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
36#define CREATE_TRACE_POINTS
37#include <trace/events/block.h>
38
39#include "blk.h"
40#include "blk-cgroup.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_unplug);
47
48DEFINE_IDA(blk_queue_ida);
49
50/*
51 * For the allocated request tables
52 */
53struct kmem_cache *request_cachep = NULL;
54
55/*
56 * For queue allocation
57 */
58struct kmem_cache *blk_requestq_cachep;
59
60/*
61 * Controlling structure to kblockd
62 */
63static struct workqueue_struct *kblockd_workqueue;
64
65void blk_queue_congestion_threshold(struct request_queue *q)
66{
67 int nr;
68
69 nr = q->nr_requests - (q->nr_requests / 8) + 1;
70 if (nr > q->nr_requests)
71 nr = q->nr_requests;
72 q->nr_congestion_on = nr;
73
74 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
75 if (nr < 1)
76 nr = 1;
77 q->nr_congestion_off = nr;
78}
79
80/**
81 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
82 * @bdev: device
83 *
84 * Locates the passed device's request queue and returns the address of its
85 * backing_dev_info
86 *
87 * Will return NULL if the request queue cannot be located.
88 */
89struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
90{
91 struct backing_dev_info *ret = NULL;
92 struct request_queue *q = bdev_get_queue(bdev);
93
94 if (q)
95 ret = &q->backing_dev_info;
96 return ret;
97}
98EXPORT_SYMBOL(blk_get_backing_dev_info);
99
100void blk_rq_init(struct request_queue *q, struct request *rq)
101{
102 memset(rq, 0, sizeof(*rq));
103
104 INIT_LIST_HEAD(&rq->queuelist);
105 INIT_LIST_HEAD(&rq->timeout_list);
106 rq->cpu = -1;
107 rq->q = q;
108 rq->__sector = (sector_t) -1;
109 INIT_HLIST_NODE(&rq->hash);
110 RB_CLEAR_NODE(&rq->rb_node);
111 rq->cmd = rq->__cmd;
112 rq->cmd_len = BLK_MAX_CDB;
113 rq->tag = -1;
114 rq->start_time = jiffies;
115 set_start_time_ns(rq);
116 rq->part = NULL;
117}
118EXPORT_SYMBOL(blk_rq_init);
119
120static void req_bio_endio(struct request *rq, struct bio *bio,
121 unsigned int nbytes, int error)
122{
123 if (error)
124 clear_bit(BIO_UPTODATE, &bio->bi_flags);
125 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
126 error = -EIO;
127
128 if (unlikely(rq->cmd_flags & REQ_QUIET))
129 set_bit(BIO_QUIET, &bio->bi_flags);
130
131 bio_advance(bio, nbytes);
132
133 /* don't actually finish bio if it's part of flush sequence */
134 if (bio->bi_iter.bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
135 bio_endio(bio, error);
136}
137
138void blk_dump_rq_flags(struct request *rq, char *msg)
139{
140 int bit;
141
142 printk(KERN_INFO "%s: dev %s: type=%x, flags=%llx\n", msg,
143 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
144 (unsigned long long) rq->cmd_flags);
145
146 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
147 (unsigned long long)blk_rq_pos(rq),
148 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
149 printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n",
150 rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq));
151
152 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
153 printk(KERN_INFO " cdb: ");
154 for (bit = 0; bit < BLK_MAX_CDB; bit++)
155 printk("%02x ", rq->cmd[bit]);
156 printk("\n");
157 }
158}
159EXPORT_SYMBOL(blk_dump_rq_flags);
160
161static void blk_delay_work(struct work_struct *work)
162{
163 struct request_queue *q;
164
165 q = container_of(work, struct request_queue, delay_work.work);
166 spin_lock_irq(q->queue_lock);
167 __blk_run_queue(q);
168 spin_unlock_irq(q->queue_lock);
169}
170
171/**
172 * blk_delay_queue - restart queueing after defined interval
173 * @q: The &struct request_queue in question
174 * @msecs: Delay in msecs
175 *
176 * Description:
177 * Sometimes queueing needs to be postponed for a little while, to allow
178 * resources to come back. This function will make sure that queueing is
179 * restarted around the specified time. Queue lock must be held.
180 */
181void blk_delay_queue(struct request_queue *q, unsigned long msecs)
182{
183 if (likely(!blk_queue_dead(q)))
184 queue_delayed_work(kblockd_workqueue, &q->delay_work,
185 msecs_to_jiffies(msecs));
186}
187EXPORT_SYMBOL(blk_delay_queue);
188
189/**
190 * blk_start_queue - restart a previously stopped queue
191 * @q: The &struct request_queue in question
192 *
193 * Description:
194 * blk_start_queue() will clear the stop flag on the queue, and call
195 * the request_fn for the queue if it was in a stopped state when
196 * entered. Also see blk_stop_queue(). Queue lock must be held.
197 **/
198void blk_start_queue(struct request_queue *q)
199{
200 WARN_ON(!irqs_disabled());
201
202 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
203 __blk_run_queue(q);
204}
205EXPORT_SYMBOL(blk_start_queue);
206
207/**
208 * blk_stop_queue - stop a queue
209 * @q: The &struct request_queue in question
210 *
211 * Description:
212 * The Linux block layer assumes that a block driver will consume all
213 * entries on the request queue when the request_fn strategy is called.
214 * Often this will not happen, because of hardware limitations (queue
215 * depth settings). If a device driver gets a 'queue full' response,
216 * or if it simply chooses not to queue more I/O at one point, it can
217 * call this function to prevent the request_fn from being called until
218 * the driver has signalled it's ready to go again. This happens by calling
219 * blk_start_queue() to restart queue operations. Queue lock must be held.
220 **/
221void blk_stop_queue(struct request_queue *q)
222{
223 cancel_delayed_work(&q->delay_work);
224 queue_flag_set(QUEUE_FLAG_STOPPED, q);
225}
226EXPORT_SYMBOL(blk_stop_queue);
227
228/**
229 * blk_sync_queue - cancel any pending callbacks on a queue
230 * @q: the queue
231 *
232 * Description:
233 * The block layer may perform asynchronous callback activity
234 * on a queue, such as calling the unplug function after a timeout.
235 * A block device may call blk_sync_queue to ensure that any
236 * such activity is cancelled, thus allowing it to release resources
237 * that the callbacks might use. The caller must already have made sure
238 * that its ->make_request_fn will not re-add plugging prior to calling
239 * this function.
240 *
241 * This function does not cancel any asynchronous activity arising
242 * out of elevator or throttling code. That would require elevaotor_exit()
243 * and blkcg_exit_queue() to be called with queue lock initialized.
244 *
245 */
246void blk_sync_queue(struct request_queue *q)
247{
248 del_timer_sync(&q->timeout);
249
250 if (q->mq_ops) {
251 struct blk_mq_hw_ctx *hctx;
252 int i;
253
254 queue_for_each_hw_ctx(q, hctx, i)
255 cancel_delayed_work_sync(&hctx->delayed_work);
256 } else {
257 cancel_delayed_work_sync(&q->delay_work);
258 }
259}
260EXPORT_SYMBOL(blk_sync_queue);
261
262/**
263 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
264 * @q: The queue to run
265 *
266 * Description:
267 * Invoke request handling on a queue if there are any pending requests.
268 * May be used to restart request handling after a request has completed.
269 * This variant runs the queue whether or not the queue has been
270 * stopped. Must be called with the queue lock held and interrupts
271 * disabled. See also @blk_run_queue.
272 */
273inline void __blk_run_queue_uncond(struct request_queue *q)
274{
275 if (unlikely(blk_queue_dead(q)))
276 return;
277
278 /*
279 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
280 * the queue lock internally. As a result multiple threads may be
281 * running such a request function concurrently. Keep track of the
282 * number of active request_fn invocations such that blk_drain_queue()
283 * can wait until all these request_fn calls have finished.
284 */
285 q->request_fn_active++;
286 q->request_fn(q);
287 q->request_fn_active--;
288}
289
290/**
291 * __blk_run_queue - run a single device queue
292 * @q: The queue to run
293 *
294 * Description:
295 * See @blk_run_queue. This variant must be called with the queue lock
296 * held and interrupts disabled.
297 */
298void __blk_run_queue(struct request_queue *q)
299{
300 if (unlikely(blk_queue_stopped(q)))
301 return;
302
303 __blk_run_queue_uncond(q);
304}
305EXPORT_SYMBOL(__blk_run_queue);
306
307/**
308 * blk_run_queue_async - run a single device queue in workqueue context
309 * @q: The queue to run
310 *
311 * Description:
312 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
313 * of us. The caller must hold the queue lock.
314 */
315void blk_run_queue_async(struct request_queue *q)
316{
317 if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
318 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
319}
320EXPORT_SYMBOL(blk_run_queue_async);
321
322/**
323 * blk_run_queue - run a single device queue
324 * @q: The queue to run
325 *
326 * Description:
327 * Invoke request handling on this queue, if it has pending work to do.
328 * May be used to restart queueing when a request has completed.
329 */
330void blk_run_queue(struct request_queue *q)
331{
332 unsigned long flags;
333
334 spin_lock_irqsave(q->queue_lock, flags);
335 __blk_run_queue(q);
336 spin_unlock_irqrestore(q->queue_lock, flags);
337}
338EXPORT_SYMBOL(blk_run_queue);
339
340void blk_put_queue(struct request_queue *q)
341{
342 kobject_put(&q->kobj);
343}
344EXPORT_SYMBOL(blk_put_queue);
345
346/**
347 * __blk_drain_queue - drain requests from request_queue
348 * @q: queue to drain
349 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
350 *
351 * Drain requests from @q. If @drain_all is set, all requests are drained.
352 * If not, only ELVPRIV requests are drained. The caller is responsible
353 * for ensuring that no new requests which need to be drained are queued.
354 */
355static void __blk_drain_queue(struct request_queue *q, bool drain_all)
356 __releases(q->queue_lock)
357 __acquires(q->queue_lock)
358{
359 int i;
360
361 lockdep_assert_held(q->queue_lock);
362
363 while (true) {
364 bool drain = false;
365
366 /*
367 * The caller might be trying to drain @q before its
368 * elevator is initialized.
369 */
370 if (q->elevator)
371 elv_drain_elevator(q);
372
373 blkcg_drain_queue(q);
374
375 /*
376 * This function might be called on a queue which failed
377 * driver init after queue creation or is not yet fully
378 * active yet. Some drivers (e.g. fd and loop) get unhappy
379 * in such cases. Kick queue iff dispatch queue has
380 * something on it and @q has request_fn set.
381 */
382 if (!list_empty(&q->queue_head) && q->request_fn)
383 __blk_run_queue(q);
384
385 drain |= q->nr_rqs_elvpriv;
386 drain |= q->request_fn_active;
387
388 /*
389 * Unfortunately, requests are queued at and tracked from
390 * multiple places and there's no single counter which can
391 * be drained. Check all the queues and counters.
392 */
393 if (drain_all) {
394 drain |= !list_empty(&q->queue_head);
395 for (i = 0; i < 2; i++) {
396 drain |= q->nr_rqs[i];
397 drain |= q->in_flight[i];
398 drain |= !list_empty(&q->flush_queue[i]);
399 }
400 }
401
402 if (!drain)
403 break;
404
405 spin_unlock_irq(q->queue_lock);
406
407 msleep(10);
408
409 spin_lock_irq(q->queue_lock);
410 }
411
412 /*
413 * With queue marked dead, any woken up waiter will fail the
414 * allocation path, so the wakeup chaining is lost and we're
415 * left with hung waiters. We need to wake up those waiters.
416 */
417 if (q->request_fn) {
418 struct request_list *rl;
419
420 blk_queue_for_each_rl(rl, q)
421 for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
422 wake_up_all(&rl->wait[i]);
423 }
424}
425
426/**
427 * blk_queue_bypass_start - enter queue bypass mode
428 * @q: queue of interest
429 *
430 * In bypass mode, only the dispatch FIFO queue of @q is used. This
431 * function makes @q enter bypass mode and drains all requests which were
432 * throttled or issued before. On return, it's guaranteed that no request
433 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
434 * inside queue or RCU read lock.
435 */
436void blk_queue_bypass_start(struct request_queue *q)
437{
438 bool drain;
439
440 spin_lock_irq(q->queue_lock);
441 drain = !q->bypass_depth++;
442 queue_flag_set(QUEUE_FLAG_BYPASS, q);
443 spin_unlock_irq(q->queue_lock);
444
445 if (drain) {
446 spin_lock_irq(q->queue_lock);
447 __blk_drain_queue(q, false);
448 spin_unlock_irq(q->queue_lock);
449
450 /* ensure blk_queue_bypass() is %true inside RCU read lock */
451 synchronize_rcu();
452 }
453}
454EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
455
456/**
457 * blk_queue_bypass_end - leave queue bypass mode
458 * @q: queue of interest
459 *
460 * Leave bypass mode and restore the normal queueing behavior.
461 */
462void blk_queue_bypass_end(struct request_queue *q)
463{
464 spin_lock_irq(q->queue_lock);
465 if (!--q->bypass_depth)
466 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
467 WARN_ON_ONCE(q->bypass_depth < 0);
468 spin_unlock_irq(q->queue_lock);
469}
470EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
471
472/**
473 * blk_cleanup_queue - shutdown a request queue
474 * @q: request queue to shutdown
475 *
476 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
477 * put it. All future requests will be failed immediately with -ENODEV.
478 */
479void blk_cleanup_queue(struct request_queue *q)
480{
481 spinlock_t *lock = q->queue_lock;
482
483 /* mark @q DYING, no new request or merges will be allowed afterwards */
484 mutex_lock(&q->sysfs_lock);
485 queue_flag_set_unlocked(QUEUE_FLAG_DYING, q);
486 spin_lock_irq(lock);
487
488 /*
489 * A dying queue is permanently in bypass mode till released. Note
490 * that, unlike blk_queue_bypass_start(), we aren't performing
491 * synchronize_rcu() after entering bypass mode to avoid the delay
492 * as some drivers create and destroy a lot of queues while
493 * probing. This is still safe because blk_release_queue() will be
494 * called only after the queue refcnt drops to zero and nothing,
495 * RCU or not, would be traversing the queue by then.
496 */
497 q->bypass_depth++;
498 queue_flag_set(QUEUE_FLAG_BYPASS, q);
499
500 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
501 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
502 queue_flag_set(QUEUE_FLAG_DYING, q);
503 spin_unlock_irq(lock);
504 mutex_unlock(&q->sysfs_lock);
505
506 /*
507 * Drain all requests queued before DYING marking. Set DEAD flag to
508 * prevent that q->request_fn() gets invoked after draining finished.
509 */
510 if (q->mq_ops) {
511 blk_mq_drain_queue(q);
512 spin_lock_irq(lock);
513 } else {
514 spin_lock_irq(lock);
515 __blk_drain_queue(q, true);
516 }
517 queue_flag_set(QUEUE_FLAG_DEAD, q);
518 spin_unlock_irq(lock);
519
520 /* @q won't process any more request, flush async actions */
521 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
522 blk_sync_queue(q);
523
524 spin_lock_irq(lock);
525 if (q->queue_lock != &q->__queue_lock)
526 q->queue_lock = &q->__queue_lock;
527 spin_unlock_irq(lock);
528
529 /* @q is and will stay empty, shutdown and put */
530 blk_put_queue(q);
531}
532EXPORT_SYMBOL(blk_cleanup_queue);
533
534int blk_init_rl(struct request_list *rl, struct request_queue *q,
535 gfp_t gfp_mask)
536{
537 if (unlikely(rl->rq_pool))
538 return 0;
539
540 rl->q = q;
541 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
542 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
543 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
544 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
545
546 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
547 mempool_free_slab, request_cachep,
548 gfp_mask, q->node);
549 if (!rl->rq_pool)
550 return -ENOMEM;
551
552 return 0;
553}
554
555void blk_exit_rl(struct request_list *rl)
556{
557 if (rl->rq_pool)
558 mempool_destroy(rl->rq_pool);
559}
560
561struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
562{
563 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
564}
565EXPORT_SYMBOL(blk_alloc_queue);
566
567struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
568{
569 struct request_queue *q;
570 int err;
571
572 q = kmem_cache_alloc_node(blk_requestq_cachep,
573 gfp_mask | __GFP_ZERO, node_id);
574 if (!q)
575 return NULL;
576
577 if (percpu_counter_init(&q->mq_usage_counter, 0))
578 goto fail_q;
579
580 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
581 if (q->id < 0)
582 goto fail_c;
583
584 q->backing_dev_info.ra_pages =
585 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
586 q->backing_dev_info.state = 0;
587 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
588 q->backing_dev_info.name = "block";
589 q->node = node_id;
590
591 err = bdi_init(&q->backing_dev_info);
592 if (err)
593 goto fail_id;
594
595 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
596 laptop_mode_timer_fn, (unsigned long) q);
597 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
598 INIT_LIST_HEAD(&q->queue_head);
599 INIT_LIST_HEAD(&q->timeout_list);
600 INIT_LIST_HEAD(&q->icq_list);
601#ifdef CONFIG_BLK_CGROUP
602 INIT_LIST_HEAD(&q->blkg_list);
603#endif
604 INIT_LIST_HEAD(&q->flush_queue[0]);
605 INIT_LIST_HEAD(&q->flush_queue[1]);
606 INIT_LIST_HEAD(&q->flush_data_in_flight);
607 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
608
609 kobject_init(&q->kobj, &blk_queue_ktype);
610
611 mutex_init(&q->sysfs_lock);
612 spin_lock_init(&q->__queue_lock);
613
614 /*
615 * By default initialize queue_lock to internal lock and driver can
616 * override it later if need be.
617 */
618 q->queue_lock = &q->__queue_lock;
619
620 /*
621 * A queue starts its life with bypass turned on to avoid
622 * unnecessary bypass on/off overhead and nasty surprises during
623 * init. The initial bypass will be finished when the queue is
624 * registered by blk_register_queue().
625 */
626 q->bypass_depth = 1;
627 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
628
629 init_waitqueue_head(&q->mq_freeze_wq);
630
631 if (blkcg_init_queue(q))
632 goto fail_bdi;
633
634 return q;
635
636fail_bdi:
637 bdi_destroy(&q->backing_dev_info);
638fail_id:
639 ida_simple_remove(&blk_queue_ida, q->id);
640fail_c:
641 percpu_counter_destroy(&q->mq_usage_counter);
642fail_q:
643 kmem_cache_free(blk_requestq_cachep, q);
644 return NULL;
645}
646EXPORT_SYMBOL(blk_alloc_queue_node);
647
648/**
649 * blk_init_queue - prepare a request queue for use with a block device
650 * @rfn: The function to be called to process requests that have been
651 * placed on the queue.
652 * @lock: Request queue spin lock
653 *
654 * Description:
655 * If a block device wishes to use the standard request handling procedures,
656 * which sorts requests and coalesces adjacent requests, then it must
657 * call blk_init_queue(). The function @rfn will be called when there
658 * are requests on the queue that need to be processed. If the device
659 * supports plugging, then @rfn may not be called immediately when requests
660 * are available on the queue, but may be called at some time later instead.
661 * Plugged queues are generally unplugged when a buffer belonging to one
662 * of the requests on the queue is needed, or due to memory pressure.
663 *
664 * @rfn is not required, or even expected, to remove all requests off the
665 * queue, but only as many as it can handle at a time. If it does leave
666 * requests on the queue, it is responsible for arranging that the requests
667 * get dealt with eventually.
668 *
669 * The queue spin lock must be held while manipulating the requests on the
670 * request queue; this lock will be taken also from interrupt context, so irq
671 * disabling is needed for it.
672 *
673 * Function returns a pointer to the initialized request queue, or %NULL if
674 * it didn't succeed.
675 *
676 * Note:
677 * blk_init_queue() must be paired with a blk_cleanup_queue() call
678 * when the block device is deactivated (such as at module unload).
679 **/
680
681struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
682{
683 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
684}
685EXPORT_SYMBOL(blk_init_queue);
686
687struct request_queue *
688blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
689{
690 struct request_queue *uninit_q, *q;
691
692 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
693 if (!uninit_q)
694 return NULL;
695
696 q = blk_init_allocated_queue(uninit_q, rfn, lock);
697 if (!q)
698 blk_cleanup_queue(uninit_q);
699
700 return q;
701}
702EXPORT_SYMBOL(blk_init_queue_node);
703
704struct request_queue *
705blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
706 spinlock_t *lock)
707{
708 if (!q)
709 return NULL;
710
711 q->flush_rq = kzalloc(sizeof(struct request), GFP_KERNEL);
712 if (!q->flush_rq)
713 return NULL;
714
715 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
716 goto fail;
717
718 q->request_fn = rfn;
719 q->prep_rq_fn = NULL;
720 q->unprep_rq_fn = NULL;
721 q->queue_flags |= QUEUE_FLAG_DEFAULT;
722
723 /* Override internal queue lock with supplied lock pointer */
724 if (lock)
725 q->queue_lock = lock;
726
727 /*
728 * This also sets hw/phys segments, boundary and size
729 */
730 blk_queue_make_request(q, blk_queue_bio);
731
732 q->sg_reserved_size = INT_MAX;
733
734 /* Protect q->elevator from elevator_change */
735 mutex_lock(&q->sysfs_lock);
736
737 /* init elevator */
738 if (elevator_init(q, NULL)) {
739 mutex_unlock(&q->sysfs_lock);
740 goto fail;
741 }
742
743 mutex_unlock(&q->sysfs_lock);
744
745 return q;
746
747fail:
748 kfree(q->flush_rq);
749 return NULL;
750}
751EXPORT_SYMBOL(blk_init_allocated_queue);
752
753bool blk_get_queue(struct request_queue *q)
754{
755 if (likely(!blk_queue_dying(q))) {
756 __blk_get_queue(q);
757 return true;
758 }
759
760 return false;
761}
762EXPORT_SYMBOL(blk_get_queue);
763
764static inline void blk_free_request(struct request_list *rl, struct request *rq)
765{
766 if (rq->cmd_flags & REQ_ELVPRIV) {
767 elv_put_request(rl->q, rq);
768 if (rq->elv.icq)
769 put_io_context(rq->elv.icq->ioc);
770 }
771
772 mempool_free(rq, rl->rq_pool);
773}
774
775/*
776 * ioc_batching returns true if the ioc is a valid batching request and
777 * should be given priority access to a request.
778 */
779static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
780{
781 if (!ioc)
782 return 0;
783
784 /*
785 * Make sure the process is able to allocate at least 1 request
786 * even if the batch times out, otherwise we could theoretically
787 * lose wakeups.
788 */
789 return ioc->nr_batch_requests == q->nr_batching ||
790 (ioc->nr_batch_requests > 0
791 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
792}
793
794/*
795 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
796 * will cause the process to be a "batcher" on all queues in the system. This
797 * is the behaviour we want though - once it gets a wakeup it should be given
798 * a nice run.
799 */
800static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
801{
802 if (!ioc || ioc_batching(q, ioc))
803 return;
804
805 ioc->nr_batch_requests = q->nr_batching;
806 ioc->last_waited = jiffies;
807}
808
809static void __freed_request(struct request_list *rl, int sync)
810{
811 struct request_queue *q = rl->q;
812
813 /*
814 * bdi isn't aware of blkcg yet. As all async IOs end up root
815 * blkcg anyway, just use root blkcg state.
816 */
817 if (rl == &q->root_rl &&
818 rl->count[sync] < queue_congestion_off_threshold(q))
819 blk_clear_queue_congested(q, sync);
820
821 if (rl->count[sync] + 1 <= q->nr_requests) {
822 if (waitqueue_active(&rl->wait[sync]))
823 wake_up(&rl->wait[sync]);
824
825 blk_clear_rl_full(rl, sync);
826 }
827}
828
829/*
830 * A request has just been released. Account for it, update the full and
831 * congestion status, wake up any waiters. Called under q->queue_lock.
832 */
833static void freed_request(struct request_list *rl, unsigned int flags)
834{
835 struct request_queue *q = rl->q;
836 int sync = rw_is_sync(flags);
837
838 q->nr_rqs[sync]--;
839 rl->count[sync]--;
840 if (flags & REQ_ELVPRIV)
841 q->nr_rqs_elvpriv--;
842
843 __freed_request(rl, sync);
844
845 if (unlikely(rl->starved[sync ^ 1]))
846 __freed_request(rl, sync ^ 1);
847}
848
849/*
850 * Determine if elevator data should be initialized when allocating the
851 * request associated with @bio.
852 */
853static bool blk_rq_should_init_elevator(struct bio *bio)
854{
855 if (!bio)
856 return true;
857
858 /*
859 * Flush requests do not use the elevator so skip initialization.
860 * This allows a request to share the flush and elevator data.
861 */
862 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
863 return false;
864
865 return true;
866}
867
868/**
869 * rq_ioc - determine io_context for request allocation
870 * @bio: request being allocated is for this bio (can be %NULL)
871 *
872 * Determine io_context to use for request allocation for @bio. May return
873 * %NULL if %current->io_context doesn't exist.
874 */
875static struct io_context *rq_ioc(struct bio *bio)
876{
877#ifdef CONFIG_BLK_CGROUP
878 if (bio && bio->bi_ioc)
879 return bio->bi_ioc;
880#endif
881 return current->io_context;
882}
883
884/**
885 * __get_request - get a free request
886 * @rl: request list to allocate from
887 * @rw_flags: RW and SYNC flags
888 * @bio: bio to allocate request for (can be %NULL)
889 * @gfp_mask: allocation mask
890 *
891 * Get a free request from @q. This function may fail under memory
892 * pressure or if @q is dead.
893 *
894 * Must be callled with @q->queue_lock held and,
895 * Returns %NULL on failure, with @q->queue_lock held.
896 * Returns !%NULL on success, with @q->queue_lock *not held*.
897 */
898static struct request *__get_request(struct request_list *rl, int rw_flags,
899 struct bio *bio, gfp_t gfp_mask)
900{
901 struct request_queue *q = rl->q;
902 struct request *rq;
903 struct elevator_type *et = q->elevator->type;
904 struct io_context *ioc = rq_ioc(bio);
905 struct io_cq *icq = NULL;
906 const bool is_sync = rw_is_sync(rw_flags) != 0;
907 int may_queue;
908
909 if (unlikely(blk_queue_dying(q)))
910 return NULL;
911
912 may_queue = elv_may_queue(q, rw_flags);
913 if (may_queue == ELV_MQUEUE_NO)
914 goto rq_starved;
915
916 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
917 if (rl->count[is_sync]+1 >= q->nr_requests) {
918 /*
919 * The queue will fill after this allocation, so set
920 * it as full, and mark this process as "batching".
921 * This process will be allowed to complete a batch of
922 * requests, others will be blocked.
923 */
924 if (!blk_rl_full(rl, is_sync)) {
925 ioc_set_batching(q, ioc);
926 blk_set_rl_full(rl, is_sync);
927 } else {
928 if (may_queue != ELV_MQUEUE_MUST
929 && !ioc_batching(q, ioc)) {
930 /*
931 * The queue is full and the allocating
932 * process is not a "batcher", and not
933 * exempted by the IO scheduler
934 */
935 return NULL;
936 }
937 }
938 }
939 /*
940 * bdi isn't aware of blkcg yet. As all async IOs end up
941 * root blkcg anyway, just use root blkcg state.
942 */
943 if (rl == &q->root_rl)
944 blk_set_queue_congested(q, is_sync);
945 }
946
947 /*
948 * Only allow batching queuers to allocate up to 50% over the defined
949 * limit of requests, otherwise we could have thousands of requests
950 * allocated with any setting of ->nr_requests
951 */
952 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
953 return NULL;
954
955 q->nr_rqs[is_sync]++;
956 rl->count[is_sync]++;
957 rl->starved[is_sync] = 0;
958
959 /*
960 * Decide whether the new request will be managed by elevator. If
961 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
962 * prevent the current elevator from being destroyed until the new
963 * request is freed. This guarantees icq's won't be destroyed and
964 * makes creating new ones safe.
965 *
966 * Also, lookup icq while holding queue_lock. If it doesn't exist,
967 * it will be created after releasing queue_lock.
968 */
969 if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
970 rw_flags |= REQ_ELVPRIV;
971 q->nr_rqs_elvpriv++;
972 if (et->icq_cache && ioc)
973 icq = ioc_lookup_icq(ioc, q);
974 }
975
976 if (blk_queue_io_stat(q))
977 rw_flags |= REQ_IO_STAT;
978 spin_unlock_irq(q->queue_lock);
979
980 /* allocate and init request */
981 rq = mempool_alloc(rl->rq_pool, gfp_mask);
982 if (!rq)
983 goto fail_alloc;
984
985 blk_rq_init(q, rq);
986 blk_rq_set_rl(rq, rl);
987 rq->cmd_flags = rw_flags | REQ_ALLOCED;
988
989 /* init elvpriv */
990 if (rw_flags & REQ_ELVPRIV) {
991 if (unlikely(et->icq_cache && !icq)) {
992 if (ioc)
993 icq = ioc_create_icq(ioc, q, gfp_mask);
994 if (!icq)
995 goto fail_elvpriv;
996 }
997
998 rq->elv.icq = icq;
999 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1000 goto fail_elvpriv;
1001
1002 /* @rq->elv.icq holds io_context until @rq is freed */
1003 if (icq)
1004 get_io_context(icq->ioc);
1005 }
1006out:
1007 /*
1008 * ioc may be NULL here, and ioc_batching will be false. That's
1009 * OK, if the queue is under the request limit then requests need
1010 * not count toward the nr_batch_requests limit. There will always
1011 * be some limit enforced by BLK_BATCH_TIME.
1012 */
1013 if (ioc_batching(q, ioc))
1014 ioc->nr_batch_requests--;
1015
1016 trace_block_getrq(q, bio, rw_flags & 1);
1017 return rq;
1018
1019fail_elvpriv:
1020 /*
1021 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1022 * and may fail indefinitely under memory pressure and thus
1023 * shouldn't stall IO. Treat this request as !elvpriv. This will
1024 * disturb iosched and blkcg but weird is bettern than dead.
1025 */
1026 printk_ratelimited(KERN_WARNING "%s: request aux data allocation failed, iosched may be disturbed\n",
1027 dev_name(q->backing_dev_info.dev));
1028
1029 rq->cmd_flags &= ~REQ_ELVPRIV;
1030 rq->elv.icq = NULL;
1031
1032 spin_lock_irq(q->queue_lock);
1033 q->nr_rqs_elvpriv--;
1034 spin_unlock_irq(q->queue_lock);
1035 goto out;
1036
1037fail_alloc:
1038 /*
1039 * Allocation failed presumably due to memory. Undo anything we
1040 * might have messed up.
1041 *
1042 * Allocating task should really be put onto the front of the wait
1043 * queue, but this is pretty rare.
1044 */
1045 spin_lock_irq(q->queue_lock);
1046 freed_request(rl, rw_flags);
1047
1048 /*
1049 * in the very unlikely event that allocation failed and no
1050 * requests for this direction was pending, mark us starved so that
1051 * freeing of a request in the other direction will notice
1052 * us. another possible fix would be to split the rq mempool into
1053 * READ and WRITE
1054 */
1055rq_starved:
1056 if (unlikely(rl->count[is_sync] == 0))
1057 rl->starved[is_sync] = 1;
1058 return NULL;
1059}
1060
1061/**
1062 * get_request - get a free request
1063 * @q: request_queue to allocate request from
1064 * @rw_flags: RW and SYNC flags
1065 * @bio: bio to allocate request for (can be %NULL)
1066 * @gfp_mask: allocation mask
1067 *
1068 * Get a free request from @q. If %__GFP_WAIT is set in @gfp_mask, this
1069 * function keeps retrying under memory pressure and fails iff @q is dead.
1070 *
1071 * Must be callled with @q->queue_lock held and,
1072 * Returns %NULL on failure, with @q->queue_lock held.
1073 * Returns !%NULL on success, with @q->queue_lock *not held*.
1074 */
1075static struct request *get_request(struct request_queue *q, int rw_flags,
1076 struct bio *bio, gfp_t gfp_mask)
1077{
1078 const bool is_sync = rw_is_sync(rw_flags) != 0;
1079 DEFINE_WAIT(wait);
1080 struct request_list *rl;
1081 struct request *rq;
1082
1083 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
1084retry:
1085 rq = __get_request(rl, rw_flags, bio, gfp_mask);
1086 if (rq)
1087 return rq;
1088
1089 if (!(gfp_mask & __GFP_WAIT) || unlikely(blk_queue_dying(q))) {
1090 blk_put_rl(rl);
1091 return NULL;
1092 }
1093
1094 /* wait on @rl and retry */
1095 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1096 TASK_UNINTERRUPTIBLE);
1097
1098 trace_block_sleeprq(q, bio, rw_flags & 1);
1099
1100 spin_unlock_irq(q->queue_lock);
1101 io_schedule();
1102
1103 /*
1104 * After sleeping, we become a "batching" process and will be able
1105 * to allocate at least one request, and up to a big batch of them
1106 * for a small period time. See ioc_batching, ioc_set_batching
1107 */
1108 ioc_set_batching(q, current->io_context);
1109
1110 spin_lock_irq(q->queue_lock);
1111 finish_wait(&rl->wait[is_sync], &wait);
1112
1113 goto retry;
1114}
1115
1116static struct request *blk_old_get_request(struct request_queue *q, int rw,
1117 gfp_t gfp_mask)
1118{
1119 struct request *rq;
1120
1121 BUG_ON(rw != READ && rw != WRITE);
1122
1123 /* create ioc upfront */
1124 create_io_context(gfp_mask, q->node);
1125
1126 spin_lock_irq(q->queue_lock);
1127 rq = get_request(q, rw, NULL, gfp_mask);
1128 if (!rq)
1129 spin_unlock_irq(q->queue_lock);
1130 /* q->queue_lock is unlocked at this point */
1131
1132 return rq;
1133}
1134
1135struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1136{
1137 if (q->mq_ops)
1138 return blk_mq_alloc_request(q, rw, gfp_mask);
1139 else
1140 return blk_old_get_request(q, rw, gfp_mask);
1141}
1142EXPORT_SYMBOL(blk_get_request);
1143
1144/**
1145 * blk_make_request - given a bio, allocate a corresponding struct request.
1146 * @q: target request queue
1147 * @bio: The bio describing the memory mappings that will be submitted for IO.
1148 * It may be a chained-bio properly constructed by block/bio layer.
1149 * @gfp_mask: gfp flags to be used for memory allocation
1150 *
1151 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1152 * type commands. Where the struct request needs to be farther initialized by
1153 * the caller. It is passed a &struct bio, which describes the memory info of
1154 * the I/O transfer.
1155 *
1156 * The caller of blk_make_request must make sure that bi_io_vec
1157 * are set to describe the memory buffers. That bio_data_dir() will return
1158 * the needed direction of the request. (And all bio's in the passed bio-chain
1159 * are properly set accordingly)
1160 *
1161 * If called under none-sleepable conditions, mapped bio buffers must not
1162 * need bouncing, by calling the appropriate masked or flagged allocator,
1163 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1164 * BUG.
1165 *
1166 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1167 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1168 * anything but the first bio in the chain. Otherwise you risk waiting for IO
1169 * completion of a bio that hasn't been submitted yet, thus resulting in a
1170 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1171 * of bio_alloc(), as that avoids the mempool deadlock.
1172 * If possible a big IO should be split into smaller parts when allocation
1173 * fails. Partial allocation should not be an error, or you risk a live-lock.
1174 */
1175struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1176 gfp_t gfp_mask)
1177{
1178 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1179
1180 if (unlikely(!rq))
1181 return ERR_PTR(-ENOMEM);
1182
1183 for_each_bio(bio) {
1184 struct bio *bounce_bio = bio;
1185 int ret;
1186
1187 blk_queue_bounce(q, &bounce_bio);
1188 ret = blk_rq_append_bio(q, rq, bounce_bio);
1189 if (unlikely(ret)) {
1190 blk_put_request(rq);
1191 return ERR_PTR(ret);
1192 }
1193 }
1194
1195 return rq;
1196}
1197EXPORT_SYMBOL(blk_make_request);
1198
1199/**
1200 * blk_requeue_request - put a request back on queue
1201 * @q: request queue where request should be inserted
1202 * @rq: request to be inserted
1203 *
1204 * Description:
1205 * Drivers often keep queueing requests until the hardware cannot accept
1206 * more, when that condition happens we need to put the request back
1207 * on the queue. Must be called with queue lock held.
1208 */
1209void blk_requeue_request(struct request_queue *q, struct request *rq)
1210{
1211 blk_delete_timer(rq);
1212 blk_clear_rq_complete(rq);
1213 trace_block_rq_requeue(q, rq);
1214
1215 if (blk_rq_tagged(rq))
1216 blk_queue_end_tag(q, rq);
1217
1218 BUG_ON(blk_queued_rq(rq));
1219
1220 elv_requeue_request(q, rq);
1221}
1222EXPORT_SYMBOL(blk_requeue_request);
1223
1224static void add_acct_request(struct request_queue *q, struct request *rq,
1225 int where)
1226{
1227 blk_account_io_start(rq, true);
1228 __elv_add_request(q, rq, where);
1229}
1230
1231static void part_round_stats_single(int cpu, struct hd_struct *part,
1232 unsigned long now)
1233{
1234 if (now == part->stamp)
1235 return;
1236
1237 if (part_in_flight(part)) {
1238 __part_stat_add(cpu, part, time_in_queue,
1239 part_in_flight(part) * (now - part->stamp));
1240 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1241 }
1242 part->stamp = now;
1243}
1244
1245/**
1246 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1247 * @cpu: cpu number for stats access
1248 * @part: target partition
1249 *
1250 * The average IO queue length and utilisation statistics are maintained
1251 * by observing the current state of the queue length and the amount of
1252 * time it has been in this state for.
1253 *
1254 * Normally, that accounting is done on IO completion, but that can result
1255 * in more than a second's worth of IO being accounted for within any one
1256 * second, leading to >100% utilisation. To deal with that, we call this
1257 * function to do a round-off before returning the results when reading
1258 * /proc/diskstats. This accounts immediately for all queue usage up to
1259 * the current jiffies and restarts the counters again.
1260 */
1261void part_round_stats(int cpu, struct hd_struct *part)
1262{
1263 unsigned long now = jiffies;
1264
1265 if (part->partno)
1266 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1267 part_round_stats_single(cpu, part, now);
1268}
1269EXPORT_SYMBOL_GPL(part_round_stats);
1270
1271#ifdef CONFIG_PM_RUNTIME
1272static void blk_pm_put_request(struct request *rq)
1273{
1274 if (rq->q->dev && !(rq->cmd_flags & REQ_PM) && !--rq->q->nr_pending)
1275 pm_runtime_mark_last_busy(rq->q->dev);
1276}
1277#else
1278static inline void blk_pm_put_request(struct request *rq) {}
1279#endif
1280
1281/*
1282 * queue lock must be held
1283 */
1284void __blk_put_request(struct request_queue *q, struct request *req)
1285{
1286 if (unlikely(!q))
1287 return;
1288
1289 if (q->mq_ops) {
1290 blk_mq_free_request(req);
1291 return;
1292 }
1293
1294 blk_pm_put_request(req);
1295
1296 elv_completed_request(q, req);
1297
1298 /* this is a bio leak */
1299 WARN_ON(req->bio != NULL);
1300
1301 /*
1302 * Request may not have originated from ll_rw_blk. if not,
1303 * it didn't come out of our reserved rq pools
1304 */
1305 if (req->cmd_flags & REQ_ALLOCED) {
1306 unsigned int flags = req->cmd_flags;
1307 struct request_list *rl = blk_rq_rl(req);
1308
1309 BUG_ON(!list_empty(&req->queuelist));
1310 BUG_ON(ELV_ON_HASH(req));
1311
1312 blk_free_request(rl, req);
1313 freed_request(rl, flags);
1314 blk_put_rl(rl);
1315 }
1316}
1317EXPORT_SYMBOL_GPL(__blk_put_request);
1318
1319void blk_put_request(struct request *req)
1320{
1321 struct request_queue *q = req->q;
1322
1323 if (q->mq_ops)
1324 blk_mq_free_request(req);
1325 else {
1326 unsigned long flags;
1327
1328 spin_lock_irqsave(q->queue_lock, flags);
1329 __blk_put_request(q, req);
1330 spin_unlock_irqrestore(q->queue_lock, flags);
1331 }
1332}
1333EXPORT_SYMBOL(blk_put_request);
1334
1335/**
1336 * blk_add_request_payload - add a payload to a request
1337 * @rq: request to update
1338 * @page: page backing the payload
1339 * @len: length of the payload.
1340 *
1341 * This allows to later add a payload to an already submitted request by
1342 * a block driver. The driver needs to take care of freeing the payload
1343 * itself.
1344 *
1345 * Note that this is a quite horrible hack and nothing but handling of
1346 * discard requests should ever use it.
1347 */
1348void blk_add_request_payload(struct request *rq, struct page *page,
1349 unsigned int len)
1350{
1351 struct bio *bio = rq->bio;
1352
1353 bio->bi_io_vec->bv_page = page;
1354 bio->bi_io_vec->bv_offset = 0;
1355 bio->bi_io_vec->bv_len = len;
1356
1357 bio->bi_iter.bi_size = len;
1358 bio->bi_vcnt = 1;
1359 bio->bi_phys_segments = 1;
1360
1361 rq->__data_len = rq->resid_len = len;
1362 rq->nr_phys_segments = 1;
1363 rq->buffer = bio_data(bio);
1364}
1365EXPORT_SYMBOL_GPL(blk_add_request_payload);
1366
1367bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1368 struct bio *bio)
1369{
1370 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1371
1372 if (!ll_back_merge_fn(q, req, bio))
1373 return false;
1374
1375 trace_block_bio_backmerge(q, req, bio);
1376
1377 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1378 blk_rq_set_mixed_merge(req);
1379
1380 req->biotail->bi_next = bio;
1381 req->biotail = bio;
1382 req->__data_len += bio->bi_iter.bi_size;
1383 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1384
1385 blk_account_io_start(req, false);
1386 return true;
1387}
1388
1389bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1390 struct bio *bio)
1391{
1392 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1393
1394 if (!ll_front_merge_fn(q, req, bio))
1395 return false;
1396
1397 trace_block_bio_frontmerge(q, req, bio);
1398
1399 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1400 blk_rq_set_mixed_merge(req);
1401
1402 bio->bi_next = req->bio;
1403 req->bio = bio;
1404
1405 /*
1406 * may not be valid. if the low level driver said
1407 * it didn't need a bounce buffer then it better
1408 * not touch req->buffer either...
1409 */
1410 req->buffer = bio_data(bio);
1411 req->__sector = bio->bi_iter.bi_sector;
1412 req->__data_len += bio->bi_iter.bi_size;
1413 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1414
1415 blk_account_io_start(req, false);
1416 return true;
1417}
1418
1419/**
1420 * blk_attempt_plug_merge - try to merge with %current's plugged list
1421 * @q: request_queue new bio is being queued at
1422 * @bio: new bio being queued
1423 * @request_count: out parameter for number of traversed plugged requests
1424 *
1425 * Determine whether @bio being queued on @q can be merged with a request
1426 * on %current's plugged list. Returns %true if merge was successful,
1427 * otherwise %false.
1428 *
1429 * Plugging coalesces IOs from the same issuer for the same purpose without
1430 * going through @q->queue_lock. As such it's more of an issuing mechanism
1431 * than scheduling, and the request, while may have elvpriv data, is not
1432 * added on the elevator at this point. In addition, we don't have
1433 * reliable access to the elevator outside queue lock. Only check basic
1434 * merging parameters without querying the elevator.
1435 */
1436bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1437 unsigned int *request_count)
1438{
1439 struct blk_plug *plug;
1440 struct request *rq;
1441 bool ret = false;
1442 struct list_head *plug_list;
1443
1444 if (blk_queue_nomerges(q))
1445 goto out;
1446
1447 plug = current->plug;
1448 if (!plug)
1449 goto out;
1450 *request_count = 0;
1451
1452 if (q->mq_ops)
1453 plug_list = &plug->mq_list;
1454 else
1455 plug_list = &plug->list;
1456
1457 list_for_each_entry_reverse(rq, plug_list, queuelist) {
1458 int el_ret;
1459
1460 if (rq->q == q)
1461 (*request_count)++;
1462
1463 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
1464 continue;
1465
1466 el_ret = blk_try_merge(rq, bio);
1467 if (el_ret == ELEVATOR_BACK_MERGE) {
1468 ret = bio_attempt_back_merge(q, rq, bio);
1469 if (ret)
1470 break;
1471 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1472 ret = bio_attempt_front_merge(q, rq, bio);
1473 if (ret)
1474 break;
1475 }
1476 }
1477out:
1478 return ret;
1479}
1480
1481void init_request_from_bio(struct request *req, struct bio *bio)
1482{
1483 req->cmd_type = REQ_TYPE_FS;
1484
1485 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1486 if (bio->bi_rw & REQ_RAHEAD)
1487 req->cmd_flags |= REQ_FAILFAST_MASK;
1488
1489 req->errors = 0;
1490 req->__sector = bio->bi_iter.bi_sector;
1491 req->ioprio = bio_prio(bio);
1492 blk_rq_bio_prep(req->q, req, bio);
1493}
1494
1495void blk_queue_bio(struct request_queue *q, struct bio *bio)
1496{
1497 const bool sync = !!(bio->bi_rw & REQ_SYNC);
1498 struct blk_plug *plug;
1499 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1500 struct request *req;
1501 unsigned int request_count = 0;
1502
1503 /*
1504 * low level driver can indicate that it wants pages above a
1505 * certain limit bounced to low memory (ie for highmem, or even
1506 * ISA dma in theory)
1507 */
1508 blk_queue_bounce(q, &bio);
1509
1510 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
1511 bio_endio(bio, -EIO);
1512 return;
1513 }
1514
1515 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
1516 spin_lock_irq(q->queue_lock);
1517 where = ELEVATOR_INSERT_FLUSH;
1518 goto get_rq;
1519 }
1520
1521 /*
1522 * Check if we can merge with the plugged list before grabbing
1523 * any locks.
1524 */
1525 if (blk_attempt_plug_merge(q, bio, &request_count))
1526 return;
1527
1528 spin_lock_irq(q->queue_lock);
1529
1530 el_ret = elv_merge(q, &req, bio);
1531 if (el_ret == ELEVATOR_BACK_MERGE) {
1532 if (bio_attempt_back_merge(q, req, bio)) {
1533 elv_bio_merged(q, req, bio);
1534 if (!attempt_back_merge(q, req))
1535 elv_merged_request(q, req, el_ret);
1536 goto out_unlock;
1537 }
1538 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1539 if (bio_attempt_front_merge(q, req, bio)) {
1540 elv_bio_merged(q, req, bio);
1541 if (!attempt_front_merge(q, req))
1542 elv_merged_request(q, req, el_ret);
1543 goto out_unlock;
1544 }
1545 }
1546
1547get_rq:
1548 /*
1549 * This sync check and mask will be re-done in init_request_from_bio(),
1550 * but we need to set it earlier to expose the sync flag to the
1551 * rq allocator and io schedulers.
1552 */
1553 rw_flags = bio_data_dir(bio);
1554 if (sync)
1555 rw_flags |= REQ_SYNC;
1556
1557 /*
1558 * Grab a free request. This is might sleep but can not fail.
1559 * Returns with the queue unlocked.
1560 */
1561 req = get_request(q, rw_flags, bio, GFP_NOIO);
1562 if (unlikely(!req)) {
1563 bio_endio(bio, -ENODEV); /* @q is dead */
1564 goto out_unlock;
1565 }
1566
1567 /*
1568 * After dropping the lock and possibly sleeping here, our request
1569 * may now be mergeable after it had proven unmergeable (above).
1570 * We don't worry about that case for efficiency. It won't happen
1571 * often, and the elevators are able to handle it.
1572 */
1573 init_request_from_bio(req, bio);
1574
1575 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
1576 req->cpu = raw_smp_processor_id();
1577
1578 plug = current->plug;
1579 if (plug) {
1580 /*
1581 * If this is the first request added after a plug, fire
1582 * of a plug trace.
1583 */
1584 if (!request_count)
1585 trace_block_plug(q);
1586 else {
1587 if (request_count >= BLK_MAX_REQUEST_COUNT) {
1588 blk_flush_plug_list(plug, false);
1589 trace_block_plug(q);
1590 }
1591 }
1592 list_add_tail(&req->queuelist, &plug->list);
1593 blk_account_io_start(req, true);
1594 } else {
1595 spin_lock_irq(q->queue_lock);
1596 add_acct_request(q, req, where);
1597 __blk_run_queue(q);
1598out_unlock:
1599 spin_unlock_irq(q->queue_lock);
1600 }
1601}
1602EXPORT_SYMBOL_GPL(blk_queue_bio); /* for device mapper only */
1603
1604/*
1605 * If bio->bi_dev is a partition, remap the location
1606 */
1607static inline void blk_partition_remap(struct bio *bio)
1608{
1609 struct block_device *bdev = bio->bi_bdev;
1610
1611 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1612 struct hd_struct *p = bdev->bd_part;
1613
1614 bio->bi_iter.bi_sector += p->start_sect;
1615 bio->bi_bdev = bdev->bd_contains;
1616
1617 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1618 bdev->bd_dev,
1619 bio->bi_iter.bi_sector - p->start_sect);
1620 }
1621}
1622
1623static void handle_bad_sector(struct bio *bio)
1624{
1625 char b[BDEVNAME_SIZE];
1626
1627 printk(KERN_INFO "attempt to access beyond end of device\n");
1628 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1629 bdevname(bio->bi_bdev, b),
1630 bio->bi_rw,
1631 (unsigned long long)bio_end_sector(bio),
1632 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1633
1634 set_bit(BIO_EOF, &bio->bi_flags);
1635}
1636
1637#ifdef CONFIG_FAIL_MAKE_REQUEST
1638
1639static DECLARE_FAULT_ATTR(fail_make_request);
1640
1641static int __init setup_fail_make_request(char *str)
1642{
1643 return setup_fault_attr(&fail_make_request, str);
1644}
1645__setup("fail_make_request=", setup_fail_make_request);
1646
1647static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1648{
1649 return part->make_it_fail && should_fail(&fail_make_request, bytes);
1650}
1651
1652static int __init fail_make_request_debugfs(void)
1653{
1654 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1655 NULL, &fail_make_request);
1656
1657 return IS_ERR(dir) ? PTR_ERR(dir) : 0;
1658}
1659
1660late_initcall(fail_make_request_debugfs);
1661
1662#else /* CONFIG_FAIL_MAKE_REQUEST */
1663
1664static inline bool should_fail_request(struct hd_struct *part,
1665 unsigned int bytes)
1666{
1667 return false;
1668}
1669
1670#endif /* CONFIG_FAIL_MAKE_REQUEST */
1671
1672/*
1673 * Check whether this bio extends beyond the end of the device.
1674 */
1675static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1676{
1677 sector_t maxsector;
1678
1679 if (!nr_sectors)
1680 return 0;
1681
1682 /* Test device or partition size, when known. */
1683 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1684 if (maxsector) {
1685 sector_t sector = bio->bi_iter.bi_sector;
1686
1687 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1688 /*
1689 * This may well happen - the kernel calls bread()
1690 * without checking the size of the device, e.g., when
1691 * mounting a device.
1692 */
1693 handle_bad_sector(bio);
1694 return 1;
1695 }
1696 }
1697
1698 return 0;
1699}
1700
1701static noinline_for_stack bool
1702generic_make_request_checks(struct bio *bio)
1703{
1704 struct request_queue *q;
1705 int nr_sectors = bio_sectors(bio);
1706 int err = -EIO;
1707 char b[BDEVNAME_SIZE];
1708 struct hd_struct *part;
1709
1710 might_sleep();
1711
1712 if (bio_check_eod(bio, nr_sectors))
1713 goto end_io;
1714
1715 q = bdev_get_queue(bio->bi_bdev);
1716 if (unlikely(!q)) {
1717 printk(KERN_ERR
1718 "generic_make_request: Trying to access "
1719 "nonexistent block-device %s (%Lu)\n",
1720 bdevname(bio->bi_bdev, b),
1721 (long long) bio->bi_iter.bi_sector);
1722 goto end_io;
1723 }
1724
1725 if (likely(bio_is_rw(bio) &&
1726 nr_sectors > queue_max_hw_sectors(q))) {
1727 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1728 bdevname(bio->bi_bdev, b),
1729 bio_sectors(bio),
1730 queue_max_hw_sectors(q));
1731 goto end_io;
1732 }
1733
1734 part = bio->bi_bdev->bd_part;
1735 if (should_fail_request(part, bio->bi_iter.bi_size) ||
1736 should_fail_request(&part_to_disk(part)->part0,
1737 bio->bi_iter.bi_size))
1738 goto end_io;
1739
1740 /*
1741 * If this device has partitions, remap block n
1742 * of partition p to block n+start(p) of the disk.
1743 */
1744 blk_partition_remap(bio);
1745
1746 if (bio_check_eod(bio, nr_sectors))
1747 goto end_io;
1748
1749 /*
1750 * Filter flush bio's early so that make_request based
1751 * drivers without flush support don't have to worry
1752 * about them.
1753 */
1754 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1755 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1756 if (!nr_sectors) {
1757 err = 0;
1758 goto end_io;
1759 }
1760 }
1761
1762 if ((bio->bi_rw & REQ_DISCARD) &&
1763 (!blk_queue_discard(q) ||
1764 ((bio->bi_rw & REQ_SECURE) && !blk_queue_secdiscard(q)))) {
1765 err = -EOPNOTSUPP;
1766 goto end_io;
1767 }
1768
1769 if (bio->bi_rw & REQ_WRITE_SAME && !bdev_write_same(bio->bi_bdev)) {
1770 err = -EOPNOTSUPP;
1771 goto end_io;
1772 }
1773
1774 /*
1775 * Various block parts want %current->io_context and lazy ioc
1776 * allocation ends up trading a lot of pain for a small amount of
1777 * memory. Just allocate it upfront. This may fail and block
1778 * layer knows how to live with it.
1779 */
1780 create_io_context(GFP_ATOMIC, q->node);
1781
1782 if (blk_throtl_bio(q, bio))
1783 return false; /* throttled, will be resubmitted later */
1784
1785 trace_block_bio_queue(q, bio);
1786 return true;
1787
1788end_io:
1789 bio_endio(bio, err);
1790 return false;
1791}
1792
1793/**
1794 * generic_make_request - hand a buffer to its device driver for I/O
1795 * @bio: The bio describing the location in memory and on the device.
1796 *
1797 * generic_make_request() is used to make I/O requests of block
1798 * devices. It is passed a &struct bio, which describes the I/O that needs
1799 * to be done.
1800 *
1801 * generic_make_request() does not return any status. The
1802 * success/failure status of the request, along with notification of
1803 * completion, is delivered asynchronously through the bio->bi_end_io
1804 * function described (one day) else where.
1805 *
1806 * The caller of generic_make_request must make sure that bi_io_vec
1807 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1808 * set to describe the device address, and the
1809 * bi_end_io and optionally bi_private are set to describe how
1810 * completion notification should be signaled.
1811 *
1812 * generic_make_request and the drivers it calls may use bi_next if this
1813 * bio happens to be merged with someone else, and may resubmit the bio to
1814 * a lower device by calling into generic_make_request recursively, which
1815 * means the bio should NOT be touched after the call to ->make_request_fn.
1816 */
1817void generic_make_request(struct bio *bio)
1818{
1819 struct bio_list bio_list_on_stack;
1820
1821 if (!generic_make_request_checks(bio))
1822 return;
1823
1824 /*
1825 * We only want one ->make_request_fn to be active at a time, else
1826 * stack usage with stacked devices could be a problem. So use
1827 * current->bio_list to keep a list of requests submited by a
1828 * make_request_fn function. current->bio_list is also used as a
1829 * flag to say if generic_make_request is currently active in this
1830 * task or not. If it is NULL, then no make_request is active. If
1831 * it is non-NULL, then a make_request is active, and new requests
1832 * should be added at the tail
1833 */
1834 if (current->bio_list) {
1835 bio_list_add(current->bio_list, bio);
1836 return;
1837 }
1838
1839 /* following loop may be a bit non-obvious, and so deserves some
1840 * explanation.
1841 * Before entering the loop, bio->bi_next is NULL (as all callers
1842 * ensure that) so we have a list with a single bio.
1843 * We pretend that we have just taken it off a longer list, so
1844 * we assign bio_list to a pointer to the bio_list_on_stack,
1845 * thus initialising the bio_list of new bios to be
1846 * added. ->make_request() may indeed add some more bios
1847 * through a recursive call to generic_make_request. If it
1848 * did, we find a non-NULL value in bio_list and re-enter the loop
1849 * from the top. In this case we really did just take the bio
1850 * of the top of the list (no pretending) and so remove it from
1851 * bio_list, and call into ->make_request() again.
1852 */
1853 BUG_ON(bio->bi_next);
1854 bio_list_init(&bio_list_on_stack);
1855 current->bio_list = &bio_list_on_stack;
1856 do {
1857 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1858
1859 q->make_request_fn(q, bio);
1860
1861 bio = bio_list_pop(current->bio_list);
1862 } while (bio);
1863 current->bio_list = NULL; /* deactivate */
1864}
1865EXPORT_SYMBOL(generic_make_request);
1866
1867/**
1868 * submit_bio - submit a bio to the block device layer for I/O
1869 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1870 * @bio: The &struct bio which describes the I/O
1871 *
1872 * submit_bio() is very similar in purpose to generic_make_request(), and
1873 * uses that function to do most of the work. Both are fairly rough
1874 * interfaces; @bio must be presetup and ready for I/O.
1875 *
1876 */
1877void submit_bio(int rw, struct bio *bio)
1878{
1879 bio->bi_rw |= rw;
1880
1881 /*
1882 * If it's a regular read/write or a barrier with data attached,
1883 * go through the normal accounting stuff before submission.
1884 */
1885 if (bio_has_data(bio)) {
1886 unsigned int count;
1887
1888 if (unlikely(rw & REQ_WRITE_SAME))
1889 count = bdev_logical_block_size(bio->bi_bdev) >> 9;
1890 else
1891 count = bio_sectors(bio);
1892
1893 if (rw & WRITE) {
1894 count_vm_events(PGPGOUT, count);
1895 } else {
1896 task_io_account_read(bio->bi_iter.bi_size);
1897 count_vm_events(PGPGIN, count);
1898 }
1899
1900 if (unlikely(block_dump)) {
1901 char b[BDEVNAME_SIZE];
1902 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1903 current->comm, task_pid_nr(current),
1904 (rw & WRITE) ? "WRITE" : "READ",
1905 (unsigned long long)bio->bi_iter.bi_sector,
1906 bdevname(bio->bi_bdev, b),
1907 count);
1908 }
1909 }
1910
1911 generic_make_request(bio);
1912}
1913EXPORT_SYMBOL(submit_bio);
1914
1915/**
1916 * blk_rq_check_limits - Helper function to check a request for the queue limit
1917 * @q: the queue
1918 * @rq: the request being checked
1919 *
1920 * Description:
1921 * @rq may have been made based on weaker limitations of upper-level queues
1922 * in request stacking drivers, and it may violate the limitation of @q.
1923 * Since the block layer and the underlying device driver trust @rq
1924 * after it is inserted to @q, it should be checked against @q before
1925 * the insertion using this generic function.
1926 *
1927 * This function should also be useful for request stacking drivers
1928 * in some cases below, so export this function.
1929 * Request stacking drivers like request-based dm may change the queue
1930 * limits while requests are in the queue (e.g. dm's table swapping).
1931 * Such request stacking drivers should check those requests against
1932 * the new queue limits again when they dispatch those requests,
1933 * although such checkings are also done against the old queue limits
1934 * when submitting requests.
1935 */
1936int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1937{
1938 if (!rq_mergeable(rq))
1939 return 0;
1940
1941 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, rq->cmd_flags)) {
1942 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1943 return -EIO;
1944 }
1945
1946 /*
1947 * queue's settings related to segment counting like q->bounce_pfn
1948 * may differ from that of other stacking queues.
1949 * Recalculate it to check the request correctly on this queue's
1950 * limitation.
1951 */
1952 blk_recalc_rq_segments(rq);
1953 if (rq->nr_phys_segments > queue_max_segments(q)) {
1954 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1955 return -EIO;
1956 }
1957
1958 return 0;
1959}
1960EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1961
1962/**
1963 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1964 * @q: the queue to submit the request
1965 * @rq: the request being queued
1966 */
1967int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1968{
1969 unsigned long flags;
1970 int where = ELEVATOR_INSERT_BACK;
1971
1972 if (blk_rq_check_limits(q, rq))
1973 return -EIO;
1974
1975 if (rq->rq_disk &&
1976 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
1977 return -EIO;
1978
1979 spin_lock_irqsave(q->queue_lock, flags);
1980 if (unlikely(blk_queue_dying(q))) {
1981 spin_unlock_irqrestore(q->queue_lock, flags);
1982 return -ENODEV;
1983 }
1984
1985 /*
1986 * Submitting request must be dequeued before calling this function
1987 * because it will be linked to another request_queue
1988 */
1989 BUG_ON(blk_queued_rq(rq));
1990
1991 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
1992 where = ELEVATOR_INSERT_FLUSH;
1993
1994 add_acct_request(q, rq, where);
1995 if (where == ELEVATOR_INSERT_FLUSH)
1996 __blk_run_queue(q);
1997 spin_unlock_irqrestore(q->queue_lock, flags);
1998
1999 return 0;
2000}
2001EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2002
2003/**
2004 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2005 * @rq: request to examine
2006 *
2007 * Description:
2008 * A request could be merge of IOs which require different failure
2009 * handling. This function determines the number of bytes which
2010 * can be failed from the beginning of the request without
2011 * crossing into area which need to be retried further.
2012 *
2013 * Return:
2014 * The number of bytes to fail.
2015 *
2016 * Context:
2017 * queue_lock must be held.
2018 */
2019unsigned int blk_rq_err_bytes(const struct request *rq)
2020{
2021 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2022 unsigned int bytes = 0;
2023 struct bio *bio;
2024
2025 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
2026 return blk_rq_bytes(rq);
2027
2028 /*
2029 * Currently the only 'mixing' which can happen is between
2030 * different fastfail types. We can safely fail portions
2031 * which have all the failfast bits that the first one has -
2032 * the ones which are at least as eager to fail as the first
2033 * one.
2034 */
2035 for (bio = rq->bio; bio; bio = bio->bi_next) {
2036 if ((bio->bi_rw & ff) != ff)
2037 break;
2038 bytes += bio->bi_iter.bi_size;
2039 }
2040
2041 /* this could lead to infinite loop */
2042 BUG_ON(blk_rq_bytes(rq) && !bytes);
2043 return bytes;
2044}
2045EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2046
2047void blk_account_io_completion(struct request *req, unsigned int bytes)
2048{
2049 if (blk_do_io_stat(req)) {
2050 const int rw = rq_data_dir(req);
2051 struct hd_struct *part;
2052 int cpu;
2053
2054 cpu = part_stat_lock();
2055 part = req->part;
2056 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2057 part_stat_unlock();
2058 }
2059}
2060
2061void blk_account_io_done(struct request *req)
2062{
2063 /*
2064 * Account IO completion. flush_rq isn't accounted as a
2065 * normal IO on queueing nor completion. Accounting the
2066 * containing request is enough.
2067 */
2068 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
2069 unsigned long duration = jiffies - req->start_time;
2070 const int rw = rq_data_dir(req);
2071 struct hd_struct *part;
2072 int cpu;
2073
2074 cpu = part_stat_lock();
2075 part = req->part;
2076
2077 part_stat_inc(cpu, part, ios[rw]);
2078 part_stat_add(cpu, part, ticks[rw], duration);
2079 part_round_stats(cpu, part);
2080 part_dec_in_flight(part, rw);
2081
2082 hd_struct_put(part);
2083 part_stat_unlock();
2084 }
2085}
2086
2087#ifdef CONFIG_PM_RUNTIME
2088/*
2089 * Don't process normal requests when queue is suspended
2090 * or in the process of suspending/resuming
2091 */
2092static struct request *blk_pm_peek_request(struct request_queue *q,
2093 struct request *rq)
2094{
2095 if (q->dev && (q->rpm_status == RPM_SUSPENDED ||
2096 (q->rpm_status != RPM_ACTIVE && !(rq->cmd_flags & REQ_PM))))
2097 return NULL;
2098 else
2099 return rq;
2100}
2101#else
2102static inline struct request *blk_pm_peek_request(struct request_queue *q,
2103 struct request *rq)
2104{
2105 return rq;
2106}
2107#endif
2108
2109void blk_account_io_start(struct request *rq, bool new_io)
2110{
2111 struct hd_struct *part;
2112 int rw = rq_data_dir(rq);
2113 int cpu;
2114
2115 if (!blk_do_io_stat(rq))
2116 return;
2117
2118 cpu = part_stat_lock();
2119
2120 if (!new_io) {
2121 part = rq->part;
2122 part_stat_inc(cpu, part, merges[rw]);
2123 } else {
2124 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2125 if (!hd_struct_try_get(part)) {
2126 /*
2127 * The partition is already being removed,
2128 * the request will be accounted on the disk only
2129 *
2130 * We take a reference on disk->part0 although that
2131 * partition will never be deleted, so we can treat
2132 * it as any other partition.
2133 */
2134 part = &rq->rq_disk->part0;
2135 hd_struct_get(part);
2136 }
2137 part_round_stats(cpu, part);
2138 part_inc_in_flight(part, rw);
2139 rq->part = part;
2140 }
2141
2142 part_stat_unlock();
2143}
2144
2145/**
2146 * blk_peek_request - peek at the top of a request queue
2147 * @q: request queue to peek at
2148 *
2149 * Description:
2150 * Return the request at the top of @q. The returned request
2151 * should be started using blk_start_request() before LLD starts
2152 * processing it.
2153 *
2154 * Return:
2155 * Pointer to the request at the top of @q if available. Null
2156 * otherwise.
2157 *
2158 * Context:
2159 * queue_lock must be held.
2160 */
2161struct request *blk_peek_request(struct request_queue *q)
2162{
2163 struct request *rq;
2164 int ret;
2165
2166 while ((rq = __elv_next_request(q)) != NULL) {
2167
2168 rq = blk_pm_peek_request(q, rq);
2169 if (!rq)
2170 break;
2171
2172 if (!(rq->cmd_flags & REQ_STARTED)) {
2173 /*
2174 * This is the first time the device driver
2175 * sees this request (possibly after
2176 * requeueing). Notify IO scheduler.
2177 */
2178 if (rq->cmd_flags & REQ_SORTED)
2179 elv_activate_rq(q, rq);
2180
2181 /*
2182 * just mark as started even if we don't start
2183 * it, a request that has been delayed should
2184 * not be passed by new incoming requests
2185 */
2186 rq->cmd_flags |= REQ_STARTED;
2187 trace_block_rq_issue(q, rq);
2188 }
2189
2190 if (!q->boundary_rq || q->boundary_rq == rq) {
2191 q->end_sector = rq_end_sector(rq);
2192 q->boundary_rq = NULL;
2193 }
2194
2195 if (rq->cmd_flags & REQ_DONTPREP)
2196 break;
2197
2198 if (q->dma_drain_size && blk_rq_bytes(rq)) {
2199 /*
2200 * make sure space for the drain appears we
2201 * know we can do this because max_hw_segments
2202 * has been adjusted to be one fewer than the
2203 * device can handle
2204 */
2205 rq->nr_phys_segments++;
2206 }
2207
2208 if (!q->prep_rq_fn)
2209 break;
2210
2211 ret = q->prep_rq_fn(q, rq);
2212 if (ret == BLKPREP_OK) {
2213 break;
2214 } else if (ret == BLKPREP_DEFER) {
2215 /*
2216 * the request may have been (partially) prepped.
2217 * we need to keep this request in the front to
2218 * avoid resource deadlock. REQ_STARTED will
2219 * prevent other fs requests from passing this one.
2220 */
2221 if (q->dma_drain_size && blk_rq_bytes(rq) &&
2222 !(rq->cmd_flags & REQ_DONTPREP)) {
2223 /*
2224 * remove the space for the drain we added
2225 * so that we don't add it again
2226 */
2227 --rq->nr_phys_segments;
2228 }
2229
2230 rq = NULL;
2231 break;
2232 } else if (ret == BLKPREP_KILL) {
2233 rq->cmd_flags |= REQ_QUIET;
2234 /*
2235 * Mark this request as started so we don't trigger
2236 * any debug logic in the end I/O path.
2237 */
2238 blk_start_request(rq);
2239 __blk_end_request_all(rq, -EIO);
2240 } else {
2241 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2242 break;
2243 }
2244 }
2245
2246 return rq;
2247}
2248EXPORT_SYMBOL(blk_peek_request);
2249
2250void blk_dequeue_request(struct request *rq)
2251{
2252 struct request_queue *q = rq->q;
2253
2254 BUG_ON(list_empty(&rq->queuelist));
2255 BUG_ON(ELV_ON_HASH(rq));
2256
2257 list_del_init(&rq->queuelist);
2258
2259 /*
2260 * the time frame between a request being removed from the lists
2261 * and to it is freed is accounted as io that is in progress at
2262 * the driver side.
2263 */
2264 if (blk_account_rq(rq)) {
2265 q->in_flight[rq_is_sync(rq)]++;
2266 set_io_start_time_ns(rq);
2267 }
2268}
2269
2270/**
2271 * blk_start_request - start request processing on the driver
2272 * @req: request to dequeue
2273 *
2274 * Description:
2275 * Dequeue @req and start timeout timer on it. This hands off the
2276 * request to the driver.
2277 *
2278 * Block internal functions which don't want to start timer should
2279 * call blk_dequeue_request().
2280 *
2281 * Context:
2282 * queue_lock must be held.
2283 */
2284void blk_start_request(struct request *req)
2285{
2286 blk_dequeue_request(req);
2287
2288 /*
2289 * We are now handing the request to the hardware, initialize
2290 * resid_len to full count and add the timeout handler.
2291 */
2292 req->resid_len = blk_rq_bytes(req);
2293 if (unlikely(blk_bidi_rq(req)))
2294 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2295
2296 BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags));
2297 blk_add_timer(req);
2298}
2299EXPORT_SYMBOL(blk_start_request);
2300
2301/**
2302 * blk_fetch_request - fetch a request from a request queue
2303 * @q: request queue to fetch a request from
2304 *
2305 * Description:
2306 * Return the request at the top of @q. The request is started on
2307 * return and LLD can start processing it immediately.
2308 *
2309 * Return:
2310 * Pointer to the request at the top of @q if available. Null
2311 * otherwise.
2312 *
2313 * Context:
2314 * queue_lock must be held.
2315 */
2316struct request *blk_fetch_request(struct request_queue *q)
2317{
2318 struct request *rq;
2319
2320 rq = blk_peek_request(q);
2321 if (rq)
2322 blk_start_request(rq);
2323 return rq;
2324}
2325EXPORT_SYMBOL(blk_fetch_request);
2326
2327/**
2328 * blk_update_request - Special helper function for request stacking drivers
2329 * @req: the request being processed
2330 * @error: %0 for success, < %0 for error
2331 * @nr_bytes: number of bytes to complete @req
2332 *
2333 * Description:
2334 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2335 * the request structure even if @req doesn't have leftover.
2336 * If @req has leftover, sets it up for the next range of segments.
2337 *
2338 * This special helper function is only for request stacking drivers
2339 * (e.g. request-based dm) so that they can handle partial completion.
2340 * Actual device drivers should use blk_end_request instead.
2341 *
2342 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2343 * %false return from this function.
2344 *
2345 * Return:
2346 * %false - this request doesn't have any more data
2347 * %true - this request has more data
2348 **/
2349bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2350{
2351 int total_bytes;
2352
2353 if (!req->bio)
2354 return false;
2355
2356 trace_block_rq_complete(req->q, req, nr_bytes);
2357
2358 /*
2359 * For fs requests, rq is just carrier of independent bio's
2360 * and each partial completion should be handled separately.
2361 * Reset per-request error on each partial completion.
2362 *
2363 * TODO: tj: This is too subtle. It would be better to let
2364 * low level drivers do what they see fit.
2365 */
2366 if (req->cmd_type == REQ_TYPE_FS)
2367 req->errors = 0;
2368
2369 if (error && req->cmd_type == REQ_TYPE_FS &&
2370 !(req->cmd_flags & REQ_QUIET)) {
2371 char *error_type;
2372
2373 switch (error) {
2374 case -ENOLINK:
2375 error_type = "recoverable transport";
2376 break;
2377 case -EREMOTEIO:
2378 error_type = "critical target";
2379 break;
2380 case -EBADE:
2381 error_type = "critical nexus";
2382 break;
2383 case -ETIMEDOUT:
2384 error_type = "timeout";
2385 break;
2386 case -ENOSPC:
2387 error_type = "critical space allocation";
2388 break;
2389 case -ENODATA:
2390 error_type = "critical medium";
2391 break;
2392 case -EIO:
2393 default:
2394 error_type = "I/O";
2395 break;
2396 }
2397 printk_ratelimited(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2398 error_type, req->rq_disk ?
2399 req->rq_disk->disk_name : "?",
2400 (unsigned long long)blk_rq_pos(req));
2401
2402 }
2403
2404 blk_account_io_completion(req, nr_bytes);
2405
2406 total_bytes = 0;
2407 while (req->bio) {
2408 struct bio *bio = req->bio;
2409 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
2410
2411 if (bio_bytes == bio->bi_iter.bi_size)
2412 req->bio = bio->bi_next;
2413
2414 req_bio_endio(req, bio, bio_bytes, error);
2415
2416 total_bytes += bio_bytes;
2417 nr_bytes -= bio_bytes;
2418
2419 if (!nr_bytes)
2420 break;
2421 }
2422
2423 /*
2424 * completely done
2425 */
2426 if (!req->bio) {
2427 /*
2428 * Reset counters so that the request stacking driver
2429 * can find how many bytes remain in the request
2430 * later.
2431 */
2432 req->__data_len = 0;
2433 return false;
2434 }
2435
2436 req->__data_len -= total_bytes;
2437 req->buffer = bio_data(req->bio);
2438
2439 /* update sector only for requests with clear definition of sector */
2440 if (req->cmd_type == REQ_TYPE_FS)
2441 req->__sector += total_bytes >> 9;
2442
2443 /* mixed attributes always follow the first bio */
2444 if (req->cmd_flags & REQ_MIXED_MERGE) {
2445 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2446 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2447 }
2448
2449 /*
2450 * If total number of sectors is less than the first segment
2451 * size, something has gone terribly wrong.
2452 */
2453 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2454 blk_dump_rq_flags(req, "request botched");
2455 req->__data_len = blk_rq_cur_bytes(req);
2456 }
2457
2458 /* recalculate the number of segments */
2459 blk_recalc_rq_segments(req);
2460
2461 return true;
2462}
2463EXPORT_SYMBOL_GPL(blk_update_request);
2464
2465static bool blk_update_bidi_request(struct request *rq, int error,
2466 unsigned int nr_bytes,
2467 unsigned int bidi_bytes)
2468{
2469 if (blk_update_request(rq, error, nr_bytes))
2470 return true;
2471
2472 /* Bidi request must be completed as a whole */
2473 if (unlikely(blk_bidi_rq(rq)) &&
2474 blk_update_request(rq->next_rq, error, bidi_bytes))
2475 return true;
2476
2477 if (blk_queue_add_random(rq->q))
2478 add_disk_randomness(rq->rq_disk);
2479
2480 return false;
2481}
2482
2483/**
2484 * blk_unprep_request - unprepare a request
2485 * @req: the request
2486 *
2487 * This function makes a request ready for complete resubmission (or
2488 * completion). It happens only after all error handling is complete,
2489 * so represents the appropriate moment to deallocate any resources
2490 * that were allocated to the request in the prep_rq_fn. The queue
2491 * lock is held when calling this.
2492 */
2493void blk_unprep_request(struct request *req)
2494{
2495 struct request_queue *q = req->q;
2496
2497 req->cmd_flags &= ~REQ_DONTPREP;
2498 if (q->unprep_rq_fn)
2499 q->unprep_rq_fn(q, req);
2500}
2501EXPORT_SYMBOL_GPL(blk_unprep_request);
2502
2503/*
2504 * queue lock must be held
2505 */
2506static void blk_finish_request(struct request *req, int error)
2507{
2508 if (blk_rq_tagged(req))
2509 blk_queue_end_tag(req->q, req);
2510
2511 BUG_ON(blk_queued_rq(req));
2512
2513 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2514 laptop_io_completion(&req->q->backing_dev_info);
2515
2516 blk_delete_timer(req);
2517
2518 if (req->cmd_flags & REQ_DONTPREP)
2519 blk_unprep_request(req);
2520
2521 blk_account_io_done(req);
2522
2523 if (req->end_io)
2524 req->end_io(req, error);
2525 else {
2526 if (blk_bidi_rq(req))
2527 __blk_put_request(req->next_rq->q, req->next_rq);
2528
2529 __blk_put_request(req->q, req);
2530 }
2531}
2532
2533/**
2534 * blk_end_bidi_request - Complete a bidi request
2535 * @rq: the request to complete
2536 * @error: %0 for success, < %0 for error
2537 * @nr_bytes: number of bytes to complete @rq
2538 * @bidi_bytes: number of bytes to complete @rq->next_rq
2539 *
2540 * Description:
2541 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2542 * Drivers that supports bidi can safely call this member for any
2543 * type of request, bidi or uni. In the later case @bidi_bytes is
2544 * just ignored.
2545 *
2546 * Return:
2547 * %false - we are done with this request
2548 * %true - still buffers pending for this request
2549 **/
2550static bool blk_end_bidi_request(struct request *rq, int error,
2551 unsigned int nr_bytes, unsigned int bidi_bytes)
2552{
2553 struct request_queue *q = rq->q;
2554 unsigned long flags;
2555
2556 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2557 return true;
2558
2559 spin_lock_irqsave(q->queue_lock, flags);
2560 blk_finish_request(rq, error);
2561 spin_unlock_irqrestore(q->queue_lock, flags);
2562
2563 return false;
2564}
2565
2566/**
2567 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2568 * @rq: the request to complete
2569 * @error: %0 for success, < %0 for error
2570 * @nr_bytes: number of bytes to complete @rq
2571 * @bidi_bytes: number of bytes to complete @rq->next_rq
2572 *
2573 * Description:
2574 * Identical to blk_end_bidi_request() except that queue lock is
2575 * assumed to be locked on entry and remains so on return.
2576 *
2577 * Return:
2578 * %false - we are done with this request
2579 * %true - still buffers pending for this request
2580 **/
2581bool __blk_end_bidi_request(struct request *rq, int error,
2582 unsigned int nr_bytes, unsigned int bidi_bytes)
2583{
2584 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2585 return true;
2586
2587 blk_finish_request(rq, error);
2588
2589 return false;
2590}
2591
2592/**
2593 * blk_end_request - Helper function for drivers to complete the request.
2594 * @rq: the request being processed
2595 * @error: %0 for success, < %0 for error
2596 * @nr_bytes: number of bytes to complete
2597 *
2598 * Description:
2599 * Ends I/O on a number of bytes attached to @rq.
2600 * If @rq has leftover, sets it up for the next range of segments.
2601 *
2602 * Return:
2603 * %false - we are done with this request
2604 * %true - still buffers pending for this request
2605 **/
2606bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2607{
2608 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2609}
2610EXPORT_SYMBOL(blk_end_request);
2611
2612/**
2613 * blk_end_request_all - Helper function for drives to finish the request.
2614 * @rq: the request to finish
2615 * @error: %0 for success, < %0 for error
2616 *
2617 * Description:
2618 * Completely finish @rq.
2619 */
2620void blk_end_request_all(struct request *rq, int error)
2621{
2622 bool pending;
2623 unsigned int bidi_bytes = 0;
2624
2625 if (unlikely(blk_bidi_rq(rq)))
2626 bidi_bytes = blk_rq_bytes(rq->next_rq);
2627
2628 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2629 BUG_ON(pending);
2630}
2631EXPORT_SYMBOL(blk_end_request_all);
2632
2633/**
2634 * blk_end_request_cur - Helper function to finish the current request chunk.
2635 * @rq: the request to finish the current chunk for
2636 * @error: %0 for success, < %0 for error
2637 *
2638 * Description:
2639 * Complete the current consecutively mapped chunk from @rq.
2640 *
2641 * Return:
2642 * %false - we are done with this request
2643 * %true - still buffers pending for this request
2644 */
2645bool blk_end_request_cur(struct request *rq, int error)
2646{
2647 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2648}
2649EXPORT_SYMBOL(blk_end_request_cur);
2650
2651/**
2652 * blk_end_request_err - Finish a request till the next failure boundary.
2653 * @rq: the request to finish till the next failure boundary for
2654 * @error: must be negative errno
2655 *
2656 * Description:
2657 * Complete @rq till the next failure boundary.
2658 *
2659 * Return:
2660 * %false - we are done with this request
2661 * %true - still buffers pending for this request
2662 */
2663bool blk_end_request_err(struct request *rq, int error)
2664{
2665 WARN_ON(error >= 0);
2666 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2667}
2668EXPORT_SYMBOL_GPL(blk_end_request_err);
2669
2670/**
2671 * __blk_end_request - Helper function for drivers to complete the request.
2672 * @rq: the request being processed
2673 * @error: %0 for success, < %0 for error
2674 * @nr_bytes: number of bytes to complete
2675 *
2676 * Description:
2677 * Must be called with queue lock held unlike blk_end_request().
2678 *
2679 * Return:
2680 * %false - we are done with this request
2681 * %true - still buffers pending for this request
2682 **/
2683bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2684{
2685 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2686}
2687EXPORT_SYMBOL(__blk_end_request);
2688
2689/**
2690 * __blk_end_request_all - Helper function for drives to finish the request.
2691 * @rq: the request to finish
2692 * @error: %0 for success, < %0 for error
2693 *
2694 * Description:
2695 * Completely finish @rq. Must be called with queue lock held.
2696 */
2697void __blk_end_request_all(struct request *rq, int error)
2698{
2699 bool pending;
2700 unsigned int bidi_bytes = 0;
2701
2702 if (unlikely(blk_bidi_rq(rq)))
2703 bidi_bytes = blk_rq_bytes(rq->next_rq);
2704
2705 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2706 BUG_ON(pending);
2707}
2708EXPORT_SYMBOL(__blk_end_request_all);
2709
2710/**
2711 * __blk_end_request_cur - Helper function to finish the current request chunk.
2712 * @rq: the request to finish the current chunk for
2713 * @error: %0 for success, < %0 for error
2714 *
2715 * Description:
2716 * Complete the current consecutively mapped chunk from @rq. Must
2717 * be called with queue lock held.
2718 *
2719 * Return:
2720 * %false - we are done with this request
2721 * %true - still buffers pending for this request
2722 */
2723bool __blk_end_request_cur(struct request *rq, int error)
2724{
2725 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2726}
2727EXPORT_SYMBOL(__blk_end_request_cur);
2728
2729/**
2730 * __blk_end_request_err - Finish a request till the next failure boundary.
2731 * @rq: the request to finish till the next failure boundary for
2732 * @error: must be negative errno
2733 *
2734 * Description:
2735 * Complete @rq till the next failure boundary. Must be called
2736 * with queue lock held.
2737 *
2738 * Return:
2739 * %false - we are done with this request
2740 * %true - still buffers pending for this request
2741 */
2742bool __blk_end_request_err(struct request *rq, int error)
2743{
2744 WARN_ON(error >= 0);
2745 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2746}
2747EXPORT_SYMBOL_GPL(__blk_end_request_err);
2748
2749void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2750 struct bio *bio)
2751{
2752 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2753 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
2754
2755 if (bio_has_data(bio)) {
2756 rq->nr_phys_segments = bio_phys_segments(q, bio);
2757 rq->buffer = bio_data(bio);
2758 }
2759 rq->__data_len = bio->bi_iter.bi_size;
2760 rq->bio = rq->biotail = bio;
2761
2762 if (bio->bi_bdev)
2763 rq->rq_disk = bio->bi_bdev->bd_disk;
2764}
2765
2766#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2767/**
2768 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2769 * @rq: the request to be flushed
2770 *
2771 * Description:
2772 * Flush all pages in @rq.
2773 */
2774void rq_flush_dcache_pages(struct request *rq)
2775{
2776 struct req_iterator iter;
2777 struct bio_vec bvec;
2778
2779 rq_for_each_segment(bvec, rq, iter)
2780 flush_dcache_page(bvec.bv_page);
2781}
2782EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2783#endif
2784
2785/**
2786 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2787 * @q : the queue of the device being checked
2788 *
2789 * Description:
2790 * Check if underlying low-level drivers of a device are busy.
2791 * If the drivers want to export their busy state, they must set own
2792 * exporting function using blk_queue_lld_busy() first.
2793 *
2794 * Basically, this function is used only by request stacking drivers
2795 * to stop dispatching requests to underlying devices when underlying
2796 * devices are busy. This behavior helps more I/O merging on the queue
2797 * of the request stacking driver and prevents I/O throughput regression
2798 * on burst I/O load.
2799 *
2800 * Return:
2801 * 0 - Not busy (The request stacking driver should dispatch request)
2802 * 1 - Busy (The request stacking driver should stop dispatching request)
2803 */
2804int blk_lld_busy(struct request_queue *q)
2805{
2806 if (q->lld_busy_fn)
2807 return q->lld_busy_fn(q);
2808
2809 return 0;
2810}
2811EXPORT_SYMBOL_GPL(blk_lld_busy);
2812
2813/**
2814 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2815 * @rq: the clone request to be cleaned up
2816 *
2817 * Description:
2818 * Free all bios in @rq for a cloned request.
2819 */
2820void blk_rq_unprep_clone(struct request *rq)
2821{
2822 struct bio *bio;
2823
2824 while ((bio = rq->bio) != NULL) {
2825 rq->bio = bio->bi_next;
2826
2827 bio_put(bio);
2828 }
2829}
2830EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2831
2832/*
2833 * Copy attributes of the original request to the clone request.
2834 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2835 */
2836static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2837{
2838 dst->cpu = src->cpu;
2839 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
2840 dst->cmd_type = src->cmd_type;
2841 dst->__sector = blk_rq_pos(src);
2842 dst->__data_len = blk_rq_bytes(src);
2843 dst->nr_phys_segments = src->nr_phys_segments;
2844 dst->ioprio = src->ioprio;
2845 dst->extra_len = src->extra_len;
2846}
2847
2848/**
2849 * blk_rq_prep_clone - Helper function to setup clone request
2850 * @rq: the request to be setup
2851 * @rq_src: original request to be cloned
2852 * @bs: bio_set that bios for clone are allocated from
2853 * @gfp_mask: memory allocation mask for bio
2854 * @bio_ctr: setup function to be called for each clone bio.
2855 * Returns %0 for success, non %0 for failure.
2856 * @data: private data to be passed to @bio_ctr
2857 *
2858 * Description:
2859 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2860 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2861 * are not copied, and copying such parts is the caller's responsibility.
2862 * Also, pages which the original bios are pointing to are not copied
2863 * and the cloned bios just point same pages.
2864 * So cloned bios must be completed before original bios, which means
2865 * the caller must complete @rq before @rq_src.
2866 */
2867int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2868 struct bio_set *bs, gfp_t gfp_mask,
2869 int (*bio_ctr)(struct bio *, struct bio *, void *),
2870 void *data)
2871{
2872 struct bio *bio, *bio_src;
2873
2874 if (!bs)
2875 bs = fs_bio_set;
2876
2877 blk_rq_init(NULL, rq);
2878
2879 __rq_for_each_bio(bio_src, rq_src) {
2880 bio = bio_clone_bioset(bio_src, gfp_mask, bs);
2881 if (!bio)
2882 goto free_and_out;
2883
2884 if (bio_ctr && bio_ctr(bio, bio_src, data))
2885 goto free_and_out;
2886
2887 if (rq->bio) {
2888 rq->biotail->bi_next = bio;
2889 rq->biotail = bio;
2890 } else
2891 rq->bio = rq->biotail = bio;
2892 }
2893
2894 __blk_rq_prep_clone(rq, rq_src);
2895
2896 return 0;
2897
2898free_and_out:
2899 if (bio)
2900 bio_put(bio);
2901 blk_rq_unprep_clone(rq);
2902
2903 return -ENOMEM;
2904}
2905EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2906
2907int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
2908{
2909 return queue_work(kblockd_workqueue, work);
2910}
2911EXPORT_SYMBOL(kblockd_schedule_work);
2912
2913int kblockd_schedule_delayed_work(struct request_queue *q,
2914 struct delayed_work *dwork, unsigned long delay)
2915{
2916 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2917}
2918EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2919
2920#define PLUG_MAGIC 0x91827364
2921
2922/**
2923 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2924 * @plug: The &struct blk_plug that needs to be initialized
2925 *
2926 * Description:
2927 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2928 * pending I/O should the task end up blocking between blk_start_plug() and
2929 * blk_finish_plug(). This is important from a performance perspective, but
2930 * also ensures that we don't deadlock. For instance, if the task is blocking
2931 * for a memory allocation, memory reclaim could end up wanting to free a
2932 * page belonging to that request that is currently residing in our private
2933 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2934 * this kind of deadlock.
2935 */
2936void blk_start_plug(struct blk_plug *plug)
2937{
2938 struct task_struct *tsk = current;
2939
2940 plug->magic = PLUG_MAGIC;
2941 INIT_LIST_HEAD(&plug->list);
2942 INIT_LIST_HEAD(&plug->mq_list);
2943 INIT_LIST_HEAD(&plug->cb_list);
2944
2945 /*
2946 * If this is a nested plug, don't actually assign it. It will be
2947 * flushed on its own.
2948 */
2949 if (!tsk->plug) {
2950 /*
2951 * Store ordering should not be needed here, since a potential
2952 * preempt will imply a full memory barrier
2953 */
2954 tsk->plug = plug;
2955 }
2956}
2957EXPORT_SYMBOL(blk_start_plug);
2958
2959static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2960{
2961 struct request *rqa = container_of(a, struct request, queuelist);
2962 struct request *rqb = container_of(b, struct request, queuelist);
2963
2964 return !(rqa->q < rqb->q ||
2965 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
2966}
2967
2968/*
2969 * If 'from_schedule' is true, then postpone the dispatch of requests
2970 * until a safe kblockd context. We due this to avoid accidental big
2971 * additional stack usage in driver dispatch, in places where the originally
2972 * plugger did not intend it.
2973 */
2974static void queue_unplugged(struct request_queue *q, unsigned int depth,
2975 bool from_schedule)
2976 __releases(q->queue_lock)
2977{
2978 trace_block_unplug(q, depth, !from_schedule);
2979
2980 if (from_schedule)
2981 blk_run_queue_async(q);
2982 else
2983 __blk_run_queue(q);
2984 spin_unlock(q->queue_lock);
2985}
2986
2987static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
2988{
2989 LIST_HEAD(callbacks);
2990
2991 while (!list_empty(&plug->cb_list)) {
2992 list_splice_init(&plug->cb_list, &callbacks);
2993
2994 while (!list_empty(&callbacks)) {
2995 struct blk_plug_cb *cb = list_first_entry(&callbacks,
2996 struct blk_plug_cb,
2997 list);
2998 list_del(&cb->list);
2999 cb->callback(cb, from_schedule);
3000 }
3001 }
3002}
3003
3004struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3005 int size)
3006{
3007 struct blk_plug *plug = current->plug;
3008 struct blk_plug_cb *cb;
3009
3010 if (!plug)
3011 return NULL;
3012
3013 list_for_each_entry(cb, &plug->cb_list, list)
3014 if (cb->callback == unplug && cb->data == data)
3015 return cb;
3016
3017 /* Not currently on the callback list */
3018 BUG_ON(size < sizeof(*cb));
3019 cb = kzalloc(size, GFP_ATOMIC);
3020 if (cb) {
3021 cb->data = data;
3022 cb->callback = unplug;
3023 list_add(&cb->list, &plug->cb_list);
3024 }
3025 return cb;
3026}
3027EXPORT_SYMBOL(blk_check_plugged);
3028
3029void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
3030{
3031 struct request_queue *q;
3032 unsigned long flags;
3033 struct request *rq;
3034 LIST_HEAD(list);
3035 unsigned int depth;
3036
3037 BUG_ON(plug->magic != PLUG_MAGIC);
3038
3039 flush_plug_callbacks(plug, from_schedule);
3040
3041 if (!list_empty(&plug->mq_list))
3042 blk_mq_flush_plug_list(plug, from_schedule);
3043
3044 if (list_empty(&plug->list))
3045 return;
3046
3047 list_splice_init(&plug->list, &list);
3048
3049 list_sort(NULL, &list, plug_rq_cmp);
3050
3051 q = NULL;
3052 depth = 0;
3053
3054 /*
3055 * Save and disable interrupts here, to avoid doing it for every
3056 * queue lock we have to take.
3057 */
3058 local_irq_save(flags);
3059 while (!list_empty(&list)) {
3060 rq = list_entry_rq(list.next);
3061 list_del_init(&rq->queuelist);
3062 BUG_ON(!rq->q);
3063 if (rq->q != q) {
3064 /*
3065 * This drops the queue lock
3066 */
3067 if (q)
3068 queue_unplugged(q, depth, from_schedule);
3069 q = rq->q;
3070 depth = 0;
3071 spin_lock(q->queue_lock);
3072 }
3073
3074 /*
3075 * Short-circuit if @q is dead
3076 */
3077 if (unlikely(blk_queue_dying(q))) {
3078 __blk_end_request_all(rq, -ENODEV);
3079 continue;
3080 }
3081
3082 /*
3083 * rq is already accounted, so use raw insert
3084 */
3085 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
3086 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3087 else
3088 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
3089
3090 depth++;
3091 }
3092
3093 /*
3094 * This drops the queue lock
3095 */
3096 if (q)
3097 queue_unplugged(q, depth, from_schedule);
3098
3099 local_irq_restore(flags);
3100}
3101
3102void blk_finish_plug(struct blk_plug *plug)
3103{
3104 blk_flush_plug_list(plug, false);
3105
3106 if (plug == current->plug)
3107 current->plug = NULL;
3108}
3109EXPORT_SYMBOL(blk_finish_plug);
3110
3111#ifdef CONFIG_PM_RUNTIME
3112/**
3113 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3114 * @q: the queue of the device
3115 * @dev: the device the queue belongs to
3116 *
3117 * Description:
3118 * Initialize runtime-PM-related fields for @q and start auto suspend for
3119 * @dev. Drivers that want to take advantage of request-based runtime PM
3120 * should call this function after @dev has been initialized, and its
3121 * request queue @q has been allocated, and runtime PM for it can not happen
3122 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3123 * cases, driver should call this function before any I/O has taken place.
3124 *
3125 * This function takes care of setting up using auto suspend for the device,
3126 * the autosuspend delay is set to -1 to make runtime suspend impossible
3127 * until an updated value is either set by user or by driver. Drivers do
3128 * not need to touch other autosuspend settings.
3129 *
3130 * The block layer runtime PM is request based, so only works for drivers
3131 * that use request as their IO unit instead of those directly use bio's.
3132 */
3133void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3134{
3135 q->dev = dev;
3136 q->rpm_status = RPM_ACTIVE;
3137 pm_runtime_set_autosuspend_delay(q->dev, -1);
3138 pm_runtime_use_autosuspend(q->dev);
3139}
3140EXPORT_SYMBOL(blk_pm_runtime_init);
3141
3142/**
3143 * blk_pre_runtime_suspend - Pre runtime suspend check
3144 * @q: the queue of the device
3145 *
3146 * Description:
3147 * This function will check if runtime suspend is allowed for the device
3148 * by examining if there are any requests pending in the queue. If there
3149 * are requests pending, the device can not be runtime suspended; otherwise,
3150 * the queue's status will be updated to SUSPENDING and the driver can
3151 * proceed to suspend the device.
3152 *
3153 * For the not allowed case, we mark last busy for the device so that
3154 * runtime PM core will try to autosuspend it some time later.
3155 *
3156 * This function should be called near the start of the device's
3157 * runtime_suspend callback.
3158 *
3159 * Return:
3160 * 0 - OK to runtime suspend the device
3161 * -EBUSY - Device should not be runtime suspended
3162 */
3163int blk_pre_runtime_suspend(struct request_queue *q)
3164{
3165 int ret = 0;
3166
3167 spin_lock_irq(q->queue_lock);
3168 if (q->nr_pending) {
3169 ret = -EBUSY;
3170 pm_runtime_mark_last_busy(q->dev);
3171 } else {
3172 q->rpm_status = RPM_SUSPENDING;
3173 }
3174 spin_unlock_irq(q->queue_lock);
3175 return ret;
3176}
3177EXPORT_SYMBOL(blk_pre_runtime_suspend);
3178
3179/**
3180 * blk_post_runtime_suspend - Post runtime suspend processing
3181 * @q: the queue of the device
3182 * @err: return value of the device's runtime_suspend function
3183 *
3184 * Description:
3185 * Update the queue's runtime status according to the return value of the
3186 * device's runtime suspend function and mark last busy for the device so
3187 * that PM core will try to auto suspend the device at a later time.
3188 *
3189 * This function should be called near the end of the device's
3190 * runtime_suspend callback.
3191 */
3192void blk_post_runtime_suspend(struct request_queue *q, int err)
3193{
3194 spin_lock_irq(q->queue_lock);
3195 if (!err) {
3196 q->rpm_status = RPM_SUSPENDED;
3197 } else {
3198 q->rpm_status = RPM_ACTIVE;
3199 pm_runtime_mark_last_busy(q->dev);
3200 }
3201 spin_unlock_irq(q->queue_lock);
3202}
3203EXPORT_SYMBOL(blk_post_runtime_suspend);
3204
3205/**
3206 * blk_pre_runtime_resume - Pre runtime resume processing
3207 * @q: the queue of the device
3208 *
3209 * Description:
3210 * Update the queue's runtime status to RESUMING in preparation for the
3211 * runtime resume of the device.
3212 *
3213 * This function should be called near the start of the device's
3214 * runtime_resume callback.
3215 */
3216void blk_pre_runtime_resume(struct request_queue *q)
3217{
3218 spin_lock_irq(q->queue_lock);
3219 q->rpm_status = RPM_RESUMING;
3220 spin_unlock_irq(q->queue_lock);
3221}
3222EXPORT_SYMBOL(blk_pre_runtime_resume);
3223
3224/**
3225 * blk_post_runtime_resume - Post runtime resume processing
3226 * @q: the queue of the device
3227 * @err: return value of the device's runtime_resume function
3228 *
3229 * Description:
3230 * Update the queue's runtime status according to the return value of the
3231 * device's runtime_resume function. If it is successfully resumed, process
3232 * the requests that are queued into the device's queue when it is resuming
3233 * and then mark last busy and initiate autosuspend for it.
3234 *
3235 * This function should be called near the end of the device's
3236 * runtime_resume callback.
3237 */
3238void blk_post_runtime_resume(struct request_queue *q, int err)
3239{
3240 spin_lock_irq(q->queue_lock);
3241 if (!err) {
3242 q->rpm_status = RPM_ACTIVE;
3243 __blk_run_queue(q);
3244 pm_runtime_mark_last_busy(q->dev);
3245 pm_request_autosuspend(q->dev);
3246 } else {
3247 q->rpm_status = RPM_SUSPENDED;
3248 }
3249 spin_unlock_irq(q->queue_lock);
3250}
3251EXPORT_SYMBOL(blk_post_runtime_resume);
3252#endif
3253
3254int __init blk_dev_init(void)
3255{
3256 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
3257 sizeof(((struct request *)0)->cmd_flags));
3258
3259 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3260 kblockd_workqueue = alloc_workqueue("kblockd",
3261 WQ_MEM_RECLAIM | WQ_HIGHPRI |
3262 WQ_POWER_EFFICIENT, 0);
3263 if (!kblockd_workqueue)
3264 panic("Failed to create kblockd\n");
3265
3266 request_cachep = kmem_cache_create("blkdev_requests",
3267 sizeof(struct request), 0, SLAB_PANIC, NULL);
3268
3269 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
3270 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
3271
3272 return 0;
3273}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) 1991, 1992 Linus Torvalds
4 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
5 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
6 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
7 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
8 * - July2000
9 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
10 */
11
12/*
13 * This handles all read/write requests to block devices
14 */
15#include <linux/kernel.h>
16#include <linux/module.h>
17#include <linux/bio.h>
18#include <linux/blkdev.h>
19#include <linux/blk-pm.h>
20#include <linux/blk-integrity.h>
21#include <linux/highmem.h>
22#include <linux/mm.h>
23#include <linux/pagemap.h>
24#include <linux/kernel_stat.h>
25#include <linux/string.h>
26#include <linux/init.h>
27#include <linux/completion.h>
28#include <linux/slab.h>
29#include <linux/swap.h>
30#include <linux/writeback.h>
31#include <linux/task_io_accounting_ops.h>
32#include <linux/fault-inject.h>
33#include <linux/list_sort.h>
34#include <linux/delay.h>
35#include <linux/ratelimit.h>
36#include <linux/pm_runtime.h>
37#include <linux/t10-pi.h>
38#include <linux/debugfs.h>
39#include <linux/bpf.h>
40#include <linux/part_stat.h>
41#include <linux/sched/sysctl.h>
42#include <linux/blk-crypto.h>
43
44#define CREATE_TRACE_POINTS
45#include <trace/events/block.h>
46
47#include "blk.h"
48#include "blk-mq-sched.h"
49#include "blk-pm.h"
50#include "blk-cgroup.h"
51#include "blk-throttle.h"
52#include "blk-ioprio.h"
53
54struct dentry *blk_debugfs_root;
55
56EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
57EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
58EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
59EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
60EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
61EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_insert);
62
63static DEFINE_IDA(blk_queue_ida);
64
65/*
66 * For queue allocation
67 */
68static struct kmem_cache *blk_requestq_cachep;
69
70/*
71 * Controlling structure to kblockd
72 */
73static struct workqueue_struct *kblockd_workqueue;
74
75/**
76 * blk_queue_flag_set - atomically set a queue flag
77 * @flag: flag to be set
78 * @q: request queue
79 */
80void blk_queue_flag_set(unsigned int flag, struct request_queue *q)
81{
82 set_bit(flag, &q->queue_flags);
83}
84EXPORT_SYMBOL(blk_queue_flag_set);
85
86/**
87 * blk_queue_flag_clear - atomically clear a queue flag
88 * @flag: flag to be cleared
89 * @q: request queue
90 */
91void blk_queue_flag_clear(unsigned int flag, struct request_queue *q)
92{
93 clear_bit(flag, &q->queue_flags);
94}
95EXPORT_SYMBOL(blk_queue_flag_clear);
96
97#define REQ_OP_NAME(name) [REQ_OP_##name] = #name
98static const char *const blk_op_name[] = {
99 REQ_OP_NAME(READ),
100 REQ_OP_NAME(WRITE),
101 REQ_OP_NAME(FLUSH),
102 REQ_OP_NAME(DISCARD),
103 REQ_OP_NAME(SECURE_ERASE),
104 REQ_OP_NAME(ZONE_RESET),
105 REQ_OP_NAME(ZONE_RESET_ALL),
106 REQ_OP_NAME(ZONE_OPEN),
107 REQ_OP_NAME(ZONE_CLOSE),
108 REQ_OP_NAME(ZONE_FINISH),
109 REQ_OP_NAME(ZONE_APPEND),
110 REQ_OP_NAME(WRITE_ZEROES),
111 REQ_OP_NAME(DRV_IN),
112 REQ_OP_NAME(DRV_OUT),
113};
114#undef REQ_OP_NAME
115
116/**
117 * blk_op_str - Return string XXX in the REQ_OP_XXX.
118 * @op: REQ_OP_XXX.
119 *
120 * Description: Centralize block layer function to convert REQ_OP_XXX into
121 * string format. Useful in the debugging and tracing bio or request. For
122 * invalid REQ_OP_XXX it returns string "UNKNOWN".
123 */
124inline const char *blk_op_str(enum req_op op)
125{
126 const char *op_str = "UNKNOWN";
127
128 if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op])
129 op_str = blk_op_name[op];
130
131 return op_str;
132}
133EXPORT_SYMBOL_GPL(blk_op_str);
134
135static const struct {
136 int errno;
137 const char *name;
138} blk_errors[] = {
139 [BLK_STS_OK] = { 0, "" },
140 [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" },
141 [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" },
142 [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" },
143 [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" },
144 [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" },
145 [BLK_STS_RESV_CONFLICT] = { -EBADE, "reservation conflict" },
146 [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" },
147 [BLK_STS_PROTECTION] = { -EILSEQ, "protection" },
148 [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" },
149 [BLK_STS_DEV_RESOURCE] = { -EBUSY, "device resource" },
150 [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" },
151 [BLK_STS_OFFLINE] = { -ENODEV, "device offline" },
152
153 /* device mapper special case, should not leak out: */
154 [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" },
155
156 /* zone device specific errors */
157 [BLK_STS_ZONE_OPEN_RESOURCE] = { -ETOOMANYREFS, "open zones exceeded" },
158 [BLK_STS_ZONE_ACTIVE_RESOURCE] = { -EOVERFLOW, "active zones exceeded" },
159
160 /* Command duration limit device-side timeout */
161 [BLK_STS_DURATION_LIMIT] = { -ETIME, "duration limit exceeded" },
162
163 [BLK_STS_INVAL] = { -EINVAL, "invalid" },
164
165 /* everything else not covered above: */
166 [BLK_STS_IOERR] = { -EIO, "I/O" },
167};
168
169blk_status_t errno_to_blk_status(int errno)
170{
171 int i;
172
173 for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
174 if (blk_errors[i].errno == errno)
175 return (__force blk_status_t)i;
176 }
177
178 return BLK_STS_IOERR;
179}
180EXPORT_SYMBOL_GPL(errno_to_blk_status);
181
182int blk_status_to_errno(blk_status_t status)
183{
184 int idx = (__force int)status;
185
186 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
187 return -EIO;
188 return blk_errors[idx].errno;
189}
190EXPORT_SYMBOL_GPL(blk_status_to_errno);
191
192const char *blk_status_to_str(blk_status_t status)
193{
194 int idx = (__force int)status;
195
196 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
197 return "<null>";
198 return blk_errors[idx].name;
199}
200EXPORT_SYMBOL_GPL(blk_status_to_str);
201
202/**
203 * blk_sync_queue - cancel any pending callbacks on a queue
204 * @q: the queue
205 *
206 * Description:
207 * The block layer may perform asynchronous callback activity
208 * on a queue, such as calling the unplug function after a timeout.
209 * A block device may call blk_sync_queue to ensure that any
210 * such activity is cancelled, thus allowing it to release resources
211 * that the callbacks might use. The caller must already have made sure
212 * that its ->submit_bio will not re-add plugging prior to calling
213 * this function.
214 *
215 * This function does not cancel any asynchronous activity arising
216 * out of elevator or throttling code. That would require elevator_exit()
217 * and blkcg_exit_queue() to be called with queue lock initialized.
218 *
219 */
220void blk_sync_queue(struct request_queue *q)
221{
222 del_timer_sync(&q->timeout);
223 cancel_work_sync(&q->timeout_work);
224}
225EXPORT_SYMBOL(blk_sync_queue);
226
227/**
228 * blk_set_pm_only - increment pm_only counter
229 * @q: request queue pointer
230 */
231void blk_set_pm_only(struct request_queue *q)
232{
233 atomic_inc(&q->pm_only);
234}
235EXPORT_SYMBOL_GPL(blk_set_pm_only);
236
237void blk_clear_pm_only(struct request_queue *q)
238{
239 int pm_only;
240
241 pm_only = atomic_dec_return(&q->pm_only);
242 WARN_ON_ONCE(pm_only < 0);
243 if (pm_only == 0)
244 wake_up_all(&q->mq_freeze_wq);
245}
246EXPORT_SYMBOL_GPL(blk_clear_pm_only);
247
248static void blk_free_queue_rcu(struct rcu_head *rcu_head)
249{
250 struct request_queue *q = container_of(rcu_head,
251 struct request_queue, rcu_head);
252
253 percpu_ref_exit(&q->q_usage_counter);
254 kmem_cache_free(blk_requestq_cachep, q);
255}
256
257static void blk_free_queue(struct request_queue *q)
258{
259 blk_free_queue_stats(q->stats);
260 if (queue_is_mq(q))
261 blk_mq_release(q);
262
263 ida_free(&blk_queue_ida, q->id);
264 lockdep_unregister_key(&q->io_lock_cls_key);
265 lockdep_unregister_key(&q->q_lock_cls_key);
266 call_rcu(&q->rcu_head, blk_free_queue_rcu);
267}
268
269/**
270 * blk_put_queue - decrement the request_queue refcount
271 * @q: the request_queue structure to decrement the refcount for
272 *
273 * Decrements the refcount of the request_queue and free it when the refcount
274 * reaches 0.
275 */
276void blk_put_queue(struct request_queue *q)
277{
278 if (refcount_dec_and_test(&q->refs))
279 blk_free_queue(q);
280}
281EXPORT_SYMBOL(blk_put_queue);
282
283bool blk_queue_start_drain(struct request_queue *q)
284{
285 /*
286 * When queue DYING flag is set, we need to block new req
287 * entering queue, so we call blk_freeze_queue_start() to
288 * prevent I/O from crossing blk_queue_enter().
289 */
290 bool freeze = __blk_freeze_queue_start(q, current);
291 if (queue_is_mq(q))
292 blk_mq_wake_waiters(q);
293 /* Make blk_queue_enter() reexamine the DYING flag. */
294 wake_up_all(&q->mq_freeze_wq);
295
296 return freeze;
297}
298
299/**
300 * blk_queue_enter() - try to increase q->q_usage_counter
301 * @q: request queue pointer
302 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PM
303 */
304int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
305{
306 const bool pm = flags & BLK_MQ_REQ_PM;
307
308 while (!blk_try_enter_queue(q, pm)) {
309 if (flags & BLK_MQ_REQ_NOWAIT)
310 return -EAGAIN;
311
312 /*
313 * read pair of barrier in blk_freeze_queue_start(), we need to
314 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
315 * reading .mq_freeze_depth or queue dying flag, otherwise the
316 * following wait may never return if the two reads are
317 * reordered.
318 */
319 smp_rmb();
320 wait_event(q->mq_freeze_wq,
321 (!q->mq_freeze_depth &&
322 blk_pm_resume_queue(pm, q)) ||
323 blk_queue_dying(q));
324 if (blk_queue_dying(q))
325 return -ENODEV;
326 }
327
328 rwsem_acquire_read(&q->q_lockdep_map, 0, 0, _RET_IP_);
329 rwsem_release(&q->q_lockdep_map, _RET_IP_);
330 return 0;
331}
332
333int __bio_queue_enter(struct request_queue *q, struct bio *bio)
334{
335 while (!blk_try_enter_queue(q, false)) {
336 struct gendisk *disk = bio->bi_bdev->bd_disk;
337
338 if (bio->bi_opf & REQ_NOWAIT) {
339 if (test_bit(GD_DEAD, &disk->state))
340 goto dead;
341 bio_wouldblock_error(bio);
342 return -EAGAIN;
343 }
344
345 /*
346 * read pair of barrier in blk_freeze_queue_start(), we need to
347 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
348 * reading .mq_freeze_depth or queue dying flag, otherwise the
349 * following wait may never return if the two reads are
350 * reordered.
351 */
352 smp_rmb();
353 wait_event(q->mq_freeze_wq,
354 (!q->mq_freeze_depth &&
355 blk_pm_resume_queue(false, q)) ||
356 test_bit(GD_DEAD, &disk->state));
357 if (test_bit(GD_DEAD, &disk->state))
358 goto dead;
359 }
360
361 rwsem_acquire_read(&q->io_lockdep_map, 0, 0, _RET_IP_);
362 rwsem_release(&q->io_lockdep_map, _RET_IP_);
363 return 0;
364dead:
365 bio_io_error(bio);
366 return -ENODEV;
367}
368
369void blk_queue_exit(struct request_queue *q)
370{
371 percpu_ref_put(&q->q_usage_counter);
372}
373
374static void blk_queue_usage_counter_release(struct percpu_ref *ref)
375{
376 struct request_queue *q =
377 container_of(ref, struct request_queue, q_usage_counter);
378
379 wake_up_all(&q->mq_freeze_wq);
380}
381
382static void blk_rq_timed_out_timer(struct timer_list *t)
383{
384 struct request_queue *q = from_timer(q, t, timeout);
385
386 kblockd_schedule_work(&q->timeout_work);
387}
388
389static void blk_timeout_work(struct work_struct *work)
390{
391}
392
393struct request_queue *blk_alloc_queue(struct queue_limits *lim, int node_id)
394{
395 struct request_queue *q;
396 int error;
397
398 q = kmem_cache_alloc_node(blk_requestq_cachep, GFP_KERNEL | __GFP_ZERO,
399 node_id);
400 if (!q)
401 return ERR_PTR(-ENOMEM);
402
403 q->last_merge = NULL;
404
405 q->id = ida_alloc(&blk_queue_ida, GFP_KERNEL);
406 if (q->id < 0) {
407 error = q->id;
408 goto fail_q;
409 }
410
411 q->stats = blk_alloc_queue_stats();
412 if (!q->stats) {
413 error = -ENOMEM;
414 goto fail_id;
415 }
416
417 error = blk_set_default_limits(lim);
418 if (error)
419 goto fail_stats;
420 q->limits = *lim;
421
422 q->node = node_id;
423
424 atomic_set(&q->nr_active_requests_shared_tags, 0);
425
426 timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
427 INIT_WORK(&q->timeout_work, blk_timeout_work);
428 INIT_LIST_HEAD(&q->icq_list);
429
430 refcount_set(&q->refs, 1);
431 mutex_init(&q->debugfs_mutex);
432 mutex_init(&q->sysfs_lock);
433 mutex_init(&q->sysfs_dir_lock);
434 mutex_init(&q->limits_lock);
435 mutex_init(&q->rq_qos_mutex);
436 spin_lock_init(&q->queue_lock);
437
438 init_waitqueue_head(&q->mq_freeze_wq);
439 mutex_init(&q->mq_freeze_lock);
440
441 blkg_init_queue(q);
442
443 /*
444 * Init percpu_ref in atomic mode so that it's faster to shutdown.
445 * See blk_register_queue() for details.
446 */
447 error = percpu_ref_init(&q->q_usage_counter,
448 blk_queue_usage_counter_release,
449 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL);
450 if (error)
451 goto fail_stats;
452 lockdep_register_key(&q->io_lock_cls_key);
453 lockdep_register_key(&q->q_lock_cls_key);
454 lockdep_init_map(&q->io_lockdep_map, "&q->q_usage_counter(io)",
455 &q->io_lock_cls_key, 0);
456 lockdep_init_map(&q->q_lockdep_map, "&q->q_usage_counter(queue)",
457 &q->q_lock_cls_key, 0);
458
459 q->nr_requests = BLKDEV_DEFAULT_RQ;
460
461 return q;
462
463fail_stats:
464 blk_free_queue_stats(q->stats);
465fail_id:
466 ida_free(&blk_queue_ida, q->id);
467fail_q:
468 kmem_cache_free(blk_requestq_cachep, q);
469 return ERR_PTR(error);
470}
471
472/**
473 * blk_get_queue - increment the request_queue refcount
474 * @q: the request_queue structure to increment the refcount for
475 *
476 * Increment the refcount of the request_queue kobject.
477 *
478 * Context: Any context.
479 */
480bool blk_get_queue(struct request_queue *q)
481{
482 if (unlikely(blk_queue_dying(q)))
483 return false;
484 refcount_inc(&q->refs);
485 return true;
486}
487EXPORT_SYMBOL(blk_get_queue);
488
489#ifdef CONFIG_FAIL_MAKE_REQUEST
490
491static DECLARE_FAULT_ATTR(fail_make_request);
492
493static int __init setup_fail_make_request(char *str)
494{
495 return setup_fault_attr(&fail_make_request, str);
496}
497__setup("fail_make_request=", setup_fail_make_request);
498
499bool should_fail_request(struct block_device *part, unsigned int bytes)
500{
501 return bdev_test_flag(part, BD_MAKE_IT_FAIL) &&
502 should_fail(&fail_make_request, bytes);
503}
504
505static int __init fail_make_request_debugfs(void)
506{
507 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
508 NULL, &fail_make_request);
509
510 return PTR_ERR_OR_ZERO(dir);
511}
512
513late_initcall(fail_make_request_debugfs);
514#endif /* CONFIG_FAIL_MAKE_REQUEST */
515
516static inline void bio_check_ro(struct bio *bio)
517{
518 if (op_is_write(bio_op(bio)) && bdev_read_only(bio->bi_bdev)) {
519 if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
520 return;
521
522 if (bdev_test_flag(bio->bi_bdev, BD_RO_WARNED))
523 return;
524
525 bdev_set_flag(bio->bi_bdev, BD_RO_WARNED);
526
527 /*
528 * Use ioctl to set underlying disk of raid/dm to read-only
529 * will trigger this.
530 */
531 pr_warn("Trying to write to read-only block-device %pg\n",
532 bio->bi_bdev);
533 }
534}
535
536static noinline int should_fail_bio(struct bio *bio)
537{
538 if (should_fail_request(bdev_whole(bio->bi_bdev), bio->bi_iter.bi_size))
539 return -EIO;
540 return 0;
541}
542ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
543
544/*
545 * Check whether this bio extends beyond the end of the device or partition.
546 * This may well happen - the kernel calls bread() without checking the size of
547 * the device, e.g., when mounting a file system.
548 */
549static inline int bio_check_eod(struct bio *bio)
550{
551 sector_t maxsector = bdev_nr_sectors(bio->bi_bdev);
552 unsigned int nr_sectors = bio_sectors(bio);
553
554 if (nr_sectors &&
555 (nr_sectors > maxsector ||
556 bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
557 pr_info_ratelimited("%s: attempt to access beyond end of device\n"
558 "%pg: rw=%d, sector=%llu, nr_sectors = %u limit=%llu\n",
559 current->comm, bio->bi_bdev, bio->bi_opf,
560 bio->bi_iter.bi_sector, nr_sectors, maxsector);
561 return -EIO;
562 }
563 return 0;
564}
565
566/*
567 * Remap block n of partition p to block n+start(p) of the disk.
568 */
569static int blk_partition_remap(struct bio *bio)
570{
571 struct block_device *p = bio->bi_bdev;
572
573 if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
574 return -EIO;
575 if (bio_sectors(bio)) {
576 bio->bi_iter.bi_sector += p->bd_start_sect;
577 trace_block_bio_remap(bio, p->bd_dev,
578 bio->bi_iter.bi_sector -
579 p->bd_start_sect);
580 }
581 bio_set_flag(bio, BIO_REMAPPED);
582 return 0;
583}
584
585/*
586 * Check write append to a zoned block device.
587 */
588static inline blk_status_t blk_check_zone_append(struct request_queue *q,
589 struct bio *bio)
590{
591 int nr_sectors = bio_sectors(bio);
592
593 /* Only applicable to zoned block devices */
594 if (!bdev_is_zoned(bio->bi_bdev))
595 return BLK_STS_NOTSUPP;
596
597 /* The bio sector must point to the start of a sequential zone */
598 if (!bdev_is_zone_start(bio->bi_bdev, bio->bi_iter.bi_sector))
599 return BLK_STS_IOERR;
600
601 /*
602 * Not allowed to cross zone boundaries. Otherwise, the BIO will be
603 * split and could result in non-contiguous sectors being written in
604 * different zones.
605 */
606 if (nr_sectors > q->limits.chunk_sectors)
607 return BLK_STS_IOERR;
608
609 /* Make sure the BIO is small enough and will not get split */
610 if (nr_sectors > q->limits.max_zone_append_sectors)
611 return BLK_STS_IOERR;
612
613 bio->bi_opf |= REQ_NOMERGE;
614
615 return BLK_STS_OK;
616}
617
618static void __submit_bio(struct bio *bio)
619{
620 /* If plug is not used, add new plug here to cache nsecs time. */
621 struct blk_plug plug;
622
623 if (unlikely(!blk_crypto_bio_prep(&bio)))
624 return;
625
626 blk_start_plug(&plug);
627
628 if (!bdev_test_flag(bio->bi_bdev, BD_HAS_SUBMIT_BIO)) {
629 blk_mq_submit_bio(bio);
630 } else if (likely(bio_queue_enter(bio) == 0)) {
631 struct gendisk *disk = bio->bi_bdev->bd_disk;
632
633 if ((bio->bi_opf & REQ_POLLED) &&
634 !(disk->queue->limits.features & BLK_FEAT_POLL)) {
635 bio->bi_status = BLK_STS_NOTSUPP;
636 bio_endio(bio);
637 } else {
638 disk->fops->submit_bio(bio);
639 }
640 blk_queue_exit(disk->queue);
641 }
642
643 blk_finish_plug(&plug);
644}
645
646/*
647 * The loop in this function may be a bit non-obvious, and so deserves some
648 * explanation:
649 *
650 * - Before entering the loop, bio->bi_next is NULL (as all callers ensure
651 * that), so we have a list with a single bio.
652 * - We pretend that we have just taken it off a longer list, so we assign
653 * bio_list to a pointer to the bio_list_on_stack, thus initialising the
654 * bio_list of new bios to be added. ->submit_bio() may indeed add some more
655 * bios through a recursive call to submit_bio_noacct. If it did, we find a
656 * non-NULL value in bio_list and re-enter the loop from the top.
657 * - In this case we really did just take the bio of the top of the list (no
658 * pretending) and so remove it from bio_list, and call into ->submit_bio()
659 * again.
660 *
661 * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio.
662 * bio_list_on_stack[1] contains bios that were submitted before the current
663 * ->submit_bio, but that haven't been processed yet.
664 */
665static void __submit_bio_noacct(struct bio *bio)
666{
667 struct bio_list bio_list_on_stack[2];
668
669 BUG_ON(bio->bi_next);
670
671 bio_list_init(&bio_list_on_stack[0]);
672 current->bio_list = bio_list_on_stack;
673
674 do {
675 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
676 struct bio_list lower, same;
677
678 /*
679 * Create a fresh bio_list for all subordinate requests.
680 */
681 bio_list_on_stack[1] = bio_list_on_stack[0];
682 bio_list_init(&bio_list_on_stack[0]);
683
684 __submit_bio(bio);
685
686 /*
687 * Sort new bios into those for a lower level and those for the
688 * same level.
689 */
690 bio_list_init(&lower);
691 bio_list_init(&same);
692 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
693 if (q == bdev_get_queue(bio->bi_bdev))
694 bio_list_add(&same, bio);
695 else
696 bio_list_add(&lower, bio);
697
698 /*
699 * Now assemble so we handle the lowest level first.
700 */
701 bio_list_merge(&bio_list_on_stack[0], &lower);
702 bio_list_merge(&bio_list_on_stack[0], &same);
703 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
704 } while ((bio = bio_list_pop(&bio_list_on_stack[0])));
705
706 current->bio_list = NULL;
707}
708
709static void __submit_bio_noacct_mq(struct bio *bio)
710{
711 struct bio_list bio_list[2] = { };
712
713 current->bio_list = bio_list;
714
715 do {
716 __submit_bio(bio);
717 } while ((bio = bio_list_pop(&bio_list[0])));
718
719 current->bio_list = NULL;
720}
721
722void submit_bio_noacct_nocheck(struct bio *bio)
723{
724 blk_cgroup_bio_start(bio);
725 blkcg_bio_issue_init(bio);
726
727 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
728 trace_block_bio_queue(bio);
729 /*
730 * Now that enqueuing has been traced, we need to trace
731 * completion as well.
732 */
733 bio_set_flag(bio, BIO_TRACE_COMPLETION);
734 }
735
736 /*
737 * We only want one ->submit_bio to be active at a time, else stack
738 * usage with stacked devices could be a problem. Use current->bio_list
739 * to collect a list of requests submited by a ->submit_bio method while
740 * it is active, and then process them after it returned.
741 */
742 if (current->bio_list)
743 bio_list_add(¤t->bio_list[0], bio);
744 else if (!bdev_test_flag(bio->bi_bdev, BD_HAS_SUBMIT_BIO))
745 __submit_bio_noacct_mq(bio);
746 else
747 __submit_bio_noacct(bio);
748}
749
750static blk_status_t blk_validate_atomic_write_op_size(struct request_queue *q,
751 struct bio *bio)
752{
753 if (bio->bi_iter.bi_size > queue_atomic_write_unit_max_bytes(q))
754 return BLK_STS_INVAL;
755
756 if (bio->bi_iter.bi_size % queue_atomic_write_unit_min_bytes(q))
757 return BLK_STS_INVAL;
758
759 return BLK_STS_OK;
760}
761
762/**
763 * submit_bio_noacct - re-submit a bio to the block device layer for I/O
764 * @bio: The bio describing the location in memory and on the device.
765 *
766 * This is a version of submit_bio() that shall only be used for I/O that is
767 * resubmitted to lower level drivers by stacking block drivers. All file
768 * systems and other upper level users of the block layer should use
769 * submit_bio() instead.
770 */
771void submit_bio_noacct(struct bio *bio)
772{
773 struct block_device *bdev = bio->bi_bdev;
774 struct request_queue *q = bdev_get_queue(bdev);
775 blk_status_t status = BLK_STS_IOERR;
776
777 might_sleep();
778
779 /*
780 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
781 * if queue does not support NOWAIT.
782 */
783 if ((bio->bi_opf & REQ_NOWAIT) && !bdev_nowait(bdev))
784 goto not_supported;
785
786 if (should_fail_bio(bio))
787 goto end_io;
788 bio_check_ro(bio);
789 if (!bio_flagged(bio, BIO_REMAPPED)) {
790 if (unlikely(bio_check_eod(bio)))
791 goto end_io;
792 if (bdev_is_partition(bdev) &&
793 unlikely(blk_partition_remap(bio)))
794 goto end_io;
795 }
796
797 /*
798 * Filter flush bio's early so that bio based drivers without flush
799 * support don't have to worry about them.
800 */
801 if (op_is_flush(bio->bi_opf)) {
802 if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_WRITE &&
803 bio_op(bio) != REQ_OP_ZONE_APPEND))
804 goto end_io;
805 if (!bdev_write_cache(bdev)) {
806 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
807 if (!bio_sectors(bio)) {
808 status = BLK_STS_OK;
809 goto end_io;
810 }
811 }
812 }
813
814 switch (bio_op(bio)) {
815 case REQ_OP_READ:
816 break;
817 case REQ_OP_WRITE:
818 if (bio->bi_opf & REQ_ATOMIC) {
819 status = blk_validate_atomic_write_op_size(q, bio);
820 if (status != BLK_STS_OK)
821 goto end_io;
822 }
823 break;
824 case REQ_OP_FLUSH:
825 /*
826 * REQ_OP_FLUSH can't be submitted through bios, it is only
827 * synthetized in struct request by the flush state machine.
828 */
829 goto not_supported;
830 case REQ_OP_DISCARD:
831 if (!bdev_max_discard_sectors(bdev))
832 goto not_supported;
833 break;
834 case REQ_OP_SECURE_ERASE:
835 if (!bdev_max_secure_erase_sectors(bdev))
836 goto not_supported;
837 break;
838 case REQ_OP_ZONE_APPEND:
839 status = blk_check_zone_append(q, bio);
840 if (status != BLK_STS_OK)
841 goto end_io;
842 break;
843 case REQ_OP_WRITE_ZEROES:
844 if (!q->limits.max_write_zeroes_sectors)
845 goto not_supported;
846 break;
847 case REQ_OP_ZONE_RESET:
848 case REQ_OP_ZONE_OPEN:
849 case REQ_OP_ZONE_CLOSE:
850 case REQ_OP_ZONE_FINISH:
851 case REQ_OP_ZONE_RESET_ALL:
852 if (!bdev_is_zoned(bio->bi_bdev))
853 goto not_supported;
854 break;
855 case REQ_OP_DRV_IN:
856 case REQ_OP_DRV_OUT:
857 /*
858 * Driver private operations are only used with passthrough
859 * requests.
860 */
861 fallthrough;
862 default:
863 goto not_supported;
864 }
865
866 if (blk_throtl_bio(bio))
867 return;
868 submit_bio_noacct_nocheck(bio);
869 return;
870
871not_supported:
872 status = BLK_STS_NOTSUPP;
873end_io:
874 bio->bi_status = status;
875 bio_endio(bio);
876}
877EXPORT_SYMBOL(submit_bio_noacct);
878
879static void bio_set_ioprio(struct bio *bio)
880{
881 /* Nobody set ioprio so far? Initialize it based on task's nice value */
882 if (IOPRIO_PRIO_CLASS(bio->bi_ioprio) == IOPRIO_CLASS_NONE)
883 bio->bi_ioprio = get_current_ioprio();
884 blkcg_set_ioprio(bio);
885}
886
887/**
888 * submit_bio - submit a bio to the block device layer for I/O
889 * @bio: The &struct bio which describes the I/O
890 *
891 * submit_bio() is used to submit I/O requests to block devices. It is passed a
892 * fully set up &struct bio that describes the I/O that needs to be done. The
893 * bio will be send to the device described by the bi_bdev field.
894 *
895 * The success/failure status of the request, along with notification of
896 * completion, is delivered asynchronously through the ->bi_end_io() callback
897 * in @bio. The bio must NOT be touched by the caller until ->bi_end_io() has
898 * been called.
899 */
900void submit_bio(struct bio *bio)
901{
902 if (bio_op(bio) == REQ_OP_READ) {
903 task_io_account_read(bio->bi_iter.bi_size);
904 count_vm_events(PGPGIN, bio_sectors(bio));
905 } else if (bio_op(bio) == REQ_OP_WRITE) {
906 count_vm_events(PGPGOUT, bio_sectors(bio));
907 }
908
909 bio_set_ioprio(bio);
910 submit_bio_noacct(bio);
911}
912EXPORT_SYMBOL(submit_bio);
913
914/**
915 * bio_poll - poll for BIO completions
916 * @bio: bio to poll for
917 * @iob: batches of IO
918 * @flags: BLK_POLL_* flags that control the behavior
919 *
920 * Poll for completions on queue associated with the bio. Returns number of
921 * completed entries found.
922 *
923 * Note: the caller must either be the context that submitted @bio, or
924 * be in a RCU critical section to prevent freeing of @bio.
925 */
926int bio_poll(struct bio *bio, struct io_comp_batch *iob, unsigned int flags)
927{
928 blk_qc_t cookie = READ_ONCE(bio->bi_cookie);
929 struct block_device *bdev;
930 struct request_queue *q;
931 int ret = 0;
932
933 bdev = READ_ONCE(bio->bi_bdev);
934 if (!bdev)
935 return 0;
936
937 q = bdev_get_queue(bdev);
938 if (cookie == BLK_QC_T_NONE)
939 return 0;
940
941 blk_flush_plug(current->plug, false);
942
943 /*
944 * We need to be able to enter a frozen queue, similar to how
945 * timeouts also need to do that. If that is blocked, then we can
946 * have pending IO when a queue freeze is started, and then the
947 * wait for the freeze to finish will wait for polled requests to
948 * timeout as the poller is preventer from entering the queue and
949 * completing them. As long as we prevent new IO from being queued,
950 * that should be all that matters.
951 */
952 if (!percpu_ref_tryget(&q->q_usage_counter))
953 return 0;
954 if (queue_is_mq(q)) {
955 ret = blk_mq_poll(q, cookie, iob, flags);
956 } else {
957 struct gendisk *disk = q->disk;
958
959 if ((q->limits.features & BLK_FEAT_POLL) && disk &&
960 disk->fops->poll_bio)
961 ret = disk->fops->poll_bio(bio, iob, flags);
962 }
963 blk_queue_exit(q);
964 return ret;
965}
966EXPORT_SYMBOL_GPL(bio_poll);
967
968/*
969 * Helper to implement file_operations.iopoll. Requires the bio to be stored
970 * in iocb->private, and cleared before freeing the bio.
971 */
972int iocb_bio_iopoll(struct kiocb *kiocb, struct io_comp_batch *iob,
973 unsigned int flags)
974{
975 struct bio *bio;
976 int ret = 0;
977
978 /*
979 * Note: the bio cache only uses SLAB_TYPESAFE_BY_RCU, so bio can
980 * point to a freshly allocated bio at this point. If that happens
981 * we have a few cases to consider:
982 *
983 * 1) the bio is beeing initialized and bi_bdev is NULL. We can just
984 * simply nothing in this case
985 * 2) the bio points to a not poll enabled device. bio_poll will catch
986 * this and return 0
987 * 3) the bio points to a poll capable device, including but not
988 * limited to the one that the original bio pointed to. In this
989 * case we will call into the actual poll method and poll for I/O,
990 * even if we don't need to, but it won't cause harm either.
991 *
992 * For cases 2) and 3) above the RCU grace period ensures that bi_bdev
993 * is still allocated. Because partitions hold a reference to the whole
994 * device bdev and thus disk, the disk is also still valid. Grabbing
995 * a reference to the queue in bio_poll() ensures the hctxs and requests
996 * are still valid as well.
997 */
998 rcu_read_lock();
999 bio = READ_ONCE(kiocb->private);
1000 if (bio)
1001 ret = bio_poll(bio, iob, flags);
1002 rcu_read_unlock();
1003
1004 return ret;
1005}
1006EXPORT_SYMBOL_GPL(iocb_bio_iopoll);
1007
1008void update_io_ticks(struct block_device *part, unsigned long now, bool end)
1009{
1010 unsigned long stamp;
1011again:
1012 stamp = READ_ONCE(part->bd_stamp);
1013 if (unlikely(time_after(now, stamp)) &&
1014 likely(try_cmpxchg(&part->bd_stamp, &stamp, now)) &&
1015 (end || part_in_flight(part)))
1016 __part_stat_add(part, io_ticks, now - stamp);
1017
1018 if (bdev_is_partition(part)) {
1019 part = bdev_whole(part);
1020 goto again;
1021 }
1022}
1023
1024unsigned long bdev_start_io_acct(struct block_device *bdev, enum req_op op,
1025 unsigned long start_time)
1026{
1027 part_stat_lock();
1028 update_io_ticks(bdev, start_time, false);
1029 part_stat_local_inc(bdev, in_flight[op_is_write(op)]);
1030 part_stat_unlock();
1031
1032 return start_time;
1033}
1034EXPORT_SYMBOL(bdev_start_io_acct);
1035
1036/**
1037 * bio_start_io_acct - start I/O accounting for bio based drivers
1038 * @bio: bio to start account for
1039 *
1040 * Returns the start time that should be passed back to bio_end_io_acct().
1041 */
1042unsigned long bio_start_io_acct(struct bio *bio)
1043{
1044 return bdev_start_io_acct(bio->bi_bdev, bio_op(bio), jiffies);
1045}
1046EXPORT_SYMBOL_GPL(bio_start_io_acct);
1047
1048void bdev_end_io_acct(struct block_device *bdev, enum req_op op,
1049 unsigned int sectors, unsigned long start_time)
1050{
1051 const int sgrp = op_stat_group(op);
1052 unsigned long now = READ_ONCE(jiffies);
1053 unsigned long duration = now - start_time;
1054
1055 part_stat_lock();
1056 update_io_ticks(bdev, now, true);
1057 part_stat_inc(bdev, ios[sgrp]);
1058 part_stat_add(bdev, sectors[sgrp], sectors);
1059 part_stat_add(bdev, nsecs[sgrp], jiffies_to_nsecs(duration));
1060 part_stat_local_dec(bdev, in_flight[op_is_write(op)]);
1061 part_stat_unlock();
1062}
1063EXPORT_SYMBOL(bdev_end_io_acct);
1064
1065void bio_end_io_acct_remapped(struct bio *bio, unsigned long start_time,
1066 struct block_device *orig_bdev)
1067{
1068 bdev_end_io_acct(orig_bdev, bio_op(bio), bio_sectors(bio), start_time);
1069}
1070EXPORT_SYMBOL_GPL(bio_end_io_acct_remapped);
1071
1072/**
1073 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1074 * @q : the queue of the device being checked
1075 *
1076 * Description:
1077 * Check if underlying low-level drivers of a device are busy.
1078 * If the drivers want to export their busy state, they must set own
1079 * exporting function using blk_queue_lld_busy() first.
1080 *
1081 * Basically, this function is used only by request stacking drivers
1082 * to stop dispatching requests to underlying devices when underlying
1083 * devices are busy. This behavior helps more I/O merging on the queue
1084 * of the request stacking driver and prevents I/O throughput regression
1085 * on burst I/O load.
1086 *
1087 * Return:
1088 * 0 - Not busy (The request stacking driver should dispatch request)
1089 * 1 - Busy (The request stacking driver should stop dispatching request)
1090 */
1091int blk_lld_busy(struct request_queue *q)
1092{
1093 if (queue_is_mq(q) && q->mq_ops->busy)
1094 return q->mq_ops->busy(q);
1095
1096 return 0;
1097}
1098EXPORT_SYMBOL_GPL(blk_lld_busy);
1099
1100int kblockd_schedule_work(struct work_struct *work)
1101{
1102 return queue_work(kblockd_workqueue, work);
1103}
1104EXPORT_SYMBOL(kblockd_schedule_work);
1105
1106int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
1107 unsigned long delay)
1108{
1109 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
1110}
1111EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
1112
1113void blk_start_plug_nr_ios(struct blk_plug *plug, unsigned short nr_ios)
1114{
1115 struct task_struct *tsk = current;
1116
1117 /*
1118 * If this is a nested plug, don't actually assign it.
1119 */
1120 if (tsk->plug)
1121 return;
1122
1123 plug->cur_ktime = 0;
1124 rq_list_init(&plug->mq_list);
1125 rq_list_init(&plug->cached_rqs);
1126 plug->nr_ios = min_t(unsigned short, nr_ios, BLK_MAX_REQUEST_COUNT);
1127 plug->rq_count = 0;
1128 plug->multiple_queues = false;
1129 plug->has_elevator = false;
1130 INIT_LIST_HEAD(&plug->cb_list);
1131
1132 /*
1133 * Store ordering should not be needed here, since a potential
1134 * preempt will imply a full memory barrier
1135 */
1136 tsk->plug = plug;
1137}
1138
1139/**
1140 * blk_start_plug - initialize blk_plug and track it inside the task_struct
1141 * @plug: The &struct blk_plug that needs to be initialized
1142 *
1143 * Description:
1144 * blk_start_plug() indicates to the block layer an intent by the caller
1145 * to submit multiple I/O requests in a batch. The block layer may use
1146 * this hint to defer submitting I/Os from the caller until blk_finish_plug()
1147 * is called. However, the block layer may choose to submit requests
1148 * before a call to blk_finish_plug() if the number of queued I/Os
1149 * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1150 * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if
1151 * the task schedules (see below).
1152 *
1153 * Tracking blk_plug inside the task_struct will help with auto-flushing the
1154 * pending I/O should the task end up blocking between blk_start_plug() and
1155 * blk_finish_plug(). This is important from a performance perspective, but
1156 * also ensures that we don't deadlock. For instance, if the task is blocking
1157 * for a memory allocation, memory reclaim could end up wanting to free a
1158 * page belonging to that request that is currently residing in our private
1159 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
1160 * this kind of deadlock.
1161 */
1162void blk_start_plug(struct blk_plug *plug)
1163{
1164 blk_start_plug_nr_ios(plug, 1);
1165}
1166EXPORT_SYMBOL(blk_start_plug);
1167
1168static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
1169{
1170 LIST_HEAD(callbacks);
1171
1172 while (!list_empty(&plug->cb_list)) {
1173 list_splice_init(&plug->cb_list, &callbacks);
1174
1175 while (!list_empty(&callbacks)) {
1176 struct blk_plug_cb *cb = list_first_entry(&callbacks,
1177 struct blk_plug_cb,
1178 list);
1179 list_del(&cb->list);
1180 cb->callback(cb, from_schedule);
1181 }
1182 }
1183}
1184
1185struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
1186 int size)
1187{
1188 struct blk_plug *plug = current->plug;
1189 struct blk_plug_cb *cb;
1190
1191 if (!plug)
1192 return NULL;
1193
1194 list_for_each_entry(cb, &plug->cb_list, list)
1195 if (cb->callback == unplug && cb->data == data)
1196 return cb;
1197
1198 /* Not currently on the callback list */
1199 BUG_ON(size < sizeof(*cb));
1200 cb = kzalloc(size, GFP_ATOMIC);
1201 if (cb) {
1202 cb->data = data;
1203 cb->callback = unplug;
1204 list_add(&cb->list, &plug->cb_list);
1205 }
1206 return cb;
1207}
1208EXPORT_SYMBOL(blk_check_plugged);
1209
1210void __blk_flush_plug(struct blk_plug *plug, bool from_schedule)
1211{
1212 if (!list_empty(&plug->cb_list))
1213 flush_plug_callbacks(plug, from_schedule);
1214 blk_mq_flush_plug_list(plug, from_schedule);
1215 /*
1216 * Unconditionally flush out cached requests, even if the unplug
1217 * event came from schedule. Since we know hold references to the
1218 * queue for cached requests, we don't want a blocked task holding
1219 * up a queue freeze/quiesce event.
1220 */
1221 if (unlikely(!rq_list_empty(&plug->cached_rqs)))
1222 blk_mq_free_plug_rqs(plug);
1223
1224 plug->cur_ktime = 0;
1225 current->flags &= ~PF_BLOCK_TS;
1226}
1227
1228/**
1229 * blk_finish_plug - mark the end of a batch of submitted I/O
1230 * @plug: The &struct blk_plug passed to blk_start_plug()
1231 *
1232 * Description:
1233 * Indicate that a batch of I/O submissions is complete. This function
1234 * must be paired with an initial call to blk_start_plug(). The intent
1235 * is to allow the block layer to optimize I/O submission. See the
1236 * documentation for blk_start_plug() for more information.
1237 */
1238void blk_finish_plug(struct blk_plug *plug)
1239{
1240 if (plug == current->plug) {
1241 __blk_flush_plug(plug, false);
1242 current->plug = NULL;
1243 }
1244}
1245EXPORT_SYMBOL(blk_finish_plug);
1246
1247void blk_io_schedule(void)
1248{
1249 /* Prevent hang_check timer from firing at us during very long I/O */
1250 unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2;
1251
1252 if (timeout)
1253 io_schedule_timeout(timeout);
1254 else
1255 io_schedule();
1256}
1257EXPORT_SYMBOL_GPL(blk_io_schedule);
1258
1259int __init blk_dev_init(void)
1260{
1261 BUILD_BUG_ON((__force u32)REQ_OP_LAST >= (1 << REQ_OP_BITS));
1262 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1263 sizeof_field(struct request, cmd_flags));
1264 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1265 sizeof_field(struct bio, bi_opf));
1266
1267 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
1268 kblockd_workqueue = alloc_workqueue("kblockd",
1269 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1270 if (!kblockd_workqueue)
1271 panic("Failed to create kblockd\n");
1272
1273 blk_requestq_cachep = KMEM_CACHE(request_queue, SLAB_PANIC);
1274
1275 blk_debugfs_root = debugfs_create_dir("block", NULL);
1276
1277 return 0;
1278}