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
86 568 : bdev_name_cmp(struct spdk_bdev_name *name1, struct spdk_bdev_name *name2)
87 : {
88 568 : return strcmp(name1->name, name2->name);
89 : }
90 :
91 2121 : RB_GENERATE_STATIC(bdev_name_tree, spdk_bdev_name, node, bdev_name_cmp);
92 :
93 : struct spdk_bdev_mgr {
94 : struct spdk_mempool *bdev_io_pool;
95 :
96 : void *zero_buffer;
97 :
98 : TAILQ_HEAD(bdev_module_list, spdk_bdev_module) bdev_modules;
99 :
100 : struct spdk_bdev_list bdevs;
101 : struct bdev_name_tree bdev_names;
102 :
103 : bool init_complete;
104 : bool module_init_complete;
105 :
106 : struct spdk_spinlock spinlock;
107 :
108 : TAILQ_HEAD(, spdk_bdev_open_async_ctx) async_bdev_opens;
109 :
110 : #ifdef SPDK_CONFIG_VTUNE
111 : __itt_domain *domain;
112 : #endif
113 : };
114 :
115 : static struct spdk_bdev_mgr g_bdev_mgr = {
116 : .bdev_modules = TAILQ_HEAD_INITIALIZER(g_bdev_mgr.bdev_modules),
117 : .bdevs = TAILQ_HEAD_INITIALIZER(g_bdev_mgr.bdevs),
118 : .bdev_names = RB_INITIALIZER(g_bdev_mgr.bdev_names),
119 : .init_complete = false,
120 : .module_init_complete = false,
121 : .async_bdev_opens = TAILQ_HEAD_INITIALIZER(g_bdev_mgr.async_bdev_opens),
122 : };
123 :
124 : static void
125 : __attribute__((constructor))
126 3 : _bdev_init(void)
127 : {
128 3 : spdk_spin_init(&g_bdev_mgr.spinlock);
129 3 : }
130 :
131 : typedef void (*lock_range_cb)(struct lba_range *range, void *ctx, int status);
132 :
133 : typedef void (*bdev_copy_bounce_buffer_cpl)(void *ctx, int rc);
134 :
135 : struct lba_range {
136 : struct spdk_bdev *bdev;
137 : uint64_t offset;
138 : uint64_t length;
139 : bool quiesce;
140 : void *locked_ctx;
141 : struct spdk_thread *owner_thread;
142 : struct spdk_bdev_channel *owner_ch;
143 : TAILQ_ENTRY(lba_range) tailq;
144 : TAILQ_ENTRY(lba_range) tailq_module;
145 : };
146 :
147 : static struct spdk_bdev_opts g_bdev_opts = {
148 : .bdev_io_pool_size = SPDK_BDEV_IO_POOL_SIZE,
149 : .bdev_io_cache_size = SPDK_BDEV_IO_CACHE_SIZE,
150 : .bdev_auto_examine = SPDK_BDEV_AUTO_EXAMINE,
151 : .iobuf_small_cache_size = BUF_SMALL_CACHE_SIZE,
152 : .iobuf_large_cache_size = BUF_LARGE_CACHE_SIZE,
153 : };
154 :
155 : static spdk_bdev_init_cb g_init_cb_fn = NULL;
156 : static void *g_init_cb_arg = NULL;
157 :
158 : static spdk_bdev_fini_cb g_fini_cb_fn = NULL;
159 : static void *g_fini_cb_arg = NULL;
160 : static struct spdk_thread *g_fini_thread = NULL;
161 :
162 : struct spdk_bdev_qos_limit {
163 : /** IOs or bytes allowed per second (i.e., 1s). */
164 : uint64_t limit;
165 :
166 : /** Remaining IOs or bytes allowed in current timeslice (e.g., 1ms).
167 : * For remaining bytes, allowed to run negative if an I/O is submitted when
168 : * some bytes are remaining, but the I/O is bigger than that amount. The
169 : * excess will be deducted from the next timeslice.
170 : */
171 : int64_t remaining_this_timeslice;
172 :
173 : /** Minimum allowed IOs or bytes to be issued in one timeslice (e.g., 1ms). */
174 : uint32_t min_per_timeslice;
175 :
176 : /** Maximum allowed IOs or bytes to be issued in one timeslice (e.g., 1ms). */
177 : uint32_t max_per_timeslice;
178 :
179 : /** Function to check whether to queue the IO.
180 : * If The IO is allowed to pass, the quota will be reduced correspondingly.
181 : */
182 : bool (*queue_io)(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io);
183 :
184 : /** Function to rewind the quota once the IO was allowed to be sent by this
185 : * limit but queued due to one of the further limits.
186 : */
187 : void (*rewind_quota)(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io);
188 : };
189 :
190 : struct spdk_bdev_qos {
191 : /** Types of structure of rate limits. */
192 : struct spdk_bdev_qos_limit rate_limits[SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES];
193 :
194 : /** The channel that all I/O are funneled through. */
195 : struct spdk_bdev_channel *ch;
196 :
197 : /** The thread on which the poller is running. */
198 : struct spdk_thread *thread;
199 :
200 : /** Size of a timeslice in tsc ticks. */
201 : uint64_t timeslice_size;
202 :
203 : /** Timestamp of start of last timeslice. */
204 : uint64_t last_timeslice;
205 :
206 : /** Poller that processes queued I/O commands each time slice. */
207 : struct spdk_poller *poller;
208 : };
209 :
210 : struct spdk_bdev_mgmt_channel {
211 : /*
212 : * Each thread keeps a cache of bdev_io - this allows
213 : * bdev threads which are *not* DPDK threads to still
214 : * benefit from a per-thread bdev_io cache. Without
215 : * this, non-DPDK threads fetching from the mempool
216 : * incur a cmpxchg on get and put.
217 : */
218 : bdev_io_stailq_t per_thread_cache;
219 : uint32_t per_thread_cache_count;
220 : uint32_t bdev_io_cache_size;
221 :
222 : struct spdk_iobuf_channel iobuf;
223 :
224 : TAILQ_HEAD(, spdk_bdev_shared_resource) shared_resources;
225 : TAILQ_HEAD(, spdk_bdev_io_wait_entry) io_wait_queue;
226 : };
227 :
228 : /*
229 : * Per-module (or per-io_device) data. Multiple bdevs built on the same io_device
230 : * will queue here their IO that awaits retry. It makes it possible to retry sending
231 : * IO to one bdev after IO from other bdev completes.
232 : */
233 : struct spdk_bdev_shared_resource {
234 : /* The bdev management channel */
235 : struct spdk_bdev_mgmt_channel *mgmt_ch;
236 :
237 : /*
238 : * Count of I/O submitted to bdev module and waiting for completion.
239 : * Incremented before submit_request() is called on an spdk_bdev_io.
240 : */
241 : uint64_t io_outstanding;
242 :
243 : /*
244 : * Queue of IO awaiting retry because of a previous NOMEM status returned
245 : * on this channel.
246 : */
247 : bdev_io_tailq_t nomem_io;
248 :
249 : /*
250 : * Threshold which io_outstanding must drop to before retrying nomem_io.
251 : */
252 : uint64_t nomem_threshold;
253 :
254 : /* I/O channel allocated by a bdev module */
255 : struct spdk_io_channel *shared_ch;
256 :
257 : struct spdk_poller *nomem_poller;
258 :
259 : /* Refcount of bdev channels using this resource */
260 : uint32_t ref;
261 :
262 : TAILQ_ENTRY(spdk_bdev_shared_resource) link;
263 : };
264 :
265 : #define BDEV_CH_RESET_IN_PROGRESS (1 << 0)
266 : #define BDEV_CH_QOS_ENABLED (1 << 1)
267 :
268 : struct spdk_bdev_channel {
269 : struct spdk_bdev *bdev;
270 :
271 : /* The channel for the underlying device */
272 : struct spdk_io_channel *channel;
273 :
274 : /* Accel channel */
275 : struct spdk_io_channel *accel_channel;
276 :
277 : /* Per io_device per thread data */
278 : struct spdk_bdev_shared_resource *shared_resource;
279 :
280 : struct spdk_bdev_io_stat *stat;
281 :
282 : /*
283 : * Count of I/O submitted to the underlying dev module through this channel
284 : * and waiting for completion.
285 : */
286 : uint64_t io_outstanding;
287 :
288 : /*
289 : * List of all submitted I/Os including I/O that are generated via splitting.
290 : */
291 : bdev_io_tailq_t io_submitted;
292 :
293 : /*
294 : * List of spdk_bdev_io that are currently queued because they write to a locked
295 : * LBA range.
296 : */
297 : bdev_io_tailq_t io_locked;
298 :
299 : /* List of I/Os with accel sequence being currently executed */
300 : bdev_io_tailq_t io_accel_exec;
301 :
302 : /* List of I/Os doing memory domain pull/push */
303 : bdev_io_tailq_t io_memory_domain;
304 :
305 : uint32_t flags;
306 :
307 : /* Counts number of bdev_io in the io_submitted TAILQ */
308 : uint16_t queue_depth;
309 :
310 : uint16_t trace_id;
311 :
312 : struct spdk_histogram_data *histogram;
313 :
314 : #ifdef SPDK_CONFIG_VTUNE
315 : uint64_t start_tsc;
316 : uint64_t interval_tsc;
317 : __itt_string_handle *handle;
318 : struct spdk_bdev_io_stat *prev_stat;
319 : #endif
320 :
321 : lba_range_tailq_t locked_ranges;
322 :
323 : /** List of I/Os queued by QoS. */
324 : bdev_io_tailq_t qos_queued_io;
325 : };
326 :
327 : struct media_event_entry {
328 : struct spdk_bdev_media_event event;
329 : TAILQ_ENTRY(media_event_entry) tailq;
330 : };
331 :
332 : #define MEDIA_EVENT_POOL_SIZE 64
333 :
334 : struct spdk_bdev_desc {
335 : struct spdk_bdev *bdev;
336 : bool write;
337 : bool memory_domains_supported;
338 : bool accel_sequence_supported[SPDK_BDEV_NUM_IO_TYPES];
339 : struct spdk_bdev_open_opts opts;
340 : struct spdk_thread *thread;
341 : struct {
342 : spdk_bdev_event_cb_t event_fn;
343 : void *ctx;
344 : } callback;
345 : bool closed;
346 : struct spdk_spinlock spinlock;
347 : uint32_t refs;
348 : TAILQ_HEAD(, media_event_entry) pending_media_events;
349 : TAILQ_HEAD(, media_event_entry) free_media_events;
350 : struct media_event_entry *media_events_buffer;
351 : TAILQ_ENTRY(spdk_bdev_desc) link;
352 :
353 : uint64_t timeout_in_sec;
354 : spdk_bdev_io_timeout_cb cb_fn;
355 : void *cb_arg;
356 : struct spdk_poller *io_timeout_poller;
357 : struct spdk_bdev_module_claim *claim;
358 : };
359 :
360 : struct spdk_bdev_iostat_ctx {
361 : struct spdk_bdev_io_stat *stat;
362 : enum spdk_bdev_reset_stat_mode reset_mode;
363 : spdk_bdev_get_device_stat_cb cb;
364 : void *cb_arg;
365 : };
366 :
367 : struct set_qos_limit_ctx {
368 : void (*cb_fn)(void *cb_arg, int status);
369 : void *cb_arg;
370 : struct spdk_bdev *bdev;
371 : };
372 :
373 : struct spdk_bdev_channel_iter {
374 : spdk_bdev_for_each_channel_msg fn;
375 : spdk_bdev_for_each_channel_done cpl;
376 : struct spdk_io_channel_iter *i;
377 : void *ctx;
378 : };
379 :
380 : struct spdk_bdev_io_error_stat {
381 : uint32_t error_status[-SPDK_MIN_BDEV_IO_STATUS];
382 : };
383 :
384 : enum bdev_io_retry_state {
385 : BDEV_IO_RETRY_STATE_INVALID,
386 : BDEV_IO_RETRY_STATE_PULL,
387 : BDEV_IO_RETRY_STATE_PULL_MD,
388 : BDEV_IO_RETRY_STATE_SUBMIT,
389 : BDEV_IO_RETRY_STATE_PUSH,
390 : BDEV_IO_RETRY_STATE_PUSH_MD,
391 : BDEV_IO_RETRY_STATE_GET_ACCEL_BUF,
392 : };
393 :
394 : #define __bdev_to_io_dev(bdev) (((char *)bdev) + 1)
395 : #define __bdev_from_io_dev(io_dev) ((struct spdk_bdev *)(((char *)io_dev) - 1))
396 : #define __io_ch_to_bdev_ch(io_ch) ((struct spdk_bdev_channel *)spdk_io_channel_get_ctx(io_ch))
397 : #define __io_ch_to_bdev_mgmt_ch(io_ch) ((struct spdk_bdev_mgmt_channel *)spdk_io_channel_get_ctx(io_ch))
398 :
399 : static inline void bdev_io_complete(void *ctx);
400 : static inline void bdev_io_complete_unsubmitted(struct spdk_bdev_io *bdev_io);
401 : static void bdev_io_push_bounce_md_buf(struct spdk_bdev_io *bdev_io);
402 : static void bdev_io_push_bounce_data(struct spdk_bdev_io *bdev_io);
403 : static void _bdev_io_get_accel_buf(struct spdk_bdev_io *bdev_io);
404 :
405 : static void bdev_write_zero_buffer_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg);
406 : static int bdev_write_zero_buffer(struct spdk_bdev_io *bdev_io);
407 :
408 : static void bdev_enable_qos_msg(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
409 : struct spdk_io_channel *ch, void *_ctx);
410 : static void bdev_enable_qos_done(struct spdk_bdev *bdev, void *_ctx, int status);
411 :
412 : static int bdev_readv_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
413 : struct iovec *iov, int iovcnt, void *md_buf, uint64_t offset_blocks,
414 : uint64_t num_blocks,
415 : struct spdk_memory_domain *domain, void *domain_ctx,
416 : struct spdk_accel_sequence *seq, uint32_t dif_check_flags,
417 : spdk_bdev_io_completion_cb cb, void *cb_arg);
418 : static int bdev_writev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
419 : struct iovec *iov, int iovcnt, void *md_buf,
420 : uint64_t offset_blocks, uint64_t num_blocks,
421 : struct spdk_memory_domain *domain, void *domain_ctx,
422 : struct spdk_accel_sequence *seq, uint32_t dif_check_flags,
423 : uint32_t nvme_cdw12_raw, uint32_t nvme_cdw13_raw,
424 : spdk_bdev_io_completion_cb cb, void *cb_arg);
425 :
426 : static int bdev_lock_lba_range(struct spdk_bdev_desc *desc, struct spdk_io_channel *_ch,
427 : uint64_t offset, uint64_t length,
428 : lock_range_cb cb_fn, void *cb_arg);
429 :
430 : static int bdev_unlock_lba_range(struct spdk_bdev_desc *desc, struct spdk_io_channel *_ch,
431 : uint64_t offset, uint64_t length,
432 : lock_range_cb cb_fn, void *cb_arg);
433 :
434 : static bool bdev_abort_queued_io(bdev_io_tailq_t *queue, struct spdk_bdev_io *bio_to_abort);
435 : static bool bdev_abort_buf_io(struct spdk_bdev_mgmt_channel *ch, struct spdk_bdev_io *bio_to_abort);
436 :
437 : static bool claim_type_is_v2(enum spdk_bdev_claim_type type);
438 : static void bdev_desc_release_claims(struct spdk_bdev_desc *desc);
439 : static void claim_reset(struct spdk_bdev *bdev);
440 :
441 : static void bdev_ch_retry_io(struct spdk_bdev_channel *bdev_ch);
442 :
443 : static bool bdev_io_should_split(struct spdk_bdev_io *bdev_io);
444 :
445 : #define bdev_get_ext_io_opt(opts, field, defval) \
446 : ((opts) != NULL ? SPDK_GET_FIELD(opts, field, defval) : (defval))
447 :
448 : static inline void
449 671 : bdev_ch_add_to_io_submitted(struct spdk_bdev_io *bdev_io)
450 : {
451 671 : TAILQ_INSERT_TAIL(&bdev_io->internal.ch->io_submitted, bdev_io, internal.ch_link);
452 671 : bdev_io->internal.ch->queue_depth++;
453 671 : }
454 :
455 : static inline void
456 671 : bdev_ch_remove_from_io_submitted(struct spdk_bdev_io *bdev_io)
457 : {
458 671 : TAILQ_REMOVE(&bdev_io->internal.ch->io_submitted, bdev_io, internal.ch_link);
459 671 : bdev_io->internal.ch->queue_depth--;
460 671 : }
461 :
462 : void
463 16 : spdk_bdev_get_opts(struct spdk_bdev_opts *opts, size_t opts_size)
464 : {
465 16 : if (!opts) {
466 0 : SPDK_ERRLOG("opts should not be NULL\n");
467 0 : return;
468 : }
469 :
470 16 : if (!opts_size) {
471 0 : SPDK_ERRLOG("opts_size should not be zero value\n");
472 0 : return;
473 : }
474 :
475 16 : opts->opts_size = opts_size;
476 :
477 : #define SET_FIELD(field) \
478 : if (offsetof(struct spdk_bdev_opts, field) + sizeof(opts->field) <= opts_size) { \
479 : opts->field = g_bdev_opts.field; \
480 : } \
481 :
482 16 : SET_FIELD(bdev_io_pool_size);
483 16 : SET_FIELD(bdev_io_cache_size);
484 16 : SET_FIELD(bdev_auto_examine);
485 16 : SET_FIELD(iobuf_small_cache_size);
486 16 : SET_FIELD(iobuf_large_cache_size);
487 :
488 : /* Do not remove this statement, you should always update this statement when you adding a new field,
489 : * and do not forget to add the SET_FIELD statement for your added field. */
490 : SPDK_STATIC_ASSERT(sizeof(struct spdk_bdev_opts) == 32, "Incorrect size");
491 :
492 : #undef SET_FIELD
493 16 : }
494 :
495 : int
496 17 : spdk_bdev_set_opts(struct spdk_bdev_opts *opts)
497 : {
498 : uint32_t min_pool_size;
499 :
500 17 : if (!opts) {
501 0 : SPDK_ERRLOG("opts cannot be NULL\n");
502 0 : return -1;
503 : }
504 :
505 17 : if (!opts->opts_size) {
506 1 : SPDK_ERRLOG("opts_size inside opts cannot be zero value\n");
507 1 : return -1;
508 : }
509 :
510 : /*
511 : * Add 1 to the thread count to account for the extra mgmt_ch that gets created during subsystem
512 : * initialization. A second mgmt_ch will be created on the same thread when the application starts
513 : * but before the deferred put_io_channel event is executed for the first mgmt_ch.
514 : */
515 16 : min_pool_size = opts->bdev_io_cache_size * (spdk_thread_get_count() + 1);
516 16 : if (opts->bdev_io_pool_size < min_pool_size) {
517 0 : SPDK_ERRLOG("bdev_io_pool_size %" PRIu32 " is not compatible with bdev_io_cache_size %" PRIu32
518 : " and %" PRIu32 " threads\n", opts->bdev_io_pool_size, opts->bdev_io_cache_size,
519 : spdk_thread_get_count());
520 0 : SPDK_ERRLOG("bdev_io_pool_size must be at least %" PRIu32 "\n", min_pool_size);
521 0 : return -1;
522 : }
523 :
524 : #define SET_FIELD(field) \
525 : if (offsetof(struct spdk_bdev_opts, field) + sizeof(opts->field) <= opts->opts_size) { \
526 : g_bdev_opts.field = opts->field; \
527 : } \
528 :
529 16 : SET_FIELD(bdev_io_pool_size);
530 16 : SET_FIELD(bdev_io_cache_size);
531 16 : SET_FIELD(bdev_auto_examine);
532 16 : SET_FIELD(iobuf_small_cache_size);
533 16 : SET_FIELD(iobuf_large_cache_size);
534 :
535 16 : g_bdev_opts.opts_size = opts->opts_size;
536 :
537 : #undef SET_FIELD
538 :
539 16 : return 0;
540 17 : }
541 :
542 : static struct spdk_bdev *
543 155 : bdev_get_by_name(const char *bdev_name)
544 : {
545 : struct spdk_bdev_name find;
546 : struct spdk_bdev_name *res;
547 :
548 155 : find.name = (char *)bdev_name;
549 155 : res = RB_FIND(bdev_name_tree, &g_bdev_mgr.bdev_names, &find);
550 155 : if (res != NULL) {
551 148 : return res->bdev;
552 : }
553 :
554 7 : return NULL;
555 155 : }
556 :
557 : struct spdk_bdev *
558 19 : spdk_bdev_get_by_name(const char *bdev_name)
559 : {
560 : struct spdk_bdev *bdev;
561 :
562 19 : spdk_spin_lock(&g_bdev_mgr.spinlock);
563 19 : bdev = bdev_get_by_name(bdev_name);
564 19 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
565 :
566 19 : return bdev;
567 : }
568 :
569 : struct bdev_io_status_string {
570 : enum spdk_bdev_io_status status;
571 : const char *str;
572 : };
573 :
574 : static const struct bdev_io_status_string bdev_io_status_strings[] = {
575 : { SPDK_BDEV_IO_STATUS_AIO_ERROR, "aio_error" },
576 : { SPDK_BDEV_IO_STATUS_ABORTED, "aborted" },
577 : { SPDK_BDEV_IO_STATUS_FIRST_FUSED_FAILED, "first_fused_failed" },
578 : { SPDK_BDEV_IO_STATUS_MISCOMPARE, "miscompare" },
579 : { SPDK_BDEV_IO_STATUS_NOMEM, "nomem" },
580 : { SPDK_BDEV_IO_STATUS_SCSI_ERROR, "scsi_error" },
581 : { SPDK_BDEV_IO_STATUS_NVME_ERROR, "nvme_error" },
582 : { SPDK_BDEV_IO_STATUS_FAILED, "failed" },
583 : { SPDK_BDEV_IO_STATUS_PENDING, "pending" },
584 : { SPDK_BDEV_IO_STATUS_SUCCESS, "success" },
585 : };
586 :
587 : static const char *
588 0 : bdev_io_status_get_string(enum spdk_bdev_io_status status)
589 : {
590 : uint32_t i;
591 :
592 0 : for (i = 0; i < SPDK_COUNTOF(bdev_io_status_strings); i++) {
593 0 : if (bdev_io_status_strings[i].status == status) {
594 0 : return bdev_io_status_strings[i].str;
595 : }
596 0 : }
597 :
598 0 : return "reserved";
599 0 : }
600 :
601 : struct spdk_bdev_wait_for_examine_ctx {
602 : struct spdk_poller *poller;
603 : spdk_bdev_wait_for_examine_cb cb_fn;
604 : void *cb_arg;
605 : };
606 :
607 : static bool bdev_module_all_actions_completed(void);
608 :
609 : static int
610 203 : bdev_wait_for_examine_cb(void *arg)
611 : {
612 203 : struct spdk_bdev_wait_for_examine_ctx *ctx = arg;
613 :
614 203 : if (!bdev_module_all_actions_completed()) {
615 0 : return SPDK_POLLER_IDLE;
616 : }
617 :
618 203 : spdk_poller_unregister(&ctx->poller);
619 203 : ctx->cb_fn(ctx->cb_arg);
620 203 : free(ctx);
621 :
622 203 : return SPDK_POLLER_BUSY;
623 203 : }
624 :
625 : int
626 203 : spdk_bdev_wait_for_examine(spdk_bdev_wait_for_examine_cb cb_fn, void *cb_arg)
627 : {
628 : struct spdk_bdev_wait_for_examine_ctx *ctx;
629 :
630 203 : ctx = calloc(1, sizeof(*ctx));
631 203 : if (ctx == NULL) {
632 0 : return -ENOMEM;
633 : }
634 203 : ctx->cb_fn = cb_fn;
635 203 : ctx->cb_arg = cb_arg;
636 203 : ctx->poller = SPDK_POLLER_REGISTER(bdev_wait_for_examine_cb, ctx, 0);
637 :
638 203 : return 0;
639 203 : }
640 :
641 : struct spdk_bdev_examine_item {
642 : char *name;
643 : TAILQ_ENTRY(spdk_bdev_examine_item) link;
644 : };
645 :
646 : TAILQ_HEAD(spdk_bdev_examine_allowlist, spdk_bdev_examine_item);
647 :
648 : struct spdk_bdev_examine_allowlist g_bdev_examine_allowlist = TAILQ_HEAD_INITIALIZER(
649 : g_bdev_examine_allowlist);
650 :
651 : static inline bool
652 24 : bdev_examine_allowlist_check(const char *name)
653 : {
654 : struct spdk_bdev_examine_item *item;
655 24 : TAILQ_FOREACH(item, &g_bdev_examine_allowlist, link) {
656 3 : if (strcmp(name, item->name) == 0) {
657 3 : return true;
658 : }
659 0 : }
660 21 : return false;
661 24 : }
662 :
663 : static inline void
664 258 : bdev_examine_allowlist_remove(const char *name)
665 : {
666 : struct spdk_bdev_examine_item *item;
667 258 : TAILQ_FOREACH(item, &g_bdev_examine_allowlist, link) {
668 3 : if (strcmp(name, item->name) == 0) {
669 3 : TAILQ_REMOVE(&g_bdev_examine_allowlist, item, link);
670 3 : free(item->name);
671 3 : free(item);
672 3 : break;
673 : }
674 0 : }
675 258 : }
676 :
677 : static inline void
678 68 : bdev_examine_allowlist_free(void)
679 : {
680 : struct spdk_bdev_examine_item *item;
681 68 : while (!TAILQ_EMPTY(&g_bdev_examine_allowlist)) {
682 0 : item = TAILQ_FIRST(&g_bdev_examine_allowlist);
683 0 : TAILQ_REMOVE(&g_bdev_examine_allowlist, item, link);
684 0 : free(item->name);
685 0 : free(item);
686 : }
687 68 : }
688 :
689 : static inline bool
690 12 : bdev_in_examine_allowlist(struct spdk_bdev *bdev)
691 : {
692 : struct spdk_bdev_alias *tmp;
693 12 : if (bdev_examine_allowlist_check(bdev->name)) {
694 3 : return true;
695 : }
696 18 : TAILQ_FOREACH(tmp, &bdev->aliases, tailq) {
697 9 : if (bdev_examine_allowlist_check(tmp->alias.name)) {
698 0 : return true;
699 : }
700 9 : }
701 9 : return false;
702 12 : }
703 :
704 : static inline bool
705 133 : bdev_ok_to_examine(struct spdk_bdev *bdev)
706 : {
707 : /* Some bdevs may not support the READ command.
708 : * Do not try to examine them.
709 : */
710 133 : if (!spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_READ)) {
711 0 : return false;
712 : }
713 :
714 133 : if (g_bdev_opts.bdev_auto_examine) {
715 121 : return true;
716 : } else {
717 12 : return bdev_in_examine_allowlist(bdev);
718 : }
719 133 : }
720 :
721 : static void
722 133 : bdev_examine(struct spdk_bdev *bdev)
723 : {
724 : struct spdk_bdev_module *module;
725 : struct spdk_bdev_module_claim *claim, *tmpclaim;
726 : uint32_t action;
727 :
728 133 : if (!bdev_ok_to_examine(bdev)) {
729 9 : return;
730 : }
731 :
732 506 : TAILQ_FOREACH(module, &g_bdev_mgr.bdev_modules, internal.tailq) {
733 382 : if (module->examine_config) {
734 258 : spdk_spin_lock(&module->internal.spinlock);
735 258 : action = module->internal.action_in_progress;
736 258 : module->internal.action_in_progress++;
737 258 : spdk_spin_unlock(&module->internal.spinlock);
738 258 : module->examine_config(bdev);
739 258 : if (action != module->internal.action_in_progress) {
740 0 : SPDK_ERRLOG("examine_config for module %s did not call "
741 : "spdk_bdev_module_examine_done()\n", module->name);
742 0 : }
743 258 : }
744 382 : }
745 :
746 124 : spdk_spin_lock(&bdev->internal.spinlock);
747 :
748 124 : switch (bdev->internal.claim_type) {
749 : case SPDK_BDEV_CLAIM_NONE:
750 : /* Examine by all bdev modules */
751 466 : TAILQ_FOREACH(module, &g_bdev_mgr.bdev_modules, internal.tailq) {
752 350 : if (module->examine_disk) {
753 225 : spdk_spin_lock(&module->internal.spinlock);
754 225 : module->internal.action_in_progress++;
755 225 : spdk_spin_unlock(&module->internal.spinlock);
756 225 : spdk_spin_unlock(&bdev->internal.spinlock);
757 225 : module->examine_disk(bdev);
758 225 : spdk_spin_lock(&bdev->internal.spinlock);
759 225 : }
760 350 : }
761 116 : break;
762 : case SPDK_BDEV_CLAIM_EXCL_WRITE:
763 : /* Examine by the one bdev module with a v1 claim */
764 1 : module = bdev->internal.claim.v1.module;
765 1 : if (module->examine_disk) {
766 1 : spdk_spin_lock(&module->internal.spinlock);
767 1 : module->internal.action_in_progress++;
768 1 : spdk_spin_unlock(&module->internal.spinlock);
769 1 : spdk_spin_unlock(&bdev->internal.spinlock);
770 1 : module->examine_disk(bdev);
771 1 : return;
772 : }
773 0 : break;
774 : default:
775 : /* Examine by all bdev modules with a v2 claim */
776 7 : assert(claim_type_is_v2(bdev->internal.claim_type));
777 : /*
778 : * Removal of tailq nodes while iterating can cause the iteration to jump out of the
779 : * list, perhaps accessing freed memory. Without protection, this could happen
780 : * while the lock is dropped during the examine callback.
781 : */
782 7 : bdev->internal.examine_in_progress++;
783 :
784 16 : TAILQ_FOREACH(claim, &bdev->internal.claim.v2.claims, link) {
785 9 : module = claim->module;
786 :
787 9 : if (module == NULL) {
788 : /* This is a vestigial claim, held by examine_count */
789 0 : continue;
790 : }
791 :
792 9 : if (module->examine_disk == NULL) {
793 0 : continue;
794 : }
795 :
796 9 : spdk_spin_lock(&module->internal.spinlock);
797 9 : module->internal.action_in_progress++;
798 9 : spdk_spin_unlock(&module->internal.spinlock);
799 :
800 : /* Call examine_disk without holding internal.spinlock. */
801 9 : spdk_spin_unlock(&bdev->internal.spinlock);
802 9 : module->examine_disk(bdev);
803 9 : spdk_spin_lock(&bdev->internal.spinlock);
804 9 : }
805 :
806 7 : assert(bdev->internal.examine_in_progress > 0);
807 7 : bdev->internal.examine_in_progress--;
808 7 : if (bdev->internal.examine_in_progress == 0) {
809 : /* Remove any claims that were released during examine_disk */
810 16 : TAILQ_FOREACH_SAFE(claim, &bdev->internal.claim.v2.claims, link, tmpclaim) {
811 9 : if (claim->desc != NULL) {
812 9 : continue;
813 : }
814 :
815 0 : TAILQ_REMOVE(&bdev->internal.claim.v2.claims, claim, link);
816 0 : free(claim);
817 0 : }
818 7 : if (TAILQ_EMPTY(&bdev->internal.claim.v2.claims)) {
819 0 : claim_reset(bdev);
820 0 : }
821 7 : }
822 7 : }
823 :
824 123 : spdk_spin_unlock(&bdev->internal.spinlock);
825 133 : }
826 :
827 : int
828 4 : spdk_bdev_examine(const char *name)
829 : {
830 : struct spdk_bdev *bdev;
831 : struct spdk_bdev_examine_item *item;
832 4 : struct spdk_thread *thread = spdk_get_thread();
833 :
834 4 : if (spdk_unlikely(!spdk_thread_is_app_thread(thread))) {
835 1 : SPDK_ERRLOG("Cannot examine bdev %s on thread %p (%s)\n", name, thread,
836 : thread ? spdk_thread_get_name(thread) : "null");
837 1 : return -EINVAL;
838 : }
839 :
840 3 : if (g_bdev_opts.bdev_auto_examine) {
841 0 : SPDK_ERRLOG("Manual examine is not allowed if auto examine is enabled\n");
842 0 : return -EINVAL;
843 : }
844 :
845 3 : if (bdev_examine_allowlist_check(name)) {
846 0 : SPDK_ERRLOG("Duplicate bdev name for manual examine: %s\n", name);
847 0 : return -EEXIST;
848 : }
849 :
850 3 : item = calloc(1, sizeof(*item));
851 3 : if (!item) {
852 0 : return -ENOMEM;
853 : }
854 3 : item->name = strdup(name);
855 3 : if (!item->name) {
856 0 : free(item);
857 0 : return -ENOMEM;
858 : }
859 3 : TAILQ_INSERT_TAIL(&g_bdev_examine_allowlist, item, link);
860 :
861 3 : bdev = spdk_bdev_get_by_name(name);
862 3 : if (bdev) {
863 3 : bdev_examine(bdev);
864 3 : }
865 3 : return 0;
866 4 : }
867 :
868 : static inline void
869 0 : bdev_examine_allowlist_config_json(struct spdk_json_write_ctx *w)
870 : {
871 : struct spdk_bdev_examine_item *item;
872 0 : TAILQ_FOREACH(item, &g_bdev_examine_allowlist, link) {
873 0 : spdk_json_write_object_begin(w);
874 0 : spdk_json_write_named_string(w, "method", "bdev_examine");
875 0 : spdk_json_write_named_object_begin(w, "params");
876 0 : spdk_json_write_named_string(w, "name", item->name);
877 0 : spdk_json_write_object_end(w);
878 0 : spdk_json_write_object_end(w);
879 0 : }
880 0 : }
881 :
882 : struct spdk_bdev *
883 1 : spdk_bdev_first(void)
884 : {
885 : struct spdk_bdev *bdev;
886 :
887 1 : bdev = TAILQ_FIRST(&g_bdev_mgr.bdevs);
888 1 : if (bdev) {
889 1 : SPDK_DEBUGLOG(bdev, "Starting bdev iteration at %s\n", bdev->name);
890 1 : }
891 :
892 1 : return bdev;
893 : }
894 :
895 : struct spdk_bdev *
896 8 : spdk_bdev_next(struct spdk_bdev *prev)
897 : {
898 : struct spdk_bdev *bdev;
899 :
900 8 : bdev = TAILQ_NEXT(prev, internal.link);
901 8 : if (bdev) {
902 7 : SPDK_DEBUGLOG(bdev, "Continuing bdev iteration at %s\n", bdev->name);
903 7 : }
904 :
905 8 : return bdev;
906 : }
907 :
908 : static struct spdk_bdev *
909 6 : _bdev_next_leaf(struct spdk_bdev *bdev)
910 : {
911 9 : while (bdev != NULL) {
912 8 : if (bdev->internal.claim_type == SPDK_BDEV_CLAIM_NONE) {
913 5 : return bdev;
914 : } else {
915 3 : bdev = TAILQ_NEXT(bdev, internal.link);
916 : }
917 : }
918 :
919 1 : return bdev;
920 6 : }
921 :
922 : struct spdk_bdev *
923 1 : spdk_bdev_first_leaf(void)
924 : {
925 : struct spdk_bdev *bdev;
926 :
927 1 : bdev = _bdev_next_leaf(TAILQ_FIRST(&g_bdev_mgr.bdevs));
928 :
929 1 : if (bdev) {
930 1 : SPDK_DEBUGLOG(bdev, "Starting bdev iteration at %s\n", bdev->name);
931 1 : }
932 :
933 1 : return bdev;
934 : }
935 :
936 : struct spdk_bdev *
937 5 : spdk_bdev_next_leaf(struct spdk_bdev *prev)
938 : {
939 : struct spdk_bdev *bdev;
940 :
941 5 : bdev = _bdev_next_leaf(TAILQ_NEXT(prev, internal.link));
942 :
943 5 : if (bdev) {
944 4 : SPDK_DEBUGLOG(bdev, "Continuing bdev iteration at %s\n", bdev->name);
945 4 : }
946 :
947 5 : return bdev;
948 : }
949 :
950 : static inline bool
951 820 : bdev_io_use_memory_domain(struct spdk_bdev_io *bdev_io)
952 : {
953 820 : return bdev_io->internal.f.has_memory_domain;
954 : }
955 :
956 : static inline bool
957 1551 : bdev_io_use_accel_sequence(struct spdk_bdev_io *bdev_io)
958 : {
959 1551 : return bdev_io->internal.f.has_accel_sequence;
960 : }
961 :
962 : static inline uint32_t
963 373 : bdev_desc_get_block_size(struct spdk_bdev_desc *desc)
964 : {
965 373 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
966 :
967 373 : if (spdk_unlikely(desc->opts.hide_metadata)) {
968 0 : return bdev->blocklen - bdev->md_len;
969 : } else {
970 373 : return bdev->blocklen;
971 : }
972 373 : }
973 :
974 : static inline uint32_t
975 110 : bdev_io_get_block_size(struct spdk_bdev_io *bdev_io)
976 : {
977 110 : struct spdk_bdev *bdev = bdev_io->bdev;
978 :
979 110 : if (bdev_io->u.bdev.dif_check_flags & SPDK_DIF_FLAGS_NVME_PRACT) {
980 0 : if (bdev->md_len == spdk_dif_pi_format_get_pi_size(bdev->dif_pi_format)) {
981 0 : return bdev->blocklen - bdev->md_len;
982 : } else {
983 0 : return bdev->blocklen;
984 : }
985 : }
986 :
987 110 : return bdev_desc_get_block_size(bdev_io->internal.desc);
988 110 : }
989 :
990 : static inline void
991 7 : bdev_queue_nomem_io_head(struct spdk_bdev_shared_resource *shared_resource,
992 : struct spdk_bdev_io *bdev_io, enum bdev_io_retry_state state)
993 : {
994 : /* Wait for some of the outstanding I/O to complete before we retry any of the nomem_io.
995 : * Normally we will wait for NOMEM_THRESHOLD_COUNT I/O to complete but for low queue depth
996 : * channels we will instead wait for half to complete.
997 : */
998 7 : shared_resource->nomem_threshold = spdk_max((int64_t)shared_resource->io_outstanding / 2,
999 : (int64_t)shared_resource->io_outstanding - NOMEM_THRESHOLD_COUNT);
1000 :
1001 7 : assert(state != BDEV_IO_RETRY_STATE_INVALID);
1002 7 : bdev_io->internal.retry_state = state;
1003 7 : TAILQ_INSERT_HEAD(&shared_resource->nomem_io, bdev_io, internal.link);
1004 7 : }
1005 :
1006 : static inline void
1007 43 : bdev_queue_nomem_io_tail(struct spdk_bdev_shared_resource *shared_resource,
1008 : struct spdk_bdev_io *bdev_io, enum bdev_io_retry_state state)
1009 : {
1010 : /* We only queue IOs at the end of the nomem_io queue if they're submitted by the user while
1011 : * the queue isn't empty, so we don't need to update the nomem_threshold here */
1012 43 : assert(!TAILQ_EMPTY(&shared_resource->nomem_io));
1013 :
1014 43 : assert(state != BDEV_IO_RETRY_STATE_INVALID);
1015 43 : bdev_io->internal.retry_state = state;
1016 43 : TAILQ_INSERT_TAIL(&shared_resource->nomem_io, bdev_io, internal.link);
1017 43 : }
1018 :
1019 : void
1020 16 : spdk_bdev_io_set_buf(struct spdk_bdev_io *bdev_io, void *buf, size_t len)
1021 : {
1022 : struct iovec *iovs;
1023 :
1024 16 : if (bdev_io->u.bdev.iovs == NULL) {
1025 3 : bdev_io->u.bdev.iovs = &bdev_io->iov;
1026 3 : bdev_io->u.bdev.iovcnt = 1;
1027 3 : }
1028 :
1029 16 : iovs = bdev_io->u.bdev.iovs;
1030 :
1031 16 : assert(iovs != NULL);
1032 16 : assert(bdev_io->u.bdev.iovcnt >= 1);
1033 :
1034 16 : iovs[0].iov_base = buf;
1035 16 : iovs[0].iov_len = len;
1036 16 : }
1037 :
1038 : void
1039 3 : spdk_bdev_io_set_md_buf(struct spdk_bdev_io *bdev_io, void *md_buf, size_t len)
1040 : {
1041 3 : assert((len / spdk_bdev_get_md_size(bdev_io->bdev)) >= bdev_io->u.bdev.num_blocks);
1042 3 : bdev_io->u.bdev.md_buf = md_buf;
1043 3 : }
1044 :
1045 : static bool
1046 167 : _is_buf_allocated(const struct iovec *iovs)
1047 : {
1048 167 : if (iovs == NULL) {
1049 6 : return false;
1050 : }
1051 :
1052 161 : return iovs[0].iov_base != NULL;
1053 167 : }
1054 :
1055 : static bool
1056 50 : _are_iovs_aligned(struct iovec *iovs, int iovcnt, uint32_t alignment)
1057 : {
1058 : int i;
1059 : uintptr_t iov_base;
1060 :
1061 50 : if (spdk_likely(alignment == 1)) {
1062 21 : return true;
1063 : }
1064 :
1065 36 : for (i = 0; i < iovcnt; i++) {
1066 29 : iov_base = (uintptr_t)iovs[i].iov_base;
1067 29 : if ((iov_base & (alignment - 1)) != 0) {
1068 22 : return false;
1069 : }
1070 7 : }
1071 :
1072 7 : return true;
1073 50 : }
1074 :
1075 : static inline bool
1076 895 : bdev_io_needs_metadata(struct spdk_bdev_desc *desc, struct spdk_bdev_io *bdev_io)
1077 : {
1078 1048 : return (bdev_io->bdev->md_len != 0) &&
1079 153 : (desc->opts.hide_metadata ||
1080 153 : (bdev_io->u.bdev.dif_check_flags & SPDK_DIF_FLAGS_NVME_PRACT));
1081 : }
1082 :
1083 : static inline bool
1084 852 : bdev_io_needs_sequence_exec(struct spdk_bdev_desc *desc, struct spdk_bdev_io *bdev_io)
1085 : {
1086 852 : if (!bdev_io_use_accel_sequence(bdev_io)) {
1087 852 : return false;
1088 : }
1089 :
1090 : /* For now, we don't allow splitting IOs with an accel sequence and will treat them as if
1091 : * bdev module didn't support accel sequences */
1092 0 : return !desc->accel_sequence_supported[bdev_io->type] || bdev_io->internal.f.split;
1093 852 : }
1094 :
1095 : static inline void
1096 592 : bdev_io_increment_outstanding(struct spdk_bdev_channel *bdev_ch,
1097 : struct spdk_bdev_shared_resource *shared_resource)
1098 : {
1099 592 : bdev_ch->io_outstanding++;
1100 592 : shared_resource->io_outstanding++;
1101 592 : }
1102 :
1103 : static inline void
1104 592 : bdev_io_decrement_outstanding(struct spdk_bdev_channel *bdev_ch,
1105 : struct spdk_bdev_shared_resource *shared_resource)
1106 : {
1107 592 : assert(bdev_ch->io_outstanding > 0);
1108 592 : assert(shared_resource->io_outstanding > 0);
1109 592 : bdev_ch->io_outstanding--;
1110 592 : shared_resource->io_outstanding--;
1111 592 : }
1112 :
1113 : static void
1114 0 : bdev_io_submit_sequence_cb(void *ctx, int status)
1115 : {
1116 0 : struct spdk_bdev_io *bdev_io = ctx;
1117 :
1118 0 : assert(bdev_io_use_accel_sequence(bdev_io));
1119 :
1120 0 : bdev_io->u.bdev.accel_sequence = NULL;
1121 0 : bdev_io->internal.f.has_accel_sequence = false;
1122 :
1123 0 : if (spdk_unlikely(status != 0)) {
1124 0 : SPDK_ERRLOG("Failed to execute accel sequence, status=%d\n", status);
1125 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
1126 0 : bdev_io_complete_unsubmitted(bdev_io);
1127 0 : return;
1128 : }
1129 :
1130 0 : bdev_io_submit(bdev_io);
1131 0 : }
1132 :
1133 : static void
1134 0 : bdev_io_exec_sequence_cb(void *ctx, int status)
1135 : {
1136 0 : struct spdk_bdev_io *bdev_io = ctx;
1137 0 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1138 :
1139 0 : TAILQ_REMOVE(&bdev_io->internal.ch->io_accel_exec, bdev_io, internal.link);
1140 0 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1141 :
1142 0 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1143 0 : bdev_ch_retry_io(ch);
1144 0 : }
1145 :
1146 0 : bdev_io->internal.data_transfer_cpl(bdev_io, status);
1147 0 : }
1148 :
1149 : static void
1150 0 : bdev_io_exec_sequence(struct spdk_bdev_io *bdev_io, void (*cb_fn)(void *ctx, int status))
1151 : {
1152 0 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1153 :
1154 0 : assert(bdev_io_needs_sequence_exec(bdev_io->internal.desc, bdev_io));
1155 0 : assert(bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE || bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
1156 0 : assert(bdev_io_use_accel_sequence(bdev_io));
1157 :
1158 : /* Since the operations are appended during submission, they're in the opposite order than
1159 : * how we want to execute them for reads (i.e. we need to execute the most recently added
1160 : * operation first), so reverse the sequence before executing it.
1161 : */
1162 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_READ) {
1163 0 : spdk_accel_sequence_reverse(bdev_io->internal.accel_sequence);
1164 0 : }
1165 :
1166 0 : TAILQ_INSERT_TAIL(&bdev_io->internal.ch->io_accel_exec, bdev_io, internal.link);
1167 0 : bdev_io_increment_outstanding(ch, ch->shared_resource);
1168 0 : bdev_io->internal.data_transfer_cpl = cb_fn;
1169 :
1170 0 : spdk_accel_sequence_finish(bdev_io->internal.accel_sequence,
1171 0 : bdev_io_exec_sequence_cb, bdev_io);
1172 0 : }
1173 :
1174 : static void
1175 42 : bdev_io_get_buf_complete(struct spdk_bdev_io *bdev_io, bool status)
1176 : {
1177 42 : struct spdk_io_channel *ch = spdk_bdev_io_get_io_channel(bdev_io);
1178 : void *buf;
1179 :
1180 42 : if (spdk_unlikely(bdev_io->internal.get_aux_buf_cb != NULL)) {
1181 0 : buf = bdev_io->internal.buf.ptr;
1182 0 : bdev_io->internal.buf.ptr = NULL;
1183 0 : bdev_io->internal.f.has_buf = false;
1184 0 : bdev_io->internal.get_aux_buf_cb(ch, bdev_io, buf);
1185 0 : bdev_io->internal.get_aux_buf_cb = NULL;
1186 0 : } else {
1187 42 : assert(bdev_io->internal.get_buf_cb != NULL);
1188 42 : bdev_io->internal.get_buf_cb(ch, bdev_io, status);
1189 42 : bdev_io->internal.get_buf_cb = NULL;
1190 : }
1191 42 : }
1192 :
1193 : static void
1194 4 : _bdev_io_pull_buffer_cpl(void *ctx, int rc)
1195 : {
1196 4 : struct spdk_bdev_io *bdev_io = ctx;
1197 :
1198 4 : if (rc) {
1199 0 : SPDK_ERRLOG("Set bounce buffer failed with rc %d\n", rc);
1200 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
1201 0 : }
1202 4 : bdev_io_get_buf_complete(bdev_io, !rc);
1203 4 : }
1204 :
1205 : static void
1206 2 : bdev_io_pull_md_buf_done(void *ctx, int status)
1207 : {
1208 2 : struct spdk_bdev_io *bdev_io = ctx;
1209 2 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1210 :
1211 2 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1212 2 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1213 :
1214 2 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1215 0 : bdev_ch_retry_io(ch);
1216 0 : }
1217 :
1218 2 : assert(bdev_io->internal.data_transfer_cpl);
1219 2 : bdev_io->internal.data_transfer_cpl(bdev_io, status);
1220 2 : }
1221 :
1222 : static void
1223 4 : bdev_io_pull_md_buf(struct spdk_bdev_io *bdev_io)
1224 : {
1225 4 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1226 4 : int rc = 0;
1227 :
1228 4 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
1229 2 : assert(bdev_io->internal.f.has_bounce_buf);
1230 2 : if (bdev_io_use_memory_domain(bdev_io)) {
1231 2 : TAILQ_INSERT_TAIL(&ch->io_memory_domain, bdev_io, internal.link);
1232 2 : bdev_io_increment_outstanding(ch, ch->shared_resource);
1233 4 : rc = spdk_memory_domain_pull_data(bdev_io->internal.memory_domain,
1234 2 : bdev_io->internal.memory_domain_ctx,
1235 2 : &bdev_io->internal.bounce_buf.orig_md_iov, 1,
1236 2 : &bdev_io->internal.bounce_buf.md_iov, 1,
1237 2 : bdev_io_pull_md_buf_done, bdev_io);
1238 2 : if (rc == 0) {
1239 : /* Continue to submit IO in completion callback */
1240 2 : return;
1241 : }
1242 0 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1243 0 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1244 0 : if (rc != -ENOMEM) {
1245 0 : SPDK_ERRLOG("Failed to pull data from memory domain %s, rc %d\n",
1246 : spdk_memory_domain_get_dma_device_id(
1247 : bdev_io->internal.memory_domain), rc);
1248 0 : }
1249 0 : } else {
1250 0 : memcpy(bdev_io->internal.bounce_buf.md_iov.iov_base,
1251 0 : bdev_io->internal.bounce_buf.orig_md_iov.iov_base,
1252 0 : bdev_io->internal.bounce_buf.orig_md_iov.iov_len);
1253 : }
1254 0 : }
1255 :
1256 2 : if (spdk_unlikely(rc == -ENOMEM)) {
1257 0 : bdev_queue_nomem_io_head(ch->shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PULL_MD);
1258 0 : } else {
1259 2 : assert(bdev_io->internal.data_transfer_cpl);
1260 2 : bdev_io->internal.data_transfer_cpl(bdev_io, rc);
1261 : }
1262 4 : }
1263 :
1264 : static void
1265 4 : _bdev_io_pull_bounce_md_buf(struct spdk_bdev_io *bdev_io, void *md_buf, size_t len)
1266 : {
1267 4 : assert(bdev_io->internal.f.has_bounce_buf);
1268 :
1269 : /* save original md_buf */
1270 4 : bdev_io->internal.bounce_buf.orig_md_iov.iov_base = bdev_io->u.bdev.md_buf;
1271 4 : bdev_io->internal.bounce_buf.orig_md_iov.iov_len = len;
1272 4 : bdev_io->internal.bounce_buf.md_iov.iov_base = md_buf;
1273 4 : bdev_io->internal.bounce_buf.md_iov.iov_len = len;
1274 : /* set bounce md_buf */
1275 4 : bdev_io->u.bdev.md_buf = md_buf;
1276 :
1277 4 : bdev_io_pull_md_buf(bdev_io);
1278 4 : }
1279 :
1280 : static void
1281 42 : _bdev_io_set_md_buf(struct spdk_bdev_io *bdev_io)
1282 : {
1283 42 : struct spdk_bdev *bdev = bdev_io->bdev;
1284 : uint64_t md_len;
1285 : void *buf;
1286 :
1287 42 : if (spdk_bdev_is_md_separate(bdev)) {
1288 7 : assert(!bdev_io_use_accel_sequence(bdev_io));
1289 :
1290 7 : buf = (char *)bdev_io->u.bdev.iovs[0].iov_base + bdev_io->u.bdev.iovs[0].iov_len;
1291 7 : md_len = bdev_io->u.bdev.num_blocks * bdev->md_len;
1292 :
1293 7 : assert(((uintptr_t)buf & (spdk_bdev_get_buf_align(bdev) - 1)) == 0);
1294 :
1295 7 : if (bdev_io->u.bdev.md_buf != NULL) {
1296 4 : _bdev_io_pull_bounce_md_buf(bdev_io, buf, md_len);
1297 4 : return;
1298 : } else {
1299 3 : spdk_bdev_io_set_md_buf(bdev_io, buf, md_len);
1300 : }
1301 3 : }
1302 :
1303 38 : bdev_io_get_buf_complete(bdev_io, true);
1304 42 : }
1305 :
1306 : static inline void
1307 26 : bdev_io_pull_data_done(struct spdk_bdev_io *bdev_io, int rc)
1308 : {
1309 26 : if (rc) {
1310 0 : SPDK_ERRLOG("Failed to get data buffer\n");
1311 0 : assert(bdev_io->internal.data_transfer_cpl);
1312 0 : bdev_io->internal.data_transfer_cpl(bdev_io, rc);
1313 0 : return;
1314 : }
1315 :
1316 26 : _bdev_io_set_md_buf(bdev_io);
1317 26 : }
1318 :
1319 : static void
1320 2 : bdev_io_pull_data_done_and_track(void *ctx, int status)
1321 : {
1322 2 : struct spdk_bdev_io *bdev_io = ctx;
1323 2 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1324 :
1325 2 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1326 2 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1327 :
1328 2 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1329 0 : bdev_ch_retry_io(ch);
1330 0 : }
1331 :
1332 2 : bdev_io_pull_data_done(bdev_io, status);
1333 2 : }
1334 :
1335 : static void
1336 27 : bdev_io_pull_data(struct spdk_bdev_io *bdev_io)
1337 : {
1338 27 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1339 27 : struct spdk_bdev_desc *desc = bdev_io->internal.desc;
1340 27 : int rc = 0;
1341 :
1342 27 : assert(bdev_io->internal.f.has_bounce_buf);
1343 :
1344 27 : if (bdev_io_needs_metadata(desc, bdev_io)) {
1345 0 : assert(bdev_io->bdev->md_interleave);
1346 :
1347 0 : bdev_io->u.bdev.dif_check_flags &= ~SPDK_DIF_FLAGS_NVME_PRACT;
1348 :
1349 0 : if (!bdev_io_use_accel_sequence(bdev_io)) {
1350 0 : bdev_io->internal.accel_sequence = NULL;
1351 0 : }
1352 :
1353 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
1354 0 : rc = spdk_accel_append_dif_generate_copy(&bdev_io->internal.accel_sequence, ch->accel_channel,
1355 0 : bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt,
1356 0 : bdev_io->u.bdev.memory_domain,
1357 0 : bdev_io->u.bdev.memory_domain_ctx,
1358 0 : bdev_io->internal.bounce_buf.orig_iovs,
1359 0 : bdev_io->internal.bounce_buf.orig_iovcnt,
1360 0 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain : NULL,
1361 0 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain_ctx : NULL,
1362 0 : bdev_io->u.bdev.num_blocks,
1363 0 : &bdev_io->u.bdev.dif_ctx,
1364 : NULL, NULL);
1365 0 : } else {
1366 0 : assert(bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
1367 0 : rc = spdk_accel_append_dif_verify_copy(&bdev_io->internal.accel_sequence, ch->accel_channel,
1368 0 : bdev_io->internal.bounce_buf.orig_iovs,
1369 0 : bdev_io->internal.bounce_buf.orig_iovcnt,
1370 0 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain : NULL,
1371 0 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain_ctx : NULL,
1372 0 : bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt,
1373 0 : bdev_io->u.bdev.memory_domain,
1374 0 : bdev_io->u.bdev.memory_domain_ctx,
1375 0 : bdev_io->u.bdev.num_blocks,
1376 0 : &bdev_io->u.bdev.dif_ctx,
1377 0 : &bdev_io->u.bdev.dif_err,
1378 : NULL, NULL);
1379 : }
1380 :
1381 0 : if (spdk_likely(rc == 0)) {
1382 0 : bdev_io->internal.f.has_accel_sequence = true;
1383 0 : bdev_io->u.bdev.accel_sequence = bdev_io->internal.accel_sequence;
1384 0 : } else if (rc != -ENOMEM) {
1385 0 : SPDK_ERRLOG("Failed to append generate/verify_copy to accel sequence: %p\n",
1386 : bdev_io->internal.accel_sequence);
1387 0 : }
1388 27 : } else if (bdev_io_needs_sequence_exec(desc, bdev_io) ||
1389 27 : (bdev_io_use_accel_sequence(bdev_io) && bdev_io_use_memory_domain(bdev_io))) {
1390 : /* If we need to exec an accel sequence or the IO uses a memory domain buffer and has a
1391 : * sequence, append a copy operation making accel change the src/dst buffers of the previous
1392 : * operation */
1393 0 : assert(bdev_io_use_accel_sequence(bdev_io));
1394 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
1395 0 : rc = spdk_accel_append_copy(&bdev_io->internal.accel_sequence, ch->accel_channel,
1396 0 : bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt,
1397 : NULL, NULL,
1398 0 : bdev_io->internal.bounce_buf.orig_iovs,
1399 0 : bdev_io->internal.bounce_buf.orig_iovcnt,
1400 0 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain : NULL,
1401 0 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain_ctx : NULL,
1402 : NULL, NULL);
1403 0 : } else {
1404 : /* We need to reverse the src/dst for reads */
1405 0 : assert(bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
1406 0 : rc = spdk_accel_append_copy(&bdev_io->internal.accel_sequence, ch->accel_channel,
1407 0 : bdev_io->internal.bounce_buf.orig_iovs,
1408 0 : bdev_io->internal.bounce_buf.orig_iovcnt,
1409 0 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain : NULL,
1410 0 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain_ctx : NULL,
1411 0 : bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt,
1412 : NULL, NULL, NULL, NULL);
1413 : }
1414 :
1415 0 : if (spdk_unlikely(rc != 0 && rc != -ENOMEM)) {
1416 0 : SPDK_ERRLOG("Failed to append copy to accel sequence: %p\n",
1417 : bdev_io->internal.accel_sequence);
1418 0 : }
1419 27 : } else if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
1420 : /* if this is write path, copy data from original buffer to bounce buffer */
1421 17 : if (bdev_io_use_memory_domain(bdev_io)) {
1422 3 : TAILQ_INSERT_TAIL(&ch->io_memory_domain, bdev_io, internal.link);
1423 3 : bdev_io_increment_outstanding(ch, ch->shared_resource);
1424 6 : rc = spdk_memory_domain_pull_data(bdev_io->internal.memory_domain,
1425 3 : bdev_io->internal.memory_domain_ctx,
1426 3 : bdev_io->internal.bounce_buf.orig_iovs,
1427 3 : (uint32_t)bdev_io->internal.bounce_buf.orig_iovcnt,
1428 3 : bdev_io->u.bdev.iovs, 1,
1429 : bdev_io_pull_data_done_and_track,
1430 3 : bdev_io);
1431 3 : if (rc == 0) {
1432 : /* Continue to submit IO in completion callback */
1433 2 : return;
1434 : }
1435 1 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1436 1 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1437 1 : if (rc != -ENOMEM) {
1438 0 : SPDK_ERRLOG("Failed to pull data from memory domain %s\n",
1439 : spdk_memory_domain_get_dma_device_id(
1440 : bdev_io->internal.memory_domain));
1441 0 : }
1442 1 : } else {
1443 14 : assert(bdev_io->u.bdev.iovcnt == 1);
1444 28 : spdk_copy_iovs_to_buf(bdev_io->u.bdev.iovs[0].iov_base,
1445 14 : bdev_io->u.bdev.iovs[0].iov_len,
1446 14 : bdev_io->internal.bounce_buf.orig_iovs,
1447 14 : bdev_io->internal.bounce_buf.orig_iovcnt);
1448 : }
1449 15 : }
1450 :
1451 25 : if (spdk_unlikely(rc == -ENOMEM)) {
1452 1 : bdev_queue_nomem_io_head(ch->shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PULL);
1453 1 : } else {
1454 24 : bdev_io_pull_data_done(bdev_io, rc);
1455 : }
1456 27 : }
1457 :
1458 : static void
1459 26 : _bdev_io_pull_bounce_data_buf(struct spdk_bdev_io *bdev_io, void *buf, size_t len,
1460 : bdev_copy_bounce_buffer_cpl cpl_cb)
1461 : {
1462 26 : struct spdk_bdev_shared_resource *shared_resource = bdev_io->internal.ch->shared_resource;
1463 :
1464 26 : assert(bdev_io->internal.f.has_bounce_buf == false);
1465 :
1466 26 : bdev_io->internal.data_transfer_cpl = cpl_cb;
1467 26 : bdev_io->internal.f.has_bounce_buf = true;
1468 : /* save original iovec */
1469 26 : bdev_io->internal.bounce_buf.orig_iovs = bdev_io->u.bdev.iovs;
1470 26 : bdev_io->internal.bounce_buf.orig_iovcnt = bdev_io->u.bdev.iovcnt;
1471 : /* zero the other data members */
1472 26 : bdev_io->internal.bounce_buf.iov.iov_base = NULL;
1473 26 : bdev_io->internal.bounce_buf.md_iov.iov_base = NULL;
1474 26 : bdev_io->internal.bounce_buf.orig_md_iov.iov_base = NULL;
1475 : /* set bounce iov */
1476 26 : bdev_io->u.bdev.iovs = &bdev_io->internal.bounce_buf.iov;
1477 26 : bdev_io->u.bdev.iovcnt = 1;
1478 : /* set bounce buffer for this operation */
1479 26 : bdev_io->u.bdev.iovs[0].iov_base = buf;
1480 26 : bdev_io->u.bdev.iovs[0].iov_len = len;
1481 : /* Now we use 1 iov, the split condition could have been changed */
1482 26 : bdev_io->internal.f.split = bdev_io_should_split(bdev_io);
1483 :
1484 26 : if (spdk_unlikely(!TAILQ_EMPTY(&shared_resource->nomem_io))) {
1485 0 : bdev_queue_nomem_io_tail(shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PULL);
1486 0 : } else {
1487 26 : bdev_io_pull_data(bdev_io);
1488 : }
1489 26 : }
1490 :
1491 : static void
1492 42 : _bdev_io_set_buf(struct spdk_bdev_io *bdev_io, void *buf, uint64_t len)
1493 : {
1494 42 : struct spdk_bdev *bdev = bdev_io->bdev;
1495 : bool buf_allocated;
1496 : uint64_t alignment;
1497 : void *aligned_buf;
1498 :
1499 42 : bdev_io->internal.buf.ptr = buf;
1500 42 : bdev_io->internal.f.has_buf = true;
1501 :
1502 42 : if (spdk_unlikely(bdev_io->internal.get_aux_buf_cb != NULL)) {
1503 0 : bdev_io_get_buf_complete(bdev_io, true);
1504 0 : return;
1505 : }
1506 :
1507 42 : alignment = spdk_bdev_get_buf_align(bdev);
1508 42 : buf_allocated = _is_buf_allocated(bdev_io->u.bdev.iovs);
1509 42 : aligned_buf = (void *)(((uintptr_t)buf + (alignment - 1)) & ~(alignment - 1));
1510 :
1511 42 : if (buf_allocated) {
1512 26 : _bdev_io_pull_bounce_data_buf(bdev_io, aligned_buf, len, _bdev_io_pull_buffer_cpl);
1513 : /* Continue in completion callback */
1514 26 : return;
1515 : } else {
1516 16 : spdk_bdev_io_set_buf(bdev_io, aligned_buf, len);
1517 : }
1518 :
1519 16 : _bdev_io_set_md_buf(bdev_io);
1520 42 : }
1521 :
1522 : static inline uint64_t
1523 42 : bdev_io_get_max_buf_len(struct spdk_bdev_io *bdev_io, uint64_t len)
1524 : {
1525 42 : struct spdk_bdev *bdev = bdev_io->bdev;
1526 : uint64_t md_len, alignment;
1527 :
1528 42 : md_len = spdk_bdev_is_md_separate(bdev) ? bdev_io->u.bdev.num_blocks * bdev->md_len : 0;
1529 :
1530 : /* 1 byte alignment needs 0 byte of extra space, 64 bytes alignment needs 63 bytes of extra space, etc. */
1531 42 : alignment = spdk_bdev_get_buf_align(bdev) - 1;
1532 :
1533 42 : return len + alignment + md_len;
1534 : }
1535 :
1536 : static void
1537 42 : bdev_io_put_accel_buf(struct spdk_bdev_io *bdev_io)
1538 : {
1539 42 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1540 :
1541 84 : spdk_accel_put_buf(ch->accel_channel,
1542 42 : bdev_io->internal.buf.ptr,
1543 42 : bdev_io->u.bdev.memory_domain,
1544 42 : bdev_io->u.bdev.memory_domain_ctx);
1545 42 : }
1546 :
1547 : static void
1548 0 : _bdev_io_put_buf(struct spdk_bdev_io *bdev_io, void *buf, uint64_t buf_len)
1549 : {
1550 : struct spdk_bdev_mgmt_channel *ch;
1551 :
1552 0 : ch = bdev_io->internal.ch->shared_resource->mgmt_ch;
1553 0 : spdk_iobuf_put(&ch->iobuf, buf, bdev_io_get_max_buf_len(bdev_io, buf_len));
1554 0 : }
1555 :
1556 : static void
1557 42 : bdev_io_put_buf(struct spdk_bdev_io *bdev_io)
1558 : {
1559 42 : assert(bdev_io->internal.f.has_buf);
1560 :
1561 42 : if (bdev_io->u.bdev.memory_domain == spdk_accel_get_memory_domain()) {
1562 42 : bdev_io_put_accel_buf(bdev_io);
1563 42 : } else {
1564 0 : assert(bdev_io->u.bdev.memory_domain == NULL);
1565 0 : _bdev_io_put_buf(bdev_io, bdev_io->internal.buf.ptr,
1566 0 : bdev_io->internal.buf.len);
1567 : }
1568 42 : bdev_io->internal.buf.ptr = NULL;
1569 42 : bdev_io->internal.f.has_buf = false;
1570 42 : }
1571 :
1572 3 : SPDK_LOG_DEPRECATION_REGISTER(spdk_bdev_io_put_aux_buf,
1573 : "spdk_bdev_io_put_aux_buf is deprecated", "v25.01", 0);
1574 :
1575 : void
1576 0 : spdk_bdev_io_put_aux_buf(struct spdk_bdev_io *bdev_io, void *buf)
1577 : {
1578 0 : uint64_t len = bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen;
1579 :
1580 0 : SPDK_LOG_DEPRECATED(spdk_bdev_io_put_aux_buf);
1581 :
1582 0 : assert(buf != NULL);
1583 0 : _bdev_io_put_buf(bdev_io, buf, len);
1584 0 : }
1585 :
1586 : static inline bool
1587 549 : bdev_io_support_dif(struct spdk_bdev_io *bdev_io)
1588 : {
1589 549 : switch (bdev_io->type) {
1590 : case SPDK_BDEV_IO_TYPE_WRITE:
1591 : case SPDK_BDEV_IO_TYPE_READ:
1592 442 : return true;
1593 : default:
1594 107 : return false;
1595 : }
1596 549 : }
1597 :
1598 : static inline void
1599 549 : bdev_submit_request(struct spdk_bdev *bdev, struct spdk_io_channel *ioch,
1600 : struct spdk_bdev_io *bdev_io)
1601 : {
1602 : /* After a request is submitted to a bdev module, the ownership of an accel sequence
1603 : * associated with that bdev_io is transferred to the bdev module. So, clear the internal
1604 : * sequence pointer to make sure we won't touch it anymore. */
1605 1016 : if ((bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE ||
1606 549 : bdev_io->type == SPDK_BDEV_IO_TYPE_READ) && bdev_io->u.bdev.accel_sequence != NULL) {
1607 0 : assert(!bdev_io_needs_sequence_exec(bdev_io->internal.desc, bdev_io));
1608 0 : bdev_io->internal.f.has_accel_sequence = false;
1609 0 : }
1610 :
1611 : /* The generic bdev layer should not pass an I/O with a dif_check_flags set that
1612 : * the underlying bdev does not support. Add an assert to check this.
1613 : */
1614 549 : assert(!bdev_io_support_dif(bdev_io) ||
1615 : ((bdev_io->u.bdev.dif_check_flags & bdev->dif_check_flags) ==
1616 : bdev_io->u.bdev.dif_check_flags));
1617 :
1618 549 : bdev->fn_table->submit_request(ioch, bdev_io);
1619 549 : }
1620 :
1621 : static inline void
1622 10 : bdev_ch_resubmit_io(struct spdk_bdev_shared_resource *shared_resource, struct spdk_bdev_io *bdev_io)
1623 : {
1624 10 : struct spdk_bdev *bdev = bdev_io->bdev;
1625 :
1626 10 : bdev_io_increment_outstanding(bdev_io->internal.ch, shared_resource);
1627 10 : bdev_io->internal.error.nvme.cdw0 = 0;
1628 10 : bdev_io->num_retries++;
1629 10 : bdev_submit_request(bdev, spdk_bdev_io_get_io_channel(bdev_io), bdev_io);
1630 10 : }
1631 :
1632 : static void
1633 63 : bdev_shared_ch_retry_io(struct spdk_bdev_shared_resource *shared_resource)
1634 : {
1635 : struct spdk_bdev_io *bdev_io;
1636 :
1637 63 : if (shared_resource->io_outstanding > shared_resource->nomem_threshold) {
1638 : /*
1639 : * Allow some more I/O to complete before retrying the nomem_io queue.
1640 : * Some drivers (such as nvme) cannot immediately take a new I/O in
1641 : * the context of a completion, because the resources for the I/O are
1642 : * not released until control returns to the bdev poller. Also, we
1643 : * may require several small I/O to complete before a larger I/O
1644 : * (that requires splitting) can be submitted.
1645 : */
1646 58 : return;
1647 : }
1648 :
1649 16 : while (!TAILQ_EMPTY(&shared_resource->nomem_io)) {
1650 12 : bdev_io = TAILQ_FIRST(&shared_resource->nomem_io);
1651 12 : TAILQ_REMOVE(&shared_resource->nomem_io, bdev_io, internal.link);
1652 :
1653 12 : switch (bdev_io->internal.retry_state) {
1654 : case BDEV_IO_RETRY_STATE_SUBMIT:
1655 10 : bdev_ch_resubmit_io(shared_resource, bdev_io);
1656 10 : break;
1657 : case BDEV_IO_RETRY_STATE_PULL:
1658 1 : bdev_io_pull_data(bdev_io);
1659 1 : break;
1660 : case BDEV_IO_RETRY_STATE_PULL_MD:
1661 0 : bdev_io_pull_md_buf(bdev_io);
1662 0 : break;
1663 : case BDEV_IO_RETRY_STATE_PUSH:
1664 1 : bdev_io_push_bounce_data(bdev_io);
1665 1 : break;
1666 : case BDEV_IO_RETRY_STATE_PUSH_MD:
1667 0 : bdev_io_push_bounce_md_buf(bdev_io);
1668 0 : break;
1669 : case BDEV_IO_RETRY_STATE_GET_ACCEL_BUF:
1670 0 : _bdev_io_get_accel_buf(bdev_io);
1671 0 : break;
1672 : default:
1673 0 : assert(0 && "invalid retry state");
1674 : break;
1675 : }
1676 :
1677 12 : if (bdev_io == TAILQ_FIRST(&shared_resource->nomem_io)) {
1678 : /* This IO completed again with NOMEM status, so break the loop and
1679 : * don't try anymore. Note that a bdev_io that fails with NOMEM
1680 : * always gets requeued at the front of the list, to maintain
1681 : * ordering.
1682 : */
1683 1 : break;
1684 : }
1685 : }
1686 63 : }
1687 :
1688 : static void
1689 63 : bdev_ch_retry_io(struct spdk_bdev_channel *bdev_ch)
1690 : {
1691 63 : bdev_shared_ch_retry_io(bdev_ch->shared_resource);
1692 63 : }
1693 :
1694 : static int
1695 0 : bdev_no_mem_poller(void *ctx)
1696 : {
1697 0 : struct spdk_bdev_shared_resource *shared_resource = ctx;
1698 :
1699 0 : spdk_poller_unregister(&shared_resource->nomem_poller);
1700 :
1701 0 : if (!TAILQ_EMPTY(&shared_resource->nomem_io)) {
1702 0 : bdev_shared_ch_retry_io(shared_resource);
1703 0 : }
1704 : /* the retry cb may re-register the poller so double check */
1705 0 : if (!TAILQ_EMPTY(&shared_resource->nomem_io) &&
1706 0 : shared_resource->io_outstanding == 0 && shared_resource->nomem_poller == NULL) {
1707 : /* No IOs were submitted, try again */
1708 0 : shared_resource->nomem_poller = SPDK_POLLER_REGISTER(bdev_no_mem_poller, shared_resource,
1709 : SPDK_BDEV_IO_POLL_INTERVAL_IN_MSEC * 10);
1710 0 : }
1711 :
1712 0 : return SPDK_POLLER_BUSY;
1713 : }
1714 :
1715 : static inline bool
1716 556 : _bdev_io_handle_no_mem(struct spdk_bdev_io *bdev_io, enum bdev_io_retry_state state)
1717 : {
1718 556 : struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
1719 556 : struct spdk_bdev_shared_resource *shared_resource = bdev_ch->shared_resource;
1720 :
1721 556 : if (spdk_unlikely(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_NOMEM)) {
1722 5 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_PENDING;
1723 5 : bdev_queue_nomem_io_head(shared_resource, bdev_io, state);
1724 :
1725 5 : if (shared_resource->io_outstanding == 0 && !shared_resource->nomem_poller) {
1726 : /* Special case when we have nomem IOs and no outstanding IOs which completions
1727 : * could trigger retry of queued IOs
1728 : * Any IOs submitted may trigger retry of queued IOs. This poller handles a case when no
1729 : * new IOs submitted, e.g. qd==1 */
1730 0 : shared_resource->nomem_poller = SPDK_POLLER_REGISTER(bdev_no_mem_poller, shared_resource,
1731 : SPDK_BDEV_IO_POLL_INTERVAL_IN_MSEC * 10);
1732 0 : }
1733 : /* If bdev module completed an I/O that has an accel sequence with NOMEM status, the
1734 : * ownership of that sequence is transferred back to the bdev layer, so we need to
1735 : * restore internal.accel_sequence to make sure that the sequence is handled
1736 : * correctly in case the I/O is later aborted. */
1737 5 : if ((bdev_io->type == SPDK_BDEV_IO_TYPE_READ ||
1738 5 : bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) && bdev_io->u.bdev.accel_sequence) {
1739 0 : assert(!bdev_io_use_accel_sequence(bdev_io));
1740 0 : bdev_io->internal.f.has_accel_sequence = true;
1741 0 : bdev_io->internal.accel_sequence = bdev_io->u.bdev.accel_sequence;
1742 0 : }
1743 :
1744 5 : return true;
1745 : }
1746 :
1747 551 : if (spdk_unlikely(!TAILQ_EMPTY(&shared_resource->nomem_io))) {
1748 63 : bdev_ch_retry_io(bdev_ch);
1749 63 : }
1750 :
1751 551 : return false;
1752 556 : }
1753 :
1754 : static void
1755 26 : _bdev_io_complete_push_bounce_done(void *ctx, int rc)
1756 : {
1757 26 : struct spdk_bdev_io *bdev_io = ctx;
1758 26 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1759 :
1760 26 : if (rc) {
1761 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
1762 0 : }
1763 : /* We want to free the bounce buffer here since we know we're done with it (as opposed
1764 : * to waiting for the conditional free of internal.buf.ptr in spdk_bdev_free_io()).
1765 : */
1766 26 : bdev_io_put_buf(bdev_io);
1767 :
1768 26 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1769 0 : bdev_ch_retry_io(ch);
1770 0 : }
1771 :
1772 : /* Continue with IO completion flow */
1773 26 : bdev_io_complete(bdev_io);
1774 26 : }
1775 :
1776 : static void
1777 2 : bdev_io_push_bounce_md_buf_done(void *ctx, int rc)
1778 : {
1779 2 : struct spdk_bdev_io *bdev_io = ctx;
1780 2 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1781 :
1782 2 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1783 2 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1784 2 : bdev_io->internal.f.has_bounce_buf = false;
1785 :
1786 2 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1787 0 : bdev_ch_retry_io(ch);
1788 0 : }
1789 :
1790 2 : bdev_io->internal.data_transfer_cpl(bdev_io, rc);
1791 2 : }
1792 :
1793 : static inline void
1794 26 : bdev_io_push_bounce_md_buf(struct spdk_bdev_io *bdev_io)
1795 : {
1796 26 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1797 26 : int rc = 0;
1798 :
1799 26 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS);
1800 26 : assert(bdev_io->internal.f.has_bounce_buf);
1801 :
1802 : /* do the same for metadata buffer */
1803 26 : if (spdk_unlikely(bdev_io->internal.bounce_buf.orig_md_iov.iov_base != NULL)) {
1804 4 : assert(spdk_bdev_is_md_separate(bdev_io->bdev));
1805 :
1806 4 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_READ) {
1807 2 : if (bdev_io_use_memory_domain(bdev_io)) {
1808 2 : TAILQ_INSERT_TAIL(&ch->io_memory_domain, bdev_io, internal.link);
1809 2 : bdev_io_increment_outstanding(ch, ch->shared_resource);
1810 : /* If memory domain is used then we need to call async push function */
1811 4 : rc = spdk_memory_domain_push_data(bdev_io->internal.memory_domain,
1812 2 : bdev_io->internal.memory_domain_ctx,
1813 2 : &bdev_io->internal.bounce_buf.orig_md_iov,
1814 2 : (uint32_t)bdev_io->internal.bounce_buf.orig_iovcnt,
1815 2 : &bdev_io->internal.bounce_buf.md_iov, 1,
1816 : bdev_io_push_bounce_md_buf_done,
1817 2 : bdev_io);
1818 2 : if (rc == 0) {
1819 : /* Continue IO completion in async callback */
1820 2 : return;
1821 : }
1822 0 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1823 0 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1824 0 : if (rc != -ENOMEM) {
1825 0 : SPDK_ERRLOG("Failed to push md to memory domain %s\n",
1826 : spdk_memory_domain_get_dma_device_id(
1827 : bdev_io->internal.memory_domain));
1828 0 : }
1829 0 : } else {
1830 0 : memcpy(bdev_io->internal.bounce_buf.orig_md_iov.iov_base, bdev_io->u.bdev.md_buf,
1831 0 : bdev_io->internal.bounce_buf.orig_md_iov.iov_len);
1832 : }
1833 0 : }
1834 2 : }
1835 :
1836 24 : if (spdk_unlikely(rc == -ENOMEM)) {
1837 0 : bdev_queue_nomem_io_head(ch->shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PUSH_MD);
1838 0 : } else {
1839 24 : assert(bdev_io->internal.data_transfer_cpl);
1840 24 : bdev_io->internal.f.has_bounce_buf = false;
1841 24 : bdev_io->internal.data_transfer_cpl(bdev_io, rc);
1842 : }
1843 26 : }
1844 :
1845 : static inline void
1846 26 : bdev_io_push_bounce_data_done(struct spdk_bdev_io *bdev_io, int rc)
1847 : {
1848 26 : assert(bdev_io->internal.data_transfer_cpl);
1849 26 : if (rc) {
1850 0 : bdev_io->internal.data_transfer_cpl(bdev_io, rc);
1851 0 : return;
1852 : }
1853 :
1854 : /* set original buffer for this io */
1855 26 : bdev_io->u.bdev.iovcnt = bdev_io->internal.bounce_buf.orig_iovcnt;
1856 26 : bdev_io->u.bdev.iovs = bdev_io->internal.bounce_buf.orig_iovs;
1857 :
1858 : /* We don't set bdev_io->internal.f.has_bounce_buf to false here because
1859 : * we still need to clear the md buf */
1860 :
1861 26 : bdev_io_push_bounce_md_buf(bdev_io);
1862 26 : }
1863 :
1864 : static void
1865 2 : bdev_io_push_bounce_data_done_and_track(void *ctx, int status)
1866 : {
1867 2 : struct spdk_bdev_io *bdev_io = ctx;
1868 2 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1869 :
1870 2 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1871 2 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1872 :
1873 2 : if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
1874 0 : bdev_ch_retry_io(ch);
1875 0 : }
1876 :
1877 2 : bdev_io_push_bounce_data_done(bdev_io, status);
1878 2 : }
1879 :
1880 : static inline void
1881 27 : bdev_io_push_bounce_data(struct spdk_bdev_io *bdev_io)
1882 : {
1883 27 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
1884 27 : int rc = 0;
1885 :
1886 27 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS);
1887 27 : assert(!bdev_io_use_accel_sequence(bdev_io));
1888 27 : assert(bdev_io->internal.f.has_bounce_buf);
1889 :
1890 : /* if this is read path, copy data from bounce buffer to original buffer */
1891 27 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_READ) {
1892 11 : if (bdev_io_use_memory_domain(bdev_io)) {
1893 3 : TAILQ_INSERT_TAIL(&ch->io_memory_domain, bdev_io, internal.link);
1894 3 : bdev_io_increment_outstanding(ch, ch->shared_resource);
1895 : /* If memory domain is used then we need to call async push function */
1896 6 : rc = spdk_memory_domain_push_data(bdev_io->internal.memory_domain,
1897 3 : bdev_io->internal.memory_domain_ctx,
1898 3 : bdev_io->internal.bounce_buf.orig_iovs,
1899 3 : (uint32_t)bdev_io->internal.bounce_buf.orig_iovcnt,
1900 3 : &bdev_io->internal.bounce_buf.iov, 1,
1901 : bdev_io_push_bounce_data_done_and_track,
1902 3 : bdev_io);
1903 3 : if (rc == 0) {
1904 : /* Continue IO completion in async callback */
1905 2 : return;
1906 : }
1907 :
1908 1 : TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
1909 1 : bdev_io_decrement_outstanding(ch, ch->shared_resource);
1910 1 : if (rc != -ENOMEM) {
1911 0 : SPDK_ERRLOG("Failed to push data to memory domain %s\n",
1912 : spdk_memory_domain_get_dma_device_id(
1913 : bdev_io->internal.memory_domain));
1914 0 : }
1915 1 : } else {
1916 16 : spdk_copy_buf_to_iovs(bdev_io->internal.bounce_buf.orig_iovs,
1917 8 : bdev_io->internal.bounce_buf.orig_iovcnt,
1918 8 : bdev_io->internal.bounce_buf.iov.iov_base,
1919 8 : bdev_io->internal.bounce_buf.iov.iov_len);
1920 : }
1921 9 : }
1922 :
1923 25 : if (spdk_unlikely(rc == -ENOMEM)) {
1924 1 : bdev_queue_nomem_io_head(ch->shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PUSH);
1925 1 : } else {
1926 24 : bdev_io_push_bounce_data_done(bdev_io, rc);
1927 : }
1928 27 : }
1929 :
1930 : static inline void
1931 26 : _bdev_io_push_bounce_data_buffer(struct spdk_bdev_io *bdev_io, bdev_copy_bounce_buffer_cpl cpl_cb)
1932 : {
1933 26 : bdev_io->internal.data_transfer_cpl = cpl_cb;
1934 26 : bdev_io_push_bounce_data(bdev_io);
1935 26 : }
1936 :
1937 : static void
1938 0 : bdev_io_get_iobuf_cb(struct spdk_iobuf_entry *iobuf, void *buf)
1939 : {
1940 : struct spdk_bdev_io *bdev_io;
1941 :
1942 0 : bdev_io = SPDK_CONTAINEROF(iobuf, struct spdk_bdev_io, internal.iobuf);
1943 0 : _bdev_io_set_buf(bdev_io, buf, bdev_io->internal.buf.len);
1944 0 : }
1945 :
1946 : static void
1947 42 : bdev_io_get_buf(struct spdk_bdev_io *bdev_io, uint64_t len)
1948 : {
1949 : struct spdk_bdev_mgmt_channel *mgmt_ch;
1950 : uint64_t max_len;
1951 : void *buf;
1952 :
1953 42 : assert(spdk_bdev_io_get_thread(bdev_io) == spdk_get_thread());
1954 42 : mgmt_ch = bdev_io->internal.ch->shared_resource->mgmt_ch;
1955 42 : max_len = bdev_io_get_max_buf_len(bdev_io, len);
1956 :
1957 42 : if (spdk_unlikely(max_len > mgmt_ch->iobuf.cache[0].large.bufsize)) {
1958 0 : SPDK_ERRLOG("Length %" PRIu64 " is larger than allowed\n", max_len);
1959 0 : bdev_io_get_buf_complete(bdev_io, false);
1960 0 : return;
1961 : }
1962 :
1963 42 : bdev_io->internal.buf.len = len;
1964 42 : buf = spdk_iobuf_get(&mgmt_ch->iobuf, max_len, &bdev_io->internal.iobuf,
1965 : bdev_io_get_iobuf_cb);
1966 42 : if (buf != NULL) {
1967 42 : _bdev_io_set_buf(bdev_io, buf, len);
1968 42 : }
1969 42 : }
1970 :
1971 : void
1972 56 : spdk_bdev_io_get_buf(struct spdk_bdev_io *bdev_io, spdk_bdev_io_get_buf_cb cb, uint64_t len)
1973 : {
1974 56 : struct spdk_bdev *bdev = bdev_io->bdev;
1975 : uint64_t alignment;
1976 :
1977 56 : assert(cb != NULL);
1978 56 : bdev_io->internal.get_buf_cb = cb;
1979 :
1980 56 : alignment = spdk_bdev_get_buf_align(bdev);
1981 :
1982 56 : if (_is_buf_allocated(bdev_io->u.bdev.iovs) &&
1983 40 : _are_iovs_aligned(bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt, alignment)) {
1984 : /* Buffer already present and aligned */
1985 18 : cb(spdk_bdev_io_get_io_channel(bdev_io), bdev_io, true);
1986 18 : return;
1987 : }
1988 :
1989 38 : bdev_io_get_buf(bdev_io, len);
1990 56 : }
1991 :
1992 : static void
1993 4 : _bdev_io_get_bounce_buf(struct spdk_bdev_io *bdev_io, spdk_bdev_io_get_buf_cb cb,
1994 : uint64_t len)
1995 : {
1996 4 : assert(cb != NULL);
1997 4 : bdev_io->internal.get_buf_cb = cb;
1998 :
1999 4 : bdev_io_get_buf(bdev_io, len);
2000 4 : }
2001 :
2002 : static void
2003 0 : _bdev_io_get_accel_buf(struct spdk_bdev_io *bdev_io)
2004 : {
2005 0 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
2006 : void *buf;
2007 : int rc;
2008 :
2009 0 : rc = spdk_accel_get_buf(ch->accel_channel,
2010 0 : bdev_io->internal.buf.len,
2011 : &buf,
2012 0 : &bdev_io->u.bdev.memory_domain,
2013 0 : &bdev_io->u.bdev.memory_domain_ctx);
2014 0 : if (rc != 0) {
2015 0 : bdev_queue_nomem_io_tail(ch->shared_resource, bdev_io,
2016 : BDEV_IO_RETRY_STATE_GET_ACCEL_BUF);
2017 0 : return;
2018 : }
2019 :
2020 0 : _bdev_io_set_buf(bdev_io, buf, bdev_io->internal.buf.len);
2021 0 : }
2022 :
2023 : static inline void
2024 0 : bdev_io_get_accel_buf(struct spdk_bdev_io *bdev_io, spdk_bdev_io_get_buf_cb cb,
2025 : uint64_t len)
2026 : {
2027 0 : bdev_io->internal.buf.len = len;
2028 0 : bdev_io->internal.get_buf_cb = cb;
2029 :
2030 0 : _bdev_io_get_accel_buf(bdev_io);
2031 0 : }
2032 :
2033 3 : SPDK_LOG_DEPRECATION_REGISTER(spdk_bdev_io_get_aux_buf,
2034 : "spdk_bdev_io_get_aux_buf is deprecated", "v25.01", 0);
2035 :
2036 : void
2037 0 : spdk_bdev_io_get_aux_buf(struct spdk_bdev_io *bdev_io, spdk_bdev_io_get_aux_buf_cb cb)
2038 : {
2039 0 : uint64_t len = bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen;
2040 :
2041 0 : SPDK_LOG_DEPRECATED(spdk_bdev_io_get_aux_buf);
2042 :
2043 0 : assert(cb != NULL);
2044 0 : assert(bdev_io->internal.get_aux_buf_cb == NULL);
2045 0 : bdev_io->internal.get_aux_buf_cb = cb;
2046 0 : bdev_io_get_buf(bdev_io, len);
2047 0 : }
2048 :
2049 : static int
2050 68 : bdev_module_get_max_ctx_size(void)
2051 : {
2052 : struct spdk_bdev_module *bdev_module;
2053 68 : int max_bdev_module_size = 0;
2054 :
2055 266 : TAILQ_FOREACH(bdev_module, &g_bdev_mgr.bdev_modules, internal.tailq) {
2056 198 : if (bdev_module->get_ctx_size && bdev_module->get_ctx_size() > max_bdev_module_size) {
2057 67 : max_bdev_module_size = bdev_module->get_ctx_size();
2058 67 : }
2059 198 : }
2060 :
2061 68 : return max_bdev_module_size;
2062 : }
2063 :
2064 : static void
2065 0 : bdev_enable_histogram_config_json(struct spdk_bdev *bdev, struct spdk_json_write_ctx *w)
2066 : {
2067 0 : if (!bdev->internal.histogram_enabled) {
2068 0 : return;
2069 : }
2070 :
2071 0 : spdk_json_write_object_begin(w);
2072 0 : spdk_json_write_named_string(w, "method", "bdev_enable_histogram");
2073 :
2074 0 : spdk_json_write_named_object_begin(w, "params");
2075 0 : spdk_json_write_named_string(w, "name", bdev->name);
2076 :
2077 0 : spdk_json_write_named_bool(w, "enable", bdev->internal.histogram_enabled);
2078 :
2079 0 : if (bdev->internal.histogram_io_type) {
2080 0 : spdk_json_write_named_string(w, "opc",
2081 0 : spdk_bdev_get_io_type_name(bdev->internal.histogram_io_type));
2082 0 : }
2083 :
2084 0 : spdk_json_write_object_end(w);
2085 :
2086 0 : spdk_json_write_object_end(w);
2087 0 : }
2088 :
2089 : static void
2090 0 : bdev_qos_config_json(struct spdk_bdev *bdev, struct spdk_json_write_ctx *w)
2091 : {
2092 : int i;
2093 0 : struct spdk_bdev_qos *qos = bdev->internal.qos;
2094 : uint64_t limits[SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES];
2095 :
2096 0 : if (!qos) {
2097 0 : return;
2098 : }
2099 :
2100 0 : spdk_bdev_get_qos_rate_limits(bdev, limits);
2101 :
2102 0 : spdk_json_write_object_begin(w);
2103 0 : spdk_json_write_named_string(w, "method", "bdev_set_qos_limit");
2104 :
2105 0 : spdk_json_write_named_object_begin(w, "params");
2106 0 : spdk_json_write_named_string(w, "name", bdev->name);
2107 0 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
2108 0 : if (limits[i] > 0) {
2109 0 : spdk_json_write_named_uint64(w, qos_rpc_type[i], limits[i]);
2110 0 : }
2111 0 : }
2112 0 : spdk_json_write_object_end(w);
2113 :
2114 0 : spdk_json_write_object_end(w);
2115 0 : }
2116 :
2117 : void
2118 0 : spdk_bdev_subsystem_config_json(struct spdk_json_write_ctx *w)
2119 : {
2120 : struct spdk_bdev_module *bdev_module;
2121 : struct spdk_bdev *bdev;
2122 :
2123 0 : assert(w != NULL);
2124 :
2125 0 : spdk_json_write_array_begin(w);
2126 :
2127 0 : spdk_json_write_object_begin(w);
2128 0 : spdk_json_write_named_string(w, "method", "bdev_set_options");
2129 0 : spdk_json_write_named_object_begin(w, "params");
2130 0 : spdk_json_write_named_uint32(w, "bdev_io_pool_size", g_bdev_opts.bdev_io_pool_size);
2131 0 : spdk_json_write_named_uint32(w, "bdev_io_cache_size", g_bdev_opts.bdev_io_cache_size);
2132 0 : spdk_json_write_named_bool(w, "bdev_auto_examine", g_bdev_opts.bdev_auto_examine);
2133 0 : spdk_json_write_named_uint32(w, "iobuf_small_cache_size", g_bdev_opts.iobuf_small_cache_size);
2134 0 : spdk_json_write_named_uint32(w, "iobuf_large_cache_size", g_bdev_opts.iobuf_large_cache_size);
2135 0 : spdk_json_write_object_end(w);
2136 0 : spdk_json_write_object_end(w);
2137 :
2138 0 : bdev_examine_allowlist_config_json(w);
2139 :
2140 0 : TAILQ_FOREACH(bdev_module, &g_bdev_mgr.bdev_modules, internal.tailq) {
2141 0 : if (bdev_module->config_json) {
2142 0 : bdev_module->config_json(w);
2143 0 : }
2144 0 : }
2145 :
2146 0 : spdk_spin_lock(&g_bdev_mgr.spinlock);
2147 :
2148 0 : TAILQ_FOREACH(bdev, &g_bdev_mgr.bdevs, internal.link) {
2149 0 : if (bdev->fn_table->write_config_json) {
2150 0 : bdev->fn_table->write_config_json(bdev, w);
2151 0 : }
2152 :
2153 0 : bdev_qos_config_json(bdev, w);
2154 0 : bdev_enable_histogram_config_json(bdev, w);
2155 0 : }
2156 :
2157 0 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
2158 :
2159 : /* This has to be last RPC in array to make sure all bdevs finished examine */
2160 0 : spdk_json_write_object_begin(w);
2161 0 : spdk_json_write_named_string(w, "method", "bdev_wait_for_examine");
2162 0 : spdk_json_write_object_end(w);
2163 :
2164 0 : spdk_json_write_array_end(w);
2165 0 : }
2166 :
2167 : static void
2168 72 : bdev_mgmt_channel_destroy(void *io_device, void *ctx_buf)
2169 : {
2170 72 : struct spdk_bdev_mgmt_channel *ch = ctx_buf;
2171 : struct spdk_bdev_io *bdev_io;
2172 :
2173 72 : spdk_iobuf_channel_fini(&ch->iobuf);
2174 :
2175 10226 : while (!STAILQ_EMPTY(&ch->per_thread_cache)) {
2176 10154 : bdev_io = STAILQ_FIRST(&ch->per_thread_cache);
2177 10154 : STAILQ_REMOVE_HEAD(&ch->per_thread_cache, internal.buf_link);
2178 10154 : ch->per_thread_cache_count--;
2179 10154 : spdk_mempool_put(g_bdev_mgr.bdev_io_pool, (void *)bdev_io);
2180 : }
2181 :
2182 72 : assert(ch->per_thread_cache_count == 0);
2183 72 : }
2184 :
2185 : static int
2186 72 : bdev_mgmt_channel_create(void *io_device, void *ctx_buf)
2187 : {
2188 72 : struct spdk_bdev_mgmt_channel *ch = ctx_buf;
2189 : struct spdk_bdev_io *bdev_io;
2190 : uint32_t i;
2191 : int rc;
2192 :
2193 144 : rc = spdk_iobuf_channel_init(&ch->iobuf, "bdev",
2194 72 : g_bdev_opts.iobuf_small_cache_size,
2195 72 : g_bdev_opts.iobuf_large_cache_size);
2196 72 : if (rc != 0) {
2197 0 : SPDK_ERRLOG("Failed to create iobuf channel: %s\n", spdk_strerror(-rc));
2198 0 : return -1;
2199 : }
2200 :
2201 72 : STAILQ_INIT(&ch->per_thread_cache);
2202 72 : ch->bdev_io_cache_size = g_bdev_opts.bdev_io_cache_size;
2203 :
2204 : /* Pre-populate bdev_io cache to ensure this thread cannot be starved. */
2205 72 : ch->per_thread_cache_count = 0;
2206 10226 : for (i = 0; i < ch->bdev_io_cache_size; i++) {
2207 10154 : bdev_io = spdk_mempool_get(g_bdev_mgr.bdev_io_pool);
2208 10154 : if (bdev_io == NULL) {
2209 0 : SPDK_ERRLOG("You need to increase bdev_io_pool_size using bdev_set_options RPC.\n");
2210 0 : assert(false);
2211 : bdev_mgmt_channel_destroy(io_device, ctx_buf);
2212 : return -1;
2213 : }
2214 10154 : ch->per_thread_cache_count++;
2215 10154 : STAILQ_INSERT_HEAD(&ch->per_thread_cache, bdev_io, internal.buf_link);
2216 10154 : }
2217 :
2218 72 : TAILQ_INIT(&ch->shared_resources);
2219 72 : TAILQ_INIT(&ch->io_wait_queue);
2220 :
2221 72 : return 0;
2222 72 : }
2223 :
2224 : static void
2225 68 : bdev_init_complete(int rc)
2226 : {
2227 68 : spdk_bdev_init_cb cb_fn = g_init_cb_fn;
2228 68 : void *cb_arg = g_init_cb_arg;
2229 : struct spdk_bdev_module *m;
2230 :
2231 68 : g_bdev_mgr.init_complete = true;
2232 68 : g_init_cb_fn = NULL;
2233 68 : g_init_cb_arg = NULL;
2234 :
2235 : /*
2236 : * For modules that need to know when subsystem init is complete,
2237 : * inform them now.
2238 : */
2239 68 : if (rc == 0) {
2240 266 : TAILQ_FOREACH(m, &g_bdev_mgr.bdev_modules, internal.tailq) {
2241 198 : if (m->init_complete) {
2242 24 : m->init_complete();
2243 24 : }
2244 198 : }
2245 68 : }
2246 :
2247 68 : cb_fn(cb_arg, rc);
2248 68 : }
2249 :
2250 : static bool
2251 271 : bdev_module_all_actions_completed(void)
2252 : {
2253 : struct spdk_bdev_module *m;
2254 :
2255 1078 : TAILQ_FOREACH(m, &g_bdev_mgr.bdev_modules, internal.tailq) {
2256 807 : if (m->internal.action_in_progress > 0) {
2257 0 : return false;
2258 : }
2259 807 : }
2260 271 : return true;
2261 271 : }
2262 :
2263 : static void
2264 629 : bdev_module_action_complete(void)
2265 : {
2266 : /*
2267 : * Don't finish bdev subsystem initialization if
2268 : * module pre-initialization is still in progress, or
2269 : * the subsystem been already initialized.
2270 : */
2271 629 : if (!g_bdev_mgr.module_init_complete || g_bdev_mgr.init_complete) {
2272 561 : return;
2273 : }
2274 :
2275 : /*
2276 : * Check all bdev modules for inits/examinations in progress. If any
2277 : * exist, return immediately since we cannot finish bdev subsystem
2278 : * initialization until all are completed.
2279 : */
2280 68 : if (!bdev_module_all_actions_completed()) {
2281 0 : return;
2282 : }
2283 :
2284 : /*
2285 : * Modules already finished initialization - now that all
2286 : * the bdev modules have finished their asynchronous I/O
2287 : * processing, the entire bdev layer can be marked as complete.
2288 : */
2289 68 : bdev_init_complete(0);
2290 629 : }
2291 :
2292 : static void
2293 561 : bdev_module_action_done(struct spdk_bdev_module *module)
2294 : {
2295 561 : spdk_spin_lock(&module->internal.spinlock);
2296 561 : assert(module->internal.action_in_progress > 0);
2297 561 : module->internal.action_in_progress--;
2298 561 : spdk_spin_unlock(&module->internal.spinlock);
2299 561 : bdev_module_action_complete();
2300 561 : }
2301 :
2302 : void
2303 68 : spdk_bdev_module_init_done(struct spdk_bdev_module *module)
2304 : {
2305 68 : assert(module->async_init);
2306 68 : bdev_module_action_done(module);
2307 68 : }
2308 :
2309 : void
2310 493 : spdk_bdev_module_examine_done(struct spdk_bdev_module *module)
2311 : {
2312 493 : bdev_module_action_done(module);
2313 493 : }
2314 :
2315 : /** The last initialized bdev module */
2316 : static struct spdk_bdev_module *g_resume_bdev_module = NULL;
2317 :
2318 : static void
2319 0 : bdev_init_failed(void *cb_arg)
2320 : {
2321 0 : struct spdk_bdev_module *module = cb_arg;
2322 :
2323 0 : spdk_spin_lock(&module->internal.spinlock);
2324 0 : assert(module->internal.action_in_progress > 0);
2325 0 : module->internal.action_in_progress--;
2326 0 : spdk_spin_unlock(&module->internal.spinlock);
2327 0 : bdev_init_complete(-1);
2328 0 : }
2329 :
2330 : static int
2331 68 : bdev_modules_init(void)
2332 : {
2333 : struct spdk_bdev_module *module;
2334 68 : int rc = 0;
2335 :
2336 266 : TAILQ_FOREACH(module, &g_bdev_mgr.bdev_modules, internal.tailq) {
2337 198 : g_resume_bdev_module = module;
2338 198 : if (module->async_init) {
2339 68 : spdk_spin_lock(&module->internal.spinlock);
2340 68 : module->internal.action_in_progress = 1;
2341 68 : spdk_spin_unlock(&module->internal.spinlock);
2342 68 : }
2343 198 : rc = module->module_init();
2344 198 : if (rc != 0) {
2345 : /* Bump action_in_progress to prevent other modules from completion of modules_init
2346 : * Send message to defer application shutdown until resources are cleaned up */
2347 0 : spdk_spin_lock(&module->internal.spinlock);
2348 0 : module->internal.action_in_progress = 1;
2349 0 : spdk_spin_unlock(&module->internal.spinlock);
2350 0 : spdk_thread_send_msg(spdk_get_thread(), bdev_init_failed, module);
2351 0 : return rc;
2352 : }
2353 198 : }
2354 :
2355 68 : g_resume_bdev_module = NULL;
2356 68 : return 0;
2357 68 : }
2358 :
2359 : void
2360 68 : spdk_bdev_initialize(spdk_bdev_init_cb cb_fn, void *cb_arg)
2361 : {
2362 68 : int rc = 0;
2363 : char mempool_name[32];
2364 :
2365 68 : assert(cb_fn != NULL);
2366 :
2367 68 : g_init_cb_fn = cb_fn;
2368 68 : g_init_cb_arg = cb_arg;
2369 :
2370 68 : spdk_notify_type_register("bdev_register");
2371 68 : spdk_notify_type_register("bdev_unregister");
2372 :
2373 68 : snprintf(mempool_name, sizeof(mempool_name), "bdev_io_%d", getpid());
2374 :
2375 68 : rc = spdk_iobuf_register_module("bdev");
2376 68 : if (rc != 0) {
2377 0 : SPDK_ERRLOG("could not register bdev iobuf module: %s\n", spdk_strerror(-rc));
2378 0 : bdev_init_complete(-1);
2379 0 : return;
2380 : }
2381 :
2382 136 : g_bdev_mgr.bdev_io_pool = spdk_mempool_create(mempool_name,
2383 68 : g_bdev_opts.bdev_io_pool_size,
2384 68 : sizeof(struct spdk_bdev_io) +
2385 68 : bdev_module_get_max_ctx_size(),
2386 : 0,
2387 : SPDK_ENV_NUMA_ID_ANY);
2388 :
2389 68 : if (g_bdev_mgr.bdev_io_pool == NULL) {
2390 0 : SPDK_ERRLOG("could not allocate spdk_bdev_io pool\n");
2391 0 : bdev_init_complete(-1);
2392 0 : return;
2393 : }
2394 :
2395 68 : g_bdev_mgr.zero_buffer = spdk_zmalloc(ZERO_BUFFER_SIZE, ZERO_BUFFER_SIZE,
2396 : NULL, SPDK_ENV_LCORE_ID_ANY, SPDK_MALLOC_DMA);
2397 68 : if (!g_bdev_mgr.zero_buffer) {
2398 0 : SPDK_ERRLOG("create bdev zero buffer failed\n");
2399 0 : bdev_init_complete(-1);
2400 0 : return;
2401 : }
2402 :
2403 : #ifdef SPDK_CONFIG_VTUNE
2404 : g_bdev_mgr.domain = __itt_domain_create("spdk_bdev");
2405 : #endif
2406 :
2407 68 : spdk_io_device_register(&g_bdev_mgr, bdev_mgmt_channel_create,
2408 : bdev_mgmt_channel_destroy,
2409 : sizeof(struct spdk_bdev_mgmt_channel),
2410 : "bdev_mgr");
2411 :
2412 68 : rc = bdev_modules_init();
2413 68 : g_bdev_mgr.module_init_complete = true;
2414 68 : if (rc != 0) {
2415 0 : SPDK_ERRLOG("bdev modules init failed\n");
2416 0 : return;
2417 : }
2418 :
2419 68 : bdev_module_action_complete();
2420 68 : }
2421 :
2422 : static void
2423 68 : bdev_mgr_unregister_cb(void *io_device)
2424 : {
2425 68 : spdk_bdev_fini_cb cb_fn = g_fini_cb_fn;
2426 :
2427 68 : if (g_bdev_mgr.bdev_io_pool) {
2428 68 : if (spdk_mempool_count(g_bdev_mgr.bdev_io_pool) != g_bdev_opts.bdev_io_pool_size) {
2429 0 : SPDK_ERRLOG("bdev IO pool count is %zu but should be %u\n",
2430 : spdk_mempool_count(g_bdev_mgr.bdev_io_pool),
2431 : g_bdev_opts.bdev_io_pool_size);
2432 0 : }
2433 :
2434 68 : spdk_mempool_free(g_bdev_mgr.bdev_io_pool);
2435 68 : }
2436 :
2437 68 : spdk_free(g_bdev_mgr.zero_buffer);
2438 :
2439 68 : bdev_examine_allowlist_free();
2440 :
2441 68 : cb_fn(g_fini_cb_arg);
2442 68 : g_fini_cb_fn = NULL;
2443 68 : g_fini_cb_arg = NULL;
2444 68 : g_bdev_mgr.init_complete = false;
2445 68 : g_bdev_mgr.module_init_complete = false;
2446 68 : }
2447 :
2448 : static void
2449 68 : bdev_module_fini_iter(void *arg)
2450 : {
2451 : struct spdk_bdev_module *bdev_module;
2452 :
2453 : /* FIXME: Handling initialization failures is broken now,
2454 : * so we won't even try cleaning up after successfully
2455 : * initialized modules. if module_init_complete is false,
2456 : * just call spdk_bdev_mgr_unregister_cb
2457 : */
2458 68 : if (!g_bdev_mgr.module_init_complete) {
2459 0 : bdev_mgr_unregister_cb(NULL);
2460 0 : return;
2461 : }
2462 :
2463 : /* Start iterating from the last touched module */
2464 68 : if (!g_resume_bdev_module) {
2465 68 : bdev_module = TAILQ_LAST(&g_bdev_mgr.bdev_modules, bdev_module_list);
2466 68 : } else {
2467 0 : bdev_module = TAILQ_PREV(g_resume_bdev_module, bdev_module_list,
2468 : internal.tailq);
2469 : }
2470 :
2471 266 : while (bdev_module) {
2472 198 : if (bdev_module->async_fini) {
2473 : /* Save our place so we can resume later. We must
2474 : * save the variable here, before calling module_fini()
2475 : * below, because in some cases the module may immediately
2476 : * call spdk_bdev_module_fini_done() and re-enter
2477 : * this function to continue iterating. */
2478 0 : g_resume_bdev_module = bdev_module;
2479 0 : }
2480 :
2481 198 : if (bdev_module->module_fini) {
2482 198 : bdev_module->module_fini();
2483 198 : }
2484 :
2485 198 : if (bdev_module->async_fini) {
2486 0 : return;
2487 : }
2488 :
2489 198 : bdev_module = TAILQ_PREV(bdev_module, bdev_module_list,
2490 : internal.tailq);
2491 : }
2492 :
2493 68 : g_resume_bdev_module = NULL;
2494 68 : spdk_io_device_unregister(&g_bdev_mgr, bdev_mgr_unregister_cb);
2495 68 : }
2496 :
2497 : void
2498 0 : spdk_bdev_module_fini_done(void)
2499 : {
2500 0 : if (spdk_get_thread() != g_fini_thread) {
2501 0 : spdk_thread_send_msg(g_fini_thread, bdev_module_fini_iter, NULL);
2502 0 : } else {
2503 0 : bdev_module_fini_iter(NULL);
2504 : }
2505 0 : }
2506 :
2507 : static void
2508 68 : bdev_finish_unregister_bdevs_iter(void *cb_arg, int bdeverrno)
2509 : {
2510 68 : struct spdk_bdev *bdev = cb_arg;
2511 :
2512 68 : if (bdeverrno && bdev) {
2513 0 : SPDK_WARNLOG("Unable to unregister bdev '%s' during spdk_bdev_finish()\n",
2514 : bdev->name);
2515 :
2516 : /*
2517 : * Since the call to spdk_bdev_unregister() failed, we have no way to free this
2518 : * bdev; try to continue by manually removing this bdev from the list and continue
2519 : * with the next bdev in the list.
2520 : */
2521 0 : TAILQ_REMOVE(&g_bdev_mgr.bdevs, bdev, internal.link);
2522 0 : }
2523 :
2524 68 : if (TAILQ_EMPTY(&g_bdev_mgr.bdevs)) {
2525 68 : SPDK_DEBUGLOG(bdev, "Done unregistering bdevs\n");
2526 : /*
2527 : * Bdev module finish need to be deferred as we might be in the middle of some context
2528 : * (like bdev part free) that will use this bdev (or private bdev driver ctx data)
2529 : * after returning.
2530 : */
2531 68 : spdk_thread_send_msg(spdk_get_thread(), bdev_module_fini_iter, NULL);
2532 68 : return;
2533 : }
2534 :
2535 : /*
2536 : * Unregister last unclaimed bdev in the list, to ensure that bdev subsystem
2537 : * shutdown proceeds top-down. The goal is to give virtual bdevs an opportunity
2538 : * to detect clean shutdown as opposed to run-time hot removal of the underlying
2539 : * base bdevs.
2540 : *
2541 : * Also, walk the list in the reverse order.
2542 : */
2543 0 : for (bdev = TAILQ_LAST(&g_bdev_mgr.bdevs, spdk_bdev_list);
2544 0 : bdev; bdev = TAILQ_PREV(bdev, spdk_bdev_list, internal.link)) {
2545 0 : spdk_spin_lock(&bdev->internal.spinlock);
2546 0 : if (bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE) {
2547 0 : LOG_ALREADY_CLAIMED_DEBUG("claimed, skipping", bdev);
2548 0 : spdk_spin_unlock(&bdev->internal.spinlock);
2549 0 : continue;
2550 : }
2551 0 : spdk_spin_unlock(&bdev->internal.spinlock);
2552 :
2553 0 : SPDK_DEBUGLOG(bdev, "Unregistering bdev '%s'\n", bdev->name);
2554 0 : spdk_bdev_unregister(bdev, bdev_finish_unregister_bdevs_iter, bdev);
2555 0 : return;
2556 : }
2557 :
2558 : /*
2559 : * If any bdev fails to unclaim underlying bdev properly, we may face the
2560 : * case of bdev list consisting of claimed bdevs only (if claims are managed
2561 : * correctly, this would mean there's a loop in the claims graph which is
2562 : * clearly impossible). Warn and unregister last bdev on the list then.
2563 : */
2564 0 : for (bdev = TAILQ_LAST(&g_bdev_mgr.bdevs, spdk_bdev_list);
2565 0 : bdev; bdev = TAILQ_PREV(bdev, spdk_bdev_list, internal.link)) {
2566 0 : SPDK_WARNLOG("Unregistering claimed bdev '%s'!\n", bdev->name);
2567 0 : spdk_bdev_unregister(bdev, bdev_finish_unregister_bdevs_iter, bdev);
2568 0 : return;
2569 : }
2570 68 : }
2571 :
2572 : static void
2573 68 : bdev_module_fini_start_iter(void *arg)
2574 : {
2575 : struct spdk_bdev_module *bdev_module;
2576 :
2577 68 : if (!g_resume_bdev_module) {
2578 68 : bdev_module = TAILQ_LAST(&g_bdev_mgr.bdev_modules, bdev_module_list);
2579 68 : } else {
2580 0 : bdev_module = TAILQ_PREV(g_resume_bdev_module, bdev_module_list, internal.tailq);
2581 : }
2582 :
2583 266 : while (bdev_module) {
2584 198 : if (bdev_module->async_fini_start) {
2585 : /* Save our place so we can resume later. We must
2586 : * save the variable here, before calling fini_start()
2587 : * below, because in some cases the module may immediately
2588 : * call spdk_bdev_module_fini_start_done() and re-enter
2589 : * this function to continue iterating. */
2590 0 : g_resume_bdev_module = bdev_module;
2591 0 : }
2592 :
2593 198 : if (bdev_module->fini_start) {
2594 24 : bdev_module->fini_start();
2595 24 : }
2596 :
2597 198 : if (bdev_module->async_fini_start) {
2598 0 : return;
2599 : }
2600 :
2601 198 : bdev_module = TAILQ_PREV(bdev_module, bdev_module_list, internal.tailq);
2602 : }
2603 :
2604 68 : g_resume_bdev_module = NULL;
2605 :
2606 68 : bdev_finish_unregister_bdevs_iter(NULL, 0);
2607 68 : }
2608 :
2609 : void
2610 0 : spdk_bdev_module_fini_start_done(void)
2611 : {
2612 0 : if (spdk_get_thread() != g_fini_thread) {
2613 0 : spdk_thread_send_msg(g_fini_thread, bdev_module_fini_start_iter, NULL);
2614 0 : } else {
2615 0 : bdev_module_fini_start_iter(NULL);
2616 : }
2617 0 : }
2618 :
2619 : static void
2620 68 : bdev_finish_wait_for_examine_done(void *cb_arg)
2621 : {
2622 68 : bdev_module_fini_start_iter(NULL);
2623 68 : }
2624 :
2625 : static void bdev_open_async_fini(void);
2626 :
2627 : void
2628 68 : spdk_bdev_finish(spdk_bdev_fini_cb cb_fn, void *cb_arg)
2629 : {
2630 : int rc;
2631 :
2632 68 : assert(cb_fn != NULL);
2633 :
2634 68 : g_fini_thread = spdk_get_thread();
2635 :
2636 68 : g_fini_cb_fn = cb_fn;
2637 68 : g_fini_cb_arg = cb_arg;
2638 :
2639 68 : bdev_open_async_fini();
2640 :
2641 68 : rc = spdk_bdev_wait_for_examine(bdev_finish_wait_for_examine_done, NULL);
2642 68 : if (rc != 0) {
2643 0 : SPDK_ERRLOG("wait_for_examine failed: %s\n", spdk_strerror(-rc));
2644 0 : bdev_finish_wait_for_examine_done(NULL);
2645 0 : }
2646 68 : }
2647 :
2648 : struct spdk_bdev_io *
2649 699 : bdev_channel_get_io(struct spdk_bdev_channel *channel)
2650 : {
2651 699 : struct spdk_bdev_mgmt_channel *ch = channel->shared_resource->mgmt_ch;
2652 : struct spdk_bdev_io *bdev_io;
2653 :
2654 699 : if (ch->per_thread_cache_count > 0) {
2655 639 : bdev_io = STAILQ_FIRST(&ch->per_thread_cache);
2656 639 : STAILQ_REMOVE_HEAD(&ch->per_thread_cache, internal.buf_link);
2657 639 : ch->per_thread_cache_count--;
2658 699 : } else if (spdk_unlikely(!TAILQ_EMPTY(&ch->io_wait_queue))) {
2659 : /*
2660 : * Don't try to look for bdev_ios in the global pool if there are
2661 : * waiters on bdev_ios - we don't want this caller to jump the line.
2662 : */
2663 0 : bdev_io = NULL;
2664 0 : } else {
2665 60 : bdev_io = spdk_mempool_get(g_bdev_mgr.bdev_io_pool);
2666 : }
2667 :
2668 699 : return bdev_io;
2669 : }
2670 :
2671 : void
2672 693 : spdk_bdev_free_io(struct spdk_bdev_io *bdev_io)
2673 : {
2674 : struct spdk_bdev_mgmt_channel *ch;
2675 :
2676 693 : assert(bdev_io != NULL);
2677 693 : assert(bdev_io->internal.status != SPDK_BDEV_IO_STATUS_PENDING);
2678 :
2679 693 : ch = bdev_io->internal.ch->shared_resource->mgmt_ch;
2680 :
2681 693 : if (bdev_io->internal.f.has_buf) {
2682 16 : bdev_io_put_buf(bdev_io);
2683 16 : }
2684 :
2685 693 : if (ch->per_thread_cache_count < ch->bdev_io_cache_size) {
2686 639 : ch->per_thread_cache_count++;
2687 639 : STAILQ_INSERT_HEAD(&ch->per_thread_cache, bdev_io, internal.buf_link);
2688 643 : while (ch->per_thread_cache_count > 0 && !TAILQ_EMPTY(&ch->io_wait_queue)) {
2689 : struct spdk_bdev_io_wait_entry *entry;
2690 :
2691 4 : entry = TAILQ_FIRST(&ch->io_wait_queue);
2692 4 : TAILQ_REMOVE(&ch->io_wait_queue, entry, link);
2693 4 : entry->cb_fn(entry->cb_arg);
2694 : }
2695 639 : } else {
2696 : /* We should never have a full cache with entries on the io wait queue. */
2697 54 : assert(TAILQ_EMPTY(&ch->io_wait_queue));
2698 54 : spdk_mempool_put(g_bdev_mgr.bdev_io_pool, (void *)bdev_io);
2699 : }
2700 693 : }
2701 :
2702 : static bool
2703 72 : bdev_qos_is_iops_rate_limit(enum spdk_bdev_qos_rate_limit_type limit)
2704 : {
2705 72 : assert(limit != SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES);
2706 :
2707 72 : switch (limit) {
2708 : case SPDK_BDEV_QOS_RW_IOPS_RATE_LIMIT:
2709 18 : return true;
2710 : case SPDK_BDEV_QOS_RW_BPS_RATE_LIMIT:
2711 : case SPDK_BDEV_QOS_R_BPS_RATE_LIMIT:
2712 : case SPDK_BDEV_QOS_W_BPS_RATE_LIMIT:
2713 54 : return false;
2714 0 : case SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES:
2715 : default:
2716 0 : return false;
2717 : }
2718 72 : }
2719 :
2720 : static bool
2721 25 : bdev_qos_io_to_limit(struct spdk_bdev_io *bdev_io)
2722 : {
2723 25 : switch (bdev_io->type) {
2724 : case SPDK_BDEV_IO_TYPE_NVME_IO:
2725 : case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
2726 : case SPDK_BDEV_IO_TYPE_READ:
2727 : case SPDK_BDEV_IO_TYPE_WRITE:
2728 23 : return true;
2729 : case SPDK_BDEV_IO_TYPE_ZCOPY:
2730 0 : if (bdev_io->u.bdev.zcopy.start) {
2731 0 : return true;
2732 : } else {
2733 0 : return false;
2734 : }
2735 : default:
2736 2 : return false;
2737 : }
2738 25 : }
2739 :
2740 : static bool
2741 33 : bdev_is_read_io(struct spdk_bdev_io *bdev_io)
2742 : {
2743 33 : switch (bdev_io->type) {
2744 : case SPDK_BDEV_IO_TYPE_NVME_IO:
2745 : case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
2746 : /* Bit 1 (0x2) set for read operation */
2747 0 : if (bdev_io->u.nvme_passthru.cmd.opc & SPDK_NVME_OPC_READ) {
2748 0 : return true;
2749 : } else {
2750 0 : return false;
2751 : }
2752 : case SPDK_BDEV_IO_TYPE_READ:
2753 30 : return true;
2754 : case SPDK_BDEV_IO_TYPE_ZCOPY:
2755 : /* Populate to read from disk */
2756 0 : if (bdev_io->u.bdev.zcopy.populate) {
2757 0 : return true;
2758 : } else {
2759 0 : return false;
2760 : }
2761 : default:
2762 3 : return false;
2763 : }
2764 33 : }
2765 :
2766 : static uint64_t
2767 43 : bdev_get_io_size_in_byte(struct spdk_bdev_io *bdev_io)
2768 : {
2769 43 : uint32_t blocklen = bdev_io_get_block_size(bdev_io);
2770 :
2771 43 : switch (bdev_io->type) {
2772 : case SPDK_BDEV_IO_TYPE_NVME_IO:
2773 : case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
2774 0 : return bdev_io->u.nvme_passthru.nbytes;
2775 : case SPDK_BDEV_IO_TYPE_READ:
2776 : case SPDK_BDEV_IO_TYPE_WRITE:
2777 43 : return bdev_io->u.bdev.num_blocks * blocklen;
2778 : case SPDK_BDEV_IO_TYPE_ZCOPY:
2779 : /* Track the data in the start phase only */
2780 0 : if (bdev_io->u.bdev.zcopy.start) {
2781 0 : return bdev_io->u.bdev.num_blocks * blocklen;
2782 : } else {
2783 0 : return 0;
2784 : }
2785 : default:
2786 0 : return 0;
2787 : }
2788 43 : }
2789 :
2790 : static inline bool
2791 64 : bdev_qos_rw_queue_io(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io, uint64_t delta)
2792 : {
2793 : int64_t remaining_this_timeslice;
2794 :
2795 64 : if (!limit->max_per_timeslice) {
2796 : /* The QoS is disabled */
2797 0 : return false;
2798 : }
2799 :
2800 64 : remaining_this_timeslice = __atomic_sub_fetch(&limit->remaining_this_timeslice, delta,
2801 : __ATOMIC_RELAXED);
2802 64 : if (remaining_this_timeslice + (int64_t)delta > 0) {
2803 : /* There was still a quota for this delta -> the IO shouldn't be queued
2804 : *
2805 : * We allow a slight quota overrun here so an IO bigger than the per-timeslice
2806 : * quota can be allowed once a while. Such overrun then taken into account in
2807 : * the QoS poller, where the next timeslice quota is calculated.
2808 : */
2809 59 : return false;
2810 : }
2811 :
2812 : /* There was no quota for this delta -> the IO should be queued
2813 : * The remaining_this_timeslice must be rewinded so it reflects the real
2814 : * amount of IOs or bytes allowed.
2815 : */
2816 5 : __atomic_add_fetch(
2817 5 : &limit->remaining_this_timeslice, delta, __ATOMIC_RELAXED);
2818 5 : return true;
2819 64 : }
2820 :
2821 : static inline void
2822 5 : bdev_qos_rw_rewind_io(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io, uint64_t delta)
2823 : {
2824 5 : __atomic_add_fetch(&limit->remaining_this_timeslice, delta, __ATOMIC_RELAXED);
2825 5 : }
2826 :
2827 : static bool
2828 23 : bdev_qos_rw_iops_queue(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2829 : {
2830 23 : return bdev_qos_rw_queue_io(limit, io, 1);
2831 : }
2832 :
2833 : static void
2834 3 : bdev_qos_rw_iops_rewind_quota(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2835 : {
2836 3 : bdev_qos_rw_rewind_io(limit, io, 1);
2837 3 : }
2838 :
2839 : static bool
2840 41 : bdev_qos_rw_bps_queue(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2841 : {
2842 41 : return bdev_qos_rw_queue_io(limit, io, bdev_get_io_size_in_byte(io));
2843 : }
2844 :
2845 : static void
2846 2 : bdev_qos_rw_bps_rewind_quota(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2847 : {
2848 2 : bdev_qos_rw_rewind_io(limit, io, bdev_get_io_size_in_byte(io));
2849 2 : }
2850 :
2851 : static bool
2852 19 : bdev_qos_r_bps_queue(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2853 : {
2854 19 : if (bdev_is_read_io(io) == false) {
2855 1 : return false;
2856 : }
2857 :
2858 18 : return bdev_qos_rw_bps_queue(limit, io);
2859 19 : }
2860 :
2861 : static void
2862 0 : bdev_qos_r_bps_rewind_quota(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2863 : {
2864 0 : if (bdev_is_read_io(io) != false) {
2865 0 : bdev_qos_rw_rewind_io(limit, io, bdev_get_io_size_in_byte(io));
2866 0 : }
2867 0 : }
2868 :
2869 : static bool
2870 14 : bdev_qos_w_bps_queue(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2871 : {
2872 14 : if (bdev_is_read_io(io) == true) {
2873 12 : return false;
2874 : }
2875 :
2876 2 : return bdev_qos_rw_bps_queue(limit, io);
2877 14 : }
2878 :
2879 : static void
2880 0 : bdev_qos_w_bps_rewind_quota(struct spdk_bdev_qos_limit *limit, struct spdk_bdev_io *io)
2881 : {
2882 0 : if (bdev_is_read_io(io) != true) {
2883 0 : bdev_qos_rw_rewind_io(limit, io, bdev_get_io_size_in_byte(io));
2884 0 : }
2885 0 : }
2886 :
2887 : static void
2888 10 : bdev_qos_set_ops(struct spdk_bdev_qos *qos)
2889 : {
2890 : int i;
2891 :
2892 50 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
2893 40 : if (qos->rate_limits[i].limit == SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
2894 15 : qos->rate_limits[i].queue_io = NULL;
2895 15 : continue;
2896 : }
2897 :
2898 25 : switch (i) {
2899 : case SPDK_BDEV_QOS_RW_IOPS_RATE_LIMIT:
2900 9 : qos->rate_limits[i].queue_io = bdev_qos_rw_iops_queue;
2901 9 : qos->rate_limits[i].rewind_quota = bdev_qos_rw_iops_rewind_quota;
2902 9 : break;
2903 : case SPDK_BDEV_QOS_RW_BPS_RATE_LIMIT:
2904 7 : qos->rate_limits[i].queue_io = bdev_qos_rw_bps_queue;
2905 7 : qos->rate_limits[i].rewind_quota = bdev_qos_rw_bps_rewind_quota;
2906 7 : break;
2907 : case SPDK_BDEV_QOS_R_BPS_RATE_LIMIT:
2908 5 : qos->rate_limits[i].queue_io = bdev_qos_r_bps_queue;
2909 5 : qos->rate_limits[i].rewind_quota = bdev_qos_r_bps_rewind_quota;
2910 5 : break;
2911 : case SPDK_BDEV_QOS_W_BPS_RATE_LIMIT:
2912 4 : qos->rate_limits[i].queue_io = bdev_qos_w_bps_queue;
2913 4 : qos->rate_limits[i].rewind_quota = bdev_qos_w_bps_rewind_quota;
2914 4 : break;
2915 : default:
2916 0 : break;
2917 : }
2918 25 : }
2919 10 : }
2920 :
2921 : static void
2922 6 : _bdev_io_complete_in_submit(struct spdk_bdev_channel *bdev_ch,
2923 : struct spdk_bdev_io *bdev_io,
2924 : enum spdk_bdev_io_status status)
2925 : {
2926 6 : bdev_io->internal.f.in_submit_request = true;
2927 6 : bdev_io_increment_outstanding(bdev_ch, bdev_ch->shared_resource);
2928 6 : spdk_bdev_io_complete(bdev_io, status);
2929 6 : bdev_io->internal.f.in_submit_request = false;
2930 6 : }
2931 :
2932 : static inline void
2933 574 : bdev_io_do_submit(struct spdk_bdev_channel *bdev_ch, struct spdk_bdev_io *bdev_io)
2934 : {
2935 574 : struct spdk_bdev *bdev = bdev_io->bdev;
2936 574 : struct spdk_io_channel *ch = bdev_ch->channel;
2937 574 : struct spdk_bdev_shared_resource *shared_resource = bdev_ch->shared_resource;
2938 :
2939 574 : if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_ABORT)) {
2940 16 : struct spdk_bdev_mgmt_channel *mgmt_channel = shared_resource->mgmt_ch;
2941 16 : struct spdk_bdev_io *bio_to_abort = bdev_io->u.abort.bio_to_abort;
2942 :
2943 16 : if (bdev_abort_queued_io(&shared_resource->nomem_io, bio_to_abort) ||
2944 16 : bdev_abort_buf_io(mgmt_channel, bio_to_abort)) {
2945 0 : _bdev_io_complete_in_submit(bdev_ch, bdev_io,
2946 : SPDK_BDEV_IO_STATUS_SUCCESS);
2947 0 : return;
2948 : }
2949 16 : }
2950 :
2951 574 : if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE &&
2952 : bdev_io->bdev->split_on_write_unit &&
2953 : bdev_io->u.bdev.num_blocks < bdev_io->bdev->write_unit_size)) {
2954 4 : SPDK_ERRLOG("IO num_blocks %lu does not match the write_unit_size %u\n",
2955 : bdev_io->u.bdev.num_blocks, bdev_io->bdev->write_unit_size);
2956 4 : _bdev_io_complete_in_submit(bdev_ch, bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
2957 4 : return;
2958 : }
2959 :
2960 570 : if (spdk_likely(TAILQ_EMPTY(&shared_resource->nomem_io))) {
2961 527 : bdev_io_increment_outstanding(bdev_ch, shared_resource);
2962 527 : bdev_io->internal.f.in_submit_request = true;
2963 527 : bdev_submit_request(bdev, ch, bdev_io);
2964 527 : bdev_io->internal.f.in_submit_request = false;
2965 527 : } else {
2966 43 : bdev_queue_nomem_io_tail(shared_resource, bdev_io, BDEV_IO_RETRY_STATE_SUBMIT);
2967 43 : if (shared_resource->nomem_threshold == 0 && shared_resource->io_outstanding == 0) {
2968 : /* Special case when we have nomem IOs and no outstanding IOs which completions
2969 : * could trigger retry of queued IOs */
2970 0 : bdev_shared_ch_retry_io(shared_resource);
2971 0 : }
2972 : }
2973 574 : }
2974 :
2975 : static bool
2976 25 : bdev_qos_queue_io(struct spdk_bdev_qos *qos, struct spdk_bdev_io *bdev_io)
2977 : {
2978 : int i;
2979 :
2980 25 : if (bdev_qos_io_to_limit(bdev_io) == true) {
2981 100 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
2982 82 : if (!qos->rate_limits[i].queue_io) {
2983 5 : continue;
2984 : }
2985 :
2986 231 : if (qos->rate_limits[i].queue_io(&qos->rate_limits[i],
2987 154 : bdev_io) == true) {
2988 10 : for (i -= 1; i >= 0 ; i--) {
2989 5 : if (!qos->rate_limits[i].queue_io) {
2990 0 : continue;
2991 : }
2992 :
2993 5 : qos->rate_limits[i].rewind_quota(&qos->rate_limits[i], bdev_io);
2994 5 : }
2995 5 : return true;
2996 : }
2997 72 : }
2998 18 : }
2999 :
3000 20 : return false;
3001 25 : }
3002 :
3003 : static int
3004 27 : bdev_qos_io_submit(struct spdk_bdev_channel *ch, struct spdk_bdev_qos *qos)
3005 : {
3006 27 : struct spdk_bdev_io *bdev_io = NULL, *tmp = NULL;
3007 27 : int submitted_ios = 0;
3008 :
3009 52 : TAILQ_FOREACH_SAFE(bdev_io, &ch->qos_queued_io, internal.link, tmp) {
3010 25 : if (!bdev_qos_queue_io(qos, bdev_io)) {
3011 20 : TAILQ_REMOVE(&ch->qos_queued_io, bdev_io, internal.link);
3012 20 : bdev_io_do_submit(ch, bdev_io);
3013 :
3014 20 : submitted_ios++;
3015 20 : }
3016 25 : }
3017 :
3018 27 : return submitted_ios;
3019 : }
3020 :
3021 : static void
3022 2 : bdev_queue_io_wait_with_cb(struct spdk_bdev_io *bdev_io, spdk_bdev_io_wait_cb cb_fn)
3023 : {
3024 : int rc;
3025 :
3026 2 : bdev_io->internal.waitq_entry.bdev = bdev_io->bdev;
3027 2 : bdev_io->internal.waitq_entry.cb_fn = cb_fn;
3028 2 : bdev_io->internal.waitq_entry.cb_arg = bdev_io;
3029 4 : rc = spdk_bdev_queue_io_wait(bdev_io->bdev, spdk_io_channel_from_ctx(bdev_io->internal.ch),
3030 2 : &bdev_io->internal.waitq_entry);
3031 2 : if (rc != 0) {
3032 0 : SPDK_ERRLOG("Queue IO failed, rc=%d\n", rc);
3033 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
3034 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
3035 0 : }
3036 2 : }
3037 :
3038 : static bool
3039 621 : bdev_rw_should_split(struct spdk_bdev_io *bdev_io)
3040 : {
3041 : uint32_t io_boundary;
3042 621 : struct spdk_bdev *bdev = bdev_io->bdev;
3043 621 : uint32_t max_segment_size = bdev->max_segment_size;
3044 621 : uint32_t max_size = bdev->max_rw_size;
3045 621 : int max_segs = bdev->max_num_segments;
3046 :
3047 621 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE && bdev->split_on_write_unit) {
3048 24 : io_boundary = bdev->write_unit_size;
3049 621 : } else if (bdev->split_on_optimal_io_boundary) {
3050 168 : io_boundary = bdev->optimal_io_boundary;
3051 168 : } else {
3052 429 : io_boundary = 0;
3053 : }
3054 :
3055 621 : if (spdk_likely(!io_boundary && !max_segs && !max_segment_size && !max_size)) {
3056 243 : return false;
3057 : }
3058 :
3059 378 : if (io_boundary) {
3060 : uint64_t start_stripe, end_stripe;
3061 :
3062 192 : start_stripe = bdev_io->u.bdev.offset_blocks;
3063 192 : end_stripe = start_stripe + bdev_io->u.bdev.num_blocks - 1;
3064 : /* Avoid expensive div operations if possible. These spdk_u32 functions are very cheap. */
3065 192 : if (spdk_likely(spdk_u32_is_pow2(io_boundary))) {
3066 192 : start_stripe >>= spdk_u32log2(io_boundary);
3067 192 : end_stripe >>= spdk_u32log2(io_boundary);
3068 192 : } else {
3069 0 : start_stripe /= io_boundary;
3070 0 : end_stripe /= io_boundary;
3071 : }
3072 :
3073 192 : if (start_stripe != end_stripe) {
3074 75 : return true;
3075 : }
3076 117 : }
3077 :
3078 303 : if (max_segs) {
3079 150 : if (bdev_io->u.bdev.iovcnt > max_segs) {
3080 15 : return true;
3081 : }
3082 135 : }
3083 :
3084 288 : if (max_segment_size) {
3085 470 : for (int i = 0; i < bdev_io->u.bdev.iovcnt; i++) {
3086 346 : if (bdev_io->u.bdev.iovs[i].iov_len > max_segment_size) {
3087 12 : return true;
3088 : }
3089 334 : }
3090 124 : }
3091 :
3092 276 : if (max_size) {
3093 52 : if (bdev_io->u.bdev.num_blocks > max_size) {
3094 7 : return true;
3095 : }
3096 45 : }
3097 :
3098 269 : return false;
3099 621 : }
3100 :
3101 : static bool
3102 24 : bdev_unmap_should_split(struct spdk_bdev_io *bdev_io)
3103 : {
3104 : uint32_t num_unmap_segments;
3105 :
3106 24 : if (!bdev_io->bdev->max_unmap || !bdev_io->bdev->max_unmap_segments) {
3107 3 : return false;
3108 : }
3109 21 : num_unmap_segments = spdk_divide_round_up(bdev_io->u.bdev.num_blocks, bdev_io->bdev->max_unmap);
3110 21 : if (num_unmap_segments > bdev_io->bdev->max_unmap_segments) {
3111 4 : return true;
3112 : }
3113 :
3114 17 : return false;
3115 24 : }
3116 :
3117 : static bool
3118 37 : bdev_write_zeroes_should_split(struct spdk_bdev_io *bdev_io)
3119 : {
3120 37 : if (!bdev_io->bdev->max_write_zeroes) {
3121 4 : return false;
3122 : }
3123 :
3124 33 : if (bdev_io->u.bdev.num_blocks > bdev_io->bdev->max_write_zeroes) {
3125 10 : return true;
3126 : }
3127 :
3128 23 : return false;
3129 37 : }
3130 :
3131 : static bool
3132 30 : bdev_copy_should_split(struct spdk_bdev_io *bdev_io)
3133 : {
3134 30 : if (bdev_io->bdev->max_copy != 0 &&
3135 25 : bdev_io->u.bdev.num_blocks > bdev_io->bdev->max_copy) {
3136 6 : return true;
3137 : }
3138 :
3139 24 : return false;
3140 30 : }
3141 :
3142 : static bool
3143 794 : bdev_io_should_split(struct spdk_bdev_io *bdev_io)
3144 : {
3145 794 : switch (bdev_io->type) {
3146 : case SPDK_BDEV_IO_TYPE_READ:
3147 : case SPDK_BDEV_IO_TYPE_WRITE:
3148 621 : return bdev_rw_should_split(bdev_io);
3149 : case SPDK_BDEV_IO_TYPE_UNMAP:
3150 24 : return bdev_unmap_should_split(bdev_io);
3151 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3152 37 : return bdev_write_zeroes_should_split(bdev_io);
3153 : case SPDK_BDEV_IO_TYPE_COPY:
3154 30 : return bdev_copy_should_split(bdev_io);
3155 : default:
3156 82 : return false;
3157 : }
3158 794 : }
3159 :
3160 : static uint32_t
3161 249 : _to_next_boundary(uint64_t offset, uint32_t boundary)
3162 : {
3163 249 : return (boundary - (offset % boundary));
3164 : }
3165 :
3166 : static void bdev_io_split_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg);
3167 :
3168 : static void _bdev_rw_split(void *_bdev_io);
3169 :
3170 : static void bdev_unmap_split(struct spdk_bdev_io *bdev_io);
3171 :
3172 : static void
3173 0 : _bdev_unmap_split(void *_bdev_io)
3174 : {
3175 0 : return bdev_unmap_split((struct spdk_bdev_io *)_bdev_io);
3176 : }
3177 :
3178 : static void bdev_write_zeroes_split(struct spdk_bdev_io *bdev_io);
3179 :
3180 : static void
3181 0 : _bdev_write_zeroes_split(void *_bdev_io)
3182 : {
3183 0 : return bdev_write_zeroes_split((struct spdk_bdev_io *)_bdev_io);
3184 : }
3185 :
3186 : static void bdev_copy_split(struct spdk_bdev_io *bdev_io);
3187 :
3188 : static void
3189 0 : _bdev_copy_split(void *_bdev_io)
3190 : {
3191 0 : return bdev_copy_split((struct spdk_bdev_io *)_bdev_io);
3192 : }
3193 :
3194 : static int
3195 305 : bdev_io_split_submit(struct spdk_bdev_io *bdev_io, struct iovec *iov, int iovcnt, void *md_buf,
3196 : uint64_t num_blocks, uint64_t *offset, uint64_t *remaining)
3197 : {
3198 : int rc;
3199 : uint64_t current_offset, current_remaining, current_src_offset;
3200 : spdk_bdev_io_wait_cb io_wait_fn;
3201 :
3202 305 : current_offset = *offset;
3203 305 : current_remaining = *remaining;
3204 :
3205 305 : assert(bdev_io->internal.f.split);
3206 :
3207 305 : bdev_io->internal.split.outstanding++;
3208 :
3209 305 : io_wait_fn = _bdev_rw_split;
3210 305 : switch (bdev_io->type) {
3211 : case SPDK_BDEV_IO_TYPE_READ:
3212 196 : assert(bdev_io->u.bdev.accel_sequence == NULL);
3213 392 : rc = bdev_readv_blocks_with_md(bdev_io->internal.desc,
3214 196 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
3215 196 : iov, iovcnt, md_buf, current_offset,
3216 196 : num_blocks,
3217 196 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain : NULL,
3218 196 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain_ctx : NULL,
3219 : NULL,
3220 196 : bdev_io->u.bdev.dif_check_flags,
3221 196 : bdev_io_split_done, bdev_io);
3222 196 : break;
3223 : case SPDK_BDEV_IO_TYPE_WRITE:
3224 50 : assert(bdev_io->u.bdev.accel_sequence == NULL);
3225 100 : rc = bdev_writev_blocks_with_md(bdev_io->internal.desc,
3226 50 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
3227 50 : iov, iovcnt, md_buf, current_offset,
3228 50 : num_blocks,
3229 50 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain : NULL,
3230 50 : bdev_io_use_memory_domain(bdev_io) ? bdev_io->internal.memory_domain_ctx : NULL,
3231 : NULL,
3232 50 : bdev_io->u.bdev.dif_check_flags,
3233 50 : bdev_io->u.bdev.nvme_cdw12.raw,
3234 50 : bdev_io->u.bdev.nvme_cdw13.raw,
3235 50 : bdev_io_split_done, bdev_io);
3236 50 : break;
3237 : case SPDK_BDEV_IO_TYPE_UNMAP:
3238 17 : io_wait_fn = _bdev_unmap_split;
3239 34 : rc = spdk_bdev_unmap_blocks(bdev_io->internal.desc,
3240 17 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
3241 17 : current_offset, num_blocks,
3242 17 : bdev_io_split_done, bdev_io);
3243 17 : break;
3244 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3245 23 : io_wait_fn = _bdev_write_zeroes_split;
3246 46 : rc = spdk_bdev_write_zeroes_blocks(bdev_io->internal.desc,
3247 23 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
3248 23 : current_offset, num_blocks,
3249 23 : bdev_io_split_done, bdev_io);
3250 23 : break;
3251 : case SPDK_BDEV_IO_TYPE_COPY:
3252 19 : io_wait_fn = _bdev_copy_split;
3253 38 : current_src_offset = bdev_io->u.bdev.copy.src_offset_blocks +
3254 19 : (current_offset - bdev_io->u.bdev.offset_blocks);
3255 38 : rc = spdk_bdev_copy_blocks(bdev_io->internal.desc,
3256 19 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
3257 19 : current_offset, current_src_offset, num_blocks,
3258 19 : bdev_io_split_done, bdev_io);
3259 19 : break;
3260 : default:
3261 0 : assert(false);
3262 : rc = -EINVAL;
3263 : break;
3264 : }
3265 :
3266 305 : if (rc == 0) {
3267 301 : current_offset += num_blocks;
3268 301 : current_remaining -= num_blocks;
3269 301 : bdev_io->internal.split.current_offset_blocks = current_offset;
3270 301 : bdev_io->internal.split.remaining_num_blocks = current_remaining;
3271 301 : *offset = current_offset;
3272 301 : *remaining = current_remaining;
3273 301 : } else {
3274 4 : bdev_io->internal.split.outstanding--;
3275 4 : if (rc == -ENOMEM) {
3276 4 : if (bdev_io->internal.split.outstanding == 0) {
3277 : /* No I/O is outstanding. Hence we should wait here. */
3278 1 : bdev_queue_io_wait_with_cb(bdev_io, io_wait_fn);
3279 1 : }
3280 4 : } else {
3281 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
3282 0 : if (bdev_io->internal.split.outstanding == 0) {
3283 0 : bdev_ch_remove_from_io_submitted(bdev_io);
3284 0 : spdk_trace_record(TRACE_BDEV_IO_DONE, bdev_io->internal.ch->trace_id,
3285 : 0, (uintptr_t)bdev_io, bdev_io->internal.caller_ctx,
3286 : bdev_io->internal.ch->queue_depth);
3287 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
3288 0 : }
3289 : }
3290 : }
3291 :
3292 305 : return rc;
3293 : }
3294 :
3295 : static void
3296 67 : _bdev_rw_split(void *_bdev_io)
3297 : {
3298 : struct iovec *parent_iov, *iov;
3299 67 : struct spdk_bdev_io *bdev_io = _bdev_io;
3300 67 : struct spdk_bdev *bdev = bdev_io->bdev;
3301 : uint64_t parent_offset, current_offset, remaining;
3302 : uint32_t parent_iov_offset, parent_iovcnt, parent_iovpos, child_iovcnt;
3303 : uint32_t to_next_boundary, to_next_boundary_bytes, to_last_block_bytes;
3304 : uint32_t iovcnt, iov_len, child_iovsize;
3305 : uint32_t blocklen;
3306 : uint32_t io_boundary;
3307 67 : uint32_t max_segment_size = bdev->max_segment_size;
3308 67 : uint32_t max_child_iovcnt = bdev->max_num_segments;
3309 67 : uint32_t max_size = bdev->max_rw_size;
3310 67 : void *md_buf = NULL;
3311 : int rc;
3312 :
3313 67 : blocklen = bdev_io_get_block_size(bdev_io);
3314 :
3315 67 : max_size = max_size ? max_size : UINT32_MAX;
3316 67 : max_segment_size = max_segment_size ? max_segment_size : UINT32_MAX;
3317 67 : max_child_iovcnt = max_child_iovcnt ? spdk_min(max_child_iovcnt, SPDK_BDEV_IO_NUM_CHILD_IOV) :
3318 : SPDK_BDEV_IO_NUM_CHILD_IOV;
3319 :
3320 67 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE && bdev->split_on_write_unit) {
3321 5 : io_boundary = bdev->write_unit_size;
3322 67 : } else if (bdev->split_on_optimal_io_boundary) {
3323 40 : io_boundary = bdev->optimal_io_boundary;
3324 40 : } else {
3325 22 : io_boundary = UINT32_MAX;
3326 : }
3327 :
3328 67 : assert(bdev_io->internal.f.split);
3329 :
3330 67 : remaining = bdev_io->internal.split.remaining_num_blocks;
3331 67 : current_offset = bdev_io->internal.split.current_offset_blocks;
3332 67 : parent_offset = bdev_io->u.bdev.offset_blocks;
3333 67 : parent_iov_offset = (current_offset - parent_offset) * blocklen;
3334 67 : parent_iovcnt = bdev_io->u.bdev.iovcnt;
3335 :
3336 420 : for (parent_iovpos = 0; parent_iovpos < parent_iovcnt; parent_iovpos++) {
3337 420 : parent_iov = &bdev_io->u.bdev.iovs[parent_iovpos];
3338 420 : if (parent_iov_offset < parent_iov->iov_len) {
3339 67 : break;
3340 : }
3341 353 : parent_iov_offset -= parent_iov->iov_len;
3342 353 : }
3343 :
3344 67 : child_iovcnt = 0;
3345 573 : while (remaining > 0 && parent_iovpos < parent_iovcnt &&
3346 264 : child_iovcnt < SPDK_BDEV_IO_NUM_CHILD_IOV) {
3347 249 : to_next_boundary = _to_next_boundary(current_offset, io_boundary);
3348 249 : to_next_boundary = spdk_min(remaining, to_next_boundary);
3349 249 : to_next_boundary = spdk_min(max_size, to_next_boundary);
3350 249 : to_next_boundary_bytes = to_next_boundary * blocklen;
3351 :
3352 249 : iov = &bdev_io->child_iov[child_iovcnt];
3353 249 : iovcnt = 0;
3354 :
3355 249 : if (bdev_io->u.bdev.md_buf) {
3356 48 : md_buf = (char *)bdev_io->u.bdev.md_buf +
3357 24 : (current_offset - parent_offset) * spdk_bdev_get_md_size(bdev);
3358 24 : }
3359 :
3360 249 : child_iovsize = spdk_min(SPDK_BDEV_IO_NUM_CHILD_IOV - child_iovcnt, max_child_iovcnt);
3361 1810 : while (to_next_boundary_bytes > 0 && parent_iovpos < parent_iovcnt &&
3362 836 : iovcnt < child_iovsize) {
3363 725 : parent_iov = &bdev_io->u.bdev.iovs[parent_iovpos];
3364 725 : iov_len = parent_iov->iov_len - parent_iov_offset;
3365 :
3366 725 : iov_len = spdk_min(iov_len, max_segment_size);
3367 725 : iov_len = spdk_min(iov_len, to_next_boundary_bytes);
3368 725 : to_next_boundary_bytes -= iov_len;
3369 :
3370 725 : bdev_io->child_iov[child_iovcnt].iov_base = parent_iov->iov_base + parent_iov_offset;
3371 725 : bdev_io->child_iov[child_iovcnt].iov_len = iov_len;
3372 :
3373 725 : if (iov_len < parent_iov->iov_len - parent_iov_offset) {
3374 183 : parent_iov_offset += iov_len;
3375 183 : } else {
3376 542 : parent_iovpos++;
3377 542 : parent_iov_offset = 0;
3378 : }
3379 725 : child_iovcnt++;
3380 725 : iovcnt++;
3381 : }
3382 :
3383 249 : if (to_next_boundary_bytes > 0) {
3384 : /* We had to stop this child I/O early because we ran out of
3385 : * child_iov space or were limited by max_num_segments.
3386 : * Ensure the iovs to be aligned with block size and
3387 : * then adjust to_next_boundary before starting the
3388 : * child I/O.
3389 : */
3390 111 : assert(child_iovcnt == SPDK_BDEV_IO_NUM_CHILD_IOV ||
3391 : iovcnt == child_iovsize);
3392 111 : to_last_block_bytes = to_next_boundary_bytes % blocklen;
3393 111 : if (to_last_block_bytes != 0) {
3394 24 : uint32_t child_iovpos = child_iovcnt - 1;
3395 : /* don't decrease child_iovcnt when it equals to SPDK_BDEV_IO_NUM_CHILD_IOV
3396 : * so the loop will naturally end
3397 : */
3398 :
3399 24 : to_last_block_bytes = blocklen - to_last_block_bytes;
3400 24 : to_next_boundary_bytes += to_last_block_bytes;
3401 53 : while (to_last_block_bytes > 0 && iovcnt > 0) {
3402 32 : iov_len = spdk_min(to_last_block_bytes,
3403 : bdev_io->child_iov[child_iovpos].iov_len);
3404 32 : bdev_io->child_iov[child_iovpos].iov_len -= iov_len;
3405 32 : if (bdev_io->child_iov[child_iovpos].iov_len == 0) {
3406 15 : child_iovpos--;
3407 15 : if (--iovcnt == 0) {
3408 : /* If the child IO is less than a block size just return.
3409 : * If the first child IO of any split round is less than
3410 : * a block size, an error exit.
3411 : */
3412 3 : if (bdev_io->internal.split.outstanding == 0) {
3413 1 : SPDK_ERRLOG("The first child io was less than a block size\n");
3414 1 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
3415 1 : bdev_ch_remove_from_io_submitted(bdev_io);
3416 1 : spdk_trace_record(TRACE_BDEV_IO_DONE, bdev_io->internal.ch->trace_id,
3417 : 0, (uintptr_t)bdev_io, bdev_io->internal.caller_ctx,
3418 : bdev_io->internal.ch->queue_depth);
3419 1 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
3420 1 : }
3421 :
3422 3 : return;
3423 : }
3424 12 : }
3425 :
3426 29 : to_last_block_bytes -= iov_len;
3427 :
3428 29 : if (parent_iov_offset == 0) {
3429 14 : parent_iovpos--;
3430 14 : parent_iov_offset = bdev_io->u.bdev.iovs[parent_iovpos].iov_len;
3431 14 : }
3432 29 : parent_iov_offset -= iov_len;
3433 : }
3434 :
3435 21 : assert(to_last_block_bytes == 0);
3436 21 : }
3437 108 : to_next_boundary -= to_next_boundary_bytes / blocklen;
3438 108 : }
3439 :
3440 246 : rc = bdev_io_split_submit(bdev_io, iov, iovcnt, md_buf, to_next_boundary,
3441 : ¤t_offset, &remaining);
3442 246 : if (spdk_unlikely(rc)) {
3443 4 : return;
3444 : }
3445 : }
3446 67 : }
3447 :
3448 : static void
3449 3 : bdev_unmap_split(struct spdk_bdev_io *bdev_io)
3450 : {
3451 : uint64_t offset, unmap_blocks, remaining, max_unmap_blocks;
3452 3 : uint32_t num_children_reqs = 0;
3453 : int rc;
3454 :
3455 3 : assert(bdev_io->internal.f.split);
3456 :
3457 3 : offset = bdev_io->internal.split.current_offset_blocks;
3458 3 : remaining = bdev_io->internal.split.remaining_num_blocks;
3459 3 : max_unmap_blocks = bdev_io->bdev->max_unmap * bdev_io->bdev->max_unmap_segments;
3460 :
3461 20 : while (remaining && (num_children_reqs < SPDK_BDEV_MAX_CHILDREN_UNMAP_WRITE_ZEROES_REQS)) {
3462 17 : unmap_blocks = spdk_min(remaining, max_unmap_blocks);
3463 :
3464 17 : rc = bdev_io_split_submit(bdev_io, NULL, 0, NULL, unmap_blocks,
3465 : &offset, &remaining);
3466 17 : if (spdk_likely(rc == 0)) {
3467 17 : num_children_reqs++;
3468 17 : } else {
3469 0 : return;
3470 : }
3471 : }
3472 3 : }
3473 :
3474 : static void
3475 6 : bdev_write_zeroes_split(struct spdk_bdev_io *bdev_io)
3476 : {
3477 : uint64_t offset, write_zeroes_blocks, remaining;
3478 6 : uint32_t num_children_reqs = 0;
3479 : int rc;
3480 :
3481 6 : assert(bdev_io->internal.f.split);
3482 :
3483 6 : offset = bdev_io->internal.split.current_offset_blocks;
3484 6 : remaining = bdev_io->internal.split.remaining_num_blocks;
3485 :
3486 29 : while (remaining && (num_children_reqs < SPDK_BDEV_MAX_CHILDREN_UNMAP_WRITE_ZEROES_REQS)) {
3487 23 : write_zeroes_blocks = spdk_min(remaining, bdev_io->bdev->max_write_zeroes);
3488 :
3489 23 : rc = bdev_io_split_submit(bdev_io, NULL, 0, NULL, write_zeroes_blocks,
3490 : &offset, &remaining);
3491 23 : if (spdk_likely(rc == 0)) {
3492 23 : num_children_reqs++;
3493 23 : } else {
3494 0 : return;
3495 : }
3496 : }
3497 6 : }
3498 :
3499 : static void
3500 4 : bdev_copy_split(struct spdk_bdev_io *bdev_io)
3501 : {
3502 : uint64_t offset, copy_blocks, remaining;
3503 4 : uint32_t num_children_reqs = 0;
3504 : int rc;
3505 :
3506 4 : assert(bdev_io->internal.f.split);
3507 :
3508 4 : offset = bdev_io->internal.split.current_offset_blocks;
3509 4 : remaining = bdev_io->internal.split.remaining_num_blocks;
3510 :
3511 4 : assert(bdev_io->bdev->max_copy != 0);
3512 23 : while (remaining && (num_children_reqs < SPDK_BDEV_MAX_CHILDREN_COPY_REQS)) {
3513 19 : copy_blocks = spdk_min(remaining, bdev_io->bdev->max_copy);
3514 :
3515 19 : rc = bdev_io_split_submit(bdev_io, NULL, 0, NULL, copy_blocks,
3516 : &offset, &remaining);
3517 19 : if (spdk_likely(rc == 0)) {
3518 19 : num_children_reqs++;
3519 19 : } else {
3520 0 : return;
3521 : }
3522 : }
3523 4 : }
3524 :
3525 : static void
3526 58 : parent_bdev_io_complete(void *ctx, int rc)
3527 : {
3528 58 : struct spdk_bdev_io *parent_io = ctx;
3529 :
3530 58 : if (rc) {
3531 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
3532 0 : }
3533 :
3534 116 : parent_io->internal.cb(parent_io, parent_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS,
3535 58 : parent_io->internal.caller_ctx);
3536 58 : }
3537 :
3538 : static void
3539 0 : bdev_io_complete_parent_sequence_cb(void *ctx, int status)
3540 : {
3541 0 : struct spdk_bdev_io *bdev_io = ctx;
3542 :
3543 : /* u.bdev.accel_sequence should have already been cleared at this point */
3544 0 : assert(bdev_io->u.bdev.accel_sequence == NULL);
3545 0 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS);
3546 0 : bdev_io->internal.f.has_accel_sequence = false;
3547 :
3548 0 : if (spdk_unlikely(status != 0)) {
3549 0 : SPDK_ERRLOG("Failed to execute accel sequence, status=%d\n", status);
3550 0 : }
3551 :
3552 0 : parent_bdev_io_complete(bdev_io, status);
3553 0 : }
3554 :
3555 : static void
3556 301 : bdev_io_split_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
3557 : {
3558 301 : struct spdk_bdev_io *parent_io = cb_arg;
3559 :
3560 301 : spdk_bdev_free_io(bdev_io);
3561 :
3562 301 : assert(parent_io->internal.f.split);
3563 :
3564 301 : if (!success) {
3565 21 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
3566 : /* If any child I/O failed, stop further splitting process. */
3567 21 : parent_io->internal.split.current_offset_blocks += parent_io->internal.split.remaining_num_blocks;
3568 21 : parent_io->internal.split.remaining_num_blocks = 0;
3569 21 : }
3570 301 : parent_io->internal.split.outstanding--;
3571 301 : if (parent_io->internal.split.outstanding != 0) {
3572 223 : return;
3573 : }
3574 :
3575 : /*
3576 : * Parent I/O finishes when all blocks are consumed.
3577 : */
3578 78 : if (parent_io->internal.split.remaining_num_blocks == 0) {
3579 58 : assert(parent_io->internal.cb != bdev_io_split_done);
3580 58 : bdev_ch_remove_from_io_submitted(parent_io);
3581 58 : spdk_trace_record(TRACE_BDEV_IO_DONE, parent_io->internal.ch->trace_id,
3582 : 0, (uintptr_t)parent_io, bdev_io->internal.caller_ctx,
3583 : parent_io->internal.ch->queue_depth);
3584 :
3585 58 : if (spdk_likely(parent_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS)) {
3586 48 : if (bdev_io_needs_sequence_exec(parent_io->internal.desc, parent_io)) {
3587 0 : bdev_io_exec_sequence(parent_io, bdev_io_complete_parent_sequence_cb);
3588 0 : return;
3589 48 : } else if (parent_io->internal.f.has_bounce_buf &&
3590 0 : !bdev_io_use_accel_sequence(bdev_io)) {
3591 : /* bdev IO will be completed in the callback */
3592 0 : _bdev_io_push_bounce_data_buffer(parent_io, parent_bdev_io_complete);
3593 0 : return;
3594 : }
3595 48 : }
3596 :
3597 58 : parent_bdev_io_complete(parent_io, 0);
3598 58 : return;
3599 : }
3600 :
3601 : /*
3602 : * Continue with the splitting process. This function will complete the parent I/O if the
3603 : * splitting is done.
3604 : */
3605 20 : switch (parent_io->type) {
3606 : case SPDK_BDEV_IO_TYPE_READ:
3607 : case SPDK_BDEV_IO_TYPE_WRITE:
3608 17 : _bdev_rw_split(parent_io);
3609 17 : break;
3610 : case SPDK_BDEV_IO_TYPE_UNMAP:
3611 1 : bdev_unmap_split(parent_io);
3612 1 : break;
3613 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3614 1 : bdev_write_zeroes_split(parent_io);
3615 1 : break;
3616 : case SPDK_BDEV_IO_TYPE_COPY:
3617 1 : bdev_copy_split(parent_io);
3618 1 : break;
3619 : default:
3620 0 : assert(false);
3621 : break;
3622 : }
3623 301 : }
3624 :
3625 : static void bdev_rw_split_get_buf_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io,
3626 : bool success);
3627 :
3628 : static void
3629 59 : bdev_io_split(struct spdk_bdev_io *bdev_io)
3630 : {
3631 59 : assert(bdev_io_should_split(bdev_io));
3632 59 : assert(bdev_io->internal.f.split);
3633 :
3634 59 : bdev_io->internal.split.current_offset_blocks = bdev_io->u.bdev.offset_blocks;
3635 59 : bdev_io->internal.split.remaining_num_blocks = bdev_io->u.bdev.num_blocks;
3636 59 : bdev_io->internal.split.outstanding = 0;
3637 59 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
3638 :
3639 59 : switch (bdev_io->type) {
3640 : case SPDK_BDEV_IO_TYPE_READ:
3641 : case SPDK_BDEV_IO_TYPE_WRITE:
3642 49 : if (_is_buf_allocated(bdev_io->u.bdev.iovs)) {
3643 49 : _bdev_rw_split(bdev_io);
3644 49 : } else {
3645 0 : assert(bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
3646 0 : spdk_bdev_io_get_buf(bdev_io, bdev_rw_split_get_buf_cb,
3647 0 : bdev_io->u.bdev.num_blocks * bdev_io_get_block_size(bdev_io));
3648 : }
3649 49 : break;
3650 : case SPDK_BDEV_IO_TYPE_UNMAP:
3651 2 : bdev_unmap_split(bdev_io);
3652 2 : break;
3653 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3654 5 : bdev_write_zeroes_split(bdev_io);
3655 5 : break;
3656 : case SPDK_BDEV_IO_TYPE_COPY:
3657 3 : bdev_copy_split(bdev_io);
3658 3 : break;
3659 : default:
3660 0 : assert(false);
3661 : break;
3662 : }
3663 59 : }
3664 :
3665 : static void
3666 0 : bdev_rw_split_get_buf_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io, bool success)
3667 : {
3668 0 : if (!success) {
3669 0 : spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
3670 0 : return;
3671 : }
3672 :
3673 0 : _bdev_rw_split(bdev_io);
3674 0 : }
3675 :
3676 : static inline void
3677 579 : _bdev_io_submit(struct spdk_bdev_io *bdev_io)
3678 : {
3679 579 : struct spdk_bdev *bdev = bdev_io->bdev;
3680 579 : struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
3681 :
3682 579 : if (spdk_likely(bdev_ch->flags == 0)) {
3683 554 : bdev_io_do_submit(bdev_ch, bdev_io);
3684 554 : return;
3685 : }
3686 :
3687 25 : if (bdev_ch->flags & BDEV_CH_RESET_IN_PROGRESS) {
3688 2 : _bdev_io_complete_in_submit(bdev_ch, bdev_io, SPDK_BDEV_IO_STATUS_ABORTED);
3689 25 : } else if (bdev_ch->flags & BDEV_CH_QOS_ENABLED) {
3690 23 : if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_ABORT) &&
3691 2 : bdev_abort_queued_io(&bdev_ch->qos_queued_io, bdev_io->u.abort.bio_to_abort)) {
3692 0 : _bdev_io_complete_in_submit(bdev_ch, bdev_io, SPDK_BDEV_IO_STATUS_SUCCESS);
3693 0 : } else {
3694 23 : TAILQ_INSERT_TAIL(&bdev_ch->qos_queued_io, bdev_io, internal.link);
3695 23 : bdev_qos_io_submit(bdev_ch, bdev->internal.qos);
3696 : }
3697 23 : } else {
3698 0 : SPDK_ERRLOG("unknown bdev_ch flag %x found\n", bdev_ch->flags);
3699 0 : _bdev_io_complete_in_submit(bdev_ch, bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
3700 : }
3701 579 : }
3702 :
3703 : bool bdev_lba_range_overlapped(struct lba_range *range1, struct lba_range *range2);
3704 :
3705 : bool
3706 23 : bdev_lba_range_overlapped(struct lba_range *range1, struct lba_range *range2)
3707 : {
3708 23 : if (range1->length == 0 || range2->length == 0) {
3709 1 : return false;
3710 : }
3711 :
3712 22 : if (range1->offset + range1->length <= range2->offset) {
3713 1 : return false;
3714 : }
3715 :
3716 21 : if (range2->offset + range2->length <= range1->offset) {
3717 3 : return false;
3718 : }
3719 :
3720 18 : return true;
3721 23 : }
3722 :
3723 : static bool
3724 11 : bdev_io_range_is_locked(struct spdk_bdev_io *bdev_io, struct lba_range *range)
3725 : {
3726 11 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
3727 : struct lba_range r;
3728 :
3729 11 : switch (bdev_io->type) {
3730 : case SPDK_BDEV_IO_TYPE_NVME_IO:
3731 : case SPDK_BDEV_IO_TYPE_NVME_IO_MD:
3732 : /* Don't try to decode the NVMe command - just assume worst-case and that
3733 : * it overlaps a locked range.
3734 : */
3735 0 : return true;
3736 : case SPDK_BDEV_IO_TYPE_READ:
3737 6 : if (!range->quiesce) {
3738 4 : return false;
3739 : }
3740 : /* fallthrough */
3741 : case SPDK_BDEV_IO_TYPE_WRITE:
3742 : case SPDK_BDEV_IO_TYPE_UNMAP:
3743 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
3744 : case SPDK_BDEV_IO_TYPE_ZCOPY:
3745 : case SPDK_BDEV_IO_TYPE_COPY:
3746 7 : r.offset = bdev_io->u.bdev.offset_blocks;
3747 7 : r.length = bdev_io->u.bdev.num_blocks;
3748 7 : if (!bdev_lba_range_overlapped(range, &r)) {
3749 : /* This I/O doesn't overlap the specified LBA range. */
3750 0 : return false;
3751 7 : } else if (range->owner_ch == ch && range->locked_ctx == bdev_io->internal.caller_ctx) {
3752 : /* This I/O overlaps, but the I/O is on the same channel that locked this
3753 : * range, and the caller_ctx is the same as the locked_ctx. This means
3754 : * that this I/O is associated with the lock, and is allowed to execute.
3755 : */
3756 2 : return false;
3757 : } else {
3758 5 : return true;
3759 : }
3760 : default:
3761 0 : return false;
3762 : }
3763 11 : }
3764 :
3765 : void
3766 639 : bdev_io_submit(struct spdk_bdev_io *bdev_io)
3767 : {
3768 639 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
3769 :
3770 639 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_PENDING);
3771 :
3772 639 : if (!TAILQ_EMPTY(&ch->locked_ranges)) {
3773 : struct lba_range *range;
3774 :
3775 13 : TAILQ_FOREACH(range, &ch->locked_ranges, tailq) {
3776 8 : if (bdev_io_range_is_locked(bdev_io, range)) {
3777 3 : TAILQ_INSERT_TAIL(&ch->io_locked, bdev_io, internal.ch_link);
3778 3 : return;
3779 : }
3780 5 : }
3781 5 : }
3782 :
3783 636 : bdev_ch_add_to_io_submitted(bdev_io);
3784 :
3785 636 : bdev_io->internal.submit_tsc = spdk_get_ticks();
3786 636 : spdk_trace_record_tsc(bdev_io->internal.submit_tsc, TRACE_BDEV_IO_START,
3787 : ch->trace_id, bdev_io->u.bdev.num_blocks,
3788 : (uintptr_t)bdev_io, (uint64_t)bdev_io->type, bdev_io->internal.caller_ctx,
3789 : bdev_io->u.bdev.offset_blocks, ch->queue_depth);
3790 :
3791 636 : if (bdev_io->internal.f.split) {
3792 59 : bdev_io_split(bdev_io);
3793 59 : return;
3794 : }
3795 :
3796 577 : _bdev_io_submit(bdev_io);
3797 639 : }
3798 :
3799 : static inline int
3800 2 : bdev_io_init_dif_ctx(struct spdk_bdev_io *bdev_io)
3801 : {
3802 2 : struct spdk_bdev *bdev = bdev_io->bdev;
3803 : struct spdk_dif_ctx_init_ext_opts dif_opts;
3804 :
3805 2 : memset(&bdev_io->u.bdev.dif_err, 0, sizeof(struct spdk_dif_error));
3806 :
3807 2 : dif_opts.size = SPDK_SIZEOF(&dif_opts, dif_pi_format);
3808 2 : dif_opts.dif_pi_format = bdev->dif_pi_format;
3809 :
3810 4 : return spdk_dif_ctx_init(&bdev_io->u.bdev.dif_ctx,
3811 2 : bdev->blocklen,
3812 2 : bdev->md_len,
3813 2 : bdev->md_interleave,
3814 2 : bdev->dif_is_head_of_md,
3815 2 : bdev->dif_type,
3816 2 : bdev_io->u.bdev.dif_check_flags,
3817 2 : bdev_io->u.bdev.offset_blocks & 0xFFFFFFFF,
3818 : 0xFFFF, 0, 0, 0, &dif_opts);
3819 : }
3820 :
3821 : static void
3822 4 : _bdev_memory_domain_get_io_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io,
3823 : bool success)
3824 : {
3825 4 : if (!success) {
3826 0 : SPDK_ERRLOG("Failed to get data buffer, completing IO\n");
3827 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
3828 0 : bdev_io_complete_unsubmitted(bdev_io);
3829 0 : return;
3830 : }
3831 :
3832 4 : if (bdev_io_needs_sequence_exec(bdev_io->internal.desc, bdev_io)) {
3833 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
3834 0 : bdev_io_exec_sequence(bdev_io, bdev_io_submit_sequence_cb);
3835 0 : return;
3836 : }
3837 : /* For reads we'll execute the sequence after the data is read, so, for now, only
3838 : * clear out accel_sequence pointer and submit the IO */
3839 0 : assert(bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
3840 0 : bdev_io->u.bdev.accel_sequence = NULL;
3841 0 : }
3842 :
3843 4 : bdev_io_submit(bdev_io);
3844 4 : }
3845 :
3846 : static inline void
3847 4 : _bdev_io_ext_use_bounce_buffer(struct spdk_bdev_io *bdev_io)
3848 : {
3849 : /* bdev doesn't support memory domains, thereby buffers in this IO request can't
3850 : * be accessed directly. It is needed to allocate buffers before issuing IO operation.
3851 : * For write operation we need to pull buffers from memory domain before submitting IO.
3852 : * Once read operation completes, we need to use memory_domain push functionality to
3853 : * update data in original memory domain IO buffer.
3854 : *
3855 : * If this I/O request is not aware of metadata, buffers in thsi IO request can't be
3856 : * accessed directly too. It is needed to allocate buffers before issuing IO operation.
3857 : * For write operation we need to insert metadata before submitting IO. Once read
3858 : * operation completes, we need to strip metadata in original IO buffer.
3859 : *
3860 : * This IO request will go through a regular IO flow, so clear memory domains pointers */
3861 4 : assert(bdev_io_use_memory_domain(bdev_io) ||
3862 : bdev_io_needs_metadata(bdev_io->internal.desc, bdev_io));
3863 :
3864 4 : bdev_io->u.bdev.memory_domain = NULL;
3865 4 : bdev_io->u.bdev.memory_domain_ctx = NULL;
3866 8 : _bdev_io_get_bounce_buf(bdev_io, _bdev_memory_domain_get_io_cb,
3867 4 : bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen);
3868 4 : }
3869 :
3870 : static inline void
3871 0 : _bdev_io_ext_use_accel_buffer(struct spdk_bdev_io *bdev_io)
3872 : {
3873 0 : assert(bdev_io_use_memory_domain(bdev_io));
3874 0 : assert(bdev_io_needs_metadata(bdev_io->internal.desc, bdev_io));
3875 :
3876 0 : bdev_io->u.bdev.memory_domain = NULL;
3877 0 : bdev_io->u.bdev.memory_domain_ctx = NULL;
3878 0 : bdev_io_get_accel_buf(bdev_io, _bdev_memory_domain_get_io_cb,
3879 0 : bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen);
3880 0 : }
3881 :
3882 : /* We need to allocate bounce buffer
3883 : * - if bdev doesn't support memory domains,
3884 : * - if it does support them, but we need to execute an accel sequence and the data buffer is
3885 : * from accel memory domain (to avoid doing a push/pull from that domain), or
3886 : * - if IO is not aware of metadata.
3887 : */
3888 : static inline bool
3889 292 : bdev_io_needs_bounce_buffer(struct spdk_bdev_desc *desc, struct spdk_bdev_io *bdev_io)
3890 : {
3891 292 : if (bdev_io_use_memory_domain(bdev_io)) {
3892 4 : if (!desc->memory_domains_supported ||
3893 0 : (bdev_io_needs_sequence_exec(desc, bdev_io) &&
3894 0 : (bdev_io->internal.memory_domain == spdk_accel_get_memory_domain() ||
3895 0 : bdev_io_needs_metadata(desc, bdev_io)))) {
3896 4 : return true;
3897 : }
3898 :
3899 0 : return false;
3900 : }
3901 :
3902 288 : if (bdev_io_needs_metadata(desc, bdev_io)) {
3903 0 : return true;
3904 : }
3905 :
3906 288 : return false;
3907 292 : }
3908 :
3909 : /* We need to allocate fake accel buffer if bdev supports memory domains but IO is not
3910 : * aware of metadata.
3911 : */
3912 : static inline bool
3913 288 : bdev_io_needs_accel_buffer(struct spdk_bdev_desc *desc, struct spdk_bdev_io *bdev_io)
3914 : {
3915 288 : if (bdev_io_needs_metadata(desc, bdev_io)) {
3916 0 : assert(bdev_io_use_memory_domain(bdev_io));
3917 0 : return true;
3918 : }
3919 :
3920 288 : return false;
3921 288 : }
3922 :
3923 : static inline void
3924 292 : _bdev_io_submit_ext(struct spdk_bdev_desc *desc, struct spdk_bdev_io *bdev_io)
3925 : {
3926 292 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
3927 : int rc;
3928 :
3929 292 : if (spdk_unlikely(ch->flags & BDEV_CH_RESET_IN_PROGRESS)) {
3930 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_ABORTED;
3931 0 : bdev_io_complete_unsubmitted(bdev_io);
3932 0 : return;
3933 : }
3934 :
3935 292 : if (bdev_io_needs_metadata(desc, bdev_io)) {
3936 0 : rc = bdev_io_init_dif_ctx(bdev_io);
3937 0 : if (spdk_unlikely(rc != 0)) {
3938 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
3939 0 : bdev_io_complete_unsubmitted(bdev_io);
3940 0 : return;
3941 : }
3942 0 : }
3943 :
3944 292 : if (bdev_io_needs_bounce_buffer(desc, bdev_io)) {
3945 4 : _bdev_io_ext_use_bounce_buffer(bdev_io);
3946 4 : return;
3947 : }
3948 :
3949 288 : if (bdev_io_needs_accel_buffer(desc, bdev_io)) {
3950 0 : _bdev_io_ext_use_accel_buffer(bdev_io);
3951 0 : return;
3952 : }
3953 :
3954 288 : if (bdev_io_needs_sequence_exec(desc, bdev_io)) {
3955 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
3956 0 : bdev_io_exec_sequence(bdev_io, bdev_io_submit_sequence_cb);
3957 0 : return;
3958 : }
3959 : /* For reads we'll execute the sequence after the data is read, so, for now, only
3960 : * clear out accel_sequence pointer and submit the IO */
3961 0 : assert(bdev_io->type == SPDK_BDEV_IO_TYPE_READ);
3962 0 : bdev_io->u.bdev.accel_sequence = NULL;
3963 0 : }
3964 :
3965 288 : bdev_io_submit(bdev_io);
3966 292 : }
3967 :
3968 : static void
3969 12 : bdev_io_submit_reset(struct spdk_bdev_io *bdev_io)
3970 : {
3971 12 : struct spdk_bdev *bdev = bdev_io->bdev;
3972 12 : struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
3973 12 : struct spdk_io_channel *ch = bdev_ch->channel;
3974 :
3975 12 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_PENDING);
3976 :
3977 12 : bdev_io->internal.f.in_submit_request = true;
3978 12 : bdev_submit_request(bdev, ch, bdev_io);
3979 12 : bdev_io->internal.f.in_submit_request = false;
3980 12 : }
3981 :
3982 : void
3983 693 : bdev_io_init(struct spdk_bdev_io *bdev_io,
3984 : struct spdk_bdev *bdev, void *cb_arg,
3985 : spdk_bdev_io_completion_cb cb)
3986 : {
3987 693 : bdev_io->bdev = bdev;
3988 693 : bdev_io->internal.f.raw = 0;
3989 693 : bdev_io->internal.caller_ctx = cb_arg;
3990 693 : bdev_io->internal.cb = cb;
3991 693 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_PENDING;
3992 693 : bdev_io->internal.f.in_submit_request = false;
3993 693 : bdev_io->internal.error.nvme.cdw0 = 0;
3994 693 : bdev_io->num_retries = 0;
3995 693 : bdev_io->internal.get_buf_cb = NULL;
3996 693 : bdev_io->internal.get_aux_buf_cb = NULL;
3997 693 : bdev_io->internal.data_transfer_cpl = NULL;
3998 693 : bdev_io->internal.f.split = bdev_io_should_split(bdev_io);
3999 693 : }
4000 :
4001 : static bool
4002 540 : bdev_io_type_supported(struct spdk_bdev *bdev, enum spdk_bdev_io_type io_type)
4003 : {
4004 540 : return bdev->fn_table->io_type_supported(bdev->ctxt, io_type);
4005 : }
4006 :
4007 : bool
4008 178 : spdk_bdev_io_type_supported(struct spdk_bdev *bdev, enum spdk_bdev_io_type io_type)
4009 : {
4010 : bool supported;
4011 :
4012 178 : supported = bdev_io_type_supported(bdev, io_type);
4013 :
4014 178 : if (!supported) {
4015 7 : switch (io_type) {
4016 : case SPDK_BDEV_IO_TYPE_WRITE_ZEROES:
4017 : /* The bdev layer will emulate write zeroes as long as write is supported. */
4018 0 : supported = bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE);
4019 0 : break;
4020 : default:
4021 7 : break;
4022 : }
4023 7 : }
4024 :
4025 178 : return supported;
4026 : }
4027 :
4028 : static const char *g_io_type_strings[] = {
4029 : [SPDK_BDEV_IO_TYPE_READ] = "read",
4030 : [SPDK_BDEV_IO_TYPE_WRITE] = "write",
4031 : [SPDK_BDEV_IO_TYPE_UNMAP] = "unmap",
4032 : [SPDK_BDEV_IO_TYPE_FLUSH] = "flush",
4033 : [SPDK_BDEV_IO_TYPE_RESET] = "reset",
4034 : [SPDK_BDEV_IO_TYPE_NVME_ADMIN] = "nvme_admin",
4035 : [SPDK_BDEV_IO_TYPE_NVME_IO] = "nvme_io",
4036 : [SPDK_BDEV_IO_TYPE_NVME_IO_MD] = "nvme_io_md",
4037 : [SPDK_BDEV_IO_TYPE_WRITE_ZEROES] = "write_zeroes",
4038 : [SPDK_BDEV_IO_TYPE_ZCOPY] = "zcopy",
4039 : [SPDK_BDEV_IO_TYPE_GET_ZONE_INFO] = "get_zone_info",
4040 : [SPDK_BDEV_IO_TYPE_ZONE_MANAGEMENT] = "zone_management",
4041 : [SPDK_BDEV_IO_TYPE_ZONE_APPEND] = "zone_append",
4042 : [SPDK_BDEV_IO_TYPE_COMPARE] = "compare",
4043 : [SPDK_BDEV_IO_TYPE_COMPARE_AND_WRITE] = "compare_and_write",
4044 : [SPDK_BDEV_IO_TYPE_ABORT] = "abort",
4045 : [SPDK_BDEV_IO_TYPE_SEEK_HOLE] = "seek_hole",
4046 : [SPDK_BDEV_IO_TYPE_SEEK_DATA] = "seek_data",
4047 : [SPDK_BDEV_IO_TYPE_COPY] = "copy",
4048 : [SPDK_BDEV_IO_TYPE_NVME_IOV_MD] = "nvme_iov_md",
4049 : };
4050 :
4051 : const char *
4052 0 : spdk_bdev_get_io_type_name(enum spdk_bdev_io_type io_type)
4053 : {
4054 0 : if (io_type <= SPDK_BDEV_IO_TYPE_INVALID || io_type >= SPDK_BDEV_NUM_IO_TYPES) {
4055 0 : return NULL;
4056 : }
4057 :
4058 0 : return g_io_type_strings[io_type];
4059 0 : }
4060 :
4061 : int
4062 0 : spdk_bdev_get_io_type(const char *io_type_string)
4063 : {
4064 : int i;
4065 :
4066 0 : for (i = SPDK_BDEV_IO_TYPE_READ; i < SPDK_BDEV_NUM_IO_TYPES; ++i) {
4067 0 : if (!strcmp(io_type_string, g_io_type_strings[i])) {
4068 0 : return i;
4069 : }
4070 0 : }
4071 :
4072 0 : return -1;
4073 0 : }
4074 :
4075 : uint64_t
4076 0 : spdk_bdev_io_get_submit_tsc(struct spdk_bdev_io *bdev_io)
4077 : {
4078 0 : return bdev_io->internal.submit_tsc;
4079 : }
4080 :
4081 : bool
4082 0 : spdk_bdev_io_hide_metadata(struct spdk_bdev_io *bdev_io)
4083 : {
4084 0 : return bdev_io->internal.desc->opts.hide_metadata;
4085 : }
4086 :
4087 : int
4088 0 : spdk_bdev_dump_info_json(struct spdk_bdev *bdev, struct spdk_json_write_ctx *w)
4089 : {
4090 0 : if (bdev->fn_table->dump_info_json) {
4091 0 : return bdev->fn_table->dump_info_json(bdev->ctxt, w);
4092 : }
4093 :
4094 0 : return 0;
4095 0 : }
4096 :
4097 : static void
4098 10 : bdev_qos_update_max_quota_per_timeslice(struct spdk_bdev_qos *qos)
4099 : {
4100 10 : uint32_t max_per_timeslice = 0;
4101 : int i;
4102 :
4103 50 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
4104 40 : if (qos->rate_limits[i].limit == SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
4105 15 : qos->rate_limits[i].max_per_timeslice = 0;
4106 15 : continue;
4107 : }
4108 :
4109 50 : max_per_timeslice = qos->rate_limits[i].limit *
4110 25 : SPDK_BDEV_QOS_TIMESLICE_IN_USEC / SPDK_SEC_TO_USEC;
4111 :
4112 25 : qos->rate_limits[i].max_per_timeslice = spdk_max(max_per_timeslice,
4113 : qos->rate_limits[i].min_per_timeslice);
4114 :
4115 50 : __atomic_store_n(&qos->rate_limits[i].remaining_this_timeslice,
4116 25 : qos->rate_limits[i].max_per_timeslice, __ATOMIC_RELEASE);
4117 25 : }
4118 :
4119 10 : bdev_qos_set_ops(qos);
4120 10 : }
4121 :
4122 : static void
4123 4 : bdev_channel_submit_qos_io(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
4124 : struct spdk_io_channel *io_ch, void *ctx)
4125 : {
4126 4 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
4127 : int status;
4128 :
4129 4 : bdev_qos_io_submit(bdev_ch, bdev->internal.qos);
4130 :
4131 : /* if all IOs were sent then continue the iteration, otherwise - stop it */
4132 : /* TODO: channels round robing */
4133 4 : status = TAILQ_EMPTY(&bdev_ch->qos_queued_io) ? 0 : 1;
4134 :
4135 4 : spdk_bdev_for_each_channel_continue(i, status);
4136 4 : }
4137 :
4138 :
4139 : static void
4140 2 : bdev_channel_submit_qos_io_done(struct spdk_bdev *bdev, void *ctx, int status)
4141 : {
4142 :
4143 2 : }
4144 :
4145 : static int
4146 3 : bdev_channel_poll_qos(void *arg)
4147 : {
4148 3 : struct spdk_bdev *bdev = arg;
4149 3 : struct spdk_bdev_qos *qos = bdev->internal.qos;
4150 3 : uint64_t now = spdk_get_ticks();
4151 : int i;
4152 : int64_t remaining_last_timeslice;
4153 :
4154 3 : if (spdk_unlikely(qos->thread == NULL)) {
4155 : /* Old QoS was unbound to remove and new QoS is not enabled yet. */
4156 1 : return SPDK_POLLER_IDLE;
4157 : }
4158 :
4159 2 : if (now < (qos->last_timeslice + qos->timeslice_size)) {
4160 : /* We received our callback earlier than expected - return
4161 : * immediately and wait to do accounting until at least one
4162 : * timeslice has actually expired. This should never happen
4163 : * with a well-behaved timer implementation.
4164 : */
4165 0 : return SPDK_POLLER_IDLE;
4166 : }
4167 :
4168 : /* Reset for next round of rate limiting */
4169 10 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
4170 : /* We may have allowed the IOs or bytes to slightly overrun in the last
4171 : * timeslice. remaining_this_timeslice is signed, so if it's negative
4172 : * here, we'll account for the overrun so that the next timeslice will
4173 : * be appropriately reduced.
4174 : */
4175 8 : remaining_last_timeslice = __atomic_exchange_n(&qos->rate_limits[i].remaining_this_timeslice,
4176 : 0, __ATOMIC_RELAXED);
4177 8 : if (remaining_last_timeslice < 0) {
4178 : /* There could be a race condition here as both bdev_qos_rw_queue_io() and bdev_channel_poll_qos()
4179 : * potentially use 2 atomic ops each, so they can intertwine.
4180 : * This race can potentially cause the limits to be a little fuzzy but won't cause any real damage.
4181 : */
4182 0 : __atomic_store_n(&qos->rate_limits[i].remaining_this_timeslice,
4183 0 : remaining_last_timeslice, __ATOMIC_RELAXED);
4184 0 : }
4185 8 : }
4186 :
4187 4 : while (now >= (qos->last_timeslice + qos->timeslice_size)) {
4188 2 : qos->last_timeslice += qos->timeslice_size;
4189 10 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
4190 16 : __atomic_add_fetch(&qos->rate_limits[i].remaining_this_timeslice,
4191 8 : qos->rate_limits[i].max_per_timeslice, __ATOMIC_RELAXED);
4192 8 : }
4193 : }
4194 :
4195 2 : spdk_bdev_for_each_channel(bdev, bdev_channel_submit_qos_io, qos,
4196 : bdev_channel_submit_qos_io_done);
4197 :
4198 2 : return SPDK_POLLER_BUSY;
4199 3 : }
4200 :
4201 : static void
4202 75 : bdev_channel_destroy_resource(struct spdk_bdev_channel *ch)
4203 : {
4204 : struct spdk_bdev_shared_resource *shared_resource;
4205 : struct lba_range *range;
4206 :
4207 75 : bdev_free_io_stat(ch->stat);
4208 : #ifdef SPDK_CONFIG_VTUNE
4209 : bdev_free_io_stat(ch->prev_stat);
4210 : #endif
4211 :
4212 75 : while (!TAILQ_EMPTY(&ch->locked_ranges)) {
4213 0 : range = TAILQ_FIRST(&ch->locked_ranges);
4214 0 : TAILQ_REMOVE(&ch->locked_ranges, range, tailq);
4215 0 : free(range);
4216 : }
4217 :
4218 75 : spdk_put_io_channel(ch->channel);
4219 75 : spdk_put_io_channel(ch->accel_channel);
4220 :
4221 75 : shared_resource = ch->shared_resource;
4222 :
4223 75 : assert(TAILQ_EMPTY(&ch->io_locked));
4224 75 : assert(TAILQ_EMPTY(&ch->io_submitted));
4225 75 : assert(TAILQ_EMPTY(&ch->io_accel_exec));
4226 75 : assert(TAILQ_EMPTY(&ch->io_memory_domain));
4227 75 : assert(ch->io_outstanding == 0);
4228 75 : assert(shared_resource->ref > 0);
4229 75 : shared_resource->ref--;
4230 75 : if (shared_resource->ref == 0) {
4231 74 : assert(shared_resource->io_outstanding == 0);
4232 74 : TAILQ_REMOVE(&shared_resource->mgmt_ch->shared_resources, shared_resource, link);
4233 74 : spdk_put_io_channel(spdk_io_channel_from_ctx(shared_resource->mgmt_ch));
4234 74 : spdk_poller_unregister(&shared_resource->nomem_poller);
4235 74 : free(shared_resource);
4236 74 : }
4237 75 : }
4238 :
4239 : static void
4240 84 : bdev_enable_qos(struct spdk_bdev *bdev, struct spdk_bdev_channel *ch)
4241 : {
4242 84 : struct spdk_bdev_qos *qos = bdev->internal.qos;
4243 : int i;
4244 :
4245 84 : assert(spdk_spin_held(&bdev->internal.spinlock));
4246 :
4247 : /* Rate limiting on this bdev enabled */
4248 84 : if (qos) {
4249 17 : if (qos->ch == NULL) {
4250 : struct spdk_io_channel *io_ch;
4251 :
4252 9 : SPDK_DEBUGLOG(bdev, "Selecting channel %p as QoS channel for bdev %s on thread %p\n", ch,
4253 : bdev->name, spdk_get_thread());
4254 :
4255 : /* No qos channel has been selected, so set one up */
4256 :
4257 : /* Take another reference to ch */
4258 9 : io_ch = spdk_get_io_channel(__bdev_to_io_dev(bdev));
4259 9 : assert(io_ch != NULL);
4260 9 : qos->ch = ch;
4261 :
4262 9 : qos->thread = spdk_io_channel_get_thread(io_ch);
4263 :
4264 45 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
4265 36 : if (bdev_qos_is_iops_rate_limit(i) == true) {
4266 9 : qos->rate_limits[i].min_per_timeslice =
4267 : SPDK_BDEV_QOS_MIN_IO_PER_TIMESLICE;
4268 9 : } else {
4269 27 : qos->rate_limits[i].min_per_timeslice =
4270 : SPDK_BDEV_QOS_MIN_BYTE_PER_TIMESLICE;
4271 : }
4272 :
4273 36 : if (qos->rate_limits[i].limit == 0) {
4274 2 : qos->rate_limits[i].limit = SPDK_BDEV_QOS_LIMIT_NOT_DEFINED;
4275 2 : }
4276 36 : }
4277 9 : bdev_qos_update_max_quota_per_timeslice(qos);
4278 9 : qos->timeslice_size =
4279 9 : SPDK_BDEV_QOS_TIMESLICE_IN_USEC * spdk_get_ticks_hz() / SPDK_SEC_TO_USEC;
4280 9 : qos->last_timeslice = spdk_get_ticks();
4281 9 : qos->poller = SPDK_POLLER_REGISTER(bdev_channel_poll_qos,
4282 : bdev,
4283 : SPDK_BDEV_QOS_TIMESLICE_IN_USEC);
4284 9 : }
4285 :
4286 17 : ch->flags |= BDEV_CH_QOS_ENABLED;
4287 17 : }
4288 84 : }
4289 :
4290 : struct poll_timeout_ctx {
4291 : struct spdk_bdev_desc *desc;
4292 : uint64_t timeout_in_sec;
4293 : spdk_bdev_io_timeout_cb cb_fn;
4294 : void *cb_arg;
4295 : };
4296 :
4297 : static void
4298 278 : bdev_desc_free(struct spdk_bdev_desc *desc)
4299 : {
4300 278 : spdk_spin_destroy(&desc->spinlock);
4301 278 : free(desc->media_events_buffer);
4302 278 : free(desc);
4303 278 : }
4304 :
4305 : static void
4306 8 : bdev_channel_poll_timeout_io_done(struct spdk_bdev *bdev, void *_ctx, int status)
4307 : {
4308 8 : struct poll_timeout_ctx *ctx = _ctx;
4309 8 : struct spdk_bdev_desc *desc = ctx->desc;
4310 :
4311 8 : free(ctx);
4312 :
4313 8 : spdk_spin_lock(&desc->spinlock);
4314 8 : desc->refs--;
4315 8 : if (desc->closed == true && desc->refs == 0) {
4316 1 : spdk_spin_unlock(&desc->spinlock);
4317 1 : bdev_desc_free(desc);
4318 1 : return;
4319 : }
4320 7 : spdk_spin_unlock(&desc->spinlock);
4321 8 : }
4322 :
4323 : static void
4324 13 : bdev_channel_poll_timeout_io(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
4325 : struct spdk_io_channel *io_ch, void *_ctx)
4326 : {
4327 13 : struct poll_timeout_ctx *ctx = _ctx;
4328 13 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
4329 13 : struct spdk_bdev_desc *desc = ctx->desc;
4330 : struct spdk_bdev_io *bdev_io;
4331 : uint64_t now;
4332 :
4333 13 : spdk_spin_lock(&desc->spinlock);
4334 13 : if (desc->closed == true) {
4335 1 : spdk_spin_unlock(&desc->spinlock);
4336 1 : spdk_bdev_for_each_channel_continue(i, -1);
4337 1 : return;
4338 : }
4339 12 : spdk_spin_unlock(&desc->spinlock);
4340 :
4341 12 : now = spdk_get_ticks();
4342 22 : TAILQ_FOREACH(bdev_io, &bdev_ch->io_submitted, internal.ch_link) {
4343 : /* Exclude any I/O that are generated via splitting. */
4344 15 : if (bdev_io->internal.cb == bdev_io_split_done) {
4345 3 : continue;
4346 : }
4347 :
4348 : /* Once we find an I/O that has not timed out, we can immediately
4349 : * exit the loop.
4350 : */
4351 24 : if (now < (bdev_io->internal.submit_tsc +
4352 12 : ctx->timeout_in_sec * spdk_get_ticks_hz())) {
4353 5 : goto end;
4354 : }
4355 :
4356 7 : if (bdev_io->internal.desc == desc) {
4357 7 : ctx->cb_fn(ctx->cb_arg, bdev_io);
4358 7 : }
4359 14 : }
4360 :
4361 : end:
4362 12 : spdk_bdev_for_each_channel_continue(i, 0);
4363 13 : }
4364 :
4365 : static int
4366 8 : bdev_poll_timeout_io(void *arg)
4367 : {
4368 8 : struct spdk_bdev_desc *desc = arg;
4369 8 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
4370 : struct poll_timeout_ctx *ctx;
4371 :
4372 8 : ctx = calloc(1, sizeof(struct poll_timeout_ctx));
4373 8 : if (!ctx) {
4374 0 : SPDK_ERRLOG("failed to allocate memory\n");
4375 0 : return SPDK_POLLER_BUSY;
4376 : }
4377 8 : ctx->desc = desc;
4378 8 : ctx->cb_arg = desc->cb_arg;
4379 8 : ctx->cb_fn = desc->cb_fn;
4380 8 : ctx->timeout_in_sec = desc->timeout_in_sec;
4381 :
4382 : /* Take a ref on the descriptor in case it gets closed while we are checking
4383 : * all of the channels.
4384 : */
4385 8 : spdk_spin_lock(&desc->spinlock);
4386 8 : desc->refs++;
4387 8 : spdk_spin_unlock(&desc->spinlock);
4388 :
4389 8 : spdk_bdev_for_each_channel(bdev, bdev_channel_poll_timeout_io, ctx,
4390 : bdev_channel_poll_timeout_io_done);
4391 :
4392 8 : return SPDK_POLLER_BUSY;
4393 8 : }
4394 :
4395 : int
4396 5 : spdk_bdev_set_timeout(struct spdk_bdev_desc *desc, uint64_t timeout_in_sec,
4397 : spdk_bdev_io_timeout_cb cb_fn, void *cb_arg)
4398 : {
4399 5 : assert(desc->thread == spdk_get_thread());
4400 :
4401 5 : spdk_poller_unregister(&desc->io_timeout_poller);
4402 :
4403 5 : if (timeout_in_sec) {
4404 4 : assert(cb_fn != NULL);
4405 4 : desc->io_timeout_poller = SPDK_POLLER_REGISTER(bdev_poll_timeout_io,
4406 : desc,
4407 : SPDK_BDEV_IO_POLL_INTERVAL_IN_MSEC * SPDK_SEC_TO_USEC /
4408 : 1000);
4409 4 : if (desc->io_timeout_poller == NULL) {
4410 0 : SPDK_ERRLOG("can not register the desc timeout IO poller\n");
4411 0 : return -1;
4412 : }
4413 4 : }
4414 :
4415 5 : desc->cb_fn = cb_fn;
4416 5 : desc->cb_arg = cb_arg;
4417 5 : desc->timeout_in_sec = timeout_in_sec;
4418 :
4419 5 : return 0;
4420 5 : }
4421 :
4422 : static int
4423 77 : bdev_channel_create(void *io_device, void *ctx_buf)
4424 : {
4425 77 : struct spdk_bdev *bdev = __bdev_from_io_dev(io_device);
4426 77 : struct spdk_bdev_channel *ch = ctx_buf;
4427 : struct spdk_io_channel *mgmt_io_ch;
4428 : struct spdk_bdev_mgmt_channel *mgmt_ch;
4429 : struct spdk_bdev_shared_resource *shared_resource;
4430 : struct lba_range *range;
4431 :
4432 77 : ch->bdev = bdev;
4433 77 : ch->channel = bdev->fn_table->get_io_channel(bdev->ctxt);
4434 77 : if (!ch->channel) {
4435 2 : return -1;
4436 : }
4437 :
4438 75 : ch->accel_channel = spdk_accel_get_io_channel();
4439 75 : if (!ch->accel_channel) {
4440 0 : spdk_put_io_channel(ch->channel);
4441 0 : return -1;
4442 : }
4443 :
4444 75 : spdk_trace_record(TRACE_BDEV_IOCH_CREATE, bdev->internal.trace_id, 0, 0,
4445 : spdk_thread_get_id(spdk_io_channel_get_thread(ch->channel)));
4446 :
4447 75 : assert(ch->histogram == NULL);
4448 75 : if (bdev->internal.histogram_enabled) {
4449 0 : ch->histogram = spdk_histogram_data_alloc();
4450 0 : if (ch->histogram == NULL) {
4451 0 : SPDK_ERRLOG("Could not allocate histogram\n");
4452 0 : }
4453 0 : }
4454 :
4455 75 : mgmt_io_ch = spdk_get_io_channel(&g_bdev_mgr);
4456 75 : if (!mgmt_io_ch) {
4457 0 : spdk_put_io_channel(ch->channel);
4458 0 : spdk_put_io_channel(ch->accel_channel);
4459 0 : return -1;
4460 : }
4461 :
4462 75 : mgmt_ch = __io_ch_to_bdev_mgmt_ch(mgmt_io_ch);
4463 77 : TAILQ_FOREACH(shared_resource, &mgmt_ch->shared_resources, link) {
4464 3 : if (shared_resource->shared_ch == ch->channel) {
4465 1 : spdk_put_io_channel(mgmt_io_ch);
4466 1 : shared_resource->ref++;
4467 1 : break;
4468 : }
4469 2 : }
4470 :
4471 75 : if (shared_resource == NULL) {
4472 74 : shared_resource = calloc(1, sizeof(*shared_resource));
4473 74 : if (shared_resource == NULL) {
4474 0 : spdk_put_io_channel(ch->channel);
4475 0 : spdk_put_io_channel(ch->accel_channel);
4476 0 : spdk_put_io_channel(mgmt_io_ch);
4477 0 : return -1;
4478 : }
4479 :
4480 74 : shared_resource->mgmt_ch = mgmt_ch;
4481 74 : shared_resource->io_outstanding = 0;
4482 74 : TAILQ_INIT(&shared_resource->nomem_io);
4483 74 : shared_resource->nomem_threshold = 0;
4484 74 : shared_resource->shared_ch = ch->channel;
4485 74 : shared_resource->ref = 1;
4486 74 : TAILQ_INSERT_TAIL(&mgmt_ch->shared_resources, shared_resource, link);
4487 74 : }
4488 :
4489 75 : ch->io_outstanding = 0;
4490 75 : TAILQ_INIT(&ch->locked_ranges);
4491 75 : TAILQ_INIT(&ch->qos_queued_io);
4492 75 : ch->flags = 0;
4493 75 : ch->trace_id = bdev->internal.trace_id;
4494 75 : ch->shared_resource = shared_resource;
4495 :
4496 75 : TAILQ_INIT(&ch->io_submitted);
4497 75 : TAILQ_INIT(&ch->io_locked);
4498 75 : TAILQ_INIT(&ch->io_accel_exec);
4499 75 : TAILQ_INIT(&ch->io_memory_domain);
4500 :
4501 75 : ch->stat = bdev_alloc_io_stat(false);
4502 75 : if (ch->stat == NULL) {
4503 0 : bdev_channel_destroy_resource(ch);
4504 0 : return -1;
4505 : }
4506 :
4507 75 : ch->stat->ticks_rate = spdk_get_ticks_hz();
4508 :
4509 : #ifdef SPDK_CONFIG_VTUNE
4510 : {
4511 : char *name;
4512 : __itt_init_ittlib(NULL, 0);
4513 : name = spdk_sprintf_alloc("spdk_bdev_%s_%p", ch->bdev->name, ch);
4514 : if (!name) {
4515 : bdev_channel_destroy_resource(ch);
4516 : return -1;
4517 : }
4518 : ch->handle = __itt_string_handle_create(name);
4519 : free(name);
4520 : ch->start_tsc = spdk_get_ticks();
4521 : ch->interval_tsc = spdk_get_ticks_hz() / 100;
4522 : ch->prev_stat = bdev_alloc_io_stat(false);
4523 : if (ch->prev_stat == NULL) {
4524 : bdev_channel_destroy_resource(ch);
4525 : return -1;
4526 : }
4527 : }
4528 : #endif
4529 :
4530 75 : spdk_spin_lock(&bdev->internal.spinlock);
4531 75 : bdev_enable_qos(bdev, ch);
4532 :
4533 76 : TAILQ_FOREACH(range, &bdev->internal.locked_ranges, tailq) {
4534 : struct lba_range *new_range;
4535 :
4536 1 : new_range = calloc(1, sizeof(*new_range));
4537 1 : if (new_range == NULL) {
4538 0 : spdk_spin_unlock(&bdev->internal.spinlock);
4539 0 : bdev_channel_destroy_resource(ch);
4540 0 : return -1;
4541 : }
4542 1 : new_range->length = range->length;
4543 1 : new_range->offset = range->offset;
4544 1 : new_range->locked_ctx = range->locked_ctx;
4545 1 : TAILQ_INSERT_TAIL(&ch->locked_ranges, new_range, tailq);
4546 1 : }
4547 :
4548 75 : spdk_spin_unlock(&bdev->internal.spinlock);
4549 :
4550 75 : return 0;
4551 77 : }
4552 :
4553 : static int
4554 0 : bdev_abort_all_buf_io_cb(struct spdk_iobuf_channel *ch, struct spdk_iobuf_entry *entry,
4555 : void *cb_ctx)
4556 : {
4557 0 : struct spdk_bdev_channel *bdev_ch = cb_ctx;
4558 : struct spdk_bdev_io *bdev_io;
4559 : uint64_t buf_len;
4560 :
4561 0 : bdev_io = SPDK_CONTAINEROF(entry, struct spdk_bdev_io, internal.iobuf);
4562 0 : if (bdev_io->internal.ch == bdev_ch) {
4563 0 : buf_len = bdev_io_get_max_buf_len(bdev_io, bdev_io->internal.buf.len);
4564 0 : spdk_iobuf_entry_abort(ch, entry, buf_len);
4565 0 : spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_ABORTED);
4566 0 : }
4567 :
4568 0 : return 0;
4569 : }
4570 :
4571 : /*
4572 : * Abort I/O that are waiting on a data buffer.
4573 : */
4574 : static void
4575 98 : bdev_abort_all_buf_io(struct spdk_bdev_mgmt_channel *mgmt_ch, struct spdk_bdev_channel *ch)
4576 : {
4577 98 : spdk_iobuf_for_each_entry(&mgmt_ch->iobuf, bdev_abort_all_buf_io_cb, ch);
4578 98 : }
4579 :
4580 : /*
4581 : * Abort I/O that are queued waiting for submission. These types of I/O are
4582 : * linked using the spdk_bdev_io link TAILQ_ENTRY.
4583 : */
4584 : static void
4585 117 : bdev_abort_all_queued_io(bdev_io_tailq_t *queue, struct spdk_bdev_channel *ch)
4586 : {
4587 : struct spdk_bdev_io *bdev_io, *tmp;
4588 :
4589 156 : TAILQ_FOREACH_SAFE(bdev_io, queue, internal.link, tmp) {
4590 39 : if (bdev_io->internal.ch == ch) {
4591 39 : TAILQ_REMOVE(queue, bdev_io, internal.link);
4592 : /*
4593 : * spdk_bdev_io_complete() assumes that the completed I/O had
4594 : * been submitted to the bdev module. Since in this case it
4595 : * hadn't, bump io_outstanding to account for the decrement
4596 : * that spdk_bdev_io_complete() will do.
4597 : */
4598 39 : if (bdev_io->type != SPDK_BDEV_IO_TYPE_RESET) {
4599 39 : bdev_io_increment_outstanding(ch, ch->shared_resource);
4600 39 : }
4601 39 : spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_ABORTED);
4602 39 : }
4603 39 : }
4604 117 : }
4605 :
4606 : static bool
4607 18 : bdev_abort_queued_io(bdev_io_tailq_t *queue, struct spdk_bdev_io *bio_to_abort)
4608 : {
4609 : struct spdk_bdev_io *bdev_io;
4610 :
4611 18 : TAILQ_FOREACH(bdev_io, queue, internal.link) {
4612 0 : if (bdev_io == bio_to_abort) {
4613 0 : TAILQ_REMOVE(queue, bio_to_abort, internal.link);
4614 0 : spdk_bdev_io_complete(bio_to_abort, SPDK_BDEV_IO_STATUS_ABORTED);
4615 0 : return true;
4616 : }
4617 0 : }
4618 :
4619 18 : return false;
4620 18 : }
4621 :
4622 : static int
4623 0 : bdev_abort_buf_io_cb(struct spdk_iobuf_channel *ch, struct spdk_iobuf_entry *entry, void *cb_ctx)
4624 : {
4625 0 : struct spdk_bdev_io *bdev_io, *bio_to_abort = cb_ctx;
4626 : uint64_t buf_len;
4627 :
4628 0 : bdev_io = SPDK_CONTAINEROF(entry, struct spdk_bdev_io, internal.iobuf);
4629 0 : if (bdev_io == bio_to_abort) {
4630 0 : buf_len = bdev_io_get_max_buf_len(bdev_io, bdev_io->internal.buf.len);
4631 0 : spdk_iobuf_entry_abort(ch, entry, buf_len);
4632 0 : spdk_bdev_io_complete(bio_to_abort, SPDK_BDEV_IO_STATUS_ABORTED);
4633 0 : return 1;
4634 : }
4635 :
4636 0 : return 0;
4637 0 : }
4638 :
4639 : static bool
4640 16 : bdev_abort_buf_io(struct spdk_bdev_mgmt_channel *mgmt_ch, struct spdk_bdev_io *bio_to_abort)
4641 : {
4642 : int rc;
4643 :
4644 16 : rc = spdk_iobuf_for_each_entry(&mgmt_ch->iobuf, bdev_abort_buf_io_cb, bio_to_abort);
4645 16 : return rc == 1;
4646 : }
4647 :
4648 : static void
4649 7 : bdev_qos_channel_destroy(void *cb_arg)
4650 : {
4651 7 : struct spdk_bdev_qos *qos = cb_arg;
4652 :
4653 7 : spdk_put_io_channel(spdk_io_channel_from_ctx(qos->ch));
4654 7 : spdk_poller_unregister(&qos->poller);
4655 :
4656 7 : SPDK_DEBUGLOG(bdev, "Free QoS %p.\n", qos);
4657 :
4658 7 : free(qos);
4659 7 : }
4660 :
4661 : static int
4662 7 : bdev_qos_destroy(struct spdk_bdev *bdev)
4663 : {
4664 : int i;
4665 :
4666 : /*
4667 : * Cleanly shutting down the QoS poller is tricky, because
4668 : * during the asynchronous operation the user could open
4669 : * a new descriptor and create a new channel, spawning
4670 : * a new QoS poller.
4671 : *
4672 : * The strategy is to create a new QoS structure here and swap it
4673 : * in. The shutdown path then continues to refer to the old one
4674 : * until it completes and then releases it.
4675 : */
4676 : struct spdk_bdev_qos *new_qos, *old_qos;
4677 :
4678 7 : old_qos = bdev->internal.qos;
4679 :
4680 7 : new_qos = calloc(1, sizeof(*new_qos));
4681 7 : if (!new_qos) {
4682 0 : SPDK_ERRLOG("Unable to allocate memory to shut down QoS.\n");
4683 0 : return -ENOMEM;
4684 : }
4685 :
4686 : /* Copy the old QoS data into the newly allocated structure */
4687 7 : memcpy(new_qos, old_qos, sizeof(*new_qos));
4688 :
4689 : /* Zero out the key parts of the QoS structure */
4690 7 : new_qos->ch = NULL;
4691 7 : new_qos->thread = NULL;
4692 7 : new_qos->poller = NULL;
4693 : /*
4694 : * The limit member of spdk_bdev_qos_limit structure is not zeroed.
4695 : * It will be used later for the new QoS structure.
4696 : */
4697 35 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
4698 28 : new_qos->rate_limits[i].remaining_this_timeslice = 0;
4699 28 : new_qos->rate_limits[i].min_per_timeslice = 0;
4700 28 : new_qos->rate_limits[i].max_per_timeslice = 0;
4701 28 : }
4702 :
4703 7 : bdev->internal.qos = new_qos;
4704 :
4705 7 : if (old_qos->thread == NULL) {
4706 0 : free(old_qos);
4707 0 : } else {
4708 7 : spdk_thread_send_msg(old_qos->thread, bdev_qos_channel_destroy, old_qos);
4709 : }
4710 :
4711 : /* It is safe to continue with destroying the bdev even though the QoS channel hasn't
4712 : * been destroyed yet. The destruction path will end up waiting for the final
4713 : * channel to be put before it releases resources. */
4714 :
4715 7 : return 0;
4716 7 : }
4717 :
4718 : void
4719 79 : spdk_bdev_add_io_stat(struct spdk_bdev_io_stat *total, struct spdk_bdev_io_stat *add)
4720 : {
4721 79 : total->bytes_read += add->bytes_read;
4722 79 : total->num_read_ops += add->num_read_ops;
4723 79 : total->bytes_written += add->bytes_written;
4724 79 : total->num_write_ops += add->num_write_ops;
4725 79 : total->bytes_unmapped += add->bytes_unmapped;
4726 79 : total->num_unmap_ops += add->num_unmap_ops;
4727 79 : total->bytes_copied += add->bytes_copied;
4728 79 : total->num_copy_ops += add->num_copy_ops;
4729 79 : total->read_latency_ticks += add->read_latency_ticks;
4730 79 : total->write_latency_ticks += add->write_latency_ticks;
4731 79 : total->unmap_latency_ticks += add->unmap_latency_ticks;
4732 79 : total->copy_latency_ticks += add->copy_latency_ticks;
4733 79 : if (total->max_read_latency_ticks < add->max_read_latency_ticks) {
4734 7 : total->max_read_latency_ticks = add->max_read_latency_ticks;
4735 7 : }
4736 79 : if (total->min_read_latency_ticks > add->min_read_latency_ticks) {
4737 39 : total->min_read_latency_ticks = add->min_read_latency_ticks;
4738 39 : }
4739 79 : if (total->max_write_latency_ticks < add->max_write_latency_ticks) {
4740 4 : total->max_write_latency_ticks = add->max_write_latency_ticks;
4741 4 : }
4742 79 : if (total->min_write_latency_ticks > add->min_write_latency_ticks) {
4743 24 : total->min_write_latency_ticks = add->min_write_latency_ticks;
4744 24 : }
4745 79 : if (total->max_unmap_latency_ticks < add->max_unmap_latency_ticks) {
4746 0 : total->max_unmap_latency_ticks = add->max_unmap_latency_ticks;
4747 0 : }
4748 79 : if (total->min_unmap_latency_ticks > add->min_unmap_latency_ticks) {
4749 3 : total->min_unmap_latency_ticks = add->min_unmap_latency_ticks;
4750 3 : }
4751 79 : if (total->max_copy_latency_ticks < add->max_copy_latency_ticks) {
4752 0 : total->max_copy_latency_ticks = add->max_copy_latency_ticks;
4753 0 : }
4754 79 : if (total->min_copy_latency_ticks > add->min_copy_latency_ticks) {
4755 4 : total->min_copy_latency_ticks = add->min_copy_latency_ticks;
4756 4 : }
4757 79 : }
4758 :
4759 : static void
4760 5 : bdev_get_io_stat(struct spdk_bdev_io_stat *to_stat, struct spdk_bdev_io_stat *from_stat)
4761 : {
4762 5 : memcpy(to_stat, from_stat, offsetof(struct spdk_bdev_io_stat, io_error));
4763 :
4764 5 : if (to_stat->io_error != NULL && from_stat->io_error != NULL) {
4765 0 : memcpy(to_stat->io_error, from_stat->io_error,
4766 : sizeof(struct spdk_bdev_io_error_stat));
4767 0 : }
4768 5 : }
4769 :
4770 : void
4771 216 : spdk_bdev_reset_io_stat(struct spdk_bdev_io_stat *stat, enum spdk_bdev_reset_stat_mode mode)
4772 : {
4773 216 : if (mode == SPDK_BDEV_RESET_STAT_NONE) {
4774 5 : return;
4775 : }
4776 :
4777 211 : stat->max_read_latency_ticks = 0;
4778 211 : stat->min_read_latency_ticks = UINT64_MAX;
4779 211 : stat->max_write_latency_ticks = 0;
4780 211 : stat->min_write_latency_ticks = UINT64_MAX;
4781 211 : stat->max_unmap_latency_ticks = 0;
4782 211 : stat->min_unmap_latency_ticks = UINT64_MAX;
4783 211 : stat->max_copy_latency_ticks = 0;
4784 211 : stat->min_copy_latency_ticks = UINT64_MAX;
4785 :
4786 211 : if (mode != SPDK_BDEV_RESET_STAT_ALL) {
4787 2 : return;
4788 : }
4789 :
4790 209 : stat->bytes_read = 0;
4791 209 : stat->num_read_ops = 0;
4792 209 : stat->bytes_written = 0;
4793 209 : stat->num_write_ops = 0;
4794 209 : stat->bytes_unmapped = 0;
4795 209 : stat->num_unmap_ops = 0;
4796 209 : stat->bytes_copied = 0;
4797 209 : stat->num_copy_ops = 0;
4798 209 : stat->read_latency_ticks = 0;
4799 209 : stat->write_latency_ticks = 0;
4800 209 : stat->unmap_latency_ticks = 0;
4801 209 : stat->copy_latency_ticks = 0;
4802 :
4803 209 : if (stat->io_error != NULL) {
4804 133 : memset(stat->io_error, 0, sizeof(struct spdk_bdev_io_error_stat));
4805 133 : }
4806 216 : }
4807 :
4808 : struct spdk_bdev_io_stat *
4809 207 : bdev_alloc_io_stat(bool io_error_stat)
4810 : {
4811 : struct spdk_bdev_io_stat *stat;
4812 :
4813 207 : stat = malloc(sizeof(struct spdk_bdev_io_stat));
4814 207 : if (stat == NULL) {
4815 0 : return NULL;
4816 : }
4817 :
4818 207 : if (io_error_stat) {
4819 132 : stat->io_error = malloc(sizeof(struct spdk_bdev_io_error_stat));
4820 132 : if (stat->io_error == NULL) {
4821 0 : free(stat);
4822 0 : return NULL;
4823 : }
4824 132 : } else {
4825 75 : stat->io_error = NULL;
4826 : }
4827 :
4828 207 : spdk_bdev_reset_io_stat(stat, SPDK_BDEV_RESET_STAT_ALL);
4829 :
4830 207 : return stat;
4831 207 : }
4832 :
4833 : void
4834 207 : bdev_free_io_stat(struct spdk_bdev_io_stat *stat)
4835 : {
4836 207 : if (stat != NULL) {
4837 207 : free(stat->io_error);
4838 207 : free(stat);
4839 207 : }
4840 207 : }
4841 :
4842 : void
4843 0 : spdk_bdev_dump_io_stat_json(struct spdk_bdev_io_stat *stat, struct spdk_json_write_ctx *w)
4844 : {
4845 : int i;
4846 :
4847 0 : spdk_json_write_named_uint64(w, "bytes_read", stat->bytes_read);
4848 0 : spdk_json_write_named_uint64(w, "num_read_ops", stat->num_read_ops);
4849 0 : spdk_json_write_named_uint64(w, "bytes_written", stat->bytes_written);
4850 0 : spdk_json_write_named_uint64(w, "num_write_ops", stat->num_write_ops);
4851 0 : spdk_json_write_named_uint64(w, "bytes_unmapped", stat->bytes_unmapped);
4852 0 : spdk_json_write_named_uint64(w, "num_unmap_ops", stat->num_unmap_ops);
4853 0 : spdk_json_write_named_uint64(w, "bytes_copied", stat->bytes_copied);
4854 0 : spdk_json_write_named_uint64(w, "num_copy_ops", stat->num_copy_ops);
4855 0 : spdk_json_write_named_uint64(w, "read_latency_ticks", stat->read_latency_ticks);
4856 0 : spdk_json_write_named_uint64(w, "max_read_latency_ticks", stat->max_read_latency_ticks);
4857 0 : spdk_json_write_named_uint64(w, "min_read_latency_ticks",
4858 0 : stat->min_read_latency_ticks != UINT64_MAX ?
4859 0 : stat->min_read_latency_ticks : 0);
4860 0 : spdk_json_write_named_uint64(w, "write_latency_ticks", stat->write_latency_ticks);
4861 0 : spdk_json_write_named_uint64(w, "max_write_latency_ticks", stat->max_write_latency_ticks);
4862 0 : spdk_json_write_named_uint64(w, "min_write_latency_ticks",
4863 0 : stat->min_write_latency_ticks != UINT64_MAX ?
4864 0 : stat->min_write_latency_ticks : 0);
4865 0 : spdk_json_write_named_uint64(w, "unmap_latency_ticks", stat->unmap_latency_ticks);
4866 0 : spdk_json_write_named_uint64(w, "max_unmap_latency_ticks", stat->max_unmap_latency_ticks);
4867 0 : spdk_json_write_named_uint64(w, "min_unmap_latency_ticks",
4868 0 : stat->min_unmap_latency_ticks != UINT64_MAX ?
4869 0 : stat->min_unmap_latency_ticks : 0);
4870 0 : spdk_json_write_named_uint64(w, "copy_latency_ticks", stat->copy_latency_ticks);
4871 0 : spdk_json_write_named_uint64(w, "max_copy_latency_ticks", stat->max_copy_latency_ticks);
4872 0 : spdk_json_write_named_uint64(w, "min_copy_latency_ticks",
4873 0 : stat->min_copy_latency_ticks != UINT64_MAX ?
4874 0 : stat->min_copy_latency_ticks : 0);
4875 :
4876 0 : if (stat->io_error != NULL) {
4877 0 : spdk_json_write_named_object_begin(w, "io_error");
4878 0 : for (i = 0; i < -SPDK_MIN_BDEV_IO_STATUS; i++) {
4879 0 : if (stat->io_error->error_status[i] != 0) {
4880 0 : spdk_json_write_named_uint32(w, bdev_io_status_get_string(-(i + 1)),
4881 0 : stat->io_error->error_status[i]);
4882 0 : }
4883 0 : }
4884 0 : spdk_json_write_object_end(w);
4885 0 : }
4886 0 : }
4887 :
4888 : static void
4889 79 : bdev_channel_abort_queued_ios(struct spdk_bdev_channel *ch)
4890 : {
4891 79 : struct spdk_bdev_shared_resource *shared_resource = ch->shared_resource;
4892 79 : struct spdk_bdev_mgmt_channel *mgmt_ch = shared_resource->mgmt_ch;
4893 :
4894 79 : bdev_abort_all_queued_io(&shared_resource->nomem_io, ch);
4895 79 : bdev_abort_all_buf_io(mgmt_ch, ch);
4896 79 : }
4897 :
4898 : static void
4899 75 : bdev_channel_destroy(void *io_device, void *ctx_buf)
4900 : {
4901 75 : struct spdk_bdev_channel *ch = ctx_buf;
4902 :
4903 75 : SPDK_DEBUGLOG(bdev, "Destroying channel %p for bdev %s on thread %p\n", ch, ch->bdev->name,
4904 : spdk_get_thread());
4905 :
4906 75 : spdk_trace_record(TRACE_BDEV_IOCH_DESTROY, ch->bdev->internal.trace_id, 0, 0,
4907 : spdk_thread_get_id(spdk_io_channel_get_thread(ch->channel)));
4908 :
4909 : /* This channel is going away, so add its statistics into the bdev so that they don't get lost. */
4910 75 : spdk_spin_lock(&ch->bdev->internal.spinlock);
4911 75 : spdk_bdev_add_io_stat(ch->bdev->internal.stat, ch->stat);
4912 75 : spdk_spin_unlock(&ch->bdev->internal.spinlock);
4913 :
4914 75 : bdev_channel_abort_queued_ios(ch);
4915 :
4916 75 : if (ch->histogram) {
4917 0 : spdk_histogram_data_free(ch->histogram);
4918 0 : }
4919 :
4920 75 : bdev_channel_destroy_resource(ch);
4921 75 : }
4922 :
4923 : /*
4924 : * If the name already exists in the global bdev name tree, RB_INSERT() returns a pointer
4925 : * to it. Hence we do not have to call bdev_get_by_name() when using this function.
4926 : */
4927 : static int
4928 267 : bdev_name_add(struct spdk_bdev_name *bdev_name, struct spdk_bdev *bdev, const char *name)
4929 : {
4930 : struct spdk_bdev_name *tmp;
4931 :
4932 267 : bdev_name->name = strdup(name);
4933 267 : if (bdev_name->name == NULL) {
4934 0 : SPDK_ERRLOG("Unable to allocate bdev name\n");
4935 0 : return -ENOMEM;
4936 : }
4937 :
4938 267 : bdev_name->bdev = bdev;
4939 :
4940 267 : spdk_spin_lock(&g_bdev_mgr.spinlock);
4941 267 : tmp = RB_INSERT(bdev_name_tree, &g_bdev_mgr.bdev_names, bdev_name);
4942 267 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
4943 :
4944 267 : if (tmp != NULL) {
4945 4 : SPDK_ERRLOG("Bdev name %s already exists\n", name);
4946 4 : free(bdev_name->name);
4947 4 : return -EEXIST;
4948 : }
4949 :
4950 263 : return 0;
4951 267 : }
4952 :
4953 : static void
4954 263 : bdev_name_del_unsafe(struct spdk_bdev_name *bdev_name)
4955 : {
4956 263 : RB_REMOVE(bdev_name_tree, &g_bdev_mgr.bdev_names, bdev_name);
4957 263 : free(bdev_name->name);
4958 263 : }
4959 :
4960 : static void
4961 5 : bdev_name_del(struct spdk_bdev_name *bdev_name)
4962 : {
4963 5 : spdk_spin_lock(&g_bdev_mgr.spinlock);
4964 5 : bdev_name_del_unsafe(bdev_name);
4965 5 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
4966 5 : }
4967 :
4968 : int
4969 138 : spdk_bdev_alias_add(struct spdk_bdev *bdev, const char *alias)
4970 : {
4971 : struct spdk_bdev_alias *tmp;
4972 : int ret;
4973 :
4974 138 : if (alias == NULL) {
4975 1 : SPDK_ERRLOG("Empty alias passed\n");
4976 1 : return -EINVAL;
4977 : }
4978 :
4979 137 : tmp = calloc(1, sizeof(*tmp));
4980 137 : if (tmp == NULL) {
4981 0 : SPDK_ERRLOG("Unable to allocate alias\n");
4982 0 : return -ENOMEM;
4983 : }
4984 :
4985 137 : ret = bdev_name_add(&tmp->alias, bdev, alias);
4986 137 : if (ret != 0) {
4987 4 : free(tmp);
4988 4 : return ret;
4989 : }
4990 :
4991 133 : TAILQ_INSERT_TAIL(&bdev->aliases, tmp, tailq);
4992 :
4993 133 : return 0;
4994 138 : }
4995 :
4996 : static int
4997 134 : bdev_alias_del(struct spdk_bdev *bdev, const char *alias,
4998 : void (*alias_del_fn)(struct spdk_bdev_name *n))
4999 : {
5000 : struct spdk_bdev_alias *tmp;
5001 :
5002 139 : TAILQ_FOREACH(tmp, &bdev->aliases, tailq) {
5003 135 : if (strcmp(alias, tmp->alias.name) == 0) {
5004 130 : TAILQ_REMOVE(&bdev->aliases, tmp, tailq);
5005 130 : alias_del_fn(&tmp->alias);
5006 130 : free(tmp);
5007 130 : return 0;
5008 : }
5009 5 : }
5010 :
5011 4 : return -ENOENT;
5012 134 : }
5013 :
5014 : int
5015 4 : spdk_bdev_alias_del(struct spdk_bdev *bdev, const char *alias)
5016 : {
5017 : int rc;
5018 :
5019 4 : rc = bdev_alias_del(bdev, alias, bdev_name_del);
5020 4 : if (rc == -ENOENT) {
5021 2 : SPDK_INFOLOG(bdev, "Alias %s does not exist\n", alias);
5022 2 : }
5023 :
5024 4 : return rc;
5025 : }
5026 :
5027 : void
5028 2 : spdk_bdev_alias_del_all(struct spdk_bdev *bdev)
5029 : {
5030 : struct spdk_bdev_alias *p, *tmp;
5031 :
5032 5 : TAILQ_FOREACH_SAFE(p, &bdev->aliases, tailq, tmp) {
5033 3 : TAILQ_REMOVE(&bdev->aliases, p, tailq);
5034 3 : bdev_name_del(&p->alias);
5035 3 : free(p);
5036 3 : }
5037 2 : }
5038 :
5039 : struct spdk_io_channel *
5040 77 : spdk_bdev_get_io_channel(struct spdk_bdev_desc *desc)
5041 : {
5042 77 : return spdk_get_io_channel(__bdev_to_io_dev(spdk_bdev_desc_get_bdev(desc)));
5043 : }
5044 :
5045 : void *
5046 0 : spdk_bdev_get_module_ctx(struct spdk_bdev_desc *desc)
5047 : {
5048 0 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5049 0 : void *ctx = NULL;
5050 :
5051 0 : if (bdev->fn_table->get_module_ctx) {
5052 0 : ctx = bdev->fn_table->get_module_ctx(bdev->ctxt);
5053 0 : }
5054 :
5055 0 : return ctx;
5056 : }
5057 :
5058 : const char *
5059 0 : spdk_bdev_get_module_name(const struct spdk_bdev *bdev)
5060 : {
5061 0 : return bdev->module->name;
5062 : }
5063 :
5064 : const char *
5065 263 : spdk_bdev_get_name(const struct spdk_bdev *bdev)
5066 : {
5067 263 : return bdev->name;
5068 : }
5069 :
5070 : const char *
5071 0 : spdk_bdev_get_product_name(const struct spdk_bdev *bdev)
5072 : {
5073 0 : return bdev->product_name;
5074 : }
5075 :
5076 : const struct spdk_bdev_aliases_list *
5077 0 : spdk_bdev_get_aliases(const struct spdk_bdev *bdev)
5078 : {
5079 0 : return &bdev->aliases;
5080 : }
5081 :
5082 : uint32_t
5083 5 : spdk_bdev_get_block_size(const struct spdk_bdev *bdev)
5084 : {
5085 5 : return bdev->blocklen;
5086 : }
5087 :
5088 : uint32_t
5089 0 : spdk_bdev_get_write_unit_size(const struct spdk_bdev *bdev)
5090 : {
5091 0 : return bdev->write_unit_size;
5092 : }
5093 :
5094 : uint64_t
5095 0 : spdk_bdev_get_num_blocks(const struct spdk_bdev *bdev)
5096 : {
5097 0 : return bdev->blockcnt;
5098 : }
5099 :
5100 : const char *
5101 0 : spdk_bdev_get_qos_rpc_type(enum spdk_bdev_qos_rate_limit_type type)
5102 : {
5103 0 : return qos_rpc_type[type];
5104 : }
5105 :
5106 : void
5107 0 : spdk_bdev_get_qos_rate_limits(struct spdk_bdev *bdev, uint64_t *limits)
5108 : {
5109 : int i;
5110 :
5111 0 : memset(limits, 0, sizeof(*limits) * SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES);
5112 :
5113 0 : spdk_spin_lock(&bdev->internal.spinlock);
5114 0 : if (bdev->internal.qos) {
5115 0 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
5116 0 : if (bdev->internal.qos->rate_limits[i].limit !=
5117 : SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
5118 0 : limits[i] = bdev->internal.qos->rate_limits[i].limit;
5119 0 : if (bdev_qos_is_iops_rate_limit(i) == false) {
5120 : /* Change from Byte to Megabyte which is user visible. */
5121 0 : limits[i] = limits[i] / 1024 / 1024;
5122 0 : }
5123 0 : }
5124 0 : }
5125 0 : }
5126 0 : spdk_spin_unlock(&bdev->internal.spinlock);
5127 0 : }
5128 :
5129 : size_t
5130 280 : spdk_bdev_get_buf_align(const struct spdk_bdev *bdev)
5131 : {
5132 280 : return 1 << bdev->required_alignment;
5133 : }
5134 :
5135 : uint32_t
5136 0 : spdk_bdev_get_optimal_io_boundary(const struct spdk_bdev *bdev)
5137 : {
5138 0 : return bdev->optimal_io_boundary;
5139 : }
5140 :
5141 : bool
5142 0 : spdk_bdev_has_write_cache(const struct spdk_bdev *bdev)
5143 : {
5144 0 : return bdev->write_cache;
5145 : }
5146 :
5147 : const struct spdk_uuid *
5148 0 : spdk_bdev_get_uuid(const struct spdk_bdev *bdev)
5149 : {
5150 0 : return &bdev->uuid;
5151 : }
5152 :
5153 : uint16_t
5154 0 : spdk_bdev_get_acwu(const struct spdk_bdev *bdev)
5155 : {
5156 0 : return bdev->acwu;
5157 : }
5158 :
5159 : uint32_t
5160 29 : spdk_bdev_get_md_size(const struct spdk_bdev *bdev)
5161 : {
5162 29 : return bdev->md_len;
5163 : }
5164 :
5165 : bool
5166 135 : spdk_bdev_is_md_interleaved(const struct spdk_bdev *bdev)
5167 : {
5168 135 : return (bdev->md_len != 0) && bdev->md_interleave;
5169 : }
5170 :
5171 : bool
5172 117 : spdk_bdev_is_md_separate(const struct spdk_bdev *bdev)
5173 : {
5174 117 : return (bdev->md_len != 0) && !bdev->md_interleave;
5175 : }
5176 :
5177 : bool
5178 0 : spdk_bdev_is_zoned(const struct spdk_bdev *bdev)
5179 : {
5180 0 : return bdev->zoned;
5181 : }
5182 :
5183 : uint32_t
5184 126 : spdk_bdev_get_data_block_size(const struct spdk_bdev *bdev)
5185 : {
5186 126 : if (spdk_bdev_is_md_interleaved(bdev)) {
5187 0 : return bdev->blocklen - bdev->md_len;
5188 : } else {
5189 126 : return bdev->blocklen;
5190 : }
5191 126 : }
5192 :
5193 : uint32_t
5194 0 : spdk_bdev_get_physical_block_size(const struct spdk_bdev *bdev)
5195 : {
5196 0 : return bdev->phys_blocklen;
5197 : }
5198 :
5199 : static uint32_t
5200 9 : _bdev_get_block_size_with_md(const struct spdk_bdev *bdev)
5201 : {
5202 9 : if (!spdk_bdev_is_md_interleaved(bdev)) {
5203 6 : return bdev->blocklen + bdev->md_len;
5204 : } else {
5205 3 : return bdev->blocklen;
5206 : }
5207 9 : }
5208 :
5209 : /* We have to use the typedef in the function declaration to appease astyle. */
5210 : typedef enum spdk_dif_type spdk_dif_type_t;
5211 : typedef enum spdk_dif_pi_format spdk_dif_pi_format_t;
5212 :
5213 : spdk_dif_type_t
5214 0 : spdk_bdev_get_dif_type(const struct spdk_bdev *bdev)
5215 : {
5216 0 : if (bdev->md_len != 0) {
5217 0 : return bdev->dif_type;
5218 : } else {
5219 0 : return SPDK_DIF_DISABLE;
5220 : }
5221 0 : }
5222 :
5223 : spdk_dif_pi_format_t
5224 0 : spdk_bdev_get_dif_pi_format(const struct spdk_bdev *bdev)
5225 : {
5226 0 : return bdev->dif_pi_format;
5227 : }
5228 :
5229 : bool
5230 0 : spdk_bdev_is_dif_head_of_md(const struct spdk_bdev *bdev)
5231 : {
5232 0 : if (spdk_bdev_get_dif_type(bdev) != SPDK_DIF_DISABLE) {
5233 0 : return bdev->dif_is_head_of_md;
5234 : } else {
5235 0 : return false;
5236 : }
5237 0 : }
5238 :
5239 : bool
5240 0 : spdk_bdev_is_dif_check_enabled(const struct spdk_bdev *bdev,
5241 : enum spdk_dif_check_type check_type)
5242 : {
5243 0 : if (spdk_bdev_get_dif_type(bdev) == SPDK_DIF_DISABLE) {
5244 0 : return false;
5245 : }
5246 :
5247 0 : switch (check_type) {
5248 : case SPDK_DIF_CHECK_TYPE_REFTAG:
5249 0 : return (bdev->dif_check_flags & SPDK_DIF_FLAGS_REFTAG_CHECK) != 0;
5250 : case SPDK_DIF_CHECK_TYPE_APPTAG:
5251 0 : return (bdev->dif_check_flags & SPDK_DIF_FLAGS_APPTAG_CHECK) != 0;
5252 : case SPDK_DIF_CHECK_TYPE_GUARD:
5253 0 : return (bdev->dif_check_flags & SPDK_DIF_FLAGS_GUARD_CHECK) != 0;
5254 : default:
5255 0 : return false;
5256 : }
5257 0 : }
5258 :
5259 : static uint32_t
5260 3 : bdev_get_max_write(const struct spdk_bdev *bdev, uint64_t num_bytes)
5261 : {
5262 : uint64_t aligned_length, max_write_blocks;
5263 :
5264 3 : aligned_length = num_bytes - (spdk_bdev_get_buf_align(bdev) - 1);
5265 3 : max_write_blocks = aligned_length / _bdev_get_block_size_with_md(bdev);
5266 3 : max_write_blocks -= max_write_blocks % bdev->write_unit_size;
5267 :
5268 3 : return max_write_blocks;
5269 : }
5270 :
5271 : uint32_t
5272 1 : spdk_bdev_get_max_copy(const struct spdk_bdev *bdev)
5273 : {
5274 1 : return bdev->max_copy;
5275 : }
5276 :
5277 : uint64_t
5278 0 : spdk_bdev_get_qd(const struct spdk_bdev *bdev)
5279 : {
5280 0 : return bdev->internal.measured_queue_depth;
5281 : }
5282 :
5283 : uint64_t
5284 0 : spdk_bdev_get_qd_sampling_period(const struct spdk_bdev *bdev)
5285 : {
5286 0 : return bdev->internal.period;
5287 : }
5288 :
5289 : uint64_t
5290 0 : spdk_bdev_get_weighted_io_time(const struct spdk_bdev *bdev)
5291 : {
5292 0 : return bdev->internal.weighted_io_time;
5293 : }
5294 :
5295 : uint64_t
5296 0 : spdk_bdev_get_io_time(const struct spdk_bdev *bdev)
5297 : {
5298 0 : return bdev->internal.io_time;
5299 : }
5300 :
5301 0 : union spdk_bdev_nvme_ctratt spdk_bdev_get_nvme_ctratt(struct spdk_bdev *bdev)
5302 : {
5303 0 : return bdev->ctratt;
5304 : }
5305 :
5306 : uint32_t
5307 0 : spdk_bdev_get_nvme_nsid(struct spdk_bdev *bdev)
5308 : {
5309 0 : return bdev->nsid;
5310 : }
5311 :
5312 : uint32_t
5313 0 : spdk_bdev_desc_get_block_size(struct spdk_bdev_desc *desc)
5314 : {
5315 0 : struct spdk_bdev *bdev = desc->bdev;
5316 :
5317 0 : return desc->opts.hide_metadata ? bdev->blocklen - bdev->md_len : bdev->blocklen;
5318 : }
5319 :
5320 : uint32_t
5321 0 : spdk_bdev_desc_get_md_size(struct spdk_bdev_desc *desc)
5322 : {
5323 0 : struct spdk_bdev *bdev = desc->bdev;
5324 :
5325 0 : return desc->opts.hide_metadata ? 0 : bdev->md_len;
5326 : }
5327 :
5328 : bool
5329 0 : spdk_bdev_desc_is_md_interleaved(struct spdk_bdev_desc *desc)
5330 : {
5331 0 : struct spdk_bdev *bdev = desc->bdev;
5332 :
5333 0 : return desc->opts.hide_metadata ? false : spdk_bdev_is_md_interleaved(bdev);
5334 : }
5335 :
5336 : bool
5337 0 : spdk_bdev_desc_is_md_separate(struct spdk_bdev_desc *desc)
5338 : {
5339 0 : struct spdk_bdev *bdev = desc->bdev;
5340 :
5341 0 : return desc->opts.hide_metadata ? false : spdk_bdev_is_md_separate(bdev);
5342 : }
5343 :
5344 : spdk_dif_type_t
5345 0 : spdk_bdev_desc_get_dif_type(struct spdk_bdev_desc *desc)
5346 : {
5347 0 : struct spdk_bdev *bdev = desc->bdev;
5348 :
5349 0 : return desc->opts.hide_metadata ? SPDK_DIF_DISABLE : spdk_bdev_get_dif_type(bdev);
5350 : }
5351 :
5352 : spdk_dif_pi_format_t
5353 0 : spdk_bdev_desc_get_dif_pi_format(struct spdk_bdev_desc *desc)
5354 : {
5355 0 : struct spdk_bdev *bdev = desc->bdev;
5356 :
5357 0 : return desc->opts.hide_metadata ? SPDK_DIF_PI_FORMAT_16 : spdk_bdev_get_dif_pi_format(bdev);
5358 : }
5359 :
5360 : bool
5361 0 : spdk_bdev_desc_is_dif_head_of_md(struct spdk_bdev_desc *desc)
5362 : {
5363 0 : struct spdk_bdev *bdev = desc->bdev;
5364 :
5365 0 : return desc->opts.hide_metadata ? false : spdk_bdev_is_dif_head_of_md(bdev);
5366 : }
5367 :
5368 : bool
5369 0 : spdk_bdev_desc_is_dif_check_enabled(struct spdk_bdev_desc *desc,
5370 : enum spdk_dif_check_type check_type)
5371 : {
5372 0 : struct spdk_bdev *bdev = desc->bdev;
5373 :
5374 0 : return desc->opts.hide_metadata ? false : spdk_bdev_is_dif_check_enabled(bdev, check_type);
5375 : }
5376 :
5377 : static void bdev_update_qd_sampling_period(void *ctx);
5378 :
5379 : static void
5380 1 : _calculate_measured_qd_cpl(struct spdk_bdev *bdev, void *_ctx, int status)
5381 : {
5382 1 : bdev->internal.measured_queue_depth = bdev->internal.temporary_queue_depth;
5383 :
5384 1 : if (bdev->internal.measured_queue_depth) {
5385 0 : bdev->internal.io_time += bdev->internal.period;
5386 0 : bdev->internal.weighted_io_time += bdev->internal.period * bdev->internal.measured_queue_depth;
5387 0 : }
5388 :
5389 1 : bdev->internal.qd_poll_in_progress = false;
5390 :
5391 1 : bdev_update_qd_sampling_period(bdev);
5392 1 : }
5393 :
5394 : static void
5395 1 : _calculate_measured_qd(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
5396 : struct spdk_io_channel *io_ch, void *_ctx)
5397 : {
5398 1 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(io_ch);
5399 :
5400 1 : bdev->internal.temporary_queue_depth += ch->io_outstanding;
5401 1 : spdk_bdev_for_each_channel_continue(i, 0);
5402 1 : }
5403 :
5404 : static int
5405 1 : bdev_calculate_measured_queue_depth(void *ctx)
5406 : {
5407 1 : struct spdk_bdev *bdev = ctx;
5408 :
5409 1 : bdev->internal.qd_poll_in_progress = true;
5410 1 : bdev->internal.temporary_queue_depth = 0;
5411 1 : spdk_bdev_for_each_channel(bdev, _calculate_measured_qd, bdev, _calculate_measured_qd_cpl);
5412 1 : return SPDK_POLLER_BUSY;
5413 : }
5414 :
5415 : static void
5416 5 : bdev_update_qd_sampling_period(void *ctx)
5417 : {
5418 5 : struct spdk_bdev *bdev = ctx;
5419 :
5420 5 : if (bdev->internal.period == bdev->internal.new_period) {
5421 0 : return;
5422 : }
5423 :
5424 5 : if (bdev->internal.qd_poll_in_progress) {
5425 1 : return;
5426 : }
5427 :
5428 4 : bdev->internal.period = bdev->internal.new_period;
5429 :
5430 4 : spdk_poller_unregister(&bdev->internal.qd_poller);
5431 4 : if (bdev->internal.period != 0) {
5432 2 : bdev->internal.qd_poller = SPDK_POLLER_REGISTER(bdev_calculate_measured_queue_depth,
5433 : bdev, bdev->internal.period);
5434 2 : } else {
5435 2 : spdk_bdev_close(bdev->internal.qd_desc);
5436 2 : bdev->internal.qd_desc = NULL;
5437 : }
5438 5 : }
5439 :
5440 : static void
5441 0 : _tmp_bdev_event_cb(enum spdk_bdev_event_type type, struct spdk_bdev *bdev, void *ctx)
5442 : {
5443 0 : SPDK_NOTICELOG("Unexpected event type: %d\n", type);
5444 0 : }
5445 :
5446 : void
5447 135 : spdk_bdev_set_qd_sampling_period(struct spdk_bdev *bdev, uint64_t period)
5448 : {
5449 : int rc;
5450 :
5451 135 : if (bdev->internal.new_period == period) {
5452 129 : return;
5453 : }
5454 :
5455 6 : bdev->internal.new_period = period;
5456 :
5457 6 : if (bdev->internal.qd_desc != NULL) {
5458 4 : assert(bdev->internal.period != 0);
5459 :
5460 8 : spdk_thread_send_msg(bdev->internal.qd_desc->thread,
5461 4 : bdev_update_qd_sampling_period, bdev);
5462 4 : return;
5463 : }
5464 :
5465 2 : assert(bdev->internal.period == 0);
5466 :
5467 4 : rc = spdk_bdev_open_ext(spdk_bdev_get_name(bdev), false, _tmp_bdev_event_cb,
5468 2 : NULL, &bdev->internal.qd_desc);
5469 2 : if (rc != 0) {
5470 0 : return;
5471 : }
5472 :
5473 2 : bdev->internal.period = period;
5474 2 : bdev->internal.qd_poller = SPDK_POLLER_REGISTER(bdev_calculate_measured_queue_depth,
5475 : bdev, period);
5476 135 : }
5477 :
5478 : struct bdev_get_current_qd_ctx {
5479 : uint64_t current_qd;
5480 : spdk_bdev_get_current_qd_cb cb_fn;
5481 : void *cb_arg;
5482 : };
5483 :
5484 : static void
5485 0 : bdev_get_current_qd_done(struct spdk_bdev *bdev, void *_ctx, int status)
5486 : {
5487 0 : struct bdev_get_current_qd_ctx *ctx = _ctx;
5488 :
5489 0 : ctx->cb_fn(bdev, ctx->current_qd, ctx->cb_arg, 0);
5490 :
5491 0 : free(ctx);
5492 0 : }
5493 :
5494 : static void
5495 0 : bdev_get_current_qd(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
5496 : struct spdk_io_channel *io_ch, void *_ctx)
5497 : {
5498 0 : struct bdev_get_current_qd_ctx *ctx = _ctx;
5499 0 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
5500 :
5501 0 : ctx->current_qd += bdev_ch->io_outstanding;
5502 :
5503 0 : spdk_bdev_for_each_channel_continue(i, 0);
5504 0 : }
5505 :
5506 : void
5507 0 : spdk_bdev_get_current_qd(struct spdk_bdev *bdev, spdk_bdev_get_current_qd_cb cb_fn,
5508 : void *cb_arg)
5509 : {
5510 : struct bdev_get_current_qd_ctx *ctx;
5511 :
5512 0 : assert(cb_fn != NULL);
5513 :
5514 0 : ctx = calloc(1, sizeof(*ctx));
5515 0 : if (ctx == NULL) {
5516 0 : cb_fn(bdev, 0, cb_arg, -ENOMEM);
5517 0 : return;
5518 : }
5519 :
5520 0 : ctx->cb_fn = cb_fn;
5521 0 : ctx->cb_arg = cb_arg;
5522 :
5523 0 : spdk_bdev_for_each_channel(bdev, bdev_get_current_qd, ctx, bdev_get_current_qd_done);
5524 0 : }
5525 :
5526 : static void
5527 25 : _event_notify(struct spdk_bdev_desc *desc, enum spdk_bdev_event_type type)
5528 : {
5529 25 : assert(desc->thread == spdk_get_thread());
5530 :
5531 25 : spdk_spin_lock(&desc->spinlock);
5532 25 : desc->refs--;
5533 25 : if (!desc->closed) {
5534 14 : spdk_spin_unlock(&desc->spinlock);
5535 28 : desc->callback.event_fn(type,
5536 14 : desc->bdev,
5537 14 : desc->callback.ctx);
5538 14 : return;
5539 11 : } else if (desc->refs == 0) {
5540 : /* This descriptor was closed after this event_notify message was sent.
5541 : * spdk_bdev_close() could not free the descriptor since this message was
5542 : * in flight, so we free it now using bdev_desc_free().
5543 : */
5544 10 : spdk_spin_unlock(&desc->spinlock);
5545 10 : bdev_desc_free(desc);
5546 10 : return;
5547 : }
5548 1 : spdk_spin_unlock(&desc->spinlock);
5549 25 : }
5550 :
5551 : static void
5552 25 : event_notify(struct spdk_bdev_desc *desc, spdk_msg_fn event_notify_fn)
5553 : {
5554 25 : spdk_spin_lock(&desc->spinlock);
5555 25 : desc->refs++;
5556 25 : spdk_thread_send_msg(desc->thread, event_notify_fn, desc);
5557 25 : spdk_spin_unlock(&desc->spinlock);
5558 25 : }
5559 :
5560 : static void
5561 6 : _resize_notify(void *ctx)
5562 : {
5563 6 : struct spdk_bdev_desc *desc = ctx;
5564 :
5565 6 : _event_notify(desc, SPDK_BDEV_EVENT_RESIZE);
5566 6 : }
5567 :
5568 : int
5569 11 : spdk_bdev_notify_blockcnt_change(struct spdk_bdev *bdev, uint64_t size)
5570 : {
5571 : struct spdk_bdev_desc *desc;
5572 : int ret;
5573 :
5574 11 : if (size == bdev->blockcnt) {
5575 0 : return 0;
5576 : }
5577 :
5578 11 : spdk_spin_lock(&bdev->internal.spinlock);
5579 :
5580 : /* bdev has open descriptors */
5581 11 : if (!TAILQ_EMPTY(&bdev->internal.open_descs) &&
5582 7 : bdev->blockcnt > size) {
5583 1 : ret = -EBUSY;
5584 1 : } else {
5585 10 : bdev->blockcnt = size;
5586 16 : TAILQ_FOREACH(desc, &bdev->internal.open_descs, link) {
5587 6 : event_notify(desc, _resize_notify);
5588 6 : }
5589 10 : ret = 0;
5590 : }
5591 :
5592 11 : spdk_spin_unlock(&bdev->internal.spinlock);
5593 :
5594 11 : return ret;
5595 11 : }
5596 :
5597 : /*
5598 : * Convert I/O offset and length from bytes to blocks.
5599 : *
5600 : * Returns zero on success or non-zero if the byte parameters aren't divisible by the block size.
5601 : */
5602 : static uint64_t
5603 20 : bdev_bytes_to_blocks(struct spdk_bdev_desc *desc, uint64_t offset_bytes,
5604 : uint64_t *offset_blocks, uint64_t num_bytes, uint64_t *num_blocks)
5605 : {
5606 20 : uint32_t block_size = bdev_desc_get_block_size(desc);
5607 : uint8_t shift_cnt;
5608 :
5609 : /* Avoid expensive div operations if possible. These spdk_u32 functions are very cheap. */
5610 20 : if (spdk_likely(spdk_u32_is_pow2(block_size))) {
5611 17 : shift_cnt = spdk_u32log2(block_size);
5612 17 : *offset_blocks = offset_bytes >> shift_cnt;
5613 17 : *num_blocks = num_bytes >> shift_cnt;
5614 34 : return (offset_bytes - (*offset_blocks << shift_cnt)) |
5615 17 : (num_bytes - (*num_blocks << shift_cnt));
5616 : } else {
5617 3 : *offset_blocks = offset_bytes / block_size;
5618 3 : *num_blocks = num_bytes / block_size;
5619 3 : return (offset_bytes % block_size) | (num_bytes % block_size);
5620 : }
5621 20 : }
5622 :
5623 : static bool
5624 689 : bdev_io_valid_blocks(struct spdk_bdev *bdev, uint64_t offset_blocks, uint64_t num_blocks)
5625 : {
5626 : /* Return failure if offset_blocks + num_blocks is less than offset_blocks; indicates there
5627 : * has been an overflow and hence the offset has been wrapped around */
5628 689 : if (offset_blocks + num_blocks < offset_blocks) {
5629 1 : return false;
5630 : }
5631 :
5632 : /* Return failure if offset_blocks + num_blocks exceeds the size of the bdev */
5633 688 : if (offset_blocks + num_blocks > bdev->blockcnt) {
5634 2 : return false;
5635 : }
5636 :
5637 686 : return true;
5638 689 : }
5639 :
5640 : static void
5641 2 : bdev_seek_complete_cb(void *ctx)
5642 : {
5643 2 : struct spdk_bdev_io *bdev_io = ctx;
5644 :
5645 2 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
5646 2 : bdev_io->internal.cb(bdev_io, true, bdev_io->internal.caller_ctx);
5647 2 : }
5648 :
5649 : static int
5650 4 : bdev_seek(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5651 : uint64_t offset_blocks, enum spdk_bdev_io_type io_type,
5652 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5653 : {
5654 4 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5655 : struct spdk_bdev_io *bdev_io;
5656 4 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5657 :
5658 4 : assert(io_type == SPDK_BDEV_IO_TYPE_SEEK_DATA || io_type == SPDK_BDEV_IO_TYPE_SEEK_HOLE);
5659 :
5660 : /* Check if offset_blocks is valid looking at the validity of one block */
5661 4 : if (!bdev_io_valid_blocks(bdev, offset_blocks, 1)) {
5662 0 : return -EINVAL;
5663 : }
5664 :
5665 4 : bdev_io = bdev_channel_get_io(channel);
5666 4 : if (!bdev_io) {
5667 0 : return -ENOMEM;
5668 : }
5669 :
5670 4 : bdev_io->internal.ch = channel;
5671 4 : bdev_io->internal.desc = desc;
5672 4 : bdev_io->type = io_type;
5673 4 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5674 4 : bdev_io->u.bdev.memory_domain = NULL;
5675 4 : bdev_io->u.bdev.memory_domain_ctx = NULL;
5676 4 : bdev_io->u.bdev.accel_sequence = NULL;
5677 4 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5678 :
5679 4 : if (!spdk_bdev_io_type_supported(bdev, io_type)) {
5680 : /* In case bdev doesn't support seek to next data/hole offset,
5681 : * it is assumed that only data and no holes are present */
5682 2 : if (io_type == SPDK_BDEV_IO_TYPE_SEEK_DATA) {
5683 1 : bdev_io->u.bdev.seek.offset = offset_blocks;
5684 1 : } else {
5685 1 : bdev_io->u.bdev.seek.offset = UINT64_MAX;
5686 : }
5687 :
5688 2 : spdk_thread_send_msg(spdk_get_thread(), bdev_seek_complete_cb, bdev_io);
5689 2 : return 0;
5690 : }
5691 :
5692 2 : bdev_io_submit(bdev_io);
5693 2 : return 0;
5694 4 : }
5695 :
5696 : int
5697 2 : spdk_bdev_seek_data(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5698 : uint64_t offset_blocks,
5699 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5700 : {
5701 2 : return bdev_seek(desc, ch, offset_blocks, SPDK_BDEV_IO_TYPE_SEEK_DATA, cb, cb_arg);
5702 : }
5703 :
5704 : int
5705 2 : spdk_bdev_seek_hole(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5706 : uint64_t offset_blocks,
5707 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5708 : {
5709 2 : return bdev_seek(desc, ch, offset_blocks, SPDK_BDEV_IO_TYPE_SEEK_HOLE, cb, cb_arg);
5710 : }
5711 :
5712 : uint64_t
5713 4 : spdk_bdev_io_get_seek_offset(const struct spdk_bdev_io *bdev_io)
5714 : {
5715 4 : return bdev_io->u.bdev.seek.offset;
5716 : }
5717 :
5718 : static int
5719 204 : bdev_read_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch, void *buf,
5720 : void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
5721 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5722 : {
5723 204 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5724 : struct spdk_bdev_io *bdev_io;
5725 204 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5726 :
5727 204 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
5728 0 : return -EINVAL;
5729 : }
5730 :
5731 204 : bdev_io = bdev_channel_get_io(channel);
5732 204 : if (!bdev_io) {
5733 1 : return -ENOMEM;
5734 : }
5735 :
5736 203 : bdev_io->internal.ch = channel;
5737 203 : bdev_io->internal.desc = desc;
5738 203 : bdev_io->type = SPDK_BDEV_IO_TYPE_READ;
5739 203 : bdev_io->u.bdev.iovs = &bdev_io->iov;
5740 203 : bdev_io->u.bdev.iovs[0].iov_base = buf;
5741 203 : bdev_io->u.bdev.iovs[0].iov_len = num_blocks * bdev_desc_get_block_size(desc);
5742 203 : bdev_io->u.bdev.iovcnt = 1;
5743 203 : bdev_io->u.bdev.md_buf = md_buf;
5744 203 : bdev_io->u.bdev.num_blocks = num_blocks;
5745 203 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5746 203 : bdev_io->u.bdev.memory_domain = NULL;
5747 203 : bdev_io->u.bdev.memory_domain_ctx = NULL;
5748 203 : bdev_io->u.bdev.accel_sequence = NULL;
5749 203 : bdev_io->u.bdev.dif_check_flags = bdev->dif_check_flags;
5750 203 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5751 :
5752 203 : bdev_io_submit(bdev_io);
5753 203 : return 0;
5754 204 : }
5755 :
5756 : int
5757 3 : spdk_bdev_read(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5758 : void *buf, uint64_t offset, uint64_t nbytes,
5759 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5760 : {
5761 : uint64_t offset_blocks, num_blocks;
5762 :
5763 3 : if (bdev_bytes_to_blocks(desc, offset, &offset_blocks, nbytes, &num_blocks) != 0) {
5764 0 : return -EINVAL;
5765 : }
5766 :
5767 3 : return spdk_bdev_read_blocks(desc, ch, buf, offset_blocks, num_blocks, cb, cb_arg);
5768 3 : }
5769 :
5770 : int
5771 200 : spdk_bdev_read_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5772 : void *buf, uint64_t offset_blocks, uint64_t num_blocks,
5773 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5774 : {
5775 200 : return bdev_read_blocks_with_md(desc, ch, buf, NULL, offset_blocks, num_blocks, cb, cb_arg);
5776 : }
5777 :
5778 : int
5779 4 : spdk_bdev_read_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5780 : void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
5781 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5782 : {
5783 8 : struct iovec iov = {
5784 4 : .iov_base = buf,
5785 : };
5786 :
5787 4 : if (md_buf && !spdk_bdev_is_md_separate(spdk_bdev_desc_get_bdev(desc))) {
5788 0 : return -EINVAL;
5789 : }
5790 :
5791 4 : if ((md_buf || desc->opts.hide_metadata) && !_is_buf_allocated(&iov)) {
5792 0 : return -EINVAL;
5793 : }
5794 :
5795 8 : return bdev_read_blocks_with_md(desc, ch, buf, md_buf, offset_blocks, num_blocks,
5796 4 : cb, cb_arg);
5797 4 : }
5798 :
5799 : int
5800 5 : spdk_bdev_readv(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5801 : struct iovec *iov, int iovcnt,
5802 : uint64_t offset, uint64_t nbytes,
5803 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5804 : {
5805 : uint64_t offset_blocks, num_blocks;
5806 :
5807 5 : if (bdev_bytes_to_blocks(desc, offset, &offset_blocks, nbytes, &num_blocks) != 0) {
5808 0 : return -EINVAL;
5809 : }
5810 :
5811 5 : return spdk_bdev_readv_blocks(desc, ch, iov, iovcnt, offset_blocks, num_blocks, cb, cb_arg);
5812 5 : }
5813 :
5814 : static int
5815 226 : bdev_readv_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5816 : struct iovec *iov, int iovcnt, void *md_buf, uint64_t offset_blocks,
5817 : uint64_t num_blocks, struct spdk_memory_domain *domain, void *domain_ctx,
5818 : struct spdk_accel_sequence *seq, uint32_t dif_check_flags,
5819 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5820 : {
5821 226 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5822 : struct spdk_bdev_io *bdev_io;
5823 226 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5824 :
5825 226 : if (spdk_unlikely(!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks))) {
5826 0 : return -EINVAL;
5827 : }
5828 :
5829 226 : bdev_io = bdev_channel_get_io(channel);
5830 226 : if (spdk_unlikely(!bdev_io)) {
5831 2 : return -ENOMEM;
5832 : }
5833 :
5834 224 : bdev_io->internal.ch = channel;
5835 224 : bdev_io->internal.desc = desc;
5836 224 : bdev_io->type = SPDK_BDEV_IO_TYPE_READ;
5837 224 : bdev_io->u.bdev.iovs = iov;
5838 224 : bdev_io->u.bdev.iovcnt = iovcnt;
5839 224 : bdev_io->u.bdev.md_buf = md_buf;
5840 224 : bdev_io->u.bdev.num_blocks = num_blocks;
5841 224 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5842 224 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
5843 :
5844 224 : if (seq != NULL) {
5845 0 : bdev_io->internal.f.has_accel_sequence = true;
5846 0 : bdev_io->internal.accel_sequence = seq;
5847 0 : }
5848 :
5849 224 : if (domain != NULL) {
5850 2 : bdev_io->internal.f.has_memory_domain = true;
5851 2 : bdev_io->internal.memory_domain = domain;
5852 2 : bdev_io->internal.memory_domain_ctx = domain_ctx;
5853 2 : }
5854 :
5855 224 : bdev_io->u.bdev.memory_domain = domain;
5856 224 : bdev_io->u.bdev.memory_domain_ctx = domain_ctx;
5857 224 : bdev_io->u.bdev.accel_sequence = seq;
5858 224 : bdev_io->u.bdev.dif_check_flags = dif_check_flags;
5859 :
5860 224 : _bdev_io_submit_ext(desc, bdev_io);
5861 :
5862 224 : return 0;
5863 226 : }
5864 :
5865 : int
5866 21 : spdk_bdev_readv_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5867 : struct iovec *iov, int iovcnt,
5868 : uint64_t offset_blocks, uint64_t num_blocks,
5869 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5870 : {
5871 21 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5872 :
5873 42 : return bdev_readv_blocks_with_md(desc, ch, iov, iovcnt, NULL, offset_blocks,
5874 21 : num_blocks, NULL, NULL, NULL, bdev->dif_check_flags, cb, cb_arg);
5875 : }
5876 :
5877 : int
5878 4 : spdk_bdev_readv_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5879 : struct iovec *iov, int iovcnt, void *md_buf,
5880 : uint64_t offset_blocks, uint64_t num_blocks,
5881 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5882 : {
5883 4 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5884 :
5885 4 : if (md_buf && !spdk_bdev_is_md_separate(bdev)) {
5886 0 : return -EINVAL;
5887 : }
5888 :
5889 4 : if (md_buf && !_is_buf_allocated(iov)) {
5890 0 : return -EINVAL;
5891 : }
5892 :
5893 8 : return bdev_readv_blocks_with_md(desc, ch, iov, iovcnt, md_buf, offset_blocks,
5894 4 : num_blocks, NULL, NULL, NULL, bdev->dif_check_flags, cb, cb_arg);
5895 4 : }
5896 :
5897 : static inline bool
5898 14 : _bdev_io_check_opts(struct spdk_bdev_ext_io_opts *opts, struct iovec *iov)
5899 : {
5900 : /*
5901 : * We check if opts size is at least of size when we first introduced
5902 : * spdk_bdev_ext_io_opts (ac6f2bdd8d) since access to those members
5903 : * are not checked internal.
5904 : */
5905 24 : return opts->size >= offsetof(struct spdk_bdev_ext_io_opts, metadata) +
5906 14 : sizeof(opts->metadata) &&
5907 10 : opts->size <= sizeof(*opts) &&
5908 : /* When memory domain is used, the user must provide data buffers */
5909 8 : (!opts->memory_domain || (iov && iov[0].iov_base));
5910 : }
5911 :
5912 : int
5913 8 : spdk_bdev_readv_blocks_ext(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5914 : struct iovec *iov, int iovcnt,
5915 : uint64_t offset_blocks, uint64_t num_blocks,
5916 : spdk_bdev_io_completion_cb cb, void *cb_arg,
5917 : struct spdk_bdev_ext_io_opts *opts)
5918 : {
5919 8 : struct spdk_memory_domain *domain = NULL;
5920 8 : struct spdk_accel_sequence *seq = NULL;
5921 8 : void *domain_ctx = NULL, *md = NULL;
5922 8 : uint32_t dif_check_flags = 0;
5923 : uint32_t nvme_cdw12_raw;
5924 8 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5925 :
5926 8 : if (opts) {
5927 7 : if (spdk_unlikely(!_bdev_io_check_opts(opts, iov))) {
5928 3 : return -EINVAL;
5929 : }
5930 :
5931 4 : md = opts->metadata;
5932 4 : domain = bdev_get_ext_io_opt(opts, memory_domain, NULL);
5933 4 : domain_ctx = bdev_get_ext_io_opt(opts, memory_domain_ctx, NULL);
5934 4 : seq = bdev_get_ext_io_opt(opts, accel_sequence, NULL);
5935 4 : nvme_cdw12_raw = bdev_get_ext_io_opt(opts, nvme_cdw12.raw, 0);
5936 4 : if (md) {
5937 4 : if (spdk_unlikely(!spdk_bdev_is_md_separate(bdev))) {
5938 0 : return -EINVAL;
5939 : }
5940 :
5941 4 : if (spdk_unlikely(!_is_buf_allocated(iov))) {
5942 0 : return -EINVAL;
5943 : }
5944 :
5945 4 : if (spdk_unlikely(seq != NULL)) {
5946 0 : return -EINVAL;
5947 : }
5948 :
5949 4 : if (nvme_cdw12_raw & SPDK_DIF_FLAGS_NVME_PRACT) {
5950 0 : SPDK_ERRLOG("Separate metadata with NVMe PRACT is not supported.\n");
5951 0 : return -ENOTSUP;
5952 : }
5953 4 : }
5954 :
5955 4 : if (nvme_cdw12_raw & SPDK_DIF_FLAGS_NVME_PRACT) {
5956 0 : dif_check_flags |= SPDK_DIF_FLAGS_NVME_PRACT;
5957 0 : }
5958 4 : }
5959 :
5960 10 : dif_check_flags |= bdev->dif_check_flags &
5961 5 : ~(bdev_get_ext_io_opt(opts, dif_check_flags_exclude_mask, 0));
5962 :
5963 10 : return bdev_readv_blocks_with_md(desc, ch, iov, iovcnt, md, offset_blocks,
5964 5 : num_blocks, domain, domain_ctx, seq, dif_check_flags, cb, cb_arg);
5965 8 : }
5966 :
5967 : static int
5968 36 : bdev_write_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
5969 : void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
5970 : spdk_bdev_io_completion_cb cb, void *cb_arg)
5971 : {
5972 36 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
5973 : struct spdk_bdev_io *bdev_io;
5974 36 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
5975 :
5976 36 : if (!desc->write) {
5977 0 : return -EBADF;
5978 : }
5979 :
5980 36 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
5981 0 : return -EINVAL;
5982 : }
5983 :
5984 36 : bdev_io = bdev_channel_get_io(channel);
5985 36 : if (!bdev_io) {
5986 0 : return -ENOMEM;
5987 : }
5988 :
5989 36 : bdev_io->internal.ch = channel;
5990 36 : bdev_io->internal.desc = desc;
5991 36 : bdev_io->type = SPDK_BDEV_IO_TYPE_WRITE;
5992 36 : bdev_io->u.bdev.iovs = &bdev_io->iov;
5993 36 : bdev_io->u.bdev.iovs[0].iov_base = buf;
5994 36 : bdev_io->u.bdev.iovs[0].iov_len = num_blocks * bdev_desc_get_block_size(desc);
5995 36 : bdev_io->u.bdev.iovcnt = 1;
5996 36 : bdev_io->u.bdev.md_buf = md_buf;
5997 36 : bdev_io->u.bdev.num_blocks = num_blocks;
5998 36 : bdev_io->u.bdev.offset_blocks = offset_blocks;
5999 36 : bdev_io->u.bdev.memory_domain = NULL;
6000 36 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6001 36 : bdev_io->u.bdev.accel_sequence = NULL;
6002 36 : bdev_io->u.bdev.dif_check_flags = bdev->dif_check_flags;
6003 36 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6004 :
6005 36 : bdev_io_submit(bdev_io);
6006 36 : return 0;
6007 36 : }
6008 :
6009 : int
6010 3 : spdk_bdev_write(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6011 : void *buf, uint64_t offset, uint64_t nbytes,
6012 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6013 : {
6014 : uint64_t offset_blocks, num_blocks;
6015 :
6016 3 : if (bdev_bytes_to_blocks(desc, offset, &offset_blocks, nbytes, &num_blocks) != 0) {
6017 0 : return -EINVAL;
6018 : }
6019 :
6020 3 : return spdk_bdev_write_blocks(desc, ch, buf, offset_blocks, num_blocks, cb, cb_arg);
6021 3 : }
6022 :
6023 : int
6024 27 : spdk_bdev_write_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6025 : void *buf, uint64_t offset_blocks, uint64_t num_blocks,
6026 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6027 : {
6028 54 : return bdev_write_blocks_with_md(desc, ch, buf, NULL, offset_blocks, num_blocks,
6029 27 : cb, cb_arg);
6030 : }
6031 :
6032 : int
6033 3 : spdk_bdev_write_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6034 : void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
6035 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6036 : {
6037 6 : struct iovec iov = {
6038 3 : .iov_base = buf,
6039 : };
6040 :
6041 3 : if (md_buf && !spdk_bdev_is_md_separate(spdk_bdev_desc_get_bdev(desc))) {
6042 0 : return -EINVAL;
6043 : }
6044 :
6045 3 : if (md_buf && !_is_buf_allocated(&iov)) {
6046 0 : return -EINVAL;
6047 : }
6048 :
6049 6 : return bdev_write_blocks_with_md(desc, ch, buf, md_buf, offset_blocks, num_blocks,
6050 3 : cb, cb_arg);
6051 3 : }
6052 :
6053 : static int
6054 70 : bdev_writev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6055 : struct iovec *iov, int iovcnt, void *md_buf,
6056 : uint64_t offset_blocks, uint64_t num_blocks,
6057 : struct spdk_memory_domain *domain, void *domain_ctx,
6058 : struct spdk_accel_sequence *seq, uint32_t dif_check_flags,
6059 : uint32_t nvme_cdw12_raw, uint32_t nvme_cdw13_raw,
6060 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6061 : {
6062 70 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6063 : struct spdk_bdev_io *bdev_io;
6064 70 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6065 :
6066 70 : if (spdk_unlikely(!desc->write)) {
6067 0 : return -EBADF;
6068 : }
6069 :
6070 70 : if (spdk_unlikely(!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks))) {
6071 0 : return -EINVAL;
6072 : }
6073 :
6074 70 : bdev_io = bdev_channel_get_io(channel);
6075 70 : if (spdk_unlikely(!bdev_io)) {
6076 2 : return -ENOMEM;
6077 : }
6078 :
6079 68 : bdev_io->internal.ch = channel;
6080 68 : bdev_io->internal.desc = desc;
6081 68 : bdev_io->type = SPDK_BDEV_IO_TYPE_WRITE;
6082 68 : bdev_io->u.bdev.iovs = iov;
6083 68 : bdev_io->u.bdev.iovcnt = iovcnt;
6084 68 : bdev_io->u.bdev.md_buf = md_buf;
6085 68 : bdev_io->u.bdev.num_blocks = num_blocks;
6086 68 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6087 68 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6088 68 : if (seq != NULL) {
6089 0 : bdev_io->internal.f.has_accel_sequence = true;
6090 0 : bdev_io->internal.accel_sequence = seq;
6091 0 : }
6092 :
6093 68 : if (domain != NULL) {
6094 2 : bdev_io->internal.f.has_memory_domain = true;
6095 2 : bdev_io->internal.memory_domain = domain;
6096 2 : bdev_io->internal.memory_domain_ctx = domain_ctx;
6097 2 : }
6098 :
6099 68 : bdev_io->u.bdev.memory_domain = domain;
6100 68 : bdev_io->u.bdev.memory_domain_ctx = domain_ctx;
6101 68 : bdev_io->u.bdev.accel_sequence = seq;
6102 68 : bdev_io->u.bdev.dif_check_flags = dif_check_flags;
6103 68 : bdev_io->u.bdev.nvme_cdw12.raw = nvme_cdw12_raw;
6104 68 : bdev_io->u.bdev.nvme_cdw13.raw = nvme_cdw13_raw;
6105 :
6106 68 : _bdev_io_submit_ext(desc, bdev_io);
6107 :
6108 68 : return 0;
6109 70 : }
6110 :
6111 : int
6112 3 : spdk_bdev_writev(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6113 : struct iovec *iov, int iovcnt,
6114 : uint64_t offset, uint64_t len,
6115 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6116 : {
6117 : uint64_t offset_blocks, num_blocks;
6118 :
6119 3 : if (bdev_bytes_to_blocks(desc, offset, &offset_blocks, len, &num_blocks) != 0) {
6120 0 : return -EINVAL;
6121 : }
6122 :
6123 3 : return spdk_bdev_writev_blocks(desc, ch, iov, iovcnt, offset_blocks, num_blocks, cb, cb_arg);
6124 3 : }
6125 :
6126 : int
6127 14 : spdk_bdev_writev_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6128 : struct iovec *iov, int iovcnt,
6129 : uint64_t offset_blocks, uint64_t num_blocks,
6130 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6131 : {
6132 14 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6133 :
6134 28 : return bdev_writev_blocks_with_md(desc, ch, iov, iovcnt, NULL, offset_blocks,
6135 14 : num_blocks, NULL, NULL, NULL, bdev->dif_check_flags, 0, 0,
6136 14 : cb, cb_arg);
6137 : }
6138 :
6139 : int
6140 1 : spdk_bdev_writev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6141 : struct iovec *iov, int iovcnt, void *md_buf,
6142 : uint64_t offset_blocks, uint64_t num_blocks,
6143 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6144 : {
6145 1 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6146 :
6147 1 : if (md_buf && !spdk_bdev_is_md_separate(bdev)) {
6148 0 : return -EINVAL;
6149 : }
6150 :
6151 1 : if (md_buf && !_is_buf_allocated(iov)) {
6152 0 : return -EINVAL;
6153 : }
6154 :
6155 2 : return bdev_writev_blocks_with_md(desc, ch, iov, iovcnt, md_buf, offset_blocks,
6156 1 : num_blocks, NULL, NULL, NULL, bdev->dif_check_flags, 0, 0,
6157 1 : cb, cb_arg);
6158 1 : }
6159 :
6160 : int
6161 8 : spdk_bdev_writev_blocks_ext(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6162 : struct iovec *iov, int iovcnt,
6163 : uint64_t offset_blocks, uint64_t num_blocks,
6164 : spdk_bdev_io_completion_cb cb, void *cb_arg,
6165 : struct spdk_bdev_ext_io_opts *opts)
6166 : {
6167 8 : struct spdk_memory_domain *domain = NULL;
6168 8 : struct spdk_accel_sequence *seq = NULL;
6169 8 : void *domain_ctx = NULL, *md = NULL;
6170 8 : uint32_t dif_check_flags = 0;
6171 8 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6172 8 : uint32_t nvme_cdw12_raw = 0;
6173 8 : uint32_t nvme_cdw13_raw = 0;
6174 :
6175 8 : if (opts) {
6176 7 : if (spdk_unlikely(!_bdev_io_check_opts(opts, iov))) {
6177 3 : return -EINVAL;
6178 : }
6179 4 : md = opts->metadata;
6180 4 : domain = bdev_get_ext_io_opt(opts, memory_domain, NULL);
6181 4 : domain_ctx = bdev_get_ext_io_opt(opts, memory_domain_ctx, NULL);
6182 4 : seq = bdev_get_ext_io_opt(opts, accel_sequence, NULL);
6183 4 : nvme_cdw12_raw = bdev_get_ext_io_opt(opts, nvme_cdw12.raw, 0);
6184 4 : nvme_cdw13_raw = bdev_get_ext_io_opt(opts, nvme_cdw13.raw, 0);
6185 4 : if (md) {
6186 4 : if (spdk_unlikely(!spdk_bdev_is_md_separate(bdev))) {
6187 0 : return -EINVAL;
6188 : }
6189 :
6190 4 : if (spdk_unlikely(!_is_buf_allocated(iov))) {
6191 0 : return -EINVAL;
6192 : }
6193 :
6194 4 : if (spdk_unlikely(seq != NULL)) {
6195 0 : return -EINVAL;
6196 : }
6197 :
6198 4 : if (nvme_cdw12_raw & SPDK_DIF_FLAGS_NVME_PRACT) {
6199 0 : SPDK_ERRLOG("Separate metadata with NVMe PRACT is not supported.\n");
6200 0 : return -ENOTSUP;
6201 : }
6202 4 : }
6203 :
6204 4 : if (nvme_cdw12_raw & SPDK_DIF_FLAGS_NVME_PRACT) {
6205 0 : dif_check_flags |= SPDK_DIF_FLAGS_NVME_PRACT;
6206 0 : }
6207 4 : }
6208 :
6209 10 : dif_check_flags |= bdev->dif_check_flags &
6210 5 : ~(bdev_get_ext_io_opt(opts, dif_check_flags_exclude_mask, 0));
6211 :
6212 10 : return bdev_writev_blocks_with_md(desc, ch, iov, iovcnt, md, offset_blocks, num_blocks,
6213 5 : domain, domain_ctx, seq, dif_check_flags,
6214 5 : nvme_cdw12_raw, nvme_cdw13_raw, cb, cb_arg);
6215 8 : }
6216 :
6217 : static void
6218 11 : bdev_compare_do_read_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
6219 : {
6220 11 : struct spdk_bdev_io *parent_io = cb_arg;
6221 11 : struct spdk_bdev *bdev = parent_io->bdev;
6222 11 : uint8_t *read_buf = bdev_io->u.bdev.iovs[0].iov_base;
6223 11 : int i, rc = 0;
6224 :
6225 11 : if (!success) {
6226 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
6227 0 : parent_io->internal.cb(parent_io, false, parent_io->internal.caller_ctx);
6228 0 : spdk_bdev_free_io(bdev_io);
6229 0 : return;
6230 : }
6231 :
6232 17 : for (i = 0; i < parent_io->u.bdev.iovcnt; i++) {
6233 22 : rc = memcmp(read_buf,
6234 11 : parent_io->u.bdev.iovs[i].iov_base,
6235 11 : parent_io->u.bdev.iovs[i].iov_len);
6236 11 : if (rc) {
6237 5 : break;
6238 : }
6239 6 : read_buf += parent_io->u.bdev.iovs[i].iov_len;
6240 6 : }
6241 :
6242 11 : if (rc == 0 && parent_io->u.bdev.md_buf && spdk_bdev_is_md_separate(bdev)) {
6243 4 : rc = memcmp(bdev_io->u.bdev.md_buf,
6244 2 : parent_io->u.bdev.md_buf,
6245 2 : spdk_bdev_get_md_size(bdev));
6246 2 : }
6247 :
6248 11 : spdk_bdev_free_io(bdev_io);
6249 :
6250 11 : if (rc == 0) {
6251 5 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
6252 5 : parent_io->internal.cb(parent_io, true, parent_io->internal.caller_ctx);
6253 5 : } else {
6254 6 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_MISCOMPARE;
6255 6 : parent_io->internal.cb(parent_io, false, parent_io->internal.caller_ctx);
6256 : }
6257 11 : }
6258 :
6259 : static void
6260 11 : bdev_compare_do_read(void *_bdev_io)
6261 : {
6262 11 : struct spdk_bdev_io *bdev_io = _bdev_io;
6263 : int rc;
6264 :
6265 22 : rc = spdk_bdev_read_blocks(bdev_io->internal.desc,
6266 11 : spdk_io_channel_from_ctx(bdev_io->internal.ch), NULL,
6267 11 : bdev_io->u.bdev.offset_blocks, bdev_io->u.bdev.num_blocks,
6268 11 : bdev_compare_do_read_done, bdev_io);
6269 :
6270 11 : if (rc == -ENOMEM) {
6271 0 : bdev_queue_io_wait_with_cb(bdev_io, bdev_compare_do_read);
6272 11 : } else if (rc != 0) {
6273 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
6274 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
6275 0 : }
6276 11 : }
6277 :
6278 : static int
6279 16 : bdev_comparev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6280 : struct iovec *iov, int iovcnt, void *md_buf,
6281 : uint64_t offset_blocks, uint64_t num_blocks,
6282 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6283 : {
6284 16 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6285 : struct spdk_bdev_io *bdev_io;
6286 16 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6287 :
6288 16 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6289 0 : return -EINVAL;
6290 : }
6291 :
6292 16 : bdev_io = bdev_channel_get_io(channel);
6293 16 : if (!bdev_io) {
6294 0 : return -ENOMEM;
6295 : }
6296 :
6297 16 : bdev_io->internal.ch = channel;
6298 16 : bdev_io->internal.desc = desc;
6299 16 : bdev_io->type = SPDK_BDEV_IO_TYPE_COMPARE;
6300 16 : bdev_io->u.bdev.iovs = iov;
6301 16 : bdev_io->u.bdev.iovcnt = iovcnt;
6302 16 : bdev_io->u.bdev.md_buf = md_buf;
6303 16 : bdev_io->u.bdev.num_blocks = num_blocks;
6304 16 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6305 16 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6306 16 : bdev_io->u.bdev.memory_domain = NULL;
6307 16 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6308 16 : bdev_io->u.bdev.accel_sequence = NULL;
6309 :
6310 16 : if (bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COMPARE)) {
6311 7 : bdev_io_submit(bdev_io);
6312 7 : return 0;
6313 : }
6314 :
6315 9 : bdev_compare_do_read(bdev_io);
6316 :
6317 9 : return 0;
6318 16 : }
6319 :
6320 : int
6321 10 : spdk_bdev_comparev_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6322 : struct iovec *iov, int iovcnt,
6323 : uint64_t offset_blocks, uint64_t num_blocks,
6324 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6325 : {
6326 20 : return bdev_comparev_blocks_with_md(desc, ch, iov, iovcnt, NULL, offset_blocks,
6327 10 : num_blocks, cb, cb_arg);
6328 : }
6329 :
6330 : int
6331 6 : spdk_bdev_comparev_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6332 : struct iovec *iov, int iovcnt, void *md_buf,
6333 : uint64_t offset_blocks, uint64_t num_blocks,
6334 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6335 : {
6336 6 : if (md_buf && !spdk_bdev_is_md_separate(spdk_bdev_desc_get_bdev(desc))) {
6337 0 : return -EINVAL;
6338 : }
6339 :
6340 6 : if (md_buf && !_is_buf_allocated(iov)) {
6341 0 : return -EINVAL;
6342 : }
6343 :
6344 12 : return bdev_comparev_blocks_with_md(desc, ch, iov, iovcnt, md_buf, offset_blocks,
6345 6 : num_blocks, cb, cb_arg);
6346 6 : }
6347 :
6348 : static int
6349 4 : bdev_compare_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6350 : void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
6351 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6352 : {
6353 4 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6354 : struct spdk_bdev_io *bdev_io;
6355 4 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6356 :
6357 4 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6358 0 : return -EINVAL;
6359 : }
6360 :
6361 4 : bdev_io = bdev_channel_get_io(channel);
6362 4 : if (!bdev_io) {
6363 0 : return -ENOMEM;
6364 : }
6365 :
6366 4 : bdev_io->internal.ch = channel;
6367 4 : bdev_io->internal.desc = desc;
6368 4 : bdev_io->type = SPDK_BDEV_IO_TYPE_COMPARE;
6369 4 : bdev_io->u.bdev.iovs = &bdev_io->iov;
6370 4 : bdev_io->u.bdev.iovs[0].iov_base = buf;
6371 4 : bdev_io->u.bdev.iovs[0].iov_len = num_blocks * bdev_desc_get_block_size(desc);
6372 4 : bdev_io->u.bdev.iovcnt = 1;
6373 4 : bdev_io->u.bdev.md_buf = md_buf;
6374 4 : bdev_io->u.bdev.num_blocks = num_blocks;
6375 4 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6376 4 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6377 4 : bdev_io->u.bdev.memory_domain = NULL;
6378 4 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6379 4 : bdev_io->u.bdev.accel_sequence = NULL;
6380 :
6381 4 : if (bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COMPARE)) {
6382 2 : bdev_io_submit(bdev_io);
6383 2 : return 0;
6384 : }
6385 :
6386 2 : bdev_compare_do_read(bdev_io);
6387 :
6388 2 : return 0;
6389 4 : }
6390 :
6391 : int
6392 4 : spdk_bdev_compare_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6393 : void *buf, uint64_t offset_blocks, uint64_t num_blocks,
6394 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6395 : {
6396 8 : return bdev_compare_blocks_with_md(desc, ch, buf, NULL, offset_blocks, num_blocks,
6397 4 : cb, cb_arg);
6398 : }
6399 :
6400 : int
6401 0 : spdk_bdev_compare_blocks_with_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6402 : void *buf, void *md_buf, uint64_t offset_blocks, uint64_t num_blocks,
6403 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6404 : {
6405 0 : struct iovec iov = {
6406 0 : .iov_base = buf,
6407 : };
6408 :
6409 0 : if (md_buf && !spdk_bdev_is_md_separate(spdk_bdev_desc_get_bdev(desc))) {
6410 0 : return -EINVAL;
6411 : }
6412 :
6413 0 : if (md_buf && !_is_buf_allocated(&iov)) {
6414 0 : return -EINVAL;
6415 : }
6416 :
6417 0 : return bdev_compare_blocks_with_md(desc, ch, buf, md_buf, offset_blocks, num_blocks,
6418 0 : cb, cb_arg);
6419 0 : }
6420 :
6421 : static void
6422 2 : bdev_comparev_and_writev_blocks_unlocked(struct lba_range *range, void *ctx, int unlock_status)
6423 : {
6424 2 : struct spdk_bdev_io *bdev_io = ctx;
6425 :
6426 2 : if (unlock_status) {
6427 0 : SPDK_ERRLOG("LBA range unlock failed\n");
6428 0 : }
6429 :
6430 4 : bdev_io->internal.cb(bdev_io, bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS ? true :
6431 2 : false, bdev_io->internal.caller_ctx);
6432 2 : }
6433 :
6434 : static void
6435 2 : bdev_comparev_and_writev_blocks_unlock(struct spdk_bdev_io *bdev_io, int status)
6436 : {
6437 2 : bdev_io->internal.status = status;
6438 :
6439 4 : bdev_unlock_lba_range(bdev_io->internal.desc, spdk_io_channel_from_ctx(bdev_io->internal.ch),
6440 2 : bdev_io->u.bdev.offset_blocks, bdev_io->u.bdev.num_blocks,
6441 2 : bdev_comparev_and_writev_blocks_unlocked, bdev_io);
6442 2 : }
6443 :
6444 : static void
6445 1 : bdev_compare_and_write_do_write_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
6446 : {
6447 1 : struct spdk_bdev_io *parent_io = cb_arg;
6448 :
6449 1 : if (!success) {
6450 0 : SPDK_ERRLOG("Compare and write operation failed\n");
6451 0 : }
6452 :
6453 1 : spdk_bdev_free_io(bdev_io);
6454 :
6455 2 : bdev_comparev_and_writev_blocks_unlock(parent_io,
6456 1 : success ? SPDK_BDEV_IO_STATUS_SUCCESS : SPDK_BDEV_IO_STATUS_FAILED);
6457 1 : }
6458 :
6459 : static void
6460 1 : bdev_compare_and_write_do_write(void *_bdev_io)
6461 : {
6462 1 : struct spdk_bdev_io *bdev_io = _bdev_io;
6463 : int rc;
6464 :
6465 2 : rc = spdk_bdev_writev_blocks(bdev_io->internal.desc,
6466 1 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
6467 1 : bdev_io->u.bdev.fused_iovs, bdev_io->u.bdev.fused_iovcnt,
6468 1 : bdev_io->u.bdev.offset_blocks, bdev_io->u.bdev.num_blocks,
6469 1 : bdev_compare_and_write_do_write_done, bdev_io);
6470 :
6471 :
6472 1 : if (rc == -ENOMEM) {
6473 0 : bdev_queue_io_wait_with_cb(bdev_io, bdev_compare_and_write_do_write);
6474 1 : } else if (rc != 0) {
6475 0 : bdev_comparev_and_writev_blocks_unlock(bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
6476 0 : }
6477 1 : }
6478 :
6479 : static void
6480 2 : bdev_compare_and_write_do_compare_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
6481 : {
6482 2 : struct spdk_bdev_io *parent_io = cb_arg;
6483 :
6484 2 : spdk_bdev_free_io(bdev_io);
6485 :
6486 2 : if (!success) {
6487 1 : bdev_comparev_and_writev_blocks_unlock(parent_io, SPDK_BDEV_IO_STATUS_MISCOMPARE);
6488 1 : return;
6489 : }
6490 :
6491 1 : bdev_compare_and_write_do_write(parent_io);
6492 2 : }
6493 :
6494 : static void
6495 2 : bdev_compare_and_write_do_compare(void *_bdev_io)
6496 : {
6497 2 : struct spdk_bdev_io *bdev_io = _bdev_io;
6498 : int rc;
6499 :
6500 4 : rc = spdk_bdev_comparev_blocks(bdev_io->internal.desc,
6501 2 : spdk_io_channel_from_ctx(bdev_io->internal.ch), bdev_io->u.bdev.iovs,
6502 2 : bdev_io->u.bdev.iovcnt, bdev_io->u.bdev.offset_blocks, bdev_io->u.bdev.num_blocks,
6503 2 : bdev_compare_and_write_do_compare_done, bdev_io);
6504 :
6505 2 : if (rc == -ENOMEM) {
6506 0 : bdev_queue_io_wait_with_cb(bdev_io, bdev_compare_and_write_do_compare);
6507 2 : } else if (rc != 0) {
6508 0 : bdev_comparev_and_writev_blocks_unlock(bdev_io, SPDK_BDEV_IO_STATUS_FIRST_FUSED_FAILED);
6509 0 : }
6510 2 : }
6511 :
6512 : static void
6513 2 : bdev_comparev_and_writev_blocks_locked(struct lba_range *range, void *ctx, int status)
6514 : {
6515 2 : struct spdk_bdev_io *bdev_io = ctx;
6516 :
6517 2 : if (status) {
6518 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FIRST_FUSED_FAILED;
6519 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
6520 0 : return;
6521 : }
6522 :
6523 2 : bdev_compare_and_write_do_compare(bdev_io);
6524 2 : }
6525 :
6526 : int
6527 2 : spdk_bdev_comparev_and_writev_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6528 : struct iovec *compare_iov, int compare_iovcnt,
6529 : struct iovec *write_iov, int write_iovcnt,
6530 : uint64_t offset_blocks, uint64_t num_blocks,
6531 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6532 : {
6533 2 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6534 : struct spdk_bdev_io *bdev_io;
6535 2 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6536 :
6537 2 : if (!desc->write) {
6538 0 : return -EBADF;
6539 : }
6540 :
6541 2 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6542 0 : return -EINVAL;
6543 : }
6544 :
6545 2 : if (num_blocks > bdev->acwu) {
6546 0 : return -EINVAL;
6547 : }
6548 :
6549 2 : bdev_io = bdev_channel_get_io(channel);
6550 2 : if (!bdev_io) {
6551 0 : return -ENOMEM;
6552 : }
6553 :
6554 2 : bdev_io->internal.ch = channel;
6555 2 : bdev_io->internal.desc = desc;
6556 2 : bdev_io->type = SPDK_BDEV_IO_TYPE_COMPARE_AND_WRITE;
6557 2 : bdev_io->u.bdev.iovs = compare_iov;
6558 2 : bdev_io->u.bdev.iovcnt = compare_iovcnt;
6559 2 : bdev_io->u.bdev.fused_iovs = write_iov;
6560 2 : bdev_io->u.bdev.fused_iovcnt = write_iovcnt;
6561 2 : bdev_io->u.bdev.md_buf = NULL;
6562 2 : bdev_io->u.bdev.num_blocks = num_blocks;
6563 2 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6564 2 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6565 2 : bdev_io->u.bdev.memory_domain = NULL;
6566 2 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6567 2 : bdev_io->u.bdev.accel_sequence = NULL;
6568 :
6569 2 : if (bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COMPARE_AND_WRITE)) {
6570 0 : bdev_io_submit(bdev_io);
6571 0 : return 0;
6572 : }
6573 :
6574 4 : return bdev_lock_lba_range(desc, ch, offset_blocks, num_blocks,
6575 2 : bdev_comparev_and_writev_blocks_locked, bdev_io);
6576 2 : }
6577 :
6578 : int
6579 2 : spdk_bdev_zcopy_start(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6580 : struct iovec *iov, int iovcnt,
6581 : uint64_t offset_blocks, uint64_t num_blocks,
6582 : bool populate,
6583 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6584 : {
6585 2 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6586 : struct spdk_bdev_io *bdev_io;
6587 2 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6588 :
6589 2 : if (!desc->write) {
6590 0 : return -EBADF;
6591 : }
6592 :
6593 2 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6594 0 : return -EINVAL;
6595 : }
6596 :
6597 2 : if (!spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_ZCOPY)) {
6598 0 : return -ENOTSUP;
6599 : }
6600 :
6601 2 : bdev_io = bdev_channel_get_io(channel);
6602 2 : if (!bdev_io) {
6603 0 : return -ENOMEM;
6604 : }
6605 :
6606 2 : bdev_io->internal.ch = channel;
6607 2 : bdev_io->internal.desc = desc;
6608 2 : bdev_io->type = SPDK_BDEV_IO_TYPE_ZCOPY;
6609 2 : bdev_io->u.bdev.num_blocks = num_blocks;
6610 2 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6611 2 : bdev_io->u.bdev.iovs = iov;
6612 2 : bdev_io->u.bdev.iovcnt = iovcnt;
6613 2 : bdev_io->u.bdev.md_buf = NULL;
6614 2 : bdev_io->u.bdev.zcopy.populate = populate ? 1 : 0;
6615 2 : bdev_io->u.bdev.zcopy.commit = 0;
6616 2 : bdev_io->u.bdev.zcopy.start = 1;
6617 2 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6618 2 : bdev_io->u.bdev.memory_domain = NULL;
6619 2 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6620 2 : bdev_io->u.bdev.accel_sequence = NULL;
6621 :
6622 2 : bdev_io_submit(bdev_io);
6623 :
6624 2 : return 0;
6625 2 : }
6626 :
6627 : int
6628 2 : spdk_bdev_zcopy_end(struct spdk_bdev_io *bdev_io, bool commit,
6629 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6630 : {
6631 2 : if (bdev_io->type != SPDK_BDEV_IO_TYPE_ZCOPY) {
6632 0 : return -EINVAL;
6633 : }
6634 :
6635 2 : bdev_io->u.bdev.zcopy.commit = commit ? 1 : 0;
6636 2 : bdev_io->u.bdev.zcopy.start = 0;
6637 2 : bdev_io->internal.caller_ctx = cb_arg;
6638 2 : bdev_io->internal.cb = cb;
6639 2 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_PENDING;
6640 :
6641 2 : bdev_io_submit(bdev_io);
6642 :
6643 2 : return 0;
6644 2 : }
6645 :
6646 : int
6647 0 : spdk_bdev_write_zeroes(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6648 : uint64_t offset, uint64_t len,
6649 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6650 : {
6651 : uint64_t offset_blocks, num_blocks;
6652 :
6653 0 : if (bdev_bytes_to_blocks(desc, offset, &offset_blocks, len, &num_blocks) != 0) {
6654 0 : return -EINVAL;
6655 : }
6656 :
6657 0 : return spdk_bdev_write_zeroes_blocks(desc, ch, offset_blocks, num_blocks, cb, cb_arg);
6658 0 : }
6659 :
6660 : int
6661 33 : spdk_bdev_write_zeroes_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6662 : uint64_t offset_blocks, uint64_t num_blocks,
6663 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6664 : {
6665 33 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6666 : struct spdk_bdev_io *bdev_io;
6667 33 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6668 :
6669 33 : if (!desc->write) {
6670 0 : return -EBADF;
6671 : }
6672 :
6673 33 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6674 0 : return -EINVAL;
6675 : }
6676 :
6677 33 : if (!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE_ZEROES) &&
6678 10 : !bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE)) {
6679 1 : return -ENOTSUP;
6680 : }
6681 :
6682 32 : bdev_io = bdev_channel_get_io(channel);
6683 :
6684 32 : if (!bdev_io) {
6685 0 : return -ENOMEM;
6686 : }
6687 :
6688 32 : bdev_io->type = SPDK_BDEV_IO_TYPE_WRITE_ZEROES;
6689 32 : bdev_io->internal.ch = channel;
6690 32 : bdev_io->internal.desc = desc;
6691 32 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6692 32 : bdev_io->u.bdev.num_blocks = num_blocks;
6693 32 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6694 32 : bdev_io->u.bdev.memory_domain = NULL;
6695 32 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6696 32 : bdev_io->u.bdev.accel_sequence = NULL;
6697 :
6698 : /* If the write_zeroes size is large and should be split, use the generic split
6699 : * logic regardless of whether SPDK_BDEV_IO_TYPE_WRITE_ZEREOS is supported or not.
6700 : *
6701 : * Then, send the write_zeroes request if SPDK_BDEV_IO_TYPE_WRITE_ZEROES is supported
6702 : * or emulate it using regular write request otherwise.
6703 : */
6704 32 : if (bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE_ZEROES) ||
6705 9 : bdev_io->internal.f.split) {
6706 26 : bdev_io_submit(bdev_io);
6707 26 : return 0;
6708 : }
6709 :
6710 6 : assert(_bdev_get_block_size_with_md(bdev) <= ZERO_BUFFER_SIZE);
6711 :
6712 6 : return bdev_write_zero_buffer(bdev_io);
6713 33 : }
6714 :
6715 : int
6716 0 : spdk_bdev_unmap(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6717 : uint64_t offset, uint64_t nbytes,
6718 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6719 : {
6720 : uint64_t offset_blocks, num_blocks;
6721 :
6722 0 : if (bdev_bytes_to_blocks(desc, offset, &offset_blocks, nbytes, &num_blocks) != 0) {
6723 0 : return -EINVAL;
6724 : }
6725 :
6726 0 : return spdk_bdev_unmap_blocks(desc, ch, offset_blocks, num_blocks, cb, cb_arg);
6727 0 : }
6728 :
6729 : static void
6730 0 : bdev_io_complete_cb(void *ctx)
6731 : {
6732 0 : struct spdk_bdev_io *bdev_io = ctx;
6733 :
6734 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
6735 0 : bdev_io->internal.cb(bdev_io, true, bdev_io->internal.caller_ctx);
6736 0 : }
6737 :
6738 : int
6739 22 : spdk_bdev_unmap_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6740 : uint64_t offset_blocks, uint64_t num_blocks,
6741 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6742 : {
6743 22 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6744 : struct spdk_bdev_io *bdev_io;
6745 22 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6746 :
6747 22 : if (!desc->write) {
6748 0 : return -EBADF;
6749 : }
6750 :
6751 22 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6752 0 : return -EINVAL;
6753 : }
6754 :
6755 22 : bdev_io = bdev_channel_get_io(channel);
6756 22 : if (!bdev_io) {
6757 0 : return -ENOMEM;
6758 : }
6759 :
6760 22 : bdev_io->internal.ch = channel;
6761 22 : bdev_io->internal.desc = desc;
6762 22 : bdev_io->type = SPDK_BDEV_IO_TYPE_UNMAP;
6763 :
6764 22 : bdev_io->u.bdev.iovs = &bdev_io->iov;
6765 22 : bdev_io->u.bdev.iovs[0].iov_base = NULL;
6766 22 : bdev_io->u.bdev.iovs[0].iov_len = 0;
6767 22 : bdev_io->u.bdev.iovcnt = 1;
6768 :
6769 22 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6770 22 : bdev_io->u.bdev.num_blocks = num_blocks;
6771 22 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6772 22 : bdev_io->u.bdev.memory_domain = NULL;
6773 22 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6774 22 : bdev_io->u.bdev.accel_sequence = NULL;
6775 :
6776 22 : if (num_blocks == 0) {
6777 0 : spdk_thread_send_msg(spdk_get_thread(), bdev_io_complete_cb, bdev_io);
6778 0 : return 0;
6779 : }
6780 :
6781 22 : bdev_io_submit(bdev_io);
6782 22 : return 0;
6783 22 : }
6784 :
6785 : int
6786 0 : spdk_bdev_flush(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6787 : uint64_t offset, uint64_t length,
6788 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6789 : {
6790 : uint64_t offset_blocks, num_blocks;
6791 :
6792 0 : if (bdev_bytes_to_blocks(desc, offset, &offset_blocks, length, &num_blocks) != 0) {
6793 0 : return -EINVAL;
6794 : }
6795 :
6796 0 : return spdk_bdev_flush_blocks(desc, ch, offset_blocks, num_blocks, cb, cb_arg);
6797 0 : }
6798 :
6799 : int
6800 2 : spdk_bdev_flush_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6801 : uint64_t offset_blocks, uint64_t num_blocks,
6802 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6803 : {
6804 2 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6805 : struct spdk_bdev_io *bdev_io;
6806 2 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6807 :
6808 2 : if (!desc->write) {
6809 0 : return -EBADF;
6810 : }
6811 :
6812 2 : if (spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_FLUSH))) {
6813 0 : return -ENOTSUP;
6814 : }
6815 :
6816 2 : if (!bdev_io_valid_blocks(bdev, offset_blocks, num_blocks)) {
6817 0 : return -EINVAL;
6818 : }
6819 :
6820 2 : bdev_io = bdev_channel_get_io(channel);
6821 2 : if (!bdev_io) {
6822 0 : return -ENOMEM;
6823 : }
6824 :
6825 2 : bdev_io->internal.ch = channel;
6826 2 : bdev_io->internal.desc = desc;
6827 2 : bdev_io->type = SPDK_BDEV_IO_TYPE_FLUSH;
6828 2 : bdev_io->u.bdev.iovs = NULL;
6829 2 : bdev_io->u.bdev.iovcnt = 0;
6830 2 : bdev_io->u.bdev.offset_blocks = offset_blocks;
6831 2 : bdev_io->u.bdev.num_blocks = num_blocks;
6832 2 : bdev_io->u.bdev.memory_domain = NULL;
6833 2 : bdev_io->u.bdev.memory_domain_ctx = NULL;
6834 2 : bdev_io->u.bdev.accel_sequence = NULL;
6835 2 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
6836 :
6837 2 : bdev_io_submit(bdev_io);
6838 2 : return 0;
6839 2 : }
6840 :
6841 : static int bdev_reset_poll_for_outstanding_io(void *ctx);
6842 :
6843 : static void
6844 13 : bdev_reset_check_outstanding_io_done(struct spdk_bdev *bdev, void *_ctx, int status)
6845 : {
6846 13 : struct spdk_bdev_io *bdev_io = _ctx;
6847 13 : struct spdk_bdev_channel *ch = bdev_io->internal.ch;
6848 :
6849 13 : if (status == -EBUSY) {
6850 9 : if (spdk_get_ticks() < bdev_io->u.reset.wait_poller.stop_time_tsc) {
6851 8 : bdev_io->u.reset.wait_poller.poller = SPDK_POLLER_REGISTER(bdev_reset_poll_for_outstanding_io,
6852 : bdev_io, BDEV_RESET_CHECK_OUTSTANDING_IO_PERIOD);
6853 8 : } else {
6854 1 : if (TAILQ_EMPTY(&ch->io_memory_domain) && TAILQ_EMPTY(&ch->io_accel_exec)) {
6855 : /* If outstanding IOs are still present and reset_io_drain_timeout
6856 : * seconds passed, start the reset. */
6857 1 : bdev_io_submit_reset(bdev_io);
6858 1 : } else {
6859 : /* We still have in progress memory domain pull/push or we're
6860 : * executing accel sequence. Since we cannot abort either of those
6861 : * operations, fail the reset request. */
6862 0 : spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED);
6863 : }
6864 : }
6865 9 : } else {
6866 4 : SPDK_DEBUGLOG(bdev,
6867 : "Skipping reset for underlying device of bdev: %s - no outstanding I/O.\n",
6868 : ch->bdev->name);
6869 : /* Mark the completion status as a SUCCESS and complete the reset. */
6870 4 : spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_SUCCESS);
6871 : }
6872 13 : }
6873 :
6874 : static void
6875 13 : bdev_reset_check_outstanding_io(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
6876 : struct spdk_io_channel *io_ch, void *_ctx)
6877 : {
6878 13 : struct spdk_bdev_channel *cur_ch = __io_ch_to_bdev_ch(io_ch);
6879 13 : int status = 0;
6880 :
6881 17 : if (cur_ch->io_outstanding > 0 ||
6882 4 : !TAILQ_EMPTY(&cur_ch->io_memory_domain) ||
6883 4 : !TAILQ_EMPTY(&cur_ch->io_accel_exec)) {
6884 : /* If a channel has outstanding IO, set status to -EBUSY code. This will stop
6885 : * further iteration over the rest of the channels and pass non-zero status
6886 : * to the callback function. */
6887 9 : status = -EBUSY;
6888 9 : }
6889 13 : spdk_bdev_for_each_channel_continue(i, status);
6890 13 : }
6891 :
6892 : static int
6893 8 : bdev_reset_poll_for_outstanding_io(void *ctx)
6894 : {
6895 8 : struct spdk_bdev_io *bdev_io = ctx;
6896 :
6897 8 : spdk_poller_unregister(&bdev_io->u.reset.wait_poller.poller);
6898 8 : spdk_bdev_for_each_channel(bdev_io->bdev, bdev_reset_check_outstanding_io, bdev_io,
6899 : bdev_reset_check_outstanding_io_done);
6900 :
6901 8 : return SPDK_POLLER_BUSY;
6902 : }
6903 :
6904 : static void
6905 16 : bdev_reset_freeze_channel_done(struct spdk_bdev *bdev, void *_ctx, int status)
6906 : {
6907 16 : struct spdk_bdev_io *bdev_io = _ctx;
6908 :
6909 16 : if (bdev->reset_io_drain_timeout == 0) {
6910 11 : bdev_io_submit_reset(bdev_io);
6911 11 : return;
6912 : }
6913 :
6914 10 : bdev_io->u.reset.wait_poller.stop_time_tsc = spdk_get_ticks() +
6915 5 : (bdev->reset_io_drain_timeout * spdk_get_ticks_hz());
6916 :
6917 : /* In case bdev->reset_io_drain_timeout is not equal to zero,
6918 : * submit the reset to the underlying module only if outstanding I/O
6919 : * remain after reset_io_drain_timeout seconds have passed. */
6920 5 : spdk_bdev_for_each_channel(bdev, bdev_reset_check_outstanding_io, bdev_io,
6921 : bdev_reset_check_outstanding_io_done);
6922 16 : }
6923 :
6924 : static void
6925 19 : bdev_reset_freeze_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
6926 : struct spdk_io_channel *ch, void *_ctx)
6927 : {
6928 : struct spdk_bdev_channel *channel;
6929 : struct spdk_bdev_mgmt_channel *mgmt_channel;
6930 : struct spdk_bdev_shared_resource *shared_resource;
6931 : bdev_io_tailq_t tmp_queued;
6932 :
6933 19 : TAILQ_INIT(&tmp_queued);
6934 :
6935 19 : channel = __io_ch_to_bdev_ch(ch);
6936 19 : shared_resource = channel->shared_resource;
6937 19 : mgmt_channel = shared_resource->mgmt_ch;
6938 :
6939 19 : channel->flags |= BDEV_CH_RESET_IN_PROGRESS;
6940 :
6941 19 : if ((channel->flags & BDEV_CH_QOS_ENABLED) != 0) {
6942 2 : TAILQ_SWAP(&channel->qos_queued_io, &tmp_queued, spdk_bdev_io, internal.link);
6943 2 : }
6944 :
6945 19 : bdev_abort_all_queued_io(&shared_resource->nomem_io, channel);
6946 19 : bdev_abort_all_buf_io(mgmt_channel, channel);
6947 19 : bdev_abort_all_queued_io(&tmp_queued, channel);
6948 :
6949 19 : spdk_bdev_for_each_channel_continue(i, 0);
6950 19 : }
6951 :
6952 : static void
6953 18 : bdev_start_reset(struct spdk_bdev_io *bdev_io)
6954 : {
6955 18 : struct spdk_bdev *bdev = bdev_io->bdev;
6956 18 : bool freeze_channel = false;
6957 :
6958 18 : bdev_ch_add_to_io_submitted(bdev_io);
6959 :
6960 : /**
6961 : * Take a channel reference for the target bdev for the life of this
6962 : * reset. This guards against the channel getting destroyed before
6963 : * the reset is completed. We will release the reference when this
6964 : * reset is completed.
6965 : */
6966 18 : bdev_io->u.reset.ch_ref = spdk_get_io_channel(__bdev_to_io_dev(bdev));
6967 :
6968 18 : spdk_spin_lock(&bdev->internal.spinlock);
6969 18 : if (bdev->internal.reset_in_progress == NULL) {
6970 16 : bdev->internal.reset_in_progress = bdev_io;
6971 16 : freeze_channel = true;
6972 16 : } else {
6973 2 : TAILQ_INSERT_TAIL(&bdev->internal.queued_resets, bdev_io, internal.link);
6974 : }
6975 18 : spdk_spin_unlock(&bdev->internal.spinlock);
6976 :
6977 18 : if (freeze_channel) {
6978 16 : spdk_bdev_for_each_channel(bdev, bdev_reset_freeze_channel, bdev_io,
6979 : bdev_reset_freeze_channel_done);
6980 16 : }
6981 18 : }
6982 :
6983 : int
6984 18 : spdk_bdev_reset(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
6985 : spdk_bdev_io_completion_cb cb, void *cb_arg)
6986 : {
6987 18 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
6988 : struct spdk_bdev_io *bdev_io;
6989 18 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
6990 :
6991 18 : bdev_io = bdev_channel_get_io(channel);
6992 18 : if (!bdev_io) {
6993 0 : return -ENOMEM;
6994 : }
6995 :
6996 18 : bdev_io->internal.ch = channel;
6997 18 : bdev_io->internal.desc = desc;
6998 18 : bdev_io->internal.submit_tsc = spdk_get_ticks();
6999 18 : bdev_io->type = SPDK_BDEV_IO_TYPE_RESET;
7000 18 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
7001 :
7002 18 : bdev_start_reset(bdev_io);
7003 18 : return 0;
7004 18 : }
7005 :
7006 : void
7007 0 : spdk_bdev_get_io_stat(struct spdk_bdev *bdev, struct spdk_io_channel *ch,
7008 : struct spdk_bdev_io_stat *stat, enum spdk_bdev_reset_stat_mode reset_mode)
7009 : {
7010 0 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
7011 :
7012 0 : bdev_get_io_stat(stat, channel->stat);
7013 0 : spdk_bdev_reset_io_stat(channel->stat, reset_mode);
7014 0 : }
7015 :
7016 : static void
7017 5 : bdev_get_device_stat_done(struct spdk_bdev *bdev, void *_ctx, int status)
7018 : {
7019 5 : struct spdk_bdev_iostat_ctx *bdev_iostat_ctx = _ctx;
7020 :
7021 10 : bdev_iostat_ctx->cb(bdev, bdev_iostat_ctx->stat,
7022 5 : bdev_iostat_ctx->cb_arg, 0);
7023 5 : free(bdev_iostat_ctx);
7024 5 : }
7025 :
7026 : static void
7027 4 : bdev_get_each_channel_stat(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
7028 : struct spdk_io_channel *ch, void *_ctx)
7029 : {
7030 4 : struct spdk_bdev_iostat_ctx *bdev_iostat_ctx = _ctx;
7031 4 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
7032 :
7033 4 : spdk_bdev_add_io_stat(bdev_iostat_ctx->stat, channel->stat);
7034 4 : spdk_bdev_reset_io_stat(channel->stat, bdev_iostat_ctx->reset_mode);
7035 4 : spdk_bdev_for_each_channel_continue(i, 0);
7036 4 : }
7037 :
7038 : void
7039 5 : spdk_bdev_get_device_stat(struct spdk_bdev *bdev, struct spdk_bdev_io_stat *stat,
7040 : enum spdk_bdev_reset_stat_mode reset_mode, spdk_bdev_get_device_stat_cb cb, void *cb_arg)
7041 : {
7042 : struct spdk_bdev_iostat_ctx *bdev_iostat_ctx;
7043 :
7044 5 : assert(bdev != NULL);
7045 5 : assert(stat != NULL);
7046 5 : assert(cb != NULL);
7047 :
7048 5 : bdev_iostat_ctx = calloc(1, sizeof(struct spdk_bdev_iostat_ctx));
7049 5 : if (bdev_iostat_ctx == NULL) {
7050 0 : SPDK_ERRLOG("Unable to allocate memory for spdk_bdev_iostat_ctx\n");
7051 0 : cb(bdev, stat, cb_arg, -ENOMEM);
7052 0 : return;
7053 : }
7054 :
7055 5 : bdev_iostat_ctx->stat = stat;
7056 5 : bdev_iostat_ctx->cb = cb;
7057 5 : bdev_iostat_ctx->cb_arg = cb_arg;
7058 5 : bdev_iostat_ctx->reset_mode = reset_mode;
7059 :
7060 : /* Start with the statistics from previously deleted channels. */
7061 5 : spdk_spin_lock(&bdev->internal.spinlock);
7062 5 : bdev_get_io_stat(bdev_iostat_ctx->stat, bdev->internal.stat);
7063 5 : spdk_bdev_reset_io_stat(bdev->internal.stat, reset_mode);
7064 5 : spdk_spin_unlock(&bdev->internal.spinlock);
7065 :
7066 : /* Then iterate and add the statistics from each existing channel. */
7067 5 : spdk_bdev_for_each_channel(bdev, bdev_get_each_channel_stat, bdev_iostat_ctx,
7068 : bdev_get_device_stat_done);
7069 5 : }
7070 :
7071 : struct bdev_iostat_reset_ctx {
7072 : enum spdk_bdev_reset_stat_mode mode;
7073 : bdev_reset_device_stat_cb cb;
7074 : void *cb_arg;
7075 : };
7076 :
7077 : static void
7078 0 : bdev_reset_device_stat_done(struct spdk_bdev *bdev, void *_ctx, int status)
7079 : {
7080 0 : struct bdev_iostat_reset_ctx *ctx = _ctx;
7081 :
7082 0 : ctx->cb(bdev, ctx->cb_arg, 0);
7083 :
7084 0 : free(ctx);
7085 0 : }
7086 :
7087 : static void
7088 0 : bdev_reset_each_channel_stat(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
7089 : struct spdk_io_channel *ch, void *_ctx)
7090 : {
7091 0 : struct bdev_iostat_reset_ctx *ctx = _ctx;
7092 0 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
7093 :
7094 0 : spdk_bdev_reset_io_stat(channel->stat, ctx->mode);
7095 :
7096 0 : spdk_bdev_for_each_channel_continue(i, 0);
7097 0 : }
7098 :
7099 : void
7100 0 : bdev_reset_device_stat(struct spdk_bdev *bdev, enum spdk_bdev_reset_stat_mode mode,
7101 : bdev_reset_device_stat_cb cb, void *cb_arg)
7102 : {
7103 : struct bdev_iostat_reset_ctx *ctx;
7104 :
7105 0 : assert(bdev != NULL);
7106 0 : assert(cb != NULL);
7107 :
7108 0 : ctx = calloc(1, sizeof(*ctx));
7109 0 : if (ctx == NULL) {
7110 0 : SPDK_ERRLOG("Unable to allocate bdev_iostat_reset_ctx.\n");
7111 0 : cb(bdev, cb_arg, -ENOMEM);
7112 0 : return;
7113 : }
7114 :
7115 0 : ctx->mode = mode;
7116 0 : ctx->cb = cb;
7117 0 : ctx->cb_arg = cb_arg;
7118 :
7119 0 : spdk_spin_lock(&bdev->internal.spinlock);
7120 0 : spdk_bdev_reset_io_stat(bdev->internal.stat, mode);
7121 0 : spdk_spin_unlock(&bdev->internal.spinlock);
7122 :
7123 0 : spdk_bdev_for_each_channel(bdev,
7124 : bdev_reset_each_channel_stat,
7125 0 : ctx,
7126 : bdev_reset_device_stat_done);
7127 0 : }
7128 :
7129 : int
7130 1 : spdk_bdev_nvme_admin_passthru(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
7131 : const struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes,
7132 : spdk_bdev_io_completion_cb cb, void *cb_arg)
7133 : {
7134 1 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
7135 : struct spdk_bdev_io *bdev_io;
7136 1 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
7137 :
7138 1 : if (!desc->write) {
7139 0 : return -EBADF;
7140 : }
7141 :
7142 1 : if (spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_ADMIN))) {
7143 1 : return -ENOTSUP;
7144 : }
7145 :
7146 0 : bdev_io = bdev_channel_get_io(channel);
7147 0 : if (!bdev_io) {
7148 0 : return -ENOMEM;
7149 : }
7150 :
7151 0 : bdev_io->internal.ch = channel;
7152 0 : bdev_io->internal.desc = desc;
7153 0 : bdev_io->type = SPDK_BDEV_IO_TYPE_NVME_ADMIN;
7154 0 : bdev_io->u.nvme_passthru.cmd = *cmd;
7155 0 : bdev_io->u.nvme_passthru.buf = buf;
7156 0 : bdev_io->u.nvme_passthru.nbytes = nbytes;
7157 0 : bdev_io->u.nvme_passthru.md_buf = NULL;
7158 0 : bdev_io->u.nvme_passthru.md_len = 0;
7159 :
7160 0 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
7161 :
7162 0 : bdev_io_submit(bdev_io);
7163 0 : return 0;
7164 1 : }
7165 :
7166 : int
7167 1 : spdk_bdev_nvme_io_passthru(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
7168 : const struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes,
7169 : spdk_bdev_io_completion_cb cb, void *cb_arg)
7170 : {
7171 1 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
7172 : struct spdk_bdev_io *bdev_io;
7173 1 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
7174 :
7175 1 : if (!desc->write) {
7176 : /*
7177 : * Do not try to parse the NVMe command - we could maybe use bits in the opcode
7178 : * to easily determine if the command is a read or write, but for now just
7179 : * do not allow io_passthru with a read-only descriptor.
7180 : */
7181 0 : return -EBADF;
7182 : }
7183 :
7184 1 : if (spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_IO))) {
7185 1 : return -ENOTSUP;
7186 : }
7187 :
7188 0 : bdev_io = bdev_channel_get_io(channel);
7189 0 : if (!bdev_io) {
7190 0 : return -ENOMEM;
7191 : }
7192 :
7193 0 : bdev_io->internal.ch = channel;
7194 0 : bdev_io->internal.desc = desc;
7195 0 : bdev_io->type = SPDK_BDEV_IO_TYPE_NVME_IO;
7196 0 : bdev_io->u.nvme_passthru.cmd = *cmd;
7197 0 : bdev_io->u.nvme_passthru.buf = buf;
7198 0 : bdev_io->u.nvme_passthru.nbytes = nbytes;
7199 0 : bdev_io->u.nvme_passthru.md_buf = NULL;
7200 0 : bdev_io->u.nvme_passthru.md_len = 0;
7201 :
7202 0 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
7203 :
7204 0 : bdev_io_submit(bdev_io);
7205 0 : return 0;
7206 1 : }
7207 :
7208 : int
7209 1 : spdk_bdev_nvme_io_passthru_md(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
7210 : const struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes, void *md_buf, size_t md_len,
7211 : spdk_bdev_io_completion_cb cb, void *cb_arg)
7212 : {
7213 1 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
7214 : struct spdk_bdev_io *bdev_io;
7215 1 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
7216 :
7217 1 : if (!desc->write) {
7218 : /*
7219 : * Do not try to parse the NVMe command - we could maybe use bits in the opcode
7220 : * to easily determine if the command is a read or write, but for now just
7221 : * do not allow io_passthru with a read-only descriptor.
7222 : */
7223 0 : return -EBADF;
7224 : }
7225 :
7226 1 : if (spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_IO_MD))) {
7227 1 : return -ENOTSUP;
7228 : }
7229 :
7230 0 : bdev_io = bdev_channel_get_io(channel);
7231 0 : if (!bdev_io) {
7232 0 : return -ENOMEM;
7233 : }
7234 :
7235 0 : bdev_io->internal.ch = channel;
7236 0 : bdev_io->internal.desc = desc;
7237 0 : bdev_io->type = SPDK_BDEV_IO_TYPE_NVME_IO_MD;
7238 0 : bdev_io->u.nvme_passthru.cmd = *cmd;
7239 0 : bdev_io->u.nvme_passthru.buf = buf;
7240 0 : bdev_io->u.nvme_passthru.nbytes = nbytes;
7241 0 : bdev_io->u.nvme_passthru.md_buf = md_buf;
7242 0 : bdev_io->u.nvme_passthru.md_len = md_len;
7243 :
7244 0 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
7245 :
7246 0 : bdev_io_submit(bdev_io);
7247 0 : return 0;
7248 1 : }
7249 :
7250 : int
7251 0 : spdk_bdev_nvme_iov_passthru_md(struct spdk_bdev_desc *desc,
7252 : struct spdk_io_channel *ch,
7253 : const struct spdk_nvme_cmd *cmd,
7254 : struct iovec *iov, int iovcnt, size_t nbytes,
7255 : void *md_buf, size_t md_len,
7256 : spdk_bdev_io_completion_cb cb, void *cb_arg)
7257 : {
7258 0 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
7259 : struct spdk_bdev_io *bdev_io;
7260 0 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
7261 :
7262 0 : if (!desc->write) {
7263 : /*
7264 : * Do not try to parse the NVMe command - we could maybe use bits in the opcode
7265 : * to easily determine if the command is a read or write, but for now just
7266 : * do not allow io_passthru with a read-only descriptor.
7267 : */
7268 0 : return -EBADF;
7269 : }
7270 :
7271 0 : if (md_buf && spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_IO_MD))) {
7272 0 : return -ENOTSUP;
7273 0 : } else if (spdk_unlikely(!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_IO))) {
7274 0 : return -ENOTSUP;
7275 : }
7276 :
7277 0 : bdev_io = bdev_channel_get_io(channel);
7278 0 : if (!bdev_io) {
7279 0 : return -ENOMEM;
7280 : }
7281 :
7282 0 : bdev_io->internal.ch = channel;
7283 0 : bdev_io->internal.desc = desc;
7284 0 : bdev_io->type = SPDK_BDEV_IO_TYPE_NVME_IOV_MD;
7285 0 : bdev_io->u.nvme_passthru.cmd = *cmd;
7286 0 : bdev_io->u.nvme_passthru.iovs = iov;
7287 0 : bdev_io->u.nvme_passthru.iovcnt = iovcnt;
7288 0 : bdev_io->u.nvme_passthru.nbytes = nbytes;
7289 0 : bdev_io->u.nvme_passthru.md_buf = md_buf;
7290 0 : bdev_io->u.nvme_passthru.md_len = md_len;
7291 :
7292 0 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
7293 :
7294 0 : bdev_io_submit(bdev_io);
7295 0 : return 0;
7296 0 : }
7297 :
7298 : static void bdev_abort_retry(void *ctx);
7299 : static void bdev_abort(struct spdk_bdev_io *parent_io);
7300 :
7301 : static void
7302 22 : bdev_abort_io_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
7303 : {
7304 22 : struct spdk_bdev_channel *channel = bdev_io->internal.ch;
7305 22 : struct spdk_bdev_io *parent_io = cb_arg;
7306 : struct spdk_bdev_io *bio_to_abort, *tmp_io;
7307 :
7308 22 : bio_to_abort = bdev_io->u.abort.bio_to_abort;
7309 :
7310 22 : spdk_bdev_free_io(bdev_io);
7311 :
7312 22 : if (!success) {
7313 : /* Check if the target I/O completed in the meantime. */
7314 2 : TAILQ_FOREACH(tmp_io, &channel->io_submitted, internal.ch_link) {
7315 1 : if (tmp_io == bio_to_abort) {
7316 0 : break;
7317 : }
7318 1 : }
7319 :
7320 : /* If the target I/O still exists, set the parent to failed. */
7321 1 : if (tmp_io != NULL) {
7322 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
7323 0 : }
7324 1 : }
7325 :
7326 22 : assert(parent_io->internal.f.split);
7327 :
7328 22 : parent_io->internal.split.outstanding--;
7329 22 : if (parent_io->internal.split.outstanding == 0) {
7330 16 : if (parent_io->internal.status == SPDK_BDEV_IO_STATUS_NOMEM) {
7331 0 : bdev_abort_retry(parent_io);
7332 0 : } else {
7333 16 : bdev_io_complete(parent_io);
7334 : }
7335 16 : }
7336 22 : }
7337 :
7338 : static int
7339 23 : bdev_abort_io(struct spdk_bdev_desc *desc, struct spdk_bdev_channel *channel,
7340 : struct spdk_bdev_io *bio_to_abort,
7341 : spdk_bdev_io_completion_cb cb, void *cb_arg)
7342 : {
7343 23 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
7344 : struct spdk_bdev_io *bdev_io;
7345 :
7346 23 : if (bio_to_abort->type == SPDK_BDEV_IO_TYPE_ABORT ||
7347 23 : bio_to_abort->type == SPDK_BDEV_IO_TYPE_RESET) {
7348 : /* TODO: Abort reset or abort request. */
7349 0 : return -ENOTSUP;
7350 : }
7351 :
7352 23 : bdev_io = bdev_channel_get_io(channel);
7353 23 : if (bdev_io == NULL) {
7354 1 : return -ENOMEM;
7355 : }
7356 :
7357 22 : bdev_io->internal.ch = channel;
7358 22 : bdev_io->internal.desc = desc;
7359 22 : bdev_io->type = SPDK_BDEV_IO_TYPE_ABORT;
7360 22 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
7361 :
7362 22 : if (bio_to_abort->internal.f.split) {
7363 6 : assert(bdev_io_should_split(bio_to_abort));
7364 6 : bdev_io->u.bdev.abort.bio_cb_arg = bio_to_abort;
7365 :
7366 : /* Parent abort request is not submitted directly, but to manage its
7367 : * execution add it to the submitted list here.
7368 : */
7369 6 : bdev_io->internal.submit_tsc = spdk_get_ticks();
7370 6 : bdev_ch_add_to_io_submitted(bdev_io);
7371 :
7372 6 : bdev_abort(bdev_io);
7373 :
7374 6 : return 0;
7375 : }
7376 :
7377 16 : bdev_io->u.abort.bio_to_abort = bio_to_abort;
7378 :
7379 : /* Submit the abort request to the underlying bdev module. */
7380 16 : bdev_io_submit(bdev_io);
7381 :
7382 16 : return 0;
7383 23 : }
7384 :
7385 : static bool
7386 46 : bdev_io_on_tailq(struct spdk_bdev_io *bdev_io, bdev_io_tailq_t *tailq)
7387 : {
7388 : struct spdk_bdev_io *iter;
7389 :
7390 46 : TAILQ_FOREACH(iter, tailq, internal.link) {
7391 0 : if (iter == bdev_io) {
7392 0 : return true;
7393 : }
7394 0 : }
7395 :
7396 46 : return false;
7397 46 : }
7398 :
7399 : static uint32_t
7400 18 : _bdev_abort(struct spdk_bdev_io *parent_io)
7401 : {
7402 18 : struct spdk_bdev_desc *desc = parent_io->internal.desc;
7403 18 : struct spdk_bdev_channel *channel = parent_io->internal.ch;
7404 : void *bio_cb_arg;
7405 : struct spdk_bdev_io *bio_to_abort;
7406 : uint32_t matched_ios;
7407 : int rc;
7408 :
7409 18 : bio_cb_arg = parent_io->u.bdev.abort.bio_cb_arg;
7410 :
7411 : /* matched_ios is returned and will be kept by the caller.
7412 : *
7413 : * This function will be used for two cases, 1) the same cb_arg is used for
7414 : * multiple I/Os, 2) a single large I/O is split into smaller ones.
7415 : * Incrementing split_outstanding directly here may confuse readers especially
7416 : * for the 1st case.
7417 : *
7418 : * Completion of I/O abort is processed after stack unwinding. Hence this trick
7419 : * works as expected.
7420 : */
7421 18 : matched_ios = 0;
7422 18 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_SUCCESS;
7423 :
7424 105 : TAILQ_FOREACH(bio_to_abort, &channel->io_submitted, internal.ch_link) {
7425 88 : if (bio_to_abort->internal.caller_ctx != bio_cb_arg) {
7426 65 : continue;
7427 : }
7428 :
7429 23 : if (bio_to_abort->internal.submit_tsc > parent_io->internal.submit_tsc) {
7430 : /* Any I/O which was submitted after this abort command should be excluded. */
7431 0 : continue;
7432 : }
7433 :
7434 : /* We can't abort a request that's being pushed/pulled or executed by accel */
7435 23 : if (bdev_io_on_tailq(bio_to_abort, &channel->io_accel_exec) ||
7436 23 : bdev_io_on_tailq(bio_to_abort, &channel->io_memory_domain)) {
7437 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
7438 0 : break;
7439 : }
7440 :
7441 23 : rc = bdev_abort_io(desc, channel, bio_to_abort, bdev_abort_io_done, parent_io);
7442 23 : if (rc != 0) {
7443 1 : if (rc == -ENOMEM) {
7444 1 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_NOMEM;
7445 1 : } else {
7446 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
7447 : }
7448 1 : break;
7449 : }
7450 22 : matched_ios++;
7451 22 : }
7452 :
7453 18 : return matched_ios;
7454 : }
7455 :
7456 : static void
7457 1 : bdev_abort_retry(void *ctx)
7458 : {
7459 1 : struct spdk_bdev_io *parent_io = ctx;
7460 : uint32_t matched_ios;
7461 :
7462 1 : matched_ios = _bdev_abort(parent_io);
7463 :
7464 1 : if (matched_ios == 0) {
7465 0 : if (parent_io->internal.status == SPDK_BDEV_IO_STATUS_NOMEM) {
7466 0 : bdev_queue_io_wait_with_cb(parent_io, bdev_abort_retry);
7467 0 : } else {
7468 : /* For retry, the case that no target I/O was found is success
7469 : * because it means target I/Os completed in the meantime.
7470 : */
7471 0 : bdev_io_complete(parent_io);
7472 : }
7473 0 : return;
7474 : }
7475 :
7476 : /* Use split_outstanding to manage the progress of aborting I/Os. */
7477 1 : parent_io->internal.f.split = true;
7478 1 : parent_io->internal.split.outstanding = matched_ios;
7479 1 : }
7480 :
7481 : static void
7482 17 : bdev_abort(struct spdk_bdev_io *parent_io)
7483 : {
7484 : uint32_t matched_ios;
7485 :
7486 17 : matched_ios = _bdev_abort(parent_io);
7487 :
7488 17 : if (matched_ios == 0) {
7489 2 : if (parent_io->internal.status == SPDK_BDEV_IO_STATUS_NOMEM) {
7490 1 : bdev_queue_io_wait_with_cb(parent_io, bdev_abort_retry);
7491 1 : } else {
7492 : /* The case the no target I/O was found is failure. */
7493 1 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
7494 1 : bdev_io_complete(parent_io);
7495 : }
7496 2 : return;
7497 : }
7498 :
7499 : /* Use split_outstanding to manage the progress of aborting I/Os. */
7500 15 : parent_io->internal.f.split = true;
7501 15 : parent_io->internal.split.outstanding = matched_ios;
7502 17 : }
7503 :
7504 : int
7505 12 : spdk_bdev_abort(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
7506 : void *bio_cb_arg,
7507 : spdk_bdev_io_completion_cb cb, void *cb_arg)
7508 : {
7509 12 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
7510 12 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
7511 : struct spdk_bdev_io *bdev_io;
7512 :
7513 12 : if (bio_cb_arg == NULL) {
7514 0 : return -EINVAL;
7515 : }
7516 :
7517 12 : if (!spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_ABORT)) {
7518 1 : return -ENOTSUP;
7519 : }
7520 :
7521 11 : bdev_io = bdev_channel_get_io(channel);
7522 11 : if (bdev_io == NULL) {
7523 0 : return -ENOMEM;
7524 : }
7525 :
7526 11 : bdev_io->internal.ch = channel;
7527 11 : bdev_io->internal.desc = desc;
7528 11 : bdev_io->internal.submit_tsc = spdk_get_ticks();
7529 11 : bdev_io->type = SPDK_BDEV_IO_TYPE_ABORT;
7530 11 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
7531 :
7532 11 : bdev_io->u.bdev.abort.bio_cb_arg = bio_cb_arg;
7533 :
7534 : /* Parent abort request is not submitted directly, but to manage its execution,
7535 : * add it to the submitted list here.
7536 : */
7537 11 : bdev_ch_add_to_io_submitted(bdev_io);
7538 :
7539 11 : bdev_abort(bdev_io);
7540 :
7541 11 : return 0;
7542 12 : }
7543 :
7544 : int
7545 4 : spdk_bdev_queue_io_wait(struct spdk_bdev *bdev, struct spdk_io_channel *ch,
7546 : struct spdk_bdev_io_wait_entry *entry)
7547 : {
7548 4 : struct spdk_bdev_channel *channel = __io_ch_to_bdev_ch(ch);
7549 4 : struct spdk_bdev_mgmt_channel *mgmt_ch = channel->shared_resource->mgmt_ch;
7550 :
7551 4 : if (bdev != entry->bdev) {
7552 0 : SPDK_ERRLOG("bdevs do not match\n");
7553 0 : return -EINVAL;
7554 : }
7555 :
7556 4 : if (mgmt_ch->per_thread_cache_count > 0) {
7557 0 : SPDK_ERRLOG("Cannot queue io_wait if spdk_bdev_io available in per-thread cache\n");
7558 0 : return -EINVAL;
7559 : }
7560 :
7561 4 : TAILQ_INSERT_TAIL(&mgmt_ch->io_wait_queue, entry, link);
7562 4 : return 0;
7563 4 : }
7564 :
7565 : static inline void
7566 612 : bdev_io_update_io_stat(struct spdk_bdev_io *bdev_io, uint64_t tsc_diff)
7567 : {
7568 612 : enum spdk_bdev_io_status io_status = bdev_io->internal.status;
7569 612 : struct spdk_bdev_io_stat *io_stat = bdev_io->internal.ch->stat;
7570 612 : uint64_t num_blocks = bdev_io->u.bdev.num_blocks;
7571 612 : uint32_t blocklen = bdev_io->bdev->blocklen;
7572 :
7573 612 : if (spdk_likely(io_status == SPDK_BDEV_IO_STATUS_SUCCESS)) {
7574 519 : switch (bdev_io->type) {
7575 : case SPDK_BDEV_IO_TYPE_READ:
7576 321 : io_stat->bytes_read += num_blocks * blocklen;
7577 321 : io_stat->num_read_ops++;
7578 321 : io_stat->read_latency_ticks += tsc_diff;
7579 321 : if (io_stat->max_read_latency_ticks < tsc_diff) {
7580 7 : io_stat->max_read_latency_ticks = tsc_diff;
7581 7 : }
7582 321 : if (io_stat->min_read_latency_ticks > tsc_diff) {
7583 42 : io_stat->min_read_latency_ticks = tsc_diff;
7584 42 : }
7585 321 : break;
7586 : case SPDK_BDEV_IO_TYPE_WRITE:
7587 75 : io_stat->bytes_written += num_blocks * blocklen;
7588 75 : io_stat->num_write_ops++;
7589 75 : io_stat->write_latency_ticks += tsc_diff;
7590 75 : if (io_stat->max_write_latency_ticks < tsc_diff) {
7591 4 : io_stat->max_write_latency_ticks = tsc_diff;
7592 4 : }
7593 75 : if (io_stat->min_write_latency_ticks > tsc_diff) {
7594 25 : io_stat->min_write_latency_ticks = tsc_diff;
7595 25 : }
7596 75 : break;
7597 : case SPDK_BDEV_IO_TYPE_UNMAP:
7598 20 : io_stat->bytes_unmapped += num_blocks * blocklen;
7599 20 : io_stat->num_unmap_ops++;
7600 20 : io_stat->unmap_latency_ticks += tsc_diff;
7601 20 : if (io_stat->max_unmap_latency_ticks < tsc_diff) {
7602 0 : io_stat->max_unmap_latency_ticks = tsc_diff;
7603 0 : }
7604 20 : if (io_stat->min_unmap_latency_ticks > tsc_diff) {
7605 3 : io_stat->min_unmap_latency_ticks = tsc_diff;
7606 3 : }
7607 20 : break;
7608 : case SPDK_BDEV_IO_TYPE_ZCOPY:
7609 : /* Track the data in the start phase only */
7610 4 : if (bdev_io->u.bdev.zcopy.start) {
7611 2 : if (bdev_io->u.bdev.zcopy.populate) {
7612 1 : io_stat->bytes_read += num_blocks * blocklen;
7613 1 : io_stat->num_read_ops++;
7614 1 : io_stat->read_latency_ticks += tsc_diff;
7615 1 : if (io_stat->max_read_latency_ticks < tsc_diff) {
7616 0 : io_stat->max_read_latency_ticks = tsc_diff;
7617 0 : }
7618 1 : if (io_stat->min_read_latency_ticks > tsc_diff) {
7619 1 : io_stat->min_read_latency_ticks = tsc_diff;
7620 1 : }
7621 1 : } else {
7622 1 : io_stat->bytes_written += num_blocks * blocklen;
7623 1 : io_stat->num_write_ops++;
7624 1 : io_stat->write_latency_ticks += tsc_diff;
7625 1 : if (io_stat->max_write_latency_ticks < tsc_diff) {
7626 0 : io_stat->max_write_latency_ticks = tsc_diff;
7627 0 : }
7628 1 : if (io_stat->min_write_latency_ticks > tsc_diff) {
7629 1 : io_stat->min_write_latency_ticks = tsc_diff;
7630 1 : }
7631 : }
7632 2 : }
7633 4 : break;
7634 : case SPDK_BDEV_IO_TYPE_COPY:
7635 21 : io_stat->bytes_copied += num_blocks * blocklen;
7636 21 : io_stat->num_copy_ops++;
7637 21 : bdev_io->internal.ch->stat->copy_latency_ticks += tsc_diff;
7638 21 : if (io_stat->max_copy_latency_ticks < tsc_diff) {
7639 0 : io_stat->max_copy_latency_ticks = tsc_diff;
7640 0 : }
7641 21 : if (io_stat->min_copy_latency_ticks > tsc_diff) {
7642 4 : io_stat->min_copy_latency_ticks = tsc_diff;
7643 4 : }
7644 21 : break;
7645 : default:
7646 78 : break;
7647 : }
7648 612 : } else if (io_status <= SPDK_BDEV_IO_STATUS_FAILED && io_status >= SPDK_MIN_BDEV_IO_STATUS) {
7649 93 : io_stat = bdev_io->bdev->internal.stat;
7650 93 : assert(io_stat->io_error != NULL);
7651 :
7652 93 : spdk_spin_lock(&bdev_io->bdev->internal.spinlock);
7653 93 : io_stat->io_error->error_status[-io_status - 1]++;
7654 93 : spdk_spin_unlock(&bdev_io->bdev->internal.spinlock);
7655 93 : }
7656 :
7657 : #ifdef SPDK_CONFIG_VTUNE
7658 : uint64_t now_tsc = spdk_get_ticks();
7659 : if (now_tsc > (bdev_io->internal.ch->start_tsc + bdev_io->internal.ch->interval_tsc)) {
7660 : uint64_t data[5];
7661 : struct spdk_bdev_io_stat *prev_stat = bdev_io->internal.ch->prev_stat;
7662 :
7663 : data[0] = io_stat->num_read_ops - prev_stat->num_read_ops;
7664 : data[1] = io_stat->bytes_read - prev_stat->bytes_read;
7665 : data[2] = io_stat->num_write_ops - prev_stat->num_write_ops;
7666 : data[3] = io_stat->bytes_written - prev_stat->bytes_written;
7667 : data[4] = bdev_io->bdev->fn_table->get_spin_time ?
7668 : bdev_io->bdev->fn_table->get_spin_time(spdk_bdev_io_get_io_channel(bdev_io)) : 0;
7669 :
7670 : __itt_metadata_add(g_bdev_mgr.domain, __itt_null, bdev_io->internal.ch->handle,
7671 : __itt_metadata_u64, 5, data);
7672 :
7673 : memcpy(prev_stat, io_stat, sizeof(struct spdk_bdev_io_stat));
7674 : bdev_io->internal.ch->start_tsc = now_tsc;
7675 : }
7676 : #endif
7677 612 : }
7678 :
7679 : static inline void
7680 612 : _bdev_io_complete(void *ctx)
7681 : {
7682 612 : struct spdk_bdev_io *bdev_io = ctx;
7683 :
7684 612 : if (spdk_unlikely(bdev_io_use_accel_sequence(bdev_io))) {
7685 0 : assert(bdev_io->internal.status != SPDK_BDEV_IO_STATUS_SUCCESS);
7686 0 : spdk_accel_sequence_abort(bdev_io->internal.accel_sequence);
7687 0 : }
7688 :
7689 612 : assert(bdev_io->internal.cb != NULL);
7690 612 : assert(spdk_get_thread() == spdk_bdev_io_get_thread(bdev_io));
7691 :
7692 1224 : bdev_io->internal.cb(bdev_io, bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS,
7693 612 : bdev_io->internal.caller_ctx);
7694 612 : }
7695 :
7696 : static inline void
7697 620 : bdev_io_complete(void *ctx)
7698 : {
7699 620 : struct spdk_bdev_io *bdev_io = ctx;
7700 620 : struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
7701 : uint64_t tsc, tsc_diff;
7702 :
7703 620 : if (spdk_unlikely(bdev_io->internal.f.in_submit_request)) {
7704 : /*
7705 : * Defer completion to avoid potential infinite recursion if the
7706 : * user's completion callback issues a new I/O.
7707 : */
7708 16 : spdk_thread_send_msg(spdk_bdev_io_get_thread(bdev_io),
7709 8 : bdev_io_complete, bdev_io);
7710 8 : return;
7711 : }
7712 :
7713 612 : tsc = spdk_get_ticks();
7714 612 : tsc_diff = tsc - bdev_io->internal.submit_tsc;
7715 :
7716 612 : bdev_ch_remove_from_io_submitted(bdev_io);
7717 612 : spdk_trace_record_tsc(tsc, TRACE_BDEV_IO_DONE, bdev_ch->trace_id, 0, (uintptr_t)bdev_io,
7718 : bdev_io->internal.caller_ctx, bdev_ch->queue_depth);
7719 :
7720 612 : if (bdev_ch->histogram) {
7721 4 : if (bdev_io->bdev->internal.histogram_io_type == 0 ||
7722 0 : bdev_io->bdev->internal.histogram_io_type == bdev_io->type) {
7723 : /*
7724 : * Tally all I/O types if the histogram_io_type is set to 0.
7725 : */
7726 4 : spdk_histogram_data_tally(bdev_ch->histogram, tsc_diff);
7727 4 : }
7728 4 : }
7729 :
7730 612 : bdev_io_update_io_stat(bdev_io, tsc_diff);
7731 612 : _bdev_io_complete(bdev_io);
7732 620 : }
7733 :
7734 : /* The difference between this function and bdev_io_complete() is that this should be called to
7735 : * complete IOs that haven't been submitted via bdev_io_submit(), as they weren't added onto the
7736 : * io_submitted list and don't have submit_tsc updated.
7737 : */
7738 : static inline void
7739 0 : bdev_io_complete_unsubmitted(struct spdk_bdev_io *bdev_io)
7740 : {
7741 : /* Since the IO hasn't been submitted it's bound to be failed */
7742 0 : assert(bdev_io->internal.status != SPDK_BDEV_IO_STATUS_SUCCESS);
7743 :
7744 : /* At this point we don't know if the IO is completed from submission context or not, but,
7745 : * since this is an error path, we can always do an spdk_thread_send_msg(). */
7746 0 : spdk_thread_send_msg(spdk_bdev_io_get_thread(bdev_io),
7747 0 : _bdev_io_complete, bdev_io);
7748 0 : }
7749 :
7750 : static void bdev_destroy_cb(void *io_device);
7751 :
7752 : static inline void
7753 18 : _bdev_reset_complete(void *ctx)
7754 : {
7755 18 : struct spdk_bdev_io *bdev_io = ctx;
7756 :
7757 : /* Put the channel reference we got in submission. */
7758 18 : assert(bdev_io->u.reset.ch_ref != NULL);
7759 18 : spdk_put_io_channel(bdev_io->u.reset.ch_ref);
7760 18 : bdev_io->u.reset.ch_ref = NULL;
7761 :
7762 18 : bdev_io_complete(bdev_io);
7763 18 : }
7764 :
7765 : static void
7766 16 : bdev_reset_complete(struct spdk_bdev *bdev, void *_ctx, int status)
7767 : {
7768 16 : struct spdk_bdev_io *bdev_io = _ctx;
7769 : bdev_io_tailq_t queued_resets;
7770 : struct spdk_bdev_io *queued_reset;
7771 :
7772 16 : assert(bdev_io == bdev->internal.reset_in_progress);
7773 :
7774 16 : TAILQ_INIT(&queued_resets);
7775 :
7776 16 : spdk_spin_lock(&bdev->internal.spinlock);
7777 16 : TAILQ_SWAP(&bdev->internal.queued_resets, &queued_resets,
7778 : spdk_bdev_io, internal.link);
7779 16 : bdev->internal.reset_in_progress = NULL;
7780 16 : spdk_spin_unlock(&bdev->internal.spinlock);
7781 :
7782 18 : while (!TAILQ_EMPTY(&queued_resets)) {
7783 2 : queued_reset = TAILQ_FIRST(&queued_resets);
7784 2 : TAILQ_REMOVE(&queued_resets, queued_reset, internal.link);
7785 2 : queued_reset->internal.status = bdev_io->internal.status;
7786 4 : spdk_thread_send_msg(spdk_bdev_io_get_thread(queued_reset),
7787 2 : _bdev_reset_complete, queued_reset);
7788 : }
7789 :
7790 16 : _bdev_reset_complete(bdev_io);
7791 :
7792 16 : if (bdev->internal.status == SPDK_BDEV_STATUS_REMOVING &&
7793 1 : TAILQ_EMPTY(&bdev->internal.open_descs)) {
7794 1 : spdk_io_device_unregister(__bdev_to_io_dev(bdev), bdev_destroy_cb);
7795 1 : }
7796 16 : }
7797 :
7798 : static void
7799 20 : bdev_unfreeze_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
7800 : struct spdk_io_channel *_ch, void *_ctx)
7801 : {
7802 20 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
7803 :
7804 20 : ch->flags &= ~BDEV_CH_RESET_IN_PROGRESS;
7805 :
7806 20 : spdk_bdev_for_each_channel_continue(i, 0);
7807 20 : }
7808 :
7809 : static void
7810 0 : bdev_io_complete_sequence_cb(void *ctx, int status)
7811 : {
7812 0 : struct spdk_bdev_io *bdev_io = ctx;
7813 :
7814 : /* u.bdev.accel_sequence should have already been cleared at this point */
7815 0 : assert(bdev_io->u.bdev.accel_sequence == NULL);
7816 0 : assert(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS);
7817 0 : bdev_io->internal.f.has_accel_sequence = false;
7818 :
7819 0 : if (spdk_unlikely(status != 0)) {
7820 0 : SPDK_ERRLOG("Failed to execute accel sequence, status=%d\n", status);
7821 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
7822 0 : }
7823 :
7824 0 : bdev_io_complete(bdev_io);
7825 0 : }
7826 :
7827 : void
7828 598 : spdk_bdev_io_complete(struct spdk_bdev_io *bdev_io, enum spdk_bdev_io_status status)
7829 : {
7830 598 : struct spdk_bdev *bdev = bdev_io->bdev;
7831 598 : struct spdk_bdev_channel *bdev_ch = bdev_io->internal.ch;
7832 598 : struct spdk_bdev_shared_resource *shared_resource = bdev_ch->shared_resource;
7833 :
7834 598 : if (spdk_unlikely(bdev_io->internal.status != SPDK_BDEV_IO_STATUS_PENDING)) {
7835 0 : SPDK_ERRLOG("Unexpected completion on IO from %s module, status was %s\n",
7836 : spdk_bdev_get_module_name(bdev),
7837 : bdev_io_status_get_string(bdev_io->internal.status));
7838 0 : assert(false);
7839 : }
7840 598 : bdev_io->internal.status = status;
7841 :
7842 598 : if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_RESET)) {
7843 16 : assert(bdev_io == bdev->internal.reset_in_progress);
7844 16 : spdk_bdev_for_each_channel(bdev, bdev_unfreeze_channel, bdev_io,
7845 : bdev_reset_complete);
7846 16 : return;
7847 : } else {
7848 582 : bdev_io_decrement_outstanding(bdev_ch, shared_resource);
7849 582 : if (spdk_likely(status == SPDK_BDEV_IO_STATUS_SUCCESS)) {
7850 485 : if (bdev_io_needs_sequence_exec(bdev_io->internal.desc, bdev_io)) {
7851 0 : bdev_io_exec_sequence(bdev_io, bdev_io_complete_sequence_cb);
7852 0 : return;
7853 485 : } else if (spdk_unlikely(bdev_io->internal.f.has_bounce_buf &&
7854 : !bdev_io_use_accel_sequence(bdev_io))) {
7855 26 : _bdev_io_push_bounce_data_buffer(bdev_io,
7856 : _bdev_io_complete_push_bounce_done);
7857 : /* bdev IO will be completed in the callback */
7858 26 : return;
7859 : }
7860 459 : }
7861 :
7862 556 : if (spdk_unlikely(_bdev_io_handle_no_mem(bdev_io, BDEV_IO_RETRY_STATE_SUBMIT))) {
7863 5 : return;
7864 : }
7865 : }
7866 :
7867 551 : bdev_io_complete(bdev_io);
7868 598 : }
7869 :
7870 : void
7871 0 : spdk_bdev_io_complete_scsi_status(struct spdk_bdev_io *bdev_io, enum spdk_scsi_status sc,
7872 : enum spdk_scsi_sense sk, uint8_t asc, uint8_t ascq)
7873 : {
7874 : enum spdk_bdev_io_status status;
7875 :
7876 0 : if (sc == SPDK_SCSI_STATUS_GOOD) {
7877 0 : status = SPDK_BDEV_IO_STATUS_SUCCESS;
7878 0 : } else {
7879 0 : status = SPDK_BDEV_IO_STATUS_SCSI_ERROR;
7880 0 : bdev_io->internal.error.scsi.sc = sc;
7881 0 : bdev_io->internal.error.scsi.sk = sk;
7882 0 : bdev_io->internal.error.scsi.asc = asc;
7883 0 : bdev_io->internal.error.scsi.ascq = ascq;
7884 : }
7885 :
7886 0 : spdk_bdev_io_complete(bdev_io, status);
7887 0 : }
7888 :
7889 : void
7890 0 : spdk_bdev_io_get_scsi_status(const struct spdk_bdev_io *bdev_io,
7891 : int *sc, int *sk, int *asc, int *ascq)
7892 : {
7893 0 : assert(sc != NULL);
7894 0 : assert(sk != NULL);
7895 0 : assert(asc != NULL);
7896 0 : assert(ascq != NULL);
7897 :
7898 0 : switch (bdev_io->internal.status) {
7899 : case SPDK_BDEV_IO_STATUS_SUCCESS:
7900 0 : *sc = SPDK_SCSI_STATUS_GOOD;
7901 0 : *sk = SPDK_SCSI_SENSE_NO_SENSE;
7902 0 : *asc = SPDK_SCSI_ASC_NO_ADDITIONAL_SENSE;
7903 0 : *ascq = SPDK_SCSI_ASCQ_CAUSE_NOT_REPORTABLE;
7904 0 : break;
7905 : case SPDK_BDEV_IO_STATUS_NVME_ERROR:
7906 0 : spdk_scsi_nvme_translate(bdev_io, sc, sk, asc, ascq);
7907 0 : break;
7908 : case SPDK_BDEV_IO_STATUS_MISCOMPARE:
7909 0 : *sc = SPDK_SCSI_STATUS_CHECK_CONDITION;
7910 0 : *sk = SPDK_SCSI_SENSE_MISCOMPARE;
7911 0 : *asc = SPDK_SCSI_ASC_MISCOMPARE_DURING_VERIFY_OPERATION;
7912 0 : *ascq = bdev_io->internal.error.scsi.ascq;
7913 0 : break;
7914 : case SPDK_BDEV_IO_STATUS_SCSI_ERROR:
7915 0 : *sc = bdev_io->internal.error.scsi.sc;
7916 0 : *sk = bdev_io->internal.error.scsi.sk;
7917 0 : *asc = bdev_io->internal.error.scsi.asc;
7918 0 : *ascq = bdev_io->internal.error.scsi.ascq;
7919 0 : break;
7920 : default:
7921 0 : *sc = SPDK_SCSI_STATUS_CHECK_CONDITION;
7922 0 : *sk = SPDK_SCSI_SENSE_ABORTED_COMMAND;
7923 0 : *asc = SPDK_SCSI_ASC_NO_ADDITIONAL_SENSE;
7924 0 : *ascq = SPDK_SCSI_ASCQ_CAUSE_NOT_REPORTABLE;
7925 0 : break;
7926 : }
7927 0 : }
7928 :
7929 : void
7930 0 : spdk_bdev_io_complete_aio_status(struct spdk_bdev_io *bdev_io, int aio_result)
7931 : {
7932 : enum spdk_bdev_io_status status;
7933 :
7934 0 : if (aio_result == 0) {
7935 0 : status = SPDK_BDEV_IO_STATUS_SUCCESS;
7936 0 : } else {
7937 0 : status = SPDK_BDEV_IO_STATUS_AIO_ERROR;
7938 : }
7939 :
7940 0 : bdev_io->internal.error.aio_result = aio_result;
7941 :
7942 0 : spdk_bdev_io_complete(bdev_io, status);
7943 0 : }
7944 :
7945 : void
7946 0 : spdk_bdev_io_get_aio_status(const struct spdk_bdev_io *bdev_io, int *aio_result)
7947 : {
7948 0 : assert(aio_result != NULL);
7949 :
7950 0 : if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_AIO_ERROR) {
7951 0 : *aio_result = bdev_io->internal.error.aio_result;
7952 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS) {
7953 0 : *aio_result = 0;
7954 0 : } else {
7955 0 : *aio_result = -EIO;
7956 : }
7957 0 : }
7958 :
7959 : void
7960 0 : spdk_bdev_io_complete_nvme_status(struct spdk_bdev_io *bdev_io, uint32_t cdw0, int sct, int sc)
7961 : {
7962 : enum spdk_bdev_io_status status;
7963 :
7964 0 : if (spdk_likely(sct == SPDK_NVME_SCT_GENERIC && sc == SPDK_NVME_SC_SUCCESS)) {
7965 0 : status = SPDK_BDEV_IO_STATUS_SUCCESS;
7966 0 : } else if (sct == SPDK_NVME_SCT_GENERIC && sc == SPDK_NVME_SC_ABORTED_BY_REQUEST) {
7967 0 : status = SPDK_BDEV_IO_STATUS_ABORTED;
7968 0 : } else {
7969 0 : status = SPDK_BDEV_IO_STATUS_NVME_ERROR;
7970 : }
7971 :
7972 0 : bdev_io->internal.error.nvme.cdw0 = cdw0;
7973 0 : bdev_io->internal.error.nvme.sct = sct;
7974 0 : bdev_io->internal.error.nvme.sc = sc;
7975 :
7976 0 : spdk_bdev_io_complete(bdev_io, status);
7977 0 : }
7978 :
7979 : void
7980 0 : spdk_bdev_io_get_nvme_status(const struct spdk_bdev_io *bdev_io, uint32_t *cdw0, int *sct, int *sc)
7981 : {
7982 0 : assert(sct != NULL);
7983 0 : assert(sc != NULL);
7984 0 : assert(cdw0 != NULL);
7985 :
7986 0 : if (spdk_unlikely(bdev_io->type == SPDK_BDEV_IO_TYPE_ABORT)) {
7987 0 : *sct = SPDK_NVME_SCT_GENERIC;
7988 0 : *sc = SPDK_NVME_SC_SUCCESS;
7989 0 : if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS) {
7990 0 : *cdw0 = 0;
7991 0 : } else {
7992 0 : *cdw0 = 1U;
7993 : }
7994 0 : return;
7995 : }
7996 :
7997 0 : if (spdk_likely(bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS)) {
7998 0 : *sct = SPDK_NVME_SCT_GENERIC;
7999 0 : *sc = SPDK_NVME_SC_SUCCESS;
8000 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_NVME_ERROR) {
8001 0 : *sct = bdev_io->internal.error.nvme.sct;
8002 0 : *sc = bdev_io->internal.error.nvme.sc;
8003 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_ABORTED) {
8004 0 : *sct = SPDK_NVME_SCT_GENERIC;
8005 0 : *sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
8006 0 : } else {
8007 0 : *sct = SPDK_NVME_SCT_GENERIC;
8008 0 : *sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
8009 : }
8010 :
8011 0 : *cdw0 = bdev_io->internal.error.nvme.cdw0;
8012 0 : }
8013 :
8014 : void
8015 0 : spdk_bdev_io_get_nvme_fused_status(const struct spdk_bdev_io *bdev_io, uint32_t *cdw0,
8016 : int *first_sct, int *first_sc, int *second_sct, int *second_sc)
8017 : {
8018 0 : assert(first_sct != NULL);
8019 0 : assert(first_sc != NULL);
8020 0 : assert(second_sct != NULL);
8021 0 : assert(second_sc != NULL);
8022 0 : assert(cdw0 != NULL);
8023 :
8024 0 : if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_NVME_ERROR) {
8025 0 : if (bdev_io->internal.error.nvme.sct == SPDK_NVME_SCT_MEDIA_ERROR &&
8026 0 : bdev_io->internal.error.nvme.sc == SPDK_NVME_SC_COMPARE_FAILURE) {
8027 0 : *first_sct = bdev_io->internal.error.nvme.sct;
8028 0 : *first_sc = bdev_io->internal.error.nvme.sc;
8029 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
8030 0 : *second_sc = SPDK_NVME_SC_ABORTED_FAILED_FUSED;
8031 0 : } else {
8032 0 : *first_sct = SPDK_NVME_SCT_GENERIC;
8033 0 : *first_sc = SPDK_NVME_SC_SUCCESS;
8034 0 : *second_sct = bdev_io->internal.error.nvme.sct;
8035 0 : *second_sc = bdev_io->internal.error.nvme.sc;
8036 : }
8037 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_ABORTED) {
8038 0 : *first_sct = SPDK_NVME_SCT_GENERIC;
8039 0 : *first_sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
8040 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
8041 0 : *second_sc = SPDK_NVME_SC_ABORTED_BY_REQUEST;
8042 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_SUCCESS) {
8043 0 : *first_sct = SPDK_NVME_SCT_GENERIC;
8044 0 : *first_sc = SPDK_NVME_SC_SUCCESS;
8045 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
8046 0 : *second_sc = SPDK_NVME_SC_SUCCESS;
8047 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_FIRST_FUSED_FAILED) {
8048 0 : *first_sct = SPDK_NVME_SCT_GENERIC;
8049 0 : *first_sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
8050 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
8051 0 : *second_sc = SPDK_NVME_SC_ABORTED_FAILED_FUSED;
8052 0 : } else if (bdev_io->internal.status == SPDK_BDEV_IO_STATUS_MISCOMPARE) {
8053 0 : *first_sct = SPDK_NVME_SCT_MEDIA_ERROR;
8054 0 : *first_sc = SPDK_NVME_SC_COMPARE_FAILURE;
8055 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
8056 0 : *second_sc = SPDK_NVME_SC_ABORTED_FAILED_FUSED;
8057 0 : } else {
8058 0 : *first_sct = SPDK_NVME_SCT_GENERIC;
8059 0 : *first_sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
8060 0 : *second_sct = SPDK_NVME_SCT_GENERIC;
8061 0 : *second_sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
8062 : }
8063 :
8064 0 : *cdw0 = bdev_io->internal.error.nvme.cdw0;
8065 0 : }
8066 :
8067 : void
8068 0 : spdk_bdev_io_complete_base_io_status(struct spdk_bdev_io *bdev_io,
8069 : const struct spdk_bdev_io *base_io)
8070 : {
8071 0 : switch (base_io->internal.status) {
8072 : case SPDK_BDEV_IO_STATUS_NVME_ERROR:
8073 0 : spdk_bdev_io_complete_nvme_status(bdev_io,
8074 0 : base_io->internal.error.nvme.cdw0,
8075 0 : base_io->internal.error.nvme.sct,
8076 0 : base_io->internal.error.nvme.sc);
8077 0 : break;
8078 : case SPDK_BDEV_IO_STATUS_SCSI_ERROR:
8079 0 : spdk_bdev_io_complete_scsi_status(bdev_io,
8080 0 : base_io->internal.error.scsi.sc,
8081 0 : base_io->internal.error.scsi.sk,
8082 0 : base_io->internal.error.scsi.asc,
8083 0 : base_io->internal.error.scsi.ascq);
8084 0 : break;
8085 : case SPDK_BDEV_IO_STATUS_AIO_ERROR:
8086 0 : spdk_bdev_io_complete_aio_status(bdev_io, base_io->internal.error.aio_result);
8087 0 : break;
8088 : default:
8089 0 : spdk_bdev_io_complete(bdev_io, base_io->internal.status);
8090 0 : break;
8091 : }
8092 0 : }
8093 :
8094 : struct spdk_thread *
8095 664 : spdk_bdev_io_get_thread(struct spdk_bdev_io *bdev_io)
8096 : {
8097 664 : return spdk_io_channel_get_thread(bdev_io->internal.ch->channel);
8098 : }
8099 :
8100 : struct spdk_io_channel *
8101 70 : spdk_bdev_io_get_io_channel(struct spdk_bdev_io *bdev_io)
8102 : {
8103 70 : return bdev_io->internal.ch->channel;
8104 : }
8105 :
8106 : static int
8107 132 : bdev_register(struct spdk_bdev *bdev)
8108 : {
8109 : char *bdev_name;
8110 : char uuid[SPDK_UUID_STRING_LEN];
8111 : struct spdk_iobuf_opts iobuf_opts;
8112 : int ret;
8113 :
8114 132 : assert(bdev->module != NULL);
8115 :
8116 132 : if (!bdev->name) {
8117 0 : SPDK_ERRLOG("Bdev name is NULL\n");
8118 0 : return -EINVAL;
8119 : }
8120 :
8121 132 : if (!strlen(bdev->name)) {
8122 0 : SPDK_ERRLOG("Bdev name must not be an empty string\n");
8123 0 : return -EINVAL;
8124 : }
8125 :
8126 : /* Users often register their own I/O devices using the bdev name. In
8127 : * order to avoid conflicts, prepend bdev_. */
8128 132 : bdev_name = spdk_sprintf_alloc("bdev_%s", bdev->name);
8129 132 : if (!bdev_name) {
8130 0 : SPDK_ERRLOG("Unable to allocate memory for internal bdev name.\n");
8131 0 : return -ENOMEM;
8132 : }
8133 :
8134 132 : bdev->internal.stat = bdev_alloc_io_stat(true);
8135 132 : if (!bdev->internal.stat) {
8136 0 : SPDK_ERRLOG("Unable to allocate I/O statistics structure.\n");
8137 0 : free(bdev_name);
8138 0 : return -ENOMEM;
8139 : }
8140 :
8141 132 : bdev->internal.status = SPDK_BDEV_STATUS_READY;
8142 132 : bdev->internal.measured_queue_depth = UINT64_MAX;
8143 132 : bdev->internal.claim_type = SPDK_BDEV_CLAIM_NONE;
8144 132 : memset(&bdev->internal.claim, 0, sizeof(bdev->internal.claim));
8145 132 : bdev->internal.qd_poller = NULL;
8146 132 : bdev->internal.qos = NULL;
8147 :
8148 132 : TAILQ_INIT(&bdev->internal.open_descs);
8149 132 : TAILQ_INIT(&bdev->internal.locked_ranges);
8150 132 : TAILQ_INIT(&bdev->internal.pending_locked_ranges);
8151 132 : TAILQ_INIT(&bdev->internal.queued_resets);
8152 132 : TAILQ_INIT(&bdev->aliases);
8153 :
8154 : /* UUID may be specified by the user or defined by bdev itself.
8155 : * Otherwise it will be generated here, so this field will never be empty. */
8156 132 : if (spdk_uuid_is_null(&bdev->uuid)) {
8157 43 : spdk_uuid_generate(&bdev->uuid);
8158 43 : }
8159 :
8160 : /* Add the UUID alias only if it's different than the name */
8161 132 : spdk_uuid_fmt_lower(uuid, sizeof(uuid), &bdev->uuid);
8162 132 : if (strcmp(bdev->name, uuid) != 0) {
8163 131 : ret = spdk_bdev_alias_add(bdev, uuid);
8164 131 : if (ret != 0) {
8165 2 : SPDK_ERRLOG("Unable to add uuid:%s alias for bdev %s\n", uuid, bdev->name);
8166 2 : bdev_free_io_stat(bdev->internal.stat);
8167 2 : free(bdev_name);
8168 2 : return ret;
8169 : }
8170 129 : }
8171 :
8172 130 : spdk_iobuf_get_opts(&iobuf_opts, sizeof(iobuf_opts));
8173 130 : if (spdk_bdev_get_buf_align(bdev) > 1) {
8174 0 : bdev->max_rw_size = spdk_min(bdev->max_rw_size ? bdev->max_rw_size : UINT32_MAX,
8175 : iobuf_opts.large_bufsize / bdev->blocklen);
8176 0 : }
8177 :
8178 : /* If the user didn't specify a write unit size, set it to one. */
8179 130 : if (bdev->write_unit_size == 0) {
8180 126 : bdev->write_unit_size = 1;
8181 126 : }
8182 :
8183 : /* Set ACWU value to the write unit size if bdev module did not set it (does not support it natively) */
8184 130 : if (bdev->acwu == 0) {
8185 126 : bdev->acwu = bdev->write_unit_size;
8186 126 : }
8187 :
8188 130 : if (bdev->phys_blocklen == 0) {
8189 126 : bdev->phys_blocklen = spdk_bdev_get_data_block_size(bdev);
8190 126 : }
8191 :
8192 130 : if (!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COPY)) {
8193 0 : bdev->max_copy = bdev_get_max_write(bdev, iobuf_opts.large_bufsize);
8194 0 : }
8195 :
8196 130 : if (!bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_WRITE_ZEROES)) {
8197 0 : bdev->max_write_zeroes = bdev_get_max_write(bdev, ZERO_BUFFER_SIZE);
8198 0 : }
8199 :
8200 130 : bdev->internal.reset_in_progress = NULL;
8201 130 : bdev->internal.qd_poll_in_progress = false;
8202 130 : bdev->internal.period = 0;
8203 130 : bdev->internal.new_period = 0;
8204 130 : bdev->internal.trace_id = spdk_trace_register_owner(OWNER_TYPE_BDEV, bdev_name);
8205 :
8206 : /*
8207 : * Initialize spinlock before registering IO device because spinlock is used in
8208 : * bdev_channel_create
8209 : */
8210 130 : spdk_spin_init(&bdev->internal.spinlock);
8211 :
8212 260 : spdk_io_device_register(__bdev_to_io_dev(bdev),
8213 : bdev_channel_create, bdev_channel_destroy,
8214 : sizeof(struct spdk_bdev_channel),
8215 130 : bdev_name);
8216 :
8217 : /*
8218 : * Register bdev name only after the bdev object is ready.
8219 : * After bdev_name_add returns, it is possible for other threads to start using the bdev,
8220 : * create IO channels...
8221 : */
8222 130 : ret = bdev_name_add(&bdev->internal.bdev_name, bdev, bdev->name);
8223 130 : if (ret != 0) {
8224 0 : spdk_io_device_unregister(__bdev_to_io_dev(bdev), NULL);
8225 0 : bdev_free_io_stat(bdev->internal.stat);
8226 0 : spdk_spin_destroy(&bdev->internal.spinlock);
8227 0 : free(bdev_name);
8228 0 : return ret;
8229 : }
8230 :
8231 130 : free(bdev_name);
8232 :
8233 130 : SPDK_DEBUGLOG(bdev, "Inserting bdev %s into list\n", bdev->name);
8234 130 : TAILQ_INSERT_TAIL(&g_bdev_mgr.bdevs, bdev, internal.link);
8235 :
8236 130 : return 0;
8237 132 : }
8238 :
8239 : static void
8240 131 : bdev_destroy_cb(void *io_device)
8241 : {
8242 : int rc;
8243 : struct spdk_bdev *bdev;
8244 : spdk_bdev_unregister_cb cb_fn;
8245 : void *cb_arg;
8246 :
8247 131 : bdev = __bdev_from_io_dev(io_device);
8248 :
8249 131 : if (bdev->internal.unregister_td != spdk_get_thread()) {
8250 1 : spdk_thread_send_msg(bdev->internal.unregister_td, bdev_destroy_cb, io_device);
8251 1 : return;
8252 : }
8253 :
8254 130 : cb_fn = bdev->internal.unregister_cb;
8255 130 : cb_arg = bdev->internal.unregister_ctx;
8256 :
8257 130 : spdk_spin_destroy(&bdev->internal.spinlock);
8258 130 : free(bdev->internal.qos);
8259 130 : bdev_free_io_stat(bdev->internal.stat);
8260 130 : spdk_trace_unregister_owner(bdev->internal.trace_id);
8261 :
8262 130 : rc = bdev->fn_table->destruct(bdev->ctxt);
8263 130 : if (rc < 0) {
8264 0 : SPDK_ERRLOG("destruct failed\n");
8265 0 : }
8266 130 : if (rc <= 0 && cb_fn != NULL) {
8267 10 : cb_fn(cb_arg, rc);
8268 10 : }
8269 131 : }
8270 :
8271 : void
8272 2 : spdk_bdev_destruct_done(struct spdk_bdev *bdev, int bdeverrno)
8273 : {
8274 2 : if (bdev->internal.unregister_cb != NULL) {
8275 0 : bdev->internal.unregister_cb(bdev->internal.unregister_ctx, bdeverrno);
8276 0 : }
8277 2 : }
8278 :
8279 : static void
8280 19 : _remove_notify(void *arg)
8281 : {
8282 19 : struct spdk_bdev_desc *desc = arg;
8283 :
8284 19 : _event_notify(desc, SPDK_BDEV_EVENT_REMOVE);
8285 19 : }
8286 :
8287 : /* returns: 0 - bdev removed and ready to be destructed.
8288 : * -EBUSY - bdev can't be destructed yet. */
8289 : static int
8290 145 : bdev_unregister_unsafe(struct spdk_bdev *bdev)
8291 : {
8292 : struct spdk_bdev_desc *desc, *tmp;
8293 : struct spdk_bdev_alias *alias;
8294 145 : int rc = 0;
8295 : char uuid[SPDK_UUID_STRING_LEN];
8296 :
8297 145 : assert(spdk_spin_held(&g_bdev_mgr.spinlock));
8298 145 : assert(spdk_spin_held(&bdev->internal.spinlock));
8299 :
8300 : /* Notify each descriptor about hotremoval */
8301 164 : TAILQ_FOREACH_SAFE(desc, &bdev->internal.open_descs, link, tmp) {
8302 19 : rc = -EBUSY;
8303 : /*
8304 : * Defer invocation of the event_cb to a separate message that will
8305 : * run later on its thread. This ensures this context unwinds and
8306 : * we don't recursively unregister this bdev again if the event_cb
8307 : * immediately closes its descriptor.
8308 : */
8309 19 : event_notify(desc, _remove_notify);
8310 19 : }
8311 :
8312 : /* If there are no descriptors, proceed removing the bdev */
8313 145 : if (rc == 0) {
8314 130 : bdev_examine_allowlist_remove(bdev->name);
8315 258 : TAILQ_FOREACH(alias, &bdev->aliases, tailq) {
8316 128 : bdev_examine_allowlist_remove(alias->alias.name);
8317 128 : }
8318 130 : TAILQ_REMOVE(&g_bdev_mgr.bdevs, bdev, internal.link);
8319 130 : SPDK_DEBUGLOG(bdev, "Removing bdev %s from list done\n", bdev->name);
8320 :
8321 : /* Delete the name and the UUID alias */
8322 130 : spdk_uuid_fmt_lower(uuid, sizeof(uuid), &bdev->uuid);
8323 130 : bdev_name_del_unsafe(&bdev->internal.bdev_name);
8324 130 : bdev_alias_del(bdev, uuid, bdev_name_del_unsafe);
8325 :
8326 130 : spdk_notify_send("bdev_unregister", spdk_bdev_get_name(bdev));
8327 :
8328 130 : if (bdev->internal.reset_in_progress != NULL) {
8329 : /* If reset is in progress, let the completion callback for reset
8330 : * unregister the bdev.
8331 : */
8332 1 : rc = -EBUSY;
8333 1 : }
8334 130 : }
8335 :
8336 145 : return rc;
8337 : }
8338 :
8339 : static void
8340 4 : bdev_unregister_abort_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
8341 : struct spdk_io_channel *io_ch, void *_ctx)
8342 : {
8343 4 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
8344 :
8345 4 : bdev_channel_abort_queued_ios(bdev_ch);
8346 4 : spdk_bdev_for_each_channel_continue(i, 0);
8347 4 : }
8348 :
8349 : static void
8350 130 : bdev_unregister(struct spdk_bdev *bdev, void *_ctx, int status)
8351 : {
8352 : int rc;
8353 :
8354 130 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8355 130 : spdk_spin_lock(&bdev->internal.spinlock);
8356 : /*
8357 : * Set the status to REMOVING after completing to abort channels. Otherwise,
8358 : * the last spdk_bdev_close() may call spdk_io_device_unregister() while
8359 : * spdk_bdev_for_each_channel() is executed and spdk_io_device_unregister()
8360 : * may fail.
8361 : */
8362 130 : bdev->internal.status = SPDK_BDEV_STATUS_REMOVING;
8363 130 : rc = bdev_unregister_unsafe(bdev);
8364 130 : spdk_spin_unlock(&bdev->internal.spinlock);
8365 130 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8366 :
8367 130 : if (rc == 0) {
8368 114 : spdk_io_device_unregister(__bdev_to_io_dev(bdev), bdev_destroy_cb);
8369 114 : }
8370 130 : }
8371 :
8372 : void
8373 137 : spdk_bdev_unregister(struct spdk_bdev *bdev, spdk_bdev_unregister_cb cb_fn, void *cb_arg)
8374 : {
8375 : struct spdk_thread *thread;
8376 :
8377 137 : SPDK_DEBUGLOG(bdev, "Removing bdev %s from list\n", bdev->name);
8378 :
8379 137 : thread = spdk_get_thread();
8380 137 : if (!thread) {
8381 : /* The user called this from a non-SPDK thread. */
8382 0 : if (cb_fn != NULL) {
8383 0 : cb_fn(cb_arg, -ENOTSUP);
8384 0 : }
8385 0 : return;
8386 : }
8387 :
8388 137 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8389 137 : if (bdev->internal.status == SPDK_BDEV_STATUS_UNREGISTERING ||
8390 137 : bdev->internal.status == SPDK_BDEV_STATUS_REMOVING) {
8391 7 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8392 7 : if (cb_fn) {
8393 0 : cb_fn(cb_arg, -EBUSY);
8394 0 : }
8395 7 : return;
8396 : }
8397 :
8398 130 : spdk_spin_lock(&bdev->internal.spinlock);
8399 130 : bdev->internal.status = SPDK_BDEV_STATUS_UNREGISTERING;
8400 130 : bdev->internal.unregister_cb = cb_fn;
8401 130 : bdev->internal.unregister_ctx = cb_arg;
8402 130 : bdev->internal.unregister_td = thread;
8403 130 : spdk_spin_unlock(&bdev->internal.spinlock);
8404 130 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8405 :
8406 130 : spdk_bdev_set_qd_sampling_period(bdev, 0);
8407 :
8408 130 : spdk_bdev_for_each_channel(bdev, bdev_unregister_abort_channel, bdev,
8409 : bdev_unregister);
8410 137 : }
8411 :
8412 : int
8413 4 : spdk_bdev_unregister_by_name(const char *bdev_name, struct spdk_bdev_module *module,
8414 : spdk_bdev_unregister_cb cb_fn, void *cb_arg)
8415 : {
8416 : struct spdk_bdev_desc *desc;
8417 : struct spdk_bdev *bdev;
8418 : int rc;
8419 :
8420 4 : rc = spdk_bdev_open_ext(bdev_name, false, _tmp_bdev_event_cb, NULL, &desc);
8421 4 : if (rc != 0) {
8422 1 : SPDK_ERRLOG("Failed to open bdev with name: %s\n", bdev_name);
8423 1 : return rc;
8424 : }
8425 :
8426 3 : bdev = spdk_bdev_desc_get_bdev(desc);
8427 :
8428 3 : if (bdev->module != module) {
8429 1 : spdk_bdev_close(desc);
8430 1 : SPDK_ERRLOG("Bdev %s was not registered by the specified module.\n",
8431 : bdev_name);
8432 1 : return -ENODEV;
8433 : }
8434 :
8435 2 : spdk_bdev_unregister(bdev, cb_fn, cb_arg);
8436 :
8437 2 : spdk_bdev_close(desc);
8438 :
8439 2 : return 0;
8440 4 : }
8441 :
8442 : static int
8443 269 : bdev_start_qos(struct spdk_bdev *bdev)
8444 : {
8445 : struct set_qos_limit_ctx *ctx;
8446 :
8447 : /* Enable QoS */
8448 269 : if (bdev->internal.qos && bdev->internal.qos->thread == NULL) {
8449 2 : ctx = calloc(1, sizeof(*ctx));
8450 2 : if (ctx == NULL) {
8451 0 : SPDK_ERRLOG("Failed to allocate memory for QoS context\n");
8452 0 : return -ENOMEM;
8453 : }
8454 2 : ctx->bdev = bdev;
8455 2 : spdk_bdev_for_each_channel(bdev, bdev_enable_qos_msg, ctx, bdev_enable_qos_done);
8456 2 : }
8457 :
8458 269 : return 0;
8459 269 : }
8460 :
8461 : static void
8462 25 : log_already_claimed(enum spdk_log_level level, const int line, const char *func, const char *detail,
8463 : struct spdk_bdev *bdev)
8464 : {
8465 : enum spdk_bdev_claim_type type;
8466 : const char *typename, *modname;
8467 : extern struct spdk_log_flag SPDK_LOG_bdev;
8468 :
8469 25 : assert(spdk_spin_held(&bdev->internal.spinlock));
8470 :
8471 25 : if (level >= SPDK_LOG_INFO && !SPDK_LOG_bdev.enabled) {
8472 0 : return;
8473 : }
8474 :
8475 25 : type = bdev->internal.claim_type;
8476 25 : typename = spdk_bdev_claim_get_name(type);
8477 :
8478 25 : if (type == SPDK_BDEV_CLAIM_EXCL_WRITE) {
8479 6 : modname = bdev->internal.claim.v1.module->name;
8480 12 : spdk_log(level, __FILE__, line, func, "bdev %s %s: type %s by module %s\n",
8481 6 : bdev->name, detail, typename, modname);
8482 6 : return;
8483 : }
8484 :
8485 19 : if (claim_type_is_v2(type)) {
8486 : struct spdk_bdev_module_claim *claim;
8487 :
8488 38 : TAILQ_FOREACH(claim, &bdev->internal.claim.v2.claims, link) {
8489 19 : modname = claim->module->name;
8490 38 : spdk_log(level, __FILE__, line, func, "bdev %s %s: type %s by module %s\n",
8491 19 : bdev->name, detail, typename, modname);
8492 19 : }
8493 19 : return;
8494 : }
8495 :
8496 0 : assert(false);
8497 25 : }
8498 :
8499 : static int
8500 278 : bdev_open(struct spdk_bdev *bdev, bool write, struct spdk_bdev_desc *desc)
8501 : {
8502 : struct spdk_thread *thread;
8503 278 : int rc = 0;
8504 :
8505 278 : thread = spdk_get_thread();
8506 278 : if (!thread) {
8507 0 : SPDK_ERRLOG("Cannot open bdev from non-SPDK thread.\n");
8508 0 : return -ENOTSUP;
8509 : }
8510 :
8511 278 : SPDK_DEBUGLOG(bdev, "Opening descriptor %p for bdev %s on thread %p\n", desc, bdev->name,
8512 : spdk_get_thread());
8513 :
8514 278 : desc->bdev = bdev;
8515 278 : desc->thread = thread;
8516 278 : desc->write = write;
8517 :
8518 278 : spdk_spin_lock(&bdev->internal.spinlock);
8519 278 : if (bdev->internal.status == SPDK_BDEV_STATUS_UNREGISTERING ||
8520 278 : bdev->internal.status == SPDK_BDEV_STATUS_REMOVING) {
8521 3 : spdk_spin_unlock(&bdev->internal.spinlock);
8522 3 : return -ENODEV;
8523 : }
8524 :
8525 275 : if (write && bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE) {
8526 6 : LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
8527 6 : spdk_spin_unlock(&bdev->internal.spinlock);
8528 6 : return -EPERM;
8529 : }
8530 :
8531 269 : rc = bdev_start_qos(bdev);
8532 269 : if (rc != 0) {
8533 0 : SPDK_ERRLOG("Failed to start QoS on bdev %s\n", bdev->name);
8534 0 : spdk_spin_unlock(&bdev->internal.spinlock);
8535 0 : return rc;
8536 : }
8537 :
8538 269 : TAILQ_INSERT_TAIL(&bdev->internal.open_descs, desc, link);
8539 :
8540 269 : spdk_spin_unlock(&bdev->internal.spinlock);
8541 :
8542 269 : return 0;
8543 278 : }
8544 :
8545 : static void
8546 279 : bdev_open_opts_get_defaults(struct spdk_bdev_open_opts *opts, size_t opts_size)
8547 : {
8548 279 : if (!opts) {
8549 0 : SPDK_ERRLOG("opts should not be NULL.\n");
8550 0 : return;
8551 : }
8552 :
8553 279 : if (!opts_size) {
8554 0 : SPDK_ERRLOG("opts_size should not be zero.\n");
8555 0 : return;
8556 : }
8557 :
8558 279 : memset(opts, 0, opts_size);
8559 279 : opts->size = opts_size;
8560 :
8561 : #define FIELD_OK(field) \
8562 : offsetof(struct spdk_bdev_open_opts, field) + sizeof(opts->field) <= opts_size
8563 :
8564 : #define SET_FIELD(field, value) \
8565 : if (FIELD_OK(field)) { \
8566 : opts->field = value; \
8567 : } \
8568 :
8569 279 : SET_FIELD(hide_metadata, false);
8570 :
8571 : #undef FIELD_OK
8572 : #undef SET_FIELD
8573 279 : }
8574 :
8575 : static void
8576 2 : bdev_open_opts_copy(struct spdk_bdev_open_opts *opts,
8577 : const struct spdk_bdev_open_opts *opts_src, size_t opts_size)
8578 : {
8579 2 : assert(opts);
8580 2 : assert(opts_src);
8581 :
8582 : #define SET_FIELD(field) \
8583 : if (offsetof(struct spdk_bdev_open_opts, field) + sizeof(opts->field) <= opts_size) { \
8584 : opts->field = opts_src->field; \
8585 : } \
8586 :
8587 2 : SET_FIELD(hide_metadata);
8588 :
8589 2 : opts->size = opts_src->size;
8590 :
8591 : /* We should not remove this statement, but need to update the assert statement
8592 : * if we add a new field, and also add a corresponding SET_FIELD statement.
8593 : */
8594 : SPDK_STATIC_ASSERT(sizeof(struct spdk_bdev_open_opts) == 16, "Incorrect size");
8595 :
8596 : #undef SET_FIELD
8597 2 : }
8598 :
8599 : void
8600 1 : spdk_bdev_open_opts_init(struct spdk_bdev_open_opts *opts, size_t opts_size)
8601 : {
8602 : struct spdk_bdev_open_opts opts_local;
8603 :
8604 1 : bdev_open_opts_get_defaults(&opts_local, sizeof(opts_local));
8605 1 : bdev_open_opts_copy(opts, &opts_local, opts_size);
8606 1 : }
8607 :
8608 : static int
8609 278 : bdev_desc_alloc(struct spdk_bdev *bdev, spdk_bdev_event_cb_t event_cb, void *event_ctx,
8610 : struct spdk_bdev_open_opts *user_opts, struct spdk_bdev_desc **_desc)
8611 : {
8612 : struct spdk_bdev_desc *desc;
8613 : struct spdk_bdev_open_opts opts;
8614 : unsigned int i;
8615 :
8616 278 : bdev_open_opts_get_defaults(&opts, sizeof(opts));
8617 278 : if (user_opts != NULL) {
8618 1 : bdev_open_opts_copy(&opts, user_opts, user_opts->size);
8619 1 : }
8620 :
8621 278 : desc = calloc(1, sizeof(*desc));
8622 278 : if (desc == NULL) {
8623 0 : SPDK_ERRLOG("Failed to allocate memory for bdev descriptor\n");
8624 0 : return -ENOMEM;
8625 : }
8626 :
8627 278 : desc->opts = opts;
8628 :
8629 278 : TAILQ_INIT(&desc->pending_media_events);
8630 278 : TAILQ_INIT(&desc->free_media_events);
8631 :
8632 278 : desc->memory_domains_supported = spdk_bdev_get_memory_domains(bdev, NULL, 0) > 0;
8633 278 : desc->callback.event_fn = event_cb;
8634 278 : desc->callback.ctx = event_ctx;
8635 278 : spdk_spin_init(&desc->spinlock);
8636 :
8637 278 : if (desc->opts.hide_metadata) {
8638 1 : if (spdk_bdev_is_md_separate(bdev)) {
8639 0 : SPDK_ERRLOG("hide_metadata option is not supported with separate metadata.\n");
8640 0 : bdev_desc_free(desc);
8641 0 : return -EINVAL;
8642 : }
8643 1 : }
8644 :
8645 278 : if (bdev->media_events) {
8646 0 : desc->media_events_buffer = calloc(MEDIA_EVENT_POOL_SIZE,
8647 : sizeof(*desc->media_events_buffer));
8648 0 : if (desc->media_events_buffer == NULL) {
8649 0 : SPDK_ERRLOG("Failed to initialize media event pool\n");
8650 0 : bdev_desc_free(desc);
8651 0 : return -ENOMEM;
8652 : }
8653 :
8654 0 : for (i = 0; i < MEDIA_EVENT_POOL_SIZE; ++i) {
8655 0 : TAILQ_INSERT_TAIL(&desc->free_media_events,
8656 : &desc->media_events_buffer[i], tailq);
8657 0 : }
8658 0 : }
8659 :
8660 278 : if (bdev->fn_table->accel_sequence_supported != NULL) {
8661 0 : for (i = 0; i < SPDK_BDEV_NUM_IO_TYPES; ++i) {
8662 0 : desc->accel_sequence_supported[i] =
8663 0 : bdev->fn_table->accel_sequence_supported(bdev->ctxt,
8664 0 : (enum spdk_bdev_io_type)i);
8665 0 : }
8666 0 : }
8667 :
8668 278 : *_desc = desc;
8669 :
8670 278 : return 0;
8671 278 : }
8672 :
8673 : static int
8674 136 : bdev_open_ext(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 : struct spdk_bdev_desc *desc;
8679 : struct spdk_bdev *bdev;
8680 : int rc;
8681 :
8682 136 : bdev = bdev_get_by_name(bdev_name);
8683 :
8684 136 : if (bdev == NULL) {
8685 1 : SPDK_NOTICELOG("Currently unable to find bdev with name: %s\n", bdev_name);
8686 1 : return -ENODEV;
8687 : }
8688 :
8689 135 : rc = bdev_desc_alloc(bdev, event_cb, event_ctx, opts, &desc);
8690 135 : if (rc != 0) {
8691 0 : return rc;
8692 : }
8693 :
8694 135 : rc = bdev_open(bdev, write, desc);
8695 135 : if (rc != 0) {
8696 7 : bdev_desc_free(desc);
8697 7 : desc = NULL;
8698 7 : }
8699 :
8700 135 : *_desc = desc;
8701 :
8702 135 : return rc;
8703 136 : }
8704 :
8705 : int
8706 138 : spdk_bdev_open_ext_v2(const char *bdev_name, bool write, spdk_bdev_event_cb_t event_cb,
8707 : void *event_ctx, struct spdk_bdev_open_opts *opts,
8708 : struct spdk_bdev_desc **_desc)
8709 : {
8710 : int rc;
8711 :
8712 138 : if (event_cb == NULL) {
8713 2 : SPDK_ERRLOG("Missing event callback function\n");
8714 2 : return -EINVAL;
8715 : }
8716 :
8717 136 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8718 136 : rc = bdev_open_ext(bdev_name, write, event_cb, event_ctx, opts, _desc);
8719 136 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8720 :
8721 136 : return rc;
8722 138 : }
8723 :
8724 : int
8725 136 : spdk_bdev_open_ext(const char *bdev_name, bool write, spdk_bdev_event_cb_t event_cb,
8726 : void *event_ctx, struct spdk_bdev_desc **_desc)
8727 : {
8728 136 : return spdk_bdev_open_ext_v2(bdev_name, write, event_cb, event_ctx, NULL, _desc);
8729 : }
8730 :
8731 : struct spdk_bdev_open_async_ctx {
8732 : char *bdev_name;
8733 : spdk_bdev_event_cb_t event_cb;
8734 : void *event_ctx;
8735 : bool write;
8736 : int rc;
8737 : spdk_bdev_open_async_cb_t cb_fn;
8738 : void *cb_arg;
8739 : struct spdk_bdev_desc *desc;
8740 : struct spdk_bdev_open_async_opts opts;
8741 : uint64_t start_ticks;
8742 : struct spdk_thread *orig_thread;
8743 : struct spdk_poller *poller;
8744 : TAILQ_ENTRY(spdk_bdev_open_async_ctx) tailq;
8745 : };
8746 :
8747 : static void
8748 0 : bdev_open_async_done(void *arg)
8749 : {
8750 0 : struct spdk_bdev_open_async_ctx *ctx = arg;
8751 :
8752 0 : ctx->cb_fn(ctx->desc, ctx->rc, ctx->cb_arg);
8753 :
8754 0 : free(ctx->bdev_name);
8755 0 : free(ctx);
8756 0 : }
8757 :
8758 : static void
8759 0 : bdev_open_async_cancel(void *arg)
8760 : {
8761 0 : struct spdk_bdev_open_async_ctx *ctx = arg;
8762 :
8763 0 : assert(ctx->rc == -ESHUTDOWN);
8764 :
8765 0 : spdk_poller_unregister(&ctx->poller);
8766 :
8767 0 : bdev_open_async_done(ctx);
8768 0 : }
8769 :
8770 : /* This is called when the bdev library finishes at shutdown. */
8771 : static void
8772 68 : bdev_open_async_fini(void)
8773 : {
8774 : struct spdk_bdev_open_async_ctx *ctx, *tmp_ctx;
8775 :
8776 68 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8777 68 : TAILQ_FOREACH_SAFE(ctx, &g_bdev_mgr.async_bdev_opens, tailq, tmp_ctx) {
8778 0 : TAILQ_REMOVE(&g_bdev_mgr.async_bdev_opens, ctx, tailq);
8779 : /*
8780 : * We have to move to ctx->orig_thread to unregister ctx->poller.
8781 : * However, there is a chance that ctx->poller is executed before
8782 : * message is executed, which could result in bdev_open_async_done()
8783 : * being called twice. To avoid such race condition, set ctx->rc to
8784 : * -ESHUTDOWN.
8785 : */
8786 0 : ctx->rc = -ESHUTDOWN;
8787 0 : spdk_thread_send_msg(ctx->orig_thread, bdev_open_async_cancel, ctx);
8788 0 : }
8789 68 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8790 68 : }
8791 :
8792 : static int bdev_open_async(void *arg);
8793 :
8794 : static void
8795 0 : _bdev_open_async(struct spdk_bdev_open_async_ctx *ctx)
8796 : {
8797 : uint64_t timeout_ticks;
8798 :
8799 0 : if (ctx->rc == -ESHUTDOWN) {
8800 : /* This context is being canceled. Do nothing. */
8801 0 : return;
8802 : }
8803 :
8804 0 : ctx->rc = bdev_open_ext(ctx->bdev_name, ctx->write, ctx->event_cb, ctx->event_ctx,
8805 0 : NULL, &ctx->desc);
8806 0 : if (ctx->rc == 0 || ctx->opts.timeout_ms == 0) {
8807 0 : goto exit;
8808 : }
8809 :
8810 0 : timeout_ticks = ctx->start_ticks + ctx->opts.timeout_ms * spdk_get_ticks_hz() / 1000ull;
8811 0 : if (spdk_get_ticks() >= timeout_ticks) {
8812 0 : SPDK_ERRLOG("Timed out while waiting for bdev '%s' to appear\n", ctx->bdev_name);
8813 0 : ctx->rc = -ETIMEDOUT;
8814 0 : goto exit;
8815 : }
8816 :
8817 0 : return;
8818 :
8819 : exit:
8820 0 : spdk_poller_unregister(&ctx->poller);
8821 0 : TAILQ_REMOVE(&g_bdev_mgr.async_bdev_opens, ctx, tailq);
8822 :
8823 : /* Completion callback is processed after stack unwinding. */
8824 0 : spdk_thread_send_msg(ctx->orig_thread, bdev_open_async_done, ctx);
8825 0 : }
8826 :
8827 : static int
8828 0 : bdev_open_async(void *arg)
8829 : {
8830 0 : struct spdk_bdev_open_async_ctx *ctx = arg;
8831 :
8832 0 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8833 :
8834 0 : _bdev_open_async(ctx);
8835 :
8836 0 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8837 :
8838 0 : return SPDK_POLLER_BUSY;
8839 : }
8840 :
8841 : static void
8842 0 : bdev_open_async_opts_copy(struct spdk_bdev_open_async_opts *opts,
8843 : struct spdk_bdev_open_async_opts *opts_src,
8844 : size_t size)
8845 : {
8846 0 : assert(opts);
8847 0 : assert(opts_src);
8848 :
8849 0 : opts->size = size;
8850 :
8851 : #define SET_FIELD(field) \
8852 : if (offsetof(struct spdk_bdev_open_async_opts, field) + sizeof(opts->field) <= size) { \
8853 : opts->field = opts_src->field; \
8854 : } \
8855 :
8856 0 : SET_FIELD(timeout_ms);
8857 :
8858 : /* Do not remove this statement, you should always update this statement when you adding a new field,
8859 : * and do not forget to add the SET_FIELD statement for your added field. */
8860 : SPDK_STATIC_ASSERT(sizeof(struct spdk_bdev_open_async_opts) == 16, "Incorrect size");
8861 :
8862 : #undef SET_FIELD
8863 0 : }
8864 :
8865 : static void
8866 0 : bdev_open_async_opts_get_default(struct spdk_bdev_open_async_opts *opts, size_t size)
8867 : {
8868 0 : assert(opts);
8869 :
8870 0 : opts->size = size;
8871 :
8872 : #define SET_FIELD(field, value) \
8873 : if (offsetof(struct spdk_bdev_open_async_opts, field) + sizeof(opts->field) <= size) { \
8874 : opts->field = value; \
8875 : } \
8876 :
8877 0 : SET_FIELD(timeout_ms, 0);
8878 :
8879 : #undef SET_FIELD
8880 0 : }
8881 :
8882 : int
8883 0 : spdk_bdev_open_async(const char *bdev_name, bool write, spdk_bdev_event_cb_t event_cb,
8884 : void *event_ctx, struct spdk_bdev_open_async_opts *opts,
8885 : spdk_bdev_open_async_cb_t open_cb, void *open_cb_arg)
8886 : {
8887 : struct spdk_bdev_open_async_ctx *ctx;
8888 :
8889 0 : if (event_cb == NULL) {
8890 0 : SPDK_ERRLOG("Missing event callback function\n");
8891 0 : return -EINVAL;
8892 : }
8893 :
8894 0 : if (open_cb == NULL) {
8895 0 : SPDK_ERRLOG("Missing open callback function\n");
8896 0 : return -EINVAL;
8897 : }
8898 :
8899 0 : if (opts != NULL && opts->size == 0) {
8900 0 : SPDK_ERRLOG("size in the options structure should not be zero\n");
8901 0 : return -EINVAL;
8902 : }
8903 :
8904 0 : ctx = calloc(1, sizeof(*ctx));
8905 0 : if (ctx == NULL) {
8906 0 : SPDK_ERRLOG("Failed to allocate open context\n");
8907 0 : return -ENOMEM;
8908 : }
8909 :
8910 0 : ctx->bdev_name = strdup(bdev_name);
8911 0 : if (ctx->bdev_name == NULL) {
8912 0 : SPDK_ERRLOG("Failed to duplicate bdev_name\n");
8913 0 : free(ctx);
8914 0 : return -ENOMEM;
8915 : }
8916 :
8917 0 : ctx->poller = SPDK_POLLER_REGISTER(bdev_open_async, ctx, 100 * 1000);
8918 0 : if (ctx->poller == NULL) {
8919 0 : SPDK_ERRLOG("Failed to register bdev_open_async poller\n");
8920 0 : free(ctx->bdev_name);
8921 0 : free(ctx);
8922 0 : return -ENOMEM;
8923 : }
8924 :
8925 0 : ctx->cb_fn = open_cb;
8926 0 : ctx->cb_arg = open_cb_arg;
8927 0 : ctx->write = write;
8928 0 : ctx->event_cb = event_cb;
8929 0 : ctx->event_ctx = event_ctx;
8930 0 : ctx->orig_thread = spdk_get_thread();
8931 0 : ctx->start_ticks = spdk_get_ticks();
8932 :
8933 0 : bdev_open_async_opts_get_default(&ctx->opts, sizeof(ctx->opts));
8934 0 : if (opts != NULL) {
8935 0 : bdev_open_async_opts_copy(&ctx->opts, opts, opts->size);
8936 0 : }
8937 :
8938 0 : spdk_spin_lock(&g_bdev_mgr.spinlock);
8939 :
8940 0 : TAILQ_INSERT_TAIL(&g_bdev_mgr.async_bdev_opens, ctx, tailq);
8941 0 : _bdev_open_async(ctx);
8942 :
8943 0 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
8944 :
8945 0 : return 0;
8946 0 : }
8947 :
8948 : static void
8949 269 : bdev_close(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc)
8950 : {
8951 : int rc;
8952 :
8953 269 : spdk_spin_lock(&bdev->internal.spinlock);
8954 269 : spdk_spin_lock(&desc->spinlock);
8955 :
8956 269 : TAILQ_REMOVE(&bdev->internal.open_descs, desc, link);
8957 :
8958 269 : desc->closed = true;
8959 :
8960 269 : if (desc->claim != NULL) {
8961 20 : bdev_desc_release_claims(desc);
8962 20 : }
8963 :
8964 269 : if (0 == desc->refs) {
8965 258 : spdk_spin_unlock(&desc->spinlock);
8966 258 : bdev_desc_free(desc);
8967 258 : } else {
8968 11 : spdk_spin_unlock(&desc->spinlock);
8969 : }
8970 :
8971 : /* If no more descriptors, kill QoS channel */
8972 269 : if (bdev->internal.qos && TAILQ_EMPTY(&bdev->internal.open_descs)) {
8973 7 : SPDK_DEBUGLOG(bdev, "Closed last descriptor for bdev %s on thread %p. Stopping QoS.\n",
8974 : bdev->name, spdk_get_thread());
8975 :
8976 7 : if (bdev_qos_destroy(bdev)) {
8977 : /* There isn't anything we can do to recover here. Just let the
8978 : * old QoS poller keep running. The QoS handling won't change
8979 : * cores when the user allocates a new channel, but it won't break. */
8980 0 : SPDK_ERRLOG("Unable to shut down QoS poller. It will continue running on the current thread.\n");
8981 0 : }
8982 7 : }
8983 :
8984 269 : if (bdev->internal.status == SPDK_BDEV_STATUS_REMOVING && TAILQ_EMPTY(&bdev->internal.open_descs)) {
8985 15 : rc = bdev_unregister_unsafe(bdev);
8986 15 : spdk_spin_unlock(&bdev->internal.spinlock);
8987 :
8988 15 : if (rc == 0) {
8989 15 : spdk_io_device_unregister(__bdev_to_io_dev(bdev), bdev_destroy_cb);
8990 15 : }
8991 15 : } else {
8992 254 : spdk_spin_unlock(&bdev->internal.spinlock);
8993 : }
8994 269 : }
8995 :
8996 : void
8997 128 : spdk_bdev_close(struct spdk_bdev_desc *desc)
8998 : {
8999 128 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
9000 :
9001 128 : SPDK_DEBUGLOG(bdev, "Closing descriptor %p for bdev %s on thread %p\n", desc, bdev->name,
9002 : spdk_get_thread());
9003 :
9004 128 : assert(desc->thread == spdk_get_thread());
9005 :
9006 128 : spdk_poller_unregister(&desc->io_timeout_poller);
9007 :
9008 128 : spdk_spin_lock(&g_bdev_mgr.spinlock);
9009 :
9010 128 : bdev_close(bdev, desc);
9011 :
9012 128 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
9013 128 : }
9014 :
9015 : int32_t
9016 3 : spdk_bdev_get_numa_id(struct spdk_bdev *bdev)
9017 : {
9018 3 : if (bdev->numa.id_valid) {
9019 2 : return bdev->numa.id;
9020 : } else {
9021 1 : return SPDK_ENV_NUMA_ID_ANY;
9022 : }
9023 3 : }
9024 :
9025 : static void
9026 130 : bdev_register_finished(void *arg)
9027 : {
9028 130 : struct spdk_bdev_desc *desc = arg;
9029 130 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
9030 :
9031 130 : spdk_notify_send("bdev_register", spdk_bdev_get_name(bdev));
9032 :
9033 130 : spdk_spin_lock(&g_bdev_mgr.spinlock);
9034 :
9035 130 : bdev_close(bdev, desc);
9036 :
9037 130 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
9038 130 : }
9039 :
9040 : int
9041 133 : spdk_bdev_register(struct spdk_bdev *bdev)
9042 : {
9043 : struct spdk_bdev_desc *desc;
9044 133 : struct spdk_thread *thread = spdk_get_thread();
9045 : int rc;
9046 :
9047 133 : if (spdk_unlikely(!spdk_thread_is_app_thread(NULL))) {
9048 1 : SPDK_ERRLOG("Cannot register bdev %s on thread %p (%s)\n", bdev->name, thread,
9049 : thread ? spdk_thread_get_name(thread) : "null");
9050 1 : return -EINVAL;
9051 : }
9052 :
9053 132 : rc = bdev_register(bdev);
9054 132 : if (rc != 0) {
9055 2 : return rc;
9056 : }
9057 :
9058 : /* A descriptor is opened to prevent bdev deletion during examination */
9059 130 : rc = bdev_desc_alloc(bdev, _tmp_bdev_event_cb, NULL, NULL, &desc);
9060 130 : if (rc != 0) {
9061 0 : spdk_bdev_unregister(bdev, NULL, NULL);
9062 0 : return rc;
9063 : }
9064 :
9065 130 : rc = bdev_open(bdev, false, desc);
9066 130 : if (rc != 0) {
9067 0 : bdev_desc_free(desc);
9068 0 : spdk_bdev_unregister(bdev, NULL, NULL);
9069 0 : return rc;
9070 : }
9071 :
9072 : /* Examine configuration before initializing I/O */
9073 130 : bdev_examine(bdev);
9074 :
9075 130 : rc = spdk_bdev_wait_for_examine(bdev_register_finished, desc);
9076 130 : if (rc != 0) {
9077 0 : bdev_close(bdev, desc);
9078 0 : spdk_bdev_unregister(bdev, NULL, NULL);
9079 0 : }
9080 :
9081 130 : return rc;
9082 133 : }
9083 :
9084 : int
9085 26 : spdk_bdev_module_claim_bdev(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc,
9086 : struct spdk_bdev_module *module)
9087 : {
9088 26 : spdk_spin_lock(&bdev->internal.spinlock);
9089 :
9090 26 : if (bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE) {
9091 6 : LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
9092 6 : spdk_spin_unlock(&bdev->internal.spinlock);
9093 6 : return -EPERM;
9094 : }
9095 :
9096 20 : if (desc && !desc->write) {
9097 5 : desc->write = true;
9098 5 : }
9099 :
9100 20 : bdev->internal.claim_type = SPDK_BDEV_CLAIM_EXCL_WRITE;
9101 20 : bdev->internal.claim.v1.module = module;
9102 :
9103 20 : spdk_spin_unlock(&bdev->internal.spinlock);
9104 20 : return 0;
9105 26 : }
9106 :
9107 : void
9108 8 : spdk_bdev_module_release_bdev(struct spdk_bdev *bdev)
9109 : {
9110 8 : spdk_spin_lock(&bdev->internal.spinlock);
9111 :
9112 8 : assert(bdev->internal.claim.v1.module != NULL);
9113 8 : assert(bdev->internal.claim_type == SPDK_BDEV_CLAIM_EXCL_WRITE);
9114 8 : bdev->internal.claim_type = SPDK_BDEV_CLAIM_NONE;
9115 8 : bdev->internal.claim.v1.module = NULL;
9116 :
9117 8 : spdk_spin_unlock(&bdev->internal.spinlock);
9118 8 : }
9119 :
9120 : /*
9121 : * Start claims v2
9122 : */
9123 :
9124 : const char *
9125 25 : spdk_bdev_claim_get_name(enum spdk_bdev_claim_type type)
9126 : {
9127 25 : switch (type) {
9128 : case SPDK_BDEV_CLAIM_NONE:
9129 0 : return "not_claimed";
9130 : case SPDK_BDEV_CLAIM_EXCL_WRITE:
9131 6 : return "exclusive_write";
9132 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE:
9133 8 : return "read_many_write_one";
9134 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_NONE:
9135 5 : return "read_many_write_none";
9136 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
9137 6 : return "read_many_write_many";
9138 : default:
9139 0 : break;
9140 : }
9141 0 : return "invalid_claim";
9142 25 : }
9143 :
9144 : static bool
9145 115 : claim_type_is_v2(enum spdk_bdev_claim_type type)
9146 : {
9147 115 : switch (type) {
9148 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE:
9149 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_NONE:
9150 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
9151 115 : return true;
9152 : default:
9153 0 : break;
9154 : }
9155 0 : return false;
9156 115 : }
9157 :
9158 : /* Returns true if taking a claim with desc->write == false should make the descriptor writable. */
9159 : static bool
9160 17 : claim_type_promotes_to_write(enum spdk_bdev_claim_type type)
9161 : {
9162 17 : switch (type) {
9163 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE:
9164 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
9165 6 : return true;
9166 : default:
9167 11 : break;
9168 : }
9169 11 : return false;
9170 17 : }
9171 :
9172 : void
9173 57 : spdk_bdev_claim_opts_init(struct spdk_bdev_claim_opts *opts, size_t size)
9174 : {
9175 57 : if (opts == NULL) {
9176 0 : SPDK_ERRLOG("opts should not be NULL\n");
9177 0 : assert(opts != NULL);
9178 0 : return;
9179 : }
9180 57 : if (size == 0) {
9181 0 : SPDK_ERRLOG("size should not be zero\n");
9182 0 : assert(size != 0);
9183 0 : return;
9184 : }
9185 :
9186 57 : memset(opts, 0, size);
9187 57 : opts->opts_size = size;
9188 :
9189 : #define FIELD_OK(field) \
9190 : offsetof(struct spdk_bdev_claim_opts, field) + sizeof(opts->field) <= size
9191 :
9192 : #define SET_FIELD(field, value) \
9193 : if (FIELD_OK(field)) { \
9194 : opts->field = value; \
9195 : } \
9196 :
9197 57 : SET_FIELD(shared_claim_key, 0);
9198 :
9199 : #undef FIELD_OK
9200 : #undef SET_FIELD
9201 57 : }
9202 :
9203 : static int
9204 22 : claim_opts_copy(struct spdk_bdev_claim_opts *src, struct spdk_bdev_claim_opts *dst)
9205 : {
9206 22 : if (src->opts_size == 0) {
9207 0 : SPDK_ERRLOG("size should not be zero\n");
9208 0 : return -1;
9209 : }
9210 :
9211 22 : memset(dst, 0, sizeof(*dst));
9212 22 : dst->opts_size = src->opts_size;
9213 :
9214 : #define FIELD_OK(field) \
9215 : offsetof(struct spdk_bdev_claim_opts, field) + sizeof(src->field) <= src->opts_size
9216 :
9217 : #define SET_FIELD(field) \
9218 : if (FIELD_OK(field)) { \
9219 : dst->field = src->field; \
9220 : } \
9221 :
9222 22 : if (FIELD_OK(name)) {
9223 22 : snprintf(dst->name, sizeof(dst->name), "%s", src->name);
9224 22 : }
9225 :
9226 22 : SET_FIELD(shared_claim_key);
9227 :
9228 : /* You should not remove this statement, but need to update the assert statement
9229 : * if you add a new field, and also add a corresponding SET_FIELD statement */
9230 : SPDK_STATIC_ASSERT(sizeof(struct spdk_bdev_claim_opts) == 48, "Incorrect size");
9231 :
9232 : #undef FIELD_OK
9233 : #undef SET_FIELD
9234 22 : return 0;
9235 22 : }
9236 :
9237 : /* Returns 0 if a read-write-once claim can be taken. */
9238 : static int
9239 10 : claim_verify_rwo(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
9240 : struct spdk_bdev_claim_opts *opts, struct spdk_bdev_module *module)
9241 : {
9242 10 : struct spdk_bdev *bdev = desc->bdev;
9243 : struct spdk_bdev_desc *open_desc;
9244 :
9245 10 : assert(spdk_spin_held(&bdev->internal.spinlock));
9246 10 : assert(type == SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE);
9247 :
9248 10 : if (opts->shared_claim_key != 0) {
9249 1 : SPDK_ERRLOG("%s: key option not supported with read-write-once claims\n",
9250 : bdev->name);
9251 1 : return -EINVAL;
9252 : }
9253 9 : if (bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE) {
9254 1 : LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
9255 1 : return -EPERM;
9256 : }
9257 8 : if (desc->claim != NULL) {
9258 0 : SPDK_NOTICELOG("%s: descriptor already claimed bdev with module %s\n",
9259 : bdev->name, desc->claim->module->name);
9260 0 : return -EPERM;
9261 : }
9262 16 : TAILQ_FOREACH(open_desc, &bdev->internal.open_descs, link) {
9263 10 : if (desc != open_desc && open_desc->write) {
9264 2 : SPDK_NOTICELOG("%s: Cannot obtain read-write-once claim while "
9265 : "another descriptor is open for writing\n",
9266 : bdev->name);
9267 2 : return -EPERM;
9268 : }
9269 8 : }
9270 :
9271 6 : return 0;
9272 10 : }
9273 :
9274 : /* Returns 0 if a read-only-many claim can be taken. */
9275 : static int
9276 15 : claim_verify_rom(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
9277 : struct spdk_bdev_claim_opts *opts, struct spdk_bdev_module *module)
9278 : {
9279 15 : struct spdk_bdev *bdev = desc->bdev;
9280 : struct spdk_bdev_desc *open_desc;
9281 :
9282 15 : assert(spdk_spin_held(&bdev->internal.spinlock));
9283 15 : assert(type == SPDK_BDEV_CLAIM_READ_MANY_WRITE_NONE);
9284 15 : assert(desc->claim == NULL);
9285 :
9286 15 : if (desc->write) {
9287 3 : SPDK_ERRLOG("%s: Cannot obtain read-only-many claim with writable descriptor\n",
9288 : bdev->name);
9289 3 : return -EINVAL;
9290 : }
9291 12 : if (opts->shared_claim_key != 0) {
9292 1 : SPDK_ERRLOG("%s: key option not supported with read-only-may claims\n", bdev->name);
9293 1 : return -EINVAL;
9294 : }
9295 11 : if (bdev->internal.claim_type == SPDK_BDEV_CLAIM_NONE) {
9296 19 : TAILQ_FOREACH(open_desc, &bdev->internal.open_descs, link) {
9297 11 : if (open_desc->write) {
9298 0 : SPDK_NOTICELOG("%s: Cannot obtain read-only-many claim while "
9299 : "another descriptor is open for writing\n",
9300 : bdev->name);
9301 0 : return -EPERM;
9302 : }
9303 11 : }
9304 8 : }
9305 :
9306 11 : return 0;
9307 15 : }
9308 :
9309 : /* Returns 0 if a read-write-many claim can be taken. */
9310 : static int
9311 8 : claim_verify_rwm(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
9312 : struct spdk_bdev_claim_opts *opts, struct spdk_bdev_module *module)
9313 : {
9314 8 : struct spdk_bdev *bdev = desc->bdev;
9315 : struct spdk_bdev_desc *open_desc;
9316 :
9317 8 : assert(spdk_spin_held(&bdev->internal.spinlock));
9318 8 : assert(type == SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED);
9319 8 : assert(desc->claim == NULL);
9320 :
9321 8 : if (opts->shared_claim_key == 0) {
9322 2 : SPDK_ERRLOG("%s: shared_claim_key option required with read-write-may claims\n",
9323 : bdev->name);
9324 2 : return -EINVAL;
9325 : }
9326 6 : switch (bdev->internal.claim_type) {
9327 : case SPDK_BDEV_CLAIM_NONE:
9328 7 : TAILQ_FOREACH(open_desc, &bdev->internal.open_descs, link) {
9329 5 : if (open_desc == desc) {
9330 3 : continue;
9331 : }
9332 2 : if (open_desc->write) {
9333 2 : SPDK_NOTICELOG("%s: Cannot obtain read-write-many claim while "
9334 : "another descriptor is open for writing without a "
9335 : "claim\n", bdev->name);
9336 2 : return -EPERM;
9337 : }
9338 0 : }
9339 2 : break;
9340 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
9341 2 : if (opts->shared_claim_key != bdev->internal.claim.v2.key) {
9342 1 : LOG_ALREADY_CLAIMED_ERROR("already claimed with another key", bdev);
9343 1 : return -EPERM;
9344 : }
9345 1 : break;
9346 : default:
9347 0 : LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
9348 0 : return -EBUSY;
9349 : }
9350 :
9351 3 : return 0;
9352 8 : }
9353 :
9354 : /* Updates desc and its bdev with a v2 claim. */
9355 : static int
9356 20 : claim_bdev(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
9357 : struct spdk_bdev_claim_opts *opts, struct spdk_bdev_module *module)
9358 : {
9359 20 : struct spdk_bdev *bdev = desc->bdev;
9360 : struct spdk_bdev_module_claim *claim;
9361 :
9362 20 : assert(spdk_spin_held(&bdev->internal.spinlock));
9363 20 : assert(claim_type_is_v2(type));
9364 20 : assert(desc->claim == NULL);
9365 :
9366 20 : claim = calloc(1, sizeof(*desc->claim));
9367 20 : if (claim == NULL) {
9368 0 : SPDK_ERRLOG("%s: out of memory while allocating claim\n", bdev->name);
9369 0 : return -ENOMEM;
9370 : }
9371 20 : claim->module = module;
9372 20 : claim->desc = desc;
9373 : SPDK_STATIC_ASSERT(sizeof(claim->name) == sizeof(opts->name), "sizes must match");
9374 20 : memcpy(claim->name, opts->name, sizeof(claim->name));
9375 20 : desc->claim = claim;
9376 :
9377 20 : if (bdev->internal.claim_type == SPDK_BDEV_CLAIM_NONE) {
9378 16 : bdev->internal.claim_type = type;
9379 16 : TAILQ_INIT(&bdev->internal.claim.v2.claims);
9380 16 : bdev->internal.claim.v2.key = opts->shared_claim_key;
9381 16 : }
9382 20 : assert(type == bdev->internal.claim_type);
9383 :
9384 20 : TAILQ_INSERT_TAIL(&bdev->internal.claim.v2.claims, claim, link);
9385 :
9386 20 : if (!desc->write && claim_type_promotes_to_write(type)) {
9387 6 : desc->write = true;
9388 6 : }
9389 :
9390 20 : return 0;
9391 20 : }
9392 :
9393 : int
9394 44 : spdk_bdev_module_claim_bdev_desc(struct spdk_bdev_desc *desc, enum spdk_bdev_claim_type type,
9395 : struct spdk_bdev_claim_opts *_opts,
9396 : struct spdk_bdev_module *module)
9397 : {
9398 : struct spdk_bdev *bdev;
9399 : struct spdk_bdev_claim_opts opts;
9400 44 : int rc = 0;
9401 :
9402 44 : if (desc == NULL) {
9403 0 : SPDK_ERRLOG("descriptor must not be NULL\n");
9404 0 : return -EINVAL;
9405 : }
9406 :
9407 44 : bdev = desc->bdev;
9408 :
9409 44 : if (_opts == NULL) {
9410 22 : spdk_bdev_claim_opts_init(&opts, sizeof(opts));
9411 44 : } else if (claim_opts_copy(_opts, &opts) != 0) {
9412 0 : return -EINVAL;
9413 : }
9414 :
9415 44 : spdk_spin_lock(&bdev->internal.spinlock);
9416 :
9417 44 : if (bdev->internal.claim_type != SPDK_BDEV_CLAIM_NONE &&
9418 17 : bdev->internal.claim_type != type) {
9419 11 : LOG_ALREADY_CLAIMED_ERROR("already claimed", bdev);
9420 11 : spdk_spin_unlock(&bdev->internal.spinlock);
9421 11 : return -EPERM;
9422 : }
9423 :
9424 33 : if (claim_type_is_v2(type) && desc->claim != NULL) {
9425 0 : SPDK_ERRLOG("%s: descriptor already has %s claim with name '%s'\n",
9426 : bdev->name, spdk_bdev_claim_get_name(type), desc->claim->name);
9427 0 : spdk_spin_unlock(&bdev->internal.spinlock);
9428 0 : return -EPERM;
9429 : }
9430 :
9431 33 : switch (type) {
9432 : case SPDK_BDEV_CLAIM_EXCL_WRITE:
9433 0 : spdk_spin_unlock(&bdev->internal.spinlock);
9434 0 : return spdk_bdev_module_claim_bdev(bdev, desc, module);
9435 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_ONE:
9436 10 : rc = claim_verify_rwo(desc, type, &opts, module);
9437 10 : break;
9438 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_NONE:
9439 15 : rc = claim_verify_rom(desc, type, &opts, module);
9440 15 : break;
9441 : case SPDK_BDEV_CLAIM_READ_MANY_WRITE_SHARED:
9442 8 : rc = claim_verify_rwm(desc, type, &opts, module);
9443 8 : break;
9444 : default:
9445 0 : SPDK_ERRLOG("%s: claim type %d not supported\n", bdev->name, type);
9446 0 : rc = -ENOTSUP;
9447 0 : }
9448 :
9449 33 : if (rc == 0) {
9450 20 : rc = claim_bdev(desc, type, &opts, module);
9451 20 : }
9452 :
9453 33 : spdk_spin_unlock(&bdev->internal.spinlock);
9454 33 : return rc;
9455 44 : }
9456 :
9457 : static void
9458 16 : claim_reset(struct spdk_bdev *bdev)
9459 : {
9460 16 : assert(spdk_spin_held(&bdev->internal.spinlock));
9461 16 : assert(claim_type_is_v2(bdev->internal.claim_type));
9462 16 : assert(TAILQ_EMPTY(&bdev->internal.claim.v2.claims));
9463 :
9464 16 : memset(&bdev->internal.claim, 0, sizeof(bdev->internal.claim));
9465 16 : bdev->internal.claim_type = SPDK_BDEV_CLAIM_NONE;
9466 16 : }
9467 :
9468 : static void
9469 20 : bdev_desc_release_claims(struct spdk_bdev_desc *desc)
9470 : {
9471 20 : struct spdk_bdev *bdev = desc->bdev;
9472 :
9473 20 : assert(spdk_spin_held(&bdev->internal.spinlock));
9474 20 : assert(claim_type_is_v2(bdev->internal.claim_type));
9475 :
9476 20 : if (bdev->internal.examine_in_progress == 0) {
9477 20 : TAILQ_REMOVE(&bdev->internal.claim.v2.claims, desc->claim, link);
9478 20 : free(desc->claim);
9479 20 : if (TAILQ_EMPTY(&bdev->internal.claim.v2.claims)) {
9480 16 : claim_reset(bdev);
9481 16 : }
9482 20 : } else {
9483 : /* This is a dead claim that will be cleaned up when bdev_examine() is done. */
9484 0 : desc->claim->module = NULL;
9485 0 : desc->claim->desc = NULL;
9486 : }
9487 20 : desc->claim = NULL;
9488 20 : }
9489 :
9490 : /*
9491 : * End claims v2
9492 : */
9493 :
9494 : struct spdk_bdev *
9495 1554 : spdk_bdev_desc_get_bdev(struct spdk_bdev_desc *desc)
9496 : {
9497 1554 : assert(desc != NULL);
9498 1554 : return desc->bdev;
9499 : }
9500 :
9501 : int
9502 1 : spdk_for_each_bdev(void *ctx, spdk_for_each_bdev_fn fn)
9503 : {
9504 : struct spdk_bdev *bdev, *tmp;
9505 : struct spdk_bdev_desc *desc;
9506 1 : int rc = 0;
9507 :
9508 1 : assert(fn != NULL);
9509 :
9510 1 : spdk_spin_lock(&g_bdev_mgr.spinlock);
9511 1 : bdev = spdk_bdev_first();
9512 9 : while (bdev != NULL) {
9513 8 : rc = bdev_desc_alloc(bdev, _tmp_bdev_event_cb, NULL, NULL, &desc);
9514 8 : if (rc != 0) {
9515 0 : break;
9516 : }
9517 8 : rc = bdev_open(bdev, false, desc);
9518 8 : if (rc != 0) {
9519 1 : bdev_desc_free(desc);
9520 1 : if (rc == -ENODEV) {
9521 : /* Ignore the error and move to the next bdev. */
9522 1 : rc = 0;
9523 1 : bdev = spdk_bdev_next(bdev);
9524 1 : continue;
9525 : }
9526 0 : break;
9527 : }
9528 7 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
9529 :
9530 7 : rc = fn(ctx, bdev);
9531 :
9532 7 : spdk_spin_lock(&g_bdev_mgr.spinlock);
9533 7 : tmp = spdk_bdev_next(bdev);
9534 7 : bdev_close(bdev, desc);
9535 7 : if (rc != 0) {
9536 0 : break;
9537 : }
9538 7 : bdev = tmp;
9539 : }
9540 1 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
9541 :
9542 1 : return rc;
9543 : }
9544 :
9545 : int
9546 1 : spdk_for_each_bdev_leaf(void *ctx, spdk_for_each_bdev_fn fn)
9547 : {
9548 : struct spdk_bdev *bdev, *tmp;
9549 : struct spdk_bdev_desc *desc;
9550 1 : int rc = 0;
9551 :
9552 1 : assert(fn != NULL);
9553 :
9554 1 : spdk_spin_lock(&g_bdev_mgr.spinlock);
9555 1 : bdev = spdk_bdev_first_leaf();
9556 6 : while (bdev != NULL) {
9557 5 : rc = bdev_desc_alloc(bdev, _tmp_bdev_event_cb, NULL, NULL, &desc);
9558 5 : if (rc != 0) {
9559 0 : break;
9560 : }
9561 5 : rc = bdev_open(bdev, false, desc);
9562 5 : if (rc != 0) {
9563 1 : bdev_desc_free(desc);
9564 1 : if (rc == -ENODEV) {
9565 : /* Ignore the error and move to the next bdev. */
9566 1 : rc = 0;
9567 1 : bdev = spdk_bdev_next_leaf(bdev);
9568 1 : continue;
9569 : }
9570 0 : break;
9571 : }
9572 4 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
9573 :
9574 4 : rc = fn(ctx, bdev);
9575 :
9576 4 : spdk_spin_lock(&g_bdev_mgr.spinlock);
9577 4 : tmp = spdk_bdev_next_leaf(bdev);
9578 4 : bdev_close(bdev, desc);
9579 4 : if (rc != 0) {
9580 0 : break;
9581 : }
9582 4 : bdev = tmp;
9583 : }
9584 1 : spdk_spin_unlock(&g_bdev_mgr.spinlock);
9585 :
9586 1 : return rc;
9587 : }
9588 :
9589 : void
9590 0 : spdk_bdev_io_get_iovec(struct spdk_bdev_io *bdev_io, struct iovec **iovp, int *iovcntp)
9591 : {
9592 : struct iovec *iovs;
9593 : int iovcnt;
9594 :
9595 0 : if (bdev_io == NULL) {
9596 0 : return;
9597 : }
9598 :
9599 0 : switch (bdev_io->type) {
9600 : case SPDK_BDEV_IO_TYPE_READ:
9601 : case SPDK_BDEV_IO_TYPE_WRITE:
9602 : case SPDK_BDEV_IO_TYPE_ZCOPY:
9603 0 : iovs = bdev_io->u.bdev.iovs;
9604 0 : iovcnt = bdev_io->u.bdev.iovcnt;
9605 0 : break;
9606 : default:
9607 0 : iovs = NULL;
9608 0 : iovcnt = 0;
9609 0 : break;
9610 : }
9611 :
9612 0 : if (iovp) {
9613 0 : *iovp = iovs;
9614 0 : }
9615 0 : if (iovcntp) {
9616 0 : *iovcntp = iovcnt;
9617 0 : }
9618 0 : }
9619 :
9620 : void *
9621 0 : spdk_bdev_io_get_md_buf(struct spdk_bdev_io *bdev_io)
9622 : {
9623 0 : if (bdev_io == NULL) {
9624 0 : return NULL;
9625 : }
9626 :
9627 0 : if (!spdk_bdev_is_md_separate(bdev_io->bdev)) {
9628 0 : return NULL;
9629 : }
9630 :
9631 0 : if (bdev_io->type == SPDK_BDEV_IO_TYPE_READ ||
9632 0 : bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
9633 0 : return bdev_io->u.bdev.md_buf;
9634 : }
9635 :
9636 0 : return NULL;
9637 0 : }
9638 :
9639 : void *
9640 0 : spdk_bdev_io_get_cb_arg(struct spdk_bdev_io *bdev_io)
9641 : {
9642 0 : if (bdev_io == NULL) {
9643 0 : assert(false);
9644 : return NULL;
9645 : }
9646 :
9647 0 : return bdev_io->internal.caller_ctx;
9648 : }
9649 :
9650 : void
9651 7 : spdk_bdev_module_list_add(struct spdk_bdev_module *bdev_module)
9652 : {
9653 :
9654 7 : if (spdk_bdev_module_list_find(bdev_module->name)) {
9655 0 : SPDK_ERRLOG("ERROR: module '%s' already registered.\n", bdev_module->name);
9656 0 : assert(false);
9657 : }
9658 :
9659 7 : spdk_spin_init(&bdev_module->internal.spinlock);
9660 7 : TAILQ_INIT(&bdev_module->internal.quiesced_ranges);
9661 :
9662 : /*
9663 : * Modules with examine callbacks must be initialized first, so they are
9664 : * ready to handle examine callbacks from later modules that will
9665 : * register physical bdevs.
9666 : */
9667 7 : if (bdev_module->examine_config != NULL || bdev_module->examine_disk != NULL) {
9668 4 : TAILQ_INSERT_HEAD(&g_bdev_mgr.bdev_modules, bdev_module, internal.tailq);
9669 4 : } else {
9670 3 : TAILQ_INSERT_TAIL(&g_bdev_mgr.bdev_modules, bdev_module, internal.tailq);
9671 : }
9672 7 : }
9673 :
9674 : struct spdk_bdev_module *
9675 7 : spdk_bdev_module_list_find(const char *name)
9676 : {
9677 : struct spdk_bdev_module *bdev_module;
9678 :
9679 14 : TAILQ_FOREACH(bdev_module, &g_bdev_mgr.bdev_modules, internal.tailq) {
9680 7 : if (strcmp(name, bdev_module->name) == 0) {
9681 0 : break;
9682 : }
9683 7 : }
9684 :
9685 7 : return bdev_module;
9686 : }
9687 :
9688 : static int
9689 6 : bdev_write_zero_buffer(struct spdk_bdev_io *bdev_io)
9690 : {
9691 : uint64_t num_blocks;
9692 6 : void *md_buf = NULL;
9693 :
9694 6 : num_blocks = bdev_io->u.bdev.num_blocks;
9695 :
9696 6 : if (spdk_bdev_is_md_separate(bdev_io->bdev)) {
9697 4 : md_buf = (char *)g_bdev_mgr.zero_buffer +
9698 2 : spdk_bdev_get_block_size(bdev_io->bdev) * num_blocks;
9699 2 : }
9700 :
9701 12 : return bdev_write_blocks_with_md(bdev_io->internal.desc,
9702 6 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
9703 6 : g_bdev_mgr.zero_buffer, md_buf,
9704 6 : bdev_io->u.bdev.offset_blocks, num_blocks,
9705 6 : bdev_write_zero_buffer_done, bdev_io);
9706 : }
9707 :
9708 : static void
9709 6 : bdev_write_zero_buffer_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
9710 : {
9711 6 : struct spdk_bdev_io *parent_io = cb_arg;
9712 :
9713 6 : spdk_bdev_free_io(bdev_io);
9714 :
9715 6 : parent_io->internal.status = success ? SPDK_BDEV_IO_STATUS_SUCCESS : SPDK_BDEV_IO_STATUS_FAILED;
9716 6 : parent_io->internal.cb(parent_io, success, parent_io->internal.caller_ctx);
9717 6 : }
9718 :
9719 : static void
9720 10 : bdev_set_qos_limit_done(struct set_qos_limit_ctx *ctx, int status)
9721 : {
9722 10 : spdk_spin_lock(&ctx->bdev->internal.spinlock);
9723 10 : ctx->bdev->internal.qos_mod_in_progress = false;
9724 10 : spdk_spin_unlock(&ctx->bdev->internal.spinlock);
9725 :
9726 10 : if (ctx->cb_fn) {
9727 8 : ctx->cb_fn(ctx->cb_arg, status);
9728 8 : }
9729 10 : free(ctx);
9730 10 : }
9731 :
9732 : static void
9733 2 : bdev_disable_qos_done(void *cb_arg)
9734 : {
9735 2 : struct set_qos_limit_ctx *ctx = cb_arg;
9736 2 : struct spdk_bdev *bdev = ctx->bdev;
9737 : struct spdk_bdev_qos *qos;
9738 :
9739 2 : spdk_spin_lock(&bdev->internal.spinlock);
9740 2 : qos = bdev->internal.qos;
9741 2 : bdev->internal.qos = NULL;
9742 2 : spdk_spin_unlock(&bdev->internal.spinlock);
9743 :
9744 2 : if (qos->thread != NULL) {
9745 2 : spdk_put_io_channel(spdk_io_channel_from_ctx(qos->ch));
9746 2 : spdk_poller_unregister(&qos->poller);
9747 2 : }
9748 :
9749 2 : free(qos);
9750 :
9751 2 : bdev_set_qos_limit_done(ctx, 0);
9752 2 : }
9753 :
9754 : static void
9755 2 : bdev_disable_qos_msg_done(struct spdk_bdev *bdev, void *_ctx, int status)
9756 : {
9757 2 : struct set_qos_limit_ctx *ctx = _ctx;
9758 : struct spdk_thread *thread;
9759 :
9760 2 : spdk_spin_lock(&bdev->internal.spinlock);
9761 2 : thread = bdev->internal.qos->thread;
9762 2 : spdk_spin_unlock(&bdev->internal.spinlock);
9763 :
9764 2 : if (thread != NULL) {
9765 2 : spdk_thread_send_msg(thread, bdev_disable_qos_done, ctx);
9766 2 : } else {
9767 0 : bdev_disable_qos_done(ctx);
9768 : }
9769 2 : }
9770 :
9771 : static void
9772 4 : bdev_disable_qos_msg(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
9773 : struct spdk_io_channel *ch, void *_ctx)
9774 : {
9775 4 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(ch);
9776 : struct spdk_bdev_io *bdev_io;
9777 :
9778 4 : bdev_ch->flags &= ~BDEV_CH_QOS_ENABLED;
9779 :
9780 6 : while (!TAILQ_EMPTY(&bdev_ch->qos_queued_io)) {
9781 : /* Re-submit the queued I/O. */
9782 2 : bdev_io = TAILQ_FIRST(&bdev_ch->qos_queued_io);
9783 2 : TAILQ_REMOVE(&bdev_ch->qos_queued_io, bdev_io, internal.link);
9784 2 : _bdev_io_submit(bdev_io);
9785 : }
9786 :
9787 4 : spdk_bdev_for_each_channel_continue(i, 0);
9788 4 : }
9789 :
9790 : static void
9791 1 : bdev_update_qos_rate_limit_msg(void *cb_arg)
9792 : {
9793 1 : struct set_qos_limit_ctx *ctx = cb_arg;
9794 1 : struct spdk_bdev *bdev = ctx->bdev;
9795 :
9796 1 : spdk_spin_lock(&bdev->internal.spinlock);
9797 1 : bdev_qos_update_max_quota_per_timeslice(bdev->internal.qos);
9798 1 : spdk_spin_unlock(&bdev->internal.spinlock);
9799 :
9800 1 : bdev_set_qos_limit_done(ctx, 0);
9801 1 : }
9802 :
9803 : static void
9804 9 : bdev_enable_qos_msg(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
9805 : struct spdk_io_channel *ch, void *_ctx)
9806 : {
9807 9 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(ch);
9808 :
9809 9 : spdk_spin_lock(&bdev->internal.spinlock);
9810 9 : bdev_enable_qos(bdev, bdev_ch);
9811 9 : spdk_spin_unlock(&bdev->internal.spinlock);
9812 9 : spdk_bdev_for_each_channel_continue(i, 0);
9813 9 : }
9814 :
9815 : static void
9816 6 : bdev_enable_qos_done(struct spdk_bdev *bdev, void *_ctx, int status)
9817 : {
9818 6 : struct set_qos_limit_ctx *ctx = _ctx;
9819 :
9820 6 : bdev_set_qos_limit_done(ctx, status);
9821 6 : }
9822 :
9823 : static void
9824 7 : bdev_set_qos_rate_limits(struct spdk_bdev *bdev, uint64_t *limits)
9825 : {
9826 : int i;
9827 :
9828 7 : assert(bdev->internal.qos != NULL);
9829 :
9830 35 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
9831 28 : if (limits[i] != SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
9832 28 : bdev->internal.qos->rate_limits[i].limit = limits[i];
9833 :
9834 28 : if (limits[i] == 0) {
9835 19 : bdev->internal.qos->rate_limits[i].limit =
9836 : SPDK_BDEV_QOS_LIMIT_NOT_DEFINED;
9837 19 : }
9838 28 : }
9839 28 : }
9840 7 : }
9841 :
9842 : void
9843 9 : spdk_bdev_set_qos_rate_limits(struct spdk_bdev *bdev, uint64_t *limits,
9844 : void (*cb_fn)(void *cb_arg, int status), void *cb_arg)
9845 : {
9846 : struct set_qos_limit_ctx *ctx;
9847 : uint32_t limit_set_complement;
9848 : uint64_t min_limit_per_sec;
9849 : int i;
9850 9 : bool disable_rate_limit = true;
9851 :
9852 45 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
9853 36 : if (limits[i] == SPDK_BDEV_QOS_LIMIT_NOT_DEFINED) {
9854 0 : continue;
9855 : }
9856 :
9857 36 : if (limits[i] > 0) {
9858 10 : disable_rate_limit = false;
9859 10 : }
9860 :
9861 36 : if (bdev_qos_is_iops_rate_limit(i) == true) {
9862 9 : min_limit_per_sec = SPDK_BDEV_QOS_MIN_IOS_PER_SEC;
9863 9 : } else {
9864 27 : if (limits[i] > SPDK_BDEV_QOS_MAX_MBYTES_PER_SEC) {
9865 0 : SPDK_WARNLOG("Requested rate limit %" PRIu64 " will result in uint64_t overflow, "
9866 : "reset to %" PRIu64 "\n", limits[i], SPDK_BDEV_QOS_MAX_MBYTES_PER_SEC);
9867 0 : limits[i] = SPDK_BDEV_QOS_MAX_MBYTES_PER_SEC;
9868 0 : }
9869 : /* Change from megabyte to byte rate limit */
9870 27 : limits[i] = limits[i] * 1024 * 1024;
9871 27 : min_limit_per_sec = SPDK_BDEV_QOS_MIN_BYTES_PER_SEC;
9872 : }
9873 :
9874 36 : limit_set_complement = limits[i] % min_limit_per_sec;
9875 36 : if (limit_set_complement) {
9876 0 : SPDK_ERRLOG("Requested rate limit %" PRIu64 " is not a multiple of %" PRIu64 "\n",
9877 : limits[i], min_limit_per_sec);
9878 0 : limits[i] += min_limit_per_sec - limit_set_complement;
9879 0 : SPDK_ERRLOG("Round up the rate limit to %" PRIu64 "\n", limits[i]);
9880 0 : }
9881 36 : }
9882 :
9883 9 : ctx = calloc(1, sizeof(*ctx));
9884 9 : if (ctx == NULL) {
9885 0 : cb_fn(cb_arg, -ENOMEM);
9886 0 : return;
9887 : }
9888 :
9889 9 : ctx->cb_fn = cb_fn;
9890 9 : ctx->cb_arg = cb_arg;
9891 9 : ctx->bdev = bdev;
9892 :
9893 9 : spdk_spin_lock(&bdev->internal.spinlock);
9894 9 : if (bdev->internal.qos_mod_in_progress) {
9895 1 : spdk_spin_unlock(&bdev->internal.spinlock);
9896 1 : free(ctx);
9897 1 : cb_fn(cb_arg, -EAGAIN);
9898 1 : return;
9899 : }
9900 8 : bdev->internal.qos_mod_in_progress = true;
9901 :
9902 8 : if (disable_rate_limit == true && bdev->internal.qos) {
9903 10 : for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
9904 8 : if (limits[i] == SPDK_BDEV_QOS_LIMIT_NOT_DEFINED &&
9905 0 : (bdev->internal.qos->rate_limits[i].limit > 0 &&
9906 0 : bdev->internal.qos->rate_limits[i].limit !=
9907 : SPDK_BDEV_QOS_LIMIT_NOT_DEFINED)) {
9908 0 : disable_rate_limit = false;
9909 0 : break;
9910 : }
9911 8 : }
9912 2 : }
9913 :
9914 8 : if (disable_rate_limit == false) {
9915 5 : if (bdev->internal.qos == NULL) {
9916 4 : bdev->internal.qos = calloc(1, sizeof(*bdev->internal.qos));
9917 4 : if (!bdev->internal.qos) {
9918 0 : spdk_spin_unlock(&bdev->internal.spinlock);
9919 0 : SPDK_ERRLOG("Unable to allocate memory for QoS tracking\n");
9920 0 : bdev_set_qos_limit_done(ctx, -ENOMEM);
9921 0 : return;
9922 : }
9923 4 : }
9924 :
9925 5 : if (bdev->internal.qos->thread == NULL) {
9926 : /* Enabling */
9927 4 : bdev_set_qos_rate_limits(bdev, limits);
9928 :
9929 4 : spdk_bdev_for_each_channel(bdev, bdev_enable_qos_msg, ctx,
9930 : bdev_enable_qos_done);
9931 4 : } else {
9932 : /* Updating */
9933 1 : bdev_set_qos_rate_limits(bdev, limits);
9934 :
9935 2 : spdk_thread_send_msg(bdev->internal.qos->thread,
9936 1 : bdev_update_qos_rate_limit_msg, ctx);
9937 : }
9938 5 : } else {
9939 3 : if (bdev->internal.qos != NULL) {
9940 2 : bdev_set_qos_rate_limits(bdev, limits);
9941 :
9942 : /* Disabling */
9943 2 : spdk_bdev_for_each_channel(bdev, bdev_disable_qos_msg, ctx,
9944 : bdev_disable_qos_msg_done);
9945 2 : } else {
9946 1 : spdk_spin_unlock(&bdev->internal.spinlock);
9947 1 : bdev_set_qos_limit_done(ctx, 0);
9948 1 : return;
9949 : }
9950 : }
9951 :
9952 7 : spdk_spin_unlock(&bdev->internal.spinlock);
9953 9 : }
9954 :
9955 : struct spdk_bdev_histogram_ctx {
9956 : spdk_bdev_histogram_status_cb cb_fn;
9957 : void *cb_arg;
9958 : struct spdk_bdev *bdev;
9959 : int status;
9960 : };
9961 :
9962 : static void
9963 2 : bdev_histogram_disable_channel_cb(struct spdk_bdev *bdev, void *_ctx, int status)
9964 : {
9965 2 : struct spdk_bdev_histogram_ctx *ctx = _ctx;
9966 :
9967 2 : spdk_spin_lock(&ctx->bdev->internal.spinlock);
9968 2 : ctx->bdev->internal.histogram_in_progress = false;
9969 2 : spdk_spin_unlock(&ctx->bdev->internal.spinlock);
9970 2 : ctx->cb_fn(ctx->cb_arg, ctx->status);
9971 2 : free(ctx);
9972 2 : }
9973 :
9974 : static void
9975 3 : bdev_histogram_disable_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 :
9980 3 : if (ch->histogram != NULL) {
9981 3 : spdk_histogram_data_free(ch->histogram);
9982 3 : ch->histogram = NULL;
9983 3 : }
9984 3 : spdk_bdev_for_each_channel_continue(i, 0);
9985 3 : }
9986 :
9987 : static void
9988 2 : bdev_histogram_enable_channel_cb(struct spdk_bdev *bdev, void *_ctx, int status)
9989 : {
9990 2 : struct spdk_bdev_histogram_ctx *ctx = _ctx;
9991 :
9992 2 : if (status != 0) {
9993 0 : ctx->status = status;
9994 0 : ctx->bdev->internal.histogram_enabled = false;
9995 0 : spdk_bdev_for_each_channel(ctx->bdev, bdev_histogram_disable_channel, ctx,
9996 : bdev_histogram_disable_channel_cb);
9997 0 : } else {
9998 2 : spdk_spin_lock(&ctx->bdev->internal.spinlock);
9999 2 : ctx->bdev->internal.histogram_in_progress = false;
10000 2 : spdk_spin_unlock(&ctx->bdev->internal.spinlock);
10001 2 : ctx->cb_fn(ctx->cb_arg, ctx->status);
10002 2 : free(ctx);
10003 : }
10004 2 : }
10005 :
10006 : static void
10007 3 : bdev_histogram_enable_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
10008 : struct spdk_io_channel *_ch, void *_ctx)
10009 : {
10010 3 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
10011 3 : int status = 0;
10012 :
10013 3 : if (ch->histogram == NULL) {
10014 3 : ch->histogram = spdk_histogram_data_alloc();
10015 3 : if (ch->histogram == NULL) {
10016 0 : status = -ENOMEM;
10017 0 : }
10018 3 : }
10019 :
10020 3 : spdk_bdev_for_each_channel_continue(i, status);
10021 3 : }
10022 :
10023 : void
10024 4 : spdk_bdev_histogram_enable_ext(struct spdk_bdev *bdev, spdk_bdev_histogram_status_cb cb_fn,
10025 : void *cb_arg, bool enable, struct spdk_bdev_enable_histogram_opts *opts)
10026 : {
10027 : struct spdk_bdev_histogram_ctx *ctx;
10028 :
10029 4 : ctx = calloc(1, sizeof(struct spdk_bdev_histogram_ctx));
10030 4 : if (ctx == NULL) {
10031 0 : cb_fn(cb_arg, -ENOMEM);
10032 0 : return;
10033 : }
10034 :
10035 4 : ctx->bdev = bdev;
10036 4 : ctx->status = 0;
10037 4 : ctx->cb_fn = cb_fn;
10038 4 : ctx->cb_arg = cb_arg;
10039 :
10040 4 : spdk_spin_lock(&bdev->internal.spinlock);
10041 4 : if (bdev->internal.histogram_in_progress) {
10042 0 : spdk_spin_unlock(&bdev->internal.spinlock);
10043 0 : free(ctx);
10044 0 : cb_fn(cb_arg, -EAGAIN);
10045 0 : return;
10046 : }
10047 :
10048 4 : bdev->internal.histogram_in_progress = true;
10049 4 : spdk_spin_unlock(&bdev->internal.spinlock);
10050 :
10051 4 : bdev->internal.histogram_enabled = enable;
10052 4 : bdev->internal.histogram_io_type = opts->io_type;
10053 :
10054 4 : if (enable) {
10055 : /* Allocate histogram for each channel */
10056 2 : spdk_bdev_for_each_channel(bdev, bdev_histogram_enable_channel, ctx,
10057 : bdev_histogram_enable_channel_cb);
10058 2 : } else {
10059 2 : spdk_bdev_for_each_channel(bdev, bdev_histogram_disable_channel, ctx,
10060 : bdev_histogram_disable_channel_cb);
10061 : }
10062 4 : }
10063 :
10064 : void
10065 4 : spdk_bdev_enable_histogram_opts_init(struct spdk_bdev_enable_histogram_opts *opts, size_t size)
10066 : {
10067 4 : if (opts == NULL) {
10068 0 : SPDK_ERRLOG("opts should not be NULL\n");
10069 0 : assert(opts != NULL);
10070 0 : return;
10071 : }
10072 4 : if (size == 0) {
10073 0 : SPDK_ERRLOG("size should not be zero\n");
10074 0 : assert(size != 0);
10075 0 : return;
10076 : }
10077 :
10078 4 : memset(opts, 0, size);
10079 4 : opts->size = size;
10080 :
10081 : #define FIELD_OK(field) \
10082 : offsetof(struct spdk_bdev_enable_histogram_opts, field) + sizeof(opts->field) <= size
10083 :
10084 : #define SET_FIELD(field, value) \
10085 : if (FIELD_OK(field)) { \
10086 : opts->field = value; \
10087 : } \
10088 :
10089 4 : SET_FIELD(io_type, 0);
10090 :
10091 : /* You should not remove this statement, but need to update the assert statement
10092 : * if you add a new field, and also add a corresponding SET_FIELD statement */
10093 : SPDK_STATIC_ASSERT(sizeof(struct spdk_bdev_enable_histogram_opts) == 9, "Incorrect size");
10094 :
10095 : #undef FIELD_OK
10096 : #undef SET_FIELD
10097 4 : }
10098 :
10099 : void
10100 4 : spdk_bdev_histogram_enable(struct spdk_bdev *bdev, spdk_bdev_histogram_status_cb cb_fn,
10101 : void *cb_arg, bool enable)
10102 : {
10103 : struct spdk_bdev_enable_histogram_opts opts;
10104 :
10105 4 : spdk_bdev_enable_histogram_opts_init(&opts, sizeof(opts));
10106 4 : spdk_bdev_histogram_enable_ext(bdev, cb_fn, cb_arg, enable, &opts);
10107 4 : }
10108 :
10109 : struct spdk_bdev_histogram_data_ctx {
10110 : spdk_bdev_histogram_data_cb cb_fn;
10111 : void *cb_arg;
10112 : struct spdk_bdev *bdev;
10113 : /** merged histogram data from all channels */
10114 : struct spdk_histogram_data *histogram;
10115 : };
10116 :
10117 : static void
10118 5 : bdev_histogram_get_channel_cb(struct spdk_bdev *bdev, void *_ctx, int status)
10119 : {
10120 5 : struct spdk_bdev_histogram_data_ctx *ctx = _ctx;
10121 :
10122 5 : ctx->cb_fn(ctx->cb_arg, status, ctx->histogram);
10123 5 : free(ctx);
10124 5 : }
10125 :
10126 : static void
10127 7 : bdev_histogram_get_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
10128 : struct spdk_io_channel *_ch, void *_ctx)
10129 : {
10130 7 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
10131 7 : struct spdk_bdev_histogram_data_ctx *ctx = _ctx;
10132 7 : int status = 0;
10133 :
10134 7 : if (ch->histogram == NULL) {
10135 1 : status = -EFAULT;
10136 1 : } else {
10137 6 : spdk_histogram_data_merge(ctx->histogram, ch->histogram);
10138 : }
10139 :
10140 7 : spdk_bdev_for_each_channel_continue(i, status);
10141 7 : }
10142 :
10143 : void
10144 5 : spdk_bdev_histogram_get(struct spdk_bdev *bdev, struct spdk_histogram_data *histogram,
10145 : spdk_bdev_histogram_data_cb cb_fn,
10146 : void *cb_arg)
10147 : {
10148 : struct spdk_bdev_histogram_data_ctx *ctx;
10149 :
10150 5 : ctx = calloc(1, sizeof(struct spdk_bdev_histogram_data_ctx));
10151 5 : if (ctx == NULL) {
10152 0 : cb_fn(cb_arg, -ENOMEM, NULL);
10153 0 : return;
10154 : }
10155 :
10156 5 : ctx->bdev = bdev;
10157 5 : ctx->cb_fn = cb_fn;
10158 5 : ctx->cb_arg = cb_arg;
10159 :
10160 5 : ctx->histogram = histogram;
10161 :
10162 5 : spdk_bdev_for_each_channel(bdev, bdev_histogram_get_channel, ctx,
10163 : bdev_histogram_get_channel_cb);
10164 5 : }
10165 :
10166 : void
10167 2 : spdk_bdev_channel_get_histogram(struct spdk_io_channel *ch, spdk_bdev_histogram_data_cb cb_fn,
10168 : void *cb_arg)
10169 : {
10170 2 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(ch);
10171 2 : int status = 0;
10172 :
10173 2 : assert(cb_fn != NULL);
10174 :
10175 2 : if (bdev_ch->histogram == NULL) {
10176 1 : status = -EFAULT;
10177 1 : }
10178 2 : cb_fn(cb_arg, status, bdev_ch->histogram);
10179 2 : }
10180 :
10181 : size_t
10182 0 : spdk_bdev_get_media_events(struct spdk_bdev_desc *desc, struct spdk_bdev_media_event *events,
10183 : size_t max_events)
10184 : {
10185 : struct media_event_entry *entry;
10186 0 : size_t num_events = 0;
10187 :
10188 0 : for (; num_events < max_events; ++num_events) {
10189 0 : entry = TAILQ_FIRST(&desc->pending_media_events);
10190 0 : if (entry == NULL) {
10191 0 : break;
10192 : }
10193 :
10194 0 : events[num_events] = entry->event;
10195 0 : TAILQ_REMOVE(&desc->pending_media_events, entry, tailq);
10196 0 : TAILQ_INSERT_TAIL(&desc->free_media_events, entry, tailq);
10197 0 : }
10198 :
10199 0 : return num_events;
10200 : }
10201 :
10202 : int
10203 0 : spdk_bdev_push_media_events(struct spdk_bdev *bdev, const struct spdk_bdev_media_event *events,
10204 : size_t num_events)
10205 : {
10206 : struct spdk_bdev_desc *desc;
10207 : struct media_event_entry *entry;
10208 : size_t event_id;
10209 0 : int rc = 0;
10210 :
10211 0 : assert(bdev->media_events);
10212 :
10213 0 : spdk_spin_lock(&bdev->internal.spinlock);
10214 0 : TAILQ_FOREACH(desc, &bdev->internal.open_descs, link) {
10215 0 : if (desc->write) {
10216 0 : break;
10217 : }
10218 0 : }
10219 :
10220 0 : if (desc == NULL || desc->media_events_buffer == NULL) {
10221 0 : rc = -ENODEV;
10222 0 : goto out;
10223 : }
10224 :
10225 0 : for (event_id = 0; event_id < num_events; ++event_id) {
10226 0 : entry = TAILQ_FIRST(&desc->free_media_events);
10227 0 : if (entry == NULL) {
10228 0 : break;
10229 : }
10230 :
10231 0 : TAILQ_REMOVE(&desc->free_media_events, entry, tailq);
10232 0 : TAILQ_INSERT_TAIL(&desc->pending_media_events, entry, tailq);
10233 0 : entry->event = events[event_id];
10234 0 : }
10235 :
10236 0 : rc = event_id;
10237 : out:
10238 0 : spdk_spin_unlock(&bdev->internal.spinlock);
10239 0 : return rc;
10240 : }
10241 :
10242 : static void
10243 0 : _media_management_notify(void *arg)
10244 : {
10245 0 : struct spdk_bdev_desc *desc = arg;
10246 :
10247 0 : _event_notify(desc, SPDK_BDEV_EVENT_MEDIA_MANAGEMENT);
10248 0 : }
10249 :
10250 : void
10251 0 : spdk_bdev_notify_media_management(struct spdk_bdev *bdev)
10252 : {
10253 : struct spdk_bdev_desc *desc;
10254 :
10255 0 : spdk_spin_lock(&bdev->internal.spinlock);
10256 0 : TAILQ_FOREACH(desc, &bdev->internal.open_descs, link) {
10257 0 : if (!TAILQ_EMPTY(&desc->pending_media_events)) {
10258 0 : event_notify(desc, _media_management_notify);
10259 0 : }
10260 0 : }
10261 0 : spdk_spin_unlock(&bdev->internal.spinlock);
10262 0 : }
10263 :
10264 : struct locked_lba_range_ctx {
10265 : struct lba_range range;
10266 : struct lba_range *current_range;
10267 : struct lba_range *owner_range;
10268 : struct spdk_poller *poller;
10269 : lock_range_cb cb_fn;
10270 : void *cb_arg;
10271 : };
10272 :
10273 : static void
10274 0 : bdev_lock_error_cleanup_cb(struct spdk_bdev *bdev, void *_ctx, int status)
10275 : {
10276 0 : struct locked_lba_range_ctx *ctx = _ctx;
10277 :
10278 0 : ctx->cb_fn(&ctx->range, ctx->cb_arg, -ENOMEM);
10279 0 : free(ctx);
10280 0 : }
10281 :
10282 : static void bdev_unlock_lba_range_get_channel(struct spdk_bdev_channel_iter *i,
10283 : struct spdk_bdev *bdev, struct spdk_io_channel *ch, void *_ctx);
10284 :
10285 : static void
10286 14 : bdev_lock_lba_range_cb(struct spdk_bdev *bdev, void *_ctx, int status)
10287 : {
10288 14 : struct locked_lba_range_ctx *ctx = _ctx;
10289 :
10290 14 : if (status == -ENOMEM) {
10291 : /* One of the channels could not allocate a range object.
10292 : * So we have to go back and clean up any ranges that were
10293 : * allocated successfully before we return error status to
10294 : * the caller. We can reuse the unlock function to do that
10295 : * clean up.
10296 : */
10297 0 : spdk_bdev_for_each_channel(bdev, bdev_unlock_lba_range_get_channel, ctx,
10298 : bdev_lock_error_cleanup_cb);
10299 0 : return;
10300 : }
10301 :
10302 : /* All channels have locked this range and no I/O overlapping the range
10303 : * are outstanding! Set the owner_ch for the range object for the
10304 : * locking channel, so that this channel will know that it is allowed
10305 : * to write to this range.
10306 : */
10307 14 : if (ctx->owner_range != NULL) {
10308 10 : ctx->owner_range->owner_ch = ctx->range.owner_ch;
10309 10 : }
10310 :
10311 14 : ctx->cb_fn(&ctx->range, ctx->cb_arg, status);
10312 :
10313 : /* Don't free the ctx here. Its range is in the bdev's global list of
10314 : * locked ranges still, and will be removed and freed when this range
10315 : * is later unlocked.
10316 : */
10317 14 : }
10318 :
10319 : static int
10320 17 : bdev_lock_lba_range_check_io(void *_i)
10321 : {
10322 17 : struct spdk_bdev_channel_iter *i = _i;
10323 17 : struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i->i);
10324 17 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
10325 17 : struct locked_lba_range_ctx *ctx = i->ctx;
10326 17 : struct lba_range *range = ctx->current_range;
10327 : struct spdk_bdev_io *bdev_io;
10328 :
10329 17 : spdk_poller_unregister(&ctx->poller);
10330 :
10331 : /* The range is now in the locked_ranges, so no new IO can be submitted to this
10332 : * range. But we need to wait until any outstanding IO overlapping with this range
10333 : * are completed.
10334 : */
10335 18 : TAILQ_FOREACH(bdev_io, &ch->io_submitted, internal.ch_link) {
10336 3 : if (bdev_io_range_is_locked(bdev_io, range)) {
10337 2 : ctx->poller = SPDK_POLLER_REGISTER(bdev_lock_lba_range_check_io, i, 100);
10338 2 : return SPDK_POLLER_BUSY;
10339 : }
10340 1 : }
10341 :
10342 15 : spdk_bdev_for_each_channel_continue(i, 0);
10343 15 : return SPDK_POLLER_BUSY;
10344 17 : }
10345 :
10346 : static void
10347 15 : bdev_lock_lba_range_get_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
10348 : struct spdk_io_channel *_ch, void *_ctx)
10349 : {
10350 15 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
10351 15 : struct locked_lba_range_ctx *ctx = _ctx;
10352 : struct lba_range *range;
10353 :
10354 16 : TAILQ_FOREACH(range, &ch->locked_ranges, tailq) {
10355 1 : if (range->length == ctx->range.length &&
10356 0 : range->offset == ctx->range.offset &&
10357 0 : range->locked_ctx == ctx->range.locked_ctx) {
10358 : /* This range already exists on this channel, so don't add
10359 : * it again. This can happen when a new channel is created
10360 : * while the for_each_channel operation is in progress.
10361 : * Do not check for outstanding I/O in that case, since the
10362 : * range was locked before any I/O could be submitted to the
10363 : * new channel.
10364 : */
10365 0 : spdk_bdev_for_each_channel_continue(i, 0);
10366 0 : return;
10367 : }
10368 1 : }
10369 :
10370 15 : range = calloc(1, sizeof(*range));
10371 15 : if (range == NULL) {
10372 0 : spdk_bdev_for_each_channel_continue(i, -ENOMEM);
10373 0 : return;
10374 : }
10375 :
10376 15 : range->length = ctx->range.length;
10377 15 : range->offset = ctx->range.offset;
10378 15 : range->locked_ctx = ctx->range.locked_ctx;
10379 15 : range->quiesce = ctx->range.quiesce;
10380 15 : ctx->current_range = range;
10381 15 : if (ctx->range.owner_ch == ch) {
10382 : /* This is the range object for the channel that will hold
10383 : * the lock. Store it in the ctx object so that we can easily
10384 : * set its owner_ch after the lock is finally acquired.
10385 : */
10386 10 : ctx->owner_range = range;
10387 10 : }
10388 15 : TAILQ_INSERT_TAIL(&ch->locked_ranges, range, tailq);
10389 15 : bdev_lock_lba_range_check_io(i);
10390 15 : }
10391 :
10392 : static void
10393 14 : bdev_lock_lba_range_ctx(struct spdk_bdev *bdev, struct locked_lba_range_ctx *ctx)
10394 : {
10395 14 : assert(spdk_get_thread() == ctx->range.owner_thread);
10396 14 : assert(ctx->range.owner_ch == NULL ||
10397 : spdk_io_channel_get_thread(ctx->range.owner_ch->channel) == ctx->range.owner_thread);
10398 :
10399 : /* We will add a copy of this range to each channel now. */
10400 14 : spdk_bdev_for_each_channel(bdev, bdev_lock_lba_range_get_channel, ctx,
10401 : bdev_lock_lba_range_cb);
10402 14 : }
10403 :
10404 : static bool
10405 17 : bdev_lba_range_overlaps_tailq(struct lba_range *range, lba_range_tailq_t *tailq)
10406 : {
10407 : struct lba_range *r;
10408 :
10409 18 : TAILQ_FOREACH(r, tailq, tailq) {
10410 4 : if (bdev_lba_range_overlapped(range, r)) {
10411 3 : return true;
10412 : }
10413 1 : }
10414 14 : return false;
10415 17 : }
10416 :
10417 : static void bdev_quiesce_range_locked(struct lba_range *range, void *ctx, int status);
10418 :
10419 : static int
10420 14 : _bdev_lock_lba_range(struct spdk_bdev *bdev, struct spdk_bdev_channel *ch,
10421 : uint64_t offset, uint64_t length,
10422 : lock_range_cb cb_fn, void *cb_arg)
10423 : {
10424 : struct locked_lba_range_ctx *ctx;
10425 :
10426 14 : ctx = calloc(1, sizeof(*ctx));
10427 14 : if (ctx == NULL) {
10428 0 : return -ENOMEM;
10429 : }
10430 :
10431 14 : ctx->range.offset = offset;
10432 14 : ctx->range.length = length;
10433 14 : ctx->range.owner_thread = spdk_get_thread();
10434 14 : ctx->range.owner_ch = ch;
10435 14 : ctx->range.locked_ctx = cb_arg;
10436 14 : ctx->range.bdev = bdev;
10437 14 : ctx->range.quiesce = (cb_fn == bdev_quiesce_range_locked);
10438 14 : ctx->cb_fn = cb_fn;
10439 14 : ctx->cb_arg = cb_arg;
10440 :
10441 14 : spdk_spin_lock(&bdev->internal.spinlock);
10442 14 : if (bdev_lba_range_overlaps_tailq(&ctx->range, &bdev->internal.locked_ranges)) {
10443 : /* There is an active lock overlapping with this range.
10444 : * Put it on the pending list until this range no
10445 : * longer overlaps with another.
10446 : */
10447 2 : TAILQ_INSERT_TAIL(&bdev->internal.pending_locked_ranges, &ctx->range, tailq);
10448 2 : } else {
10449 12 : TAILQ_INSERT_TAIL(&bdev->internal.locked_ranges, &ctx->range, tailq);
10450 12 : bdev_lock_lba_range_ctx(bdev, ctx);
10451 : }
10452 14 : spdk_spin_unlock(&bdev->internal.spinlock);
10453 14 : return 0;
10454 14 : }
10455 :
10456 : static int
10457 10 : bdev_lock_lba_range(struct spdk_bdev_desc *desc, struct spdk_io_channel *_ch,
10458 : uint64_t offset, uint64_t length,
10459 : lock_range_cb cb_fn, void *cb_arg)
10460 : {
10461 10 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
10462 10 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
10463 :
10464 10 : if (cb_arg == NULL) {
10465 0 : SPDK_ERRLOG("cb_arg must not be NULL\n");
10466 0 : return -EINVAL;
10467 : }
10468 :
10469 10 : return _bdev_lock_lba_range(bdev, ch, offset, length, cb_fn, cb_arg);
10470 10 : }
10471 :
10472 : static void
10473 2 : bdev_lock_lba_range_ctx_msg(void *_ctx)
10474 : {
10475 2 : struct locked_lba_range_ctx *ctx = _ctx;
10476 :
10477 2 : bdev_lock_lba_range_ctx(ctx->range.bdev, ctx);
10478 2 : }
10479 :
10480 : static void
10481 14 : bdev_unlock_lba_range_cb(struct spdk_bdev *bdev, void *_ctx, int status)
10482 : {
10483 14 : struct locked_lba_range_ctx *ctx = _ctx;
10484 : struct locked_lba_range_ctx *pending_ctx;
10485 : struct lba_range *range, *tmp;
10486 :
10487 14 : spdk_spin_lock(&bdev->internal.spinlock);
10488 : /* Check if there are any pending locked ranges that overlap with this range
10489 : * that was just unlocked. If there are, check that it doesn't overlap with any
10490 : * other locked ranges before calling bdev_lock_lba_range_ctx which will start
10491 : * the lock process.
10492 : */
10493 17 : TAILQ_FOREACH_SAFE(range, &bdev->internal.pending_locked_ranges, tailq, tmp) {
10494 3 : if (bdev_lba_range_overlapped(range, &ctx->range) &&
10495 3 : !bdev_lba_range_overlaps_tailq(range, &bdev->internal.locked_ranges)) {
10496 2 : TAILQ_REMOVE(&bdev->internal.pending_locked_ranges, range, tailq);
10497 2 : pending_ctx = SPDK_CONTAINEROF(range, struct locked_lba_range_ctx, range);
10498 2 : TAILQ_INSERT_TAIL(&bdev->internal.locked_ranges, range, tailq);
10499 4 : spdk_thread_send_msg(pending_ctx->range.owner_thread,
10500 2 : bdev_lock_lba_range_ctx_msg, pending_ctx);
10501 2 : }
10502 3 : }
10503 14 : spdk_spin_unlock(&bdev->internal.spinlock);
10504 :
10505 14 : ctx->cb_fn(&ctx->range, ctx->cb_arg, status);
10506 14 : free(ctx);
10507 14 : }
10508 :
10509 : static void
10510 16 : bdev_unlock_lba_range_get_channel(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
10511 : struct spdk_io_channel *_ch, void *_ctx)
10512 : {
10513 16 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
10514 16 : struct locked_lba_range_ctx *ctx = _ctx;
10515 : TAILQ_HEAD(, spdk_bdev_io) io_locked;
10516 : struct spdk_bdev_io *bdev_io;
10517 : struct lba_range *range;
10518 :
10519 16 : TAILQ_FOREACH(range, &ch->locked_ranges, tailq) {
10520 32 : if (ctx->range.offset == range->offset &&
10521 16 : ctx->range.length == range->length &&
10522 16 : ctx->range.locked_ctx == range->locked_ctx) {
10523 16 : TAILQ_REMOVE(&ch->locked_ranges, range, tailq);
10524 16 : free(range);
10525 16 : break;
10526 : }
10527 0 : }
10528 :
10529 : /* Note: we should almost always be able to assert that the range specified
10530 : * was found. But there are some very rare corner cases where a new channel
10531 : * gets created simultaneously with a range unlock, where this function
10532 : * would execute on that new channel and wouldn't have the range.
10533 : * We also use this to clean up range allocations when a later allocation
10534 : * fails in the locking path.
10535 : * So we can't actually assert() here.
10536 : */
10537 :
10538 : /* Swap the locked IO into a temporary list, and then try to submit them again.
10539 : * We could hyper-optimize this to only resubmit locked I/O that overlap
10540 : * with the range that was just unlocked, but this isn't a performance path so
10541 : * we go for simplicity here.
10542 : */
10543 16 : TAILQ_INIT(&io_locked);
10544 16 : TAILQ_SWAP(&ch->io_locked, &io_locked, spdk_bdev_io, internal.ch_link);
10545 19 : while (!TAILQ_EMPTY(&io_locked)) {
10546 3 : bdev_io = TAILQ_FIRST(&io_locked);
10547 3 : TAILQ_REMOVE(&io_locked, bdev_io, internal.ch_link);
10548 3 : bdev_io_submit(bdev_io);
10549 : }
10550 :
10551 16 : spdk_bdev_for_each_channel_continue(i, 0);
10552 16 : }
10553 :
10554 : static int
10555 14 : _bdev_unlock_lba_range(struct spdk_bdev *bdev, uint64_t offset, uint64_t length,
10556 : lock_range_cb cb_fn, void *cb_arg)
10557 : {
10558 : struct locked_lba_range_ctx *ctx;
10559 : struct lba_range *range;
10560 :
10561 14 : spdk_spin_lock(&bdev->internal.spinlock);
10562 : /* To start the unlock the process, we find the range in the bdev's locked_ranges
10563 : * and remove it. This ensures new channels don't inherit the locked range.
10564 : * Then we will send a message to each channel to remove the range from its
10565 : * per-channel list.
10566 : */
10567 14 : TAILQ_FOREACH(range, &bdev->internal.locked_ranges, tailq) {
10568 24 : if (range->offset == offset && range->length == length &&
10569 14 : (range->owner_ch == NULL || range->locked_ctx == cb_arg)) {
10570 14 : break;
10571 : }
10572 0 : }
10573 14 : if (range == NULL) {
10574 0 : assert(false);
10575 : spdk_spin_unlock(&bdev->internal.spinlock);
10576 : return -EINVAL;
10577 : }
10578 14 : TAILQ_REMOVE(&bdev->internal.locked_ranges, range, tailq);
10579 14 : ctx = SPDK_CONTAINEROF(range, struct locked_lba_range_ctx, range);
10580 14 : spdk_spin_unlock(&bdev->internal.spinlock);
10581 :
10582 14 : ctx->cb_fn = cb_fn;
10583 14 : ctx->cb_arg = cb_arg;
10584 :
10585 14 : spdk_bdev_for_each_channel(bdev, bdev_unlock_lba_range_get_channel, ctx,
10586 : bdev_unlock_lba_range_cb);
10587 14 : return 0;
10588 : }
10589 :
10590 : static int
10591 12 : bdev_unlock_lba_range(struct spdk_bdev_desc *desc, struct spdk_io_channel *_ch,
10592 : uint64_t offset, uint64_t length,
10593 : lock_range_cb cb_fn, void *cb_arg)
10594 : {
10595 12 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
10596 12 : struct spdk_bdev_channel *ch = __io_ch_to_bdev_ch(_ch);
10597 : struct lba_range *range;
10598 12 : bool range_found = false;
10599 :
10600 : /* Let's make sure the specified channel actually has a lock on
10601 : * the specified range. Note that the range must match exactly.
10602 : */
10603 14 : TAILQ_FOREACH(range, &ch->locked_ranges, tailq) {
10604 22 : if (range->offset == offset && range->length == length &&
10605 11 : range->owner_ch == ch && range->locked_ctx == cb_arg) {
10606 10 : range_found = true;
10607 10 : break;
10608 : }
10609 2 : }
10610 :
10611 12 : if (!range_found) {
10612 2 : return -EINVAL;
10613 : }
10614 :
10615 10 : return _bdev_unlock_lba_range(bdev, offset, length, cb_fn, cb_arg);
10616 12 : }
10617 :
10618 : struct bdev_quiesce_ctx {
10619 : spdk_bdev_quiesce_cb cb_fn;
10620 : void *cb_arg;
10621 : };
10622 :
10623 : static void
10624 4 : bdev_unquiesce_range_unlocked(struct lba_range *range, void *ctx, int status)
10625 : {
10626 4 : struct bdev_quiesce_ctx *quiesce_ctx = ctx;
10627 :
10628 4 : if (quiesce_ctx->cb_fn != NULL) {
10629 4 : quiesce_ctx->cb_fn(quiesce_ctx->cb_arg, status);
10630 4 : }
10631 :
10632 4 : free(quiesce_ctx);
10633 4 : }
10634 :
10635 : static void
10636 4 : bdev_quiesce_range_locked(struct lba_range *range, void *ctx, int status)
10637 : {
10638 4 : struct bdev_quiesce_ctx *quiesce_ctx = ctx;
10639 4 : struct spdk_bdev_module *module = range->bdev->module;
10640 :
10641 4 : if (status != 0) {
10642 0 : if (quiesce_ctx->cb_fn != NULL) {
10643 0 : quiesce_ctx->cb_fn(quiesce_ctx->cb_arg, status);
10644 0 : }
10645 0 : free(quiesce_ctx);
10646 0 : return;
10647 : }
10648 :
10649 4 : spdk_spin_lock(&module->internal.spinlock);
10650 4 : TAILQ_INSERT_TAIL(&module->internal.quiesced_ranges, range, tailq_module);
10651 4 : spdk_spin_unlock(&module->internal.spinlock);
10652 :
10653 4 : if (quiesce_ctx->cb_fn != NULL) {
10654 : /* copy the context in case the range is unlocked by the callback */
10655 4 : struct bdev_quiesce_ctx tmp = *quiesce_ctx;
10656 :
10657 4 : quiesce_ctx->cb_fn = NULL;
10658 4 : quiesce_ctx->cb_arg = NULL;
10659 :
10660 4 : tmp.cb_fn(tmp.cb_arg, status);
10661 4 : }
10662 : /* quiesce_ctx will be freed on unquiesce */
10663 4 : }
10664 :
10665 : static int
10666 9 : _spdk_bdev_quiesce(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
10667 : uint64_t offset, uint64_t length,
10668 : spdk_bdev_quiesce_cb cb_fn, void *cb_arg,
10669 : bool unquiesce)
10670 : {
10671 : struct bdev_quiesce_ctx *quiesce_ctx;
10672 : int rc;
10673 :
10674 9 : if (module != bdev->module) {
10675 0 : SPDK_ERRLOG("Bdev does not belong to specified module.\n");
10676 0 : return -EINVAL;
10677 : }
10678 :
10679 9 : if (!bdev_io_valid_blocks(bdev, offset, length)) {
10680 0 : return -EINVAL;
10681 : }
10682 :
10683 9 : if (unquiesce) {
10684 : struct lba_range *range;
10685 :
10686 : /* Make sure the specified range is actually quiesced in the specified module and
10687 : * then remove it from the list. Note that the range must match exactly.
10688 : */
10689 5 : spdk_spin_lock(&module->internal.spinlock);
10690 6 : TAILQ_FOREACH(range, &module->internal.quiesced_ranges, tailq_module) {
10691 5 : if (range->bdev == bdev && range->offset == offset && range->length == length) {
10692 4 : TAILQ_REMOVE(&module->internal.quiesced_ranges, range, tailq_module);
10693 4 : break;
10694 : }
10695 1 : }
10696 5 : spdk_spin_unlock(&module->internal.spinlock);
10697 :
10698 5 : if (range == NULL) {
10699 1 : SPDK_ERRLOG("The range to unquiesce was not found.\n");
10700 1 : return -EINVAL;
10701 : }
10702 :
10703 4 : quiesce_ctx = range->locked_ctx;
10704 4 : quiesce_ctx->cb_fn = cb_fn;
10705 4 : quiesce_ctx->cb_arg = cb_arg;
10706 :
10707 4 : rc = _bdev_unlock_lba_range(bdev, offset, length, bdev_unquiesce_range_unlocked, quiesce_ctx);
10708 4 : } else {
10709 4 : quiesce_ctx = malloc(sizeof(*quiesce_ctx));
10710 4 : if (quiesce_ctx == NULL) {
10711 0 : return -ENOMEM;
10712 : }
10713 :
10714 4 : quiesce_ctx->cb_fn = cb_fn;
10715 4 : quiesce_ctx->cb_arg = cb_arg;
10716 :
10717 4 : rc = _bdev_lock_lba_range(bdev, NULL, offset, length, bdev_quiesce_range_locked, quiesce_ctx);
10718 4 : if (rc != 0) {
10719 0 : free(quiesce_ctx);
10720 0 : }
10721 : }
10722 :
10723 8 : return rc;
10724 9 : }
10725 :
10726 : int
10727 3 : spdk_bdev_quiesce(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
10728 : spdk_bdev_quiesce_cb cb_fn, void *cb_arg)
10729 : {
10730 3 : return _spdk_bdev_quiesce(bdev, module, 0, bdev->blockcnt, cb_fn, cb_arg, false);
10731 : }
10732 :
10733 : int
10734 3 : spdk_bdev_unquiesce(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
10735 : spdk_bdev_quiesce_cb cb_fn, void *cb_arg)
10736 : {
10737 3 : return _spdk_bdev_quiesce(bdev, module, 0, bdev->blockcnt, cb_fn, cb_arg, true);
10738 : }
10739 :
10740 : int
10741 1 : spdk_bdev_quiesce_range(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
10742 : uint64_t offset, uint64_t length,
10743 : spdk_bdev_quiesce_cb cb_fn, void *cb_arg)
10744 : {
10745 1 : return _spdk_bdev_quiesce(bdev, module, offset, length, cb_fn, cb_arg, false);
10746 : }
10747 :
10748 : int
10749 2 : spdk_bdev_unquiesce_range(struct spdk_bdev *bdev, struct spdk_bdev_module *module,
10750 : uint64_t offset, uint64_t length,
10751 : spdk_bdev_quiesce_cb cb_fn, void *cb_arg)
10752 : {
10753 2 : return _spdk_bdev_quiesce(bdev, module, offset, length, cb_fn, cb_arg, true);
10754 : }
10755 :
10756 : int
10757 283 : spdk_bdev_get_memory_domains(struct spdk_bdev *bdev, struct spdk_memory_domain **domains,
10758 : int array_size)
10759 : {
10760 283 : if (!bdev) {
10761 1 : return -EINVAL;
10762 : }
10763 :
10764 282 : if (bdev->fn_table->get_memory_domains) {
10765 3 : return bdev->fn_table->get_memory_domains(bdev->ctxt, domains, array_size);
10766 : }
10767 :
10768 279 : return 0;
10769 283 : }
10770 :
10771 : struct spdk_bdev_for_each_io_ctx {
10772 : void *ctx;
10773 : spdk_bdev_io_fn fn;
10774 : spdk_bdev_for_each_io_cb cb;
10775 : };
10776 :
10777 : static void
10778 0 : bdev_channel_for_each_io(struct spdk_bdev_channel_iter *i, struct spdk_bdev *bdev,
10779 : struct spdk_io_channel *io_ch, void *_ctx)
10780 : {
10781 0 : struct spdk_bdev_for_each_io_ctx *ctx = _ctx;
10782 0 : struct spdk_bdev_channel *bdev_ch = __io_ch_to_bdev_ch(io_ch);
10783 : struct spdk_bdev_io *bdev_io;
10784 0 : int rc = 0;
10785 :
10786 0 : TAILQ_FOREACH(bdev_io, &bdev_ch->io_submitted, internal.ch_link) {
10787 0 : rc = ctx->fn(ctx->ctx, bdev_io);
10788 0 : if (rc != 0) {
10789 0 : break;
10790 : }
10791 0 : }
10792 :
10793 0 : spdk_bdev_for_each_channel_continue(i, rc);
10794 0 : }
10795 :
10796 : static void
10797 0 : bdev_for_each_io_done(struct spdk_bdev *bdev, void *_ctx, int status)
10798 : {
10799 0 : struct spdk_bdev_for_each_io_ctx *ctx = _ctx;
10800 :
10801 0 : ctx->cb(ctx->ctx, status);
10802 :
10803 0 : free(ctx);
10804 0 : }
10805 :
10806 : void
10807 0 : spdk_bdev_for_each_bdev_io(struct spdk_bdev *bdev, void *_ctx, spdk_bdev_io_fn fn,
10808 : spdk_bdev_for_each_io_cb cb)
10809 : {
10810 : struct spdk_bdev_for_each_io_ctx *ctx;
10811 :
10812 0 : assert(fn != NULL && cb != NULL);
10813 :
10814 0 : ctx = calloc(1, sizeof(*ctx));
10815 0 : if (ctx == NULL) {
10816 0 : SPDK_ERRLOG("Failed to allocate context.\n");
10817 0 : cb(_ctx, -ENOMEM);
10818 0 : return;
10819 : }
10820 :
10821 0 : ctx->ctx = _ctx;
10822 0 : ctx->fn = fn;
10823 0 : ctx->cb = cb;
10824 :
10825 0 : spdk_bdev_for_each_channel(bdev, bdev_channel_for_each_io, ctx,
10826 : bdev_for_each_io_done);
10827 0 : }
10828 :
10829 : void
10830 135 : spdk_bdev_for_each_channel_continue(struct spdk_bdev_channel_iter *iter, int status)
10831 : {
10832 135 : spdk_for_each_channel_continue(iter->i, status);
10833 135 : }
10834 :
10835 : static struct spdk_bdev *
10836 371 : io_channel_iter_get_bdev(struct spdk_io_channel_iter *i)
10837 : {
10838 371 : void *io_device = spdk_io_channel_iter_get_io_device(i);
10839 :
10840 371 : return __bdev_from_io_dev(io_device);
10841 : }
10842 :
10843 : static void
10844 135 : bdev_each_channel_msg(struct spdk_io_channel_iter *i)
10845 : {
10846 135 : struct spdk_bdev_channel_iter *iter = spdk_io_channel_iter_get_ctx(i);
10847 135 : struct spdk_bdev *bdev = io_channel_iter_get_bdev(i);
10848 135 : struct spdk_io_channel *ch = spdk_io_channel_iter_get_channel(i);
10849 :
10850 135 : iter->i = i;
10851 135 : iter->fn(iter, bdev, ch, iter->ctx);
10852 135 : }
10853 :
10854 : static void
10855 236 : bdev_each_channel_cpl(struct spdk_io_channel_iter *i, int status)
10856 : {
10857 236 : struct spdk_bdev_channel_iter *iter = spdk_io_channel_iter_get_ctx(i);
10858 236 : struct spdk_bdev *bdev = io_channel_iter_get_bdev(i);
10859 :
10860 236 : iter->i = i;
10861 236 : iter->cpl(bdev, iter->ctx, status);
10862 :
10863 236 : free(iter);
10864 236 : }
10865 :
10866 : void
10867 236 : spdk_bdev_for_each_channel(struct spdk_bdev *bdev, spdk_bdev_for_each_channel_msg fn,
10868 : void *ctx, spdk_bdev_for_each_channel_done cpl)
10869 : {
10870 : struct spdk_bdev_channel_iter *iter;
10871 :
10872 236 : assert(bdev != NULL && fn != NULL && ctx != NULL);
10873 :
10874 236 : iter = calloc(1, sizeof(struct spdk_bdev_channel_iter));
10875 236 : if (iter == NULL) {
10876 0 : SPDK_ERRLOG("Unable to allocate iterator\n");
10877 0 : assert(false);
10878 : return;
10879 : }
10880 :
10881 236 : iter->fn = fn;
10882 236 : iter->cpl = cpl;
10883 236 : iter->ctx = ctx;
10884 :
10885 472 : spdk_for_each_channel(__bdev_to_io_dev(bdev), bdev_each_channel_msg,
10886 236 : iter, bdev_each_channel_cpl);
10887 236 : }
10888 :
10889 : static void
10890 3 : bdev_copy_do_write_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
10891 : {
10892 3 : struct spdk_bdev_io *parent_io = cb_arg;
10893 :
10894 3 : spdk_bdev_free_io(bdev_io);
10895 :
10896 : /* Check return status of write */
10897 3 : parent_io->internal.status = success ? SPDK_BDEV_IO_STATUS_SUCCESS : SPDK_BDEV_IO_STATUS_FAILED;
10898 3 : parent_io->internal.cb(parent_io, success, parent_io->internal.caller_ctx);
10899 3 : }
10900 :
10901 : static void
10902 3 : bdev_copy_do_write(void *_bdev_io)
10903 : {
10904 3 : struct spdk_bdev_io *bdev_io = _bdev_io;
10905 : int rc;
10906 :
10907 : /* Write blocks */
10908 6 : rc = spdk_bdev_write_blocks_with_md(bdev_io->internal.desc,
10909 3 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
10910 3 : bdev_io->u.bdev.iovs[0].iov_base,
10911 3 : bdev_io->u.bdev.md_buf, bdev_io->u.bdev.offset_blocks,
10912 3 : bdev_io->u.bdev.num_blocks, bdev_copy_do_write_done, bdev_io);
10913 :
10914 3 : if (rc == -ENOMEM) {
10915 0 : bdev_queue_io_wait_with_cb(bdev_io, bdev_copy_do_write);
10916 3 : } else if (rc != 0) {
10917 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
10918 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
10919 0 : }
10920 3 : }
10921 :
10922 : static void
10923 3 : bdev_copy_do_read_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
10924 : {
10925 3 : struct spdk_bdev_io *parent_io = cb_arg;
10926 :
10927 3 : spdk_bdev_free_io(bdev_io);
10928 :
10929 : /* Check return status of read */
10930 3 : if (!success) {
10931 0 : parent_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
10932 0 : parent_io->internal.cb(parent_io, false, parent_io->internal.caller_ctx);
10933 0 : return;
10934 : }
10935 :
10936 : /* Do write */
10937 3 : bdev_copy_do_write(parent_io);
10938 3 : }
10939 :
10940 : static void
10941 3 : bdev_copy_do_read(void *_bdev_io)
10942 : {
10943 3 : struct spdk_bdev_io *bdev_io = _bdev_io;
10944 : int rc;
10945 :
10946 : /* Read blocks */
10947 6 : rc = spdk_bdev_read_blocks_with_md(bdev_io->internal.desc,
10948 3 : spdk_io_channel_from_ctx(bdev_io->internal.ch),
10949 3 : bdev_io->u.bdev.iovs[0].iov_base,
10950 3 : bdev_io->u.bdev.md_buf, bdev_io->u.bdev.copy.src_offset_blocks,
10951 3 : bdev_io->u.bdev.num_blocks, bdev_copy_do_read_done, bdev_io);
10952 :
10953 3 : if (rc == -ENOMEM) {
10954 0 : bdev_queue_io_wait_with_cb(bdev_io, bdev_copy_do_read);
10955 3 : } else if (rc != 0) {
10956 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
10957 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
10958 0 : }
10959 3 : }
10960 :
10961 : static void
10962 3 : bdev_copy_get_buf_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io, bool success)
10963 : {
10964 3 : if (!success) {
10965 0 : bdev_io->internal.status = SPDK_BDEV_IO_STATUS_FAILED;
10966 0 : bdev_io->internal.cb(bdev_io, false, bdev_io->internal.caller_ctx);
10967 0 : return;
10968 : }
10969 :
10970 3 : bdev_copy_do_read(bdev_io);
10971 3 : }
10972 :
10973 : int
10974 27 : spdk_bdev_copy_blocks(struct spdk_bdev_desc *desc, struct spdk_io_channel *ch,
10975 : uint64_t dst_offset_blocks, uint64_t src_offset_blocks, uint64_t num_blocks,
10976 : spdk_bdev_io_completion_cb cb, void *cb_arg)
10977 : {
10978 27 : struct spdk_bdev *bdev = spdk_bdev_desc_get_bdev(desc);
10979 : struct spdk_bdev_io *bdev_io;
10980 27 : struct spdk_bdev_channel *channel = spdk_io_channel_get_ctx(ch);
10981 :
10982 27 : if (!desc->write) {
10983 0 : return -EBADF;
10984 : }
10985 :
10986 27 : if (!bdev_io_valid_blocks(bdev, dst_offset_blocks, num_blocks) ||
10987 27 : !bdev_io_valid_blocks(bdev, src_offset_blocks, num_blocks)) {
10988 0 : SPDK_DEBUGLOG(bdev,
10989 : "Invalid offset or number of blocks: dst %lu, src %lu, count %lu\n",
10990 : dst_offset_blocks, src_offset_blocks, num_blocks);
10991 0 : return -EINVAL;
10992 : }
10993 :
10994 27 : bdev_io = bdev_channel_get_io(channel);
10995 27 : if (!bdev_io) {
10996 0 : return -ENOMEM;
10997 : }
10998 :
10999 27 : bdev_io->internal.ch = channel;
11000 27 : bdev_io->internal.desc = desc;
11001 27 : bdev_io->type = SPDK_BDEV_IO_TYPE_COPY;
11002 :
11003 27 : bdev_io->u.bdev.offset_blocks = dst_offset_blocks;
11004 27 : bdev_io->u.bdev.copy.src_offset_blocks = src_offset_blocks;
11005 27 : bdev_io->u.bdev.num_blocks = num_blocks;
11006 27 : bdev_io->u.bdev.memory_domain = NULL;
11007 27 : bdev_io->u.bdev.memory_domain_ctx = NULL;
11008 27 : bdev_io->u.bdev.iovs = NULL;
11009 27 : bdev_io->u.bdev.iovcnt = 0;
11010 27 : bdev_io->u.bdev.md_buf = NULL;
11011 27 : bdev_io->u.bdev.accel_sequence = NULL;
11012 27 : bdev_io_init(bdev_io, bdev, cb_arg, cb);
11013 :
11014 27 : if (dst_offset_blocks == src_offset_blocks || num_blocks == 0) {
11015 0 : spdk_thread_send_msg(spdk_get_thread(), bdev_io_complete_cb, bdev_io);
11016 0 : return 0;
11017 : }
11018 :
11019 :
11020 : /* If the copy size is large and should be split, use the generic split logic
11021 : * regardless of whether SPDK_BDEV_IO_TYPE_COPY is supported or not.
11022 : *
11023 : * Then, send the copy request if SPDK_BDEV_IO_TYPE_COPY is supported or
11024 : * emulate it using regular read and write requests otherwise.
11025 : */
11026 27 : if (spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_COPY) ||
11027 4 : bdev_io->internal.f.split) {
11028 24 : bdev_io_submit(bdev_io);
11029 24 : return 0;
11030 : }
11031 :
11032 3 : spdk_bdev_io_get_buf(bdev_io, bdev_copy_get_buf_cb, num_blocks * spdk_bdev_get_block_size(bdev));
11033 :
11034 3 : return 0;
11035 27 : }
11036 :
11037 3 : SPDK_LOG_REGISTER_COMPONENT(bdev)
11038 :
11039 : static void
11040 0 : bdev_trace(void)
11041 : {
11042 0 : struct spdk_trace_tpoint_opts opts[] = {
11043 : {
11044 : "BDEV_IO_START", TRACE_BDEV_IO_START,
11045 : OWNER_TYPE_BDEV, OBJECT_BDEV_IO, 1,
11046 : {
11047 : { "type", SPDK_TRACE_ARG_TYPE_INT, 8 },
11048 : { "ctx", SPDK_TRACE_ARG_TYPE_PTR, 8 },
11049 : { "offset", SPDK_TRACE_ARG_TYPE_INT, 8 },
11050 : { "qd", SPDK_TRACE_ARG_TYPE_INT, 4 }
11051 : }
11052 : },
11053 : {
11054 : "BDEV_IO_DONE", TRACE_BDEV_IO_DONE,
11055 : OWNER_TYPE_BDEV, OBJECT_BDEV_IO, 0,
11056 : {
11057 : { "ctx", SPDK_TRACE_ARG_TYPE_PTR, 8 },
11058 : { "qd", SPDK_TRACE_ARG_TYPE_INT, 4 }
11059 : }
11060 : },
11061 : {
11062 : "BDEV_IOCH_CREATE", TRACE_BDEV_IOCH_CREATE,
11063 : OWNER_TYPE_BDEV, OBJECT_NONE, 0,
11064 : {
11065 : { "tid", SPDK_TRACE_ARG_TYPE_INT, 8 }
11066 : }
11067 : },
11068 : {
11069 : "BDEV_IOCH_DESTROY", TRACE_BDEV_IOCH_DESTROY,
11070 : OWNER_TYPE_BDEV, OBJECT_NONE, 0,
11071 : {
11072 : { "tid", SPDK_TRACE_ARG_TYPE_INT, 8 }
11073 : }
11074 : },
11075 : };
11076 :
11077 :
11078 0 : spdk_trace_register_owner_type(OWNER_TYPE_BDEV, 'b');
11079 0 : spdk_trace_register_object(OBJECT_BDEV_IO, 'i');
11080 0 : spdk_trace_register_description_ext(opts, SPDK_COUNTOF(opts));
11081 0 : spdk_trace_tpoint_register_relation(TRACE_BDEV_NVME_IO_START, OBJECT_BDEV_IO, 0);
11082 0 : spdk_trace_tpoint_register_relation(TRACE_BDEV_NVME_IO_DONE, OBJECT_BDEV_IO, 0);
11083 0 : spdk_trace_tpoint_register_relation(TRACE_BLOB_REQ_SET_START, OBJECT_BDEV_IO, 0);
11084 0 : spdk_trace_tpoint_register_relation(TRACE_BLOB_REQ_SET_COMPLETE, OBJECT_BDEV_IO, 0);
11085 0 : spdk_trace_tpoint_register_relation(TRACE_BDEV_RAID_IO_START, OBJECT_BDEV_IO, 0);
11086 0 : spdk_trace_tpoint_register_relation(TRACE_BDEV_RAID_IO_DONE, OBJECT_BDEV_IO, 0);
11087 0 : }
11088 3 : SPDK_TRACE_REGISTER_FN(bdev_trace, "bdev", TRACE_GROUP_BDEV)
|