Line data Source code
1 : /* SPDX-License-Identifier: BSD-3-Clause
2 : * Copyright (C) 2016 Intel Corporation. All rights reserved.
3 : * Copyright (c) 2019 Mellanox Technologies LTD. All rights reserved.
4 : * Copyright (c) 2021-2023 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
5 : */
6 :
7 : #include "spdk/stdinc.h"
8 :
9 : #include "spdk/bdev.h"
10 :
11 : #include "spdk/accel.h"
12 : #include "spdk/config.h"
13 : #include "spdk/env.h"
14 : #include "spdk/thread.h"
15 : #include "spdk/likely.h"
16 : #include "spdk/queue.h"
17 : #include "spdk/nvme_spec.h"
18 : #include "spdk/scsi_spec.h"
19 : #include "spdk/notify.h"
20 : #include "spdk/util.h"
21 : #include "spdk/trace.h"
22 : #include "spdk/dma.h"
23 :
24 : #include "spdk/bdev_module.h"
25 : #include "spdk/log.h"
26 : #include "spdk/string.h"
27 :
28 : #include "bdev_internal.h"
29 : #include "spdk_internal/trace_defs.h"
30 : #include "spdk_internal/assert.h"
31 :
32 : #ifdef SPDK_CONFIG_VTUNE
33 : #include "ittnotify.h"
34 : #include "ittnotify_types.h"
35 : int __itt_init_ittlib(const char *, __itt_group_id);
36 : #endif
37 :
38 : #define SPDK_BDEV_IO_POOL_SIZE (64 * 1024 - 1)
39 : #define SPDK_BDEV_IO_CACHE_SIZE 256
40 : #define SPDK_BDEV_AUTO_EXAMINE true
41 : #define BUF_SMALL_CACHE_SIZE 128
42 : #define BUF_LARGE_CACHE_SIZE 16
43 : #define NOMEM_THRESHOLD_COUNT 8
44 :
45 : #define SPDK_BDEV_QOS_TIMESLICE_IN_USEC 1000
46 : #define SPDK_BDEV_QOS_MIN_IO_PER_TIMESLICE 1
47 : #define SPDK_BDEV_QOS_MIN_BYTE_PER_TIMESLICE 512
48 : #define SPDK_BDEV_QOS_MIN_IOS_PER_SEC 1000
49 : #define SPDK_BDEV_QOS_MIN_BYTES_PER_SEC (1024 * 1024)
50 : #define SPDK_BDEV_QOS_MAX_MBYTES_PER_SEC (UINT64_MAX / (1024 * 1024))
51 : #define SPDK_BDEV_QOS_LIMIT_NOT_DEFINED UINT64_MAX
52 : #define SPDK_BDEV_IO_POLL_INTERVAL_IN_MSEC 1000
53 :
54 : /* The maximum number of children requests for a UNMAP or WRITE ZEROES command
55 : * when splitting into children requests at a time.
56 : */
57 : #define SPDK_BDEV_MAX_CHILDREN_UNMAP_WRITE_ZEROES_REQS (8)
58 : #define BDEV_RESET_CHECK_OUTSTANDING_IO_PERIOD 1000000
59 :
60 : /* The maximum number of children requests for a COPY command
61 : * when splitting into children requests at a time.
62 : */
63 : #define SPDK_BDEV_MAX_CHILDREN_COPY_REQS (8)
64 :
65 : #define LOG_ALREADY_CLAIMED_ERROR(detail, bdev) \
66 : log_already_claimed(SPDK_LOG_ERROR, __LINE__, __func__, detail, bdev)
67 : #ifdef DEBUG
68 : #define LOG_ALREADY_CLAIMED_DEBUG(detail, bdev) \
69 : log_already_claimed(SPDK_LOG_DEBUG, __LINE__, __func__, detail, bdev)
70 : #else
71 : #define LOG_ALREADY_CLAIMED_DEBUG(detail, bdev) do {} while(0)
72 : #endif
73 :
74 : static void log_already_claimed(enum spdk_log_level level, const int line, const char *func,
75 : const char *detail, struct spdk_bdev *bdev);
76 :
77 : static const char *qos_rpc_type[] = {"rw_ios_per_sec",
78 : "rw_mbytes_per_sec", "r_mbytes_per_sec", "w_mbytes_per_sec"
79 : };
80 :
81 : TAILQ_HEAD(spdk_bdev_list, spdk_bdev);
82 :
83 : RB_HEAD(bdev_name_tree, spdk_bdev_name);
84 :
85 : static int
86 546 : bdev_name_cmp(struct spdk_bdev_name *name1, struct spdk_bdev_name *name2)
87 : {
88 546 : return strcmp(name1->name, name2->name);
89 : }
90 :
91 2065 : RB_GENERATE_STATIC(bdev_name_tree, spdk_bdev_name, node, bdev_name_cmp);
92 :
93 : struct spdk_bdev_mgr {
94 : struct spdk_mempool *bdev_io_pool;
95 :
96 : void *zero_buffer;
97 :
98 : TAILQ_HEAD(bdev_module_list, spdk_bdev_module) bdev_modules;
99 :
100 : struct spdk_bdev_list bdevs;
101 : struct bdev_name_tree bdev_names;
102 :
103 : bool init_complete;
104 : bool module_init_complete;
105 :
106 : struct spdk_spinlock spinlock;
107 :
108 : TAILQ_HEAD(, spdk_bdev_open_async_ctx) async_bdev_opens;
109 :
110 : #ifdef SPDK_CONFIG_VTUNE
111 : __itt_domain *domain;
112 : #endif
113 : };
114 :
115 : static struct spdk_bdev_mgr g_bdev_mgr = {
116 : .bdev_modules = TAILQ_HEAD_INITIALIZER(g_bdev_mgr.bdev_modules),
117 : .bdevs = TAILQ_HEAD_INITIALIZER(g_bdev_mgr.bdevs),
118 : .bdev_names = RB_INITIALIZER(g_bdev_mgr.bdev_names),
119 : .init_complete = false,
120 : .module_init_complete = false,
121 : .async_bdev_opens = TAILQ_HEAD_INITIALIZER(g_bdev_mgr.async_bdev_opens),
122 : };
123 :
124 : static void
125 : __attribute__((constructor))
126 3 : _bdev_init(void)
127 : {
128 3 : spdk_spin_init(&g_bdev_mgr.spinlock);
129 3 : }
130 :
131 : typedef void (*lock_range_cb)(struct lba_range *range, void *ctx, int status);
132 :
133 : typedef void (*bdev_copy_bounce_buffer_cpl)(void *ctx, int rc);
134 :
135 : struct lba_range {
136 : struct spdk_bdev *bdev;
137 : uint64_t offset;
138 : uint64_t length;
139 : bool quiesce;
140 : void *locked_ctx;
141 : struct spdk_thread *owner_thread;
142 : struct spdk_bdev_channel *owner_ch;
143 : TAILQ_ENTRY(lba_range) tailq;
144 : TAILQ_ENTRY(lba_range) tailq_module;
145 : };
146 :
147 : static struct spdk_bdev_opts g_bdev_opts = {
148 : .bdev_io_pool_size = SPDK_BDEV_IO_POOL_SIZE,
149 : .bdev_io_cache_size = SPDK_BDEV_IO_CACHE_SIZE,
150 : .bdev_auto_examine = SPDK_BDEV_AUTO_EXAMINE,
151 : .iobuf_small_cache_size = BUF_SMALL_CACHE_SIZE,
152 : .iobuf_large_cache_size = BUF_LARGE_CACHE_SIZE,
153 : };
154 :
155 : static spdk_bdev_init_cb g_init_cb_fn = NULL;
156 : static void *g_init_cb_arg = NULL;
157 :
158 : static spdk_bdev_fini_cb g_fini_cb_fn = NULL;
159 : static void *g_fini_cb_arg = NULL;
160 : static struct spdk_thread *g_fini_thread = NULL;
161 :
162 : struct spdk_bdev_qos_limit {
163 : /** IOs or bytes allowed per second (i.e., 1s). */
164 : uint64_t limit;
165 :
166 : /** Remaining IOs or bytes allowed in current timeslice (e.g., 1ms).
167 : * For remaining bytes, allowed to run negative if an I/O is submitted when
168 : * some bytes are remaining, but the I/O is bigger than that amount. The
169 : * excess will be deducted from the next timeslice.
170 : */
171 : int64_t remaining_this_timeslice;
172 :
173 : /** Minimum allowed IOs or bytes to be issued in one timeslice (e.g., 1ms). */
174 : uint32_t min_per_timeslice;
175 :
176 : /** Maximum allowed IOs or bytes to be issued in one timeslice (e.g., 1ms). */
177 : uint32_t max_per_timeslice;
178 :
179 : /** Function to check whether to queue the IO.
180 : * If The IO is allowed to pass, the quota will be reduced correspondingly.
181 : */
182 : bool (*queue_io)(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io);
183 :
184 : /** Function to rewind the quota once the IO was allowed to be sent by this
185 : * limit but queued due to one of the further limits.
186 : */
187 : void (*rewind_quota)(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io);
188 : };
189 :
190 : struct spdk_bdev_qos {
191 : /** Types of structure of rate limits. */
192 : struct spdk_bdev_qos_limit rate_limits[SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES];
193 :
194 : /** The channel that all I/O are funneled through. */
195 : struct spdk_bdev_channel *ch;
196 :
197 : /** The thread on which the poller is running. */
198 : struct spdk_thread *thread;
199 :
200 : /** Size of a timeslice in tsc ticks. */
201 : uint64_t timeslice_size;
202 :
203 : /** Timestamp of start of last timeslice. */
204 : uint64_t last_timeslice;
205 :
206 : /** Poller that processes queued I/O commands each time slice. */
207 : struct spdk_poller *poller;
208 : };
209 :
210 : struct spdk_bdev_mgmt_channel {
211 : /*
212 : * Each thread keeps a cache of bdev_io - this allows
213 : * bdev threads which are *not* DPDK threads to still
214 : * benefit from a per-thread bdev_io cache. Without
215 : * this, non-DPDK threads fetching from the mempool
216 : * incur a cmpxchg on get and put.
217 : */
218 : bdev_io_stailq_t per_thread_cache;
219 : uint32_t per_thread_cache_count;
220 : uint32_t bdev_io_cache_size;
221 :
222 : struct spdk_iobuf_channel iobuf;
223 :
224 : TAILQ_HEAD(, spdk_bdev_shared_resource) shared_resources;
225 : TAILQ_HEAD(, spdk_bdev_io_wait_entry) io_wait_queue;
226 : };
227 :
228 : /*
229 : * Per-module (or per-io_device) data. Multiple bdevs built on the same io_device
230 : * will queue here their IO that awaits retry. It makes it possible to retry sending
231 : * IO to one bdev after IO from other bdev completes.
232 : */
233 : struct spdk_bdev_shared_resource {
234 : /* The bdev management channel */
235 : struct spdk_bdev_mgmt_channel *mgmt_ch;
236 :
237 : /*
238 : * Count of I/O submitted to bdev module and waiting for completion.
239 : * Incremented before submit_request() is called on an spdk_bdev_io.
240 : */
241 : uint64_t io_outstanding;
242 :
243 : /*
244 : * Queue of IO awaiting retry because of a previous NOMEM status returned
245 : * on this channel.
246 : */
247 : bdev_io_tailq_t nomem_io;
248 :
249 : /*
250 : * Threshold which io_outstanding must drop to before retrying nomem_io.
251 : */
252 : uint64_t nomem_threshold;
253 :
254 : /* I/O channel allocated by a bdev module */
255 : struct spdk_io_channel *shared_ch;
256 :
257 : struct spdk_poller *nomem_poller;
258 :
259 : /* Refcount of bdev channels using this resource */
260 : uint32_t ref;
261 :
262 : TAILQ_ENTRY(spdk_bdev_shared_resource) link;
263 : };
264 :
265 : #define BDEV_CH_RESET_IN_PROGRESS (1 << 0)
266 : #define BDEV_CH_QOS_ENABLED (1 << 1)
267 :
268 : struct spdk_bdev_channel {
269 : struct spdk_bdev *bdev;
270 :
271 : /* The channel for the underlying device */
272 : struct spdk_io_channel *channel;
273 :
274 : /* Accel channel */
275 : struct spdk_io_channel *accel_channel;
276 :
277 : /* Per io_device per thread data */
278 : struct spdk_bdev_shared_resource *shared_resource;
279 :
280 : struct spdk_bdev_io_stat *stat;
281 :
282 : /*
283 : * Count of I/O submitted to the underlying dev module through this channel
284 : * and waiting for completion.
285 : */
286 : uint64_t io_outstanding;
287 :
288 : /*
289 : * List of all submitted I/Os including I/O that are generated via splitting.
290 : */
291 : bdev_io_tailq_t io_submitted;
292 :
293 : /*
294 : * List of spdk_bdev_io that are currently queued because they write to a locked
295 : * LBA range.
296 : */
297 : bdev_io_tailq_t io_locked;
298 :
299 : /* List of I/Os with accel sequence being currently executed */
300 : bdev_io_tailq_t io_accel_exec;
301 :
302 : /* List of I/Os doing memory domain pull/push */
303 : bdev_io_tailq_t io_memory_domain;
304 :
305 : uint32_t flags;
306 :
307 : /* Counts number of bdev_io in the io_submitted TAILQ */
308 : uint16_t queue_depth;
309 :
310 : uint16_t trace_id;
311 :
312 : struct spdk_histogram_data *histogram;
313 :
314 : #ifdef SPDK_CONFIG_VTUNE
315 : uint64_t start_tsc;
316 : uint64_t interval_tsc;
317 : __itt_string_handle *handle;
318 : struct spdk_bdev_io_stat *prev_stat;
319 : #endif
320 :
321 : bdev_io_tailq_t queued_resets;
322 :
323 : lba_range_tailq_t locked_ranges;
324 :
325 : /** List of I/Os queued by QoS. */
326 : bdev_io_tailq_t qos_queued_io;
327 : };
328 :
329 : struct media_event_entry {
330 : struct spdk_bdev_media_event event;
331 : TAILQ_ENTRY(media_event_entry) tailq;
332 : };
333 :
334 : #define MEDIA_EVENT_POOL_SIZE 64
335 :
336 : struct spdk_bdev_desc {
337 : struct spdk_bdev *bdev;
338 : struct spdk_thread *thread;
339 : struct {
340 : spdk_bdev_event_cb_t event_fn;
341 : void *ctx;
342 : } callback;
343 : bool closed;
344 : bool write;
345 : bool memory_domains_supported;
346 : bool accel_sequence_supported[SPDK_BDEV_NUM_IO_TYPES];
347 : struct spdk_spinlock spinlock;
348 : uint32_t refs;
349 : struct spdk_bdev_open_opts opts;
350 : TAILQ_HEAD(, media_event_entry) pending_media_events;
351 : TAILQ_HEAD(, media_event_entry) free_media_events;
352 : struct media_event_entry *media_events_buffer;
353 : TAILQ_ENTRY(spdk_bdev_desc) link;
354 :
355 : uint64_t timeout_in_sec;
356 : spdk_bdev_io_timeout_cb cb_fn;
357 : void *cb_arg;
358 : struct spdk_poller *io_timeout_poller;
359 : struct spdk_bdev_module_claim *claim;
360 : };
361 :
362 : struct spdk_bdev_iostat_ctx {
363 : struct spdk_bdev_io_stat *stat;
364 : enum spdk_bdev_reset_stat_mode reset_mode;
365 : spdk_bdev_get_device_stat_cb cb;
366 : void *cb_arg;
367 : };
368 :
369 : struct set_qos_limit_ctx {
370 : void (*cb_fn)(void *cb_arg, int status);
371 : void *cb_arg;
372 : struct spdk_bdev *bdev;
373 : };
374 :
375 : struct spdk_bdev_channel_iter {
376 : spdk_bdev_for_each_channel_msg fn;
377 : spdk_bdev_for_each_channel_done cpl;
378 : struct spdk_io_channel_iter *i;
379 : void *ctx;
380 : };
381 :
382 : struct spdk_bdev_io_error_stat {
383 : uint32_t error_status[-SPDK_MIN_BDEV_IO_STATUS];
384 : };
385 :
386 : enum bdev_io_retry_state {
387 : BDEV_IO_RETRY_STATE_INVALID,
388 : BDEV_IO_RETRY_STATE_PULL,
389 : BDEV_IO_RETRY_STATE_PULL_MD,
390 : BDEV_IO_RETRY_STATE_SUBMIT,
391 : BDEV_IO_RETRY_STATE_PUSH,
392 : BDEV_IO_RETRY_STATE_PUSH_MD,
393 : BDEV_IO_RETRY_STATE_GET_ACCEL_BUF,
394 : };
395 :
396 : #define __bdev_to_io_dev(bdev) (((char *)bdev) + 1)
397 : #define __bdev_from_io_dev(io_dev) ((struct spdk_bdev *)(((char *)io_dev) - 1))
398 : #define __io_ch_to_bdev_ch(io_ch) ((struct spdk_bdev_channel *)spdk_io_channel_get_ctx(io_ch))
399 : #define __io_ch_to_bdev_mgmt_ch(io_ch) ((struct spdk_bdev_mgmt_channel *)spdk_io_channel_get_ctx(io_ch))
400 :
401 : static inline void bdev_io_complete(void *ctx);
402 : static inline void bdev_io_complete_unsubmitted(struct spdk_bdev_io *bdev_io);
403 : static void bdev_io_push_bounce_md_buf(struct spdk_bdev_io *bdev_io);
404 : static void bdev_io_push_bounce_data(struct spdk_bdev_io *bdev_io);
405 : static void _bdev_io_get_accel_buf(struct spdk_bdev_io *bdev_io);
406 :
407 : static void bdev_write_zero_buffer_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg);
408 : static int bdev_write_zero_buffer(struct spdk_bdev_io *bdev_io);
409 :
410 : static void bdev_enable_qos_msg(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
411 : struct spdk_io_channel *ch, void *_ctx);
412 : static void bdev_enable_qos_done(struct spdk_bdev *bdev, void *_ctx, int status);
413 :
414 : static int bdev_readv_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
415 : struct iovec *iov, int iovcnt, void *md_buf, uint64_t offset_blocks,
416 : uint64_t num_blocks,
417 : struct spdk_memory_domain *domain, void *domain_ctx,
418 : struct spdk_accel_sequence *seq, uint32_t dif_check_flags,
419 : spdk_bdev_io_completion_cb cb, void *cb_arg);
420 : static int bdev_writev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
421 : struct iovec *iov, int iovcnt, void *md_buf,
422 : uint64_t offset_blocks, uint64_t num_blocks,
423 : struct spdk_memory_domain *domain, void *domain_ctx,
424 : struct spdk_accel_sequence *seq, uint32_t dif_check_flags,
425 : uint32_t nvme_cdw12_raw, uint32_t nvme_cdw13_raw,
426 : spdk_bdev_io_completion_cb cb, void *cb_arg);
427 :
428 : static int bdev_lock_lba_range(struct spdk_bdev_desc *desc, struct spdk_io_channel *_ch,
429 : uint64_t offset, uint64_t length,
430 : lock_range_cb cb_fn, void *cb_arg);
431 :
432 : static int bdev_unlock_lba_range(struct spdk_bdev_desc *desc, struct spdk_io_channel *_ch,
433 : uint64_t offset, uint64_t length,
434 : lock_range_cb cb_fn, void *cb_arg);
435 :
436 : static bool bdev_abort_queued_io(bdev_io_tailq_t *queue, struct spdk_bdev_io *bio_to_abort);
437 : static bool bdev_abort_buf_io(struct spdk_bdev_mgmt_channel *ch, struct spdk_bdev_io *bio_to_abort);
438 :
439 : static bool claim_type_is_v2(enum spdk_bdev_claim_type type);
440 : static void bdev_desc_release_claims(struct spdk_bdev_desc *desc);
441 : static void claim_reset(struct spdk_bdev *bdev);
442 :
443 : static void bdev_ch_retry_io(struct spdk_bdev_channel *bdev_ch);
444 :
445 : static bool bdev_io_should_split(struct spdk_bdev_io *bdev_io);
446 :
447 : #define bdev_get_ext_io_opt(opts, field, defval) \
448 : ((opts) != NULL ? SPDK_GET_FIELD(opts, field, defval) : (defval))
449 :
450 : static inline void
451 669 : bdev_ch_add_to_io_submitted(struct spdk_bdev_io *bdev_io)
452 : {
453 669 : TAILQ_INSERT_TAIL(&bdev_io->internal.ch->io_submitted, bdev_io, internal.ch_link);
454 669 : bdev_io->internal.ch->queue_depth++;
455 669 : }
456 :
457 : static inline void
458 669 : bdev_ch_remove_from_io_submitted(struct spdk_bdev_io *bdev_io)
459 : {
460 669 : TAILQ_REMOVE(&bdev_io->internal.ch->io_submitted, bdev_io, internal.ch_link);
461 669 : bdev_io->internal.ch->queue_depth--;
462 669 : }
463 :
464 : void
465 14 : spdk_bdev_get_opts(struct spdk_bdev_opts *opts, size_t opts_size)
466 : {
467 14 : if (!opts) {
468 0 : SPDK_ERRLOG("opts should not be NULL\n");
469 0 : return;
470 : }
471 :
472 14 : if (!opts_size) {
473 0 : SPDK_ERRLOG("opts_size should not be zero value\n");
474 0 : return;
475 : }
476 :
477 14 : opts->opts_size = opts_size;
478 :
479 : #define SET_FIELD(field) \
480 : if (offsetof(struct spdk_bdev_opts, field) + sizeof(opts->field) <= opts_size) { \
481 : opts->field = g_bdev_opts.field; \
482 : } \
483 :
484 14 : SET_FIELD(bdev_io_pool_size);
485 14 : SET_FIELD(bdev_io_cache_size);
486 14 : SET_FIELD(bdev_auto_examine);
487 14 : SET_FIELD(iobuf_small_cache_size);
488 14 : SET_FIELD(iobuf_large_cache_size);
489 :
490 : /* Do not remove this statement, you should always update this statement when you adding a new field,
491 : * and do not forget to add the SET_FIELD statement for your added field. */
492 : SPDK_STATIC_ASSERT(sizeof(struct spdk_bdev_opts) == 32, "Incorrect size");
493 :
494 : #undef SET_FIELD
495 14 : }
496 :
497 : int
498 15 : spdk_bdev_set_opts(struct spdk_bdev_opts *opts)
499 : {
500 : uint32_t min_pool_size;
501 :
502 15 : if (!opts) {
503 0 : SPDK_ERRLOG("opts cannot be NULL\n");
504 0 : return -1;
505 : }
506 :
507 15 : if (!opts->opts_size) {
508 1 : SPDK_ERRLOG("opts_size inside opts cannot be zero value\n");
509 1 : return -1;
510 : }
511 :
512 : /*
513 : * Add 1 to the thread count to account for the extra mgmt_ch that gets created during subsystem
514 : * initialization. A second mgmt_ch will be created on the same thread when the application starts
515 : * but before the deferred put_io_channel event is executed for the first mgmt_ch.
516 : */
517 14 : min_pool_size = opts->bdev_io_cache_size * (spdk_thread_get_count() + 1);
518 14 : if (opts->bdev_io_pool_size < min_pool_size) {
519 0 : SPDK_ERRLOG("bdev_io_pool_size %" PRIu32 " is not compatible with bdev_io_cache_size %" PRIu32
520 : " and %" PRIu32 " threads\n", opts->bdev_io_pool_size, opts->bdev_io_cache_size,
521 : spdk_thread_get_count());
522 0 : SPDK_ERRLOG("bdev_io_pool_size must be at least %" PRIu32 "\n", min_pool_size);
523 0 : return -1;
524 : }
525 :
526 : #define SET_FIELD(field) \
527 : if (offsetof(struct spdk_bdev_opts, field) + sizeof(opts->field) <= opts->opts_size) { \
528 : g_bdev_opts.field = opts->field; \
529 : } \
530 :
531 14 : SET_FIELD(bdev_io_pool_size);
532 14 : SET_FIELD(bdev_io_cache_size);
533 14 : SET_FIELD(bdev_auto_examine);
534 14 : SET_FIELD(iobuf_small_cache_size);
535 14 : SET_FIELD(iobuf_large_cache_size);
536 :
537 14 : g_bdev_opts.opts_size = opts->opts_size;
538 :
539 : #undef SET_FIELD
540 :
541 14 : return 0;
542 15 : }
543 :
544 : static struct spdk_bdev *
545 146 : bdev_get_by_name(const char *bdev_name)
546 : {
547 : struct spdk_bdev_name find;
548 : struct spdk_bdev_name *res;
549 :
550 146 : find.name = (char *)bdev_name;
551 146 : res = RB_FIND(bdev_name_tree, &g_bdev_mgr.bdev_names, &find);
552 146 : if (res != NULL) {
553 139 : return res->bdev;
554 : }
555 :
556 7 : return NULL;
557 146 : }
558 :
559 : struct spdk_bdev *
560 16 : spdk_bdev_get_by_name(const char *bdev_name)
561 : {
562 : struct spdk_bdev *bdev;
563 :
564 16 : spdk_spin_lock(&g_bdev_mgr.spinlock);
565 16 : bdev = bdev_get_by_name(bdev_name);
566 16 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
567 :
568 16 : return bdev;
569 : }
570 :
571 : struct bdev_io_status_string {
572 : enum spdk_bdev_io_status status;
573 : const char *str;
574 : };
575 :
576 : static const struct bdev_io_status_string bdev_io_status_strings[] = {
577 : { SPDK_BDEV_IO_STATUS_AIO_ERROR, "aio_error" },
578 : { SPDK_BDEV_IO_STATUS_ABORTED, "aborted" },
579 : { SPDK_BDEV_IO_STATUS_FIRST_FUSED_FAILED, "first_fused_failed" },
580 : { SPDK_BDEV_IO_STATUS_MISCOMPARE, "miscompare" },
581 : { SPDK_BDEV_IO_STATUS_NOMEM, "nomem" },
582 : { SPDK_BDEV_IO_STATUS_SCSI_ERROR, "scsi_error" },
583 : { SPDK_BDEV_IO_STATUS_NVME_ERROR, "nvme_error" },
584 : { SPDK_BDEV_IO_STATUS_FAILED, "failed" },
585 : { SPDK_BDEV_IO_STATUS_PENDING, "pending" },
586 : { SPDK_BDEV_IO_STATUS_SUCCESS, "success" },
587 : };
588 :
589 : static const char *
590 0 : bdev_io_status_get_string(enum spdk_bdev_io_status status)
591 : {
592 : uint32_t i;
593 :
594 0 : for (i = 0; i < SPDK_COUNTOF(bdev_io_status_strings); i++) {
595 0 : if (bdev_io_status_strings[i].status == status) {
596 0 : return bdev_io_status_strings[i].str;
597 : }
598 0 : }
599 :
600 0 : return "reserved";
601 0 : }
602 :
603 : struct spdk_bdev_wait_for_examine_ctx {
604 : struct spdk_poller *poller;
605 : spdk_bdev_wait_for_examine_cb cb_fn;
606 : void *cb_arg;
607 : };
608 :
609 : static bool bdev_module_all_actions_completed(void);
610 :
611 : static int
612 193 : bdev_wait_for_examine_cb(void *arg)
613 : {
614 193 : struct spdk_bdev_wait_for_examine_ctx *ctx = arg;
615 :
616 193 : if (!bdev_module_all_actions_completed()) {
617 0 : return SPDK_POLLER_IDLE;
618 : }
619 :
620 193 : spdk_poller_unregister(&ctx->poller);
621 193 : ctx->cb_fn(ctx->cb_arg);
622 193 : free(ctx);
623 :
624 193 : return SPDK_POLLER_BUSY;
625 193 : }
626 :
627 : int
628 193 : spdk_bdev_wait_for_examine(spdk_bdev_wait_for_examine_cb cb_fn, void *cb_arg)
629 : {
630 : struct spdk_bdev_wait_for_examine_ctx *ctx;
631 :
632 193 : ctx = calloc(1, sizeof(*ctx));
633 193 : if (ctx == NULL) {
634 0 : return -ENOMEM;
635 : }
636 193 : ctx->cb_fn = cb_fn;
637 193 : ctx->cb_arg = cb_arg;
638 193 : ctx->poller = SPDK_POLLER_REGISTER(bdev_wait_for_examine_cb, ctx, 0);
639 :
640 193 : return 0;
641 193 : }
642 :
643 : struct spdk_bdev_examine_item {
644 : char *name;
645 : TAILQ_ENTRY(spdk_bdev_examine_item) link;
646 : };
647 :
648 : TAILQ_HEAD(spdk_bdev_examine_allowlist, spdk_bdev_examine_item);
649 :
650 : struct spdk_bdev_examine_allowlist g_bdev_examine_allowlist = TAILQ_HEAD_INITIALIZER(
651 : g_bdev_examine_allowlist);
652 :
653 : static inline bool
654 6 : bdev_examine_allowlist_check(const char *name)
655 : {
656 : struct spdk_bdev_examine_item *item;
657 6 : TAILQ_FOREACH(item, &g_bdev_examine_allowlist, link) {
658 0 : if (strcmp(name, item->name) == 0) {
659 0 : return true;
660 : }
661 0 : }
662 6 : return false;
663 6 : }
664 :
665 : static inline void
666 65 : bdev_examine_allowlist_free(void)
667 : {
668 : struct spdk_bdev_examine_item *item;
669 65 : while (!TAILQ_EMPTY(&g_bdev_examine_allowlist)) {
670 0 : item = TAILQ_FIRST(&g_bdev_examine_allowlist);
671 0 : TAILQ_REMOVE(&g_bdev_examine_allowlist, item, link);
672 0 : free(item->name);
673 0 : free(item);
674 : }
675 65 : }
676 :
677 : static inline bool
678 3 : bdev_in_examine_allowlist(struct spdk_bdev *bdev)
679 : {
680 : struct spdk_bdev_alias *tmp;
681 3 : if (bdev_examine_allowlist_check(bdev->name)) {
682 0 : return true;
683 : }
684 6 : TAILQ_FOREACH(tmp, &bdev->aliases, tailq) {
685 3 : if (bdev_examine_allowlist_check(tmp->alias.name)) {
686 0 : return true;
687 : }
688 3 : }
689 3 : return false;
690 3 : }
691 :
692 : static inline bool
693 126 : bdev_ok_to_examine(struct spdk_bdev *bdev)
694 : {
695 : /* Some bdevs may not support the READ command.
696 : * Do not try to examine them.
697 : */
698 126 : if (!spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_READ)) {
699 0 : return false;
700 : }
701 :
702 126 : if (g_bdev_opts.bdev_auto_examine) {
703 123 : return true;
704 : } else {
705 3 : return bdev_in_examine_allowlist(bdev);
706 : }
707 126 : }
708 :
709 : static void
710 126 : bdev_examine(struct spdk_bdev *bdev)
711 : {
712 : struct spdk_bdev_module *module;
713 : struct spdk_bdev_module_claim *claim, *tmpclaim;
714 : uint32_t action;
715 :
716 126 : if (!bdev_ok_to_examine(bdev)) {
717 3 : return;
718 : }
719 :
720 504 : TAILQ_FOREACH(module, &g_bdev_mgr.bdev_modules, internal.tailq) {
721 381 : if (module->examine_config) {
722 258 : spdk_spin_lock(&module->internal.spinlock);
723 258 : action = module->internal.action_in_progress;
724 258 : module->internal.action_in_progress++;
725 258 : spdk_spin_unlock(&module->internal.spinlock);
726 258 : module->examine_config(bdev);
727 258 : if (action != module->internal.action_in_progress) {
728 0 : SPDK_ERRLOG("examine_config for module %s did not call "
729 : "spdk_bdev_module_examine_done()\n", module->name);
730 0 : }
731 258 : }
732 381 : }
733 :
734 123 : spdk_spin_lock(&bdev->internal.spinlock);
735 :
736 123 : switch (bdev->internal.claim_type) {
737 : case SPDK_BDEV_CLAIM_NONE:
738 : /* Examine by all bdev modules */
739 479 : TAILQ_FOREACH(module, &g_bdev_mgr.bdev_modules, internal.tailq) {
740 361 : if (module->examine_disk) {
741 234 : spdk_spin_lock(&module->internal.spinlock);
742 234 : module->internal.action_in_progress++;
743 234 : spdk_spin_unlock(&module->internal.spinlock);
744 234 : spdk_spin_unlock(&bdev->internal.spinlock);
745 234 : module->examine_disk(bdev);
746 234 : spdk_spin_lock(&bdev->internal.spinlock);
747 234 : }
748 361 : }
749 118 : break;
750 : case SPDK_BDEV_CLAIM_EXCL_WRITE:
751 : /* Examine by the one bdev module with a v1 claim */
752 1 : module = bdev->internal.claim.v1.module;
753 1 : if (module->examine_disk) {
754 1 : spdk_spin_lock(&module->internal.spinlock);
755 1 : module->internal.action_in_progress++;
756 1 : spdk_spin_unlock(&module->internal.spinlock);
757 1 : spdk_spin_unlock(&bdev->internal.spinlock);
758 1 : module->examine_disk(bdev);
759 1 : return;
760 : }
761 0 : break;
762 : default:
763 : /* Examine by all bdev modules with a v2 claim */
764 4 : assert(claim_type_is_v2(bdev->internal.claim_type));
765 : /*
766 : * Removal of tailq nodes while iterating can cause the iteration to jump out of the
767 : * list, perhaps accessing freed memory. Without protection, this could happen
768 : * while the lock is dropped during the examine callback.
769 : */
770 4 : bdev->internal.examine_in_progress++;
771 :
772 9 : TAILQ_FOREACH(claim, &bdev->internal.claim.v2.claims, link) {
773 5 : module = claim->module;
774 :
775 5 : if (module == NULL) {
776 : /* This is a vestigial claim, held by examine_count */
777 0 : continue;
778 : }
779 :
780 5 : if (module->examine_disk == NULL) {
781 0 : continue;
782 : }
783 :
784 5 : spdk_spin_lock(&module->internal.spinlock);
785 5 : module->internal.action_in_progress++;
786 5 : spdk_spin_unlock(&module->internal.spinlock);
787 :
788 : /* Call examine_disk without holding internal.spinlock. */
789 5 : spdk_spin_unlock(&bdev->internal.spinlock);
790 5 : module->examine_disk(bdev);
791 5 : spdk_spin_lock(&bdev->internal.spinlock);
792 5 : }
793 :
794 4 : assert(bdev->internal.examine_in_progress > 0);
795 4 : bdev->internal.examine_in_progress--;
796 4 : if (bdev->internal.examine_in_progress == 0) {
797 : /* Remove any claims that were released during examine_disk */
798 9 : TAILQ_FOREACH_SAFE(claim, &bdev->internal.claim.v2.claims, link, tmpclaim) {
799 5 : if (claim->desc != NULL) {
800 5 : continue;
801 : }
802 :
803 0 : TAILQ_REMOVE(&bdev->internal.claim.v2.claims, claim, link);
804 0 : free(claim);
805 0 : }
806 4 : if (TAILQ_EMPTY(&bdev->internal.claim.v2.claims)) {
807 0 : claim_reset(bdev);
808 0 : }
809 4 : }
810 4 : }
811 :
812 122 : spdk_spin_unlock(&bdev->internal.spinlock);
813 126 : }
814 :
815 : int
816 1 : spdk_bdev_examine(const char *name)
817 : {
818 : struct spdk_bdev *bdev;
819 : struct spdk_bdev_examine_item *item;
820 1 : struct spdk_thread *thread = spdk_get_thread();
821 :
822 1 : if (spdk_unlikely(!spdk_thread_is_app_thread(thread))) {
823 1 : SPDK_ERRLOG("Cannot examine bdev %s on thread %p (%s)\n", name, thread,
824 : thread ? spdk_thread_get_name(thread) : "null");
825 1 : return -EINVAL;
826 : }
827 :
828 0 : if (g_bdev_opts.bdev_auto_examine) {
829 0 : SPDK_ERRLOG("Manual examine is not allowed if auto examine is enabled\n");
830 0 : return -EINVAL;
831 : }
832 :
833 0 : if (bdev_examine_allowlist_check(name)) {
834 0 : SPDK_ERRLOG("Duplicate bdev name for manual examine: %s\n", name);
835 0 : return -EEXIST;
836 : }
837 :
838 0 : item = calloc(1, sizeof(*item));
839 0 : if (!item) {
840 0 : return -ENOMEM;
841 : }
842 0 : item->name = strdup(name);
843 0 : if (!item->name) {
844 0 : free(item);
845 0 : return -ENOMEM;
846 : }
847 0 : TAILQ_INSERT_TAIL(&g_bdev_examine_allowlist, item, link);
848 :
849 0 : bdev = spdk_bdev_get_by_name(name);
850 0 : if (bdev) {
851 0 : bdev_examine(bdev);
852 0 : }
853 0 : return 0;
854 1 : }
855 :
856 : static inline void
857 0 : bdev_examine_allowlist_config_json(struct spdk_json_write_ctx *w)
858 : {
859 : struct spdk_bdev_examine_item *item;
860 0 : TAILQ_FOREACH(item, &g_bdev_examine_allowlist, link) {
861 0 : spdk_json_write_object_begin(w);
862 0 : spdk_json_write_named_string(w, "method", "bdev_examine");
863 0 : spdk_json_write_named_object_begin(w, "params");
864 0 : spdk_json_write_named_string(w, "name", item->name);
865 0 : spdk_json_write_object_end(w);
866 0 : spdk_json_write_object_end(w);
867 0 : }
868 0 : }
869 :
870 : struct spdk_bdev *
871 1 : spdk_bdev_first(void)
872 : {
873 : struct spdk_bdev *bdev;
874 :
875 1 : bdev = TAILQ_FIRST(&g_bdev_mgr.bdevs);
876 1 : if (bdev) {
877 1 : SPDK_DEBUGLOG(bdev, "Starting bdev iteration at %s\n", bdev->name);
878 1 : }
879 :
880 1 : return bdev;
881 : }
882 :
883 : struct spdk_bdev *
884 8 : spdk_bdev_next(struct spdk_bdev *prev)
885 : {
886 : struct spdk_bdev *bdev;
887 :
888 8 : bdev = TAILQ_NEXT(prev, internal.link);
889 8 : if (bdev) {
890 7 : SPDK_DEBUGLOG(bdev, "Continuing bdev iteration at %s\n", bdev->name);
891 7 : }
892 :
893 8 : return bdev;
894 : }
895 :
896 : static struct spdk_bdev *
897 6 : _bdev_next_leaf(struct spdk_bdev *bdev)
898 : {
899 9 : while (bdev != NULL) {
900 8 : if (bdev->internal.claim_type == SPDK_BDEV_CLAIM_NONE) {
901 5 : return bdev;
902 : } else {
903 3 : bdev = TAILQ_NEXT(bdev, internal.link);
904 : }
905 : }
906 :
907 1 : return bdev;
908 6 : }
909 :
910 : struct spdk_bdev *
911 1 : spdk_bdev_first_leaf(void)
912 : {
913 : struct spdk_bdev *bdev;
914 :
915 1 : bdev = _bdev_next_leaf(TAILQ_FIRST(&g_bdev_mgr.bdevs));
916 :
917 1 : if (bdev) {
918 1 : SPDK_DEBUGLOG(bdev, "Starting bdev iteration at %s\n", bdev->name);
919 1 : }
920 :
921 1 : return bdev;
922 : }
923 :
924 : struct spdk_bdev *
925 5 : spdk_bdev_next_leaf(struct spdk_bdev *prev)
926 : {
927 : struct spdk_bdev *bdev;
928 :
929 5 : bdev = _bdev_next_leaf(TAILQ_NEXT(prev, internal.link));
930 :
931 5 : if (bdev) {
932 4 : SPDK_DEBUGLOG(bdev, "Continuing bdev iteration at %s\n", bdev->name);
933 4 : }
934 :
935 5 : return bdev;
936 : }
937 :
938 : static inline bool
939 820 : bdev_io_use_memory_domain(struct spdk_bdev_io *bdev_io)
940 : {
941 820 : return bdev_io->internal.f.has_memory_domain;
942 : }
943 :
944 : static inline bool
945 1549 : bdev_io_use_accel_sequence(struct spdk_bdev_io *bdev_io)
946 : {
947 1549 : return bdev_io->internal.f.has_accel_sequence;
948 : }
949 :
950 : static inline uint32_t
951 373 : bdev_desc_get_block_size(struct spdk_bdev_desc *desc)
952 : {
953 373 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
954 :
955 373 : if (spdk_unlikely(desc->opts.no_metadata)) {
956 0 : return bdev->blocklen - bdev->md_len;
957 : } else {
958 373 : return bdev->blocklen;
959 : }
960 373 : }
961 :
962 : static inline uint32_t
963 110 : bdev_io_get_block_size(struct spdk_bdev_io *bdev_io)
964 : {
965 110 : struct spdk_bdev *bdev = bdev_io->bdev;
966 :
967 110 : if (bdev_io->u.bdev.dif_check_flags & SPDK_DIF_FLAGS_NVME_PRACT) {
968 0 : if (bdev->md_len == spdk_dif_pi_format_get_pi_size(bdev->dif_pi_format)) {
969 0 : return bdev->blocklen - bdev->md_len;
970 : } else {
971 0 : return bdev->blocklen;
972 : }
973 : }
974 :
975 110 : return bdev_desc_get_block_size(bdev_io->internal.desc);
976 110 : }
977 :
978 : static inline void
979 7 : bdev_queue_nomem_io_head(struct spdk_bdev_shared_resource *shared_resource,
980 : struct spdk_bdev_io *bdev_io, enum bdev_io_retry_state state)
981 : {
982 : /* Wait for some of the outstanding I/O to complete before we retry any of the nomem_io.
983 : * Normally we will wait for NOMEM_THRESHOLD_COUNT I/O to complete but for low queue depth
984 : * channels we will instead wait for half to complete.
985 : */
986 7 : shared_resource->nomem_threshold = spdk_max((int64_t)shared_resource->io_outstanding / 2,
987 : (int64_t)shared_resource->io_outstanding - NOMEM_THRESHOLD_COUNT);
988 :
989 7 : assert(state != BDEV_IO_RETRY_STATE_INVALID);
990 7 : bdev_io->internal.retry_state = state;
991 7 : TAILQ_INSERT_HEAD(&shared_resource->nomem_io, bdev_io, internal.link);
992 7 : }
993 :
994 : static inline void
995 43 : bdev_queue_nomem_io_tail(struct spdk_bdev_shared_resource *shared_resource,
996 : struct spdk_bdev_io *bdev_io, enum bdev_io_retry_state state)
997 : {
998 : /* We only queue IOs at the end of the nomem_io queue if they're submitted by the user while
999 : * the queue isn't empty, so we don't need to update the nomem_threshold here */
1000 43 : assert(!TAILQ_EMPTY(&shared_resource->nomem_io));
1001 :
1002 43 : assert(state != BDEV_IO_RETRY_STATE_INVALID);
1003 43 : bdev_io->internal.retry_state = state;
1004 43 : TAILQ_INSERT_TAIL(&shared_resource->nomem_io, bdev_io, internal.link);
1005 43 : }
1006 :
1007 : void
1008 16 : spdk_bdev_io_set_buf(struct spdk_bdev_io *bdev_io, void *buf, size_t len)
1009 : {
1010 : struct iovec *iovs;
1011 :
1012 16 : if (bdev_io->u.bdev.iovs == NULL) {
1013 3 : bdev_io->u.bdev.iovs = &bdev_io->iov;
1014 3 : bdev_io->u.bdev.iovcnt = 1;
1015 3 : }
1016 :
1017 16 : iovs = bdev_io->u.bdev.iovs;
1018 :
1019 16 : assert(iovs != NULL);
1020 16 : assert(bdev_io->u.bdev.iovcnt >= 1);
1021 :
1022 16 : iovs[0].iov_base = buf;
1023 16 : iovs[0].iov_len = len;
1024 16 : }
1025 :
1026 : void
1027 3 : spdk_bdev_io_set_md_buf(struct spdk_bdev_io *bdev_io, void *md_buf, size_t len)
1028 : {
1029 3 : assert((len / spdk_bdev_get_md_size(bdev_io->bdev)) >= bdev_io->u.bdev.num_blocks);
1030 3 : bdev_io->u.bdev.md_buf = md_buf;
1031 3 : }
1032 :
1033 : static bool
1034 167 : _is_buf_allocated(const struct iovec *iovs)
1035 : {
1036 167 : if (iovs == NULL) {
1037 6 : return false;
1038 : }
1039 :
1040 161 : return iovs[0].iov_base != NULL;
1041 167 : }
1042 :
1043 : static bool
1044 50 : _are_iovs_aligned(struct iovec *iovs, int iovcnt, uint32_t alignment)
1045 : {
1046 : int i;
1047 : uintptr_t iov_base;
1048 :
1049 50 : if (spdk_likely(alignment == 1)) {
1050 21 : return true;
1051 : }
1052 :
1053 36 : for (i = 0; i < iovcnt; i++) {
1054 29 : iov_base = (uintptr_t)iovs[i].iov_base;
1055 29 : if ((iov_base & (alignment - 1)) != 0) {
1056 22 : return false;
1057 : }
1058 7 : }
1059 :
1060 7 : return true;
1061 50 : }
1062 :
1063 : static inline bool
1064 895 : bdev_io_needs_metadata(struct spdk_bdev_desc *desc, struct spdk_bdev_io *bdev_io)
1065 : {
1066 1048 : return (bdev_io->bdev->md_len != 0) &&
1067 153 : (desc->opts.no_metadata ||
1068 153 : (bdev_io->u.bdev.dif_check_flags & SPDK_DIF_FLAGS_NVME_PRACT));
1069 : }
1070 :
1071 : static inline bool
1072 852 : bdev_io_needs_sequence_exec(struct spdk_bdev_desc *desc, struct spdk_bdev_io *bdev_io)
1073 : {
1074 852 : if (!bdev_io_use_accel_sequence(bdev_io)) {
1075 852 : return false;
1076 : }
1077 :
1078 : /* For now, we don't allow splitting IOs with an accel sequence and will treat them as if
1079 : * bdev module didn't support accel sequences */
1080 0 : return !desc->accel_sequence_supported[bdev_io->type] || bdev_io->internal.f.split;
1081 852 : }
1082 :
1083 : static inline void
1084 592 : bdev_io_increment_outstanding(struct spdk_bdev_channel *bdev_ch,
1085 : struct spdk_bdev_shared_resource *shared_resource)
1086 : {
1087 592 : bdev_ch->io_outstanding++;
1088 592 : shared_resource->io_outstanding++;
1089 592 : }
1090 :
1091 : static inline void
1092 592 : bdev_io_decrement_outstanding(struct spdk_bdev_channel *bdev_ch,
1093 : struct spdk_bdev_shared_resource *shared_resource)
1094 : {
1095 592 : assert(bdev_ch->io_outstanding > 0);
1096 592 : assert(shared_resource->io_outstanding > 0);
1097 592 : bdev_ch->io_outstanding--;
1098 592 : shared_resource->io_outstanding--;
1099 592 : }
1100 :
1101 : static void
1102 0 : bdev_io_submit_sequence_cb(void *ctx, int status)
1103 : {
1104 0 : struct spdk_bdev_io *bdev_io = ctx;
1105 :
1106 0 : assert(bdev_io_use_accel_sequence(bdev_io));
1107 :
1108 0 : bdev_io->u.bdev.accel_sequence = NULL;
1109 0 : bdev_io->internal.f.has_accel_sequence = false;
1110 :
1111 0 : if (spdk_unlikely(status != 0)) {
1112 0 : SPDK_ERRLOG("Failed to execute accel sequence, status=%d\n", status);
1113 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
1114 0 : bdev_io_complete_unsubmitted(bdev_io);
1115 0 : return;
1116 : }
1117 :
1118 0 : bdev_io_submit(bdev_io);
1119 0 : }
1120 :
1121 : static void
1122 0 : bdev_io_exec_sequence_cb(void *ctx, int status)
1123 : {
1124 0 : struct spdk_bdev_io *bdev_io = ctx;
1125 0 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1126 :
1127 0 : TAILQ_REMOVE(&bdev_io->internal.ch->io_accel_exec, bdev_io, internal.link);
1128 0 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1129 :
1130 0 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1131 0 : bdev_ch_retry_io(ch);
1132 0 : }
1133 :
1134 0 : bdev_io->internal.data_transfer_cpl(bdev_io, status);
1135 0 : }
1136 :
1137 : static void
1138 0 : bdev_io_exec_sequence(struct spdk_bdev_io *bdev_io, void (*cb_fn)(void *ctx, int status))
1139 : {
1140 0 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1141 :
1142 0 : assert(bdev_io_needs_sequence_exec(bdev_io->internal.desc, bdev_io));
1143 0 : assert(bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE || bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
1144 0 : assert(bdev_io_use_accel_sequence(bdev_io));
1145 :
1146 : /* Since the operations are appended during submission, they're in the opposite order than
1147 : * how we want to execute them for reads (i.e. we need to execute the most recently added
1148 : * operation first), so reverse the sequence before executing it.
1149 : */
1150 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_READ) {
1151 0 : spdk_accel_sequence_reverse(bdev_io->internal.accel_sequence);
1152 0 : }
1153 :
1154 0 : TAILQ_INSERT_TAIL(&bdev_io->internal.ch->io_accel_exec, bdev_io, internal.link);
1155 0 : bdev_io_increment_outstanding(ch, ch->shared_resource);
1156 0 : bdev_io->internal.data_transfer_cpl = cb_fn;
1157 :
1158 0 : spdk_accel_sequence_finish(bdev_io->internal.accel_sequence,
1159 0 : bdev_io_exec_sequence_cb, bdev_io);
1160 0 : }
1161 :
1162 : static void
1163 42 : bdev_io_get_buf_complete(struct spdk_bdev_io *bdev_io, bool status)
1164 : {
1165 42 : struct spdk_io_channel *ch = spdk_bdev_io_get_io_channel(bdev_io);
1166 : void *buf;
1167 :
1168 42 : if (spdk_unlikely(bdev_io->internal.get_aux_buf_cb != NULL)) {
1169 0 : buf = bdev_io->internal.buf.ptr;
1170 0 : bdev_io->internal.buf.ptr = NULL;
1171 0 : bdev_io->internal.f.has_buf = false;
1172 0 : bdev_io->internal.get_aux_buf_cb(ch, bdev_io, buf);
1173 0 : bdev_io->internal.get_aux_buf_cb = NULL;
1174 0 : } else {
1175 42 : assert(bdev_io->internal.get_buf_cb != NULL);
1176 42 : bdev_io->internal.get_buf_cb(ch, bdev_io, status);
1177 42 : bdev_io->internal.get_buf_cb = NULL;
1178 : }
1179 42 : }
1180 :
1181 : static void
1182 4 : _bdev_io_pull_buffer_cpl(void *ctx, int rc)
1183 : {
1184 4 : struct spdk_bdev_io *bdev_io = ctx;
1185 :
1186 4 : if (rc) {
1187 0 : SPDK_ERRLOG("Set bounce buffer failed with rc %d\n", rc);
1188 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
1189 0 : }
1190 4 : bdev_io_get_buf_complete(bdev_io, !rc);
1191 4 : }
1192 :
1193 : static void
1194 2 : bdev_io_pull_md_buf_done(void *ctx, int status)
1195 : {
1196 2 : struct spdk_bdev_io *bdev_io = ctx;
1197 2 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1198 :
1199 2 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1200 2 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1201 :
1202 2 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1203 0 : bdev_ch_retry_io(ch);
1204 0 : }
1205 :
1206 2 : assert(bdev_io->internal.data_transfer_cpl);
1207 2 : bdev_io->internal.data_transfer_cpl(bdev_io, status);
1208 2 : }
1209 :
1210 : static void
1211 4 : bdev_io_pull_md_buf(struct spdk_bdev_io *bdev_io)
1212 : {
1213 4 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1214 4 : int rc = 0;
1215 :
1216 4 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
1217 2 : assert(bdev_io->internal.f.has_bounce_buf);
1218 2 : if (bdev_io_use_memory_domain(bdev_io)) {
1219 2 : TAILQ_INSERT_TAIL(&ch->io_memory_domain, bdev_io, internal.link);
1220 2 : bdev_io_increment_outstanding(ch, ch->shared_resource);
1221 4 : rc = spdk_memory_domain_pull_data(bdev_io->internal.memory_domain,
1222 2 : bdev_io->internal.memory_domain_ctx,
1223 2 : &bdev_io->internal.bounce_buf.orig_md_iov, 1,
1224 2 : &bdev_io->internal.bounce_buf.md_iov, 1,
1225 2 : bdev_io_pull_md_buf_done, bdev_io);
1226 2 : if (rc == 0) {
1227 : /* Continue to submit IO in completion callback */
1228 2 : return;
1229 : }
1230 0 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1231 0 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1232 0 : if (rc != -ENOMEM) {
1233 0 : SPDK_ERRLOG("Failed to pull data from memory domain %s, rc %d\n",
1234 : spdk_memory_domain_get_dma_device_id(
1235 : bdev_io->internal.memory_domain), rc);
1236 0 : }
1237 0 : } else {
1238 0 : memcpy(bdev_io->internal.bounce_buf.md_iov.iov_base,
1239 0 : bdev_io->internal.bounce_buf.orig_md_iov.iov_base,
1240 0 : bdev_io->internal.bounce_buf.orig_md_iov.iov_len);
1241 : }
1242 0 : }
1243 :
1244 2 : if (spdk_unlikely(rc == -ENOMEM)) {
1245 0 : bdev_queue_nomem_io_head(ch->shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PULL_MD);
1246 0 : } else {
1247 2 : assert(bdev_io->internal.data_transfer_cpl);
1248 2 : bdev_io->internal.data_transfer_cpl(bdev_io, rc);
1249 : }
1250 4 : }
1251 :
1252 : static void
1253 4 : _bdev_io_pull_bounce_md_buf(struct spdk_bdev_io *bdev_io, void *md_buf, size_t len)
1254 : {
1255 4 : assert(bdev_io->internal.f.has_bounce_buf);
1256 :
1257 : /* save original md_buf */
1258 4 : bdev_io->internal.bounce_buf.orig_md_iov.iov_base = bdev_io->u.bdev.md_buf;
1259 4 : bdev_io->internal.bounce_buf.orig_md_iov.iov_len = len;
1260 4 : bdev_io->internal.bounce_buf.md_iov.iov_base = md_buf;
1261 4 : bdev_io->internal.bounce_buf.md_iov.iov_len = len;
1262 : /* set bounce md_buf */
1263 4 : bdev_io->u.bdev.md_buf = md_buf;
1264 :
1265 4 : bdev_io_pull_md_buf(bdev_io);
1266 4 : }
1267 :
1268 : static void
1269 42 : _bdev_io_set_md_buf(struct spdk_bdev_io *bdev_io)
1270 : {
1271 42 : struct spdk_bdev *bdev = bdev_io->bdev;
1272 : uint64_t md_len;
1273 : void *buf;
1274 :
1275 42 : if (spdk_bdev_is_md_separate(bdev)) {
1276 7 : assert(!bdev_io_use_accel_sequence(bdev_io));
1277 :
1278 7 : buf = (char *)bdev_io->u.bdev.iovs[0].iov_base + bdev_io->u.bdev.iovs[0].iov_len;
1279 7 : md_len = bdev_io->u.bdev.num_blocks * bdev->md_len;
1280 :
1281 7 : assert(((uintptr_t)buf & (spdk_bdev_get_buf_align(bdev) - 1)) == 0);
1282 :
1283 7 : if (bdev_io->u.bdev.md_buf != NULL) {
1284 4 : _bdev_io_pull_bounce_md_buf(bdev_io, buf, md_len);
1285 4 : return;
1286 : } else {
1287 3 : spdk_bdev_io_set_md_buf(bdev_io, buf, md_len);
1288 : }
1289 3 : }
1290 :
1291 38 : bdev_io_get_buf_complete(bdev_io, true);
1292 42 : }
1293 :
1294 : static inline void
1295 26 : bdev_io_pull_data_done(struct spdk_bdev_io *bdev_io, int rc)
1296 : {
1297 26 : if (rc) {
1298 0 : SPDK_ERRLOG("Failed to get data buffer\n");
1299 0 : assert(bdev_io->internal.data_transfer_cpl);
1300 0 : bdev_io->internal.data_transfer_cpl(bdev_io, rc);
1301 0 : return;
1302 : }
1303 :
1304 26 : _bdev_io_set_md_buf(bdev_io);
1305 26 : }
1306 :
1307 : static void
1308 2 : bdev_io_pull_data_done_and_track(void *ctx, int status)
1309 : {
1310 2 : struct spdk_bdev_io *bdev_io = ctx;
1311 2 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1312 :
1313 2 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1314 2 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1315 :
1316 2 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1317 0 : bdev_ch_retry_io(ch);
1318 0 : }
1319 :
1320 2 : bdev_io_pull_data_done(bdev_io, status);
1321 2 : }
1322 :
1323 : static void
1324 27 : bdev_io_pull_data(struct spdk_bdev_io *bdev_io)
1325 : {
1326 27 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1327 27 : struct spdk_bdev_desc *desc = bdev_io->internal.desc;
1328 27 : int rc = 0;
1329 :
1330 27 : assert(bdev_io->internal.f.has_bounce_buf);
1331 :
1332 27 : if (bdev_io_needs_metadata(desc, bdev_io)) {
1333 0 : assert(bdev_io->bdev->md_interleave);
1334 :
1335 0 : bdev_io->u.bdev.dif_check_flags &= ~SPDK_DIF_FLAGS_NVME_PRACT;
1336 :
1337 0 : if (!bdev_io_use_accel_sequence(bdev_io)) {
1338 0 : bdev_io->internal.accel_sequence = NULL;
1339 0 : }
1340 :
1341 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
1342 0 : rc = spdk_accel_append_dif_generate_copy(&bdev_io->internal.accel_sequence, ch->accel_channel,
1343 0 : bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt,
1344 0 : bdev_io->u.bdev.memory_domain,
1345 0 : bdev_io->u.bdev.memory_domain_ctx,
1346 0 : bdev_io->internal.bounce_buf.orig_iovs,
1347 0 : bdev_io->internal.bounce_buf.orig_iovcnt,
1348 0 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain : NULL,
1349 0 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain_ctx : NULL,
1350 0 : bdev_io->u.bdev.num_blocks,
1351 0 : &bdev_io->u.bdev.dif_ctx,
1352 : NULL, NULL);
1353 0 : } else {
1354 0 : assert(bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
1355 0 : rc = spdk_accel_append_dif_verify_copy(&bdev_io->internal.accel_sequence, ch->accel_channel,
1356 0 : bdev_io->internal.bounce_buf.orig_iovs,
1357 0 : bdev_io->internal.bounce_buf.orig_iovcnt,
1358 0 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain : NULL,
1359 0 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain_ctx : NULL,
1360 0 : bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt,
1361 0 : bdev_io->u.bdev.memory_domain,
1362 0 : bdev_io->u.bdev.memory_domain_ctx,
1363 0 : bdev_io->u.bdev.num_blocks,
1364 0 : &bdev_io->u.bdev.dif_ctx,
1365 0 : &bdev_io->u.bdev.dif_err,
1366 : NULL, NULL);
1367 : }
1368 :
1369 0 : if (spdk_likely(rc == 0)) {
1370 0 : bdev_io->internal.f.has_accel_sequence = true;
1371 0 : bdev_io->u.bdev.accel_sequence = bdev_io->internal.accel_sequence;
1372 0 : } else if (rc != -ENOMEM) {
1373 0 : SPDK_ERRLOG("Failed to append generate/verify_copy to accel sequence: %p\n",
1374 : bdev_io->internal.accel_sequence);
1375 0 : }
1376 27 : } else if (bdev_io_needs_sequence_exec(desc, bdev_io) ||
1377 27 : (bdev_io_use_accel_sequence(bdev_io) && bdev_io_use_memory_domain(bdev_io))) {
1378 : /* If we need to exec an accel sequence or the IO uses a memory domain buffer and has a
1379 : * sequence, append a copy operation making accel change the src/dst buffers of the previous
1380 : * operation */
1381 0 : assert(bdev_io_use_accel_sequence(bdev_io));
1382 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
1383 0 : rc = spdk_accel_append_copy(&bdev_io->internal.accel_sequence, ch->accel_channel,
1384 0 : bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt,
1385 : NULL, NULL,
1386 0 : bdev_io->internal.bounce_buf.orig_iovs,
1387 0 : bdev_io->internal.bounce_buf.orig_iovcnt,
1388 0 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain : NULL,
1389 0 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain_ctx : NULL,
1390 : NULL, NULL);
1391 0 : } else {
1392 : /* We need to reverse the src/dst for reads */
1393 0 : assert(bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
1394 0 : rc = spdk_accel_append_copy(&bdev_io->internal.accel_sequence, ch->accel_channel,
1395 0 : bdev_io->internal.bounce_buf.orig_iovs,
1396 0 : bdev_io->internal.bounce_buf.orig_iovcnt,
1397 0 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain : NULL,
1398 0 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain_ctx : NULL,
1399 0 : bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt,
1400 : NULL, NULL, NULL, NULL);
1401 : }
1402 :
1403 0 : if (spdk_unlikely(rc != 0 && rc != -ENOMEM)) {
1404 0 : SPDK_ERRLOG("Failed to append copy to accel sequence: %p\n",
1405 : bdev_io->internal.accel_sequence);
1406 0 : }
1407 27 : } else if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
1408 : /* if this is write path, copy data from original buffer to bounce buffer */
1409 17 : if (bdev_io_use_memory_domain(bdev_io)) {
1410 3 : TAILQ_INSERT_TAIL(&ch->io_memory_domain, bdev_io, internal.link);
1411 3 : bdev_io_increment_outstanding(ch, ch->shared_resource);
1412 6 : rc = spdk_memory_domain_pull_data(bdev_io->internal.memory_domain,
1413 3 : bdev_io->internal.memory_domain_ctx,
1414 3 : bdev_io->internal.bounce_buf.orig_iovs,
1415 3 : (uint32_t)bdev_io->internal.bounce_buf.orig_iovcnt,
1416 3 : bdev_io->u.bdev.iovs, 1,
1417 : bdev_io_pull_data_done_and_track,
1418 3 : bdev_io);
1419 3 : if (rc == 0) {
1420 : /* Continue to submit IO in completion callback */
1421 2 : return;
1422 : }
1423 1 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1424 1 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1425 1 : if (rc != -ENOMEM) {
1426 0 : SPDK_ERRLOG("Failed to pull data from memory domain %s\n",
1427 : spdk_memory_domain_get_dma_device_id(
1428 : bdev_io->internal.memory_domain));
1429 0 : }
1430 1 : } else {
1431 14 : assert(bdev_io->u.bdev.iovcnt == 1);
1432 28 : spdk_copy_iovs_to_buf(bdev_io->u.bdev.iovs[0].iov_base,
1433 14 : bdev_io->u.bdev.iovs[0].iov_len,
1434 14 : bdev_io->internal.bounce_buf.orig_iovs,
1435 14 : bdev_io->internal.bounce_buf.orig_iovcnt);
1436 : }
1437 15 : }
1438 :
1439 25 : if (spdk_unlikely(rc == -ENOMEM)) {
1440 1 : bdev_queue_nomem_io_head(ch->shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PULL);
1441 1 : } else {
1442 24 : bdev_io_pull_data_done(bdev_io, rc);
1443 : }
1444 27 : }
1445 :
1446 : static void
1447 26 : _bdev_io_pull_bounce_data_buf(struct spdk_bdev_io *bdev_io, void *buf, size_t len,
1448 : bdev_copy_bounce_buffer_cpl cpl_cb)
1449 : {
1450 26 : struct spdk_bdev_shared_resource *shared_resource = bdev_io->internal.ch->shared_resource;
1451 :
1452 26 : assert(bdev_io->internal.f.has_bounce_buf == false);
1453 :
1454 26 : bdev_io->internal.data_transfer_cpl = cpl_cb;
1455 26 : bdev_io->internal.f.has_bounce_buf = true;
1456 : /* save original iovec */
1457 26 : bdev_io->internal.bounce_buf.orig_iovs = bdev_io->u.bdev.iovs;
1458 26 : bdev_io->internal.bounce_buf.orig_iovcnt = bdev_io->u.bdev.iovcnt;
1459 : /* zero the other data members */
1460 26 : bdev_io->internal.bounce_buf.iov.iov_base = NULL;
1461 26 : bdev_io->internal.bounce_buf.md_iov.iov_base = NULL;
1462 26 : bdev_io->internal.bounce_buf.orig_md_iov.iov_base = NULL;
1463 : /* set bounce iov */
1464 26 : bdev_io->u.bdev.iovs = &bdev_io->internal.bounce_buf.iov;
1465 26 : bdev_io->u.bdev.iovcnt = 1;
1466 : /* set bounce buffer for this operation */
1467 26 : bdev_io->u.bdev.iovs[0].iov_base = buf;
1468 26 : bdev_io->u.bdev.iovs[0].iov_len = len;
1469 : /* Now we use 1 iov, the split condition could have been changed */
1470 26 : bdev_io->internal.f.split = bdev_io_should_split(bdev_io);
1471 :
1472 26 : if (spdk_unlikely(!TAILQ_EMPTY(&shared_resource->nomem_io))) {
1473 0 : bdev_queue_nomem_io_tail(shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PULL);
1474 0 : } else {
1475 26 : bdev_io_pull_data(bdev_io);
1476 : }
1477 26 : }
1478 :
1479 : static void
1480 42 : _bdev_io_set_buf(struct spdk_bdev_io *bdev_io, void *buf, uint64_t len)
1481 : {
1482 42 : struct spdk_bdev *bdev = bdev_io->bdev;
1483 : bool buf_allocated;
1484 : uint64_t alignment;
1485 : void *aligned_buf;
1486 :
1487 42 : bdev_io->internal.buf.ptr = buf;
1488 42 : bdev_io->internal.f.has_buf = true;
1489 :
1490 42 : if (spdk_unlikely(bdev_io->internal.get_aux_buf_cb != NULL)) {
1491 0 : bdev_io_get_buf_complete(bdev_io, true);
1492 0 : return;
1493 : }
1494 :
1495 42 : alignment = spdk_bdev_get_buf_align(bdev);
1496 42 : buf_allocated = _is_buf_allocated(bdev_io->u.bdev.iovs);
1497 42 : aligned_buf = (void *)(((uintptr_t)buf + (alignment - 1)) & ~(alignment - 1));
1498 :
1499 42 : if (buf_allocated) {
1500 26 : _bdev_io_pull_bounce_data_buf(bdev_io, aligned_buf, len, _bdev_io_pull_buffer_cpl);
1501 : /* Continue in completion callback */
1502 26 : return;
1503 : } else {
1504 16 : spdk_bdev_io_set_buf(bdev_io, aligned_buf, len);
1505 : }
1506 :
1507 16 : _bdev_io_set_md_buf(bdev_io);
1508 42 : }
1509 :
1510 : static inline uint64_t
1511 42 : bdev_io_get_max_buf_len(struct spdk_bdev_io *bdev_io, uint64_t len)
1512 : {
1513 42 : struct spdk_bdev *bdev = bdev_io->bdev;
1514 : uint64_t md_len, alignment;
1515 :
1516 42 : md_len = spdk_bdev_is_md_separate(bdev) ? bdev_io->u.bdev.num_blocks * bdev->md_len : 0;
1517 :
1518 : /* 1 byte alignment needs 0 byte of extra space, 64 bytes alignment needs 63 bytes of extra space, etc. */
1519 42 : alignment = spdk_bdev_get_buf_align(bdev) - 1;
1520 :
1521 42 : return len + alignment + md_len;
1522 : }
1523 :
1524 : static void
1525 42 : bdev_io_put_accel_buf(struct spdk_bdev_io *bdev_io)
1526 : {
1527 42 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1528 :
1529 84 : spdk_accel_put_buf(ch->accel_channel,
1530 42 : bdev_io->internal.buf.ptr,
1531 42 : bdev_io->u.bdev.memory_domain,
1532 42 : bdev_io->u.bdev.memory_domain_ctx);
1533 42 : }
1534 :
1535 : static void
1536 0 : _bdev_io_put_buf(struct spdk_bdev_io *bdev_io, void *buf, uint64_t buf_len)
1537 : {
1538 : struct spdk_bdev_mgmt_channel *ch;
1539 :
1540 0 : ch = bdev_io->internal.ch->shared_resource->mgmt_ch;
1541 0 : spdk_iobuf_put(&ch->iobuf, buf, bdev_io_get_max_buf_len(bdev_io, buf_len));
1542 0 : }
1543 :
1544 : static void
1545 42 : bdev_io_put_buf(struct spdk_bdev_io *bdev_io)
1546 : {
1547 42 : assert(bdev_io->internal.f.has_buf);
1548 :
1549 42 : if (bdev_io->u.bdev.memory_domain == spdk_accel_get_memory_domain()) {
1550 42 : bdev_io_put_accel_buf(bdev_io);
1551 42 : } else {
1552 0 : assert(bdev_io->u.bdev.memory_domain == NULL);
1553 0 : _bdev_io_put_buf(bdev_io, bdev_io->internal.buf.ptr,
1554 0 : bdev_io->internal.buf.len);
1555 : }
1556 42 : bdev_io->internal.buf.ptr = NULL;
1557 42 : bdev_io->internal.f.has_buf = false;
1558 42 : }
1559 :
1560 3 : SPDK_LOG_DEPRECATION_REGISTER(spdk_bdev_io_put_aux_buf,
1561 : "spdk_bdev_io_put_aux_buf is deprecated", "v25.01", 0);
1562 :
1563 : void
1564 0 : spdk_bdev_io_put_aux_buf(struct spdk_bdev_io *bdev_io, void *buf)
1565 : {
1566 0 : uint64_t len = bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen;
1567 :
1568 0 : SPDK_LOG_DEPRECATED(spdk_bdev_io_put_aux_buf);
1569 :
1570 0 : assert(buf != NULL);
1571 0 : _bdev_io_put_buf(bdev_io, buf, len);
1572 0 : }
1573 :
1574 : static inline void
1575 548 : bdev_submit_request(struct spdk_bdev *bdev, struct spdk_io_channel *ioch,
1576 : struct spdk_bdev_io *bdev_io)
1577 : {
1578 : /* After a request is submitted to a bdev module, the ownership of an accel sequence
1579 : * associated with that bdev_io is transferred to the bdev module. So, clear the internal
1580 : * sequence pointer to make sure we won't touch it anymore. */
1581 1014 : if ((bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE ||
1582 548 : bdev_io->type == SPDK_BDEV_IO_TYPE_READ) && bdev_io->u.bdev.accel_sequence != NULL) {
1583 0 : assert(!bdev_io_needs_sequence_exec(bdev_io->internal.desc, bdev_io));
1584 0 : bdev_io->internal.f.has_accel_sequence = false;
1585 0 : }
1586 :
1587 548 : bdev->fn_table->submit_request(ioch, bdev_io);
1588 548 : }
1589 :
1590 : static inline void
1591 10 : bdev_ch_resubmit_io(struct spdk_bdev_shared_resource *shared_resource, struct spdk_bdev_io *bdev_io)
1592 : {
1593 10 : struct spdk_bdev *bdev = bdev_io->bdev;
1594 :
1595 10 : bdev_io_increment_outstanding(bdev_io->internal.ch, shared_resource);
1596 10 : bdev_io->internal.error.nvme.cdw0 = 0;
1597 10 : bdev_io->num_retries++;
1598 10 : bdev_submit_request(bdev, spdk_bdev_io_get_io_channel(bdev_io), bdev_io);
1599 10 : }
1600 :
1601 : static void
1602 63 : bdev_shared_ch_retry_io(struct spdk_bdev_shared_resource *shared_resource)
1603 : {
1604 : struct spdk_bdev_io *bdev_io;
1605 :
1606 63 : if (shared_resource->io_outstanding > shared_resource->nomem_threshold) {
1607 : /*
1608 : * Allow some more I/O to complete before retrying the nomem_io queue.
1609 : * Some drivers (such as nvme) cannot immediately take a new I/O in
1610 : * the context of a completion, because the resources for the I/O are
1611 : * not released until control returns to the bdev poller. Also, we
1612 : * may require several small I/O to complete before a larger I/O
1613 : * (that requires splitting) can be submitted.
1614 : */
1615 58 : return;
1616 : }
1617 :
1618 16 : while (!TAILQ_EMPTY(&shared_resource->nomem_io)) {
1619 12 : bdev_io = TAILQ_FIRST(&shared_resource->nomem_io);
1620 12 : TAILQ_REMOVE(&shared_resource->nomem_io, bdev_io, internal.link);
1621 :
1622 12 : switch (bdev_io->internal.retry_state) {
1623 : case BDEV_IO_RETRY_STATE_SUBMIT:
1624 10 : bdev_ch_resubmit_io(shared_resource, bdev_io);
1625 10 : break;
1626 : case BDEV_IO_RETRY_STATE_PULL:
1627 1 : bdev_io_pull_data(bdev_io);
1628 1 : break;
1629 : case BDEV_IO_RETRY_STATE_PULL_MD:
1630 0 : bdev_io_pull_md_buf(bdev_io);
1631 0 : break;
1632 : case BDEV_IO_RETRY_STATE_PUSH:
1633 1 : bdev_io_push_bounce_data(bdev_io);
1634 1 : break;
1635 : case BDEV_IO_RETRY_STATE_PUSH_MD:
1636 0 : bdev_io_push_bounce_md_buf(bdev_io);
1637 0 : break;
1638 : case BDEV_IO_RETRY_STATE_GET_ACCEL_BUF:
1639 0 : _bdev_io_get_accel_buf(bdev_io);
1640 0 : break;
1641 : default:
1642 0 : assert(0 && "invalid retry state");
1643 : break;
1644 : }
1645 :
1646 12 : if (bdev_io == TAILQ_FIRST(&shared_resource->nomem_io)) {
1647 : /* This IO completed again with NOMEM status, so break the loop and
1648 : * don't try anymore. Note that a bdev_io that fails with NOMEM
1649 : * always gets requeued at the front of the list, to maintain
1650 : * ordering.
1651 : */
1652 1 : break;
1653 : }
1654 : }
1655 63 : }
1656 :
1657 : static void
1658 63 : bdev_ch_retry_io(struct spdk_bdev_channel *bdev_ch)
1659 : {
1660 63 : bdev_shared_ch_retry_io(bdev_ch->shared_resource);
1661 63 : }
1662 :
1663 : static int
1664 0 : bdev_no_mem_poller(void *ctx)
1665 : {
1666 0 : struct spdk_bdev_shared_resource *shared_resource = ctx;
1667 :
1668 0 : spdk_poller_unregister(&shared_resource->nomem_poller);
1669 :
1670 0 : if (!TAILQ_EMPTY(&shared_resource->nomem_io)) {
1671 0 : bdev_shared_ch_retry_io(shared_resource);
1672 0 : }
1673 : /* the retry cb may re-register the poller so double check */
1674 0 : if (!TAILQ_EMPTY(&shared_resource->nomem_io) &&
1675 0 : shared_resource->io_outstanding == 0 && shared_resource->nomem_poller == NULL) {
1676 : /* No IOs were submitted, try again */
1677 0 : shared_resource->nomem_poller = SPDK_POLLER_REGISTER(bdev_no_mem_poller, shared_resource,
1678 : SPDK_BDEV_IO_POLL_INTERVAL_IN_MSEC * 10);
1679 0 : }
1680 :
1681 0 : return SPDK_POLLER_BUSY;
1682 : }
1683 :
1684 : static inline bool
1685 556 : _bdev_io_handle_no_mem(struct spdk_bdev_io *bdev_io, enum bdev_io_retry_state state)
1686 : {
1687 556 : struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
1688 556 : struct spdk_bdev_shared_resource *shared_resource = bdev_ch->shared_resource;
1689 :
1690 556 : if (spdk_unlikely(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_NOMEM)) {
1691 5 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_PENDING;
1692 5 : bdev_queue_nomem_io_head(shared_resource, bdev_io, state);
1693 :
1694 5 : if (shared_resource->io_outstanding == 0 && !shared_resource->nomem_poller) {
1695 : /* Special case when we have nomem IOs and no outstanding IOs which completions
1696 : * could trigger retry of queued IOs
1697 : * Any IOs submitted may trigger retry of queued IOs. This poller handles a case when no
1698 : * new IOs submitted, e.g. qd==1 */
1699 0 : shared_resource->nomem_poller = SPDK_POLLER_REGISTER(bdev_no_mem_poller, shared_resource,
1700 : SPDK_BDEV_IO_POLL_INTERVAL_IN_MSEC * 10);
1701 0 : }
1702 : /* If bdev module completed an I/O that has an accel sequence with NOMEM status, the
1703 : * ownership of that sequence is transferred back to the bdev layer, so we need to
1704 : * restore internal.accel_sequence to make sure that the sequence is handled
1705 : * correctly in case the I/O is later aborted. */
1706 5 : if ((bdev_io->type == SPDK_BDEV_IO_TYPE_READ ||
1707 5 : bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) && bdev_io->u.bdev.accel_sequence) {
1708 0 : assert(!bdev_io_use_accel_sequence(bdev_io));
1709 0 : bdev_io->internal.f.has_accel_sequence = true;
1710 0 : bdev_io->internal.accel_sequence = bdev_io->u.bdev.accel_sequence;
1711 0 : }
1712 :
1713 5 : return true;
1714 : }
1715 :
1716 551 : if (spdk_unlikely(!TAILQ_EMPTY(&shared_resource->nomem_io))) {
1717 63 : bdev_ch_retry_io(bdev_ch);
1718 63 : }
1719 :
1720 551 : return false;
1721 556 : }
1722 :
1723 : static void
1724 26 : _bdev_io_complete_push_bounce_done(void *ctx, int rc)
1725 : {
1726 26 : struct spdk_bdev_io *bdev_io = ctx;
1727 26 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1728 :
1729 26 : if (rc) {
1730 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
1731 0 : }
1732 : /* We want to free the bounce buffer here since we know we're done with it (as opposed
1733 : * to waiting for the conditional free of internal.buf.ptr in spdk_bdev_free_io()).
1734 : */
1735 26 : bdev_io_put_buf(bdev_io);
1736 :
1737 26 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1738 0 : bdev_ch_retry_io(ch);
1739 0 : }
1740 :
1741 : /* Continue with IO completion flow */
1742 26 : bdev_io_complete(bdev_io);
1743 26 : }
1744 :
1745 : static void
1746 2 : bdev_io_push_bounce_md_buf_done(void *ctx, int rc)
1747 : {
1748 2 : struct spdk_bdev_io *bdev_io = ctx;
1749 2 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1750 :
1751 2 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1752 2 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1753 2 : bdev_io->internal.f.has_bounce_buf = false;
1754 :
1755 2 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1756 0 : bdev_ch_retry_io(ch);
1757 0 : }
1758 :
1759 2 : bdev_io->internal.data_transfer_cpl(bdev_io, rc);
1760 2 : }
1761 :
1762 : static inline void
1763 26 : bdev_io_push_bounce_md_buf(struct spdk_bdev_io *bdev_io)
1764 : {
1765 26 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1766 26 : int rc = 0;
1767 :
1768 26 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS);
1769 26 : assert(bdev_io->internal.f.has_bounce_buf);
1770 :
1771 : /* do the same for metadata buffer */
1772 26 : if (spdk_unlikely(bdev_io->internal.bounce_buf.orig_md_iov.iov_base != NULL)) {
1773 4 : assert(spdk_bdev_is_md_separate(bdev_io->bdev));
1774 :
1775 4 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_READ) {
1776 2 : if (bdev_io_use_memory_domain(bdev_io)) {
1777 2 : TAILQ_INSERT_TAIL(&ch->io_memory_domain, bdev_io, internal.link);
1778 2 : bdev_io_increment_outstanding(ch, ch->shared_resource);
1779 : /* If memory domain is used then we need to call async push function */
1780 4 : rc = spdk_memory_domain_push_data(bdev_io->internal.memory_domain,
1781 2 : bdev_io->internal.memory_domain_ctx,
1782 2 : &bdev_io->internal.bounce_buf.orig_md_iov,
1783 2 : (uint32_t)bdev_io->internal.bounce_buf.orig_iovcnt,
1784 2 : &bdev_io->internal.bounce_buf.md_iov, 1,
1785 : bdev_io_push_bounce_md_buf_done,
1786 2 : bdev_io);
1787 2 : if (rc == 0) {
1788 : /* Continue IO completion in async callback */
1789 2 : return;
1790 : }
1791 0 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1792 0 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1793 0 : if (rc != -ENOMEM) {
1794 0 : SPDK_ERRLOG("Failed to push md to memory domain %s\n",
1795 : spdk_memory_domain_get_dma_device_id(
1796 : bdev_io->internal.memory_domain));
1797 0 : }
1798 0 : } else {
1799 0 : memcpy(bdev_io->internal.bounce_buf.orig_md_iov.iov_base, bdev_io->u.bdev.md_buf,
1800 0 : bdev_io->internal.bounce_buf.orig_md_iov.iov_len);
1801 : }
1802 0 : }
1803 2 : }
1804 :
1805 24 : if (spdk_unlikely(rc == -ENOMEM)) {
1806 0 : bdev_queue_nomem_io_head(ch->shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PUSH_MD);
1807 0 : } else {
1808 24 : assert(bdev_io->internal.data_transfer_cpl);
1809 24 : bdev_io->internal.f.has_bounce_buf = false;
1810 24 : bdev_io->internal.data_transfer_cpl(bdev_io, rc);
1811 : }
1812 26 : }
1813 :
1814 : static inline void
1815 26 : bdev_io_push_bounce_data_done(struct spdk_bdev_io *bdev_io, int rc)
1816 : {
1817 26 : assert(bdev_io->internal.data_transfer_cpl);
1818 26 : if (rc) {
1819 0 : bdev_io->internal.data_transfer_cpl(bdev_io, rc);
1820 0 : return;
1821 : }
1822 :
1823 : /* set original buffer for this io */
1824 26 : bdev_io->u.bdev.iovcnt = bdev_io->internal.bounce_buf.orig_iovcnt;
1825 26 : bdev_io->u.bdev.iovs = bdev_io->internal.bounce_buf.orig_iovs;
1826 :
1827 : /* We don't set bdev_io->internal.f.has_bounce_buf to false here because
1828 : * we still need to clear the md buf */
1829 :
1830 26 : bdev_io_push_bounce_md_buf(bdev_io);
1831 26 : }
1832 :
1833 : static void
1834 2 : bdev_io_push_bounce_data_done_and_track(void *ctx, int status)
1835 : {
1836 2 : struct spdk_bdev_io *bdev_io = ctx;
1837 2 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1838 :
1839 2 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1840 2 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1841 :
1842 2 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1843 0 : bdev_ch_retry_io(ch);
1844 0 : }
1845 :
1846 2 : bdev_io_push_bounce_data_done(bdev_io, status);
1847 2 : }
1848 :
1849 : static inline void
1850 27 : bdev_io_push_bounce_data(struct spdk_bdev_io *bdev_io)
1851 : {
1852 27 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1853 27 : int rc = 0;
1854 :
1855 27 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS);
1856 27 : assert(!bdev_io_use_accel_sequence(bdev_io));
1857 27 : assert(bdev_io->internal.f.has_bounce_buf);
1858 :
1859 : /* if this is read path, copy data from bounce buffer to original buffer */
1860 27 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_READ) {
1861 11 : if (bdev_io_use_memory_domain(bdev_io)) {
1862 3 : TAILQ_INSERT_TAIL(&ch->io_memory_domain, bdev_io, internal.link);
1863 3 : bdev_io_increment_outstanding(ch, ch->shared_resource);
1864 : /* If memory domain is used then we need to call async push function */
1865 6 : rc = spdk_memory_domain_push_data(bdev_io->internal.memory_domain,
1866 3 : bdev_io->internal.memory_domain_ctx,
1867 3 : bdev_io->internal.bounce_buf.orig_iovs,
1868 3 : (uint32_t)bdev_io->internal.bounce_buf.orig_iovcnt,
1869 3 : &bdev_io->internal.bounce_buf.iov, 1,
1870 : bdev_io_push_bounce_data_done_and_track,
1871 3 : bdev_io);
1872 3 : if (rc == 0) {
1873 : /* Continue IO completion in async callback */
1874 2 : return;
1875 : }
1876 :
1877 1 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1878 1 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1879 1 : if (rc != -ENOMEM) {
1880 0 : SPDK_ERRLOG("Failed to push data to memory domain %s\n",
1881 : spdk_memory_domain_get_dma_device_id(
1882 : bdev_io->internal.memory_domain));
1883 0 : }
1884 1 : } else {
1885 16 : spdk_copy_buf_to_iovs(bdev_io->internal.bounce_buf.orig_iovs,
1886 8 : bdev_io->internal.bounce_buf.orig_iovcnt,
1887 8 : bdev_io->internal.bounce_buf.iov.iov_base,
1888 8 : bdev_io->internal.bounce_buf.iov.iov_len);
1889 : }
1890 9 : }
1891 :
1892 25 : if (spdk_unlikely(rc == -ENOMEM)) {
1893 1 : bdev_queue_nomem_io_head(ch->shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PUSH);
1894 1 : } else {
1895 24 : bdev_io_push_bounce_data_done(bdev_io, rc);
1896 : }
1897 27 : }
1898 :
1899 : static inline void
1900 26 : _bdev_io_push_bounce_data_buffer(struct spdk_bdev_io *bdev_io, bdev_copy_bounce_buffer_cpl cpl_cb)
1901 : {
1902 26 : bdev_io->internal.data_transfer_cpl = cpl_cb;
1903 26 : bdev_io_push_bounce_data(bdev_io);
1904 26 : }
1905 :
1906 : static void
1907 0 : bdev_io_get_iobuf_cb(struct spdk_iobuf_entry *iobuf, void *buf)
1908 : {
1909 : struct spdk_bdev_io *bdev_io;
1910 :
1911 0 : bdev_io = SPDK_CONTAINEROF(iobuf, struct spdk_bdev_io, internal.iobuf);
1912 0 : _bdev_io_set_buf(bdev_io, buf, bdev_io->internal.buf.len);
1913 0 : }
1914 :
1915 : static void
1916 42 : bdev_io_get_buf(struct spdk_bdev_io *bdev_io, uint64_t len)
1917 : {
1918 : struct spdk_bdev_mgmt_channel *mgmt_ch;
1919 : uint64_t max_len;
1920 : void *buf;
1921 :
1922 42 : assert(spdk_bdev_io_get_thread(bdev_io) == spdk_get_thread());
1923 42 : mgmt_ch = bdev_io->internal.ch->shared_resource->mgmt_ch;
1924 42 : max_len = bdev_io_get_max_buf_len(bdev_io, len);
1925 :
1926 42 : if (spdk_unlikely(max_len > mgmt_ch->iobuf.cache[0].large.bufsize)) {
1927 0 : SPDK_ERRLOG("Length %" PRIu64 " is larger than allowed\n", max_len);
1928 0 : bdev_io_get_buf_complete(bdev_io, false);
1929 0 : return;
1930 : }
1931 :
1932 42 : bdev_io->internal.buf.len = len;
1933 42 : buf = spdk_iobuf_get(&mgmt_ch->iobuf, max_len, &bdev_io->internal.iobuf,
1934 : bdev_io_get_iobuf_cb);
1935 42 : if (buf != NULL) {
1936 42 : _bdev_io_set_buf(bdev_io, buf, len);
1937 42 : }
1938 42 : }
1939 :
1940 : void
1941 56 : spdk_bdev_io_get_buf(struct spdk_bdev_io *bdev_io, spdk_bdev_io_get_buf_cb cb, uint64_t len)
1942 : {
1943 56 : struct spdk_bdev *bdev = bdev_io->bdev;
1944 : uint64_t alignment;
1945 :
1946 56 : assert(cb != NULL);
1947 56 : bdev_io->internal.get_buf_cb = cb;
1948 :
1949 56 : alignment = spdk_bdev_get_buf_align(bdev);
1950 :
1951 56 : if (_is_buf_allocated(bdev_io->u.bdev.iovs) &&
1952 40 : _are_iovs_aligned(bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt, alignment)) {
1953 : /* Buffer already present and aligned */
1954 18 : cb(spdk_bdev_io_get_io_channel(bdev_io), bdev_io, true);
1955 18 : return;
1956 : }
1957 :
1958 38 : bdev_io_get_buf(bdev_io, len);
1959 56 : }
1960 :
1961 : static void
1962 4 : _bdev_io_get_bounce_buf(struct spdk_bdev_io *bdev_io, spdk_bdev_io_get_buf_cb cb,
1963 : uint64_t len)
1964 : {
1965 4 : assert(cb != NULL);
1966 4 : bdev_io->internal.get_buf_cb = cb;
1967 :
1968 4 : bdev_io_get_buf(bdev_io, len);
1969 4 : }
1970 :
1971 : static void
1972 0 : _bdev_io_get_accel_buf(struct spdk_bdev_io *bdev_io)
1973 : {
1974 0 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1975 : void *buf;
1976 : int rc;
1977 :
1978 0 : rc = spdk_accel_get_buf(ch->accel_channel,
1979 0 : bdev_io->internal.buf.len,
1980 : &buf,
1981 0 : &bdev_io->u.bdev.memory_domain,
1982 0 : &bdev_io->u.bdev.memory_domain_ctx);
1983 0 : if (rc != 0) {
1984 0 : bdev_queue_nomem_io_tail(ch->shared_resource, bdev_io,
1985 : BDEV_IO_RETRY_STATE_GET_ACCEL_BUF);
1986 0 : return;
1987 : }
1988 :
1989 0 : _bdev_io_set_buf(bdev_io, buf, bdev_io->internal.buf.len);
1990 0 : }
1991 :
1992 : static inline void
1993 0 : bdev_io_get_accel_buf(struct spdk_bdev_io *bdev_io, spdk_bdev_io_get_buf_cb cb,
1994 : uint64_t len)
1995 : {
1996 0 : bdev_io->internal.buf.len = len;
1997 0 : bdev_io->internal.get_buf_cb = cb;
1998 :
1999 0 : _bdev_io_get_accel_buf(bdev_io);
2000 0 : }
2001 :
2002 3 : SPDK_LOG_DEPRECATION_REGISTER(spdk_bdev_io_get_aux_buf,
2003 : "spdk_bdev_io_get_aux_buf is deprecated", "v25.01", 0);
2004 :
2005 : void
2006 0 : spdk_bdev_io_get_aux_buf(struct spdk_bdev_io *bdev_io, spdk_bdev_io_get_aux_buf_cb cb)
2007 : {
2008 0 : uint64_t len = bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen;
2009 :
2010 0 : SPDK_LOG_DEPRECATED(spdk_bdev_io_get_aux_buf);
2011 :
2012 0 : assert(cb != NULL);
2013 0 : assert(bdev_io->internal.get_aux_buf_cb == NULL);
2014 0 : bdev_io->internal.get_aux_buf_cb = cb;
2015 0 : bdev_io_get_buf(bdev_io, len);
2016 0 : }
2017 :
2018 : static int
2019 65 : bdev_module_get_max_ctx_size(void)
2020 : {
2021 : struct spdk_bdev_module *bdev_module;
2022 65 : int max_bdev_module_size = 0;
2023 :
2024 254 : TAILQ_FOREACH(bdev_module, &g_bdev_mgr.bdev_modules, internal.tailq) {
2025 189 : if (bdev_module->get_ctx_size && bdev_module->get_ctx_size() > max_bdev_module_size) {
2026 64 : max_bdev_module_size = bdev_module->get_ctx_size();
2027 64 : }
2028 189 : }
2029 :
2030 65 : return max_bdev_module_size;
2031 : }
2032 :
2033 : static void
2034 0 : bdev_enable_histogram_config_json(struct spdk_bdev *bdev, struct spdk_json_write_ctx *w)
2035 : {
2036 0 : if (!bdev->internal.histogram_enabled) {
2037 0 : return;
2038 : }
2039 :
2040 0 : spdk_json_write_object_begin(w);
2041 0 : spdk_json_write_named_string(w, "method", "bdev_enable_histogram");
2042 :
2043 0 : spdk_json_write_named_object_begin(w, "params");
2044 0 : spdk_json_write_named_string(w, "name", bdev->name);
2045 :
2046 0 : spdk_json_write_named_bool(w, "enable", bdev->internal.histogram_enabled);
2047 :
2048 0 : if (bdev->internal.histogram_io_type) {
2049 0 : spdk_json_write_named_string(w, "opc",
2050 0 : spdk_bdev_get_io_type_name(bdev->internal.histogram_io_type));
2051 0 : }
2052 :
2053 0 : spdk_json_write_object_end(w);
2054 :
2055 0 : spdk_json_write_object_end(w);
2056 0 : }
2057 :
2058 : static void
2059 0 : bdev_qos_config_json(struct spdk_bdev *bdev, struct spdk_json_write_ctx *w)
2060 : {
2061 : int i;
2062 0 : struct spdk_bdev_qos *qos = bdev->internal.qos;
2063 : uint64_t limits[SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES];
2064 :
2065 0 : if (!qos) {
2066 0 : return;
2067 : }
2068 :
2069 0 : spdk_bdev_get_qos_rate_limits(bdev, limits);
2070 :
2071 0 : spdk_json_write_object_begin(w);
2072 0 : spdk_json_write_named_string(w, "method", "bdev_set_qos_limit");
2073 :
2074 0 : spdk_json_write_named_object_begin(w, "params");
2075 0 : spdk_json_write_named_string(w, "name", bdev->name);
2076 0 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
2077 0 : if (limits[i] > 0) {
2078 0 : spdk_json_write_named_uint64(w, qos_rpc_type[i], limits[i]);
2079 0 : }
2080 0 : }
2081 0 : spdk_json_write_object_end(w);
2082 :
2083 0 : spdk_json_write_object_end(w);
2084 0 : }
2085 :
2086 : void
2087 0 : spdk_bdev_subsystem_config_json(struct spdk_json_write_ctx *w)
2088 : {
2089 : struct spdk_bdev_module *bdev_module;
2090 : struct spdk_bdev *bdev;
2091 :
2092 0 : assert(w != NULL);
2093 :
2094 0 : spdk_json_write_array_begin(w);
2095 :
2096 0 : spdk_json_write_object_begin(w);
2097 0 : spdk_json_write_named_string(w, "method", "bdev_set_options");
2098 0 : spdk_json_write_named_object_begin(w, "params");
2099 0 : spdk_json_write_named_uint32(w, "bdev_io_pool_size", g_bdev_opts.bdev_io_pool_size);
2100 0 : spdk_json_write_named_uint32(w, "bdev_io_cache_size", g_bdev_opts.bdev_io_cache_size);
2101 0 : spdk_json_write_named_bool(w, "bdev_auto_examine", g_bdev_opts.bdev_auto_examine);
2102 0 : spdk_json_write_named_uint32(w, "iobuf_small_cache_size", g_bdev_opts.iobuf_small_cache_size);
2103 0 : spdk_json_write_named_uint32(w, "iobuf_large_cache_size", g_bdev_opts.iobuf_large_cache_size);
2104 0 : spdk_json_write_object_end(w);
2105 0 : spdk_json_write_object_end(w);
2106 :
2107 0 : bdev_examine_allowlist_config_json(w);
2108 :
2109 0 : TAILQ_FOREACH(bdev_module, &g_bdev_mgr.bdev_modules, internal.tailq) {
2110 0 : if (bdev_module->config_json) {
2111 0 : bdev_module->config_json(w);
2112 0 : }
2113 0 : }
2114 :
2115 0 : spdk_spin_lock(&g_bdev_mgr.spinlock);
2116 :
2117 0 : TAILQ_FOREACH(bdev, &g_bdev_mgr.bdevs, internal.link) {
2118 0 : if (bdev->fn_table->write_config_json) {
2119 0 : bdev->fn_table->write_config_json(bdev, w);
2120 0 : }
2121 :
2122 0 : bdev_qos_config_json(bdev, w);
2123 0 : bdev_enable_histogram_config_json(bdev, w);
2124 0 : }
2125 :
2126 0 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
2127 :
2128 : /* This has to be last RPC in array to make sure all bdevs finished examine */
2129 0 : spdk_json_write_object_begin(w);
2130 0 : spdk_json_write_named_string(w, "method", "bdev_wait_for_examine");
2131 0 : spdk_json_write_object_end(w);
2132 :
2133 0 : spdk_json_write_array_end(w);
2134 0 : }
2135 :
2136 : static void
2137 71 : bdev_mgmt_channel_destroy(void *io_device, void *ctx_buf)
2138 : {
2139 71 : struct spdk_bdev_mgmt_channel *ch = ctx_buf;
2140 : struct spdk_bdev_io *bdev_io;
2141 :
2142 71 : spdk_iobuf_channel_fini(&ch->iobuf);
2143 :
2144 9969 : while (!STAILQ_EMPTY(&ch->per_thread_cache)) {
2145 9898 : bdev_io = STAILQ_FIRST(&ch->per_thread_cache);
2146 9898 : STAILQ_REMOVE_HEAD(&ch->per_thread_cache, internal.buf_link);
2147 9898 : ch->per_thread_cache_count--;
2148 9898 : spdk_mempool_put(g_bdev_mgr.bdev_io_pool, (void *)bdev_io);
2149 : }
2150 :
2151 71 : assert(ch->per_thread_cache_count == 0);
2152 71 : }
2153 :
2154 : static int
2155 71 : bdev_mgmt_channel_create(void *io_device, void *ctx_buf)
2156 : {
2157 71 : struct spdk_bdev_mgmt_channel *ch = ctx_buf;
2158 : struct spdk_bdev_io *bdev_io;
2159 : uint32_t i;
2160 : int rc;
2161 :
2162 142 : rc = spdk_iobuf_channel_init(&ch->iobuf, "bdev",
2163 71 : g_bdev_opts.iobuf_small_cache_size,
2164 71 : g_bdev_opts.iobuf_large_cache_size);
2165 71 : if (rc != 0) {
2166 0 : SPDK_ERRLOG("Failed to create iobuf channel: %s\n", spdk_strerror(-rc));
2167 0 : return -1;
2168 : }
2169 :
2170 71 : STAILQ_INIT(&ch->per_thread_cache);
2171 71 : ch->bdev_io_cache_size = g_bdev_opts.bdev_io_cache_size;
2172 :
2173 : /* Pre-populate bdev_io cache to ensure this thread cannot be starved. */
2174 71 : ch->per_thread_cache_count = 0;
2175 9969 : for (i = 0; i < ch->bdev_io_cache_size; i++) {
2176 9898 : bdev_io = spdk_mempool_get(g_bdev_mgr.bdev_io_pool);
2177 9898 : if (bdev_io == NULL) {
2178 0 : SPDK_ERRLOG("You need to increase bdev_io_pool_size using bdev_set_options RPC.\n");
2179 0 : assert(false);
2180 : bdev_mgmt_channel_destroy(io_device, ctx_buf);
2181 : return -1;
2182 : }
2183 9898 : ch->per_thread_cache_count++;
2184 9898 : STAILQ_INSERT_HEAD(&ch->per_thread_cache, bdev_io, internal.buf_link);
2185 9898 : }
2186 :
2187 71 : TAILQ_INIT(&ch->shared_resources);
2188 71 : TAILQ_INIT(&ch->io_wait_queue);
2189 :
2190 71 : return 0;
2191 71 : }
2192 :
2193 : static void
2194 65 : bdev_init_complete(int rc)
2195 : {
2196 65 : spdk_bdev_init_cb cb_fn = g_init_cb_fn;
2197 65 : void *cb_arg = g_init_cb_arg;
2198 : struct spdk_bdev_module *m;
2199 :
2200 65 : g_bdev_mgr.init_complete = true;
2201 65 : g_init_cb_fn = NULL;
2202 65 : g_init_cb_arg = NULL;
2203 :
2204 : /*
2205 : * For modules that need to know when subsystem init is complete,
2206 : * inform them now.
2207 : */
2208 65 : if (rc == 0) {
2209 254 : TAILQ_FOREACH(m, &g_bdev_mgr.bdev_modules, internal.tailq) {
2210 189 : if (m->init_complete) {
2211 23 : m->init_complete();
2212 23 : }
2213 189 : }
2214 65 : }
2215 :
2216 65 : cb_fn(cb_arg, rc);
2217 65 : }
2218 :
2219 : static bool
2220 258 : bdev_module_all_actions_completed(void)
2221 : {
2222 : struct spdk_bdev_module *m;
2223 :
2224 1022 : TAILQ_FOREACH(m, &g_bdev_mgr.bdev_modules, internal.tailq) {
2225 764 : if (m->internal.action_in_progress > 0) {
2226 0 : return false;
2227 : }
2228 764 : }
2229 258 : return true;
2230 258 : }
2231 :
2232 : static void
2233 628 : bdev_module_action_complete(void)
2234 : {
2235 : /*
2236 : * Don't finish bdev subsystem initialization if
2237 : * module pre-initialization is still in progress, or
2238 : * the subsystem been already initialized.
2239 : */
2240 628 : if (!g_bdev_mgr.module_init_complete || g_bdev_mgr.init_complete) {
2241 563 : return;
2242 : }
2243 :
2244 : /*
2245 : * Check all bdev modules for inits/examinations in progress. If any
2246 : * exist, return immediately since we cannot finish bdev subsystem
2247 : * initialization until all are completed.
2248 : */
2249 65 : if (!bdev_module_all_actions_completed()) {
2250 0 : return;
2251 : }
2252 :
2253 : /*
2254 : * Modules already finished initialization - now that all
2255 : * the bdev modules have finished their asynchronous I/O
2256 : * processing, the entire bdev layer can be marked as complete.
2257 : */
2258 65 : bdev_init_complete(0);
2259 628 : }
2260 :
2261 : static void
2262 563 : bdev_module_action_done(struct spdk_bdev_module *module)
2263 : {
2264 563 : spdk_spin_lock(&module->internal.spinlock);
2265 563 : assert(module->internal.action_in_progress > 0);
2266 563 : module->internal.action_in_progress--;
2267 563 : spdk_spin_unlock(&module->internal.spinlock);
2268 563 : bdev_module_action_complete();
2269 563 : }
2270 :
2271 : void
2272 65 : spdk_bdev_module_init_done(struct spdk_bdev_module *module)
2273 : {
2274 65 : assert(module->async_init);
2275 65 : bdev_module_action_done(module);
2276 65 : }
2277 :
2278 : void
2279 498 : spdk_bdev_module_examine_done(struct spdk_bdev_module *module)
2280 : {
2281 498 : bdev_module_action_done(module);
2282 498 : }
2283 :
2284 : /** The last initialized bdev module */
2285 : static struct spdk_bdev_module *g_resume_bdev_module = NULL;
2286 :
2287 : static void
2288 0 : bdev_init_failed(void *cb_arg)
2289 : {
2290 0 : struct spdk_bdev_module *module = cb_arg;
2291 :
2292 0 : spdk_spin_lock(&module->internal.spinlock);
2293 0 : assert(module->internal.action_in_progress > 0);
2294 0 : module->internal.action_in_progress--;
2295 0 : spdk_spin_unlock(&module->internal.spinlock);
2296 0 : bdev_init_complete(-1);
2297 0 : }
2298 :
2299 : static int
2300 65 : bdev_modules_init(void)
2301 : {
2302 : struct spdk_bdev_module *module;
2303 65 : int rc = 0;
2304 :
2305 254 : TAILQ_FOREACH(module, &g_bdev_mgr.bdev_modules, internal.tailq) {
2306 189 : g_resume_bdev_module = module;
2307 189 : if (module->async_init) {
2308 65 : spdk_spin_lock(&module->internal.spinlock);
2309 65 : module->internal.action_in_progress = 1;
2310 65 : spdk_spin_unlock(&module->internal.spinlock);
2311 65 : }
2312 189 : rc = module->module_init();
2313 189 : if (rc != 0) {
2314 : /* Bump action_in_progress to prevent other modules from completion of modules_init
2315 : * Send message to defer application shutdown until resources are cleaned up */
2316 0 : spdk_spin_lock(&module->internal.spinlock);
2317 0 : module->internal.action_in_progress = 1;
2318 0 : spdk_spin_unlock(&module->internal.spinlock);
2319 0 : spdk_thread_send_msg(spdk_get_thread(), bdev_init_failed, module);
2320 0 : return rc;
2321 : }
2322 189 : }
2323 :
2324 65 : g_resume_bdev_module = NULL;
2325 65 : return 0;
2326 65 : }
2327 :
2328 : void
2329 65 : spdk_bdev_initialize(spdk_bdev_init_cb cb_fn, void *cb_arg)
2330 : {
2331 65 : int rc = 0;
2332 : char mempool_name[32];
2333 :
2334 65 : assert(cb_fn != NULL);
2335 :
2336 65 : g_init_cb_fn = cb_fn;
2337 65 : g_init_cb_arg = cb_arg;
2338 :
2339 65 : spdk_notify_type_register("bdev_register");
2340 65 : spdk_notify_type_register("bdev_unregister");
2341 :
2342 65 : snprintf(mempool_name, sizeof(mempool_name), "bdev_io_%d", getpid());
2343 :
2344 65 : rc = spdk_iobuf_register_module("bdev");
2345 65 : if (rc != 0) {
2346 0 : SPDK_ERRLOG("could not register bdev iobuf module: %s\n", spdk_strerror(-rc));
2347 0 : bdev_init_complete(-1);
2348 0 : return;
2349 : }
2350 :
2351 130 : g_bdev_mgr.bdev_io_pool = spdk_mempool_create(mempool_name,
2352 65 : g_bdev_opts.bdev_io_pool_size,
2353 65 : sizeof(struct spdk_bdev_io) +
2354 65 : bdev_module_get_max_ctx_size(),
2355 : 0,
2356 : SPDK_ENV_NUMA_ID_ANY);
2357 :
2358 65 : if (g_bdev_mgr.bdev_io_pool == NULL) {
2359 0 : SPDK_ERRLOG("could not allocate spdk_bdev_io pool\n");
2360 0 : bdev_init_complete(-1);
2361 0 : return;
2362 : }
2363 :
2364 65 : g_bdev_mgr.zero_buffer = spdk_zmalloc(ZERO_BUFFER_SIZE, ZERO_BUFFER_SIZE,
2365 : NULL, SPDK_ENV_LCORE_ID_ANY, SPDK_MALLOC_DMA);
2366 65 : if (!g_bdev_mgr.zero_buffer) {
2367 0 : SPDK_ERRLOG("create bdev zero buffer failed\n");
2368 0 : bdev_init_complete(-1);
2369 0 : return;
2370 : }
2371 :
2372 : #ifdef SPDK_CONFIG_VTUNE
2373 : g_bdev_mgr.domain = __itt_domain_create("spdk_bdev");
2374 : #endif
2375 :
2376 65 : spdk_io_device_register(&g_bdev_mgr, bdev_mgmt_channel_create,
2377 : bdev_mgmt_channel_destroy,
2378 : sizeof(struct spdk_bdev_mgmt_channel),
2379 : "bdev_mgr");
2380 :
2381 65 : rc = bdev_modules_init();
2382 65 : g_bdev_mgr.module_init_complete = true;
2383 65 : if (rc != 0) {
2384 0 : SPDK_ERRLOG("bdev modules init failed\n");
2385 0 : return;
2386 : }
2387 :
2388 65 : bdev_module_action_complete();
2389 65 : }
2390 :
2391 : static void
2392 65 : bdev_mgr_unregister_cb(void *io_device)
2393 : {
2394 65 : spdk_bdev_fini_cb cb_fn = g_fini_cb_fn;
2395 :
2396 65 : if (g_bdev_mgr.bdev_io_pool) {
2397 65 : if (spdk_mempool_count(g_bdev_mgr.bdev_io_pool) != g_bdev_opts.bdev_io_pool_size) {
2398 0 : SPDK_ERRLOG("bdev IO pool count is %zu but should be %u\n",
2399 : spdk_mempool_count(g_bdev_mgr.bdev_io_pool),
2400 : g_bdev_opts.bdev_io_pool_size);
2401 0 : }
2402 :
2403 65 : spdk_mempool_free(g_bdev_mgr.bdev_io_pool);
2404 65 : }
2405 :
2406 65 : spdk_free(g_bdev_mgr.zero_buffer);
2407 :
2408 65 : bdev_examine_allowlist_free();
2409 :
2410 65 : cb_fn(g_fini_cb_arg);
2411 65 : g_fini_cb_fn = NULL;
2412 65 : g_fini_cb_arg = NULL;
2413 65 : g_bdev_mgr.init_complete = false;
2414 65 : g_bdev_mgr.module_init_complete = false;
2415 65 : }
2416 :
2417 : static void
2418 65 : bdev_module_fini_iter(void *arg)
2419 : {
2420 : struct spdk_bdev_module *bdev_module;
2421 :
2422 : /* FIXME: Handling initialization failures is broken now,
2423 : * so we won't even try cleaning up after successfully
2424 : * initialized modules. if module_init_complete is false,
2425 : * just call spdk_bdev_mgr_unregister_cb
2426 : */
2427 65 : if (!g_bdev_mgr.module_init_complete) {
2428 0 : bdev_mgr_unregister_cb(NULL);
2429 0 : return;
2430 : }
2431 :
2432 : /* Start iterating from the last touched module */
2433 65 : if (!g_resume_bdev_module) {
2434 65 : bdev_module = TAILQ_LAST(&g_bdev_mgr.bdev_modules, bdev_module_list);
2435 65 : } else {
2436 0 : bdev_module = TAILQ_PREV(g_resume_bdev_module, bdev_module_list,
2437 : internal.tailq);
2438 : }
2439 :
2440 254 : while (bdev_module) {
2441 189 : if (bdev_module->async_fini) {
2442 : /* Save our place so we can resume later. We must
2443 : * save the variable here, before calling module_fini()
2444 : * below, because in some cases the module may immediately
2445 : * call spdk_bdev_module_fini_done() and re-enter
2446 : * this function to continue iterating. */
2447 0 : g_resume_bdev_module = bdev_module;
2448 0 : }
2449 :
2450 189 : if (bdev_module->module_fini) {
2451 189 : bdev_module->module_fini();
2452 189 : }
2453 :
2454 189 : if (bdev_module->async_fini) {
2455 0 : return;
2456 : }
2457 :
2458 189 : bdev_module = TAILQ_PREV(bdev_module, bdev_module_list,
2459 : internal.tailq);
2460 : }
2461 :
2462 65 : g_resume_bdev_module = NULL;
2463 65 : spdk_io_device_unregister(&g_bdev_mgr, bdev_mgr_unregister_cb);
2464 65 : }
2465 :
2466 : void
2467 0 : spdk_bdev_module_fini_done(void)
2468 : {
2469 0 : if (spdk_get_thread() != g_fini_thread) {
2470 0 : spdk_thread_send_msg(g_fini_thread, bdev_module_fini_iter, NULL);
2471 0 : } else {
2472 0 : bdev_module_fini_iter(NULL);
2473 : }
2474 0 : }
2475 :
2476 : static void
2477 65 : bdev_finish_unregister_bdevs_iter(void *cb_arg, int bdeverrno)
2478 : {
2479 65 : struct spdk_bdev *bdev = cb_arg;
2480 :
2481 65 : if (bdeverrno && bdev) {
2482 0 : SPDK_WARNLOG("Unable to unregister bdev '%s' during spdk_bdev_finish()\n",
2483 : bdev->name);
2484 :
2485 : /*
2486 : * Since the call to spdk_bdev_unregister() failed, we have no way to free this
2487 : * bdev; try to continue by manually removing this bdev from the list and continue
2488 : * with the next bdev in the list.
2489 : */
2490 0 : TAILQ_REMOVE(&g_bdev_mgr.bdevs, bdev, internal.link);
2491 0 : }
2492 :
2493 65 : if (TAILQ_EMPTY(&g_bdev_mgr.bdevs)) {
2494 65 : SPDK_DEBUGLOG(bdev, "Done unregistering bdevs\n");
2495 : /*
2496 : * Bdev module finish need to be deferred as we might be in the middle of some context
2497 : * (like bdev part free) that will use this bdev (or private bdev driver ctx data)
2498 : * after returning.
2499 : */
2500 65 : spdk_thread_send_msg(spdk_get_thread(), bdev_module_fini_iter, NULL);
2501 65 : return;
2502 : }
2503 :
2504 : /*
2505 : * Unregister last unclaimed bdev in the list, to ensure that bdev subsystem
2506 : * shutdown proceeds top-down. The goal is to give virtual bdevs an opportunity
2507 : * to detect clean shutdown as opposed to run-time hot removal of the underlying
2508 : * base bdevs.
2509 : *
2510 : * Also, walk the list in the reverse order.
2511 : */
2512 0 : for (bdev = TAILQ_LAST(&g_bdev_mgr.bdevs, spdk_bdev_list);
2513 0 : bdev; bdev = TAILQ_PREV(bdev, spdk_bdev_list, internal.link)) {
2514 0 : spdk_spin_lock(&bdev->internal.spinlock);
2515 0 : if (bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE) {
2516 0 : LOG_ALREADY_CLAIMED_DEBUG("claimed, skipping", bdev);
2517 0 : spdk_spin_unlock(&bdev->internal.spinlock);
2518 0 : continue;
2519 : }
2520 0 : spdk_spin_unlock(&bdev->internal.spinlock);
2521 :
2522 0 : SPDK_DEBUGLOG(bdev, "Unregistering bdev '%s'\n", bdev->name);
2523 0 : spdk_bdev_unregister(bdev, bdev_finish_unregister_bdevs_iter, bdev);
2524 0 : return;
2525 : }
2526 :
2527 : /*
2528 : * If any bdev fails to unclaim underlying bdev properly, we may face the
2529 : * case of bdev list consisting of claimed bdevs only (if claims are managed
2530 : * correctly, this would mean there's a loop in the claims graph which is
2531 : * clearly impossible). Warn and unregister last bdev on the list then.
2532 : */
2533 0 : for (bdev = TAILQ_LAST(&g_bdev_mgr.bdevs, spdk_bdev_list);
2534 0 : bdev; bdev = TAILQ_PREV(bdev, spdk_bdev_list, internal.link)) {
2535 0 : SPDK_WARNLOG("Unregistering claimed bdev '%s'!\n", bdev->name);
2536 0 : spdk_bdev_unregister(bdev, bdev_finish_unregister_bdevs_iter, bdev);
2537 0 : return;
2538 : }
2539 65 : }
2540 :
2541 : static void
2542 65 : bdev_module_fini_start_iter(void *arg)
2543 : {
2544 : struct spdk_bdev_module *bdev_module;
2545 :
2546 65 : if (!g_resume_bdev_module) {
2547 65 : bdev_module = TAILQ_LAST(&g_bdev_mgr.bdev_modules, bdev_module_list);
2548 65 : } else {
2549 0 : bdev_module = TAILQ_PREV(g_resume_bdev_module, bdev_module_list, internal.tailq);
2550 : }
2551 :
2552 254 : while (bdev_module) {
2553 189 : if (bdev_module->async_fini_start) {
2554 : /* Save our place so we can resume later. We must
2555 : * save the variable here, before calling fini_start()
2556 : * below, because in some cases the module may immediately
2557 : * call spdk_bdev_module_fini_start_done() and re-enter
2558 : * this function to continue iterating. */
2559 0 : g_resume_bdev_module = bdev_module;
2560 0 : }
2561 :
2562 189 : if (bdev_module->fini_start) {
2563 23 : bdev_module->fini_start();
2564 23 : }
2565 :
2566 189 : if (bdev_module->async_fini_start) {
2567 0 : return;
2568 : }
2569 :
2570 189 : bdev_module = TAILQ_PREV(bdev_module, bdev_module_list, internal.tailq);
2571 : }
2572 :
2573 65 : g_resume_bdev_module = NULL;
2574 :
2575 65 : bdev_finish_unregister_bdevs_iter(NULL, 0);
2576 65 : }
2577 :
2578 : void
2579 0 : spdk_bdev_module_fini_start_done(void)
2580 : {
2581 0 : if (spdk_get_thread() != g_fini_thread) {
2582 0 : spdk_thread_send_msg(g_fini_thread, bdev_module_fini_start_iter, NULL);
2583 0 : } else {
2584 0 : bdev_module_fini_start_iter(NULL);
2585 : }
2586 0 : }
2587 :
2588 : static void
2589 65 : bdev_finish_wait_for_examine_done(void *cb_arg)
2590 : {
2591 65 : bdev_module_fini_start_iter(NULL);
2592 65 : }
2593 :
2594 : static void bdev_open_async_fini(void);
2595 :
2596 : void
2597 65 : spdk_bdev_finish(spdk_bdev_fini_cb cb_fn, void *cb_arg)
2598 : {
2599 : int rc;
2600 :
2601 65 : assert(cb_fn != NULL);
2602 :
2603 65 : g_fini_thread = spdk_get_thread();
2604 :
2605 65 : g_fini_cb_fn = cb_fn;
2606 65 : g_fini_cb_arg = cb_arg;
2607 :
2608 65 : bdev_open_async_fini();
2609 :
2610 65 : rc = spdk_bdev_wait_for_examine(bdev_finish_wait_for_examine_done, NULL);
2611 65 : if (rc != 0) {
2612 0 : SPDK_ERRLOG("wait_for_examine failed: %s\n", spdk_strerror(-rc));
2613 0 : bdev_finish_wait_for_examine_done(NULL);
2614 0 : }
2615 65 : }
2616 :
2617 : struct spdk_bdev_io *
2618 697 : bdev_channel_get_io(struct spdk_bdev_channel *channel)
2619 : {
2620 697 : struct spdk_bdev_mgmt_channel *ch = channel->shared_resource->mgmt_ch;
2621 : struct spdk_bdev_io *bdev_io;
2622 :
2623 697 : if (ch->per_thread_cache_count > 0) {
2624 637 : bdev_io = STAILQ_FIRST(&ch->per_thread_cache);
2625 637 : STAILQ_REMOVE_HEAD(&ch->per_thread_cache, internal.buf_link);
2626 637 : ch->per_thread_cache_count--;
2627 697 : } else if (spdk_unlikely(!TAILQ_EMPTY(&ch->io_wait_queue))) {
2628 : /*
2629 : * Don't try to look for bdev_ios in the global pool if there are
2630 : * waiters on bdev_ios - we don't want this caller to jump the line.
2631 : */
2632 0 : bdev_io = NULL;
2633 0 : } else {
2634 60 : bdev_io = spdk_mempool_get(g_bdev_mgr.bdev_io_pool);
2635 : }
2636 :
2637 697 : return bdev_io;
2638 : }
2639 :
2640 : void
2641 691 : spdk_bdev_free_io(struct spdk_bdev_io *bdev_io)
2642 : {
2643 : struct spdk_bdev_mgmt_channel *ch;
2644 :
2645 691 : assert(bdev_io != NULL);
2646 691 : assert(bdev_io->internal.status != SPDK_BDEV_IO_STATUS_PENDING);
2647 :
2648 691 : ch = bdev_io->internal.ch->shared_resource->mgmt_ch;
2649 :
2650 691 : if (bdev_io->internal.f.has_buf) {
2651 16 : bdev_io_put_buf(bdev_io);
2652 16 : }
2653 :
2654 691 : if (ch->per_thread_cache_count < ch->bdev_io_cache_size) {
2655 637 : ch->per_thread_cache_count++;
2656 637 : STAILQ_INSERT_HEAD(&ch->per_thread_cache, bdev_io, internal.buf_link);
2657 641 : while (ch->per_thread_cache_count > 0 && !TAILQ_EMPTY(&ch->io_wait_queue)) {
2658 : struct spdk_bdev_io_wait_entry *entry;
2659 :
2660 4 : entry = TAILQ_FIRST(&ch->io_wait_queue);
2661 4 : TAILQ_REMOVE(&ch->io_wait_queue, entry, link);
2662 4 : entry->cb_fn(entry->cb_arg);
2663 : }
2664 637 : } else {
2665 : /* We should never have a full cache with entries on the io wait queue. */
2666 54 : assert(TAILQ_EMPTY(&ch->io_wait_queue));
2667 54 : spdk_mempool_put(g_bdev_mgr.bdev_io_pool, (void *)bdev_io);
2668 : }
2669 691 : }
2670 :
2671 : static bool
2672 72 : bdev_qos_is_iops_rate_limit(enum spdk_bdev_qos_rate_limit_type limit)
2673 : {
2674 72 : assert(limit != SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES);
2675 :
2676 72 : switch (limit) {
2677 : case SPDK_BDEV_QOS_RW_IOPS_RATE_LIMIT:
2678 18 : return true;
2679 : case SPDK_BDEV_QOS_RW_BPS_RATE_LIMIT:
2680 : case SPDK_BDEV_QOS_R_BPS_RATE_LIMIT:
2681 : case SPDK_BDEV_QOS_W_BPS_RATE_LIMIT:
2682 54 : return false;
2683 0 : case SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES:
2684 : default:
2685 0 : return false;
2686 : }
2687 72 : }
2688 :
2689 : static bool
2690 25 : bdev_qos_io_to_limit(struct spdk_bdev_io *bdev_io)
2691 : {
2692 25 : switch (bdev_io->type) {
2693 : case SPDK_BDEV_IO_TYPE_NVME_IO:
2694 : case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
2695 : case SPDK_BDEV_IO_TYPE_READ:
2696 : case SPDK_BDEV_IO_TYPE_WRITE:
2697 23 : return true;
2698 : case SPDK_BDEV_IO_TYPE_ZCOPY:
2699 0 : if (bdev_io->u.bdev.zcopy.start) {
2700 0 : return true;
2701 : } else {
2702 0 : return false;
2703 : }
2704 : default:
2705 2 : return false;
2706 : }
2707 25 : }
2708 :
2709 : static bool
2710 33 : bdev_is_read_io(struct spdk_bdev_io *bdev_io)
2711 : {
2712 33 : switch (bdev_io->type) {
2713 : case SPDK_BDEV_IO_TYPE_NVME_IO:
2714 : case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
2715 : /* Bit 1 (0x2) set for read operation */
2716 0 : if (bdev_io->u.nvme_passthru.cmd.opc & SPDK_NVME_OPC_READ) {
2717 0 : return true;
2718 : } else {
2719 0 : return false;
2720 : }
2721 : case SPDK_BDEV_IO_TYPE_READ:
2722 30 : return true;
2723 : case SPDK_BDEV_IO_TYPE_ZCOPY:
2724 : /* Populate to read from disk */
2725 0 : if (bdev_io->u.bdev.zcopy.populate) {
2726 0 : return true;
2727 : } else {
2728 0 : return false;
2729 : }
2730 : default:
2731 3 : return false;
2732 : }
2733 33 : }
2734 :
2735 : static uint64_t
2736 43 : bdev_get_io_size_in_byte(struct spdk_bdev_io *bdev_io)
2737 : {
2738 43 : uint32_t blocklen = bdev_io_get_block_size(bdev_io);
2739 :
2740 43 : switch (bdev_io->type) {
2741 : case SPDK_BDEV_IO_TYPE_NVME_IO:
2742 : case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
2743 0 : return bdev_io->u.nvme_passthru.nbytes;
2744 : case SPDK_BDEV_IO_TYPE_READ:
2745 : case SPDK_BDEV_IO_TYPE_WRITE:
2746 43 : return bdev_io->u.bdev.num_blocks * blocklen;
2747 : case SPDK_BDEV_IO_TYPE_ZCOPY:
2748 : /* Track the data in the start phase only */
2749 0 : if (bdev_io->u.bdev.zcopy.start) {
2750 0 : return bdev_io->u.bdev.num_blocks * blocklen;
2751 : } else {
2752 0 : return 0;
2753 : }
2754 : default:
2755 0 : return 0;
2756 : }
2757 43 : }
2758 :
2759 : static inline bool
2760 64 : bdev_qos_rw_queue_io(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io, uint64_t delta)
2761 : {
2762 : int64_t remaining_this_timeslice;
2763 :
2764 64 : if (!limit->max_per_timeslice) {
2765 : /* The QoS is disabled */
2766 0 : return false;
2767 : }
2768 :
2769 64 : remaining_this_timeslice = __atomic_sub_fetch(&limit->remaining_this_timeslice, delta,
2770 : __ATOMIC_RELAXED);
2771 64 : if (remaining_this_timeslice + (int64_t)delta > 0) {
2772 : /* There was still a quota for this delta -> the IO shouldn't be queued
2773 : *
2774 : * We allow a slight quota overrun here so an IO bigger than the per-timeslice
2775 : * quota can be allowed once a while. Such overrun then taken into account in
2776 : * the QoS poller, where the next timeslice quota is calculated.
2777 : */
2778 59 : return false;
2779 : }
2780 :
2781 : /* There was no quota for this delta -> the IO should be queued
2782 : * The remaining_this_timeslice must be rewinded so it reflects the real
2783 : * amount of IOs or bytes allowed.
2784 : */
2785 5 : __atomic_add_fetch(
2786 5 : &limit->remaining_this_timeslice, delta, __ATOMIC_RELAXED);
2787 5 : return true;
2788 64 : }
2789 :
2790 : static inline void
2791 5 : bdev_qos_rw_rewind_io(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io, uint64_t delta)
2792 : {
2793 5 : __atomic_add_fetch(&limit->remaining_this_timeslice, delta, __ATOMIC_RELAXED);
2794 5 : }
2795 :
2796 : static bool
2797 23 : bdev_qos_rw_iops_queue(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2798 : {
2799 23 : return bdev_qos_rw_queue_io(limit, io, 1);
2800 : }
2801 :
2802 : static void
2803 3 : bdev_qos_rw_iops_rewind_quota(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2804 : {
2805 3 : bdev_qos_rw_rewind_io(limit, io, 1);
2806 3 : }
2807 :
2808 : static bool
2809 41 : bdev_qos_rw_bps_queue(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2810 : {
2811 41 : return bdev_qos_rw_queue_io(limit, io, bdev_get_io_size_in_byte(io));
2812 : }
2813 :
2814 : static void
2815 2 : bdev_qos_rw_bps_rewind_quota(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2816 : {
2817 2 : bdev_qos_rw_rewind_io(limit, io, bdev_get_io_size_in_byte(io));
2818 2 : }
2819 :
2820 : static bool
2821 19 : bdev_qos_r_bps_queue(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2822 : {
2823 19 : if (bdev_is_read_io(io) == false) {
2824 1 : return false;
2825 : }
2826 :
2827 18 : return bdev_qos_rw_bps_queue(limit, io);
2828 19 : }
2829 :
2830 : static void
2831 0 : bdev_qos_r_bps_rewind_quota(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2832 : {
2833 0 : if (bdev_is_read_io(io) != false) {
2834 0 : bdev_qos_rw_rewind_io(limit, io, bdev_get_io_size_in_byte(io));
2835 0 : }
2836 0 : }
2837 :
2838 : static bool
2839 14 : bdev_qos_w_bps_queue(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2840 : {
2841 14 : if (bdev_is_read_io(io) == true) {
2842 12 : return false;
2843 : }
2844 :
2845 2 : return bdev_qos_rw_bps_queue(limit, io);
2846 14 : }
2847 :
2848 : static void
2849 0 : bdev_qos_w_bps_rewind_quota(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2850 : {
2851 0 : if (bdev_is_read_io(io) != true) {
2852 0 : bdev_qos_rw_rewind_io(limit, io, bdev_get_io_size_in_byte(io));
2853 0 : }
2854 0 : }
2855 :
2856 : static void
2857 10 : bdev_qos_set_ops(struct spdk_bdev_qos *qos)
2858 : {
2859 : int i;
2860 :
2861 50 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
2862 40 : if (qos->rate_limits[i].limit == SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
2863 15 : qos->rate_limits[i].queue_io = NULL;
2864 15 : continue;
2865 : }
2866 :
2867 25 : switch (i) {
2868 : case SPDK_BDEV_QOS_RW_IOPS_RATE_LIMIT:
2869 9 : qos->rate_limits[i].queue_io = bdev_qos_rw_iops_queue;
2870 9 : qos->rate_limits[i].rewind_quota = bdev_qos_rw_iops_rewind_quota;
2871 9 : break;
2872 : case SPDK_BDEV_QOS_RW_BPS_RATE_LIMIT:
2873 7 : qos->rate_limits[i].queue_io = bdev_qos_rw_bps_queue;
2874 7 : qos->rate_limits[i].rewind_quota = bdev_qos_rw_bps_rewind_quota;
2875 7 : break;
2876 : case SPDK_BDEV_QOS_R_BPS_RATE_LIMIT:
2877 5 : qos->rate_limits[i].queue_io = bdev_qos_r_bps_queue;
2878 5 : qos->rate_limits[i].rewind_quota = bdev_qos_r_bps_rewind_quota;
2879 5 : break;
2880 : case SPDK_BDEV_QOS_W_BPS_RATE_LIMIT:
2881 4 : qos->rate_limits[i].queue_io = bdev_qos_w_bps_queue;
2882 4 : qos->rate_limits[i].rewind_quota = bdev_qos_w_bps_rewind_quota;
2883 4 : break;
2884 : default:
2885 0 : break;
2886 : }
2887 25 : }
2888 10 : }
2889 :
2890 : static void
2891 6 : _bdev_io_complete_in_submit(struct spdk_bdev_channel *bdev_ch,
2892 : struct spdk_bdev_io *bdev_io,
2893 : enum spdk_bdev_io_status status)
2894 : {
2895 6 : bdev_io->internal.f.in_submit_request = true;
2896 6 : bdev_io_increment_outstanding(bdev_ch, bdev_ch->shared_resource);
2897 6 : spdk_bdev_io_complete(bdev_io, status);
2898 6 : bdev_io->internal.f.in_submit_request = false;
2899 6 : }
2900 :
2901 : static inline void
2902 574 : bdev_io_do_submit(struct spdk_bdev_channel *bdev_ch, struct spdk_bdev_io *bdev_io)
2903 : {
2904 574 : struct spdk_bdev *bdev = bdev_io->bdev;
2905 574 : struct spdk_io_channel *ch = bdev_ch->channel;
2906 574 : struct spdk_bdev_shared_resource *shared_resource = bdev_ch->shared_resource;
2907 :
2908 574 : if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_ABORT)) {
2909 16 : struct spdk_bdev_mgmt_channel *mgmt_channel = shared_resource->mgmt_ch;
2910 16 : struct spdk_bdev_io *bio_to_abort = bdev_io->u.abort.bio_to_abort;
2911 :
2912 16 : if (bdev_abort_queued_io(&shared_resource->nomem_io, bio_to_abort) ||
2913 16 : bdev_abort_buf_io(mgmt_channel, bio_to_abort)) {
2914 0 : _bdev_io_complete_in_submit(bdev_ch, bdev_io,
2915 : SPDK_BDEV_IO_STATUS_SUCCESS);
2916 0 : return;
2917 : }
2918 16 : }
2919 :
2920 574 : if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE &&
2921 : bdev_io->bdev->split_on_write_unit &&
2922 : bdev_io->u.bdev.num_blocks < bdev_io->bdev->write_unit_size)) {
2923 4 : SPDK_ERRLOG("IO num_blocks %lu does not match the write_unit_size %u\n",
2924 : bdev_io->u.bdev.num_blocks, bdev_io->bdev->write_unit_size);
2925 4 : _bdev_io_complete_in_submit(bdev_ch, bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
2926 4 : return;
2927 : }
2928 :
2929 570 : if (spdk_likely(TAILQ_EMPTY(&shared_resource->nomem_io))) {
2930 527 : bdev_io_increment_outstanding(bdev_ch, shared_resource);
2931 527 : bdev_io->internal.f.in_submit_request = true;
2932 527 : bdev_submit_request(bdev, ch, bdev_io);
2933 527 : bdev_io->internal.f.in_submit_request = false;
2934 527 : } else {
2935 43 : bdev_queue_nomem_io_tail(shared_resource, bdev_io, BDEV_IO_RETRY_STATE_SUBMIT);
2936 43 : if (shared_resource->nomem_threshold == 0 && shared_resource->io_outstanding == 0) {
2937 : /* Special case when we have nomem IOs and no outstanding IOs which completions
2938 : * could trigger retry of queued IOs */
2939 0 : bdev_shared_ch_retry_io(shared_resource);
2940 0 : }
2941 : }
2942 574 : }
2943 :
2944 : static bool
2945 25 : bdev_qos_queue_io(struct spdk_bdev_qos *qos, struct spdk_bdev_io *bdev_io)
2946 : {
2947 : int i;
2948 :
2949 25 : if (bdev_qos_io_to_limit(bdev_io) == true) {
2950 100 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
2951 82 : if (!qos->rate_limits[i].queue_io) {
2952 5 : continue;
2953 : }
2954 :
2955 231 : if (qos->rate_limits[i].queue_io(&qos->rate_limits[i],
2956 154 : bdev_io) == true) {
2957 10 : for (i -= 1; i >= 0 ; i--) {
2958 5 : if (!qos->rate_limits[i].queue_io) {
2959 0 : continue;
2960 : }
2961 :
2962 5 : qos->rate_limits[i].rewind_quota(&qos->rate_limits[i], bdev_io);
2963 5 : }
2964 5 : return true;
2965 : }
2966 72 : }
2967 18 : }
2968 :
2969 20 : return false;
2970 25 : }
2971 :
2972 : static int
2973 27 : bdev_qos_io_submit(struct spdk_bdev_channel *ch, struct spdk_bdev_qos *qos)
2974 : {
2975 27 : struct spdk_bdev_io *bdev_io = NULL, *tmp = NULL;
2976 27 : int submitted_ios = 0;
2977 :
2978 52 : TAILQ_FOREACH_SAFE(bdev_io, &ch->qos_queued_io, internal.link, tmp) {
2979 25 : if (!bdev_qos_queue_io(qos, bdev_io)) {
2980 20 : TAILQ_REMOVE(&ch->qos_queued_io, bdev_io, internal.link);
2981 20 : bdev_io_do_submit(ch, bdev_io);
2982 :
2983 20 : submitted_ios++;
2984 20 : }
2985 25 : }
2986 :
2987 27 : return submitted_ios;
2988 : }
2989 :
2990 : static void
2991 2 : bdev_queue_io_wait_with_cb(struct spdk_bdev_io *bdev_io, spdk_bdev_io_wait_cb cb_fn)
2992 : {
2993 : int rc;
2994 :
2995 2 : bdev_io->internal.waitq_entry.bdev = bdev_io->bdev;
2996 2 : bdev_io->internal.waitq_entry.cb_fn = cb_fn;
2997 2 : bdev_io->internal.waitq_entry.cb_arg = bdev_io;
2998 4 : rc = spdk_bdev_queue_io_wait(bdev_io->bdev, spdk_io_channel_from_ctx(bdev_io->internal.ch),
2999 2 : &bdev_io->internal.waitq_entry);
3000 2 : if (rc != 0) {
3001 0 : SPDK_ERRLOG("Queue IO failed, rc=%d\n", rc);
3002 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
3003 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
3004 0 : }
3005 2 : }
3006 :
3007 : static bool
3008 621 : bdev_rw_should_split(struct spdk_bdev_io *bdev_io)
3009 : {
3010 : uint32_t io_boundary;
3011 621 : struct spdk_bdev *bdev = bdev_io->bdev;
3012 621 : uint32_t max_segment_size = bdev->max_segment_size;
3013 621 : uint32_t max_size = bdev->max_rw_size;
3014 621 : int max_segs = bdev->max_num_segments;
3015 :
3016 621 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE && bdev->split_on_write_unit) {
3017 24 : io_boundary = bdev->write_unit_size;
3018 621 : } else if (bdev->split_on_optimal_io_boundary) {
3019 168 : io_boundary = bdev->optimal_io_boundary;
3020 168 : } else {
3021 429 : io_boundary = 0;
3022 : }
3023 :
3024 621 : if (spdk_likely(!io_boundary && !max_segs && !max_segment_size && !max_size)) {
3025 243 : return false;
3026 : }
3027 :
3028 378 : if (io_boundary) {
3029 : uint64_t start_stripe, end_stripe;
3030 :
3031 192 : start_stripe = bdev_io->u.bdev.offset_blocks;
3032 192 : end_stripe = start_stripe + bdev_io->u.bdev.num_blocks - 1;
3033 : /* Avoid expensive div operations if possible. These spdk_u32 functions are very cheap. */
3034 192 : if (spdk_likely(spdk_u32_is_pow2(io_boundary))) {
3035 192 : start_stripe >>= spdk_u32log2(io_boundary);
3036 192 : end_stripe >>= spdk_u32log2(io_boundary);
3037 192 : } else {
3038 0 : start_stripe /= io_boundary;
3039 0 : end_stripe /= io_boundary;
3040 : }
3041 :
3042 192 : if (start_stripe != end_stripe) {
3043 75 : return true;
3044 : }
3045 117 : }
3046 :
3047 303 : if (max_segs) {
3048 150 : if (bdev_io->u.bdev.iovcnt > max_segs) {
3049 15 : return true;
3050 : }
3051 135 : }
3052 :
3053 288 : if (max_segment_size) {
3054 470 : for (int i = 0; i < bdev_io->u.bdev.iovcnt; i++) {
3055 346 : if (bdev_io->u.bdev.iovs[i].iov_len > max_segment_size) {
3056 12 : return true;
3057 : }
3058 334 : }
3059 124 : }
3060 :
3061 276 : if (max_size) {
3062 52 : if (bdev_io->u.bdev.num_blocks > max_size) {
3063 7 : return true;
3064 : }
3065 45 : }
3066 :
3067 269 : return false;
3068 621 : }
3069 :
3070 : static bool
3071 24 : bdev_unmap_should_split(struct spdk_bdev_io *bdev_io)
3072 : {
3073 : uint32_t num_unmap_segments;
3074 :
3075 24 : if (!bdev_io->bdev->max_unmap || !bdev_io->bdev->max_unmap_segments) {
3076 3 : return false;
3077 : }
3078 21 : num_unmap_segments = spdk_divide_round_up(bdev_io->u.bdev.num_blocks, bdev_io->bdev->max_unmap);
3079 21 : if (num_unmap_segments > bdev_io->bdev->max_unmap_segments) {
3080 4 : return true;
3081 : }
3082 :
3083 17 : return false;
3084 24 : }
3085 :
3086 : static bool
3087 37 : bdev_write_zeroes_should_split(struct spdk_bdev_io *bdev_io)
3088 : {
3089 37 : if (!bdev_io->bdev->max_write_zeroes) {
3090 4 : return false;
3091 : }
3092 :
3093 33 : if (bdev_io->u.bdev.num_blocks > bdev_io->bdev->max_write_zeroes) {
3094 10 : return true;
3095 : }
3096 :
3097 23 : return false;
3098 37 : }
3099 :
3100 : static bool
3101 30 : bdev_copy_should_split(struct spdk_bdev_io *bdev_io)
3102 : {
3103 30 : if (bdev_io->bdev->max_copy != 0 &&
3104 25 : bdev_io->u.bdev.num_blocks > bdev_io->bdev->max_copy) {
3105 6 : return true;
3106 : }
3107 :
3108 24 : return false;
3109 30 : }
3110 :
3111 : static bool
3112 792 : bdev_io_should_split(struct spdk_bdev_io *bdev_io)
3113 : {
3114 792 : switch (bdev_io->type) {
3115 : case SPDK_BDEV_IO_TYPE_READ:
3116 : case SPDK_BDEV_IO_TYPE_WRITE:
3117 621 : return bdev_rw_should_split(bdev_io);
3118 : case SPDK_BDEV_IO_TYPE_UNMAP:
3119 24 : return bdev_unmap_should_split(bdev_io);
3120 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3121 37 : return bdev_write_zeroes_should_split(bdev_io);
3122 : case SPDK_BDEV_IO_TYPE_COPY:
3123 30 : return bdev_copy_should_split(bdev_io);
3124 : default:
3125 80 : return false;
3126 : }
3127 792 : }
3128 :
3129 : static uint32_t
3130 249 : _to_next_boundary(uint64_t offset, uint32_t boundary)
3131 : {
3132 249 : return (boundary - (offset % boundary));
3133 : }
3134 :
3135 : static void bdev_io_split_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg);
3136 :
3137 : static void _bdev_rw_split(void *_bdev_io);
3138 :
3139 : static void bdev_unmap_split(struct spdk_bdev_io *bdev_io);
3140 :
3141 : static void
3142 0 : _bdev_unmap_split(void *_bdev_io)
3143 : {
3144 0 : return bdev_unmap_split((struct spdk_bdev_io *)_bdev_io);
3145 : }
3146 :
3147 : static void bdev_write_zeroes_split(struct spdk_bdev_io *bdev_io);
3148 :
3149 : static void
3150 0 : _bdev_write_zeroes_split(void *_bdev_io)
3151 : {
3152 0 : return bdev_write_zeroes_split((struct spdk_bdev_io *)_bdev_io);
3153 : }
3154 :
3155 : static void bdev_copy_split(struct spdk_bdev_io *bdev_io);
3156 :
3157 : static void
3158 0 : _bdev_copy_split(void *_bdev_io)
3159 : {
3160 0 : return bdev_copy_split((struct spdk_bdev_io *)_bdev_io);
3161 : }
3162 :
3163 : static int
3164 305 : bdev_io_split_submit(struct spdk_bdev_io *bdev_io, struct iovec *iov, int iovcnt, void *md_buf,
3165 : uint64_t num_blocks, uint64_t *offset, uint64_t *remaining)
3166 : {
3167 : int rc;
3168 : uint64_t current_offset, current_remaining, current_src_offset;
3169 : spdk_bdev_io_wait_cb io_wait_fn;
3170 :
3171 305 : current_offset = *offset;
3172 305 : current_remaining = *remaining;
3173 :
3174 305 : assert(bdev_io->internal.f.split);
3175 :
3176 305 : bdev_io->internal.split.outstanding++;
3177 :
3178 305 : io_wait_fn = _bdev_rw_split;
3179 305 : switch (bdev_io->type) {
3180 : case SPDK_BDEV_IO_TYPE_READ:
3181 196 : assert(bdev_io->u.bdev.accel_sequence == NULL);
3182 392 : rc = bdev_readv_blocks_with_md(bdev_io->internal.desc,
3183 196 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
3184 196 : iov, iovcnt, md_buf, current_offset,
3185 196 : num_blocks,
3186 196 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain : NULL,
3187 196 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain_ctx : NULL,
3188 : NULL,
3189 196 : bdev_io->u.bdev.dif_check_flags,
3190 196 : bdev_io_split_done, bdev_io);
3191 196 : break;
3192 : case SPDK_BDEV_IO_TYPE_WRITE:
3193 50 : assert(bdev_io->u.bdev.accel_sequence == NULL);
3194 100 : rc = bdev_writev_blocks_with_md(bdev_io->internal.desc,
3195 50 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
3196 50 : iov, iovcnt, md_buf, current_offset,
3197 50 : num_blocks,
3198 50 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain : NULL,
3199 50 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain_ctx : NULL,
3200 : NULL,
3201 50 : bdev_io->u.bdev.dif_check_flags,
3202 50 : bdev_io->u.bdev.nvme_cdw12.raw,
3203 50 : bdev_io->u.bdev.nvme_cdw13.raw,
3204 50 : bdev_io_split_done, bdev_io);
3205 50 : break;
3206 : case SPDK_BDEV_IO_TYPE_UNMAP:
3207 17 : io_wait_fn = _bdev_unmap_split;
3208 34 : rc = spdk_bdev_unmap_blocks(bdev_io->internal.desc,
3209 17 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
3210 17 : current_offset, num_blocks,
3211 17 : bdev_io_split_done, bdev_io);
3212 17 : break;
3213 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3214 23 : io_wait_fn = _bdev_write_zeroes_split;
3215 46 : rc = spdk_bdev_write_zeroes_blocks(bdev_io->internal.desc,
3216 23 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
3217 23 : current_offset, num_blocks,
3218 23 : bdev_io_split_done, bdev_io);
3219 23 : break;
3220 : case SPDK_BDEV_IO_TYPE_COPY:
3221 19 : io_wait_fn = _bdev_copy_split;
3222 38 : current_src_offset = bdev_io->u.bdev.copy.src_offset_blocks +
3223 19 : (current_offset - bdev_io->u.bdev.offset_blocks);
3224 38 : rc = spdk_bdev_copy_blocks(bdev_io->internal.desc,
3225 19 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
3226 19 : current_offset, current_src_offset, num_blocks,
3227 19 : bdev_io_split_done, bdev_io);
3228 19 : break;
3229 : default:
3230 0 : assert(false);
3231 : rc = -EINVAL;
3232 : break;
3233 : }
3234 :
3235 305 : if (rc == 0) {
3236 301 : current_offset += num_blocks;
3237 301 : current_remaining -= num_blocks;
3238 301 : bdev_io->internal.split.current_offset_blocks = current_offset;
3239 301 : bdev_io->internal.split.remaining_num_blocks = current_remaining;
3240 301 : *offset = current_offset;
3241 301 : *remaining = current_remaining;
3242 301 : } else {
3243 4 : bdev_io->internal.split.outstanding--;
3244 4 : if (rc == -ENOMEM) {
3245 4 : if (bdev_io->internal.split.outstanding == 0) {
3246 : /* No I/O is outstanding. Hence we should wait here. */
3247 1 : bdev_queue_io_wait_with_cb(bdev_io, io_wait_fn);
3248 1 : }
3249 4 : } else {
3250 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
3251 0 : if (bdev_io->internal.split.outstanding == 0) {
3252 0 : bdev_ch_remove_from_io_submitted(bdev_io);
3253 0 : spdk_trace_record(TRACE_BDEV_IO_DONE, bdev_io->internal.ch->trace_id,
3254 : 0, (uintptr_t)bdev_io, bdev_io->internal.caller_ctx,
3255 : bdev_io->internal.ch->queue_depth);
3256 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
3257 0 : }
3258 : }
3259 : }
3260 :
3261 305 : return rc;
3262 : }
3263 :
3264 : static void
3265 67 : _bdev_rw_split(void *_bdev_io)
3266 : {
3267 : struct iovec *parent_iov, *iov;
3268 67 : struct spdk_bdev_io *bdev_io = _bdev_io;
3269 67 : struct spdk_bdev *bdev = bdev_io->bdev;
3270 : uint64_t parent_offset, current_offset, remaining;
3271 : uint32_t parent_iov_offset, parent_iovcnt, parent_iovpos, child_iovcnt;
3272 : uint32_t to_next_boundary, to_next_boundary_bytes, to_last_block_bytes;
3273 : uint32_t iovcnt, iov_len, child_iovsize;
3274 : uint32_t blocklen;
3275 : uint32_t io_boundary;
3276 67 : uint32_t max_segment_size = bdev->max_segment_size;
3277 67 : uint32_t max_child_iovcnt = bdev->max_num_segments;
3278 67 : uint32_t max_size = bdev->max_rw_size;
3279 67 : void *md_buf = NULL;
3280 : int rc;
3281 :
3282 67 : blocklen = bdev_io_get_block_size(bdev_io);
3283 :
3284 67 : max_size = max_size ? max_size : UINT32_MAX;
3285 67 : max_segment_size = max_segment_size ? max_segment_size : UINT32_MAX;
3286 67 : max_child_iovcnt = max_child_iovcnt ? spdk_min(max_child_iovcnt, SPDK_BDEV_IO_NUM_CHILD_IOV) :
3287 : SPDK_BDEV_IO_NUM_CHILD_IOV;
3288 :
3289 67 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE && bdev->split_on_write_unit) {
3290 5 : io_boundary = bdev->write_unit_size;
3291 67 : } else if (bdev->split_on_optimal_io_boundary) {
3292 40 : io_boundary = bdev->optimal_io_boundary;
3293 40 : } else {
3294 22 : io_boundary = UINT32_MAX;
3295 : }
3296 :
3297 67 : assert(bdev_io->internal.f.split);
3298 :
3299 67 : remaining = bdev_io->internal.split.remaining_num_blocks;
3300 67 : current_offset = bdev_io->internal.split.current_offset_blocks;
3301 67 : parent_offset = bdev_io->u.bdev.offset_blocks;
3302 67 : parent_iov_offset = (current_offset - parent_offset) * blocklen;
3303 67 : parent_iovcnt = bdev_io->u.bdev.iovcnt;
3304 :
3305 420 : for (parent_iovpos = 0; parent_iovpos < parent_iovcnt; parent_iovpos++) {
3306 420 : parent_iov = &bdev_io->u.bdev.iovs[parent_iovpos];
3307 420 : if (parent_iov_offset < parent_iov->iov_len) {
3308 67 : break;
3309 : }
3310 353 : parent_iov_offset -= parent_iov->iov_len;
3311 353 : }
3312 :
3313 67 : child_iovcnt = 0;
3314 573 : while (remaining > 0 && parent_iovpos < parent_iovcnt &&
3315 264 : child_iovcnt < SPDK_BDEV_IO_NUM_CHILD_IOV) {
3316 249 : to_next_boundary = _to_next_boundary(current_offset, io_boundary);
3317 249 : to_next_boundary = spdk_min(remaining, to_next_boundary);
3318 249 : to_next_boundary = spdk_min(max_size, to_next_boundary);
3319 249 : to_next_boundary_bytes = to_next_boundary * blocklen;
3320 :
3321 249 : iov = &bdev_io->child_iov[child_iovcnt];
3322 249 : iovcnt = 0;
3323 :
3324 249 : if (bdev_io->u.bdev.md_buf) {
3325 48 : md_buf = (char *)bdev_io->u.bdev.md_buf +
3326 24 : (current_offset - parent_offset) * spdk_bdev_get_md_size(bdev);
3327 24 : }
3328 :
3329 249 : child_iovsize = spdk_min(SPDK_BDEV_IO_NUM_CHILD_IOV - child_iovcnt, max_child_iovcnt);
3330 1810 : while (to_next_boundary_bytes > 0 && parent_iovpos < parent_iovcnt &&
3331 836 : iovcnt < child_iovsize) {
3332 725 : parent_iov = &bdev_io->u.bdev.iovs[parent_iovpos];
3333 725 : iov_len = parent_iov->iov_len - parent_iov_offset;
3334 :
3335 725 : iov_len = spdk_min(iov_len, max_segment_size);
3336 725 : iov_len = spdk_min(iov_len, to_next_boundary_bytes);
3337 725 : to_next_boundary_bytes -= iov_len;
3338 :
3339 725 : bdev_io->child_iov[child_iovcnt].iov_base = parent_iov->iov_base + parent_iov_offset;
3340 725 : bdev_io->child_iov[child_iovcnt].iov_len = iov_len;
3341 :
3342 725 : if (iov_len < parent_iov->iov_len - parent_iov_offset) {
3343 183 : parent_iov_offset += iov_len;
3344 183 : } else {
3345 542 : parent_iovpos++;
3346 542 : parent_iov_offset = 0;
3347 : }
3348 725 : child_iovcnt++;
3349 725 : iovcnt++;
3350 : }
3351 :
3352 249 : if (to_next_boundary_bytes > 0) {
3353 : /* We had to stop this child I/O early because we ran out of
3354 : * child_iov space or were limited by max_num_segments.
3355 : * Ensure the iovs to be aligned with block size and
3356 : * then adjust to_next_boundary before starting the
3357 : * child I/O.
3358 : */
3359 111 : assert(child_iovcnt == SPDK_BDEV_IO_NUM_CHILD_IOV ||
3360 : iovcnt == child_iovsize);
3361 111 : to_last_block_bytes = to_next_boundary_bytes % blocklen;
3362 111 : if (to_last_block_bytes != 0) {
3363 24 : uint32_t child_iovpos = child_iovcnt - 1;
3364 : /* don't decrease child_iovcnt when it equals to SPDK_BDEV_IO_NUM_CHILD_IOV
3365 : * so the loop will naturally end
3366 : */
3367 :
3368 24 : to_last_block_bytes = blocklen - to_last_block_bytes;
3369 24 : to_next_boundary_bytes += to_last_block_bytes;
3370 53 : while (to_last_block_bytes > 0 && iovcnt > 0) {
3371 32 : iov_len = spdk_min(to_last_block_bytes,
3372 : bdev_io->child_iov[child_iovpos].iov_len);
3373 32 : bdev_io->child_iov[child_iovpos].iov_len -= iov_len;
3374 32 : if (bdev_io->child_iov[child_iovpos].iov_len == 0) {
3375 15 : child_iovpos--;
3376 15 : if (--iovcnt == 0) {
3377 : /* If the child IO is less than a block size just return.
3378 : * If the first child IO of any split round is less than
3379 : * a block size, an error exit.
3380 : */
3381 3 : if (bdev_io->internal.split.outstanding == 0) {
3382 1 : SPDK_ERRLOG("The first child io was less than a block size\n");
3383 1 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
3384 1 : bdev_ch_remove_from_io_submitted(bdev_io);
3385 1 : spdk_trace_record(TRACE_BDEV_IO_DONE, bdev_io->internal.ch->trace_id,
3386 : 0, (uintptr_t)bdev_io, bdev_io->internal.caller_ctx,
3387 : bdev_io->internal.ch->queue_depth);
3388 1 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
3389 1 : }
3390 :
3391 3 : return;
3392 : }
3393 12 : }
3394 :
3395 29 : to_last_block_bytes -= iov_len;
3396 :
3397 29 : if (parent_iov_offset == 0) {
3398 14 : parent_iovpos--;
3399 14 : parent_iov_offset = bdev_io->u.bdev.iovs[parent_iovpos].iov_len;
3400 14 : }
3401 29 : parent_iov_offset -= iov_len;
3402 : }
3403 :
3404 21 : assert(to_last_block_bytes == 0);
3405 21 : }
3406 108 : to_next_boundary -= to_next_boundary_bytes / blocklen;
3407 108 : }
3408 :
3409 246 : rc = bdev_io_split_submit(bdev_io, iov, iovcnt, md_buf, to_next_boundary,
3410 : ¤t_offset, &remaining);
3411 246 : if (spdk_unlikely(rc)) {
3412 4 : return;
3413 : }
3414 : }
3415 67 : }
3416 :
3417 : static void
3418 3 : bdev_unmap_split(struct spdk_bdev_io *bdev_io)
3419 : {
3420 : uint64_t offset, unmap_blocks, remaining, max_unmap_blocks;
3421 3 : uint32_t num_children_reqs = 0;
3422 : int rc;
3423 :
3424 3 : assert(bdev_io->internal.f.split);
3425 :
3426 3 : offset = bdev_io->internal.split.current_offset_blocks;
3427 3 : remaining = bdev_io->internal.split.remaining_num_blocks;
3428 3 : max_unmap_blocks = bdev_io->bdev->max_unmap * bdev_io->bdev->max_unmap_segments;
3429 :
3430 20 : while (remaining && (num_children_reqs < SPDK_BDEV_MAX_CHILDREN_UNMAP_WRITE_ZEROES_REQS)) {
3431 17 : unmap_blocks = spdk_min(remaining, max_unmap_blocks);
3432 :
3433 17 : rc = bdev_io_split_submit(bdev_io, NULL, 0, NULL, unmap_blocks,
3434 : &offset, &remaining);
3435 17 : if (spdk_likely(rc == 0)) {
3436 17 : num_children_reqs++;
3437 17 : } else {
3438 0 : return;
3439 : }
3440 : }
3441 3 : }
3442 :
3443 : static void
3444 6 : bdev_write_zeroes_split(struct spdk_bdev_io *bdev_io)
3445 : {
3446 : uint64_t offset, write_zeroes_blocks, remaining;
3447 6 : uint32_t num_children_reqs = 0;
3448 : int rc;
3449 :
3450 6 : assert(bdev_io->internal.f.split);
3451 :
3452 6 : offset = bdev_io->internal.split.current_offset_blocks;
3453 6 : remaining = bdev_io->internal.split.remaining_num_blocks;
3454 :
3455 29 : while (remaining && (num_children_reqs < SPDK_BDEV_MAX_CHILDREN_UNMAP_WRITE_ZEROES_REQS)) {
3456 23 : write_zeroes_blocks = spdk_min(remaining, bdev_io->bdev->max_write_zeroes);
3457 :
3458 23 : rc = bdev_io_split_submit(bdev_io, NULL, 0, NULL, write_zeroes_blocks,
3459 : &offset, &remaining);
3460 23 : if (spdk_likely(rc == 0)) {
3461 23 : num_children_reqs++;
3462 23 : } else {
3463 0 : return;
3464 : }
3465 : }
3466 6 : }
3467 :
3468 : static void
3469 4 : bdev_copy_split(struct spdk_bdev_io *bdev_io)
3470 : {
3471 : uint64_t offset, copy_blocks, remaining;
3472 4 : uint32_t num_children_reqs = 0;
3473 : int rc;
3474 :
3475 4 : assert(bdev_io->internal.f.split);
3476 :
3477 4 : offset = bdev_io->internal.split.current_offset_blocks;
3478 4 : remaining = bdev_io->internal.split.remaining_num_blocks;
3479 :
3480 4 : assert(bdev_io->bdev->max_copy != 0);
3481 23 : while (remaining && (num_children_reqs < SPDK_BDEV_MAX_CHILDREN_COPY_REQS)) {
3482 19 : copy_blocks = spdk_min(remaining, bdev_io->bdev->max_copy);
3483 :
3484 19 : rc = bdev_io_split_submit(bdev_io, NULL, 0, NULL, copy_blocks,
3485 : &offset, &remaining);
3486 19 : if (spdk_likely(rc == 0)) {
3487 19 : num_children_reqs++;
3488 19 : } else {
3489 0 : return;
3490 : }
3491 : }
3492 4 : }
3493 :
3494 : static void
3495 58 : parent_bdev_io_complete(void *ctx, int rc)
3496 : {
3497 58 : struct spdk_bdev_io *parent_io = ctx;
3498 :
3499 58 : if (rc) {
3500 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
3501 0 : }
3502 :
3503 116 : parent_io->internal.cb(parent_io, parent_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS,
3504 58 : parent_io->internal.caller_ctx);
3505 58 : }
3506 :
3507 : static void
3508 0 : bdev_io_complete_parent_sequence_cb(void *ctx, int status)
3509 : {
3510 0 : struct spdk_bdev_io *bdev_io = ctx;
3511 :
3512 : /* u.bdev.accel_sequence should have already been cleared at this point */
3513 0 : assert(bdev_io->u.bdev.accel_sequence == NULL);
3514 0 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS);
3515 0 : bdev_io->internal.f.has_accel_sequence = false;
3516 :
3517 0 : if (spdk_unlikely(status != 0)) {
3518 0 : SPDK_ERRLOG("Failed to execute accel sequence, status=%d\n", status);
3519 0 : }
3520 :
3521 0 : parent_bdev_io_complete(bdev_io, status);
3522 0 : }
3523 :
3524 : static void
3525 301 : bdev_io_split_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
3526 : {
3527 301 : struct spdk_bdev_io *parent_io = cb_arg;
3528 :
3529 301 : spdk_bdev_free_io(bdev_io);
3530 :
3531 301 : assert(parent_io->internal.f.split);
3532 :
3533 301 : if (!success) {
3534 21 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
3535 : /* If any child I/O failed, stop further splitting process. */
3536 21 : parent_io->internal.split.current_offset_blocks += parent_io->internal.split.remaining_num_blocks;
3537 21 : parent_io->internal.split.remaining_num_blocks = 0;
3538 21 : }
3539 301 : parent_io->internal.split.outstanding--;
3540 301 : if (parent_io->internal.split.outstanding != 0) {
3541 223 : return;
3542 : }
3543 :
3544 : /*
3545 : * Parent I/O finishes when all blocks are consumed.
3546 : */
3547 78 : if (parent_io->internal.split.remaining_num_blocks == 0) {
3548 58 : assert(parent_io->internal.cb != bdev_io_split_done);
3549 58 : bdev_ch_remove_from_io_submitted(parent_io);
3550 58 : spdk_trace_record(TRACE_BDEV_IO_DONE, parent_io->internal.ch->trace_id,
3551 : 0, (uintptr_t)parent_io, bdev_io->internal.caller_ctx,
3552 : parent_io->internal.ch->queue_depth);
3553 :
3554 58 : if (spdk_likely(parent_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS)) {
3555 48 : if (bdev_io_needs_sequence_exec(parent_io->internal.desc, parent_io)) {
3556 0 : bdev_io_exec_sequence(parent_io, bdev_io_complete_parent_sequence_cb);
3557 0 : return;
3558 48 : } else if (parent_io->internal.f.has_bounce_buf &&
3559 0 : !bdev_io_use_accel_sequence(bdev_io)) {
3560 : /* bdev IO will be completed in the callback */
3561 0 : _bdev_io_push_bounce_data_buffer(parent_io, parent_bdev_io_complete);
3562 0 : return;
3563 : }
3564 48 : }
3565 :
3566 58 : parent_bdev_io_complete(parent_io, 0);
3567 58 : return;
3568 : }
3569 :
3570 : /*
3571 : * Continue with the splitting process. This function will complete the parent I/O if the
3572 : * splitting is done.
3573 : */
3574 20 : switch (parent_io->type) {
3575 : case SPDK_BDEV_IO_TYPE_READ:
3576 : case SPDK_BDEV_IO_TYPE_WRITE:
3577 17 : _bdev_rw_split(parent_io);
3578 17 : break;
3579 : case SPDK_BDEV_IO_TYPE_UNMAP:
3580 1 : bdev_unmap_split(parent_io);
3581 1 : break;
3582 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3583 1 : bdev_write_zeroes_split(parent_io);
3584 1 : break;
3585 : case SPDK_BDEV_IO_TYPE_COPY:
3586 1 : bdev_copy_split(parent_io);
3587 1 : break;
3588 : default:
3589 0 : assert(false);
3590 : break;
3591 : }
3592 301 : }
3593 :
3594 : static void bdev_rw_split_get_buf_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io,
3595 : bool success);
3596 :
3597 : static void
3598 59 : bdev_io_split(struct spdk_bdev_io *bdev_io)
3599 : {
3600 59 : assert(bdev_io_should_split(bdev_io));
3601 59 : assert(bdev_io->internal.f.split);
3602 :
3603 59 : bdev_io->internal.split.current_offset_blocks = bdev_io->u.bdev.offset_blocks;
3604 59 : bdev_io->internal.split.remaining_num_blocks = bdev_io->u.bdev.num_blocks;
3605 59 : bdev_io->internal.split.outstanding = 0;
3606 59 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
3607 :
3608 59 : switch (bdev_io->type) {
3609 : case SPDK_BDEV_IO_TYPE_READ:
3610 : case SPDK_BDEV_IO_TYPE_WRITE:
3611 49 : if (_is_buf_allocated(bdev_io->u.bdev.iovs)) {
3612 49 : _bdev_rw_split(bdev_io);
3613 49 : } else {
3614 0 : assert(bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
3615 0 : spdk_bdev_io_get_buf(bdev_io, bdev_rw_split_get_buf_cb,
3616 0 : bdev_io->u.bdev.num_blocks * bdev_io_get_block_size(bdev_io));
3617 : }
3618 49 : break;
3619 : case SPDK_BDEV_IO_TYPE_UNMAP:
3620 2 : bdev_unmap_split(bdev_io);
3621 2 : break;
3622 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3623 5 : bdev_write_zeroes_split(bdev_io);
3624 5 : break;
3625 : case SPDK_BDEV_IO_TYPE_COPY:
3626 3 : bdev_copy_split(bdev_io);
3627 3 : break;
3628 : default:
3629 0 : assert(false);
3630 : break;
3631 : }
3632 59 : }
3633 :
3634 : static void
3635 0 : bdev_rw_split_get_buf_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io, bool success)
3636 : {
3637 0 : if (!success) {
3638 0 : spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
3639 0 : return;
3640 : }
3641 :
3642 0 : _bdev_rw_split(bdev_io);
3643 0 : }
3644 :
3645 : static inline void
3646 579 : _bdev_io_submit(struct spdk_bdev_io *bdev_io)
3647 : {
3648 579 : struct spdk_bdev *bdev = bdev_io->bdev;
3649 579 : struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
3650 :
3651 579 : if (spdk_likely(bdev_ch->flags == 0)) {
3652 554 : bdev_io_do_submit(bdev_ch, bdev_io);
3653 554 : return;
3654 : }
3655 :
3656 25 : if (bdev_ch->flags & BDEV_CH_RESET_IN_PROGRESS) {
3657 2 : _bdev_io_complete_in_submit(bdev_ch, bdev_io, SPDK_BDEV_IO_STATUS_ABORTED);
3658 25 : } else if (bdev_ch->flags & BDEV_CH_QOS_ENABLED) {
3659 23 : if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_ABORT) &&
3660 2 : bdev_abort_queued_io(&bdev_ch->qos_queued_io, bdev_io->u.abort.bio_to_abort)) {
3661 0 : _bdev_io_complete_in_submit(bdev_ch, bdev_io, SPDK_BDEV_IO_STATUS_SUCCESS);
3662 0 : } else {
3663 23 : TAILQ_INSERT_TAIL(&bdev_ch->qos_queued_io, bdev_io, internal.link);
3664 23 : bdev_qos_io_submit(bdev_ch, bdev->internal.qos);
3665 : }
3666 23 : } else {
3667 0 : SPDK_ERRLOG("unknown bdev_ch flag %x found\n", bdev_ch->flags);
3668 0 : _bdev_io_complete_in_submit(bdev_ch, bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
3669 : }
3670 579 : }
3671 :
3672 : bool bdev_lba_range_overlapped(struct lba_range *range1, struct lba_range *range2);
3673 :
3674 : bool
3675 23 : bdev_lba_range_overlapped(struct lba_range *range1, struct lba_range *range2)
3676 : {
3677 23 : if (range1->length == 0 || range2->length == 0) {
3678 1 : return false;
3679 : }
3680 :
3681 22 : if (range1->offset + range1->length <= range2->offset) {
3682 1 : return false;
3683 : }
3684 :
3685 21 : if (range2->offset + range2->length <= range1->offset) {
3686 3 : return false;
3687 : }
3688 :
3689 18 : return true;
3690 23 : }
3691 :
3692 : static bool
3693 11 : bdev_io_range_is_locked(struct spdk_bdev_io *bdev_io, struct lba_range *range)
3694 : {
3695 11 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
3696 : struct lba_range r;
3697 :
3698 11 : switch (bdev_io->type) {
3699 : case SPDK_BDEV_IO_TYPE_NVME_IO:
3700 : case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
3701 : /* Don't try to decode the NVMe command - just assume worst-case and that
3702 : * it overlaps a locked range.
3703 : */
3704 0 : return true;
3705 : case SPDK_BDEV_IO_TYPE_READ:
3706 6 : if (!range->quiesce) {
3707 4 : return false;
3708 : }
3709 : /* fallthrough */
3710 : case SPDK_BDEV_IO_TYPE_WRITE:
3711 : case SPDK_BDEV_IO_TYPE_UNMAP:
3712 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3713 : case SPDK_BDEV_IO_TYPE_ZCOPY:
3714 : case SPDK_BDEV_IO_TYPE_COPY:
3715 7 : r.offset = bdev_io->u.bdev.offset_blocks;
3716 7 : r.length = bdev_io->u.bdev.num_blocks;
3717 7 : if (!bdev_lba_range_overlapped(range, &r)) {
3718 : /* This I/O doesn't overlap the specified LBA range. */
3719 0 : return false;
3720 7 : } else if (range->owner_ch == ch && range->locked_ctx == bdev_io->internal.caller_ctx) {
3721 : /* This I/O overlaps, but the I/O is on the same channel that locked this
3722 : * range, and the caller_ctx is the same as the locked_ctx. This means
3723 : * that this I/O is associated with the lock, and is allowed to execute.
3724 : */
3725 2 : return false;
3726 : } else {
3727 5 : return true;
3728 : }
3729 : default:
3730 0 : return false;
3731 : }
3732 11 : }
3733 :
3734 : void
3735 639 : bdev_io_submit(struct spdk_bdev_io *bdev_io)
3736 : {
3737 639 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
3738 :
3739 639 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_PENDING);
3740 :
3741 639 : if (!TAILQ_EMPTY(&ch->locked_ranges)) {
3742 : struct lba_range *range;
3743 :
3744 13 : TAILQ_FOREACH(range, &ch->locked_ranges, tailq) {
3745 8 : if (bdev_io_range_is_locked(bdev_io, range)) {
3746 3 : TAILQ_INSERT_TAIL(&ch->io_locked, bdev_io, internal.ch_link);
3747 3 : return;
3748 : }
3749 5 : }
3750 5 : }
3751 :
3752 636 : bdev_ch_add_to_io_submitted(bdev_io);
3753 :
3754 636 : bdev_io->internal.submit_tsc = spdk_get_ticks();
3755 636 : spdk_trace_record_tsc(bdev_io->internal.submit_tsc, TRACE_BDEV_IO_START,
3756 : ch->trace_id, bdev_io->u.bdev.num_blocks,
3757 : (uintptr_t)bdev_io, (uint64_t)bdev_io->type, bdev_io->internal.caller_ctx,
3758 : bdev_io->u.bdev.offset_blocks, ch->queue_depth);
3759 :
3760 636 : if (bdev_io->internal.f.split) {
3761 59 : bdev_io_split(bdev_io);
3762 59 : return;
3763 : }
3764 :
3765 577 : _bdev_io_submit(bdev_io);
3766 639 : }
3767 :
3768 : static inline int
3769 2 : bdev_io_init_dif_ctx(struct spdk_bdev_io *bdev_io)
3770 : {
3771 2 : struct spdk_bdev *bdev = bdev_io->bdev;
3772 : struct spdk_dif_ctx_init_ext_opts dif_opts;
3773 :
3774 2 : memset(&bdev_io->u.bdev.dif_err, 0, sizeof(struct spdk_dif_error));
3775 :
3776 2 : dif_opts.size = SPDK_SIZEOF(&dif_opts, dif_pi_format);
3777 2 : dif_opts.dif_pi_format = bdev->dif_pi_format;
3778 :
3779 4 : return spdk_dif_ctx_init(&bdev_io->u.bdev.dif_ctx,
3780 2 : bdev->blocklen,
3781 2 : bdev->md_len,
3782 2 : bdev->md_interleave,
3783 2 : bdev->dif_is_head_of_md,
3784 2 : bdev->dif_type,
3785 2 : bdev_io->u.bdev.dif_check_flags,
3786 2 : bdev_io->u.bdev.offset_blocks & 0xFFFFFFFF,
3787 : 0xFFFF, 0, 0, 0, &dif_opts);
3788 : }
3789 :
3790 : static void
3791 4 : _bdev_memory_domain_get_io_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io,
3792 : bool success)
3793 : {
3794 4 : if (!success) {
3795 0 : SPDK_ERRLOG("Failed to get data buffer, completing IO\n");
3796 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
3797 0 : bdev_io_complete_unsubmitted(bdev_io);
3798 0 : return;
3799 : }
3800 :
3801 4 : if (bdev_io_needs_sequence_exec(bdev_io->internal.desc, bdev_io)) {
3802 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
3803 0 : bdev_io_exec_sequence(bdev_io, bdev_io_submit_sequence_cb);
3804 0 : return;
3805 : }
3806 : /* For reads we'll execute the sequence after the data is read, so, for now, only
3807 : * clear out accel_sequence pointer and submit the IO */
3808 0 : assert(bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
3809 0 : bdev_io->u.bdev.accel_sequence = NULL;
3810 0 : }
3811 :
3812 4 : bdev_io_submit(bdev_io);
3813 4 : }
3814 :
3815 : static inline void
3816 4 : _bdev_io_ext_use_bounce_buffer(struct spdk_bdev_io *bdev_io)
3817 : {
3818 : /* bdev doesn't support memory domains, thereby buffers in this IO request can't
3819 : * be accessed directly. It is needed to allocate buffers before issuing IO operation.
3820 : * For write operation we need to pull buffers from memory domain before submitting IO.
3821 : * Once read operation completes, we need to use memory_domain push functionality to
3822 : * update data in original memory domain IO buffer.
3823 : *
3824 : * If this I/O request is not aware of metadata, buffers in thsi IO request can't be
3825 : * accessed directly too. It is needed to allocate buffers before issuing IO operation.
3826 : * For write operation we need to insert metadata before submitting IO. Once read
3827 : * operation completes, we need to strip metadata in original IO buffer.
3828 : *
3829 : * This IO request will go through a regular IO flow, so clear memory domains pointers */
3830 4 : assert(bdev_io_use_memory_domain(bdev_io) ||
3831 : bdev_io_needs_metadata(bdev_io->internal.desc, bdev_io));
3832 :
3833 4 : bdev_io->u.bdev.memory_domain = NULL;
3834 4 : bdev_io->u.bdev.memory_domain_ctx = NULL;
3835 8 : _bdev_io_get_bounce_buf(bdev_io, _bdev_memory_domain_get_io_cb,
3836 4 : bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen);
3837 4 : }
3838 :
3839 : static inline void
3840 0 : _bdev_io_ext_use_accel_buffer(struct spdk_bdev_io *bdev_io)
3841 : {
3842 0 : assert(bdev_io_use_memory_domain(bdev_io));
3843 0 : assert(bdev_io_needs_metadata(bdev_io->internal.desc, bdev_io));
3844 :
3845 0 : bdev_io->u.bdev.memory_domain = NULL;
3846 0 : bdev_io->u.bdev.memory_domain_ctx = NULL;
3847 0 : bdev_io_get_accel_buf(bdev_io, _bdev_memory_domain_get_io_cb,
3848 0 : bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen);
3849 0 : }
3850 :
3851 : /* We need to allocate bounce buffer
3852 : * - if bdev doesn't support memory domains,
3853 : * - if it does support them, but we need to execute an accel sequence and the data buffer is
3854 : * from accel memory domain (to avoid doing a push/pull from that domain), or
3855 : * - if IO is not aware of metadata.
3856 : */
3857 : static inline bool
3858 292 : bdev_io_needs_bounce_buffer(struct spdk_bdev_desc *desc, struct spdk_bdev_io *bdev_io)
3859 : {
3860 292 : if (bdev_io_use_memory_domain(bdev_io)) {
3861 4 : if (!desc->memory_domains_supported ||
3862 0 : (bdev_io_needs_sequence_exec(desc, bdev_io) &&
3863 0 : (bdev_io->internal.memory_domain == spdk_accel_get_memory_domain() ||
3864 0 : bdev_io_needs_metadata(desc, bdev_io)))) {
3865 4 : return true;
3866 : }
3867 :
3868 0 : return false;
3869 : }
3870 :
3871 288 : if (bdev_io_needs_metadata(desc, bdev_io)) {
3872 0 : return true;
3873 : }
3874 :
3875 288 : return false;
3876 292 : }
3877 :
3878 : /* We need to allocate fake accel buffer if bdev supports memory domains but IO is not
3879 : * aware of metadata.
3880 : */
3881 : static inline bool
3882 288 : bdev_io_needs_accel_buffer(struct spdk_bdev_desc *desc, struct spdk_bdev_io *bdev_io)
3883 : {
3884 288 : if (bdev_io_needs_metadata(desc, bdev_io)) {
3885 0 : assert(bdev_io_use_memory_domain(bdev_io));
3886 0 : return true;
3887 : }
3888 :
3889 288 : return false;
3890 288 : }
3891 :
3892 : static inline void
3893 292 : _bdev_io_submit_ext(struct spdk_bdev_desc *desc, struct spdk_bdev_io *bdev_io)
3894 : {
3895 292 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
3896 : int rc;
3897 :
3898 292 : if (spdk_unlikely(ch->flags & BDEV_CH_RESET_IN_PROGRESS)) {
3899 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_ABORTED;
3900 0 : bdev_io_complete_unsubmitted(bdev_io);
3901 0 : return;
3902 : }
3903 :
3904 292 : if (bdev_io_needs_metadata(desc, bdev_io)) {
3905 0 : rc = bdev_io_init_dif_ctx(bdev_io);
3906 0 : if (spdk_unlikely(rc != 0)) {
3907 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
3908 0 : bdev_io_complete_unsubmitted(bdev_io);
3909 0 : return;
3910 : }
3911 0 : }
3912 :
3913 292 : if (bdev_io_needs_bounce_buffer(desc, bdev_io)) {
3914 4 : _bdev_io_ext_use_bounce_buffer(bdev_io);
3915 4 : return;
3916 : }
3917 :
3918 288 : if (bdev_io_needs_accel_buffer(desc, bdev_io)) {
3919 0 : _bdev_io_ext_use_accel_buffer(bdev_io);
3920 0 : return;
3921 : }
3922 :
3923 288 : if (bdev_io_needs_sequence_exec(desc, bdev_io)) {
3924 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
3925 0 : bdev_io_exec_sequence(bdev_io, bdev_io_submit_sequence_cb);
3926 0 : return;
3927 : }
3928 : /* For reads we'll execute the sequence after the data is read, so, for now, only
3929 : * clear out accel_sequence pointer and submit the IO */
3930 0 : assert(bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
3931 0 : bdev_io->u.bdev.accel_sequence = NULL;
3932 0 : }
3933 :
3934 288 : bdev_io_submit(bdev_io);
3935 292 : }
3936 :
3937 : static void
3938 11 : bdev_io_submit_reset(struct spdk_bdev_io *bdev_io)
3939 : {
3940 11 : struct spdk_bdev *bdev = bdev_io->bdev;
3941 11 : struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
3942 11 : struct spdk_io_channel *ch = bdev_ch->channel;
3943 :
3944 11 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_PENDING);
3945 :
3946 11 : bdev_io->internal.f.in_submit_request = true;
3947 11 : bdev_submit_request(bdev, ch, bdev_io);
3948 11 : bdev_io->internal.f.in_submit_request = false;
3949 11 : }
3950 :
3951 : void
3952 691 : bdev_io_init(struct spdk_bdev_io *bdev_io,
3953 : struct spdk_bdev *bdev, void *cb_arg,
3954 : spdk_bdev_io_completion_cb cb)
3955 : {
3956 691 : bdev_io->bdev = bdev;
3957 691 : bdev_io->internal.f.raw = 0;
3958 691 : bdev_io->internal.caller_ctx = cb_arg;
3959 691 : bdev_io->internal.cb = cb;
3960 691 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_PENDING;
3961 691 : bdev_io->internal.f.in_submit_request = false;
3962 691 : bdev_io->internal.error.nvme.cdw0 = 0;
3963 691 : bdev_io->num_retries = 0;
3964 691 : bdev_io->internal.get_buf_cb = NULL;
3965 691 : bdev_io->internal.get_aux_buf_cb = NULL;
3966 691 : bdev_io->internal.data_transfer_cpl = NULL;
3967 691 : bdev_io->internal.f.split = bdev_io_should_split(bdev_io);
3968 691 : }
3969 :
3970 : static bool
3971 523 : bdev_io_type_supported(struct spdk_bdev *bdev, enum spdk_bdev_io_type io_type)
3972 : {
3973 523 : return bdev->fn_table->io_type_supported(bdev->ctxt, io_type);
3974 : }
3975 :
3976 : bool
3977 171 : spdk_bdev_io_type_supported(struct spdk_bdev *bdev, enum spdk_bdev_io_type io_type)
3978 : {
3979 : bool supported;
3980 :
3981 171 : supported = bdev_io_type_supported(bdev, io_type);
3982 :
3983 171 : if (!supported) {
3984 7 : switch (io_type) {
3985 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3986 : /* The bdev layer will emulate write zeroes as long as write is supported. */
3987 0 : supported = bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE);
3988 0 : break;
3989 : default:
3990 7 : break;
3991 : }
3992 7 : }
3993 :
3994 171 : return supported;
3995 : }
3996 :
3997 : static const char *g_io_type_strings[] = {
3998 : [SPDK_BDEV_IO_TYPE_READ] = "read",
3999 : [SPDK_BDEV_IO_TYPE_WRITE] = "write",
4000 : [SPDK_BDEV_IO_TYPE_UNMAP] = "unmap",
4001 : [SPDK_BDEV_IO_TYPE_FLUSH] = "flush",
4002 : [SPDK_BDEV_IO_TYPE_RESET] = "reset",
4003 : [SPDK_BDEV_IO_TYPE_NVME_ADMIN] = "nvme_admin",
4004 : [SPDK_BDEV_IO_TYPE_NVME_IO] = "nvme_io",
4005 : [SPDK_BDEV_IO_TYPE_NVME_IO_MD] = "nvme_io_md",
4006 : [SPDK_BDEV_IO_TYPE_WRITE_ZEROES] = "write_zeroes",
4007 : [SPDK_BDEV_IO_TYPE_ZCOPY] = "zcopy",
4008 : [SPDK_BDEV_IO_TYPE_GET_ZONE_INFO] = "get_zone_info",
4009 : [SPDK_BDEV_IO_TYPE_ZONE_MANAGEMENT] = "zone_management",
4010 : [SPDK_BDEV_IO_TYPE_ZONE_APPEND] = "zone_append",
4011 : [SPDK_BDEV_IO_TYPE_COMPARE] = "compare",
4012 : [SPDK_BDEV_IO_TYPE_COMPARE_AND_WRITE] = "compare_and_write",
4013 : [SPDK_BDEV_IO_TYPE_ABORT] = "abort",
4014 : [SPDK_BDEV_IO_TYPE_SEEK_HOLE] = "seek_hole",
4015 : [SPDK_BDEV_IO_TYPE_SEEK_DATA] = "seek_data",
4016 : [SPDK_BDEV_IO_TYPE_COPY] = "copy",
4017 : [SPDK_BDEV_IO_TYPE_NVME_IOV_MD] = "nvme_iov_md",
4018 : };
4019 :
4020 : const char *
4021 0 : spdk_bdev_get_io_type_name(enum spdk_bdev_io_type io_type)
4022 : {
4023 0 : if (io_type <= SPDK_BDEV_IO_TYPE_INVALID || io_type >= SPDK_BDEV_NUM_IO_TYPES) {
4024 0 : return NULL;
4025 : }
4026 :
4027 0 : return g_io_type_strings[io_type];
4028 0 : }
4029 :
4030 : int
4031 0 : spdk_bdev_get_io_type(const char *io_type_string)
4032 : {
4033 : int i;
4034 :
4035 0 : for (i = SPDK_BDEV_IO_TYPE_READ; i < SPDK_BDEV_NUM_IO_TYPES; ++i) {
4036 0 : if (!strcmp(io_type_string, g_io_type_strings[i])) {
4037 0 : return i;
4038 : }
4039 0 : }
4040 :
4041 0 : return -1;
4042 0 : }
4043 :
4044 : uint64_t
4045 0 : spdk_bdev_io_get_submit_tsc(struct spdk_bdev_io *bdev_io)
4046 : {
4047 0 : return bdev_io->internal.submit_tsc;
4048 : }
4049 :
4050 : bool
4051 0 : spdk_bdev_io_has_no_metadata(struct spdk_bdev_io *bdev_io)
4052 : {
4053 0 : return bdev_io->internal.desc->opts.no_metadata;
4054 : }
4055 :
4056 : int
4057 0 : spdk_bdev_dump_info_json(struct spdk_bdev *bdev, struct spdk_json_write_ctx *w)
4058 : {
4059 0 : if (bdev->fn_table->dump_info_json) {
4060 0 : return bdev->fn_table->dump_info_json(bdev->ctxt, w);
4061 : }
4062 :
4063 0 : return 0;
4064 0 : }
4065 :
4066 : static void
4067 10 : bdev_qos_update_max_quota_per_timeslice(struct spdk_bdev_qos *qos)
4068 : {
4069 10 : uint32_t max_per_timeslice = 0;
4070 : int i;
4071 :
4072 50 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
4073 40 : if (qos->rate_limits[i].limit == SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
4074 15 : qos->rate_limits[i].max_per_timeslice = 0;
4075 15 : continue;
4076 : }
4077 :
4078 50 : max_per_timeslice = qos->rate_limits[i].limit *
4079 25 : SPDK_BDEV_QOS_TIMESLICE_IN_USEC / SPDK_SEC_TO_USEC;
4080 :
4081 25 : qos->rate_limits[i].max_per_timeslice = spdk_max(max_per_timeslice,
4082 : qos->rate_limits[i].min_per_timeslice);
4083 :
4084 50 : __atomic_store_n(&qos->rate_limits[i].remaining_this_timeslice,
4085 25 : qos->rate_limits[i].max_per_timeslice, __ATOMIC_RELEASE);
4086 25 : }
4087 :
4088 10 : bdev_qos_set_ops(qos);
4089 10 : }
4090 :
4091 : static void
4092 4 : bdev_channel_submit_qos_io(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
4093 : struct spdk_io_channel *io_ch, void *ctx)
4094 : {
4095 4 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
4096 : int status;
4097 :
4098 4 : bdev_qos_io_submit(bdev_ch, bdev->internal.qos);
4099 :
4100 : /* if all IOs were sent then continue the iteration, otherwise - stop it */
4101 : /* TODO: channels round robing */
4102 4 : status = TAILQ_EMPTY(&bdev_ch->qos_queued_io) ? 0 : 1;
4103 :
4104 4 : spdk_bdev_for_each_channel_continue(i, status);
4105 4 : }
4106 :
4107 :
4108 : static void
4109 2 : bdev_channel_submit_qos_io_done(struct spdk_bdev *bdev, void *ctx, int status)
4110 : {
4111 :
4112 2 : }
4113 :
4114 : static int
4115 3 : bdev_channel_poll_qos(void *arg)
4116 : {
4117 3 : struct spdk_bdev *bdev = arg;
4118 3 : struct spdk_bdev_qos *qos = bdev->internal.qos;
4119 3 : uint64_t now = spdk_get_ticks();
4120 : int i;
4121 : int64_t remaining_last_timeslice;
4122 :
4123 3 : if (spdk_unlikely(qos->thread == NULL)) {
4124 : /* Old QoS was unbound to remove and new QoS is not enabled yet. */
4125 1 : return SPDK_POLLER_IDLE;
4126 : }
4127 :
4128 2 : if (now < (qos->last_timeslice + qos->timeslice_size)) {
4129 : /* We received our callback earlier than expected - return
4130 : * immediately and wait to do accounting until at least one
4131 : * timeslice has actually expired. This should never happen
4132 : * with a well-behaved timer implementation.
4133 : */
4134 0 : return SPDK_POLLER_IDLE;
4135 : }
4136 :
4137 : /* Reset for next round of rate limiting */
4138 10 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
4139 : /* We may have allowed the IOs or bytes to slightly overrun in the last
4140 : * timeslice. remaining_this_timeslice is signed, so if it's negative
4141 : * here, we'll account for the overrun so that the next timeslice will
4142 : * be appropriately reduced.
4143 : */
4144 8 : remaining_last_timeslice = __atomic_exchange_n(&qos->rate_limits[i].remaining_this_timeslice,
4145 : 0, __ATOMIC_RELAXED);
4146 8 : if (remaining_last_timeslice < 0) {
4147 : /* There could be a race condition here as both bdev_qos_rw_queue_io() and bdev_channel_poll_qos()
4148 : * potentially use 2 atomic ops each, so they can intertwine.
4149 : * This race can potentially cause the limits to be a little fuzzy but won't cause any real damage.
4150 : */
4151 0 : __atomic_store_n(&qos->rate_limits[i].remaining_this_timeslice,
4152 0 : remaining_last_timeslice, __ATOMIC_RELAXED);
4153 0 : }
4154 8 : }
4155 :
4156 4 : while (now >= (qos->last_timeslice + qos->timeslice_size)) {
4157 2 : qos->last_timeslice += qos->timeslice_size;
4158 10 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
4159 16 : __atomic_add_fetch(&qos->rate_limits[i].remaining_this_timeslice,
4160 8 : qos->rate_limits[i].max_per_timeslice, __ATOMIC_RELAXED);
4161 8 : }
4162 : }
4163 :
4164 2 : spdk_bdev_for_each_channel(bdev, bdev_channel_submit_qos_io, qos,
4165 : bdev_channel_submit_qos_io_done);
4166 :
4167 2 : return SPDK_POLLER_BUSY;
4168 3 : }
4169 :
4170 : static void
4171 74 : bdev_channel_destroy_resource(struct spdk_bdev_channel *ch)
4172 : {
4173 : struct spdk_bdev_shared_resource *shared_resource;
4174 : struct lba_range *range;
4175 :
4176 74 : bdev_free_io_stat(ch->stat);
4177 : #ifdef SPDK_CONFIG_VTUNE
4178 : bdev_free_io_stat(ch->prev_stat);
4179 : #endif
4180 :
4181 74 : while (!TAILQ_EMPTY(&ch->locked_ranges)) {
4182 0 : range = TAILQ_FIRST(&ch->locked_ranges);
4183 0 : TAILQ_REMOVE(&ch->locked_ranges, range, tailq);
4184 0 : free(range);
4185 : }
4186 :
4187 74 : spdk_put_io_channel(ch->channel);
4188 74 : spdk_put_io_channel(ch->accel_channel);
4189 :
4190 74 : shared_resource = ch->shared_resource;
4191 :
4192 74 : assert(TAILQ_EMPTY(&ch->io_locked));
4193 74 : assert(TAILQ_EMPTY(&ch->io_submitted));
4194 74 : assert(TAILQ_EMPTY(&ch->io_accel_exec));
4195 74 : assert(TAILQ_EMPTY(&ch->io_memory_domain));
4196 74 : assert(ch->io_outstanding == 0);
4197 74 : assert(shared_resource->ref > 0);
4198 74 : shared_resource->ref--;
4199 74 : if (shared_resource->ref == 0) {
4200 73 : assert(shared_resource->io_outstanding == 0);
4201 73 : TAILQ_REMOVE(&shared_resource->mgmt_ch->shared_resources, shared_resource, link);
4202 73 : spdk_put_io_channel(spdk_io_channel_from_ctx(shared_resource->mgmt_ch));
4203 73 : spdk_poller_unregister(&shared_resource->nomem_poller);
4204 73 : free(shared_resource);
4205 73 : }
4206 74 : }
4207 :
4208 : static void
4209 83 : bdev_enable_qos(struct spdk_bdev *bdev, struct spdk_bdev_channel *ch)
4210 : {
4211 83 : struct spdk_bdev_qos *qos = bdev->internal.qos;
4212 : int i;
4213 :
4214 83 : assert(spdk_spin_held(&bdev->internal.spinlock));
4215 :
4216 : /* Rate limiting on this bdev enabled */
4217 83 : if (qos) {
4218 17 : if (qos->ch == NULL) {
4219 : struct spdk_io_channel *io_ch;
4220 :
4221 9 : SPDK_DEBUGLOG(bdev, "Selecting channel %p as QoS channel for bdev %s on thread %p\n", ch,
4222 : bdev->name, spdk_get_thread());
4223 :
4224 : /* No qos channel has been selected, so set one up */
4225 :
4226 : /* Take another reference to ch */
4227 9 : io_ch = spdk_get_io_channel(__bdev_to_io_dev(bdev));
4228 9 : assert(io_ch != NULL);
4229 9 : qos->ch = ch;
4230 :
4231 9 : qos->thread = spdk_io_channel_get_thread(io_ch);
4232 :
4233 45 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
4234 36 : if (bdev_qos_is_iops_rate_limit(i) == true) {
4235 9 : qos->rate_limits[i].min_per_timeslice =
4236 : SPDK_BDEV_QOS_MIN_IO_PER_TIMESLICE;
4237 9 : } else {
4238 27 : qos->rate_limits[i].min_per_timeslice =
4239 : SPDK_BDEV_QOS_MIN_BYTE_PER_TIMESLICE;
4240 : }
4241 :
4242 36 : if (qos->rate_limits[i].limit == 0) {
4243 2 : qos->rate_limits[i].limit = SPDK_BDEV_QOS_LIMIT_NOT_DEFINED;
4244 2 : }
4245 36 : }
4246 9 : bdev_qos_update_max_quota_per_timeslice(qos);
4247 9 : qos->timeslice_size =
4248 9 : SPDK_BDEV_QOS_TIMESLICE_IN_USEC * spdk_get_ticks_hz() / SPDK_SEC_TO_USEC;
4249 9 : qos->last_timeslice = spdk_get_ticks();
4250 9 : qos->poller = SPDK_POLLER_REGISTER(bdev_channel_poll_qos,
4251 : bdev,
4252 : SPDK_BDEV_QOS_TIMESLICE_IN_USEC);
4253 9 : }
4254 :
4255 17 : ch->flags |= BDEV_CH_QOS_ENABLED;
4256 17 : }
4257 83 : }
4258 :
4259 : struct poll_timeout_ctx {
4260 : struct spdk_bdev_desc *desc;
4261 : uint64_t timeout_in_sec;
4262 : spdk_bdev_io_timeout_cb cb_fn;
4263 : void *cb_arg;
4264 : };
4265 :
4266 : static void
4267 268 : bdev_desc_free(struct spdk_bdev_desc *desc)
4268 : {
4269 268 : spdk_spin_destroy(&desc->spinlock);
4270 268 : free(desc->media_events_buffer);
4271 268 : free(desc);
4272 268 : }
4273 :
4274 : static void
4275 8 : bdev_channel_poll_timeout_io_done(struct spdk_bdev *bdev, void *_ctx, int status)
4276 : {
4277 8 : struct poll_timeout_ctx *ctx = _ctx;
4278 8 : struct spdk_bdev_desc *desc = ctx->desc;
4279 :
4280 8 : free(ctx);
4281 :
4282 8 : spdk_spin_lock(&desc->spinlock);
4283 8 : desc->refs--;
4284 8 : if (desc->closed == true && desc->refs == 0) {
4285 1 : spdk_spin_unlock(&desc->spinlock);
4286 1 : bdev_desc_free(desc);
4287 1 : return;
4288 : }
4289 7 : spdk_spin_unlock(&desc->spinlock);
4290 8 : }
4291 :
4292 : static void
4293 13 : bdev_channel_poll_timeout_io(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
4294 : struct spdk_io_channel *io_ch, void *_ctx)
4295 : {
4296 13 : struct poll_timeout_ctx *ctx = _ctx;
4297 13 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
4298 13 : struct spdk_bdev_desc *desc = ctx->desc;
4299 : struct spdk_bdev_io *bdev_io;
4300 : uint64_t now;
4301 :
4302 13 : spdk_spin_lock(&desc->spinlock);
4303 13 : if (desc->closed == true) {
4304 1 : spdk_spin_unlock(&desc->spinlock);
4305 1 : spdk_bdev_for_each_channel_continue(i, -1);
4306 1 : return;
4307 : }
4308 12 : spdk_spin_unlock(&desc->spinlock);
4309 :
4310 12 : now = spdk_get_ticks();
4311 22 : TAILQ_FOREACH(bdev_io, &bdev_ch->io_submitted, internal.ch_link) {
4312 : /* Exclude any I/O that are generated via splitting. */
4313 15 : if (bdev_io->internal.cb == bdev_io_split_done) {
4314 3 : continue;
4315 : }
4316 :
4317 : /* Once we find an I/O that has not timed out, we can immediately
4318 : * exit the loop.
4319 : */
4320 24 : if (now < (bdev_io->internal.submit_tsc +
4321 12 : ctx->timeout_in_sec * spdk_get_ticks_hz())) {
4322 5 : goto end;
4323 : }
4324 :
4325 7 : if (bdev_io->internal.desc == desc) {
4326 7 : ctx->cb_fn(ctx->cb_arg, bdev_io);
4327 7 : }
4328 14 : }
4329 :
4330 : end:
4331 12 : spdk_bdev_for_each_channel_continue(i, 0);
4332 13 : }
4333 :
4334 : static int
4335 8 : bdev_poll_timeout_io(void *arg)
4336 : {
4337 8 : struct spdk_bdev_desc *desc = arg;
4338 8 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
4339 : struct poll_timeout_ctx *ctx;
4340 :
4341 8 : ctx = calloc(1, sizeof(struct poll_timeout_ctx));
4342 8 : if (!ctx) {
4343 0 : SPDK_ERRLOG("failed to allocate memory\n");
4344 0 : return SPDK_POLLER_BUSY;
4345 : }
4346 8 : ctx->desc = desc;
4347 8 : ctx->cb_arg = desc->cb_arg;
4348 8 : ctx->cb_fn = desc->cb_fn;
4349 8 : ctx->timeout_in_sec = desc->timeout_in_sec;
4350 :
4351 : /* Take a ref on the descriptor in case it gets closed while we are checking
4352 : * all of the channels.
4353 : */
4354 8 : spdk_spin_lock(&desc->spinlock);
4355 8 : desc->refs++;
4356 8 : spdk_spin_unlock(&desc->spinlock);
4357 :
4358 8 : spdk_bdev_for_each_channel(bdev, bdev_channel_poll_timeout_io, ctx,
4359 : bdev_channel_poll_timeout_io_done);
4360 :
4361 8 : return SPDK_POLLER_BUSY;
4362 8 : }
4363 :
4364 : int
4365 5 : spdk_bdev_set_timeout(struct spdk_bdev_desc *desc, uint64_t timeout_in_sec,
4366 : spdk_bdev_io_timeout_cb cb_fn, void *cb_arg)
4367 : {
4368 5 : assert(desc->thread == spdk_get_thread());
4369 :
4370 5 : spdk_poller_unregister(&desc->io_timeout_poller);
4371 :
4372 5 : if (timeout_in_sec) {
4373 4 : assert(cb_fn != NULL);
4374 4 : desc->io_timeout_poller = SPDK_POLLER_REGISTER(bdev_poll_timeout_io,
4375 : desc,
4376 : SPDK_BDEV_IO_POLL_INTERVAL_IN_MSEC * SPDK_SEC_TO_USEC /
4377 : 1000);
4378 4 : if (desc->io_timeout_poller == NULL) {
4379 0 : SPDK_ERRLOG("can not register the desc timeout IO poller\n");
4380 0 : return -1;
4381 : }
4382 4 : }
4383 :
4384 5 : desc->cb_fn = cb_fn;
4385 5 : desc->cb_arg = cb_arg;
4386 5 : desc->timeout_in_sec = timeout_in_sec;
4387 :
4388 5 : return 0;
4389 5 : }
4390 :
4391 : static int
4392 76 : bdev_channel_create(void *io_device, void *ctx_buf)
4393 : {
4394 76 : struct spdk_bdev *bdev = __bdev_from_io_dev(io_device);
4395 76 : struct spdk_bdev_channel *ch = ctx_buf;
4396 : struct spdk_io_channel *mgmt_io_ch;
4397 : struct spdk_bdev_mgmt_channel *mgmt_ch;
4398 : struct spdk_bdev_shared_resource *shared_resource;
4399 : struct lba_range *range;
4400 :
4401 76 : ch->bdev = bdev;
4402 76 : ch->channel = bdev->fn_table->get_io_channel(bdev->ctxt);
4403 76 : if (!ch->channel) {
4404 2 : return -1;
4405 : }
4406 :
4407 74 : ch->accel_channel = spdk_accel_get_io_channel();
4408 74 : if (!ch->accel_channel) {
4409 0 : spdk_put_io_channel(ch->channel);
4410 0 : return -1;
4411 : }
4412 :
4413 74 : spdk_trace_record(TRACE_BDEV_IOCH_CREATE, bdev->internal.trace_id, 0, 0,
4414 : spdk_thread_get_id(spdk_io_channel_get_thread(ch->channel)));
4415 :
4416 74 : assert(ch->histogram == NULL);
4417 74 : if (bdev->internal.histogram_enabled) {
4418 0 : ch->histogram = spdk_histogram_data_alloc();
4419 0 : if (ch->histogram == NULL) {
4420 0 : SPDK_ERRLOG("Could not allocate histogram\n");
4421 0 : }
4422 0 : }
4423 :
4424 74 : mgmt_io_ch = spdk_get_io_channel(&g_bdev_mgr);
4425 74 : if (!mgmt_io_ch) {
4426 0 : spdk_put_io_channel(ch->channel);
4427 0 : spdk_put_io_channel(ch->accel_channel);
4428 0 : return -1;
4429 : }
4430 :
4431 74 : mgmt_ch = __io_ch_to_bdev_mgmt_ch(mgmt_io_ch);
4432 76 : TAILQ_FOREACH(shared_resource, &mgmt_ch->shared_resources, link) {
4433 3 : if (shared_resource->shared_ch == ch->channel) {
4434 1 : spdk_put_io_channel(mgmt_io_ch);
4435 1 : shared_resource->ref++;
4436 1 : break;
4437 : }
4438 2 : }
4439 :
4440 74 : if (shared_resource == NULL) {
4441 73 : shared_resource = calloc(1, sizeof(*shared_resource));
4442 73 : if (shared_resource == NULL) {
4443 0 : spdk_put_io_channel(ch->channel);
4444 0 : spdk_put_io_channel(ch->accel_channel);
4445 0 : spdk_put_io_channel(mgmt_io_ch);
4446 0 : return -1;
4447 : }
4448 :
4449 73 : shared_resource->mgmt_ch = mgmt_ch;
4450 73 : shared_resource->io_outstanding = 0;
4451 73 : TAILQ_INIT(&shared_resource->nomem_io);
4452 73 : shared_resource->nomem_threshold = 0;
4453 73 : shared_resource->shared_ch = ch->channel;
4454 73 : shared_resource->ref = 1;
4455 73 : TAILQ_INSERT_TAIL(&mgmt_ch->shared_resources, shared_resource, link);
4456 73 : }
4457 :
4458 74 : ch->io_outstanding = 0;
4459 74 : TAILQ_INIT(&ch->queued_resets);
4460 74 : TAILQ_INIT(&ch->locked_ranges);
4461 74 : TAILQ_INIT(&ch->qos_queued_io);
4462 74 : ch->flags = 0;
4463 74 : ch->trace_id = bdev->internal.trace_id;
4464 74 : ch->shared_resource = shared_resource;
4465 :
4466 74 : TAILQ_INIT(&ch->io_submitted);
4467 74 : TAILQ_INIT(&ch->io_locked);
4468 74 : TAILQ_INIT(&ch->io_accel_exec);
4469 74 : TAILQ_INIT(&ch->io_memory_domain);
4470 :
4471 74 : ch->stat = bdev_alloc_io_stat(false);
4472 74 : if (ch->stat == NULL) {
4473 0 : bdev_channel_destroy_resource(ch);
4474 0 : return -1;
4475 : }
4476 :
4477 74 : ch->stat->ticks_rate = spdk_get_ticks_hz();
4478 :
4479 : #ifdef SPDK_CONFIG_VTUNE
4480 : {
4481 : char *name;
4482 : __itt_init_ittlib(NULL, 0);
4483 : name = spdk_sprintf_alloc("spdk_bdev_%s_%p", ch->bdev->name, ch);
4484 : if (!name) {
4485 : bdev_channel_destroy_resource(ch);
4486 : return -1;
4487 : }
4488 : ch->handle = __itt_string_handle_create(name);
4489 : free(name);
4490 : ch->start_tsc = spdk_get_ticks();
4491 : ch->interval_tsc = spdk_get_ticks_hz() / 100;
4492 : ch->prev_stat = bdev_alloc_io_stat(false);
4493 : if (ch->prev_stat == NULL) {
4494 : bdev_channel_destroy_resource(ch);
4495 : return -1;
4496 : }
4497 : }
4498 : #endif
4499 :
4500 74 : spdk_spin_lock(&bdev->internal.spinlock);
4501 74 : bdev_enable_qos(bdev, ch);
4502 :
4503 75 : TAILQ_FOREACH(range, &bdev->internal.locked_ranges, tailq) {
4504 : struct lba_range *new_range;
4505 :
4506 1 : new_range = calloc(1, sizeof(*new_range));
4507 1 : if (new_range == NULL) {
4508 0 : spdk_spin_unlock(&bdev->internal.spinlock);
4509 0 : bdev_channel_destroy_resource(ch);
4510 0 : return -1;
4511 : }
4512 1 : new_range->length = range->length;
4513 1 : new_range->offset = range->offset;
4514 1 : new_range->locked_ctx = range->locked_ctx;
4515 1 : TAILQ_INSERT_TAIL(&ch->locked_ranges, new_range, tailq);
4516 1 : }
4517 :
4518 74 : spdk_spin_unlock(&bdev->internal.spinlock);
4519 :
4520 74 : return 0;
4521 76 : }
4522 :
4523 : static int
4524 0 : bdev_abort_all_buf_io_cb(struct spdk_iobuf_channel *ch, struct spdk_iobuf_entry *entry,
4525 : void *cb_ctx)
4526 : {
4527 0 : struct spdk_bdev_channel *bdev_ch = cb_ctx;
4528 : struct spdk_bdev_io *bdev_io;
4529 : uint64_t buf_len;
4530 :
4531 0 : bdev_io = SPDK_CONTAINEROF(entry, struct spdk_bdev_io, internal.iobuf);
4532 0 : if (bdev_io->internal.ch == bdev_ch) {
4533 0 : buf_len = bdev_io_get_max_buf_len(bdev_io, bdev_io->internal.buf.len);
4534 0 : spdk_iobuf_entry_abort(ch, entry, buf_len);
4535 0 : spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_ABORTED);
4536 0 : }
4537 :
4538 0 : return 0;
4539 : }
4540 :
4541 : /*
4542 : * Abort I/O that are waiting on a data buffer.
4543 : */
4544 : static void
4545 96 : bdev_abort_all_buf_io(struct spdk_bdev_mgmt_channel *mgmt_ch, struct spdk_bdev_channel *ch)
4546 : {
4547 96 : spdk_iobuf_for_each_entry(&mgmt_ch->iobuf, bdev_abort_all_buf_io_cb, ch);
4548 96 : }
4549 :
4550 : /*
4551 : * Abort I/O that are queued waiting for submission. These types of I/O are
4552 : * linked using the spdk_bdev_io link TAILQ_ENTRY.
4553 : */
4554 : static void
4555 188 : bdev_abort_all_queued_io(bdev_io_tailq_t *queue, struct spdk_bdev_channel *ch)
4556 : {
4557 : struct spdk_bdev_io *bdev_io, *tmp;
4558 :
4559 228 : TAILQ_FOREACH_SAFE(bdev_io, queue, internal.link, tmp) {
4560 40 : if (bdev_io->internal.ch == ch) {
4561 40 : TAILQ_REMOVE(queue, bdev_io, internal.link);
4562 : /*
4563 : * spdk_bdev_io_complete() assumes that the completed I/O had
4564 : * been submitted to the bdev module. Since in this case it
4565 : * hadn't, bump io_outstanding to account for the decrement
4566 : * that spdk_bdev_io_complete() will do.
4567 : */
4568 40 : if (bdev_io->type != SPDK_BDEV_IO_TYPE_RESET) {
4569 39 : bdev_io_increment_outstanding(ch, ch->shared_resource);
4570 39 : }
4571 40 : spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_ABORTED);
4572 40 : }
4573 40 : }
4574 188 : }
4575 :
4576 : static bool
4577 18 : bdev_abort_queued_io(bdev_io_tailq_t *queue, struct spdk_bdev_io *bio_to_abort)
4578 : {
4579 : struct spdk_bdev_io *bdev_io;
4580 :
4581 18 : TAILQ_FOREACH(bdev_io, queue, internal.link) {
4582 0 : if (bdev_io == bio_to_abort) {
4583 0 : TAILQ_REMOVE(queue, bio_to_abort, internal.link);
4584 0 : spdk_bdev_io_complete(bio_to_abort, SPDK_BDEV_IO_STATUS_ABORTED);
4585 0 : return true;
4586 : }
4587 0 : }
4588 :
4589 18 : return false;
4590 18 : }
4591 :
4592 : static int
4593 0 : bdev_abort_buf_io_cb(struct spdk_iobuf_channel *ch, struct spdk_iobuf_entry *entry, void *cb_ctx)
4594 : {
4595 0 : struct spdk_bdev_io *bdev_io, *bio_to_abort = cb_ctx;
4596 : uint64_t buf_len;
4597 :
4598 0 : bdev_io = SPDK_CONTAINEROF(entry, struct spdk_bdev_io, internal.iobuf);
4599 0 : if (bdev_io == bio_to_abort) {
4600 0 : buf_len = bdev_io_get_max_buf_len(bdev_io, bdev_io->internal.buf.len);
4601 0 : spdk_iobuf_entry_abort(ch, entry, buf_len);
4602 0 : spdk_bdev_io_complete(bio_to_abort, SPDK_BDEV_IO_STATUS_ABORTED);
4603 0 : return 1;
4604 : }
4605 :
4606 0 : return 0;
4607 0 : }
4608 :
4609 : static bool
4610 16 : bdev_abort_buf_io(struct spdk_bdev_mgmt_channel *mgmt_ch, struct spdk_bdev_io *bio_to_abort)
4611 : {
4612 : int rc;
4613 :
4614 16 : rc = spdk_iobuf_for_each_entry(&mgmt_ch->iobuf, bdev_abort_buf_io_cb, bio_to_abort);
4615 16 : return rc == 1;
4616 : }
4617 :
4618 : static void
4619 7 : bdev_qos_channel_destroy(void *cb_arg)
4620 : {
4621 7 : struct spdk_bdev_qos *qos = cb_arg;
4622 :
4623 7 : spdk_put_io_channel(spdk_io_channel_from_ctx(qos->ch));
4624 7 : spdk_poller_unregister(&qos->poller);
4625 :
4626 7 : SPDK_DEBUGLOG(bdev, "Free QoS %p.\n", qos);
4627 :
4628 7 : free(qos);
4629 7 : }
4630 :
4631 : static int
4632 7 : bdev_qos_destroy(struct spdk_bdev *bdev)
4633 : {
4634 : int i;
4635 :
4636 : /*
4637 : * Cleanly shutting down the QoS poller is tricky, because
4638 : * during the asynchronous operation the user could open
4639 : * a new descriptor and create a new channel, spawning
4640 : * a new QoS poller.
4641 : *
4642 : * The strategy is to create a new QoS structure here and swap it
4643 : * in. The shutdown path then continues to refer to the old one
4644 : * until it completes and then releases it.
4645 : */
4646 : struct spdk_bdev_qos *new_qos, *old_qos;
4647 :
4648 7 : old_qos = bdev->internal.qos;
4649 :
4650 7 : new_qos = calloc(1, sizeof(*new_qos));
4651 7 : if (!new_qos) {
4652 0 : SPDK_ERRLOG("Unable to allocate memory to shut down QoS.\n");
4653 0 : return -ENOMEM;
4654 : }
4655 :
4656 : /* Copy the old QoS data into the newly allocated structure */
4657 7 : memcpy(new_qos, old_qos, sizeof(*new_qos));
4658 :
4659 : /* Zero out the key parts of the QoS structure */
4660 7 : new_qos->ch = NULL;
4661 7 : new_qos->thread = NULL;
4662 7 : new_qos->poller = NULL;
4663 : /*
4664 : * The limit member of spdk_bdev_qos_limit structure is not zeroed.
4665 : * It will be used later for the new QoS structure.
4666 : */
4667 35 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
4668 28 : new_qos->rate_limits[i].remaining_this_timeslice = 0;
4669 28 : new_qos->rate_limits[i].min_per_timeslice = 0;
4670 28 : new_qos->rate_limits[i].max_per_timeslice = 0;
4671 28 : }
4672 :
4673 7 : bdev->internal.qos = new_qos;
4674 :
4675 7 : if (old_qos->thread == NULL) {
4676 0 : free(old_qos);
4677 0 : } else {
4678 7 : spdk_thread_send_msg(old_qos->thread, bdev_qos_channel_destroy, old_qos);
4679 : }
4680 :
4681 : /* It is safe to continue with destroying the bdev even though the QoS channel hasn't
4682 : * been destroyed yet. The destruction path will end up waiting for the final
4683 : * channel to be put before it releases resources. */
4684 :
4685 7 : return 0;
4686 7 : }
4687 :
4688 : void
4689 78 : spdk_bdev_add_io_stat(struct spdk_bdev_io_stat *total, struct spdk_bdev_io_stat *add)
4690 : {
4691 78 : total->bytes_read += add->bytes_read;
4692 78 : total->num_read_ops += add->num_read_ops;
4693 78 : total->bytes_written += add->bytes_written;
4694 78 : total->num_write_ops += add->num_write_ops;
4695 78 : total->bytes_unmapped += add->bytes_unmapped;
4696 78 : total->num_unmap_ops += add->num_unmap_ops;
4697 78 : total->bytes_copied += add->bytes_copied;
4698 78 : total->num_copy_ops += add->num_copy_ops;
4699 78 : total->read_latency_ticks += add->read_latency_ticks;
4700 78 : total->write_latency_ticks += add->write_latency_ticks;
4701 78 : total->unmap_latency_ticks += add->unmap_latency_ticks;
4702 78 : total->copy_latency_ticks += add->copy_latency_ticks;
4703 78 : if (total->max_read_latency_ticks < add->max_read_latency_ticks) {
4704 7 : total->max_read_latency_ticks = add->max_read_latency_ticks;
4705 7 : }
4706 78 : if (total->min_read_latency_ticks > add->min_read_latency_ticks) {
4707 39 : total->min_read_latency_ticks = add->min_read_latency_ticks;
4708 39 : }
4709 78 : if (total->max_write_latency_ticks < add->max_write_latency_ticks) {
4710 4 : total->max_write_latency_ticks = add->max_write_latency_ticks;
4711 4 : }
4712 78 : if (total->min_write_latency_ticks > add->min_write_latency_ticks) {
4713 24 : total->min_write_latency_ticks = add->min_write_latency_ticks;
4714 24 : }
4715 78 : if (total->max_unmap_latency_ticks < add->max_unmap_latency_ticks) {
4716 0 : total->max_unmap_latency_ticks = add->max_unmap_latency_ticks;
4717 0 : }
4718 78 : if (total->min_unmap_latency_ticks > add->min_unmap_latency_ticks) {
4719 3 : total->min_unmap_latency_ticks = add->min_unmap_latency_ticks;
4720 3 : }
4721 78 : if (total->max_copy_latency_ticks < add->max_copy_latency_ticks) {
4722 0 : total->max_copy_latency_ticks = add->max_copy_latency_ticks;
4723 0 : }
4724 78 : if (total->min_copy_latency_ticks > add->min_copy_latency_ticks) {
4725 4 : total->min_copy_latency_ticks = add->min_copy_latency_ticks;
4726 4 : }
4727 78 : }
4728 :
4729 : static void
4730 5 : bdev_get_io_stat(struct spdk_bdev_io_stat *to_stat, struct spdk_bdev_io_stat *from_stat)
4731 : {
4732 5 : memcpy(to_stat, from_stat, offsetof(struct spdk_bdev_io_stat, io_error));
4733 :
4734 5 : if (to_stat->io_error != NULL && from_stat->io_error != NULL) {
4735 0 : memcpy(to_stat->io_error, from_stat->io_error,
4736 : sizeof(struct spdk_bdev_io_error_stat));
4737 0 : }
4738 5 : }
4739 :
4740 : void
4741 211 : spdk_bdev_reset_io_stat(struct spdk_bdev_io_stat *stat, enum spdk_bdev_reset_stat_mode mode)
4742 : {
4743 211 : if (mode == SPDK_BDEV_RESET_STAT_NONE) {
4744 5 : return;
4745 : }
4746 :
4747 206 : stat->max_read_latency_ticks = 0;
4748 206 : stat->min_read_latency_ticks = UINT64_MAX;
4749 206 : stat->max_write_latency_ticks = 0;
4750 206 : stat->min_write_latency_ticks = UINT64_MAX;
4751 206 : stat->max_unmap_latency_ticks = 0;
4752 206 : stat->min_unmap_latency_ticks = UINT64_MAX;
4753 206 : stat->max_copy_latency_ticks = 0;
4754 206 : stat->min_copy_latency_ticks = UINT64_MAX;
4755 :
4756 206 : if (mode != SPDK_BDEV_RESET_STAT_ALL) {
4757 2 : return;
4758 : }
4759 :
4760 204 : stat->bytes_read = 0;
4761 204 : stat->num_read_ops = 0;
4762 204 : stat->bytes_written = 0;
4763 204 : stat->num_write_ops = 0;
4764 204 : stat->bytes_unmapped = 0;
4765 204 : stat->num_unmap_ops = 0;
4766 204 : stat->bytes_copied = 0;
4767 204 : stat->num_copy_ops = 0;
4768 204 : stat->read_latency_ticks = 0;
4769 204 : stat->write_latency_ticks = 0;
4770 204 : stat->unmap_latency_ticks = 0;
4771 204 : stat->copy_latency_ticks = 0;
4772 :
4773 204 : if (stat->io_error != NULL) {
4774 129 : memset(stat->io_error, 0, sizeof(struct spdk_bdev_io_error_stat));
4775 129 : }
4776 211 : }
4777 :
4778 : struct spdk_bdev_io_stat *
4779 202 : bdev_alloc_io_stat(bool io_error_stat)
4780 : {
4781 : struct spdk_bdev_io_stat *stat;
4782 :
4783 202 : stat = malloc(sizeof(struct spdk_bdev_io_stat));
4784 202 : if (stat == NULL) {
4785 0 : return NULL;
4786 : }
4787 :
4788 202 : if (io_error_stat) {
4789 128 : stat->io_error = malloc(sizeof(struct spdk_bdev_io_error_stat));
4790 128 : if (stat->io_error == NULL) {
4791 0 : free(stat);
4792 0 : return NULL;
4793 : }
4794 128 : } else {
4795 74 : stat->io_error = NULL;
4796 : }
4797 :
4798 202 : spdk_bdev_reset_io_stat(stat, SPDK_BDEV_RESET_STAT_ALL);
4799 :
4800 202 : return stat;
4801 202 : }
4802 :
4803 : void
4804 202 : bdev_free_io_stat(struct spdk_bdev_io_stat *stat)
4805 : {
4806 202 : if (stat != NULL) {
4807 202 : free(stat->io_error);
4808 202 : free(stat);
4809 202 : }
4810 202 : }
4811 :
4812 : void
4813 0 : spdk_bdev_dump_io_stat_json(struct spdk_bdev_io_stat *stat, struct spdk_json_write_ctx *w)
4814 : {
4815 : int i;
4816 :
4817 0 : spdk_json_write_named_uint64(w, "bytes_read", stat->bytes_read);
4818 0 : spdk_json_write_named_uint64(w, "num_read_ops", stat->num_read_ops);
4819 0 : spdk_json_write_named_uint64(w, "bytes_written", stat->bytes_written);
4820 0 : spdk_json_write_named_uint64(w, "num_write_ops", stat->num_write_ops);
4821 0 : spdk_json_write_named_uint64(w, "bytes_unmapped", stat->bytes_unmapped);
4822 0 : spdk_json_write_named_uint64(w, "num_unmap_ops", stat->num_unmap_ops);
4823 0 : spdk_json_write_named_uint64(w, "bytes_copied", stat->bytes_copied);
4824 0 : spdk_json_write_named_uint64(w, "num_copy_ops", stat->num_copy_ops);
4825 0 : spdk_json_write_named_uint64(w, "read_latency_ticks", stat->read_latency_ticks);
4826 0 : spdk_json_write_named_uint64(w, "max_read_latency_ticks", stat->max_read_latency_ticks);
4827 0 : spdk_json_write_named_uint64(w, "min_read_latency_ticks",
4828 0 : stat->min_read_latency_ticks != UINT64_MAX ?
4829 0 : stat->min_read_latency_ticks : 0);
4830 0 : spdk_json_write_named_uint64(w, "write_latency_ticks", stat->write_latency_ticks);
4831 0 : spdk_json_write_named_uint64(w, "max_write_latency_ticks", stat->max_write_latency_ticks);
4832 0 : spdk_json_write_named_uint64(w, "min_write_latency_ticks",
4833 0 : stat->min_write_latency_ticks != UINT64_MAX ?
4834 0 : stat->min_write_latency_ticks : 0);
4835 0 : spdk_json_write_named_uint64(w, "unmap_latency_ticks", stat->unmap_latency_ticks);
4836 0 : spdk_json_write_named_uint64(w, "max_unmap_latency_ticks", stat->max_unmap_latency_ticks);
4837 0 : spdk_json_write_named_uint64(w, "min_unmap_latency_ticks",
4838 0 : stat->min_unmap_latency_ticks != UINT64_MAX ?
4839 0 : stat->min_unmap_latency_ticks : 0);
4840 0 : spdk_json_write_named_uint64(w, "copy_latency_ticks", stat->copy_latency_ticks);
4841 0 : spdk_json_write_named_uint64(w, "max_copy_latency_ticks", stat->max_copy_latency_ticks);
4842 0 : spdk_json_write_named_uint64(w, "min_copy_latency_ticks",
4843 0 : stat->min_copy_latency_ticks != UINT64_MAX ?
4844 0 : stat->min_copy_latency_ticks : 0);
4845 :
4846 0 : if (stat->io_error != NULL) {
4847 0 : spdk_json_write_named_object_begin(w, "io_error");
4848 0 : for (i = 0; i < -SPDK_MIN_BDEV_IO_STATUS; i++) {
4849 0 : if (stat->io_error->error_status[i] != 0) {
4850 0 : spdk_json_write_named_uint32(w, bdev_io_status_get_string(-(i + 1)),
4851 0 : stat->io_error->error_status[i]);
4852 0 : }
4853 0 : }
4854 0 : spdk_json_write_object_end(w);
4855 0 : }
4856 0 : }
4857 :
4858 : static void
4859 78 : bdev_channel_abort_queued_ios(struct spdk_bdev_channel *ch)
4860 : {
4861 78 : struct spdk_bdev_shared_resource *shared_resource = ch->shared_resource;
4862 78 : struct spdk_bdev_mgmt_channel *mgmt_ch = shared_resource->mgmt_ch;
4863 :
4864 78 : bdev_abort_all_queued_io(&shared_resource->nomem_io, ch);
4865 78 : bdev_abort_all_buf_io(mgmt_ch, ch);
4866 78 : }
4867 :
4868 : static void
4869 74 : bdev_channel_destroy(void *io_device, void *ctx_buf)
4870 : {
4871 74 : struct spdk_bdev_channel *ch = ctx_buf;
4872 :
4873 74 : SPDK_DEBUGLOG(bdev, "Destroying channel %p for bdev %s on thread %p\n", ch, ch->bdev->name,
4874 : spdk_get_thread());
4875 :
4876 74 : spdk_trace_record(TRACE_BDEV_IOCH_DESTROY, ch->bdev->internal.trace_id, 0, 0,
4877 : spdk_thread_get_id(spdk_io_channel_get_thread(ch->channel)));
4878 :
4879 : /* This channel is going away, so add its statistics into the bdev so that they don't get lost. */
4880 74 : spdk_spin_lock(&ch->bdev->internal.spinlock);
4881 74 : spdk_bdev_add_io_stat(ch->bdev->internal.stat, ch->stat);
4882 74 : spdk_spin_unlock(&ch->bdev->internal.spinlock);
4883 :
4884 74 : bdev_abort_all_queued_io(&ch->queued_resets, ch);
4885 :
4886 74 : bdev_channel_abort_queued_ios(ch);
4887 :
4888 74 : if (ch->histogram) {
4889 0 : spdk_histogram_data_free(ch->histogram);
4890 0 : }
4891 :
4892 74 : bdev_channel_destroy_resource(ch);
4893 74 : }
4894 :
4895 : /*
4896 : * If the name already exists in the global bdev name tree, RB_INSERT() returns a pointer
4897 : * to it. Hence we do not have to call bdev_get_by_name() when using this function.
4898 : */
4899 : static int
4900 259 : bdev_name_add(struct spdk_bdev_name *bdev_name, struct spdk_bdev *bdev, const char *name)
4901 : {
4902 : struct spdk_bdev_name *tmp;
4903 :
4904 259 : bdev_name->name = strdup(name);
4905 259 : if (bdev_name->name == NULL) {
4906 0 : SPDK_ERRLOG("Unable to allocate bdev name\n");
4907 0 : return -ENOMEM;
4908 : }
4909 :
4910 259 : bdev_name->bdev = bdev;
4911 :
4912 259 : spdk_spin_lock(&g_bdev_mgr.spinlock);
4913 259 : tmp = RB_INSERT(bdev_name_tree, &g_bdev_mgr.bdev_names, bdev_name);
4914 259 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
4915 :
4916 259 : if (tmp != NULL) {
4917 4 : SPDK_ERRLOG("Bdev name %s already exists\n", name);
4918 4 : free(bdev_name->name);
4919 4 : return -EEXIST;
4920 : }
4921 :
4922 255 : return 0;
4923 259 : }
4924 :
4925 : static void
4926 255 : bdev_name_del_unsafe(struct spdk_bdev_name *bdev_name)
4927 : {
4928 255 : RB_REMOVE(bdev_name_tree, &g_bdev_mgr.bdev_names, bdev_name);
4929 255 : free(bdev_name->name);
4930 255 : }
4931 :
4932 : static void
4933 5 : bdev_name_del(struct spdk_bdev_name *bdev_name)
4934 : {
4935 5 : spdk_spin_lock(&g_bdev_mgr.spinlock);
4936 5 : bdev_name_del_unsafe(bdev_name);
4937 5 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
4938 5 : }
4939 :
4940 : int
4941 134 : spdk_bdev_alias_add(struct spdk_bdev *bdev, const char *alias)
4942 : {
4943 : struct spdk_bdev_alias *tmp;
4944 : int ret;
4945 :
4946 134 : if (alias == NULL) {
4947 1 : SPDK_ERRLOG("Empty alias passed\n");
4948 1 : return -EINVAL;
4949 : }
4950 :
4951 133 : tmp = calloc(1, sizeof(*tmp));
4952 133 : if (tmp == NULL) {
4953 0 : SPDK_ERRLOG("Unable to allocate alias\n");
4954 0 : return -ENOMEM;
4955 : }
4956 :
4957 133 : ret = bdev_name_add(&tmp->alias, bdev, alias);
4958 133 : if (ret != 0) {
4959 4 : free(tmp);
4960 4 : return ret;
4961 : }
4962 :
4963 129 : TAILQ_INSERT_TAIL(&bdev->aliases, tmp, tailq);
4964 :
4965 129 : return 0;
4966 134 : }
4967 :
4968 : static int
4969 130 : bdev_alias_del(struct spdk_bdev *bdev, const char *alias,
4970 : void (*alias_del_fn)(struct spdk_bdev_name *n))
4971 : {
4972 : struct spdk_bdev_alias *tmp;
4973 :
4974 135 : TAILQ_FOREACH(tmp, &bdev->aliases, tailq) {
4975 131 : if (strcmp(alias, tmp->alias.name) == 0) {
4976 126 : TAILQ_REMOVE(&bdev->aliases, tmp, tailq);
4977 126 : alias_del_fn(&tmp->alias);
4978 126 : free(tmp);
4979 126 : return 0;
4980 : }
4981 5 : }
4982 :
4983 4 : return -ENOENT;
4984 130 : }
4985 :
4986 : int
4987 4 : spdk_bdev_alias_del(struct spdk_bdev *bdev, const char *alias)
4988 : {
4989 : int rc;
4990 :
4991 4 : rc = bdev_alias_del(bdev, alias, bdev_name_del);
4992 4 : if (rc == -ENOENT) {
4993 2 : SPDK_INFOLOG(bdev, "Alias %s does not exist\n", alias);
4994 2 : }
4995 :
4996 4 : return rc;
4997 : }
4998 :
4999 : void
5000 2 : spdk_bdev_alias_del_all(struct spdk_bdev *bdev)
5001 : {
5002 : struct spdk_bdev_alias *p, *tmp;
5003 :
5004 5 : TAILQ_FOREACH_SAFE(p, &bdev->aliases, tailq, tmp) {
5005 3 : TAILQ_REMOVE(&bdev->aliases, p, tailq);
5006 3 : bdev_name_del(&p->alias);
5007 3 : free(p);
5008 3 : }
5009 2 : }
5010 :
5011 : struct spdk_io_channel *
5012 76 : spdk_bdev_get_io_channel(struct spdk_bdev_desc *desc)
5013 : {
5014 76 : return spdk_get_io_channel(__bdev_to_io_dev(spdk_bdev_desc_get_bdev(desc)));
5015 : }
5016 :
5017 : void *
5018 0 : spdk_bdev_get_module_ctx(struct spdk_bdev_desc *desc)
5019 : {
5020 0 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5021 0 : void *ctx = NULL;
5022 :
5023 0 : if (bdev->fn_table->get_module_ctx) {
5024 0 : ctx = bdev->fn_table->get_module_ctx(bdev->ctxt);
5025 0 : }
5026 :
5027 0 : return ctx;
5028 : }
5029 :
5030 : const char *
5031 0 : spdk_bdev_get_module_name(const struct spdk_bdev *bdev)
5032 : {
5033 0 : return bdev->module->name;
5034 : }
5035 :
5036 : const char *
5037 255 : spdk_bdev_get_name(const struct spdk_bdev *bdev)
5038 : {
5039 255 : return bdev->name;
5040 : }
5041 :
5042 : const char *
5043 0 : spdk_bdev_get_product_name(const struct spdk_bdev *bdev)
5044 : {
5045 0 : return bdev->product_name;
5046 : }
5047 :
5048 : const struct spdk_bdev_aliases_list *
5049 0 : spdk_bdev_get_aliases(const struct spdk_bdev *bdev)
5050 : {
5051 0 : return &bdev->aliases;
5052 : }
5053 :
5054 : uint32_t
5055 5 : spdk_bdev_get_block_size(const struct spdk_bdev *bdev)
5056 : {
5057 5 : return bdev->blocklen;
5058 : }
5059 :
5060 : uint32_t
5061 0 : spdk_bdev_get_write_unit_size(const struct spdk_bdev *bdev)
5062 : {
5063 0 : return bdev->write_unit_size;
5064 : }
5065 :
5066 : uint64_t
5067 0 : spdk_bdev_get_num_blocks(const struct spdk_bdev *bdev)
5068 : {
5069 0 : return bdev->blockcnt;
5070 : }
5071 :
5072 : const char *
5073 0 : spdk_bdev_get_qos_rpc_type(enum spdk_bdev_qos_rate_limit_type type)
5074 : {
5075 0 : return qos_rpc_type[type];
5076 : }
5077 :
5078 : void
5079 0 : spdk_bdev_get_qos_rate_limits(struct spdk_bdev *bdev, uint64_t *limits)
5080 : {
5081 : int i;
5082 :
5083 0 : memset(limits, 0, sizeof(*limits) * SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES);
5084 :
5085 0 : spdk_spin_lock(&bdev->internal.spinlock);
5086 0 : if (bdev->internal.qos) {
5087 0 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
5088 0 : if (bdev->internal.qos->rate_limits[i].limit !=
5089 : SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
5090 0 : limits[i] = bdev->internal.qos->rate_limits[i].limit;
5091 0 : if (bdev_qos_is_iops_rate_limit(i) == false) {
5092 : /* Change from Byte to Megabyte which is user visible. */
5093 0 : limits[i] = limits[i] / 1024 / 1024;
5094 0 : }
5095 0 : }
5096 0 : }
5097 0 : }
5098 0 : spdk_spin_unlock(&bdev->internal.spinlock);
5099 0 : }
5100 :
5101 : size_t
5102 276 : spdk_bdev_get_buf_align(const struct spdk_bdev *bdev)
5103 : {
5104 276 : return 1 << bdev->required_alignment;
5105 : }
5106 :
5107 : uint32_t
5108 0 : spdk_bdev_get_optimal_io_boundary(const struct spdk_bdev *bdev)
5109 : {
5110 0 : return bdev->optimal_io_boundary;
5111 : }
5112 :
5113 : bool
5114 0 : spdk_bdev_has_write_cache(const struct spdk_bdev *bdev)
5115 : {
5116 0 : return bdev->write_cache;
5117 : }
5118 :
5119 : const struct spdk_uuid *
5120 0 : spdk_bdev_get_uuid(const struct spdk_bdev *bdev)
5121 : {
5122 0 : return &bdev->uuid;
5123 : }
5124 :
5125 : uint16_t
5126 0 : spdk_bdev_get_acwu(const struct spdk_bdev *bdev)
5127 : {
5128 0 : return bdev->acwu;
5129 : }
5130 :
5131 : uint32_t
5132 29 : spdk_bdev_get_md_size(const struct spdk_bdev *bdev)
5133 : {
5134 29 : return bdev->md_len;
5135 : }
5136 :
5137 : bool
5138 131 : spdk_bdev_is_md_interleaved(const struct spdk_bdev *bdev)
5139 : {
5140 131 : return (bdev->md_len != 0) && bdev->md_interleave;
5141 : }
5142 :
5143 : bool
5144 117 : spdk_bdev_is_md_separate(const struct spdk_bdev *bdev)
5145 : {
5146 117 : return (bdev->md_len != 0) && !bdev->md_interleave;
5147 : }
5148 :
5149 : bool
5150 0 : spdk_bdev_is_zoned(const struct spdk_bdev *bdev)
5151 : {
5152 0 : return bdev->zoned;
5153 : }
5154 :
5155 : uint32_t
5156 122 : spdk_bdev_get_data_block_size(const struct spdk_bdev *bdev)
5157 : {
5158 122 : if (spdk_bdev_is_md_interleaved(bdev)) {
5159 0 : return bdev->blocklen - bdev->md_len;
5160 : } else {
5161 122 : return bdev->blocklen;
5162 : }
5163 122 : }
5164 :
5165 : uint32_t
5166 0 : spdk_bdev_get_physical_block_size(const struct spdk_bdev *bdev)
5167 : {
5168 0 : return bdev->phys_blocklen;
5169 : }
5170 :
5171 : static uint32_t
5172 9 : _bdev_get_block_size_with_md(const struct spdk_bdev *bdev)
5173 : {
5174 9 : if (!spdk_bdev_is_md_interleaved(bdev)) {
5175 6 : return bdev->blocklen + bdev->md_len;
5176 : } else {
5177 3 : return bdev->blocklen;
5178 : }
5179 9 : }
5180 :
5181 : /* We have to use the typedef in the function declaration to appease astyle. */
5182 : typedef enum spdk_dif_type spdk_dif_type_t;
5183 : typedef enum spdk_dif_pi_format spdk_dif_pi_format_t;
5184 :
5185 : spdk_dif_type_t
5186 0 : spdk_bdev_get_dif_type(const struct spdk_bdev *bdev)
5187 : {
5188 0 : if (bdev->md_len != 0) {
5189 0 : return bdev->dif_type;
5190 : } else {
5191 0 : return SPDK_DIF_DISABLE;
5192 : }
5193 0 : }
5194 :
5195 : spdk_dif_pi_format_t
5196 0 : spdk_bdev_get_dif_pi_format(const struct spdk_bdev *bdev)
5197 : {
5198 0 : return bdev->dif_pi_format;
5199 : }
5200 :
5201 : bool
5202 0 : spdk_bdev_is_dif_head_of_md(const struct spdk_bdev *bdev)
5203 : {
5204 0 : if (spdk_bdev_get_dif_type(bdev) != SPDK_DIF_DISABLE) {
5205 0 : return bdev->dif_is_head_of_md;
5206 : } else {
5207 0 : return false;
5208 : }
5209 0 : }
5210 :
5211 : bool
5212 0 : spdk_bdev_is_dif_check_enabled(const struct spdk_bdev *bdev,
5213 : enum spdk_dif_check_type check_type)
5214 : {
5215 0 : if (spdk_bdev_get_dif_type(bdev) == SPDK_DIF_DISABLE) {
5216 0 : return false;
5217 : }
5218 :
5219 0 : switch (check_type) {
5220 : case SPDK_DIF_CHECK_TYPE_REFTAG:
5221 0 : return (bdev->dif_check_flags & SPDK_DIF_FLAGS_REFTAG_CHECK) != 0;
5222 : case SPDK_DIF_CHECK_TYPE_APPTAG:
5223 0 : return (bdev->dif_check_flags & SPDK_DIF_FLAGS_APPTAG_CHECK) != 0;
5224 : case SPDK_DIF_CHECK_TYPE_GUARD:
5225 0 : return (bdev->dif_check_flags & SPDK_DIF_FLAGS_GUARD_CHECK) != 0;
5226 : default:
5227 0 : return false;
5228 : }
5229 0 : }
5230 :
5231 : static uint32_t
5232 3 : bdev_get_max_write(const struct spdk_bdev *bdev, uint64_t num_bytes)
5233 : {
5234 : uint64_t aligned_length, max_write_blocks;
5235 :
5236 3 : aligned_length = num_bytes - (spdk_bdev_get_buf_align(bdev) - 1);
5237 3 : max_write_blocks = aligned_length / _bdev_get_block_size_with_md(bdev);
5238 3 : max_write_blocks -= max_write_blocks % bdev->write_unit_size;
5239 :
5240 3 : return max_write_blocks;
5241 : }
5242 :
5243 : uint32_t
5244 1 : spdk_bdev_get_max_copy(const struct spdk_bdev *bdev)
5245 : {
5246 1 : return bdev->max_copy;
5247 : }
5248 :
5249 : uint64_t
5250 0 : spdk_bdev_get_qd(const struct spdk_bdev *bdev)
5251 : {
5252 0 : return bdev->internal.measured_queue_depth;
5253 : }
5254 :
5255 : uint64_t
5256 0 : spdk_bdev_get_qd_sampling_period(const struct spdk_bdev *bdev)
5257 : {
5258 0 : return bdev->internal.period;
5259 : }
5260 :
5261 : uint64_t
5262 0 : spdk_bdev_get_weighted_io_time(const struct spdk_bdev *bdev)
5263 : {
5264 0 : return bdev->internal.weighted_io_time;
5265 : }
5266 :
5267 : uint64_t
5268 0 : spdk_bdev_get_io_time(const struct spdk_bdev *bdev)
5269 : {
5270 0 : return bdev->internal.io_time;
5271 : }
5272 :
5273 0 : union spdk_bdev_nvme_ctratt spdk_bdev_get_nvme_ctratt(struct spdk_bdev *bdev)
5274 : {
5275 0 : return bdev->ctratt;
5276 : }
5277 :
5278 : uint32_t
5279 0 : spdk_bdev_get_nvme_nsid(struct spdk_bdev *bdev)
5280 : {
5281 0 : return bdev->nsid;
5282 : }
5283 :
5284 : uint32_t
5285 0 : spdk_bdev_desc_get_block_size(struct spdk_bdev_desc *desc)
5286 : {
5287 0 : struct spdk_bdev *bdev = desc->bdev;
5288 :
5289 0 : return desc->opts.no_metadata ? bdev->blocklen - bdev->md_len : bdev->blocklen;
5290 : }
5291 :
5292 : uint32_t
5293 0 : spdk_bdev_desc_get_md_size(struct spdk_bdev_desc *desc)
5294 : {
5295 0 : struct spdk_bdev *bdev = desc->bdev;
5296 :
5297 0 : return desc->opts.no_metadata ? 0 : bdev->md_len;
5298 : }
5299 :
5300 : bool
5301 0 : spdk_bdev_desc_is_md_interleaved(struct spdk_bdev_desc *desc)
5302 : {
5303 0 : struct spdk_bdev *bdev = desc->bdev;
5304 :
5305 0 : return desc->opts.no_metadata ? false : spdk_bdev_is_md_interleaved(bdev);
5306 : }
5307 :
5308 : bool
5309 0 : spdk_bdev_desc_is_md_separate(struct spdk_bdev_desc *desc)
5310 : {
5311 0 : struct spdk_bdev *bdev = desc->bdev;
5312 :
5313 0 : return desc->opts.no_metadata ? false : spdk_bdev_is_md_separate(bdev);
5314 : }
5315 :
5316 : spdk_dif_type_t
5317 0 : spdk_bdev_desc_get_dif_type(struct spdk_bdev_desc *desc)
5318 : {
5319 0 : struct spdk_bdev *bdev = desc->bdev;
5320 :
5321 0 : return desc->opts.no_metadata ? SPDK_DIF_DISABLE : spdk_bdev_get_dif_type(bdev);
5322 : }
5323 :
5324 : spdk_dif_pi_format_t
5325 0 : spdk_bdev_desc_get_dif_pi_format(struct spdk_bdev_desc *desc)
5326 : {
5327 0 : struct spdk_bdev *bdev = desc->bdev;
5328 :
5329 0 : return desc->opts.no_metadata ? SPDK_DIF_PI_FORMAT_16 : spdk_bdev_get_dif_pi_format(bdev);
5330 : }
5331 :
5332 : bool
5333 0 : spdk_bdev_desc_is_dif_head_of_md(struct spdk_bdev_desc *desc)
5334 : {
5335 0 : struct spdk_bdev *bdev = desc->bdev;
5336 :
5337 0 : return desc->opts.no_metadata ? false : spdk_bdev_is_dif_head_of_md(bdev);
5338 : }
5339 :
5340 : bool
5341 0 : spdk_bdev_desc_is_dif_check_enabled(struct spdk_bdev_desc *desc,
5342 : enum spdk_dif_check_type check_type)
5343 : {
5344 0 : struct spdk_bdev *bdev = desc->bdev;
5345 :
5346 0 : return desc->opts.no_metadata ? false : spdk_bdev_is_dif_check_enabled(bdev, check_type);
5347 : }
5348 :
5349 : static void bdev_update_qd_sampling_period(void *ctx);
5350 :
5351 : static void
5352 1 : _calculate_measured_qd_cpl(struct spdk_bdev *bdev, void *_ctx, int status)
5353 : {
5354 1 : bdev->internal.measured_queue_depth = bdev->internal.temporary_queue_depth;
5355 :
5356 1 : if (bdev->internal.measured_queue_depth) {
5357 0 : bdev->internal.io_time += bdev->internal.period;
5358 0 : bdev->internal.weighted_io_time += bdev->internal.period * bdev->internal.measured_queue_depth;
5359 0 : }
5360 :
5361 1 : bdev->internal.qd_poll_in_progress = false;
5362 :
5363 1 : bdev_update_qd_sampling_period(bdev);
5364 1 : }
5365 :
5366 : static void
5367 1 : _calculate_measured_qd(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
5368 : struct spdk_io_channel *io_ch, void *_ctx)
5369 : {
5370 1 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(io_ch);
5371 :
5372 1 : bdev->internal.temporary_queue_depth += ch->io_outstanding;
5373 1 : spdk_bdev_for_each_channel_continue(i, 0);
5374 1 : }
5375 :
5376 : static int
5377 1 : bdev_calculate_measured_queue_depth(void *ctx)
5378 : {
5379 1 : struct spdk_bdev *bdev = ctx;
5380 :
5381 1 : bdev->internal.qd_poll_in_progress = true;
5382 1 : bdev->internal.temporary_queue_depth = 0;
5383 1 : spdk_bdev_for_each_channel(bdev, _calculate_measured_qd, bdev, _calculate_measured_qd_cpl);
5384 1 : return SPDK_POLLER_BUSY;
5385 : }
5386 :
5387 : static void
5388 5 : bdev_update_qd_sampling_period(void *ctx)
5389 : {
5390 5 : struct spdk_bdev *bdev = ctx;
5391 :
5392 5 : if (bdev->internal.period == bdev->internal.new_period) {
5393 0 : return;
5394 : }
5395 :
5396 5 : if (bdev->internal.qd_poll_in_progress) {
5397 1 : return;
5398 : }
5399 :
5400 4 : bdev->internal.period = bdev->internal.new_period;
5401 :
5402 4 : spdk_poller_unregister(&bdev->internal.qd_poller);
5403 4 : if (bdev->internal.period != 0) {
5404 2 : bdev->internal.qd_poller = SPDK_POLLER_REGISTER(bdev_calculate_measured_queue_depth,
5405 : bdev, bdev->internal.period);
5406 2 : } else {
5407 2 : spdk_bdev_close(bdev->internal.qd_desc);
5408 2 : bdev->internal.qd_desc = NULL;
5409 : }
5410 5 : }
5411 :
5412 : static void
5413 0 : _tmp_bdev_event_cb(enum spdk_bdev_event_type type, struct spdk_bdev *bdev, void *ctx)
5414 : {
5415 0 : SPDK_NOTICELOG("Unexpected event type: %d\n", type);
5416 0 : }
5417 :
5418 : void
5419 131 : spdk_bdev_set_qd_sampling_period(struct spdk_bdev *bdev, uint64_t period)
5420 : {
5421 : int rc;
5422 :
5423 131 : if (bdev->internal.new_period == period) {
5424 125 : return;
5425 : }
5426 :
5427 6 : bdev->internal.new_period = period;
5428 :
5429 6 : if (bdev->internal.qd_desc != NULL) {
5430 4 : assert(bdev->internal.period != 0);
5431 :
5432 8 : spdk_thread_send_msg(bdev->internal.qd_desc->thread,
5433 4 : bdev_update_qd_sampling_period, bdev);
5434 4 : return;
5435 : }
5436 :
5437 2 : assert(bdev->internal.period == 0);
5438 :
5439 4 : rc = spdk_bdev_open_ext(spdk_bdev_get_name(bdev), false, _tmp_bdev_event_cb,
5440 2 : NULL, &bdev->internal.qd_desc);
5441 2 : if (rc != 0) {
5442 0 : return;
5443 : }
5444 :
5445 2 : bdev->internal.period = period;
5446 2 : bdev->internal.qd_poller = SPDK_POLLER_REGISTER(bdev_calculate_measured_queue_depth,
5447 : bdev, period);
5448 131 : }
5449 :
5450 : struct bdev_get_current_qd_ctx {
5451 : uint64_t current_qd;
5452 : spdk_bdev_get_current_qd_cb cb_fn;
5453 : void *cb_arg;
5454 : };
5455 :
5456 : static void
5457 0 : bdev_get_current_qd_done(struct spdk_bdev *bdev, void *_ctx, int status)
5458 : {
5459 0 : struct bdev_get_current_qd_ctx *ctx = _ctx;
5460 :
5461 0 : ctx->cb_fn(bdev, ctx->current_qd, ctx->cb_arg, 0);
5462 :
5463 0 : free(ctx);
5464 0 : }
5465 :
5466 : static void
5467 0 : bdev_get_current_qd(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
5468 : struct spdk_io_channel *io_ch, void *_ctx)
5469 : {
5470 0 : struct bdev_get_current_qd_ctx *ctx = _ctx;
5471 0 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
5472 :
5473 0 : ctx->current_qd += bdev_ch->io_outstanding;
5474 :
5475 0 : spdk_bdev_for_each_channel_continue(i, 0);
5476 0 : }
5477 :
5478 : void
5479 0 : spdk_bdev_get_current_qd(struct spdk_bdev *bdev, spdk_bdev_get_current_qd_cb cb_fn,
5480 : void *cb_arg)
5481 : {
5482 : struct bdev_get_current_qd_ctx *ctx;
5483 :
5484 0 : assert(cb_fn != NULL);
5485 :
5486 0 : ctx = calloc(1, sizeof(*ctx));
5487 0 : if (ctx == NULL) {
5488 0 : cb_fn(bdev, 0, cb_arg, -ENOMEM);
5489 0 : return;
5490 : }
5491 :
5492 0 : ctx->cb_fn = cb_fn;
5493 0 : ctx->cb_arg = cb_arg;
5494 :
5495 0 : spdk_bdev_for_each_channel(bdev, bdev_get_current_qd, ctx, bdev_get_current_qd_done);
5496 0 : }
5497 :
5498 : static void
5499 25 : _event_notify(struct spdk_bdev_desc *desc, enum spdk_bdev_event_type type)
5500 : {
5501 25 : assert(desc->thread == spdk_get_thread());
5502 :
5503 25 : spdk_spin_lock(&desc->spinlock);
5504 25 : desc->refs--;
5505 25 : if (!desc->closed) {
5506 14 : spdk_spin_unlock(&desc->spinlock);
5507 28 : desc->callback.event_fn(type,
5508 14 : desc->bdev,
5509 14 : desc->callback.ctx);
5510 14 : return;
5511 11 : } else if (desc->refs == 0) {
5512 : /* This descriptor was closed after this event_notify message was sent.
5513 : * spdk_bdev_close() could not free the descriptor since this message was
5514 : * in flight, so we free it now using bdev_desc_free().
5515 : */
5516 10 : spdk_spin_unlock(&desc->spinlock);
5517 10 : bdev_desc_free(desc);
5518 10 : return;
5519 : }
5520 1 : spdk_spin_unlock(&desc->spinlock);
5521 25 : }
5522 :
5523 : static void
5524 25 : event_notify(struct spdk_bdev_desc *desc, spdk_msg_fn event_notify_fn)
5525 : {
5526 25 : spdk_spin_lock(&desc->spinlock);
5527 25 : desc->refs++;
5528 25 : spdk_thread_send_msg(desc->thread, event_notify_fn, desc);
5529 25 : spdk_spin_unlock(&desc->spinlock);
5530 25 : }
5531 :
5532 : static void
5533 6 : _resize_notify(void *ctx)
5534 : {
5535 6 : struct spdk_bdev_desc *desc = ctx;
5536 :
5537 6 : _event_notify(desc, SPDK_BDEV_EVENT_RESIZE);
5538 6 : }
5539 :
5540 : int
5541 11 : spdk_bdev_notify_blockcnt_change(struct spdk_bdev *bdev, uint64_t size)
5542 : {
5543 : struct spdk_bdev_desc *desc;
5544 : int ret;
5545 :
5546 11 : if (size == bdev->blockcnt) {
5547 0 : return 0;
5548 : }
5549 :
5550 11 : spdk_spin_lock(&bdev->internal.spinlock);
5551 :
5552 : /* bdev has open descriptors */
5553 11 : if (!TAILQ_EMPTY(&bdev->internal.open_descs) &&
5554 7 : bdev->blockcnt > size) {
5555 1 : ret = -EBUSY;
5556 1 : } else {
5557 10 : bdev->blockcnt = size;
5558 16 : TAILQ_FOREACH(desc, &bdev->internal.open_descs, link) {
5559 6 : event_notify(desc, _resize_notify);
5560 6 : }
5561 10 : ret = 0;
5562 : }
5563 :
5564 11 : spdk_spin_unlock(&bdev->internal.spinlock);
5565 :
5566 11 : return ret;
5567 11 : }
5568 :
5569 : /*
5570 : * Convert I/O offset and length from bytes to blocks.
5571 : *
5572 : * Returns zero on success or non-zero if the byte parameters aren't divisible by the block size.
5573 : */
5574 : static uint64_t
5575 20 : bdev_bytes_to_blocks(struct spdk_bdev_desc *desc, uint64_t offset_bytes,
5576 : uint64_t *offset_blocks, uint64_t num_bytes, uint64_t *num_blocks)
5577 : {
5578 20 : uint32_t block_size = bdev_desc_get_block_size(desc);
5579 : uint8_t shift_cnt;
5580 :
5581 : /* Avoid expensive div operations if possible. These spdk_u32 functions are very cheap. */
5582 20 : if (spdk_likely(spdk_u32_is_pow2(block_size))) {
5583 17 : shift_cnt = spdk_u32log2(block_size);
5584 17 : *offset_blocks = offset_bytes >> shift_cnt;
5585 17 : *num_blocks = num_bytes >> shift_cnt;
5586 34 : return (offset_bytes - (*offset_blocks << shift_cnt)) |
5587 17 : (num_bytes - (*num_blocks << shift_cnt));
5588 : } else {
5589 3 : *offset_blocks = offset_bytes / block_size;
5590 3 : *num_blocks = num_bytes / block_size;
5591 3 : return (offset_bytes % block_size) | (num_bytes % block_size);
5592 : }
5593 20 : }
5594 :
5595 : static bool
5596 689 : bdev_io_valid_blocks(struct spdk_bdev *bdev, uint64_t offset_blocks, uint64_t num_blocks)
5597 : {
5598 : /* Return failure if offset_blocks + num_blocks is less than offset_blocks; indicates there
5599 : * has been an overflow and hence the offset has been wrapped around */
5600 689 : if (offset_blocks + num_blocks < offset_blocks) {
5601 1 : return false;
5602 : }
5603 :
5604 : /* Return failure if offset_blocks + num_blocks exceeds the size of the bdev */
5605 688 : if (offset_blocks + num_blocks > bdev->blockcnt) {
5606 2 : return false;
5607 : }
5608 :
5609 686 : return true;
5610 689 : }
5611 :
5612 : static void
5613 2 : bdev_seek_complete_cb(void *ctx)
5614 : {
5615 2 : struct spdk_bdev_io *bdev_io = ctx;
5616 :
5617 2 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
5618 2 : bdev_io->internal.cb(bdev_io, true, bdev_io->internal.caller_ctx);
5619 2 : }
5620 :
5621 : static int
5622 4 : bdev_seek(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5623 : uint64_t offset_blocks, enum spdk_bdev_io_type io_type,
5624 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5625 : {
5626 4 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5627 : struct spdk_bdev_io *bdev_io;
5628 4 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5629 :
5630 4 : assert(io_type == SPDK_BDEV_IO_TYPE_SEEK_DATA || io_type == SPDK_BDEV_IO_TYPE_SEEK_HOLE);
5631 :
5632 : /* Check if offset_blocks is valid looking at the validity of one block */
5633 4 : if (!bdev_io_valid_blocks(bdev, offset_blocks, 1)) {
5634 0 : return -EINVAL;
5635 : }
5636 :
5637 4 : bdev_io = bdev_channel_get_io(channel);
5638 4 : if (!bdev_io) {
5639 0 : return -ENOMEM;
5640 : }
5641 :
5642 4 : bdev_io->internal.ch = channel;
5643 4 : bdev_io->internal.desc = desc;
5644 4 : bdev_io->type = io_type;
5645 4 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5646 4 : bdev_io->u.bdev.memory_domain = NULL;
5647 4 : bdev_io->u.bdev.memory_domain_ctx = NULL;
5648 4 : bdev_io->u.bdev.accel_sequence = NULL;
5649 4 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5650 :
5651 4 : if (!spdk_bdev_io_type_supported(bdev, io_type)) {
5652 : /* In case bdev doesn't support seek to next data/hole offset,
5653 : * it is assumed that only data and no holes are present */
5654 2 : if (io_type == SPDK_BDEV_IO_TYPE_SEEK_DATA) {
5655 1 : bdev_io->u.bdev.seek.offset = offset_blocks;
5656 1 : } else {
5657 1 : bdev_io->u.bdev.seek.offset = UINT64_MAX;
5658 : }
5659 :
5660 2 : spdk_thread_send_msg(spdk_get_thread(), bdev_seek_complete_cb, bdev_io);
5661 2 : return 0;
5662 : }
5663 :
5664 2 : bdev_io_submit(bdev_io);
5665 2 : return 0;
5666 4 : }
5667 :
5668 : int
5669 2 : spdk_bdev_seek_data(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5670 : uint64_t offset_blocks,
5671 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5672 : {
5673 2 : return bdev_seek(desc, ch, offset_blocks, SPDK_BDEV_IO_TYPE_SEEK_DATA, cb, cb_arg);
5674 : }
5675 :
5676 : int
5677 2 : spdk_bdev_seek_hole(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5678 : uint64_t offset_blocks,
5679 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5680 : {
5681 2 : return bdev_seek(desc, ch, offset_blocks, SPDK_BDEV_IO_TYPE_SEEK_HOLE, cb, cb_arg);
5682 : }
5683 :
5684 : uint64_t
5685 4 : spdk_bdev_io_get_seek_offset(const struct spdk_bdev_io *bdev_io)
5686 : {
5687 4 : return bdev_io->u.bdev.seek.offset;
5688 : }
5689 :
5690 : static int
5691 204 : bdev_read_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch, void *buf,
5692 : void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
5693 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5694 : {
5695 204 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5696 : struct spdk_bdev_io *bdev_io;
5697 204 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5698 :
5699 204 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
5700 0 : return -EINVAL;
5701 : }
5702 :
5703 204 : bdev_io = bdev_channel_get_io(channel);
5704 204 : if (!bdev_io) {
5705 1 : return -ENOMEM;
5706 : }
5707 :
5708 203 : bdev_io->internal.ch = channel;
5709 203 : bdev_io->internal.desc = desc;
5710 203 : bdev_io->type = SPDK_BDEV_IO_TYPE_READ;
5711 203 : bdev_io->u.bdev.iovs = &bdev_io->iov;
5712 203 : bdev_io->u.bdev.iovs[0].iov_base = buf;
5713 203 : bdev_io->u.bdev.iovs[0].iov_len = num_blocks * bdev_desc_get_block_size(desc);
5714 203 : bdev_io->u.bdev.iovcnt = 1;
5715 203 : bdev_io->u.bdev.md_buf = md_buf;
5716 203 : bdev_io->u.bdev.num_blocks = num_blocks;
5717 203 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5718 203 : bdev_io->u.bdev.memory_domain = NULL;
5719 203 : bdev_io->u.bdev.memory_domain_ctx = NULL;
5720 203 : bdev_io->u.bdev.accel_sequence = NULL;
5721 203 : bdev_io->u.bdev.dif_check_flags = bdev->dif_check_flags;
5722 203 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5723 :
5724 203 : bdev_io_submit(bdev_io);
5725 203 : return 0;
5726 204 : }
5727 :
5728 : int
5729 3 : spdk_bdev_read(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5730 : void *buf, uint64_t offset, uint64_t nbytes,
5731 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5732 : {
5733 : uint64_t offset_blocks, num_blocks;
5734 :
5735 3 : if (bdev_bytes_to_blocks(desc, offset, &offset_blocks, nbytes, &num_blocks) != 0) {
5736 0 : return -EINVAL;
5737 : }
5738 :
5739 3 : return spdk_bdev_read_blocks(desc, ch, buf, offset_blocks, num_blocks, cb, cb_arg);
5740 3 : }
5741 :
5742 : int
5743 200 : spdk_bdev_read_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5744 : void *buf, uint64_t offset_blocks, uint64_t num_blocks,
5745 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5746 : {
5747 200 : return bdev_read_blocks_with_md(desc, ch, buf, NULL, offset_blocks, num_blocks, cb, cb_arg);
5748 : }
5749 :
5750 : int
5751 4 : spdk_bdev_read_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5752 : void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
5753 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5754 : {
5755 8 : struct iovec iov = {
5756 4 : .iov_base = buf,
5757 : };
5758 :
5759 4 : if (md_buf && !spdk_bdev_is_md_separate(spdk_bdev_desc_get_bdev(desc))) {
5760 0 : return -EINVAL;
5761 : }
5762 :
5763 4 : if ((md_buf || desc->opts.no_metadata) && !_is_buf_allocated(&iov)) {
5764 0 : return -EINVAL;
5765 : }
5766 :
5767 8 : return bdev_read_blocks_with_md(desc, ch, buf, md_buf, offset_blocks, num_blocks,
5768 4 : cb, cb_arg);
5769 4 : }
5770 :
5771 : int
5772 5 : spdk_bdev_readv(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5773 : struct iovec *iov, int iovcnt,
5774 : uint64_t offset, uint64_t nbytes,
5775 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5776 : {
5777 : uint64_t offset_blocks, num_blocks;
5778 :
5779 5 : if (bdev_bytes_to_blocks(desc, offset, &offset_blocks, nbytes, &num_blocks) != 0) {
5780 0 : return -EINVAL;
5781 : }
5782 :
5783 5 : return spdk_bdev_readv_blocks(desc, ch, iov, iovcnt, offset_blocks, num_blocks, cb, cb_arg);
5784 5 : }
5785 :
5786 : static int
5787 226 : bdev_readv_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5788 : struct iovec *iov, int iovcnt, void *md_buf, uint64_t offset_blocks,
5789 : uint64_t num_blocks, struct spdk_memory_domain *domain, void *domain_ctx,
5790 : struct spdk_accel_sequence *seq, uint32_t dif_check_flags,
5791 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5792 : {
5793 226 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5794 : struct spdk_bdev_io *bdev_io;
5795 226 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5796 :
5797 226 : if (spdk_unlikely(!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks))) {
5798 0 : return -EINVAL;
5799 : }
5800 :
5801 226 : bdev_io = bdev_channel_get_io(channel);
5802 226 : if (spdk_unlikely(!bdev_io)) {
5803 2 : return -ENOMEM;
5804 : }
5805 :
5806 224 : bdev_io->internal.ch = channel;
5807 224 : bdev_io->internal.desc = desc;
5808 224 : bdev_io->type = SPDK_BDEV_IO_TYPE_READ;
5809 224 : bdev_io->u.bdev.iovs = iov;
5810 224 : bdev_io->u.bdev.iovcnt = iovcnt;
5811 224 : bdev_io->u.bdev.md_buf = md_buf;
5812 224 : bdev_io->u.bdev.num_blocks = num_blocks;
5813 224 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5814 224 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5815 :
5816 224 : if (seq != NULL) {
5817 0 : bdev_io->internal.f.has_accel_sequence = true;
5818 0 : bdev_io->internal.accel_sequence = seq;
5819 0 : }
5820 :
5821 224 : if (domain != NULL) {
5822 2 : bdev_io->internal.f.has_memory_domain = true;
5823 2 : bdev_io->internal.memory_domain = domain;
5824 2 : bdev_io->internal.memory_domain_ctx = domain_ctx;
5825 2 : }
5826 :
5827 224 : bdev_io->u.bdev.memory_domain = domain;
5828 224 : bdev_io->u.bdev.memory_domain_ctx = domain_ctx;
5829 224 : bdev_io->u.bdev.accel_sequence = seq;
5830 224 : bdev_io->u.bdev.dif_check_flags = dif_check_flags;
5831 :
5832 224 : _bdev_io_submit_ext(desc, bdev_io);
5833 :
5834 224 : return 0;
5835 226 : }
5836 :
5837 : int
5838 21 : spdk_bdev_readv_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5839 : struct iovec *iov, int iovcnt,
5840 : uint64_t offset_blocks, uint64_t num_blocks,
5841 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5842 : {
5843 21 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5844 :
5845 42 : return bdev_readv_blocks_with_md(desc, ch, iov, iovcnt, NULL, offset_blocks,
5846 21 : num_blocks, NULL, NULL, NULL, bdev->dif_check_flags, cb, cb_arg);
5847 : }
5848 :
5849 : int
5850 4 : spdk_bdev_readv_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5851 : struct iovec *iov, int iovcnt, void *md_buf,
5852 : uint64_t offset_blocks, uint64_t num_blocks,
5853 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5854 : {
5855 4 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5856 :
5857 4 : if (md_buf && !spdk_bdev_is_md_separate(bdev)) {
5858 0 : return -EINVAL;
5859 : }
5860 :
5861 4 : if (md_buf && !_is_buf_allocated(iov)) {
5862 0 : return -EINVAL;
5863 : }
5864 :
5865 8 : return bdev_readv_blocks_with_md(desc, ch, iov, iovcnt, md_buf, offset_blocks,
5866 4 : num_blocks, NULL, NULL, NULL, bdev->dif_check_flags, cb, cb_arg);
5867 4 : }
5868 :
5869 : static inline bool
5870 14 : _bdev_io_check_opts(struct spdk_bdev_ext_io_opts *opts, struct iovec *iov)
5871 : {
5872 : /*
5873 : * We check if opts size is at least of size when we first introduced
5874 : * spdk_bdev_ext_io_opts (ac6f2bdd8d) since access to those members
5875 : * are not checked internal.
5876 : */
5877 24 : return opts->size >= offsetof(struct spdk_bdev_ext_io_opts, metadata) +
5878 14 : sizeof(opts->metadata) &&
5879 10 : opts->size <= sizeof(*opts) &&
5880 : /* When memory domain is used, the user must provide data buffers */
5881 8 : (!opts->memory_domain || (iov && iov[0].iov_base));
5882 : }
5883 :
5884 : int
5885 8 : spdk_bdev_readv_blocks_ext(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5886 : struct iovec *iov, int iovcnt,
5887 : uint64_t offset_blocks, uint64_t num_blocks,
5888 : spdk_bdev_io_completion_cb cb, void *cb_arg,
5889 : struct spdk_bdev_ext_io_opts *opts)
5890 : {
5891 8 : struct spdk_memory_domain *domain = NULL;
5892 8 : struct spdk_accel_sequence *seq = NULL;
5893 8 : void *domain_ctx = NULL, *md = NULL;
5894 8 : uint32_t dif_check_flags = 0;
5895 : uint32_t nvme_cdw12_raw;
5896 8 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5897 :
5898 8 : if (opts) {
5899 7 : if (spdk_unlikely(!_bdev_io_check_opts(opts, iov))) {
5900 3 : return -EINVAL;
5901 : }
5902 :
5903 4 : md = opts->metadata;
5904 4 : domain = bdev_get_ext_io_opt(opts, memory_domain, NULL);
5905 4 : domain_ctx = bdev_get_ext_io_opt(opts, memory_domain_ctx, NULL);
5906 4 : seq = bdev_get_ext_io_opt(opts, accel_sequence, NULL);
5907 4 : nvme_cdw12_raw = bdev_get_ext_io_opt(opts, nvme_cdw12.raw, 0);
5908 4 : if (md) {
5909 4 : if (spdk_unlikely(!spdk_bdev_is_md_separate(bdev))) {
5910 0 : return -EINVAL;
5911 : }
5912 :
5913 4 : if (spdk_unlikely(!_is_buf_allocated(iov))) {
5914 0 : return -EINVAL;
5915 : }
5916 :
5917 4 : if (spdk_unlikely(seq != NULL)) {
5918 0 : return -EINVAL;
5919 : }
5920 :
5921 4 : if (nvme_cdw12_raw & SPDK_DIF_FLAGS_NVME_PRACT) {
5922 0 : SPDK_ERRLOG("Separate metadata with NVMe PRACT is not supported.\n");
5923 0 : return -ENOTSUP;
5924 : }
5925 4 : }
5926 :
5927 4 : if (nvme_cdw12_raw & SPDK_DIF_FLAGS_NVME_PRACT) {
5928 0 : dif_check_flags |= SPDK_DIF_FLAGS_NVME_PRACT;
5929 0 : }
5930 4 : }
5931 :
5932 10 : dif_check_flags |= bdev->dif_check_flags &
5933 5 : ~(bdev_get_ext_io_opt(opts, dif_check_flags_exclude_mask, 0));
5934 :
5935 10 : return bdev_readv_blocks_with_md(desc, ch, iov, iovcnt, md, offset_blocks,
5936 5 : num_blocks, domain, domain_ctx, seq, dif_check_flags, cb, cb_arg);
5937 8 : }
5938 :
5939 : static int
5940 36 : bdev_write_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5941 : void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
5942 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5943 : {
5944 36 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5945 : struct spdk_bdev_io *bdev_io;
5946 36 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5947 :
5948 36 : if (!desc->write) {
5949 0 : return -EBADF;
5950 : }
5951 :
5952 36 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
5953 0 : return -EINVAL;
5954 : }
5955 :
5956 36 : bdev_io = bdev_channel_get_io(channel);
5957 36 : if (!bdev_io) {
5958 0 : return -ENOMEM;
5959 : }
5960 :
5961 36 : bdev_io->internal.ch = channel;
5962 36 : bdev_io->internal.desc = desc;
5963 36 : bdev_io->type = SPDK_BDEV_IO_TYPE_WRITE;
5964 36 : bdev_io->u.bdev.iovs = &bdev_io->iov;
5965 36 : bdev_io->u.bdev.iovs[0].iov_base = buf;
5966 36 : bdev_io->u.bdev.iovs[0].iov_len = num_blocks * bdev_desc_get_block_size(desc);
5967 36 : bdev_io->u.bdev.iovcnt = 1;
5968 36 : bdev_io->u.bdev.md_buf = md_buf;
5969 36 : bdev_io->u.bdev.num_blocks = num_blocks;
5970 36 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5971 36 : bdev_io->u.bdev.memory_domain = NULL;
5972 36 : bdev_io->u.bdev.memory_domain_ctx = NULL;
5973 36 : bdev_io->u.bdev.accel_sequence = NULL;
5974 36 : bdev_io->u.bdev.dif_check_flags = bdev->dif_check_flags;
5975 36 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5976 :
5977 36 : bdev_io_submit(bdev_io);
5978 36 : return 0;
5979 36 : }
5980 :
5981 : int
5982 3 : spdk_bdev_write(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5983 : void *buf, uint64_t offset, uint64_t nbytes,
5984 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5985 : {
5986 : uint64_t offset_blocks, num_blocks;
5987 :
5988 3 : if (bdev_bytes_to_blocks(desc, offset, &offset_blocks, nbytes, &num_blocks) != 0) {
5989 0 : return -EINVAL;
5990 : }
5991 :
5992 3 : return spdk_bdev_write_blocks(desc, ch, buf, offset_blocks, num_blocks, cb, cb_arg);
5993 3 : }
5994 :
5995 : int
5996 27 : spdk_bdev_write_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5997 : void *buf, uint64_t offset_blocks, uint64_t num_blocks,
5998 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5999 : {
6000 54 : return bdev_write_blocks_with_md(desc, ch, buf, NULL, offset_blocks, num_blocks,
6001 27 : cb, cb_arg);
6002 : }
6003 :
6004 : int
6005 3 : spdk_bdev_write_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6006 : void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
6007 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6008 : {
6009 6 : struct iovec iov = {
6010 3 : .iov_base = buf,
6011 : };
6012 :
6013 3 : if (md_buf && !spdk_bdev_is_md_separate(spdk_bdev_desc_get_bdev(desc))) {
6014 0 : return -EINVAL;
6015 : }
6016 :
6017 3 : if (md_buf && !_is_buf_allocated(&iov)) {
6018 0 : return -EINVAL;
6019 : }
6020 :
6021 6 : return bdev_write_blocks_with_md(desc, ch, buf, md_buf, offset_blocks, num_blocks,
6022 3 : cb, cb_arg);
6023 3 : }
6024 :
6025 : static int
6026 70 : bdev_writev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6027 : struct iovec *iov, int iovcnt, void *md_buf,
6028 : uint64_t offset_blocks, uint64_t num_blocks,
6029 : struct spdk_memory_domain *domain, void *domain_ctx,
6030 : struct spdk_accel_sequence *seq, uint32_t dif_check_flags,
6031 : uint32_t nvme_cdw12_raw, uint32_t nvme_cdw13_raw,
6032 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6033 : {
6034 70 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6035 : struct spdk_bdev_io *bdev_io;
6036 70 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6037 :
6038 70 : if (spdk_unlikely(!desc->write)) {
6039 0 : return -EBADF;
6040 : }
6041 :
6042 70 : if (spdk_unlikely(!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks))) {
6043 0 : return -EINVAL;
6044 : }
6045 :
6046 70 : bdev_io = bdev_channel_get_io(channel);
6047 70 : if (spdk_unlikely(!bdev_io)) {
6048 2 : return -ENOMEM;
6049 : }
6050 :
6051 68 : bdev_io->internal.ch = channel;
6052 68 : bdev_io->internal.desc = desc;
6053 68 : bdev_io->type = SPDK_BDEV_IO_TYPE_WRITE;
6054 68 : bdev_io->u.bdev.iovs = iov;
6055 68 : bdev_io->u.bdev.iovcnt = iovcnt;
6056 68 : bdev_io->u.bdev.md_buf = md_buf;
6057 68 : bdev_io->u.bdev.num_blocks = num_blocks;
6058 68 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6059 68 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6060 68 : if (seq != NULL) {
6061 0 : bdev_io->internal.f.has_accel_sequence = true;
6062 0 : bdev_io->internal.accel_sequence = seq;
6063 0 : }
6064 :
6065 68 : if (domain != NULL) {
6066 2 : bdev_io->internal.f.has_memory_domain = true;
6067 2 : bdev_io->internal.memory_domain = domain;
6068 2 : bdev_io->internal.memory_domain_ctx = domain_ctx;
6069 2 : }
6070 :
6071 68 : bdev_io->u.bdev.memory_domain = domain;
6072 68 : bdev_io->u.bdev.memory_domain_ctx = domain_ctx;
6073 68 : bdev_io->u.bdev.accel_sequence = seq;
6074 68 : bdev_io->u.bdev.dif_check_flags = dif_check_flags;
6075 68 : bdev_io->u.bdev.nvme_cdw12.raw = nvme_cdw12_raw;
6076 68 : bdev_io->u.bdev.nvme_cdw13.raw = nvme_cdw13_raw;
6077 :
6078 68 : _bdev_io_submit_ext(desc, bdev_io);
6079 :
6080 68 : return 0;
6081 70 : }
6082 :
6083 : int
6084 3 : spdk_bdev_writev(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6085 : struct iovec *iov, int iovcnt,
6086 : uint64_t offset, uint64_t len,
6087 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6088 : {
6089 : uint64_t offset_blocks, num_blocks;
6090 :
6091 3 : if (bdev_bytes_to_blocks(desc, offset, &offset_blocks, len, &num_blocks) != 0) {
6092 0 : return -EINVAL;
6093 : }
6094 :
6095 3 : return spdk_bdev_writev_blocks(desc, ch, iov, iovcnt, offset_blocks, num_blocks, cb, cb_arg);
6096 3 : }
6097 :
6098 : int
6099 14 : spdk_bdev_writev_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6100 : struct iovec *iov, int iovcnt,
6101 : uint64_t offset_blocks, uint64_t num_blocks,
6102 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6103 : {
6104 14 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6105 :
6106 28 : return bdev_writev_blocks_with_md(desc, ch, iov, iovcnt, NULL, offset_blocks,
6107 14 : num_blocks, NULL, NULL, NULL, bdev->dif_check_flags, 0, 0,
6108 14 : cb, cb_arg);
6109 : }
6110 :
6111 : int
6112 1 : spdk_bdev_writev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6113 : struct iovec *iov, int iovcnt, void *md_buf,
6114 : uint64_t offset_blocks, uint64_t num_blocks,
6115 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6116 : {
6117 1 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6118 :
6119 1 : if (md_buf && !spdk_bdev_is_md_separate(bdev)) {
6120 0 : return -EINVAL;
6121 : }
6122 :
6123 1 : if (md_buf && !_is_buf_allocated(iov)) {
6124 0 : return -EINVAL;
6125 : }
6126 :
6127 2 : return bdev_writev_blocks_with_md(desc, ch, iov, iovcnt, md_buf, offset_blocks,
6128 1 : num_blocks, NULL, NULL, NULL, bdev->dif_check_flags, 0, 0,
6129 1 : cb, cb_arg);
6130 1 : }
6131 :
6132 : int
6133 8 : spdk_bdev_writev_blocks_ext(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6134 : struct iovec *iov, int iovcnt,
6135 : uint64_t offset_blocks, uint64_t num_blocks,
6136 : spdk_bdev_io_completion_cb cb, void *cb_arg,
6137 : struct spdk_bdev_ext_io_opts *opts)
6138 : {
6139 8 : struct spdk_memory_domain *domain = NULL;
6140 8 : struct spdk_accel_sequence *seq = NULL;
6141 8 : void *domain_ctx = NULL, *md = NULL;
6142 8 : uint32_t dif_check_flags = 0;
6143 8 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6144 8 : uint32_t nvme_cdw12_raw = 0;
6145 8 : uint32_t nvme_cdw13_raw = 0;
6146 :
6147 8 : if (opts) {
6148 7 : if (spdk_unlikely(!_bdev_io_check_opts(opts, iov))) {
6149 3 : return -EINVAL;
6150 : }
6151 4 : md = opts->metadata;
6152 4 : domain = bdev_get_ext_io_opt(opts, memory_domain, NULL);
6153 4 : domain_ctx = bdev_get_ext_io_opt(opts, memory_domain_ctx, NULL);
6154 4 : seq = bdev_get_ext_io_opt(opts, accel_sequence, NULL);
6155 4 : nvme_cdw12_raw = bdev_get_ext_io_opt(opts, nvme_cdw12.raw, 0);
6156 4 : nvme_cdw13_raw = bdev_get_ext_io_opt(opts, nvme_cdw13.raw, 0);
6157 4 : if (md) {
6158 4 : if (spdk_unlikely(!spdk_bdev_is_md_separate(bdev))) {
6159 0 : return -EINVAL;
6160 : }
6161 :
6162 4 : if (spdk_unlikely(!_is_buf_allocated(iov))) {
6163 0 : return -EINVAL;
6164 : }
6165 :
6166 4 : if (spdk_unlikely(seq != NULL)) {
6167 0 : return -EINVAL;
6168 : }
6169 :
6170 4 : if (nvme_cdw12_raw & SPDK_DIF_FLAGS_NVME_PRACT) {
6171 0 : SPDK_ERRLOG("Separate metadata with NVMe PRACT is not supported.\n");
6172 0 : return -ENOTSUP;
6173 : }
6174 4 : }
6175 :
6176 4 : if (nvme_cdw12_raw & SPDK_DIF_FLAGS_NVME_PRACT) {
6177 0 : dif_check_flags |= SPDK_DIF_FLAGS_NVME_PRACT;
6178 0 : }
6179 4 : }
6180 :
6181 10 : dif_check_flags |= bdev->dif_check_flags &
6182 5 : ~(bdev_get_ext_io_opt(opts, dif_check_flags_exclude_mask, 0));
6183 :
6184 10 : return bdev_writev_blocks_with_md(desc, ch, iov, iovcnt, md, offset_blocks, num_blocks,
6185 5 : domain, domain_ctx, seq, dif_check_flags,
6186 5 : nvme_cdw12_raw, nvme_cdw13_raw, cb, cb_arg);
6187 8 : }
6188 :
6189 : static void
6190 11 : bdev_compare_do_read_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
6191 : {
6192 11 : struct spdk_bdev_io *parent_io = cb_arg;
6193 11 : struct spdk_bdev *bdev = parent_io->bdev;
6194 11 : uint8_t *read_buf = bdev_io->u.bdev.iovs[0].iov_base;
6195 11 : int i, rc = 0;
6196 :
6197 11 : if (!success) {
6198 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
6199 0 : parent_io->internal.cb(parent_io, false, parent_io->internal.caller_ctx);
6200 0 : spdk_bdev_free_io(bdev_io);
6201 0 : return;
6202 : }
6203 :
6204 17 : for (i = 0; i < parent_io->u.bdev.iovcnt; i++) {
6205 22 : rc = memcmp(read_buf,
6206 11 : parent_io->u.bdev.iovs[i].iov_base,
6207 11 : parent_io->u.bdev.iovs[i].iov_len);
6208 11 : if (rc) {
6209 5 : break;
6210 : }
6211 6 : read_buf += parent_io->u.bdev.iovs[i].iov_len;
6212 6 : }
6213 :
6214 11 : if (rc == 0 && parent_io->u.bdev.md_buf && spdk_bdev_is_md_separate(bdev)) {
6215 4 : rc = memcmp(bdev_io->u.bdev.md_buf,
6216 2 : parent_io->u.bdev.md_buf,
6217 2 : spdk_bdev_get_md_size(bdev));
6218 2 : }
6219 :
6220 11 : spdk_bdev_free_io(bdev_io);
6221 :
6222 11 : if (rc == 0) {
6223 5 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
6224 5 : parent_io->internal.cb(parent_io, true, parent_io->internal.caller_ctx);
6225 5 : } else {
6226 6 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_MISCOMPARE;
6227 6 : parent_io->internal.cb(parent_io, false, parent_io->internal.caller_ctx);
6228 : }
6229 11 : }
6230 :
6231 : static void
6232 11 : bdev_compare_do_read(void *_bdev_io)
6233 : {
6234 11 : struct spdk_bdev_io *bdev_io = _bdev_io;
6235 : int rc;
6236 :
6237 22 : rc = spdk_bdev_read_blocks(bdev_io->internal.desc,
6238 11 : spdk_io_channel_from_ctx(bdev_io->internal.ch), NULL,
6239 11 : bdev_io->u.bdev.offset_blocks, bdev_io->u.bdev.num_blocks,
6240 11 : bdev_compare_do_read_done, bdev_io);
6241 :
6242 11 : if (rc == -ENOMEM) {
6243 0 : bdev_queue_io_wait_with_cb(bdev_io, bdev_compare_do_read);
6244 11 : } else if (rc != 0) {
6245 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
6246 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
6247 0 : }
6248 11 : }
6249 :
6250 : static int
6251 16 : bdev_comparev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6252 : struct iovec *iov, int iovcnt, void *md_buf,
6253 : uint64_t offset_blocks, uint64_t num_blocks,
6254 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6255 : {
6256 16 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6257 : struct spdk_bdev_io *bdev_io;
6258 16 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6259 :
6260 16 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6261 0 : return -EINVAL;
6262 : }
6263 :
6264 16 : bdev_io = bdev_channel_get_io(channel);
6265 16 : if (!bdev_io) {
6266 0 : return -ENOMEM;
6267 : }
6268 :
6269 16 : bdev_io->internal.ch = channel;
6270 16 : bdev_io->internal.desc = desc;
6271 16 : bdev_io->type = SPDK_BDEV_IO_TYPE_COMPARE;
6272 16 : bdev_io->u.bdev.iovs = iov;
6273 16 : bdev_io->u.bdev.iovcnt = iovcnt;
6274 16 : bdev_io->u.bdev.md_buf = md_buf;
6275 16 : bdev_io->u.bdev.num_blocks = num_blocks;
6276 16 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6277 16 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6278 16 : bdev_io->u.bdev.memory_domain = NULL;
6279 16 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6280 16 : bdev_io->u.bdev.accel_sequence = NULL;
6281 :
6282 16 : if (bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COMPARE)) {
6283 7 : bdev_io_submit(bdev_io);
6284 7 : return 0;
6285 : }
6286 :
6287 9 : bdev_compare_do_read(bdev_io);
6288 :
6289 9 : return 0;
6290 16 : }
6291 :
6292 : int
6293 10 : spdk_bdev_comparev_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6294 : struct iovec *iov, int iovcnt,
6295 : uint64_t offset_blocks, uint64_t num_blocks,
6296 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6297 : {
6298 20 : return bdev_comparev_blocks_with_md(desc, ch, iov, iovcnt, NULL, offset_blocks,
6299 10 : num_blocks, cb, cb_arg);
6300 : }
6301 :
6302 : int
6303 6 : spdk_bdev_comparev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6304 : struct iovec *iov, int iovcnt, void *md_buf,
6305 : uint64_t offset_blocks, uint64_t num_blocks,
6306 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6307 : {
6308 6 : if (md_buf && !spdk_bdev_is_md_separate(spdk_bdev_desc_get_bdev(desc))) {
6309 0 : return -EINVAL;
6310 : }
6311 :
6312 6 : if (md_buf && !_is_buf_allocated(iov)) {
6313 0 : return -EINVAL;
6314 : }
6315 :
6316 12 : return bdev_comparev_blocks_with_md(desc, ch, iov, iovcnt, md_buf, offset_blocks,
6317 6 : num_blocks, cb, cb_arg);
6318 6 : }
6319 :
6320 : static int
6321 4 : bdev_compare_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6322 : void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
6323 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6324 : {
6325 4 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6326 : struct spdk_bdev_io *bdev_io;
6327 4 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6328 :
6329 4 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6330 0 : return -EINVAL;
6331 : }
6332 :
6333 4 : bdev_io = bdev_channel_get_io(channel);
6334 4 : if (!bdev_io) {
6335 0 : return -ENOMEM;
6336 : }
6337 :
6338 4 : bdev_io->internal.ch = channel;
6339 4 : bdev_io->internal.desc = desc;
6340 4 : bdev_io->type = SPDK_BDEV_IO_TYPE_COMPARE;
6341 4 : bdev_io->u.bdev.iovs = &bdev_io->iov;
6342 4 : bdev_io->u.bdev.iovs[0].iov_base = buf;
6343 4 : bdev_io->u.bdev.iovs[0].iov_len = num_blocks * bdev_desc_get_block_size(desc);
6344 4 : bdev_io->u.bdev.iovcnt = 1;
6345 4 : bdev_io->u.bdev.md_buf = md_buf;
6346 4 : bdev_io->u.bdev.num_blocks = num_blocks;
6347 4 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6348 4 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6349 4 : bdev_io->u.bdev.memory_domain = NULL;
6350 4 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6351 4 : bdev_io->u.bdev.accel_sequence = NULL;
6352 :
6353 4 : if (bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COMPARE)) {
6354 2 : bdev_io_submit(bdev_io);
6355 2 : return 0;
6356 : }
6357 :
6358 2 : bdev_compare_do_read(bdev_io);
6359 :
6360 2 : return 0;
6361 4 : }
6362 :
6363 : int
6364 4 : spdk_bdev_compare_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6365 : void *buf, uint64_t offset_blocks, uint64_t num_blocks,
6366 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6367 : {
6368 8 : return bdev_compare_blocks_with_md(desc, ch, buf, NULL, offset_blocks, num_blocks,
6369 4 : cb, cb_arg);
6370 : }
6371 :
6372 : int
6373 0 : spdk_bdev_compare_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6374 : void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
6375 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6376 : {
6377 0 : struct iovec iov = {
6378 0 : .iov_base = buf,
6379 : };
6380 :
6381 0 : if (md_buf && !spdk_bdev_is_md_separate(spdk_bdev_desc_get_bdev(desc))) {
6382 0 : return -EINVAL;
6383 : }
6384 :
6385 0 : if (md_buf && !_is_buf_allocated(&iov)) {
6386 0 : return -EINVAL;
6387 : }
6388 :
6389 0 : return bdev_compare_blocks_with_md(desc, ch, buf, md_buf, offset_blocks, num_blocks,
6390 0 : cb, cb_arg);
6391 0 : }
6392 :
6393 : static void
6394 2 : bdev_comparev_and_writev_blocks_unlocked(struct lba_range *range, void *ctx, int unlock_status)
6395 : {
6396 2 : struct spdk_bdev_io *bdev_io = ctx;
6397 :
6398 2 : if (unlock_status) {
6399 0 : SPDK_ERRLOG("LBA range unlock failed\n");
6400 0 : }
6401 :
6402 4 : bdev_io->internal.cb(bdev_io, bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS ? true :
6403 2 : false, bdev_io->internal.caller_ctx);
6404 2 : }
6405 :
6406 : static void
6407 2 : bdev_comparev_and_writev_blocks_unlock(struct spdk_bdev_io *bdev_io, int status)
6408 : {
6409 2 : bdev_io->internal.status = status;
6410 :
6411 4 : bdev_unlock_lba_range(bdev_io->internal.desc, spdk_io_channel_from_ctx(bdev_io->internal.ch),
6412 2 : bdev_io->u.bdev.offset_blocks, bdev_io->u.bdev.num_blocks,
6413 2 : bdev_comparev_and_writev_blocks_unlocked, bdev_io);
6414 2 : }
6415 :
6416 : static void
6417 1 : bdev_compare_and_write_do_write_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
6418 : {
6419 1 : struct spdk_bdev_io *parent_io = cb_arg;
6420 :
6421 1 : if (!success) {
6422 0 : SPDK_ERRLOG("Compare and write operation failed\n");
6423 0 : }
6424 :
6425 1 : spdk_bdev_free_io(bdev_io);
6426 :
6427 2 : bdev_comparev_and_writev_blocks_unlock(parent_io,
6428 1 : success ? SPDK_BDEV_IO_STATUS_SUCCESS : SPDK_BDEV_IO_STATUS_FAILED);
6429 1 : }
6430 :
6431 : static void
6432 1 : bdev_compare_and_write_do_write(void *_bdev_io)
6433 : {
6434 1 : struct spdk_bdev_io *bdev_io = _bdev_io;
6435 : int rc;
6436 :
6437 2 : rc = spdk_bdev_writev_blocks(bdev_io->internal.desc,
6438 1 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
6439 1 : bdev_io->u.bdev.fused_iovs, bdev_io->u.bdev.fused_iovcnt,
6440 1 : bdev_io->u.bdev.offset_blocks, bdev_io->u.bdev.num_blocks,
6441 1 : bdev_compare_and_write_do_write_done, bdev_io);
6442 :
6443 :
6444 1 : if (rc == -ENOMEM) {
6445 0 : bdev_queue_io_wait_with_cb(bdev_io, bdev_compare_and_write_do_write);
6446 1 : } else if (rc != 0) {
6447 0 : bdev_comparev_and_writev_blocks_unlock(bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
6448 0 : }
6449 1 : }
6450 :
6451 : static void
6452 2 : bdev_compare_and_write_do_compare_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
6453 : {
6454 2 : struct spdk_bdev_io *parent_io = cb_arg;
6455 :
6456 2 : spdk_bdev_free_io(bdev_io);
6457 :
6458 2 : if (!success) {
6459 1 : bdev_comparev_and_writev_blocks_unlock(parent_io, SPDK_BDEV_IO_STATUS_MISCOMPARE);
6460 1 : return;
6461 : }
6462 :
6463 1 : bdev_compare_and_write_do_write(parent_io);
6464 2 : }
6465 :
6466 : static void
6467 2 : bdev_compare_and_write_do_compare(void *_bdev_io)
6468 : {
6469 2 : struct spdk_bdev_io *bdev_io = _bdev_io;
6470 : int rc;
6471 :
6472 4 : rc = spdk_bdev_comparev_blocks(bdev_io->internal.desc,
6473 2 : spdk_io_channel_from_ctx(bdev_io->internal.ch), bdev_io->u.bdev.iovs,
6474 2 : bdev_io->u.bdev.iovcnt, bdev_io->u.bdev.offset_blocks, bdev_io->u.bdev.num_blocks,
6475 2 : bdev_compare_and_write_do_compare_done, bdev_io);
6476 :
6477 2 : if (rc == -ENOMEM) {
6478 0 : bdev_queue_io_wait_with_cb(bdev_io, bdev_compare_and_write_do_compare);
6479 2 : } else if (rc != 0) {
6480 0 : bdev_comparev_and_writev_blocks_unlock(bdev_io, SPDK_BDEV_IO_STATUS_FIRST_FUSED_FAILED);
6481 0 : }
6482 2 : }
6483 :
6484 : static void
6485 2 : bdev_comparev_and_writev_blocks_locked(struct lba_range *range, void *ctx, int status)
6486 : {
6487 2 : struct spdk_bdev_io *bdev_io = ctx;
6488 :
6489 2 : if (status) {
6490 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FIRST_FUSED_FAILED;
6491 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
6492 0 : return;
6493 : }
6494 :
6495 2 : bdev_compare_and_write_do_compare(bdev_io);
6496 2 : }
6497 :
6498 : int
6499 2 : spdk_bdev_comparev_and_writev_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6500 : struct iovec *compare_iov, int compare_iovcnt,
6501 : struct iovec *write_iov, int write_iovcnt,
6502 : uint64_t offset_blocks, uint64_t num_blocks,
6503 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6504 : {
6505 2 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6506 : struct spdk_bdev_io *bdev_io;
6507 2 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6508 :
6509 2 : if (!desc->write) {
6510 0 : return -EBADF;
6511 : }
6512 :
6513 2 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6514 0 : return -EINVAL;
6515 : }
6516 :
6517 2 : if (num_blocks > bdev->acwu) {
6518 0 : return -EINVAL;
6519 : }
6520 :
6521 2 : bdev_io = bdev_channel_get_io(channel);
6522 2 : if (!bdev_io) {
6523 0 : return -ENOMEM;
6524 : }
6525 :
6526 2 : bdev_io->internal.ch = channel;
6527 2 : bdev_io->internal.desc = desc;
6528 2 : bdev_io->type = SPDK_BDEV_IO_TYPE_COMPARE_AND_WRITE;
6529 2 : bdev_io->u.bdev.iovs = compare_iov;
6530 2 : bdev_io->u.bdev.iovcnt = compare_iovcnt;
6531 2 : bdev_io->u.bdev.fused_iovs = write_iov;
6532 2 : bdev_io->u.bdev.fused_iovcnt = write_iovcnt;
6533 2 : bdev_io->u.bdev.md_buf = NULL;
6534 2 : bdev_io->u.bdev.num_blocks = num_blocks;
6535 2 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6536 2 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6537 2 : bdev_io->u.bdev.memory_domain = NULL;
6538 2 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6539 2 : bdev_io->u.bdev.accel_sequence = NULL;
6540 :
6541 2 : if (bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COMPARE_AND_WRITE)) {
6542 0 : bdev_io_submit(bdev_io);
6543 0 : return 0;
6544 : }
6545 :
6546 4 : return bdev_lock_lba_range(desc, ch, offset_blocks, num_blocks,
6547 2 : bdev_comparev_and_writev_blocks_locked, bdev_io);
6548 2 : }
6549 :
6550 : int
6551 2 : spdk_bdev_zcopy_start(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6552 : struct iovec *iov, int iovcnt,
6553 : uint64_t offset_blocks, uint64_t num_blocks,
6554 : bool populate,
6555 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6556 : {
6557 2 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6558 : struct spdk_bdev_io *bdev_io;
6559 2 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6560 :
6561 2 : if (!desc->write) {
6562 0 : return -EBADF;
6563 : }
6564 :
6565 2 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6566 0 : return -EINVAL;
6567 : }
6568 :
6569 2 : if (!spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_ZCOPY)) {
6570 0 : return -ENOTSUP;
6571 : }
6572 :
6573 2 : bdev_io = bdev_channel_get_io(channel);
6574 2 : if (!bdev_io) {
6575 0 : return -ENOMEM;
6576 : }
6577 :
6578 2 : bdev_io->internal.ch = channel;
6579 2 : bdev_io->internal.desc = desc;
6580 2 : bdev_io->type = SPDK_BDEV_IO_TYPE_ZCOPY;
6581 2 : bdev_io->u.bdev.num_blocks = num_blocks;
6582 2 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6583 2 : bdev_io->u.bdev.iovs = iov;
6584 2 : bdev_io->u.bdev.iovcnt = iovcnt;
6585 2 : bdev_io->u.bdev.md_buf = NULL;
6586 2 : bdev_io->u.bdev.zcopy.populate = populate ? 1 : 0;
6587 2 : bdev_io->u.bdev.zcopy.commit = 0;
6588 2 : bdev_io->u.bdev.zcopy.start = 1;
6589 2 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6590 2 : bdev_io->u.bdev.memory_domain = NULL;
6591 2 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6592 2 : bdev_io->u.bdev.accel_sequence = NULL;
6593 :
6594 2 : bdev_io_submit(bdev_io);
6595 :
6596 2 : return 0;
6597 2 : }
6598 :
6599 : int
6600 2 : spdk_bdev_zcopy_end(struct spdk_bdev_io *bdev_io, bool commit,
6601 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6602 : {
6603 2 : if (bdev_io->type != SPDK_BDEV_IO_TYPE_ZCOPY) {
6604 0 : return -EINVAL;
6605 : }
6606 :
6607 2 : bdev_io->u.bdev.zcopy.commit = commit ? 1 : 0;
6608 2 : bdev_io->u.bdev.zcopy.start = 0;
6609 2 : bdev_io->internal.caller_ctx = cb_arg;
6610 2 : bdev_io->internal.cb = cb;
6611 2 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_PENDING;
6612 :
6613 2 : bdev_io_submit(bdev_io);
6614 :
6615 2 : return 0;
6616 2 : }
6617 :
6618 : int
6619 0 : spdk_bdev_write_zeroes(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6620 : uint64_t offset, uint64_t len,
6621 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6622 : {
6623 : uint64_t offset_blocks, num_blocks;
6624 :
6625 0 : if (bdev_bytes_to_blocks(desc, offset, &offset_blocks, len, &num_blocks) != 0) {
6626 0 : return -EINVAL;
6627 : }
6628 :
6629 0 : return spdk_bdev_write_zeroes_blocks(desc, ch, offset_blocks, num_blocks, cb, cb_arg);
6630 0 : }
6631 :
6632 : int
6633 33 : spdk_bdev_write_zeroes_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6634 : uint64_t offset_blocks, uint64_t num_blocks,
6635 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6636 : {
6637 33 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6638 : struct spdk_bdev_io *bdev_io;
6639 33 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6640 :
6641 33 : if (!desc->write) {
6642 0 : return -EBADF;
6643 : }
6644 :
6645 33 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6646 0 : return -EINVAL;
6647 : }
6648 :
6649 33 : if (!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE_ZEROES) &&
6650 10 : !bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE)) {
6651 1 : return -ENOTSUP;
6652 : }
6653 :
6654 32 : bdev_io = bdev_channel_get_io(channel);
6655 :
6656 32 : if (!bdev_io) {
6657 0 : return -ENOMEM;
6658 : }
6659 :
6660 32 : bdev_io->type = SPDK_BDEV_IO_TYPE_WRITE_ZEROES;
6661 32 : bdev_io->internal.ch = channel;
6662 32 : bdev_io->internal.desc = desc;
6663 32 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6664 32 : bdev_io->u.bdev.num_blocks = num_blocks;
6665 32 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6666 32 : bdev_io->u.bdev.memory_domain = NULL;
6667 32 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6668 32 : bdev_io->u.bdev.accel_sequence = NULL;
6669 :
6670 : /* If the write_zeroes size is large and should be split, use the generic split
6671 : * logic regardless of whether SPDK_BDEV_IO_TYPE_WRITE_ZEREOS is supported or not.
6672 : *
6673 : * Then, send the write_zeroes request if SPDK_BDEV_IO_TYPE_WRITE_ZEROES is supported
6674 : * or emulate it using regular write request otherwise.
6675 : */
6676 32 : if (bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE_ZEROES) ||
6677 9 : bdev_io->internal.f.split) {
6678 26 : bdev_io_submit(bdev_io);
6679 26 : return 0;
6680 : }
6681 :
6682 6 : assert(_bdev_get_block_size_with_md(bdev) <= ZERO_BUFFER_SIZE);
6683 :
6684 6 : return bdev_write_zero_buffer(bdev_io);
6685 33 : }
6686 :
6687 : int
6688 0 : spdk_bdev_unmap(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6689 : uint64_t offset, uint64_t nbytes,
6690 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6691 : {
6692 : uint64_t offset_blocks, num_blocks;
6693 :
6694 0 : if (bdev_bytes_to_blocks(desc, offset, &offset_blocks, nbytes, &num_blocks) != 0) {
6695 0 : return -EINVAL;
6696 : }
6697 :
6698 0 : return spdk_bdev_unmap_blocks(desc, ch, offset_blocks, num_blocks, cb, cb_arg);
6699 0 : }
6700 :
6701 : static void
6702 0 : bdev_io_complete_cb(void *ctx)
6703 : {
6704 0 : struct spdk_bdev_io *bdev_io = ctx;
6705 :
6706 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
6707 0 : bdev_io->internal.cb(bdev_io, true, bdev_io->internal.caller_ctx);
6708 0 : }
6709 :
6710 : int
6711 22 : spdk_bdev_unmap_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6712 : uint64_t offset_blocks, uint64_t num_blocks,
6713 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6714 : {
6715 22 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6716 : struct spdk_bdev_io *bdev_io;
6717 22 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6718 :
6719 22 : if (!desc->write) {
6720 0 : return -EBADF;
6721 : }
6722 :
6723 22 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6724 0 : return -EINVAL;
6725 : }
6726 :
6727 22 : bdev_io = bdev_channel_get_io(channel);
6728 22 : if (!bdev_io) {
6729 0 : return -ENOMEM;
6730 : }
6731 :
6732 22 : bdev_io->internal.ch = channel;
6733 22 : bdev_io->internal.desc = desc;
6734 22 : bdev_io->type = SPDK_BDEV_IO_TYPE_UNMAP;
6735 :
6736 22 : bdev_io->u.bdev.iovs = &bdev_io->iov;
6737 22 : bdev_io->u.bdev.iovs[0].iov_base = NULL;
6738 22 : bdev_io->u.bdev.iovs[0].iov_len = 0;
6739 22 : bdev_io->u.bdev.iovcnt = 1;
6740 :
6741 22 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6742 22 : bdev_io->u.bdev.num_blocks = num_blocks;
6743 22 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6744 22 : bdev_io->u.bdev.memory_domain = NULL;
6745 22 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6746 22 : bdev_io->u.bdev.accel_sequence = NULL;
6747 :
6748 22 : if (num_blocks == 0) {
6749 0 : spdk_thread_send_msg(spdk_get_thread(), bdev_io_complete_cb, bdev_io);
6750 0 : return 0;
6751 : }
6752 :
6753 22 : bdev_io_submit(bdev_io);
6754 22 : return 0;
6755 22 : }
6756 :
6757 : int
6758 0 : spdk_bdev_flush(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6759 : uint64_t offset, uint64_t length,
6760 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6761 : {
6762 : uint64_t offset_blocks, num_blocks;
6763 :
6764 0 : if (bdev_bytes_to_blocks(desc, offset, &offset_blocks, length, &num_blocks) != 0) {
6765 0 : return -EINVAL;
6766 : }
6767 :
6768 0 : return spdk_bdev_flush_blocks(desc, ch, offset_blocks, num_blocks, cb, cb_arg);
6769 0 : }
6770 :
6771 : int
6772 2 : spdk_bdev_flush_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6773 : uint64_t offset_blocks, uint64_t num_blocks,
6774 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6775 : {
6776 2 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6777 : struct spdk_bdev_io *bdev_io;
6778 2 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6779 :
6780 2 : if (!desc->write) {
6781 0 : return -EBADF;
6782 : }
6783 :
6784 2 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6785 0 : return -EINVAL;
6786 : }
6787 :
6788 2 : bdev_io = bdev_channel_get_io(channel);
6789 2 : if (!bdev_io) {
6790 0 : return -ENOMEM;
6791 : }
6792 :
6793 2 : bdev_io->internal.ch = channel;
6794 2 : bdev_io->internal.desc = desc;
6795 2 : bdev_io->type = SPDK_BDEV_IO_TYPE_FLUSH;
6796 2 : bdev_io->u.bdev.iovs = NULL;
6797 2 : bdev_io->u.bdev.iovcnt = 0;
6798 2 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6799 2 : bdev_io->u.bdev.num_blocks = num_blocks;
6800 2 : bdev_io->u.bdev.memory_domain = NULL;
6801 2 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6802 2 : bdev_io->u.bdev.accel_sequence = NULL;
6803 2 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6804 :
6805 2 : bdev_io_submit(bdev_io);
6806 2 : return 0;
6807 2 : }
6808 :
6809 : static int bdev_reset_poll_for_outstanding_io(void *ctx);
6810 :
6811 : static void
6812 13 : bdev_reset_check_outstanding_io_done(struct spdk_bdev *bdev, void *_ctx, int status)
6813 : {
6814 13 : struct spdk_bdev_channel *ch = _ctx;
6815 : struct spdk_bdev_io *bdev_io;
6816 :
6817 13 : bdev_io = TAILQ_FIRST(&ch->queued_resets);
6818 :
6819 13 : if (status == -EBUSY) {
6820 9 : if (spdk_get_ticks() < bdev_io->u.reset.wait_poller.stop_time_tsc) {
6821 8 : bdev_io->u.reset.wait_poller.poller = SPDK_POLLER_REGISTER(bdev_reset_poll_for_outstanding_io,
6822 : ch, BDEV_RESET_CHECK_OUTSTANDING_IO_PERIOD);
6823 8 : } else {
6824 1 : TAILQ_REMOVE(&ch->queued_resets, bdev_io, internal.link);
6825 :
6826 1 : if (TAILQ_EMPTY(&ch->io_memory_domain) && TAILQ_EMPTY(&ch->io_accel_exec)) {
6827 : /* If outstanding IOs are still present and reset_io_drain_timeout
6828 : * seconds passed, start the reset. */
6829 1 : bdev_io_submit_reset(bdev_io);
6830 1 : } else {
6831 : /* We still have in progress memory domain pull/push or we're
6832 : * executing accel sequence. Since we cannot abort either of those
6833 : * operations, fail the reset request. */
6834 0 : spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
6835 : }
6836 : }
6837 9 : } else {
6838 4 : TAILQ_REMOVE(&ch->queued_resets, bdev_io, internal.link);
6839 4 : SPDK_DEBUGLOG(bdev,
6840 : "Skipping reset for underlying device of bdev: %s - no outstanding I/O.\n",
6841 : ch->bdev->name);
6842 : /* Mark the completion status as a SUCCESS and complete the reset. */
6843 4 : spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_SUCCESS);
6844 : }
6845 13 : }
6846 :
6847 : static void
6848 13 : bdev_reset_check_outstanding_io(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
6849 : struct spdk_io_channel *io_ch, void *_ctx)
6850 : {
6851 13 : struct spdk_bdev_channel *cur_ch = __io_ch_to_bdev_ch(io_ch);
6852 13 : int status = 0;
6853 :
6854 17 : if (cur_ch->io_outstanding > 0 ||
6855 4 : !TAILQ_EMPTY(&cur_ch->io_memory_domain) ||
6856 4 : !TAILQ_EMPTY(&cur_ch->io_accel_exec)) {
6857 : /* If a channel has outstanding IO, set status to -EBUSY code. This will stop
6858 : * further iteration over the rest of the channels and pass non-zero status
6859 : * to the callback function. */
6860 9 : status = -EBUSY;
6861 9 : }
6862 13 : spdk_bdev_for_each_channel_continue(i, status);
6863 13 : }
6864 :
6865 : static int
6866 8 : bdev_reset_poll_for_outstanding_io(void *ctx)
6867 : {
6868 8 : struct spdk_bdev_channel *ch = ctx;
6869 : struct spdk_bdev_io *bdev_io;
6870 :
6871 8 : bdev_io = TAILQ_FIRST(&ch->queued_resets);
6872 :
6873 8 : spdk_poller_unregister(&bdev_io->u.reset.wait_poller.poller);
6874 8 : spdk_bdev_for_each_channel(ch->bdev, bdev_reset_check_outstanding_io, ch,
6875 : bdev_reset_check_outstanding_io_done);
6876 :
6877 8 : return SPDK_POLLER_BUSY;
6878 : }
6879 :
6880 : static void
6881 15 : bdev_reset_freeze_channel_done(struct spdk_bdev *bdev, void *_ctx, int status)
6882 : {
6883 15 : struct spdk_bdev_channel *ch = _ctx;
6884 : struct spdk_bdev_io *bdev_io;
6885 :
6886 15 : bdev_io = TAILQ_FIRST(&ch->queued_resets);
6887 :
6888 15 : if (bdev->reset_io_drain_timeout == 0) {
6889 10 : TAILQ_REMOVE(&ch->queued_resets, bdev_io, internal.link);
6890 :
6891 10 : bdev_io_submit_reset(bdev_io);
6892 10 : return;
6893 : }
6894 :
6895 10 : bdev_io->u.reset.wait_poller.stop_time_tsc = spdk_get_ticks() +
6896 5 : (ch->bdev->reset_io_drain_timeout * spdk_get_ticks_hz());
6897 :
6898 : /* In case bdev->reset_io_drain_timeout is not equal to zero,
6899 : * submit the reset to the underlying module only if outstanding I/O
6900 : * remain after reset_io_drain_timeout seconds have passed. */
6901 5 : spdk_bdev_for_each_channel(ch->bdev, bdev_reset_check_outstanding_io, ch,
6902 : bdev_reset_check_outstanding_io_done);
6903 15 : }
6904 :
6905 : static void
6906 18 : bdev_reset_freeze_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
6907 : struct spdk_io_channel *ch, void *_ctx)
6908 : {
6909 : struct spdk_bdev_channel *channel;
6910 : struct spdk_bdev_mgmt_channel *mgmt_channel;
6911 : struct spdk_bdev_shared_resource *shared_resource;
6912 : bdev_io_tailq_t tmp_queued;
6913 :
6914 18 : TAILQ_INIT(&tmp_queued);
6915 :
6916 18 : channel = __io_ch_to_bdev_ch(ch);
6917 18 : shared_resource = channel->shared_resource;
6918 18 : mgmt_channel = shared_resource->mgmt_ch;
6919 :
6920 18 : channel->flags |= BDEV_CH_RESET_IN_PROGRESS;
6921 :
6922 18 : if ((channel->flags & BDEV_CH_QOS_ENABLED) != 0) {
6923 2 : TAILQ_SWAP(&channel->qos_queued_io, &tmp_queued, spdk_bdev_io, internal.link);
6924 2 : }
6925 :
6926 18 : bdev_abort_all_queued_io(&shared_resource->nomem_io, channel);
6927 18 : bdev_abort_all_buf_io(mgmt_channel, channel);
6928 18 : bdev_abort_all_queued_io(&tmp_queued, channel);
6929 :
6930 18 : spdk_bdev_for_each_channel_continue(i, 0);
6931 18 : }
6932 :
6933 : static void
6934 15 : bdev_start_reset(void *ctx)
6935 : {
6936 15 : struct spdk_bdev_channel *ch = ctx;
6937 :
6938 15 : spdk_bdev_for_each_channel(ch->bdev, bdev_reset_freeze_channel, ch,
6939 : bdev_reset_freeze_channel_done);
6940 15 : }
6941 :
6942 : static void
6943 16 : bdev_channel_start_reset(struct spdk_bdev_channel *ch)
6944 : {
6945 16 : struct spdk_bdev *bdev = ch->bdev;
6946 :
6947 16 : assert(!TAILQ_EMPTY(&ch->queued_resets));
6948 :
6949 16 : spdk_spin_lock(&bdev->internal.spinlock);
6950 16 : if (bdev->internal.reset_in_progress == NULL) {
6951 15 : bdev->internal.reset_in_progress = TAILQ_FIRST(&ch->queued_resets);
6952 : /*
6953 : * Take a channel reference for the target bdev for the life of this
6954 : * reset. This guards against the channel getting destroyed while
6955 : * spdk_bdev_for_each_channel() calls related to this reset IO are in
6956 : * progress. We will release the reference when this reset is
6957 : * completed.
6958 : */
6959 15 : bdev->internal.reset_in_progress->u.reset.ch_ref = spdk_get_io_channel(__bdev_to_io_dev(bdev));
6960 15 : bdev_start_reset(ch);
6961 15 : }
6962 16 : spdk_spin_unlock(&bdev->internal.spinlock);
6963 16 : }
6964 :
6965 : int
6966 16 : spdk_bdev_reset(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6967 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6968 : {
6969 16 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6970 : struct spdk_bdev_io *bdev_io;
6971 16 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6972 :
6973 16 : bdev_io = bdev_channel_get_io(channel);
6974 16 : if (!bdev_io) {
6975 0 : return -ENOMEM;
6976 : }
6977 :
6978 16 : bdev_io->internal.ch = channel;
6979 16 : bdev_io->internal.desc = desc;
6980 16 : bdev_io->internal.submit_tsc = spdk_get_ticks();
6981 16 : bdev_io->type = SPDK_BDEV_IO_TYPE_RESET;
6982 16 : bdev_io->u.reset.ch_ref = NULL;
6983 16 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6984 :
6985 16 : spdk_spin_lock(&bdev->internal.spinlock);
6986 16 : TAILQ_INSERT_TAIL(&channel->queued_resets, bdev_io, internal.link);
6987 16 : spdk_spin_unlock(&bdev->internal.spinlock);
6988 :
6989 16 : bdev_ch_add_to_io_submitted(bdev_io);
6990 :
6991 16 : bdev_channel_start_reset(channel);
6992 :
6993 16 : return 0;
6994 16 : }
6995 :
6996 : void
6997 0 : spdk_bdev_get_io_stat(struct spdk_bdev *bdev, struct spdk_io_channel *ch,
6998 : struct spdk_bdev_io_stat *stat, enum spdk_bdev_reset_stat_mode reset_mode)
6999 : {
7000 0 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
7001 :
7002 0 : bdev_get_io_stat(stat, channel->stat);
7003 0 : spdk_bdev_reset_io_stat(stat, reset_mode);
7004 0 : }
7005 :
7006 : static void
7007 5 : bdev_get_device_stat_done(struct spdk_bdev *bdev, void *_ctx, int status)
7008 : {
7009 5 : struct spdk_bdev_iostat_ctx *bdev_iostat_ctx = _ctx;
7010 :
7011 10 : bdev_iostat_ctx->cb(bdev, bdev_iostat_ctx->stat,
7012 5 : bdev_iostat_ctx->cb_arg, 0);
7013 5 : free(bdev_iostat_ctx);
7014 5 : }
7015 :
7016 : static void
7017 4 : bdev_get_each_channel_stat(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
7018 : struct spdk_io_channel *ch, void *_ctx)
7019 : {
7020 4 : struct spdk_bdev_iostat_ctx *bdev_iostat_ctx = _ctx;
7021 4 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
7022 :
7023 4 : spdk_bdev_add_io_stat(bdev_iostat_ctx->stat, channel->stat);
7024 4 : spdk_bdev_reset_io_stat(channel->stat, bdev_iostat_ctx->reset_mode);
7025 4 : spdk_bdev_for_each_channel_continue(i, 0);
7026 4 : }
7027 :
7028 : void
7029 5 : spdk_bdev_get_device_stat(struct spdk_bdev *bdev, struct spdk_bdev_io_stat *stat,
7030 : enum spdk_bdev_reset_stat_mode reset_mode, spdk_bdev_get_device_stat_cb cb, void *cb_arg)
7031 : {
7032 : struct spdk_bdev_iostat_ctx *bdev_iostat_ctx;
7033 :
7034 5 : assert(bdev != NULL);
7035 5 : assert(stat != NULL);
7036 5 : assert(cb != NULL);
7037 :
7038 5 : bdev_iostat_ctx = calloc(1, sizeof(struct spdk_bdev_iostat_ctx));
7039 5 : if (bdev_iostat_ctx == NULL) {
7040 0 : SPDK_ERRLOG("Unable to allocate memory for spdk_bdev_iostat_ctx\n");
7041 0 : cb(bdev, stat, cb_arg, -ENOMEM);
7042 0 : return;
7043 : }
7044 :
7045 5 : bdev_iostat_ctx->stat = stat;
7046 5 : bdev_iostat_ctx->cb = cb;
7047 5 : bdev_iostat_ctx->cb_arg = cb_arg;
7048 5 : bdev_iostat_ctx->reset_mode = reset_mode;
7049 :
7050 : /* Start with the statistics from previously deleted channels. */
7051 5 : spdk_spin_lock(&bdev->internal.spinlock);
7052 5 : bdev_get_io_stat(bdev_iostat_ctx->stat, bdev->internal.stat);
7053 5 : spdk_bdev_reset_io_stat(bdev->internal.stat, reset_mode);
7054 5 : spdk_spin_unlock(&bdev->internal.spinlock);
7055 :
7056 : /* Then iterate and add the statistics from each existing channel. */
7057 5 : spdk_bdev_for_each_channel(bdev, bdev_get_each_channel_stat, bdev_iostat_ctx,
7058 : bdev_get_device_stat_done);
7059 5 : }
7060 :
7061 : struct bdev_iostat_reset_ctx {
7062 : enum spdk_bdev_reset_stat_mode mode;
7063 : bdev_reset_device_stat_cb cb;
7064 : void *cb_arg;
7065 : };
7066 :
7067 : static void
7068 0 : bdev_reset_device_stat_done(struct spdk_bdev *bdev, void *_ctx, int status)
7069 : {
7070 0 : struct bdev_iostat_reset_ctx *ctx = _ctx;
7071 :
7072 0 : ctx->cb(bdev, ctx->cb_arg, 0);
7073 :
7074 0 : free(ctx);
7075 0 : }
7076 :
7077 : static void
7078 0 : bdev_reset_each_channel_stat(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
7079 : struct spdk_io_channel *ch, void *_ctx)
7080 : {
7081 0 : struct bdev_iostat_reset_ctx *ctx = _ctx;
7082 0 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
7083 :
7084 0 : spdk_bdev_reset_io_stat(channel->stat, ctx->mode);
7085 :
7086 0 : spdk_bdev_for_each_channel_continue(i, 0);
7087 0 : }
7088 :
7089 : void
7090 0 : bdev_reset_device_stat(struct spdk_bdev *bdev, enum spdk_bdev_reset_stat_mode mode,
7091 : bdev_reset_device_stat_cb cb, void *cb_arg)
7092 : {
7093 : struct bdev_iostat_reset_ctx *ctx;
7094 :
7095 0 : assert(bdev != NULL);
7096 0 : assert(cb != NULL);
7097 :
7098 0 : ctx = calloc(1, sizeof(*ctx));
7099 0 : if (ctx == NULL) {
7100 0 : SPDK_ERRLOG("Unable to allocate bdev_iostat_reset_ctx.\n");
7101 0 : cb(bdev, cb_arg, -ENOMEM);
7102 0 : return;
7103 : }
7104 :
7105 0 : ctx->mode = mode;
7106 0 : ctx->cb = cb;
7107 0 : ctx->cb_arg = cb_arg;
7108 :
7109 0 : spdk_spin_lock(&bdev->internal.spinlock);
7110 0 : spdk_bdev_reset_io_stat(bdev->internal.stat, mode);
7111 0 : spdk_spin_unlock(&bdev->internal.spinlock);
7112 :
7113 0 : spdk_bdev_for_each_channel(bdev,
7114 : bdev_reset_each_channel_stat,
7115 0 : ctx,
7116 : bdev_reset_device_stat_done);
7117 0 : }
7118 :
7119 : int
7120 1 : spdk_bdev_nvme_admin_passthru(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
7121 : const struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes,
7122 : spdk_bdev_io_completion_cb cb, void *cb_arg)
7123 : {
7124 1 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
7125 : struct spdk_bdev_io *bdev_io;
7126 1 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
7127 :
7128 1 : if (!desc->write) {
7129 0 : return -EBADF;
7130 : }
7131 :
7132 1 : if (spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_ADMIN))) {
7133 1 : return -ENOTSUP;
7134 : }
7135 :
7136 0 : bdev_io = bdev_channel_get_io(channel);
7137 0 : if (!bdev_io) {
7138 0 : return -ENOMEM;
7139 : }
7140 :
7141 0 : bdev_io->internal.ch = channel;
7142 0 : bdev_io->internal.desc = desc;
7143 0 : bdev_io->type = SPDK_BDEV_IO_TYPE_NVME_ADMIN;
7144 0 : bdev_io->u.nvme_passthru.cmd = *cmd;
7145 0 : bdev_io->u.nvme_passthru.buf = buf;
7146 0 : bdev_io->u.nvme_passthru.nbytes = nbytes;
7147 0 : bdev_io->u.nvme_passthru.md_buf = NULL;
7148 0 : bdev_io->u.nvme_passthru.md_len = 0;
7149 :
7150 0 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
7151 :
7152 0 : bdev_io_submit(bdev_io);
7153 0 : return 0;
7154 1 : }
7155 :
7156 : int
7157 1 : spdk_bdev_nvme_io_passthru(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
7158 : const struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes,
7159 : spdk_bdev_io_completion_cb cb, void *cb_arg)
7160 : {
7161 1 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
7162 : struct spdk_bdev_io *bdev_io;
7163 1 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
7164 :
7165 1 : if (!desc->write) {
7166 : /*
7167 : * Do not try to parse the NVMe command - we could maybe use bits in the opcode
7168 : * to easily determine if the command is a read or write, but for now just
7169 : * do not allow io_passthru with a read-only descriptor.
7170 : */
7171 0 : return -EBADF;
7172 : }
7173 :
7174 1 : if (spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_IO))) {
7175 1 : return -ENOTSUP;
7176 : }
7177 :
7178 0 : bdev_io = bdev_channel_get_io(channel);
7179 0 : if (!bdev_io) {
7180 0 : return -ENOMEM;
7181 : }
7182 :
7183 0 : bdev_io->internal.ch = channel;
7184 0 : bdev_io->internal.desc = desc;
7185 0 : bdev_io->type = SPDK_BDEV_IO_TYPE_NVME_IO;
7186 0 : bdev_io->u.nvme_passthru.cmd = *cmd;
7187 0 : bdev_io->u.nvme_passthru.buf = buf;
7188 0 : bdev_io->u.nvme_passthru.nbytes = nbytes;
7189 0 : bdev_io->u.nvme_passthru.md_buf = NULL;
7190 0 : bdev_io->u.nvme_passthru.md_len = 0;
7191 :
7192 0 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
7193 :
7194 0 : bdev_io_submit(bdev_io);
7195 0 : return 0;
7196 1 : }
7197 :
7198 : int
7199 1 : spdk_bdev_nvme_io_passthru_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
7200 : const struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes, void *md_buf, size_t md_len,
7201 : spdk_bdev_io_completion_cb cb, void *cb_arg)
7202 : {
7203 1 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
7204 : struct spdk_bdev_io *bdev_io;
7205 1 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
7206 :
7207 1 : if (!desc->write) {
7208 : /*
7209 : * Do not try to parse the NVMe command - we could maybe use bits in the opcode
7210 : * to easily determine if the command is a read or write, but for now just
7211 : * do not allow io_passthru with a read-only descriptor.
7212 : */
7213 0 : return -EBADF;
7214 : }
7215 :
7216 1 : if (spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_IO_MD))) {
7217 1 : return -ENOTSUP;
7218 : }
7219 :
7220 0 : bdev_io = bdev_channel_get_io(channel);
7221 0 : if (!bdev_io) {
7222 0 : return -ENOMEM;
7223 : }
7224 :
7225 0 : bdev_io->internal.ch = channel;
7226 0 : bdev_io->internal.desc = desc;
7227 0 : bdev_io->type = SPDK_BDEV_IO_TYPE_NVME_IO_MD;
7228 0 : bdev_io->u.nvme_passthru.cmd = *cmd;
7229 0 : bdev_io->u.nvme_passthru.buf = buf;
7230 0 : bdev_io->u.nvme_passthru.nbytes = nbytes;
7231 0 : bdev_io->u.nvme_passthru.md_buf = md_buf;
7232 0 : bdev_io->u.nvme_passthru.md_len = md_len;
7233 :
7234 0 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
7235 :
7236 0 : bdev_io_submit(bdev_io);
7237 0 : return 0;
7238 1 : }
7239 :
7240 : int
7241 0 : spdk_bdev_nvme_iov_passthru_md(struct spdk_bdev_desc *desc,
7242 : struct spdk_io_channel *ch,
7243 : const struct spdk_nvme_cmd *cmd,
7244 : struct iovec *iov, int iovcnt, size_t nbytes,
7245 : void *md_buf, size_t md_len,
7246 : spdk_bdev_io_completion_cb cb, void *cb_arg)
7247 : {
7248 0 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
7249 : struct spdk_bdev_io *bdev_io;
7250 0 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
7251 :
7252 0 : if (!desc->write) {
7253 : /*
7254 : * Do not try to parse the NVMe command - we could maybe use bits in the opcode
7255 : * to easily determine if the command is a read or write, but for now just
7256 : * do not allow io_passthru with a read-only descriptor.
7257 : */
7258 0 : return -EBADF;
7259 : }
7260 :
7261 0 : if (md_buf && spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_IO_MD))) {
7262 0 : return -ENOTSUP;
7263 0 : } else if (spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_IO))) {
7264 0 : return -ENOTSUP;
7265 : }
7266 :
7267 0 : bdev_io = bdev_channel_get_io(channel);
7268 0 : if (!bdev_io) {
7269 0 : return -ENOMEM;
7270 : }
7271 :
7272 0 : bdev_io->internal.ch = channel;
7273 0 : bdev_io->internal.desc = desc;
7274 0 : bdev_io->type = SPDK_BDEV_IO_TYPE_NVME_IOV_MD;
7275 0 : bdev_io->u.nvme_passthru.cmd = *cmd;
7276 0 : bdev_io->u.nvme_passthru.iovs = iov;
7277 0 : bdev_io->u.nvme_passthru.iovcnt = iovcnt;
7278 0 : bdev_io->u.nvme_passthru.nbytes = nbytes;
7279 0 : bdev_io->u.nvme_passthru.md_buf = md_buf;
7280 0 : bdev_io->u.nvme_passthru.md_len = md_len;
7281 :
7282 0 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
7283 :
7284 0 : bdev_io_submit(bdev_io);
7285 0 : return 0;
7286 0 : }
7287 :
7288 : static void bdev_abort_retry(void *ctx);
7289 : static void bdev_abort(struct spdk_bdev_io *parent_io);
7290 :
7291 : static void
7292 22 : bdev_abort_io_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
7293 : {
7294 22 : struct spdk_bdev_channel *channel = bdev_io->internal.ch;
7295 22 : struct spdk_bdev_io *parent_io = cb_arg;
7296 : struct spdk_bdev_io *bio_to_abort, *tmp_io;
7297 :
7298 22 : bio_to_abort = bdev_io->u.abort.bio_to_abort;
7299 :
7300 22 : spdk_bdev_free_io(bdev_io);
7301 :
7302 22 : if (!success) {
7303 : /* Check if the target I/O completed in the meantime. */
7304 2 : TAILQ_FOREACH(tmp_io, &channel->io_submitted, internal.ch_link) {
7305 1 : if (tmp_io == bio_to_abort) {
7306 0 : break;
7307 : }
7308 1 : }
7309 :
7310 : /* If the target I/O still exists, set the parent to failed. */
7311 1 : if (tmp_io != NULL) {
7312 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
7313 0 : }
7314 1 : }
7315 :
7316 22 : assert(parent_io->internal.f.split);
7317 :
7318 22 : parent_io->internal.split.outstanding--;
7319 22 : if (parent_io->internal.split.outstanding == 0) {
7320 16 : if (parent_io->internal.status == SPDK_BDEV_IO_STATUS_NOMEM) {
7321 0 : bdev_abort_retry(parent_io);
7322 0 : } else {
7323 16 : bdev_io_complete(parent_io);
7324 : }
7325 16 : }
7326 22 : }
7327 :
7328 : static int
7329 23 : bdev_abort_io(struct spdk_bdev_desc *desc, struct spdk_bdev_channel *channel,
7330 : struct spdk_bdev_io *bio_to_abort,
7331 : spdk_bdev_io_completion_cb cb, void *cb_arg)
7332 : {
7333 23 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
7334 : struct spdk_bdev_io *bdev_io;
7335 :
7336 23 : if (bio_to_abort->type == SPDK_BDEV_IO_TYPE_ABORT ||
7337 23 : bio_to_abort->type == SPDK_BDEV_IO_TYPE_RESET) {
7338 : /* TODO: Abort reset or abort request. */
7339 0 : return -ENOTSUP;
7340 : }
7341 :
7342 23 : bdev_io = bdev_channel_get_io(channel);
7343 23 : if (bdev_io == NULL) {
7344 1 : return -ENOMEM;
7345 : }
7346 :
7347 22 : bdev_io->internal.ch = channel;
7348 22 : bdev_io->internal.desc = desc;
7349 22 : bdev_io->type = SPDK_BDEV_IO_TYPE_ABORT;
7350 22 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
7351 :
7352 22 : if (bio_to_abort->internal.f.split) {
7353 6 : assert(bdev_io_should_split(bio_to_abort));
7354 6 : bdev_io->u.bdev.abort.bio_cb_arg = bio_to_abort;
7355 :
7356 : /* Parent abort request is not submitted directly, but to manage its
7357 : * execution add it to the submitted list here.
7358 : */
7359 6 : bdev_io->internal.submit_tsc = spdk_get_ticks();
7360 6 : bdev_ch_add_to_io_submitted(bdev_io);
7361 :
7362 6 : bdev_abort(bdev_io);
7363 :
7364 6 : return 0;
7365 : }
7366 :
7367 16 : bdev_io->u.abort.bio_to_abort = bio_to_abort;
7368 :
7369 : /* Submit the abort request to the underlying bdev module. */
7370 16 : bdev_io_submit(bdev_io);
7371 :
7372 16 : return 0;
7373 23 : }
7374 :
7375 : static bool
7376 46 : bdev_io_on_tailq(struct spdk_bdev_io *bdev_io, bdev_io_tailq_t *tailq)
7377 : {
7378 : struct spdk_bdev_io *iter;
7379 :
7380 46 : TAILQ_FOREACH(iter, tailq, internal.link) {
7381 0 : if (iter == bdev_io) {
7382 0 : return true;
7383 : }
7384 0 : }
7385 :
7386 46 : return false;
7387 46 : }
7388 :
7389 : static uint32_t
7390 18 : _bdev_abort(struct spdk_bdev_io *parent_io)
7391 : {
7392 18 : struct spdk_bdev_desc *desc = parent_io->internal.desc;
7393 18 : struct spdk_bdev_channel *channel = parent_io->internal.ch;
7394 : void *bio_cb_arg;
7395 : struct spdk_bdev_io *bio_to_abort;
7396 : uint32_t matched_ios;
7397 : int rc;
7398 :
7399 18 : bio_cb_arg = parent_io->u.bdev.abort.bio_cb_arg;
7400 :
7401 : /* matched_ios is returned and will be kept by the caller.
7402 : *
7403 : * This function will be used for two cases, 1) the same cb_arg is used for
7404 : * multiple I/Os, 2) a single large I/O is split into smaller ones.
7405 : * Incrementing split_outstanding directly here may confuse readers especially
7406 : * for the 1st case.
7407 : *
7408 : * Completion of I/O abort is processed after stack unwinding. Hence this trick
7409 : * works as expected.
7410 : */
7411 18 : matched_ios = 0;
7412 18 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
7413 :
7414 105 : TAILQ_FOREACH(bio_to_abort, &channel->io_submitted, internal.ch_link) {
7415 88 : if (bio_to_abort->internal.caller_ctx != bio_cb_arg) {
7416 65 : continue;
7417 : }
7418 :
7419 23 : if (bio_to_abort->internal.submit_tsc > parent_io->internal.submit_tsc) {
7420 : /* Any I/O which was submitted after this abort command should be excluded. */
7421 0 : continue;
7422 : }
7423 :
7424 : /* We can't abort a request that's being pushed/pulled or executed by accel */
7425 23 : if (bdev_io_on_tailq(bio_to_abort, &channel->io_accel_exec) ||
7426 23 : bdev_io_on_tailq(bio_to_abort, &channel->io_memory_domain)) {
7427 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
7428 0 : break;
7429 : }
7430 :
7431 23 : rc = bdev_abort_io(desc, channel, bio_to_abort, bdev_abort_io_done, parent_io);
7432 23 : if (rc != 0) {
7433 1 : if (rc == -ENOMEM) {
7434 1 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_NOMEM;
7435 1 : } else {
7436 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
7437 : }
7438 1 : break;
7439 : }
7440 22 : matched_ios++;
7441 22 : }
7442 :
7443 18 : return matched_ios;
7444 : }
7445 :
7446 : static void
7447 1 : bdev_abort_retry(void *ctx)
7448 : {
7449 1 : struct spdk_bdev_io *parent_io = ctx;
7450 : uint32_t matched_ios;
7451 :
7452 1 : matched_ios = _bdev_abort(parent_io);
7453 :
7454 1 : if (matched_ios == 0) {
7455 0 : if (parent_io->internal.status == SPDK_BDEV_IO_STATUS_NOMEM) {
7456 0 : bdev_queue_io_wait_with_cb(parent_io, bdev_abort_retry);
7457 0 : } else {
7458 : /* For retry, the case that no target I/O was found is success
7459 : * because it means target I/Os completed in the meantime.
7460 : */
7461 0 : bdev_io_complete(parent_io);
7462 : }
7463 0 : return;
7464 : }
7465 :
7466 : /* Use split_outstanding to manage the progress of aborting I/Os. */
7467 1 : parent_io->internal.f.split = true;
7468 1 : parent_io->internal.split.outstanding = matched_ios;
7469 1 : }
7470 :
7471 : static void
7472 17 : bdev_abort(struct spdk_bdev_io *parent_io)
7473 : {
7474 : uint32_t matched_ios;
7475 :
7476 17 : matched_ios = _bdev_abort(parent_io);
7477 :
7478 17 : if (matched_ios == 0) {
7479 2 : if (parent_io->internal.status == SPDK_BDEV_IO_STATUS_NOMEM) {
7480 1 : bdev_queue_io_wait_with_cb(parent_io, bdev_abort_retry);
7481 1 : } else {
7482 : /* The case the no target I/O was found is failure. */
7483 1 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
7484 1 : bdev_io_complete(parent_io);
7485 : }
7486 2 : return;
7487 : }
7488 :
7489 : /* Use split_outstanding to manage the progress of aborting I/Os. */
7490 15 : parent_io->internal.f.split = true;
7491 15 : parent_io->internal.split.outstanding = matched_ios;
7492 17 : }
7493 :
7494 : int
7495 12 : spdk_bdev_abort(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
7496 : void *bio_cb_arg,
7497 : spdk_bdev_io_completion_cb cb, void *cb_arg)
7498 : {
7499 12 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
7500 12 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
7501 : struct spdk_bdev_io *bdev_io;
7502 :
7503 12 : if (bio_cb_arg == NULL) {
7504 0 : return -EINVAL;
7505 : }
7506 :
7507 12 : if (!spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_ABORT)) {
7508 1 : return -ENOTSUP;
7509 : }
7510 :
7511 11 : bdev_io = bdev_channel_get_io(channel);
7512 11 : if (bdev_io == NULL) {
7513 0 : return -ENOMEM;
7514 : }
7515 :
7516 11 : bdev_io->internal.ch = channel;
7517 11 : bdev_io->internal.desc = desc;
7518 11 : bdev_io->internal.submit_tsc = spdk_get_ticks();
7519 11 : bdev_io->type = SPDK_BDEV_IO_TYPE_ABORT;
7520 11 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
7521 :
7522 11 : bdev_io->u.bdev.abort.bio_cb_arg = bio_cb_arg;
7523 :
7524 : /* Parent abort request is not submitted directly, but to manage its execution,
7525 : * add it to the submitted list here.
7526 : */
7527 11 : bdev_ch_add_to_io_submitted(bdev_io);
7528 :
7529 11 : bdev_abort(bdev_io);
7530 :
7531 11 : return 0;
7532 12 : }
7533 :
7534 : int
7535 4 : spdk_bdev_queue_io_wait(struct spdk_bdev *bdev, struct spdk_io_channel *ch,
7536 : struct spdk_bdev_io_wait_entry *entry)
7537 : {
7538 4 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
7539 4 : struct spdk_bdev_mgmt_channel *mgmt_ch = channel->shared_resource->mgmt_ch;
7540 :
7541 4 : if (bdev != entry->bdev) {
7542 0 : SPDK_ERRLOG("bdevs do not match\n");
7543 0 : return -EINVAL;
7544 : }
7545 :
7546 4 : if (mgmt_ch->per_thread_cache_count > 0) {
7547 0 : SPDK_ERRLOG("Cannot queue io_wait if spdk_bdev_io available in per-thread cache\n");
7548 0 : return -EINVAL;
7549 : }
7550 :
7551 4 : TAILQ_INSERT_TAIL(&mgmt_ch->io_wait_queue, entry, link);
7552 4 : return 0;
7553 4 : }
7554 :
7555 : static inline void
7556 610 : bdev_io_update_io_stat(struct spdk_bdev_io *bdev_io, uint64_t tsc_diff)
7557 : {
7558 610 : enum spdk_bdev_io_status io_status = bdev_io->internal.status;
7559 610 : struct spdk_bdev_io_stat *io_stat = bdev_io->internal.ch->stat;
7560 610 : uint64_t num_blocks = bdev_io->u.bdev.num_blocks;
7561 610 : uint32_t blocklen = bdev_io->bdev->blocklen;
7562 :
7563 610 : if (spdk_likely(io_status == SPDK_BDEV_IO_STATUS_SUCCESS)) {
7564 516 : switch (bdev_io->type) {
7565 : case SPDK_BDEV_IO_TYPE_READ:
7566 321 : io_stat->bytes_read += num_blocks * blocklen;
7567 321 : io_stat->num_read_ops++;
7568 321 : io_stat->read_latency_ticks += tsc_diff;
7569 321 : if (io_stat->max_read_latency_ticks < tsc_diff) {
7570 7 : io_stat->max_read_latency_ticks = tsc_diff;
7571 7 : }
7572 321 : if (io_stat->min_read_latency_ticks > tsc_diff) {
7573 42 : io_stat->min_read_latency_ticks = tsc_diff;
7574 42 : }
7575 321 : break;
7576 : case SPDK_BDEV_IO_TYPE_WRITE:
7577 75 : io_stat->bytes_written += num_blocks * blocklen;
7578 75 : io_stat->num_write_ops++;
7579 75 : io_stat->write_latency_ticks += tsc_diff;
7580 75 : if (io_stat->max_write_latency_ticks < tsc_diff) {
7581 4 : io_stat->max_write_latency_ticks = tsc_diff;
7582 4 : }
7583 75 : if (io_stat->min_write_latency_ticks > tsc_diff) {
7584 25 : io_stat->min_write_latency_ticks = tsc_diff;
7585 25 : }
7586 75 : break;
7587 : case SPDK_BDEV_IO_TYPE_UNMAP:
7588 20 : io_stat->bytes_unmapped += num_blocks * blocklen;
7589 20 : io_stat->num_unmap_ops++;
7590 20 : io_stat->unmap_latency_ticks += tsc_diff;
7591 20 : if (io_stat->max_unmap_latency_ticks < tsc_diff) {
7592 0 : io_stat->max_unmap_latency_ticks = tsc_diff;
7593 0 : }
7594 20 : if (io_stat->min_unmap_latency_ticks > tsc_diff) {
7595 3 : io_stat->min_unmap_latency_ticks = tsc_diff;
7596 3 : }
7597 20 : break;
7598 : case SPDK_BDEV_IO_TYPE_ZCOPY:
7599 : /* Track the data in the start phase only */
7600 4 : if (bdev_io->u.bdev.zcopy.start) {
7601 2 : if (bdev_io->u.bdev.zcopy.populate) {
7602 1 : io_stat->bytes_read += num_blocks * blocklen;
7603 1 : io_stat->num_read_ops++;
7604 1 : io_stat->read_latency_ticks += tsc_diff;
7605 1 : if (io_stat->max_read_latency_ticks < tsc_diff) {
7606 0 : io_stat->max_read_latency_ticks = tsc_diff;
7607 0 : }
7608 1 : if (io_stat->min_read_latency_ticks > tsc_diff) {
7609 1 : io_stat->min_read_latency_ticks = tsc_diff;
7610 1 : }
7611 1 : } else {
7612 1 : io_stat->bytes_written += num_blocks * blocklen;
7613 1 : io_stat->num_write_ops++;
7614 1 : io_stat->write_latency_ticks += tsc_diff;
7615 1 : if (io_stat->max_write_latency_ticks < tsc_diff) {
7616 0 : io_stat->max_write_latency_ticks = tsc_diff;
7617 0 : }
7618 1 : if (io_stat->min_write_latency_ticks > tsc_diff) {
7619 1 : io_stat->min_write_latency_ticks = tsc_diff;
7620 1 : }
7621 : }
7622 2 : }
7623 4 : break;
7624 : case SPDK_BDEV_IO_TYPE_COPY:
7625 21 : io_stat->bytes_copied += num_blocks * blocklen;
7626 21 : io_stat->num_copy_ops++;
7627 21 : bdev_io->internal.ch->stat->copy_latency_ticks += tsc_diff;
7628 21 : if (io_stat->max_copy_latency_ticks < tsc_diff) {
7629 0 : io_stat->max_copy_latency_ticks = tsc_diff;
7630 0 : }
7631 21 : if (io_stat->min_copy_latency_ticks > tsc_diff) {
7632 4 : io_stat->min_copy_latency_ticks = tsc_diff;
7633 4 : }
7634 21 : break;
7635 : default:
7636 75 : break;
7637 : }
7638 610 : } else if (io_status <= SPDK_BDEV_IO_STATUS_FAILED && io_status >= SPDK_MIN_BDEV_IO_STATUS) {
7639 94 : io_stat = bdev_io->bdev->internal.stat;
7640 94 : assert(io_stat->io_error != NULL);
7641 :
7642 94 : spdk_spin_lock(&bdev_io->bdev->internal.spinlock);
7643 94 : io_stat->io_error->error_status[-io_status - 1]++;
7644 94 : spdk_spin_unlock(&bdev_io->bdev->internal.spinlock);
7645 94 : }
7646 :
7647 : #ifdef SPDK_CONFIG_VTUNE
7648 : uint64_t now_tsc = spdk_get_ticks();
7649 : if (now_tsc > (bdev_io->internal.ch->start_tsc + bdev_io->internal.ch->interval_tsc)) {
7650 : uint64_t data[5];
7651 : struct spdk_bdev_io_stat *prev_stat = bdev_io->internal.ch->prev_stat;
7652 :
7653 : data[0] = io_stat->num_read_ops - prev_stat->num_read_ops;
7654 : data[1] = io_stat->bytes_read - prev_stat->bytes_read;
7655 : data[2] = io_stat->num_write_ops - prev_stat->num_write_ops;
7656 : data[3] = io_stat->bytes_written - prev_stat->bytes_written;
7657 : data[4] = bdev_io->bdev->fn_table->get_spin_time ?
7658 : bdev_io->bdev->fn_table->get_spin_time(spdk_bdev_io_get_io_channel(bdev_io)) : 0;
7659 :
7660 : __itt_metadata_add(g_bdev_mgr.domain, __itt_null, bdev_io->internal.ch->handle,
7661 : __itt_metadata_u64, 5, data);
7662 :
7663 : memcpy(prev_stat, io_stat, sizeof(struct spdk_bdev_io_stat));
7664 : bdev_io->internal.ch->start_tsc = now_tsc;
7665 : }
7666 : #endif
7667 610 : }
7668 :
7669 : static inline void
7670 610 : _bdev_io_complete(void *ctx)
7671 : {
7672 610 : struct spdk_bdev_io *bdev_io = ctx;
7673 :
7674 610 : if (spdk_unlikely(bdev_io_use_accel_sequence(bdev_io))) {
7675 0 : assert(bdev_io->internal.status != SPDK_BDEV_IO_STATUS_SUCCESS);
7676 0 : spdk_accel_sequence_abort(bdev_io->internal.accel_sequence);
7677 0 : }
7678 :
7679 610 : assert(bdev_io->internal.cb != NULL);
7680 610 : assert(spdk_get_thread() == spdk_bdev_io_get_thread(bdev_io));
7681 :
7682 1220 : bdev_io->internal.cb(bdev_io, bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS,
7683 610 : bdev_io->internal.caller_ctx);
7684 610 : }
7685 :
7686 : static inline void
7687 618 : bdev_io_complete(void *ctx)
7688 : {
7689 618 : struct spdk_bdev_io *bdev_io = ctx;
7690 618 : struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
7691 : uint64_t tsc, tsc_diff;
7692 :
7693 618 : if (spdk_unlikely(bdev_io->internal.f.in_submit_request)) {
7694 : /*
7695 : * Defer completion to avoid potential infinite recursion if the
7696 : * user's completion callback issues a new I/O.
7697 : */
7698 16 : spdk_thread_send_msg(spdk_bdev_io_get_thread(bdev_io),
7699 8 : bdev_io_complete, bdev_io);
7700 8 : return;
7701 : }
7702 :
7703 610 : tsc = spdk_get_ticks();
7704 610 : tsc_diff = tsc - bdev_io->internal.submit_tsc;
7705 :
7706 610 : bdev_ch_remove_from_io_submitted(bdev_io);
7707 610 : spdk_trace_record_tsc(tsc, TRACE_BDEV_IO_DONE, bdev_ch->trace_id, 0, (uintptr_t)bdev_io,
7708 : bdev_io->internal.caller_ctx, bdev_ch->queue_depth);
7709 :
7710 610 : if (bdev_ch->histogram) {
7711 4 : if (bdev_io->bdev->internal.histogram_io_type == 0 ||
7712 0 : bdev_io->bdev->internal.histogram_io_type == bdev_io->type) {
7713 : /*
7714 : * Tally all I/O types if the histogram_io_type is set to 0.
7715 : */
7716 4 : spdk_histogram_data_tally(bdev_ch->histogram, tsc_diff);
7717 4 : }
7718 4 : }
7719 :
7720 610 : bdev_io_update_io_stat(bdev_io, tsc_diff);
7721 610 : _bdev_io_complete(bdev_io);
7722 618 : }
7723 :
7724 : /* The difference between this function and bdev_io_complete() is that this should be called to
7725 : * complete IOs that haven't been submitted via bdev_io_submit(), as they weren't added onto the
7726 : * io_submitted list and don't have submit_tsc updated.
7727 : */
7728 : static inline void
7729 0 : bdev_io_complete_unsubmitted(struct spdk_bdev_io *bdev_io)
7730 : {
7731 : /* Since the IO hasn't been submitted it's bound to be failed */
7732 0 : assert(bdev_io->internal.status != SPDK_BDEV_IO_STATUS_SUCCESS);
7733 :
7734 : /* At this point we don't know if the IO is completed from submission context or not, but,
7735 : * since this is an error path, we can always do an spdk_thread_send_msg(). */
7736 0 : spdk_thread_send_msg(spdk_bdev_io_get_thread(bdev_io),
7737 0 : _bdev_io_complete, bdev_io);
7738 0 : }
7739 :
7740 : static void bdev_destroy_cb(void *io_device);
7741 :
7742 : static void
7743 15 : bdev_reset_complete(struct spdk_bdev *bdev, void *_ctx, int status)
7744 : {
7745 15 : struct spdk_bdev_io *bdev_io = _ctx;
7746 :
7747 15 : if (bdev_io->u.reset.ch_ref != NULL) {
7748 15 : spdk_put_io_channel(bdev_io->u.reset.ch_ref);
7749 15 : bdev_io->u.reset.ch_ref = NULL;
7750 15 : }
7751 :
7752 15 : bdev_io_complete(bdev_io);
7753 :
7754 15 : if (bdev->internal.status == SPDK_BDEV_STATUS_REMOVING &&
7755 1 : TAILQ_EMPTY(&bdev->internal.open_descs)) {
7756 1 : spdk_io_device_unregister(__bdev_to_io_dev(bdev), bdev_destroy_cb);
7757 1 : }
7758 15 : }
7759 :
7760 : static void
7761 18 : bdev_unfreeze_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
7762 : struct spdk_io_channel *_ch, void *_ctx)
7763 : {
7764 18 : struct spdk_bdev_io *bdev_io = _ctx;
7765 18 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
7766 : struct spdk_bdev_io *queued_reset;
7767 :
7768 18 : ch->flags &= ~BDEV_CH_RESET_IN_PROGRESS;
7769 18 : while (!TAILQ_EMPTY(&ch->queued_resets)) {
7770 0 : queued_reset = TAILQ_FIRST(&ch->queued_resets);
7771 0 : TAILQ_REMOVE(&ch->queued_resets, queued_reset, internal.link);
7772 0 : spdk_bdev_io_complete(queued_reset, bdev_io->internal.status);
7773 : }
7774 :
7775 18 : spdk_bdev_for_each_channel_continue(i, 0);
7776 18 : }
7777 :
7778 : static void
7779 0 : bdev_io_complete_sequence_cb(void *ctx, int status)
7780 : {
7781 0 : struct spdk_bdev_io *bdev_io = ctx;
7782 :
7783 : /* u.bdev.accel_sequence should have already been cleared at this point */
7784 0 : assert(bdev_io->u.bdev.accel_sequence == NULL);
7785 0 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS);
7786 0 : bdev_io->internal.f.has_accel_sequence = false;
7787 :
7788 0 : if (spdk_unlikely(status != 0)) {
7789 0 : SPDK_ERRLOG("Failed to execute accel sequence, status=%d\n", status);
7790 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
7791 0 : }
7792 :
7793 0 : bdev_io_complete(bdev_io);
7794 0 : }
7795 :
7796 : void
7797 598 : spdk_bdev_io_complete(struct spdk_bdev_io *bdev_io, enum spdk_bdev_io_status status)
7798 : {
7799 598 : struct spdk_bdev *bdev = bdev_io->bdev;
7800 598 : struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
7801 598 : struct spdk_bdev_shared_resource *shared_resource = bdev_ch->shared_resource;
7802 :
7803 598 : if (spdk_unlikely(bdev_io->internal.status != SPDK_BDEV_IO_STATUS_PENDING)) {
7804 0 : SPDK_ERRLOG("Unexpected completion on IO from %s module, status was %s\n",
7805 : spdk_bdev_get_module_name(bdev),
7806 : bdev_io_status_get_string(bdev_io->internal.status));
7807 0 : assert(false);
7808 : }
7809 598 : bdev_io->internal.status = status;
7810 :
7811 598 : if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_RESET)) {
7812 16 : bool unlock_channels = false;
7813 :
7814 16 : if (status == SPDK_BDEV_IO_STATUS_NOMEM) {
7815 0 : SPDK_ERRLOG("NOMEM returned for reset\n");
7816 0 : }
7817 16 : spdk_spin_lock(&bdev->internal.spinlock);
7818 16 : if (bdev_io == bdev->internal.reset_in_progress) {
7819 15 : bdev->internal.reset_in_progress = NULL;
7820 15 : unlock_channels = true;
7821 15 : }
7822 16 : spdk_spin_unlock(&bdev->internal.spinlock);
7823 :
7824 16 : if (unlock_channels) {
7825 15 : spdk_bdev_for_each_channel(bdev, bdev_unfreeze_channel, bdev_io,
7826 : bdev_reset_complete);
7827 15 : return;
7828 : }
7829 1 : } else {
7830 582 : bdev_io_decrement_outstanding(bdev_ch, shared_resource);
7831 582 : if (spdk_likely(status == SPDK_BDEV_IO_STATUS_SUCCESS)) {
7832 485 : if (bdev_io_needs_sequence_exec(bdev_io->internal.desc, bdev_io)) {
7833 0 : bdev_io_exec_sequence(bdev_io, bdev_io_complete_sequence_cb);
7834 0 : return;
7835 485 : } else if (spdk_unlikely(bdev_io->internal.f.has_bounce_buf &&
7836 : !bdev_io_use_accel_sequence(bdev_io))) {
7837 26 : _bdev_io_push_bounce_data_buffer(bdev_io,
7838 : _bdev_io_complete_push_bounce_done);
7839 : /* bdev IO will be completed in the callback */
7840 26 : return;
7841 : }
7842 459 : }
7843 :
7844 556 : if (spdk_unlikely(_bdev_io_handle_no_mem(bdev_io, BDEV_IO_RETRY_STATE_SUBMIT))) {
7845 5 : return;
7846 : }
7847 : }
7848 :
7849 552 : bdev_io_complete(bdev_io);
7850 598 : }
7851 :
7852 : void
7853 0 : spdk_bdev_io_complete_scsi_status(struct spdk_bdev_io *bdev_io, enum spdk_scsi_status sc,
7854 : enum spdk_scsi_sense sk, uint8_t asc, uint8_t ascq)
7855 : {
7856 : enum spdk_bdev_io_status status;
7857 :
7858 0 : if (sc == SPDK_SCSI_STATUS_GOOD) {
7859 0 : status = SPDK_BDEV_IO_STATUS_SUCCESS;
7860 0 : } else {
7861 0 : status = SPDK_BDEV_IO_STATUS_SCSI_ERROR;
7862 0 : bdev_io->internal.error.scsi.sc = sc;
7863 0 : bdev_io->internal.error.scsi.sk = sk;
7864 0 : bdev_io->internal.error.scsi.asc = asc;
7865 0 : bdev_io->internal.error.scsi.ascq = ascq;
7866 : }
7867 :
7868 0 : spdk_bdev_io_complete(bdev_io, status);
7869 0 : }
7870 :
7871 : void
7872 0 : spdk_bdev_io_get_scsi_status(const struct spdk_bdev_io *bdev_io,
7873 : int *sc, int *sk, int *asc, int *ascq)
7874 : {
7875 0 : assert(sc != NULL);
7876 0 : assert(sk != NULL);
7877 0 : assert(asc != NULL);
7878 0 : assert(ascq != NULL);
7879 :
7880 0 : switch (bdev_io->internal.status) {
7881 : case SPDK_BDEV_IO_STATUS_SUCCESS:
7882 0 : *sc = SPDK_SCSI_STATUS_GOOD;
7883 0 : *sk = SPDK_SCSI_SENSE_NO_SENSE;
7884 0 : *asc = SPDK_SCSI_ASC_NO_ADDITIONAL_SENSE;
7885 0 : *ascq = SPDK_SCSI_ASCQ_CAUSE_NOT_REPORTABLE;
7886 0 : break;
7887 : case SPDK_BDEV_IO_STATUS_NVME_ERROR:
7888 0 : spdk_scsi_nvme_translate(bdev_io, sc, sk, asc, ascq);
7889 0 : break;
7890 : case SPDK_BDEV_IO_STATUS_MISCOMPARE:
7891 0 : *sc = SPDK_SCSI_STATUS_CHECK_CONDITION;
7892 0 : *sk = SPDK_SCSI_SENSE_MISCOMPARE;
7893 0 : *asc = SPDK_SCSI_ASC_MISCOMPARE_DURING_VERIFY_OPERATION;
7894 0 : *ascq = bdev_io->internal.error.scsi.ascq;
7895 0 : break;
7896 : case SPDK_BDEV_IO_STATUS_SCSI_ERROR:
7897 0 : *sc = bdev_io->internal.error.scsi.sc;
7898 0 : *sk = bdev_io->internal.error.scsi.sk;
7899 0 : *asc = bdev_io->internal.error.scsi.asc;
7900 0 : *ascq = bdev_io->internal.error.scsi.ascq;
7901 0 : break;
7902 : default:
7903 0 : *sc = SPDK_SCSI_STATUS_CHECK_CONDITION;
7904 0 : *sk = SPDK_SCSI_SENSE_ABORTED_COMMAND;
7905 0 : *asc = SPDK_SCSI_ASC_NO_ADDITIONAL_SENSE;
7906 0 : *ascq = SPDK_SCSI_ASCQ_CAUSE_NOT_REPORTABLE;
7907 0 : break;
7908 : }
7909 0 : }
7910 :
7911 : void
7912 0 : spdk_bdev_io_complete_aio_status(struct spdk_bdev_io *bdev_io, int aio_result)
7913 : {
7914 : enum spdk_bdev_io_status status;
7915 :
7916 0 : if (aio_result == 0) {
7917 0 : status = SPDK_BDEV_IO_STATUS_SUCCESS;
7918 0 : } else {
7919 0 : status = SPDK_BDEV_IO_STATUS_AIO_ERROR;
7920 : }
7921 :
7922 0 : bdev_io->internal.error.aio_result = aio_result;
7923 :
7924 0 : spdk_bdev_io_complete(bdev_io, status);
7925 0 : }
7926 :
7927 : void
7928 0 : spdk_bdev_io_get_aio_status(const struct spdk_bdev_io *bdev_io, int *aio_result)
7929 : {
7930 0 : assert(aio_result != NULL);
7931 :
7932 0 : if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_AIO_ERROR) {
7933 0 : *aio_result = bdev_io->internal.error.aio_result;
7934 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS) {
7935 0 : *aio_result = 0;
7936 0 : } else {
7937 0 : *aio_result = -EIO;
7938 : }
7939 0 : }
7940 :
7941 : void
7942 0 : spdk_bdev_io_complete_nvme_status(struct spdk_bdev_io *bdev_io, uint32_t cdw0, int sct, int sc)
7943 : {
7944 : enum spdk_bdev_io_status status;
7945 :
7946 0 : if (spdk_likely(sct == SPDK_NVME_SCT_GENERIC && sc == SPDK_NVME_SC_SUCCESS)) {
7947 0 : status = SPDK_BDEV_IO_STATUS_SUCCESS;
7948 0 : } else if (sct == SPDK_NVME_SCT_GENERIC && sc == SPDK_NVME_SC_ABORTED_BY_REQUEST) {
7949 0 : status = SPDK_BDEV_IO_STATUS_ABORTED;
7950 0 : } else {
7951 0 : status = SPDK_BDEV_IO_STATUS_NVME_ERROR;
7952 : }
7953 :
7954 0 : bdev_io->internal.error.nvme.cdw0 = cdw0;
7955 0 : bdev_io->internal.error.nvme.sct = sct;
7956 0 : bdev_io->internal.error.nvme.sc = sc;
7957 :
7958 0 : spdk_bdev_io_complete(bdev_io, status);
7959 0 : }
7960 :
7961 : void
7962 0 : spdk_bdev_io_get_nvme_status(const struct spdk_bdev_io *bdev_io, uint32_t *cdw0, int *sct, int *sc)
7963 : {
7964 0 : assert(sct != NULL);
7965 0 : assert(sc != NULL);
7966 0 : assert(cdw0 != NULL);
7967 :
7968 0 : if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_ABORT)) {
7969 0 : *sct = SPDK_NVME_SCT_GENERIC;
7970 0 : *sc = SPDK_NVME_SC_SUCCESS;
7971 0 : if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS) {
7972 0 : *cdw0 = 0;
7973 0 : } else {
7974 0 : *cdw0 = 1U;
7975 : }
7976 0 : return;
7977 : }
7978 :
7979 0 : if (spdk_likely(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS)) {
7980 0 : *sct = SPDK_NVME_SCT_GENERIC;
7981 0 : *sc = SPDK_NVME_SC_SUCCESS;
7982 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_NVME_ERROR) {
7983 0 : *sct = bdev_io->internal.error.nvme.sct;
7984 0 : *sc = bdev_io->internal.error.nvme.sc;
7985 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_ABORTED) {
7986 0 : *sct = SPDK_NVME_SCT_GENERIC;
7987 0 : *sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
7988 0 : } else {
7989 0 : *sct = SPDK_NVME_SCT_GENERIC;
7990 0 : *sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
7991 : }
7992 :
7993 0 : *cdw0 = bdev_io->internal.error.nvme.cdw0;
7994 0 : }
7995 :
7996 : void
7997 0 : spdk_bdev_io_get_nvme_fused_status(const struct spdk_bdev_io *bdev_io, uint32_t *cdw0,
7998 : int *first_sct, int *first_sc, int *second_sct, int *second_sc)
7999 : {
8000 0 : assert(first_sct != NULL);
8001 0 : assert(first_sc != NULL);
8002 0 : assert(second_sct != NULL);
8003 0 : assert(second_sc != NULL);
8004 0 : assert(cdw0 != NULL);
8005 :
8006 0 : if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_NVME_ERROR) {
8007 0 : if (bdev_io->internal.error.nvme.sct == SPDK_NVME_SCT_MEDIA_ERROR &&
8008 0 : bdev_io->internal.error.nvme.sc == SPDK_NVME_SC_COMPARE_FAILURE) {
8009 0 : *first_sct = bdev_io->internal.error.nvme.sct;
8010 0 : *first_sc = bdev_io->internal.error.nvme.sc;
8011 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
8012 0 : *second_sc = SPDK_NVME_SC_ABORTED_FAILED_FUSED;
8013 0 : } else {
8014 0 : *first_sct = SPDK_NVME_SCT_GENERIC;
8015 0 : *first_sc = SPDK_NVME_SC_SUCCESS;
8016 0 : *second_sct = bdev_io->internal.error.nvme.sct;
8017 0 : *second_sc = bdev_io->internal.error.nvme.sc;
8018 : }
8019 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_ABORTED) {
8020 0 : *first_sct = SPDK_NVME_SCT_GENERIC;
8021 0 : *first_sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
8022 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
8023 0 : *second_sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
8024 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS) {
8025 0 : *first_sct = SPDK_NVME_SCT_GENERIC;
8026 0 : *first_sc = SPDK_NVME_SC_SUCCESS;
8027 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
8028 0 : *second_sc = SPDK_NVME_SC_SUCCESS;
8029 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_FIRST_FUSED_FAILED) {
8030 0 : *first_sct = SPDK_NVME_SCT_GENERIC;
8031 0 : *first_sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
8032 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
8033 0 : *second_sc = SPDK_NVME_SC_ABORTED_FAILED_FUSED;
8034 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_MISCOMPARE) {
8035 0 : *first_sct = SPDK_NVME_SCT_MEDIA_ERROR;
8036 0 : *first_sc = SPDK_NVME_SC_COMPARE_FAILURE;
8037 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
8038 0 : *second_sc = SPDK_NVME_SC_ABORTED_FAILED_FUSED;
8039 0 : } else {
8040 0 : *first_sct = SPDK_NVME_SCT_GENERIC;
8041 0 : *first_sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
8042 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
8043 0 : *second_sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
8044 : }
8045 :
8046 0 : *cdw0 = bdev_io->internal.error.nvme.cdw0;
8047 0 : }
8048 :
8049 : void
8050 0 : spdk_bdev_io_complete_base_io_status(struct spdk_bdev_io *bdev_io,
8051 : const struct spdk_bdev_io *base_io)
8052 : {
8053 0 : switch (base_io->internal.status) {
8054 : case SPDK_BDEV_IO_STATUS_NVME_ERROR:
8055 0 : spdk_bdev_io_complete_nvme_status(bdev_io,
8056 0 : base_io->internal.error.nvme.cdw0,
8057 0 : base_io->internal.error.nvme.sct,
8058 0 : base_io->internal.error.nvme.sc);
8059 0 : break;
8060 : case SPDK_BDEV_IO_STATUS_SCSI_ERROR:
8061 0 : spdk_bdev_io_complete_scsi_status(bdev_io,
8062 0 : base_io->internal.error.scsi.sc,
8063 0 : base_io->internal.error.scsi.sk,
8064 0 : base_io->internal.error.scsi.asc,
8065 0 : base_io->internal.error.scsi.ascq);
8066 0 : break;
8067 : case SPDK_BDEV_IO_STATUS_AIO_ERROR:
8068 0 : spdk_bdev_io_complete_aio_status(bdev_io, base_io->internal.error.aio_result);
8069 0 : break;
8070 : default:
8071 0 : spdk_bdev_io_complete(bdev_io, base_io->internal.status);
8072 0 : break;
8073 : }
8074 0 : }
8075 :
8076 : struct spdk_thread *
8077 660 : spdk_bdev_io_get_thread(struct spdk_bdev_io *bdev_io)
8078 : {
8079 660 : return spdk_io_channel_get_thread(bdev_io->internal.ch->channel);
8080 : }
8081 :
8082 : struct spdk_io_channel *
8083 70 : spdk_bdev_io_get_io_channel(struct spdk_bdev_io *bdev_io)
8084 : {
8085 70 : return bdev_io->internal.ch->channel;
8086 : }
8087 :
8088 : static int
8089 128 : bdev_register(struct spdk_bdev *bdev)
8090 : {
8091 : char *bdev_name;
8092 : char uuid[SPDK_UUID_STRING_LEN];
8093 : struct spdk_iobuf_opts iobuf_opts;
8094 : int ret;
8095 :
8096 128 : assert(bdev->module != NULL);
8097 :
8098 128 : if (!bdev->name) {
8099 0 : SPDK_ERRLOG("Bdev name is NULL\n");
8100 0 : return -EINVAL;
8101 : }
8102 :
8103 128 : if (!strlen(bdev->name)) {
8104 0 : SPDK_ERRLOG("Bdev name must not be an empty string\n");
8105 0 : return -EINVAL;
8106 : }
8107 :
8108 : /* Users often register their own I/O devices using the bdev name. In
8109 : * order to avoid conflicts, prepend bdev_. */
8110 128 : bdev_name = spdk_sprintf_alloc("bdev_%s", bdev->name);
8111 128 : if (!bdev_name) {
8112 0 : SPDK_ERRLOG("Unable to allocate memory for internal bdev name.\n");
8113 0 : return -ENOMEM;
8114 : }
8115 :
8116 128 : bdev->internal.stat = bdev_alloc_io_stat(true);
8117 128 : if (!bdev->internal.stat) {
8118 0 : SPDK_ERRLOG("Unable to allocate I/O statistics structure.\n");
8119 0 : free(bdev_name);
8120 0 : return -ENOMEM;
8121 : }
8122 :
8123 128 : bdev->internal.status = SPDK_BDEV_STATUS_READY;
8124 128 : bdev->internal.measured_queue_depth = UINT64_MAX;
8125 128 : bdev->internal.claim_type = SPDK_BDEV_CLAIM_NONE;
8126 128 : memset(&bdev->internal.claim, 0, sizeof(bdev->internal.claim));
8127 128 : bdev->internal.qd_poller = NULL;
8128 128 : bdev->internal.qos = NULL;
8129 :
8130 128 : TAILQ_INIT(&bdev->internal.open_descs);
8131 128 : TAILQ_INIT(&bdev->internal.locked_ranges);
8132 128 : TAILQ_INIT(&bdev->internal.pending_locked_ranges);
8133 128 : TAILQ_INIT(&bdev->aliases);
8134 :
8135 : /* UUID may be specified by the user or defined by bdev itself.
8136 : * Otherwise it will be generated here, so this field will never be empty. */
8137 128 : if (spdk_uuid_is_null(&bdev->uuid)) {
8138 42 : spdk_uuid_generate(&bdev->uuid);
8139 42 : }
8140 :
8141 : /* Add the UUID alias only if it's different than the name */
8142 128 : spdk_uuid_fmt_lower(uuid, sizeof(uuid), &bdev->uuid);
8143 128 : if (strcmp(bdev->name, uuid) != 0) {
8144 127 : ret = spdk_bdev_alias_add(bdev, uuid);
8145 127 : if (ret != 0) {
8146 2 : SPDK_ERRLOG("Unable to add uuid:%s alias for bdev %s\n", uuid, bdev->name);
8147 2 : bdev_free_io_stat(bdev->internal.stat);
8148 2 : free(bdev_name);
8149 2 : return ret;
8150 : }
8151 125 : }
8152 :
8153 126 : spdk_iobuf_get_opts(&iobuf_opts, sizeof(iobuf_opts));
8154 126 : if (spdk_bdev_get_buf_align(bdev) > 1) {
8155 0 : bdev->max_rw_size = spdk_min(bdev->max_rw_size ? bdev->max_rw_size : UINT32_MAX,
8156 : iobuf_opts.large_bufsize / bdev->blocklen);
8157 0 : }
8158 :
8159 : /* If the user didn't specify a write unit size, set it to one. */
8160 126 : if (bdev->write_unit_size == 0) {
8161 122 : bdev->write_unit_size = 1;
8162 122 : }
8163 :
8164 : /* Set ACWU value to the write unit size if bdev module did not set it (does not support it natively) */
8165 126 : if (bdev->acwu == 0) {
8166 122 : bdev->acwu = bdev->write_unit_size;
8167 122 : }
8168 :
8169 126 : if (bdev->phys_blocklen == 0) {
8170 122 : bdev->phys_blocklen = spdk_bdev_get_data_block_size(bdev);
8171 122 : }
8172 :
8173 126 : if (!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COPY)) {
8174 0 : bdev->max_copy = bdev_get_max_write(bdev, iobuf_opts.large_bufsize);
8175 0 : }
8176 :
8177 126 : if (!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE_ZEROES)) {
8178 0 : bdev->max_write_zeroes = bdev_get_max_write(bdev, ZERO_BUFFER_SIZE);
8179 0 : }
8180 :
8181 126 : bdev->internal.reset_in_progress = NULL;
8182 126 : bdev->internal.qd_poll_in_progress = false;
8183 126 : bdev->internal.period = 0;
8184 126 : bdev->internal.new_period = 0;
8185 126 : bdev->internal.trace_id = spdk_trace_register_owner(OWNER_TYPE_BDEV, bdev_name);
8186 :
8187 : /*
8188 : * Initialize spinlock before registering IO device because spinlock is used in
8189 : * bdev_channel_create
8190 : */
8191 126 : spdk_spin_init(&bdev->internal.spinlock);
8192 :
8193 252 : spdk_io_device_register(__bdev_to_io_dev(bdev),
8194 : bdev_channel_create, bdev_channel_destroy,
8195 : sizeof(struct spdk_bdev_channel),
8196 126 : bdev_name);
8197 :
8198 : /*
8199 : * Register bdev name only after the bdev object is ready.
8200 : * After bdev_name_add returns, it is possible for other threads to start using the bdev,
8201 : * create IO channels...
8202 : */
8203 126 : ret = bdev_name_add(&bdev->internal.bdev_name, bdev, bdev->name);
8204 126 : if (ret != 0) {
8205 0 : spdk_io_device_unregister(__bdev_to_io_dev(bdev), NULL);
8206 0 : bdev_free_io_stat(bdev->internal.stat);
8207 0 : spdk_spin_destroy(&bdev->internal.spinlock);
8208 0 : free(bdev_name);
8209 0 : return ret;
8210 : }
8211 :
8212 126 : free(bdev_name);
8213 :
8214 126 : SPDK_DEBUGLOG(bdev, "Inserting bdev %s into list\n", bdev->name);
8215 126 : TAILQ_INSERT_TAIL(&g_bdev_mgr.bdevs, bdev, internal.link);
8216 :
8217 126 : return 0;
8218 128 : }
8219 :
8220 : static void
8221 127 : bdev_destroy_cb(void *io_device)
8222 : {
8223 : int rc;
8224 : struct spdk_bdev *bdev;
8225 : spdk_bdev_unregister_cb cb_fn;
8226 : void *cb_arg;
8227 :
8228 127 : bdev = __bdev_from_io_dev(io_device);
8229 :
8230 127 : if (bdev->internal.unregister_td != spdk_get_thread()) {
8231 1 : spdk_thread_send_msg(bdev->internal.unregister_td, bdev_destroy_cb, io_device);
8232 1 : return;
8233 : }
8234 :
8235 126 : cb_fn = bdev->internal.unregister_cb;
8236 126 : cb_arg = bdev->internal.unregister_ctx;
8237 :
8238 126 : spdk_spin_destroy(&bdev->internal.spinlock);
8239 126 : free(bdev->internal.qos);
8240 126 : bdev_free_io_stat(bdev->internal.stat);
8241 126 : spdk_trace_unregister_owner(bdev->internal.trace_id);
8242 :
8243 126 : rc = bdev->fn_table->destruct(bdev->ctxt);
8244 126 : if (rc < 0) {
8245 0 : SPDK_ERRLOG("destruct failed\n");
8246 0 : }
8247 126 : if (rc <= 0 && cb_fn != NULL) {
8248 10 : cb_fn(cb_arg, rc);
8249 10 : }
8250 127 : }
8251 :
8252 : void
8253 2 : spdk_bdev_destruct_done(struct spdk_bdev *bdev, int bdeverrno)
8254 : {
8255 2 : if (bdev->internal.unregister_cb != NULL) {
8256 0 : bdev->internal.unregister_cb(bdev->internal.unregister_ctx, bdeverrno);
8257 0 : }
8258 2 : }
8259 :
8260 : static void
8261 19 : _remove_notify(void *arg)
8262 : {
8263 19 : struct spdk_bdev_desc *desc = arg;
8264 :
8265 19 : _event_notify(desc, SPDK_BDEV_EVENT_REMOVE);
8266 19 : }
8267 :
8268 : /* returns: 0 - bdev removed and ready to be destructed.
8269 : * -EBUSY - bdev can't be destructed yet. */
8270 : static int
8271 141 : bdev_unregister_unsafe(struct spdk_bdev *bdev)
8272 : {
8273 : struct spdk_bdev_desc *desc, *tmp;
8274 141 : int rc = 0;
8275 : char uuid[SPDK_UUID_STRING_LEN];
8276 :
8277 141 : assert(spdk_spin_held(&g_bdev_mgr.spinlock));
8278 141 : assert(spdk_spin_held(&bdev->internal.spinlock));
8279 :
8280 : /* Notify each descriptor about hotremoval */
8281 160 : TAILQ_FOREACH_SAFE(desc, &bdev->internal.open_descs, link, tmp) {
8282 19 : rc = -EBUSY;
8283 : /*
8284 : * Defer invocation of the event_cb to a separate message that will
8285 : * run later on its thread. This ensures this context unwinds and
8286 : * we don't recursively unregister this bdev again if the event_cb
8287 : * immediately closes its descriptor.
8288 : */
8289 19 : event_notify(desc, _remove_notify);
8290 19 : }
8291 :
8292 : /* If there are no descriptors, proceed removing the bdev */
8293 141 : if (rc == 0) {
8294 126 : TAILQ_REMOVE(&g_bdev_mgr.bdevs, bdev, internal.link);
8295 126 : SPDK_DEBUGLOG(bdev, "Removing bdev %s from list done\n", bdev->name);
8296 :
8297 : /* Delete the name and the UUID alias */
8298 126 : spdk_uuid_fmt_lower(uuid, sizeof(uuid), &bdev->uuid);
8299 126 : bdev_name_del_unsafe(&bdev->internal.bdev_name);
8300 126 : bdev_alias_del(bdev, uuid, bdev_name_del_unsafe);
8301 :
8302 126 : spdk_notify_send("bdev_unregister", spdk_bdev_get_name(bdev));
8303 :
8304 126 : if (bdev->internal.reset_in_progress != NULL) {
8305 : /* If reset is in progress, let the completion callback for reset
8306 : * unregister the bdev.
8307 : */
8308 1 : rc = -EBUSY;
8309 1 : }
8310 126 : }
8311 :
8312 141 : return rc;
8313 : }
8314 :
8315 : static void
8316 4 : bdev_unregister_abort_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
8317 : struct spdk_io_channel *io_ch, void *_ctx)
8318 : {
8319 4 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
8320 :
8321 4 : bdev_channel_abort_queued_ios(bdev_ch);
8322 4 : spdk_bdev_for_each_channel_continue(i, 0);
8323 4 : }
8324 :
8325 : static void
8326 126 : bdev_unregister(struct spdk_bdev *bdev, void *_ctx, int status)
8327 : {
8328 : int rc;
8329 :
8330 126 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8331 126 : spdk_spin_lock(&bdev->internal.spinlock);
8332 : /*
8333 : * Set the status to REMOVING after completing to abort channels. Otherwise,
8334 : * the last spdk_bdev_close() may call spdk_io_device_unregister() while
8335 : * spdk_bdev_for_each_channel() is executed and spdk_io_device_unregister()
8336 : * may fail.
8337 : */
8338 126 : bdev->internal.status = SPDK_BDEV_STATUS_REMOVING;
8339 126 : rc = bdev_unregister_unsafe(bdev);
8340 126 : spdk_spin_unlock(&bdev->internal.spinlock);
8341 126 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8342 :
8343 126 : if (rc == 0) {
8344 110 : spdk_io_device_unregister(__bdev_to_io_dev(bdev), bdev_destroy_cb);
8345 110 : }
8346 126 : }
8347 :
8348 : void
8349 133 : spdk_bdev_unregister(struct spdk_bdev *bdev, spdk_bdev_unregister_cb cb_fn, void *cb_arg)
8350 : {
8351 : struct spdk_thread *thread;
8352 :
8353 133 : SPDK_DEBUGLOG(bdev, "Removing bdev %s from list\n", bdev->name);
8354 :
8355 133 : thread = spdk_get_thread();
8356 133 : if (!thread) {
8357 : /* The user called this from a non-SPDK thread. */
8358 0 : if (cb_fn != NULL) {
8359 0 : cb_fn(cb_arg, -ENOTSUP);
8360 0 : }
8361 0 : return;
8362 : }
8363 :
8364 133 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8365 133 : if (bdev->internal.status == SPDK_BDEV_STATUS_UNREGISTERING ||
8366 133 : bdev->internal.status == SPDK_BDEV_STATUS_REMOVING) {
8367 7 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8368 7 : if (cb_fn) {
8369 0 : cb_fn(cb_arg, -EBUSY);
8370 0 : }
8371 7 : return;
8372 : }
8373 :
8374 126 : spdk_spin_lock(&bdev->internal.spinlock);
8375 126 : bdev->internal.status = SPDK_BDEV_STATUS_UNREGISTERING;
8376 126 : bdev->internal.unregister_cb = cb_fn;
8377 126 : bdev->internal.unregister_ctx = cb_arg;
8378 126 : bdev->internal.unregister_td = thread;
8379 126 : spdk_spin_unlock(&bdev->internal.spinlock);
8380 126 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8381 :
8382 126 : spdk_bdev_set_qd_sampling_period(bdev, 0);
8383 :
8384 126 : spdk_bdev_for_each_channel(bdev, bdev_unregister_abort_channel, bdev,
8385 : bdev_unregister);
8386 133 : }
8387 :
8388 : int
8389 4 : spdk_bdev_unregister_by_name(const char *bdev_name, struct spdk_bdev_module *module,
8390 : spdk_bdev_unregister_cb cb_fn, void *cb_arg)
8391 : {
8392 : struct spdk_bdev_desc *desc;
8393 : struct spdk_bdev *bdev;
8394 : int rc;
8395 :
8396 4 : rc = spdk_bdev_open_ext(bdev_name, false, _tmp_bdev_event_cb, NULL, &desc);
8397 4 : if (rc != 0) {
8398 1 : SPDK_ERRLOG("Failed to open bdev with name: %s\n", bdev_name);
8399 1 : return rc;
8400 : }
8401 :
8402 3 : bdev = spdk_bdev_desc_get_bdev(desc);
8403 :
8404 3 : if (bdev->module != module) {
8405 1 : spdk_bdev_close(desc);
8406 1 : SPDK_ERRLOG("Bdev %s was not registered by the specified module.\n",
8407 : bdev_name);
8408 1 : return -ENODEV;
8409 : }
8410 :
8411 2 : spdk_bdev_unregister(bdev, cb_fn, cb_arg);
8412 :
8413 2 : spdk_bdev_close(desc);
8414 :
8415 2 : return 0;
8416 4 : }
8417 :
8418 : static int
8419 259 : bdev_start_qos(struct spdk_bdev *bdev)
8420 : {
8421 : struct set_qos_limit_ctx *ctx;
8422 :
8423 : /* Enable QoS */
8424 259 : if (bdev->internal.qos && bdev->internal.qos->thread == NULL) {
8425 2 : ctx = calloc(1, sizeof(*ctx));
8426 2 : if (ctx == NULL) {
8427 0 : SPDK_ERRLOG("Failed to allocate memory for QoS context\n");
8428 0 : return -ENOMEM;
8429 : }
8430 2 : ctx->bdev = bdev;
8431 2 : spdk_bdev_for_each_channel(bdev, bdev_enable_qos_msg, ctx, bdev_enable_qos_done);
8432 2 : }
8433 :
8434 259 : return 0;
8435 259 : }
8436 :
8437 : static void
8438 24 : log_already_claimed(enum spdk_log_level level, const int line, const char *func, const char *detail,
8439 : struct spdk_bdev *bdev)
8440 : {
8441 : enum spdk_bdev_claim_type type;
8442 : const char *typename, *modname;
8443 : extern struct spdk_log_flag SPDK_LOG_bdev;
8444 :
8445 24 : assert(spdk_spin_held(&bdev->internal.spinlock));
8446 :
8447 24 : if (level >= SPDK_LOG_INFO && !SPDK_LOG_bdev.enabled) {
8448 0 : return;
8449 : }
8450 :
8451 24 : type = bdev->internal.claim_type;
8452 24 : typename = spdk_bdev_claim_get_name(type);
8453 :
8454 24 : if (type == SPDK_BDEV_CLAIM_EXCL_WRITE) {
8455 6 : modname = bdev->internal.claim.v1.module->name;
8456 12 : spdk_log(level, __FILE__, line, func, "bdev %s %s: type %s by module %s\n",
8457 6 : bdev->name, detail, typename, modname);
8458 6 : return;
8459 : }
8460 :
8461 18 : if (claim_type_is_v2(type)) {
8462 : struct spdk_bdev_module_claim *claim;
8463 :
8464 36 : TAILQ_FOREACH(claim, &bdev->internal.claim.v2.claims, link) {
8465 18 : modname = claim->module->name;
8466 36 : spdk_log(level, __FILE__, line, func, "bdev %s %s: type %s by module %s\n",
8467 18 : bdev->name, detail, typename, modname);
8468 18 : }
8469 18 : return;
8470 : }
8471 :
8472 0 : assert(false);
8473 24 : }
8474 :
8475 : static int
8476 268 : bdev_open(struct spdk_bdev *bdev, bool write, struct spdk_bdev_desc *desc)
8477 : {
8478 : struct spdk_thread *thread;
8479 268 : int rc = 0;
8480 :
8481 268 : thread = spdk_get_thread();
8482 268 : if (!thread) {
8483 0 : SPDK_ERRLOG("Cannot open bdev from non-SPDK thread.\n");
8484 0 : return -ENOTSUP;
8485 : }
8486 :
8487 268 : SPDK_DEBUGLOG(bdev, "Opening descriptor %p for bdev %s on thread %p\n", desc, bdev->name,
8488 : spdk_get_thread());
8489 :
8490 268 : desc->bdev = bdev;
8491 268 : desc->thread = thread;
8492 268 : desc->write = write;
8493 :
8494 268 : spdk_spin_lock(&bdev->internal.spinlock);
8495 268 : if (bdev->internal.status == SPDK_BDEV_STATUS_UNREGISTERING ||
8496 268 : bdev->internal.status == SPDK_BDEV_STATUS_REMOVING) {
8497 3 : spdk_spin_unlock(&bdev->internal.spinlock);
8498 3 : return -ENODEV;
8499 : }
8500 :
8501 265 : if (write && bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE) {
8502 6 : LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
8503 6 : spdk_spin_unlock(&bdev->internal.spinlock);
8504 6 : return -EPERM;
8505 : }
8506 :
8507 259 : rc = bdev_start_qos(bdev);
8508 259 : if (rc != 0) {
8509 0 : SPDK_ERRLOG("Failed to start QoS on bdev %s\n", bdev->name);
8510 0 : spdk_spin_unlock(&bdev->internal.spinlock);
8511 0 : return rc;
8512 : }
8513 :
8514 259 : TAILQ_INSERT_TAIL(&bdev->internal.open_descs, desc, link);
8515 :
8516 259 : spdk_spin_unlock(&bdev->internal.spinlock);
8517 :
8518 259 : return 0;
8519 268 : }
8520 :
8521 : static void
8522 268 : bdev_open_opts_get_defaults(struct spdk_bdev_open_opts *opts, size_t opts_size)
8523 : {
8524 268 : if (!opts) {
8525 0 : SPDK_ERRLOG("opts should not be NULL.\n");
8526 0 : return;
8527 : }
8528 :
8529 268 : if (!opts_size) {
8530 0 : SPDK_ERRLOG("opts_size should not be zero.\n");
8531 0 : return;
8532 : }
8533 :
8534 268 : memset(opts, 0, opts_size);
8535 268 : opts->size = opts_size;
8536 :
8537 : #define FIELD_OK(field) \
8538 : offsetof(struct spdk_bdev_open_opts, field) + sizeof(opts->field) <= opts_size
8539 :
8540 : #define SET_FIELD(field, value) \
8541 : if (FIELD_OK(field)) { \
8542 : opts->field = value; \
8543 : } \
8544 :
8545 268 : SET_FIELD(no_metadata, false);
8546 :
8547 : #undef FIELD_OK
8548 : #undef SET_FIELD
8549 268 : }
8550 :
8551 : static void
8552 1 : bdev_open_opts_copy(struct spdk_bdev_open_opts *opts,
8553 : const struct spdk_bdev_open_opts *user_opts)
8554 : {
8555 : #define FIELD_OK(field) \
8556 : offsetof(struct spdk_bdev_open_opts, field) + sizeof(opts->field) <= user_opts->size
8557 :
8558 : #define SET_FIELD(field) \
8559 : if (FIELD_OK(field)) { \
8560 : opts->field = user_opts->field; \
8561 : } \
8562 :
8563 1 : SET_FIELD(no_metadata);
8564 :
8565 1 : opts->size = user_opts->size;
8566 :
8567 : /* We should not remove this statement, but need to update the assert statement
8568 : * if we add a new field, and also add a corresponding SET_FIELD statement.
8569 : */
8570 : SPDK_STATIC_ASSERT(sizeof(struct spdk_bdev_open_opts) == 16, "Incorrect size");
8571 :
8572 : #undef FIELD_OK
8573 : #undef SET_FIELD
8574 1 : }
8575 :
8576 : static int
8577 268 : bdev_desc_alloc(struct spdk_bdev *bdev, spdk_bdev_event_cb_t event_cb, void *event_ctx,
8578 : struct spdk_bdev_open_opts *user_opts, struct spdk_bdev_desc **_desc)
8579 : {
8580 : struct spdk_bdev_desc *desc;
8581 : struct spdk_bdev_open_opts opts;
8582 : unsigned int i;
8583 :
8584 268 : bdev_open_opts_get_defaults(&opts, sizeof(opts));
8585 268 : if (user_opts != NULL) {
8586 1 : bdev_open_opts_copy(&opts, user_opts);
8587 1 : }
8588 :
8589 268 : desc = calloc(1, sizeof(*desc));
8590 268 : if (desc == NULL) {
8591 0 : SPDK_ERRLOG("Failed to allocate memory for bdev descriptor\n");
8592 0 : return -ENOMEM;
8593 : }
8594 :
8595 268 : desc->opts = opts;
8596 :
8597 268 : TAILQ_INIT(&desc->pending_media_events);
8598 268 : TAILQ_INIT(&desc->free_media_events);
8599 :
8600 268 : desc->memory_domains_supported = spdk_bdev_get_memory_domains(bdev, NULL, 0) > 0;
8601 268 : desc->callback.event_fn = event_cb;
8602 268 : desc->callback.ctx = event_ctx;
8603 268 : spdk_spin_init(&desc->spinlock);
8604 :
8605 268 : if (desc->opts.no_metadata) {
8606 1 : if (spdk_bdev_is_md_separate(bdev)) {
8607 0 : SPDK_ERRLOG("no_metadata option is not supported with separate metadata.\n");
8608 0 : bdev_desc_free(desc);
8609 0 : return -EINVAL;
8610 : }
8611 1 : }
8612 :
8613 268 : if (bdev->media_events) {
8614 0 : desc->media_events_buffer = calloc(MEDIA_EVENT_POOL_SIZE,
8615 : sizeof(*desc->media_events_buffer));
8616 0 : if (desc->media_events_buffer == NULL) {
8617 0 : SPDK_ERRLOG("Failed to initialize media event pool\n");
8618 0 : bdev_desc_free(desc);
8619 0 : return -ENOMEM;
8620 : }
8621 :
8622 0 : for (i = 0; i < MEDIA_EVENT_POOL_SIZE; ++i) {
8623 0 : TAILQ_INSERT_TAIL(&desc->free_media_events,
8624 : &desc->media_events_buffer[i], tailq);
8625 0 : }
8626 0 : }
8627 :
8628 268 : if (bdev->fn_table->accel_sequence_supported != NULL) {
8629 0 : for (i = 0; i < SPDK_BDEV_NUM_IO_TYPES; ++i) {
8630 0 : desc->accel_sequence_supported[i] =
8631 0 : bdev->fn_table->accel_sequence_supported(bdev->ctxt,
8632 0 : (enum spdk_bdev_io_type)i);
8633 0 : }
8634 0 : }
8635 :
8636 268 : *_desc = desc;
8637 :
8638 268 : return 0;
8639 268 : }
8640 :
8641 : static int
8642 130 : bdev_open_ext(const char *bdev_name, bool write, spdk_bdev_event_cb_t event_cb,
8643 : void *event_ctx, struct spdk_bdev_open_opts *opts,
8644 : struct spdk_bdev_desc **_desc)
8645 : {
8646 : struct spdk_bdev_desc *desc;
8647 : struct spdk_bdev *bdev;
8648 : int rc;
8649 :
8650 130 : bdev = bdev_get_by_name(bdev_name);
8651 :
8652 130 : if (bdev == NULL) {
8653 1 : SPDK_NOTICELOG("Currently unable to find bdev with name: %s\n", bdev_name);
8654 1 : return -ENODEV;
8655 : }
8656 :
8657 129 : rc = bdev_desc_alloc(bdev, event_cb, event_ctx, opts, &desc);
8658 129 : if (rc != 0) {
8659 0 : return rc;
8660 : }
8661 :
8662 129 : rc = bdev_open(bdev, write, desc);
8663 129 : if (rc != 0) {
8664 7 : bdev_desc_free(desc);
8665 7 : desc = NULL;
8666 7 : }
8667 :
8668 129 : *_desc = desc;
8669 :
8670 129 : return rc;
8671 130 : }
8672 :
8673 : int
8674 132 : spdk_bdev_open_ext_v2(const char *bdev_name, bool write, spdk_bdev_event_cb_t event_cb,
8675 : void *event_ctx, struct spdk_bdev_open_opts *opts,
8676 : struct spdk_bdev_desc **_desc)
8677 : {
8678 : int rc;
8679 :
8680 132 : if (event_cb == NULL) {
8681 2 : SPDK_ERRLOG("Missing event callback function\n");
8682 2 : return -EINVAL;
8683 : }
8684 :
8685 130 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8686 130 : rc = bdev_open_ext(bdev_name, write, event_cb, event_ctx, opts, _desc);
8687 130 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8688 :
8689 130 : return rc;
8690 132 : }
8691 :
8692 : int
8693 130 : spdk_bdev_open_ext(const char *bdev_name, bool write, spdk_bdev_event_cb_t event_cb,
8694 : void *event_ctx, struct spdk_bdev_desc **_desc)
8695 : {
8696 130 : return spdk_bdev_open_ext_v2(bdev_name, write, event_cb, event_ctx, NULL, _desc);
8697 : }
8698 :
8699 : struct spdk_bdev_open_async_ctx {
8700 : char *bdev_name;
8701 : spdk_bdev_event_cb_t event_cb;
8702 : void *event_ctx;
8703 : bool write;
8704 : int rc;
8705 : spdk_bdev_open_async_cb_t cb_fn;
8706 : void *cb_arg;
8707 : struct spdk_bdev_desc *desc;
8708 : struct spdk_bdev_open_async_opts opts;
8709 : uint64_t start_ticks;
8710 : struct spdk_thread *orig_thread;
8711 : struct spdk_poller *poller;
8712 : TAILQ_ENTRY(spdk_bdev_open_async_ctx) tailq;
8713 : };
8714 :
8715 : static void
8716 0 : bdev_open_async_done(void *arg)
8717 : {
8718 0 : struct spdk_bdev_open_async_ctx *ctx = arg;
8719 :
8720 0 : ctx->cb_fn(ctx->desc, ctx->rc, ctx->cb_arg);
8721 :
8722 0 : free(ctx->bdev_name);
8723 0 : free(ctx);
8724 0 : }
8725 :
8726 : static void
8727 0 : bdev_open_async_cancel(void *arg)
8728 : {
8729 0 : struct spdk_bdev_open_async_ctx *ctx = arg;
8730 :
8731 0 : assert(ctx->rc == -ESHUTDOWN);
8732 :
8733 0 : spdk_poller_unregister(&ctx->poller);
8734 :
8735 0 : bdev_open_async_done(ctx);
8736 0 : }
8737 :
8738 : /* This is called when the bdev library finishes at shutdown. */
8739 : static void
8740 65 : bdev_open_async_fini(void)
8741 : {
8742 : struct spdk_bdev_open_async_ctx *ctx, *tmp_ctx;
8743 :
8744 65 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8745 65 : TAILQ_FOREACH_SAFE(ctx, &g_bdev_mgr.async_bdev_opens, tailq, tmp_ctx) {
8746 0 : TAILQ_REMOVE(&g_bdev_mgr.async_bdev_opens, ctx, tailq);
8747 : /*
8748 : * We have to move to ctx->orig_thread to unregister ctx->poller.
8749 : * However, there is a chance that ctx->poller is executed before
8750 : * message is executed, which could result in bdev_open_async_done()
8751 : * being called twice. To avoid such race condition, set ctx->rc to
8752 : * -ESHUTDOWN.
8753 : */
8754 0 : ctx->rc = -ESHUTDOWN;
8755 0 : spdk_thread_send_msg(ctx->orig_thread, bdev_open_async_cancel, ctx);
8756 0 : }
8757 65 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8758 65 : }
8759 :
8760 : static int bdev_open_async(void *arg);
8761 :
8762 : static void
8763 0 : _bdev_open_async(struct spdk_bdev_open_async_ctx *ctx)
8764 : {
8765 : uint64_t timeout_ticks;
8766 :
8767 0 : if (ctx->rc == -ESHUTDOWN) {
8768 : /* This context is being canceled. Do nothing. */
8769 0 : return;
8770 : }
8771 :
8772 0 : ctx->rc = bdev_open_ext(ctx->bdev_name, ctx->write, ctx->event_cb, ctx->event_ctx,
8773 0 : NULL, &ctx->desc);
8774 0 : if (ctx->rc == 0 || ctx->opts.timeout_ms == 0) {
8775 0 : goto exit;
8776 : }
8777 :
8778 0 : timeout_ticks = ctx->start_ticks + ctx->opts.timeout_ms * spdk_get_ticks_hz() / 1000ull;
8779 0 : if (spdk_get_ticks() >= timeout_ticks) {
8780 0 : SPDK_ERRLOG("Timed out while waiting for bdev '%s' to appear\n", ctx->bdev_name);
8781 0 : ctx->rc = -ETIMEDOUT;
8782 0 : goto exit;
8783 : }
8784 :
8785 0 : return;
8786 :
8787 : exit:
8788 0 : spdk_poller_unregister(&ctx->poller);
8789 0 : TAILQ_REMOVE(&g_bdev_mgr.async_bdev_opens, ctx, tailq);
8790 :
8791 : /* Completion callback is processed after stack unwinding. */
8792 0 : spdk_thread_send_msg(ctx->orig_thread, bdev_open_async_done, ctx);
8793 0 : }
8794 :
8795 : static int
8796 0 : bdev_open_async(void *arg)
8797 : {
8798 0 : struct spdk_bdev_open_async_ctx *ctx = arg;
8799 :
8800 0 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8801 :
8802 0 : _bdev_open_async(ctx);
8803 :
8804 0 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8805 :
8806 0 : return SPDK_POLLER_BUSY;
8807 : }
8808 :
8809 : static void
8810 0 : bdev_open_async_opts_copy(struct spdk_bdev_open_async_opts *opts,
8811 : struct spdk_bdev_open_async_opts *opts_src,
8812 : size_t size)
8813 : {
8814 0 : assert(opts);
8815 0 : assert(opts_src);
8816 :
8817 0 : opts->size = size;
8818 :
8819 : #define SET_FIELD(field) \
8820 : if (offsetof(struct spdk_bdev_open_async_opts, field) + sizeof(opts->field) <= size) { \
8821 : opts->field = opts_src->field; \
8822 : } \
8823 :
8824 0 : SET_FIELD(timeout_ms);
8825 :
8826 : /* Do not remove this statement, you should always update this statement when you adding a new field,
8827 : * and do not forget to add the SET_FIELD statement for your added field. */
8828 : SPDK_STATIC_ASSERT(sizeof(struct spdk_bdev_open_async_opts) == 16, "Incorrect size");
8829 :
8830 : #undef SET_FIELD
8831 0 : }
8832 :
8833 : static void
8834 0 : bdev_open_async_opts_get_default(struct spdk_bdev_open_async_opts *opts, size_t size)
8835 : {
8836 0 : assert(opts);
8837 :
8838 0 : opts->size = size;
8839 :
8840 : #define SET_FIELD(field, value) \
8841 : if (offsetof(struct spdk_bdev_open_async_opts, field) + sizeof(opts->field) <= size) { \
8842 : opts->field = value; \
8843 : } \
8844 :
8845 0 : SET_FIELD(timeout_ms, 0);
8846 :
8847 : #undef SET_FIELD
8848 0 : }
8849 :
8850 : int
8851 0 : spdk_bdev_open_async(const char *bdev_name, bool write, spdk_bdev_event_cb_t event_cb,
8852 : void *event_ctx, struct spdk_bdev_open_async_opts *opts,
8853 : spdk_bdev_open_async_cb_t open_cb, void *open_cb_arg)
8854 : {
8855 : struct spdk_bdev_open_async_ctx *ctx;
8856 :
8857 0 : if (event_cb == NULL) {
8858 0 : SPDK_ERRLOG("Missing event callback function\n");
8859 0 : return -EINVAL;
8860 : }
8861 :
8862 0 : if (open_cb == NULL) {
8863 0 : SPDK_ERRLOG("Missing open callback function\n");
8864 0 : return -EINVAL;
8865 : }
8866 :
8867 0 : if (opts != NULL && opts->size == 0) {
8868 0 : SPDK_ERRLOG("size in the options structure should not be zero\n");
8869 0 : return -EINVAL;
8870 : }
8871 :
8872 0 : ctx = calloc(1, sizeof(*ctx));
8873 0 : if (ctx == NULL) {
8874 0 : SPDK_ERRLOG("Failed to allocate open context\n");
8875 0 : return -ENOMEM;
8876 : }
8877 :
8878 0 : ctx->bdev_name = strdup(bdev_name);
8879 0 : if (ctx->bdev_name == NULL) {
8880 0 : SPDK_ERRLOG("Failed to duplicate bdev_name\n");
8881 0 : free(ctx);
8882 0 : return -ENOMEM;
8883 : }
8884 :
8885 0 : ctx->poller = SPDK_POLLER_REGISTER(bdev_open_async, ctx, 100 * 1000);
8886 0 : if (ctx->poller == NULL) {
8887 0 : SPDK_ERRLOG("Failed to register bdev_open_async poller\n");
8888 0 : free(ctx->bdev_name);
8889 0 : free(ctx);
8890 0 : return -ENOMEM;
8891 : }
8892 :
8893 0 : ctx->cb_fn = open_cb;
8894 0 : ctx->cb_arg = open_cb_arg;
8895 0 : ctx->write = write;
8896 0 : ctx->event_cb = event_cb;
8897 0 : ctx->event_ctx = event_ctx;
8898 0 : ctx->orig_thread = spdk_get_thread();
8899 0 : ctx->start_ticks = spdk_get_ticks();
8900 :
8901 0 : bdev_open_async_opts_get_default(&ctx->opts, sizeof(ctx->opts));
8902 0 : if (opts != NULL) {
8903 0 : bdev_open_async_opts_copy(&ctx->opts, opts, opts->size);
8904 0 : }
8905 :
8906 0 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8907 :
8908 0 : TAILQ_INSERT_TAIL(&g_bdev_mgr.async_bdev_opens, ctx, tailq);
8909 0 : _bdev_open_async(ctx);
8910 :
8911 0 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8912 :
8913 0 : return 0;
8914 0 : }
8915 :
8916 : static void
8917 259 : bdev_close(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc)
8918 : {
8919 : int rc;
8920 :
8921 259 : spdk_spin_lock(&bdev->internal.spinlock);
8922 259 : spdk_spin_lock(&desc->spinlock);
8923 :
8924 259 : TAILQ_REMOVE(&bdev->internal.open_descs, desc, link);
8925 :
8926 259 : desc->closed = true;
8927 :
8928 259 : if (desc->claim != NULL) {
8929 16 : bdev_desc_release_claims(desc);
8930 16 : }
8931 :
8932 259 : if (0 == desc->refs) {
8933 248 : spdk_spin_unlock(&desc->spinlock);
8934 248 : bdev_desc_free(desc);
8935 248 : } else {
8936 11 : spdk_spin_unlock(&desc->spinlock);
8937 : }
8938 :
8939 : /* If no more descriptors, kill QoS channel */
8940 259 : if (bdev->internal.qos && TAILQ_EMPTY(&bdev->internal.open_descs)) {
8941 7 : SPDK_DEBUGLOG(bdev, "Closed last descriptor for bdev %s on thread %p. Stopping QoS.\n",
8942 : bdev->name, spdk_get_thread());
8943 :
8944 7 : if (bdev_qos_destroy(bdev)) {
8945 : /* There isn't anything we can do to recover here. Just let the
8946 : * old QoS poller keep running. The QoS handling won't change
8947 : * cores when the user allocates a new channel, but it won't break. */
8948 0 : SPDK_ERRLOG("Unable to shut down QoS poller. It will continue running on the current thread.\n");
8949 0 : }
8950 7 : }
8951 :
8952 259 : if (bdev->internal.status == SPDK_BDEV_STATUS_REMOVING && TAILQ_EMPTY(&bdev->internal.open_descs)) {
8953 15 : rc = bdev_unregister_unsafe(bdev);
8954 15 : spdk_spin_unlock(&bdev->internal.spinlock);
8955 :
8956 15 : if (rc == 0) {
8957 15 : spdk_io_device_unregister(__bdev_to_io_dev(bdev), bdev_destroy_cb);
8958 15 : }
8959 15 : } else {
8960 244 : spdk_spin_unlock(&bdev->internal.spinlock);
8961 : }
8962 259 : }
8963 :
8964 : void
8965 122 : spdk_bdev_close(struct spdk_bdev_desc *desc)
8966 : {
8967 122 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
8968 :
8969 122 : SPDK_DEBUGLOG(bdev, "Closing descriptor %p for bdev %s on thread %p\n", desc, bdev->name,
8970 : spdk_get_thread());
8971 :
8972 122 : assert(desc->thread == spdk_get_thread());
8973 :
8974 122 : spdk_poller_unregister(&desc->io_timeout_poller);
8975 :
8976 122 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8977 :
8978 122 : bdev_close(bdev, desc);
8979 :
8980 122 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8981 122 : }
8982 :
8983 : int32_t
8984 3 : spdk_bdev_get_numa_id(struct spdk_bdev *bdev)
8985 : {
8986 3 : if (bdev->numa.id_valid) {
8987 2 : return bdev->numa.id;
8988 : } else {
8989 1 : return SPDK_ENV_NUMA_ID_ANY;
8990 : }
8991 3 : }
8992 :
8993 : static void
8994 126 : bdev_register_finished(void *arg)
8995 : {
8996 126 : struct spdk_bdev_desc *desc = arg;
8997 126 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
8998 :
8999 126 : spdk_notify_send("bdev_register", spdk_bdev_get_name(bdev));
9000 :
9001 126 : spdk_spin_lock(&g_bdev_mgr.spinlock);
9002 :
9003 126 : bdev_close(bdev, desc);
9004 :
9005 126 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
9006 126 : }
9007 :
9008 : int
9009 129 : spdk_bdev_register(struct spdk_bdev *bdev)
9010 : {
9011 : struct spdk_bdev_desc *desc;
9012 129 : struct spdk_thread *thread = spdk_get_thread();
9013 : int rc;
9014 :
9015 129 : if (spdk_unlikely(!spdk_thread_is_app_thread(NULL))) {
9016 1 : SPDK_ERRLOG("Cannot register bdev %s on thread %p (%s)\n", bdev->name, thread,
9017 : thread ? spdk_thread_get_name(thread) : "null");
9018 1 : return -EINVAL;
9019 : }
9020 :
9021 128 : rc = bdev_register(bdev);
9022 128 : if (rc != 0) {
9023 2 : return rc;
9024 : }
9025 :
9026 : /* A descriptor is opened to prevent bdev deletion during examination */
9027 126 : rc = bdev_desc_alloc(bdev, _tmp_bdev_event_cb, NULL, NULL, &desc);
9028 126 : if (rc != 0) {
9029 0 : spdk_bdev_unregister(bdev, NULL, NULL);
9030 0 : return rc;
9031 : }
9032 :
9033 126 : rc = bdev_open(bdev, false, desc);
9034 126 : if (rc != 0) {
9035 0 : bdev_desc_free(desc);
9036 0 : spdk_bdev_unregister(bdev, NULL, NULL);
9037 0 : return rc;
9038 : }
9039 :
9040 : /* Examine configuration before initializing I/O */
9041 126 : bdev_examine(bdev);
9042 :
9043 126 : rc = spdk_bdev_wait_for_examine(bdev_register_finished, desc);
9044 126 : if (rc != 0) {
9045 0 : bdev_close(bdev, desc);
9046 0 : spdk_bdev_unregister(bdev, NULL, NULL);
9047 0 : }
9048 :
9049 126 : return rc;
9050 129 : }
9051 :
9052 : int
9053 26 : spdk_bdev_module_claim_bdev(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc,
9054 : struct spdk_bdev_module *module)
9055 : {
9056 26 : spdk_spin_lock(&bdev->internal.spinlock);
9057 :
9058 26 : if (bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE) {
9059 6 : LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
9060 6 : spdk_spin_unlock(&bdev->internal.spinlock);
9061 6 : return -EPERM;
9062 : }
9063 :
9064 20 : if (desc && !desc->write) {
9065 5 : desc->write = true;
9066 5 : }
9067 :
9068 20 : bdev->internal.claim_type = SPDK_BDEV_CLAIM_EXCL_WRITE;
9069 20 : bdev->internal.claim.v1.module = module;
9070 :
9071 20 : spdk_spin_unlock(&bdev->internal.spinlock);
9072 20 : return 0;
9073 26 : }
9074 :
9075 : void
9076 8 : spdk_bdev_module_release_bdev(struct spdk_bdev *bdev)
9077 : {
9078 8 : spdk_spin_lock(&bdev->internal.spinlock);
9079 :
9080 8 : assert(bdev->internal.claim.v1.module != NULL);
9081 8 : assert(bdev->internal.claim_type == SPDK_BDEV_CLAIM_EXCL_WRITE);
9082 8 : bdev->internal.claim_type = SPDK_BDEV_CLAIM_NONE;
9083 8 : bdev->internal.claim.v1.module = NULL;
9084 :
9085 8 : spdk_spin_unlock(&bdev->internal.spinlock);
9086 8 : }
9087 :
9088 : /*
9089 : * Start claims v2
9090 : */
9091 :
9092 : const char *
9093 24 : spdk_bdev_claim_get_name(enum spdk_bdev_claim_type type)
9094 : {
9095 24 : switch (type) {
9096 : case SPDK_BDEV_CLAIM_NONE:
9097 0 : return "not_claimed";
9098 : case SPDK_BDEV_CLAIM_EXCL_WRITE:
9099 6 : return "exclusive_write";
9100 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE:
9101 7 : return "read_many_write_one";
9102 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_NONE:
9103 5 : return "read_many_write_none";
9104 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
9105 6 : return "read_many_write_many";
9106 : default:
9107 0 : break;
9108 : }
9109 0 : return "invalid_claim";
9110 24 : }
9111 :
9112 : static bool
9113 96 : claim_type_is_v2(enum spdk_bdev_claim_type type)
9114 : {
9115 96 : switch (type) {
9116 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE:
9117 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_NONE:
9118 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
9119 96 : return true;
9120 : default:
9121 0 : break;
9122 : }
9123 0 : return false;
9124 96 : }
9125 :
9126 : /* Returns true if taking a claim with desc->write == false should make the descriptor writable. */
9127 : static bool
9128 13 : claim_type_promotes_to_write(enum spdk_bdev_claim_type type)
9129 : {
9130 13 : switch (type) {
9131 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE:
9132 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
9133 5 : return true;
9134 : default:
9135 8 : break;
9136 : }
9137 8 : return false;
9138 13 : }
9139 :
9140 : void
9141 44 : spdk_bdev_claim_opts_init(struct spdk_bdev_claim_opts *opts, size_t size)
9142 : {
9143 44 : if (opts == NULL) {
9144 0 : SPDK_ERRLOG("opts should not be NULL\n");
9145 0 : assert(opts != NULL);
9146 0 : return;
9147 : }
9148 44 : if (size == 0) {
9149 0 : SPDK_ERRLOG("size should not be zero\n");
9150 0 : assert(size != 0);
9151 0 : return;
9152 : }
9153 :
9154 44 : memset(opts, 0, size);
9155 44 : opts->opts_size = size;
9156 :
9157 : #define FIELD_OK(field) \
9158 : offsetof(struct spdk_bdev_claim_opts, field) + sizeof(opts->field) <= size
9159 :
9160 : #define SET_FIELD(field, value) \
9161 : if (FIELD_OK(field)) { \
9162 : opts->field = value; \
9163 : } \
9164 :
9165 44 : SET_FIELD(shared_claim_key, 0);
9166 :
9167 : #undef FIELD_OK
9168 : #undef SET_FIELD
9169 44 : }
9170 :
9171 : static int
9172 22 : claim_opts_copy(struct spdk_bdev_claim_opts *src, struct spdk_bdev_claim_opts *dst)
9173 : {
9174 22 : if (src->opts_size == 0) {
9175 0 : SPDK_ERRLOG("size should not be zero\n");
9176 0 : return -1;
9177 : }
9178 :
9179 22 : memset(dst, 0, sizeof(*dst));
9180 22 : dst->opts_size = src->opts_size;
9181 :
9182 : #define FIELD_OK(field) \
9183 : offsetof(struct spdk_bdev_claim_opts, field) + sizeof(src->field) <= src->opts_size
9184 :
9185 : #define SET_FIELD(field) \
9186 : if (FIELD_OK(field)) { \
9187 : dst->field = src->field; \
9188 : } \
9189 :
9190 22 : if (FIELD_OK(name)) {
9191 22 : snprintf(dst->name, sizeof(dst->name), "%s", src->name);
9192 22 : }
9193 :
9194 22 : SET_FIELD(shared_claim_key);
9195 :
9196 : /* You should not remove this statement, but need to update the assert statement
9197 : * if you add a new field, and also add a corresponding SET_FIELD statement */
9198 : SPDK_STATIC_ASSERT(sizeof(struct spdk_bdev_claim_opts) == 48, "Incorrect size");
9199 :
9200 : #undef FIELD_OK
9201 : #undef SET_FIELD
9202 22 : return 0;
9203 22 : }
9204 :
9205 : /* Returns 0 if a read-write-once claim can be taken. */
9206 : static int
9207 9 : claim_verify_rwo(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
9208 : struct spdk_bdev_claim_opts *opts, struct spdk_bdev_module *module)
9209 : {
9210 9 : struct spdk_bdev *bdev = desc->bdev;
9211 : struct spdk_bdev_desc *open_desc;
9212 :
9213 9 : assert(spdk_spin_held(&bdev->internal.spinlock));
9214 9 : assert(type == SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE);
9215 :
9216 9 : if (opts->shared_claim_key != 0) {
9217 1 : SPDK_ERRLOG("%s: key option not supported with read-write-once claims\n",
9218 : bdev->name);
9219 1 : return -EINVAL;
9220 : }
9221 8 : if (bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE) {
9222 1 : LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
9223 1 : return -EPERM;
9224 : }
9225 7 : if (desc->claim != NULL) {
9226 0 : SPDK_NOTICELOG("%s: descriptor already claimed bdev with module %s\n",
9227 : bdev->name, desc->claim->module->name);
9228 0 : return -EPERM;
9229 : }
9230 14 : TAILQ_FOREACH(open_desc, &bdev->internal.open_descs, link) {
9231 9 : if (desc != open_desc && open_desc->write) {
9232 2 : SPDK_NOTICELOG("%s: Cannot obtain read-write-once claim while "
9233 : "another descriptor is open for writing\n",
9234 : bdev->name);
9235 2 : return -EPERM;
9236 : }
9237 7 : }
9238 :
9239 5 : return 0;
9240 9 : }
9241 :
9242 : /* Returns 0 if a read-only-many claim can be taken. */
9243 : static int
9244 12 : claim_verify_rom(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
9245 : struct spdk_bdev_claim_opts *opts, struct spdk_bdev_module *module)
9246 : {
9247 12 : struct spdk_bdev *bdev = desc->bdev;
9248 : struct spdk_bdev_desc *open_desc;
9249 :
9250 12 : assert(spdk_spin_held(&bdev->internal.spinlock));
9251 12 : assert(type == SPDK_BDEV_CLAIM_READ_MANY_WRITE_NONE);
9252 12 : assert(desc->claim == NULL);
9253 :
9254 12 : if (desc->write) {
9255 3 : SPDK_ERRLOG("%s: Cannot obtain read-only-many claim with writable descriptor\n",
9256 : bdev->name);
9257 3 : return -EINVAL;
9258 : }
9259 9 : if (opts->shared_claim_key != 0) {
9260 1 : SPDK_ERRLOG("%s: key option not supported with read-only-may claims\n", bdev->name);
9261 1 : return -EINVAL;
9262 : }
9263 8 : if (bdev->internal.claim_type == SPDK_BDEV_CLAIM_NONE) {
9264 15 : TAILQ_FOREACH(open_desc, &bdev->internal.open_descs, link) {
9265 9 : if (open_desc->write) {
9266 0 : SPDK_NOTICELOG("%s: Cannot obtain read-only-many claim while "
9267 : "another descriptor is open for writing\n",
9268 : bdev->name);
9269 0 : return -EPERM;
9270 : }
9271 9 : }
9272 6 : }
9273 :
9274 8 : return 0;
9275 12 : }
9276 :
9277 : /* Returns 0 if a read-write-many claim can be taken. */
9278 : static int
9279 8 : claim_verify_rwm(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
9280 : struct spdk_bdev_claim_opts *opts, struct spdk_bdev_module *module)
9281 : {
9282 8 : struct spdk_bdev *bdev = desc->bdev;
9283 : struct spdk_bdev_desc *open_desc;
9284 :
9285 8 : assert(spdk_spin_held(&bdev->internal.spinlock));
9286 8 : assert(type == SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED);
9287 8 : assert(desc->claim == NULL);
9288 :
9289 8 : if (opts->shared_claim_key == 0) {
9290 2 : SPDK_ERRLOG("%s: shared_claim_key option required with read-write-may claims\n",
9291 : bdev->name);
9292 2 : return -EINVAL;
9293 : }
9294 6 : switch (bdev->internal.claim_type) {
9295 : case SPDK_BDEV_CLAIM_NONE:
9296 7 : TAILQ_FOREACH(open_desc, &bdev->internal.open_descs, link) {
9297 5 : if (open_desc == desc) {
9298 3 : continue;
9299 : }
9300 2 : if (open_desc->write) {
9301 2 : SPDK_NOTICELOG("%s: Cannot obtain read-write-many claim while "
9302 : "another descriptor is open for writing without a "
9303 : "claim\n", bdev->name);
9304 2 : return -EPERM;
9305 : }
9306 0 : }
9307 2 : break;
9308 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
9309 2 : if (opts->shared_claim_key != bdev->internal.claim.v2.key) {
9310 1 : LOG_ALREADY_CLAIMED_ERROR("already claimed with another key", bdev);
9311 1 : return -EPERM;
9312 : }
9313 1 : break;
9314 : default:
9315 0 : LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
9316 0 : return -EBUSY;
9317 : }
9318 :
9319 3 : return 0;
9320 8 : }
9321 :
9322 : /* Updates desc and its bdev with a v2 claim. */
9323 : static int
9324 16 : claim_bdev(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
9325 : struct spdk_bdev_claim_opts *opts, struct spdk_bdev_module *module)
9326 : {
9327 16 : struct spdk_bdev *bdev = desc->bdev;
9328 : struct spdk_bdev_module_claim *claim;
9329 :
9330 16 : assert(spdk_spin_held(&bdev->internal.spinlock));
9331 16 : assert(claim_type_is_v2(type));
9332 16 : assert(desc->claim == NULL);
9333 :
9334 16 : claim = calloc(1, sizeof(*desc->claim));
9335 16 : if (claim == NULL) {
9336 0 : SPDK_ERRLOG("%s: out of memory while allocating claim\n", bdev->name);
9337 0 : return -ENOMEM;
9338 : }
9339 16 : claim->module = module;
9340 16 : claim->desc = desc;
9341 : SPDK_STATIC_ASSERT(sizeof(claim->name) == sizeof(opts->name), "sizes must match");
9342 16 : memcpy(claim->name, opts->name, sizeof(claim->name));
9343 16 : desc->claim = claim;
9344 :
9345 16 : if (bdev->internal.claim_type == SPDK_BDEV_CLAIM_NONE) {
9346 13 : bdev->internal.claim_type = type;
9347 13 : TAILQ_INIT(&bdev->internal.claim.v2.claims);
9348 13 : bdev->internal.claim.v2.key = opts->shared_claim_key;
9349 13 : }
9350 16 : assert(type == bdev->internal.claim_type);
9351 :
9352 16 : TAILQ_INSERT_TAIL(&bdev->internal.claim.v2.claims, claim, link);
9353 :
9354 16 : if (!desc->write && claim_type_promotes_to_write(type)) {
9355 5 : desc->write = true;
9356 5 : }
9357 :
9358 16 : return 0;
9359 16 : }
9360 :
9361 : int
9362 39 : spdk_bdev_module_claim_bdev_desc(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
9363 : struct spdk_bdev_claim_opts *_opts,
9364 : struct spdk_bdev_module *module)
9365 : {
9366 : struct spdk_bdev *bdev;
9367 : struct spdk_bdev_claim_opts opts;
9368 39 : int rc = 0;
9369 :
9370 39 : if (desc == NULL) {
9371 0 : SPDK_ERRLOG("descriptor must not be NULL\n");
9372 0 : return -EINVAL;
9373 : }
9374 :
9375 39 : bdev = desc->bdev;
9376 :
9377 39 : if (_opts == NULL) {
9378 17 : spdk_bdev_claim_opts_init(&opts, sizeof(opts));
9379 39 : } else if (claim_opts_copy(_opts, &opts) != 0) {
9380 0 : return -EINVAL;
9381 : }
9382 :
9383 39 : spdk_spin_lock(&bdev->internal.spinlock);
9384 :
9385 39 : if (bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE &&
9386 15 : bdev->internal.claim_type != type) {
9387 10 : LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
9388 10 : spdk_spin_unlock(&bdev->internal.spinlock);
9389 10 : return -EPERM;
9390 : }
9391 :
9392 29 : if (claim_type_is_v2(type) && desc->claim != NULL) {
9393 0 : SPDK_ERRLOG("%s: descriptor already has %s claim with name '%s'\n",
9394 : bdev->name, spdk_bdev_claim_get_name(type), desc->claim->name);
9395 0 : spdk_spin_unlock(&bdev->internal.spinlock);
9396 0 : return -EPERM;
9397 : }
9398 :
9399 29 : switch (type) {
9400 : case SPDK_BDEV_CLAIM_EXCL_WRITE:
9401 0 : spdk_spin_unlock(&bdev->internal.spinlock);
9402 0 : return spdk_bdev_module_claim_bdev(bdev, desc, module);
9403 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE:
9404 9 : rc = claim_verify_rwo(desc, type, &opts, module);
9405 9 : break;
9406 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_NONE:
9407 12 : rc = claim_verify_rom(desc, type, &opts, module);
9408 12 : break;
9409 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
9410 8 : rc = claim_verify_rwm(desc, type, &opts, module);
9411 8 : break;
9412 : default:
9413 0 : SPDK_ERRLOG("%s: claim type %d not supported\n", bdev->name, type);
9414 0 : rc = -ENOTSUP;
9415 0 : }
9416 :
9417 29 : if (rc == 0) {
9418 16 : rc = claim_bdev(desc, type, &opts, module);
9419 16 : }
9420 :
9421 29 : spdk_spin_unlock(&bdev->internal.spinlock);
9422 29 : return rc;
9423 39 : }
9424 :
9425 : static void
9426 13 : claim_reset(struct spdk_bdev *bdev)
9427 : {
9428 13 : assert(spdk_spin_held(&bdev->internal.spinlock));
9429 13 : assert(claim_type_is_v2(bdev->internal.claim_type));
9430 13 : assert(TAILQ_EMPTY(&bdev->internal.claim.v2.claims));
9431 :
9432 13 : memset(&bdev->internal.claim, 0, sizeof(bdev->internal.claim));
9433 13 : bdev->internal.claim_type = SPDK_BDEV_CLAIM_NONE;
9434 13 : }
9435 :
9436 : static void
9437 16 : bdev_desc_release_claims(struct spdk_bdev_desc *desc)
9438 : {
9439 16 : struct spdk_bdev *bdev = desc->bdev;
9440 :
9441 16 : assert(spdk_spin_held(&bdev->internal.spinlock));
9442 16 : assert(claim_type_is_v2(bdev->internal.claim_type));
9443 :
9444 16 : if (bdev->internal.examine_in_progress == 0) {
9445 16 : TAILQ_REMOVE(&bdev->internal.claim.v2.claims, desc->claim, link);
9446 16 : free(desc->claim);
9447 16 : if (TAILQ_EMPTY(&bdev->internal.claim.v2.claims)) {
9448 13 : claim_reset(bdev);
9449 13 : }
9450 16 : } else {
9451 : /* This is a dead claim that will be cleaned up when bdev_examine() is done. */
9452 0 : desc->claim->module = NULL;
9453 0 : desc->claim->desc = NULL;
9454 : }
9455 16 : desc->claim = NULL;
9456 16 : }
9457 :
9458 : /*
9459 : * End claims v2
9460 : */
9461 :
9462 : struct spdk_bdev *
9463 1541 : spdk_bdev_desc_get_bdev(struct spdk_bdev_desc *desc)
9464 : {
9465 1541 : assert(desc != NULL);
9466 1541 : return desc->bdev;
9467 : }
9468 :
9469 : int
9470 1 : spdk_for_each_bdev(void *ctx, spdk_for_each_bdev_fn fn)
9471 : {
9472 : struct spdk_bdev *bdev, *tmp;
9473 : struct spdk_bdev_desc *desc;
9474 1 : int rc = 0;
9475 :
9476 1 : assert(fn != NULL);
9477 :
9478 1 : spdk_spin_lock(&g_bdev_mgr.spinlock);
9479 1 : bdev = spdk_bdev_first();
9480 9 : while (bdev != NULL) {
9481 8 : rc = bdev_desc_alloc(bdev, _tmp_bdev_event_cb, NULL, NULL, &desc);
9482 8 : if (rc != 0) {
9483 0 : break;
9484 : }
9485 8 : rc = bdev_open(bdev, false, desc);
9486 8 : if (rc != 0) {
9487 1 : bdev_desc_free(desc);
9488 1 : if (rc == -ENODEV) {
9489 : /* Ignore the error and move to the next bdev. */
9490 1 : rc = 0;
9491 1 : bdev = spdk_bdev_next(bdev);
9492 1 : continue;
9493 : }
9494 0 : break;
9495 : }
9496 7 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
9497 :
9498 7 : rc = fn(ctx, bdev);
9499 :
9500 7 : spdk_spin_lock(&g_bdev_mgr.spinlock);
9501 7 : tmp = spdk_bdev_next(bdev);
9502 7 : bdev_close(bdev, desc);
9503 7 : if (rc != 0) {
9504 0 : break;
9505 : }
9506 7 : bdev = tmp;
9507 : }
9508 1 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
9509 :
9510 1 : return rc;
9511 : }
9512 :
9513 : int
9514 1 : spdk_for_each_bdev_leaf(void *ctx, spdk_for_each_bdev_fn fn)
9515 : {
9516 : struct spdk_bdev *bdev, *tmp;
9517 : struct spdk_bdev_desc *desc;
9518 1 : int rc = 0;
9519 :
9520 1 : assert(fn != NULL);
9521 :
9522 1 : spdk_spin_lock(&g_bdev_mgr.spinlock);
9523 1 : bdev = spdk_bdev_first_leaf();
9524 6 : while (bdev != NULL) {
9525 5 : rc = bdev_desc_alloc(bdev, _tmp_bdev_event_cb, NULL, NULL, &desc);
9526 5 : if (rc != 0) {
9527 0 : break;
9528 : }
9529 5 : rc = bdev_open(bdev, false, desc);
9530 5 : if (rc != 0) {
9531 1 : bdev_desc_free(desc);
9532 1 : if (rc == -ENODEV) {
9533 : /* Ignore the error and move to the next bdev. */
9534 1 : rc = 0;
9535 1 : bdev = spdk_bdev_next_leaf(bdev);
9536 1 : continue;
9537 : }
9538 0 : break;
9539 : }
9540 4 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
9541 :
9542 4 : rc = fn(ctx, bdev);
9543 :
9544 4 : spdk_spin_lock(&g_bdev_mgr.spinlock);
9545 4 : tmp = spdk_bdev_next_leaf(bdev);
9546 4 : bdev_close(bdev, desc);
9547 4 : if (rc != 0) {
9548 0 : break;
9549 : }
9550 4 : bdev = tmp;
9551 : }
9552 1 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
9553 :
9554 1 : return rc;
9555 : }
9556 :
9557 : void
9558 0 : spdk_bdev_io_get_iovec(struct spdk_bdev_io *bdev_io, struct iovec **iovp, int *iovcntp)
9559 : {
9560 : struct iovec *iovs;
9561 : int iovcnt;
9562 :
9563 0 : if (bdev_io == NULL) {
9564 0 : return;
9565 : }
9566 :
9567 0 : switch (bdev_io->type) {
9568 : case SPDK_BDEV_IO_TYPE_READ:
9569 : case SPDK_BDEV_IO_TYPE_WRITE:
9570 : case SPDK_BDEV_IO_TYPE_ZCOPY:
9571 0 : iovs = bdev_io->u.bdev.iovs;
9572 0 : iovcnt = bdev_io->u.bdev.iovcnt;
9573 0 : break;
9574 : default:
9575 0 : iovs = NULL;
9576 0 : iovcnt = 0;
9577 0 : break;
9578 : }
9579 :
9580 0 : if (iovp) {
9581 0 : *iovp = iovs;
9582 0 : }
9583 0 : if (iovcntp) {
9584 0 : *iovcntp = iovcnt;
9585 0 : }
9586 0 : }
9587 :
9588 : void *
9589 0 : spdk_bdev_io_get_md_buf(struct spdk_bdev_io *bdev_io)
9590 : {
9591 0 : if (bdev_io == NULL) {
9592 0 : return NULL;
9593 : }
9594 :
9595 0 : if (!spdk_bdev_is_md_separate(bdev_io->bdev)) {
9596 0 : return NULL;
9597 : }
9598 :
9599 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_READ ||
9600 0 : bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
9601 0 : return bdev_io->u.bdev.md_buf;
9602 : }
9603 :
9604 0 : return NULL;
9605 0 : }
9606 :
9607 : void *
9608 0 : spdk_bdev_io_get_cb_arg(struct spdk_bdev_io *bdev_io)
9609 : {
9610 0 : if (bdev_io == NULL) {
9611 0 : assert(false);
9612 : return NULL;
9613 : }
9614 :
9615 0 : return bdev_io->internal.caller_ctx;
9616 : }
9617 :
9618 : void
9619 7 : spdk_bdev_module_list_add(struct spdk_bdev_module *bdev_module)
9620 : {
9621 :
9622 7 : if (spdk_bdev_module_list_find(bdev_module->name)) {
9623 0 : SPDK_ERRLOG("ERROR: module '%s' already registered.\n", bdev_module->name);
9624 0 : assert(false);
9625 : }
9626 :
9627 7 : spdk_spin_init(&bdev_module->internal.spinlock);
9628 7 : TAILQ_INIT(&bdev_module->internal.quiesced_ranges);
9629 :
9630 : /*
9631 : * Modules with examine callbacks must be initialized first, so they are
9632 : * ready to handle examine callbacks from later modules that will
9633 : * register physical bdevs.
9634 : */
9635 7 : if (bdev_module->examine_config != NULL || bdev_module->examine_disk != NULL) {
9636 4 : TAILQ_INSERT_HEAD(&g_bdev_mgr.bdev_modules, bdev_module, internal.tailq);
9637 4 : } else {
9638 3 : TAILQ_INSERT_TAIL(&g_bdev_mgr.bdev_modules, bdev_module, internal.tailq);
9639 : }
9640 7 : }
9641 :
9642 : struct spdk_bdev_module *
9643 7 : spdk_bdev_module_list_find(const char *name)
9644 : {
9645 : struct spdk_bdev_module *bdev_module;
9646 :
9647 14 : TAILQ_FOREACH(bdev_module, &g_bdev_mgr.bdev_modules, internal.tailq) {
9648 7 : if (strcmp(name, bdev_module->name) == 0) {
9649 0 : break;
9650 : }
9651 7 : }
9652 :
9653 7 : return bdev_module;
9654 : }
9655 :
9656 : static int
9657 6 : bdev_write_zero_buffer(struct spdk_bdev_io *bdev_io)
9658 : {
9659 : uint64_t num_blocks;
9660 6 : void *md_buf = NULL;
9661 :
9662 6 : num_blocks = bdev_io->u.bdev.num_blocks;
9663 :
9664 6 : if (spdk_bdev_is_md_separate(bdev_io->bdev)) {
9665 4 : md_buf = (char *)g_bdev_mgr.zero_buffer +
9666 2 : spdk_bdev_get_block_size(bdev_io->bdev) * num_blocks;
9667 2 : }
9668 :
9669 12 : return bdev_write_blocks_with_md(bdev_io->internal.desc,
9670 6 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
9671 6 : g_bdev_mgr.zero_buffer, md_buf,
9672 6 : bdev_io->u.bdev.offset_blocks, num_blocks,
9673 6 : bdev_write_zero_buffer_done, bdev_io);
9674 : }
9675 :
9676 : static void
9677 6 : bdev_write_zero_buffer_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
9678 : {
9679 6 : struct spdk_bdev_io *parent_io = cb_arg;
9680 :
9681 6 : spdk_bdev_free_io(bdev_io);
9682 :
9683 6 : parent_io->internal.status = success ? SPDK_BDEV_IO_STATUS_SUCCESS : SPDK_BDEV_IO_STATUS_FAILED;
9684 6 : parent_io->internal.cb(parent_io, success, parent_io->internal.caller_ctx);
9685 6 : }
9686 :
9687 : static void
9688 10 : bdev_set_qos_limit_done(struct set_qos_limit_ctx *ctx, int status)
9689 : {
9690 10 : spdk_spin_lock(&ctx->bdev->internal.spinlock);
9691 10 : ctx->bdev->internal.qos_mod_in_progress = false;
9692 10 : spdk_spin_unlock(&ctx->bdev->internal.spinlock);
9693 :
9694 10 : if (ctx->cb_fn) {
9695 8 : ctx->cb_fn(ctx->cb_arg, status);
9696 8 : }
9697 10 : free(ctx);
9698 10 : }
9699 :
9700 : static void
9701 2 : bdev_disable_qos_done(void *cb_arg)
9702 : {
9703 2 : struct set_qos_limit_ctx *ctx = cb_arg;
9704 2 : struct spdk_bdev *bdev = ctx->bdev;
9705 : struct spdk_bdev_qos *qos;
9706 :
9707 2 : spdk_spin_lock(&bdev->internal.spinlock);
9708 2 : qos = bdev->internal.qos;
9709 2 : bdev->internal.qos = NULL;
9710 2 : spdk_spin_unlock(&bdev->internal.spinlock);
9711 :
9712 2 : if (qos->thread != NULL) {
9713 2 : spdk_put_io_channel(spdk_io_channel_from_ctx(qos->ch));
9714 2 : spdk_poller_unregister(&qos->poller);
9715 2 : }
9716 :
9717 2 : free(qos);
9718 :
9719 2 : bdev_set_qos_limit_done(ctx, 0);
9720 2 : }
9721 :
9722 : static void
9723 2 : bdev_disable_qos_msg_done(struct spdk_bdev *bdev, void *_ctx, int status)
9724 : {
9725 2 : struct set_qos_limit_ctx *ctx = _ctx;
9726 : struct spdk_thread *thread;
9727 :
9728 2 : spdk_spin_lock(&bdev->internal.spinlock);
9729 2 : thread = bdev->internal.qos->thread;
9730 2 : spdk_spin_unlock(&bdev->internal.spinlock);
9731 :
9732 2 : if (thread != NULL) {
9733 2 : spdk_thread_send_msg(thread, bdev_disable_qos_done, ctx);
9734 2 : } else {
9735 0 : bdev_disable_qos_done(ctx);
9736 : }
9737 2 : }
9738 :
9739 : static void
9740 4 : bdev_disable_qos_msg(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
9741 : struct spdk_io_channel *ch, void *_ctx)
9742 : {
9743 4 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(ch);
9744 : struct spdk_bdev_io *bdev_io;
9745 :
9746 4 : bdev_ch->flags &= ~BDEV_CH_QOS_ENABLED;
9747 :
9748 6 : while (!TAILQ_EMPTY(&bdev_ch->qos_queued_io)) {
9749 : /* Re-submit the queued I/O. */
9750 2 : bdev_io = TAILQ_FIRST(&bdev_ch->qos_queued_io);
9751 2 : TAILQ_REMOVE(&bdev_ch->qos_queued_io, bdev_io, internal.link);
9752 2 : _bdev_io_submit(bdev_io);
9753 : }
9754 :
9755 4 : spdk_bdev_for_each_channel_continue(i, 0);
9756 4 : }
9757 :
9758 : static void
9759 1 : bdev_update_qos_rate_limit_msg(void *cb_arg)
9760 : {
9761 1 : struct set_qos_limit_ctx *ctx = cb_arg;
9762 1 : struct spdk_bdev *bdev = ctx->bdev;
9763 :
9764 1 : spdk_spin_lock(&bdev->internal.spinlock);
9765 1 : bdev_qos_update_max_quota_per_timeslice(bdev->internal.qos);
9766 1 : spdk_spin_unlock(&bdev->internal.spinlock);
9767 :
9768 1 : bdev_set_qos_limit_done(ctx, 0);
9769 1 : }
9770 :
9771 : static void
9772 9 : bdev_enable_qos_msg(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
9773 : struct spdk_io_channel *ch, void *_ctx)
9774 : {
9775 9 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(ch);
9776 :
9777 9 : spdk_spin_lock(&bdev->internal.spinlock);
9778 9 : bdev_enable_qos(bdev, bdev_ch);
9779 9 : spdk_spin_unlock(&bdev->internal.spinlock);
9780 9 : spdk_bdev_for_each_channel_continue(i, 0);
9781 9 : }
9782 :
9783 : static void
9784 6 : bdev_enable_qos_done(struct spdk_bdev *bdev, void *_ctx, int status)
9785 : {
9786 6 : struct set_qos_limit_ctx *ctx = _ctx;
9787 :
9788 6 : bdev_set_qos_limit_done(ctx, status);
9789 6 : }
9790 :
9791 : static void
9792 7 : bdev_set_qos_rate_limits(struct spdk_bdev *bdev, uint64_t *limits)
9793 : {
9794 : int i;
9795 :
9796 7 : assert(bdev->internal.qos != NULL);
9797 :
9798 35 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
9799 28 : if (limits[i] != SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
9800 28 : bdev->internal.qos->rate_limits[i].limit = limits[i];
9801 :
9802 28 : if (limits[i] == 0) {
9803 19 : bdev->internal.qos->rate_limits[i].limit =
9804 : SPDK_BDEV_QOS_LIMIT_NOT_DEFINED;
9805 19 : }
9806 28 : }
9807 28 : }
9808 7 : }
9809 :
9810 : void
9811 9 : spdk_bdev_set_qos_rate_limits(struct spdk_bdev *bdev, uint64_t *limits,
9812 : void (*cb_fn)(void *cb_arg, int status), void *cb_arg)
9813 : {
9814 : struct set_qos_limit_ctx *ctx;
9815 : uint32_t limit_set_complement;
9816 : uint64_t min_limit_per_sec;
9817 : int i;
9818 9 : bool disable_rate_limit = true;
9819 :
9820 45 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
9821 36 : if (limits[i] == SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
9822 0 : continue;
9823 : }
9824 :
9825 36 : if (limits[i] > 0) {
9826 10 : disable_rate_limit = false;
9827 10 : }
9828 :
9829 36 : if (bdev_qos_is_iops_rate_limit(i) == true) {
9830 9 : min_limit_per_sec = SPDK_BDEV_QOS_MIN_IOS_PER_SEC;
9831 9 : } else {
9832 27 : if (limits[i] > SPDK_BDEV_QOS_MAX_MBYTES_PER_SEC) {
9833 0 : SPDK_WARNLOG("Requested rate limit %" PRIu64 " will result in uint64_t overflow, "
9834 : "reset to %" PRIu64 "\n", limits[i], SPDK_BDEV_QOS_MAX_MBYTES_PER_SEC);
9835 0 : limits[i] = SPDK_BDEV_QOS_MAX_MBYTES_PER_SEC;
9836 0 : }
9837 : /* Change from megabyte to byte rate limit */
9838 27 : limits[i] = limits[i] * 1024 * 1024;
9839 27 : min_limit_per_sec = SPDK_BDEV_QOS_MIN_BYTES_PER_SEC;
9840 : }
9841 :
9842 36 : limit_set_complement = limits[i] % min_limit_per_sec;
9843 36 : if (limit_set_complement) {
9844 0 : SPDK_ERRLOG("Requested rate limit %" PRIu64 " is not a multiple of %" PRIu64 "\n",
9845 : limits[i], min_limit_per_sec);
9846 0 : limits[i] += min_limit_per_sec - limit_set_complement;
9847 0 : SPDK_ERRLOG("Round up the rate limit to %" PRIu64 "\n", limits[i]);
9848 0 : }
9849 36 : }
9850 :
9851 9 : ctx = calloc(1, sizeof(*ctx));
9852 9 : if (ctx == NULL) {
9853 0 : cb_fn(cb_arg, -ENOMEM);
9854 0 : return;
9855 : }
9856 :
9857 9 : ctx->cb_fn = cb_fn;
9858 9 : ctx->cb_arg = cb_arg;
9859 9 : ctx->bdev = bdev;
9860 :
9861 9 : spdk_spin_lock(&bdev->internal.spinlock);
9862 9 : if (bdev->internal.qos_mod_in_progress) {
9863 1 : spdk_spin_unlock(&bdev->internal.spinlock);
9864 1 : free(ctx);
9865 1 : cb_fn(cb_arg, -EAGAIN);
9866 1 : return;
9867 : }
9868 8 : bdev->internal.qos_mod_in_progress = true;
9869 :
9870 8 : if (disable_rate_limit == true && bdev->internal.qos) {
9871 10 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
9872 8 : if (limits[i] == SPDK_BDEV_QOS_LIMIT_NOT_DEFINED &&
9873 0 : (bdev->internal.qos->rate_limits[i].limit > 0 &&
9874 0 : bdev->internal.qos->rate_limits[i].limit !=
9875 : SPDK_BDEV_QOS_LIMIT_NOT_DEFINED)) {
9876 0 : disable_rate_limit = false;
9877 0 : break;
9878 : }
9879 8 : }
9880 2 : }
9881 :
9882 8 : if (disable_rate_limit == false) {
9883 5 : if (bdev->internal.qos == NULL) {
9884 4 : bdev->internal.qos = calloc(1, sizeof(*bdev->internal.qos));
9885 4 : if (!bdev->internal.qos) {
9886 0 : spdk_spin_unlock(&bdev->internal.spinlock);
9887 0 : SPDK_ERRLOG("Unable to allocate memory for QoS tracking\n");
9888 0 : bdev_set_qos_limit_done(ctx, -ENOMEM);
9889 0 : return;
9890 : }
9891 4 : }
9892 :
9893 5 : if (bdev->internal.qos->thread == NULL) {
9894 : /* Enabling */
9895 4 : bdev_set_qos_rate_limits(bdev, limits);
9896 :
9897 4 : spdk_bdev_for_each_channel(bdev, bdev_enable_qos_msg, ctx,
9898 : bdev_enable_qos_done);
9899 4 : } else {
9900 : /* Updating */
9901 1 : bdev_set_qos_rate_limits(bdev, limits);
9902 :
9903 2 : spdk_thread_send_msg(bdev->internal.qos->thread,
9904 1 : bdev_update_qos_rate_limit_msg, ctx);
9905 : }
9906 5 : } else {
9907 3 : if (bdev->internal.qos != NULL) {
9908 2 : bdev_set_qos_rate_limits(bdev, limits);
9909 :
9910 : /* Disabling */
9911 2 : spdk_bdev_for_each_channel(bdev, bdev_disable_qos_msg, ctx,
9912 : bdev_disable_qos_msg_done);
9913 2 : } else {
9914 1 : spdk_spin_unlock(&bdev->internal.spinlock);
9915 1 : bdev_set_qos_limit_done(ctx, 0);
9916 1 : return;
9917 : }
9918 : }
9919 :
9920 7 : spdk_spin_unlock(&bdev->internal.spinlock);
9921 9 : }
9922 :
9923 : struct spdk_bdev_histogram_ctx {
9924 : spdk_bdev_histogram_status_cb cb_fn;
9925 : void *cb_arg;
9926 : struct spdk_bdev *bdev;
9927 : int status;
9928 : };
9929 :
9930 : static void
9931 2 : bdev_histogram_disable_channel_cb(struct spdk_bdev *bdev, void *_ctx, int status)
9932 : {
9933 2 : struct spdk_bdev_histogram_ctx *ctx = _ctx;
9934 :
9935 2 : spdk_spin_lock(&ctx->bdev->internal.spinlock);
9936 2 : ctx->bdev->internal.histogram_in_progress = false;
9937 2 : spdk_spin_unlock(&ctx->bdev->internal.spinlock);
9938 2 : ctx->cb_fn(ctx->cb_arg, ctx->status);
9939 2 : free(ctx);
9940 2 : }
9941 :
9942 : static void
9943 3 : bdev_histogram_disable_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
9944 : struct spdk_io_channel *_ch, void *_ctx)
9945 : {
9946 3 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
9947 :
9948 3 : if (ch->histogram != NULL) {
9949 3 : spdk_histogram_data_free(ch->histogram);
9950 3 : ch->histogram = NULL;
9951 3 : }
9952 3 : spdk_bdev_for_each_channel_continue(i, 0);
9953 3 : }
9954 :
9955 : static void
9956 2 : bdev_histogram_enable_channel_cb(struct spdk_bdev *bdev, void *_ctx, int status)
9957 : {
9958 2 : struct spdk_bdev_histogram_ctx *ctx = _ctx;
9959 :
9960 2 : if (status != 0) {
9961 0 : ctx->status = status;
9962 0 : ctx->bdev->internal.histogram_enabled = false;
9963 0 : spdk_bdev_for_each_channel(ctx->bdev, bdev_histogram_disable_channel, ctx,
9964 : bdev_histogram_disable_channel_cb);
9965 0 : } else {
9966 2 : spdk_spin_lock(&ctx->bdev->internal.spinlock);
9967 2 : ctx->bdev->internal.histogram_in_progress = false;
9968 2 : spdk_spin_unlock(&ctx->bdev->internal.spinlock);
9969 2 : ctx->cb_fn(ctx->cb_arg, ctx->status);
9970 2 : free(ctx);
9971 : }
9972 2 : }
9973 :
9974 : static void
9975 3 : bdev_histogram_enable_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
9976 : struct spdk_io_channel *_ch, void *_ctx)
9977 : {
9978 3 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
9979 3 : int status = 0;
9980 :
9981 3 : if (ch->histogram == NULL) {
9982 3 : ch->histogram = spdk_histogram_data_alloc();
9983 3 : if (ch->histogram == NULL) {
9984 0 : status = -ENOMEM;
9985 0 : }
9986 3 : }
9987 :
9988 3 : spdk_bdev_for_each_channel_continue(i, status);
9989 3 : }
9990 :
9991 : void
9992 4 : spdk_bdev_histogram_enable_ext(struct spdk_bdev *bdev, spdk_bdev_histogram_status_cb cb_fn,
9993 : void *cb_arg, bool enable, struct spdk_bdev_enable_histogram_opts *opts)
9994 : {
9995 : struct spdk_bdev_histogram_ctx *ctx;
9996 :
9997 4 : ctx = calloc(1, sizeof(struct spdk_bdev_histogram_ctx));
9998 4 : if (ctx == NULL) {
9999 0 : cb_fn(cb_arg, -ENOMEM);
10000 0 : return;
10001 : }
10002 :
10003 4 : ctx->bdev = bdev;
10004 4 : ctx->status = 0;
10005 4 : ctx->cb_fn = cb_fn;
10006 4 : ctx->cb_arg = cb_arg;
10007 :
10008 4 : spdk_spin_lock(&bdev->internal.spinlock);
10009 4 : if (bdev->internal.histogram_in_progress) {
10010 0 : spdk_spin_unlock(&bdev->internal.spinlock);
10011 0 : free(ctx);
10012 0 : cb_fn(cb_arg, -EAGAIN);
10013 0 : return;
10014 : }
10015 :
10016 4 : bdev->internal.histogram_in_progress = true;
10017 4 : spdk_spin_unlock(&bdev->internal.spinlock);
10018 :
10019 4 : bdev->internal.histogram_enabled = enable;
10020 4 : bdev->internal.histogram_io_type = opts->io_type;
10021 :
10022 4 : if (enable) {
10023 : /* Allocate histogram for each channel */
10024 2 : spdk_bdev_for_each_channel(bdev, bdev_histogram_enable_channel, ctx,
10025 : bdev_histogram_enable_channel_cb);
10026 2 : } else {
10027 2 : spdk_bdev_for_each_channel(bdev, bdev_histogram_disable_channel, ctx,
10028 : bdev_histogram_disable_channel_cb);
10029 : }
10030 4 : }
10031 :
10032 : void
10033 4 : spdk_bdev_enable_histogram_opts_init(struct spdk_bdev_enable_histogram_opts *opts, size_t size)
10034 : {
10035 4 : if (opts == NULL) {
10036 0 : SPDK_ERRLOG("opts should not be NULL\n");
10037 0 : assert(opts != NULL);
10038 0 : return;
10039 : }
10040 4 : if (size == 0) {
10041 0 : SPDK_ERRLOG("size should not be zero\n");
10042 0 : assert(size != 0);
10043 0 : return;
10044 : }
10045 :
10046 4 : memset(opts, 0, size);
10047 4 : opts->size = size;
10048 :
10049 : #define FIELD_OK(field) \
10050 : offsetof(struct spdk_bdev_enable_histogram_opts, field) + sizeof(opts->field) <= size
10051 :
10052 : #define SET_FIELD(field, value) \
10053 : if (FIELD_OK(field)) { \
10054 : opts->field = value; \
10055 : } \
10056 :
10057 4 : SET_FIELD(io_type, 0);
10058 :
10059 : /* You should not remove this statement, but need to update the assert statement
10060 : * if you add a new field, and also add a corresponding SET_FIELD statement */
10061 : SPDK_STATIC_ASSERT(sizeof(struct spdk_bdev_enable_histogram_opts) == 9, "Incorrect size");
10062 :
10063 : #undef FIELD_OK
10064 : #undef SET_FIELD
10065 4 : }
10066 :
10067 : void
10068 4 : spdk_bdev_histogram_enable(struct spdk_bdev *bdev, spdk_bdev_histogram_status_cb cb_fn,
10069 : void *cb_arg, bool enable)
10070 : {
10071 : struct spdk_bdev_enable_histogram_opts opts;
10072 :
10073 4 : spdk_bdev_enable_histogram_opts_init(&opts, sizeof(opts));
10074 4 : spdk_bdev_histogram_enable_ext(bdev, cb_fn, cb_arg, enable, &opts);
10075 4 : }
10076 :
10077 : struct spdk_bdev_histogram_data_ctx {
10078 : spdk_bdev_histogram_data_cb cb_fn;
10079 : void *cb_arg;
10080 : struct spdk_bdev *bdev;
10081 : /** merged histogram data from all channels */
10082 : struct spdk_histogram_data *histogram;
10083 : };
10084 :
10085 : static void
10086 5 : bdev_histogram_get_channel_cb(struct spdk_bdev *bdev, void *_ctx, int status)
10087 : {
10088 5 : struct spdk_bdev_histogram_data_ctx *ctx = _ctx;
10089 :
10090 5 : ctx->cb_fn(ctx->cb_arg, status, ctx->histogram);
10091 5 : free(ctx);
10092 5 : }
10093 :
10094 : static void
10095 7 : bdev_histogram_get_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
10096 : struct spdk_io_channel *_ch, void *_ctx)
10097 : {
10098 7 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
10099 7 : struct spdk_bdev_histogram_data_ctx *ctx = _ctx;
10100 7 : int status = 0;
10101 :
10102 7 : if (ch->histogram == NULL) {
10103 1 : status = -EFAULT;
10104 1 : } else {
10105 6 : spdk_histogram_data_merge(ctx->histogram, ch->histogram);
10106 : }
10107 :
10108 7 : spdk_bdev_for_each_channel_continue(i, status);
10109 7 : }
10110 :
10111 : void
10112 5 : spdk_bdev_histogram_get(struct spdk_bdev *bdev, struct spdk_histogram_data *histogram,
10113 : spdk_bdev_histogram_data_cb cb_fn,
10114 : void *cb_arg)
10115 : {
10116 : struct spdk_bdev_histogram_data_ctx *ctx;
10117 :
10118 5 : ctx = calloc(1, sizeof(struct spdk_bdev_histogram_data_ctx));
10119 5 : if (ctx == NULL) {
10120 0 : cb_fn(cb_arg, -ENOMEM, NULL);
10121 0 : return;
10122 : }
10123 :
10124 5 : ctx->bdev = bdev;
10125 5 : ctx->cb_fn = cb_fn;
10126 5 : ctx->cb_arg = cb_arg;
10127 :
10128 5 : ctx->histogram = histogram;
10129 :
10130 5 : spdk_bdev_for_each_channel(bdev, bdev_histogram_get_channel, ctx,
10131 : bdev_histogram_get_channel_cb);
10132 5 : }
10133 :
10134 : void
10135 2 : spdk_bdev_channel_get_histogram(struct spdk_io_channel *ch, spdk_bdev_histogram_data_cb cb_fn,
10136 : void *cb_arg)
10137 : {
10138 2 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(ch);
10139 2 : int status = 0;
10140 :
10141 2 : assert(cb_fn != NULL);
10142 :
10143 2 : if (bdev_ch->histogram == NULL) {
10144 1 : status = -EFAULT;
10145 1 : }
10146 2 : cb_fn(cb_arg, status, bdev_ch->histogram);
10147 2 : }
10148 :
10149 : size_t
10150 0 : spdk_bdev_get_media_events(struct spdk_bdev_desc *desc, struct spdk_bdev_media_event *events,
10151 : size_t max_events)
10152 : {
10153 : struct media_event_entry *entry;
10154 0 : size_t num_events = 0;
10155 :
10156 0 : for (; num_events < max_events; ++num_events) {
10157 0 : entry = TAILQ_FIRST(&desc->pending_media_events);
10158 0 : if (entry == NULL) {
10159 0 : break;
10160 : }
10161 :
10162 0 : events[num_events] = entry->event;
10163 0 : TAILQ_REMOVE(&desc->pending_media_events, entry, tailq);
10164 0 : TAILQ_INSERT_TAIL(&desc->free_media_events, entry, tailq);
10165 0 : }
10166 :
10167 0 : return num_events;
10168 : }
10169 :
10170 : int
10171 0 : spdk_bdev_push_media_events(struct spdk_bdev *bdev, const struct spdk_bdev_media_event *events,
10172 : size_t num_events)
10173 : {
10174 : struct spdk_bdev_desc *desc;
10175 : struct media_event_entry *entry;
10176 : size_t event_id;
10177 0 : int rc = 0;
10178 :
10179 0 : assert(bdev->media_events);
10180 :
10181 0 : spdk_spin_lock(&bdev->internal.spinlock);
10182 0 : TAILQ_FOREACH(desc, &bdev->internal.open_descs, link) {
10183 0 : if (desc->write) {
10184 0 : break;
10185 : }
10186 0 : }
10187 :
10188 0 : if (desc == NULL || desc->media_events_buffer == NULL) {
10189 0 : rc = -ENODEV;
10190 0 : goto out;
10191 : }
10192 :
10193 0 : for (event_id = 0; event_id < num_events; ++event_id) {
10194 0 : entry = TAILQ_FIRST(&desc->free_media_events);
10195 0 : if (entry == NULL) {
10196 0 : break;
10197 : }
10198 :
10199 0 : TAILQ_REMOVE(&desc->free_media_events, entry, tailq);
10200 0 : TAILQ_INSERT_TAIL(&desc->pending_media_events, entry, tailq);
10201 0 : entry->event = events[event_id];
10202 0 : }
10203 :
10204 0 : rc = event_id;
10205 : out:
10206 0 : spdk_spin_unlock(&bdev->internal.spinlock);
10207 0 : return rc;
10208 : }
10209 :
10210 : static void
10211 0 : _media_management_notify(void *arg)
10212 : {
10213 0 : struct spdk_bdev_desc *desc = arg;
10214 :
10215 0 : _event_notify(desc, SPDK_BDEV_EVENT_MEDIA_MANAGEMENT);
10216 0 : }
10217 :
10218 : void
10219 0 : spdk_bdev_notify_media_management(struct spdk_bdev *bdev)
10220 : {
10221 : struct spdk_bdev_desc *desc;
10222 :
10223 0 : spdk_spin_lock(&bdev->internal.spinlock);
10224 0 : TAILQ_FOREACH(desc, &bdev->internal.open_descs, link) {
10225 0 : if (!TAILQ_EMPTY(&desc->pending_media_events)) {
10226 0 : event_notify(desc, _media_management_notify);
10227 0 : }
10228 0 : }
10229 0 : spdk_spin_unlock(&bdev->internal.spinlock);
10230 0 : }
10231 :
10232 : struct locked_lba_range_ctx {
10233 : struct lba_range range;
10234 : struct lba_range *current_range;
10235 : struct lba_range *owner_range;
10236 : struct spdk_poller *poller;
10237 : lock_range_cb cb_fn;
10238 : void *cb_arg;
10239 : };
10240 :
10241 : static void
10242 0 : bdev_lock_error_cleanup_cb(struct spdk_bdev *bdev, void *_ctx, int status)
10243 : {
10244 0 : struct locked_lba_range_ctx *ctx = _ctx;
10245 :
10246 0 : ctx->cb_fn(&ctx->range, ctx->cb_arg, -ENOMEM);
10247 0 : free(ctx);
10248 0 : }
10249 :
10250 : static void bdev_unlock_lba_range_get_channel(struct spdk_bdev_channel_iter *i,
10251 : struct spdk_bdev *bdev, struct spdk_io_channel *ch, void *_ctx);
10252 :
10253 : static void
10254 14 : bdev_lock_lba_range_cb(struct spdk_bdev *bdev, void *_ctx, int status)
10255 : {
10256 14 : struct locked_lba_range_ctx *ctx = _ctx;
10257 :
10258 14 : if (status == -ENOMEM) {
10259 : /* One of the channels could not allocate a range object.
10260 : * So we have to go back and clean up any ranges that were
10261 : * allocated successfully before we return error status to
10262 : * the caller. We can reuse the unlock function to do that
10263 : * clean up.
10264 : */
10265 0 : spdk_bdev_for_each_channel(bdev, bdev_unlock_lba_range_get_channel, ctx,
10266 : bdev_lock_error_cleanup_cb);
10267 0 : return;
10268 : }
10269 :
10270 : /* All channels have locked this range and no I/O overlapping the range
10271 : * are outstanding! Set the owner_ch for the range object for the
10272 : * locking channel, so that this channel will know that it is allowed
10273 : * to write to this range.
10274 : */
10275 14 : if (ctx->owner_range != NULL) {
10276 10 : ctx->owner_range->owner_ch = ctx->range.owner_ch;
10277 10 : }
10278 :
10279 14 : ctx->cb_fn(&ctx->range, ctx->cb_arg, status);
10280 :
10281 : /* Don't free the ctx here. Its range is in the bdev's global list of
10282 : * locked ranges still, and will be removed and freed when this range
10283 : * is later unlocked.
10284 : */
10285 14 : }
10286 :
10287 : static int
10288 17 : bdev_lock_lba_range_check_io(void *_i)
10289 : {
10290 17 : struct spdk_bdev_channel_iter *i = _i;
10291 17 : struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i->i);
10292 17 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
10293 17 : struct locked_lba_range_ctx *ctx = i->ctx;
10294 17 : struct lba_range *range = ctx->current_range;
10295 : struct spdk_bdev_io *bdev_io;
10296 :
10297 17 : spdk_poller_unregister(&ctx->poller);
10298 :
10299 : /* The range is now in the locked_ranges, so no new IO can be submitted to this
10300 : * range. But we need to wait until any outstanding IO overlapping with this range
10301 : * are completed.
10302 : */
10303 18 : TAILQ_FOREACH(bdev_io, &ch->io_submitted, internal.ch_link) {
10304 3 : if (bdev_io_range_is_locked(bdev_io, range)) {
10305 2 : ctx->poller = SPDK_POLLER_REGISTER(bdev_lock_lba_range_check_io, i, 100);
10306 2 : return SPDK_POLLER_BUSY;
10307 : }
10308 1 : }
10309 :
10310 15 : spdk_bdev_for_each_channel_continue(i, 0);
10311 15 : return SPDK_POLLER_BUSY;
10312 17 : }
10313 :
10314 : static void
10315 15 : bdev_lock_lba_range_get_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
10316 : struct spdk_io_channel *_ch, void *_ctx)
10317 : {
10318 15 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
10319 15 : struct locked_lba_range_ctx *ctx = _ctx;
10320 : struct lba_range *range;
10321 :
10322 16 : TAILQ_FOREACH(range, &ch->locked_ranges, tailq) {
10323 1 : if (range->length == ctx->range.length &&
10324 0 : range->offset == ctx->range.offset &&
10325 0 : range->locked_ctx == ctx->range.locked_ctx) {
10326 : /* This range already exists on this channel, so don't add
10327 : * it again. This can happen when a new channel is created
10328 : * while the for_each_channel operation is in progress.
10329 : * Do not check for outstanding I/O in that case, since the
10330 : * range was locked before any I/O could be submitted to the
10331 : * new channel.
10332 : */
10333 0 : spdk_bdev_for_each_channel_continue(i, 0);
10334 0 : return;
10335 : }
10336 1 : }
10337 :
10338 15 : range = calloc(1, sizeof(*range));
10339 15 : if (range == NULL) {
10340 0 : spdk_bdev_for_each_channel_continue(i, -ENOMEM);
10341 0 : return;
10342 : }
10343 :
10344 15 : range->length = ctx->range.length;
10345 15 : range->offset = ctx->range.offset;
10346 15 : range->locked_ctx = ctx->range.locked_ctx;
10347 15 : range->quiesce = ctx->range.quiesce;
10348 15 : ctx->current_range = range;
10349 15 : if (ctx->range.owner_ch == ch) {
10350 : /* This is the range object for the channel that will hold
10351 : * the lock. Store it in the ctx object so that we can easily
10352 : * set its owner_ch after the lock is finally acquired.
10353 : */
10354 10 : ctx->owner_range = range;
10355 10 : }
10356 15 : TAILQ_INSERT_TAIL(&ch->locked_ranges, range, tailq);
10357 15 : bdev_lock_lba_range_check_io(i);
10358 15 : }
10359 :
10360 : static void
10361 14 : bdev_lock_lba_range_ctx(struct spdk_bdev *bdev, struct locked_lba_range_ctx *ctx)
10362 : {
10363 14 : assert(spdk_get_thread() == ctx->range.owner_thread);
10364 14 : assert(ctx->range.owner_ch == NULL ||
10365 : spdk_io_channel_get_thread(ctx->range.owner_ch->channel) == ctx->range.owner_thread);
10366 :
10367 : /* We will add a copy of this range to each channel now. */
10368 14 : spdk_bdev_for_each_channel(bdev, bdev_lock_lba_range_get_channel, ctx,
10369 : bdev_lock_lba_range_cb);
10370 14 : }
10371 :
10372 : static bool
10373 17 : bdev_lba_range_overlaps_tailq(struct lba_range *range, lba_range_tailq_t *tailq)
10374 : {
10375 : struct lba_range *r;
10376 :
10377 18 : TAILQ_FOREACH(r, tailq, tailq) {
10378 4 : if (bdev_lba_range_overlapped(range, r)) {
10379 3 : return true;
10380 : }
10381 1 : }
10382 14 : return false;
10383 17 : }
10384 :
10385 : static void bdev_quiesce_range_locked(struct lba_range *range, void *ctx, int status);
10386 :
10387 : static int
10388 14 : _bdev_lock_lba_range(struct spdk_bdev *bdev, struct spdk_bdev_channel *ch,
10389 : uint64_t offset, uint64_t length,
10390 : lock_range_cb cb_fn, void *cb_arg)
10391 : {
10392 : struct locked_lba_range_ctx *ctx;
10393 :
10394 14 : ctx = calloc(1, sizeof(*ctx));
10395 14 : if (ctx == NULL) {
10396 0 : return -ENOMEM;
10397 : }
10398 :
10399 14 : ctx->range.offset = offset;
10400 14 : ctx->range.length = length;
10401 14 : ctx->range.owner_thread = spdk_get_thread();
10402 14 : ctx->range.owner_ch = ch;
10403 14 : ctx->range.locked_ctx = cb_arg;
10404 14 : ctx->range.bdev = bdev;
10405 14 : ctx->range.quiesce = (cb_fn == bdev_quiesce_range_locked);
10406 14 : ctx->cb_fn = cb_fn;
10407 14 : ctx->cb_arg = cb_arg;
10408 :
10409 14 : spdk_spin_lock(&bdev->internal.spinlock);
10410 14 : if (bdev_lba_range_overlaps_tailq(&ctx->range, &bdev->internal.locked_ranges)) {
10411 : /* There is an active lock overlapping with this range.
10412 : * Put it on the pending list until this range no
10413 : * longer overlaps with another.
10414 : */
10415 2 : TAILQ_INSERT_TAIL(&bdev->internal.pending_locked_ranges, &ctx->range, tailq);
10416 2 : } else {
10417 12 : TAILQ_INSERT_TAIL(&bdev->internal.locked_ranges, &ctx->range, tailq);
10418 12 : bdev_lock_lba_range_ctx(bdev, ctx);
10419 : }
10420 14 : spdk_spin_unlock(&bdev->internal.spinlock);
10421 14 : return 0;
10422 14 : }
10423 :
10424 : static int
10425 10 : bdev_lock_lba_range(struct spdk_bdev_desc *desc, struct spdk_io_channel *_ch,
10426 : uint64_t offset, uint64_t length,
10427 : lock_range_cb cb_fn, void *cb_arg)
10428 : {
10429 10 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
10430 10 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
10431 :
10432 10 : if (cb_arg == NULL) {
10433 0 : SPDK_ERRLOG("cb_arg must not be NULL\n");
10434 0 : return -EINVAL;
10435 : }
10436 :
10437 10 : return _bdev_lock_lba_range(bdev, ch, offset, length, cb_fn, cb_arg);
10438 10 : }
10439 :
10440 : static void
10441 2 : bdev_lock_lba_range_ctx_msg(void *_ctx)
10442 : {
10443 2 : struct locked_lba_range_ctx *ctx = _ctx;
10444 :
10445 2 : bdev_lock_lba_range_ctx(ctx->range.bdev, ctx);
10446 2 : }
10447 :
10448 : static void
10449 14 : bdev_unlock_lba_range_cb(struct spdk_bdev *bdev, void *_ctx, int status)
10450 : {
10451 14 : struct locked_lba_range_ctx *ctx = _ctx;
10452 : struct locked_lba_range_ctx *pending_ctx;
10453 : struct lba_range *range, *tmp;
10454 :
10455 14 : spdk_spin_lock(&bdev->internal.spinlock);
10456 : /* Check if there are any pending locked ranges that overlap with this range
10457 : * that was just unlocked. If there are, check that it doesn't overlap with any
10458 : * other locked ranges before calling bdev_lock_lba_range_ctx which will start
10459 : * the lock process.
10460 : */
10461 17 : TAILQ_FOREACH_SAFE(range, &bdev->internal.pending_locked_ranges, tailq, tmp) {
10462 3 : if (bdev_lba_range_overlapped(range, &ctx->range) &&
10463 3 : !bdev_lba_range_overlaps_tailq(range, &bdev->internal.locked_ranges)) {
10464 2 : TAILQ_REMOVE(&bdev->internal.pending_locked_ranges, range, tailq);
10465 2 : pending_ctx = SPDK_CONTAINEROF(range, struct locked_lba_range_ctx, range);
10466 2 : TAILQ_INSERT_TAIL(&bdev->internal.locked_ranges, range, tailq);
10467 4 : spdk_thread_send_msg(pending_ctx->range.owner_thread,
10468 2 : bdev_lock_lba_range_ctx_msg, pending_ctx);
10469 2 : }
10470 3 : }
10471 14 : spdk_spin_unlock(&bdev->internal.spinlock);
10472 :
10473 14 : ctx->cb_fn(&ctx->range, ctx->cb_arg, status);
10474 14 : free(ctx);
10475 14 : }
10476 :
10477 : static void
10478 16 : bdev_unlock_lba_range_get_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
10479 : struct spdk_io_channel *_ch, void *_ctx)
10480 : {
10481 16 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
10482 16 : struct locked_lba_range_ctx *ctx = _ctx;
10483 : TAILQ_HEAD(, spdk_bdev_io) io_locked;
10484 : struct spdk_bdev_io *bdev_io;
10485 : struct lba_range *range;
10486 :
10487 16 : TAILQ_FOREACH(range, &ch->locked_ranges, tailq) {
10488 32 : if (ctx->range.offset == range->offset &&
10489 16 : ctx->range.length == range->length &&
10490 16 : ctx->range.locked_ctx == range->locked_ctx) {
10491 16 : TAILQ_REMOVE(&ch->locked_ranges, range, tailq);
10492 16 : free(range);
10493 16 : break;
10494 : }
10495 0 : }
10496 :
10497 : /* Note: we should almost always be able to assert that the range specified
10498 : * was found. But there are some very rare corner cases where a new channel
10499 : * gets created simultaneously with a range unlock, where this function
10500 : * would execute on that new channel and wouldn't have the range.
10501 : * We also use this to clean up range allocations when a later allocation
10502 : * fails in the locking path.
10503 : * So we can't actually assert() here.
10504 : */
10505 :
10506 : /* Swap the locked IO into a temporary list, and then try to submit them again.
10507 : * We could hyper-optimize this to only resubmit locked I/O that overlap
10508 : * with the range that was just unlocked, but this isn't a performance path so
10509 : * we go for simplicity here.
10510 : */
10511 16 : TAILQ_INIT(&io_locked);
10512 16 : TAILQ_SWAP(&ch->io_locked, &io_locked, spdk_bdev_io, internal.ch_link);
10513 19 : while (!TAILQ_EMPTY(&io_locked)) {
10514 3 : bdev_io = TAILQ_FIRST(&io_locked);
10515 3 : TAILQ_REMOVE(&io_locked, bdev_io, internal.ch_link);
10516 3 : bdev_io_submit(bdev_io);
10517 : }
10518 :
10519 16 : spdk_bdev_for_each_channel_continue(i, 0);
10520 16 : }
10521 :
10522 : static int
10523 14 : _bdev_unlock_lba_range(struct spdk_bdev *bdev, uint64_t offset, uint64_t length,
10524 : lock_range_cb cb_fn, void *cb_arg)
10525 : {
10526 : struct locked_lba_range_ctx *ctx;
10527 : struct lba_range *range;
10528 :
10529 14 : spdk_spin_lock(&bdev->internal.spinlock);
10530 : /* To start the unlock the process, we find the range in the bdev's locked_ranges
10531 : * and remove it. This ensures new channels don't inherit the locked range.
10532 : * Then we will send a message to each channel to remove the range from its
10533 : * per-channel list.
10534 : */
10535 14 : TAILQ_FOREACH(range, &bdev->internal.locked_ranges, tailq) {
10536 24 : if (range->offset == offset && range->length == length &&
10537 14 : (range->owner_ch == NULL || range->locked_ctx == cb_arg)) {
10538 14 : break;
10539 : }
10540 0 : }
10541 14 : if (range == NULL) {
10542 0 : assert(false);
10543 : spdk_spin_unlock(&bdev->internal.spinlock);
10544 : return -EINVAL;
10545 : }
10546 14 : TAILQ_REMOVE(&bdev->internal.locked_ranges, range, tailq);
10547 14 : ctx = SPDK_CONTAINEROF(range, struct locked_lba_range_ctx, range);
10548 14 : spdk_spin_unlock(&bdev->internal.spinlock);
10549 :
10550 14 : ctx->cb_fn = cb_fn;
10551 14 : ctx->cb_arg = cb_arg;
10552 :
10553 14 : spdk_bdev_for_each_channel(bdev, bdev_unlock_lba_range_get_channel, ctx,
10554 : bdev_unlock_lba_range_cb);
10555 14 : return 0;
10556 : }
10557 :
10558 : static int
10559 12 : bdev_unlock_lba_range(struct spdk_bdev_desc *desc, struct spdk_io_channel *_ch,
10560 : uint64_t offset, uint64_t length,
10561 : lock_range_cb cb_fn, void *cb_arg)
10562 : {
10563 12 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
10564 12 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
10565 : struct lba_range *range;
10566 12 : bool range_found = false;
10567 :
10568 : /* Let's make sure the specified channel actually has a lock on
10569 : * the specified range. Note that the range must match exactly.
10570 : */
10571 14 : TAILQ_FOREACH(range, &ch->locked_ranges, tailq) {
10572 22 : if (range->offset == offset && range->length == length &&
10573 11 : range->owner_ch == ch && range->locked_ctx == cb_arg) {
10574 10 : range_found = true;
10575 10 : break;
10576 : }
10577 2 : }
10578 :
10579 12 : if (!range_found) {
10580 2 : return -EINVAL;
10581 : }
10582 :
10583 10 : return _bdev_unlock_lba_range(bdev, offset, length, cb_fn, cb_arg);
10584 12 : }
10585 :
10586 : struct bdev_quiesce_ctx {
10587 : spdk_bdev_quiesce_cb cb_fn;
10588 : void *cb_arg;
10589 : };
10590 :
10591 : static void
10592 4 : bdev_unquiesce_range_unlocked(struct lba_range *range, void *ctx, int status)
10593 : {
10594 4 : struct bdev_quiesce_ctx *quiesce_ctx = ctx;
10595 :
10596 4 : if (quiesce_ctx->cb_fn != NULL) {
10597 4 : quiesce_ctx->cb_fn(quiesce_ctx->cb_arg, status);
10598 4 : }
10599 :
10600 4 : free(quiesce_ctx);
10601 4 : }
10602 :
10603 : static void
10604 4 : bdev_quiesce_range_locked(struct lba_range *range, void *ctx, int status)
10605 : {
10606 4 : struct bdev_quiesce_ctx *quiesce_ctx = ctx;
10607 4 : struct spdk_bdev_module *module = range->bdev->module;
10608 :
10609 4 : if (status != 0) {
10610 0 : if (quiesce_ctx->cb_fn != NULL) {
10611 0 : quiesce_ctx->cb_fn(quiesce_ctx->cb_arg, status);
10612 0 : }
10613 0 : free(quiesce_ctx);
10614 0 : return;
10615 : }
10616 :
10617 4 : spdk_spin_lock(&module->internal.spinlock);
10618 4 : TAILQ_INSERT_TAIL(&module->internal.quiesced_ranges, range, tailq_module);
10619 4 : spdk_spin_unlock(&module->internal.spinlock);
10620 :
10621 4 : if (quiesce_ctx->cb_fn != NULL) {
10622 : /* copy the context in case the range is unlocked by the callback */
10623 4 : struct bdev_quiesce_ctx tmp = *quiesce_ctx;
10624 :
10625 4 : quiesce_ctx->cb_fn = NULL;
10626 4 : quiesce_ctx->cb_arg = NULL;
10627 :
10628 4 : tmp.cb_fn(tmp.cb_arg, status);
10629 4 : }
10630 : /* quiesce_ctx will be freed on unquiesce */
10631 4 : }
10632 :
10633 : static int
10634 9 : _spdk_bdev_quiesce(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
10635 : uint64_t offset, uint64_t length,
10636 : spdk_bdev_quiesce_cb cb_fn, void *cb_arg,
10637 : bool unquiesce)
10638 : {
10639 : struct bdev_quiesce_ctx *quiesce_ctx;
10640 : int rc;
10641 :
10642 9 : if (module != bdev->module) {
10643 0 : SPDK_ERRLOG("Bdev does not belong to specified module.\n");
10644 0 : return -EINVAL;
10645 : }
10646 :
10647 9 : if (!bdev_io_valid_blocks(bdev, offset, length)) {
10648 0 : return -EINVAL;
10649 : }
10650 :
10651 9 : if (unquiesce) {
10652 : struct lba_range *range;
10653 :
10654 : /* Make sure the specified range is actually quiesced in the specified module and
10655 : * then remove it from the list. Note that the range must match exactly.
10656 : */
10657 5 : spdk_spin_lock(&module->internal.spinlock);
10658 6 : TAILQ_FOREACH(range, &module->internal.quiesced_ranges, tailq_module) {
10659 5 : if (range->bdev == bdev && range->offset == offset && range->length == length) {
10660 4 : TAILQ_REMOVE(&module->internal.quiesced_ranges, range, tailq_module);
10661 4 : break;
10662 : }
10663 1 : }
10664 5 : spdk_spin_unlock(&module->internal.spinlock);
10665 :
10666 5 : if (range == NULL) {
10667 1 : SPDK_ERRLOG("The range to unquiesce was not found.\n");
10668 1 : return -EINVAL;
10669 : }
10670 :
10671 4 : quiesce_ctx = range->locked_ctx;
10672 4 : quiesce_ctx->cb_fn = cb_fn;
10673 4 : quiesce_ctx->cb_arg = cb_arg;
10674 :
10675 4 : rc = _bdev_unlock_lba_range(bdev, offset, length, bdev_unquiesce_range_unlocked, quiesce_ctx);
10676 4 : } else {
10677 4 : quiesce_ctx = malloc(sizeof(*quiesce_ctx));
10678 4 : if (quiesce_ctx == NULL) {
10679 0 : return -ENOMEM;
10680 : }
10681 :
10682 4 : quiesce_ctx->cb_fn = cb_fn;
10683 4 : quiesce_ctx->cb_arg = cb_arg;
10684 :
10685 4 : rc = _bdev_lock_lba_range(bdev, NULL, offset, length, bdev_quiesce_range_locked, quiesce_ctx);
10686 4 : if (rc != 0) {
10687 0 : free(quiesce_ctx);
10688 0 : }
10689 : }
10690 :
10691 8 : return rc;
10692 9 : }
10693 :
10694 : int
10695 3 : spdk_bdev_quiesce(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
10696 : spdk_bdev_quiesce_cb cb_fn, void *cb_arg)
10697 : {
10698 3 : return _spdk_bdev_quiesce(bdev, module, 0, bdev->blockcnt, cb_fn, cb_arg, false);
10699 : }
10700 :
10701 : int
10702 3 : spdk_bdev_unquiesce(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
10703 : spdk_bdev_quiesce_cb cb_fn, void *cb_arg)
10704 : {
10705 3 : return _spdk_bdev_quiesce(bdev, module, 0, bdev->blockcnt, cb_fn, cb_arg, true);
10706 : }
10707 :
10708 : int
10709 1 : spdk_bdev_quiesce_range(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
10710 : uint64_t offset, uint64_t length,
10711 : spdk_bdev_quiesce_cb cb_fn, void *cb_arg)
10712 : {
10713 1 : return _spdk_bdev_quiesce(bdev, module, offset, length, cb_fn, cb_arg, false);
10714 : }
10715 :
10716 : int
10717 2 : spdk_bdev_unquiesce_range(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
10718 : uint64_t offset, uint64_t length,
10719 : spdk_bdev_quiesce_cb cb_fn, void *cb_arg)
10720 : {
10721 2 : return _spdk_bdev_quiesce(bdev, module, offset, length, cb_fn, cb_arg, true);
10722 : }
10723 :
10724 : int
10725 273 : spdk_bdev_get_memory_domains(struct spdk_bdev *bdev, struct spdk_memory_domain **domains,
10726 : int array_size)
10727 : {
10728 273 : if (!bdev) {
10729 1 : return -EINVAL;
10730 : }
10731 :
10732 272 : if (bdev->fn_table->get_memory_domains) {
10733 3 : return bdev->fn_table->get_memory_domains(bdev->ctxt, domains, array_size);
10734 : }
10735 :
10736 269 : return 0;
10737 273 : }
10738 :
10739 : struct spdk_bdev_for_each_io_ctx {
10740 : void *ctx;
10741 : spdk_bdev_io_fn fn;
10742 : spdk_bdev_for_each_io_cb cb;
10743 : };
10744 :
10745 : static void
10746 0 : bdev_channel_for_each_io(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
10747 : struct spdk_io_channel *io_ch, void *_ctx)
10748 : {
10749 0 : struct spdk_bdev_for_each_io_ctx *ctx = _ctx;
10750 0 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
10751 : struct spdk_bdev_io *bdev_io;
10752 0 : int rc = 0;
10753 :
10754 0 : TAILQ_FOREACH(bdev_io, &bdev_ch->io_submitted, internal.ch_link) {
10755 0 : rc = ctx->fn(ctx->ctx, bdev_io);
10756 0 : if (rc != 0) {
10757 0 : break;
10758 : }
10759 0 : }
10760 :
10761 0 : spdk_bdev_for_each_channel_continue(i, rc);
10762 0 : }
10763 :
10764 : static void
10765 0 : bdev_for_each_io_done(struct spdk_bdev *bdev, void *_ctx, int status)
10766 : {
10767 0 : struct spdk_bdev_for_each_io_ctx *ctx = _ctx;
10768 :
10769 0 : ctx->cb(ctx->ctx, status);
10770 :
10771 0 : free(ctx);
10772 0 : }
10773 :
10774 : void
10775 0 : spdk_bdev_for_each_bdev_io(struct spdk_bdev *bdev, void *_ctx, spdk_bdev_io_fn fn,
10776 : spdk_bdev_for_each_io_cb cb)
10777 : {
10778 : struct spdk_bdev_for_each_io_ctx *ctx;
10779 :
10780 0 : assert(fn != NULL && cb != NULL);
10781 :
10782 0 : ctx = calloc(1, sizeof(*ctx));
10783 0 : if (ctx == NULL) {
10784 0 : SPDK_ERRLOG("Failed to allocate context.\n");
10785 0 : cb(_ctx, -ENOMEM);
10786 0 : return;
10787 : }
10788 :
10789 0 : ctx->ctx = _ctx;
10790 0 : ctx->fn = fn;
10791 0 : ctx->cb = cb;
10792 :
10793 0 : spdk_bdev_for_each_channel(bdev, bdev_channel_for_each_io, ctx,
10794 : bdev_for_each_io_done);
10795 0 : }
10796 :
10797 : void
10798 132 : spdk_bdev_for_each_channel_continue(struct spdk_bdev_channel_iter *iter, int status)
10799 : {
10800 132 : spdk_for_each_channel_continue(iter->i, status);
10801 132 : }
10802 :
10803 : static struct spdk_bdev *
10804 362 : io_channel_iter_get_bdev(struct spdk_io_channel_iter *i)
10805 : {
10806 362 : void *io_device = spdk_io_channel_iter_get_io_device(i);
10807 :
10808 362 : return __bdev_from_io_dev(io_device);
10809 : }
10810 :
10811 : static void
10812 132 : bdev_each_channel_msg(struct spdk_io_channel_iter *i)
10813 : {
10814 132 : struct spdk_bdev_channel_iter *iter = spdk_io_channel_iter_get_ctx(i);
10815 132 : struct spdk_bdev *bdev = io_channel_iter_get_bdev(i);
10816 132 : struct spdk_io_channel *ch = spdk_io_channel_iter_get_channel(i);
10817 :
10818 132 : iter->i = i;
10819 132 : iter->fn(iter, bdev, ch, iter->ctx);
10820 132 : }
10821 :
10822 : static void
10823 230 : bdev_each_channel_cpl(struct spdk_io_channel_iter *i, int status)
10824 : {
10825 230 : struct spdk_bdev_channel_iter *iter = spdk_io_channel_iter_get_ctx(i);
10826 230 : struct spdk_bdev *bdev = io_channel_iter_get_bdev(i);
10827 :
10828 230 : iter->i = i;
10829 230 : iter->cpl(bdev, iter->ctx, status);
10830 :
10831 230 : free(iter);
10832 230 : }
10833 :
10834 : void
10835 230 : spdk_bdev_for_each_channel(struct spdk_bdev *bdev, spdk_bdev_for_each_channel_msg fn,
10836 : void *ctx, spdk_bdev_for_each_channel_done cpl)
10837 : {
10838 : struct spdk_bdev_channel_iter *iter;
10839 :
10840 230 : assert(bdev != NULL && fn != NULL && ctx != NULL);
10841 :
10842 230 : iter = calloc(1, sizeof(struct spdk_bdev_channel_iter));
10843 230 : if (iter == NULL) {
10844 0 : SPDK_ERRLOG("Unable to allocate iterator\n");
10845 0 : assert(false);
10846 : return;
10847 : }
10848 :
10849 230 : iter->fn = fn;
10850 230 : iter->cpl = cpl;
10851 230 : iter->ctx = ctx;
10852 :
10853 460 : spdk_for_each_channel(__bdev_to_io_dev(bdev), bdev_each_channel_msg,
10854 230 : iter, bdev_each_channel_cpl);
10855 230 : }
10856 :
10857 : static void
10858 3 : bdev_copy_do_write_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
10859 : {
10860 3 : struct spdk_bdev_io *parent_io = cb_arg;
10861 :
10862 3 : spdk_bdev_free_io(bdev_io);
10863 :
10864 : /* Check return status of write */
10865 3 : parent_io->internal.status = success ? SPDK_BDEV_IO_STATUS_SUCCESS : SPDK_BDEV_IO_STATUS_FAILED;
10866 3 : parent_io->internal.cb(parent_io, success, parent_io->internal.caller_ctx);
10867 3 : }
10868 :
10869 : static void
10870 3 : bdev_copy_do_write(void *_bdev_io)
10871 : {
10872 3 : struct spdk_bdev_io *bdev_io = _bdev_io;
10873 : int rc;
10874 :
10875 : /* Write blocks */
10876 6 : rc = spdk_bdev_write_blocks_with_md(bdev_io->internal.desc,
10877 3 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
10878 3 : bdev_io->u.bdev.iovs[0].iov_base,
10879 3 : bdev_io->u.bdev.md_buf, bdev_io->u.bdev.offset_blocks,
10880 3 : bdev_io->u.bdev.num_blocks, bdev_copy_do_write_done, bdev_io);
10881 :
10882 3 : if (rc == -ENOMEM) {
10883 0 : bdev_queue_io_wait_with_cb(bdev_io, bdev_copy_do_write);
10884 3 : } else if (rc != 0) {
10885 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
10886 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
10887 0 : }
10888 3 : }
10889 :
10890 : static void
10891 3 : bdev_copy_do_read_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
10892 : {
10893 3 : struct spdk_bdev_io *parent_io = cb_arg;
10894 :
10895 3 : spdk_bdev_free_io(bdev_io);
10896 :
10897 : /* Check return status of read */
10898 3 : if (!success) {
10899 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
10900 0 : parent_io->internal.cb(parent_io, false, parent_io->internal.caller_ctx);
10901 0 : return;
10902 : }
10903 :
10904 : /* Do write */
10905 3 : bdev_copy_do_write(parent_io);
10906 3 : }
10907 :
10908 : static void
10909 3 : bdev_copy_do_read(void *_bdev_io)
10910 : {
10911 3 : struct spdk_bdev_io *bdev_io = _bdev_io;
10912 : int rc;
10913 :
10914 : /* Read blocks */
10915 6 : rc = spdk_bdev_read_blocks_with_md(bdev_io->internal.desc,
10916 3 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
10917 3 : bdev_io->u.bdev.iovs[0].iov_base,
10918 3 : bdev_io->u.bdev.md_buf, bdev_io->u.bdev.copy.src_offset_blocks,
10919 3 : bdev_io->u.bdev.num_blocks, bdev_copy_do_read_done, bdev_io);
10920 :
10921 3 : if (rc == -ENOMEM) {
10922 0 : bdev_queue_io_wait_with_cb(bdev_io, bdev_copy_do_read);
10923 3 : } else if (rc != 0) {
10924 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
10925 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
10926 0 : }
10927 3 : }
10928 :
10929 : static void
10930 3 : bdev_copy_get_buf_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io, bool success)
10931 : {
10932 3 : if (!success) {
10933 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
10934 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
10935 0 : return;
10936 : }
10937 :
10938 3 : bdev_copy_do_read(bdev_io);
10939 3 : }
10940 :
10941 : int
10942 27 : spdk_bdev_copy_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
10943 : uint64_t dst_offset_blocks, uint64_t src_offset_blocks, uint64_t num_blocks,
10944 : spdk_bdev_io_completion_cb cb, void *cb_arg)
10945 : {
10946 27 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
10947 : struct spdk_bdev_io *bdev_io;
10948 27 : struct spdk_bdev_channel *channel = spdk_io_channel_get_ctx(ch);
10949 :
10950 27 : if (!desc->write) {
10951 0 : return -EBADF;
10952 : }
10953 :
10954 27 : if (!bdev_io_valid_blocks(bdev, dst_offset_blocks, num_blocks) ||
10955 27 : !bdev_io_valid_blocks(bdev, src_offset_blocks, num_blocks)) {
10956 0 : SPDK_DEBUGLOG(bdev,
10957 : "Invalid offset or number of blocks: dst %lu, src %lu, count %lu\n",
10958 : dst_offset_blocks, src_offset_blocks, num_blocks);
10959 0 : return -EINVAL;
10960 : }
10961 :
10962 27 : bdev_io = bdev_channel_get_io(channel);
10963 27 : if (!bdev_io) {
10964 0 : return -ENOMEM;
10965 : }
10966 :
10967 27 : bdev_io->internal.ch = channel;
10968 27 : bdev_io->internal.desc = desc;
10969 27 : bdev_io->type = SPDK_BDEV_IO_TYPE_COPY;
10970 :
10971 27 : bdev_io->u.bdev.offset_blocks = dst_offset_blocks;
10972 27 : bdev_io->u.bdev.copy.src_offset_blocks = src_offset_blocks;
10973 27 : bdev_io->u.bdev.num_blocks = num_blocks;
10974 27 : bdev_io->u.bdev.memory_domain = NULL;
10975 27 : bdev_io->u.bdev.memory_domain_ctx = NULL;
10976 27 : bdev_io->u.bdev.iovs = NULL;
10977 27 : bdev_io->u.bdev.iovcnt = 0;
10978 27 : bdev_io->u.bdev.md_buf = NULL;
10979 27 : bdev_io->u.bdev.accel_sequence = NULL;
10980 27 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
10981 :
10982 27 : if (dst_offset_blocks == src_offset_blocks || num_blocks == 0) {
10983 0 : spdk_thread_send_msg(spdk_get_thread(), bdev_io_complete_cb, bdev_io);
10984 0 : return 0;
10985 : }
10986 :
10987 :
10988 : /* If the copy size is large and should be split, use the generic split logic
10989 : * regardless of whether SPDK_BDEV_IO_TYPE_COPY is supported or not.
10990 : *
10991 : * Then, send the copy request if SPDK_BDEV_IO_TYPE_COPY is supported or
10992 : * emulate it using regular read and write requests otherwise.
10993 : */
10994 27 : if (spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COPY) ||
10995 4 : bdev_io->internal.f.split) {
10996 24 : bdev_io_submit(bdev_io);
10997 24 : return 0;
10998 : }
10999 :
11000 3 : spdk_bdev_io_get_buf(bdev_io, bdev_copy_get_buf_cb, num_blocks * spdk_bdev_get_block_size(bdev));
11001 :
11002 3 : return 0;
11003 27 : }
11004 :
11005 3 : SPDK_LOG_REGISTER_COMPONENT(bdev)
11006 :
11007 : static void
11008 0 : bdev_trace(void)
11009 : {
11010 0 : struct spdk_trace_tpoint_opts opts[] = {
11011 : {
11012 : "BDEV_IO_START", TRACE_BDEV_IO_START,
11013 : OWNER_TYPE_BDEV, OBJECT_BDEV_IO, 1,
11014 : {
11015 : { "type", SPDK_TRACE_ARG_TYPE_INT, 8 },
11016 : { "ctx", SPDK_TRACE_ARG_TYPE_PTR, 8 },
11017 : { "offset", SPDK_TRACE_ARG_TYPE_INT, 8 },
11018 : { "qd", SPDK_TRACE_ARG_TYPE_INT, 4 }
11019 : }
11020 : },
11021 : {
11022 : "BDEV_IO_DONE", TRACE_BDEV_IO_DONE,
11023 : OWNER_TYPE_BDEV, OBJECT_BDEV_IO, 0,
11024 : {
11025 : { "ctx", SPDK_TRACE_ARG_TYPE_PTR, 8 },
11026 : { "qd", SPDK_TRACE_ARG_TYPE_INT, 4 }
11027 : }
11028 : },
11029 : {
11030 : "BDEV_IOCH_CREATE", TRACE_BDEV_IOCH_CREATE,
11031 : OWNER_TYPE_BDEV, OBJECT_NONE, 0,
11032 : {
11033 : { "tid", SPDK_TRACE_ARG_TYPE_INT, 8 }
11034 : }
11035 : },
11036 : {
11037 : "BDEV_IOCH_DESTROY", TRACE_BDEV_IOCH_DESTROY,
11038 : OWNER_TYPE_BDEV, OBJECT_NONE, 0,
11039 : {
11040 : { "tid", SPDK_TRACE_ARG_TYPE_INT, 8 }
11041 : }
11042 : },
11043 : };
11044 :
11045 :
11046 0 : spdk_trace_register_owner_type(OWNER_TYPE_BDEV, 'b');
11047 0 : spdk_trace_register_object(OBJECT_BDEV_IO, 'i');
11048 0 : spdk_trace_register_description_ext(opts, SPDK_COUNTOF(opts));
11049 0 : spdk_trace_tpoint_register_relation(TRACE_BDEV_NVME_IO_START, OBJECT_BDEV_IO, 0);
11050 0 : spdk_trace_tpoint_register_relation(TRACE_BDEV_NVME_IO_DONE, OBJECT_BDEV_IO, 0);
11051 0 : spdk_trace_tpoint_register_relation(TRACE_BLOB_REQ_SET_START, OBJECT_BDEV_IO, 0);
11052 0 : spdk_trace_tpoint_register_relation(TRACE_BLOB_REQ_SET_COMPLETE, OBJECT_BDEV_IO, 0);
11053 0 : spdk_trace_tpoint_register_relation(TRACE_BDEV_RAID_IO_START, OBJECT_BDEV_IO, 0);
11054 0 : spdk_trace_tpoint_register_relation(TRACE_BDEV_RAID_IO_DONE, OBJECT_BDEV_IO, 0);
11055 0 : }
11056 3 : SPDK_TRACE_REGISTER_FN(bdev_trace, "bdev", TRACE_GROUP_BDEV)
|