]>
Commit | Line | Data |
---|---|---|
1 | // SPDX-License-Identifier: GPL-2.0 | |
2 | /* | |
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> | |
7 | * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au> | |
8 | * - July2000 | |
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 | */ | |
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> | |
20 | #include <linux/blk-mq.h> | |
21 | #include <linux/blk-pm.h> | |
22 | #include <linux/highmem.h> | |
23 | #include <linux/mm.h> | |
24 | #include <linux/pagemap.h> | |
25 | #include <linux/kernel_stat.h> | |
26 | #include <linux/string.h> | |
27 | #include <linux/init.h> | |
28 | #include <linux/completion.h> | |
29 | #include <linux/slab.h> | |
30 | #include <linux/swap.h> | |
31 | #include <linux/writeback.h> | |
32 | #include <linux/task_io_accounting_ops.h> | |
33 | #include <linux/fault-inject.h> | |
34 | #include <linux/list_sort.h> | |
35 | #include <linux/delay.h> | |
36 | #include <linux/ratelimit.h> | |
37 | #include <linux/pm_runtime.h> | |
38 | #include <linux/blk-cgroup.h> | |
39 | #include <linux/t10-pi.h> | |
40 | #include <linux/debugfs.h> | |
41 | #include <linux/bpf.h> | |
42 | #include <linux/psi.h> | |
43 | #include <linux/sched/sysctl.h> | |
44 | #include <linux/blk-crypto.h> | |
45 | ||
46 | #define CREATE_TRACE_POINTS | |
47 | #include <trace/events/block.h> | |
48 | ||
49 | #include "blk.h" | |
50 | #include "blk-mq.h" | |
51 | #include "blk-mq-sched.h" | |
52 | #include "blk-pm.h" | |
53 | #include "blk-rq-qos.h" | |
54 | ||
55 | struct dentry *blk_debugfs_root; | |
56 | ||
57 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap); | |
58 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap); | |
59 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete); | |
60 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_split); | |
61 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug); | |
62 | ||
63 | DEFINE_IDA(blk_queue_ida); | |
64 | ||
65 | /* | |
66 | * For queue allocation | |
67 | */ | |
68 | struct kmem_cache *blk_requestq_cachep; | |
69 | ||
70 | /* | |
71 | * Controlling structure to kblockd | |
72 | */ | |
73 | static struct workqueue_struct *kblockd_workqueue; | |
74 | ||
75 | /** | |
76 | * blk_queue_flag_set - atomically set a queue flag | |
77 | * @flag: flag to be set | |
78 | * @q: request queue | |
79 | */ | |
80 | void blk_queue_flag_set(unsigned int flag, struct request_queue *q) | |
81 | { | |
82 | set_bit(flag, &q->queue_flags); | |
83 | } | |
84 | EXPORT_SYMBOL(blk_queue_flag_set); | |
85 | ||
86 | /** | |
87 | * blk_queue_flag_clear - atomically clear a queue flag | |
88 | * @flag: flag to be cleared | |
89 | * @q: request queue | |
90 | */ | |
91 | void blk_queue_flag_clear(unsigned int flag, struct request_queue *q) | |
92 | { | |
93 | clear_bit(flag, &q->queue_flags); | |
94 | } | |
95 | EXPORT_SYMBOL(blk_queue_flag_clear); | |
96 | ||
97 | /** | |
98 | * blk_queue_flag_test_and_set - atomically test and set a queue flag | |
99 | * @flag: flag to be set | |
100 | * @q: request queue | |
101 | * | |
102 | * Returns the previous value of @flag - 0 if the flag was not set and 1 if | |
103 | * the flag was already set. | |
104 | */ | |
105 | bool blk_queue_flag_test_and_set(unsigned int flag, struct request_queue *q) | |
106 | { | |
107 | return test_and_set_bit(flag, &q->queue_flags); | |
108 | } | |
109 | EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set); | |
110 | ||
111 | void blk_rq_init(struct request_queue *q, struct request *rq) | |
112 | { | |
113 | memset(rq, 0, sizeof(*rq)); | |
114 | ||
115 | INIT_LIST_HEAD(&rq->queuelist); | |
116 | rq->q = q; | |
117 | rq->__sector = (sector_t) -1; | |
118 | INIT_HLIST_NODE(&rq->hash); | |
119 | RB_CLEAR_NODE(&rq->rb_node); | |
120 | rq->tag = BLK_MQ_NO_TAG; | |
121 | rq->internal_tag = BLK_MQ_NO_TAG; | |
122 | rq->start_time_ns = ktime_get_ns(); | |
123 | rq->part = NULL; | |
124 | refcount_set(&rq->ref, 1); | |
125 | blk_crypto_rq_set_defaults(rq); | |
126 | } | |
127 | EXPORT_SYMBOL(blk_rq_init); | |
128 | ||
129 | #define REQ_OP_NAME(name) [REQ_OP_##name] = #name | |
130 | static const char *const blk_op_name[] = { | |
131 | REQ_OP_NAME(READ), | |
132 | REQ_OP_NAME(WRITE), | |
133 | REQ_OP_NAME(FLUSH), | |
134 | REQ_OP_NAME(DISCARD), | |
135 | REQ_OP_NAME(SECURE_ERASE), | |
136 | REQ_OP_NAME(ZONE_RESET), | |
137 | REQ_OP_NAME(ZONE_RESET_ALL), | |
138 | REQ_OP_NAME(ZONE_OPEN), | |
139 | REQ_OP_NAME(ZONE_CLOSE), | |
140 | REQ_OP_NAME(ZONE_FINISH), | |
141 | REQ_OP_NAME(ZONE_APPEND), | |
142 | REQ_OP_NAME(WRITE_SAME), | |
143 | REQ_OP_NAME(WRITE_ZEROES), | |
144 | REQ_OP_NAME(SCSI_IN), | |
145 | REQ_OP_NAME(SCSI_OUT), | |
146 | REQ_OP_NAME(DRV_IN), | |
147 | REQ_OP_NAME(DRV_OUT), | |
148 | }; | |
149 | #undef REQ_OP_NAME | |
150 | ||
151 | /** | |
152 | * blk_op_str - Return string XXX in the REQ_OP_XXX. | |
153 | * @op: REQ_OP_XXX. | |
154 | * | |
155 | * Description: Centralize block layer function to convert REQ_OP_XXX into | |
156 | * string format. Useful in the debugging and tracing bio or request. For | |
157 | * invalid REQ_OP_XXX it returns string "UNKNOWN". | |
158 | */ | |
159 | inline const char *blk_op_str(unsigned int op) | |
160 | { | |
161 | const char *op_str = "UNKNOWN"; | |
162 | ||
163 | if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op]) | |
164 | op_str = blk_op_name[op]; | |
165 | ||
166 | return op_str; | |
167 | } | |
168 | EXPORT_SYMBOL_GPL(blk_op_str); | |
169 | ||
170 | static const struct { | |
171 | int errno; | |
172 | const char *name; | |
173 | } blk_errors[] = { | |
174 | [BLK_STS_OK] = { 0, "" }, | |
175 | [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" }, | |
176 | [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" }, | |
177 | [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" }, | |
178 | [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" }, | |
179 | [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" }, | |
180 | [BLK_STS_NEXUS] = { -EBADE, "critical nexus" }, | |
181 | [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" }, | |
182 | [BLK_STS_PROTECTION] = { -EILSEQ, "protection" }, | |
183 | [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" }, | |
184 | [BLK_STS_DEV_RESOURCE] = { -EBUSY, "device resource" }, | |
185 | [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" }, | |
186 | ||
187 | /* device mapper special case, should not leak out: */ | |
188 | [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" }, | |
189 | ||
190 | /* zone device specific errors */ | |
191 | [BLK_STS_ZONE_OPEN_RESOURCE] = { -ETOOMANYREFS, "open zones exceeded" }, | |
192 | [BLK_STS_ZONE_ACTIVE_RESOURCE] = { -EOVERFLOW, "active zones exceeded" }, | |
193 | ||
194 | /* everything else not covered above: */ | |
195 | [BLK_STS_IOERR] = { -EIO, "I/O" }, | |
196 | }; | |
197 | ||
198 | blk_status_t errno_to_blk_status(int errno) | |
199 | { | |
200 | int i; | |
201 | ||
202 | for (i = 0; i < ARRAY_SIZE(blk_errors); i++) { | |
203 | if (blk_errors[i].errno == errno) | |
204 | return (__force blk_status_t)i; | |
205 | } | |
206 | ||
207 | return BLK_STS_IOERR; | |
208 | } | |
209 | EXPORT_SYMBOL_GPL(errno_to_blk_status); | |
210 | ||
211 | int blk_status_to_errno(blk_status_t status) | |
212 | { | |
213 | int idx = (__force int)status; | |
214 | ||
215 | if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors))) | |
216 | return -EIO; | |
217 | return blk_errors[idx].errno; | |
218 | } | |
219 | EXPORT_SYMBOL_GPL(blk_status_to_errno); | |
220 | ||
221 | static void print_req_error(struct request *req, blk_status_t status, | |
222 | const char *caller) | |
223 | { | |
224 | int idx = (__force int)status; | |
225 | ||
226 | if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors))) | |
227 | return; | |
228 | ||
229 | printk_ratelimited(KERN_ERR | |
230 | "%s: %s error, dev %s, sector %llu op 0x%x:(%s) flags 0x%x " | |
231 | "phys_seg %u prio class %u\n", | |
232 | caller, blk_errors[idx].name, | |
233 | req->rq_disk ? req->rq_disk->disk_name : "?", | |
234 | blk_rq_pos(req), req_op(req), blk_op_str(req_op(req)), | |
235 | req->cmd_flags & ~REQ_OP_MASK, | |
236 | req->nr_phys_segments, | |
237 | IOPRIO_PRIO_CLASS(req->ioprio)); | |
238 | } | |
239 | ||
240 | static void req_bio_endio(struct request *rq, struct bio *bio, | |
241 | unsigned int nbytes, blk_status_t error) | |
242 | { | |
243 | if (error) | |
244 | bio->bi_status = error; | |
245 | ||
246 | if (unlikely(rq->rq_flags & RQF_QUIET)) | |
247 | bio_set_flag(bio, BIO_QUIET); | |
248 | ||
249 | bio_advance(bio, nbytes); | |
250 | ||
251 | if (req_op(rq) == REQ_OP_ZONE_APPEND && error == BLK_STS_OK) { | |
252 | /* | |
253 | * Partial zone append completions cannot be supported as the | |
254 | * BIO fragments may end up not being written sequentially. | |
255 | */ | |
256 | if (bio->bi_iter.bi_size) | |
257 | bio->bi_status = BLK_STS_IOERR; | |
258 | else | |
259 | bio->bi_iter.bi_sector = rq->__sector; | |
260 | } | |
261 | ||
262 | /* don't actually finish bio if it's part of flush sequence */ | |
263 | if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ)) | |
264 | bio_endio(bio); | |
265 | } | |
266 | ||
267 | void blk_dump_rq_flags(struct request *rq, char *msg) | |
268 | { | |
269 | printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg, | |
270 | rq->rq_disk ? rq->rq_disk->disk_name : "?", | |
271 | (unsigned long long) rq->cmd_flags); | |
272 | ||
273 | printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n", | |
274 | (unsigned long long)blk_rq_pos(rq), | |
275 | blk_rq_sectors(rq), blk_rq_cur_sectors(rq)); | |
276 | printk(KERN_INFO " bio %p, biotail %p, len %u\n", | |
277 | rq->bio, rq->biotail, blk_rq_bytes(rq)); | |
278 | } | |
279 | EXPORT_SYMBOL(blk_dump_rq_flags); | |
280 | ||
281 | /** | |
282 | * blk_sync_queue - cancel any pending callbacks on a queue | |
283 | * @q: the queue | |
284 | * | |
285 | * Description: | |
286 | * The block layer may perform asynchronous callback activity | |
287 | * on a queue, such as calling the unplug function after a timeout. | |
288 | * A block device may call blk_sync_queue to ensure that any | |
289 | * such activity is cancelled, thus allowing it to release resources | |
290 | * that the callbacks might use. The caller must already have made sure | |
291 | * that its ->submit_bio will not re-add plugging prior to calling | |
292 | * this function. | |
293 | * | |
294 | * This function does not cancel any asynchronous activity arising | |
295 | * out of elevator or throttling code. That would require elevator_exit() | |
296 | * and blkcg_exit_queue() to be called with queue lock initialized. | |
297 | * | |
298 | */ | |
299 | void blk_sync_queue(struct request_queue *q) | |
300 | { | |
301 | del_timer_sync(&q->timeout); | |
302 | cancel_work_sync(&q->timeout_work); | |
303 | } | |
304 | EXPORT_SYMBOL(blk_sync_queue); | |
305 | ||
306 | /** | |
307 | * blk_set_pm_only - increment pm_only counter | |
308 | * @q: request queue pointer | |
309 | */ | |
310 | void blk_set_pm_only(struct request_queue *q) | |
311 | { | |
312 | atomic_inc(&q->pm_only); | |
313 | } | |
314 | EXPORT_SYMBOL_GPL(blk_set_pm_only); | |
315 | ||
316 | void blk_clear_pm_only(struct request_queue *q) | |
317 | { | |
318 | int pm_only; | |
319 | ||
320 | pm_only = atomic_dec_return(&q->pm_only); | |
321 | WARN_ON_ONCE(pm_only < 0); | |
322 | if (pm_only == 0) | |
323 | wake_up_all(&q->mq_freeze_wq); | |
324 | } | |
325 | EXPORT_SYMBOL_GPL(blk_clear_pm_only); | |
326 | ||
327 | /** | |
328 | * blk_put_queue - decrement the request_queue refcount | |
329 | * @q: the request_queue structure to decrement the refcount for | |
330 | * | |
331 | * Decrements the refcount of the request_queue kobject. When this reaches 0 | |
332 | * we'll have blk_release_queue() called. | |
333 | * | |
334 | * Context: Any context, but the last reference must not be dropped from | |
335 | * atomic context. | |
336 | */ | |
337 | void blk_put_queue(struct request_queue *q) | |
338 | { | |
339 | kobject_put(&q->kobj); | |
340 | } | |
341 | EXPORT_SYMBOL(blk_put_queue); | |
342 | ||
343 | void blk_set_queue_dying(struct request_queue *q) | |
344 | { | |
345 | blk_queue_flag_set(QUEUE_FLAG_DYING, q); | |
346 | ||
347 | /* | |
348 | * When queue DYING flag is set, we need to block new req | |
349 | * entering queue, so we call blk_freeze_queue_start() to | |
350 | * prevent I/O from crossing blk_queue_enter(). | |
351 | */ | |
352 | blk_freeze_queue_start(q); | |
353 | ||
354 | if (queue_is_mq(q)) | |
355 | blk_mq_wake_waiters(q); | |
356 | ||
357 | /* Make blk_queue_enter() reexamine the DYING flag. */ | |
358 | wake_up_all(&q->mq_freeze_wq); | |
359 | } | |
360 | EXPORT_SYMBOL_GPL(blk_set_queue_dying); | |
361 | ||
362 | /** | |
363 | * blk_cleanup_queue - shutdown a request queue | |
364 | * @q: request queue to shutdown | |
365 | * | |
366 | * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and | |
367 | * put it. All future requests will be failed immediately with -ENODEV. | |
368 | * | |
369 | * Context: can sleep | |
370 | */ | |
371 | void blk_cleanup_queue(struct request_queue *q) | |
372 | { | |
373 | /* cannot be called from atomic context */ | |
374 | might_sleep(); | |
375 | ||
376 | WARN_ON_ONCE(blk_queue_registered(q)); | |
377 | ||
378 | /* mark @q DYING, no new request or merges will be allowed afterwards */ | |
379 | blk_set_queue_dying(q); | |
380 | ||
381 | blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q); | |
382 | blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q); | |
383 | ||
384 | /* | |
385 | * Drain all requests queued before DYING marking. Set DEAD flag to | |
386 | * prevent that blk_mq_run_hw_queues() accesses the hardware queues | |
387 | * after draining finished. | |
388 | */ | |
389 | blk_freeze_queue(q); | |
390 | ||
391 | rq_qos_exit(q); | |
392 | ||
393 | blk_queue_flag_set(QUEUE_FLAG_DEAD, q); | |
394 | ||
395 | /* for synchronous bio-based driver finish in-flight integrity i/o */ | |
396 | blk_flush_integrity(); | |
397 | ||
398 | /* @q won't process any more request, flush async actions */ | |
399 | del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer); | |
400 | blk_sync_queue(q); | |
401 | ||
402 | if (queue_is_mq(q)) | |
403 | blk_mq_exit_queue(q); | |
404 | ||
405 | /* | |
406 | * In theory, request pool of sched_tags belongs to request queue. | |
407 | * However, the current implementation requires tag_set for freeing | |
408 | * requests, so free the pool now. | |
409 | * | |
410 | * Queue has become frozen, there can't be any in-queue requests, so | |
411 | * it is safe to free requests now. | |
412 | */ | |
413 | mutex_lock(&q->sysfs_lock); | |
414 | if (q->elevator) | |
415 | blk_mq_sched_free_requests(q); | |
416 | mutex_unlock(&q->sysfs_lock); | |
417 | ||
418 | percpu_ref_exit(&q->q_usage_counter); | |
419 | ||
420 | /* @q is and will stay empty, shutdown and put */ | |
421 | blk_put_queue(q); | |
422 | } | |
423 | EXPORT_SYMBOL(blk_cleanup_queue); | |
424 | ||
425 | /** | |
426 | * blk_queue_enter() - try to increase q->q_usage_counter | |
427 | * @q: request queue pointer | |
428 | * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PM | |
429 | */ | |
430 | int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags) | |
431 | { | |
432 | const bool pm = flags & BLK_MQ_REQ_PM; | |
433 | ||
434 | while (true) { | |
435 | bool success = false; | |
436 | ||
437 | rcu_read_lock(); | |
438 | if (percpu_ref_tryget_live(&q->q_usage_counter)) { | |
439 | /* | |
440 | * The code that increments the pm_only counter is | |
441 | * responsible for ensuring that that counter is | |
442 | * globally visible before the queue is unfrozen. | |
443 | */ | |
444 | if ((pm && queue_rpm_status(q) != RPM_SUSPENDED) || | |
445 | !blk_queue_pm_only(q)) { | |
446 | success = true; | |
447 | } else { | |
448 | percpu_ref_put(&q->q_usage_counter); | |
449 | } | |
450 | } | |
451 | rcu_read_unlock(); | |
452 | ||
453 | if (success) | |
454 | return 0; | |
455 | ||
456 | if (flags & BLK_MQ_REQ_NOWAIT) | |
457 | return -EBUSY; | |
458 | ||
459 | /* | |
460 | * read pair of barrier in blk_freeze_queue_start(), | |
461 | * we need to order reading __PERCPU_REF_DEAD flag of | |
462 | * .q_usage_counter and reading .mq_freeze_depth or | |
463 | * queue dying flag, otherwise the following wait may | |
464 | * never return if the two reads are reordered. | |
465 | */ | |
466 | smp_rmb(); | |
467 | ||
468 | wait_event(q->mq_freeze_wq, | |
469 | (!q->mq_freeze_depth && | |
470 | blk_pm_resume_queue(pm, q)) || | |
471 | blk_queue_dying(q)); | |
472 | if (blk_queue_dying(q)) | |
473 | return -ENODEV; | |
474 | } | |
475 | } | |
476 | ||
477 | static inline int bio_queue_enter(struct bio *bio) | |
478 | { | |
479 | struct request_queue *q = bio->bi_disk->queue; | |
480 | bool nowait = bio->bi_opf & REQ_NOWAIT; | |
481 | int ret; | |
482 | ||
483 | ret = blk_queue_enter(q, nowait ? BLK_MQ_REQ_NOWAIT : 0); | |
484 | if (unlikely(ret)) { | |
485 | if (nowait && !blk_queue_dying(q)) | |
486 | bio_wouldblock_error(bio); | |
487 | else | |
488 | bio_io_error(bio); | |
489 | } | |
490 | ||
491 | return ret; | |
492 | } | |
493 | ||
494 | void blk_queue_exit(struct request_queue *q) | |
495 | { | |
496 | percpu_ref_put(&q->q_usage_counter); | |
497 | } | |
498 | ||
499 | static void blk_queue_usage_counter_release(struct percpu_ref *ref) | |
500 | { | |
501 | struct request_queue *q = | |
502 | container_of(ref, struct request_queue, q_usage_counter); | |
503 | ||
504 | wake_up_all(&q->mq_freeze_wq); | |
505 | } | |
506 | ||
507 | static void blk_rq_timed_out_timer(struct timer_list *t) | |
508 | { | |
509 | struct request_queue *q = from_timer(q, t, timeout); | |
510 | ||
511 | kblockd_schedule_work(&q->timeout_work); | |
512 | } | |
513 | ||
514 | static void blk_timeout_work(struct work_struct *work) | |
515 | { | |
516 | } | |
517 | ||
518 | struct request_queue *blk_alloc_queue(int node_id) | |
519 | { | |
520 | struct request_queue *q; | |
521 | int ret; | |
522 | ||
523 | q = kmem_cache_alloc_node(blk_requestq_cachep, | |
524 | GFP_KERNEL | __GFP_ZERO, node_id); | |
525 | if (!q) | |
526 | return NULL; | |
527 | ||
528 | q->last_merge = NULL; | |
529 | ||
530 | q->id = ida_simple_get(&blk_queue_ida, 0, 0, GFP_KERNEL); | |
531 | if (q->id < 0) | |
532 | goto fail_q; | |
533 | ||
534 | ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS); | |
535 | if (ret) | |
536 | goto fail_id; | |
537 | ||
538 | q->backing_dev_info = bdi_alloc(node_id); | |
539 | if (!q->backing_dev_info) | |
540 | goto fail_split; | |
541 | ||
542 | q->stats = blk_alloc_queue_stats(); | |
543 | if (!q->stats) | |
544 | goto fail_stats; | |
545 | ||
546 | q->node = node_id; | |
547 | ||
548 | atomic_set(&q->nr_active_requests_shared_sbitmap, 0); | |
549 | ||
550 | timer_setup(&q->backing_dev_info->laptop_mode_wb_timer, | |
551 | laptop_mode_timer_fn, 0); | |
552 | timer_setup(&q->timeout, blk_rq_timed_out_timer, 0); | |
553 | INIT_WORK(&q->timeout_work, blk_timeout_work); | |
554 | INIT_LIST_HEAD(&q->icq_list); | |
555 | #ifdef CONFIG_BLK_CGROUP | |
556 | INIT_LIST_HEAD(&q->blkg_list); | |
557 | #endif | |
558 | ||
559 | kobject_init(&q->kobj, &blk_queue_ktype); | |
560 | ||
561 | mutex_init(&q->debugfs_mutex); | |
562 | mutex_init(&q->sysfs_lock); | |
563 | mutex_init(&q->sysfs_dir_lock); | |
564 | spin_lock_init(&q->queue_lock); | |
565 | ||
566 | init_waitqueue_head(&q->mq_freeze_wq); | |
567 | mutex_init(&q->mq_freeze_lock); | |
568 | ||
569 | /* | |
570 | * Init percpu_ref in atomic mode so that it's faster to shutdown. | |
571 | * See blk_register_queue() for details. | |
572 | */ | |
573 | if (percpu_ref_init(&q->q_usage_counter, | |
574 | blk_queue_usage_counter_release, | |
575 | PERCPU_REF_INIT_ATOMIC, GFP_KERNEL)) | |
576 | goto fail_bdi; | |
577 | ||
578 | if (blkcg_init_queue(q)) | |
579 | goto fail_ref; | |
580 | ||
581 | blk_queue_dma_alignment(q, 511); | |
582 | blk_set_default_limits(&q->limits); | |
583 | q->nr_requests = BLKDEV_MAX_RQ; | |
584 | ||
585 | return q; | |
586 | ||
587 | fail_ref: | |
588 | percpu_ref_exit(&q->q_usage_counter); | |
589 | fail_bdi: | |
590 | blk_free_queue_stats(q->stats); | |
591 | fail_stats: | |
592 | bdi_put(q->backing_dev_info); | |
593 | fail_split: | |
594 | bioset_exit(&q->bio_split); | |
595 | fail_id: | |
596 | ida_simple_remove(&blk_queue_ida, q->id); | |
597 | fail_q: | |
598 | kmem_cache_free(blk_requestq_cachep, q); | |
599 | return NULL; | |
600 | } | |
601 | EXPORT_SYMBOL(blk_alloc_queue); | |
602 | ||
603 | /** | |
604 | * blk_get_queue - increment the request_queue refcount | |
605 | * @q: the request_queue structure to increment the refcount for | |
606 | * | |
607 | * Increment the refcount of the request_queue kobject. | |
608 | * | |
609 | * Context: Any context. | |
610 | */ | |
611 | bool blk_get_queue(struct request_queue *q) | |
612 | { | |
613 | if (likely(!blk_queue_dying(q))) { | |
614 | __blk_get_queue(q); | |
615 | return true; | |
616 | } | |
617 | ||
618 | return false; | |
619 | } | |
620 | EXPORT_SYMBOL(blk_get_queue); | |
621 | ||
622 | /** | |
623 | * blk_get_request - allocate a request | |
624 | * @q: request queue to allocate a request for | |
625 | * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC. | |
626 | * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT. | |
627 | */ | |
628 | struct request *blk_get_request(struct request_queue *q, unsigned int op, | |
629 | blk_mq_req_flags_t flags) | |
630 | { | |
631 | struct request *req; | |
632 | ||
633 | WARN_ON_ONCE(op & REQ_NOWAIT); | |
634 | WARN_ON_ONCE(flags & ~(BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_PM)); | |
635 | ||
636 | req = blk_mq_alloc_request(q, op, flags); | |
637 | if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn) | |
638 | q->mq_ops->initialize_rq_fn(req); | |
639 | ||
640 | return req; | |
641 | } | |
642 | EXPORT_SYMBOL(blk_get_request); | |
643 | ||
644 | void blk_put_request(struct request *req) | |
645 | { | |
646 | blk_mq_free_request(req); | |
647 | } | |
648 | EXPORT_SYMBOL(blk_put_request); | |
649 | ||
650 | static void handle_bad_sector(struct bio *bio, sector_t maxsector) | |
651 | { | |
652 | char b[BDEVNAME_SIZE]; | |
653 | ||
654 | pr_info_ratelimited("attempt to access beyond end of device\n" | |
655 | "%s: rw=%d, want=%llu, limit=%llu\n", | |
656 | bio_devname(bio, b), bio->bi_opf, | |
657 | bio_end_sector(bio), maxsector); | |
658 | } | |
659 | ||
660 | #ifdef CONFIG_FAIL_MAKE_REQUEST | |
661 | ||
662 | static DECLARE_FAULT_ATTR(fail_make_request); | |
663 | ||
664 | static int __init setup_fail_make_request(char *str) | |
665 | { | |
666 | return setup_fault_attr(&fail_make_request, str); | |
667 | } | |
668 | __setup("fail_make_request=", setup_fail_make_request); | |
669 | ||
670 | static bool should_fail_request(struct block_device *part, unsigned int bytes) | |
671 | { | |
672 | return part->bd_make_it_fail && should_fail(&fail_make_request, bytes); | |
673 | } | |
674 | ||
675 | static int __init fail_make_request_debugfs(void) | |
676 | { | |
677 | struct dentry *dir = fault_create_debugfs_attr("fail_make_request", | |
678 | NULL, &fail_make_request); | |
679 | ||
680 | return PTR_ERR_OR_ZERO(dir); | |
681 | } | |
682 | ||
683 | late_initcall(fail_make_request_debugfs); | |
684 | ||
685 | #else /* CONFIG_FAIL_MAKE_REQUEST */ | |
686 | ||
687 | static inline bool should_fail_request(struct block_device *part, | |
688 | unsigned int bytes) | |
689 | { | |
690 | return false; | |
691 | } | |
692 | ||
693 | #endif /* CONFIG_FAIL_MAKE_REQUEST */ | |
694 | ||
695 | static inline bool bio_check_ro(struct bio *bio, struct block_device *part) | |
696 | { | |
697 | const int op = bio_op(bio); | |
698 | ||
699 | if (part->bd_read_only && op_is_write(op)) { | |
700 | char b[BDEVNAME_SIZE]; | |
701 | ||
702 | if (op_is_flush(bio->bi_opf) && !bio_sectors(bio)) | |
703 | return false; | |
704 | ||
705 | WARN_ONCE(1, | |
706 | "Trying to write to read-only block-device %s (partno %d)\n", | |
707 | bio_devname(bio, b), part->bd_partno); | |
708 | /* Older lvm-tools actually trigger this */ | |
709 | return false; | |
710 | } | |
711 | ||
712 | return false; | |
713 | } | |
714 | ||
715 | static noinline int should_fail_bio(struct bio *bio) | |
716 | { | |
717 | if (should_fail_request(bio->bi_disk->part0, bio->bi_iter.bi_size)) | |
718 | return -EIO; | |
719 | return 0; | |
720 | } | |
721 | ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO); | |
722 | ||
723 | /* | |
724 | * Check whether this bio extends beyond the end of the device or partition. | |
725 | * This may well happen - the kernel calls bread() without checking the size of | |
726 | * the device, e.g., when mounting a file system. | |
727 | */ | |
728 | static inline int bio_check_eod(struct bio *bio, sector_t maxsector) | |
729 | { | |
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 | ||
741 | /* | |
742 | * Remap block n of partition p to block n+start(p) of the disk. | |
743 | */ | |
744 | static inline int blk_partition_remap(struct bio *bio) | |
745 | { | |
746 | struct block_device *p; | |
747 | int ret = -EIO; | |
748 | ||
749 | rcu_read_lock(); | |
750 | p = __disk_get_part(bio->bi_disk, bio->bi_partno); | |
751 | if (unlikely(!p)) | |
752 | goto out; | |
753 | if (unlikely(should_fail_request(p, bio->bi_iter.bi_size))) | |
754 | goto out; | |
755 | if (unlikely(bio_check_ro(bio, p))) | |
756 | goto out; | |
757 | ||
758 | if (bio_sectors(bio)) { | |
759 | if (bio_check_eod(bio, bdev_nr_sectors(p))) | |
760 | goto out; | |
761 | bio->bi_iter.bi_sector += p->bd_start_sect; | |
762 | trace_block_bio_remap(bio, p->bd_dev, | |
763 | bio->bi_iter.bi_sector - | |
764 | p->bd_start_sect); | |
765 | } | |
766 | bio->bi_partno = 0; | |
767 | ret = 0; | |
768 | out: | |
769 | rcu_read_unlock(); | |
770 | return ret; | |
771 | } | |
772 | ||
773 | /* | |
774 | * Check write append to a zoned block device. | |
775 | */ | |
776 | static inline blk_status_t blk_check_zone_append(struct request_queue *q, | |
777 | struct bio *bio) | |
778 | { | |
779 | sector_t pos = bio->bi_iter.bi_sector; | |
780 | int nr_sectors = bio_sectors(bio); | |
781 | ||
782 | /* Only applicable to zoned block devices */ | |
783 | if (!blk_queue_is_zoned(q)) | |
784 | return BLK_STS_NOTSUPP; | |
785 | ||
786 | /* The bio sector must point to the start of a sequential zone */ | |
787 | if (pos & (blk_queue_zone_sectors(q) - 1) || | |
788 | !blk_queue_zone_is_seq(q, pos)) | |
789 | return BLK_STS_IOERR; | |
790 | ||
791 | /* | |
792 | * Not allowed to cross zone boundaries. Otherwise, the BIO will be | |
793 | * split and could result in non-contiguous sectors being written in | |
794 | * different zones. | |
795 | */ | |
796 | if (nr_sectors > q->limits.chunk_sectors) | |
797 | return BLK_STS_IOERR; | |
798 | ||
799 | /* Make sure the BIO is small enough and will not get split */ | |
800 | if (nr_sectors > q->limits.max_zone_append_sectors) | |
801 | return BLK_STS_IOERR; | |
802 | ||
803 | bio->bi_opf |= REQ_NOMERGE; | |
804 | ||
805 | return BLK_STS_OK; | |
806 | } | |
807 | ||
808 | static noinline_for_stack bool submit_bio_checks(struct bio *bio) | |
809 | { | |
810 | struct request_queue *q = bio->bi_disk->queue; | |
811 | blk_status_t status = BLK_STS_IOERR; | |
812 | struct blk_plug *plug; | |
813 | ||
814 | might_sleep(); | |
815 | ||
816 | plug = blk_mq_plug(q, bio); | |
817 | if (plug && plug->nowait) | |
818 | bio->bi_opf |= REQ_NOWAIT; | |
819 | ||
820 | /* | |
821 | * For a REQ_NOWAIT based request, return -EOPNOTSUPP | |
822 | * if queue does not support NOWAIT. | |
823 | */ | |
824 | if ((bio->bi_opf & REQ_NOWAIT) && !blk_queue_nowait(q)) | |
825 | goto not_supported; | |
826 | ||
827 | if (should_fail_bio(bio)) | |
828 | goto end_io; | |
829 | ||
830 | if (bio->bi_partno) { | |
831 | if (unlikely(blk_partition_remap(bio))) | |
832 | goto end_io; | |
833 | } else { | |
834 | if (unlikely(bio_check_ro(bio, bio->bi_disk->part0))) | |
835 | goto end_io; | |
836 | if (unlikely(bio_check_eod(bio, get_capacity(bio->bi_disk)))) | |
837 | goto end_io; | |
838 | } | |
839 | ||
840 | /* | |
841 | * Filter flush bio's early so that bio based drivers without flush | |
842 | * support don't have to worry about them. | |
843 | */ | |
844 | if (op_is_flush(bio->bi_opf) && | |
845 | !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) { | |
846 | bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA); | |
847 | if (!bio_sectors(bio)) { | |
848 | status = BLK_STS_OK; | |
849 | goto end_io; | |
850 | } | |
851 | } | |
852 | ||
853 | if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags)) | |
854 | bio->bi_opf &= ~REQ_HIPRI; | |
855 | ||
856 | switch (bio_op(bio)) { | |
857 | case REQ_OP_DISCARD: | |
858 | if (!blk_queue_discard(q)) | |
859 | goto not_supported; | |
860 | break; | |
861 | case REQ_OP_SECURE_ERASE: | |
862 | if (!blk_queue_secure_erase(q)) | |
863 | goto not_supported; | |
864 | break; | |
865 | case REQ_OP_WRITE_SAME: | |
866 | if (!q->limits.max_write_same_sectors) | |
867 | goto not_supported; | |
868 | break; | |
869 | case REQ_OP_ZONE_APPEND: | |
870 | status = blk_check_zone_append(q, bio); | |
871 | if (status != BLK_STS_OK) | |
872 | goto end_io; | |
873 | break; | |
874 | case REQ_OP_ZONE_RESET: | |
875 | case REQ_OP_ZONE_OPEN: | |
876 | case REQ_OP_ZONE_CLOSE: | |
877 | case REQ_OP_ZONE_FINISH: | |
878 | if (!blk_queue_is_zoned(q)) | |
879 | goto not_supported; | |
880 | break; | |
881 | case REQ_OP_ZONE_RESET_ALL: | |
882 | if (!blk_queue_is_zoned(q) || !blk_queue_zone_resetall(q)) | |
883 | goto not_supported; | |
884 | break; | |
885 | case REQ_OP_WRITE_ZEROES: | |
886 | if (!q->limits.max_write_zeroes_sectors) | |
887 | goto not_supported; | |
888 | break; | |
889 | default: | |
890 | break; | |
891 | } | |
892 | ||
893 | /* | |
894 | * Various block parts want %current->io_context, so allocate it up | |
895 | * front rather than dealing with lots of pain to allocate it only | |
896 | * where needed. This may fail and the block layer knows how to live | |
897 | * with it. | |
898 | */ | |
899 | if (unlikely(!current->io_context)) | |
900 | create_task_io_context(current, GFP_ATOMIC, q->node); | |
901 | ||
902 | if (blk_throtl_bio(bio)) { | |
903 | blkcg_bio_issue_init(bio); | |
904 | return false; | |
905 | } | |
906 | ||
907 | blk_cgroup_bio_start(bio); | |
908 | blkcg_bio_issue_init(bio); | |
909 | ||
910 | if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) { | |
911 | trace_block_bio_queue(bio); | |
912 | /* Now that enqueuing has been traced, we need to trace | |
913 | * completion as well. | |
914 | */ | |
915 | bio_set_flag(bio, BIO_TRACE_COMPLETION); | |
916 | } | |
917 | return true; | |
918 | ||
919 | not_supported: | |
920 | status = BLK_STS_NOTSUPP; | |
921 | end_io: | |
922 | bio->bi_status = status; | |
923 | bio_endio(bio); | |
924 | return false; | |
925 | } | |
926 | ||
927 | static blk_qc_t __submit_bio(struct bio *bio) | |
928 | { | |
929 | struct gendisk *disk = bio->bi_disk; | |
930 | blk_qc_t ret = BLK_QC_T_NONE; | |
931 | ||
932 | if (blk_crypto_bio_prep(&bio)) { | |
933 | if (!disk->fops->submit_bio) | |
934 | return blk_mq_submit_bio(bio); | |
935 | ret = disk->fops->submit_bio(bio); | |
936 | } | |
937 | blk_queue_exit(disk->queue); | |
938 | return ret; | |
939 | } | |
940 | ||
941 | /* | |
942 | * The loop in this function may be a bit non-obvious, and so deserves some | |
943 | * explanation: | |
944 | * | |
945 | * - Before entering the loop, bio->bi_next is NULL (as all callers ensure | |
946 | * that), so we have a list with a single bio. | |
947 | * - We pretend that we have just taken it off a longer list, so we assign | |
948 | * bio_list to a pointer to the bio_list_on_stack, thus initialising the | |
949 | * bio_list of new bios to be added. ->submit_bio() may indeed add some more | |
950 | * bios through a recursive call to submit_bio_noacct. If it did, we find a | |
951 | * non-NULL value in bio_list and re-enter the loop from the top. | |
952 | * - In this case we really did just take the bio of the top of the list (no | |
953 | * pretending) and so remove it from bio_list, and call into ->submit_bio() | |
954 | * again. | |
955 | * | |
956 | * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio. | |
957 | * bio_list_on_stack[1] contains bios that were submitted before the current | |
958 | * ->submit_bio_bio, but that haven't been processed yet. | |
959 | */ | |
960 | static blk_qc_t __submit_bio_noacct(struct bio *bio) | |
961 | { | |
962 | struct bio_list bio_list_on_stack[2]; | |
963 | blk_qc_t ret = BLK_QC_T_NONE; | |
964 | ||
965 | BUG_ON(bio->bi_next); | |
966 | ||
967 | bio_list_init(&bio_list_on_stack[0]); | |
968 | current->bio_list = bio_list_on_stack; | |
969 | ||
970 | do { | |
971 | struct request_queue *q = bio->bi_disk->queue; | |
972 | struct bio_list lower, same; | |
973 | ||
974 | if (unlikely(bio_queue_enter(bio) != 0)) | |
975 | continue; | |
976 | ||
977 | /* | |
978 | * Create a fresh bio_list for all subordinate requests. | |
979 | */ | |
980 | bio_list_on_stack[1] = bio_list_on_stack[0]; | |
981 | bio_list_init(&bio_list_on_stack[0]); | |
982 | ||
983 | ret = __submit_bio(bio); | |
984 | ||
985 | /* | |
986 | * Sort new bios into those for a lower level and those for the | |
987 | * same level. | |
988 | */ | |
989 | bio_list_init(&lower); | |
990 | bio_list_init(&same); | |
991 | while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL) | |
992 | if (q == bio->bi_disk->queue) | |
993 | bio_list_add(&same, bio); | |
994 | else | |
995 | bio_list_add(&lower, bio); | |
996 | ||
997 | /* | |
998 | * Now assemble so we handle the lowest level first. | |
999 | */ | |
1000 | bio_list_merge(&bio_list_on_stack[0], &lower); | |
1001 | bio_list_merge(&bio_list_on_stack[0], &same); | |
1002 | bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]); | |
1003 | } while ((bio = bio_list_pop(&bio_list_on_stack[0]))); | |
1004 | ||
1005 | current->bio_list = NULL; | |
1006 | return ret; | |
1007 | } | |
1008 | ||
1009 | static blk_qc_t __submit_bio_noacct_mq(struct bio *bio) | |
1010 | { | |
1011 | struct bio_list bio_list[2] = { }; | |
1012 | blk_qc_t ret = BLK_QC_T_NONE; | |
1013 | ||
1014 | current->bio_list = bio_list; | |
1015 | ||
1016 | do { | |
1017 | struct gendisk *disk = bio->bi_disk; | |
1018 | ||
1019 | if (unlikely(bio_queue_enter(bio) != 0)) | |
1020 | continue; | |
1021 | ||
1022 | if (!blk_crypto_bio_prep(&bio)) { | |
1023 | blk_queue_exit(disk->queue); | |
1024 | ret = BLK_QC_T_NONE; | |
1025 | continue; | |
1026 | } | |
1027 | ||
1028 | ret = blk_mq_submit_bio(bio); | |
1029 | } while ((bio = bio_list_pop(&bio_list[0]))); | |
1030 | ||
1031 | current->bio_list = NULL; | |
1032 | return ret; | |
1033 | } | |
1034 | ||
1035 | /** | |
1036 | * submit_bio_noacct - re-submit a bio to the block device layer for I/O | |
1037 | * @bio: The bio describing the location in memory and on the device. | |
1038 | * | |
1039 | * This is a version of submit_bio() that shall only be used for I/O that is | |
1040 | * resubmitted to lower level drivers by stacking block drivers. All file | |
1041 | * systems and other upper level users of the block layer should use | |
1042 | * submit_bio() instead. | |
1043 | */ | |
1044 | blk_qc_t submit_bio_noacct(struct bio *bio) | |
1045 | { | |
1046 | if (!submit_bio_checks(bio)) | |
1047 | return BLK_QC_T_NONE; | |
1048 | ||
1049 | /* | |
1050 | * We only want one ->submit_bio to be active at a time, else stack | |
1051 | * usage with stacked devices could be a problem. Use current->bio_list | |
1052 | * to collect a list of requests submited by a ->submit_bio method while | |
1053 | * it is active, and then process them after it returned. | |
1054 | */ | |
1055 | if (current->bio_list) { | |
1056 | bio_list_add(¤t->bio_list[0], bio); | |
1057 | return BLK_QC_T_NONE; | |
1058 | } | |
1059 | ||
1060 | if (!bio->bi_disk->fops->submit_bio) | |
1061 | return __submit_bio_noacct_mq(bio); | |
1062 | return __submit_bio_noacct(bio); | |
1063 | } | |
1064 | EXPORT_SYMBOL(submit_bio_noacct); | |
1065 | ||
1066 | /** | |
1067 | * submit_bio - submit a bio to the block device layer for I/O | |
1068 | * @bio: The &struct bio which describes the I/O | |
1069 | * | |
1070 | * submit_bio() is used to submit I/O requests to block devices. It is passed a | |
1071 | * fully set up &struct bio that describes the I/O that needs to be done. The | |
1072 | * bio will be send to the device described by the bi_disk and bi_partno fields. | |
1073 | * | |
1074 | * The success/failure status of the request, along with notification of | |
1075 | * completion, is delivered asynchronously through the ->bi_end_io() callback | |
1076 | * in @bio. The bio must NOT be touched by thecaller until ->bi_end_io() has | |
1077 | * been called. | |
1078 | */ | |
1079 | blk_qc_t submit_bio(struct bio *bio) | |
1080 | { | |
1081 | if (blkcg_punt_bio_submit(bio)) | |
1082 | return BLK_QC_T_NONE; | |
1083 | ||
1084 | /* | |
1085 | * If it's a regular read/write or a barrier with data attached, | |
1086 | * go through the normal accounting stuff before submission. | |
1087 | */ | |
1088 | if (bio_has_data(bio)) { | |
1089 | unsigned int count; | |
1090 | ||
1091 | if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME)) | |
1092 | count = queue_logical_block_size(bio->bi_disk->queue) >> 9; | |
1093 | else | |
1094 | count = bio_sectors(bio); | |
1095 | ||
1096 | if (op_is_write(bio_op(bio))) { | |
1097 | count_vm_events(PGPGOUT, count); | |
1098 | } else { | |
1099 | task_io_account_read(bio->bi_iter.bi_size); | |
1100 | count_vm_events(PGPGIN, count); | |
1101 | } | |
1102 | ||
1103 | if (unlikely(block_dump)) { | |
1104 | char b[BDEVNAME_SIZE]; | |
1105 | printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n", | |
1106 | current->comm, task_pid_nr(current), | |
1107 | op_is_write(bio_op(bio)) ? "WRITE" : "READ", | |
1108 | (unsigned long long)bio->bi_iter.bi_sector, | |
1109 | bio_devname(bio, b), count); | |
1110 | } | |
1111 | } | |
1112 | ||
1113 | /* | |
1114 | * If we're reading data that is part of the userspace workingset, count | |
1115 | * submission time as memory stall. When the device is congested, or | |
1116 | * the submitting cgroup IO-throttled, submission can be a significant | |
1117 | * part of overall IO time. | |
1118 | */ | |
1119 | if (unlikely(bio_op(bio) == REQ_OP_READ && | |
1120 | bio_flagged(bio, BIO_WORKINGSET))) { | |
1121 | unsigned long pflags; | |
1122 | blk_qc_t ret; | |
1123 | ||
1124 | psi_memstall_enter(&pflags); | |
1125 | ret = submit_bio_noacct(bio); | |
1126 | psi_memstall_leave(&pflags); | |
1127 | ||
1128 | return ret; | |
1129 | } | |
1130 | ||
1131 | return submit_bio_noacct(bio); | |
1132 | } | |
1133 | EXPORT_SYMBOL(submit_bio); | |
1134 | ||
1135 | /** | |
1136 | * blk_cloned_rq_check_limits - Helper function to check a cloned request | |
1137 | * for the new queue limits | |
1138 | * @q: the queue | |
1139 | * @rq: the request being checked | |
1140 | * | |
1141 | * Description: | |
1142 | * @rq may have been made based on weaker limitations of upper-level queues | |
1143 | * in request stacking drivers, and it may violate the limitation of @q. | |
1144 | * Since the block layer and the underlying device driver trust @rq | |
1145 | * after it is inserted to @q, it should be checked against @q before | |
1146 | * the insertion using this generic function. | |
1147 | * | |
1148 | * Request stacking drivers like request-based dm may change the queue | |
1149 | * limits when retrying requests on other queues. Those requests need | |
1150 | * to be checked against the new queue limits again during dispatch. | |
1151 | */ | |
1152 | static blk_status_t blk_cloned_rq_check_limits(struct request_queue *q, | |
1153 | struct request *rq) | |
1154 | { | |
1155 | unsigned int max_sectors = blk_queue_get_max_sectors(q, req_op(rq)); | |
1156 | ||
1157 | if (blk_rq_sectors(rq) > max_sectors) { | |
1158 | /* | |
1159 | * SCSI device does not have a good way to return if | |
1160 | * Write Same/Zero is actually supported. If a device rejects | |
1161 | * a non-read/write command (discard, write same,etc.) the | |
1162 | * low-level device driver will set the relevant queue limit to | |
1163 | * 0 to prevent blk-lib from issuing more of the offending | |
1164 | * operations. Commands queued prior to the queue limit being | |
1165 | * reset need to be completed with BLK_STS_NOTSUPP to avoid I/O | |
1166 | * errors being propagated to upper layers. | |
1167 | */ | |
1168 | if (max_sectors == 0) | |
1169 | return BLK_STS_NOTSUPP; | |
1170 | ||
1171 | printk(KERN_ERR "%s: over max size limit. (%u > %u)\n", | |
1172 | __func__, blk_rq_sectors(rq), max_sectors); | |
1173 | return BLK_STS_IOERR; | |
1174 | } | |
1175 | ||
1176 | /* | |
1177 | * queue's settings related to segment counting like q->bounce_pfn | |
1178 | * may differ from that of other stacking queues. | |
1179 | * Recalculate it to check the request correctly on this queue's | |
1180 | * limitation. | |
1181 | */ | |
1182 | rq->nr_phys_segments = blk_recalc_rq_segments(rq); | |
1183 | if (rq->nr_phys_segments > queue_max_segments(q)) { | |
1184 | printk(KERN_ERR "%s: over max segments limit. (%hu > %hu)\n", | |
1185 | __func__, rq->nr_phys_segments, queue_max_segments(q)); | |
1186 | return BLK_STS_IOERR; | |
1187 | } | |
1188 | ||
1189 | return BLK_STS_OK; | |
1190 | } | |
1191 | ||
1192 | /** | |
1193 | * blk_insert_cloned_request - Helper for stacking drivers to submit a request | |
1194 | * @q: the queue to submit the request | |
1195 | * @rq: the request being queued | |
1196 | */ | |
1197 | blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq) | |
1198 | { | |
1199 | blk_status_t ret; | |
1200 | ||
1201 | ret = blk_cloned_rq_check_limits(q, rq); | |
1202 | if (ret != BLK_STS_OK) | |
1203 | return ret; | |
1204 | ||
1205 | if (rq->rq_disk && | |
1206 | should_fail_request(rq->rq_disk->part0, blk_rq_bytes(rq))) | |
1207 | return BLK_STS_IOERR; | |
1208 | ||
1209 | if (blk_crypto_insert_cloned_request(rq)) | |
1210 | return BLK_STS_IOERR; | |
1211 | ||
1212 | if (blk_queue_io_stat(q)) | |
1213 | blk_account_io_start(rq); | |
1214 | ||
1215 | /* | |
1216 | * Since we have a scheduler attached on the top device, | |
1217 | * bypass a potential scheduler on the bottom device for | |
1218 | * insert. | |
1219 | */ | |
1220 | return blk_mq_request_issue_directly(rq, true); | |
1221 | } | |
1222 | EXPORT_SYMBOL_GPL(blk_insert_cloned_request); | |
1223 | ||
1224 | /** | |
1225 | * blk_rq_err_bytes - determine number of bytes till the next failure boundary | |
1226 | * @rq: request to examine | |
1227 | * | |
1228 | * Description: | |
1229 | * A request could be merge of IOs which require different failure | |
1230 | * handling. This function determines the number of bytes which | |
1231 | * can be failed from the beginning of the request without | |
1232 | * crossing into area which need to be retried further. | |
1233 | * | |
1234 | * Return: | |
1235 | * The number of bytes to fail. | |
1236 | */ | |
1237 | unsigned int blk_rq_err_bytes(const struct request *rq) | |
1238 | { | |
1239 | unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK; | |
1240 | unsigned int bytes = 0; | |
1241 | struct bio *bio; | |
1242 | ||
1243 | if (!(rq->rq_flags & RQF_MIXED_MERGE)) | |
1244 | return blk_rq_bytes(rq); | |
1245 | ||
1246 | /* | |
1247 | * Currently the only 'mixing' which can happen is between | |
1248 | * different fastfail types. We can safely fail portions | |
1249 | * which have all the failfast bits that the first one has - | |
1250 | * the ones which are at least as eager to fail as the first | |
1251 | * one. | |
1252 | */ | |
1253 | for (bio = rq->bio; bio; bio = bio->bi_next) { | |
1254 | if ((bio->bi_opf & ff) != ff) | |
1255 | break; | |
1256 | bytes += bio->bi_iter.bi_size; | |
1257 | } | |
1258 | ||
1259 | /* this could lead to infinite loop */ | |
1260 | BUG_ON(blk_rq_bytes(rq) && !bytes); | |
1261 | return bytes; | |
1262 | } | |
1263 | EXPORT_SYMBOL_GPL(blk_rq_err_bytes); | |
1264 | ||
1265 | static void update_io_ticks(struct block_device *part, unsigned long now, | |
1266 | bool end) | |
1267 | { | |
1268 | unsigned long stamp; | |
1269 | again: | |
1270 | stamp = READ_ONCE(part->bd_stamp); | |
1271 | if (unlikely(stamp != now)) { | |
1272 | if (likely(cmpxchg(&part->bd_stamp, stamp, now) == stamp)) | |
1273 | __part_stat_add(part, io_ticks, end ? now - stamp : 1); | |
1274 | } | |
1275 | if (part->bd_partno) { | |
1276 | part = bdev_whole(part); | |
1277 | goto again; | |
1278 | } | |
1279 | } | |
1280 | ||
1281 | static void blk_account_io_completion(struct request *req, unsigned int bytes) | |
1282 | { | |
1283 | if (req->part && blk_do_io_stat(req)) { | |
1284 | const int sgrp = op_stat_group(req_op(req)); | |
1285 | ||
1286 | part_stat_lock(); | |
1287 | part_stat_add(req->part, sectors[sgrp], bytes >> 9); | |
1288 | part_stat_unlock(); | |
1289 | } | |
1290 | } | |
1291 | ||
1292 | void blk_account_io_done(struct request *req, u64 now) | |
1293 | { | |
1294 | /* | |
1295 | * Account IO completion. flush_rq isn't accounted as a | |
1296 | * normal IO on queueing nor completion. Accounting the | |
1297 | * containing request is enough. | |
1298 | */ | |
1299 | if (req->part && blk_do_io_stat(req) && | |
1300 | !(req->rq_flags & RQF_FLUSH_SEQ)) { | |
1301 | const int sgrp = op_stat_group(req_op(req)); | |
1302 | ||
1303 | part_stat_lock(); | |
1304 | update_io_ticks(req->part, jiffies, true); | |
1305 | part_stat_inc(req->part, ios[sgrp]); | |
1306 | part_stat_add(req->part, nsecs[sgrp], now - req->start_time_ns); | |
1307 | part_stat_unlock(); | |
1308 | } | |
1309 | } | |
1310 | ||
1311 | void blk_account_io_start(struct request *rq) | |
1312 | { | |
1313 | if (!blk_do_io_stat(rq)) | |
1314 | return; | |
1315 | ||
1316 | rq->part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq)); | |
1317 | ||
1318 | part_stat_lock(); | |
1319 | update_io_ticks(rq->part, jiffies, false); | |
1320 | part_stat_unlock(); | |
1321 | } | |
1322 | ||
1323 | static unsigned long __part_start_io_acct(struct block_device *part, | |
1324 | unsigned int sectors, unsigned int op) | |
1325 | { | |
1326 | const int sgrp = op_stat_group(op); | |
1327 | unsigned long now = READ_ONCE(jiffies); | |
1328 | ||
1329 | part_stat_lock(); | |
1330 | update_io_ticks(part, now, false); | |
1331 | part_stat_inc(part, ios[sgrp]); | |
1332 | part_stat_add(part, sectors[sgrp], sectors); | |
1333 | part_stat_local_inc(part, in_flight[op_is_write(op)]); | |
1334 | part_stat_unlock(); | |
1335 | ||
1336 | return now; | |
1337 | } | |
1338 | ||
1339 | unsigned long part_start_io_acct(struct gendisk *disk, struct block_device **part, | |
1340 | struct bio *bio) | |
1341 | { | |
1342 | *part = disk_map_sector_rcu(disk, bio->bi_iter.bi_sector); | |
1343 | ||
1344 | return __part_start_io_acct(*part, bio_sectors(bio), bio_op(bio)); | |
1345 | } | |
1346 | EXPORT_SYMBOL_GPL(part_start_io_acct); | |
1347 | ||
1348 | unsigned long disk_start_io_acct(struct gendisk *disk, unsigned int sectors, | |
1349 | unsigned int op) | |
1350 | { | |
1351 | return __part_start_io_acct(disk->part0, sectors, op); | |
1352 | } | |
1353 | EXPORT_SYMBOL(disk_start_io_acct); | |
1354 | ||
1355 | static void __part_end_io_acct(struct block_device *part, unsigned int op, | |
1356 | unsigned long start_time) | |
1357 | { | |
1358 | const int sgrp = op_stat_group(op); | |
1359 | unsigned long now = READ_ONCE(jiffies); | |
1360 | unsigned long duration = now - start_time; | |
1361 | ||
1362 | part_stat_lock(); | |
1363 | update_io_ticks(part, now, true); | |
1364 | part_stat_add(part, nsecs[sgrp], jiffies_to_nsecs(duration)); | |
1365 | part_stat_local_dec(part, in_flight[op_is_write(op)]); | |
1366 | part_stat_unlock(); | |
1367 | } | |
1368 | ||
1369 | void part_end_io_acct(struct block_device *part, struct bio *bio, | |
1370 | unsigned long start_time) | |
1371 | { | |
1372 | __part_end_io_acct(part, bio_op(bio), start_time); | |
1373 | } | |
1374 | EXPORT_SYMBOL_GPL(part_end_io_acct); | |
1375 | ||
1376 | void disk_end_io_acct(struct gendisk *disk, unsigned int op, | |
1377 | unsigned long start_time) | |
1378 | { | |
1379 | __part_end_io_acct(disk->part0, op, start_time); | |
1380 | } | |
1381 | EXPORT_SYMBOL(disk_end_io_acct); | |
1382 | ||
1383 | /* | |
1384 | * Steal bios from a request and add them to a bio list. | |
1385 | * The request must not have been partially completed before. | |
1386 | */ | |
1387 | void blk_steal_bios(struct bio_list *list, struct request *rq) | |
1388 | { | |
1389 | if (rq->bio) { | |
1390 | if (list->tail) | |
1391 | list->tail->bi_next = rq->bio; | |
1392 | else | |
1393 | list->head = rq->bio; | |
1394 | list->tail = rq->biotail; | |
1395 | ||
1396 | rq->bio = NULL; | |
1397 | rq->biotail = NULL; | |
1398 | } | |
1399 | ||
1400 | rq->__data_len = 0; | |
1401 | } | |
1402 | EXPORT_SYMBOL_GPL(blk_steal_bios); | |
1403 | ||
1404 | /** | |
1405 | * blk_update_request - Special helper function for request stacking drivers | |
1406 | * @req: the request being processed | |
1407 | * @error: block status code | |
1408 | * @nr_bytes: number of bytes to complete @req | |
1409 | * | |
1410 | * Description: | |
1411 | * Ends I/O on a number of bytes attached to @req, but doesn't complete | |
1412 | * the request structure even if @req doesn't have leftover. | |
1413 | * If @req has leftover, sets it up for the next range of segments. | |
1414 | * | |
1415 | * This special helper function is only for request stacking drivers | |
1416 | * (e.g. request-based dm) so that they can handle partial completion. | |
1417 | * Actual device drivers should use blk_mq_end_request instead. | |
1418 | * | |
1419 | * Passing the result of blk_rq_bytes() as @nr_bytes guarantees | |
1420 | * %false return from this function. | |
1421 | * | |
1422 | * Note: | |
1423 | * The RQF_SPECIAL_PAYLOAD flag is ignored on purpose in both | |
1424 | * blk_rq_bytes() and in blk_update_request(). | |
1425 | * | |
1426 | * Return: | |
1427 | * %false - this request doesn't have any more data | |
1428 | * %true - this request has more data | |
1429 | **/ | |
1430 | bool blk_update_request(struct request *req, blk_status_t error, | |
1431 | unsigned int nr_bytes) | |
1432 | { | |
1433 | int total_bytes; | |
1434 | ||
1435 | trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes); | |
1436 | ||
1437 | if (!req->bio) | |
1438 | return false; | |
1439 | ||
1440 | #ifdef CONFIG_BLK_DEV_INTEGRITY | |
1441 | if (blk_integrity_rq(req) && req_op(req) == REQ_OP_READ && | |
1442 | error == BLK_STS_OK) | |
1443 | req->q->integrity.profile->complete_fn(req, nr_bytes); | |
1444 | #endif | |
1445 | ||
1446 | if (unlikely(error && !blk_rq_is_passthrough(req) && | |
1447 | !(req->rq_flags & RQF_QUIET))) | |
1448 | print_req_error(req, error, __func__); | |
1449 | ||
1450 | blk_account_io_completion(req, nr_bytes); | |
1451 | ||
1452 | total_bytes = 0; | |
1453 | while (req->bio) { | |
1454 | struct bio *bio = req->bio; | |
1455 | unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes); | |
1456 | ||
1457 | if (bio_bytes == bio->bi_iter.bi_size) | |
1458 | req->bio = bio->bi_next; | |
1459 | ||
1460 | /* Completion has already been traced */ | |
1461 | bio_clear_flag(bio, BIO_TRACE_COMPLETION); | |
1462 | req_bio_endio(req, bio, bio_bytes, error); | |
1463 | ||
1464 | total_bytes += bio_bytes; | |
1465 | nr_bytes -= bio_bytes; | |
1466 | ||
1467 | if (!nr_bytes) | |
1468 | break; | |
1469 | } | |
1470 | ||
1471 | /* | |
1472 | * completely done | |
1473 | */ | |
1474 | if (!req->bio) { | |
1475 | /* | |
1476 | * Reset counters so that the request stacking driver | |
1477 | * can find how many bytes remain in the request | |
1478 | * later. | |
1479 | */ | |
1480 | req->__data_len = 0; | |
1481 | return false; | |
1482 | } | |
1483 | ||
1484 | req->__data_len -= total_bytes; | |
1485 | ||
1486 | /* update sector only for requests with clear definition of sector */ | |
1487 | if (!blk_rq_is_passthrough(req)) | |
1488 | req->__sector += total_bytes >> 9; | |
1489 | ||
1490 | /* mixed attributes always follow the first bio */ | |
1491 | if (req->rq_flags & RQF_MIXED_MERGE) { | |
1492 | req->cmd_flags &= ~REQ_FAILFAST_MASK; | |
1493 | req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK; | |
1494 | } | |
1495 | ||
1496 | if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) { | |
1497 | /* | |
1498 | * If total number of sectors is less than the first segment | |
1499 | * size, something has gone terribly wrong. | |
1500 | */ | |
1501 | if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) { | |
1502 | blk_dump_rq_flags(req, "request botched"); | |
1503 | req->__data_len = blk_rq_cur_bytes(req); | |
1504 | } | |
1505 | ||
1506 | /* recalculate the number of segments */ | |
1507 | req->nr_phys_segments = blk_recalc_rq_segments(req); | |
1508 | } | |
1509 | ||
1510 | return true; | |
1511 | } | |
1512 | EXPORT_SYMBOL_GPL(blk_update_request); | |
1513 | ||
1514 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE | |
1515 | /** | |
1516 | * rq_flush_dcache_pages - Helper function to flush all pages in a request | |
1517 | * @rq: the request to be flushed | |
1518 | * | |
1519 | * Description: | |
1520 | * Flush all pages in @rq. | |
1521 | */ | |
1522 | void rq_flush_dcache_pages(struct request *rq) | |
1523 | { | |
1524 | struct req_iterator iter; | |
1525 | struct bio_vec bvec; | |
1526 | ||
1527 | rq_for_each_segment(bvec, rq, iter) | |
1528 | flush_dcache_page(bvec.bv_page); | |
1529 | } | |
1530 | EXPORT_SYMBOL_GPL(rq_flush_dcache_pages); | |
1531 | #endif | |
1532 | ||
1533 | /** | |
1534 | * blk_lld_busy - Check if underlying low-level drivers of a device are busy | |
1535 | * @q : the queue of the device being checked | |
1536 | * | |
1537 | * Description: | |
1538 | * Check if underlying low-level drivers of a device are busy. | |
1539 | * If the drivers want to export their busy state, they must set own | |
1540 | * exporting function using blk_queue_lld_busy() first. | |
1541 | * | |
1542 | * Basically, this function is used only by request stacking drivers | |
1543 | * to stop dispatching requests to underlying devices when underlying | |
1544 | * devices are busy. This behavior helps more I/O merging on the queue | |
1545 | * of the request stacking driver and prevents I/O throughput regression | |
1546 | * on burst I/O load. | |
1547 | * | |
1548 | * Return: | |
1549 | * 0 - Not busy (The request stacking driver should dispatch request) | |
1550 | * 1 - Busy (The request stacking driver should stop dispatching request) | |
1551 | */ | |
1552 | int blk_lld_busy(struct request_queue *q) | |
1553 | { | |
1554 | if (queue_is_mq(q) && q->mq_ops->busy) | |
1555 | return q->mq_ops->busy(q); | |
1556 | ||
1557 | return 0; | |
1558 | } | |
1559 | EXPORT_SYMBOL_GPL(blk_lld_busy); | |
1560 | ||
1561 | /** | |
1562 | * blk_rq_unprep_clone - Helper function to free all bios in a cloned request | |
1563 | * @rq: the clone request to be cleaned up | |
1564 | * | |
1565 | * Description: | |
1566 | * Free all bios in @rq for a cloned request. | |
1567 | */ | |
1568 | void blk_rq_unprep_clone(struct request *rq) | |
1569 | { | |
1570 | struct bio *bio; | |
1571 | ||
1572 | while ((bio = rq->bio) != NULL) { | |
1573 | rq->bio = bio->bi_next; | |
1574 | ||
1575 | bio_put(bio); | |
1576 | } | |
1577 | } | |
1578 | EXPORT_SYMBOL_GPL(blk_rq_unprep_clone); | |
1579 | ||
1580 | /** | |
1581 | * blk_rq_prep_clone - Helper function to setup clone request | |
1582 | * @rq: the request to be setup | |
1583 | * @rq_src: original request to be cloned | |
1584 | * @bs: bio_set that bios for clone are allocated from | |
1585 | * @gfp_mask: memory allocation mask for bio | |
1586 | * @bio_ctr: setup function to be called for each clone bio. | |
1587 | * Returns %0 for success, non %0 for failure. | |
1588 | * @data: private data to be passed to @bio_ctr | |
1589 | * | |
1590 | * Description: | |
1591 | * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq. | |
1592 | * Also, pages which the original bios are pointing to are not copied | |
1593 | * and the cloned bios just point same pages. | |
1594 | * So cloned bios must be completed before original bios, which means | |
1595 | * the caller must complete @rq before @rq_src. | |
1596 | */ | |
1597 | int blk_rq_prep_clone(struct request *rq, struct request *rq_src, | |
1598 | struct bio_set *bs, gfp_t gfp_mask, | |
1599 | int (*bio_ctr)(struct bio *, struct bio *, void *), | |
1600 | void *data) | |
1601 | { | |
1602 | struct bio *bio, *bio_src; | |
1603 | ||
1604 | if (!bs) | |
1605 | bs = &fs_bio_set; | |
1606 | ||
1607 | __rq_for_each_bio(bio_src, rq_src) { | |
1608 | bio = bio_clone_fast(bio_src, gfp_mask, bs); | |
1609 | if (!bio) | |
1610 | goto free_and_out; | |
1611 | ||
1612 | if (bio_ctr && bio_ctr(bio, bio_src, data)) | |
1613 | goto free_and_out; | |
1614 | ||
1615 | if (rq->bio) { | |
1616 | rq->biotail->bi_next = bio; | |
1617 | rq->biotail = bio; | |
1618 | } else { | |
1619 | rq->bio = rq->biotail = bio; | |
1620 | } | |
1621 | bio = NULL; | |
1622 | } | |
1623 | ||
1624 | /* Copy attributes of the original request to the clone request. */ | |
1625 | rq->__sector = blk_rq_pos(rq_src); | |
1626 | rq->__data_len = blk_rq_bytes(rq_src); | |
1627 | if (rq_src->rq_flags & RQF_SPECIAL_PAYLOAD) { | |
1628 | rq->rq_flags |= RQF_SPECIAL_PAYLOAD; | |
1629 | rq->special_vec = rq_src->special_vec; | |
1630 | } | |
1631 | rq->nr_phys_segments = rq_src->nr_phys_segments; | |
1632 | rq->ioprio = rq_src->ioprio; | |
1633 | ||
1634 | if (rq->bio && blk_crypto_rq_bio_prep(rq, rq->bio, gfp_mask) < 0) | |
1635 | goto free_and_out; | |
1636 | ||
1637 | return 0; | |
1638 | ||
1639 | free_and_out: | |
1640 | if (bio) | |
1641 | bio_put(bio); | |
1642 | blk_rq_unprep_clone(rq); | |
1643 | ||
1644 | return -ENOMEM; | |
1645 | } | |
1646 | EXPORT_SYMBOL_GPL(blk_rq_prep_clone); | |
1647 | ||
1648 | int kblockd_schedule_work(struct work_struct *work) | |
1649 | { | |
1650 | return queue_work(kblockd_workqueue, work); | |
1651 | } | |
1652 | EXPORT_SYMBOL(kblockd_schedule_work); | |
1653 | ||
1654 | int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork, | |
1655 | unsigned long delay) | |
1656 | { | |
1657 | return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay); | |
1658 | } | |
1659 | EXPORT_SYMBOL(kblockd_mod_delayed_work_on); | |
1660 | ||
1661 | /** | |
1662 | * blk_start_plug - initialize blk_plug and track it inside the task_struct | |
1663 | * @plug: The &struct blk_plug that needs to be initialized | |
1664 | * | |
1665 | * Description: | |
1666 | * blk_start_plug() indicates to the block layer an intent by the caller | |
1667 | * to submit multiple I/O requests in a batch. The block layer may use | |
1668 | * this hint to defer submitting I/Os from the caller until blk_finish_plug() | |
1669 | * is called. However, the block layer may choose to submit requests | |
1670 | * before a call to blk_finish_plug() if the number of queued I/Os | |
1671 | * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than | |
1672 | * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if | |
1673 | * the task schedules (see below). | |
1674 | * | |
1675 | * Tracking blk_plug inside the task_struct will help with auto-flushing the | |
1676 | * pending I/O should the task end up blocking between blk_start_plug() and | |
1677 | * blk_finish_plug(). This is important from a performance perspective, but | |
1678 | * also ensures that we don't deadlock. For instance, if the task is blocking | |
1679 | * for a memory allocation, memory reclaim could end up wanting to free a | |
1680 | * page belonging to that request that is currently residing in our private | |
1681 | * plug. By flushing the pending I/O when the process goes to sleep, we avoid | |
1682 | * this kind of deadlock. | |
1683 | */ | |
1684 | void blk_start_plug(struct blk_plug *plug) | |
1685 | { | |
1686 | struct task_struct *tsk = current; | |
1687 | ||
1688 | /* | |
1689 | * If this is a nested plug, don't actually assign it. | |
1690 | */ | |
1691 | if (tsk->plug) | |
1692 | return; | |
1693 | ||
1694 | INIT_LIST_HEAD(&plug->mq_list); | |
1695 | INIT_LIST_HEAD(&plug->cb_list); | |
1696 | plug->rq_count = 0; | |
1697 | plug->multiple_queues = false; | |
1698 | plug->nowait = false; | |
1699 | ||
1700 | /* | |
1701 | * Store ordering should not be needed here, since a potential | |
1702 | * preempt will imply a full memory barrier | |
1703 | */ | |
1704 | tsk->plug = plug; | |
1705 | } | |
1706 | EXPORT_SYMBOL(blk_start_plug); | |
1707 | ||
1708 | static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule) | |
1709 | { | |
1710 | LIST_HEAD(callbacks); | |
1711 | ||
1712 | while (!list_empty(&plug->cb_list)) { | |
1713 | list_splice_init(&plug->cb_list, &callbacks); | |
1714 | ||
1715 | while (!list_empty(&callbacks)) { | |
1716 | struct blk_plug_cb *cb = list_first_entry(&callbacks, | |
1717 | struct blk_plug_cb, | |
1718 | list); | |
1719 | list_del(&cb->list); | |
1720 | cb->callback(cb, from_schedule); | |
1721 | } | |
1722 | } | |
1723 | } | |
1724 | ||
1725 | struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data, | |
1726 | int size) | |
1727 | { | |
1728 | struct blk_plug *plug = current->plug; | |
1729 | struct blk_plug_cb *cb; | |
1730 | ||
1731 | if (!plug) | |
1732 | return NULL; | |
1733 | ||
1734 | list_for_each_entry(cb, &plug->cb_list, list) | |
1735 | if (cb->callback == unplug && cb->data == data) | |
1736 | return cb; | |
1737 | ||
1738 | /* Not currently on the callback list */ | |
1739 | BUG_ON(size < sizeof(*cb)); | |
1740 | cb = kzalloc(size, GFP_ATOMIC); | |
1741 | if (cb) { | |
1742 | cb->data = data; | |
1743 | cb->callback = unplug; | |
1744 | list_add(&cb->list, &plug->cb_list); | |
1745 | } | |
1746 | return cb; | |
1747 | } | |
1748 | EXPORT_SYMBOL(blk_check_plugged); | |
1749 | ||
1750 | void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule) | |
1751 | { | |
1752 | flush_plug_callbacks(plug, from_schedule); | |
1753 | ||
1754 | if (!list_empty(&plug->mq_list)) | |
1755 | blk_mq_flush_plug_list(plug, from_schedule); | |
1756 | } | |
1757 | ||
1758 | /** | |
1759 | * blk_finish_plug - mark the end of a batch of submitted I/O | |
1760 | * @plug: The &struct blk_plug passed to blk_start_plug() | |
1761 | * | |
1762 | * Description: | |
1763 | * Indicate that a batch of I/O submissions is complete. This function | |
1764 | * must be paired with an initial call to blk_start_plug(). The intent | |
1765 | * is to allow the block layer to optimize I/O submission. See the | |
1766 | * documentation for blk_start_plug() for more information. | |
1767 | */ | |
1768 | void blk_finish_plug(struct blk_plug *plug) | |
1769 | { | |
1770 | if (plug != current->plug) | |
1771 | return; | |
1772 | blk_flush_plug_list(plug, false); | |
1773 | ||
1774 | current->plug = NULL; | |
1775 | } | |
1776 | EXPORT_SYMBOL(blk_finish_plug); | |
1777 | ||
1778 | void blk_io_schedule(void) | |
1779 | { | |
1780 | /* Prevent hang_check timer from firing at us during very long I/O */ | |
1781 | unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2; | |
1782 | ||
1783 | if (timeout) | |
1784 | io_schedule_timeout(timeout); | |
1785 | else | |
1786 | io_schedule(); | |
1787 | } | |
1788 | EXPORT_SYMBOL_GPL(blk_io_schedule); | |
1789 | ||
1790 | int __init blk_dev_init(void) | |
1791 | { | |
1792 | BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS)); | |
1793 | BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * | |
1794 | sizeof_field(struct request, cmd_flags)); | |
1795 | BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * | |
1796 | sizeof_field(struct bio, bi_opf)); | |
1797 | ||
1798 | /* used for unplugging and affects IO latency/throughput - HIGHPRI */ | |
1799 | kblockd_workqueue = alloc_workqueue("kblockd", | |
1800 | WQ_MEM_RECLAIM | WQ_HIGHPRI, 0); | |
1801 | if (!kblockd_workqueue) | |
1802 | panic("Failed to create kblockd\n"); | |
1803 | ||
1804 | blk_requestq_cachep = kmem_cache_create("request_queue", | |
1805 | sizeof(struct request_queue), 0, SLAB_PANIC, NULL); | |
1806 | ||
1807 | blk_debugfs_root = debugfs_create_dir("block", NULL); | |
1808 | ||
1809 | return 0; | |
1810 | } |