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>
8 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
12 * This handles all read/write requests to block devices
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>
22 #include <linux/kernel_stat.h>
23 #include <linux/string.h>
24 #include <linux/init.h>
25 #include <linux/completion.h>
26 #include <linux/slab.h>
27 #include <linux/swap.h>
28 #include <linux/writeback.h>
29 #include <linux/task_io_accounting_ops.h>
30 #include <linux/fault-inject.h>
31 #include <linux/list_sort.h>
32 #include <linux/delay.h>
33 #include <linux/ratelimit.h>
34 #include <linux/pm_runtime.h>
35 #include <linux/blk-cgroup.h>
36 #include <linux/debugfs.h>
38 #define CREATE_TRACE_POINTS
39 #include <trace/events/block.h>
43 #include "blk-mq-sched.h"
46 #ifdef CONFIG_DEBUG_FS
47 struct dentry
*blk_debugfs_root
;
50 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap
);
51 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap
);
52 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete
);
53 EXPORT_TRACEPOINT_SYMBOL_GPL(block_split
);
54 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug
);
56 DEFINE_IDA(blk_queue_ida
);
59 * For the allocated request tables
61 struct kmem_cache
*request_cachep
;
64 * For queue allocation
66 struct kmem_cache
*blk_requestq_cachep
;
69 * Controlling structure to kblockd
71 static struct workqueue_struct
*kblockd_workqueue
;
73 static void blk_clear_congested(struct request_list
*rl
, int sync
)
75 #ifdef CONFIG_CGROUP_WRITEBACK
76 clear_wb_congested(rl
->blkg
->wb_congested
, sync
);
79 * If !CGROUP_WRITEBACK, all blkg's map to bdi->wb and we shouldn't
80 * flip its congestion state for events on other blkcgs.
82 if (rl
== &rl
->q
->root_rl
)
83 clear_wb_congested(rl
->q
->backing_dev_info
->wb
.congested
, sync
);
87 static void blk_set_congested(struct request_list
*rl
, int sync
)
89 #ifdef CONFIG_CGROUP_WRITEBACK
90 set_wb_congested(rl
->blkg
->wb_congested
, sync
);
92 /* see blk_clear_congested() */
93 if (rl
== &rl
->q
->root_rl
)
94 set_wb_congested(rl
->q
->backing_dev_info
->wb
.congested
, sync
);
98 void blk_queue_congestion_threshold(struct request_queue
*q
)
102 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) + 1;
103 if (nr
> q
->nr_requests
)
105 q
->nr_congestion_on
= nr
;
107 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) - (q
->nr_requests
/ 16) - 1;
110 q
->nr_congestion_off
= nr
;
113 void blk_rq_init(struct request_queue
*q
, struct request
*rq
)
115 memset(rq
, 0, sizeof(*rq
));
117 INIT_LIST_HEAD(&rq
->queuelist
);
118 INIT_LIST_HEAD(&rq
->timeout_list
);
121 rq
->__sector
= (sector_t
) -1;
122 INIT_HLIST_NODE(&rq
->hash
);
123 RB_CLEAR_NODE(&rq
->rb_node
);
125 rq
->internal_tag
= -1;
126 rq
->start_time
= jiffies
;
127 set_start_time_ns(rq
);
130 EXPORT_SYMBOL(blk_rq_init
);
132 static const struct {
136 [BLK_STS_OK
] = { 0, "" },
137 [BLK_STS_NOTSUPP
] = { -EOPNOTSUPP
, "operation not supported" },
138 [BLK_STS_TIMEOUT
] = { -ETIMEDOUT
, "timeout" },
139 [BLK_STS_NOSPC
] = { -ENOSPC
, "critical space allocation" },
140 [BLK_STS_TRANSPORT
] = { -ENOLINK
, "recoverable transport" },
141 [BLK_STS_TARGET
] = { -EREMOTEIO
, "critical target" },
142 [BLK_STS_NEXUS
] = { -EBADE
, "critical nexus" },
143 [BLK_STS_MEDIUM
] = { -ENODATA
, "critical medium" },
144 [BLK_STS_PROTECTION
] = { -EILSEQ
, "protection" },
145 [BLK_STS_RESOURCE
] = { -ENOMEM
, "kernel resource" },
146 [BLK_STS_AGAIN
] = { -EAGAIN
, "nonblocking retry" },
148 /* device mapper special case, should not leak out: */
149 [BLK_STS_DM_REQUEUE
] = { -EREMCHG
, "dm internal retry" },
151 /* everything else not covered above: */
152 [BLK_STS_IOERR
] = { -EIO
, "I/O" },
155 blk_status_t
errno_to_blk_status(int errno
)
159 for (i
= 0; i
< ARRAY_SIZE(blk_errors
); i
++) {
160 if (blk_errors
[i
].errno
== errno
)
161 return (__force blk_status_t
)i
;
164 return BLK_STS_IOERR
;
166 EXPORT_SYMBOL_GPL(errno_to_blk_status
);
168 int blk_status_to_errno(blk_status_t status
)
170 int idx
= (__force
int)status
;
172 if (WARN_ON_ONCE(idx
>= ARRAY_SIZE(blk_errors
)))
174 return blk_errors
[idx
].errno
;
176 EXPORT_SYMBOL_GPL(blk_status_to_errno
);
178 static void print_req_error(struct request
*req
, blk_status_t status
)
180 int idx
= (__force
int)status
;
182 if (WARN_ON_ONCE(idx
>= ARRAY_SIZE(blk_errors
)))
185 printk_ratelimited(KERN_ERR
"%s: %s error, dev %s, sector %llu\n",
186 __func__
, blk_errors
[idx
].name
, req
->rq_disk
?
187 req
->rq_disk
->disk_name
: "?",
188 (unsigned long long)blk_rq_pos(req
));
191 static void req_bio_endio(struct request
*rq
, struct bio
*bio
,
192 unsigned int nbytes
, blk_status_t error
)
195 bio
->bi_status
= error
;
197 if (unlikely(rq
->rq_flags
& RQF_QUIET
))
198 bio_set_flag(bio
, BIO_QUIET
);
200 bio_advance(bio
, nbytes
);
202 /* don't actually finish bio if it's part of flush sequence */
203 if (bio
->bi_iter
.bi_size
== 0 && !(rq
->rq_flags
& RQF_FLUSH_SEQ
))
207 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
209 printk(KERN_INFO
"%s: dev %s: flags=%llx\n", msg
,
210 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?",
211 (unsigned long long) rq
->cmd_flags
);
213 printk(KERN_INFO
" sector %llu, nr/cnr %u/%u\n",
214 (unsigned long long)blk_rq_pos(rq
),
215 blk_rq_sectors(rq
), blk_rq_cur_sectors(rq
));
216 printk(KERN_INFO
" bio %p, biotail %p, len %u\n",
217 rq
->bio
, rq
->biotail
, blk_rq_bytes(rq
));
219 EXPORT_SYMBOL(blk_dump_rq_flags
);
221 static void blk_delay_work(struct work_struct
*work
)
223 struct request_queue
*q
;
225 q
= container_of(work
, struct request_queue
, delay_work
.work
);
226 spin_lock_irq(q
->queue_lock
);
228 spin_unlock_irq(q
->queue_lock
);
232 * blk_delay_queue - restart queueing after defined interval
233 * @q: The &struct request_queue in question
234 * @msecs: Delay in msecs
237 * Sometimes queueing needs to be postponed for a little while, to allow
238 * resources to come back. This function will make sure that queueing is
239 * restarted around the specified time.
241 void blk_delay_queue(struct request_queue
*q
, unsigned long msecs
)
243 lockdep_assert_held(q
->queue_lock
);
244 WARN_ON_ONCE(q
->mq_ops
);
246 if (likely(!blk_queue_dead(q
)))
247 queue_delayed_work(kblockd_workqueue
, &q
->delay_work
,
248 msecs_to_jiffies(msecs
));
250 EXPORT_SYMBOL(blk_delay_queue
);
253 * blk_start_queue_async - asynchronously restart a previously stopped queue
254 * @q: The &struct request_queue in question
257 * blk_start_queue_async() will clear the stop flag on the queue, and
258 * ensure that the request_fn for the queue is run from an async
261 void blk_start_queue_async(struct request_queue
*q
)
263 lockdep_assert_held(q
->queue_lock
);
264 WARN_ON_ONCE(q
->mq_ops
);
266 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
267 blk_run_queue_async(q
);
269 EXPORT_SYMBOL(blk_start_queue_async
);
272 * blk_start_queue - restart a previously stopped queue
273 * @q: The &struct request_queue in question
276 * blk_start_queue() will clear the stop flag on the queue, and call
277 * the request_fn for the queue if it was in a stopped state when
278 * entered. Also see blk_stop_queue().
280 void blk_start_queue(struct request_queue
*q
)
282 lockdep_assert_held(q
->queue_lock
);
283 WARN_ON(!in_interrupt() && !irqs_disabled());
284 WARN_ON_ONCE(q
->mq_ops
);
286 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
289 EXPORT_SYMBOL(blk_start_queue
);
292 * blk_stop_queue - stop a queue
293 * @q: The &struct request_queue in question
296 * The Linux block layer assumes that a block driver will consume all
297 * entries on the request queue when the request_fn strategy is called.
298 * Often this will not happen, because of hardware limitations (queue
299 * depth settings). If a device driver gets a 'queue full' response,
300 * or if it simply chooses not to queue more I/O at one point, it can
301 * call this function to prevent the request_fn from being called until
302 * the driver has signalled it's ready to go again. This happens by calling
303 * blk_start_queue() to restart queue operations.
305 void blk_stop_queue(struct request_queue
*q
)
307 lockdep_assert_held(q
->queue_lock
);
308 WARN_ON_ONCE(q
->mq_ops
);
310 cancel_delayed_work(&q
->delay_work
);
311 queue_flag_set(QUEUE_FLAG_STOPPED
, q
);
313 EXPORT_SYMBOL(blk_stop_queue
);
316 * blk_sync_queue - cancel any pending callbacks on a queue
320 * The block layer may perform asynchronous callback activity
321 * on a queue, such as calling the unplug function after a timeout.
322 * A block device may call blk_sync_queue to ensure that any
323 * such activity is cancelled, thus allowing it to release resources
324 * that the callbacks might use. The caller must already have made sure
325 * that its ->make_request_fn will not re-add plugging prior to calling
328 * This function does not cancel any asynchronous activity arising
329 * out of elevator or throttling code. That would require elevator_exit()
330 * and blkcg_exit_queue() to be called with queue lock initialized.
333 void blk_sync_queue(struct request_queue
*q
)
335 del_timer_sync(&q
->timeout
);
336 cancel_work_sync(&q
->timeout_work
);
339 struct blk_mq_hw_ctx
*hctx
;
342 cancel_delayed_work_sync(&q
->requeue_work
);
343 queue_for_each_hw_ctx(q
, hctx
, i
)
344 cancel_delayed_work_sync(&hctx
->run_work
);
346 cancel_delayed_work_sync(&q
->delay_work
);
349 EXPORT_SYMBOL(blk_sync_queue
);
352 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
353 * @q: The queue to run
356 * Invoke request handling on a queue if there are any pending requests.
357 * May be used to restart request handling after a request has completed.
358 * This variant runs the queue whether or not the queue has been
359 * stopped. Must be called with the queue lock held and interrupts
360 * disabled. See also @blk_run_queue.
362 inline void __blk_run_queue_uncond(struct request_queue
*q
)
364 lockdep_assert_held(q
->queue_lock
);
365 WARN_ON_ONCE(q
->mq_ops
);
367 if (unlikely(blk_queue_dead(q
)))
371 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
372 * the queue lock internally. As a result multiple threads may be
373 * running such a request function concurrently. Keep track of the
374 * number of active request_fn invocations such that blk_drain_queue()
375 * can wait until all these request_fn calls have finished.
377 q
->request_fn_active
++;
379 q
->request_fn_active
--;
381 EXPORT_SYMBOL_GPL(__blk_run_queue_uncond
);
384 * __blk_run_queue - run a single device queue
385 * @q: The queue to run
388 * See @blk_run_queue.
390 void __blk_run_queue(struct request_queue
*q
)
392 lockdep_assert_held(q
->queue_lock
);
393 WARN_ON_ONCE(q
->mq_ops
);
395 if (unlikely(blk_queue_stopped(q
)))
398 __blk_run_queue_uncond(q
);
400 EXPORT_SYMBOL(__blk_run_queue
);
403 * blk_run_queue_async - run a single device queue in workqueue context
404 * @q: The queue to run
407 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
411 * Since it is not allowed to run q->delay_work after blk_cleanup_queue()
412 * has canceled q->delay_work, callers must hold the queue lock to avoid
413 * race conditions between blk_cleanup_queue() and blk_run_queue_async().
415 void blk_run_queue_async(struct request_queue
*q
)
417 lockdep_assert_held(q
->queue_lock
);
418 WARN_ON_ONCE(q
->mq_ops
);
420 if (likely(!blk_queue_stopped(q
) && !blk_queue_dead(q
)))
421 mod_delayed_work(kblockd_workqueue
, &q
->delay_work
, 0);
423 EXPORT_SYMBOL(blk_run_queue_async
);
426 * blk_run_queue - run a single device queue
427 * @q: The queue to run
430 * Invoke request handling on this queue, if it has pending work to do.
431 * May be used to restart queueing when a request has completed.
433 void blk_run_queue(struct request_queue
*q
)
437 WARN_ON_ONCE(q
->mq_ops
);
439 spin_lock_irqsave(q
->queue_lock
, flags
);
441 spin_unlock_irqrestore(q
->queue_lock
, flags
);
443 EXPORT_SYMBOL(blk_run_queue
);
445 void blk_put_queue(struct request_queue
*q
)
447 kobject_put(&q
->kobj
);
449 EXPORT_SYMBOL(blk_put_queue
);
452 * __blk_drain_queue - drain requests from request_queue
454 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
456 * Drain requests from @q. If @drain_all is set, all requests are drained.
457 * If not, only ELVPRIV requests are drained. The caller is responsible
458 * for ensuring that no new requests which need to be drained are queued.
460 static void __blk_drain_queue(struct request_queue
*q
, bool drain_all
)
461 __releases(q
->queue_lock
)
462 __acquires(q
->queue_lock
)
466 lockdep_assert_held(q
->queue_lock
);
467 WARN_ON_ONCE(q
->mq_ops
);
473 * The caller might be trying to drain @q before its
474 * elevator is initialized.
477 elv_drain_elevator(q
);
479 blkcg_drain_queue(q
);
482 * This function might be called on a queue which failed
483 * driver init after queue creation or is not yet fully
484 * active yet. Some drivers (e.g. fd and loop) get unhappy
485 * in such cases. Kick queue iff dispatch queue has
486 * something on it and @q has request_fn set.
488 if (!list_empty(&q
->queue_head
) && q
->request_fn
)
491 drain
|= q
->nr_rqs_elvpriv
;
492 drain
|= q
->request_fn_active
;
495 * Unfortunately, requests are queued at and tracked from
496 * multiple places and there's no single counter which can
497 * be drained. Check all the queues and counters.
500 struct blk_flush_queue
*fq
= blk_get_flush_queue(q
, NULL
);
501 drain
|= !list_empty(&q
->queue_head
);
502 for (i
= 0; i
< 2; i
++) {
503 drain
|= q
->nr_rqs
[i
];
504 drain
|= q
->in_flight
[i
];
506 drain
|= !list_empty(&fq
->flush_queue
[i
]);
513 spin_unlock_irq(q
->queue_lock
);
517 spin_lock_irq(q
->queue_lock
);
521 * With queue marked dead, any woken up waiter will fail the
522 * allocation path, so the wakeup chaining is lost and we're
523 * left with hung waiters. We need to wake up those waiters.
526 struct request_list
*rl
;
528 blk_queue_for_each_rl(rl
, q
)
529 for (i
= 0; i
< ARRAY_SIZE(rl
->wait
); i
++)
530 wake_up_all(&rl
->wait
[i
]);
535 * blk_queue_bypass_start - enter queue bypass mode
536 * @q: queue of interest
538 * In bypass mode, only the dispatch FIFO queue of @q is used. This
539 * function makes @q enter bypass mode and drains all requests which were
540 * throttled or issued before. On return, it's guaranteed that no request
541 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
542 * inside queue or RCU read lock.
544 void blk_queue_bypass_start(struct request_queue
*q
)
546 WARN_ON_ONCE(q
->mq_ops
);
548 spin_lock_irq(q
->queue_lock
);
550 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
551 spin_unlock_irq(q
->queue_lock
);
554 * Queues start drained. Skip actual draining till init is
555 * complete. This avoids lenghty delays during queue init which
556 * can happen many times during boot.
558 if (blk_queue_init_done(q
)) {
559 spin_lock_irq(q
->queue_lock
);
560 __blk_drain_queue(q
, false);
561 spin_unlock_irq(q
->queue_lock
);
563 /* ensure blk_queue_bypass() is %true inside RCU read lock */
567 EXPORT_SYMBOL_GPL(blk_queue_bypass_start
);
570 * blk_queue_bypass_end - leave queue bypass mode
571 * @q: queue of interest
573 * Leave bypass mode and restore the normal queueing behavior.
575 * Note: although blk_queue_bypass_start() is only called for blk-sq queues,
576 * this function is called for both blk-sq and blk-mq queues.
578 void blk_queue_bypass_end(struct request_queue
*q
)
580 spin_lock_irq(q
->queue_lock
);
581 if (!--q
->bypass_depth
)
582 queue_flag_clear(QUEUE_FLAG_BYPASS
, q
);
583 WARN_ON_ONCE(q
->bypass_depth
< 0);
584 spin_unlock_irq(q
->queue_lock
);
586 EXPORT_SYMBOL_GPL(blk_queue_bypass_end
);
588 void blk_set_queue_dying(struct request_queue
*q
)
590 spin_lock_irq(q
->queue_lock
);
591 queue_flag_set(QUEUE_FLAG_DYING
, q
);
592 spin_unlock_irq(q
->queue_lock
);
595 * When queue DYING flag is set, we need to block new req
596 * entering queue, so we call blk_freeze_queue_start() to
597 * prevent I/O from crossing blk_queue_enter().
599 blk_freeze_queue_start(q
);
602 blk_mq_wake_waiters(q
);
604 struct request_list
*rl
;
606 spin_lock_irq(q
->queue_lock
);
607 blk_queue_for_each_rl(rl
, q
) {
609 wake_up(&rl
->wait
[BLK_RW_SYNC
]);
610 wake_up(&rl
->wait
[BLK_RW_ASYNC
]);
613 spin_unlock_irq(q
->queue_lock
);
616 /* Make blk_queue_enter() reexamine the DYING flag. */
617 wake_up_all(&q
->mq_freeze_wq
);
619 EXPORT_SYMBOL_GPL(blk_set_queue_dying
);
622 * blk_cleanup_queue - shutdown a request queue
623 * @q: request queue to shutdown
625 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
626 * put it. All future requests will be failed immediately with -ENODEV.
628 void blk_cleanup_queue(struct request_queue
*q
)
630 spinlock_t
*lock
= q
->queue_lock
;
632 /* mark @q DYING, no new request or merges will be allowed afterwards */
633 mutex_lock(&q
->sysfs_lock
);
634 blk_set_queue_dying(q
);
638 * A dying queue is permanently in bypass mode till released. Note
639 * that, unlike blk_queue_bypass_start(), we aren't performing
640 * synchronize_rcu() after entering bypass mode to avoid the delay
641 * as some drivers create and destroy a lot of queues while
642 * probing. This is still safe because blk_release_queue() will be
643 * called only after the queue refcnt drops to zero and nothing,
644 * RCU or not, would be traversing the queue by then.
647 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
649 queue_flag_set(QUEUE_FLAG_NOMERGES
, q
);
650 queue_flag_set(QUEUE_FLAG_NOXMERGES
, q
);
651 queue_flag_set(QUEUE_FLAG_DYING
, q
);
652 spin_unlock_irq(lock
);
653 mutex_unlock(&q
->sysfs_lock
);
656 * Drain all requests queued before DYING marking. Set DEAD flag to
657 * prevent that q->request_fn() gets invoked after draining finished.
662 __blk_drain_queue(q
, true);
663 queue_flag_set(QUEUE_FLAG_DEAD
, q
);
664 spin_unlock_irq(lock
);
666 /* for synchronous bio-based driver finish in-flight integrity i/o */
667 blk_flush_integrity();
669 /* @q won't process any more request, flush async actions */
670 del_timer_sync(&q
->backing_dev_info
->laptop_mode_wb_timer
);
674 blk_mq_free_queue(q
);
675 percpu_ref_exit(&q
->q_usage_counter
);
678 if (q
->queue_lock
!= &q
->__queue_lock
)
679 q
->queue_lock
= &q
->__queue_lock
;
680 spin_unlock_irq(lock
);
682 /* @q is and will stay empty, shutdown and put */
685 EXPORT_SYMBOL(blk_cleanup_queue
);
687 /* Allocate memory local to the request queue */
688 static void *alloc_request_simple(gfp_t gfp_mask
, void *data
)
690 struct request_queue
*q
= data
;
692 return kmem_cache_alloc_node(request_cachep
, gfp_mask
, q
->node
);
695 static void free_request_simple(void *element
, void *data
)
697 kmem_cache_free(request_cachep
, element
);
700 static void *alloc_request_size(gfp_t gfp_mask
, void *data
)
702 struct request_queue
*q
= data
;
705 rq
= kmalloc_node(sizeof(struct request
) + q
->cmd_size
, gfp_mask
,
707 if (rq
&& q
->init_rq_fn
&& q
->init_rq_fn(q
, rq
, gfp_mask
) < 0) {
714 static void free_request_size(void *element
, void *data
)
716 struct request_queue
*q
= data
;
719 q
->exit_rq_fn(q
, element
);
723 int blk_init_rl(struct request_list
*rl
, struct request_queue
*q
,
726 if (unlikely(rl
->rq_pool
) || q
->mq_ops
)
730 rl
->count
[BLK_RW_SYNC
] = rl
->count
[BLK_RW_ASYNC
] = 0;
731 rl
->starved
[BLK_RW_SYNC
] = rl
->starved
[BLK_RW_ASYNC
] = 0;
732 init_waitqueue_head(&rl
->wait
[BLK_RW_SYNC
]);
733 init_waitqueue_head(&rl
->wait
[BLK_RW_ASYNC
]);
736 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
,
737 alloc_request_size
, free_request_size
,
738 q
, gfp_mask
, q
->node
);
740 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
,
741 alloc_request_simple
, free_request_simple
,
742 q
, gfp_mask
, q
->node
);
747 if (rl
!= &q
->root_rl
)
748 WARN_ON_ONCE(!blk_get_queue(q
));
753 void blk_exit_rl(struct request_queue
*q
, struct request_list
*rl
)
756 mempool_destroy(rl
->rq_pool
);
757 if (rl
!= &q
->root_rl
)
762 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
764 return blk_alloc_queue_node(gfp_mask
, NUMA_NO_NODE
);
766 EXPORT_SYMBOL(blk_alloc_queue
);
768 int blk_queue_enter(struct request_queue
*q
, bool nowait
)
773 if (percpu_ref_tryget_live(&q
->q_usage_counter
))
780 * read pair of barrier in blk_freeze_queue_start(),
781 * we need to order reading __PERCPU_REF_DEAD flag of
782 * .q_usage_counter and reading .mq_freeze_depth or
783 * queue dying flag, otherwise the following wait may
784 * never return if the two reads are reordered.
788 ret
= wait_event_interruptible(q
->mq_freeze_wq
,
789 !atomic_read(&q
->mq_freeze_depth
) ||
791 if (blk_queue_dying(q
))
798 void blk_queue_exit(struct request_queue
*q
)
800 percpu_ref_put(&q
->q_usage_counter
);
803 static void blk_queue_usage_counter_release(struct percpu_ref
*ref
)
805 struct request_queue
*q
=
806 container_of(ref
, struct request_queue
, q_usage_counter
);
808 wake_up_all(&q
->mq_freeze_wq
);
811 static void blk_rq_timed_out_timer(unsigned long data
)
813 struct request_queue
*q
= (struct request_queue
*)data
;
815 kblockd_schedule_work(&q
->timeout_work
);
818 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
820 struct request_queue
*q
;
822 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
823 gfp_mask
| __GFP_ZERO
, node_id
);
827 q
->id
= ida_simple_get(&blk_queue_ida
, 0, 0, gfp_mask
);
831 q
->bio_split
= bioset_create(BIO_POOL_SIZE
, 0, BIOSET_NEED_BVECS
);
835 q
->backing_dev_info
= bdi_alloc_node(gfp_mask
, node_id
);
836 if (!q
->backing_dev_info
)
839 q
->stats
= blk_alloc_queue_stats();
843 q
->backing_dev_info
->ra_pages
=
844 (VM_MAX_READAHEAD
* 1024) / PAGE_SIZE
;
845 q
->backing_dev_info
->capabilities
= BDI_CAP_CGROUP_WRITEBACK
;
846 q
->backing_dev_info
->name
= "block";
849 setup_timer(&q
->backing_dev_info
->laptop_mode_wb_timer
,
850 laptop_mode_timer_fn
, (unsigned long) q
);
851 setup_timer(&q
->timeout
, blk_rq_timed_out_timer
, (unsigned long) q
);
852 INIT_WORK(&q
->timeout_work
, NULL
);
853 INIT_LIST_HEAD(&q
->queue_head
);
854 INIT_LIST_HEAD(&q
->timeout_list
);
855 INIT_LIST_HEAD(&q
->icq_list
);
856 #ifdef CONFIG_BLK_CGROUP
857 INIT_LIST_HEAD(&q
->blkg_list
);
859 INIT_DELAYED_WORK(&q
->delay_work
, blk_delay_work
);
861 kobject_init(&q
->kobj
, &blk_queue_ktype
);
863 #ifdef CONFIG_BLK_DEV_IO_TRACE
864 mutex_init(&q
->blk_trace_mutex
);
866 mutex_init(&q
->sysfs_lock
);
867 spin_lock_init(&q
->__queue_lock
);
870 * By default initialize queue_lock to internal lock and driver can
871 * override it later if need be.
873 q
->queue_lock
= &q
->__queue_lock
;
876 * A queue starts its life with bypass turned on to avoid
877 * unnecessary bypass on/off overhead and nasty surprises during
878 * init. The initial bypass will be finished when the queue is
879 * registered by blk_register_queue().
882 __set_bit(QUEUE_FLAG_BYPASS
, &q
->queue_flags
);
884 init_waitqueue_head(&q
->mq_freeze_wq
);
887 * Init percpu_ref in atomic mode so that it's faster to shutdown.
888 * See blk_register_queue() for details.
890 if (percpu_ref_init(&q
->q_usage_counter
,
891 blk_queue_usage_counter_release
,
892 PERCPU_REF_INIT_ATOMIC
, GFP_KERNEL
))
895 if (blkcg_init_queue(q
))
901 percpu_ref_exit(&q
->q_usage_counter
);
903 blk_free_queue_stats(q
->stats
);
905 bdi_put(q
->backing_dev_info
);
907 bioset_free(q
->bio_split
);
909 ida_simple_remove(&blk_queue_ida
, q
->id
);
911 kmem_cache_free(blk_requestq_cachep
, q
);
914 EXPORT_SYMBOL(blk_alloc_queue_node
);
917 * blk_init_queue - prepare a request queue for use with a block device
918 * @rfn: The function to be called to process requests that have been
919 * placed on the queue.
920 * @lock: Request queue spin lock
923 * If a block device wishes to use the standard request handling procedures,
924 * which sorts requests and coalesces adjacent requests, then it must
925 * call blk_init_queue(). The function @rfn will be called when there
926 * are requests on the queue that need to be processed. If the device
927 * supports plugging, then @rfn may not be called immediately when requests
928 * are available on the queue, but may be called at some time later instead.
929 * Plugged queues are generally unplugged when a buffer belonging to one
930 * of the requests on the queue is needed, or due to memory pressure.
932 * @rfn is not required, or even expected, to remove all requests off the
933 * queue, but only as many as it can handle at a time. If it does leave
934 * requests on the queue, it is responsible for arranging that the requests
935 * get dealt with eventually.
937 * The queue spin lock must be held while manipulating the requests on the
938 * request queue; this lock will be taken also from interrupt context, so irq
939 * disabling is needed for it.
941 * Function returns a pointer to the initialized request queue, or %NULL if
945 * blk_init_queue() must be paired with a blk_cleanup_queue() call
946 * when the block device is deactivated (such as at module unload).
949 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
951 return blk_init_queue_node(rfn
, lock
, NUMA_NO_NODE
);
953 EXPORT_SYMBOL(blk_init_queue
);
955 struct request_queue
*
956 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
958 struct request_queue
*q
;
960 q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
966 q
->queue_lock
= lock
;
967 if (blk_init_allocated_queue(q
) < 0) {
968 blk_cleanup_queue(q
);
974 EXPORT_SYMBOL(blk_init_queue_node
);
976 static blk_qc_t
blk_queue_bio(struct request_queue
*q
, struct bio
*bio
);
979 int blk_init_allocated_queue(struct request_queue
*q
)
981 WARN_ON_ONCE(q
->mq_ops
);
983 q
->fq
= blk_alloc_flush_queue(q
, NUMA_NO_NODE
, q
->cmd_size
);
987 if (q
->init_rq_fn
&& q
->init_rq_fn(q
, q
->fq
->flush_rq
, GFP_KERNEL
))
988 goto out_free_flush_queue
;
990 if (blk_init_rl(&q
->root_rl
, q
, GFP_KERNEL
))
991 goto out_exit_flush_rq
;
993 INIT_WORK(&q
->timeout_work
, blk_timeout_work
);
994 q
->queue_flags
|= QUEUE_FLAG_DEFAULT
;
997 * This also sets hw/phys segments, boundary and size
999 blk_queue_make_request(q
, blk_queue_bio
);
1001 q
->sg_reserved_size
= INT_MAX
;
1003 /* Protect q->elevator from elevator_change */
1004 mutex_lock(&q
->sysfs_lock
);
1007 if (elevator_init(q
, NULL
)) {
1008 mutex_unlock(&q
->sysfs_lock
);
1009 goto out_exit_flush_rq
;
1012 mutex_unlock(&q
->sysfs_lock
);
1017 q
->exit_rq_fn(q
, q
->fq
->flush_rq
);
1018 out_free_flush_queue
:
1019 blk_free_flush_queue(q
->fq
);
1022 EXPORT_SYMBOL(blk_init_allocated_queue
);
1024 bool blk_get_queue(struct request_queue
*q
)
1026 if (likely(!blk_queue_dying(q
))) {
1033 EXPORT_SYMBOL(blk_get_queue
);
1035 static inline void blk_free_request(struct request_list
*rl
, struct request
*rq
)
1037 if (rq
->rq_flags
& RQF_ELVPRIV
) {
1038 elv_put_request(rl
->q
, rq
);
1040 put_io_context(rq
->elv
.icq
->ioc
);
1043 mempool_free(rq
, rl
->rq_pool
);
1047 * ioc_batching returns true if the ioc is a valid batching request and
1048 * should be given priority access to a request.
1050 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
1056 * Make sure the process is able to allocate at least 1 request
1057 * even if the batch times out, otherwise we could theoretically
1060 return ioc
->nr_batch_requests
== q
->nr_batching
||
1061 (ioc
->nr_batch_requests
> 0
1062 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
1066 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
1067 * will cause the process to be a "batcher" on all queues in the system. This
1068 * is the behaviour we want though - once it gets a wakeup it should be given
1071 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
1073 if (!ioc
|| ioc_batching(q
, ioc
))
1076 ioc
->nr_batch_requests
= q
->nr_batching
;
1077 ioc
->last_waited
= jiffies
;
1080 static void __freed_request(struct request_list
*rl
, int sync
)
1082 struct request_queue
*q
= rl
->q
;
1084 if (rl
->count
[sync
] < queue_congestion_off_threshold(q
))
1085 blk_clear_congested(rl
, sync
);
1087 if (rl
->count
[sync
] + 1 <= q
->nr_requests
) {
1088 if (waitqueue_active(&rl
->wait
[sync
]))
1089 wake_up(&rl
->wait
[sync
]);
1091 blk_clear_rl_full(rl
, sync
);
1096 * A request has just been released. Account for it, update the full and
1097 * congestion status, wake up any waiters. Called under q->queue_lock.
1099 static void freed_request(struct request_list
*rl
, bool sync
,
1100 req_flags_t rq_flags
)
1102 struct request_queue
*q
= rl
->q
;
1106 if (rq_flags
& RQF_ELVPRIV
)
1107 q
->nr_rqs_elvpriv
--;
1109 __freed_request(rl
, sync
);
1111 if (unlikely(rl
->starved
[sync
^ 1]))
1112 __freed_request(rl
, sync
^ 1);
1115 int blk_update_nr_requests(struct request_queue
*q
, unsigned int nr
)
1117 struct request_list
*rl
;
1118 int on_thresh
, off_thresh
;
1120 WARN_ON_ONCE(q
->mq_ops
);
1122 spin_lock_irq(q
->queue_lock
);
1123 q
->nr_requests
= nr
;
1124 blk_queue_congestion_threshold(q
);
1125 on_thresh
= queue_congestion_on_threshold(q
);
1126 off_thresh
= queue_congestion_off_threshold(q
);
1128 blk_queue_for_each_rl(rl
, q
) {
1129 if (rl
->count
[BLK_RW_SYNC
] >= on_thresh
)
1130 blk_set_congested(rl
, BLK_RW_SYNC
);
1131 else if (rl
->count
[BLK_RW_SYNC
] < off_thresh
)
1132 blk_clear_congested(rl
, BLK_RW_SYNC
);
1134 if (rl
->count
[BLK_RW_ASYNC
] >= on_thresh
)
1135 blk_set_congested(rl
, BLK_RW_ASYNC
);
1136 else if (rl
->count
[BLK_RW_ASYNC
] < off_thresh
)
1137 blk_clear_congested(rl
, BLK_RW_ASYNC
);
1139 if (rl
->count
[BLK_RW_SYNC
] >= q
->nr_requests
) {
1140 blk_set_rl_full(rl
, BLK_RW_SYNC
);
1142 blk_clear_rl_full(rl
, BLK_RW_SYNC
);
1143 wake_up(&rl
->wait
[BLK_RW_SYNC
]);
1146 if (rl
->count
[BLK_RW_ASYNC
] >= q
->nr_requests
) {
1147 blk_set_rl_full(rl
, BLK_RW_ASYNC
);
1149 blk_clear_rl_full(rl
, BLK_RW_ASYNC
);
1150 wake_up(&rl
->wait
[BLK_RW_ASYNC
]);
1154 spin_unlock_irq(q
->queue_lock
);
1159 * __get_request - get a free request
1160 * @rl: request list to allocate from
1161 * @op: operation and flags
1162 * @bio: bio to allocate request for (can be %NULL)
1163 * @flags: BLQ_MQ_REQ_* flags
1165 * Get a free request from @q. This function may fail under memory
1166 * pressure or if @q is dead.
1168 * Must be called with @q->queue_lock held and,
1169 * Returns ERR_PTR on failure, with @q->queue_lock held.
1170 * Returns request pointer on success, with @q->queue_lock *not held*.
1172 static struct request
*__get_request(struct request_list
*rl
, unsigned int op
,
1173 struct bio
*bio
, unsigned int flags
)
1175 struct request_queue
*q
= rl
->q
;
1177 struct elevator_type
*et
= q
->elevator
->type
;
1178 struct io_context
*ioc
= rq_ioc(bio
);
1179 struct io_cq
*icq
= NULL
;
1180 const bool is_sync
= op_is_sync(op
);
1182 gfp_t gfp_mask
= flags
& BLK_MQ_REQ_NOWAIT
? GFP_ATOMIC
:
1183 __GFP_DIRECT_RECLAIM
;
1184 req_flags_t rq_flags
= RQF_ALLOCED
;
1186 lockdep_assert_held(q
->queue_lock
);
1188 if (unlikely(blk_queue_dying(q
)))
1189 return ERR_PTR(-ENODEV
);
1191 may_queue
= elv_may_queue(q
, op
);
1192 if (may_queue
== ELV_MQUEUE_NO
)
1195 if (rl
->count
[is_sync
]+1 >= queue_congestion_on_threshold(q
)) {
1196 if (rl
->count
[is_sync
]+1 >= q
->nr_requests
) {
1198 * The queue will fill after this allocation, so set
1199 * it as full, and mark this process as "batching".
1200 * This process will be allowed to complete a batch of
1201 * requests, others will be blocked.
1203 if (!blk_rl_full(rl
, is_sync
)) {
1204 ioc_set_batching(q
, ioc
);
1205 blk_set_rl_full(rl
, is_sync
);
1207 if (may_queue
!= ELV_MQUEUE_MUST
1208 && !ioc_batching(q
, ioc
)) {
1210 * The queue is full and the allocating
1211 * process is not a "batcher", and not
1212 * exempted by the IO scheduler
1214 return ERR_PTR(-ENOMEM
);
1218 blk_set_congested(rl
, is_sync
);
1222 * Only allow batching queuers to allocate up to 50% over the defined
1223 * limit of requests, otherwise we could have thousands of requests
1224 * allocated with any setting of ->nr_requests
1226 if (rl
->count
[is_sync
] >= (3 * q
->nr_requests
/ 2))
1227 return ERR_PTR(-ENOMEM
);
1229 q
->nr_rqs
[is_sync
]++;
1230 rl
->count
[is_sync
]++;
1231 rl
->starved
[is_sync
] = 0;
1234 * Decide whether the new request will be managed by elevator. If
1235 * so, mark @rq_flags and increment elvpriv. Non-zero elvpriv will
1236 * prevent the current elevator from being destroyed until the new
1237 * request is freed. This guarantees icq's won't be destroyed and
1238 * makes creating new ones safe.
1240 * Flush requests do not use the elevator so skip initialization.
1241 * This allows a request to share the flush and elevator data.
1243 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1244 * it will be created after releasing queue_lock.
1246 if (!op_is_flush(op
) && !blk_queue_bypass(q
)) {
1247 rq_flags
|= RQF_ELVPRIV
;
1248 q
->nr_rqs_elvpriv
++;
1249 if (et
->icq_cache
&& ioc
)
1250 icq
= ioc_lookup_icq(ioc
, q
);
1253 if (blk_queue_io_stat(q
))
1254 rq_flags
|= RQF_IO_STAT
;
1255 spin_unlock_irq(q
->queue_lock
);
1257 /* allocate and init request */
1258 rq
= mempool_alloc(rl
->rq_pool
, gfp_mask
);
1263 blk_rq_set_rl(rq
, rl
);
1265 rq
->rq_flags
= rq_flags
;
1266 if (flags
& BLK_MQ_REQ_PREEMPT
)
1267 rq
->rq_flags
|= RQF_PREEMPT
;
1270 if (rq_flags
& RQF_ELVPRIV
) {
1271 if (unlikely(et
->icq_cache
&& !icq
)) {
1273 icq
= ioc_create_icq(ioc
, q
, gfp_mask
);
1279 if (unlikely(elv_set_request(q
, rq
, bio
, gfp_mask
)))
1282 /* @rq->elv.icq holds io_context until @rq is freed */
1284 get_io_context(icq
->ioc
);
1288 * ioc may be NULL here, and ioc_batching will be false. That's
1289 * OK, if the queue is under the request limit then requests need
1290 * not count toward the nr_batch_requests limit. There will always
1291 * be some limit enforced by BLK_BATCH_TIME.
1293 if (ioc_batching(q
, ioc
))
1294 ioc
->nr_batch_requests
--;
1296 trace_block_getrq(q
, bio
, op
);
1301 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1302 * and may fail indefinitely under memory pressure and thus
1303 * shouldn't stall IO. Treat this request as !elvpriv. This will
1304 * disturb iosched and blkcg but weird is bettern than dead.
1306 printk_ratelimited(KERN_WARNING
"%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1307 __func__
, dev_name(q
->backing_dev_info
->dev
));
1309 rq
->rq_flags
&= ~RQF_ELVPRIV
;
1312 spin_lock_irq(q
->queue_lock
);
1313 q
->nr_rqs_elvpriv
--;
1314 spin_unlock_irq(q
->queue_lock
);
1319 * Allocation failed presumably due to memory. Undo anything we
1320 * might have messed up.
1322 * Allocating task should really be put onto the front of the wait
1323 * queue, but this is pretty rare.
1325 spin_lock_irq(q
->queue_lock
);
1326 freed_request(rl
, is_sync
, rq_flags
);
1329 * in the very unlikely event that allocation failed and no
1330 * requests for this direction was pending, mark us starved so that
1331 * freeing of a request in the other direction will notice
1332 * us. another possible fix would be to split the rq mempool into
1336 if (unlikely(rl
->count
[is_sync
] == 0))
1337 rl
->starved
[is_sync
] = 1;
1338 return ERR_PTR(-ENOMEM
);
1342 * get_request - get a free request
1343 * @q: request_queue to allocate request from
1344 * @op: operation and flags
1345 * @bio: bio to allocate request for (can be %NULL)
1346 * @flags: BLK_MQ_REQ_* flags.
1348 * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask,
1349 * this function keeps retrying under memory pressure and fails iff @q is dead.
1351 * Must be called with @q->queue_lock held and,
1352 * Returns ERR_PTR on failure, with @q->queue_lock held.
1353 * Returns request pointer on success, with @q->queue_lock *not held*.
1355 static struct request
*get_request(struct request_queue
*q
, unsigned int op
,
1356 struct bio
*bio
, unsigned int flags
)
1358 const bool is_sync
= op_is_sync(op
);
1360 struct request_list
*rl
;
1363 lockdep_assert_held(q
->queue_lock
);
1364 WARN_ON_ONCE(q
->mq_ops
);
1366 rl
= blk_get_rl(q
, bio
); /* transferred to @rq on success */
1368 rq
= __get_request(rl
, op
, bio
, flags
);
1372 if (op
& REQ_NOWAIT
) {
1374 return ERR_PTR(-EAGAIN
);
1377 if ((flags
& BLK_MQ_REQ_NOWAIT
) || unlikely(blk_queue_dying(q
))) {
1382 /* wait on @rl and retry */
1383 prepare_to_wait_exclusive(&rl
->wait
[is_sync
], &wait
,
1384 TASK_UNINTERRUPTIBLE
);
1386 trace_block_sleeprq(q
, bio
, op
);
1388 spin_unlock_irq(q
->queue_lock
);
1392 * After sleeping, we become a "batching" process and will be able
1393 * to allocate at least one request, and up to a big batch of them
1394 * for a small period time. See ioc_batching, ioc_set_batching
1396 ioc_set_batching(q
, current
->io_context
);
1398 spin_lock_irq(q
->queue_lock
);
1399 finish_wait(&rl
->wait
[is_sync
], &wait
);
1404 /* flags: BLK_MQ_REQ_PREEMPT and/or BLK_MQ_REQ_NOWAIT. */
1405 static struct request
*blk_old_get_request(struct request_queue
*q
,
1406 unsigned int op
, unsigned int flags
)
1409 gfp_t gfp_mask
= flags
& BLK_MQ_REQ_NOWAIT
? GFP_ATOMIC
:
1410 __GFP_DIRECT_RECLAIM
;
1413 WARN_ON_ONCE(q
->mq_ops
);
1415 /* create ioc upfront */
1416 create_io_context(gfp_mask
, q
->node
);
1418 ret
= blk_queue_enter(q
, !(gfp_mask
& __GFP_DIRECT_RECLAIM
) ||
1421 return ERR_PTR(ret
);
1422 spin_lock_irq(q
->queue_lock
);
1423 rq
= get_request(q
, op
, NULL
, flags
);
1425 spin_unlock_irq(q
->queue_lock
);
1430 /* q->queue_lock is unlocked at this point */
1432 rq
->__sector
= (sector_t
) -1;
1433 rq
->bio
= rq
->biotail
= NULL
;
1438 * blk_get_request_flags - allocate a request
1439 * @q: request queue to allocate a request for
1440 * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC.
1441 * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
1443 struct request
*blk_get_request_flags(struct request_queue
*q
, unsigned int op
,
1446 struct request
*req
;
1448 WARN_ON_ONCE(op
& REQ_NOWAIT
);
1449 WARN_ON_ONCE(flags
& ~(BLK_MQ_REQ_NOWAIT
| BLK_MQ_REQ_PREEMPT
));
1452 req
= blk_mq_alloc_request(q
, op
, flags
);
1453 if (!IS_ERR(req
) && q
->mq_ops
->initialize_rq_fn
)
1454 q
->mq_ops
->initialize_rq_fn(req
);
1456 req
= blk_old_get_request(q
, op
, flags
);
1457 if (!IS_ERR(req
) && q
->initialize_rq_fn
)
1458 q
->initialize_rq_fn(req
);
1463 EXPORT_SYMBOL(blk_get_request_flags
);
1465 struct request
*blk_get_request(struct request_queue
*q
, unsigned int op
,
1468 return blk_get_request_flags(q
, op
, gfp_mask
& __GFP_DIRECT_RECLAIM
?
1469 0 : BLK_MQ_REQ_NOWAIT
);
1471 EXPORT_SYMBOL(blk_get_request
);
1474 * blk_requeue_request - put a request back on queue
1475 * @q: request queue where request should be inserted
1476 * @rq: request to be inserted
1479 * Drivers often keep queueing requests until the hardware cannot accept
1480 * more, when that condition happens we need to put the request back
1481 * on the queue. Must be called with queue lock held.
1483 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
1485 lockdep_assert_held(q
->queue_lock
);
1486 WARN_ON_ONCE(q
->mq_ops
);
1488 blk_delete_timer(rq
);
1489 blk_clear_rq_complete(rq
);
1490 trace_block_rq_requeue(q
, rq
);
1491 wbt_requeue(q
->rq_wb
, &rq
->issue_stat
);
1493 if (rq
->rq_flags
& RQF_QUEUED
)
1494 blk_queue_end_tag(q
, rq
);
1496 BUG_ON(blk_queued_rq(rq
));
1498 elv_requeue_request(q
, rq
);
1500 EXPORT_SYMBOL(blk_requeue_request
);
1502 static void add_acct_request(struct request_queue
*q
, struct request
*rq
,
1505 blk_account_io_start(rq
, true);
1506 __elv_add_request(q
, rq
, where
);
1509 static void part_round_stats_single(struct request_queue
*q
, int cpu
,
1510 struct hd_struct
*part
, unsigned long now
,
1511 unsigned int inflight
)
1514 __part_stat_add(cpu
, part
, time_in_queue
,
1515 inflight
* (now
- part
->stamp
));
1516 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
1522 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1523 * @q: target block queue
1524 * @cpu: cpu number for stats access
1525 * @part: target partition
1527 * The average IO queue length and utilisation statistics are maintained
1528 * by observing the current state of the queue length and the amount of
1529 * time it has been in this state for.
1531 * Normally, that accounting is done on IO completion, but that can result
1532 * in more than a second's worth of IO being accounted for within any one
1533 * second, leading to >100% utilisation. To deal with that, we call this
1534 * function to do a round-off before returning the results when reading
1535 * /proc/diskstats. This accounts immediately for all queue usage up to
1536 * the current jiffies and restarts the counters again.
1538 void part_round_stats(struct request_queue
*q
, int cpu
, struct hd_struct
*part
)
1540 struct hd_struct
*part2
= NULL
;
1541 unsigned long now
= jiffies
;
1542 unsigned int inflight
[2];
1545 if (part
->stamp
!= now
)
1549 part2
= &part_to_disk(part
)->part0
;
1550 if (part2
->stamp
!= now
)
1557 part_in_flight(q
, part
, inflight
);
1560 part_round_stats_single(q
, cpu
, part2
, now
, inflight
[1]);
1562 part_round_stats_single(q
, cpu
, part
, now
, inflight
[0]);
1564 EXPORT_SYMBOL_GPL(part_round_stats
);
1567 static void blk_pm_put_request(struct request
*rq
)
1569 if (rq
->q
->dev
&& !(rq
->rq_flags
& RQF_PM
) && !--rq
->q
->nr_pending
)
1570 pm_runtime_mark_last_busy(rq
->q
->dev
);
1573 static inline void blk_pm_put_request(struct request
*rq
) {}
1576 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1578 req_flags_t rq_flags
= req
->rq_flags
;
1584 blk_mq_free_request(req
);
1588 lockdep_assert_held(q
->queue_lock
);
1590 blk_pm_put_request(req
);
1592 elv_completed_request(q
, req
);
1594 /* this is a bio leak */
1595 WARN_ON(req
->bio
!= NULL
);
1597 wbt_done(q
->rq_wb
, &req
->issue_stat
);
1600 * Request may not have originated from ll_rw_blk. if not,
1601 * it didn't come out of our reserved rq pools
1603 if (rq_flags
& RQF_ALLOCED
) {
1604 struct request_list
*rl
= blk_rq_rl(req
);
1605 bool sync
= op_is_sync(req
->cmd_flags
);
1607 BUG_ON(!list_empty(&req
->queuelist
));
1608 BUG_ON(ELV_ON_HASH(req
));
1610 blk_free_request(rl
, req
);
1611 freed_request(rl
, sync
, rq_flags
);
1616 EXPORT_SYMBOL_GPL(__blk_put_request
);
1618 void blk_put_request(struct request
*req
)
1620 struct request_queue
*q
= req
->q
;
1623 blk_mq_free_request(req
);
1625 unsigned long flags
;
1627 spin_lock_irqsave(q
->queue_lock
, flags
);
1628 __blk_put_request(q
, req
);
1629 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1632 EXPORT_SYMBOL(blk_put_request
);
1634 bool bio_attempt_back_merge(struct request_queue
*q
, struct request
*req
,
1637 const int ff
= bio
->bi_opf
& REQ_FAILFAST_MASK
;
1639 if (!ll_back_merge_fn(q
, req
, bio
))
1642 trace_block_bio_backmerge(q
, req
, bio
);
1644 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1645 blk_rq_set_mixed_merge(req
);
1647 req
->biotail
->bi_next
= bio
;
1649 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1650 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1652 blk_account_io_start(req
, false);
1656 bool bio_attempt_front_merge(struct request_queue
*q
, struct request
*req
,
1659 const int ff
= bio
->bi_opf
& REQ_FAILFAST_MASK
;
1661 if (!ll_front_merge_fn(q
, req
, bio
))
1664 trace_block_bio_frontmerge(q
, req
, bio
);
1666 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1667 blk_rq_set_mixed_merge(req
);
1669 bio
->bi_next
= req
->bio
;
1672 req
->__sector
= bio
->bi_iter
.bi_sector
;
1673 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1674 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1676 blk_account_io_start(req
, false);
1680 bool bio_attempt_discard_merge(struct request_queue
*q
, struct request
*req
,
1683 unsigned short segments
= blk_rq_nr_discard_segments(req
);
1685 if (segments
>= queue_max_discard_segments(q
))
1687 if (blk_rq_sectors(req
) + bio_sectors(bio
) >
1688 blk_rq_get_max_sectors(req
, blk_rq_pos(req
)))
1691 req
->biotail
->bi_next
= bio
;
1693 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1694 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1695 req
->nr_phys_segments
= segments
+ 1;
1697 blk_account_io_start(req
, false);
1700 req_set_nomerge(q
, req
);
1705 * blk_attempt_plug_merge - try to merge with %current's plugged list
1706 * @q: request_queue new bio is being queued at
1707 * @bio: new bio being queued
1708 * @request_count: out parameter for number of traversed plugged requests
1709 * @same_queue_rq: pointer to &struct request that gets filled in when
1710 * another request associated with @q is found on the plug list
1711 * (optional, may be %NULL)
1713 * Determine whether @bio being queued on @q can be merged with a request
1714 * on %current's plugged list. Returns %true if merge was successful,
1717 * Plugging coalesces IOs from the same issuer for the same purpose without
1718 * going through @q->queue_lock. As such it's more of an issuing mechanism
1719 * than scheduling, and the request, while may have elvpriv data, is not
1720 * added on the elevator at this point. In addition, we don't have
1721 * reliable access to the elevator outside queue lock. Only check basic
1722 * merging parameters without querying the elevator.
1724 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1726 bool blk_attempt_plug_merge(struct request_queue
*q
, struct bio
*bio
,
1727 unsigned int *request_count
,
1728 struct request
**same_queue_rq
)
1730 struct blk_plug
*plug
;
1732 struct list_head
*plug_list
;
1734 plug
= current
->plug
;
1740 plug_list
= &plug
->mq_list
;
1742 plug_list
= &plug
->list
;
1744 list_for_each_entry_reverse(rq
, plug_list
, queuelist
) {
1745 bool merged
= false;
1750 * Only blk-mq multiple hardware queues case checks the
1751 * rq in the same queue, there should be only one such
1755 *same_queue_rq
= rq
;
1758 if (rq
->q
!= q
|| !blk_rq_merge_ok(rq
, bio
))
1761 switch (blk_try_merge(rq
, bio
)) {
1762 case ELEVATOR_BACK_MERGE
:
1763 merged
= bio_attempt_back_merge(q
, rq
, bio
);
1765 case ELEVATOR_FRONT_MERGE
:
1766 merged
= bio_attempt_front_merge(q
, rq
, bio
);
1768 case ELEVATOR_DISCARD_MERGE
:
1769 merged
= bio_attempt_discard_merge(q
, rq
, bio
);
1782 unsigned int blk_plug_queued_count(struct request_queue
*q
)
1784 struct blk_plug
*plug
;
1786 struct list_head
*plug_list
;
1787 unsigned int ret
= 0;
1789 plug
= current
->plug
;
1794 plug_list
= &plug
->mq_list
;
1796 plug_list
= &plug
->list
;
1798 list_for_each_entry(rq
, plug_list
, queuelist
) {
1806 void blk_init_request_from_bio(struct request
*req
, struct bio
*bio
)
1808 struct io_context
*ioc
= rq_ioc(bio
);
1810 if (bio
->bi_opf
& REQ_RAHEAD
)
1811 req
->cmd_flags
|= REQ_FAILFAST_MASK
;
1813 req
->__sector
= bio
->bi_iter
.bi_sector
;
1814 if (ioprio_valid(bio_prio(bio
)))
1815 req
->ioprio
= bio_prio(bio
);
1817 req
->ioprio
= ioc
->ioprio
;
1819 req
->ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE
, 0);
1820 req
->write_hint
= bio
->bi_write_hint
;
1821 blk_rq_bio_prep(req
->q
, req
, bio
);
1823 EXPORT_SYMBOL_GPL(blk_init_request_from_bio
);
1825 static blk_qc_t
blk_queue_bio(struct request_queue
*q
, struct bio
*bio
)
1827 struct blk_plug
*plug
;
1828 int where
= ELEVATOR_INSERT_SORT
;
1829 struct request
*req
, *free
;
1830 unsigned int request_count
= 0;
1831 unsigned int wb_acct
;
1834 * low level driver can indicate that it wants pages above a
1835 * certain limit bounced to low memory (ie for highmem, or even
1836 * ISA dma in theory)
1838 blk_queue_bounce(q
, &bio
);
1840 blk_queue_split(q
, &bio
);
1842 if (!bio_integrity_prep(bio
))
1843 return BLK_QC_T_NONE
;
1845 if (op_is_flush(bio
->bi_opf
)) {
1846 spin_lock_irq(q
->queue_lock
);
1847 where
= ELEVATOR_INSERT_FLUSH
;
1852 * Check if we can merge with the plugged list before grabbing
1855 if (!blk_queue_nomerges(q
)) {
1856 if (blk_attempt_plug_merge(q
, bio
, &request_count
, NULL
))
1857 return BLK_QC_T_NONE
;
1859 request_count
= blk_plug_queued_count(q
);
1861 spin_lock_irq(q
->queue_lock
);
1863 switch (elv_merge(q
, &req
, bio
)) {
1864 case ELEVATOR_BACK_MERGE
:
1865 if (!bio_attempt_back_merge(q
, req
, bio
))
1867 elv_bio_merged(q
, req
, bio
);
1868 free
= attempt_back_merge(q
, req
);
1870 __blk_put_request(q
, free
);
1872 elv_merged_request(q
, req
, ELEVATOR_BACK_MERGE
);
1874 case ELEVATOR_FRONT_MERGE
:
1875 if (!bio_attempt_front_merge(q
, req
, bio
))
1877 elv_bio_merged(q
, req
, bio
);
1878 free
= attempt_front_merge(q
, req
);
1880 __blk_put_request(q
, free
);
1882 elv_merged_request(q
, req
, ELEVATOR_FRONT_MERGE
);
1889 wb_acct
= wbt_wait(q
->rq_wb
, bio
, q
->queue_lock
);
1892 * Grab a free request. This is might sleep but can not fail.
1893 * Returns with the queue unlocked.
1895 blk_queue_enter_live(q
);
1896 req
= get_request(q
, bio
->bi_opf
, bio
, 0);
1899 __wbt_done(q
->rq_wb
, wb_acct
);
1900 if (PTR_ERR(req
) == -ENOMEM
)
1901 bio
->bi_status
= BLK_STS_RESOURCE
;
1903 bio
->bi_status
= BLK_STS_IOERR
;
1908 wbt_track(&req
->issue_stat
, wb_acct
);
1911 * After dropping the lock and possibly sleeping here, our request
1912 * may now be mergeable after it had proven unmergeable (above).
1913 * We don't worry about that case for efficiency. It won't happen
1914 * often, and the elevators are able to handle it.
1916 blk_init_request_from_bio(req
, bio
);
1918 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
))
1919 req
->cpu
= raw_smp_processor_id();
1921 plug
= current
->plug
;
1924 * If this is the first request added after a plug, fire
1927 * @request_count may become stale because of schedule
1928 * out, so check plug list again.
1930 if (!request_count
|| list_empty(&plug
->list
))
1931 trace_block_plug(q
);
1933 struct request
*last
= list_entry_rq(plug
->list
.prev
);
1934 if (request_count
>= BLK_MAX_REQUEST_COUNT
||
1935 blk_rq_bytes(last
) >= BLK_PLUG_FLUSH_SIZE
) {
1936 blk_flush_plug_list(plug
, false);
1937 trace_block_plug(q
);
1940 list_add_tail(&req
->queuelist
, &plug
->list
);
1941 blk_account_io_start(req
, true);
1943 spin_lock_irq(q
->queue_lock
);
1944 add_acct_request(q
, req
, where
);
1947 spin_unlock_irq(q
->queue_lock
);
1950 return BLK_QC_T_NONE
;
1953 static void handle_bad_sector(struct bio
*bio
)
1955 char b
[BDEVNAME_SIZE
];
1957 printk(KERN_INFO
"attempt to access beyond end of device\n");
1958 printk(KERN_INFO
"%s: rw=%d, want=%Lu, limit=%Lu\n",
1959 bio_devname(bio
, b
), bio
->bi_opf
,
1960 (unsigned long long)bio_end_sector(bio
),
1961 (long long)get_capacity(bio
->bi_disk
));
1964 #ifdef CONFIG_FAIL_MAKE_REQUEST
1966 static DECLARE_FAULT_ATTR(fail_make_request
);
1968 static int __init
setup_fail_make_request(char *str
)
1970 return setup_fault_attr(&fail_make_request
, str
);
1972 __setup("fail_make_request=", setup_fail_make_request
);
1974 static bool should_fail_request(struct hd_struct
*part
, unsigned int bytes
)
1976 return part
->make_it_fail
&& should_fail(&fail_make_request
, bytes
);
1979 static int __init
fail_make_request_debugfs(void)
1981 struct dentry
*dir
= fault_create_debugfs_attr("fail_make_request",
1982 NULL
, &fail_make_request
);
1984 return PTR_ERR_OR_ZERO(dir
);
1987 late_initcall(fail_make_request_debugfs
);
1989 #else /* CONFIG_FAIL_MAKE_REQUEST */
1991 static inline bool should_fail_request(struct hd_struct
*part
,
1997 #endif /* CONFIG_FAIL_MAKE_REQUEST */
2000 * Remap block n of partition p to block n+start(p) of the disk.
2002 static inline int blk_partition_remap(struct bio
*bio
)
2004 struct hd_struct
*p
;
2008 * Zone reset does not include bi_size so bio_sectors() is always 0.
2009 * Include a test for the reset op code and perform the remap if needed.
2011 if (!bio
->bi_partno
||
2012 (!bio_sectors(bio
) && bio_op(bio
) != REQ_OP_ZONE_RESET
))
2016 p
= __disk_get_part(bio
->bi_disk
, bio
->bi_partno
);
2017 if (likely(p
&& !should_fail_request(p
, bio
->bi_iter
.bi_size
))) {
2018 bio
->bi_iter
.bi_sector
+= p
->start_sect
;
2020 trace_block_bio_remap(bio
->bi_disk
->queue
, bio
, part_devt(p
),
2021 bio
->bi_iter
.bi_sector
- p
->start_sect
);
2023 printk("%s: fail for partition %d\n", __func__
, bio
->bi_partno
);
2032 * Check whether this bio extends beyond the end of the device.
2034 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
2041 /* Test device or partition size, when known. */
2042 maxsector
= get_capacity(bio
->bi_disk
);
2044 sector_t sector
= bio
->bi_iter
.bi_sector
;
2046 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
2048 * This may well happen - the kernel calls bread()
2049 * without checking the size of the device, e.g., when
2050 * mounting a device.
2052 handle_bad_sector(bio
);
2060 static noinline_for_stack
bool
2061 generic_make_request_checks(struct bio
*bio
)
2063 struct request_queue
*q
;
2064 int nr_sectors
= bio_sectors(bio
);
2065 blk_status_t status
= BLK_STS_IOERR
;
2066 char b
[BDEVNAME_SIZE
];
2070 if (bio_check_eod(bio
, nr_sectors
))
2073 q
= bio
->bi_disk
->queue
;
2076 "generic_make_request: Trying to access "
2077 "nonexistent block-device %s (%Lu)\n",
2078 bio_devname(bio
, b
), (long long)bio
->bi_iter
.bi_sector
);
2083 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
2084 * if queue is not a request based queue.
2087 if ((bio
->bi_opf
& REQ_NOWAIT
) && !queue_is_rq_based(q
))
2090 if (should_fail_request(&bio
->bi_disk
->part0
, bio
->bi_iter
.bi_size
))
2093 if (blk_partition_remap(bio
))
2096 if (bio_check_eod(bio
, nr_sectors
))
2100 * Filter flush bio's early so that make_request based
2101 * drivers without flush support don't have to worry
2104 if (op_is_flush(bio
->bi_opf
) &&
2105 !test_bit(QUEUE_FLAG_WC
, &q
->queue_flags
)) {
2106 bio
->bi_opf
&= ~(REQ_PREFLUSH
| REQ_FUA
);
2108 status
= BLK_STS_OK
;
2113 switch (bio_op(bio
)) {
2114 case REQ_OP_DISCARD
:
2115 if (!blk_queue_discard(q
))
2118 case REQ_OP_SECURE_ERASE
:
2119 if (!blk_queue_secure_erase(q
))
2122 case REQ_OP_WRITE_SAME
:
2123 if (!q
->limits
.max_write_same_sectors
)
2126 case REQ_OP_ZONE_REPORT
:
2127 case REQ_OP_ZONE_RESET
:
2128 if (!blk_queue_is_zoned(q
))
2131 case REQ_OP_WRITE_ZEROES
:
2132 if (!q
->limits
.max_write_zeroes_sectors
)
2140 * Various block parts want %current->io_context and lazy ioc
2141 * allocation ends up trading a lot of pain for a small amount of
2142 * memory. Just allocate it upfront. This may fail and block
2143 * layer knows how to live with it.
2145 create_io_context(GFP_ATOMIC
, q
->node
);
2147 if (!blkcg_bio_issue_check(q
, bio
))
2150 if (!bio_flagged(bio
, BIO_TRACE_COMPLETION
)) {
2151 trace_block_bio_queue(q
, bio
);
2152 /* Now that enqueuing has been traced, we need to trace
2153 * completion as well.
2155 bio_set_flag(bio
, BIO_TRACE_COMPLETION
);
2160 status
= BLK_STS_NOTSUPP
;
2162 bio
->bi_status
= status
;
2168 * generic_make_request - hand a buffer to its device driver for I/O
2169 * @bio: The bio describing the location in memory and on the device.
2171 * generic_make_request() is used to make I/O requests of block
2172 * devices. It is passed a &struct bio, which describes the I/O that needs
2175 * generic_make_request() does not return any status. The
2176 * success/failure status of the request, along with notification of
2177 * completion, is delivered asynchronously through the bio->bi_end_io
2178 * function described (one day) else where.
2180 * The caller of generic_make_request must make sure that bi_io_vec
2181 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2182 * set to describe the device address, and the
2183 * bi_end_io and optionally bi_private are set to describe how
2184 * completion notification should be signaled.
2186 * generic_make_request and the drivers it calls may use bi_next if this
2187 * bio happens to be merged with someone else, and may resubmit the bio to
2188 * a lower device by calling into generic_make_request recursively, which
2189 * means the bio should NOT be touched after the call to ->make_request_fn.
2191 blk_qc_t
generic_make_request(struct bio
*bio
)
2194 * bio_list_on_stack[0] contains bios submitted by the current
2196 * bio_list_on_stack[1] contains bios that were submitted before
2197 * the current make_request_fn, but that haven't been processed
2200 struct bio_list bio_list_on_stack
[2];
2201 blk_qc_t ret
= BLK_QC_T_NONE
;
2203 if (!generic_make_request_checks(bio
))
2207 * We only want one ->make_request_fn to be active at a time, else
2208 * stack usage with stacked devices could be a problem. So use
2209 * current->bio_list to keep a list of requests submited by a
2210 * make_request_fn function. current->bio_list is also used as a
2211 * flag to say if generic_make_request is currently active in this
2212 * task or not. If it is NULL, then no make_request is active. If
2213 * it is non-NULL, then a make_request is active, and new requests
2214 * should be added at the tail
2216 if (current
->bio_list
) {
2217 bio_list_add(¤t
->bio_list
[0], bio
);
2221 /* following loop may be a bit non-obvious, and so deserves some
2223 * Before entering the loop, bio->bi_next is NULL (as all callers
2224 * ensure that) so we have a list with a single bio.
2225 * We pretend that we have just taken it off a longer list, so
2226 * we assign bio_list to a pointer to the bio_list_on_stack,
2227 * thus initialising the bio_list of new bios to be
2228 * added. ->make_request() may indeed add some more bios
2229 * through a recursive call to generic_make_request. If it
2230 * did, we find a non-NULL value in bio_list and re-enter the loop
2231 * from the top. In this case we really did just take the bio
2232 * of the top of the list (no pretending) and so remove it from
2233 * bio_list, and call into ->make_request() again.
2235 BUG_ON(bio
->bi_next
);
2236 bio_list_init(&bio_list_on_stack
[0]);
2237 current
->bio_list
= bio_list_on_stack
;
2239 struct request_queue
*q
= bio
->bi_disk
->queue
;
2241 if (likely(blk_queue_enter(q
, bio
->bi_opf
& REQ_NOWAIT
) == 0)) {
2242 struct bio_list lower
, same
;
2244 /* Create a fresh bio_list for all subordinate requests */
2245 bio_list_on_stack
[1] = bio_list_on_stack
[0];
2246 bio_list_init(&bio_list_on_stack
[0]);
2247 ret
= q
->make_request_fn(q
, bio
);
2251 /* sort new bios into those for a lower level
2252 * and those for the same level
2254 bio_list_init(&lower
);
2255 bio_list_init(&same
);
2256 while ((bio
= bio_list_pop(&bio_list_on_stack
[0])) != NULL
)
2257 if (q
== bio
->bi_disk
->queue
)
2258 bio_list_add(&same
, bio
);
2260 bio_list_add(&lower
, bio
);
2261 /* now assemble so we handle the lowest level first */
2262 bio_list_merge(&bio_list_on_stack
[0], &lower
);
2263 bio_list_merge(&bio_list_on_stack
[0], &same
);
2264 bio_list_merge(&bio_list_on_stack
[0], &bio_list_on_stack
[1]);
2266 if (unlikely(!blk_queue_dying(q
) &&
2267 (bio
->bi_opf
& REQ_NOWAIT
)))
2268 bio_wouldblock_error(bio
);
2272 bio
= bio_list_pop(&bio_list_on_stack
[0]);
2274 current
->bio_list
= NULL
; /* deactivate */
2279 EXPORT_SYMBOL(generic_make_request
);
2282 * direct_make_request - hand a buffer directly to its device driver for I/O
2283 * @bio: The bio describing the location in memory and on the device.
2285 * This function behaves like generic_make_request(), but does not protect
2286 * against recursion. Must only be used if the called driver is known
2287 * to not call generic_make_request (or direct_make_request) again from
2288 * its make_request function. (Calling direct_make_request again from
2289 * a workqueue is perfectly fine as that doesn't recurse).
2291 blk_qc_t
direct_make_request(struct bio
*bio
)
2293 struct request_queue
*q
= bio
->bi_disk
->queue
;
2294 bool nowait
= bio
->bi_opf
& REQ_NOWAIT
;
2297 if (!generic_make_request_checks(bio
))
2298 return BLK_QC_T_NONE
;
2300 if (unlikely(blk_queue_enter(q
, nowait
))) {
2301 if (nowait
&& !blk_queue_dying(q
))
2302 bio
->bi_status
= BLK_STS_AGAIN
;
2304 bio
->bi_status
= BLK_STS_IOERR
;
2306 return BLK_QC_T_NONE
;
2309 ret
= q
->make_request_fn(q
, bio
);
2313 EXPORT_SYMBOL_GPL(direct_make_request
);
2316 * submit_bio - submit a bio to the block device layer for I/O
2317 * @bio: The &struct bio which describes the I/O
2319 * submit_bio() is very similar in purpose to generic_make_request(), and
2320 * uses that function to do most of the work. Both are fairly rough
2321 * interfaces; @bio must be presetup and ready for I/O.
2324 blk_qc_t
submit_bio(struct bio
*bio
)
2327 * If it's a regular read/write or a barrier with data attached,
2328 * go through the normal accounting stuff before submission.
2330 if (bio_has_data(bio
)) {
2333 if (unlikely(bio_op(bio
) == REQ_OP_WRITE_SAME
))
2334 count
= queue_logical_block_size(bio
->bi_disk
->queue
);
2336 count
= bio_sectors(bio
);
2338 if (op_is_write(bio_op(bio
))) {
2339 count_vm_events(PGPGOUT
, count
);
2341 task_io_account_read(bio
->bi_iter
.bi_size
);
2342 count_vm_events(PGPGIN
, count
);
2345 if (unlikely(block_dump
)) {
2346 char b
[BDEVNAME_SIZE
];
2347 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s (%u sectors)\n",
2348 current
->comm
, task_pid_nr(current
),
2349 op_is_write(bio_op(bio
)) ? "WRITE" : "READ",
2350 (unsigned long long)bio
->bi_iter
.bi_sector
,
2351 bio_devname(bio
, b
), count
);
2355 return generic_make_request(bio
);
2357 EXPORT_SYMBOL(submit_bio
);
2359 bool blk_poll(struct request_queue
*q
, blk_qc_t cookie
)
2361 if (!q
->poll_fn
|| !blk_qc_t_valid(cookie
))
2365 blk_flush_plug_list(current
->plug
, false);
2366 return q
->poll_fn(q
, cookie
);
2368 EXPORT_SYMBOL_GPL(blk_poll
);
2371 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2372 * for new the queue limits
2374 * @rq: the request being checked
2377 * @rq may have been made based on weaker limitations of upper-level queues
2378 * in request stacking drivers, and it may violate the limitation of @q.
2379 * Since the block layer and the underlying device driver trust @rq
2380 * after it is inserted to @q, it should be checked against @q before
2381 * the insertion using this generic function.
2383 * Request stacking drivers like request-based dm may change the queue
2384 * limits when retrying requests on other queues. Those requests need
2385 * to be checked against the new queue limits again during dispatch.
2387 static int blk_cloned_rq_check_limits(struct request_queue
*q
,
2390 if (blk_rq_sectors(rq
) > blk_queue_get_max_sectors(q
, req_op(rq
))) {
2391 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
2396 * queue's settings related to segment counting like q->bounce_pfn
2397 * may differ from that of other stacking queues.
2398 * Recalculate it to check the request correctly on this queue's
2401 blk_recalc_rq_segments(rq
);
2402 if (rq
->nr_phys_segments
> queue_max_segments(q
)) {
2403 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
2411 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2412 * @q: the queue to submit the request
2413 * @rq: the request being queued
2415 blk_status_t
blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
2417 unsigned long flags
;
2418 int where
= ELEVATOR_INSERT_BACK
;
2420 if (blk_cloned_rq_check_limits(q
, rq
))
2421 return BLK_STS_IOERR
;
2424 should_fail_request(&rq
->rq_disk
->part0
, blk_rq_bytes(rq
)))
2425 return BLK_STS_IOERR
;
2428 if (blk_queue_io_stat(q
))
2429 blk_account_io_start(rq
, true);
2431 * Since we have a scheduler attached on the top device,
2432 * bypass a potential scheduler on the bottom device for
2435 blk_mq_request_bypass_insert(rq
, true);
2439 spin_lock_irqsave(q
->queue_lock
, flags
);
2440 if (unlikely(blk_queue_dying(q
))) {
2441 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2442 return BLK_STS_IOERR
;
2446 * Submitting request must be dequeued before calling this function
2447 * because it will be linked to another request_queue
2449 BUG_ON(blk_queued_rq(rq
));
2451 if (op_is_flush(rq
->cmd_flags
))
2452 where
= ELEVATOR_INSERT_FLUSH
;
2454 add_acct_request(q
, rq
, where
);
2455 if (where
== ELEVATOR_INSERT_FLUSH
)
2457 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2461 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
2464 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2465 * @rq: request to examine
2468 * A request could be merge of IOs which require different failure
2469 * handling. This function determines the number of bytes which
2470 * can be failed from the beginning of the request without
2471 * crossing into area which need to be retried further.
2474 * The number of bytes to fail.
2476 unsigned int blk_rq_err_bytes(const struct request
*rq
)
2478 unsigned int ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
2479 unsigned int bytes
= 0;
2482 if (!(rq
->rq_flags
& RQF_MIXED_MERGE
))
2483 return blk_rq_bytes(rq
);
2486 * Currently the only 'mixing' which can happen is between
2487 * different fastfail types. We can safely fail portions
2488 * which have all the failfast bits that the first one has -
2489 * the ones which are at least as eager to fail as the first
2492 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
2493 if ((bio
->bi_opf
& ff
) != ff
)
2495 bytes
+= bio
->bi_iter
.bi_size
;
2498 /* this could lead to infinite loop */
2499 BUG_ON(blk_rq_bytes(rq
) && !bytes
);
2502 EXPORT_SYMBOL_GPL(blk_rq_err_bytes
);
2504 void blk_account_io_completion(struct request
*req
, unsigned int bytes
)
2506 if (blk_do_io_stat(req
)) {
2507 const int rw
= rq_data_dir(req
);
2508 struct hd_struct
*part
;
2511 cpu
= part_stat_lock();
2513 part_stat_add(cpu
, part
, sectors
[rw
], bytes
>> 9);
2518 void blk_account_io_done(struct request
*req
)
2521 * Account IO completion. flush_rq isn't accounted as a
2522 * normal IO on queueing nor completion. Accounting the
2523 * containing request is enough.
2525 if (blk_do_io_stat(req
) && !(req
->rq_flags
& RQF_FLUSH_SEQ
)) {
2526 unsigned long duration
= jiffies
- req
->start_time
;
2527 const int rw
= rq_data_dir(req
);
2528 struct hd_struct
*part
;
2531 cpu
= part_stat_lock();
2534 part_stat_inc(cpu
, part
, ios
[rw
]);
2535 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
2536 part_round_stats(req
->q
, cpu
, part
);
2537 part_dec_in_flight(req
->q
, part
, rw
);
2539 hd_struct_put(part
);
2546 * Don't process normal requests when queue is suspended
2547 * or in the process of suspending/resuming
2549 static bool blk_pm_allow_request(struct request
*rq
)
2551 switch (rq
->q
->rpm_status
) {
2553 case RPM_SUSPENDING
:
2554 return rq
->rq_flags
& RQF_PM
;
2562 static bool blk_pm_allow_request(struct request
*rq
)
2568 void blk_account_io_start(struct request
*rq
, bool new_io
)
2570 struct hd_struct
*part
;
2571 int rw
= rq_data_dir(rq
);
2574 if (!blk_do_io_stat(rq
))
2577 cpu
= part_stat_lock();
2581 part_stat_inc(cpu
, part
, merges
[rw
]);
2583 part
= disk_map_sector_rcu(rq
->rq_disk
, blk_rq_pos(rq
));
2584 if (!hd_struct_try_get(part
)) {
2586 * The partition is already being removed,
2587 * the request will be accounted on the disk only
2589 * We take a reference on disk->part0 although that
2590 * partition will never be deleted, so we can treat
2591 * it as any other partition.
2593 part
= &rq
->rq_disk
->part0
;
2594 hd_struct_get(part
);
2596 part_round_stats(rq
->q
, cpu
, part
);
2597 part_inc_in_flight(rq
->q
, part
, rw
);
2604 static struct request
*elv_next_request(struct request_queue
*q
)
2607 struct blk_flush_queue
*fq
= blk_get_flush_queue(q
, NULL
);
2609 WARN_ON_ONCE(q
->mq_ops
);
2612 list_for_each_entry(rq
, &q
->queue_head
, queuelist
) {
2613 if (blk_pm_allow_request(rq
))
2616 if (rq
->rq_flags
& RQF_SOFTBARRIER
)
2621 * Flush request is running and flush request isn't queueable
2622 * in the drive, we can hold the queue till flush request is
2623 * finished. Even we don't do this, driver can't dispatch next
2624 * requests and will requeue them. And this can improve
2625 * throughput too. For example, we have request flush1, write1,
2626 * flush 2. flush1 is dispatched, then queue is hold, write1
2627 * isn't inserted to queue. After flush1 is finished, flush2
2628 * will be dispatched. Since disk cache is already clean,
2629 * flush2 will be finished very soon, so looks like flush2 is
2631 * Since the queue is hold, a flag is set to indicate the queue
2632 * should be restarted later. Please see flush_end_io() for
2635 if (fq
->flush_pending_idx
!= fq
->flush_running_idx
&&
2636 !queue_flush_queueable(q
)) {
2637 fq
->flush_queue_delayed
= 1;
2640 if (unlikely(blk_queue_bypass(q
)) ||
2641 !q
->elevator
->type
->ops
.sq
.elevator_dispatch_fn(q
, 0))
2647 * blk_peek_request - peek at the top of a request queue
2648 * @q: request queue to peek at
2651 * Return the request at the top of @q. The returned request
2652 * should be started using blk_start_request() before LLD starts
2656 * Pointer to the request at the top of @q if available. Null
2659 struct request
*blk_peek_request(struct request_queue
*q
)
2664 lockdep_assert_held(q
->queue_lock
);
2665 WARN_ON_ONCE(q
->mq_ops
);
2667 while ((rq
= elv_next_request(q
)) != NULL
) {
2668 if (!(rq
->rq_flags
& RQF_STARTED
)) {
2670 * This is the first time the device driver
2671 * sees this request (possibly after
2672 * requeueing). Notify IO scheduler.
2674 if (rq
->rq_flags
& RQF_SORTED
)
2675 elv_activate_rq(q
, rq
);
2678 * just mark as started even if we don't start
2679 * it, a request that has been delayed should
2680 * not be passed by new incoming requests
2682 rq
->rq_flags
|= RQF_STARTED
;
2683 trace_block_rq_issue(q
, rq
);
2686 if (!q
->boundary_rq
|| q
->boundary_rq
== rq
) {
2687 q
->end_sector
= rq_end_sector(rq
);
2688 q
->boundary_rq
= NULL
;
2691 if (rq
->rq_flags
& RQF_DONTPREP
)
2694 if (q
->dma_drain_size
&& blk_rq_bytes(rq
)) {
2696 * make sure space for the drain appears we
2697 * know we can do this because max_hw_segments
2698 * has been adjusted to be one fewer than the
2701 rq
->nr_phys_segments
++;
2707 ret
= q
->prep_rq_fn(q
, rq
);
2708 if (ret
== BLKPREP_OK
) {
2710 } else if (ret
== BLKPREP_DEFER
) {
2712 * the request may have been (partially) prepped.
2713 * we need to keep this request in the front to
2714 * avoid resource deadlock. RQF_STARTED will
2715 * prevent other fs requests from passing this one.
2717 if (q
->dma_drain_size
&& blk_rq_bytes(rq
) &&
2718 !(rq
->rq_flags
& RQF_DONTPREP
)) {
2720 * remove the space for the drain we added
2721 * so that we don't add it again
2723 --rq
->nr_phys_segments
;
2728 } else if (ret
== BLKPREP_KILL
|| ret
== BLKPREP_INVALID
) {
2729 rq
->rq_flags
|= RQF_QUIET
;
2731 * Mark this request as started so we don't trigger
2732 * any debug logic in the end I/O path.
2734 blk_start_request(rq
);
2735 __blk_end_request_all(rq
, ret
== BLKPREP_INVALID
?
2736 BLK_STS_TARGET
: BLK_STS_IOERR
);
2738 printk(KERN_ERR
"%s: bad return=%d\n", __func__
, ret
);
2745 EXPORT_SYMBOL(blk_peek_request
);
2747 static void blk_dequeue_request(struct request
*rq
)
2749 struct request_queue
*q
= rq
->q
;
2751 BUG_ON(list_empty(&rq
->queuelist
));
2752 BUG_ON(ELV_ON_HASH(rq
));
2754 list_del_init(&rq
->queuelist
);
2757 * the time frame between a request being removed from the lists
2758 * and to it is freed is accounted as io that is in progress at
2761 if (blk_account_rq(rq
)) {
2762 q
->in_flight
[rq_is_sync(rq
)]++;
2763 set_io_start_time_ns(rq
);
2768 * blk_start_request - start request processing on the driver
2769 * @req: request to dequeue
2772 * Dequeue @req and start timeout timer on it. This hands off the
2773 * request to the driver.
2775 void blk_start_request(struct request
*req
)
2777 lockdep_assert_held(req
->q
->queue_lock
);
2778 WARN_ON_ONCE(req
->q
->mq_ops
);
2780 blk_dequeue_request(req
);
2782 if (test_bit(QUEUE_FLAG_STATS
, &req
->q
->queue_flags
)) {
2783 blk_stat_set_issue(&req
->issue_stat
, blk_rq_sectors(req
));
2784 req
->rq_flags
|= RQF_STATS
;
2785 wbt_issue(req
->q
->rq_wb
, &req
->issue_stat
);
2788 BUG_ON(test_bit(REQ_ATOM_COMPLETE
, &req
->atomic_flags
));
2791 EXPORT_SYMBOL(blk_start_request
);
2794 * blk_fetch_request - fetch a request from a request queue
2795 * @q: request queue to fetch a request from
2798 * Return the request at the top of @q. The request is started on
2799 * return and LLD can start processing it immediately.
2802 * Pointer to the request at the top of @q if available. Null
2805 struct request
*blk_fetch_request(struct request_queue
*q
)
2809 lockdep_assert_held(q
->queue_lock
);
2810 WARN_ON_ONCE(q
->mq_ops
);
2812 rq
= blk_peek_request(q
);
2814 blk_start_request(rq
);
2817 EXPORT_SYMBOL(blk_fetch_request
);
2820 * Steal bios from a request and add them to a bio list.
2821 * The request must not have been partially completed before.
2823 void blk_steal_bios(struct bio_list
*list
, struct request
*rq
)
2827 list
->tail
->bi_next
= rq
->bio
;
2829 list
->head
= rq
->bio
;
2830 list
->tail
= rq
->biotail
;
2838 EXPORT_SYMBOL_GPL(blk_steal_bios
);
2841 * blk_update_request - Special helper function for request stacking drivers
2842 * @req: the request being processed
2843 * @error: block status code
2844 * @nr_bytes: number of bytes to complete @req
2847 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2848 * the request structure even if @req doesn't have leftover.
2849 * If @req has leftover, sets it up for the next range of segments.
2851 * This special helper function is only for request stacking drivers
2852 * (e.g. request-based dm) so that they can handle partial completion.
2853 * Actual device drivers should use blk_end_request instead.
2855 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2856 * %false return from this function.
2859 * %false - this request doesn't have any more data
2860 * %true - this request has more data
2862 bool blk_update_request(struct request
*req
, blk_status_t error
,
2863 unsigned int nr_bytes
)
2867 trace_block_rq_complete(req
, blk_status_to_errno(error
), nr_bytes
);
2872 if (unlikely(error
&& !blk_rq_is_passthrough(req
) &&
2873 !(req
->rq_flags
& RQF_QUIET
)))
2874 print_req_error(req
, error
);
2876 blk_account_io_completion(req
, nr_bytes
);
2880 struct bio
*bio
= req
->bio
;
2881 unsigned bio_bytes
= min(bio
->bi_iter
.bi_size
, nr_bytes
);
2883 if (bio_bytes
== bio
->bi_iter
.bi_size
)
2884 req
->bio
= bio
->bi_next
;
2886 /* Completion has already been traced */
2887 bio_clear_flag(bio
, BIO_TRACE_COMPLETION
);
2888 req_bio_endio(req
, bio
, bio_bytes
, error
);
2890 total_bytes
+= bio_bytes
;
2891 nr_bytes
-= bio_bytes
;
2902 * Reset counters so that the request stacking driver
2903 * can find how many bytes remain in the request
2906 req
->__data_len
= 0;
2910 req
->__data_len
-= total_bytes
;
2912 /* update sector only for requests with clear definition of sector */
2913 if (!blk_rq_is_passthrough(req
))
2914 req
->__sector
+= total_bytes
>> 9;
2916 /* mixed attributes always follow the first bio */
2917 if (req
->rq_flags
& RQF_MIXED_MERGE
) {
2918 req
->cmd_flags
&= ~REQ_FAILFAST_MASK
;
2919 req
->cmd_flags
|= req
->bio
->bi_opf
& REQ_FAILFAST_MASK
;
2922 if (!(req
->rq_flags
& RQF_SPECIAL_PAYLOAD
)) {
2924 * If total number of sectors is less than the first segment
2925 * size, something has gone terribly wrong.
2927 if (blk_rq_bytes(req
) < blk_rq_cur_bytes(req
)) {
2928 blk_dump_rq_flags(req
, "request botched");
2929 req
->__data_len
= blk_rq_cur_bytes(req
);
2932 /* recalculate the number of segments */
2933 blk_recalc_rq_segments(req
);
2938 EXPORT_SYMBOL_GPL(blk_update_request
);
2940 static bool blk_update_bidi_request(struct request
*rq
, blk_status_t error
,
2941 unsigned int nr_bytes
,
2942 unsigned int bidi_bytes
)
2944 if (blk_update_request(rq
, error
, nr_bytes
))
2947 /* Bidi request must be completed as a whole */
2948 if (unlikely(blk_bidi_rq(rq
)) &&
2949 blk_update_request(rq
->next_rq
, error
, bidi_bytes
))
2952 if (blk_queue_add_random(rq
->q
))
2953 add_disk_randomness(rq
->rq_disk
);
2959 * blk_unprep_request - unprepare a request
2962 * This function makes a request ready for complete resubmission (or
2963 * completion). It happens only after all error handling is complete,
2964 * so represents the appropriate moment to deallocate any resources
2965 * that were allocated to the request in the prep_rq_fn. The queue
2966 * lock is held when calling this.
2968 void blk_unprep_request(struct request
*req
)
2970 struct request_queue
*q
= req
->q
;
2972 req
->rq_flags
&= ~RQF_DONTPREP
;
2973 if (q
->unprep_rq_fn
)
2974 q
->unprep_rq_fn(q
, req
);
2976 EXPORT_SYMBOL_GPL(blk_unprep_request
);
2978 void blk_finish_request(struct request
*req
, blk_status_t error
)
2980 struct request_queue
*q
= req
->q
;
2982 lockdep_assert_held(req
->q
->queue_lock
);
2983 WARN_ON_ONCE(q
->mq_ops
);
2985 if (req
->rq_flags
& RQF_STATS
)
2988 if (req
->rq_flags
& RQF_QUEUED
)
2989 blk_queue_end_tag(q
, req
);
2991 BUG_ON(blk_queued_rq(req
));
2993 if (unlikely(laptop_mode
) && !blk_rq_is_passthrough(req
))
2994 laptop_io_completion(req
->q
->backing_dev_info
);
2996 blk_delete_timer(req
);
2998 if (req
->rq_flags
& RQF_DONTPREP
)
2999 blk_unprep_request(req
);
3001 blk_account_io_done(req
);
3004 wbt_done(req
->q
->rq_wb
, &req
->issue_stat
);
3005 req
->end_io(req
, error
);
3007 if (blk_bidi_rq(req
))
3008 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
3010 __blk_put_request(q
, req
);
3013 EXPORT_SYMBOL(blk_finish_request
);
3016 * blk_end_bidi_request - Complete a bidi request
3017 * @rq: the request to complete
3018 * @error: block status code
3019 * @nr_bytes: number of bytes to complete @rq
3020 * @bidi_bytes: number of bytes to complete @rq->next_rq
3023 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
3024 * Drivers that supports bidi can safely call this member for any
3025 * type of request, bidi or uni. In the later case @bidi_bytes is
3029 * %false - we are done with this request
3030 * %true - still buffers pending for this request
3032 static bool blk_end_bidi_request(struct request
*rq
, blk_status_t error
,
3033 unsigned int nr_bytes
, unsigned int bidi_bytes
)
3035 struct request_queue
*q
= rq
->q
;
3036 unsigned long flags
;
3038 WARN_ON_ONCE(q
->mq_ops
);
3040 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
3043 spin_lock_irqsave(q
->queue_lock
, flags
);
3044 blk_finish_request(rq
, error
);
3045 spin_unlock_irqrestore(q
->queue_lock
, flags
);
3051 * __blk_end_bidi_request - Complete a bidi request with queue lock held
3052 * @rq: the request to complete
3053 * @error: block status code
3054 * @nr_bytes: number of bytes to complete @rq
3055 * @bidi_bytes: number of bytes to complete @rq->next_rq
3058 * Identical to blk_end_bidi_request() except that queue lock is
3059 * assumed to be locked on entry and remains so on return.
3062 * %false - we are done with this request
3063 * %true - still buffers pending for this request
3065 static bool __blk_end_bidi_request(struct request
*rq
, blk_status_t error
,
3066 unsigned int nr_bytes
, unsigned int bidi_bytes
)
3068 lockdep_assert_held(rq
->q
->queue_lock
);
3069 WARN_ON_ONCE(rq
->q
->mq_ops
);
3071 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
3074 blk_finish_request(rq
, error
);
3080 * blk_end_request - Helper function for drivers to complete the request.
3081 * @rq: the request being processed
3082 * @error: block status code
3083 * @nr_bytes: number of bytes to complete
3086 * Ends I/O on a number of bytes attached to @rq.
3087 * If @rq has leftover, sets it up for the next range of segments.
3090 * %false - we are done with this request
3091 * %true - still buffers pending for this request
3093 bool blk_end_request(struct request
*rq
, blk_status_t error
,
3094 unsigned int nr_bytes
)
3096 WARN_ON_ONCE(rq
->q
->mq_ops
);
3097 return blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
3099 EXPORT_SYMBOL(blk_end_request
);
3102 * blk_end_request_all - Helper function for drives to finish the request.
3103 * @rq: the request to finish
3104 * @error: block status code
3107 * Completely finish @rq.
3109 void blk_end_request_all(struct request
*rq
, blk_status_t error
)
3112 unsigned int bidi_bytes
= 0;
3114 if (unlikely(blk_bidi_rq(rq
)))
3115 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
3117 pending
= blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
3120 EXPORT_SYMBOL(blk_end_request_all
);
3123 * __blk_end_request - Helper function for drivers to complete the request.
3124 * @rq: the request being processed
3125 * @error: block status code
3126 * @nr_bytes: number of bytes to complete
3129 * Must be called with queue lock held unlike blk_end_request().
3132 * %false - we are done with this request
3133 * %true - still buffers pending for this request
3135 bool __blk_end_request(struct request
*rq
, blk_status_t error
,
3136 unsigned int nr_bytes
)
3138 lockdep_assert_held(rq
->q
->queue_lock
);
3139 WARN_ON_ONCE(rq
->q
->mq_ops
);
3141 return __blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
3143 EXPORT_SYMBOL(__blk_end_request
);
3146 * __blk_end_request_all - Helper function for drives to finish the request.
3147 * @rq: the request to finish
3148 * @error: block status code
3151 * Completely finish @rq. Must be called with queue lock held.
3153 void __blk_end_request_all(struct request
*rq
, blk_status_t error
)
3156 unsigned int bidi_bytes
= 0;
3158 lockdep_assert_held(rq
->q
->queue_lock
);
3159 WARN_ON_ONCE(rq
->q
->mq_ops
);
3161 if (unlikely(blk_bidi_rq(rq
)))
3162 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
3164 pending
= __blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
3167 EXPORT_SYMBOL(__blk_end_request_all
);
3170 * __blk_end_request_cur - Helper function to finish the current request chunk.
3171 * @rq: the request to finish the current chunk for
3172 * @error: block status code
3175 * Complete the current consecutively mapped chunk from @rq. Must
3176 * be called with queue lock held.
3179 * %false - we are done with this request
3180 * %true - still buffers pending for this request
3182 bool __blk_end_request_cur(struct request
*rq
, blk_status_t error
)
3184 return __blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
3186 EXPORT_SYMBOL(__blk_end_request_cur
);
3188 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
3191 if (bio_has_data(bio
))
3192 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
3194 rq
->__data_len
= bio
->bi_iter
.bi_size
;
3195 rq
->bio
= rq
->biotail
= bio
;
3198 rq
->rq_disk
= bio
->bi_disk
;
3201 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
3203 * rq_flush_dcache_pages - Helper function to flush all pages in a request
3204 * @rq: the request to be flushed
3207 * Flush all pages in @rq.
3209 void rq_flush_dcache_pages(struct request
*rq
)
3211 struct req_iterator iter
;
3212 struct bio_vec bvec
;
3214 rq_for_each_segment(bvec
, rq
, iter
)
3215 flush_dcache_page(bvec
.bv_page
);
3217 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages
);
3221 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
3222 * @q : the queue of the device being checked
3225 * Check if underlying low-level drivers of a device are busy.
3226 * If the drivers want to export their busy state, they must set own
3227 * exporting function using blk_queue_lld_busy() first.
3229 * Basically, this function is used only by request stacking drivers
3230 * to stop dispatching requests to underlying devices when underlying
3231 * devices are busy. This behavior helps more I/O merging on the queue
3232 * of the request stacking driver and prevents I/O throughput regression
3233 * on burst I/O load.
3236 * 0 - Not busy (The request stacking driver should dispatch request)
3237 * 1 - Busy (The request stacking driver should stop dispatching request)
3239 int blk_lld_busy(struct request_queue
*q
)
3242 return q
->lld_busy_fn(q
);
3246 EXPORT_SYMBOL_GPL(blk_lld_busy
);
3249 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
3250 * @rq: the clone request to be cleaned up
3253 * Free all bios in @rq for a cloned request.
3255 void blk_rq_unprep_clone(struct request
*rq
)
3259 while ((bio
= rq
->bio
) != NULL
) {
3260 rq
->bio
= bio
->bi_next
;
3265 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone
);
3268 * Copy attributes of the original request to the clone request.
3269 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3271 static void __blk_rq_prep_clone(struct request
*dst
, struct request
*src
)
3273 dst
->cpu
= src
->cpu
;
3274 dst
->__sector
= blk_rq_pos(src
);
3275 dst
->__data_len
= blk_rq_bytes(src
);
3276 dst
->nr_phys_segments
= src
->nr_phys_segments
;
3277 dst
->ioprio
= src
->ioprio
;
3278 dst
->extra_len
= src
->extra_len
;
3282 * blk_rq_prep_clone - Helper function to setup clone request
3283 * @rq: the request to be setup
3284 * @rq_src: original request to be cloned
3285 * @bs: bio_set that bios for clone are allocated from
3286 * @gfp_mask: memory allocation mask for bio
3287 * @bio_ctr: setup function to be called for each clone bio.
3288 * Returns %0 for success, non %0 for failure.
3289 * @data: private data to be passed to @bio_ctr
3292 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3293 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3294 * are not copied, and copying such parts is the caller's responsibility.
3295 * Also, pages which the original bios are pointing to are not copied
3296 * and the cloned bios just point same pages.
3297 * So cloned bios must be completed before original bios, which means
3298 * the caller must complete @rq before @rq_src.
3300 int blk_rq_prep_clone(struct request
*rq
, struct request
*rq_src
,
3301 struct bio_set
*bs
, gfp_t gfp_mask
,
3302 int (*bio_ctr
)(struct bio
*, struct bio
*, void *),
3305 struct bio
*bio
, *bio_src
;
3310 __rq_for_each_bio(bio_src
, rq_src
) {
3311 bio
= bio_clone_fast(bio_src
, gfp_mask
, bs
);
3315 if (bio_ctr
&& bio_ctr(bio
, bio_src
, data
))
3319 rq
->biotail
->bi_next
= bio
;
3322 rq
->bio
= rq
->biotail
= bio
;
3325 __blk_rq_prep_clone(rq
, rq_src
);
3332 blk_rq_unprep_clone(rq
);
3336 EXPORT_SYMBOL_GPL(blk_rq_prep_clone
);
3338 int kblockd_schedule_work(struct work_struct
*work
)
3340 return queue_work(kblockd_workqueue
, work
);
3342 EXPORT_SYMBOL(kblockd_schedule_work
);
3344 int kblockd_schedule_work_on(int cpu
, struct work_struct
*work
)
3346 return queue_work_on(cpu
, kblockd_workqueue
, work
);
3348 EXPORT_SYMBOL(kblockd_schedule_work_on
);
3350 int kblockd_mod_delayed_work_on(int cpu
, struct delayed_work
*dwork
,
3351 unsigned long delay
)
3353 return mod_delayed_work_on(cpu
, kblockd_workqueue
, dwork
, delay
);
3355 EXPORT_SYMBOL(kblockd_mod_delayed_work_on
);
3357 int kblockd_schedule_delayed_work(struct delayed_work
*dwork
,
3358 unsigned long delay
)
3360 return queue_delayed_work(kblockd_workqueue
, dwork
, delay
);
3362 EXPORT_SYMBOL(kblockd_schedule_delayed_work
);
3364 int kblockd_schedule_delayed_work_on(int cpu
, struct delayed_work
*dwork
,
3365 unsigned long delay
)
3367 return queue_delayed_work_on(cpu
, kblockd_workqueue
, dwork
, delay
);
3369 EXPORT_SYMBOL(kblockd_schedule_delayed_work_on
);
3372 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3373 * @plug: The &struct blk_plug that needs to be initialized
3376 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3377 * pending I/O should the task end up blocking between blk_start_plug() and
3378 * blk_finish_plug(). This is important from a performance perspective, but
3379 * also ensures that we don't deadlock. For instance, if the task is blocking
3380 * for a memory allocation, memory reclaim could end up wanting to free a
3381 * page belonging to that request that is currently residing in our private
3382 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3383 * this kind of deadlock.
3385 void blk_start_plug(struct blk_plug
*plug
)
3387 struct task_struct
*tsk
= current
;
3390 * If this is a nested plug, don't actually assign it.
3395 INIT_LIST_HEAD(&plug
->list
);
3396 INIT_LIST_HEAD(&plug
->mq_list
);
3397 INIT_LIST_HEAD(&plug
->cb_list
);
3399 * Store ordering should not be needed here, since a potential
3400 * preempt will imply a full memory barrier
3404 EXPORT_SYMBOL(blk_start_plug
);
3406 static int plug_rq_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
3408 struct request
*rqa
= container_of(a
, struct request
, queuelist
);
3409 struct request
*rqb
= container_of(b
, struct request
, queuelist
);
3411 return !(rqa
->q
< rqb
->q
||
3412 (rqa
->q
== rqb
->q
&& blk_rq_pos(rqa
) < blk_rq_pos(rqb
)));
3416 * If 'from_schedule' is true, then postpone the dispatch of requests
3417 * until a safe kblockd context. We due this to avoid accidental big
3418 * additional stack usage in driver dispatch, in places where the originally
3419 * plugger did not intend it.
3421 static void queue_unplugged(struct request_queue
*q
, unsigned int depth
,
3423 __releases(q
->queue_lock
)
3425 lockdep_assert_held(q
->queue_lock
);
3427 trace_block_unplug(q
, depth
, !from_schedule
);
3430 blk_run_queue_async(q
);
3433 spin_unlock(q
->queue_lock
);
3436 static void flush_plug_callbacks(struct blk_plug
*plug
, bool from_schedule
)
3438 LIST_HEAD(callbacks
);
3440 while (!list_empty(&plug
->cb_list
)) {
3441 list_splice_init(&plug
->cb_list
, &callbacks
);
3443 while (!list_empty(&callbacks
)) {
3444 struct blk_plug_cb
*cb
= list_first_entry(&callbacks
,
3447 list_del(&cb
->list
);
3448 cb
->callback(cb
, from_schedule
);
3453 struct blk_plug_cb
*blk_check_plugged(blk_plug_cb_fn unplug
, void *data
,
3456 struct blk_plug
*plug
= current
->plug
;
3457 struct blk_plug_cb
*cb
;
3462 list_for_each_entry(cb
, &plug
->cb_list
, list
)
3463 if (cb
->callback
== unplug
&& cb
->data
== data
)
3466 /* Not currently on the callback list */
3467 BUG_ON(size
< sizeof(*cb
));
3468 cb
= kzalloc(size
, GFP_ATOMIC
);
3471 cb
->callback
= unplug
;
3472 list_add(&cb
->list
, &plug
->cb_list
);
3476 EXPORT_SYMBOL(blk_check_plugged
);
3478 void blk_flush_plug_list(struct blk_plug
*plug
, bool from_schedule
)
3480 struct request_queue
*q
;
3481 unsigned long flags
;
3486 flush_plug_callbacks(plug
, from_schedule
);
3488 if (!list_empty(&plug
->mq_list
))
3489 blk_mq_flush_plug_list(plug
, from_schedule
);
3491 if (list_empty(&plug
->list
))
3494 list_splice_init(&plug
->list
, &list
);
3496 list_sort(NULL
, &list
, plug_rq_cmp
);
3502 * Save and disable interrupts here, to avoid doing it for every
3503 * queue lock we have to take.
3505 local_irq_save(flags
);
3506 while (!list_empty(&list
)) {
3507 rq
= list_entry_rq(list
.next
);
3508 list_del_init(&rq
->queuelist
);
3512 * This drops the queue lock
3515 queue_unplugged(q
, depth
, from_schedule
);
3518 spin_lock(q
->queue_lock
);
3522 * Short-circuit if @q is dead
3524 if (unlikely(blk_queue_dying(q
))) {
3525 __blk_end_request_all(rq
, BLK_STS_IOERR
);
3530 * rq is already accounted, so use raw insert
3532 if (op_is_flush(rq
->cmd_flags
))
3533 __elv_add_request(q
, rq
, ELEVATOR_INSERT_FLUSH
);
3535 __elv_add_request(q
, rq
, ELEVATOR_INSERT_SORT_MERGE
);
3541 * This drops the queue lock
3544 queue_unplugged(q
, depth
, from_schedule
);
3546 local_irq_restore(flags
);
3549 void blk_finish_plug(struct blk_plug
*plug
)
3551 if (plug
!= current
->plug
)
3553 blk_flush_plug_list(plug
, false);
3555 current
->plug
= NULL
;
3557 EXPORT_SYMBOL(blk_finish_plug
);
3561 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3562 * @q: the queue of the device
3563 * @dev: the device the queue belongs to
3566 * Initialize runtime-PM-related fields for @q and start auto suspend for
3567 * @dev. Drivers that want to take advantage of request-based runtime PM
3568 * should call this function after @dev has been initialized, and its
3569 * request queue @q has been allocated, and runtime PM for it can not happen
3570 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3571 * cases, driver should call this function before any I/O has taken place.
3573 * This function takes care of setting up using auto suspend for the device,
3574 * the autosuspend delay is set to -1 to make runtime suspend impossible
3575 * until an updated value is either set by user or by driver. Drivers do
3576 * not need to touch other autosuspend settings.
3578 * The block layer runtime PM is request based, so only works for drivers
3579 * that use request as their IO unit instead of those directly use bio's.
3581 void blk_pm_runtime_init(struct request_queue
*q
, struct device
*dev
)
3583 /* not support for RQF_PM and ->rpm_status in blk-mq yet */
3588 q
->rpm_status
= RPM_ACTIVE
;
3589 pm_runtime_set_autosuspend_delay(q
->dev
, -1);
3590 pm_runtime_use_autosuspend(q
->dev
);
3592 EXPORT_SYMBOL(blk_pm_runtime_init
);
3595 * blk_pre_runtime_suspend - Pre runtime suspend check
3596 * @q: the queue of the device
3599 * This function will check if runtime suspend is allowed for the device
3600 * by examining if there are any requests pending in the queue. If there
3601 * are requests pending, the device can not be runtime suspended; otherwise,
3602 * the queue's status will be updated to SUSPENDING and the driver can
3603 * proceed to suspend the device.
3605 * For the not allowed case, we mark last busy for the device so that
3606 * runtime PM core will try to autosuspend it some time later.
3608 * This function should be called near the start of the device's
3609 * runtime_suspend callback.
3612 * 0 - OK to runtime suspend the device
3613 * -EBUSY - Device should not be runtime suspended
3615 int blk_pre_runtime_suspend(struct request_queue
*q
)
3622 spin_lock_irq(q
->queue_lock
);
3623 if (q
->nr_pending
) {
3625 pm_runtime_mark_last_busy(q
->dev
);
3627 q
->rpm_status
= RPM_SUSPENDING
;
3629 spin_unlock_irq(q
->queue_lock
);
3632 EXPORT_SYMBOL(blk_pre_runtime_suspend
);
3635 * blk_post_runtime_suspend - Post runtime suspend processing
3636 * @q: the queue of the device
3637 * @err: return value of the device's runtime_suspend function
3640 * Update the queue's runtime status according to the return value of the
3641 * device's runtime suspend function and mark last busy for the device so
3642 * that PM core will try to auto suspend the device at a later time.
3644 * This function should be called near the end of the device's
3645 * runtime_suspend callback.
3647 void blk_post_runtime_suspend(struct request_queue
*q
, int err
)
3652 spin_lock_irq(q
->queue_lock
);
3654 q
->rpm_status
= RPM_SUSPENDED
;
3656 q
->rpm_status
= RPM_ACTIVE
;
3657 pm_runtime_mark_last_busy(q
->dev
);
3659 spin_unlock_irq(q
->queue_lock
);
3661 EXPORT_SYMBOL(blk_post_runtime_suspend
);
3664 * blk_pre_runtime_resume - Pre runtime resume processing
3665 * @q: the queue of the device
3668 * Update the queue's runtime status to RESUMING in preparation for the
3669 * runtime resume of the device.
3671 * This function should be called near the start of the device's
3672 * runtime_resume callback.
3674 void blk_pre_runtime_resume(struct request_queue
*q
)
3679 spin_lock_irq(q
->queue_lock
);
3680 q
->rpm_status
= RPM_RESUMING
;
3681 spin_unlock_irq(q
->queue_lock
);
3683 EXPORT_SYMBOL(blk_pre_runtime_resume
);
3686 * blk_post_runtime_resume - Post runtime resume processing
3687 * @q: the queue of the device
3688 * @err: return value of the device's runtime_resume function
3691 * Update the queue's runtime status according to the return value of the
3692 * device's runtime_resume function. If it is successfully resumed, process
3693 * the requests that are queued into the device's queue when it is resuming
3694 * and then mark last busy and initiate autosuspend for it.
3696 * This function should be called near the end of the device's
3697 * runtime_resume callback.
3699 void blk_post_runtime_resume(struct request_queue
*q
, int err
)
3704 spin_lock_irq(q
->queue_lock
);
3706 q
->rpm_status
= RPM_ACTIVE
;
3708 pm_runtime_mark_last_busy(q
->dev
);
3709 pm_request_autosuspend(q
->dev
);
3711 q
->rpm_status
= RPM_SUSPENDED
;
3713 spin_unlock_irq(q
->queue_lock
);
3715 EXPORT_SYMBOL(blk_post_runtime_resume
);
3718 * blk_set_runtime_active - Force runtime status of the queue to be active
3719 * @q: the queue of the device
3721 * If the device is left runtime suspended during system suspend the resume
3722 * hook typically resumes the device and corrects runtime status
3723 * accordingly. However, that does not affect the queue runtime PM status
3724 * which is still "suspended". This prevents processing requests from the
3727 * This function can be used in driver's resume hook to correct queue
3728 * runtime PM status and re-enable peeking requests from the queue. It
3729 * should be called before first request is added to the queue.
3731 void blk_set_runtime_active(struct request_queue
*q
)
3733 spin_lock_irq(q
->queue_lock
);
3734 q
->rpm_status
= RPM_ACTIVE
;
3735 pm_runtime_mark_last_busy(q
->dev
);
3736 pm_request_autosuspend(q
->dev
);
3737 spin_unlock_irq(q
->queue_lock
);
3739 EXPORT_SYMBOL(blk_set_runtime_active
);
3742 int __init
blk_dev_init(void)
3744 BUILD_BUG_ON(REQ_OP_LAST
>= (1 << REQ_OP_BITS
));
3745 BUILD_BUG_ON(REQ_OP_BITS
+ REQ_FLAG_BITS
> 8 *
3746 FIELD_SIZEOF(struct request
, cmd_flags
));
3747 BUILD_BUG_ON(REQ_OP_BITS
+ REQ_FLAG_BITS
> 8 *
3748 FIELD_SIZEOF(struct bio
, bi_opf
));
3750 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3751 kblockd_workqueue
= alloc_workqueue("kblockd",
3752 WQ_MEM_RECLAIM
| WQ_HIGHPRI
, 0);
3753 if (!kblockd_workqueue
)
3754 panic("Failed to create kblockd\n");
3756 request_cachep
= kmem_cache_create("blkdev_requests",
3757 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
3759 blk_requestq_cachep
= kmem_cache_create("request_queue",
3760 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);
3762 #ifdef CONFIG_DEBUG_FS
3763 blk_debugfs_root
= debugfs_create_dir("block", NULL
);