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> | |
320ae51f | 19 | #include <linux/blk-mq.h> |
52abca64 | 20 | #include <linux/blk-pm.h> |
fe45e630 | 21 | #include <linux/blk-integrity.h> |
1da177e4 LT |
22 | #include <linux/highmem.h> |
23 | #include <linux/mm.h> | |
cee9a0c4 | 24 | #include <linux/pagemap.h> |
1da177e4 LT |
25 | #include <linux/kernel_stat.h> |
26 | #include <linux/string.h> | |
27 | #include <linux/init.h> | |
1da177e4 LT |
28 | #include <linux/completion.h> |
29 | #include <linux/slab.h> | |
30 | #include <linux/swap.h> | |
31 | #include <linux/writeback.h> | |
faccbd4b | 32 | #include <linux/task_io_accounting_ops.h> |
c17bb495 | 33 | #include <linux/fault-inject.h> |
73c10101 | 34 | #include <linux/list_sort.h> |
e3c78ca5 | 35 | #include <linux/delay.h> |
aaf7c680 | 36 | #include <linux/ratelimit.h> |
6c954667 | 37 | #include <linux/pm_runtime.h> |
eea8f41c | 38 | #include <linux/blk-cgroup.h> |
54d4e6ab | 39 | #include <linux/t10-pi.h> |
18fbda91 | 40 | #include <linux/debugfs.h> |
30abb3a6 | 41 | #include <linux/bpf.h> |
b8e24a93 | 42 | #include <linux/psi.h> |
71ac860a | 43 | #include <linux/sched/sysctl.h> |
a892c8d5 | 44 | #include <linux/blk-crypto.h> |
55782138 LZ |
45 | |
46 | #define CREATE_TRACE_POINTS | |
47 | #include <trace/events/block.h> | |
1da177e4 | 48 | |
8324aa91 | 49 | #include "blk.h" |
43a5e4e2 | 50 | #include "blk-mq.h" |
bd166ef1 | 51 | #include "blk-mq-sched.h" |
bca6b067 | 52 | #include "blk-pm.h" |
a7b36ee6 | 53 | #include "blk-throttle.h" |
8324aa91 | 54 | |
18fbda91 | 55 | struct dentry *blk_debugfs_root; |
18fbda91 | 56 | |
d07335e5 | 57 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap); |
b0da3f0d | 58 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap); |
0a82a8d1 | 59 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete); |
3291fa57 | 60 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_split); |
cbae8d45 | 61 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug); |
b357e4a6 | 62 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_insert); |
0bfc2455 | 63 | |
a73f730d TH |
64 | DEFINE_IDA(blk_queue_ida); |
65 | ||
1da177e4 LT |
66 | /* |
67 | * For queue allocation | |
68 | */ | |
6728cb0e | 69 | struct kmem_cache *blk_requestq_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 | ||
2a4aa30c | 112 | void blk_rq_init(struct request_queue *q, struct request *rq) |
1da177e4 | 113 | { |
1afb20f3 FT |
114 | memset(rq, 0, sizeof(*rq)); |
115 | ||
1da177e4 | 116 | INIT_LIST_HEAD(&rq->queuelist); |
63a71386 | 117 | rq->q = q; |
a2dec7b3 | 118 | rq->__sector = (sector_t) -1; |
2e662b65 JA |
119 | INIT_HLIST_NODE(&rq->hash); |
120 | RB_CLEAR_NODE(&rq->rb_node); | |
e44a6a23 XT |
121 | rq->tag = BLK_MQ_NO_TAG; |
122 | rq->internal_tag = BLK_MQ_NO_TAG; | |
522a7775 | 123 | rq->start_time_ns = ktime_get_ns(); |
09e099d4 | 124 | rq->part = NULL; |
a892c8d5 | 125 | blk_crypto_rq_set_defaults(rq); |
1da177e4 | 126 | } |
2a4aa30c | 127 | EXPORT_SYMBOL(blk_rq_init); |
1da177e4 | 128 | |
e47bc4ed CK |
129 | #define REQ_OP_NAME(name) [REQ_OP_##name] = #name |
130 | static const char *const blk_op_name[] = { | |
131 | REQ_OP_NAME(READ), | |
132 | REQ_OP_NAME(WRITE), | |
133 | REQ_OP_NAME(FLUSH), | |
134 | REQ_OP_NAME(DISCARD), | |
135 | REQ_OP_NAME(SECURE_ERASE), | |
136 | REQ_OP_NAME(ZONE_RESET), | |
6e33dbf2 | 137 | REQ_OP_NAME(ZONE_RESET_ALL), |
6c1b1da5 AJ |
138 | REQ_OP_NAME(ZONE_OPEN), |
139 | REQ_OP_NAME(ZONE_CLOSE), | |
140 | REQ_OP_NAME(ZONE_FINISH), | |
0512a75b | 141 | REQ_OP_NAME(ZONE_APPEND), |
e47bc4ed CK |
142 | REQ_OP_NAME(WRITE_SAME), |
143 | REQ_OP_NAME(WRITE_ZEROES), | |
e47bc4ed CK |
144 | REQ_OP_NAME(DRV_IN), |
145 | REQ_OP_NAME(DRV_OUT), | |
146 | }; | |
147 | #undef REQ_OP_NAME | |
148 | ||
149 | /** | |
150 | * blk_op_str - Return string XXX in the REQ_OP_XXX. | |
151 | * @op: REQ_OP_XXX. | |
152 | * | |
153 | * Description: Centralize block layer function to convert REQ_OP_XXX into | |
154 | * string format. Useful in the debugging and tracing bio or request. For | |
155 | * invalid REQ_OP_XXX it returns string "UNKNOWN". | |
156 | */ | |
157 | inline const char *blk_op_str(unsigned int op) | |
158 | { | |
159 | const char *op_str = "UNKNOWN"; | |
160 | ||
161 | if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op]) | |
162 | op_str = blk_op_name[op]; | |
163 | ||
164 | return op_str; | |
165 | } | |
166 | EXPORT_SYMBOL_GPL(blk_op_str); | |
167 | ||
2a842aca CH |
168 | static const struct { |
169 | int errno; | |
170 | const char *name; | |
171 | } blk_errors[] = { | |
172 | [BLK_STS_OK] = { 0, "" }, | |
173 | [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" }, | |
174 | [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" }, | |
175 | [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" }, | |
176 | [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" }, | |
177 | [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" }, | |
178 | [BLK_STS_NEXUS] = { -EBADE, "critical nexus" }, | |
179 | [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" }, | |
180 | [BLK_STS_PROTECTION] = { -EILSEQ, "protection" }, | |
181 | [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" }, | |
86ff7c2a | 182 | [BLK_STS_DEV_RESOURCE] = { -EBUSY, "device resource" }, |
03a07c92 | 183 | [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" }, |
2a842aca | 184 | |
4e4cbee9 CH |
185 | /* device mapper special case, should not leak out: */ |
186 | [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" }, | |
187 | ||
3b481d91 KB |
188 | /* zone device specific errors */ |
189 | [BLK_STS_ZONE_OPEN_RESOURCE] = { -ETOOMANYREFS, "open zones exceeded" }, | |
190 | [BLK_STS_ZONE_ACTIVE_RESOURCE] = { -EOVERFLOW, "active zones exceeded" }, | |
191 | ||
2a842aca CH |
192 | /* everything else not covered above: */ |
193 | [BLK_STS_IOERR] = { -EIO, "I/O" }, | |
194 | }; | |
195 | ||
196 | blk_status_t errno_to_blk_status(int errno) | |
197 | { | |
198 | int i; | |
199 | ||
200 | for (i = 0; i < ARRAY_SIZE(blk_errors); i++) { | |
201 | if (blk_errors[i].errno == errno) | |
202 | return (__force blk_status_t)i; | |
203 | } | |
204 | ||
205 | return BLK_STS_IOERR; | |
206 | } | |
207 | EXPORT_SYMBOL_GPL(errno_to_blk_status); | |
208 | ||
209 | int blk_status_to_errno(blk_status_t status) | |
210 | { | |
211 | int idx = (__force int)status; | |
212 | ||
34bd9c1c | 213 | if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors))) |
2a842aca CH |
214 | return -EIO; |
215 | return blk_errors[idx].errno; | |
216 | } | |
217 | EXPORT_SYMBOL_GPL(blk_status_to_errno); | |
218 | ||
9be3e06f | 219 | void blk_print_req_error(struct request *req, blk_status_t status) |
2a842aca CH |
220 | { |
221 | int idx = (__force int)status; | |
222 | ||
34bd9c1c | 223 | if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors))) |
2a842aca CH |
224 | return; |
225 | ||
178cc590 | 226 | printk_ratelimited(KERN_ERR |
c477b797 | 227 | "%s error, dev %s, sector %llu op 0x%x:(%s) flags 0x%x " |
b0e5168a | 228 | "phys_seg %u prio class %u\n", |
c477b797 | 229 | blk_errors[idx].name, |
b0e5168a CK |
230 | req->rq_disk ? req->rq_disk->disk_name : "?", |
231 | blk_rq_pos(req), req_op(req), blk_op_str(req_op(req)), | |
232 | req->cmd_flags & ~REQ_OP_MASK, | |
233 | req->nr_phys_segments, | |
234 | IOPRIO_PRIO_CLASS(req->ioprio)); | |
2a842aca CH |
235 | } |
236 | ||
1da177e4 LT |
237 | void blk_dump_rq_flags(struct request *rq, char *msg) |
238 | { | |
aebf526b CH |
239 | printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg, |
240 | rq->rq_disk ? rq->rq_disk->disk_name : "?", | |
5953316d | 241 | (unsigned long long) rq->cmd_flags); |
1da177e4 | 242 | |
83096ebf TH |
243 | printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n", |
244 | (unsigned long long)blk_rq_pos(rq), | |
245 | blk_rq_sectors(rq), blk_rq_cur_sectors(rq)); | |
b4f42e28 JA |
246 | printk(KERN_INFO " bio %p, biotail %p, len %u\n", |
247 | rq->bio, rq->biotail, blk_rq_bytes(rq)); | |
1da177e4 | 248 | } |
1da177e4 LT |
249 | EXPORT_SYMBOL(blk_dump_rq_flags); |
250 | ||
1da177e4 LT |
251 | /** |
252 | * blk_sync_queue - cancel any pending callbacks on a queue | |
253 | * @q: the queue | |
254 | * | |
255 | * Description: | |
256 | * The block layer may perform asynchronous callback activity | |
257 | * on a queue, such as calling the unplug function after a timeout. | |
258 | * A block device may call blk_sync_queue to ensure that any | |
259 | * such activity is cancelled, thus allowing it to release resources | |
59c51591 | 260 | * that the callbacks might use. The caller must already have made sure |
c62b37d9 | 261 | * that its ->submit_bio will not re-add plugging prior to calling |
1da177e4 LT |
262 | * this function. |
263 | * | |
da527770 | 264 | * This function does not cancel any asynchronous activity arising |
da3dae54 | 265 | * out of elevator or throttling code. That would require elevator_exit() |
5efd6113 | 266 | * and blkcg_exit_queue() to be called with queue lock initialized. |
da527770 | 267 | * |
1da177e4 LT |
268 | */ |
269 | void blk_sync_queue(struct request_queue *q) | |
270 | { | |
70ed28b9 | 271 | del_timer_sync(&q->timeout); |
4e9b6f20 | 272 | cancel_work_sync(&q->timeout_work); |
1da177e4 LT |
273 | } |
274 | EXPORT_SYMBOL(blk_sync_queue); | |
275 | ||
c9254f2d | 276 | /** |
cd84a62e | 277 | * blk_set_pm_only - increment pm_only counter |
c9254f2d | 278 | * @q: request queue pointer |
c9254f2d | 279 | */ |
cd84a62e | 280 | void blk_set_pm_only(struct request_queue *q) |
c9254f2d | 281 | { |
cd84a62e | 282 | atomic_inc(&q->pm_only); |
c9254f2d | 283 | } |
cd84a62e | 284 | EXPORT_SYMBOL_GPL(blk_set_pm_only); |
c9254f2d | 285 | |
cd84a62e | 286 | void blk_clear_pm_only(struct request_queue *q) |
c9254f2d | 287 | { |
cd84a62e BVA |
288 | int pm_only; |
289 | ||
290 | pm_only = atomic_dec_return(&q->pm_only); | |
291 | WARN_ON_ONCE(pm_only < 0); | |
292 | if (pm_only == 0) | |
293 | wake_up_all(&q->mq_freeze_wq); | |
c9254f2d | 294 | } |
cd84a62e | 295 | EXPORT_SYMBOL_GPL(blk_clear_pm_only); |
c9254f2d | 296 | |
b5bd357c LC |
297 | /** |
298 | * blk_put_queue - decrement the request_queue refcount | |
299 | * @q: the request_queue structure to decrement the refcount for | |
300 | * | |
301 | * Decrements the refcount of the request_queue kobject. When this reaches 0 | |
302 | * we'll have blk_release_queue() called. | |
e8c7d14a LC |
303 | * |
304 | * Context: Any context, but the last reference must not be dropped from | |
305 | * atomic context. | |
b5bd357c | 306 | */ |
165125e1 | 307 | void blk_put_queue(struct request_queue *q) |
483f4afc AV |
308 | { |
309 | kobject_put(&q->kobj); | |
310 | } | |
d86e0e83 | 311 | EXPORT_SYMBOL(blk_put_queue); |
483f4afc | 312 | |
8e141f9e | 313 | void blk_queue_start_drain(struct request_queue *q) |
aed3ea94 | 314 | { |
d3cfb2a0 ML |
315 | /* |
316 | * When queue DYING flag is set, we need to block new req | |
317 | * entering queue, so we call blk_freeze_queue_start() to | |
318 | * prevent I/O from crossing blk_queue_enter(). | |
319 | */ | |
320 | blk_freeze_queue_start(q); | |
344e9ffc | 321 | if (queue_is_mq(q)) |
aed3ea94 | 322 | blk_mq_wake_waiters(q); |
055f6e18 ML |
323 | /* Make blk_queue_enter() reexamine the DYING flag. */ |
324 | wake_up_all(&q->mq_freeze_wq); | |
aed3ea94 | 325 | } |
8e141f9e CH |
326 | |
327 | void blk_set_queue_dying(struct request_queue *q) | |
328 | { | |
329 | blk_queue_flag_set(QUEUE_FLAG_DYING, q); | |
330 | blk_queue_start_drain(q); | |
331 | } | |
aed3ea94 JA |
332 | EXPORT_SYMBOL_GPL(blk_set_queue_dying); |
333 | ||
c9a929dd TH |
334 | /** |
335 | * blk_cleanup_queue - shutdown a request queue | |
336 | * @q: request queue to shutdown | |
337 | * | |
c246e80d BVA |
338 | * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and |
339 | * put it. All future requests will be failed immediately with -ENODEV. | |
e8c7d14a LC |
340 | * |
341 | * Context: can sleep | |
c94a96ac | 342 | */ |
6728cb0e | 343 | void blk_cleanup_queue(struct request_queue *q) |
483f4afc | 344 | { |
e8c7d14a LC |
345 | /* cannot be called from atomic context */ |
346 | might_sleep(); | |
347 | ||
bae85c15 BVA |
348 | WARN_ON_ONCE(blk_queue_registered(q)); |
349 | ||
3f3299d5 | 350 | /* mark @q DYING, no new request or merges will be allowed afterwards */ |
aed3ea94 | 351 | blk_set_queue_dying(q); |
6ecf23af | 352 | |
57d74df9 CH |
353 | blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q); |
354 | blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q); | |
c9a929dd | 355 | |
c246e80d BVA |
356 | /* |
357 | * Drain all requests queued before DYING marking. Set DEAD flag to | |
67ed8b73 BVA |
358 | * prevent that blk_mq_run_hw_queues() accesses the hardware queues |
359 | * after draining finished. | |
c246e80d | 360 | */ |
3ef28e83 | 361 | blk_freeze_queue(q); |
c57cdf7a | 362 | |
57d74df9 | 363 | blk_queue_flag_set(QUEUE_FLAG_DEAD, q); |
c9a929dd | 364 | |
c9a929dd | 365 | blk_sync_queue(q); |
344e9ffc | 366 | if (queue_is_mq(q)) |
c7e2d94b | 367 | blk_mq_exit_queue(q); |
a1ce35fa | 368 | |
c3e22192 ML |
369 | /* |
370 | * In theory, request pool of sched_tags belongs to request queue. | |
371 | * However, the current implementation requires tag_set for freeing | |
372 | * requests, so free the pool now. | |
373 | * | |
374 | * Queue has become frozen, there can't be any in-queue requests, so | |
375 | * it is safe to free requests now. | |
376 | */ | |
377 | mutex_lock(&q->sysfs_lock); | |
378 | if (q->elevator) | |
1820f4f0 | 379 | blk_mq_sched_free_rqs(q); |
c3e22192 ML |
380 | mutex_unlock(&q->sysfs_lock); |
381 | ||
3ef28e83 | 382 | percpu_ref_exit(&q->q_usage_counter); |
45a9c9d9 | 383 | |
c9a929dd | 384 | /* @q is and will stay empty, shutdown and put */ |
483f4afc AV |
385 | blk_put_queue(q); |
386 | } | |
1da177e4 LT |
387 | EXPORT_SYMBOL(blk_cleanup_queue); |
388 | ||
1f14a098 CH |
389 | static bool blk_try_enter_queue(struct request_queue *q, bool pm) |
390 | { | |
391 | rcu_read_lock(); | |
392 | if (!percpu_ref_tryget_live(&q->q_usage_counter)) | |
393 | goto fail; | |
394 | ||
395 | /* | |
396 | * The code that increments the pm_only counter must ensure that the | |
397 | * counter is globally visible before the queue is unfrozen. | |
398 | */ | |
399 | if (blk_queue_pm_only(q) && | |
400 | (!pm || queue_rpm_status(q) == RPM_SUSPENDED)) | |
401 | goto fail_put; | |
402 | ||
403 | rcu_read_unlock(); | |
404 | return true; | |
405 | ||
406 | fail_put: | |
407 | percpu_ref_put(&q->q_usage_counter); | |
408 | fail: | |
409 | rcu_read_unlock(); | |
410 | return false; | |
411 | } | |
412 | ||
3a0a5299 BVA |
413 | /** |
414 | * blk_queue_enter() - try to increase q->q_usage_counter | |
415 | * @q: request queue pointer | |
a4d34da7 | 416 | * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PM |
3a0a5299 | 417 | */ |
9a95e4ef | 418 | int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags) |
3ef28e83 | 419 | { |
a4d34da7 | 420 | const bool pm = flags & BLK_MQ_REQ_PM; |
3a0a5299 | 421 | |
1f14a098 | 422 | while (!blk_try_enter_queue(q, pm)) { |
3a0a5299 | 423 | if (flags & BLK_MQ_REQ_NOWAIT) |
3ef28e83 DW |
424 | return -EBUSY; |
425 | ||
5ed61d3f | 426 | /* |
1f14a098 CH |
427 | * read pair of barrier in blk_freeze_queue_start(), we need to |
428 | * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and | |
429 | * reading .mq_freeze_depth or queue dying flag, otherwise the | |
430 | * following wait may never return if the two reads are | |
431 | * reordered. | |
5ed61d3f ML |
432 | */ |
433 | smp_rmb(); | |
1dc3039b | 434 | wait_event(q->mq_freeze_wq, |
7996a8b5 | 435 | (!q->mq_freeze_depth && |
52abca64 | 436 | blk_pm_resume_queue(pm, q)) || |
1dc3039b | 437 | blk_queue_dying(q)); |
3ef28e83 DW |
438 | if (blk_queue_dying(q)) |
439 | return -ENODEV; | |
3ef28e83 | 440 | } |
1f14a098 CH |
441 | |
442 | return 0; | |
3ef28e83 DW |
443 | } |
444 | ||
accea322 CH |
445 | static inline int bio_queue_enter(struct bio *bio) |
446 | { | |
eab4e027 | 447 | struct request_queue *q = bdev_get_queue(bio->bi_bdev); |
accea322 | 448 | |
a6741536 | 449 | while (!blk_try_enter_queue(q, false)) { |
eab4e027 PB |
450 | struct gendisk *disk = bio->bi_bdev->bd_disk; |
451 | ||
a6741536 | 452 | if (bio->bi_opf & REQ_NOWAIT) { |
8e141f9e | 453 | if (test_bit(GD_DEAD, &disk->state)) |
a6741536 | 454 | goto dead; |
accea322 | 455 | bio_wouldblock_error(bio); |
a6741536 CH |
456 | return -EBUSY; |
457 | } | |
458 | ||
459 | /* | |
460 | * read pair of barrier in blk_freeze_queue_start(), we need to | |
461 | * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and | |
462 | * reading .mq_freeze_depth or queue dying flag, otherwise the | |
463 | * following wait may never return if the two reads are | |
464 | * reordered. | |
465 | */ | |
466 | smp_rmb(); | |
467 | wait_event(q->mq_freeze_wq, | |
468 | (!q->mq_freeze_depth && | |
469 | blk_pm_resume_queue(false, q)) || | |
8e141f9e CH |
470 | test_bit(GD_DEAD, &disk->state)); |
471 | if (test_bit(GD_DEAD, &disk->state)) | |
a6741536 | 472 | goto dead; |
accea322 CH |
473 | } |
474 | ||
a6741536 CH |
475 | return 0; |
476 | dead: | |
477 | bio_io_error(bio); | |
478 | return -ENODEV; | |
accea322 CH |
479 | } |
480 | ||
3ef28e83 DW |
481 | void blk_queue_exit(struct request_queue *q) |
482 | { | |
483 | percpu_ref_put(&q->q_usage_counter); | |
484 | } | |
485 | ||
486 | static void blk_queue_usage_counter_release(struct percpu_ref *ref) | |
487 | { | |
488 | struct request_queue *q = | |
489 | container_of(ref, struct request_queue, q_usage_counter); | |
490 | ||
491 | wake_up_all(&q->mq_freeze_wq); | |
492 | } | |
493 | ||
bca237a5 | 494 | static void blk_rq_timed_out_timer(struct timer_list *t) |
287922eb | 495 | { |
bca237a5 | 496 | struct request_queue *q = from_timer(q, t, timeout); |
287922eb CH |
497 | |
498 | kblockd_schedule_work(&q->timeout_work); | |
499 | } | |
500 | ||
2e3c18d0 TH |
501 | static void blk_timeout_work(struct work_struct *work) |
502 | { | |
503 | } | |
504 | ||
c62b37d9 | 505 | struct request_queue *blk_alloc_queue(int node_id) |
1946089a | 506 | { |
165125e1 | 507 | struct request_queue *q; |
338aa96d | 508 | int ret; |
1946089a | 509 | |
8324aa91 | 510 | q = kmem_cache_alloc_node(blk_requestq_cachep, |
3d745ea5 | 511 | GFP_KERNEL | __GFP_ZERO, node_id); |
1da177e4 LT |
512 | if (!q) |
513 | return NULL; | |
514 | ||
cbf62af3 | 515 | q->last_merge = NULL; |
cbf62af3 | 516 | |
3d745ea5 | 517 | q->id = ida_simple_get(&blk_queue_ida, 0, 0, GFP_KERNEL); |
a73f730d | 518 | if (q->id < 0) |
3d2936f4 | 519 | goto fail_q; |
a73f730d | 520 | |
c495a176 | 521 | ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, 0); |
338aa96d | 522 | if (ret) |
54efd50b KO |
523 | goto fail_id; |
524 | ||
a83b576c JA |
525 | q->stats = blk_alloc_queue_stats(); |
526 | if (!q->stats) | |
edb0872f | 527 | goto fail_split; |
a83b576c | 528 | |
5151412d | 529 | q->node = node_id; |
0989a025 | 530 | |
079a2e3e | 531 | atomic_set(&q->nr_active_requests_shared_tags, 0); |
bccf5e26 | 532 | |
bca237a5 | 533 | timer_setup(&q->timeout, blk_rq_timed_out_timer, 0); |
2e3c18d0 | 534 | INIT_WORK(&q->timeout_work, blk_timeout_work); |
a612fddf | 535 | INIT_LIST_HEAD(&q->icq_list); |
4eef3049 | 536 | #ifdef CONFIG_BLK_CGROUP |
e8989fae | 537 | INIT_LIST_HEAD(&q->blkg_list); |
4eef3049 | 538 | #endif |
483f4afc | 539 | |
8324aa91 | 540 | kobject_init(&q->kobj, &blk_queue_ktype); |
1da177e4 | 541 | |
85e0cbbb | 542 | mutex_init(&q->debugfs_mutex); |
483f4afc | 543 | mutex_init(&q->sysfs_lock); |
cecf5d87 | 544 | mutex_init(&q->sysfs_dir_lock); |
0d945c1f | 545 | spin_lock_init(&q->queue_lock); |
c94a96ac | 546 | |
320ae51f | 547 | init_waitqueue_head(&q->mq_freeze_wq); |
7996a8b5 | 548 | mutex_init(&q->mq_freeze_lock); |
320ae51f | 549 | |
3ef28e83 DW |
550 | /* |
551 | * Init percpu_ref in atomic mode so that it's faster to shutdown. | |
552 | * See blk_register_queue() for details. | |
553 | */ | |
554 | if (percpu_ref_init(&q->q_usage_counter, | |
555 | blk_queue_usage_counter_release, | |
556 | PERCPU_REF_INIT_ATOMIC, GFP_KERNEL)) | |
edb0872f | 557 | goto fail_stats; |
f51b802c | 558 | |
3ef28e83 DW |
559 | if (blkcg_init_queue(q)) |
560 | goto fail_ref; | |
561 | ||
3d745ea5 CH |
562 | blk_queue_dma_alignment(q, 511); |
563 | blk_set_default_limits(&q->limits); | |
d2a27964 | 564 | q->nr_requests = BLKDEV_DEFAULT_RQ; |
3d745ea5 | 565 | |
1da177e4 | 566 | return q; |
a73f730d | 567 | |
3ef28e83 DW |
568 | fail_ref: |
569 | percpu_ref_exit(&q->q_usage_counter); | |
a83b576c | 570 | fail_stats: |
edb0872f | 571 | blk_free_queue_stats(q->stats); |
54efd50b | 572 | fail_split: |
338aa96d | 573 | bioset_exit(&q->bio_split); |
a73f730d TH |
574 | fail_id: |
575 | ida_simple_remove(&blk_queue_ida, q->id); | |
576 | fail_q: | |
577 | kmem_cache_free(blk_requestq_cachep, q); | |
578 | return NULL; | |
1da177e4 | 579 | } |
1da177e4 | 580 | |
b5bd357c LC |
581 | /** |
582 | * blk_get_queue - increment the request_queue refcount | |
583 | * @q: the request_queue structure to increment the refcount for | |
584 | * | |
585 | * Increment the refcount of the request_queue kobject. | |
763b5892 LC |
586 | * |
587 | * Context: Any context. | |
b5bd357c | 588 | */ |
09ac46c4 | 589 | bool blk_get_queue(struct request_queue *q) |
1da177e4 | 590 | { |
3f3299d5 | 591 | if (likely(!blk_queue_dying(q))) { |
09ac46c4 TH |
592 | __blk_get_queue(q); |
593 | return true; | |
1da177e4 LT |
594 | } |
595 | ||
09ac46c4 | 596 | return false; |
1da177e4 | 597 | } |
d86e0e83 | 598 | EXPORT_SYMBOL(blk_get_queue); |
1da177e4 | 599 | |
a1ce35fa JA |
600 | /** |
601 | * blk_get_request - allocate a request | |
602 | * @q: request queue to allocate a request for | |
603 | * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC. | |
604 | * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT. | |
1da177e4 | 605 | */ |
a1ce35fa JA |
606 | struct request *blk_get_request(struct request_queue *q, unsigned int op, |
607 | blk_mq_req_flags_t flags) | |
1da177e4 | 608 | { |
a1ce35fa | 609 | struct request *req; |
1da177e4 | 610 | |
a1ce35fa | 611 | WARN_ON_ONCE(op & REQ_NOWAIT); |
a4d34da7 | 612 | WARN_ON_ONCE(flags & ~(BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_PM)); |
1da177e4 | 613 | |
a1ce35fa JA |
614 | req = blk_mq_alloc_request(q, op, flags); |
615 | if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn) | |
616 | q->mq_ops->initialize_rq_fn(req); | |
1da177e4 | 617 | |
a1ce35fa | 618 | return req; |
1da177e4 | 619 | } |
a1ce35fa | 620 | EXPORT_SYMBOL(blk_get_request); |
1da177e4 | 621 | |
1da177e4 LT |
622 | void blk_put_request(struct request *req) |
623 | { | |
a1ce35fa | 624 | blk_mq_free_request(req); |
1da177e4 | 625 | } |
1da177e4 LT |
626 | EXPORT_SYMBOL(blk_put_request); |
627 | ||
52c5e62d | 628 | static void handle_bad_sector(struct bio *bio, sector_t maxsector) |
1da177e4 LT |
629 | { |
630 | char b[BDEVNAME_SIZE]; | |
631 | ||
8a3ee677 | 632 | pr_info_ratelimited("%s: attempt to access beyond end of device\n" |
f4ac712e | 633 | "%s: rw=%d, want=%llu, limit=%llu\n", |
8a3ee677 | 634 | current->comm, |
f4ac712e TH |
635 | bio_devname(bio, b), bio->bi_opf, |
636 | bio_end_sector(bio), maxsector); | |
1da177e4 LT |
637 | } |
638 | ||
c17bb495 AM |
639 | #ifdef CONFIG_FAIL_MAKE_REQUEST |
640 | ||
641 | static DECLARE_FAULT_ATTR(fail_make_request); | |
642 | ||
643 | static int __init setup_fail_make_request(char *str) | |
644 | { | |
645 | return setup_fault_attr(&fail_make_request, str); | |
646 | } | |
647 | __setup("fail_make_request=", setup_fail_make_request); | |
648 | ||
8446fe92 | 649 | static bool should_fail_request(struct block_device *part, unsigned int bytes) |
c17bb495 | 650 | { |
8446fe92 | 651 | return part->bd_make_it_fail && should_fail(&fail_make_request, bytes); |
c17bb495 AM |
652 | } |
653 | ||
654 | static int __init fail_make_request_debugfs(void) | |
655 | { | |
dd48c085 AM |
656 | struct dentry *dir = fault_create_debugfs_attr("fail_make_request", |
657 | NULL, &fail_make_request); | |
658 | ||
21f9fcd8 | 659 | return PTR_ERR_OR_ZERO(dir); |
c17bb495 AM |
660 | } |
661 | ||
662 | late_initcall(fail_make_request_debugfs); | |
663 | ||
664 | #else /* CONFIG_FAIL_MAKE_REQUEST */ | |
665 | ||
8446fe92 | 666 | static inline bool should_fail_request(struct block_device *part, |
b2c9cd37 | 667 | unsigned int bytes) |
c17bb495 | 668 | { |
b2c9cd37 | 669 | return false; |
c17bb495 AM |
670 | } |
671 | ||
672 | #endif /* CONFIG_FAIL_MAKE_REQUEST */ | |
673 | ||
2f9f6221 | 674 | static inline bool bio_check_ro(struct bio *bio) |
721c7fc7 | 675 | { |
2f9f6221 | 676 | if (op_is_write(bio_op(bio)) && bdev_read_only(bio->bi_bdev)) { |
721c7fc7 ID |
677 | char b[BDEVNAME_SIZE]; |
678 | ||
8b2ded1c MP |
679 | if (op_is_flush(bio->bi_opf) && !bio_sectors(bio)) |
680 | return false; | |
681 | ||
a32e236e | 682 | WARN_ONCE(1, |
c8178674 | 683 | "Trying to write to read-only block-device %s (partno %d)\n", |
2f9f6221 | 684 | bio_devname(bio, b), bio->bi_bdev->bd_partno); |
a32e236e LT |
685 | /* Older lvm-tools actually trigger this */ |
686 | return false; | |
721c7fc7 ID |
687 | } |
688 | ||
689 | return false; | |
690 | } | |
691 | ||
30abb3a6 HM |
692 | static noinline int should_fail_bio(struct bio *bio) |
693 | { | |
309dca30 | 694 | if (should_fail_request(bdev_whole(bio->bi_bdev), bio->bi_iter.bi_size)) |
30abb3a6 HM |
695 | return -EIO; |
696 | return 0; | |
697 | } | |
698 | ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO); | |
699 | ||
52c5e62d CH |
700 | /* |
701 | * Check whether this bio extends beyond the end of the device or partition. | |
702 | * This may well happen - the kernel calls bread() without checking the size of | |
703 | * the device, e.g., when mounting a file system. | |
704 | */ | |
2f9f6221 | 705 | static inline int bio_check_eod(struct bio *bio) |
52c5e62d | 706 | { |
2f9f6221 | 707 | sector_t maxsector = bdev_nr_sectors(bio->bi_bdev); |
52c5e62d CH |
708 | unsigned int nr_sectors = bio_sectors(bio); |
709 | ||
710 | if (nr_sectors && maxsector && | |
711 | (nr_sectors > maxsector || | |
712 | bio->bi_iter.bi_sector > maxsector - nr_sectors)) { | |
713 | handle_bad_sector(bio, maxsector); | |
714 | return -EIO; | |
715 | } | |
716 | return 0; | |
717 | } | |
718 | ||
74d46992 CH |
719 | /* |
720 | * Remap block n of partition p to block n+start(p) of the disk. | |
721 | */ | |
2f9f6221 | 722 | static int blk_partition_remap(struct bio *bio) |
74d46992 | 723 | { |
309dca30 | 724 | struct block_device *p = bio->bi_bdev; |
74d46992 | 725 | |
52c5e62d | 726 | if (unlikely(should_fail_request(p, bio->bi_iter.bi_size))) |
2f9f6221 | 727 | return -EIO; |
5eac3eb3 | 728 | if (bio_sectors(bio)) { |
8446fe92 | 729 | bio->bi_iter.bi_sector += p->bd_start_sect; |
1c02fca6 | 730 | trace_block_bio_remap(bio, p->bd_dev, |
29ff57c6 | 731 | bio->bi_iter.bi_sector - |
8446fe92 | 732 | p->bd_start_sect); |
52c5e62d | 733 | } |
30c5d345 | 734 | bio_set_flag(bio, BIO_REMAPPED); |
2f9f6221 | 735 | return 0; |
74d46992 CH |
736 | } |
737 | ||
0512a75b KB |
738 | /* |
739 | * Check write append to a zoned block device. | |
740 | */ | |
741 | static inline blk_status_t blk_check_zone_append(struct request_queue *q, | |
742 | struct bio *bio) | |
743 | { | |
744 | sector_t pos = bio->bi_iter.bi_sector; | |
745 | int nr_sectors = bio_sectors(bio); | |
746 | ||
747 | /* Only applicable to zoned block devices */ | |
748 | if (!blk_queue_is_zoned(q)) | |
749 | return BLK_STS_NOTSUPP; | |
750 | ||
751 | /* The bio sector must point to the start of a sequential zone */ | |
752 | if (pos & (blk_queue_zone_sectors(q) - 1) || | |
753 | !blk_queue_zone_is_seq(q, pos)) | |
754 | return BLK_STS_IOERR; | |
755 | ||
756 | /* | |
757 | * Not allowed to cross zone boundaries. Otherwise, the BIO will be | |
758 | * split and could result in non-contiguous sectors being written in | |
759 | * different zones. | |
760 | */ | |
761 | if (nr_sectors > q->limits.chunk_sectors) | |
762 | return BLK_STS_IOERR; | |
763 | ||
764 | /* Make sure the BIO is small enough and will not get split */ | |
765 | if (nr_sectors > q->limits.max_zone_append_sectors) | |
766 | return BLK_STS_IOERR; | |
767 | ||
768 | bio->bi_opf |= REQ_NOMERGE; | |
769 | ||
770 | return BLK_STS_OK; | |
771 | } | |
772 | ||
ed00aabd | 773 | static noinline_for_stack bool submit_bio_checks(struct bio *bio) |
1da177e4 | 774 | { |
309dca30 | 775 | struct block_device *bdev = bio->bi_bdev; |
eab4e027 | 776 | struct request_queue *q = bdev_get_queue(bdev); |
4e4cbee9 | 777 | blk_status_t status = BLK_STS_IOERR; |
5a473e83 | 778 | struct blk_plug *plug; |
1da177e4 LT |
779 | |
780 | might_sleep(); | |
1da177e4 | 781 | |
5a473e83 JA |
782 | plug = blk_mq_plug(q, bio); |
783 | if (plug && plug->nowait) | |
784 | bio->bi_opf |= REQ_NOWAIT; | |
785 | ||
03a07c92 | 786 | /* |
b0beb280 | 787 | * For a REQ_NOWAIT based request, return -EOPNOTSUPP |
021a2446 | 788 | * if queue does not support NOWAIT. |
03a07c92 | 789 | */ |
021a2446 | 790 | if ((bio->bi_opf & REQ_NOWAIT) && !blk_queue_nowait(q)) |
b0beb280 | 791 | goto not_supported; |
03a07c92 | 792 | |
30abb3a6 | 793 | if (should_fail_bio(bio)) |
5a7bbad2 | 794 | goto end_io; |
2f9f6221 CH |
795 | if (unlikely(bio_check_ro(bio))) |
796 | goto end_io; | |
3a905c37 CH |
797 | if (!bio_flagged(bio, BIO_REMAPPED)) { |
798 | if (unlikely(bio_check_eod(bio))) | |
799 | goto end_io; | |
800 | if (bdev->bd_partno && unlikely(blk_partition_remap(bio))) | |
801 | goto end_io; | |
802 | } | |
2056a782 | 803 | |
5a7bbad2 | 804 | /* |
ed00aabd CH |
805 | * Filter flush bio's early so that bio based drivers without flush |
806 | * support don't have to worry about them. | |
5a7bbad2 | 807 | */ |
f3a8ab7d | 808 | if (op_is_flush(bio->bi_opf) && |
c888a8f9 | 809 | !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) { |
1eff9d32 | 810 | bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA); |
e439ab71 | 811 | if (!bio_sectors(bio)) { |
4e4cbee9 | 812 | status = BLK_STS_OK; |
51fd77bd JA |
813 | goto end_io; |
814 | } | |
5a7bbad2 | 815 | } |
5ddfe969 | 816 | |
d04c406f | 817 | if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags)) |
6ce913fe | 818 | bio_clear_polled(bio); |
d04c406f | 819 | |
288dab8a CH |
820 | switch (bio_op(bio)) { |
821 | case REQ_OP_DISCARD: | |
822 | if (!blk_queue_discard(q)) | |
823 | goto not_supported; | |
824 | break; | |
825 | case REQ_OP_SECURE_ERASE: | |
826 | if (!blk_queue_secure_erase(q)) | |
827 | goto not_supported; | |
828 | break; | |
829 | case REQ_OP_WRITE_SAME: | |
74d46992 | 830 | if (!q->limits.max_write_same_sectors) |
288dab8a | 831 | goto not_supported; |
58886785 | 832 | break; |
0512a75b KB |
833 | case REQ_OP_ZONE_APPEND: |
834 | status = blk_check_zone_append(q, bio); | |
835 | if (status != BLK_STS_OK) | |
836 | goto end_io; | |
837 | break; | |
2d253440 | 838 | case REQ_OP_ZONE_RESET: |
6c1b1da5 AJ |
839 | case REQ_OP_ZONE_OPEN: |
840 | case REQ_OP_ZONE_CLOSE: | |
841 | case REQ_OP_ZONE_FINISH: | |
74d46992 | 842 | if (!blk_queue_is_zoned(q)) |
2d253440 | 843 | goto not_supported; |
288dab8a | 844 | break; |
6e33dbf2 CK |
845 | case REQ_OP_ZONE_RESET_ALL: |
846 | if (!blk_queue_is_zoned(q) || !blk_queue_zone_resetall(q)) | |
847 | goto not_supported; | |
848 | break; | |
a6f0788e | 849 | case REQ_OP_WRITE_ZEROES: |
74d46992 | 850 | if (!q->limits.max_write_zeroes_sectors) |
a6f0788e CK |
851 | goto not_supported; |
852 | break; | |
288dab8a CH |
853 | default: |
854 | break; | |
5a7bbad2 | 855 | } |
01edede4 | 856 | |
7f4b35d1 | 857 | /* |
3e82c348 CH |
858 | * Various block parts want %current->io_context, so allocate it up |
859 | * front rather than dealing with lots of pain to allocate it only | |
860 | * where needed. This may fail and the block layer knows how to live | |
861 | * with it. | |
7f4b35d1 | 862 | */ |
3e82c348 CH |
863 | if (unlikely(!current->io_context)) |
864 | create_task_io_context(current, GFP_ATOMIC, q->node); | |
7f4b35d1 | 865 | |
db18a53e CH |
866 | if (blk_throtl_bio(bio)) { |
867 | blkcg_bio_issue_init(bio); | |
ae118896 | 868 | return false; |
db18a53e CH |
869 | } |
870 | ||
871 | blk_cgroup_bio_start(bio); | |
872 | blkcg_bio_issue_init(bio); | |
27a84d54 | 873 | |
fbbaf700 | 874 | if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) { |
e8a676d6 | 875 | trace_block_bio_queue(bio); |
fbbaf700 N |
876 | /* Now that enqueuing has been traced, we need to trace |
877 | * completion as well. | |
878 | */ | |
879 | bio_set_flag(bio, BIO_TRACE_COMPLETION); | |
880 | } | |
27a84d54 | 881 | return true; |
a7384677 | 882 | |
288dab8a | 883 | not_supported: |
4e4cbee9 | 884 | status = BLK_STS_NOTSUPP; |
a7384677 | 885 | end_io: |
4e4cbee9 | 886 | bio->bi_status = status; |
4246a0b6 | 887 | bio_endio(bio); |
27a84d54 | 888 | return false; |
1da177e4 LT |
889 | } |
890 | ||
3e08773c | 891 | static void __submit_bio(struct bio *bio) |
ac7c5675 | 892 | { |
309dca30 | 893 | struct gendisk *disk = bio->bi_bdev->bd_disk; |
ac7c5675 | 894 | |
cc9c884d | 895 | if (unlikely(bio_queue_enter(bio) != 0)) |
3e08773c | 896 | return; |
cc9c884d CH |
897 | |
898 | if (!submit_bio_checks(bio) || !blk_crypto_bio_prep(&bio)) | |
899 | goto queue_exit; | |
3e08773c CH |
900 | if (!disk->fops->submit_bio) { |
901 | blk_mq_submit_bio(bio); | |
902 | return; | |
ac7c5675 | 903 | } |
3e08773c | 904 | disk->fops->submit_bio(bio); |
cc9c884d | 905 | queue_exit: |
c62b37d9 | 906 | blk_queue_exit(disk->queue); |
ac7c5675 CH |
907 | } |
908 | ||
566acf2d CH |
909 | /* |
910 | * The loop in this function may be a bit non-obvious, and so deserves some | |
911 | * explanation: | |
912 | * | |
913 | * - Before entering the loop, bio->bi_next is NULL (as all callers ensure | |
914 | * that), so we have a list with a single bio. | |
915 | * - We pretend that we have just taken it off a longer list, so we assign | |
916 | * bio_list to a pointer to the bio_list_on_stack, thus initialising the | |
917 | * bio_list of new bios to be added. ->submit_bio() may indeed add some more | |
918 | * bios through a recursive call to submit_bio_noacct. If it did, we find a | |
919 | * non-NULL value in bio_list and re-enter the loop from the top. | |
920 | * - In this case we really did just take the bio of the top of the list (no | |
921 | * pretending) and so remove it from bio_list, and call into ->submit_bio() | |
922 | * again. | |
923 | * | |
924 | * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio. | |
925 | * bio_list_on_stack[1] contains bios that were submitted before the current | |
926 | * ->submit_bio_bio, but that haven't been processed yet. | |
927 | */ | |
3e08773c | 928 | static void __submit_bio_noacct(struct bio *bio) |
566acf2d CH |
929 | { |
930 | struct bio_list bio_list_on_stack[2]; | |
566acf2d CH |
931 | |
932 | BUG_ON(bio->bi_next); | |
933 | ||
934 | bio_list_init(&bio_list_on_stack[0]); | |
935 | current->bio_list = bio_list_on_stack; | |
936 | ||
937 | do { | |
eab4e027 | 938 | struct request_queue *q = bdev_get_queue(bio->bi_bdev); |
566acf2d CH |
939 | struct bio_list lower, same; |
940 | ||
566acf2d CH |
941 | /* |
942 | * Create a fresh bio_list for all subordinate requests. | |
943 | */ | |
944 | bio_list_on_stack[1] = bio_list_on_stack[0]; | |
945 | bio_list_init(&bio_list_on_stack[0]); | |
946 | ||
3e08773c | 947 | __submit_bio(bio); |
566acf2d CH |
948 | |
949 | /* | |
950 | * Sort new bios into those for a lower level and those for the | |
951 | * same level. | |
952 | */ | |
953 | bio_list_init(&lower); | |
954 | bio_list_init(&same); | |
955 | while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL) | |
eab4e027 | 956 | if (q == bdev_get_queue(bio->bi_bdev)) |
566acf2d CH |
957 | bio_list_add(&same, bio); |
958 | else | |
959 | bio_list_add(&lower, bio); | |
960 | ||
961 | /* | |
962 | * Now assemble so we handle the lowest level first. | |
963 | */ | |
964 | bio_list_merge(&bio_list_on_stack[0], &lower); | |
965 | bio_list_merge(&bio_list_on_stack[0], &same); | |
966 | bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]); | |
967 | } while ((bio = bio_list_pop(&bio_list_on_stack[0]))); | |
968 | ||
969 | current->bio_list = NULL; | |
566acf2d CH |
970 | } |
971 | ||
3e08773c | 972 | static void __submit_bio_noacct_mq(struct bio *bio) |
ff93ea0c | 973 | { |
7c792f33 | 974 | struct bio_list bio_list[2] = { }; |
ff93ea0c | 975 | |
7c792f33 | 976 | current->bio_list = bio_list; |
ff93ea0c CH |
977 | |
978 | do { | |
3e08773c | 979 | __submit_bio(bio); |
7c792f33 | 980 | } while ((bio = bio_list_pop(&bio_list[0]))); |
ff93ea0c CH |
981 | |
982 | current->bio_list = NULL; | |
ff93ea0c CH |
983 | } |
984 | ||
27a84d54 | 985 | /** |
ed00aabd | 986 | * submit_bio_noacct - re-submit a bio to the block device layer for I/O |
27a84d54 CH |
987 | * @bio: The bio describing the location in memory and on the device. |
988 | * | |
3fdd4086 CH |
989 | * This is a version of submit_bio() that shall only be used for I/O that is |
990 | * resubmitted to lower level drivers by stacking block drivers. All file | |
991 | * systems and other upper level users of the block layer should use | |
992 | * submit_bio() instead. | |
d89d8796 | 993 | */ |
3e08773c | 994 | void submit_bio_noacct(struct bio *bio) |
d89d8796 | 995 | { |
27a84d54 | 996 | /* |
566acf2d CH |
997 | * We only want one ->submit_bio to be active at a time, else stack |
998 | * usage with stacked devices could be a problem. Use current->bio_list | |
999 | * to collect a list of requests submited by a ->submit_bio method while | |
1000 | * it is active, and then process them after it returned. | |
27a84d54 | 1001 | */ |
3e08773c | 1002 | if (current->bio_list) |
f5fe1b51 | 1003 | bio_list_add(¤t->bio_list[0], bio); |
3e08773c CH |
1004 | else if (!bio->bi_bdev->bd_disk->fops->submit_bio) |
1005 | __submit_bio_noacct_mq(bio); | |
1006 | else | |
1007 | __submit_bio_noacct(bio); | |
d89d8796 | 1008 | } |
ed00aabd | 1009 | EXPORT_SYMBOL(submit_bio_noacct); |
1da177e4 LT |
1010 | |
1011 | /** | |
710027a4 | 1012 | * submit_bio - submit a bio to the block device layer for I/O |
1da177e4 LT |
1013 | * @bio: The &struct bio which describes the I/O |
1014 | * | |
3fdd4086 CH |
1015 | * submit_bio() is used to submit I/O requests to block devices. It is passed a |
1016 | * fully set up &struct bio that describes the I/O that needs to be done. The | |
309dca30 | 1017 | * bio will be send to the device described by the bi_bdev field. |
1da177e4 | 1018 | * |
3fdd4086 CH |
1019 | * The success/failure status of the request, along with notification of |
1020 | * completion, is delivered asynchronously through the ->bi_end_io() callback | |
1021 | * in @bio. The bio must NOT be touched by thecaller until ->bi_end_io() has | |
1022 | * been called. | |
1da177e4 | 1023 | */ |
3e08773c | 1024 | void submit_bio(struct bio *bio) |
1da177e4 | 1025 | { |
d3f77dfd | 1026 | if (blkcg_punt_bio_submit(bio)) |
3e08773c | 1027 | return; |
d3f77dfd | 1028 | |
bf2de6f5 JA |
1029 | /* |
1030 | * If it's a regular read/write or a barrier with data attached, | |
1031 | * go through the normal accounting stuff before submission. | |
1032 | */ | |
e2a60da7 | 1033 | if (bio_has_data(bio)) { |
4363ac7c MP |
1034 | unsigned int count; |
1035 | ||
95fe6c1a | 1036 | if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME)) |
309dca30 | 1037 | count = queue_logical_block_size( |
eab4e027 | 1038 | bdev_get_queue(bio->bi_bdev)) >> 9; |
4363ac7c MP |
1039 | else |
1040 | count = bio_sectors(bio); | |
1041 | ||
a8ebb056 | 1042 | if (op_is_write(bio_op(bio))) { |
bf2de6f5 JA |
1043 | count_vm_events(PGPGOUT, count); |
1044 | } else { | |
4f024f37 | 1045 | task_io_account_read(bio->bi_iter.bi_size); |
bf2de6f5 JA |
1046 | count_vm_events(PGPGIN, count); |
1047 | } | |
1da177e4 LT |
1048 | } |
1049 | ||
b8e24a93 | 1050 | /* |
760f83ea CH |
1051 | * If we're reading data that is part of the userspace workingset, count |
1052 | * submission time as memory stall. When the device is congested, or | |
1053 | * the submitting cgroup IO-throttled, submission can be a significant | |
1054 | * part of overall IO time. | |
b8e24a93 | 1055 | */ |
760f83ea CH |
1056 | if (unlikely(bio_op(bio) == REQ_OP_READ && |
1057 | bio_flagged(bio, BIO_WORKINGSET))) { | |
1058 | unsigned long pflags; | |
b8e24a93 | 1059 | |
760f83ea | 1060 | psi_memstall_enter(&pflags); |
3e08773c | 1061 | submit_bio_noacct(bio); |
b8e24a93 | 1062 | psi_memstall_leave(&pflags); |
3e08773c | 1063 | return; |
760f83ea CH |
1064 | } |
1065 | ||
3e08773c | 1066 | submit_bio_noacct(bio); |
1da177e4 | 1067 | } |
1da177e4 LT |
1068 | EXPORT_SYMBOL(submit_bio); |
1069 | ||
3e08773c CH |
1070 | /** |
1071 | * bio_poll - poll for BIO completions | |
1072 | * @bio: bio to poll for | |
1073 | * @flags: BLK_POLL_* flags that control the behavior | |
1074 | * | |
1075 | * Poll for completions on queue associated with the bio. Returns number of | |
1076 | * completed entries found. | |
1077 | * | |
1078 | * Note: the caller must either be the context that submitted @bio, or | |
1079 | * be in a RCU critical section to prevent freeing of @bio. | |
1080 | */ | |
5a72e899 | 1081 | int bio_poll(struct bio *bio, struct io_comp_batch *iob, unsigned int flags) |
3e08773c CH |
1082 | { |
1083 | struct request_queue *q = bio->bi_bdev->bd_disk->queue; | |
1084 | blk_qc_t cookie = READ_ONCE(bio->bi_cookie); | |
1085 | int ret; | |
1086 | ||
1087 | if (cookie == BLK_QC_T_NONE || | |
1088 | !test_bit(QUEUE_FLAG_POLL, &q->queue_flags)) | |
1089 | return 0; | |
1090 | ||
1091 | if (current->plug) | |
1092 | blk_flush_plug_list(current->plug, false); | |
1093 | ||
1094 | if (blk_queue_enter(q, BLK_MQ_REQ_NOWAIT)) | |
1095 | return 0; | |
1096 | if (WARN_ON_ONCE(!queue_is_mq(q))) | |
1097 | ret = 0; /* not yet implemented, should not happen */ | |
1098 | else | |
5a72e899 | 1099 | ret = blk_mq_poll(q, cookie, iob, flags); |
3e08773c CH |
1100 | blk_queue_exit(q); |
1101 | return ret; | |
1102 | } | |
1103 | EXPORT_SYMBOL_GPL(bio_poll); | |
1104 | ||
1105 | /* | |
1106 | * Helper to implement file_operations.iopoll. Requires the bio to be stored | |
1107 | * in iocb->private, and cleared before freeing the bio. | |
1108 | */ | |
5a72e899 JA |
1109 | int iocb_bio_iopoll(struct kiocb *kiocb, struct io_comp_batch *iob, |
1110 | unsigned int flags) | |
3e08773c CH |
1111 | { |
1112 | struct bio *bio; | |
1113 | int ret = 0; | |
1114 | ||
1115 | /* | |
1116 | * Note: the bio cache only uses SLAB_TYPESAFE_BY_RCU, so bio can | |
1117 | * point to a freshly allocated bio at this point. If that happens | |
1118 | * we have a few cases to consider: | |
1119 | * | |
1120 | * 1) the bio is beeing initialized and bi_bdev is NULL. We can just | |
1121 | * simply nothing in this case | |
1122 | * 2) the bio points to a not poll enabled device. bio_poll will catch | |
1123 | * this and return 0 | |
1124 | * 3) the bio points to a poll capable device, including but not | |
1125 | * limited to the one that the original bio pointed to. In this | |
1126 | * case we will call into the actual poll method and poll for I/O, | |
1127 | * even if we don't need to, but it won't cause harm either. | |
1128 | * | |
1129 | * For cases 2) and 3) above the RCU grace period ensures that bi_bdev | |
1130 | * is still allocated. Because partitions hold a reference to the whole | |
1131 | * device bdev and thus disk, the disk is also still valid. Grabbing | |
1132 | * a reference to the queue in bio_poll() ensures the hctxs and requests | |
1133 | * are still valid as well. | |
1134 | */ | |
1135 | rcu_read_lock(); | |
1136 | bio = READ_ONCE(kiocb->private); | |
1137 | if (bio && bio->bi_bdev) | |
5a72e899 | 1138 | ret = bio_poll(bio, iob, flags); |
3e08773c CH |
1139 | rcu_read_unlock(); |
1140 | ||
1141 | return ret; | |
1142 | } | |
1143 | EXPORT_SYMBOL_GPL(iocb_bio_iopoll); | |
1144 | ||
82124d60 | 1145 | /** |
bf4e6b4e | 1146 | * blk_cloned_rq_check_limits - Helper function to check a cloned request |
0d720318 | 1147 | * for the new queue limits |
82124d60 KU |
1148 | * @q: the queue |
1149 | * @rq: the request being checked | |
1150 | * | |
1151 | * Description: | |
1152 | * @rq may have been made based on weaker limitations of upper-level queues | |
1153 | * in request stacking drivers, and it may violate the limitation of @q. | |
1154 | * Since the block layer and the underlying device driver trust @rq | |
1155 | * after it is inserted to @q, it should be checked against @q before | |
1156 | * the insertion using this generic function. | |
1157 | * | |
82124d60 | 1158 | * Request stacking drivers like request-based dm may change the queue |
bf4e6b4e HR |
1159 | * limits when retrying requests on other queues. Those requests need |
1160 | * to be checked against the new queue limits again during dispatch. | |
82124d60 | 1161 | */ |
143d2600 | 1162 | static blk_status_t blk_cloned_rq_check_limits(struct request_queue *q, |
bf4e6b4e | 1163 | struct request *rq) |
82124d60 | 1164 | { |
8327cce5 RS |
1165 | unsigned int max_sectors = blk_queue_get_max_sectors(q, req_op(rq)); |
1166 | ||
1167 | if (blk_rq_sectors(rq) > max_sectors) { | |
1168 | /* | |
1169 | * SCSI device does not have a good way to return if | |
1170 | * Write Same/Zero is actually supported. If a device rejects | |
1171 | * a non-read/write command (discard, write same,etc.) the | |
1172 | * low-level device driver will set the relevant queue limit to | |
1173 | * 0 to prevent blk-lib from issuing more of the offending | |
1174 | * operations. Commands queued prior to the queue limit being | |
1175 | * reset need to be completed with BLK_STS_NOTSUPP to avoid I/O | |
1176 | * errors being propagated to upper layers. | |
1177 | */ | |
1178 | if (max_sectors == 0) | |
1179 | return BLK_STS_NOTSUPP; | |
1180 | ||
61939b12 | 1181 | printk(KERN_ERR "%s: over max size limit. (%u > %u)\n", |
8327cce5 | 1182 | __func__, blk_rq_sectors(rq), max_sectors); |
143d2600 | 1183 | return BLK_STS_IOERR; |
82124d60 KU |
1184 | } |
1185 | ||
1186 | /* | |
9bb33f24 CH |
1187 | * The queue settings related to segment counting may differ from the |
1188 | * original queue. | |
82124d60 | 1189 | */ |
e9cd19c0 | 1190 | rq->nr_phys_segments = blk_recalc_rq_segments(rq); |
8a78362c | 1191 | if (rq->nr_phys_segments > queue_max_segments(q)) { |
61939b12 JP |
1192 | printk(KERN_ERR "%s: over max segments limit. (%hu > %hu)\n", |
1193 | __func__, rq->nr_phys_segments, queue_max_segments(q)); | |
143d2600 | 1194 | return BLK_STS_IOERR; |
82124d60 KU |
1195 | } |
1196 | ||
143d2600 | 1197 | return BLK_STS_OK; |
82124d60 | 1198 | } |
82124d60 KU |
1199 | |
1200 | /** | |
1201 | * blk_insert_cloned_request - Helper for stacking drivers to submit a request | |
1202 | * @q: the queue to submit the request | |
1203 | * @rq: the request being queued | |
1204 | */ | |
2a842aca | 1205 | blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq) |
82124d60 | 1206 | { |
8327cce5 RS |
1207 | blk_status_t ret; |
1208 | ||
1209 | ret = blk_cloned_rq_check_limits(q, rq); | |
1210 | if (ret != BLK_STS_OK) | |
1211 | return ret; | |
82124d60 | 1212 | |
b2c9cd37 | 1213 | if (rq->rq_disk && |
8446fe92 | 1214 | should_fail_request(rq->rq_disk->part0, blk_rq_bytes(rq))) |
2a842aca | 1215 | return BLK_STS_IOERR; |
82124d60 | 1216 | |
a892c8d5 ST |
1217 | if (blk_crypto_insert_cloned_request(rq)) |
1218 | return BLK_STS_IOERR; | |
1219 | ||
be6bfe36 | 1220 | blk_account_io_start(rq); |
82124d60 KU |
1221 | |
1222 | /* | |
a1ce35fa JA |
1223 | * Since we have a scheduler attached on the top device, |
1224 | * bypass a potential scheduler on the bottom device for | |
1225 | * insert. | |
82124d60 | 1226 | */ |
fd9c40f6 | 1227 | return blk_mq_request_issue_directly(rq, true); |
82124d60 KU |
1228 | } |
1229 | EXPORT_SYMBOL_GPL(blk_insert_cloned_request); | |
1230 | ||
80a761fd TH |
1231 | /** |
1232 | * blk_rq_err_bytes - determine number of bytes till the next failure boundary | |
1233 | * @rq: request to examine | |
1234 | * | |
1235 | * Description: | |
1236 | * A request could be merge of IOs which require different failure | |
1237 | * handling. This function determines the number of bytes which | |
1238 | * can be failed from the beginning of the request without | |
1239 | * crossing into area which need to be retried further. | |
1240 | * | |
1241 | * Return: | |
1242 | * The number of bytes to fail. | |
80a761fd TH |
1243 | */ |
1244 | unsigned int blk_rq_err_bytes(const struct request *rq) | |
1245 | { | |
1246 | unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK; | |
1247 | unsigned int bytes = 0; | |
1248 | struct bio *bio; | |
1249 | ||
e8064021 | 1250 | if (!(rq->rq_flags & RQF_MIXED_MERGE)) |
80a761fd TH |
1251 | return blk_rq_bytes(rq); |
1252 | ||
1253 | /* | |
1254 | * Currently the only 'mixing' which can happen is between | |
1255 | * different fastfail types. We can safely fail portions | |
1256 | * which have all the failfast bits that the first one has - | |
1257 | * the ones which are at least as eager to fail as the first | |
1258 | * one. | |
1259 | */ | |
1260 | for (bio = rq->bio; bio; bio = bio->bi_next) { | |
1eff9d32 | 1261 | if ((bio->bi_opf & ff) != ff) |
80a761fd | 1262 | break; |
4f024f37 | 1263 | bytes += bio->bi_iter.bi_size; |
80a761fd TH |
1264 | } |
1265 | ||
1266 | /* this could lead to infinite loop */ | |
1267 | BUG_ON(blk_rq_bytes(rq) && !bytes); | |
1268 | return bytes; | |
1269 | } | |
1270 | EXPORT_SYMBOL_GPL(blk_rq_err_bytes); | |
1271 | ||
8446fe92 CH |
1272 | static void update_io_ticks(struct block_device *part, unsigned long now, |
1273 | bool end) | |
9123bf6f CH |
1274 | { |
1275 | unsigned long stamp; | |
1276 | again: | |
8446fe92 | 1277 | stamp = READ_ONCE(part->bd_stamp); |
d80c228d | 1278 | if (unlikely(time_after(now, stamp))) { |
8446fe92 | 1279 | if (likely(cmpxchg(&part->bd_stamp, stamp, now) == stamp)) |
9123bf6f CH |
1280 | __part_stat_add(part, io_ticks, end ? now - stamp : 1); |
1281 | } | |
8446fe92 CH |
1282 | if (part->bd_partno) { |
1283 | part = bdev_whole(part); | |
9123bf6f CH |
1284 | goto again; |
1285 | } | |
1286 | } | |
1287 | ||
be6bfe36 | 1288 | void __blk_account_io_done(struct request *req, u64 now) |
bc58ba94 | 1289 | { |
be6bfe36 | 1290 | const int sgrp = op_stat_group(req_op(req)); |
bc58ba94 | 1291 | |
be6bfe36 PB |
1292 | part_stat_lock(); |
1293 | update_io_ticks(req->part, jiffies, true); | |
1294 | part_stat_inc(req->part, ios[sgrp]); | |
1295 | part_stat_add(req->part, nsecs[sgrp], now - req->start_time_ns); | |
1296 | part_stat_unlock(); | |
bc58ba94 JA |
1297 | } |
1298 | ||
be6bfe36 | 1299 | void __blk_account_io_start(struct request *rq) |
320ae51f | 1300 | { |
0b6e522c CH |
1301 | /* passthrough requests can hold bios that do not have ->bi_bdev set */ |
1302 | if (rq->bio && rq->bio->bi_bdev) | |
1303 | rq->part = rq->bio->bi_bdev; | |
1304 | else | |
1305 | rq->part = rq->rq_disk->part0; | |
524f9ffd | 1306 | |
112f158f | 1307 | part_stat_lock(); |
76268f3a | 1308 | update_io_ticks(rq->part, jiffies, false); |
320ae51f JA |
1309 | part_stat_unlock(); |
1310 | } | |
320ae51f | 1311 | |
8446fe92 | 1312 | static unsigned long __part_start_io_acct(struct block_device *part, |
7b26410b | 1313 | unsigned int sectors, unsigned int op) |
956d510e | 1314 | { |
956d510e CH |
1315 | const int sgrp = op_stat_group(op); |
1316 | unsigned long now = READ_ONCE(jiffies); | |
1317 | ||
1318 | part_stat_lock(); | |
1319 | update_io_ticks(part, now, false); | |
1320 | part_stat_inc(part, ios[sgrp]); | |
1321 | part_stat_add(part, sectors[sgrp], sectors); | |
1322 | part_stat_local_inc(part, in_flight[op_is_write(op)]); | |
1323 | part_stat_unlock(); | |
320ae51f | 1324 | |
956d510e CH |
1325 | return now; |
1326 | } | |
7b26410b | 1327 | |
99dfc43e CH |
1328 | /** |
1329 | * bio_start_io_acct - start I/O accounting for bio based drivers | |
1330 | * @bio: bio to start account for | |
1331 | * | |
1332 | * Returns the start time that should be passed back to bio_end_io_acct(). | |
1333 | */ | |
1334 | unsigned long bio_start_io_acct(struct bio *bio) | |
7b26410b | 1335 | { |
99dfc43e | 1336 | return __part_start_io_acct(bio->bi_bdev, bio_sectors(bio), bio_op(bio)); |
7b26410b | 1337 | } |
99dfc43e | 1338 | EXPORT_SYMBOL_GPL(bio_start_io_acct); |
7b26410b SL |
1339 | |
1340 | unsigned long disk_start_io_acct(struct gendisk *disk, unsigned int sectors, | |
1341 | unsigned int op) | |
1342 | { | |
8446fe92 | 1343 | return __part_start_io_acct(disk->part0, sectors, op); |
7b26410b | 1344 | } |
956d510e CH |
1345 | EXPORT_SYMBOL(disk_start_io_acct); |
1346 | ||
8446fe92 | 1347 | static void __part_end_io_acct(struct block_device *part, unsigned int op, |
7b26410b | 1348 | unsigned long start_time) |
956d510e | 1349 | { |
956d510e CH |
1350 | const int sgrp = op_stat_group(op); |
1351 | unsigned long now = READ_ONCE(jiffies); | |
1352 | unsigned long duration = now - start_time; | |
5b18b5a7 | 1353 | |
956d510e CH |
1354 | part_stat_lock(); |
1355 | update_io_ticks(part, now, true); | |
1356 | part_stat_add(part, nsecs[sgrp], jiffies_to_nsecs(duration)); | |
1357 | part_stat_local_dec(part, in_flight[op_is_write(op)]); | |
320ae51f JA |
1358 | part_stat_unlock(); |
1359 | } | |
7b26410b | 1360 | |
99dfc43e CH |
1361 | void bio_end_io_acct_remapped(struct bio *bio, unsigned long start_time, |
1362 | struct block_device *orig_bdev) | |
7b26410b | 1363 | { |
99dfc43e | 1364 | __part_end_io_acct(orig_bdev, bio_op(bio), start_time); |
7b26410b | 1365 | } |
99dfc43e | 1366 | EXPORT_SYMBOL_GPL(bio_end_io_acct_remapped); |
7b26410b SL |
1367 | |
1368 | void disk_end_io_acct(struct gendisk *disk, unsigned int op, | |
1369 | unsigned long start_time) | |
1370 | { | |
8446fe92 | 1371 | __part_end_io_acct(disk->part0, op, start_time); |
7b26410b | 1372 | } |
956d510e | 1373 | EXPORT_SYMBOL(disk_end_io_acct); |
320ae51f | 1374 | |
ef71de8b CH |
1375 | /* |
1376 | * Steal bios from a request and add them to a bio list. | |
1377 | * The request must not have been partially completed before. | |
1378 | */ | |
1379 | void blk_steal_bios(struct bio_list *list, struct request *rq) | |
1380 | { | |
1381 | if (rq->bio) { | |
1382 | if (list->tail) | |
1383 | list->tail->bi_next = rq->bio; | |
1384 | else | |
1385 | list->head = rq->bio; | |
1386 | list->tail = rq->biotail; | |
1387 | ||
1388 | rq->bio = NULL; | |
1389 | rq->biotail = NULL; | |
1390 | } | |
1391 | ||
1392 | rq->__data_len = 0; | |
1393 | } | |
1394 | EXPORT_SYMBOL_GPL(blk_steal_bios); | |
1395 | ||
2d4dc890 IL |
1396 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE |
1397 | /** | |
1398 | * rq_flush_dcache_pages - Helper function to flush all pages in a request | |
1399 | * @rq: the request to be flushed | |
1400 | * | |
1401 | * Description: | |
1402 | * Flush all pages in @rq. | |
1403 | */ | |
1404 | void rq_flush_dcache_pages(struct request *rq) | |
1405 | { | |
1406 | struct req_iterator iter; | |
7988613b | 1407 | struct bio_vec bvec; |
2d4dc890 IL |
1408 | |
1409 | rq_for_each_segment(bvec, rq, iter) | |
7988613b | 1410 | flush_dcache_page(bvec.bv_page); |
2d4dc890 IL |
1411 | } |
1412 | EXPORT_SYMBOL_GPL(rq_flush_dcache_pages); | |
1413 | #endif | |
1414 | ||
ef9e3fac KU |
1415 | /** |
1416 | * blk_lld_busy - Check if underlying low-level drivers of a device are busy | |
1417 | * @q : the queue of the device being checked | |
1418 | * | |
1419 | * Description: | |
1420 | * Check if underlying low-level drivers of a device are busy. | |
1421 | * If the drivers want to export their busy state, they must set own | |
1422 | * exporting function using blk_queue_lld_busy() first. | |
1423 | * | |
1424 | * Basically, this function is used only by request stacking drivers | |
1425 | * to stop dispatching requests to underlying devices when underlying | |
1426 | * devices are busy. This behavior helps more I/O merging on the queue | |
1427 | * of the request stacking driver and prevents I/O throughput regression | |
1428 | * on burst I/O load. | |
1429 | * | |
1430 | * Return: | |
1431 | * 0 - Not busy (The request stacking driver should dispatch request) | |
1432 | * 1 - Busy (The request stacking driver should stop dispatching request) | |
1433 | */ | |
1434 | int blk_lld_busy(struct request_queue *q) | |
1435 | { | |
344e9ffc | 1436 | if (queue_is_mq(q) && q->mq_ops->busy) |
9ba20527 | 1437 | return q->mq_ops->busy(q); |
ef9e3fac KU |
1438 | |
1439 | return 0; | |
1440 | } | |
1441 | EXPORT_SYMBOL_GPL(blk_lld_busy); | |
1442 | ||
78d8e58a MS |
1443 | /** |
1444 | * blk_rq_unprep_clone - Helper function to free all bios in a cloned request | |
1445 | * @rq: the clone request to be cleaned up | |
1446 | * | |
1447 | * Description: | |
1448 | * Free all bios in @rq for a cloned request. | |
1449 | */ | |
1450 | void blk_rq_unprep_clone(struct request *rq) | |
1451 | { | |
1452 | struct bio *bio; | |
1453 | ||
1454 | while ((bio = rq->bio) != NULL) { | |
1455 | rq->bio = bio->bi_next; | |
1456 | ||
1457 | bio_put(bio); | |
1458 | } | |
1459 | } | |
1460 | EXPORT_SYMBOL_GPL(blk_rq_unprep_clone); | |
1461 | ||
78d8e58a MS |
1462 | /** |
1463 | * blk_rq_prep_clone - Helper function to setup clone request | |
1464 | * @rq: the request to be setup | |
1465 | * @rq_src: original request to be cloned | |
1466 | * @bs: bio_set that bios for clone are allocated from | |
1467 | * @gfp_mask: memory allocation mask for bio | |
1468 | * @bio_ctr: setup function to be called for each clone bio. | |
1469 | * Returns %0 for success, non %0 for failure. | |
1470 | * @data: private data to be passed to @bio_ctr | |
1471 | * | |
1472 | * Description: | |
1473 | * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq. | |
78d8e58a MS |
1474 | * Also, pages which the original bios are pointing to are not copied |
1475 | * and the cloned bios just point same pages. | |
1476 | * So cloned bios must be completed before original bios, which means | |
1477 | * the caller must complete @rq before @rq_src. | |
1478 | */ | |
1479 | int blk_rq_prep_clone(struct request *rq, struct request *rq_src, | |
1480 | struct bio_set *bs, gfp_t gfp_mask, | |
1481 | int (*bio_ctr)(struct bio *, struct bio *, void *), | |
1482 | void *data) | |
1483 | { | |
1484 | struct bio *bio, *bio_src; | |
1485 | ||
1486 | if (!bs) | |
f4f8154a | 1487 | bs = &fs_bio_set; |
78d8e58a MS |
1488 | |
1489 | __rq_for_each_bio(bio_src, rq_src) { | |
1490 | bio = bio_clone_fast(bio_src, gfp_mask, bs); | |
1491 | if (!bio) | |
1492 | goto free_and_out; | |
1493 | ||
1494 | if (bio_ctr && bio_ctr(bio, bio_src, data)) | |
1495 | goto free_and_out; | |
1496 | ||
1497 | if (rq->bio) { | |
1498 | rq->biotail->bi_next = bio; | |
1499 | rq->biotail = bio; | |
93f221ae | 1500 | } else { |
78d8e58a | 1501 | rq->bio = rq->biotail = bio; |
93f221ae EB |
1502 | } |
1503 | bio = NULL; | |
78d8e58a MS |
1504 | } |
1505 | ||
361301a2 GJ |
1506 | /* Copy attributes of the original request to the clone request. */ |
1507 | rq->__sector = blk_rq_pos(rq_src); | |
1508 | rq->__data_len = blk_rq_bytes(rq_src); | |
1509 | if (rq_src->rq_flags & RQF_SPECIAL_PAYLOAD) { | |
1510 | rq->rq_flags |= RQF_SPECIAL_PAYLOAD; | |
1511 | rq->special_vec = rq_src->special_vec; | |
1512 | } | |
1513 | rq->nr_phys_segments = rq_src->nr_phys_segments; | |
1514 | rq->ioprio = rq_src->ioprio; | |
78d8e58a | 1515 | |
93f221ae EB |
1516 | if (rq->bio && blk_crypto_rq_bio_prep(rq, rq->bio, gfp_mask) < 0) |
1517 | goto free_and_out; | |
78d8e58a MS |
1518 | |
1519 | return 0; | |
1520 | ||
1521 | free_and_out: | |
1522 | if (bio) | |
1523 | bio_put(bio); | |
1524 | blk_rq_unprep_clone(rq); | |
1525 | ||
1526 | return -ENOMEM; | |
b0fd271d KU |
1527 | } |
1528 | EXPORT_SYMBOL_GPL(blk_rq_prep_clone); | |
1529 | ||
59c3d45e | 1530 | int kblockd_schedule_work(struct work_struct *work) |
1da177e4 LT |
1531 | { |
1532 | return queue_work(kblockd_workqueue, work); | |
1533 | } | |
1da177e4 LT |
1534 | EXPORT_SYMBOL(kblockd_schedule_work); |
1535 | ||
818cd1cb JA |
1536 | int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork, |
1537 | unsigned long delay) | |
1538 | { | |
1539 | return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay); | |
1540 | } | |
1541 | EXPORT_SYMBOL(kblockd_mod_delayed_work_on); | |
1542 | ||
47c122e3 JA |
1543 | void blk_start_plug_nr_ios(struct blk_plug *plug, unsigned short nr_ios) |
1544 | { | |
1545 | struct task_struct *tsk = current; | |
1546 | ||
1547 | /* | |
1548 | * If this is a nested plug, don't actually assign it. | |
1549 | */ | |
1550 | if (tsk->plug) | |
1551 | return; | |
1552 | ||
1553 | INIT_LIST_HEAD(&plug->mq_list); | |
1554 | plug->cached_rq = NULL; | |
1555 | plug->nr_ios = min_t(unsigned short, nr_ios, BLK_MAX_REQUEST_COUNT); | |
1556 | plug->rq_count = 0; | |
1557 | plug->multiple_queues = false; | |
1558 | plug->nowait = false; | |
1559 | INIT_LIST_HEAD(&plug->cb_list); | |
1560 | ||
1561 | /* | |
1562 | * Store ordering should not be needed here, since a potential | |
1563 | * preempt will imply a full memory barrier | |
1564 | */ | |
1565 | tsk->plug = plug; | |
1566 | } | |
1567 | ||
75df7136 SJ |
1568 | /** |
1569 | * blk_start_plug - initialize blk_plug and track it inside the task_struct | |
1570 | * @plug: The &struct blk_plug that needs to be initialized | |
1571 | * | |
1572 | * Description: | |
40405851 JM |
1573 | * blk_start_plug() indicates to the block layer an intent by the caller |
1574 | * to submit multiple I/O requests in a batch. The block layer may use | |
1575 | * this hint to defer submitting I/Os from the caller until blk_finish_plug() | |
1576 | * is called. However, the block layer may choose to submit requests | |
1577 | * before a call to blk_finish_plug() if the number of queued I/Os | |
1578 | * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than | |
1579 | * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if | |
1580 | * the task schedules (see below). | |
1581 | * | |
75df7136 SJ |
1582 | * Tracking blk_plug inside the task_struct will help with auto-flushing the |
1583 | * pending I/O should the task end up blocking between blk_start_plug() and | |
1584 | * blk_finish_plug(). This is important from a performance perspective, but | |
1585 | * also ensures that we don't deadlock. For instance, if the task is blocking | |
1586 | * for a memory allocation, memory reclaim could end up wanting to free a | |
1587 | * page belonging to that request that is currently residing in our private | |
1588 | * plug. By flushing the pending I/O when the process goes to sleep, we avoid | |
1589 | * this kind of deadlock. | |
1590 | */ | |
73c10101 JA |
1591 | void blk_start_plug(struct blk_plug *plug) |
1592 | { | |
47c122e3 | 1593 | blk_start_plug_nr_ios(plug, 1); |
73c10101 JA |
1594 | } |
1595 | EXPORT_SYMBOL(blk_start_plug); | |
1596 | ||
74018dc3 | 1597 | static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule) |
048c9374 N |
1598 | { |
1599 | LIST_HEAD(callbacks); | |
1600 | ||
2a7d5559 SL |
1601 | while (!list_empty(&plug->cb_list)) { |
1602 | list_splice_init(&plug->cb_list, &callbacks); | |
048c9374 | 1603 | |
2a7d5559 SL |
1604 | while (!list_empty(&callbacks)) { |
1605 | struct blk_plug_cb *cb = list_first_entry(&callbacks, | |
048c9374 N |
1606 | struct blk_plug_cb, |
1607 | list); | |
2a7d5559 | 1608 | list_del(&cb->list); |
74018dc3 | 1609 | cb->callback(cb, from_schedule); |
2a7d5559 | 1610 | } |
048c9374 N |
1611 | } |
1612 | } | |
1613 | ||
9cbb1750 N |
1614 | struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data, |
1615 | int size) | |
1616 | { | |
1617 | struct blk_plug *plug = current->plug; | |
1618 | struct blk_plug_cb *cb; | |
1619 | ||
1620 | if (!plug) | |
1621 | return NULL; | |
1622 | ||
1623 | list_for_each_entry(cb, &plug->cb_list, list) | |
1624 | if (cb->callback == unplug && cb->data == data) | |
1625 | return cb; | |
1626 | ||
1627 | /* Not currently on the callback list */ | |
1628 | BUG_ON(size < sizeof(*cb)); | |
1629 | cb = kzalloc(size, GFP_ATOMIC); | |
1630 | if (cb) { | |
1631 | cb->data = data; | |
1632 | cb->callback = unplug; | |
1633 | list_add(&cb->list, &plug->cb_list); | |
1634 | } | |
1635 | return cb; | |
1636 | } | |
1637 | EXPORT_SYMBOL(blk_check_plugged); | |
1638 | ||
49cac01e | 1639 | void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule) |
73c10101 | 1640 | { |
74018dc3 | 1641 | flush_plug_callbacks(plug, from_schedule); |
320ae51f JA |
1642 | |
1643 | if (!list_empty(&plug->mq_list)) | |
1644 | blk_mq_flush_plug_list(plug, from_schedule); | |
47c122e3 JA |
1645 | if (unlikely(!from_schedule && plug->cached_rq)) |
1646 | blk_mq_free_plug_rqs(plug); | |
73c10101 | 1647 | } |
73c10101 | 1648 | |
40405851 JM |
1649 | /** |
1650 | * blk_finish_plug - mark the end of a batch of submitted I/O | |
1651 | * @plug: The &struct blk_plug passed to blk_start_plug() | |
1652 | * | |
1653 | * Description: | |
1654 | * Indicate that a batch of I/O submissions is complete. This function | |
1655 | * must be paired with an initial call to blk_start_plug(). The intent | |
1656 | * is to allow the block layer to optimize I/O submission. See the | |
1657 | * documentation for blk_start_plug() for more information. | |
1658 | */ | |
73c10101 JA |
1659 | void blk_finish_plug(struct blk_plug *plug) |
1660 | { | |
dd6cf3e1 SL |
1661 | if (plug != current->plug) |
1662 | return; | |
f6603783 | 1663 | blk_flush_plug_list(plug, false); |
73c10101 | 1664 | |
dd6cf3e1 | 1665 | current->plug = NULL; |
73c10101 | 1666 | } |
88b996cd | 1667 | EXPORT_SYMBOL(blk_finish_plug); |
73c10101 | 1668 | |
71ac860a ML |
1669 | void blk_io_schedule(void) |
1670 | { | |
1671 | /* Prevent hang_check timer from firing at us during very long I/O */ | |
1672 | unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2; | |
1673 | ||
1674 | if (timeout) | |
1675 | io_schedule_timeout(timeout); | |
1676 | else | |
1677 | io_schedule(); | |
1678 | } | |
1679 | EXPORT_SYMBOL_GPL(blk_io_schedule); | |
1680 | ||
1da177e4 LT |
1681 | int __init blk_dev_init(void) |
1682 | { | |
ef295ecf CH |
1683 | BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS)); |
1684 | BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * | |
c593642c | 1685 | sizeof_field(struct request, cmd_flags)); |
ef295ecf | 1686 | BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * |
c593642c | 1687 | sizeof_field(struct bio, bi_opf)); |
9eb55b03 | 1688 | |
89b90be2 TH |
1689 | /* used for unplugging and affects IO latency/throughput - HIGHPRI */ |
1690 | kblockd_workqueue = alloc_workqueue("kblockd", | |
28747fcd | 1691 | WQ_MEM_RECLAIM | WQ_HIGHPRI, 0); |
1da177e4 LT |
1692 | if (!kblockd_workqueue) |
1693 | panic("Failed to create kblockd\n"); | |
1694 | ||
c2789bd4 | 1695 | blk_requestq_cachep = kmem_cache_create("request_queue", |
165125e1 | 1696 | sizeof(struct request_queue), 0, SLAB_PANIC, NULL); |
1da177e4 | 1697 | |
18fbda91 | 1698 | blk_debugfs_root = debugfs_create_dir("block", NULL); |
18fbda91 | 1699 | |
d38ecf93 | 1700 | return 0; |
1da177e4 | 1701 | } |