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
37#define CREATE_TRACE_POINTS
38#include <trace/events/block.h>
39
40#include "blk.h"
41#include "blk-mq.h"
42
43EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
44EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
45EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
46EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
47EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
48
49DEFINE_IDA(blk_queue_ida);
50
51/*
52 * For the allocated request tables
53 */
54struct kmem_cache *request_cachep;
55
56/*
57 * For queue allocation
58 */
59struct kmem_cache *blk_requestq_cachep;
60
61/*
62 * Controlling structure to kblockd
63 */
64static struct workqueue_struct *kblockd_workqueue;
65
66static void blk_clear_congested(struct request_list *rl, int sync)
67{
68#ifdef CONFIG_CGROUP_WRITEBACK
69 clear_wb_congested(rl->blkg->wb_congested, sync);
70#else
71 /*
72 * If !CGROUP_WRITEBACK, all blkg's map to bdi->wb and we shouldn't
73 * flip its congestion state for events on other blkcgs.
74 */
75 if (rl == &rl->q->root_rl)
76 clear_wb_congested(rl->q->backing_dev_info.wb.congested, sync);
77#endif
78}
79
80static void blk_set_congested(struct request_list *rl, int sync)
81{
82#ifdef CONFIG_CGROUP_WRITEBACK
83 set_wb_congested(rl->blkg->wb_congested, sync);
84#else
85 /* see blk_clear_congested() */
86 if (rl == &rl->q->root_rl)
87 set_wb_congested(rl->q->backing_dev_info.wb.congested, sync);
88#endif
89}
90
91void blk_queue_congestion_threshold(struct request_queue *q)
92{
93 int nr;
94
95 nr = q->nr_requests - (q->nr_requests / 8) + 1;
96 if (nr > q->nr_requests)
97 nr = q->nr_requests;
98 q->nr_congestion_on = nr;
99
100 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
101 if (nr < 1)
102 nr = 1;
103 q->nr_congestion_off = nr;
104}
105
106/**
107 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
108 * @bdev: device
109 *
110 * Locates the passed device's request queue and returns the address of its
111 * backing_dev_info. This function can only be called if @bdev is opened
112 * and the return value is never NULL.
113 */
114struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
115{
116 struct request_queue *q = bdev_get_queue(bdev);
117
118 return &q->backing_dev_info;
119}
120EXPORT_SYMBOL(blk_get_backing_dev_info);
121
122void blk_rq_init(struct request_queue *q, struct request *rq)
123{
124 memset(rq, 0, sizeof(*rq));
125
126 INIT_LIST_HEAD(&rq->queuelist);
127 INIT_LIST_HEAD(&rq->timeout_list);
128 rq->cpu = -1;
129 rq->q = q;
130 rq->__sector = (sector_t) -1;
131 INIT_HLIST_NODE(&rq->hash);
132 RB_CLEAR_NODE(&rq->rb_node);
133 rq->cmd = rq->__cmd;
134 rq->cmd_len = BLK_MAX_CDB;
135 rq->tag = -1;
136 rq->start_time = jiffies;
137 set_start_time_ns(rq);
138 rq->part = NULL;
139}
140EXPORT_SYMBOL(blk_rq_init);
141
142static void req_bio_endio(struct request *rq, struct bio *bio,
143 unsigned int nbytes, int error)
144{
145 if (error)
146 bio->bi_error = error;
147
148 if (unlikely(rq->cmd_flags & REQ_QUIET))
149 bio_set_flag(bio, BIO_QUIET);
150
151 bio_advance(bio, nbytes);
152
153 /* don't actually finish bio if it's part of flush sequence */
154 if (bio->bi_iter.bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
155 bio_endio(bio);
156}
157
158void blk_dump_rq_flags(struct request *rq, char *msg)
159{
160 int bit;
161
162 printk(KERN_INFO "%s: dev %s: type=%x, flags=%llx\n", msg,
163 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
164 (unsigned long long) rq->cmd_flags);
165
166 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
167 (unsigned long long)blk_rq_pos(rq),
168 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
169 printk(KERN_INFO " bio %p, biotail %p, len %u\n",
170 rq->bio, rq->biotail, blk_rq_bytes(rq));
171
172 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
173 printk(KERN_INFO " cdb: ");
174 for (bit = 0; bit < BLK_MAX_CDB; bit++)
175 printk("%02x ", rq->cmd[bit]);
176 printk("\n");
177 }
178}
179EXPORT_SYMBOL(blk_dump_rq_flags);
180
181static void blk_delay_work(struct work_struct *work)
182{
183 struct request_queue *q;
184
185 q = container_of(work, struct request_queue, delay_work.work);
186 spin_lock_irq(q->queue_lock);
187 __blk_run_queue(q);
188 spin_unlock_irq(q->queue_lock);
189}
190
191/**
192 * blk_delay_queue - restart queueing after defined interval
193 * @q: The &struct request_queue in question
194 * @msecs: Delay in msecs
195 *
196 * Description:
197 * Sometimes queueing needs to be postponed for a little while, to allow
198 * resources to come back. This function will make sure that queueing is
199 * restarted around the specified time. Queue lock must be held.
200 */
201void blk_delay_queue(struct request_queue *q, unsigned long msecs)
202{
203 if (likely(!blk_queue_dead(q)))
204 queue_delayed_work(kblockd_workqueue, &q->delay_work,
205 msecs_to_jiffies(msecs));
206}
207EXPORT_SYMBOL(blk_delay_queue);
208
209/**
210 * blk_start_queue_async - asynchronously restart a previously stopped queue
211 * @q: The &struct request_queue in question
212 *
213 * Description:
214 * blk_start_queue_async() will clear the stop flag on the queue, and
215 * ensure that the request_fn for the queue is run from an async
216 * context.
217 **/
218void blk_start_queue_async(struct request_queue *q)
219{
220 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
221 blk_run_queue_async(q);
222}
223EXPORT_SYMBOL(blk_start_queue_async);
224
225/**
226 * blk_start_queue - restart a previously stopped queue
227 * @q: The &struct request_queue in question
228 *
229 * Description:
230 * blk_start_queue() will clear the stop flag on the queue, and call
231 * the request_fn for the queue if it was in a stopped state when
232 * entered. Also see blk_stop_queue(). Queue lock must be held.
233 **/
234void blk_start_queue(struct request_queue *q)
235{
236 WARN_ON(!irqs_disabled());
237
238 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
239 __blk_run_queue(q);
240}
241EXPORT_SYMBOL(blk_start_queue);
242
243/**
244 * blk_stop_queue - stop a queue
245 * @q: The &struct request_queue in question
246 *
247 * Description:
248 * The Linux block layer assumes that a block driver will consume all
249 * entries on the request queue when the request_fn strategy is called.
250 * Often this will not happen, because of hardware limitations (queue
251 * depth settings). If a device driver gets a 'queue full' response,
252 * or if it simply chooses not to queue more I/O at one point, it can
253 * call this function to prevent the request_fn from being called until
254 * the driver has signalled it's ready to go again. This happens by calling
255 * blk_start_queue() to restart queue operations. Queue lock must be held.
256 **/
257void blk_stop_queue(struct request_queue *q)
258{
259 cancel_delayed_work(&q->delay_work);
260 queue_flag_set(QUEUE_FLAG_STOPPED, q);
261}
262EXPORT_SYMBOL(blk_stop_queue);
263
264/**
265 * blk_sync_queue - cancel any pending callbacks on a queue
266 * @q: the queue
267 *
268 * Description:
269 * The block layer may perform asynchronous callback activity
270 * on a queue, such as calling the unplug function after a timeout.
271 * A block device may call blk_sync_queue to ensure that any
272 * such activity is cancelled, thus allowing it to release resources
273 * that the callbacks might use. The caller must already have made sure
274 * that its ->make_request_fn will not re-add plugging prior to calling
275 * this function.
276 *
277 * This function does not cancel any asynchronous activity arising
278 * out of elevator or throttling code. That would require elevator_exit()
279 * and blkcg_exit_queue() to be called with queue lock initialized.
280 *
281 */
282void blk_sync_queue(struct request_queue *q)
283{
284 del_timer_sync(&q->timeout);
285
286 if (q->mq_ops) {
287 struct blk_mq_hw_ctx *hctx;
288 int i;
289
290 queue_for_each_hw_ctx(q, hctx, i) {
291 cancel_delayed_work_sync(&hctx->run_work);
292 cancel_delayed_work_sync(&hctx->delay_work);
293 }
294 } else {
295 cancel_delayed_work_sync(&q->delay_work);
296 }
297}
298EXPORT_SYMBOL(blk_sync_queue);
299
300/**
301 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
302 * @q: The queue to run
303 *
304 * Description:
305 * Invoke request handling on a queue if there are any pending requests.
306 * May be used to restart request handling after a request has completed.
307 * This variant runs the queue whether or not the queue has been
308 * stopped. Must be called with the queue lock held and interrupts
309 * disabled. See also @blk_run_queue.
310 */
311inline void __blk_run_queue_uncond(struct request_queue *q)
312{
313 if (unlikely(blk_queue_dead(q)))
314 return;
315
316 /*
317 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
318 * the queue lock internally. As a result multiple threads may be
319 * running such a request function concurrently. Keep track of the
320 * number of active request_fn invocations such that blk_drain_queue()
321 * can wait until all these request_fn calls have finished.
322 */
323 q->request_fn_active++;
324 q->request_fn(q);
325 q->request_fn_active--;
326}
327EXPORT_SYMBOL_GPL(__blk_run_queue_uncond);
328
329/**
330 * __blk_run_queue - run a single device queue
331 * @q: The queue to run
332 *
333 * Description:
334 * See @blk_run_queue. This variant must be called with the queue lock
335 * held and interrupts disabled.
336 */
337void __blk_run_queue(struct request_queue *q)
338{
339 if (unlikely(blk_queue_stopped(q)))
340 return;
341
342 __blk_run_queue_uncond(q);
343}
344EXPORT_SYMBOL(__blk_run_queue);
345
346/**
347 * blk_run_queue_async - run a single device queue in workqueue context
348 * @q: The queue to run
349 *
350 * Description:
351 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
352 * of us. The caller must hold the queue lock.
353 */
354void blk_run_queue_async(struct request_queue *q)
355{
356 if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
357 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
358}
359EXPORT_SYMBOL(blk_run_queue_async);
360
361/**
362 * blk_run_queue - run a single device queue
363 * @q: The queue to run
364 *
365 * Description:
366 * Invoke request handling on this queue, if it has pending work to do.
367 * May be used to restart queueing when a request has completed.
368 */
369void blk_run_queue(struct request_queue *q)
370{
371 unsigned long flags;
372
373 spin_lock_irqsave(q->queue_lock, flags);
374 __blk_run_queue(q);
375 spin_unlock_irqrestore(q->queue_lock, flags);
376}
377EXPORT_SYMBOL(blk_run_queue);
378
379void blk_put_queue(struct request_queue *q)
380{
381 kobject_put(&q->kobj);
382}
383EXPORT_SYMBOL(blk_put_queue);
384
385/**
386 * __blk_drain_queue - drain requests from request_queue
387 * @q: queue to drain
388 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
389 *
390 * Drain requests from @q. If @drain_all is set, all requests are drained.
391 * If not, only ELVPRIV requests are drained. The caller is responsible
392 * for ensuring that no new requests which need to be drained are queued.
393 */
394static void __blk_drain_queue(struct request_queue *q, bool drain_all)
395 __releases(q->queue_lock)
396 __acquires(q->queue_lock)
397{
398 int i;
399
400 lockdep_assert_held(q->queue_lock);
401
402 while (true) {
403 bool drain = false;
404
405 /*
406 * The caller might be trying to drain @q before its
407 * elevator is initialized.
408 */
409 if (q->elevator)
410 elv_drain_elevator(q);
411
412 blkcg_drain_queue(q);
413
414 /*
415 * This function might be called on a queue which failed
416 * driver init after queue creation or is not yet fully
417 * active yet. Some drivers (e.g. fd and loop) get unhappy
418 * in such cases. Kick queue iff dispatch queue has
419 * something on it and @q has request_fn set.
420 */
421 if (!list_empty(&q->queue_head) && q->request_fn)
422 __blk_run_queue(q);
423
424 drain |= q->nr_rqs_elvpriv;
425 drain |= q->request_fn_active;
426
427 /*
428 * Unfortunately, requests are queued at and tracked from
429 * multiple places and there's no single counter which can
430 * be drained. Check all the queues and counters.
431 */
432 if (drain_all) {
433 struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
434 drain |= !list_empty(&q->queue_head);
435 for (i = 0; i < 2; i++) {
436 drain |= q->nr_rqs[i];
437 drain |= q->in_flight[i];
438 if (fq)
439 drain |= !list_empty(&fq->flush_queue[i]);
440 }
441 }
442
443 if (!drain)
444 break;
445
446 spin_unlock_irq(q->queue_lock);
447
448 msleep(10);
449
450 spin_lock_irq(q->queue_lock);
451 }
452
453 /*
454 * With queue marked dead, any woken up waiter will fail the
455 * allocation path, so the wakeup chaining is lost and we're
456 * left with hung waiters. We need to wake up those waiters.
457 */
458 if (q->request_fn) {
459 struct request_list *rl;
460
461 blk_queue_for_each_rl(rl, q)
462 for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
463 wake_up_all(&rl->wait[i]);
464 }
465}
466
467/**
468 * blk_queue_bypass_start - enter queue bypass mode
469 * @q: queue of interest
470 *
471 * In bypass mode, only the dispatch FIFO queue of @q is used. This
472 * function makes @q enter bypass mode and drains all requests which were
473 * throttled or issued before. On return, it's guaranteed that no request
474 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
475 * inside queue or RCU read lock.
476 */
477void blk_queue_bypass_start(struct request_queue *q)
478{
479 spin_lock_irq(q->queue_lock);
480 q->bypass_depth++;
481 queue_flag_set(QUEUE_FLAG_BYPASS, q);
482 spin_unlock_irq(q->queue_lock);
483
484 /*
485 * Queues start drained. Skip actual draining till init is
486 * complete. This avoids lenghty delays during queue init which
487 * can happen many times during boot.
488 */
489 if (blk_queue_init_done(q)) {
490 spin_lock_irq(q->queue_lock);
491 __blk_drain_queue(q, false);
492 spin_unlock_irq(q->queue_lock);
493
494 /* ensure blk_queue_bypass() is %true inside RCU read lock */
495 synchronize_rcu();
496 }
497}
498EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
499
500/**
501 * blk_queue_bypass_end - leave queue bypass mode
502 * @q: queue of interest
503 *
504 * Leave bypass mode and restore the normal queueing behavior.
505 */
506void blk_queue_bypass_end(struct request_queue *q)
507{
508 spin_lock_irq(q->queue_lock);
509 if (!--q->bypass_depth)
510 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
511 WARN_ON_ONCE(q->bypass_depth < 0);
512 spin_unlock_irq(q->queue_lock);
513}
514EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
515
516void blk_set_queue_dying(struct request_queue *q)
517{
518 queue_flag_set_unlocked(QUEUE_FLAG_DYING, q);
519
520 if (q->mq_ops)
521 blk_mq_wake_waiters(q);
522 else {
523 struct request_list *rl;
524
525 blk_queue_for_each_rl(rl, q) {
526 if (rl->rq_pool) {
527 wake_up(&rl->wait[BLK_RW_SYNC]);
528 wake_up(&rl->wait[BLK_RW_ASYNC]);
529 }
530 }
531 }
532}
533EXPORT_SYMBOL_GPL(blk_set_queue_dying);
534
535/**
536 * blk_cleanup_queue - shutdown a request queue
537 * @q: request queue to shutdown
538 *
539 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
540 * put it. All future requests will be failed immediately with -ENODEV.
541 */
542void blk_cleanup_queue(struct request_queue *q)
543{
544 spinlock_t *lock = q->queue_lock;
545
546 /* mark @q DYING, no new request or merges will be allowed afterwards */
547 mutex_lock(&q->sysfs_lock);
548 blk_set_queue_dying(q);
549 spin_lock_irq(lock);
550
551 /*
552 * A dying queue is permanently in bypass mode till released. Note
553 * that, unlike blk_queue_bypass_start(), we aren't performing
554 * synchronize_rcu() after entering bypass mode to avoid the delay
555 * as some drivers create and destroy a lot of queues while
556 * probing. This is still safe because blk_release_queue() will be
557 * called only after the queue refcnt drops to zero and nothing,
558 * RCU or not, would be traversing the queue by then.
559 */
560 q->bypass_depth++;
561 queue_flag_set(QUEUE_FLAG_BYPASS, q);
562
563 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
564 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
565 queue_flag_set(QUEUE_FLAG_DYING, q);
566 spin_unlock_irq(lock);
567 mutex_unlock(&q->sysfs_lock);
568
569 /*
570 * Drain all requests queued before DYING marking. Set DEAD flag to
571 * prevent that q->request_fn() gets invoked after draining finished.
572 */
573 blk_freeze_queue(q);
574 spin_lock_irq(lock);
575 if (!q->mq_ops)
576 __blk_drain_queue(q, true);
577 queue_flag_set(QUEUE_FLAG_DEAD, q);
578 spin_unlock_irq(lock);
579
580 /* for synchronous bio-based driver finish in-flight integrity i/o */
581 blk_flush_integrity();
582
583 /* @q won't process any more request, flush async actions */
584 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
585 blk_sync_queue(q);
586
587 if (q->mq_ops)
588 blk_mq_free_queue(q);
589 percpu_ref_exit(&q->q_usage_counter);
590
591 spin_lock_irq(lock);
592 if (q->queue_lock != &q->__queue_lock)
593 q->queue_lock = &q->__queue_lock;
594 spin_unlock_irq(lock);
595
596 bdi_unregister(&q->backing_dev_info);
597
598 /* @q is and will stay empty, shutdown and put */
599 blk_put_queue(q);
600}
601EXPORT_SYMBOL(blk_cleanup_queue);
602
603/* Allocate memory local to the request queue */
604static void *alloc_request_struct(gfp_t gfp_mask, void *data)
605{
606 int nid = (int)(long)data;
607 return kmem_cache_alloc_node(request_cachep, gfp_mask, nid);
608}
609
610static void free_request_struct(void *element, void *unused)
611{
612 kmem_cache_free(request_cachep, element);
613}
614
615int blk_init_rl(struct request_list *rl, struct request_queue *q,
616 gfp_t gfp_mask)
617{
618 if (unlikely(rl->rq_pool))
619 return 0;
620
621 rl->q = q;
622 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
623 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
624 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
625 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
626
627 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, alloc_request_struct,
628 free_request_struct,
629 (void *)(long)q->node, gfp_mask,
630 q->node);
631 if (!rl->rq_pool)
632 return -ENOMEM;
633
634 return 0;
635}
636
637void blk_exit_rl(struct request_list *rl)
638{
639 if (rl->rq_pool)
640 mempool_destroy(rl->rq_pool);
641}
642
643struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
644{
645 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
646}
647EXPORT_SYMBOL(blk_alloc_queue);
648
649int blk_queue_enter(struct request_queue *q, bool nowait)
650{
651 while (true) {
652 int ret;
653
654 if (percpu_ref_tryget_live(&q->q_usage_counter))
655 return 0;
656
657 if (nowait)
658 return -EBUSY;
659
660 ret = wait_event_interruptible(q->mq_freeze_wq,
661 !atomic_read(&q->mq_freeze_depth) ||
662 blk_queue_dying(q));
663 if (blk_queue_dying(q))
664 return -ENODEV;
665 if (ret)
666 return ret;
667 }
668}
669
670void blk_queue_exit(struct request_queue *q)
671{
672 percpu_ref_put(&q->q_usage_counter);
673}
674
675static void blk_queue_usage_counter_release(struct percpu_ref *ref)
676{
677 struct request_queue *q =
678 container_of(ref, struct request_queue, q_usage_counter);
679
680 wake_up_all(&q->mq_freeze_wq);
681}
682
683static void blk_rq_timed_out_timer(unsigned long data)
684{
685 struct request_queue *q = (struct request_queue *)data;
686
687 kblockd_schedule_work(&q->timeout_work);
688}
689
690struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
691{
692 struct request_queue *q;
693 int err;
694
695 q = kmem_cache_alloc_node(blk_requestq_cachep,
696 gfp_mask | __GFP_ZERO, node_id);
697 if (!q)
698 return NULL;
699
700 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
701 if (q->id < 0)
702 goto fail_q;
703
704 q->bio_split = bioset_create(BIO_POOL_SIZE, 0);
705 if (!q->bio_split)
706 goto fail_id;
707
708 q->backing_dev_info.ra_pages =
709 (VM_MAX_READAHEAD * 1024) / PAGE_SIZE;
710 q->backing_dev_info.capabilities = BDI_CAP_CGROUP_WRITEBACK;
711 q->backing_dev_info.name = "block";
712 q->node = node_id;
713
714 err = bdi_init(&q->backing_dev_info);
715 if (err)
716 goto fail_split;
717
718 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
719 laptop_mode_timer_fn, (unsigned long) q);
720 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
721 INIT_LIST_HEAD(&q->queue_head);
722 INIT_LIST_HEAD(&q->timeout_list);
723 INIT_LIST_HEAD(&q->icq_list);
724#ifdef CONFIG_BLK_CGROUP
725 INIT_LIST_HEAD(&q->blkg_list);
726#endif
727 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
728
729 kobject_init(&q->kobj, &blk_queue_ktype);
730
731 mutex_init(&q->sysfs_lock);
732 spin_lock_init(&q->__queue_lock);
733
734 /*
735 * By default initialize queue_lock to internal lock and driver can
736 * override it later if need be.
737 */
738 q->queue_lock = &q->__queue_lock;
739
740 /*
741 * A queue starts its life with bypass turned on to avoid
742 * unnecessary bypass on/off overhead and nasty surprises during
743 * init. The initial bypass will be finished when the queue is
744 * registered by blk_register_queue().
745 */
746 q->bypass_depth = 1;
747 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
748
749 init_waitqueue_head(&q->mq_freeze_wq);
750
751 /*
752 * Init percpu_ref in atomic mode so that it's faster to shutdown.
753 * See blk_register_queue() for details.
754 */
755 if (percpu_ref_init(&q->q_usage_counter,
756 blk_queue_usage_counter_release,
757 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
758 goto fail_bdi;
759
760 if (blkcg_init_queue(q))
761 goto fail_ref;
762
763 return q;
764
765fail_ref:
766 percpu_ref_exit(&q->q_usage_counter);
767fail_bdi:
768 bdi_destroy(&q->backing_dev_info);
769fail_split:
770 bioset_free(q->bio_split);
771fail_id:
772 ida_simple_remove(&blk_queue_ida, q->id);
773fail_q:
774 kmem_cache_free(blk_requestq_cachep, q);
775 return NULL;
776}
777EXPORT_SYMBOL(blk_alloc_queue_node);
778
779/**
780 * blk_init_queue - prepare a request queue for use with a block device
781 * @rfn: The function to be called to process requests that have been
782 * placed on the queue.
783 * @lock: Request queue spin lock
784 *
785 * Description:
786 * If a block device wishes to use the standard request handling procedures,
787 * which sorts requests and coalesces adjacent requests, then it must
788 * call blk_init_queue(). The function @rfn will be called when there
789 * are requests on the queue that need to be processed. If the device
790 * supports plugging, then @rfn may not be called immediately when requests
791 * are available on the queue, but may be called at some time later instead.
792 * Plugged queues are generally unplugged when a buffer belonging to one
793 * of the requests on the queue is needed, or due to memory pressure.
794 *
795 * @rfn is not required, or even expected, to remove all requests off the
796 * queue, but only as many as it can handle at a time. If it does leave
797 * requests on the queue, it is responsible for arranging that the requests
798 * get dealt with eventually.
799 *
800 * The queue spin lock must be held while manipulating the requests on the
801 * request queue; this lock will be taken also from interrupt context, so irq
802 * disabling is needed for it.
803 *
804 * Function returns a pointer to the initialized request queue, or %NULL if
805 * it didn't succeed.
806 *
807 * Note:
808 * blk_init_queue() must be paired with a blk_cleanup_queue() call
809 * when the block device is deactivated (such as at module unload).
810 **/
811
812struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
813{
814 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
815}
816EXPORT_SYMBOL(blk_init_queue);
817
818struct request_queue *
819blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
820{
821 struct request_queue *uninit_q, *q;
822
823 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
824 if (!uninit_q)
825 return NULL;
826
827 q = blk_init_allocated_queue(uninit_q, rfn, lock);
828 if (!q)
829 blk_cleanup_queue(uninit_q);
830
831 return q;
832}
833EXPORT_SYMBOL(blk_init_queue_node);
834
835static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio);
836
837struct request_queue *
838blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
839 spinlock_t *lock)
840{
841 if (!q)
842 return NULL;
843
844 q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, 0);
845 if (!q->fq)
846 return NULL;
847
848 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
849 goto fail;
850
851 INIT_WORK(&q->timeout_work, blk_timeout_work);
852 q->request_fn = rfn;
853 q->prep_rq_fn = NULL;
854 q->unprep_rq_fn = NULL;
855 q->queue_flags |= QUEUE_FLAG_DEFAULT;
856
857 /* Override internal queue lock with supplied lock pointer */
858 if (lock)
859 q->queue_lock = lock;
860
861 /*
862 * This also sets hw/phys segments, boundary and size
863 */
864 blk_queue_make_request(q, blk_queue_bio);
865
866 q->sg_reserved_size = INT_MAX;
867
868 /* Protect q->elevator from elevator_change */
869 mutex_lock(&q->sysfs_lock);
870
871 /* init elevator */
872 if (elevator_init(q, NULL)) {
873 mutex_unlock(&q->sysfs_lock);
874 goto fail;
875 }
876
877 mutex_unlock(&q->sysfs_lock);
878
879 return q;
880
881fail:
882 blk_free_flush_queue(q->fq);
883 return NULL;
884}
885EXPORT_SYMBOL(blk_init_allocated_queue);
886
887bool blk_get_queue(struct request_queue *q)
888{
889 if (likely(!blk_queue_dying(q))) {
890 __blk_get_queue(q);
891 return true;
892 }
893
894 return false;
895}
896EXPORT_SYMBOL(blk_get_queue);
897
898static inline void blk_free_request(struct request_list *rl, struct request *rq)
899{
900 if (rq->cmd_flags & REQ_ELVPRIV) {
901 elv_put_request(rl->q, rq);
902 if (rq->elv.icq)
903 put_io_context(rq->elv.icq->ioc);
904 }
905
906 mempool_free(rq, rl->rq_pool);
907}
908
909/*
910 * ioc_batching returns true if the ioc is a valid batching request and
911 * should be given priority access to a request.
912 */
913static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
914{
915 if (!ioc)
916 return 0;
917
918 /*
919 * Make sure the process is able to allocate at least 1 request
920 * even if the batch times out, otherwise we could theoretically
921 * lose wakeups.
922 */
923 return ioc->nr_batch_requests == q->nr_batching ||
924 (ioc->nr_batch_requests > 0
925 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
926}
927
928/*
929 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
930 * will cause the process to be a "batcher" on all queues in the system. This
931 * is the behaviour we want though - once it gets a wakeup it should be given
932 * a nice run.
933 */
934static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
935{
936 if (!ioc || ioc_batching(q, ioc))
937 return;
938
939 ioc->nr_batch_requests = q->nr_batching;
940 ioc->last_waited = jiffies;
941}
942
943static void __freed_request(struct request_list *rl, int sync)
944{
945 struct request_queue *q = rl->q;
946
947 if (rl->count[sync] < queue_congestion_off_threshold(q))
948 blk_clear_congested(rl, sync);
949
950 if (rl->count[sync] + 1 <= q->nr_requests) {
951 if (waitqueue_active(&rl->wait[sync]))
952 wake_up(&rl->wait[sync]);
953
954 blk_clear_rl_full(rl, sync);
955 }
956}
957
958/*
959 * A request has just been released. Account for it, update the full and
960 * congestion status, wake up any waiters. Called under q->queue_lock.
961 */
962static void freed_request(struct request_list *rl, unsigned int flags)
963{
964 struct request_queue *q = rl->q;
965 int sync = rw_is_sync(flags);
966
967 q->nr_rqs[sync]--;
968 rl->count[sync]--;
969 if (flags & REQ_ELVPRIV)
970 q->nr_rqs_elvpriv--;
971
972 __freed_request(rl, sync);
973
974 if (unlikely(rl->starved[sync ^ 1]))
975 __freed_request(rl, sync ^ 1);
976}
977
978int blk_update_nr_requests(struct request_queue *q, unsigned int nr)
979{
980 struct request_list *rl;
981 int on_thresh, off_thresh;
982
983 spin_lock_irq(q->queue_lock);
984 q->nr_requests = nr;
985 blk_queue_congestion_threshold(q);
986 on_thresh = queue_congestion_on_threshold(q);
987 off_thresh = queue_congestion_off_threshold(q);
988
989 blk_queue_for_each_rl(rl, q) {
990 if (rl->count[BLK_RW_SYNC] >= on_thresh)
991 blk_set_congested(rl, BLK_RW_SYNC);
992 else if (rl->count[BLK_RW_SYNC] < off_thresh)
993 blk_clear_congested(rl, BLK_RW_SYNC);
994
995 if (rl->count[BLK_RW_ASYNC] >= on_thresh)
996 blk_set_congested(rl, BLK_RW_ASYNC);
997 else if (rl->count[BLK_RW_ASYNC] < off_thresh)
998 blk_clear_congested(rl, BLK_RW_ASYNC);
999
1000 if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
1001 blk_set_rl_full(rl, BLK_RW_SYNC);
1002 } else {
1003 blk_clear_rl_full(rl, BLK_RW_SYNC);
1004 wake_up(&rl->wait[BLK_RW_SYNC]);
1005 }
1006
1007 if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
1008 blk_set_rl_full(rl, BLK_RW_ASYNC);
1009 } else {
1010 blk_clear_rl_full(rl, BLK_RW_ASYNC);
1011 wake_up(&rl->wait[BLK_RW_ASYNC]);
1012 }
1013 }
1014
1015 spin_unlock_irq(q->queue_lock);
1016 return 0;
1017}
1018
1019/*
1020 * Determine if elevator data should be initialized when allocating the
1021 * request associated with @bio.
1022 */
1023static bool blk_rq_should_init_elevator(struct bio *bio)
1024{
1025 if (!bio)
1026 return true;
1027
1028 /*
1029 * Flush requests do not use the elevator so skip initialization.
1030 * This allows a request to share the flush and elevator data.
1031 */
1032 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
1033 return false;
1034
1035 return true;
1036}
1037
1038/**
1039 * rq_ioc - determine io_context for request allocation
1040 * @bio: request being allocated is for this bio (can be %NULL)
1041 *
1042 * Determine io_context to use for request allocation for @bio. May return
1043 * %NULL if %current->io_context doesn't exist.
1044 */
1045static struct io_context *rq_ioc(struct bio *bio)
1046{
1047#ifdef CONFIG_BLK_CGROUP
1048 if (bio && bio->bi_ioc)
1049 return bio->bi_ioc;
1050#endif
1051 return current->io_context;
1052}
1053
1054/**
1055 * __get_request - get a free request
1056 * @rl: request list to allocate from
1057 * @rw_flags: RW and SYNC flags
1058 * @bio: bio to allocate request for (can be %NULL)
1059 * @gfp_mask: allocation mask
1060 *
1061 * Get a free request from @q. This function may fail under memory
1062 * pressure or if @q is dead.
1063 *
1064 * Must be called with @q->queue_lock held and,
1065 * Returns ERR_PTR on failure, with @q->queue_lock held.
1066 * Returns request pointer on success, with @q->queue_lock *not held*.
1067 */
1068static struct request *__get_request(struct request_list *rl, int rw_flags,
1069 struct bio *bio, gfp_t gfp_mask)
1070{
1071 struct request_queue *q = rl->q;
1072 struct request *rq;
1073 struct elevator_type *et = q->elevator->type;
1074 struct io_context *ioc = rq_ioc(bio);
1075 struct io_cq *icq = NULL;
1076 const bool is_sync = rw_is_sync(rw_flags) != 0;
1077 int may_queue;
1078
1079 if (unlikely(blk_queue_dying(q)))
1080 return ERR_PTR(-ENODEV);
1081
1082 may_queue = elv_may_queue(q, rw_flags);
1083 if (may_queue == ELV_MQUEUE_NO)
1084 goto rq_starved;
1085
1086 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
1087 if (rl->count[is_sync]+1 >= q->nr_requests) {
1088 /*
1089 * The queue will fill after this allocation, so set
1090 * it as full, and mark this process as "batching".
1091 * This process will be allowed to complete a batch of
1092 * requests, others will be blocked.
1093 */
1094 if (!blk_rl_full(rl, is_sync)) {
1095 ioc_set_batching(q, ioc);
1096 blk_set_rl_full(rl, is_sync);
1097 } else {
1098 if (may_queue != ELV_MQUEUE_MUST
1099 && !ioc_batching(q, ioc)) {
1100 /*
1101 * The queue is full and the allocating
1102 * process is not a "batcher", and not
1103 * exempted by the IO scheduler
1104 */
1105 return ERR_PTR(-ENOMEM);
1106 }
1107 }
1108 }
1109 blk_set_congested(rl, is_sync);
1110 }
1111
1112 /*
1113 * Only allow batching queuers to allocate up to 50% over the defined
1114 * limit of requests, otherwise we could have thousands of requests
1115 * allocated with any setting of ->nr_requests
1116 */
1117 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
1118 return ERR_PTR(-ENOMEM);
1119
1120 q->nr_rqs[is_sync]++;
1121 rl->count[is_sync]++;
1122 rl->starved[is_sync] = 0;
1123
1124 /*
1125 * Decide whether the new request will be managed by elevator. If
1126 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
1127 * prevent the current elevator from being destroyed until the new
1128 * request is freed. This guarantees icq's won't be destroyed and
1129 * makes creating new ones safe.
1130 *
1131 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1132 * it will be created after releasing queue_lock.
1133 */
1134 if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
1135 rw_flags |= REQ_ELVPRIV;
1136 q->nr_rqs_elvpriv++;
1137 if (et->icq_cache && ioc)
1138 icq = ioc_lookup_icq(ioc, q);
1139 }
1140
1141 if (blk_queue_io_stat(q))
1142 rw_flags |= REQ_IO_STAT;
1143 spin_unlock_irq(q->queue_lock);
1144
1145 /* allocate and init request */
1146 rq = mempool_alloc(rl->rq_pool, gfp_mask);
1147 if (!rq)
1148 goto fail_alloc;
1149
1150 blk_rq_init(q, rq);
1151 blk_rq_set_rl(rq, rl);
1152 rq->cmd_flags = rw_flags | REQ_ALLOCED;
1153
1154 /* init elvpriv */
1155 if (rw_flags & REQ_ELVPRIV) {
1156 if (unlikely(et->icq_cache && !icq)) {
1157 if (ioc)
1158 icq = ioc_create_icq(ioc, q, gfp_mask);
1159 if (!icq)
1160 goto fail_elvpriv;
1161 }
1162
1163 rq->elv.icq = icq;
1164 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1165 goto fail_elvpriv;
1166
1167 /* @rq->elv.icq holds io_context until @rq is freed */
1168 if (icq)
1169 get_io_context(icq->ioc);
1170 }
1171out:
1172 /*
1173 * ioc may be NULL here, and ioc_batching will be false. That's
1174 * OK, if the queue is under the request limit then requests need
1175 * not count toward the nr_batch_requests limit. There will always
1176 * be some limit enforced by BLK_BATCH_TIME.
1177 */
1178 if (ioc_batching(q, ioc))
1179 ioc->nr_batch_requests--;
1180
1181 trace_block_getrq(q, bio, rw_flags & 1);
1182 return rq;
1183
1184fail_elvpriv:
1185 /*
1186 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1187 * and may fail indefinitely under memory pressure and thus
1188 * shouldn't stall IO. Treat this request as !elvpriv. This will
1189 * disturb iosched and blkcg but weird is bettern than dead.
1190 */
1191 printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1192 __func__, dev_name(q->backing_dev_info.dev));
1193
1194 rq->cmd_flags &= ~REQ_ELVPRIV;
1195 rq->elv.icq = NULL;
1196
1197 spin_lock_irq(q->queue_lock);
1198 q->nr_rqs_elvpriv--;
1199 spin_unlock_irq(q->queue_lock);
1200 goto out;
1201
1202fail_alloc:
1203 /*
1204 * Allocation failed presumably due to memory. Undo anything we
1205 * might have messed up.
1206 *
1207 * Allocating task should really be put onto the front of the wait
1208 * queue, but this is pretty rare.
1209 */
1210 spin_lock_irq(q->queue_lock);
1211 freed_request(rl, rw_flags);
1212
1213 /*
1214 * in the very unlikely event that allocation failed and no
1215 * requests for this direction was pending, mark us starved so that
1216 * freeing of a request in the other direction will notice
1217 * us. another possible fix would be to split the rq mempool into
1218 * READ and WRITE
1219 */
1220rq_starved:
1221 if (unlikely(rl->count[is_sync] == 0))
1222 rl->starved[is_sync] = 1;
1223 return ERR_PTR(-ENOMEM);
1224}
1225
1226/**
1227 * get_request - get a free request
1228 * @q: request_queue to allocate request from
1229 * @rw_flags: RW and SYNC flags
1230 * @bio: bio to allocate request for (can be %NULL)
1231 * @gfp_mask: allocation mask
1232 *
1233 * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask,
1234 * this function keeps retrying under memory pressure and fails iff @q is dead.
1235 *
1236 * Must be called with @q->queue_lock held and,
1237 * Returns ERR_PTR on failure, with @q->queue_lock held.
1238 * Returns request pointer on success, with @q->queue_lock *not held*.
1239 */
1240static struct request *get_request(struct request_queue *q, int rw_flags,
1241 struct bio *bio, gfp_t gfp_mask)
1242{
1243 const bool is_sync = rw_is_sync(rw_flags) != 0;
1244 DEFINE_WAIT(wait);
1245 struct request_list *rl;
1246 struct request *rq;
1247
1248 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
1249retry:
1250 rq = __get_request(rl, rw_flags, bio, gfp_mask);
1251 if (!IS_ERR(rq))
1252 return rq;
1253
1254 if (!gfpflags_allow_blocking(gfp_mask) || unlikely(blk_queue_dying(q))) {
1255 blk_put_rl(rl);
1256 return rq;
1257 }
1258
1259 /* wait on @rl and retry */
1260 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1261 TASK_UNINTERRUPTIBLE);
1262
1263 trace_block_sleeprq(q, bio, rw_flags & 1);
1264
1265 spin_unlock_irq(q->queue_lock);
1266 io_schedule();
1267
1268 /*
1269 * After sleeping, we become a "batching" process and will be able
1270 * to allocate at least one request, and up to a big batch of them
1271 * for a small period time. See ioc_batching, ioc_set_batching
1272 */
1273 ioc_set_batching(q, current->io_context);
1274
1275 spin_lock_irq(q->queue_lock);
1276 finish_wait(&rl->wait[is_sync], &wait);
1277
1278 goto retry;
1279}
1280
1281static struct request *blk_old_get_request(struct request_queue *q, int rw,
1282 gfp_t gfp_mask)
1283{
1284 struct request *rq;
1285
1286 BUG_ON(rw != READ && rw != WRITE);
1287
1288 /* create ioc upfront */
1289 create_io_context(gfp_mask, q->node);
1290
1291 spin_lock_irq(q->queue_lock);
1292 rq = get_request(q, rw, NULL, gfp_mask);
1293 if (IS_ERR(rq))
1294 spin_unlock_irq(q->queue_lock);
1295 /* q->queue_lock is unlocked at this point */
1296
1297 return rq;
1298}
1299
1300struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1301{
1302 if (q->mq_ops)
1303 return blk_mq_alloc_request(q, rw,
1304 (gfp_mask & __GFP_DIRECT_RECLAIM) ?
1305 0 : BLK_MQ_REQ_NOWAIT);
1306 else
1307 return blk_old_get_request(q, rw, gfp_mask);
1308}
1309EXPORT_SYMBOL(blk_get_request);
1310
1311/**
1312 * blk_make_request - given a bio, allocate a corresponding struct request.
1313 * @q: target request queue
1314 * @bio: The bio describing the memory mappings that will be submitted for IO.
1315 * It may be a chained-bio properly constructed by block/bio layer.
1316 * @gfp_mask: gfp flags to be used for memory allocation
1317 *
1318 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1319 * type commands. Where the struct request needs to be farther initialized by
1320 * the caller. It is passed a &struct bio, which describes the memory info of
1321 * the I/O transfer.
1322 *
1323 * The caller of blk_make_request must make sure that bi_io_vec
1324 * are set to describe the memory buffers. That bio_data_dir() will return
1325 * the needed direction of the request. (And all bio's in the passed bio-chain
1326 * are properly set accordingly)
1327 *
1328 * If called under none-sleepable conditions, mapped bio buffers must not
1329 * need bouncing, by calling the appropriate masked or flagged allocator,
1330 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1331 * BUG.
1332 *
1333 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1334 * given to how you allocate bios. In particular, you cannot use
1335 * __GFP_DIRECT_RECLAIM for anything but the first bio in the chain. Otherwise
1336 * you risk waiting for IO completion of a bio that hasn't been submitted yet,
1337 * thus resulting in a deadlock. Alternatively bios should be allocated using
1338 * bio_kmalloc() instead of bio_alloc(), as that avoids the mempool deadlock.
1339 * If possible a big IO should be split into smaller parts when allocation
1340 * fails. Partial allocation should not be an error, or you risk a live-lock.
1341 */
1342struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1343 gfp_t gfp_mask)
1344{
1345 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1346
1347 if (IS_ERR(rq))
1348 return rq;
1349
1350 blk_rq_set_block_pc(rq);
1351
1352 for_each_bio(bio) {
1353 struct bio *bounce_bio = bio;
1354 int ret;
1355
1356 blk_queue_bounce(q, &bounce_bio);
1357 ret = blk_rq_append_bio(q, rq, bounce_bio);
1358 if (unlikely(ret)) {
1359 blk_put_request(rq);
1360 return ERR_PTR(ret);
1361 }
1362 }
1363
1364 return rq;
1365}
1366EXPORT_SYMBOL(blk_make_request);
1367
1368/**
1369 * blk_rq_set_block_pc - initialize a request to type BLOCK_PC
1370 * @rq: request to be initialized
1371 *
1372 */
1373void blk_rq_set_block_pc(struct request *rq)
1374{
1375 rq->cmd_type = REQ_TYPE_BLOCK_PC;
1376 rq->__data_len = 0;
1377 rq->__sector = (sector_t) -1;
1378 rq->bio = rq->biotail = NULL;
1379 memset(rq->__cmd, 0, sizeof(rq->__cmd));
1380}
1381EXPORT_SYMBOL(blk_rq_set_block_pc);
1382
1383/**
1384 * blk_requeue_request - put a request back on queue
1385 * @q: request queue where request should be inserted
1386 * @rq: request to be inserted
1387 *
1388 * Description:
1389 * Drivers often keep queueing requests until the hardware cannot accept
1390 * more, when that condition happens we need to put the request back
1391 * on the queue. Must be called with queue lock held.
1392 */
1393void blk_requeue_request(struct request_queue *q, struct request *rq)
1394{
1395 blk_delete_timer(rq);
1396 blk_clear_rq_complete(rq);
1397 trace_block_rq_requeue(q, rq);
1398
1399 if (rq->cmd_flags & REQ_QUEUED)
1400 blk_queue_end_tag(q, rq);
1401
1402 BUG_ON(blk_queued_rq(rq));
1403
1404 elv_requeue_request(q, rq);
1405}
1406EXPORT_SYMBOL(blk_requeue_request);
1407
1408static void add_acct_request(struct request_queue *q, struct request *rq,
1409 int where)
1410{
1411 blk_account_io_start(rq, true);
1412 __elv_add_request(q, rq, where);
1413}
1414
1415static void part_round_stats_single(int cpu, struct hd_struct *part,
1416 unsigned long now)
1417{
1418 int inflight;
1419
1420 if (now == part->stamp)
1421 return;
1422
1423 inflight = part_in_flight(part);
1424 if (inflight) {
1425 __part_stat_add(cpu, part, time_in_queue,
1426 inflight * (now - part->stamp));
1427 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1428 }
1429 part->stamp = now;
1430}
1431
1432/**
1433 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1434 * @cpu: cpu number for stats access
1435 * @part: target partition
1436 *
1437 * The average IO queue length and utilisation statistics are maintained
1438 * by observing the current state of the queue length and the amount of
1439 * time it has been in this state for.
1440 *
1441 * Normally, that accounting is done on IO completion, but that can result
1442 * in more than a second's worth of IO being accounted for within any one
1443 * second, leading to >100% utilisation. To deal with that, we call this
1444 * function to do a round-off before returning the results when reading
1445 * /proc/diskstats. This accounts immediately for all queue usage up to
1446 * the current jiffies and restarts the counters again.
1447 */
1448void part_round_stats(int cpu, struct hd_struct *part)
1449{
1450 unsigned long now = jiffies;
1451
1452 if (part->partno)
1453 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1454 part_round_stats_single(cpu, part, now);
1455}
1456EXPORT_SYMBOL_GPL(part_round_stats);
1457
1458#ifdef CONFIG_PM
1459static void blk_pm_put_request(struct request *rq)
1460{
1461 if (rq->q->dev && !(rq->cmd_flags & REQ_PM) && !--rq->q->nr_pending)
1462 pm_runtime_mark_last_busy(rq->q->dev);
1463}
1464#else
1465static inline void blk_pm_put_request(struct request *rq) {}
1466#endif
1467
1468/*
1469 * queue lock must be held
1470 */
1471void __blk_put_request(struct request_queue *q, struct request *req)
1472{
1473 if (unlikely(!q))
1474 return;
1475
1476 if (q->mq_ops) {
1477 blk_mq_free_request(req);
1478 return;
1479 }
1480
1481 blk_pm_put_request(req);
1482
1483 elv_completed_request(q, req);
1484
1485 /* this is a bio leak */
1486 WARN_ON(req->bio != NULL);
1487
1488 /*
1489 * Request may not have originated from ll_rw_blk. if not,
1490 * it didn't come out of our reserved rq pools
1491 */
1492 if (req->cmd_flags & REQ_ALLOCED) {
1493 unsigned int flags = req->cmd_flags;
1494 struct request_list *rl = blk_rq_rl(req);
1495
1496 BUG_ON(!list_empty(&req->queuelist));
1497 BUG_ON(ELV_ON_HASH(req));
1498
1499 blk_free_request(rl, req);
1500 freed_request(rl, flags);
1501 blk_put_rl(rl);
1502 }
1503}
1504EXPORT_SYMBOL_GPL(__blk_put_request);
1505
1506void blk_put_request(struct request *req)
1507{
1508 struct request_queue *q = req->q;
1509
1510 if (q->mq_ops)
1511 blk_mq_free_request(req);
1512 else {
1513 unsigned long flags;
1514
1515 spin_lock_irqsave(q->queue_lock, flags);
1516 __blk_put_request(q, req);
1517 spin_unlock_irqrestore(q->queue_lock, flags);
1518 }
1519}
1520EXPORT_SYMBOL(blk_put_request);
1521
1522/**
1523 * blk_add_request_payload - add a payload to a request
1524 * @rq: request to update
1525 * @page: page backing the payload
1526 * @len: length of the payload.
1527 *
1528 * This allows to later add a payload to an already submitted request by
1529 * a block driver. The driver needs to take care of freeing the payload
1530 * itself.
1531 *
1532 * Note that this is a quite horrible hack and nothing but handling of
1533 * discard requests should ever use it.
1534 */
1535void blk_add_request_payload(struct request *rq, struct page *page,
1536 unsigned int len)
1537{
1538 struct bio *bio = rq->bio;
1539
1540 bio->bi_io_vec->bv_page = page;
1541 bio->bi_io_vec->bv_offset = 0;
1542 bio->bi_io_vec->bv_len = len;
1543
1544 bio->bi_iter.bi_size = len;
1545 bio->bi_vcnt = 1;
1546 bio->bi_phys_segments = 1;
1547
1548 rq->__data_len = rq->resid_len = len;
1549 rq->nr_phys_segments = 1;
1550}
1551EXPORT_SYMBOL_GPL(blk_add_request_payload);
1552
1553bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1554 struct bio *bio)
1555{
1556 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1557
1558 if (!ll_back_merge_fn(q, req, bio))
1559 return false;
1560
1561 trace_block_bio_backmerge(q, req, bio);
1562
1563 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1564 blk_rq_set_mixed_merge(req);
1565
1566 req->biotail->bi_next = bio;
1567 req->biotail = bio;
1568 req->__data_len += bio->bi_iter.bi_size;
1569 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1570
1571 blk_account_io_start(req, false);
1572 return true;
1573}
1574
1575bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1576 struct bio *bio)
1577{
1578 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1579
1580 if (!ll_front_merge_fn(q, req, bio))
1581 return false;
1582
1583 trace_block_bio_frontmerge(q, req, bio);
1584
1585 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1586 blk_rq_set_mixed_merge(req);
1587
1588 bio->bi_next = req->bio;
1589 req->bio = bio;
1590
1591 req->__sector = bio->bi_iter.bi_sector;
1592 req->__data_len += bio->bi_iter.bi_size;
1593 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1594
1595 blk_account_io_start(req, false);
1596 return true;
1597}
1598
1599/**
1600 * blk_attempt_plug_merge - try to merge with %current's plugged list
1601 * @q: request_queue new bio is being queued at
1602 * @bio: new bio being queued
1603 * @request_count: out parameter for number of traversed plugged requests
1604 * @same_queue_rq: pointer to &struct request that gets filled in when
1605 * another request associated with @q is found on the plug list
1606 * (optional, may be %NULL)
1607 *
1608 * Determine whether @bio being queued on @q can be merged with a request
1609 * on %current's plugged list. Returns %true if merge was successful,
1610 * otherwise %false.
1611 *
1612 * Plugging coalesces IOs from the same issuer for the same purpose without
1613 * going through @q->queue_lock. As such it's more of an issuing mechanism
1614 * than scheduling, and the request, while may have elvpriv data, is not
1615 * added on the elevator at this point. In addition, we don't have
1616 * reliable access to the elevator outside queue lock. Only check basic
1617 * merging parameters without querying the elevator.
1618 *
1619 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1620 */
1621bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1622 unsigned int *request_count,
1623 struct request **same_queue_rq)
1624{
1625 struct blk_plug *plug;
1626 struct request *rq;
1627 bool ret = false;
1628 struct list_head *plug_list;
1629
1630 plug = current->plug;
1631 if (!plug)
1632 goto out;
1633 *request_count = 0;
1634
1635 if (q->mq_ops)
1636 plug_list = &plug->mq_list;
1637 else
1638 plug_list = &plug->list;
1639
1640 list_for_each_entry_reverse(rq, plug_list, queuelist) {
1641 int el_ret;
1642
1643 if (rq->q == q) {
1644 (*request_count)++;
1645 /*
1646 * Only blk-mq multiple hardware queues case checks the
1647 * rq in the same queue, there should be only one such
1648 * rq in a queue
1649 **/
1650 if (same_queue_rq)
1651 *same_queue_rq = rq;
1652 }
1653
1654 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
1655 continue;
1656
1657 el_ret = blk_try_merge(rq, bio);
1658 if (el_ret == ELEVATOR_BACK_MERGE) {
1659 ret = bio_attempt_back_merge(q, rq, bio);
1660 if (ret)
1661 break;
1662 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1663 ret = bio_attempt_front_merge(q, rq, bio);
1664 if (ret)
1665 break;
1666 }
1667 }
1668out:
1669 return ret;
1670}
1671
1672unsigned int blk_plug_queued_count(struct request_queue *q)
1673{
1674 struct blk_plug *plug;
1675 struct request *rq;
1676 struct list_head *plug_list;
1677 unsigned int ret = 0;
1678
1679 plug = current->plug;
1680 if (!plug)
1681 goto out;
1682
1683 if (q->mq_ops)
1684 plug_list = &plug->mq_list;
1685 else
1686 plug_list = &plug->list;
1687
1688 list_for_each_entry(rq, plug_list, queuelist) {
1689 if (rq->q == q)
1690 ret++;
1691 }
1692out:
1693 return ret;
1694}
1695
1696void init_request_from_bio(struct request *req, struct bio *bio)
1697{
1698 req->cmd_type = REQ_TYPE_FS;
1699
1700 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1701 if (bio->bi_rw & REQ_RAHEAD)
1702 req->cmd_flags |= REQ_FAILFAST_MASK;
1703
1704 req->errors = 0;
1705 req->__sector = bio->bi_iter.bi_sector;
1706 req->ioprio = bio_prio(bio);
1707 blk_rq_bio_prep(req->q, req, bio);
1708}
1709
1710static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio)
1711{
1712 const bool sync = !!(bio->bi_rw & REQ_SYNC);
1713 struct blk_plug *plug;
1714 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1715 struct request *req;
1716 unsigned int request_count = 0;
1717
1718 /*
1719 * low level driver can indicate that it wants pages above a
1720 * certain limit bounced to low memory (ie for highmem, or even
1721 * ISA dma in theory)
1722 */
1723 blk_queue_bounce(q, &bio);
1724
1725 blk_queue_split(q, &bio, q->bio_split);
1726
1727 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
1728 bio->bi_error = -EIO;
1729 bio_endio(bio);
1730 return BLK_QC_T_NONE;
1731 }
1732
1733 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
1734 spin_lock_irq(q->queue_lock);
1735 where = ELEVATOR_INSERT_FLUSH;
1736 goto get_rq;
1737 }
1738
1739 /*
1740 * Check if we can merge with the plugged list before grabbing
1741 * any locks.
1742 */
1743 if (!blk_queue_nomerges(q)) {
1744 if (blk_attempt_plug_merge(q, bio, &request_count, NULL))
1745 return BLK_QC_T_NONE;
1746 } else
1747 request_count = blk_plug_queued_count(q);
1748
1749 spin_lock_irq(q->queue_lock);
1750
1751 el_ret = elv_merge(q, &req, bio);
1752 if (el_ret == ELEVATOR_BACK_MERGE) {
1753 if (bio_attempt_back_merge(q, req, bio)) {
1754 elv_bio_merged(q, req, bio);
1755 if (!attempt_back_merge(q, req))
1756 elv_merged_request(q, req, el_ret);
1757 goto out_unlock;
1758 }
1759 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1760 if (bio_attempt_front_merge(q, req, bio)) {
1761 elv_bio_merged(q, req, bio);
1762 if (!attempt_front_merge(q, req))
1763 elv_merged_request(q, req, el_ret);
1764 goto out_unlock;
1765 }
1766 }
1767
1768get_rq:
1769 /*
1770 * This sync check and mask will be re-done in init_request_from_bio(),
1771 * but we need to set it earlier to expose the sync flag to the
1772 * rq allocator and io schedulers.
1773 */
1774 rw_flags = bio_data_dir(bio);
1775 if (sync)
1776 rw_flags |= REQ_SYNC;
1777
1778 /*
1779 * Grab a free request. This is might sleep but can not fail.
1780 * Returns with the queue unlocked.
1781 */
1782 req = get_request(q, rw_flags, bio, GFP_NOIO);
1783 if (IS_ERR(req)) {
1784 bio->bi_error = PTR_ERR(req);
1785 bio_endio(bio);
1786 goto out_unlock;
1787 }
1788
1789 /*
1790 * After dropping the lock and possibly sleeping here, our request
1791 * may now be mergeable after it had proven unmergeable (above).
1792 * We don't worry about that case for efficiency. It won't happen
1793 * often, and the elevators are able to handle it.
1794 */
1795 init_request_from_bio(req, bio);
1796
1797 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
1798 req->cpu = raw_smp_processor_id();
1799
1800 plug = current->plug;
1801 if (plug) {
1802 /*
1803 * If this is the first request added after a plug, fire
1804 * of a plug trace.
1805 */
1806 if (!request_count)
1807 trace_block_plug(q);
1808 else {
1809 if (request_count >= BLK_MAX_REQUEST_COUNT) {
1810 blk_flush_plug_list(plug, false);
1811 trace_block_plug(q);
1812 }
1813 }
1814 list_add_tail(&req->queuelist, &plug->list);
1815 blk_account_io_start(req, true);
1816 } else {
1817 spin_lock_irq(q->queue_lock);
1818 add_acct_request(q, req, where);
1819 __blk_run_queue(q);
1820out_unlock:
1821 spin_unlock_irq(q->queue_lock);
1822 }
1823
1824 return BLK_QC_T_NONE;
1825}
1826
1827/*
1828 * If bio->bi_dev is a partition, remap the location
1829 */
1830static inline void blk_partition_remap(struct bio *bio)
1831{
1832 struct block_device *bdev = bio->bi_bdev;
1833
1834 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1835 struct hd_struct *p = bdev->bd_part;
1836
1837 bio->bi_iter.bi_sector += p->start_sect;
1838 bio->bi_bdev = bdev->bd_contains;
1839
1840 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1841 bdev->bd_dev,
1842 bio->bi_iter.bi_sector - p->start_sect);
1843 }
1844}
1845
1846static void handle_bad_sector(struct bio *bio)
1847{
1848 char b[BDEVNAME_SIZE];
1849
1850 printk(KERN_INFO "attempt to access beyond end of device\n");
1851 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1852 bdevname(bio->bi_bdev, b),
1853 bio->bi_rw,
1854 (unsigned long long)bio_end_sector(bio),
1855 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1856}
1857
1858#ifdef CONFIG_FAIL_MAKE_REQUEST
1859
1860static DECLARE_FAULT_ATTR(fail_make_request);
1861
1862static int __init setup_fail_make_request(char *str)
1863{
1864 return setup_fault_attr(&fail_make_request, str);
1865}
1866__setup("fail_make_request=", setup_fail_make_request);
1867
1868static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1869{
1870 return part->make_it_fail && should_fail(&fail_make_request, bytes);
1871}
1872
1873static int __init fail_make_request_debugfs(void)
1874{
1875 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1876 NULL, &fail_make_request);
1877
1878 return PTR_ERR_OR_ZERO(dir);
1879}
1880
1881late_initcall(fail_make_request_debugfs);
1882
1883#else /* CONFIG_FAIL_MAKE_REQUEST */
1884
1885static inline bool should_fail_request(struct hd_struct *part,
1886 unsigned int bytes)
1887{
1888 return false;
1889}
1890
1891#endif /* CONFIG_FAIL_MAKE_REQUEST */
1892
1893/*
1894 * Check whether this bio extends beyond the end of the device.
1895 */
1896static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1897{
1898 sector_t maxsector;
1899
1900 if (!nr_sectors)
1901 return 0;
1902
1903 /* Test device or partition size, when known. */
1904 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1905 if (maxsector) {
1906 sector_t sector = bio->bi_iter.bi_sector;
1907
1908 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1909 /*
1910 * This may well happen - the kernel calls bread()
1911 * without checking the size of the device, e.g., when
1912 * mounting a device.
1913 */
1914 handle_bad_sector(bio);
1915 return 1;
1916 }
1917 }
1918
1919 return 0;
1920}
1921
1922static noinline_for_stack bool
1923generic_make_request_checks(struct bio *bio)
1924{
1925 struct request_queue *q;
1926 int nr_sectors = bio_sectors(bio);
1927 int err = -EIO;
1928 char b[BDEVNAME_SIZE];
1929 struct hd_struct *part;
1930
1931 might_sleep();
1932
1933 if (bio_check_eod(bio, nr_sectors))
1934 goto end_io;
1935
1936 q = bdev_get_queue(bio->bi_bdev);
1937 if (unlikely(!q)) {
1938 printk(KERN_ERR
1939 "generic_make_request: Trying to access "
1940 "nonexistent block-device %s (%Lu)\n",
1941 bdevname(bio->bi_bdev, b),
1942 (long long) bio->bi_iter.bi_sector);
1943 goto end_io;
1944 }
1945
1946 part = bio->bi_bdev->bd_part;
1947 if (should_fail_request(part, bio->bi_iter.bi_size) ||
1948 should_fail_request(&part_to_disk(part)->part0,
1949 bio->bi_iter.bi_size))
1950 goto end_io;
1951
1952 /*
1953 * If this device has partitions, remap block n
1954 * of partition p to block n+start(p) of the disk.
1955 */
1956 blk_partition_remap(bio);
1957
1958 if (bio_check_eod(bio, nr_sectors))
1959 goto end_io;
1960
1961 /*
1962 * Filter flush bio's early so that make_request based
1963 * drivers without flush support don't have to worry
1964 * about them.
1965 */
1966 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1967 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1968 if (!nr_sectors) {
1969 err = 0;
1970 goto end_io;
1971 }
1972 }
1973
1974 if ((bio->bi_rw & REQ_DISCARD) &&
1975 (!blk_queue_discard(q) ||
1976 ((bio->bi_rw & REQ_SECURE) && !blk_queue_secdiscard(q)))) {
1977 err = -EOPNOTSUPP;
1978 goto end_io;
1979 }
1980
1981 if (bio->bi_rw & REQ_WRITE_SAME && !bdev_write_same(bio->bi_bdev)) {
1982 err = -EOPNOTSUPP;
1983 goto end_io;
1984 }
1985
1986 /*
1987 * Various block parts want %current->io_context and lazy ioc
1988 * allocation ends up trading a lot of pain for a small amount of
1989 * memory. Just allocate it upfront. This may fail and block
1990 * layer knows how to live with it.
1991 */
1992 create_io_context(GFP_ATOMIC, q->node);
1993
1994 if (!blkcg_bio_issue_check(q, bio))
1995 return false;
1996
1997 trace_block_bio_queue(q, bio);
1998 return true;
1999
2000end_io:
2001 bio->bi_error = err;
2002 bio_endio(bio);
2003 return false;
2004}
2005
2006/**
2007 * generic_make_request - hand a buffer to its device driver for I/O
2008 * @bio: The bio describing the location in memory and on the device.
2009 *
2010 * generic_make_request() is used to make I/O requests of block
2011 * devices. It is passed a &struct bio, which describes the I/O that needs
2012 * to be done.
2013 *
2014 * generic_make_request() does not return any status. The
2015 * success/failure status of the request, along with notification of
2016 * completion, is delivered asynchronously through the bio->bi_end_io
2017 * function described (one day) else where.
2018 *
2019 * The caller of generic_make_request must make sure that bi_io_vec
2020 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2021 * set to describe the device address, and the
2022 * bi_end_io and optionally bi_private are set to describe how
2023 * completion notification should be signaled.
2024 *
2025 * generic_make_request and the drivers it calls may use bi_next if this
2026 * bio happens to be merged with someone else, and may resubmit the bio to
2027 * a lower device by calling into generic_make_request recursively, which
2028 * means the bio should NOT be touched after the call to ->make_request_fn.
2029 */
2030blk_qc_t generic_make_request(struct bio *bio)
2031{
2032 struct bio_list bio_list_on_stack;
2033 blk_qc_t ret = BLK_QC_T_NONE;
2034
2035 if (!generic_make_request_checks(bio))
2036 goto out;
2037
2038 /*
2039 * We only want one ->make_request_fn to be active at a time, else
2040 * stack usage with stacked devices could be a problem. So use
2041 * current->bio_list to keep a list of requests submited by a
2042 * make_request_fn function. current->bio_list is also used as a
2043 * flag to say if generic_make_request is currently active in this
2044 * task or not. If it is NULL, then no make_request is active. If
2045 * it is non-NULL, then a make_request is active, and new requests
2046 * should be added at the tail
2047 */
2048 if (current->bio_list) {
2049 bio_list_add(current->bio_list, bio);
2050 goto out;
2051 }
2052
2053 /* following loop may be a bit non-obvious, and so deserves some
2054 * explanation.
2055 * Before entering the loop, bio->bi_next is NULL (as all callers
2056 * ensure that) so we have a list with a single bio.
2057 * We pretend that we have just taken it off a longer list, so
2058 * we assign bio_list to a pointer to the bio_list_on_stack,
2059 * thus initialising the bio_list of new bios to be
2060 * added. ->make_request() may indeed add some more bios
2061 * through a recursive call to generic_make_request. If it
2062 * did, we find a non-NULL value in bio_list and re-enter the loop
2063 * from the top. In this case we really did just take the bio
2064 * of the top of the list (no pretending) and so remove it from
2065 * bio_list, and call into ->make_request() again.
2066 */
2067 BUG_ON(bio->bi_next);
2068 bio_list_init(&bio_list_on_stack);
2069 current->bio_list = &bio_list_on_stack;
2070 do {
2071 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
2072
2073 if (likely(blk_queue_enter(q, false) == 0)) {
2074 ret = q->make_request_fn(q, bio);
2075
2076 blk_queue_exit(q);
2077
2078 bio = bio_list_pop(current->bio_list);
2079 } else {
2080 struct bio *bio_next = bio_list_pop(current->bio_list);
2081
2082 bio_io_error(bio);
2083 bio = bio_next;
2084 }
2085 } while (bio);
2086 current->bio_list = NULL; /* deactivate */
2087
2088out:
2089 return ret;
2090}
2091EXPORT_SYMBOL(generic_make_request);
2092
2093/**
2094 * submit_bio - submit a bio to the block device layer for I/O
2095 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
2096 * @bio: The &struct bio which describes the I/O
2097 *
2098 * submit_bio() is very similar in purpose to generic_make_request(), and
2099 * uses that function to do most of the work. Both are fairly rough
2100 * interfaces; @bio must be presetup and ready for I/O.
2101 *
2102 */
2103blk_qc_t submit_bio(int rw, struct bio *bio)
2104{
2105 bio->bi_rw |= rw;
2106
2107 /*
2108 * If it's a regular read/write or a barrier with data attached,
2109 * go through the normal accounting stuff before submission.
2110 */
2111 if (bio_has_data(bio)) {
2112 unsigned int count;
2113
2114 if (unlikely(rw & REQ_WRITE_SAME))
2115 count = bdev_logical_block_size(bio->bi_bdev) >> 9;
2116 else
2117 count = bio_sectors(bio);
2118
2119 if (rw & WRITE) {
2120 count_vm_events(PGPGOUT, count);
2121 } else {
2122 task_io_account_read(bio->bi_iter.bi_size);
2123 count_vm_events(PGPGIN, count);
2124 }
2125
2126 if (unlikely(block_dump)) {
2127 char b[BDEVNAME_SIZE];
2128 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
2129 current->comm, task_pid_nr(current),
2130 (rw & WRITE) ? "WRITE" : "READ",
2131 (unsigned long long)bio->bi_iter.bi_sector,
2132 bdevname(bio->bi_bdev, b),
2133 count);
2134 }
2135 }
2136
2137 return generic_make_request(bio);
2138}
2139EXPORT_SYMBOL(submit_bio);
2140
2141/**
2142 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2143 * for new the queue limits
2144 * @q: the queue
2145 * @rq: the request being checked
2146 *
2147 * Description:
2148 * @rq may have been made based on weaker limitations of upper-level queues
2149 * in request stacking drivers, and it may violate the limitation of @q.
2150 * Since the block layer and the underlying device driver trust @rq
2151 * after it is inserted to @q, it should be checked against @q before
2152 * the insertion using this generic function.
2153 *
2154 * Request stacking drivers like request-based dm may change the queue
2155 * limits when retrying requests on other queues. Those requests need
2156 * to be checked against the new queue limits again during dispatch.
2157 */
2158static int blk_cloned_rq_check_limits(struct request_queue *q,
2159 struct request *rq)
2160{
2161 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, rq->cmd_flags)) {
2162 printk(KERN_ERR "%s: over max size limit.\n", __func__);
2163 return -EIO;
2164 }
2165
2166 /*
2167 * queue's settings related to segment counting like q->bounce_pfn
2168 * may differ from that of other stacking queues.
2169 * Recalculate it to check the request correctly on this queue's
2170 * limitation.
2171 */
2172 blk_recalc_rq_segments(rq);
2173 if (rq->nr_phys_segments > queue_max_segments(q)) {
2174 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2175 return -EIO;
2176 }
2177
2178 return 0;
2179}
2180
2181/**
2182 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2183 * @q: the queue to submit the request
2184 * @rq: the request being queued
2185 */
2186int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
2187{
2188 unsigned long flags;
2189 int where = ELEVATOR_INSERT_BACK;
2190
2191 if (blk_cloned_rq_check_limits(q, rq))
2192 return -EIO;
2193
2194 if (rq->rq_disk &&
2195 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
2196 return -EIO;
2197
2198 if (q->mq_ops) {
2199 if (blk_queue_io_stat(q))
2200 blk_account_io_start(rq, true);
2201 blk_mq_insert_request(rq, false, true, false);
2202 return 0;
2203 }
2204
2205 spin_lock_irqsave(q->queue_lock, flags);
2206 if (unlikely(blk_queue_dying(q))) {
2207 spin_unlock_irqrestore(q->queue_lock, flags);
2208 return -ENODEV;
2209 }
2210
2211 /*
2212 * Submitting request must be dequeued before calling this function
2213 * because it will be linked to another request_queue
2214 */
2215 BUG_ON(blk_queued_rq(rq));
2216
2217 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
2218 where = ELEVATOR_INSERT_FLUSH;
2219
2220 add_acct_request(q, rq, where);
2221 if (where == ELEVATOR_INSERT_FLUSH)
2222 __blk_run_queue(q);
2223 spin_unlock_irqrestore(q->queue_lock, flags);
2224
2225 return 0;
2226}
2227EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2228
2229/**
2230 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2231 * @rq: request to examine
2232 *
2233 * Description:
2234 * A request could be merge of IOs which require different failure
2235 * handling. This function determines the number of bytes which
2236 * can be failed from the beginning of the request without
2237 * crossing into area which need to be retried further.
2238 *
2239 * Return:
2240 * The number of bytes to fail.
2241 *
2242 * Context:
2243 * queue_lock must be held.
2244 */
2245unsigned int blk_rq_err_bytes(const struct request *rq)
2246{
2247 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2248 unsigned int bytes = 0;
2249 struct bio *bio;
2250
2251 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
2252 return blk_rq_bytes(rq);
2253
2254 /*
2255 * Currently the only 'mixing' which can happen is between
2256 * different fastfail types. We can safely fail portions
2257 * which have all the failfast bits that the first one has -
2258 * the ones which are at least as eager to fail as the first
2259 * one.
2260 */
2261 for (bio = rq->bio; bio; bio = bio->bi_next) {
2262 if ((bio->bi_rw & ff) != ff)
2263 break;
2264 bytes += bio->bi_iter.bi_size;
2265 }
2266
2267 /* this could lead to infinite loop */
2268 BUG_ON(blk_rq_bytes(rq) && !bytes);
2269 return bytes;
2270}
2271EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2272
2273void blk_account_io_completion(struct request *req, unsigned int bytes)
2274{
2275 if (blk_do_io_stat(req)) {
2276 const int rw = rq_data_dir(req);
2277 struct hd_struct *part;
2278 int cpu;
2279
2280 cpu = part_stat_lock();
2281 part = req->part;
2282 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2283 part_stat_unlock();
2284 }
2285}
2286
2287void blk_account_io_done(struct request *req)
2288{
2289 /*
2290 * Account IO completion. flush_rq isn't accounted as a
2291 * normal IO on queueing nor completion. Accounting the
2292 * containing request is enough.
2293 */
2294 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
2295 unsigned long duration = jiffies - req->start_time;
2296 const int rw = rq_data_dir(req);
2297 struct hd_struct *part;
2298 int cpu;
2299
2300 cpu = part_stat_lock();
2301 part = req->part;
2302
2303 part_stat_inc(cpu, part, ios[rw]);
2304 part_stat_add(cpu, part, ticks[rw], duration);
2305 part_round_stats(cpu, part);
2306 part_dec_in_flight(part, rw);
2307
2308 hd_struct_put(part);
2309 part_stat_unlock();
2310 }
2311}
2312
2313#ifdef CONFIG_PM
2314/*
2315 * Don't process normal requests when queue is suspended
2316 * or in the process of suspending/resuming
2317 */
2318static struct request *blk_pm_peek_request(struct request_queue *q,
2319 struct request *rq)
2320{
2321 if (q->dev && (q->rpm_status == RPM_SUSPENDED ||
2322 (q->rpm_status != RPM_ACTIVE && !(rq->cmd_flags & REQ_PM))))
2323 return NULL;
2324 else
2325 return rq;
2326}
2327#else
2328static inline struct request *blk_pm_peek_request(struct request_queue *q,
2329 struct request *rq)
2330{
2331 return rq;
2332}
2333#endif
2334
2335void blk_account_io_start(struct request *rq, bool new_io)
2336{
2337 struct hd_struct *part;
2338 int rw = rq_data_dir(rq);
2339 int cpu;
2340
2341 if (!blk_do_io_stat(rq))
2342 return;
2343
2344 cpu = part_stat_lock();
2345
2346 if (!new_io) {
2347 part = rq->part;
2348 part_stat_inc(cpu, part, merges[rw]);
2349 } else {
2350 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2351 if (!hd_struct_try_get(part)) {
2352 /*
2353 * The partition is already being removed,
2354 * the request will be accounted on the disk only
2355 *
2356 * We take a reference on disk->part0 although that
2357 * partition will never be deleted, so we can treat
2358 * it as any other partition.
2359 */
2360 part = &rq->rq_disk->part0;
2361 hd_struct_get(part);
2362 }
2363 part_round_stats(cpu, part);
2364 part_inc_in_flight(part, rw);
2365 rq->part = part;
2366 }
2367
2368 part_stat_unlock();
2369}
2370
2371/**
2372 * blk_peek_request - peek at the top of a request queue
2373 * @q: request queue to peek at
2374 *
2375 * Description:
2376 * Return the request at the top of @q. The returned request
2377 * should be started using blk_start_request() before LLD starts
2378 * processing it.
2379 *
2380 * Return:
2381 * Pointer to the request at the top of @q if available. Null
2382 * otherwise.
2383 *
2384 * Context:
2385 * queue_lock must be held.
2386 */
2387struct request *blk_peek_request(struct request_queue *q)
2388{
2389 struct request *rq;
2390 int ret;
2391
2392 while ((rq = __elv_next_request(q)) != NULL) {
2393
2394 rq = blk_pm_peek_request(q, rq);
2395 if (!rq)
2396 break;
2397
2398 if (!(rq->cmd_flags & REQ_STARTED)) {
2399 /*
2400 * This is the first time the device driver
2401 * sees this request (possibly after
2402 * requeueing). Notify IO scheduler.
2403 */
2404 if (rq->cmd_flags & REQ_SORTED)
2405 elv_activate_rq(q, rq);
2406
2407 /*
2408 * just mark as started even if we don't start
2409 * it, a request that has been delayed should
2410 * not be passed by new incoming requests
2411 */
2412 rq->cmd_flags |= REQ_STARTED;
2413 trace_block_rq_issue(q, rq);
2414 }
2415
2416 if (!q->boundary_rq || q->boundary_rq == rq) {
2417 q->end_sector = rq_end_sector(rq);
2418 q->boundary_rq = NULL;
2419 }
2420
2421 if (rq->cmd_flags & REQ_DONTPREP)
2422 break;
2423
2424 if (q->dma_drain_size && blk_rq_bytes(rq)) {
2425 /*
2426 * make sure space for the drain appears we
2427 * know we can do this because max_hw_segments
2428 * has been adjusted to be one fewer than the
2429 * device can handle
2430 */
2431 rq->nr_phys_segments++;
2432 }
2433
2434 if (!q->prep_rq_fn)
2435 break;
2436
2437 ret = q->prep_rq_fn(q, rq);
2438 if (ret == BLKPREP_OK) {
2439 break;
2440 } else if (ret == BLKPREP_DEFER) {
2441 /*
2442 * the request may have been (partially) prepped.
2443 * we need to keep this request in the front to
2444 * avoid resource deadlock. REQ_STARTED will
2445 * prevent other fs requests from passing this one.
2446 */
2447 if (q->dma_drain_size && blk_rq_bytes(rq) &&
2448 !(rq->cmd_flags & REQ_DONTPREP)) {
2449 /*
2450 * remove the space for the drain we added
2451 * so that we don't add it again
2452 */
2453 --rq->nr_phys_segments;
2454 }
2455
2456 rq = NULL;
2457 break;
2458 } else if (ret == BLKPREP_KILL || ret == BLKPREP_INVALID) {
2459 int err = (ret == BLKPREP_INVALID) ? -EREMOTEIO : -EIO;
2460
2461 rq->cmd_flags |= REQ_QUIET;
2462 /*
2463 * Mark this request as started so we don't trigger
2464 * any debug logic in the end I/O path.
2465 */
2466 blk_start_request(rq);
2467 __blk_end_request_all(rq, err);
2468 } else {
2469 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2470 break;
2471 }
2472 }
2473
2474 return rq;
2475}
2476EXPORT_SYMBOL(blk_peek_request);
2477
2478void blk_dequeue_request(struct request *rq)
2479{
2480 struct request_queue *q = rq->q;
2481
2482 BUG_ON(list_empty(&rq->queuelist));
2483 BUG_ON(ELV_ON_HASH(rq));
2484
2485 list_del_init(&rq->queuelist);
2486
2487 /*
2488 * the time frame between a request being removed from the lists
2489 * and to it is freed is accounted as io that is in progress at
2490 * the driver side.
2491 */
2492 if (blk_account_rq(rq)) {
2493 q->in_flight[rq_is_sync(rq)]++;
2494 set_io_start_time_ns(rq);
2495 }
2496}
2497
2498/**
2499 * blk_start_request - start request processing on the driver
2500 * @req: request to dequeue
2501 *
2502 * Description:
2503 * Dequeue @req and start timeout timer on it. This hands off the
2504 * request to the driver.
2505 *
2506 * Block internal functions which don't want to start timer should
2507 * call blk_dequeue_request().
2508 *
2509 * Context:
2510 * queue_lock must be held.
2511 */
2512void blk_start_request(struct request *req)
2513{
2514 blk_dequeue_request(req);
2515
2516 /*
2517 * We are now handing the request to the hardware, initialize
2518 * resid_len to full count and add the timeout handler.
2519 */
2520 req->resid_len = blk_rq_bytes(req);
2521 if (unlikely(blk_bidi_rq(req)))
2522 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2523
2524 BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags));
2525 blk_add_timer(req);
2526}
2527EXPORT_SYMBOL(blk_start_request);
2528
2529/**
2530 * blk_fetch_request - fetch a request from a request queue
2531 * @q: request queue to fetch a request from
2532 *
2533 * Description:
2534 * Return the request at the top of @q. The request is started on
2535 * return and LLD can start processing it immediately.
2536 *
2537 * Return:
2538 * Pointer to the request at the top of @q if available. Null
2539 * otherwise.
2540 *
2541 * Context:
2542 * queue_lock must be held.
2543 */
2544struct request *blk_fetch_request(struct request_queue *q)
2545{
2546 struct request *rq;
2547
2548 rq = blk_peek_request(q);
2549 if (rq)
2550 blk_start_request(rq);
2551 return rq;
2552}
2553EXPORT_SYMBOL(blk_fetch_request);
2554
2555/**
2556 * blk_update_request - Special helper function for request stacking drivers
2557 * @req: the request being processed
2558 * @error: %0 for success, < %0 for error
2559 * @nr_bytes: number of bytes to complete @req
2560 *
2561 * Description:
2562 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2563 * the request structure even if @req doesn't have leftover.
2564 * If @req has leftover, sets it up for the next range of segments.
2565 *
2566 * This special helper function is only for request stacking drivers
2567 * (e.g. request-based dm) so that they can handle partial completion.
2568 * Actual device drivers should use blk_end_request instead.
2569 *
2570 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2571 * %false return from this function.
2572 *
2573 * Return:
2574 * %false - this request doesn't have any more data
2575 * %true - this request has more data
2576 **/
2577bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2578{
2579 int total_bytes;
2580
2581 trace_block_rq_complete(req->q, req, nr_bytes);
2582
2583 if (!req->bio)
2584 return false;
2585
2586 /*
2587 * For fs requests, rq is just carrier of independent bio's
2588 * and each partial completion should be handled separately.
2589 * Reset per-request error on each partial completion.
2590 *
2591 * TODO: tj: This is too subtle. It would be better to let
2592 * low level drivers do what they see fit.
2593 */
2594 if (req->cmd_type == REQ_TYPE_FS)
2595 req->errors = 0;
2596
2597 if (error && req->cmd_type == REQ_TYPE_FS &&
2598 !(req->cmd_flags & REQ_QUIET)) {
2599 char *error_type;
2600
2601 switch (error) {
2602 case -ENOLINK:
2603 error_type = "recoverable transport";
2604 break;
2605 case -EREMOTEIO:
2606 error_type = "critical target";
2607 break;
2608 case -EBADE:
2609 error_type = "critical nexus";
2610 break;
2611 case -ETIMEDOUT:
2612 error_type = "timeout";
2613 break;
2614 case -ENOSPC:
2615 error_type = "critical space allocation";
2616 break;
2617 case -ENODATA:
2618 error_type = "critical medium";
2619 break;
2620 case -EIO:
2621 default:
2622 error_type = "I/O";
2623 break;
2624 }
2625 printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n",
2626 __func__, error_type, req->rq_disk ?
2627 req->rq_disk->disk_name : "?",
2628 (unsigned long long)blk_rq_pos(req));
2629
2630 }
2631
2632 blk_account_io_completion(req, nr_bytes);
2633
2634 total_bytes = 0;
2635 while (req->bio) {
2636 struct bio *bio = req->bio;
2637 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
2638
2639 if (bio_bytes == bio->bi_iter.bi_size)
2640 req->bio = bio->bi_next;
2641
2642 req_bio_endio(req, bio, bio_bytes, error);
2643
2644 total_bytes += bio_bytes;
2645 nr_bytes -= bio_bytes;
2646
2647 if (!nr_bytes)
2648 break;
2649 }
2650
2651 /*
2652 * completely done
2653 */
2654 if (!req->bio) {
2655 /*
2656 * Reset counters so that the request stacking driver
2657 * can find how many bytes remain in the request
2658 * later.
2659 */
2660 req->__data_len = 0;
2661 return false;
2662 }
2663
2664 req->__data_len -= total_bytes;
2665
2666 /* update sector only for requests with clear definition of sector */
2667 if (req->cmd_type == REQ_TYPE_FS)
2668 req->__sector += total_bytes >> 9;
2669
2670 /* mixed attributes always follow the first bio */
2671 if (req->cmd_flags & REQ_MIXED_MERGE) {
2672 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2673 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2674 }
2675
2676 /*
2677 * If total number of sectors is less than the first segment
2678 * size, something has gone terribly wrong.
2679 */
2680 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2681 blk_dump_rq_flags(req, "request botched");
2682 req->__data_len = blk_rq_cur_bytes(req);
2683 }
2684
2685 /* recalculate the number of segments */
2686 blk_recalc_rq_segments(req);
2687
2688 return true;
2689}
2690EXPORT_SYMBOL_GPL(blk_update_request);
2691
2692static bool blk_update_bidi_request(struct request *rq, int error,
2693 unsigned int nr_bytes,
2694 unsigned int bidi_bytes)
2695{
2696 if (blk_update_request(rq, error, nr_bytes))
2697 return true;
2698
2699 /* Bidi request must be completed as a whole */
2700 if (unlikely(blk_bidi_rq(rq)) &&
2701 blk_update_request(rq->next_rq, error, bidi_bytes))
2702 return true;
2703
2704 if (blk_queue_add_random(rq->q))
2705 add_disk_randomness(rq->rq_disk);
2706
2707 return false;
2708}
2709
2710/**
2711 * blk_unprep_request - unprepare a request
2712 * @req: the request
2713 *
2714 * This function makes a request ready for complete resubmission (or
2715 * completion). It happens only after all error handling is complete,
2716 * so represents the appropriate moment to deallocate any resources
2717 * that were allocated to the request in the prep_rq_fn. The queue
2718 * lock is held when calling this.
2719 */
2720void blk_unprep_request(struct request *req)
2721{
2722 struct request_queue *q = req->q;
2723
2724 req->cmd_flags &= ~REQ_DONTPREP;
2725 if (q->unprep_rq_fn)
2726 q->unprep_rq_fn(q, req);
2727}
2728EXPORT_SYMBOL_GPL(blk_unprep_request);
2729
2730/*
2731 * queue lock must be held
2732 */
2733void blk_finish_request(struct request *req, int error)
2734{
2735 if (req->cmd_flags & REQ_QUEUED)
2736 blk_queue_end_tag(req->q, req);
2737
2738 BUG_ON(blk_queued_rq(req));
2739
2740 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2741 laptop_io_completion(&req->q->backing_dev_info);
2742
2743 blk_delete_timer(req);
2744
2745 if (req->cmd_flags & REQ_DONTPREP)
2746 blk_unprep_request(req);
2747
2748 blk_account_io_done(req);
2749
2750 if (req->end_io)
2751 req->end_io(req, error);
2752 else {
2753 if (blk_bidi_rq(req))
2754 __blk_put_request(req->next_rq->q, req->next_rq);
2755
2756 __blk_put_request(req->q, req);
2757 }
2758}
2759EXPORT_SYMBOL(blk_finish_request);
2760
2761/**
2762 * blk_end_bidi_request - Complete a bidi request
2763 * @rq: the request to complete
2764 * @error: %0 for success, < %0 for error
2765 * @nr_bytes: number of bytes to complete @rq
2766 * @bidi_bytes: number of bytes to complete @rq->next_rq
2767 *
2768 * Description:
2769 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2770 * Drivers that supports bidi can safely call this member for any
2771 * type of request, bidi or uni. In the later case @bidi_bytes is
2772 * just ignored.
2773 *
2774 * Return:
2775 * %false - we are done with this request
2776 * %true - still buffers pending for this request
2777 **/
2778static bool blk_end_bidi_request(struct request *rq, int error,
2779 unsigned int nr_bytes, unsigned int bidi_bytes)
2780{
2781 struct request_queue *q = rq->q;
2782 unsigned long flags;
2783
2784 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2785 return true;
2786
2787 spin_lock_irqsave(q->queue_lock, flags);
2788 blk_finish_request(rq, error);
2789 spin_unlock_irqrestore(q->queue_lock, flags);
2790
2791 return false;
2792}
2793
2794/**
2795 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2796 * @rq: the request to complete
2797 * @error: %0 for success, < %0 for error
2798 * @nr_bytes: number of bytes to complete @rq
2799 * @bidi_bytes: number of bytes to complete @rq->next_rq
2800 *
2801 * Description:
2802 * Identical to blk_end_bidi_request() except that queue lock is
2803 * assumed to be locked on entry and remains so on return.
2804 *
2805 * Return:
2806 * %false - we are done with this request
2807 * %true - still buffers pending for this request
2808 **/
2809bool __blk_end_bidi_request(struct request *rq, int error,
2810 unsigned int nr_bytes, unsigned int bidi_bytes)
2811{
2812 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2813 return true;
2814
2815 blk_finish_request(rq, error);
2816
2817 return false;
2818}
2819
2820/**
2821 * blk_end_request - Helper function for drivers to complete the request.
2822 * @rq: the request being processed
2823 * @error: %0 for success, < %0 for error
2824 * @nr_bytes: number of bytes to complete
2825 *
2826 * Description:
2827 * Ends I/O on a number of bytes attached to @rq.
2828 * If @rq has leftover, sets it up for the next range of segments.
2829 *
2830 * Return:
2831 * %false - we are done with this request
2832 * %true - still buffers pending for this request
2833 **/
2834bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2835{
2836 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2837}
2838EXPORT_SYMBOL(blk_end_request);
2839
2840/**
2841 * blk_end_request_all - Helper function for drives to finish the request.
2842 * @rq: the request to finish
2843 * @error: %0 for success, < %0 for error
2844 *
2845 * Description:
2846 * Completely finish @rq.
2847 */
2848void blk_end_request_all(struct request *rq, int error)
2849{
2850 bool pending;
2851 unsigned int bidi_bytes = 0;
2852
2853 if (unlikely(blk_bidi_rq(rq)))
2854 bidi_bytes = blk_rq_bytes(rq->next_rq);
2855
2856 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2857 BUG_ON(pending);
2858}
2859EXPORT_SYMBOL(blk_end_request_all);
2860
2861/**
2862 * blk_end_request_cur - Helper function to finish the current request chunk.
2863 * @rq: the request to finish the current chunk for
2864 * @error: %0 for success, < %0 for error
2865 *
2866 * Description:
2867 * Complete the current consecutively mapped chunk from @rq.
2868 *
2869 * Return:
2870 * %false - we are done with this request
2871 * %true - still buffers pending for this request
2872 */
2873bool blk_end_request_cur(struct request *rq, int error)
2874{
2875 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2876}
2877EXPORT_SYMBOL(blk_end_request_cur);
2878
2879/**
2880 * blk_end_request_err - Finish a request till the next failure boundary.
2881 * @rq: the request to finish till the next failure boundary for
2882 * @error: must be negative errno
2883 *
2884 * Description:
2885 * Complete @rq till the next failure boundary.
2886 *
2887 * Return:
2888 * %false - we are done with this request
2889 * %true - still buffers pending for this request
2890 */
2891bool blk_end_request_err(struct request *rq, int error)
2892{
2893 WARN_ON(error >= 0);
2894 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2895}
2896EXPORT_SYMBOL_GPL(blk_end_request_err);
2897
2898/**
2899 * __blk_end_request - Helper function for drivers to complete the request.
2900 * @rq: the request being processed
2901 * @error: %0 for success, < %0 for error
2902 * @nr_bytes: number of bytes to complete
2903 *
2904 * Description:
2905 * Must be called with queue lock held unlike blk_end_request().
2906 *
2907 * Return:
2908 * %false - we are done with this request
2909 * %true - still buffers pending for this request
2910 **/
2911bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2912{
2913 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2914}
2915EXPORT_SYMBOL(__blk_end_request);
2916
2917/**
2918 * __blk_end_request_all - Helper function for drives to finish the request.
2919 * @rq: the request to finish
2920 * @error: %0 for success, < %0 for error
2921 *
2922 * Description:
2923 * Completely finish @rq. Must be called with queue lock held.
2924 */
2925void __blk_end_request_all(struct request *rq, int error)
2926{
2927 bool pending;
2928 unsigned int bidi_bytes = 0;
2929
2930 if (unlikely(blk_bidi_rq(rq)))
2931 bidi_bytes = blk_rq_bytes(rq->next_rq);
2932
2933 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2934 BUG_ON(pending);
2935}
2936EXPORT_SYMBOL(__blk_end_request_all);
2937
2938/**
2939 * __blk_end_request_cur - Helper function to finish the current request chunk.
2940 * @rq: the request to finish the current chunk for
2941 * @error: %0 for success, < %0 for error
2942 *
2943 * Description:
2944 * Complete the current consecutively mapped chunk from @rq. Must
2945 * be called with queue lock held.
2946 *
2947 * Return:
2948 * %false - we are done with this request
2949 * %true - still buffers pending for this request
2950 */
2951bool __blk_end_request_cur(struct request *rq, int error)
2952{
2953 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2954}
2955EXPORT_SYMBOL(__blk_end_request_cur);
2956
2957/**
2958 * __blk_end_request_err - Finish a request till the next failure boundary.
2959 * @rq: the request to finish till the next failure boundary for
2960 * @error: must be negative errno
2961 *
2962 * Description:
2963 * Complete @rq till the next failure boundary. Must be called
2964 * with queue lock held.
2965 *
2966 * Return:
2967 * %false - we are done with this request
2968 * %true - still buffers pending for this request
2969 */
2970bool __blk_end_request_err(struct request *rq, int error)
2971{
2972 WARN_ON(error >= 0);
2973 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2974}
2975EXPORT_SYMBOL_GPL(__blk_end_request_err);
2976
2977void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2978 struct bio *bio)
2979{
2980 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2981 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
2982
2983 if (bio_has_data(bio))
2984 rq->nr_phys_segments = bio_phys_segments(q, bio);
2985
2986 rq->__data_len = bio->bi_iter.bi_size;
2987 rq->bio = rq->biotail = bio;
2988
2989 if (bio->bi_bdev)
2990 rq->rq_disk = bio->bi_bdev->bd_disk;
2991}
2992
2993#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2994/**
2995 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2996 * @rq: the request to be flushed
2997 *
2998 * Description:
2999 * Flush all pages in @rq.
3000 */
3001void rq_flush_dcache_pages(struct request *rq)
3002{
3003 struct req_iterator iter;
3004 struct bio_vec bvec;
3005
3006 rq_for_each_segment(bvec, rq, iter)
3007 flush_dcache_page(bvec.bv_page);
3008}
3009EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
3010#endif
3011
3012/**
3013 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
3014 * @q : the queue of the device being checked
3015 *
3016 * Description:
3017 * Check if underlying low-level drivers of a device are busy.
3018 * If the drivers want to export their busy state, they must set own
3019 * exporting function using blk_queue_lld_busy() first.
3020 *
3021 * Basically, this function is used only by request stacking drivers
3022 * to stop dispatching requests to underlying devices when underlying
3023 * devices are busy. This behavior helps more I/O merging on the queue
3024 * of the request stacking driver and prevents I/O throughput regression
3025 * on burst I/O load.
3026 *
3027 * Return:
3028 * 0 - Not busy (The request stacking driver should dispatch request)
3029 * 1 - Busy (The request stacking driver should stop dispatching request)
3030 */
3031int blk_lld_busy(struct request_queue *q)
3032{
3033 if (q->lld_busy_fn)
3034 return q->lld_busy_fn(q);
3035
3036 return 0;
3037}
3038EXPORT_SYMBOL_GPL(blk_lld_busy);
3039
3040/**
3041 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
3042 * @rq: the clone request to be cleaned up
3043 *
3044 * Description:
3045 * Free all bios in @rq for a cloned request.
3046 */
3047void blk_rq_unprep_clone(struct request *rq)
3048{
3049 struct bio *bio;
3050
3051 while ((bio = rq->bio) != NULL) {
3052 rq->bio = bio->bi_next;
3053
3054 bio_put(bio);
3055 }
3056}
3057EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
3058
3059/*
3060 * Copy attributes of the original request to the clone request.
3061 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3062 */
3063static void __blk_rq_prep_clone(struct request *dst, struct request *src)
3064{
3065 dst->cpu = src->cpu;
3066 dst->cmd_flags |= (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
3067 dst->cmd_type = src->cmd_type;
3068 dst->__sector = blk_rq_pos(src);
3069 dst->__data_len = blk_rq_bytes(src);
3070 dst->nr_phys_segments = src->nr_phys_segments;
3071 dst->ioprio = src->ioprio;
3072 dst->extra_len = src->extra_len;
3073}
3074
3075/**
3076 * blk_rq_prep_clone - Helper function to setup clone request
3077 * @rq: the request to be setup
3078 * @rq_src: original request to be cloned
3079 * @bs: bio_set that bios for clone are allocated from
3080 * @gfp_mask: memory allocation mask for bio
3081 * @bio_ctr: setup function to be called for each clone bio.
3082 * Returns %0 for success, non %0 for failure.
3083 * @data: private data to be passed to @bio_ctr
3084 *
3085 * Description:
3086 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3087 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3088 * are not copied, and copying such parts is the caller's responsibility.
3089 * Also, pages which the original bios are pointing to are not copied
3090 * and the cloned bios just point same pages.
3091 * So cloned bios must be completed before original bios, which means
3092 * the caller must complete @rq before @rq_src.
3093 */
3094int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
3095 struct bio_set *bs, gfp_t gfp_mask,
3096 int (*bio_ctr)(struct bio *, struct bio *, void *),
3097 void *data)
3098{
3099 struct bio *bio, *bio_src;
3100
3101 if (!bs)
3102 bs = fs_bio_set;
3103
3104 __rq_for_each_bio(bio_src, rq_src) {
3105 bio = bio_clone_fast(bio_src, gfp_mask, bs);
3106 if (!bio)
3107 goto free_and_out;
3108
3109 if (bio_ctr && bio_ctr(bio, bio_src, data))
3110 goto free_and_out;
3111
3112 if (rq->bio) {
3113 rq->biotail->bi_next = bio;
3114 rq->biotail = bio;
3115 } else
3116 rq->bio = rq->biotail = bio;
3117 }
3118
3119 __blk_rq_prep_clone(rq, rq_src);
3120
3121 return 0;
3122
3123free_and_out:
3124 if (bio)
3125 bio_put(bio);
3126 blk_rq_unprep_clone(rq);
3127
3128 return -ENOMEM;
3129}
3130EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
3131
3132int kblockd_schedule_work(struct work_struct *work)
3133{
3134 return queue_work(kblockd_workqueue, work);
3135}
3136EXPORT_SYMBOL(kblockd_schedule_work);
3137
3138int kblockd_schedule_delayed_work(struct delayed_work *dwork,
3139 unsigned long delay)
3140{
3141 return queue_delayed_work(kblockd_workqueue, dwork, delay);
3142}
3143EXPORT_SYMBOL(kblockd_schedule_delayed_work);
3144
3145int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork,
3146 unsigned long delay)
3147{
3148 return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3149}
3150EXPORT_SYMBOL(kblockd_schedule_delayed_work_on);
3151
3152/**
3153 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3154 * @plug: The &struct blk_plug that needs to be initialized
3155 *
3156 * Description:
3157 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3158 * pending I/O should the task end up blocking between blk_start_plug() and
3159 * blk_finish_plug(). This is important from a performance perspective, but
3160 * also ensures that we don't deadlock. For instance, if the task is blocking
3161 * for a memory allocation, memory reclaim could end up wanting to free a
3162 * page belonging to that request that is currently residing in our private
3163 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3164 * this kind of deadlock.
3165 */
3166void blk_start_plug(struct blk_plug *plug)
3167{
3168 struct task_struct *tsk = current;
3169
3170 /*
3171 * If this is a nested plug, don't actually assign it.
3172 */
3173 if (tsk->plug)
3174 return;
3175
3176 INIT_LIST_HEAD(&plug->list);
3177 INIT_LIST_HEAD(&plug->mq_list);
3178 INIT_LIST_HEAD(&plug->cb_list);
3179 /*
3180 * Store ordering should not be needed here, since a potential
3181 * preempt will imply a full memory barrier
3182 */
3183 tsk->plug = plug;
3184}
3185EXPORT_SYMBOL(blk_start_plug);
3186
3187static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
3188{
3189 struct request *rqa = container_of(a, struct request, queuelist);
3190 struct request *rqb = container_of(b, struct request, queuelist);
3191
3192 return !(rqa->q < rqb->q ||
3193 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
3194}
3195
3196/*
3197 * If 'from_schedule' is true, then postpone the dispatch of requests
3198 * until a safe kblockd context. We due this to avoid accidental big
3199 * additional stack usage in driver dispatch, in places where the originally
3200 * plugger did not intend it.
3201 */
3202static void queue_unplugged(struct request_queue *q, unsigned int depth,
3203 bool from_schedule)
3204 __releases(q->queue_lock)
3205{
3206 trace_block_unplug(q, depth, !from_schedule);
3207
3208 if (from_schedule)
3209 blk_run_queue_async(q);
3210 else
3211 __blk_run_queue(q);
3212 spin_unlock(q->queue_lock);
3213}
3214
3215static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
3216{
3217 LIST_HEAD(callbacks);
3218
3219 while (!list_empty(&plug->cb_list)) {
3220 list_splice_init(&plug->cb_list, &callbacks);
3221
3222 while (!list_empty(&callbacks)) {
3223 struct blk_plug_cb *cb = list_first_entry(&callbacks,
3224 struct blk_plug_cb,
3225 list);
3226 list_del(&cb->list);
3227 cb->callback(cb, from_schedule);
3228 }
3229 }
3230}
3231
3232struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3233 int size)
3234{
3235 struct blk_plug *plug = current->plug;
3236 struct blk_plug_cb *cb;
3237
3238 if (!plug)
3239 return NULL;
3240
3241 list_for_each_entry(cb, &plug->cb_list, list)
3242 if (cb->callback == unplug && cb->data == data)
3243 return cb;
3244
3245 /* Not currently on the callback list */
3246 BUG_ON(size < sizeof(*cb));
3247 cb = kzalloc(size, GFP_ATOMIC);
3248 if (cb) {
3249 cb->data = data;
3250 cb->callback = unplug;
3251 list_add(&cb->list, &plug->cb_list);
3252 }
3253 return cb;
3254}
3255EXPORT_SYMBOL(blk_check_plugged);
3256
3257void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
3258{
3259 struct request_queue *q;
3260 unsigned long flags;
3261 struct request *rq;
3262 LIST_HEAD(list);
3263 unsigned int depth;
3264
3265 flush_plug_callbacks(plug, from_schedule);
3266
3267 if (!list_empty(&plug->mq_list))
3268 blk_mq_flush_plug_list(plug, from_schedule);
3269
3270 if (list_empty(&plug->list))
3271 return;
3272
3273 list_splice_init(&plug->list, &list);
3274
3275 list_sort(NULL, &list, plug_rq_cmp);
3276
3277 q = NULL;
3278 depth = 0;
3279
3280 /*
3281 * Save and disable interrupts here, to avoid doing it for every
3282 * queue lock we have to take.
3283 */
3284 local_irq_save(flags);
3285 while (!list_empty(&list)) {
3286 rq = list_entry_rq(list.next);
3287 list_del_init(&rq->queuelist);
3288 BUG_ON(!rq->q);
3289 if (rq->q != q) {
3290 /*
3291 * This drops the queue lock
3292 */
3293 if (q)
3294 queue_unplugged(q, depth, from_schedule);
3295 q = rq->q;
3296 depth = 0;
3297 spin_lock(q->queue_lock);
3298 }
3299
3300 /*
3301 * Short-circuit if @q is dead
3302 */
3303 if (unlikely(blk_queue_dying(q))) {
3304 __blk_end_request_all(rq, -ENODEV);
3305 continue;
3306 }
3307
3308 /*
3309 * rq is already accounted, so use raw insert
3310 */
3311 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
3312 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3313 else
3314 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
3315
3316 depth++;
3317 }
3318
3319 /*
3320 * This drops the queue lock
3321 */
3322 if (q)
3323 queue_unplugged(q, depth, from_schedule);
3324
3325 local_irq_restore(flags);
3326}
3327
3328void blk_finish_plug(struct blk_plug *plug)
3329{
3330 if (plug != current->plug)
3331 return;
3332 blk_flush_plug_list(plug, false);
3333
3334 current->plug = NULL;
3335}
3336EXPORT_SYMBOL(blk_finish_plug);
3337
3338bool blk_poll(struct request_queue *q, blk_qc_t cookie)
3339{
3340 struct blk_plug *plug;
3341 long state;
3342
3343 if (!q->mq_ops || !q->mq_ops->poll || !blk_qc_t_valid(cookie) ||
3344 !test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
3345 return false;
3346
3347 plug = current->plug;
3348 if (plug)
3349 blk_flush_plug_list(plug, false);
3350
3351 state = current->state;
3352 while (!need_resched()) {
3353 unsigned int queue_num = blk_qc_t_to_queue_num(cookie);
3354 struct blk_mq_hw_ctx *hctx = q->queue_hw_ctx[queue_num];
3355 int ret;
3356
3357 hctx->poll_invoked++;
3358
3359 ret = q->mq_ops->poll(hctx, blk_qc_t_to_tag(cookie));
3360 if (ret > 0) {
3361 hctx->poll_success++;
3362 set_current_state(TASK_RUNNING);
3363 return true;
3364 }
3365
3366 if (signal_pending_state(state, current))
3367 set_current_state(TASK_RUNNING);
3368
3369 if (current->state == TASK_RUNNING)
3370 return true;
3371 if (ret < 0)
3372 break;
3373 cpu_relax();
3374 }
3375
3376 return false;
3377}
3378
3379#ifdef CONFIG_PM
3380/**
3381 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3382 * @q: the queue of the device
3383 * @dev: the device the queue belongs to
3384 *
3385 * Description:
3386 * Initialize runtime-PM-related fields for @q and start auto suspend for
3387 * @dev. Drivers that want to take advantage of request-based runtime PM
3388 * should call this function after @dev has been initialized, and its
3389 * request queue @q has been allocated, and runtime PM for it can not happen
3390 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3391 * cases, driver should call this function before any I/O has taken place.
3392 *
3393 * This function takes care of setting up using auto suspend for the device,
3394 * the autosuspend delay is set to -1 to make runtime suspend impossible
3395 * until an updated value is either set by user or by driver. Drivers do
3396 * not need to touch other autosuspend settings.
3397 *
3398 * The block layer runtime PM is request based, so only works for drivers
3399 * that use request as their IO unit instead of those directly use bio's.
3400 */
3401void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3402{
3403 q->dev = dev;
3404 q->rpm_status = RPM_ACTIVE;
3405 pm_runtime_set_autosuspend_delay(q->dev, -1);
3406 pm_runtime_use_autosuspend(q->dev);
3407}
3408EXPORT_SYMBOL(blk_pm_runtime_init);
3409
3410/**
3411 * blk_pre_runtime_suspend - Pre runtime suspend check
3412 * @q: the queue of the device
3413 *
3414 * Description:
3415 * This function will check if runtime suspend is allowed for the device
3416 * by examining if there are any requests pending in the queue. If there
3417 * are requests pending, the device can not be runtime suspended; otherwise,
3418 * the queue's status will be updated to SUSPENDING and the driver can
3419 * proceed to suspend the device.
3420 *
3421 * For the not allowed case, we mark last busy for the device so that
3422 * runtime PM core will try to autosuspend it some time later.
3423 *
3424 * This function should be called near the start of the device's
3425 * runtime_suspend callback.
3426 *
3427 * Return:
3428 * 0 - OK to runtime suspend the device
3429 * -EBUSY - Device should not be runtime suspended
3430 */
3431int blk_pre_runtime_suspend(struct request_queue *q)
3432{
3433 int ret = 0;
3434
3435 if (!q->dev)
3436 return ret;
3437
3438 spin_lock_irq(q->queue_lock);
3439 if (q->nr_pending) {
3440 ret = -EBUSY;
3441 pm_runtime_mark_last_busy(q->dev);
3442 } else {
3443 q->rpm_status = RPM_SUSPENDING;
3444 }
3445 spin_unlock_irq(q->queue_lock);
3446 return ret;
3447}
3448EXPORT_SYMBOL(blk_pre_runtime_suspend);
3449
3450/**
3451 * blk_post_runtime_suspend - Post runtime suspend processing
3452 * @q: the queue of the device
3453 * @err: return value of the device's runtime_suspend function
3454 *
3455 * Description:
3456 * Update the queue's runtime status according to the return value of the
3457 * device's runtime suspend function and mark last busy for the device so
3458 * that PM core will try to auto suspend the device at a later time.
3459 *
3460 * This function should be called near the end of the device's
3461 * runtime_suspend callback.
3462 */
3463void blk_post_runtime_suspend(struct request_queue *q, int err)
3464{
3465 if (!q->dev)
3466 return;
3467
3468 spin_lock_irq(q->queue_lock);
3469 if (!err) {
3470 q->rpm_status = RPM_SUSPENDED;
3471 } else {
3472 q->rpm_status = RPM_ACTIVE;
3473 pm_runtime_mark_last_busy(q->dev);
3474 }
3475 spin_unlock_irq(q->queue_lock);
3476}
3477EXPORT_SYMBOL(blk_post_runtime_suspend);
3478
3479/**
3480 * blk_pre_runtime_resume - Pre runtime resume processing
3481 * @q: the queue of the device
3482 *
3483 * Description:
3484 * Update the queue's runtime status to RESUMING in preparation for the
3485 * runtime resume of the device.
3486 *
3487 * This function should be called near the start of the device's
3488 * runtime_resume callback.
3489 */
3490void blk_pre_runtime_resume(struct request_queue *q)
3491{
3492 if (!q->dev)
3493 return;
3494
3495 spin_lock_irq(q->queue_lock);
3496 q->rpm_status = RPM_RESUMING;
3497 spin_unlock_irq(q->queue_lock);
3498}
3499EXPORT_SYMBOL(blk_pre_runtime_resume);
3500
3501/**
3502 * blk_post_runtime_resume - Post runtime resume processing
3503 * @q: the queue of the device
3504 * @err: return value of the device's runtime_resume function
3505 *
3506 * Description:
3507 * Update the queue's runtime status according to the return value of the
3508 * device's runtime_resume function. If it is successfully resumed, process
3509 * the requests that are queued into the device's queue when it is resuming
3510 * and then mark last busy and initiate autosuspend for it.
3511 *
3512 * This function should be called near the end of the device's
3513 * runtime_resume callback.
3514 */
3515void blk_post_runtime_resume(struct request_queue *q, int err)
3516{
3517 if (!q->dev)
3518 return;
3519
3520 spin_lock_irq(q->queue_lock);
3521 if (!err) {
3522 q->rpm_status = RPM_ACTIVE;
3523 __blk_run_queue(q);
3524 pm_runtime_mark_last_busy(q->dev);
3525 pm_request_autosuspend(q->dev);
3526 } else {
3527 q->rpm_status = RPM_SUSPENDED;
3528 }
3529 spin_unlock_irq(q->queue_lock);
3530}
3531EXPORT_SYMBOL(blk_post_runtime_resume);
3532
3533/**
3534 * blk_set_runtime_active - Force runtime status of the queue to be active
3535 * @q: the queue of the device
3536 *
3537 * If the device is left runtime suspended during system suspend the resume
3538 * hook typically resumes the device and corrects runtime status
3539 * accordingly. However, that does not affect the queue runtime PM status
3540 * which is still "suspended". This prevents processing requests from the
3541 * queue.
3542 *
3543 * This function can be used in driver's resume hook to correct queue
3544 * runtime PM status and re-enable peeking requests from the queue. It
3545 * should be called before first request is added to the queue.
3546 */
3547void blk_set_runtime_active(struct request_queue *q)
3548{
3549 spin_lock_irq(q->queue_lock);
3550 q->rpm_status = RPM_ACTIVE;
3551 pm_runtime_mark_last_busy(q->dev);
3552 pm_request_autosuspend(q->dev);
3553 spin_unlock_irq(q->queue_lock);
3554}
3555EXPORT_SYMBOL(blk_set_runtime_active);
3556#endif
3557
3558int __init blk_dev_init(void)
3559{
3560 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
3561 FIELD_SIZEOF(struct request, cmd_flags));
3562
3563 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3564 kblockd_workqueue = alloc_workqueue("kblockd",
3565 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
3566 if (!kblockd_workqueue)
3567 panic("Failed to create kblockd\n");
3568
3569 request_cachep = kmem_cache_create("blkdev_requests",
3570 sizeof(struct request), 0, SLAB_PANIC, NULL);
3571
3572 blk_requestq_cachep = kmem_cache_create("request_queue",
3573 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
3574
3575 return 0;
3576}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) 1991, 1992 Linus Torvalds
4 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
5 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
6 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
7 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
8 * - July2000
9 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
10 */
11
12/*
13 * This handles all read/write requests to block devices
14 */
15#include <linux/kernel.h>
16#include <linux/module.h>
17#include <linux/backing-dev.h>
18#include <linux/bio.h>
19#include <linux/blkdev.h>
20#include <linux/blk-mq.h>
21#include <linux/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/blk-cgroup.h>
38#include <linux/t10-pi.h>
39#include <linux/debugfs.h>
40#include <linux/bpf.h>
41#include <linux/psi.h>
42#include <linux/sched/sysctl.h>
43#include <linux/blk-crypto.h>
44
45#define CREATE_TRACE_POINTS
46#include <trace/events/block.h>
47
48#include "blk.h"
49#include "blk-mq.h"
50#include "blk-mq-sched.h"
51#include "blk-pm.h"
52#include "blk-rq-qos.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);
61
62DEFINE_IDA(blk_queue_ida);
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 set_bit(flag, &q->queue_flags);
82}
83EXPORT_SYMBOL(blk_queue_flag_set);
84
85/**
86 * blk_queue_flag_clear - atomically clear a queue flag
87 * @flag: flag to be cleared
88 * @q: request queue
89 */
90void blk_queue_flag_clear(unsigned int flag, struct request_queue *q)
91{
92 clear_bit(flag, &q->queue_flags);
93}
94EXPORT_SYMBOL(blk_queue_flag_clear);
95
96/**
97 * blk_queue_flag_test_and_set - atomically test and set a queue flag
98 * @flag: flag to be set
99 * @q: request queue
100 *
101 * Returns the previous value of @flag - 0 if the flag was not set and 1 if
102 * the flag was already set.
103 */
104bool blk_queue_flag_test_and_set(unsigned int flag, struct request_queue *q)
105{
106 return test_and_set_bit(flag, &q->queue_flags);
107}
108EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set);
109
110void blk_rq_init(struct request_queue *q, struct request *rq)
111{
112 memset(rq, 0, sizeof(*rq));
113
114 INIT_LIST_HEAD(&rq->queuelist);
115 rq->q = q;
116 rq->__sector = (sector_t) -1;
117 INIT_HLIST_NODE(&rq->hash);
118 RB_CLEAR_NODE(&rq->rb_node);
119 rq->tag = -1;
120 rq->internal_tag = -1;
121 rq->start_time_ns = ktime_get_ns();
122 rq->part = NULL;
123 refcount_set(&rq->ref, 1);
124 blk_crypto_rq_set_defaults(rq);
125}
126EXPORT_SYMBOL(blk_rq_init);
127
128#define REQ_OP_NAME(name) [REQ_OP_##name] = #name
129static const char *const blk_op_name[] = {
130 REQ_OP_NAME(READ),
131 REQ_OP_NAME(WRITE),
132 REQ_OP_NAME(FLUSH),
133 REQ_OP_NAME(DISCARD),
134 REQ_OP_NAME(SECURE_ERASE),
135 REQ_OP_NAME(ZONE_RESET),
136 REQ_OP_NAME(ZONE_RESET_ALL),
137 REQ_OP_NAME(ZONE_OPEN),
138 REQ_OP_NAME(ZONE_CLOSE),
139 REQ_OP_NAME(ZONE_FINISH),
140 REQ_OP_NAME(ZONE_APPEND),
141 REQ_OP_NAME(WRITE_SAME),
142 REQ_OP_NAME(WRITE_ZEROES),
143 REQ_OP_NAME(SCSI_IN),
144 REQ_OP_NAME(SCSI_OUT),
145 REQ_OP_NAME(DRV_IN),
146 REQ_OP_NAME(DRV_OUT),
147};
148#undef REQ_OP_NAME
149
150/**
151 * blk_op_str - Return string XXX in the REQ_OP_XXX.
152 * @op: REQ_OP_XXX.
153 *
154 * Description: Centralize block layer function to convert REQ_OP_XXX into
155 * string format. Useful in the debugging and tracing bio or request. For
156 * invalid REQ_OP_XXX it returns string "UNKNOWN".
157 */
158inline const char *blk_op_str(unsigned int op)
159{
160 const char *op_str = "UNKNOWN";
161
162 if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op])
163 op_str = blk_op_name[op];
164
165 return op_str;
166}
167EXPORT_SYMBOL_GPL(blk_op_str);
168
169static const struct {
170 int errno;
171 const char *name;
172} blk_errors[] = {
173 [BLK_STS_OK] = { 0, "" },
174 [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" },
175 [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" },
176 [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" },
177 [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" },
178 [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" },
179 [BLK_STS_NEXUS] = { -EBADE, "critical nexus" },
180 [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" },
181 [BLK_STS_PROTECTION] = { -EILSEQ, "protection" },
182 [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" },
183 [BLK_STS_DEV_RESOURCE] = { -EBUSY, "device resource" },
184 [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" },
185
186 /* device mapper special case, should not leak out: */
187 [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" },
188
189 /* everything else not covered above: */
190 [BLK_STS_IOERR] = { -EIO, "I/O" },
191};
192
193blk_status_t errno_to_blk_status(int errno)
194{
195 int i;
196
197 for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
198 if (blk_errors[i].errno == errno)
199 return (__force blk_status_t)i;
200 }
201
202 return BLK_STS_IOERR;
203}
204EXPORT_SYMBOL_GPL(errno_to_blk_status);
205
206int blk_status_to_errno(blk_status_t status)
207{
208 int idx = (__force int)status;
209
210 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
211 return -EIO;
212 return blk_errors[idx].errno;
213}
214EXPORT_SYMBOL_GPL(blk_status_to_errno);
215
216static void print_req_error(struct request *req, blk_status_t status,
217 const char *caller)
218{
219 int idx = (__force int)status;
220
221 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
222 return;
223
224 printk_ratelimited(KERN_ERR
225 "%s: %s error, dev %s, sector %llu op 0x%x:(%s) flags 0x%x "
226 "phys_seg %u prio class %u\n",
227 caller, blk_errors[idx].name,
228 req->rq_disk ? req->rq_disk->disk_name : "?",
229 blk_rq_pos(req), req_op(req), blk_op_str(req_op(req)),
230 req->cmd_flags & ~REQ_OP_MASK,
231 req->nr_phys_segments,
232 IOPRIO_PRIO_CLASS(req->ioprio));
233}
234
235static void req_bio_endio(struct request *rq, struct bio *bio,
236 unsigned int nbytes, blk_status_t error)
237{
238 if (error)
239 bio->bi_status = error;
240
241 if (unlikely(rq->rq_flags & RQF_QUIET))
242 bio_set_flag(bio, BIO_QUIET);
243
244 bio_advance(bio, nbytes);
245
246 if (req_op(rq) == REQ_OP_ZONE_APPEND && error == BLK_STS_OK) {
247 /*
248 * Partial zone append completions cannot be supported as the
249 * BIO fragments may end up not being written sequentially.
250 */
251 if (bio->bi_iter.bi_size)
252 bio->bi_status = BLK_STS_IOERR;
253 else
254 bio->bi_iter.bi_sector = rq->__sector;
255 }
256
257 /* don't actually finish bio if it's part of flush sequence */
258 if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ))
259 bio_endio(bio);
260}
261
262void blk_dump_rq_flags(struct request *rq, char *msg)
263{
264 printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg,
265 rq->rq_disk ? rq->rq_disk->disk_name : "?",
266 (unsigned long long) rq->cmd_flags);
267
268 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
269 (unsigned long long)blk_rq_pos(rq),
270 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
271 printk(KERN_INFO " bio %p, biotail %p, len %u\n",
272 rq->bio, rq->biotail, blk_rq_bytes(rq));
273}
274EXPORT_SYMBOL(blk_dump_rq_flags);
275
276/**
277 * blk_sync_queue - cancel any pending callbacks on a queue
278 * @q: the queue
279 *
280 * Description:
281 * The block layer may perform asynchronous callback activity
282 * on a queue, such as calling the unplug function after a timeout.
283 * A block device may call blk_sync_queue to ensure that any
284 * such activity is cancelled, thus allowing it to release resources
285 * that the callbacks might use. The caller must already have made sure
286 * that its ->submit_bio will not re-add plugging prior to calling
287 * this function.
288 *
289 * This function does not cancel any asynchronous activity arising
290 * out of elevator or throttling code. That would require elevator_exit()
291 * and blkcg_exit_queue() to be called with queue lock initialized.
292 *
293 */
294void blk_sync_queue(struct request_queue *q)
295{
296 del_timer_sync(&q->timeout);
297 cancel_work_sync(&q->timeout_work);
298}
299EXPORT_SYMBOL(blk_sync_queue);
300
301/**
302 * blk_set_pm_only - increment pm_only counter
303 * @q: request queue pointer
304 */
305void blk_set_pm_only(struct request_queue *q)
306{
307 atomic_inc(&q->pm_only);
308}
309EXPORT_SYMBOL_GPL(blk_set_pm_only);
310
311void blk_clear_pm_only(struct request_queue *q)
312{
313 int pm_only;
314
315 pm_only = atomic_dec_return(&q->pm_only);
316 WARN_ON_ONCE(pm_only < 0);
317 if (pm_only == 0)
318 wake_up_all(&q->mq_freeze_wq);
319}
320EXPORT_SYMBOL_GPL(blk_clear_pm_only);
321
322/**
323 * blk_put_queue - decrement the request_queue refcount
324 * @q: the request_queue structure to decrement the refcount for
325 *
326 * Decrements the refcount of the request_queue kobject. When this reaches 0
327 * we'll have blk_release_queue() called.
328 *
329 * Context: Any context, but the last reference must not be dropped from
330 * atomic context.
331 */
332void blk_put_queue(struct request_queue *q)
333{
334 kobject_put(&q->kobj);
335}
336EXPORT_SYMBOL(blk_put_queue);
337
338void blk_set_queue_dying(struct request_queue *q)
339{
340 blk_queue_flag_set(QUEUE_FLAG_DYING, q);
341
342 /*
343 * When queue DYING flag is set, we need to block new req
344 * entering queue, so we call blk_freeze_queue_start() to
345 * prevent I/O from crossing blk_queue_enter().
346 */
347 blk_freeze_queue_start(q);
348
349 if (queue_is_mq(q))
350 blk_mq_wake_waiters(q);
351
352 /* Make blk_queue_enter() reexamine the DYING flag. */
353 wake_up_all(&q->mq_freeze_wq);
354}
355EXPORT_SYMBOL_GPL(blk_set_queue_dying);
356
357/**
358 * blk_cleanup_queue - shutdown a request queue
359 * @q: request queue to shutdown
360 *
361 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
362 * put it. All future requests will be failed immediately with -ENODEV.
363 *
364 * Context: can sleep
365 */
366void blk_cleanup_queue(struct request_queue *q)
367{
368 /* cannot be called from atomic context */
369 might_sleep();
370
371 WARN_ON_ONCE(blk_queue_registered(q));
372
373 /* mark @q DYING, no new request or merges will be allowed afterwards */
374 blk_set_queue_dying(q);
375
376 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q);
377 blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
378
379 /*
380 * Drain all requests queued before DYING marking. Set DEAD flag to
381 * prevent that blk_mq_run_hw_queues() accesses the hardware queues
382 * after draining finished.
383 */
384 blk_freeze_queue(q);
385
386 rq_qos_exit(q);
387
388 blk_queue_flag_set(QUEUE_FLAG_DEAD, q);
389
390 /* for synchronous bio-based driver finish in-flight integrity i/o */
391 blk_flush_integrity();
392
393 /* @q won't process any more request, flush async actions */
394 del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer);
395 blk_sync_queue(q);
396
397 if (queue_is_mq(q))
398 blk_mq_exit_queue(q);
399
400 /*
401 * In theory, request pool of sched_tags belongs to request queue.
402 * However, the current implementation requires tag_set for freeing
403 * requests, so free the pool now.
404 *
405 * Queue has become frozen, there can't be any in-queue requests, so
406 * it is safe to free requests now.
407 */
408 mutex_lock(&q->sysfs_lock);
409 if (q->elevator)
410 blk_mq_sched_free_requests(q);
411 mutex_unlock(&q->sysfs_lock);
412
413 percpu_ref_exit(&q->q_usage_counter);
414
415 /* @q is and will stay empty, shutdown and put */
416 blk_put_queue(q);
417}
418EXPORT_SYMBOL(blk_cleanup_queue);
419
420/**
421 * blk_queue_enter() - try to increase q->q_usage_counter
422 * @q: request queue pointer
423 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PREEMPT
424 */
425int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
426{
427 const bool pm = flags & BLK_MQ_REQ_PREEMPT;
428
429 while (true) {
430 bool success = false;
431
432 rcu_read_lock();
433 if (percpu_ref_tryget_live(&q->q_usage_counter)) {
434 /*
435 * The code that increments the pm_only counter is
436 * responsible for ensuring that that counter is
437 * globally visible before the queue is unfrozen.
438 */
439 if (pm || !blk_queue_pm_only(q)) {
440 success = true;
441 } else {
442 percpu_ref_put(&q->q_usage_counter);
443 }
444 }
445 rcu_read_unlock();
446
447 if (success)
448 return 0;
449
450 if (flags & BLK_MQ_REQ_NOWAIT)
451 return -EBUSY;
452
453 /*
454 * read pair of barrier in blk_freeze_queue_start(),
455 * we need to order reading __PERCPU_REF_DEAD flag of
456 * .q_usage_counter and reading .mq_freeze_depth or
457 * queue dying flag, otherwise the following wait may
458 * never return if the two reads are reordered.
459 */
460 smp_rmb();
461
462 wait_event(q->mq_freeze_wq,
463 (!q->mq_freeze_depth &&
464 (pm || (blk_pm_request_resume(q),
465 !blk_queue_pm_only(q)))) ||
466 blk_queue_dying(q));
467 if (blk_queue_dying(q))
468 return -ENODEV;
469 }
470}
471
472static inline int bio_queue_enter(struct bio *bio)
473{
474 struct request_queue *q = bio->bi_disk->queue;
475 bool nowait = bio->bi_opf & REQ_NOWAIT;
476 int ret;
477
478 ret = blk_queue_enter(q, nowait ? BLK_MQ_REQ_NOWAIT : 0);
479 if (unlikely(ret)) {
480 if (nowait && !blk_queue_dying(q))
481 bio_wouldblock_error(bio);
482 else
483 bio_io_error(bio);
484 }
485
486 return ret;
487}
488
489void blk_queue_exit(struct request_queue *q)
490{
491 percpu_ref_put(&q->q_usage_counter);
492}
493
494static void blk_queue_usage_counter_release(struct percpu_ref *ref)
495{
496 struct request_queue *q =
497 container_of(ref, struct request_queue, q_usage_counter);
498
499 wake_up_all(&q->mq_freeze_wq);
500}
501
502static void blk_rq_timed_out_timer(struct timer_list *t)
503{
504 struct request_queue *q = from_timer(q, t, timeout);
505
506 kblockd_schedule_work(&q->timeout_work);
507}
508
509static void blk_timeout_work(struct work_struct *work)
510{
511}
512
513struct request_queue *blk_alloc_queue(int node_id)
514{
515 struct request_queue *q;
516 int ret;
517
518 q = kmem_cache_alloc_node(blk_requestq_cachep,
519 GFP_KERNEL | __GFP_ZERO, node_id);
520 if (!q)
521 return NULL;
522
523 q->last_merge = NULL;
524
525 q->id = ida_simple_get(&blk_queue_ida, 0, 0, GFP_KERNEL);
526 if (q->id < 0)
527 goto fail_q;
528
529 ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
530 if (ret)
531 goto fail_id;
532
533 q->backing_dev_info = bdi_alloc(node_id);
534 if (!q->backing_dev_info)
535 goto fail_split;
536
537 q->stats = blk_alloc_queue_stats();
538 if (!q->stats)
539 goto fail_stats;
540
541 q->backing_dev_info->ra_pages = VM_READAHEAD_PAGES;
542 q->backing_dev_info->io_pages = VM_READAHEAD_PAGES;
543 q->backing_dev_info->capabilities = BDI_CAP_CGROUP_WRITEBACK;
544 q->node = node_id;
545
546 timer_setup(&q->backing_dev_info->laptop_mode_wb_timer,
547 laptop_mode_timer_fn, 0);
548 timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
549 INIT_WORK(&q->timeout_work, blk_timeout_work);
550 INIT_LIST_HEAD(&q->icq_list);
551#ifdef CONFIG_BLK_CGROUP
552 INIT_LIST_HEAD(&q->blkg_list);
553#endif
554
555 kobject_init(&q->kobj, &blk_queue_ktype);
556
557 mutex_init(&q->debugfs_mutex);
558 mutex_init(&q->sysfs_lock);
559 mutex_init(&q->sysfs_dir_lock);
560 spin_lock_init(&q->queue_lock);
561
562 init_waitqueue_head(&q->mq_freeze_wq);
563 mutex_init(&q->mq_freeze_lock);
564
565 /*
566 * Init percpu_ref in atomic mode so that it's faster to shutdown.
567 * See blk_register_queue() for details.
568 */
569 if (percpu_ref_init(&q->q_usage_counter,
570 blk_queue_usage_counter_release,
571 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
572 goto fail_bdi;
573
574 if (blkcg_init_queue(q))
575 goto fail_ref;
576
577 blk_queue_dma_alignment(q, 511);
578 blk_set_default_limits(&q->limits);
579 q->nr_requests = BLKDEV_MAX_RQ;
580
581 return q;
582
583fail_ref:
584 percpu_ref_exit(&q->q_usage_counter);
585fail_bdi:
586 blk_free_queue_stats(q->stats);
587fail_stats:
588 bdi_put(q->backing_dev_info);
589fail_split:
590 bioset_exit(&q->bio_split);
591fail_id:
592 ida_simple_remove(&blk_queue_ida, q->id);
593fail_q:
594 kmem_cache_free(blk_requestq_cachep, q);
595 return NULL;
596}
597EXPORT_SYMBOL(blk_alloc_queue);
598
599/**
600 * blk_get_queue - increment the request_queue refcount
601 * @q: the request_queue structure to increment the refcount for
602 *
603 * Increment the refcount of the request_queue kobject.
604 *
605 * Context: Any context.
606 */
607bool blk_get_queue(struct request_queue *q)
608{
609 if (likely(!blk_queue_dying(q))) {
610 __blk_get_queue(q);
611 return true;
612 }
613
614 return false;
615}
616EXPORT_SYMBOL(blk_get_queue);
617
618/**
619 * blk_get_request - allocate a request
620 * @q: request queue to allocate a request for
621 * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC.
622 * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
623 */
624struct request *blk_get_request(struct request_queue *q, unsigned int op,
625 blk_mq_req_flags_t flags)
626{
627 struct request *req;
628
629 WARN_ON_ONCE(op & REQ_NOWAIT);
630 WARN_ON_ONCE(flags & ~(BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_PREEMPT));
631
632 req = blk_mq_alloc_request(q, op, flags);
633 if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn)
634 q->mq_ops->initialize_rq_fn(req);
635
636 return req;
637}
638EXPORT_SYMBOL(blk_get_request);
639
640void blk_put_request(struct request *req)
641{
642 blk_mq_free_request(req);
643}
644EXPORT_SYMBOL(blk_put_request);
645
646static void blk_account_io_merge_bio(struct request *req)
647{
648 if (!blk_do_io_stat(req))
649 return;
650
651 part_stat_lock();
652 part_stat_inc(req->part, merges[op_stat_group(req_op(req))]);
653 part_stat_unlock();
654}
655
656bool bio_attempt_back_merge(struct request *req, struct bio *bio,
657 unsigned int nr_segs)
658{
659 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
660
661 if (!ll_back_merge_fn(req, bio, nr_segs))
662 return false;
663
664 trace_block_bio_backmerge(req->q, req, bio);
665 rq_qos_merge(req->q, req, bio);
666
667 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
668 blk_rq_set_mixed_merge(req);
669
670 req->biotail->bi_next = bio;
671 req->biotail = bio;
672 req->__data_len += bio->bi_iter.bi_size;
673
674 bio_crypt_free_ctx(bio);
675
676 blk_account_io_merge_bio(req);
677 return true;
678}
679
680bool bio_attempt_front_merge(struct request *req, struct bio *bio,
681 unsigned int nr_segs)
682{
683 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
684
685 if (!ll_front_merge_fn(req, bio, nr_segs))
686 return false;
687
688 trace_block_bio_frontmerge(req->q, req, bio);
689 rq_qos_merge(req->q, req, bio);
690
691 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
692 blk_rq_set_mixed_merge(req);
693
694 bio->bi_next = req->bio;
695 req->bio = bio;
696
697 req->__sector = bio->bi_iter.bi_sector;
698 req->__data_len += bio->bi_iter.bi_size;
699
700 bio_crypt_do_front_merge(req, bio);
701
702 blk_account_io_merge_bio(req);
703 return true;
704}
705
706bool bio_attempt_discard_merge(struct request_queue *q, struct request *req,
707 struct bio *bio)
708{
709 unsigned short segments = blk_rq_nr_discard_segments(req);
710
711 if (segments >= queue_max_discard_segments(q))
712 goto no_merge;
713 if (blk_rq_sectors(req) + bio_sectors(bio) >
714 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
715 goto no_merge;
716
717 rq_qos_merge(q, req, bio);
718
719 req->biotail->bi_next = bio;
720 req->biotail = bio;
721 req->__data_len += bio->bi_iter.bi_size;
722 req->nr_phys_segments = segments + 1;
723
724 blk_account_io_merge_bio(req);
725 return true;
726no_merge:
727 req_set_nomerge(q, req);
728 return false;
729}
730
731/**
732 * blk_attempt_plug_merge - try to merge with %current's plugged list
733 * @q: request_queue new bio is being queued at
734 * @bio: new bio being queued
735 * @nr_segs: number of segments in @bio
736 * @same_queue_rq: pointer to &struct request that gets filled in when
737 * another request associated with @q is found on the plug list
738 * (optional, may be %NULL)
739 *
740 * Determine whether @bio being queued on @q can be merged with a request
741 * on %current's plugged list. Returns %true if merge was successful,
742 * otherwise %false.
743 *
744 * Plugging coalesces IOs from the same issuer for the same purpose without
745 * going through @q->queue_lock. As such it's more of an issuing mechanism
746 * than scheduling, and the request, while may have elvpriv data, is not
747 * added on the elevator at this point. In addition, we don't have
748 * reliable access to the elevator outside queue lock. Only check basic
749 * merging parameters without querying the elevator.
750 *
751 * Caller must ensure !blk_queue_nomerges(q) beforehand.
752 */
753bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
754 unsigned int nr_segs, struct request **same_queue_rq)
755{
756 struct blk_plug *plug;
757 struct request *rq;
758 struct list_head *plug_list;
759
760 plug = blk_mq_plug(q, bio);
761 if (!plug)
762 return false;
763
764 plug_list = &plug->mq_list;
765
766 list_for_each_entry_reverse(rq, plug_list, queuelist) {
767 bool merged = false;
768
769 if (rq->q == q && same_queue_rq) {
770 /*
771 * Only blk-mq multiple hardware queues case checks the
772 * rq in the same queue, there should be only one such
773 * rq in a queue
774 **/
775 *same_queue_rq = rq;
776 }
777
778 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
779 continue;
780
781 switch (blk_try_merge(rq, bio)) {
782 case ELEVATOR_BACK_MERGE:
783 merged = bio_attempt_back_merge(rq, bio, nr_segs);
784 break;
785 case ELEVATOR_FRONT_MERGE:
786 merged = bio_attempt_front_merge(rq, bio, nr_segs);
787 break;
788 case ELEVATOR_DISCARD_MERGE:
789 merged = bio_attempt_discard_merge(q, rq, bio);
790 break;
791 default:
792 break;
793 }
794
795 if (merged)
796 return true;
797 }
798
799 return false;
800}
801
802static void handle_bad_sector(struct bio *bio, sector_t maxsector)
803{
804 char b[BDEVNAME_SIZE];
805
806 printk(KERN_INFO "attempt to access beyond end of device\n");
807 printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n",
808 bio_devname(bio, b), bio->bi_opf,
809 (unsigned long long)bio_end_sector(bio),
810 (long long)maxsector);
811}
812
813#ifdef CONFIG_FAIL_MAKE_REQUEST
814
815static DECLARE_FAULT_ATTR(fail_make_request);
816
817static int __init setup_fail_make_request(char *str)
818{
819 return setup_fault_attr(&fail_make_request, str);
820}
821__setup("fail_make_request=", setup_fail_make_request);
822
823static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
824{
825 return part->make_it_fail && should_fail(&fail_make_request, bytes);
826}
827
828static int __init fail_make_request_debugfs(void)
829{
830 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
831 NULL, &fail_make_request);
832
833 return PTR_ERR_OR_ZERO(dir);
834}
835
836late_initcall(fail_make_request_debugfs);
837
838#else /* CONFIG_FAIL_MAKE_REQUEST */
839
840static inline bool should_fail_request(struct hd_struct *part,
841 unsigned int bytes)
842{
843 return false;
844}
845
846#endif /* CONFIG_FAIL_MAKE_REQUEST */
847
848static inline bool bio_check_ro(struct bio *bio, struct hd_struct *part)
849{
850 const int op = bio_op(bio);
851
852 if (part->policy && op_is_write(op)) {
853 char b[BDEVNAME_SIZE];
854
855 if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
856 return false;
857
858 WARN_ONCE(1,
859 "Trying to write to read-only block-device %s (partno %d)\n",
860 bio_devname(bio, b), part->partno);
861 /* Older lvm-tools actually trigger this */
862 return false;
863 }
864
865 return false;
866}
867
868static noinline int should_fail_bio(struct bio *bio)
869{
870 if (should_fail_request(&bio->bi_disk->part0, bio->bi_iter.bi_size))
871 return -EIO;
872 return 0;
873}
874ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
875
876/*
877 * Check whether this bio extends beyond the end of the device or partition.
878 * This may well happen - the kernel calls bread() without checking the size of
879 * the device, e.g., when mounting a file system.
880 */
881static inline int bio_check_eod(struct bio *bio, sector_t maxsector)
882{
883 unsigned int nr_sectors = bio_sectors(bio);
884
885 if (nr_sectors && maxsector &&
886 (nr_sectors > maxsector ||
887 bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
888 handle_bad_sector(bio, maxsector);
889 return -EIO;
890 }
891 return 0;
892}
893
894/*
895 * Remap block n of partition p to block n+start(p) of the disk.
896 */
897static inline int blk_partition_remap(struct bio *bio)
898{
899 struct hd_struct *p;
900 int ret = -EIO;
901
902 rcu_read_lock();
903 p = __disk_get_part(bio->bi_disk, bio->bi_partno);
904 if (unlikely(!p))
905 goto out;
906 if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
907 goto out;
908 if (unlikely(bio_check_ro(bio, p)))
909 goto out;
910
911 if (bio_sectors(bio)) {
912 if (bio_check_eod(bio, part_nr_sects_read(p)))
913 goto out;
914 bio->bi_iter.bi_sector += p->start_sect;
915 trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p),
916 bio->bi_iter.bi_sector - p->start_sect);
917 }
918 bio->bi_partno = 0;
919 ret = 0;
920out:
921 rcu_read_unlock();
922 return ret;
923}
924
925/*
926 * Check write append to a zoned block device.
927 */
928static inline blk_status_t blk_check_zone_append(struct request_queue *q,
929 struct bio *bio)
930{
931 sector_t pos = bio->bi_iter.bi_sector;
932 int nr_sectors = bio_sectors(bio);
933
934 /* Only applicable to zoned block devices */
935 if (!blk_queue_is_zoned(q))
936 return BLK_STS_NOTSUPP;
937
938 /* The bio sector must point to the start of a sequential zone */
939 if (pos & (blk_queue_zone_sectors(q) - 1) ||
940 !blk_queue_zone_is_seq(q, pos))
941 return BLK_STS_IOERR;
942
943 /*
944 * Not allowed to cross zone boundaries. Otherwise, the BIO will be
945 * split and could result in non-contiguous sectors being written in
946 * different zones.
947 */
948 if (nr_sectors > q->limits.chunk_sectors)
949 return BLK_STS_IOERR;
950
951 /* Make sure the BIO is small enough and will not get split */
952 if (nr_sectors > q->limits.max_zone_append_sectors)
953 return BLK_STS_IOERR;
954
955 bio->bi_opf |= REQ_NOMERGE;
956
957 return BLK_STS_OK;
958}
959
960static noinline_for_stack bool submit_bio_checks(struct bio *bio)
961{
962 struct request_queue *q = bio->bi_disk->queue;
963 blk_status_t status = BLK_STS_IOERR;
964 struct blk_plug *plug;
965
966 might_sleep();
967
968 plug = blk_mq_plug(q, bio);
969 if (plug && plug->nowait)
970 bio->bi_opf |= REQ_NOWAIT;
971
972 /*
973 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
974 * if queue is not a request based queue.
975 */
976 if ((bio->bi_opf & REQ_NOWAIT) && !queue_is_mq(q))
977 goto not_supported;
978
979 if (should_fail_bio(bio))
980 goto end_io;
981
982 if (bio->bi_partno) {
983 if (unlikely(blk_partition_remap(bio)))
984 goto end_io;
985 } else {
986 if (unlikely(bio_check_ro(bio, &bio->bi_disk->part0)))
987 goto end_io;
988 if (unlikely(bio_check_eod(bio, get_capacity(bio->bi_disk))))
989 goto end_io;
990 }
991
992 /*
993 * Filter flush bio's early so that bio based drivers without flush
994 * support don't have to worry about them.
995 */
996 if (op_is_flush(bio->bi_opf) &&
997 !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
998 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
999 if (!bio_sectors(bio)) {
1000 status = BLK_STS_OK;
1001 goto end_io;
1002 }
1003 }
1004
1005 if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
1006 bio->bi_opf &= ~REQ_HIPRI;
1007
1008 switch (bio_op(bio)) {
1009 case REQ_OP_DISCARD:
1010 if (!blk_queue_discard(q))
1011 goto not_supported;
1012 break;
1013 case REQ_OP_SECURE_ERASE:
1014 if (!blk_queue_secure_erase(q))
1015 goto not_supported;
1016 break;
1017 case REQ_OP_WRITE_SAME:
1018 if (!q->limits.max_write_same_sectors)
1019 goto not_supported;
1020 break;
1021 case REQ_OP_ZONE_APPEND:
1022 status = blk_check_zone_append(q, bio);
1023 if (status != BLK_STS_OK)
1024 goto end_io;
1025 break;
1026 case REQ_OP_ZONE_RESET:
1027 case REQ_OP_ZONE_OPEN:
1028 case REQ_OP_ZONE_CLOSE:
1029 case REQ_OP_ZONE_FINISH:
1030 if (!blk_queue_is_zoned(q))
1031 goto not_supported;
1032 break;
1033 case REQ_OP_ZONE_RESET_ALL:
1034 if (!blk_queue_is_zoned(q) || !blk_queue_zone_resetall(q))
1035 goto not_supported;
1036 break;
1037 case REQ_OP_WRITE_ZEROES:
1038 if (!q->limits.max_write_zeroes_sectors)
1039 goto not_supported;
1040 break;
1041 default:
1042 break;
1043 }
1044
1045 /*
1046 * Various block parts want %current->io_context, so allocate it up
1047 * front rather than dealing with lots of pain to allocate it only
1048 * where needed. This may fail and the block layer knows how to live
1049 * with it.
1050 */
1051 if (unlikely(!current->io_context))
1052 create_task_io_context(current, GFP_ATOMIC, q->node);
1053
1054 if (blk_throtl_bio(bio)) {
1055 blkcg_bio_issue_init(bio);
1056 return false;
1057 }
1058
1059 blk_cgroup_bio_start(bio);
1060 blkcg_bio_issue_init(bio);
1061
1062 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
1063 trace_block_bio_queue(q, bio);
1064 /* Now that enqueuing has been traced, we need to trace
1065 * completion as well.
1066 */
1067 bio_set_flag(bio, BIO_TRACE_COMPLETION);
1068 }
1069 return true;
1070
1071not_supported:
1072 status = BLK_STS_NOTSUPP;
1073end_io:
1074 bio->bi_status = status;
1075 bio_endio(bio);
1076 return false;
1077}
1078
1079static blk_qc_t __submit_bio(struct bio *bio)
1080{
1081 struct gendisk *disk = bio->bi_disk;
1082 blk_qc_t ret = BLK_QC_T_NONE;
1083
1084 if (blk_crypto_bio_prep(&bio)) {
1085 if (!disk->fops->submit_bio)
1086 return blk_mq_submit_bio(bio);
1087 ret = disk->fops->submit_bio(bio);
1088 }
1089 blk_queue_exit(disk->queue);
1090 return ret;
1091}
1092
1093/*
1094 * The loop in this function may be a bit non-obvious, and so deserves some
1095 * explanation:
1096 *
1097 * - Before entering the loop, bio->bi_next is NULL (as all callers ensure
1098 * that), so we have a list with a single bio.
1099 * - We pretend that we have just taken it off a longer list, so we assign
1100 * bio_list to a pointer to the bio_list_on_stack, thus initialising the
1101 * bio_list of new bios to be added. ->submit_bio() may indeed add some more
1102 * bios through a recursive call to submit_bio_noacct. If it did, we find a
1103 * non-NULL value in bio_list and re-enter the loop from the top.
1104 * - In this case we really did just take the bio of the top of the list (no
1105 * pretending) and so remove it from bio_list, and call into ->submit_bio()
1106 * again.
1107 *
1108 * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio.
1109 * bio_list_on_stack[1] contains bios that were submitted before the current
1110 * ->submit_bio_bio, but that haven't been processed yet.
1111 */
1112static blk_qc_t __submit_bio_noacct(struct bio *bio)
1113{
1114 struct bio_list bio_list_on_stack[2];
1115 blk_qc_t ret = BLK_QC_T_NONE;
1116
1117 BUG_ON(bio->bi_next);
1118
1119 bio_list_init(&bio_list_on_stack[0]);
1120 current->bio_list = bio_list_on_stack;
1121
1122 do {
1123 struct request_queue *q = bio->bi_disk->queue;
1124 struct bio_list lower, same;
1125
1126 if (unlikely(bio_queue_enter(bio) != 0))
1127 continue;
1128
1129 /*
1130 * Create a fresh bio_list for all subordinate requests.
1131 */
1132 bio_list_on_stack[1] = bio_list_on_stack[0];
1133 bio_list_init(&bio_list_on_stack[0]);
1134
1135 ret = __submit_bio(bio);
1136
1137 /*
1138 * Sort new bios into those for a lower level and those for the
1139 * same level.
1140 */
1141 bio_list_init(&lower);
1142 bio_list_init(&same);
1143 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
1144 if (q == bio->bi_disk->queue)
1145 bio_list_add(&same, bio);
1146 else
1147 bio_list_add(&lower, bio);
1148
1149 /*
1150 * Now assemble so we handle the lowest level first.
1151 */
1152 bio_list_merge(&bio_list_on_stack[0], &lower);
1153 bio_list_merge(&bio_list_on_stack[0], &same);
1154 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
1155 } while ((bio = bio_list_pop(&bio_list_on_stack[0])));
1156
1157 current->bio_list = NULL;
1158 return ret;
1159}
1160
1161static blk_qc_t __submit_bio_noacct_mq(struct bio *bio)
1162{
1163 struct bio_list bio_list[2] = { };
1164 blk_qc_t ret = BLK_QC_T_NONE;
1165
1166 current->bio_list = bio_list;
1167
1168 do {
1169 struct gendisk *disk = bio->bi_disk;
1170
1171 if (unlikely(bio_queue_enter(bio) != 0))
1172 continue;
1173
1174 if (!blk_crypto_bio_prep(&bio)) {
1175 blk_queue_exit(disk->queue);
1176 ret = BLK_QC_T_NONE;
1177 continue;
1178 }
1179
1180 ret = blk_mq_submit_bio(bio);
1181 } while ((bio = bio_list_pop(&bio_list[0])));
1182
1183 current->bio_list = NULL;
1184 return ret;
1185}
1186
1187/**
1188 * submit_bio_noacct - re-submit a bio to the block device layer for I/O
1189 * @bio: The bio describing the location in memory and on the device.
1190 *
1191 * This is a version of submit_bio() that shall only be used for I/O that is
1192 * resubmitted to lower level drivers by stacking block drivers. All file
1193 * systems and other upper level users of the block layer should use
1194 * submit_bio() instead.
1195 */
1196blk_qc_t submit_bio_noacct(struct bio *bio)
1197{
1198 if (!submit_bio_checks(bio))
1199 return BLK_QC_T_NONE;
1200
1201 /*
1202 * We only want one ->submit_bio to be active at a time, else stack
1203 * usage with stacked devices could be a problem. Use current->bio_list
1204 * to collect a list of requests submited by a ->submit_bio method while
1205 * it is active, and then process them after it returned.
1206 */
1207 if (current->bio_list) {
1208 bio_list_add(¤t->bio_list[0], bio);
1209 return BLK_QC_T_NONE;
1210 }
1211
1212 if (!bio->bi_disk->fops->submit_bio)
1213 return __submit_bio_noacct_mq(bio);
1214 return __submit_bio_noacct(bio);
1215}
1216EXPORT_SYMBOL(submit_bio_noacct);
1217
1218/**
1219 * submit_bio - submit a bio to the block device layer for I/O
1220 * @bio: The &struct bio which describes the I/O
1221 *
1222 * submit_bio() is used to submit I/O requests to block devices. It is passed a
1223 * fully set up &struct bio that describes the I/O that needs to be done. The
1224 * bio will be send to the device described by the bi_disk and bi_partno fields.
1225 *
1226 * The success/failure status of the request, along with notification of
1227 * completion, is delivered asynchronously through the ->bi_end_io() callback
1228 * in @bio. The bio must NOT be touched by thecaller until ->bi_end_io() has
1229 * been called.
1230 */
1231blk_qc_t submit_bio(struct bio *bio)
1232{
1233 if (blkcg_punt_bio_submit(bio))
1234 return BLK_QC_T_NONE;
1235
1236 /*
1237 * If it's a regular read/write or a barrier with data attached,
1238 * go through the normal accounting stuff before submission.
1239 */
1240 if (bio_has_data(bio)) {
1241 unsigned int count;
1242
1243 if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
1244 count = queue_logical_block_size(bio->bi_disk->queue) >> 9;
1245 else
1246 count = bio_sectors(bio);
1247
1248 if (op_is_write(bio_op(bio))) {
1249 count_vm_events(PGPGOUT, count);
1250 } else {
1251 task_io_account_read(bio->bi_iter.bi_size);
1252 count_vm_events(PGPGIN, count);
1253 }
1254
1255 if (unlikely(block_dump)) {
1256 char b[BDEVNAME_SIZE];
1257 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1258 current->comm, task_pid_nr(current),
1259 op_is_write(bio_op(bio)) ? "WRITE" : "READ",
1260 (unsigned long long)bio->bi_iter.bi_sector,
1261 bio_devname(bio, b), count);
1262 }
1263 }
1264
1265 /*
1266 * If we're reading data that is part of the userspace workingset, count
1267 * submission time as memory stall. When the device is congested, or
1268 * the submitting cgroup IO-throttled, submission can be a significant
1269 * part of overall IO time.
1270 */
1271 if (unlikely(bio_op(bio) == REQ_OP_READ &&
1272 bio_flagged(bio, BIO_WORKINGSET))) {
1273 unsigned long pflags;
1274 blk_qc_t ret;
1275
1276 psi_memstall_enter(&pflags);
1277 ret = submit_bio_noacct(bio);
1278 psi_memstall_leave(&pflags);
1279
1280 return ret;
1281 }
1282
1283 return submit_bio_noacct(bio);
1284}
1285EXPORT_SYMBOL(submit_bio);
1286
1287/**
1288 * blk_cloned_rq_check_limits - Helper function to check a cloned request
1289 * for the new queue limits
1290 * @q: the queue
1291 * @rq: the request being checked
1292 *
1293 * Description:
1294 * @rq may have been made based on weaker limitations of upper-level queues
1295 * in request stacking drivers, and it may violate the limitation of @q.
1296 * Since the block layer and the underlying device driver trust @rq
1297 * after it is inserted to @q, it should be checked against @q before
1298 * the insertion using this generic function.
1299 *
1300 * Request stacking drivers like request-based dm may change the queue
1301 * limits when retrying requests on other queues. Those requests need
1302 * to be checked against the new queue limits again during dispatch.
1303 */
1304static int blk_cloned_rq_check_limits(struct request_queue *q,
1305 struct request *rq)
1306{
1307 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) {
1308 printk(KERN_ERR "%s: over max size limit. (%u > %u)\n",
1309 __func__, blk_rq_sectors(rq),
1310 blk_queue_get_max_sectors(q, req_op(rq)));
1311 return -EIO;
1312 }
1313
1314 /*
1315 * queue's settings related to segment counting like q->bounce_pfn
1316 * may differ from that of other stacking queues.
1317 * Recalculate it to check the request correctly on this queue's
1318 * limitation.
1319 */
1320 rq->nr_phys_segments = blk_recalc_rq_segments(rq);
1321 if (rq->nr_phys_segments > queue_max_segments(q)) {
1322 printk(KERN_ERR "%s: over max segments limit. (%hu > %hu)\n",
1323 __func__, rq->nr_phys_segments, queue_max_segments(q));
1324 return -EIO;
1325 }
1326
1327 return 0;
1328}
1329
1330/**
1331 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1332 * @q: the queue to submit the request
1333 * @rq: the request being queued
1334 */
1335blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1336{
1337 if (blk_cloned_rq_check_limits(q, rq))
1338 return BLK_STS_IOERR;
1339
1340 if (rq->rq_disk &&
1341 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
1342 return BLK_STS_IOERR;
1343
1344 if (blk_crypto_insert_cloned_request(rq))
1345 return BLK_STS_IOERR;
1346
1347 if (blk_queue_io_stat(q))
1348 blk_account_io_start(rq);
1349
1350 /*
1351 * Since we have a scheduler attached on the top device,
1352 * bypass a potential scheduler on the bottom device for
1353 * insert.
1354 */
1355 return blk_mq_request_issue_directly(rq, true);
1356}
1357EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1358
1359/**
1360 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1361 * @rq: request to examine
1362 *
1363 * Description:
1364 * A request could be merge of IOs which require different failure
1365 * handling. This function determines the number of bytes which
1366 * can be failed from the beginning of the request without
1367 * crossing into area which need to be retried further.
1368 *
1369 * Return:
1370 * The number of bytes to fail.
1371 */
1372unsigned int blk_rq_err_bytes(const struct request *rq)
1373{
1374 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1375 unsigned int bytes = 0;
1376 struct bio *bio;
1377
1378 if (!(rq->rq_flags & RQF_MIXED_MERGE))
1379 return blk_rq_bytes(rq);
1380
1381 /*
1382 * Currently the only 'mixing' which can happen is between
1383 * different fastfail types. We can safely fail portions
1384 * which have all the failfast bits that the first one has -
1385 * the ones which are at least as eager to fail as the first
1386 * one.
1387 */
1388 for (bio = rq->bio; bio; bio = bio->bi_next) {
1389 if ((bio->bi_opf & ff) != ff)
1390 break;
1391 bytes += bio->bi_iter.bi_size;
1392 }
1393
1394 /* this could lead to infinite loop */
1395 BUG_ON(blk_rq_bytes(rq) && !bytes);
1396 return bytes;
1397}
1398EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1399
1400static void update_io_ticks(struct hd_struct *part, unsigned long now, bool end)
1401{
1402 unsigned long stamp;
1403again:
1404 stamp = READ_ONCE(part->stamp);
1405 if (unlikely(stamp != now)) {
1406 if (likely(cmpxchg(&part->stamp, stamp, now) == stamp))
1407 __part_stat_add(part, io_ticks, end ? now - stamp : 1);
1408 }
1409 if (part->partno) {
1410 part = &part_to_disk(part)->part0;
1411 goto again;
1412 }
1413}
1414
1415static void blk_account_io_completion(struct request *req, unsigned int bytes)
1416{
1417 if (req->part && blk_do_io_stat(req)) {
1418 const int sgrp = op_stat_group(req_op(req));
1419 struct hd_struct *part;
1420
1421 part_stat_lock();
1422 part = req->part;
1423 part_stat_add(part, sectors[sgrp], bytes >> 9);
1424 part_stat_unlock();
1425 }
1426}
1427
1428void blk_account_io_done(struct request *req, u64 now)
1429{
1430 /*
1431 * Account IO completion. flush_rq isn't accounted as a
1432 * normal IO on queueing nor completion. Accounting the
1433 * containing request is enough.
1434 */
1435 if (req->part && blk_do_io_stat(req) &&
1436 !(req->rq_flags & RQF_FLUSH_SEQ)) {
1437 const int sgrp = op_stat_group(req_op(req));
1438 struct hd_struct *part;
1439
1440 part_stat_lock();
1441 part = req->part;
1442
1443 update_io_ticks(part, jiffies, true);
1444 part_stat_inc(part, ios[sgrp]);
1445 part_stat_add(part, nsecs[sgrp], now - req->start_time_ns);
1446 part_stat_unlock();
1447
1448 hd_struct_put(part);
1449 }
1450}
1451
1452void blk_account_io_start(struct request *rq)
1453{
1454 if (!blk_do_io_stat(rq))
1455 return;
1456
1457 rq->part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
1458
1459 part_stat_lock();
1460 update_io_ticks(rq->part, jiffies, false);
1461 part_stat_unlock();
1462}
1463
1464unsigned long disk_start_io_acct(struct gendisk *disk, unsigned int sectors,
1465 unsigned int op)
1466{
1467 struct hd_struct *part = &disk->part0;
1468 const int sgrp = op_stat_group(op);
1469 unsigned long now = READ_ONCE(jiffies);
1470
1471 part_stat_lock();
1472 update_io_ticks(part, now, false);
1473 part_stat_inc(part, ios[sgrp]);
1474 part_stat_add(part, sectors[sgrp], sectors);
1475 part_stat_local_inc(part, in_flight[op_is_write(op)]);
1476 part_stat_unlock();
1477
1478 return now;
1479}
1480EXPORT_SYMBOL(disk_start_io_acct);
1481
1482void disk_end_io_acct(struct gendisk *disk, unsigned int op,
1483 unsigned long start_time)
1484{
1485 struct hd_struct *part = &disk->part0;
1486 const int sgrp = op_stat_group(op);
1487 unsigned long now = READ_ONCE(jiffies);
1488 unsigned long duration = now - start_time;
1489
1490 part_stat_lock();
1491 update_io_ticks(part, now, true);
1492 part_stat_add(part, nsecs[sgrp], jiffies_to_nsecs(duration));
1493 part_stat_local_dec(part, in_flight[op_is_write(op)]);
1494 part_stat_unlock();
1495}
1496EXPORT_SYMBOL(disk_end_io_acct);
1497
1498/*
1499 * Steal bios from a request and add them to a bio list.
1500 * The request must not have been partially completed before.
1501 */
1502void blk_steal_bios(struct bio_list *list, struct request *rq)
1503{
1504 if (rq->bio) {
1505 if (list->tail)
1506 list->tail->bi_next = rq->bio;
1507 else
1508 list->head = rq->bio;
1509 list->tail = rq->biotail;
1510
1511 rq->bio = NULL;
1512 rq->biotail = NULL;
1513 }
1514
1515 rq->__data_len = 0;
1516}
1517EXPORT_SYMBOL_GPL(blk_steal_bios);
1518
1519/**
1520 * blk_update_request - Special helper function for request stacking drivers
1521 * @req: the request being processed
1522 * @error: block status code
1523 * @nr_bytes: number of bytes to complete @req
1524 *
1525 * Description:
1526 * Ends I/O on a number of bytes attached to @req, but doesn't complete
1527 * the request structure even if @req doesn't have leftover.
1528 * If @req has leftover, sets it up for the next range of segments.
1529 *
1530 * This special helper function is only for request stacking drivers
1531 * (e.g. request-based dm) so that they can handle partial completion.
1532 * Actual device drivers should use blk_mq_end_request instead.
1533 *
1534 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
1535 * %false return from this function.
1536 *
1537 * Note:
1538 * The RQF_SPECIAL_PAYLOAD flag is ignored on purpose in both
1539 * blk_rq_bytes() and in blk_update_request().
1540 *
1541 * Return:
1542 * %false - this request doesn't have any more data
1543 * %true - this request has more data
1544 **/
1545bool blk_update_request(struct request *req, blk_status_t error,
1546 unsigned int nr_bytes)
1547{
1548 int total_bytes;
1549
1550 trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes);
1551
1552 if (!req->bio)
1553 return false;
1554
1555#ifdef CONFIG_BLK_DEV_INTEGRITY
1556 if (blk_integrity_rq(req) && req_op(req) == REQ_OP_READ &&
1557 error == BLK_STS_OK)
1558 req->q->integrity.profile->complete_fn(req, nr_bytes);
1559#endif
1560
1561 if (unlikely(error && !blk_rq_is_passthrough(req) &&
1562 !(req->rq_flags & RQF_QUIET)))
1563 print_req_error(req, error, __func__);
1564
1565 blk_account_io_completion(req, nr_bytes);
1566
1567 total_bytes = 0;
1568 while (req->bio) {
1569 struct bio *bio = req->bio;
1570 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
1571
1572 if (bio_bytes == bio->bi_iter.bi_size)
1573 req->bio = bio->bi_next;
1574
1575 /* Completion has already been traced */
1576 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1577 req_bio_endio(req, bio, bio_bytes, error);
1578
1579 total_bytes += bio_bytes;
1580 nr_bytes -= bio_bytes;
1581
1582 if (!nr_bytes)
1583 break;
1584 }
1585
1586 /*
1587 * completely done
1588 */
1589 if (!req->bio) {
1590 /*
1591 * Reset counters so that the request stacking driver
1592 * can find how many bytes remain in the request
1593 * later.
1594 */
1595 req->__data_len = 0;
1596 return false;
1597 }
1598
1599 req->__data_len -= total_bytes;
1600
1601 /* update sector only for requests with clear definition of sector */
1602 if (!blk_rq_is_passthrough(req))
1603 req->__sector += total_bytes >> 9;
1604
1605 /* mixed attributes always follow the first bio */
1606 if (req->rq_flags & RQF_MIXED_MERGE) {
1607 req->cmd_flags &= ~REQ_FAILFAST_MASK;
1608 req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
1609 }
1610
1611 if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) {
1612 /*
1613 * If total number of sectors is less than the first segment
1614 * size, something has gone terribly wrong.
1615 */
1616 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
1617 blk_dump_rq_flags(req, "request botched");
1618 req->__data_len = blk_rq_cur_bytes(req);
1619 }
1620
1621 /* recalculate the number of segments */
1622 req->nr_phys_segments = blk_recalc_rq_segments(req);
1623 }
1624
1625 return true;
1626}
1627EXPORT_SYMBOL_GPL(blk_update_request);
1628
1629#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1630/**
1631 * rq_flush_dcache_pages - Helper function to flush all pages in a request
1632 * @rq: the request to be flushed
1633 *
1634 * Description:
1635 * Flush all pages in @rq.
1636 */
1637void rq_flush_dcache_pages(struct request *rq)
1638{
1639 struct req_iterator iter;
1640 struct bio_vec bvec;
1641
1642 rq_for_each_segment(bvec, rq, iter)
1643 flush_dcache_page(bvec.bv_page);
1644}
1645EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
1646#endif
1647
1648/**
1649 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1650 * @q : the queue of the device being checked
1651 *
1652 * Description:
1653 * Check if underlying low-level drivers of a device are busy.
1654 * If the drivers want to export their busy state, they must set own
1655 * exporting function using blk_queue_lld_busy() first.
1656 *
1657 * Basically, this function is used only by request stacking drivers
1658 * to stop dispatching requests to underlying devices when underlying
1659 * devices are busy. This behavior helps more I/O merging on the queue
1660 * of the request stacking driver and prevents I/O throughput regression
1661 * on burst I/O load.
1662 *
1663 * Return:
1664 * 0 - Not busy (The request stacking driver should dispatch request)
1665 * 1 - Busy (The request stacking driver should stop dispatching request)
1666 */
1667int blk_lld_busy(struct request_queue *q)
1668{
1669 if (queue_is_mq(q) && q->mq_ops->busy)
1670 return q->mq_ops->busy(q);
1671
1672 return 0;
1673}
1674EXPORT_SYMBOL_GPL(blk_lld_busy);
1675
1676/**
1677 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
1678 * @rq: the clone request to be cleaned up
1679 *
1680 * Description:
1681 * Free all bios in @rq for a cloned request.
1682 */
1683void blk_rq_unprep_clone(struct request *rq)
1684{
1685 struct bio *bio;
1686
1687 while ((bio = rq->bio) != NULL) {
1688 rq->bio = bio->bi_next;
1689
1690 bio_put(bio);
1691 }
1692}
1693EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
1694
1695/**
1696 * blk_rq_prep_clone - Helper function to setup clone request
1697 * @rq: the request to be setup
1698 * @rq_src: original request to be cloned
1699 * @bs: bio_set that bios for clone are allocated from
1700 * @gfp_mask: memory allocation mask for bio
1701 * @bio_ctr: setup function to be called for each clone bio.
1702 * Returns %0 for success, non %0 for failure.
1703 * @data: private data to be passed to @bio_ctr
1704 *
1705 * Description:
1706 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
1707 * Also, pages which the original bios are pointing to are not copied
1708 * and the cloned bios just point same pages.
1709 * So cloned bios must be completed before original bios, which means
1710 * the caller must complete @rq before @rq_src.
1711 */
1712int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
1713 struct bio_set *bs, gfp_t gfp_mask,
1714 int (*bio_ctr)(struct bio *, struct bio *, void *),
1715 void *data)
1716{
1717 struct bio *bio, *bio_src;
1718
1719 if (!bs)
1720 bs = &fs_bio_set;
1721
1722 __rq_for_each_bio(bio_src, rq_src) {
1723 bio = bio_clone_fast(bio_src, gfp_mask, bs);
1724 if (!bio)
1725 goto free_and_out;
1726
1727 if (bio_ctr && bio_ctr(bio, bio_src, data))
1728 goto free_and_out;
1729
1730 if (rq->bio) {
1731 rq->biotail->bi_next = bio;
1732 rq->biotail = bio;
1733 } else
1734 rq->bio = rq->biotail = bio;
1735 }
1736
1737 /* Copy attributes of the original request to the clone request. */
1738 rq->__sector = blk_rq_pos(rq_src);
1739 rq->__data_len = blk_rq_bytes(rq_src);
1740 if (rq_src->rq_flags & RQF_SPECIAL_PAYLOAD) {
1741 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
1742 rq->special_vec = rq_src->special_vec;
1743 }
1744 rq->nr_phys_segments = rq_src->nr_phys_segments;
1745 rq->ioprio = rq_src->ioprio;
1746
1747 if (rq->bio)
1748 blk_crypto_rq_bio_prep(rq, rq->bio, gfp_mask);
1749
1750 return 0;
1751
1752free_and_out:
1753 if (bio)
1754 bio_put(bio);
1755 blk_rq_unprep_clone(rq);
1756
1757 return -ENOMEM;
1758}
1759EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
1760
1761int kblockd_schedule_work(struct work_struct *work)
1762{
1763 return queue_work(kblockd_workqueue, work);
1764}
1765EXPORT_SYMBOL(kblockd_schedule_work);
1766
1767int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
1768 unsigned long delay)
1769{
1770 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
1771}
1772EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
1773
1774/**
1775 * blk_start_plug - initialize blk_plug and track it inside the task_struct
1776 * @plug: The &struct blk_plug that needs to be initialized
1777 *
1778 * Description:
1779 * blk_start_plug() indicates to the block layer an intent by the caller
1780 * to submit multiple I/O requests in a batch. The block layer may use
1781 * this hint to defer submitting I/Os from the caller until blk_finish_plug()
1782 * is called. However, the block layer may choose to submit requests
1783 * before a call to blk_finish_plug() if the number of queued I/Os
1784 * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1785 * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if
1786 * the task schedules (see below).
1787 *
1788 * Tracking blk_plug inside the task_struct will help with auto-flushing the
1789 * pending I/O should the task end up blocking between blk_start_plug() and
1790 * blk_finish_plug(). This is important from a performance perspective, but
1791 * also ensures that we don't deadlock. For instance, if the task is blocking
1792 * for a memory allocation, memory reclaim could end up wanting to free a
1793 * page belonging to that request that is currently residing in our private
1794 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
1795 * this kind of deadlock.
1796 */
1797void blk_start_plug(struct blk_plug *plug)
1798{
1799 struct task_struct *tsk = current;
1800
1801 /*
1802 * If this is a nested plug, don't actually assign it.
1803 */
1804 if (tsk->plug)
1805 return;
1806
1807 INIT_LIST_HEAD(&plug->mq_list);
1808 INIT_LIST_HEAD(&plug->cb_list);
1809 plug->rq_count = 0;
1810 plug->multiple_queues = false;
1811 plug->nowait = false;
1812
1813 /*
1814 * Store ordering should not be needed here, since a potential
1815 * preempt will imply a full memory barrier
1816 */
1817 tsk->plug = plug;
1818}
1819EXPORT_SYMBOL(blk_start_plug);
1820
1821static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
1822{
1823 LIST_HEAD(callbacks);
1824
1825 while (!list_empty(&plug->cb_list)) {
1826 list_splice_init(&plug->cb_list, &callbacks);
1827
1828 while (!list_empty(&callbacks)) {
1829 struct blk_plug_cb *cb = list_first_entry(&callbacks,
1830 struct blk_plug_cb,
1831 list);
1832 list_del(&cb->list);
1833 cb->callback(cb, from_schedule);
1834 }
1835 }
1836}
1837
1838struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
1839 int size)
1840{
1841 struct blk_plug *plug = current->plug;
1842 struct blk_plug_cb *cb;
1843
1844 if (!plug)
1845 return NULL;
1846
1847 list_for_each_entry(cb, &plug->cb_list, list)
1848 if (cb->callback == unplug && cb->data == data)
1849 return cb;
1850
1851 /* Not currently on the callback list */
1852 BUG_ON(size < sizeof(*cb));
1853 cb = kzalloc(size, GFP_ATOMIC);
1854 if (cb) {
1855 cb->data = data;
1856 cb->callback = unplug;
1857 list_add(&cb->list, &plug->cb_list);
1858 }
1859 return cb;
1860}
1861EXPORT_SYMBOL(blk_check_plugged);
1862
1863void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
1864{
1865 flush_plug_callbacks(plug, from_schedule);
1866
1867 if (!list_empty(&plug->mq_list))
1868 blk_mq_flush_plug_list(plug, from_schedule);
1869}
1870
1871/**
1872 * blk_finish_plug - mark the end of a batch of submitted I/O
1873 * @plug: The &struct blk_plug passed to blk_start_plug()
1874 *
1875 * Description:
1876 * Indicate that a batch of I/O submissions is complete. This function
1877 * must be paired with an initial call to blk_start_plug(). The intent
1878 * is to allow the block layer to optimize I/O submission. See the
1879 * documentation for blk_start_plug() for more information.
1880 */
1881void blk_finish_plug(struct blk_plug *plug)
1882{
1883 if (plug != current->plug)
1884 return;
1885 blk_flush_plug_list(plug, false);
1886
1887 current->plug = NULL;
1888}
1889EXPORT_SYMBOL(blk_finish_plug);
1890
1891void blk_io_schedule(void)
1892{
1893 /* Prevent hang_check timer from firing at us during very long I/O */
1894 unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2;
1895
1896 if (timeout)
1897 io_schedule_timeout(timeout);
1898 else
1899 io_schedule();
1900}
1901EXPORT_SYMBOL_GPL(blk_io_schedule);
1902
1903int __init blk_dev_init(void)
1904{
1905 BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
1906 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1907 sizeof_field(struct request, cmd_flags));
1908 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1909 sizeof_field(struct bio, bi_opf));
1910
1911 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
1912 kblockd_workqueue = alloc_workqueue("kblockd",
1913 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1914 if (!kblockd_workqueue)
1915 panic("Failed to create kblockd\n");
1916
1917 blk_requestq_cachep = kmem_cache_create("request_queue",
1918 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
1919
1920 blk_debugfs_root = debugfs_create_dir("block", NULL);
1921
1922 return 0;
1923}