Loading...
1/*
2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
6 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
7 * - July2000
8 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
9 */
10
11/*
12 * This handles all read/write requests to block devices
13 */
14#include <linux/kernel.h>
15#include <linux/module.h>
16#include <linux/backing-dev.h>
17#include <linux/bio.h>
18#include <linux/blkdev.h>
19#include <linux/blk-mq.h>
20#include <linux/highmem.h>
21#include <linux/mm.h>
22#include <linux/kernel_stat.h>
23#include <linux/string.h>
24#include <linux/init.h>
25#include <linux/completion.h>
26#include <linux/slab.h>
27#include <linux/swap.h>
28#include <linux/writeback.h>
29#include <linux/task_io_accounting_ops.h>
30#include <linux/fault-inject.h>
31#include <linux/list_sort.h>
32#include <linux/delay.h>
33#include <linux/ratelimit.h>
34#include <linux/pm_runtime.h>
35#include <linux/blk-cgroup.h>
36#include <linux/debugfs.h>
37#include <linux/bpf.h>
38
39#define CREATE_TRACE_POINTS
40#include <trace/events/block.h>
41
42#include "blk.h"
43#include "blk-mq.h"
44#include "blk-mq-sched.h"
45#include "blk-wbt.h"
46
47#ifdef CONFIG_DEBUG_FS
48struct dentry *blk_debugfs_root;
49#endif
50
51EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
52EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
53EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
54EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
55EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
56
57DEFINE_IDA(blk_queue_ida);
58
59/*
60 * For the allocated request tables
61 */
62struct kmem_cache *request_cachep;
63
64/*
65 * For queue allocation
66 */
67struct kmem_cache *blk_requestq_cachep;
68
69/*
70 * Controlling structure to kblockd
71 */
72static struct workqueue_struct *kblockd_workqueue;
73
74/**
75 * blk_queue_flag_set - atomically set a queue flag
76 * @flag: flag to be set
77 * @q: request queue
78 */
79void blk_queue_flag_set(unsigned int flag, struct request_queue *q)
80{
81 unsigned long flags;
82
83 spin_lock_irqsave(q->queue_lock, flags);
84 queue_flag_set(flag, q);
85 spin_unlock_irqrestore(q->queue_lock, flags);
86}
87EXPORT_SYMBOL(blk_queue_flag_set);
88
89/**
90 * blk_queue_flag_clear - atomically clear a queue flag
91 * @flag: flag to be cleared
92 * @q: request queue
93 */
94void blk_queue_flag_clear(unsigned int flag, struct request_queue *q)
95{
96 unsigned long flags;
97
98 spin_lock_irqsave(q->queue_lock, flags);
99 queue_flag_clear(flag, q);
100 spin_unlock_irqrestore(q->queue_lock, flags);
101}
102EXPORT_SYMBOL(blk_queue_flag_clear);
103
104/**
105 * blk_queue_flag_test_and_set - atomically test and set a queue flag
106 * @flag: flag to be set
107 * @q: request queue
108 *
109 * Returns the previous value of @flag - 0 if the flag was not set and 1 if
110 * the flag was already set.
111 */
112bool blk_queue_flag_test_and_set(unsigned int flag, struct request_queue *q)
113{
114 unsigned long flags;
115 bool res;
116
117 spin_lock_irqsave(q->queue_lock, flags);
118 res = queue_flag_test_and_set(flag, q);
119 spin_unlock_irqrestore(q->queue_lock, flags);
120
121 return res;
122}
123EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set);
124
125/**
126 * blk_queue_flag_test_and_clear - atomically test and clear a queue flag
127 * @flag: flag to be cleared
128 * @q: request queue
129 *
130 * Returns the previous value of @flag - 0 if the flag was not set and 1 if
131 * the flag was set.
132 */
133bool blk_queue_flag_test_and_clear(unsigned int flag, struct request_queue *q)
134{
135 unsigned long flags;
136 bool res;
137
138 spin_lock_irqsave(q->queue_lock, flags);
139 res = queue_flag_test_and_clear(flag, q);
140 spin_unlock_irqrestore(q->queue_lock, flags);
141
142 return res;
143}
144EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_clear);
145
146static void blk_clear_congested(struct request_list *rl, int sync)
147{
148#ifdef CONFIG_CGROUP_WRITEBACK
149 clear_wb_congested(rl->blkg->wb_congested, sync);
150#else
151 /*
152 * If !CGROUP_WRITEBACK, all blkg's map to bdi->wb and we shouldn't
153 * flip its congestion state for events on other blkcgs.
154 */
155 if (rl == &rl->q->root_rl)
156 clear_wb_congested(rl->q->backing_dev_info->wb.congested, sync);
157#endif
158}
159
160static void blk_set_congested(struct request_list *rl, int sync)
161{
162#ifdef CONFIG_CGROUP_WRITEBACK
163 set_wb_congested(rl->blkg->wb_congested, sync);
164#else
165 /* see blk_clear_congested() */
166 if (rl == &rl->q->root_rl)
167 set_wb_congested(rl->q->backing_dev_info->wb.congested, sync);
168#endif
169}
170
171void blk_queue_congestion_threshold(struct request_queue *q)
172{
173 int nr;
174
175 nr = q->nr_requests - (q->nr_requests / 8) + 1;
176 if (nr > q->nr_requests)
177 nr = q->nr_requests;
178 q->nr_congestion_on = nr;
179
180 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
181 if (nr < 1)
182 nr = 1;
183 q->nr_congestion_off = nr;
184}
185
186void blk_rq_init(struct request_queue *q, struct request *rq)
187{
188 memset(rq, 0, sizeof(*rq));
189
190 INIT_LIST_HEAD(&rq->queuelist);
191 INIT_LIST_HEAD(&rq->timeout_list);
192 rq->cpu = -1;
193 rq->q = q;
194 rq->__sector = (sector_t) -1;
195 INIT_HLIST_NODE(&rq->hash);
196 RB_CLEAR_NODE(&rq->rb_node);
197 rq->tag = -1;
198 rq->internal_tag = -1;
199 rq->start_time = jiffies;
200 set_start_time_ns(rq);
201 rq->part = NULL;
202 seqcount_init(&rq->gstate_seq);
203 u64_stats_init(&rq->aborted_gstate_sync);
204 /*
205 * See comment of blk_mq_init_request
206 */
207 WRITE_ONCE(rq->gstate, MQ_RQ_GEN_INC);
208}
209EXPORT_SYMBOL(blk_rq_init);
210
211static const struct {
212 int errno;
213 const char *name;
214} blk_errors[] = {
215 [BLK_STS_OK] = { 0, "" },
216 [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" },
217 [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" },
218 [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" },
219 [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" },
220 [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" },
221 [BLK_STS_NEXUS] = { -EBADE, "critical nexus" },
222 [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" },
223 [BLK_STS_PROTECTION] = { -EILSEQ, "protection" },
224 [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" },
225 [BLK_STS_DEV_RESOURCE] = { -EBUSY, "device resource" },
226 [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" },
227
228 /* device mapper special case, should not leak out: */
229 [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" },
230
231 /* everything else not covered above: */
232 [BLK_STS_IOERR] = { -EIO, "I/O" },
233};
234
235blk_status_t errno_to_blk_status(int errno)
236{
237 int i;
238
239 for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
240 if (blk_errors[i].errno == errno)
241 return (__force blk_status_t)i;
242 }
243
244 return BLK_STS_IOERR;
245}
246EXPORT_SYMBOL_GPL(errno_to_blk_status);
247
248int blk_status_to_errno(blk_status_t status)
249{
250 int idx = (__force int)status;
251
252 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
253 return -EIO;
254 return blk_errors[idx].errno;
255}
256EXPORT_SYMBOL_GPL(blk_status_to_errno);
257
258static void print_req_error(struct request *req, blk_status_t status)
259{
260 int idx = (__force int)status;
261
262 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
263 return;
264
265 printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n",
266 __func__, blk_errors[idx].name, req->rq_disk ?
267 req->rq_disk->disk_name : "?",
268 (unsigned long long)blk_rq_pos(req));
269}
270
271static void req_bio_endio(struct request *rq, struct bio *bio,
272 unsigned int nbytes, blk_status_t error)
273{
274 if (error)
275 bio->bi_status = error;
276
277 if (unlikely(rq->rq_flags & RQF_QUIET))
278 bio_set_flag(bio, BIO_QUIET);
279
280 bio_advance(bio, nbytes);
281
282 /* don't actually finish bio if it's part of flush sequence */
283 if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ))
284 bio_endio(bio);
285}
286
287void blk_dump_rq_flags(struct request *rq, char *msg)
288{
289 printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg,
290 rq->rq_disk ? rq->rq_disk->disk_name : "?",
291 (unsigned long long) rq->cmd_flags);
292
293 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
294 (unsigned long long)blk_rq_pos(rq),
295 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
296 printk(KERN_INFO " bio %p, biotail %p, len %u\n",
297 rq->bio, rq->biotail, blk_rq_bytes(rq));
298}
299EXPORT_SYMBOL(blk_dump_rq_flags);
300
301static void blk_delay_work(struct work_struct *work)
302{
303 struct request_queue *q;
304
305 q = container_of(work, struct request_queue, delay_work.work);
306 spin_lock_irq(q->queue_lock);
307 __blk_run_queue(q);
308 spin_unlock_irq(q->queue_lock);
309}
310
311/**
312 * blk_delay_queue - restart queueing after defined interval
313 * @q: The &struct request_queue in question
314 * @msecs: Delay in msecs
315 *
316 * Description:
317 * Sometimes queueing needs to be postponed for a little while, to allow
318 * resources to come back. This function will make sure that queueing is
319 * restarted around the specified time.
320 */
321void blk_delay_queue(struct request_queue *q, unsigned long msecs)
322{
323 lockdep_assert_held(q->queue_lock);
324 WARN_ON_ONCE(q->mq_ops);
325
326 if (likely(!blk_queue_dead(q)))
327 queue_delayed_work(kblockd_workqueue, &q->delay_work,
328 msecs_to_jiffies(msecs));
329}
330EXPORT_SYMBOL(blk_delay_queue);
331
332/**
333 * blk_start_queue_async - asynchronously restart a previously stopped queue
334 * @q: The &struct request_queue in question
335 *
336 * Description:
337 * blk_start_queue_async() will clear the stop flag on the queue, and
338 * ensure that the request_fn for the queue is run from an async
339 * context.
340 **/
341void blk_start_queue_async(struct request_queue *q)
342{
343 lockdep_assert_held(q->queue_lock);
344 WARN_ON_ONCE(q->mq_ops);
345
346 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
347 blk_run_queue_async(q);
348}
349EXPORT_SYMBOL(blk_start_queue_async);
350
351/**
352 * blk_start_queue - restart a previously stopped queue
353 * @q: The &struct request_queue in question
354 *
355 * Description:
356 * blk_start_queue() will clear the stop flag on the queue, and call
357 * the request_fn for the queue if it was in a stopped state when
358 * entered. Also see blk_stop_queue().
359 **/
360void blk_start_queue(struct request_queue *q)
361{
362 lockdep_assert_held(q->queue_lock);
363 WARN_ON(!in_interrupt() && !irqs_disabled());
364 WARN_ON_ONCE(q->mq_ops);
365
366 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
367 __blk_run_queue(q);
368}
369EXPORT_SYMBOL(blk_start_queue);
370
371/**
372 * blk_stop_queue - stop a queue
373 * @q: The &struct request_queue in question
374 *
375 * Description:
376 * The Linux block layer assumes that a block driver will consume all
377 * entries on the request queue when the request_fn strategy is called.
378 * Often this will not happen, because of hardware limitations (queue
379 * depth settings). If a device driver gets a 'queue full' response,
380 * or if it simply chooses not to queue more I/O at one point, it can
381 * call this function to prevent the request_fn from being called until
382 * the driver has signalled it's ready to go again. This happens by calling
383 * blk_start_queue() to restart queue operations.
384 **/
385void blk_stop_queue(struct request_queue *q)
386{
387 lockdep_assert_held(q->queue_lock);
388 WARN_ON_ONCE(q->mq_ops);
389
390 cancel_delayed_work(&q->delay_work);
391 queue_flag_set(QUEUE_FLAG_STOPPED, q);
392}
393EXPORT_SYMBOL(blk_stop_queue);
394
395/**
396 * blk_sync_queue - cancel any pending callbacks on a queue
397 * @q: the queue
398 *
399 * Description:
400 * The block layer may perform asynchronous callback activity
401 * on a queue, such as calling the unplug function after a timeout.
402 * A block device may call blk_sync_queue to ensure that any
403 * such activity is cancelled, thus allowing it to release resources
404 * that the callbacks might use. The caller must already have made sure
405 * that its ->make_request_fn will not re-add plugging prior to calling
406 * this function.
407 *
408 * This function does not cancel any asynchronous activity arising
409 * out of elevator or throttling code. That would require elevator_exit()
410 * and blkcg_exit_queue() to be called with queue lock initialized.
411 *
412 */
413void blk_sync_queue(struct request_queue *q)
414{
415 del_timer_sync(&q->timeout);
416 cancel_work_sync(&q->timeout_work);
417
418 if (q->mq_ops) {
419 struct blk_mq_hw_ctx *hctx;
420 int i;
421
422 cancel_delayed_work_sync(&q->requeue_work);
423 queue_for_each_hw_ctx(q, hctx, i)
424 cancel_delayed_work_sync(&hctx->run_work);
425 } else {
426 cancel_delayed_work_sync(&q->delay_work);
427 }
428}
429EXPORT_SYMBOL(blk_sync_queue);
430
431/**
432 * blk_set_preempt_only - set QUEUE_FLAG_PREEMPT_ONLY
433 * @q: request queue pointer
434 *
435 * Returns the previous value of the PREEMPT_ONLY flag - 0 if the flag was not
436 * set and 1 if the flag was already set.
437 */
438int blk_set_preempt_only(struct request_queue *q)
439{
440 return blk_queue_flag_test_and_set(QUEUE_FLAG_PREEMPT_ONLY, q);
441}
442EXPORT_SYMBOL_GPL(blk_set_preempt_only);
443
444void blk_clear_preempt_only(struct request_queue *q)
445{
446 blk_queue_flag_clear(QUEUE_FLAG_PREEMPT_ONLY, q);
447 wake_up_all(&q->mq_freeze_wq);
448}
449EXPORT_SYMBOL_GPL(blk_clear_preempt_only);
450
451/**
452 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
453 * @q: The queue to run
454 *
455 * Description:
456 * Invoke request handling on a queue if there are any pending requests.
457 * May be used to restart request handling after a request has completed.
458 * This variant runs the queue whether or not the queue has been
459 * stopped. Must be called with the queue lock held and interrupts
460 * disabled. See also @blk_run_queue.
461 */
462inline void __blk_run_queue_uncond(struct request_queue *q)
463{
464 lockdep_assert_held(q->queue_lock);
465 WARN_ON_ONCE(q->mq_ops);
466
467 if (unlikely(blk_queue_dead(q)))
468 return;
469
470 /*
471 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
472 * the queue lock internally. As a result multiple threads may be
473 * running such a request function concurrently. Keep track of the
474 * number of active request_fn invocations such that blk_drain_queue()
475 * can wait until all these request_fn calls have finished.
476 */
477 q->request_fn_active++;
478 q->request_fn(q);
479 q->request_fn_active--;
480}
481EXPORT_SYMBOL_GPL(__blk_run_queue_uncond);
482
483/**
484 * __blk_run_queue - run a single device queue
485 * @q: The queue to run
486 *
487 * Description:
488 * See @blk_run_queue.
489 */
490void __blk_run_queue(struct request_queue *q)
491{
492 lockdep_assert_held(q->queue_lock);
493 WARN_ON_ONCE(q->mq_ops);
494
495 if (unlikely(blk_queue_stopped(q)))
496 return;
497
498 __blk_run_queue_uncond(q);
499}
500EXPORT_SYMBOL(__blk_run_queue);
501
502/**
503 * blk_run_queue_async - run a single device queue in workqueue context
504 * @q: The queue to run
505 *
506 * Description:
507 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
508 * of us.
509 *
510 * Note:
511 * Since it is not allowed to run q->delay_work after blk_cleanup_queue()
512 * has canceled q->delay_work, callers must hold the queue lock to avoid
513 * race conditions between blk_cleanup_queue() and blk_run_queue_async().
514 */
515void blk_run_queue_async(struct request_queue *q)
516{
517 lockdep_assert_held(q->queue_lock);
518 WARN_ON_ONCE(q->mq_ops);
519
520 if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
521 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
522}
523EXPORT_SYMBOL(blk_run_queue_async);
524
525/**
526 * blk_run_queue - run a single device queue
527 * @q: The queue to run
528 *
529 * Description:
530 * Invoke request handling on this queue, if it has pending work to do.
531 * May be used to restart queueing when a request has completed.
532 */
533void blk_run_queue(struct request_queue *q)
534{
535 unsigned long flags;
536
537 WARN_ON_ONCE(q->mq_ops);
538
539 spin_lock_irqsave(q->queue_lock, flags);
540 __blk_run_queue(q);
541 spin_unlock_irqrestore(q->queue_lock, flags);
542}
543EXPORT_SYMBOL(blk_run_queue);
544
545void blk_put_queue(struct request_queue *q)
546{
547 kobject_put(&q->kobj);
548}
549EXPORT_SYMBOL(blk_put_queue);
550
551/**
552 * __blk_drain_queue - drain requests from request_queue
553 * @q: queue to drain
554 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
555 *
556 * Drain requests from @q. If @drain_all is set, all requests are drained.
557 * If not, only ELVPRIV requests are drained. The caller is responsible
558 * for ensuring that no new requests which need to be drained are queued.
559 */
560static void __blk_drain_queue(struct request_queue *q, bool drain_all)
561 __releases(q->queue_lock)
562 __acquires(q->queue_lock)
563{
564 int i;
565
566 lockdep_assert_held(q->queue_lock);
567 WARN_ON_ONCE(q->mq_ops);
568
569 while (true) {
570 bool drain = false;
571
572 /*
573 * The caller might be trying to drain @q before its
574 * elevator is initialized.
575 */
576 if (q->elevator)
577 elv_drain_elevator(q);
578
579 blkcg_drain_queue(q);
580
581 /*
582 * This function might be called on a queue which failed
583 * driver init after queue creation or is not yet fully
584 * active yet. Some drivers (e.g. fd and loop) get unhappy
585 * in such cases. Kick queue iff dispatch queue has
586 * something on it and @q has request_fn set.
587 */
588 if (!list_empty(&q->queue_head) && q->request_fn)
589 __blk_run_queue(q);
590
591 drain |= q->nr_rqs_elvpriv;
592 drain |= q->request_fn_active;
593
594 /*
595 * Unfortunately, requests are queued at and tracked from
596 * multiple places and there's no single counter which can
597 * be drained. Check all the queues and counters.
598 */
599 if (drain_all) {
600 struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
601 drain |= !list_empty(&q->queue_head);
602 for (i = 0; i < 2; i++) {
603 drain |= q->nr_rqs[i];
604 drain |= q->in_flight[i];
605 if (fq)
606 drain |= !list_empty(&fq->flush_queue[i]);
607 }
608 }
609
610 if (!drain)
611 break;
612
613 spin_unlock_irq(q->queue_lock);
614
615 msleep(10);
616
617 spin_lock_irq(q->queue_lock);
618 }
619
620 /*
621 * With queue marked dead, any woken up waiter will fail the
622 * allocation path, so the wakeup chaining is lost and we're
623 * left with hung waiters. We need to wake up those waiters.
624 */
625 if (q->request_fn) {
626 struct request_list *rl;
627
628 blk_queue_for_each_rl(rl, q)
629 for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
630 wake_up_all(&rl->wait[i]);
631 }
632}
633
634void blk_drain_queue(struct request_queue *q)
635{
636 spin_lock_irq(q->queue_lock);
637 __blk_drain_queue(q, true);
638 spin_unlock_irq(q->queue_lock);
639}
640
641/**
642 * blk_queue_bypass_start - enter queue bypass mode
643 * @q: queue of interest
644 *
645 * In bypass mode, only the dispatch FIFO queue of @q is used. This
646 * function makes @q enter bypass mode and drains all requests which were
647 * throttled or issued before. On return, it's guaranteed that no request
648 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
649 * inside queue or RCU read lock.
650 */
651void blk_queue_bypass_start(struct request_queue *q)
652{
653 WARN_ON_ONCE(q->mq_ops);
654
655 spin_lock_irq(q->queue_lock);
656 q->bypass_depth++;
657 queue_flag_set(QUEUE_FLAG_BYPASS, q);
658 spin_unlock_irq(q->queue_lock);
659
660 /*
661 * Queues start drained. Skip actual draining till init is
662 * complete. This avoids lenghty delays during queue init which
663 * can happen many times during boot.
664 */
665 if (blk_queue_init_done(q)) {
666 spin_lock_irq(q->queue_lock);
667 __blk_drain_queue(q, false);
668 spin_unlock_irq(q->queue_lock);
669
670 /* ensure blk_queue_bypass() is %true inside RCU read lock */
671 synchronize_rcu();
672 }
673}
674EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
675
676/**
677 * blk_queue_bypass_end - leave queue bypass mode
678 * @q: queue of interest
679 *
680 * Leave bypass mode and restore the normal queueing behavior.
681 *
682 * Note: although blk_queue_bypass_start() is only called for blk-sq queues,
683 * this function is called for both blk-sq and blk-mq queues.
684 */
685void blk_queue_bypass_end(struct request_queue *q)
686{
687 spin_lock_irq(q->queue_lock);
688 if (!--q->bypass_depth)
689 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
690 WARN_ON_ONCE(q->bypass_depth < 0);
691 spin_unlock_irq(q->queue_lock);
692}
693EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
694
695void blk_set_queue_dying(struct request_queue *q)
696{
697 blk_queue_flag_set(QUEUE_FLAG_DYING, q);
698
699 /*
700 * When queue DYING flag is set, we need to block new req
701 * entering queue, so we call blk_freeze_queue_start() to
702 * prevent I/O from crossing blk_queue_enter().
703 */
704 blk_freeze_queue_start(q);
705
706 if (q->mq_ops)
707 blk_mq_wake_waiters(q);
708 else {
709 struct request_list *rl;
710
711 spin_lock_irq(q->queue_lock);
712 blk_queue_for_each_rl(rl, q) {
713 if (rl->rq_pool) {
714 wake_up_all(&rl->wait[BLK_RW_SYNC]);
715 wake_up_all(&rl->wait[BLK_RW_ASYNC]);
716 }
717 }
718 spin_unlock_irq(q->queue_lock);
719 }
720
721 /* Make blk_queue_enter() reexamine the DYING flag. */
722 wake_up_all(&q->mq_freeze_wq);
723}
724EXPORT_SYMBOL_GPL(blk_set_queue_dying);
725
726/**
727 * blk_cleanup_queue - shutdown a request queue
728 * @q: request queue to shutdown
729 *
730 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
731 * put it. All future requests will be failed immediately with -ENODEV.
732 */
733void blk_cleanup_queue(struct request_queue *q)
734{
735 spinlock_t *lock = q->queue_lock;
736
737 /* mark @q DYING, no new request or merges will be allowed afterwards */
738 mutex_lock(&q->sysfs_lock);
739 blk_set_queue_dying(q);
740 spin_lock_irq(lock);
741
742 /*
743 * A dying queue is permanently in bypass mode till released. Note
744 * that, unlike blk_queue_bypass_start(), we aren't performing
745 * synchronize_rcu() after entering bypass mode to avoid the delay
746 * as some drivers create and destroy a lot of queues while
747 * probing. This is still safe because blk_release_queue() will be
748 * called only after the queue refcnt drops to zero and nothing,
749 * RCU or not, would be traversing the queue by then.
750 */
751 q->bypass_depth++;
752 queue_flag_set(QUEUE_FLAG_BYPASS, q);
753
754 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
755 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
756 queue_flag_set(QUEUE_FLAG_DYING, q);
757 spin_unlock_irq(lock);
758 mutex_unlock(&q->sysfs_lock);
759
760 /*
761 * Drain all requests queued before DYING marking. Set DEAD flag to
762 * prevent that q->request_fn() gets invoked after draining finished.
763 */
764 blk_freeze_queue(q);
765 spin_lock_irq(lock);
766 queue_flag_set(QUEUE_FLAG_DEAD, q);
767 spin_unlock_irq(lock);
768
769 /*
770 * make sure all in-progress dispatch are completed because
771 * blk_freeze_queue() can only complete all requests, and
772 * dispatch may still be in-progress since we dispatch requests
773 * from more than one contexts
774 */
775 if (q->mq_ops)
776 blk_mq_quiesce_queue(q);
777
778 /* for synchronous bio-based driver finish in-flight integrity i/o */
779 blk_flush_integrity();
780
781 /* @q won't process any more request, flush async actions */
782 del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer);
783 blk_sync_queue(q);
784
785 /*
786 * I/O scheduler exit is only safe after the sysfs scheduler attribute
787 * has been removed.
788 */
789 WARN_ON_ONCE(q->kobj.state_in_sysfs);
790
791 /*
792 * Since the I/O scheduler exit code may access cgroup information,
793 * perform I/O scheduler exit before disassociating from the block
794 * cgroup controller.
795 */
796 if (q->elevator) {
797 ioc_clear_queue(q);
798 elevator_exit(q, q->elevator);
799 q->elevator = NULL;
800 }
801
802 /*
803 * Remove all references to @q from the block cgroup controller before
804 * restoring @q->queue_lock to avoid that restoring this pointer causes
805 * e.g. blkcg_print_blkgs() to crash.
806 */
807 blkcg_exit_queue(q);
808
809 /*
810 * Since the cgroup code may dereference the @q->backing_dev_info
811 * pointer, only decrease its reference count after having removed the
812 * association with the block cgroup controller.
813 */
814 bdi_put(q->backing_dev_info);
815
816 if (q->mq_ops)
817 blk_mq_free_queue(q);
818 percpu_ref_exit(&q->q_usage_counter);
819
820 spin_lock_irq(lock);
821 if (q->queue_lock != &q->__queue_lock)
822 q->queue_lock = &q->__queue_lock;
823 spin_unlock_irq(lock);
824
825 /* @q is and will stay empty, shutdown and put */
826 blk_put_queue(q);
827}
828EXPORT_SYMBOL(blk_cleanup_queue);
829
830/* Allocate memory local to the request queue */
831static void *alloc_request_simple(gfp_t gfp_mask, void *data)
832{
833 struct request_queue *q = data;
834
835 return kmem_cache_alloc_node(request_cachep, gfp_mask, q->node);
836}
837
838static void free_request_simple(void *element, void *data)
839{
840 kmem_cache_free(request_cachep, element);
841}
842
843static void *alloc_request_size(gfp_t gfp_mask, void *data)
844{
845 struct request_queue *q = data;
846 struct request *rq;
847
848 rq = kmalloc_node(sizeof(struct request) + q->cmd_size, gfp_mask,
849 q->node);
850 if (rq && q->init_rq_fn && q->init_rq_fn(q, rq, gfp_mask) < 0) {
851 kfree(rq);
852 rq = NULL;
853 }
854 return rq;
855}
856
857static void free_request_size(void *element, void *data)
858{
859 struct request_queue *q = data;
860
861 if (q->exit_rq_fn)
862 q->exit_rq_fn(q, element);
863 kfree(element);
864}
865
866int blk_init_rl(struct request_list *rl, struct request_queue *q,
867 gfp_t gfp_mask)
868{
869 if (unlikely(rl->rq_pool) || q->mq_ops)
870 return 0;
871
872 rl->q = q;
873 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
874 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
875 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
876 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
877
878 if (q->cmd_size) {
879 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ,
880 alloc_request_size, free_request_size,
881 q, gfp_mask, q->node);
882 } else {
883 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ,
884 alloc_request_simple, free_request_simple,
885 q, gfp_mask, q->node);
886 }
887 if (!rl->rq_pool)
888 return -ENOMEM;
889
890 if (rl != &q->root_rl)
891 WARN_ON_ONCE(!blk_get_queue(q));
892
893 return 0;
894}
895
896void blk_exit_rl(struct request_queue *q, struct request_list *rl)
897{
898 if (rl->rq_pool) {
899 mempool_destroy(rl->rq_pool);
900 if (rl != &q->root_rl)
901 blk_put_queue(q);
902 }
903}
904
905struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
906{
907 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE, NULL);
908}
909EXPORT_SYMBOL(blk_alloc_queue);
910
911/**
912 * blk_queue_enter() - try to increase q->q_usage_counter
913 * @q: request queue pointer
914 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PREEMPT
915 */
916int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
917{
918 const bool preempt = flags & BLK_MQ_REQ_PREEMPT;
919
920 while (true) {
921 bool success = false;
922
923 rcu_read_lock();
924 if (percpu_ref_tryget_live(&q->q_usage_counter)) {
925 /*
926 * The code that sets the PREEMPT_ONLY flag is
927 * responsible for ensuring that that flag is globally
928 * visible before the queue is unfrozen.
929 */
930 if (preempt || !blk_queue_preempt_only(q)) {
931 success = true;
932 } else {
933 percpu_ref_put(&q->q_usage_counter);
934 }
935 }
936 rcu_read_unlock();
937
938 if (success)
939 return 0;
940
941 if (flags & BLK_MQ_REQ_NOWAIT)
942 return -EBUSY;
943
944 /*
945 * read pair of barrier in blk_freeze_queue_start(),
946 * we need to order reading __PERCPU_REF_DEAD flag of
947 * .q_usage_counter and reading .mq_freeze_depth or
948 * queue dying flag, otherwise the following wait may
949 * never return if the two reads are reordered.
950 */
951 smp_rmb();
952
953 wait_event(q->mq_freeze_wq,
954 (atomic_read(&q->mq_freeze_depth) == 0 &&
955 (preempt || !blk_queue_preempt_only(q))) ||
956 blk_queue_dying(q));
957 if (blk_queue_dying(q))
958 return -ENODEV;
959 }
960}
961
962void blk_queue_exit(struct request_queue *q)
963{
964 percpu_ref_put(&q->q_usage_counter);
965}
966
967static void blk_queue_usage_counter_release(struct percpu_ref *ref)
968{
969 struct request_queue *q =
970 container_of(ref, struct request_queue, q_usage_counter);
971
972 wake_up_all(&q->mq_freeze_wq);
973}
974
975static void blk_rq_timed_out_timer(struct timer_list *t)
976{
977 struct request_queue *q = from_timer(q, t, timeout);
978
979 kblockd_schedule_work(&q->timeout_work);
980}
981
982/**
983 * blk_alloc_queue_node - allocate a request queue
984 * @gfp_mask: memory allocation flags
985 * @node_id: NUMA node to allocate memory from
986 * @lock: For legacy queues, pointer to a spinlock that will be used to e.g.
987 * serialize calls to the legacy .request_fn() callback. Ignored for
988 * blk-mq request queues.
989 *
990 * Note: pass the queue lock as the third argument to this function instead of
991 * setting the queue lock pointer explicitly to avoid triggering a sporadic
992 * crash in the blkcg code. This function namely calls blkcg_init_queue() and
993 * the queue lock pointer must be set before blkcg_init_queue() is called.
994 */
995struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id,
996 spinlock_t *lock)
997{
998 struct request_queue *q;
999
1000 q = kmem_cache_alloc_node(blk_requestq_cachep,
1001 gfp_mask | __GFP_ZERO, node_id);
1002 if (!q)
1003 return NULL;
1004
1005 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
1006 if (q->id < 0)
1007 goto fail_q;
1008
1009 q->bio_split = bioset_create(BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
1010 if (!q->bio_split)
1011 goto fail_id;
1012
1013 q->backing_dev_info = bdi_alloc_node(gfp_mask, node_id);
1014 if (!q->backing_dev_info)
1015 goto fail_split;
1016
1017 q->stats = blk_alloc_queue_stats();
1018 if (!q->stats)
1019 goto fail_stats;
1020
1021 q->backing_dev_info->ra_pages =
1022 (VM_MAX_READAHEAD * 1024) / PAGE_SIZE;
1023 q->backing_dev_info->capabilities = BDI_CAP_CGROUP_WRITEBACK;
1024 q->backing_dev_info->name = "block";
1025 q->node = node_id;
1026
1027 timer_setup(&q->backing_dev_info->laptop_mode_wb_timer,
1028 laptop_mode_timer_fn, 0);
1029 timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
1030 INIT_WORK(&q->timeout_work, NULL);
1031 INIT_LIST_HEAD(&q->queue_head);
1032 INIT_LIST_HEAD(&q->timeout_list);
1033 INIT_LIST_HEAD(&q->icq_list);
1034#ifdef CONFIG_BLK_CGROUP
1035 INIT_LIST_HEAD(&q->blkg_list);
1036#endif
1037 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
1038
1039 kobject_init(&q->kobj, &blk_queue_ktype);
1040
1041#ifdef CONFIG_BLK_DEV_IO_TRACE
1042 mutex_init(&q->blk_trace_mutex);
1043#endif
1044 mutex_init(&q->sysfs_lock);
1045 spin_lock_init(&q->__queue_lock);
1046
1047 if (!q->mq_ops)
1048 q->queue_lock = lock ? : &q->__queue_lock;
1049
1050 /*
1051 * A queue starts its life with bypass turned on to avoid
1052 * unnecessary bypass on/off overhead and nasty surprises during
1053 * init. The initial bypass will be finished when the queue is
1054 * registered by blk_register_queue().
1055 */
1056 q->bypass_depth = 1;
1057 queue_flag_set_unlocked(QUEUE_FLAG_BYPASS, q);
1058
1059 init_waitqueue_head(&q->mq_freeze_wq);
1060
1061 /*
1062 * Init percpu_ref in atomic mode so that it's faster to shutdown.
1063 * See blk_register_queue() for details.
1064 */
1065 if (percpu_ref_init(&q->q_usage_counter,
1066 blk_queue_usage_counter_release,
1067 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
1068 goto fail_bdi;
1069
1070 if (blkcg_init_queue(q))
1071 goto fail_ref;
1072
1073 return q;
1074
1075fail_ref:
1076 percpu_ref_exit(&q->q_usage_counter);
1077fail_bdi:
1078 blk_free_queue_stats(q->stats);
1079fail_stats:
1080 bdi_put(q->backing_dev_info);
1081fail_split:
1082 bioset_free(q->bio_split);
1083fail_id:
1084 ida_simple_remove(&blk_queue_ida, q->id);
1085fail_q:
1086 kmem_cache_free(blk_requestq_cachep, q);
1087 return NULL;
1088}
1089EXPORT_SYMBOL(blk_alloc_queue_node);
1090
1091/**
1092 * blk_init_queue - prepare a request queue for use with a block device
1093 * @rfn: The function to be called to process requests that have been
1094 * placed on the queue.
1095 * @lock: Request queue spin lock
1096 *
1097 * Description:
1098 * If a block device wishes to use the standard request handling procedures,
1099 * which sorts requests and coalesces adjacent requests, then it must
1100 * call blk_init_queue(). The function @rfn will be called when there
1101 * are requests on the queue that need to be processed. If the device
1102 * supports plugging, then @rfn may not be called immediately when requests
1103 * are available on the queue, but may be called at some time later instead.
1104 * Plugged queues are generally unplugged when a buffer belonging to one
1105 * of the requests on the queue is needed, or due to memory pressure.
1106 *
1107 * @rfn is not required, or even expected, to remove all requests off the
1108 * queue, but only as many as it can handle at a time. If it does leave
1109 * requests on the queue, it is responsible for arranging that the requests
1110 * get dealt with eventually.
1111 *
1112 * The queue spin lock must be held while manipulating the requests on the
1113 * request queue; this lock will be taken also from interrupt context, so irq
1114 * disabling is needed for it.
1115 *
1116 * Function returns a pointer to the initialized request queue, or %NULL if
1117 * it didn't succeed.
1118 *
1119 * Note:
1120 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1121 * when the block device is deactivated (such as at module unload).
1122 **/
1123
1124struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
1125{
1126 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
1127}
1128EXPORT_SYMBOL(blk_init_queue);
1129
1130struct request_queue *
1131blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
1132{
1133 struct request_queue *q;
1134
1135 q = blk_alloc_queue_node(GFP_KERNEL, node_id, lock);
1136 if (!q)
1137 return NULL;
1138
1139 q->request_fn = rfn;
1140 if (blk_init_allocated_queue(q) < 0) {
1141 blk_cleanup_queue(q);
1142 return NULL;
1143 }
1144
1145 return q;
1146}
1147EXPORT_SYMBOL(blk_init_queue_node);
1148
1149static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio);
1150
1151
1152int blk_init_allocated_queue(struct request_queue *q)
1153{
1154 WARN_ON_ONCE(q->mq_ops);
1155
1156 q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, q->cmd_size);
1157 if (!q->fq)
1158 return -ENOMEM;
1159
1160 if (q->init_rq_fn && q->init_rq_fn(q, q->fq->flush_rq, GFP_KERNEL))
1161 goto out_free_flush_queue;
1162
1163 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
1164 goto out_exit_flush_rq;
1165
1166 INIT_WORK(&q->timeout_work, blk_timeout_work);
1167 q->queue_flags |= QUEUE_FLAG_DEFAULT;
1168
1169 /*
1170 * This also sets hw/phys segments, boundary and size
1171 */
1172 blk_queue_make_request(q, blk_queue_bio);
1173
1174 q->sg_reserved_size = INT_MAX;
1175
1176 /* Protect q->elevator from elevator_change */
1177 mutex_lock(&q->sysfs_lock);
1178
1179 /* init elevator */
1180 if (elevator_init(q, NULL)) {
1181 mutex_unlock(&q->sysfs_lock);
1182 goto out_exit_flush_rq;
1183 }
1184
1185 mutex_unlock(&q->sysfs_lock);
1186 return 0;
1187
1188out_exit_flush_rq:
1189 if (q->exit_rq_fn)
1190 q->exit_rq_fn(q, q->fq->flush_rq);
1191out_free_flush_queue:
1192 blk_free_flush_queue(q->fq);
1193 return -ENOMEM;
1194}
1195EXPORT_SYMBOL(blk_init_allocated_queue);
1196
1197bool blk_get_queue(struct request_queue *q)
1198{
1199 if (likely(!blk_queue_dying(q))) {
1200 __blk_get_queue(q);
1201 return true;
1202 }
1203
1204 return false;
1205}
1206EXPORT_SYMBOL(blk_get_queue);
1207
1208static inline void blk_free_request(struct request_list *rl, struct request *rq)
1209{
1210 if (rq->rq_flags & RQF_ELVPRIV) {
1211 elv_put_request(rl->q, rq);
1212 if (rq->elv.icq)
1213 put_io_context(rq->elv.icq->ioc);
1214 }
1215
1216 mempool_free(rq, rl->rq_pool);
1217}
1218
1219/*
1220 * ioc_batching returns true if the ioc is a valid batching request and
1221 * should be given priority access to a request.
1222 */
1223static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
1224{
1225 if (!ioc)
1226 return 0;
1227
1228 /*
1229 * Make sure the process is able to allocate at least 1 request
1230 * even if the batch times out, otherwise we could theoretically
1231 * lose wakeups.
1232 */
1233 return ioc->nr_batch_requests == q->nr_batching ||
1234 (ioc->nr_batch_requests > 0
1235 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
1236}
1237
1238/*
1239 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
1240 * will cause the process to be a "batcher" on all queues in the system. This
1241 * is the behaviour we want though - once it gets a wakeup it should be given
1242 * a nice run.
1243 */
1244static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
1245{
1246 if (!ioc || ioc_batching(q, ioc))
1247 return;
1248
1249 ioc->nr_batch_requests = q->nr_batching;
1250 ioc->last_waited = jiffies;
1251}
1252
1253static void __freed_request(struct request_list *rl, int sync)
1254{
1255 struct request_queue *q = rl->q;
1256
1257 if (rl->count[sync] < queue_congestion_off_threshold(q))
1258 blk_clear_congested(rl, sync);
1259
1260 if (rl->count[sync] + 1 <= q->nr_requests) {
1261 if (waitqueue_active(&rl->wait[sync]))
1262 wake_up(&rl->wait[sync]);
1263
1264 blk_clear_rl_full(rl, sync);
1265 }
1266}
1267
1268/*
1269 * A request has just been released. Account for it, update the full and
1270 * congestion status, wake up any waiters. Called under q->queue_lock.
1271 */
1272static void freed_request(struct request_list *rl, bool sync,
1273 req_flags_t rq_flags)
1274{
1275 struct request_queue *q = rl->q;
1276
1277 q->nr_rqs[sync]--;
1278 rl->count[sync]--;
1279 if (rq_flags & RQF_ELVPRIV)
1280 q->nr_rqs_elvpriv--;
1281
1282 __freed_request(rl, sync);
1283
1284 if (unlikely(rl->starved[sync ^ 1]))
1285 __freed_request(rl, sync ^ 1);
1286}
1287
1288int blk_update_nr_requests(struct request_queue *q, unsigned int nr)
1289{
1290 struct request_list *rl;
1291 int on_thresh, off_thresh;
1292
1293 WARN_ON_ONCE(q->mq_ops);
1294
1295 spin_lock_irq(q->queue_lock);
1296 q->nr_requests = nr;
1297 blk_queue_congestion_threshold(q);
1298 on_thresh = queue_congestion_on_threshold(q);
1299 off_thresh = queue_congestion_off_threshold(q);
1300
1301 blk_queue_for_each_rl(rl, q) {
1302 if (rl->count[BLK_RW_SYNC] >= on_thresh)
1303 blk_set_congested(rl, BLK_RW_SYNC);
1304 else if (rl->count[BLK_RW_SYNC] < off_thresh)
1305 blk_clear_congested(rl, BLK_RW_SYNC);
1306
1307 if (rl->count[BLK_RW_ASYNC] >= on_thresh)
1308 blk_set_congested(rl, BLK_RW_ASYNC);
1309 else if (rl->count[BLK_RW_ASYNC] < off_thresh)
1310 blk_clear_congested(rl, BLK_RW_ASYNC);
1311
1312 if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
1313 blk_set_rl_full(rl, BLK_RW_SYNC);
1314 } else {
1315 blk_clear_rl_full(rl, BLK_RW_SYNC);
1316 wake_up(&rl->wait[BLK_RW_SYNC]);
1317 }
1318
1319 if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
1320 blk_set_rl_full(rl, BLK_RW_ASYNC);
1321 } else {
1322 blk_clear_rl_full(rl, BLK_RW_ASYNC);
1323 wake_up(&rl->wait[BLK_RW_ASYNC]);
1324 }
1325 }
1326
1327 spin_unlock_irq(q->queue_lock);
1328 return 0;
1329}
1330
1331/**
1332 * __get_request - get a free request
1333 * @rl: request list to allocate from
1334 * @op: operation and flags
1335 * @bio: bio to allocate request for (can be %NULL)
1336 * @flags: BLQ_MQ_REQ_* flags
1337 *
1338 * Get a free request from @q. This function may fail under memory
1339 * pressure or if @q is dead.
1340 *
1341 * Must be called with @q->queue_lock held and,
1342 * Returns ERR_PTR on failure, with @q->queue_lock held.
1343 * Returns request pointer on success, with @q->queue_lock *not held*.
1344 */
1345static struct request *__get_request(struct request_list *rl, unsigned int op,
1346 struct bio *bio, blk_mq_req_flags_t flags)
1347{
1348 struct request_queue *q = rl->q;
1349 struct request *rq;
1350 struct elevator_type *et = q->elevator->type;
1351 struct io_context *ioc = rq_ioc(bio);
1352 struct io_cq *icq = NULL;
1353 const bool is_sync = op_is_sync(op);
1354 int may_queue;
1355 gfp_t gfp_mask = flags & BLK_MQ_REQ_NOWAIT ? GFP_ATOMIC :
1356 __GFP_DIRECT_RECLAIM;
1357 req_flags_t rq_flags = RQF_ALLOCED;
1358
1359 lockdep_assert_held(q->queue_lock);
1360
1361 if (unlikely(blk_queue_dying(q)))
1362 return ERR_PTR(-ENODEV);
1363
1364 may_queue = elv_may_queue(q, op);
1365 if (may_queue == ELV_MQUEUE_NO)
1366 goto rq_starved;
1367
1368 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
1369 if (rl->count[is_sync]+1 >= q->nr_requests) {
1370 /*
1371 * The queue will fill after this allocation, so set
1372 * it as full, and mark this process as "batching".
1373 * This process will be allowed to complete a batch of
1374 * requests, others will be blocked.
1375 */
1376 if (!blk_rl_full(rl, is_sync)) {
1377 ioc_set_batching(q, ioc);
1378 blk_set_rl_full(rl, is_sync);
1379 } else {
1380 if (may_queue != ELV_MQUEUE_MUST
1381 && !ioc_batching(q, ioc)) {
1382 /*
1383 * The queue is full and the allocating
1384 * process is not a "batcher", and not
1385 * exempted by the IO scheduler
1386 */
1387 return ERR_PTR(-ENOMEM);
1388 }
1389 }
1390 }
1391 blk_set_congested(rl, is_sync);
1392 }
1393
1394 /*
1395 * Only allow batching queuers to allocate up to 50% over the defined
1396 * limit of requests, otherwise we could have thousands of requests
1397 * allocated with any setting of ->nr_requests
1398 */
1399 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
1400 return ERR_PTR(-ENOMEM);
1401
1402 q->nr_rqs[is_sync]++;
1403 rl->count[is_sync]++;
1404 rl->starved[is_sync] = 0;
1405
1406 /*
1407 * Decide whether the new request will be managed by elevator. If
1408 * so, mark @rq_flags and increment elvpriv. Non-zero elvpriv will
1409 * prevent the current elevator from being destroyed until the new
1410 * request is freed. This guarantees icq's won't be destroyed and
1411 * makes creating new ones safe.
1412 *
1413 * Flush requests do not use the elevator so skip initialization.
1414 * This allows a request to share the flush and elevator data.
1415 *
1416 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1417 * it will be created after releasing queue_lock.
1418 */
1419 if (!op_is_flush(op) && !blk_queue_bypass(q)) {
1420 rq_flags |= RQF_ELVPRIV;
1421 q->nr_rqs_elvpriv++;
1422 if (et->icq_cache && ioc)
1423 icq = ioc_lookup_icq(ioc, q);
1424 }
1425
1426 if (blk_queue_io_stat(q))
1427 rq_flags |= RQF_IO_STAT;
1428 spin_unlock_irq(q->queue_lock);
1429
1430 /* allocate and init request */
1431 rq = mempool_alloc(rl->rq_pool, gfp_mask);
1432 if (!rq)
1433 goto fail_alloc;
1434
1435 blk_rq_init(q, rq);
1436 blk_rq_set_rl(rq, rl);
1437 rq->cmd_flags = op;
1438 rq->rq_flags = rq_flags;
1439 if (flags & BLK_MQ_REQ_PREEMPT)
1440 rq->rq_flags |= RQF_PREEMPT;
1441
1442 /* init elvpriv */
1443 if (rq_flags & RQF_ELVPRIV) {
1444 if (unlikely(et->icq_cache && !icq)) {
1445 if (ioc)
1446 icq = ioc_create_icq(ioc, q, gfp_mask);
1447 if (!icq)
1448 goto fail_elvpriv;
1449 }
1450
1451 rq->elv.icq = icq;
1452 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1453 goto fail_elvpriv;
1454
1455 /* @rq->elv.icq holds io_context until @rq is freed */
1456 if (icq)
1457 get_io_context(icq->ioc);
1458 }
1459out:
1460 /*
1461 * ioc may be NULL here, and ioc_batching will be false. That's
1462 * OK, if the queue is under the request limit then requests need
1463 * not count toward the nr_batch_requests limit. There will always
1464 * be some limit enforced by BLK_BATCH_TIME.
1465 */
1466 if (ioc_batching(q, ioc))
1467 ioc->nr_batch_requests--;
1468
1469 trace_block_getrq(q, bio, op);
1470 return rq;
1471
1472fail_elvpriv:
1473 /*
1474 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1475 * and may fail indefinitely under memory pressure and thus
1476 * shouldn't stall IO. Treat this request as !elvpriv. This will
1477 * disturb iosched and blkcg but weird is bettern than dead.
1478 */
1479 printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1480 __func__, dev_name(q->backing_dev_info->dev));
1481
1482 rq->rq_flags &= ~RQF_ELVPRIV;
1483 rq->elv.icq = NULL;
1484
1485 spin_lock_irq(q->queue_lock);
1486 q->nr_rqs_elvpriv--;
1487 spin_unlock_irq(q->queue_lock);
1488 goto out;
1489
1490fail_alloc:
1491 /*
1492 * Allocation failed presumably due to memory. Undo anything we
1493 * might have messed up.
1494 *
1495 * Allocating task should really be put onto the front of the wait
1496 * queue, but this is pretty rare.
1497 */
1498 spin_lock_irq(q->queue_lock);
1499 freed_request(rl, is_sync, rq_flags);
1500
1501 /*
1502 * in the very unlikely event that allocation failed and no
1503 * requests for this direction was pending, mark us starved so that
1504 * freeing of a request in the other direction will notice
1505 * us. another possible fix would be to split the rq mempool into
1506 * READ and WRITE
1507 */
1508rq_starved:
1509 if (unlikely(rl->count[is_sync] == 0))
1510 rl->starved[is_sync] = 1;
1511 return ERR_PTR(-ENOMEM);
1512}
1513
1514/**
1515 * get_request - get a free request
1516 * @q: request_queue to allocate request from
1517 * @op: operation and flags
1518 * @bio: bio to allocate request for (can be %NULL)
1519 * @flags: BLK_MQ_REQ_* flags.
1520 *
1521 * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask,
1522 * this function keeps retrying under memory pressure and fails iff @q is dead.
1523 *
1524 * Must be called with @q->queue_lock held and,
1525 * Returns ERR_PTR on failure, with @q->queue_lock held.
1526 * Returns request pointer on success, with @q->queue_lock *not held*.
1527 */
1528static struct request *get_request(struct request_queue *q, unsigned int op,
1529 struct bio *bio, blk_mq_req_flags_t flags)
1530{
1531 const bool is_sync = op_is_sync(op);
1532 DEFINE_WAIT(wait);
1533 struct request_list *rl;
1534 struct request *rq;
1535
1536 lockdep_assert_held(q->queue_lock);
1537 WARN_ON_ONCE(q->mq_ops);
1538
1539 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
1540retry:
1541 rq = __get_request(rl, op, bio, flags);
1542 if (!IS_ERR(rq))
1543 return rq;
1544
1545 if (op & REQ_NOWAIT) {
1546 blk_put_rl(rl);
1547 return ERR_PTR(-EAGAIN);
1548 }
1549
1550 if ((flags & BLK_MQ_REQ_NOWAIT) || unlikely(blk_queue_dying(q))) {
1551 blk_put_rl(rl);
1552 return rq;
1553 }
1554
1555 /* wait on @rl and retry */
1556 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1557 TASK_UNINTERRUPTIBLE);
1558
1559 trace_block_sleeprq(q, bio, op);
1560
1561 spin_unlock_irq(q->queue_lock);
1562 io_schedule();
1563
1564 /*
1565 * After sleeping, we become a "batching" process and will be able
1566 * to allocate at least one request, and up to a big batch of them
1567 * for a small period time. See ioc_batching, ioc_set_batching
1568 */
1569 ioc_set_batching(q, current->io_context);
1570
1571 spin_lock_irq(q->queue_lock);
1572 finish_wait(&rl->wait[is_sync], &wait);
1573
1574 goto retry;
1575}
1576
1577/* flags: BLK_MQ_REQ_PREEMPT and/or BLK_MQ_REQ_NOWAIT. */
1578static struct request *blk_old_get_request(struct request_queue *q,
1579 unsigned int op, blk_mq_req_flags_t flags)
1580{
1581 struct request *rq;
1582 gfp_t gfp_mask = flags & BLK_MQ_REQ_NOWAIT ? GFP_ATOMIC :
1583 __GFP_DIRECT_RECLAIM;
1584 int ret = 0;
1585
1586 WARN_ON_ONCE(q->mq_ops);
1587
1588 /* create ioc upfront */
1589 create_io_context(gfp_mask, q->node);
1590
1591 ret = blk_queue_enter(q, flags);
1592 if (ret)
1593 return ERR_PTR(ret);
1594 spin_lock_irq(q->queue_lock);
1595 rq = get_request(q, op, NULL, flags);
1596 if (IS_ERR(rq)) {
1597 spin_unlock_irq(q->queue_lock);
1598 blk_queue_exit(q);
1599 return rq;
1600 }
1601
1602 /* q->queue_lock is unlocked at this point */
1603 rq->__data_len = 0;
1604 rq->__sector = (sector_t) -1;
1605 rq->bio = rq->biotail = NULL;
1606 return rq;
1607}
1608
1609/**
1610 * blk_get_request_flags - allocate a request
1611 * @q: request queue to allocate a request for
1612 * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC.
1613 * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
1614 */
1615struct request *blk_get_request_flags(struct request_queue *q, unsigned int op,
1616 blk_mq_req_flags_t flags)
1617{
1618 struct request *req;
1619
1620 WARN_ON_ONCE(op & REQ_NOWAIT);
1621 WARN_ON_ONCE(flags & ~(BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_PREEMPT));
1622
1623 if (q->mq_ops) {
1624 req = blk_mq_alloc_request(q, op, flags);
1625 if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn)
1626 q->mq_ops->initialize_rq_fn(req);
1627 } else {
1628 req = blk_old_get_request(q, op, flags);
1629 if (!IS_ERR(req) && q->initialize_rq_fn)
1630 q->initialize_rq_fn(req);
1631 }
1632
1633 return req;
1634}
1635EXPORT_SYMBOL(blk_get_request_flags);
1636
1637struct request *blk_get_request(struct request_queue *q, unsigned int op,
1638 gfp_t gfp_mask)
1639{
1640 return blk_get_request_flags(q, op, gfp_mask & __GFP_DIRECT_RECLAIM ?
1641 0 : BLK_MQ_REQ_NOWAIT);
1642}
1643EXPORT_SYMBOL(blk_get_request);
1644
1645/**
1646 * blk_requeue_request - put a request back on queue
1647 * @q: request queue where request should be inserted
1648 * @rq: request to be inserted
1649 *
1650 * Description:
1651 * Drivers often keep queueing requests until the hardware cannot accept
1652 * more, when that condition happens we need to put the request back
1653 * on the queue. Must be called with queue lock held.
1654 */
1655void blk_requeue_request(struct request_queue *q, struct request *rq)
1656{
1657 lockdep_assert_held(q->queue_lock);
1658 WARN_ON_ONCE(q->mq_ops);
1659
1660 blk_delete_timer(rq);
1661 blk_clear_rq_complete(rq);
1662 trace_block_rq_requeue(q, rq);
1663 wbt_requeue(q->rq_wb, &rq->issue_stat);
1664
1665 if (rq->rq_flags & RQF_QUEUED)
1666 blk_queue_end_tag(q, rq);
1667
1668 BUG_ON(blk_queued_rq(rq));
1669
1670 elv_requeue_request(q, rq);
1671}
1672EXPORT_SYMBOL(blk_requeue_request);
1673
1674static void add_acct_request(struct request_queue *q, struct request *rq,
1675 int where)
1676{
1677 blk_account_io_start(rq, true);
1678 __elv_add_request(q, rq, where);
1679}
1680
1681static void part_round_stats_single(struct request_queue *q, int cpu,
1682 struct hd_struct *part, unsigned long now,
1683 unsigned int inflight)
1684{
1685 if (inflight) {
1686 __part_stat_add(cpu, part, time_in_queue,
1687 inflight * (now - part->stamp));
1688 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1689 }
1690 part->stamp = now;
1691}
1692
1693/**
1694 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1695 * @q: target block queue
1696 * @cpu: cpu number for stats access
1697 * @part: target partition
1698 *
1699 * The average IO queue length and utilisation statistics are maintained
1700 * by observing the current state of the queue length and the amount of
1701 * time it has been in this state for.
1702 *
1703 * Normally, that accounting is done on IO completion, but that can result
1704 * in more than a second's worth of IO being accounted for within any one
1705 * second, leading to >100% utilisation. To deal with that, we call this
1706 * function to do a round-off before returning the results when reading
1707 * /proc/diskstats. This accounts immediately for all queue usage up to
1708 * the current jiffies and restarts the counters again.
1709 */
1710void part_round_stats(struct request_queue *q, int cpu, struct hd_struct *part)
1711{
1712 struct hd_struct *part2 = NULL;
1713 unsigned long now = jiffies;
1714 unsigned int inflight[2];
1715 int stats = 0;
1716
1717 if (part->stamp != now)
1718 stats |= 1;
1719
1720 if (part->partno) {
1721 part2 = &part_to_disk(part)->part0;
1722 if (part2->stamp != now)
1723 stats |= 2;
1724 }
1725
1726 if (!stats)
1727 return;
1728
1729 part_in_flight(q, part, inflight);
1730
1731 if (stats & 2)
1732 part_round_stats_single(q, cpu, part2, now, inflight[1]);
1733 if (stats & 1)
1734 part_round_stats_single(q, cpu, part, now, inflight[0]);
1735}
1736EXPORT_SYMBOL_GPL(part_round_stats);
1737
1738#ifdef CONFIG_PM
1739static void blk_pm_put_request(struct request *rq)
1740{
1741 if (rq->q->dev && !(rq->rq_flags & RQF_PM) && !--rq->q->nr_pending)
1742 pm_runtime_mark_last_busy(rq->q->dev);
1743}
1744#else
1745static inline void blk_pm_put_request(struct request *rq) {}
1746#endif
1747
1748void __blk_put_request(struct request_queue *q, struct request *req)
1749{
1750 req_flags_t rq_flags = req->rq_flags;
1751
1752 if (unlikely(!q))
1753 return;
1754
1755 if (q->mq_ops) {
1756 blk_mq_free_request(req);
1757 return;
1758 }
1759
1760 lockdep_assert_held(q->queue_lock);
1761
1762 blk_req_zone_write_unlock(req);
1763 blk_pm_put_request(req);
1764
1765 elv_completed_request(q, req);
1766
1767 /* this is a bio leak */
1768 WARN_ON(req->bio != NULL);
1769
1770 wbt_done(q->rq_wb, &req->issue_stat);
1771
1772 /*
1773 * Request may not have originated from ll_rw_blk. if not,
1774 * it didn't come out of our reserved rq pools
1775 */
1776 if (rq_flags & RQF_ALLOCED) {
1777 struct request_list *rl = blk_rq_rl(req);
1778 bool sync = op_is_sync(req->cmd_flags);
1779
1780 BUG_ON(!list_empty(&req->queuelist));
1781 BUG_ON(ELV_ON_HASH(req));
1782
1783 blk_free_request(rl, req);
1784 freed_request(rl, sync, rq_flags);
1785 blk_put_rl(rl);
1786 blk_queue_exit(q);
1787 }
1788}
1789EXPORT_SYMBOL_GPL(__blk_put_request);
1790
1791void blk_put_request(struct request *req)
1792{
1793 struct request_queue *q = req->q;
1794
1795 if (q->mq_ops)
1796 blk_mq_free_request(req);
1797 else {
1798 unsigned long flags;
1799
1800 spin_lock_irqsave(q->queue_lock, flags);
1801 __blk_put_request(q, req);
1802 spin_unlock_irqrestore(q->queue_lock, flags);
1803 }
1804}
1805EXPORT_SYMBOL(blk_put_request);
1806
1807bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1808 struct bio *bio)
1809{
1810 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
1811
1812 if (!ll_back_merge_fn(q, req, bio))
1813 return false;
1814
1815 trace_block_bio_backmerge(q, req, bio);
1816
1817 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1818 blk_rq_set_mixed_merge(req);
1819
1820 req->biotail->bi_next = bio;
1821 req->biotail = bio;
1822 req->__data_len += bio->bi_iter.bi_size;
1823 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1824
1825 blk_account_io_start(req, false);
1826 return true;
1827}
1828
1829bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1830 struct bio *bio)
1831{
1832 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
1833
1834 if (!ll_front_merge_fn(q, req, bio))
1835 return false;
1836
1837 trace_block_bio_frontmerge(q, req, bio);
1838
1839 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1840 blk_rq_set_mixed_merge(req);
1841
1842 bio->bi_next = req->bio;
1843 req->bio = bio;
1844
1845 req->__sector = bio->bi_iter.bi_sector;
1846 req->__data_len += bio->bi_iter.bi_size;
1847 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1848
1849 blk_account_io_start(req, false);
1850 return true;
1851}
1852
1853bool bio_attempt_discard_merge(struct request_queue *q, struct request *req,
1854 struct bio *bio)
1855{
1856 unsigned short segments = blk_rq_nr_discard_segments(req);
1857
1858 if (segments >= queue_max_discard_segments(q))
1859 goto no_merge;
1860 if (blk_rq_sectors(req) + bio_sectors(bio) >
1861 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
1862 goto no_merge;
1863
1864 req->biotail->bi_next = bio;
1865 req->biotail = bio;
1866 req->__data_len += bio->bi_iter.bi_size;
1867 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1868 req->nr_phys_segments = segments + 1;
1869
1870 blk_account_io_start(req, false);
1871 return true;
1872no_merge:
1873 req_set_nomerge(q, req);
1874 return false;
1875}
1876
1877/**
1878 * blk_attempt_plug_merge - try to merge with %current's plugged list
1879 * @q: request_queue new bio is being queued at
1880 * @bio: new bio being queued
1881 * @request_count: out parameter for number of traversed plugged requests
1882 * @same_queue_rq: pointer to &struct request that gets filled in when
1883 * another request associated with @q is found on the plug list
1884 * (optional, may be %NULL)
1885 *
1886 * Determine whether @bio being queued on @q can be merged with a request
1887 * on %current's plugged list. Returns %true if merge was successful,
1888 * otherwise %false.
1889 *
1890 * Plugging coalesces IOs from the same issuer for the same purpose without
1891 * going through @q->queue_lock. As such it's more of an issuing mechanism
1892 * than scheduling, and the request, while may have elvpriv data, is not
1893 * added on the elevator at this point. In addition, we don't have
1894 * reliable access to the elevator outside queue lock. Only check basic
1895 * merging parameters without querying the elevator.
1896 *
1897 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1898 */
1899bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1900 unsigned int *request_count,
1901 struct request **same_queue_rq)
1902{
1903 struct blk_plug *plug;
1904 struct request *rq;
1905 struct list_head *plug_list;
1906
1907 plug = current->plug;
1908 if (!plug)
1909 return false;
1910 *request_count = 0;
1911
1912 if (q->mq_ops)
1913 plug_list = &plug->mq_list;
1914 else
1915 plug_list = &plug->list;
1916
1917 list_for_each_entry_reverse(rq, plug_list, queuelist) {
1918 bool merged = false;
1919
1920 if (rq->q == q) {
1921 (*request_count)++;
1922 /*
1923 * Only blk-mq multiple hardware queues case checks the
1924 * rq in the same queue, there should be only one such
1925 * rq in a queue
1926 **/
1927 if (same_queue_rq)
1928 *same_queue_rq = rq;
1929 }
1930
1931 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
1932 continue;
1933
1934 switch (blk_try_merge(rq, bio)) {
1935 case ELEVATOR_BACK_MERGE:
1936 merged = bio_attempt_back_merge(q, rq, bio);
1937 break;
1938 case ELEVATOR_FRONT_MERGE:
1939 merged = bio_attempt_front_merge(q, rq, bio);
1940 break;
1941 case ELEVATOR_DISCARD_MERGE:
1942 merged = bio_attempt_discard_merge(q, rq, bio);
1943 break;
1944 default:
1945 break;
1946 }
1947
1948 if (merged)
1949 return true;
1950 }
1951
1952 return false;
1953}
1954
1955unsigned int blk_plug_queued_count(struct request_queue *q)
1956{
1957 struct blk_plug *plug;
1958 struct request *rq;
1959 struct list_head *plug_list;
1960 unsigned int ret = 0;
1961
1962 plug = current->plug;
1963 if (!plug)
1964 goto out;
1965
1966 if (q->mq_ops)
1967 plug_list = &plug->mq_list;
1968 else
1969 plug_list = &plug->list;
1970
1971 list_for_each_entry(rq, plug_list, queuelist) {
1972 if (rq->q == q)
1973 ret++;
1974 }
1975out:
1976 return ret;
1977}
1978
1979void blk_init_request_from_bio(struct request *req, struct bio *bio)
1980{
1981 struct io_context *ioc = rq_ioc(bio);
1982
1983 if (bio->bi_opf & REQ_RAHEAD)
1984 req->cmd_flags |= REQ_FAILFAST_MASK;
1985
1986 req->__sector = bio->bi_iter.bi_sector;
1987 if (ioprio_valid(bio_prio(bio)))
1988 req->ioprio = bio_prio(bio);
1989 else if (ioc)
1990 req->ioprio = ioc->ioprio;
1991 else
1992 req->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0);
1993 req->write_hint = bio->bi_write_hint;
1994 blk_rq_bio_prep(req->q, req, bio);
1995}
1996EXPORT_SYMBOL_GPL(blk_init_request_from_bio);
1997
1998static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio)
1999{
2000 struct blk_plug *plug;
2001 int where = ELEVATOR_INSERT_SORT;
2002 struct request *req, *free;
2003 unsigned int request_count = 0;
2004 unsigned int wb_acct;
2005
2006 /*
2007 * low level driver can indicate that it wants pages above a
2008 * certain limit bounced to low memory (ie for highmem, or even
2009 * ISA dma in theory)
2010 */
2011 blk_queue_bounce(q, &bio);
2012
2013 blk_queue_split(q, &bio);
2014
2015 if (!bio_integrity_prep(bio))
2016 return BLK_QC_T_NONE;
2017
2018 if (op_is_flush(bio->bi_opf)) {
2019 spin_lock_irq(q->queue_lock);
2020 where = ELEVATOR_INSERT_FLUSH;
2021 goto get_rq;
2022 }
2023
2024 /*
2025 * Check if we can merge with the plugged list before grabbing
2026 * any locks.
2027 */
2028 if (!blk_queue_nomerges(q)) {
2029 if (blk_attempt_plug_merge(q, bio, &request_count, NULL))
2030 return BLK_QC_T_NONE;
2031 } else
2032 request_count = blk_plug_queued_count(q);
2033
2034 spin_lock_irq(q->queue_lock);
2035
2036 switch (elv_merge(q, &req, bio)) {
2037 case ELEVATOR_BACK_MERGE:
2038 if (!bio_attempt_back_merge(q, req, bio))
2039 break;
2040 elv_bio_merged(q, req, bio);
2041 free = attempt_back_merge(q, req);
2042 if (free)
2043 __blk_put_request(q, free);
2044 else
2045 elv_merged_request(q, req, ELEVATOR_BACK_MERGE);
2046 goto out_unlock;
2047 case ELEVATOR_FRONT_MERGE:
2048 if (!bio_attempt_front_merge(q, req, bio))
2049 break;
2050 elv_bio_merged(q, req, bio);
2051 free = attempt_front_merge(q, req);
2052 if (free)
2053 __blk_put_request(q, free);
2054 else
2055 elv_merged_request(q, req, ELEVATOR_FRONT_MERGE);
2056 goto out_unlock;
2057 default:
2058 break;
2059 }
2060
2061get_rq:
2062 wb_acct = wbt_wait(q->rq_wb, bio, q->queue_lock);
2063
2064 /*
2065 * Grab a free request. This is might sleep but can not fail.
2066 * Returns with the queue unlocked.
2067 */
2068 blk_queue_enter_live(q);
2069 req = get_request(q, bio->bi_opf, bio, 0);
2070 if (IS_ERR(req)) {
2071 blk_queue_exit(q);
2072 __wbt_done(q->rq_wb, wb_acct);
2073 if (PTR_ERR(req) == -ENOMEM)
2074 bio->bi_status = BLK_STS_RESOURCE;
2075 else
2076 bio->bi_status = BLK_STS_IOERR;
2077 bio_endio(bio);
2078 goto out_unlock;
2079 }
2080
2081 wbt_track(&req->issue_stat, wb_acct);
2082
2083 /*
2084 * After dropping the lock and possibly sleeping here, our request
2085 * may now be mergeable after it had proven unmergeable (above).
2086 * We don't worry about that case for efficiency. It won't happen
2087 * often, and the elevators are able to handle it.
2088 */
2089 blk_init_request_from_bio(req, bio);
2090
2091 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
2092 req->cpu = raw_smp_processor_id();
2093
2094 plug = current->plug;
2095 if (plug) {
2096 /*
2097 * If this is the first request added after a plug, fire
2098 * of a plug trace.
2099 *
2100 * @request_count may become stale because of schedule
2101 * out, so check plug list again.
2102 */
2103 if (!request_count || list_empty(&plug->list))
2104 trace_block_plug(q);
2105 else {
2106 struct request *last = list_entry_rq(plug->list.prev);
2107 if (request_count >= BLK_MAX_REQUEST_COUNT ||
2108 blk_rq_bytes(last) >= BLK_PLUG_FLUSH_SIZE) {
2109 blk_flush_plug_list(plug, false);
2110 trace_block_plug(q);
2111 }
2112 }
2113 list_add_tail(&req->queuelist, &plug->list);
2114 blk_account_io_start(req, true);
2115 } else {
2116 spin_lock_irq(q->queue_lock);
2117 add_acct_request(q, req, where);
2118 __blk_run_queue(q);
2119out_unlock:
2120 spin_unlock_irq(q->queue_lock);
2121 }
2122
2123 return BLK_QC_T_NONE;
2124}
2125
2126static void handle_bad_sector(struct bio *bio, sector_t maxsector)
2127{
2128 char b[BDEVNAME_SIZE];
2129
2130 printk(KERN_INFO "attempt to access beyond end of device\n");
2131 printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n",
2132 bio_devname(bio, b), bio->bi_opf,
2133 (unsigned long long)bio_end_sector(bio),
2134 (long long)maxsector);
2135}
2136
2137#ifdef CONFIG_FAIL_MAKE_REQUEST
2138
2139static DECLARE_FAULT_ATTR(fail_make_request);
2140
2141static int __init setup_fail_make_request(char *str)
2142{
2143 return setup_fault_attr(&fail_make_request, str);
2144}
2145__setup("fail_make_request=", setup_fail_make_request);
2146
2147static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
2148{
2149 return part->make_it_fail && should_fail(&fail_make_request, bytes);
2150}
2151
2152static int __init fail_make_request_debugfs(void)
2153{
2154 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
2155 NULL, &fail_make_request);
2156
2157 return PTR_ERR_OR_ZERO(dir);
2158}
2159
2160late_initcall(fail_make_request_debugfs);
2161
2162#else /* CONFIG_FAIL_MAKE_REQUEST */
2163
2164static inline bool should_fail_request(struct hd_struct *part,
2165 unsigned int bytes)
2166{
2167 return false;
2168}
2169
2170#endif /* CONFIG_FAIL_MAKE_REQUEST */
2171
2172static inline bool bio_check_ro(struct bio *bio, struct hd_struct *part)
2173{
2174 if (part->policy && op_is_write(bio_op(bio))) {
2175 char b[BDEVNAME_SIZE];
2176
2177 printk(KERN_ERR
2178 "generic_make_request: Trying to write "
2179 "to read-only block-device %s (partno %d)\n",
2180 bio_devname(bio, b), part->partno);
2181 return true;
2182 }
2183
2184 return false;
2185}
2186
2187static noinline int should_fail_bio(struct bio *bio)
2188{
2189 if (should_fail_request(&bio->bi_disk->part0, bio->bi_iter.bi_size))
2190 return -EIO;
2191 return 0;
2192}
2193ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
2194
2195/*
2196 * Check whether this bio extends beyond the end of the device or partition.
2197 * This may well happen - the kernel calls bread() without checking the size of
2198 * the device, e.g., when mounting a file system.
2199 */
2200static inline int bio_check_eod(struct bio *bio, sector_t maxsector)
2201{
2202 unsigned int nr_sectors = bio_sectors(bio);
2203
2204 if (nr_sectors && maxsector &&
2205 (nr_sectors > maxsector ||
2206 bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
2207 handle_bad_sector(bio, maxsector);
2208 return -EIO;
2209 }
2210 return 0;
2211}
2212
2213/*
2214 * Remap block n of partition p to block n+start(p) of the disk.
2215 */
2216static inline int blk_partition_remap(struct bio *bio)
2217{
2218 struct hd_struct *p;
2219 int ret = -EIO;
2220
2221 rcu_read_lock();
2222 p = __disk_get_part(bio->bi_disk, bio->bi_partno);
2223 if (unlikely(!p))
2224 goto out;
2225 if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
2226 goto out;
2227 if (unlikely(bio_check_ro(bio, p)))
2228 goto out;
2229
2230 /*
2231 * Zone reset does not include bi_size so bio_sectors() is always 0.
2232 * Include a test for the reset op code and perform the remap if needed.
2233 */
2234 if (bio_sectors(bio) || bio_op(bio) == REQ_OP_ZONE_RESET) {
2235 if (bio_check_eod(bio, part_nr_sects_read(p)))
2236 goto out;
2237 bio->bi_iter.bi_sector += p->start_sect;
2238 bio->bi_partno = 0;
2239 trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p),
2240 bio->bi_iter.bi_sector - p->start_sect);
2241 }
2242 ret = 0;
2243out:
2244 rcu_read_unlock();
2245 return ret;
2246}
2247
2248static noinline_for_stack bool
2249generic_make_request_checks(struct bio *bio)
2250{
2251 struct request_queue *q;
2252 int nr_sectors = bio_sectors(bio);
2253 blk_status_t status = BLK_STS_IOERR;
2254 char b[BDEVNAME_SIZE];
2255
2256 might_sleep();
2257
2258 q = bio->bi_disk->queue;
2259 if (unlikely(!q)) {
2260 printk(KERN_ERR
2261 "generic_make_request: Trying to access "
2262 "nonexistent block-device %s (%Lu)\n",
2263 bio_devname(bio, b), (long long)bio->bi_iter.bi_sector);
2264 goto end_io;
2265 }
2266
2267 /*
2268 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
2269 * if queue is not a request based queue.
2270 */
2271 if ((bio->bi_opf & REQ_NOWAIT) && !queue_is_rq_based(q))
2272 goto not_supported;
2273
2274 if (should_fail_bio(bio))
2275 goto end_io;
2276
2277 if (bio->bi_partno) {
2278 if (unlikely(blk_partition_remap(bio)))
2279 goto end_io;
2280 } else {
2281 if (unlikely(bio_check_ro(bio, &bio->bi_disk->part0)))
2282 goto end_io;
2283 if (unlikely(bio_check_eod(bio, get_capacity(bio->bi_disk))))
2284 goto end_io;
2285 }
2286
2287 /*
2288 * Filter flush bio's early so that make_request based
2289 * drivers without flush support don't have to worry
2290 * about them.
2291 */
2292 if (op_is_flush(bio->bi_opf) &&
2293 !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
2294 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
2295 if (!nr_sectors) {
2296 status = BLK_STS_OK;
2297 goto end_io;
2298 }
2299 }
2300
2301 switch (bio_op(bio)) {
2302 case REQ_OP_DISCARD:
2303 if (!blk_queue_discard(q))
2304 goto not_supported;
2305 break;
2306 case REQ_OP_SECURE_ERASE:
2307 if (!blk_queue_secure_erase(q))
2308 goto not_supported;
2309 break;
2310 case REQ_OP_WRITE_SAME:
2311 if (!q->limits.max_write_same_sectors)
2312 goto not_supported;
2313 break;
2314 case REQ_OP_ZONE_REPORT:
2315 case REQ_OP_ZONE_RESET:
2316 if (!blk_queue_is_zoned(q))
2317 goto not_supported;
2318 break;
2319 case REQ_OP_WRITE_ZEROES:
2320 if (!q->limits.max_write_zeroes_sectors)
2321 goto not_supported;
2322 break;
2323 default:
2324 break;
2325 }
2326
2327 /*
2328 * Various block parts want %current->io_context and lazy ioc
2329 * allocation ends up trading a lot of pain for a small amount of
2330 * memory. Just allocate it upfront. This may fail and block
2331 * layer knows how to live with it.
2332 */
2333 create_io_context(GFP_ATOMIC, q->node);
2334
2335 if (!blkcg_bio_issue_check(q, bio))
2336 return false;
2337
2338 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
2339 trace_block_bio_queue(q, bio);
2340 /* Now that enqueuing has been traced, we need to trace
2341 * completion as well.
2342 */
2343 bio_set_flag(bio, BIO_TRACE_COMPLETION);
2344 }
2345 return true;
2346
2347not_supported:
2348 status = BLK_STS_NOTSUPP;
2349end_io:
2350 bio->bi_status = status;
2351 bio_endio(bio);
2352 return false;
2353}
2354
2355/**
2356 * generic_make_request - hand a buffer to its device driver for I/O
2357 * @bio: The bio describing the location in memory and on the device.
2358 *
2359 * generic_make_request() is used to make I/O requests of block
2360 * devices. It is passed a &struct bio, which describes the I/O that needs
2361 * to be done.
2362 *
2363 * generic_make_request() does not return any status. The
2364 * success/failure status of the request, along with notification of
2365 * completion, is delivered asynchronously through the bio->bi_end_io
2366 * function described (one day) else where.
2367 *
2368 * The caller of generic_make_request must make sure that bi_io_vec
2369 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2370 * set to describe the device address, and the
2371 * bi_end_io and optionally bi_private are set to describe how
2372 * completion notification should be signaled.
2373 *
2374 * generic_make_request and the drivers it calls may use bi_next if this
2375 * bio happens to be merged with someone else, and may resubmit the bio to
2376 * a lower device by calling into generic_make_request recursively, which
2377 * means the bio should NOT be touched after the call to ->make_request_fn.
2378 */
2379blk_qc_t generic_make_request(struct bio *bio)
2380{
2381 /*
2382 * bio_list_on_stack[0] contains bios submitted by the current
2383 * make_request_fn.
2384 * bio_list_on_stack[1] contains bios that were submitted before
2385 * the current make_request_fn, but that haven't been processed
2386 * yet.
2387 */
2388 struct bio_list bio_list_on_stack[2];
2389 blk_mq_req_flags_t flags = 0;
2390 struct request_queue *q = bio->bi_disk->queue;
2391 blk_qc_t ret = BLK_QC_T_NONE;
2392
2393 if (bio->bi_opf & REQ_NOWAIT)
2394 flags = BLK_MQ_REQ_NOWAIT;
2395 if (blk_queue_enter(q, flags) < 0) {
2396 if (!blk_queue_dying(q) && (bio->bi_opf & REQ_NOWAIT))
2397 bio_wouldblock_error(bio);
2398 else
2399 bio_io_error(bio);
2400 return ret;
2401 }
2402
2403 if (!generic_make_request_checks(bio))
2404 goto out;
2405
2406 /*
2407 * We only want one ->make_request_fn to be active at a time, else
2408 * stack usage with stacked devices could be a problem. So use
2409 * current->bio_list to keep a list of requests submited by a
2410 * make_request_fn function. current->bio_list is also used as a
2411 * flag to say if generic_make_request is currently active in this
2412 * task or not. If it is NULL, then no make_request is active. If
2413 * it is non-NULL, then a make_request is active, and new requests
2414 * should be added at the tail
2415 */
2416 if (current->bio_list) {
2417 bio_list_add(¤t->bio_list[0], bio);
2418 goto out;
2419 }
2420
2421 /* following loop may be a bit non-obvious, and so deserves some
2422 * explanation.
2423 * Before entering the loop, bio->bi_next is NULL (as all callers
2424 * ensure that) so we have a list with a single bio.
2425 * We pretend that we have just taken it off a longer list, so
2426 * we assign bio_list to a pointer to the bio_list_on_stack,
2427 * thus initialising the bio_list of new bios to be
2428 * added. ->make_request() may indeed add some more bios
2429 * through a recursive call to generic_make_request. If it
2430 * did, we find a non-NULL value in bio_list and re-enter the loop
2431 * from the top. In this case we really did just take the bio
2432 * of the top of the list (no pretending) and so remove it from
2433 * bio_list, and call into ->make_request() again.
2434 */
2435 BUG_ON(bio->bi_next);
2436 bio_list_init(&bio_list_on_stack[0]);
2437 current->bio_list = bio_list_on_stack;
2438 do {
2439 bool enter_succeeded = true;
2440
2441 if (unlikely(q != bio->bi_disk->queue)) {
2442 if (q)
2443 blk_queue_exit(q);
2444 q = bio->bi_disk->queue;
2445 flags = 0;
2446 if (bio->bi_opf & REQ_NOWAIT)
2447 flags = BLK_MQ_REQ_NOWAIT;
2448 if (blk_queue_enter(q, flags) < 0) {
2449 enter_succeeded = false;
2450 q = NULL;
2451 }
2452 }
2453
2454 if (enter_succeeded) {
2455 struct bio_list lower, same;
2456
2457 /* Create a fresh bio_list for all subordinate requests */
2458 bio_list_on_stack[1] = bio_list_on_stack[0];
2459 bio_list_init(&bio_list_on_stack[0]);
2460 ret = q->make_request_fn(q, bio);
2461
2462 /* sort new bios into those for a lower level
2463 * and those for the same level
2464 */
2465 bio_list_init(&lower);
2466 bio_list_init(&same);
2467 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
2468 if (q == bio->bi_disk->queue)
2469 bio_list_add(&same, bio);
2470 else
2471 bio_list_add(&lower, bio);
2472 /* now assemble so we handle the lowest level first */
2473 bio_list_merge(&bio_list_on_stack[0], &lower);
2474 bio_list_merge(&bio_list_on_stack[0], &same);
2475 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
2476 } else {
2477 if (unlikely(!blk_queue_dying(q) &&
2478 (bio->bi_opf & REQ_NOWAIT)))
2479 bio_wouldblock_error(bio);
2480 else
2481 bio_io_error(bio);
2482 }
2483 bio = bio_list_pop(&bio_list_on_stack[0]);
2484 } while (bio);
2485 current->bio_list = NULL; /* deactivate */
2486
2487out:
2488 if (q)
2489 blk_queue_exit(q);
2490 return ret;
2491}
2492EXPORT_SYMBOL(generic_make_request);
2493
2494/**
2495 * direct_make_request - hand a buffer directly to its device driver for I/O
2496 * @bio: The bio describing the location in memory and on the device.
2497 *
2498 * This function behaves like generic_make_request(), but does not protect
2499 * against recursion. Must only be used if the called driver is known
2500 * to not call generic_make_request (or direct_make_request) again from
2501 * its make_request function. (Calling direct_make_request again from
2502 * a workqueue is perfectly fine as that doesn't recurse).
2503 */
2504blk_qc_t direct_make_request(struct bio *bio)
2505{
2506 struct request_queue *q = bio->bi_disk->queue;
2507 bool nowait = bio->bi_opf & REQ_NOWAIT;
2508 blk_qc_t ret;
2509
2510 if (!generic_make_request_checks(bio))
2511 return BLK_QC_T_NONE;
2512
2513 if (unlikely(blk_queue_enter(q, nowait ? BLK_MQ_REQ_NOWAIT : 0))) {
2514 if (nowait && !blk_queue_dying(q))
2515 bio->bi_status = BLK_STS_AGAIN;
2516 else
2517 bio->bi_status = BLK_STS_IOERR;
2518 bio_endio(bio);
2519 return BLK_QC_T_NONE;
2520 }
2521
2522 ret = q->make_request_fn(q, bio);
2523 blk_queue_exit(q);
2524 return ret;
2525}
2526EXPORT_SYMBOL_GPL(direct_make_request);
2527
2528/**
2529 * submit_bio - submit a bio to the block device layer for I/O
2530 * @bio: The &struct bio which describes the I/O
2531 *
2532 * submit_bio() is very similar in purpose to generic_make_request(), and
2533 * uses that function to do most of the work. Both are fairly rough
2534 * interfaces; @bio must be presetup and ready for I/O.
2535 *
2536 */
2537blk_qc_t submit_bio(struct bio *bio)
2538{
2539 /*
2540 * If it's a regular read/write or a barrier with data attached,
2541 * go through the normal accounting stuff before submission.
2542 */
2543 if (bio_has_data(bio)) {
2544 unsigned int count;
2545
2546 if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
2547 count = queue_logical_block_size(bio->bi_disk->queue) >> 9;
2548 else
2549 count = bio_sectors(bio);
2550
2551 if (op_is_write(bio_op(bio))) {
2552 count_vm_events(PGPGOUT, count);
2553 } else {
2554 task_io_account_read(bio->bi_iter.bi_size);
2555 count_vm_events(PGPGIN, count);
2556 }
2557
2558 if (unlikely(block_dump)) {
2559 char b[BDEVNAME_SIZE];
2560 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
2561 current->comm, task_pid_nr(current),
2562 op_is_write(bio_op(bio)) ? "WRITE" : "READ",
2563 (unsigned long long)bio->bi_iter.bi_sector,
2564 bio_devname(bio, b), count);
2565 }
2566 }
2567
2568 return generic_make_request(bio);
2569}
2570EXPORT_SYMBOL(submit_bio);
2571
2572bool blk_poll(struct request_queue *q, blk_qc_t cookie)
2573{
2574 if (!q->poll_fn || !blk_qc_t_valid(cookie))
2575 return false;
2576
2577 if (current->plug)
2578 blk_flush_plug_list(current->plug, false);
2579 return q->poll_fn(q, cookie);
2580}
2581EXPORT_SYMBOL_GPL(blk_poll);
2582
2583/**
2584 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2585 * for new the queue limits
2586 * @q: the queue
2587 * @rq: the request being checked
2588 *
2589 * Description:
2590 * @rq may have been made based on weaker limitations of upper-level queues
2591 * in request stacking drivers, and it may violate the limitation of @q.
2592 * Since the block layer and the underlying device driver trust @rq
2593 * after it is inserted to @q, it should be checked against @q before
2594 * the insertion using this generic function.
2595 *
2596 * Request stacking drivers like request-based dm may change the queue
2597 * limits when retrying requests on other queues. Those requests need
2598 * to be checked against the new queue limits again during dispatch.
2599 */
2600static int blk_cloned_rq_check_limits(struct request_queue *q,
2601 struct request *rq)
2602{
2603 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) {
2604 printk(KERN_ERR "%s: over max size limit.\n", __func__);
2605 return -EIO;
2606 }
2607
2608 /*
2609 * queue's settings related to segment counting like q->bounce_pfn
2610 * may differ from that of other stacking queues.
2611 * Recalculate it to check the request correctly on this queue's
2612 * limitation.
2613 */
2614 blk_recalc_rq_segments(rq);
2615 if (rq->nr_phys_segments > queue_max_segments(q)) {
2616 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2617 return -EIO;
2618 }
2619
2620 return 0;
2621}
2622
2623/**
2624 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2625 * @q: the queue to submit the request
2626 * @rq: the request being queued
2627 */
2628blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq)
2629{
2630 unsigned long flags;
2631 int where = ELEVATOR_INSERT_BACK;
2632
2633 if (blk_cloned_rq_check_limits(q, rq))
2634 return BLK_STS_IOERR;
2635
2636 if (rq->rq_disk &&
2637 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
2638 return BLK_STS_IOERR;
2639
2640 if (q->mq_ops) {
2641 if (blk_queue_io_stat(q))
2642 blk_account_io_start(rq, true);
2643 /*
2644 * Since we have a scheduler attached on the top device,
2645 * bypass a potential scheduler on the bottom device for
2646 * insert.
2647 */
2648 return blk_mq_request_issue_directly(rq);
2649 }
2650
2651 spin_lock_irqsave(q->queue_lock, flags);
2652 if (unlikely(blk_queue_dying(q))) {
2653 spin_unlock_irqrestore(q->queue_lock, flags);
2654 return BLK_STS_IOERR;
2655 }
2656
2657 /*
2658 * Submitting request must be dequeued before calling this function
2659 * because it will be linked to another request_queue
2660 */
2661 BUG_ON(blk_queued_rq(rq));
2662
2663 if (op_is_flush(rq->cmd_flags))
2664 where = ELEVATOR_INSERT_FLUSH;
2665
2666 add_acct_request(q, rq, where);
2667 if (where == ELEVATOR_INSERT_FLUSH)
2668 __blk_run_queue(q);
2669 spin_unlock_irqrestore(q->queue_lock, flags);
2670
2671 return BLK_STS_OK;
2672}
2673EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2674
2675/**
2676 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2677 * @rq: request to examine
2678 *
2679 * Description:
2680 * A request could be merge of IOs which require different failure
2681 * handling. This function determines the number of bytes which
2682 * can be failed from the beginning of the request without
2683 * crossing into area which need to be retried further.
2684 *
2685 * Return:
2686 * The number of bytes to fail.
2687 */
2688unsigned int blk_rq_err_bytes(const struct request *rq)
2689{
2690 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2691 unsigned int bytes = 0;
2692 struct bio *bio;
2693
2694 if (!(rq->rq_flags & RQF_MIXED_MERGE))
2695 return blk_rq_bytes(rq);
2696
2697 /*
2698 * Currently the only 'mixing' which can happen is between
2699 * different fastfail types. We can safely fail portions
2700 * which have all the failfast bits that the first one has -
2701 * the ones which are at least as eager to fail as the first
2702 * one.
2703 */
2704 for (bio = rq->bio; bio; bio = bio->bi_next) {
2705 if ((bio->bi_opf & ff) != ff)
2706 break;
2707 bytes += bio->bi_iter.bi_size;
2708 }
2709
2710 /* this could lead to infinite loop */
2711 BUG_ON(blk_rq_bytes(rq) && !bytes);
2712 return bytes;
2713}
2714EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2715
2716void blk_account_io_completion(struct request *req, unsigned int bytes)
2717{
2718 if (blk_do_io_stat(req)) {
2719 const int rw = rq_data_dir(req);
2720 struct hd_struct *part;
2721 int cpu;
2722
2723 cpu = part_stat_lock();
2724 part = req->part;
2725 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2726 part_stat_unlock();
2727 }
2728}
2729
2730void blk_account_io_done(struct request *req)
2731{
2732 /*
2733 * Account IO completion. flush_rq isn't accounted as a
2734 * normal IO on queueing nor completion. Accounting the
2735 * containing request is enough.
2736 */
2737 if (blk_do_io_stat(req) && !(req->rq_flags & RQF_FLUSH_SEQ)) {
2738 unsigned long duration = jiffies - req->start_time;
2739 const int rw = rq_data_dir(req);
2740 struct hd_struct *part;
2741 int cpu;
2742
2743 cpu = part_stat_lock();
2744 part = req->part;
2745
2746 part_stat_inc(cpu, part, ios[rw]);
2747 part_stat_add(cpu, part, ticks[rw], duration);
2748 part_round_stats(req->q, cpu, part);
2749 part_dec_in_flight(req->q, part, rw);
2750
2751 hd_struct_put(part);
2752 part_stat_unlock();
2753 }
2754}
2755
2756#ifdef CONFIG_PM
2757/*
2758 * Don't process normal requests when queue is suspended
2759 * or in the process of suspending/resuming
2760 */
2761static bool blk_pm_allow_request(struct request *rq)
2762{
2763 switch (rq->q->rpm_status) {
2764 case RPM_RESUMING:
2765 case RPM_SUSPENDING:
2766 return rq->rq_flags & RQF_PM;
2767 case RPM_SUSPENDED:
2768 return false;
2769 }
2770
2771 return true;
2772}
2773#else
2774static bool blk_pm_allow_request(struct request *rq)
2775{
2776 return true;
2777}
2778#endif
2779
2780void blk_account_io_start(struct request *rq, bool new_io)
2781{
2782 struct hd_struct *part;
2783 int rw = rq_data_dir(rq);
2784 int cpu;
2785
2786 if (!blk_do_io_stat(rq))
2787 return;
2788
2789 cpu = part_stat_lock();
2790
2791 if (!new_io) {
2792 part = rq->part;
2793 part_stat_inc(cpu, part, merges[rw]);
2794 } else {
2795 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2796 if (!hd_struct_try_get(part)) {
2797 /*
2798 * The partition is already being removed,
2799 * the request will be accounted on the disk only
2800 *
2801 * We take a reference on disk->part0 although that
2802 * partition will never be deleted, so we can treat
2803 * it as any other partition.
2804 */
2805 part = &rq->rq_disk->part0;
2806 hd_struct_get(part);
2807 }
2808 part_round_stats(rq->q, cpu, part);
2809 part_inc_in_flight(rq->q, part, rw);
2810 rq->part = part;
2811 }
2812
2813 part_stat_unlock();
2814}
2815
2816static struct request *elv_next_request(struct request_queue *q)
2817{
2818 struct request *rq;
2819 struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
2820
2821 WARN_ON_ONCE(q->mq_ops);
2822
2823 while (1) {
2824 list_for_each_entry(rq, &q->queue_head, queuelist) {
2825 if (blk_pm_allow_request(rq))
2826 return rq;
2827
2828 if (rq->rq_flags & RQF_SOFTBARRIER)
2829 break;
2830 }
2831
2832 /*
2833 * Flush request is running and flush request isn't queueable
2834 * in the drive, we can hold the queue till flush request is
2835 * finished. Even we don't do this, driver can't dispatch next
2836 * requests and will requeue them. And this can improve
2837 * throughput too. For example, we have request flush1, write1,
2838 * flush 2. flush1 is dispatched, then queue is hold, write1
2839 * isn't inserted to queue. After flush1 is finished, flush2
2840 * will be dispatched. Since disk cache is already clean,
2841 * flush2 will be finished very soon, so looks like flush2 is
2842 * folded to flush1.
2843 * Since the queue is hold, a flag is set to indicate the queue
2844 * should be restarted later. Please see flush_end_io() for
2845 * details.
2846 */
2847 if (fq->flush_pending_idx != fq->flush_running_idx &&
2848 !queue_flush_queueable(q)) {
2849 fq->flush_queue_delayed = 1;
2850 return NULL;
2851 }
2852 if (unlikely(blk_queue_bypass(q)) ||
2853 !q->elevator->type->ops.sq.elevator_dispatch_fn(q, 0))
2854 return NULL;
2855 }
2856}
2857
2858/**
2859 * blk_peek_request - peek at the top of a request queue
2860 * @q: request queue to peek at
2861 *
2862 * Description:
2863 * Return the request at the top of @q. The returned request
2864 * should be started using blk_start_request() before LLD starts
2865 * processing it.
2866 *
2867 * Return:
2868 * Pointer to the request at the top of @q if available. Null
2869 * otherwise.
2870 */
2871struct request *blk_peek_request(struct request_queue *q)
2872{
2873 struct request *rq;
2874 int ret;
2875
2876 lockdep_assert_held(q->queue_lock);
2877 WARN_ON_ONCE(q->mq_ops);
2878
2879 while ((rq = elv_next_request(q)) != NULL) {
2880 if (!(rq->rq_flags & RQF_STARTED)) {
2881 /*
2882 * This is the first time the device driver
2883 * sees this request (possibly after
2884 * requeueing). Notify IO scheduler.
2885 */
2886 if (rq->rq_flags & RQF_SORTED)
2887 elv_activate_rq(q, rq);
2888
2889 /*
2890 * just mark as started even if we don't start
2891 * it, a request that has been delayed should
2892 * not be passed by new incoming requests
2893 */
2894 rq->rq_flags |= RQF_STARTED;
2895 trace_block_rq_issue(q, rq);
2896 }
2897
2898 if (!q->boundary_rq || q->boundary_rq == rq) {
2899 q->end_sector = rq_end_sector(rq);
2900 q->boundary_rq = NULL;
2901 }
2902
2903 if (rq->rq_flags & RQF_DONTPREP)
2904 break;
2905
2906 if (q->dma_drain_size && blk_rq_bytes(rq)) {
2907 /*
2908 * make sure space for the drain appears we
2909 * know we can do this because max_hw_segments
2910 * has been adjusted to be one fewer than the
2911 * device can handle
2912 */
2913 rq->nr_phys_segments++;
2914 }
2915
2916 if (!q->prep_rq_fn)
2917 break;
2918
2919 ret = q->prep_rq_fn(q, rq);
2920 if (ret == BLKPREP_OK) {
2921 break;
2922 } else if (ret == BLKPREP_DEFER) {
2923 /*
2924 * the request may have been (partially) prepped.
2925 * we need to keep this request in the front to
2926 * avoid resource deadlock. RQF_STARTED will
2927 * prevent other fs requests from passing this one.
2928 */
2929 if (q->dma_drain_size && blk_rq_bytes(rq) &&
2930 !(rq->rq_flags & RQF_DONTPREP)) {
2931 /*
2932 * remove the space for the drain we added
2933 * so that we don't add it again
2934 */
2935 --rq->nr_phys_segments;
2936 }
2937
2938 rq = NULL;
2939 break;
2940 } else if (ret == BLKPREP_KILL || ret == BLKPREP_INVALID) {
2941 rq->rq_flags |= RQF_QUIET;
2942 /*
2943 * Mark this request as started so we don't trigger
2944 * any debug logic in the end I/O path.
2945 */
2946 blk_start_request(rq);
2947 __blk_end_request_all(rq, ret == BLKPREP_INVALID ?
2948 BLK_STS_TARGET : BLK_STS_IOERR);
2949 } else {
2950 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2951 break;
2952 }
2953 }
2954
2955 return rq;
2956}
2957EXPORT_SYMBOL(blk_peek_request);
2958
2959static void blk_dequeue_request(struct request *rq)
2960{
2961 struct request_queue *q = rq->q;
2962
2963 BUG_ON(list_empty(&rq->queuelist));
2964 BUG_ON(ELV_ON_HASH(rq));
2965
2966 list_del_init(&rq->queuelist);
2967
2968 /*
2969 * the time frame between a request being removed from the lists
2970 * and to it is freed is accounted as io that is in progress at
2971 * the driver side.
2972 */
2973 if (blk_account_rq(rq)) {
2974 q->in_flight[rq_is_sync(rq)]++;
2975 set_io_start_time_ns(rq);
2976 }
2977}
2978
2979/**
2980 * blk_start_request - start request processing on the driver
2981 * @req: request to dequeue
2982 *
2983 * Description:
2984 * Dequeue @req and start timeout timer on it. This hands off the
2985 * request to the driver.
2986 */
2987void blk_start_request(struct request *req)
2988{
2989 lockdep_assert_held(req->q->queue_lock);
2990 WARN_ON_ONCE(req->q->mq_ops);
2991
2992 blk_dequeue_request(req);
2993
2994 if (test_bit(QUEUE_FLAG_STATS, &req->q->queue_flags)) {
2995 blk_stat_set_issue(&req->issue_stat, blk_rq_sectors(req));
2996 req->rq_flags |= RQF_STATS;
2997 wbt_issue(req->q->rq_wb, &req->issue_stat);
2998 }
2999
3000 BUG_ON(blk_rq_is_complete(req));
3001 blk_add_timer(req);
3002}
3003EXPORT_SYMBOL(blk_start_request);
3004
3005/**
3006 * blk_fetch_request - fetch a request from a request queue
3007 * @q: request queue to fetch a request from
3008 *
3009 * Description:
3010 * Return the request at the top of @q. The request is started on
3011 * return and LLD can start processing it immediately.
3012 *
3013 * Return:
3014 * Pointer to the request at the top of @q if available. Null
3015 * otherwise.
3016 */
3017struct request *blk_fetch_request(struct request_queue *q)
3018{
3019 struct request *rq;
3020
3021 lockdep_assert_held(q->queue_lock);
3022 WARN_ON_ONCE(q->mq_ops);
3023
3024 rq = blk_peek_request(q);
3025 if (rq)
3026 blk_start_request(rq);
3027 return rq;
3028}
3029EXPORT_SYMBOL(blk_fetch_request);
3030
3031/*
3032 * Steal bios from a request and add them to a bio list.
3033 * The request must not have been partially completed before.
3034 */
3035void blk_steal_bios(struct bio_list *list, struct request *rq)
3036{
3037 if (rq->bio) {
3038 if (list->tail)
3039 list->tail->bi_next = rq->bio;
3040 else
3041 list->head = rq->bio;
3042 list->tail = rq->biotail;
3043
3044 rq->bio = NULL;
3045 rq->biotail = NULL;
3046 }
3047
3048 rq->__data_len = 0;
3049}
3050EXPORT_SYMBOL_GPL(blk_steal_bios);
3051
3052/**
3053 * blk_update_request - Special helper function for request stacking drivers
3054 * @req: the request being processed
3055 * @error: block status code
3056 * @nr_bytes: number of bytes to complete @req
3057 *
3058 * Description:
3059 * Ends I/O on a number of bytes attached to @req, but doesn't complete
3060 * the request structure even if @req doesn't have leftover.
3061 * If @req has leftover, sets it up for the next range of segments.
3062 *
3063 * This special helper function is only for request stacking drivers
3064 * (e.g. request-based dm) so that they can handle partial completion.
3065 * Actual device drivers should use blk_end_request instead.
3066 *
3067 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
3068 * %false return from this function.
3069 *
3070 * Return:
3071 * %false - this request doesn't have any more data
3072 * %true - this request has more data
3073 **/
3074bool blk_update_request(struct request *req, blk_status_t error,
3075 unsigned int nr_bytes)
3076{
3077 int total_bytes;
3078
3079 trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes);
3080
3081 if (!req->bio)
3082 return false;
3083
3084 if (unlikely(error && !blk_rq_is_passthrough(req) &&
3085 !(req->rq_flags & RQF_QUIET)))
3086 print_req_error(req, error);
3087
3088 blk_account_io_completion(req, nr_bytes);
3089
3090 total_bytes = 0;
3091 while (req->bio) {
3092 struct bio *bio = req->bio;
3093 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
3094
3095 if (bio_bytes == bio->bi_iter.bi_size)
3096 req->bio = bio->bi_next;
3097
3098 /* Completion has already been traced */
3099 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
3100 req_bio_endio(req, bio, bio_bytes, error);
3101
3102 total_bytes += bio_bytes;
3103 nr_bytes -= bio_bytes;
3104
3105 if (!nr_bytes)
3106 break;
3107 }
3108
3109 /*
3110 * completely done
3111 */
3112 if (!req->bio) {
3113 /*
3114 * Reset counters so that the request stacking driver
3115 * can find how many bytes remain in the request
3116 * later.
3117 */
3118 req->__data_len = 0;
3119 return false;
3120 }
3121
3122 req->__data_len -= total_bytes;
3123
3124 /* update sector only for requests with clear definition of sector */
3125 if (!blk_rq_is_passthrough(req))
3126 req->__sector += total_bytes >> 9;
3127
3128 /* mixed attributes always follow the first bio */
3129 if (req->rq_flags & RQF_MIXED_MERGE) {
3130 req->cmd_flags &= ~REQ_FAILFAST_MASK;
3131 req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
3132 }
3133
3134 if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) {
3135 /*
3136 * If total number of sectors is less than the first segment
3137 * size, something has gone terribly wrong.
3138 */
3139 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
3140 blk_dump_rq_flags(req, "request botched");
3141 req->__data_len = blk_rq_cur_bytes(req);
3142 }
3143
3144 /* recalculate the number of segments */
3145 blk_recalc_rq_segments(req);
3146 }
3147
3148 return true;
3149}
3150EXPORT_SYMBOL_GPL(blk_update_request);
3151
3152static bool blk_update_bidi_request(struct request *rq, blk_status_t error,
3153 unsigned int nr_bytes,
3154 unsigned int bidi_bytes)
3155{
3156 if (blk_update_request(rq, error, nr_bytes))
3157 return true;
3158
3159 /* Bidi request must be completed as a whole */
3160 if (unlikely(blk_bidi_rq(rq)) &&
3161 blk_update_request(rq->next_rq, error, bidi_bytes))
3162 return true;
3163
3164 if (blk_queue_add_random(rq->q))
3165 add_disk_randomness(rq->rq_disk);
3166
3167 return false;
3168}
3169
3170/**
3171 * blk_unprep_request - unprepare a request
3172 * @req: the request
3173 *
3174 * This function makes a request ready for complete resubmission (or
3175 * completion). It happens only after all error handling is complete,
3176 * so represents the appropriate moment to deallocate any resources
3177 * that were allocated to the request in the prep_rq_fn. The queue
3178 * lock is held when calling this.
3179 */
3180void blk_unprep_request(struct request *req)
3181{
3182 struct request_queue *q = req->q;
3183
3184 req->rq_flags &= ~RQF_DONTPREP;
3185 if (q->unprep_rq_fn)
3186 q->unprep_rq_fn(q, req);
3187}
3188EXPORT_SYMBOL_GPL(blk_unprep_request);
3189
3190void blk_finish_request(struct request *req, blk_status_t error)
3191{
3192 struct request_queue *q = req->q;
3193
3194 lockdep_assert_held(req->q->queue_lock);
3195 WARN_ON_ONCE(q->mq_ops);
3196
3197 if (req->rq_flags & RQF_STATS)
3198 blk_stat_add(req);
3199
3200 if (req->rq_flags & RQF_QUEUED)
3201 blk_queue_end_tag(q, req);
3202
3203 BUG_ON(blk_queued_rq(req));
3204
3205 if (unlikely(laptop_mode) && !blk_rq_is_passthrough(req))
3206 laptop_io_completion(req->q->backing_dev_info);
3207
3208 blk_delete_timer(req);
3209
3210 if (req->rq_flags & RQF_DONTPREP)
3211 blk_unprep_request(req);
3212
3213 blk_account_io_done(req);
3214
3215 if (req->end_io) {
3216 wbt_done(req->q->rq_wb, &req->issue_stat);
3217 req->end_io(req, error);
3218 } else {
3219 if (blk_bidi_rq(req))
3220 __blk_put_request(req->next_rq->q, req->next_rq);
3221
3222 __blk_put_request(q, req);
3223 }
3224}
3225EXPORT_SYMBOL(blk_finish_request);
3226
3227/**
3228 * blk_end_bidi_request - Complete a bidi request
3229 * @rq: the request to complete
3230 * @error: block status code
3231 * @nr_bytes: number of bytes to complete @rq
3232 * @bidi_bytes: number of bytes to complete @rq->next_rq
3233 *
3234 * Description:
3235 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
3236 * Drivers that supports bidi can safely call this member for any
3237 * type of request, bidi or uni. In the later case @bidi_bytes is
3238 * just ignored.
3239 *
3240 * Return:
3241 * %false - we are done with this request
3242 * %true - still buffers pending for this request
3243 **/
3244static bool blk_end_bidi_request(struct request *rq, blk_status_t error,
3245 unsigned int nr_bytes, unsigned int bidi_bytes)
3246{
3247 struct request_queue *q = rq->q;
3248 unsigned long flags;
3249
3250 WARN_ON_ONCE(q->mq_ops);
3251
3252 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
3253 return true;
3254
3255 spin_lock_irqsave(q->queue_lock, flags);
3256 blk_finish_request(rq, error);
3257 spin_unlock_irqrestore(q->queue_lock, flags);
3258
3259 return false;
3260}
3261
3262/**
3263 * __blk_end_bidi_request - Complete a bidi request with queue lock held
3264 * @rq: the request to complete
3265 * @error: block status code
3266 * @nr_bytes: number of bytes to complete @rq
3267 * @bidi_bytes: number of bytes to complete @rq->next_rq
3268 *
3269 * Description:
3270 * Identical to blk_end_bidi_request() except that queue lock is
3271 * assumed to be locked on entry and remains so on return.
3272 *
3273 * Return:
3274 * %false - we are done with this request
3275 * %true - still buffers pending for this request
3276 **/
3277static bool __blk_end_bidi_request(struct request *rq, blk_status_t error,
3278 unsigned int nr_bytes, unsigned int bidi_bytes)
3279{
3280 lockdep_assert_held(rq->q->queue_lock);
3281 WARN_ON_ONCE(rq->q->mq_ops);
3282
3283 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
3284 return true;
3285
3286 blk_finish_request(rq, error);
3287
3288 return false;
3289}
3290
3291/**
3292 * blk_end_request - Helper function for drivers to complete the request.
3293 * @rq: the request being processed
3294 * @error: block status code
3295 * @nr_bytes: number of bytes to complete
3296 *
3297 * Description:
3298 * Ends I/O on a number of bytes attached to @rq.
3299 * If @rq has leftover, sets it up for the next range of segments.
3300 *
3301 * Return:
3302 * %false - we are done with this request
3303 * %true - still buffers pending for this request
3304 **/
3305bool blk_end_request(struct request *rq, blk_status_t error,
3306 unsigned int nr_bytes)
3307{
3308 WARN_ON_ONCE(rq->q->mq_ops);
3309 return blk_end_bidi_request(rq, error, nr_bytes, 0);
3310}
3311EXPORT_SYMBOL(blk_end_request);
3312
3313/**
3314 * blk_end_request_all - Helper function for drives to finish the request.
3315 * @rq: the request to finish
3316 * @error: block status code
3317 *
3318 * Description:
3319 * Completely finish @rq.
3320 */
3321void blk_end_request_all(struct request *rq, blk_status_t error)
3322{
3323 bool pending;
3324 unsigned int bidi_bytes = 0;
3325
3326 if (unlikely(blk_bidi_rq(rq)))
3327 bidi_bytes = blk_rq_bytes(rq->next_rq);
3328
3329 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
3330 BUG_ON(pending);
3331}
3332EXPORT_SYMBOL(blk_end_request_all);
3333
3334/**
3335 * __blk_end_request - Helper function for drivers to complete the request.
3336 * @rq: the request being processed
3337 * @error: block status code
3338 * @nr_bytes: number of bytes to complete
3339 *
3340 * Description:
3341 * Must be called with queue lock held unlike blk_end_request().
3342 *
3343 * Return:
3344 * %false - we are done with this request
3345 * %true - still buffers pending for this request
3346 **/
3347bool __blk_end_request(struct request *rq, blk_status_t error,
3348 unsigned int nr_bytes)
3349{
3350 lockdep_assert_held(rq->q->queue_lock);
3351 WARN_ON_ONCE(rq->q->mq_ops);
3352
3353 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
3354}
3355EXPORT_SYMBOL(__blk_end_request);
3356
3357/**
3358 * __blk_end_request_all - Helper function for drives to finish the request.
3359 * @rq: the request to finish
3360 * @error: block status code
3361 *
3362 * Description:
3363 * Completely finish @rq. Must be called with queue lock held.
3364 */
3365void __blk_end_request_all(struct request *rq, blk_status_t error)
3366{
3367 bool pending;
3368 unsigned int bidi_bytes = 0;
3369
3370 lockdep_assert_held(rq->q->queue_lock);
3371 WARN_ON_ONCE(rq->q->mq_ops);
3372
3373 if (unlikely(blk_bidi_rq(rq)))
3374 bidi_bytes = blk_rq_bytes(rq->next_rq);
3375
3376 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
3377 BUG_ON(pending);
3378}
3379EXPORT_SYMBOL(__blk_end_request_all);
3380
3381/**
3382 * __blk_end_request_cur - Helper function to finish the current request chunk.
3383 * @rq: the request to finish the current chunk for
3384 * @error: block status code
3385 *
3386 * Description:
3387 * Complete the current consecutively mapped chunk from @rq. Must
3388 * be called with queue lock held.
3389 *
3390 * Return:
3391 * %false - we are done with this request
3392 * %true - still buffers pending for this request
3393 */
3394bool __blk_end_request_cur(struct request *rq, blk_status_t error)
3395{
3396 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
3397}
3398EXPORT_SYMBOL(__blk_end_request_cur);
3399
3400void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
3401 struct bio *bio)
3402{
3403 if (bio_has_data(bio))
3404 rq->nr_phys_segments = bio_phys_segments(q, bio);
3405 else if (bio_op(bio) == REQ_OP_DISCARD)
3406 rq->nr_phys_segments = 1;
3407
3408 rq->__data_len = bio->bi_iter.bi_size;
3409 rq->bio = rq->biotail = bio;
3410
3411 if (bio->bi_disk)
3412 rq->rq_disk = bio->bi_disk;
3413}
3414
3415#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
3416/**
3417 * rq_flush_dcache_pages - Helper function to flush all pages in a request
3418 * @rq: the request to be flushed
3419 *
3420 * Description:
3421 * Flush all pages in @rq.
3422 */
3423void rq_flush_dcache_pages(struct request *rq)
3424{
3425 struct req_iterator iter;
3426 struct bio_vec bvec;
3427
3428 rq_for_each_segment(bvec, rq, iter)
3429 flush_dcache_page(bvec.bv_page);
3430}
3431EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
3432#endif
3433
3434/**
3435 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
3436 * @q : the queue of the device being checked
3437 *
3438 * Description:
3439 * Check if underlying low-level drivers of a device are busy.
3440 * If the drivers want to export their busy state, they must set own
3441 * exporting function using blk_queue_lld_busy() first.
3442 *
3443 * Basically, this function is used only by request stacking drivers
3444 * to stop dispatching requests to underlying devices when underlying
3445 * devices are busy. This behavior helps more I/O merging on the queue
3446 * of the request stacking driver and prevents I/O throughput regression
3447 * on burst I/O load.
3448 *
3449 * Return:
3450 * 0 - Not busy (The request stacking driver should dispatch request)
3451 * 1 - Busy (The request stacking driver should stop dispatching request)
3452 */
3453int blk_lld_busy(struct request_queue *q)
3454{
3455 if (q->lld_busy_fn)
3456 return q->lld_busy_fn(q);
3457
3458 return 0;
3459}
3460EXPORT_SYMBOL_GPL(blk_lld_busy);
3461
3462/**
3463 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
3464 * @rq: the clone request to be cleaned up
3465 *
3466 * Description:
3467 * Free all bios in @rq for a cloned request.
3468 */
3469void blk_rq_unprep_clone(struct request *rq)
3470{
3471 struct bio *bio;
3472
3473 while ((bio = rq->bio) != NULL) {
3474 rq->bio = bio->bi_next;
3475
3476 bio_put(bio);
3477 }
3478}
3479EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
3480
3481/*
3482 * Copy attributes of the original request to the clone request.
3483 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3484 */
3485static void __blk_rq_prep_clone(struct request *dst, struct request *src)
3486{
3487 dst->cpu = src->cpu;
3488 dst->__sector = blk_rq_pos(src);
3489 dst->__data_len = blk_rq_bytes(src);
3490 dst->nr_phys_segments = src->nr_phys_segments;
3491 dst->ioprio = src->ioprio;
3492 dst->extra_len = src->extra_len;
3493}
3494
3495/**
3496 * blk_rq_prep_clone - Helper function to setup clone request
3497 * @rq: the request to be setup
3498 * @rq_src: original request to be cloned
3499 * @bs: bio_set that bios for clone are allocated from
3500 * @gfp_mask: memory allocation mask for bio
3501 * @bio_ctr: setup function to be called for each clone bio.
3502 * Returns %0 for success, non %0 for failure.
3503 * @data: private data to be passed to @bio_ctr
3504 *
3505 * Description:
3506 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3507 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3508 * are not copied, and copying such parts is the caller's responsibility.
3509 * Also, pages which the original bios are pointing to are not copied
3510 * and the cloned bios just point same pages.
3511 * So cloned bios must be completed before original bios, which means
3512 * the caller must complete @rq before @rq_src.
3513 */
3514int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
3515 struct bio_set *bs, gfp_t gfp_mask,
3516 int (*bio_ctr)(struct bio *, struct bio *, void *),
3517 void *data)
3518{
3519 struct bio *bio, *bio_src;
3520
3521 if (!bs)
3522 bs = fs_bio_set;
3523
3524 __rq_for_each_bio(bio_src, rq_src) {
3525 bio = bio_clone_fast(bio_src, gfp_mask, bs);
3526 if (!bio)
3527 goto free_and_out;
3528
3529 if (bio_ctr && bio_ctr(bio, bio_src, data))
3530 goto free_and_out;
3531
3532 if (rq->bio) {
3533 rq->biotail->bi_next = bio;
3534 rq->biotail = bio;
3535 } else
3536 rq->bio = rq->biotail = bio;
3537 }
3538
3539 __blk_rq_prep_clone(rq, rq_src);
3540
3541 return 0;
3542
3543free_and_out:
3544 if (bio)
3545 bio_put(bio);
3546 blk_rq_unprep_clone(rq);
3547
3548 return -ENOMEM;
3549}
3550EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
3551
3552int kblockd_schedule_work(struct work_struct *work)
3553{
3554 return queue_work(kblockd_workqueue, work);
3555}
3556EXPORT_SYMBOL(kblockd_schedule_work);
3557
3558int kblockd_schedule_work_on(int cpu, struct work_struct *work)
3559{
3560 return queue_work_on(cpu, kblockd_workqueue, work);
3561}
3562EXPORT_SYMBOL(kblockd_schedule_work_on);
3563
3564int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
3565 unsigned long delay)
3566{
3567 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3568}
3569EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
3570
3571/**
3572 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3573 * @plug: The &struct blk_plug that needs to be initialized
3574 *
3575 * Description:
3576 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3577 * pending I/O should the task end up blocking between blk_start_plug() and
3578 * blk_finish_plug(). This is important from a performance perspective, but
3579 * also ensures that we don't deadlock. For instance, if the task is blocking
3580 * for a memory allocation, memory reclaim could end up wanting to free a
3581 * page belonging to that request that is currently residing in our private
3582 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3583 * this kind of deadlock.
3584 */
3585void blk_start_plug(struct blk_plug *plug)
3586{
3587 struct task_struct *tsk = current;
3588
3589 /*
3590 * If this is a nested plug, don't actually assign it.
3591 */
3592 if (tsk->plug)
3593 return;
3594
3595 INIT_LIST_HEAD(&plug->list);
3596 INIT_LIST_HEAD(&plug->mq_list);
3597 INIT_LIST_HEAD(&plug->cb_list);
3598 /*
3599 * Store ordering should not be needed here, since a potential
3600 * preempt will imply a full memory barrier
3601 */
3602 tsk->plug = plug;
3603}
3604EXPORT_SYMBOL(blk_start_plug);
3605
3606static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
3607{
3608 struct request *rqa = container_of(a, struct request, queuelist);
3609 struct request *rqb = container_of(b, struct request, queuelist);
3610
3611 return !(rqa->q < rqb->q ||
3612 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
3613}
3614
3615/*
3616 * If 'from_schedule' is true, then postpone the dispatch of requests
3617 * until a safe kblockd context. We due this to avoid accidental big
3618 * additional stack usage in driver dispatch, in places where the originally
3619 * plugger did not intend it.
3620 */
3621static void queue_unplugged(struct request_queue *q, unsigned int depth,
3622 bool from_schedule)
3623 __releases(q->queue_lock)
3624{
3625 lockdep_assert_held(q->queue_lock);
3626
3627 trace_block_unplug(q, depth, !from_schedule);
3628
3629 if (from_schedule)
3630 blk_run_queue_async(q);
3631 else
3632 __blk_run_queue(q);
3633 spin_unlock(q->queue_lock);
3634}
3635
3636static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
3637{
3638 LIST_HEAD(callbacks);
3639
3640 while (!list_empty(&plug->cb_list)) {
3641 list_splice_init(&plug->cb_list, &callbacks);
3642
3643 while (!list_empty(&callbacks)) {
3644 struct blk_plug_cb *cb = list_first_entry(&callbacks,
3645 struct blk_plug_cb,
3646 list);
3647 list_del(&cb->list);
3648 cb->callback(cb, from_schedule);
3649 }
3650 }
3651}
3652
3653struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3654 int size)
3655{
3656 struct blk_plug *plug = current->plug;
3657 struct blk_plug_cb *cb;
3658
3659 if (!plug)
3660 return NULL;
3661
3662 list_for_each_entry(cb, &plug->cb_list, list)
3663 if (cb->callback == unplug && cb->data == data)
3664 return cb;
3665
3666 /* Not currently on the callback list */
3667 BUG_ON(size < sizeof(*cb));
3668 cb = kzalloc(size, GFP_ATOMIC);
3669 if (cb) {
3670 cb->data = data;
3671 cb->callback = unplug;
3672 list_add(&cb->list, &plug->cb_list);
3673 }
3674 return cb;
3675}
3676EXPORT_SYMBOL(blk_check_plugged);
3677
3678void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
3679{
3680 struct request_queue *q;
3681 unsigned long flags;
3682 struct request *rq;
3683 LIST_HEAD(list);
3684 unsigned int depth;
3685
3686 flush_plug_callbacks(plug, from_schedule);
3687
3688 if (!list_empty(&plug->mq_list))
3689 blk_mq_flush_plug_list(plug, from_schedule);
3690
3691 if (list_empty(&plug->list))
3692 return;
3693
3694 list_splice_init(&plug->list, &list);
3695
3696 list_sort(NULL, &list, plug_rq_cmp);
3697
3698 q = NULL;
3699 depth = 0;
3700
3701 /*
3702 * Save and disable interrupts here, to avoid doing it for every
3703 * queue lock we have to take.
3704 */
3705 local_irq_save(flags);
3706 while (!list_empty(&list)) {
3707 rq = list_entry_rq(list.next);
3708 list_del_init(&rq->queuelist);
3709 BUG_ON(!rq->q);
3710 if (rq->q != q) {
3711 /*
3712 * This drops the queue lock
3713 */
3714 if (q)
3715 queue_unplugged(q, depth, from_schedule);
3716 q = rq->q;
3717 depth = 0;
3718 spin_lock(q->queue_lock);
3719 }
3720
3721 /*
3722 * Short-circuit if @q is dead
3723 */
3724 if (unlikely(blk_queue_dying(q))) {
3725 __blk_end_request_all(rq, BLK_STS_IOERR);
3726 continue;
3727 }
3728
3729 /*
3730 * rq is already accounted, so use raw insert
3731 */
3732 if (op_is_flush(rq->cmd_flags))
3733 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3734 else
3735 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
3736
3737 depth++;
3738 }
3739
3740 /*
3741 * This drops the queue lock
3742 */
3743 if (q)
3744 queue_unplugged(q, depth, from_schedule);
3745
3746 local_irq_restore(flags);
3747}
3748
3749void blk_finish_plug(struct blk_plug *plug)
3750{
3751 if (plug != current->plug)
3752 return;
3753 blk_flush_plug_list(plug, false);
3754
3755 current->plug = NULL;
3756}
3757EXPORT_SYMBOL(blk_finish_plug);
3758
3759#ifdef CONFIG_PM
3760/**
3761 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3762 * @q: the queue of the device
3763 * @dev: the device the queue belongs to
3764 *
3765 * Description:
3766 * Initialize runtime-PM-related fields for @q and start auto suspend for
3767 * @dev. Drivers that want to take advantage of request-based runtime PM
3768 * should call this function after @dev has been initialized, and its
3769 * request queue @q has been allocated, and runtime PM for it can not happen
3770 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3771 * cases, driver should call this function before any I/O has taken place.
3772 *
3773 * This function takes care of setting up using auto suspend for the device,
3774 * the autosuspend delay is set to -1 to make runtime suspend impossible
3775 * until an updated value is either set by user or by driver. Drivers do
3776 * not need to touch other autosuspend settings.
3777 *
3778 * The block layer runtime PM is request based, so only works for drivers
3779 * that use request as their IO unit instead of those directly use bio's.
3780 */
3781void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3782{
3783 /* not support for RQF_PM and ->rpm_status in blk-mq yet */
3784 if (q->mq_ops)
3785 return;
3786
3787 q->dev = dev;
3788 q->rpm_status = RPM_ACTIVE;
3789 pm_runtime_set_autosuspend_delay(q->dev, -1);
3790 pm_runtime_use_autosuspend(q->dev);
3791}
3792EXPORT_SYMBOL(blk_pm_runtime_init);
3793
3794/**
3795 * blk_pre_runtime_suspend - Pre runtime suspend check
3796 * @q: the queue of the device
3797 *
3798 * Description:
3799 * This function will check if runtime suspend is allowed for the device
3800 * by examining if there are any requests pending in the queue. If there
3801 * are requests pending, the device can not be runtime suspended; otherwise,
3802 * the queue's status will be updated to SUSPENDING and the driver can
3803 * proceed to suspend the device.
3804 *
3805 * For the not allowed case, we mark last busy for the device so that
3806 * runtime PM core will try to autosuspend it some time later.
3807 *
3808 * This function should be called near the start of the device's
3809 * runtime_suspend callback.
3810 *
3811 * Return:
3812 * 0 - OK to runtime suspend the device
3813 * -EBUSY - Device should not be runtime suspended
3814 */
3815int blk_pre_runtime_suspend(struct request_queue *q)
3816{
3817 int ret = 0;
3818
3819 if (!q->dev)
3820 return ret;
3821
3822 spin_lock_irq(q->queue_lock);
3823 if (q->nr_pending) {
3824 ret = -EBUSY;
3825 pm_runtime_mark_last_busy(q->dev);
3826 } else {
3827 q->rpm_status = RPM_SUSPENDING;
3828 }
3829 spin_unlock_irq(q->queue_lock);
3830 return ret;
3831}
3832EXPORT_SYMBOL(blk_pre_runtime_suspend);
3833
3834/**
3835 * blk_post_runtime_suspend - Post runtime suspend processing
3836 * @q: the queue of the device
3837 * @err: return value of the device's runtime_suspend function
3838 *
3839 * Description:
3840 * Update the queue's runtime status according to the return value of the
3841 * device's runtime suspend function and mark last busy for the device so
3842 * that PM core will try to auto suspend the device at a later time.
3843 *
3844 * This function should be called near the end of the device's
3845 * runtime_suspend callback.
3846 */
3847void blk_post_runtime_suspend(struct request_queue *q, int err)
3848{
3849 if (!q->dev)
3850 return;
3851
3852 spin_lock_irq(q->queue_lock);
3853 if (!err) {
3854 q->rpm_status = RPM_SUSPENDED;
3855 } else {
3856 q->rpm_status = RPM_ACTIVE;
3857 pm_runtime_mark_last_busy(q->dev);
3858 }
3859 spin_unlock_irq(q->queue_lock);
3860}
3861EXPORT_SYMBOL(blk_post_runtime_suspend);
3862
3863/**
3864 * blk_pre_runtime_resume - Pre runtime resume processing
3865 * @q: the queue of the device
3866 *
3867 * Description:
3868 * Update the queue's runtime status to RESUMING in preparation for the
3869 * runtime resume of the device.
3870 *
3871 * This function should be called near the start of the device's
3872 * runtime_resume callback.
3873 */
3874void blk_pre_runtime_resume(struct request_queue *q)
3875{
3876 if (!q->dev)
3877 return;
3878
3879 spin_lock_irq(q->queue_lock);
3880 q->rpm_status = RPM_RESUMING;
3881 spin_unlock_irq(q->queue_lock);
3882}
3883EXPORT_SYMBOL(blk_pre_runtime_resume);
3884
3885/**
3886 * blk_post_runtime_resume - Post runtime resume processing
3887 * @q: the queue of the device
3888 * @err: return value of the device's runtime_resume function
3889 *
3890 * Description:
3891 * Update the queue's runtime status according to the return value of the
3892 * device's runtime_resume function. If it is successfully resumed, process
3893 * the requests that are queued into the device's queue when it is resuming
3894 * and then mark last busy and initiate autosuspend for it.
3895 *
3896 * This function should be called near the end of the device's
3897 * runtime_resume callback.
3898 */
3899void blk_post_runtime_resume(struct request_queue *q, int err)
3900{
3901 if (!q->dev)
3902 return;
3903
3904 spin_lock_irq(q->queue_lock);
3905 if (!err) {
3906 q->rpm_status = RPM_ACTIVE;
3907 __blk_run_queue(q);
3908 pm_runtime_mark_last_busy(q->dev);
3909 pm_request_autosuspend(q->dev);
3910 } else {
3911 q->rpm_status = RPM_SUSPENDED;
3912 }
3913 spin_unlock_irq(q->queue_lock);
3914}
3915EXPORT_SYMBOL(blk_post_runtime_resume);
3916
3917/**
3918 * blk_set_runtime_active - Force runtime status of the queue to be active
3919 * @q: the queue of the device
3920 *
3921 * If the device is left runtime suspended during system suspend the resume
3922 * hook typically resumes the device and corrects runtime status
3923 * accordingly. However, that does not affect the queue runtime PM status
3924 * which is still "suspended". This prevents processing requests from the
3925 * queue.
3926 *
3927 * This function can be used in driver's resume hook to correct queue
3928 * runtime PM status and re-enable peeking requests from the queue. It
3929 * should be called before first request is added to the queue.
3930 */
3931void blk_set_runtime_active(struct request_queue *q)
3932{
3933 spin_lock_irq(q->queue_lock);
3934 q->rpm_status = RPM_ACTIVE;
3935 pm_runtime_mark_last_busy(q->dev);
3936 pm_request_autosuspend(q->dev);
3937 spin_unlock_irq(q->queue_lock);
3938}
3939EXPORT_SYMBOL(blk_set_runtime_active);
3940#endif
3941
3942int __init blk_dev_init(void)
3943{
3944 BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
3945 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
3946 FIELD_SIZEOF(struct request, cmd_flags));
3947 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
3948 FIELD_SIZEOF(struct bio, bi_opf));
3949
3950 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3951 kblockd_workqueue = alloc_workqueue("kblockd",
3952 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
3953 if (!kblockd_workqueue)
3954 panic("Failed to create kblockd\n");
3955
3956 request_cachep = kmem_cache_create("blkdev_requests",
3957 sizeof(struct request), 0, SLAB_PANIC, NULL);
3958
3959 blk_requestq_cachep = kmem_cache_create("request_queue",
3960 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
3961
3962#ifdef CONFIG_DEBUG_FS
3963 blk_debugfs_root = debugfs_create_dir("block", NULL);
3964#endif
3965
3966 return 0;
3967}
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/bio.h>
18#include <linux/blkdev.h>
19#include <linux/blk-pm.h>
20#include <linux/blk-integrity.h>
21#include <linux/highmem.h>
22#include <linux/mm.h>
23#include <linux/pagemap.h>
24#include <linux/kernel_stat.h>
25#include <linux/string.h>
26#include <linux/init.h>
27#include <linux/completion.h>
28#include <linux/slab.h>
29#include <linux/swap.h>
30#include <linux/writeback.h>
31#include <linux/task_io_accounting_ops.h>
32#include <linux/fault-inject.h>
33#include <linux/list_sort.h>
34#include <linux/delay.h>
35#include <linux/ratelimit.h>
36#include <linux/pm_runtime.h>
37#include <linux/t10-pi.h>
38#include <linux/debugfs.h>
39#include <linux/bpf.h>
40#include <linux/part_stat.h>
41#include <linux/sched/sysctl.h>
42#include <linux/blk-crypto.h>
43
44#define CREATE_TRACE_POINTS
45#include <trace/events/block.h>
46
47#include "blk.h"
48#include "blk-mq-sched.h"
49#include "blk-pm.h"
50#include "blk-cgroup.h"
51#include "blk-throttle.h"
52#include "blk-ioprio.h"
53
54struct dentry *blk_debugfs_root;
55
56EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
57EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
58EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
59EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
60EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
61EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_insert);
62
63static DEFINE_IDA(blk_queue_ida);
64
65/*
66 * For queue allocation
67 */
68static struct kmem_cache *blk_requestq_cachep;
69
70/*
71 * Controlling structure to kblockd
72 */
73static 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 */
80void blk_queue_flag_set(unsigned int flag, struct request_queue *q)
81{
82 set_bit(flag, &q->queue_flags);
83}
84EXPORT_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 */
91void blk_queue_flag_clear(unsigned int flag, struct request_queue *q)
92{
93 clear_bit(flag, &q->queue_flags);
94}
95EXPORT_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 */
105bool 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}
109EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set);
110
111#define REQ_OP_NAME(name) [REQ_OP_##name] = #name
112static const char *const blk_op_name[] = {
113 REQ_OP_NAME(READ),
114 REQ_OP_NAME(WRITE),
115 REQ_OP_NAME(FLUSH),
116 REQ_OP_NAME(DISCARD),
117 REQ_OP_NAME(SECURE_ERASE),
118 REQ_OP_NAME(ZONE_RESET),
119 REQ_OP_NAME(ZONE_RESET_ALL),
120 REQ_OP_NAME(ZONE_OPEN),
121 REQ_OP_NAME(ZONE_CLOSE),
122 REQ_OP_NAME(ZONE_FINISH),
123 REQ_OP_NAME(ZONE_APPEND),
124 REQ_OP_NAME(WRITE_ZEROES),
125 REQ_OP_NAME(DRV_IN),
126 REQ_OP_NAME(DRV_OUT),
127};
128#undef REQ_OP_NAME
129
130/**
131 * blk_op_str - Return string XXX in the REQ_OP_XXX.
132 * @op: REQ_OP_XXX.
133 *
134 * Description: Centralize block layer function to convert REQ_OP_XXX into
135 * string format. Useful in the debugging and tracing bio or request. For
136 * invalid REQ_OP_XXX it returns string "UNKNOWN".
137 */
138inline const char *blk_op_str(enum req_op op)
139{
140 const char *op_str = "UNKNOWN";
141
142 if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op])
143 op_str = blk_op_name[op];
144
145 return op_str;
146}
147EXPORT_SYMBOL_GPL(blk_op_str);
148
149static const struct {
150 int errno;
151 const char *name;
152} blk_errors[] = {
153 [BLK_STS_OK] = { 0, "" },
154 [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" },
155 [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" },
156 [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" },
157 [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" },
158 [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" },
159 [BLK_STS_RESV_CONFLICT] = { -EBADE, "reservation conflict" },
160 [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" },
161 [BLK_STS_PROTECTION] = { -EILSEQ, "protection" },
162 [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" },
163 [BLK_STS_DEV_RESOURCE] = { -EBUSY, "device resource" },
164 [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" },
165 [BLK_STS_OFFLINE] = { -ENODEV, "device offline" },
166
167 /* device mapper special case, should not leak out: */
168 [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" },
169
170 /* zone device specific errors */
171 [BLK_STS_ZONE_OPEN_RESOURCE] = { -ETOOMANYREFS, "open zones exceeded" },
172 [BLK_STS_ZONE_ACTIVE_RESOURCE] = { -EOVERFLOW, "active zones exceeded" },
173
174 /* Command duration limit device-side timeout */
175 [BLK_STS_DURATION_LIMIT] = { -ETIME, "duration limit exceeded" },
176
177 /* everything else not covered above: */
178 [BLK_STS_IOERR] = { -EIO, "I/O" },
179};
180
181blk_status_t errno_to_blk_status(int errno)
182{
183 int i;
184
185 for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
186 if (blk_errors[i].errno == errno)
187 return (__force blk_status_t)i;
188 }
189
190 return BLK_STS_IOERR;
191}
192EXPORT_SYMBOL_GPL(errno_to_blk_status);
193
194int blk_status_to_errno(blk_status_t status)
195{
196 int idx = (__force int)status;
197
198 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
199 return -EIO;
200 return blk_errors[idx].errno;
201}
202EXPORT_SYMBOL_GPL(blk_status_to_errno);
203
204const char *blk_status_to_str(blk_status_t status)
205{
206 int idx = (__force int)status;
207
208 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
209 return "<null>";
210 return blk_errors[idx].name;
211}
212EXPORT_SYMBOL_GPL(blk_status_to_str);
213
214/**
215 * blk_sync_queue - cancel any pending callbacks on a queue
216 * @q: the queue
217 *
218 * Description:
219 * The block layer may perform asynchronous callback activity
220 * on a queue, such as calling the unplug function after a timeout.
221 * A block device may call blk_sync_queue to ensure that any
222 * such activity is cancelled, thus allowing it to release resources
223 * that the callbacks might use. The caller must already have made sure
224 * that its ->submit_bio will not re-add plugging prior to calling
225 * this function.
226 *
227 * This function does not cancel any asynchronous activity arising
228 * out of elevator or throttling code. That would require elevator_exit()
229 * and blkcg_exit_queue() to be called with queue lock initialized.
230 *
231 */
232void blk_sync_queue(struct request_queue *q)
233{
234 del_timer_sync(&q->timeout);
235 cancel_work_sync(&q->timeout_work);
236}
237EXPORT_SYMBOL(blk_sync_queue);
238
239/**
240 * blk_set_pm_only - increment pm_only counter
241 * @q: request queue pointer
242 */
243void blk_set_pm_only(struct request_queue *q)
244{
245 atomic_inc(&q->pm_only);
246}
247EXPORT_SYMBOL_GPL(blk_set_pm_only);
248
249void blk_clear_pm_only(struct request_queue *q)
250{
251 int pm_only;
252
253 pm_only = atomic_dec_return(&q->pm_only);
254 WARN_ON_ONCE(pm_only < 0);
255 if (pm_only == 0)
256 wake_up_all(&q->mq_freeze_wq);
257}
258EXPORT_SYMBOL_GPL(blk_clear_pm_only);
259
260static void blk_free_queue_rcu(struct rcu_head *rcu_head)
261{
262 struct request_queue *q = container_of(rcu_head,
263 struct request_queue, rcu_head);
264
265 percpu_ref_exit(&q->q_usage_counter);
266 kmem_cache_free(blk_requestq_cachep, q);
267}
268
269static void blk_free_queue(struct request_queue *q)
270{
271 blk_free_queue_stats(q->stats);
272 if (queue_is_mq(q))
273 blk_mq_release(q);
274
275 ida_free(&blk_queue_ida, q->id);
276 call_rcu(&q->rcu_head, blk_free_queue_rcu);
277}
278
279/**
280 * blk_put_queue - decrement the request_queue refcount
281 * @q: the request_queue structure to decrement the refcount for
282 *
283 * Decrements the refcount of the request_queue and free it when the refcount
284 * reaches 0.
285 */
286void blk_put_queue(struct request_queue *q)
287{
288 if (refcount_dec_and_test(&q->refs))
289 blk_free_queue(q);
290}
291EXPORT_SYMBOL(blk_put_queue);
292
293void blk_queue_start_drain(struct request_queue *q)
294{
295 /*
296 * When queue DYING flag is set, we need to block new req
297 * entering queue, so we call blk_freeze_queue_start() to
298 * prevent I/O from crossing blk_queue_enter().
299 */
300 blk_freeze_queue_start(q);
301 if (queue_is_mq(q))
302 blk_mq_wake_waiters(q);
303 /* Make blk_queue_enter() reexamine the DYING flag. */
304 wake_up_all(&q->mq_freeze_wq);
305}
306
307/**
308 * blk_queue_enter() - try to increase q->q_usage_counter
309 * @q: request queue pointer
310 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PM
311 */
312int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
313{
314 const bool pm = flags & BLK_MQ_REQ_PM;
315
316 while (!blk_try_enter_queue(q, pm)) {
317 if (flags & BLK_MQ_REQ_NOWAIT)
318 return -EAGAIN;
319
320 /*
321 * read pair of barrier in blk_freeze_queue_start(), we need to
322 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
323 * reading .mq_freeze_depth or queue dying flag, otherwise the
324 * following wait may never return if the two reads are
325 * reordered.
326 */
327 smp_rmb();
328 wait_event(q->mq_freeze_wq,
329 (!q->mq_freeze_depth &&
330 blk_pm_resume_queue(pm, q)) ||
331 blk_queue_dying(q));
332 if (blk_queue_dying(q))
333 return -ENODEV;
334 }
335
336 return 0;
337}
338
339int __bio_queue_enter(struct request_queue *q, struct bio *bio)
340{
341 while (!blk_try_enter_queue(q, false)) {
342 struct gendisk *disk = bio->bi_bdev->bd_disk;
343
344 if (bio->bi_opf & REQ_NOWAIT) {
345 if (test_bit(GD_DEAD, &disk->state))
346 goto dead;
347 bio_wouldblock_error(bio);
348 return -EAGAIN;
349 }
350
351 /*
352 * read pair of barrier in blk_freeze_queue_start(), we need to
353 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
354 * reading .mq_freeze_depth or queue dying flag, otherwise the
355 * following wait may never return if the two reads are
356 * reordered.
357 */
358 smp_rmb();
359 wait_event(q->mq_freeze_wq,
360 (!q->mq_freeze_depth &&
361 blk_pm_resume_queue(false, q)) ||
362 test_bit(GD_DEAD, &disk->state));
363 if (test_bit(GD_DEAD, &disk->state))
364 goto dead;
365 }
366
367 return 0;
368dead:
369 bio_io_error(bio);
370 return -ENODEV;
371}
372
373void blk_queue_exit(struct request_queue *q)
374{
375 percpu_ref_put(&q->q_usage_counter);
376}
377
378static void blk_queue_usage_counter_release(struct percpu_ref *ref)
379{
380 struct request_queue *q =
381 container_of(ref, struct request_queue, q_usage_counter);
382
383 wake_up_all(&q->mq_freeze_wq);
384}
385
386static void blk_rq_timed_out_timer(struct timer_list *t)
387{
388 struct request_queue *q = from_timer(q, t, timeout);
389
390 kblockd_schedule_work(&q->timeout_work);
391}
392
393static void blk_timeout_work(struct work_struct *work)
394{
395}
396
397struct request_queue *blk_alloc_queue(struct queue_limits *lim, int node_id)
398{
399 struct request_queue *q;
400 int error;
401
402 q = kmem_cache_alloc_node(blk_requestq_cachep, GFP_KERNEL | __GFP_ZERO,
403 node_id);
404 if (!q)
405 return ERR_PTR(-ENOMEM);
406
407 q->last_merge = NULL;
408
409 q->id = ida_alloc(&blk_queue_ida, GFP_KERNEL);
410 if (q->id < 0) {
411 error = q->id;
412 goto fail_q;
413 }
414
415 q->stats = blk_alloc_queue_stats();
416 if (!q->stats) {
417 error = -ENOMEM;
418 goto fail_id;
419 }
420
421 error = blk_set_default_limits(lim);
422 if (error)
423 goto fail_stats;
424 q->limits = *lim;
425
426 q->node = node_id;
427
428 atomic_set(&q->nr_active_requests_shared_tags, 0);
429
430 timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
431 INIT_WORK(&q->timeout_work, blk_timeout_work);
432 INIT_LIST_HEAD(&q->icq_list);
433
434 refcount_set(&q->refs, 1);
435 mutex_init(&q->debugfs_mutex);
436 mutex_init(&q->sysfs_lock);
437 mutex_init(&q->sysfs_dir_lock);
438 mutex_init(&q->limits_lock);
439 mutex_init(&q->rq_qos_mutex);
440 spin_lock_init(&q->queue_lock);
441
442 init_waitqueue_head(&q->mq_freeze_wq);
443 mutex_init(&q->mq_freeze_lock);
444
445 blkg_init_queue(q);
446
447 /*
448 * Init percpu_ref in atomic mode so that it's faster to shutdown.
449 * See blk_register_queue() for details.
450 */
451 error = percpu_ref_init(&q->q_usage_counter,
452 blk_queue_usage_counter_release,
453 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL);
454 if (error)
455 goto fail_stats;
456
457 q->nr_requests = BLKDEV_DEFAULT_RQ;
458
459 return q;
460
461fail_stats:
462 blk_free_queue_stats(q->stats);
463fail_id:
464 ida_free(&blk_queue_ida, q->id);
465fail_q:
466 kmem_cache_free(blk_requestq_cachep, q);
467 return ERR_PTR(error);
468}
469
470/**
471 * blk_get_queue - increment the request_queue refcount
472 * @q: the request_queue structure to increment the refcount for
473 *
474 * Increment the refcount of the request_queue kobject.
475 *
476 * Context: Any context.
477 */
478bool blk_get_queue(struct request_queue *q)
479{
480 if (unlikely(blk_queue_dying(q)))
481 return false;
482 refcount_inc(&q->refs);
483 return true;
484}
485EXPORT_SYMBOL(blk_get_queue);
486
487#ifdef CONFIG_FAIL_MAKE_REQUEST
488
489static DECLARE_FAULT_ATTR(fail_make_request);
490
491static int __init setup_fail_make_request(char *str)
492{
493 return setup_fault_attr(&fail_make_request, str);
494}
495__setup("fail_make_request=", setup_fail_make_request);
496
497bool should_fail_request(struct block_device *part, unsigned int bytes)
498{
499 return part->bd_make_it_fail && should_fail(&fail_make_request, bytes);
500}
501
502static int __init fail_make_request_debugfs(void)
503{
504 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
505 NULL, &fail_make_request);
506
507 return PTR_ERR_OR_ZERO(dir);
508}
509
510late_initcall(fail_make_request_debugfs);
511#endif /* CONFIG_FAIL_MAKE_REQUEST */
512
513static inline void bio_check_ro(struct bio *bio)
514{
515 if (op_is_write(bio_op(bio)) && bdev_read_only(bio->bi_bdev)) {
516 if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
517 return;
518
519 if (bio->bi_bdev->bd_ro_warned)
520 return;
521
522 bio->bi_bdev->bd_ro_warned = true;
523 /*
524 * Use ioctl to set underlying disk of raid/dm to read-only
525 * will trigger this.
526 */
527 pr_warn("Trying to write to read-only block-device %pg\n",
528 bio->bi_bdev);
529 }
530}
531
532static noinline int should_fail_bio(struct bio *bio)
533{
534 if (should_fail_request(bdev_whole(bio->bi_bdev), bio->bi_iter.bi_size))
535 return -EIO;
536 return 0;
537}
538ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
539
540/*
541 * Check whether this bio extends beyond the end of the device or partition.
542 * This may well happen - the kernel calls bread() without checking the size of
543 * the device, e.g., when mounting a file system.
544 */
545static inline int bio_check_eod(struct bio *bio)
546{
547 sector_t maxsector = bdev_nr_sectors(bio->bi_bdev);
548 unsigned int nr_sectors = bio_sectors(bio);
549
550 if (nr_sectors &&
551 (nr_sectors > maxsector ||
552 bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
553 pr_info_ratelimited("%s: attempt to access beyond end of device\n"
554 "%pg: rw=%d, sector=%llu, nr_sectors = %u limit=%llu\n",
555 current->comm, bio->bi_bdev, bio->bi_opf,
556 bio->bi_iter.bi_sector, nr_sectors, maxsector);
557 return -EIO;
558 }
559 return 0;
560}
561
562/*
563 * Remap block n of partition p to block n+start(p) of the disk.
564 */
565static int blk_partition_remap(struct bio *bio)
566{
567 struct block_device *p = bio->bi_bdev;
568
569 if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
570 return -EIO;
571 if (bio_sectors(bio)) {
572 bio->bi_iter.bi_sector += p->bd_start_sect;
573 trace_block_bio_remap(bio, p->bd_dev,
574 bio->bi_iter.bi_sector -
575 p->bd_start_sect);
576 }
577 bio_set_flag(bio, BIO_REMAPPED);
578 return 0;
579}
580
581/*
582 * Check write append to a zoned block device.
583 */
584static inline blk_status_t blk_check_zone_append(struct request_queue *q,
585 struct bio *bio)
586{
587 int nr_sectors = bio_sectors(bio);
588
589 /* Only applicable to zoned block devices */
590 if (!bdev_is_zoned(bio->bi_bdev))
591 return BLK_STS_NOTSUPP;
592
593 /* The bio sector must point to the start of a sequential zone */
594 if (!bdev_is_zone_start(bio->bi_bdev, bio->bi_iter.bi_sector) ||
595 !bio_zone_is_seq(bio))
596 return BLK_STS_IOERR;
597
598 /*
599 * Not allowed to cross zone boundaries. Otherwise, the BIO will be
600 * split and could result in non-contiguous sectors being written in
601 * different zones.
602 */
603 if (nr_sectors > q->limits.chunk_sectors)
604 return BLK_STS_IOERR;
605
606 /* Make sure the BIO is small enough and will not get split */
607 if (nr_sectors > q->limits.max_zone_append_sectors)
608 return BLK_STS_IOERR;
609
610 bio->bi_opf |= REQ_NOMERGE;
611
612 return BLK_STS_OK;
613}
614
615static void __submit_bio(struct bio *bio)
616{
617 if (unlikely(!blk_crypto_bio_prep(&bio)))
618 return;
619
620 if (!bio->bi_bdev->bd_has_submit_bio) {
621 blk_mq_submit_bio(bio);
622 } else if (likely(bio_queue_enter(bio) == 0)) {
623 struct gendisk *disk = bio->bi_bdev->bd_disk;
624
625 disk->fops->submit_bio(bio);
626 blk_queue_exit(disk->queue);
627 }
628}
629
630/*
631 * The loop in this function may be a bit non-obvious, and so deserves some
632 * explanation:
633 *
634 * - Before entering the loop, bio->bi_next is NULL (as all callers ensure
635 * that), so we have a list with a single bio.
636 * - We pretend that we have just taken it off a longer list, so we assign
637 * bio_list to a pointer to the bio_list_on_stack, thus initialising the
638 * bio_list of new bios to be added. ->submit_bio() may indeed add some more
639 * bios through a recursive call to submit_bio_noacct. If it did, we find a
640 * non-NULL value in bio_list and re-enter the loop from the top.
641 * - In this case we really did just take the bio of the top of the list (no
642 * pretending) and so remove it from bio_list, and call into ->submit_bio()
643 * again.
644 *
645 * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio.
646 * bio_list_on_stack[1] contains bios that were submitted before the current
647 * ->submit_bio, but that haven't been processed yet.
648 */
649static void __submit_bio_noacct(struct bio *bio)
650{
651 struct bio_list bio_list_on_stack[2];
652
653 BUG_ON(bio->bi_next);
654
655 bio_list_init(&bio_list_on_stack[0]);
656 current->bio_list = bio_list_on_stack;
657
658 do {
659 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
660 struct bio_list lower, same;
661
662 /*
663 * Create a fresh bio_list for all subordinate requests.
664 */
665 bio_list_on_stack[1] = bio_list_on_stack[0];
666 bio_list_init(&bio_list_on_stack[0]);
667
668 __submit_bio(bio);
669
670 /*
671 * Sort new bios into those for a lower level and those for the
672 * same level.
673 */
674 bio_list_init(&lower);
675 bio_list_init(&same);
676 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
677 if (q == bdev_get_queue(bio->bi_bdev))
678 bio_list_add(&same, bio);
679 else
680 bio_list_add(&lower, bio);
681
682 /*
683 * Now assemble so we handle the lowest level first.
684 */
685 bio_list_merge(&bio_list_on_stack[0], &lower);
686 bio_list_merge(&bio_list_on_stack[0], &same);
687 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
688 } while ((bio = bio_list_pop(&bio_list_on_stack[0])));
689
690 current->bio_list = NULL;
691}
692
693static void __submit_bio_noacct_mq(struct bio *bio)
694{
695 struct bio_list bio_list[2] = { };
696
697 current->bio_list = bio_list;
698
699 do {
700 __submit_bio(bio);
701 } while ((bio = bio_list_pop(&bio_list[0])));
702
703 current->bio_list = NULL;
704}
705
706void submit_bio_noacct_nocheck(struct bio *bio)
707{
708 blk_cgroup_bio_start(bio);
709 blkcg_bio_issue_init(bio);
710
711 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
712 trace_block_bio_queue(bio);
713 /*
714 * Now that enqueuing has been traced, we need to trace
715 * completion as well.
716 */
717 bio_set_flag(bio, BIO_TRACE_COMPLETION);
718 }
719
720 /*
721 * We only want one ->submit_bio to be active at a time, else stack
722 * usage with stacked devices could be a problem. Use current->bio_list
723 * to collect a list of requests submited by a ->submit_bio method while
724 * it is active, and then process them after it returned.
725 */
726 if (current->bio_list)
727 bio_list_add(¤t->bio_list[0], bio);
728 else if (!bio->bi_bdev->bd_has_submit_bio)
729 __submit_bio_noacct_mq(bio);
730 else
731 __submit_bio_noacct(bio);
732}
733
734/**
735 * submit_bio_noacct - re-submit a bio to the block device layer for I/O
736 * @bio: The bio describing the location in memory and on the device.
737 *
738 * This is a version of submit_bio() that shall only be used for I/O that is
739 * resubmitted to lower level drivers by stacking block drivers. All file
740 * systems and other upper level users of the block layer should use
741 * submit_bio() instead.
742 */
743void submit_bio_noacct(struct bio *bio)
744{
745 struct block_device *bdev = bio->bi_bdev;
746 struct request_queue *q = bdev_get_queue(bdev);
747 blk_status_t status = BLK_STS_IOERR;
748
749 might_sleep();
750
751 /*
752 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
753 * if queue does not support NOWAIT.
754 */
755 if ((bio->bi_opf & REQ_NOWAIT) && !bdev_nowait(bdev))
756 goto not_supported;
757
758 if (should_fail_bio(bio))
759 goto end_io;
760 bio_check_ro(bio);
761 if (!bio_flagged(bio, BIO_REMAPPED)) {
762 if (unlikely(bio_check_eod(bio)))
763 goto end_io;
764 if (bdev->bd_partno && unlikely(blk_partition_remap(bio)))
765 goto end_io;
766 }
767
768 /*
769 * Filter flush bio's early so that bio based drivers without flush
770 * support don't have to worry about them.
771 */
772 if (op_is_flush(bio->bi_opf)) {
773 if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_WRITE &&
774 bio_op(bio) != REQ_OP_ZONE_APPEND))
775 goto end_io;
776 if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
777 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
778 if (!bio_sectors(bio)) {
779 status = BLK_STS_OK;
780 goto end_io;
781 }
782 }
783 }
784
785 if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
786 bio_clear_polled(bio);
787
788 switch (bio_op(bio)) {
789 case REQ_OP_READ:
790 case REQ_OP_WRITE:
791 break;
792 case REQ_OP_FLUSH:
793 /*
794 * REQ_OP_FLUSH can't be submitted through bios, it is only
795 * synthetized in struct request by the flush state machine.
796 */
797 goto not_supported;
798 case REQ_OP_DISCARD:
799 if (!bdev_max_discard_sectors(bdev))
800 goto not_supported;
801 break;
802 case REQ_OP_SECURE_ERASE:
803 if (!bdev_max_secure_erase_sectors(bdev))
804 goto not_supported;
805 break;
806 case REQ_OP_ZONE_APPEND:
807 status = blk_check_zone_append(q, bio);
808 if (status != BLK_STS_OK)
809 goto end_io;
810 break;
811 case REQ_OP_WRITE_ZEROES:
812 if (!q->limits.max_write_zeroes_sectors)
813 goto not_supported;
814 break;
815 case REQ_OP_ZONE_RESET:
816 case REQ_OP_ZONE_OPEN:
817 case REQ_OP_ZONE_CLOSE:
818 case REQ_OP_ZONE_FINISH:
819 if (!bdev_is_zoned(bio->bi_bdev))
820 goto not_supported;
821 break;
822 case REQ_OP_ZONE_RESET_ALL:
823 if (!bdev_is_zoned(bio->bi_bdev) || !blk_queue_zone_resetall(q))
824 goto not_supported;
825 break;
826 case REQ_OP_DRV_IN:
827 case REQ_OP_DRV_OUT:
828 /*
829 * Driver private operations are only used with passthrough
830 * requests.
831 */
832 fallthrough;
833 default:
834 goto not_supported;
835 }
836
837 if (blk_throtl_bio(bio))
838 return;
839 submit_bio_noacct_nocheck(bio);
840 return;
841
842not_supported:
843 status = BLK_STS_NOTSUPP;
844end_io:
845 bio->bi_status = status;
846 bio_endio(bio);
847}
848EXPORT_SYMBOL(submit_bio_noacct);
849
850static void bio_set_ioprio(struct bio *bio)
851{
852 /* Nobody set ioprio so far? Initialize it based on task's nice value */
853 if (IOPRIO_PRIO_CLASS(bio->bi_ioprio) == IOPRIO_CLASS_NONE)
854 bio->bi_ioprio = get_current_ioprio();
855 blkcg_set_ioprio(bio);
856}
857
858/**
859 * submit_bio - submit a bio to the block device layer for I/O
860 * @bio: The &struct bio which describes the I/O
861 *
862 * submit_bio() is used to submit I/O requests to block devices. It is passed a
863 * fully set up &struct bio that describes the I/O that needs to be done. The
864 * bio will be send to the device described by the bi_bdev field.
865 *
866 * The success/failure status of the request, along with notification of
867 * completion, is delivered asynchronously through the ->bi_end_io() callback
868 * in @bio. The bio must NOT be touched by the caller until ->bi_end_io() has
869 * been called.
870 */
871void submit_bio(struct bio *bio)
872{
873 if (bio_op(bio) == REQ_OP_READ) {
874 task_io_account_read(bio->bi_iter.bi_size);
875 count_vm_events(PGPGIN, bio_sectors(bio));
876 } else if (bio_op(bio) == REQ_OP_WRITE) {
877 count_vm_events(PGPGOUT, bio_sectors(bio));
878 }
879
880 bio_set_ioprio(bio);
881 submit_bio_noacct(bio);
882}
883EXPORT_SYMBOL(submit_bio);
884
885/**
886 * bio_poll - poll for BIO completions
887 * @bio: bio to poll for
888 * @iob: batches of IO
889 * @flags: BLK_POLL_* flags that control the behavior
890 *
891 * Poll for completions on queue associated with the bio. Returns number of
892 * completed entries found.
893 *
894 * Note: the caller must either be the context that submitted @bio, or
895 * be in a RCU critical section to prevent freeing of @bio.
896 */
897int bio_poll(struct bio *bio, struct io_comp_batch *iob, unsigned int flags)
898{
899 blk_qc_t cookie = READ_ONCE(bio->bi_cookie);
900 struct block_device *bdev;
901 struct request_queue *q;
902 int ret = 0;
903
904 bdev = READ_ONCE(bio->bi_bdev);
905 if (!bdev)
906 return 0;
907
908 q = bdev_get_queue(bdev);
909 if (cookie == BLK_QC_T_NONE ||
910 !test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
911 return 0;
912
913 /*
914 * As the requests that require a zone lock are not plugged in the
915 * first place, directly accessing the plug instead of using
916 * blk_mq_plug() should not have any consequences during flushing for
917 * zoned devices.
918 */
919 blk_flush_plug(current->plug, false);
920
921 /*
922 * We need to be able to enter a frozen queue, similar to how
923 * timeouts also need to do that. If that is blocked, then we can
924 * have pending IO when a queue freeze is started, and then the
925 * wait for the freeze to finish will wait for polled requests to
926 * timeout as the poller is preventer from entering the queue and
927 * completing them. As long as we prevent new IO from being queued,
928 * that should be all that matters.
929 */
930 if (!percpu_ref_tryget(&q->q_usage_counter))
931 return 0;
932 if (queue_is_mq(q)) {
933 ret = blk_mq_poll(q, cookie, iob, flags);
934 } else {
935 struct gendisk *disk = q->disk;
936
937 if (disk && disk->fops->poll_bio)
938 ret = disk->fops->poll_bio(bio, iob, flags);
939 }
940 blk_queue_exit(q);
941 return ret;
942}
943EXPORT_SYMBOL_GPL(bio_poll);
944
945/*
946 * Helper to implement file_operations.iopoll. Requires the bio to be stored
947 * in iocb->private, and cleared before freeing the bio.
948 */
949int iocb_bio_iopoll(struct kiocb *kiocb, struct io_comp_batch *iob,
950 unsigned int flags)
951{
952 struct bio *bio;
953 int ret = 0;
954
955 /*
956 * Note: the bio cache only uses SLAB_TYPESAFE_BY_RCU, so bio can
957 * point to a freshly allocated bio at this point. If that happens
958 * we have a few cases to consider:
959 *
960 * 1) the bio is beeing initialized and bi_bdev is NULL. We can just
961 * simply nothing in this case
962 * 2) the bio points to a not poll enabled device. bio_poll will catch
963 * this and return 0
964 * 3) the bio points to a poll capable device, including but not
965 * limited to the one that the original bio pointed to. In this
966 * case we will call into the actual poll method and poll for I/O,
967 * even if we don't need to, but it won't cause harm either.
968 *
969 * For cases 2) and 3) above the RCU grace period ensures that bi_bdev
970 * is still allocated. Because partitions hold a reference to the whole
971 * device bdev and thus disk, the disk is also still valid. Grabbing
972 * a reference to the queue in bio_poll() ensures the hctxs and requests
973 * are still valid as well.
974 */
975 rcu_read_lock();
976 bio = READ_ONCE(kiocb->private);
977 if (bio)
978 ret = bio_poll(bio, iob, flags);
979 rcu_read_unlock();
980
981 return ret;
982}
983EXPORT_SYMBOL_GPL(iocb_bio_iopoll);
984
985void update_io_ticks(struct block_device *part, unsigned long now, bool end)
986{
987 unsigned long stamp;
988again:
989 stamp = READ_ONCE(part->bd_stamp);
990 if (unlikely(time_after(now, stamp)) &&
991 likely(try_cmpxchg(&part->bd_stamp, &stamp, now)) &&
992 (end || part_in_flight(part)))
993 __part_stat_add(part, io_ticks, now - stamp);
994
995 if (part->bd_partno) {
996 part = bdev_whole(part);
997 goto again;
998 }
999}
1000
1001unsigned long bdev_start_io_acct(struct block_device *bdev, enum req_op op,
1002 unsigned long start_time)
1003{
1004 part_stat_lock();
1005 update_io_ticks(bdev, start_time, false);
1006 part_stat_local_inc(bdev, in_flight[op_is_write(op)]);
1007 part_stat_unlock();
1008
1009 return start_time;
1010}
1011EXPORT_SYMBOL(bdev_start_io_acct);
1012
1013/**
1014 * bio_start_io_acct - start I/O accounting for bio based drivers
1015 * @bio: bio to start account for
1016 *
1017 * Returns the start time that should be passed back to bio_end_io_acct().
1018 */
1019unsigned long bio_start_io_acct(struct bio *bio)
1020{
1021 return bdev_start_io_acct(bio->bi_bdev, bio_op(bio), jiffies);
1022}
1023EXPORT_SYMBOL_GPL(bio_start_io_acct);
1024
1025void bdev_end_io_acct(struct block_device *bdev, enum req_op op,
1026 unsigned int sectors, unsigned long start_time)
1027{
1028 const int sgrp = op_stat_group(op);
1029 unsigned long now = READ_ONCE(jiffies);
1030 unsigned long duration = now - start_time;
1031
1032 part_stat_lock();
1033 update_io_ticks(bdev, now, true);
1034 part_stat_inc(bdev, ios[sgrp]);
1035 part_stat_add(bdev, sectors[sgrp], sectors);
1036 part_stat_add(bdev, nsecs[sgrp], jiffies_to_nsecs(duration));
1037 part_stat_local_dec(bdev, in_flight[op_is_write(op)]);
1038 part_stat_unlock();
1039}
1040EXPORT_SYMBOL(bdev_end_io_acct);
1041
1042void bio_end_io_acct_remapped(struct bio *bio, unsigned long start_time,
1043 struct block_device *orig_bdev)
1044{
1045 bdev_end_io_acct(orig_bdev, bio_op(bio), bio_sectors(bio), start_time);
1046}
1047EXPORT_SYMBOL_GPL(bio_end_io_acct_remapped);
1048
1049/**
1050 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1051 * @q : the queue of the device being checked
1052 *
1053 * Description:
1054 * Check if underlying low-level drivers of a device are busy.
1055 * If the drivers want to export their busy state, they must set own
1056 * exporting function using blk_queue_lld_busy() first.
1057 *
1058 * Basically, this function is used only by request stacking drivers
1059 * to stop dispatching requests to underlying devices when underlying
1060 * devices are busy. This behavior helps more I/O merging on the queue
1061 * of the request stacking driver and prevents I/O throughput regression
1062 * on burst I/O load.
1063 *
1064 * Return:
1065 * 0 - Not busy (The request stacking driver should dispatch request)
1066 * 1 - Busy (The request stacking driver should stop dispatching request)
1067 */
1068int blk_lld_busy(struct request_queue *q)
1069{
1070 if (queue_is_mq(q) && q->mq_ops->busy)
1071 return q->mq_ops->busy(q);
1072
1073 return 0;
1074}
1075EXPORT_SYMBOL_GPL(blk_lld_busy);
1076
1077int kblockd_schedule_work(struct work_struct *work)
1078{
1079 return queue_work(kblockd_workqueue, work);
1080}
1081EXPORT_SYMBOL(kblockd_schedule_work);
1082
1083int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
1084 unsigned long delay)
1085{
1086 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
1087}
1088EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
1089
1090void blk_start_plug_nr_ios(struct blk_plug *plug, unsigned short nr_ios)
1091{
1092 struct task_struct *tsk = current;
1093
1094 /*
1095 * If this is a nested plug, don't actually assign it.
1096 */
1097 if (tsk->plug)
1098 return;
1099
1100 plug->cur_ktime = 0;
1101 plug->mq_list = NULL;
1102 plug->cached_rq = NULL;
1103 plug->nr_ios = min_t(unsigned short, nr_ios, BLK_MAX_REQUEST_COUNT);
1104 plug->rq_count = 0;
1105 plug->multiple_queues = false;
1106 plug->has_elevator = false;
1107 INIT_LIST_HEAD(&plug->cb_list);
1108
1109 /*
1110 * Store ordering should not be needed here, since a potential
1111 * preempt will imply a full memory barrier
1112 */
1113 tsk->plug = plug;
1114}
1115
1116/**
1117 * blk_start_plug - initialize blk_plug and track it inside the task_struct
1118 * @plug: The &struct blk_plug that needs to be initialized
1119 *
1120 * Description:
1121 * blk_start_plug() indicates to the block layer an intent by the caller
1122 * to submit multiple I/O requests in a batch. The block layer may use
1123 * this hint to defer submitting I/Os from the caller until blk_finish_plug()
1124 * is called. However, the block layer may choose to submit requests
1125 * before a call to blk_finish_plug() if the number of queued I/Os
1126 * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1127 * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if
1128 * the task schedules (see below).
1129 *
1130 * Tracking blk_plug inside the task_struct will help with auto-flushing the
1131 * pending I/O should the task end up blocking between blk_start_plug() and
1132 * blk_finish_plug(). This is important from a performance perspective, but
1133 * also ensures that we don't deadlock. For instance, if the task is blocking
1134 * for a memory allocation, memory reclaim could end up wanting to free a
1135 * page belonging to that request that is currently residing in our private
1136 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
1137 * this kind of deadlock.
1138 */
1139void blk_start_plug(struct blk_plug *plug)
1140{
1141 blk_start_plug_nr_ios(plug, 1);
1142}
1143EXPORT_SYMBOL(blk_start_plug);
1144
1145static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
1146{
1147 LIST_HEAD(callbacks);
1148
1149 while (!list_empty(&plug->cb_list)) {
1150 list_splice_init(&plug->cb_list, &callbacks);
1151
1152 while (!list_empty(&callbacks)) {
1153 struct blk_plug_cb *cb = list_first_entry(&callbacks,
1154 struct blk_plug_cb,
1155 list);
1156 list_del(&cb->list);
1157 cb->callback(cb, from_schedule);
1158 }
1159 }
1160}
1161
1162struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
1163 int size)
1164{
1165 struct blk_plug *plug = current->plug;
1166 struct blk_plug_cb *cb;
1167
1168 if (!plug)
1169 return NULL;
1170
1171 list_for_each_entry(cb, &plug->cb_list, list)
1172 if (cb->callback == unplug && cb->data == data)
1173 return cb;
1174
1175 /* Not currently on the callback list */
1176 BUG_ON(size < sizeof(*cb));
1177 cb = kzalloc(size, GFP_ATOMIC);
1178 if (cb) {
1179 cb->data = data;
1180 cb->callback = unplug;
1181 list_add(&cb->list, &plug->cb_list);
1182 }
1183 return cb;
1184}
1185EXPORT_SYMBOL(blk_check_plugged);
1186
1187void __blk_flush_plug(struct blk_plug *plug, bool from_schedule)
1188{
1189 if (!list_empty(&plug->cb_list))
1190 flush_plug_callbacks(plug, from_schedule);
1191 blk_mq_flush_plug_list(plug, from_schedule);
1192 /*
1193 * Unconditionally flush out cached requests, even if the unplug
1194 * event came from schedule. Since we know hold references to the
1195 * queue for cached requests, we don't want a blocked task holding
1196 * up a queue freeze/quiesce event.
1197 */
1198 if (unlikely(!rq_list_empty(plug->cached_rq)))
1199 blk_mq_free_plug_rqs(plug);
1200
1201 plug->cur_ktime = 0;
1202 current->flags &= ~PF_BLOCK_TS;
1203}
1204
1205/**
1206 * blk_finish_plug - mark the end of a batch of submitted I/O
1207 * @plug: The &struct blk_plug passed to blk_start_plug()
1208 *
1209 * Description:
1210 * Indicate that a batch of I/O submissions is complete. This function
1211 * must be paired with an initial call to blk_start_plug(). The intent
1212 * is to allow the block layer to optimize I/O submission. See the
1213 * documentation for blk_start_plug() for more information.
1214 */
1215void blk_finish_plug(struct blk_plug *plug)
1216{
1217 if (plug == current->plug) {
1218 __blk_flush_plug(plug, false);
1219 current->plug = NULL;
1220 }
1221}
1222EXPORT_SYMBOL(blk_finish_plug);
1223
1224void blk_io_schedule(void)
1225{
1226 /* Prevent hang_check timer from firing at us during very long I/O */
1227 unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2;
1228
1229 if (timeout)
1230 io_schedule_timeout(timeout);
1231 else
1232 io_schedule();
1233}
1234EXPORT_SYMBOL_GPL(blk_io_schedule);
1235
1236int __init blk_dev_init(void)
1237{
1238 BUILD_BUG_ON((__force u32)REQ_OP_LAST >= (1 << REQ_OP_BITS));
1239 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1240 sizeof_field(struct request, cmd_flags));
1241 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1242 sizeof_field(struct bio, bi_opf));
1243
1244 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
1245 kblockd_workqueue = alloc_workqueue("kblockd",
1246 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1247 if (!kblockd_workqueue)
1248 panic("Failed to create kblockd\n");
1249
1250 blk_requestq_cachep = KMEM_CACHE(request_queue, SLAB_PANIC);
1251
1252 blk_debugfs_root = debugfs_create_dir("block", NULL);
1253
1254 return 0;
1255}