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
;
1268 if (rq_flags
& RQF_ELVPRIV
) {
1269 if (unlikely(et
->icq_cache
&& !icq
)) {
1271 icq
= ioc_create_icq(ioc
, q
, gfp_mask
);
1277 if (unlikely(elv_set_request(q
, rq
, bio
, gfp_mask
)))
1280 /* @rq->elv.icq holds io_context until @rq is freed */
1282 get_io_context(icq
->ioc
);
1286 * ioc may be NULL here, and ioc_batching will be false. That's
1287 * OK, if the queue is under the request limit then requests need
1288 * not count toward the nr_batch_requests limit. There will always
1289 * be some limit enforced by BLK_BATCH_TIME.
1291 if (ioc_batching(q
, ioc
))
1292 ioc
->nr_batch_requests
--;
1294 trace_block_getrq(q
, bio
, op
);
1299 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1300 * and may fail indefinitely under memory pressure and thus
1301 * shouldn't stall IO. Treat this request as !elvpriv. This will
1302 * disturb iosched and blkcg but weird is bettern than dead.
1304 printk_ratelimited(KERN_WARNING
"%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1305 __func__
, dev_name(q
->backing_dev_info
->dev
));
1307 rq
->rq_flags
&= ~RQF_ELVPRIV
;
1310 spin_lock_irq(q
->queue_lock
);
1311 q
->nr_rqs_elvpriv
--;
1312 spin_unlock_irq(q
->queue_lock
);
1317 * Allocation failed presumably due to memory. Undo anything we
1318 * might have messed up.
1320 * Allocating task should really be put onto the front of the wait
1321 * queue, but this is pretty rare.
1323 spin_lock_irq(q
->queue_lock
);
1324 freed_request(rl
, is_sync
, rq_flags
);
1327 * in the very unlikely event that allocation failed and no
1328 * requests for this direction was pending, mark us starved so that
1329 * freeing of a request in the other direction will notice
1330 * us. another possible fix would be to split the rq mempool into
1334 if (unlikely(rl
->count
[is_sync
] == 0))
1335 rl
->starved
[is_sync
] = 1;
1336 return ERR_PTR(-ENOMEM
);
1340 * get_request - get a free request
1341 * @q: request_queue to allocate request from
1342 * @op: operation and flags
1343 * @bio: bio to allocate request for (can be %NULL)
1344 * @flags: BLK_MQ_REQ_* flags.
1346 * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask,
1347 * this function keeps retrying under memory pressure and fails iff @q is dead.
1349 * Must be called with @q->queue_lock held and,
1350 * Returns ERR_PTR on failure, with @q->queue_lock held.
1351 * Returns request pointer on success, with @q->queue_lock *not held*.
1353 static struct request
*get_request(struct request_queue
*q
, unsigned int op
,
1354 struct bio
*bio
, unsigned int flags
)
1356 const bool is_sync
= op_is_sync(op
);
1358 struct request_list
*rl
;
1361 lockdep_assert_held(q
->queue_lock
);
1362 WARN_ON_ONCE(q
->mq_ops
);
1364 rl
= blk_get_rl(q
, bio
); /* transferred to @rq on success */
1366 rq
= __get_request(rl
, op
, bio
, flags
);
1370 if (op
& REQ_NOWAIT
) {
1372 return ERR_PTR(-EAGAIN
);
1375 if ((flags
& BLK_MQ_REQ_NOWAIT
) || unlikely(blk_queue_dying(q
))) {
1380 /* wait on @rl and retry */
1381 prepare_to_wait_exclusive(&rl
->wait
[is_sync
], &wait
,
1382 TASK_UNINTERRUPTIBLE
);
1384 trace_block_sleeprq(q
, bio
, op
);
1386 spin_unlock_irq(q
->queue_lock
);
1390 * After sleeping, we become a "batching" process and will be able
1391 * to allocate at least one request, and up to a big batch of them
1392 * for a small period time. See ioc_batching, ioc_set_batching
1394 ioc_set_batching(q
, current
->io_context
);
1396 spin_lock_irq(q
->queue_lock
);
1397 finish_wait(&rl
->wait
[is_sync
], &wait
);
1402 /* flags: BLK_MQ_REQ_PREEMPT and/or BLK_MQ_REQ_NOWAIT. */
1403 static struct request
*blk_old_get_request(struct request_queue
*q
,
1404 unsigned int op
, unsigned int flags
)
1407 gfp_t gfp_mask
= flags
& BLK_MQ_REQ_NOWAIT
? GFP_ATOMIC
:
1408 __GFP_DIRECT_RECLAIM
;
1411 WARN_ON_ONCE(q
->mq_ops
);
1413 /* create ioc upfront */
1414 create_io_context(gfp_mask
, q
->node
);
1416 ret
= blk_queue_enter(q
, !(gfp_mask
& __GFP_DIRECT_RECLAIM
) ||
1419 return ERR_PTR(ret
);
1420 spin_lock_irq(q
->queue_lock
);
1421 rq
= get_request(q
, op
, NULL
, flags
);
1423 spin_unlock_irq(q
->queue_lock
);
1428 /* q->queue_lock is unlocked at this point */
1430 rq
->__sector
= (sector_t
) -1;
1431 rq
->bio
= rq
->biotail
= NULL
;
1436 * blk_get_request_flags - allocate a request
1437 * @q: request queue to allocate a request for
1438 * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC.
1439 * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
1441 struct request
*blk_get_request_flags(struct request_queue
*q
, unsigned int op
,
1444 struct request
*req
;
1446 WARN_ON_ONCE(op
& REQ_NOWAIT
);
1447 WARN_ON_ONCE(flags
& ~BLK_MQ_REQ_NOWAIT
);
1450 req
= blk_mq_alloc_request(q
, op
, flags
);
1451 if (!IS_ERR(req
) && q
->mq_ops
->initialize_rq_fn
)
1452 q
->mq_ops
->initialize_rq_fn(req
);
1454 req
= blk_old_get_request(q
, op
, flags
);
1455 if (!IS_ERR(req
) && q
->initialize_rq_fn
)
1456 q
->initialize_rq_fn(req
);
1461 EXPORT_SYMBOL(blk_get_request_flags
);
1463 struct request
*blk_get_request(struct request_queue
*q
, unsigned int op
,
1466 return blk_get_request_flags(q
, op
, gfp_mask
& __GFP_DIRECT_RECLAIM
?
1467 0 : BLK_MQ_REQ_NOWAIT
);
1469 EXPORT_SYMBOL(blk_get_request
);
1472 * blk_requeue_request - put a request back on queue
1473 * @q: request queue where request should be inserted
1474 * @rq: request to be inserted
1477 * Drivers often keep queueing requests until the hardware cannot accept
1478 * more, when that condition happens we need to put the request back
1479 * on the queue. Must be called with queue lock held.
1481 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
1483 lockdep_assert_held(q
->queue_lock
);
1484 WARN_ON_ONCE(q
->mq_ops
);
1486 blk_delete_timer(rq
);
1487 blk_clear_rq_complete(rq
);
1488 trace_block_rq_requeue(q
, rq
);
1489 wbt_requeue(q
->rq_wb
, &rq
->issue_stat
);
1491 if (rq
->rq_flags
& RQF_QUEUED
)
1492 blk_queue_end_tag(q
, rq
);
1494 BUG_ON(blk_queued_rq(rq
));
1496 elv_requeue_request(q
, rq
);
1498 EXPORT_SYMBOL(blk_requeue_request
);
1500 static void add_acct_request(struct request_queue
*q
, struct request
*rq
,
1503 blk_account_io_start(rq
, true);
1504 __elv_add_request(q
, rq
, where
);
1507 static void part_round_stats_single(struct request_queue
*q
, int cpu
,
1508 struct hd_struct
*part
, unsigned long now
,
1509 unsigned int inflight
)
1512 __part_stat_add(cpu
, part
, time_in_queue
,
1513 inflight
* (now
- part
->stamp
));
1514 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
1520 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1521 * @q: target block queue
1522 * @cpu: cpu number for stats access
1523 * @part: target partition
1525 * The average IO queue length and utilisation statistics are maintained
1526 * by observing the current state of the queue length and the amount of
1527 * time it has been in this state for.
1529 * Normally, that accounting is done on IO completion, but that can result
1530 * in more than a second's worth of IO being accounted for within any one
1531 * second, leading to >100% utilisation. To deal with that, we call this
1532 * function to do a round-off before returning the results when reading
1533 * /proc/diskstats. This accounts immediately for all queue usage up to
1534 * the current jiffies and restarts the counters again.
1536 void part_round_stats(struct request_queue
*q
, int cpu
, struct hd_struct
*part
)
1538 struct hd_struct
*part2
= NULL
;
1539 unsigned long now
= jiffies
;
1540 unsigned int inflight
[2];
1543 if (part
->stamp
!= now
)
1547 part2
= &part_to_disk(part
)->part0
;
1548 if (part2
->stamp
!= now
)
1555 part_in_flight(q
, part
, inflight
);
1558 part_round_stats_single(q
, cpu
, part2
, now
, inflight
[1]);
1560 part_round_stats_single(q
, cpu
, part
, now
, inflight
[0]);
1562 EXPORT_SYMBOL_GPL(part_round_stats
);
1565 static void blk_pm_put_request(struct request
*rq
)
1567 if (rq
->q
->dev
&& !(rq
->rq_flags
& RQF_PM
) && !--rq
->q
->nr_pending
)
1568 pm_runtime_mark_last_busy(rq
->q
->dev
);
1571 static inline void blk_pm_put_request(struct request
*rq
) {}
1574 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1576 req_flags_t rq_flags
= req
->rq_flags
;
1582 blk_mq_free_request(req
);
1586 lockdep_assert_held(q
->queue_lock
);
1588 blk_pm_put_request(req
);
1590 elv_completed_request(q
, req
);
1592 /* this is a bio leak */
1593 WARN_ON(req
->bio
!= NULL
);
1595 wbt_done(q
->rq_wb
, &req
->issue_stat
);
1598 * Request may not have originated from ll_rw_blk. if not,
1599 * it didn't come out of our reserved rq pools
1601 if (rq_flags
& RQF_ALLOCED
) {
1602 struct request_list
*rl
= blk_rq_rl(req
);
1603 bool sync
= op_is_sync(req
->cmd_flags
);
1605 BUG_ON(!list_empty(&req
->queuelist
));
1606 BUG_ON(ELV_ON_HASH(req
));
1608 blk_free_request(rl
, req
);
1609 freed_request(rl
, sync
, rq_flags
);
1614 EXPORT_SYMBOL_GPL(__blk_put_request
);
1616 void blk_put_request(struct request
*req
)
1618 struct request_queue
*q
= req
->q
;
1621 blk_mq_free_request(req
);
1623 unsigned long flags
;
1625 spin_lock_irqsave(q
->queue_lock
, flags
);
1626 __blk_put_request(q
, req
);
1627 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1630 EXPORT_SYMBOL(blk_put_request
);
1632 bool bio_attempt_back_merge(struct request_queue
*q
, struct request
*req
,
1635 const int ff
= bio
->bi_opf
& REQ_FAILFAST_MASK
;
1637 if (!ll_back_merge_fn(q
, req
, bio
))
1640 trace_block_bio_backmerge(q
, req
, bio
);
1642 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1643 blk_rq_set_mixed_merge(req
);
1645 req
->biotail
->bi_next
= bio
;
1647 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1648 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1650 blk_account_io_start(req
, false);
1654 bool bio_attempt_front_merge(struct request_queue
*q
, struct request
*req
,
1657 const int ff
= bio
->bi_opf
& REQ_FAILFAST_MASK
;
1659 if (!ll_front_merge_fn(q
, req
, bio
))
1662 trace_block_bio_frontmerge(q
, req
, bio
);
1664 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1665 blk_rq_set_mixed_merge(req
);
1667 bio
->bi_next
= req
->bio
;
1670 req
->__sector
= bio
->bi_iter
.bi_sector
;
1671 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1672 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1674 blk_account_io_start(req
, false);
1678 bool bio_attempt_discard_merge(struct request_queue
*q
, struct request
*req
,
1681 unsigned short segments
= blk_rq_nr_discard_segments(req
);
1683 if (segments
>= queue_max_discard_segments(q
))
1685 if (blk_rq_sectors(req
) + bio_sectors(bio
) >
1686 blk_rq_get_max_sectors(req
, blk_rq_pos(req
)))
1689 req
->biotail
->bi_next
= bio
;
1691 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1692 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1693 req
->nr_phys_segments
= segments
+ 1;
1695 blk_account_io_start(req
, false);
1698 req_set_nomerge(q
, req
);
1703 * blk_attempt_plug_merge - try to merge with %current's plugged list
1704 * @q: request_queue new bio is being queued at
1705 * @bio: new bio being queued
1706 * @request_count: out parameter for number of traversed plugged requests
1707 * @same_queue_rq: pointer to &struct request that gets filled in when
1708 * another request associated with @q is found on the plug list
1709 * (optional, may be %NULL)
1711 * Determine whether @bio being queued on @q can be merged with a request
1712 * on %current's plugged list. Returns %true if merge was successful,
1715 * Plugging coalesces IOs from the same issuer for the same purpose without
1716 * going through @q->queue_lock. As such it's more of an issuing mechanism
1717 * than scheduling, and the request, while may have elvpriv data, is not
1718 * added on the elevator at this point. In addition, we don't have
1719 * reliable access to the elevator outside queue lock. Only check basic
1720 * merging parameters without querying the elevator.
1722 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1724 bool blk_attempt_plug_merge(struct request_queue
*q
, struct bio
*bio
,
1725 unsigned int *request_count
,
1726 struct request
**same_queue_rq
)
1728 struct blk_plug
*plug
;
1730 struct list_head
*plug_list
;
1732 plug
= current
->plug
;
1738 plug_list
= &plug
->mq_list
;
1740 plug_list
= &plug
->list
;
1742 list_for_each_entry_reverse(rq
, plug_list
, queuelist
) {
1743 bool merged
= false;
1748 * Only blk-mq multiple hardware queues case checks the
1749 * rq in the same queue, there should be only one such
1753 *same_queue_rq
= rq
;
1756 if (rq
->q
!= q
|| !blk_rq_merge_ok(rq
, bio
))
1759 switch (blk_try_merge(rq
, bio
)) {
1760 case ELEVATOR_BACK_MERGE
:
1761 merged
= bio_attempt_back_merge(q
, rq
, bio
);
1763 case ELEVATOR_FRONT_MERGE
:
1764 merged
= bio_attempt_front_merge(q
, rq
, bio
);
1766 case ELEVATOR_DISCARD_MERGE
:
1767 merged
= bio_attempt_discard_merge(q
, rq
, bio
);
1780 unsigned int blk_plug_queued_count(struct request_queue
*q
)
1782 struct blk_plug
*plug
;
1784 struct list_head
*plug_list
;
1785 unsigned int ret
= 0;
1787 plug
= current
->plug
;
1792 plug_list
= &plug
->mq_list
;
1794 plug_list
= &plug
->list
;
1796 list_for_each_entry(rq
, plug_list
, queuelist
) {
1804 void blk_init_request_from_bio(struct request
*req
, struct bio
*bio
)
1806 struct io_context
*ioc
= rq_ioc(bio
);
1808 if (bio
->bi_opf
& REQ_RAHEAD
)
1809 req
->cmd_flags
|= REQ_FAILFAST_MASK
;
1811 req
->__sector
= bio
->bi_iter
.bi_sector
;
1812 if (ioprio_valid(bio_prio(bio
)))
1813 req
->ioprio
= bio_prio(bio
);
1815 req
->ioprio
= ioc
->ioprio
;
1817 req
->ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE
, 0);
1818 req
->write_hint
= bio
->bi_write_hint
;
1819 blk_rq_bio_prep(req
->q
, req
, bio
);
1821 EXPORT_SYMBOL_GPL(blk_init_request_from_bio
);
1823 static blk_qc_t
blk_queue_bio(struct request_queue
*q
, struct bio
*bio
)
1825 struct blk_plug
*plug
;
1826 int where
= ELEVATOR_INSERT_SORT
;
1827 struct request
*req
, *free
;
1828 unsigned int request_count
= 0;
1829 unsigned int wb_acct
;
1832 * low level driver can indicate that it wants pages above a
1833 * certain limit bounced to low memory (ie for highmem, or even
1834 * ISA dma in theory)
1836 blk_queue_bounce(q
, &bio
);
1838 blk_queue_split(q
, &bio
);
1840 if (!bio_integrity_prep(bio
))
1841 return BLK_QC_T_NONE
;
1843 if (op_is_flush(bio
->bi_opf
)) {
1844 spin_lock_irq(q
->queue_lock
);
1845 where
= ELEVATOR_INSERT_FLUSH
;
1850 * Check if we can merge with the plugged list before grabbing
1853 if (!blk_queue_nomerges(q
)) {
1854 if (blk_attempt_plug_merge(q
, bio
, &request_count
, NULL
))
1855 return BLK_QC_T_NONE
;
1857 request_count
= blk_plug_queued_count(q
);
1859 spin_lock_irq(q
->queue_lock
);
1861 switch (elv_merge(q
, &req
, bio
)) {
1862 case ELEVATOR_BACK_MERGE
:
1863 if (!bio_attempt_back_merge(q
, req
, bio
))
1865 elv_bio_merged(q
, req
, bio
);
1866 free
= attempt_back_merge(q
, req
);
1868 __blk_put_request(q
, free
);
1870 elv_merged_request(q
, req
, ELEVATOR_BACK_MERGE
);
1872 case ELEVATOR_FRONT_MERGE
:
1873 if (!bio_attempt_front_merge(q
, req
, bio
))
1875 elv_bio_merged(q
, req
, bio
);
1876 free
= attempt_front_merge(q
, req
);
1878 __blk_put_request(q
, free
);
1880 elv_merged_request(q
, req
, ELEVATOR_FRONT_MERGE
);
1887 wb_acct
= wbt_wait(q
->rq_wb
, bio
, q
->queue_lock
);
1890 * Grab a free request. This is might sleep but can not fail.
1891 * Returns with the queue unlocked.
1893 blk_queue_enter_live(q
);
1894 req
= get_request(q
, bio
->bi_opf
, bio
, 0);
1897 __wbt_done(q
->rq_wb
, wb_acct
);
1898 if (PTR_ERR(req
) == -ENOMEM
)
1899 bio
->bi_status
= BLK_STS_RESOURCE
;
1901 bio
->bi_status
= BLK_STS_IOERR
;
1906 wbt_track(&req
->issue_stat
, wb_acct
);
1909 * After dropping the lock and possibly sleeping here, our request
1910 * may now be mergeable after it had proven unmergeable (above).
1911 * We don't worry about that case for efficiency. It won't happen
1912 * often, and the elevators are able to handle it.
1914 blk_init_request_from_bio(req
, bio
);
1916 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
))
1917 req
->cpu
= raw_smp_processor_id();
1919 plug
= current
->plug
;
1922 * If this is the first request added after a plug, fire
1925 * @request_count may become stale because of schedule
1926 * out, so check plug list again.
1928 if (!request_count
|| list_empty(&plug
->list
))
1929 trace_block_plug(q
);
1931 struct request
*last
= list_entry_rq(plug
->list
.prev
);
1932 if (request_count
>= BLK_MAX_REQUEST_COUNT
||
1933 blk_rq_bytes(last
) >= BLK_PLUG_FLUSH_SIZE
) {
1934 blk_flush_plug_list(plug
, false);
1935 trace_block_plug(q
);
1938 list_add_tail(&req
->queuelist
, &plug
->list
);
1939 blk_account_io_start(req
, true);
1941 spin_lock_irq(q
->queue_lock
);
1942 add_acct_request(q
, req
, where
);
1945 spin_unlock_irq(q
->queue_lock
);
1948 return BLK_QC_T_NONE
;
1951 static void handle_bad_sector(struct bio
*bio
)
1953 char b
[BDEVNAME_SIZE
];
1955 printk(KERN_INFO
"attempt to access beyond end of device\n");
1956 printk(KERN_INFO
"%s: rw=%d, want=%Lu, limit=%Lu\n",
1957 bio_devname(bio
, b
), bio
->bi_opf
,
1958 (unsigned long long)bio_end_sector(bio
),
1959 (long long)get_capacity(bio
->bi_disk
));
1962 #ifdef CONFIG_FAIL_MAKE_REQUEST
1964 static DECLARE_FAULT_ATTR(fail_make_request
);
1966 static int __init
setup_fail_make_request(char *str
)
1968 return setup_fault_attr(&fail_make_request
, str
);
1970 __setup("fail_make_request=", setup_fail_make_request
);
1972 static bool should_fail_request(struct hd_struct
*part
, unsigned int bytes
)
1974 return part
->make_it_fail
&& should_fail(&fail_make_request
, bytes
);
1977 static int __init
fail_make_request_debugfs(void)
1979 struct dentry
*dir
= fault_create_debugfs_attr("fail_make_request",
1980 NULL
, &fail_make_request
);
1982 return PTR_ERR_OR_ZERO(dir
);
1985 late_initcall(fail_make_request_debugfs
);
1987 #else /* CONFIG_FAIL_MAKE_REQUEST */
1989 static inline bool should_fail_request(struct hd_struct
*part
,
1995 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1998 * Remap block n of partition p to block n+start(p) of the disk.
2000 static inline int blk_partition_remap(struct bio
*bio
)
2002 struct hd_struct
*p
;
2006 * Zone reset does not include bi_size so bio_sectors() is always 0.
2007 * Include a test for the reset op code and perform the remap if needed.
2009 if (!bio
->bi_partno
||
2010 (!bio_sectors(bio
) && bio_op(bio
) != REQ_OP_ZONE_RESET
))
2014 p
= __disk_get_part(bio
->bi_disk
, bio
->bi_partno
);
2015 if (likely(p
&& !should_fail_request(p
, bio
->bi_iter
.bi_size
))) {
2016 bio
->bi_iter
.bi_sector
+= p
->start_sect
;
2018 trace_block_bio_remap(bio
->bi_disk
->queue
, bio
, part_devt(p
),
2019 bio
->bi_iter
.bi_sector
- p
->start_sect
);
2021 printk("%s: fail for partition %d\n", __func__
, bio
->bi_partno
);
2030 * Check whether this bio extends beyond the end of the device.
2032 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
2039 /* Test device or partition size, when known. */
2040 maxsector
= get_capacity(bio
->bi_disk
);
2042 sector_t sector
= bio
->bi_iter
.bi_sector
;
2044 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
2046 * This may well happen - the kernel calls bread()
2047 * without checking the size of the device, e.g., when
2048 * mounting a device.
2050 handle_bad_sector(bio
);
2058 static noinline_for_stack
bool
2059 generic_make_request_checks(struct bio
*bio
)
2061 struct request_queue
*q
;
2062 int nr_sectors
= bio_sectors(bio
);
2063 blk_status_t status
= BLK_STS_IOERR
;
2064 char b
[BDEVNAME_SIZE
];
2068 if (bio_check_eod(bio
, nr_sectors
))
2071 q
= bio
->bi_disk
->queue
;
2074 "generic_make_request: Trying to access "
2075 "nonexistent block-device %s (%Lu)\n",
2076 bio_devname(bio
, b
), (long long)bio
->bi_iter
.bi_sector
);
2081 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
2082 * if queue is not a request based queue.
2085 if ((bio
->bi_opf
& REQ_NOWAIT
) && !queue_is_rq_based(q
))
2088 if (should_fail_request(&bio
->bi_disk
->part0
, bio
->bi_iter
.bi_size
))
2091 if (blk_partition_remap(bio
))
2094 if (bio_check_eod(bio
, nr_sectors
))
2098 * Filter flush bio's early so that make_request based
2099 * drivers without flush support don't have to worry
2102 if (op_is_flush(bio
->bi_opf
) &&
2103 !test_bit(QUEUE_FLAG_WC
, &q
->queue_flags
)) {
2104 bio
->bi_opf
&= ~(REQ_PREFLUSH
| REQ_FUA
);
2106 status
= BLK_STS_OK
;
2111 switch (bio_op(bio
)) {
2112 case REQ_OP_DISCARD
:
2113 if (!blk_queue_discard(q
))
2116 case REQ_OP_SECURE_ERASE
:
2117 if (!blk_queue_secure_erase(q
))
2120 case REQ_OP_WRITE_SAME
:
2121 if (!q
->limits
.max_write_same_sectors
)
2124 case REQ_OP_ZONE_REPORT
:
2125 case REQ_OP_ZONE_RESET
:
2126 if (!blk_queue_is_zoned(q
))
2129 case REQ_OP_WRITE_ZEROES
:
2130 if (!q
->limits
.max_write_zeroes_sectors
)
2138 * Various block parts want %current->io_context and lazy ioc
2139 * allocation ends up trading a lot of pain for a small amount of
2140 * memory. Just allocate it upfront. This may fail and block
2141 * layer knows how to live with it.
2143 create_io_context(GFP_ATOMIC
, q
->node
);
2145 if (!blkcg_bio_issue_check(q
, bio
))
2148 if (!bio_flagged(bio
, BIO_TRACE_COMPLETION
)) {
2149 trace_block_bio_queue(q
, bio
);
2150 /* Now that enqueuing has been traced, we need to trace
2151 * completion as well.
2153 bio_set_flag(bio
, BIO_TRACE_COMPLETION
);
2158 status
= BLK_STS_NOTSUPP
;
2160 bio
->bi_status
= status
;
2166 * generic_make_request - hand a buffer to its device driver for I/O
2167 * @bio: The bio describing the location in memory and on the device.
2169 * generic_make_request() is used to make I/O requests of block
2170 * devices. It is passed a &struct bio, which describes the I/O that needs
2173 * generic_make_request() does not return any status. The
2174 * success/failure status of the request, along with notification of
2175 * completion, is delivered asynchronously through the bio->bi_end_io
2176 * function described (one day) else where.
2178 * The caller of generic_make_request must make sure that bi_io_vec
2179 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2180 * set to describe the device address, and the
2181 * bi_end_io and optionally bi_private are set to describe how
2182 * completion notification should be signaled.
2184 * generic_make_request and the drivers it calls may use bi_next if this
2185 * bio happens to be merged with someone else, and may resubmit the bio to
2186 * a lower device by calling into generic_make_request recursively, which
2187 * means the bio should NOT be touched after the call to ->make_request_fn.
2189 blk_qc_t
generic_make_request(struct bio
*bio
)
2192 * bio_list_on_stack[0] contains bios submitted by the current
2194 * bio_list_on_stack[1] contains bios that were submitted before
2195 * the current make_request_fn, but that haven't been processed
2198 struct bio_list bio_list_on_stack
[2];
2199 blk_qc_t ret
= BLK_QC_T_NONE
;
2201 if (!generic_make_request_checks(bio
))
2205 * We only want one ->make_request_fn to be active at a time, else
2206 * stack usage with stacked devices could be a problem. So use
2207 * current->bio_list to keep a list of requests submited by a
2208 * make_request_fn function. current->bio_list is also used as a
2209 * flag to say if generic_make_request is currently active in this
2210 * task or not. If it is NULL, then no make_request is active. If
2211 * it is non-NULL, then a make_request is active, and new requests
2212 * should be added at the tail
2214 if (current
->bio_list
) {
2215 bio_list_add(¤t
->bio_list
[0], bio
);
2219 /* following loop may be a bit non-obvious, and so deserves some
2221 * Before entering the loop, bio->bi_next is NULL (as all callers
2222 * ensure that) so we have a list with a single bio.
2223 * We pretend that we have just taken it off a longer list, so
2224 * we assign bio_list to a pointer to the bio_list_on_stack,
2225 * thus initialising the bio_list of new bios to be
2226 * added. ->make_request() may indeed add some more bios
2227 * through a recursive call to generic_make_request. If it
2228 * did, we find a non-NULL value in bio_list and re-enter the loop
2229 * from the top. In this case we really did just take the bio
2230 * of the top of the list (no pretending) and so remove it from
2231 * bio_list, and call into ->make_request() again.
2233 BUG_ON(bio
->bi_next
);
2234 bio_list_init(&bio_list_on_stack
[0]);
2235 current
->bio_list
= bio_list_on_stack
;
2237 struct request_queue
*q
= bio
->bi_disk
->queue
;
2239 if (likely(blk_queue_enter(q
, bio
->bi_opf
& REQ_NOWAIT
) == 0)) {
2240 struct bio_list lower
, same
;
2242 /* Create a fresh bio_list for all subordinate requests */
2243 bio_list_on_stack
[1] = bio_list_on_stack
[0];
2244 bio_list_init(&bio_list_on_stack
[0]);
2245 ret
= q
->make_request_fn(q
, bio
);
2249 /* sort new bios into those for a lower level
2250 * and those for the same level
2252 bio_list_init(&lower
);
2253 bio_list_init(&same
);
2254 while ((bio
= bio_list_pop(&bio_list_on_stack
[0])) != NULL
)
2255 if (q
== bio
->bi_disk
->queue
)
2256 bio_list_add(&same
, bio
);
2258 bio_list_add(&lower
, bio
);
2259 /* now assemble so we handle the lowest level first */
2260 bio_list_merge(&bio_list_on_stack
[0], &lower
);
2261 bio_list_merge(&bio_list_on_stack
[0], &same
);
2262 bio_list_merge(&bio_list_on_stack
[0], &bio_list_on_stack
[1]);
2264 if (unlikely(!blk_queue_dying(q
) &&
2265 (bio
->bi_opf
& REQ_NOWAIT
)))
2266 bio_wouldblock_error(bio
);
2270 bio
= bio_list_pop(&bio_list_on_stack
[0]);
2272 current
->bio_list
= NULL
; /* deactivate */
2277 EXPORT_SYMBOL(generic_make_request
);
2280 * direct_make_request - hand a buffer directly to its device driver for I/O
2281 * @bio: The bio describing the location in memory and on the device.
2283 * This function behaves like generic_make_request(), but does not protect
2284 * against recursion. Must only be used if the called driver is known
2285 * to not call generic_make_request (or direct_make_request) again from
2286 * its make_request function. (Calling direct_make_request again from
2287 * a workqueue is perfectly fine as that doesn't recurse).
2289 blk_qc_t
direct_make_request(struct bio
*bio
)
2291 struct request_queue
*q
= bio
->bi_disk
->queue
;
2292 bool nowait
= bio
->bi_opf
& REQ_NOWAIT
;
2295 if (!generic_make_request_checks(bio
))
2296 return BLK_QC_T_NONE
;
2298 if (unlikely(blk_queue_enter(q
, nowait
))) {
2299 if (nowait
&& !blk_queue_dying(q
))
2300 bio
->bi_status
= BLK_STS_AGAIN
;
2302 bio
->bi_status
= BLK_STS_IOERR
;
2304 return BLK_QC_T_NONE
;
2307 ret
= q
->make_request_fn(q
, bio
);
2311 EXPORT_SYMBOL_GPL(direct_make_request
);
2314 * submit_bio - submit a bio to the block device layer for I/O
2315 * @bio: The &struct bio which describes the I/O
2317 * submit_bio() is very similar in purpose to generic_make_request(), and
2318 * uses that function to do most of the work. Both are fairly rough
2319 * interfaces; @bio must be presetup and ready for I/O.
2322 blk_qc_t
submit_bio(struct bio
*bio
)
2325 * If it's a regular read/write or a barrier with data attached,
2326 * go through the normal accounting stuff before submission.
2328 if (bio_has_data(bio
)) {
2331 if (unlikely(bio_op(bio
) == REQ_OP_WRITE_SAME
))
2332 count
= queue_logical_block_size(bio
->bi_disk
->queue
);
2334 count
= bio_sectors(bio
);
2336 if (op_is_write(bio_op(bio
))) {
2337 count_vm_events(PGPGOUT
, count
);
2339 task_io_account_read(bio
->bi_iter
.bi_size
);
2340 count_vm_events(PGPGIN
, count
);
2343 if (unlikely(block_dump
)) {
2344 char b
[BDEVNAME_SIZE
];
2345 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s (%u sectors)\n",
2346 current
->comm
, task_pid_nr(current
),
2347 op_is_write(bio_op(bio
)) ? "WRITE" : "READ",
2348 (unsigned long long)bio
->bi_iter
.bi_sector
,
2349 bio_devname(bio
, b
), count
);
2353 return generic_make_request(bio
);
2355 EXPORT_SYMBOL(submit_bio
);
2357 bool blk_poll(struct request_queue
*q
, blk_qc_t cookie
)
2359 if (!q
->poll_fn
|| !blk_qc_t_valid(cookie
))
2363 blk_flush_plug_list(current
->plug
, false);
2364 return q
->poll_fn(q
, cookie
);
2366 EXPORT_SYMBOL_GPL(blk_poll
);
2369 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2370 * for new the queue limits
2372 * @rq: the request being checked
2375 * @rq may have been made based on weaker limitations of upper-level queues
2376 * in request stacking drivers, and it may violate the limitation of @q.
2377 * Since the block layer and the underlying device driver trust @rq
2378 * after it is inserted to @q, it should be checked against @q before
2379 * the insertion using this generic function.
2381 * Request stacking drivers like request-based dm may change the queue
2382 * limits when retrying requests on other queues. Those requests need
2383 * to be checked against the new queue limits again during dispatch.
2385 static int blk_cloned_rq_check_limits(struct request_queue
*q
,
2388 if (blk_rq_sectors(rq
) > blk_queue_get_max_sectors(q
, req_op(rq
))) {
2389 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
2394 * queue's settings related to segment counting like q->bounce_pfn
2395 * may differ from that of other stacking queues.
2396 * Recalculate it to check the request correctly on this queue's
2399 blk_recalc_rq_segments(rq
);
2400 if (rq
->nr_phys_segments
> queue_max_segments(q
)) {
2401 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
2409 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2410 * @q: the queue to submit the request
2411 * @rq: the request being queued
2413 blk_status_t
blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
2415 unsigned long flags
;
2416 int where
= ELEVATOR_INSERT_BACK
;
2418 if (blk_cloned_rq_check_limits(q
, rq
))
2419 return BLK_STS_IOERR
;
2422 should_fail_request(&rq
->rq_disk
->part0
, blk_rq_bytes(rq
)))
2423 return BLK_STS_IOERR
;
2426 if (blk_queue_io_stat(q
))
2427 blk_account_io_start(rq
, true);
2429 * Since we have a scheduler attached on the top device,
2430 * bypass a potential scheduler on the bottom device for
2433 blk_mq_request_bypass_insert(rq
, true);
2437 spin_lock_irqsave(q
->queue_lock
, flags
);
2438 if (unlikely(blk_queue_dying(q
))) {
2439 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2440 return BLK_STS_IOERR
;
2444 * Submitting request must be dequeued before calling this function
2445 * because it will be linked to another request_queue
2447 BUG_ON(blk_queued_rq(rq
));
2449 if (op_is_flush(rq
->cmd_flags
))
2450 where
= ELEVATOR_INSERT_FLUSH
;
2452 add_acct_request(q
, rq
, where
);
2453 if (where
== ELEVATOR_INSERT_FLUSH
)
2455 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2459 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
2462 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2463 * @rq: request to examine
2466 * A request could be merge of IOs which require different failure
2467 * handling. This function determines the number of bytes which
2468 * can be failed from the beginning of the request without
2469 * crossing into area which need to be retried further.
2472 * The number of bytes to fail.
2474 unsigned int blk_rq_err_bytes(const struct request
*rq
)
2476 unsigned int ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
2477 unsigned int bytes
= 0;
2480 if (!(rq
->rq_flags
& RQF_MIXED_MERGE
))
2481 return blk_rq_bytes(rq
);
2484 * Currently the only 'mixing' which can happen is between
2485 * different fastfail types. We can safely fail portions
2486 * which have all the failfast bits that the first one has -
2487 * the ones which are at least as eager to fail as the first
2490 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
2491 if ((bio
->bi_opf
& ff
) != ff
)
2493 bytes
+= bio
->bi_iter
.bi_size
;
2496 /* this could lead to infinite loop */
2497 BUG_ON(blk_rq_bytes(rq
) && !bytes
);
2500 EXPORT_SYMBOL_GPL(blk_rq_err_bytes
);
2502 void blk_account_io_completion(struct request
*req
, unsigned int bytes
)
2504 if (blk_do_io_stat(req
)) {
2505 const int rw
= rq_data_dir(req
);
2506 struct hd_struct
*part
;
2509 cpu
= part_stat_lock();
2511 part_stat_add(cpu
, part
, sectors
[rw
], bytes
>> 9);
2516 void blk_account_io_done(struct request
*req
)
2519 * Account IO completion. flush_rq isn't accounted as a
2520 * normal IO on queueing nor completion. Accounting the
2521 * containing request is enough.
2523 if (blk_do_io_stat(req
) && !(req
->rq_flags
& RQF_FLUSH_SEQ
)) {
2524 unsigned long duration
= jiffies
- req
->start_time
;
2525 const int rw
= rq_data_dir(req
);
2526 struct hd_struct
*part
;
2529 cpu
= part_stat_lock();
2532 part_stat_inc(cpu
, part
, ios
[rw
]);
2533 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
2534 part_round_stats(req
->q
, cpu
, part
);
2535 part_dec_in_flight(req
->q
, part
, rw
);
2537 hd_struct_put(part
);
2544 * Don't process normal requests when queue is suspended
2545 * or in the process of suspending/resuming
2547 static bool blk_pm_allow_request(struct request
*rq
)
2549 switch (rq
->q
->rpm_status
) {
2551 case RPM_SUSPENDING
:
2552 return rq
->rq_flags
& RQF_PM
;
2560 static bool blk_pm_allow_request(struct request
*rq
)
2566 void blk_account_io_start(struct request
*rq
, bool new_io
)
2568 struct hd_struct
*part
;
2569 int rw
= rq_data_dir(rq
);
2572 if (!blk_do_io_stat(rq
))
2575 cpu
= part_stat_lock();
2579 part_stat_inc(cpu
, part
, merges
[rw
]);
2581 part
= disk_map_sector_rcu(rq
->rq_disk
, blk_rq_pos(rq
));
2582 if (!hd_struct_try_get(part
)) {
2584 * The partition is already being removed,
2585 * the request will be accounted on the disk only
2587 * We take a reference on disk->part0 although that
2588 * partition will never be deleted, so we can treat
2589 * it as any other partition.
2591 part
= &rq
->rq_disk
->part0
;
2592 hd_struct_get(part
);
2594 part_round_stats(rq
->q
, cpu
, part
);
2595 part_inc_in_flight(rq
->q
, part
, rw
);
2602 static struct request
*elv_next_request(struct request_queue
*q
)
2605 struct blk_flush_queue
*fq
= blk_get_flush_queue(q
, NULL
);
2607 WARN_ON_ONCE(q
->mq_ops
);
2610 list_for_each_entry(rq
, &q
->queue_head
, queuelist
) {
2611 if (blk_pm_allow_request(rq
))
2614 if (rq
->rq_flags
& RQF_SOFTBARRIER
)
2619 * Flush request is running and flush request isn't queueable
2620 * in the drive, we can hold the queue till flush request is
2621 * finished. Even we don't do this, driver can't dispatch next
2622 * requests and will requeue them. And this can improve
2623 * throughput too. For example, we have request flush1, write1,
2624 * flush 2. flush1 is dispatched, then queue is hold, write1
2625 * isn't inserted to queue. After flush1 is finished, flush2
2626 * will be dispatched. Since disk cache is already clean,
2627 * flush2 will be finished very soon, so looks like flush2 is
2629 * Since the queue is hold, a flag is set to indicate the queue
2630 * should be restarted later. Please see flush_end_io() for
2633 if (fq
->flush_pending_idx
!= fq
->flush_running_idx
&&
2634 !queue_flush_queueable(q
)) {
2635 fq
->flush_queue_delayed
= 1;
2638 if (unlikely(blk_queue_bypass(q
)) ||
2639 !q
->elevator
->type
->ops
.sq
.elevator_dispatch_fn(q
, 0))
2645 * blk_peek_request - peek at the top of a request queue
2646 * @q: request queue to peek at
2649 * Return the request at the top of @q. The returned request
2650 * should be started using blk_start_request() before LLD starts
2654 * Pointer to the request at the top of @q if available. Null
2657 struct request
*blk_peek_request(struct request_queue
*q
)
2662 lockdep_assert_held(q
->queue_lock
);
2663 WARN_ON_ONCE(q
->mq_ops
);
2665 while ((rq
= elv_next_request(q
)) != NULL
) {
2666 if (!(rq
->rq_flags
& RQF_STARTED
)) {
2668 * This is the first time the device driver
2669 * sees this request (possibly after
2670 * requeueing). Notify IO scheduler.
2672 if (rq
->rq_flags
& RQF_SORTED
)
2673 elv_activate_rq(q
, rq
);
2676 * just mark as started even if we don't start
2677 * it, a request that has been delayed should
2678 * not be passed by new incoming requests
2680 rq
->rq_flags
|= RQF_STARTED
;
2681 trace_block_rq_issue(q
, rq
);
2684 if (!q
->boundary_rq
|| q
->boundary_rq
== rq
) {
2685 q
->end_sector
= rq_end_sector(rq
);
2686 q
->boundary_rq
= NULL
;
2689 if (rq
->rq_flags
& RQF_DONTPREP
)
2692 if (q
->dma_drain_size
&& blk_rq_bytes(rq
)) {
2694 * make sure space for the drain appears we
2695 * know we can do this because max_hw_segments
2696 * has been adjusted to be one fewer than the
2699 rq
->nr_phys_segments
++;
2705 ret
= q
->prep_rq_fn(q
, rq
);
2706 if (ret
== BLKPREP_OK
) {
2708 } else if (ret
== BLKPREP_DEFER
) {
2710 * the request may have been (partially) prepped.
2711 * we need to keep this request in the front to
2712 * avoid resource deadlock. RQF_STARTED will
2713 * prevent other fs requests from passing this one.
2715 if (q
->dma_drain_size
&& blk_rq_bytes(rq
) &&
2716 !(rq
->rq_flags
& RQF_DONTPREP
)) {
2718 * remove the space for the drain we added
2719 * so that we don't add it again
2721 --rq
->nr_phys_segments
;
2726 } else if (ret
== BLKPREP_KILL
|| ret
== BLKPREP_INVALID
) {
2727 rq
->rq_flags
|= RQF_QUIET
;
2729 * Mark this request as started so we don't trigger
2730 * any debug logic in the end I/O path.
2732 blk_start_request(rq
);
2733 __blk_end_request_all(rq
, ret
== BLKPREP_INVALID
?
2734 BLK_STS_TARGET
: BLK_STS_IOERR
);
2736 printk(KERN_ERR
"%s: bad return=%d\n", __func__
, ret
);
2743 EXPORT_SYMBOL(blk_peek_request
);
2745 static void blk_dequeue_request(struct request
*rq
)
2747 struct request_queue
*q
= rq
->q
;
2749 BUG_ON(list_empty(&rq
->queuelist
));
2750 BUG_ON(ELV_ON_HASH(rq
));
2752 list_del_init(&rq
->queuelist
);
2755 * the time frame between a request being removed from the lists
2756 * and to it is freed is accounted as io that is in progress at
2759 if (blk_account_rq(rq
)) {
2760 q
->in_flight
[rq_is_sync(rq
)]++;
2761 set_io_start_time_ns(rq
);
2766 * blk_start_request - start request processing on the driver
2767 * @req: request to dequeue
2770 * Dequeue @req and start timeout timer on it. This hands off the
2771 * request to the driver.
2773 void blk_start_request(struct request
*req
)
2775 lockdep_assert_held(req
->q
->queue_lock
);
2776 WARN_ON_ONCE(req
->q
->mq_ops
);
2778 blk_dequeue_request(req
);
2780 if (test_bit(QUEUE_FLAG_STATS
, &req
->q
->queue_flags
)) {
2781 blk_stat_set_issue(&req
->issue_stat
, blk_rq_sectors(req
));
2782 req
->rq_flags
|= RQF_STATS
;
2783 wbt_issue(req
->q
->rq_wb
, &req
->issue_stat
);
2786 BUG_ON(test_bit(REQ_ATOM_COMPLETE
, &req
->atomic_flags
));
2789 EXPORT_SYMBOL(blk_start_request
);
2792 * blk_fetch_request - fetch a request from a request queue
2793 * @q: request queue to fetch a request from
2796 * Return the request at the top of @q. The request is started on
2797 * return and LLD can start processing it immediately.
2800 * Pointer to the request at the top of @q if available. Null
2803 struct request
*blk_fetch_request(struct request_queue
*q
)
2807 lockdep_assert_held(q
->queue_lock
);
2808 WARN_ON_ONCE(q
->mq_ops
);
2810 rq
= blk_peek_request(q
);
2812 blk_start_request(rq
);
2815 EXPORT_SYMBOL(blk_fetch_request
);
2818 * Steal bios from a request and add them to a bio list.
2819 * The request must not have been partially completed before.
2821 void blk_steal_bios(struct bio_list
*list
, struct request
*rq
)
2825 list
->tail
->bi_next
= rq
->bio
;
2827 list
->head
= rq
->bio
;
2828 list
->tail
= rq
->biotail
;
2836 EXPORT_SYMBOL_GPL(blk_steal_bios
);
2839 * blk_update_request - Special helper function for request stacking drivers
2840 * @req: the request being processed
2841 * @error: block status code
2842 * @nr_bytes: number of bytes to complete @req
2845 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2846 * the request structure even if @req doesn't have leftover.
2847 * If @req has leftover, sets it up for the next range of segments.
2849 * This special helper function is only for request stacking drivers
2850 * (e.g. request-based dm) so that they can handle partial completion.
2851 * Actual device drivers should use blk_end_request instead.
2853 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2854 * %false return from this function.
2857 * %false - this request doesn't have any more data
2858 * %true - this request has more data
2860 bool blk_update_request(struct request
*req
, blk_status_t error
,
2861 unsigned int nr_bytes
)
2865 trace_block_rq_complete(req
, blk_status_to_errno(error
), nr_bytes
);
2870 if (unlikely(error
&& !blk_rq_is_passthrough(req
) &&
2871 !(req
->rq_flags
& RQF_QUIET
)))
2872 print_req_error(req
, error
);
2874 blk_account_io_completion(req
, nr_bytes
);
2878 struct bio
*bio
= req
->bio
;
2879 unsigned bio_bytes
= min(bio
->bi_iter
.bi_size
, nr_bytes
);
2881 if (bio_bytes
== bio
->bi_iter
.bi_size
)
2882 req
->bio
= bio
->bi_next
;
2884 /* Completion has already been traced */
2885 bio_clear_flag(bio
, BIO_TRACE_COMPLETION
);
2886 req_bio_endio(req
, bio
, bio_bytes
, error
);
2888 total_bytes
+= bio_bytes
;
2889 nr_bytes
-= bio_bytes
;
2900 * Reset counters so that the request stacking driver
2901 * can find how many bytes remain in the request
2904 req
->__data_len
= 0;
2908 req
->__data_len
-= total_bytes
;
2910 /* update sector only for requests with clear definition of sector */
2911 if (!blk_rq_is_passthrough(req
))
2912 req
->__sector
+= total_bytes
>> 9;
2914 /* mixed attributes always follow the first bio */
2915 if (req
->rq_flags
& RQF_MIXED_MERGE
) {
2916 req
->cmd_flags
&= ~REQ_FAILFAST_MASK
;
2917 req
->cmd_flags
|= req
->bio
->bi_opf
& REQ_FAILFAST_MASK
;
2920 if (!(req
->rq_flags
& RQF_SPECIAL_PAYLOAD
)) {
2922 * If total number of sectors is less than the first segment
2923 * size, something has gone terribly wrong.
2925 if (blk_rq_bytes(req
) < blk_rq_cur_bytes(req
)) {
2926 blk_dump_rq_flags(req
, "request botched");
2927 req
->__data_len
= blk_rq_cur_bytes(req
);
2930 /* recalculate the number of segments */
2931 blk_recalc_rq_segments(req
);
2936 EXPORT_SYMBOL_GPL(blk_update_request
);
2938 static bool blk_update_bidi_request(struct request
*rq
, blk_status_t error
,
2939 unsigned int nr_bytes
,
2940 unsigned int bidi_bytes
)
2942 if (blk_update_request(rq
, error
, nr_bytes
))
2945 /* Bidi request must be completed as a whole */
2946 if (unlikely(blk_bidi_rq(rq
)) &&
2947 blk_update_request(rq
->next_rq
, error
, bidi_bytes
))
2950 if (blk_queue_add_random(rq
->q
))
2951 add_disk_randomness(rq
->rq_disk
);
2957 * blk_unprep_request - unprepare a request
2960 * This function makes a request ready for complete resubmission (or
2961 * completion). It happens only after all error handling is complete,
2962 * so represents the appropriate moment to deallocate any resources
2963 * that were allocated to the request in the prep_rq_fn. The queue
2964 * lock is held when calling this.
2966 void blk_unprep_request(struct request
*req
)
2968 struct request_queue
*q
= req
->q
;
2970 req
->rq_flags
&= ~RQF_DONTPREP
;
2971 if (q
->unprep_rq_fn
)
2972 q
->unprep_rq_fn(q
, req
);
2974 EXPORT_SYMBOL_GPL(blk_unprep_request
);
2976 void blk_finish_request(struct request
*req
, blk_status_t error
)
2978 struct request_queue
*q
= req
->q
;
2980 lockdep_assert_held(req
->q
->queue_lock
);
2981 WARN_ON_ONCE(q
->mq_ops
);
2983 if (req
->rq_flags
& RQF_STATS
)
2986 if (req
->rq_flags
& RQF_QUEUED
)
2987 blk_queue_end_tag(q
, req
);
2989 BUG_ON(blk_queued_rq(req
));
2991 if (unlikely(laptop_mode
) && !blk_rq_is_passthrough(req
))
2992 laptop_io_completion(req
->q
->backing_dev_info
);
2994 blk_delete_timer(req
);
2996 if (req
->rq_flags
& RQF_DONTPREP
)
2997 blk_unprep_request(req
);
2999 blk_account_io_done(req
);
3002 wbt_done(req
->q
->rq_wb
, &req
->issue_stat
);
3003 req
->end_io(req
, error
);
3005 if (blk_bidi_rq(req
))
3006 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
3008 __blk_put_request(q
, req
);
3011 EXPORT_SYMBOL(blk_finish_request
);
3014 * blk_end_bidi_request - Complete a bidi request
3015 * @rq: the request to complete
3016 * @error: block status code
3017 * @nr_bytes: number of bytes to complete @rq
3018 * @bidi_bytes: number of bytes to complete @rq->next_rq
3021 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
3022 * Drivers that supports bidi can safely call this member for any
3023 * type of request, bidi or uni. In the later case @bidi_bytes is
3027 * %false - we are done with this request
3028 * %true - still buffers pending for this request
3030 static bool blk_end_bidi_request(struct request
*rq
, blk_status_t error
,
3031 unsigned int nr_bytes
, unsigned int bidi_bytes
)
3033 struct request_queue
*q
= rq
->q
;
3034 unsigned long flags
;
3036 WARN_ON_ONCE(q
->mq_ops
);
3038 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
3041 spin_lock_irqsave(q
->queue_lock
, flags
);
3042 blk_finish_request(rq
, error
);
3043 spin_unlock_irqrestore(q
->queue_lock
, flags
);
3049 * __blk_end_bidi_request - Complete a bidi request with queue lock held
3050 * @rq: the request to complete
3051 * @error: block status code
3052 * @nr_bytes: number of bytes to complete @rq
3053 * @bidi_bytes: number of bytes to complete @rq->next_rq
3056 * Identical to blk_end_bidi_request() except that queue lock is
3057 * assumed to be locked on entry and remains so on return.
3060 * %false - we are done with this request
3061 * %true - still buffers pending for this request
3063 static bool __blk_end_bidi_request(struct request
*rq
, blk_status_t error
,
3064 unsigned int nr_bytes
, unsigned int bidi_bytes
)
3066 lockdep_assert_held(rq
->q
->queue_lock
);
3067 WARN_ON_ONCE(rq
->q
->mq_ops
);
3069 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
3072 blk_finish_request(rq
, error
);
3078 * blk_end_request - Helper function for drivers to complete the request.
3079 * @rq: the request being processed
3080 * @error: block status code
3081 * @nr_bytes: number of bytes to complete
3084 * Ends I/O on a number of bytes attached to @rq.
3085 * If @rq has leftover, sets it up for the next range of segments.
3088 * %false - we are done with this request
3089 * %true - still buffers pending for this request
3091 bool blk_end_request(struct request
*rq
, blk_status_t error
,
3092 unsigned int nr_bytes
)
3094 WARN_ON_ONCE(rq
->q
->mq_ops
);
3095 return blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
3097 EXPORT_SYMBOL(blk_end_request
);
3100 * blk_end_request_all - Helper function for drives to finish the request.
3101 * @rq: the request to finish
3102 * @error: block status code
3105 * Completely finish @rq.
3107 void blk_end_request_all(struct request
*rq
, blk_status_t error
)
3110 unsigned int bidi_bytes
= 0;
3112 if (unlikely(blk_bidi_rq(rq
)))
3113 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
3115 pending
= blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
3118 EXPORT_SYMBOL(blk_end_request_all
);
3121 * __blk_end_request - Helper function for drivers to complete the request.
3122 * @rq: the request being processed
3123 * @error: block status code
3124 * @nr_bytes: number of bytes to complete
3127 * Must be called with queue lock held unlike blk_end_request().
3130 * %false - we are done with this request
3131 * %true - still buffers pending for this request
3133 bool __blk_end_request(struct request
*rq
, blk_status_t error
,
3134 unsigned int nr_bytes
)
3136 lockdep_assert_held(rq
->q
->queue_lock
);
3137 WARN_ON_ONCE(rq
->q
->mq_ops
);
3139 return __blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
3141 EXPORT_SYMBOL(__blk_end_request
);
3144 * __blk_end_request_all - Helper function for drives to finish the request.
3145 * @rq: the request to finish
3146 * @error: block status code
3149 * Completely finish @rq. Must be called with queue lock held.
3151 void __blk_end_request_all(struct request
*rq
, blk_status_t error
)
3154 unsigned int bidi_bytes
= 0;
3156 lockdep_assert_held(rq
->q
->queue_lock
);
3157 WARN_ON_ONCE(rq
->q
->mq_ops
);
3159 if (unlikely(blk_bidi_rq(rq
)))
3160 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
3162 pending
= __blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
3165 EXPORT_SYMBOL(__blk_end_request_all
);
3168 * __blk_end_request_cur - Helper function to finish the current request chunk.
3169 * @rq: the request to finish the current chunk for
3170 * @error: block status code
3173 * Complete the current consecutively mapped chunk from @rq. Must
3174 * be called with queue lock held.
3177 * %false - we are done with this request
3178 * %true - still buffers pending for this request
3180 bool __blk_end_request_cur(struct request
*rq
, blk_status_t error
)
3182 return __blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
3184 EXPORT_SYMBOL(__blk_end_request_cur
);
3186 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
3189 if (bio_has_data(bio
))
3190 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
3192 rq
->__data_len
= bio
->bi_iter
.bi_size
;
3193 rq
->bio
= rq
->biotail
= bio
;
3196 rq
->rq_disk
= bio
->bi_disk
;
3199 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
3201 * rq_flush_dcache_pages - Helper function to flush all pages in a request
3202 * @rq: the request to be flushed
3205 * Flush all pages in @rq.
3207 void rq_flush_dcache_pages(struct request
*rq
)
3209 struct req_iterator iter
;
3210 struct bio_vec bvec
;
3212 rq_for_each_segment(bvec
, rq
, iter
)
3213 flush_dcache_page(bvec
.bv_page
);
3215 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages
);
3219 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
3220 * @q : the queue of the device being checked
3223 * Check if underlying low-level drivers of a device are busy.
3224 * If the drivers want to export their busy state, they must set own
3225 * exporting function using blk_queue_lld_busy() first.
3227 * Basically, this function is used only by request stacking drivers
3228 * to stop dispatching requests to underlying devices when underlying
3229 * devices are busy. This behavior helps more I/O merging on the queue
3230 * of the request stacking driver and prevents I/O throughput regression
3231 * on burst I/O load.
3234 * 0 - Not busy (The request stacking driver should dispatch request)
3235 * 1 - Busy (The request stacking driver should stop dispatching request)
3237 int blk_lld_busy(struct request_queue
*q
)
3240 return q
->lld_busy_fn(q
);
3244 EXPORT_SYMBOL_GPL(blk_lld_busy
);
3247 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
3248 * @rq: the clone request to be cleaned up
3251 * Free all bios in @rq for a cloned request.
3253 void blk_rq_unprep_clone(struct request
*rq
)
3257 while ((bio
= rq
->bio
) != NULL
) {
3258 rq
->bio
= bio
->bi_next
;
3263 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone
);
3266 * Copy attributes of the original request to the clone request.
3267 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3269 static void __blk_rq_prep_clone(struct request
*dst
, struct request
*src
)
3271 dst
->cpu
= src
->cpu
;
3272 dst
->__sector
= blk_rq_pos(src
);
3273 dst
->__data_len
= blk_rq_bytes(src
);
3274 dst
->nr_phys_segments
= src
->nr_phys_segments
;
3275 dst
->ioprio
= src
->ioprio
;
3276 dst
->extra_len
= src
->extra_len
;
3280 * blk_rq_prep_clone - Helper function to setup clone request
3281 * @rq: the request to be setup
3282 * @rq_src: original request to be cloned
3283 * @bs: bio_set that bios for clone are allocated from
3284 * @gfp_mask: memory allocation mask for bio
3285 * @bio_ctr: setup function to be called for each clone bio.
3286 * Returns %0 for success, non %0 for failure.
3287 * @data: private data to be passed to @bio_ctr
3290 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3291 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3292 * are not copied, and copying such parts is the caller's responsibility.
3293 * Also, pages which the original bios are pointing to are not copied
3294 * and the cloned bios just point same pages.
3295 * So cloned bios must be completed before original bios, which means
3296 * the caller must complete @rq before @rq_src.
3298 int blk_rq_prep_clone(struct request
*rq
, struct request
*rq_src
,
3299 struct bio_set
*bs
, gfp_t gfp_mask
,
3300 int (*bio_ctr
)(struct bio
*, struct bio
*, void *),
3303 struct bio
*bio
, *bio_src
;
3308 __rq_for_each_bio(bio_src
, rq_src
) {
3309 bio
= bio_clone_fast(bio_src
, gfp_mask
, bs
);
3313 if (bio_ctr
&& bio_ctr(bio
, bio_src
, data
))
3317 rq
->biotail
->bi_next
= bio
;
3320 rq
->bio
= rq
->biotail
= bio
;
3323 __blk_rq_prep_clone(rq
, rq_src
);
3330 blk_rq_unprep_clone(rq
);
3334 EXPORT_SYMBOL_GPL(blk_rq_prep_clone
);
3336 int kblockd_schedule_work(struct work_struct
*work
)
3338 return queue_work(kblockd_workqueue
, work
);
3340 EXPORT_SYMBOL(kblockd_schedule_work
);
3342 int kblockd_schedule_work_on(int cpu
, struct work_struct
*work
)
3344 return queue_work_on(cpu
, kblockd_workqueue
, work
);
3346 EXPORT_SYMBOL(kblockd_schedule_work_on
);
3348 int kblockd_mod_delayed_work_on(int cpu
, struct delayed_work
*dwork
,
3349 unsigned long delay
)
3351 return mod_delayed_work_on(cpu
, kblockd_workqueue
, dwork
, delay
);
3353 EXPORT_SYMBOL(kblockd_mod_delayed_work_on
);
3355 int kblockd_schedule_delayed_work(struct delayed_work
*dwork
,
3356 unsigned long delay
)
3358 return queue_delayed_work(kblockd_workqueue
, dwork
, delay
);
3360 EXPORT_SYMBOL(kblockd_schedule_delayed_work
);
3362 int kblockd_schedule_delayed_work_on(int cpu
, struct delayed_work
*dwork
,
3363 unsigned long delay
)
3365 return queue_delayed_work_on(cpu
, kblockd_workqueue
, dwork
, delay
);
3367 EXPORT_SYMBOL(kblockd_schedule_delayed_work_on
);
3370 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3371 * @plug: The &struct blk_plug that needs to be initialized
3374 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3375 * pending I/O should the task end up blocking between blk_start_plug() and
3376 * blk_finish_plug(). This is important from a performance perspective, but
3377 * also ensures that we don't deadlock. For instance, if the task is blocking
3378 * for a memory allocation, memory reclaim could end up wanting to free a
3379 * page belonging to that request that is currently residing in our private
3380 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3381 * this kind of deadlock.
3383 void blk_start_plug(struct blk_plug
*plug
)
3385 struct task_struct
*tsk
= current
;
3388 * If this is a nested plug, don't actually assign it.
3393 INIT_LIST_HEAD(&plug
->list
);
3394 INIT_LIST_HEAD(&plug
->mq_list
);
3395 INIT_LIST_HEAD(&plug
->cb_list
);
3397 * Store ordering should not be needed here, since a potential
3398 * preempt will imply a full memory barrier
3402 EXPORT_SYMBOL(blk_start_plug
);
3404 static int plug_rq_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
3406 struct request
*rqa
= container_of(a
, struct request
, queuelist
);
3407 struct request
*rqb
= container_of(b
, struct request
, queuelist
);
3409 return !(rqa
->q
< rqb
->q
||
3410 (rqa
->q
== rqb
->q
&& blk_rq_pos(rqa
) < blk_rq_pos(rqb
)));
3414 * If 'from_schedule' is true, then postpone the dispatch of requests
3415 * until a safe kblockd context. We due this to avoid accidental big
3416 * additional stack usage in driver dispatch, in places where the originally
3417 * plugger did not intend it.
3419 static void queue_unplugged(struct request_queue
*q
, unsigned int depth
,
3421 __releases(q
->queue_lock
)
3423 lockdep_assert_held(q
->queue_lock
);
3425 trace_block_unplug(q
, depth
, !from_schedule
);
3428 blk_run_queue_async(q
);
3431 spin_unlock(q
->queue_lock
);
3434 static void flush_plug_callbacks(struct blk_plug
*plug
, bool from_schedule
)
3436 LIST_HEAD(callbacks
);
3438 while (!list_empty(&plug
->cb_list
)) {
3439 list_splice_init(&plug
->cb_list
, &callbacks
);
3441 while (!list_empty(&callbacks
)) {
3442 struct blk_plug_cb
*cb
= list_first_entry(&callbacks
,
3445 list_del(&cb
->list
);
3446 cb
->callback(cb
, from_schedule
);
3451 struct blk_plug_cb
*blk_check_plugged(blk_plug_cb_fn unplug
, void *data
,
3454 struct blk_plug
*plug
= current
->plug
;
3455 struct blk_plug_cb
*cb
;
3460 list_for_each_entry(cb
, &plug
->cb_list
, list
)
3461 if (cb
->callback
== unplug
&& cb
->data
== data
)
3464 /* Not currently on the callback list */
3465 BUG_ON(size
< sizeof(*cb
));
3466 cb
= kzalloc(size
, GFP_ATOMIC
);
3469 cb
->callback
= unplug
;
3470 list_add(&cb
->list
, &plug
->cb_list
);
3474 EXPORT_SYMBOL(blk_check_plugged
);
3476 void blk_flush_plug_list(struct blk_plug
*plug
, bool from_schedule
)
3478 struct request_queue
*q
;
3479 unsigned long flags
;
3484 flush_plug_callbacks(plug
, from_schedule
);
3486 if (!list_empty(&plug
->mq_list
))
3487 blk_mq_flush_plug_list(plug
, from_schedule
);
3489 if (list_empty(&plug
->list
))
3492 list_splice_init(&plug
->list
, &list
);
3494 list_sort(NULL
, &list
, plug_rq_cmp
);
3500 * Save and disable interrupts here, to avoid doing it for every
3501 * queue lock we have to take.
3503 local_irq_save(flags
);
3504 while (!list_empty(&list
)) {
3505 rq
= list_entry_rq(list
.next
);
3506 list_del_init(&rq
->queuelist
);
3510 * This drops the queue lock
3513 queue_unplugged(q
, depth
, from_schedule
);
3516 spin_lock(q
->queue_lock
);
3520 * Short-circuit if @q is dead
3522 if (unlikely(blk_queue_dying(q
))) {
3523 __blk_end_request_all(rq
, BLK_STS_IOERR
);
3528 * rq is already accounted, so use raw insert
3530 if (op_is_flush(rq
->cmd_flags
))
3531 __elv_add_request(q
, rq
, ELEVATOR_INSERT_FLUSH
);
3533 __elv_add_request(q
, rq
, ELEVATOR_INSERT_SORT_MERGE
);
3539 * This drops the queue lock
3542 queue_unplugged(q
, depth
, from_schedule
);
3544 local_irq_restore(flags
);
3547 void blk_finish_plug(struct blk_plug
*plug
)
3549 if (plug
!= current
->plug
)
3551 blk_flush_plug_list(plug
, false);
3553 current
->plug
= NULL
;
3555 EXPORT_SYMBOL(blk_finish_plug
);
3559 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3560 * @q: the queue of the device
3561 * @dev: the device the queue belongs to
3564 * Initialize runtime-PM-related fields for @q and start auto suspend for
3565 * @dev. Drivers that want to take advantage of request-based runtime PM
3566 * should call this function after @dev has been initialized, and its
3567 * request queue @q has been allocated, and runtime PM for it can not happen
3568 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3569 * cases, driver should call this function before any I/O has taken place.
3571 * This function takes care of setting up using auto suspend for the device,
3572 * the autosuspend delay is set to -1 to make runtime suspend impossible
3573 * until an updated value is either set by user or by driver. Drivers do
3574 * not need to touch other autosuspend settings.
3576 * The block layer runtime PM is request based, so only works for drivers
3577 * that use request as their IO unit instead of those directly use bio's.
3579 void blk_pm_runtime_init(struct request_queue
*q
, struct device
*dev
)
3581 /* not support for RQF_PM and ->rpm_status in blk-mq yet */
3586 q
->rpm_status
= RPM_ACTIVE
;
3587 pm_runtime_set_autosuspend_delay(q
->dev
, -1);
3588 pm_runtime_use_autosuspend(q
->dev
);
3590 EXPORT_SYMBOL(blk_pm_runtime_init
);
3593 * blk_pre_runtime_suspend - Pre runtime suspend check
3594 * @q: the queue of the device
3597 * This function will check if runtime suspend is allowed for the device
3598 * by examining if there are any requests pending in the queue. If there
3599 * are requests pending, the device can not be runtime suspended; otherwise,
3600 * the queue's status will be updated to SUSPENDING and the driver can
3601 * proceed to suspend the device.
3603 * For the not allowed case, we mark last busy for the device so that
3604 * runtime PM core will try to autosuspend it some time later.
3606 * This function should be called near the start of the device's
3607 * runtime_suspend callback.
3610 * 0 - OK to runtime suspend the device
3611 * -EBUSY - Device should not be runtime suspended
3613 int blk_pre_runtime_suspend(struct request_queue
*q
)
3620 spin_lock_irq(q
->queue_lock
);
3621 if (q
->nr_pending
) {
3623 pm_runtime_mark_last_busy(q
->dev
);
3625 q
->rpm_status
= RPM_SUSPENDING
;
3627 spin_unlock_irq(q
->queue_lock
);
3630 EXPORT_SYMBOL(blk_pre_runtime_suspend
);
3633 * blk_post_runtime_suspend - Post runtime suspend processing
3634 * @q: the queue of the device
3635 * @err: return value of the device's runtime_suspend function
3638 * Update the queue's runtime status according to the return value of the
3639 * device's runtime suspend function and mark last busy for the device so
3640 * that PM core will try to auto suspend the device at a later time.
3642 * This function should be called near the end of the device's
3643 * runtime_suspend callback.
3645 void blk_post_runtime_suspend(struct request_queue
*q
, int err
)
3650 spin_lock_irq(q
->queue_lock
);
3652 q
->rpm_status
= RPM_SUSPENDED
;
3654 q
->rpm_status
= RPM_ACTIVE
;
3655 pm_runtime_mark_last_busy(q
->dev
);
3657 spin_unlock_irq(q
->queue_lock
);
3659 EXPORT_SYMBOL(blk_post_runtime_suspend
);
3662 * blk_pre_runtime_resume - Pre runtime resume processing
3663 * @q: the queue of the device
3666 * Update the queue's runtime status to RESUMING in preparation for the
3667 * runtime resume of the device.
3669 * This function should be called near the start of the device's
3670 * runtime_resume callback.
3672 void blk_pre_runtime_resume(struct request_queue
*q
)
3677 spin_lock_irq(q
->queue_lock
);
3678 q
->rpm_status
= RPM_RESUMING
;
3679 spin_unlock_irq(q
->queue_lock
);
3681 EXPORT_SYMBOL(blk_pre_runtime_resume
);
3684 * blk_post_runtime_resume - Post runtime resume processing
3685 * @q: the queue of the device
3686 * @err: return value of the device's runtime_resume function
3689 * Update the queue's runtime status according to the return value of the
3690 * device's runtime_resume function. If it is successfully resumed, process
3691 * the requests that are queued into the device's queue when it is resuming
3692 * and then mark last busy and initiate autosuspend for it.
3694 * This function should be called near the end of the device's
3695 * runtime_resume callback.
3697 void blk_post_runtime_resume(struct request_queue
*q
, int err
)
3702 spin_lock_irq(q
->queue_lock
);
3704 q
->rpm_status
= RPM_ACTIVE
;
3706 pm_runtime_mark_last_busy(q
->dev
);
3707 pm_request_autosuspend(q
->dev
);
3709 q
->rpm_status
= RPM_SUSPENDED
;
3711 spin_unlock_irq(q
->queue_lock
);
3713 EXPORT_SYMBOL(blk_post_runtime_resume
);
3716 * blk_set_runtime_active - Force runtime status of the queue to be active
3717 * @q: the queue of the device
3719 * If the device is left runtime suspended during system suspend the resume
3720 * hook typically resumes the device and corrects runtime status
3721 * accordingly. However, that does not affect the queue runtime PM status
3722 * which is still "suspended". This prevents processing requests from the
3725 * This function can be used in driver's resume hook to correct queue
3726 * runtime PM status and re-enable peeking requests from the queue. It
3727 * should be called before first request is added to the queue.
3729 void blk_set_runtime_active(struct request_queue
*q
)
3731 spin_lock_irq(q
->queue_lock
);
3732 q
->rpm_status
= RPM_ACTIVE
;
3733 pm_runtime_mark_last_busy(q
->dev
);
3734 pm_request_autosuspend(q
->dev
);
3735 spin_unlock_irq(q
->queue_lock
);
3737 EXPORT_SYMBOL(blk_set_runtime_active
);
3740 int __init
blk_dev_init(void)
3742 BUILD_BUG_ON(REQ_OP_LAST
>= (1 << REQ_OP_BITS
));
3743 BUILD_BUG_ON(REQ_OP_BITS
+ REQ_FLAG_BITS
> 8 *
3744 FIELD_SIZEOF(struct request
, cmd_flags
));
3745 BUILD_BUG_ON(REQ_OP_BITS
+ REQ_FLAG_BITS
> 8 *
3746 FIELD_SIZEOF(struct bio
, bi_opf
));
3748 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3749 kblockd_workqueue
= alloc_workqueue("kblockd",
3750 WQ_MEM_RECLAIM
| WQ_HIGHPRI
, 0);
3751 if (!kblockd_workqueue
)
3752 panic("Failed to create kblockd\n");
3754 request_cachep
= kmem_cache_create("blkdev_requests",
3755 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
3757 blk_requestq_cachep
= kmem_cache_create("request_queue",
3758 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);
3760 #ifdef CONFIG_DEBUG_FS
3761 blk_debugfs_root
= debugfs_create_dir("block", NULL
);