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> |
1da177e4 LT |
21 | #include <linux/highmem.h> |
22 | #include <linux/mm.h> | |
23 | #include <linux/kernel_stat.h> | |
24 | #include <linux/string.h> | |
25 | #include <linux/init.h> | |
1da177e4 LT |
26 | #include <linux/completion.h> |
27 | #include <linux/slab.h> | |
28 | #include <linux/swap.h> | |
29 | #include <linux/writeback.h> | |
faccbd4b | 30 | #include <linux/task_io_accounting_ops.h> |
c17bb495 | 31 | #include <linux/fault-inject.h> |
73c10101 | 32 | #include <linux/list_sort.h> |
e3c78ca5 | 33 | #include <linux/delay.h> |
aaf7c680 | 34 | #include <linux/ratelimit.h> |
6c954667 | 35 | #include <linux/pm_runtime.h> |
eea8f41c | 36 | #include <linux/blk-cgroup.h> |
18fbda91 | 37 | #include <linux/debugfs.h> |
30abb3a6 | 38 | #include <linux/bpf.h> |
55782138 LZ |
39 | |
40 | #define CREATE_TRACE_POINTS | |
41 | #include <trace/events/block.h> | |
1da177e4 | 42 | |
8324aa91 | 43 | #include "blk.h" |
43a5e4e2 | 44 | #include "blk-mq.h" |
bd166ef1 | 45 | #include "blk-mq-sched.h" |
bca6b067 | 46 | #include "blk-pm.h" |
c1c80384 | 47 | #include "blk-rq-qos.h" |
8324aa91 | 48 | |
18fbda91 OS |
49 | #ifdef CONFIG_DEBUG_FS |
50 | struct dentry *blk_debugfs_root; | |
51 | #endif | |
52 | ||
d07335e5 | 53 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap); |
b0da3f0d | 54 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap); |
0a82a8d1 | 55 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete); |
3291fa57 | 56 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_split); |
cbae8d45 | 57 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug); |
0bfc2455 | 58 | |
a73f730d TH |
59 | DEFINE_IDA(blk_queue_ida); |
60 | ||
1da177e4 LT |
61 | /* |
62 | * For queue allocation | |
63 | */ | |
6728cb0e | 64 | struct kmem_cache *blk_requestq_cachep; |
1da177e4 | 65 | |
1da177e4 LT |
66 | /* |
67 | * Controlling structure to kblockd | |
68 | */ | |
ff856bad | 69 | static struct workqueue_struct *kblockd_workqueue; |
1da177e4 | 70 | |
8814ce8a BVA |
71 | /** |
72 | * blk_queue_flag_set - atomically set a queue flag | |
73 | * @flag: flag to be set | |
74 | * @q: request queue | |
75 | */ | |
76 | void blk_queue_flag_set(unsigned int flag, struct request_queue *q) | |
77 | { | |
57d74df9 | 78 | set_bit(flag, &q->queue_flags); |
8814ce8a BVA |
79 | } |
80 | EXPORT_SYMBOL(blk_queue_flag_set); | |
81 | ||
82 | /** | |
83 | * blk_queue_flag_clear - atomically clear a queue flag | |
84 | * @flag: flag to be cleared | |
85 | * @q: request queue | |
86 | */ | |
87 | void blk_queue_flag_clear(unsigned int flag, struct request_queue *q) | |
88 | { | |
57d74df9 | 89 | clear_bit(flag, &q->queue_flags); |
8814ce8a BVA |
90 | } |
91 | EXPORT_SYMBOL(blk_queue_flag_clear); | |
92 | ||
93 | /** | |
94 | * blk_queue_flag_test_and_set - atomically test and set a queue flag | |
95 | * @flag: flag to be set | |
96 | * @q: request queue | |
97 | * | |
98 | * Returns the previous value of @flag - 0 if the flag was not set and 1 if | |
99 | * the flag was already set. | |
100 | */ | |
101 | bool blk_queue_flag_test_and_set(unsigned int flag, struct request_queue *q) | |
102 | { | |
57d74df9 | 103 | return test_and_set_bit(flag, &q->queue_flags); |
8814ce8a BVA |
104 | } |
105 | EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set); | |
106 | ||
2a4aa30c | 107 | void blk_rq_init(struct request_queue *q, struct request *rq) |
1da177e4 | 108 | { |
1afb20f3 FT |
109 | memset(rq, 0, sizeof(*rq)); |
110 | ||
1da177e4 | 111 | INIT_LIST_HEAD(&rq->queuelist); |
63a71386 | 112 | rq->q = q; |
a2dec7b3 | 113 | rq->__sector = (sector_t) -1; |
2e662b65 JA |
114 | INIT_HLIST_NODE(&rq->hash); |
115 | RB_CLEAR_NODE(&rq->rb_node); | |
63a71386 | 116 | rq->tag = -1; |
bd166ef1 | 117 | rq->internal_tag = -1; |
522a7775 | 118 | rq->start_time_ns = ktime_get_ns(); |
09e099d4 | 119 | rq->part = NULL; |
b554db14 | 120 | refcount_set(&rq->ref, 1); |
1da177e4 | 121 | } |
2a4aa30c | 122 | EXPORT_SYMBOL(blk_rq_init); |
1da177e4 | 123 | |
e47bc4ed CK |
124 | #define REQ_OP_NAME(name) [REQ_OP_##name] = #name |
125 | static const char *const blk_op_name[] = { | |
126 | REQ_OP_NAME(READ), | |
127 | REQ_OP_NAME(WRITE), | |
128 | REQ_OP_NAME(FLUSH), | |
129 | REQ_OP_NAME(DISCARD), | |
130 | REQ_OP_NAME(SECURE_ERASE), | |
131 | REQ_OP_NAME(ZONE_RESET), | |
132 | REQ_OP_NAME(WRITE_SAME), | |
133 | REQ_OP_NAME(WRITE_ZEROES), | |
134 | REQ_OP_NAME(SCSI_IN), | |
135 | REQ_OP_NAME(SCSI_OUT), | |
136 | REQ_OP_NAME(DRV_IN), | |
137 | REQ_OP_NAME(DRV_OUT), | |
138 | }; | |
139 | #undef REQ_OP_NAME | |
140 | ||
141 | /** | |
142 | * blk_op_str - Return string XXX in the REQ_OP_XXX. | |
143 | * @op: REQ_OP_XXX. | |
144 | * | |
145 | * Description: Centralize block layer function to convert REQ_OP_XXX into | |
146 | * string format. Useful in the debugging and tracing bio or request. For | |
147 | * invalid REQ_OP_XXX it returns string "UNKNOWN". | |
148 | */ | |
149 | inline const char *blk_op_str(unsigned int op) | |
150 | { | |
151 | const char *op_str = "UNKNOWN"; | |
152 | ||
153 | if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op]) | |
154 | op_str = blk_op_name[op]; | |
155 | ||
156 | return op_str; | |
157 | } | |
158 | EXPORT_SYMBOL_GPL(blk_op_str); | |
159 | ||
2a842aca CH |
160 | static const struct { |
161 | int errno; | |
162 | const char *name; | |
163 | } blk_errors[] = { | |
164 | [BLK_STS_OK] = { 0, "" }, | |
165 | [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" }, | |
166 | [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" }, | |
167 | [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" }, | |
168 | [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" }, | |
169 | [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" }, | |
170 | [BLK_STS_NEXUS] = { -EBADE, "critical nexus" }, | |
171 | [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" }, | |
172 | [BLK_STS_PROTECTION] = { -EILSEQ, "protection" }, | |
173 | [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" }, | |
86ff7c2a | 174 | [BLK_STS_DEV_RESOURCE] = { -EBUSY, "device resource" }, |
03a07c92 | 175 | [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" }, |
2a842aca | 176 | |
4e4cbee9 CH |
177 | /* device mapper special case, should not leak out: */ |
178 | [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" }, | |
179 | ||
2a842aca CH |
180 | /* everything else not covered above: */ |
181 | [BLK_STS_IOERR] = { -EIO, "I/O" }, | |
182 | }; | |
183 | ||
184 | blk_status_t errno_to_blk_status(int errno) | |
185 | { | |
186 | int i; | |
187 | ||
188 | for (i = 0; i < ARRAY_SIZE(blk_errors); i++) { | |
189 | if (blk_errors[i].errno == errno) | |
190 | return (__force blk_status_t)i; | |
191 | } | |
192 | ||
193 | return BLK_STS_IOERR; | |
194 | } | |
195 | EXPORT_SYMBOL_GPL(errno_to_blk_status); | |
196 | ||
197 | int blk_status_to_errno(blk_status_t status) | |
198 | { | |
199 | int idx = (__force int)status; | |
200 | ||
34bd9c1c | 201 | if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors))) |
2a842aca CH |
202 | return -EIO; |
203 | return blk_errors[idx].errno; | |
204 | } | |
205 | EXPORT_SYMBOL_GPL(blk_status_to_errno); | |
206 | ||
178cc590 CH |
207 | static void print_req_error(struct request *req, blk_status_t status, |
208 | const char *caller) | |
2a842aca CH |
209 | { |
210 | int idx = (__force int)status; | |
211 | ||
34bd9c1c | 212 | if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors))) |
2a842aca CH |
213 | return; |
214 | ||
178cc590 | 215 | printk_ratelimited(KERN_ERR |
b0e5168a CK |
216 | "%s: %s error, dev %s, sector %llu op 0x%x:(%s) flags 0x%x " |
217 | "phys_seg %u prio class %u\n", | |
178cc590 | 218 | caller, blk_errors[idx].name, |
b0e5168a CK |
219 | req->rq_disk ? req->rq_disk->disk_name : "?", |
220 | blk_rq_pos(req), req_op(req), blk_op_str(req_op(req)), | |
221 | req->cmd_flags & ~REQ_OP_MASK, | |
222 | req->nr_phys_segments, | |
223 | IOPRIO_PRIO_CLASS(req->ioprio)); | |
2a842aca CH |
224 | } |
225 | ||
5bb23a68 | 226 | static void req_bio_endio(struct request *rq, struct bio *bio, |
2a842aca | 227 | unsigned int nbytes, blk_status_t error) |
1da177e4 | 228 | { |
78d8e58a | 229 | if (error) |
4e4cbee9 | 230 | bio->bi_status = error; |
797e7dbb | 231 | |
e8064021 | 232 | if (unlikely(rq->rq_flags & RQF_QUIET)) |
b7c44ed9 | 233 | bio_set_flag(bio, BIO_QUIET); |
08bafc03 | 234 | |
f79ea416 | 235 | bio_advance(bio, nbytes); |
7ba1ba12 | 236 | |
143a87f4 | 237 | /* don't actually finish bio if it's part of flush sequence */ |
e8064021 | 238 | if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ)) |
4246a0b6 | 239 | bio_endio(bio); |
1da177e4 | 240 | } |
1da177e4 | 241 | |
1da177e4 LT |
242 | void blk_dump_rq_flags(struct request *rq, char *msg) |
243 | { | |
aebf526b CH |
244 | printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg, |
245 | rq->rq_disk ? rq->rq_disk->disk_name : "?", | |
5953316d | 246 | (unsigned long long) rq->cmd_flags); |
1da177e4 | 247 | |
83096ebf TH |
248 | printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n", |
249 | (unsigned long long)blk_rq_pos(rq), | |
250 | blk_rq_sectors(rq), blk_rq_cur_sectors(rq)); | |
b4f42e28 JA |
251 | printk(KERN_INFO " bio %p, biotail %p, len %u\n", |
252 | rq->bio, rq->biotail, blk_rq_bytes(rq)); | |
1da177e4 | 253 | } |
1da177e4 LT |
254 | EXPORT_SYMBOL(blk_dump_rq_flags); |
255 | ||
1da177e4 LT |
256 | /** |
257 | * blk_sync_queue - cancel any pending callbacks on a queue | |
258 | * @q: the queue | |
259 | * | |
260 | * Description: | |
261 | * The block layer may perform asynchronous callback activity | |
262 | * on a queue, such as calling the unplug function after a timeout. | |
263 | * A block device may call blk_sync_queue to ensure that any | |
264 | * such activity is cancelled, thus allowing it to release resources | |
59c51591 | 265 | * that the callbacks might use. The caller must already have made sure |
1da177e4 LT |
266 | * that its ->make_request_fn will not re-add plugging prior to calling |
267 | * this function. | |
268 | * | |
da527770 | 269 | * This function does not cancel any asynchronous activity arising |
da3dae54 | 270 | * out of elevator or throttling code. That would require elevator_exit() |
5efd6113 | 271 | * and blkcg_exit_queue() to be called with queue lock initialized. |
da527770 | 272 | * |
1da177e4 LT |
273 | */ |
274 | void blk_sync_queue(struct request_queue *q) | |
275 | { | |
70ed28b9 | 276 | del_timer_sync(&q->timeout); |
4e9b6f20 | 277 | cancel_work_sync(&q->timeout_work); |
1da177e4 LT |
278 | } |
279 | EXPORT_SYMBOL(blk_sync_queue); | |
280 | ||
c9254f2d | 281 | /** |
cd84a62e | 282 | * blk_set_pm_only - increment pm_only counter |
c9254f2d | 283 | * @q: request queue pointer |
c9254f2d | 284 | */ |
cd84a62e | 285 | void blk_set_pm_only(struct request_queue *q) |
c9254f2d | 286 | { |
cd84a62e | 287 | atomic_inc(&q->pm_only); |
c9254f2d | 288 | } |
cd84a62e | 289 | EXPORT_SYMBOL_GPL(blk_set_pm_only); |
c9254f2d | 290 | |
cd84a62e | 291 | void blk_clear_pm_only(struct request_queue *q) |
c9254f2d | 292 | { |
cd84a62e BVA |
293 | int pm_only; |
294 | ||
295 | pm_only = atomic_dec_return(&q->pm_only); | |
296 | WARN_ON_ONCE(pm_only < 0); | |
297 | if (pm_only == 0) | |
298 | wake_up_all(&q->mq_freeze_wq); | |
c9254f2d | 299 | } |
cd84a62e | 300 | EXPORT_SYMBOL_GPL(blk_clear_pm_only); |
c9254f2d | 301 | |
165125e1 | 302 | void blk_put_queue(struct request_queue *q) |
483f4afc AV |
303 | { |
304 | kobject_put(&q->kobj); | |
305 | } | |
d86e0e83 | 306 | EXPORT_SYMBOL(blk_put_queue); |
483f4afc | 307 | |
aed3ea94 JA |
308 | void blk_set_queue_dying(struct request_queue *q) |
309 | { | |
8814ce8a | 310 | blk_queue_flag_set(QUEUE_FLAG_DYING, q); |
aed3ea94 | 311 | |
d3cfb2a0 ML |
312 | /* |
313 | * When queue DYING flag is set, we need to block new req | |
314 | * entering queue, so we call blk_freeze_queue_start() to | |
315 | * prevent I/O from crossing blk_queue_enter(). | |
316 | */ | |
317 | blk_freeze_queue_start(q); | |
318 | ||
344e9ffc | 319 | if (queue_is_mq(q)) |
aed3ea94 | 320 | blk_mq_wake_waiters(q); |
055f6e18 ML |
321 | |
322 | /* Make blk_queue_enter() reexamine the DYING flag. */ | |
323 | wake_up_all(&q->mq_freeze_wq); | |
aed3ea94 JA |
324 | } |
325 | EXPORT_SYMBOL_GPL(blk_set_queue_dying); | |
326 | ||
c9a929dd TH |
327 | /** |
328 | * blk_cleanup_queue - shutdown a request queue | |
329 | * @q: request queue to shutdown | |
330 | * | |
c246e80d BVA |
331 | * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and |
332 | * put it. All future requests will be failed immediately with -ENODEV. | |
c94a96ac | 333 | */ |
6728cb0e | 334 | void blk_cleanup_queue(struct request_queue *q) |
483f4afc | 335 | { |
3f3299d5 | 336 | /* mark @q DYING, no new request or merges will be allowed afterwards */ |
483f4afc | 337 | mutex_lock(&q->sysfs_lock); |
aed3ea94 | 338 | blk_set_queue_dying(q); |
6ecf23af | 339 | |
57d74df9 CH |
340 | blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q); |
341 | blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q); | |
342 | blk_queue_flag_set(QUEUE_FLAG_DYING, q); | |
c9a929dd TH |
343 | mutex_unlock(&q->sysfs_lock); |
344 | ||
c246e80d BVA |
345 | /* |
346 | * Drain all requests queued before DYING marking. Set DEAD flag to | |
347 | * prevent that q->request_fn() gets invoked after draining finished. | |
348 | */ | |
3ef28e83 | 349 | blk_freeze_queue(q); |
c57cdf7a ML |
350 | |
351 | rq_qos_exit(q); | |
352 | ||
57d74df9 | 353 | blk_queue_flag_set(QUEUE_FLAG_DEAD, q); |
c9a929dd | 354 | |
5a48fc14 DW |
355 | /* for synchronous bio-based driver finish in-flight integrity i/o */ |
356 | blk_flush_integrity(); | |
357 | ||
c9a929dd | 358 | /* @q won't process any more request, flush async actions */ |
dc3b17cc | 359 | del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer); |
c9a929dd TH |
360 | blk_sync_queue(q); |
361 | ||
344e9ffc | 362 | if (queue_is_mq(q)) |
c7e2d94b | 363 | blk_mq_exit_queue(q); |
a1ce35fa | 364 | |
c3e22192 ML |
365 | /* |
366 | * In theory, request pool of sched_tags belongs to request queue. | |
367 | * However, the current implementation requires tag_set for freeing | |
368 | * requests, so free the pool now. | |
369 | * | |
370 | * Queue has become frozen, there can't be any in-queue requests, so | |
371 | * it is safe to free requests now. | |
372 | */ | |
373 | mutex_lock(&q->sysfs_lock); | |
374 | if (q->elevator) | |
375 | blk_mq_sched_free_requests(q); | |
376 | mutex_unlock(&q->sysfs_lock); | |
377 | ||
3ef28e83 | 378 | percpu_ref_exit(&q->q_usage_counter); |
45a9c9d9 | 379 | |
c9a929dd | 380 | /* @q is and will stay empty, shutdown and put */ |
483f4afc AV |
381 | blk_put_queue(q); |
382 | } | |
1da177e4 LT |
383 | EXPORT_SYMBOL(blk_cleanup_queue); |
384 | ||
165125e1 | 385 | struct request_queue *blk_alloc_queue(gfp_t gfp_mask) |
1da177e4 | 386 | { |
6d469642 | 387 | return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE); |
1946089a CL |
388 | } |
389 | EXPORT_SYMBOL(blk_alloc_queue); | |
1da177e4 | 390 | |
3a0a5299 BVA |
391 | /** |
392 | * blk_queue_enter() - try to increase q->q_usage_counter | |
393 | * @q: request queue pointer | |
394 | * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PREEMPT | |
395 | */ | |
9a95e4ef | 396 | int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags) |
3ef28e83 | 397 | { |
cd84a62e | 398 | const bool pm = flags & BLK_MQ_REQ_PREEMPT; |
3a0a5299 | 399 | |
3ef28e83 | 400 | while (true) { |
3a0a5299 | 401 | bool success = false; |
3ef28e83 | 402 | |
818e0fa2 | 403 | rcu_read_lock(); |
3a0a5299 BVA |
404 | if (percpu_ref_tryget_live(&q->q_usage_counter)) { |
405 | /* | |
cd84a62e BVA |
406 | * The code that increments the pm_only counter is |
407 | * responsible for ensuring that that counter is | |
408 | * globally visible before the queue is unfrozen. | |
3a0a5299 | 409 | */ |
cd84a62e | 410 | if (pm || !blk_queue_pm_only(q)) { |
3a0a5299 BVA |
411 | success = true; |
412 | } else { | |
413 | percpu_ref_put(&q->q_usage_counter); | |
414 | } | |
415 | } | |
818e0fa2 | 416 | rcu_read_unlock(); |
3a0a5299 BVA |
417 | |
418 | if (success) | |
3ef28e83 DW |
419 | return 0; |
420 | ||
3a0a5299 | 421 | if (flags & BLK_MQ_REQ_NOWAIT) |
3ef28e83 DW |
422 | return -EBUSY; |
423 | ||
5ed61d3f | 424 | /* |
1671d522 | 425 | * read pair of barrier in blk_freeze_queue_start(), |
5ed61d3f | 426 | * we need to order reading __PERCPU_REF_DEAD flag of |
d3cfb2a0 ML |
427 | * .q_usage_counter and reading .mq_freeze_depth or |
428 | * queue dying flag, otherwise the following wait may | |
429 | * never return if the two reads are reordered. | |
5ed61d3f ML |
430 | */ |
431 | smp_rmb(); | |
432 | ||
1dc3039b | 433 | wait_event(q->mq_freeze_wq, |
7996a8b5 | 434 | (!q->mq_freeze_depth && |
0d25bd07 BVA |
435 | (pm || (blk_pm_request_resume(q), |
436 | !blk_queue_pm_only(q)))) || | |
1dc3039b | 437 | blk_queue_dying(q)); |
3ef28e83 DW |
438 | if (blk_queue_dying(q)) |
439 | return -ENODEV; | |
3ef28e83 DW |
440 | } |
441 | } | |
442 | ||
443 | void blk_queue_exit(struct request_queue *q) | |
444 | { | |
445 | percpu_ref_put(&q->q_usage_counter); | |
446 | } | |
447 | ||
448 | static void blk_queue_usage_counter_release(struct percpu_ref *ref) | |
449 | { | |
450 | struct request_queue *q = | |
451 | container_of(ref, struct request_queue, q_usage_counter); | |
452 | ||
453 | wake_up_all(&q->mq_freeze_wq); | |
454 | } | |
455 | ||
bca237a5 | 456 | static void blk_rq_timed_out_timer(struct timer_list *t) |
287922eb | 457 | { |
bca237a5 | 458 | struct request_queue *q = from_timer(q, t, timeout); |
287922eb CH |
459 | |
460 | kblockd_schedule_work(&q->timeout_work); | |
461 | } | |
462 | ||
2e3c18d0 TH |
463 | static void blk_timeout_work(struct work_struct *work) |
464 | { | |
465 | } | |
466 | ||
498f6650 BVA |
467 | /** |
468 | * blk_alloc_queue_node - allocate a request queue | |
469 | * @gfp_mask: memory allocation flags | |
470 | * @node_id: NUMA node to allocate memory from | |
498f6650 | 471 | */ |
6d469642 | 472 | struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id) |
1946089a | 473 | { |
165125e1 | 474 | struct request_queue *q; |
338aa96d | 475 | int ret; |
1946089a | 476 | |
8324aa91 | 477 | q = kmem_cache_alloc_node(blk_requestq_cachep, |
94f6030c | 478 | gfp_mask | __GFP_ZERO, node_id); |
1da177e4 LT |
479 | if (!q) |
480 | return NULL; | |
481 | ||
cbf62af3 CH |
482 | INIT_LIST_HEAD(&q->queue_head); |
483 | q->last_merge = NULL; | |
cbf62af3 | 484 | |
00380a40 | 485 | q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask); |
a73f730d | 486 | if (q->id < 0) |
3d2936f4 | 487 | goto fail_q; |
a73f730d | 488 | |
338aa96d KO |
489 | ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS); |
490 | if (ret) | |
54efd50b KO |
491 | goto fail_id; |
492 | ||
d03f6cdc JK |
493 | q->backing_dev_info = bdi_alloc_node(gfp_mask, node_id); |
494 | if (!q->backing_dev_info) | |
495 | goto fail_split; | |
496 | ||
a83b576c JA |
497 | q->stats = blk_alloc_queue_stats(); |
498 | if (!q->stats) | |
499 | goto fail_stats; | |
500 | ||
b5420237 | 501 | q->backing_dev_info->ra_pages = VM_READAHEAD_PAGES; |
dc3b17cc JK |
502 | q->backing_dev_info->capabilities = BDI_CAP_CGROUP_WRITEBACK; |
503 | q->backing_dev_info->name = "block"; | |
5151412d | 504 | q->node = node_id; |
0989a025 | 505 | |
bca237a5 KC |
506 | timer_setup(&q->backing_dev_info->laptop_mode_wb_timer, |
507 | laptop_mode_timer_fn, 0); | |
508 | timer_setup(&q->timeout, blk_rq_timed_out_timer, 0); | |
2e3c18d0 | 509 | INIT_WORK(&q->timeout_work, blk_timeout_work); |
a612fddf | 510 | INIT_LIST_HEAD(&q->icq_list); |
4eef3049 | 511 | #ifdef CONFIG_BLK_CGROUP |
e8989fae | 512 | INIT_LIST_HEAD(&q->blkg_list); |
4eef3049 | 513 | #endif |
483f4afc | 514 | |
8324aa91 | 515 | kobject_init(&q->kobj, &blk_queue_ktype); |
1da177e4 | 516 | |
5acb3cc2 WL |
517 | #ifdef CONFIG_BLK_DEV_IO_TRACE |
518 | mutex_init(&q->blk_trace_mutex); | |
519 | #endif | |
483f4afc | 520 | mutex_init(&q->sysfs_lock); |
0d945c1f | 521 | spin_lock_init(&q->queue_lock); |
c94a96ac | 522 | |
320ae51f | 523 | init_waitqueue_head(&q->mq_freeze_wq); |
7996a8b5 | 524 | mutex_init(&q->mq_freeze_lock); |
320ae51f | 525 | |
3ef28e83 DW |
526 | /* |
527 | * Init percpu_ref in atomic mode so that it's faster to shutdown. | |
528 | * See blk_register_queue() for details. | |
529 | */ | |
530 | if (percpu_ref_init(&q->q_usage_counter, | |
531 | blk_queue_usage_counter_release, | |
532 | PERCPU_REF_INIT_ATOMIC, GFP_KERNEL)) | |
fff4996b | 533 | goto fail_bdi; |
f51b802c | 534 | |
3ef28e83 DW |
535 | if (blkcg_init_queue(q)) |
536 | goto fail_ref; | |
537 | ||
1da177e4 | 538 | return q; |
a73f730d | 539 | |
3ef28e83 DW |
540 | fail_ref: |
541 | percpu_ref_exit(&q->q_usage_counter); | |
fff4996b | 542 | fail_bdi: |
a83b576c JA |
543 | blk_free_queue_stats(q->stats); |
544 | fail_stats: | |
d03f6cdc | 545 | bdi_put(q->backing_dev_info); |
54efd50b | 546 | fail_split: |
338aa96d | 547 | bioset_exit(&q->bio_split); |
a73f730d TH |
548 | fail_id: |
549 | ida_simple_remove(&blk_queue_ida, q->id); | |
550 | fail_q: | |
551 | kmem_cache_free(blk_requestq_cachep, q); | |
552 | return NULL; | |
1da177e4 | 553 | } |
1946089a | 554 | EXPORT_SYMBOL(blk_alloc_queue_node); |
1da177e4 | 555 | |
09ac46c4 | 556 | bool blk_get_queue(struct request_queue *q) |
1da177e4 | 557 | { |
3f3299d5 | 558 | if (likely(!blk_queue_dying(q))) { |
09ac46c4 TH |
559 | __blk_get_queue(q); |
560 | return true; | |
1da177e4 LT |
561 | } |
562 | ||
09ac46c4 | 563 | return false; |
1da177e4 | 564 | } |
d86e0e83 | 565 | EXPORT_SYMBOL(blk_get_queue); |
1da177e4 | 566 | |
a1ce35fa JA |
567 | /** |
568 | * blk_get_request - allocate a request | |
569 | * @q: request queue to allocate a request for | |
570 | * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC. | |
571 | * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT. | |
1da177e4 | 572 | */ |
a1ce35fa JA |
573 | struct request *blk_get_request(struct request_queue *q, unsigned int op, |
574 | blk_mq_req_flags_t flags) | |
1da177e4 | 575 | { |
a1ce35fa | 576 | struct request *req; |
1da177e4 | 577 | |
a1ce35fa JA |
578 | WARN_ON_ONCE(op & REQ_NOWAIT); |
579 | WARN_ON_ONCE(flags & ~(BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_PREEMPT)); | |
1da177e4 | 580 | |
a1ce35fa JA |
581 | req = blk_mq_alloc_request(q, op, flags); |
582 | if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn) | |
583 | q->mq_ops->initialize_rq_fn(req); | |
1da177e4 | 584 | |
a1ce35fa | 585 | return req; |
1da177e4 | 586 | } |
a1ce35fa | 587 | EXPORT_SYMBOL(blk_get_request); |
1da177e4 | 588 | |
1da177e4 LT |
589 | void blk_put_request(struct request *req) |
590 | { | |
a1ce35fa | 591 | blk_mq_free_request(req); |
1da177e4 | 592 | } |
1da177e4 LT |
593 | EXPORT_SYMBOL(blk_put_request); |
594 | ||
14ccb66b CH |
595 | bool bio_attempt_back_merge(struct request *req, struct bio *bio, |
596 | unsigned int nr_segs) | |
73c10101 | 597 | { |
1eff9d32 | 598 | const int ff = bio->bi_opf & REQ_FAILFAST_MASK; |
73c10101 | 599 | |
14ccb66b | 600 | if (!ll_back_merge_fn(req, bio, nr_segs)) |
73c10101 JA |
601 | return false; |
602 | ||
14ccb66b | 603 | trace_block_bio_backmerge(req->q, req, bio); |
73c10101 JA |
604 | |
605 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
606 | blk_rq_set_mixed_merge(req); | |
607 | ||
608 | req->biotail->bi_next = bio; | |
609 | req->biotail = bio; | |
4f024f37 | 610 | req->__data_len += bio->bi_iter.bi_size; |
73c10101 | 611 | |
320ae51f | 612 | blk_account_io_start(req, false); |
73c10101 JA |
613 | return true; |
614 | } | |
615 | ||
14ccb66b CH |
616 | bool bio_attempt_front_merge(struct request *req, struct bio *bio, |
617 | unsigned int nr_segs) | |
73c10101 | 618 | { |
1eff9d32 | 619 | const int ff = bio->bi_opf & REQ_FAILFAST_MASK; |
73c10101 | 620 | |
14ccb66b | 621 | if (!ll_front_merge_fn(req, bio, nr_segs)) |
73c10101 JA |
622 | return false; |
623 | ||
14ccb66b | 624 | trace_block_bio_frontmerge(req->q, req, bio); |
73c10101 JA |
625 | |
626 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
627 | blk_rq_set_mixed_merge(req); | |
628 | ||
73c10101 JA |
629 | bio->bi_next = req->bio; |
630 | req->bio = bio; | |
631 | ||
4f024f37 KO |
632 | req->__sector = bio->bi_iter.bi_sector; |
633 | req->__data_len += bio->bi_iter.bi_size; | |
73c10101 | 634 | |
320ae51f | 635 | blk_account_io_start(req, false); |
73c10101 JA |
636 | return true; |
637 | } | |
638 | ||
1e739730 CH |
639 | bool bio_attempt_discard_merge(struct request_queue *q, struct request *req, |
640 | struct bio *bio) | |
641 | { | |
642 | unsigned short segments = blk_rq_nr_discard_segments(req); | |
643 | ||
644 | if (segments >= queue_max_discard_segments(q)) | |
645 | goto no_merge; | |
646 | if (blk_rq_sectors(req) + bio_sectors(bio) > | |
647 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) | |
648 | goto no_merge; | |
649 | ||
650 | req->biotail->bi_next = bio; | |
651 | req->biotail = bio; | |
652 | req->__data_len += bio->bi_iter.bi_size; | |
1e739730 CH |
653 | req->nr_phys_segments = segments + 1; |
654 | ||
655 | blk_account_io_start(req, false); | |
656 | return true; | |
657 | no_merge: | |
658 | req_set_nomerge(q, req); | |
659 | return false; | |
660 | } | |
661 | ||
bd87b589 | 662 | /** |
320ae51f | 663 | * blk_attempt_plug_merge - try to merge with %current's plugged list |
bd87b589 TH |
664 | * @q: request_queue new bio is being queued at |
665 | * @bio: new bio being queued | |
14ccb66b | 666 | * @nr_segs: number of segments in @bio |
ccc2600b RD |
667 | * @same_queue_rq: pointer to &struct request that gets filled in when |
668 | * another request associated with @q is found on the plug list | |
669 | * (optional, may be %NULL) | |
bd87b589 TH |
670 | * |
671 | * Determine whether @bio being queued on @q can be merged with a request | |
672 | * on %current's plugged list. Returns %true if merge was successful, | |
673 | * otherwise %false. | |
674 | * | |
07c2bd37 TH |
675 | * Plugging coalesces IOs from the same issuer for the same purpose without |
676 | * going through @q->queue_lock. As such it's more of an issuing mechanism | |
677 | * than scheduling, and the request, while may have elvpriv data, is not | |
678 | * added on the elevator at this point. In addition, we don't have | |
679 | * reliable access to the elevator outside queue lock. Only check basic | |
680 | * merging parameters without querying the elevator. | |
da41a589 RE |
681 | * |
682 | * Caller must ensure !blk_queue_nomerges(q) beforehand. | |
73c10101 | 683 | */ |
320ae51f | 684 | bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio, |
14ccb66b | 685 | unsigned int nr_segs, struct request **same_queue_rq) |
73c10101 JA |
686 | { |
687 | struct blk_plug *plug; | |
688 | struct request *rq; | |
92f399c7 | 689 | struct list_head *plug_list; |
73c10101 | 690 | |
b49773e7 | 691 | plug = blk_mq_plug(q, bio); |
73c10101 | 692 | if (!plug) |
34fe7c05 | 693 | return false; |
73c10101 | 694 | |
a1ce35fa | 695 | plug_list = &plug->mq_list; |
92f399c7 SL |
696 | |
697 | list_for_each_entry_reverse(rq, plug_list, queuelist) { | |
34fe7c05 | 698 | bool merged = false; |
73c10101 | 699 | |
5f0ed774 | 700 | if (rq->q == q && same_queue_rq) { |
5b3f341f SL |
701 | /* |
702 | * Only blk-mq multiple hardware queues case checks the | |
703 | * rq in the same queue, there should be only one such | |
704 | * rq in a queue | |
705 | **/ | |
5f0ed774 | 706 | *same_queue_rq = rq; |
5b3f341f | 707 | } |
56ebdaf2 | 708 | |
07c2bd37 | 709 | if (rq->q != q || !blk_rq_merge_ok(rq, bio)) |
73c10101 JA |
710 | continue; |
711 | ||
34fe7c05 CH |
712 | switch (blk_try_merge(rq, bio)) { |
713 | case ELEVATOR_BACK_MERGE: | |
14ccb66b | 714 | merged = bio_attempt_back_merge(rq, bio, nr_segs); |
34fe7c05 CH |
715 | break; |
716 | case ELEVATOR_FRONT_MERGE: | |
14ccb66b | 717 | merged = bio_attempt_front_merge(rq, bio, nr_segs); |
34fe7c05 | 718 | break; |
1e739730 CH |
719 | case ELEVATOR_DISCARD_MERGE: |
720 | merged = bio_attempt_discard_merge(q, rq, bio); | |
721 | break; | |
34fe7c05 CH |
722 | default: |
723 | break; | |
73c10101 | 724 | } |
34fe7c05 CH |
725 | |
726 | if (merged) | |
727 | return true; | |
73c10101 | 728 | } |
34fe7c05 CH |
729 | |
730 | return false; | |
73c10101 JA |
731 | } |
732 | ||
52c5e62d | 733 | static void handle_bad_sector(struct bio *bio, sector_t maxsector) |
1da177e4 LT |
734 | { |
735 | char b[BDEVNAME_SIZE]; | |
736 | ||
737 | printk(KERN_INFO "attempt to access beyond end of device\n"); | |
6296b960 | 738 | printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n", |
74d46992 | 739 | bio_devname(bio, b), bio->bi_opf, |
f73a1c7d | 740 | (unsigned long long)bio_end_sector(bio), |
52c5e62d | 741 | (long long)maxsector); |
1da177e4 LT |
742 | } |
743 | ||
c17bb495 AM |
744 | #ifdef CONFIG_FAIL_MAKE_REQUEST |
745 | ||
746 | static DECLARE_FAULT_ATTR(fail_make_request); | |
747 | ||
748 | static int __init setup_fail_make_request(char *str) | |
749 | { | |
750 | return setup_fault_attr(&fail_make_request, str); | |
751 | } | |
752 | __setup("fail_make_request=", setup_fail_make_request); | |
753 | ||
b2c9cd37 | 754 | static bool should_fail_request(struct hd_struct *part, unsigned int bytes) |
c17bb495 | 755 | { |
b2c9cd37 | 756 | return part->make_it_fail && should_fail(&fail_make_request, bytes); |
c17bb495 AM |
757 | } |
758 | ||
759 | static int __init fail_make_request_debugfs(void) | |
760 | { | |
dd48c085 AM |
761 | struct dentry *dir = fault_create_debugfs_attr("fail_make_request", |
762 | NULL, &fail_make_request); | |
763 | ||
21f9fcd8 | 764 | return PTR_ERR_OR_ZERO(dir); |
c17bb495 AM |
765 | } |
766 | ||
767 | late_initcall(fail_make_request_debugfs); | |
768 | ||
769 | #else /* CONFIG_FAIL_MAKE_REQUEST */ | |
770 | ||
b2c9cd37 AM |
771 | static inline bool should_fail_request(struct hd_struct *part, |
772 | unsigned int bytes) | |
c17bb495 | 773 | { |
b2c9cd37 | 774 | return false; |
c17bb495 AM |
775 | } |
776 | ||
777 | #endif /* CONFIG_FAIL_MAKE_REQUEST */ | |
778 | ||
721c7fc7 ID |
779 | static inline bool bio_check_ro(struct bio *bio, struct hd_struct *part) |
780 | { | |
b089cfd9 JA |
781 | const int op = bio_op(bio); |
782 | ||
8b2ded1c | 783 | if (part->policy && op_is_write(op)) { |
721c7fc7 ID |
784 | char b[BDEVNAME_SIZE]; |
785 | ||
8b2ded1c MP |
786 | if (op_is_flush(bio->bi_opf) && !bio_sectors(bio)) |
787 | return false; | |
788 | ||
a32e236e | 789 | WARN_ONCE(1, |
721c7fc7 ID |
790 | "generic_make_request: Trying to write " |
791 | "to read-only block-device %s (partno %d)\n", | |
792 | bio_devname(bio, b), part->partno); | |
a32e236e LT |
793 | /* Older lvm-tools actually trigger this */ |
794 | return false; | |
721c7fc7 ID |
795 | } |
796 | ||
797 | return false; | |
798 | } | |
799 | ||
30abb3a6 HM |
800 | static noinline int should_fail_bio(struct bio *bio) |
801 | { | |
802 | if (should_fail_request(&bio->bi_disk->part0, bio->bi_iter.bi_size)) | |
803 | return -EIO; | |
804 | return 0; | |
805 | } | |
806 | ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO); | |
807 | ||
52c5e62d CH |
808 | /* |
809 | * Check whether this bio extends beyond the end of the device or partition. | |
810 | * This may well happen - the kernel calls bread() without checking the size of | |
811 | * the device, e.g., when mounting a file system. | |
812 | */ | |
813 | static inline int bio_check_eod(struct bio *bio, sector_t maxsector) | |
814 | { | |
815 | unsigned int nr_sectors = bio_sectors(bio); | |
816 | ||
817 | if (nr_sectors && maxsector && | |
818 | (nr_sectors > maxsector || | |
819 | bio->bi_iter.bi_sector > maxsector - nr_sectors)) { | |
820 | handle_bad_sector(bio, maxsector); | |
821 | return -EIO; | |
822 | } | |
823 | return 0; | |
824 | } | |
825 | ||
74d46992 CH |
826 | /* |
827 | * Remap block n of partition p to block n+start(p) of the disk. | |
828 | */ | |
829 | static inline int blk_partition_remap(struct bio *bio) | |
830 | { | |
831 | struct hd_struct *p; | |
52c5e62d | 832 | int ret = -EIO; |
74d46992 | 833 | |
721c7fc7 ID |
834 | rcu_read_lock(); |
835 | p = __disk_get_part(bio->bi_disk, bio->bi_partno); | |
52c5e62d CH |
836 | if (unlikely(!p)) |
837 | goto out; | |
838 | if (unlikely(should_fail_request(p, bio->bi_iter.bi_size))) | |
839 | goto out; | |
840 | if (unlikely(bio_check_ro(bio, p))) | |
721c7fc7 | 841 | goto out; |
721c7fc7 | 842 | |
74d46992 CH |
843 | /* |
844 | * Zone reset does not include bi_size so bio_sectors() is always 0. | |
845 | * Include a test for the reset op code and perform the remap if needed. | |
846 | */ | |
52c5e62d CH |
847 | if (bio_sectors(bio) || bio_op(bio) == REQ_OP_ZONE_RESET) { |
848 | if (bio_check_eod(bio, part_nr_sects_read(p))) | |
849 | goto out; | |
850 | bio->bi_iter.bi_sector += p->start_sect; | |
52c5e62d CH |
851 | trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p), |
852 | bio->bi_iter.bi_sector - p->start_sect); | |
853 | } | |
c04fa44b | 854 | bio->bi_partno = 0; |
52c5e62d | 855 | ret = 0; |
721c7fc7 ID |
856 | out: |
857 | rcu_read_unlock(); | |
74d46992 CH |
858 | return ret; |
859 | } | |
860 | ||
27a84d54 CH |
861 | static noinline_for_stack bool |
862 | generic_make_request_checks(struct bio *bio) | |
1da177e4 | 863 | { |
165125e1 | 864 | struct request_queue *q; |
5a7bbad2 | 865 | int nr_sectors = bio_sectors(bio); |
4e4cbee9 | 866 | blk_status_t status = BLK_STS_IOERR; |
5a7bbad2 | 867 | char b[BDEVNAME_SIZE]; |
1da177e4 LT |
868 | |
869 | might_sleep(); | |
1da177e4 | 870 | |
74d46992 | 871 | q = bio->bi_disk->queue; |
5a7bbad2 CH |
872 | if (unlikely(!q)) { |
873 | printk(KERN_ERR | |
874 | "generic_make_request: Trying to access " | |
875 | "nonexistent block-device %s (%Lu)\n", | |
74d46992 | 876 | bio_devname(bio, b), (long long)bio->bi_iter.bi_sector); |
5a7bbad2 CH |
877 | goto end_io; |
878 | } | |
c17bb495 | 879 | |
03a07c92 GR |
880 | /* |
881 | * For a REQ_NOWAIT based request, return -EOPNOTSUPP | |
882 | * if queue is not a request based queue. | |
883 | */ | |
344e9ffc | 884 | if ((bio->bi_opf & REQ_NOWAIT) && !queue_is_mq(q)) |
03a07c92 GR |
885 | goto not_supported; |
886 | ||
30abb3a6 | 887 | if (should_fail_bio(bio)) |
5a7bbad2 | 888 | goto end_io; |
2056a782 | 889 | |
52c5e62d CH |
890 | if (bio->bi_partno) { |
891 | if (unlikely(blk_partition_remap(bio))) | |
721c7fc7 ID |
892 | goto end_io; |
893 | } else { | |
52c5e62d CH |
894 | if (unlikely(bio_check_ro(bio, &bio->bi_disk->part0))) |
895 | goto end_io; | |
896 | if (unlikely(bio_check_eod(bio, get_capacity(bio->bi_disk)))) | |
721c7fc7 ID |
897 | goto end_io; |
898 | } | |
2056a782 | 899 | |
5a7bbad2 CH |
900 | /* |
901 | * Filter flush bio's early so that make_request based | |
902 | * drivers without flush support don't have to worry | |
903 | * about them. | |
904 | */ | |
f3a8ab7d | 905 | if (op_is_flush(bio->bi_opf) && |
c888a8f9 | 906 | !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) { |
1eff9d32 | 907 | bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA); |
5a7bbad2 | 908 | if (!nr_sectors) { |
4e4cbee9 | 909 | status = BLK_STS_OK; |
51fd77bd JA |
910 | goto end_io; |
911 | } | |
5a7bbad2 | 912 | } |
5ddfe969 | 913 | |
d04c406f CH |
914 | if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags)) |
915 | bio->bi_opf &= ~REQ_HIPRI; | |
916 | ||
288dab8a CH |
917 | switch (bio_op(bio)) { |
918 | case REQ_OP_DISCARD: | |
919 | if (!blk_queue_discard(q)) | |
920 | goto not_supported; | |
921 | break; | |
922 | case REQ_OP_SECURE_ERASE: | |
923 | if (!blk_queue_secure_erase(q)) | |
924 | goto not_supported; | |
925 | break; | |
926 | case REQ_OP_WRITE_SAME: | |
74d46992 | 927 | if (!q->limits.max_write_same_sectors) |
288dab8a | 928 | goto not_supported; |
58886785 | 929 | break; |
2d253440 | 930 | case REQ_OP_ZONE_RESET: |
74d46992 | 931 | if (!blk_queue_is_zoned(q)) |
2d253440 | 932 | goto not_supported; |
288dab8a | 933 | break; |
a6f0788e | 934 | case REQ_OP_WRITE_ZEROES: |
74d46992 | 935 | if (!q->limits.max_write_zeroes_sectors) |
a6f0788e CK |
936 | goto not_supported; |
937 | break; | |
288dab8a CH |
938 | default: |
939 | break; | |
5a7bbad2 | 940 | } |
01edede4 | 941 | |
7f4b35d1 TH |
942 | /* |
943 | * Various block parts want %current->io_context and lazy ioc | |
944 | * allocation ends up trading a lot of pain for a small amount of | |
945 | * memory. Just allocate it upfront. This may fail and block | |
946 | * layer knows how to live with it. | |
947 | */ | |
948 | create_io_context(GFP_ATOMIC, q->node); | |
949 | ||
ae118896 TH |
950 | if (!blkcg_bio_issue_check(q, bio)) |
951 | return false; | |
27a84d54 | 952 | |
fbbaf700 N |
953 | if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) { |
954 | trace_block_bio_queue(q, bio); | |
955 | /* Now that enqueuing has been traced, we need to trace | |
956 | * completion as well. | |
957 | */ | |
958 | bio_set_flag(bio, BIO_TRACE_COMPLETION); | |
959 | } | |
27a84d54 | 960 | return true; |
a7384677 | 961 | |
288dab8a | 962 | not_supported: |
4e4cbee9 | 963 | status = BLK_STS_NOTSUPP; |
a7384677 | 964 | end_io: |
4e4cbee9 | 965 | bio->bi_status = status; |
4246a0b6 | 966 | bio_endio(bio); |
27a84d54 | 967 | return false; |
1da177e4 LT |
968 | } |
969 | ||
27a84d54 CH |
970 | /** |
971 | * generic_make_request - hand a buffer to its device driver for I/O | |
972 | * @bio: The bio describing the location in memory and on the device. | |
973 | * | |
974 | * generic_make_request() is used to make I/O requests of block | |
975 | * devices. It is passed a &struct bio, which describes the I/O that needs | |
976 | * to be done. | |
977 | * | |
978 | * generic_make_request() does not return any status. The | |
979 | * success/failure status of the request, along with notification of | |
980 | * completion, is delivered asynchronously through the bio->bi_end_io | |
981 | * function described (one day) else where. | |
982 | * | |
983 | * The caller of generic_make_request must make sure that bi_io_vec | |
984 | * are set to describe the memory buffer, and that bi_dev and bi_sector are | |
985 | * set to describe the device address, and the | |
986 | * bi_end_io and optionally bi_private are set to describe how | |
987 | * completion notification should be signaled. | |
988 | * | |
989 | * generic_make_request and the drivers it calls may use bi_next if this | |
990 | * bio happens to be merged with someone else, and may resubmit the bio to | |
991 | * a lower device by calling into generic_make_request recursively, which | |
992 | * means the bio should NOT be touched after the call to ->make_request_fn. | |
d89d8796 | 993 | */ |
dece1635 | 994 | blk_qc_t generic_make_request(struct bio *bio) |
d89d8796 | 995 | { |
f5fe1b51 N |
996 | /* |
997 | * bio_list_on_stack[0] contains bios submitted by the current | |
998 | * make_request_fn. | |
999 | * bio_list_on_stack[1] contains bios that were submitted before | |
1000 | * the current make_request_fn, but that haven't been processed | |
1001 | * yet. | |
1002 | */ | |
1003 | struct bio_list bio_list_on_stack[2]; | |
dece1635 | 1004 | blk_qc_t ret = BLK_QC_T_NONE; |
bddd87c7 | 1005 | |
27a84d54 | 1006 | if (!generic_make_request_checks(bio)) |
dece1635 | 1007 | goto out; |
27a84d54 CH |
1008 | |
1009 | /* | |
1010 | * We only want one ->make_request_fn to be active at a time, else | |
1011 | * stack usage with stacked devices could be a problem. So use | |
1012 | * current->bio_list to keep a list of requests submited by a | |
1013 | * make_request_fn function. current->bio_list is also used as a | |
1014 | * flag to say if generic_make_request is currently active in this | |
1015 | * task or not. If it is NULL, then no make_request is active. If | |
1016 | * it is non-NULL, then a make_request is active, and new requests | |
1017 | * should be added at the tail | |
1018 | */ | |
bddd87c7 | 1019 | if (current->bio_list) { |
f5fe1b51 | 1020 | bio_list_add(¤t->bio_list[0], bio); |
dece1635 | 1021 | goto out; |
d89d8796 | 1022 | } |
27a84d54 | 1023 | |
d89d8796 NB |
1024 | /* following loop may be a bit non-obvious, and so deserves some |
1025 | * explanation. | |
1026 | * Before entering the loop, bio->bi_next is NULL (as all callers | |
1027 | * ensure that) so we have a list with a single bio. | |
1028 | * We pretend that we have just taken it off a longer list, so | |
bddd87c7 AM |
1029 | * we assign bio_list to a pointer to the bio_list_on_stack, |
1030 | * thus initialising the bio_list of new bios to be | |
27a84d54 | 1031 | * added. ->make_request() may indeed add some more bios |
d89d8796 NB |
1032 | * through a recursive call to generic_make_request. If it |
1033 | * did, we find a non-NULL value in bio_list and re-enter the loop | |
1034 | * from the top. In this case we really did just take the bio | |
bddd87c7 | 1035 | * of the top of the list (no pretending) and so remove it from |
27a84d54 | 1036 | * bio_list, and call into ->make_request() again. |
d89d8796 NB |
1037 | */ |
1038 | BUG_ON(bio->bi_next); | |
f5fe1b51 N |
1039 | bio_list_init(&bio_list_on_stack[0]); |
1040 | current->bio_list = bio_list_on_stack; | |
d89d8796 | 1041 | do { |
fe200864 ML |
1042 | struct request_queue *q = bio->bi_disk->queue; |
1043 | blk_mq_req_flags_t flags = bio->bi_opf & REQ_NOWAIT ? | |
1044 | BLK_MQ_REQ_NOWAIT : 0; | |
27a84d54 | 1045 | |
fe200864 | 1046 | if (likely(blk_queue_enter(q, flags) == 0)) { |
79bd9959 N |
1047 | struct bio_list lower, same; |
1048 | ||
1049 | /* Create a fresh bio_list for all subordinate requests */ | |
f5fe1b51 N |
1050 | bio_list_on_stack[1] = bio_list_on_stack[0]; |
1051 | bio_list_init(&bio_list_on_stack[0]); | |
dece1635 | 1052 | ret = q->make_request_fn(q, bio); |
3ef28e83 | 1053 | |
fe200864 ML |
1054 | blk_queue_exit(q); |
1055 | ||
79bd9959 N |
1056 | /* sort new bios into those for a lower level |
1057 | * and those for the same level | |
1058 | */ | |
1059 | bio_list_init(&lower); | |
1060 | bio_list_init(&same); | |
f5fe1b51 | 1061 | while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL) |
74d46992 | 1062 | if (q == bio->bi_disk->queue) |
79bd9959 N |
1063 | bio_list_add(&same, bio); |
1064 | else | |
1065 | bio_list_add(&lower, bio); | |
1066 | /* now assemble so we handle the lowest level first */ | |
f5fe1b51 N |
1067 | bio_list_merge(&bio_list_on_stack[0], &lower); |
1068 | bio_list_merge(&bio_list_on_stack[0], &same); | |
1069 | bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]); | |
3ef28e83 | 1070 | } else { |
03a07c92 GR |
1071 | if (unlikely(!blk_queue_dying(q) && |
1072 | (bio->bi_opf & REQ_NOWAIT))) | |
1073 | bio_wouldblock_error(bio); | |
1074 | else | |
1075 | bio_io_error(bio); | |
3ef28e83 | 1076 | } |
f5fe1b51 | 1077 | bio = bio_list_pop(&bio_list_on_stack[0]); |
d89d8796 | 1078 | } while (bio); |
bddd87c7 | 1079 | current->bio_list = NULL; /* deactivate */ |
dece1635 JA |
1080 | |
1081 | out: | |
1082 | return ret; | |
d89d8796 | 1083 | } |
1da177e4 LT |
1084 | EXPORT_SYMBOL(generic_make_request); |
1085 | ||
f421e1d9 CH |
1086 | /** |
1087 | * direct_make_request - hand a buffer directly to its device driver for I/O | |
1088 | * @bio: The bio describing the location in memory and on the device. | |
1089 | * | |
1090 | * This function behaves like generic_make_request(), but does not protect | |
1091 | * against recursion. Must only be used if the called driver is known | |
1092 | * to not call generic_make_request (or direct_make_request) again from | |
1093 | * its make_request function. (Calling direct_make_request again from | |
1094 | * a workqueue is perfectly fine as that doesn't recurse). | |
1095 | */ | |
1096 | blk_qc_t direct_make_request(struct bio *bio) | |
1097 | { | |
1098 | struct request_queue *q = bio->bi_disk->queue; | |
1099 | bool nowait = bio->bi_opf & REQ_NOWAIT; | |
1100 | blk_qc_t ret; | |
1101 | ||
1102 | if (!generic_make_request_checks(bio)) | |
1103 | return BLK_QC_T_NONE; | |
1104 | ||
3a0a5299 | 1105 | if (unlikely(blk_queue_enter(q, nowait ? BLK_MQ_REQ_NOWAIT : 0))) { |
f421e1d9 CH |
1106 | if (nowait && !blk_queue_dying(q)) |
1107 | bio->bi_status = BLK_STS_AGAIN; | |
1108 | else | |
1109 | bio->bi_status = BLK_STS_IOERR; | |
1110 | bio_endio(bio); | |
1111 | return BLK_QC_T_NONE; | |
1112 | } | |
1113 | ||
1114 | ret = q->make_request_fn(q, bio); | |
1115 | blk_queue_exit(q); | |
1116 | return ret; | |
1117 | } | |
1118 | EXPORT_SYMBOL_GPL(direct_make_request); | |
1119 | ||
1da177e4 | 1120 | /** |
710027a4 | 1121 | * submit_bio - submit a bio to the block device layer for I/O |
1da177e4 LT |
1122 | * @bio: The &struct bio which describes the I/O |
1123 | * | |
1124 | * submit_bio() is very similar in purpose to generic_make_request(), and | |
1125 | * uses that function to do most of the work. Both are fairly rough | |
710027a4 | 1126 | * interfaces; @bio must be presetup and ready for I/O. |
1da177e4 LT |
1127 | * |
1128 | */ | |
4e49ea4a | 1129 | blk_qc_t submit_bio(struct bio *bio) |
1da177e4 | 1130 | { |
d3f77dfd TH |
1131 | if (blkcg_punt_bio_submit(bio)) |
1132 | return BLK_QC_T_NONE; | |
1133 | ||
bf2de6f5 JA |
1134 | /* |
1135 | * If it's a regular read/write or a barrier with data attached, | |
1136 | * go through the normal accounting stuff before submission. | |
1137 | */ | |
e2a60da7 | 1138 | if (bio_has_data(bio)) { |
4363ac7c MP |
1139 | unsigned int count; |
1140 | ||
95fe6c1a | 1141 | if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME)) |
7c5a0dcf | 1142 | count = queue_logical_block_size(bio->bi_disk->queue) >> 9; |
4363ac7c MP |
1143 | else |
1144 | count = bio_sectors(bio); | |
1145 | ||
a8ebb056 | 1146 | if (op_is_write(bio_op(bio))) { |
bf2de6f5 JA |
1147 | count_vm_events(PGPGOUT, count); |
1148 | } else { | |
4f024f37 | 1149 | task_io_account_read(bio->bi_iter.bi_size); |
bf2de6f5 JA |
1150 | count_vm_events(PGPGIN, count); |
1151 | } | |
1152 | ||
1153 | if (unlikely(block_dump)) { | |
1154 | char b[BDEVNAME_SIZE]; | |
8dcbdc74 | 1155 | printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n", |
ba25f9dc | 1156 | current->comm, task_pid_nr(current), |
a8ebb056 | 1157 | op_is_write(bio_op(bio)) ? "WRITE" : "READ", |
4f024f37 | 1158 | (unsigned long long)bio->bi_iter.bi_sector, |
74d46992 | 1159 | bio_devname(bio, b), count); |
bf2de6f5 | 1160 | } |
1da177e4 LT |
1161 | } |
1162 | ||
dece1635 | 1163 | return generic_make_request(bio); |
1da177e4 | 1164 | } |
1da177e4 LT |
1165 | EXPORT_SYMBOL(submit_bio); |
1166 | ||
82124d60 | 1167 | /** |
bf4e6b4e HR |
1168 | * blk_cloned_rq_check_limits - Helper function to check a cloned request |
1169 | * for new the queue limits | |
82124d60 KU |
1170 | * @q: the queue |
1171 | * @rq: the request being checked | |
1172 | * | |
1173 | * Description: | |
1174 | * @rq may have been made based on weaker limitations of upper-level queues | |
1175 | * in request stacking drivers, and it may violate the limitation of @q. | |
1176 | * Since the block layer and the underlying device driver trust @rq | |
1177 | * after it is inserted to @q, it should be checked against @q before | |
1178 | * the insertion using this generic function. | |
1179 | * | |
82124d60 | 1180 | * Request stacking drivers like request-based dm may change the queue |
bf4e6b4e HR |
1181 | * limits when retrying requests on other queues. Those requests need |
1182 | * to be checked against the new queue limits again during dispatch. | |
82124d60 | 1183 | */ |
bf4e6b4e HR |
1184 | static int blk_cloned_rq_check_limits(struct request_queue *q, |
1185 | struct request *rq) | |
82124d60 | 1186 | { |
8fe0d473 | 1187 | if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) { |
61939b12 JP |
1188 | printk(KERN_ERR "%s: over max size limit. (%u > %u)\n", |
1189 | __func__, blk_rq_sectors(rq), | |
1190 | blk_queue_get_max_sectors(q, req_op(rq))); | |
82124d60 KU |
1191 | return -EIO; |
1192 | } | |
1193 | ||
1194 | /* | |
1195 | * queue's settings related to segment counting like q->bounce_pfn | |
1196 | * may differ from that of other stacking queues. | |
1197 | * Recalculate it to check the request correctly on this queue's | |
1198 | * limitation. | |
1199 | */ | |
e9cd19c0 | 1200 | rq->nr_phys_segments = blk_recalc_rq_segments(rq); |
8a78362c | 1201 | if (rq->nr_phys_segments > queue_max_segments(q)) { |
61939b12 JP |
1202 | printk(KERN_ERR "%s: over max segments limit. (%hu > %hu)\n", |
1203 | __func__, rq->nr_phys_segments, queue_max_segments(q)); | |
82124d60 KU |
1204 | return -EIO; |
1205 | } | |
1206 | ||
1207 | return 0; | |
1208 | } | |
82124d60 KU |
1209 | |
1210 | /** | |
1211 | * blk_insert_cloned_request - Helper for stacking drivers to submit a request | |
1212 | * @q: the queue to submit the request | |
1213 | * @rq: the request being queued | |
1214 | */ | |
2a842aca | 1215 | blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq) |
82124d60 | 1216 | { |
bf4e6b4e | 1217 | if (blk_cloned_rq_check_limits(q, rq)) |
2a842aca | 1218 | return BLK_STS_IOERR; |
82124d60 | 1219 | |
b2c9cd37 AM |
1220 | if (rq->rq_disk && |
1221 | should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq))) | |
2a842aca | 1222 | return BLK_STS_IOERR; |
82124d60 | 1223 | |
a1ce35fa JA |
1224 | if (blk_queue_io_stat(q)) |
1225 | blk_account_io_start(rq, true); | |
82124d60 KU |
1226 | |
1227 | /* | |
a1ce35fa JA |
1228 | * Since we have a scheduler attached on the top device, |
1229 | * bypass a potential scheduler on the bottom device for | |
1230 | * insert. | |
82124d60 | 1231 | */ |
fd9c40f6 | 1232 | return blk_mq_request_issue_directly(rq, true); |
82124d60 KU |
1233 | } |
1234 | EXPORT_SYMBOL_GPL(blk_insert_cloned_request); | |
1235 | ||
80a761fd TH |
1236 | /** |
1237 | * blk_rq_err_bytes - determine number of bytes till the next failure boundary | |
1238 | * @rq: request to examine | |
1239 | * | |
1240 | * Description: | |
1241 | * A request could be merge of IOs which require different failure | |
1242 | * handling. This function determines the number of bytes which | |
1243 | * can be failed from the beginning of the request without | |
1244 | * crossing into area which need to be retried further. | |
1245 | * | |
1246 | * Return: | |
1247 | * The number of bytes to fail. | |
80a761fd TH |
1248 | */ |
1249 | unsigned int blk_rq_err_bytes(const struct request *rq) | |
1250 | { | |
1251 | unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK; | |
1252 | unsigned int bytes = 0; | |
1253 | struct bio *bio; | |
1254 | ||
e8064021 | 1255 | if (!(rq->rq_flags & RQF_MIXED_MERGE)) |
80a761fd TH |
1256 | return blk_rq_bytes(rq); |
1257 | ||
1258 | /* | |
1259 | * Currently the only 'mixing' which can happen is between | |
1260 | * different fastfail types. We can safely fail portions | |
1261 | * which have all the failfast bits that the first one has - | |
1262 | * the ones which are at least as eager to fail as the first | |
1263 | * one. | |
1264 | */ | |
1265 | for (bio = rq->bio; bio; bio = bio->bi_next) { | |
1eff9d32 | 1266 | if ((bio->bi_opf & ff) != ff) |
80a761fd | 1267 | break; |
4f024f37 | 1268 | bytes += bio->bi_iter.bi_size; |
80a761fd TH |
1269 | } |
1270 | ||
1271 | /* this could lead to infinite loop */ | |
1272 | BUG_ON(blk_rq_bytes(rq) && !bytes); | |
1273 | return bytes; | |
1274 | } | |
1275 | EXPORT_SYMBOL_GPL(blk_rq_err_bytes); | |
1276 | ||
320ae51f | 1277 | void blk_account_io_completion(struct request *req, unsigned int bytes) |
bc58ba94 | 1278 | { |
c2553b58 | 1279 | if (blk_do_io_stat(req)) { |
ddcf35d3 | 1280 | const int sgrp = op_stat_group(req_op(req)); |
bc58ba94 | 1281 | struct hd_struct *part; |
bc58ba94 | 1282 | |
112f158f | 1283 | part_stat_lock(); |
09e099d4 | 1284 | part = req->part; |
112f158f | 1285 | part_stat_add(part, sectors[sgrp], bytes >> 9); |
bc58ba94 JA |
1286 | part_stat_unlock(); |
1287 | } | |
1288 | } | |
1289 | ||
522a7775 | 1290 | void blk_account_io_done(struct request *req, u64 now) |
bc58ba94 | 1291 | { |
bc58ba94 | 1292 | /* |
dd4c133f TH |
1293 | * Account IO completion. flush_rq isn't accounted as a |
1294 | * normal IO on queueing nor completion. Accounting the | |
1295 | * containing request is enough. | |
bc58ba94 | 1296 | */ |
e8064021 | 1297 | if (blk_do_io_stat(req) && !(req->rq_flags & RQF_FLUSH_SEQ)) { |
ddcf35d3 | 1298 | const int sgrp = op_stat_group(req_op(req)); |
bc58ba94 | 1299 | struct hd_struct *part; |
bc58ba94 | 1300 | |
112f158f | 1301 | part_stat_lock(); |
09e099d4 | 1302 | part = req->part; |
bc58ba94 | 1303 | |
5b18b5a7 | 1304 | update_io_ticks(part, jiffies); |
112f158f MS |
1305 | part_stat_inc(part, ios[sgrp]); |
1306 | part_stat_add(part, nsecs[sgrp], now - req->start_time_ns); | |
5b18b5a7 | 1307 | part_stat_add(part, time_in_queue, nsecs_to_jiffies64(now - req->start_time_ns)); |
ddcf35d3 | 1308 | part_dec_in_flight(req->q, part, rq_data_dir(req)); |
bc58ba94 | 1309 | |
6c23a968 | 1310 | hd_struct_put(part); |
bc58ba94 JA |
1311 | part_stat_unlock(); |
1312 | } | |
1313 | } | |
1314 | ||
320ae51f JA |
1315 | void blk_account_io_start(struct request *rq, bool new_io) |
1316 | { | |
1317 | struct hd_struct *part; | |
1318 | int rw = rq_data_dir(rq); | |
320ae51f JA |
1319 | |
1320 | if (!blk_do_io_stat(rq)) | |
1321 | return; | |
1322 | ||
112f158f | 1323 | part_stat_lock(); |
320ae51f JA |
1324 | |
1325 | if (!new_io) { | |
1326 | part = rq->part; | |
112f158f | 1327 | part_stat_inc(part, merges[rw]); |
320ae51f JA |
1328 | } else { |
1329 | part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq)); | |
1330 | if (!hd_struct_try_get(part)) { | |
1331 | /* | |
1332 | * The partition is already being removed, | |
1333 | * the request will be accounted on the disk only | |
1334 | * | |
1335 | * We take a reference on disk->part0 although that | |
1336 | * partition will never be deleted, so we can treat | |
1337 | * it as any other partition. | |
1338 | */ | |
1339 | part = &rq->rq_disk->part0; | |
1340 | hd_struct_get(part); | |
1341 | } | |
d62e26b3 | 1342 | part_inc_in_flight(rq->q, part, rw); |
320ae51f JA |
1343 | rq->part = part; |
1344 | } | |
1345 | ||
5b18b5a7 MP |
1346 | update_io_ticks(part, jiffies); |
1347 | ||
320ae51f JA |
1348 | part_stat_unlock(); |
1349 | } | |
1350 | ||
ef71de8b CH |
1351 | /* |
1352 | * Steal bios from a request and add them to a bio list. | |
1353 | * The request must not have been partially completed before. | |
1354 | */ | |
1355 | void blk_steal_bios(struct bio_list *list, struct request *rq) | |
1356 | { | |
1357 | if (rq->bio) { | |
1358 | if (list->tail) | |
1359 | list->tail->bi_next = rq->bio; | |
1360 | else | |
1361 | list->head = rq->bio; | |
1362 | list->tail = rq->biotail; | |
1363 | ||
1364 | rq->bio = NULL; | |
1365 | rq->biotail = NULL; | |
1366 | } | |
1367 | ||
1368 | rq->__data_len = 0; | |
1369 | } | |
1370 | EXPORT_SYMBOL_GPL(blk_steal_bios); | |
1371 | ||
3bcddeac | 1372 | /** |
2e60e022 | 1373 | * blk_update_request - Special helper function for request stacking drivers |
8ebf9756 | 1374 | * @req: the request being processed |
2a842aca | 1375 | * @error: block status code |
8ebf9756 | 1376 | * @nr_bytes: number of bytes to complete @req |
3bcddeac KU |
1377 | * |
1378 | * Description: | |
8ebf9756 RD |
1379 | * Ends I/O on a number of bytes attached to @req, but doesn't complete |
1380 | * the request structure even if @req doesn't have leftover. | |
1381 | * If @req has leftover, sets it up for the next range of segments. | |
2e60e022 TH |
1382 | * |
1383 | * This special helper function is only for request stacking drivers | |
1384 | * (e.g. request-based dm) so that they can handle partial completion. | |
3a211b71 | 1385 | * Actual device drivers should use blk_mq_end_request instead. |
2e60e022 TH |
1386 | * |
1387 | * Passing the result of blk_rq_bytes() as @nr_bytes guarantees | |
1388 | * %false return from this function. | |
3bcddeac | 1389 | * |
1954e9a9 BVA |
1390 | * Note: |
1391 | * The RQF_SPECIAL_PAYLOAD flag is ignored on purpose in both | |
1392 | * blk_rq_bytes() and in blk_update_request(). | |
1393 | * | |
3bcddeac | 1394 | * Return: |
2e60e022 TH |
1395 | * %false - this request doesn't have any more data |
1396 | * %true - this request has more data | |
3bcddeac | 1397 | **/ |
2a842aca CH |
1398 | bool blk_update_request(struct request *req, blk_status_t error, |
1399 | unsigned int nr_bytes) | |
1da177e4 | 1400 | { |
f79ea416 | 1401 | int total_bytes; |
1da177e4 | 1402 | |
2a842aca | 1403 | trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes); |
4a0efdc9 | 1404 | |
2e60e022 TH |
1405 | if (!req->bio) |
1406 | return false; | |
1407 | ||
2a842aca CH |
1408 | if (unlikely(error && !blk_rq_is_passthrough(req) && |
1409 | !(req->rq_flags & RQF_QUIET))) | |
178cc590 | 1410 | print_req_error(req, error, __func__); |
1da177e4 | 1411 | |
bc58ba94 | 1412 | blk_account_io_completion(req, nr_bytes); |
d72d904a | 1413 | |
f79ea416 KO |
1414 | total_bytes = 0; |
1415 | while (req->bio) { | |
1416 | struct bio *bio = req->bio; | |
4f024f37 | 1417 | unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes); |
1da177e4 | 1418 | |
9c24c10a | 1419 | if (bio_bytes == bio->bi_iter.bi_size) |
1da177e4 | 1420 | req->bio = bio->bi_next; |
1da177e4 | 1421 | |
fbbaf700 N |
1422 | /* Completion has already been traced */ |
1423 | bio_clear_flag(bio, BIO_TRACE_COMPLETION); | |
f79ea416 | 1424 | req_bio_endio(req, bio, bio_bytes, error); |
1da177e4 | 1425 | |
f79ea416 KO |
1426 | total_bytes += bio_bytes; |
1427 | nr_bytes -= bio_bytes; | |
1da177e4 | 1428 | |
f79ea416 KO |
1429 | if (!nr_bytes) |
1430 | break; | |
1da177e4 LT |
1431 | } |
1432 | ||
1433 | /* | |
1434 | * completely done | |
1435 | */ | |
2e60e022 TH |
1436 | if (!req->bio) { |
1437 | /* | |
1438 | * Reset counters so that the request stacking driver | |
1439 | * can find how many bytes remain in the request | |
1440 | * later. | |
1441 | */ | |
a2dec7b3 | 1442 | req->__data_len = 0; |
2e60e022 TH |
1443 | return false; |
1444 | } | |
1da177e4 | 1445 | |
a2dec7b3 | 1446 | req->__data_len -= total_bytes; |
2e46e8b2 TH |
1447 | |
1448 | /* update sector only for requests with clear definition of sector */ | |
57292b58 | 1449 | if (!blk_rq_is_passthrough(req)) |
a2dec7b3 | 1450 | req->__sector += total_bytes >> 9; |
2e46e8b2 | 1451 | |
80a761fd | 1452 | /* mixed attributes always follow the first bio */ |
e8064021 | 1453 | if (req->rq_flags & RQF_MIXED_MERGE) { |
80a761fd | 1454 | req->cmd_flags &= ~REQ_FAILFAST_MASK; |
1eff9d32 | 1455 | req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK; |
80a761fd TH |
1456 | } |
1457 | ||
ed6565e7 CH |
1458 | if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) { |
1459 | /* | |
1460 | * If total number of sectors is less than the first segment | |
1461 | * size, something has gone terribly wrong. | |
1462 | */ | |
1463 | if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) { | |
1464 | blk_dump_rq_flags(req, "request botched"); | |
1465 | req->__data_len = blk_rq_cur_bytes(req); | |
1466 | } | |
2e46e8b2 | 1467 | |
ed6565e7 | 1468 | /* recalculate the number of segments */ |
e9cd19c0 | 1469 | req->nr_phys_segments = blk_recalc_rq_segments(req); |
ed6565e7 | 1470 | } |
2e46e8b2 | 1471 | |
2e60e022 | 1472 | return true; |
1da177e4 | 1473 | } |
2e60e022 | 1474 | EXPORT_SYMBOL_GPL(blk_update_request); |
1da177e4 | 1475 | |
2d4dc890 IL |
1476 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE |
1477 | /** | |
1478 | * rq_flush_dcache_pages - Helper function to flush all pages in a request | |
1479 | * @rq: the request to be flushed | |
1480 | * | |
1481 | * Description: | |
1482 | * Flush all pages in @rq. | |
1483 | */ | |
1484 | void rq_flush_dcache_pages(struct request *rq) | |
1485 | { | |
1486 | struct req_iterator iter; | |
7988613b | 1487 | struct bio_vec bvec; |
2d4dc890 IL |
1488 | |
1489 | rq_for_each_segment(bvec, rq, iter) | |
7988613b | 1490 | flush_dcache_page(bvec.bv_page); |
2d4dc890 IL |
1491 | } |
1492 | EXPORT_SYMBOL_GPL(rq_flush_dcache_pages); | |
1493 | #endif | |
1494 | ||
ef9e3fac KU |
1495 | /** |
1496 | * blk_lld_busy - Check if underlying low-level drivers of a device are busy | |
1497 | * @q : the queue of the device being checked | |
1498 | * | |
1499 | * Description: | |
1500 | * Check if underlying low-level drivers of a device are busy. | |
1501 | * If the drivers want to export their busy state, they must set own | |
1502 | * exporting function using blk_queue_lld_busy() first. | |
1503 | * | |
1504 | * Basically, this function is used only by request stacking drivers | |
1505 | * to stop dispatching requests to underlying devices when underlying | |
1506 | * devices are busy. This behavior helps more I/O merging on the queue | |
1507 | * of the request stacking driver and prevents I/O throughput regression | |
1508 | * on burst I/O load. | |
1509 | * | |
1510 | * Return: | |
1511 | * 0 - Not busy (The request stacking driver should dispatch request) | |
1512 | * 1 - Busy (The request stacking driver should stop dispatching request) | |
1513 | */ | |
1514 | int blk_lld_busy(struct request_queue *q) | |
1515 | { | |
344e9ffc | 1516 | if (queue_is_mq(q) && q->mq_ops->busy) |
9ba20527 | 1517 | return q->mq_ops->busy(q); |
ef9e3fac KU |
1518 | |
1519 | return 0; | |
1520 | } | |
1521 | EXPORT_SYMBOL_GPL(blk_lld_busy); | |
1522 | ||
78d8e58a MS |
1523 | /** |
1524 | * blk_rq_unprep_clone - Helper function to free all bios in a cloned request | |
1525 | * @rq: the clone request to be cleaned up | |
1526 | * | |
1527 | * Description: | |
1528 | * Free all bios in @rq for a cloned request. | |
1529 | */ | |
1530 | void blk_rq_unprep_clone(struct request *rq) | |
1531 | { | |
1532 | struct bio *bio; | |
1533 | ||
1534 | while ((bio = rq->bio) != NULL) { | |
1535 | rq->bio = bio->bi_next; | |
1536 | ||
1537 | bio_put(bio); | |
1538 | } | |
1539 | } | |
1540 | EXPORT_SYMBOL_GPL(blk_rq_unprep_clone); | |
1541 | ||
1542 | /* | |
1543 | * Copy attributes of the original request to the clone request. | |
1544 | * The actual data parts (e.g. ->cmd, ->sense) are not copied. | |
1545 | */ | |
1546 | static void __blk_rq_prep_clone(struct request *dst, struct request *src) | |
b0fd271d | 1547 | { |
b0fd271d KU |
1548 | dst->__sector = blk_rq_pos(src); |
1549 | dst->__data_len = blk_rq_bytes(src); | |
297ba57d BVA |
1550 | if (src->rq_flags & RQF_SPECIAL_PAYLOAD) { |
1551 | dst->rq_flags |= RQF_SPECIAL_PAYLOAD; | |
1552 | dst->special_vec = src->special_vec; | |
1553 | } | |
b0fd271d KU |
1554 | dst->nr_phys_segments = src->nr_phys_segments; |
1555 | dst->ioprio = src->ioprio; | |
1556 | dst->extra_len = src->extra_len; | |
78d8e58a MS |
1557 | } |
1558 | ||
1559 | /** | |
1560 | * blk_rq_prep_clone - Helper function to setup clone request | |
1561 | * @rq: the request to be setup | |
1562 | * @rq_src: original request to be cloned | |
1563 | * @bs: bio_set that bios for clone are allocated from | |
1564 | * @gfp_mask: memory allocation mask for bio | |
1565 | * @bio_ctr: setup function to be called for each clone bio. | |
1566 | * Returns %0 for success, non %0 for failure. | |
1567 | * @data: private data to be passed to @bio_ctr | |
1568 | * | |
1569 | * Description: | |
1570 | * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq. | |
1571 | * The actual data parts of @rq_src (e.g. ->cmd, ->sense) | |
1572 | * are not copied, and copying such parts is the caller's responsibility. | |
1573 | * Also, pages which the original bios are pointing to are not copied | |
1574 | * and the cloned bios just point same pages. | |
1575 | * So cloned bios must be completed before original bios, which means | |
1576 | * the caller must complete @rq before @rq_src. | |
1577 | */ | |
1578 | int blk_rq_prep_clone(struct request *rq, struct request *rq_src, | |
1579 | struct bio_set *bs, gfp_t gfp_mask, | |
1580 | int (*bio_ctr)(struct bio *, struct bio *, void *), | |
1581 | void *data) | |
1582 | { | |
1583 | struct bio *bio, *bio_src; | |
1584 | ||
1585 | if (!bs) | |
f4f8154a | 1586 | bs = &fs_bio_set; |
78d8e58a MS |
1587 | |
1588 | __rq_for_each_bio(bio_src, rq_src) { | |
1589 | bio = bio_clone_fast(bio_src, gfp_mask, bs); | |
1590 | if (!bio) | |
1591 | goto free_and_out; | |
1592 | ||
1593 | if (bio_ctr && bio_ctr(bio, bio_src, data)) | |
1594 | goto free_and_out; | |
1595 | ||
1596 | if (rq->bio) { | |
1597 | rq->biotail->bi_next = bio; | |
1598 | rq->biotail = bio; | |
1599 | } else | |
1600 | rq->bio = rq->biotail = bio; | |
1601 | } | |
1602 | ||
1603 | __blk_rq_prep_clone(rq, rq_src); | |
1604 | ||
1605 | return 0; | |
1606 | ||
1607 | free_and_out: | |
1608 | if (bio) | |
1609 | bio_put(bio); | |
1610 | blk_rq_unprep_clone(rq); | |
1611 | ||
1612 | return -ENOMEM; | |
b0fd271d KU |
1613 | } |
1614 | EXPORT_SYMBOL_GPL(blk_rq_prep_clone); | |
1615 | ||
59c3d45e | 1616 | int kblockd_schedule_work(struct work_struct *work) |
1da177e4 LT |
1617 | { |
1618 | return queue_work(kblockd_workqueue, work); | |
1619 | } | |
1da177e4 LT |
1620 | EXPORT_SYMBOL(kblockd_schedule_work); |
1621 | ||
ee63cfa7 JA |
1622 | int kblockd_schedule_work_on(int cpu, struct work_struct *work) |
1623 | { | |
1624 | return queue_work_on(cpu, kblockd_workqueue, work); | |
1625 | } | |
1626 | EXPORT_SYMBOL(kblockd_schedule_work_on); | |
1627 | ||
818cd1cb JA |
1628 | int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork, |
1629 | unsigned long delay) | |
1630 | { | |
1631 | return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay); | |
1632 | } | |
1633 | EXPORT_SYMBOL(kblockd_mod_delayed_work_on); | |
1634 | ||
75df7136 SJ |
1635 | /** |
1636 | * blk_start_plug - initialize blk_plug and track it inside the task_struct | |
1637 | * @plug: The &struct blk_plug that needs to be initialized | |
1638 | * | |
1639 | * Description: | |
40405851 JM |
1640 | * blk_start_plug() indicates to the block layer an intent by the caller |
1641 | * to submit multiple I/O requests in a batch. The block layer may use | |
1642 | * this hint to defer submitting I/Os from the caller until blk_finish_plug() | |
1643 | * is called. However, the block layer may choose to submit requests | |
1644 | * before a call to blk_finish_plug() if the number of queued I/Os | |
1645 | * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than | |
1646 | * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if | |
1647 | * the task schedules (see below). | |
1648 | * | |
75df7136 SJ |
1649 | * Tracking blk_plug inside the task_struct will help with auto-flushing the |
1650 | * pending I/O should the task end up blocking between blk_start_plug() and | |
1651 | * blk_finish_plug(). This is important from a performance perspective, but | |
1652 | * also ensures that we don't deadlock. For instance, if the task is blocking | |
1653 | * for a memory allocation, memory reclaim could end up wanting to free a | |
1654 | * page belonging to that request that is currently residing in our private | |
1655 | * plug. By flushing the pending I/O when the process goes to sleep, we avoid | |
1656 | * this kind of deadlock. | |
1657 | */ | |
73c10101 JA |
1658 | void blk_start_plug(struct blk_plug *plug) |
1659 | { | |
1660 | struct task_struct *tsk = current; | |
1661 | ||
dd6cf3e1 SL |
1662 | /* |
1663 | * If this is a nested plug, don't actually assign it. | |
1664 | */ | |
1665 | if (tsk->plug) | |
1666 | return; | |
1667 | ||
320ae51f | 1668 | INIT_LIST_HEAD(&plug->mq_list); |
048c9374 | 1669 | INIT_LIST_HEAD(&plug->cb_list); |
5f0ed774 | 1670 | plug->rq_count = 0; |
ce5b009c | 1671 | plug->multiple_queues = false; |
5f0ed774 | 1672 | |
73c10101 | 1673 | /* |
dd6cf3e1 SL |
1674 | * Store ordering should not be needed here, since a potential |
1675 | * preempt will imply a full memory barrier | |
73c10101 | 1676 | */ |
dd6cf3e1 | 1677 | tsk->plug = plug; |
73c10101 JA |
1678 | } |
1679 | EXPORT_SYMBOL(blk_start_plug); | |
1680 | ||
74018dc3 | 1681 | static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule) |
048c9374 N |
1682 | { |
1683 | LIST_HEAD(callbacks); | |
1684 | ||
2a7d5559 SL |
1685 | while (!list_empty(&plug->cb_list)) { |
1686 | list_splice_init(&plug->cb_list, &callbacks); | |
048c9374 | 1687 | |
2a7d5559 SL |
1688 | while (!list_empty(&callbacks)) { |
1689 | struct blk_plug_cb *cb = list_first_entry(&callbacks, | |
048c9374 N |
1690 | struct blk_plug_cb, |
1691 | list); | |
2a7d5559 | 1692 | list_del(&cb->list); |
74018dc3 | 1693 | cb->callback(cb, from_schedule); |
2a7d5559 | 1694 | } |
048c9374 N |
1695 | } |
1696 | } | |
1697 | ||
9cbb1750 N |
1698 | struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data, |
1699 | int size) | |
1700 | { | |
1701 | struct blk_plug *plug = current->plug; | |
1702 | struct blk_plug_cb *cb; | |
1703 | ||
1704 | if (!plug) | |
1705 | return NULL; | |
1706 | ||
1707 | list_for_each_entry(cb, &plug->cb_list, list) | |
1708 | if (cb->callback == unplug && cb->data == data) | |
1709 | return cb; | |
1710 | ||
1711 | /* Not currently on the callback list */ | |
1712 | BUG_ON(size < sizeof(*cb)); | |
1713 | cb = kzalloc(size, GFP_ATOMIC); | |
1714 | if (cb) { | |
1715 | cb->data = data; | |
1716 | cb->callback = unplug; | |
1717 | list_add(&cb->list, &plug->cb_list); | |
1718 | } | |
1719 | return cb; | |
1720 | } | |
1721 | EXPORT_SYMBOL(blk_check_plugged); | |
1722 | ||
49cac01e | 1723 | void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule) |
73c10101 | 1724 | { |
74018dc3 | 1725 | flush_plug_callbacks(plug, from_schedule); |
320ae51f JA |
1726 | |
1727 | if (!list_empty(&plug->mq_list)) | |
1728 | blk_mq_flush_plug_list(plug, from_schedule); | |
73c10101 | 1729 | } |
73c10101 | 1730 | |
40405851 JM |
1731 | /** |
1732 | * blk_finish_plug - mark the end of a batch of submitted I/O | |
1733 | * @plug: The &struct blk_plug passed to blk_start_plug() | |
1734 | * | |
1735 | * Description: | |
1736 | * Indicate that a batch of I/O submissions is complete. This function | |
1737 | * must be paired with an initial call to blk_start_plug(). The intent | |
1738 | * is to allow the block layer to optimize I/O submission. See the | |
1739 | * documentation for blk_start_plug() for more information. | |
1740 | */ | |
73c10101 JA |
1741 | void blk_finish_plug(struct blk_plug *plug) |
1742 | { | |
dd6cf3e1 SL |
1743 | if (plug != current->plug) |
1744 | return; | |
f6603783 | 1745 | blk_flush_plug_list(plug, false); |
73c10101 | 1746 | |
dd6cf3e1 | 1747 | current->plug = NULL; |
73c10101 | 1748 | } |
88b996cd | 1749 | EXPORT_SYMBOL(blk_finish_plug); |
73c10101 | 1750 | |
1da177e4 LT |
1751 | int __init blk_dev_init(void) |
1752 | { | |
ef295ecf CH |
1753 | BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS)); |
1754 | BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * | |
0762b23d | 1755 | FIELD_SIZEOF(struct request, cmd_flags)); |
ef295ecf CH |
1756 | BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * |
1757 | FIELD_SIZEOF(struct bio, bi_opf)); | |
9eb55b03 | 1758 | |
89b90be2 TH |
1759 | /* used for unplugging and affects IO latency/throughput - HIGHPRI */ |
1760 | kblockd_workqueue = alloc_workqueue("kblockd", | |
28747fcd | 1761 | WQ_MEM_RECLAIM | WQ_HIGHPRI, 0); |
1da177e4 LT |
1762 | if (!kblockd_workqueue) |
1763 | panic("Failed to create kblockd\n"); | |
1764 | ||
c2789bd4 | 1765 | blk_requestq_cachep = kmem_cache_create("request_queue", |
165125e1 | 1766 | sizeof(struct request_queue), 0, SLAB_PANIC, NULL); |
1da177e4 | 1767 | |
18fbda91 OS |
1768 | #ifdef CONFIG_DEBUG_FS |
1769 | blk_debugfs_root = debugfs_create_dir("block", NULL); | |
1770 | #endif | |
1771 | ||
d38ecf93 | 1772 | return 0; |
1da177e4 | 1773 | } |