
5143 0 : ctx->current_qd += bdev_ch->io_outstanding;
5144 :
5145 0 : spdk_bdev_for_each_channel_continue(i, 0);
5146 0 : }
5147 :
5148 : void
5149 0 : spdk_bdev_get_current_qd(struct spdk_bdev *bdev, spdk_bdev_get_current_qd_cb cb_fn,
5150 : void *cb_arg)
5151 : {
5152 : struct bdev_get_current_qd_ctx *ctx;
5153 :
5154 0 : assert(cb_fn != NULL);
5155 :
5156 0 : ctx = calloc(1, sizeof(*ctx));
5157 0 : if (ctx == NULL) {
5158 0 : cb_fn(bdev, 0, cb_arg, -ENOMEM);
5159 0 : return;
5160 : }
5161 :
5162 0 : ctx->cb_fn = cb_fn;
5163 0 : ctx->cb_arg = cb_arg;
5164 :
5165 0 : spdk_bdev_for_each_channel(bdev, bdev_get_current_qd, ctx, bdev_get_current_qd_done);
5166 : }
5167 :
5168 : static void
5169 25 : _event_notify(struct spdk_bdev_desc *desc, enum spdk_bdev_event_type type)
5170 : {
5171 25 : assert(desc->thread == spdk_get_thread());
5172 :
5173 25 : spdk_spin_lock(&desc->spinlock);
5174 25 : desc->refs--;
5175 25 : if (!desc->closed) {
5176 14 : spdk_spin_unlock(&desc->spinlock);
5177 14 : desc->callback.event_fn(type,
5178 : desc->bdev,
5179 : desc->callback.ctx);
5180 14 : return;
5181 11 : } else if (desc->refs == 0) {
5182 : /* This descriptor was closed after this event_notify message was sent.
5183 : * spdk_bdev_close() could not free the descriptor since this message was
5184 : * in flight, so we free it now using bdev_desc_free().
5185 : */
5186 10 : spdk_spin_unlock(&desc->spinlock);
5187 10 : bdev_desc_free(desc);
5188 10 : return;
5189 : }
5190 1 : spdk_spin_unlock(&desc->spinlock);
5191 : }
5192 :
5193 : static void
5194 25 : event_notify(struct spdk_bdev_desc *desc, spdk_msg_fn event_notify_fn)
5195 : {
5196 25 : spdk_spin_lock(&desc->spinlock);
5197 25 : desc->refs++;
5198 25 : spdk_thread_send_msg(desc->thread, event_notify_fn, desc);
5199 25 : spdk_spin_unlock(&desc->spinlock);
5200 25 : }
5201 :
5202 : static void
5203 6 : _resize_notify(void *ctx)
5204 : {
5205 6 : struct spdk_bdev_desc *desc = ctx;
5206 :
5207 6 : _event_notify(desc, SPDK_BDEV_EVENT_RESIZE);
5208 6 : }
5209 :
5210 : int
5211 11 : spdk_bdev_notify_blockcnt_change(struct spdk_bdev *bdev, uint64_t size)
5212 : {
5213 : struct spdk_bdev_desc *desc;
5214 : int ret;
5215 :
5216 11 : if (size == bdev->blockcnt) {
5217 0 : return 0;
5218 : }
5219 :
5220 11 : spdk_spin_lock(&bdev->internal.spinlock);
5221 :
5222 : /* bdev has open descriptors */
5223 11 : if (!TAILQ_EMPTY(&bdev->internal.open_descs) &&
5224 7 : bdev->blockcnt > size) {
5225 1 : ret = -EBUSY;
5226 : } else {
5227 10 : bdev->blockcnt = size;
5228 16 : TAILQ_FOREACH(desc, &bdev->internal.open_descs, link) {
5229 6 : event_notify(desc, _resize_notify);
5230 : }
5231 10 : ret = 0;
5232 : }
5233 :
5234 11 : spdk_spin_unlock(&bdev->internal.spinlock);
5235 :
5236 11 : return ret;
5237 : }
5238 :
5239 : /*
5240 : * Convert I/O offset and length from bytes to blocks.
5241 : *
5242 : * Returns zero on success or non-zero if the byte parameters aren't divisible by the block size.
5243 : */
5244 : static uint64_t
5245 19 : bdev_bytes_to_blocks(struct spdk_bdev *bdev, uint64_t offset_bytes, uint64_t *offset_blocks,
5246 : uint64_t num_bytes, uint64_t *num_blocks)
5247 : {
5248 19 : uint32_t block_size = bdev->blocklen;
5249 : uint8_t shift_cnt;
5250 :
5251 : /* Avoid expensive div operations if possible. These spdk_u32 functions are very cheap. */
5252 19 : if (spdk_likely(spdk_u32_is_pow2(block_size))) {
5253 16 : shift_cnt = spdk_u32log2(block_size);
5254 16 : *offset_blocks = offset_bytes >> shift_cnt;
5255 16 : *num_blocks = num_bytes >> shift_cnt;
5256 16 : return (offset_bytes - (*offset_blocks << shift_cnt)) |
5257 16 : (num_bytes - (*num_blocks << shift_cnt));
5258 : } else {
5259 3 : *offset_blocks = offset_bytes / block_size;
5260 3 : *num_blocks = num_bytes / block_size;
5261 3 : return (offset_bytes % block_size) | (num_bytes % block_size);
5262 : }
5263 : }
5264 :
5265 : static bool
5266 688 : bdev_io_valid_blocks(struct spdk_bdev *bdev, uint64_t offset_blocks, uint64_t num_blocks)
5267 : {
5268 : /* Return failure if offset_blocks + num_blocks is less than offset_blocks; indicates there
5269 : * has been an overflow and hence the offset has been wrapped around */
5270 688 : if (offset_blocks + num_blocks < offset_blocks) {
5271 1 : return false;
5272 : }
5273 :
5274 : /* Return failure if offset_blocks + num_blocks exceeds the size of the bdev */
5275 687 : if (offset_blocks + num_blocks > bdev->blockcnt) {
5276 2 : return false;
5277 : }
5278 :
5279 685 : return true;
5280 : }
5281 :
5282 : static void
5283 2 : bdev_seek_complete_cb(void *ctx)
5284 : {
5285 2 : struct spdk_bdev_io *bdev_io = ctx;
5286 :
5287 2 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
5288 2 : bdev_io->internal.cb(bdev_io, true, bdev_io->internal.caller_ctx);
5289 2 : }
5290 :
5291 : static int
5292 4 : bdev_seek(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5293 : uint64_t offset_blocks, enum spdk_bdev_io_type io_type,
5294 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5295 : {
5296 4 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5297 : struct spdk_bdev_io *bdev_io;
5298 4 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5299 :
5300 4 : assert(io_type == SPDK_BDEV_IO_TYPE_SEEK_DATA || io_type == SPDK_BDEV_IO_TYPE_SEEK_HOLE);
5301 :
5302 : /* Check if offset_blocks is valid looking at the validity of one block */
5303 4 : if (!bdev_io_valid_blocks(bdev, offset_blocks, 1)) {
5304 0 : return -EINVAL;
5305 : }
5306 :
5307 4 : bdev_io = bdev_channel_get_io(channel);
5308 4 : if (!bdev_io) {
5309 0 : return -ENOMEM;
5310 : }
5311 :
5312 4 : bdev_io->internal.ch = channel;
5313 4 : bdev_io->internal.desc = desc;
5314 4 : bdev_io->type = io_type;
5315 4 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5316 4 : bdev_io->u.bdev.memory_domain = NULL;
5317 4 : bdev_io->u.bdev.memory_domain_ctx = NULL;
5318 4 : bdev_io->u.bdev.accel_sequence = NULL;
5319 4 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5320 :
5321 4 : if (!spdk_bdev_io_type_supported(bdev, io_type)) {
5322 : /* In case bdev doesn't support seek to next data/hole offset,
5323 : * it is assumed that only data and no holes are present */
5324 2 : if (io_type == SPDK_BDEV_IO_TYPE_SEEK_DATA) {
5325 1 : bdev_io->u.bdev.seek.offset = offset_blocks;
5326 : } else {
5327 1 : bdev_io->u.bdev.seek.offset = UINT64_MAX;
5328 : }
5329 :
5330 2 : spdk_thread_send_msg(spdk_get_thread(), bdev_seek_complete_cb, bdev_io);
5331 2 : return 0;
5332 : }
5333 :
5334 2 : bdev_io_submit(bdev_io);
5335 2 : return 0;
5336 : }
5337 :
5338 : int
5339 2 : spdk_bdev_seek_data(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5340 : uint64_t offset_blocks,
5341 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5342 : {
5343 2 : return bdev_seek(desc, ch, offset_blocks, SPDK_BDEV_IO_TYPE_SEEK_DATA, cb, cb_arg);
5344 : }
5345 :
5346 : int
5347 2 : spdk_bdev_seek_hole(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5348 : uint64_t offset_blocks,
5349 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5350 : {
5351 2 : return bdev_seek(desc, ch, offset_blocks, SPDK_BDEV_IO_TYPE_SEEK_HOLE, cb, cb_arg);
5352 : }
5353 :
5354 : uint64_t
5355 4 : spdk_bdev_io_get_seek_offset(const struct spdk_bdev_io *bdev_io)
5356 : {
5357 4 : return bdev_io->u.bdev.seek.offset;
5358 : }
5359 :
5360 : static int
5361 203 : bdev_read_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch, void *buf,
5362 : void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
5363 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5364 : {
5365 203 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5366 : struct spdk_bdev_io *bdev_io;
5367 203 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5368 :
5369 203 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
5370 0 : return -EINVAL;
5371 : }
5372 :
5373 203 : bdev_io = bdev_channel_get_io(channel);
5374 203 : if (!bdev_io) {
5375 1 : return -ENOMEM;
5376 : }
5377 :
5378 202 : bdev_io->internal.ch = channel;
5379 202 : bdev_io->internal.desc = desc;
5380 202 : bdev_io->type = SPDK_BDEV_IO_TYPE_READ;
5381 202 : bdev_io->u.bdev.iovs = &bdev_io->iov;
5382 202 : bdev_io->u.bdev.iovs[0].iov_base = buf;
5383 202 : bdev_io->u.bdev.iovs[0].iov_len = num_blocks * bdev->blocklen;
5384 202 : bdev_io->u.bdev.iovcnt = 1;
5385 202 : bdev_io->u.bdev.md_buf = md_buf;
5386 202 : bdev_io->u.bdev.num_blocks = num_blocks;
5387 202 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5388 202 : bdev_io->u.bdev.memory_domain = NULL;
5389 202 : bdev_io->u.bdev.memory_domain_ctx = NULL;
5390 202 : bdev_io->u.bdev.accel_sequence = NULL;
5391 202 : bdev_io->u.bdev.dif_check_flags = bdev->dif_check_flags;
5392 202 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5393 :
5394 202 : bdev_io_submit(bdev_io);
5395 202 : return 0;
5396 : }
5397 :
5398 : int
5399 2 : spdk_bdev_read(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5400 : void *buf, uint64_t offset, uint64_t nbytes,
5401 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5402 : {
5403 2 : uint64_t offset_blocks, num_blocks;
5404 :
5405 2 : if (bdev_bytes_to_blocks(spdk_bdev_desc_get_bdev(desc), offset, &offset_blocks,
5406 : nbytes, &num_blocks) != 0) {
5407 0 : return -EINVAL;
5408 : }
5409 :
5410 2 : return spdk_bdev_read_blocks(desc, ch, buf, offset_blocks, num_blocks, cb, cb_arg);
5411 : }
5412 :
5413 : int
5414 199 : spdk_bdev_read_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5415 : void *buf, uint64_t offset_blocks, uint64_t num_blocks,
5416 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5417 : {
5418 199 : return bdev_read_blocks_with_md(desc, ch, buf, NULL, offset_blocks, num_blocks, cb, cb_arg);
5419 : }
5420 :
5421 : int
5422 4 : spdk_bdev_read_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5423 : void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
5424 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5425 : {
5426 4 : struct iovec iov = {
5427 : .iov_base = buf,
5428 : };
5429 :
5430 4 : if (md_buf && !spdk_bdev_is_md_separate(spdk_bdev_desc_get_bdev(desc))) {
5431 0 : return -EINVAL;
5432 : }
5433 :
5434 4 : if (md_buf && !_is_buf_allocated(&iov)) {
5435 0 : return -EINVAL;
5436 : }
5437 :
5438 4 : return bdev_read_blocks_with_md(desc, ch, buf, md_buf, offset_blocks, num_blocks,
5439 : cb, cb_arg);
5440 : }
5441 :
5442 : int
5443 5 : spdk_bdev_readv(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5444 : struct iovec *iov, int iovcnt,
5445 : uint64_t offset, uint64_t nbytes,
5446 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5447 : {
5448 5 : uint64_t offset_blocks, num_blocks;
5449 :
5450 5 : if (bdev_bytes_to_blocks(spdk_bdev_desc_get_bdev(desc), offset, &offset_blocks,
5451 : nbytes, &num_blocks) != 0) {
5452 0 : return -EINVAL;
5453 : }
5454 :
5455 5 : return spdk_bdev_readv_blocks(desc, ch, iov, iovcnt, offset_blocks, num_blocks, cb, cb_arg);
5456 : }
5457 :
5458 : static int
5459 235 : bdev_readv_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5460 : struct iovec *iov, int iovcnt, void *md_buf, uint64_t offset_blocks,
5461 : uint64_t num_blocks, struct spdk_memory_domain *domain, void *domain_ctx,
5462 : struct spdk_accel_sequence *seq, uint32_t dif_check_flags,
5463 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5464 : {
5465 235 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5466 : struct spdk_bdev_io *bdev_io;
5467 235 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5468 :
5469 235 : if (spdk_unlikely(!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks))) {
5470 0 : return -EINVAL;
5471 : }
5472 :
5473 235 : bdev_io = bdev_channel_get_io(channel);
5474 235 : if (spdk_unlikely(!bdev_io)) {
5475 2 : return -ENOMEM;
5476 : }
5477 :
5478 233 : bdev_io->internal.ch = channel;
5479 233 : bdev_io->internal.desc = desc;
5480 233 : bdev_io->type = SPDK_BDEV_IO_TYPE_READ;
5481 233 : bdev_io->u.bdev.iovs = iov;
5482 233 : bdev_io->u.bdev.iovcnt = iovcnt;
5483 233 : bdev_io->u.bdev.md_buf = md_buf;
5484 233 : bdev_io->u.bdev.num_blocks = num_blocks;
5485 233 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5486 233 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5487 :
5488 233 : if (seq != NULL) {
5489 0 : bdev_io->internal.f.has_accel_sequence = true;
5490 0 : bdev_io->internal.accel_sequence = seq;
5491 : }
5492 :
5493 233 : if (domain != NULL) {
5494 2 : bdev_io->internal.f.has_memory_domain = true;
5495 2 : bdev_io->internal.memory_domain = domain;
5496 2 : bdev_io->internal.memory_domain_ctx = domain_ctx;
5497 : }
5498 :
5499 233 : bdev_io->u.bdev.memory_domain = domain;
5500 233 : bdev_io->u.bdev.memory_domain_ctx = domain_ctx;
5501 233 : bdev_io->u.bdev.accel_sequence = seq;
5502 233 : bdev_io->u.bdev.dif_check_flags = dif_check_flags;
5503 :
5504 233 : _bdev_io_submit_ext(desc, bdev_io);
5505 :
5506 233 : return 0;
5507 : }
5508 :
5509 : int
5510 24 : spdk_bdev_readv_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5511 : struct iovec *iov, int iovcnt,
5512 : uint64_t offset_blocks, uint64_t num_blocks,
5513 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5514 : {
5515 24 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5516 :
5517 24 : return bdev_readv_blocks_with_md(desc, ch, iov, iovcnt, NULL, offset_blocks,
5518 : num_blocks, NULL, NULL, NULL, bdev->dif_check_flags, cb, cb_arg);
5519 : }
5520 :
5521 : int
5522 4 : spdk_bdev_readv_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5523 : struct iovec *iov, int iovcnt, void *md_buf,
5524 : uint64_t offset_blocks, uint64_t num_blocks,
5525 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5526 : {
5527 4 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5528 :
5529 4 : if (md_buf && !spdk_bdev_is_md_separate(bdev)) {
5530 0 : return -EINVAL;
5531 : }
5532 :
5533 4 : if (md_buf && !_is_buf_allocated(iov)) {
5534 0 : return -EINVAL;
5535 : }
5536 :
5537 4 : return bdev_readv_blocks_with_md(desc, ch, iov, iovcnt, md_buf, offset_blocks,
5538 : num_blocks, NULL, NULL, NULL, bdev->dif_check_flags, cb, cb_arg);
5539 : }
5540 :
5541 : static inline bool
5542 14 : _bdev_io_check_opts(struct spdk_bdev_ext_io_opts *opts, struct iovec *iov)
5543 : {
5544 : /*
5545 : * We check if opts size is at least of size when we first introduced
5546 : * spdk_bdev_ext_io_opts (ac6f2bdd8d) since access to those members
5547 : * are not checked internal.
5548 : */
5549 14 : return opts->size >= offsetof(struct spdk_bdev_ext_io_opts, metadata) +
5550 10 : sizeof(opts->metadata) &&
5551 22 : opts->size <= sizeof(*opts) &&
5552 : /* When memory domain is used, the user must provide data buffers */
5553 8 : (!opts->memory_domain || (iov && iov[0].iov_base));
5554 : }
5555 :
5556 : int
5557 8 : spdk_bdev_readv_blocks_ext(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5558 : struct iovec *iov, int iovcnt,
5559 : uint64_t offset_blocks, uint64_t num_blocks,
5560 : spdk_bdev_io_completion_cb cb, void *cb_arg,
5561 : struct spdk_bdev_ext_io_opts *opts)
5562 : {
5563 8 : struct spdk_memory_domain *domain = NULL;
5564 8 : struct spdk_accel_sequence *seq = NULL;
5565 8 : void *domain_ctx = NULL, *md = NULL;
5566 8 : uint32_t dif_check_flags = 0;
5567 8 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5568 :
5569 8 : if (opts) {
5570 7 : if (spdk_unlikely(!_bdev_io_check_opts(opts, iov))) {
5571 3 : return -EINVAL;
5572 : }
5573 :
5574 4 : md = opts->metadata;
5575 4 : domain = bdev_get_ext_io_opt(opts, memory_domain, NULL);
5576 4 : domain_ctx = bdev_get_ext_io_opt(opts, memory_domain_ctx, NULL);
5577 4 : seq = bdev_get_ext_io_opt(opts, accel_sequence, NULL);
5578 4 : if (md) {
5579 4 : if (spdk_unlikely(!spdk_bdev_is_md_separate(bdev))) {
5580 0 : return -EINVAL;
5581 : }
5582 :
5583 4 : if (spdk_unlikely(!_is_buf_allocated(iov))) {
5584 0 : return -EINVAL;
5585 : }
5586 :
5587 4 : if (spdk_unlikely(seq != NULL)) {
5588 0 : return -EINVAL;
5589 : }
5590 : }
5591 : }
5592 :
5593 10 : dif_check_flags = bdev->dif_check_flags &
5594 5 : ~(bdev_get_ext_io_opt(opts, dif_check_flags_exclude_mask, 0));
5595 :
5596 5 : return bdev_readv_blocks_with_md(desc, ch, iov, iovcnt, md, offset_blocks,
5597 : num_blocks, domain, domain_ctx, seq, dif_check_flags, cb, cb_arg);
5598 : }
5599 :
5600 : static int
5601 33 : bdev_write_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5602 : void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
5603 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5604 : {
5605 33 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5606 : struct spdk_bdev_io *bdev_io;
5607 33 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5608 :
5609 33 : if (!desc->write) {
5610 0 : return -EBADF;
5611 : }
5612 :
5613 33 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
5614 0 : return -EINVAL;
5615 : }
5616 :
5617 33 : bdev_io = bdev_channel_get_io(channel);
5618 33 : if (!bdev_io) {
5619 0 : return -ENOMEM;
5620 : }
5621 :
5622 33 : bdev_io->internal.ch = channel;
5623 33 : bdev_io->internal.desc = desc;
5624 33 : bdev_io->type = SPDK_BDEV_IO_TYPE_WRITE;
5625 33 : bdev_io->u.bdev.iovs = &bdev_io->iov;
5626 33 : bdev_io->u.bdev.iovs[0].iov_base = buf;
5627 33 : bdev_io->u.bdev.iovs[0].iov_len = num_blocks * bdev->blocklen;
5628 33 : bdev_io->u.bdev.iovcnt = 1;
5629 33 : bdev_io->u.bdev.md_buf = md_buf;
5630 33 : bdev_io->u.bdev.num_blocks = num_blocks;
5631 33 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5632 33 : bdev_io->u.bdev.memory_domain = NULL;
5633 33 : bdev_io->u.bdev.memory_domain_ctx = NULL;
5634 33 : bdev_io->u.bdev.accel_sequence = NULL;
5635 33 : bdev_io->u.bdev.dif_check_flags = bdev->dif_check_flags;
5636 33 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5637 :
5638 33 : bdev_io_submit(bdev_io);
5639 33 : return 0;
5640 : }
5641 :
5642 : int
5643 3 : spdk_bdev_write(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5644 : void *buf, uint64_t offset, uint64_t nbytes,
5645 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5646 : {
5647 3 : uint64_t offset_blocks, num_blocks;
5648 :
5649 3 : if (bdev_bytes_to_blocks(spdk_bdev_desc_get_bdev(desc), offset, &offset_blocks,
5650 : nbytes, &num_blocks) != 0) {
5651 0 : return -EINVAL;
5652 : }
5653 :
5654 3 : return spdk_bdev_write_blocks(desc, ch, buf, offset_blocks, num_blocks, cb, cb_arg);
5655 : }
5656 :
5657 : int
5658 24 : spdk_bdev_write_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5659 : void *buf, uint64_t offset_blocks, uint64_t num_blocks,
5660 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5661 : {
5662 24 : return bdev_write_blocks_with_md(desc, ch, buf, NULL, offset_blocks, num_blocks,
5663 : cb, cb_arg);
5664 : }
5665 :
5666 : int
5667 3 : spdk_bdev_write_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5668 : void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
5669 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5670 : {
5671 3 : struct iovec iov = {
5672 : .iov_base = buf,
5673 : };
5674 :
5675 3 : if (md_buf && !spdk_bdev_is_md_separate(spdk_bdev_desc_get_bdev(desc))) {
5676 0 : return -EINVAL;
5677 : }
5678 :
5679 3 : if (md_buf && !_is_buf_allocated(&iov)) {
5680 0 : return -EINVAL;
5681 : }
5682 :
5683 3 : return bdev_write_blocks_with_md(desc, ch, buf, md_buf, offset_blocks, num_blocks,
5684 : cb, cb_arg);
5685 : }
5686 :
5687 : static int
5688 64 : bdev_writev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5689 : struct iovec *iov, int iovcnt, void *md_buf,
5690 : uint64_t offset_blocks, uint64_t num_blocks,
5691 : struct spdk_memory_domain *domain, void *domain_ctx,
5692 : struct spdk_accel_sequence *seq, uint32_t dif_check_flags,
5693 : uint32_t nvme_cdw12_raw, uint32_t nvme_cdw13_raw,
5694 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5695 : {
5696 64 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5697 : struct spdk_bdev_io *bdev_io;
5698 64 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5699 :
5700 64 : if (spdk_unlikely(!desc->write)) {
5701 0 : return -EBADF;
5702 : }
5703 :
5704 64 : if (spdk_unlikely(!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks))) {
5705 0 : return -EINVAL;
5706 : }
5707 :
5708 64 : bdev_io = bdev_channel_get_io(channel);
5709 64 : if (spdk_unlikely(!bdev_io)) {
5710 2 : return -ENOMEM;
5711 : }
5712 :
5713 62 : bdev_io->internal.ch = channel;
5714 62 : bdev_io->internal.desc = desc;
5715 62 : bdev_io->type = SPDK_BDEV_IO_TYPE_WRITE;
5716 62 : bdev_io->u.bdev.iovs = iov;
5717 62 : bdev_io->u.bdev.iovcnt = iovcnt;
5718 62 : bdev_io->u.bdev.md_buf = md_buf;
5719 62 : bdev_io->u.bdev.num_blocks = num_blocks;
5720 62 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5721 62 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5722 62 : if (seq != NULL) {
5723 0 : bdev_io->internal.f.has_accel_sequence = true;
5724 0 : bdev_io->internal.accel_sequence = seq;
5725 : }
5726 :
5727 62 : if (domain != NULL) {
5728 2 : bdev_io->internal.f.has_memory_domain = true;
5729 2 : bdev_io->internal.memory_domain = domain;
5730 2 : bdev_io->internal.memory_domain_ctx = domain_ctx;
5731 : }
5732 :
5733 62 : bdev_io->u.bdev.memory_domain = domain;
5734 62 : bdev_io->u.bdev.memory_domain_ctx = domain_ctx;
5735 62 : bdev_io->u.bdev.accel_sequence = seq;
5736 62 : bdev_io->u.bdev.dif_check_flags = dif_check_flags;
5737 62 : bdev_io->u.bdev.nvme_cdw12.raw = nvme_cdw12_raw;
5738 62 : bdev_io->u.bdev.nvme_cdw13.raw = nvme_cdw13_raw;
5739 :
5740 62 : _bdev_io_submit_ext(desc, bdev_io);
5741 :
5742 62 : return 0;
5743 : }
5744 :
5745 : int
5746 3 : spdk_bdev_writev(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5747 : struct iovec *iov, int iovcnt,
5748 : uint64_t offset, uint64_t len,
5749 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5750 : {
5751 3 : uint64_t offset_blocks, num_blocks;
5752 :
5753 3 : if (bdev_bytes_to_blocks(spdk_bdev_desc_get_bdev(desc), offset, &offset_blocks,
5754 : len, &num_blocks) != 0) {
5755 0 : return -EINVAL;
5756 : }
5757 :
5758 3 : return spdk_bdev_writev_blocks(desc, ch, iov, iovcnt, offset_blocks, num_blocks, cb, cb_arg);
5759 : }
5760 :
5761 : int
5762 14 : spdk_bdev_writev_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5763 : struct iovec *iov, int iovcnt,
5764 : uint64_t offset_blocks, uint64_t num_blocks,
5765 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5766 : {
5767 14 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5768 :
5769 14 : return bdev_writev_blocks_with_md(desc, ch, iov, iovcnt, NULL, offset_blocks,
5770 : num_blocks, NULL, NULL, NULL, bdev->dif_check_flags, 0, 0,
5771 : cb, cb_arg);
5772 : }
5773 :
5774 : int
5775 1 : spdk_bdev_writev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5776 : struct iovec *iov, int iovcnt, void *md_buf,
5777 : uint64_t offset_blocks, uint64_t num_blocks,
5778 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5779 : {
5780 1 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5781 :
5782 1 : if (md_buf && !spdk_bdev_is_md_separate(bdev)) {
5783 0 : return -EINVAL;
5784 : }
5785 :
5786 1 : if (md_buf && !_is_buf_allocated(iov)) {
5787 0 : return -EINVAL;
5788 : }
5789 :
5790 1 : return bdev_writev_blocks_with_md(desc, ch, iov, iovcnt, md_buf, offset_blocks,
5791 : num_blocks, NULL, NULL, NULL, bdev->dif_check_flags, 0, 0,
5792 : cb, cb_arg);
5793 : }
5794 :
5795 : int
5796 8 : spdk_bdev_writev_blocks_ext(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5797 : struct iovec *iov, int iovcnt,
5798 : uint64_t offset_blocks, uint64_t num_blocks,
5799 : spdk_bdev_io_completion_cb cb, void *cb_arg,
5800 : struct spdk_bdev_ext_io_opts *opts)
5801 : {
5802 8 : struct spdk_memory_domain *domain = NULL;
5803 8 : struct spdk_accel_sequence *seq = NULL;
5804 8 : void *domain_ctx = NULL, *md = NULL;
5805 8 : uint32_t dif_check_flags = 0;
5806 8 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5807 8 : uint32_t nvme_cdw12_raw = 0;
5808 8 : uint32_t nvme_cdw13_raw = 0;
5809 :
5810 8 : if (opts) {
5811 7 : if (spdk_unlikely(!_bdev_io_check_opts(opts, iov))) {
5812 3 : return -EINVAL;
5813 : }
5814 4 : md = opts->metadata;
5815 4 : domain = bdev_get_ext_io_opt(opts, memory_domain, NULL);
5816 4 : domain_ctx = bdev_get_ext_io_opt(opts, memory_domain_ctx, NULL);
5817 4 : seq = bdev_get_ext_io_opt(opts, accel_sequence, NULL);
5818 4 : nvme_cdw12_raw = bdev_get_ext_io_opt(opts, nvme_cdw12.raw, 0);
5819 4 : nvme_cdw13_raw = bdev_get_ext_io_opt(opts, nvme_cdw13.raw, 0);
5820 4 : if (md) {
5821 4 : if (spdk_unlikely(!spdk_bdev_is_md_separate(bdev))) {
5822 0 : return -EINVAL;
5823 : }
5824 :
5825 4 : if (spdk_unlikely(!_is_buf_allocated(iov))) {
5826 0 : return -EINVAL;
5827 : }
5828 :
5829 4 : if (spdk_unlikely(seq != NULL)) {
5830 0 : return -EINVAL;
5831 : }
5832 : }
5833 : }
5834 :
5835 10 : dif_check_flags = bdev->dif_check_flags &
5836 5 : ~(bdev_get_ext_io_opt(opts, dif_check_flags_exclude_mask, 0));
5837 :
5838 5 : return bdev_writev_blocks_with_md(desc, ch, iov, iovcnt, md, offset_blocks, num_blocks,
5839 : domain, domain_ctx, seq, dif_check_flags,
5840 : nvme_cdw12_raw, nvme_cdw13_raw, cb, cb_arg);
5841 : }
5842 :
5843 : static void
5844 11 : bdev_compare_do_read_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
5845 : {
5846 11 : struct spdk_bdev_io *parent_io = cb_arg;
5847 11 : struct spdk_bdev *bdev = parent_io->bdev;
5848 11 : uint8_t *read_buf = bdev_io->u.bdev.iovs[0].iov_base;
5849 11 : int i, rc = 0;
5850 :
5851 11 : if (!success) {
5852 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
5853 0 : parent_io->internal.cb(parent_io, false, parent_io->internal.caller_ctx);
5854 0 : spdk_bdev_free_io(bdev_io);
5855 0 : return;
5856 : }
5857 :
5858 17 : for (i = 0; i < parent_io->u.bdev.iovcnt; i++) {
5859 11 : rc = memcmp(read_buf,
5860 11 : parent_io->u.bdev.iovs[i].iov_base,
5861 11 : parent_io->u.bdev.iovs[i].iov_len);
5862 11 : if (rc) {
5863 5 : break;
5864 : }
5865 6 : read_buf += parent_io->u.bdev.iovs[i].iov_len;
5866 : }
5867 :
5868 11 : if (rc == 0 && parent_io->u.bdev.md_buf && spdk_bdev_is_md_separate(bdev)) {
5869 2 : rc = memcmp(bdev_io->u.bdev.md_buf,
5870 2 : parent_io->u.bdev.md_buf,
5871 2 : spdk_bdev_get_md_size(bdev));
5872 : }
5873 :
5874 11 : spdk_bdev_free_io(bdev_io);
5875 :
5876 11 : if (rc == 0) {
5877 5 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
5878 5 : parent_io->internal.cb(parent_io, true, parent_io->internal.caller_ctx);
5879 : } else {
5880 6 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_MISCOMPARE;
5881 6 : parent_io->internal.cb(parent_io, false, parent_io->internal.caller_ctx);
5882 : }
5883 : }
5884 :
5885 : static void
5886 11 : bdev_compare_do_read(void *_bdev_io)
5887 : {
5888 11 : struct spdk_bdev_io *bdev_io = _bdev_io;
5889 : int rc;
5890 :
5891 11 : rc = spdk_bdev_read_blocks(bdev_io->internal.desc,
5892 11 : spdk_io_channel_from_ctx(bdev_io->internal.ch), NULL,
5893 : bdev_io->u.bdev.offset_blocks, bdev_io->u.bdev.num_blocks,
5894 : bdev_compare_do_read_done, bdev_io);
5895 :
5896 11 : if (rc == -ENOMEM) {
5897 0 : bdev_queue_io_wait_with_cb(bdev_io, bdev_compare_do_read);
5898 11 : } else if (rc != 0) {
5899 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
5900 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
5901 : }
5902 11 : }
5903 :
5904 : static int
5905 16 : bdev_comparev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5906 : struct iovec *iov, int iovcnt, void *md_buf,
5907 : uint64_t offset_blocks, uint64_t num_blocks,
5908 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5909 : {
5910 16 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5911 : struct spdk_bdev_io *bdev_io;
5912 16 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5913 :
5914 16 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
5915 0 : return -EINVAL;
5916 : }
5917 :
5918 16 : bdev_io = bdev_channel_get_io(channel);
5919 16 : if (!bdev_io) {
5920 0 : return -ENOMEM;
5921 : }
5922 :
5923 16 : bdev_io->internal.ch = channel;
5924 16 : bdev_io->internal.desc = desc;
5925 16 : bdev_io->type = SPDK_BDEV_IO_TYPE_COMPARE;
5926 16 : bdev_io->u.bdev.iovs = iov;
5927 16 : bdev_io->u.bdev.iovcnt = iovcnt;
5928 16 : bdev_io->u.bdev.md_buf = md_buf;
5929 16 : bdev_io->u.bdev.num_blocks = num_blocks;
5930 16 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5931 16 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5932 16 : bdev_io->u.bdev.memory_domain = NULL;
5933 16 : bdev_io->u.bdev.memory_domain_ctx = NULL;
5934 16 : bdev_io->u.bdev.accel_sequence = NULL;
5935 :
5936 16 : if (bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COMPARE)) {
5937 7 : bdev_io_submit(bdev_io);
5938 7 : return 0;
5939 : }
5940 :
5941 9 : bdev_compare_do_read(bdev_io);
5942 :
5943 9 : return 0;
5944 : }
5945 :
5946 : int
5947 10 : spdk_bdev_comparev_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5948 : struct iovec *iov, int iovcnt,
5949 : uint64_t offset_blocks, uint64_t num_blocks,
5950 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5951 : {
5952 10 : return bdev_comparev_blocks_with_md(desc, ch, iov, iovcnt, NULL, offset_blocks,
5953 : num_blocks, cb, cb_arg);
5954 : }
5955 :
5956 : int
5957 6 : spdk_bdev_comparev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5958 : struct iovec *iov, int iovcnt, void *md_buf,
5959 : uint64_t offset_blocks, uint64_t num_blocks,
5960 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5961 : {
5962 6 : if (md_buf && !spdk_bdev_is_md_separate(spdk_bdev_desc_get_bdev(desc))) {
5963 0 : return -EINVAL;
5964 : }
5965 :
5966 6 : if (md_buf && !_is_buf_allocated(iov)) {
5967 0 : return -EINVAL;
5968 : }
5969 :
5970 6 : return bdev_comparev_blocks_with_md(desc, ch, iov, iovcnt, md_buf, offset_blocks,
5971 : num_blocks, cb, cb_arg);
5972 : }
5973 :
5974 : static int
5975 4 : bdev_compare_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5976 : void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
5977 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5978 : {
5979 4 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5980 : struct spdk_bdev_io *bdev_io;
5981 4 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5982 :
5983 4 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
5984 0 : return -EINVAL;
5985 : }
5986 :
5987 4 : bdev_io = bdev_channel_get_io(channel);
5988 4 : if (!bdev_io) {
5989 0 : return -ENOMEM;
5990 : }
5991 :
5992 4 : bdev_io->internal.ch = channel;
5993 4 : bdev_io->internal.desc = desc;
5994 4 : bdev_io->type = SPDK_BDEV_IO_TYPE_COMPARE;
5995 4 : bdev_io->u.bdev.iovs = &bdev_io->iov;
5996 4 : bdev_io->u.bdev.iovs[0].iov_base = buf;
5997 4 : bdev_io->u.bdev.iovs[0].iov_len = num_blocks * bdev->blocklen;
5998 4 : bdev_io->u.bdev.iovcnt = 1;
5999 4 : bdev_io->u.bdev.md_buf = md_buf;
6000 4 : bdev_io->u.bdev.num_blocks = num_blocks;
6001 4 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6002 4 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6003 4 : bdev_io->u.bdev.memory_domain = NULL;
6004 4 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6005 4 : bdev_io->u.bdev.accel_sequence = NULL;
6006 :
6007 4 : if (bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COMPARE)) {
6008 2 : bdev_io_submit(bdev_io);
6009 2 : return 0;
6010 : }
6011 :
6012 2 : bdev_compare_do_read(bdev_io);
6013 :
6014 2 : return 0;
6015 : }
6016 :
6017 : int
6018 4 : spdk_bdev_compare_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6019 : void *buf, uint64_t offset_blocks, uint64_t num_blocks,
6020 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6021 : {
6022 4 : return bdev_compare_blocks_with_md(desc, ch, buf, NULL, offset_blocks, num_blocks,
6023 : cb, cb_arg);
6024 : }
6025 :
6026 : int
6027 0 : spdk_bdev_compare_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6028 : void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
6029 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6030 : {
6031 0 : struct iovec iov = {
6032 : .iov_base = buf,
6033 : };
6034 :
6035 0 : if (md_buf && !spdk_bdev_is_md_separate(spdk_bdev_desc_get_bdev(desc))) {
6036 0 : return -EINVAL;
6037 : }
6038 :
6039 0 : if (md_buf && !_is_buf_allocated(&iov)) {
6040 0 : return -EINVAL;
6041 : }
6042 :
6043 0 : return bdev_compare_blocks_with_md(desc, ch, buf, md_buf, offset_blocks, num_blocks,
6044 : cb, cb_arg);
6045 : }
6046 :
6047 : static void
6048 2 : bdev_comparev_and_writev_blocks_unlocked(struct lba_range *range, void *ctx, int unlock_status)
6049 : {
6050 2 : struct spdk_bdev_io *bdev_io = ctx;
6051 :
6052 2 : if (unlock_status) {
6053 0 : SPDK_ERRLOG("LBA range unlock failed\n");
6054 : }
6055 :
6056 2 : bdev_io->internal.cb(bdev_io, bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS ? true :
6057 : false, bdev_io->internal.caller_ctx);
6058 2 : }
6059 :
6060 : static void
6061 2 : bdev_comparev_and_writev_blocks_unlock(struct spdk_bdev_io *bdev_io, int status)
6062 : {
6063 2 : bdev_io->internal.status = status;
6064 :
6065 2 : bdev_unlock_lba_range(bdev_io->internal.desc, spdk_io_channel_from_ctx(bdev_io->internal.ch),
6066 : bdev_io->u.bdev.offset_blocks, bdev_io->u.bdev.num_blocks,
6067 : bdev_comparev_and_writev_blocks_unlocked, bdev_io);
6068 2 : }
6069 :
6070 : static void
6071 1 : bdev_compare_and_write_do_write_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
6072 : {
6073 1 : struct spdk_bdev_io *parent_io = cb_arg;
6074 :
6075 1 : if (!success) {
6076 0 : SPDK_ERRLOG("Compare and write operation failed\n");
6077 : }
6078 :
6079 1 : spdk_bdev_free_io(bdev_io);
6080 :
6081 1 : bdev_comparev_and_writev_blocks_unlock(parent_io,
6082 : success ? SPDK_BDEV_IO_STATUS_SUCCESS : SPDK_BDEV_IO_STATUS_FAILED);
6083 1 : }
6084 :
6085 : static void
6086 1 : bdev_compare_and_write_do_write(void *_bdev_io)
6087 : {
6088 1 : struct spdk_bdev_io *bdev_io = _bdev_io;
6089 : int rc;
6090 :
6091 1 : rc = spdk_bdev_writev_blocks(bdev_io->internal.desc,
6092 1 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
6093 : bdev_io->u.bdev.fused_iovs, bdev_io->u.bdev.fused_iovcnt,
6094 : bdev_io->u.bdev.offset_blocks, bdev_io->u.bdev.num_blocks,
6095 : bdev_compare_and_write_do_write_done, bdev_io);
6096 :
6097 :
6098 1 : if (rc == -ENOMEM) {
6099 0 : bdev_queue_io_wait_with_cb(bdev_io, bdev_compare_and_write_do_write);
6100 1 : } else if (rc != 0) {
6101 0 : bdev_comparev_and_writev_blocks_unlock(bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
6102 : }
6103 1 : }
6104 :
6105 : static void
6106 2 : bdev_compare_and_write_do_compare_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
6107 : {
6108 2 : struct spdk_bdev_io *parent_io = cb_arg;
6109 :
6110 2 : spdk_bdev_free_io(bdev_io);
6111 :
6112 2 : if (!success) {
6113 1 : bdev_comparev_and_writev_blocks_unlock(parent_io, SPDK_BDEV_IO_STATUS_MISCOMPARE);
6114 1 : return;
6115 : }
6116 :
6117 1 : bdev_compare_and_write_do_write(parent_io);
6118 : }
6119 :
6120 : static void
6121 2 : bdev_compare_and_write_do_compare(void *_bdev_io)
6122 : {
6123 2 : struct spdk_bdev_io *bdev_io = _bdev_io;
6124 : int rc;
6125 :
6126 2 : rc = spdk_bdev_comparev_blocks(bdev_io->internal.desc,
6127 2 : spdk_io_channel_from_ctx(bdev_io->internal.ch), bdev_io->u.bdev.iovs,
6128 : bdev_io->u.bdev.iovcnt, bdev_io->u.bdev.offset_blocks, bdev_io->u.bdev.num_blocks,
6129 : bdev_compare_and_write_do_compare_done, bdev_io);
6130 :
6131 2 : if (rc == -ENOMEM) {
6132 0 : bdev_queue_io_wait_with_cb(bdev_io, bdev_compare_and_write_do_compare);
6133 2 : } else if (rc != 0) {
6134 0 : bdev_comparev_and_writev_blocks_unlock(bdev_io, SPDK_BDEV_IO_STATUS_FIRST_FUSED_FAILED);
6135 : }
6136 2 : }
6137 :
6138 : static void
6139 2 : bdev_comparev_and_writev_blocks_locked(struct lba_range *range, void *ctx, int status)
6140 : {
6141 2 : struct spdk_bdev_io *bdev_io = ctx;
6142 :
6143 2 : if (status) {
6144 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FIRST_FUSED_FAILED;
6145 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
6146 0 : return;
6147 : }
6148 :
6149 2 : bdev_compare_and_write_do_compare(bdev_io);
6150 : }
6151 :
6152 : int
6153 2 : spdk_bdev_comparev_and_writev_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6154 : struct iovec *compare_iov, int compare_iovcnt,
6155 : struct iovec *write_iov, int write_iovcnt,
6156 : uint64_t offset_blocks, uint64_t num_blocks,
6157 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6158 : {
6159 2 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6160 : struct spdk_bdev_io *bdev_io;
6161 2 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6162 :
6163 2 : if (!desc->write) {
6164 0 : return -EBADF;
6165 : }
6166 :
6167 2 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6168 0 : return -EINVAL;
6169 : }
6170 :
6171 2 : if (num_blocks > bdev->acwu) {
6172 0 : return -EINVAL;
6173 : }
6174 :
6175 2 : bdev_io = bdev_channel_get_io(channel);
6176 2 : if (!bdev_io) {
6177 0 : return -ENOMEM;
6178 : }
6179 :
6180 2 : bdev_io->internal.ch = channel;
6181 2 : bdev_io->internal.desc = desc;
6182 2 : bdev_io->type = SPDK_BDEV_IO_TYPE_COMPARE_AND_WRITE;
6183 2 : bdev_io->u.bdev.iovs = compare_iov;
6184 2 : bdev_io->u.bdev.iovcnt = compare_iovcnt;
6185 2 : bdev_io->u.bdev.fused_iovs = write_iov;
6186 2 : bdev_io->u.bdev.fused_iovcnt = write_iovcnt;
6187 2 : bdev_io->u.bdev.md_buf = NULL;
6188 2 : bdev_io->u.bdev.num_blocks = num_blocks;
6189 2 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6190 2 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6191 2 : bdev_io->u.bdev.memory_domain = NULL;
6192 2 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6193 2 : bdev_io->u.bdev.accel_sequence = NULL;
6194 :
6195 2 : if (bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COMPARE_AND_WRITE)) {
6196 0 : bdev_io_submit(bdev_io);
6197 0 : return 0;
6198 : }
6199 :
6200 2 : return bdev_lock_lba_range(desc, ch, offset_blocks, num_blocks,
6201 : bdev_comparev_and_writev_blocks_locked, bdev_io);
6202 : }
6203 :
6204 : int
6205 2 : spdk_bdev_zcopy_start(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6206 : struct iovec *iov, int iovcnt,
6207 : uint64_t offset_blocks, uint64_t num_blocks,
6208 : bool populate,
6209 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6210 : {
6211 2 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6212 : struct spdk_bdev_io *bdev_io;
6213 2 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6214 :
6215 2 : if (!desc->write) {
6216 0 : return -EBADF;
6217 : }
6218 :
6219 2 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6220 0 : return -EINVAL;
6221 : }
6222 :
6223 2 : if (!spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_ZCOPY)) {
6224 0 : return -ENOTSUP;
6225 : }
6226 :
6227 2 : bdev_io = bdev_channel_get_io(channel);
6228 2 : if (!bdev_io) {
6229 0 : return -ENOMEM;
6230 : }
6231 :
6232 2 : bdev_io->internal.ch = channel;
6233 2 : bdev_io->internal.desc = desc;
6234 2 : bdev_io->type = SPDK_BDEV_IO_TYPE_ZCOPY;
6235 2 : bdev_io->u.bdev.num_blocks = num_blocks;
6236 2 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6237 2 : bdev_io->u.bdev.iovs = iov;
6238 2 : bdev_io->u.bdev.iovcnt = iovcnt;
6239 2 : bdev_io->u.bdev.md_buf = NULL;
6240 2 : bdev_io->u.bdev.zcopy.populate = populate ? 1 : 0;
6241 2 : bdev_io->u.bdev.zcopy.commit = 0;
6242 2 : bdev_io->u.bdev.zcopy.start = 1;
6243 2 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6244 2 : bdev_io->u.bdev.memory_domain = NULL;
6245 2 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6246 2 : bdev_io->u.bdev.accel_sequence = NULL;
6247 :
6248 2 : bdev_io_submit(bdev_io);
6249 :
6250 2 : return 0;
6251 : }
6252 :
6253 : int
6254 2 : spdk_bdev_zcopy_end(struct spdk_bdev_io *bdev_io, bool commit,
6255 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6256 : {
6257 2 : if (bdev_io->type != SPDK_BDEV_IO_TYPE_ZCOPY) {
6258 0 : return -EINVAL;
6259 : }
6260 :
6261 2 : bdev_io->u.bdev.zcopy.commit = commit ? 1 : 0;
6262 2 : bdev_io->u.bdev.zcopy.start = 0;
6263 2 : bdev_io->internal.caller_ctx = cb_arg;
6264 2 : bdev_io->internal.cb = cb;
6265 2 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_PENDING;
6266 :
6267 2 : bdev_io_submit(bdev_io);
6268 :
6269 2 : return 0;
6270 : }
6271 :
6272 : int
6273 0 : spdk_bdev_write_zeroes(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6274 : uint64_t offset, uint64_t len,
6275 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6276 : {
6277 0 : uint64_t offset_blocks, num_blocks;
6278 :
6279 0 : if (bdev_bytes_to_blocks(spdk_bdev_desc_get_bdev(desc), offset, &offset_blocks,
6280 : len, &num_blocks) != 0) {
6281 0 : return -EINVAL;
6282 : }
6283 :
6284 0 : return spdk_bdev_write_zeroes_blocks(desc, ch, offset_blocks, num_blocks, cb, cb_arg);
6285 : }
6286 :
6287 : int
6288 33 : spdk_bdev_write_zeroes_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6289 : uint64_t offset_blocks, uint64_t num_blocks,
6290 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6291 : {
6292 33 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6293 : struct spdk_bdev_io *bdev_io;
6294 33 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6295 :
6296 33 : if (!desc->write) {
6297 0 : return -EBADF;
6298 : }
6299 :
6300 33 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6301 0 : return -EINVAL;
6302 : }
6303 :
6304 33 : if (!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE_ZEROES) &&
6305 10 : !bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE)) {
6306 1 : return -ENOTSUP;
6307 : }
6308 :
6309 32 : bdev_io = bdev_channel_get_io(channel);
6310 :
6311 32 : if (!bdev_io) {
6312 0 : return -ENOMEM;
6313 : }
6314 :
6315 32 : bdev_io->type = SPDK_BDEV_IO_TYPE_WRITE_ZEROES;
6316 32 : bdev_io->internal.ch = channel;
6317 32 : bdev_io->internal.desc = desc;
6318 32 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6319 32 : bdev_io->u.bdev.num_blocks = num_blocks;
6320 32 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6321 32 : bdev_io->u.bdev.memory_domain = NULL;
6322 32 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6323 32 : bdev_io->u.bdev.accel_sequence = NULL;
6324 :
6325 : /* If the write_zeroes size is large and should be split, use the generic split
6326 : * logic regardless of whether SPDK_BDEV_IO_TYPE_WRITE_ZEREOS is supported or not.
6327 : *
6328 : * Then, send the write_zeroes request if SPDK_BDEV_IO_TYPE_WRITE_ZEROES is supported
6329 : * or emulate it using regular write request otherwise.
6330 : */
6331 32 : if (bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE_ZEROES) ||
6332 : bdev_io->internal.f.split) {
6333 26 : bdev_io_submit(bdev_io);
6334 26 : return 0;
6335 : }
6336 :
6337 6 : assert(_bdev_get_block_size_with_md(bdev) <= ZERO_BUFFER_SIZE);
6338 :
6339 6 : return bdev_write_zero_buffer(bdev_io);
6340 : }
6341 :
6342 : int
6343 0 : spdk_bdev_unmap(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6344 : uint64_t offset, uint64_t nbytes,
6345 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6346 : {
6347 0 : uint64_t offset_blocks, num_blocks;
6348 :
6349 0 : if (bdev_bytes_to_blocks(spdk_bdev_desc_get_bdev(desc), offset, &offset_blocks,
6350 : nbytes, &num_blocks) != 0) {
6351 0 : return -EINVAL;
6352 : }
6353 :
6354 0 : return spdk_bdev_unmap_blocks(desc, ch, offset_blocks, num_blocks, cb, cb_arg);
6355 : }
6356 :
6357 : static void
6358 0 : bdev_io_complete_cb(void *ctx)
6359 : {
6360 0 : struct spdk_bdev_io *bdev_io = ctx;
6361 :
6362 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
6363 0 : bdev_io->internal.cb(bdev_io, true, bdev_io->internal.caller_ctx);
6364 0 : }
6365 :
6366 : int
6367 22 : spdk_bdev_unmap_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6368 : uint64_t offset_blocks, uint64_t num_blocks,
6369 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6370 : {
6371 22 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6372 : struct spdk_bdev_io *bdev_io;
6373 22 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6374 :
6375 22 : if (!desc->write) {
6376 0 : return -EBADF;
6377 : }
6378 :
6379 22 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6380 0 : return -EINVAL;
6381 : }
6382 :
6383 22 : bdev_io = bdev_channel_get_io(channel);
6384 22 : if (!bdev_io) {
6385 0 : return -ENOMEM;
6386 : }
6387 :
6388 22 : bdev_io->internal.ch = channel;
6389 22 : bdev_io->internal.desc = desc;
6390 22 : bdev_io->type = SPDK_BDEV_IO_TYPE_UNMAP;
6391 :
6392 22 : bdev_io->u.bdev.iovs = &bdev_io->iov;
6393 22 : bdev_io->u.bdev.iovs[0].iov_base = NULL;
6394 22 : bdev_io->u.bdev.iovs[0].iov_len = 0;
6395 22 : bdev_io->u.bdev.iovcnt = 1;
6396 :
6397 22 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6398 22 : bdev_io->u.bdev.num_blocks = num_blocks;
6399 22 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6400 22 : bdev_io->u.bdev.memory_domain = NULL;
6401 22 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6402 22 : bdev_io->u.bdev.accel_sequence = NULL;
6403 :
6404 22 : if (num_blocks == 0) {
6405 0 : spdk_thread_send_msg(spdk_get_thread(), bdev_io_complete_cb, bdev_io);
6406 0 : return 0;
6407 : }
6408 :
6409 22 : bdev_io_submit(bdev_io);
6410 22 : return 0;
6411 : }
6412 :
6413 : int
6414 0 : spdk_bdev_flush(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6415 : uint64_t offset, uint64_t length,
6416 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6417 : {
6418 0 : uint64_t offset_blocks, num_blocks;
6419 :
6420 0 : if (bdev_bytes_to_blocks(spdk_bdev_desc_get_bdev(desc), offset, &offset_blocks,
6421 : length, &num_blocks) != 0) {
6422 0 : return -EINVAL;
6423 : }
6424 :
6425 0 : return spdk_bdev_flush_blocks(desc, ch, offset_blocks, num_blocks, cb, cb_arg);
6426 : }
6427 :
6428 : int
6429 2 : spdk_bdev_flush_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6430 : uint64_t offset_blocks, uint64_t num_blocks,
6431 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6432 : {
6433 2 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6434 : struct spdk_bdev_io *bdev_io;
6435 2 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6436 :
6437 2 : if (!desc->write) {
6438 0 : return -EBADF;
6439 : }
6440 :
6441 2 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6442 0 : return -EINVAL;
6443 : }
6444 :
6445 2 : bdev_io = bdev_channel_get_io(channel);
6446 2 : if (!bdev_io) {
6447 0 : return -ENOMEM;
6448 : }
6449 :
6450 2 : bdev_io->internal.ch = channel;
6451 2 : bdev_io->internal.desc = desc;
6452 2 : bdev_io->type = SPDK_BDEV_IO_TYPE_FLUSH;
6453 2 : bdev_io->u.bdev.iovs = NULL;
6454 2 : bdev_io->u.bdev.iovcnt = 0;
6455 2 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6456 2 : bdev_io->u.bdev.num_blocks = num_blocks;
6457 2 : bdev_io->u.bdev.memory_domain = NULL;
6458 2 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6459 2 : bdev_io->u.bdev.accel_sequence = NULL;
6460 2 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6461 :
6462 2 : bdev_io_submit(bdev_io);
6463 2 : return 0;
6464 : }
6465 :
6466 : static int bdev_reset_poll_for_outstanding_io(void *ctx);
6467 :
6468 : static void
6469 13 : bdev_reset_check_outstanding_io_done(struct spdk_bdev *bdev, void *_ctx, int status)
6470 : {
6471 13 : struct spdk_bdev_channel *ch = _ctx;
6472 : struct spdk_bdev_io *bdev_io;
6473 :
6474 13 : bdev_io = TAILQ_FIRST(&ch->queued_resets);
6475 :
6476 13 : if (status == -EBUSY) {
6477 9 : if (spdk_get_ticks() < bdev_io->u.reset.wait_poller.stop_time_tsc) {
6478 8 : bdev_io->u.reset.wait_poller.poller = SPDK_POLLER_REGISTER(bdev_reset_poll_for_outstanding_io,
6479 : ch, BDEV_RESET_CHECK_OUTSTANDING_IO_PERIOD);
6480 : } else {
6481 1 : TAILQ_REMOVE(&ch->queued_resets, bdev_io, internal.link);
6482 :
6483 1 : if (TAILQ_EMPTY(&ch->io_memory_domain) && TAILQ_EMPTY(&ch->io_accel_exec)) {
6484 : /* If outstanding IOs are still present and reset_io_drain_timeout
6485 : * seconds passed, start the reset. */
6486 1 : bdev_io_submit_reset(bdev_io);
6487 : } else {
6488 : /* We still have in progress memory domain pull/push or we're
6489 : * executing accel sequence. Since we cannot abort either of those
6490 : * operations, fail the reset request. */
6491 0 : spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
6492 : }
6493 : }
6494 : } else {
6495 4 : TAILQ_REMOVE(&ch->queued_resets, bdev_io, internal.link);
6496 4 : SPDK_DEBUGLOG(bdev,
6497 : "Skipping reset for underlying device of bdev: %s - no outstanding I/O.\n",
6498 : ch->bdev->name);
6499 : /* Mark the completion status as a SUCCESS and complete the reset. */
6500 4 : spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_SUCCESS);
6501 : }
6502 13 : }
6503 :
6504 : static void
6505 13 : bdev_reset_check_outstanding_io(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
6506 : struct spdk_io_channel *io_ch, void *_ctx)
6507 : {
6508 13 : struct spdk_bdev_channel *cur_ch = __io_ch_to_bdev_ch(io_ch);
6509 13 : int status = 0;
6510 :
6511 13 : if (cur_ch->io_outstanding > 0 ||
6512 4 : !TAILQ_EMPTY(&cur_ch->io_memory_domain) ||
6513 4 : !TAILQ_EMPTY(&cur_ch->io_accel_exec)) {
6514 : /* If a channel has outstanding IO, set status to -EBUSY code. This will stop
6515 : * further iteration over the rest of the channels and pass non-zero status
6516 : * to the callback function. */
6517 9 : status = -EBUSY;
6518 : }
6519 13 : spdk_bdev_for_each_channel_continue(i, status);
6520 13 : }
6521 :
6522 : static int
6523 8 : bdev_reset_poll_for_outstanding_io(void *ctx)
6524 : {
6525 8 : struct spdk_bdev_channel *ch = ctx;
6526 : struct spdk_bdev_io *bdev_io;
6527 :
6528 8 : bdev_io = TAILQ_FIRST(&ch->queued_resets);
6529 :
6530 8 : spdk_poller_unregister(&bdev_io->u.reset.wait_poller.poller);
6531 8 : spdk_bdev_for_each_channel(ch->bdev, bdev_reset_check_outstanding_io, ch,
6532 : bdev_reset_check_outstanding_io_done);
6533 :
6534 8 : return SPDK_POLLER_BUSY;
6535 : }
6536 :
6537 : static void
6538 15 : bdev_reset_freeze_channel_done(struct spdk_bdev *bdev, void *_ctx, int status)
6539 : {
6540 15 : struct spdk_bdev_channel *ch = _ctx;
6541 : struct spdk_bdev_io *bdev_io;
6542 :
6543 15 : bdev_io = TAILQ_FIRST(&ch->queued_resets);
6544 :
6545 15 : if (bdev->reset_io_drain_timeout == 0) {
6546 10 : TAILQ_REMOVE(&ch->queued_resets, bdev_io, internal.link);
6547 :
6548 10 : bdev_io_submit_reset(bdev_io);
6549 10 : return;
6550 : }
6551 :
6552 5 : bdev_io->u.reset.wait_poller.stop_time_tsc = spdk_get_ticks() +
6553 5 : (ch->bdev->reset_io_drain_timeout * spdk_get_ticks_hz());
6554 :
6555 : /* In case bdev->reset_io_drain_timeout is not equal to zero,
6556 : * submit the reset to the underlying module only if outstanding I/O
6557 : * remain after reset_io_drain_timeout seconds have passed. */
6558 5 : spdk_bdev_for_each_channel(ch->bdev, bdev_reset_check_outstanding_io, ch,
6559 : bdev_reset_check_outstanding_io_done);
6560 : }
6561 :
6562 : static void
6563 18 : bdev_reset_freeze_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
6564 : struct spdk_io_channel *ch, void *_ctx)
6565 : {
6566 : struct spdk_bdev_channel *channel;
6567 : struct spdk_bdev_mgmt_channel *mgmt_channel;
6568 : struct spdk_bdev_shared_resource *shared_resource;
6569 18 : bdev_io_tailq_t tmp_queued;
6570 :
6571 18 : TAILQ_INIT(&tmp_queued);
6572 :
6573 18 : channel = __io_ch_to_bdev_ch(ch);
6574 18 : shared_resource = channel->shared_resource;
6575 18 : mgmt_channel = shared_resource->mgmt_ch;
6576 :
6577 18 : channel->flags |= BDEV_CH_RESET_IN_PROGRESS;
6578 :
6579 18 : if ((channel->flags & BDEV_CH_QOS_ENABLED) != 0) {
6580 2 : TAILQ_SWAP(&channel->qos_queued_io, &tmp_queued, spdk_bdev_io, internal.link);
6581 : }
6582 :
6583 18 : bdev_abort_all_queued_io(&shared_resource->nomem_io, channel);
6584 18 : bdev_abort_all_buf_io(mgmt_channel, channel);
6585 18 : bdev_abort_all_queued_io(&tmp_queued, channel);
6586 :
6587 18 : spdk_bdev_for_each_channel_continue(i, 0);
6588 18 : }
6589 :
6590 : static void
6591 15 : bdev_start_reset(void *ctx)
6592 : {
6593 15 : struct spdk_bdev_channel *ch = ctx;
6594 :
6595 15 : spdk_bdev_for_each_channel(ch->bdev, bdev_reset_freeze_channel, ch,
6596 : bdev_reset_freeze_channel_done);
6597 15 : }
6598 :
6599 : static void
6600 16 : bdev_channel_start_reset(struct spdk_bdev_channel *ch)
6601 : {
6602 16 : struct spdk_bdev *bdev = ch->bdev;
6603 :
6604 16 : assert(!TAILQ_EMPTY(&ch->queued_resets));
6605 :
6606 16 : spdk_spin_lock(&bdev->internal.spinlock);
6607 16 : if (bdev->internal.reset_in_progress == NULL) {
6608 15 : bdev->internal.reset_in_progress = TAILQ_FIRST(&ch->queued_resets);
6609 : /*
6610 : * Take a channel reference for the target bdev for the life of this
6611 : * reset. This guards against the channel getting destroyed while
6612 : * spdk_bdev_for_each_channel() calls related to this reset IO are in
6613 : * progress. We will release the reference when this reset is
6614 : * completed.
6615 : */
6616 15 : bdev->internal.reset_in_progress->u.reset.ch_ref = spdk_get_io_channel(__bdev_to_io_dev(bdev));
6617 15 : bdev_start_reset(ch);
6618 : }
6619 16 : spdk_spin_unlock(&bdev->internal.spinlock);
6620 16 : }
6621 :
6622 : int
6623 16 : spdk_bdev_reset(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6624 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6625 : {
6626 16 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6627 : struct spdk_bdev_io *bdev_io;
6628 16 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6629 :
6630 16 : bdev_io = bdev_channel_get_io(channel);
6631 16 : if (!bdev_io) {
6632 0 : return -ENOMEM;
6633 : }
6634 :
6635 16 : bdev_io->internal.ch = channel;
6636 16 : bdev_io->internal.desc = desc;
6637 16 : bdev_io->internal.submit_tsc = spdk_get_ticks();
6638 16 : bdev_io->type = SPDK_BDEV_IO_TYPE_RESET;
6639 16 : bdev_io->u.reset.ch_ref = NULL;
6640 16 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6641 :
6642 16 : spdk_spin_lock(&bdev->internal.spinlock);
6643 16 : TAILQ_INSERT_TAIL(&channel->queued_resets, bdev_io, internal.link);
6644 16 : spdk_spin_unlock(&bdev->internal.spinlock);
6645 :
6646 16 : bdev_ch_add_to_io_submitted(bdev_io);
6647 :
6648 16 : bdev_channel_start_reset(channel);
6649 :
6650 16 : return 0;
6651 : }
6652 :
6653 : void
6654 0 : spdk_bdev_get_io_stat(struct spdk_bdev *bdev, struct spdk_io_channel *ch,
6655 : struct spdk_bdev_io_stat *stat)
6656 : {
6657 0 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6658 :
6659 0 : bdev_get_io_stat(stat, channel->stat);
6660 0 : }
6661 :
6662 : static void
6663 1 : bdev_get_device_stat_done(struct spdk_bdev *bdev, void *_ctx, int status)
6664 : {
6665 1 : struct spdk_bdev_iostat_ctx *bdev_iostat_ctx = _ctx;
6666 :
6667 1 : bdev_iostat_ctx->cb(bdev, bdev_iostat_ctx->stat,
6668 : bdev_iostat_ctx->cb_arg, 0);
6669 1 : free(bdev_iostat_ctx);
6670 1 : }
6671 :
6672 : static void
6673 0 : bdev_get_each_channel_stat(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
6674 : struct spdk_io_channel *ch, void *_ctx)
6675 : {
6676 0 : struct spdk_bdev_iostat_ctx *bdev_iostat_ctx = _ctx;
6677 0 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6678 :
6679 0 : spdk_bdev_add_io_stat(bdev_iostat_ctx->stat, channel->stat);
6680 0 : spdk_bdev_for_each_channel_continue(i, 0);
6681 0 : }
6682 :
6683 : void
6684 1 : spdk_bdev_get_device_stat(struct spdk_bdev *bdev, struct spdk_bdev_io_stat *stat,
6685 : spdk_bdev_get_device_stat_cb cb, void *cb_arg)
6686 : {
6687 : struct spdk_bdev_iostat_ctx *bdev_iostat_ctx;
6688 :
6689 1 : assert(bdev != NULL);
6690 1 : assert(stat != NULL);
6691 1 : assert(cb != NULL);
6692 :
6693 1 : bdev_iostat_ctx = calloc(1, sizeof(struct spdk_bdev_iostat_ctx));
6694 1 : if (bdev_iostat_ctx == NULL) {
6695 0 : SPDK_ERRLOG("Unable to allocate memory for spdk_bdev_iostat_ctx\n");
6696 0 : cb(bdev, stat, cb_arg, -ENOMEM);
6697 0 : return;
6698 : }
6699 :
6700 1 : bdev_iostat_ctx->stat = stat;
6701 1 : bdev_iostat_ctx->cb = cb;
6702 1 : bdev_iostat_ctx->cb_arg = cb_arg;
6703 :
6704 : /* Start with the statistics from previously deleted channels. */
6705 1 : spdk_spin_lock(&bdev->internal.spinlock);
6706 1 : bdev_get_io_stat(bdev_iostat_ctx->stat, bdev->internal.stat);
6707 1 : spdk_spin_unlock(&bdev->internal.spinlock);
6708 :
6709 : /* Then iterate and add the statistics from each existing channel. */
6710 1 : spdk_bdev_for_each_channel(bdev, bdev_get_each_channel_stat, bdev_iostat_ctx,
6711 : bdev_get_device_stat_done);
6712 : }
6713 :
6714 : struct bdev_iostat_reset_ctx {
6715 : enum spdk_bdev_reset_stat_mode mode;
6716 : bdev_reset_device_stat_cb cb;
6717 : void *cb_arg;
6718 : };
6719 :
6720 : static void
6721 0 : bdev_reset_device_stat_done(struct spdk_bdev *bdev, void *_ctx, int status)
6722 : {
6723 0 : struct bdev_iostat_reset_ctx *ctx = _ctx;
6724 :
6725 0 : ctx->cb(bdev, ctx->cb_arg, 0);
6726 :
6727 0 : free(ctx);
6728 0 : }
6729 :
6730 : static void
6731 0 : bdev_reset_each_channel_stat(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
6732 : struct spdk_io_channel *ch, void *_ctx)
6733 : {
6734 0 : struct bdev_iostat_reset_ctx *ctx = _ctx;
6735 0 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6736 :
6737 0 : spdk_bdev_reset_io_stat(channel->stat, ctx->mode);
6738 :
6739 0 : spdk_bdev_for_each_channel_continue(i, 0);
6740 0 : }
6741 :
6742 : void
6743 0 : bdev_reset_device_stat(struct spdk_bdev *bdev, enum spdk_bdev_reset_stat_mode mode,
6744 : bdev_reset_device_stat_cb cb, void *cb_arg)
6745 : {
6746 : struct bdev_iostat_reset_ctx *ctx;
6747 :
6748 0 : assert(bdev != NULL);
6749 0 : assert(cb != NULL);
6750 :
6751 0 : ctx = calloc(1, sizeof(*ctx));
6752 0 : if (ctx == NULL) {
6753 0 : SPDK_ERRLOG("Unable to allocate bdev_iostat_reset_ctx.\n");
6754 0 : cb(bdev, cb_arg, -ENOMEM);
6755 0 : return;
6756 : }
6757 :
6758 0 : ctx->mode = mode;
6759 0 : ctx->cb = cb;
6760 0 : ctx->cb_arg = cb_arg;
6761 :
6762 0 : spdk_spin_lock(&bdev->internal.spinlock);
6763 0 : spdk_bdev_reset_io_stat(bdev->internal.stat, mode);
6764 0 : spdk_spin_unlock(&bdev->internal.spinlock);
6765 :
6766 0 : spdk_bdev_for_each_channel(bdev,
6767 : bdev_reset_each_channel_stat,
6768 : ctx,
6769 : bdev_reset_device_stat_done);
6770 : }
6771 :
6772 : int
6773 1 : spdk_bdev_nvme_admin_passthru(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6774 : const struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes,
6775 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6776 : {
6777 1 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6778 : struct spdk_bdev_io *bdev_io;
6779 1 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6780 :
6781 1 : if (!desc->write) {
6782 0 : return -EBADF;
6783 : }
6784 :
6785 1 : if (spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_ADMIN))) {
6786 1 : return -ENOTSUP;
6787 : }
6788 :
6789 0 : bdev_io = bdev_channel_get_io(channel);
6790 0 : if (!bdev_io) {
6791 0 : return -ENOMEM;
6792 : }
6793 :
6794 0 : bdev_io->internal.ch = channel;
6795 0 : bdev_io->internal.desc = desc;
6796 0 : bdev_io->type = SPDK_BDEV_IO_TYPE_NVME_ADMIN;
6797 0 : bdev_io->u.nvme_passthru.cmd = *cmd;
6798 0 : bdev_io->u.nvme_passthru.buf = buf;
6799 0 : bdev_io->u.nvme_passthru.nbytes = nbytes;
6800 0 : bdev_io->u.nvme_passthru.md_buf = NULL;
6801 0 : bdev_io->u.nvme_passthru.md_len = 0;
6802 :
6803 0 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6804 :
6805 0 : bdev_io_submit(bdev_io);
6806 0 : return 0;
6807 : }
6808 :
6809 : int
6810 1 : spdk_bdev_nvme_io_passthru(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6811 : const struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes,
6812 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6813 : {
6814 1 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6815 : struct spdk_bdev_io *bdev_io;
6816 1 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6817 :
6818 1 : if (!desc->write) {
6819 : /*
6820 : * Do not try to parse the NVMe command - we could maybe use bits in the opcode
6821 : * to easily determine if the command is a read or write, but for now just
6822 : * do not allow io_passthru with a read-only descriptor.
6823 : */
6824 0 : return -EBADF;
6825 : }
6826 :
6827 1 : if (spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_IO))) {
6828 1 : return -ENOTSUP;
6829 : }
6830 :
6831 0 : bdev_io = bdev_channel_get_io(channel);
6832 0 : if (!bdev_io) {
6833 0 : return -ENOMEM;
6834 : }
6835 :
6836 0 : bdev_io->internal.ch = channel;
6837 0 : bdev_io->internal.desc = desc;
6838 0 : bdev_io->type = SPDK_BDEV_IO_TYPE_NVME_IO;
6839 0 : bdev_io->u.nvme_passthru.cmd = *cmd;
6840 0 : bdev_io->u.nvme_passthru.buf = buf;
6841 0 : bdev_io->u.nvme_passthru.nbytes = nbytes;
6842 0 : bdev_io->u.nvme_passthru.md_buf = NULL;
6843 0 : bdev_io->u.nvme_passthru.md_len = 0;
6844 :
6845 0 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6846 :
6847 0 : bdev_io_submit(bdev_io);
6848 0 : return 0;
6849 : }
6850 :
6851 : int
6852 1 : spdk_bdev_nvme_io_passthru_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6853 : const struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes, void *md_buf, size_t md_len,
6854 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6855 : {
6856 1 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6857 : struct spdk_bdev_io *bdev_io;
6858 1 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6859 :
6860 1 : if (!desc->write) {
6861 : /*
6862 : * Do not try to parse the NVMe command - we could maybe use bits in the opcode
6863 : * to easily determine if the command is a read or write, but for now just
6864 : * do not allow io_passthru with a read-only descriptor.
6865 : */
6866 0 : return -EBADF;
6867 : }
6868 :
6869 1 : if (spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_IO_MD))) {
6870 1 : return -ENOTSUP;
6871 : }
6872 :
6873 0 : bdev_io = bdev_channel_get_io(channel);
6874 0 : if (!bdev_io) {
6875 0 : return -ENOMEM;
6876 : }
6877 :
6878 0 : bdev_io->internal.ch = channel;
6879 0 : bdev_io->internal.desc = desc;
6880 0 : bdev_io->type = SPDK_BDEV_IO_TYPE_NVME_IO_MD;
6881 0 : bdev_io->u.nvme_passthru.cmd = *cmd;
6882 0 : bdev_io->u.nvme_passthru.buf = buf;
6883 0 : bdev_io->u.nvme_passthru.nbytes = nbytes;
6884 0 : bdev_io->u.nvme_passthru.md_buf = md_buf;
6885 0 : bdev_io->u.nvme_passthru.md_len = md_len;
6886 :
6887 0 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6888 :
6889 0 : bdev_io_submit(bdev_io);
6890 0 : return 0;
6891 : }
6892 :
6893 : int
6894 0 : spdk_bdev_nvme_iov_passthru_md(struct spdk_bdev_desc *desc,
6895 : struct spdk_io_channel *ch,
6896 : const struct spdk_nvme_cmd *cmd,
6897 : struct iovec *iov, int iovcnt, size_t nbytes,
6898 : void *md_buf, size_t md_len,
6899 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6900 : {
6901 0 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6902 : struct spdk_bdev_io *bdev_io;
6903 0 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6904 :
6905 0 : if (!desc->write) {
6906 : /*
6907 : * Do not try to parse the NVMe command - we could maybe use bits in the opcode
6908 : * to easily determine if the command is a read or write, but for now just
6909 : * do not allow io_passthru with a read-only descriptor.
6910 : */
6911 0 : return -EBADF;
6912 : }
6913 :
6914 0 : if (md_buf && spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_IO_MD))) {
6915 0 : return -ENOTSUP;
6916 0 : } else if (spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_IO))) {
6917 0 : return -ENOTSUP;
6918 : }
6919 :
6920 0 : bdev_io = bdev_channel_get_io(channel);
6921 0 : if (!bdev_io) {
6922 0 : return -ENOMEM;
6923 : }
6924 :
6925 0 : bdev_io->internal.ch = channel;
6926 0 : bdev_io->internal.desc = desc;
6927 0 : bdev_io->type = SPDK_BDEV_IO_TYPE_NVME_IOV_MD;
6928 0 : bdev_io->u.nvme_passthru.cmd = *cmd;
6929 0 : bdev_io->u.nvme_passthru.iovs = iov;
6930 0 : bdev_io->u.nvme_passthru.iovcnt = iovcnt;
6931 0 : bdev_io->u.nvme_passthru.nbytes = nbytes;
6932 0 : bdev_io->u.nvme_passthru.md_buf = md_buf;
6933 0 : bdev_io->u.nvme_passthru.md_len = md_len;
6934 :
6935 0 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6936 :
6937 0 : bdev_io_submit(bdev_io);
6938 0 : return 0;
6939 : }
6940 :
6941 : static void bdev_abort_retry(void *ctx);
6942 : static void bdev_abort(struct spdk_bdev_io *parent_io);
6943 :
6944 : static void
6945 13 : bdev_abort_io_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
6946 : {
6947 13 : struct spdk_bdev_channel *channel = bdev_io->internal.ch;
6948 13 : struct spdk_bdev_io *parent_io = cb_arg;
6949 : struct spdk_bdev_io *bio_to_abort, *tmp_io;
6950 :
6951 13 : bio_to_abort = bdev_io->u.abort.bio_to_abort;
6952 :
6953 13 : spdk_bdev_free_io(bdev_io);
6954 :
6955 13 : if (!success) {
6956 : /* Check if the target I/O completed in the meantime. */
6957 2 : TAILQ_FOREACH(tmp_io, &channel->io_submitted, internal.ch_link) {
6958 1 : if (tmp_io == bio_to_abort) {
6959 0 : break;
6960 : }
6961 : }
6962 :
6963 : /* If the target I/O still exists, set the parent to failed. */
6964 1 : if (tmp_io != NULL) {
6965 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
6966 : }
6967 : }
6968 :
6969 13 : assert(parent_io->internal.f.split);
6970 :
6971 13 : parent_io->internal.split.outstanding--;
6972 13 : if (parent_io->internal.split.outstanding == 0) {
6973 10 : if (parent_io->internal.status == SPDK_BDEV_IO_STATUS_NOMEM) {
6974 0 : bdev_abort_retry(parent_io);
6975 : } else {
6976 10 : bdev_io_complete(parent_io);
6977 : }
6978 : }
6979 13 : }
6980 :
6981 : static int
6982 14 : bdev_abort_io(struct spdk_bdev_desc *desc, struct spdk_bdev_channel *channel,
6983 : struct spdk_bdev_io *bio_to_abort,
6984 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6985 : {
6986 14 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6987 : struct spdk_bdev_io *bdev_io;
6988 :
6989 14 : if (bio_to_abort->type == SPDK_BDEV_IO_TYPE_ABORT ||
6990 14 : bio_to_abort->type == SPDK_BDEV_IO_TYPE_RESET) {
6991 : /* TODO: Abort reset or abort request. */
6992 0 : return -ENOTSUP;
6993 : }
6994 :
6995 14 : bdev_io = bdev_channel_get_io(channel);
6996 14 : if (bdev_io == NULL) {
6997 1 : return -ENOMEM;
6998 : }
6999 :
7000 13 : bdev_io->internal.ch = channel;
7001 13 : bdev_io->internal.desc = desc;
7002 13 : bdev_io->type = SPDK_BDEV_IO_TYPE_ABORT;
7003 13 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
7004 :
7005 13 : if (bdev->split_on_optimal_io_boundary && bio_to_abort->internal.f.split) {
7006 3 : assert(bdev_io_should_split(bio_to_abort));
7007 3 : bdev_io->u.bdev.abort.bio_cb_arg = bio_to_abort;
7008 :
7009 : /* Parent abort request is not submitted directly, but to manage its
7010 : * execution add it to the submitted list here.
7011 : */
7012 3 : bdev_io->internal.submit_tsc = spdk_get_ticks();
7013 3 : bdev_ch_add_to_io_submitted(bdev_io);
7014 :
7015 3 : bdev_abort(bdev_io);
7016 :
7017 3 : return 0;
7018 : }
7019 :
7020 10 : bdev_io->u.abort.bio_to_abort = bio_to_abort;
7021 :
7022 : /* Submit the abort request to the underlying bdev module. */
7023 10 : bdev_io_submit(bdev_io);
7024 :
7025 10 : return 0;
7026 : }
7027 :
7028 : static bool
7029 28 : bdev_io_on_tailq(struct spdk_bdev_io *bdev_io, bdev_io_tailq_t *tailq)
7030 : {
7031 : struct spdk_bdev_io *iter;
7032 :
7033 28 : TAILQ_FOREACH(iter, tailq, internal.link) {
7034 0 : if (iter == bdev_io) {
7035 0 : return true;
7036 : }
7037 : }
7038 :
7039 28 : return false;
7040 : }
7041 :
7042 : static uint32_t
7043 12 : _bdev_abort(struct spdk_bdev_io *parent_io)
7044 : {
7045 12 : struct spdk_bdev_desc *desc = parent_io->internal.desc;
7046 12 : struct spdk_bdev_channel *channel = parent_io->internal.ch;
7047 : void *bio_cb_arg;
7048 : struct spdk_bdev_io *bio_to_abort;
7049 : uint32_t matched_ios;
7050 : int rc;
7051 :
7052 12 : bio_cb_arg = parent_io->u.bdev.abort.bio_cb_arg;
7053 :
7054 : /* matched_ios is returned and will be kept by the caller.
7055 : *
7056 : * This function will be used for two cases, 1) the same cb_arg is used for
7057 : * multiple I/Os, 2) a single large I/O is split into smaller ones.
7058 : * Incrementing split_outstanding directly here may confuse readers especially
7059 : * for the 1st case.
7060 : *
7061 : * Completion of I/O abort is processed after stack unwinding. Hence this trick
7062 : * works as expected.
7063 : */
7064 12 : matched_ios = 0;
7065 12 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
7066 :
7067 63 : TAILQ_FOREACH(bio_to_abort, &channel->io_submitted, internal.ch_link) {
7068 52 : if (bio_to_abort->internal.caller_ctx != bio_cb_arg) {
7069 38 : continue;
7070 : }
7071 :
7072 14 : if (bio_to_abort->internal.submit_tsc > parent_io->internal.submit_tsc) {
7073 : /* Any I/O which was submitted after this abort command should be excluded. */
7074 0 : continue;
7075 : }
7076 :
7077 : /* We can't abort a request that's being pushed/pulled or executed by accel */
7078 28 : if (bdev_io_on_tailq(bio_to_abort, &channel->io_accel_exec) ||
7079 14 : bdev_io_on_tailq(bio_to_abort, &channel->io_memory_domain)) {
7080 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
7081 0 : break;
7082 : }
7083 :
7084 14 : rc = bdev_abort_io(desc, channel, bio_to_abort, bdev_abort_io_done, parent_io);
7085 14 : if (rc != 0) {
7086 1 : if (rc == -ENOMEM) {
7087 1 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_NOMEM;
7088 : } else {
7089 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
7090 : }
7091 1 : break;
7092 : }
7093 13 : matched_ios++;
7094 : }
7095 :
7096 12 : return matched_ios;
7097 : }
7098 :
7099 : static void
7100 1 : bdev_abort_retry(void *ctx)
7101 : {
7102 1 : struct spdk_bdev_io *parent_io = ctx;
7103 : uint32_t matched_ios;
7104 :
7105 1 : matched_ios = _bdev_abort(parent_io);
7106 :
7107 1 : if (matched_ios == 0) {
7108 0 : if (parent_io->internal.status == SPDK_BDEV_IO_STATUS_NOMEM) {
7109 0 : bdev_queue_io_wait_with_cb(parent_io, bdev_abort_retry);
7110 : } else {
7111 : /* For retry, the case that no target I/O was found is success
7112 : * because it means target I/Os completed in the meantime.
7113 : */
7114 0 : bdev_io_complete(parent_io);
7115 : }
7116 0 : return;
7117 : }
7118 :
7119 : /* Use split_outstanding to manage the progress of aborting I/Os. */
7120 1 : parent_io->internal.f.split = true;
7121 1 : parent_io->internal.split.outstanding = matched_ios;
7122 : }
7123 :
7124 : static void
7125 11 : bdev_abort(struct spdk_bdev_io *parent_io)
7126 : {
7127 : uint32_t matched_ios;
7128 :
7129 11 : matched_ios = _bdev_abort(parent_io);
7130 :
7131 11 : if (matched_ios == 0) {
7132 2 : if (parent_io->internal.status == SPDK_BDEV_IO_STATUS_NOMEM) {
7133 1 : bdev_queue_io_wait_with_cb(parent_io, bdev_abort_retry);
7134 : } else {
7135 : /* The case the no target I/O was found is failure. */
7136 1 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
7137 1 : bdev_io_complete(parent_io);
7138 : }
7139 2 : return;
7140 : }
7141 :
7142 : /* Use split_outstanding to manage the progress of aborting I/Os. */
7143 9 : parent_io->internal.f.split = true;
7144 9 : parent_io->internal.split.outstanding = matched_ios;
7145 : }
7146 :
7147 : int
7148 9 : spdk_bdev_abort(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
7149 : void *bio_cb_arg,
7150 : spdk_bdev_io_completion_cb cb, void *cb_arg)
7151 : {
7152 9 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
7153 9 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
7154 : struct spdk_bdev_io *bdev_io;
7155 :
7156 9 : if (bio_cb_arg == NULL) {
7157 0 : return -EINVAL;
7158 : }
7159 :
7160 9 : if (!spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_ABORT)) {
7161 1 : return -ENOTSUP;
7162 : }
7163 :
7164 8 : bdev_io = bdev_channel_get_io(channel);
7165 8 : if (bdev_io == NULL) {
7166 0 : return -ENOMEM;
7167 : }
7168 :
7169 8 : bdev_io->internal.ch = channel;
7170 8 : bdev_io->internal.desc = desc;
7171 8 : bdev_io->internal.submit_tsc = spdk_get_ticks();
7172 8 : bdev_io->type = SPDK_BDEV_IO_TYPE_ABORT;
7173 8 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
7174 :
7175 8 : bdev_io->u.bdev.abort.bio_cb_arg = bio_cb_arg;
7176 :
7177 : /* Parent abort request is not submitted directly, but to manage its execution,
7178 : * add it to the submitted list here.
7179 : */
7180 8 : bdev_ch_add_to_io_submitted(bdev_io);
7181 :
7182 8 : bdev_abort(bdev_io);
7183 :
7184 8 : return 0;
7185 : }
7186 :
7187 : int
7188 4 : spdk_bdev_queue_io_wait(struct spdk_bdev *bdev, struct spdk_io_channel *ch,
7189 : struct spdk_bdev_io_wait_entry *entry)
7190 : {
7191 4 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
7192 4 : struct spdk_bdev_mgmt_channel *mgmt_ch = channel->shared_resource->mgmt_ch;
7193 :
7194 4 : if (bdev != entry->bdev) {
7195 0 : SPDK_ERRLOG("bdevs do not match\n");
7196 0 : return -EINVAL;
7197 : }
7198 :
7199 4 : if (mgmt_ch->per_thread_cache_count > 0) {
7200 0 : SPDK_ERRLOG("Cannot queue io_wait if spdk_bdev_io available in per-thread cache\n");
7201 0 : return -EINVAL;
7202 : }
7203 :
7204 4 : TAILQ_INSERT_TAIL(&mgmt_ch->io_wait_queue, entry, link);
7205 4 : return 0;
7206 : }
7207 :
7208 : static inline void
7209 597 : bdev_io_update_io_stat(struct spdk_bdev_io *bdev_io, uint64_t tsc_diff)
7210 : {
7211 597 : enum spdk_bdev_io_status io_status = bdev_io->internal.status;
7212 597 : struct spdk_bdev_io_stat *io_stat = bdev_io->internal.ch->stat;
7213 597 : uint64_t num_blocks = bdev_io->u.bdev.num_blocks;
7214 597 : uint32_t blocklen = bdev_io->bdev->blocklen;
7215 :
7216 597 : if (spdk_likely(io_status == SPDK_BDEV_IO_STATUS_SUCCESS)) {
7217 509 : switch (bdev_io->type) {
7218 326 : case SPDK_BDEV_IO_TYPE_READ:
7219 326 : io_stat->bytes_read += num_blocks * blocklen;
7220 326 : io_stat->num_read_ops++;
7221 326 : io_stat->read_latency_ticks += tsc_diff;
7222 326 : if (io_stat->max_read_latency_ticks < tsc_diff) {
7223 6 : io_stat->max_read_latency_ticks = tsc_diff;
7224 : }
7225 326 : if (io_stat->min_read_latency_ticks > tsc_diff) {
7226 41 : io_stat->min_read_latency_ticks = tsc_diff;
7227 : }
7228 326 : break;
7229 75 : case SPDK_BDEV_IO_TYPE_WRITE:
7230 75 : io_stat->bytes_written += num_blocks * blocklen;
7231 75 : io_stat->num_write_ops++;
7232 75 : io_stat->write_latency_ticks += tsc_diff;
7233 75 : if (io_stat->max_write_latency_ticks < tsc_diff) {
7234 4 : io_stat->max_write_latency_ticks = tsc_diff;
7235 : }
7236 75 : if (io_stat->min_write_latency_ticks > tsc_diff) {
7237 25 : io_stat->min_write_latency_ticks = tsc_diff;
7238 : }
7239 75 : break;
7240 20 : case SPDK_BDEV_IO_TYPE_UNMAP:
7241 20 : io_stat->bytes_unmapped += num_blocks * blocklen;
7242 20 : io_stat->num_unmap_ops++;
7243 20 : io_stat->unmap_latency_ticks += tsc_diff;
7244 20 : if (io_stat->max_unmap_latency_ticks < tsc_diff) {
7245 0 : io_stat->max_unmap_latency_ticks = tsc_diff;
7246 : }
7247 20 : if (io_stat->min_unmap_latency_ticks > tsc_diff) {
7248 3 : io_stat->min_unmap_latency_ticks = tsc_diff;
7249 : }
7250 20 : break;
7251 4 : case SPDK_BDEV_IO_TYPE_ZCOPY:
7252 : /* Track the data in the start phase only */
7253 4 : if (bdev_io->u.bdev.zcopy.start) {
7254 2 : if (bdev_io->u.bdev.zcopy.populate) {
7255 1 : io_stat->bytes_read += num_blocks * blocklen;
7256 1 : io_stat->num_read_ops++;
7257 1 : io_stat->read_latency_ticks += tsc_diff;
7258 1 : if (io_stat->max_read_latency_ticks < tsc_diff) {
7259 0 : io_stat->max_read_latency_ticks = tsc_diff;
7260 : }
7261 1 : if (io_stat->min_read_latency_ticks > tsc_diff) {
7262 1 : io_stat->min_read_latency_ticks = tsc_diff;
7263 : }
7264 : } else {
7265 1 : io_stat->bytes_written += num_blocks * blocklen;
7266 1 : io_stat->num_write_ops++;
7267 1 : io_stat->write_latency_ticks += tsc_diff;
7268 1 : if (io_stat->max_write_latency_ticks < tsc_diff) {
7269 0 : io_stat->max_write_latency_ticks = tsc_diff;
7270 : }
7271 1 : if (io_stat->min_write_latency_ticks > tsc_diff) {
7272 1 : io_stat->min_write_latency_ticks = tsc_diff;
7273 : }
7274 : }
7275 : }
7276 4 : break;
7277 21 : case SPDK_BDEV_IO_TYPE_COPY:
7278 21 : io_stat->bytes_copied += num_blocks * blocklen;
7279 21 : io_stat->num_copy_ops++;
7280 21 : bdev_io->internal.ch->stat->copy_latency_ticks += tsc_diff;
7281 21 : if (io_stat->max_copy_latency_ticks < tsc_diff) {
7282 0 : io_stat->max_copy_latency_ticks = tsc_diff;
7283 : }
7284 21 : if (io_stat->min_copy_latency_ticks > tsc_diff) {
7285 4 : io_stat->min_copy_latency_ticks = tsc_diff;
7286 : }
7287 21 : break;
7288 63 : default:
7289 63 : break;
7290 : }
7291 88 : } else if (io_status <= SPDK_BDEV_IO_STATUS_FAILED && io_status >= SPDK_MIN_BDEV_IO_STATUS) {
7292 88 : io_stat = bdev_io->bdev->internal.stat;
7293 88 : assert(io_stat->io_error != NULL);
7294 :
7295 88 : spdk_spin_lock(&bdev_io->bdev->internal.spinlock);
7296 88 : io_stat->io_error->error_status[-io_status - 1]++;
7297 88 : spdk_spin_unlock(&bdev_io->bdev->internal.spinlock);
7298 : }
7299 :
7300 : #ifdef SPDK_CONFIG_VTUNE
7301 : uint64_t now_tsc = spdk_get_ticks();
7302 : if (now_tsc > (bdev_io->internal.ch->start_tsc + bdev_io->internal.ch->interval_tsc)) {
7303 : uint64_t data[5];
7304 : struct spdk_bdev_io_stat *prev_stat = bdev_io->internal.ch->prev_stat;
7305 :
7306 : data[0] = io_stat->num_read_ops - prev_stat->num_read_ops;
7307 : data[1] = io_stat->bytes_read - prev_stat->bytes_read;
7308 : data[2] = io_stat->num_write_ops - prev_stat->num_write_ops;
7309 : data[3] = io_stat->bytes_written - prev_stat->bytes_written;
7310 : data[4] = bdev_io->bdev->fn_table->get_spin_time ?
7311 : bdev_io->bdev->fn_table->get_spin_time(spdk_bdev_io_get_io_channel(bdev_io)) : 0;
7312 :
7313 : __itt_metadata_add(g_bdev_mgr.domain, __itt_null, bdev_io->internal.ch->handle,
7314 : __itt_metadata_u64, 5, data);
7315 :
7316 : memcpy(prev_stat, io_stat, sizeof(struct spdk_bdev_io_stat));
7317 : bdev_io->internal.ch->start_tsc = now_tsc;
7318 : }
7319 : #endif
7320 597 : }
7321 :
7322 : static inline void
7323 597 : _bdev_io_complete(void *ctx)
7324 : {
7325 597 : struct spdk_bdev_io *bdev_io = ctx;
7326 :
7327 597 : if (spdk_unlikely(bdev_io_use_accel_sequence(bdev_io))) {
7328 0 : assert(bdev_io->internal.status != SPDK_BDEV_IO_STATUS_SUCCESS);
7329 0 : spdk_accel_sequence_abort(bdev_io->internal.accel_sequence);
7330 : }
7331 :
7332 597 : assert(bdev_io->internal.cb != NULL);
7333 597 : assert(spdk_get_thread() == spdk_bdev_io_get_thread(bdev_io));
7334 :
7335 597 : bdev_io->internal.cb(bdev_io, bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS,
7336 : bdev_io->internal.caller_ctx);
7337 597 : }
7338 :
7339 : static inline void
7340 605 : bdev_io_complete(void *ctx)
7341 : {
7342 605 : struct spdk_bdev_io *bdev_io = ctx;
7343 605 : struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
7344 : uint64_t tsc, tsc_diff;
7345 :
7346 605 : if (spdk_unlikely(bdev_io->internal.in_submit_request)) {
7347 : /*
7348 : * Defer completion to avoid potential infinite recursion if the
7349 : * user's completion callback issues a new I/O.
7350 : */
7351 8 : spdk_thread_send_msg(spdk_bdev_io_get_thread(bdev_io),
7352 : bdev_io_complete, bdev_io);
7353 8 : return;
7354 : }
7355 :
7356 597 : tsc = spdk_get_ticks();
7357 597 : tsc_diff = tsc - bdev_io->internal.submit_tsc;
7358 :
7359 597 : bdev_ch_remove_from_io_submitted(bdev_io);
7360 597 : spdk_trace_record_tsc(tsc, TRACE_BDEV_IO_DONE, bdev_ch->trace_id, 0, (uintptr_t)bdev_io,
7361 : bdev_io->internal.caller_ctx, bdev_ch->queue_depth);
7362 :
7363 597 : if (bdev_ch->histogram) {
7364 4 : if (bdev_io->bdev->internal.histogram_io_type == 0 ||
7365 0 : bdev_io->bdev->internal.histogram_io_type == bdev_io->type) {
7366 : /*
7367 : * Tally all I/O types if the histogram_io_type is set to 0.
7368 : */
7369 4 : spdk_histogram_data_tally(bdev_ch->histogram, tsc_diff);
7370 : }
7371 : }
7372 :
7373 597 : bdev_io_update_io_stat(bdev_io, tsc_diff);
7374 597 : _bdev_io_complete(bdev_io);
7375 : }
7376 :
7377 : /* The difference between this function and bdev_io_complete() is that this should be called to
7378 : * complete IOs that haven't been submitted via bdev_io_submit(), as they weren't added onto the
7379 : * io_submitted list and don't have submit_tsc updated.
7380 : */
7381 : static inline void
7382 0 : bdev_io_complete_unsubmitted(struct spdk_bdev_io *bdev_io)
7383 : {
7384 : /* Since the IO hasn't been submitted it's bound to be failed */
7385 0 : assert(bdev_io->internal.status != SPDK_BDEV_IO_STATUS_SUCCESS);
7386 :
7387 : /* At this point we don't know if the IO is completed from submission context or not, but,
7388 : * since this is an error path, we can always do an spdk_thread_send_msg(). */
7389 0 : spdk_thread_send_msg(spdk_bdev_io_get_thread(bdev_io),
7390 : _bdev_io_complete, bdev_io);
7391 0 : }
7392 :
7393 : static void bdev_destroy_cb(void *io_device);
7394 :
7395 : static void
7396 15 : bdev_reset_complete(struct spdk_bdev *bdev, void *_ctx, int status)
7397 : {
7398 15 : struct spdk_bdev_io *bdev_io = _ctx;
7399 :
7400 15 : if (bdev_io->u.reset.ch_ref != NULL) {
7401 15 : spdk_put_io_channel(bdev_io->u.reset.ch_ref);
7402 15 : bdev_io->u.reset.ch_ref = NULL;
7403 : }
7404 :
7405 15 : bdev_io_complete(bdev_io);
7406 :
7407 15 : if (bdev->internal.status == SPDK_BDEV_STATUS_REMOVING &&
7408 1 : TAILQ_EMPTY(&bdev->internal.open_descs)) {
7409 1 : spdk_io_device_unregister(__bdev_to_io_dev(bdev), bdev_destroy_cb);
7410 : }
7411 15 : }
7412 :
7413 : static void
7414 18 : bdev_unfreeze_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
7415 : struct spdk_io_channel *_ch, void *_ctx)
7416 : {
7417 18 : struct spdk_bdev_io *bdev_io = _ctx;
7418 18 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
7419 : struct spdk_bdev_io *queued_reset;
7420 :
7421 18 : ch->flags &= ~BDEV_CH_RESET_IN_PROGRESS;
7422 18 : while (!TAILQ_EMPTY(&ch->queued_resets)) {
7423 0 : queued_reset = TAILQ_FIRST(&ch->queued_resets);
7424 0 : TAILQ_REMOVE(&ch->queued_resets, queued_reset, internal.link);
7425 0 : spdk_bdev_io_complete(queued_reset, bdev_io->internal.status);
7426 : }
7427 :
7428 18 : spdk_bdev_for_each_channel_continue(i, 0);
7429 18 : }
7430 :
7431 : static void
7432 0 : bdev_io_complete_sequence_cb(void *ctx, int status)
7433 : {
7434 0 : struct spdk_bdev_io *bdev_io = ctx;
7435 :
7436 : /* u.bdev.accel_sequence should have already been cleared at this point */
7437 0 : assert(bdev_io->u.bdev.accel_sequence == NULL);
7438 0 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS);
7439 0 : bdev_io->internal.f.has_accel_sequence = false;
7440 :
7441 0 : if (spdk_unlikely(status != 0)) {
7442 0 : SPDK_ERRLOG("Failed to execute accel sequence, status=%d\n", status);
7443 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
7444 : }
7445 :
7446 0 : bdev_io_complete(bdev_io);
7447 0 : }
7448 :
7449 : void
7450 591 : spdk_bdev_io_complete(struct spdk_bdev_io *bdev_io, enum spdk_bdev_io_status status)
7451 : {
7452 591 : struct spdk_bdev *bdev = bdev_io->bdev;
7453 591 : struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
7454 591 : struct spdk_bdev_shared_resource *shared_resource = bdev_ch->shared_resource;
7455 :
7456 591 : if (spdk_unlikely(bdev_io->internal.status != SPDK_BDEV_IO_STATUS_PENDING)) {
7457 0 : SPDK_ERRLOG("Unexpected completion on IO from %s module, status was %s\n",
7458 : spdk_bdev_get_module_name(bdev),
7459 : bdev_io_status_get_string(bdev_io->internal.status));
7460 0 : assert(false);
7461 : }
7462 591 : bdev_io->internal.status = status;
7463 :
7464 591 : if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_RESET)) {
7465 16 : bool unlock_channels = false;
7466 :
7467 16 : if (status == SPDK_BDEV_IO_STATUS_NOMEM) {
7468 0 : SPDK_ERRLOG("NOMEM returned for reset\n");
7469 : }
7470 16 : spdk_spin_lock(&bdev->internal.spinlock);
7471 16 : if (bdev_io == bdev->internal.reset_in_progress) {
7472 15 : bdev->internal.reset_in_progress = NULL;
7473 15 : unlock_channels = true;
7474 : }
7475 16 : spdk_spin_unlock(&bdev->internal.spinlock);
7476 :
7477 16 : if (unlock_channels) {
7478 15 : spdk_bdev_for_each_channel(bdev, bdev_unfreeze_channel, bdev_io,
7479 : bdev_reset_complete);
7480 15 : return;
7481 : }
7482 : } else {
7483 575 : bdev_io_decrement_outstanding(bdev_ch, shared_resource);
7484 575 : if (spdk_likely(status == SPDK_BDEV_IO_STATUS_SUCCESS)) {
7485 484 : if (bdev_io_needs_sequence_exec(bdev_io->internal.desc, bdev_io)) {
7486 0 : bdev_io_exec_sequence(bdev_io, bdev_io_complete_sequence_cb);
7487 0 : return;
7488 484 : } else if (spdk_unlikely(bdev_io->internal.orig_iovcnt != 0 &&
7489 : !bdev_io_use_accel_sequence(bdev_io))) {
7490 28 : _bdev_io_push_bounce_data_buffer(bdev_io,
7491 : _bdev_io_complete_push_bounce_done);
7492 : /* bdev IO will be completed in the callback */
7493 28 : return;
7494 : }
7495 : }
7496 :
7497 547 : if (spdk_unlikely(_bdev_io_handle_no_mem(bdev_io, BDEV_IO_RETRY_STATE_SUBMIT))) {
7498 5 : return;
7499 : }
7500 : }
7501 :
7502 543 : bdev_io_complete(bdev_io);
7503 : }
7504 :
7505 : void
7506 0 : spdk_bdev_io_complete_scsi_status(struct spdk_bdev_io *bdev_io, enum spdk_scsi_status sc,
7507 : enum spdk_scsi_sense sk, uint8_t asc, uint8_t ascq)
7508 : {
7509 : enum spdk_bdev_io_status status;
7510 :
7511 0 : if (sc == SPDK_SCSI_STATUS_GOOD) {
7512 0 : status = SPDK_BDEV_IO_STATUS_SUCCESS;
7513 : } else {
7514 0 : status = SPDK_BDEV_IO_STATUS_SCSI_ERROR;
7515 0 : bdev_io->internal.error.scsi.sc = sc;
7516 0 : bdev_io->internal.error.scsi.sk = sk;
7517 0 : bdev_io->internal.error.scsi.asc = asc;
7518 0 : bdev_io->internal.error.scsi.ascq = ascq;
7519 : }
7520 :
7521 0 : spdk_bdev_io_complete(bdev_io, status);
7522 0 : }
7523 :
7524 : void
7525 0 : spdk_bdev_io_get_scsi_status(const struct spdk_bdev_io *bdev_io,
7526 : int *sc, int *sk, int *asc, int *ascq)
7527 : {
7528 0 : assert(sc != NULL);
7529 0 : assert(sk != NULL);
7530 0 : assert(asc != NULL);
7531 0 : assert(ascq != NULL);
7532 :
7533 0 : switch (bdev_io->internal.status) {
7534 0 : case SPDK_BDEV_IO_STATUS_SUCCESS:
7535 0 : *sc = SPDK_SCSI_STATUS_GOOD;
7536 0 : *sk = SPDK_SCSI_SENSE_NO_SENSE;
7537 0 : *asc = SPDK_SCSI_ASC_NO_ADDITIONAL_SENSE;
7538 0 : *ascq = SPDK_SCSI_ASCQ_CAUSE_NOT_REPORTABLE;
7539 0 : break;
7540 0 : case SPDK_BDEV_IO_STATUS_NVME_ERROR:
7541 0 : spdk_scsi_nvme_translate(bdev_io, sc, sk, asc, ascq);
7542 0 : break;
7543 0 : case SPDK_BDEV_IO_STATUS_MISCOMPARE:
7544 0 : *sc = SPDK_SCSI_STATUS_CHECK_CONDITION;
7545 0 : *sk = SPDK_SCSI_SENSE_MISCOMPARE;
7546 0 : *asc = SPDK_SCSI_ASC_MISCOMPARE_DURING_VERIFY_OPERATION;
7547 0 : *ascq = bdev_io->internal.error.scsi.ascq;
7548 0 : break;
7549 0 : case SPDK_BDEV_IO_STATUS_SCSI_ERROR:
7550 0 : *sc = bdev_io->internal.error.scsi.sc;
7551 0 : *sk = bdev_io->internal.error.scsi.sk;
7552 0 : *asc = bdev_io->internal.error.scsi.asc;
7553 0 : *ascq = bdev_io->internal.error.scsi.ascq;
7554 0 : break;
7555 0 : default:
7556 0 : *sc = SPDK_SCSI_STATUS_CHECK_CONDITION;
7557 0 : *sk = SPDK_SCSI_SENSE_ABORTED_COMMAND;
7558 0 : *asc = SPDK_SCSI_ASC_NO_ADDITIONAL_SENSE;
7559 0 : *ascq = SPDK_SCSI_ASCQ_CAUSE_NOT_REPORTABLE;
7560 0 : break;
7561 : }
7562 0 : }
7563 :
7564 : void
7565 0 : spdk_bdev_io_complete_aio_status(struct spdk_bdev_io *bdev_io, int aio_result)
7566 : {
7567 : enum spdk_bdev_io_status status;
7568 :
7569 0 : if (aio_result == 0) {
7570 0 : status = SPDK_BDEV_IO_STATUS_SUCCESS;
7571 : } else {
7572 0 : status = SPDK_BDEV_IO_STATUS_AIO_ERROR;
7573 : }
7574 :
7575 0 : bdev_io->internal.error.aio_result = aio_result;
7576 :
7577 0 : spdk_bdev_io_complete(bdev_io, status);
7578 0 : }
7579 :
7580 : void
7581 0 : spdk_bdev_io_get_aio_status(const struct spdk_bdev_io *bdev_io, int *aio_result)
7582 : {
7583 0 : assert(aio_result != NULL);
7584 :
7585 0 : if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_AIO_ERROR) {
7586 0 : *aio_result = bdev_io->internal.error.aio_result;
7587 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS) {
7588 0 : *aio_result = 0;
7589 : } else {
7590 0 : *aio_result = -EIO;
7591 : }
7592 0 : }
7593 :
7594 : void
7595 0 : spdk_bdev_io_complete_nvme_status(struct spdk_bdev_io *bdev_io, uint32_t cdw0, int sct, int sc)
7596 : {
7597 : enum spdk_bdev_io_status status;
7598 :
7599 0 : if (spdk_likely(sct == SPDK_NVME_SCT_GENERIC && sc == SPDK_NVME_SC_SUCCESS)) {
7600 0 : status = SPDK_BDEV_IO_STATUS_SUCCESS;
7601 0 : } else if (sct == SPDK_NVME_SCT_GENERIC && sc == SPDK_NVME_SC_ABORTED_BY_REQUEST) {
7602 0 : status = SPDK_BDEV_IO_STATUS_ABORTED;
7603 : } else {
7604 0 : status = SPDK_BDEV_IO_STATUS_NVME_ERROR;
7605 : }
7606 :
7607 0 : bdev_io->internal.error.nvme.cdw0 = cdw0;
7608 0 : bdev_io->internal.error.nvme.sct = sct;
7609 0 : bdev_io->internal.error.nvme.sc = sc;
7610 :
7611 0 : spdk_bdev_io_complete(bdev_io, status);
7612 0 : }
7613 :
7614 : void
7615 0 : spdk_bdev_io_get_nvme_status(const struct spdk_bdev_io *bdev_io, uint32_t *cdw0, int *sct, int *sc)
7616 : {
7617 0 : assert(sct != NULL);
7618 0 : assert(sc != NULL);
7619 0 : assert(cdw0 != NULL);
7620 :
7621 0 : if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_ABORT)) {
7622 0 : *sct = SPDK_NVME_SCT_GENERIC;
7623 0 : *sc = SPDK_NVME_SC_SUCCESS;
7624 0 : if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS) {
7625 0 : *cdw0 = 0;
7626 : } else {
7627 0 : *cdw0 = 1U;
7628 : }
7629 0 : return;
7630 : }
7631 :
7632 0 : if (spdk_likely(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS)) {
7633 0 : *sct = SPDK_NVME_SCT_GENERIC;
7634 0 : *sc = SPDK_NVME_SC_SUCCESS;
7635 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_NVME_ERROR) {
7636 0 : *sct = bdev_io->internal.error.nvme.sct;
7637 0 : *sc = bdev_io->internal.error.nvme.sc;
7638 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_ABORTED) {
7639 0 : *sct = SPDK_NVME_SCT_GENERIC;
7640 0 : *sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
7641 : } else {
7642 0 : *sct = SPDK_NVME_SCT_GENERIC;
7643 0 : *sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
7644 : }
7645 :
7646 0 : *cdw0 = bdev_io->internal.error.nvme.cdw0;
7647 : }
7648 :
7649 : void
7650 0 : spdk_bdev_io_get_nvme_fused_status(const struct spdk_bdev_io *bdev_io, uint32_t *cdw0,
7651 : int *first_sct, int *first_sc, int *second_sct, int *second_sc)
7652 : {
7653 0 : assert(first_sct != NULL);
7654 0 : assert(first_sc != NULL);
7655 0 : assert(second_sct != NULL);
7656 0 : assert(second_sc != NULL);
7657 0 : assert(cdw0 != NULL);
7658 :
7659 0 : if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_NVME_ERROR) {
7660 0 : if (bdev_io->internal.error.nvme.sct == SPDK_NVME_SCT_MEDIA_ERROR &&
7661 0 : bdev_io->internal.error.nvme.sc == SPDK_NVME_SC_COMPARE_FAILURE) {
7662 0 : *first_sct = bdev_io->internal.error.nvme.sct;
7663 0 : *first_sc = bdev_io->internal.error.nvme.sc;
7664 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
7665 0 : *second_sc = SPDK_NVME_SC_ABORTED_FAILED_FUSED;
7666 : } else {
7667 0 : *first_sct = SPDK_NVME_SCT_GENERIC;
7668 0 : *first_sc = SPDK_NVME_SC_SUCCESS;
7669 0 : *second_sct = bdev_io->internal.error.nvme.sct;
7670 0 : *second_sc = bdev_io->internal.error.nvme.sc;
7671 : }
7672 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_ABORTED) {
7673 0 : *first_sct = SPDK_NVME_SCT_GENERIC;
7674 0 : *first_sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
7675 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
7676 0 : *second_sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
7677 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS) {
7678 0 : *first_sct = SPDK_NVME_SCT_GENERIC;
7679 0 : *first_sc = SPDK_NVME_SC_SUCCESS;
7680 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
7681 0 : *second_sc = SPDK_NVME_SC_SUCCESS;
7682 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_FIRST_FUSED_FAILED) {
7683 0 : *first_sct = SPDK_NVME_SCT_GENERIC;
7684 0 : *first_sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
7685 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
7686 0 : *second_sc = SPDK_NVME_SC_ABORTED_FAILED_FUSED;
7687 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_MISCOMPARE) {
7688 0 : *first_sct = SPDK_NVME_SCT_MEDIA_ERROR;
7689 0 : *first_sc = SPDK_NVME_SC_COMPARE_FAILURE;
7690 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
7691 0 : *second_sc = SPDK_NVME_SC_ABORTED_FAILED_FUSED;
7692 : } else {
7693 0 : *first_sct = SPDK_NVME_SCT_GENERIC;
7694 0 : *first_sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
7695 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
7696 0 : *second_sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
7697 : }
7698 :
7699 0 : *cdw0 = bdev_io->internal.error.nvme.cdw0;
7700 0 : }
7701 :
7702 : void
7703 0 : spdk_bdev_io_complete_base_io_status(struct spdk_bdev_io *bdev_io,
7704 : const struct spdk_bdev_io *base_io)
7705 : {
7706 0 : switch (base_io->internal.status) {
7707 0 : case SPDK_BDEV_IO_STATUS_NVME_ERROR:
7708 0 : spdk_bdev_io_complete_nvme_status(bdev_io,
7709 0 : base_io->internal.error.nvme.cdw0,
7710 0 : base_io->internal.error.nvme.sct,
7711 0 : base_io->internal.error.nvme.sc);
7712 0 : break;
7713 0 : case SPDK_BDEV_IO_STATUS_SCSI_ERROR:
7714 0 : spdk_bdev_io_complete_scsi_status(bdev_io,
7715 0 : base_io->internal.error.scsi.sc,
7716 0 : base_io->internal.error.scsi.sk,
7717 0 : base_io->internal.error.scsi.asc,
7718 0 : base_io->internal.error.scsi.ascq);
7719 0 : break;
7720 0 : case SPDK_BDEV_IO_STATUS_AIO_ERROR:
7721 0 : spdk_bdev_io_complete_aio_status(bdev_io, base_io->internal.error.aio_result);
7722 0 : break;
7723 0 : default:
7724 0 : spdk_bdev_io_complete(bdev_io, base_io->internal.status);
7725 0 : break;
7726 : }
7727 0 : }
7728 :
7729 : struct spdk_thread *
7730 651 : spdk_bdev_io_get_thread(struct spdk_bdev_io *bdev_io)
7731 : {
7732 651 : return spdk_io_channel_get_thread(bdev_io->internal.ch->channel);
7733 : }
7734 :
7735 : struct spdk_io_channel *
7736 78 : spdk_bdev_io_get_io_channel(struct spdk_bdev_io *bdev_io)
7737 : {
7738 78 : return bdev_io->internal.ch->channel;
7739 : }
7740 :
7741 : static int
7742 125 : bdev_register(struct spdk_bdev *bdev)
7743 : {
7744 : char *bdev_name;
7745 125 : char uuid[SPDK_UUID_STRING_LEN];
7746 125 : struct spdk_iobuf_opts iobuf_opts;
7747 : int ret;
7748 :
7749 125 : assert(bdev->module != NULL);
7750 :
7751 125 : if (!bdev->name) {
7752 0 : SPDK_ERRLOG("Bdev name is NULL\n");
7753 0 : return -EINVAL;
7754 : }
7755 :
7756 125 : if (!strlen(bdev->name)) {
7757 0 : SPDK_ERRLOG("Bdev name must not be an empty string\n");
7758 0 : return -EINVAL;
7759 : }
7760 :
7761 : /* Users often register their own I/O devices using the bdev name. In
7762 : * order to avoid conflicts, prepend bdev_. */
7763 125 : bdev_name = spdk_sprintf_alloc("bdev_%s", bdev->name);
7764 125 : if (!bdev_name) {
7765 0 : SPDK_ERRLOG("Unable to allocate memory for internal bdev name.\n");
7766 0 : return -ENOMEM;
7767 : }
7768 :
7769 125 : bdev->internal.stat = bdev_alloc_io_stat(true);
7770 125 : if (!bdev->internal.stat) {
7771 0 : SPDK_ERRLOG("Unable to allocate I/O statistics structure.\n");
7772 0 : free(bdev_name);
7773 0 : return -ENOMEM;
7774 : }
7775 :
7776 125 : bdev->internal.status = SPDK_BDEV_STATUS_READY;
7777 125 : bdev->internal.measured_queue_depth = UINT64_MAX;
7778 125 : bdev->internal.claim_type = SPDK_BDEV_CLAIM_NONE;
7779 125 : memset(&bdev->internal.claim, 0, sizeof(bdev->internal.claim));
7780 125 : bdev->internal.qd_poller = NULL;
7781 125 : bdev->internal.qos = NULL;
7782 :
7783 125 : TAILQ_INIT(&bdev->internal.open_descs);
7784 125 : TAILQ_INIT(&bdev->internal.locked_ranges);
7785 125 : TAILQ_INIT(&bdev->internal.pending_locked_ranges);
7786 125 : TAILQ_INIT(&bdev->aliases);
7787 :
7788 : /* UUID may be specified by the user or defined by bdev itself.
7789 : * Otherwise it will be generated here, so this field will never be empty. */
7790 125 : if (spdk_uuid_is_null(&bdev->uuid)) {
7791 42 : spdk_uuid_generate(&bdev->uuid);
7792 : }
7793 :
7794 : /* Add the UUID alias only if it's different than the name */
7795 125 : spdk_uuid_fmt_lower(uuid, sizeof(uuid), &bdev->uuid);
7796 125 : if (strcmp(bdev->name, uuid) != 0) {
7797 124 : ret = spdk_bdev_alias_add(bdev, uuid);
7798 124 : if (ret != 0) {
7799 2 : SPDK_ERRLOG("Unable to add uuid:%s alias for bdev %s\n", uuid, bdev->name);
7800 2 : bdev_free_io_stat(bdev->internal.stat);
7801 2 : free(bdev_name);
7802 2 : return ret;
7803 : }
7804 : }
7805 :
7806 123 : spdk_iobuf_get_opts(&iobuf_opts, sizeof(iobuf_opts));
7807 123 : if (spdk_bdev_get_buf_align(bdev) > 1) {
7808 0 : bdev->max_rw_size = spdk_min(bdev->max_rw_size ? bdev->max_rw_size : UINT32_MAX,
7809 : iobuf_opts.large_bufsize / bdev->blocklen);
7810 : }
7811 :
7812 : /* If the user didn't specify a write unit size, set it to one. */
7813 123 : if (bdev->write_unit_size == 0) {
7814 119 : bdev->write_unit_size = 1;
7815 : }
7816 :
7817 : /* Set ACWU value to the write unit size if bdev module did not set it (does not support it natively) */
7818 123 : if (bdev->acwu == 0) {
7819 119 : bdev->acwu = bdev->write_unit_size;
7820 : }
7821 :
7822 123 : if (bdev->phys_blocklen == 0) {
7823 119 : bdev->phys_blocklen = spdk_bdev_get_data_block_size(bdev);
7824 : }
7825 :
7826 123 : if (!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COPY)) {
7827 0 : bdev->max_copy = bdev_get_max_write(bdev, iobuf_opts.large_bufsize);
7828 : }
7829 :
7830 123 : if (!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE_ZEROES)) {
7831 0 : bdev->max_write_zeroes = bdev_get_max_write(bdev, ZERO_BUFFER_SIZE);
7832 : }
7833 :
7834 123 : bdev->internal.reset_in_progress = NULL;
7835 123 : bdev->internal.qd_poll_in_progress = false;
7836 123 : bdev->internal.period = 0;
7837 123 : bdev->internal.new_period = 0;
7838 123 : bdev->internal.trace_id = spdk_trace_register_owner(OWNER_TYPE_BDEV, bdev_name);
7839 :
7840 : /*
7841 : * Initialize spinlock before registering IO device because spinlock is used in
7842 : * bdev_channel_create
7843 : */
7844 123 : spdk_spin_init(&bdev->internal.spinlock);
7845 :
7846 123 : spdk_io_device_register(__bdev_to_io_dev(bdev),
7847 : bdev_channel_create, bdev_channel_destroy,
7848 : sizeof(struct spdk_bdev_channel),
7849 : bdev_name);
7850 :
7851 : /*
7852 : * Register bdev name only after the bdev object is ready.
7853 : * After bdev_name_add returns, it is possible for other threads to start using the bdev,
7854 : * create IO channels...
7855 : */
7856 123 : ret = bdev_name_add(&bdev->internal.bdev_name, bdev, bdev->name);
7857 123 : if (ret != 0) {
7858 0 : spdk_io_device_unregister(__bdev_to_io_dev(bdev), NULL);
7859 0 : bdev_free_io_stat(bdev->internal.stat);
7860 0 : spdk_spin_destroy(&bdev->internal.spinlock);
7861 0 : free(bdev_name);
7862 0 : return ret;
7863 : }
7864 :
7865 123 : free(bdev_name);
7866 :
7867 123 : SPDK_DEBUGLOG(bdev, "Inserting bdev %s into list\n", bdev->name);
7868 123 : TAILQ_INSERT_TAIL(&g_bdev_mgr.bdevs, bdev, internal.link);
7869 :
7870 123 : return 0;
7871 : }
7872 :
7873 : static void
7874 124 : bdev_destroy_cb(void *io_device)
7875 : {
7876 : int rc;
7877 : struct spdk_bdev *bdev;
7878 : spdk_bdev_unregister_cb cb_fn;
7879 : void *cb_arg;
7880 :
7881 124 : bdev = __bdev_from_io_dev(io_device);
7882 :
7883 124 : if (bdev->internal.unregister_td != spdk_get_thread()) {
7884 1 : spdk_thread_send_msg(bdev->internal.unregister_td, bdev_destroy_cb, io_device);
7885 1 : return;
7886 : }
7887 :
7888 123 : cb_fn = bdev->internal.unregister_cb;
7889 123 : cb_arg = bdev->internal.unregister_ctx;
7890 :
7891 123 : spdk_spin_destroy(&bdev->internal.spinlock);
7892 123 : free(bdev->internal.qos);
7893 123 : bdev_free_io_stat(bdev->internal.stat);
7894 123 : spdk_trace_unregister_owner(bdev->internal.trace_id);
7895 :
7896 123 : rc = bdev->fn_table->destruct(bdev->ctxt);
7897 123 : if (rc < 0) {
7898 0 : SPDK_ERRLOG("destruct failed\n");
7899 : }
7900 123 : if (rc <= 0 && cb_fn != NULL) {
7901 10 : cb_fn(cb_arg, rc);
7902 : }
7903 : }
7904 :
7905 : void
7906 2 : spdk_bdev_destruct_done(struct spdk_bdev *bdev, int bdeverrno)
7907 : {
7908 2 : if (bdev->internal.unregister_cb != NULL) {
7909 0 : bdev->internal.unregister_cb(bdev->internal.unregister_ctx, bdeverrno);
7910 : }
7911 2 : }
7912 :
7913 : static void
7914 19 : _remove_notify(void *arg)
7915 : {
7916 19 : struct spdk_bdev_desc *desc = arg;
7917 :
7918 19 : _event_notify(desc, SPDK_BDEV_EVENT_REMOVE);
7919 19 : }
7920 :
7921 : /* returns: 0 - bdev removed and ready to be destructed.
7922 : * -EBUSY - bdev can't be destructed yet. */
7923 : static int
7924 138 : bdev_unregister_unsafe(struct spdk_bdev *bdev)
7925 : {
7926 : struct spdk_bdev_desc *desc, *tmp;
7927 138 : int rc = 0;
7928 138 : char uuid[SPDK_UUID_STRING_LEN];
7929 :
7930 138 : assert(spdk_spin_held(&g_bdev_mgr.spinlock));
7931 138 : assert(spdk_spin_held(&bdev->internal.spinlock));
7932 :
7933 : /* Notify each descriptor about hotremoval */
7934 157 : TAILQ_FOREACH_SAFE(desc, &bdev->internal.open_descs, link, tmp) {
7935 19 : rc = -EBUSY;
7936 : /*
7937 : * Defer invocation of the event_cb to a separate message that will
7938 : * run later on its thread. This ensures this context unwinds and
7939 : * we don't recursively unregister this bdev again if the event_cb
7940 : * immediately closes its descriptor.
7941 : */
7942 19 : event_notify(desc, _remove_notify);
7943 : }
7944 :
7945 : /* If there are no descriptors, proceed removing the bdev */
7946 138 : if (rc == 0) {
7947 123 : TAILQ_REMOVE(&g_bdev_mgr.bdevs, bdev, internal.link);
7948 123 : SPDK_DEBUGLOG(bdev, "Removing bdev %s from list done\n", bdev->name);
7949 :
7950 : /* Delete the name and the UUID alias */
7951 123 : spdk_uuid_fmt_lower(uuid, sizeof(uuid), &bdev->uuid);
7952 123 : bdev_name_del_unsafe(&bdev->internal.bdev_name);
7953 123 : bdev_alias_del(bdev, uuid, bdev_name_del_unsafe);
7954 :
7955 123 : spdk_notify_send("bdev_unregister", spdk_bdev_get_name(bdev));
7956 :
7957 123 : if (bdev->internal.reset_in_progress != NULL) {
7958 : /* If reset is in progress, let the completion callback for reset
7959 : * unregister the bdev.
7960 : */
7961 1 : rc = -EBUSY;
7962 : }
7963 : }
7964 :
7965 138 : return rc;
7966 : }
7967 :
7968 : static void
7969 4 : bdev_unregister_abort_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
7970 : struct spdk_io_channel *io_ch, void *_ctx)
7971 : {
7972 4 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
7973 :
7974 4 : bdev_channel_abort_queued_ios(bdev_ch);
7975 4 : spdk_bdev_for_each_channel_continue(i, 0);
7976 4 : }
7977 :
7978 : static void
7979 123 : bdev_unregister(struct spdk_bdev *bdev, void *_ctx, int status)
7980 : {
7981 : int rc;
7982 :
7983 123 : spdk_spin_lock(&g_bdev_mgr.spinlock);
7984 123 : spdk_spin_lock(&bdev->internal.spinlock);
7985 : /*
7986 : * Set the status to REMOVING after completing to abort channels. Otherwise,
7987 : * the last spdk_bdev_close() may call spdk_io_device_unregister() while
7988 : * spdk_bdev_for_each_channel() is executed and spdk_io_device_unregister()
7989 : * may fail.
7990 : */
7991 123 : bdev->internal.status = SPDK_BDEV_STATUS_REMOVING;
7992 123 : rc = bdev_unregister_unsafe(bdev);
7993 123 : spdk_spin_unlock(&bdev->internal.spinlock);
7994 123 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
7995 :
7996 123 : if (rc == 0) {
7997 107 : spdk_io_device_unregister(__bdev_to_io_dev(bdev), bdev_destroy_cb);
7998 : }
7999 123 : }
8000 :
8001 : void
8002 130 : spdk_bdev_unregister(struct spdk_bdev *bdev, spdk_bdev_unregister_cb cb_fn, void *cb_arg)
8003 : {
8004 : struct spdk_thread *thread;
8005 :
8006 130 : SPDK_DEBUGLOG(bdev, "Removing bdev %s from list\n", bdev->name);
8007 :
8008 130 : thread = spdk_get_thread();
8009 130 : if (!thread) {
8010 : /* The user called this from a non-SPDK thread. */
8011 0 : if (cb_fn != NULL) {
8012 0 : cb_fn(cb_arg, -ENOTSUP);
8013 : }
8014 0 : return;
8015 : }
8016 :
8017 130 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8018 130 : if (bdev->internal.status == SPDK_BDEV_STATUS_UNREGISTERING ||
8019 130 : bdev->internal.status == SPDK_BDEV_STATUS_REMOVING) {
8020 7 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8021 7 : if (cb_fn) {
8022 0 : cb_fn(cb_arg, -EBUSY);
8023 : }
8024 7 : return;
8025 : }
8026 :
8027 123 : spdk_spin_lock(&bdev->internal.spinlock);
8028 123 : bdev->internal.status = SPDK_BDEV_STATUS_UNREGISTERING;
8029 123 : bdev->internal.unregister_cb = cb_fn;
8030 123 : bdev->internal.unregister_ctx = cb_arg;
8031 123 : bdev->internal.unregister_td = thread;
8032 123 : spdk_spin_unlock(&bdev->internal.spinlock);
8033 123 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8034 :
8035 123 : spdk_bdev_set_qd_sampling_period(bdev, 0);
8036 :
8037 123 : spdk_bdev_for_each_channel(bdev, bdev_unregister_abort_channel, bdev,
8038 : bdev_unregister);
8039 : }
8040 :
8041 : int
8042 4 : spdk_bdev_unregister_by_name(const char *bdev_name, struct spdk_bdev_module *module,
8043 : spdk_bdev_unregister_cb cb_fn, void *cb_arg)
8044 : {
8045 4 : struct spdk_bdev_desc *desc;
8046 : struct spdk_bdev *bdev;
8047 : int rc;
8048 :
8049 4 : rc = spdk_bdev_open_ext(bdev_name, false, _tmp_bdev_event_cb, NULL, &desc);
8050 4 : if (rc != 0) {
8051 1 : SPDK_ERRLOG("Failed to open bdev with name: %s\n", bdev_name);
8052 1 : return rc;
8053 : }
8054 :
8055 3 : bdev = spdk_bdev_desc_get_bdev(desc);
8056 :
8057 3 : if (bdev->module != module) {
8058 1 : spdk_bdev_close(desc);
8059 1 : SPDK_ERRLOG("Bdev %s was not registered by the specified module.\n",
8060 : bdev_name);
8061 1 : return -ENODEV;
8062 : }
8063 :
8064 2 : spdk_bdev_unregister(bdev, cb_fn, cb_arg);
8065 :
8066 2 : spdk_bdev_close(desc);
8067 :
8068 2 : return 0;
8069 : }
8070 :
8071 : static int
8072 253 : bdev_start_qos(struct spdk_bdev *bdev)
8073 : {
8074 : struct set_qos_limit_ctx *ctx;
8075 :
8076 : /* Enable QoS */
8077 253 : if (bdev->internal.qos && bdev->internal.qos->thread == NULL) {
8078 2 : ctx = calloc(1, sizeof(*ctx));
8079 2 : if (ctx == NULL) {
8080 0 : SPDK_ERRLOG("Failed to allocate memory for QoS context\n");
8081 0 : return -ENOMEM;
8082 : }
8083 2 : ctx->bdev = bdev;
8084 2 : spdk_bdev_for_each_channel(bdev, bdev_enable_qos_msg, ctx, bdev_enable_qos_done);
8085 : }
8086 :
8087 253 : return 0;
8088 : }
8089 :
8090 : static void
8091 24 : log_already_claimed(enum spdk_log_level level, const int line, const char *func, const char *detail,
8092 : struct spdk_bdev *bdev)
8093 : {
8094 : enum spdk_bdev_claim_type type;
8095 : const char *typename, *modname;
8096 : extern struct spdk_log_flag SPDK_LOG_bdev;
8097 :
8098 24 : assert(spdk_spin_held(&bdev->internal.spinlock));
8099 :
8100 24 : if (level >= SPDK_LOG_INFO && !SPDK_LOG_bdev.enabled) {
8101 0 : return;
8102 : }
8103 :
8104 24 : type = bdev->internal.claim_type;
8105 24 : typename = spdk_bdev_claim_get_name(type);
8106 :
8107 24 : if (type == SPDK_BDEV_CLAIM_EXCL_WRITE) {
8108 6 : modname = bdev->internal.claim.v1.module->name;
8109 6 : spdk_log(level, __FILE__, line, func, "bdev %s %s: type %s by module %s\n",
8110 : bdev->name, detail, typename, modname);
8111 6 : return;
8112 : }
8113 :
8114 18 : if (claim_type_is_v2(type)) {
8115 : struct spdk_bdev_module_claim *claim;
8116 :
8117 36 : TAILQ_FOREACH(claim, &bdev->internal.claim.v2.claims, link) {
8118 18 : modname = claim->module->name;
8119 18 : spdk_log(level, __FILE__, line, func, "bdev %s %s: type %s by module %s\n",
8120 : bdev->name, detail, typename, modname);
8121 : }
8122 18 : return;
8123 : }
8124 :
8125 0 : assert(false);
8126 : }
8127 :
8128 : static int
8129 262 : bdev_open(struct spdk_bdev *bdev, bool write, struct spdk_bdev_desc *desc)
8130 : {
8131 : struct spdk_thread *thread;
8132 262 : int rc = 0;
8133 :
8134 262 : thread = spdk_get_thread();
8135 262 : if (!thread) {
8136 0 : SPDK_ERRLOG("Cannot open bdev from non-SPDK thread.\n");
8137 0 : return -ENOTSUP;
8138 : }
8139 :
8140 262 : SPDK_DEBUGLOG(bdev, "Opening descriptor %p for bdev %s on thread %p\n", desc, bdev->name,
8141 : spdk_get_thread());
8142 :
8143 262 : desc->bdev = bdev;
8144 262 : desc->thread = thread;
8145 262 : desc->write = write;
8146 :
8147 262 : spdk_spin_lock(&bdev->internal.spinlock);
8148 262 : if (bdev->internal.status == SPDK_BDEV_STATUS_UNREGISTERING ||
8149 262 : bdev->internal.status == SPDK_BDEV_STATUS_REMOVING) {
8150 3 : spdk_spin_unlock(&bdev->internal.spinlock);
8151 3 : return -ENODEV;
8152 : }
8153 :
8154 259 : if (write && bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE) {
8155 6 : LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
8156 6 : spdk_spin_unlock(&bdev->internal.spinlock);
8157 6 : return -EPERM;
8158 : }
8159 :
8160 253 : rc = bdev_start_qos(bdev);
8161 253 : if (rc != 0) {
8162 0 : SPDK_ERRLOG("Failed to start QoS on bdev %s\n", bdev->name);
8163 0 : spdk_spin_unlock(&bdev->internal.spinlock);
8164 0 : return rc;
8165 : }
8166 :
8167 253 : TAILQ_INSERT_TAIL(&bdev->internal.open_descs, desc, link);
8168 :
8169 253 : spdk_spin_unlock(&bdev->internal.spinlock);
8170 :
8171 253 : return 0;
8172 : }
8173 :
8174 : static int
8175 262 : bdev_desc_alloc(struct spdk_bdev *bdev, spdk_bdev_event_cb_t event_cb, void *event_ctx,
8176 : struct spdk_bdev_desc **_desc)
8177 : {
8178 : struct spdk_bdev_desc *desc;
8179 : unsigned int i;
8180 :
8181 262 : desc = calloc(1, sizeof(*desc));
8182 262 : if (desc == NULL) {
8183 0 : SPDK_ERRLOG("Failed to allocate memory for bdev descriptor\n");
8184 0 : return -ENOMEM;
8185 : }
8186 :
8187 262 : TAILQ_INIT(&desc->pending_media_events);
8188 262 : TAILQ_INIT(&desc->free_media_events);
8189 :
8190 262 : desc->memory_domains_supported = spdk_bdev_get_memory_domains(bdev, NULL, 0) > 0;
8191 262 : desc->callback.event_fn = event_cb;
8192 262 : desc->callback.ctx = event_ctx;
8193 262 : spdk_spin_init(&desc->spinlock);
8194 :
8195 262 : if (bdev->media_events) {
8196 0 : desc->media_events_buffer = calloc(MEDIA_EVENT_POOL_SIZE,
8197 : sizeof(*desc->media_events_buffer));
8198 0 : if (desc->media_events_buffer == NULL) {
8199 0 : SPDK_ERRLOG("Failed to initialize media event pool\n");
8200 0 : bdev_desc_free(desc);
8201 0 : return -ENOMEM;
8202 : }
8203 :
8204 0 : for (i = 0; i < MEDIA_EVENT_POOL_SIZE; ++i) {
8205 0 : TAILQ_INSERT_TAIL(&desc->free_media_events,
8206 : &desc->media_events_buffer[i], tailq);
8207 : }
8208 : }
8209 :
8210 262 : if (bdev->fn_table->accel_sequence_supported != NULL) {
8211 0 : for (i = 0; i < SPDK_BDEV_NUM_IO_TYPES; ++i) {
8212 0 : desc->accel_sequence_supported[i] =
8213 0 : bdev->fn_table->accel_sequence_supported(bdev->ctxt,
8214 : (enum spdk_bdev_io_type)i);
8215 : }
8216 : }
8217 :
8218 262 : *_desc = desc;
8219 :
8220 262 : return 0;
8221 : }
8222 :
8223 : static int
8224 127 : bdev_open_ext(const char *bdev_name, bool write, spdk_bdev_event_cb_t event_cb,
8225 : void *event_ctx, struct spdk_bdev_desc **_desc)
8226 : {
8227 127 : struct spdk_bdev_desc *desc;
8228 : struct spdk_bdev *bdev;
8229 : int rc;
8230 :
8231 127 : bdev = bdev_get_by_name(bdev_name);
8232 :
8233 127 : if (bdev == NULL) {
8234 1 : SPDK_NOTICELOG("Currently unable to find bdev with name: %s\n", bdev_name);
8235 1 : return -ENODEV;
8236 : }
8237 :
8238 126 : rc = bdev_desc_alloc(bdev, event_cb, event_ctx, &desc);
8239 126 : if (rc != 0) {
8240 0 : return rc;
8241 : }
8242 :
8243 126 : rc = bdev_open(bdev, write, desc);
8244 126 : if (rc != 0) {
8245 7 : bdev_desc_free(desc);
8246 7 : desc = NULL;
8247 : }
8248 :
8249 126 : *_desc = desc;
8250 :
8251 126 : return rc;
8252 : }
8253 :
8254 : int
8255 129 : spdk_bdev_open_ext(const char *bdev_name, bool write, spdk_bdev_event_cb_t event_cb,
8256 : void *event_ctx, struct spdk_bdev_desc **_desc)
8257 : {
8258 : int rc;
8259 :
8260 129 : if (event_cb == NULL) {
8261 2 : SPDK_ERRLOG("Missing event callback function\n");
8262 2 : return -EINVAL;
8263 : }
8264 :
8265 127 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8266 127 : rc = bdev_open_ext(bdev_name, write, event_cb, event_ctx, _desc);
8267 127 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8268 :
8269 127 : return rc;
8270 : }
8271 :
8272 : struct spdk_bdev_open_async_ctx {
8273 : char *bdev_name;
8274 : spdk_bdev_event_cb_t event_cb;
8275 : void *event_ctx;
8276 : bool write;
8277 : int rc;
8278 : spdk_bdev_open_async_cb_t cb_fn;
8279 : void *cb_arg;
8280 : struct spdk_bdev_desc *desc;
8281 : struct spdk_bdev_open_async_opts opts;
8282 : uint64_t start_ticks;
8283 : struct spdk_thread *orig_thread;
8284 : struct spdk_poller *poller;
8285 : TAILQ_ENTRY(spdk_bdev_open_async_ctx) tailq;
8286 : };
8287 :
8288 : static void
8289 0 : bdev_open_async_done(void *arg)
8290 : {
8291 0 : struct spdk_bdev_open_async_ctx *ctx = arg;
8292 :
8293 0 : ctx->cb_fn(ctx->desc, ctx->rc, ctx->cb_arg);
8294 :
8295 0 : free(ctx->bdev_name);
8296 0 : free(ctx);
8297 0 : }
8298 :
8299 : static void
8300 0 : bdev_open_async_cancel(void *arg)
8301 : {
8302 0 : struct spdk_bdev_open_async_ctx *ctx = arg;
8303 :
8304 0 : assert(ctx->rc == -ESHUTDOWN);
8305 :
8306 0 : spdk_poller_unregister(&ctx->poller);
8307 :
8308 0 : bdev_open_async_done(ctx);
8309 0 : }
8310 :
8311 : /* This is called when the bdev library finishes at shutdown. */
8312 : static void
8313 64 : bdev_open_async_fini(void)
8314 : {
8315 : struct spdk_bdev_open_async_ctx *ctx, *tmp_ctx;
8316 :
8317 64 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8318 64 : TAILQ_FOREACH_SAFE(ctx, &g_bdev_mgr.async_bdev_opens, tailq, tmp_ctx) {
8319 0 : TAILQ_REMOVE(&g_bdev_mgr.async_bdev_opens, ctx, tailq);
8320 : /*
8321 : * We have to move to ctx->orig_thread to unregister ctx->poller.
8322 : * However, there is a chance that ctx->poller is executed before
8323 : * message is executed, which could result in bdev_open_async_done()
8324 : * being called twice. To avoid such race condition, set ctx->rc to
8325 : * -ESHUTDOWN.
8326 : */
8327 0 : ctx->rc = -ESHUTDOWN;
8328 0 : spdk_thread_send_msg(ctx->orig_thread, bdev_open_async_cancel, ctx);
8329 : }
8330 64 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8331 64 : }
8332 :
8333 : static int bdev_open_async(void *arg);
8334 :
8335 : static void
8336 0 : _bdev_open_async(struct spdk_bdev_open_async_ctx *ctx)
8337 : {
8338 : uint64_t timeout_ticks;
8339 :
8340 0 : if (ctx->rc == -ESHUTDOWN) {
8341 : /* This context is being canceled. Do nothing. */
8342 0 : return;
8343 : }
8344 :
8345 0 : ctx->rc = bdev_open_ext(ctx->bdev_name, ctx->write, ctx->event_cb, ctx->event_ctx,
8346 : &ctx->desc);
8347 0 : if (ctx->rc == 0 || ctx->opts.timeout_ms == 0) {
8348 0 : goto exit;
8349 : }
8350 :
8351 0 : timeout_ticks = ctx->start_ticks + ctx->opts.timeout_ms * spdk_get_ticks_hz() / 1000ull;
8352 0 : if (spdk_get_ticks() >= timeout_ticks) {
8353 0 : SPDK_ERRLOG("Timed out while waiting for bdev '%s' to appear\n", ctx->bdev_name);
8354 0 : ctx->rc = -ETIMEDOUT;
8355 0 : goto exit;
8356 : }
8357 :
8358 0 : return;
8359 :
8360 0 : exit:
8361 0 : spdk_poller_unregister(&ctx->poller);
8362 0 : TAILQ_REMOVE(&g_bdev_mgr.async_bdev_opens, ctx, tailq);
8363 :
8364 : /* Completion callback is processed after stack unwinding. */
8365 0 : spdk_thread_send_msg(ctx->orig_thread, bdev_open_async_done, ctx);
8366 : }
8367 :
8368 : static int
8369 0 : bdev_open_async(void *arg)
8370 : {
8371 0 : struct spdk_bdev_open_async_ctx *ctx = arg;
8372 :
8373 0 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8374 :
8375 0 : _bdev_open_async(ctx);
8376 :
8377 0 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8378 :
8379 0 : return SPDK_POLLER_BUSY;
8380 : }
8381 :
8382 : static void
8383 0 : bdev_open_async_opts_copy(struct spdk_bdev_open_async_opts *opts,
8384 : struct spdk_bdev_open_async_opts *opts_src,
8385 : size_t size)
8386 : {
8387 0 : assert(opts);
8388 0 : assert(opts_src);
8389 :
8390 0 : opts->size = size;
8391 :
8392 : #define SET_FIELD(field) \
8393 : if (offsetof(struct spdk_bdev_open_async_opts, field) + sizeof(opts->field) <= size) { \
8394 : opts->field = opts_src->field; \
8395 : } \
8396 :
8397 0 : SET_FIELD(timeout_ms);
8398 :
8399 : /* Do not remove this statement, you should always update this statement when you adding a new field,
8400 : * and do not forget to add the SET_FIELD statement for your added field. */
8401 : SPDK_STATIC_ASSERT(sizeof(struct spdk_bdev_open_async_opts) == 16, "Incorrect size");
8402 :
8403 : #undef SET_FIELD
8404 0 : }
8405 :
8406 : static void
8407 0 : bdev_open_async_opts_get_default(struct spdk_bdev_open_async_opts *opts, size_t size)
8408 : {
8409 0 : assert(opts);
8410 :
8411 0 : opts->size = size;
8412 :
8413 : #define SET_FIELD(field, value) \
8414 : if (offsetof(struct spdk_bdev_open_async_opts, field) + sizeof(opts->field) <= size) { \
8415 : opts->field = value; \
8416 : } \
8417 :
8418 0 : SET_FIELD(timeout_ms, 0);
8419 :
8420 : #undef SET_FIELD
8421 0 : }
8422 :
8423 : int
8424 0 : spdk_bdev_open_async(const char *bdev_name, bool write, spdk_bdev_event_cb_t event_cb,
8425 : void *event_ctx, struct spdk_bdev_open_async_opts *opts,
8426 : spdk_bdev_open_async_cb_t open_cb, void *open_cb_arg)
8427 : {
8428 : struct spdk_bdev_open_async_ctx *ctx;
8429 :
8430 0 : if (event_cb == NULL) {
8431 0 : SPDK_ERRLOG("Missing event callback function\n");
8432 0 : return -EINVAL;
8433 : }
8434 :
8435 0 : if (open_cb == NULL) {
8436 0 : SPDK_ERRLOG("Missing open callback function\n");
8437 0 : return -EINVAL;
8438 : }
8439 :
8440 0 : if (opts != NULL && opts->size == 0) {
8441 0 : SPDK_ERRLOG("size in the options structure should not be zero\n");
8442 0 : return -EINVAL;
8443 : }
8444 :
8445 0 : ctx = calloc(1, sizeof(*ctx));
8446 0 : if (ctx == NULL) {
8447 0 : SPDK_ERRLOG("Failed to allocate open context\n");
8448 0 : return -ENOMEM;
8449 : }
8450 :
8451 0 : ctx->bdev_name = strdup(bdev_name);
8452 0 : if (ctx->bdev_name == NULL) {
8453 0 : SPDK_ERRLOG("Failed to duplicate bdev_name\n");
8454 0 : free(ctx);
8455 0 : return -ENOMEM;
8456 : }
8457 :
8458 0 : ctx->poller = SPDK_POLLER_REGISTER(bdev_open_async, ctx, 100 * 1000);
8459 0 : if (ctx->poller == NULL) {
8460 0 : SPDK_ERRLOG("Failed to register bdev_open_async poller\n");
8461 0 : free(ctx->bdev_name);
8462 0 : free(ctx);
8463 0 : return -ENOMEM;
8464 : }
8465 :
8466 0 : ctx->cb_fn = open_cb;
8467 0 : ctx->cb_arg = open_cb_arg;
8468 0 : ctx->write = write;
8469 0 : ctx->event_cb = event_cb;
8470 0 : ctx->event_ctx = event_ctx;
8471 0 : ctx->orig_thread = spdk_get_thread();
8472 0 : ctx->start_ticks = spdk_get_ticks();
8473 :
8474 0 : bdev_open_async_opts_get_default(&ctx->opts, sizeof(ctx->opts));
8475 0 : if (opts != NULL) {
8476 0 : bdev_open_async_opts_copy(&ctx->opts, opts, opts->size);
8477 : }
8478 :
8479 0 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8480 :
8481 0 : TAILQ_INSERT_TAIL(&g_bdev_mgr.async_bdev_opens, ctx, tailq);
8482 0 : _bdev_open_async(ctx);
8483 :
8484 0 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8485 :
8486 0 : return 0;
8487 : }
8488 :
8489 : static void
8490 253 : bdev_close(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc)
8491 : {
8492 : int rc;
8493 :
8494 253 : spdk_spin_lock(&bdev->internal.spinlock);
8495 253 : spdk_spin_lock(&desc->spinlock);
8496 :
8497 253 : TAILQ_REMOVE(&bdev->internal.open_descs, desc, link);
8498 :
8499 253 : desc->closed = true;
8500 :
8501 253 : if (desc->claim != NULL) {
8502 16 : bdev_desc_release_claims(desc);
8503 : }
8504 :
8505 253 : if (0 == desc->refs) {
8506 242 : spdk_spin_unlock(&desc->spinlock);
8507 242 : bdev_desc_free(desc);
8508 : } else {
8509 11 : spdk_spin_unlock(&desc->spinlock);
8510 : }
8511 :
8512 : /* If no more descriptors, kill QoS channel */
8513 253 : if (bdev->internal.qos && TAILQ_EMPTY(&bdev->internal.open_descs)) {
8514 7 : SPDK_DEBUGLOG(bdev, "Closed last descriptor for bdev %s on thread %p. Stopping QoS.\n",
8515 : bdev->name, spdk_get_thread());
8516 :
8517 7 : if (bdev_qos_destroy(bdev)) {
8518 : /* There isn't anything we can do to recover here. Just let the
8519 : * old QoS poller keep running. The QoS handling won't change
8520 : * cores when the user allocates a new channel, but it won't break. */
8521 0 : SPDK_ERRLOG("Unable to shut down QoS poller. It will continue running on the current thread.\n");
8522 : }
8523 : }
8524 :
8525 253 : if (bdev->internal.status == SPDK_BDEV_STATUS_REMOVING && TAILQ_EMPTY(&bdev->internal.open_descs)) {
8526 15 : rc = bdev_unregister_unsafe(bdev);
8527 15 : spdk_spin_unlock(&bdev->internal.spinlock);
8528 :
8529 15 : if (rc == 0) {
8530 15 : spdk_io_device_unregister(__bdev_to_io_dev(bdev), bdev_destroy_cb);
8531 : }
8532 : } else {
8533 238 : spdk_spin_unlock(&bdev->internal.spinlock);
8534 : }
8535 253 : }
8536 :
8537 : void
8538 119 : spdk_bdev_close(struct spdk_bdev_desc *desc)
8539 : {
8540 119 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
8541 :
8542 119 : SPDK_DEBUGLOG(bdev, "Closing descriptor %p for bdev %s on thread %p\n", desc, bdev->name,
8543 : spdk_get_thread());
8544 :
8545 119 : assert(desc->thread == spdk_get_thread());
8546 :
8547 119 : spdk_poller_unregister(&desc->io_timeout_poller);
8548 :
8549 119 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8550 :
8551 119 : bdev_close(bdev, desc);
8552 :
8553 119 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8554 119 : }
8555 :
8556 : static void
8557 123 : bdev_register_finished(void *arg)
8558 : {
8559 123 : struct spdk_bdev_desc *desc = arg;
8560 123 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
8561 :
8562 123 : spdk_notify_send("bdev_register", spdk_bdev_get_name(bdev));
8563 :
8564 123 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8565 :
8566 123 : bdev_close(bdev, desc);
8567 :
8568 123 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8569 123 : }
8570 :
8571 : int
8572 126 : spdk_bdev_register(struct spdk_bdev *bdev)
8573 : {
8574 126 : struct spdk_bdev_desc *desc;
8575 126 : struct spdk_thread *thread = spdk_get_thread();
8576 : int rc;
8577 :
8578 126 : if (spdk_unlikely(!spdk_thread_is_app_thread(NULL))) {
8579 1 : SPDK_ERRLOG("Cannot register bdev %s on thread %p (%s)\n", bdev->name, thread,
8580 : thread ? spdk_thread_get_name(thread) : "null");
8581 1 : return -EINVAL;
8582 : }
8583 :
8584 125 : rc = bdev_register(bdev);
8585 125 : if (rc != 0) {
8586 2 : return rc;
8587 : }
8588 :
8589 : /* A descriptor is opened to prevent bdev deletion during examination */
8590 123 : rc = bdev_desc_alloc(bdev, _tmp_bdev_event_cb, NULL, &desc);
8591 123 : if (rc != 0) {
8592 0 : spdk_bdev_unregister(bdev, NULL, NULL);
8593 0 : return rc;
8594 : }
8595 :
8596 123 : rc = bdev_open(bdev, false, desc);
8597 123 : if (rc != 0) {
8598 0 : bdev_desc_free(desc);
8599 0 : spdk_bdev_unregister(bdev, NULL, NULL);
8600 0 : return rc;
8601 : }
8602 :
8603 : /* Examine configuration before initializing I/O */
8604 123 : bdev_examine(bdev);
8605 :
8606 123 : rc = spdk_bdev_wait_for_examine(bdev_register_finished, desc);
8607 123 : if (rc != 0) {
8608 0 : bdev_close(bdev, desc);
8609 0 : spdk_bdev_unregister(bdev, NULL, NULL);
8610 : }
8611 :
8612 123 : return rc;
8613 : }
8614 :
8615 : int
8616 26 : spdk_bdev_module_claim_bdev(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc,
8617 : struct spdk_bdev_module *module)
8618 : {
8619 26 : spdk_spin_lock(&bdev->internal.spinlock);
8620 :
8621 26 : if (bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE) {
8622 6 : LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
8623 6 : spdk_spin_unlock(&bdev->internal.spinlock);
8624 6 : return -EPERM;
8625 : }
8626 :
8627 20 : if (desc && !desc->write) {
8628 5 : desc->write = true;
8629 : }
8630 :
8631 20 : bdev->internal.claim_type = SPDK_BDEV_CLAIM_EXCL_WRITE;
8632 20 : bdev->internal.claim.v1.module = module;
8633 :
8634 20 : spdk_spin_unlock(&bdev->internal.spinlock);
8635 20 : return 0;
8636 : }
8637 :
8638 : void
8639 8 : spdk_bdev_module_release_bdev(struct spdk_bdev *bdev)
8640 : {
8641 8 : spdk_spin_lock(&bdev->internal.spinlock);
8642 :
8643 8 : assert(bdev->internal.claim.v1.module != NULL);
8644 8 : assert(bdev->internal.claim_type == SPDK_BDEV_CLAIM_EXCL_WRITE);
8645 8 : bdev->internal.claim_type = SPDK_BDEV_CLAIM_NONE;
8646 8 : bdev->internal.claim.v1.module = NULL;
8647 :
8648 8 : spdk_spin_unlock(&bdev->internal.spinlock);
8649 8 : }
8650 :
8651 : /*
8652 : * Start claims v2
8653 : */
8654 :
8655 : const char *
8656 24 : spdk_bdev_claim_get_name(enum spdk_bdev_claim_type type)
8657 : {
8658 24 : switch (type) {
8659 0 : case SPDK_BDEV_CLAIM_NONE:
8660 0 : return "not_claimed";
8661 6 : case SPDK_BDEV_CLAIM_EXCL_WRITE:
8662 6 : return "exclusive_write";
8663 7 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE:
8664 7 : return "read_many_write_one";
8665 5 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_NONE:
8666 5 : return "read_many_write_none";
8667 6 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
8668 6 : return "read_many_write_many";
8669 0 : default:
8670 0 : break;
8671 : }
8672 0 : return "invalid_claim";
8673 : }
8674 :
8675 : static bool
8676 96 : claim_type_is_v2(enum spdk_bdev_claim_type type)
8677 : {
8678 96 : switch (type) {
8679 96 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE:
8680 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_NONE:
8681 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
8682 96 : return true;
8683 0 : default:
8684 0 : break;
8685 : }
8686 0 : return false;
8687 : }
8688 :
8689 : /* Returns true if taking a claim with desc->write == false should make the descriptor writable. */
8690 : static bool
8691 13 : claim_type_promotes_to_write(enum spdk_bdev_claim_type type)
8692 : {
8693 13 : switch (type) {
8694 5 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE:
8695 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
8696 5 : return true;
8697 8 : default:
8698 8 : break;
8699 : }
8700 8 : return false;
8701 : }
8702 :
8703 : void
8704 44 : spdk_bdev_claim_opts_init(struct spdk_bdev_claim_opts *opts, size_t size)
8705 : {
8706 44 : if (opts == NULL) {
8707 0 : SPDK_ERRLOG("opts should not be NULL\n");
8708 0 : assert(opts != NULL);
8709 0 : return;
8710 : }
8711 44 : if (size == 0) {
8712 0 : SPDK_ERRLOG("size should not be zero\n");
8713 0 : assert(size != 0);
8714 0 : return;
8715 : }
8716 :
8717 44 : memset(opts, 0, size);
8718 44 : opts->opts_size = size;
8719 :
8720 : #define FIELD_OK(field) \
8721 : offsetof(struct spdk_bdev_claim_opts, field) + sizeof(opts->field) <= size
8722 :
8723 : #define SET_FIELD(field, value) \
8724 : if (FIELD_OK(field)) { \
8725 : opts->field = value; \
8726 : } \
8727 :
8728 44 : SET_FIELD(shared_claim_key, 0);
8729 :
8730 : #undef FIELD_OK
8731 : #undef SET_FIELD
8732 : }
8733 :
8734 : static int
8735 22 : claim_opts_copy(struct spdk_bdev_claim_opts *src, struct spdk_bdev_claim_opts *dst)
8736 : {
8737 22 : if (src->opts_size == 0) {
8738 0 : SPDK_ERRLOG("size should not be zero\n");
8739 0 : return -1;
8740 : }
8741 :
8742 22 : memset(dst, 0, sizeof(*dst));
8743 22 : dst->opts_size = src->opts_size;
8744 :
8745 : #define FIELD_OK(field) \
8746 : offsetof(struct spdk_bdev_claim_opts, field) + sizeof(src->field) <= src->opts_size
8747 :
8748 : #define SET_FIELD(field) \
8749 : if (FIELD_OK(field)) { \
8750 : dst->field = src->field; \
8751 : } \
8752 :
8753 22 : if (FIELD_OK(name)) {
8754 22 : snprintf(dst->name, sizeof(dst->name), "%s", src->name);
8755 : }
8756 :
8757 22 : SET_FIELD(shared_claim_key);
8758 :
8759 : /* You should not remove this statement, but need to update the assert statement
8760 : * if you add a new field, and also add a corresponding SET_FIELD statement */
8761 : SPDK_STATIC_ASSERT(sizeof(struct spdk_bdev_claim_opts) == 48, "Incorrect size");
8762 :
8763 : #undef FIELD_OK
8764 : #undef SET_FIELD
8765 22 : return 0;
8766 : }
8767 :
8768 : /* Returns 0 if a read-write-once claim can be taken. */
8769 : static int
8770 9 : claim_verify_rwo(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
8771 : struct spdk_bdev_claim_opts *opts, struct spdk_bdev_module *module)
8772 : {
8773 9 : struct spdk_bdev *bdev = desc->bdev;
8774 : struct spdk_bdev_desc *open_desc;
8775 :
8776 9 : assert(spdk_spin_held(&bdev->internal.spinlock));
8777 9 : assert(type == SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE);
8778 :
8779 9 : if (opts->shared_claim_key != 0) {
8780 1 : SPDK_ERRLOG("%s: key option not supported with read-write-once claims\n",
8781 : bdev->name);
8782 1 : return -EINVAL;
8783 : }
8784 8 : if (bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE) {
8785 1 : LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
8786 1 : return -EPERM;
8787 : }
8788 7 : if (desc->claim != NULL) {
8789 0 : SPDK_NOTICELOG("%s: descriptor already claimed bdev with module %s\n",
8790 : bdev->name, desc->claim->module->name);
8791 0 : return -EPERM;
8792 : }
8793 14 : TAILQ_FOREACH(open_desc, &bdev->internal.open_descs, link) {
8794 9 : if (desc != open_desc && open_desc->write) {
8795 2 : SPDK_NOTICELOG("%s: Cannot obtain read-write-once claim while "
8796 : "another descriptor is open for writing\n",
8797 : bdev->name);
8798 2 : return -EPERM;
8799 : }
8800 : }
8801 :
8802 5 : return 0;
8803 : }
8804 :
8805 : /* Returns 0 if a read-only-many claim can be taken. */
8806 : static int
8807 12 : claim_verify_rom(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
8808 : struct spdk_bdev_claim_opts *opts, struct spdk_bdev_module *module)
8809 : {
8810 12 : struct spdk_bdev *bdev = desc->bdev;
8811 : struct spdk_bdev_desc *open_desc;
8812 :
8813 12 : assert(spdk_spin_held(&bdev->internal.spinlock));
8814 12 : assert(type == SPDK_BDEV_CLAIM_READ_MANY_WRITE_NONE);
8815 12 : assert(desc->claim == NULL);
8816 :
8817 12 : if (desc->write) {
8818 3 : SPDK_ERRLOG("%s: Cannot obtain read-only-many claim with writable descriptor\n",
8819 : bdev->name);
8820 3 : return -EINVAL;
8821 : }
8822 9 : if (opts->shared_claim_key != 0) {
8823 1 : SPDK_ERRLOG("%s: key option not supported with read-only-may claims\n", bdev->name);
8824 1 : return -EINVAL;
8825 : }
8826 8 : if (bdev->internal.claim_type == SPDK_BDEV_CLAIM_NONE) {
8827 15 : TAILQ_FOREACH(open_desc, &bdev->internal.open_descs, link) {
8828 9 : if (open_desc->write) {
8829 0 : SPDK_NOTICELOG("%s: Cannot obtain read-only-many claim while "
8830 : "another descriptor is open for writing\n",
8831 : bdev->name);
8832 0 : return -EPERM;
8833 : }
8834 : }
8835 : }
8836 :
8837 8 : return 0;
8838 : }
8839 :
8840 : /* Returns 0 if a read-write-many claim can be taken. */
8841 : static int
8842 8 : claim_verify_rwm(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
8843 : struct spdk_bdev_claim_opts *opts, struct spdk_bdev_module *module)
8844 : {
8845 8 : struct spdk_bdev *bdev = desc->bdev;
8846 : struct spdk_bdev_desc *open_desc;
8847 :
8848 8 : assert(spdk_spin_held(&bdev->internal.spinlock));
8849 8 : assert(type == SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED);
8850 8 : assert(desc->claim == NULL);
8851 :
8852 8 : if (opts->shared_claim_key == 0) {
8853 2 : SPDK_ERRLOG("%s: shared_claim_key option required with read-write-may claims\n",
8854 : bdev->name);
8855 2 : return -EINVAL;
8856 : }
8857 6 : switch (bdev->internal.claim_type) {
8858 4 : case SPDK_BDEV_CLAIM_NONE:
8859 7 : TAILQ_FOREACH(open_desc, &bdev->internal.open_descs, link) {
8860 5 : if (open_desc == desc) {
8861 3 : continue;
8862 : }
8863 2 : if (open_desc->write) {
8864 2 : SPDK_NOTICELOG("%s: Cannot obtain read-write-many claim while "
8865 : "another descriptor is open for writing without a "
8866 : "claim\n", bdev->name);
8867 2 : return -EPERM;
8868 : }
8869 : }
8870 2 : break;
8871 2 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
8872 2 : if (opts->shared_claim_key != bdev->internal.claim.v2.key) {
8873 1 : LOG_ALREADY_CLAIMED_ERROR("already claimed with another key", bdev);
8874 1 : return -EPERM;
8875 : }
8876 1 : break;
8877 0 : default:
8878 0 : LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
8879 0 : return -EBUSY;
8880 : }
8881 :
8882 3 : return 0;
8883 : }
8884 :
8885 : /* Updates desc and its bdev with a v2 claim. */
8886 : static int
8887 16 : claim_bdev(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
8888 : struct spdk_bdev_claim_opts *opts, struct spdk_bdev_module *module)
8889 : {
8890 16 : struct spdk_bdev *bdev = desc->bdev;
8891 : struct spdk_bdev_module_claim *claim;
8892 :
8893 16 : assert(spdk_spin_held(&bdev->internal.spinlock));
8894 16 : assert(claim_type_is_v2(type));
8895 16 : assert(desc->claim == NULL);
8896 :
8897 16 : claim = calloc(1, sizeof(*desc->claim));
8898 16 : if (claim == NULL) {
8899 0 : SPDK_ERRLOG("%s: out of memory while allocating claim\n", bdev->name);
8900 0 : return -ENOMEM;
8901 : }
8902 16 : claim->module = module;
8903 16 : claim->desc = desc;
8904 : SPDK_STATIC_ASSERT(sizeof(claim->name) == sizeof(opts->name), "sizes must match");
8905 16 : memcpy(claim->name, opts->name, sizeof(claim->name));
8906 16 : desc->claim = claim;
8907 :
8908 16 : if (bdev->internal.claim_type == SPDK_BDEV_CLAIM_NONE) {
8909 13 : bdev->internal.claim_type = type;
8910 13 : TAILQ_INIT(&bdev->internal.claim.v2.claims);
8911 13 : bdev->internal.claim.v2.key = opts->shared_claim_key;
8912 : }
8913 16 : assert(type == bdev->internal.claim_type);
8914 :
8915 16 : TAILQ_INSERT_TAIL(&bdev->internal.claim.v2.claims, claim, link);
8916 :
8917 16 : if (!desc->write && claim_type_promotes_to_write(type)) {
8918 5 : desc->write = true;
8919 : }
8920 :
8921 16 : return 0;
8922 : }
8923 :
8924 : int
8925 39 : spdk_bdev_module_claim_bdev_desc(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
8926 : struct spdk_bdev_claim_opts *_opts,
8927 : struct spdk_bdev_module *module)
8928 : {
8929 : struct spdk_bdev *bdev;
8930 39 : struct spdk_bdev_claim_opts opts;
8931 39 : int rc = 0;
8932 :
8933 39 : if (desc == NULL) {
8934 0 : SPDK_ERRLOG("descriptor must not be NULL\n");
8935 0 : return -EINVAL;
8936 : }
8937 :
8938 39 : bdev = desc->bdev;
8939 :
8940 39 : if (_opts == NULL) {
8941 17 : spdk_bdev_claim_opts_init(&opts, sizeof(opts));
8942 22 : } else if (claim_opts_copy(_opts, &opts) != 0) {
8943 0 : return -EINVAL;
8944 : }
8945 :
8946 39 : spdk_spin_lock(&bdev->internal.spinlock);
8947 :
8948 39 : if (bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE &&
8949 15 : bdev->internal.claim_type != type) {
8950 10 : LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
8951 10 : spdk_spin_unlock(&bdev->internal.spinlock);
8952 10 : return -EPERM;
8953 : }
8954 :
8955 29 : if (claim_type_is_v2(type) && desc->claim != NULL) {
8956 0 : SPDK_ERRLOG("%s: descriptor already has %s claim with name '%s'\n",
8957 : bdev->name, spdk_bdev_claim_get_name(type), desc->claim->name);
8958 0 : spdk_spin_unlock(&bdev->internal.spinlock);
8959 0 : return -EPERM;
8960 : }
8961 :
8962 29 : switch (type) {
8963 0 : case SPDK_BDEV_CLAIM_EXCL_WRITE:
8964 0 : spdk_spin_unlock(&bdev->internal.spinlock);
8965 0 : return spdk_bdev_module_claim_bdev(bdev, desc, module);
8966 9 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE:
8967 9 : rc = claim_verify_rwo(desc, type, &opts, module);
8968 9 : break;
8969 12 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_NONE:
8970 12 : rc = claim_verify_rom(desc, type, &opts, module);
8971 12 : break;
8972 8 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
8973 8 : rc = claim_verify_rwm(desc, type, &opts, module);
8974 8 : break;
8975 0 : default:
8976 0 : SPDK_ERRLOG("%s: claim type %d not supported\n", bdev->name, type);
8977 0 : rc = -ENOTSUP;
8978 : }
8979 :
8980 29 : if (rc == 0) {
8981 16 : rc = claim_bdev(desc, type, &opts, module);
8982 : }
8983 :
8984 29 : spdk_spin_unlock(&bdev->internal.spinlock);
8985 29 : return rc;
8986 : }
8987 :
8988 : static void
8989 13 : claim_reset(struct spdk_bdev *bdev)
8990 : {
8991 13 : assert(spdk_spin_held(&bdev->internal.spinlock));
8992 13 : assert(claim_type_is_v2(bdev->internal.claim_type));
8993 13 : assert(TAILQ_EMPTY(&bdev->internal.claim.v2.claims));
8994 :
8995 13 : memset(&bdev->internal.claim, 0, sizeof(bdev->internal.claim));
8996 13 : bdev->internal.claim_type = SPDK_BDEV_CLAIM_NONE;
8997 13 : }
8998 :
8999 : static void
9000 16 : bdev_desc_release_claims(struct spdk_bdev_desc *desc)
9001 : {
9002 16 : struct spdk_bdev *bdev = desc->bdev;
9003 :
9004 16 : assert(spdk_spin_held(&bdev->internal.spinlock));
9005 16 : assert(claim_type_is_v2(bdev->internal.claim_type));
9006 :
9007 16 : if (bdev->internal.examine_in_progress == 0) {
9008 16 : TAILQ_REMOVE(&bdev->internal.claim.v2.claims, desc->claim, link);
9009 16 : free(desc->claim);
9010 16 : if (TAILQ_EMPTY(&bdev->internal.claim.v2.claims)) {
9011 13 : claim_reset(bdev);
9012 : }
9013 : } else {
9014 : /* This is a dead claim that will be cleaned up when bdev_examine() is done. */
9015 0 : desc->claim->module = NULL;
9016 0 : desc->claim->desc = NULL;
9017 : }
9018 16 : desc->claim = NULL;
9019 16 : }
9020 :
9021 : /*
9022 : * End claims v2
9023 : */
9024 :
9025 : struct spdk_bdev *
9026 1163 : spdk_bdev_desc_get_bdev(struct spdk_bdev_desc *desc)
9027 : {
9028 1163 : assert(desc != NULL);
9029 1163 : return desc->bdev;
9030 : }
9031 :
9032 : int
9033 1 : spdk_for_each_bdev(void *ctx, spdk_for_each_bdev_fn fn)
9034 : {
9035 : struct spdk_bdev *bdev, *tmp;
9036 1 : struct spdk_bdev_desc *desc;
9037 1 : int rc = 0;
9038 :
9039 1 : assert(fn != NULL);
9040 :
9041 1 : spdk_spin_lock(&g_bdev_mgr.spinlock);
9042 1 : bdev = spdk_bdev_first();
9043 9 : while (bdev != NULL) {
9044 8 : rc = bdev_desc_alloc(bdev, _tmp_bdev_event_cb, NULL, &desc);
9045 8 : if (rc != 0) {
9046 0 : break;
9047 : }
9048 8 : rc = bdev_open(bdev, false, desc);
9049 8 : if (rc != 0) {
9050 1 : bdev_desc_free(desc);
9051 1 : if (rc == -ENODEV) {
9052 : /* Ignore the error and move to the next bdev. */
9053 1 : rc = 0;
9054 1 : bdev = spdk_bdev_next(bdev);
9055 1 : continue;
9056 : }
9057 0 : break;
9058 : }
9059 7 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
9060 :
9061 7 : rc = fn(ctx, bdev);
9062 :
9063 7 : spdk_spin_lock(&g_bdev_mgr.spinlock);
9064 7 : tmp = spdk_bdev_next(bdev);
9065 7 : bdev_close(bdev, desc);
9066 7 : if (rc != 0) {
9067 0 : break;
9068 : }
9069 7 : bdev = tmp;
9070 : }
9071 1 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
9072 :
9073 1 : return rc;
9074 : }
9075 :
9076 : int
9077 1 : spdk_for_each_bdev_leaf(void *ctx, spdk_for_each_bdev_fn fn)
9078 : {
9079 : struct spdk_bdev *bdev, *tmp;
9080 1 : struct spdk_bdev_desc *desc;
9081 1 : int rc = 0;
9082 :
9083 1 : assert(fn != NULL);
9084 :
9085 1 : spdk_spin_lock(&g_bdev_mgr.spinlock);
9086 1 : bdev = spdk_bdev_first_leaf();
9087 6 : while (bdev != NULL) {
9088 5 : rc = bdev_desc_alloc(bdev, _tmp_bdev_event_cb, NULL, &desc);
9089 5 : if (rc != 0) {
9090 0 : break;
9091 : }
9092 5 : rc = bdev_open(bdev, false, desc);
9093 5 : if (rc != 0) {
9094 1 : bdev_desc_free(desc);
9095 1 : if (rc == -ENODEV) {
9096 : /* Ignore the error and move to the next bdev. */
9097 1 : rc = 0;
9098 1 : bdev = spdk_bdev_next_leaf(bdev);
9099 1 : continue;
9100 : }
9101 0 : break;
9102 : }
9103 4 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
9104 :
9105 4 : rc = fn(ctx, bdev);
9106 :
9107 4 : spdk_spin_lock(&g_bdev_mgr.spinlock);
9108 4 : tmp = spdk_bdev_next_leaf(bdev);
9109 4 : bdev_close(bdev, desc);
9110 4 : if (rc != 0) {
9111 0 : break;
9112 : }
9113 4 : bdev = tmp;
9114 : }
9115 1 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
9116 :
9117 1 : return rc;
9118 : }
9119 :
9120 : void
9121 0 : spdk_bdev_io_get_iovec(struct spdk_bdev_io *bdev_io, struct iovec **iovp, int *iovcntp)
9122 : {
9123 : struct iovec *iovs;
9124 : int iovcnt;
9125 :
9126 0 : if (bdev_io == NULL) {
9127 0 : return;
9128 : }
9129 :
9130 0 : switch (bdev_io->type) {
9131 0 : case SPDK_BDEV_IO_TYPE_READ:
9132 : case SPDK_BDEV_IO_TYPE_WRITE:
9133 : case SPDK_BDEV_IO_TYPE_ZCOPY:
9134 0 : iovs = bdev_io->u.bdev.iovs;
9135 0 : iovcnt = bdev_io->u.bdev.iovcnt;
9136 0 : break;
9137 0 : default:
9138 0 : iovs = NULL;
9139 0 : iovcnt = 0;
9140 0 : break;
9141 : }
9142 :
9143 0 : if (iovp) {
9144 0 : *iovp = iovs;
9145 : }
9146 0 : if (iovcntp) {
9147 0 : *iovcntp = iovcnt;
9148 : }
9149 : }
9150 :
9151 : void *
9152 0 : spdk_bdev_io_get_md_buf(struct spdk_bdev_io *bdev_io)
9153 : {
9154 0 : if (bdev_io == NULL) {
9155 0 : return NULL;
9156 : }
9157 :
9158 0 : if (!spdk_bdev_is_md_separate(bdev_io->bdev)) {
9159 0 : return NULL;
9160 : }
9161 :
9162 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_READ ||
9163 0 : bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
9164 0 : return bdev_io->u.bdev.md_buf;
9165 : }
9166 :
9167 0 : return NULL;
9168 : }
9169 :
9170 : void *
9171 0 : spdk_bdev_io_get_cb_arg(struct spdk_bdev_io *bdev_io)
9172 : {
9173 0 : if (bdev_io == NULL) {
9174 0 : assert(false);
9175 : return NULL;
9176 : }
9177 :
9178 0 : return bdev_io->internal.caller_ctx;
9179 : }
9180 :
9181 : void
9182 7 : spdk_bdev_module_list_add(struct spdk_bdev_module *bdev_module)
9183 : {
9184 :
9185 7 : if (spdk_bdev_module_list_find(bdev_module->name)) {
9186 0 : SPDK_ERRLOG("ERROR: module '%s' already registered.\n", bdev_module->name);
9187 0 : assert(false);
9188 : }
9189 :
9190 7 : spdk_spin_init(&bdev_module->internal.spinlock);
9191 7 : TAILQ_INIT(&bdev_module->internal.quiesced_ranges);
9192 :
9193 : /*
9194 : * Modules with examine callbacks must be initialized first, so they are
9195 : * ready to handle examine callbacks from later modules that will
9196 : * register physical bdevs.
9197 : */
9198 7 : if (bdev_module->examine_config != NULL || bdev_module->examine_disk != NULL) {
9199 4 : TAILQ_INSERT_HEAD(&g_bdev_mgr.bdev_modules, bdev_module, internal.tailq);
9200 : } else {
9201 3 : TAILQ_INSERT_TAIL(&g_bdev_mgr.bdev_modules, bdev_module, internal.tailq);
9202 : }
9203 7 : }
9204 :
9205 : struct spdk_bdev_module *
9206 7 : spdk_bdev_module_list_find(const char *name)
9207 : {
9208 : struct spdk_bdev_module *bdev_module;
9209 :
9210 14 : TAILQ_FOREACH(bdev_module, &g_bdev_mgr.bdev_modules, internal.tailq) {
9211 7 : if (strcmp(name, bdev_module->name) == 0) {
9212 0 : break;
9213 : }
9214 : }
9215 :
9216 7 : return bdev_module;
9217 : }
9218 :
9219 : static int
9220 6 : bdev_write_zero_buffer(struct spdk_bdev_io *bdev_io)
9221 : {
9222 : uint64_t num_blocks;
9223 6 : void *md_buf = NULL;
9224 :
9225 6 : num_blocks = bdev_io->u.bdev.num_blocks;
9226 :
9227 6 : if (spdk_bdev_is_md_separate(bdev_io->bdev)) {
9228 4 : md_buf = (char *)g_bdev_mgr.zero_buffer +
9229 2 : spdk_bdev_get_block_size(bdev_io->bdev) * num_blocks;
9230 : }
9231 :
9232 6 : return bdev_write_blocks_with_md(bdev_io->internal.desc,
9233 6 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
9234 : g_bdev_mgr.zero_buffer, md_buf,
9235 : bdev_io->u.bdev.offset_blocks, num_blocks,
9236 : bdev_write_zero_buffer_done, bdev_io);
9237 : }
9238 :
9239 : static void
9240 6 : bdev_write_zero_buffer_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
9241 : {
9242 6 : struct spdk_bdev_io *parent_io = cb_arg;
9243 :
9244 6 : spdk_bdev_free_io(bdev_io);
9245 :
9246 6 : parent_io->internal.status = success ? SPDK_BDEV_IO_STATUS_SUCCESS : SPDK_BDEV_IO_STATUS_FAILED;
9247 6 : parent_io->internal.cb(parent_io, success, parent_io->internal.caller_ctx);
9248 6 : }
9249 :
9250 : static void
9251 10 : bdev_set_qos_limit_done(struct set_qos_limit_ctx *ctx, int status)
9252 : {
9253 10 : spdk_spin_lock(&ctx->bdev->internal.spinlock);
9254 10 : ctx->bdev->internal.qos_mod_in_progress = false;
9255 10 : spdk_spin_unlock(&ctx->bdev->internal.spinlock);
9256 :
9257 10 : if (ctx->cb_fn) {
9258 8 : ctx->cb_fn(ctx->cb_arg, status);
9259 : }
9260 10 : free(ctx);
9261 10 : }
9262 :
9263 : static void
9264 2 : bdev_disable_qos_done(void *cb_arg)
9265 : {
9266 2 : struct set_qos_limit_ctx *ctx = cb_arg;
9267 2 : struct spdk_bdev *bdev = ctx->bdev;
9268 : struct spdk_bdev_qos *qos;
9269 :
9270 2 : spdk_spin_lock(&bdev->internal.spinlock);
9271 2 : qos = bdev->internal.qos;
9272 2 : bdev->internal.qos = NULL;
9273 2 : spdk_spin_unlock(&bdev->internal.spinlock);
9274 :
9275 2 : if (qos->thread != NULL) {
9276 2 : spdk_put_io_channel(spdk_io_channel_from_ctx(qos->ch));
9277 2 : spdk_poller_unregister(&qos->poller);
9278 : }
9279 :
9280 2 : free(qos);
9281 :
9282 2 : bdev_set_qos_limit_done(ctx, 0);
9283 2 : }
9284 :
9285 : static void
9286 2 : bdev_disable_qos_msg_done(struct spdk_bdev *bdev, void *_ctx, int status)
9287 : {
9288 2 : struct set_qos_limit_ctx *ctx = _ctx;
9289 : struct spdk_thread *thread;
9290 :
9291 2 : spdk_spin_lock(&bdev->internal.spinlock);
9292 2 : thread = bdev->internal.qos->thread;
9293 2 : spdk_spin_unlock(&bdev->internal.spinlock);
9294 :
9295 2 : if (thread != NULL) {
9296 2 : spdk_thread_send_msg(thread, bdev_disable_qos_done, ctx);
9297 : } else {
9298 0 : bdev_disable_qos_done(ctx);
9299 : }
9300 2 : }
9301 :
9302 : static void
9303 4 : bdev_disable_qos_msg(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
9304 : struct spdk_io_channel *ch, void *_ctx)
9305 : {
9306 4 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(ch);
9307 : struct spdk_bdev_io *bdev_io;
9308 :
9309 4 : bdev_ch->flags &= ~BDEV_CH_QOS_ENABLED;
9310 :
9311 6 : while (!TAILQ_EMPTY(&bdev_ch->qos_queued_io)) {
9312 : /* Re-submit the queued I/O. */
9313 2 : bdev_io = TAILQ_FIRST(&bdev_ch->qos_queued_io);
9314 2 : TAILQ_REMOVE(&bdev_ch->qos_queued_io, bdev_io, internal.link);
9315 2 : _bdev_io_submit(bdev_io);
9316 : }
9317 :
9318 4 : spdk_bdev_for_each_channel_continue(i, 0);
9319 4 : }
9320 :
9321 : static void
9322 1 : bdev_update_qos_rate_limit_msg(void *cb_arg)
9323 : {
9324 1 : struct set_qos_limit_ctx *ctx = cb_arg;
9325 1 : struct spdk_bdev *bdev = ctx->bdev;
9326 :
9327 1 : spdk_spin_lock(&bdev->internal.spinlock);
9328 1 : bdev_qos_update_max_quota_per_timeslice(bdev->internal.qos);
9329 1 : spdk_spin_unlock(&bdev->internal.spinlock);
9330 :
9331 1 : bdev_set_qos_limit_done(ctx, 0);
9332 1 : }
9333 :
9334 : static void
9335 9 : bdev_enable_qos_msg(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
9336 : struct spdk_io_channel *ch, void *_ctx)
9337 : {
9338 9 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(ch);
9339 :
9340 9 : spdk_spin_lock(&bdev->internal.spinlock);
9341 9 : bdev_enable_qos(bdev, bdev_ch);
9342 9 : spdk_spin_unlock(&bdev->internal.spinlock);
9343 9 : spdk_bdev_for_each_channel_continue(i, 0);
9344 9 : }
9345 :
9346 : static void
9347 6 : bdev_enable_qos_done(struct spdk_bdev *bdev, void *_ctx, int status)
9348 : {
9349 6 : struct set_qos_limit_ctx *ctx = _ctx;
9350 :
9351 6 : bdev_set_qos_limit_done(ctx, status);
9352 6 : }
9353 :
9354 : static void
9355 7 : bdev_set_qos_rate_limits(struct spdk_bdev *bdev, uint64_t *limits)
9356 : {
9357 : int i;
9358 :
9359 7 : assert(bdev->internal.qos != NULL);
9360 :
9361 35 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
9362 28 : if (limits[i] != SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
9363 28 : bdev->internal.qos->rate_limits[i].limit = limits[i];
9364 :
9365 28 : if (limits[i] == 0) {
9366 19 : bdev->internal.qos->rate_limits[i].limit =
9367 : SPDK_BDEV_QOS_LIMIT_NOT_DEFINED;
9368 : }
9369 : }
9370 : }
9371 7 : }
9372 :
9373 : void
9374 9 : spdk_bdev_set_qos_rate_limits(struct spdk_bdev *bdev, uint64_t *limits,
9375 : void (*cb_fn)(void *cb_arg, int status), void *cb_arg)
9376 : {
9377 : struct set_qos_limit_ctx *ctx;
9378 : uint32_t limit_set_complement;
9379 : uint64_t min_limit_per_sec;
9380 : int i;
9381 9 : bool disable_rate_limit = true;
9382 :
9383 45 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
9384 36 : if (limits[i] == SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
9385 0 : continue;
9386 : }
9387 :
9388 36 : if (limits[i] > 0) {
9389 10 : disable_rate_limit = false;
9390 : }
9391 :
9392 36 : if (bdev_qos_is_iops_rate_limit(i) == true) {
9393 9 : min_limit_per_sec = SPDK_BDEV_QOS_MIN_IOS_PER_SEC;
9394 : } else {
9395 27 : if (limits[i] > SPDK_BDEV_QOS_MAX_MBYTES_PER_SEC) {
9396 0 : SPDK_WARNLOG("Requested rate limit %" PRIu64 " will result in uint64_t overflow, "
9397 : "reset to %" PRIu64 "\n", limits[i], SPDK_BDEV_QOS_MAX_MBYTES_PER_SEC);
9398 0 : limits[i] = SPDK_BDEV_QOS_MAX_MBYTES_PER_SEC;
9399 : }
9400 : /* Change from megabyte to byte rate limit */
9401 27 : limits[i] = limits[i] * 1024 * 1024;
9402 27 : min_limit_per_sec = SPDK_BDEV_QOS_MIN_BYTES_PER_SEC;
9403 : }
9404 :
9405 36 : limit_set_complement = limits[i] % min_limit_per_sec;
9406 36 : if (limit_set_complement) {
9407 0 : SPDK_ERRLOG("Requested rate limit %" PRIu64 " is not a multiple of %" PRIu64 "\n",
9408 : limits[i], min_limit_per_sec);
9409 0 : limits[i] += min_limit_per_sec - limit_set_complement;
9410 0 : SPDK_ERRLOG("Round up the rate limit to %" PRIu64 "\n", limits[i]);
9411 : }
9412 : }
9413 :
9414 9 : ctx = calloc(1, sizeof(*ctx));
9415 9 : if (ctx == NULL) {
9416 0 : cb_fn(cb_arg, -ENOMEM);
9417 0 : return;
9418 : }
9419 :
9420 9 : ctx->cb_fn = cb_fn;
9421 9 : ctx->cb_arg = cb_arg;
9422 9 : ctx->bdev = bdev;
9423 :
9424 9 : spdk_spin_lock(&bdev->internal.spinlock);
9425 9 : if (bdev->internal.qos_mod_in_progress) {
9426 1 : spdk_spin_unlock(&bdev->internal.spinlock);
9427 1 : free(ctx);
9428 1 : cb_fn(cb_arg, -EAGAIN);
9429 1 : return;
9430 : }
9431 8 : bdev->internal.qos_mod_in_progress = true;
9432 :
9433 8 : if (disable_rate_limit == true && bdev->internal.qos) {
9434 10 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
9435 8 : if (limits[i] == SPDK_BDEV_QOS_LIMIT_NOT_DEFINED &&
9436 0 : (bdev->internal.qos->rate_limits[i].limit > 0 &&
9437 0 : bdev->internal.qos->rate_limits[i].limit !=
9438 : SPDK_BDEV_QOS_LIMIT_NOT_DEFINED)) {
9439 0 : disable_rate_limit = false;
9440 0 : break;
9441 : }
9442 : }
9443 : }
9444 :
9445 8 : if (disable_rate_limit == false) {
9446 5 : if (bdev->internal.qos == NULL) {
9447 4 : bdev->internal.qos = calloc(1, sizeof(*bdev->internal.qos));
9448 4 : if (!bdev->internal.qos) {
9449 0 : spdk_spin_unlock(&bdev->internal.spinlock);
9450 0 : SPDK_ERRLOG("Unable to allocate memory for QoS tracking\n");
9451 0 : bdev_set_qos_limit_done(ctx, -ENOMEM);
9452 0 : return;
9453 : }
9454 : }
9455 :
9456 5 : if (bdev->internal.qos->thread == NULL) {
9457 : /* Enabling */
9458 4 : bdev_set_qos_rate_limits(bdev, limits);
9459 :
9460 4 : spdk_bdev_for_each_channel(bdev, bdev_enable_qos_msg, ctx,
9461 : bdev_enable_qos_done);
9462 : } else {
9463 : /* Updating */
9464 1 : bdev_set_qos_rate_limits(bdev, limits);
9465 :
9466 1 : spdk_thread_send_msg(bdev->internal.qos->thread,
9467 : bdev_update_qos_rate_limit_msg, ctx);
9468 : }
9469 : } else {
9470 3 : if (bdev->internal.qos != NULL) {
9471 2 : bdev_set_qos_rate_limits(bdev, limits);
9472 :
9473 : /* Disabling */
9474 2 : spdk_bdev_for_each_channel(bdev, bdev_disable_qos_msg, ctx,
9475 : bdev_disable_qos_msg_done);
9476 : } else {
9477 1 : spdk_spin_unlock(&bdev->internal.spinlock);
9478 1 : bdev_set_qos_limit_done(ctx, 0);
9479 1 : return;
9480 : }
9481 : }
9482 :
9483 7 : spdk_spin_unlock(&bdev->internal.spinlock);
9484 : }
9485 :
9486 : struct spdk_bdev_histogram_ctx {
9487 : spdk_bdev_histogram_status_cb cb_fn;
9488 : void *cb_arg;
9489 : struct spdk_bdev *bdev;
9490 : int status;
9491 : };
9492 :
9493 : static void
9494 2 : bdev_histogram_disable_channel_cb(struct spdk_bdev *bdev, void *_ctx, int status)
9495 : {
9496 2 : struct spdk_bdev_histogram_ctx *ctx = _ctx;
9497 :
9498 2 : spdk_spin_lock(&ctx->bdev->internal.spinlock);
9499 2 : ctx->bdev->internal.histogram_in_progress = false;
9500 2 : spdk_spin_unlock(&ctx->bdev->internal.spinlock);
9501 2 : ctx->cb_fn(ctx->cb_arg, ctx->status);
9502 2 : free(ctx);
9503 2 : }
9504 :
9505 : static void
9506 3 : bdev_histogram_disable_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
9507 : struct spdk_io_channel *_ch, void *_ctx)
9508 : {
9509 3 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
9510 :
9511 3 : if (ch->histogram != NULL) {
9512 3 : spdk_histogram_data_free(ch->histogram);
9513 3 : ch->histogram = NULL;
9514 : }
9515 3 : spdk_bdev_for_each_channel_continue(i, 0);
9516 3 : }
9517 :
9518 : static void
9519 2 : bdev_histogram_enable_channel_cb(struct spdk_bdev *bdev, void *_ctx, int status)
9520 : {
9521 2 : struct spdk_bdev_histogram_ctx *ctx = _ctx;
9522 :
9523 2 : if (status != 0) {
9524 0 : ctx->status = status;
9525 0 : ctx->bdev->internal.histogram_enabled = false;
9526 0 : spdk_bdev_for_each_channel(ctx->bdev, bdev_histogram_disable_channel, ctx,
9527 : bdev_histogram_disable_channel_cb);
9528 : } else {
9529 2 : spdk_spin_lock(&ctx->bdev->internal.spinlock);
9530 2 : ctx->bdev->internal.histogram_in_progress = false;
9531 2 : spdk_spin_unlock(&ctx->bdev->internal.spinlock);
9532 2 : ctx->cb_fn(ctx->cb_arg, ctx->status);
9533 2 : free(ctx);
9534 : }
9535 2 : }
9536 :
9537 : static void
9538 3 : bdev_histogram_enable_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
9539 : struct spdk_io_channel *_ch, void *_ctx)
9540 : {
9541 3 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
9542 3 : int status = 0;
9543 :
9544 3 : if (ch->histogram == NULL) {
9545 3 : ch->histogram = spdk_histogram_data_alloc();
9546 3 : if (ch->histogram == NULL) {
9547 0 : status = -ENOMEM;
9548 : }
9549 : }
9550 :
9551 3 : spdk_bdev_for_each_channel_continue(i, status);
9552 3 : }
9553 :
9554 : void
9555 4 : spdk_bdev_histogram_enable_ext(struct spdk_bdev *bdev, spdk_bdev_histogram_status_cb cb_fn,
9556 : void *cb_arg, bool enable, struct spdk_bdev_enable_histogram_opts *opts)
9557 : {
9558 : struct spdk_bdev_histogram_ctx *ctx;
9559 :
9560 4 : ctx = calloc(1, sizeof(struct spdk_bdev_histogram_ctx));
9561 4 : if (ctx == NULL) {
9562 0 : cb_fn(cb_arg, -ENOMEM);
9563 0 : return;
9564 : }
9565 :
9566 4 : ctx->bdev = bdev;
9567 4 : ctx->status = 0;
9568 4 : ctx->cb_fn = cb_fn;
9569 4 : ctx->cb_arg = cb_arg;
9570 :
9571 4 : spdk_spin_lock(&bdev->internal.spinlock);
9572 4 : if (bdev->internal.histogram_in_progress) {
9573 0 : spdk_spin_unlock(&bdev->internal.spinlock);
9574 0 : free(ctx);
9575 0 : cb_fn(cb_arg, -EAGAIN);
9576 0 : return;
9577 : }
9578 :
9579 4 : bdev->internal.histogram_in_progress = true;
9580 4 : spdk_spin_unlock(&bdev->internal.spinlock);
9581 :
9582 4 : bdev->internal.histogram_enabled = enable;
9583 4 : bdev->internal.histogram_io_type = opts->io_type;
9584 :
9585 4 : if (enable) {
9586 : /* Allocate histogram for each channel */
9587 2 : spdk_bdev_for_each_channel(bdev, bdev_histogram_enable_channel, ctx,
9588 : bdev_histogram_enable_channel_cb);
9589 : } else {
9590 2 : spdk_bdev_for_each_channel(bdev, bdev_histogram_disable_channel, ctx,
9591 : bdev_histogram_disable_channel_cb);
9592 : }
9593 : }
9594 :
9595 : void
9596 4 : spdk_bdev_enable_histogram_opts_init(struct spdk_bdev_enable_histogram_opts *opts, size_t size)
9597 : {
9598 4 : if (opts == NULL) {
9599 0 : SPDK_ERRLOG("opts should not be NULL\n");
9600 0 : assert(opts != NULL);
9601 0 : return;
9602 : }
9603 4 : if (size == 0) {
9604 0 : SPDK_ERRLOG("size should not be zero\n");
9605 0 : assert(size != 0);
9606 0 : return;
9607 : }
9608 :
9609 4 : memset(opts, 0, size);
9610 4 : opts->size = size;
9611 :
9612 : #define FIELD_OK(field) \
9613 : offsetof(struct spdk_bdev_enable_histogram_opts, field) + sizeof(opts->field) <= size
9614 :
9615 : #define SET_FIELD(field, value) \
9616 : if (FIELD_OK(field)) { \
9617 : opts->field = value; \
9618 : } \
9619 :
9620 4 : SET_FIELD(io_type, 0);
9621 :
9622 : /* You should not remove this statement, but need to update the assert statement
9623 : * if you add a new field, and also add a corresponding SET_FIELD statement */
9624 : SPDK_STATIC_ASSERT(sizeof(struct spdk_bdev_enable_histogram_opts) == 9, "Incorrect size");
9625 :
9626 : #undef FIELD_OK
9627 : #undef SET_FIELD
9628 : }
9629 :
9630 : void
9631 4 : spdk_bdev_histogram_enable(struct spdk_bdev *bdev, spdk_bdev_histogram_status_cb cb_fn,
9632 : void *cb_arg, bool enable)
9633 : {
9634 4 : struct spdk_bdev_enable_histogram_opts opts;
9635 :
9636 4 : spdk_bdev_enable_histogram_opts_init(&opts, sizeof(opts));
9637 4 : spdk_bdev_histogram_enable_ext(bdev, cb_fn, cb_arg, enable, &opts);
9638 4 : }
9639 :
9640 : struct spdk_bdev_histogram_data_ctx {
9641 : spdk_bdev_histogram_data_cb cb_fn;
9642 : void *cb_arg;
9643 : struct spdk_bdev *bdev;
9644 : /** merged histogram data from all channels */
9645 : struct spdk_histogram_data *histogram;
9646 : };
9647 :
9648 : static void
9649 5 : bdev_histogram_get_channel_cb(struct spdk_bdev *bdev, void *_ctx, int status)
9650 : {
9651 5 : struct spdk_bdev_histogram_data_ctx *ctx = _ctx;
9652 :
9653 5 : ctx->cb_fn(ctx->cb_arg, status, ctx->histogram);
9654 5 : free(ctx);
9655 5 : }
9656 :
9657 : static void
9658 7 : bdev_histogram_get_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
9659 : struct spdk_io_channel *_ch, void *_ctx)
9660 : {
9661 7 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
9662 7 : struct spdk_bdev_histogram_data_ctx *ctx = _ctx;
9663 7 : int status = 0;
9664 :
9665 7 : if (ch->histogram == NULL) {
9666 1 : status = -EFAULT;
9667 : } else {
9668 6 : spdk_histogram_data_merge(ctx->histogram, ch->histogram);
9669 : }
9670 :
9671 7 : spdk_bdev_for_each_channel_continue(i, status);
9672 7 : }
9673 :
9674 : void
9675 5 : spdk_bdev_histogram_get(struct spdk_bdev *bdev, struct spdk_histogram_data *histogram,
9676 : spdk_bdev_histogram_data_cb cb_fn,
9677 : void *cb_arg)
9678 : {
9679 : struct spdk_bdev_histogram_data_ctx *ctx;
9680 :
9681 5 : ctx = calloc(1, sizeof(struct spdk_bdev_histogram_data_ctx));
9682 5 : if (ctx == NULL) {
9683 0 : cb_fn(cb_arg, -ENOMEM, NULL);
9684 0 : return;
9685 : }
9686 :
9687 5 : ctx->bdev = bdev;
9688 5 : ctx->cb_fn = cb_fn;
9689 5 : ctx->cb_arg = cb_arg;
9690 :
9691 5 : ctx->histogram = histogram;
9692 :
9693 5 : spdk_bdev_for_each_channel(bdev, bdev_histogram_get_channel, ctx,
9694 : bdev_histogram_get_channel_cb);
9695 : }
9696 :
9697 : void
9698 2 : spdk_bdev_channel_get_histogram(struct spdk_io_channel *ch, spdk_bdev_histogram_data_cb cb_fn,
9699 : void *cb_arg)
9700 : {
9701 2 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(ch);
9702 2 : int status = 0;
9703 :
9704 2 : assert(cb_fn != NULL);
9705 :
9706 2 : if (bdev_ch->histogram == NULL) {
9707 1 : status = -EFAULT;
9708 : }
9709 2 : cb_fn(cb_arg, status, bdev_ch->histogram);
9710 2 : }
9711 :
9712 : size_t
9713 0 : spdk_bdev_get_media_events(struct spdk_bdev_desc *desc, struct spdk_bdev_media_event *events,
9714 : size_t max_events)
9715 : {
9716 : struct media_event_entry *entry;
9717 0 : size_t num_events = 0;
9718 :
9719 0 : for (; num_events < max_events; ++num_events) {
9720 0 : entry = TAILQ_FIRST(&desc->pending_media_events);
9721 0 : if (entry == NULL) {
9722 0 : break;
9723 : }
9724 :
9725 0 : events[num_events] = entry->event;
9726 0 : TAILQ_REMOVE(&desc->pending_media_events, entry, tailq);
9727 0 : TAILQ_INSERT_TAIL(&desc->free_media_events, entry, tailq);
9728 : }
9729 :
9730 0 : return num_events;
9731 : }
9732 :
9733 : int
9734 0 : spdk_bdev_push_media_events(struct spdk_bdev *bdev, const struct spdk_bdev_media_event *events,
9735 : size_t num_events)
9736 : {
9737 : struct spdk_bdev_desc *desc;
9738 : struct media_event_entry *entry;
9739 : size_t event_id;
9740 0 : int rc = 0;
9741 :
9742 0 : assert(bdev->media_events);
9743 :
9744 0 : spdk_spin_lock(&bdev->internal.spinlock);
9745 0 : TAILQ_FOREACH(desc, &bdev->internal.open_descs, link) {
9746 0 : if (desc->write) {
9747 0 : break;
9748 : }
9749 : }
9750 :
9751 0 : if (desc == NULL || desc->media_events_buffer == NULL) {
9752 0 : rc = -ENODEV;
9753 0 : goto out;
9754 : }
9755 :
9756 0 : for (event_id = 0; event_id < num_events; ++event_id) {
9757 0 : entry = TAILQ_FIRST(&desc->free_media_events);
9758 0 : if (entry == NULL) {
9759 0 : break;
9760 : }
9761 :
9762 0 : TAILQ_REMOVE(&desc->free_media_events, entry, tailq);
9763 0 : TAILQ_INSERT_TAIL(&desc->pending_media_events, entry, tailq);
9764 0 : entry->event = events[event_id];
9765 : }
9766 :
9767 0 : rc = event_id;
9768 0 : out:
9769 0 : spdk_spin_unlock(&bdev->internal.spinlock);
9770 0 : return rc;
9771 : }
9772 :
9773 : static void
9774 0 : _media_management_notify(void *arg)
9775 : {
9776 0 : struct spdk_bdev_desc *desc = arg;
9777 :
9778 0 : _event_notify(desc, SPDK_BDEV_EVENT_MEDIA_MANAGEMENT);
9779 0 : }
9780 :
9781 : void
9782 0 : spdk_bdev_notify_media_management(struct spdk_bdev *bdev)
9783 : {
9784 : struct spdk_bdev_desc *desc;
9785 :
9786 0 : spdk_spin_lock(&bdev->internal.spinlock);
9787 0 : TAILQ_FOREACH(desc, &bdev->internal.open_descs, link) {
9788 0 : if (!TAILQ_EMPTY(&desc->pending_media_events)) {
9789 0 : event_notify(desc, _media_management_notify);
9790 : }
9791 : }
9792 0 : spdk_spin_unlock(&bdev->internal.spinlock);
9793 0 : }
9794 :
9795 : struct locked_lba_range_ctx {
9796 : struct lba_range range;
9797 : struct lba_range *current_range;
9798 : struct lba_range *owner_range;
9799 : struct spdk_poller *poller;
9800 : lock_range_cb cb_fn;
9801 : void *cb_arg;
9802 : };
9803 :
9804 : static void
9805 0 : bdev_lock_error_cleanup_cb(struct spdk_bdev *bdev, void *_ctx, int status)
9806 : {
9807 0 : struct locked_lba_range_ctx *ctx = _ctx;
9808 :
9809 0 : ctx->cb_fn(&ctx->range, ctx->cb_arg, -ENOMEM);
9810 0 : free(ctx);
9811 0 : }
9812 :
9813 : static void bdev_unlock_lba_range_get_channel(struct spdk_bdev_channel_iter *i,
9814 : struct spdk_bdev *bdev, struct spdk_io_channel *ch, void *_ctx);
9815 :
9816 : static void
9817 14 : bdev_lock_lba_range_cb(struct spdk_bdev *bdev, void *_ctx, int status)
9818 : {
9819 14 : struct locked_lba_range_ctx *ctx = _ctx;
9820 :
9821 14 : if (status == -ENOMEM) {
9822 : /* One of the channels could not allocate a range object.
9823 : * So we have to go back and clean up any ranges that were
9824 : * allocated successfully before we return error status to
9825 : * the caller. We can reuse the unlock function to do that
9826 : * clean up.
9827 : */
9828 0 : spdk_bdev_for_each_channel(bdev, bdev_unlock_lba_range_get_channel, ctx,
9829 : bdev_lock_error_cleanup_cb);
9830 0 : return;
9831 : }
9832 :
9833 : /* All channels have locked this range and no I/O overlapping the range
9834 : * are outstanding! Set the owner_ch for the range object for the
9835 : * locking channel, so that this channel will know that it is allowed
9836 : * to write to this range.
9837 : */
9838 14 : if (ctx->owner_range != NULL) {
9839 10 : ctx->owner_range->owner_ch = ctx->range.owner_ch;
9840 : }
9841 :
9842 14 : ctx->cb_fn(&ctx->range, ctx->cb_arg, status);
9843 :
9844 : /* Don't free the ctx here. Its range is in the bdev's global list of
9845 : * locked ranges still, and will be removed and freed when this range
9846 : * is later unlocked.
9847 : */
9848 : }
9849 :
9850 : static int
9851 17 : bdev_lock_lba_range_check_io(void *_i)
9852 : {
9853 17 : struct spdk_bdev_channel_iter *i = _i;
9854 17 : struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i->i);
9855 17 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
9856 17 : struct locked_lba_range_ctx *ctx = i->ctx;
9857 17 : struct lba_range *range = ctx->current_range;
9858 : struct spdk_bdev_io *bdev_io;
9859 :
9860 17 : spdk_poller_unregister(&ctx->poller);
9861 :
9862 : /* The range is now in the locked_ranges, so no new IO can be submitted to this
9863 : * range. But we need to wait until any outstanding IO overlapping with this range
9864 : * are completed.
9865 : */
9866 18 : TAILQ_FOREACH(bdev_io, &ch->io_submitted, internal.ch_link) {
9867 3 : if (bdev_io_range_is_locked(bdev_io, range)) {
9868 2 : ctx->poller = SPDK_POLLER_REGISTER(bdev_lock_lba_range_check_io, i, 100);
9869 2 : return SPDK_POLLER_BUSY;
9870 : }
9871 : }
9872 :
9873 15 : spdk_bdev_for_each_channel_continue(i, 0);
9874 15 : return SPDK_POLLER_BUSY;
9875 : }
9876 :
9877 : static void
9878 15 : bdev_lock_lba_range_get_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
9879 : struct spdk_io_channel *_ch, void *_ctx)
9880 : {
9881 15 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
9882 15 : struct locked_lba_range_ctx *ctx = _ctx;
9883 : struct lba_range *range;
9884 :
9885 16 : TAILQ_FOREACH(range, &ch->locked_ranges, tailq) {
9886 1 : if (range->length == ctx->range.length &&
9887 0 : range->offset == ctx->range.offset &&
9888 0 : range->locked_ctx == ctx->range.locked_ctx) {
9889 : /* This range already exists on this channel, so don't add
9890 : * it again. This can happen when a new channel is created
9891 : * while the for_each_channel operation is in progress.
9892 : * Do not check for outstanding I/O in that case, since the
9893 : * range was locked before any I/O could be submitted to the
9894 : * new channel.
9895 : */
9896 0 : spdk_bdev_for_each_channel_continue(i, 0);
9897 0 : return;
9898 : }
9899 : }
9900 :
9901 15 : range = calloc(1, sizeof(*range));
9902 15 : if (range == NULL) {
9903 0 : spdk_bdev_for_each_channel_continue(i, -ENOMEM);
9904 0 : return;
9905 : }
9906 :
9907 15 : range->length = ctx->range.length;
9908 15 : range->offset = ctx->range.offset;
9909 15 : range->locked_ctx = ctx->range.locked_ctx;
9910 15 : range->quiesce = ctx->range.quiesce;
9911 15 : ctx->current_range = range;
9912 15 : if (ctx->range.owner_ch == ch) {
9913 : /* This is the range object for the channel that will hold
9914 : * the lock. Store it in the ctx object so that we can easily
9915 : * set its owner_ch after the lock is finally acquired.
9916 : */
9917 10 : ctx->owner_range = range;
9918 : }
9919 15 : TAILQ_INSERT_TAIL(&ch->locked_ranges, range, tailq);
9920 15 : bdev_lock_lba_range_check_io(i);
9921 : }
9922 :
9923 : static void
9924 14 : bdev_lock_lba_range_ctx(struct spdk_bdev *bdev, struct locked_lba_range_ctx *ctx)
9925 : {
9926 14 : assert(spdk_get_thread() == ctx->range.owner_thread);
9927 14 : assert(ctx->range.owner_ch == NULL ||
9928 : spdk_io_channel_get_thread(ctx->range.owner_ch->channel) == ctx->range.owner_thread);
9929 :
9930 : /* We will add a copy of this range to each channel now. */
9931 14 : spdk_bdev_for_each_channel(bdev, bdev_lock_lba_range_get_channel, ctx,
9932 : bdev_lock_lba_range_cb);
9933 14 : }
9934 :
9935 : static bool
9936 17 : bdev_lba_range_overlaps_tailq(struct lba_range *range, lba_range_tailq_t *tailq)
9937 : {
9938 : struct lba_range *r;
9939 :
9940 18 : TAILQ_FOREACH(r, tailq, tailq) {
9941 4 : if (bdev_lba_range_overlapped(range, r)) {
9942 3 : return true;
9943 : }
9944 : }
9945 14 : return false;
9946 : }
9947 :
9948 : static void bdev_quiesce_range_locked(struct lba_range *range, void *ctx, int status);
9949 :
9950 : static int
9951 14 : _bdev_lock_lba_range(struct spdk_bdev *bdev, struct spdk_bdev_channel *ch,
9952 : uint64_t offset, uint64_t length,
9953 : lock_range_cb cb_fn, void *cb_arg)
9954 : {
9955 : struct locked_lba_range_ctx *ctx;
9956 :
9957 14 : ctx = calloc(1, sizeof(*ctx));
9958 14 : if (ctx == NULL) {
9959 0 : return -ENOMEM;
9960 : }
9961 :
9962 14 : ctx->range.offset = offset;
9963 14 : ctx->range.length = length;
9964 14 : ctx->range.owner_thread = spdk_get_thread();
9965 14 : ctx->range.owner_ch = ch;
9966 14 : ctx->range.locked_ctx = cb_arg;
9967 14 : ctx->range.bdev = bdev;
9968 14 : ctx->range.quiesce = (cb_fn == bdev_quiesce_range_locked);
9969 14 : ctx->cb_fn = cb_fn;
9970 14 : ctx->cb_arg = cb_arg;
9971 :
9972 14 : spdk_spin_lock(&bdev->internal.spinlock);
9973 14 : if (bdev_lba_range_overlaps_tailq(&ctx->range, &bdev->internal.locked_ranges)) {
9974 : /* There is an active lock overlapping with this range.
9975 : * Put it on the pending list until this range no
9976 : * longer overlaps with another.
9977 : */
9978 2 : TAILQ_INSERT_TAIL(&bdev->internal.pending_locked_ranges, &ctx->range, tailq);
9979 : } else {
9980 12 : TAILQ_INSERT_TAIL(&bdev->internal.locked_ranges, &ctx->range, tailq);
9981 12 : bdev_lock_lba_range_ctx(bdev, ctx);
9982 : }
9983 14 : spdk_spin_unlock(&bdev->internal.spinlock);
9984 14 : return 0;
9985 : }
9986 :
9987 : static int
9988 10 : bdev_lock_lba_range(struct spdk_bdev_desc *desc, struct spdk_io_channel *_ch,
9989 : uint64_t offset, uint64_t length,
9990 : lock_range_cb cb_fn, void *cb_arg)
9991 : {
9992 10 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
9993 10 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
9994 :
9995 10 : if (cb_arg == NULL) {
9996 0 : SPDK_ERRLOG("cb_arg must not be NULL\n");
9997 0 : return -EINVAL;
9998 : }
9999 :
10000 10 : return _bdev_lock_lba_range(bdev, ch, offset, length, cb_fn, cb_arg);
10001 : }
10002 :
10003 : static void
10004 2 : bdev_lock_lba_range_ctx_msg(void *_ctx)
10005 : {
10006 2 : struct locked_lba_range_ctx *ctx = _ctx;
10007 :
10008 2 : bdev_lock_lba_range_ctx(ctx->range.bdev, ctx);
10009 2 : }
10010 :
10011 : static void
10012 14 : bdev_unlock_lba_range_cb(struct spdk_bdev *bdev, void *_ctx, int status)
10013 : {
10014 14 : struct locked_lba_range_ctx *ctx = _ctx;
10015 : struct locked_lba_range_ctx *pending_ctx;
10016 : struct lba_range *range, *tmp;
10017 :
10018 14 : spdk_spin_lock(&bdev->internal.spinlock);
10019 : /* Check if there are any pending locked ranges that overlap with this range
10020 : * that was just unlocked. If there are, check that it doesn't overlap with any
10021 : * other locked ranges before calling bdev_lock_lba_range_ctx which will start
10022 : * the lock process.
10023 : */
10024 17 : TAILQ_FOREACH_SAFE(range, &bdev->internal.pending_locked_ranges, tailq, tmp) {
10025 3 : if (bdev_lba_range_overlapped(range, &ctx->range) &&
10026 3 : !bdev_lba_range_overlaps_tailq(range, &bdev->internal.locked_ranges)) {
10027 2 : TAILQ_REMOVE(&bdev->internal.pending_locked_ranges, range, tailq);
10028 2 : pending_ctx = SPDK_CONTAINEROF(range, struct locked_lba_range_ctx, range);
10029 2 : TAILQ_INSERT_TAIL(&bdev->internal.locked_ranges, range, tailq);
10030 2 : spdk_thread_send_msg(pending_ctx->range.owner_thread,
10031 : bdev_lock_lba_range_ctx_msg, pending_ctx);
10032 : }
10033 : }
10034 14 : spdk_spin_unlock(&bdev->internal.spinlock);
10035 :
10036 14 : ctx->cb_fn(&ctx->range, ctx->cb_arg, status);
10037 14 : free(ctx);
10038 14 : }
10039 :
10040 : static void
10041 16 : bdev_unlock_lba_range_get_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
10042 : struct spdk_io_channel *_ch, void *_ctx)
10043 : {
10044 16 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
10045 16 : struct locked_lba_range_ctx *ctx = _ctx;
10046 16 : TAILQ_HEAD(, spdk_bdev_io) io_locked;
10047 : struct spdk_bdev_io *bdev_io;
10048 : struct lba_range *range;
10049 :
10050 16 : TAILQ_FOREACH(range, &ch->locked_ranges, tailq) {
10051 16 : if (ctx->range.offset == range->offset &&
10052 16 : ctx->range.length == range->length &&
10053 16 : ctx->range.locked_ctx == range->locked_ctx) {
10054 16 : TAILQ_REMOVE(&ch->locked_ranges, range, tailq);
10055 16 : free(range);
10056 16 : break;
10057 : }
10058 : }
10059 :
10060 : /* Note: we should almost always be able to assert that the range specified
10061 : * was found. But there are some very rare corner cases where a new channel
10062 : * gets created simultaneously with a range unlock, where this function
10063 : * would execute on that new channel and wouldn't have the range.
10064 : * We also use this to clean up range allocations when a later allocation
10065 : * fails in the locking path.
10066 : * So we can't actually assert() here.
10067 : */
10068 :
10069 : /* Swap the locked IO into a temporary list, and then try to submit them again.
10070 : * We could hyper-optimize this to only resubmit locked I/O that overlap
10071 : * with the range that was just unlocked, but this isn't a performance path so
10072 : * we go for simplicity here.
10073 : */
10074 16 : TAILQ_INIT(&io_locked);
10075 16 : TAILQ_SWAP(&ch->io_locked, &io_locked, spdk_bdev_io, internal.ch_link);
10076 19 : while (!TAILQ_EMPTY(&io_locked)) {
10077 3 : bdev_io = TAILQ_FIRST(&io_locked);
10078 3 : TAILQ_REMOVE(&io_locked, bdev_io, internal.ch_link);
10079 3 : bdev_io_submit(bdev_io);
10080 : }
10081 :
10082 16 : spdk_bdev_for_each_channel_continue(i, 0);
10083 16 : }
10084 :
10085 : static int
10086 14 : _bdev_unlock_lba_range(struct spdk_bdev *bdev, uint64_t offset, uint64_t length,
10087 : lock_range_cb cb_fn, void *cb_arg)
10088 : {
10089 : struct locked_lba_range_ctx *ctx;
10090 : struct lba_range *range;
10091 :
10092 14 : spdk_spin_lock(&bdev->internal.spinlock);
10093 : /* To start the unlock the process, we find the range in the bdev's locked_ranges
10094 : * and remove it. This ensures new channels don't inherit the locked range.
10095 : * Then we will send a message to each channel to remove the range from its
10096 : * per-channel list.
10097 : */
10098 14 : TAILQ_FOREACH(range, &bdev->internal.locked_ranges, tailq) {
10099 14 : if (range->offset == offset && range->length == length &&
10100 14 : (range->owner_ch == NULL || range->locked_ctx == cb_arg)) {
10101 : break;
10102 : }
10103 : }
10104 14 : if (range == NULL) {
10105 0 : assert(false);
10106 : spdk_spin_unlock(&bdev->internal.spinlock);
10107 : return -EINVAL;
10108 : }
10109 14 : TAILQ_REMOVE(&bdev->internal.locked_ranges, range, tailq);
10110 14 : ctx = SPDK_CONTAINEROF(range, struct locked_lba_range_ctx, range);
10111 14 : spdk_spin_unlock(&bdev->internal.spinlock);
10112 :
10113 14 : ctx->cb_fn = cb_fn;
10114 14 : ctx->cb_arg = cb_arg;
10115 :
10116 14 : spdk_bdev_for_each_channel(bdev, bdev_unlock_lba_range_get_channel, ctx,
10117 : bdev_unlock_lba_range_cb);
10118 14 : return 0;
10119 : }
10120 :
10121 : static int
10122 12 : bdev_unlock_lba_range(struct spdk_bdev_desc *desc, struct spdk_io_channel *_ch,
10123 : uint64_t offset, uint64_t length,
10124 : lock_range_cb cb_fn, void *cb_arg)
10125 : {
10126 12 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
10127 12 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
10128 : struct lba_range *range;
10129 12 : bool range_found = false;
10130 :
10131 : /* Let's make sure the specified channel actually has a lock on
10132 : * the specified range. Note that the range must match exactly.
10133 : */
10134 14 : TAILQ_FOREACH(range, &ch->locked_ranges, tailq) {
10135 12 : if (range->offset == offset && range->length == length &&
10136 11 : range->owner_ch == ch && range->locked_ctx == cb_arg) {
10137 10 : range_found = true;
10138 10 : break;
10139 : }
10140 : }
10141 :
10142 12 : if (!range_found) {
10143 2 : return -EINVAL;
10144 : }
10145 :
10146 10 : return _bdev_unlock_lba_range(bdev, offset, length, cb_fn, cb_arg);
10147 : }
10148 :
10149 : struct bdev_quiesce_ctx {
10150 : spdk_bdev_quiesce_cb cb_fn;
10151 : void *cb_arg;
10152 : };
10153 :
10154 : static void
10155 4 : bdev_unquiesce_range_unlocked(struct lba_range *range, void *ctx, int status)
10156 : {
10157 4 : struct bdev_quiesce_ctx *quiesce_ctx = ctx;
10158 :
10159 4 : if (quiesce_ctx->cb_fn != NULL) {
10160 4 : quiesce_ctx->cb_fn(quiesce_ctx->cb_arg, status);
10161 : }
10162 :
10163 4 : free(quiesce_ctx);
10164 4 : }
10165 :
10166 : static void
10167 4 : bdev_quiesce_range_locked(struct lba_range *range, void *ctx, int status)
10168 : {
10169 4 : struct bdev_quiesce_ctx *quiesce_ctx = ctx;
10170 4 : struct spdk_bdev_module *module = range->bdev->module;
10171 :
10172 4 : if (status != 0) {
10173 0 : if (quiesce_ctx->cb_fn != NULL) {
10174 0 : quiesce_ctx->cb_fn(quiesce_ctx->cb_arg, status);
10175 : }
10176 0 : free(quiesce_ctx);
10177 0 : return;
10178 : }
10179 :
10180 4 : spdk_spin_lock(&module->internal.spinlock);
10181 4 : TAILQ_INSERT_TAIL(&module->internal.quiesced_ranges, range, tailq_module);
10182 4 : spdk_spin_unlock(&module->internal.spinlock);
10183 :
10184 4 : if (quiesce_ctx->cb_fn != NULL) {
10185 : /* copy the context in case the range is unlocked by the callback */
10186 4 : struct bdev_quiesce_ctx tmp = *quiesce_ctx;
10187 :
10188 4 : quiesce_ctx->cb_fn = NULL;
10189 4 : quiesce_ctx->cb_arg = NULL;
10190 :
10191 4 : tmp.cb_fn(tmp.cb_arg, status);
10192 : }
10193 : /* quiesce_ctx will be freed on unquiesce */
10194 : }
10195 :
10196 : static int
10197 9 : _spdk_bdev_quiesce(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
10198 : uint64_t offset, uint64_t length,
10199 : spdk_bdev_quiesce_cb cb_fn, void *cb_arg,
10200 : bool unquiesce)
10201 : {
10202 : struct bdev_quiesce_ctx *quiesce_ctx;
10203 : int rc;
10204 :
10205 9 : if (module != bdev->module) {
10206 0 : SPDK_ERRLOG("Bdev does not belong to specified module.\n");
10207 0 : return -EINVAL;
10208 : }
10209 :
10210 9 : if (!bdev_io_valid_blocks(bdev, offset, length)) {
10211 0 : return -EINVAL;
10212 : }
10213 :
10214 9 : if (unquiesce) {
10215 : struct lba_range *range;
10216 :
10217 : /* Make sure the specified range is actually quiesced in the specified module and
10218 : * then remove it from the list. Note that the range must match exactly.
10219 : */
10220 5 : spdk_spin_lock(&module->internal.spinlock);
10221 6 : TAILQ_FOREACH(range, &module->internal.quiesced_ranges, tailq_module) {
10222 5 : if (range->bdev == bdev && range->offset == offset && range->length == length) {
10223 4 : TAILQ_REMOVE(&module->internal.quiesced_ranges, range, tailq_module);
10224 4 : break;
10225 : }
10226 : }
10227 5 : spdk_spin_unlock(&module->internal.spinlock);
10228 :
10229 5 : if (range == NULL) {
10230 1 : SPDK_ERRLOG("The range to unquiesce was not found.\n");
10231 1 : return -EINVAL;
10232 : }
10233 :
10234 4 : quiesce_ctx = range->locked_ctx;
10235 4 : quiesce_ctx->cb_fn = cb_fn;
10236 4 : quiesce_ctx->cb_arg = cb_arg;
10237 :
10238 4 : rc = _bdev_unlock_lba_range(bdev, offset, length, bdev_unquiesce_range_unlocked, quiesce_ctx);
10239 : } else {
10240 4 : quiesce_ctx = malloc(sizeof(*quiesce_ctx));
10241 4 : if (quiesce_ctx == NULL) {
10242 0 : return -ENOMEM;
10243 : }
10244 :
10245 4 : quiesce_ctx->cb_fn = cb_fn;
10246 4 : quiesce_ctx->cb_arg = cb_arg;
10247 :
10248 4 : rc = _bdev_lock_lba_range(bdev, NULL, offset, length, bdev_quiesce_range_locked, quiesce_ctx);
10249 4 : if (rc != 0) {
10250 0 : free(quiesce_ctx);
10251 : }
10252 : }
10253 :
10254 8 : return rc;
10255 : }
10256 :
10257 : int
10258 3 : spdk_bdev_quiesce(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
10259 : spdk_bdev_quiesce_cb cb_fn, void *cb_arg)
10260 : {
10261 3 : return _spdk_bdev_quiesce(bdev, module, 0, bdev->blockcnt, cb_fn, cb_arg, false);
10262 : }
10263 :
10264 : int
10265 3 : spdk_bdev_unquiesce(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
10266 : spdk_bdev_quiesce_cb cb_fn, void *cb_arg)
10267 : {
10268 3 : return _spdk_bdev_quiesce(bdev, module, 0, bdev->blockcnt, cb_fn, cb_arg, true);
10269 : }
10270 :
10271 : int
10272 1 : spdk_bdev_quiesce_range(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
10273 : uint64_t offset, uint64_t length,
10274 : spdk_bdev_quiesce_cb cb_fn, void *cb_arg)
10275 : {
10276 1 : return _spdk_bdev_quiesce(bdev, module, offset, length, cb_fn, cb_arg, false);
10277 : }
10278 :
10279 : int
10280 2 : spdk_bdev_unquiesce_range(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
10281 : uint64_t offset, uint64_t length,
10282 : spdk_bdev_quiesce_cb cb_fn, void *cb_arg)
10283 : {
10284 2 : return _spdk_bdev_quiesce(bdev, module, offset, length, cb_fn, cb_arg, true);
10285 : }
10286 :
10287 : int
10288 267 : spdk_bdev_get_memory_domains(struct spdk_bdev *bdev, struct spdk_memory_domain **domains,
10289 : int array_size)
10290 : {
10291 267 : if (!bdev) {
10292 1 : return -EINVAL;
10293 : }
10294 :
10295 266 : if (bdev->fn_table->get_memory_domains) {
10296 3 : return bdev->fn_table->get_memory_domains(bdev->ctxt, domains, array_size);
10297 : }
10298 :
10299 263 : return 0;
10300 : }
10301 :
10302 : struct spdk_bdev_for_each_io_ctx {
10303 : void *ctx;
10304 : spdk_bdev_io_fn fn;
10305 : spdk_bdev_for_each_io_cb cb;
10306 : };
10307 :
10308 : static void
10309 0 : bdev_channel_for_each_io(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
10310 : struct spdk_io_channel *io_ch, void *_ctx)
10311 : {
10312 0 : struct spdk_bdev_for_each_io_ctx *ctx = _ctx;
10313 0 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
10314 : struct spdk_bdev_io *bdev_io;
10315 0 : int rc = 0;
10316 :
10317 0 : TAILQ_FOREACH(bdev_io, &bdev_ch->io_submitted, internal.ch_link) {
10318 0 : rc = ctx->fn(ctx->ctx, bdev_io);
10319 0 : if (rc != 0) {
10320 0 : break;
10321 : }
10322 : }
10323 :
10324 0 : spdk_bdev_for_each_channel_continue(i, rc);
10325 0 : }
10326 :
10327 : static void
10328 0 : bdev_for_each_io_done(struct spdk_bdev *bdev, void *_ctx, int status)
10329 : {
10330 0 : struct spdk_bdev_for_each_io_ctx *ctx = _ctx;
10331 :
10332 0 : ctx->cb(ctx->ctx, status);
10333 :
10334 0 : free(ctx);
10335 0 : }
10336 :
10337 : void
10338 0 : spdk_bdev_for_each_bdev_io(struct spdk_bdev *bdev, void *_ctx, spdk_bdev_io_fn fn,
10339 : spdk_bdev_for_each_io_cb cb)
10340 : {
10341 : struct spdk_bdev_for_each_io_ctx *ctx;
10342 :
10343 0 : assert(fn != NULL && cb != NULL);
10344 :
10345 0 : ctx = calloc(1, sizeof(*ctx));
10346 0 : if (ctx == NULL) {
10347 0 : SPDK_ERRLOG("Failed to allocate context.\n");
10348 0 : cb(_ctx, -ENOMEM);
10349 0 : return;
10350 : }
10351 :
10352 0 : ctx->ctx = _ctx;
10353 0 : ctx->fn = fn;
10354 0 : ctx->cb = cb;
10355 :
10356 0 : spdk_bdev_for_each_channel(bdev, bdev_channel_for_each_io, ctx,
10357 : bdev_for_each_io_done);
10358 : }
10359 :
10360 : void
10361 128 : spdk_bdev_for_each_channel_continue(struct spdk_bdev_channel_iter *iter, int status)
10362 : {
10363 128 : spdk_for_each_channel_continue(iter->i, status);
10364 128 : }
10365 :
10366 : static struct spdk_bdev *
10367 351 : io_channel_iter_get_bdev(struct spdk_io_channel_iter *i)
10368 : {
10369 351 : void *io_device = spdk_io_channel_iter_get_io_device(i);
10370 :
10371 351 : return __bdev_from_io_dev(io_device);
10372 : }
10373 :
10374 : static void
10375 128 : bdev_each_channel_msg(struct spdk_io_channel_iter *i)
10376 : {
10377 128 : struct spdk_bdev_channel_iter *iter = spdk_io_channel_iter_get_ctx(i);
10378 128 : struct spdk_bdev *bdev = io_channel_iter_get_bdev(i);
10379 128 : struct spdk_io_channel *ch = spdk_io_channel_iter_get_channel(i);
10380 :
10381 128 : iter->i = i;
10382 128 : iter->fn(iter, bdev, ch, iter->ctx);
10383 128 : }
10384 :
10385 : static void
10386 223 : bdev_each_channel_cpl(struct spdk_io_channel_iter *i, int status)
10387 : {
10388 223 : struct spdk_bdev_channel_iter *iter = spdk_io_channel_iter_get_ctx(i);
10389 223 : struct spdk_bdev *bdev = io_channel_iter_get_bdev(i);
10390 :
10391 223 : iter->i = i;
10392 223 : iter->cpl(bdev, iter->ctx, status);
10393 :
10394 223 : free(iter);
10395 223 : }
10396 :
10397 : void
10398 223 : spdk_bdev_for_each_channel(struct spdk_bdev *bdev, spdk_bdev_for_each_channel_msg fn,
10399 : void *ctx, spdk_bdev_for_each_channel_done cpl)
10400 : {
10401 : struct spdk_bdev_channel_iter *iter;
10402 :
10403 223 : assert(bdev != NULL && fn != NULL && ctx != NULL);
10404 :
10405 223 : iter = calloc(1, sizeof(struct spdk_bdev_channel_iter));
10406 223 : if (iter == NULL) {
10407 0 : SPDK_ERRLOG("Unable to allocate iterator\n");
10408 0 : assert(false);
10409 : return;
10410 : }
10411 :
10412 223 : iter->fn = fn;
10413 223 : iter->cpl = cpl;
10414 223 : iter->ctx = ctx;
10415 :
10416 223 : spdk_for_each_channel(__bdev_to_io_dev(bdev), bdev_each_channel_msg,
10417 : iter, bdev_each_channel_cpl);
10418 : }
10419 :
10420 : static void
10421 3 : bdev_copy_do_write_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
10422 : {
10423 3 : struct spdk_bdev_io *parent_io = cb_arg;
10424 :
10425 3 : spdk_bdev_free_io(bdev_io);
10426 :
10427 : /* Check return status of write */
10428 3 : parent_io->internal.status = success ? SPDK_BDEV_IO_STATUS_SUCCESS : SPDK_BDEV_IO_STATUS_FAILED;
10429 3 : parent_io->internal.cb(parent_io, success, parent_io->internal.caller_ctx);
10430 3 : }
10431 :
10432 : static void
10433 3 : bdev_copy_do_write(void *_bdev_io)
10434 : {
10435 3 : struct spdk_bdev_io *bdev_io = _bdev_io;
10436 : int rc;
10437 :
10438 : /* Write blocks */
10439 3 : rc = spdk_bdev_write_blocks_with_md(bdev_io->internal.desc,
10440 3 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
10441 3 : bdev_io->u.bdev.iovs[0].iov_base,
10442 : bdev_io->u.bdev.md_buf, bdev_io->u.bdev.offset_blocks,
10443 : bdev_io->u.bdev.num_blocks, bdev_copy_do_write_done, bdev_io);
10444 :
10445 3 : if (rc == -ENOMEM) {
10446 0 : bdev_queue_io_wait_with_cb(bdev_io, bdev_copy_do_write);
10447 3 : } else if (rc != 0) {
10448 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
10449 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
10450 : }
10451 3 : }
10452 :
10453 : static void
10454 3 : bdev_copy_do_read_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
10455 : {
10456 3 : struct spdk_bdev_io *parent_io = cb_arg;
10457 :
10458 3 : spdk_bdev_free_io(bdev_io);
10459 :
10460 : /* Check return status of read */
10461 3 : if (!success) {
10462 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
10463 0 : parent_io->internal.cb(parent_io, false, parent_io->internal.caller_ctx);
10464 0 : return;
10465 : }
10466 :
10467 : /* Do write */
10468 3 : bdev_copy_do_write(parent_io);
10469 : }
10470 :
10471 : static void
10472 3 : bdev_copy_do_read(void *_bdev_io)
10473 : {
10474 3 : struct spdk_bdev_io *bdev_io = _bdev_io;
10475 : int rc;
10476 :
10477 : /* Read blocks */
10478 3 : rc = spdk_bdev_read_blocks_with_md(bdev_io->internal.desc,
10479 3 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
10480 3 : bdev_io->u.bdev.iovs[0].iov_base,
10481 : bdev_io->u.bdev.md_buf, bdev_io->u.bdev.copy.src_offset_blocks,
10482 : bdev_io->u.bdev.num_blocks, bdev_copy_do_read_done, bdev_io);
10483 :
10484 3 : if (rc == -ENOMEM) {
10485 0 : bdev_queue_io_wait_with_cb(bdev_io, bdev_copy_do_read);
10486 3 : } else if (rc != 0) {
10487 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
10488 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
10489 : }
10490 3 : }
10491 :
10492 : static void
10493 3 : bdev_copy_get_buf_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io, bool success)
10494 : {
10495 3 : if (!success) {
10496 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
10497 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
10498 0 : return;
10499 : }
10500 :
10501 3 : bdev_copy_do_read(bdev_io);
10502 : }
10503 :
10504 : int
10505 27 : spdk_bdev_copy_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
10506 : uint64_t dst_offset_blocks, uint64_t src_offset_blocks, uint64_t num_blocks,
10507 : spdk_bdev_io_completion_cb cb, void *cb_arg)
10508 : {
10509 27 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
10510 : struct spdk_bdev_io *bdev_io;
10511 27 : struct spdk_bdev_channel *channel = spdk_io_channel_get_ctx(ch);
10512 :
10513 27 : if (!desc->write) {
10514 0 : return -EBADF;
10515 : }
10516 :
10517 27 : if (!bdev_io_valid_blocks(bdev, dst_offset_blocks, num_blocks) ||
10518 27 : !bdev_io_valid_blocks(bdev, src_offset_blocks, num_blocks)) {
10519 0 : SPDK_DEBUGLOG(bdev,
10520 : "Invalid offset or number of blocks: dst %lu, src %lu, count %lu\n",
10521 : dst_offset_blocks, src_offset_blocks, num_blocks);
10522 0 : return -EINVAL;
10523 : }
10524 :
10525 27 : bdev_io = bdev_channel_get_io(channel);
10526 27 : if (!bdev_io) {
10527 0 : return -ENOMEM;
10528 : }
10529 :
10530 27 : bdev_io->internal.ch = channel;
10531 27 : bdev_io->internal.desc = desc;
10532 27 : bdev_io->type = SPDK_BDEV_IO_TYPE_COPY;
10533 :
10534 27 : bdev_io->u.bdev.offset_blocks = dst_offset_blocks;
10535 27 : bdev_io->u.bdev.copy.src_offset_blocks = src_offset_blocks;
10536 27 : bdev_io->u.bdev.num_blocks = num_blocks;
10537 27 : bdev_io->u.bdev.memory_domain = NULL;
10538 27 : bdev_io->u.bdev.memory_domain_ctx = NULL;
10539 27 : bdev_io->u.bdev.iovs = NULL;
10540 27 : bdev_io->u.bdev.iovcnt = 0;
10541 27 : bdev_io->u.bdev.md_buf = NULL;
10542 27 : bdev_io->u.bdev.accel_sequence = NULL;
10543 27 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
10544 :
10545 27 : if (dst_offset_blocks == src_offset_blocks || num_blocks == 0) {
10546 0 : spdk_thread_send_msg(spdk_get_thread(), bdev_io_complete_cb, bdev_io);
10547 0 : return 0;
10548 : }
10549 :
10550 :
10551 : /* If the copy size is large and should be split, use the generic split logic
10552 : * regardless of whether SPDK_BDEV_IO_TYPE_COPY is supported or not.
10553 : *
10554 : * Then, send the copy request if SPDK_BDEV_IO_TYPE_COPY is supported or
10555 : * emulate it using regular read and write requests otherwise.
10556 : */
10557 27 : if (spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COPY) ||
10558 : bdev_io->internal.f.split) {
10559 24 : bdev_io_submit(bdev_io);
10560 24 : return 0;
10561 : }
10562 :
10563 3 : spdk_bdev_io_get_buf(bdev_io, bdev_copy_get_buf_cb, num_blocks * spdk_bdev_get_block_size(bdev));
10564 :
10565 3 : return 0;
10566 : }
10567 :
10568 3 : SPDK_LOG_REGISTER_COMPONENT(bdev)
10569 :
10570 3 : SPDK_TRACE_REGISTER_FN(bdev_trace, "bdev", TRACE_GROUP_BDEV)
10571 : {
10572 0 : struct spdk_trace_tpoint_opts opts[] = {
10573 : {
10574 : "BDEV_IO_START", TRACE_BDEV_IO_START,
10575 : OWNER_TYPE_BDEV, OBJECT_BDEV_IO, 1,
10576 : {
10577 : { "type", SPDK_TRACE_ARG_TYPE_INT, 8 },
10578 : { "ctx", SPDK_TRACE_ARG_TYPE_PTR, 8 },
10579 : { "offset", SPDK_TRACE_ARG_TYPE_INT, 8 },
10580 : { "qd", SPDK_TRACE_ARG_TYPE_INT, 4 }
10581 : }
10582 : },
10583 : {
10584 : "BDEV_IO_DONE", TRACE_BDEV_IO_DONE,
10585 : OWNER_TYPE_BDEV, OBJECT_BDEV_IO, 0,
10586 : {
10587 : { "ctx", SPDK_TRACE_ARG_TYPE_PTR, 8 },
10588 : { "qd", SPDK_TRACE_ARG_TYPE_INT, 4 }
10589 : }
10590 : },
10591 : {
10592 : "BDEV_IOCH_CREATE", TRACE_BDEV_IOCH_CREATE,
10593 : OWNER_TYPE_BDEV, OBJECT_NONE, 0,
10594 : {
10595 : { "tid", SPDK_TRACE_ARG_TYPE_INT, 8 }
10596 : }
10597 : },
10598 : {
10599 : "BDEV_IOCH_DESTROY", TRACE_BDEV_IOCH_DESTROY,
10600 : OWNER_TYPE_BDEV, OBJECT_NONE, 0,
10601 : {
10602 : { "tid", SPDK_TRACE_ARG_TYPE_INT, 8 }
10603 : }
10604 : },
10605 : };
10606 :
10607 :
10608 0 : spdk_trace_register_owner_type(OWNER_TYPE_BDEV, 'b');
10609 0 : spdk_trace_register_object(OBJECT_BDEV_IO, 'i');
10610 0 : spdk_trace_register_description_ext(opts, SPDK_COUNTOF(opts));
10611 0 : spdk_trace_tpoint_register_relation(TRACE_BDEV_NVME_IO_START, OBJECT_BDEV_IO, 0);
10612 0 : spdk_trace_tpoint_register_relation(TRACE_BDEV_NVME_IO_DONE, OBJECT_BDEV_IO, 0);
10613 0 : }
|