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_set_preempt_only - set QUEUE_FLAG_PREEMPT_ONLY
353 * @q: request queue pointer
355 * Returns the previous value of the PREEMPT_ONLY flag - 0 if the flag was not
356 * set and 1 if the flag was already set.
358 int blk_set_preempt_only(struct request_queue
*q
)
363 spin_lock_irqsave(q
->queue_lock
, flags
);
364 res
= queue_flag_test_and_set(QUEUE_FLAG_PREEMPT_ONLY
, q
);
365 spin_unlock_irqrestore(q
->queue_lock
, flags
);
369 EXPORT_SYMBOL_GPL(blk_set_preempt_only
);
371 void blk_clear_preempt_only(struct request_queue
*q
)
375 spin_lock_irqsave(q
->queue_lock
, flags
);
376 queue_flag_clear(QUEUE_FLAG_PREEMPT_ONLY
, q
);
377 spin_unlock_irqrestore(q
->queue_lock
, flags
);
379 EXPORT_SYMBOL_GPL(blk_clear_preempt_only
);
382 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
383 * @q: The queue to run
386 * Invoke request handling on a queue if there are any pending requests.
387 * May be used to restart request handling after a request has completed.
388 * This variant runs the queue whether or not the queue has been
389 * stopped. Must be called with the queue lock held and interrupts
390 * disabled. See also @blk_run_queue.
392 inline void __blk_run_queue_uncond(struct request_queue
*q
)
394 lockdep_assert_held(q
->queue_lock
);
395 WARN_ON_ONCE(q
->mq_ops
);
397 if (unlikely(blk_queue_dead(q
)))
401 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
402 * the queue lock internally. As a result multiple threads may be
403 * running such a request function concurrently. Keep track of the
404 * number of active request_fn invocations such that blk_drain_queue()
405 * can wait until all these request_fn calls have finished.
407 q
->request_fn_active
++;
409 q
->request_fn_active
--;
411 EXPORT_SYMBOL_GPL(__blk_run_queue_uncond
);
414 * __blk_run_queue - run a single device queue
415 * @q: The queue to run
418 * See @blk_run_queue.
420 void __blk_run_queue(struct request_queue
*q
)
422 lockdep_assert_held(q
->queue_lock
);
423 WARN_ON_ONCE(q
->mq_ops
);
425 if (unlikely(blk_queue_stopped(q
)))
428 __blk_run_queue_uncond(q
);
430 EXPORT_SYMBOL(__blk_run_queue
);
433 * blk_run_queue_async - run a single device queue in workqueue context
434 * @q: The queue to run
437 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
441 * Since it is not allowed to run q->delay_work after blk_cleanup_queue()
442 * has canceled q->delay_work, callers must hold the queue lock to avoid
443 * race conditions between blk_cleanup_queue() and blk_run_queue_async().
445 void blk_run_queue_async(struct request_queue
*q
)
447 lockdep_assert_held(q
->queue_lock
);
448 WARN_ON_ONCE(q
->mq_ops
);
450 if (likely(!blk_queue_stopped(q
) && !blk_queue_dead(q
)))
451 mod_delayed_work(kblockd_workqueue
, &q
->delay_work
, 0);
453 EXPORT_SYMBOL(blk_run_queue_async
);
456 * blk_run_queue - run a single device queue
457 * @q: The queue to run
460 * Invoke request handling on this queue, if it has pending work to do.
461 * May be used to restart queueing when a request has completed.
463 void blk_run_queue(struct request_queue
*q
)
467 WARN_ON_ONCE(q
->mq_ops
);
469 spin_lock_irqsave(q
->queue_lock
, flags
);
471 spin_unlock_irqrestore(q
->queue_lock
, flags
);
473 EXPORT_SYMBOL(blk_run_queue
);
475 void blk_put_queue(struct request_queue
*q
)
477 kobject_put(&q
->kobj
);
479 EXPORT_SYMBOL(blk_put_queue
);
482 * __blk_drain_queue - drain requests from request_queue
484 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
486 * Drain requests from @q. If @drain_all is set, all requests are drained.
487 * If not, only ELVPRIV requests are drained. The caller is responsible
488 * for ensuring that no new requests which need to be drained are queued.
490 static void __blk_drain_queue(struct request_queue
*q
, bool drain_all
)
491 __releases(q
->queue_lock
)
492 __acquires(q
->queue_lock
)
496 lockdep_assert_held(q
->queue_lock
);
497 WARN_ON_ONCE(q
->mq_ops
);
503 * The caller might be trying to drain @q before its
504 * elevator is initialized.
507 elv_drain_elevator(q
);
509 blkcg_drain_queue(q
);
512 * This function might be called on a queue which failed
513 * driver init after queue creation or is not yet fully
514 * active yet. Some drivers (e.g. fd and loop) get unhappy
515 * in such cases. Kick queue iff dispatch queue has
516 * something on it and @q has request_fn set.
518 if (!list_empty(&q
->queue_head
) && q
->request_fn
)
521 drain
|= q
->nr_rqs_elvpriv
;
522 drain
|= q
->request_fn_active
;
525 * Unfortunately, requests are queued at and tracked from
526 * multiple places and there's no single counter which can
527 * be drained. Check all the queues and counters.
530 struct blk_flush_queue
*fq
= blk_get_flush_queue(q
, NULL
);
531 drain
|= !list_empty(&q
->queue_head
);
532 for (i
= 0; i
< 2; i
++) {
533 drain
|= q
->nr_rqs
[i
];
534 drain
|= q
->in_flight
[i
];
536 drain
|= !list_empty(&fq
->flush_queue
[i
]);
543 spin_unlock_irq(q
->queue_lock
);
547 spin_lock_irq(q
->queue_lock
);
551 * With queue marked dead, any woken up waiter will fail the
552 * allocation path, so the wakeup chaining is lost and we're
553 * left with hung waiters. We need to wake up those waiters.
556 struct request_list
*rl
;
558 blk_queue_for_each_rl(rl
, q
)
559 for (i
= 0; i
< ARRAY_SIZE(rl
->wait
); i
++)
560 wake_up_all(&rl
->wait
[i
]);
565 * blk_queue_bypass_start - enter queue bypass mode
566 * @q: queue of interest
568 * In bypass mode, only the dispatch FIFO queue of @q is used. This
569 * function makes @q enter bypass mode and drains all requests which were
570 * throttled or issued before. On return, it's guaranteed that no request
571 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
572 * inside queue or RCU read lock.
574 void blk_queue_bypass_start(struct request_queue
*q
)
576 WARN_ON_ONCE(q
->mq_ops
);
578 spin_lock_irq(q
->queue_lock
);
580 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
581 spin_unlock_irq(q
->queue_lock
);
584 * Queues start drained. Skip actual draining till init is
585 * complete. This avoids lenghty delays during queue init which
586 * can happen many times during boot.
588 if (blk_queue_init_done(q
)) {
589 spin_lock_irq(q
->queue_lock
);
590 __blk_drain_queue(q
, false);
591 spin_unlock_irq(q
->queue_lock
);
593 /* ensure blk_queue_bypass() is %true inside RCU read lock */
597 EXPORT_SYMBOL_GPL(blk_queue_bypass_start
);
600 * blk_queue_bypass_end - leave queue bypass mode
601 * @q: queue of interest
603 * Leave bypass mode and restore the normal queueing behavior.
605 * Note: although blk_queue_bypass_start() is only called for blk-sq queues,
606 * this function is called for both blk-sq and blk-mq queues.
608 void blk_queue_bypass_end(struct request_queue
*q
)
610 spin_lock_irq(q
->queue_lock
);
611 if (!--q
->bypass_depth
)
612 queue_flag_clear(QUEUE_FLAG_BYPASS
, q
);
613 WARN_ON_ONCE(q
->bypass_depth
< 0);
614 spin_unlock_irq(q
->queue_lock
);
616 EXPORT_SYMBOL_GPL(blk_queue_bypass_end
);
618 void blk_set_queue_dying(struct request_queue
*q
)
620 spin_lock_irq(q
->queue_lock
);
621 queue_flag_set(QUEUE_FLAG_DYING
, q
);
622 spin_unlock_irq(q
->queue_lock
);
625 * When queue DYING flag is set, we need to block new req
626 * entering queue, so we call blk_freeze_queue_start() to
627 * prevent I/O from crossing blk_queue_enter().
629 blk_freeze_queue_start(q
);
632 blk_mq_wake_waiters(q
);
634 struct request_list
*rl
;
636 spin_lock_irq(q
->queue_lock
);
637 blk_queue_for_each_rl(rl
, q
) {
639 wake_up(&rl
->wait
[BLK_RW_SYNC
]);
640 wake_up(&rl
->wait
[BLK_RW_ASYNC
]);
643 spin_unlock_irq(q
->queue_lock
);
646 /* Make blk_queue_enter() reexamine the DYING flag. */
647 wake_up_all(&q
->mq_freeze_wq
);
649 EXPORT_SYMBOL_GPL(blk_set_queue_dying
);
652 * blk_cleanup_queue - shutdown a request queue
653 * @q: request queue to shutdown
655 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
656 * put it. All future requests will be failed immediately with -ENODEV.
658 void blk_cleanup_queue(struct request_queue
*q
)
660 spinlock_t
*lock
= q
->queue_lock
;
662 /* mark @q DYING, no new request or merges will be allowed afterwards */
663 mutex_lock(&q
->sysfs_lock
);
664 blk_set_queue_dying(q
);
668 * A dying queue is permanently in bypass mode till released. Note
669 * that, unlike blk_queue_bypass_start(), we aren't performing
670 * synchronize_rcu() after entering bypass mode to avoid the delay
671 * as some drivers create and destroy a lot of queues while
672 * probing. This is still safe because blk_release_queue() will be
673 * called only after the queue refcnt drops to zero and nothing,
674 * RCU or not, would be traversing the queue by then.
677 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
679 queue_flag_set(QUEUE_FLAG_NOMERGES
, q
);
680 queue_flag_set(QUEUE_FLAG_NOXMERGES
, q
);
681 queue_flag_set(QUEUE_FLAG_DYING
, q
);
682 spin_unlock_irq(lock
);
683 mutex_unlock(&q
->sysfs_lock
);
686 * Drain all requests queued before DYING marking. Set DEAD flag to
687 * prevent that q->request_fn() gets invoked after draining finished.
692 __blk_drain_queue(q
, true);
693 queue_flag_set(QUEUE_FLAG_DEAD
, q
);
694 spin_unlock_irq(lock
);
696 /* for synchronous bio-based driver finish in-flight integrity i/o */
697 blk_flush_integrity();
699 /* @q won't process any more request, flush async actions */
700 del_timer_sync(&q
->backing_dev_info
->laptop_mode_wb_timer
);
704 blk_mq_free_queue(q
);
705 percpu_ref_exit(&q
->q_usage_counter
);
708 if (q
->queue_lock
!= &q
->__queue_lock
)
709 q
->queue_lock
= &q
->__queue_lock
;
710 spin_unlock_irq(lock
);
712 /* @q is and will stay empty, shutdown and put */
715 EXPORT_SYMBOL(blk_cleanup_queue
);
717 /* Allocate memory local to the request queue */
718 static void *alloc_request_simple(gfp_t gfp_mask
, void *data
)
720 struct request_queue
*q
= data
;
722 return kmem_cache_alloc_node(request_cachep
, gfp_mask
, q
->node
);
725 static void free_request_simple(void *element
, void *data
)
727 kmem_cache_free(request_cachep
, element
);
730 static void *alloc_request_size(gfp_t gfp_mask
, void *data
)
732 struct request_queue
*q
= data
;
735 rq
= kmalloc_node(sizeof(struct request
) + q
->cmd_size
, gfp_mask
,
737 if (rq
&& q
->init_rq_fn
&& q
->init_rq_fn(q
, rq
, gfp_mask
) < 0) {
744 static void free_request_size(void *element
, void *data
)
746 struct request_queue
*q
= data
;
749 q
->exit_rq_fn(q
, element
);
753 int blk_init_rl(struct request_list
*rl
, struct request_queue
*q
,
756 if (unlikely(rl
->rq_pool
) || q
->mq_ops
)
760 rl
->count
[BLK_RW_SYNC
] = rl
->count
[BLK_RW_ASYNC
] = 0;
761 rl
->starved
[BLK_RW_SYNC
] = rl
->starved
[BLK_RW_ASYNC
] = 0;
762 init_waitqueue_head(&rl
->wait
[BLK_RW_SYNC
]);
763 init_waitqueue_head(&rl
->wait
[BLK_RW_ASYNC
]);
766 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
,
767 alloc_request_size
, free_request_size
,
768 q
, gfp_mask
, q
->node
);
770 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
,
771 alloc_request_simple
, free_request_simple
,
772 q
, gfp_mask
, q
->node
);
777 if (rl
!= &q
->root_rl
)
778 WARN_ON_ONCE(!blk_get_queue(q
));
783 void blk_exit_rl(struct request_queue
*q
, struct request_list
*rl
)
786 mempool_destroy(rl
->rq_pool
);
787 if (rl
!= &q
->root_rl
)
792 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
794 return blk_alloc_queue_node(gfp_mask
, NUMA_NO_NODE
);
796 EXPORT_SYMBOL(blk_alloc_queue
);
798 int blk_queue_enter(struct request_queue
*q
, bool nowait
)
803 if (percpu_ref_tryget_live(&q
->q_usage_counter
))
810 * read pair of barrier in blk_freeze_queue_start(),
811 * we need to order reading __PERCPU_REF_DEAD flag of
812 * .q_usage_counter and reading .mq_freeze_depth or
813 * queue dying flag, otherwise the following wait may
814 * never return if the two reads are reordered.
818 ret
= wait_event_interruptible(q
->mq_freeze_wq
,
819 !atomic_read(&q
->mq_freeze_depth
) ||
821 if (blk_queue_dying(q
))
828 void blk_queue_exit(struct request_queue
*q
)
830 percpu_ref_put(&q
->q_usage_counter
);
833 static void blk_queue_usage_counter_release(struct percpu_ref
*ref
)
835 struct request_queue
*q
=
836 container_of(ref
, struct request_queue
, q_usage_counter
);
838 wake_up_all(&q
->mq_freeze_wq
);
841 static void blk_rq_timed_out_timer(unsigned long data
)
843 struct request_queue
*q
= (struct request_queue
*)data
;
845 kblockd_schedule_work(&q
->timeout_work
);
848 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
850 struct request_queue
*q
;
852 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
853 gfp_mask
| __GFP_ZERO
, node_id
);
857 q
->id
= ida_simple_get(&blk_queue_ida
, 0, 0, gfp_mask
);
861 q
->bio_split
= bioset_create(BIO_POOL_SIZE
, 0, BIOSET_NEED_BVECS
);
865 q
->backing_dev_info
= bdi_alloc_node(gfp_mask
, node_id
);
866 if (!q
->backing_dev_info
)
869 q
->stats
= blk_alloc_queue_stats();
873 q
->backing_dev_info
->ra_pages
=
874 (VM_MAX_READAHEAD
* 1024) / PAGE_SIZE
;
875 q
->backing_dev_info
->capabilities
= BDI_CAP_CGROUP_WRITEBACK
;
876 q
->backing_dev_info
->name
= "block";
879 setup_timer(&q
->backing_dev_info
->laptop_mode_wb_timer
,
880 laptop_mode_timer_fn
, (unsigned long) q
);
881 setup_timer(&q
->timeout
, blk_rq_timed_out_timer
, (unsigned long) q
);
882 INIT_WORK(&q
->timeout_work
, NULL
);
883 INIT_LIST_HEAD(&q
->queue_head
);
884 INIT_LIST_HEAD(&q
->timeout_list
);
885 INIT_LIST_HEAD(&q
->icq_list
);
886 #ifdef CONFIG_BLK_CGROUP
887 INIT_LIST_HEAD(&q
->blkg_list
);
889 INIT_DELAYED_WORK(&q
->delay_work
, blk_delay_work
);
891 kobject_init(&q
->kobj
, &blk_queue_ktype
);
893 #ifdef CONFIG_BLK_DEV_IO_TRACE
894 mutex_init(&q
->blk_trace_mutex
);
896 mutex_init(&q
->sysfs_lock
);
897 spin_lock_init(&q
->__queue_lock
);
900 * By default initialize queue_lock to internal lock and driver can
901 * override it later if need be.
903 q
->queue_lock
= &q
->__queue_lock
;
906 * A queue starts its life with bypass turned on to avoid
907 * unnecessary bypass on/off overhead and nasty surprises during
908 * init. The initial bypass will be finished when the queue is
909 * registered by blk_register_queue().
912 __set_bit(QUEUE_FLAG_BYPASS
, &q
->queue_flags
);
914 init_waitqueue_head(&q
->mq_freeze_wq
);
917 * Init percpu_ref in atomic mode so that it's faster to shutdown.
918 * See blk_register_queue() for details.
920 if (percpu_ref_init(&q
->q_usage_counter
,
921 blk_queue_usage_counter_release
,
922 PERCPU_REF_INIT_ATOMIC
, GFP_KERNEL
))
925 if (blkcg_init_queue(q
))
931 percpu_ref_exit(&q
->q_usage_counter
);
933 blk_free_queue_stats(q
->stats
);
935 bdi_put(q
->backing_dev_info
);
937 bioset_free(q
->bio_split
);
939 ida_simple_remove(&blk_queue_ida
, q
->id
);
941 kmem_cache_free(blk_requestq_cachep
, q
);
944 EXPORT_SYMBOL(blk_alloc_queue_node
);
947 * blk_init_queue - prepare a request queue for use with a block device
948 * @rfn: The function to be called to process requests that have been
949 * placed on the queue.
950 * @lock: Request queue spin lock
953 * If a block device wishes to use the standard request handling procedures,
954 * which sorts requests and coalesces adjacent requests, then it must
955 * call blk_init_queue(). The function @rfn will be called when there
956 * are requests on the queue that need to be processed. If the device
957 * supports plugging, then @rfn may not be called immediately when requests
958 * are available on the queue, but may be called at some time later instead.
959 * Plugged queues are generally unplugged when a buffer belonging to one
960 * of the requests on the queue is needed, or due to memory pressure.
962 * @rfn is not required, or even expected, to remove all requests off the
963 * queue, but only as many as it can handle at a time. If it does leave
964 * requests on the queue, it is responsible for arranging that the requests
965 * get dealt with eventually.
967 * The queue spin lock must be held while manipulating the requests on the
968 * request queue; this lock will be taken also from interrupt context, so irq
969 * disabling is needed for it.
971 * Function returns a pointer to the initialized request queue, or %NULL if
975 * blk_init_queue() must be paired with a blk_cleanup_queue() call
976 * when the block device is deactivated (such as at module unload).
979 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
981 return blk_init_queue_node(rfn
, lock
, NUMA_NO_NODE
);
983 EXPORT_SYMBOL(blk_init_queue
);
985 struct request_queue
*
986 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
988 struct request_queue
*q
;
990 q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
996 q
->queue_lock
= lock
;
997 if (blk_init_allocated_queue(q
) < 0) {
998 blk_cleanup_queue(q
);
1004 EXPORT_SYMBOL(blk_init_queue_node
);
1006 static blk_qc_t
blk_queue_bio(struct request_queue
*q
, struct bio
*bio
);
1009 int blk_init_allocated_queue(struct request_queue
*q
)
1011 WARN_ON_ONCE(q
->mq_ops
);
1013 q
->fq
= blk_alloc_flush_queue(q
, NUMA_NO_NODE
, q
->cmd_size
);
1017 if (q
->init_rq_fn
&& q
->init_rq_fn(q
, q
->fq
->flush_rq
, GFP_KERNEL
))
1018 goto out_free_flush_queue
;
1020 if (blk_init_rl(&q
->root_rl
, q
, GFP_KERNEL
))
1021 goto out_exit_flush_rq
;
1023 INIT_WORK(&q
->timeout_work
, blk_timeout_work
);
1024 q
->queue_flags
|= QUEUE_FLAG_DEFAULT
;
1027 * This also sets hw/phys segments, boundary and size
1029 blk_queue_make_request(q
, blk_queue_bio
);
1031 q
->sg_reserved_size
= INT_MAX
;
1033 /* Protect q->elevator from elevator_change */
1034 mutex_lock(&q
->sysfs_lock
);
1037 if (elevator_init(q
, NULL
)) {
1038 mutex_unlock(&q
->sysfs_lock
);
1039 goto out_exit_flush_rq
;
1042 mutex_unlock(&q
->sysfs_lock
);
1047 q
->exit_rq_fn(q
, q
->fq
->flush_rq
);
1048 out_free_flush_queue
:
1049 blk_free_flush_queue(q
->fq
);
1052 EXPORT_SYMBOL(blk_init_allocated_queue
);
1054 bool blk_get_queue(struct request_queue
*q
)
1056 if (likely(!blk_queue_dying(q
))) {
1063 EXPORT_SYMBOL(blk_get_queue
);
1065 static inline void blk_free_request(struct request_list
*rl
, struct request
*rq
)
1067 if (rq
->rq_flags
& RQF_ELVPRIV
) {
1068 elv_put_request(rl
->q
, rq
);
1070 put_io_context(rq
->elv
.icq
->ioc
);
1073 mempool_free(rq
, rl
->rq_pool
);
1077 * ioc_batching returns true if the ioc is a valid batching request and
1078 * should be given priority access to a request.
1080 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
1086 * Make sure the process is able to allocate at least 1 request
1087 * even if the batch times out, otherwise we could theoretically
1090 return ioc
->nr_batch_requests
== q
->nr_batching
||
1091 (ioc
->nr_batch_requests
> 0
1092 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
1096 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
1097 * will cause the process to be a "batcher" on all queues in the system. This
1098 * is the behaviour we want though - once it gets a wakeup it should be given
1101 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
1103 if (!ioc
|| ioc_batching(q
, ioc
))
1106 ioc
->nr_batch_requests
= q
->nr_batching
;
1107 ioc
->last_waited
= jiffies
;
1110 static void __freed_request(struct request_list
*rl
, int sync
)
1112 struct request_queue
*q
= rl
->q
;
1114 if (rl
->count
[sync
] < queue_congestion_off_threshold(q
))
1115 blk_clear_congested(rl
, sync
);
1117 if (rl
->count
[sync
] + 1 <= q
->nr_requests
) {
1118 if (waitqueue_active(&rl
->wait
[sync
]))
1119 wake_up(&rl
->wait
[sync
]);
1121 blk_clear_rl_full(rl
, sync
);
1126 * A request has just been released. Account for it, update the full and
1127 * congestion status, wake up any waiters. Called under q->queue_lock.
1129 static void freed_request(struct request_list
*rl
, bool sync
,
1130 req_flags_t rq_flags
)
1132 struct request_queue
*q
= rl
->q
;
1136 if (rq_flags
& RQF_ELVPRIV
)
1137 q
->nr_rqs_elvpriv
--;
1139 __freed_request(rl
, sync
);
1141 if (unlikely(rl
->starved
[sync
^ 1]))
1142 __freed_request(rl
, sync
^ 1);
1145 int blk_update_nr_requests(struct request_queue
*q
, unsigned int nr
)
1147 struct request_list
*rl
;
1148 int on_thresh
, off_thresh
;
1150 WARN_ON_ONCE(q
->mq_ops
);
1152 spin_lock_irq(q
->queue_lock
);
1153 q
->nr_requests
= nr
;
1154 blk_queue_congestion_threshold(q
);
1155 on_thresh
= queue_congestion_on_threshold(q
);
1156 off_thresh
= queue_congestion_off_threshold(q
);
1158 blk_queue_for_each_rl(rl
, q
) {
1159 if (rl
->count
[BLK_RW_SYNC
] >= on_thresh
)
1160 blk_set_congested(rl
, BLK_RW_SYNC
);
1161 else if (rl
->count
[BLK_RW_SYNC
] < off_thresh
)
1162 blk_clear_congested(rl
, BLK_RW_SYNC
);
1164 if (rl
->count
[BLK_RW_ASYNC
] >= on_thresh
)
1165 blk_set_congested(rl
, BLK_RW_ASYNC
);
1166 else if (rl
->count
[BLK_RW_ASYNC
] < off_thresh
)
1167 blk_clear_congested(rl
, BLK_RW_ASYNC
);
1169 if (rl
->count
[BLK_RW_SYNC
] >= q
->nr_requests
) {
1170 blk_set_rl_full(rl
, BLK_RW_SYNC
);
1172 blk_clear_rl_full(rl
, BLK_RW_SYNC
);
1173 wake_up(&rl
->wait
[BLK_RW_SYNC
]);
1176 if (rl
->count
[BLK_RW_ASYNC
] >= q
->nr_requests
) {
1177 blk_set_rl_full(rl
, BLK_RW_ASYNC
);
1179 blk_clear_rl_full(rl
, BLK_RW_ASYNC
);
1180 wake_up(&rl
->wait
[BLK_RW_ASYNC
]);
1184 spin_unlock_irq(q
->queue_lock
);
1189 * __get_request - get a free request
1190 * @rl: request list to allocate from
1191 * @op: operation and flags
1192 * @bio: bio to allocate request for (can be %NULL)
1193 * @flags: BLQ_MQ_REQ_* flags
1195 * Get a free request from @q. This function may fail under memory
1196 * pressure or if @q is dead.
1198 * Must be called with @q->queue_lock held and,
1199 * Returns ERR_PTR on failure, with @q->queue_lock held.
1200 * Returns request pointer on success, with @q->queue_lock *not held*.
1202 static struct request
*__get_request(struct request_list
*rl
, unsigned int op
,
1203 struct bio
*bio
, unsigned int flags
)
1205 struct request_queue
*q
= rl
->q
;
1207 struct elevator_type
*et
= q
->elevator
->type
;
1208 struct io_context
*ioc
= rq_ioc(bio
);
1209 struct io_cq
*icq
= NULL
;
1210 const bool is_sync
= op_is_sync(op
);
1212 gfp_t gfp_mask
= flags
& BLK_MQ_REQ_NOWAIT
? GFP_ATOMIC
:
1213 __GFP_DIRECT_RECLAIM
;
1214 req_flags_t rq_flags
= RQF_ALLOCED
;
1216 lockdep_assert_held(q
->queue_lock
);
1218 if (unlikely(blk_queue_dying(q
)))
1219 return ERR_PTR(-ENODEV
);
1221 may_queue
= elv_may_queue(q
, op
);
1222 if (may_queue
== ELV_MQUEUE_NO
)
1225 if (rl
->count
[is_sync
]+1 >= queue_congestion_on_threshold(q
)) {
1226 if (rl
->count
[is_sync
]+1 >= q
->nr_requests
) {
1228 * The queue will fill after this allocation, so set
1229 * it as full, and mark this process as "batching".
1230 * This process will be allowed to complete a batch of
1231 * requests, others will be blocked.
1233 if (!blk_rl_full(rl
, is_sync
)) {
1234 ioc_set_batching(q
, ioc
);
1235 blk_set_rl_full(rl
, is_sync
);
1237 if (may_queue
!= ELV_MQUEUE_MUST
1238 && !ioc_batching(q
, ioc
)) {
1240 * The queue is full and the allocating
1241 * process is not a "batcher", and not
1242 * exempted by the IO scheduler
1244 return ERR_PTR(-ENOMEM
);
1248 blk_set_congested(rl
, is_sync
);
1252 * Only allow batching queuers to allocate up to 50% over the defined
1253 * limit of requests, otherwise we could have thousands of requests
1254 * allocated with any setting of ->nr_requests
1256 if (rl
->count
[is_sync
] >= (3 * q
->nr_requests
/ 2))
1257 return ERR_PTR(-ENOMEM
);
1259 q
->nr_rqs
[is_sync
]++;
1260 rl
->count
[is_sync
]++;
1261 rl
->starved
[is_sync
] = 0;
1264 * Decide whether the new request will be managed by elevator. If
1265 * so, mark @rq_flags and increment elvpriv. Non-zero elvpriv will
1266 * prevent the current elevator from being destroyed until the new
1267 * request is freed. This guarantees icq's won't be destroyed and
1268 * makes creating new ones safe.
1270 * Flush requests do not use the elevator so skip initialization.
1271 * This allows a request to share the flush and elevator data.
1273 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1274 * it will be created after releasing queue_lock.
1276 if (!op_is_flush(op
) && !blk_queue_bypass(q
)) {
1277 rq_flags
|= RQF_ELVPRIV
;
1278 q
->nr_rqs_elvpriv
++;
1279 if (et
->icq_cache
&& ioc
)
1280 icq
= ioc_lookup_icq(ioc
, q
);
1283 if (blk_queue_io_stat(q
))
1284 rq_flags
|= RQF_IO_STAT
;
1285 spin_unlock_irq(q
->queue_lock
);
1287 /* allocate and init request */
1288 rq
= mempool_alloc(rl
->rq_pool
, gfp_mask
);
1293 blk_rq_set_rl(rq
, rl
);
1295 rq
->rq_flags
= rq_flags
;
1296 if (flags
& BLK_MQ_REQ_PREEMPT
)
1297 rq
->rq_flags
|= RQF_PREEMPT
;
1300 if (rq_flags
& RQF_ELVPRIV
) {
1301 if (unlikely(et
->icq_cache
&& !icq
)) {
1303 icq
= ioc_create_icq(ioc
, q
, gfp_mask
);
1309 if (unlikely(elv_set_request(q
, rq
, bio
, gfp_mask
)))
1312 /* @rq->elv.icq holds io_context until @rq is freed */
1314 get_io_context(icq
->ioc
);
1318 * ioc may be NULL here, and ioc_batching will be false. That's
1319 * OK, if the queue is under the request limit then requests need
1320 * not count toward the nr_batch_requests limit. There will always
1321 * be some limit enforced by BLK_BATCH_TIME.
1323 if (ioc_batching(q
, ioc
))
1324 ioc
->nr_batch_requests
--;
1326 trace_block_getrq(q
, bio
, op
);
1331 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1332 * and may fail indefinitely under memory pressure and thus
1333 * shouldn't stall IO. Treat this request as !elvpriv. This will
1334 * disturb iosched and blkcg but weird is bettern than dead.
1336 printk_ratelimited(KERN_WARNING
"%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1337 __func__
, dev_name(q
->backing_dev_info
->dev
));
1339 rq
->rq_flags
&= ~RQF_ELVPRIV
;
1342 spin_lock_irq(q
->queue_lock
);
1343 q
->nr_rqs_elvpriv
--;
1344 spin_unlock_irq(q
->queue_lock
);
1349 * Allocation failed presumably due to memory. Undo anything we
1350 * might have messed up.
1352 * Allocating task should really be put onto the front of the wait
1353 * queue, but this is pretty rare.
1355 spin_lock_irq(q
->queue_lock
);
1356 freed_request(rl
, is_sync
, rq_flags
);
1359 * in the very unlikely event that allocation failed and no
1360 * requests for this direction was pending, mark us starved so that
1361 * freeing of a request in the other direction will notice
1362 * us. another possible fix would be to split the rq mempool into
1366 if (unlikely(rl
->count
[is_sync
] == 0))
1367 rl
->starved
[is_sync
] = 1;
1368 return ERR_PTR(-ENOMEM
);
1372 * get_request - get a free request
1373 * @q: request_queue to allocate request from
1374 * @op: operation and flags
1375 * @bio: bio to allocate request for (can be %NULL)
1376 * @flags: BLK_MQ_REQ_* flags.
1378 * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask,
1379 * this function keeps retrying under memory pressure and fails iff @q is dead.
1381 * Must be called with @q->queue_lock held and,
1382 * Returns ERR_PTR on failure, with @q->queue_lock held.
1383 * Returns request pointer on success, with @q->queue_lock *not held*.
1385 static struct request
*get_request(struct request_queue
*q
, unsigned int op
,
1386 struct bio
*bio
, unsigned int flags
)
1388 const bool is_sync
= op_is_sync(op
);
1390 struct request_list
*rl
;
1393 lockdep_assert_held(q
->queue_lock
);
1394 WARN_ON_ONCE(q
->mq_ops
);
1396 rl
= blk_get_rl(q
, bio
); /* transferred to @rq on success */
1398 rq
= __get_request(rl
, op
, bio
, flags
);
1402 if (op
& REQ_NOWAIT
) {
1404 return ERR_PTR(-EAGAIN
);
1407 if ((flags
& BLK_MQ_REQ_NOWAIT
) || unlikely(blk_queue_dying(q
))) {
1412 /* wait on @rl and retry */
1413 prepare_to_wait_exclusive(&rl
->wait
[is_sync
], &wait
,
1414 TASK_UNINTERRUPTIBLE
);
1416 trace_block_sleeprq(q
, bio
, op
);
1418 spin_unlock_irq(q
->queue_lock
);
1422 * After sleeping, we become a "batching" process and will be able
1423 * to allocate at least one request, and up to a big batch of them
1424 * for a small period time. See ioc_batching, ioc_set_batching
1426 ioc_set_batching(q
, current
->io_context
);
1428 spin_lock_irq(q
->queue_lock
);
1429 finish_wait(&rl
->wait
[is_sync
], &wait
);
1434 /* flags: BLK_MQ_REQ_PREEMPT and/or BLK_MQ_REQ_NOWAIT. */
1435 static struct request
*blk_old_get_request(struct request_queue
*q
,
1436 unsigned int op
, unsigned int flags
)
1439 gfp_t gfp_mask
= flags
& BLK_MQ_REQ_NOWAIT
? GFP_ATOMIC
:
1440 __GFP_DIRECT_RECLAIM
;
1443 WARN_ON_ONCE(q
->mq_ops
);
1445 /* create ioc upfront */
1446 create_io_context(gfp_mask
, q
->node
);
1448 ret
= blk_queue_enter(q
, !(gfp_mask
& __GFP_DIRECT_RECLAIM
) ||
1451 return ERR_PTR(ret
);
1452 spin_lock_irq(q
->queue_lock
);
1453 rq
= get_request(q
, op
, NULL
, flags
);
1455 spin_unlock_irq(q
->queue_lock
);
1460 /* q->queue_lock is unlocked at this point */
1462 rq
->__sector
= (sector_t
) -1;
1463 rq
->bio
= rq
->biotail
= NULL
;
1468 * blk_get_request_flags - allocate a request
1469 * @q: request queue to allocate a request for
1470 * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC.
1471 * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
1473 struct request
*blk_get_request_flags(struct request_queue
*q
, unsigned int op
,
1476 struct request
*req
;
1478 WARN_ON_ONCE(op
& REQ_NOWAIT
);
1479 WARN_ON_ONCE(flags
& ~(BLK_MQ_REQ_NOWAIT
| BLK_MQ_REQ_PREEMPT
));
1482 req
= blk_mq_alloc_request(q
, op
, flags
);
1483 if (!IS_ERR(req
) && q
->mq_ops
->initialize_rq_fn
)
1484 q
->mq_ops
->initialize_rq_fn(req
);
1486 req
= blk_old_get_request(q
, op
, flags
);
1487 if (!IS_ERR(req
) && q
->initialize_rq_fn
)
1488 q
->initialize_rq_fn(req
);
1493 EXPORT_SYMBOL(blk_get_request_flags
);
1495 struct request
*blk_get_request(struct request_queue
*q
, unsigned int op
,
1498 return blk_get_request_flags(q
, op
, gfp_mask
& __GFP_DIRECT_RECLAIM
?
1499 0 : BLK_MQ_REQ_NOWAIT
);
1501 EXPORT_SYMBOL(blk_get_request
);
1504 * blk_requeue_request - put a request back on queue
1505 * @q: request queue where request should be inserted
1506 * @rq: request to be inserted
1509 * Drivers often keep queueing requests until the hardware cannot accept
1510 * more, when that condition happens we need to put the request back
1511 * on the queue. Must be called with queue lock held.
1513 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
1515 lockdep_assert_held(q
->queue_lock
);
1516 WARN_ON_ONCE(q
->mq_ops
);
1518 blk_delete_timer(rq
);
1519 blk_clear_rq_complete(rq
);
1520 trace_block_rq_requeue(q
, rq
);
1521 wbt_requeue(q
->rq_wb
, &rq
->issue_stat
);
1523 if (rq
->rq_flags
& RQF_QUEUED
)
1524 blk_queue_end_tag(q
, rq
);
1526 BUG_ON(blk_queued_rq(rq
));
1528 elv_requeue_request(q
, rq
);
1530 EXPORT_SYMBOL(blk_requeue_request
);
1532 static void add_acct_request(struct request_queue
*q
, struct request
*rq
,
1535 blk_account_io_start(rq
, true);
1536 __elv_add_request(q
, rq
, where
);
1539 static void part_round_stats_single(struct request_queue
*q
, int cpu
,
1540 struct hd_struct
*part
, unsigned long now
,
1541 unsigned int inflight
)
1544 __part_stat_add(cpu
, part
, time_in_queue
,
1545 inflight
* (now
- part
->stamp
));
1546 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
1552 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1553 * @q: target block queue
1554 * @cpu: cpu number for stats access
1555 * @part: target partition
1557 * The average IO queue length and utilisation statistics are maintained
1558 * by observing the current state of the queue length and the amount of
1559 * time it has been in this state for.
1561 * Normally, that accounting is done on IO completion, but that can result
1562 * in more than a second's worth of IO being accounted for within any one
1563 * second, leading to >100% utilisation. To deal with that, we call this
1564 * function to do a round-off before returning the results when reading
1565 * /proc/diskstats. This accounts immediately for all queue usage up to
1566 * the current jiffies and restarts the counters again.
1568 void part_round_stats(struct request_queue
*q
, int cpu
, struct hd_struct
*part
)
1570 struct hd_struct
*part2
= NULL
;
1571 unsigned long now
= jiffies
;
1572 unsigned int inflight
[2];
1575 if (part
->stamp
!= now
)
1579 part2
= &part_to_disk(part
)->part0
;
1580 if (part2
->stamp
!= now
)
1587 part_in_flight(q
, part
, inflight
);
1590 part_round_stats_single(q
, cpu
, part2
, now
, inflight
[1]);
1592 part_round_stats_single(q
, cpu
, part
, now
, inflight
[0]);
1594 EXPORT_SYMBOL_GPL(part_round_stats
);
1597 static void blk_pm_put_request(struct request
*rq
)
1599 if (rq
->q
->dev
&& !(rq
->rq_flags
& RQF_PM
) && !--rq
->q
->nr_pending
)
1600 pm_runtime_mark_last_busy(rq
->q
->dev
);
1603 static inline void blk_pm_put_request(struct request
*rq
) {}
1606 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1608 req_flags_t rq_flags
= req
->rq_flags
;
1614 blk_mq_free_request(req
);
1618 lockdep_assert_held(q
->queue_lock
);
1620 blk_pm_put_request(req
);
1622 elv_completed_request(q
, req
);
1624 /* this is a bio leak */
1625 WARN_ON(req
->bio
!= NULL
);
1627 wbt_done(q
->rq_wb
, &req
->issue_stat
);
1630 * Request may not have originated from ll_rw_blk. if not,
1631 * it didn't come out of our reserved rq pools
1633 if (rq_flags
& RQF_ALLOCED
) {
1634 struct request_list
*rl
= blk_rq_rl(req
);
1635 bool sync
= op_is_sync(req
->cmd_flags
);
1637 BUG_ON(!list_empty(&req
->queuelist
));
1638 BUG_ON(ELV_ON_HASH(req
));
1640 blk_free_request(rl
, req
);
1641 freed_request(rl
, sync
, rq_flags
);
1646 EXPORT_SYMBOL_GPL(__blk_put_request
);
1648 void blk_put_request(struct request
*req
)
1650 struct request_queue
*q
= req
->q
;
1653 blk_mq_free_request(req
);
1655 unsigned long flags
;
1657 spin_lock_irqsave(q
->queue_lock
, flags
);
1658 __blk_put_request(q
, req
);
1659 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1662 EXPORT_SYMBOL(blk_put_request
);
1664 bool bio_attempt_back_merge(struct request_queue
*q
, struct request
*req
,
1667 const int ff
= bio
->bi_opf
& REQ_FAILFAST_MASK
;
1669 if (!ll_back_merge_fn(q
, req
, bio
))
1672 trace_block_bio_backmerge(q
, req
, bio
);
1674 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1675 blk_rq_set_mixed_merge(req
);
1677 req
->biotail
->bi_next
= bio
;
1679 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1680 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1682 blk_account_io_start(req
, false);
1686 bool bio_attempt_front_merge(struct request_queue
*q
, struct request
*req
,
1689 const int ff
= bio
->bi_opf
& REQ_FAILFAST_MASK
;
1691 if (!ll_front_merge_fn(q
, req
, bio
))
1694 trace_block_bio_frontmerge(q
, req
, bio
);
1696 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1697 blk_rq_set_mixed_merge(req
);
1699 bio
->bi_next
= req
->bio
;
1702 req
->__sector
= bio
->bi_iter
.bi_sector
;
1703 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1704 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1706 blk_account_io_start(req
, false);
1710 bool bio_attempt_discard_merge(struct request_queue
*q
, struct request
*req
,
1713 unsigned short segments
= blk_rq_nr_discard_segments(req
);
1715 if (segments
>= queue_max_discard_segments(q
))
1717 if (blk_rq_sectors(req
) + bio_sectors(bio
) >
1718 blk_rq_get_max_sectors(req
, blk_rq_pos(req
)))
1721 req
->biotail
->bi_next
= bio
;
1723 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1724 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1725 req
->nr_phys_segments
= segments
+ 1;
1727 blk_account_io_start(req
, false);
1730 req_set_nomerge(q
, req
);
1735 * blk_attempt_plug_merge - try to merge with %current's plugged list
1736 * @q: request_queue new bio is being queued at
1737 * @bio: new bio being queued
1738 * @request_count: out parameter for number of traversed plugged requests
1739 * @same_queue_rq: pointer to &struct request that gets filled in when
1740 * another request associated with @q is found on the plug list
1741 * (optional, may be %NULL)
1743 * Determine whether @bio being queued on @q can be merged with a request
1744 * on %current's plugged list. Returns %true if merge was successful,
1747 * Plugging coalesces IOs from the same issuer for the same purpose without
1748 * going through @q->queue_lock. As such it's more of an issuing mechanism
1749 * than scheduling, and the request, while may have elvpriv data, is not
1750 * added on the elevator at this point. In addition, we don't have
1751 * reliable access to the elevator outside queue lock. Only check basic
1752 * merging parameters without querying the elevator.
1754 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1756 bool blk_attempt_plug_merge(struct request_queue
*q
, struct bio
*bio
,
1757 unsigned int *request_count
,
1758 struct request
**same_queue_rq
)
1760 struct blk_plug
*plug
;
1762 struct list_head
*plug_list
;
1764 plug
= current
->plug
;
1770 plug_list
= &plug
->mq_list
;
1772 plug_list
= &plug
->list
;
1774 list_for_each_entry_reverse(rq
, plug_list
, queuelist
) {
1775 bool merged
= false;
1780 * Only blk-mq multiple hardware queues case checks the
1781 * rq in the same queue, there should be only one such
1785 *same_queue_rq
= rq
;
1788 if (rq
->q
!= q
|| !blk_rq_merge_ok(rq
, bio
))
1791 switch (blk_try_merge(rq
, bio
)) {
1792 case ELEVATOR_BACK_MERGE
:
1793 merged
= bio_attempt_back_merge(q
, rq
, bio
);
1795 case ELEVATOR_FRONT_MERGE
:
1796 merged
= bio_attempt_front_merge(q
, rq
, bio
);
1798 case ELEVATOR_DISCARD_MERGE
:
1799 merged
= bio_attempt_discard_merge(q
, rq
, bio
);
1812 unsigned int blk_plug_queued_count(struct request_queue
*q
)
1814 struct blk_plug
*plug
;
1816 struct list_head
*plug_list
;
1817 unsigned int ret
= 0;
1819 plug
= current
->plug
;
1824 plug_list
= &plug
->mq_list
;
1826 plug_list
= &plug
->list
;
1828 list_for_each_entry(rq
, plug_list
, queuelist
) {
1836 void blk_init_request_from_bio(struct request
*req
, struct bio
*bio
)
1838 struct io_context
*ioc
= rq_ioc(bio
);
1840 if (bio
->bi_opf
& REQ_RAHEAD
)
1841 req
->cmd_flags
|= REQ_FAILFAST_MASK
;
1843 req
->__sector
= bio
->bi_iter
.bi_sector
;
1844 if (ioprio_valid(bio_prio(bio
)))
1845 req
->ioprio
= bio_prio(bio
);
1847 req
->ioprio
= ioc
->ioprio
;
1849 req
->ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE
, 0);
1850 req
->write_hint
= bio
->bi_write_hint
;
1851 blk_rq_bio_prep(req
->q
, req
, bio
);
1853 EXPORT_SYMBOL_GPL(blk_init_request_from_bio
);
1855 static blk_qc_t
blk_queue_bio(struct request_queue
*q
, struct bio
*bio
)
1857 struct blk_plug
*plug
;
1858 int where
= ELEVATOR_INSERT_SORT
;
1859 struct request
*req
, *free
;
1860 unsigned int request_count
= 0;
1861 unsigned int wb_acct
;
1864 * low level driver can indicate that it wants pages above a
1865 * certain limit bounced to low memory (ie for highmem, or even
1866 * ISA dma in theory)
1868 blk_queue_bounce(q
, &bio
);
1870 blk_queue_split(q
, &bio
);
1872 if (!bio_integrity_prep(bio
))
1873 return BLK_QC_T_NONE
;
1875 if (op_is_flush(bio
->bi_opf
)) {
1876 spin_lock_irq(q
->queue_lock
);
1877 where
= ELEVATOR_INSERT_FLUSH
;
1882 * Check if we can merge with the plugged list before grabbing
1885 if (!blk_queue_nomerges(q
)) {
1886 if (blk_attempt_plug_merge(q
, bio
, &request_count
, NULL
))
1887 return BLK_QC_T_NONE
;
1889 request_count
= blk_plug_queued_count(q
);
1891 spin_lock_irq(q
->queue_lock
);
1893 switch (elv_merge(q
, &req
, bio
)) {
1894 case ELEVATOR_BACK_MERGE
:
1895 if (!bio_attempt_back_merge(q
, req
, bio
))
1897 elv_bio_merged(q
, req
, bio
);
1898 free
= attempt_back_merge(q
, req
);
1900 __blk_put_request(q
, free
);
1902 elv_merged_request(q
, req
, ELEVATOR_BACK_MERGE
);
1904 case ELEVATOR_FRONT_MERGE
:
1905 if (!bio_attempt_front_merge(q
, req
, bio
))
1907 elv_bio_merged(q
, req
, bio
);
1908 free
= attempt_front_merge(q
, req
);
1910 __blk_put_request(q
, free
);
1912 elv_merged_request(q
, req
, ELEVATOR_FRONT_MERGE
);
1919 wb_acct
= wbt_wait(q
->rq_wb
, bio
, q
->queue_lock
);
1922 * Grab a free request. This is might sleep but can not fail.
1923 * Returns with the queue unlocked.
1925 blk_queue_enter_live(q
);
1926 req
= get_request(q
, bio
->bi_opf
, bio
, 0);
1929 __wbt_done(q
->rq_wb
, wb_acct
);
1930 if (PTR_ERR(req
) == -ENOMEM
)
1931 bio
->bi_status
= BLK_STS_RESOURCE
;
1933 bio
->bi_status
= BLK_STS_IOERR
;
1938 wbt_track(&req
->issue_stat
, wb_acct
);
1941 * After dropping the lock and possibly sleeping here, our request
1942 * may now be mergeable after it had proven unmergeable (above).
1943 * We don't worry about that case for efficiency. It won't happen
1944 * often, and the elevators are able to handle it.
1946 blk_init_request_from_bio(req
, bio
);
1948 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
))
1949 req
->cpu
= raw_smp_processor_id();
1951 plug
= current
->plug
;
1954 * If this is the first request added after a plug, fire
1957 * @request_count may become stale because of schedule
1958 * out, so check plug list again.
1960 if (!request_count
|| list_empty(&plug
->list
))
1961 trace_block_plug(q
);
1963 struct request
*last
= list_entry_rq(plug
->list
.prev
);
1964 if (request_count
>= BLK_MAX_REQUEST_COUNT
||
1965 blk_rq_bytes(last
) >= BLK_PLUG_FLUSH_SIZE
) {
1966 blk_flush_plug_list(plug
, false);
1967 trace_block_plug(q
);
1970 list_add_tail(&req
->queuelist
, &plug
->list
);
1971 blk_account_io_start(req
, true);
1973 spin_lock_irq(q
->queue_lock
);
1974 add_acct_request(q
, req
, where
);
1977 spin_unlock_irq(q
->queue_lock
);
1980 return BLK_QC_T_NONE
;
1983 static void handle_bad_sector(struct bio
*bio
)
1985 char b
[BDEVNAME_SIZE
];
1987 printk(KERN_INFO
"attempt to access beyond end of device\n");
1988 printk(KERN_INFO
"%s: rw=%d, want=%Lu, limit=%Lu\n",
1989 bio_devname(bio
, b
), bio
->bi_opf
,
1990 (unsigned long long)bio_end_sector(bio
),
1991 (long long)get_capacity(bio
->bi_disk
));
1994 #ifdef CONFIG_FAIL_MAKE_REQUEST
1996 static DECLARE_FAULT_ATTR(fail_make_request
);
1998 static int __init
setup_fail_make_request(char *str
)
2000 return setup_fault_attr(&fail_make_request
, str
);
2002 __setup("fail_make_request=", setup_fail_make_request
);
2004 static bool should_fail_request(struct hd_struct
*part
, unsigned int bytes
)
2006 return part
->make_it_fail
&& should_fail(&fail_make_request
, bytes
);
2009 static int __init
fail_make_request_debugfs(void)
2011 struct dentry
*dir
= fault_create_debugfs_attr("fail_make_request",
2012 NULL
, &fail_make_request
);
2014 return PTR_ERR_OR_ZERO(dir
);
2017 late_initcall(fail_make_request_debugfs
);
2019 #else /* CONFIG_FAIL_MAKE_REQUEST */
2021 static inline bool should_fail_request(struct hd_struct
*part
,
2027 #endif /* CONFIG_FAIL_MAKE_REQUEST */
2030 * Remap block n of partition p to block n+start(p) of the disk.
2032 static inline int blk_partition_remap(struct bio
*bio
)
2034 struct hd_struct
*p
;
2038 * Zone reset does not include bi_size so bio_sectors() is always 0.
2039 * Include a test for the reset op code and perform the remap if needed.
2041 if (!bio
->bi_partno
||
2042 (!bio_sectors(bio
) && bio_op(bio
) != REQ_OP_ZONE_RESET
))
2046 p
= __disk_get_part(bio
->bi_disk
, bio
->bi_partno
);
2047 if (likely(p
&& !should_fail_request(p
, bio
->bi_iter
.bi_size
))) {
2048 bio
->bi_iter
.bi_sector
+= p
->start_sect
;
2050 trace_block_bio_remap(bio
->bi_disk
->queue
, bio
, part_devt(p
),
2051 bio
->bi_iter
.bi_sector
- p
->start_sect
);
2053 printk("%s: fail for partition %d\n", __func__
, bio
->bi_partno
);
2062 * Check whether this bio extends beyond the end of the device.
2064 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
2071 /* Test device or partition size, when known. */
2072 maxsector
= get_capacity(bio
->bi_disk
);
2074 sector_t sector
= bio
->bi_iter
.bi_sector
;
2076 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
2078 * This may well happen - the kernel calls bread()
2079 * without checking the size of the device, e.g., when
2080 * mounting a device.
2082 handle_bad_sector(bio
);
2090 static noinline_for_stack
bool
2091 generic_make_request_checks(struct bio
*bio
)
2093 struct request_queue
*q
;
2094 int nr_sectors
= bio_sectors(bio
);
2095 blk_status_t status
= BLK_STS_IOERR
;
2096 char b
[BDEVNAME_SIZE
];
2100 if (bio_check_eod(bio
, nr_sectors
))
2103 q
= bio
->bi_disk
->queue
;
2106 "generic_make_request: Trying to access "
2107 "nonexistent block-device %s (%Lu)\n",
2108 bio_devname(bio
, b
), (long long)bio
->bi_iter
.bi_sector
);
2113 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
2114 * if queue is not a request based queue.
2117 if ((bio
->bi_opf
& REQ_NOWAIT
) && !queue_is_rq_based(q
))
2120 if (should_fail_request(&bio
->bi_disk
->part0
, bio
->bi_iter
.bi_size
))
2123 if (blk_partition_remap(bio
))
2126 if (bio_check_eod(bio
, nr_sectors
))
2130 * Filter flush bio's early so that make_request based
2131 * drivers without flush support don't have to worry
2134 if (op_is_flush(bio
->bi_opf
) &&
2135 !test_bit(QUEUE_FLAG_WC
, &q
->queue_flags
)) {
2136 bio
->bi_opf
&= ~(REQ_PREFLUSH
| REQ_FUA
);
2138 status
= BLK_STS_OK
;
2143 switch (bio_op(bio
)) {
2144 case REQ_OP_DISCARD
:
2145 if (!blk_queue_discard(q
))
2148 case REQ_OP_SECURE_ERASE
:
2149 if (!blk_queue_secure_erase(q
))
2152 case REQ_OP_WRITE_SAME
:
2153 if (!q
->limits
.max_write_same_sectors
)
2156 case REQ_OP_ZONE_REPORT
:
2157 case REQ_OP_ZONE_RESET
:
2158 if (!blk_queue_is_zoned(q
))
2161 case REQ_OP_WRITE_ZEROES
:
2162 if (!q
->limits
.max_write_zeroes_sectors
)
2170 * Various block parts want %current->io_context and lazy ioc
2171 * allocation ends up trading a lot of pain for a small amount of
2172 * memory. Just allocate it upfront. This may fail and block
2173 * layer knows how to live with it.
2175 create_io_context(GFP_ATOMIC
, q
->node
);
2177 if (!blkcg_bio_issue_check(q
, bio
))
2180 if (!bio_flagged(bio
, BIO_TRACE_COMPLETION
)) {
2181 trace_block_bio_queue(q
, bio
);
2182 /* Now that enqueuing has been traced, we need to trace
2183 * completion as well.
2185 bio_set_flag(bio
, BIO_TRACE_COMPLETION
);
2190 status
= BLK_STS_NOTSUPP
;
2192 bio
->bi_status
= status
;
2198 * generic_make_request - hand a buffer to its device driver for I/O
2199 * @bio: The bio describing the location in memory and on the device.
2201 * generic_make_request() is used to make I/O requests of block
2202 * devices. It is passed a &struct bio, which describes the I/O that needs
2205 * generic_make_request() does not return any status. The
2206 * success/failure status of the request, along with notification of
2207 * completion, is delivered asynchronously through the bio->bi_end_io
2208 * function described (one day) else where.
2210 * The caller of generic_make_request must make sure that bi_io_vec
2211 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2212 * set to describe the device address, and the
2213 * bi_end_io and optionally bi_private are set to describe how
2214 * completion notification should be signaled.
2216 * generic_make_request and the drivers it calls may use bi_next if this
2217 * bio happens to be merged with someone else, and may resubmit the bio to
2218 * a lower device by calling into generic_make_request recursively, which
2219 * means the bio should NOT be touched after the call to ->make_request_fn.
2221 blk_qc_t
generic_make_request(struct bio
*bio
)
2224 * bio_list_on_stack[0] contains bios submitted by the current
2226 * bio_list_on_stack[1] contains bios that were submitted before
2227 * the current make_request_fn, but that haven't been processed
2230 struct bio_list bio_list_on_stack
[2];
2231 blk_qc_t ret
= BLK_QC_T_NONE
;
2233 if (!generic_make_request_checks(bio
))
2237 * We only want one ->make_request_fn to be active at a time, else
2238 * stack usage with stacked devices could be a problem. So use
2239 * current->bio_list to keep a list of requests submited by a
2240 * make_request_fn function. current->bio_list is also used as a
2241 * flag to say if generic_make_request is currently active in this
2242 * task or not. If it is NULL, then no make_request is active. If
2243 * it is non-NULL, then a make_request is active, and new requests
2244 * should be added at the tail
2246 if (current
->bio_list
) {
2247 bio_list_add(¤t
->bio_list
[0], bio
);
2251 /* following loop may be a bit non-obvious, and so deserves some
2253 * Before entering the loop, bio->bi_next is NULL (as all callers
2254 * ensure that) so we have a list with a single bio.
2255 * We pretend that we have just taken it off a longer list, so
2256 * we assign bio_list to a pointer to the bio_list_on_stack,
2257 * thus initialising the bio_list of new bios to be
2258 * added. ->make_request() may indeed add some more bios
2259 * through a recursive call to generic_make_request. If it
2260 * did, we find a non-NULL value in bio_list and re-enter the loop
2261 * from the top. In this case we really did just take the bio
2262 * of the top of the list (no pretending) and so remove it from
2263 * bio_list, and call into ->make_request() again.
2265 BUG_ON(bio
->bi_next
);
2266 bio_list_init(&bio_list_on_stack
[0]);
2267 current
->bio_list
= bio_list_on_stack
;
2269 struct request_queue
*q
= bio
->bi_disk
->queue
;
2271 if (likely(blk_queue_enter(q
, bio
->bi_opf
& REQ_NOWAIT
) == 0)) {
2272 struct bio_list lower
, same
;
2274 /* Create a fresh bio_list for all subordinate requests */
2275 bio_list_on_stack
[1] = bio_list_on_stack
[0];
2276 bio_list_init(&bio_list_on_stack
[0]);
2277 ret
= q
->make_request_fn(q
, bio
);
2281 /* sort new bios into those for a lower level
2282 * and those for the same level
2284 bio_list_init(&lower
);
2285 bio_list_init(&same
);
2286 while ((bio
= bio_list_pop(&bio_list_on_stack
[0])) != NULL
)
2287 if (q
== bio
->bi_disk
->queue
)
2288 bio_list_add(&same
, bio
);
2290 bio_list_add(&lower
, bio
);
2291 /* now assemble so we handle the lowest level first */
2292 bio_list_merge(&bio_list_on_stack
[0], &lower
);
2293 bio_list_merge(&bio_list_on_stack
[0], &same
);
2294 bio_list_merge(&bio_list_on_stack
[0], &bio_list_on_stack
[1]);
2296 if (unlikely(!blk_queue_dying(q
) &&
2297 (bio
->bi_opf
& REQ_NOWAIT
)))
2298 bio_wouldblock_error(bio
);
2302 bio
= bio_list_pop(&bio_list_on_stack
[0]);
2304 current
->bio_list
= NULL
; /* deactivate */
2309 EXPORT_SYMBOL(generic_make_request
);
2312 * direct_make_request - hand a buffer directly to its device driver for I/O
2313 * @bio: The bio describing the location in memory and on the device.
2315 * This function behaves like generic_make_request(), but does not protect
2316 * against recursion. Must only be used if the called driver is known
2317 * to not call generic_make_request (or direct_make_request) again from
2318 * its make_request function. (Calling direct_make_request again from
2319 * a workqueue is perfectly fine as that doesn't recurse).
2321 blk_qc_t
direct_make_request(struct bio
*bio
)
2323 struct request_queue
*q
= bio
->bi_disk
->queue
;
2324 bool nowait
= bio
->bi_opf
& REQ_NOWAIT
;
2327 if (!generic_make_request_checks(bio
))
2328 return BLK_QC_T_NONE
;
2330 if (unlikely(blk_queue_enter(q
, nowait
))) {
2331 if (nowait
&& !blk_queue_dying(q
))
2332 bio
->bi_status
= BLK_STS_AGAIN
;
2334 bio
->bi_status
= BLK_STS_IOERR
;
2336 return BLK_QC_T_NONE
;
2339 ret
= q
->make_request_fn(q
, bio
);
2343 EXPORT_SYMBOL_GPL(direct_make_request
);
2346 * submit_bio - submit a bio to the block device layer for I/O
2347 * @bio: The &struct bio which describes the I/O
2349 * submit_bio() is very similar in purpose to generic_make_request(), and
2350 * uses that function to do most of the work. Both are fairly rough
2351 * interfaces; @bio must be presetup and ready for I/O.
2354 blk_qc_t
submit_bio(struct bio
*bio
)
2357 * If it's a regular read/write or a barrier with data attached,
2358 * go through the normal accounting stuff before submission.
2360 if (bio_has_data(bio
)) {
2363 if (unlikely(bio_op(bio
) == REQ_OP_WRITE_SAME
))
2364 count
= queue_logical_block_size(bio
->bi_disk
->queue
);
2366 count
= bio_sectors(bio
);
2368 if (op_is_write(bio_op(bio
))) {
2369 count_vm_events(PGPGOUT
, count
);
2371 task_io_account_read(bio
->bi_iter
.bi_size
);
2372 count_vm_events(PGPGIN
, count
);
2375 if (unlikely(block_dump
)) {
2376 char b
[BDEVNAME_SIZE
];
2377 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s (%u sectors)\n",
2378 current
->comm
, task_pid_nr(current
),
2379 op_is_write(bio_op(bio
)) ? "WRITE" : "READ",
2380 (unsigned long long)bio
->bi_iter
.bi_sector
,
2381 bio_devname(bio
, b
), count
);
2385 return generic_make_request(bio
);
2387 EXPORT_SYMBOL(submit_bio
);
2389 bool blk_poll(struct request_queue
*q
, blk_qc_t cookie
)
2391 if (!q
->poll_fn
|| !blk_qc_t_valid(cookie
))
2395 blk_flush_plug_list(current
->plug
, false);
2396 return q
->poll_fn(q
, cookie
);
2398 EXPORT_SYMBOL_GPL(blk_poll
);
2401 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2402 * for new the queue limits
2404 * @rq: the request being checked
2407 * @rq may have been made based on weaker limitations of upper-level queues
2408 * in request stacking drivers, and it may violate the limitation of @q.
2409 * Since the block layer and the underlying device driver trust @rq
2410 * after it is inserted to @q, it should be checked against @q before
2411 * the insertion using this generic function.
2413 * Request stacking drivers like request-based dm may change the queue
2414 * limits when retrying requests on other queues. Those requests need
2415 * to be checked against the new queue limits again during dispatch.
2417 static int blk_cloned_rq_check_limits(struct request_queue
*q
,
2420 if (blk_rq_sectors(rq
) > blk_queue_get_max_sectors(q
, req_op(rq
))) {
2421 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
2426 * queue's settings related to segment counting like q->bounce_pfn
2427 * may differ from that of other stacking queues.
2428 * Recalculate it to check the request correctly on this queue's
2431 blk_recalc_rq_segments(rq
);
2432 if (rq
->nr_phys_segments
> queue_max_segments(q
)) {
2433 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
2441 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2442 * @q: the queue to submit the request
2443 * @rq: the request being queued
2445 blk_status_t
blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
2447 unsigned long flags
;
2448 int where
= ELEVATOR_INSERT_BACK
;
2450 if (blk_cloned_rq_check_limits(q
, rq
))
2451 return BLK_STS_IOERR
;
2454 should_fail_request(&rq
->rq_disk
->part0
, blk_rq_bytes(rq
)))
2455 return BLK_STS_IOERR
;
2458 if (blk_queue_io_stat(q
))
2459 blk_account_io_start(rq
, true);
2461 * Since we have a scheduler attached on the top device,
2462 * bypass a potential scheduler on the bottom device for
2465 blk_mq_request_bypass_insert(rq
, true);
2469 spin_lock_irqsave(q
->queue_lock
, flags
);
2470 if (unlikely(blk_queue_dying(q
))) {
2471 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2472 return BLK_STS_IOERR
;
2476 * Submitting request must be dequeued before calling this function
2477 * because it will be linked to another request_queue
2479 BUG_ON(blk_queued_rq(rq
));
2481 if (op_is_flush(rq
->cmd_flags
))
2482 where
= ELEVATOR_INSERT_FLUSH
;
2484 add_acct_request(q
, rq
, where
);
2485 if (where
== ELEVATOR_INSERT_FLUSH
)
2487 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2491 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
2494 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2495 * @rq: request to examine
2498 * A request could be merge of IOs which require different failure
2499 * handling. This function determines the number of bytes which
2500 * can be failed from the beginning of the request without
2501 * crossing into area which need to be retried further.
2504 * The number of bytes to fail.
2506 unsigned int blk_rq_err_bytes(const struct request
*rq
)
2508 unsigned int ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
2509 unsigned int bytes
= 0;
2512 if (!(rq
->rq_flags
& RQF_MIXED_MERGE
))
2513 return blk_rq_bytes(rq
);
2516 * Currently the only 'mixing' which can happen is between
2517 * different fastfail types. We can safely fail portions
2518 * which have all the failfast bits that the first one has -
2519 * the ones which are at least as eager to fail as the first
2522 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
2523 if ((bio
->bi_opf
& ff
) != ff
)
2525 bytes
+= bio
->bi_iter
.bi_size
;
2528 /* this could lead to infinite loop */
2529 BUG_ON(blk_rq_bytes(rq
) && !bytes
);
2532 EXPORT_SYMBOL_GPL(blk_rq_err_bytes
);
2534 void blk_account_io_completion(struct request
*req
, unsigned int bytes
)
2536 if (blk_do_io_stat(req
)) {
2537 const int rw
= rq_data_dir(req
);
2538 struct hd_struct
*part
;
2541 cpu
= part_stat_lock();
2543 part_stat_add(cpu
, part
, sectors
[rw
], bytes
>> 9);
2548 void blk_account_io_done(struct request
*req
)
2551 * Account IO completion. flush_rq isn't accounted as a
2552 * normal IO on queueing nor completion. Accounting the
2553 * containing request is enough.
2555 if (blk_do_io_stat(req
) && !(req
->rq_flags
& RQF_FLUSH_SEQ
)) {
2556 unsigned long duration
= jiffies
- req
->start_time
;
2557 const int rw
= rq_data_dir(req
);
2558 struct hd_struct
*part
;
2561 cpu
= part_stat_lock();
2564 part_stat_inc(cpu
, part
, ios
[rw
]);
2565 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
2566 part_round_stats(req
->q
, cpu
, part
);
2567 part_dec_in_flight(req
->q
, part
, rw
);
2569 hd_struct_put(part
);
2576 * Don't process normal requests when queue is suspended
2577 * or in the process of suspending/resuming
2579 static bool blk_pm_allow_request(struct request
*rq
)
2581 switch (rq
->q
->rpm_status
) {
2583 case RPM_SUSPENDING
:
2584 return rq
->rq_flags
& RQF_PM
;
2592 static bool blk_pm_allow_request(struct request
*rq
)
2598 void blk_account_io_start(struct request
*rq
, bool new_io
)
2600 struct hd_struct
*part
;
2601 int rw
= rq_data_dir(rq
);
2604 if (!blk_do_io_stat(rq
))
2607 cpu
= part_stat_lock();
2611 part_stat_inc(cpu
, part
, merges
[rw
]);
2613 part
= disk_map_sector_rcu(rq
->rq_disk
, blk_rq_pos(rq
));
2614 if (!hd_struct_try_get(part
)) {
2616 * The partition is already being removed,
2617 * the request will be accounted on the disk only
2619 * We take a reference on disk->part0 although that
2620 * partition will never be deleted, so we can treat
2621 * it as any other partition.
2623 part
= &rq
->rq_disk
->part0
;
2624 hd_struct_get(part
);
2626 part_round_stats(rq
->q
, cpu
, part
);
2627 part_inc_in_flight(rq
->q
, part
, rw
);
2634 static struct request
*elv_next_request(struct request_queue
*q
)
2637 struct blk_flush_queue
*fq
= blk_get_flush_queue(q
, NULL
);
2639 WARN_ON_ONCE(q
->mq_ops
);
2642 list_for_each_entry(rq
, &q
->queue_head
, queuelist
) {
2643 if (blk_pm_allow_request(rq
))
2646 if (rq
->rq_flags
& RQF_SOFTBARRIER
)
2651 * Flush request is running and flush request isn't queueable
2652 * in the drive, we can hold the queue till flush request is
2653 * finished. Even we don't do this, driver can't dispatch next
2654 * requests and will requeue them. And this can improve
2655 * throughput too. For example, we have request flush1, write1,
2656 * flush 2. flush1 is dispatched, then queue is hold, write1
2657 * isn't inserted to queue. After flush1 is finished, flush2
2658 * will be dispatched. Since disk cache is already clean,
2659 * flush2 will be finished very soon, so looks like flush2 is
2661 * Since the queue is hold, a flag is set to indicate the queue
2662 * should be restarted later. Please see flush_end_io() for
2665 if (fq
->flush_pending_idx
!= fq
->flush_running_idx
&&
2666 !queue_flush_queueable(q
)) {
2667 fq
->flush_queue_delayed
= 1;
2670 if (unlikely(blk_queue_bypass(q
)) ||
2671 !q
->elevator
->type
->ops
.sq
.elevator_dispatch_fn(q
, 0))
2677 * blk_peek_request - peek at the top of a request queue
2678 * @q: request queue to peek at
2681 * Return the request at the top of @q. The returned request
2682 * should be started using blk_start_request() before LLD starts
2686 * Pointer to the request at the top of @q if available. Null
2689 struct request
*blk_peek_request(struct request_queue
*q
)
2694 lockdep_assert_held(q
->queue_lock
);
2695 WARN_ON_ONCE(q
->mq_ops
);
2697 while ((rq
= elv_next_request(q
)) != NULL
) {
2698 if (!(rq
->rq_flags
& RQF_STARTED
)) {
2700 * This is the first time the device driver
2701 * sees this request (possibly after
2702 * requeueing). Notify IO scheduler.
2704 if (rq
->rq_flags
& RQF_SORTED
)
2705 elv_activate_rq(q
, rq
);
2708 * just mark as started even if we don't start
2709 * it, a request that has been delayed should
2710 * not be passed by new incoming requests
2712 rq
->rq_flags
|= RQF_STARTED
;
2713 trace_block_rq_issue(q
, rq
);
2716 if (!q
->boundary_rq
|| q
->boundary_rq
== rq
) {
2717 q
->end_sector
= rq_end_sector(rq
);
2718 q
->boundary_rq
= NULL
;
2721 if (rq
->rq_flags
& RQF_DONTPREP
)
2724 if (q
->dma_drain_size
&& blk_rq_bytes(rq
)) {
2726 * make sure space for the drain appears we
2727 * know we can do this because max_hw_segments
2728 * has been adjusted to be one fewer than the
2731 rq
->nr_phys_segments
++;
2737 ret
= q
->prep_rq_fn(q
, rq
);
2738 if (ret
== BLKPREP_OK
) {
2740 } else if (ret
== BLKPREP_DEFER
) {
2742 * the request may have been (partially) prepped.
2743 * we need to keep this request in the front to
2744 * avoid resource deadlock. RQF_STARTED will
2745 * prevent other fs requests from passing this one.
2747 if (q
->dma_drain_size
&& blk_rq_bytes(rq
) &&
2748 !(rq
->rq_flags
& RQF_DONTPREP
)) {
2750 * remove the space for the drain we added
2751 * so that we don't add it again
2753 --rq
->nr_phys_segments
;
2758 } else if (ret
== BLKPREP_KILL
|| ret
== BLKPREP_INVALID
) {
2759 rq
->rq_flags
|= RQF_QUIET
;
2761 * Mark this request as started so we don't trigger
2762 * any debug logic in the end I/O path.
2764 blk_start_request(rq
);
2765 __blk_end_request_all(rq
, ret
== BLKPREP_INVALID
?
2766 BLK_STS_TARGET
: BLK_STS_IOERR
);
2768 printk(KERN_ERR
"%s: bad return=%d\n", __func__
, ret
);
2775 EXPORT_SYMBOL(blk_peek_request
);
2777 static void blk_dequeue_request(struct request
*rq
)
2779 struct request_queue
*q
= rq
->q
;
2781 BUG_ON(list_empty(&rq
->queuelist
));
2782 BUG_ON(ELV_ON_HASH(rq
));
2784 list_del_init(&rq
->queuelist
);
2787 * the time frame between a request being removed from the lists
2788 * and to it is freed is accounted as io that is in progress at
2791 if (blk_account_rq(rq
)) {
2792 q
->in_flight
[rq_is_sync(rq
)]++;
2793 set_io_start_time_ns(rq
);
2798 * blk_start_request - start request processing on the driver
2799 * @req: request to dequeue
2802 * Dequeue @req and start timeout timer on it. This hands off the
2803 * request to the driver.
2805 void blk_start_request(struct request
*req
)
2807 lockdep_assert_held(req
->q
->queue_lock
);
2808 WARN_ON_ONCE(req
->q
->mq_ops
);
2810 blk_dequeue_request(req
);
2812 if (test_bit(QUEUE_FLAG_STATS
, &req
->q
->queue_flags
)) {
2813 blk_stat_set_issue(&req
->issue_stat
, blk_rq_sectors(req
));
2814 req
->rq_flags
|= RQF_STATS
;
2815 wbt_issue(req
->q
->rq_wb
, &req
->issue_stat
);
2818 BUG_ON(test_bit(REQ_ATOM_COMPLETE
, &req
->atomic_flags
));
2821 EXPORT_SYMBOL(blk_start_request
);
2824 * blk_fetch_request - fetch a request from a request queue
2825 * @q: request queue to fetch a request from
2828 * Return the request at the top of @q. The request is started on
2829 * return and LLD can start processing it immediately.
2832 * Pointer to the request at the top of @q if available. Null
2835 struct request
*blk_fetch_request(struct request_queue
*q
)
2839 lockdep_assert_held(q
->queue_lock
);
2840 WARN_ON_ONCE(q
->mq_ops
);
2842 rq
= blk_peek_request(q
);
2844 blk_start_request(rq
);
2847 EXPORT_SYMBOL(blk_fetch_request
);
2850 * Steal bios from a request and add them to a bio list.
2851 * The request must not have been partially completed before.
2853 void blk_steal_bios(struct bio_list
*list
, struct request
*rq
)
2857 list
->tail
->bi_next
= rq
->bio
;
2859 list
->head
= rq
->bio
;
2860 list
->tail
= rq
->biotail
;
2868 EXPORT_SYMBOL_GPL(blk_steal_bios
);
2871 * blk_update_request - Special helper function for request stacking drivers
2872 * @req: the request being processed
2873 * @error: block status code
2874 * @nr_bytes: number of bytes to complete @req
2877 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2878 * the request structure even if @req doesn't have leftover.
2879 * If @req has leftover, sets it up for the next range of segments.
2881 * This special helper function is only for request stacking drivers
2882 * (e.g. request-based dm) so that they can handle partial completion.
2883 * Actual device drivers should use blk_end_request instead.
2885 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2886 * %false return from this function.
2889 * %false - this request doesn't have any more data
2890 * %true - this request has more data
2892 bool blk_update_request(struct request
*req
, blk_status_t error
,
2893 unsigned int nr_bytes
)
2897 trace_block_rq_complete(req
, blk_status_to_errno(error
), nr_bytes
);
2902 if (unlikely(error
&& !blk_rq_is_passthrough(req
) &&
2903 !(req
->rq_flags
& RQF_QUIET
)))
2904 print_req_error(req
, error
);
2906 blk_account_io_completion(req
, nr_bytes
);
2910 struct bio
*bio
= req
->bio
;
2911 unsigned bio_bytes
= min(bio
->bi_iter
.bi_size
, nr_bytes
);
2913 if (bio_bytes
== bio
->bi_iter
.bi_size
)
2914 req
->bio
= bio
->bi_next
;
2916 /* Completion has already been traced */
2917 bio_clear_flag(bio
, BIO_TRACE_COMPLETION
);
2918 req_bio_endio(req
, bio
, bio_bytes
, error
);
2920 total_bytes
+= bio_bytes
;
2921 nr_bytes
-= bio_bytes
;
2932 * Reset counters so that the request stacking driver
2933 * can find how many bytes remain in the request
2936 req
->__data_len
= 0;
2940 req
->__data_len
-= total_bytes
;
2942 /* update sector only for requests with clear definition of sector */
2943 if (!blk_rq_is_passthrough(req
))
2944 req
->__sector
+= total_bytes
>> 9;
2946 /* mixed attributes always follow the first bio */
2947 if (req
->rq_flags
& RQF_MIXED_MERGE
) {
2948 req
->cmd_flags
&= ~REQ_FAILFAST_MASK
;
2949 req
->cmd_flags
|= req
->bio
->bi_opf
& REQ_FAILFAST_MASK
;
2952 if (!(req
->rq_flags
& RQF_SPECIAL_PAYLOAD
)) {
2954 * If total number of sectors is less than the first segment
2955 * size, something has gone terribly wrong.
2957 if (blk_rq_bytes(req
) < blk_rq_cur_bytes(req
)) {
2958 blk_dump_rq_flags(req
, "request botched");
2959 req
->__data_len
= blk_rq_cur_bytes(req
);
2962 /* recalculate the number of segments */
2963 blk_recalc_rq_segments(req
);
2968 EXPORT_SYMBOL_GPL(blk_update_request
);
2970 static bool blk_update_bidi_request(struct request
*rq
, blk_status_t error
,
2971 unsigned int nr_bytes
,
2972 unsigned int bidi_bytes
)
2974 if (blk_update_request(rq
, error
, nr_bytes
))
2977 /* Bidi request must be completed as a whole */
2978 if (unlikely(blk_bidi_rq(rq
)) &&
2979 blk_update_request(rq
->next_rq
, error
, bidi_bytes
))
2982 if (blk_queue_add_random(rq
->q
))
2983 add_disk_randomness(rq
->rq_disk
);
2989 * blk_unprep_request - unprepare a request
2992 * This function makes a request ready for complete resubmission (or
2993 * completion). It happens only after all error handling is complete,
2994 * so represents the appropriate moment to deallocate any resources
2995 * that were allocated to the request in the prep_rq_fn. The queue
2996 * lock is held when calling this.
2998 void blk_unprep_request(struct request
*req
)
3000 struct request_queue
*q
= req
->q
;
3002 req
->rq_flags
&= ~RQF_DONTPREP
;
3003 if (q
->unprep_rq_fn
)
3004 q
->unprep_rq_fn(q
, req
);
3006 EXPORT_SYMBOL_GPL(blk_unprep_request
);
3008 void blk_finish_request(struct request
*req
, blk_status_t error
)
3010 struct request_queue
*q
= req
->q
;
3012 lockdep_assert_held(req
->q
->queue_lock
);
3013 WARN_ON_ONCE(q
->mq_ops
);
3015 if (req
->rq_flags
& RQF_STATS
)
3018 if (req
->rq_flags
& RQF_QUEUED
)
3019 blk_queue_end_tag(q
, req
);
3021 BUG_ON(blk_queued_rq(req
));
3023 if (unlikely(laptop_mode
) && !blk_rq_is_passthrough(req
))
3024 laptop_io_completion(req
->q
->backing_dev_info
);
3026 blk_delete_timer(req
);
3028 if (req
->rq_flags
& RQF_DONTPREP
)
3029 blk_unprep_request(req
);
3031 blk_account_io_done(req
);
3034 wbt_done(req
->q
->rq_wb
, &req
->issue_stat
);
3035 req
->end_io(req
, error
);
3037 if (blk_bidi_rq(req
))
3038 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
3040 __blk_put_request(q
, req
);
3043 EXPORT_SYMBOL(blk_finish_request
);
3046 * blk_end_bidi_request - Complete a bidi request
3047 * @rq: the request to complete
3048 * @error: block status code
3049 * @nr_bytes: number of bytes to complete @rq
3050 * @bidi_bytes: number of bytes to complete @rq->next_rq
3053 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
3054 * Drivers that supports bidi can safely call this member for any
3055 * type of request, bidi or uni. In the later case @bidi_bytes is
3059 * %false - we are done with this request
3060 * %true - still buffers pending for this request
3062 static bool blk_end_bidi_request(struct request
*rq
, blk_status_t error
,
3063 unsigned int nr_bytes
, unsigned int bidi_bytes
)
3065 struct request_queue
*q
= rq
->q
;
3066 unsigned long flags
;
3068 WARN_ON_ONCE(q
->mq_ops
);
3070 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
3073 spin_lock_irqsave(q
->queue_lock
, flags
);
3074 blk_finish_request(rq
, error
);
3075 spin_unlock_irqrestore(q
->queue_lock
, flags
);
3081 * __blk_end_bidi_request - Complete a bidi request with queue lock held
3082 * @rq: the request to complete
3083 * @error: block status code
3084 * @nr_bytes: number of bytes to complete @rq
3085 * @bidi_bytes: number of bytes to complete @rq->next_rq
3088 * Identical to blk_end_bidi_request() except that queue lock is
3089 * assumed to be locked on entry and remains so on return.
3092 * %false - we are done with this request
3093 * %true - still buffers pending for this request
3095 static bool __blk_end_bidi_request(struct request
*rq
, blk_status_t error
,
3096 unsigned int nr_bytes
, unsigned int bidi_bytes
)
3098 lockdep_assert_held(rq
->q
->queue_lock
);
3099 WARN_ON_ONCE(rq
->q
->mq_ops
);
3101 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
3104 blk_finish_request(rq
, error
);
3110 * blk_end_request - Helper function for drivers to complete the request.
3111 * @rq: the request being processed
3112 * @error: block status code
3113 * @nr_bytes: number of bytes to complete
3116 * Ends I/O on a number of bytes attached to @rq.
3117 * If @rq has leftover, sets it up for the next range of segments.
3120 * %false - we are done with this request
3121 * %true - still buffers pending for this request
3123 bool blk_end_request(struct request
*rq
, blk_status_t error
,
3124 unsigned int nr_bytes
)
3126 WARN_ON_ONCE(rq
->q
->mq_ops
);
3127 return blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
3129 EXPORT_SYMBOL(blk_end_request
);
3132 * blk_end_request_all - Helper function for drives to finish the request.
3133 * @rq: the request to finish
3134 * @error: block status code
3137 * Completely finish @rq.
3139 void blk_end_request_all(struct request
*rq
, blk_status_t error
)
3142 unsigned int bidi_bytes
= 0;
3144 if (unlikely(blk_bidi_rq(rq
)))
3145 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
3147 pending
= blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
3150 EXPORT_SYMBOL(blk_end_request_all
);
3153 * __blk_end_request - Helper function for drivers to complete the request.
3154 * @rq: the request being processed
3155 * @error: block status code
3156 * @nr_bytes: number of bytes to complete
3159 * Must be called with queue lock held unlike blk_end_request().
3162 * %false - we are done with this request
3163 * %true - still buffers pending for this request
3165 bool __blk_end_request(struct request
*rq
, blk_status_t error
,
3166 unsigned int nr_bytes
)
3168 lockdep_assert_held(rq
->q
->queue_lock
);
3169 WARN_ON_ONCE(rq
->q
->mq_ops
);
3171 return __blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
3173 EXPORT_SYMBOL(__blk_end_request
);
3176 * __blk_end_request_all - Helper function for drives to finish the request.
3177 * @rq: the request to finish
3178 * @error: block status code
3181 * Completely finish @rq. Must be called with queue lock held.
3183 void __blk_end_request_all(struct request
*rq
, blk_status_t error
)
3186 unsigned int bidi_bytes
= 0;
3188 lockdep_assert_held(rq
->q
->queue_lock
);
3189 WARN_ON_ONCE(rq
->q
->mq_ops
);
3191 if (unlikely(blk_bidi_rq(rq
)))
3192 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
3194 pending
= __blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
3197 EXPORT_SYMBOL(__blk_end_request_all
);
3200 * __blk_end_request_cur - Helper function to finish the current request chunk.
3201 * @rq: the request to finish the current chunk for
3202 * @error: block status code
3205 * Complete the current consecutively mapped chunk from @rq. Must
3206 * be called with queue lock held.
3209 * %false - we are done with this request
3210 * %true - still buffers pending for this request
3212 bool __blk_end_request_cur(struct request
*rq
, blk_status_t error
)
3214 return __blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
3216 EXPORT_SYMBOL(__blk_end_request_cur
);
3218 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
3221 if (bio_has_data(bio
))
3222 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
3224 rq
->__data_len
= bio
->bi_iter
.bi_size
;
3225 rq
->bio
= rq
->biotail
= bio
;
3228 rq
->rq_disk
= bio
->bi_disk
;
3231 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
3233 * rq_flush_dcache_pages - Helper function to flush all pages in a request
3234 * @rq: the request to be flushed
3237 * Flush all pages in @rq.
3239 void rq_flush_dcache_pages(struct request
*rq
)
3241 struct req_iterator iter
;
3242 struct bio_vec bvec
;
3244 rq_for_each_segment(bvec
, rq
, iter
)
3245 flush_dcache_page(bvec
.bv_page
);
3247 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages
);
3251 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
3252 * @q : the queue of the device being checked
3255 * Check if underlying low-level drivers of a device are busy.
3256 * If the drivers want to export their busy state, they must set own
3257 * exporting function using blk_queue_lld_busy() first.
3259 * Basically, this function is used only by request stacking drivers
3260 * to stop dispatching requests to underlying devices when underlying
3261 * devices are busy. This behavior helps more I/O merging on the queue
3262 * of the request stacking driver and prevents I/O throughput regression
3263 * on burst I/O load.
3266 * 0 - Not busy (The request stacking driver should dispatch request)
3267 * 1 - Busy (The request stacking driver should stop dispatching request)
3269 int blk_lld_busy(struct request_queue
*q
)
3272 return q
->lld_busy_fn(q
);
3276 EXPORT_SYMBOL_GPL(blk_lld_busy
);
3279 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
3280 * @rq: the clone request to be cleaned up
3283 * Free all bios in @rq for a cloned request.
3285 void blk_rq_unprep_clone(struct request
*rq
)
3289 while ((bio
= rq
->bio
) != NULL
) {
3290 rq
->bio
= bio
->bi_next
;
3295 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone
);
3298 * Copy attributes of the original request to the clone request.
3299 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3301 static void __blk_rq_prep_clone(struct request
*dst
, struct request
*src
)
3303 dst
->cpu
= src
->cpu
;
3304 dst
->__sector
= blk_rq_pos(src
);
3305 dst
->__data_len
= blk_rq_bytes(src
);
3306 dst
->nr_phys_segments
= src
->nr_phys_segments
;
3307 dst
->ioprio
= src
->ioprio
;
3308 dst
->extra_len
= src
->extra_len
;
3312 * blk_rq_prep_clone - Helper function to setup clone request
3313 * @rq: the request to be setup
3314 * @rq_src: original request to be cloned
3315 * @bs: bio_set that bios for clone are allocated from
3316 * @gfp_mask: memory allocation mask for bio
3317 * @bio_ctr: setup function to be called for each clone bio.
3318 * Returns %0 for success, non %0 for failure.
3319 * @data: private data to be passed to @bio_ctr
3322 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3323 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3324 * are not copied, and copying such parts is the caller's responsibility.
3325 * Also, pages which the original bios are pointing to are not copied
3326 * and the cloned bios just point same pages.
3327 * So cloned bios must be completed before original bios, which means
3328 * the caller must complete @rq before @rq_src.
3330 int blk_rq_prep_clone(struct request
*rq
, struct request
*rq_src
,
3331 struct bio_set
*bs
, gfp_t gfp_mask
,
3332 int (*bio_ctr
)(struct bio
*, struct bio
*, void *),
3335 struct bio
*bio
, *bio_src
;
3340 __rq_for_each_bio(bio_src
, rq_src
) {
3341 bio
= bio_clone_fast(bio_src
, gfp_mask
, bs
);
3345 if (bio_ctr
&& bio_ctr(bio
, bio_src
, data
))
3349 rq
->biotail
->bi_next
= bio
;
3352 rq
->bio
= rq
->biotail
= bio
;
3355 __blk_rq_prep_clone(rq
, rq_src
);
3362 blk_rq_unprep_clone(rq
);
3366 EXPORT_SYMBOL_GPL(blk_rq_prep_clone
);
3368 int kblockd_schedule_work(struct work_struct
*work
)
3370 return queue_work(kblockd_workqueue
, work
);
3372 EXPORT_SYMBOL(kblockd_schedule_work
);
3374 int kblockd_schedule_work_on(int cpu
, struct work_struct
*work
)
3376 return queue_work_on(cpu
, kblockd_workqueue
, work
);
3378 EXPORT_SYMBOL(kblockd_schedule_work_on
);
3380 int kblockd_mod_delayed_work_on(int cpu
, struct delayed_work
*dwork
,
3381 unsigned long delay
)
3383 return mod_delayed_work_on(cpu
, kblockd_workqueue
, dwork
, delay
);
3385 EXPORT_SYMBOL(kblockd_mod_delayed_work_on
);
3387 int kblockd_schedule_delayed_work(struct delayed_work
*dwork
,
3388 unsigned long delay
)
3390 return queue_delayed_work(kblockd_workqueue
, dwork
, delay
);
3392 EXPORT_SYMBOL(kblockd_schedule_delayed_work
);
3394 int kblockd_schedule_delayed_work_on(int cpu
, struct delayed_work
*dwork
,
3395 unsigned long delay
)
3397 return queue_delayed_work_on(cpu
, kblockd_workqueue
, dwork
, delay
);
3399 EXPORT_SYMBOL(kblockd_schedule_delayed_work_on
);
3402 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3403 * @plug: The &struct blk_plug that needs to be initialized
3406 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3407 * pending I/O should the task end up blocking between blk_start_plug() and
3408 * blk_finish_plug(). This is important from a performance perspective, but
3409 * also ensures that we don't deadlock. For instance, if the task is blocking
3410 * for a memory allocation, memory reclaim could end up wanting to free a
3411 * page belonging to that request that is currently residing in our private
3412 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3413 * this kind of deadlock.
3415 void blk_start_plug(struct blk_plug
*plug
)
3417 struct task_struct
*tsk
= current
;
3420 * If this is a nested plug, don't actually assign it.
3425 INIT_LIST_HEAD(&plug
->list
);
3426 INIT_LIST_HEAD(&plug
->mq_list
);
3427 INIT_LIST_HEAD(&plug
->cb_list
);
3429 * Store ordering should not be needed here, since a potential
3430 * preempt will imply a full memory barrier
3434 EXPORT_SYMBOL(blk_start_plug
);
3436 static int plug_rq_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
3438 struct request
*rqa
= container_of(a
, struct request
, queuelist
);
3439 struct request
*rqb
= container_of(b
, struct request
, queuelist
);
3441 return !(rqa
->q
< rqb
->q
||
3442 (rqa
->q
== rqb
->q
&& blk_rq_pos(rqa
) < blk_rq_pos(rqb
)));
3446 * If 'from_schedule' is true, then postpone the dispatch of requests
3447 * until a safe kblockd context. We due this to avoid accidental big
3448 * additional stack usage in driver dispatch, in places where the originally
3449 * plugger did not intend it.
3451 static void queue_unplugged(struct request_queue
*q
, unsigned int depth
,
3453 __releases(q
->queue_lock
)
3455 lockdep_assert_held(q
->queue_lock
);
3457 trace_block_unplug(q
, depth
, !from_schedule
);
3460 blk_run_queue_async(q
);
3463 spin_unlock(q
->queue_lock
);
3466 static void flush_plug_callbacks(struct blk_plug
*plug
, bool from_schedule
)
3468 LIST_HEAD(callbacks
);
3470 while (!list_empty(&plug
->cb_list
)) {
3471 list_splice_init(&plug
->cb_list
, &callbacks
);
3473 while (!list_empty(&callbacks
)) {
3474 struct blk_plug_cb
*cb
= list_first_entry(&callbacks
,
3477 list_del(&cb
->list
);
3478 cb
->callback(cb
, from_schedule
);
3483 struct blk_plug_cb
*blk_check_plugged(blk_plug_cb_fn unplug
, void *data
,
3486 struct blk_plug
*plug
= current
->plug
;
3487 struct blk_plug_cb
*cb
;
3492 list_for_each_entry(cb
, &plug
->cb_list
, list
)
3493 if (cb
->callback
== unplug
&& cb
->data
== data
)
3496 /* Not currently on the callback list */
3497 BUG_ON(size
< sizeof(*cb
));
3498 cb
= kzalloc(size
, GFP_ATOMIC
);
3501 cb
->callback
= unplug
;
3502 list_add(&cb
->list
, &plug
->cb_list
);
3506 EXPORT_SYMBOL(blk_check_plugged
);
3508 void blk_flush_plug_list(struct blk_plug
*plug
, bool from_schedule
)
3510 struct request_queue
*q
;
3511 unsigned long flags
;
3516 flush_plug_callbacks(plug
, from_schedule
);
3518 if (!list_empty(&plug
->mq_list
))
3519 blk_mq_flush_plug_list(plug
, from_schedule
);
3521 if (list_empty(&plug
->list
))
3524 list_splice_init(&plug
->list
, &list
);
3526 list_sort(NULL
, &list
, plug_rq_cmp
);
3532 * Save and disable interrupts here, to avoid doing it for every
3533 * queue lock we have to take.
3535 local_irq_save(flags
);
3536 while (!list_empty(&list
)) {
3537 rq
= list_entry_rq(list
.next
);
3538 list_del_init(&rq
->queuelist
);
3542 * This drops the queue lock
3545 queue_unplugged(q
, depth
, from_schedule
);
3548 spin_lock(q
->queue_lock
);
3552 * Short-circuit if @q is dead
3554 if (unlikely(blk_queue_dying(q
))) {
3555 __blk_end_request_all(rq
, BLK_STS_IOERR
);
3560 * rq is already accounted, so use raw insert
3562 if (op_is_flush(rq
->cmd_flags
))
3563 __elv_add_request(q
, rq
, ELEVATOR_INSERT_FLUSH
);
3565 __elv_add_request(q
, rq
, ELEVATOR_INSERT_SORT_MERGE
);
3571 * This drops the queue lock
3574 queue_unplugged(q
, depth
, from_schedule
);
3576 local_irq_restore(flags
);
3579 void blk_finish_plug(struct blk_plug
*plug
)
3581 if (plug
!= current
->plug
)
3583 blk_flush_plug_list(plug
, false);
3585 current
->plug
= NULL
;
3587 EXPORT_SYMBOL(blk_finish_plug
);
3591 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3592 * @q: the queue of the device
3593 * @dev: the device the queue belongs to
3596 * Initialize runtime-PM-related fields for @q and start auto suspend for
3597 * @dev. Drivers that want to take advantage of request-based runtime PM
3598 * should call this function after @dev has been initialized, and its
3599 * request queue @q has been allocated, and runtime PM for it can not happen
3600 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3601 * cases, driver should call this function before any I/O has taken place.
3603 * This function takes care of setting up using auto suspend for the device,
3604 * the autosuspend delay is set to -1 to make runtime suspend impossible
3605 * until an updated value is either set by user or by driver. Drivers do
3606 * not need to touch other autosuspend settings.
3608 * The block layer runtime PM is request based, so only works for drivers
3609 * that use request as their IO unit instead of those directly use bio's.
3611 void blk_pm_runtime_init(struct request_queue
*q
, struct device
*dev
)
3613 /* not support for RQF_PM and ->rpm_status in blk-mq yet */
3618 q
->rpm_status
= RPM_ACTIVE
;
3619 pm_runtime_set_autosuspend_delay(q
->dev
, -1);
3620 pm_runtime_use_autosuspend(q
->dev
);
3622 EXPORT_SYMBOL(blk_pm_runtime_init
);
3625 * blk_pre_runtime_suspend - Pre runtime suspend check
3626 * @q: the queue of the device
3629 * This function will check if runtime suspend is allowed for the device
3630 * by examining if there are any requests pending in the queue. If there
3631 * are requests pending, the device can not be runtime suspended; otherwise,
3632 * the queue's status will be updated to SUSPENDING and the driver can
3633 * proceed to suspend the device.
3635 * For the not allowed case, we mark last busy for the device so that
3636 * runtime PM core will try to autosuspend it some time later.
3638 * This function should be called near the start of the device's
3639 * runtime_suspend callback.
3642 * 0 - OK to runtime suspend the device
3643 * -EBUSY - Device should not be runtime suspended
3645 int blk_pre_runtime_suspend(struct request_queue
*q
)
3652 spin_lock_irq(q
->queue_lock
);
3653 if (q
->nr_pending
) {
3655 pm_runtime_mark_last_busy(q
->dev
);
3657 q
->rpm_status
= RPM_SUSPENDING
;
3659 spin_unlock_irq(q
->queue_lock
);
3662 EXPORT_SYMBOL(blk_pre_runtime_suspend
);
3665 * blk_post_runtime_suspend - Post runtime suspend processing
3666 * @q: the queue of the device
3667 * @err: return value of the device's runtime_suspend function
3670 * Update the queue's runtime status according to the return value of the
3671 * device's runtime suspend function and mark last busy for the device so
3672 * that PM core will try to auto suspend the device at a later time.
3674 * This function should be called near the end of the device's
3675 * runtime_suspend callback.
3677 void blk_post_runtime_suspend(struct request_queue
*q
, int err
)
3682 spin_lock_irq(q
->queue_lock
);
3684 q
->rpm_status
= RPM_SUSPENDED
;
3686 q
->rpm_status
= RPM_ACTIVE
;
3687 pm_runtime_mark_last_busy(q
->dev
);
3689 spin_unlock_irq(q
->queue_lock
);
3691 EXPORT_SYMBOL(blk_post_runtime_suspend
);
3694 * blk_pre_runtime_resume - Pre runtime resume processing
3695 * @q: the queue of the device
3698 * Update the queue's runtime status to RESUMING in preparation for the
3699 * runtime resume of the device.
3701 * This function should be called near the start of the device's
3702 * runtime_resume callback.
3704 void blk_pre_runtime_resume(struct request_queue
*q
)
3709 spin_lock_irq(q
->queue_lock
);
3710 q
->rpm_status
= RPM_RESUMING
;
3711 spin_unlock_irq(q
->queue_lock
);
3713 EXPORT_SYMBOL(blk_pre_runtime_resume
);
3716 * blk_post_runtime_resume - Post runtime resume processing
3717 * @q: the queue of the device
3718 * @err: return value of the device's runtime_resume function
3721 * Update the queue's runtime status according to the return value of the
3722 * device's runtime_resume function. If it is successfully resumed, process
3723 * the requests that are queued into the device's queue when it is resuming
3724 * and then mark last busy and initiate autosuspend for it.
3726 * This function should be called near the end of the device's
3727 * runtime_resume callback.
3729 void blk_post_runtime_resume(struct request_queue
*q
, int err
)
3734 spin_lock_irq(q
->queue_lock
);
3736 q
->rpm_status
= RPM_ACTIVE
;
3738 pm_runtime_mark_last_busy(q
->dev
);
3739 pm_request_autosuspend(q
->dev
);
3741 q
->rpm_status
= RPM_SUSPENDED
;
3743 spin_unlock_irq(q
->queue_lock
);
3745 EXPORT_SYMBOL(blk_post_runtime_resume
);
3748 * blk_set_runtime_active - Force runtime status of the queue to be active
3749 * @q: the queue of the device
3751 * If the device is left runtime suspended during system suspend the resume
3752 * hook typically resumes the device and corrects runtime status
3753 * accordingly. However, that does not affect the queue runtime PM status
3754 * which is still "suspended". This prevents processing requests from the
3757 * This function can be used in driver's resume hook to correct queue
3758 * runtime PM status and re-enable peeking requests from the queue. It
3759 * should be called before first request is added to the queue.
3761 void blk_set_runtime_active(struct request_queue
*q
)
3763 spin_lock_irq(q
->queue_lock
);
3764 q
->rpm_status
= RPM_ACTIVE
;
3765 pm_runtime_mark_last_busy(q
->dev
);
3766 pm_request_autosuspend(q
->dev
);
3767 spin_unlock_irq(q
->queue_lock
);
3769 EXPORT_SYMBOL(blk_set_runtime_active
);
3772 int __init
blk_dev_init(void)
3774 BUILD_BUG_ON(REQ_OP_LAST
>= (1 << REQ_OP_BITS
));
3775 BUILD_BUG_ON(REQ_OP_BITS
+ REQ_FLAG_BITS
> 8 *
3776 FIELD_SIZEOF(struct request
, cmd_flags
));
3777 BUILD_BUG_ON(REQ_OP_BITS
+ REQ_FLAG_BITS
> 8 *
3778 FIELD_SIZEOF(struct bio
, bi_opf
));
3780 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3781 kblockd_workqueue
= alloc_workqueue("kblockd",
3782 WQ_MEM_RECLAIM
| WQ_HIGHPRI
, 0);
3783 if (!kblockd_workqueue
)
3784 panic("Failed to create kblockd\n");
3786 request_cachep
= kmem_cache_create("blkdev_requests",
3787 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
3789 blk_requestq_cachep
= kmem_cache_create("request_queue",
3790 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);
3792 #ifdef CONFIG_DEBUG_FS
3793 blk_debugfs_root
= debugfs_create_dir("block", NULL
);