Commit | Line | Data |
---|---|---|
3dcf60bc | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 | 2 | /* |
1da177e4 LT |
3 | * Copyright (C) 1991, 1992 Linus Torvalds |
4 | * Copyright (C) 1994, Karl Keyte: Added support for disk statistics | |
5 | * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE | |
6 | * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de> | |
6728cb0e JA |
7 | * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au> |
8 | * - July2000 | |
1da177e4 LT |
9 | * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001 |
10 | */ | |
11 | ||
12 | /* | |
13 | * This handles all read/write requests to block devices | |
14 | */ | |
1da177e4 LT |
15 | #include <linux/kernel.h> |
16 | #include <linux/module.h> | |
1da177e4 LT |
17 | #include <linux/bio.h> |
18 | #include <linux/blkdev.h> | |
52abca64 | 19 | #include <linux/blk-pm.h> |
fe45e630 | 20 | #include <linux/blk-integrity.h> |
1da177e4 LT |
21 | #include <linux/highmem.h> |
22 | #include <linux/mm.h> | |
cee9a0c4 | 23 | #include <linux/pagemap.h> |
1da177e4 LT |
24 | #include <linux/kernel_stat.h> |
25 | #include <linux/string.h> | |
26 | #include <linux/init.h> | |
1da177e4 LT |
27 | #include <linux/completion.h> |
28 | #include <linux/slab.h> | |
29 | #include <linux/swap.h> | |
30 | #include <linux/writeback.h> | |
faccbd4b | 31 | #include <linux/task_io_accounting_ops.h> |
c17bb495 | 32 | #include <linux/fault-inject.h> |
73c10101 | 33 | #include <linux/list_sort.h> |
e3c78ca5 | 34 | #include <linux/delay.h> |
aaf7c680 | 35 | #include <linux/ratelimit.h> |
6c954667 | 36 | #include <linux/pm_runtime.h> |
54d4e6ab | 37 | #include <linux/t10-pi.h> |
18fbda91 | 38 | #include <linux/debugfs.h> |
30abb3a6 | 39 | #include <linux/bpf.h> |
b8e24a93 | 40 | #include <linux/psi.h> |
82d981d4 | 41 | #include <linux/part_stat.h> |
71ac860a | 42 | #include <linux/sched/sysctl.h> |
a892c8d5 | 43 | #include <linux/blk-crypto.h> |
55782138 LZ |
44 | |
45 | #define CREATE_TRACE_POINTS | |
46 | #include <trace/events/block.h> | |
1da177e4 | 47 | |
8324aa91 | 48 | #include "blk.h" |
2aa7745b | 49 | #include "blk-mq-sched.h" |
bca6b067 | 50 | #include "blk-pm.h" |
672fdcf0 | 51 | #include "blk-cgroup.h" |
a7b36ee6 | 52 | #include "blk-throttle.h" |
8324aa91 | 53 | |
18fbda91 | 54 | struct dentry *blk_debugfs_root; |
18fbda91 | 55 | |
d07335e5 | 56 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap); |
b0da3f0d | 57 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap); |
0a82a8d1 | 58 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete); |
3291fa57 | 59 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_split); |
cbae8d45 | 60 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug); |
b357e4a6 | 61 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_insert); |
0bfc2455 | 62 | |
a73f730d TH |
63 | DEFINE_IDA(blk_queue_ida); |
64 | ||
1da177e4 LT |
65 | /* |
66 | * For queue allocation | |
67 | */ | |
6728cb0e | 68 | struct kmem_cache *blk_requestq_cachep; |
704b914f | 69 | struct kmem_cache *blk_requestq_srcu_cachep; |
1da177e4 | 70 | |
1da177e4 LT |
71 | /* |
72 | * Controlling structure to kblockd | |
73 | */ | |
ff856bad | 74 | static struct workqueue_struct *kblockd_workqueue; |
1da177e4 | 75 | |
8814ce8a BVA |
76 | /** |
77 | * blk_queue_flag_set - atomically set a queue flag | |
78 | * @flag: flag to be set | |
79 | * @q: request queue | |
80 | */ | |
81 | void blk_queue_flag_set(unsigned int flag, struct request_queue *q) | |
82 | { | |
57d74df9 | 83 | set_bit(flag, &q->queue_flags); |
8814ce8a BVA |
84 | } |
85 | EXPORT_SYMBOL(blk_queue_flag_set); | |
86 | ||
87 | /** | |
88 | * blk_queue_flag_clear - atomically clear a queue flag | |
89 | * @flag: flag to be cleared | |
90 | * @q: request queue | |
91 | */ | |
92 | void blk_queue_flag_clear(unsigned int flag, struct request_queue *q) | |
93 | { | |
57d74df9 | 94 | clear_bit(flag, &q->queue_flags); |
8814ce8a BVA |
95 | } |
96 | EXPORT_SYMBOL(blk_queue_flag_clear); | |
97 | ||
98 | /** | |
99 | * blk_queue_flag_test_and_set - atomically test and set a queue flag | |
100 | * @flag: flag to be set | |
101 | * @q: request queue | |
102 | * | |
103 | * Returns the previous value of @flag - 0 if the flag was not set and 1 if | |
104 | * the flag was already set. | |
105 | */ | |
106 | bool blk_queue_flag_test_and_set(unsigned int flag, struct request_queue *q) | |
107 | { | |
57d74df9 | 108 | return test_and_set_bit(flag, &q->queue_flags); |
8814ce8a BVA |
109 | } |
110 | EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set); | |
111 | ||
e47bc4ed CK |
112 | #define REQ_OP_NAME(name) [REQ_OP_##name] = #name |
113 | static const char *const blk_op_name[] = { | |
114 | REQ_OP_NAME(READ), | |
115 | REQ_OP_NAME(WRITE), | |
116 | REQ_OP_NAME(FLUSH), | |
117 | REQ_OP_NAME(DISCARD), | |
118 | REQ_OP_NAME(SECURE_ERASE), | |
119 | REQ_OP_NAME(ZONE_RESET), | |
6e33dbf2 | 120 | REQ_OP_NAME(ZONE_RESET_ALL), |
6c1b1da5 AJ |
121 | REQ_OP_NAME(ZONE_OPEN), |
122 | REQ_OP_NAME(ZONE_CLOSE), | |
123 | REQ_OP_NAME(ZONE_FINISH), | |
0512a75b | 124 | REQ_OP_NAME(ZONE_APPEND), |
e47bc4ed | 125 | REQ_OP_NAME(WRITE_ZEROES), |
e47bc4ed CK |
126 | REQ_OP_NAME(DRV_IN), |
127 | REQ_OP_NAME(DRV_OUT), | |
128 | }; | |
129 | #undef REQ_OP_NAME | |
130 | ||
131 | /** | |
132 | * blk_op_str - Return string XXX in the REQ_OP_XXX. | |
133 | * @op: REQ_OP_XXX. | |
134 | * | |
135 | * Description: Centralize block layer function to convert REQ_OP_XXX into | |
136 | * string format. Useful in the debugging and tracing bio or request. For | |
137 | * invalid REQ_OP_XXX it returns string "UNKNOWN". | |
138 | */ | |
139 | inline const char *blk_op_str(unsigned int op) | |
140 | { | |
141 | const char *op_str = "UNKNOWN"; | |
142 | ||
143 | if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op]) | |
144 | op_str = blk_op_name[op]; | |
145 | ||
146 | return op_str; | |
147 | } | |
148 | EXPORT_SYMBOL_GPL(blk_op_str); | |
149 | ||
2a842aca CH |
150 | static const struct { |
151 | int errno; | |
152 | const char *name; | |
153 | } blk_errors[] = { | |
154 | [BLK_STS_OK] = { 0, "" }, | |
155 | [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" }, | |
156 | [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" }, | |
157 | [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" }, | |
158 | [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" }, | |
159 | [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" }, | |
160 | [BLK_STS_NEXUS] = { -EBADE, "critical nexus" }, | |
161 | [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" }, | |
162 | [BLK_STS_PROTECTION] = { -EILSEQ, "protection" }, | |
163 | [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" }, | |
86ff7c2a | 164 | [BLK_STS_DEV_RESOURCE] = { -EBUSY, "device resource" }, |
03a07c92 | 165 | [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" }, |
7d32c027 | 166 | [BLK_STS_OFFLINE] = { -ENODEV, "device offline" }, |
2a842aca | 167 | |
4e4cbee9 CH |
168 | /* device mapper special case, should not leak out: */ |
169 | [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" }, | |
170 | ||
3b481d91 KB |
171 | /* zone device specific errors */ |
172 | [BLK_STS_ZONE_OPEN_RESOURCE] = { -ETOOMANYREFS, "open zones exceeded" }, | |
173 | [BLK_STS_ZONE_ACTIVE_RESOURCE] = { -EOVERFLOW, "active zones exceeded" }, | |
174 | ||
2a842aca CH |
175 | /* everything else not covered above: */ |
176 | [BLK_STS_IOERR] = { -EIO, "I/O" }, | |
177 | }; | |
178 | ||
179 | blk_status_t errno_to_blk_status(int errno) | |
180 | { | |
181 | int i; | |
182 | ||
183 | for (i = 0; i < ARRAY_SIZE(blk_errors); i++) { | |
184 | if (blk_errors[i].errno == errno) | |
185 | return (__force blk_status_t)i; | |
186 | } | |
187 | ||
188 | return BLK_STS_IOERR; | |
189 | } | |
190 | EXPORT_SYMBOL_GPL(errno_to_blk_status); | |
191 | ||
192 | int blk_status_to_errno(blk_status_t status) | |
193 | { | |
194 | int idx = (__force int)status; | |
195 | ||
34bd9c1c | 196 | if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors))) |
2a842aca CH |
197 | return -EIO; |
198 | return blk_errors[idx].errno; | |
199 | } | |
200 | EXPORT_SYMBOL_GPL(blk_status_to_errno); | |
201 | ||
0d7a29a2 | 202 | const char *blk_status_to_str(blk_status_t status) |
2a842aca CH |
203 | { |
204 | int idx = (__force int)status; | |
205 | ||
34bd9c1c | 206 | if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors))) |
0d7a29a2 CH |
207 | return "<null>"; |
208 | return blk_errors[idx].name; | |
2a842aca CH |
209 | } |
210 | ||
1da177e4 LT |
211 | /** |
212 | * blk_sync_queue - cancel any pending callbacks on a queue | |
213 | * @q: the queue | |
214 | * | |
215 | * Description: | |
216 | * The block layer may perform asynchronous callback activity | |
217 | * on a queue, such as calling the unplug function after a timeout. | |
218 | * A block device may call blk_sync_queue to ensure that any | |
219 | * such activity is cancelled, thus allowing it to release resources | |
59c51591 | 220 | * that the callbacks might use. The caller must already have made sure |
c62b37d9 | 221 | * that its ->submit_bio will not re-add plugging prior to calling |
1da177e4 LT |
222 | * this function. |
223 | * | |
da527770 | 224 | * This function does not cancel any asynchronous activity arising |
da3dae54 | 225 | * out of elevator or throttling code. That would require elevator_exit() |
5efd6113 | 226 | * and blkcg_exit_queue() to be called with queue lock initialized. |
da527770 | 227 | * |
1da177e4 LT |
228 | */ |
229 | void blk_sync_queue(struct request_queue *q) | |
230 | { | |
70ed28b9 | 231 | del_timer_sync(&q->timeout); |
4e9b6f20 | 232 | cancel_work_sync(&q->timeout_work); |
1da177e4 LT |
233 | } |
234 | EXPORT_SYMBOL(blk_sync_queue); | |
235 | ||
c9254f2d | 236 | /** |
cd84a62e | 237 | * blk_set_pm_only - increment pm_only counter |
c9254f2d | 238 | * @q: request queue pointer |
c9254f2d | 239 | */ |
cd84a62e | 240 | void blk_set_pm_only(struct request_queue *q) |
c9254f2d | 241 | { |
cd84a62e | 242 | atomic_inc(&q->pm_only); |
c9254f2d | 243 | } |
cd84a62e | 244 | EXPORT_SYMBOL_GPL(blk_set_pm_only); |
c9254f2d | 245 | |
cd84a62e | 246 | void blk_clear_pm_only(struct request_queue *q) |
c9254f2d | 247 | { |
cd84a62e BVA |
248 | int pm_only; |
249 | ||
250 | pm_only = atomic_dec_return(&q->pm_only); | |
251 | WARN_ON_ONCE(pm_only < 0); | |
252 | if (pm_only == 0) | |
253 | wake_up_all(&q->mq_freeze_wq); | |
c9254f2d | 254 | } |
cd84a62e | 255 | EXPORT_SYMBOL_GPL(blk_clear_pm_only); |
c9254f2d | 256 | |
b5bd357c LC |
257 | /** |
258 | * blk_put_queue - decrement the request_queue refcount | |
259 | * @q: the request_queue structure to decrement the refcount for | |
260 | * | |
261 | * Decrements the refcount of the request_queue kobject. When this reaches 0 | |
262 | * we'll have blk_release_queue() called. | |
e8c7d14a LC |
263 | * |
264 | * Context: Any context, but the last reference must not be dropped from | |
265 | * atomic context. | |
b5bd357c | 266 | */ |
165125e1 | 267 | void blk_put_queue(struct request_queue *q) |
483f4afc AV |
268 | { |
269 | kobject_put(&q->kobj); | |
270 | } | |
d86e0e83 | 271 | EXPORT_SYMBOL(blk_put_queue); |
483f4afc | 272 | |
8e141f9e | 273 | void blk_queue_start_drain(struct request_queue *q) |
aed3ea94 | 274 | { |
d3cfb2a0 ML |
275 | /* |
276 | * When queue DYING flag is set, we need to block new req | |
277 | * entering queue, so we call blk_freeze_queue_start() to | |
278 | * prevent I/O from crossing blk_queue_enter(). | |
279 | */ | |
280 | blk_freeze_queue_start(q); | |
344e9ffc | 281 | if (queue_is_mq(q)) |
aed3ea94 | 282 | blk_mq_wake_waiters(q); |
055f6e18 ML |
283 | /* Make blk_queue_enter() reexamine the DYING flag. */ |
284 | wake_up_all(&q->mq_freeze_wq); | |
aed3ea94 | 285 | } |
8e141f9e | 286 | |
3a0a5299 BVA |
287 | /** |
288 | * blk_queue_enter() - try to increase q->q_usage_counter | |
289 | * @q: request queue pointer | |
a4d34da7 | 290 | * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PM |
3a0a5299 | 291 | */ |
9a95e4ef | 292 | int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags) |
3ef28e83 | 293 | { |
a4d34da7 | 294 | const bool pm = flags & BLK_MQ_REQ_PM; |
3a0a5299 | 295 | |
1f14a098 | 296 | while (!blk_try_enter_queue(q, pm)) { |
3a0a5299 | 297 | if (flags & BLK_MQ_REQ_NOWAIT) |
3ef28e83 DW |
298 | return -EBUSY; |
299 | ||
5ed61d3f | 300 | /* |
1f14a098 CH |
301 | * read pair of barrier in blk_freeze_queue_start(), we need to |
302 | * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and | |
303 | * reading .mq_freeze_depth or queue dying flag, otherwise the | |
304 | * following wait may never return if the two reads are | |
305 | * reordered. | |
5ed61d3f ML |
306 | */ |
307 | smp_rmb(); | |
1dc3039b | 308 | wait_event(q->mq_freeze_wq, |
7996a8b5 | 309 | (!q->mq_freeze_depth && |
52abca64 | 310 | blk_pm_resume_queue(pm, q)) || |
1dc3039b | 311 | blk_queue_dying(q)); |
3ef28e83 DW |
312 | if (blk_queue_dying(q)) |
313 | return -ENODEV; | |
3ef28e83 | 314 | } |
1f14a098 CH |
315 | |
316 | return 0; | |
3ef28e83 DW |
317 | } |
318 | ||
c98cb5bb | 319 | int __bio_queue_enter(struct request_queue *q, struct bio *bio) |
accea322 | 320 | { |
a6741536 | 321 | while (!blk_try_enter_queue(q, false)) { |
eab4e027 PB |
322 | struct gendisk *disk = bio->bi_bdev->bd_disk; |
323 | ||
a6741536 | 324 | if (bio->bi_opf & REQ_NOWAIT) { |
8e141f9e | 325 | if (test_bit(GD_DEAD, &disk->state)) |
a6741536 | 326 | goto dead; |
accea322 | 327 | bio_wouldblock_error(bio); |
a6741536 CH |
328 | return -EBUSY; |
329 | } | |
330 | ||
331 | /* | |
332 | * read pair of barrier in blk_freeze_queue_start(), we need to | |
333 | * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and | |
334 | * reading .mq_freeze_depth or queue dying flag, otherwise the | |
335 | * following wait may never return if the two reads are | |
336 | * reordered. | |
337 | */ | |
338 | smp_rmb(); | |
339 | wait_event(q->mq_freeze_wq, | |
340 | (!q->mq_freeze_depth && | |
341 | blk_pm_resume_queue(false, q)) || | |
8e141f9e CH |
342 | test_bit(GD_DEAD, &disk->state)); |
343 | if (test_bit(GD_DEAD, &disk->state)) | |
a6741536 | 344 | goto dead; |
accea322 CH |
345 | } |
346 | ||
a6741536 CH |
347 | return 0; |
348 | dead: | |
349 | bio_io_error(bio); | |
350 | return -ENODEV; | |
accea322 CH |
351 | } |
352 | ||
3ef28e83 DW |
353 | void blk_queue_exit(struct request_queue *q) |
354 | { | |
355 | percpu_ref_put(&q->q_usage_counter); | |
356 | } | |
357 | ||
358 | static void blk_queue_usage_counter_release(struct percpu_ref *ref) | |
359 | { | |
360 | struct request_queue *q = | |
361 | container_of(ref, struct request_queue, q_usage_counter); | |
362 | ||
363 | wake_up_all(&q->mq_freeze_wq); | |
364 | } | |
365 | ||
bca237a5 | 366 | static void blk_rq_timed_out_timer(struct timer_list *t) |
287922eb | 367 | { |
bca237a5 | 368 | struct request_queue *q = from_timer(q, t, timeout); |
287922eb CH |
369 | |
370 | kblockd_schedule_work(&q->timeout_work); | |
371 | } | |
372 | ||
2e3c18d0 TH |
373 | static void blk_timeout_work(struct work_struct *work) |
374 | { | |
375 | } | |
376 | ||
704b914f | 377 | struct request_queue *blk_alloc_queue(int node_id, bool alloc_srcu) |
1946089a | 378 | { |
165125e1 | 379 | struct request_queue *q; |
338aa96d | 380 | int ret; |
1946089a | 381 | |
704b914f ML |
382 | q = kmem_cache_alloc_node(blk_get_queue_kmem_cache(alloc_srcu), |
383 | GFP_KERNEL | __GFP_ZERO, node_id); | |
1da177e4 LT |
384 | if (!q) |
385 | return NULL; | |
386 | ||
704b914f ML |
387 | if (alloc_srcu) { |
388 | blk_queue_flag_set(QUEUE_FLAG_HAS_SRCU, q); | |
389 | if (init_srcu_struct(q->srcu) != 0) | |
390 | goto fail_q; | |
391 | } | |
392 | ||
cbf62af3 | 393 | q->last_merge = NULL; |
cbf62af3 | 394 | |
798f2a6f | 395 | q->id = ida_alloc(&blk_queue_ida, GFP_KERNEL); |
a73f730d | 396 | if (q->id < 0) |
704b914f | 397 | goto fail_srcu; |
a73f730d | 398 | |
c495a176 | 399 | ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, 0); |
338aa96d | 400 | if (ret) |
54efd50b KO |
401 | goto fail_id; |
402 | ||
a83b576c JA |
403 | q->stats = blk_alloc_queue_stats(); |
404 | if (!q->stats) | |
edb0872f | 405 | goto fail_split; |
a83b576c | 406 | |
5151412d | 407 | q->node = node_id; |
0989a025 | 408 | |
079a2e3e | 409 | atomic_set(&q->nr_active_requests_shared_tags, 0); |
bccf5e26 | 410 | |
bca237a5 | 411 | timer_setup(&q->timeout, blk_rq_timed_out_timer, 0); |
2e3c18d0 | 412 | INIT_WORK(&q->timeout_work, blk_timeout_work); |
a612fddf | 413 | INIT_LIST_HEAD(&q->icq_list); |
483f4afc | 414 | |
8324aa91 | 415 | kobject_init(&q->kobj, &blk_queue_ktype); |
1da177e4 | 416 | |
85e0cbbb | 417 | mutex_init(&q->debugfs_mutex); |
483f4afc | 418 | mutex_init(&q->sysfs_lock); |
cecf5d87 | 419 | mutex_init(&q->sysfs_dir_lock); |
0d945c1f | 420 | spin_lock_init(&q->queue_lock); |
c94a96ac | 421 | |
320ae51f | 422 | init_waitqueue_head(&q->mq_freeze_wq); |
7996a8b5 | 423 | mutex_init(&q->mq_freeze_lock); |
320ae51f | 424 | |
3ef28e83 DW |
425 | /* |
426 | * Init percpu_ref in atomic mode so that it's faster to shutdown. | |
427 | * See blk_register_queue() for details. | |
428 | */ | |
429 | if (percpu_ref_init(&q->q_usage_counter, | |
430 | blk_queue_usage_counter_release, | |
431 | PERCPU_REF_INIT_ATOMIC, GFP_KERNEL)) | |
edb0872f | 432 | goto fail_stats; |
f51b802c | 433 | |
3d745ea5 CH |
434 | blk_queue_dma_alignment(q, 511); |
435 | blk_set_default_limits(&q->limits); | |
d2a27964 | 436 | q->nr_requests = BLKDEV_DEFAULT_RQ; |
3d745ea5 | 437 | |
1da177e4 | 438 | return q; |
a73f730d | 439 | |
a83b576c | 440 | fail_stats: |
edb0872f | 441 | blk_free_queue_stats(q->stats); |
54efd50b | 442 | fail_split: |
338aa96d | 443 | bioset_exit(&q->bio_split); |
a73f730d | 444 | fail_id: |
798f2a6f | 445 | ida_free(&blk_queue_ida, q->id); |
704b914f ML |
446 | fail_srcu: |
447 | if (alloc_srcu) | |
448 | cleanup_srcu_struct(q->srcu); | |
a73f730d | 449 | fail_q: |
704b914f | 450 | kmem_cache_free(blk_get_queue_kmem_cache(alloc_srcu), q); |
a73f730d | 451 | return NULL; |
1da177e4 | 452 | } |
1da177e4 | 453 | |
b5bd357c LC |
454 | /** |
455 | * blk_get_queue - increment the request_queue refcount | |
456 | * @q: the request_queue structure to increment the refcount for | |
457 | * | |
458 | * Increment the refcount of the request_queue kobject. | |
763b5892 LC |
459 | * |
460 | * Context: Any context. | |
b5bd357c | 461 | */ |
09ac46c4 | 462 | bool blk_get_queue(struct request_queue *q) |
1da177e4 | 463 | { |
3f3299d5 | 464 | if (likely(!blk_queue_dying(q))) { |
09ac46c4 TH |
465 | __blk_get_queue(q); |
466 | return true; | |
1da177e4 LT |
467 | } |
468 | ||
09ac46c4 | 469 | return false; |
1da177e4 | 470 | } |
d86e0e83 | 471 | EXPORT_SYMBOL(blk_get_queue); |
1da177e4 | 472 | |
c17bb495 AM |
473 | #ifdef CONFIG_FAIL_MAKE_REQUEST |
474 | ||
475 | static DECLARE_FAULT_ATTR(fail_make_request); | |
476 | ||
477 | static int __init setup_fail_make_request(char *str) | |
478 | { | |
479 | return setup_fault_attr(&fail_make_request, str); | |
480 | } | |
481 | __setup("fail_make_request=", setup_fail_make_request); | |
482 | ||
06c8c691 | 483 | bool should_fail_request(struct block_device *part, unsigned int bytes) |
c17bb495 | 484 | { |
8446fe92 | 485 | return part->bd_make_it_fail && should_fail(&fail_make_request, bytes); |
c17bb495 AM |
486 | } |
487 | ||
488 | static int __init fail_make_request_debugfs(void) | |
489 | { | |
dd48c085 AM |
490 | struct dentry *dir = fault_create_debugfs_attr("fail_make_request", |
491 | NULL, &fail_make_request); | |
492 | ||
21f9fcd8 | 493 | return PTR_ERR_OR_ZERO(dir); |
c17bb495 AM |
494 | } |
495 | ||
496 | late_initcall(fail_make_request_debugfs); | |
c17bb495 AM |
497 | #endif /* CONFIG_FAIL_MAKE_REQUEST */ |
498 | ||
2f9f6221 | 499 | static inline bool bio_check_ro(struct bio *bio) |
721c7fc7 | 500 | { |
2f9f6221 | 501 | if (op_is_write(bio_op(bio)) && bdev_read_only(bio->bi_bdev)) { |
8b2ded1c MP |
502 | if (op_is_flush(bio->bi_opf) && !bio_sectors(bio)) |
503 | return false; | |
57e95e46 CH |
504 | pr_warn("Trying to write to read-only block-device %pg\n", |
505 | bio->bi_bdev); | |
a32e236e LT |
506 | /* Older lvm-tools actually trigger this */ |
507 | return false; | |
721c7fc7 ID |
508 | } |
509 | ||
510 | return false; | |
511 | } | |
512 | ||
30abb3a6 HM |
513 | static noinline int should_fail_bio(struct bio *bio) |
514 | { | |
309dca30 | 515 | if (should_fail_request(bdev_whole(bio->bi_bdev), bio->bi_iter.bi_size)) |
30abb3a6 HM |
516 | return -EIO; |
517 | return 0; | |
518 | } | |
519 | ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO); | |
520 | ||
52c5e62d CH |
521 | /* |
522 | * Check whether this bio extends beyond the end of the device or partition. | |
523 | * This may well happen - the kernel calls bread() without checking the size of | |
524 | * the device, e.g., when mounting a file system. | |
525 | */ | |
2f9f6221 | 526 | static inline int bio_check_eod(struct bio *bio) |
52c5e62d | 527 | { |
2f9f6221 | 528 | sector_t maxsector = bdev_nr_sectors(bio->bi_bdev); |
52c5e62d CH |
529 | unsigned int nr_sectors = bio_sectors(bio); |
530 | ||
531 | if (nr_sectors && maxsector && | |
532 | (nr_sectors > maxsector || | |
533 | bio->bi_iter.bi_sector > maxsector - nr_sectors)) { | |
ad740780 | 534 | pr_info_ratelimited("%s: attempt to access beyond end of device\n" |
069adbac CH |
535 | "%pg: rw=%d, sector=%llu, nr_sectors = %u limit=%llu\n", |
536 | current->comm, bio->bi_bdev, bio->bi_opf, | |
537 | bio->bi_iter.bi_sector, nr_sectors, maxsector); | |
52c5e62d CH |
538 | return -EIO; |
539 | } | |
540 | return 0; | |
541 | } | |
542 | ||
74d46992 CH |
543 | /* |
544 | * Remap block n of partition p to block n+start(p) of the disk. | |
545 | */ | |
2f9f6221 | 546 | static int blk_partition_remap(struct bio *bio) |
74d46992 | 547 | { |
309dca30 | 548 | struct block_device *p = bio->bi_bdev; |
74d46992 | 549 | |
52c5e62d | 550 | if (unlikely(should_fail_request(p, bio->bi_iter.bi_size))) |
2f9f6221 | 551 | return -EIO; |
5eac3eb3 | 552 | if (bio_sectors(bio)) { |
8446fe92 | 553 | bio->bi_iter.bi_sector += p->bd_start_sect; |
1c02fca6 | 554 | trace_block_bio_remap(bio, p->bd_dev, |
29ff57c6 | 555 | bio->bi_iter.bi_sector - |
8446fe92 | 556 | p->bd_start_sect); |
52c5e62d | 557 | } |
30c5d345 | 558 | bio_set_flag(bio, BIO_REMAPPED); |
2f9f6221 | 559 | return 0; |
74d46992 CH |
560 | } |
561 | ||
0512a75b KB |
562 | /* |
563 | * Check write append to a zoned block device. | |
564 | */ | |
565 | static inline blk_status_t blk_check_zone_append(struct request_queue *q, | |
566 | struct bio *bio) | |
567 | { | |
568 | sector_t pos = bio->bi_iter.bi_sector; | |
569 | int nr_sectors = bio_sectors(bio); | |
570 | ||
571 | /* Only applicable to zoned block devices */ | |
edd1dbc8 | 572 | if (!bdev_is_zoned(bio->bi_bdev)) |
0512a75b KB |
573 | return BLK_STS_NOTSUPP; |
574 | ||
575 | /* The bio sector must point to the start of a sequential zone */ | |
576 | if (pos & (blk_queue_zone_sectors(q) - 1) || | |
577 | !blk_queue_zone_is_seq(q, pos)) | |
578 | return BLK_STS_IOERR; | |
579 | ||
580 | /* | |
581 | * Not allowed to cross zone boundaries. Otherwise, the BIO will be | |
582 | * split and could result in non-contiguous sectors being written in | |
583 | * different zones. | |
584 | */ | |
585 | if (nr_sectors > q->limits.chunk_sectors) | |
586 | return BLK_STS_IOERR; | |
587 | ||
588 | /* Make sure the BIO is small enough and will not get split */ | |
589 | if (nr_sectors > q->limits.max_zone_append_sectors) | |
590 | return BLK_STS_IOERR; | |
591 | ||
592 | bio->bi_opf |= REQ_NOMERGE; | |
593 | ||
594 | return BLK_STS_OK; | |
595 | } | |
596 | ||
900e0807 JA |
597 | static void __submit_bio(struct bio *bio) |
598 | { | |
599 | struct gendisk *disk = bio->bi_bdev->bd_disk; | |
cc9c884d | 600 | |
7f36b7d0 ML |
601 | if (unlikely(!blk_crypto_bio_prep(&bio))) |
602 | return; | |
603 | ||
604 | if (!disk->fops->submit_bio) { | |
3e08773c | 605 | blk_mq_submit_bio(bio); |
7f36b7d0 ML |
606 | } else if (likely(bio_queue_enter(bio) == 0)) { |
607 | disk->fops->submit_bio(bio); | |
608 | blk_queue_exit(disk->queue); | |
609 | } | |
ac7c5675 CH |
610 | } |
611 | ||
566acf2d CH |
612 | /* |
613 | * The loop in this function may be a bit non-obvious, and so deserves some | |
614 | * explanation: | |
615 | * | |
616 | * - Before entering the loop, bio->bi_next is NULL (as all callers ensure | |
617 | * that), so we have a list with a single bio. | |
618 | * - We pretend that we have just taken it off a longer list, so we assign | |
619 | * bio_list to a pointer to the bio_list_on_stack, thus initialising the | |
620 | * bio_list of new bios to be added. ->submit_bio() may indeed add some more | |
621 | * bios through a recursive call to submit_bio_noacct. If it did, we find a | |
622 | * non-NULL value in bio_list and re-enter the loop from the top. | |
623 | * - In this case we really did just take the bio of the top of the list (no | |
624 | * pretending) and so remove it from bio_list, and call into ->submit_bio() | |
625 | * again. | |
626 | * | |
627 | * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio. | |
628 | * bio_list_on_stack[1] contains bios that were submitted before the current | |
69fe0f29 | 629 | * ->submit_bio, but that haven't been processed yet. |
566acf2d | 630 | */ |
3e08773c | 631 | static void __submit_bio_noacct(struct bio *bio) |
566acf2d CH |
632 | { |
633 | struct bio_list bio_list_on_stack[2]; | |
566acf2d CH |
634 | |
635 | BUG_ON(bio->bi_next); | |
636 | ||
637 | bio_list_init(&bio_list_on_stack[0]); | |
638 | current->bio_list = bio_list_on_stack; | |
639 | ||
640 | do { | |
eab4e027 | 641 | struct request_queue *q = bdev_get_queue(bio->bi_bdev); |
566acf2d CH |
642 | struct bio_list lower, same; |
643 | ||
566acf2d CH |
644 | /* |
645 | * Create a fresh bio_list for all subordinate requests. | |
646 | */ | |
647 | bio_list_on_stack[1] = bio_list_on_stack[0]; | |
648 | bio_list_init(&bio_list_on_stack[0]); | |
649 | ||
3e08773c | 650 | __submit_bio(bio); |
566acf2d CH |
651 | |
652 | /* | |
653 | * Sort new bios into those for a lower level and those for the | |
654 | * same level. | |
655 | */ | |
656 | bio_list_init(&lower); | |
657 | bio_list_init(&same); | |
658 | while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL) | |
eab4e027 | 659 | if (q == bdev_get_queue(bio->bi_bdev)) |
566acf2d CH |
660 | bio_list_add(&same, bio); |
661 | else | |
662 | bio_list_add(&lower, bio); | |
663 | ||
664 | /* | |
665 | * Now assemble so we handle the lowest level first. | |
666 | */ | |
667 | bio_list_merge(&bio_list_on_stack[0], &lower); | |
668 | bio_list_merge(&bio_list_on_stack[0], &same); | |
669 | bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]); | |
670 | } while ((bio = bio_list_pop(&bio_list_on_stack[0]))); | |
671 | ||
672 | current->bio_list = NULL; | |
566acf2d CH |
673 | } |
674 | ||
3e08773c | 675 | static void __submit_bio_noacct_mq(struct bio *bio) |
ff93ea0c | 676 | { |
7c792f33 | 677 | struct bio_list bio_list[2] = { }; |
ff93ea0c | 678 | |
7c792f33 | 679 | current->bio_list = bio_list; |
ff93ea0c CH |
680 | |
681 | do { | |
3e08773c | 682 | __submit_bio(bio); |
7c792f33 | 683 | } while ((bio = bio_list_pop(&bio_list[0]))); |
ff93ea0c CH |
684 | |
685 | current->bio_list = NULL; | |
ff93ea0c CH |
686 | } |
687 | ||
3f98c753 | 688 | void submit_bio_noacct_nocheck(struct bio *bio) |
d89d8796 | 689 | { |
27a84d54 | 690 | /* |
566acf2d CH |
691 | * We only want one ->submit_bio to be active at a time, else stack |
692 | * usage with stacked devices could be a problem. Use current->bio_list | |
693 | * to collect a list of requests submited by a ->submit_bio method while | |
694 | * it is active, and then process them after it returned. | |
27a84d54 | 695 | */ |
3e08773c | 696 | if (current->bio_list) |
f5fe1b51 | 697 | bio_list_add(¤t->bio_list[0], bio); |
3e08773c CH |
698 | else if (!bio->bi_bdev->bd_disk->fops->submit_bio) |
699 | __submit_bio_noacct_mq(bio); | |
700 | else | |
701 | __submit_bio_noacct(bio); | |
d89d8796 | 702 | } |
3f98c753 ML |
703 | |
704 | /** | |
705 | * submit_bio_noacct - re-submit a bio to the block device layer for I/O | |
706 | * @bio: The bio describing the location in memory and on the device. | |
707 | * | |
708 | * This is a version of submit_bio() that shall only be used for I/O that is | |
709 | * resubmitted to lower level drivers by stacking block drivers. All file | |
710 | * systems and other upper level users of the block layer should use | |
711 | * submit_bio() instead. | |
712 | */ | |
713 | void submit_bio_noacct(struct bio *bio) | |
1da177e4 | 714 | { |
309dca30 | 715 | struct block_device *bdev = bio->bi_bdev; |
eab4e027 | 716 | struct request_queue *q = bdev_get_queue(bdev); |
4e4cbee9 | 717 | blk_status_t status = BLK_STS_IOERR; |
5a473e83 | 718 | struct blk_plug *plug; |
1da177e4 LT |
719 | |
720 | might_sleep(); | |
1da177e4 | 721 | |
5a473e83 JA |
722 | plug = blk_mq_plug(q, bio); |
723 | if (plug && plug->nowait) | |
724 | bio->bi_opf |= REQ_NOWAIT; | |
725 | ||
03a07c92 | 726 | /* |
b0beb280 | 727 | * For a REQ_NOWAIT based request, return -EOPNOTSUPP |
021a2446 | 728 | * if queue does not support NOWAIT. |
03a07c92 | 729 | */ |
021a2446 | 730 | if ((bio->bi_opf & REQ_NOWAIT) && !blk_queue_nowait(q)) |
b0beb280 | 731 | goto not_supported; |
03a07c92 | 732 | |
30abb3a6 | 733 | if (should_fail_bio(bio)) |
5a7bbad2 | 734 | goto end_io; |
2f9f6221 CH |
735 | if (unlikely(bio_check_ro(bio))) |
736 | goto end_io; | |
3a905c37 CH |
737 | if (!bio_flagged(bio, BIO_REMAPPED)) { |
738 | if (unlikely(bio_check_eod(bio))) | |
739 | goto end_io; | |
740 | if (bdev->bd_partno && unlikely(blk_partition_remap(bio))) | |
741 | goto end_io; | |
742 | } | |
2056a782 | 743 | |
5a7bbad2 | 744 | /* |
ed00aabd CH |
745 | * Filter flush bio's early so that bio based drivers without flush |
746 | * support don't have to worry about them. | |
5a7bbad2 | 747 | */ |
f3a8ab7d | 748 | if (op_is_flush(bio->bi_opf) && |
c888a8f9 | 749 | !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) { |
1eff9d32 | 750 | bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA); |
e439ab71 | 751 | if (!bio_sectors(bio)) { |
4e4cbee9 | 752 | status = BLK_STS_OK; |
51fd77bd JA |
753 | goto end_io; |
754 | } | |
5a7bbad2 | 755 | } |
5ddfe969 | 756 | |
d04c406f | 757 | if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags)) |
6ce913fe | 758 | bio_clear_polled(bio); |
d04c406f | 759 | |
288dab8a CH |
760 | switch (bio_op(bio)) { |
761 | case REQ_OP_DISCARD: | |
70200574 | 762 | if (!bdev_max_discard_sectors(bdev)) |
288dab8a CH |
763 | goto not_supported; |
764 | break; | |
765 | case REQ_OP_SECURE_ERASE: | |
44abff2c | 766 | if (!bdev_max_secure_erase_sectors(bdev)) |
288dab8a CH |
767 | goto not_supported; |
768 | break; | |
0512a75b KB |
769 | case REQ_OP_ZONE_APPEND: |
770 | status = blk_check_zone_append(q, bio); | |
771 | if (status != BLK_STS_OK) | |
772 | goto end_io; | |
773 | break; | |
2d253440 | 774 | case REQ_OP_ZONE_RESET: |
6c1b1da5 AJ |
775 | case REQ_OP_ZONE_OPEN: |
776 | case REQ_OP_ZONE_CLOSE: | |
777 | case REQ_OP_ZONE_FINISH: | |
edd1dbc8 | 778 | if (!bdev_is_zoned(bio->bi_bdev)) |
2d253440 | 779 | goto not_supported; |
288dab8a | 780 | break; |
6e33dbf2 | 781 | case REQ_OP_ZONE_RESET_ALL: |
edd1dbc8 | 782 | if (!bdev_is_zoned(bio->bi_bdev) || !blk_queue_zone_resetall(q)) |
6e33dbf2 CK |
783 | goto not_supported; |
784 | break; | |
a6f0788e | 785 | case REQ_OP_WRITE_ZEROES: |
74d46992 | 786 | if (!q->limits.max_write_zeroes_sectors) |
a6f0788e CK |
787 | goto not_supported; |
788 | break; | |
288dab8a CH |
789 | default: |
790 | break; | |
5a7bbad2 | 791 | } |
01edede4 | 792 | |
b781d8db | 793 | if (blk_throtl_bio(bio)) |
3f98c753 | 794 | return; |
db18a53e CH |
795 | |
796 | blk_cgroup_bio_start(bio); | |
797 | blkcg_bio_issue_init(bio); | |
27a84d54 | 798 | |
fbbaf700 | 799 | if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) { |
e8a676d6 | 800 | trace_block_bio_queue(bio); |
fbbaf700 N |
801 | /* Now that enqueuing has been traced, we need to trace |
802 | * completion as well. | |
803 | */ | |
804 | bio_set_flag(bio, BIO_TRACE_COMPLETION); | |
805 | } | |
3f98c753 | 806 | submit_bio_noacct_nocheck(bio); |
d24c670e | 807 | return; |
a7384677 | 808 | |
288dab8a | 809 | not_supported: |
4e4cbee9 | 810 | status = BLK_STS_NOTSUPP; |
a7384677 | 811 | end_io: |
4e4cbee9 | 812 | bio->bi_status = status; |
4246a0b6 | 813 | bio_endio(bio); |
d89d8796 | 814 | } |
ed00aabd | 815 | EXPORT_SYMBOL(submit_bio_noacct); |
1da177e4 LT |
816 | |
817 | /** | |
710027a4 | 818 | * submit_bio - submit a bio to the block device layer for I/O |
1da177e4 LT |
819 | * @bio: The &struct bio which describes the I/O |
820 | * | |
3fdd4086 CH |
821 | * submit_bio() is used to submit I/O requests to block devices. It is passed a |
822 | * fully set up &struct bio that describes the I/O that needs to be done. The | |
309dca30 | 823 | * bio will be send to the device described by the bi_bdev field. |
1da177e4 | 824 | * |
3fdd4086 CH |
825 | * The success/failure status of the request, along with notification of |
826 | * completion, is delivered asynchronously through the ->bi_end_io() callback | |
827 | * in @bio. The bio must NOT be touched by thecaller until ->bi_end_io() has | |
828 | * been called. | |
1da177e4 | 829 | */ |
3e08773c | 830 | void submit_bio(struct bio *bio) |
1da177e4 | 831 | { |
d3f77dfd | 832 | if (blkcg_punt_bio_submit(bio)) |
3e08773c | 833 | return; |
d3f77dfd | 834 | |
a3e7689b CH |
835 | if (bio_op(bio) == REQ_OP_READ) { |
836 | task_io_account_read(bio->bi_iter.bi_size); | |
837 | count_vm_events(PGPGIN, bio_sectors(bio)); | |
838 | } else if (bio_op(bio) == REQ_OP_WRITE) { | |
839 | count_vm_events(PGPGOUT, bio_sectors(bio)); | |
1da177e4 LT |
840 | } |
841 | ||
b8e24a93 | 842 | /* |
760f83ea CH |
843 | * If we're reading data that is part of the userspace workingset, count |
844 | * submission time as memory stall. When the device is congested, or | |
845 | * the submitting cgroup IO-throttled, submission can be a significant | |
846 | * part of overall IO time. | |
b8e24a93 | 847 | */ |
760f83ea CH |
848 | if (unlikely(bio_op(bio) == REQ_OP_READ && |
849 | bio_flagged(bio, BIO_WORKINGSET))) { | |
850 | unsigned long pflags; | |
b8e24a93 | 851 | |
760f83ea | 852 | psi_memstall_enter(&pflags); |
3e08773c | 853 | submit_bio_noacct(bio); |
b8e24a93 | 854 | psi_memstall_leave(&pflags); |
3e08773c | 855 | return; |
760f83ea CH |
856 | } |
857 | ||
3e08773c | 858 | submit_bio_noacct(bio); |
1da177e4 | 859 | } |
1da177e4 LT |
860 | EXPORT_SYMBOL(submit_bio); |
861 | ||
3e08773c CH |
862 | /** |
863 | * bio_poll - poll for BIO completions | |
864 | * @bio: bio to poll for | |
e30028ac | 865 | * @iob: batches of IO |
3e08773c CH |
866 | * @flags: BLK_POLL_* flags that control the behavior |
867 | * | |
868 | * Poll for completions on queue associated with the bio. Returns number of | |
869 | * completed entries found. | |
870 | * | |
871 | * Note: the caller must either be the context that submitted @bio, or | |
872 | * be in a RCU critical section to prevent freeing of @bio. | |
873 | */ | |
5a72e899 | 874 | int bio_poll(struct bio *bio, struct io_comp_batch *iob, unsigned int flags) |
3e08773c | 875 | { |
859897c3 | 876 | struct request_queue *q = bdev_get_queue(bio->bi_bdev); |
3e08773c | 877 | blk_qc_t cookie = READ_ONCE(bio->bi_cookie); |
69fe0f29 | 878 | int ret = 0; |
3e08773c CH |
879 | |
880 | if (cookie == BLK_QC_T_NONE || | |
881 | !test_bit(QUEUE_FLAG_POLL, &q->queue_flags)) | |
882 | return 0; | |
883 | ||
aa8dccca | 884 | blk_flush_plug(current->plug, false); |
3e08773c | 885 | |
ebd076bf | 886 | if (bio_queue_enter(bio)) |
3e08773c | 887 | return 0; |
69fe0f29 | 888 | if (queue_is_mq(q)) { |
5a72e899 | 889 | ret = blk_mq_poll(q, cookie, iob, flags); |
69fe0f29 ML |
890 | } else { |
891 | struct gendisk *disk = q->disk; | |
892 | ||
893 | if (disk && disk->fops->poll_bio) | |
894 | ret = disk->fops->poll_bio(bio, iob, flags); | |
895 | } | |
3e08773c CH |
896 | blk_queue_exit(q); |
897 | return ret; | |
898 | } | |
899 | EXPORT_SYMBOL_GPL(bio_poll); | |
900 | ||
901 | /* | |
902 | * Helper to implement file_operations.iopoll. Requires the bio to be stored | |
903 | * in iocb->private, and cleared before freeing the bio. | |
904 | */ | |
5a72e899 JA |
905 | int iocb_bio_iopoll(struct kiocb *kiocb, struct io_comp_batch *iob, |
906 | unsigned int flags) | |
3e08773c CH |
907 | { |
908 | struct bio *bio; | |
909 | int ret = 0; | |
910 | ||
911 | /* | |
912 | * Note: the bio cache only uses SLAB_TYPESAFE_BY_RCU, so bio can | |
913 | * point to a freshly allocated bio at this point. If that happens | |
914 | * we have a few cases to consider: | |
915 | * | |
916 | * 1) the bio is beeing initialized and bi_bdev is NULL. We can just | |
917 | * simply nothing in this case | |
918 | * 2) the bio points to a not poll enabled device. bio_poll will catch | |
919 | * this and return 0 | |
920 | * 3) the bio points to a poll capable device, including but not | |
921 | * limited to the one that the original bio pointed to. In this | |
922 | * case we will call into the actual poll method and poll for I/O, | |
923 | * even if we don't need to, but it won't cause harm either. | |
924 | * | |
925 | * For cases 2) and 3) above the RCU grace period ensures that bi_bdev | |
926 | * is still allocated. Because partitions hold a reference to the whole | |
927 | * device bdev and thus disk, the disk is also still valid. Grabbing | |
928 | * a reference to the queue in bio_poll() ensures the hctxs and requests | |
929 | * are still valid as well. | |
930 | */ | |
931 | rcu_read_lock(); | |
932 | bio = READ_ONCE(kiocb->private); | |
933 | if (bio && bio->bi_bdev) | |
5a72e899 | 934 | ret = bio_poll(bio, iob, flags); |
3e08773c CH |
935 | rcu_read_unlock(); |
936 | ||
937 | return ret; | |
938 | } | |
939 | EXPORT_SYMBOL_GPL(iocb_bio_iopoll); | |
940 | ||
450b7879 | 941 | void update_io_ticks(struct block_device *part, unsigned long now, bool end) |
9123bf6f CH |
942 | { |
943 | unsigned long stamp; | |
944 | again: | |
8446fe92 | 945 | stamp = READ_ONCE(part->bd_stamp); |
d80c228d | 946 | if (unlikely(time_after(now, stamp))) { |
8446fe92 | 947 | if (likely(cmpxchg(&part->bd_stamp, stamp, now) == stamp)) |
9123bf6f CH |
948 | __part_stat_add(part, io_ticks, end ? now - stamp : 1); |
949 | } | |
8446fe92 CH |
950 | if (part->bd_partno) { |
951 | part = bdev_whole(part); | |
9123bf6f CH |
952 | goto again; |
953 | } | |
954 | } | |
955 | ||
5f0614a5 ML |
956 | unsigned long bdev_start_io_acct(struct block_device *bdev, |
957 | unsigned int sectors, unsigned int op, | |
958 | unsigned long start_time) | |
956d510e | 959 | { |
956d510e | 960 | const int sgrp = op_stat_group(op); |
956d510e CH |
961 | |
962 | part_stat_lock(); | |
5f0614a5 ML |
963 | update_io_ticks(bdev, start_time, false); |
964 | part_stat_inc(bdev, ios[sgrp]); | |
965 | part_stat_add(bdev, sectors[sgrp], sectors); | |
966 | part_stat_local_inc(bdev, in_flight[op_is_write(op)]); | |
956d510e | 967 | part_stat_unlock(); |
320ae51f | 968 | |
e45c47d1 MS |
969 | return start_time; |
970 | } | |
5f0614a5 | 971 | EXPORT_SYMBOL(bdev_start_io_acct); |
e45c47d1 MS |
972 | |
973 | /** | |
974 | * bio_start_io_acct_time - start I/O accounting for bio based drivers | |
975 | * @bio: bio to start account for | |
976 | * @start_time: start time that should be passed back to bio_end_io_acct(). | |
977 | */ | |
978 | void bio_start_io_acct_time(struct bio *bio, unsigned long start_time) | |
979 | { | |
5f0614a5 ML |
980 | bdev_start_io_acct(bio->bi_bdev, bio_sectors(bio), |
981 | bio_op(bio), start_time); | |
956d510e | 982 | } |
e45c47d1 | 983 | EXPORT_SYMBOL_GPL(bio_start_io_acct_time); |
7b26410b | 984 | |
99dfc43e CH |
985 | /** |
986 | * bio_start_io_acct - start I/O accounting for bio based drivers | |
987 | * @bio: bio to start account for | |
988 | * | |
989 | * Returns the start time that should be passed back to bio_end_io_acct(). | |
990 | */ | |
991 | unsigned long bio_start_io_acct(struct bio *bio) | |
7b26410b | 992 | { |
5f0614a5 ML |
993 | return bdev_start_io_acct(bio->bi_bdev, bio_sectors(bio), |
994 | bio_op(bio), jiffies); | |
7b26410b | 995 | } |
99dfc43e | 996 | EXPORT_SYMBOL_GPL(bio_start_io_acct); |
7b26410b | 997 | |
5f0614a5 ML |
998 | void bdev_end_io_acct(struct block_device *bdev, unsigned int op, |
999 | unsigned long start_time) | |
956d510e | 1000 | { |
956d510e CH |
1001 | const int sgrp = op_stat_group(op); |
1002 | unsigned long now = READ_ONCE(jiffies); | |
1003 | unsigned long duration = now - start_time; | |
5b18b5a7 | 1004 | |
956d510e | 1005 | part_stat_lock(); |
5f0614a5 ML |
1006 | update_io_ticks(bdev, now, true); |
1007 | part_stat_add(bdev, nsecs[sgrp], jiffies_to_nsecs(duration)); | |
1008 | part_stat_local_dec(bdev, in_flight[op_is_write(op)]); | |
320ae51f JA |
1009 | part_stat_unlock(); |
1010 | } | |
5f0614a5 | 1011 | EXPORT_SYMBOL(bdev_end_io_acct); |
7b26410b | 1012 | |
99dfc43e | 1013 | void bio_end_io_acct_remapped(struct bio *bio, unsigned long start_time, |
5f0614a5 | 1014 | struct block_device *orig_bdev) |
7b26410b | 1015 | { |
5f0614a5 | 1016 | bdev_end_io_acct(orig_bdev, bio_op(bio), start_time); |
7b26410b | 1017 | } |
99dfc43e | 1018 | EXPORT_SYMBOL_GPL(bio_end_io_acct_remapped); |
7b26410b | 1019 | |
ef9e3fac KU |
1020 | /** |
1021 | * blk_lld_busy - Check if underlying low-level drivers of a device are busy | |
1022 | * @q : the queue of the device being checked | |
1023 | * | |
1024 | * Description: | |
1025 | * Check if underlying low-level drivers of a device are busy. | |
1026 | * If the drivers want to export their busy state, they must set own | |
1027 | * exporting function using blk_queue_lld_busy() first. | |
1028 | * | |
1029 | * Basically, this function is used only by request stacking drivers | |
1030 | * to stop dispatching requests to underlying devices when underlying | |
1031 | * devices are busy. This behavior helps more I/O merging on the queue | |
1032 | * of the request stacking driver and prevents I/O throughput regression | |
1033 | * on burst I/O load. | |
1034 | * | |
1035 | * Return: | |
1036 | * 0 - Not busy (The request stacking driver should dispatch request) | |
1037 | * 1 - Busy (The request stacking driver should stop dispatching request) | |
1038 | */ | |
1039 | int blk_lld_busy(struct request_queue *q) | |
1040 | { | |
344e9ffc | 1041 | if (queue_is_mq(q) && q->mq_ops->busy) |
9ba20527 | 1042 | return q->mq_ops->busy(q); |
ef9e3fac KU |
1043 | |
1044 | return 0; | |
1045 | } | |
1046 | EXPORT_SYMBOL_GPL(blk_lld_busy); | |
1047 | ||
59c3d45e | 1048 | int kblockd_schedule_work(struct work_struct *work) |
1da177e4 LT |
1049 | { |
1050 | return queue_work(kblockd_workqueue, work); | |
1051 | } | |
1da177e4 LT |
1052 | EXPORT_SYMBOL(kblockd_schedule_work); |
1053 | ||
818cd1cb JA |
1054 | int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork, |
1055 | unsigned long delay) | |
1056 | { | |
1057 | return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay); | |
1058 | } | |
1059 | EXPORT_SYMBOL(kblockd_mod_delayed_work_on); | |
1060 | ||
47c122e3 JA |
1061 | void blk_start_plug_nr_ios(struct blk_plug *plug, unsigned short nr_ios) |
1062 | { | |
1063 | struct task_struct *tsk = current; | |
1064 | ||
1065 | /* | |
1066 | * If this is a nested plug, don't actually assign it. | |
1067 | */ | |
1068 | if (tsk->plug) | |
1069 | return; | |
1070 | ||
bc490f81 | 1071 | plug->mq_list = NULL; |
47c122e3 JA |
1072 | plug->cached_rq = NULL; |
1073 | plug->nr_ios = min_t(unsigned short, nr_ios, BLK_MAX_REQUEST_COUNT); | |
1074 | plug->rq_count = 0; | |
1075 | plug->multiple_queues = false; | |
dc5fc361 | 1076 | plug->has_elevator = false; |
47c122e3 JA |
1077 | plug->nowait = false; |
1078 | INIT_LIST_HEAD(&plug->cb_list); | |
1079 | ||
1080 | /* | |
1081 | * Store ordering should not be needed here, since a potential | |
1082 | * preempt will imply a full memory barrier | |
1083 | */ | |
1084 | tsk->plug = plug; | |
1085 | } | |
1086 | ||
75df7136 SJ |
1087 | /** |
1088 | * blk_start_plug - initialize blk_plug and track it inside the task_struct | |
1089 | * @plug: The &struct blk_plug that needs to be initialized | |
1090 | * | |
1091 | * Description: | |
40405851 JM |
1092 | * blk_start_plug() indicates to the block layer an intent by the caller |
1093 | * to submit multiple I/O requests in a batch. The block layer may use | |
1094 | * this hint to defer submitting I/Os from the caller until blk_finish_plug() | |
1095 | * is called. However, the block layer may choose to submit requests | |
1096 | * before a call to blk_finish_plug() if the number of queued I/Os | |
1097 | * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than | |
1098 | * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if | |
1099 | * the task schedules (see below). | |
1100 | * | |
75df7136 SJ |
1101 | * Tracking blk_plug inside the task_struct will help with auto-flushing the |
1102 | * pending I/O should the task end up blocking between blk_start_plug() and | |
1103 | * blk_finish_plug(). This is important from a performance perspective, but | |
1104 | * also ensures that we don't deadlock. For instance, if the task is blocking | |
1105 | * for a memory allocation, memory reclaim could end up wanting to free a | |
1106 | * page belonging to that request that is currently residing in our private | |
1107 | * plug. By flushing the pending I/O when the process goes to sleep, we avoid | |
1108 | * this kind of deadlock. | |
1109 | */ | |
73c10101 JA |
1110 | void blk_start_plug(struct blk_plug *plug) |
1111 | { | |
47c122e3 | 1112 | blk_start_plug_nr_ios(plug, 1); |
73c10101 JA |
1113 | } |
1114 | EXPORT_SYMBOL(blk_start_plug); | |
1115 | ||
74018dc3 | 1116 | static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule) |
048c9374 N |
1117 | { |
1118 | LIST_HEAD(callbacks); | |
1119 | ||
2a7d5559 SL |
1120 | while (!list_empty(&plug->cb_list)) { |
1121 | list_splice_init(&plug->cb_list, &callbacks); | |
048c9374 | 1122 | |
2a7d5559 SL |
1123 | while (!list_empty(&callbacks)) { |
1124 | struct blk_plug_cb *cb = list_first_entry(&callbacks, | |
048c9374 N |
1125 | struct blk_plug_cb, |
1126 | list); | |
2a7d5559 | 1127 | list_del(&cb->list); |
74018dc3 | 1128 | cb->callback(cb, from_schedule); |
2a7d5559 | 1129 | } |
048c9374 N |
1130 | } |
1131 | } | |
1132 | ||
9cbb1750 N |
1133 | struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data, |
1134 | int size) | |
1135 | { | |
1136 | struct blk_plug *plug = current->plug; | |
1137 | struct blk_plug_cb *cb; | |
1138 | ||
1139 | if (!plug) | |
1140 | return NULL; | |
1141 | ||
1142 | list_for_each_entry(cb, &plug->cb_list, list) | |
1143 | if (cb->callback == unplug && cb->data == data) | |
1144 | return cb; | |
1145 | ||
1146 | /* Not currently on the callback list */ | |
1147 | BUG_ON(size < sizeof(*cb)); | |
1148 | cb = kzalloc(size, GFP_ATOMIC); | |
1149 | if (cb) { | |
1150 | cb->data = data; | |
1151 | cb->callback = unplug; | |
1152 | list_add(&cb->list, &plug->cb_list); | |
1153 | } | |
1154 | return cb; | |
1155 | } | |
1156 | EXPORT_SYMBOL(blk_check_plugged); | |
1157 | ||
aa8dccca | 1158 | void __blk_flush_plug(struct blk_plug *plug, bool from_schedule) |
73c10101 | 1159 | { |
b600455d PB |
1160 | if (!list_empty(&plug->cb_list)) |
1161 | flush_plug_callbacks(plug, from_schedule); | |
bc490f81 | 1162 | if (!rq_list_empty(plug->mq_list)) |
320ae51f | 1163 | blk_mq_flush_plug_list(plug, from_schedule); |
c5fc7b93 JA |
1164 | /* |
1165 | * Unconditionally flush out cached requests, even if the unplug | |
1166 | * event came from schedule. Since we know hold references to the | |
1167 | * queue for cached requests, we don't want a blocked task holding | |
1168 | * up a queue freeze/quiesce event. | |
1169 | */ | |
1170 | if (unlikely(!rq_list_empty(plug->cached_rq))) | |
47c122e3 | 1171 | blk_mq_free_plug_rqs(plug); |
73c10101 | 1172 | } |
73c10101 | 1173 | |
40405851 JM |
1174 | /** |
1175 | * blk_finish_plug - mark the end of a batch of submitted I/O | |
1176 | * @plug: The &struct blk_plug passed to blk_start_plug() | |
1177 | * | |
1178 | * Description: | |
1179 | * Indicate that a batch of I/O submissions is complete. This function | |
1180 | * must be paired with an initial call to blk_start_plug(). The intent | |
1181 | * is to allow the block layer to optimize I/O submission. See the | |
1182 | * documentation for blk_start_plug() for more information. | |
1183 | */ | |
73c10101 JA |
1184 | void blk_finish_plug(struct blk_plug *plug) |
1185 | { | |
008f75a2 | 1186 | if (plug == current->plug) { |
aa8dccca | 1187 | __blk_flush_plug(plug, false); |
008f75a2 CH |
1188 | current->plug = NULL; |
1189 | } | |
73c10101 | 1190 | } |
88b996cd | 1191 | EXPORT_SYMBOL(blk_finish_plug); |
73c10101 | 1192 | |
71ac860a ML |
1193 | void blk_io_schedule(void) |
1194 | { | |
1195 | /* Prevent hang_check timer from firing at us during very long I/O */ | |
1196 | unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2; | |
1197 | ||
1198 | if (timeout) | |
1199 | io_schedule_timeout(timeout); | |
1200 | else | |
1201 | io_schedule(); | |
1202 | } | |
1203 | EXPORT_SYMBOL_GPL(blk_io_schedule); | |
1204 | ||
1da177e4 LT |
1205 | int __init blk_dev_init(void) |
1206 | { | |
ef295ecf CH |
1207 | BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS)); |
1208 | BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * | |
c593642c | 1209 | sizeof_field(struct request, cmd_flags)); |
ef295ecf | 1210 | BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * |
c593642c | 1211 | sizeof_field(struct bio, bi_opf)); |
704b914f ML |
1212 | BUILD_BUG_ON(ALIGN(offsetof(struct request_queue, srcu), |
1213 | __alignof__(struct request_queue)) != | |
1214 | sizeof(struct request_queue)); | |
9eb55b03 | 1215 | |
89b90be2 TH |
1216 | /* used for unplugging and affects IO latency/throughput - HIGHPRI */ |
1217 | kblockd_workqueue = alloc_workqueue("kblockd", | |
28747fcd | 1218 | WQ_MEM_RECLAIM | WQ_HIGHPRI, 0); |
1da177e4 LT |
1219 | if (!kblockd_workqueue) |
1220 | panic("Failed to create kblockd\n"); | |
1221 | ||
c2789bd4 | 1222 | blk_requestq_cachep = kmem_cache_create("request_queue", |
165125e1 | 1223 | sizeof(struct request_queue), 0, SLAB_PANIC, NULL); |
1da177e4 | 1224 | |
704b914f ML |
1225 | blk_requestq_srcu_cachep = kmem_cache_create("request_queue_srcu", |
1226 | sizeof(struct request_queue) + | |
1227 | sizeof(struct srcu_struct), 0, SLAB_PANIC, NULL); | |
1228 | ||
18fbda91 | 1229 | blk_debugfs_root = debugfs_create_dir("block", NULL); |
18fbda91 | 1230 | |
d38ecf93 | 1231 | return 0; |
1da177e4 | 1232 | } |