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); |
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; |
b554db14 | 125 | refcount_set(&rq->ref, 1); |
a892c8d5 | 126 | blk_crypto_rq_set_defaults(rq); |
1da177e4 | 127 | } |
2a4aa30c | 128 | EXPORT_SYMBOL(blk_rq_init); |
1da177e4 | 129 | |
e47bc4ed CK |
130 | #define REQ_OP_NAME(name) [REQ_OP_##name] = #name |
131 | static const char *const blk_op_name[] = { | |
132 | REQ_OP_NAME(READ), | |
133 | REQ_OP_NAME(WRITE), | |
134 | REQ_OP_NAME(FLUSH), | |
135 | REQ_OP_NAME(DISCARD), | |
136 | REQ_OP_NAME(SECURE_ERASE), | |
137 | REQ_OP_NAME(ZONE_RESET), | |
6e33dbf2 | 138 | REQ_OP_NAME(ZONE_RESET_ALL), |
6c1b1da5 AJ |
139 | REQ_OP_NAME(ZONE_OPEN), |
140 | REQ_OP_NAME(ZONE_CLOSE), | |
141 | REQ_OP_NAME(ZONE_FINISH), | |
0512a75b | 142 | REQ_OP_NAME(ZONE_APPEND), |
e47bc4ed CK |
143 | REQ_OP_NAME(WRITE_SAME), |
144 | REQ_OP_NAME(WRITE_ZEROES), | |
145 | REQ_OP_NAME(SCSI_IN), | |
146 | REQ_OP_NAME(SCSI_OUT), | |
147 | REQ_OP_NAME(DRV_IN), | |
148 | REQ_OP_NAME(DRV_OUT), | |
149 | }; | |
150 | #undef REQ_OP_NAME | |
151 | ||
152 | /** | |
153 | * blk_op_str - Return string XXX in the REQ_OP_XXX. | |
154 | * @op: REQ_OP_XXX. | |
155 | * | |
156 | * Description: Centralize block layer function to convert REQ_OP_XXX into | |
157 | * string format. Useful in the debugging and tracing bio or request. For | |
158 | * invalid REQ_OP_XXX it returns string "UNKNOWN". | |
159 | */ | |
160 | inline const char *blk_op_str(unsigned int op) | |
161 | { | |
162 | const char *op_str = "UNKNOWN"; | |
163 | ||
164 | if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op]) | |
165 | op_str = blk_op_name[op]; | |
166 | ||
167 | return op_str; | |
168 | } | |
169 | EXPORT_SYMBOL_GPL(blk_op_str); | |
170 | ||
2a842aca CH |
171 | static const struct { |
172 | int errno; | |
173 | const char *name; | |
174 | } blk_errors[] = { | |
175 | [BLK_STS_OK] = { 0, "" }, | |
176 | [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" }, | |
177 | [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" }, | |
178 | [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" }, | |
179 | [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" }, | |
180 | [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" }, | |
181 | [BLK_STS_NEXUS] = { -EBADE, "critical nexus" }, | |
182 | [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" }, | |
183 | [BLK_STS_PROTECTION] = { -EILSEQ, "protection" }, | |
184 | [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" }, | |
86ff7c2a | 185 | [BLK_STS_DEV_RESOURCE] = { -EBUSY, "device resource" }, |
03a07c92 | 186 | [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" }, |
2a842aca | 187 | |
4e4cbee9 CH |
188 | /* device mapper special case, should not leak out: */ |
189 | [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" }, | |
190 | ||
3b481d91 KB |
191 | /* zone device specific errors */ |
192 | [BLK_STS_ZONE_OPEN_RESOURCE] = { -ETOOMANYREFS, "open zones exceeded" }, | |
193 | [BLK_STS_ZONE_ACTIVE_RESOURCE] = { -EOVERFLOW, "active zones exceeded" }, | |
194 | ||
2a842aca CH |
195 | /* everything else not covered above: */ |
196 | [BLK_STS_IOERR] = { -EIO, "I/O" }, | |
197 | }; | |
198 | ||
199 | blk_status_t errno_to_blk_status(int errno) | |
200 | { | |
201 | int i; | |
202 | ||
203 | for (i = 0; i < ARRAY_SIZE(blk_errors); i++) { | |
204 | if (blk_errors[i].errno == errno) | |
205 | return (__force blk_status_t)i; | |
206 | } | |
207 | ||
208 | return BLK_STS_IOERR; | |
209 | } | |
210 | EXPORT_SYMBOL_GPL(errno_to_blk_status); | |
211 | ||
212 | int blk_status_to_errno(blk_status_t status) | |
213 | { | |
214 | int idx = (__force int)status; | |
215 | ||
34bd9c1c | 216 | if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors))) |
2a842aca CH |
217 | return -EIO; |
218 | return blk_errors[idx].errno; | |
219 | } | |
220 | EXPORT_SYMBOL_GPL(blk_status_to_errno); | |
221 | ||
178cc590 CH |
222 | static void print_req_error(struct request *req, blk_status_t status, |
223 | const char *caller) | |
2a842aca CH |
224 | { |
225 | int idx = (__force int)status; | |
226 | ||
34bd9c1c | 227 | if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors))) |
2a842aca CH |
228 | return; |
229 | ||
178cc590 | 230 | printk_ratelimited(KERN_ERR |
b0e5168a CK |
231 | "%s: %s error, dev %s, sector %llu op 0x%x:(%s) flags 0x%x " |
232 | "phys_seg %u prio class %u\n", | |
178cc590 | 233 | caller, blk_errors[idx].name, |
b0e5168a CK |
234 | req->rq_disk ? req->rq_disk->disk_name : "?", |
235 | blk_rq_pos(req), req_op(req), blk_op_str(req_op(req)), | |
236 | req->cmd_flags & ~REQ_OP_MASK, | |
237 | req->nr_phys_segments, | |
238 | IOPRIO_PRIO_CLASS(req->ioprio)); | |
2a842aca CH |
239 | } |
240 | ||
5bb23a68 | 241 | static void req_bio_endio(struct request *rq, struct bio *bio, |
2a842aca | 242 | unsigned int nbytes, blk_status_t error) |
1da177e4 | 243 | { |
78d8e58a | 244 | if (error) |
4e4cbee9 | 245 | bio->bi_status = error; |
797e7dbb | 246 | |
e8064021 | 247 | if (unlikely(rq->rq_flags & RQF_QUIET)) |
b7c44ed9 | 248 | bio_set_flag(bio, BIO_QUIET); |
08bafc03 | 249 | |
f79ea416 | 250 | bio_advance(bio, nbytes); |
7ba1ba12 | 251 | |
0512a75b KB |
252 | if (req_op(rq) == REQ_OP_ZONE_APPEND && error == BLK_STS_OK) { |
253 | /* | |
254 | * Partial zone append completions cannot be supported as the | |
255 | * BIO fragments may end up not being written sequentially. | |
256 | */ | |
257 | if (bio->bi_iter.bi_size) | |
258 | bio->bi_status = BLK_STS_IOERR; | |
259 | else | |
260 | bio->bi_iter.bi_sector = rq->__sector; | |
261 | } | |
262 | ||
143a87f4 | 263 | /* don't actually finish bio if it's part of flush sequence */ |
e8064021 | 264 | if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ)) |
4246a0b6 | 265 | bio_endio(bio); |
1da177e4 | 266 | } |
1da177e4 | 267 | |
1da177e4 LT |
268 | void blk_dump_rq_flags(struct request *rq, char *msg) |
269 | { | |
aebf526b CH |
270 | printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg, |
271 | rq->rq_disk ? rq->rq_disk->disk_name : "?", | |
5953316d | 272 | (unsigned long long) rq->cmd_flags); |
1da177e4 | 273 | |
83096ebf TH |
274 | printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n", |
275 | (unsigned long long)blk_rq_pos(rq), | |
276 | blk_rq_sectors(rq), blk_rq_cur_sectors(rq)); | |
b4f42e28 JA |
277 | printk(KERN_INFO " bio %p, biotail %p, len %u\n", |
278 | rq->bio, rq->biotail, blk_rq_bytes(rq)); | |
1da177e4 | 279 | } |
1da177e4 LT |
280 | EXPORT_SYMBOL(blk_dump_rq_flags); |
281 | ||
1da177e4 LT |
282 | /** |
283 | * blk_sync_queue - cancel any pending callbacks on a queue | |
284 | * @q: the queue | |
285 | * | |
286 | * Description: | |
287 | * The block layer may perform asynchronous callback activity | |
288 | * on a queue, such as calling the unplug function after a timeout. | |
289 | * A block device may call blk_sync_queue to ensure that any | |
290 | * such activity is cancelled, thus allowing it to release resources | |
59c51591 | 291 | * that the callbacks might use. The caller must already have made sure |
c62b37d9 | 292 | * that its ->submit_bio will not re-add plugging prior to calling |
1da177e4 LT |
293 | * this function. |
294 | * | |
da527770 | 295 | * This function does not cancel any asynchronous activity arising |
da3dae54 | 296 | * out of elevator or throttling code. That would require elevator_exit() |
5efd6113 | 297 | * and blkcg_exit_queue() to be called with queue lock initialized. |
da527770 | 298 | * |
1da177e4 LT |
299 | */ |
300 | void blk_sync_queue(struct request_queue *q) | |
301 | { | |
70ed28b9 | 302 | del_timer_sync(&q->timeout); |
4e9b6f20 | 303 | cancel_work_sync(&q->timeout_work); |
1da177e4 LT |
304 | } |
305 | EXPORT_SYMBOL(blk_sync_queue); | |
306 | ||
c9254f2d | 307 | /** |
cd84a62e | 308 | * blk_set_pm_only - increment pm_only counter |
c9254f2d | 309 | * @q: request queue pointer |
c9254f2d | 310 | */ |
cd84a62e | 311 | void blk_set_pm_only(struct request_queue *q) |
c9254f2d | 312 | { |
cd84a62e | 313 | atomic_inc(&q->pm_only); |
c9254f2d | 314 | } |
cd84a62e | 315 | EXPORT_SYMBOL_GPL(blk_set_pm_only); |
c9254f2d | 316 | |
cd84a62e | 317 | void blk_clear_pm_only(struct request_queue *q) |
c9254f2d | 318 | { |
cd84a62e BVA |
319 | int pm_only; |
320 | ||
321 | pm_only = atomic_dec_return(&q->pm_only); | |
322 | WARN_ON_ONCE(pm_only < 0); | |
323 | if (pm_only == 0) | |
324 | wake_up_all(&q->mq_freeze_wq); | |
c9254f2d | 325 | } |
cd84a62e | 326 | EXPORT_SYMBOL_GPL(blk_clear_pm_only); |
c9254f2d | 327 | |
b5bd357c LC |
328 | /** |
329 | * blk_put_queue - decrement the request_queue refcount | |
330 | * @q: the request_queue structure to decrement the refcount for | |
331 | * | |
332 | * Decrements the refcount of the request_queue kobject. When this reaches 0 | |
333 | * we'll have blk_release_queue() called. | |
e8c7d14a LC |
334 | * |
335 | * Context: Any context, but the last reference must not be dropped from | |
336 | * atomic context. | |
b5bd357c | 337 | */ |
165125e1 | 338 | void blk_put_queue(struct request_queue *q) |
483f4afc AV |
339 | { |
340 | kobject_put(&q->kobj); | |
341 | } | |
d86e0e83 | 342 | EXPORT_SYMBOL(blk_put_queue); |
483f4afc | 343 | |
aed3ea94 JA |
344 | void blk_set_queue_dying(struct request_queue *q) |
345 | { | |
8814ce8a | 346 | blk_queue_flag_set(QUEUE_FLAG_DYING, q); |
aed3ea94 | 347 | |
d3cfb2a0 ML |
348 | /* |
349 | * When queue DYING flag is set, we need to block new req | |
350 | * entering queue, so we call blk_freeze_queue_start() to | |
351 | * prevent I/O from crossing blk_queue_enter(). | |
352 | */ | |
353 | blk_freeze_queue_start(q); | |
354 | ||
344e9ffc | 355 | if (queue_is_mq(q)) |
aed3ea94 | 356 | blk_mq_wake_waiters(q); |
055f6e18 ML |
357 | |
358 | /* Make blk_queue_enter() reexamine the DYING flag. */ | |
359 | wake_up_all(&q->mq_freeze_wq); | |
aed3ea94 JA |
360 | } |
361 | EXPORT_SYMBOL_GPL(blk_set_queue_dying); | |
362 | ||
c9a929dd TH |
363 | /** |
364 | * blk_cleanup_queue - shutdown a request queue | |
365 | * @q: request queue to shutdown | |
366 | * | |
c246e80d BVA |
367 | * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and |
368 | * put it. All future requests will be failed immediately with -ENODEV. | |
e8c7d14a LC |
369 | * |
370 | * Context: can sleep | |
c94a96ac | 371 | */ |
6728cb0e | 372 | void blk_cleanup_queue(struct request_queue *q) |
483f4afc | 373 | { |
e8c7d14a LC |
374 | /* cannot be called from atomic context */ |
375 | might_sleep(); | |
376 | ||
bae85c15 BVA |
377 | WARN_ON_ONCE(blk_queue_registered(q)); |
378 | ||
3f3299d5 | 379 | /* mark @q DYING, no new request or merges will be allowed afterwards */ |
aed3ea94 | 380 | blk_set_queue_dying(q); |
6ecf23af | 381 | |
57d74df9 CH |
382 | blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q); |
383 | blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q); | |
c9a929dd | 384 | |
c246e80d BVA |
385 | /* |
386 | * Drain all requests queued before DYING marking. Set DEAD flag to | |
67ed8b73 BVA |
387 | * prevent that blk_mq_run_hw_queues() accesses the hardware queues |
388 | * after draining finished. | |
c246e80d | 389 | */ |
3ef28e83 | 390 | blk_freeze_queue(q); |
c57cdf7a ML |
391 | |
392 | rq_qos_exit(q); | |
393 | ||
57d74df9 | 394 | blk_queue_flag_set(QUEUE_FLAG_DEAD, q); |
c9a929dd | 395 | |
5a48fc14 DW |
396 | /* for synchronous bio-based driver finish in-flight integrity i/o */ |
397 | blk_flush_integrity(); | |
398 | ||
c9a929dd | 399 | /* @q won't process any more request, flush async actions */ |
dc3b17cc | 400 | del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer); |
c9a929dd TH |
401 | blk_sync_queue(q); |
402 | ||
344e9ffc | 403 | if (queue_is_mq(q)) |
c7e2d94b | 404 | blk_mq_exit_queue(q); |
a1ce35fa | 405 | |
c3e22192 ML |
406 | /* |
407 | * In theory, request pool of sched_tags belongs to request queue. | |
408 | * However, the current implementation requires tag_set for freeing | |
409 | * requests, so free the pool now. | |
410 | * | |
411 | * Queue has become frozen, there can't be any in-queue requests, so | |
412 | * it is safe to free requests now. | |
413 | */ | |
414 | mutex_lock(&q->sysfs_lock); | |
415 | if (q->elevator) | |
416 | blk_mq_sched_free_requests(q); | |
417 | mutex_unlock(&q->sysfs_lock); | |
418 | ||
3ef28e83 | 419 | percpu_ref_exit(&q->q_usage_counter); |
45a9c9d9 | 420 | |
c9a929dd | 421 | /* @q is and will stay empty, shutdown and put */ |
483f4afc AV |
422 | blk_put_queue(q); |
423 | } | |
1da177e4 LT |
424 | EXPORT_SYMBOL(blk_cleanup_queue); |
425 | ||
3a0a5299 BVA |
426 | /** |
427 | * blk_queue_enter() - try to increase q->q_usage_counter | |
428 | * @q: request queue pointer | |
a4d34da7 | 429 | * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PM |
3a0a5299 | 430 | */ |
9a95e4ef | 431 | int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags) |
3ef28e83 | 432 | { |
a4d34da7 | 433 | const bool pm = flags & BLK_MQ_REQ_PM; |
3a0a5299 | 434 | |
3ef28e83 | 435 | while (true) { |
3a0a5299 | 436 | bool success = false; |
3ef28e83 | 437 | |
818e0fa2 | 438 | rcu_read_lock(); |
3a0a5299 BVA |
439 | if (percpu_ref_tryget_live(&q->q_usage_counter)) { |
440 | /* | |
cd84a62e BVA |
441 | * The code that increments the pm_only counter is |
442 | * responsible for ensuring that that counter is | |
443 | * globally visible before the queue is unfrozen. | |
3a0a5299 | 444 | */ |
52abca64 AS |
445 | if ((pm && queue_rpm_status(q) != RPM_SUSPENDED) || |
446 | !blk_queue_pm_only(q)) { | |
3a0a5299 BVA |
447 | success = true; |
448 | } else { | |
449 | percpu_ref_put(&q->q_usage_counter); | |
450 | } | |
451 | } | |
818e0fa2 | 452 | rcu_read_unlock(); |
3a0a5299 BVA |
453 | |
454 | if (success) | |
3ef28e83 DW |
455 | return 0; |
456 | ||
3a0a5299 | 457 | if (flags & BLK_MQ_REQ_NOWAIT) |
3ef28e83 DW |
458 | return -EBUSY; |
459 | ||
5ed61d3f | 460 | /* |
1671d522 | 461 | * read pair of barrier in blk_freeze_queue_start(), |
5ed61d3f | 462 | * we need to order reading __PERCPU_REF_DEAD flag of |
d3cfb2a0 ML |
463 | * .q_usage_counter and reading .mq_freeze_depth or |
464 | * queue dying flag, otherwise the following wait may | |
465 | * never return if the two reads are reordered. | |
5ed61d3f ML |
466 | */ |
467 | smp_rmb(); | |
468 | ||
1dc3039b | 469 | wait_event(q->mq_freeze_wq, |
7996a8b5 | 470 | (!q->mq_freeze_depth && |
52abca64 | 471 | blk_pm_resume_queue(pm, q)) || |
1dc3039b | 472 | blk_queue_dying(q)); |
3ef28e83 DW |
473 | if (blk_queue_dying(q)) |
474 | return -ENODEV; | |
3ef28e83 DW |
475 | } |
476 | } | |
477 | ||
accea322 CH |
478 | static inline int bio_queue_enter(struct bio *bio) |
479 | { | |
309dca30 | 480 | struct request_queue *q = bio->bi_bdev->bd_disk->queue; |
accea322 CH |
481 | bool nowait = bio->bi_opf & REQ_NOWAIT; |
482 | int ret; | |
483 | ||
484 | ret = blk_queue_enter(q, nowait ? BLK_MQ_REQ_NOWAIT : 0); | |
485 | if (unlikely(ret)) { | |
486 | if (nowait && !blk_queue_dying(q)) | |
487 | bio_wouldblock_error(bio); | |
488 | else | |
489 | bio_io_error(bio); | |
490 | } | |
491 | ||
492 | return ret; | |
493 | } | |
494 | ||
3ef28e83 DW |
495 | void blk_queue_exit(struct request_queue *q) |
496 | { | |
497 | percpu_ref_put(&q->q_usage_counter); | |
498 | } | |
499 | ||
500 | static void blk_queue_usage_counter_release(struct percpu_ref *ref) | |
501 | { | |
502 | struct request_queue *q = | |
503 | container_of(ref, struct request_queue, q_usage_counter); | |
504 | ||
505 | wake_up_all(&q->mq_freeze_wq); | |
506 | } | |
507 | ||
bca237a5 | 508 | static void blk_rq_timed_out_timer(struct timer_list *t) |
287922eb | 509 | { |
bca237a5 | 510 | struct request_queue *q = from_timer(q, t, timeout); |
287922eb CH |
511 | |
512 | kblockd_schedule_work(&q->timeout_work); | |
513 | } | |
514 | ||
2e3c18d0 TH |
515 | static void blk_timeout_work(struct work_struct *work) |
516 | { | |
517 | } | |
518 | ||
c62b37d9 | 519 | struct request_queue *blk_alloc_queue(int node_id) |
1946089a | 520 | { |
165125e1 | 521 | struct request_queue *q; |
338aa96d | 522 | int ret; |
1946089a | 523 | |
8324aa91 | 524 | q = kmem_cache_alloc_node(blk_requestq_cachep, |
3d745ea5 | 525 | GFP_KERNEL | __GFP_ZERO, node_id); |
1da177e4 LT |
526 | if (!q) |
527 | return NULL; | |
528 | ||
cbf62af3 | 529 | q->last_merge = NULL; |
cbf62af3 | 530 | |
3d745ea5 | 531 | q->id = ida_simple_get(&blk_queue_ida, 0, 0, GFP_KERNEL); |
a73f730d | 532 | if (q->id < 0) |
3d2936f4 | 533 | goto fail_q; |
a73f730d | 534 | |
c495a176 | 535 | ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, 0); |
338aa96d | 536 | if (ret) |
54efd50b KO |
537 | goto fail_id; |
538 | ||
aef33c2f | 539 | q->backing_dev_info = bdi_alloc(node_id); |
d03f6cdc JK |
540 | if (!q->backing_dev_info) |
541 | goto fail_split; | |
542 | ||
a83b576c JA |
543 | q->stats = blk_alloc_queue_stats(); |
544 | if (!q->stats) | |
545 | goto fail_stats; | |
546 | ||
5151412d | 547 | q->node = node_id; |
0989a025 | 548 | |
bccf5e26 JG |
549 | atomic_set(&q->nr_active_requests_shared_sbitmap, 0); |
550 | ||
bca237a5 KC |
551 | timer_setup(&q->backing_dev_info->laptop_mode_wb_timer, |
552 | laptop_mode_timer_fn, 0); | |
553 | timer_setup(&q->timeout, blk_rq_timed_out_timer, 0); | |
2e3c18d0 | 554 | INIT_WORK(&q->timeout_work, blk_timeout_work); |
a612fddf | 555 | INIT_LIST_HEAD(&q->icq_list); |
4eef3049 | 556 | #ifdef CONFIG_BLK_CGROUP |
e8989fae | 557 | INIT_LIST_HEAD(&q->blkg_list); |
4eef3049 | 558 | #endif |
483f4afc | 559 | |
8324aa91 | 560 | kobject_init(&q->kobj, &blk_queue_ktype); |
1da177e4 | 561 | |
85e0cbbb | 562 | mutex_init(&q->debugfs_mutex); |
483f4afc | 563 | mutex_init(&q->sysfs_lock); |
cecf5d87 | 564 | mutex_init(&q->sysfs_dir_lock); |
0d945c1f | 565 | spin_lock_init(&q->queue_lock); |
c94a96ac | 566 | |
320ae51f | 567 | init_waitqueue_head(&q->mq_freeze_wq); |
7996a8b5 | 568 | mutex_init(&q->mq_freeze_lock); |
320ae51f | 569 | |
3ef28e83 DW |
570 | /* |
571 | * Init percpu_ref in atomic mode so that it's faster to shutdown. | |
572 | * See blk_register_queue() for details. | |
573 | */ | |
574 | if (percpu_ref_init(&q->q_usage_counter, | |
575 | blk_queue_usage_counter_release, | |
576 | PERCPU_REF_INIT_ATOMIC, GFP_KERNEL)) | |
fff4996b | 577 | goto fail_bdi; |
f51b802c | 578 | |
3ef28e83 DW |
579 | if (blkcg_init_queue(q)) |
580 | goto fail_ref; | |
581 | ||
3d745ea5 CH |
582 | blk_queue_dma_alignment(q, 511); |
583 | blk_set_default_limits(&q->limits); | |
c62b37d9 | 584 | q->nr_requests = BLKDEV_MAX_RQ; |
3d745ea5 | 585 | |
1da177e4 | 586 | return q; |
a73f730d | 587 | |
3ef28e83 DW |
588 | fail_ref: |
589 | percpu_ref_exit(&q->q_usage_counter); | |
fff4996b | 590 | fail_bdi: |
a83b576c JA |
591 | blk_free_queue_stats(q->stats); |
592 | fail_stats: | |
d03f6cdc | 593 | bdi_put(q->backing_dev_info); |
54efd50b | 594 | fail_split: |
338aa96d | 595 | bioset_exit(&q->bio_split); |
a73f730d TH |
596 | fail_id: |
597 | ida_simple_remove(&blk_queue_ida, q->id); | |
598 | fail_q: | |
599 | kmem_cache_free(blk_requestq_cachep, q); | |
600 | return NULL; | |
1da177e4 | 601 | } |
3d745ea5 | 602 | EXPORT_SYMBOL(blk_alloc_queue); |
1da177e4 | 603 | |
b5bd357c LC |
604 | /** |
605 | * blk_get_queue - increment the request_queue refcount | |
606 | * @q: the request_queue structure to increment the refcount for | |
607 | * | |
608 | * Increment the refcount of the request_queue kobject. | |
763b5892 LC |
609 | * |
610 | * Context: Any context. | |
b5bd357c | 611 | */ |
09ac46c4 | 612 | bool blk_get_queue(struct request_queue *q) |
1da177e4 | 613 | { |
3f3299d5 | 614 | if (likely(!blk_queue_dying(q))) { |
09ac46c4 TH |
615 | __blk_get_queue(q); |
616 | return true; | |
1da177e4 LT |
617 | } |
618 | ||
09ac46c4 | 619 | return false; |
1da177e4 | 620 | } |
d86e0e83 | 621 | EXPORT_SYMBOL(blk_get_queue); |
1da177e4 | 622 | |
a1ce35fa JA |
623 | /** |
624 | * blk_get_request - allocate a request | |
625 | * @q: request queue to allocate a request for | |
626 | * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC. | |
627 | * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT. | |
1da177e4 | 628 | */ |
a1ce35fa JA |
629 | struct request *blk_get_request(struct request_queue *q, unsigned int op, |
630 | blk_mq_req_flags_t flags) | |
1da177e4 | 631 | { |
a1ce35fa | 632 | struct request *req; |
1da177e4 | 633 | |
a1ce35fa | 634 | WARN_ON_ONCE(op & REQ_NOWAIT); |
a4d34da7 | 635 | WARN_ON_ONCE(flags & ~(BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_PM)); |
1da177e4 | 636 | |
a1ce35fa JA |
637 | req = blk_mq_alloc_request(q, op, flags); |
638 | if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn) | |
639 | q->mq_ops->initialize_rq_fn(req); | |
1da177e4 | 640 | |
a1ce35fa | 641 | return req; |
1da177e4 | 642 | } |
a1ce35fa | 643 | EXPORT_SYMBOL(blk_get_request); |
1da177e4 | 644 | |
1da177e4 LT |
645 | void blk_put_request(struct request *req) |
646 | { | |
a1ce35fa | 647 | blk_mq_free_request(req); |
1da177e4 | 648 | } |
1da177e4 LT |
649 | EXPORT_SYMBOL(blk_put_request); |
650 | ||
52c5e62d | 651 | static void handle_bad_sector(struct bio *bio, sector_t maxsector) |
1da177e4 LT |
652 | { |
653 | char b[BDEVNAME_SIZE]; | |
654 | ||
f4ac712e TH |
655 | pr_info_ratelimited("attempt to access beyond end of device\n" |
656 | "%s: rw=%d, want=%llu, limit=%llu\n", | |
657 | bio_devname(bio, b), bio->bi_opf, | |
658 | bio_end_sector(bio), maxsector); | |
1da177e4 LT |
659 | } |
660 | ||
c17bb495 AM |
661 | #ifdef CONFIG_FAIL_MAKE_REQUEST |
662 | ||
663 | static DECLARE_FAULT_ATTR(fail_make_request); | |
664 | ||
665 | static int __init setup_fail_make_request(char *str) | |
666 | { | |
667 | return setup_fault_attr(&fail_make_request, str); | |
668 | } | |
669 | __setup("fail_make_request=", setup_fail_make_request); | |
670 | ||
8446fe92 | 671 | static bool should_fail_request(struct block_device *part, unsigned int bytes) |
c17bb495 | 672 | { |
8446fe92 | 673 | return part->bd_make_it_fail && should_fail(&fail_make_request, bytes); |
c17bb495 AM |
674 | } |
675 | ||
676 | static int __init fail_make_request_debugfs(void) | |
677 | { | |
dd48c085 AM |
678 | struct dentry *dir = fault_create_debugfs_attr("fail_make_request", |
679 | NULL, &fail_make_request); | |
680 | ||
21f9fcd8 | 681 | return PTR_ERR_OR_ZERO(dir); |
c17bb495 AM |
682 | } |
683 | ||
684 | late_initcall(fail_make_request_debugfs); | |
685 | ||
686 | #else /* CONFIG_FAIL_MAKE_REQUEST */ | |
687 | ||
8446fe92 | 688 | static inline bool should_fail_request(struct block_device *part, |
b2c9cd37 | 689 | unsigned int bytes) |
c17bb495 | 690 | { |
b2c9cd37 | 691 | return false; |
c17bb495 AM |
692 | } |
693 | ||
694 | #endif /* CONFIG_FAIL_MAKE_REQUEST */ | |
695 | ||
2f9f6221 | 696 | static inline bool bio_check_ro(struct bio *bio) |
721c7fc7 | 697 | { |
2f9f6221 | 698 | if (op_is_write(bio_op(bio)) && bdev_read_only(bio->bi_bdev)) { |
721c7fc7 ID |
699 | char b[BDEVNAME_SIZE]; |
700 | ||
8b2ded1c MP |
701 | if (op_is_flush(bio->bi_opf) && !bio_sectors(bio)) |
702 | return false; | |
703 | ||
a32e236e | 704 | WARN_ONCE(1, |
c8178674 | 705 | "Trying to write to read-only block-device %s (partno %d)\n", |
2f9f6221 | 706 | bio_devname(bio, b), bio->bi_bdev->bd_partno); |
a32e236e LT |
707 | /* Older lvm-tools actually trigger this */ |
708 | return false; | |
721c7fc7 ID |
709 | } |
710 | ||
711 | return false; | |
712 | } | |
713 | ||
30abb3a6 HM |
714 | static noinline int should_fail_bio(struct bio *bio) |
715 | { | |
309dca30 | 716 | if (should_fail_request(bdev_whole(bio->bi_bdev), bio->bi_iter.bi_size)) |
30abb3a6 HM |
717 | return -EIO; |
718 | return 0; | |
719 | } | |
720 | ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO); | |
721 | ||
52c5e62d CH |
722 | /* |
723 | * Check whether this bio extends beyond the end of the device or partition. | |
724 | * This may well happen - the kernel calls bread() without checking the size of | |
725 | * the device, e.g., when mounting a file system. | |
726 | */ | |
2f9f6221 | 727 | static inline int bio_check_eod(struct bio *bio) |
52c5e62d | 728 | { |
2f9f6221 | 729 | sector_t maxsector = bdev_nr_sectors(bio->bi_bdev); |
52c5e62d CH |
730 | unsigned int nr_sectors = bio_sectors(bio); |
731 | ||
732 | if (nr_sectors && maxsector && | |
733 | (nr_sectors > maxsector || | |
734 | bio->bi_iter.bi_sector > maxsector - nr_sectors)) { | |
735 | handle_bad_sector(bio, maxsector); | |
736 | return -EIO; | |
737 | } | |
738 | return 0; | |
739 | } | |
740 | ||
74d46992 CH |
741 | /* |
742 | * Remap block n of partition p to block n+start(p) of the disk. | |
743 | */ | |
2f9f6221 | 744 | static int blk_partition_remap(struct bio *bio) |
74d46992 | 745 | { |
309dca30 | 746 | struct block_device *p = bio->bi_bdev; |
74d46992 | 747 | |
52c5e62d | 748 | if (unlikely(should_fail_request(p, bio->bi_iter.bi_size))) |
2f9f6221 | 749 | return -EIO; |
5eac3eb3 | 750 | if (bio_sectors(bio)) { |
8446fe92 | 751 | bio->bi_iter.bi_sector += p->bd_start_sect; |
1c02fca6 | 752 | trace_block_bio_remap(bio, p->bd_dev, |
29ff57c6 | 753 | bio->bi_iter.bi_sector - |
8446fe92 | 754 | p->bd_start_sect); |
52c5e62d | 755 | } |
30c5d345 | 756 | bio_set_flag(bio, BIO_REMAPPED); |
2f9f6221 | 757 | return 0; |
74d46992 CH |
758 | } |
759 | ||
0512a75b KB |
760 | /* |
761 | * Check write append to a zoned block device. | |
762 | */ | |
763 | static inline blk_status_t blk_check_zone_append(struct request_queue *q, | |
764 | struct bio *bio) | |
765 | { | |
766 | sector_t pos = bio->bi_iter.bi_sector; | |
767 | int nr_sectors = bio_sectors(bio); | |
768 | ||
769 | /* Only applicable to zoned block devices */ | |
770 | if (!blk_queue_is_zoned(q)) | |
771 | return BLK_STS_NOTSUPP; | |
772 | ||
773 | /* The bio sector must point to the start of a sequential zone */ | |
774 | if (pos & (blk_queue_zone_sectors(q) - 1) || | |
775 | !blk_queue_zone_is_seq(q, pos)) | |
776 | return BLK_STS_IOERR; | |
777 | ||
778 | /* | |
779 | * Not allowed to cross zone boundaries. Otherwise, the BIO will be | |
780 | * split and could result in non-contiguous sectors being written in | |
781 | * different zones. | |
782 | */ | |
783 | if (nr_sectors > q->limits.chunk_sectors) | |
784 | return BLK_STS_IOERR; | |
785 | ||
786 | /* Make sure the BIO is small enough and will not get split */ | |
787 | if (nr_sectors > q->limits.max_zone_append_sectors) | |
788 | return BLK_STS_IOERR; | |
789 | ||
790 | bio->bi_opf |= REQ_NOMERGE; | |
791 | ||
792 | return BLK_STS_OK; | |
793 | } | |
794 | ||
ed00aabd | 795 | static noinline_for_stack bool submit_bio_checks(struct bio *bio) |
1da177e4 | 796 | { |
309dca30 CH |
797 | struct block_device *bdev = bio->bi_bdev; |
798 | struct request_queue *q = bdev->bd_disk->queue; | |
4e4cbee9 | 799 | blk_status_t status = BLK_STS_IOERR; |
5a473e83 | 800 | struct blk_plug *plug; |
1da177e4 LT |
801 | |
802 | might_sleep(); | |
1da177e4 | 803 | |
5a473e83 JA |
804 | plug = blk_mq_plug(q, bio); |
805 | if (plug && plug->nowait) | |
806 | bio->bi_opf |= REQ_NOWAIT; | |
807 | ||
03a07c92 | 808 | /* |
b0beb280 | 809 | * For a REQ_NOWAIT based request, return -EOPNOTSUPP |
021a2446 | 810 | * if queue does not support NOWAIT. |
03a07c92 | 811 | */ |
021a2446 | 812 | if ((bio->bi_opf & REQ_NOWAIT) && !blk_queue_nowait(q)) |
b0beb280 | 813 | goto not_supported; |
03a07c92 | 814 | |
30abb3a6 | 815 | if (should_fail_bio(bio)) |
5a7bbad2 | 816 | goto end_io; |
2f9f6221 CH |
817 | if (unlikely(bio_check_ro(bio))) |
818 | goto end_io; | |
3a905c37 CH |
819 | if (!bio_flagged(bio, BIO_REMAPPED)) { |
820 | if (unlikely(bio_check_eod(bio))) | |
821 | goto end_io; | |
822 | if (bdev->bd_partno && unlikely(blk_partition_remap(bio))) | |
823 | goto end_io; | |
824 | } | |
2056a782 | 825 | |
5a7bbad2 | 826 | /* |
ed00aabd CH |
827 | * Filter flush bio's early so that bio based drivers without flush |
828 | * support don't have to worry about them. | |
5a7bbad2 | 829 | */ |
f3a8ab7d | 830 | if (op_is_flush(bio->bi_opf) && |
c888a8f9 | 831 | !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) { |
1eff9d32 | 832 | bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA); |
e439ab71 | 833 | if (!bio_sectors(bio)) { |
4e4cbee9 | 834 | status = BLK_STS_OK; |
51fd77bd JA |
835 | goto end_io; |
836 | } | |
5a7bbad2 | 837 | } |
5ddfe969 | 838 | |
d04c406f CH |
839 | if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags)) |
840 | bio->bi_opf &= ~REQ_HIPRI; | |
841 | ||
288dab8a CH |
842 | switch (bio_op(bio)) { |
843 | case REQ_OP_DISCARD: | |
844 | if (!blk_queue_discard(q)) | |
845 | goto not_supported; | |
846 | break; | |
847 | case REQ_OP_SECURE_ERASE: | |
848 | if (!blk_queue_secure_erase(q)) | |
849 | goto not_supported; | |
850 | break; | |
851 | case REQ_OP_WRITE_SAME: | |
74d46992 | 852 | if (!q->limits.max_write_same_sectors) |
288dab8a | 853 | goto not_supported; |
58886785 | 854 | break; |
0512a75b KB |
855 | case REQ_OP_ZONE_APPEND: |
856 | status = blk_check_zone_append(q, bio); | |
857 | if (status != BLK_STS_OK) | |
858 | goto end_io; | |
859 | break; | |
2d253440 | 860 | case REQ_OP_ZONE_RESET: |
6c1b1da5 AJ |
861 | case REQ_OP_ZONE_OPEN: |
862 | case REQ_OP_ZONE_CLOSE: | |
863 | case REQ_OP_ZONE_FINISH: | |
74d46992 | 864 | if (!blk_queue_is_zoned(q)) |
2d253440 | 865 | goto not_supported; |
288dab8a | 866 | break; |
6e33dbf2 CK |
867 | case REQ_OP_ZONE_RESET_ALL: |
868 | if (!blk_queue_is_zoned(q) || !blk_queue_zone_resetall(q)) | |
869 | goto not_supported; | |
870 | break; | |
a6f0788e | 871 | case REQ_OP_WRITE_ZEROES: |
74d46992 | 872 | if (!q->limits.max_write_zeroes_sectors) |
a6f0788e CK |
873 | goto not_supported; |
874 | break; | |
288dab8a CH |
875 | default: |
876 | break; | |
5a7bbad2 | 877 | } |
01edede4 | 878 | |
7f4b35d1 | 879 | /* |
3e82c348 CH |
880 | * Various block parts want %current->io_context, so allocate it up |
881 | * front rather than dealing with lots of pain to allocate it only | |
882 | * where needed. This may fail and the block layer knows how to live | |
883 | * with it. | |
7f4b35d1 | 884 | */ |
3e82c348 CH |
885 | if (unlikely(!current->io_context)) |
886 | create_task_io_context(current, GFP_ATOMIC, q->node); | |
7f4b35d1 | 887 | |
db18a53e CH |
888 | if (blk_throtl_bio(bio)) { |
889 | blkcg_bio_issue_init(bio); | |
ae118896 | 890 | return false; |
db18a53e CH |
891 | } |
892 | ||
893 | blk_cgroup_bio_start(bio); | |
894 | blkcg_bio_issue_init(bio); | |
27a84d54 | 895 | |
fbbaf700 | 896 | if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) { |
e8a676d6 | 897 | trace_block_bio_queue(bio); |
fbbaf700 N |
898 | /* Now that enqueuing has been traced, we need to trace |
899 | * completion as well. | |
900 | */ | |
901 | bio_set_flag(bio, BIO_TRACE_COMPLETION); | |
902 | } | |
27a84d54 | 903 | return true; |
a7384677 | 904 | |
288dab8a | 905 | not_supported: |
4e4cbee9 | 906 | status = BLK_STS_NOTSUPP; |
a7384677 | 907 | end_io: |
4e4cbee9 | 908 | bio->bi_status = status; |
4246a0b6 | 909 | bio_endio(bio); |
27a84d54 | 910 | return false; |
1da177e4 LT |
911 | } |
912 | ||
ed00aabd | 913 | static blk_qc_t __submit_bio(struct bio *bio) |
ac7c5675 | 914 | { |
309dca30 | 915 | struct gendisk *disk = bio->bi_bdev->bd_disk; |
ac7c5675 CH |
916 | blk_qc_t ret = BLK_QC_T_NONE; |
917 | ||
918 | if (blk_crypto_bio_prep(&bio)) { | |
c62b37d9 CH |
919 | if (!disk->fops->submit_bio) |
920 | return blk_mq_submit_bio(bio); | |
921 | ret = disk->fops->submit_bio(bio); | |
ac7c5675 | 922 | } |
c62b37d9 | 923 | blk_queue_exit(disk->queue); |
ac7c5675 CH |
924 | return ret; |
925 | } | |
926 | ||
566acf2d CH |
927 | /* |
928 | * The loop in this function may be a bit non-obvious, and so deserves some | |
929 | * explanation: | |
930 | * | |
931 | * - Before entering the loop, bio->bi_next is NULL (as all callers ensure | |
932 | * that), so we have a list with a single bio. | |
933 | * - We pretend that we have just taken it off a longer list, so we assign | |
934 | * bio_list to a pointer to the bio_list_on_stack, thus initialising the | |
935 | * bio_list of new bios to be added. ->submit_bio() may indeed add some more | |
936 | * bios through a recursive call to submit_bio_noacct. If it did, we find a | |
937 | * non-NULL value in bio_list and re-enter the loop from the top. | |
938 | * - In this case we really did just take the bio of the top of the list (no | |
939 | * pretending) and so remove it from bio_list, and call into ->submit_bio() | |
940 | * again. | |
941 | * | |
942 | * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio. | |
943 | * bio_list_on_stack[1] contains bios that were submitted before the current | |
944 | * ->submit_bio_bio, but that haven't been processed yet. | |
945 | */ | |
946 | static blk_qc_t __submit_bio_noacct(struct bio *bio) | |
947 | { | |
948 | struct bio_list bio_list_on_stack[2]; | |
949 | blk_qc_t ret = BLK_QC_T_NONE; | |
950 | ||
951 | BUG_ON(bio->bi_next); | |
952 | ||
953 | bio_list_init(&bio_list_on_stack[0]); | |
954 | current->bio_list = bio_list_on_stack; | |
955 | ||
956 | do { | |
309dca30 | 957 | struct request_queue *q = bio->bi_bdev->bd_disk->queue; |
566acf2d CH |
958 | struct bio_list lower, same; |
959 | ||
960 | if (unlikely(bio_queue_enter(bio) != 0)) | |
961 | continue; | |
962 | ||
963 | /* | |
964 | * Create a fresh bio_list for all subordinate requests. | |
965 | */ | |
966 | bio_list_on_stack[1] = bio_list_on_stack[0]; | |
967 | bio_list_init(&bio_list_on_stack[0]); | |
968 | ||
969 | ret = __submit_bio(bio); | |
970 | ||
971 | /* | |
972 | * Sort new bios into those for a lower level and those for the | |
973 | * same level. | |
974 | */ | |
975 | bio_list_init(&lower); | |
976 | bio_list_init(&same); | |
977 | while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL) | |
309dca30 | 978 | if (q == bio->bi_bdev->bd_disk->queue) |
566acf2d CH |
979 | bio_list_add(&same, bio); |
980 | else | |
981 | bio_list_add(&lower, bio); | |
982 | ||
983 | /* | |
984 | * Now assemble so we handle the lowest level first. | |
985 | */ | |
986 | bio_list_merge(&bio_list_on_stack[0], &lower); | |
987 | bio_list_merge(&bio_list_on_stack[0], &same); | |
988 | bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]); | |
989 | } while ((bio = bio_list_pop(&bio_list_on_stack[0]))); | |
990 | ||
991 | current->bio_list = NULL; | |
992 | return ret; | |
993 | } | |
994 | ||
ff93ea0c CH |
995 | static blk_qc_t __submit_bio_noacct_mq(struct bio *bio) |
996 | { | |
7c792f33 | 997 | struct bio_list bio_list[2] = { }; |
ff93ea0c CH |
998 | blk_qc_t ret = BLK_QC_T_NONE; |
999 | ||
7c792f33 | 1000 | current->bio_list = bio_list; |
ff93ea0c CH |
1001 | |
1002 | do { | |
309dca30 | 1003 | struct gendisk *disk = bio->bi_bdev->bd_disk; |
ff93ea0c CH |
1004 | |
1005 | if (unlikely(bio_queue_enter(bio) != 0)) | |
1006 | continue; | |
1007 | ||
1008 | if (!blk_crypto_bio_prep(&bio)) { | |
1009 | blk_queue_exit(disk->queue); | |
1010 | ret = BLK_QC_T_NONE; | |
1011 | continue; | |
1012 | } | |
1013 | ||
1014 | ret = blk_mq_submit_bio(bio); | |
7c792f33 | 1015 | } while ((bio = bio_list_pop(&bio_list[0]))); |
ff93ea0c CH |
1016 | |
1017 | current->bio_list = NULL; | |
1018 | return ret; | |
1019 | } | |
1020 | ||
27a84d54 | 1021 | /** |
ed00aabd | 1022 | * submit_bio_noacct - re-submit a bio to the block device layer for I/O |
27a84d54 CH |
1023 | * @bio: The bio describing the location in memory and on the device. |
1024 | * | |
3fdd4086 CH |
1025 | * This is a version of submit_bio() that shall only be used for I/O that is |
1026 | * resubmitted to lower level drivers by stacking block drivers. All file | |
1027 | * systems and other upper level users of the block layer should use | |
1028 | * submit_bio() instead. | |
d89d8796 | 1029 | */ |
ed00aabd | 1030 | blk_qc_t submit_bio_noacct(struct bio *bio) |
d89d8796 | 1031 | { |
ed00aabd | 1032 | if (!submit_bio_checks(bio)) |
566acf2d | 1033 | return BLK_QC_T_NONE; |
27a84d54 CH |
1034 | |
1035 | /* | |
566acf2d CH |
1036 | * We only want one ->submit_bio to be active at a time, else stack |
1037 | * usage with stacked devices could be a problem. Use current->bio_list | |
1038 | * to collect a list of requests submited by a ->submit_bio method while | |
1039 | * it is active, and then process them after it returned. | |
27a84d54 | 1040 | */ |
bddd87c7 | 1041 | if (current->bio_list) { |
f5fe1b51 | 1042 | bio_list_add(¤t->bio_list[0], bio); |
566acf2d | 1043 | return BLK_QC_T_NONE; |
d89d8796 | 1044 | } |
27a84d54 | 1045 | |
309dca30 | 1046 | if (!bio->bi_bdev->bd_disk->fops->submit_bio) |
ff93ea0c | 1047 | return __submit_bio_noacct_mq(bio); |
566acf2d | 1048 | return __submit_bio_noacct(bio); |
d89d8796 | 1049 | } |
ed00aabd | 1050 | EXPORT_SYMBOL(submit_bio_noacct); |
1da177e4 LT |
1051 | |
1052 | /** | |
710027a4 | 1053 | * submit_bio - submit a bio to the block device layer for I/O |
1da177e4 LT |
1054 | * @bio: The &struct bio which describes the I/O |
1055 | * | |
3fdd4086 CH |
1056 | * submit_bio() is used to submit I/O requests to block devices. It is passed a |
1057 | * fully set up &struct bio that describes the I/O that needs to be done. The | |
309dca30 | 1058 | * bio will be send to the device described by the bi_bdev field. |
1da177e4 | 1059 | * |
3fdd4086 CH |
1060 | * The success/failure status of the request, along with notification of |
1061 | * completion, is delivered asynchronously through the ->bi_end_io() callback | |
1062 | * in @bio. The bio must NOT be touched by thecaller until ->bi_end_io() has | |
1063 | * been called. | |
1da177e4 | 1064 | */ |
4e49ea4a | 1065 | blk_qc_t submit_bio(struct bio *bio) |
1da177e4 | 1066 | { |
d3f77dfd TH |
1067 | if (blkcg_punt_bio_submit(bio)) |
1068 | return BLK_QC_T_NONE; | |
1069 | ||
bf2de6f5 JA |
1070 | /* |
1071 | * If it's a regular read/write or a barrier with data attached, | |
1072 | * go through the normal accounting stuff before submission. | |
1073 | */ | |
e2a60da7 | 1074 | if (bio_has_data(bio)) { |
4363ac7c MP |
1075 | unsigned int count; |
1076 | ||
95fe6c1a | 1077 | if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME)) |
309dca30 CH |
1078 | count = queue_logical_block_size( |
1079 | bio->bi_bdev->bd_disk->queue) >> 9; | |
4363ac7c MP |
1080 | else |
1081 | count = bio_sectors(bio); | |
1082 | ||
a8ebb056 | 1083 | if (op_is_write(bio_op(bio))) { |
bf2de6f5 JA |
1084 | count_vm_events(PGPGOUT, count); |
1085 | } else { | |
4f024f37 | 1086 | task_io_account_read(bio->bi_iter.bi_size); |
bf2de6f5 JA |
1087 | count_vm_events(PGPGIN, count); |
1088 | } | |
1089 | ||
1090 | if (unlikely(block_dump)) { | |
1091 | char b[BDEVNAME_SIZE]; | |
8dcbdc74 | 1092 | printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n", |
ba25f9dc | 1093 | current->comm, task_pid_nr(current), |
a8ebb056 | 1094 | op_is_write(bio_op(bio)) ? "WRITE" : "READ", |
4f024f37 | 1095 | (unsigned long long)bio->bi_iter.bi_sector, |
74d46992 | 1096 | bio_devname(bio, b), count); |
bf2de6f5 | 1097 | } |
1da177e4 LT |
1098 | } |
1099 | ||
b8e24a93 | 1100 | /* |
760f83ea CH |
1101 | * If we're reading data that is part of the userspace workingset, count |
1102 | * submission time as memory stall. When the device is congested, or | |
1103 | * the submitting cgroup IO-throttled, submission can be a significant | |
1104 | * part of overall IO time. | |
b8e24a93 | 1105 | */ |
760f83ea CH |
1106 | if (unlikely(bio_op(bio) == REQ_OP_READ && |
1107 | bio_flagged(bio, BIO_WORKINGSET))) { | |
1108 | unsigned long pflags; | |
1109 | blk_qc_t ret; | |
b8e24a93 | 1110 | |
760f83ea | 1111 | psi_memstall_enter(&pflags); |
ed00aabd | 1112 | ret = submit_bio_noacct(bio); |
b8e24a93 JW |
1113 | psi_memstall_leave(&pflags); |
1114 | ||
760f83ea CH |
1115 | return ret; |
1116 | } | |
1117 | ||
ed00aabd | 1118 | return submit_bio_noacct(bio); |
1da177e4 | 1119 | } |
1da177e4 LT |
1120 | EXPORT_SYMBOL(submit_bio); |
1121 | ||
82124d60 | 1122 | /** |
bf4e6b4e | 1123 | * blk_cloned_rq_check_limits - Helper function to check a cloned request |
0d720318 | 1124 | * for the new queue limits |
82124d60 KU |
1125 | * @q: the queue |
1126 | * @rq: the request being checked | |
1127 | * | |
1128 | * Description: | |
1129 | * @rq may have been made based on weaker limitations of upper-level queues | |
1130 | * in request stacking drivers, and it may violate the limitation of @q. | |
1131 | * Since the block layer and the underlying device driver trust @rq | |
1132 | * after it is inserted to @q, it should be checked against @q before | |
1133 | * the insertion using this generic function. | |
1134 | * | |
82124d60 | 1135 | * Request stacking drivers like request-based dm may change the queue |
bf4e6b4e HR |
1136 | * limits when retrying requests on other queues. Those requests need |
1137 | * to be checked against the new queue limits again during dispatch. | |
82124d60 | 1138 | */ |
143d2600 | 1139 | static blk_status_t blk_cloned_rq_check_limits(struct request_queue *q, |
bf4e6b4e | 1140 | struct request *rq) |
82124d60 | 1141 | { |
8327cce5 RS |
1142 | unsigned int max_sectors = blk_queue_get_max_sectors(q, req_op(rq)); |
1143 | ||
1144 | if (blk_rq_sectors(rq) > max_sectors) { | |
1145 | /* | |
1146 | * SCSI device does not have a good way to return if | |
1147 | * Write Same/Zero is actually supported. If a device rejects | |
1148 | * a non-read/write command (discard, write same,etc.) the | |
1149 | * low-level device driver will set the relevant queue limit to | |
1150 | * 0 to prevent blk-lib from issuing more of the offending | |
1151 | * operations. Commands queued prior to the queue limit being | |
1152 | * reset need to be completed with BLK_STS_NOTSUPP to avoid I/O | |
1153 | * errors being propagated to upper layers. | |
1154 | */ | |
1155 | if (max_sectors == 0) | |
1156 | return BLK_STS_NOTSUPP; | |
1157 | ||
61939b12 | 1158 | printk(KERN_ERR "%s: over max size limit. (%u > %u)\n", |
8327cce5 | 1159 | __func__, blk_rq_sectors(rq), max_sectors); |
143d2600 | 1160 | return BLK_STS_IOERR; |
82124d60 KU |
1161 | } |
1162 | ||
1163 | /* | |
1164 | * queue's settings related to segment counting like q->bounce_pfn | |
1165 | * may differ from that of other stacking queues. | |
1166 | * Recalculate it to check the request correctly on this queue's | |
1167 | * limitation. | |
1168 | */ | |
e9cd19c0 | 1169 | rq->nr_phys_segments = blk_recalc_rq_segments(rq); |
8a78362c | 1170 | if (rq->nr_phys_segments > queue_max_segments(q)) { |
61939b12 JP |
1171 | printk(KERN_ERR "%s: over max segments limit. (%hu > %hu)\n", |
1172 | __func__, rq->nr_phys_segments, queue_max_segments(q)); | |
143d2600 | 1173 | return BLK_STS_IOERR; |
82124d60 KU |
1174 | } |
1175 | ||
143d2600 | 1176 | return BLK_STS_OK; |
82124d60 | 1177 | } |
82124d60 KU |
1178 | |
1179 | /** | |
1180 | * blk_insert_cloned_request - Helper for stacking drivers to submit a request | |
1181 | * @q: the queue to submit the request | |
1182 | * @rq: the request being queued | |
1183 | */ | |
2a842aca | 1184 | blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq) |
82124d60 | 1185 | { |
8327cce5 RS |
1186 | blk_status_t ret; |
1187 | ||
1188 | ret = blk_cloned_rq_check_limits(q, rq); | |
1189 | if (ret != BLK_STS_OK) | |
1190 | return ret; | |
82124d60 | 1191 | |
b2c9cd37 | 1192 | if (rq->rq_disk && |
8446fe92 | 1193 | should_fail_request(rq->rq_disk->part0, blk_rq_bytes(rq))) |
2a842aca | 1194 | return BLK_STS_IOERR; |
82124d60 | 1195 | |
a892c8d5 ST |
1196 | if (blk_crypto_insert_cloned_request(rq)) |
1197 | return BLK_STS_IOERR; | |
1198 | ||
a1ce35fa | 1199 | if (blk_queue_io_stat(q)) |
b5af37ab | 1200 | blk_account_io_start(rq); |
82124d60 KU |
1201 | |
1202 | /* | |
a1ce35fa JA |
1203 | * Since we have a scheduler attached on the top device, |
1204 | * bypass a potential scheduler on the bottom device for | |
1205 | * insert. | |
82124d60 | 1206 | */ |
fd9c40f6 | 1207 | return blk_mq_request_issue_directly(rq, true); |
82124d60 KU |
1208 | } |
1209 | EXPORT_SYMBOL_GPL(blk_insert_cloned_request); | |
1210 | ||
80a761fd TH |
1211 | /** |
1212 | * blk_rq_err_bytes - determine number of bytes till the next failure boundary | |
1213 | * @rq: request to examine | |
1214 | * | |
1215 | * Description: | |
1216 | * A request could be merge of IOs which require different failure | |
1217 | * handling. This function determines the number of bytes which | |
1218 | * can be failed from the beginning of the request without | |
1219 | * crossing into area which need to be retried further. | |
1220 | * | |
1221 | * Return: | |
1222 | * The number of bytes to fail. | |
80a761fd TH |
1223 | */ |
1224 | unsigned int blk_rq_err_bytes(const struct request *rq) | |
1225 | { | |
1226 | unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK; | |
1227 | unsigned int bytes = 0; | |
1228 | struct bio *bio; | |
1229 | ||
e8064021 | 1230 | if (!(rq->rq_flags & RQF_MIXED_MERGE)) |
80a761fd TH |
1231 | return blk_rq_bytes(rq); |
1232 | ||
1233 | /* | |
1234 | * Currently the only 'mixing' which can happen is between | |
1235 | * different fastfail types. We can safely fail portions | |
1236 | * which have all the failfast bits that the first one has - | |
1237 | * the ones which are at least as eager to fail as the first | |
1238 | * one. | |
1239 | */ | |
1240 | for (bio = rq->bio; bio; bio = bio->bi_next) { | |
1eff9d32 | 1241 | if ((bio->bi_opf & ff) != ff) |
80a761fd | 1242 | break; |
4f024f37 | 1243 | bytes += bio->bi_iter.bi_size; |
80a761fd TH |
1244 | } |
1245 | ||
1246 | /* this could lead to infinite loop */ | |
1247 | BUG_ON(blk_rq_bytes(rq) && !bytes); | |
1248 | return bytes; | |
1249 | } | |
1250 | EXPORT_SYMBOL_GPL(blk_rq_err_bytes); | |
1251 | ||
8446fe92 CH |
1252 | static void update_io_ticks(struct block_device *part, unsigned long now, |
1253 | bool end) | |
9123bf6f CH |
1254 | { |
1255 | unsigned long stamp; | |
1256 | again: | |
8446fe92 | 1257 | stamp = READ_ONCE(part->bd_stamp); |
9123bf6f | 1258 | if (unlikely(stamp != now)) { |
8446fe92 | 1259 | if (likely(cmpxchg(&part->bd_stamp, stamp, now) == stamp)) |
9123bf6f CH |
1260 | __part_stat_add(part, io_ticks, end ? now - stamp : 1); |
1261 | } | |
8446fe92 CH |
1262 | if (part->bd_partno) { |
1263 | part = bdev_whole(part); | |
9123bf6f CH |
1264 | goto again; |
1265 | } | |
1266 | } | |
1267 | ||
f1394b79 | 1268 | static void blk_account_io_completion(struct request *req, unsigned int bytes) |
bc58ba94 | 1269 | { |
ecb6186c | 1270 | if (req->part && blk_do_io_stat(req)) { |
ddcf35d3 | 1271 | const int sgrp = op_stat_group(req_op(req)); |
bc58ba94 | 1272 | |
112f158f | 1273 | part_stat_lock(); |
8446fe92 | 1274 | part_stat_add(req->part, sectors[sgrp], bytes >> 9); |
bc58ba94 JA |
1275 | part_stat_unlock(); |
1276 | } | |
1277 | } | |
1278 | ||
522a7775 | 1279 | void blk_account_io_done(struct request *req, u64 now) |
bc58ba94 | 1280 | { |
bc58ba94 | 1281 | /* |
dd4c133f TH |
1282 | * Account IO completion. flush_rq isn't accounted as a |
1283 | * normal IO on queueing nor completion. Accounting the | |
1284 | * containing request is enough. | |
bc58ba94 | 1285 | */ |
ecb6186c LG |
1286 | if (req->part && blk_do_io_stat(req) && |
1287 | !(req->rq_flags & RQF_FLUSH_SEQ)) { | |
ddcf35d3 | 1288 | const int sgrp = op_stat_group(req_op(req)); |
bc58ba94 | 1289 | |
112f158f | 1290 | part_stat_lock(); |
8446fe92 CH |
1291 | update_io_ticks(req->part, jiffies, true); |
1292 | part_stat_inc(req->part, ios[sgrp]); | |
1293 | part_stat_add(req->part, nsecs[sgrp], now - req->start_time_ns); | |
524f9ffd | 1294 | part_stat_unlock(); |
bc58ba94 JA |
1295 | } |
1296 | } | |
1297 | ||
b5af37ab | 1298 | void blk_account_io_start(struct request *rq) |
320ae51f | 1299 | { |
320ae51f JA |
1300 | if (!blk_do_io_stat(rq)) |
1301 | return; | |
1302 | ||
0b6e522c CH |
1303 | /* passthrough requests can hold bios that do not have ->bi_bdev set */ |
1304 | if (rq->bio && rq->bio->bi_bdev) | |
1305 | rq->part = rq->bio->bi_bdev; | |
1306 | else | |
1307 | rq->part = rq->rq_disk->part0; | |
524f9ffd | 1308 | |
112f158f | 1309 | part_stat_lock(); |
76268f3a | 1310 | update_io_ticks(rq->part, jiffies, false); |
320ae51f JA |
1311 | part_stat_unlock(); |
1312 | } | |
320ae51f | 1313 | |
8446fe92 | 1314 | static unsigned long __part_start_io_acct(struct block_device *part, |
7b26410b | 1315 | unsigned int sectors, unsigned int op) |
956d510e | 1316 | { |
956d510e CH |
1317 | const int sgrp = op_stat_group(op); |
1318 | unsigned long now = READ_ONCE(jiffies); | |
1319 | ||
1320 | part_stat_lock(); | |
1321 | update_io_ticks(part, now, false); | |
1322 | part_stat_inc(part, ios[sgrp]); | |
1323 | part_stat_add(part, sectors[sgrp], sectors); | |
1324 | part_stat_local_inc(part, in_flight[op_is_write(op)]); | |
1325 | part_stat_unlock(); | |
320ae51f | 1326 | |
956d510e CH |
1327 | return now; |
1328 | } | |
7b26410b | 1329 | |
99dfc43e CH |
1330 | /** |
1331 | * bio_start_io_acct - start I/O accounting for bio based drivers | |
1332 | * @bio: bio to start account for | |
1333 | * | |
1334 | * Returns the start time that should be passed back to bio_end_io_acct(). | |
1335 | */ | |
1336 | unsigned long bio_start_io_acct(struct bio *bio) | |
7b26410b | 1337 | { |
99dfc43e | 1338 | return __part_start_io_acct(bio->bi_bdev, bio_sectors(bio), bio_op(bio)); |
7b26410b | 1339 | } |
99dfc43e | 1340 | EXPORT_SYMBOL_GPL(bio_start_io_acct); |
7b26410b SL |
1341 | |
1342 | unsigned long disk_start_io_acct(struct gendisk *disk, unsigned int sectors, | |
1343 | unsigned int op) | |
1344 | { | |
8446fe92 | 1345 | return __part_start_io_acct(disk->part0, sectors, op); |
7b26410b | 1346 | } |
956d510e CH |
1347 | EXPORT_SYMBOL(disk_start_io_acct); |
1348 | ||
8446fe92 | 1349 | static void __part_end_io_acct(struct block_device *part, unsigned int op, |
7b26410b | 1350 | unsigned long start_time) |
956d510e | 1351 | { |
956d510e CH |
1352 | const int sgrp = op_stat_group(op); |
1353 | unsigned long now = READ_ONCE(jiffies); | |
1354 | unsigned long duration = now - start_time; | |
5b18b5a7 | 1355 | |
956d510e CH |
1356 | part_stat_lock(); |
1357 | update_io_ticks(part, now, true); | |
1358 | part_stat_add(part, nsecs[sgrp], jiffies_to_nsecs(duration)); | |
1359 | part_stat_local_dec(part, in_flight[op_is_write(op)]); | |
320ae51f JA |
1360 | part_stat_unlock(); |
1361 | } | |
7b26410b | 1362 | |
99dfc43e CH |
1363 | void bio_end_io_acct_remapped(struct bio *bio, unsigned long start_time, |
1364 | struct block_device *orig_bdev) | |
7b26410b | 1365 | { |
99dfc43e | 1366 | __part_end_io_acct(orig_bdev, bio_op(bio), start_time); |
7b26410b | 1367 | } |
99dfc43e | 1368 | EXPORT_SYMBOL_GPL(bio_end_io_acct_remapped); |
7b26410b SL |
1369 | |
1370 | void disk_end_io_acct(struct gendisk *disk, unsigned int op, | |
1371 | unsigned long start_time) | |
1372 | { | |
8446fe92 | 1373 | __part_end_io_acct(disk->part0, op, start_time); |
7b26410b | 1374 | } |
956d510e | 1375 | EXPORT_SYMBOL(disk_end_io_acct); |
320ae51f | 1376 | |
ef71de8b CH |
1377 | /* |
1378 | * Steal bios from a request and add them to a bio list. | |
1379 | * The request must not have been partially completed before. | |
1380 | */ | |
1381 | void blk_steal_bios(struct bio_list *list, struct request *rq) | |
1382 | { | |
1383 | if (rq->bio) { | |
1384 | if (list->tail) | |
1385 | list->tail->bi_next = rq->bio; | |
1386 | else | |
1387 | list->head = rq->bio; | |
1388 | list->tail = rq->biotail; | |
1389 | ||
1390 | rq->bio = NULL; | |
1391 | rq->biotail = NULL; | |
1392 | } | |
1393 | ||
1394 | rq->__data_len = 0; | |
1395 | } | |
1396 | EXPORT_SYMBOL_GPL(blk_steal_bios); | |
1397 | ||
3bcddeac | 1398 | /** |
2e60e022 | 1399 | * blk_update_request - Special helper function for request stacking drivers |
8ebf9756 | 1400 | * @req: the request being processed |
2a842aca | 1401 | * @error: block status code |
8ebf9756 | 1402 | * @nr_bytes: number of bytes to complete @req |
3bcddeac KU |
1403 | * |
1404 | * Description: | |
8ebf9756 RD |
1405 | * Ends I/O on a number of bytes attached to @req, but doesn't complete |
1406 | * the request structure even if @req doesn't have leftover. | |
1407 | * If @req has leftover, sets it up for the next range of segments. | |
2e60e022 TH |
1408 | * |
1409 | * This special helper function is only for request stacking drivers | |
1410 | * (e.g. request-based dm) so that they can handle partial completion. | |
3a211b71 | 1411 | * Actual device drivers should use blk_mq_end_request instead. |
2e60e022 TH |
1412 | * |
1413 | * Passing the result of blk_rq_bytes() as @nr_bytes guarantees | |
1414 | * %false return from this function. | |
3bcddeac | 1415 | * |
1954e9a9 BVA |
1416 | * Note: |
1417 | * The RQF_SPECIAL_PAYLOAD flag is ignored on purpose in both | |
1418 | * blk_rq_bytes() and in blk_update_request(). | |
1419 | * | |
3bcddeac | 1420 | * Return: |
2e60e022 TH |
1421 | * %false - this request doesn't have any more data |
1422 | * %true - this request has more data | |
3bcddeac | 1423 | **/ |
2a842aca CH |
1424 | bool blk_update_request(struct request *req, blk_status_t error, |
1425 | unsigned int nr_bytes) | |
1da177e4 | 1426 | { |
f79ea416 | 1427 | int total_bytes; |
1da177e4 | 1428 | |
2a842aca | 1429 | trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes); |
4a0efdc9 | 1430 | |
2e60e022 TH |
1431 | if (!req->bio) |
1432 | return false; | |
1433 | ||
54d4e6ab MG |
1434 | #ifdef CONFIG_BLK_DEV_INTEGRITY |
1435 | if (blk_integrity_rq(req) && req_op(req) == REQ_OP_READ && | |
1436 | error == BLK_STS_OK) | |
1437 | req->q->integrity.profile->complete_fn(req, nr_bytes); | |
1438 | #endif | |
1439 | ||
2a842aca CH |
1440 | if (unlikely(error && !blk_rq_is_passthrough(req) && |
1441 | !(req->rq_flags & RQF_QUIET))) | |
178cc590 | 1442 | print_req_error(req, error, __func__); |
1da177e4 | 1443 | |
bc58ba94 | 1444 | blk_account_io_completion(req, nr_bytes); |
d72d904a | 1445 | |
f79ea416 KO |
1446 | total_bytes = 0; |
1447 | while (req->bio) { | |
1448 | struct bio *bio = req->bio; | |
4f024f37 | 1449 | unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes); |
1da177e4 | 1450 | |
9c24c10a | 1451 | if (bio_bytes == bio->bi_iter.bi_size) |
1da177e4 | 1452 | req->bio = bio->bi_next; |
1da177e4 | 1453 | |
fbbaf700 N |
1454 | /* Completion has already been traced */ |
1455 | bio_clear_flag(bio, BIO_TRACE_COMPLETION); | |
f79ea416 | 1456 | req_bio_endio(req, bio, bio_bytes, error); |
1da177e4 | 1457 | |
f79ea416 KO |
1458 | total_bytes += bio_bytes; |
1459 | nr_bytes -= bio_bytes; | |
1da177e4 | 1460 | |
f79ea416 KO |
1461 | if (!nr_bytes) |
1462 | break; | |
1da177e4 LT |
1463 | } |
1464 | ||
1465 | /* | |
1466 | * completely done | |
1467 | */ | |
2e60e022 TH |
1468 | if (!req->bio) { |
1469 | /* | |
1470 | * Reset counters so that the request stacking driver | |
1471 | * can find how many bytes remain in the request | |
1472 | * later. | |
1473 | */ | |
a2dec7b3 | 1474 | req->__data_len = 0; |
2e60e022 TH |
1475 | return false; |
1476 | } | |
1da177e4 | 1477 | |
a2dec7b3 | 1478 | req->__data_len -= total_bytes; |
2e46e8b2 TH |
1479 | |
1480 | /* update sector only for requests with clear definition of sector */ | |
57292b58 | 1481 | if (!blk_rq_is_passthrough(req)) |
a2dec7b3 | 1482 | req->__sector += total_bytes >> 9; |
2e46e8b2 | 1483 | |
80a761fd | 1484 | /* mixed attributes always follow the first bio */ |
e8064021 | 1485 | if (req->rq_flags & RQF_MIXED_MERGE) { |
80a761fd | 1486 | req->cmd_flags &= ~REQ_FAILFAST_MASK; |
1eff9d32 | 1487 | req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK; |
80a761fd TH |
1488 | } |
1489 | ||
ed6565e7 CH |
1490 | if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) { |
1491 | /* | |
1492 | * If total number of sectors is less than the first segment | |
1493 | * size, something has gone terribly wrong. | |
1494 | */ | |
1495 | if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) { | |
1496 | blk_dump_rq_flags(req, "request botched"); | |
1497 | req->__data_len = blk_rq_cur_bytes(req); | |
1498 | } | |
2e46e8b2 | 1499 | |
ed6565e7 | 1500 | /* recalculate the number of segments */ |
e9cd19c0 | 1501 | req->nr_phys_segments = blk_recalc_rq_segments(req); |
ed6565e7 | 1502 | } |
2e46e8b2 | 1503 | |
2e60e022 | 1504 | return true; |
1da177e4 | 1505 | } |
2e60e022 | 1506 | EXPORT_SYMBOL_GPL(blk_update_request); |
1da177e4 | 1507 | |
2d4dc890 IL |
1508 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE |
1509 | /** | |
1510 | * rq_flush_dcache_pages - Helper function to flush all pages in a request | |
1511 | * @rq: the request to be flushed | |
1512 | * | |
1513 | * Description: | |
1514 | * Flush all pages in @rq. | |
1515 | */ | |
1516 | void rq_flush_dcache_pages(struct request *rq) | |
1517 | { | |
1518 | struct req_iterator iter; | |
7988613b | 1519 | struct bio_vec bvec; |
2d4dc890 IL |
1520 | |
1521 | rq_for_each_segment(bvec, rq, iter) | |
7988613b | 1522 | flush_dcache_page(bvec.bv_page); |
2d4dc890 IL |
1523 | } |
1524 | EXPORT_SYMBOL_GPL(rq_flush_dcache_pages); | |
1525 | #endif | |
1526 | ||
ef9e3fac KU |
1527 | /** |
1528 | * blk_lld_busy - Check if underlying low-level drivers of a device are busy | |
1529 | * @q : the queue of the device being checked | |
1530 | * | |
1531 | * Description: | |
1532 | * Check if underlying low-level drivers of a device are busy. | |
1533 | * If the drivers want to export their busy state, they must set own | |
1534 | * exporting function using blk_queue_lld_busy() first. | |
1535 | * | |
1536 | * Basically, this function is used only by request stacking drivers | |
1537 | * to stop dispatching requests to underlying devices when underlying | |
1538 | * devices are busy. This behavior helps more I/O merging on the queue | |
1539 | * of the request stacking driver and prevents I/O throughput regression | |
1540 | * on burst I/O load. | |
1541 | * | |
1542 | * Return: | |
1543 | * 0 - Not busy (The request stacking driver should dispatch request) | |
1544 | * 1 - Busy (The request stacking driver should stop dispatching request) | |
1545 | */ | |
1546 | int blk_lld_busy(struct request_queue *q) | |
1547 | { | |
344e9ffc | 1548 | if (queue_is_mq(q) && q->mq_ops->busy) |
9ba20527 | 1549 | return q->mq_ops->busy(q); |
ef9e3fac KU |
1550 | |
1551 | return 0; | |
1552 | } | |
1553 | EXPORT_SYMBOL_GPL(blk_lld_busy); | |
1554 | ||
78d8e58a MS |
1555 | /** |
1556 | * blk_rq_unprep_clone - Helper function to free all bios in a cloned request | |
1557 | * @rq: the clone request to be cleaned up | |
1558 | * | |
1559 | * Description: | |
1560 | * Free all bios in @rq for a cloned request. | |
1561 | */ | |
1562 | void blk_rq_unprep_clone(struct request *rq) | |
1563 | { | |
1564 | struct bio *bio; | |
1565 | ||
1566 | while ((bio = rq->bio) != NULL) { | |
1567 | rq->bio = bio->bi_next; | |
1568 | ||
1569 | bio_put(bio); | |
1570 | } | |
1571 | } | |
1572 | EXPORT_SYMBOL_GPL(blk_rq_unprep_clone); | |
1573 | ||
78d8e58a MS |
1574 | /** |
1575 | * blk_rq_prep_clone - Helper function to setup clone request | |
1576 | * @rq: the request to be setup | |
1577 | * @rq_src: original request to be cloned | |
1578 | * @bs: bio_set that bios for clone are allocated from | |
1579 | * @gfp_mask: memory allocation mask for bio | |
1580 | * @bio_ctr: setup function to be called for each clone bio. | |
1581 | * Returns %0 for success, non %0 for failure. | |
1582 | * @data: private data to be passed to @bio_ctr | |
1583 | * | |
1584 | * Description: | |
1585 | * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq. | |
78d8e58a MS |
1586 | * Also, pages which the original bios are pointing to are not copied |
1587 | * and the cloned bios just point same pages. | |
1588 | * So cloned bios must be completed before original bios, which means | |
1589 | * the caller must complete @rq before @rq_src. | |
1590 | */ | |
1591 | int blk_rq_prep_clone(struct request *rq, struct request *rq_src, | |
1592 | struct bio_set *bs, gfp_t gfp_mask, | |
1593 | int (*bio_ctr)(struct bio *, struct bio *, void *), | |
1594 | void *data) | |
1595 | { | |
1596 | struct bio *bio, *bio_src; | |
1597 | ||
1598 | if (!bs) | |
f4f8154a | 1599 | bs = &fs_bio_set; |
78d8e58a MS |
1600 | |
1601 | __rq_for_each_bio(bio_src, rq_src) { | |
1602 | bio = bio_clone_fast(bio_src, gfp_mask, bs); | |
1603 | if (!bio) | |
1604 | goto free_and_out; | |
1605 | ||
1606 | if (bio_ctr && bio_ctr(bio, bio_src, data)) | |
1607 | goto free_and_out; | |
1608 | ||
1609 | if (rq->bio) { | |
1610 | rq->biotail->bi_next = bio; | |
1611 | rq->biotail = bio; | |
93f221ae | 1612 | } else { |
78d8e58a | 1613 | rq->bio = rq->biotail = bio; |
93f221ae EB |
1614 | } |
1615 | bio = NULL; | |
78d8e58a MS |
1616 | } |
1617 | ||
361301a2 GJ |
1618 | /* Copy attributes of the original request to the clone request. */ |
1619 | rq->__sector = blk_rq_pos(rq_src); | |
1620 | rq->__data_len = blk_rq_bytes(rq_src); | |
1621 | if (rq_src->rq_flags & RQF_SPECIAL_PAYLOAD) { | |
1622 | rq->rq_flags |= RQF_SPECIAL_PAYLOAD; | |
1623 | rq->special_vec = rq_src->special_vec; | |
1624 | } | |
1625 | rq->nr_phys_segments = rq_src->nr_phys_segments; | |
1626 | rq->ioprio = rq_src->ioprio; | |
78d8e58a | 1627 | |
93f221ae EB |
1628 | if (rq->bio && blk_crypto_rq_bio_prep(rq, rq->bio, gfp_mask) < 0) |
1629 | goto free_and_out; | |
78d8e58a MS |
1630 | |
1631 | return 0; | |
1632 | ||
1633 | free_and_out: | |
1634 | if (bio) | |
1635 | bio_put(bio); | |
1636 | blk_rq_unprep_clone(rq); | |
1637 | ||
1638 | return -ENOMEM; | |
b0fd271d KU |
1639 | } |
1640 | EXPORT_SYMBOL_GPL(blk_rq_prep_clone); | |
1641 | ||
59c3d45e | 1642 | int kblockd_schedule_work(struct work_struct *work) |
1da177e4 LT |
1643 | { |
1644 | return queue_work(kblockd_workqueue, work); | |
1645 | } | |
1da177e4 LT |
1646 | EXPORT_SYMBOL(kblockd_schedule_work); |
1647 | ||
818cd1cb JA |
1648 | int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork, |
1649 | unsigned long delay) | |
1650 | { | |
1651 | return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay); | |
1652 | } | |
1653 | EXPORT_SYMBOL(kblockd_mod_delayed_work_on); | |
1654 | ||
75df7136 SJ |
1655 | /** |
1656 | * blk_start_plug - initialize blk_plug and track it inside the task_struct | |
1657 | * @plug: The &struct blk_plug that needs to be initialized | |
1658 | * | |
1659 | * Description: | |
40405851 JM |
1660 | * blk_start_plug() indicates to the block layer an intent by the caller |
1661 | * to submit multiple I/O requests in a batch. The block layer may use | |
1662 | * this hint to defer submitting I/Os from the caller until blk_finish_plug() | |
1663 | * is called. However, the block layer may choose to submit requests | |
1664 | * before a call to blk_finish_plug() if the number of queued I/Os | |
1665 | * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than | |
1666 | * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if | |
1667 | * the task schedules (see below). | |
1668 | * | |
75df7136 SJ |
1669 | * Tracking blk_plug inside the task_struct will help with auto-flushing the |
1670 | * pending I/O should the task end up blocking between blk_start_plug() and | |
1671 | * blk_finish_plug(). This is important from a performance perspective, but | |
1672 | * also ensures that we don't deadlock. For instance, if the task is blocking | |
1673 | * for a memory allocation, memory reclaim could end up wanting to free a | |
1674 | * page belonging to that request that is currently residing in our private | |
1675 | * plug. By flushing the pending I/O when the process goes to sleep, we avoid | |
1676 | * this kind of deadlock. | |
1677 | */ | |
73c10101 JA |
1678 | void blk_start_plug(struct blk_plug *plug) |
1679 | { | |
1680 | struct task_struct *tsk = current; | |
1681 | ||
dd6cf3e1 SL |
1682 | /* |
1683 | * If this is a nested plug, don't actually assign it. | |
1684 | */ | |
1685 | if (tsk->plug) | |
1686 | return; | |
1687 | ||
320ae51f | 1688 | INIT_LIST_HEAD(&plug->mq_list); |
048c9374 | 1689 | INIT_LIST_HEAD(&plug->cb_list); |
5f0ed774 | 1690 | plug->rq_count = 0; |
ce5b009c | 1691 | plug->multiple_queues = false; |
5a473e83 | 1692 | plug->nowait = false; |
5f0ed774 | 1693 | |
73c10101 | 1694 | /* |
dd6cf3e1 SL |
1695 | * Store ordering should not be needed here, since a potential |
1696 | * preempt will imply a full memory barrier | |
73c10101 | 1697 | */ |
dd6cf3e1 | 1698 | tsk->plug = plug; |
73c10101 JA |
1699 | } |
1700 | EXPORT_SYMBOL(blk_start_plug); | |
1701 | ||
74018dc3 | 1702 | static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule) |
048c9374 N |
1703 | { |
1704 | LIST_HEAD(callbacks); | |
1705 | ||
2a7d5559 SL |
1706 | while (!list_empty(&plug->cb_list)) { |
1707 | list_splice_init(&plug->cb_list, &callbacks); | |
048c9374 | 1708 | |
2a7d5559 SL |
1709 | while (!list_empty(&callbacks)) { |
1710 | struct blk_plug_cb *cb = list_first_entry(&callbacks, | |
048c9374 N |
1711 | struct blk_plug_cb, |
1712 | list); | |
2a7d5559 | 1713 | list_del(&cb->list); |
74018dc3 | 1714 | cb->callback(cb, from_schedule); |
2a7d5559 | 1715 | } |
048c9374 N |
1716 | } |
1717 | } | |
1718 | ||
9cbb1750 N |
1719 | struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data, |
1720 | int size) | |
1721 | { | |
1722 | struct blk_plug *plug = current->plug; | |
1723 | struct blk_plug_cb *cb; | |
1724 | ||
1725 | if (!plug) | |
1726 | return NULL; | |
1727 | ||
1728 | list_for_each_entry(cb, &plug->cb_list, list) | |
1729 | if (cb->callback == unplug && cb->data == data) | |
1730 | return cb; | |
1731 | ||
1732 | /* Not currently on the callback list */ | |
1733 | BUG_ON(size < sizeof(*cb)); | |
1734 | cb = kzalloc(size, GFP_ATOMIC); | |
1735 | if (cb) { | |
1736 | cb->data = data; | |
1737 | cb->callback = unplug; | |
1738 | list_add(&cb->list, &plug->cb_list); | |
1739 | } | |
1740 | return cb; | |
1741 | } | |
1742 | EXPORT_SYMBOL(blk_check_plugged); | |
1743 | ||
49cac01e | 1744 | void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule) |
73c10101 | 1745 | { |
74018dc3 | 1746 | flush_plug_callbacks(plug, from_schedule); |
320ae51f JA |
1747 | |
1748 | if (!list_empty(&plug->mq_list)) | |
1749 | blk_mq_flush_plug_list(plug, from_schedule); | |
73c10101 | 1750 | } |
73c10101 | 1751 | |
40405851 JM |
1752 | /** |
1753 | * blk_finish_plug - mark the end of a batch of submitted I/O | |
1754 | * @plug: The &struct blk_plug passed to blk_start_plug() | |
1755 | * | |
1756 | * Description: | |
1757 | * Indicate that a batch of I/O submissions is complete. This function | |
1758 | * must be paired with an initial call to blk_start_plug(). The intent | |
1759 | * is to allow the block layer to optimize I/O submission. See the | |
1760 | * documentation for blk_start_plug() for more information. | |
1761 | */ | |
73c10101 JA |
1762 | void blk_finish_plug(struct blk_plug *plug) |
1763 | { | |
dd6cf3e1 SL |
1764 | if (plug != current->plug) |
1765 | return; | |
f6603783 | 1766 | blk_flush_plug_list(plug, false); |
73c10101 | 1767 | |
dd6cf3e1 | 1768 | current->plug = NULL; |
73c10101 | 1769 | } |
88b996cd | 1770 | EXPORT_SYMBOL(blk_finish_plug); |
73c10101 | 1771 | |
71ac860a ML |
1772 | void blk_io_schedule(void) |
1773 | { | |
1774 | /* Prevent hang_check timer from firing at us during very long I/O */ | |
1775 | unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2; | |
1776 | ||
1777 | if (timeout) | |
1778 | io_schedule_timeout(timeout); | |
1779 | else | |
1780 | io_schedule(); | |
1781 | } | |
1782 | EXPORT_SYMBOL_GPL(blk_io_schedule); | |
1783 | ||
1da177e4 LT |
1784 | int __init blk_dev_init(void) |
1785 | { | |
ef295ecf CH |
1786 | BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS)); |
1787 | BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * | |
c593642c | 1788 | sizeof_field(struct request, cmd_flags)); |
ef295ecf | 1789 | BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * |
c593642c | 1790 | sizeof_field(struct bio, bi_opf)); |
9eb55b03 | 1791 | |
89b90be2 TH |
1792 | /* used for unplugging and affects IO latency/throughput - HIGHPRI */ |
1793 | kblockd_workqueue = alloc_workqueue("kblockd", | |
28747fcd | 1794 | WQ_MEM_RECLAIM | WQ_HIGHPRI, 0); |
1da177e4 LT |
1795 | if (!kblockd_workqueue) |
1796 | panic("Failed to create kblockd\n"); | |
1797 | ||
c2789bd4 | 1798 | blk_requestq_cachep = kmem_cache_create("request_queue", |
165125e1 | 1799 | sizeof(struct request_queue), 0, SLAB_PANIC, NULL); |
1da177e4 | 1800 | |
18fbda91 | 1801 | blk_debugfs_root = debugfs_create_dir("block", NULL); |
18fbda91 | 1802 | |
d38ecf93 | 1803 | return 0; |
1da177e4 | 1804 | } |