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> |
82d981d4 | 40 | #include <linux/part_stat.h> |
71ac860a | 41 | #include <linux/sched/sysctl.h> |
a892c8d5 | 42 | #include <linux/blk-crypto.h> |
55782138 LZ |
43 | |
44 | #define CREATE_TRACE_POINTS | |
45 | #include <trace/events/block.h> | |
1da177e4 | 46 | |
8324aa91 | 47 | #include "blk.h" |
2aa7745b | 48 | #include "blk-mq-sched.h" |
bca6b067 | 49 | #include "blk-pm.h" |
672fdcf0 | 50 | #include "blk-cgroup.h" |
a7b36ee6 | 51 | #include "blk-throttle.h" |
f3c89983 | 52 | #include "blk-ioprio.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 | |
2bd85221 | 63 | static DEFINE_IDA(blk_queue_ida); |
a73f730d | 64 | |
1da177e4 LT |
65 | /* |
66 | * For queue allocation | |
67 | */ | |
2bd85221 | 68 | static struct kmem_cache *blk_requestq_cachep; |
1da177e4 | 69 | |
1da177e4 LT |
70 | /* |
71 | * Controlling structure to kblockd | |
72 | */ | |
ff856bad | 73 | static struct workqueue_struct *kblockd_workqueue; |
1da177e4 | 74 | |
8814ce8a BVA |
75 | /** |
76 | * blk_queue_flag_set - atomically set a queue flag | |
77 | * @flag: flag to be set | |
78 | * @q: request queue | |
79 | */ | |
80 | void blk_queue_flag_set(unsigned int flag, struct request_queue *q) | |
81 | { | |
57d74df9 | 82 | set_bit(flag, &q->queue_flags); |
8814ce8a BVA |
83 | } |
84 | EXPORT_SYMBOL(blk_queue_flag_set); | |
85 | ||
86 | /** | |
87 | * blk_queue_flag_clear - atomically clear a queue flag | |
88 | * @flag: flag to be cleared | |
89 | * @q: request queue | |
90 | */ | |
91 | void blk_queue_flag_clear(unsigned int flag, struct request_queue *q) | |
92 | { | |
57d74df9 | 93 | clear_bit(flag, &q->queue_flags); |
8814ce8a BVA |
94 | } |
95 | EXPORT_SYMBOL(blk_queue_flag_clear); | |
96 | ||
97 | /** | |
98 | * blk_queue_flag_test_and_set - atomically test and set a queue flag | |
99 | * @flag: flag to be set | |
100 | * @q: request queue | |
101 | * | |
102 | * Returns the previous value of @flag - 0 if the flag was not set and 1 if | |
103 | * the flag was already set. | |
104 | */ | |
105 | bool blk_queue_flag_test_and_set(unsigned int flag, struct request_queue *q) | |
106 | { | |
57d74df9 | 107 | return test_and_set_bit(flag, &q->queue_flags); |
8814ce8a BVA |
108 | } |
109 | EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set); | |
110 | ||
e47bc4ed CK |
111 | #define REQ_OP_NAME(name) [REQ_OP_##name] = #name |
112 | static const char *const blk_op_name[] = { | |
113 | REQ_OP_NAME(READ), | |
114 | REQ_OP_NAME(WRITE), | |
115 | REQ_OP_NAME(FLUSH), | |
116 | REQ_OP_NAME(DISCARD), | |
117 | REQ_OP_NAME(SECURE_ERASE), | |
118 | REQ_OP_NAME(ZONE_RESET), | |
6e33dbf2 | 119 | REQ_OP_NAME(ZONE_RESET_ALL), |
6c1b1da5 AJ |
120 | REQ_OP_NAME(ZONE_OPEN), |
121 | REQ_OP_NAME(ZONE_CLOSE), | |
122 | REQ_OP_NAME(ZONE_FINISH), | |
0512a75b | 123 | REQ_OP_NAME(ZONE_APPEND), |
e47bc4ed | 124 | REQ_OP_NAME(WRITE_ZEROES), |
e47bc4ed CK |
125 | REQ_OP_NAME(DRV_IN), |
126 | REQ_OP_NAME(DRV_OUT), | |
127 | }; | |
128 | #undef REQ_OP_NAME | |
129 | ||
130 | /** | |
131 | * blk_op_str - Return string XXX in the REQ_OP_XXX. | |
132 | * @op: REQ_OP_XXX. | |
133 | * | |
134 | * Description: Centralize block layer function to convert REQ_OP_XXX into | |
135 | * string format. Useful in the debugging and tracing bio or request. For | |
136 | * invalid REQ_OP_XXX it returns string "UNKNOWN". | |
137 | */ | |
77e7ffd7 | 138 | inline const char *blk_op_str(enum req_op op) |
e47bc4ed CK |
139 | { |
140 | const char *op_str = "UNKNOWN"; | |
141 | ||
142 | if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op]) | |
143 | op_str = blk_op_name[op]; | |
144 | ||
145 | return op_str; | |
146 | } | |
147 | EXPORT_SYMBOL_GPL(blk_op_str); | |
148 | ||
2a842aca CH |
149 | static const struct { |
150 | int errno; | |
151 | const char *name; | |
152 | } blk_errors[] = { | |
153 | [BLK_STS_OK] = { 0, "" }, | |
154 | [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" }, | |
155 | [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" }, | |
156 | [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" }, | |
157 | [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" }, | |
158 | [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" }, | |
7ba15083 | 159 | [BLK_STS_RESV_CONFLICT] = { -EBADE, "reservation conflict" }, |
2a842aca CH |
160 | [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" }, |
161 | [BLK_STS_PROTECTION] = { -EILSEQ, "protection" }, | |
162 | [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" }, | |
86ff7c2a | 163 | [BLK_STS_DEV_RESOURCE] = { -EBUSY, "device resource" }, |
03a07c92 | 164 | [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" }, |
7d32c027 | 165 | [BLK_STS_OFFLINE] = { -ENODEV, "device offline" }, |
2a842aca | 166 | |
4e4cbee9 CH |
167 | /* device mapper special case, should not leak out: */ |
168 | [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" }, | |
169 | ||
3b481d91 KB |
170 | /* zone device specific errors */ |
171 | [BLK_STS_ZONE_OPEN_RESOURCE] = { -ETOOMANYREFS, "open zones exceeded" }, | |
172 | [BLK_STS_ZONE_ACTIVE_RESOURCE] = { -EOVERFLOW, "active zones exceeded" }, | |
173 | ||
dffc480d DLM |
174 | /* Command duration limit device-side timeout */ |
175 | [BLK_STS_DURATION_LIMIT] = { -ETIME, "duration limit exceeded" }, | |
176 | ||
2a842aca CH |
177 | /* everything else not covered above: */ |
178 | [BLK_STS_IOERR] = { -EIO, "I/O" }, | |
179 | }; | |
180 | ||
181 | blk_status_t errno_to_blk_status(int errno) | |
182 | { | |
183 | int i; | |
184 | ||
185 | for (i = 0; i < ARRAY_SIZE(blk_errors); i++) { | |
186 | if (blk_errors[i].errno == errno) | |
187 | return (__force blk_status_t)i; | |
188 | } | |
189 | ||
190 | return BLK_STS_IOERR; | |
191 | } | |
192 | EXPORT_SYMBOL_GPL(errno_to_blk_status); | |
193 | ||
194 | int blk_status_to_errno(blk_status_t status) | |
195 | { | |
196 | int idx = (__force int)status; | |
197 | ||
34bd9c1c | 198 | if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors))) |
2a842aca CH |
199 | return -EIO; |
200 | return blk_errors[idx].errno; | |
201 | } | |
202 | EXPORT_SYMBOL_GPL(blk_status_to_errno); | |
203 | ||
0d7a29a2 | 204 | const char *blk_status_to_str(blk_status_t status) |
2a842aca CH |
205 | { |
206 | int idx = (__force int)status; | |
207 | ||
34bd9c1c | 208 | if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors))) |
0d7a29a2 CH |
209 | return "<null>"; |
210 | return blk_errors[idx].name; | |
2a842aca | 211 | } |
7ba37927 | 212 | EXPORT_SYMBOL_GPL(blk_status_to_str); |
2a842aca | 213 | |
1da177e4 LT |
214 | /** |
215 | * blk_sync_queue - cancel any pending callbacks on a queue | |
216 | * @q: the queue | |
217 | * | |
218 | * Description: | |
219 | * The block layer may perform asynchronous callback activity | |
220 | * on a queue, such as calling the unplug function after a timeout. | |
221 | * A block device may call blk_sync_queue to ensure that any | |
222 | * such activity is cancelled, thus allowing it to release resources | |
59c51591 | 223 | * that the callbacks might use. The caller must already have made sure |
c62b37d9 | 224 | * that its ->submit_bio will not re-add plugging prior to calling |
1da177e4 LT |
225 | * this function. |
226 | * | |
da527770 | 227 | * This function does not cancel any asynchronous activity arising |
da3dae54 | 228 | * out of elevator or throttling code. That would require elevator_exit() |
5efd6113 | 229 | * and blkcg_exit_queue() to be called with queue lock initialized. |
da527770 | 230 | * |
1da177e4 LT |
231 | */ |
232 | void blk_sync_queue(struct request_queue *q) | |
233 | { | |
70ed28b9 | 234 | del_timer_sync(&q->timeout); |
4e9b6f20 | 235 | cancel_work_sync(&q->timeout_work); |
1da177e4 LT |
236 | } |
237 | EXPORT_SYMBOL(blk_sync_queue); | |
238 | ||
c9254f2d | 239 | /** |
cd84a62e | 240 | * blk_set_pm_only - increment pm_only counter |
c9254f2d | 241 | * @q: request queue pointer |
c9254f2d | 242 | */ |
cd84a62e | 243 | void blk_set_pm_only(struct request_queue *q) |
c9254f2d | 244 | { |
cd84a62e | 245 | atomic_inc(&q->pm_only); |
c9254f2d | 246 | } |
cd84a62e | 247 | EXPORT_SYMBOL_GPL(blk_set_pm_only); |
c9254f2d | 248 | |
cd84a62e | 249 | void blk_clear_pm_only(struct request_queue *q) |
c9254f2d | 250 | { |
cd84a62e BVA |
251 | int pm_only; |
252 | ||
253 | pm_only = atomic_dec_return(&q->pm_only); | |
254 | WARN_ON_ONCE(pm_only < 0); | |
255 | if (pm_only == 0) | |
256 | wake_up_all(&q->mq_freeze_wq); | |
c9254f2d | 257 | } |
cd84a62e | 258 | EXPORT_SYMBOL_GPL(blk_clear_pm_only); |
c9254f2d | 259 | |
2bd85221 CH |
260 | static void blk_free_queue_rcu(struct rcu_head *rcu_head) |
261 | { | |
d36a9ea5 ML |
262 | struct request_queue *q = container_of(rcu_head, |
263 | struct request_queue, rcu_head); | |
264 | ||
265 | percpu_ref_exit(&q->q_usage_counter); | |
266 | kmem_cache_free(blk_requestq_cachep, q); | |
2bd85221 CH |
267 | } |
268 | ||
269 | static void blk_free_queue(struct request_queue *q) | |
270 | { | |
2bd85221 | 271 | blk_free_queue_stats(q->stats); |
2bd85221 CH |
272 | if (queue_is_mq(q)) |
273 | blk_mq_release(q); | |
274 | ||
275 | ida_free(&blk_queue_ida, q->id); | |
276 | call_rcu(&q->rcu_head, blk_free_queue_rcu); | |
277 | } | |
278 | ||
b5bd357c LC |
279 | /** |
280 | * blk_put_queue - decrement the request_queue refcount | |
281 | * @q: the request_queue structure to decrement the refcount for | |
282 | * | |
2bd85221 CH |
283 | * Decrements the refcount of the request_queue and free it when the refcount |
284 | * reaches 0. | |
b5bd357c | 285 | */ |
165125e1 | 286 | void blk_put_queue(struct request_queue *q) |
483f4afc | 287 | { |
2bd85221 CH |
288 | if (refcount_dec_and_test(&q->refs)) |
289 | blk_free_queue(q); | |
483f4afc | 290 | } |
d86e0e83 | 291 | EXPORT_SYMBOL(blk_put_queue); |
483f4afc | 292 | |
8e141f9e | 293 | void blk_queue_start_drain(struct request_queue *q) |
aed3ea94 | 294 | { |
d3cfb2a0 ML |
295 | /* |
296 | * When queue DYING flag is set, we need to block new req | |
297 | * entering queue, so we call blk_freeze_queue_start() to | |
298 | * prevent I/O from crossing blk_queue_enter(). | |
299 | */ | |
300 | blk_freeze_queue_start(q); | |
344e9ffc | 301 | if (queue_is_mq(q)) |
aed3ea94 | 302 | blk_mq_wake_waiters(q); |
055f6e18 ML |
303 | /* Make blk_queue_enter() reexamine the DYING flag. */ |
304 | wake_up_all(&q->mq_freeze_wq); | |
aed3ea94 | 305 | } |
8e141f9e | 306 | |
3a0a5299 BVA |
307 | /** |
308 | * blk_queue_enter() - try to increase q->q_usage_counter | |
309 | * @q: request queue pointer | |
a4d34da7 | 310 | * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PM |
3a0a5299 | 311 | */ |
9a95e4ef | 312 | int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags) |
3ef28e83 | 313 | { |
a4d34da7 | 314 | const bool pm = flags & BLK_MQ_REQ_PM; |
3a0a5299 | 315 | |
1f14a098 | 316 | while (!blk_try_enter_queue(q, pm)) { |
3a0a5299 | 317 | if (flags & BLK_MQ_REQ_NOWAIT) |
56f99b8d | 318 | return -EAGAIN; |
3ef28e83 | 319 | |
5ed61d3f | 320 | /* |
1f14a098 CH |
321 | * read pair of barrier in blk_freeze_queue_start(), we need to |
322 | * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and | |
323 | * reading .mq_freeze_depth or queue dying flag, otherwise the | |
324 | * following wait may never return if the two reads are | |
325 | * reordered. | |
5ed61d3f ML |
326 | */ |
327 | smp_rmb(); | |
1dc3039b | 328 | wait_event(q->mq_freeze_wq, |
7996a8b5 | 329 | (!q->mq_freeze_depth && |
52abca64 | 330 | blk_pm_resume_queue(pm, q)) || |
1dc3039b | 331 | blk_queue_dying(q)); |
3ef28e83 DW |
332 | if (blk_queue_dying(q)) |
333 | return -ENODEV; | |
3ef28e83 | 334 | } |
1f14a098 CH |
335 | |
336 | return 0; | |
3ef28e83 DW |
337 | } |
338 | ||
c98cb5bb | 339 | int __bio_queue_enter(struct request_queue *q, struct bio *bio) |
accea322 | 340 | { |
a6741536 | 341 | while (!blk_try_enter_queue(q, false)) { |
eab4e027 PB |
342 | struct gendisk *disk = bio->bi_bdev->bd_disk; |
343 | ||
a6741536 | 344 | if (bio->bi_opf & REQ_NOWAIT) { |
8e141f9e | 345 | if (test_bit(GD_DEAD, &disk->state)) |
a6741536 | 346 | goto dead; |
accea322 | 347 | bio_wouldblock_error(bio); |
56f99b8d | 348 | return -EAGAIN; |
a6741536 CH |
349 | } |
350 | ||
351 | /* | |
352 | * read pair of barrier in blk_freeze_queue_start(), we need to | |
353 | * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and | |
354 | * reading .mq_freeze_depth or queue dying flag, otherwise the | |
355 | * following wait may never return if the two reads are | |
356 | * reordered. | |
357 | */ | |
358 | smp_rmb(); | |
359 | wait_event(q->mq_freeze_wq, | |
360 | (!q->mq_freeze_depth && | |
361 | blk_pm_resume_queue(false, q)) || | |
8e141f9e CH |
362 | test_bit(GD_DEAD, &disk->state)); |
363 | if (test_bit(GD_DEAD, &disk->state)) | |
a6741536 | 364 | goto dead; |
accea322 CH |
365 | } |
366 | ||
a6741536 CH |
367 | return 0; |
368 | dead: | |
369 | bio_io_error(bio); | |
370 | return -ENODEV; | |
accea322 CH |
371 | } |
372 | ||
3ef28e83 DW |
373 | void blk_queue_exit(struct request_queue *q) |
374 | { | |
375 | percpu_ref_put(&q->q_usage_counter); | |
376 | } | |
377 | ||
378 | static void blk_queue_usage_counter_release(struct percpu_ref *ref) | |
379 | { | |
380 | struct request_queue *q = | |
381 | container_of(ref, struct request_queue, q_usage_counter); | |
382 | ||
383 | wake_up_all(&q->mq_freeze_wq); | |
384 | } | |
385 | ||
bca237a5 | 386 | static void blk_rq_timed_out_timer(struct timer_list *t) |
287922eb | 387 | { |
bca237a5 | 388 | struct request_queue *q = from_timer(q, t, timeout); |
287922eb CH |
389 | |
390 | kblockd_schedule_work(&q->timeout_work); | |
391 | } | |
392 | ||
2e3c18d0 TH |
393 | static void blk_timeout_work(struct work_struct *work) |
394 | { | |
395 | } | |
396 | ||
ad751ba1 | 397 | struct request_queue *blk_alloc_queue(struct queue_limits *lim, int node_id) |
1946089a | 398 | { |
165125e1 | 399 | struct request_queue *q; |
ad751ba1 | 400 | int error; |
1946089a | 401 | |
80bd4a7a CH |
402 | q = kmem_cache_alloc_node(blk_requestq_cachep, GFP_KERNEL | __GFP_ZERO, |
403 | node_id); | |
1da177e4 | 404 | if (!q) |
ad751ba1 | 405 | return ERR_PTR(-ENOMEM); |
1da177e4 | 406 | |
cbf62af3 | 407 | q->last_merge = NULL; |
cbf62af3 | 408 | |
798f2a6f | 409 | q->id = ida_alloc(&blk_queue_ida, GFP_KERNEL); |
ad751ba1 CH |
410 | if (q->id < 0) { |
411 | error = q->id; | |
80bd4a7a | 412 | goto fail_q; |
ad751ba1 | 413 | } |
a73f730d | 414 | |
a83b576c | 415 | q->stats = blk_alloc_queue_stats(); |
ad751ba1 CH |
416 | if (!q->stats) { |
417 | error = -ENOMEM; | |
46754bd0 | 418 | goto fail_id; |
ad751ba1 CH |
419 | } |
420 | ||
421 | error = blk_set_default_limits(lim); | |
422 | if (error) | |
423 | goto fail_stats; | |
424 | q->limits = *lim; | |
a83b576c | 425 | |
5151412d | 426 | q->node = node_id; |
0989a025 | 427 | |
079a2e3e | 428 | atomic_set(&q->nr_active_requests_shared_tags, 0); |
bccf5e26 | 429 | |
bca237a5 | 430 | timer_setup(&q->timeout, blk_rq_timed_out_timer, 0); |
2e3c18d0 | 431 | INIT_WORK(&q->timeout_work, blk_timeout_work); |
a612fddf | 432 | INIT_LIST_HEAD(&q->icq_list); |
483f4afc | 433 | |
2bd85221 | 434 | refcount_set(&q->refs, 1); |
85e0cbbb | 435 | mutex_init(&q->debugfs_mutex); |
483f4afc | 436 | mutex_init(&q->sysfs_lock); |
cecf5d87 | 437 | mutex_init(&q->sysfs_dir_lock); |
d690cb8a | 438 | mutex_init(&q->limits_lock); |
a13bd91b | 439 | mutex_init(&q->rq_qos_mutex); |
0d945c1f | 440 | spin_lock_init(&q->queue_lock); |
c94a96ac | 441 | |
320ae51f | 442 | init_waitqueue_head(&q->mq_freeze_wq); |
7996a8b5 | 443 | mutex_init(&q->mq_freeze_lock); |
320ae51f | 444 | |
8b8ace08 ML |
445 | blkg_init_queue(q); |
446 | ||
3ef28e83 DW |
447 | /* |
448 | * Init percpu_ref in atomic mode so that it's faster to shutdown. | |
449 | * See blk_register_queue() for details. | |
450 | */ | |
ad751ba1 | 451 | error = percpu_ref_init(&q->q_usage_counter, |
3ef28e83 | 452 | blk_queue_usage_counter_release, |
ad751ba1 CH |
453 | PERCPU_REF_INIT_ATOMIC, GFP_KERNEL); |
454 | if (error) | |
edb0872f | 455 | goto fail_stats; |
f51b802c | 456 | |
d2a27964 | 457 | q->nr_requests = BLKDEV_DEFAULT_RQ; |
3d745ea5 | 458 | |
1da177e4 | 459 | return q; |
a73f730d | 460 | |
a83b576c | 461 | fail_stats: |
edb0872f | 462 | blk_free_queue_stats(q->stats); |
a73f730d | 463 | fail_id: |
798f2a6f | 464 | ida_free(&blk_queue_ida, q->id); |
a73f730d | 465 | fail_q: |
80bd4a7a | 466 | kmem_cache_free(blk_requestq_cachep, q); |
ad751ba1 | 467 | return ERR_PTR(error); |
1da177e4 | 468 | } |
1da177e4 | 469 | |
b5bd357c LC |
470 | /** |
471 | * blk_get_queue - increment the request_queue refcount | |
472 | * @q: the request_queue structure to increment the refcount for | |
473 | * | |
474 | * Increment the refcount of the request_queue kobject. | |
763b5892 LC |
475 | * |
476 | * Context: Any context. | |
b5bd357c | 477 | */ |
09ac46c4 | 478 | bool blk_get_queue(struct request_queue *q) |
1da177e4 | 479 | { |
828b5f01 CH |
480 | if (unlikely(blk_queue_dying(q))) |
481 | return false; | |
2bd85221 | 482 | refcount_inc(&q->refs); |
828b5f01 | 483 | return true; |
1da177e4 | 484 | } |
d86e0e83 | 485 | EXPORT_SYMBOL(blk_get_queue); |
1da177e4 | 486 | |
c17bb495 AM |
487 | #ifdef CONFIG_FAIL_MAKE_REQUEST |
488 | ||
489 | static DECLARE_FAULT_ATTR(fail_make_request); | |
490 | ||
491 | static int __init setup_fail_make_request(char *str) | |
492 | { | |
493 | return setup_fault_attr(&fail_make_request, str); | |
494 | } | |
495 | __setup("fail_make_request=", setup_fail_make_request); | |
496 | ||
06c8c691 | 497 | bool should_fail_request(struct block_device *part, unsigned int bytes) |
c17bb495 | 498 | { |
811ba89a AV |
499 | return bdev_test_flag(part, BD_MAKE_IT_FAIL) && |
500 | should_fail(&fail_make_request, bytes); | |
c17bb495 AM |
501 | } |
502 | ||
503 | static int __init fail_make_request_debugfs(void) | |
504 | { | |
dd48c085 AM |
505 | struct dentry *dir = fault_create_debugfs_attr("fail_make_request", |
506 | NULL, &fail_make_request); | |
507 | ||
21f9fcd8 | 508 | return PTR_ERR_OR_ZERO(dir); |
c17bb495 AM |
509 | } |
510 | ||
511 | late_initcall(fail_make_request_debugfs); | |
c17bb495 AM |
512 | #endif /* CONFIG_FAIL_MAKE_REQUEST */ |
513 | ||
bdb7d420 | 514 | static inline void bio_check_ro(struct bio *bio) |
721c7fc7 | 515 | { |
2f9f6221 | 516 | if (op_is_write(bio_op(bio)) && bdev_read_only(bio->bi_bdev)) { |
8b2ded1c | 517 | if (op_is_flush(bio->bi_opf) && !bio_sectors(bio)) |
bdb7d420 | 518 | return; |
67d995e0 | 519 | |
49a43dae | 520 | if (bdev_test_flag(bio->bi_bdev, BD_RO_WARNED)) |
67d995e0 YK |
521 | return; |
522 | ||
49a43dae AV |
523 | bdev_set_flag(bio->bi_bdev, BD_RO_WARNED); |
524 | ||
67d995e0 YK |
525 | /* |
526 | * Use ioctl to set underlying disk of raid/dm to read-only | |
527 | * will trigger this. | |
528 | */ | |
529 | pr_warn("Trying to write to read-only block-device %pg\n", | |
530 | bio->bi_bdev); | |
721c7fc7 | 531 | } |
721c7fc7 ID |
532 | } |
533 | ||
30abb3a6 HM |
534 | static noinline int should_fail_bio(struct bio *bio) |
535 | { | |
309dca30 | 536 | if (should_fail_request(bdev_whole(bio->bi_bdev), bio->bi_iter.bi_size)) |
30abb3a6 HM |
537 | return -EIO; |
538 | return 0; | |
539 | } | |
540 | ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO); | |
541 | ||
52c5e62d CH |
542 | /* |
543 | * Check whether this bio extends beyond the end of the device or partition. | |
544 | * This may well happen - the kernel calls bread() without checking the size of | |
545 | * the device, e.g., when mounting a file system. | |
546 | */ | |
2f9f6221 | 547 | static inline int bio_check_eod(struct bio *bio) |
52c5e62d | 548 | { |
2f9f6221 | 549 | sector_t maxsector = bdev_nr_sectors(bio->bi_bdev); |
52c5e62d CH |
550 | unsigned int nr_sectors = bio_sectors(bio); |
551 | ||
3eb96946 | 552 | if (nr_sectors && |
52c5e62d CH |
553 | (nr_sectors > maxsector || |
554 | bio->bi_iter.bi_sector > maxsector - nr_sectors)) { | |
ad740780 | 555 | pr_info_ratelimited("%s: attempt to access beyond end of device\n" |
069adbac CH |
556 | "%pg: rw=%d, sector=%llu, nr_sectors = %u limit=%llu\n", |
557 | current->comm, bio->bi_bdev, bio->bi_opf, | |
558 | bio->bi_iter.bi_sector, nr_sectors, maxsector); | |
52c5e62d CH |
559 | return -EIO; |
560 | } | |
561 | return 0; | |
562 | } | |
563 | ||
74d46992 CH |
564 | /* |
565 | * Remap block n of partition p to block n+start(p) of the disk. | |
566 | */ | |
2f9f6221 | 567 | static int blk_partition_remap(struct bio *bio) |
74d46992 | 568 | { |
309dca30 | 569 | struct block_device *p = bio->bi_bdev; |
74d46992 | 570 | |
52c5e62d | 571 | if (unlikely(should_fail_request(p, bio->bi_iter.bi_size))) |
2f9f6221 | 572 | return -EIO; |
5eac3eb3 | 573 | if (bio_sectors(bio)) { |
8446fe92 | 574 | bio->bi_iter.bi_sector += p->bd_start_sect; |
1c02fca6 | 575 | trace_block_bio_remap(bio, p->bd_dev, |
29ff57c6 | 576 | bio->bi_iter.bi_sector - |
8446fe92 | 577 | p->bd_start_sect); |
52c5e62d | 578 | } |
30c5d345 | 579 | bio_set_flag(bio, BIO_REMAPPED); |
2f9f6221 | 580 | return 0; |
74d46992 CH |
581 | } |
582 | ||
0512a75b KB |
583 | /* |
584 | * Check write append to a zoned block device. | |
585 | */ | |
586 | static inline blk_status_t blk_check_zone_append(struct request_queue *q, | |
587 | struct bio *bio) | |
588 | { | |
0512a75b KB |
589 | int nr_sectors = bio_sectors(bio); |
590 | ||
591 | /* Only applicable to zoned block devices */ | |
edd1dbc8 | 592 | if (!bdev_is_zoned(bio->bi_bdev)) |
0512a75b KB |
593 | return BLK_STS_NOTSUPP; |
594 | ||
595 | /* The bio sector must point to the start of a sequential zone */ | |
bca150f0 | 596 | if (!bdev_is_zone_start(bio->bi_bdev, bio->bi_iter.bi_sector)) |
0512a75b KB |
597 | return BLK_STS_IOERR; |
598 | ||
599 | /* | |
600 | * Not allowed to cross zone boundaries. Otherwise, the BIO will be | |
601 | * split and could result in non-contiguous sectors being written in | |
602 | * different zones. | |
603 | */ | |
604 | if (nr_sectors > q->limits.chunk_sectors) | |
605 | return BLK_STS_IOERR; | |
606 | ||
607 | /* Make sure the BIO is small enough and will not get split */ | |
ccdbf0aa | 608 | if (nr_sectors > queue_max_zone_append_sectors(q)) |
0512a75b KB |
609 | return BLK_STS_IOERR; |
610 | ||
611 | bio->bi_opf |= REQ_NOMERGE; | |
612 | ||
613 | return BLK_STS_OK; | |
614 | } | |
615 | ||
900e0807 JA |
616 | static void __submit_bio(struct bio *bio) |
617 | { | |
9a42891c YK |
618 | /* If plug is not used, add new plug here to cache nsecs time. */ |
619 | struct blk_plug plug; | |
620 | ||
7f36b7d0 ML |
621 | if (unlikely(!blk_crypto_bio_prep(&bio))) |
622 | return; | |
623 | ||
9a42891c YK |
624 | blk_start_plug(&plug); |
625 | ||
ac2b6f9d | 626 | if (!bdev_test_flag(bio->bi_bdev, BD_HAS_SUBMIT_BIO)) { |
3e08773c | 627 | blk_mq_submit_bio(bio); |
7f36b7d0 | 628 | } else if (likely(bio_queue_enter(bio) == 0)) { |
9f4107b0 JA |
629 | struct gendisk *disk = bio->bi_bdev->bd_disk; |
630 | ||
7f36b7d0 ML |
631 | disk->fops->submit_bio(bio); |
632 | blk_queue_exit(disk->queue); | |
633 | } | |
9a42891c YK |
634 | |
635 | blk_finish_plug(&plug); | |
ac7c5675 CH |
636 | } |
637 | ||
566acf2d CH |
638 | /* |
639 | * The loop in this function may be a bit non-obvious, and so deserves some | |
640 | * explanation: | |
641 | * | |
642 | * - Before entering the loop, bio->bi_next is NULL (as all callers ensure | |
643 | * that), so we have a list with a single bio. | |
644 | * - We pretend that we have just taken it off a longer list, so we assign | |
645 | * bio_list to a pointer to the bio_list_on_stack, thus initialising the | |
646 | * bio_list of new bios to be added. ->submit_bio() may indeed add some more | |
647 | * bios through a recursive call to submit_bio_noacct. If it did, we find a | |
648 | * non-NULL value in bio_list and re-enter the loop from the top. | |
649 | * - In this case we really did just take the bio of the top of the list (no | |
650 | * pretending) and so remove it from bio_list, and call into ->submit_bio() | |
651 | * again. | |
652 | * | |
653 | * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio. | |
654 | * bio_list_on_stack[1] contains bios that were submitted before the current | |
69fe0f29 | 655 | * ->submit_bio, but that haven't been processed yet. |
566acf2d | 656 | */ |
3e08773c | 657 | static void __submit_bio_noacct(struct bio *bio) |
566acf2d CH |
658 | { |
659 | struct bio_list bio_list_on_stack[2]; | |
566acf2d CH |
660 | |
661 | BUG_ON(bio->bi_next); | |
662 | ||
663 | bio_list_init(&bio_list_on_stack[0]); | |
664 | current->bio_list = bio_list_on_stack; | |
665 | ||
666 | do { | |
eab4e027 | 667 | struct request_queue *q = bdev_get_queue(bio->bi_bdev); |
566acf2d CH |
668 | struct bio_list lower, same; |
669 | ||
566acf2d CH |
670 | /* |
671 | * Create a fresh bio_list for all subordinate requests. | |
672 | */ | |
673 | bio_list_on_stack[1] = bio_list_on_stack[0]; | |
674 | bio_list_init(&bio_list_on_stack[0]); | |
675 | ||
3e08773c | 676 | __submit_bio(bio); |
566acf2d CH |
677 | |
678 | /* | |
679 | * Sort new bios into those for a lower level and those for the | |
680 | * same level. | |
681 | */ | |
682 | bio_list_init(&lower); | |
683 | bio_list_init(&same); | |
684 | while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL) | |
eab4e027 | 685 | if (q == bdev_get_queue(bio->bi_bdev)) |
566acf2d CH |
686 | bio_list_add(&same, bio); |
687 | else | |
688 | bio_list_add(&lower, bio); | |
689 | ||
690 | /* | |
691 | * Now assemble so we handle the lowest level first. | |
692 | */ | |
693 | bio_list_merge(&bio_list_on_stack[0], &lower); | |
694 | bio_list_merge(&bio_list_on_stack[0], &same); | |
695 | bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]); | |
696 | } while ((bio = bio_list_pop(&bio_list_on_stack[0]))); | |
697 | ||
698 | current->bio_list = NULL; | |
566acf2d CH |
699 | } |
700 | ||
3e08773c | 701 | static void __submit_bio_noacct_mq(struct bio *bio) |
ff93ea0c | 702 | { |
7c792f33 | 703 | struct bio_list bio_list[2] = { }; |
ff93ea0c | 704 | |
7c792f33 | 705 | current->bio_list = bio_list; |
ff93ea0c CH |
706 | |
707 | do { | |
3e08773c | 708 | __submit_bio(bio); |
7c792f33 | 709 | } while ((bio = bio_list_pop(&bio_list[0]))); |
ff93ea0c CH |
710 | |
711 | current->bio_list = NULL; | |
ff93ea0c CH |
712 | } |
713 | ||
3f98c753 | 714 | void submit_bio_noacct_nocheck(struct bio *bio) |
d89d8796 | 715 | { |
0f7c8f0f JH |
716 | blk_cgroup_bio_start(bio); |
717 | blkcg_bio_issue_init(bio); | |
718 | ||
719 | if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) { | |
720 | trace_block_bio_queue(bio); | |
721 | /* | |
722 | * Now that enqueuing has been traced, we need to trace | |
723 | * completion as well. | |
724 | */ | |
725 | bio_set_flag(bio, BIO_TRACE_COMPLETION); | |
726 | } | |
727 | ||
27a84d54 | 728 | /* |
566acf2d CH |
729 | * We only want one ->submit_bio to be active at a time, else stack |
730 | * usage with stacked devices could be a problem. Use current->bio_list | |
731 | * to collect a list of requests submited by a ->submit_bio method while | |
732 | * it is active, and then process them after it returned. | |
27a84d54 | 733 | */ |
3e08773c | 734 | if (current->bio_list) |
f5fe1b51 | 735 | bio_list_add(¤t->bio_list[0], bio); |
ac2b6f9d | 736 | else if (!bdev_test_flag(bio->bi_bdev, BD_HAS_SUBMIT_BIO)) |
3e08773c CH |
737 | __submit_bio_noacct_mq(bio); |
738 | else | |
739 | __submit_bio_noacct(bio); | |
d89d8796 | 740 | } |
3f98c753 ML |
741 | |
742 | /** | |
743 | * submit_bio_noacct - re-submit a bio to the block device layer for I/O | |
744 | * @bio: The bio describing the location in memory and on the device. | |
745 | * | |
746 | * This is a version of submit_bio() that shall only be used for I/O that is | |
747 | * resubmitted to lower level drivers by stacking block drivers. All file | |
748 | * systems and other upper level users of the block layer should use | |
749 | * submit_bio() instead. | |
750 | */ | |
751 | void submit_bio_noacct(struct bio *bio) | |
1da177e4 | 752 | { |
309dca30 | 753 | struct block_device *bdev = bio->bi_bdev; |
eab4e027 | 754 | struct request_queue *q = bdev_get_queue(bdev); |
4e4cbee9 | 755 | blk_status_t status = BLK_STS_IOERR; |
1da177e4 LT |
756 | |
757 | might_sleep(); | |
1da177e4 | 758 | |
03a07c92 | 759 | /* |
b0beb280 | 760 | * For a REQ_NOWAIT based request, return -EOPNOTSUPP |
021a2446 | 761 | * if queue does not support NOWAIT. |
03a07c92 | 762 | */ |
568ec936 | 763 | if ((bio->bi_opf & REQ_NOWAIT) && !bdev_nowait(bdev)) |
b0beb280 | 764 | goto not_supported; |
03a07c92 | 765 | |
30abb3a6 | 766 | if (should_fail_bio(bio)) |
5a7bbad2 | 767 | goto end_io; |
bdb7d420 | 768 | bio_check_ro(bio); |
3a905c37 CH |
769 | if (!bio_flagged(bio, BIO_REMAPPED)) { |
770 | if (unlikely(bio_check_eod(bio))) | |
771 | goto end_io; | |
3f9b8fb4 AV |
772 | if (bdev_is_partition(bdev) && |
773 | unlikely(blk_partition_remap(bio))) | |
3a905c37 CH |
774 | goto end_io; |
775 | } | |
2056a782 | 776 | |
5a7bbad2 | 777 | /* |
ed00aabd CH |
778 | * Filter flush bio's early so that bio based drivers without flush |
779 | * support don't have to worry about them. | |
5a7bbad2 | 780 | */ |
b4a6bb3a CH |
781 | if (op_is_flush(bio->bi_opf)) { |
782 | if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_WRITE && | |
783 | bio_op(bio) != REQ_OP_ZONE_APPEND)) | |
51fd77bd | 784 | goto end_io; |
b4a6bb3a CH |
785 | if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags)) { |
786 | bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA); | |
787 | if (!bio_sectors(bio)) { | |
788 | status = BLK_STS_OK; | |
789 | goto end_io; | |
790 | } | |
51fd77bd | 791 | } |
5a7bbad2 | 792 | } |
5ddfe969 | 793 | |
d04c406f | 794 | if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags)) |
6ce913fe | 795 | bio_clear_polled(bio); |
d04c406f | 796 | |
288dab8a | 797 | switch (bio_op(bio)) { |
1c042f8d CH |
798 | case REQ_OP_READ: |
799 | case REQ_OP_WRITE: | |
800 | break; | |
801 | case REQ_OP_FLUSH: | |
802 | /* | |
803 | * REQ_OP_FLUSH can't be submitted through bios, it is only | |
804 | * synthetized in struct request by the flush state machine. | |
805 | */ | |
806 | goto not_supported; | |
288dab8a | 807 | case REQ_OP_DISCARD: |
70200574 | 808 | if (!bdev_max_discard_sectors(bdev)) |
288dab8a CH |
809 | goto not_supported; |
810 | break; | |
811 | case REQ_OP_SECURE_ERASE: | |
44abff2c | 812 | if (!bdev_max_secure_erase_sectors(bdev)) |
288dab8a CH |
813 | goto not_supported; |
814 | break; | |
0512a75b KB |
815 | case REQ_OP_ZONE_APPEND: |
816 | status = blk_check_zone_append(q, bio); | |
817 | if (status != BLK_STS_OK) | |
818 | goto end_io; | |
819 | break; | |
1c042f8d CH |
820 | case REQ_OP_WRITE_ZEROES: |
821 | if (!q->limits.max_write_zeroes_sectors) | |
822 | goto not_supported; | |
823 | break; | |
2d253440 | 824 | case REQ_OP_ZONE_RESET: |
6c1b1da5 AJ |
825 | case REQ_OP_ZONE_OPEN: |
826 | case REQ_OP_ZONE_CLOSE: | |
827 | case REQ_OP_ZONE_FINISH: | |
edd1dbc8 | 828 | if (!bdev_is_zoned(bio->bi_bdev)) |
2d253440 | 829 | goto not_supported; |
288dab8a | 830 | break; |
6e33dbf2 | 831 | case REQ_OP_ZONE_RESET_ALL: |
edd1dbc8 | 832 | if (!bdev_is_zoned(bio->bi_bdev) || !blk_queue_zone_resetall(q)) |
6e33dbf2 CK |
833 | goto not_supported; |
834 | break; | |
1c042f8d CH |
835 | case REQ_OP_DRV_IN: |
836 | case REQ_OP_DRV_OUT: | |
837 | /* | |
838 | * Driver private operations are only used with passthrough | |
839 | * requests. | |
840 | */ | |
841 | fallthrough; | |
288dab8a | 842 | default: |
1c042f8d | 843 | goto not_supported; |
5a7bbad2 | 844 | } |
01edede4 | 845 | |
b781d8db | 846 | if (blk_throtl_bio(bio)) |
3f98c753 ML |
847 | return; |
848 | submit_bio_noacct_nocheck(bio); | |
d24c670e | 849 | return; |
a7384677 | 850 | |
288dab8a | 851 | not_supported: |
4e4cbee9 | 852 | status = BLK_STS_NOTSUPP; |
a7384677 | 853 | end_io: |
4e4cbee9 | 854 | bio->bi_status = status; |
4246a0b6 | 855 | bio_endio(bio); |
d89d8796 | 856 | } |
ed00aabd | 857 | EXPORT_SYMBOL(submit_bio_noacct); |
1da177e4 | 858 | |
f3c89983 HJ |
859 | static void bio_set_ioprio(struct bio *bio) |
860 | { | |
861 | /* Nobody set ioprio so far? Initialize it based on task's nice value */ | |
862 | if (IOPRIO_PRIO_CLASS(bio->bi_ioprio) == IOPRIO_CLASS_NONE) | |
863 | bio->bi_ioprio = get_current_ioprio(); | |
864 | blkcg_set_ioprio(bio); | |
865 | } | |
866 | ||
1da177e4 | 867 | /** |
710027a4 | 868 | * submit_bio - submit a bio to the block device layer for I/O |
1da177e4 LT |
869 | * @bio: The &struct bio which describes the I/O |
870 | * | |
3fdd4086 CH |
871 | * submit_bio() is used to submit I/O requests to block devices. It is passed a |
872 | * fully set up &struct bio that describes the I/O that needs to be done. The | |
309dca30 | 873 | * bio will be send to the device described by the bi_bdev field. |
1da177e4 | 874 | * |
3fdd4086 CH |
875 | * The success/failure status of the request, along with notification of |
876 | * completion, is delivered asynchronously through the ->bi_end_io() callback | |
e8848087 | 877 | * in @bio. The bio must NOT be touched by the caller until ->bi_end_io() has |
3fdd4086 | 878 | * been called. |
1da177e4 | 879 | */ |
3e08773c | 880 | void submit_bio(struct bio *bio) |
1da177e4 | 881 | { |
a3e7689b CH |
882 | if (bio_op(bio) == REQ_OP_READ) { |
883 | task_io_account_read(bio->bi_iter.bi_size); | |
884 | count_vm_events(PGPGIN, bio_sectors(bio)); | |
885 | } else if (bio_op(bio) == REQ_OP_WRITE) { | |
886 | count_vm_events(PGPGOUT, bio_sectors(bio)); | |
1da177e4 LT |
887 | } |
888 | ||
f3c89983 | 889 | bio_set_ioprio(bio); |
3e08773c | 890 | submit_bio_noacct(bio); |
1da177e4 | 891 | } |
1da177e4 LT |
892 | EXPORT_SYMBOL(submit_bio); |
893 | ||
3e08773c CH |
894 | /** |
895 | * bio_poll - poll for BIO completions | |
896 | * @bio: bio to poll for | |
e30028ac | 897 | * @iob: batches of IO |
3e08773c CH |
898 | * @flags: BLK_POLL_* flags that control the behavior |
899 | * | |
900 | * Poll for completions on queue associated with the bio. Returns number of | |
901 | * completed entries found. | |
902 | * | |
903 | * Note: the caller must either be the context that submitted @bio, or | |
904 | * be in a RCU critical section to prevent freeing of @bio. | |
905 | */ | |
5a72e899 | 906 | int bio_poll(struct bio *bio, struct io_comp_batch *iob, unsigned int flags) |
3e08773c | 907 | { |
3e08773c | 908 | blk_qc_t cookie = READ_ONCE(bio->bi_cookie); |
310726c3 JA |
909 | struct block_device *bdev; |
910 | struct request_queue *q; | |
69fe0f29 | 911 | int ret = 0; |
3e08773c | 912 | |
310726c3 JA |
913 | bdev = READ_ONCE(bio->bi_bdev); |
914 | if (!bdev) | |
915 | return 0; | |
916 | ||
917 | q = bdev_get_queue(bdev); | |
3e08773c CH |
918 | if (cookie == BLK_QC_T_NONE || |
919 | !test_bit(QUEUE_FLAG_POLL, &q->queue_flags)) | |
920 | return 0; | |
921 | ||
aa8dccca | 922 | blk_flush_plug(current->plug, false); |
3e08773c | 923 | |
33391eec JA |
924 | /* |
925 | * We need to be able to enter a frozen queue, similar to how | |
926 | * timeouts also need to do that. If that is blocked, then we can | |
927 | * have pending IO when a queue freeze is started, and then the | |
928 | * wait for the freeze to finish will wait for polled requests to | |
929 | * timeout as the poller is preventer from entering the queue and | |
930 | * completing them. As long as we prevent new IO from being queued, | |
931 | * that should be all that matters. | |
932 | */ | |
933 | if (!percpu_ref_tryget(&q->q_usage_counter)) | |
3e08773c | 934 | return 0; |
69fe0f29 | 935 | if (queue_is_mq(q)) { |
5a72e899 | 936 | ret = blk_mq_poll(q, cookie, iob, flags); |
69fe0f29 ML |
937 | } else { |
938 | struct gendisk *disk = q->disk; | |
939 | ||
940 | if (disk && disk->fops->poll_bio) | |
941 | ret = disk->fops->poll_bio(bio, iob, flags); | |
942 | } | |
3e08773c CH |
943 | blk_queue_exit(q); |
944 | return ret; | |
945 | } | |
946 | EXPORT_SYMBOL_GPL(bio_poll); | |
947 | ||
948 | /* | |
949 | * Helper to implement file_operations.iopoll. Requires the bio to be stored | |
950 | * in iocb->private, and cleared before freeing the bio. | |
951 | */ | |
5a72e899 JA |
952 | int iocb_bio_iopoll(struct kiocb *kiocb, struct io_comp_batch *iob, |
953 | unsigned int flags) | |
3e08773c CH |
954 | { |
955 | struct bio *bio; | |
956 | int ret = 0; | |
957 | ||
958 | /* | |
959 | * Note: the bio cache only uses SLAB_TYPESAFE_BY_RCU, so bio can | |
960 | * point to a freshly allocated bio at this point. If that happens | |
961 | * we have a few cases to consider: | |
962 | * | |
963 | * 1) the bio is beeing initialized and bi_bdev is NULL. We can just | |
964 | * simply nothing in this case | |
965 | * 2) the bio points to a not poll enabled device. bio_poll will catch | |
966 | * this and return 0 | |
967 | * 3) the bio points to a poll capable device, including but not | |
968 | * limited to the one that the original bio pointed to. In this | |
969 | * case we will call into the actual poll method and poll for I/O, | |
970 | * even if we don't need to, but it won't cause harm either. | |
971 | * | |
972 | * For cases 2) and 3) above the RCU grace period ensures that bi_bdev | |
973 | * is still allocated. Because partitions hold a reference to the whole | |
974 | * device bdev and thus disk, the disk is also still valid. Grabbing | |
975 | * a reference to the queue in bio_poll() ensures the hctxs and requests | |
976 | * are still valid as well. | |
977 | */ | |
978 | rcu_read_lock(); | |
979 | bio = READ_ONCE(kiocb->private); | |
310726c3 | 980 | if (bio) |
5a72e899 | 981 | ret = bio_poll(bio, iob, flags); |
3e08773c CH |
982 | rcu_read_unlock(); |
983 | ||
984 | return ret; | |
985 | } | |
986 | EXPORT_SYMBOL_GPL(iocb_bio_iopoll); | |
987 | ||
450b7879 | 988 | void update_io_ticks(struct block_device *part, unsigned long now, bool end) |
9123bf6f CH |
989 | { |
990 | unsigned long stamp; | |
991 | again: | |
8446fe92 | 992 | stamp = READ_ONCE(part->bd_stamp); |
99dc4223 YK |
993 | if (unlikely(time_after(now, stamp)) && |
994 | likely(try_cmpxchg(&part->bd_stamp, &stamp, now)) && | |
995 | (end || part_in_flight(part))) | |
996 | __part_stat_add(part, io_ticks, now - stamp); | |
997 | ||
3f9b8fb4 | 998 | if (bdev_is_partition(part)) { |
8446fe92 | 999 | part = bdev_whole(part); |
9123bf6f CH |
1000 | goto again; |
1001 | } | |
1002 | } | |
1003 | ||
5f275713 | 1004 | unsigned long bdev_start_io_acct(struct block_device *bdev, enum req_op op, |
5f0614a5 | 1005 | unsigned long start_time) |
956d510e | 1006 | { |
956d510e | 1007 | part_stat_lock(); |
5f0614a5 | 1008 | update_io_ticks(bdev, start_time, false); |
5f0614a5 | 1009 | part_stat_local_inc(bdev, in_flight[op_is_write(op)]); |
956d510e | 1010 | part_stat_unlock(); |
320ae51f | 1011 | |
e45c47d1 MS |
1012 | return start_time; |
1013 | } | |
5f0614a5 | 1014 | EXPORT_SYMBOL(bdev_start_io_acct); |
e45c47d1 | 1015 | |
99dfc43e CH |
1016 | /** |
1017 | * bio_start_io_acct - start I/O accounting for bio based drivers | |
1018 | * @bio: bio to start account for | |
1019 | * | |
1020 | * Returns the start time that should be passed back to bio_end_io_acct(). | |
1021 | */ | |
1022 | unsigned long bio_start_io_acct(struct bio *bio) | |
7b26410b | 1023 | { |
5f275713 | 1024 | return bdev_start_io_acct(bio->bi_bdev, bio_op(bio), jiffies); |
7b26410b | 1025 | } |
99dfc43e | 1026 | EXPORT_SYMBOL_GPL(bio_start_io_acct); |
7b26410b | 1027 | |
77e7ffd7 | 1028 | void bdev_end_io_acct(struct block_device *bdev, enum req_op op, |
5f275713 | 1029 | unsigned int sectors, unsigned long start_time) |
956d510e | 1030 | { |
956d510e CH |
1031 | const int sgrp = op_stat_group(op); |
1032 | unsigned long now = READ_ONCE(jiffies); | |
1033 | unsigned long duration = now - start_time; | |
5b18b5a7 | 1034 | |
956d510e | 1035 | part_stat_lock(); |
5f0614a5 | 1036 | update_io_ticks(bdev, now, true); |
5f275713 YK |
1037 | part_stat_inc(bdev, ios[sgrp]); |
1038 | part_stat_add(bdev, sectors[sgrp], sectors); | |
5f0614a5 ML |
1039 | part_stat_add(bdev, nsecs[sgrp], jiffies_to_nsecs(duration)); |
1040 | part_stat_local_dec(bdev, in_flight[op_is_write(op)]); | |
320ae51f JA |
1041 | part_stat_unlock(); |
1042 | } | |
5f0614a5 | 1043 | EXPORT_SYMBOL(bdev_end_io_acct); |
7b26410b | 1044 | |
99dfc43e | 1045 | void bio_end_io_acct_remapped(struct bio *bio, unsigned long start_time, |
5f0614a5 | 1046 | struct block_device *orig_bdev) |
7b26410b | 1047 | { |
5f275713 | 1048 | bdev_end_io_acct(orig_bdev, bio_op(bio), bio_sectors(bio), start_time); |
7b26410b | 1049 | } |
99dfc43e | 1050 | EXPORT_SYMBOL_GPL(bio_end_io_acct_remapped); |
7b26410b | 1051 | |
ef9e3fac KU |
1052 | /** |
1053 | * blk_lld_busy - Check if underlying low-level drivers of a device are busy | |
1054 | * @q : the queue of the device being checked | |
1055 | * | |
1056 | * Description: | |
1057 | * Check if underlying low-level drivers of a device are busy. | |
1058 | * If the drivers want to export their busy state, they must set own | |
1059 | * exporting function using blk_queue_lld_busy() first. | |
1060 | * | |
1061 | * Basically, this function is used only by request stacking drivers | |
1062 | * to stop dispatching requests to underlying devices when underlying | |
1063 | * devices are busy. This behavior helps more I/O merging on the queue | |
1064 | * of the request stacking driver and prevents I/O throughput regression | |
1065 | * on burst I/O load. | |
1066 | * | |
1067 | * Return: | |
1068 | * 0 - Not busy (The request stacking driver should dispatch request) | |
1069 | * 1 - Busy (The request stacking driver should stop dispatching request) | |
1070 | */ | |
1071 | int blk_lld_busy(struct request_queue *q) | |
1072 | { | |
344e9ffc | 1073 | if (queue_is_mq(q) && q->mq_ops->busy) |
9ba20527 | 1074 | return q->mq_ops->busy(q); |
ef9e3fac KU |
1075 | |
1076 | return 0; | |
1077 | } | |
1078 | EXPORT_SYMBOL_GPL(blk_lld_busy); | |
1079 | ||
59c3d45e | 1080 | int kblockd_schedule_work(struct work_struct *work) |
1da177e4 LT |
1081 | { |
1082 | return queue_work(kblockd_workqueue, work); | |
1083 | } | |
1da177e4 LT |
1084 | EXPORT_SYMBOL(kblockd_schedule_work); |
1085 | ||
818cd1cb JA |
1086 | int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork, |
1087 | unsigned long delay) | |
1088 | { | |
1089 | return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay); | |
1090 | } | |
1091 | EXPORT_SYMBOL(kblockd_mod_delayed_work_on); | |
1092 | ||
47c122e3 JA |
1093 | void blk_start_plug_nr_ios(struct blk_plug *plug, unsigned short nr_ios) |
1094 | { | |
1095 | struct task_struct *tsk = current; | |
1096 | ||
1097 | /* | |
1098 | * If this is a nested plug, don't actually assign it. | |
1099 | */ | |
1100 | if (tsk->plug) | |
1101 | return; | |
1102 | ||
da4c8c3d | 1103 | plug->cur_ktime = 0; |
bc490f81 | 1104 | plug->mq_list = NULL; |
47c122e3 JA |
1105 | plug->cached_rq = NULL; |
1106 | plug->nr_ios = min_t(unsigned short, nr_ios, BLK_MAX_REQUEST_COUNT); | |
1107 | plug->rq_count = 0; | |
1108 | plug->multiple_queues = false; | |
dc5fc361 | 1109 | plug->has_elevator = false; |
47c122e3 JA |
1110 | INIT_LIST_HEAD(&plug->cb_list); |
1111 | ||
1112 | /* | |
1113 | * Store ordering should not be needed here, since a potential | |
1114 | * preempt will imply a full memory barrier | |
1115 | */ | |
1116 | tsk->plug = plug; | |
1117 | } | |
1118 | ||
75df7136 SJ |
1119 | /** |
1120 | * blk_start_plug - initialize blk_plug and track it inside the task_struct | |
1121 | * @plug: The &struct blk_plug that needs to be initialized | |
1122 | * | |
1123 | * Description: | |
40405851 JM |
1124 | * blk_start_plug() indicates to the block layer an intent by the caller |
1125 | * to submit multiple I/O requests in a batch. The block layer may use | |
1126 | * this hint to defer submitting I/Os from the caller until blk_finish_plug() | |
1127 | * is called. However, the block layer may choose to submit requests | |
1128 | * before a call to blk_finish_plug() if the number of queued I/Os | |
1129 | * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than | |
1130 | * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if | |
1131 | * the task schedules (see below). | |
1132 | * | |
75df7136 SJ |
1133 | * Tracking blk_plug inside the task_struct will help with auto-flushing the |
1134 | * pending I/O should the task end up blocking between blk_start_plug() and | |
1135 | * blk_finish_plug(). This is important from a performance perspective, but | |
1136 | * also ensures that we don't deadlock. For instance, if the task is blocking | |
1137 | * for a memory allocation, memory reclaim could end up wanting to free a | |
1138 | * page belonging to that request that is currently residing in our private | |
1139 | * plug. By flushing the pending I/O when the process goes to sleep, we avoid | |
1140 | * this kind of deadlock. | |
1141 | */ | |
73c10101 JA |
1142 | void blk_start_plug(struct blk_plug *plug) |
1143 | { | |
47c122e3 | 1144 | blk_start_plug_nr_ios(plug, 1); |
73c10101 JA |
1145 | } |
1146 | EXPORT_SYMBOL(blk_start_plug); | |
1147 | ||
74018dc3 | 1148 | static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule) |
048c9374 N |
1149 | { |
1150 | LIST_HEAD(callbacks); | |
1151 | ||
2a7d5559 SL |
1152 | while (!list_empty(&plug->cb_list)) { |
1153 | list_splice_init(&plug->cb_list, &callbacks); | |
048c9374 | 1154 | |
2a7d5559 SL |
1155 | while (!list_empty(&callbacks)) { |
1156 | struct blk_plug_cb *cb = list_first_entry(&callbacks, | |
048c9374 N |
1157 | struct blk_plug_cb, |
1158 | list); | |
2a7d5559 | 1159 | list_del(&cb->list); |
74018dc3 | 1160 | cb->callback(cb, from_schedule); |
2a7d5559 | 1161 | } |
048c9374 N |
1162 | } |
1163 | } | |
1164 | ||
9cbb1750 N |
1165 | struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data, |
1166 | int size) | |
1167 | { | |
1168 | struct blk_plug *plug = current->plug; | |
1169 | struct blk_plug_cb *cb; | |
1170 | ||
1171 | if (!plug) | |
1172 | return NULL; | |
1173 | ||
1174 | list_for_each_entry(cb, &plug->cb_list, list) | |
1175 | if (cb->callback == unplug && cb->data == data) | |
1176 | return cb; | |
1177 | ||
1178 | /* Not currently on the callback list */ | |
1179 | BUG_ON(size < sizeof(*cb)); | |
1180 | cb = kzalloc(size, GFP_ATOMIC); | |
1181 | if (cb) { | |
1182 | cb->data = data; | |
1183 | cb->callback = unplug; | |
1184 | list_add(&cb->list, &plug->cb_list); | |
1185 | } | |
1186 | return cb; | |
1187 | } | |
1188 | EXPORT_SYMBOL(blk_check_plugged); | |
1189 | ||
aa8dccca | 1190 | void __blk_flush_plug(struct blk_plug *plug, bool from_schedule) |
73c10101 | 1191 | { |
b600455d PB |
1192 | if (!list_empty(&plug->cb_list)) |
1193 | flush_plug_callbacks(plug, from_schedule); | |
70904263 | 1194 | blk_mq_flush_plug_list(plug, from_schedule); |
c5fc7b93 JA |
1195 | /* |
1196 | * Unconditionally flush out cached requests, even if the unplug | |
1197 | * event came from schedule. Since we know hold references to the | |
1198 | * queue for cached requests, we don't want a blocked task holding | |
1199 | * up a queue freeze/quiesce event. | |
1200 | */ | |
1201 | if (unlikely(!rq_list_empty(plug->cached_rq))) | |
47c122e3 | 1202 | blk_mq_free_plug_rqs(plug); |
06b23f92 | 1203 | |
3ec48489 | 1204 | plug->cur_ktime = 0; |
06b23f92 | 1205 | current->flags &= ~PF_BLOCK_TS; |
73c10101 | 1206 | } |
73c10101 | 1207 | |
40405851 JM |
1208 | /** |
1209 | * blk_finish_plug - mark the end of a batch of submitted I/O | |
1210 | * @plug: The &struct blk_plug passed to blk_start_plug() | |
1211 | * | |
1212 | * Description: | |
1213 | * Indicate that a batch of I/O submissions is complete. This function | |
1214 | * must be paired with an initial call to blk_start_plug(). The intent | |
1215 | * is to allow the block layer to optimize I/O submission. See the | |
1216 | * documentation for blk_start_plug() for more information. | |
1217 | */ | |
73c10101 JA |
1218 | void blk_finish_plug(struct blk_plug *plug) |
1219 | { | |
008f75a2 | 1220 | if (plug == current->plug) { |
aa8dccca | 1221 | __blk_flush_plug(plug, false); |
008f75a2 CH |
1222 | current->plug = NULL; |
1223 | } | |
73c10101 | 1224 | } |
88b996cd | 1225 | EXPORT_SYMBOL(blk_finish_plug); |
73c10101 | 1226 | |
71ac860a ML |
1227 | void blk_io_schedule(void) |
1228 | { | |
1229 | /* Prevent hang_check timer from firing at us during very long I/O */ | |
1230 | unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2; | |
1231 | ||
1232 | if (timeout) | |
1233 | io_schedule_timeout(timeout); | |
1234 | else | |
1235 | io_schedule(); | |
1236 | } | |
1237 | EXPORT_SYMBOL_GPL(blk_io_schedule); | |
1238 | ||
1da177e4 LT |
1239 | int __init blk_dev_init(void) |
1240 | { | |
16458cf3 | 1241 | BUILD_BUG_ON((__force u32)REQ_OP_LAST >= (1 << REQ_OP_BITS)); |
ef295ecf | 1242 | BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * |
c593642c | 1243 | sizeof_field(struct request, cmd_flags)); |
ef295ecf | 1244 | BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * |
c593642c | 1245 | sizeof_field(struct bio, bi_opf)); |
9eb55b03 | 1246 | |
89b90be2 TH |
1247 | /* used for unplugging and affects IO latency/throughput - HIGHPRI */ |
1248 | kblockd_workqueue = alloc_workqueue("kblockd", | |
28747fcd | 1249 | WQ_MEM_RECLAIM | WQ_HIGHPRI, 0); |
1da177e4 LT |
1250 | if (!kblockd_workqueue) |
1251 | panic("Failed to create kblockd\n"); | |
1252 | ||
48ff13a6 | 1253 | blk_requestq_cachep = KMEM_CACHE(request_queue, SLAB_PANIC); |
1da177e4 | 1254 | |
18fbda91 | 1255 | blk_debugfs_root = debugfs_create_dir("block", NULL); |
18fbda91 | 1256 | |
d38ecf93 | 1257 | return 0; |
1da177e4 | 1258 | } |