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_POOL_SIZE 8191
42 : #define BUF_LARGE_POOL_SIZE 1023
43 : #define BUF_SMALL_CACHE_SIZE 128
44 : #define BUF_LARGE_CACHE_SIZE 16
45 : #define NOMEM_THRESHOLD_COUNT 8
46 :
47 : #define SPDK_BDEV_QOS_TIMESLICE_IN_USEC 1000
48 : #define SPDK_BDEV_QOS_MIN_IO_PER_TIMESLICE 1
49 : #define SPDK_BDEV_QOS_MIN_BYTE_PER_TIMESLICE 512
50 : #define SPDK_BDEV_QOS_MIN_IOS_PER_SEC 1000
51 : #define SPDK_BDEV_QOS_MIN_BYTES_PER_SEC (1024 * 1024)
52 : #define SPDK_BDEV_QOS_MAX_MBYTES_PER_SEC (UINT64_MAX / (1024 * 1024))
53 : #define SPDK_BDEV_QOS_LIMIT_NOT_DEFINED UINT64_MAX
54 : #define SPDK_BDEV_IO_POLL_INTERVAL_IN_MSEC 1000
55 :
56 : /* The maximum number of children requests for a UNMAP or WRITE ZEROES command
57 : * when splitting into children requests at a time.
58 : */
59 : #define SPDK_BDEV_MAX_CHILDREN_UNMAP_WRITE_ZEROES_REQS (8)
60 : #define BDEV_RESET_CHECK_OUTSTANDING_IO_PERIOD 1000000
61 :
62 : /* The maximum number of children requests for a COPY command
63 : * when splitting into children requests at a time.
64 : */
65 : #define SPDK_BDEV_MAX_CHILDREN_COPY_REQS (8)
66 :
67 : #define LOG_ALREADY_CLAIMED_ERROR(detail, bdev) \
68 : log_already_claimed(SPDK_LOG_ERROR, __LINE__, __func__, detail, bdev)
69 : #ifdef DEBUG
70 : #define LOG_ALREADY_CLAIMED_DEBUG(detail, bdev) \
71 : log_already_claimed(SPDK_LOG_DEBUG, __LINE__, __func__, detail, bdev)
72 : #else
73 : #define LOG_ALREADY_CLAIMED_DEBUG(detail, bdev) do {} while(0)
74 : #endif
75 :
76 : static void log_already_claimed(enum spdk_log_level level, const int line, const char *func,
77 : const char *detail, struct spdk_bdev *bdev);
78 :
79 : static const char *qos_rpc_type[] = {"rw_ios_per_sec",
80 : "rw_mbytes_per_sec", "r_mbytes_per_sec", "w_mbytes_per_sec"
81 : };
82 :
83 : TAILQ_HEAD(spdk_bdev_list, spdk_bdev);
84 :
85 : RB_HEAD(bdev_name_tree, spdk_bdev_name);
86 :
87 : static int
88 411 : bdev_name_cmp(struct spdk_bdev_name *name1, struct spdk_bdev_name *name2)
89 : {
90 411 : return strcmp(name1->name, name2->name);
91 : }
92 :
93 1467 : RB_GENERATE_STATIC(bdev_name_tree, spdk_bdev_name, node, bdev_name_cmp);
94 :
95 : struct spdk_bdev_mgr {
96 : struct spdk_mempool *bdev_io_pool;
97 :
98 : void *zero_buffer;
99 :
100 : TAILQ_HEAD(bdev_module_list, spdk_bdev_module) bdev_modules;
101 :
102 : struct spdk_bdev_list bdevs;
103 : struct bdev_name_tree bdev_names;
104 :
105 : bool init_complete;
106 : bool module_init_complete;
107 :
108 : struct spdk_spinlock spinlock;
109 :
110 : TAILQ_HEAD(, spdk_bdev_open_async_ctx) async_bdev_opens;
111 :
112 : #ifdef SPDK_CONFIG_VTUNE
113 : __itt_domain *domain;
114 : #endif
115 : };
116 :
117 : static struct spdk_bdev_mgr g_bdev_mgr = {
118 : .bdev_modules = TAILQ_HEAD_INITIALIZER(g_bdev_mgr.bdev_modules),
119 : .bdevs = TAILQ_HEAD_INITIALIZER(g_bdev_mgr.bdevs),
120 : .bdev_names = RB_INITIALIZER(g_bdev_mgr.bdev_names),
121 : .init_complete = false,
122 : .module_init_complete = false,
123 : .async_bdev_opens = TAILQ_HEAD_INITIALIZER(g_bdev_mgr.async_bdev_opens),
124 : };
125 :
126 : static void
127 : __attribute__((constructor))
128 3 : _bdev_init(void)
129 : {
130 3 : spdk_spin_init(&g_bdev_mgr.spinlock);
131 3 : }
132 :
133 : typedef void (*lock_range_cb)(struct lba_range *range, void *ctx, int status);
134 :
135 : typedef void (*bdev_copy_bounce_buffer_cpl)(void *ctx, int rc);
136 :
137 : struct lba_range {
138 : struct spdk_bdev *bdev;
139 : uint64_t offset;
140 : uint64_t length;
141 : bool quiesce;
142 : void *locked_ctx;
143 : struct spdk_thread *owner_thread;
144 : struct spdk_bdev_channel *owner_ch;
145 : TAILQ_ENTRY(lba_range) tailq;
146 : TAILQ_ENTRY(lba_range) tailq_module;
147 : };
148 :
149 : static struct spdk_bdev_opts g_bdev_opts = {
150 : .bdev_io_pool_size = SPDK_BDEV_IO_POOL_SIZE,
151 : .bdev_io_cache_size = SPDK_BDEV_IO_CACHE_SIZE,
152 : .bdev_auto_examine = SPDK_BDEV_AUTO_EXAMINE,
153 : .iobuf_small_cache_size = BUF_SMALL_CACHE_SIZE,
154 : .iobuf_large_cache_size = BUF_LARGE_CACHE_SIZE,
155 : };
156 :
157 : static spdk_bdev_init_cb g_init_cb_fn = NULL;
158 : static void *g_init_cb_arg = NULL;
159 :
160 : static spdk_bdev_fini_cb g_fini_cb_fn = NULL;
161 : static void *g_fini_cb_arg = NULL;
162 : static struct spdk_thread *g_fini_thread = NULL;
163 :
164 : struct spdk_bdev_qos_limit {
165 : /** IOs or bytes allowed per second (i.e., 1s). */
166 : uint64_t limit;
167 :
168 : /** Remaining IOs or bytes allowed in current timeslice (e.g., 1ms).
169 : * For remaining bytes, allowed to run negative if an I/O is submitted when
170 : * some bytes are remaining, but the I/O is bigger than that amount. The
171 : * excess will be deducted from the next timeslice.
172 : */
173 : int64_t remaining_this_timeslice;
174 :
175 : /** Minimum allowed IOs or bytes to be issued in one timeslice (e.g., 1ms). */
176 : uint32_t min_per_timeslice;
177 :
178 : /** Maximum allowed IOs or bytes to be issued in one timeslice (e.g., 1ms). */
179 : uint32_t max_per_timeslice;
180 :
181 : /** Function to check whether to queue the IO.
182 : * If The IO is allowed to pass, the quota will be reduced correspondingly.
183 : */
184 : bool (*queue_io)(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io);
185 :
186 : /** Function to rewind the quota once the IO was allowed to be sent by this
187 : * limit but queued due to one of the further limits.
188 : */
189 : void (*rewind_quota)(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io);
190 : };
191 :
192 : struct spdk_bdev_qos {
193 : /** Types of structure of rate limits. */
194 : struct spdk_bdev_qos_limit rate_limits[SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES];
195 :
196 : /** The channel that all I/O are funneled through. */
197 : struct spdk_bdev_channel *ch;
198 :
199 : /** The thread on which the poller is running. */
200 : struct spdk_thread *thread;
201 :
202 : /** Size of a timeslice in tsc ticks. */
203 : uint64_t timeslice_size;
204 :
205 : /** Timestamp of start of last timeslice. */
206 : uint64_t last_timeslice;
207 :
208 : /** Poller that processes queued I/O commands each time slice. */
209 : struct spdk_poller *poller;
210 : };
211 :
212 : struct spdk_bdev_mgmt_channel {
213 : /*
214 : * Each thread keeps a cache of bdev_io - this allows
215 : * bdev threads which are *not* DPDK threads to still
216 : * benefit from a per-thread bdev_io cache. Without
217 : * this, non-DPDK threads fetching from the mempool
218 : * incur a cmpxchg on get and put.
219 : */
220 : bdev_io_stailq_t per_thread_cache;
221 : uint32_t per_thread_cache_count;
222 : uint32_t bdev_io_cache_size;
223 :
224 : struct spdk_iobuf_channel iobuf;
225 :
226 : TAILQ_HEAD(, spdk_bdev_shared_resource) shared_resources;
227 : TAILQ_HEAD(, spdk_bdev_io_wait_entry) io_wait_queue;
228 : };
229 :
230 : /*
231 : * Per-module (or per-io_device) data. Multiple bdevs built on the same io_device
232 : * will queue here their IO that awaits retry. It makes it possible to retry sending
233 : * IO to one bdev after IO from other bdev completes.
234 : */
235 : struct spdk_bdev_shared_resource {
236 : /* The bdev management channel */
237 : struct spdk_bdev_mgmt_channel *mgmt_ch;
238 :
239 : /*
240 : * Count of I/O submitted to bdev module and waiting for completion.
241 : * Incremented before submit_request() is called on an spdk_bdev_io.
242 : */
243 : uint64_t io_outstanding;
244 :
245 : /*
246 : * Queue of IO awaiting retry because of a previous NOMEM status returned
247 : * on this channel.
248 : */
249 : bdev_io_tailq_t nomem_io;
250 :
251 : /*
252 : * Threshold which io_outstanding must drop to before retrying nomem_io.
253 : */
254 : uint64_t nomem_threshold;
255 :
256 : /* I/O channel allocated by a bdev module */
257 : struct spdk_io_channel *shared_ch;
258 :
259 : struct spdk_poller *nomem_poller;
260 :
261 : /* Refcount of bdev channels using this resource */
262 : uint32_t ref;
263 :
264 : TAILQ_ENTRY(spdk_bdev_shared_resource) link;
265 : };
266 :
267 : #define BDEV_CH_RESET_IN_PROGRESS (1 << 0)
268 : #define BDEV_CH_QOS_ENABLED (1 << 1)
269 :
270 : struct spdk_bdev_channel {
271 : struct spdk_bdev *bdev;
272 :
273 : /* The channel for the underlying device */
274 : struct spdk_io_channel *channel;
275 :
276 : /* Accel channel */
277 : struct spdk_io_channel *accel_channel;
278 :
279 : /* Per io_device per thread data */
280 : struct spdk_bdev_shared_resource *shared_resource;
281 :
282 : struct spdk_bdev_io_stat *stat;
283 :
284 : /*
285 : * Count of I/O submitted to the underlying dev module through this channel
286 : * and waiting for completion.
287 : */
288 : uint64_t io_outstanding;
289 :
290 : /*
291 : * List of all submitted I/Os including I/O that are generated via splitting.
292 : */
293 : bdev_io_tailq_t io_submitted;
294 :
295 : /*
296 : * List of spdk_bdev_io that are currently queued because they write to a locked
297 : * LBA range.
298 : */
299 : bdev_io_tailq_t io_locked;
300 :
301 : /* List of I/Os with accel sequence being currently executed */
302 : bdev_io_tailq_t io_accel_exec;
303 :
304 : /* List of I/Os doing memory domain pull/push */
305 : bdev_io_tailq_t io_memory_domain;
306 :
307 : uint32_t flags;
308 :
309 : struct spdk_histogram_data *histogram;
310 :
311 : #ifdef SPDK_CONFIG_VTUNE
312 : uint64_t start_tsc;
313 : uint64_t interval_tsc;
314 : __itt_string_handle *handle;
315 : struct spdk_bdev_io_stat *prev_stat;
316 : #endif
317 :
318 : bdev_io_tailq_t queued_resets;
319 :
320 : lba_range_tailq_t locked_ranges;
321 :
322 : /** List of I/Os queued by QoS. */
323 : bdev_io_tailq_t qos_queued_io;
324 : };
325 :
326 : struct media_event_entry {
327 : struct spdk_bdev_media_event event;
328 : TAILQ_ENTRY(media_event_entry) tailq;
329 : };
330 :
331 : #define MEDIA_EVENT_POOL_SIZE 64
332 :
333 : struct spdk_bdev_desc {
334 : struct spdk_bdev *bdev;
335 : struct spdk_thread *thread;
336 : struct {
337 : spdk_bdev_event_cb_t event_fn;
338 : void *ctx;
339 : } callback;
340 : bool closed;
341 : bool write;
342 : bool memory_domains_supported;
343 : bool accel_sequence_supported[SPDK_BDEV_NUM_IO_TYPES];
344 : struct spdk_spinlock spinlock;
345 : uint32_t refs;
346 : TAILQ_HEAD(, media_event_entry) pending_media_events;
347 : TAILQ_HEAD(, media_event_entry) free_media_events;
348 : struct media_event_entry *media_events_buffer;
349 : TAILQ_ENTRY(spdk_bdev_desc) link;
350 :
351 : uint64_t timeout_in_sec;
352 : spdk_bdev_io_timeout_cb cb_fn;
353 : void *cb_arg;
354 : struct spdk_poller *io_timeout_poller;
355 : struct spdk_bdev_module_claim *claim;
356 : };
357 :
358 : struct spdk_bdev_iostat_ctx {
359 : struct spdk_bdev_io_stat *stat;
360 : spdk_bdev_get_device_stat_cb cb;
361 : void *cb_arg;
362 : };
363 :
364 : struct set_qos_limit_ctx {
365 : void (*cb_fn)(void *cb_arg, int status);
366 : void *cb_arg;
367 : struct spdk_bdev *bdev;
368 : };
369 :
370 : struct spdk_bdev_channel_iter {
371 : spdk_bdev_for_each_channel_msg fn;
372 : spdk_bdev_for_each_channel_done cpl;
373 : struct spdk_io_channel_iter *i;
374 : void *ctx;
375 : };
376 :
377 : struct spdk_bdev_io_error_stat {
378 : uint32_t error_status[-SPDK_MIN_BDEV_IO_STATUS];
379 : };
380 :
381 : enum bdev_io_retry_state {
382 : BDEV_IO_RETRY_STATE_INVALID,
383 : BDEV_IO_RETRY_STATE_PULL,
384 : BDEV_IO_RETRY_STATE_PULL_MD,
385 : BDEV_IO_RETRY_STATE_SUBMIT,
386 : BDEV_IO_RETRY_STATE_PUSH,
387 : BDEV_IO_RETRY_STATE_PUSH_MD,
388 : };
389 :
390 : #define __bdev_to_io_dev(bdev) (((char *)bdev) + 1)
391 : #define __bdev_from_io_dev(io_dev) ((struct spdk_bdev *)(((char *)io_dev) - 1))
392 : #define __io_ch_to_bdev_ch(io_ch) ((struct spdk_bdev_channel *)spdk_io_channel_get_ctx(io_ch))
393 : #define __io_ch_to_bdev_mgmt_ch(io_ch) ((struct spdk_bdev_mgmt_channel *)spdk_io_channel_get_ctx(io_ch))
394 :
395 : static inline void bdev_io_complete(void *ctx);
396 : static inline void bdev_io_complete_unsubmitted(struct spdk_bdev_io *bdev_io);
397 : static void bdev_io_push_bounce_md_buf(struct spdk_bdev_io *bdev_io);
398 : static void bdev_io_push_bounce_data(struct spdk_bdev_io *bdev_io);
399 :
400 : static void bdev_write_zero_buffer_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg);
401 : static int bdev_write_zero_buffer(struct spdk_bdev_io *bdev_io);
402 :
403 : static void bdev_enable_qos_msg(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
404 : struct spdk_io_channel *ch, void *_ctx);
405 : static void bdev_enable_qos_done(struct spdk_bdev *bdev, void *_ctx, int status);
406 :
407 : static int bdev_readv_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
408 : struct iovec *iov, int iovcnt, void *md_buf, uint64_t offset_blocks,
409 : uint64_t num_blocks,
410 : struct spdk_memory_domain *domain, void *domain_ctx,
411 : struct spdk_accel_sequence *seq,
412 : spdk_bdev_io_completion_cb cb, void *cb_arg);
413 : static int bdev_writev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
414 : struct iovec *iov, int iovcnt, void *md_buf,
415 : uint64_t offset_blocks, uint64_t num_blocks,
416 : struct spdk_memory_domain *domain, void *domain_ctx,
417 : struct spdk_accel_sequence *seq,
418 : spdk_bdev_io_completion_cb cb, void *cb_arg);
419 :
420 : static int bdev_lock_lba_range(struct spdk_bdev_desc *desc, struct spdk_io_channel *_ch,
421 : uint64_t offset, uint64_t length,
422 : lock_range_cb cb_fn, void *cb_arg);
423 :
424 : static int bdev_unlock_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 bool bdev_abort_queued_io(bdev_io_tailq_t *queue, struct spdk_bdev_io *bio_to_abort);
429 : static bool bdev_abort_buf_io(struct spdk_bdev_mgmt_channel *ch, struct spdk_bdev_io *bio_to_abort);
430 :
431 : static bool claim_type_is_v2(enum spdk_bdev_claim_type type);
432 : static void bdev_desc_release_claims(struct spdk_bdev_desc *desc);
433 : static void claim_reset(struct spdk_bdev *bdev);
434 :
435 : static void bdev_ch_retry_io(struct spdk_bdev_channel *bdev_ch);
436 :
437 : #define bdev_get_ext_io_opt(opts, field, defval) \
438 : (((opts) != NULL && offsetof(struct spdk_bdev_ext_io_opts, field) + \
439 : sizeof((opts)->field) <= (opts)->size) ? (opts)->field : (defval))
440 :
441 : void
442 13 : spdk_bdev_get_opts(struct spdk_bdev_opts *opts, size_t opts_size)
443 : {
444 13 : if (!opts) {
445 0 : SPDK_ERRLOG("opts should not be NULL\n");
446 0 : return;
447 : }
448 :
449 13 : if (!opts_size) {
450 0 : SPDK_ERRLOG("opts_size should not be zero value\n");
451 0 : return;
452 : }
453 :
454 13 : opts->opts_size = opts_size;
455 :
456 : #define SET_FIELD(field) \
457 : if (offsetof(struct spdk_bdev_opts, field) + sizeof(opts->field) <= opts_size) { \
458 : opts->field = g_bdev_opts.field; \
459 : } \
460 :
461 13 : SET_FIELD(bdev_io_pool_size);
462 13 : SET_FIELD(bdev_io_cache_size);
463 13 : SET_FIELD(bdev_auto_examine);
464 13 : SET_FIELD(iobuf_small_cache_size);
465 13 : SET_FIELD(iobuf_large_cache_size);
466 :
467 : /* Do not remove this statement, you should always update this statement when you adding a new field,
468 : * and do not forget to add the SET_FIELD statement for your added field. */
469 : SPDK_STATIC_ASSERT(sizeof(struct spdk_bdev_opts) == 32, "Incorrect size");
470 :
471 : #undef SET_FIELD
472 : }
473 :
474 : int
475 14 : spdk_bdev_set_opts(struct spdk_bdev_opts *opts)
476 : {
477 : uint32_t min_pool_size;
478 :
479 14 : if (!opts) {
480 0 : SPDK_ERRLOG("opts cannot be NULL\n");
481 0 : return -1;
482 : }
483 :
484 14 : if (!opts->opts_size) {
485 1 : SPDK_ERRLOG("opts_size inside opts cannot be zero value\n");
486 1 : return -1;
487 : }
488 :
489 : /*
490 : * Add 1 to the thread count to account for the extra mgmt_ch that gets created during subsystem
491 : * initialization. A second mgmt_ch will be created on the same thread when the application starts
492 : * but before the deferred put_io_channel event is executed for the first mgmt_ch.
493 : */
494 13 : min_pool_size = opts->bdev_io_cache_size * (spdk_thread_get_count() + 1);
495 13 : if (opts->bdev_io_pool_size < min_pool_size) {
496 0 : SPDK_ERRLOG("bdev_io_pool_size %" PRIu32 " is not compatible with bdev_io_cache_size %" PRIu32
497 : " and %" PRIu32 " threads\n", opts->bdev_io_pool_size, opts->bdev_io_cache_size,
498 : spdk_thread_get_count());
499 0 : SPDK_ERRLOG("bdev_io_pool_size must be at least %" PRIu32 "\n", min_pool_size);
500 0 : return -1;
501 : }
502 :
503 : #define SET_FIELD(field) \
504 : if (offsetof(struct spdk_bdev_opts, field) + sizeof(opts->field) <= opts->opts_size) { \
505 : g_bdev_opts.field = opts->field; \
506 : } \
507 :
508 13 : SET_FIELD(bdev_io_pool_size);
509 13 : SET_FIELD(bdev_io_cache_size);
510 13 : SET_FIELD(bdev_auto_examine);
511 13 : SET_FIELD(iobuf_small_cache_size);
512 13 : SET_FIELD(iobuf_large_cache_size);
513 :
514 13 : g_bdev_opts.opts_size = opts->opts_size;
515 :
516 : #undef SET_FIELD
517 :
518 13 : return 0;
519 : }
520 :
521 : static struct spdk_bdev *
522 143 : bdev_get_by_name(const char *bdev_name)
523 : {
524 143 : struct spdk_bdev_name find;
525 : struct spdk_bdev_name *res;
526 :
527 143 : find.name = (char *)bdev_name;
528 143 : res = RB_FIND(bdev_name_tree, &g_bdev_mgr.bdev_names, &find);
529 143 : if (res != NULL) {
530 136 : return res->bdev;
531 : }
532 :
533 7 : return NULL;
534 : }
535 :
536 : struct spdk_bdev *
537 16 : spdk_bdev_get_by_name(const char *bdev_name)
538 : {
539 : struct spdk_bdev *bdev;
540 :
541 16 : spdk_spin_lock(&g_bdev_mgr.spinlock);
542 16 : bdev = bdev_get_by_name(bdev_name);
543 16 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
544 :
545 16 : return bdev;
546 : }
547 :
548 : struct bdev_io_status_string {
549 : enum spdk_bdev_io_status status;
550 : const char *str;
551 : };
552 :
553 : static const struct bdev_io_status_string bdev_io_status_strings[] = {
554 : { SPDK_BDEV_IO_STATUS_AIO_ERROR, "aio_error" },
555 : { SPDK_BDEV_IO_STATUS_ABORTED, "aborted" },
556 : { SPDK_BDEV_IO_STATUS_FIRST_FUSED_FAILED, "first_fused_failed" },
557 : { SPDK_BDEV_IO_STATUS_MISCOMPARE, "miscompare" },
558 : { SPDK_BDEV_IO_STATUS_NOMEM, "nomem" },
559 : { SPDK_BDEV_IO_STATUS_SCSI_ERROR, "scsi_error" },
560 : { SPDK_BDEV_IO_STATUS_NVME_ERROR, "nvme_error" },
561 : { SPDK_BDEV_IO_STATUS_FAILED, "failed" },
562 : { SPDK_BDEV_IO_STATUS_PENDING, "pending" },
563 : { SPDK_BDEV_IO_STATUS_SUCCESS, "success" },
564 : };
565 :
566 : static const char *
567 0 : bdev_io_status_get_string(enum spdk_bdev_io_status status)
568 : {
569 : uint32_t i;
570 :
571 0 : for (i = 0; i < SPDK_COUNTOF(bdev_io_status_strings); i++) {
572 0 : if (bdev_io_status_strings[i].status == status) {
573 0 : return bdev_io_status_strings[i].str;
574 : }
575 : }
576 :
577 0 : return "reserved";
578 : }
579 :
580 : struct spdk_bdev_wait_for_examine_ctx {
581 : struct spdk_poller *poller;
582 : spdk_bdev_wait_for_examine_cb cb_fn;
583 : void *cb_arg;
584 : };
585 :
586 : static bool bdev_module_all_actions_completed(void);
587 :
588 : static int
589 189 : bdev_wait_for_examine_cb(void *arg)
590 : {
591 189 : struct spdk_bdev_wait_for_examine_ctx *ctx = arg;
592 :
593 189 : if (!bdev_module_all_actions_completed()) {
594 0 : return SPDK_POLLER_IDLE;
595 : }
596 :
597 189 : spdk_poller_unregister(&ctx->poller);
598 189 : ctx->cb_fn(ctx->cb_arg);
599 189 : free(ctx);
600 :
601 189 : return SPDK_POLLER_BUSY;
602 : }
603 :
604 : int
605 189 : spdk_bdev_wait_for_examine(spdk_bdev_wait_for_examine_cb cb_fn, void *cb_arg)
606 : {
607 : struct spdk_bdev_wait_for_examine_ctx *ctx;
608 :
609 189 : ctx = calloc(1, sizeof(*ctx));
610 189 : if (ctx == NULL) {
611 0 : return -ENOMEM;
612 : }
613 189 : ctx->cb_fn = cb_fn;
614 189 : ctx->cb_arg = cb_arg;
615 189 : ctx->poller = SPDK_POLLER_REGISTER(bdev_wait_for_examine_cb, ctx, 0);
616 :
617 189 : return 0;
618 : }
619 :
620 : struct spdk_bdev_examine_item {
621 : char *name;
622 : TAILQ_ENTRY(spdk_bdev_examine_item) link;
623 : };
624 :
625 : TAILQ_HEAD(spdk_bdev_examine_allowlist, spdk_bdev_examine_item);
626 :
627 : struct spdk_bdev_examine_allowlist g_bdev_examine_allowlist = TAILQ_HEAD_INITIALIZER(
628 : g_bdev_examine_allowlist);
629 :
630 : static inline bool
631 6 : bdev_examine_allowlist_check(const char *name)
632 : {
633 : struct spdk_bdev_examine_item *item;
634 6 : TAILQ_FOREACH(item, &g_bdev_examine_allowlist, link) {
635 0 : if (strcmp(name, item->name) == 0) {
636 0 : return true;
637 : }
638 : }
639 6 : return false;
640 : }
641 :
642 : static inline void
643 64 : bdev_examine_allowlist_free(void)
644 : {
645 : struct spdk_bdev_examine_item *item;
646 64 : while (!TAILQ_EMPTY(&g_bdev_examine_allowlist)) {
647 0 : item = TAILQ_FIRST(&g_bdev_examine_allowlist);
648 0 : TAILQ_REMOVE(&g_bdev_examine_allowlist, item, link);
649 0 : free(item->name);
650 0 : free(item);
651 : }
652 64 : }
653 :
654 : static inline bool
655 3 : bdev_in_examine_allowlist(struct spdk_bdev *bdev)
656 : {
657 : struct spdk_bdev_alias *tmp;
658 3 : if (bdev_examine_allowlist_check(bdev->name)) {
659 0 : return true;
660 : }
661 6 : TAILQ_FOREACH(tmp, &bdev->aliases, tailq) {
662 3 : if (bdev_examine_allowlist_check(tmp->alias.name)) {
663 0 : return true;
664 : }
665 : }
666 3 : return false;
667 : }
668 :
669 : static inline bool
670 123 : bdev_ok_to_examine(struct spdk_bdev *bdev)
671 : {
672 123 : if (g_bdev_opts.bdev_auto_examine) {
673 120 : return true;
674 : } else {
675 3 : return bdev_in_examine_allowlist(bdev);
676 : }
677 : }
678 :
679 : static void
680 123 : bdev_examine(struct spdk_bdev *bdev)
681 : {
682 : struct spdk_bdev_module *module;
683 : struct spdk_bdev_module_claim *claim, *tmpclaim;
684 : uint32_t action;
685 :
686 123 : if (!bdev_ok_to_examine(bdev)) {
687 3 : return;
688 : }
689 :
690 489 : TAILQ_FOREACH(module, &g_bdev_mgr.bdev_modules, internal.tailq) {
691 369 : if (module->examine_config) {
692 249 : spdk_spin_lock(&module->internal.spinlock);
693 249 : action = module->internal.action_in_progress;
694 249 : module->internal.action_in_progress++;
695 249 : spdk_spin_unlock(&module->internal.spinlock);
696 249 : module->examine_config(bdev);
697 249 : if (action != module->internal.action_in_progress) {
698 0 : SPDK_ERRLOG("examine_config for module %s did not call "
699 : "spdk_bdev_module_examine_done()\n", module->name);
700 : }
701 : }
702 : }
703 :
704 120 : spdk_spin_lock(&bdev->internal.spinlock);
705 :
706 120 : switch (bdev->internal.claim_type) {
707 115 : case SPDK_BDEV_CLAIM_NONE:
708 : /* Examine by all bdev modules */
709 464 : TAILQ_FOREACH(module, &g_bdev_mgr.bdev_modules, internal.tailq) {
710 349 : if (module->examine_disk) {
711 225 : spdk_spin_lock(&module->internal.spinlock);
712 225 : module->internal.action_in_progress++;
713 225 : spdk_spin_unlock(&module->internal.spinlock);
714 225 : spdk_spin_unlock(&bdev->internal.spinlock);
715 225 : module->examine_disk(bdev);
716 225 : spdk_spin_lock(&bdev->internal.spinlock);
717 : }
718 : }
719 115 : break;
720 1 : case SPDK_BDEV_CLAIM_EXCL_WRITE:
721 : /* Examine by the one bdev module with a v1 claim */
722 1 : module = bdev->internal.claim.v1.module;
723 1 : if (module->examine_disk) {
724 1 : spdk_spin_lock(&module->internal.spinlock);
725 1 : module->internal.action_in_progress++;
726 1 : spdk_spin_unlock(&module->internal.spinlock);
727 1 : spdk_spin_unlock(&bdev->internal.spinlock);
728 1 : module->examine_disk(bdev);
729 1 : return;
730 : }
731 0 : break;
732 4 : default:
733 : /* Examine by all bdev modules with a v2 claim */
734 4 : assert(claim_type_is_v2(bdev->internal.claim_type));
735 : /*
736 : * Removal of tailq nodes while iterating can cause the iteration to jump out of the
737 : * list, perhaps accessing freed memory. Without protection, this could happen
738 : * while the lock is dropped during the examine callback.
739 : */
740 4 : bdev->internal.examine_in_progress++;
741 :
742 9 : TAILQ_FOREACH(claim, &bdev->internal.claim.v2.claims, link) {
743 5 : module = claim->module;
744 :
745 5 : if (module == NULL) {
746 : /* This is a vestigial claim, held by examine_count */
747 0 : continue;
748 : }
749 :
750 5 : if (module->examine_disk == NULL) {
751 0 : continue;
752 : }
753 :
754 5 : spdk_spin_lock(&module->internal.spinlock);
755 5 : module->internal.action_in_progress++;
756 5 : spdk_spin_unlock(&module->internal.spinlock);
757 :
758 : /* Call examine_disk without holding internal.spinlock. */
759 5 : spdk_spin_unlock(&bdev->internal.spinlock);
760 5 : module->examine_disk(bdev);
761 5 : spdk_spin_lock(&bdev->internal.spinlock);
762 : }
763 :
764 4 : assert(bdev->internal.examine_in_progress > 0);
765 4 : bdev->internal.examine_in_progress--;
766 4 : if (bdev->internal.examine_in_progress == 0) {
767 : /* Remove any claims that were released during examine_disk */
768 9 : TAILQ_FOREACH_SAFE(claim, &bdev->internal.claim.v2.claims, link, tmpclaim) {
769 5 : if (claim->desc != NULL) {
770 5 : continue;
771 : }
772 :
773 0 : TAILQ_REMOVE(&bdev->internal.claim.v2.claims, claim, link);
774 0 : free(claim);
775 : }
776 4 : if (TAILQ_EMPTY(&bdev->internal.claim.v2.claims)) {
777 0 : claim_reset(bdev);
778 : }
779 : }
780 : }
781 :
782 119 : spdk_spin_unlock(&bdev->internal.spinlock);
783 : }
784 :
785 : int
786 1 : spdk_bdev_examine(const char *name)
787 : {
788 : struct spdk_bdev *bdev;
789 : struct spdk_bdev_examine_item *item;
790 1 : struct spdk_thread *thread = spdk_get_thread();
791 :
792 1 : if (spdk_unlikely(!spdk_thread_is_app_thread(thread))) {
793 1 : SPDK_ERRLOG("Cannot examine bdev %s on thread %p (%s)\n", name, thread,
794 : thread ? spdk_thread_get_name(thread) : "null");
795 1 : return -EINVAL;
796 : }
797 :
798 0 : if (g_bdev_opts.bdev_auto_examine) {
799 0 : SPDK_ERRLOG("Manual examine is not allowed if auto examine is enabled");
800 0 : return -EINVAL;
801 : }
802 :
803 0 : if (bdev_examine_allowlist_check(name)) {
804 0 : SPDK_ERRLOG("Duplicate bdev name for manual examine: %s\n", name);
805 0 : return -EEXIST;
806 : }
807 :
808 0 : item = calloc(1, sizeof(*item));
809 0 : if (!item) {
810 0 : return -ENOMEM;
811 : }
812 0 : item->name = strdup(name);
813 0 : if (!item->name) {
814 0 : free(item);
815 0 : return -ENOMEM;
816 : }
817 0 : TAILQ_INSERT_TAIL(&g_bdev_examine_allowlist, item, link);
818 :
819 0 : bdev = spdk_bdev_get_by_name(name);
820 0 : if (bdev) {
821 0 : bdev_examine(bdev);
822 : }
823 0 : return 0;
824 : }
825 :
826 : static inline void
827 0 : bdev_examine_allowlist_config_json(struct spdk_json_write_ctx *w)
828 : {
829 : struct spdk_bdev_examine_item *item;
830 0 : TAILQ_FOREACH(item, &g_bdev_examine_allowlist, link) {
831 0 : spdk_json_write_object_begin(w);
832 0 : spdk_json_write_named_string(w, "method", "bdev_examine");
833 0 : spdk_json_write_named_object_begin(w, "params");
834 0 : spdk_json_write_named_string(w, "name", item->name);
835 0 : spdk_json_write_object_end(w);
836 0 : spdk_json_write_object_end(w);
837 : }
838 0 : }
839 :
840 : struct spdk_bdev *
841 1 : spdk_bdev_first(void)
842 : {
843 : struct spdk_bdev *bdev;
844 :
845 1 : bdev = TAILQ_FIRST(&g_bdev_mgr.bdevs);
846 1 : if (bdev) {
847 1 : SPDK_DEBUGLOG(bdev, "Starting bdev iteration at %s\n", bdev->name);
848 : }
849 :
850 1 : return bdev;
851 : }
852 :
853 : struct spdk_bdev *
854 8 : spdk_bdev_next(struct spdk_bdev *prev)
855 : {
856 : struct spdk_bdev *bdev;
857 :
858 8 : bdev = TAILQ_NEXT(prev, internal.link);
859 8 : if (bdev) {
860 7 : SPDK_DEBUGLOG(bdev, "Continuing bdev iteration at %s\n", bdev->name);
861 : }
862 :
863 8 : return bdev;
864 : }
865 :
866 : static struct spdk_bdev *
867 6 : _bdev_next_leaf(struct spdk_bdev *bdev)
868 : {
869 9 : while (bdev != NULL) {
870 8 : if (bdev->internal.claim_type == SPDK_BDEV_CLAIM_NONE) {
871 5 : return bdev;
872 : } else {
873 3 : bdev = TAILQ_NEXT(bdev, internal.link);
874 : }
875 : }
876 :
877 1 : return bdev;
878 : }
879 :
880 : struct spdk_bdev *
881 1 : spdk_bdev_first_leaf(void)
882 : {
883 : struct spdk_bdev *bdev;
884 :
885 1 : bdev = _bdev_next_leaf(TAILQ_FIRST(&g_bdev_mgr.bdevs));
886 :
887 1 : if (bdev) {
888 1 : SPDK_DEBUGLOG(bdev, "Starting bdev iteration at %s\n", bdev->name);
889 : }
890 :
891 1 : return bdev;
892 : }
893 :
894 : struct spdk_bdev *
895 5 : spdk_bdev_next_leaf(struct spdk_bdev *prev)
896 : {
897 : struct spdk_bdev *bdev;
898 :
899 5 : bdev = _bdev_next_leaf(TAILQ_NEXT(prev, internal.link));
900 :
901 5 : if (bdev) {
902 4 : SPDK_DEBUGLOG(bdev, "Continuing bdev iteration at %s\n", bdev->name);
903 : }
904 :
905 5 : return bdev;
906 : }
907 :
908 : static inline bool
909 34 : bdev_io_use_memory_domain(struct spdk_bdev_io *bdev_io)
910 : {
911 34 : return bdev_io->internal.memory_domain;
912 : }
913 :
914 : static inline bool
915 93 : bdev_io_use_accel_sequence(struct spdk_bdev_io *bdev_io)
916 : {
917 93 : return bdev_io->internal.has_accel_sequence;
918 : }
919 :
920 : static inline void
921 7 : bdev_queue_nomem_io_head(struct spdk_bdev_shared_resource *shared_resource,
922 : struct spdk_bdev_io *bdev_io, enum bdev_io_retry_state state)
923 : {
924 : /* Wait for some of the outstanding I/O to complete before we retry any of the nomem_io.
925 : * Normally we will wait for NOMEM_THRESHOLD_COUNT I/O to complete but for low queue depth
926 : * channels we will instead wait for half to complete.
927 : */
928 7 : shared_resource->nomem_threshold = spdk_max((int64_t)shared_resource->io_outstanding / 2,
929 : (int64_t)shared_resource->io_outstanding - NOMEM_THRESHOLD_COUNT);
930 :
931 7 : assert(state != BDEV_IO_RETRY_STATE_INVALID);
932 7 : bdev_io->internal.retry_state = state;
933 7 : TAILQ_INSERT_HEAD(&shared_resource->nomem_io, bdev_io, internal.link);
934 7 : }
935 :
936 : static inline void
937 43 : bdev_queue_nomem_io_tail(struct spdk_bdev_shared_resource *shared_resource,
938 : struct spdk_bdev_io *bdev_io, enum bdev_io_retry_state state)
939 : {
940 : /* We only queue IOs at the end of the nomem_io queue if they're submitted by the user while
941 : * the queue isn't empty, so we don't need to update the nomem_threshold here */
942 43 : assert(!TAILQ_EMPTY(&shared_resource->nomem_io));
943 :
944 43 : assert(state != BDEV_IO_RETRY_STATE_INVALID);
945 43 : bdev_io->internal.retry_state = state;
946 43 : TAILQ_INSERT_TAIL(&shared_resource->nomem_io, bdev_io, internal.link);
947 43 : }
948 :
949 : void
950 18 : spdk_bdev_io_set_buf(struct spdk_bdev_io *bdev_io, void *buf, size_t len)
951 : {
952 : struct iovec *iovs;
953 :
954 18 : if (bdev_io->u.bdev.iovs == NULL) {
955 3 : bdev_io->u.bdev.iovs = &bdev_io->iov;
956 3 : bdev_io->u.bdev.iovcnt = 1;
957 : }
958 :
959 18 : iovs = bdev_io->u.bdev.iovs;
960 :
961 18 : assert(iovs != NULL);
962 18 : assert(bdev_io->u.bdev.iovcnt >= 1);
963 :
964 18 : iovs[0].iov_base = buf;
965 18 : iovs[0].iov_len = len;
966 18 : }
967 :
968 : void
969 3 : spdk_bdev_io_set_md_buf(struct spdk_bdev_io *bdev_io, void *md_buf, size_t len)
970 : {
971 3 : assert((len / spdk_bdev_get_md_size(bdev_io->bdev)) >= bdev_io->u.bdev.num_blocks);
972 3 : bdev_io->u.bdev.md_buf = md_buf;
973 3 : }
974 :
975 : static bool
976 179 : _is_buf_allocated(const struct iovec *iovs)
977 : {
978 179 : if (iovs == NULL) {
979 6 : return false;
980 : }
981 :
982 173 : return iovs[0].iov_base != NULL;
983 : }
984 :
985 : static bool
986 56 : _are_iovs_aligned(struct iovec *iovs, int iovcnt, uint32_t alignment)
987 : {
988 : int i;
989 : uintptr_t iov_base;
990 :
991 56 : if (spdk_likely(alignment == 1)) {
992 21 : return true;
993 : }
994 :
995 46 : for (i = 0; i < iovcnt; i++) {
996 35 : iov_base = (uintptr_t)iovs[i].iov_base;
997 35 : if ((iov_base & (alignment - 1)) != 0) {
998 24 : return false;
999 : }
1000 : }
1001 :
1002 11 : return true;
1003 : }
1004 :
1005 : static inline bool
1006 863 : bdev_io_needs_sequence_exec(struct spdk_bdev_desc *desc, struct spdk_bdev_io *bdev_io)
1007 : {
1008 863 : if (!bdev_io->internal.accel_sequence) {
1009 863 : return false;
1010 : }
1011 :
1012 : /* For now, we don't allow splitting IOs with an accel sequence and will treat them as if
1013 : * bdev module didn't support accel sequences */
1014 0 : return !desc->accel_sequence_supported[bdev_io->type] || bdev_io->internal.split;
1015 : }
1016 :
1017 : static inline void
1018 585 : bdev_io_increment_outstanding(struct spdk_bdev_channel *bdev_ch,
1019 : struct spdk_bdev_shared_resource *shared_resource)
1020 : {
1021 585 : bdev_ch->io_outstanding++;
1022 585 : shared_resource->io_outstanding++;
1023 585 : }
1024 :
1025 : static inline void
1026 585 : bdev_io_decrement_outstanding(struct spdk_bdev_channel *bdev_ch,
1027 : struct spdk_bdev_shared_resource *shared_resource)
1028 : {
1029 585 : assert(bdev_ch->io_outstanding > 0);
1030 585 : assert(shared_resource->io_outstanding > 0);
1031 585 : bdev_ch->io_outstanding--;
1032 585 : shared_resource->io_outstanding--;
1033 585 : }
1034 :
1035 : static void
1036 0 : bdev_io_submit_sequence_cb(void *ctx, int status)
1037 : {
1038 0 : struct spdk_bdev_io *bdev_io = ctx;
1039 :
1040 0 : bdev_io->u.bdev.accel_sequence = NULL;
1041 0 : bdev_io->internal.accel_sequence = NULL;
1042 :
1043 0 : if (spdk_unlikely(status != 0)) {
1044 0 : SPDK_ERRLOG("Failed to execute accel sequence, status=%d\n", status);
1045 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
1046 0 : bdev_io_complete_unsubmitted(bdev_io);
1047 0 : return;
1048 : }
1049 :
1050 0 : bdev_io_submit(bdev_io);
1051 : }
1052 :
1053 : static void
1054 0 : bdev_io_exec_sequence_cb(void *ctx, int status)
1055 : {
1056 0 : struct spdk_bdev_io *bdev_io = ctx;
1057 0 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1058 :
1059 0 : TAILQ_REMOVE(&bdev_io->internal.ch->io_accel_exec, bdev_io, internal.link);
1060 0 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1061 :
1062 0 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1063 0 : bdev_ch_retry_io(ch);
1064 : }
1065 :
1066 0 : bdev_io->internal.data_transfer_cpl(bdev_io, status);
1067 0 : }
1068 :
1069 : static void
1070 0 : bdev_io_exec_sequence(struct spdk_bdev_io *bdev_io, void (*cb_fn)(void *ctx, int status))
1071 : {
1072 0 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1073 :
1074 0 : assert(bdev_io_needs_sequence_exec(bdev_io->internal.desc, bdev_io));
1075 0 : assert(bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE || bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
1076 :
1077 : /* Since the operations are appended during submission, they're in the opposite order than
1078 : * how we want to execute them for reads (i.e. we need to execute the most recently added
1079 : * operation first), so reverse the sequence before executing it.
1080 : */
1081 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_READ) {
1082 0 : spdk_accel_sequence_reverse(bdev_io->internal.accel_sequence);
1083 : }
1084 :
1085 0 : TAILQ_INSERT_TAIL(&bdev_io->internal.ch->io_accel_exec, bdev_io, internal.link);
1086 0 : bdev_io_increment_outstanding(ch, ch->shared_resource);
1087 0 : bdev_io->internal.data_transfer_cpl = cb_fn;
1088 :
1089 0 : spdk_accel_sequence_finish(bdev_io->internal.accel_sequence,
1090 : bdev_io_exec_sequence_cb, bdev_io);
1091 0 : }
1092 :
1093 : static void
1094 46 : bdev_io_get_buf_complete(struct spdk_bdev_io *bdev_io, bool status)
1095 : {
1096 46 : struct spdk_io_channel *ch = spdk_bdev_io_get_io_channel(bdev_io);
1097 : void *buf;
1098 :
1099 46 : if (spdk_unlikely(bdev_io->internal.get_aux_buf_cb != NULL)) {
1100 0 : buf = bdev_io->internal.buf;
1101 0 : bdev_io->internal.buf = NULL;
1102 0 : bdev_io->internal.get_aux_buf_cb(ch, bdev_io, buf);
1103 0 : bdev_io->internal.get_aux_buf_cb = NULL;
1104 : } else {
1105 46 : assert(bdev_io->internal.get_buf_cb != NULL);
1106 46 : bdev_io->internal.get_buf_cb(ch, bdev_io, status);
1107 46 : bdev_io->internal.get_buf_cb = NULL;
1108 : }
1109 46 : }
1110 :
1111 : static void
1112 4 : _bdev_io_pull_buffer_cpl(void *ctx, int rc)
1113 : {
1114 4 : struct spdk_bdev_io *bdev_io = ctx;
1115 :
1116 4 : if (rc) {
1117 0 : SPDK_ERRLOG("Set bounce buffer failed with rc %d\n", rc);
1118 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
1119 : }
1120 4 : bdev_io_get_buf_complete(bdev_io, !rc);
1121 4 : }
1122 :
1123 : static void
1124 2 : bdev_io_pull_md_buf_done(void *ctx, int status)
1125 : {
1126 2 : struct spdk_bdev_io *bdev_io = ctx;
1127 2 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1128 :
1129 2 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1130 2 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1131 :
1132 2 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1133 0 : bdev_ch_retry_io(ch);
1134 : }
1135 :
1136 2 : assert(bdev_io->internal.data_transfer_cpl);
1137 2 : bdev_io->internal.data_transfer_cpl(bdev_io, status);
1138 2 : }
1139 :
1140 : static void
1141 4 : bdev_io_pull_md_buf(struct spdk_bdev_io *bdev_io)
1142 : {
1143 4 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1144 4 : int rc = 0;
1145 :
1146 4 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
1147 2 : if (bdev_io_use_memory_domain(bdev_io)) {
1148 2 : TAILQ_INSERT_TAIL(&ch->io_memory_domain, bdev_io, internal.link);
1149 2 : bdev_io_increment_outstanding(ch, ch->shared_resource);
1150 2 : rc = spdk_memory_domain_pull_data(bdev_io->internal.memory_domain,
1151 : bdev_io->internal.memory_domain_ctx,
1152 : &bdev_io->internal.orig_md_iov, 1,
1153 : &bdev_io->internal.bounce_md_iov, 1,
1154 : bdev_io_pull_md_buf_done, bdev_io);
1155 2 : if (rc == 0) {
1156 : /* Continue to submit IO in completion callback */
1157 2 : return;
1158 : }
1159 0 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1160 0 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1161 0 : if (rc != -ENOMEM) {
1162 0 : SPDK_ERRLOG("Failed to pull data from memory domain %s, rc %d\n",
1163 : spdk_memory_domain_get_dma_device_id(
1164 : bdev_io->internal.memory_domain), rc);
1165 : }
1166 : } else {
1167 0 : memcpy(bdev_io->internal.bounce_md_iov.iov_base,
1168 0 : bdev_io->internal.orig_md_iov.iov_base,
1169 : bdev_io->internal.orig_md_iov.iov_len);
1170 : }
1171 : }
1172 :
1173 2 : if (spdk_unlikely(rc == -ENOMEM)) {
1174 0 : bdev_queue_nomem_io_head(ch->shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PULL_MD);
1175 : } else {
1176 2 : assert(bdev_io->internal.data_transfer_cpl);
1177 2 : bdev_io->internal.data_transfer_cpl(bdev_io, rc);
1178 : }
1179 : }
1180 :
1181 : static void
1182 4 : _bdev_io_pull_bounce_md_buf(struct spdk_bdev_io *bdev_io, void *md_buf, size_t len)
1183 : {
1184 : /* save original md_buf */
1185 4 : bdev_io->internal.orig_md_iov.iov_base = bdev_io->u.bdev.md_buf;
1186 4 : bdev_io->internal.orig_md_iov.iov_len = len;
1187 4 : bdev_io->internal.bounce_md_iov.iov_base = md_buf;
1188 4 : bdev_io->internal.bounce_md_iov.iov_len = len;
1189 : /* set bounce md_buf */
1190 4 : bdev_io->u.bdev.md_buf = md_buf;
1191 :
1192 4 : bdev_io_pull_md_buf(bdev_io);
1193 4 : }
1194 :
1195 : static void
1196 46 : _bdev_io_set_md_buf(struct spdk_bdev_io *bdev_io)
1197 : {
1198 46 : struct spdk_bdev *bdev = bdev_io->bdev;
1199 : uint64_t md_len;
1200 : void *buf;
1201 :
1202 46 : if (spdk_bdev_is_md_separate(bdev)) {
1203 7 : assert(!bdev_io_use_accel_sequence(bdev_io));
1204 :
1205 7 : buf = (char *)bdev_io->u.bdev.iovs[0].iov_base + bdev_io->u.bdev.iovs[0].iov_len;
1206 7 : md_len = bdev_io->u.bdev.num_blocks * bdev->md_len;
1207 :
1208 7 : assert(((uintptr_t)buf & (spdk_bdev_get_buf_align(bdev) - 1)) == 0);
1209 :
1210 7 : if (bdev_io->u.bdev.md_buf != NULL) {
1211 4 : _bdev_io_pull_bounce_md_buf(bdev_io, buf, md_len);
1212 4 : return;
1213 : } else {
1214 3 : spdk_bdev_io_set_md_buf(bdev_io, buf, md_len);
1215 : }
1216 : }
1217 :
1218 42 : bdev_io_get_buf_complete(bdev_io, true);
1219 : }
1220 :
1221 : static inline void
1222 28 : bdev_io_pull_data_done(struct spdk_bdev_io *bdev_io, int rc)
1223 : {
1224 28 : if (rc) {
1225 0 : SPDK_ERRLOG("Failed to get data buffer\n");
1226 0 : assert(bdev_io->internal.data_transfer_cpl);
1227 0 : bdev_io->internal.data_transfer_cpl(bdev_io, rc);
1228 0 : return;
1229 : }
1230 :
1231 28 : _bdev_io_set_md_buf(bdev_io);
1232 : }
1233 :
1234 : static void
1235 2 : bdev_io_pull_data_done_and_track(void *ctx, int status)
1236 : {
1237 2 : struct spdk_bdev_io *bdev_io = ctx;
1238 2 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1239 :
1240 2 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1241 2 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1242 :
1243 2 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1244 0 : bdev_ch_retry_io(ch);
1245 : }
1246 :
1247 2 : bdev_io_pull_data_done(bdev_io, status);
1248 2 : }
1249 :
1250 : static void
1251 29 : bdev_io_pull_data(struct spdk_bdev_io *bdev_io)
1252 : {
1253 29 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1254 29 : int rc = 0;
1255 :
1256 : /* If we need to exec an accel sequence or the IO uses a memory domain buffer and has a
1257 : * sequence, append a copy operation making accel change the src/dst buffers of the previous
1258 : * operation */
1259 58 : if (bdev_io_needs_sequence_exec(bdev_io->internal.desc, bdev_io) ||
1260 29 : (bdev_io_use_accel_sequence(bdev_io) && bdev_io_use_memory_domain(bdev_io))) {
1261 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
1262 0 : rc = spdk_accel_append_copy(&bdev_io->internal.accel_sequence, ch->accel_channel,
1263 0 : bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt,
1264 : NULL, NULL,
1265 : bdev_io->internal.orig_iovs,
1266 0 : bdev_io->internal.orig_iovcnt,
1267 : bdev_io->internal.memory_domain,
1268 : bdev_io->internal.memory_domain_ctx,
1269 : 0, NULL, NULL);
1270 : } else {
1271 : /* We need to reverse the src/dst for reads */
1272 0 : assert(bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
1273 0 : rc = spdk_accel_append_copy(&bdev_io->internal.accel_sequence, ch->accel_channel,
1274 : bdev_io->internal.orig_iovs,
1275 0 : bdev_io->internal.orig_iovcnt,
1276 : bdev_io->internal.memory_domain,
1277 : bdev_io->internal.memory_domain_ctx,
1278 0 : bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt,
1279 : NULL, NULL, 0, NULL, NULL);
1280 : }
1281 :
1282 0 : if (spdk_unlikely(rc != 0 && rc != -ENOMEM)) {
1283 0 : SPDK_ERRLOG("Failed to append copy to accel sequence: %p\n",
1284 : bdev_io->internal.accel_sequence);
1285 : }
1286 29 : } else if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
1287 : /* if this is write path, copy data from original buffer to bounce buffer */
1288 17 : if (bdev_io_use_memory_domain(bdev_io)) {
1289 3 : TAILQ_INSERT_TAIL(&ch->io_memory_domain, bdev_io, internal.link);
1290 3 : bdev_io_increment_outstanding(ch, ch->shared_resource);
1291 3 : rc = spdk_memory_domain_pull_data(bdev_io->internal.memory_domain,
1292 : bdev_io->internal.memory_domain_ctx,
1293 : bdev_io->internal.orig_iovs,
1294 3 : (uint32_t) bdev_io->internal.orig_iovcnt,
1295 : bdev_io->u.bdev.iovs, 1,
1296 : bdev_io_pull_data_done_and_track,
1297 : bdev_io);
1298 3 : if (rc == 0) {
1299 : /* Continue to submit IO in completion callback */
1300 2 : return;
1301 : }
1302 1 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1303 1 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1304 1 : if (rc != -ENOMEM) {
1305 0 : SPDK_ERRLOG("Failed to pull data from memory domain %s\n",
1306 : spdk_memory_domain_get_dma_device_id(
1307 : bdev_io->internal.memory_domain));
1308 : }
1309 : } else {
1310 14 : assert(bdev_io->u.bdev.iovcnt == 1);
1311 14 : spdk_copy_iovs_to_buf(bdev_io->u.bdev.iovs[0].iov_base,
1312 14 : bdev_io->u.bdev.iovs[0].iov_len,
1313 : bdev_io->internal.orig_iovs,
1314 : bdev_io->internal.orig_iovcnt);
1315 : }
1316 : }
1317 :
1318 27 : if (spdk_unlikely(rc == -ENOMEM)) {
1319 1 : bdev_queue_nomem_io_head(ch->shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PULL);
1320 : } else {
1321 26 : bdev_io_pull_data_done(bdev_io, rc);
1322 : }
1323 : }
1324 :
1325 : static void
1326 28 : _bdev_io_pull_bounce_data_buf(struct spdk_bdev_io *bdev_io, void *buf, size_t len,
1327 : bdev_copy_bounce_buffer_cpl cpl_cb)
1328 : {
1329 28 : struct spdk_bdev_shared_resource *shared_resource = bdev_io->internal.ch->shared_resource;
1330 :
1331 28 : bdev_io->internal.data_transfer_cpl = cpl_cb;
1332 : /* save original iovec */
1333 28 : bdev_io->internal.orig_iovs = bdev_io->u.bdev.iovs;
1334 28 : bdev_io->internal.orig_iovcnt = bdev_io->u.bdev.iovcnt;
1335 : /* set bounce iov */
1336 28 : bdev_io->u.bdev.iovs = &bdev_io->internal.bounce_iov;
1337 28 : bdev_io->u.bdev.iovcnt = 1;
1338 : /* set bounce buffer for this operation */
1339 28 : bdev_io->u.bdev.iovs[0].iov_base = buf;
1340 28 : bdev_io->u.bdev.iovs[0].iov_len = len;
1341 :
1342 28 : if (spdk_unlikely(!TAILQ_EMPTY(&shared_resource->nomem_io))) {
1343 0 : bdev_queue_nomem_io_tail(shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PULL);
1344 : } else {
1345 28 : bdev_io_pull_data(bdev_io);
1346 : }
1347 28 : }
1348 :
1349 : static void
1350 46 : _bdev_io_set_buf(struct spdk_bdev_io *bdev_io, void *buf, uint64_t len)
1351 : {
1352 46 : struct spdk_bdev *bdev = bdev_io->bdev;
1353 : bool buf_allocated;
1354 : uint64_t alignment;
1355 : void *aligned_buf;
1356 :
1357 46 : bdev_io->internal.buf = buf;
1358 :
1359 46 : if (spdk_unlikely(bdev_io->internal.get_aux_buf_cb != NULL)) {
1360 0 : bdev_io_get_buf_complete(bdev_io, true);
1361 0 : return;
1362 : }
1363 :
1364 46 : alignment = spdk_bdev_get_buf_align(bdev);
1365 46 : buf_allocated = _is_buf_allocated(bdev_io->u.bdev.iovs);
1366 46 : aligned_buf = (void *)(((uintptr_t)buf + (alignment - 1)) & ~(alignment - 1));
1367 :
1368 46 : if (buf_allocated) {
1369 28 : _bdev_io_pull_bounce_data_buf(bdev_io, aligned_buf, len, _bdev_io_pull_buffer_cpl);
1370 : /* Continue in completion callback */
1371 28 : return;
1372 : } else {
1373 18 : spdk_bdev_io_set_buf(bdev_io, aligned_buf, len);
1374 : }
1375 :
1376 18 : _bdev_io_set_md_buf(bdev_io);
1377 : }
1378 :
1379 : static inline uint64_t
1380 92 : bdev_io_get_max_buf_len(struct spdk_bdev_io *bdev_io, uint64_t len)
1381 : {
1382 92 : struct spdk_bdev *bdev = bdev_io->bdev;
1383 : uint64_t md_len, alignment;
1384 :
1385 92 : md_len = spdk_bdev_is_md_separate(bdev) ? bdev_io->u.bdev.num_blocks * bdev->md_len : 0;
1386 :
1387 : /* 1 byte alignment needs 0 byte of extra space, 64 bytes alignment needs 63 bytes of extra space, etc. */
1388 92 : alignment = spdk_bdev_get_buf_align(bdev) - 1;
1389 :
1390 92 : return len + alignment + md_len;
1391 : }
1392 :
1393 : static void
1394 46 : _bdev_io_put_buf(struct spdk_bdev_io *bdev_io, void *buf, uint64_t buf_len)
1395 : {
1396 : struct spdk_bdev_mgmt_channel *ch;
1397 :
1398 46 : ch = bdev_io->internal.ch->shared_resource->mgmt_ch;
1399 46 : spdk_iobuf_put(&ch->iobuf, buf, bdev_io_get_max_buf_len(bdev_io, buf_len));
1400 46 : }
1401 :
1402 : static void
1403 46 : bdev_io_put_buf(struct spdk_bdev_io *bdev_io)
1404 : {
1405 46 : assert(bdev_io->internal.buf != NULL);
1406 46 : _bdev_io_put_buf(bdev_io, bdev_io->internal.buf, bdev_io->internal.buf_len);
1407 46 : bdev_io->internal.buf = NULL;
1408 46 : }
1409 :
1410 : void
1411 0 : spdk_bdev_io_put_aux_buf(struct spdk_bdev_io *bdev_io, void *buf)
1412 : {
1413 0 : uint64_t len = bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen;
1414 :
1415 0 : assert(buf != NULL);
1416 0 : _bdev_io_put_buf(bdev_io, buf, len);
1417 0 : }
1418 :
1419 : static inline void
1420 541 : bdev_submit_request(struct spdk_bdev *bdev, struct spdk_io_channel *ioch,
1421 : struct spdk_bdev_io *bdev_io)
1422 : {
1423 : /* After a request is submitted to a bdev module, the ownership of an accel sequence
1424 : * associated with that bdev_io is transferred to the bdev module. So, clear the internal
1425 : * sequence pointer to make sure we won't touch it anymore. */
1426 541 : if ((bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE ||
1427 541 : bdev_io->type == SPDK_BDEV_IO_TYPE_READ) && bdev_io->u.bdev.accel_sequence != NULL) {
1428 0 : assert(!bdev_io_needs_sequence_exec(bdev_io->internal.desc, bdev_io));
1429 0 : bdev_io->internal.accel_sequence = NULL;
1430 : }
1431 :
1432 541 : bdev->fn_table->submit_request(ioch, bdev_io);
1433 541 : }
1434 :
1435 : static inline void
1436 10 : bdev_ch_resubmit_io(struct spdk_bdev_shared_resource *shared_resource, struct spdk_bdev_io *bdev_io)
1437 : {
1438 10 : struct spdk_bdev *bdev = bdev_io->bdev;
1439 :
1440 10 : bdev_io_increment_outstanding(bdev_io->internal.ch, shared_resource);
1441 10 : bdev_io->internal.error.nvme.cdw0 = 0;
1442 10 : bdev_io->num_retries++;
1443 10 : bdev_submit_request(bdev, spdk_bdev_io_get_io_channel(bdev_io), bdev_io);
1444 10 : }
1445 :
1446 : static void
1447 63 : bdev_shared_ch_retry_io(struct spdk_bdev_shared_resource *shared_resource)
1448 : {
1449 : struct spdk_bdev_io *bdev_io;
1450 :
1451 63 : if (shared_resource->io_outstanding > shared_resource->nomem_threshold) {
1452 : /*
1453 : * Allow some more I/O to complete before retrying the nomem_io queue.
1454 : * Some drivers (such as nvme) cannot immediately take a new I/O in
1455 : * the context of a completion, because the resources for the I/O are
1456 : * not released until control returns to the bdev poller. Also, we
1457 : * may require several small I/O to complete before a larger I/O
1458 : * (that requires splitting) can be submitted.
1459 : */
1460 58 : return;
1461 : }
1462 :
1463 16 : while (!TAILQ_EMPTY(&shared_resource->nomem_io)) {
1464 12 : bdev_io = TAILQ_FIRST(&shared_resource->nomem_io);
1465 12 : TAILQ_REMOVE(&shared_resource->nomem_io, bdev_io, internal.link);
1466 :
1467 12 : switch (bdev_io->internal.retry_state) {
1468 10 : case BDEV_IO_RETRY_STATE_SUBMIT:
1469 10 : bdev_ch_resubmit_io(shared_resource, bdev_io);
1470 10 : break;
1471 1 : case BDEV_IO_RETRY_STATE_PULL:
1472 1 : bdev_io_pull_data(bdev_io);
1473 1 : break;
1474 0 : case BDEV_IO_RETRY_STATE_PULL_MD:
1475 0 : bdev_io_pull_md_buf(bdev_io);
1476 0 : break;
1477 1 : case BDEV_IO_RETRY_STATE_PUSH:
1478 1 : bdev_io_push_bounce_data(bdev_io);
1479 1 : break;
1480 0 : case BDEV_IO_RETRY_STATE_PUSH_MD:
1481 0 : bdev_io_push_bounce_md_buf(bdev_io);
1482 0 : break;
1483 0 : default:
1484 0 : assert(0 && "invalid retry state");
1485 : break;
1486 : }
1487 :
1488 12 : if (bdev_io == TAILQ_FIRST(&shared_resource->nomem_io)) {
1489 : /* This IO completed again with NOMEM status, so break the loop and
1490 : * don't try anymore. Note that a bdev_io that fails with NOMEM
1491 : * always gets requeued at the front of the list, to maintain
1492 : * ordering.
1493 : */
1494 1 : break;
1495 : }
1496 : }
1497 : }
1498 :
1499 : static void
1500 63 : bdev_ch_retry_io(struct spdk_bdev_channel *bdev_ch)
1501 : {
1502 63 : bdev_shared_ch_retry_io(bdev_ch->shared_resource);
1503 63 : }
1504 :
1505 : static int
1506 0 : bdev_no_mem_poller(void *ctx)
1507 : {
1508 0 : struct spdk_bdev_shared_resource *shared_resource = ctx;
1509 :
1510 0 : spdk_poller_unregister(&shared_resource->nomem_poller);
1511 :
1512 0 : if (!TAILQ_EMPTY(&shared_resource->nomem_io)) {
1513 0 : bdev_shared_ch_retry_io(shared_resource);
1514 : }
1515 : /* the retry cb may re-register the poller so double check */
1516 0 : if (!TAILQ_EMPTY(&shared_resource->nomem_io) &&
1517 0 : shared_resource->io_outstanding == 0 && shared_resource->nomem_poller == NULL) {
1518 : /* No IOs were submitted, try again */
1519 0 : shared_resource->nomem_poller = SPDK_POLLER_REGISTER(bdev_no_mem_poller, shared_resource,
1520 : SPDK_BDEV_IO_POLL_INTERVAL_IN_MSEC * 10);
1521 : }
1522 :
1523 0 : return SPDK_POLLER_BUSY;
1524 : }
1525 :
1526 : static inline bool
1527 547 : _bdev_io_handle_no_mem(struct spdk_bdev_io *bdev_io, enum bdev_io_retry_state state)
1528 : {
1529 547 : struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
1530 547 : struct spdk_bdev_shared_resource *shared_resource = bdev_ch->shared_resource;
1531 :
1532 547 : if (spdk_unlikely(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_NOMEM)) {
1533 5 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_PENDING;
1534 5 : bdev_queue_nomem_io_head(shared_resource, bdev_io, state);
1535 :
1536 5 : if (shared_resource->io_outstanding == 0 && !shared_resource->nomem_poller) {
1537 : /* Special case when we have nomem IOs and no outstanding IOs which completions
1538 : * could trigger retry of queued IOs
1539 : * Any IOs submitted may trigger retry of queued IOs. This poller handles a case when no
1540 : * new IOs submitted, e.g. qd==1 */
1541 0 : shared_resource->nomem_poller = SPDK_POLLER_REGISTER(bdev_no_mem_poller, shared_resource,
1542 : SPDK_BDEV_IO_POLL_INTERVAL_IN_MSEC * 10);
1543 : }
1544 : /* If bdev module completed an I/O that has an accel sequence with NOMEM status, the
1545 : * ownership of that sequence is transferred back to the bdev layer, so we need to
1546 : * restore internal.accel_sequence to make sure that the sequence is handled
1547 : * correctly in case the I/O is later aborted. */
1548 5 : if ((bdev_io->type == SPDK_BDEV_IO_TYPE_READ ||
1549 5 : bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) && bdev_io->u.bdev.accel_sequence) {
1550 0 : assert(bdev_io->internal.accel_sequence == NULL);
1551 0 : bdev_io->internal.accel_sequence = bdev_io->u.bdev.accel_sequence;
1552 : }
1553 :
1554 5 : return true;
1555 : }
1556 :
1557 542 : if (spdk_unlikely(!TAILQ_EMPTY(&shared_resource->nomem_io))) {
1558 63 : bdev_ch_retry_io(bdev_ch);
1559 : }
1560 :
1561 542 : return false;
1562 : }
1563 :
1564 : static void
1565 28 : _bdev_io_complete_push_bounce_done(void *ctx, int rc)
1566 : {
1567 28 : struct spdk_bdev_io *bdev_io = ctx;
1568 28 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1569 :
1570 28 : if (rc) {
1571 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
1572 : }
1573 : /* We want to free the bounce buffer here since we know we're done with it (as opposed
1574 : * to waiting for the conditional free of internal.buf in spdk_bdev_free_io()).
1575 : */
1576 28 : bdev_io_put_buf(bdev_io);
1577 :
1578 28 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1579 0 : bdev_ch_retry_io(ch);
1580 : }
1581 :
1582 : /* Continue with IO completion flow */
1583 28 : bdev_io_complete(bdev_io);
1584 28 : }
1585 :
1586 : static void
1587 2 : bdev_io_push_bounce_md_buf_done(void *ctx, int rc)
1588 : {
1589 2 : struct spdk_bdev_io *bdev_io = ctx;
1590 2 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1591 :
1592 2 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1593 2 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1594 :
1595 2 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1596 0 : bdev_ch_retry_io(ch);
1597 : }
1598 :
1599 2 : bdev_io->internal.data_transfer_cpl(bdev_io, rc);
1600 2 : }
1601 :
1602 : static inline void
1603 28 : bdev_io_push_bounce_md_buf(struct spdk_bdev_io *bdev_io)
1604 : {
1605 28 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1606 28 : int rc = 0;
1607 :
1608 28 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS);
1609 : /* do the same for metadata buffer */
1610 28 : if (spdk_unlikely(bdev_io->internal.orig_md_iov.iov_base != NULL)) {
1611 4 : assert(spdk_bdev_is_md_separate(bdev_io->bdev));
1612 :
1613 4 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_READ) {
1614 2 : if (bdev_io_use_memory_domain(bdev_io)) {
1615 2 : TAILQ_INSERT_TAIL(&ch->io_memory_domain, bdev_io, internal.link);
1616 2 : bdev_io_increment_outstanding(ch, ch->shared_resource);
1617 : /* If memory domain is used then we need to call async push function */
1618 2 : rc = spdk_memory_domain_push_data(bdev_io->internal.memory_domain,
1619 : bdev_io->internal.memory_domain_ctx,
1620 : &bdev_io->internal.orig_md_iov,
1621 2 : (uint32_t)bdev_io->internal.orig_iovcnt,
1622 : &bdev_io->internal.bounce_md_iov, 1,
1623 : bdev_io_push_bounce_md_buf_done,
1624 : bdev_io);
1625 2 : if (rc == 0) {
1626 : /* Continue IO completion in async callback */
1627 2 : return;
1628 : }
1629 0 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1630 0 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1631 0 : if (rc != -ENOMEM) {
1632 0 : SPDK_ERRLOG("Failed to push md to memory domain %s\n",
1633 : spdk_memory_domain_get_dma_device_id(
1634 : bdev_io->internal.memory_domain));
1635 : }
1636 : } else {
1637 0 : memcpy(bdev_io->internal.orig_md_iov.iov_base, bdev_io->u.bdev.md_buf,
1638 : bdev_io->internal.orig_md_iov.iov_len);
1639 : }
1640 : }
1641 : }
1642 :
1643 26 : if (spdk_unlikely(rc == -ENOMEM)) {
1644 0 : bdev_queue_nomem_io_head(ch->shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PUSH_MD);
1645 : } else {
1646 26 : assert(bdev_io->internal.data_transfer_cpl);
1647 26 : bdev_io->internal.data_transfer_cpl(bdev_io, rc);
1648 : }
1649 : }
1650 :
1651 : static inline void
1652 28 : bdev_io_push_bounce_data_done(struct spdk_bdev_io *bdev_io, int rc)
1653 : {
1654 28 : assert(bdev_io->internal.data_transfer_cpl);
1655 28 : if (rc) {
1656 0 : bdev_io->internal.data_transfer_cpl(bdev_io, rc);
1657 0 : return;
1658 : }
1659 :
1660 : /* set original buffer for this io */
1661 28 : bdev_io->u.bdev.iovcnt = bdev_io->internal.orig_iovcnt;
1662 28 : bdev_io->u.bdev.iovs = bdev_io->internal.orig_iovs;
1663 : /* disable bouncing buffer for this io */
1664 28 : bdev_io->internal.orig_iovcnt = 0;
1665 28 : bdev_io->internal.orig_iovs = NULL;
1666 :
1667 28 : bdev_io_push_bounce_md_buf(bdev_io);
1668 : }
1669 :
1670 : static void
1671 2 : bdev_io_push_bounce_data_done_and_track(void *ctx, int status)
1672 : {
1673 2 : struct spdk_bdev_io *bdev_io = ctx;
1674 2 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1675 :
1676 2 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1677 2 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1678 :
1679 2 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1680 0 : bdev_ch_retry_io(ch);
1681 : }
1682 :
1683 2 : bdev_io_push_bounce_data_done(bdev_io, status);
1684 2 : }
1685 :
1686 : static inline void
1687 29 : bdev_io_push_bounce_data(struct spdk_bdev_io *bdev_io)
1688 : {
1689 29 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1690 29 : int rc = 0;
1691 :
1692 29 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS);
1693 29 : assert(!bdev_io_use_accel_sequence(bdev_io));
1694 :
1695 : /* if this is read path, copy data from bounce buffer to original buffer */
1696 29 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_READ) {
1697 13 : if (bdev_io_use_memory_domain(bdev_io)) {
1698 3 : TAILQ_INSERT_TAIL(&ch->io_memory_domain, bdev_io, internal.link);
1699 3 : bdev_io_increment_outstanding(ch, ch->shared_resource);
1700 : /* If memory domain is used then we need to call async push function */
1701 3 : rc = spdk_memory_domain_push_data(bdev_io->internal.memory_domain,
1702 : bdev_io->internal.memory_domain_ctx,
1703 : bdev_io->internal.orig_iovs,
1704 3 : (uint32_t)bdev_io->internal.orig_iovcnt,
1705 : &bdev_io->internal.bounce_iov, 1,
1706 : bdev_io_push_bounce_data_done_and_track,
1707 : bdev_io);
1708 3 : if (rc == 0) {
1709 : /* Continue IO completion in async callback */
1710 2 : return;
1711 : }
1712 :
1713 1 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1714 1 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1715 1 : if (rc != -ENOMEM) {
1716 0 : SPDK_ERRLOG("Failed to push data to memory domain %s\n",
1717 : spdk_memory_domain_get_dma_device_id(
1718 : bdev_io->internal.memory_domain));
1719 : }
1720 : } else {
1721 10 : spdk_copy_buf_to_iovs(bdev_io->internal.orig_iovs,
1722 : bdev_io->internal.orig_iovcnt,
1723 : bdev_io->internal.bounce_iov.iov_base,
1724 : bdev_io->internal.bounce_iov.iov_len);
1725 : }
1726 : }
1727 :
1728 27 : if (spdk_unlikely(rc == -ENOMEM)) {
1729 1 : bdev_queue_nomem_io_head(ch->shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PUSH);
1730 : } else {
1731 26 : bdev_io_push_bounce_data_done(bdev_io, rc);
1732 : }
1733 : }
1734 :
1735 : static inline void
1736 28 : _bdev_io_push_bounce_data_buffer(struct spdk_bdev_io *bdev_io, bdev_copy_bounce_buffer_cpl cpl_cb)
1737 : {
1738 28 : bdev_io->internal.data_transfer_cpl = cpl_cb;
1739 28 : bdev_io_push_bounce_data(bdev_io);
1740 28 : }
1741 :
1742 : static void
1743 0 : bdev_io_get_iobuf_cb(struct spdk_iobuf_entry *iobuf, void *buf)
1744 : {
1745 : struct spdk_bdev_io *bdev_io;
1746 :
1747 0 : bdev_io = SPDK_CONTAINEROF(iobuf, struct spdk_bdev_io, internal.iobuf);
1748 0 : _bdev_io_set_buf(bdev_io, buf, bdev_io->internal.buf_len);
1749 0 : }
1750 :
1751 : static void
1752 46 : bdev_io_get_buf(struct spdk_bdev_io *bdev_io, uint64_t len)
1753 : {
1754 : struct spdk_bdev_mgmt_channel *mgmt_ch;
1755 : uint64_t max_len;
1756 : void *buf;
1757 :
1758 46 : assert(spdk_bdev_io_get_thread(bdev_io) == spdk_get_thread());
1759 46 : mgmt_ch = bdev_io->internal.ch->shared_resource->mgmt_ch;
1760 46 : max_len = bdev_io_get_max_buf_len(bdev_io, len);
1761 :
1762 46 : if (spdk_unlikely(max_len > mgmt_ch->iobuf.large.bufsize)) {
1763 0 : SPDK_ERRLOG("Length %" PRIu64 " is larger than allowed\n", max_len);
1764 0 : bdev_io_get_buf_complete(bdev_io, false);
1765 0 : return;
1766 : }
1767 :
1768 46 : bdev_io->internal.buf_len = len;
1769 46 : buf = spdk_iobuf_get(&mgmt_ch->iobuf, max_len, &bdev_io->internal.iobuf,
1770 : bdev_io_get_iobuf_cb);
1771 46 : if (buf != NULL) {
1772 46 : _bdev_io_set_buf(bdev_io, buf, len);
1773 : }
1774 : }
1775 :
1776 : void
1777 64 : spdk_bdev_io_get_buf(struct spdk_bdev_io *bdev_io, spdk_bdev_io_get_buf_cb cb, uint64_t len)
1778 : {
1779 64 : struct spdk_bdev *bdev = bdev_io->bdev;
1780 : uint64_t alignment;
1781 :
1782 64 : assert(cb != NULL);
1783 64 : bdev_io->internal.get_buf_cb = cb;
1784 :
1785 64 : alignment = spdk_bdev_get_buf_align(bdev);
1786 :
1787 110 : if (_is_buf_allocated(bdev_io->u.bdev.iovs) &&
1788 46 : _are_iovs_aligned(bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt, alignment)) {
1789 : /* Buffer already present and aligned */
1790 22 : cb(spdk_bdev_io_get_io_channel(bdev_io), bdev_io, true);
1791 22 : return;
1792 : }
1793 :
1794 42 : bdev_io_get_buf(bdev_io, len);
1795 : }
1796 :
1797 : static void
1798 4 : _bdev_memory_domain_get_io_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io,
1799 : bool success)
1800 : {
1801 4 : if (!success) {
1802 0 : SPDK_ERRLOG("Failed to get data buffer, completing IO\n");
1803 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
1804 0 : bdev_io_complete_unsubmitted(bdev_io);
1805 0 : return;
1806 : }
1807 :
1808 4 : if (bdev_io_needs_sequence_exec(bdev_io->internal.desc, bdev_io)) {
1809 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
1810 0 : bdev_io_exec_sequence(bdev_io, bdev_io_submit_sequence_cb);
1811 0 : return;
1812 : }
1813 : /* For reads we'll execute the sequence after the data is read, so, for now, only
1814 : * clear out accel_sequence pointer and submit the IO */
1815 0 : assert(bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
1816 0 : bdev_io->u.bdev.accel_sequence = NULL;
1817 : }
1818 :
1819 4 : bdev_io_submit(bdev_io);
1820 : }
1821 :
1822 : static void
1823 4 : _bdev_memory_domain_io_get_buf(struct spdk_bdev_io *bdev_io, spdk_bdev_io_get_buf_cb cb,
1824 : uint64_t len)
1825 : {
1826 4 : assert(cb != NULL);
1827 4 : bdev_io->internal.get_buf_cb = cb;
1828 :
1829 4 : bdev_io_get_buf(bdev_io, len);
1830 4 : }
1831 :
1832 : void
1833 0 : spdk_bdev_io_get_aux_buf(struct spdk_bdev_io *bdev_io, spdk_bdev_io_get_aux_buf_cb cb)
1834 : {
1835 0 : uint64_t len = bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen;
1836 :
1837 0 : assert(cb != NULL);
1838 0 : assert(bdev_io->internal.get_aux_buf_cb == NULL);
1839 0 : bdev_io->internal.get_aux_buf_cb = cb;
1840 0 : bdev_io_get_buf(bdev_io, len);
1841 0 : }
1842 :
1843 : static int
1844 64 : bdev_module_get_max_ctx_size(void)
1845 : {
1846 : struct spdk_bdev_module *bdev_module;
1847 64 : int max_bdev_module_size = 0;
1848 :
1849 249 : TAILQ_FOREACH(bdev_module, &g_bdev_mgr.bdev_modules, internal.tailq) {
1850 185 : if (bdev_module->get_ctx_size && bdev_module->get_ctx_size() > max_bdev_module_size) {
1851 0 : max_bdev_module_size = bdev_module->get_ctx_size();
1852 : }
1853 : }
1854 :
1855 64 : return max_bdev_module_size;
1856 : }
1857 :
1858 : static void
1859 0 : bdev_enable_histogram_config_json(struct spdk_bdev *bdev, struct spdk_json_write_ctx *w)
1860 : {
1861 0 : if (!bdev->internal.histogram_enabled) {
1862 0 : return;
1863 : }
1864 :
1865 0 : spdk_json_write_object_begin(w);
1866 0 : spdk_json_write_named_string(w, "method", "bdev_enable_histogram");
1867 :
1868 0 : spdk_json_write_named_object_begin(w, "params");
1869 0 : spdk_json_write_named_string(w, "name", bdev->name);
1870 :
1871 0 : spdk_json_write_named_bool(w, "enable", bdev->internal.histogram_enabled);
1872 0 : spdk_json_write_object_end(w);
1873 :
1874 0 : spdk_json_write_object_end(w);
1875 : }
1876 :
1877 : static void
1878 0 : bdev_qos_config_json(struct spdk_bdev *bdev, struct spdk_json_write_ctx *w)
1879 : {
1880 : int i;
1881 0 : struct spdk_bdev_qos *qos = bdev->internal.qos;
1882 0 : uint64_t limits[SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES];
1883 :
1884 0 : if (!qos) {
1885 0 : return;
1886 : }
1887 :
1888 0 : spdk_bdev_get_qos_rate_limits(bdev, limits);
1889 :
1890 0 : spdk_json_write_object_begin(w);
1891 0 : spdk_json_write_named_string(w, "method", "bdev_set_qos_limit");
1892 :
1893 0 : spdk_json_write_named_object_begin(w, "params");
1894 0 : spdk_json_write_named_string(w, "name", bdev->name);
1895 0 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
1896 0 : if (limits[i] > 0) {
1897 0 : spdk_json_write_named_uint64(w, qos_rpc_type[i], limits[i]);
1898 : }
1899 : }
1900 0 : spdk_json_write_object_end(w);
1901 :
1902 0 : spdk_json_write_object_end(w);
1903 : }
1904 :
1905 : void
1906 0 : spdk_bdev_subsystem_config_json(struct spdk_json_write_ctx *w)
1907 : {
1908 : struct spdk_bdev_module *bdev_module;
1909 : struct spdk_bdev *bdev;
1910 :
1911 0 : assert(w != NULL);
1912 :
1913 0 : spdk_json_write_array_begin(w);
1914 :
1915 0 : spdk_json_write_object_begin(w);
1916 0 : spdk_json_write_named_string(w, "method", "bdev_set_options");
1917 0 : spdk_json_write_named_object_begin(w, "params");
1918 0 : spdk_json_write_named_uint32(w, "bdev_io_pool_size", g_bdev_opts.bdev_io_pool_size);
1919 0 : spdk_json_write_named_uint32(w, "bdev_io_cache_size", g_bdev_opts.bdev_io_cache_size);
1920 0 : spdk_json_write_named_bool(w, "bdev_auto_examine", g_bdev_opts.bdev_auto_examine);
1921 0 : spdk_json_write_named_uint32(w, "iobuf_small_cache_size", g_bdev_opts.iobuf_small_cache_size);
1922 0 : spdk_json_write_named_uint32(w, "iobuf_large_cache_size", g_bdev_opts.iobuf_large_cache_size);
1923 0 : spdk_json_write_object_end(w);
1924 0 : spdk_json_write_object_end(w);
1925 :
1926 0 : bdev_examine_allowlist_config_json(w);
1927 :
1928 0 : TAILQ_FOREACH(bdev_module, &g_bdev_mgr.bdev_modules, internal.tailq) {
1929 0 : if (bdev_module->config_json) {
1930 0 : bdev_module->config_json(w);
1931 : }
1932 : }
1933 :
1934 0 : spdk_spin_lock(&g_bdev_mgr.spinlock);
1935 :
1936 0 : TAILQ_FOREACH(bdev, &g_bdev_mgr.bdevs, internal.link) {
1937 0 : if (bdev->fn_table->write_config_json) {
1938 0 : bdev->fn_table->write_config_json(bdev, w);
1939 : }
1940 :
1941 0 : bdev_qos_config_json(bdev, w);
1942 0 : bdev_enable_histogram_config_json(bdev, w);
1943 : }
1944 :
1945 0 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
1946 :
1947 : /* This has to be last RPC in array to make sure all bdevs finished examine */
1948 0 : spdk_json_write_object_begin(w);
1949 0 : spdk_json_write_named_string(w, "method", "bdev_wait_for_examine");
1950 0 : spdk_json_write_object_end(w);
1951 :
1952 0 : spdk_json_write_array_end(w);
1953 0 : }
1954 :
1955 : static void
1956 70 : bdev_mgmt_channel_destroy(void *io_device, void *ctx_buf)
1957 : {
1958 70 : struct spdk_bdev_mgmt_channel *ch = ctx_buf;
1959 : struct spdk_bdev_io *bdev_io;
1960 :
1961 70 : spdk_iobuf_channel_fini(&ch->iobuf);
1962 :
1963 9967 : while (!STAILQ_EMPTY(&ch->per_thread_cache)) {
1964 9897 : bdev_io = STAILQ_FIRST(&ch->per_thread_cache);
1965 9897 : STAILQ_REMOVE_HEAD(&ch->per_thread_cache, internal.buf_link);
1966 9897 : ch->per_thread_cache_count--;
1967 9897 : spdk_mempool_put(g_bdev_mgr.bdev_io_pool, (void *)bdev_io);
1968 : }
1969 :
1970 70 : assert(ch->per_thread_cache_count == 0);
1971 70 : }
1972 :
1973 : static int
1974 70 : bdev_mgmt_channel_create(void *io_device, void *ctx_buf)
1975 : {
1976 70 : struct spdk_bdev_mgmt_channel *ch = ctx_buf;
1977 : struct spdk_bdev_io *bdev_io;
1978 : uint32_t i;
1979 : int rc;
1980 :
1981 70 : rc = spdk_iobuf_channel_init(&ch->iobuf, "bdev",
1982 : g_bdev_opts.iobuf_small_cache_size,
1983 : g_bdev_opts.iobuf_large_cache_size);
1984 70 : if (rc != 0) {
1985 0 : SPDK_ERRLOG("Failed to create iobuf channel: %s\n", spdk_strerror(-rc));
1986 0 : return -1;
1987 : }
1988 :
1989 70 : STAILQ_INIT(&ch->per_thread_cache);
1990 70 : ch->bdev_io_cache_size = g_bdev_opts.bdev_io_cache_size;
1991 :
1992 : /* Pre-populate bdev_io cache to ensure this thread cannot be starved. */
1993 70 : ch->per_thread_cache_count = 0;
1994 9967 : for (i = 0; i < ch->bdev_io_cache_size; i++) {
1995 9897 : bdev_io = spdk_mempool_get(g_bdev_mgr.bdev_io_pool);
1996 9897 : if (bdev_io == NULL) {
1997 0 : SPDK_ERRLOG("You need to increase bdev_io_pool_size using bdev_set_options RPC.\n");
1998 0 : assert(false);
1999 : bdev_mgmt_channel_destroy(io_device, ctx_buf);
2000 : return -1;
2001 : }
2002 9897 : ch->per_thread_cache_count++;
2003 9897 : STAILQ_INSERT_HEAD(&ch->per_thread_cache, bdev_io, internal.buf_link);
2004 : }
2005 :
2006 70 : TAILQ_INIT(&ch->shared_resources);
2007 70 : TAILQ_INIT(&ch->io_wait_queue);
2008 :
2009 70 : return 0;
2010 : }
2011 :
2012 : static void
2013 64 : bdev_init_complete(int rc)
2014 : {
2015 64 : spdk_bdev_init_cb cb_fn = g_init_cb_fn;
2016 64 : void *cb_arg = g_init_cb_arg;
2017 : struct spdk_bdev_module *m;
2018 :
2019 64 : g_bdev_mgr.init_complete = true;
2020 64 : g_init_cb_fn = NULL;
2021 64 : g_init_cb_arg = NULL;
2022 :
2023 : /*
2024 : * For modules that need to know when subsystem init is complete,
2025 : * inform them now.
2026 : */
2027 64 : if (rc == 0) {
2028 249 : TAILQ_FOREACH(m, &g_bdev_mgr.bdev_modules, internal.tailq) {
2029 185 : if (m->init_complete) {
2030 23 : m->init_complete();
2031 : }
2032 : }
2033 : }
2034 :
2035 64 : cb_fn(cb_arg, rc);
2036 64 : }
2037 :
2038 : static bool
2039 253 : bdev_module_all_actions_completed(void)
2040 : {
2041 : struct spdk_bdev_module *m;
2042 :
2043 997 : TAILQ_FOREACH(m, &g_bdev_mgr.bdev_modules, internal.tailq) {
2044 744 : if (m->internal.action_in_progress > 0) {
2045 0 : return false;
2046 : }
2047 : }
2048 253 : return true;
2049 : }
2050 :
2051 : static void
2052 608 : bdev_module_action_complete(void)
2053 : {
2054 : /*
2055 : * Don't finish bdev subsystem initialization if
2056 : * module pre-initialization is still in progress, or
2057 : * the subsystem been already initialized.
2058 : */
2059 608 : if (!g_bdev_mgr.module_init_complete || g_bdev_mgr.init_complete) {
2060 544 : return;
2061 : }
2062 :
2063 : /*
2064 : * Check all bdev modules for inits/examinations in progress. If any
2065 : * exist, return immediately since we cannot finish bdev subsystem
2066 : * initialization until all are completed.
2067 : */
2068 64 : if (!bdev_module_all_actions_completed()) {
2069 0 : return;
2070 : }
2071 :
2072 : /*
2073 : * Modules already finished initialization - now that all
2074 : * the bdev modules have finished their asynchronous I/O
2075 : * processing, the entire bdev layer can be marked as complete.
2076 : */
2077 64 : bdev_init_complete(0);
2078 : }
2079 :
2080 : static void
2081 544 : bdev_module_action_done(struct spdk_bdev_module *module)
2082 : {
2083 544 : spdk_spin_lock(&module->internal.spinlock);
2084 544 : assert(module->internal.action_in_progress > 0);
2085 544 : module->internal.action_in_progress--;
2086 544 : spdk_spin_unlock(&module->internal.spinlock);
2087 544 : bdev_module_action_complete();
2088 544 : }
2089 :
2090 : void
2091 64 : spdk_bdev_module_init_done(struct spdk_bdev_module *module)
2092 : {
2093 64 : assert(module->async_init);
2094 64 : bdev_module_action_done(module);
2095 64 : }
2096 :
2097 : void
2098 480 : spdk_bdev_module_examine_done(struct spdk_bdev_module *module)
2099 : {
2100 480 : bdev_module_action_done(module);
2101 480 : }
2102 :
2103 : /** The last initialized bdev module */
2104 : static struct spdk_bdev_module *g_resume_bdev_module = NULL;
2105 :
2106 : static void
2107 0 : bdev_init_failed(void *cb_arg)
2108 : {
2109 0 : struct spdk_bdev_module *module = cb_arg;
2110 :
2111 0 : spdk_spin_lock(&module->internal.spinlock);
2112 0 : assert(module->internal.action_in_progress > 0);
2113 0 : module->internal.action_in_progress--;
2114 0 : spdk_spin_unlock(&module->internal.spinlock);
2115 0 : bdev_init_complete(-1);
2116 0 : }
2117 :
2118 : static int
2119 64 : bdev_modules_init(void)
2120 : {
2121 : struct spdk_bdev_module *module;
2122 64 : int rc = 0;
2123 :
2124 249 : TAILQ_FOREACH(module, &g_bdev_mgr.bdev_modules, internal.tailq) {
2125 185 : g_resume_bdev_module = module;
2126 185 : if (module->async_init) {
2127 64 : spdk_spin_lock(&module->internal.spinlock);
2128 64 : module->internal.action_in_progress = 1;
2129 64 : spdk_spin_unlock(&module->internal.spinlock);
2130 : }
2131 185 : rc = module->module_init();
2132 185 : if (rc != 0) {
2133 : /* Bump action_in_progress to prevent other modules from completion of modules_init
2134 : * Send message to defer application shutdown until resources are cleaned up */
2135 0 : spdk_spin_lock(&module->internal.spinlock);
2136 0 : module->internal.action_in_progress = 1;
2137 0 : spdk_spin_unlock(&module->internal.spinlock);
2138 0 : spdk_thread_send_msg(spdk_get_thread(), bdev_init_failed, module);
2139 0 : return rc;
2140 : }
2141 : }
2142 :
2143 64 : g_resume_bdev_module = NULL;
2144 64 : return 0;
2145 : }
2146 :
2147 : void
2148 64 : spdk_bdev_initialize(spdk_bdev_init_cb cb_fn, void *cb_arg)
2149 : {
2150 64 : int rc = 0;
2151 64 : char mempool_name[32];
2152 :
2153 64 : assert(cb_fn != NULL);
2154 :
2155 64 : g_init_cb_fn = cb_fn;
2156 64 : g_init_cb_arg = cb_arg;
2157 :
2158 64 : spdk_notify_type_register("bdev_register");
2159 64 : spdk_notify_type_register("bdev_unregister");
2160 :
2161 64 : snprintf(mempool_name, sizeof(mempool_name), "bdev_io_%d", getpid());
2162 :
2163 64 : rc = spdk_iobuf_register_module("bdev");
2164 64 : if (rc != 0) {
2165 0 : SPDK_ERRLOG("could not register bdev iobuf module: %s\n", spdk_strerror(-rc));
2166 0 : bdev_init_complete(-1);
2167 0 : return;
2168 : }
2169 :
2170 128 : g_bdev_mgr.bdev_io_pool = spdk_mempool_create(mempool_name,
2171 64 : g_bdev_opts.bdev_io_pool_size,
2172 : sizeof(struct spdk_bdev_io) +
2173 64 : bdev_module_get_max_ctx_size(),
2174 : 0,
2175 : SPDK_ENV_SOCKET_ID_ANY);
2176 :
2177 64 : if (g_bdev_mgr.bdev_io_pool == NULL) {
2178 0 : SPDK_ERRLOG("could not allocate spdk_bdev_io pool\n");
2179 0 : bdev_init_complete(-1);
2180 0 : return;
2181 : }
2182 :
2183 64 : g_bdev_mgr.zero_buffer = spdk_zmalloc(ZERO_BUFFER_SIZE, ZERO_BUFFER_SIZE,
2184 : NULL, SPDK_ENV_LCORE_ID_ANY, SPDK_MALLOC_DMA);
2185 64 : if (!g_bdev_mgr.zero_buffer) {
2186 0 : SPDK_ERRLOG("create bdev zero buffer failed\n");
2187 0 : bdev_init_complete(-1);
2188 0 : return;
2189 : }
2190 :
2191 : #ifdef SPDK_CONFIG_VTUNE
2192 : g_bdev_mgr.domain = __itt_domain_create("spdk_bdev");
2193 : #endif
2194 :
2195 64 : spdk_io_device_register(&g_bdev_mgr, bdev_mgmt_channel_create,
2196 : bdev_mgmt_channel_destroy,
2197 : sizeof(struct spdk_bdev_mgmt_channel),
2198 : "bdev_mgr");
2199 :
2200 64 : rc = bdev_modules_init();
2201 64 : g_bdev_mgr.module_init_complete = true;
2202 64 : if (rc != 0) {
2203 0 : SPDK_ERRLOG("bdev modules init failed\n");
2204 0 : return;
2205 : }
2206 :
2207 64 : bdev_module_action_complete();
2208 : }
2209 :
2210 : static void
2211 64 : bdev_mgr_unregister_cb(void *io_device)
2212 : {
2213 64 : spdk_bdev_fini_cb cb_fn = g_fini_cb_fn;
2214 :
2215 64 : if (g_bdev_mgr.bdev_io_pool) {
2216 64 : if (spdk_mempool_count(g_bdev_mgr.bdev_io_pool) != g_bdev_opts.bdev_io_pool_size) {
2217 0 : SPDK_ERRLOG("bdev IO pool count is %zu but should be %u\n",
2218 : spdk_mempool_count(g_bdev_mgr.bdev_io_pool),
2219 : g_bdev_opts.bdev_io_pool_size);
2220 : }
2221 :
2222 64 : spdk_mempool_free(g_bdev_mgr.bdev_io_pool);
2223 : }
2224 :
2225 64 : spdk_free(g_bdev_mgr.zero_buffer);
2226 :
2227 64 : bdev_examine_allowlist_free();
2228 :
2229 64 : cb_fn(g_fini_cb_arg);
2230 64 : g_fini_cb_fn = NULL;
2231 64 : g_fini_cb_arg = NULL;
2232 64 : g_bdev_mgr.init_complete = false;
2233 64 : g_bdev_mgr.module_init_complete = false;
2234 64 : }
2235 :
2236 : static void
2237 64 : bdev_module_fini_iter(void *arg)
2238 : {
2239 : struct spdk_bdev_module *bdev_module;
2240 :
2241 : /* FIXME: Handling initialization failures is broken now,
2242 : * so we won't even try cleaning up after successfully
2243 : * initialized modules. if module_init_complete is false,
2244 : * just call spdk_bdev_mgr_unregister_cb
2245 : */
2246 64 : if (!g_bdev_mgr.module_init_complete) {
2247 0 : bdev_mgr_unregister_cb(NULL);
2248 0 : return;
2249 : }
2250 :
2251 : /* Start iterating from the last touched module */
2252 64 : if (!g_resume_bdev_module) {
2253 64 : bdev_module = TAILQ_LAST(&g_bdev_mgr.bdev_modules, bdev_module_list);
2254 : } else {
2255 0 : bdev_module = TAILQ_PREV(g_resume_bdev_module, bdev_module_list,
2256 : internal.tailq);
2257 : }
2258 :
2259 249 : while (bdev_module) {
2260 185 : if (bdev_module->async_fini) {
2261 : /* Save our place so we can resume later. We must
2262 : * save the variable here, before calling module_fini()
2263 : * below, because in some cases the module may immediately
2264 : * call spdk_bdev_module_fini_done() and re-enter
2265 : * this function to continue iterating. */
2266 0 : g_resume_bdev_module = bdev_module;
2267 : }
2268 :
2269 185 : if (bdev_module->module_fini) {
2270 185 : bdev_module->module_fini();
2271 : }
2272 :
2273 185 : if (bdev_module->async_fini) {
2274 0 : return;
2275 : }
2276 :
2277 185 : bdev_module = TAILQ_PREV(bdev_module, bdev_module_list,
2278 : internal.tailq);
2279 : }
2280 :
2281 64 : g_resume_bdev_module = NULL;
2282 64 : spdk_io_device_unregister(&g_bdev_mgr, bdev_mgr_unregister_cb);
2283 : }
2284 :
2285 : void
2286 0 : spdk_bdev_module_fini_done(void)
2287 : {
2288 0 : if (spdk_get_thread() != g_fini_thread) {
2289 0 : spdk_thread_send_msg(g_fini_thread, bdev_module_fini_iter, NULL);
2290 : } else {
2291 0 : bdev_module_fini_iter(NULL);
2292 : }
2293 0 : }
2294 :
2295 : static void
2296 64 : bdev_finish_unregister_bdevs_iter(void *cb_arg, int bdeverrno)
2297 : {
2298 64 : struct spdk_bdev *bdev = cb_arg;
2299 :
2300 64 : if (bdeverrno && bdev) {
2301 0 : SPDK_WARNLOG("Unable to unregister bdev '%s' during spdk_bdev_finish()\n",
2302 : bdev->name);
2303 :
2304 : /*
2305 : * Since the call to spdk_bdev_unregister() failed, we have no way to free this
2306 : * bdev; try to continue by manually removing this bdev from the list and continue
2307 : * with the next bdev in the list.
2308 : */
2309 0 : TAILQ_REMOVE(&g_bdev_mgr.bdevs, bdev, internal.link);
2310 : }
2311 :
2312 64 : if (TAILQ_EMPTY(&g_bdev_mgr.bdevs)) {
2313 64 : SPDK_DEBUGLOG(bdev, "Done unregistering bdevs\n");
2314 : /*
2315 : * Bdev module finish need to be deferred as we might be in the middle of some context
2316 : * (like bdev part free) that will use this bdev (or private bdev driver ctx data)
2317 : * after returning.
2318 : */
2319 64 : spdk_thread_send_msg(spdk_get_thread(), bdev_module_fini_iter, NULL);
2320 64 : return;
2321 : }
2322 :
2323 : /*
2324 : * Unregister last unclaimed bdev in the list, to ensure that bdev subsystem
2325 : * shutdown proceeds top-down. The goal is to give virtual bdevs an opportunity
2326 : * to detect clean shutdown as opposed to run-time hot removal of the underlying
2327 : * base bdevs.
2328 : *
2329 : * Also, walk the list in the reverse order.
2330 : */
2331 0 : for (bdev = TAILQ_LAST(&g_bdev_mgr.bdevs, spdk_bdev_list);
2332 0 : bdev; bdev = TAILQ_PREV(bdev, spdk_bdev_list, internal.link)) {
2333 0 : spdk_spin_lock(&bdev->internal.spinlock);
2334 0 : if (bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE) {
2335 0 : LOG_ALREADY_CLAIMED_DEBUG("claimed, skipping", bdev);
2336 0 : spdk_spin_unlock(&bdev->internal.spinlock);
2337 0 : continue;
2338 : }
2339 0 : spdk_spin_unlock(&bdev->internal.spinlock);
2340 :
2341 0 : SPDK_DEBUGLOG(bdev, "Unregistering bdev '%s'\n", bdev->name);
2342 0 : spdk_bdev_unregister(bdev, bdev_finish_unregister_bdevs_iter, bdev);
2343 0 : return;
2344 : }
2345 :
2346 : /*
2347 : * If any bdev fails to unclaim underlying bdev properly, we may face the
2348 : * case of bdev list consisting of claimed bdevs only (if claims are managed
2349 : * correctly, this would mean there's a loop in the claims graph which is
2350 : * clearly impossible). Warn and unregister last bdev on the list then.
2351 : */
2352 0 : for (bdev = TAILQ_LAST(&g_bdev_mgr.bdevs, spdk_bdev_list);
2353 : bdev; bdev = TAILQ_PREV(bdev, spdk_bdev_list, internal.link)) {
2354 0 : SPDK_WARNLOG("Unregistering claimed bdev '%s'!\n", bdev->name);
2355 0 : spdk_bdev_unregister(bdev, bdev_finish_unregister_bdevs_iter, bdev);
2356 0 : return;
2357 : }
2358 : }
2359 :
2360 : static void
2361 64 : bdev_module_fini_start_iter(void *arg)
2362 : {
2363 : struct spdk_bdev_module *bdev_module;
2364 :
2365 64 : if (!g_resume_bdev_module) {
2366 64 : bdev_module = TAILQ_LAST(&g_bdev_mgr.bdev_modules, bdev_module_list);
2367 : } else {
2368 0 : bdev_module = TAILQ_PREV(g_resume_bdev_module, bdev_module_list, internal.tailq);
2369 : }
2370 :
2371 249 : while (bdev_module) {
2372 185 : if (bdev_module->async_fini_start) {
2373 : /* Save our place so we can resume later. We must
2374 : * save the variable here, before calling fini_start()
2375 : * below, because in some cases the module may immediately
2376 : * call spdk_bdev_module_fini_start_done() and re-enter
2377 : * this function to continue iterating. */
2378 0 : g_resume_bdev_module = bdev_module;
2379 : }
2380 :
2381 185 : if (bdev_module->fini_start) {
2382 23 : bdev_module->fini_start();
2383 : }
2384 :
2385 185 : if (bdev_module->async_fini_start) {
2386 0 : return;
2387 : }
2388 :
2389 185 : bdev_module = TAILQ_PREV(bdev_module, bdev_module_list, internal.tailq);
2390 : }
2391 :
2392 64 : g_resume_bdev_module = NULL;
2393 :
2394 64 : bdev_finish_unregister_bdevs_iter(NULL, 0);
2395 : }
2396 :
2397 : void
2398 0 : spdk_bdev_module_fini_start_done(void)
2399 : {
2400 0 : if (spdk_get_thread() != g_fini_thread) {
2401 0 : spdk_thread_send_msg(g_fini_thread, bdev_module_fini_start_iter, NULL);
2402 : } else {
2403 0 : bdev_module_fini_start_iter(NULL);
2404 : }
2405 0 : }
2406 :
2407 : static void
2408 64 : bdev_finish_wait_for_examine_done(void *cb_arg)
2409 : {
2410 64 : bdev_module_fini_start_iter(NULL);
2411 64 : }
2412 :
2413 : static void bdev_open_async_fini(void);
2414 :
2415 : void
2416 64 : spdk_bdev_finish(spdk_bdev_fini_cb cb_fn, void *cb_arg)
2417 : {
2418 : int rc;
2419 :
2420 64 : assert(cb_fn != NULL);
2421 :
2422 64 : g_fini_thread = spdk_get_thread();
2423 :
2424 64 : g_fini_cb_fn = cb_fn;
2425 64 : g_fini_cb_arg = cb_arg;
2426 :
2427 64 : bdev_open_async_fini();
2428 :
2429 64 : rc = spdk_bdev_wait_for_examine(bdev_finish_wait_for_examine_done, NULL);
2430 64 : if (rc != 0) {
2431 0 : SPDK_ERRLOG("wait_for_examine failed: %s\n", spdk_strerror(-rc));
2432 0 : bdev_finish_wait_for_examine_done(NULL);
2433 : }
2434 64 : }
2435 :
2436 : struct spdk_bdev_io *
2437 684 : bdev_channel_get_io(struct spdk_bdev_channel *channel)
2438 : {
2439 684 : struct spdk_bdev_mgmt_channel *ch = channel->shared_resource->mgmt_ch;
2440 : struct spdk_bdev_io *bdev_io;
2441 :
2442 684 : if (ch->per_thread_cache_count > 0) {
2443 633 : bdev_io = STAILQ_FIRST(&ch->per_thread_cache);
2444 633 : STAILQ_REMOVE_HEAD(&ch->per_thread_cache, internal.buf_link);
2445 633 : ch->per_thread_cache_count--;
2446 51 : } else if (spdk_unlikely(!TAILQ_EMPTY(&ch->io_wait_queue))) {
2447 : /*
2448 : * Don't try to look for bdev_ios in the global pool if there are
2449 : * waiters on bdev_ios - we don't want this caller to jump the line.
2450 : */
2451 0 : bdev_io = NULL;
2452 : } else {
2453 51 : bdev_io = spdk_mempool_get(g_bdev_mgr.bdev_io_pool);
2454 : }
2455 :
2456 684 : return bdev_io;
2457 : }
2458 :
2459 : void
2460 678 : spdk_bdev_free_io(struct spdk_bdev_io *bdev_io)
2461 : {
2462 : struct spdk_bdev_mgmt_channel *ch;
2463 :
2464 678 : assert(bdev_io != NULL);
2465 678 : assert(bdev_io->internal.status != SPDK_BDEV_IO_STATUS_PENDING);
2466 :
2467 678 : ch = bdev_io->internal.ch->shared_resource->mgmt_ch;
2468 :
2469 678 : if (bdev_io->internal.buf != NULL) {
2470 18 : bdev_io_put_buf(bdev_io);
2471 : }
2472 :
2473 678 : if (ch->per_thread_cache_count < ch->bdev_io_cache_size) {
2474 633 : ch->per_thread_cache_count++;
2475 633 : STAILQ_INSERT_HEAD(&ch->per_thread_cache, bdev_io, internal.buf_link);
2476 637 : while (ch->per_thread_cache_count > 0 && !TAILQ_EMPTY(&ch->io_wait_queue)) {
2477 : struct spdk_bdev_io_wait_entry *entry;
2478 :
2479 4 : entry = TAILQ_FIRST(&ch->io_wait_queue);
2480 4 : TAILQ_REMOVE(&ch->io_wait_queue, entry, link);
2481 4 : entry->cb_fn(entry->cb_arg);
2482 : }
2483 : } else {
2484 : /* We should never have a full cache with entries on the io wait queue. */
2485 45 : assert(TAILQ_EMPTY(&ch->io_wait_queue));
2486 45 : spdk_mempool_put(g_bdev_mgr.bdev_io_pool, (void *)bdev_io);
2487 : }
2488 678 : }
2489 :
2490 : static bool
2491 72 : bdev_qos_is_iops_rate_limit(enum spdk_bdev_qos_rate_limit_type limit)
2492 : {
2493 72 : assert(limit != SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES);
2494 :
2495 72 : switch (limit) {
2496 18 : case SPDK_BDEV_QOS_RW_IOPS_RATE_LIMIT:
2497 18 : return true;
2498 54 : case SPDK_BDEV_QOS_RW_BPS_RATE_LIMIT:
2499 : case SPDK_BDEV_QOS_R_BPS_RATE_LIMIT:
2500 : case SPDK_BDEV_QOS_W_BPS_RATE_LIMIT:
2501 54 : return false;
2502 0 : case SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES:
2503 : default:
2504 0 : return false;
2505 : }
2506 : }
2507 :
2508 : static bool
2509 25 : bdev_qos_io_to_limit(struct spdk_bdev_io *bdev_io)
2510 : {
2511 25 : switch (bdev_io->type) {
2512 23 : case SPDK_BDEV_IO_TYPE_NVME_IO:
2513 : case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
2514 : case SPDK_BDEV_IO_TYPE_READ:
2515 : case SPDK_BDEV_IO_TYPE_WRITE:
2516 23 : return true;
2517 0 : case SPDK_BDEV_IO_TYPE_ZCOPY:
2518 0 : if (bdev_io->u.bdev.zcopy.start) {
2519 0 : return true;
2520 : } else {
2521 0 : return false;
2522 : }
2523 2 : default:
2524 2 : return false;
2525 : }
2526 : }
2527 :
2528 : static bool
2529 33 : bdev_is_read_io(struct spdk_bdev_io *bdev_io)
2530 : {
2531 33 : switch (bdev_io->type) {
2532 0 : case SPDK_BDEV_IO_TYPE_NVME_IO:
2533 : case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
2534 : /* Bit 1 (0x2) set for read operation */
2535 0 : if (bdev_io->u.nvme_passthru.cmd.opc & SPDK_NVME_OPC_READ) {
2536 0 : return true;
2537 : } else {
2538 0 : return false;
2539 : }
2540 30 : case SPDK_BDEV_IO_TYPE_READ:
2541 30 : return true;
2542 0 : case SPDK_BDEV_IO_TYPE_ZCOPY:
2543 : /* Populate to read from disk */
2544 0 : if (bdev_io->u.bdev.zcopy.populate) {
2545 0 : return true;
2546 : } else {
2547 0 : return false;
2548 : }
2549 3 : default:
2550 3 : return false;
2551 : }
2552 : }
2553 :
2554 : static uint64_t
2555 43 : bdev_get_io_size_in_byte(struct spdk_bdev_io *bdev_io)
2556 : {
2557 43 : struct spdk_bdev *bdev = bdev_io->bdev;
2558 :
2559 43 : switch (bdev_io->type) {
2560 0 : case SPDK_BDEV_IO_TYPE_NVME_IO:
2561 : case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
2562 0 : return bdev_io->u.nvme_passthru.nbytes;
2563 43 : case SPDK_BDEV_IO_TYPE_READ:
2564 : case SPDK_BDEV_IO_TYPE_WRITE:
2565 43 : return bdev_io->u.bdev.num_blocks * bdev->blocklen;
2566 0 : case SPDK_BDEV_IO_TYPE_ZCOPY:
2567 : /* Track the data in the start phase only */
2568 0 : if (bdev_io->u.bdev.zcopy.start) {
2569 0 : return bdev_io->u.bdev.num_blocks * bdev->blocklen;
2570 : } else {
2571 0 : return 0;
2572 : }
2573 0 : default:
2574 0 : return 0;
2575 : }
2576 : }
2577 :
2578 : static inline bool
2579 64 : bdev_qos_rw_queue_io(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io, uint64_t delta)
2580 : {
2581 : int64_t remaining_this_timeslice;
2582 :
2583 64 : if (!limit->max_per_timeslice) {
2584 : /* The QoS is disabled */
2585 0 : return false;
2586 : }
2587 :
2588 64 : remaining_this_timeslice = __atomic_sub_fetch(&limit->remaining_this_timeslice, delta,
2589 : __ATOMIC_RELAXED);
2590 64 : if (remaining_this_timeslice + (int64_t)delta > 0) {
2591 : /* There was still a quota for this delta -> the IO shouldn't be queued
2592 : *
2593 : * We allow a slight quota overrun here so an IO bigger than the per-timeslice
2594 : * quota can be allowed once a while. Such overrun then taken into account in
2595 : * the QoS poller, where the next timeslice quota is calculated.
2596 : */
2597 59 : return false;
2598 : }
2599 :
2600 : /* There was no quota for this delta -> the IO should be queued
2601 : * The remaining_this_timeslice must be rewinded so it reflects the real
2602 : * amount of IOs or bytes allowed.
2603 : */
2604 5 : __atomic_add_fetch(
2605 5 : &limit->remaining_this_timeslice, delta, __ATOMIC_RELAXED);
2606 5 : return true;
2607 : }
2608 :
2609 : static inline void
2610 5 : bdev_qos_rw_rewind_io(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io, uint64_t delta)
2611 : {
2612 5 : __atomic_add_fetch(&limit->remaining_this_timeslice, delta, __ATOMIC_RELAXED);
2613 5 : }
2614 :
2615 : static bool
2616 23 : bdev_qos_rw_iops_queue(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2617 : {
2618 23 : return bdev_qos_rw_queue_io(limit, io, 1);
2619 : }
2620 :
2621 : static void
2622 3 : bdev_qos_rw_iops_rewind_quota(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2623 : {
2624 3 : bdev_qos_rw_rewind_io(limit, io, 1);
2625 3 : }
2626 :
2627 : static bool
2628 41 : bdev_qos_rw_bps_queue(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2629 : {
2630 41 : return bdev_qos_rw_queue_io(limit, io, bdev_get_io_size_in_byte(io));
2631 : }
2632 :
2633 : static void
2634 2 : bdev_qos_rw_bps_rewind_quota(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2635 : {
2636 2 : bdev_qos_rw_rewind_io(limit, io, bdev_get_io_size_in_byte(io));
2637 2 : }
2638 :
2639 : static bool
2640 19 : bdev_qos_r_bps_queue(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2641 : {
2642 19 : if (bdev_is_read_io(io) == false) {
2643 1 : return false;
2644 : }
2645 :
2646 18 : return bdev_qos_rw_bps_queue(limit, io);
2647 : }
2648 :
2649 : static void
2650 0 : bdev_qos_r_bps_rewind_quota(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2651 : {
2652 0 : if (bdev_is_read_io(io) != false) {
2653 0 : bdev_qos_rw_rewind_io(limit, io, bdev_get_io_size_in_byte(io));
2654 : }
2655 0 : }
2656 :
2657 : static bool
2658 14 : bdev_qos_w_bps_queue(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2659 : {
2660 14 : if (bdev_is_read_io(io) == true) {
2661 12 : return false;
2662 : }
2663 :
2664 2 : return bdev_qos_rw_bps_queue(limit, io);
2665 : }
2666 :
2667 : static void
2668 0 : bdev_qos_w_bps_rewind_quota(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2669 : {
2670 0 : if (bdev_is_read_io(io) != true) {
2671 0 : bdev_qos_rw_rewind_io(limit, io, bdev_get_io_size_in_byte(io));
2672 : }
2673 0 : }
2674 :
2675 : static void
2676 10 : bdev_qos_set_ops(struct spdk_bdev_qos *qos)
2677 : {
2678 : int i;
2679 :
2680 50 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
2681 40 : if (qos->rate_limits[i].limit == SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
2682 15 : qos->rate_limits[i].queue_io = NULL;
2683 15 : continue;
2684 : }
2685 :
2686 25 : switch (i) {
2687 9 : case SPDK_BDEV_QOS_RW_IOPS_RATE_LIMIT:
2688 9 : qos->rate_limits[i].queue_io = bdev_qos_rw_iops_queue;
2689 9 : qos->rate_limits[i].rewind_quota = bdev_qos_rw_iops_rewind_quota;
2690 9 : break;
2691 7 : case SPDK_BDEV_QOS_RW_BPS_RATE_LIMIT:
2692 7 : qos->rate_limits[i].queue_io = bdev_qos_rw_bps_queue;
2693 7 : qos->rate_limits[i].rewind_quota = bdev_qos_rw_bps_rewind_quota;
2694 7 : break;
2695 5 : case SPDK_BDEV_QOS_R_BPS_RATE_LIMIT:
2696 5 : qos->rate_limits[i].queue_io = bdev_qos_r_bps_queue;
2697 5 : qos->rate_limits[i].rewind_quota = bdev_qos_r_bps_rewind_quota;
2698 5 : break;
2699 4 : case SPDK_BDEV_QOS_W_BPS_RATE_LIMIT:
2700 4 : qos->rate_limits[i].queue_io = bdev_qos_w_bps_queue;
2701 4 : qos->rate_limits[i].rewind_quota = bdev_qos_w_bps_rewind_quota;
2702 4 : break;
2703 0 : default:
2704 0 : break;
2705 : }
2706 : }
2707 10 : }
2708 :
2709 : static void
2710 6 : _bdev_io_complete_in_submit(struct spdk_bdev_channel *bdev_ch,
2711 : struct spdk_bdev_io *bdev_io,
2712 : enum spdk_bdev_io_status status)
2713 : {
2714 6 : bdev_io->internal.in_submit_request = true;
2715 6 : bdev_io_increment_outstanding(bdev_ch, bdev_ch->shared_resource);
2716 6 : spdk_bdev_io_complete(bdev_io, status);
2717 6 : bdev_io->internal.in_submit_request = false;
2718 6 : }
2719 :
2720 : static inline void
2721 567 : bdev_io_do_submit(struct spdk_bdev_channel *bdev_ch, struct spdk_bdev_io *bdev_io)
2722 : {
2723 567 : struct spdk_bdev *bdev = bdev_io->bdev;
2724 567 : struct spdk_io_channel *ch = bdev_ch->channel;
2725 567 : struct spdk_bdev_shared_resource *shared_resource = bdev_ch->shared_resource;
2726 :
2727 567 : if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_ABORT)) {
2728 10 : struct spdk_bdev_mgmt_channel *mgmt_channel = shared_resource->mgmt_ch;
2729 10 : struct spdk_bdev_io *bio_to_abort = bdev_io->u.abort.bio_to_abort;
2730 :
2731 20 : if (bdev_abort_queued_io(&shared_resource->nomem_io, bio_to_abort) ||
2732 10 : bdev_abort_buf_io(mgmt_channel, bio_to_abort)) {
2733 0 : _bdev_io_complete_in_submit(bdev_ch, bdev_io,
2734 : SPDK_BDEV_IO_STATUS_SUCCESS);
2735 0 : return;
2736 : }
2737 : }
2738 :
2739 567 : if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE &&
2740 : bdev_io->bdev->split_on_write_unit &&
2741 : bdev_io->u.bdev.num_blocks < bdev_io->bdev->write_unit_size)) {
2742 4 : SPDK_ERRLOG("IO num_blocks %lu does not match the write_unit_size %u\n",
2743 : bdev_io->u.bdev.num_blocks, bdev_io->bdev->write_unit_size);
2744 4 : _bdev_io_complete_in_submit(bdev_ch, bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
2745 4 : return;
2746 : }
2747 :
2748 563 : if (spdk_likely(TAILQ_EMPTY(&shared_resource->nomem_io))) {
2749 520 : bdev_io_increment_outstanding(bdev_ch, shared_resource);
2750 520 : bdev_io->internal.in_submit_request = true;
2751 520 : bdev_submit_request(bdev, ch, bdev_io);
2752 520 : bdev_io->internal.in_submit_request = false;
2753 : } else {
2754 43 : bdev_queue_nomem_io_tail(shared_resource, bdev_io, BDEV_IO_RETRY_STATE_SUBMIT);
2755 43 : if (shared_resource->nomem_threshold == 0 && shared_resource->io_outstanding == 0) {
2756 : /* Special case when we have nomem IOs and no outstanding IOs which completions
2757 : * could trigger retry of queued IOs */
2758 0 : bdev_shared_ch_retry_io(shared_resource);
2759 : }
2760 : }
2761 : }
2762 :
2763 : static bool
2764 25 : bdev_qos_queue_io(struct spdk_bdev_qos *qos, struct spdk_bdev_io *bdev_io)
2765 : {
2766 : int i;
2767 :
2768 25 : if (bdev_qos_io_to_limit(bdev_io) == true) {
2769 100 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
2770 82 : if (!qos->rate_limits[i].queue_io) {
2771 5 : continue;
2772 : }
2773 :
2774 77 : if (qos->rate_limits[i].queue_io(&qos->rate_limits[i],
2775 : bdev_io) == true) {
2776 10 : for (i -= 1; i >= 0 ; i--) {
2777 5 : if (!qos->rate_limits[i].queue_io) {
2778 0 : continue;
2779 : }
2780 :
2781 5 : qos->rate_limits[i].rewind_quota(&qos->rate_limits[i], bdev_io);
2782 : }
2783 5 : return true;
2784 : }
2785 : }
2786 : }
2787 :
2788 20 : return false;
2789 : }
2790 :
2791 : static int
2792 27 : bdev_qos_io_submit(struct spdk_bdev_channel *ch, struct spdk_bdev_qos *qos)
2793 : {
2794 27 : struct spdk_bdev_io *bdev_io = NULL, *tmp = NULL;
2795 27 : int submitted_ios = 0;
2796 :
2797 52 : TAILQ_FOREACH_SAFE(bdev_io, &ch->qos_queued_io, internal.link, tmp) {
2798 25 : if (!bdev_qos_queue_io(qos, bdev_io)) {
2799 20 : TAILQ_REMOVE(&ch->qos_queued_io, bdev_io, internal.link);
2800 20 : bdev_io_do_submit(ch, bdev_io);
2801 :
2802 20 : submitted_ios++;
2803 : }
2804 : }
2805 :
2806 27 : return submitted_ios;
2807 : }
2808 :
2809 : static void
2810 2 : bdev_queue_io_wait_with_cb(struct spdk_bdev_io *bdev_io, spdk_bdev_io_wait_cb cb_fn)
2811 : {
2812 : int rc;
2813 :
2814 2 : bdev_io->internal.waitq_entry.bdev = bdev_io->bdev;
2815 2 : bdev_io->internal.waitq_entry.cb_fn = cb_fn;
2816 2 : bdev_io->internal.waitq_entry.cb_arg = bdev_io;
2817 2 : rc = spdk_bdev_queue_io_wait(bdev_io->bdev, spdk_io_channel_from_ctx(bdev_io->internal.ch),
2818 : &bdev_io->internal.waitq_entry);
2819 2 : if (rc != 0) {
2820 0 : SPDK_ERRLOG("Queue IO failed, rc=%d\n", rc);
2821 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
2822 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
2823 : }
2824 2 : }
2825 :
2826 : static bool
2827 591 : bdev_rw_should_split(struct spdk_bdev_io *bdev_io)
2828 : {
2829 : uint32_t io_boundary;
2830 591 : struct spdk_bdev *bdev = bdev_io->bdev;
2831 591 : uint32_t max_segment_size = bdev->max_segment_size;
2832 591 : uint32_t max_size = bdev->max_rw_size;
2833 591 : int max_segs = bdev->max_num_segments;
2834 :
2835 591 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE && bdev->split_on_write_unit) {
2836 19 : io_boundary = bdev->write_unit_size;
2837 572 : } else if (bdev->split_on_optimal_io_boundary) {
2838 164 : io_boundary = bdev->optimal_io_boundary;
2839 : } else {
2840 408 : io_boundary = 0;
2841 : }
2842 :
2843 591 : if (spdk_likely(!io_boundary && !max_segs && !max_segment_size && !max_size)) {
2844 232 : return false;
2845 : }
2846 :
2847 359 : if (io_boundary) {
2848 : uint64_t start_stripe, end_stripe;
2849 :
2850 183 : start_stripe = bdev_io->u.bdev.offset_blocks;
2851 183 : end_stripe = start_stripe + bdev_io->u.bdev.num_blocks - 1;
2852 : /* Avoid expensive div operations if possible. These spdk_u32 functions are very cheap. */
2853 183 : if (spdk_likely(spdk_u32_is_pow2(io_boundary))) {
2854 183 : start_stripe >>= spdk_u32log2(io_boundary);
2855 183 : end_stripe >>= spdk_u32log2(io_boundary);
2856 : } else {
2857 0 : start_stripe /= io_boundary;
2858 0 : end_stripe /= io_boundary;
2859 : }
2860 :
2861 183 : if (start_stripe != end_stripe) {
2862 78 : return true;
2863 : }
2864 : }
2865 :
2866 281 : if (max_segs) {
2867 145 : if (bdev_io->u.bdev.iovcnt > max_segs) {
2868 15 : return true;
2869 : }
2870 : }
2871 :
2872 266 : if (max_segment_size) {
2873 463 : for (int i = 0; i < bdev_io->u.bdev.iovcnt; i++) {
2874 341 : if (bdev_io->u.bdev.iovs[i].iov_len > max_segment_size) {
2875 9 : return true;
2876 : }
2877 : }
2878 : }
2879 :
2880 257 : if (max_size) {
2881 47 : if (bdev_io->u.bdev.num_blocks > max_size) {
2882 4 : return true;
2883 : }
2884 : }
2885 :
2886 253 : return false;
2887 : }
2888 :
2889 : static bool
2890 24 : bdev_unmap_should_split(struct spdk_bdev_io *bdev_io)
2891 : {
2892 : uint32_t num_unmap_segments;
2893 :
2894 24 : if (!bdev_io->bdev->max_unmap || !bdev_io->bdev->max_unmap_segments) {
2895 3 : return false;
2896 : }
2897 21 : num_unmap_segments = spdk_divide_round_up(bdev_io->u.bdev.num_blocks, bdev_io->bdev->max_unmap);
2898 21 : if (num_unmap_segments > bdev_io->bdev->max_unmap_segments) {
2899 4 : return true;
2900 : }
2901 :
2902 17 : return false;
2903 : }
2904 :
2905 : static bool
2906 37 : bdev_write_zeroes_should_split(struct spdk_bdev_io *bdev_io)
2907 : {
2908 37 : if (!bdev_io->bdev->max_write_zeroes) {
2909 4 : return false;
2910 : }
2911 :
2912 33 : if (bdev_io->u.bdev.num_blocks > bdev_io->bdev->max_write_zeroes) {
2913 10 : return true;
2914 : }
2915 :
2916 23 : return false;
2917 : }
2918 :
2919 : static bool
2920 30 : bdev_copy_should_split(struct spdk_bdev_io *bdev_io)
2921 : {
2922 30 : if (bdev_io->bdev->max_copy != 0 &&
2923 25 : bdev_io->u.bdev.num_blocks > bdev_io->bdev->max_copy) {
2924 6 : return true;
2925 : }
2926 :
2927 24 : return false;
2928 : }
2929 :
2930 : static bool
2931 750 : bdev_io_should_split(struct spdk_bdev_io *bdev_io)
2932 : {
2933 750 : switch (bdev_io->type) {
2934 591 : case SPDK_BDEV_IO_TYPE_READ:
2935 : case SPDK_BDEV_IO_TYPE_WRITE:
2936 591 : return bdev_rw_should_split(bdev_io);
2937 24 : case SPDK_BDEV_IO_TYPE_UNMAP:
2938 24 : return bdev_unmap_should_split(bdev_io);
2939 37 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
2940 37 : return bdev_write_zeroes_should_split(bdev_io);
2941 30 : case SPDK_BDEV_IO_TYPE_COPY:
2942 30 : return bdev_copy_should_split(bdev_io);
2943 68 : default:
2944 68 : return false;
2945 : }
2946 : }
2947 :
2948 : static uint32_t
2949 249 : _to_next_boundary(uint64_t offset, uint32_t boundary)
2950 : {
2951 249 : return (boundary - (offset % boundary));
2952 : }
2953 :
2954 : static void bdev_io_split_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg);
2955 :
2956 : static void _bdev_rw_split(void *_bdev_io);
2957 :
2958 : static void bdev_unmap_split(struct spdk_bdev_io *bdev_io);
2959 :
2960 : static void
2961 0 : _bdev_unmap_split(void *_bdev_io)
2962 : {
2963 0 : return bdev_unmap_split((struct spdk_bdev_io *)_bdev_io);
2964 : }
2965 :
2966 : static void bdev_write_zeroes_split(struct spdk_bdev_io *bdev_io);
2967 :
2968 : static void
2969 0 : _bdev_write_zeroes_split(void *_bdev_io)
2970 : {
2971 0 : return bdev_write_zeroes_split((struct spdk_bdev_io *)_bdev_io);
2972 : }
2973 :
2974 : static void bdev_copy_split(struct spdk_bdev_io *bdev_io);
2975 :
2976 : static void
2977 0 : _bdev_copy_split(void *_bdev_io)
2978 : {
2979 0 : return bdev_copy_split((struct spdk_bdev_io *)_bdev_io);
2980 : }
2981 :
2982 : static int
2983 305 : bdev_io_split_submit(struct spdk_bdev_io *bdev_io, struct iovec *iov, int iovcnt, void *md_buf,
2984 : uint64_t num_blocks, uint64_t *offset, uint64_t *remaining)
2985 : {
2986 : int rc;
2987 : uint64_t current_offset, current_remaining, current_src_offset;
2988 : spdk_bdev_io_wait_cb io_wait_fn;
2989 :
2990 305 : current_offset = *offset;
2991 305 : current_remaining = *remaining;
2992 :
2993 305 : bdev_io->u.bdev.split_outstanding++;
2994 :
2995 305 : io_wait_fn = _bdev_rw_split;
2996 305 : switch (bdev_io->type) {
2997 202 : case SPDK_BDEV_IO_TYPE_READ:
2998 202 : assert(bdev_io->u.bdev.accel_sequence == NULL);
2999 202 : rc = bdev_readv_blocks_with_md(bdev_io->internal.desc,
3000 202 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
3001 : iov, iovcnt, md_buf, current_offset,
3002 : num_blocks, bdev_io->internal.memory_domain,
3003 : bdev_io->internal.memory_domain_ctx, NULL,
3004 : bdev_io_split_done, bdev_io);
3005 202 : break;
3006 44 : case SPDK_BDEV_IO_TYPE_WRITE:
3007 44 : assert(bdev_io->u.bdev.accel_sequence == NULL);
3008 44 : rc = bdev_writev_blocks_with_md(bdev_io->internal.desc,
3009 44 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
3010 : iov, iovcnt, md_buf, current_offset,
3011 : num_blocks, bdev_io->internal.memory_domain,
3012 : bdev_io->internal.memory_domain_ctx, NULL,
3013 : bdev_io_split_done, bdev_io);
3014 44 : break;
3015 17 : case SPDK_BDEV_IO_TYPE_UNMAP:
3016 17 : io_wait_fn = _bdev_unmap_split;
3017 17 : rc = spdk_bdev_unmap_blocks(bdev_io->internal.desc,
3018 17 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
3019 : current_offset, num_blocks,
3020 : bdev_io_split_done, bdev_io);
3021 17 : break;
3022 23 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3023 23 : io_wait_fn = _bdev_write_zeroes_split;
3024 23 : rc = spdk_bdev_write_zeroes_blocks(bdev_io->internal.desc,
3025 23 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
3026 : current_offset, num_blocks,
3027 : bdev_io_split_done, bdev_io);
3028 23 : break;
3029 19 : case SPDK_BDEV_IO_TYPE_COPY:
3030 19 : io_wait_fn = _bdev_copy_split;
3031 19 : current_src_offset = bdev_io->u.bdev.copy.src_offset_blocks +
3032 19 : (current_offset - bdev_io->u.bdev.offset_blocks);
3033 19 : rc = spdk_bdev_copy_blocks(bdev_io->internal.desc,
3034 19 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
3035 : current_offset, current_src_offset, num_blocks,
3036 : bdev_io_split_done, bdev_io);
3037 19 : break;
3038 0 : default:
3039 0 : assert(false);
3040 : rc = -EINVAL;
3041 : break;
3042 : }
3043 :
3044 305 : if (rc == 0) {
3045 301 : current_offset += num_blocks;
3046 301 : current_remaining -= num_blocks;
3047 301 : bdev_io->u.bdev.split_current_offset_blocks = current_offset;
3048 301 : bdev_io->u.bdev.split_remaining_num_blocks = current_remaining;
3049 301 : *offset = current_offset;
3050 301 : *remaining = current_remaining;
3051 : } else {
3052 4 : bdev_io->u.bdev.split_outstanding--;
3053 4 : if (rc == -ENOMEM) {
3054 4 : if (bdev_io->u.bdev.split_outstanding == 0) {
3055 : /* No I/O is outstanding. Hence we should wait here. */
3056 1 : bdev_queue_io_wait_with_cb(bdev_io, io_wait_fn);
3057 : }
3058 : } else {
3059 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
3060 0 : if (bdev_io->u.bdev.split_outstanding == 0) {
3061 0 : spdk_trace_record(TRACE_BDEV_IO_DONE, 0, 0, (uintptr_t)bdev_io, bdev_io->internal.caller_ctx);
3062 0 : TAILQ_REMOVE(&bdev_io->internal.ch->io_submitted, bdev_io, internal.ch_link);
3063 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
3064 : }
3065 : }
3066 : }
3067 :
3068 305 : return rc;
3069 : }
3070 :
3071 : static void
3072 67 : _bdev_rw_split(void *_bdev_io)
3073 : {
3074 : struct iovec *parent_iov, *iov;
3075 67 : struct spdk_bdev_io *bdev_io = _bdev_io;
3076 67 : struct spdk_bdev *bdev = bdev_io->bdev;
3077 67 : uint64_t parent_offset, current_offset, remaining;
3078 : uint32_t parent_iov_offset, parent_iovcnt, parent_iovpos, child_iovcnt;
3079 : uint32_t to_next_boundary, to_next_boundary_bytes, to_last_block_bytes;
3080 : uint32_t iovcnt, iov_len, child_iovsize;
3081 67 : uint32_t blocklen = bdev->blocklen;
3082 : uint32_t io_boundary;
3083 67 : uint32_t max_segment_size = bdev->max_segment_size;
3084 67 : uint32_t max_child_iovcnt = bdev->max_num_segments;
3085 67 : uint32_t max_size = bdev->max_rw_size;
3086 67 : void *md_buf = NULL;
3087 : int rc;
3088 :
3089 67 : max_size = max_size ? max_size : UINT32_MAX;
3090 67 : max_segment_size = max_segment_size ? max_segment_size : UINT32_MAX;
3091 67 : max_child_iovcnt = max_child_iovcnt ? spdk_min(max_child_iovcnt, SPDK_BDEV_IO_NUM_CHILD_IOV) :
3092 : SPDK_BDEV_IO_NUM_CHILD_IOV;
3093 :
3094 67 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE && bdev->split_on_write_unit) {
3095 4 : io_boundary = bdev->write_unit_size;
3096 63 : } else if (bdev->split_on_optimal_io_boundary) {
3097 43 : io_boundary = bdev->optimal_io_boundary;
3098 : } else {
3099 20 : io_boundary = UINT32_MAX;
3100 : }
3101 :
3102 67 : remaining = bdev_io->u.bdev.split_remaining_num_blocks;
3103 67 : current_offset = bdev_io->u.bdev.split_current_offset_blocks;
3104 67 : parent_offset = bdev_io->u.bdev.offset_blocks;
3105 67 : parent_iov_offset = (current_offset - parent_offset) * blocklen;
3106 67 : parent_iovcnt = bdev_io->u.bdev.iovcnt;
3107 :
3108 420 : for (parent_iovpos = 0; parent_iovpos < parent_iovcnt; parent_iovpos++) {
3109 420 : parent_iov = &bdev_io->u.bdev.iovs[parent_iovpos];
3110 420 : if (parent_iov_offset < parent_iov->iov_len) {
3111 67 : break;
3112 : }
3113 353 : parent_iov_offset -= parent_iov->iov_len;
3114 : }
3115 :
3116 67 : child_iovcnt = 0;
3117 309 : while (remaining > 0 && parent_iovpos < parent_iovcnt &&
3118 : child_iovcnt < SPDK_BDEV_IO_NUM_CHILD_IOV) {
3119 249 : to_next_boundary = _to_next_boundary(current_offset, io_boundary);
3120 249 : to_next_boundary = spdk_min(remaining, to_next_boundary);
3121 249 : to_next_boundary = spdk_min(max_size, to_next_boundary);
3122 249 : to_next_boundary_bytes = to_next_boundary * blocklen;
3123 :
3124 249 : iov = &bdev_io->child_iov[child_iovcnt];
3125 249 : iovcnt = 0;
3126 :
3127 249 : if (bdev_io->u.bdev.md_buf) {
3128 48 : md_buf = (char *)bdev_io->u.bdev.md_buf +
3129 24 : (current_offset - parent_offset) * spdk_bdev_get_md_size(bdev);
3130 : }
3131 :
3132 249 : child_iovsize = spdk_min(SPDK_BDEV_IO_NUM_CHILD_IOV - child_iovcnt, max_child_iovcnt);
3133 974 : while (to_next_boundary_bytes > 0 && parent_iovpos < parent_iovcnt &&
3134 : iovcnt < child_iovsize) {
3135 725 : parent_iov = &bdev_io->u.bdev.iovs[parent_iovpos];
3136 725 : iov_len = parent_iov->iov_len - parent_iov_offset;
3137 :
3138 725 : iov_len = spdk_min(iov_len, max_segment_size);
3139 725 : iov_len = spdk_min(iov_len, to_next_boundary_bytes);
3140 725 : to_next_boundary_bytes -= iov_len;
3141 :
3142 725 : bdev_io->child_iov[child_iovcnt].iov_base = parent_iov->iov_base + parent_iov_offset;
3143 725 : bdev_io->child_iov[child_iovcnt].iov_len = iov_len;
3144 :
3145 725 : if (iov_len < parent_iov->iov_len - parent_iov_offset) {
3146 183 : parent_iov_offset += iov_len;
3147 : } else {
3148 542 : parent_iovpos++;
3149 542 : parent_iov_offset = 0;
3150 : }
3151 725 : child_iovcnt++;
3152 725 : iovcnt++;
3153 : }
3154 :
3155 249 : if (to_next_boundary_bytes > 0) {
3156 : /* We had to stop this child I/O early because we ran out of
3157 : * child_iov space or were limited by max_num_segments.
3158 : * Ensure the iovs to be aligned with block size and
3159 : * then adjust to_next_boundary before starting the
3160 : * child I/O.
3161 : */
3162 110 : assert(child_iovcnt == SPDK_BDEV_IO_NUM_CHILD_IOV ||
3163 : iovcnt == child_iovsize);
3164 110 : to_last_block_bytes = to_next_boundary_bytes % blocklen;
3165 110 : if (to_last_block_bytes != 0) {
3166 24 : uint32_t child_iovpos = child_iovcnt - 1;
3167 : /* don't decrease child_iovcnt when it equals to SPDK_BDEV_IO_NUM_CHILD_IOV
3168 : * so the loop will naturally end
3169 : */
3170 :
3171 24 : to_last_block_bytes = blocklen - to_last_block_bytes;
3172 24 : to_next_boundary_bytes += to_last_block_bytes;
3173 53 : while (to_last_block_bytes > 0 && iovcnt > 0) {
3174 32 : iov_len = spdk_min(to_last_block_bytes,
3175 : bdev_io->child_iov[child_iovpos].iov_len);
3176 32 : bdev_io->child_iov[child_iovpos].iov_len -= iov_len;
3177 32 : if (bdev_io->child_iov[child_iovpos].iov_len == 0) {
3178 15 : child_iovpos--;
3179 15 : if (--iovcnt == 0) {
3180 : /* If the child IO is less than a block size just return.
3181 : * If the first child IO of any split round is less than
3182 : * a block size, an error exit.
3183 : */
3184 3 : if (bdev_io->u.bdev.split_outstanding == 0) {
3185 1 : SPDK_ERRLOG("The first child io was less than a block size\n");
3186 1 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
3187 1 : spdk_trace_record(TRACE_BDEV_IO_DONE, 0, 0, (uintptr_t)bdev_io, bdev_io->internal.caller_ctx);
3188 1 : TAILQ_REMOVE(&bdev_io->internal.ch->io_submitted, bdev_io, internal.ch_link);
3189 1 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
3190 : }
3191 :
3192 3 : return;
3193 : }
3194 : }
3195 :
3196 29 : to_last_block_bytes -= iov_len;
3197 :
3198 29 : if (parent_iov_offset == 0) {
3199 14 : parent_iovpos--;
3200 14 : parent_iov_offset = bdev_io->u.bdev.iovs[parent_iovpos].iov_len;
3201 : }
3202 29 : parent_iov_offset -= iov_len;
3203 : }
3204 :
3205 21 : assert(to_last_block_bytes == 0);
3206 : }
3207 107 : to_next_boundary -= to_next_boundary_bytes / blocklen;
3208 : }
3209 :
3210 246 : rc = bdev_io_split_submit(bdev_io, iov, iovcnt, md_buf, to_next_boundary,
3211 : ¤t_offset, &remaining);
3212 246 : if (spdk_unlikely(rc)) {
3213 4 : return;
3214 : }
3215 : }
3216 : }
3217 :
3218 : static void
3219 3 : bdev_unmap_split(struct spdk_bdev_io *bdev_io)
3220 : {
3221 3 : uint64_t offset, unmap_blocks, remaining, max_unmap_blocks;
3222 3 : uint32_t num_children_reqs = 0;
3223 : int rc;
3224 :
3225 3 : offset = bdev_io->u.bdev.split_current_offset_blocks;
3226 3 : remaining = bdev_io->u.bdev.split_remaining_num_blocks;
3227 3 : max_unmap_blocks = bdev_io->bdev->max_unmap * bdev_io->bdev->max_unmap_segments;
3228 :
3229 20 : while (remaining && (num_children_reqs < SPDK_BDEV_MAX_CHILDREN_UNMAP_WRITE_ZEROES_REQS)) {
3230 17 : unmap_blocks = spdk_min(remaining, max_unmap_blocks);
3231 :
3232 17 : rc = bdev_io_split_submit(bdev_io, NULL, 0, NULL, unmap_blocks,
3233 : &offset, &remaining);
3234 17 : if (spdk_likely(rc == 0)) {
3235 17 : num_children_reqs++;
3236 : } else {
3237 0 : return;
3238 : }
3239 : }
3240 : }
3241 :
3242 : static void
3243 6 : bdev_write_zeroes_split(struct spdk_bdev_io *bdev_io)
3244 : {
3245 6 : uint64_t offset, write_zeroes_blocks, remaining;
3246 6 : uint32_t num_children_reqs = 0;
3247 : int rc;
3248 :
3249 6 : offset = bdev_io->u.bdev.split_current_offset_blocks;
3250 6 : remaining = bdev_io->u.bdev.split_remaining_num_blocks;
3251 :
3252 29 : while (remaining && (num_children_reqs < SPDK_BDEV_MAX_CHILDREN_UNMAP_WRITE_ZEROES_REQS)) {
3253 23 : write_zeroes_blocks = spdk_min(remaining, bdev_io->bdev->max_write_zeroes);
3254 :
3255 23 : rc = bdev_io_split_submit(bdev_io, NULL, 0, NULL, write_zeroes_blocks,
3256 : &offset, &remaining);
3257 23 : if (spdk_likely(rc == 0)) {
3258 23 : num_children_reqs++;
3259 : } else {
3260 0 : return;
3261 : }
3262 : }
3263 : }
3264 :
3265 : static void
3266 4 : bdev_copy_split(struct spdk_bdev_io *bdev_io)
3267 : {
3268 4 : uint64_t offset, copy_blocks, remaining;
3269 4 : uint32_t num_children_reqs = 0;
3270 : int rc;
3271 :
3272 4 : offset = bdev_io->u.bdev.split_current_offset_blocks;
3273 4 : remaining = bdev_io->u.bdev.split_remaining_num_blocks;
3274 :
3275 4 : assert(bdev_io->bdev->max_copy != 0);
3276 23 : while (remaining && (num_children_reqs < SPDK_BDEV_MAX_CHILDREN_COPY_REQS)) {
3277 19 : copy_blocks = spdk_min(remaining, bdev_io->bdev->max_copy);
3278 :
3279 19 : rc = bdev_io_split_submit(bdev_io, NULL, 0, NULL, copy_blocks,
3280 : &offset, &remaining);
3281 19 : if (spdk_likely(rc == 0)) {
3282 19 : num_children_reqs++;
3283 : } else {
3284 0 : return;
3285 : }
3286 : }
3287 : }
3288 :
3289 : static void
3290 58 : parent_bdev_io_complete(void *ctx, int rc)
3291 : {
3292 58 : struct spdk_bdev_io *parent_io = ctx;
3293 :
3294 58 : if (rc) {
3295 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
3296 : }
3297 :
3298 58 : parent_io->internal.cb(parent_io, parent_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS,
3299 : parent_io->internal.caller_ctx);
3300 58 : }
3301 :
3302 : static void
3303 0 : bdev_io_complete_parent_sequence_cb(void *ctx, int status)
3304 : {
3305 0 : struct spdk_bdev_io *bdev_io = ctx;
3306 :
3307 : /* u.bdev.accel_sequence should have already been cleared at this point */
3308 0 : assert(bdev_io->u.bdev.accel_sequence == NULL);
3309 0 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS);
3310 0 : bdev_io->internal.accel_sequence = NULL;
3311 :
3312 0 : if (spdk_unlikely(status != 0)) {
3313 0 : SPDK_ERRLOG("Failed to execute accel sequence, status=%d\n", status);
3314 : }
3315 :
3316 0 : parent_bdev_io_complete(bdev_io, status);
3317 0 : }
3318 :
3319 : static void
3320 301 : bdev_io_split_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
3321 : {
3322 301 : struct spdk_bdev_io *parent_io = cb_arg;
3323 :
3324 301 : spdk_bdev_free_io(bdev_io);
3325 :
3326 301 : if (!success) {
3327 15 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
3328 : /* If any child I/O failed, stop further splitting process. */
3329 15 : parent_io->u.bdev.split_current_offset_blocks += parent_io->u.bdev.split_remaining_num_blocks;
3330 15 : parent_io->u.bdev.split_remaining_num_blocks = 0;
3331 : }
3332 301 : parent_io->u.bdev.split_outstanding--;
3333 301 : if (parent_io->u.bdev.split_outstanding != 0) {
3334 223 : return;
3335 : }
3336 :
3337 : /*
3338 : * Parent I/O finishes when all blocks are consumed.
3339 : */
3340 78 : if (parent_io->u.bdev.split_remaining_num_blocks == 0) {
3341 58 : assert(parent_io->internal.cb != bdev_io_split_done);
3342 58 : spdk_trace_record(TRACE_BDEV_IO_DONE, 0, 0, (uintptr_t)parent_io, bdev_io->internal.caller_ctx);
3343 58 : TAILQ_REMOVE(&parent_io->internal.ch->io_submitted, parent_io, internal.ch_link);
3344 :
3345 58 : if (spdk_likely(parent_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS)) {
3346 51 : if (bdev_io_needs_sequence_exec(parent_io->internal.desc, parent_io)) {
3347 0 : bdev_io_exec_sequence(parent_io, bdev_io_complete_parent_sequence_cb);
3348 0 : return;
3349 51 : } else if (parent_io->internal.orig_iovcnt != 0 &&
3350 0 : !bdev_io_use_accel_sequence(bdev_io)) {
3351 : /* bdev IO will be completed in the callback */
3352 0 : _bdev_io_push_bounce_data_buffer(parent_io, parent_bdev_io_complete);
3353 0 : return;
3354 : }
3355 : }
3356 :
3357 58 : parent_bdev_io_complete(parent_io, 0);
3358 58 : return;
3359 : }
3360 :
3361 : /*
3362 : * Continue with the splitting process. This function will complete the parent I/O if the
3363 : * splitting is done.
3364 : */
3365 20 : switch (parent_io->type) {
3366 17 : case SPDK_BDEV_IO_TYPE_READ:
3367 : case SPDK_BDEV_IO_TYPE_WRITE:
3368 17 : _bdev_rw_split(parent_io);
3369 17 : break;
3370 1 : case SPDK_BDEV_IO_TYPE_UNMAP:
3371 1 : bdev_unmap_split(parent_io);
3372 1 : break;
3373 1 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3374 1 : bdev_write_zeroes_split(parent_io);
3375 1 : break;
3376 1 : case SPDK_BDEV_IO_TYPE_COPY:
3377 1 : bdev_copy_split(parent_io);
3378 1 : break;
3379 0 : default:
3380 0 : assert(false);
3381 : break;
3382 : }
3383 : }
3384 :
3385 : static void bdev_rw_split_get_buf_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io,
3386 : bool success);
3387 :
3388 : static void
3389 59 : bdev_io_split(struct spdk_bdev_io *bdev_io)
3390 : {
3391 59 : assert(bdev_io_should_split(bdev_io));
3392 :
3393 59 : bdev_io->u.bdev.split_current_offset_blocks = bdev_io->u.bdev.offset_blocks;
3394 59 : bdev_io->u.bdev.split_remaining_num_blocks = bdev_io->u.bdev.num_blocks;
3395 59 : bdev_io->u.bdev.split_outstanding = 0;
3396 59 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
3397 :
3398 59 : switch (bdev_io->type) {
3399 49 : case SPDK_BDEV_IO_TYPE_READ:
3400 : case SPDK_BDEV_IO_TYPE_WRITE:
3401 49 : if (_is_buf_allocated(bdev_io->u.bdev.iovs)) {
3402 47 : _bdev_rw_split(bdev_io);
3403 : } else {
3404 2 : assert(bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
3405 2 : spdk_bdev_io_get_buf(bdev_io, bdev_rw_split_get_buf_cb,
3406 2 : bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen);
3407 : }
3408 49 : break;
3409 2 : case SPDK_BDEV_IO_TYPE_UNMAP:
3410 2 : bdev_unmap_split(bdev_io);
3411 2 : break;
3412 5 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3413 5 : bdev_write_zeroes_split(bdev_io);
3414 5 : break;
3415 3 : case SPDK_BDEV_IO_TYPE_COPY:
3416 3 : bdev_copy_split(bdev_io);
3417 3 : break;
3418 0 : default:
3419 0 : assert(false);
3420 : break;
3421 : }
3422 59 : }
3423 :
3424 : static void
3425 2 : bdev_rw_split_get_buf_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io, bool success)
3426 : {
3427 2 : if (!success) {
3428 0 : spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
3429 0 : return;
3430 : }
3431 :
3432 2 : _bdev_rw_split(bdev_io);
3433 : }
3434 :
3435 : /* Explicitly mark this inline, since it's used as a function pointer and otherwise won't
3436 : * be inlined, at least on some compilers.
3437 : */
3438 : static inline void
3439 572 : _bdev_io_submit(void *ctx)
3440 : {
3441 572 : struct spdk_bdev_io *bdev_io = ctx;
3442 572 : struct spdk_bdev *bdev = bdev_io->bdev;
3443 572 : struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
3444 :
3445 572 : if (spdk_likely(bdev_ch->flags == 0)) {
3446 547 : bdev_io_do_submit(bdev_ch, bdev_io);
3447 547 : return;
3448 : }
3449 :
3450 25 : if (bdev_ch->flags & BDEV_CH_RESET_IN_PROGRESS) {
3451 2 : _bdev_io_complete_in_submit(bdev_ch, bdev_io, SPDK_BDEV_IO_STATUS_ABORTED);
3452 23 : } else if (bdev_ch->flags & BDEV_CH_QOS_ENABLED) {
3453 25 : if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_ABORT) &&
3454 2 : bdev_abort_queued_io(&bdev_ch->qos_queued_io, bdev_io->u.abort.bio_to_abort)) {
3455 0 : _bdev_io_complete_in_submit(bdev_ch, bdev_io, SPDK_BDEV_IO_STATUS_SUCCESS);
3456 : } else {
3457 23 : TAILQ_INSERT_TAIL(&bdev_ch->qos_queued_io, bdev_io, internal.link);
3458 23 : bdev_qos_io_submit(bdev_ch, bdev->internal.qos);
3459 : }
3460 : } else {
3461 0 : SPDK_ERRLOG("unknown bdev_ch flag %x found\n", bdev_ch->flags);
3462 0 : _bdev_io_complete_in_submit(bdev_ch, bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
3463 : }
3464 : }
3465 :
3466 : bool bdev_lba_range_overlapped(struct lba_range *range1, struct lba_range *range2);
3467 :
3468 : bool
3469 23 : bdev_lba_range_overlapped(struct lba_range *range1, struct lba_range *range2)
3470 : {
3471 23 : if (range1->length == 0 || range2->length == 0) {
3472 1 : return false;
3473 : }
3474 :
3475 22 : if (range1->offset + range1->length <= range2->offset) {
3476 1 : return false;
3477 : }
3478 :
3479 21 : if (range2->offset + range2->length <= range1->offset) {
3480 3 : return false;
3481 : }
3482 :
3483 18 : return true;
3484 : }
3485 :
3486 : static bool
3487 11 : bdev_io_range_is_locked(struct spdk_bdev_io *bdev_io, struct lba_range *range)
3488 : {
3489 11 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
3490 11 : struct lba_range r;
3491 :
3492 11 : switch (bdev_io->type) {
3493 0 : case SPDK_BDEV_IO_TYPE_NVME_IO:
3494 : case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
3495 : /* Don't try to decode the NVMe command - just assume worst-case and that
3496 : * it overlaps a locked range.
3497 : */
3498 0 : return true;
3499 6 : case SPDK_BDEV_IO_TYPE_READ:
3500 6 : if (!range->quiesce) {
3501 4 : return false;
3502 : }
3503 : /* fallthrough */
3504 : case SPDK_BDEV_IO_TYPE_WRITE:
3505 : case SPDK_BDEV_IO_TYPE_UNMAP:
3506 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3507 : case SPDK_BDEV_IO_TYPE_ZCOPY:
3508 : case SPDK_BDEV_IO_TYPE_COPY:
3509 7 : r.offset = bdev_io->u.bdev.offset_blocks;
3510 7 : r.length = bdev_io->u.bdev.num_blocks;
3511 7 : if (!bdev_lba_range_overlapped(range, &r)) {
3512 : /* This I/O doesn't overlap the specified LBA range. */
3513 0 : return false;
3514 7 : } else if (range->owner_ch == ch && range->locked_ctx == bdev_io->internal.caller_ctx) {
3515 : /* This I/O overlaps, but the I/O is on the same channel that locked this
3516 : * range, and the caller_ctx is the same as the locked_ctx. This means
3517 : * that this I/O is associated with the lock, and is allowed to execute.
3518 : */
3519 2 : return false;
3520 : } else {
3521 5 : return true;
3522 : }
3523 0 : default:
3524 0 : return false;
3525 : }
3526 : }
3527 :
3528 : void
3529 632 : bdev_io_submit(struct spdk_bdev_io *bdev_io)
3530 : {
3531 632 : struct spdk_bdev *bdev = bdev_io->bdev;
3532 632 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
3533 :
3534 632 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_PENDING);
3535 :
3536 632 : if (!TAILQ_EMPTY(&ch->locked_ranges)) {
3537 : struct lba_range *range;
3538 :
3539 13 : TAILQ_FOREACH(range, &ch->locked_ranges, tailq) {
3540 8 : if (bdev_io_range_is_locked(bdev_io, range)) {
3541 3 : TAILQ_INSERT_TAIL(&ch->io_locked, bdev_io, internal.ch_link);
3542 3 : return;
3543 : }
3544 : }
3545 : }
3546 :
3547 629 : TAILQ_INSERT_TAIL(&ch->io_submitted, bdev_io, internal.ch_link);
3548 :
3549 629 : bdev_io->internal.submit_tsc = spdk_get_ticks();
3550 629 : spdk_trace_record_tsc(bdev_io->internal.submit_tsc, TRACE_BDEV_IO_START, 0, 0,
3551 : (uintptr_t)bdev_io, (uint64_t)bdev_io->type, bdev_io->internal.caller_ctx,
3552 : bdev_io->u.bdev.offset_blocks, bdev_io->u.bdev.num_blocks,
3553 : spdk_bdev_get_name(bdev));
3554 :
3555 629 : if (bdev_io->internal.split) {
3556 59 : bdev_io_split(bdev_io);
3557 59 : return;
3558 : }
3559 :
3560 570 : _bdev_io_submit(bdev_io);
3561 : }
3562 :
3563 : static inline void
3564 4 : _bdev_io_ext_use_bounce_buffer(struct spdk_bdev_io *bdev_io)
3565 : {
3566 : /* bdev doesn't support memory domains, thereby buffers in this IO request can't
3567 : * be accessed directly. It is needed to allocate buffers before issuing IO operation.
3568 : * For write operation we need to pull buffers from memory domain before submitting IO.
3569 : * Once read operation completes, we need to use memory_domain push functionality to
3570 : * update data in original memory domain IO buffer
3571 : * This IO request will go through a regular IO flow, so clear memory domains pointers */
3572 4 : bdev_io->u.bdev.memory_domain = NULL;
3573 4 : bdev_io->u.bdev.memory_domain_ctx = NULL;
3574 4 : _bdev_memory_domain_io_get_buf(bdev_io, _bdev_memory_domain_get_io_cb,
3575 4 : bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen);
3576 4 : }
3577 :
3578 : static inline void
3579 295 : _bdev_io_submit_ext(struct spdk_bdev_desc *desc, struct spdk_bdev_io *bdev_io)
3580 : {
3581 295 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
3582 295 : bool needs_exec = bdev_io_needs_sequence_exec(desc, bdev_io);
3583 :
3584 295 : if (spdk_unlikely(ch->flags & BDEV_CH_RESET_IN_PROGRESS)) {
3585 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_ABORTED;
3586 0 : bdev_io_complete_unsubmitted(bdev_io);
3587 0 : return;
3588 : }
3589 :
3590 : /* We need to allocate bounce buffer if bdev doesn't support memory domains, or if it does
3591 : * support them, but we need to execute an accel sequence and the data buffer is from accel
3592 : * memory domain (to avoid doing a push/pull from that domain).
3593 : */
3594 295 : if ((bdev_io->internal.memory_domain && !desc->memory_domains_supported) ||
3595 0 : (needs_exec && bdev_io->internal.memory_domain == spdk_accel_get_memory_domain())) {
3596 4 : _bdev_io_ext_use_bounce_buffer(bdev_io);
3597 4 : return;
3598 : }
3599 :
3600 291 : if (needs_exec) {
3601 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
3602 0 : bdev_io_exec_sequence(bdev_io, bdev_io_submit_sequence_cb);
3603 0 : return;
3604 : }
3605 : /* For reads we'll execute the sequence after the data is read, so, for now, only
3606 : * clear out accel_sequence pointer and submit the IO */
3607 0 : assert(bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
3608 0 : bdev_io->u.bdev.accel_sequence = NULL;
3609 : }
3610 :
3611 291 : bdev_io_submit(bdev_io);
3612 : }
3613 :
3614 : static void
3615 11 : bdev_io_submit_reset(struct spdk_bdev_io *bdev_io)
3616 : {
3617 11 : struct spdk_bdev *bdev = bdev_io->bdev;
3618 11 : struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
3619 11 : struct spdk_io_channel *ch = bdev_ch->channel;
3620 :
3621 11 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_PENDING);
3622 :
3623 11 : bdev_io->internal.in_submit_request = true;
3624 11 : bdev_submit_request(bdev, ch, bdev_io);
3625 11 : bdev_io->internal.in_submit_request = false;
3626 11 : }
3627 :
3628 : void
3629 678 : bdev_io_init(struct spdk_bdev_io *bdev_io,
3630 : struct spdk_bdev *bdev, void *cb_arg,
3631 : spdk_bdev_io_completion_cb cb)
3632 : {
3633 678 : bdev_io->bdev = bdev;
3634 678 : bdev_io->internal.caller_ctx = cb_arg;
3635 678 : bdev_io->internal.cb = cb;
3636 678 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_PENDING;
3637 678 : bdev_io->internal.in_submit_request = false;
3638 678 : bdev_io->internal.buf = NULL;
3639 678 : bdev_io->internal.orig_iovs = NULL;
3640 678 : bdev_io->internal.orig_iovcnt = 0;
3641 678 : bdev_io->internal.orig_md_iov.iov_base = NULL;
3642 678 : bdev_io->internal.error.nvme.cdw0 = 0;
3643 678 : bdev_io->num_retries = 0;
3644 678 : bdev_io->internal.get_buf_cb = NULL;
3645 678 : bdev_io->internal.get_aux_buf_cb = NULL;
3646 678 : bdev_io->internal.memory_domain = NULL;
3647 678 : bdev_io->internal.memory_domain_ctx = NULL;
3648 678 : bdev_io->internal.data_transfer_cpl = NULL;
3649 678 : bdev_io->internal.split = bdev_io_should_split(bdev_io);
3650 678 : bdev_io->internal.accel_sequence = NULL;
3651 678 : bdev_io->internal.has_accel_sequence = false;
3652 678 : }
3653 :
3654 : static bool
3655 388 : bdev_io_type_supported(struct spdk_bdev *bdev, enum spdk_bdev_io_type io_type)
3656 : {
3657 388 : return bdev->fn_table->io_type_supported(bdev->ctxt, io_type);
3658 : }
3659 :
3660 : bool
3661 42 : spdk_bdev_io_type_supported(struct spdk_bdev *bdev, enum spdk_bdev_io_type io_type)
3662 : {
3663 : bool supported;
3664 :
3665 42 : supported = bdev_io_type_supported(bdev, io_type);
3666 :
3667 42 : if (!supported) {
3668 7 : switch (io_type) {
3669 0 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3670 : /* The bdev layer will emulate write zeroes as long as write is supported. */
3671 0 : supported = bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE);
3672 0 : break;
3673 7 : default:
3674 7 : break;
3675 : }
3676 35 : }
3677 :
3678 42 : return supported;
3679 : }
3680 :
3681 : uint64_t
3682 0 : spdk_bdev_io_get_submit_tsc(struct spdk_bdev_io *bdev_io)
3683 : {
3684 0 : return bdev_io->internal.submit_tsc;
3685 : }
3686 :
3687 : int
3688 0 : spdk_bdev_dump_info_json(struct spdk_bdev *bdev, struct spdk_json_write_ctx *w)
3689 : {
3690 0 : if (bdev->fn_table->dump_info_json) {
3691 0 : return bdev->fn_table->dump_info_json(bdev->ctxt, w);
3692 : }
3693 :
3694 0 : return 0;
3695 : }
3696 :
3697 : static void
3698 10 : bdev_qos_update_max_quota_per_timeslice(struct spdk_bdev_qos *qos)
3699 : {
3700 10 : uint32_t max_per_timeslice = 0;
3701 : int i;
3702 :
3703 50 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
3704 40 : if (qos->rate_limits[i].limit == SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
3705 15 : qos->rate_limits[i].max_per_timeslice = 0;
3706 15 : continue;
3707 : }
3708 :
3709 25 : max_per_timeslice = qos->rate_limits[i].limit *
3710 25 : SPDK_BDEV_QOS_TIMESLICE_IN_USEC / SPDK_SEC_TO_USEC;
3711 :
3712 25 : qos->rate_limits[i].max_per_timeslice = spdk_max(max_per_timeslice,
3713 : qos->rate_limits[i].min_per_timeslice);
3714 :
3715 25 : __atomic_store_n(&qos->rate_limits[i].remaining_this_timeslice,
3716 25 : qos->rate_limits[i].max_per_timeslice, __ATOMIC_RELEASE);
3717 : }
3718 :
3719 10 : bdev_qos_set_ops(qos);
3720 10 : }
3721 :
3722 : static void
3723 4 : bdev_channel_submit_qos_io(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
3724 : struct spdk_io_channel *io_ch, void *ctx)
3725 : {
3726 4 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
3727 : int status;
3728 :
3729 4 : bdev_qos_io_submit(bdev_ch, bdev->internal.qos);
3730 :
3731 : /* if all IOs were sent then continue the iteration, otherwise - stop it */
3732 : /* TODO: channels round robing */
3733 4 : status = TAILQ_EMPTY(&bdev_ch->qos_queued_io) ? 0 : 1;
3734 :
3735 4 : spdk_bdev_for_each_channel_continue(i, status);
3736 4 : }
3737 :
3738 :
3739 : static void
3740 2 : bdev_channel_submit_qos_io_done(struct spdk_bdev *bdev, void *ctx, int status)
3741 : {
3742 :
3743 2 : }
3744 :
3745 : static int
3746 3 : bdev_channel_poll_qos(void *arg)
3747 : {
3748 3 : struct spdk_bdev *bdev = arg;
3749 3 : struct spdk_bdev_qos *qos = bdev->internal.qos;
3750 3 : uint64_t now = spdk_get_ticks();
3751 : int i;
3752 : int64_t remaining_last_timeslice;
3753 :
3754 3 : if (spdk_unlikely(qos->thread == NULL)) {
3755 : /* Old QoS was unbound to remove and new QoS is not enabled yet. */
3756 1 : return SPDK_POLLER_IDLE;
3757 : }
3758 :
3759 2 : if (now < (qos->last_timeslice + qos->timeslice_size)) {
3760 : /* We received our callback earlier than expected - return
3761 : * immediately and wait to do accounting until at least one
3762 : * timeslice has actually expired. This should never happen
3763 : * with a well-behaved timer implementation.
3764 : */
3765 0 : return SPDK_POLLER_IDLE;
3766 : }
3767 :
3768 : /* Reset for next round of rate limiting */
3769 10 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
3770 : /* We may have allowed the IOs or bytes to slightly overrun in the last
3771 : * timeslice. remaining_this_timeslice is signed, so if it's negative
3772 : * here, we'll account for the overrun so that the next timeslice will
3773 : * be appropriately reduced.
3774 : */
3775 8 : remaining_last_timeslice = __atomic_exchange_n(&qos->rate_limits[i].remaining_this_timeslice,
3776 : 0, __ATOMIC_RELAXED);
3777 8 : if (remaining_last_timeslice < 0) {
3778 : /* There could be a race condition here as both bdev_qos_rw_queue_io() and bdev_channel_poll_qos()
3779 : * potentially use 2 atomic ops each, so they can intertwine.
3780 : * This race can potentialy cause the limits to be a little fuzzy but won't cause any real damage.
3781 : */
3782 0 : __atomic_store_n(&qos->rate_limits[i].remaining_this_timeslice,
3783 : remaining_last_timeslice, __ATOMIC_RELAXED);
3784 : }
3785 : }
3786 :
3787 4 : while (now >= (qos->last_timeslice + qos->timeslice_size)) {
3788 2 : qos->last_timeslice += qos->timeslice_size;
3789 10 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
3790 8 : __atomic_add_fetch(&qos->rate_limits[i].remaining_this_timeslice,
3791 8 : qos->rate_limits[i].max_per_timeslice, __ATOMIC_RELAXED);
3792 : }
3793 : }
3794 :
3795 2 : spdk_bdev_for_each_channel(bdev, bdev_channel_submit_qos_io, qos,
3796 : bdev_channel_submit_qos_io_done);
3797 :
3798 2 : return SPDK_POLLER_BUSY;
3799 : }
3800 :
3801 : static void
3802 73 : bdev_channel_destroy_resource(struct spdk_bdev_channel *ch)
3803 : {
3804 : struct spdk_bdev_shared_resource *shared_resource;
3805 : struct lba_range *range;
3806 :
3807 73 : bdev_free_io_stat(ch->stat);
3808 : #ifdef SPDK_CONFIG_VTUNE
3809 : bdev_free_io_stat(ch->prev_stat);
3810 : #endif
3811 :
3812 73 : while (!TAILQ_EMPTY(&ch->locked_ranges)) {
3813 0 : range = TAILQ_FIRST(&ch->locked_ranges);
3814 0 : TAILQ_REMOVE(&ch->locked_ranges, range, tailq);
3815 0 : free(range);
3816 : }
3817 :
3818 73 : spdk_put_io_channel(ch->channel);
3819 73 : spdk_put_io_channel(ch->accel_channel);
3820 :
3821 73 : shared_resource = ch->shared_resource;
3822 :
3823 73 : assert(TAILQ_EMPTY(&ch->io_locked));
3824 73 : assert(TAILQ_EMPTY(&ch->io_submitted));
3825 73 : assert(TAILQ_EMPTY(&ch->io_accel_exec));
3826 73 : assert(TAILQ_EMPTY(&ch->io_memory_domain));
3827 73 : assert(ch->io_outstanding == 0);
3828 73 : assert(shared_resource->ref > 0);
3829 73 : shared_resource->ref--;
3830 73 : if (shared_resource->ref == 0) {
3831 72 : assert(shared_resource->io_outstanding == 0);
3832 72 : TAILQ_REMOVE(&shared_resource->mgmt_ch->shared_resources, shared_resource, link);
3833 72 : spdk_put_io_channel(spdk_io_channel_from_ctx(shared_resource->mgmt_ch));
3834 72 : spdk_poller_unregister(&shared_resource->nomem_poller);
3835 72 : free(shared_resource);
3836 : }
3837 73 : }
3838 :
3839 : static void
3840 82 : bdev_enable_qos(struct spdk_bdev *bdev, struct spdk_bdev_channel *ch)
3841 : {
3842 82 : struct spdk_bdev_qos *qos = bdev->internal.qos;
3843 : int i;
3844 :
3845 82 : assert(spdk_spin_held(&bdev->internal.spinlock));
3846 :
3847 : /* Rate limiting on this bdev enabled */
3848 82 : if (qos) {
3849 17 : if (qos->ch == NULL) {
3850 : struct spdk_io_channel *io_ch;
3851 :
3852 9 : SPDK_DEBUGLOG(bdev, "Selecting channel %p as QoS channel for bdev %s on thread %p\n", ch,
3853 : bdev->name, spdk_get_thread());
3854 :
3855 : /* No qos channel has been selected, so set one up */
3856 :
3857 : /* Take another reference to ch */
3858 9 : io_ch = spdk_get_io_channel(__bdev_to_io_dev(bdev));
3859 9 : assert(io_ch != NULL);
3860 9 : qos->ch = ch;
3861 :
3862 9 : qos->thread = spdk_io_channel_get_thread(io_ch);
3863 :
3864 45 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
3865 36 : if (bdev_qos_is_iops_rate_limit(i) == true) {
3866 9 : qos->rate_limits[i].min_per_timeslice =
3867 : SPDK_BDEV_QOS_MIN_IO_PER_TIMESLICE;
3868 : } else {
3869 27 : qos->rate_limits[i].min_per_timeslice =
3870 : SPDK_BDEV_QOS_MIN_BYTE_PER_TIMESLICE;
3871 : }
3872 :
3873 36 : if (qos->rate_limits[i].limit == 0) {
3874 2 : qos->rate_limits[i].limit = SPDK_BDEV_QOS_LIMIT_NOT_DEFINED;
3875 : }
3876 : }
3877 9 : bdev_qos_update_max_quota_per_timeslice(qos);
3878 9 : qos->timeslice_size =
3879 9 : SPDK_BDEV_QOS_TIMESLICE_IN_USEC * spdk_get_ticks_hz() / SPDK_SEC_TO_USEC;
3880 9 : qos->last_timeslice = spdk_get_ticks();
3881 9 : qos->poller = SPDK_POLLER_REGISTER(bdev_channel_poll_qos,
3882 : bdev,
3883 : SPDK_BDEV_QOS_TIMESLICE_IN_USEC);
3884 : }
3885 :
3886 17 : ch->flags |= BDEV_CH_QOS_ENABLED;
3887 : }
3888 82 : }
3889 :
3890 : struct poll_timeout_ctx {
3891 : struct spdk_bdev_desc *desc;
3892 : uint64_t timeout_in_sec;
3893 : spdk_bdev_io_timeout_cb cb_fn;
3894 : void *cb_arg;
3895 : };
3896 :
3897 : static void
3898 262 : bdev_desc_free(struct spdk_bdev_desc *desc)
3899 : {
3900 262 : spdk_spin_destroy(&desc->spinlock);
3901 262 : free(desc->media_events_buffer);
3902 262 : free(desc);
3903 262 : }
3904 :
3905 : static void
3906 8 : bdev_channel_poll_timeout_io_done(struct spdk_bdev *bdev, void *_ctx, int status)
3907 : {
3908 8 : struct poll_timeout_ctx *ctx = _ctx;
3909 8 : struct spdk_bdev_desc *desc = ctx->desc;
3910 :
3911 8 : free(ctx);
3912 :
3913 8 : spdk_spin_lock(&desc->spinlock);
3914 8 : desc->refs--;
3915 8 : if (desc->closed == true && desc->refs == 0) {
3916 1 : spdk_spin_unlock(&desc->spinlock);
3917 1 : bdev_desc_free(desc);
3918 1 : return;
3919 : }
3920 7 : spdk_spin_unlock(&desc->spinlock);
3921 : }
3922 :
3923 : static void
3924 13 : bdev_channel_poll_timeout_io(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
3925 : struct spdk_io_channel *io_ch, void *_ctx)
3926 : {
3927 13 : struct poll_timeout_ctx *ctx = _ctx;
3928 13 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
3929 13 : struct spdk_bdev_desc *desc = ctx->desc;
3930 : struct spdk_bdev_io *bdev_io;
3931 : uint64_t now;
3932 :
3933 13 : spdk_spin_lock(&desc->spinlock);
3934 13 : if (desc->closed == true) {
3935 1 : spdk_spin_unlock(&desc->spinlock);
3936 1 : spdk_bdev_for_each_channel_continue(i, -1);
3937 1 : return;
3938 : }
3939 12 : spdk_spin_unlock(&desc->spinlock);
3940 :
3941 12 : now = spdk_get_ticks();
3942 22 : TAILQ_FOREACH(bdev_io, &bdev_ch->io_submitted, internal.ch_link) {
3943 : /* Exclude any I/O that are generated via splitting. */
3944 15 : if (bdev_io->internal.cb == bdev_io_split_done) {
3945 3 : continue;
3946 : }
3947 :
3948 : /* Once we find an I/O that has not timed out, we can immediately
3949 : * exit the loop.
3950 : */
3951 24 : if (now < (bdev_io->internal.submit_tsc +
3952 12 : ctx->timeout_in_sec * spdk_get_ticks_hz())) {
3953 5 : goto end;
3954 : }
3955 :
3956 7 : if (bdev_io->internal.desc == desc) {
3957 7 : ctx->cb_fn(ctx->cb_arg, bdev_io);
3958 : }
3959 : }
3960 :
3961 7 : end:
3962 12 : spdk_bdev_for_each_channel_continue(i, 0);
3963 : }
3964 :
3965 : static int
3966 8 : bdev_poll_timeout_io(void *arg)
3967 : {
3968 8 : struct spdk_bdev_desc *desc = arg;
3969 8 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
3970 : struct poll_timeout_ctx *ctx;
3971 :
3972 8 : ctx = calloc(1, sizeof(struct poll_timeout_ctx));
3973 8 : if (!ctx) {
3974 0 : SPDK_ERRLOG("failed to allocate memory\n");
3975 0 : return SPDK_POLLER_BUSY;
3976 : }
3977 8 : ctx->desc = desc;
3978 8 : ctx->cb_arg = desc->cb_arg;
3979 8 : ctx->cb_fn = desc->cb_fn;
3980 8 : ctx->timeout_in_sec = desc->timeout_in_sec;
3981 :
3982 : /* Take a ref on the descriptor in case it gets closed while we are checking
3983 : * all of the channels.
3984 : */
3985 8 : spdk_spin_lock(&desc->spinlock);
3986 8 : desc->refs++;
3987 8 : spdk_spin_unlock(&desc->spinlock);
3988 :
3989 8 : spdk_bdev_for_each_channel(bdev, bdev_channel_poll_timeout_io, ctx,
3990 : bdev_channel_poll_timeout_io_done);
3991 :
3992 8 : return SPDK_POLLER_BUSY;
3993 : }
3994 :
3995 : int
3996 5 : spdk_bdev_set_timeout(struct spdk_bdev_desc *desc, uint64_t timeout_in_sec,
3997 : spdk_bdev_io_timeout_cb cb_fn, void *cb_arg)
3998 : {
3999 5 : assert(desc->thread == spdk_get_thread());
4000 :
4001 5 : spdk_poller_unregister(&desc->io_timeout_poller);
4002 :
4003 5 : if (timeout_in_sec) {
4004 4 : assert(cb_fn != NULL);
4005 4 : desc->io_timeout_poller = SPDK_POLLER_REGISTER(bdev_poll_timeout_io,
4006 : desc,
4007 : SPDK_BDEV_IO_POLL_INTERVAL_IN_MSEC * SPDK_SEC_TO_USEC /
4008 : 1000);
4009 4 : if (desc->io_timeout_poller == NULL) {
4010 0 : SPDK_ERRLOG("can not register the desc timeout IO poller\n");
4011 0 : return -1;
4012 : }
4013 : }
4014 :
4015 5 : desc->cb_fn = cb_fn;
4016 5 : desc->cb_arg = cb_arg;
4017 5 : desc->timeout_in_sec = timeout_in_sec;
4018 :
4019 5 : return 0;
4020 : }
4021 :
4022 : static int
4023 75 : bdev_channel_create(void *io_device, void *ctx_buf)
4024 : {
4025 75 : struct spdk_bdev *bdev = __bdev_from_io_dev(io_device);
4026 75 : struct spdk_bdev_channel *ch = ctx_buf;
4027 : struct spdk_io_channel *mgmt_io_ch;
4028 : struct spdk_bdev_mgmt_channel *mgmt_ch;
4029 : struct spdk_bdev_shared_resource *shared_resource;
4030 : struct lba_range *range;
4031 :
4032 75 : ch->bdev = bdev;
4033 75 : ch->channel = bdev->fn_table->get_io_channel(bdev->ctxt);
4034 75 : if (!ch->channel) {
4035 2 : return -1;
4036 : }
4037 :
4038 73 : ch->accel_channel = spdk_accel_get_io_channel();
4039 73 : if (!ch->accel_channel) {
4040 0 : spdk_put_io_channel(ch->channel);
4041 0 : return -1;
4042 : }
4043 :
4044 73 : spdk_trace_record(TRACE_BDEV_IOCH_CREATE, 0, 0, 0, ch->bdev->name,
4045 : spdk_thread_get_id(spdk_io_channel_get_thread(ch->channel)));
4046 :
4047 73 : assert(ch->histogram == NULL);
4048 73 : if (bdev->internal.histogram_enabled) {
4049 0 : ch->histogram = spdk_histogram_data_alloc();
4050 0 : if (ch->histogram == NULL) {
4051 0 : SPDK_ERRLOG("Could not allocate histogram\n");
4052 : }
4053 : }
4054 :
4055 73 : mgmt_io_ch = spdk_get_io_channel(&g_bdev_mgr);
4056 73 : if (!mgmt_io_ch) {
4057 0 : spdk_put_io_channel(ch->channel);
4058 0 : spdk_put_io_channel(ch->accel_channel);
4059 0 : return -1;
4060 : }
4061 :
4062 73 : mgmt_ch = __io_ch_to_bdev_mgmt_ch(mgmt_io_ch);
4063 75 : TAILQ_FOREACH(shared_resource, &mgmt_ch->shared_resources, link) {
4064 3 : if (shared_resource->shared_ch == ch->channel) {
4065 1 : spdk_put_io_channel(mgmt_io_ch);
4066 1 : shared_resource->ref++;
4067 1 : break;
4068 : }
4069 : }
4070 :
4071 73 : if (shared_resource == NULL) {
4072 72 : shared_resource = calloc(1, sizeof(*shared_resource));
4073 72 : if (shared_resource == NULL) {
4074 0 : spdk_put_io_channel(ch->channel);
4075 0 : spdk_put_io_channel(ch->accel_channel);
4076 0 : spdk_put_io_channel(mgmt_io_ch);
4077 0 : return -1;
4078 : }
4079 :
4080 72 : shared_resource->mgmt_ch = mgmt_ch;
4081 72 : shared_resource->io_outstanding = 0;
4082 72 : TAILQ_INIT(&shared_resource->nomem_io);
4083 72 : shared_resource->nomem_threshold = 0;
4084 72 : shared_resource->shared_ch = ch->channel;
4085 72 : shared_resource->ref = 1;
4086 72 : TAILQ_INSERT_TAIL(&mgmt_ch->shared_resources, shared_resource, link);
4087 : }
4088 :
4089 73 : ch->io_outstanding = 0;
4090 73 : TAILQ_INIT(&ch->queued_resets);
4091 73 : TAILQ_INIT(&ch->locked_ranges);
4092 73 : TAILQ_INIT(&ch->qos_queued_io);
4093 73 : ch->flags = 0;
4094 73 : ch->shared_resource = shared_resource;
4095 :
4096 73 : TAILQ_INIT(&ch->io_submitted);
4097 73 : TAILQ_INIT(&ch->io_locked);
4098 73 : TAILQ_INIT(&ch->io_accel_exec);
4099 73 : TAILQ_INIT(&ch->io_memory_domain);
4100 :
4101 73 : ch->stat = bdev_alloc_io_stat(false);
4102 73 : if (ch->stat == NULL) {
4103 0 : bdev_channel_destroy_resource(ch);
4104 0 : return -1;
4105 : }
4106 :
4107 73 : ch->stat->ticks_rate = spdk_get_ticks_hz();
4108 :
4109 : #ifdef SPDK_CONFIG_VTUNE
4110 : {
4111 : char *name;
4112 : __itt_init_ittlib(NULL, 0);
4113 : name = spdk_sprintf_alloc("spdk_bdev_%s_%p", ch->bdev->name, ch);
4114 : if (!name) {
4115 : bdev_channel_destroy_resource(ch);
4116 : return -1;
4117 : }
4118 : ch->handle = __itt_string_handle_create(name);
4119 : free(name);
4120 : ch->start_tsc = spdk_get_ticks();
4121 : ch->interval_tsc = spdk_get_ticks_hz() / 100;
4122 : ch->prev_stat = bdev_alloc_io_stat(false);
4123 : if (ch->prev_stat == NULL) {
4124 : bdev_channel_destroy_resource(ch);
4125 : return -1;
4126 : }
4127 : }
4128 : #endif
4129 :
4130 73 : spdk_spin_lock(&bdev->internal.spinlock);
4131 73 : bdev_enable_qos(bdev, ch);
4132 :
4133 74 : TAILQ_FOREACH(range, &bdev->internal.locked_ranges, tailq) {
4134 : struct lba_range *new_range;
4135 :
4136 1 : new_range = calloc(1, sizeof(*new_range));
4137 1 : if (new_range == NULL) {
4138 0 : spdk_spin_unlock(&bdev->internal.spinlock);
4139 0 : bdev_channel_destroy_resource(ch);
4140 0 : return -1;
4141 : }
4142 1 : new_range->length = range->length;
4143 1 : new_range->offset = range->offset;
4144 1 : new_range->locked_ctx = range->locked_ctx;
4145 1 : TAILQ_INSERT_TAIL(&ch->locked_ranges, new_range, tailq);
4146 : }
4147 :
4148 73 : spdk_spin_unlock(&bdev->internal.spinlock);
4149 :
4150 73 : return 0;
4151 : }
4152 :
4153 : static int
4154 0 : bdev_abort_all_buf_io_cb(struct spdk_iobuf_channel *ch, struct spdk_iobuf_entry *entry,
4155 : void *cb_ctx)
4156 : {
4157 0 : struct spdk_bdev_channel *bdev_ch = cb_ctx;
4158 : struct spdk_bdev_io *bdev_io;
4159 : uint64_t buf_len;
4160 :
4161 0 : bdev_io = SPDK_CONTAINEROF(entry, struct spdk_bdev_io, internal.iobuf);
4162 0 : if (bdev_io->internal.ch == bdev_ch) {
4163 0 : buf_len = bdev_io_get_max_buf_len(bdev_io, bdev_io->internal.buf_len);
4164 0 : spdk_iobuf_entry_abort(ch, entry, buf_len);
4165 0 : spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_ABORTED);
4166 : }
4167 :
4168 0 : return 0;
4169 : }
4170 :
4171 : /*
4172 : * Abort I/O that are waiting on a data buffer.
4173 : */
4174 : static void
4175 95 : bdev_abort_all_buf_io(struct spdk_bdev_mgmt_channel *mgmt_ch, struct spdk_bdev_channel *ch)
4176 : {
4177 95 : spdk_iobuf_for_each_entry(&mgmt_ch->iobuf, &mgmt_ch->iobuf.small,
4178 : bdev_abort_all_buf_io_cb, ch);
4179 95 : spdk_iobuf_for_each_entry(&mgmt_ch->iobuf, &mgmt_ch->iobuf.large,
4180 : bdev_abort_all_buf_io_cb, ch);
4181 95 : }
4182 :
4183 : /*
4184 : * Abort I/O that are queued waiting for submission. These types of I/O are
4185 : * linked using the spdk_bdev_io link TAILQ_ENTRY.
4186 : */
4187 : static void
4188 186 : bdev_abort_all_queued_io(bdev_io_tailq_t *queue, struct spdk_bdev_channel *ch)
4189 : {
4190 : struct spdk_bdev_io *bdev_io, *tmp;
4191 :
4192 226 : TAILQ_FOREACH_SAFE(bdev_io, queue, internal.link, tmp) {
4193 40 : if (bdev_io->internal.ch == ch) {
4194 40 : TAILQ_REMOVE(queue, bdev_io, internal.link);
4195 : /*
4196 : * spdk_bdev_io_complete() assumes that the completed I/O had
4197 : * been submitted to the bdev module. Since in this case it
4198 : * hadn't, bump io_outstanding to account for the decrement
4199 : * that spdk_bdev_io_complete() will do.
4200 : */
4201 40 : if (bdev_io->type != SPDK_BDEV_IO_TYPE_RESET) {
4202 39 : bdev_io_increment_outstanding(ch, ch->shared_resource);
4203 : }
4204 40 : spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_ABORTED);
4205 : }
4206 : }
4207 186 : }
4208 :
4209 : static bool
4210 12 : bdev_abort_queued_io(bdev_io_tailq_t *queue, struct spdk_bdev_io *bio_to_abort)
4211 : {
4212 : struct spdk_bdev_io *bdev_io;
4213 :
4214 12 : TAILQ_FOREACH(bdev_io, queue, internal.link) {
4215 0 : if (bdev_io == bio_to_abort) {
4216 0 : TAILQ_REMOVE(queue, bio_to_abort, internal.link);
4217 0 : spdk_bdev_io_complete(bio_to_abort, SPDK_BDEV_IO_STATUS_ABORTED);
4218 0 : return true;
4219 : }
4220 : }
4221 :
4222 12 : return false;
4223 : }
4224 :
4225 : static int
4226 0 : bdev_abort_buf_io_cb(struct spdk_iobuf_channel *ch, struct spdk_iobuf_entry *entry, void *cb_ctx)
4227 : {
4228 0 : struct spdk_bdev_io *bdev_io, *bio_to_abort = cb_ctx;
4229 : uint64_t buf_len;
4230 :
4231 0 : bdev_io = SPDK_CONTAINEROF(entry, struct spdk_bdev_io, internal.iobuf);
4232 0 : if (bdev_io == bio_to_abort) {
4233 0 : buf_len = bdev_io_get_max_buf_len(bdev_io, bdev_io->internal.buf_len);
4234 0 : spdk_iobuf_entry_abort(ch, entry, buf_len);
4235 0 : spdk_bdev_io_complete(bio_to_abort, SPDK_BDEV_IO_STATUS_ABORTED);
4236 0 : return 1;
4237 : }
4238 :
4239 0 : return 0;
4240 : }
4241 :
4242 : static bool
4243 10 : bdev_abort_buf_io(struct spdk_bdev_mgmt_channel *mgmt_ch, struct spdk_bdev_io *bio_to_abort)
4244 : {
4245 : int rc;
4246 :
4247 10 : rc = spdk_iobuf_for_each_entry(&mgmt_ch->iobuf, &mgmt_ch->iobuf.small,
4248 : bdev_abort_buf_io_cb, bio_to_abort);
4249 10 : if (rc == 1) {
4250 0 : return true;
4251 : }
4252 :
4253 10 : rc = spdk_iobuf_for_each_entry(&mgmt_ch->iobuf, &mgmt_ch->iobuf.large,
4254 : bdev_abort_buf_io_cb, bio_to_abort);
4255 10 : return rc == 1;
4256 : }
4257 :
4258 : static void
4259 7 : bdev_qos_channel_destroy(void *cb_arg)
4260 : {
4261 7 : struct spdk_bdev_qos *qos = cb_arg;
4262 :
4263 7 : spdk_put_io_channel(spdk_io_channel_from_ctx(qos->ch));
4264 7 : spdk_poller_unregister(&qos->poller);
4265 :
4266 7 : SPDK_DEBUGLOG(bdev, "Free QoS %p.\n", qos);
4267 :
4268 7 : free(qos);
4269 7 : }
4270 :
4271 : static int
4272 7 : bdev_qos_destroy(struct spdk_bdev *bdev)
4273 : {
4274 : int i;
4275 :
4276 : /*
4277 : * Cleanly shutting down the QoS poller is tricky, because
4278 : * during the asynchronous operation the user could open
4279 : * a new descriptor and create a new channel, spawning
4280 : * a new QoS poller.
4281 : *
4282 : * The strategy is to create a new QoS structure here and swap it
4283 : * in. The shutdown path then continues to refer to the old one
4284 : * until it completes and then releases it.
4285 : */
4286 : struct spdk_bdev_qos *new_qos, *old_qos;
4287 :
4288 7 : old_qos = bdev->internal.qos;
4289 :
4290 7 : new_qos = calloc(1, sizeof(*new_qos));
4291 7 : if (!new_qos) {
4292 0 : SPDK_ERRLOG("Unable to allocate memory to shut down QoS.\n");
4293 0 : return -ENOMEM;
4294 : }
4295 :
4296 : /* Copy the old QoS data into the newly allocated structure */
4297 7 : memcpy(new_qos, old_qos, sizeof(*new_qos));
4298 :
4299 : /* Zero out the key parts of the QoS structure */
4300 7 : new_qos->ch = NULL;
4301 7 : new_qos->thread = NULL;
4302 7 : new_qos->poller = NULL;
4303 : /*
4304 : * The limit member of spdk_bdev_qos_limit structure is not zeroed.
4305 : * It will be used later for the new QoS structure.
4306 : */
4307 35 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
4308 28 : new_qos->rate_limits[i].remaining_this_timeslice = 0;
4309 28 : new_qos->rate_limits[i].min_per_timeslice = 0;
4310 28 : new_qos->rate_limits[i].max_per_timeslice = 0;
4311 : }
4312 :
4313 7 : bdev->internal.qos = new_qos;
4314 :
4315 7 : if (old_qos->thread == NULL) {
4316 0 : free(old_qos);
4317 : } else {
4318 7 : spdk_thread_send_msg(old_qos->thread, bdev_qos_channel_destroy, old_qos);
4319 : }
4320 :
4321 : /* It is safe to continue with destroying the bdev even though the QoS channel hasn't
4322 : * been destroyed yet. The destruction path will end up waiting for the final
4323 : * channel to be put before it releases resources. */
4324 :
4325 7 : return 0;
4326 : }
4327 :
4328 : void
4329 73 : spdk_bdev_add_io_stat(struct spdk_bdev_io_stat *total, struct spdk_bdev_io_stat *add)
4330 : {
4331 73 : total->bytes_read += add->bytes_read;
4332 73 : total->num_read_ops += add->num_read_ops;
4333 73 : total->bytes_written += add->bytes_written;
4334 73 : total->num_write_ops += add->num_write_ops;
4335 73 : total->bytes_unmapped += add->bytes_unmapped;
4336 73 : total->num_unmap_ops += add->num_unmap_ops;
4337 73 : total->bytes_copied += add->bytes_copied;
4338 73 : total->num_copy_ops += add->num_copy_ops;
4339 73 : total->read_latency_ticks += add->read_latency_ticks;
4340 73 : total->write_latency_ticks += add->write_latency_ticks;
4341 73 : total->unmap_latency_ticks += add->unmap_latency_ticks;
4342 73 : total->copy_latency_ticks += add->copy_latency_ticks;
4343 73 : if (total->max_read_latency_ticks < add->max_read_latency_ticks) {
4344 5 : total->max_read_latency_ticks = add->max_read_latency_ticks;
4345 : }
4346 73 : if (total->min_read_latency_ticks > add->min_read_latency_ticks) {
4347 37 : total->min_read_latency_ticks = add->min_read_latency_ticks;
4348 : }
4349 73 : if (total->max_write_latency_ticks < add->max_write_latency_ticks) {
4350 4 : total->max_write_latency_ticks = add->max_write_latency_ticks;
4351 : }
4352 73 : if (total->min_write_latency_ticks > add->min_write_latency_ticks) {
4353 24 : total->min_write_latency_ticks = add->min_write_latency_ticks;
4354 : }
4355 73 : if (total->max_unmap_latency_ticks < add->max_unmap_latency_ticks) {
4356 0 : total->max_unmap_latency_ticks = add->max_unmap_latency_ticks;
4357 : }
4358 73 : if (total->min_unmap_latency_ticks > add->min_unmap_latency_ticks) {
4359 3 : total->min_unmap_latency_ticks = add->min_unmap_latency_ticks;
4360 : }
4361 73 : if (total->max_copy_latency_ticks < add->max_copy_latency_ticks) {
4362 0 : total->max_copy_latency_ticks = add->max_copy_latency_ticks;
4363 : }
4364 73 : if (total->min_copy_latency_ticks > add->min_copy_latency_ticks) {
4365 4 : total->min_copy_latency_ticks = add->min_copy_latency_ticks;
4366 : }
4367 73 : }
4368 :
4369 : static void
4370 1 : bdev_get_io_stat(struct spdk_bdev_io_stat *to_stat, struct spdk_bdev_io_stat *from_stat)
4371 : {
4372 1 : memcpy(to_stat, from_stat, offsetof(struct spdk_bdev_io_stat, io_error));
4373 :
4374 1 : if (to_stat->io_error != NULL && from_stat->io_error != NULL) {
4375 0 : memcpy(to_stat->io_error, from_stat->io_error,
4376 : sizeof(struct spdk_bdev_io_error_stat));
4377 : }
4378 1 : }
4379 :
4380 : void
4381 198 : spdk_bdev_reset_io_stat(struct spdk_bdev_io_stat *stat, enum spdk_bdev_reset_stat_mode mode)
4382 : {
4383 198 : stat->max_read_latency_ticks = 0;
4384 198 : stat->min_read_latency_ticks = UINT64_MAX;
4385 198 : stat->max_write_latency_ticks = 0;
4386 198 : stat->min_write_latency_ticks = UINT64_MAX;
4387 198 : stat->max_unmap_latency_ticks = 0;
4388 198 : stat->min_unmap_latency_ticks = UINT64_MAX;
4389 198 : stat->max_copy_latency_ticks = 0;
4390 198 : stat->min_copy_latency_ticks = UINT64_MAX;
4391 :
4392 198 : if (mode != SPDK_BDEV_RESET_STAT_ALL) {
4393 0 : return;
4394 : }
4395 :
4396 198 : stat->bytes_read = 0;
4397 198 : stat->num_read_ops = 0;
4398 198 : stat->bytes_written = 0;
4399 198 : stat->num_write_ops = 0;
4400 198 : stat->bytes_unmapped = 0;
4401 198 : stat->num_unmap_ops = 0;
4402 198 : stat->bytes_copied = 0;
4403 198 : stat->num_copy_ops = 0;
4404 198 : stat->read_latency_ticks = 0;
4405 198 : stat->write_latency_ticks = 0;
4406 198 : stat->unmap_latency_ticks = 0;
4407 198 : stat->copy_latency_ticks = 0;
4408 :
4409 198 : if (stat->io_error != NULL) {
4410 125 : memset(stat->io_error, 0, sizeof(struct spdk_bdev_io_error_stat));
4411 : }
4412 : }
4413 :
4414 : struct spdk_bdev_io_stat *
4415 198 : bdev_alloc_io_stat(bool io_error_stat)
4416 : {
4417 : struct spdk_bdev_io_stat *stat;
4418 :
4419 198 : stat = malloc(sizeof(struct spdk_bdev_io_stat));
4420 198 : if (stat == NULL) {
4421 0 : return NULL;
4422 : }
4423 :
4424 198 : if (io_error_stat) {
4425 125 : stat->io_error = malloc(sizeof(struct spdk_bdev_io_error_stat));
4426 125 : if (stat->io_error == NULL) {
4427 0 : free(stat);
4428 0 : return NULL;
4429 : }
4430 : } else {
4431 73 : stat->io_error = NULL;
4432 : }
4433 :
4434 198 : spdk_bdev_reset_io_stat(stat, SPDK_BDEV_RESET_STAT_ALL);
4435 :
4436 198 : return stat;
4437 : }
4438 :
4439 : void
4440 198 : bdev_free_io_stat(struct spdk_bdev_io_stat *stat)
4441 : {
4442 198 : if (stat != NULL) {
4443 198 : free(stat->io_error);
4444 198 : free(stat);
4445 : }
4446 198 : }
4447 :
4448 : void
4449 0 : spdk_bdev_dump_io_stat_json(struct spdk_bdev_io_stat *stat, struct spdk_json_write_ctx *w)
4450 : {
4451 : int i;
4452 :
4453 0 : spdk_json_write_named_uint64(w, "bytes_read", stat->bytes_read);
4454 0 : spdk_json_write_named_uint64(w, "num_read_ops", stat->num_read_ops);
4455 0 : spdk_json_write_named_uint64(w, "bytes_written", stat->bytes_written);
4456 0 : spdk_json_write_named_uint64(w, "num_write_ops", stat->num_write_ops);
4457 0 : spdk_json_write_named_uint64(w, "bytes_unmapped", stat->bytes_unmapped);
4458 0 : spdk_json_write_named_uint64(w, "num_unmap_ops", stat->num_unmap_ops);
4459 0 : spdk_json_write_named_uint64(w, "bytes_copied", stat->bytes_copied);
4460 0 : spdk_json_write_named_uint64(w, "num_copy_ops", stat->num_copy_ops);
4461 0 : spdk_json_write_named_uint64(w, "read_latency_ticks", stat->read_latency_ticks);
4462 0 : spdk_json_write_named_uint64(w, "max_read_latency_ticks", stat->max_read_latency_ticks);
4463 0 : spdk_json_write_named_uint64(w, "min_read_latency_ticks",
4464 0 : stat->min_read_latency_ticks != UINT64_MAX ?
4465 : stat->min_read_latency_ticks : 0);
4466 0 : spdk_json_write_named_uint64(w, "write_latency_ticks", stat->write_latency_ticks);
4467 0 : spdk_json_write_named_uint64(w, "max_write_latency_ticks", stat->max_write_latency_ticks);
4468 0 : spdk_json_write_named_uint64(w, "min_write_latency_ticks",
4469 0 : stat->min_write_latency_ticks != UINT64_MAX ?
4470 : stat->min_write_latency_ticks : 0);
4471 0 : spdk_json_write_named_uint64(w, "unmap_latency_ticks", stat->unmap_latency_ticks);
4472 0 : spdk_json_write_named_uint64(w, "max_unmap_latency_ticks", stat->max_unmap_latency_ticks);
4473 0 : spdk_json_write_named_uint64(w, "min_unmap_latency_ticks",
4474 0 : stat->min_unmap_latency_ticks != UINT64_MAX ?
4475 : stat->min_unmap_latency_ticks : 0);
4476 0 : spdk_json_write_named_uint64(w, "copy_latency_ticks", stat->copy_latency_ticks);
4477 0 : spdk_json_write_named_uint64(w, "max_copy_latency_ticks", stat->max_copy_latency_ticks);
4478 0 : spdk_json_write_named_uint64(w, "min_copy_latency_ticks",
4479 0 : stat->min_copy_latency_ticks != UINT64_MAX ?
4480 : stat->min_copy_latency_ticks : 0);
4481 :
4482 0 : if (stat->io_error != NULL) {
4483 0 : spdk_json_write_named_object_begin(w, "io_error");
4484 0 : for (i = 0; i < -SPDK_MIN_BDEV_IO_STATUS; i++) {
4485 0 : if (stat->io_error->error_status[i] != 0) {
4486 0 : spdk_json_write_named_uint32(w, bdev_io_status_get_string(-(i + 1)),
4487 0 : stat->io_error->error_status[i]);
4488 : }
4489 : }
4490 0 : spdk_json_write_object_end(w);
4491 : }
4492 0 : }
4493 :
4494 : static void
4495 77 : bdev_channel_abort_queued_ios(struct spdk_bdev_channel *ch)
4496 : {
4497 77 : struct spdk_bdev_shared_resource *shared_resource = ch->shared_resource;
4498 77 : struct spdk_bdev_mgmt_channel *mgmt_ch = shared_resource->mgmt_ch;
4499 :
4500 77 : bdev_abort_all_queued_io(&shared_resource->nomem_io, ch);
4501 77 : bdev_abort_all_buf_io(mgmt_ch, ch);
4502 77 : }
4503 :
4504 : static void
4505 73 : bdev_channel_destroy(void *io_device, void *ctx_buf)
4506 : {
4507 73 : struct spdk_bdev_channel *ch = ctx_buf;
4508 :
4509 73 : SPDK_DEBUGLOG(bdev, "Destroying channel %p for bdev %s on thread %p\n", ch, ch->bdev->name,
4510 : spdk_get_thread());
4511 :
4512 73 : spdk_trace_record(TRACE_BDEV_IOCH_DESTROY, 0, 0, 0, ch->bdev->name,
4513 : spdk_thread_get_id(spdk_io_channel_get_thread(ch->channel)));
4514 :
4515 : /* This channel is going away, so add its statistics into the bdev so that they don't get lost. */
4516 73 : spdk_spin_lock(&ch->bdev->internal.spinlock);
4517 73 : spdk_bdev_add_io_stat(ch->bdev->internal.stat, ch->stat);
4518 73 : spdk_spin_unlock(&ch->bdev->internal.spinlock);
4519 :
4520 73 : bdev_abort_all_queued_io(&ch->queued_resets, ch);
4521 :
4522 73 : bdev_channel_abort_queued_ios(ch);
4523 :
4524 73 : if (ch->histogram) {
4525 0 : spdk_histogram_data_free(ch->histogram);
4526 : }
4527 :
4528 73 : bdev_channel_destroy_resource(ch);
4529 73 : }
4530 :
4531 : /*
4532 : * If the name already exists in the global bdev name tree, RB_INSERT() returns a pointer
4533 : * to it. Hence we do not have to call bdev_get_by_name() when using this function.
4534 : */
4535 : static int
4536 254 : bdev_name_add(struct spdk_bdev_name *bdev_name, struct spdk_bdev *bdev, const char *name)
4537 : {
4538 : struct spdk_bdev_name *tmp;
4539 :
4540 254 : bdev_name->name = strdup(name);
4541 254 : if (bdev_name->name == NULL) {
4542 0 : SPDK_ERRLOG("Unable to allocate bdev name\n");
4543 0 : return -ENOMEM;
4544 : }
4545 :
4546 254 : bdev_name->bdev = bdev;
4547 :
4548 254 : spdk_spin_lock(&g_bdev_mgr.spinlock);
4549 254 : tmp = RB_INSERT(bdev_name_tree, &g_bdev_mgr.bdev_names, bdev_name);
4550 254 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
4551 :
4552 254 : if (tmp != NULL) {
4553 4 : SPDK_ERRLOG("Bdev name %s already exists\n", name);
4554 4 : free(bdev_name->name);
4555 4 : return -EEXIST;
4556 : }
4557 :
4558 250 : return 0;
4559 : }
4560 :
4561 : static void
4562 250 : bdev_name_del_unsafe(struct spdk_bdev_name *bdev_name)
4563 : {
4564 250 : RB_REMOVE(bdev_name_tree, &g_bdev_mgr.bdev_names, bdev_name);
4565 250 : free(bdev_name->name);
4566 250 : }
4567 :
4568 : static void
4569 6 : bdev_name_del(struct spdk_bdev_name *bdev_name)
4570 : {
4571 6 : spdk_spin_lock(&g_bdev_mgr.spinlock);
4572 6 : bdev_name_del_unsafe(bdev_name);
4573 6 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
4574 6 : }
4575 :
4576 : int
4577 130 : spdk_bdev_alias_add(struct spdk_bdev *bdev, const char *alias)
4578 : {
4579 : struct spdk_bdev_alias *tmp;
4580 : int ret;
4581 :
4582 130 : if (alias == NULL) {
4583 1 : SPDK_ERRLOG("Empty alias passed\n");
4584 1 : return -EINVAL;
4585 : }
4586 :
4587 129 : tmp = calloc(1, sizeof(*tmp));
4588 129 : if (tmp == NULL) {
4589 0 : SPDK_ERRLOG("Unable to allocate alias\n");
4590 0 : return -ENOMEM;
4591 : }
4592 :
4593 129 : ret = bdev_name_add(&tmp->alias, bdev, alias);
4594 129 : if (ret != 0) {
4595 3 : free(tmp);
4596 3 : return ret;
4597 : }
4598 :
4599 126 : TAILQ_INSERT_TAIL(&bdev->aliases, tmp, tailq);
4600 :
4601 126 : return 0;
4602 : }
4603 :
4604 : static int
4605 127 : bdev_alias_del(struct spdk_bdev *bdev, const char *alias,
4606 : void (*alias_del_fn)(struct spdk_bdev_name *n))
4607 : {
4608 : struct spdk_bdev_alias *tmp;
4609 :
4610 132 : TAILQ_FOREACH(tmp, &bdev->aliases, tailq) {
4611 128 : if (strcmp(alias, tmp->alias.name) == 0) {
4612 123 : TAILQ_REMOVE(&bdev->aliases, tmp, tailq);
4613 123 : alias_del_fn(&tmp->alias);
4614 123 : free(tmp);
4615 123 : return 0;
4616 : }
4617 : }
4618 :
4619 4 : return -ENOENT;
4620 : }
4621 :
4622 : int
4623 4 : spdk_bdev_alias_del(struct spdk_bdev *bdev, const char *alias)
4624 : {
4625 : int rc;
4626 :
4627 4 : rc = bdev_alias_del(bdev, alias, bdev_name_del);
4628 4 : if (rc == -ENOENT) {
4629 2 : SPDK_INFOLOG(bdev, "Alias %s does not exist\n", alias);
4630 : }
4631 :
4632 4 : return rc;
4633 : }
4634 :
4635 : void
4636 2 : spdk_bdev_alias_del_all(struct spdk_bdev *bdev)
4637 : {
4638 : struct spdk_bdev_alias *p, *tmp;
4639 :
4640 5 : TAILQ_FOREACH_SAFE(p, &bdev->aliases, tailq, tmp) {
4641 3 : TAILQ_REMOVE(&bdev->aliases, p, tailq);
4642 3 : bdev_name_del(&p->alias);
4643 3 : free(p);
4644 : }
4645 2 : }
4646 :
4647 : struct spdk_io_channel *
4648 75 : spdk_bdev_get_io_channel(struct spdk_bdev_desc *desc)
4649 : {
4650 75 : return spdk_get_io_channel(__bdev_to_io_dev(spdk_bdev_desc_get_bdev(desc)));
4651 : }
4652 :
4653 : void *
4654 0 : spdk_bdev_get_module_ctx(struct spdk_bdev_desc *desc)
4655 : {
4656 0 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
4657 0 : void *ctx = NULL;
4658 :
4659 0 : if (bdev->fn_table->get_module_ctx) {
4660 0 : ctx = bdev->fn_table->get_module_ctx(bdev->ctxt);
4661 : }
4662 :
4663 0 : return ctx;
4664 : }
4665 :
4666 : const char *
4667 0 : spdk_bdev_get_module_name(const struct spdk_bdev *bdev)
4668 : {
4669 0 : return bdev->module->name;
4670 : }
4671 :
4672 : const char *
4673 249 : spdk_bdev_get_name(const struct spdk_bdev *bdev)
4674 : {
4675 249 : return bdev->name;
4676 : }
4677 :
4678 : const char *
4679 0 : spdk_bdev_get_product_name(const struct spdk_bdev *bdev)
4680 : {
4681 0 : return bdev->product_name;
4682 : }
4683 :
4684 : const struct spdk_bdev_aliases_list *
4685 0 : spdk_bdev_get_aliases(const struct spdk_bdev *bdev)
4686 : {
4687 0 : return &bdev->aliases;
4688 : }
4689 :
4690 : uint32_t
4691 5 : spdk_bdev_get_block_size(const struct spdk_bdev *bdev)
4692 : {
4693 5 : return bdev->blocklen;
4694 : }
4695 :
4696 : uint32_t
4697 0 : spdk_bdev_get_write_unit_size(const struct spdk_bdev *bdev)
4698 : {
4699 0 : return bdev->write_unit_size;
4700 : }
4701 :
4702 : uint64_t
4703 0 : spdk_bdev_get_num_blocks(const struct spdk_bdev *bdev)
4704 : {
4705 0 : return bdev->blockcnt;
4706 : }
4707 :
4708 : const char *
4709 0 : spdk_bdev_get_qos_rpc_type(enum spdk_bdev_qos_rate_limit_type type)
4710 : {
4711 0 : return qos_rpc_type[type];
4712 : }
4713 :
4714 : void
4715 0 : spdk_bdev_get_qos_rate_limits(struct spdk_bdev *bdev, uint64_t *limits)
4716 : {
4717 : int i;
4718 :
4719 0 : memset(limits, 0, sizeof(*limits) * SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES);
4720 :
4721 0 : spdk_spin_lock(&bdev->internal.spinlock);
4722 0 : if (bdev->internal.qos) {
4723 0 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
4724 0 : if (bdev->internal.qos->rate_limits[i].limit !=
4725 : SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
4726 0 : limits[i] = bdev->internal.qos->rate_limits[i].limit;
4727 0 : if (bdev_qos_is_iops_rate_limit(i) == false) {
4728 : /* Change from Byte to Megabyte which is user visible. */
4729 0 : limits[i] = limits[i] / 1024 / 1024;
4730 : }
4731 : }
4732 : }
4733 : }
4734 0 : spdk_spin_unlock(&bdev->internal.spinlock);
4735 0 : }
4736 :
4737 : size_t
4738 335 : spdk_bdev_get_buf_align(const struct spdk_bdev *bdev)
4739 : {
4740 335 : return 1 << bdev->required_alignment;
4741 : }
4742 :
4743 : uint32_t
4744 0 : spdk_bdev_get_optimal_io_boundary(const struct spdk_bdev *bdev)
4745 : {
4746 0 : return bdev->optimal_io_boundary;
4747 : }
4748 :
4749 : bool
4750 0 : spdk_bdev_has_write_cache(const struct spdk_bdev *bdev)
4751 : {
4752 0 : return bdev->write_cache;
4753 : }
4754 :
4755 : const struct spdk_uuid *
4756 0 : spdk_bdev_get_uuid(const struct spdk_bdev *bdev)
4757 : {
4758 0 : return &bdev->uuid;
4759 : }
4760 :
4761 : uint16_t
4762 0 : spdk_bdev_get_acwu(const struct spdk_bdev *bdev)
4763 : {
4764 0 : return bdev->acwu;
4765 : }
4766 :
4767 : uint32_t
4768 29 : spdk_bdev_get_md_size(const struct spdk_bdev *bdev)
4769 : {
4770 29 : return bdev->md_len;
4771 : }
4772 :
4773 : bool
4774 128 : spdk_bdev_is_md_interleaved(const struct spdk_bdev *bdev)
4775 : {
4776 128 : return (bdev->md_len != 0) && bdev->md_interleave;
4777 : }
4778 :
4779 : bool
4780 170 : spdk_bdev_is_md_separate(const struct spdk_bdev *bdev)
4781 : {
4782 170 : return (bdev->md_len != 0) && !bdev->md_interleave;
4783 : }
4784 :
4785 : bool
4786 0 : spdk_bdev_is_zoned(const struct spdk_bdev *bdev)
4787 : {
4788 0 : return bdev->zoned;
4789 : }
4790 :
4791 : uint32_t
4792 119 : spdk_bdev_get_data_block_size(const struct spdk_bdev *bdev)
4793 : {
4794 119 : if (spdk_bdev_is_md_interleaved(bdev)) {
4795 0 : return bdev->blocklen - bdev->md_len;
4796 : } else {
4797 119 : return bdev->blocklen;
4798 : }
4799 : }
4800 :
4801 : uint32_t
4802 0 : spdk_bdev_get_physical_block_size(const struct spdk_bdev *bdev)
4803 : {
4804 0 : return bdev->phys_blocklen;
4805 : }
4806 :
4807 : static uint32_t
4808 9 : _bdev_get_block_size_with_md(const struct spdk_bdev *bdev)
4809 : {
4810 9 : if (!spdk_bdev_is_md_interleaved(bdev)) {
4811 6 : return bdev->blocklen + bdev->md_len;
4812 : } else {
4813 3 : return bdev->blocklen;
4814 : }
4815 : }
4816 :
4817 : /* We have to use the typedef in the function declaration to appease astyle. */
4818 : typedef enum spdk_dif_type spdk_dif_type_t;
4819 :
4820 : spdk_dif_type_t
4821 0 : spdk_bdev_get_dif_type(const struct spdk_bdev *bdev)
4822 : {
4823 0 : if (bdev->md_len != 0) {
4824 0 : return bdev->dif_type;
4825 : } else {
4826 0 : return SPDK_DIF_DISABLE;
4827 : }
4828 : }
4829 :
4830 : bool
4831 0 : spdk_bdev_is_dif_head_of_md(const struct spdk_bdev *bdev)
4832 : {
4833 0 : if (spdk_bdev_get_dif_type(bdev) != SPDK_DIF_DISABLE) {
4834 0 : return bdev->dif_is_head_of_md;
4835 : } else {
4836 0 : return false;
4837 : }
4838 : }
4839 :
4840 : bool
4841 0 : spdk_bdev_is_dif_check_enabled(const struct spdk_bdev *bdev,
4842 : enum spdk_dif_check_type check_type)
4843 : {
4844 0 : if (spdk_bdev_get_dif_type(bdev) == SPDK_DIF_DISABLE) {
4845 0 : return false;
4846 : }
4847 :
4848 0 : switch (check_type) {
4849 0 : case SPDK_DIF_CHECK_TYPE_REFTAG:
4850 0 : return (bdev->dif_check_flags & SPDK_DIF_FLAGS_REFTAG_CHECK) != 0;
4851 0 : case SPDK_DIF_CHECK_TYPE_APPTAG:
4852 0 : return (bdev->dif_check_flags & SPDK_DIF_FLAGS_APPTAG_CHECK) != 0;
4853 0 : case SPDK_DIF_CHECK_TYPE_GUARD:
4854 0 : return (bdev->dif_check_flags & SPDK_DIF_FLAGS_GUARD_CHECK) != 0;
4855 0 : default:
4856 0 : return false;
4857 : }
4858 : }
4859 :
4860 : static uint32_t
4861 3 : bdev_get_max_write(const struct spdk_bdev *bdev, uint64_t num_bytes)
4862 : {
4863 : uint64_t aligned_length, max_write_blocks;
4864 :
4865 3 : aligned_length = num_bytes - (spdk_bdev_get_buf_align(bdev) - 1);
4866 3 : max_write_blocks = aligned_length / _bdev_get_block_size_with_md(bdev);
4867 3 : max_write_blocks -= max_write_blocks % bdev->write_unit_size;
4868 :
4869 3 : return max_write_blocks;
4870 : }
4871 :
4872 : uint32_t
4873 1 : spdk_bdev_get_max_copy(const struct spdk_bdev *bdev)
4874 : {
4875 1 : return bdev->max_copy;
4876 : }
4877 :
4878 : uint64_t
4879 0 : spdk_bdev_get_qd(const struct spdk_bdev *bdev)
4880 : {
4881 0 : return bdev->internal.measured_queue_depth;
4882 : }
4883 :
4884 : uint64_t
4885 0 : spdk_bdev_get_qd_sampling_period(const struct spdk_bdev *bdev)
4886 : {
4887 0 : return bdev->internal.period;
4888 : }
4889 :
4890 : uint64_t
4891 0 : spdk_bdev_get_weighted_io_time(const struct spdk_bdev *bdev)
4892 : {
4893 0 : return bdev->internal.weighted_io_time;
4894 : }
4895 :
4896 : uint64_t
4897 0 : spdk_bdev_get_io_time(const struct spdk_bdev *bdev)
4898 : {
4899 0 : return bdev->internal.io_time;
4900 : }
4901 :
4902 : static void bdev_update_qd_sampling_period(void *ctx);
4903 :
4904 : static void
4905 1 : _calculate_measured_qd_cpl(struct spdk_bdev *bdev, void *_ctx, int status)
4906 : {
4907 1 : bdev->internal.measured_queue_depth = bdev->internal.temporary_queue_depth;
4908 :
4909 1 : if (bdev->internal.measured_queue_depth) {
4910 0 : bdev->internal.io_time += bdev->internal.period;
4911 0 : bdev->internal.weighted_io_time += bdev->internal.period * bdev->internal.measured_queue_depth;
4912 : }
4913 :
4914 1 : bdev->internal.qd_poll_in_progress = false;
4915 :
4916 1 : bdev_update_qd_sampling_period(bdev);
4917 1 : }
4918 :
4919 : static void
4920 1 : _calculate_measured_qd(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
4921 : struct spdk_io_channel *io_ch, void *_ctx)
4922 : {
4923 1 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(io_ch);
4924 :
4925 1 : bdev->internal.temporary_queue_depth += ch->io_outstanding;
4926 1 : spdk_bdev_for_each_channel_continue(i, 0);
4927 1 : }
4928 :
4929 : static int
4930 1 : bdev_calculate_measured_queue_depth(void *ctx)
4931 : {
4932 1 : struct spdk_bdev *bdev = ctx;
4933 :
4934 1 : bdev->internal.qd_poll_in_progress = true;
4935 1 : bdev->internal.temporary_queue_depth = 0;
4936 1 : spdk_bdev_for_each_channel(bdev, _calculate_measured_qd, bdev, _calculate_measured_qd_cpl);
4937 1 : return SPDK_POLLER_BUSY;
4938 : }
4939 :
4940 : static void
4941 5 : bdev_update_qd_sampling_period(void *ctx)
4942 : {
4943 5 : struct spdk_bdev *bdev = ctx;
4944 :
4945 5 : if (bdev->internal.period == bdev->internal.new_period) {
4946 0 : return;
4947 : }
4948 :
4949 5 : if (bdev->internal.qd_poll_in_progress) {
4950 1 : return;
4951 : }
4952 :
4953 4 : bdev->internal.period = bdev->internal.new_period;
4954 :
4955 4 : spdk_poller_unregister(&bdev->internal.qd_poller);
4956 4 : if (bdev->internal.period != 0) {
4957 2 : bdev->internal.qd_poller = SPDK_POLLER_REGISTER(bdev_calculate_measured_queue_depth,
4958 : bdev, bdev->internal.period);
4959 : } else {
4960 2 : spdk_bdev_close(bdev->internal.qd_desc);
4961 2 : bdev->internal.qd_desc = NULL;
4962 : }
4963 : }
4964 :
4965 : static void
4966 0 : _tmp_bdev_event_cb(enum spdk_bdev_event_type type, struct spdk_bdev *bdev, void *ctx)
4967 : {
4968 0 : SPDK_NOTICELOG("Unexpected event type: %d\n", type);
4969 0 : }
4970 :
4971 : void
4972 128 : spdk_bdev_set_qd_sampling_period(struct spdk_bdev *bdev, uint64_t period)
4973 : {
4974 : int rc;
4975 :
4976 128 : if (bdev->internal.new_period == period) {
4977 122 : return;
4978 : }
4979 :
4980 6 : bdev->internal.new_period = period;
4981 :
4982 6 : if (bdev->internal.qd_desc != NULL) {
4983 4 : assert(bdev->internal.period != 0);
4984 :
4985 4 : spdk_thread_send_msg(bdev->internal.qd_desc->thread,
4986 : bdev_update_qd_sampling_period, bdev);
4987 4 : return;
4988 : }
4989 :
4990 2 : assert(bdev->internal.period == 0);
4991 :
4992 2 : rc = spdk_bdev_open_ext(spdk_bdev_get_name(bdev), false, _tmp_bdev_event_cb,
4993 : NULL, &bdev->internal.qd_desc);
4994 2 : if (rc != 0) {
4995 0 : return;
4996 : }
4997 :
4998 2 : bdev->internal.period = period;
4999 2 : bdev->internal.qd_poller = SPDK_POLLER_REGISTER(bdev_calculate_measured_queue_depth,
5000 : bdev, period);
5001 : }
5002 :
5003 : struct bdev_get_current_qd_ctx {
5004 : uint64_t current_qd;
5005 : spdk_bdev_get_current_qd_cb cb_fn;
5006 : void *cb_arg;
5007 : };
5008 :
5009 : static void
5010 0 : bdev_get_current_qd_done(struct spdk_bdev *bdev, void *_ctx, int status)
5011 : {
5012 0 : struct bdev_get_current_qd_ctx *ctx = _ctx;
5013 :
5014 0 : ctx->cb_fn(bdev, ctx->current_qd, ctx->cb_arg, 0);
5015 :
5016 0 : free(ctx);
5017 0 : }
5018 :
5019 : static void
5020 0 : bdev_get_current_qd(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
5021 : struct spdk_io_channel *io_ch, void *_ctx)
5022 : {
5023 0 : struct bdev_get_current_qd_ctx *ctx = _ctx;
5024 0 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
5025 :
5026 0 : ctx->current_qd += bdev_ch->io_outstanding;
5027 :
5028 0 : spdk_bdev_for_each_channel_continue(i, 0);
5029 0 : }
5030 :
5031 : void
5032 0 : spdk_bdev_get_current_qd(struct spdk_bdev *bdev, spdk_bdev_get_current_qd_cb cb_fn,
5033 : void *cb_arg)
5034 : {
5035 : struct bdev_get_current_qd_ctx *ctx;
5036 :
5037 0 : assert(cb_fn != NULL);
5038 :
5039 0 : ctx = calloc(1, sizeof(*ctx));
5040 0 : if (ctx == NULL) {
5041 0 : cb_fn(bdev, 0, cb_arg, -ENOMEM);
5042 0 : return;
5043 : }
5044 :
5045 0 : ctx->cb_fn = cb_fn;
5046 0 : ctx->cb_arg = cb_arg;
5047 :
5048 0 : spdk_bdev_for_each_channel(bdev, bdev_get_current_qd, ctx, bdev_get_current_qd_done);
5049 : }
5050 :
5051 : static void
5052 25 : _event_notify(struct spdk_bdev_desc *desc, enum spdk_bdev_event_type type)
5053 : {
5054 25 : assert(desc->thread == spdk_get_thread());
5055 :
5056 25 : spdk_spin_lock(&desc->spinlock);
5057 25 : desc->refs--;
5058 25 : if (!desc->closed) {
5059 14 : spdk_spin_unlock(&desc->spinlock);
5060 14 : desc->callback.event_fn(type,
5061 : desc->bdev,
5062 : desc->callback.ctx);
5063 14 : return;
5064 11 : } else if (desc->refs == 0) {
5065 : /* This descriptor was closed after this event_notify message was sent.
5066 : * spdk_bdev_close() could not free the descriptor since this message was
5067 : * in flight, so we free it now using bdev_desc_free().
5068 : */
5069 10 : spdk_spin_unlock(&desc->spinlock);
5070 10 : bdev_desc_free(desc);
5071 10 : return;
5072 : }
5073 1 : spdk_spin_unlock(&desc->spinlock);
5074 : }
5075 :
5076 : static void
5077 25 : event_notify(struct spdk_bdev_desc *desc, spdk_msg_fn event_notify_fn)
5078 : {
5079 25 : spdk_spin_lock(&desc->spinlock);
5080 25 : desc->refs++;
5081 25 : spdk_thread_send_msg(desc->thread, event_notify_fn, desc);
5082 25 : spdk_spin_unlock(&desc->spinlock);
5083 25 : }
5084 :
5085 : static void
5086 6 : _resize_notify(void *ctx)
5087 : {
5088 6 : struct spdk_bdev_desc *desc = ctx;
5089 :
5090 6 : _event_notify(desc, SPDK_BDEV_EVENT_RESIZE);
5091 6 : }
5092 :
5093 : int
5094 11 : spdk_bdev_notify_blockcnt_change(struct spdk_bdev *bdev, uint64_t size)
5095 : {
5096 : struct spdk_bdev_desc *desc;
5097 : int ret;
5098 :
5099 11 : if (size == bdev->blockcnt) {
5100 0 : return 0;
5101 : }
5102 :
5103 11 : spdk_spin_lock(&bdev->internal.spinlock);
5104 :
5105 : /* bdev has open descriptors */
5106 11 : if (!TAILQ_EMPTY(&bdev->internal.open_descs) &&
5107 7 : bdev->blockcnt > size) {
5108 1 : ret = -EBUSY;
5109 : } else {
5110 10 : bdev->blockcnt = size;
5111 16 : TAILQ_FOREACH(desc, &bdev->internal.open_descs, link) {
5112 6 : event_notify(desc, _resize_notify);
5113 : }
5114 10 : ret = 0;
5115 : }
5116 :
5117 11 : spdk_spin_unlock(&bdev->internal.spinlock);
5118 :
5119 11 : return ret;
5120 : }
5121 :
5122 : /*
5123 : * Convert I/O offset and length from bytes to blocks.
5124 : *
5125 : * Returns zero on success or non-zero if the byte parameters aren't divisible by the block size.
5126 : */
5127 : static uint64_t
5128 19 : bdev_bytes_to_blocks(struct spdk_bdev *bdev, uint64_t offset_bytes, uint64_t *offset_blocks,
5129 : uint64_t num_bytes, uint64_t *num_blocks)
5130 : {
5131 19 : uint32_t block_size = bdev->blocklen;
5132 : uint8_t shift_cnt;
5133 :
5134 : /* Avoid expensive div operations if possible. These spdk_u32 functions are very cheap. */
5135 19 : if (spdk_likely(spdk_u32_is_pow2(block_size))) {
5136 16 : shift_cnt = spdk_u32log2(block_size);
5137 16 : *offset_blocks = offset_bytes >> shift_cnt;
5138 16 : *num_blocks = num_bytes >> shift_cnt;
5139 16 : return (offset_bytes - (*offset_blocks << shift_cnt)) |
5140 16 : (num_bytes - (*num_blocks << shift_cnt));
5141 : } else {
5142 3 : *offset_blocks = offset_bytes / block_size;
5143 3 : *num_blocks = num_bytes / block_size;
5144 3 : return (offset_bytes % block_size) | (num_bytes % block_size);
5145 : }
5146 : }
5147 :
5148 : static bool
5149 688 : bdev_io_valid_blocks(struct spdk_bdev *bdev, uint64_t offset_blocks, uint64_t num_blocks)
5150 : {
5151 : /* Return failure if offset_blocks + num_blocks is less than offset_blocks; indicates there
5152 : * has been an overflow and hence the offset has been wrapped around */
5153 688 : if (offset_blocks + num_blocks < offset_blocks) {
5154 1 : return false;
5155 : }
5156 :
5157 : /* Return failure if offset_blocks + num_blocks exceeds the size of the bdev */
5158 687 : if (offset_blocks + num_blocks > bdev->blockcnt) {
5159 2 : return false;
5160 : }
5161 :
5162 685 : return true;
5163 : }
5164 :
5165 : static void
5166 2 : bdev_seek_complete_cb(void *ctx)
5167 : {
5168 2 : struct spdk_bdev_io *bdev_io = ctx;
5169 :
5170 2 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
5171 2 : bdev_io->internal.cb(bdev_io, true, bdev_io->internal.caller_ctx);
5172 2 : }
5173 :
5174 : static int
5175 4 : bdev_seek(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5176 : uint64_t offset_blocks, enum spdk_bdev_io_type io_type,
5177 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5178 : {
5179 4 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5180 : struct spdk_bdev_io *bdev_io;
5181 4 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5182 :
5183 4 : assert(io_type == SPDK_BDEV_IO_TYPE_SEEK_DATA || io_type == SPDK_BDEV_IO_TYPE_SEEK_HOLE);
5184 :
5185 : /* Check if offset_blocks is valid looking at the validity of one block */
5186 4 : if (!bdev_io_valid_blocks(bdev, offset_blocks, 1)) {
5187 0 : return -EINVAL;
5188 : }
5189 :
5190 4 : bdev_io = bdev_channel_get_io(channel);
5191 4 : if (!bdev_io) {
5192 0 : return -ENOMEM;
5193 : }
5194 :
5195 4 : bdev_io->internal.ch = channel;
5196 4 : bdev_io->internal.desc = desc;
5197 4 : bdev_io->type = io_type;
5198 4 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5199 4 : bdev_io->u.bdev.memory_domain = NULL;
5200 4 : bdev_io->u.bdev.memory_domain_ctx = NULL;
5201 4 : bdev_io->u.bdev.accel_sequence = NULL;
5202 4 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5203 :
5204 4 : if (!spdk_bdev_io_type_supported(bdev, io_type)) {
5205 : /* In case bdev doesn't support seek to next data/hole offset,
5206 : * it is assumed that only data and no holes are present */
5207 2 : if (io_type == SPDK_BDEV_IO_TYPE_SEEK_DATA) {
5208 1 : bdev_io->u.bdev.seek.offset = offset_blocks;
5209 : } else {
5210 1 : bdev_io->u.bdev.seek.offset = UINT64_MAX;
5211 : }
5212 :
5213 2 : spdk_thread_send_msg(spdk_get_thread(), bdev_seek_complete_cb, bdev_io);
5214 2 : return 0;
5215 : }
5216 :
5217 2 : bdev_io_submit(bdev_io);
5218 2 : return 0;
5219 : }
5220 :
5221 : int
5222 2 : spdk_bdev_seek_data(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5223 : uint64_t offset_blocks,
5224 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5225 : {
5226 2 : return bdev_seek(desc, ch, offset_blocks, SPDK_BDEV_IO_TYPE_SEEK_DATA, cb, cb_arg);
5227 : }
5228 :
5229 : int
5230 2 : spdk_bdev_seek_hole(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5231 : uint64_t offset_blocks,
5232 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5233 : {
5234 2 : return bdev_seek(desc, ch, offset_blocks, SPDK_BDEV_IO_TYPE_SEEK_HOLE, cb, cb_arg);
5235 : }
5236 :
5237 : uint64_t
5238 4 : spdk_bdev_io_get_seek_offset(const struct spdk_bdev_io *bdev_io)
5239 : {
5240 4 : return bdev_io->u.bdev.seek.offset;
5241 : }
5242 :
5243 : static int
5244 203 : bdev_read_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch, void *buf,
5245 : void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
5246 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5247 : {
5248 203 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5249 : struct spdk_bdev_io *bdev_io;
5250 203 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5251 :
5252 203 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
5253 0 : return -EINVAL;
5254 : }
5255 :
5256 203 : bdev_io = bdev_channel_get_io(channel);
5257 203 : if (!bdev_io) {
5258 1 : return -ENOMEM;
5259 : }
5260 :
5261 202 : bdev_io->internal.ch = channel;
5262 202 : bdev_io->internal.desc = desc;
5263 202 : bdev_io->type = SPDK_BDEV_IO_TYPE_READ;
5264 202 : bdev_io->u.bdev.iovs = &bdev_io->iov;
5265 202 : bdev_io->u.bdev.iovs[0].iov_base = buf;
5266 202 : bdev_io->u.bdev.iovs[0].iov_len = num_blocks * bdev->blocklen;
5267 202 : bdev_io->u.bdev.iovcnt = 1;
5268 202 : bdev_io->u.bdev.md_buf = md_buf;
5269 202 : bdev_io->u.bdev.num_blocks = num_blocks;
5270 202 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5271 202 : bdev_io->u.bdev.memory_domain = NULL;
5272 202 : bdev_io->u.bdev.memory_domain_ctx = NULL;
5273 202 : bdev_io->u.bdev.accel_sequence = NULL;
5274 202 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5275 :
5276 202 : bdev_io_submit(bdev_io);
5277 202 : return 0;
5278 : }
5279 :
5280 : int
5281 2 : spdk_bdev_read(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5282 : void *buf, uint64_t offset, uint64_t nbytes,
5283 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5284 : {
5285 2 : uint64_t offset_blocks, num_blocks;
5286 :
5287 2 : if (bdev_bytes_to_blocks(spdk_bdev_desc_get_bdev(desc), offset, &offset_blocks,
5288 : nbytes, &num_blocks) != 0) {
5289 0 : return -EINVAL;
5290 : }
5291 :
5292 2 : return spdk_bdev_read_blocks(desc, ch, buf, offset_blocks, num_blocks, cb, cb_arg);
5293 : }
5294 :
5295 : int
5296 199 : spdk_bdev_read_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5297 : void *buf, uint64_t offset_blocks, uint64_t num_blocks,
5298 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5299 : {
5300 199 : return bdev_read_blocks_with_md(desc, ch, buf, NULL, offset_blocks, num_blocks, cb, cb_arg);
5301 : }
5302 :
5303 : int
5304 4 : spdk_bdev_read_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5305 : void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
5306 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5307 : {
5308 4 : struct iovec iov = {
5309 : .iov_base = buf,
5310 : };
5311 :
5312 4 : if (md_buf && !spdk_bdev_is_md_separate(spdk_bdev_desc_get_bdev(desc))) {
5313 0 : return -EINVAL;
5314 : }
5315 :
5316 4 : if (md_buf && !_is_buf_allocated(&iov)) {
5317 0 : return -EINVAL;
5318 : }
5319 :
5320 4 : return bdev_read_blocks_with_md(desc, ch, buf, md_buf, offset_blocks, num_blocks,
5321 : cb, cb_arg);
5322 : }
5323 :
5324 : int
5325 5 : spdk_bdev_readv(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5326 : struct iovec *iov, int iovcnt,
5327 : uint64_t offset, uint64_t nbytes,
5328 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5329 : {
5330 5 : uint64_t offset_blocks, num_blocks;
5331 :
5332 5 : if (bdev_bytes_to_blocks(spdk_bdev_desc_get_bdev(desc), offset, &offset_blocks,
5333 : nbytes, &num_blocks) != 0) {
5334 0 : return -EINVAL;
5335 : }
5336 :
5337 5 : return spdk_bdev_readv_blocks(desc, ch, iov, iovcnt, offset_blocks, num_blocks, cb, cb_arg);
5338 : }
5339 :
5340 : static int
5341 235 : bdev_readv_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5342 : struct iovec *iov, int iovcnt, void *md_buf, uint64_t offset_blocks,
5343 : uint64_t num_blocks, struct spdk_memory_domain *domain, void *domain_ctx,
5344 : struct spdk_accel_sequence *seq,
5345 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5346 : {
5347 235 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5348 : struct spdk_bdev_io *bdev_io;
5349 235 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5350 :
5351 235 : if (spdk_unlikely(!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks))) {
5352 0 : return -EINVAL;
5353 : }
5354 :
5355 235 : bdev_io = bdev_channel_get_io(channel);
5356 235 : if (spdk_unlikely(!bdev_io)) {
5357 2 : return -ENOMEM;
5358 : }
5359 :
5360 233 : bdev_io->internal.ch = channel;
5361 233 : bdev_io->internal.desc = desc;
5362 233 : bdev_io->type = SPDK_BDEV_IO_TYPE_READ;
5363 233 : bdev_io->u.bdev.iovs = iov;
5364 233 : bdev_io->u.bdev.iovcnt = iovcnt;
5365 233 : bdev_io->u.bdev.md_buf = md_buf;
5366 233 : bdev_io->u.bdev.num_blocks = num_blocks;
5367 233 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5368 233 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5369 233 : bdev_io->internal.memory_domain = domain;
5370 233 : bdev_io->internal.memory_domain_ctx = domain_ctx;
5371 233 : bdev_io->internal.accel_sequence = seq;
5372 233 : bdev_io->internal.has_accel_sequence = seq != NULL;
5373 233 : bdev_io->u.bdev.memory_domain = domain;
5374 233 : bdev_io->u.bdev.memory_domain_ctx = domain_ctx;
5375 233 : bdev_io->u.bdev.accel_sequence = seq;
5376 :
5377 233 : _bdev_io_submit_ext(desc, bdev_io);
5378 :
5379 233 : return 0;
5380 : }
5381 :
5382 : int
5383 24 : spdk_bdev_readv_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5384 : struct iovec *iov, int iovcnt,
5385 : uint64_t offset_blocks, uint64_t num_blocks,
5386 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5387 : {
5388 24 : return bdev_readv_blocks_with_md(desc, ch, iov, iovcnt, NULL, offset_blocks,
5389 : num_blocks, NULL, NULL, NULL, cb, cb_arg);
5390 : }
5391 :
5392 : int
5393 4 : spdk_bdev_readv_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5394 : struct iovec *iov, int iovcnt, void *md_buf,
5395 : uint64_t offset_blocks, uint64_t num_blocks,
5396 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5397 : {
5398 4 : if (md_buf && !spdk_bdev_is_md_separate(spdk_bdev_desc_get_bdev(desc))) {
5399 0 : return -EINVAL;
5400 : }
5401 :
5402 4 : if (md_buf && !_is_buf_allocated(iov)) {
5403 0 : return -EINVAL;
5404 : }
5405 :
5406 4 : return bdev_readv_blocks_with_md(desc, ch, iov, iovcnt, md_buf, offset_blocks,
5407 : num_blocks, NULL, NULL, NULL, cb, cb_arg);
5408 : }
5409 :
5410 : static inline bool
5411 14 : _bdev_io_check_opts(struct spdk_bdev_ext_io_opts *opts, struct iovec *iov)
5412 : {
5413 : /*
5414 : * We check if opts size is at least of size when we first introduced
5415 : * spdk_bdev_ext_io_opts (ac6f2bdd8d) since access to those members
5416 : * are not checked internal.
5417 : */
5418 14 : return opts->size >= offsetof(struct spdk_bdev_ext_io_opts, metadata) +
5419 10 : sizeof(opts->metadata) &&
5420 22 : opts->size <= sizeof(*opts) &&
5421 : /* When memory domain is used, the user must provide data buffers */
5422 8 : (!opts->memory_domain || (iov && iov[0].iov_base));
5423 : }
5424 :
5425 : int
5426 8 : spdk_bdev_readv_blocks_ext(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5427 : struct iovec *iov, int iovcnt,
5428 : uint64_t offset_blocks, uint64_t num_blocks,
5429 : spdk_bdev_io_completion_cb cb, void *cb_arg,
5430 : struct spdk_bdev_ext_io_opts *opts)
5431 : {
5432 8 : void *md = NULL;
5433 :
5434 8 : if (opts) {
5435 7 : if (spdk_unlikely(!_bdev_io_check_opts(opts, iov))) {
5436 3 : return -EINVAL;
5437 : }
5438 :
5439 4 : md = opts->metadata;
5440 4 : if (md) {
5441 4 : if (spdk_unlikely(!spdk_bdev_is_md_separate(spdk_bdev_desc_get_bdev(desc)))) {
5442 0 : return -EINVAL;
5443 : }
5444 :
5445 4 : if (spdk_unlikely(!_is_buf_allocated(iov))) {
5446 0 : return -EINVAL;
5447 : }
5448 : }
5449 : }
5450 :
5451 17 : return bdev_readv_blocks_with_md(desc, ch, iov, iovcnt, md, offset_blocks,
5452 : num_blocks,
5453 4 : bdev_get_ext_io_opt(opts, memory_domain, NULL),
5454 4 : bdev_get_ext_io_opt(opts, memory_domain_ctx, NULL),
5455 4 : bdev_get_ext_io_opt(opts, accel_sequence, NULL),
5456 : cb, cb_arg);
5457 : }
5458 :
5459 : static int
5460 33 : bdev_write_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5461 : void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
5462 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5463 : {
5464 33 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5465 : struct spdk_bdev_io *bdev_io;
5466 33 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5467 :
5468 33 : if (!desc->write) {
5469 0 : return -EBADF;
5470 : }
5471 :
5472 33 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
5473 0 : return -EINVAL;
5474 : }
5475 :
5476 33 : bdev_io = bdev_channel_get_io(channel);
5477 33 : if (!bdev_io) {
5478 0 : return -ENOMEM;
5479 : }
5480 :
5481 33 : bdev_io->internal.ch = channel;
5482 33 : bdev_io->internal.desc = desc;
5483 33 : bdev_io->type = SPDK_BDEV_IO_TYPE_WRITE;
5484 33 : bdev_io->u.bdev.iovs = &bdev_io->iov;
5485 33 : bdev_io->u.bdev.iovs[0].iov_base = buf;
5486 33 : bdev_io->u.bdev.iovs[0].iov_len = num_blocks * bdev->blocklen;
5487 33 : bdev_io->u.bdev.iovcnt = 1;
5488 33 : bdev_io->u.bdev.md_buf = md_buf;
5489 33 : bdev_io->u.bdev.num_blocks = num_blocks;
5490 33 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5491 33 : bdev_io->u.bdev.memory_domain = NULL;
5492 33 : bdev_io->u.bdev.memory_domain_ctx = NULL;
5493 33 : bdev_io->u.bdev.accel_sequence = NULL;
5494 33 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5495 :
5496 33 : bdev_io_submit(bdev_io);
5497 33 : return 0;
5498 : }
5499 :
5500 : int
5501 3 : spdk_bdev_write(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5502 : void *buf, uint64_t offset, uint64_t nbytes,
5503 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5504 : {
5505 3 : uint64_t offset_blocks, num_blocks;
5506 :
5507 3 : if (bdev_bytes_to_blocks(spdk_bdev_desc_get_bdev(desc), offset, &offset_blocks,
5508 : nbytes, &num_blocks) != 0) {
5509 0 : return -EINVAL;
5510 : }
5511 :
5512 3 : return spdk_bdev_write_blocks(desc, ch, buf, offset_blocks, num_blocks, cb, cb_arg);
5513 : }
5514 :
5515 : int
5516 24 : spdk_bdev_write_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5517 : void *buf, uint64_t offset_blocks, uint64_t num_blocks,
5518 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5519 : {
5520 24 : return bdev_write_blocks_with_md(desc, ch, buf, NULL, offset_blocks, num_blocks,
5521 : cb, cb_arg);
5522 : }
5523 :
5524 : int
5525 3 : spdk_bdev_write_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5526 : void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
5527 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5528 : {
5529 3 : struct iovec iov = {
5530 : .iov_base = buf,
5531 : };
5532 :
5533 3 : if (md_buf && !spdk_bdev_is_md_separate(spdk_bdev_desc_get_bdev(desc))) {
5534 0 : return -EINVAL;
5535 : }
5536 :
5537 3 : if (md_buf && !_is_buf_allocated(&iov)) {
5538 0 : return -EINVAL;
5539 : }
5540 :
5541 3 : return bdev_write_blocks_with_md(desc, ch, buf, md_buf, offset_blocks, num_blocks,
5542 : cb, cb_arg);
5543 : }
5544 :
5545 : static int
5546 64 : bdev_writev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5547 : struct iovec *iov, int iovcnt, void *md_buf,
5548 : uint64_t offset_blocks, uint64_t num_blocks,
5549 : struct spdk_memory_domain *domain, void *domain_ctx,
5550 : struct spdk_accel_sequence *seq,
5551 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5552 : {
5553 64 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5554 : struct spdk_bdev_io *bdev_io;
5555 64 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5556 :
5557 64 : if (spdk_unlikely(!desc->write)) {
5558 0 : return -EBADF;
5559 : }
5560 :
5561 64 : if (spdk_unlikely(!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks))) {
5562 0 : return -EINVAL;
5563 : }
5564 :
5565 64 : bdev_io = bdev_channel_get_io(channel);
5566 64 : if (spdk_unlikely(!bdev_io)) {
5567 2 : return -ENOMEM;
5568 : }
5569 :
5570 62 : bdev_io->internal.ch = channel;
5571 62 : bdev_io->internal.desc = desc;
5572 62 : bdev_io->type = SPDK_BDEV_IO_TYPE_WRITE;
5573 62 : bdev_io->u.bdev.iovs = iov;
5574 62 : bdev_io->u.bdev.iovcnt = iovcnt;
5575 62 : bdev_io->u.bdev.md_buf = md_buf;
5576 62 : bdev_io->u.bdev.num_blocks = num_blocks;
5577 62 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5578 62 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5579 62 : bdev_io->internal.memory_domain = domain;
5580 62 : bdev_io->internal.memory_domain_ctx = domain_ctx;
5581 62 : bdev_io->internal.accel_sequence = seq;
5582 62 : bdev_io->internal.has_accel_sequence = seq != NULL;
5583 62 : bdev_io->u.bdev.memory_domain = domain;
5584 62 : bdev_io->u.bdev.memory_domain_ctx = domain_ctx;
5585 62 : bdev_io->u.bdev.accel_sequence = seq;
5586 :
5587 62 : _bdev_io_submit_ext(desc, bdev_io);
5588 :
5589 62 : return 0;
5590 : }
5591 :
5592 : int
5593 3 : spdk_bdev_writev(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5594 : struct iovec *iov, int iovcnt,
5595 : uint64_t offset, uint64_t len,
5596 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5597 : {
5598 3 : uint64_t offset_blocks, num_blocks;
5599 :
5600 3 : if (bdev_bytes_to_blocks(spdk_bdev_desc_get_bdev(desc), offset, &offset_blocks,
5601 : len, &num_blocks) != 0) {
5602 0 : return -EINVAL;
5603 : }
5604 :
5605 3 : return spdk_bdev_writev_blocks(desc, ch, iov, iovcnt, offset_blocks, num_blocks, cb, cb_arg);
5606 : }
5607 :
5608 : int
5609 14 : spdk_bdev_writev_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5610 : struct iovec *iov, int iovcnt,
5611 : uint64_t offset_blocks, uint64_t num_blocks,
5612 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5613 : {
5614 14 : return bdev_writev_blocks_with_md(desc, ch, iov, iovcnt, NULL, offset_blocks,
5615 : num_blocks, NULL, NULL, NULL, cb, cb_arg);
5616 : }
5617 :
5618 : int
5619 1 : spdk_bdev_writev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5620 : struct iovec *iov, int iovcnt, void *md_buf,
5621 : uint64_t offset_blocks, uint64_t num_blocks,
5622 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5623 : {
5624 1 : if (md_buf && !spdk_bdev_is_md_separate(spdk_bdev_desc_get_bdev(desc))) {
5625 0 : return -EINVAL;
5626 : }
5627 :
5628 1 : if (md_buf && !_is_buf_allocated(iov)) {
5629 0 : return -EINVAL;
5630 : }
5631 :
5632 1 : return bdev_writev_blocks_with_md(desc, ch, iov, iovcnt, md_buf, offset_blocks,
5633 : num_blocks, NULL, NULL, NULL, cb, cb_arg);
5634 : }
5635 :
5636 : int
5637 8 : spdk_bdev_writev_blocks_ext(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5638 : struct iovec *iov, int iovcnt,
5639 : uint64_t offset_blocks, uint64_t num_blocks,
5640 : spdk_bdev_io_completion_cb cb, void *cb_arg,
5641 : struct spdk_bdev_ext_io_opts *opts)
5642 : {
5643 8 : void *md = NULL;
5644 :
5645 8 : if (opts) {
5646 7 : if (spdk_unlikely(!_bdev_io_check_opts(opts, iov))) {
5647 3 : return -EINVAL;
5648 : }
5649 :
5650 4 : md = opts->metadata;
5651 4 : if (md) {
5652 4 : if (spdk_unlikely(!spdk_bdev_is_md_separate(spdk_bdev_desc_get_bdev(desc)))) {
5653 0 : return -EINVAL;
5654 : }
5655 :
5656 4 : if (spdk_unlikely(!_is_buf_allocated(iov))) {
5657 0 : return -EINVAL;
5658 : }
5659 : }
5660 : }
5661 :
5662 17 : return bdev_writev_blocks_with_md(desc, ch, iov, iovcnt, md, offset_blocks, num_blocks,
5663 4 : bdev_get_ext_io_opt(opts, memory_domain, NULL),
5664 4 : bdev_get_ext_io_opt(opts, memory_domain_ctx, NULL),
5665 4 : bdev_get_ext_io_opt(opts, accel_sequence, NULL),
5666 : cb, cb_arg);
5667 : }
5668 :
5669 : static void
5670 11 : bdev_compare_do_read_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
5671 : {
5672 11 : struct spdk_bdev_io *parent_io = cb_arg;
5673 11 : struct spdk_bdev *bdev = parent_io->bdev;
5674 11 : uint8_t *read_buf = bdev_io->u.bdev.iovs[0].iov_base;
5675 11 : int i, rc = 0;
5676 :
5677 11 : if (!success) {
5678 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
5679 0 : parent_io->internal.cb(parent_io, false, parent_io->internal.caller_ctx);
5680 0 : spdk_bdev_free_io(bdev_io);
5681 0 : return;
5682 : }
5683 :
5684 17 : for (i = 0; i < parent_io->u.bdev.iovcnt; i++) {
5685 11 : rc = memcmp(read_buf,
5686 11 : parent_io->u.bdev.iovs[i].iov_base,
5687 11 : parent_io->u.bdev.iovs[i].iov_len);
5688 11 : if (rc) {
5689 5 : break;
5690 : }
5691 6 : read_buf += parent_io->u.bdev.iovs[i].iov_len;
5692 : }
5693 :
5694 11 : if (rc == 0 && parent_io->u.bdev.md_buf && spdk_bdev_is_md_separate(bdev)) {
5695 2 : rc = memcmp(bdev_io->u.bdev.md_buf,
5696 2 : parent_io->u.bdev.md_buf,
5697 2 : spdk_bdev_get_md_size(bdev));
5698 : }
5699 :
5700 11 : spdk_bdev_free_io(bdev_io);
5701 :
5702 11 : if (rc == 0) {
5703 5 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
5704 5 : parent_io->internal.cb(parent_io, true, parent_io->internal.caller_ctx);
5705 : } else {
5706 6 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_MISCOMPARE;
5707 6 : parent_io->internal.cb(parent_io, false, parent_io->internal.caller_ctx);
5708 : }
5709 : }
5710 :
5711 : static void
5712 11 : bdev_compare_do_read(void *_bdev_io)
5713 : {
5714 11 : struct spdk_bdev_io *bdev_io = _bdev_io;
5715 : int rc;
5716 :
5717 11 : rc = spdk_bdev_read_blocks(bdev_io->internal.desc,
5718 11 : spdk_io_channel_from_ctx(bdev_io->internal.ch), NULL,
5719 : bdev_io->u.bdev.offset_blocks, bdev_io->u.bdev.num_blocks,
5720 : bdev_compare_do_read_done, bdev_io);
5721 :
5722 11 : if (rc == -ENOMEM) {
5723 0 : bdev_queue_io_wait_with_cb(bdev_io, bdev_compare_do_read);
5724 11 : } else if (rc != 0) {
5725 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
5726 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
5727 : }
5728 11 : }
5729 :
5730 : static int
5731 16 : bdev_comparev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5732 : struct iovec *iov, int iovcnt, void *md_buf,
5733 : uint64_t offset_blocks, uint64_t num_blocks,
5734 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5735 : {
5736 16 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5737 : struct spdk_bdev_io *bdev_io;
5738 16 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5739 :
5740 16 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
5741 0 : return -EINVAL;
5742 : }
5743 :
5744 16 : bdev_io = bdev_channel_get_io(channel);
5745 16 : if (!bdev_io) {
5746 0 : return -ENOMEM;
5747 : }
5748 :
5749 16 : bdev_io->internal.ch = channel;
5750 16 : bdev_io->internal.desc = desc;
5751 16 : bdev_io->type = SPDK_BDEV_IO_TYPE_COMPARE;
5752 16 : bdev_io->u.bdev.iovs = iov;
5753 16 : bdev_io->u.bdev.iovcnt = iovcnt;
5754 16 : bdev_io->u.bdev.md_buf = md_buf;
5755 16 : bdev_io->u.bdev.num_blocks = num_blocks;
5756 16 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5757 16 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5758 16 : bdev_io->u.bdev.memory_domain = NULL;
5759 16 : bdev_io->u.bdev.memory_domain_ctx = NULL;
5760 16 : bdev_io->u.bdev.accel_sequence = NULL;
5761 :
5762 16 : if (bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COMPARE)) {
5763 7 : bdev_io_submit(bdev_io);
5764 7 : return 0;
5765 : }
5766 :
5767 9 : bdev_compare_do_read(bdev_io);
5768 :
5769 9 : return 0;
5770 : }
5771 :
5772 : int
5773 10 : spdk_bdev_comparev_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5774 : struct iovec *iov, int iovcnt,
5775 : uint64_t offset_blocks, uint64_t num_blocks,
5776 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5777 : {
5778 10 : return bdev_comparev_blocks_with_md(desc, ch, iov, iovcnt, NULL, offset_blocks,
5779 : num_blocks, cb, cb_arg);
5780 : }
5781 :
5782 : int
5783 6 : spdk_bdev_comparev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5784 : struct iovec *iov, int iovcnt, void *md_buf,
5785 : uint64_t offset_blocks, uint64_t num_blocks,
5786 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5787 : {
5788 6 : if (md_buf && !spdk_bdev_is_md_separate(spdk_bdev_desc_get_bdev(desc))) {
5789 0 : return -EINVAL;
5790 : }
5791 :
5792 6 : if (md_buf && !_is_buf_allocated(iov)) {
5793 0 : return -EINVAL;
5794 : }
5795 :
5796 6 : return bdev_comparev_blocks_with_md(desc, ch, iov, iovcnt, md_buf, offset_blocks,
5797 : num_blocks, cb, cb_arg);
5798 : }
5799 :
5800 : static int
5801 4 : bdev_compare_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5802 : void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
5803 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5804 : {
5805 4 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5806 : struct spdk_bdev_io *bdev_io;
5807 4 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5808 :
5809 4 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
5810 0 : return -EINVAL;
5811 : }
5812 :
5813 4 : bdev_io = bdev_channel_get_io(channel);
5814 4 : if (!bdev_io) {
5815 0 : return -ENOMEM;
5816 : }
5817 :
5818 4 : bdev_io->internal.ch = channel;
5819 4 : bdev_io->internal.desc = desc;
5820 4 : bdev_io->type = SPDK_BDEV_IO_TYPE_COMPARE;
5821 4 : bdev_io->u.bdev.iovs = &bdev_io->iov;
5822 4 : bdev_io->u.bdev.iovs[0].iov_base = buf;
5823 4 : bdev_io->u.bdev.iovs[0].iov_len = num_blocks * bdev->blocklen;
5824 4 : bdev_io->u.bdev.iovcnt = 1;
5825 4 : bdev_io->u.bdev.md_buf = md_buf;
5826 4 : bdev_io->u.bdev.num_blocks = num_blocks;
5827 4 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5828 4 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5829 4 : bdev_io->u.bdev.memory_domain = NULL;
5830 4 : bdev_io->u.bdev.memory_domain_ctx = NULL;
5831 4 : bdev_io->u.bdev.accel_sequence = NULL;
5832 :
5833 4 : if (bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COMPARE)) {
5834 2 : bdev_io_submit(bdev_io);
5835 2 : return 0;
5836 : }
5837 :
5838 2 : bdev_compare_do_read(bdev_io);
5839 :
5840 2 : return 0;
5841 : }
5842 :
5843 : int
5844 4 : spdk_bdev_compare_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5845 : void *buf, uint64_t offset_blocks, uint64_t num_blocks,
5846 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5847 : {
5848 4 : return bdev_compare_blocks_with_md(desc, ch, buf, NULL, offset_blocks, num_blocks,
5849 : cb, cb_arg);
5850 : }
5851 :
5852 : int
5853 0 : spdk_bdev_compare_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5854 : void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
5855 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5856 : {
5857 0 : struct iovec iov = {
5858 : .iov_base = buf,
5859 : };
5860 :
5861 0 : if (md_buf && !spdk_bdev_is_md_separate(spdk_bdev_desc_get_bdev(desc))) {
5862 0 : return -EINVAL;
5863 : }
5864 :
5865 0 : if (md_buf && !_is_buf_allocated(&iov)) {
5866 0 : return -EINVAL;
5867 : }
5868 :
5869 0 : return bdev_compare_blocks_with_md(desc, ch, buf, md_buf, offset_blocks, num_blocks,
5870 : cb, cb_arg);
5871 : }
5872 :
5873 : static void
5874 2 : bdev_comparev_and_writev_blocks_unlocked(struct lba_range *range, void *ctx, int unlock_status)
5875 : {
5876 2 : struct spdk_bdev_io *bdev_io = ctx;
5877 :
5878 2 : if (unlock_status) {
5879 0 : SPDK_ERRLOG("LBA range unlock failed\n");
5880 : }
5881 :
5882 2 : bdev_io->internal.cb(bdev_io, bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS ? true :
5883 : false, bdev_io->internal.caller_ctx);
5884 2 : }
5885 :
5886 : static void
5887 2 : bdev_comparev_and_writev_blocks_unlock(struct spdk_bdev_io *bdev_io, int status)
5888 : {
5889 2 : bdev_io->internal.status = status;
5890 :
5891 2 : bdev_unlock_lba_range(bdev_io->internal.desc, spdk_io_channel_from_ctx(bdev_io->internal.ch),
5892 : bdev_io->u.bdev.offset_blocks, bdev_io->u.bdev.num_blocks,
5893 : bdev_comparev_and_writev_blocks_unlocked, bdev_io);
5894 2 : }
5895 :
5896 : static void
5897 1 : bdev_compare_and_write_do_write_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
5898 : {
5899 1 : struct spdk_bdev_io *parent_io = cb_arg;
5900 :
5901 1 : if (!success) {
5902 0 : SPDK_ERRLOG("Compare and write operation failed\n");
5903 : }
5904 :
5905 1 : spdk_bdev_free_io(bdev_io);
5906 :
5907 1 : bdev_comparev_and_writev_blocks_unlock(parent_io,
5908 : success ? SPDK_BDEV_IO_STATUS_SUCCESS : SPDK_BDEV_IO_STATUS_FAILED);
5909 1 : }
5910 :
5911 : static void
5912 1 : bdev_compare_and_write_do_write(void *_bdev_io)
5913 : {
5914 1 : struct spdk_bdev_io *bdev_io = _bdev_io;
5915 : int rc;
5916 :
5917 1 : rc = spdk_bdev_writev_blocks(bdev_io->internal.desc,
5918 1 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
5919 : bdev_io->u.bdev.fused_iovs, bdev_io->u.bdev.fused_iovcnt,
5920 : bdev_io->u.bdev.offset_blocks, bdev_io->u.bdev.num_blocks,
5921 : bdev_compare_and_write_do_write_done, bdev_io);
5922 :
5923 :
5924 1 : if (rc == -ENOMEM) {
5925 0 : bdev_queue_io_wait_with_cb(bdev_io, bdev_compare_and_write_do_write);
5926 1 : } else if (rc != 0) {
5927 0 : bdev_comparev_and_writev_blocks_unlock(bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
5928 : }
5929 1 : }
5930 :
5931 : static void
5932 2 : bdev_compare_and_write_do_compare_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
5933 : {
5934 2 : struct spdk_bdev_io *parent_io = cb_arg;
5935 :
5936 2 : spdk_bdev_free_io(bdev_io);
5937 :
5938 2 : if (!success) {
5939 1 : bdev_comparev_and_writev_blocks_unlock(parent_io, SPDK_BDEV_IO_STATUS_MISCOMPARE);
5940 1 : return;
5941 : }
5942 :
5943 1 : bdev_compare_and_write_do_write(parent_io);
5944 : }
5945 :
5946 : static void
5947 2 : bdev_compare_and_write_do_compare(void *_bdev_io)
5948 : {
5949 2 : struct spdk_bdev_io *bdev_io = _bdev_io;
5950 : int rc;
5951 :
5952 2 : rc = spdk_bdev_comparev_blocks(bdev_io->internal.desc,
5953 2 : spdk_io_channel_from_ctx(bdev_io->internal.ch), bdev_io->u.bdev.iovs,
5954 : bdev_io->u.bdev.iovcnt, bdev_io->u.bdev.offset_blocks, bdev_io->u.bdev.num_blocks,
5955 : bdev_compare_and_write_do_compare_done, bdev_io);
5956 :
5957 2 : if (rc == -ENOMEM) {
5958 0 : bdev_queue_io_wait_with_cb(bdev_io, bdev_compare_and_write_do_compare);
5959 2 : } else if (rc != 0) {
5960 0 : bdev_comparev_and_writev_blocks_unlock(bdev_io, SPDK_BDEV_IO_STATUS_FIRST_FUSED_FAILED);
5961 : }
5962 2 : }
5963 :
5964 : static void
5965 2 : bdev_comparev_and_writev_blocks_locked(struct lba_range *range, void *ctx, int status)
5966 : {
5967 2 : struct spdk_bdev_io *bdev_io = ctx;
5968 :
5969 2 : if (status) {
5970 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FIRST_FUSED_FAILED;
5971 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
5972 0 : return;
5973 : }
5974 :
5975 2 : bdev_compare_and_write_do_compare(bdev_io);
5976 : }
5977 :
5978 : int
5979 2 : spdk_bdev_comparev_and_writev_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5980 : struct iovec *compare_iov, int compare_iovcnt,
5981 : struct iovec *write_iov, int write_iovcnt,
5982 : uint64_t offset_blocks, uint64_t num_blocks,
5983 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5984 : {
5985 2 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5986 : struct spdk_bdev_io *bdev_io;
5987 2 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5988 :
5989 2 : if (!desc->write) {
5990 0 : return -EBADF;
5991 : }
5992 :
5993 2 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
5994 0 : return -EINVAL;
5995 : }
5996 :
5997 2 : if (num_blocks > bdev->acwu) {
5998 0 : return -EINVAL;
5999 : }
6000 :
6001 2 : bdev_io = bdev_channel_get_io(channel);
6002 2 : if (!bdev_io) {
6003 0 : return -ENOMEM;
6004 : }
6005 :
6006 2 : bdev_io->internal.ch = channel;
6007 2 : bdev_io->internal.desc = desc;
6008 2 : bdev_io->type = SPDK_BDEV_IO_TYPE_COMPARE_AND_WRITE;
6009 2 : bdev_io->u.bdev.iovs = compare_iov;
6010 2 : bdev_io->u.bdev.iovcnt = compare_iovcnt;
6011 2 : bdev_io->u.bdev.fused_iovs = write_iov;
6012 2 : bdev_io->u.bdev.fused_iovcnt = write_iovcnt;
6013 2 : bdev_io->u.bdev.md_buf = NULL;
6014 2 : bdev_io->u.bdev.num_blocks = num_blocks;
6015 2 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6016 2 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6017 2 : bdev_io->u.bdev.memory_domain = NULL;
6018 2 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6019 2 : bdev_io->u.bdev.accel_sequence = NULL;
6020 :
6021 2 : if (bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COMPARE_AND_WRITE)) {
6022 0 : bdev_io_submit(bdev_io);
6023 0 : return 0;
6024 : }
6025 :
6026 2 : return bdev_lock_lba_range(desc, ch, offset_blocks, num_blocks,
6027 : bdev_comparev_and_writev_blocks_locked, bdev_io);
6028 : }
6029 :
6030 : int
6031 2 : spdk_bdev_zcopy_start(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6032 : struct iovec *iov, int iovcnt,
6033 : uint64_t offset_blocks, uint64_t num_blocks,
6034 : bool populate,
6035 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6036 : {
6037 2 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6038 : struct spdk_bdev_io *bdev_io;
6039 2 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6040 :
6041 2 : if (!desc->write) {
6042 0 : return -EBADF;
6043 : }
6044 :
6045 2 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6046 0 : return -EINVAL;
6047 : }
6048 :
6049 2 : if (!spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_ZCOPY)) {
6050 0 : return -ENOTSUP;
6051 : }
6052 :
6053 2 : bdev_io = bdev_channel_get_io(channel);
6054 2 : if (!bdev_io) {
6055 0 : return -ENOMEM;
6056 : }
6057 :
6058 2 : bdev_io->internal.ch = channel;
6059 2 : bdev_io->internal.desc = desc;
6060 2 : bdev_io->type = SPDK_BDEV_IO_TYPE_ZCOPY;
6061 2 : bdev_io->u.bdev.num_blocks = num_blocks;
6062 2 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6063 2 : bdev_io->u.bdev.iovs = iov;
6064 2 : bdev_io->u.bdev.iovcnt = iovcnt;
6065 2 : bdev_io->u.bdev.md_buf = NULL;
6066 2 : bdev_io->u.bdev.zcopy.populate = populate ? 1 : 0;
6067 2 : bdev_io->u.bdev.zcopy.commit = 0;
6068 2 : bdev_io->u.bdev.zcopy.start = 1;
6069 2 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6070 2 : bdev_io->u.bdev.memory_domain = NULL;
6071 2 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6072 2 : bdev_io->u.bdev.accel_sequence = NULL;
6073 :
6074 2 : bdev_io_submit(bdev_io);
6075 :
6076 2 : return 0;
6077 : }
6078 :
6079 : int
6080 2 : spdk_bdev_zcopy_end(struct spdk_bdev_io *bdev_io, bool commit,
6081 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6082 : {
6083 2 : if (bdev_io->type != SPDK_BDEV_IO_TYPE_ZCOPY) {
6084 0 : return -EINVAL;
6085 : }
6086 :
6087 2 : bdev_io->u.bdev.zcopy.commit = commit ? 1 : 0;
6088 2 : bdev_io->u.bdev.zcopy.start = 0;
6089 2 : bdev_io->internal.caller_ctx = cb_arg;
6090 2 : bdev_io->internal.cb = cb;
6091 2 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_PENDING;
6092 :
6093 2 : bdev_io_submit(bdev_io);
6094 :
6095 2 : return 0;
6096 : }
6097 :
6098 : int
6099 0 : spdk_bdev_write_zeroes(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6100 : uint64_t offset, uint64_t len,
6101 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6102 : {
6103 0 : uint64_t offset_blocks, num_blocks;
6104 :
6105 0 : if (bdev_bytes_to_blocks(spdk_bdev_desc_get_bdev(desc), offset, &offset_blocks,
6106 : len, &num_blocks) != 0) {
6107 0 : return -EINVAL;
6108 : }
6109 :
6110 0 : return spdk_bdev_write_zeroes_blocks(desc, ch, offset_blocks, num_blocks, cb, cb_arg);
6111 : }
6112 :
6113 : int
6114 33 : spdk_bdev_write_zeroes_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6115 : uint64_t offset_blocks, uint64_t num_blocks,
6116 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6117 : {
6118 33 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6119 : struct spdk_bdev_io *bdev_io;
6120 33 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6121 :
6122 33 : if (!desc->write) {
6123 0 : return -EBADF;
6124 : }
6125 :
6126 33 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6127 0 : return -EINVAL;
6128 : }
6129 :
6130 33 : if (!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE_ZEROES) &&
6131 10 : !bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE)) {
6132 1 : return -ENOTSUP;
6133 : }
6134 :
6135 32 : bdev_io = bdev_channel_get_io(channel);
6136 :
6137 32 : if (!bdev_io) {
6138 0 : return -ENOMEM;
6139 : }
6140 :
6141 32 : bdev_io->type = SPDK_BDEV_IO_TYPE_WRITE_ZEROES;
6142 32 : bdev_io->internal.ch = channel;
6143 32 : bdev_io->internal.desc = desc;
6144 32 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6145 32 : bdev_io->u.bdev.num_blocks = num_blocks;
6146 32 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6147 32 : bdev_io->u.bdev.memory_domain = NULL;
6148 32 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6149 32 : bdev_io->u.bdev.accel_sequence = NULL;
6150 :
6151 : /* If the write_zeroes size is large and should be split, use the generic split
6152 : * logic regardless of whether SPDK_BDEV_IO_TYPE_WRITE_ZEREOS is supported or not.
6153 : *
6154 : * Then, send the write_zeroes request if SPDK_BDEV_IO_TYPE_WRITE_ZEROES is supported
6155 : * or emulate it using regular write request otherwise.
6156 : */
6157 32 : if (bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE_ZEROES) ||
6158 9 : bdev_io->internal.split) {
6159 26 : bdev_io_submit(bdev_io);
6160 26 : return 0;
6161 : }
6162 :
6163 6 : assert(_bdev_get_block_size_with_md(bdev) <= ZERO_BUFFER_SIZE);
6164 :
6165 6 : return bdev_write_zero_buffer(bdev_io);
6166 : }
6167 :
6168 : int
6169 0 : spdk_bdev_unmap(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6170 : uint64_t offset, uint64_t nbytes,
6171 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6172 : {
6173 0 : uint64_t offset_blocks, num_blocks;
6174 :
6175 0 : if (bdev_bytes_to_blocks(spdk_bdev_desc_get_bdev(desc), offset, &offset_blocks,
6176 : nbytes, &num_blocks) != 0) {
6177 0 : return -EINVAL;
6178 : }
6179 :
6180 0 : return spdk_bdev_unmap_blocks(desc, ch, offset_blocks, num_blocks, cb, cb_arg);
6181 : }
6182 :
6183 : int
6184 22 : spdk_bdev_unmap_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6185 : uint64_t offset_blocks, uint64_t num_blocks,
6186 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6187 : {
6188 22 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6189 : struct spdk_bdev_io *bdev_io;
6190 22 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6191 :
6192 22 : if (!desc->write) {
6193 0 : return -EBADF;
6194 : }
6195 :
6196 22 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6197 0 : return -EINVAL;
6198 : }
6199 :
6200 22 : if (num_blocks == 0) {
6201 0 : SPDK_ERRLOG("Can't unmap 0 bytes\n");
6202 0 : return -EINVAL;
6203 : }
6204 :
6205 22 : bdev_io = bdev_channel_get_io(channel);
6206 22 : if (!bdev_io) {
6207 0 : return -ENOMEM;
6208 : }
6209 :
6210 22 : bdev_io->internal.ch = channel;
6211 22 : bdev_io->internal.desc = desc;
6212 22 : bdev_io->type = SPDK_BDEV_IO_TYPE_UNMAP;
6213 :
6214 22 : bdev_io->u.bdev.iovs = &bdev_io->iov;
6215 22 : bdev_io->u.bdev.iovs[0].iov_base = NULL;
6216 22 : bdev_io->u.bdev.iovs[0].iov_len = 0;
6217 22 : bdev_io->u.bdev.iovcnt = 1;
6218 :
6219 22 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6220 22 : bdev_io->u.bdev.num_blocks = num_blocks;
6221 22 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6222 22 : bdev_io->u.bdev.memory_domain = NULL;
6223 22 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6224 22 : bdev_io->u.bdev.accel_sequence = NULL;
6225 :
6226 22 : bdev_io_submit(bdev_io);
6227 22 : return 0;
6228 : }
6229 :
6230 : int
6231 0 : spdk_bdev_flush(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6232 : uint64_t offset, uint64_t length,
6233 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6234 : {
6235 0 : uint64_t offset_blocks, num_blocks;
6236 :
6237 0 : if (bdev_bytes_to_blocks(spdk_bdev_desc_get_bdev(desc), offset, &offset_blocks,
6238 : length, &num_blocks) != 0) {
6239 0 : return -EINVAL;
6240 : }
6241 :
6242 0 : return spdk_bdev_flush_blocks(desc, ch, offset_blocks, num_blocks, cb, cb_arg);
6243 : }
6244 :
6245 : int
6246 2 : spdk_bdev_flush_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6247 : uint64_t offset_blocks, uint64_t num_blocks,
6248 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6249 : {
6250 2 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6251 : struct spdk_bdev_io *bdev_io;
6252 2 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6253 :
6254 2 : if (!desc->write) {
6255 0 : return -EBADF;
6256 : }
6257 :
6258 2 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6259 0 : return -EINVAL;
6260 : }
6261 :
6262 2 : bdev_io = bdev_channel_get_io(channel);
6263 2 : if (!bdev_io) {
6264 0 : return -ENOMEM;
6265 : }
6266 :
6267 2 : bdev_io->internal.ch = channel;
6268 2 : bdev_io->internal.desc = desc;
6269 2 : bdev_io->type = SPDK_BDEV_IO_TYPE_FLUSH;
6270 2 : bdev_io->u.bdev.iovs = NULL;
6271 2 : bdev_io->u.bdev.iovcnt = 0;
6272 2 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6273 2 : bdev_io->u.bdev.num_blocks = num_blocks;
6274 2 : bdev_io->u.bdev.memory_domain = NULL;
6275 2 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6276 2 : bdev_io->u.bdev.accel_sequence = NULL;
6277 2 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6278 :
6279 2 : bdev_io_submit(bdev_io);
6280 2 : return 0;
6281 : }
6282 :
6283 : static int bdev_reset_poll_for_outstanding_io(void *ctx);
6284 :
6285 : static void
6286 13 : bdev_reset_check_outstanding_io_done(struct spdk_bdev *bdev, void *_ctx, int status)
6287 : {
6288 13 : struct spdk_bdev_channel *ch = _ctx;
6289 : struct spdk_bdev_io *bdev_io;
6290 :
6291 13 : bdev_io = TAILQ_FIRST(&ch->queued_resets);
6292 :
6293 13 : if (status == -EBUSY) {
6294 9 : if (spdk_get_ticks() < bdev_io->u.reset.wait_poller.stop_time_tsc) {
6295 8 : bdev_io->u.reset.wait_poller.poller = SPDK_POLLER_REGISTER(bdev_reset_poll_for_outstanding_io,
6296 : ch, BDEV_RESET_CHECK_OUTSTANDING_IO_PERIOD);
6297 : } else {
6298 1 : TAILQ_REMOVE(&ch->queued_resets, bdev_io, internal.link);
6299 :
6300 1 : if (TAILQ_EMPTY(&ch->io_memory_domain) && TAILQ_EMPTY(&ch->io_accel_exec)) {
6301 : /* If outstanding IOs are still present and reset_io_drain_timeout
6302 : * seconds passed, start the reset. */
6303 1 : bdev_io_submit_reset(bdev_io);
6304 : } else {
6305 : /* We still have in progress memory domain pull/push or we're
6306 : * executing accel sequence. Since we cannot abort either of those
6307 : * operaions, fail the reset request. */
6308 0 : spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
6309 : }
6310 : }
6311 : } else {
6312 4 : TAILQ_REMOVE(&ch->queued_resets, bdev_io, internal.link);
6313 4 : SPDK_DEBUGLOG(bdev,
6314 : "Skipping reset for underlying device of bdev: %s - no outstanding I/O.\n",
6315 : ch->bdev->name);
6316 : /* Mark the completion status as a SUCCESS and complete the reset. */
6317 4 : spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_SUCCESS);
6318 : }
6319 13 : }
6320 :
6321 : static void
6322 13 : bdev_reset_check_outstanding_io(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
6323 : struct spdk_io_channel *io_ch, void *_ctx)
6324 : {
6325 13 : struct spdk_bdev_channel *cur_ch = __io_ch_to_bdev_ch(io_ch);
6326 13 : int status = 0;
6327 :
6328 13 : if (cur_ch->io_outstanding > 0 ||
6329 4 : !TAILQ_EMPTY(&cur_ch->io_memory_domain) ||
6330 4 : !TAILQ_EMPTY(&cur_ch->io_accel_exec)) {
6331 : /* If a channel has outstanding IO, set status to -EBUSY code. This will stop
6332 : * further iteration over the rest of the channels and pass non-zero status
6333 : * to the callback function. */
6334 9 : status = -EBUSY;
6335 : }
6336 13 : spdk_bdev_for_each_channel_continue(i, status);
6337 13 : }
6338 :
6339 : static int
6340 8 : bdev_reset_poll_for_outstanding_io(void *ctx)
6341 : {
6342 8 : struct spdk_bdev_channel *ch = ctx;
6343 : struct spdk_bdev_io *bdev_io;
6344 :
6345 8 : bdev_io = TAILQ_FIRST(&ch->queued_resets);
6346 :
6347 8 : spdk_poller_unregister(&bdev_io->u.reset.wait_poller.poller);
6348 8 : spdk_bdev_for_each_channel(ch->bdev, bdev_reset_check_outstanding_io, ch,
6349 : bdev_reset_check_outstanding_io_done);
6350 :
6351 8 : return SPDK_POLLER_BUSY;
6352 : }
6353 :
6354 : static void
6355 15 : bdev_reset_freeze_channel_done(struct spdk_bdev *bdev, void *_ctx, int status)
6356 : {
6357 15 : struct spdk_bdev_channel *ch = _ctx;
6358 : struct spdk_bdev_io *bdev_io;
6359 :
6360 15 : bdev_io = TAILQ_FIRST(&ch->queued_resets);
6361 :
6362 15 : if (bdev->reset_io_drain_timeout == 0) {
6363 10 : TAILQ_REMOVE(&ch->queued_resets, bdev_io, internal.link);
6364 :
6365 10 : bdev_io_submit_reset(bdev_io);
6366 10 : return;
6367 : }
6368 :
6369 5 : bdev_io->u.reset.wait_poller.stop_time_tsc = spdk_get_ticks() +
6370 5 : (ch->bdev->reset_io_drain_timeout * spdk_get_ticks_hz());
6371 :
6372 : /* In case bdev->reset_io_drain_timeout is not equal to zero,
6373 : * submit the reset to the underlying module only if outstanding I/O
6374 : * remain after reset_io_drain_timeout seconds have passed. */
6375 5 : spdk_bdev_for_each_channel(ch->bdev, bdev_reset_check_outstanding_io, ch,
6376 : bdev_reset_check_outstanding_io_done);
6377 : }
6378 :
6379 : static void
6380 18 : bdev_reset_freeze_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
6381 : struct spdk_io_channel *ch, void *_ctx)
6382 : {
6383 : struct spdk_bdev_channel *channel;
6384 : struct spdk_bdev_mgmt_channel *mgmt_channel;
6385 : struct spdk_bdev_shared_resource *shared_resource;
6386 18 : bdev_io_tailq_t tmp_queued;
6387 :
6388 18 : TAILQ_INIT(&tmp_queued);
6389 :
6390 18 : channel = __io_ch_to_bdev_ch(ch);
6391 18 : shared_resource = channel->shared_resource;
6392 18 : mgmt_channel = shared_resource->mgmt_ch;
6393 :
6394 18 : channel->flags |= BDEV_CH_RESET_IN_PROGRESS;
6395 :
6396 18 : if ((channel->flags & BDEV_CH_QOS_ENABLED) != 0) {
6397 2 : TAILQ_SWAP(&channel->qos_queued_io, &tmp_queued, spdk_bdev_io, internal.link);
6398 : }
6399 :
6400 18 : bdev_abort_all_queued_io(&shared_resource->nomem_io, channel);
6401 18 : bdev_abort_all_buf_io(mgmt_channel, channel);
6402 18 : bdev_abort_all_queued_io(&tmp_queued, channel);
6403 :
6404 18 : spdk_bdev_for_each_channel_continue(i, 0);
6405 18 : }
6406 :
6407 : static void
6408 15 : bdev_start_reset(void *ctx)
6409 : {
6410 15 : struct spdk_bdev_channel *ch = ctx;
6411 :
6412 15 : spdk_bdev_for_each_channel(ch->bdev, bdev_reset_freeze_channel, ch,
6413 : bdev_reset_freeze_channel_done);
6414 15 : }
6415 :
6416 : static void
6417 16 : bdev_channel_start_reset(struct spdk_bdev_channel *ch)
6418 : {
6419 16 : struct spdk_bdev *bdev = ch->bdev;
6420 :
6421 16 : assert(!TAILQ_EMPTY(&ch->queued_resets));
6422 :
6423 16 : spdk_spin_lock(&bdev->internal.spinlock);
6424 16 : if (bdev->internal.reset_in_progress == NULL) {
6425 15 : bdev->internal.reset_in_progress = TAILQ_FIRST(&ch->queued_resets);
6426 : /*
6427 : * Take a channel reference for the target bdev for the life of this
6428 : * reset. This guards against the channel getting destroyed while
6429 : * spdk_bdev_for_each_channel() calls related to this reset IO are in
6430 : * progress. We will release the reference when this reset is
6431 : * completed.
6432 : */
6433 15 : bdev->internal.reset_in_progress->u.reset.ch_ref = spdk_get_io_channel(__bdev_to_io_dev(bdev));
6434 15 : bdev_start_reset(ch);
6435 : }
6436 16 : spdk_spin_unlock(&bdev->internal.spinlock);
6437 16 : }
6438 :
6439 : int
6440 16 : spdk_bdev_reset(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6441 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6442 : {
6443 16 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6444 : struct spdk_bdev_io *bdev_io;
6445 16 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6446 :
6447 16 : bdev_io = bdev_channel_get_io(channel);
6448 16 : if (!bdev_io) {
6449 0 : return -ENOMEM;
6450 : }
6451 :
6452 16 : bdev_io->internal.ch = channel;
6453 16 : bdev_io->internal.desc = desc;
6454 16 : bdev_io->internal.submit_tsc = spdk_get_ticks();
6455 16 : bdev_io->type = SPDK_BDEV_IO_TYPE_RESET;
6456 16 : bdev_io->u.reset.ch_ref = NULL;
6457 16 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6458 :
6459 16 : spdk_spin_lock(&bdev->internal.spinlock);
6460 16 : TAILQ_INSERT_TAIL(&channel->queued_resets, bdev_io, internal.link);
6461 16 : spdk_spin_unlock(&bdev->internal.spinlock);
6462 :
6463 16 : TAILQ_INSERT_TAIL(&bdev_io->internal.ch->io_submitted, bdev_io,
6464 : internal.ch_link);
6465 :
6466 16 : bdev_channel_start_reset(channel);
6467 :
6468 16 : return 0;
6469 : }
6470 :
6471 : void
6472 0 : spdk_bdev_get_io_stat(struct spdk_bdev *bdev, struct spdk_io_channel *ch,
6473 : struct spdk_bdev_io_stat *stat)
6474 : {
6475 0 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6476 :
6477 0 : bdev_get_io_stat(stat, channel->stat);
6478 0 : }
6479 :
6480 : static void
6481 1 : bdev_get_device_stat_done(struct spdk_bdev *bdev, void *_ctx, int status)
6482 : {
6483 1 : struct spdk_bdev_iostat_ctx *bdev_iostat_ctx = _ctx;
6484 :
6485 1 : bdev_iostat_ctx->cb(bdev, bdev_iostat_ctx->stat,
6486 : bdev_iostat_ctx->cb_arg, 0);
6487 1 : free(bdev_iostat_ctx);
6488 1 : }
6489 :
6490 : static void
6491 0 : bdev_get_each_channel_stat(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
6492 : struct spdk_io_channel *ch, void *_ctx)
6493 : {
6494 0 : struct spdk_bdev_iostat_ctx *bdev_iostat_ctx = _ctx;
6495 0 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6496 :
6497 0 : spdk_bdev_add_io_stat(bdev_iostat_ctx->stat, channel->stat);
6498 0 : spdk_bdev_for_each_channel_continue(i, 0);
6499 0 : }
6500 :
6501 : void
6502 1 : spdk_bdev_get_device_stat(struct spdk_bdev *bdev, struct spdk_bdev_io_stat *stat,
6503 : spdk_bdev_get_device_stat_cb cb, void *cb_arg)
6504 : {
6505 : struct spdk_bdev_iostat_ctx *bdev_iostat_ctx;
6506 :
6507 1 : assert(bdev != NULL);
6508 1 : assert(stat != NULL);
6509 1 : assert(cb != NULL);
6510 :
6511 1 : bdev_iostat_ctx = calloc(1, sizeof(struct spdk_bdev_iostat_ctx));
6512 1 : if (bdev_iostat_ctx == NULL) {
6513 0 : SPDK_ERRLOG("Unable to allocate memory for spdk_bdev_iostat_ctx\n");
6514 0 : cb(bdev, stat, cb_arg, -ENOMEM);
6515 0 : return;
6516 : }
6517 :
6518 1 : bdev_iostat_ctx->stat = stat;
6519 1 : bdev_iostat_ctx->cb = cb;
6520 1 : bdev_iostat_ctx->cb_arg = cb_arg;
6521 :
6522 : /* Start with the statistics from previously deleted channels. */
6523 1 : spdk_spin_lock(&bdev->internal.spinlock);
6524 1 : bdev_get_io_stat(bdev_iostat_ctx->stat, bdev->internal.stat);
6525 1 : spdk_spin_unlock(&bdev->internal.spinlock);
6526 :
6527 : /* Then iterate and add the statistics from each existing channel. */
6528 1 : spdk_bdev_for_each_channel(bdev, bdev_get_each_channel_stat, bdev_iostat_ctx,
6529 : bdev_get_device_stat_done);
6530 : }
6531 :
6532 : struct bdev_iostat_reset_ctx {
6533 : enum spdk_bdev_reset_stat_mode mode;
6534 : bdev_reset_device_stat_cb cb;
6535 : void *cb_arg;
6536 : };
6537 :
6538 : static void
6539 0 : bdev_reset_device_stat_done(struct spdk_bdev *bdev, void *_ctx, int status)
6540 : {
6541 0 : struct bdev_iostat_reset_ctx *ctx = _ctx;
6542 :
6543 0 : ctx->cb(bdev, ctx->cb_arg, 0);
6544 :
6545 0 : free(ctx);
6546 0 : }
6547 :
6548 : static void
6549 0 : bdev_reset_each_channel_stat(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
6550 : struct spdk_io_channel *ch, void *_ctx)
6551 : {
6552 0 : struct bdev_iostat_reset_ctx *ctx = _ctx;
6553 0 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6554 :
6555 0 : spdk_bdev_reset_io_stat(channel->stat, ctx->mode);
6556 :
6557 0 : spdk_bdev_for_each_channel_continue(i, 0);
6558 0 : }
6559 :
6560 : void
6561 0 : bdev_reset_device_stat(struct spdk_bdev *bdev, enum spdk_bdev_reset_stat_mode mode,
6562 : bdev_reset_device_stat_cb cb, void *cb_arg)
6563 : {
6564 : struct bdev_iostat_reset_ctx *ctx;
6565 :
6566 0 : assert(bdev != NULL);
6567 0 : assert(cb != NULL);
6568 :
6569 0 : ctx = calloc(1, sizeof(*ctx));
6570 0 : if (ctx == NULL) {
6571 0 : SPDK_ERRLOG("Unable to allocate bdev_iostat_reset_ctx.\n");
6572 0 : cb(bdev, cb_arg, -ENOMEM);
6573 0 : return;
6574 : }
6575 :
6576 0 : ctx->mode = mode;
6577 0 : ctx->cb = cb;
6578 0 : ctx->cb_arg = cb_arg;
6579 :
6580 0 : spdk_spin_lock(&bdev->internal.spinlock);
6581 0 : spdk_bdev_reset_io_stat(bdev->internal.stat, mode);
6582 0 : spdk_spin_unlock(&bdev->internal.spinlock);
6583 :
6584 0 : spdk_bdev_for_each_channel(bdev,
6585 : bdev_reset_each_channel_stat,
6586 : ctx,
6587 : bdev_reset_device_stat_done);
6588 : }
6589 :
6590 : int
6591 1 : spdk_bdev_nvme_admin_passthru(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6592 : const struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes,
6593 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6594 : {
6595 1 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6596 : struct spdk_bdev_io *bdev_io;
6597 1 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6598 :
6599 1 : if (!desc->write) {
6600 0 : return -EBADF;
6601 : }
6602 :
6603 1 : if (spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_ADMIN))) {
6604 1 : return -ENOTSUP;
6605 : }
6606 :
6607 0 : bdev_io = bdev_channel_get_io(channel);
6608 0 : if (!bdev_io) {
6609 0 : return -ENOMEM;
6610 : }
6611 :
6612 0 : bdev_io->internal.ch = channel;
6613 0 : bdev_io->internal.desc = desc;
6614 0 : bdev_io->type = SPDK_BDEV_IO_TYPE_NVME_ADMIN;
6615 0 : bdev_io->u.nvme_passthru.cmd = *cmd;
6616 0 : bdev_io->u.nvme_passthru.buf = buf;
6617 0 : bdev_io->u.nvme_passthru.nbytes = nbytes;
6618 0 : bdev_io->u.nvme_passthru.md_buf = NULL;
6619 0 : bdev_io->u.nvme_passthru.md_len = 0;
6620 :
6621 0 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6622 :
6623 0 : bdev_io_submit(bdev_io);
6624 0 : return 0;
6625 : }
6626 :
6627 : int
6628 1 : spdk_bdev_nvme_io_passthru(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6629 : const struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes,
6630 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6631 : {
6632 1 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6633 : struct spdk_bdev_io *bdev_io;
6634 1 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6635 :
6636 1 : if (!desc->write) {
6637 : /*
6638 : * Do not try to parse the NVMe command - we could maybe use bits in the opcode
6639 : * to easily determine if the command is a read or write, but for now just
6640 : * do not allow io_passthru with a read-only descriptor.
6641 : */
6642 0 : return -EBADF;
6643 : }
6644 :
6645 1 : if (spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_IO))) {
6646 1 : return -ENOTSUP;
6647 : }
6648 :
6649 0 : bdev_io = bdev_channel_get_io(channel);
6650 0 : if (!bdev_io) {
6651 0 : return -ENOMEM;
6652 : }
6653 :
6654 0 : bdev_io->internal.ch = channel;
6655 0 : bdev_io->internal.desc = desc;
6656 0 : bdev_io->type = SPDK_BDEV_IO_TYPE_NVME_IO;
6657 0 : bdev_io->u.nvme_passthru.cmd = *cmd;
6658 0 : bdev_io->u.nvme_passthru.buf = buf;
6659 0 : bdev_io->u.nvme_passthru.nbytes = nbytes;
6660 0 : bdev_io->u.nvme_passthru.md_buf = NULL;
6661 0 : bdev_io->u.nvme_passthru.md_len = 0;
6662 :
6663 0 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6664 :
6665 0 : bdev_io_submit(bdev_io);
6666 0 : return 0;
6667 : }
6668 :
6669 : int
6670 1 : spdk_bdev_nvme_io_passthru_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6671 : const struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes, void *md_buf, size_t md_len,
6672 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6673 : {
6674 1 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6675 : struct spdk_bdev_io *bdev_io;
6676 1 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6677 :
6678 1 : if (!desc->write) {
6679 : /*
6680 : * Do not try to parse the NVMe command - we could maybe use bits in the opcode
6681 : * to easily determine if the command is a read or write, but for now just
6682 : * do not allow io_passthru with a read-only descriptor.
6683 : */
6684 0 : return -EBADF;
6685 : }
6686 :
6687 1 : if (spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_IO_MD))) {
6688 1 : return -ENOTSUP;
6689 : }
6690 :
6691 0 : bdev_io = bdev_channel_get_io(channel);
6692 0 : if (!bdev_io) {
6693 0 : return -ENOMEM;
6694 : }
6695 :
6696 0 : bdev_io->internal.ch = channel;
6697 0 : bdev_io->internal.desc = desc;
6698 0 : bdev_io->type = SPDK_BDEV_IO_TYPE_NVME_IO_MD;
6699 0 : bdev_io->u.nvme_passthru.cmd = *cmd;
6700 0 : bdev_io->u.nvme_passthru.buf = buf;
6701 0 : bdev_io->u.nvme_passthru.nbytes = nbytes;
6702 0 : bdev_io->u.nvme_passthru.md_buf = md_buf;
6703 0 : bdev_io->u.nvme_passthru.md_len = md_len;
6704 :
6705 0 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6706 :
6707 0 : bdev_io_submit(bdev_io);
6708 0 : return 0;
6709 : }
6710 :
6711 : int
6712 0 : spdk_bdev_nvme_iov_passthru_md(struct spdk_bdev_desc *desc,
6713 : struct spdk_io_channel *ch,
6714 : const struct spdk_nvme_cmd *cmd,
6715 : struct iovec *iov, int iovcnt, size_t nbytes,
6716 : void *md_buf, size_t md_len,
6717 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6718 : {
6719 0 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6720 : struct spdk_bdev_io *bdev_io;
6721 0 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6722 :
6723 0 : if (!desc->write) {
6724 : /*
6725 : * Do not try to parse the NVMe command - we could maybe use bits in the opcode
6726 : * to easily determine if the command is a read or write, but for now just
6727 : * do not allow io_passthru with a read-only descriptor.
6728 : */
6729 0 : return -EBADF;
6730 : }
6731 :
6732 0 : if (md_buf && spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_IO_MD))) {
6733 0 : return -ENOTSUP;
6734 0 : } else if (spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_IO))) {
6735 0 : return -ENOTSUP;
6736 : }
6737 :
6738 0 : bdev_io = bdev_channel_get_io(channel);
6739 0 : if (!bdev_io) {
6740 0 : return -ENOMEM;
6741 : }
6742 :
6743 0 : bdev_io->internal.ch = channel;
6744 0 : bdev_io->internal.desc = desc;
6745 0 : bdev_io->type = SPDK_BDEV_IO_TYPE_NVME_IOV_MD;
6746 0 : bdev_io->u.nvme_passthru.cmd = *cmd;
6747 0 : bdev_io->u.nvme_passthru.iovs = iov;
6748 0 : bdev_io->u.nvme_passthru.iovcnt = iovcnt;
6749 0 : bdev_io->u.nvme_passthru.nbytes = nbytes;
6750 0 : bdev_io->u.nvme_passthru.md_buf = md_buf;
6751 0 : bdev_io->u.nvme_passthru.md_len = md_len;
6752 :
6753 0 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6754 :
6755 0 : bdev_io_submit(bdev_io);
6756 0 : return 0;
6757 : }
6758 :
6759 : static void bdev_abort_retry(void *ctx);
6760 : static void bdev_abort(struct spdk_bdev_io *parent_io);
6761 :
6762 : static void
6763 13 : bdev_abort_io_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
6764 : {
6765 13 : struct spdk_bdev_channel *channel = bdev_io->internal.ch;
6766 13 : struct spdk_bdev_io *parent_io = cb_arg;
6767 : struct spdk_bdev_io *bio_to_abort, *tmp_io;
6768 :
6769 13 : bio_to_abort = bdev_io->u.abort.bio_to_abort;
6770 :
6771 13 : spdk_bdev_free_io(bdev_io);
6772 :
6773 13 : if (!success) {
6774 : /* Check if the target I/O completed in the meantime. */
6775 2 : TAILQ_FOREACH(tmp_io, &channel->io_submitted, internal.ch_link) {
6776 1 : if (tmp_io == bio_to_abort) {
6777 0 : break;
6778 : }
6779 : }
6780 :
6781 : /* If the target I/O still exists, set the parent to failed. */
6782 1 : if (tmp_io != NULL) {
6783 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
6784 : }
6785 : }
6786 :
6787 13 : parent_io->u.bdev.split_outstanding--;
6788 13 : if (parent_io->u.bdev.split_outstanding == 0) {
6789 10 : if (parent_io->internal.status == SPDK_BDEV_IO_STATUS_NOMEM) {
6790 0 : bdev_abort_retry(parent_io);
6791 : } else {
6792 10 : bdev_io_complete(parent_io);
6793 : }
6794 : }
6795 13 : }
6796 :
6797 : static int
6798 14 : bdev_abort_io(struct spdk_bdev_desc *desc, struct spdk_bdev_channel *channel,
6799 : struct spdk_bdev_io *bio_to_abort,
6800 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6801 : {
6802 14 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6803 : struct spdk_bdev_io *bdev_io;
6804 :
6805 14 : if (bio_to_abort->type == SPDK_BDEV_IO_TYPE_ABORT ||
6806 14 : bio_to_abort->type == SPDK_BDEV_IO_TYPE_RESET) {
6807 : /* TODO: Abort reset or abort request. */
6808 0 : return -ENOTSUP;
6809 : }
6810 :
6811 14 : bdev_io = bdev_channel_get_io(channel);
6812 14 : if (bdev_io == NULL) {
6813 1 : return -ENOMEM;
6814 : }
6815 :
6816 13 : bdev_io->internal.ch = channel;
6817 13 : bdev_io->internal.desc = desc;
6818 13 : bdev_io->type = SPDK_BDEV_IO_TYPE_ABORT;
6819 13 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6820 :
6821 13 : if (bdev->split_on_optimal_io_boundary && bio_to_abort->internal.split) {
6822 3 : assert(bdev_io_should_split(bio_to_abort));
6823 3 : bdev_io->u.bdev.abort.bio_cb_arg = bio_to_abort;
6824 :
6825 : /* Parent abort request is not submitted directly, but to manage its
6826 : * execution add it to the submitted list here.
6827 : */
6828 3 : bdev_io->internal.submit_tsc = spdk_get_ticks();
6829 3 : TAILQ_INSERT_TAIL(&channel->io_submitted, bdev_io, internal.ch_link);
6830 :
6831 3 : bdev_abort(bdev_io);
6832 :
6833 3 : return 0;
6834 : }
6835 :
6836 10 : bdev_io->u.abort.bio_to_abort = bio_to_abort;
6837 :
6838 : /* Submit the abort request to the underlying bdev module. */
6839 10 : bdev_io_submit(bdev_io);
6840 :
6841 10 : return 0;
6842 : }
6843 :
6844 : static bool
6845 28 : bdev_io_on_tailq(struct spdk_bdev_io *bdev_io, bdev_io_tailq_t *tailq)
6846 : {
6847 : struct spdk_bdev_io *iter;
6848 :
6849 28 : TAILQ_FOREACH(iter, tailq, internal.link) {
6850 0 : if (iter == bdev_io) {
6851 0 : return true;
6852 : }
6853 : }
6854 :
6855 28 : return false;
6856 : }
6857 :
6858 : static uint32_t
6859 12 : _bdev_abort(struct spdk_bdev_io *parent_io)
6860 : {
6861 12 : struct spdk_bdev_desc *desc = parent_io->internal.desc;
6862 12 : struct spdk_bdev_channel *channel = parent_io->internal.ch;
6863 : void *bio_cb_arg;
6864 : struct spdk_bdev_io *bio_to_abort;
6865 : uint32_t matched_ios;
6866 : int rc;
6867 :
6868 12 : bio_cb_arg = parent_io->u.bdev.abort.bio_cb_arg;
6869 :
6870 : /* matched_ios is returned and will be kept by the caller.
6871 : *
6872 : * This function will be used for two cases, 1) the same cb_arg is used for
6873 : * multiple I/Os, 2) a single large I/O is split into smaller ones.
6874 : * Incrementing split_outstanding directly here may confuse readers especially
6875 : * for the 1st case.
6876 : *
6877 : * Completion of I/O abort is processed after stack unwinding. Hence this trick
6878 : * works as expected.
6879 : */
6880 12 : matched_ios = 0;
6881 12 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
6882 :
6883 63 : TAILQ_FOREACH(bio_to_abort, &channel->io_submitted, internal.ch_link) {
6884 52 : if (bio_to_abort->internal.caller_ctx != bio_cb_arg) {
6885 38 : continue;
6886 : }
6887 :
6888 14 : if (bio_to_abort->internal.submit_tsc > parent_io->internal.submit_tsc) {
6889 : /* Any I/O which was submitted after this abort command should be excluded. */
6890 0 : continue;
6891 : }
6892 :
6893 : /* We can't abort a request that's being pushed/pulled or executed by accel */
6894 28 : if (bdev_io_on_tailq(bio_to_abort, &channel->io_accel_exec) ||
6895 14 : bdev_io_on_tailq(bio_to_abort, &channel->io_memory_domain)) {
6896 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
6897 0 : break;
6898 : }
6899 :
6900 14 : rc = bdev_abort_io(desc, channel, bio_to_abort, bdev_abort_io_done, parent_io);
6901 14 : if (rc != 0) {
6902 1 : if (rc == -ENOMEM) {
6903 1 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_NOMEM;
6904 : } else {
6905 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
6906 : }
6907 1 : break;
6908 : }
6909 13 : matched_ios++;
6910 : }
6911 :
6912 12 : return matched_ios;
6913 : }
6914 :
6915 : static void
6916 1 : bdev_abort_retry(void *ctx)
6917 : {
6918 1 : struct spdk_bdev_io *parent_io = ctx;
6919 : uint32_t matched_ios;
6920 :
6921 1 : matched_ios = _bdev_abort(parent_io);
6922 :
6923 1 : if (matched_ios == 0) {
6924 0 : if (parent_io->internal.status == SPDK_BDEV_IO_STATUS_NOMEM) {
6925 0 : bdev_queue_io_wait_with_cb(parent_io, bdev_abort_retry);
6926 : } else {
6927 : /* For retry, the case that no target I/O was found is success
6928 : * because it means target I/Os completed in the meantime.
6929 : */
6930 0 : bdev_io_complete(parent_io);
6931 : }
6932 0 : return;
6933 : }
6934 :
6935 : /* Use split_outstanding to manage the progress of aborting I/Os. */
6936 1 : parent_io->u.bdev.split_outstanding = matched_ios;
6937 : }
6938 :
6939 : static void
6940 11 : bdev_abort(struct spdk_bdev_io *parent_io)
6941 : {
6942 : uint32_t matched_ios;
6943 :
6944 11 : matched_ios = _bdev_abort(parent_io);
6945 :
6946 11 : if (matched_ios == 0) {
6947 2 : if (parent_io->internal.status == SPDK_BDEV_IO_STATUS_NOMEM) {
6948 1 : bdev_queue_io_wait_with_cb(parent_io, bdev_abort_retry);
6949 : } else {
6950 : /* The case the no target I/O was found is failure. */
6951 1 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
6952 1 : bdev_io_complete(parent_io);
6953 : }
6954 2 : return;
6955 : }
6956 :
6957 : /* Use split_outstanding to manage the progress of aborting I/Os. */
6958 9 : parent_io->u.bdev.split_outstanding = matched_ios;
6959 : }
6960 :
6961 : int
6962 9 : spdk_bdev_abort(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6963 : void *bio_cb_arg,
6964 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6965 : {
6966 9 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6967 9 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6968 : struct spdk_bdev_io *bdev_io;
6969 :
6970 9 : if (bio_cb_arg == NULL) {
6971 0 : return -EINVAL;
6972 : }
6973 :
6974 9 : if (!spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_ABORT)) {
6975 1 : return -ENOTSUP;
6976 : }
6977 :
6978 8 : bdev_io = bdev_channel_get_io(channel);
6979 8 : if (bdev_io == NULL) {
6980 0 : return -ENOMEM;
6981 : }
6982 :
6983 8 : bdev_io->internal.ch = channel;
6984 8 : bdev_io->internal.desc = desc;
6985 8 : bdev_io->internal.submit_tsc = spdk_get_ticks();
6986 8 : bdev_io->type = SPDK_BDEV_IO_TYPE_ABORT;
6987 8 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6988 :
6989 8 : bdev_io->u.bdev.abort.bio_cb_arg = bio_cb_arg;
6990 :
6991 : /* Parent abort request is not submitted directly, but to manage its execution,
6992 : * add it to the submitted list here.
6993 : */
6994 8 : TAILQ_INSERT_TAIL(&channel->io_submitted, bdev_io, internal.ch_link);
6995 :
6996 8 : bdev_abort(bdev_io);
6997 :
6998 8 : return 0;
6999 : }
7000 :
7001 : int
7002 4 : spdk_bdev_queue_io_wait(struct spdk_bdev *bdev, struct spdk_io_channel *ch,
7003 : struct spdk_bdev_io_wait_entry *entry)
7004 : {
7005 4 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
7006 4 : struct spdk_bdev_mgmt_channel *mgmt_ch = channel->shared_resource->mgmt_ch;
7007 :
7008 4 : if (bdev != entry->bdev) {
7009 0 : SPDK_ERRLOG("bdevs do not match\n");
7010 0 : return -EINVAL;
7011 : }
7012 :
7013 4 : if (mgmt_ch->per_thread_cache_count > 0) {
7014 0 : SPDK_ERRLOG("Cannot queue io_wait if spdk_bdev_io available in per-thread cache\n");
7015 0 : return -EINVAL;
7016 : }
7017 :
7018 4 : TAILQ_INSERT_TAIL(&mgmt_ch->io_wait_queue, entry, link);
7019 4 : return 0;
7020 : }
7021 :
7022 : static inline void
7023 597 : bdev_io_update_io_stat(struct spdk_bdev_io *bdev_io, uint64_t tsc_diff)
7024 : {
7025 597 : enum spdk_bdev_io_status io_status = bdev_io->internal.status;
7026 597 : struct spdk_bdev_io_stat *io_stat = bdev_io->internal.ch->stat;
7027 597 : uint64_t num_blocks = bdev_io->u.bdev.num_blocks;
7028 597 : uint32_t blocklen = bdev_io->bdev->blocklen;
7029 :
7030 597 : if (spdk_likely(io_status == SPDK_BDEV_IO_STATUS_SUCCESS)) {
7031 509 : switch (bdev_io->type) {
7032 326 : case SPDK_BDEV_IO_TYPE_READ:
7033 326 : io_stat->bytes_read += num_blocks * blocklen;
7034 326 : io_stat->num_read_ops++;
7035 326 : io_stat->read_latency_ticks += tsc_diff;
7036 326 : if (io_stat->max_read_latency_ticks < tsc_diff) {
7037 6 : io_stat->max_read_latency_ticks = tsc_diff;
7038 : }
7039 326 : if (io_stat->min_read_latency_ticks > tsc_diff) {
7040 41 : io_stat->min_read_latency_ticks = tsc_diff;
7041 : }
7042 326 : break;
7043 75 : case SPDK_BDEV_IO_TYPE_WRITE:
7044 75 : io_stat->bytes_written += num_blocks * blocklen;
7045 75 : io_stat->num_write_ops++;
7046 75 : io_stat->write_latency_ticks += tsc_diff;
7047 75 : if (io_stat->max_write_latency_ticks < tsc_diff) {
7048 4 : io_stat->max_write_latency_ticks = tsc_diff;
7049 : }
7050 75 : if (io_stat->min_write_latency_ticks > tsc_diff) {
7051 25 : io_stat->min_write_latency_ticks = tsc_diff;
7052 : }
7053 75 : break;
7054 20 : case SPDK_BDEV_IO_TYPE_UNMAP:
7055 20 : io_stat->bytes_unmapped += num_blocks * blocklen;
7056 20 : io_stat->num_unmap_ops++;
7057 20 : io_stat->unmap_latency_ticks += tsc_diff;
7058 20 : if (io_stat->max_unmap_latency_ticks < tsc_diff) {
7059 0 : io_stat->max_unmap_latency_ticks = tsc_diff;
7060 : }
7061 20 : if (io_stat->min_unmap_latency_ticks > tsc_diff) {
7062 3 : io_stat->min_unmap_latency_ticks = tsc_diff;
7063 : }
7064 20 : break;
7065 4 : case SPDK_BDEV_IO_TYPE_ZCOPY:
7066 : /* Track the data in the start phase only */
7067 4 : if (bdev_io->u.bdev.zcopy.start) {
7068 2 : if (bdev_io->u.bdev.zcopy.populate) {
7069 1 : io_stat->bytes_read += num_blocks * blocklen;
7070 1 : io_stat->num_read_ops++;
7071 1 : io_stat->read_latency_ticks += tsc_diff;
7072 1 : if (io_stat->max_read_latency_ticks < tsc_diff) {
7073 0 : io_stat->max_read_latency_ticks = tsc_diff;
7074 : }
7075 1 : if (io_stat->min_read_latency_ticks > tsc_diff) {
7076 1 : io_stat->min_read_latency_ticks = tsc_diff;
7077 : }
7078 : } else {
7079 1 : io_stat->bytes_written += num_blocks * blocklen;
7080 1 : io_stat->num_write_ops++;
7081 1 : io_stat->write_latency_ticks += tsc_diff;
7082 1 : if (io_stat->max_write_latency_ticks < tsc_diff) {
7083 0 : io_stat->max_write_latency_ticks = tsc_diff;
7084 : }
7085 1 : if (io_stat->min_write_latency_ticks > tsc_diff) {
7086 1 : io_stat->min_write_latency_ticks = tsc_diff;
7087 : }
7088 : }
7089 : }
7090 4 : break;
7091 21 : case SPDK_BDEV_IO_TYPE_COPY:
7092 21 : io_stat->bytes_copied += num_blocks * blocklen;
7093 21 : io_stat->num_copy_ops++;
7094 21 : bdev_io->internal.ch->stat->copy_latency_ticks += tsc_diff;
7095 21 : if (io_stat->max_copy_latency_ticks < tsc_diff) {
7096 0 : io_stat->max_copy_latency_ticks = tsc_diff;
7097 : }
7098 21 : if (io_stat->min_copy_latency_ticks > tsc_diff) {
7099 4 : io_stat->min_copy_latency_ticks = tsc_diff;
7100 : }
7101 21 : break;
7102 63 : default:
7103 63 : break;
7104 : }
7105 88 : } else if (io_status <= SPDK_BDEV_IO_STATUS_FAILED && io_status >= SPDK_MIN_BDEV_IO_STATUS) {
7106 88 : io_stat = bdev_io->bdev->internal.stat;
7107 88 : assert(io_stat->io_error != NULL);
7108 :
7109 88 : spdk_spin_lock(&bdev_io->bdev->internal.spinlock);
7110 88 : io_stat->io_error->error_status[-io_status - 1]++;
7111 88 : spdk_spin_unlock(&bdev_io->bdev->internal.spinlock);
7112 : }
7113 :
7114 : #ifdef SPDK_CONFIG_VTUNE
7115 : uint64_t now_tsc = spdk_get_ticks();
7116 : if (now_tsc > (bdev_io->internal.ch->start_tsc + bdev_io->internal.ch->interval_tsc)) {
7117 : uint64_t data[5];
7118 : struct spdk_bdev_io_stat *prev_stat = bdev_io->internal.ch->prev_stat;
7119 :
7120 : data[0] = io_stat->num_read_ops - prev_stat->num_read_ops;
7121 : data[1] = io_stat->bytes_read - prev_stat->bytes_read;
7122 : data[2] = io_stat->num_write_ops - prev_stat->num_write_ops;
7123 : data[3] = io_stat->bytes_written - prev_stat->bytes_written;
7124 : data[4] = bdev_io->bdev->fn_table->get_spin_time ?
7125 : bdev_io->bdev->fn_table->get_spin_time(spdk_bdev_io_get_io_channel(bdev_io)) : 0;
7126 :
7127 : __itt_metadata_add(g_bdev_mgr.domain, __itt_null, bdev_io->internal.ch->handle,
7128 : __itt_metadata_u64, 5, data);
7129 :
7130 : memcpy(prev_stat, io_stat, sizeof(struct spdk_bdev_io_stat));
7131 : bdev_io->internal.ch->start_tsc = now_tsc;
7132 : }
7133 : #endif
7134 597 : }
7135 :
7136 : static inline void
7137 597 : _bdev_io_complete(void *ctx)
7138 : {
7139 597 : struct spdk_bdev_io *bdev_io = ctx;
7140 :
7141 597 : if (spdk_unlikely(bdev_io->internal.accel_sequence != NULL)) {
7142 0 : assert(bdev_io->internal.status != SPDK_BDEV_IO_STATUS_SUCCESS);
7143 0 : spdk_accel_sequence_abort(bdev_io->internal.accel_sequence);
7144 : }
7145 :
7146 597 : assert(bdev_io->internal.cb != NULL);
7147 597 : assert(spdk_get_thread() == spdk_bdev_io_get_thread(bdev_io));
7148 :
7149 597 : bdev_io->internal.cb(bdev_io, bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS,
7150 : bdev_io->internal.caller_ctx);
7151 597 : }
7152 :
7153 : static inline void
7154 605 : bdev_io_complete(void *ctx)
7155 : {
7156 605 : struct spdk_bdev_io *bdev_io = ctx;
7157 605 : struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
7158 : uint64_t tsc, tsc_diff;
7159 :
7160 605 : if (spdk_unlikely(bdev_io->internal.in_submit_request)) {
7161 : /*
7162 : * Defer completion to avoid potential infinite recursion if the
7163 : * user's completion callback issues a new I/O.
7164 : */
7165 8 : spdk_thread_send_msg(spdk_bdev_io_get_thread(bdev_io),
7166 : bdev_io_complete, bdev_io);
7167 8 : return;
7168 : }
7169 :
7170 597 : tsc = spdk_get_ticks();
7171 597 : tsc_diff = tsc - bdev_io->internal.submit_tsc;
7172 597 : spdk_trace_record_tsc(tsc, TRACE_BDEV_IO_DONE, 0, 0, (uintptr_t)bdev_io,
7173 : bdev_io->internal.caller_ctx);
7174 :
7175 597 : TAILQ_REMOVE(&bdev_ch->io_submitted, bdev_io, internal.ch_link);
7176 :
7177 597 : if (bdev_io->internal.ch->histogram) {
7178 4 : spdk_histogram_data_tally(bdev_io->internal.ch->histogram, tsc_diff);
7179 : }
7180 :
7181 597 : bdev_io_update_io_stat(bdev_io, tsc_diff);
7182 597 : _bdev_io_complete(bdev_io);
7183 : }
7184 :
7185 : /* The difference between this function and bdev_io_complete() is that this should be called to
7186 : * complete IOs that haven't been submitted via bdev_io_submit(), as they weren't added onto the
7187 : * io_submitted list and don't have submit_tsc updated.
7188 : */
7189 : static inline void
7190 0 : bdev_io_complete_unsubmitted(struct spdk_bdev_io *bdev_io)
7191 : {
7192 : /* Since the IO hasn't been submitted it's bound to be failed */
7193 0 : assert(bdev_io->internal.status != SPDK_BDEV_IO_STATUS_SUCCESS);
7194 :
7195 : /* At this point we don't know if the IO is completed from submission context or not, but,
7196 : * since this is an error path, we can always do an spdk_thread_send_msg(). */
7197 0 : spdk_thread_send_msg(spdk_bdev_io_get_thread(bdev_io),
7198 : _bdev_io_complete, bdev_io);
7199 0 : }
7200 :
7201 : static void bdev_destroy_cb(void *io_device);
7202 :
7203 : static void
7204 15 : bdev_reset_complete(struct spdk_bdev *bdev, void *_ctx, int status)
7205 : {
7206 15 : struct spdk_bdev_io *bdev_io = _ctx;
7207 :
7208 15 : if (bdev_io->u.reset.ch_ref != NULL) {
7209 15 : spdk_put_io_channel(bdev_io->u.reset.ch_ref);
7210 15 : bdev_io->u.reset.ch_ref = NULL;
7211 : }
7212 :
7213 15 : bdev_io_complete(bdev_io);
7214 :
7215 15 : if (bdev->internal.status == SPDK_BDEV_STATUS_REMOVING &&
7216 1 : TAILQ_EMPTY(&bdev->internal.open_descs)) {
7217 1 : spdk_io_device_unregister(__bdev_to_io_dev(bdev), bdev_destroy_cb);
7218 : }
7219 15 : }
7220 :
7221 : static void
7222 18 : bdev_unfreeze_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
7223 : struct spdk_io_channel *_ch, void *_ctx)
7224 : {
7225 18 : struct spdk_bdev_io *bdev_io = _ctx;
7226 18 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
7227 : struct spdk_bdev_io *queued_reset;
7228 :
7229 18 : ch->flags &= ~BDEV_CH_RESET_IN_PROGRESS;
7230 18 : while (!TAILQ_EMPTY(&ch->queued_resets)) {
7231 0 : queued_reset = TAILQ_FIRST(&ch->queued_resets);
7232 0 : TAILQ_REMOVE(&ch->queued_resets, queued_reset, internal.link);
7233 0 : spdk_bdev_io_complete(queued_reset, bdev_io->internal.status);
7234 : }
7235 :
7236 18 : spdk_bdev_for_each_channel_continue(i, 0);
7237 18 : }
7238 :
7239 : static void
7240 0 : bdev_io_complete_sequence_cb(void *ctx, int status)
7241 : {
7242 0 : struct spdk_bdev_io *bdev_io = ctx;
7243 :
7244 : /* u.bdev.accel_sequence should have already been cleared at this point */
7245 0 : assert(bdev_io->u.bdev.accel_sequence == NULL);
7246 0 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS);
7247 0 : bdev_io->internal.accel_sequence = NULL;
7248 :
7249 0 : if (spdk_unlikely(status != 0)) {
7250 0 : SPDK_ERRLOG("Failed to execute accel sequence, status=%d\n", status);
7251 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
7252 : }
7253 :
7254 0 : bdev_io_complete(bdev_io);
7255 0 : }
7256 :
7257 : void
7258 591 : spdk_bdev_io_complete(struct spdk_bdev_io *bdev_io, enum spdk_bdev_io_status status)
7259 : {
7260 591 : struct spdk_bdev *bdev = bdev_io->bdev;
7261 591 : struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
7262 591 : struct spdk_bdev_shared_resource *shared_resource = bdev_ch->shared_resource;
7263 :
7264 591 : if (spdk_unlikely(bdev_io->internal.status != SPDK_BDEV_IO_STATUS_PENDING)) {
7265 0 : SPDK_ERRLOG("Unexpected completion on IO from %s module, status was %s\n",
7266 : spdk_bdev_get_module_name(bdev),
7267 : bdev_io_status_get_string(bdev_io->internal.status));
7268 0 : assert(false);
7269 : }
7270 591 : bdev_io->internal.status = status;
7271 :
7272 591 : if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_RESET)) {
7273 16 : bool unlock_channels = false;
7274 :
7275 16 : if (status == SPDK_BDEV_IO_STATUS_NOMEM) {
7276 0 : SPDK_ERRLOG("NOMEM returned for reset\n");
7277 : }
7278 16 : spdk_spin_lock(&bdev->internal.spinlock);
7279 16 : if (bdev_io == bdev->internal.reset_in_progress) {
7280 15 : bdev->internal.reset_in_progress = NULL;
7281 15 : unlock_channels = true;
7282 : }
7283 16 : spdk_spin_unlock(&bdev->internal.spinlock);
7284 :
7285 16 : if (unlock_channels) {
7286 15 : spdk_bdev_for_each_channel(bdev, bdev_unfreeze_channel, bdev_io,
7287 : bdev_reset_complete);
7288 15 : return;
7289 : }
7290 : } else {
7291 575 : bdev_io_decrement_outstanding(bdev_ch, shared_resource);
7292 575 : if (spdk_likely(status == SPDK_BDEV_IO_STATUS_SUCCESS)) {
7293 484 : if (bdev_io_needs_sequence_exec(bdev_io->internal.desc, bdev_io)) {
7294 0 : bdev_io_exec_sequence(bdev_io, bdev_io_complete_sequence_cb);
7295 0 : return;
7296 484 : } else if (spdk_unlikely(bdev_io->internal.orig_iovcnt != 0 &&
7297 : !bdev_io_use_accel_sequence(bdev_io))) {
7298 28 : _bdev_io_push_bounce_data_buffer(bdev_io,
7299 : _bdev_io_complete_push_bounce_done);
7300 : /* bdev IO will be completed in the callback */
7301 28 : return;
7302 : }
7303 : }
7304 :
7305 547 : if (spdk_unlikely(_bdev_io_handle_no_mem(bdev_io, BDEV_IO_RETRY_STATE_SUBMIT))) {
7306 5 : return;
7307 : }
7308 : }
7309 :
7310 543 : bdev_io_complete(bdev_io);
7311 : }
7312 :
7313 : void
7314 0 : spdk_bdev_io_complete_scsi_status(struct spdk_bdev_io *bdev_io, enum spdk_scsi_status sc,
7315 : enum spdk_scsi_sense sk, uint8_t asc, uint8_t ascq)
7316 : {
7317 : enum spdk_bdev_io_status status;
7318 :
7319 0 : if (sc == SPDK_SCSI_STATUS_GOOD) {
7320 0 : status = SPDK_BDEV_IO_STATUS_SUCCESS;
7321 : } else {
7322 0 : status = SPDK_BDEV_IO_STATUS_SCSI_ERROR;
7323 0 : bdev_io->internal.error.scsi.sc = sc;
7324 0 : bdev_io->internal.error.scsi.sk = sk;
7325 0 : bdev_io->internal.error.scsi.asc = asc;
7326 0 : bdev_io->internal.error.scsi.ascq = ascq;
7327 : }
7328 :
7329 0 : spdk_bdev_io_complete(bdev_io, status);
7330 0 : }
7331 :
7332 : void
7333 0 : spdk_bdev_io_get_scsi_status(const struct spdk_bdev_io *bdev_io,
7334 : int *sc, int *sk, int *asc, int *ascq)
7335 : {
7336 0 : assert(sc != NULL);
7337 0 : assert(sk != NULL);
7338 0 : assert(asc != NULL);
7339 0 : assert(ascq != NULL);
7340 :
7341 0 : switch (bdev_io->internal.status) {
7342 0 : case SPDK_BDEV_IO_STATUS_SUCCESS:
7343 0 : *sc = SPDK_SCSI_STATUS_GOOD;
7344 0 : *sk = SPDK_SCSI_SENSE_NO_SENSE;
7345 0 : *asc = SPDK_SCSI_ASC_NO_ADDITIONAL_SENSE;
7346 0 : *ascq = SPDK_SCSI_ASCQ_CAUSE_NOT_REPORTABLE;
7347 0 : break;
7348 0 : case SPDK_BDEV_IO_STATUS_NVME_ERROR:
7349 0 : spdk_scsi_nvme_translate(bdev_io, sc, sk, asc, ascq);
7350 0 : break;
7351 0 : case SPDK_BDEV_IO_STATUS_MISCOMPARE:
7352 0 : *sc = SPDK_SCSI_STATUS_CHECK_CONDITION;
7353 0 : *sk = SPDK_SCSI_SENSE_MISCOMPARE;
7354 0 : *asc = SPDK_SCSI_ASC_MISCOMPARE_DURING_VERIFY_OPERATION;
7355 0 : *ascq = bdev_io->internal.error.scsi.ascq;
7356 0 : break;
7357 0 : case SPDK_BDEV_IO_STATUS_SCSI_ERROR:
7358 0 : *sc = bdev_io->internal.error.scsi.sc;
7359 0 : *sk = bdev_io->internal.error.scsi.sk;
7360 0 : *asc = bdev_io->internal.error.scsi.asc;
7361 0 : *ascq = bdev_io->internal.error.scsi.ascq;
7362 0 : break;
7363 0 : default:
7364 0 : *sc = SPDK_SCSI_STATUS_CHECK_CONDITION;
7365 0 : *sk = SPDK_SCSI_SENSE_ABORTED_COMMAND;
7366 0 : *asc = SPDK_SCSI_ASC_NO_ADDITIONAL_SENSE;
7367 0 : *ascq = SPDK_SCSI_ASCQ_CAUSE_NOT_REPORTABLE;
7368 0 : break;
7369 : }
7370 0 : }
7371 :
7372 : void
7373 0 : spdk_bdev_io_complete_aio_status(struct spdk_bdev_io *bdev_io, int aio_result)
7374 : {
7375 : enum spdk_bdev_io_status status;
7376 :
7377 0 : if (aio_result == 0) {
7378 0 : status = SPDK_BDEV_IO_STATUS_SUCCESS;
7379 : } else {
7380 0 : status = SPDK_BDEV_IO_STATUS_AIO_ERROR;
7381 : }
7382 :
7383 0 : bdev_io->internal.error.aio_result = aio_result;
7384 :
7385 0 : spdk_bdev_io_complete(bdev_io, status);
7386 0 : }
7387 :
7388 : void
7389 0 : spdk_bdev_io_get_aio_status(const struct spdk_bdev_io *bdev_io, int *aio_result)
7390 : {
7391 0 : assert(aio_result != NULL);
7392 :
7393 0 : if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_AIO_ERROR) {
7394 0 : *aio_result = bdev_io->internal.error.aio_result;
7395 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS) {
7396 0 : *aio_result = 0;
7397 : } else {
7398 0 : *aio_result = -EIO;
7399 : }
7400 0 : }
7401 :
7402 : void
7403 0 : spdk_bdev_io_complete_nvme_status(struct spdk_bdev_io *bdev_io, uint32_t cdw0, int sct, int sc)
7404 : {
7405 : enum spdk_bdev_io_status status;
7406 :
7407 0 : if (spdk_likely(sct == SPDK_NVME_SCT_GENERIC && sc == SPDK_NVME_SC_SUCCESS)) {
7408 0 : status = SPDK_BDEV_IO_STATUS_SUCCESS;
7409 0 : } else if (sct == SPDK_NVME_SCT_GENERIC && sc == SPDK_NVME_SC_ABORTED_BY_REQUEST) {
7410 0 : status = SPDK_BDEV_IO_STATUS_ABORTED;
7411 : } else {
7412 0 : status = SPDK_BDEV_IO_STATUS_NVME_ERROR;
7413 : }
7414 :
7415 0 : bdev_io->internal.error.nvme.cdw0 = cdw0;
7416 0 : bdev_io->internal.error.nvme.sct = sct;
7417 0 : bdev_io->internal.error.nvme.sc = sc;
7418 :
7419 0 : spdk_bdev_io_complete(bdev_io, status);
7420 0 : }
7421 :
7422 : void
7423 0 : spdk_bdev_io_get_nvme_status(const struct spdk_bdev_io *bdev_io, uint32_t *cdw0, int *sct, int *sc)
7424 : {
7425 0 : assert(sct != NULL);
7426 0 : assert(sc != NULL);
7427 0 : assert(cdw0 != NULL);
7428 :
7429 0 : if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_ABORT)) {
7430 0 : *sct = SPDK_NVME_SCT_GENERIC;
7431 0 : *sc = SPDK_NVME_SC_SUCCESS;
7432 0 : if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS) {
7433 0 : *cdw0 = 0;
7434 : } else {
7435 0 : *cdw0 = 1U;
7436 : }
7437 0 : return;
7438 : }
7439 :
7440 0 : if (spdk_likely(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS)) {
7441 0 : *sct = SPDK_NVME_SCT_GENERIC;
7442 0 : *sc = SPDK_NVME_SC_SUCCESS;
7443 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_NVME_ERROR) {
7444 0 : *sct = bdev_io->internal.error.nvme.sct;
7445 0 : *sc = bdev_io->internal.error.nvme.sc;
7446 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_ABORTED) {
7447 0 : *sct = SPDK_NVME_SCT_GENERIC;
7448 0 : *sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
7449 : } else {
7450 0 : *sct = SPDK_NVME_SCT_GENERIC;
7451 0 : *sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
7452 : }
7453 :
7454 0 : *cdw0 = bdev_io->internal.error.nvme.cdw0;
7455 : }
7456 :
7457 : void
7458 0 : spdk_bdev_io_get_nvme_fused_status(const struct spdk_bdev_io *bdev_io, uint32_t *cdw0,
7459 : int *first_sct, int *first_sc, int *second_sct, int *second_sc)
7460 : {
7461 0 : assert(first_sct != NULL);
7462 0 : assert(first_sc != NULL);
7463 0 : assert(second_sct != NULL);
7464 0 : assert(second_sc != NULL);
7465 0 : assert(cdw0 != NULL);
7466 :
7467 0 : if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_NVME_ERROR) {
7468 0 : if (bdev_io->internal.error.nvme.sct == SPDK_NVME_SCT_MEDIA_ERROR &&
7469 0 : bdev_io->internal.error.nvme.sc == SPDK_NVME_SC_COMPARE_FAILURE) {
7470 0 : *first_sct = bdev_io->internal.error.nvme.sct;
7471 0 : *first_sc = bdev_io->internal.error.nvme.sc;
7472 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
7473 0 : *second_sc = SPDK_NVME_SC_ABORTED_FAILED_FUSED;
7474 : } else {
7475 0 : *first_sct = SPDK_NVME_SCT_GENERIC;
7476 0 : *first_sc = SPDK_NVME_SC_SUCCESS;
7477 0 : *second_sct = bdev_io->internal.error.nvme.sct;
7478 0 : *second_sc = bdev_io->internal.error.nvme.sc;
7479 : }
7480 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_ABORTED) {
7481 0 : *first_sct = SPDK_NVME_SCT_GENERIC;
7482 0 : *first_sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
7483 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
7484 0 : *second_sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
7485 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS) {
7486 0 : *first_sct = SPDK_NVME_SCT_GENERIC;
7487 0 : *first_sc = SPDK_NVME_SC_SUCCESS;
7488 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
7489 0 : *second_sc = SPDK_NVME_SC_SUCCESS;
7490 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_FIRST_FUSED_FAILED) {
7491 0 : *first_sct = SPDK_NVME_SCT_GENERIC;
7492 0 : *first_sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
7493 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
7494 0 : *second_sc = SPDK_NVME_SC_ABORTED_FAILED_FUSED;
7495 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_MISCOMPARE) {
7496 0 : *first_sct = SPDK_NVME_SCT_MEDIA_ERROR;
7497 0 : *first_sc = SPDK_NVME_SC_COMPARE_FAILURE;
7498 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
7499 0 : *second_sc = SPDK_NVME_SC_ABORTED_FAILED_FUSED;
7500 : } else {
7501 0 : *first_sct = SPDK_NVME_SCT_GENERIC;
7502 0 : *first_sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
7503 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
7504 0 : *second_sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
7505 : }
7506 :
7507 0 : *cdw0 = bdev_io->internal.error.nvme.cdw0;
7508 0 : }
7509 :
7510 : struct spdk_thread *
7511 651 : spdk_bdev_io_get_thread(struct spdk_bdev_io *bdev_io)
7512 : {
7513 651 : return spdk_io_channel_get_thread(bdev_io->internal.ch->channel);
7514 : }
7515 :
7516 : struct spdk_io_channel *
7517 78 : spdk_bdev_io_get_io_channel(struct spdk_bdev_io *bdev_io)
7518 : {
7519 78 : return bdev_io->internal.ch->channel;
7520 : }
7521 :
7522 : static int
7523 125 : bdev_register(struct spdk_bdev *bdev)
7524 : {
7525 : char *bdev_name;
7526 125 : char uuid[SPDK_UUID_STRING_LEN];
7527 125 : struct spdk_iobuf_opts iobuf_opts;
7528 : int ret, i;
7529 :
7530 125 : assert(bdev->module != NULL);
7531 :
7532 125 : if (!bdev->name) {
7533 0 : SPDK_ERRLOG("Bdev name is NULL\n");
7534 0 : return -EINVAL;
7535 : }
7536 :
7537 125 : if (!strlen(bdev->name)) {
7538 0 : SPDK_ERRLOG("Bdev name must not be an empty string\n");
7539 0 : return -EINVAL;
7540 : }
7541 :
7542 2750 : for (i = 0; i < SPDK_BDEV_NUM_IO_TYPES; ++i) {
7543 2625 : if (bdev->fn_table->accel_sequence_supported == NULL) {
7544 2625 : continue;
7545 : }
7546 0 : if (!bdev->fn_table->accel_sequence_supported(bdev->ctxt,
7547 : (enum spdk_bdev_io_type)i)) {
7548 0 : continue;
7549 : }
7550 :
7551 0 : if (spdk_bdev_is_md_separate(bdev)) {
7552 0 : SPDK_ERRLOG("Separate metadata is currently unsupported for bdevs with "
7553 : "accel sequence support\n");
7554 0 : return -EINVAL;
7555 : }
7556 : }
7557 :
7558 : /* Users often register their own I/O devices using the bdev name. In
7559 : * order to avoid conflicts, prepend bdev_. */
7560 125 : bdev_name = spdk_sprintf_alloc("bdev_%s", bdev->name);
7561 125 : if (!bdev_name) {
7562 0 : SPDK_ERRLOG("Unable to allocate memory for internal bdev name.\n");
7563 0 : return -ENOMEM;
7564 : }
7565 :
7566 125 : bdev->internal.stat = bdev_alloc_io_stat(true);
7567 125 : if (!bdev->internal.stat) {
7568 0 : SPDK_ERRLOG("Unable to allocate I/O statistics structure.\n");
7569 0 : free(bdev_name);
7570 0 : return -ENOMEM;
7571 : }
7572 :
7573 125 : bdev->internal.status = SPDK_BDEV_STATUS_READY;
7574 125 : bdev->internal.measured_queue_depth = UINT64_MAX;
7575 125 : bdev->internal.claim_type = SPDK_BDEV_CLAIM_NONE;
7576 125 : memset(&bdev->internal.claim, 0, sizeof(bdev->internal.claim));
7577 125 : bdev->internal.qd_poller = NULL;
7578 125 : bdev->internal.qos = NULL;
7579 :
7580 125 : TAILQ_INIT(&bdev->internal.open_descs);
7581 125 : TAILQ_INIT(&bdev->internal.locked_ranges);
7582 125 : TAILQ_INIT(&bdev->internal.pending_locked_ranges);
7583 125 : TAILQ_INIT(&bdev->aliases);
7584 :
7585 125 : ret = bdev_name_add(&bdev->internal.bdev_name, bdev, bdev->name);
7586 125 : if (ret != 0) {
7587 1 : bdev_free_io_stat(bdev->internal.stat);
7588 1 : free(bdev_name);
7589 1 : return ret;
7590 : }
7591 :
7592 : /* UUID may be specified by the user or defined by bdev itself.
7593 : * Otherwise it will be generated here, so this field will never be empty. */
7594 124 : if (spdk_uuid_is_null(&bdev->uuid)) {
7595 42 : spdk_uuid_generate(&bdev->uuid);
7596 : }
7597 :
7598 : /* Add the UUID alias only if it's different than the name */
7599 124 : spdk_uuid_fmt_lower(uuid, sizeof(uuid), &bdev->uuid);
7600 124 : if (strcmp(bdev->name, uuid) != 0) {
7601 123 : ret = spdk_bdev_alias_add(bdev, uuid);
7602 123 : if (ret != 0) {
7603 1 : SPDK_ERRLOG("Unable to add uuid:%s alias for bdev %s\n", uuid, bdev->name);
7604 1 : bdev_name_del(&bdev->internal.bdev_name);
7605 1 : bdev_free_io_stat(bdev->internal.stat);
7606 1 : free(bdev_name);
7607 1 : return ret;
7608 : }
7609 : }
7610 :
7611 123 : spdk_iobuf_get_opts(&iobuf_opts);
7612 123 : if (spdk_bdev_get_buf_align(bdev) > 1) {
7613 0 : bdev->max_rw_size = spdk_min(bdev->max_rw_size ? bdev->max_rw_size : UINT32_MAX,
7614 : iobuf_opts.large_bufsize / bdev->blocklen);
7615 : }
7616 :
7617 : /* If the user didn't specify a write unit size, set it to one. */
7618 123 : if (bdev->write_unit_size == 0) {
7619 119 : bdev->write_unit_size = 1;
7620 : }
7621 :
7622 : /* Set ACWU value to the write unit size if bdev module did not set it (does not support it natively) */
7623 123 : if (bdev->acwu == 0) {
7624 119 : bdev->acwu = bdev->write_unit_size;
7625 : }
7626 :
7627 123 : if (bdev->phys_blocklen == 0) {
7628 119 : bdev->phys_blocklen = spdk_bdev_get_data_block_size(bdev);
7629 : }
7630 :
7631 123 : if (!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COPY)) {
7632 0 : bdev->max_copy = bdev_get_max_write(bdev, iobuf_opts.large_bufsize);
7633 : }
7634 :
7635 123 : if (!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE_ZEROES)) {
7636 0 : bdev->max_write_zeroes = bdev_get_max_write(bdev, ZERO_BUFFER_SIZE);
7637 : }
7638 :
7639 123 : bdev->internal.reset_in_progress = NULL;
7640 123 : bdev->internal.qd_poll_in_progress = false;
7641 123 : bdev->internal.period = 0;
7642 123 : bdev->internal.new_period = 0;
7643 :
7644 123 : spdk_io_device_register(__bdev_to_io_dev(bdev),
7645 : bdev_channel_create, bdev_channel_destroy,
7646 : sizeof(struct spdk_bdev_channel),
7647 : bdev_name);
7648 :
7649 123 : free(bdev_name);
7650 :
7651 123 : spdk_spin_init(&bdev->internal.spinlock);
7652 :
7653 123 : SPDK_DEBUGLOG(bdev, "Inserting bdev %s into list\n", bdev->name);
7654 123 : TAILQ_INSERT_TAIL(&g_bdev_mgr.bdevs, bdev, internal.link);
7655 :
7656 123 : return 0;
7657 : }
7658 :
7659 : static void
7660 124 : bdev_destroy_cb(void *io_device)
7661 : {
7662 : int rc;
7663 : struct spdk_bdev *bdev;
7664 : spdk_bdev_unregister_cb cb_fn;
7665 : void *cb_arg;
7666 :
7667 124 : bdev = __bdev_from_io_dev(io_device);
7668 :
7669 124 : if (bdev->internal.unregister_td != spdk_get_thread()) {
7670 1 : spdk_thread_send_msg(bdev->internal.unregister_td, bdev_destroy_cb, io_device);
7671 1 : return;
7672 : }
7673 :
7674 123 : cb_fn = bdev->internal.unregister_cb;
7675 123 : cb_arg = bdev->internal.unregister_ctx;
7676 :
7677 123 : spdk_spin_destroy(&bdev->internal.spinlock);
7678 123 : free(bdev->internal.qos);
7679 123 : bdev_free_io_stat(bdev->internal.stat);
7680 :
7681 123 : rc = bdev->fn_table->destruct(bdev->ctxt);
7682 123 : if (rc < 0) {
7683 0 : SPDK_ERRLOG("destruct failed\n");
7684 : }
7685 123 : if (rc <= 0 && cb_fn != NULL) {
7686 10 : cb_fn(cb_arg, rc);
7687 : }
7688 : }
7689 :
7690 : void
7691 2 : spdk_bdev_destruct_done(struct spdk_bdev *bdev, int bdeverrno)
7692 : {
7693 2 : if (bdev->internal.unregister_cb != NULL) {
7694 0 : bdev->internal.unregister_cb(bdev->internal.unregister_ctx, bdeverrno);
7695 : }
7696 2 : }
7697 :
7698 : static void
7699 19 : _remove_notify(void *arg)
7700 : {
7701 19 : struct spdk_bdev_desc *desc = arg;
7702 :
7703 19 : _event_notify(desc, SPDK_BDEV_EVENT_REMOVE);
7704 19 : }
7705 :
7706 : /* returns: 0 - bdev removed and ready to be destructed.
7707 : * -EBUSY - bdev can't be destructed yet. */
7708 : static int
7709 138 : bdev_unregister_unsafe(struct spdk_bdev *bdev)
7710 : {
7711 : struct spdk_bdev_desc *desc, *tmp;
7712 138 : int rc = 0;
7713 138 : char uuid[SPDK_UUID_STRING_LEN];
7714 :
7715 138 : assert(spdk_spin_held(&g_bdev_mgr.spinlock));
7716 138 : assert(spdk_spin_held(&bdev->internal.spinlock));
7717 :
7718 : /* Notify each descriptor about hotremoval */
7719 157 : TAILQ_FOREACH_SAFE(desc, &bdev->internal.open_descs, link, tmp) {
7720 19 : rc = -EBUSY;
7721 : /*
7722 : * Defer invocation of the event_cb to a separate message that will
7723 : * run later on its thread. This ensures this context unwinds and
7724 : * we don't recursively unregister this bdev again if the event_cb
7725 : * immediately closes its descriptor.
7726 : */
7727 19 : event_notify(desc, _remove_notify);
7728 : }
7729 :
7730 : /* If there are no descriptors, proceed removing the bdev */
7731 138 : if (rc == 0) {
7732 123 : TAILQ_REMOVE(&g_bdev_mgr.bdevs, bdev, internal.link);
7733 123 : SPDK_DEBUGLOG(bdev, "Removing bdev %s from list done\n", bdev->name);
7734 :
7735 : /* Delete the name and the UUID alias */
7736 123 : spdk_uuid_fmt_lower(uuid, sizeof(uuid), &bdev->uuid);
7737 123 : bdev_name_del_unsafe(&bdev->internal.bdev_name);
7738 123 : bdev_alias_del(bdev, uuid, bdev_name_del_unsafe);
7739 :
7740 123 : spdk_notify_send("bdev_unregister", spdk_bdev_get_name(bdev));
7741 :
7742 123 : if (bdev->internal.reset_in_progress != NULL) {
7743 : /* If reset is in progress, let the completion callback for reset
7744 : * unregister the bdev.
7745 : */
7746 1 : rc = -EBUSY;
7747 : }
7748 : }
7749 :
7750 138 : return rc;
7751 : }
7752 :
7753 : static void
7754 4 : bdev_unregister_abort_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
7755 : struct spdk_io_channel *io_ch, void *_ctx)
7756 : {
7757 4 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
7758 :
7759 4 : bdev_channel_abort_queued_ios(bdev_ch);
7760 4 : spdk_bdev_for_each_channel_continue(i, 0);
7761 4 : }
7762 :
7763 : static void
7764 123 : bdev_unregister(struct spdk_bdev *bdev, void *_ctx, int status)
7765 : {
7766 : int rc;
7767 :
7768 123 : spdk_spin_lock(&g_bdev_mgr.spinlock);
7769 123 : spdk_spin_lock(&bdev->internal.spinlock);
7770 : /*
7771 : * Set the status to REMOVING after completing to abort channels. Otherwise,
7772 : * the last spdk_bdev_close() may call spdk_io_device_unregister() while
7773 : * spdk_bdev_for_each_channel() is executed and spdk_io_device_unregister()
7774 : * may fail.
7775 : */
7776 123 : bdev->internal.status = SPDK_BDEV_STATUS_REMOVING;
7777 123 : rc = bdev_unregister_unsafe(bdev);
7778 123 : spdk_spin_unlock(&bdev->internal.spinlock);
7779 123 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
7780 :
7781 123 : if (rc == 0) {
7782 107 : spdk_io_device_unregister(__bdev_to_io_dev(bdev), bdev_destroy_cb);
7783 : }
7784 123 : }
7785 :
7786 : void
7787 130 : spdk_bdev_unregister(struct spdk_bdev *bdev, spdk_bdev_unregister_cb cb_fn, void *cb_arg)
7788 : {
7789 : struct spdk_thread *thread;
7790 :
7791 130 : SPDK_DEBUGLOG(bdev, "Removing bdev %s from list\n", bdev->name);
7792 :
7793 130 : thread = spdk_get_thread();
7794 130 : if (!thread) {
7795 : /* The user called this from a non-SPDK thread. */
7796 0 : if (cb_fn != NULL) {
7797 0 : cb_fn(cb_arg, -ENOTSUP);
7798 : }
7799 0 : return;
7800 : }
7801 :
7802 130 : spdk_spin_lock(&g_bdev_mgr.spinlock);
7803 130 : if (bdev->internal.status == SPDK_BDEV_STATUS_UNREGISTERING ||
7804 130 : bdev->internal.status == SPDK_BDEV_STATUS_REMOVING) {
7805 7 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
7806 7 : if (cb_fn) {
7807 0 : cb_fn(cb_arg, -EBUSY);
7808 : }
7809 7 : return;
7810 : }
7811 :
7812 123 : spdk_spin_lock(&bdev->internal.spinlock);
7813 123 : bdev->internal.status = SPDK_BDEV_STATUS_UNREGISTERING;
7814 123 : bdev->internal.unregister_cb = cb_fn;
7815 123 : bdev->internal.unregister_ctx = cb_arg;
7816 123 : bdev->internal.unregister_td = thread;
7817 123 : spdk_spin_unlock(&bdev->internal.spinlock);
7818 123 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
7819 :
7820 123 : spdk_bdev_set_qd_sampling_period(bdev, 0);
7821 :
7822 123 : spdk_bdev_for_each_channel(bdev, bdev_unregister_abort_channel, bdev,
7823 : bdev_unregister);
7824 : }
7825 :
7826 : int
7827 4 : spdk_bdev_unregister_by_name(const char *bdev_name, struct spdk_bdev_module *module,
7828 : spdk_bdev_unregister_cb cb_fn, void *cb_arg)
7829 : {
7830 4 : struct spdk_bdev_desc *desc;
7831 : struct spdk_bdev *bdev;
7832 : int rc;
7833 :
7834 4 : rc = spdk_bdev_open_ext(bdev_name, false, _tmp_bdev_event_cb, NULL, &desc);
7835 4 : if (rc != 0) {
7836 1 : SPDK_ERRLOG("Failed to open bdev with name: %s\n", bdev_name);
7837 1 : return rc;
7838 : }
7839 :
7840 3 : bdev = spdk_bdev_desc_get_bdev(desc);
7841 :
7842 3 : if (bdev->module != module) {
7843 1 : spdk_bdev_close(desc);
7844 1 : SPDK_ERRLOG("Bdev %s was not registered by the specified module.\n",
7845 : bdev_name);
7846 1 : return -ENODEV;
7847 : }
7848 :
7849 2 : spdk_bdev_unregister(bdev, cb_fn, cb_arg);
7850 :
7851 2 : spdk_bdev_close(desc);
7852 :
7853 2 : return 0;
7854 : }
7855 :
7856 : static int
7857 253 : bdev_start_qos(struct spdk_bdev *bdev)
7858 : {
7859 : struct set_qos_limit_ctx *ctx;
7860 :
7861 : /* Enable QoS */
7862 253 : if (bdev->internal.qos && bdev->internal.qos->thread == NULL) {
7863 2 : ctx = calloc(1, sizeof(*ctx));
7864 2 : if (ctx == NULL) {
7865 0 : SPDK_ERRLOG("Failed to allocate memory for QoS context\n");
7866 0 : return -ENOMEM;
7867 : }
7868 2 : ctx->bdev = bdev;
7869 2 : spdk_bdev_for_each_channel(bdev, bdev_enable_qos_msg, ctx, bdev_enable_qos_done);
7870 : }
7871 :
7872 253 : return 0;
7873 : }
7874 :
7875 : static void
7876 24 : log_already_claimed(enum spdk_log_level level, const int line, const char *func, const char *detail,
7877 : struct spdk_bdev *bdev)
7878 : {
7879 : enum spdk_bdev_claim_type type;
7880 : const char *typename, *modname;
7881 : extern struct spdk_log_flag SPDK_LOG_bdev;
7882 :
7883 24 : assert(spdk_spin_held(&bdev->internal.spinlock));
7884 :
7885 24 : if (level >= SPDK_LOG_INFO && !SPDK_LOG_bdev.enabled) {
7886 0 : return;
7887 : }
7888 :
7889 24 : type = bdev->internal.claim_type;
7890 24 : typename = spdk_bdev_claim_get_name(type);
7891 :
7892 24 : if (type == SPDK_BDEV_CLAIM_EXCL_WRITE) {
7893 6 : modname = bdev->internal.claim.v1.module->name;
7894 6 : spdk_log(level, __FILE__, line, func, "bdev %s %s: type %s by module %s\n",
7895 : bdev->name, detail, typename, modname);
7896 6 : return;
7897 : }
7898 :
7899 18 : if (claim_type_is_v2(type)) {
7900 : struct spdk_bdev_module_claim *claim;
7901 :
7902 36 : TAILQ_FOREACH(claim, &bdev->internal.claim.v2.claims, link) {
7903 18 : modname = claim->module->name;
7904 18 : spdk_log(level, __FILE__, line, func, "bdev %s %s: type %s by module %s\n",
7905 : bdev->name, detail, typename, modname);
7906 : }
7907 18 : return;
7908 : }
7909 :
7910 0 : assert(false);
7911 : }
7912 :
7913 : static int
7914 262 : bdev_open(struct spdk_bdev *bdev, bool write, struct spdk_bdev_desc *desc)
7915 : {
7916 : struct spdk_thread *thread;
7917 262 : int rc = 0;
7918 :
7919 262 : thread = spdk_get_thread();
7920 262 : if (!thread) {
7921 0 : SPDK_ERRLOG("Cannot open bdev from non-SPDK thread.\n");
7922 0 : return -ENOTSUP;
7923 : }
7924 :
7925 262 : SPDK_DEBUGLOG(bdev, "Opening descriptor %p for bdev %s on thread %p\n", desc, bdev->name,
7926 : spdk_get_thread());
7927 :
7928 262 : desc->bdev = bdev;
7929 262 : desc->thread = thread;
7930 262 : desc->write = write;
7931 :
7932 262 : spdk_spin_lock(&bdev->internal.spinlock);
7933 262 : if (bdev->internal.status == SPDK_BDEV_STATUS_UNREGISTERING ||
7934 262 : bdev->internal.status == SPDK_BDEV_STATUS_REMOVING) {
7935 3 : spdk_spin_unlock(&bdev->internal.spinlock);
7936 3 : return -ENODEV;
7937 : }
7938 :
7939 259 : if (write && bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE) {
7940 6 : LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
7941 6 : spdk_spin_unlock(&bdev->internal.spinlock);
7942 6 : return -EPERM;
7943 : }
7944 :
7945 253 : rc = bdev_start_qos(bdev);
7946 253 : if (rc != 0) {
7947 0 : SPDK_ERRLOG("Failed to start QoS on bdev %s\n", bdev->name);
7948 0 : spdk_spin_unlock(&bdev->internal.spinlock);
7949 0 : return rc;
7950 : }
7951 :
7952 253 : TAILQ_INSERT_TAIL(&bdev->internal.open_descs, desc, link);
7953 :
7954 253 : spdk_spin_unlock(&bdev->internal.spinlock);
7955 :
7956 253 : return 0;
7957 : }
7958 :
7959 : static int
7960 262 : bdev_desc_alloc(struct spdk_bdev *bdev, spdk_bdev_event_cb_t event_cb, void *event_ctx,
7961 : struct spdk_bdev_desc **_desc)
7962 : {
7963 : struct spdk_bdev_desc *desc;
7964 : unsigned int i;
7965 :
7966 262 : desc = calloc(1, sizeof(*desc));
7967 262 : if (desc == NULL) {
7968 0 : SPDK_ERRLOG("Failed to allocate memory for bdev descriptor\n");
7969 0 : return -ENOMEM;
7970 : }
7971 :
7972 262 : TAILQ_INIT(&desc->pending_media_events);
7973 262 : TAILQ_INIT(&desc->free_media_events);
7974 :
7975 262 : desc->memory_domains_supported = spdk_bdev_get_memory_domains(bdev, NULL, 0) > 0;
7976 262 : desc->callback.event_fn = event_cb;
7977 262 : desc->callback.ctx = event_ctx;
7978 262 : spdk_spin_init(&desc->spinlock);
7979 :
7980 262 : if (bdev->media_events) {
7981 0 : desc->media_events_buffer = calloc(MEDIA_EVENT_POOL_SIZE,
7982 : sizeof(*desc->media_events_buffer));
7983 0 : if (desc->media_events_buffer == NULL) {
7984 0 : SPDK_ERRLOG("Failed to initialize media event pool\n");
7985 0 : bdev_desc_free(desc);
7986 0 : return -ENOMEM;
7987 : }
7988 :
7989 0 : for (i = 0; i < MEDIA_EVENT_POOL_SIZE; ++i) {
7990 0 : TAILQ_INSERT_TAIL(&desc->free_media_events,
7991 : &desc->media_events_buffer[i], tailq);
7992 : }
7993 : }
7994 :
7995 262 : if (bdev->fn_table->accel_sequence_supported != NULL) {
7996 0 : for (i = 0; i < SPDK_BDEV_NUM_IO_TYPES; ++i) {
7997 0 : desc->accel_sequence_supported[i] =
7998 0 : bdev->fn_table->accel_sequence_supported(bdev->ctxt,
7999 : (enum spdk_bdev_io_type)i);
8000 : }
8001 : }
8002 :
8003 262 : *_desc = desc;
8004 :
8005 262 : return 0;
8006 : }
8007 :
8008 : static int
8009 127 : bdev_open_ext(const char *bdev_name, bool write, spdk_bdev_event_cb_t event_cb,
8010 : void *event_ctx, struct spdk_bdev_desc **_desc)
8011 : {
8012 127 : struct spdk_bdev_desc *desc;
8013 : struct spdk_bdev *bdev;
8014 : int rc;
8015 :
8016 127 : bdev = bdev_get_by_name(bdev_name);
8017 :
8018 127 : if (bdev == NULL) {
8019 1 : SPDK_NOTICELOG("Currently unable to find bdev with name: %s\n", bdev_name);
8020 1 : return -ENODEV;
8021 : }
8022 :
8023 126 : rc = bdev_desc_alloc(bdev, event_cb, event_ctx, &desc);
8024 126 : if (rc != 0) {
8025 0 : return rc;
8026 : }
8027 :
8028 126 : rc = bdev_open(bdev, write, desc);
8029 126 : if (rc != 0) {
8030 7 : bdev_desc_free(desc);
8031 7 : desc = NULL;
8032 : }
8033 :
8034 126 : *_desc = desc;
8035 :
8036 126 : return rc;
8037 : }
8038 :
8039 : int
8040 129 : spdk_bdev_open_ext(const char *bdev_name, bool write, spdk_bdev_event_cb_t event_cb,
8041 : void *event_ctx, struct spdk_bdev_desc **_desc)
8042 : {
8043 : int rc;
8044 :
8045 129 : if (event_cb == NULL) {
8046 2 : SPDK_ERRLOG("Missing event callback function\n");
8047 2 : return -EINVAL;
8048 : }
8049 :
8050 127 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8051 127 : rc = bdev_open_ext(bdev_name, write, event_cb, event_ctx, _desc);
8052 127 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8053 :
8054 127 : return rc;
8055 : }
8056 :
8057 : struct spdk_bdev_open_async_ctx {
8058 : char *bdev_name;
8059 : spdk_bdev_event_cb_t event_cb;
8060 : void *event_ctx;
8061 : bool write;
8062 : int rc;
8063 : spdk_bdev_open_async_cb_t cb_fn;
8064 : void *cb_arg;
8065 : struct spdk_bdev_desc *desc;
8066 : struct spdk_bdev_open_async_opts opts;
8067 : uint64_t start_ticks;
8068 : struct spdk_thread *orig_thread;
8069 : struct spdk_poller *poller;
8070 : TAILQ_ENTRY(spdk_bdev_open_async_ctx) tailq;
8071 : };
8072 :
8073 : static void
8074 0 : bdev_open_async_done(void *arg)
8075 : {
8076 0 : struct spdk_bdev_open_async_ctx *ctx = arg;
8077 :
8078 0 : ctx->cb_fn(ctx->desc, ctx->rc, ctx->cb_arg);
8079 :
8080 0 : free(ctx->bdev_name);
8081 0 : free(ctx);
8082 0 : }
8083 :
8084 : static void
8085 0 : bdev_open_async_cancel(void *arg)
8086 : {
8087 0 : struct spdk_bdev_open_async_ctx *ctx = arg;
8088 :
8089 0 : assert(ctx->rc == -ESHUTDOWN);
8090 :
8091 0 : spdk_poller_unregister(&ctx->poller);
8092 :
8093 0 : bdev_open_async_done(ctx);
8094 0 : }
8095 :
8096 : /* This is called when the bdev library finishes at shutdown. */
8097 : static void
8098 64 : bdev_open_async_fini(void)
8099 : {
8100 : struct spdk_bdev_open_async_ctx *ctx, *tmp_ctx;
8101 :
8102 64 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8103 64 : TAILQ_FOREACH_SAFE(ctx, &g_bdev_mgr.async_bdev_opens, tailq, tmp_ctx) {
8104 0 : TAILQ_REMOVE(&g_bdev_mgr.async_bdev_opens, ctx, tailq);
8105 : /*
8106 : * We have to move to ctx->orig_thread to unregister ctx->poller.
8107 : * However, there is a chance that ctx->poller is executed before
8108 : * message is executed, which could result in bdev_open_async_done()
8109 : * being called twice. To avoid such race condition, set ctx->rc to
8110 : * -ESHUTDOWN.
8111 : */
8112 0 : ctx->rc = -ESHUTDOWN;
8113 0 : spdk_thread_send_msg(ctx->orig_thread, bdev_open_async_cancel, ctx);
8114 : }
8115 64 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8116 64 : }
8117 :
8118 : static int bdev_open_async(void *arg);
8119 :
8120 : static void
8121 0 : _bdev_open_async(struct spdk_bdev_open_async_ctx *ctx)
8122 : {
8123 : uint64_t timeout_ticks;
8124 :
8125 0 : if (ctx->rc == -ESHUTDOWN) {
8126 : /* This context is being canceled. Do nothing. */
8127 0 : return;
8128 : }
8129 :
8130 0 : ctx->rc = bdev_open_ext(ctx->bdev_name, ctx->write, ctx->event_cb, ctx->event_ctx,
8131 : &ctx->desc);
8132 0 : if (ctx->rc == 0 || ctx->opts.timeout_ms == 0) {
8133 0 : goto exit;
8134 : }
8135 :
8136 0 : timeout_ticks = ctx->start_ticks + ctx->opts.timeout_ms * spdk_get_ticks_hz() / 1000ull;
8137 0 : if (spdk_get_ticks() >= timeout_ticks) {
8138 0 : SPDK_ERRLOG("Timed out while waiting for bdev '%s' to appear\n", ctx->bdev_name);
8139 0 : ctx->rc = -ETIMEDOUT;
8140 0 : goto exit;
8141 : }
8142 :
8143 0 : return;
8144 :
8145 0 : exit:
8146 0 : spdk_poller_unregister(&ctx->poller);
8147 0 : TAILQ_REMOVE(&g_bdev_mgr.async_bdev_opens, ctx, tailq);
8148 :
8149 : /* Completion callback is processed after stack unwinding. */
8150 0 : spdk_thread_send_msg(ctx->orig_thread, bdev_open_async_done, ctx);
8151 : }
8152 :
8153 : static int
8154 0 : bdev_open_async(void *arg)
8155 : {
8156 0 : struct spdk_bdev_open_async_ctx *ctx = arg;
8157 :
8158 0 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8159 :
8160 0 : _bdev_open_async(ctx);
8161 :
8162 0 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8163 :
8164 0 : return SPDK_POLLER_BUSY;
8165 : }
8166 :
8167 : static void
8168 0 : bdev_open_async_opts_copy(struct spdk_bdev_open_async_opts *opts,
8169 : struct spdk_bdev_open_async_opts *opts_src,
8170 : size_t size)
8171 : {
8172 0 : assert(opts);
8173 0 : assert(opts_src);
8174 :
8175 0 : opts->size = size;
8176 :
8177 : #define SET_FIELD(field) \
8178 : if (offsetof(struct spdk_bdev_open_async_opts, field) + sizeof(opts->field) <= size) { \
8179 : opts->field = opts_src->field; \
8180 : } \
8181 :
8182 0 : SET_FIELD(timeout_ms);
8183 :
8184 : /* Do not remove this statement, you should always update this statement when you adding a new field,
8185 : * and do not forget to add the SET_FIELD statement for your added field. */
8186 : SPDK_STATIC_ASSERT(sizeof(struct spdk_bdev_open_async_opts) == 16, "Incorrect size");
8187 :
8188 : #undef SET_FIELD
8189 0 : }
8190 :
8191 : static void
8192 0 : bdev_open_async_opts_get_default(struct spdk_bdev_open_async_opts *opts, size_t size)
8193 : {
8194 0 : assert(opts);
8195 :
8196 0 : opts->size = size;
8197 :
8198 : #define SET_FIELD(field, value) \
8199 : if (offsetof(struct spdk_bdev_open_async_opts, field) + sizeof(opts->field) <= size) { \
8200 : opts->field = value; \
8201 : } \
8202 :
8203 0 : SET_FIELD(timeout_ms, 0);
8204 :
8205 : #undef SET_FIELD
8206 0 : }
8207 :
8208 : int
8209 0 : spdk_bdev_open_async(const char *bdev_name, bool write, spdk_bdev_event_cb_t event_cb,
8210 : void *event_ctx, struct spdk_bdev_open_async_opts *opts,
8211 : spdk_bdev_open_async_cb_t open_cb, void *open_cb_arg)
8212 : {
8213 : struct spdk_bdev_open_async_ctx *ctx;
8214 :
8215 0 : if (event_cb == NULL) {
8216 0 : SPDK_ERRLOG("Missing event callback function\n");
8217 0 : return -EINVAL;
8218 : }
8219 :
8220 0 : if (open_cb == NULL) {
8221 0 : SPDK_ERRLOG("Missing open callback function\n");
8222 0 : return -EINVAL;
8223 : }
8224 :
8225 0 : if (opts != NULL && opts->size == 0) {
8226 0 : SPDK_ERRLOG("size in the options structure should not be zero\n");
8227 0 : return -EINVAL;
8228 : }
8229 :
8230 0 : ctx = calloc(1, sizeof(*ctx));
8231 0 : if (ctx == NULL) {
8232 0 : SPDK_ERRLOG("Failed to allocate open context\n");
8233 0 : return -ENOMEM;
8234 : }
8235 :
8236 0 : ctx->bdev_name = strdup(bdev_name);
8237 0 : if (ctx->bdev_name == NULL) {
8238 0 : SPDK_ERRLOG("Failed to duplicate bdev_name\n");
8239 0 : free(ctx);
8240 0 : return -ENOMEM;
8241 : }
8242 :
8243 0 : ctx->poller = SPDK_POLLER_REGISTER(bdev_open_async, ctx, 100 * 1000);
8244 0 : if (ctx->poller == NULL) {
8245 0 : SPDK_ERRLOG("Failed to register bdev_open_async poller\n");
8246 0 : free(ctx->bdev_name);
8247 0 : free(ctx);
8248 0 : return -ENOMEM;
8249 : }
8250 :
8251 0 : ctx->cb_fn = open_cb;
8252 0 : ctx->cb_arg = open_cb_arg;
8253 0 : ctx->write = write;
8254 0 : ctx->event_cb = event_cb;
8255 0 : ctx->event_ctx = event_ctx;
8256 0 : ctx->orig_thread = spdk_get_thread();
8257 0 : ctx->start_ticks = spdk_get_ticks();
8258 :
8259 0 : bdev_open_async_opts_get_default(&ctx->opts, sizeof(ctx->opts));
8260 0 : if (opts != NULL) {
8261 0 : bdev_open_async_opts_copy(&ctx->opts, opts, opts->size);
8262 : }
8263 :
8264 0 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8265 :
8266 0 : TAILQ_INSERT_TAIL(&g_bdev_mgr.async_bdev_opens, ctx, tailq);
8267 0 : _bdev_open_async(ctx);
8268 :
8269 0 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8270 :
8271 0 : return 0;
8272 : }
8273 :
8274 : static void
8275 253 : bdev_close(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc)
8276 : {
8277 : int rc;
8278 :
8279 253 : spdk_spin_lock(&bdev->internal.spinlock);
8280 253 : spdk_spin_lock(&desc->spinlock);
8281 :
8282 253 : TAILQ_REMOVE(&bdev->internal.open_descs, desc, link);
8283 :
8284 253 : desc->closed = true;
8285 :
8286 253 : if (desc->claim != NULL) {
8287 16 : bdev_desc_release_claims(desc);
8288 : }
8289 :
8290 253 : if (0 == desc->refs) {
8291 242 : spdk_spin_unlock(&desc->spinlock);
8292 242 : bdev_desc_free(desc);
8293 : } else {
8294 11 : spdk_spin_unlock(&desc->spinlock);
8295 : }
8296 :
8297 : /* If no more descriptors, kill QoS channel */
8298 253 : if (bdev->internal.qos && TAILQ_EMPTY(&bdev->internal.open_descs)) {
8299 7 : SPDK_DEBUGLOG(bdev, "Closed last descriptor for bdev %s on thread %p. Stopping QoS.\n",
8300 : bdev->name, spdk_get_thread());
8301 :
8302 7 : if (bdev_qos_destroy(bdev)) {
8303 : /* There isn't anything we can do to recover here. Just let the
8304 : * old QoS poller keep running. The QoS handling won't change
8305 : * cores when the user allocates a new channel, but it won't break. */
8306 0 : SPDK_ERRLOG("Unable to shut down QoS poller. It will continue running on the current thread.\n");
8307 : }
8308 : }
8309 :
8310 253 : if (bdev->internal.status == SPDK_BDEV_STATUS_REMOVING && TAILQ_EMPTY(&bdev->internal.open_descs)) {
8311 15 : rc = bdev_unregister_unsafe(bdev);
8312 15 : spdk_spin_unlock(&bdev->internal.spinlock);
8313 :
8314 15 : if (rc == 0) {
8315 15 : spdk_io_device_unregister(__bdev_to_io_dev(bdev), bdev_destroy_cb);
8316 : }
8317 : } else {
8318 238 : spdk_spin_unlock(&bdev->internal.spinlock);
8319 : }
8320 253 : }
8321 :
8322 : void
8323 119 : spdk_bdev_close(struct spdk_bdev_desc *desc)
8324 : {
8325 119 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
8326 :
8327 119 : SPDK_DEBUGLOG(bdev, "Closing descriptor %p for bdev %s on thread %p\n", desc, bdev->name,
8328 : spdk_get_thread());
8329 :
8330 119 : assert(desc->thread == spdk_get_thread());
8331 :
8332 119 : spdk_poller_unregister(&desc->io_timeout_poller);
8333 :
8334 119 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8335 :
8336 119 : bdev_close(bdev, desc);
8337 :
8338 119 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8339 119 : }
8340 :
8341 : static void
8342 123 : bdev_register_finished(void *arg)
8343 : {
8344 123 : struct spdk_bdev_desc *desc = arg;
8345 123 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
8346 :
8347 123 : spdk_notify_send("bdev_register", spdk_bdev_get_name(bdev));
8348 :
8349 123 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8350 :
8351 123 : bdev_close(bdev, desc);
8352 :
8353 123 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8354 123 : }
8355 :
8356 : int
8357 126 : spdk_bdev_register(struct spdk_bdev *bdev)
8358 : {
8359 126 : struct spdk_bdev_desc *desc;
8360 126 : struct spdk_thread *thread = spdk_get_thread();
8361 : int rc;
8362 :
8363 126 : if (spdk_unlikely(!spdk_thread_is_app_thread(NULL))) {
8364 1 : SPDK_ERRLOG("Cannot examine bdev %s on thread %p (%s)\n", bdev->name, thread,
8365 : thread ? spdk_thread_get_name(thread) : "null");
8366 1 : return -EINVAL;
8367 : }
8368 :
8369 125 : rc = bdev_register(bdev);
8370 125 : if (rc != 0) {
8371 2 : return rc;
8372 : }
8373 :
8374 : /* A descriptor is opened to prevent bdev deletion during examination */
8375 123 : rc = bdev_desc_alloc(bdev, _tmp_bdev_event_cb, NULL, &desc);
8376 123 : if (rc != 0) {
8377 0 : spdk_bdev_unregister(bdev, NULL, NULL);
8378 0 : return rc;
8379 : }
8380 :
8381 123 : rc = bdev_open(bdev, false, desc);
8382 123 : if (rc != 0) {
8383 0 : bdev_desc_free(desc);
8384 0 : spdk_bdev_unregister(bdev, NULL, NULL);
8385 0 : return rc;
8386 : }
8387 :
8388 : /* Examine configuration before initializing I/O */
8389 123 : bdev_examine(bdev);
8390 :
8391 123 : rc = spdk_bdev_wait_for_examine(bdev_register_finished, desc);
8392 123 : if (rc != 0) {
8393 0 : bdev_close(bdev, desc);
8394 0 : spdk_bdev_unregister(bdev, NULL, NULL);
8395 : }
8396 :
8397 123 : return rc;
8398 : }
8399 :
8400 : int
8401 26 : spdk_bdev_module_claim_bdev(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc,
8402 : struct spdk_bdev_module *module)
8403 : {
8404 26 : spdk_spin_lock(&bdev->internal.spinlock);
8405 :
8406 26 : if (bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE) {
8407 6 : LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
8408 6 : spdk_spin_unlock(&bdev->internal.spinlock);
8409 6 : return -EPERM;
8410 : }
8411 :
8412 20 : if (desc && !desc->write) {
8413 5 : desc->write = true;
8414 : }
8415 :
8416 20 : bdev->internal.claim_type = SPDK_BDEV_CLAIM_EXCL_WRITE;
8417 20 : bdev->internal.claim.v1.module = module;
8418 :
8419 20 : spdk_spin_unlock(&bdev->internal.spinlock);
8420 20 : return 0;
8421 : }
8422 :
8423 : void
8424 8 : spdk_bdev_module_release_bdev(struct spdk_bdev *bdev)
8425 : {
8426 8 : spdk_spin_lock(&bdev->internal.spinlock);
8427 :
8428 8 : assert(bdev->internal.claim.v1.module != NULL);
8429 8 : assert(bdev->internal.claim_type == SPDK_BDEV_CLAIM_EXCL_WRITE);
8430 8 : bdev->internal.claim_type = SPDK_BDEV_CLAIM_NONE;
8431 8 : bdev->internal.claim.v1.module = NULL;
8432 :
8433 8 : spdk_spin_unlock(&bdev->internal.spinlock);
8434 8 : }
8435 :
8436 : /*
8437 : * Start claims v2
8438 : */
8439 :
8440 : const char *
8441 24 : spdk_bdev_claim_get_name(enum spdk_bdev_claim_type type)
8442 : {
8443 24 : switch (type) {
8444 0 : case SPDK_BDEV_CLAIM_NONE:
8445 0 : return "not_claimed";
8446 6 : case SPDK_BDEV_CLAIM_EXCL_WRITE:
8447 6 : return "exclusive_write";
8448 7 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE:
8449 7 : return "read_many_write_one";
8450 5 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_NONE:
8451 5 : return "read_many_write_none";
8452 6 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
8453 6 : return "read_many_write_many";
8454 0 : default:
8455 0 : break;
8456 : }
8457 0 : return "invalid_claim";
8458 : }
8459 :
8460 : static bool
8461 96 : claim_type_is_v2(enum spdk_bdev_claim_type type)
8462 : {
8463 96 : switch (type) {
8464 96 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE:
8465 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_NONE:
8466 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
8467 96 : return true;
8468 0 : default:
8469 0 : break;
8470 : }
8471 0 : return false;
8472 : }
8473 :
8474 : /* Returns true if taking a claim with desc->write == false should make the descriptor writable. */
8475 : static bool
8476 13 : claim_type_promotes_to_write(enum spdk_bdev_claim_type type)
8477 : {
8478 13 : switch (type) {
8479 5 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE:
8480 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
8481 5 : return true;
8482 8 : default:
8483 8 : break;
8484 : }
8485 8 : return false;
8486 : }
8487 :
8488 : void
8489 44 : spdk_bdev_claim_opts_init(struct spdk_bdev_claim_opts *opts, size_t size)
8490 : {
8491 44 : if (opts == NULL) {
8492 0 : SPDK_ERRLOG("opts should not be NULL\n");
8493 0 : assert(opts != NULL);
8494 0 : return;
8495 : }
8496 44 : if (size == 0) {
8497 0 : SPDK_ERRLOG("size should not be zero\n");
8498 0 : assert(size != 0);
8499 0 : return;
8500 : }
8501 :
8502 44 : memset(opts, 0, size);
8503 44 : opts->opts_size = size;
8504 :
8505 : #define FIELD_OK(field) \
8506 : offsetof(struct spdk_bdev_claim_opts, field) + sizeof(opts->field) <= size
8507 :
8508 : #define SET_FIELD(field, value) \
8509 : if (FIELD_OK(field)) { \
8510 : opts->field = value; \
8511 : } \
8512 :
8513 44 : SET_FIELD(shared_claim_key, 0);
8514 :
8515 : #undef FIELD_OK
8516 : #undef SET_FIELD
8517 : }
8518 :
8519 : static int
8520 22 : claim_opts_copy(struct spdk_bdev_claim_opts *src, struct spdk_bdev_claim_opts *dst)
8521 : {
8522 22 : if (src->opts_size == 0) {
8523 0 : SPDK_ERRLOG("size should not be zero\n");
8524 0 : return -1;
8525 : }
8526 :
8527 22 : memset(dst, 0, sizeof(*dst));
8528 22 : dst->opts_size = src->opts_size;
8529 :
8530 : #define FIELD_OK(field) \
8531 : offsetof(struct spdk_bdev_claim_opts, field) + sizeof(src->field) <= src->opts_size
8532 :
8533 : #define SET_FIELD(field) \
8534 : if (FIELD_OK(field)) { \
8535 : dst->field = src->field; \
8536 : } \
8537 :
8538 22 : if (FIELD_OK(name)) {
8539 22 : snprintf(dst->name, sizeof(dst->name), "%s", src->name);
8540 : }
8541 :
8542 22 : SET_FIELD(shared_claim_key);
8543 :
8544 : /* You should not remove this statement, but need to update the assert statement
8545 : * if you add a new field, and also add a corresponding SET_FIELD statement */
8546 : SPDK_STATIC_ASSERT(sizeof(struct spdk_bdev_claim_opts) == 48, "Incorrect size");
8547 :
8548 : #undef FIELD_OK
8549 : #undef SET_FIELD
8550 22 : return 0;
8551 : }
8552 :
8553 : /* Returns 0 if a read-write-once claim can be taken. */
8554 : static int
8555 9 : claim_verify_rwo(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
8556 : struct spdk_bdev_claim_opts *opts, struct spdk_bdev_module *module)
8557 : {
8558 9 : struct spdk_bdev *bdev = desc->bdev;
8559 : struct spdk_bdev_desc *open_desc;
8560 :
8561 9 : assert(spdk_spin_held(&bdev->internal.spinlock));
8562 9 : assert(type == SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE);
8563 :
8564 9 : if (opts->shared_claim_key != 0) {
8565 1 : SPDK_ERRLOG("%s: key option not supported with read-write-once claims\n",
8566 : bdev->name);
8567 1 : return -EINVAL;
8568 : }
8569 8 : if (bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE) {
8570 1 : LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
8571 1 : return -EPERM;
8572 : }
8573 7 : if (desc->claim != NULL) {
8574 0 : SPDK_NOTICELOG("%s: descriptor already claimed bdev with module %s\n",
8575 : bdev->name, desc->claim->module->name);
8576 0 : return -EPERM;
8577 : }
8578 14 : TAILQ_FOREACH(open_desc, &bdev->internal.open_descs, link) {
8579 9 : if (desc != open_desc && open_desc->write) {
8580 2 : SPDK_NOTICELOG("%s: Cannot obtain read-write-once claim while "
8581 : "another descriptor is open for writing\n",
8582 : bdev->name);
8583 2 : return -EPERM;
8584 : }
8585 : }
8586 :
8587 5 : return 0;
8588 : }
8589 :
8590 : /* Returns 0 if a read-only-many claim can be taken. */
8591 : static int
8592 12 : claim_verify_rom(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
8593 : struct spdk_bdev_claim_opts *opts, struct spdk_bdev_module *module)
8594 : {
8595 12 : struct spdk_bdev *bdev = desc->bdev;
8596 : struct spdk_bdev_desc *open_desc;
8597 :
8598 12 : assert(spdk_spin_held(&bdev->internal.spinlock));
8599 12 : assert(type == SPDK_BDEV_CLAIM_READ_MANY_WRITE_NONE);
8600 12 : assert(desc->claim == NULL);
8601 :
8602 12 : if (desc->write) {
8603 3 : SPDK_ERRLOG("%s: Cannot obtain read-only-many claim with writable descriptor\n",
8604 : bdev->name);
8605 3 : return -EINVAL;
8606 : }
8607 9 : if (opts->shared_claim_key != 0) {
8608 1 : SPDK_ERRLOG("%s: key option not supported with read-only-may claims\n", bdev->name);
8609 1 : return -EINVAL;
8610 : }
8611 8 : if (bdev->internal.claim_type == SPDK_BDEV_CLAIM_NONE) {
8612 15 : TAILQ_FOREACH(open_desc, &bdev->internal.open_descs, link) {
8613 9 : if (open_desc->write) {
8614 0 : SPDK_NOTICELOG("%s: Cannot obtain read-only-many claim while "
8615 : "another descriptor is open for writing\n",
8616 : bdev->name);
8617 0 : return -EPERM;
8618 : }
8619 : }
8620 : }
8621 :
8622 8 : return 0;
8623 : }
8624 :
8625 : /* Returns 0 if a read-write-many claim can be taken. */
8626 : static int
8627 8 : claim_verify_rwm(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
8628 : struct spdk_bdev_claim_opts *opts, struct spdk_bdev_module *module)
8629 : {
8630 8 : struct spdk_bdev *bdev = desc->bdev;
8631 : struct spdk_bdev_desc *open_desc;
8632 :
8633 8 : assert(spdk_spin_held(&bdev->internal.spinlock));
8634 8 : assert(type == SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED);
8635 8 : assert(desc->claim == NULL);
8636 :
8637 8 : if (opts->shared_claim_key == 0) {
8638 2 : SPDK_ERRLOG("%s: shared_claim_key option required with read-write-may claims\n",
8639 : bdev->name);
8640 2 : return -EINVAL;
8641 : }
8642 6 : switch (bdev->internal.claim_type) {
8643 4 : case SPDK_BDEV_CLAIM_NONE:
8644 7 : TAILQ_FOREACH(open_desc, &bdev->internal.open_descs, link) {
8645 5 : if (open_desc == desc) {
8646 3 : continue;
8647 : }
8648 2 : if (open_desc->write) {
8649 2 : SPDK_NOTICELOG("%s: Cannot obtain read-write-many claim while "
8650 : "another descriptor is open for writing without a "
8651 : "claim\n", bdev->name);
8652 2 : return -EPERM;
8653 : }
8654 : }
8655 2 : break;
8656 2 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
8657 2 : if (opts->shared_claim_key != bdev->internal.claim.v2.key) {
8658 1 : LOG_ALREADY_CLAIMED_ERROR("already claimed with another key", bdev);
8659 1 : return -EPERM;
8660 : }
8661 1 : break;
8662 0 : default:
8663 0 : LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
8664 0 : return -EBUSY;
8665 : }
8666 :
8667 3 : return 0;
8668 : }
8669 :
8670 : /* Updates desc and its bdev with a v2 claim. */
8671 : static int
8672 16 : claim_bdev(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
8673 : struct spdk_bdev_claim_opts *opts, struct spdk_bdev_module *module)
8674 : {
8675 16 : struct spdk_bdev *bdev = desc->bdev;
8676 : struct spdk_bdev_module_claim *claim;
8677 :
8678 16 : assert(spdk_spin_held(&bdev->internal.spinlock));
8679 16 : assert(claim_type_is_v2(type));
8680 16 : assert(desc->claim == NULL);
8681 :
8682 16 : claim = calloc(1, sizeof(*desc->claim));
8683 16 : if (claim == NULL) {
8684 0 : SPDK_ERRLOG("%s: out of memory while allocating claim\n", bdev->name);
8685 0 : return -ENOMEM;
8686 : }
8687 16 : claim->module = module;
8688 16 : claim->desc = desc;
8689 : SPDK_STATIC_ASSERT(sizeof(claim->name) == sizeof(opts->name), "sizes must match");
8690 16 : memcpy(claim->name, opts->name, sizeof(claim->name));
8691 16 : desc->claim = claim;
8692 :
8693 16 : if (bdev->internal.claim_type == SPDK_BDEV_CLAIM_NONE) {
8694 13 : bdev->internal.claim_type = type;
8695 13 : TAILQ_INIT(&bdev->internal.claim.v2.claims);
8696 13 : bdev->internal.claim.v2.key = opts->shared_claim_key;
8697 : }
8698 16 : assert(type == bdev->internal.claim_type);
8699 :
8700 16 : TAILQ_INSERT_TAIL(&bdev->internal.claim.v2.claims, claim, link);
8701 :
8702 16 : if (!desc->write && claim_type_promotes_to_write(type)) {
8703 5 : desc->write = true;
8704 : }
8705 :
8706 16 : return 0;
8707 : }
8708 :
8709 : int
8710 39 : spdk_bdev_module_claim_bdev_desc(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
8711 : struct spdk_bdev_claim_opts *_opts,
8712 : struct spdk_bdev_module *module)
8713 : {
8714 : struct spdk_bdev *bdev;
8715 39 : struct spdk_bdev_claim_opts opts;
8716 39 : int rc = 0;
8717 :
8718 39 : if (desc == NULL) {
8719 0 : SPDK_ERRLOG("descriptor must not be NULL\n");
8720 0 : return -EINVAL;
8721 : }
8722 :
8723 39 : bdev = desc->bdev;
8724 :
8725 39 : if (_opts == NULL) {
8726 17 : spdk_bdev_claim_opts_init(&opts, sizeof(opts));
8727 22 : } else if (claim_opts_copy(_opts, &opts) != 0) {
8728 0 : return -EINVAL;
8729 : }
8730 :
8731 39 : spdk_spin_lock(&bdev->internal.spinlock);
8732 :
8733 39 : if (bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE &&
8734 15 : bdev->internal.claim_type != type) {
8735 10 : LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
8736 10 : spdk_spin_unlock(&bdev->internal.spinlock);
8737 10 : return -EPERM;
8738 : }
8739 :
8740 29 : if (claim_type_is_v2(type) && desc->claim != NULL) {
8741 0 : SPDK_ERRLOG("%s: descriptor already has %s claim with name '%s'\n",
8742 : bdev->name, spdk_bdev_claim_get_name(type), desc->claim->name);
8743 0 : spdk_spin_unlock(&bdev->internal.spinlock);
8744 0 : return -EPERM;
8745 : }
8746 :
8747 29 : switch (type) {
8748 0 : case SPDK_BDEV_CLAIM_EXCL_WRITE:
8749 0 : spdk_spin_unlock(&bdev->internal.spinlock);
8750 0 : return spdk_bdev_module_claim_bdev(bdev, desc, module);
8751 9 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE:
8752 9 : rc = claim_verify_rwo(desc, type, &opts, module);
8753 9 : break;
8754 12 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_NONE:
8755 12 : rc = claim_verify_rom(desc, type, &opts, module);
8756 12 : break;
8757 8 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
8758 8 : rc = claim_verify_rwm(desc, type, &opts, module);
8759 8 : break;
8760 0 : default:
8761 0 : SPDK_ERRLOG("%s: claim type %d not supported\n", bdev->name, type);
8762 0 : rc = -ENOTSUP;
8763 : }
8764 :
8765 29 : if (rc == 0) {
8766 16 : rc = claim_bdev(desc, type, &opts, module);
8767 : }
8768 :
8769 29 : spdk_spin_unlock(&bdev->internal.spinlock);
8770 29 : return rc;
8771 : }
8772 :
8773 : static void
8774 13 : claim_reset(struct spdk_bdev *bdev)
8775 : {
8776 13 : assert(spdk_spin_held(&bdev->internal.spinlock));
8777 13 : assert(claim_type_is_v2(bdev->internal.claim_type));
8778 13 : assert(TAILQ_EMPTY(&bdev->internal.claim.v2.claims));
8779 :
8780 13 : memset(&bdev->internal.claim, 0, sizeof(bdev->internal.claim));
8781 13 : bdev->internal.claim_type = SPDK_BDEV_CLAIM_NONE;
8782 13 : }
8783 :
8784 : static void
8785 16 : bdev_desc_release_claims(struct spdk_bdev_desc *desc)
8786 : {
8787 16 : struct spdk_bdev *bdev = desc->bdev;
8788 :
8789 16 : assert(spdk_spin_held(&bdev->internal.spinlock));
8790 16 : assert(claim_type_is_v2(bdev->internal.claim_type));
8791 :
8792 16 : if (bdev->internal.examine_in_progress == 0) {
8793 16 : TAILQ_REMOVE(&bdev->internal.claim.v2.claims, desc->claim, link);
8794 16 : free(desc->claim);
8795 16 : if (TAILQ_EMPTY(&bdev->internal.claim.v2.claims)) {
8796 13 : claim_reset(bdev);
8797 : }
8798 : } else {
8799 : /* This is a dead claim that will be cleaned up when bdev_examine() is done. */
8800 0 : desc->claim->module = NULL;
8801 0 : desc->claim->desc = NULL;
8802 : }
8803 16 : desc->claim = NULL;
8804 16 : }
8805 :
8806 : /*
8807 : * End claims v2
8808 : */
8809 :
8810 : struct spdk_bdev *
8811 1117 : spdk_bdev_desc_get_bdev(struct spdk_bdev_desc *desc)
8812 : {
8813 1117 : assert(desc != NULL);
8814 1117 : return desc->bdev;
8815 : }
8816 :
8817 : int
8818 1 : spdk_for_each_bdev(void *ctx, spdk_for_each_bdev_fn fn)
8819 : {
8820 : struct spdk_bdev *bdev, *tmp;
8821 1 : struct spdk_bdev_desc *desc;
8822 1 : int rc = 0;
8823 :
8824 1 : assert(fn != NULL);
8825 :
8826 1 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8827 1 : bdev = spdk_bdev_first();
8828 9 : while (bdev != NULL) {
8829 8 : rc = bdev_desc_alloc(bdev, _tmp_bdev_event_cb, NULL, &desc);
8830 8 : if (rc != 0) {
8831 0 : break;
8832 : }
8833 8 : rc = bdev_open(bdev, false, desc);
8834 8 : if (rc != 0) {
8835 1 : bdev_desc_free(desc);
8836 1 : if (rc == -ENODEV) {
8837 : /* Ignore the error and move to the next bdev. */
8838 1 : rc = 0;
8839 1 : bdev = spdk_bdev_next(bdev);
8840 1 : continue;
8841 : }
8842 0 : break;
8843 : }
8844 7 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8845 :
8846 7 : rc = fn(ctx, bdev);
8847 :
8848 7 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8849 7 : tmp = spdk_bdev_next(bdev);
8850 7 : bdev_close(bdev, desc);
8851 7 : if (rc != 0) {
8852 0 : break;
8853 : }
8854 7 : bdev = tmp;
8855 : }
8856 1 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8857 :
8858 1 : return rc;
8859 : }
8860 :
8861 : int
8862 1 : spdk_for_each_bdev_leaf(void *ctx, spdk_for_each_bdev_fn fn)
8863 : {
8864 : struct spdk_bdev *bdev, *tmp;
8865 1 : struct spdk_bdev_desc *desc;
8866 1 : int rc = 0;
8867 :
8868 1 : assert(fn != NULL);
8869 :
8870 1 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8871 1 : bdev = spdk_bdev_first_leaf();
8872 6 : while (bdev != NULL) {
8873 5 : rc = bdev_desc_alloc(bdev, _tmp_bdev_event_cb, NULL, &desc);
8874 5 : if (rc != 0) {
8875 0 : break;
8876 : }
8877 5 : rc = bdev_open(bdev, false, desc);
8878 5 : if (rc != 0) {
8879 1 : bdev_desc_free(desc);
8880 1 : if (rc == -ENODEV) {
8881 : /* Ignore the error and move to the next bdev. */
8882 1 : rc = 0;
8883 1 : bdev = spdk_bdev_next_leaf(bdev);
8884 1 : continue;
8885 : }
8886 0 : break;
8887 : }
8888 4 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8889 :
8890 4 : rc = fn(ctx, bdev);
8891 :
8892 4 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8893 4 : tmp = spdk_bdev_next_leaf(bdev);
8894 4 : bdev_close(bdev, desc);
8895 4 : if (rc != 0) {
8896 0 : break;
8897 : }
8898 4 : bdev = tmp;
8899 : }
8900 1 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8901 :
8902 1 : return rc;
8903 : }
8904 :
8905 : void
8906 0 : spdk_bdev_io_get_iovec(struct spdk_bdev_io *bdev_io, struct iovec **iovp, int *iovcntp)
8907 : {
8908 : struct iovec *iovs;
8909 : int iovcnt;
8910 :
8911 0 : if (bdev_io == NULL) {
8912 0 : return;
8913 : }
8914 :
8915 0 : switch (bdev_io->type) {
8916 0 : case SPDK_BDEV_IO_TYPE_READ:
8917 : case SPDK_BDEV_IO_TYPE_WRITE:
8918 : case SPDK_BDEV_IO_TYPE_ZCOPY:
8919 0 : iovs = bdev_io->u.bdev.iovs;
8920 0 : iovcnt = bdev_io->u.bdev.iovcnt;
8921 0 : break;
8922 0 : default:
8923 0 : iovs = NULL;
8924 0 : iovcnt = 0;
8925 0 : break;
8926 : }
8927 :
8928 0 : if (iovp) {
8929 0 : *iovp = iovs;
8930 : }
8931 0 : if (iovcntp) {
8932 0 : *iovcntp = iovcnt;
8933 : }
8934 : }
8935 :
8936 : void *
8937 0 : spdk_bdev_io_get_md_buf(struct spdk_bdev_io *bdev_io)
8938 : {
8939 0 : if (bdev_io == NULL) {
8940 0 : return NULL;
8941 : }
8942 :
8943 0 : if (!spdk_bdev_is_md_separate(bdev_io->bdev)) {
8944 0 : return NULL;
8945 : }
8946 :
8947 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_READ ||
8948 0 : bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
8949 0 : return bdev_io->u.bdev.md_buf;
8950 : }
8951 :
8952 0 : return NULL;
8953 : }
8954 :
8955 : void *
8956 0 : spdk_bdev_io_get_cb_arg(struct spdk_bdev_io *bdev_io)
8957 : {
8958 0 : if (bdev_io == NULL) {
8959 0 : assert(false);
8960 : return NULL;
8961 : }
8962 :
8963 0 : return bdev_io->internal.caller_ctx;
8964 : }
8965 :
8966 : void
8967 7 : spdk_bdev_module_list_add(struct spdk_bdev_module *bdev_module)
8968 : {
8969 :
8970 7 : if (spdk_bdev_module_list_find(bdev_module->name)) {
8971 0 : SPDK_ERRLOG("ERROR: module '%s' already registered.\n", bdev_module->name);
8972 0 : assert(false);
8973 : }
8974 :
8975 7 : spdk_spin_init(&bdev_module->internal.spinlock);
8976 7 : TAILQ_INIT(&bdev_module->internal.quiesced_ranges);
8977 :
8978 : /*
8979 : * Modules with examine callbacks must be initialized first, so they are
8980 : * ready to handle examine callbacks from later modules that will
8981 : * register physical bdevs.
8982 : */
8983 7 : if (bdev_module->examine_config != NULL || bdev_module->examine_disk != NULL) {
8984 4 : TAILQ_INSERT_HEAD(&g_bdev_mgr.bdev_modules, bdev_module, internal.tailq);
8985 : } else {
8986 3 : TAILQ_INSERT_TAIL(&g_bdev_mgr.bdev_modules, bdev_module, internal.tailq);
8987 : }
8988 7 : }
8989 :
8990 : struct spdk_bdev_module *
8991 7 : spdk_bdev_module_list_find(const char *name)
8992 : {
8993 : struct spdk_bdev_module *bdev_module;
8994 :
8995 14 : TAILQ_FOREACH(bdev_module, &g_bdev_mgr.bdev_modules, internal.tailq) {
8996 7 : if (strcmp(name, bdev_module->name) == 0) {
8997 0 : break;
8998 : }
8999 : }
9000 :
9001 7 : return bdev_module;
9002 : }
9003 :
9004 : static int
9005 6 : bdev_write_zero_buffer(struct spdk_bdev_io *bdev_io)
9006 : {
9007 : uint64_t num_blocks;
9008 6 : void *md_buf = NULL;
9009 :
9010 6 : num_blocks = bdev_io->u.bdev.num_blocks;
9011 :
9012 6 : if (spdk_bdev_is_md_separate(bdev_io->bdev)) {
9013 4 : md_buf = (char *)g_bdev_mgr.zero_buffer +
9014 2 : spdk_bdev_get_block_size(bdev_io->bdev) * num_blocks;
9015 : }
9016 :
9017 6 : return bdev_write_blocks_with_md(bdev_io->internal.desc,
9018 6 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
9019 : g_bdev_mgr.zero_buffer, md_buf,
9020 : bdev_io->u.bdev.offset_blocks, num_blocks,
9021 : bdev_write_zero_buffer_done, bdev_io);
9022 : }
9023 :
9024 : static void
9025 6 : bdev_write_zero_buffer_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
9026 : {
9027 6 : struct spdk_bdev_io *parent_io = cb_arg;
9028 :
9029 6 : spdk_bdev_free_io(bdev_io);
9030 :
9031 6 : parent_io->internal.status = success ? SPDK_BDEV_IO_STATUS_SUCCESS : SPDK_BDEV_IO_STATUS_FAILED;
9032 6 : parent_io->internal.cb(parent_io, success, parent_io->internal.caller_ctx);
9033 6 : }
9034 :
9035 : static void
9036 10 : bdev_set_qos_limit_done(struct set_qos_limit_ctx *ctx, int status)
9037 : {
9038 10 : spdk_spin_lock(&ctx->bdev->internal.spinlock);
9039 10 : ctx->bdev->internal.qos_mod_in_progress = false;
9040 10 : spdk_spin_unlock(&ctx->bdev->internal.spinlock);
9041 :
9042 10 : if (ctx->cb_fn) {
9043 8 : ctx->cb_fn(ctx->cb_arg, status);
9044 : }
9045 10 : free(ctx);
9046 10 : }
9047 :
9048 : static void
9049 2 : bdev_disable_qos_done(void *cb_arg)
9050 : {
9051 2 : struct set_qos_limit_ctx *ctx = cb_arg;
9052 2 : struct spdk_bdev *bdev = ctx->bdev;
9053 : struct spdk_bdev_qos *qos;
9054 :
9055 2 : spdk_spin_lock(&bdev->internal.spinlock);
9056 2 : qos = bdev->internal.qos;
9057 2 : bdev->internal.qos = NULL;
9058 2 : spdk_spin_unlock(&bdev->internal.spinlock);
9059 :
9060 2 : if (qos->thread != NULL) {
9061 2 : spdk_put_io_channel(spdk_io_channel_from_ctx(qos->ch));
9062 2 : spdk_poller_unregister(&qos->poller);
9063 : }
9064 :
9065 2 : free(qos);
9066 :
9067 2 : bdev_set_qos_limit_done(ctx, 0);
9068 2 : }
9069 :
9070 : static void
9071 2 : bdev_disable_qos_msg_done(struct spdk_bdev *bdev, void *_ctx, int status)
9072 : {
9073 2 : struct set_qos_limit_ctx *ctx = _ctx;
9074 : struct spdk_thread *thread;
9075 :
9076 2 : spdk_spin_lock(&bdev->internal.spinlock);
9077 2 : thread = bdev->internal.qos->thread;
9078 2 : spdk_spin_unlock(&bdev->internal.spinlock);
9079 :
9080 2 : if (thread != NULL) {
9081 2 : spdk_thread_send_msg(thread, bdev_disable_qos_done, ctx);
9082 : } else {
9083 0 : bdev_disable_qos_done(ctx);
9084 : }
9085 2 : }
9086 :
9087 : static void
9088 4 : bdev_disable_qos_msg(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
9089 : struct spdk_io_channel *ch, void *_ctx)
9090 : {
9091 4 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(ch);
9092 : struct spdk_bdev_io *bdev_io;
9093 :
9094 4 : bdev_ch->flags &= ~BDEV_CH_QOS_ENABLED;
9095 :
9096 6 : while (!TAILQ_EMPTY(&bdev_ch->qos_queued_io)) {
9097 : /* Re-submit the queued I/O. */
9098 2 : bdev_io = TAILQ_FIRST(&bdev_ch->qos_queued_io);
9099 2 : TAILQ_REMOVE(&bdev_ch->qos_queued_io, bdev_io, internal.link);
9100 2 : _bdev_io_submit(bdev_io);
9101 : }
9102 :
9103 4 : spdk_bdev_for_each_channel_continue(i, 0);
9104 4 : }
9105 :
9106 : static void
9107 1 : bdev_update_qos_rate_limit_msg(void *cb_arg)
9108 : {
9109 1 : struct set_qos_limit_ctx *ctx = cb_arg;
9110 1 : struct spdk_bdev *bdev = ctx->bdev;
9111 :
9112 1 : spdk_spin_lock(&bdev->internal.spinlock);
9113 1 : bdev_qos_update_max_quota_per_timeslice(bdev->internal.qos);
9114 1 : spdk_spin_unlock(&bdev->internal.spinlock);
9115 :
9116 1 : bdev_set_qos_limit_done(ctx, 0);
9117 1 : }
9118 :
9119 : static void
9120 9 : bdev_enable_qos_msg(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
9121 : struct spdk_io_channel *ch, void *_ctx)
9122 : {
9123 9 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(ch);
9124 :
9125 9 : spdk_spin_lock(&bdev->internal.spinlock);
9126 9 : bdev_enable_qos(bdev, bdev_ch);
9127 9 : spdk_spin_unlock(&bdev->internal.spinlock);
9128 9 : spdk_bdev_for_each_channel_continue(i, 0);
9129 9 : }
9130 :
9131 : static void
9132 6 : bdev_enable_qos_done(struct spdk_bdev *bdev, void *_ctx, int status)
9133 : {
9134 6 : struct set_qos_limit_ctx *ctx = _ctx;
9135 :
9136 6 : bdev_set_qos_limit_done(ctx, status);
9137 6 : }
9138 :
9139 : static void
9140 7 : bdev_set_qos_rate_limits(struct spdk_bdev *bdev, uint64_t *limits)
9141 : {
9142 : int i;
9143 :
9144 7 : assert(bdev->internal.qos != NULL);
9145 :
9146 35 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
9147 28 : if (limits[i] != SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
9148 28 : bdev->internal.qos->rate_limits[i].limit = limits[i];
9149 :
9150 28 : if (limits[i] == 0) {
9151 19 : bdev->internal.qos->rate_limits[i].limit =
9152 : SPDK_BDEV_QOS_LIMIT_NOT_DEFINED;
9153 : }
9154 : }
9155 : }
9156 7 : }
9157 :
9158 : void
9159 9 : spdk_bdev_set_qos_rate_limits(struct spdk_bdev *bdev, uint64_t *limits,
9160 : void (*cb_fn)(void *cb_arg, int status), void *cb_arg)
9161 : {
9162 : struct set_qos_limit_ctx *ctx;
9163 : uint32_t limit_set_complement;
9164 : uint64_t min_limit_per_sec;
9165 : int i;
9166 9 : bool disable_rate_limit = true;
9167 :
9168 45 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
9169 36 : if (limits[i] == SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
9170 0 : continue;
9171 : }
9172 :
9173 36 : if (limits[i] > 0) {
9174 10 : disable_rate_limit = false;
9175 : }
9176 :
9177 36 : if (bdev_qos_is_iops_rate_limit(i) == true) {
9178 9 : min_limit_per_sec = SPDK_BDEV_QOS_MIN_IOS_PER_SEC;
9179 : } else {
9180 27 : if (limits[i] > SPDK_BDEV_QOS_MAX_MBYTES_PER_SEC) {
9181 0 : SPDK_WARNLOG("Requested rate limit %" PRIu64 " will result in uint64_t overflow, "
9182 : "reset to %" PRIu64 "\n", limits[i], SPDK_BDEV_QOS_MAX_MBYTES_PER_SEC);
9183 0 : limits[i] = SPDK_BDEV_QOS_MAX_MBYTES_PER_SEC;
9184 : }
9185 : /* Change from megabyte to byte rate limit */
9186 27 : limits[i] = limits[i] * 1024 * 1024;
9187 27 : min_limit_per_sec = SPDK_BDEV_QOS_MIN_BYTES_PER_SEC;
9188 : }
9189 :
9190 36 : limit_set_complement = limits[i] % min_limit_per_sec;
9191 36 : if (limit_set_complement) {
9192 0 : SPDK_ERRLOG("Requested rate limit %" PRIu64 " is not a multiple of %" PRIu64 "\n",
9193 : limits[i], min_limit_per_sec);
9194 0 : limits[i] += min_limit_per_sec - limit_set_complement;
9195 0 : SPDK_ERRLOG("Round up the rate limit to %" PRIu64 "\n", limits[i]);
9196 : }
9197 : }
9198 :
9199 9 : ctx = calloc(1, sizeof(*ctx));
9200 9 : if (ctx == NULL) {
9201 0 : cb_fn(cb_arg, -ENOMEM);
9202 0 : return;
9203 : }
9204 :
9205 9 : ctx->cb_fn = cb_fn;
9206 9 : ctx->cb_arg = cb_arg;
9207 9 : ctx->bdev = bdev;
9208 :
9209 9 : spdk_spin_lock(&bdev->internal.spinlock);
9210 9 : if (bdev->internal.qos_mod_in_progress) {
9211 1 : spdk_spin_unlock(&bdev->internal.spinlock);
9212 1 : free(ctx);
9213 1 : cb_fn(cb_arg, -EAGAIN);
9214 1 : return;
9215 : }
9216 8 : bdev->internal.qos_mod_in_progress = true;
9217 :
9218 8 : if (disable_rate_limit == true && bdev->internal.qos) {
9219 10 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
9220 8 : if (limits[i] == SPDK_BDEV_QOS_LIMIT_NOT_DEFINED &&
9221 0 : (bdev->internal.qos->rate_limits[i].limit > 0 &&
9222 0 : bdev->internal.qos->rate_limits[i].limit !=
9223 : SPDK_BDEV_QOS_LIMIT_NOT_DEFINED)) {
9224 0 : disable_rate_limit = false;
9225 0 : break;
9226 : }
9227 : }
9228 : }
9229 :
9230 8 : if (disable_rate_limit == false) {
9231 5 : if (bdev->internal.qos == NULL) {
9232 4 : bdev->internal.qos = calloc(1, sizeof(*bdev->internal.qos));
9233 4 : if (!bdev->internal.qos) {
9234 0 : spdk_spin_unlock(&bdev->internal.spinlock);
9235 0 : SPDK_ERRLOG("Unable to allocate memory for QoS tracking\n");
9236 0 : bdev_set_qos_limit_done(ctx, -ENOMEM);
9237 0 : return;
9238 : }
9239 : }
9240 :
9241 5 : if (bdev->internal.qos->thread == NULL) {
9242 : /* Enabling */
9243 4 : bdev_set_qos_rate_limits(bdev, limits);
9244 :
9245 4 : spdk_bdev_for_each_channel(bdev, bdev_enable_qos_msg, ctx,
9246 : bdev_enable_qos_done);
9247 : } else {
9248 : /* Updating */
9249 1 : bdev_set_qos_rate_limits(bdev, limits);
9250 :
9251 1 : spdk_thread_send_msg(bdev->internal.qos->thread,
9252 : bdev_update_qos_rate_limit_msg, ctx);
9253 : }
9254 : } else {
9255 3 : if (bdev->internal.qos != NULL) {
9256 2 : bdev_set_qos_rate_limits(bdev, limits);
9257 :
9258 : /* Disabling */
9259 2 : spdk_bdev_for_each_channel(bdev, bdev_disable_qos_msg, ctx,
9260 : bdev_disable_qos_msg_done);
9261 : } else {
9262 1 : spdk_spin_unlock(&bdev->internal.spinlock);
9263 1 : bdev_set_qos_limit_done(ctx, 0);
9264 1 : return;
9265 : }
9266 : }
9267 :
9268 7 : spdk_spin_unlock(&bdev->internal.spinlock);
9269 : }
9270 :
9271 : struct spdk_bdev_histogram_ctx {
9272 : spdk_bdev_histogram_status_cb cb_fn;
9273 : void *cb_arg;
9274 : struct spdk_bdev *bdev;
9275 : int status;
9276 : };
9277 :
9278 : static void
9279 2 : bdev_histogram_disable_channel_cb(struct spdk_bdev *bdev, void *_ctx, int status)
9280 : {
9281 2 : struct spdk_bdev_histogram_ctx *ctx = _ctx;
9282 :
9283 2 : spdk_spin_lock(&ctx->bdev->internal.spinlock);
9284 2 : ctx->bdev->internal.histogram_in_progress = false;
9285 2 : spdk_spin_unlock(&ctx->bdev->internal.spinlock);
9286 2 : ctx->cb_fn(ctx->cb_arg, ctx->status);
9287 2 : free(ctx);
9288 2 : }
9289 :
9290 : static void
9291 3 : bdev_histogram_disable_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
9292 : struct spdk_io_channel *_ch, void *_ctx)
9293 : {
9294 3 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
9295 :
9296 3 : if (ch->histogram != NULL) {
9297 3 : spdk_histogram_data_free(ch->histogram);
9298 3 : ch->histogram = NULL;
9299 : }
9300 3 : spdk_bdev_for_each_channel_continue(i, 0);
9301 3 : }
9302 :
9303 : static void
9304 2 : bdev_histogram_enable_channel_cb(struct spdk_bdev *bdev, void *_ctx, int status)
9305 : {
9306 2 : struct spdk_bdev_histogram_ctx *ctx = _ctx;
9307 :
9308 2 : if (status != 0) {
9309 0 : ctx->status = status;
9310 0 : ctx->bdev->internal.histogram_enabled = false;
9311 0 : spdk_bdev_for_each_channel(ctx->bdev, bdev_histogram_disable_channel, ctx,
9312 : bdev_histogram_disable_channel_cb);
9313 : } else {
9314 2 : spdk_spin_lock(&ctx->bdev->internal.spinlock);
9315 2 : ctx->bdev->internal.histogram_in_progress = false;
9316 2 : spdk_spin_unlock(&ctx->bdev->internal.spinlock);
9317 2 : ctx->cb_fn(ctx->cb_arg, ctx->status);
9318 2 : free(ctx);
9319 : }
9320 2 : }
9321 :
9322 : static void
9323 3 : bdev_histogram_enable_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
9324 : struct spdk_io_channel *_ch, void *_ctx)
9325 : {
9326 3 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
9327 3 : int status = 0;
9328 :
9329 3 : if (ch->histogram == NULL) {
9330 3 : ch->histogram = spdk_histogram_data_alloc();
9331 3 : if (ch->histogram == NULL) {
9332 0 : status = -ENOMEM;
9333 : }
9334 : }
9335 :
9336 3 : spdk_bdev_for_each_channel_continue(i, status);
9337 3 : }
9338 :
9339 : void
9340 4 : spdk_bdev_histogram_enable(struct spdk_bdev *bdev, spdk_bdev_histogram_status_cb cb_fn,
9341 : void *cb_arg, bool enable)
9342 : {
9343 : struct spdk_bdev_histogram_ctx *ctx;
9344 :
9345 4 : ctx = calloc(1, sizeof(struct spdk_bdev_histogram_ctx));
9346 4 : if (ctx == NULL) {
9347 0 : cb_fn(cb_arg, -ENOMEM);
9348 0 : return;
9349 : }
9350 :
9351 4 : ctx->bdev = bdev;
9352 4 : ctx->status = 0;
9353 4 : ctx->cb_fn = cb_fn;
9354 4 : ctx->cb_arg = cb_arg;
9355 :
9356 4 : spdk_spin_lock(&bdev->internal.spinlock);
9357 4 : if (bdev->internal.histogram_in_progress) {
9358 0 : spdk_spin_unlock(&bdev->internal.spinlock);
9359 0 : free(ctx);
9360 0 : cb_fn(cb_arg, -EAGAIN);
9361 0 : return;
9362 : }
9363 :
9364 4 : bdev->internal.histogram_in_progress = true;
9365 4 : spdk_spin_unlock(&bdev->internal.spinlock);
9366 :
9367 4 : bdev->internal.histogram_enabled = enable;
9368 :
9369 4 : if (enable) {
9370 : /* Allocate histogram for each channel */
9371 2 : spdk_bdev_for_each_channel(bdev, bdev_histogram_enable_channel, ctx,
9372 : bdev_histogram_enable_channel_cb);
9373 : } else {
9374 2 : spdk_bdev_for_each_channel(bdev, bdev_histogram_disable_channel, ctx,
9375 : bdev_histogram_disable_channel_cb);
9376 : }
9377 : }
9378 :
9379 : struct spdk_bdev_histogram_data_ctx {
9380 : spdk_bdev_histogram_data_cb cb_fn;
9381 : void *cb_arg;
9382 : struct spdk_bdev *bdev;
9383 : /** merged histogram data from all channels */
9384 : struct spdk_histogram_data *histogram;
9385 : };
9386 :
9387 : static void
9388 5 : bdev_histogram_get_channel_cb(struct spdk_bdev *bdev, void *_ctx, int status)
9389 : {
9390 5 : struct spdk_bdev_histogram_data_ctx *ctx = _ctx;
9391 :
9392 5 : ctx->cb_fn(ctx->cb_arg, status, ctx->histogram);
9393 5 : free(ctx);
9394 5 : }
9395 :
9396 : static void
9397 7 : bdev_histogram_get_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
9398 : struct spdk_io_channel *_ch, void *_ctx)
9399 : {
9400 7 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
9401 7 : struct spdk_bdev_histogram_data_ctx *ctx = _ctx;
9402 7 : int status = 0;
9403 :
9404 7 : if (ch->histogram == NULL) {
9405 1 : status = -EFAULT;
9406 : } else {
9407 6 : spdk_histogram_data_merge(ctx->histogram, ch->histogram);
9408 : }
9409 :
9410 7 : spdk_bdev_for_each_channel_continue(i, status);
9411 7 : }
9412 :
9413 : void
9414 5 : spdk_bdev_histogram_get(struct spdk_bdev *bdev, struct spdk_histogram_data *histogram,
9415 : spdk_bdev_histogram_data_cb cb_fn,
9416 : void *cb_arg)
9417 : {
9418 : struct spdk_bdev_histogram_data_ctx *ctx;
9419 :
9420 5 : ctx = calloc(1, sizeof(struct spdk_bdev_histogram_data_ctx));
9421 5 : if (ctx == NULL) {
9422 0 : cb_fn(cb_arg, -ENOMEM, NULL);
9423 0 : return;
9424 : }
9425 :
9426 5 : ctx->bdev = bdev;
9427 5 : ctx->cb_fn = cb_fn;
9428 5 : ctx->cb_arg = cb_arg;
9429 :
9430 5 : ctx->histogram = histogram;
9431 :
9432 5 : spdk_bdev_for_each_channel(bdev, bdev_histogram_get_channel, ctx,
9433 : bdev_histogram_get_channel_cb);
9434 : }
9435 :
9436 : void
9437 2 : spdk_bdev_channel_get_histogram(struct spdk_io_channel *ch, spdk_bdev_histogram_data_cb cb_fn,
9438 : void *cb_arg)
9439 : {
9440 2 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(ch);
9441 2 : int status = 0;
9442 :
9443 2 : assert(cb_fn != NULL);
9444 :
9445 2 : if (bdev_ch->histogram == NULL) {
9446 1 : status = -EFAULT;
9447 : }
9448 2 : cb_fn(cb_arg, status, bdev_ch->histogram);
9449 2 : }
9450 :
9451 : size_t
9452 0 : spdk_bdev_get_media_events(struct spdk_bdev_desc *desc, struct spdk_bdev_media_event *events,
9453 : size_t max_events)
9454 : {
9455 : struct media_event_entry *entry;
9456 0 : size_t num_events = 0;
9457 :
9458 0 : for (; num_events < max_events; ++num_events) {
9459 0 : entry = TAILQ_FIRST(&desc->pending_media_events);
9460 0 : if (entry == NULL) {
9461 0 : break;
9462 : }
9463 :
9464 0 : events[num_events] = entry->event;
9465 0 : TAILQ_REMOVE(&desc->pending_media_events, entry, tailq);
9466 0 : TAILQ_INSERT_TAIL(&desc->free_media_events, entry, tailq);
9467 : }
9468 :
9469 0 : return num_events;
9470 : }
9471 :
9472 : int
9473 0 : spdk_bdev_push_media_events(struct spdk_bdev *bdev, const struct spdk_bdev_media_event *events,
9474 : size_t num_events)
9475 : {
9476 : struct spdk_bdev_desc *desc;
9477 : struct media_event_entry *entry;
9478 : size_t event_id;
9479 0 : int rc = 0;
9480 :
9481 0 : assert(bdev->media_events);
9482 :
9483 0 : spdk_spin_lock(&bdev->internal.spinlock);
9484 0 : TAILQ_FOREACH(desc, &bdev->internal.open_descs, link) {
9485 0 : if (desc->write) {
9486 0 : break;
9487 : }
9488 : }
9489 :
9490 0 : if (desc == NULL || desc->media_events_buffer == NULL) {
9491 0 : rc = -ENODEV;
9492 0 : goto out;
9493 : }
9494 :
9495 0 : for (event_id = 0; event_id < num_events; ++event_id) {
9496 0 : entry = TAILQ_FIRST(&desc->free_media_events);
9497 0 : if (entry == NULL) {
9498 0 : break;
9499 : }
9500 :
9501 0 : TAILQ_REMOVE(&desc->free_media_events, entry, tailq);
9502 0 : TAILQ_INSERT_TAIL(&desc->pending_media_events, entry, tailq);
9503 0 : entry->event = events[event_id];
9504 : }
9505 :
9506 0 : rc = event_id;
9507 0 : out:
9508 0 : spdk_spin_unlock(&bdev->internal.spinlock);
9509 0 : return rc;
9510 : }
9511 :
9512 : static void
9513 0 : _media_management_notify(void *arg)
9514 : {
9515 0 : struct spdk_bdev_desc *desc = arg;
9516 :
9517 0 : _event_notify(desc, SPDK_BDEV_EVENT_MEDIA_MANAGEMENT);
9518 0 : }
9519 :
9520 : void
9521 0 : spdk_bdev_notify_media_management(struct spdk_bdev *bdev)
9522 : {
9523 : struct spdk_bdev_desc *desc;
9524 :
9525 0 : spdk_spin_lock(&bdev->internal.spinlock);
9526 0 : TAILQ_FOREACH(desc, &bdev->internal.open_descs, link) {
9527 0 : if (!TAILQ_EMPTY(&desc->pending_media_events)) {
9528 0 : event_notify(desc, _media_management_notify);
9529 : }
9530 : }
9531 0 : spdk_spin_unlock(&bdev->internal.spinlock);
9532 0 : }
9533 :
9534 : struct locked_lba_range_ctx {
9535 : struct lba_range range;
9536 : struct lba_range *current_range;
9537 : struct lba_range *owner_range;
9538 : struct spdk_poller *poller;
9539 : lock_range_cb cb_fn;
9540 : void *cb_arg;
9541 : };
9542 :
9543 : static void
9544 0 : bdev_lock_error_cleanup_cb(struct spdk_bdev *bdev, void *_ctx, int status)
9545 : {
9546 0 : struct locked_lba_range_ctx *ctx = _ctx;
9547 :
9548 0 : ctx->cb_fn(&ctx->range, ctx->cb_arg, -ENOMEM);
9549 0 : free(ctx);
9550 0 : }
9551 :
9552 : static void bdev_unlock_lba_range_get_channel(struct spdk_bdev_channel_iter *i,
9553 : struct spdk_bdev *bdev, struct spdk_io_channel *ch, void *_ctx);
9554 :
9555 : static void
9556 14 : bdev_lock_lba_range_cb(struct spdk_bdev *bdev, void *_ctx, int status)
9557 : {
9558 14 : struct locked_lba_range_ctx *ctx = _ctx;
9559 :
9560 14 : if (status == -ENOMEM) {
9561 : /* One of the channels could not allocate a range object.
9562 : * So we have to go back and clean up any ranges that were
9563 : * allocated successfully before we return error status to
9564 : * the caller. We can reuse the unlock function to do that
9565 : * clean up.
9566 : */
9567 0 : spdk_bdev_for_each_channel(bdev, bdev_unlock_lba_range_get_channel, ctx,
9568 : bdev_lock_error_cleanup_cb);
9569 0 : return;
9570 : }
9571 :
9572 : /* All channels have locked this range and no I/O overlapping the range
9573 : * are outstanding! Set the owner_ch for the range object for the
9574 : * locking channel, so that this channel will know that it is allowed
9575 : * to write to this range.
9576 : */
9577 14 : if (ctx->owner_range != NULL) {
9578 10 : ctx->owner_range->owner_ch = ctx->range.owner_ch;
9579 : }
9580 :
9581 14 : ctx->cb_fn(&ctx->range, ctx->cb_arg, status);
9582 :
9583 : /* Don't free the ctx here. Its range is in the bdev's global list of
9584 : * locked ranges still, and will be removed and freed when this range
9585 : * is later unlocked.
9586 : */
9587 : }
9588 :
9589 : static int
9590 17 : bdev_lock_lba_range_check_io(void *_i)
9591 : {
9592 17 : struct spdk_bdev_channel_iter *i = _i;
9593 17 : struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i->i);
9594 17 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
9595 17 : struct locked_lba_range_ctx *ctx = i->ctx;
9596 17 : struct lba_range *range = ctx->current_range;
9597 : struct spdk_bdev_io *bdev_io;
9598 :
9599 17 : spdk_poller_unregister(&ctx->poller);
9600 :
9601 : /* The range is now in the locked_ranges, so no new IO can be submitted to this
9602 : * range. But we need to wait until any outstanding IO overlapping with this range
9603 : * are completed.
9604 : */
9605 18 : TAILQ_FOREACH(bdev_io, &ch->io_submitted, internal.ch_link) {
9606 3 : if (bdev_io_range_is_locked(bdev_io, range)) {
9607 2 : ctx->poller = SPDK_POLLER_REGISTER(bdev_lock_lba_range_check_io, i, 100);
9608 2 : return SPDK_POLLER_BUSY;
9609 : }
9610 : }
9611 :
9612 15 : spdk_bdev_for_each_channel_continue(i, 0);
9613 15 : return SPDK_POLLER_BUSY;
9614 : }
9615 :
9616 : static void
9617 15 : bdev_lock_lba_range_get_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
9618 : struct spdk_io_channel *_ch, void *_ctx)
9619 : {
9620 15 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
9621 15 : struct locked_lba_range_ctx *ctx = _ctx;
9622 : struct lba_range *range;
9623 :
9624 16 : TAILQ_FOREACH(range, &ch->locked_ranges, tailq) {
9625 1 : if (range->length == ctx->range.length &&
9626 0 : range->offset == ctx->range.offset &&
9627 0 : range->locked_ctx == ctx->range.locked_ctx) {
9628 : /* This range already exists on this channel, so don't add
9629 : * it again. This can happen when a new channel is created
9630 : * while the for_each_channel operation is in progress.
9631 : * Do not check for outstanding I/O in that case, since the
9632 : * range was locked before any I/O could be submitted to the
9633 : * new channel.
9634 : */
9635 0 : spdk_bdev_for_each_channel_continue(i, 0);
9636 0 : return;
9637 : }
9638 : }
9639 :
9640 15 : range = calloc(1, sizeof(*range));
9641 15 : if (range == NULL) {
9642 0 : spdk_bdev_for_each_channel_continue(i, -ENOMEM);
9643 0 : return;
9644 : }
9645 :
9646 15 : range->length = ctx->range.length;
9647 15 : range->offset = ctx->range.offset;
9648 15 : range->locked_ctx = ctx->range.locked_ctx;
9649 15 : range->quiesce = ctx->range.quiesce;
9650 15 : ctx->current_range = range;
9651 15 : if (ctx->range.owner_ch == ch) {
9652 : /* This is the range object for the channel that will hold
9653 : * the lock. Store it in the ctx object so that we can easily
9654 : * set its owner_ch after the lock is finally acquired.
9655 : */
9656 10 : ctx->owner_range = range;
9657 : }
9658 15 : TAILQ_INSERT_TAIL(&ch->locked_ranges, range, tailq);
9659 15 : bdev_lock_lba_range_check_io(i);
9660 : }
9661 :
9662 : static void
9663 14 : bdev_lock_lba_range_ctx(struct spdk_bdev *bdev, struct locked_lba_range_ctx *ctx)
9664 : {
9665 14 : assert(spdk_get_thread() == ctx->range.owner_thread);
9666 14 : assert(ctx->range.owner_ch == NULL ||
9667 : spdk_io_channel_get_thread(ctx->range.owner_ch->channel) == ctx->range.owner_thread);
9668 :
9669 : /* We will add a copy of this range to each channel now. */
9670 14 : spdk_bdev_for_each_channel(bdev, bdev_lock_lba_range_get_channel, ctx,
9671 : bdev_lock_lba_range_cb);
9672 14 : }
9673 :
9674 : static bool
9675 17 : bdev_lba_range_overlaps_tailq(struct lba_range *range, lba_range_tailq_t *tailq)
9676 : {
9677 : struct lba_range *r;
9678 :
9679 18 : TAILQ_FOREACH(r, tailq, tailq) {
9680 4 : if (bdev_lba_range_overlapped(range, r)) {
9681 3 : return true;
9682 : }
9683 : }
9684 14 : return false;
9685 : }
9686 :
9687 : static void bdev_quiesce_range_locked(struct lba_range *range, void *ctx, int status);
9688 :
9689 : static int
9690 14 : _bdev_lock_lba_range(struct spdk_bdev *bdev, struct spdk_bdev_channel *ch,
9691 : uint64_t offset, uint64_t length,
9692 : lock_range_cb cb_fn, void *cb_arg)
9693 : {
9694 : struct locked_lba_range_ctx *ctx;
9695 :
9696 14 : ctx = calloc(1, sizeof(*ctx));
9697 14 : if (ctx == NULL) {
9698 0 : return -ENOMEM;
9699 : }
9700 :
9701 14 : ctx->range.offset = offset;
9702 14 : ctx->range.length = length;
9703 14 : ctx->range.owner_thread = spdk_get_thread();
9704 14 : ctx->range.owner_ch = ch;
9705 14 : ctx->range.locked_ctx = cb_arg;
9706 14 : ctx->range.bdev = bdev;
9707 14 : ctx->range.quiesce = (cb_fn == bdev_quiesce_range_locked);
9708 14 : ctx->cb_fn = cb_fn;
9709 14 : ctx->cb_arg = cb_arg;
9710 :
9711 14 : spdk_spin_lock(&bdev->internal.spinlock);
9712 14 : if (bdev_lba_range_overlaps_tailq(&ctx->range, &bdev->internal.locked_ranges)) {
9713 : /* There is an active lock overlapping with this range.
9714 : * Put it on the pending list until this range no
9715 : * longer overlaps with another.
9716 : */
9717 2 : TAILQ_INSERT_TAIL(&bdev->internal.pending_locked_ranges, &ctx->range, tailq);
9718 : } else {
9719 12 : TAILQ_INSERT_TAIL(&bdev->internal.locked_ranges, &ctx->range, tailq);
9720 12 : bdev_lock_lba_range_ctx(bdev, ctx);
9721 : }
9722 14 : spdk_spin_unlock(&bdev->internal.spinlock);
9723 14 : return 0;
9724 : }
9725 :
9726 : static int
9727 10 : bdev_lock_lba_range(struct spdk_bdev_desc *desc, struct spdk_io_channel *_ch,
9728 : uint64_t offset, uint64_t length,
9729 : lock_range_cb cb_fn, void *cb_arg)
9730 : {
9731 10 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
9732 10 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
9733 :
9734 10 : if (cb_arg == NULL) {
9735 0 : SPDK_ERRLOG("cb_arg must not be NULL\n");
9736 0 : return -EINVAL;
9737 : }
9738 :
9739 10 : return _bdev_lock_lba_range(bdev, ch, offset, length, cb_fn, cb_arg);
9740 : }
9741 :
9742 : static void
9743 2 : bdev_lock_lba_range_ctx_msg(void *_ctx)
9744 : {
9745 2 : struct locked_lba_range_ctx *ctx = _ctx;
9746 :
9747 2 : bdev_lock_lba_range_ctx(ctx->range.bdev, ctx);
9748 2 : }
9749 :
9750 : static void
9751 14 : bdev_unlock_lba_range_cb(struct spdk_bdev *bdev, void *_ctx, int status)
9752 : {
9753 14 : struct locked_lba_range_ctx *ctx = _ctx;
9754 : struct locked_lba_range_ctx *pending_ctx;
9755 : struct lba_range *range, *tmp;
9756 :
9757 14 : spdk_spin_lock(&bdev->internal.spinlock);
9758 : /* Check if there are any pending locked ranges that overlap with this range
9759 : * that was just unlocked. If there are, check that it doesn't overlap with any
9760 : * other locked ranges before calling bdev_lock_lba_range_ctx which will start
9761 : * the lock process.
9762 : */
9763 17 : TAILQ_FOREACH_SAFE(range, &bdev->internal.pending_locked_ranges, tailq, tmp) {
9764 3 : if (bdev_lba_range_overlapped(range, &ctx->range) &&
9765 3 : !bdev_lba_range_overlaps_tailq(range, &bdev->internal.locked_ranges)) {
9766 2 : TAILQ_REMOVE(&bdev->internal.pending_locked_ranges, range, tailq);
9767 2 : pending_ctx = SPDK_CONTAINEROF(range, struct locked_lba_range_ctx, range);
9768 2 : TAILQ_INSERT_TAIL(&bdev->internal.locked_ranges, range, tailq);
9769 2 : spdk_thread_send_msg(pending_ctx->range.owner_thread,
9770 : bdev_lock_lba_range_ctx_msg, pending_ctx);
9771 : }
9772 : }
9773 14 : spdk_spin_unlock(&bdev->internal.spinlock);
9774 :
9775 14 : ctx->cb_fn(&ctx->range, ctx->cb_arg, status);
9776 14 : free(ctx);
9777 14 : }
9778 :
9779 : static void
9780 16 : bdev_unlock_lba_range_get_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
9781 : struct spdk_io_channel *_ch, void *_ctx)
9782 : {
9783 16 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
9784 16 : struct locked_lba_range_ctx *ctx = _ctx;
9785 16 : TAILQ_HEAD(, spdk_bdev_io) io_locked;
9786 : struct spdk_bdev_io *bdev_io;
9787 : struct lba_range *range;
9788 :
9789 16 : TAILQ_FOREACH(range, &ch->locked_ranges, tailq) {
9790 16 : if (ctx->range.offset == range->offset &&
9791 16 : ctx->range.length == range->length &&
9792 16 : ctx->range.locked_ctx == range->locked_ctx) {
9793 16 : TAILQ_REMOVE(&ch->locked_ranges, range, tailq);
9794 16 : free(range);
9795 16 : break;
9796 : }
9797 : }
9798 :
9799 : /* Note: we should almost always be able to assert that the range specified
9800 : * was found. But there are some very rare corner cases where a new channel
9801 : * gets created simultaneously with a range unlock, where this function
9802 : * would execute on that new channel and wouldn't have the range.
9803 : * We also use this to clean up range allocations when a later allocation
9804 : * fails in the locking path.
9805 : * So we can't actually assert() here.
9806 : */
9807 :
9808 : /* Swap the locked IO into a temporary list, and then try to submit them again.
9809 : * We could hyper-optimize this to only resubmit locked I/O that overlap
9810 : * with the range that was just unlocked, but this isn't a performance path so
9811 : * we go for simplicity here.
9812 : */
9813 16 : TAILQ_INIT(&io_locked);
9814 16 : TAILQ_SWAP(&ch->io_locked, &io_locked, spdk_bdev_io, internal.ch_link);
9815 19 : while (!TAILQ_EMPTY(&io_locked)) {
9816 3 : bdev_io = TAILQ_FIRST(&io_locked);
9817 3 : TAILQ_REMOVE(&io_locked, bdev_io, internal.ch_link);
9818 3 : bdev_io_submit(bdev_io);
9819 : }
9820 :
9821 16 : spdk_bdev_for_each_channel_continue(i, 0);
9822 16 : }
9823 :
9824 : static int
9825 14 : _bdev_unlock_lba_range(struct spdk_bdev *bdev, uint64_t offset, uint64_t length,
9826 : lock_range_cb cb_fn, void *cb_arg)
9827 : {
9828 : struct locked_lba_range_ctx *ctx;
9829 : struct lba_range *range;
9830 :
9831 14 : spdk_spin_lock(&bdev->internal.spinlock);
9832 : /* To start the unlock the process, we find the range in the bdev's locked_ranges
9833 : * and remove it. This ensures new channels don't inherit the locked range.
9834 : * Then we will send a message to each channel to remove the range from its
9835 : * per-channel list.
9836 : */
9837 14 : TAILQ_FOREACH(range, &bdev->internal.locked_ranges, tailq) {
9838 14 : if (range->offset == offset && range->length == length &&
9839 14 : (range->owner_ch == NULL || range->locked_ctx == cb_arg)) {
9840 : break;
9841 : }
9842 : }
9843 14 : if (range == NULL) {
9844 0 : assert(false);
9845 : spdk_spin_unlock(&bdev->internal.spinlock);
9846 : return -EINVAL;
9847 : }
9848 14 : TAILQ_REMOVE(&bdev->internal.locked_ranges, range, tailq);
9849 14 : ctx = SPDK_CONTAINEROF(range, struct locked_lba_range_ctx, range);
9850 14 : spdk_spin_unlock(&bdev->internal.spinlock);
9851 :
9852 14 : ctx->cb_fn = cb_fn;
9853 14 : ctx->cb_arg = cb_arg;
9854 :
9855 14 : spdk_bdev_for_each_channel(bdev, bdev_unlock_lba_range_get_channel, ctx,
9856 : bdev_unlock_lba_range_cb);
9857 14 : return 0;
9858 : }
9859 :
9860 : static int
9861 12 : bdev_unlock_lba_range(struct spdk_bdev_desc *desc, struct spdk_io_channel *_ch,
9862 : uint64_t offset, uint64_t length,
9863 : lock_range_cb cb_fn, void *cb_arg)
9864 : {
9865 12 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
9866 12 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
9867 : struct lba_range *range;
9868 12 : bool range_found = false;
9869 :
9870 : /* Let's make sure the specified channel actually has a lock on
9871 : * the specified range. Note that the range must match exactly.
9872 : */
9873 14 : TAILQ_FOREACH(range, &ch->locked_ranges, tailq) {
9874 12 : if (range->offset == offset && range->length == length &&
9875 11 : range->owner_ch == ch && range->locked_ctx == cb_arg) {
9876 10 : range_found = true;
9877 10 : break;
9878 : }
9879 : }
9880 :
9881 12 : if (!range_found) {
9882 2 : return -EINVAL;
9883 : }
9884 :
9885 10 : return _bdev_unlock_lba_range(bdev, offset, length, cb_fn, cb_arg);
9886 : }
9887 :
9888 : struct bdev_quiesce_ctx {
9889 : spdk_bdev_quiesce_cb cb_fn;
9890 : void *cb_arg;
9891 : };
9892 :
9893 : static void
9894 4 : bdev_unquiesce_range_unlocked(struct lba_range *range, void *ctx, int status)
9895 : {
9896 4 : struct bdev_quiesce_ctx *quiesce_ctx = ctx;
9897 :
9898 4 : if (quiesce_ctx->cb_fn != NULL) {
9899 4 : quiesce_ctx->cb_fn(quiesce_ctx->cb_arg, status);
9900 : }
9901 :
9902 4 : free(quiesce_ctx);
9903 4 : }
9904 :
9905 : static void
9906 4 : bdev_quiesce_range_locked(struct lba_range *range, void *ctx, int status)
9907 : {
9908 4 : struct bdev_quiesce_ctx *quiesce_ctx = ctx;
9909 4 : struct spdk_bdev_module *module = range->bdev->module;
9910 :
9911 4 : if (status != 0) {
9912 0 : if (quiesce_ctx->cb_fn != NULL) {
9913 0 : quiesce_ctx->cb_fn(quiesce_ctx->cb_arg, status);
9914 : }
9915 0 : free(quiesce_ctx);
9916 0 : return;
9917 : }
9918 :
9919 4 : spdk_spin_lock(&module->internal.spinlock);
9920 4 : TAILQ_INSERT_TAIL(&module->internal.quiesced_ranges, range, tailq_module);
9921 4 : spdk_spin_unlock(&module->internal.spinlock);
9922 :
9923 4 : if (quiesce_ctx->cb_fn != NULL) {
9924 : /* copy the context in case the range is unlocked by the callback */
9925 4 : struct bdev_quiesce_ctx tmp = *quiesce_ctx;
9926 :
9927 4 : quiesce_ctx->cb_fn = NULL;
9928 4 : quiesce_ctx->cb_arg = NULL;
9929 :
9930 4 : tmp.cb_fn(tmp.cb_arg, status);
9931 : }
9932 : /* quiesce_ctx will be freed on unquiesce */
9933 : }
9934 :
9935 : static int
9936 9 : _spdk_bdev_quiesce(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
9937 : uint64_t offset, uint64_t length,
9938 : spdk_bdev_quiesce_cb cb_fn, void *cb_arg,
9939 : bool unquiesce)
9940 : {
9941 : struct bdev_quiesce_ctx *quiesce_ctx;
9942 : int rc;
9943 :
9944 9 : if (module != bdev->module) {
9945 0 : SPDK_ERRLOG("Bdev does not belong to specified module.\n");
9946 0 : return -EINVAL;
9947 : }
9948 :
9949 9 : if (!bdev_io_valid_blocks(bdev, offset, length)) {
9950 0 : return -EINVAL;
9951 : }
9952 :
9953 9 : if (unquiesce) {
9954 : struct lba_range *range;
9955 :
9956 : /* Make sure the specified range is actually quiesced in the specified module and
9957 : * then remove it from the list. Note that the range must match exactly.
9958 : */
9959 5 : spdk_spin_lock(&module->internal.spinlock);
9960 6 : TAILQ_FOREACH(range, &module->internal.quiesced_ranges, tailq_module) {
9961 5 : if (range->bdev == bdev && range->offset == offset && range->length == length) {
9962 4 : TAILQ_REMOVE(&module->internal.quiesced_ranges, range, tailq_module);
9963 4 : break;
9964 : }
9965 : }
9966 5 : spdk_spin_unlock(&module->internal.spinlock);
9967 :
9968 5 : if (range == NULL) {
9969 1 : SPDK_ERRLOG("The range to unquiesce was not found.\n");
9970 1 : return -EINVAL;
9971 : }
9972 :
9973 4 : quiesce_ctx = range->locked_ctx;
9974 4 : quiesce_ctx->cb_fn = cb_fn;
9975 4 : quiesce_ctx->cb_arg = cb_arg;
9976 :
9977 4 : rc = _bdev_unlock_lba_range(bdev, offset, length, bdev_unquiesce_range_unlocked, quiesce_ctx);
9978 : } else {
9979 4 : quiesce_ctx = malloc(sizeof(*quiesce_ctx));
9980 4 : if (quiesce_ctx == NULL) {
9981 0 : return -ENOMEM;
9982 : }
9983 :
9984 4 : quiesce_ctx->cb_fn = cb_fn;
9985 4 : quiesce_ctx->cb_arg = cb_arg;
9986 :
9987 4 : rc = _bdev_lock_lba_range(bdev, NULL, offset, length, bdev_quiesce_range_locked, quiesce_ctx);
9988 4 : if (rc != 0) {
9989 0 : free(quiesce_ctx);
9990 : }
9991 : }
9992 :
9993 8 : return rc;
9994 : }
9995 :
9996 : int
9997 3 : spdk_bdev_quiesce(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
9998 : spdk_bdev_quiesce_cb cb_fn, void *cb_arg)
9999 : {
10000 3 : return _spdk_bdev_quiesce(bdev, module, 0, bdev->blockcnt, cb_fn, cb_arg, false);
10001 : }
10002 :
10003 : int
10004 3 : spdk_bdev_unquiesce(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
10005 : spdk_bdev_quiesce_cb cb_fn, void *cb_arg)
10006 : {
10007 3 : return _spdk_bdev_quiesce(bdev, module, 0, bdev->blockcnt, cb_fn, cb_arg, true);
10008 : }
10009 :
10010 : int
10011 1 : spdk_bdev_quiesce_range(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
10012 : uint64_t offset, uint64_t length,
10013 : spdk_bdev_quiesce_cb cb_fn, void *cb_arg)
10014 : {
10015 1 : return _spdk_bdev_quiesce(bdev, module, offset, length, cb_fn, cb_arg, false);
10016 : }
10017 :
10018 : int
10019 2 : spdk_bdev_unquiesce_range(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
10020 : uint64_t offset, uint64_t length,
10021 : spdk_bdev_quiesce_cb cb_fn, void *cb_arg)
10022 : {
10023 2 : return _spdk_bdev_quiesce(bdev, module, offset, length, cb_fn, cb_arg, true);
10024 : }
10025 :
10026 : int
10027 267 : spdk_bdev_get_memory_domains(struct spdk_bdev *bdev, struct spdk_memory_domain **domains,
10028 : int array_size)
10029 : {
10030 267 : if (!bdev) {
10031 1 : return -EINVAL;
10032 : }
10033 :
10034 266 : if (bdev->fn_table->get_memory_domains) {
10035 3 : return bdev->fn_table->get_memory_domains(bdev->ctxt, domains, array_size);
10036 : }
10037 :
10038 263 : return 0;
10039 : }
10040 :
10041 : struct spdk_bdev_for_each_io_ctx {
10042 : void *ctx;
10043 : spdk_bdev_io_fn fn;
10044 : spdk_bdev_for_each_io_cb cb;
10045 : };
10046 :
10047 : static void
10048 0 : bdev_channel_for_each_io(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
10049 : struct spdk_io_channel *io_ch, void *_ctx)
10050 : {
10051 0 : struct spdk_bdev_for_each_io_ctx *ctx = _ctx;
10052 0 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
10053 : struct spdk_bdev_io *bdev_io;
10054 0 : int rc = 0;
10055 :
10056 0 : TAILQ_FOREACH(bdev_io, &bdev_ch->io_submitted, internal.ch_link) {
10057 0 : rc = ctx->fn(ctx->ctx, bdev_io);
10058 0 : if (rc != 0) {
10059 0 : break;
10060 : }
10061 : }
10062 :
10063 0 : spdk_bdev_for_each_channel_continue(i, rc);
10064 0 : }
10065 :
10066 : static void
10067 0 : bdev_for_each_io_done(struct spdk_bdev *bdev, void *_ctx, int status)
10068 : {
10069 0 : struct spdk_bdev_for_each_io_ctx *ctx = _ctx;
10070 :
10071 0 : ctx->cb(ctx->ctx, status);
10072 :
10073 0 : free(ctx);
10074 0 : }
10075 :
10076 : void
10077 0 : spdk_bdev_for_each_bdev_io(struct spdk_bdev *bdev, void *_ctx, spdk_bdev_io_fn fn,
10078 : spdk_bdev_for_each_io_cb cb)
10079 : {
10080 : struct spdk_bdev_for_each_io_ctx *ctx;
10081 :
10082 0 : assert(fn != NULL && cb != NULL);
10083 :
10084 0 : ctx = calloc(1, sizeof(*ctx));
10085 0 : if (ctx == NULL) {
10086 0 : SPDK_ERRLOG("Failed to allocate context.\n");
10087 0 : cb(_ctx, -ENOMEM);
10088 0 : return;
10089 : }
10090 :
10091 0 : ctx->ctx = _ctx;
10092 0 : ctx->fn = fn;
10093 0 : ctx->cb = cb;
10094 :
10095 0 : spdk_bdev_for_each_channel(bdev, bdev_channel_for_each_io, ctx,
10096 : bdev_for_each_io_done);
10097 : }
10098 :
10099 : void
10100 128 : spdk_bdev_for_each_channel_continue(struct spdk_bdev_channel_iter *iter, int status)
10101 : {
10102 128 : spdk_for_each_channel_continue(iter->i, status);
10103 128 : }
10104 :
10105 : static struct spdk_bdev *
10106 351 : io_channel_iter_get_bdev(struct spdk_io_channel_iter *i)
10107 : {
10108 351 : void *io_device = spdk_io_channel_iter_get_io_device(i);
10109 :
10110 351 : return __bdev_from_io_dev(io_device);
10111 : }
10112 :
10113 : static void
10114 128 : bdev_each_channel_msg(struct spdk_io_channel_iter *i)
10115 : {
10116 128 : struct spdk_bdev_channel_iter *iter = spdk_io_channel_iter_get_ctx(i);
10117 128 : struct spdk_bdev *bdev = io_channel_iter_get_bdev(i);
10118 128 : struct spdk_io_channel *ch = spdk_io_channel_iter_get_channel(i);
10119 :
10120 128 : iter->i = i;
10121 128 : iter->fn(iter, bdev, ch, iter->ctx);
10122 128 : }
10123 :
10124 : static void
10125 223 : bdev_each_channel_cpl(struct spdk_io_channel_iter *i, int status)
10126 : {
10127 223 : struct spdk_bdev_channel_iter *iter = spdk_io_channel_iter_get_ctx(i);
10128 223 : struct spdk_bdev *bdev = io_channel_iter_get_bdev(i);
10129 :
10130 223 : iter->i = i;
10131 223 : iter->cpl(bdev, iter->ctx, status);
10132 :
10133 223 : free(iter);
10134 223 : }
10135 :
10136 : void
10137 223 : spdk_bdev_for_each_channel(struct spdk_bdev *bdev, spdk_bdev_for_each_channel_msg fn,
10138 : void *ctx, spdk_bdev_for_each_channel_done cpl)
10139 : {
10140 : struct spdk_bdev_channel_iter *iter;
10141 :
10142 223 : assert(bdev != NULL && fn != NULL && ctx != NULL);
10143 :
10144 223 : iter = calloc(1, sizeof(struct spdk_bdev_channel_iter));
10145 223 : if (iter == NULL) {
10146 0 : SPDK_ERRLOG("Unable to allocate iterator\n");
10147 0 : assert(false);
10148 : return;
10149 : }
10150 :
10151 223 : iter->fn = fn;
10152 223 : iter->cpl = cpl;
10153 223 : iter->ctx = ctx;
10154 :
10155 223 : spdk_for_each_channel(__bdev_to_io_dev(bdev), bdev_each_channel_msg,
10156 : iter, bdev_each_channel_cpl);
10157 : }
10158 :
10159 : static void
10160 3 : bdev_copy_do_write_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
10161 : {
10162 3 : struct spdk_bdev_io *parent_io = cb_arg;
10163 :
10164 3 : spdk_bdev_free_io(bdev_io);
10165 :
10166 : /* Check return status of write */
10167 3 : parent_io->internal.status = success ? SPDK_BDEV_IO_STATUS_SUCCESS : SPDK_BDEV_IO_STATUS_FAILED;
10168 3 : parent_io->internal.cb(parent_io, success, parent_io->internal.caller_ctx);
10169 3 : }
10170 :
10171 : static void
10172 3 : bdev_copy_do_write(void *_bdev_io)
10173 : {
10174 3 : struct spdk_bdev_io *bdev_io = _bdev_io;
10175 : int rc;
10176 :
10177 : /* Write blocks */
10178 3 : rc = spdk_bdev_write_blocks_with_md(bdev_io->internal.desc,
10179 3 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
10180 3 : bdev_io->u.bdev.iovs[0].iov_base,
10181 : bdev_io->u.bdev.md_buf, bdev_io->u.bdev.offset_blocks,
10182 : bdev_io->u.bdev.num_blocks, bdev_copy_do_write_done, bdev_io);
10183 :
10184 3 : if (rc == -ENOMEM) {
10185 0 : bdev_queue_io_wait_with_cb(bdev_io, bdev_copy_do_write);
10186 3 : } else if (rc != 0) {
10187 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
10188 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
10189 : }
10190 3 : }
10191 :
10192 : static void
10193 3 : bdev_copy_do_read_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
10194 : {
10195 3 : struct spdk_bdev_io *parent_io = cb_arg;
10196 :
10197 3 : spdk_bdev_free_io(bdev_io);
10198 :
10199 : /* Check return status of read */
10200 3 : if (!success) {
10201 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
10202 0 : parent_io->internal.cb(parent_io, false, parent_io->internal.caller_ctx);
10203 0 : return;
10204 : }
10205 :
10206 : /* Do write */
10207 3 : bdev_copy_do_write(parent_io);
10208 : }
10209 :
10210 : static void
10211 3 : bdev_copy_do_read(void *_bdev_io)
10212 : {
10213 3 : struct spdk_bdev_io *bdev_io = _bdev_io;
10214 : int rc;
10215 :
10216 : /* Read blocks */
10217 3 : rc = spdk_bdev_read_blocks_with_md(bdev_io->internal.desc,
10218 3 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
10219 3 : bdev_io->u.bdev.iovs[0].iov_base,
10220 : bdev_io->u.bdev.md_buf, bdev_io->u.bdev.copy.src_offset_blocks,
10221 : bdev_io->u.bdev.num_blocks, bdev_copy_do_read_done, bdev_io);
10222 :
10223 3 : if (rc == -ENOMEM) {
10224 0 : bdev_queue_io_wait_with_cb(bdev_io, bdev_copy_do_read);
10225 3 : } else if (rc != 0) {
10226 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
10227 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
10228 : }
10229 3 : }
10230 :
10231 : static void
10232 3 : bdev_copy_get_buf_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io, bool success)
10233 : {
10234 3 : if (!success) {
10235 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
10236 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
10237 0 : return;
10238 : }
10239 :
10240 3 : bdev_copy_do_read(bdev_io);
10241 : }
10242 :
10243 : int
10244 27 : spdk_bdev_copy_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
10245 : uint64_t dst_offset_blocks, uint64_t src_offset_blocks, uint64_t num_blocks,
10246 : spdk_bdev_io_completion_cb cb, void *cb_arg)
10247 : {
10248 27 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
10249 : struct spdk_bdev_io *bdev_io;
10250 27 : struct spdk_bdev_channel *channel = spdk_io_channel_get_ctx(ch);
10251 :
10252 27 : if (!desc->write) {
10253 0 : return -EBADF;
10254 : }
10255 :
10256 27 : if (num_blocks == 0) {
10257 0 : SPDK_ERRLOG("Can't copy 0 blocks\n");
10258 0 : return -EINVAL;
10259 : }
10260 :
10261 27 : if (!bdev_io_valid_blocks(bdev, dst_offset_blocks, num_blocks) ||
10262 27 : !bdev_io_valid_blocks(bdev, src_offset_blocks, num_blocks)) {
10263 0 : SPDK_DEBUGLOG(bdev,
10264 : "Invalid offset or number of blocks: dst %lu, src %lu, count %lu\n",
10265 : dst_offset_blocks, src_offset_blocks, num_blocks);
10266 0 : return -EINVAL;
10267 : }
10268 :
10269 27 : bdev_io = bdev_channel_get_io(channel);
10270 27 : if (!bdev_io) {
10271 0 : return -ENOMEM;
10272 : }
10273 :
10274 27 : bdev_io->internal.ch = channel;
10275 27 : bdev_io->internal.desc = desc;
10276 27 : bdev_io->type = SPDK_BDEV_IO_TYPE_COPY;
10277 :
10278 27 : bdev_io->u.bdev.offset_blocks = dst_offset_blocks;
10279 27 : bdev_io->u.bdev.copy.src_offset_blocks = src_offset_blocks;
10280 27 : bdev_io->u.bdev.num_blocks = num_blocks;
10281 27 : bdev_io->u.bdev.memory_domain = NULL;
10282 27 : bdev_io->u.bdev.memory_domain_ctx = NULL;
10283 27 : bdev_io->u.bdev.iovs = NULL;
10284 27 : bdev_io->u.bdev.iovcnt = 0;
10285 27 : bdev_io->u.bdev.md_buf = NULL;
10286 27 : bdev_io->u.bdev.accel_sequence = NULL;
10287 27 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
10288 :
10289 27 : if (dst_offset_blocks == src_offset_blocks) {
10290 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
10291 0 : bdev_io->internal.cb(bdev_io, true, bdev_io->internal.caller_ctx);
10292 :
10293 0 : return 0;
10294 : }
10295 :
10296 :
10297 : /* If the copy size is large and should be split, use the generic split logic
10298 : * regardless of whether SPDK_BDEV_IO_TYPE_COPY is supported or not.
10299 : *
10300 : * Then, send the copy request if SPDK_BDEV_IO_TYPE_COPY is supported or
10301 : * emulate it using regular read and write requests otherwise.
10302 : */
10303 27 : if (spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COPY) ||
10304 4 : bdev_io->internal.split) {
10305 24 : bdev_io_submit(bdev_io);
10306 24 : return 0;
10307 : }
10308 :
10309 3 : spdk_bdev_io_get_buf(bdev_io, bdev_copy_get_buf_cb, num_blocks * spdk_bdev_get_block_size(bdev));
10310 :
10311 3 : return 0;
10312 : }
10313 :
10314 3 : SPDK_LOG_REGISTER_COMPONENT(bdev)
10315 :
10316 3 : SPDK_TRACE_REGISTER_FN(bdev_trace, "bdev", TRACE_GROUP_BDEV)
10317 : {
10318 0 : struct spdk_trace_tpoint_opts opts[] = {
10319 : {
10320 : "BDEV_IO_START", TRACE_BDEV_IO_START,
10321 : OWNER_BDEV, OBJECT_BDEV_IO, 1,
10322 : {
10323 : { "type", SPDK_TRACE_ARG_TYPE_INT, 8 },
10324 : { "ctx", SPDK_TRACE_ARG_TYPE_PTR, 8 },
10325 : { "offset", SPDK_TRACE_ARG_TYPE_INT, 8 },
10326 : { "len", SPDK_TRACE_ARG_TYPE_INT, 8 },
10327 : { "name", SPDK_TRACE_ARG_TYPE_STR, 40}
10328 : }
10329 : },
10330 : {
10331 : "BDEV_IO_DONE", TRACE_BDEV_IO_DONE,
10332 : OWNER_BDEV, OBJECT_BDEV_IO, 0,
10333 : {{ "ctx", SPDK_TRACE_ARG_TYPE_PTR, 8 }}
10334 : },
10335 : {
10336 : "BDEV_IOCH_CREATE", TRACE_BDEV_IOCH_CREATE,
10337 : OWNER_BDEV, OBJECT_NONE, 1,
10338 : {
10339 : { "name", SPDK_TRACE_ARG_TYPE_STR, 40 },
10340 : { "thread_id", SPDK_TRACE_ARG_TYPE_INT, 8}
10341 : }
10342 : },
10343 : {
10344 : "BDEV_IOCH_DESTROY", TRACE_BDEV_IOCH_DESTROY,
10345 : OWNER_BDEV, OBJECT_NONE, 0,
10346 : {
10347 : { "name", SPDK_TRACE_ARG_TYPE_STR, 40 },
10348 : { "thread_id", SPDK_TRACE_ARG_TYPE_INT, 8}
10349 : }
10350 : },
10351 : };
10352 :
10353 :
10354 0 : spdk_trace_register_owner(OWNER_BDEV, 'b');
10355 0 : spdk_trace_register_object(OBJECT_BDEV_IO, 'i');
10356 0 : spdk_trace_register_description_ext(opts, SPDK_COUNTOF(opts));
10357 0 : spdk_trace_tpoint_register_relation(TRACE_BDEV_NVME_IO_START, OBJECT_BDEV_IO, 0);
10358 0 : spdk_trace_tpoint_register_relation(TRACE_BDEV_NVME_IO_DONE, OBJECT_BDEV_IO, 0);
10359 0 : }
|