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> | |
17 | #include <linux/backing-dev.h> | |
18 | #include <linux/bio.h> | |
19 | #include <linux/blkdev.h> | |
320ae51f | 20 | #include <linux/blk-mq.h> |
52abca64 | 21 | #include <linux/blk-pm.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" |
c1c80384 | 53 | #include "blk-rq-qos.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); |
0bfc2455 | 62 | |
a73f730d TH |
63 | DEFINE_IDA(blk_queue_ida); |
64 | ||
1da177e4 LT |
65 | /* |
66 | * For queue allocation | |
67 | */ | |
6728cb0e | 68 | struct kmem_cache *blk_requestq_cachep; |
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 | ||
2a4aa30c | 111 | void blk_rq_init(struct request_queue *q, struct request *rq) |
1da177e4 | 112 | { |
1afb20f3 FT |
113 | memset(rq, 0, sizeof(*rq)); |
114 | ||
1da177e4 | 115 | INIT_LIST_HEAD(&rq->queuelist); |
63a71386 | 116 | rq->q = q; |
a2dec7b3 | 117 | rq->__sector = (sector_t) -1; |
2e662b65 JA |
118 | INIT_HLIST_NODE(&rq->hash); |
119 | RB_CLEAR_NODE(&rq->rb_node); | |
e44a6a23 XT |
120 | rq->tag = BLK_MQ_NO_TAG; |
121 | rq->internal_tag = BLK_MQ_NO_TAG; | |
522a7775 | 122 | rq->start_time_ns = ktime_get_ns(); |
09e099d4 | 123 | rq->part = NULL; |
b554db14 | 124 | refcount_set(&rq->ref, 1); |
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), | |
144 | REQ_OP_NAME(SCSI_IN), | |
145 | REQ_OP_NAME(SCSI_OUT), | |
146 | REQ_OP_NAME(DRV_IN), | |
147 | REQ_OP_NAME(DRV_OUT), | |
148 | }; | |
149 | #undef REQ_OP_NAME | |
150 | ||
151 | /** | |
152 | * blk_op_str - Return string XXX in the REQ_OP_XXX. | |
153 | * @op: REQ_OP_XXX. | |
154 | * | |
155 | * Description: Centralize block layer function to convert REQ_OP_XXX into | |
156 | * string format. Useful in the debugging and tracing bio or request. For | |
157 | * invalid REQ_OP_XXX it returns string "UNKNOWN". | |
158 | */ | |
159 | inline const char *blk_op_str(unsigned int op) | |
160 | { | |
161 | const char *op_str = "UNKNOWN"; | |
162 | ||
163 | if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op]) | |
164 | op_str = blk_op_name[op]; | |
165 | ||
166 | return op_str; | |
167 | } | |
168 | EXPORT_SYMBOL_GPL(blk_op_str); | |
169 | ||
2a842aca CH |
170 | static const struct { |
171 | int errno; | |
172 | const char *name; | |
173 | } blk_errors[] = { | |
174 | [BLK_STS_OK] = { 0, "" }, | |
175 | [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" }, | |
176 | [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" }, | |
177 | [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" }, | |
178 | [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" }, | |
179 | [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" }, | |
180 | [BLK_STS_NEXUS] = { -EBADE, "critical nexus" }, | |
181 | [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" }, | |
182 | [BLK_STS_PROTECTION] = { -EILSEQ, "protection" }, | |
183 | [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" }, | |
86ff7c2a | 184 | [BLK_STS_DEV_RESOURCE] = { -EBUSY, "device resource" }, |
03a07c92 | 185 | [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" }, |
2a842aca | 186 | |
4e4cbee9 CH |
187 | /* device mapper special case, should not leak out: */ |
188 | [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" }, | |
189 | ||
3b481d91 KB |
190 | /* zone device specific errors */ |
191 | [BLK_STS_ZONE_OPEN_RESOURCE] = { -ETOOMANYREFS, "open zones exceeded" }, | |
192 | [BLK_STS_ZONE_ACTIVE_RESOURCE] = { -EOVERFLOW, "active zones exceeded" }, | |
193 | ||
2a842aca CH |
194 | /* everything else not covered above: */ |
195 | [BLK_STS_IOERR] = { -EIO, "I/O" }, | |
196 | }; | |
197 | ||
198 | blk_status_t errno_to_blk_status(int errno) | |
199 | { | |
200 | int i; | |
201 | ||
202 | for (i = 0; i < ARRAY_SIZE(blk_errors); i++) { | |
203 | if (blk_errors[i].errno == errno) | |
204 | return (__force blk_status_t)i; | |
205 | } | |
206 | ||
207 | return BLK_STS_IOERR; | |
208 | } | |
209 | EXPORT_SYMBOL_GPL(errno_to_blk_status); | |
210 | ||
211 | int blk_status_to_errno(blk_status_t status) | |
212 | { | |
213 | int idx = (__force int)status; | |
214 | ||
34bd9c1c | 215 | if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors))) |
2a842aca CH |
216 | return -EIO; |
217 | return blk_errors[idx].errno; | |
218 | } | |
219 | EXPORT_SYMBOL_GPL(blk_status_to_errno); | |
220 | ||
178cc590 CH |
221 | static void print_req_error(struct request *req, blk_status_t status, |
222 | const char *caller) | |
2a842aca CH |
223 | { |
224 | int idx = (__force int)status; | |
225 | ||
34bd9c1c | 226 | if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors))) |
2a842aca CH |
227 | return; |
228 | ||
178cc590 | 229 | printk_ratelimited(KERN_ERR |
b0e5168a CK |
230 | "%s: %s error, dev %s, sector %llu op 0x%x:(%s) flags 0x%x " |
231 | "phys_seg %u prio class %u\n", | |
178cc590 | 232 | caller, blk_errors[idx].name, |
b0e5168a CK |
233 | req->rq_disk ? req->rq_disk->disk_name : "?", |
234 | blk_rq_pos(req), req_op(req), blk_op_str(req_op(req)), | |
235 | req->cmd_flags & ~REQ_OP_MASK, | |
236 | req->nr_phys_segments, | |
237 | IOPRIO_PRIO_CLASS(req->ioprio)); | |
2a842aca CH |
238 | } |
239 | ||
5bb23a68 | 240 | static void req_bio_endio(struct request *rq, struct bio *bio, |
2a842aca | 241 | unsigned int nbytes, blk_status_t error) |
1da177e4 | 242 | { |
78d8e58a | 243 | if (error) |
4e4cbee9 | 244 | bio->bi_status = error; |
797e7dbb | 245 | |
e8064021 | 246 | if (unlikely(rq->rq_flags & RQF_QUIET)) |
b7c44ed9 | 247 | bio_set_flag(bio, BIO_QUIET); |
08bafc03 | 248 | |
f79ea416 | 249 | bio_advance(bio, nbytes); |
7ba1ba12 | 250 | |
0512a75b KB |
251 | if (req_op(rq) == REQ_OP_ZONE_APPEND && error == BLK_STS_OK) { |
252 | /* | |
253 | * Partial zone append completions cannot be supported as the | |
254 | * BIO fragments may end up not being written sequentially. | |
255 | */ | |
256 | if (bio->bi_iter.bi_size) | |
257 | bio->bi_status = BLK_STS_IOERR; | |
258 | else | |
259 | bio->bi_iter.bi_sector = rq->__sector; | |
260 | } | |
261 | ||
143a87f4 | 262 | /* don't actually finish bio if it's part of flush sequence */ |
e8064021 | 263 | if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ)) |
4246a0b6 | 264 | bio_endio(bio); |
1da177e4 | 265 | } |
1da177e4 | 266 | |
1da177e4 LT |
267 | void blk_dump_rq_flags(struct request *rq, char *msg) |
268 | { | |
aebf526b CH |
269 | printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg, |
270 | rq->rq_disk ? rq->rq_disk->disk_name : "?", | |
5953316d | 271 | (unsigned long long) rq->cmd_flags); |
1da177e4 | 272 | |
83096ebf TH |
273 | printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n", |
274 | (unsigned long long)blk_rq_pos(rq), | |
275 | blk_rq_sectors(rq), blk_rq_cur_sectors(rq)); | |
b4f42e28 JA |
276 | printk(KERN_INFO " bio %p, biotail %p, len %u\n", |
277 | rq->bio, rq->biotail, blk_rq_bytes(rq)); | |
1da177e4 | 278 | } |
1da177e4 LT |
279 | EXPORT_SYMBOL(blk_dump_rq_flags); |
280 | ||
1da177e4 LT |
281 | /** |
282 | * blk_sync_queue - cancel any pending callbacks on a queue | |
283 | * @q: the queue | |
284 | * | |
285 | * Description: | |
286 | * The block layer may perform asynchronous callback activity | |
287 | * on a queue, such as calling the unplug function after a timeout. | |
288 | * A block device may call blk_sync_queue to ensure that any | |
289 | * such activity is cancelled, thus allowing it to release resources | |
59c51591 | 290 | * that the callbacks might use. The caller must already have made sure |
c62b37d9 | 291 | * that its ->submit_bio will not re-add plugging prior to calling |
1da177e4 LT |
292 | * this function. |
293 | * | |
da527770 | 294 | * This function does not cancel any asynchronous activity arising |
da3dae54 | 295 | * out of elevator or throttling code. That would require elevator_exit() |
5efd6113 | 296 | * and blkcg_exit_queue() to be called with queue lock initialized. |
da527770 | 297 | * |
1da177e4 LT |
298 | */ |
299 | void blk_sync_queue(struct request_queue *q) | |
300 | { | |
70ed28b9 | 301 | del_timer_sync(&q->timeout); |
4e9b6f20 | 302 | cancel_work_sync(&q->timeout_work); |
1da177e4 LT |
303 | } |
304 | EXPORT_SYMBOL(blk_sync_queue); | |
305 | ||
c9254f2d | 306 | /** |
cd84a62e | 307 | * blk_set_pm_only - increment pm_only counter |
c9254f2d | 308 | * @q: request queue pointer |
c9254f2d | 309 | */ |
cd84a62e | 310 | void blk_set_pm_only(struct request_queue *q) |
c9254f2d | 311 | { |
cd84a62e | 312 | atomic_inc(&q->pm_only); |
c9254f2d | 313 | } |
cd84a62e | 314 | EXPORT_SYMBOL_GPL(blk_set_pm_only); |
c9254f2d | 315 | |
cd84a62e | 316 | void blk_clear_pm_only(struct request_queue *q) |
c9254f2d | 317 | { |
cd84a62e BVA |
318 | int pm_only; |
319 | ||
320 | pm_only = atomic_dec_return(&q->pm_only); | |
321 | WARN_ON_ONCE(pm_only < 0); | |
322 | if (pm_only == 0) | |
323 | wake_up_all(&q->mq_freeze_wq); | |
c9254f2d | 324 | } |
cd84a62e | 325 | EXPORT_SYMBOL_GPL(blk_clear_pm_only); |
c9254f2d | 326 | |
b5bd357c LC |
327 | /** |
328 | * blk_put_queue - decrement the request_queue refcount | |
329 | * @q: the request_queue structure to decrement the refcount for | |
330 | * | |
331 | * Decrements the refcount of the request_queue kobject. When this reaches 0 | |
332 | * we'll have blk_release_queue() called. | |
e8c7d14a LC |
333 | * |
334 | * Context: Any context, but the last reference must not be dropped from | |
335 | * atomic context. | |
b5bd357c | 336 | */ |
165125e1 | 337 | void blk_put_queue(struct request_queue *q) |
483f4afc AV |
338 | { |
339 | kobject_put(&q->kobj); | |
340 | } | |
d86e0e83 | 341 | EXPORT_SYMBOL(blk_put_queue); |
483f4afc | 342 | |
aed3ea94 JA |
343 | void blk_set_queue_dying(struct request_queue *q) |
344 | { | |
8814ce8a | 345 | blk_queue_flag_set(QUEUE_FLAG_DYING, q); |
aed3ea94 | 346 | |
d3cfb2a0 ML |
347 | /* |
348 | * When queue DYING flag is set, we need to block new req | |
349 | * entering queue, so we call blk_freeze_queue_start() to | |
350 | * prevent I/O from crossing blk_queue_enter(). | |
351 | */ | |
352 | blk_freeze_queue_start(q); | |
353 | ||
344e9ffc | 354 | if (queue_is_mq(q)) |
aed3ea94 | 355 | blk_mq_wake_waiters(q); |
055f6e18 ML |
356 | |
357 | /* Make blk_queue_enter() reexamine the DYING flag. */ | |
358 | wake_up_all(&q->mq_freeze_wq); | |
aed3ea94 JA |
359 | } |
360 | EXPORT_SYMBOL_GPL(blk_set_queue_dying); | |
361 | ||
c9a929dd TH |
362 | /** |
363 | * blk_cleanup_queue - shutdown a request queue | |
364 | * @q: request queue to shutdown | |
365 | * | |
c246e80d BVA |
366 | * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and |
367 | * put it. All future requests will be failed immediately with -ENODEV. | |
e8c7d14a LC |
368 | * |
369 | * Context: can sleep | |
c94a96ac | 370 | */ |
6728cb0e | 371 | void blk_cleanup_queue(struct request_queue *q) |
483f4afc | 372 | { |
e8c7d14a LC |
373 | /* cannot be called from atomic context */ |
374 | might_sleep(); | |
375 | ||
bae85c15 BVA |
376 | WARN_ON_ONCE(blk_queue_registered(q)); |
377 | ||
3f3299d5 | 378 | /* mark @q DYING, no new request or merges will be allowed afterwards */ |
aed3ea94 | 379 | blk_set_queue_dying(q); |
6ecf23af | 380 | |
57d74df9 CH |
381 | blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q); |
382 | blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q); | |
c9a929dd | 383 | |
c246e80d BVA |
384 | /* |
385 | * Drain all requests queued before DYING marking. Set DEAD flag to | |
67ed8b73 BVA |
386 | * prevent that blk_mq_run_hw_queues() accesses the hardware queues |
387 | * after draining finished. | |
c246e80d | 388 | */ |
3ef28e83 | 389 | blk_freeze_queue(q); |
c57cdf7a ML |
390 | |
391 | rq_qos_exit(q); | |
392 | ||
57d74df9 | 393 | blk_queue_flag_set(QUEUE_FLAG_DEAD, q); |
c9a929dd | 394 | |
5a48fc14 DW |
395 | /* for synchronous bio-based driver finish in-flight integrity i/o */ |
396 | blk_flush_integrity(); | |
397 | ||
c9a929dd | 398 | /* @q won't process any more request, flush async actions */ |
dc3b17cc | 399 | del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer); |
c9a929dd TH |
400 | blk_sync_queue(q); |
401 | ||
344e9ffc | 402 | if (queue_is_mq(q)) |
c7e2d94b | 403 | blk_mq_exit_queue(q); |
a1ce35fa | 404 | |
c3e22192 ML |
405 | /* |
406 | * In theory, request pool of sched_tags belongs to request queue. | |
407 | * However, the current implementation requires tag_set for freeing | |
408 | * requests, so free the pool now. | |
409 | * | |
410 | * Queue has become frozen, there can't be any in-queue requests, so | |
411 | * it is safe to free requests now. | |
412 | */ | |
413 | mutex_lock(&q->sysfs_lock); | |
414 | if (q->elevator) | |
415 | blk_mq_sched_free_requests(q); | |
416 | mutex_unlock(&q->sysfs_lock); | |
417 | ||
3ef28e83 | 418 | percpu_ref_exit(&q->q_usage_counter); |
45a9c9d9 | 419 | |
c9a929dd | 420 | /* @q is and will stay empty, shutdown and put */ |
483f4afc AV |
421 | blk_put_queue(q); |
422 | } | |
1da177e4 LT |
423 | EXPORT_SYMBOL(blk_cleanup_queue); |
424 | ||
3a0a5299 BVA |
425 | /** |
426 | * blk_queue_enter() - try to increase q->q_usage_counter | |
427 | * @q: request queue pointer | |
a4d34da7 | 428 | * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PM |
3a0a5299 | 429 | */ |
9a95e4ef | 430 | int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags) |
3ef28e83 | 431 | { |
a4d34da7 | 432 | const bool pm = flags & BLK_MQ_REQ_PM; |
3a0a5299 | 433 | |
3ef28e83 | 434 | while (true) { |
3a0a5299 | 435 | bool success = false; |
3ef28e83 | 436 | |
818e0fa2 | 437 | rcu_read_lock(); |
3a0a5299 BVA |
438 | if (percpu_ref_tryget_live(&q->q_usage_counter)) { |
439 | /* | |
cd84a62e BVA |
440 | * The code that increments the pm_only counter is |
441 | * responsible for ensuring that that counter is | |
442 | * globally visible before the queue is unfrozen. | |
3a0a5299 | 443 | */ |
52abca64 AS |
444 | if ((pm && queue_rpm_status(q) != RPM_SUSPENDED) || |
445 | !blk_queue_pm_only(q)) { | |
3a0a5299 BVA |
446 | success = true; |
447 | } else { | |
448 | percpu_ref_put(&q->q_usage_counter); | |
449 | } | |
450 | } | |
818e0fa2 | 451 | rcu_read_unlock(); |
3a0a5299 BVA |
452 | |
453 | if (success) | |
3ef28e83 DW |
454 | return 0; |
455 | ||
3a0a5299 | 456 | if (flags & BLK_MQ_REQ_NOWAIT) |
3ef28e83 DW |
457 | return -EBUSY; |
458 | ||
5ed61d3f | 459 | /* |
1671d522 | 460 | * read pair of barrier in blk_freeze_queue_start(), |
5ed61d3f | 461 | * we need to order reading __PERCPU_REF_DEAD flag of |
d3cfb2a0 ML |
462 | * .q_usage_counter and reading .mq_freeze_depth or |
463 | * queue dying flag, otherwise the following wait may | |
464 | * never return if the two reads are reordered. | |
5ed61d3f ML |
465 | */ |
466 | smp_rmb(); | |
467 | ||
1dc3039b | 468 | wait_event(q->mq_freeze_wq, |
7996a8b5 | 469 | (!q->mq_freeze_depth && |
52abca64 | 470 | blk_pm_resume_queue(pm, q)) || |
1dc3039b | 471 | blk_queue_dying(q)); |
3ef28e83 DW |
472 | if (blk_queue_dying(q)) |
473 | return -ENODEV; | |
3ef28e83 DW |
474 | } |
475 | } | |
476 | ||
accea322 CH |
477 | static inline int bio_queue_enter(struct bio *bio) |
478 | { | |
309dca30 | 479 | struct request_queue *q = bio->bi_bdev->bd_disk->queue; |
accea322 CH |
480 | bool nowait = bio->bi_opf & REQ_NOWAIT; |
481 | int ret; | |
482 | ||
483 | ret = blk_queue_enter(q, nowait ? BLK_MQ_REQ_NOWAIT : 0); | |
484 | if (unlikely(ret)) { | |
485 | if (nowait && !blk_queue_dying(q)) | |
486 | bio_wouldblock_error(bio); | |
487 | else | |
488 | bio_io_error(bio); | |
489 | } | |
490 | ||
491 | return ret; | |
492 | } | |
493 | ||
3ef28e83 DW |
494 | void blk_queue_exit(struct request_queue *q) |
495 | { | |
496 | percpu_ref_put(&q->q_usage_counter); | |
497 | } | |
498 | ||
499 | static void blk_queue_usage_counter_release(struct percpu_ref *ref) | |
500 | { | |
501 | struct request_queue *q = | |
502 | container_of(ref, struct request_queue, q_usage_counter); | |
503 | ||
504 | wake_up_all(&q->mq_freeze_wq); | |
505 | } | |
506 | ||
bca237a5 | 507 | static void blk_rq_timed_out_timer(struct timer_list *t) |
287922eb | 508 | { |
bca237a5 | 509 | struct request_queue *q = from_timer(q, t, timeout); |
287922eb CH |
510 | |
511 | kblockd_schedule_work(&q->timeout_work); | |
512 | } | |
513 | ||
2e3c18d0 TH |
514 | static void blk_timeout_work(struct work_struct *work) |
515 | { | |
516 | } | |
517 | ||
c62b37d9 | 518 | struct request_queue *blk_alloc_queue(int node_id) |
1946089a | 519 | { |
165125e1 | 520 | struct request_queue *q; |
338aa96d | 521 | int ret; |
1946089a | 522 | |
8324aa91 | 523 | q = kmem_cache_alloc_node(blk_requestq_cachep, |
3d745ea5 | 524 | GFP_KERNEL | __GFP_ZERO, node_id); |
1da177e4 LT |
525 | if (!q) |
526 | return NULL; | |
527 | ||
cbf62af3 | 528 | q->last_merge = NULL; |
cbf62af3 | 529 | |
3d745ea5 | 530 | q->id = ida_simple_get(&blk_queue_ida, 0, 0, GFP_KERNEL); |
a73f730d | 531 | if (q->id < 0) |
3d2936f4 | 532 | goto fail_q; |
a73f730d | 533 | |
338aa96d KO |
534 | ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS); |
535 | if (ret) | |
54efd50b KO |
536 | goto fail_id; |
537 | ||
aef33c2f | 538 | q->backing_dev_info = bdi_alloc(node_id); |
d03f6cdc JK |
539 | if (!q->backing_dev_info) |
540 | goto fail_split; | |
541 | ||
a83b576c JA |
542 | q->stats = blk_alloc_queue_stats(); |
543 | if (!q->stats) | |
544 | goto fail_stats; | |
545 | ||
5151412d | 546 | q->node = node_id; |
0989a025 | 547 | |
bccf5e26 JG |
548 | atomic_set(&q->nr_active_requests_shared_sbitmap, 0); |
549 | ||
bca237a5 KC |
550 | timer_setup(&q->backing_dev_info->laptop_mode_wb_timer, |
551 | laptop_mode_timer_fn, 0); | |
552 | timer_setup(&q->timeout, blk_rq_timed_out_timer, 0); | |
2e3c18d0 | 553 | INIT_WORK(&q->timeout_work, blk_timeout_work); |
a612fddf | 554 | INIT_LIST_HEAD(&q->icq_list); |
4eef3049 | 555 | #ifdef CONFIG_BLK_CGROUP |
e8989fae | 556 | INIT_LIST_HEAD(&q->blkg_list); |
4eef3049 | 557 | #endif |
483f4afc | 558 | |
8324aa91 | 559 | kobject_init(&q->kobj, &blk_queue_ktype); |
1da177e4 | 560 | |
85e0cbbb | 561 | mutex_init(&q->debugfs_mutex); |
483f4afc | 562 | mutex_init(&q->sysfs_lock); |
cecf5d87 | 563 | mutex_init(&q->sysfs_dir_lock); |
0d945c1f | 564 | spin_lock_init(&q->queue_lock); |
c94a96ac | 565 | |
320ae51f | 566 | init_waitqueue_head(&q->mq_freeze_wq); |
7996a8b5 | 567 | mutex_init(&q->mq_freeze_lock); |
320ae51f | 568 | |
3ef28e83 DW |
569 | /* |
570 | * Init percpu_ref in atomic mode so that it's faster to shutdown. | |
571 | * See blk_register_queue() for details. | |
572 | */ | |
573 | if (percpu_ref_init(&q->q_usage_counter, | |
574 | blk_queue_usage_counter_release, | |
575 | PERCPU_REF_INIT_ATOMIC, GFP_KERNEL)) | |
fff4996b | 576 | goto fail_bdi; |
f51b802c | 577 | |
3ef28e83 DW |
578 | if (blkcg_init_queue(q)) |
579 | goto fail_ref; | |
580 | ||
3d745ea5 CH |
581 | blk_queue_dma_alignment(q, 511); |
582 | blk_set_default_limits(&q->limits); | |
c62b37d9 | 583 | q->nr_requests = BLKDEV_MAX_RQ; |
3d745ea5 | 584 | |
1da177e4 | 585 | return q; |
a73f730d | 586 | |
3ef28e83 DW |
587 | fail_ref: |
588 | percpu_ref_exit(&q->q_usage_counter); | |
fff4996b | 589 | fail_bdi: |
a83b576c JA |
590 | blk_free_queue_stats(q->stats); |
591 | fail_stats: | |
d03f6cdc | 592 | bdi_put(q->backing_dev_info); |
54efd50b | 593 | fail_split: |
338aa96d | 594 | bioset_exit(&q->bio_split); |
a73f730d TH |
595 | fail_id: |
596 | ida_simple_remove(&blk_queue_ida, q->id); | |
597 | fail_q: | |
598 | kmem_cache_free(blk_requestq_cachep, q); | |
599 | return NULL; | |
1da177e4 | 600 | } |
3d745ea5 | 601 | EXPORT_SYMBOL(blk_alloc_queue); |
1da177e4 | 602 | |
b5bd357c LC |
603 | /** |
604 | * blk_get_queue - increment the request_queue refcount | |
605 | * @q: the request_queue structure to increment the refcount for | |
606 | * | |
607 | * Increment the refcount of the request_queue kobject. | |
763b5892 LC |
608 | * |
609 | * Context: Any context. | |
b5bd357c | 610 | */ |
09ac46c4 | 611 | bool blk_get_queue(struct request_queue *q) |
1da177e4 | 612 | { |
3f3299d5 | 613 | if (likely(!blk_queue_dying(q))) { |
09ac46c4 TH |
614 | __blk_get_queue(q); |
615 | return true; | |
1da177e4 LT |
616 | } |
617 | ||
09ac46c4 | 618 | return false; |
1da177e4 | 619 | } |
d86e0e83 | 620 | EXPORT_SYMBOL(blk_get_queue); |
1da177e4 | 621 | |
a1ce35fa JA |
622 | /** |
623 | * blk_get_request - allocate a request | |
624 | * @q: request queue to allocate a request for | |
625 | * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC. | |
626 | * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT. | |
1da177e4 | 627 | */ |
a1ce35fa JA |
628 | struct request *blk_get_request(struct request_queue *q, unsigned int op, |
629 | blk_mq_req_flags_t flags) | |
1da177e4 | 630 | { |
a1ce35fa | 631 | struct request *req; |
1da177e4 | 632 | |
a1ce35fa | 633 | WARN_ON_ONCE(op & REQ_NOWAIT); |
a4d34da7 | 634 | WARN_ON_ONCE(flags & ~(BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_PM)); |
1da177e4 | 635 | |
a1ce35fa JA |
636 | req = blk_mq_alloc_request(q, op, flags); |
637 | if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn) | |
638 | q->mq_ops->initialize_rq_fn(req); | |
1da177e4 | 639 | |
a1ce35fa | 640 | return req; |
1da177e4 | 641 | } |
a1ce35fa | 642 | EXPORT_SYMBOL(blk_get_request); |
1da177e4 | 643 | |
1da177e4 LT |
644 | void blk_put_request(struct request *req) |
645 | { | |
a1ce35fa | 646 | blk_mq_free_request(req); |
1da177e4 | 647 | } |
1da177e4 LT |
648 | EXPORT_SYMBOL(blk_put_request); |
649 | ||
52c5e62d | 650 | static void handle_bad_sector(struct bio *bio, sector_t maxsector) |
1da177e4 LT |
651 | { |
652 | char b[BDEVNAME_SIZE]; | |
653 | ||
f4ac712e TH |
654 | pr_info_ratelimited("attempt to access beyond end of device\n" |
655 | "%s: rw=%d, want=%llu, limit=%llu\n", | |
656 | bio_devname(bio, b), bio->bi_opf, | |
657 | bio_end_sector(bio), maxsector); | |
1da177e4 LT |
658 | } |
659 | ||
c17bb495 AM |
660 | #ifdef CONFIG_FAIL_MAKE_REQUEST |
661 | ||
662 | static DECLARE_FAULT_ATTR(fail_make_request); | |
663 | ||
664 | static int __init setup_fail_make_request(char *str) | |
665 | { | |
666 | return setup_fault_attr(&fail_make_request, str); | |
667 | } | |
668 | __setup("fail_make_request=", setup_fail_make_request); | |
669 | ||
8446fe92 | 670 | static bool should_fail_request(struct block_device *part, unsigned int bytes) |
c17bb495 | 671 | { |
8446fe92 | 672 | return part->bd_make_it_fail && should_fail(&fail_make_request, bytes); |
c17bb495 AM |
673 | } |
674 | ||
675 | static int __init fail_make_request_debugfs(void) | |
676 | { | |
dd48c085 AM |
677 | struct dentry *dir = fault_create_debugfs_attr("fail_make_request", |
678 | NULL, &fail_make_request); | |
679 | ||
21f9fcd8 | 680 | return PTR_ERR_OR_ZERO(dir); |
c17bb495 AM |
681 | } |
682 | ||
683 | late_initcall(fail_make_request_debugfs); | |
684 | ||
685 | #else /* CONFIG_FAIL_MAKE_REQUEST */ | |
686 | ||
8446fe92 | 687 | static inline bool should_fail_request(struct block_device *part, |
b2c9cd37 | 688 | unsigned int bytes) |
c17bb495 | 689 | { |
b2c9cd37 | 690 | return false; |
c17bb495 AM |
691 | } |
692 | ||
693 | #endif /* CONFIG_FAIL_MAKE_REQUEST */ | |
694 | ||
2f9f6221 | 695 | static inline bool bio_check_ro(struct bio *bio) |
721c7fc7 | 696 | { |
2f9f6221 | 697 | if (op_is_write(bio_op(bio)) && bdev_read_only(bio->bi_bdev)) { |
721c7fc7 ID |
698 | char b[BDEVNAME_SIZE]; |
699 | ||
8b2ded1c MP |
700 | if (op_is_flush(bio->bi_opf) && !bio_sectors(bio)) |
701 | return false; | |
702 | ||
a32e236e | 703 | WARN_ONCE(1, |
c8178674 | 704 | "Trying to write to read-only block-device %s (partno %d)\n", |
2f9f6221 | 705 | bio_devname(bio, b), bio->bi_bdev->bd_partno); |
a32e236e LT |
706 | /* Older lvm-tools actually trigger this */ |
707 | return false; | |
721c7fc7 ID |
708 | } |
709 | ||
710 | return false; | |
711 | } | |
712 | ||
30abb3a6 HM |
713 | static noinline int should_fail_bio(struct bio *bio) |
714 | { | |
309dca30 | 715 | if (should_fail_request(bdev_whole(bio->bi_bdev), bio->bi_iter.bi_size)) |
30abb3a6 HM |
716 | return -EIO; |
717 | return 0; | |
718 | } | |
719 | ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO); | |
720 | ||
52c5e62d CH |
721 | /* |
722 | * Check whether this bio extends beyond the end of the device or partition. | |
723 | * This may well happen - the kernel calls bread() without checking the size of | |
724 | * the device, e.g., when mounting a file system. | |
725 | */ | |
2f9f6221 | 726 | static inline int bio_check_eod(struct bio *bio) |
52c5e62d | 727 | { |
2f9f6221 | 728 | sector_t maxsector = bdev_nr_sectors(bio->bi_bdev); |
52c5e62d CH |
729 | unsigned int nr_sectors = bio_sectors(bio); |
730 | ||
731 | if (nr_sectors && maxsector && | |
732 | (nr_sectors > maxsector || | |
733 | bio->bi_iter.bi_sector > maxsector - nr_sectors)) { | |
734 | handle_bad_sector(bio, maxsector); | |
735 | return -EIO; | |
736 | } | |
737 | return 0; | |
738 | } | |
739 | ||
74d46992 CH |
740 | /* |
741 | * Remap block n of partition p to block n+start(p) of the disk. | |
742 | */ | |
2f9f6221 | 743 | static int blk_partition_remap(struct bio *bio) |
74d46992 | 744 | { |
309dca30 | 745 | struct block_device *p = bio->bi_bdev; |
74d46992 | 746 | |
52c5e62d | 747 | if (unlikely(should_fail_request(p, bio->bi_iter.bi_size))) |
2f9f6221 | 748 | return -EIO; |
5eac3eb3 | 749 | if (bio_sectors(bio)) { |
8446fe92 | 750 | bio->bi_iter.bi_sector += p->bd_start_sect; |
1c02fca6 | 751 | trace_block_bio_remap(bio, p->bd_dev, |
29ff57c6 | 752 | bio->bi_iter.bi_sector - |
8446fe92 | 753 | p->bd_start_sect); |
52c5e62d | 754 | } |
309dca30 | 755 | bio->bi_bdev = bdev_whole(p); |
2f9f6221 | 756 | return 0; |
74d46992 CH |
757 | } |
758 | ||
0512a75b KB |
759 | /* |
760 | * Check write append to a zoned block device. | |
761 | */ | |
762 | static inline blk_status_t blk_check_zone_append(struct request_queue *q, | |
763 | struct bio *bio) | |
764 | { | |
765 | sector_t pos = bio->bi_iter.bi_sector; | |
766 | int nr_sectors = bio_sectors(bio); | |
767 | ||
768 | /* Only applicable to zoned block devices */ | |
769 | if (!blk_queue_is_zoned(q)) | |
770 | return BLK_STS_NOTSUPP; | |
771 | ||
772 | /* The bio sector must point to the start of a sequential zone */ | |
773 | if (pos & (blk_queue_zone_sectors(q) - 1) || | |
774 | !blk_queue_zone_is_seq(q, pos)) | |
775 | return BLK_STS_IOERR; | |
776 | ||
777 | /* | |
778 | * Not allowed to cross zone boundaries. Otherwise, the BIO will be | |
779 | * split and could result in non-contiguous sectors being written in | |
780 | * different zones. | |
781 | */ | |
782 | if (nr_sectors > q->limits.chunk_sectors) | |
783 | return BLK_STS_IOERR; | |
784 | ||
785 | /* Make sure the BIO is small enough and will not get split */ | |
786 | if (nr_sectors > q->limits.max_zone_append_sectors) | |
787 | return BLK_STS_IOERR; | |
788 | ||
789 | bio->bi_opf |= REQ_NOMERGE; | |
790 | ||
791 | return BLK_STS_OK; | |
792 | } | |
793 | ||
ed00aabd | 794 | static noinline_for_stack bool submit_bio_checks(struct bio *bio) |
1da177e4 | 795 | { |
309dca30 CH |
796 | struct block_device *bdev = bio->bi_bdev; |
797 | struct request_queue *q = bdev->bd_disk->queue; | |
4e4cbee9 | 798 | blk_status_t status = BLK_STS_IOERR; |
5a473e83 | 799 | struct blk_plug *plug; |
1da177e4 LT |
800 | |
801 | might_sleep(); | |
1da177e4 | 802 | |
5a473e83 JA |
803 | plug = blk_mq_plug(q, bio); |
804 | if (plug && plug->nowait) | |
805 | bio->bi_opf |= REQ_NOWAIT; | |
806 | ||
03a07c92 | 807 | /* |
b0beb280 | 808 | * For a REQ_NOWAIT based request, return -EOPNOTSUPP |
021a2446 | 809 | * if queue does not support NOWAIT. |
03a07c92 | 810 | */ |
021a2446 | 811 | if ((bio->bi_opf & REQ_NOWAIT) && !blk_queue_nowait(q)) |
b0beb280 | 812 | goto not_supported; |
03a07c92 | 813 | |
30abb3a6 | 814 | if (should_fail_bio(bio)) |
5a7bbad2 | 815 | goto end_io; |
2f9f6221 CH |
816 | if (unlikely(bio_check_ro(bio))) |
817 | goto end_io; | |
818 | if (unlikely(bio_check_eod(bio))) | |
819 | goto end_io; | |
820 | if (bio->bi_bdev->bd_partno && unlikely(blk_partition_remap(bio))) | |
821 | goto end_io; | |
2056a782 | 822 | |
5a7bbad2 | 823 | /* |
ed00aabd CH |
824 | * Filter flush bio's early so that bio based drivers without flush |
825 | * support don't have to worry about them. | |
5a7bbad2 | 826 | */ |
f3a8ab7d | 827 | if (op_is_flush(bio->bi_opf) && |
c888a8f9 | 828 | !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) { |
1eff9d32 | 829 | bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA); |
e439ab71 | 830 | if (!bio_sectors(bio)) { |
4e4cbee9 | 831 | status = BLK_STS_OK; |
51fd77bd JA |
832 | goto end_io; |
833 | } | |
5a7bbad2 | 834 | } |
5ddfe969 | 835 | |
d04c406f CH |
836 | if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags)) |
837 | bio->bi_opf &= ~REQ_HIPRI; | |
838 | ||
288dab8a CH |
839 | switch (bio_op(bio)) { |
840 | case REQ_OP_DISCARD: | |
841 | if (!blk_queue_discard(q)) | |
842 | goto not_supported; | |
843 | break; | |
844 | case REQ_OP_SECURE_ERASE: | |
845 | if (!blk_queue_secure_erase(q)) | |
846 | goto not_supported; | |
847 | break; | |
848 | case REQ_OP_WRITE_SAME: | |
74d46992 | 849 | if (!q->limits.max_write_same_sectors) |
288dab8a | 850 | goto not_supported; |
58886785 | 851 | break; |
0512a75b KB |
852 | case REQ_OP_ZONE_APPEND: |
853 | status = blk_check_zone_append(q, bio); | |
854 | if (status != BLK_STS_OK) | |
855 | goto end_io; | |
856 | break; | |
2d253440 | 857 | case REQ_OP_ZONE_RESET: |
6c1b1da5 AJ |
858 | case REQ_OP_ZONE_OPEN: |
859 | case REQ_OP_ZONE_CLOSE: | |
860 | case REQ_OP_ZONE_FINISH: | |
74d46992 | 861 | if (!blk_queue_is_zoned(q)) |
2d253440 | 862 | goto not_supported; |
288dab8a | 863 | break; |
6e33dbf2 CK |
864 | case REQ_OP_ZONE_RESET_ALL: |
865 | if (!blk_queue_is_zoned(q) || !blk_queue_zone_resetall(q)) | |
866 | goto not_supported; | |
867 | break; | |
a6f0788e | 868 | case REQ_OP_WRITE_ZEROES: |
74d46992 | 869 | if (!q->limits.max_write_zeroes_sectors) |
a6f0788e CK |
870 | goto not_supported; |
871 | break; | |
288dab8a CH |
872 | default: |
873 | break; | |
5a7bbad2 | 874 | } |
01edede4 | 875 | |
7f4b35d1 | 876 | /* |
3e82c348 CH |
877 | * Various block parts want %current->io_context, so allocate it up |
878 | * front rather than dealing with lots of pain to allocate it only | |
879 | * where needed. This may fail and the block layer knows how to live | |
880 | * with it. | |
7f4b35d1 | 881 | */ |
3e82c348 CH |
882 | if (unlikely(!current->io_context)) |
883 | create_task_io_context(current, GFP_ATOMIC, q->node); | |
7f4b35d1 | 884 | |
db18a53e CH |
885 | if (blk_throtl_bio(bio)) { |
886 | blkcg_bio_issue_init(bio); | |
ae118896 | 887 | return false; |
db18a53e CH |
888 | } |
889 | ||
890 | blk_cgroup_bio_start(bio); | |
891 | blkcg_bio_issue_init(bio); | |
27a84d54 | 892 | |
fbbaf700 | 893 | if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) { |
e8a676d6 | 894 | trace_block_bio_queue(bio); |
fbbaf700 N |
895 | /* Now that enqueuing has been traced, we need to trace |
896 | * completion as well. | |
897 | */ | |
898 | bio_set_flag(bio, BIO_TRACE_COMPLETION); | |
899 | } | |
27a84d54 | 900 | return true; |
a7384677 | 901 | |
288dab8a | 902 | not_supported: |
4e4cbee9 | 903 | status = BLK_STS_NOTSUPP; |
a7384677 | 904 | end_io: |
4e4cbee9 | 905 | bio->bi_status = status; |
4246a0b6 | 906 | bio_endio(bio); |
27a84d54 | 907 | return false; |
1da177e4 LT |
908 | } |
909 | ||
ed00aabd | 910 | static blk_qc_t __submit_bio(struct bio *bio) |
ac7c5675 | 911 | { |
309dca30 | 912 | struct gendisk *disk = bio->bi_bdev->bd_disk; |
ac7c5675 CH |
913 | blk_qc_t ret = BLK_QC_T_NONE; |
914 | ||
915 | if (blk_crypto_bio_prep(&bio)) { | |
c62b37d9 CH |
916 | if (!disk->fops->submit_bio) |
917 | return blk_mq_submit_bio(bio); | |
918 | ret = disk->fops->submit_bio(bio); | |
ac7c5675 | 919 | } |
c62b37d9 | 920 | blk_queue_exit(disk->queue); |
ac7c5675 CH |
921 | return ret; |
922 | } | |
923 | ||
566acf2d CH |
924 | /* |
925 | * The loop in this function may be a bit non-obvious, and so deserves some | |
926 | * explanation: | |
927 | * | |
928 | * - Before entering the loop, bio->bi_next is NULL (as all callers ensure | |
929 | * that), so we have a list with a single bio. | |
930 | * - We pretend that we have just taken it off a longer list, so we assign | |
931 | * bio_list to a pointer to the bio_list_on_stack, thus initialising the | |
932 | * bio_list of new bios to be added. ->submit_bio() may indeed add some more | |
933 | * bios through a recursive call to submit_bio_noacct. If it did, we find a | |
934 | * non-NULL value in bio_list and re-enter the loop from the top. | |
935 | * - In this case we really did just take the bio of the top of the list (no | |
936 | * pretending) and so remove it from bio_list, and call into ->submit_bio() | |
937 | * again. | |
938 | * | |
939 | * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio. | |
940 | * bio_list_on_stack[1] contains bios that were submitted before the current | |
941 | * ->submit_bio_bio, but that haven't been processed yet. | |
942 | */ | |
943 | static blk_qc_t __submit_bio_noacct(struct bio *bio) | |
944 | { | |
945 | struct bio_list bio_list_on_stack[2]; | |
946 | blk_qc_t ret = BLK_QC_T_NONE; | |
947 | ||
948 | BUG_ON(bio->bi_next); | |
949 | ||
950 | bio_list_init(&bio_list_on_stack[0]); | |
951 | current->bio_list = bio_list_on_stack; | |
952 | ||
953 | do { | |
309dca30 | 954 | struct request_queue *q = bio->bi_bdev->bd_disk->queue; |
566acf2d CH |
955 | struct bio_list lower, same; |
956 | ||
957 | if (unlikely(bio_queue_enter(bio) != 0)) | |
958 | continue; | |
959 | ||
960 | /* | |
961 | * Create a fresh bio_list for all subordinate requests. | |
962 | */ | |
963 | bio_list_on_stack[1] = bio_list_on_stack[0]; | |
964 | bio_list_init(&bio_list_on_stack[0]); | |
965 | ||
966 | ret = __submit_bio(bio); | |
967 | ||
968 | /* | |
969 | * Sort new bios into those for a lower level and those for the | |
970 | * same level. | |
971 | */ | |
972 | bio_list_init(&lower); | |
973 | bio_list_init(&same); | |
974 | while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL) | |
309dca30 | 975 | if (q == bio->bi_bdev->bd_disk->queue) |
566acf2d CH |
976 | bio_list_add(&same, bio); |
977 | else | |
978 | bio_list_add(&lower, bio); | |
979 | ||
980 | /* | |
981 | * Now assemble so we handle the lowest level first. | |
982 | */ | |
983 | bio_list_merge(&bio_list_on_stack[0], &lower); | |
984 | bio_list_merge(&bio_list_on_stack[0], &same); | |
985 | bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]); | |
986 | } while ((bio = bio_list_pop(&bio_list_on_stack[0]))); | |
987 | ||
988 | current->bio_list = NULL; | |
989 | return ret; | |
990 | } | |
991 | ||
ff93ea0c CH |
992 | static blk_qc_t __submit_bio_noacct_mq(struct bio *bio) |
993 | { | |
7c792f33 | 994 | struct bio_list bio_list[2] = { }; |
ff93ea0c CH |
995 | blk_qc_t ret = BLK_QC_T_NONE; |
996 | ||
7c792f33 | 997 | current->bio_list = bio_list; |
ff93ea0c CH |
998 | |
999 | do { | |
309dca30 | 1000 | struct gendisk *disk = bio->bi_bdev->bd_disk; |
ff93ea0c CH |
1001 | |
1002 | if (unlikely(bio_queue_enter(bio) != 0)) | |
1003 | continue; | |
1004 | ||
1005 | if (!blk_crypto_bio_prep(&bio)) { | |
1006 | blk_queue_exit(disk->queue); | |
1007 | ret = BLK_QC_T_NONE; | |
1008 | continue; | |
1009 | } | |
1010 | ||
1011 | ret = blk_mq_submit_bio(bio); | |
7c792f33 | 1012 | } while ((bio = bio_list_pop(&bio_list[0]))); |
ff93ea0c CH |
1013 | |
1014 | current->bio_list = NULL; | |
1015 | return ret; | |
1016 | } | |
1017 | ||
27a84d54 | 1018 | /** |
ed00aabd | 1019 | * submit_bio_noacct - re-submit a bio to the block device layer for I/O |
27a84d54 CH |
1020 | * @bio: The bio describing the location in memory and on the device. |
1021 | * | |
3fdd4086 CH |
1022 | * This is a version of submit_bio() that shall only be used for I/O that is |
1023 | * resubmitted to lower level drivers by stacking block drivers. All file | |
1024 | * systems and other upper level users of the block layer should use | |
1025 | * submit_bio() instead. | |
d89d8796 | 1026 | */ |
ed00aabd | 1027 | blk_qc_t submit_bio_noacct(struct bio *bio) |
d89d8796 | 1028 | { |
ed00aabd | 1029 | if (!submit_bio_checks(bio)) |
566acf2d | 1030 | return BLK_QC_T_NONE; |
27a84d54 CH |
1031 | |
1032 | /* | |
566acf2d CH |
1033 | * We only want one ->submit_bio to be active at a time, else stack |
1034 | * usage with stacked devices could be a problem. Use current->bio_list | |
1035 | * to collect a list of requests submited by a ->submit_bio method while | |
1036 | * it is active, and then process them after it returned. | |
27a84d54 | 1037 | */ |
bddd87c7 | 1038 | if (current->bio_list) { |
f5fe1b51 | 1039 | bio_list_add(¤t->bio_list[0], bio); |
566acf2d | 1040 | return BLK_QC_T_NONE; |
d89d8796 | 1041 | } |
27a84d54 | 1042 | |
309dca30 | 1043 | if (!bio->bi_bdev->bd_disk->fops->submit_bio) |
ff93ea0c | 1044 | return __submit_bio_noacct_mq(bio); |
566acf2d | 1045 | return __submit_bio_noacct(bio); |
d89d8796 | 1046 | } |
ed00aabd | 1047 | EXPORT_SYMBOL(submit_bio_noacct); |
1da177e4 LT |
1048 | |
1049 | /** | |
710027a4 | 1050 | * submit_bio - submit a bio to the block device layer for I/O |
1da177e4 LT |
1051 | * @bio: The &struct bio which describes the I/O |
1052 | * | |
3fdd4086 CH |
1053 | * submit_bio() is used to submit I/O requests to block devices. It is passed a |
1054 | * fully set up &struct bio that describes the I/O that needs to be done. The | |
309dca30 | 1055 | * bio will be send to the device described by the bi_bdev field. |
1da177e4 | 1056 | * |
3fdd4086 CH |
1057 | * The success/failure status of the request, along with notification of |
1058 | * completion, is delivered asynchronously through the ->bi_end_io() callback | |
1059 | * in @bio. The bio must NOT be touched by thecaller until ->bi_end_io() has | |
1060 | * been called. | |
1da177e4 | 1061 | */ |
4e49ea4a | 1062 | blk_qc_t submit_bio(struct bio *bio) |
1da177e4 | 1063 | { |
d3f77dfd TH |
1064 | if (blkcg_punt_bio_submit(bio)) |
1065 | return BLK_QC_T_NONE; | |
1066 | ||
bf2de6f5 JA |
1067 | /* |
1068 | * If it's a regular read/write or a barrier with data attached, | |
1069 | * go through the normal accounting stuff before submission. | |
1070 | */ | |
e2a60da7 | 1071 | if (bio_has_data(bio)) { |
4363ac7c MP |
1072 | unsigned int count; |
1073 | ||
95fe6c1a | 1074 | if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME)) |
309dca30 CH |
1075 | count = queue_logical_block_size( |
1076 | bio->bi_bdev->bd_disk->queue) >> 9; | |
4363ac7c MP |
1077 | else |
1078 | count = bio_sectors(bio); | |
1079 | ||
a8ebb056 | 1080 | if (op_is_write(bio_op(bio))) { |
bf2de6f5 JA |
1081 | count_vm_events(PGPGOUT, count); |
1082 | } else { | |
4f024f37 | 1083 | task_io_account_read(bio->bi_iter.bi_size); |
bf2de6f5 JA |
1084 | count_vm_events(PGPGIN, count); |
1085 | } | |
1086 | ||
1087 | if (unlikely(block_dump)) { | |
1088 | char b[BDEVNAME_SIZE]; | |
8dcbdc74 | 1089 | printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n", |
ba25f9dc | 1090 | current->comm, task_pid_nr(current), |
a8ebb056 | 1091 | op_is_write(bio_op(bio)) ? "WRITE" : "READ", |
4f024f37 | 1092 | (unsigned long long)bio->bi_iter.bi_sector, |
74d46992 | 1093 | bio_devname(bio, b), count); |
bf2de6f5 | 1094 | } |
1da177e4 LT |
1095 | } |
1096 | ||
b8e24a93 | 1097 | /* |
760f83ea CH |
1098 | * If we're reading data that is part of the userspace workingset, count |
1099 | * submission time as memory stall. When the device is congested, or | |
1100 | * the submitting cgroup IO-throttled, submission can be a significant | |
1101 | * part of overall IO time. | |
b8e24a93 | 1102 | */ |
760f83ea CH |
1103 | if (unlikely(bio_op(bio) == REQ_OP_READ && |
1104 | bio_flagged(bio, BIO_WORKINGSET))) { | |
1105 | unsigned long pflags; | |
1106 | blk_qc_t ret; | |
b8e24a93 | 1107 | |
760f83ea | 1108 | psi_memstall_enter(&pflags); |
ed00aabd | 1109 | ret = submit_bio_noacct(bio); |
b8e24a93 JW |
1110 | psi_memstall_leave(&pflags); |
1111 | ||
760f83ea CH |
1112 | return ret; |
1113 | } | |
1114 | ||
ed00aabd | 1115 | return submit_bio_noacct(bio); |
1da177e4 | 1116 | } |
1da177e4 LT |
1117 | EXPORT_SYMBOL(submit_bio); |
1118 | ||
82124d60 | 1119 | /** |
bf4e6b4e | 1120 | * blk_cloned_rq_check_limits - Helper function to check a cloned request |
0d720318 | 1121 | * for the new queue limits |
82124d60 KU |
1122 | * @q: the queue |
1123 | * @rq: the request being checked | |
1124 | * | |
1125 | * Description: | |
1126 | * @rq may have been made based on weaker limitations of upper-level queues | |
1127 | * in request stacking drivers, and it may violate the limitation of @q. | |
1128 | * Since the block layer and the underlying device driver trust @rq | |
1129 | * after it is inserted to @q, it should be checked against @q before | |
1130 | * the insertion using this generic function. | |
1131 | * | |
82124d60 | 1132 | * Request stacking drivers like request-based dm may change the queue |
bf4e6b4e HR |
1133 | * limits when retrying requests on other queues. Those requests need |
1134 | * to be checked against the new queue limits again during dispatch. | |
82124d60 | 1135 | */ |
143d2600 | 1136 | static blk_status_t blk_cloned_rq_check_limits(struct request_queue *q, |
bf4e6b4e | 1137 | struct request *rq) |
82124d60 | 1138 | { |
8327cce5 RS |
1139 | unsigned int max_sectors = blk_queue_get_max_sectors(q, req_op(rq)); |
1140 | ||
1141 | if (blk_rq_sectors(rq) > max_sectors) { | |
1142 | /* | |
1143 | * SCSI device does not have a good way to return if | |
1144 | * Write Same/Zero is actually supported. If a device rejects | |
1145 | * a non-read/write command (discard, write same,etc.) the | |
1146 | * low-level device driver will set the relevant queue limit to | |
1147 | * 0 to prevent blk-lib from issuing more of the offending | |
1148 | * operations. Commands queued prior to the queue limit being | |
1149 | * reset need to be completed with BLK_STS_NOTSUPP to avoid I/O | |
1150 | * errors being propagated to upper layers. | |
1151 | */ | |
1152 | if (max_sectors == 0) | |
1153 | return BLK_STS_NOTSUPP; | |
1154 | ||
61939b12 | 1155 | printk(KERN_ERR "%s: over max size limit. (%u > %u)\n", |
8327cce5 | 1156 | __func__, blk_rq_sectors(rq), max_sectors); |
143d2600 | 1157 | return BLK_STS_IOERR; |
82124d60 KU |
1158 | } |
1159 | ||
1160 | /* | |
1161 | * queue's settings related to segment counting like q->bounce_pfn | |
1162 | * may differ from that of other stacking queues. | |
1163 | * Recalculate it to check the request correctly on this queue's | |
1164 | * limitation. | |
1165 | */ | |
e9cd19c0 | 1166 | rq->nr_phys_segments = blk_recalc_rq_segments(rq); |
8a78362c | 1167 | if (rq->nr_phys_segments > queue_max_segments(q)) { |
61939b12 JP |
1168 | printk(KERN_ERR "%s: over max segments limit. (%hu > %hu)\n", |
1169 | __func__, rq->nr_phys_segments, queue_max_segments(q)); | |
143d2600 | 1170 | return BLK_STS_IOERR; |
82124d60 KU |
1171 | } |
1172 | ||
143d2600 | 1173 | return BLK_STS_OK; |
82124d60 | 1174 | } |
82124d60 KU |
1175 | |
1176 | /** | |
1177 | * blk_insert_cloned_request - Helper for stacking drivers to submit a request | |
1178 | * @q: the queue to submit the request | |
1179 | * @rq: the request being queued | |
1180 | */ | |
2a842aca | 1181 | blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq) |
82124d60 | 1182 | { |
8327cce5 RS |
1183 | blk_status_t ret; |
1184 | ||
1185 | ret = blk_cloned_rq_check_limits(q, rq); | |
1186 | if (ret != BLK_STS_OK) | |
1187 | return ret; | |
82124d60 | 1188 | |
b2c9cd37 | 1189 | if (rq->rq_disk && |
8446fe92 | 1190 | should_fail_request(rq->rq_disk->part0, blk_rq_bytes(rq))) |
2a842aca | 1191 | return BLK_STS_IOERR; |
82124d60 | 1192 | |
a892c8d5 ST |
1193 | if (blk_crypto_insert_cloned_request(rq)) |
1194 | return BLK_STS_IOERR; | |
1195 | ||
a1ce35fa | 1196 | if (blk_queue_io_stat(q)) |
b5af37ab | 1197 | blk_account_io_start(rq); |
82124d60 KU |
1198 | |
1199 | /* | |
a1ce35fa JA |
1200 | * Since we have a scheduler attached on the top device, |
1201 | * bypass a potential scheduler on the bottom device for | |
1202 | * insert. | |
82124d60 | 1203 | */ |
fd9c40f6 | 1204 | return blk_mq_request_issue_directly(rq, true); |
82124d60 KU |
1205 | } |
1206 | EXPORT_SYMBOL_GPL(blk_insert_cloned_request); | |
1207 | ||
80a761fd TH |
1208 | /** |
1209 | * blk_rq_err_bytes - determine number of bytes till the next failure boundary | |
1210 | * @rq: request to examine | |
1211 | * | |
1212 | * Description: | |
1213 | * A request could be merge of IOs which require different failure | |
1214 | * handling. This function determines the number of bytes which | |
1215 | * can be failed from the beginning of the request without | |
1216 | * crossing into area which need to be retried further. | |
1217 | * | |
1218 | * Return: | |
1219 | * The number of bytes to fail. | |
80a761fd TH |
1220 | */ |
1221 | unsigned int blk_rq_err_bytes(const struct request *rq) | |
1222 | { | |
1223 | unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK; | |
1224 | unsigned int bytes = 0; | |
1225 | struct bio *bio; | |
1226 | ||
e8064021 | 1227 | if (!(rq->rq_flags & RQF_MIXED_MERGE)) |
80a761fd TH |
1228 | return blk_rq_bytes(rq); |
1229 | ||
1230 | /* | |
1231 | * Currently the only 'mixing' which can happen is between | |
1232 | * different fastfail types. We can safely fail portions | |
1233 | * which have all the failfast bits that the first one has - | |
1234 | * the ones which are at least as eager to fail as the first | |
1235 | * one. | |
1236 | */ | |
1237 | for (bio = rq->bio; bio; bio = bio->bi_next) { | |
1eff9d32 | 1238 | if ((bio->bi_opf & ff) != ff) |
80a761fd | 1239 | break; |
4f024f37 | 1240 | bytes += bio->bi_iter.bi_size; |
80a761fd TH |
1241 | } |
1242 | ||
1243 | /* this could lead to infinite loop */ | |
1244 | BUG_ON(blk_rq_bytes(rq) && !bytes); | |
1245 | return bytes; | |
1246 | } | |
1247 | EXPORT_SYMBOL_GPL(blk_rq_err_bytes); | |
1248 | ||
8446fe92 CH |
1249 | static void update_io_ticks(struct block_device *part, unsigned long now, |
1250 | bool end) | |
9123bf6f CH |
1251 | { |
1252 | unsigned long stamp; | |
1253 | again: | |
8446fe92 | 1254 | stamp = READ_ONCE(part->bd_stamp); |
9123bf6f | 1255 | if (unlikely(stamp != now)) { |
8446fe92 | 1256 | if (likely(cmpxchg(&part->bd_stamp, stamp, now) == stamp)) |
9123bf6f CH |
1257 | __part_stat_add(part, io_ticks, end ? now - stamp : 1); |
1258 | } | |
8446fe92 CH |
1259 | if (part->bd_partno) { |
1260 | part = bdev_whole(part); | |
9123bf6f CH |
1261 | goto again; |
1262 | } | |
1263 | } | |
1264 | ||
f1394b79 | 1265 | static void blk_account_io_completion(struct request *req, unsigned int bytes) |
bc58ba94 | 1266 | { |
ecb6186c | 1267 | if (req->part && blk_do_io_stat(req)) { |
ddcf35d3 | 1268 | const int sgrp = op_stat_group(req_op(req)); |
bc58ba94 | 1269 | |
112f158f | 1270 | part_stat_lock(); |
8446fe92 | 1271 | part_stat_add(req->part, sectors[sgrp], bytes >> 9); |
bc58ba94 JA |
1272 | part_stat_unlock(); |
1273 | } | |
1274 | } | |
1275 | ||
522a7775 | 1276 | void blk_account_io_done(struct request *req, u64 now) |
bc58ba94 | 1277 | { |
bc58ba94 | 1278 | /* |
dd4c133f TH |
1279 | * Account IO completion. flush_rq isn't accounted as a |
1280 | * normal IO on queueing nor completion. Accounting the | |
1281 | * containing request is enough. | |
bc58ba94 | 1282 | */ |
ecb6186c LG |
1283 | if (req->part && blk_do_io_stat(req) && |
1284 | !(req->rq_flags & RQF_FLUSH_SEQ)) { | |
ddcf35d3 | 1285 | const int sgrp = op_stat_group(req_op(req)); |
bc58ba94 | 1286 | |
112f158f | 1287 | part_stat_lock(); |
8446fe92 CH |
1288 | update_io_ticks(req->part, jiffies, true); |
1289 | part_stat_inc(req->part, ios[sgrp]); | |
1290 | part_stat_add(req->part, nsecs[sgrp], now - req->start_time_ns); | |
524f9ffd | 1291 | part_stat_unlock(); |
bc58ba94 JA |
1292 | } |
1293 | } | |
1294 | ||
b5af37ab | 1295 | void blk_account_io_start(struct request *rq) |
320ae51f | 1296 | { |
320ae51f JA |
1297 | if (!blk_do_io_stat(rq)) |
1298 | return; | |
1299 | ||
b5af37ab | 1300 | rq->part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq)); |
524f9ffd | 1301 | |
112f158f | 1302 | part_stat_lock(); |
76268f3a | 1303 | update_io_ticks(rq->part, jiffies, false); |
320ae51f JA |
1304 | part_stat_unlock(); |
1305 | } | |
320ae51f | 1306 | |
8446fe92 | 1307 | static unsigned long __part_start_io_acct(struct block_device *part, |
7b26410b | 1308 | unsigned int sectors, unsigned int op) |
956d510e | 1309 | { |
956d510e CH |
1310 | const int sgrp = op_stat_group(op); |
1311 | unsigned long now = READ_ONCE(jiffies); | |
1312 | ||
1313 | part_stat_lock(); | |
1314 | update_io_ticks(part, now, false); | |
1315 | part_stat_inc(part, ios[sgrp]); | |
1316 | part_stat_add(part, sectors[sgrp], sectors); | |
1317 | part_stat_local_inc(part, in_flight[op_is_write(op)]); | |
1318 | part_stat_unlock(); | |
320ae51f | 1319 | |
956d510e CH |
1320 | return now; |
1321 | } | |
7b26410b | 1322 | |
8446fe92 | 1323 | unsigned long part_start_io_acct(struct gendisk *disk, struct block_device **part, |
7b26410b SL |
1324 | struct bio *bio) |
1325 | { | |
1326 | *part = disk_map_sector_rcu(disk, bio->bi_iter.bi_sector); | |
1327 | ||
1328 | return __part_start_io_acct(*part, bio_sectors(bio), bio_op(bio)); | |
1329 | } | |
1330 | EXPORT_SYMBOL_GPL(part_start_io_acct); | |
1331 | ||
1332 | unsigned long disk_start_io_acct(struct gendisk *disk, unsigned int sectors, | |
1333 | unsigned int op) | |
1334 | { | |
8446fe92 | 1335 | return __part_start_io_acct(disk->part0, sectors, op); |
7b26410b | 1336 | } |
956d510e CH |
1337 | EXPORT_SYMBOL(disk_start_io_acct); |
1338 | ||
8446fe92 | 1339 | static void __part_end_io_acct(struct block_device *part, unsigned int op, |
7b26410b | 1340 | unsigned long start_time) |
956d510e | 1341 | { |
956d510e CH |
1342 | const int sgrp = op_stat_group(op); |
1343 | unsigned long now = READ_ONCE(jiffies); | |
1344 | unsigned long duration = now - start_time; | |
5b18b5a7 | 1345 | |
956d510e CH |
1346 | part_stat_lock(); |
1347 | update_io_ticks(part, now, true); | |
1348 | part_stat_add(part, nsecs[sgrp], jiffies_to_nsecs(duration)); | |
1349 | part_stat_local_dec(part, in_flight[op_is_write(op)]); | |
320ae51f JA |
1350 | part_stat_unlock(); |
1351 | } | |
7b26410b | 1352 | |
8446fe92 | 1353 | void part_end_io_acct(struct block_device *part, struct bio *bio, |
7b26410b SL |
1354 | unsigned long start_time) |
1355 | { | |
1356 | __part_end_io_acct(part, bio_op(bio), start_time); | |
7b26410b SL |
1357 | } |
1358 | EXPORT_SYMBOL_GPL(part_end_io_acct); | |
1359 | ||
1360 | void disk_end_io_acct(struct gendisk *disk, unsigned int op, | |
1361 | unsigned long start_time) | |
1362 | { | |
8446fe92 | 1363 | __part_end_io_acct(disk->part0, op, start_time); |
7b26410b | 1364 | } |
956d510e | 1365 | EXPORT_SYMBOL(disk_end_io_acct); |
320ae51f | 1366 | |
ef71de8b CH |
1367 | /* |
1368 | * Steal bios from a request and add them to a bio list. | |
1369 | * The request must not have been partially completed before. | |
1370 | */ | |
1371 | void blk_steal_bios(struct bio_list *list, struct request *rq) | |
1372 | { | |
1373 | if (rq->bio) { | |
1374 | if (list->tail) | |
1375 | list->tail->bi_next = rq->bio; | |
1376 | else | |
1377 | list->head = rq->bio; | |
1378 | list->tail = rq->biotail; | |
1379 | ||
1380 | rq->bio = NULL; | |
1381 | rq->biotail = NULL; | |
1382 | } | |
1383 | ||
1384 | rq->__data_len = 0; | |
1385 | } | |
1386 | EXPORT_SYMBOL_GPL(blk_steal_bios); | |
1387 | ||
3bcddeac | 1388 | /** |
2e60e022 | 1389 | * blk_update_request - Special helper function for request stacking drivers |
8ebf9756 | 1390 | * @req: the request being processed |
2a842aca | 1391 | * @error: block status code |
8ebf9756 | 1392 | * @nr_bytes: number of bytes to complete @req |
3bcddeac KU |
1393 | * |
1394 | * Description: | |
8ebf9756 RD |
1395 | * Ends I/O on a number of bytes attached to @req, but doesn't complete |
1396 | * the request structure even if @req doesn't have leftover. | |
1397 | * If @req has leftover, sets it up for the next range of segments. | |
2e60e022 TH |
1398 | * |
1399 | * This special helper function is only for request stacking drivers | |
1400 | * (e.g. request-based dm) so that they can handle partial completion. | |
3a211b71 | 1401 | * Actual device drivers should use blk_mq_end_request instead. |
2e60e022 TH |
1402 | * |
1403 | * Passing the result of blk_rq_bytes() as @nr_bytes guarantees | |
1404 | * %false return from this function. | |
3bcddeac | 1405 | * |
1954e9a9 BVA |
1406 | * Note: |
1407 | * The RQF_SPECIAL_PAYLOAD flag is ignored on purpose in both | |
1408 | * blk_rq_bytes() and in blk_update_request(). | |
1409 | * | |
3bcddeac | 1410 | * Return: |
2e60e022 TH |
1411 | * %false - this request doesn't have any more data |
1412 | * %true - this request has more data | |
3bcddeac | 1413 | **/ |
2a842aca CH |
1414 | bool blk_update_request(struct request *req, blk_status_t error, |
1415 | unsigned int nr_bytes) | |
1da177e4 | 1416 | { |
f79ea416 | 1417 | int total_bytes; |
1da177e4 | 1418 | |
2a842aca | 1419 | trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes); |
4a0efdc9 | 1420 | |
2e60e022 TH |
1421 | if (!req->bio) |
1422 | return false; | |
1423 | ||
54d4e6ab MG |
1424 | #ifdef CONFIG_BLK_DEV_INTEGRITY |
1425 | if (blk_integrity_rq(req) && req_op(req) == REQ_OP_READ && | |
1426 | error == BLK_STS_OK) | |
1427 | req->q->integrity.profile->complete_fn(req, nr_bytes); | |
1428 | #endif | |
1429 | ||
2a842aca CH |
1430 | if (unlikely(error && !blk_rq_is_passthrough(req) && |
1431 | !(req->rq_flags & RQF_QUIET))) | |
178cc590 | 1432 | print_req_error(req, error, __func__); |
1da177e4 | 1433 | |
bc58ba94 | 1434 | blk_account_io_completion(req, nr_bytes); |
d72d904a | 1435 | |
f79ea416 KO |
1436 | total_bytes = 0; |
1437 | while (req->bio) { | |
1438 | struct bio *bio = req->bio; | |
4f024f37 | 1439 | unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes); |
1da177e4 | 1440 | |
9c24c10a | 1441 | if (bio_bytes == bio->bi_iter.bi_size) |
1da177e4 | 1442 | req->bio = bio->bi_next; |
1da177e4 | 1443 | |
fbbaf700 N |
1444 | /* Completion has already been traced */ |
1445 | bio_clear_flag(bio, BIO_TRACE_COMPLETION); | |
f79ea416 | 1446 | req_bio_endio(req, bio, bio_bytes, error); |
1da177e4 | 1447 | |
f79ea416 KO |
1448 | total_bytes += bio_bytes; |
1449 | nr_bytes -= bio_bytes; | |
1da177e4 | 1450 | |
f79ea416 KO |
1451 | if (!nr_bytes) |
1452 | break; | |
1da177e4 LT |
1453 | } |
1454 | ||
1455 | /* | |
1456 | * completely done | |
1457 | */ | |
2e60e022 TH |
1458 | if (!req->bio) { |
1459 | /* | |
1460 | * Reset counters so that the request stacking driver | |
1461 | * can find how many bytes remain in the request | |
1462 | * later. | |
1463 | */ | |
a2dec7b3 | 1464 | req->__data_len = 0; |
2e60e022 TH |
1465 | return false; |
1466 | } | |
1da177e4 | 1467 | |
a2dec7b3 | 1468 | req->__data_len -= total_bytes; |
2e46e8b2 TH |
1469 | |
1470 | /* update sector only for requests with clear definition of sector */ | |
57292b58 | 1471 | if (!blk_rq_is_passthrough(req)) |
a2dec7b3 | 1472 | req->__sector += total_bytes >> 9; |
2e46e8b2 | 1473 | |
80a761fd | 1474 | /* mixed attributes always follow the first bio */ |
e8064021 | 1475 | if (req->rq_flags & RQF_MIXED_MERGE) { |
80a761fd | 1476 | req->cmd_flags &= ~REQ_FAILFAST_MASK; |
1eff9d32 | 1477 | req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK; |
80a761fd TH |
1478 | } |
1479 | ||
ed6565e7 CH |
1480 | if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) { |
1481 | /* | |
1482 | * If total number of sectors is less than the first segment | |
1483 | * size, something has gone terribly wrong. | |
1484 | */ | |
1485 | if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) { | |
1486 | blk_dump_rq_flags(req, "request botched"); | |
1487 | req->__data_len = blk_rq_cur_bytes(req); | |
1488 | } | |
2e46e8b2 | 1489 | |
ed6565e7 | 1490 | /* recalculate the number of segments */ |
e9cd19c0 | 1491 | req->nr_phys_segments = blk_recalc_rq_segments(req); |
ed6565e7 | 1492 | } |
2e46e8b2 | 1493 | |
2e60e022 | 1494 | return true; |
1da177e4 | 1495 | } |
2e60e022 | 1496 | EXPORT_SYMBOL_GPL(blk_update_request); |
1da177e4 | 1497 | |
2d4dc890 IL |
1498 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE |
1499 | /** | |
1500 | * rq_flush_dcache_pages - Helper function to flush all pages in a request | |
1501 | * @rq: the request to be flushed | |
1502 | * | |
1503 | * Description: | |
1504 | * Flush all pages in @rq. | |
1505 | */ | |
1506 | void rq_flush_dcache_pages(struct request *rq) | |
1507 | { | |
1508 | struct req_iterator iter; | |
7988613b | 1509 | struct bio_vec bvec; |
2d4dc890 IL |
1510 | |
1511 | rq_for_each_segment(bvec, rq, iter) | |
7988613b | 1512 | flush_dcache_page(bvec.bv_page); |
2d4dc890 IL |
1513 | } |
1514 | EXPORT_SYMBOL_GPL(rq_flush_dcache_pages); | |
1515 | #endif | |
1516 | ||
ef9e3fac KU |
1517 | /** |
1518 | * blk_lld_busy - Check if underlying low-level drivers of a device are busy | |
1519 | * @q : the queue of the device being checked | |
1520 | * | |
1521 | * Description: | |
1522 | * Check if underlying low-level drivers of a device are busy. | |
1523 | * If the drivers want to export their busy state, they must set own | |
1524 | * exporting function using blk_queue_lld_busy() first. | |
1525 | * | |
1526 | * Basically, this function is used only by request stacking drivers | |
1527 | * to stop dispatching requests to underlying devices when underlying | |
1528 | * devices are busy. This behavior helps more I/O merging on the queue | |
1529 | * of the request stacking driver and prevents I/O throughput regression | |
1530 | * on burst I/O load. | |
1531 | * | |
1532 | * Return: | |
1533 | * 0 - Not busy (The request stacking driver should dispatch request) | |
1534 | * 1 - Busy (The request stacking driver should stop dispatching request) | |
1535 | */ | |
1536 | int blk_lld_busy(struct request_queue *q) | |
1537 | { | |
344e9ffc | 1538 | if (queue_is_mq(q) && q->mq_ops->busy) |
9ba20527 | 1539 | return q->mq_ops->busy(q); |
ef9e3fac KU |
1540 | |
1541 | return 0; | |
1542 | } | |
1543 | EXPORT_SYMBOL_GPL(blk_lld_busy); | |
1544 | ||
78d8e58a MS |
1545 | /** |
1546 | * blk_rq_unprep_clone - Helper function to free all bios in a cloned request | |
1547 | * @rq: the clone request to be cleaned up | |
1548 | * | |
1549 | * Description: | |
1550 | * Free all bios in @rq for a cloned request. | |
1551 | */ | |
1552 | void blk_rq_unprep_clone(struct request *rq) | |
1553 | { | |
1554 | struct bio *bio; | |
1555 | ||
1556 | while ((bio = rq->bio) != NULL) { | |
1557 | rq->bio = bio->bi_next; | |
1558 | ||
1559 | bio_put(bio); | |
1560 | } | |
1561 | } | |
1562 | EXPORT_SYMBOL_GPL(blk_rq_unprep_clone); | |
1563 | ||
78d8e58a MS |
1564 | /** |
1565 | * blk_rq_prep_clone - Helper function to setup clone request | |
1566 | * @rq: the request to be setup | |
1567 | * @rq_src: original request to be cloned | |
1568 | * @bs: bio_set that bios for clone are allocated from | |
1569 | * @gfp_mask: memory allocation mask for bio | |
1570 | * @bio_ctr: setup function to be called for each clone bio. | |
1571 | * Returns %0 for success, non %0 for failure. | |
1572 | * @data: private data to be passed to @bio_ctr | |
1573 | * | |
1574 | * Description: | |
1575 | * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq. | |
78d8e58a MS |
1576 | * Also, pages which the original bios are pointing to are not copied |
1577 | * and the cloned bios just point same pages. | |
1578 | * So cloned bios must be completed before original bios, which means | |
1579 | * the caller must complete @rq before @rq_src. | |
1580 | */ | |
1581 | int blk_rq_prep_clone(struct request *rq, struct request *rq_src, | |
1582 | struct bio_set *bs, gfp_t gfp_mask, | |
1583 | int (*bio_ctr)(struct bio *, struct bio *, void *), | |
1584 | void *data) | |
1585 | { | |
1586 | struct bio *bio, *bio_src; | |
1587 | ||
1588 | if (!bs) | |
f4f8154a | 1589 | bs = &fs_bio_set; |
78d8e58a MS |
1590 | |
1591 | __rq_for_each_bio(bio_src, rq_src) { | |
1592 | bio = bio_clone_fast(bio_src, gfp_mask, bs); | |
1593 | if (!bio) | |
1594 | goto free_and_out; | |
1595 | ||
1596 | if (bio_ctr && bio_ctr(bio, bio_src, data)) | |
1597 | goto free_and_out; | |
1598 | ||
1599 | if (rq->bio) { | |
1600 | rq->biotail->bi_next = bio; | |
1601 | rq->biotail = bio; | |
93f221ae | 1602 | } else { |
78d8e58a | 1603 | rq->bio = rq->biotail = bio; |
93f221ae EB |
1604 | } |
1605 | bio = NULL; | |
78d8e58a MS |
1606 | } |
1607 | ||
361301a2 GJ |
1608 | /* Copy attributes of the original request to the clone request. */ |
1609 | rq->__sector = blk_rq_pos(rq_src); | |
1610 | rq->__data_len = blk_rq_bytes(rq_src); | |
1611 | if (rq_src->rq_flags & RQF_SPECIAL_PAYLOAD) { | |
1612 | rq->rq_flags |= RQF_SPECIAL_PAYLOAD; | |
1613 | rq->special_vec = rq_src->special_vec; | |
1614 | } | |
1615 | rq->nr_phys_segments = rq_src->nr_phys_segments; | |
1616 | rq->ioprio = rq_src->ioprio; | |
78d8e58a | 1617 | |
93f221ae EB |
1618 | if (rq->bio && blk_crypto_rq_bio_prep(rq, rq->bio, gfp_mask) < 0) |
1619 | goto free_and_out; | |
78d8e58a MS |
1620 | |
1621 | return 0; | |
1622 | ||
1623 | free_and_out: | |
1624 | if (bio) | |
1625 | bio_put(bio); | |
1626 | blk_rq_unprep_clone(rq); | |
1627 | ||
1628 | return -ENOMEM; | |
b0fd271d KU |
1629 | } |
1630 | EXPORT_SYMBOL_GPL(blk_rq_prep_clone); | |
1631 | ||
59c3d45e | 1632 | int kblockd_schedule_work(struct work_struct *work) |
1da177e4 LT |
1633 | { |
1634 | return queue_work(kblockd_workqueue, work); | |
1635 | } | |
1da177e4 LT |
1636 | EXPORT_SYMBOL(kblockd_schedule_work); |
1637 | ||
818cd1cb JA |
1638 | int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork, |
1639 | unsigned long delay) | |
1640 | { | |
1641 | return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay); | |
1642 | } | |
1643 | EXPORT_SYMBOL(kblockd_mod_delayed_work_on); | |
1644 | ||
75df7136 SJ |
1645 | /** |
1646 | * blk_start_plug - initialize blk_plug and track it inside the task_struct | |
1647 | * @plug: The &struct blk_plug that needs to be initialized | |
1648 | * | |
1649 | * Description: | |
40405851 JM |
1650 | * blk_start_plug() indicates to the block layer an intent by the caller |
1651 | * to submit multiple I/O requests in a batch. The block layer may use | |
1652 | * this hint to defer submitting I/Os from the caller until blk_finish_plug() | |
1653 | * is called. However, the block layer may choose to submit requests | |
1654 | * before a call to blk_finish_plug() if the number of queued I/Os | |
1655 | * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than | |
1656 | * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if | |
1657 | * the task schedules (see below). | |
1658 | * | |
75df7136 SJ |
1659 | * Tracking blk_plug inside the task_struct will help with auto-flushing the |
1660 | * pending I/O should the task end up blocking between blk_start_plug() and | |
1661 | * blk_finish_plug(). This is important from a performance perspective, but | |
1662 | * also ensures that we don't deadlock. For instance, if the task is blocking | |
1663 | * for a memory allocation, memory reclaim could end up wanting to free a | |
1664 | * page belonging to that request that is currently residing in our private | |
1665 | * plug. By flushing the pending I/O when the process goes to sleep, we avoid | |
1666 | * this kind of deadlock. | |
1667 | */ | |
73c10101 JA |
1668 | void blk_start_plug(struct blk_plug *plug) |
1669 | { | |
1670 | struct task_struct *tsk = current; | |
1671 | ||
dd6cf3e1 SL |
1672 | /* |
1673 | * If this is a nested plug, don't actually assign it. | |
1674 | */ | |
1675 | if (tsk->plug) | |
1676 | return; | |
1677 | ||
320ae51f | 1678 | INIT_LIST_HEAD(&plug->mq_list); |
048c9374 | 1679 | INIT_LIST_HEAD(&plug->cb_list); |
5f0ed774 | 1680 | plug->rq_count = 0; |
ce5b009c | 1681 | plug->multiple_queues = false; |
5a473e83 | 1682 | plug->nowait = false; |
5f0ed774 | 1683 | |
73c10101 | 1684 | /* |
dd6cf3e1 SL |
1685 | * Store ordering should not be needed here, since a potential |
1686 | * preempt will imply a full memory barrier | |
73c10101 | 1687 | */ |
dd6cf3e1 | 1688 | tsk->plug = plug; |
73c10101 JA |
1689 | } |
1690 | EXPORT_SYMBOL(blk_start_plug); | |
1691 | ||
74018dc3 | 1692 | static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule) |
048c9374 N |
1693 | { |
1694 | LIST_HEAD(callbacks); | |
1695 | ||
2a7d5559 SL |
1696 | while (!list_empty(&plug->cb_list)) { |
1697 | list_splice_init(&plug->cb_list, &callbacks); | |
048c9374 | 1698 | |
2a7d5559 SL |
1699 | while (!list_empty(&callbacks)) { |
1700 | struct blk_plug_cb *cb = list_first_entry(&callbacks, | |
048c9374 N |
1701 | struct blk_plug_cb, |
1702 | list); | |
2a7d5559 | 1703 | list_del(&cb->list); |
74018dc3 | 1704 | cb->callback(cb, from_schedule); |
2a7d5559 | 1705 | } |
048c9374 N |
1706 | } |
1707 | } | |
1708 | ||
9cbb1750 N |
1709 | struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data, |
1710 | int size) | |
1711 | { | |
1712 | struct blk_plug *plug = current->plug; | |
1713 | struct blk_plug_cb *cb; | |
1714 | ||
1715 | if (!plug) | |
1716 | return NULL; | |
1717 | ||
1718 | list_for_each_entry(cb, &plug->cb_list, list) | |
1719 | if (cb->callback == unplug && cb->data == data) | |
1720 | return cb; | |
1721 | ||
1722 | /* Not currently on the callback list */ | |
1723 | BUG_ON(size < sizeof(*cb)); | |
1724 | cb = kzalloc(size, GFP_ATOMIC); | |
1725 | if (cb) { | |
1726 | cb->data = data; | |
1727 | cb->callback = unplug; | |
1728 | list_add(&cb->list, &plug->cb_list); | |
1729 | } | |
1730 | return cb; | |
1731 | } | |
1732 | EXPORT_SYMBOL(blk_check_plugged); | |
1733 | ||
49cac01e | 1734 | void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule) |
73c10101 | 1735 | { |
74018dc3 | 1736 | flush_plug_callbacks(plug, from_schedule); |
320ae51f JA |
1737 | |
1738 | if (!list_empty(&plug->mq_list)) | |
1739 | blk_mq_flush_plug_list(plug, from_schedule); | |
73c10101 | 1740 | } |
73c10101 | 1741 | |
40405851 JM |
1742 | /** |
1743 | * blk_finish_plug - mark the end of a batch of submitted I/O | |
1744 | * @plug: The &struct blk_plug passed to blk_start_plug() | |
1745 | * | |
1746 | * Description: | |
1747 | * Indicate that a batch of I/O submissions is complete. This function | |
1748 | * must be paired with an initial call to blk_start_plug(). The intent | |
1749 | * is to allow the block layer to optimize I/O submission. See the | |
1750 | * documentation for blk_start_plug() for more information. | |
1751 | */ | |
73c10101 JA |
1752 | void blk_finish_plug(struct blk_plug *plug) |
1753 | { | |
dd6cf3e1 SL |
1754 | if (plug != current->plug) |
1755 | return; | |
f6603783 | 1756 | blk_flush_plug_list(plug, false); |
73c10101 | 1757 | |
dd6cf3e1 | 1758 | current->plug = NULL; |
73c10101 | 1759 | } |
88b996cd | 1760 | EXPORT_SYMBOL(blk_finish_plug); |
73c10101 | 1761 | |
71ac860a ML |
1762 | void blk_io_schedule(void) |
1763 | { | |
1764 | /* Prevent hang_check timer from firing at us during very long I/O */ | |
1765 | unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2; | |
1766 | ||
1767 | if (timeout) | |
1768 | io_schedule_timeout(timeout); | |
1769 | else | |
1770 | io_schedule(); | |
1771 | } | |
1772 | EXPORT_SYMBOL_GPL(blk_io_schedule); | |
1773 | ||
1da177e4 LT |
1774 | int __init blk_dev_init(void) |
1775 | { | |
ef295ecf CH |
1776 | BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS)); |
1777 | BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * | |
c593642c | 1778 | sizeof_field(struct request, cmd_flags)); |
ef295ecf | 1779 | BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * |
c593642c | 1780 | sizeof_field(struct bio, bi_opf)); |
9eb55b03 | 1781 | |
89b90be2 TH |
1782 | /* used for unplugging and affects IO latency/throughput - HIGHPRI */ |
1783 | kblockd_workqueue = alloc_workqueue("kblockd", | |
28747fcd | 1784 | WQ_MEM_RECLAIM | WQ_HIGHPRI, 0); |
1da177e4 LT |
1785 | if (!kblockd_workqueue) |
1786 | panic("Failed to create kblockd\n"); | |
1787 | ||
c2789bd4 | 1788 | blk_requestq_cachep = kmem_cache_create("request_queue", |
165125e1 | 1789 | sizeof(struct request_queue), 0, SLAB_PANIC, NULL); |
1da177e4 | 1790 | |
18fbda91 | 1791 | blk_debugfs_root = debugfs_create_dir("block", NULL); |
18fbda91 | 1792 | |
d38ecf93 | 1793 | return 0; |
1da177e4 | 1794 | } |