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>
37 #define CREATE_TRACE_POINTS
38 #include <trace/events/block.h>
43 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap
);
44 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap
);
45 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete
);
46 EXPORT_TRACEPOINT_SYMBOL_GPL(block_split
);
47 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug
);
49 DEFINE_IDA(blk_queue_ida
);
52 * For the allocated request tables
54 struct kmem_cache
*request_cachep
= NULL
;
57 * For queue allocation
59 struct kmem_cache
*blk_requestq_cachep
;
62 * Controlling structure to kblockd
64 static struct workqueue_struct
*kblockd_workqueue
;
66 static void blk_clear_congested(struct request_list
*rl
, int sync
)
68 #ifdef CONFIG_CGROUP_WRITEBACK
69 clear_wb_congested(rl
->blkg
->wb_congested
, sync
);
72 * If !CGROUP_WRITEBACK, all blkg's map to bdi->wb and we shouldn't
73 * flip its congestion state for events on other blkcgs.
75 if (rl
== &rl
->q
->root_rl
)
76 clear_wb_congested(rl
->q
->backing_dev_info
.wb
.congested
, sync
);
80 static void blk_set_congested(struct request_list
*rl
, int sync
)
82 #ifdef CONFIG_CGROUP_WRITEBACK
83 set_wb_congested(rl
->blkg
->wb_congested
, sync
);
85 /* see blk_clear_congested() */
86 if (rl
== &rl
->q
->root_rl
)
87 set_wb_congested(rl
->q
->backing_dev_info
.wb
.congested
, sync
);
91 void blk_queue_congestion_threshold(struct request_queue
*q
)
95 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) + 1;
96 if (nr
> q
->nr_requests
)
98 q
->nr_congestion_on
= nr
;
100 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) - (q
->nr_requests
/ 16) - 1;
103 q
->nr_congestion_off
= nr
;
107 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
110 * Locates the passed device's request queue and returns the address of its
111 * backing_dev_info. This function can only be called if @bdev is opened
112 * and the return value is never NULL.
114 struct backing_dev_info
*blk_get_backing_dev_info(struct block_device
*bdev
)
116 struct request_queue
*q
= bdev_get_queue(bdev
);
118 return &q
->backing_dev_info
;
120 EXPORT_SYMBOL(blk_get_backing_dev_info
);
122 void blk_rq_init(struct request_queue
*q
, struct request
*rq
)
124 memset(rq
, 0, sizeof(*rq
));
126 INIT_LIST_HEAD(&rq
->queuelist
);
127 INIT_LIST_HEAD(&rq
->timeout_list
);
130 rq
->__sector
= (sector_t
) -1;
131 INIT_HLIST_NODE(&rq
->hash
);
132 RB_CLEAR_NODE(&rq
->rb_node
);
134 rq
->cmd_len
= BLK_MAX_CDB
;
136 rq
->start_time
= jiffies
;
137 set_start_time_ns(rq
);
140 EXPORT_SYMBOL(blk_rq_init
);
142 static void req_bio_endio(struct request
*rq
, struct bio
*bio
,
143 unsigned int nbytes
, int error
)
146 bio
->bi_error
= error
;
148 if (unlikely(rq
->cmd_flags
& REQ_QUIET
))
149 bio_set_flag(bio
, BIO_QUIET
);
151 bio_advance(bio
, nbytes
);
153 /* don't actually finish bio if it's part of flush sequence */
154 if (bio
->bi_iter
.bi_size
== 0 && !(rq
->cmd_flags
& REQ_FLUSH_SEQ
))
158 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
162 printk(KERN_INFO
"%s: dev %s: type=%x, flags=%llx\n", msg
,
163 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->cmd_type
,
164 (unsigned long long) rq
->cmd_flags
);
166 printk(KERN_INFO
" sector %llu, nr/cnr %u/%u\n",
167 (unsigned long long)blk_rq_pos(rq
),
168 blk_rq_sectors(rq
), blk_rq_cur_sectors(rq
));
169 printk(KERN_INFO
" bio %p, biotail %p, len %u\n",
170 rq
->bio
, rq
->biotail
, blk_rq_bytes(rq
));
172 if (rq
->cmd_type
== REQ_TYPE_BLOCK_PC
) {
173 printk(KERN_INFO
" cdb: ");
174 for (bit
= 0; bit
< BLK_MAX_CDB
; bit
++)
175 printk("%02x ", rq
->cmd
[bit
]);
179 EXPORT_SYMBOL(blk_dump_rq_flags
);
181 static void blk_delay_work(struct work_struct
*work
)
183 struct request_queue
*q
;
185 q
= container_of(work
, struct request_queue
, delay_work
.work
);
186 spin_lock_irq(q
->queue_lock
);
188 spin_unlock_irq(q
->queue_lock
);
192 * blk_delay_queue - restart queueing after defined interval
193 * @q: The &struct request_queue in question
194 * @msecs: Delay in msecs
197 * Sometimes queueing needs to be postponed for a little while, to allow
198 * resources to come back. This function will make sure that queueing is
199 * restarted around the specified time. Queue lock must be held.
201 void blk_delay_queue(struct request_queue
*q
, unsigned long msecs
)
203 if (likely(!blk_queue_dead(q
)))
204 queue_delayed_work(kblockd_workqueue
, &q
->delay_work
,
205 msecs_to_jiffies(msecs
));
207 EXPORT_SYMBOL(blk_delay_queue
);
210 * blk_start_queue - restart a previously stopped queue
211 * @q: The &struct request_queue in question
214 * blk_start_queue() will clear the stop flag on the queue, and call
215 * the request_fn for the queue if it was in a stopped state when
216 * entered. Also see blk_stop_queue(). Queue lock must be held.
218 void blk_start_queue(struct request_queue
*q
)
220 WARN_ON(!irqs_disabled());
222 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
225 EXPORT_SYMBOL(blk_start_queue
);
228 * blk_stop_queue - stop a queue
229 * @q: The &struct request_queue in question
232 * The Linux block layer assumes that a block driver will consume all
233 * entries on the request queue when the request_fn strategy is called.
234 * Often this will not happen, because of hardware limitations (queue
235 * depth settings). If a device driver gets a 'queue full' response,
236 * or if it simply chooses not to queue more I/O at one point, it can
237 * call this function to prevent the request_fn from being called until
238 * the driver has signalled it's ready to go again. This happens by calling
239 * blk_start_queue() to restart queue operations. Queue lock must be held.
241 void blk_stop_queue(struct request_queue
*q
)
243 cancel_delayed_work(&q
->delay_work
);
244 queue_flag_set(QUEUE_FLAG_STOPPED
, q
);
246 EXPORT_SYMBOL(blk_stop_queue
);
249 * blk_sync_queue - cancel any pending callbacks on a queue
253 * The block layer may perform asynchronous callback activity
254 * on a queue, such as calling the unplug function after a timeout.
255 * A block device may call blk_sync_queue to ensure that any
256 * such activity is cancelled, thus allowing it to release resources
257 * that the callbacks might use. The caller must already have made sure
258 * that its ->make_request_fn will not re-add plugging prior to calling
261 * This function does not cancel any asynchronous activity arising
262 * out of elevator or throttling code. That would require elevator_exit()
263 * and blkcg_exit_queue() to be called with queue lock initialized.
266 void blk_sync_queue(struct request_queue
*q
)
268 del_timer_sync(&q
->timeout
);
271 struct blk_mq_hw_ctx
*hctx
;
274 queue_for_each_hw_ctx(q
, hctx
, i
) {
275 cancel_delayed_work_sync(&hctx
->run_work
);
276 cancel_delayed_work_sync(&hctx
->delay_work
);
279 cancel_delayed_work_sync(&q
->delay_work
);
282 EXPORT_SYMBOL(blk_sync_queue
);
285 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
286 * @q: The queue to run
289 * Invoke request handling on a queue if there are any pending requests.
290 * May be used to restart request handling after a request has completed.
291 * This variant runs the queue whether or not the queue has been
292 * stopped. Must be called with the queue lock held and interrupts
293 * disabled. See also @blk_run_queue.
295 inline void __blk_run_queue_uncond(struct request_queue
*q
)
297 if (unlikely(blk_queue_dead(q
)))
301 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
302 * the queue lock internally. As a result multiple threads may be
303 * running such a request function concurrently. Keep track of the
304 * number of active request_fn invocations such that blk_drain_queue()
305 * can wait until all these request_fn calls have finished.
307 q
->request_fn_active
++;
309 q
->request_fn_active
--;
311 EXPORT_SYMBOL_GPL(__blk_run_queue_uncond
);
314 * __blk_run_queue - run a single device queue
315 * @q: The queue to run
318 * See @blk_run_queue. This variant must be called with the queue lock
319 * held and interrupts disabled.
321 void __blk_run_queue(struct request_queue
*q
)
323 if (unlikely(blk_queue_stopped(q
)))
326 __blk_run_queue_uncond(q
);
328 EXPORT_SYMBOL(__blk_run_queue
);
331 * blk_run_queue_async - run a single device queue in workqueue context
332 * @q: The queue to run
335 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
336 * of us. The caller must hold the queue lock.
338 void blk_run_queue_async(struct request_queue
*q
)
340 if (likely(!blk_queue_stopped(q
) && !blk_queue_dead(q
)))
341 mod_delayed_work(kblockd_workqueue
, &q
->delay_work
, 0);
343 EXPORT_SYMBOL(blk_run_queue_async
);
346 * blk_run_queue - run a single device queue
347 * @q: The queue to run
350 * Invoke request handling on this queue, if it has pending work to do.
351 * May be used to restart queueing when a request has completed.
353 void blk_run_queue(struct request_queue
*q
)
357 spin_lock_irqsave(q
->queue_lock
, flags
);
359 spin_unlock_irqrestore(q
->queue_lock
, flags
);
361 EXPORT_SYMBOL(blk_run_queue
);
363 void blk_put_queue(struct request_queue
*q
)
365 kobject_put(&q
->kobj
);
367 EXPORT_SYMBOL(blk_put_queue
);
370 * __blk_drain_queue - drain requests from request_queue
372 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
374 * Drain requests from @q. If @drain_all is set, all requests are drained.
375 * If not, only ELVPRIV requests are drained. The caller is responsible
376 * for ensuring that no new requests which need to be drained are queued.
378 static void __blk_drain_queue(struct request_queue
*q
, bool drain_all
)
379 __releases(q
->queue_lock
)
380 __acquires(q
->queue_lock
)
384 lockdep_assert_held(q
->queue_lock
);
390 * The caller might be trying to drain @q before its
391 * elevator is initialized.
394 elv_drain_elevator(q
);
396 blkcg_drain_queue(q
);
399 * This function might be called on a queue which failed
400 * driver init after queue creation or is not yet fully
401 * active yet. Some drivers (e.g. fd and loop) get unhappy
402 * in such cases. Kick queue iff dispatch queue has
403 * something on it and @q has request_fn set.
405 if (!list_empty(&q
->queue_head
) && q
->request_fn
)
408 drain
|= q
->nr_rqs_elvpriv
;
409 drain
|= q
->request_fn_active
;
412 * Unfortunately, requests are queued at and tracked from
413 * multiple places and there's no single counter which can
414 * be drained. Check all the queues and counters.
417 struct blk_flush_queue
*fq
= blk_get_flush_queue(q
, NULL
);
418 drain
|= !list_empty(&q
->queue_head
);
419 for (i
= 0; i
< 2; i
++) {
420 drain
|= q
->nr_rqs
[i
];
421 drain
|= q
->in_flight
[i
];
423 drain
|= !list_empty(&fq
->flush_queue
[i
]);
430 spin_unlock_irq(q
->queue_lock
);
434 spin_lock_irq(q
->queue_lock
);
438 * With queue marked dead, any woken up waiter will fail the
439 * allocation path, so the wakeup chaining is lost and we're
440 * left with hung waiters. We need to wake up those waiters.
443 struct request_list
*rl
;
445 blk_queue_for_each_rl(rl
, q
)
446 for (i
= 0; i
< ARRAY_SIZE(rl
->wait
); i
++)
447 wake_up_all(&rl
->wait
[i
]);
452 * blk_queue_bypass_start - enter queue bypass mode
453 * @q: queue of interest
455 * In bypass mode, only the dispatch FIFO queue of @q is used. This
456 * function makes @q enter bypass mode and drains all requests which were
457 * throttled or issued before. On return, it's guaranteed that no request
458 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
459 * inside queue or RCU read lock.
461 void blk_queue_bypass_start(struct request_queue
*q
)
463 spin_lock_irq(q
->queue_lock
);
465 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
466 spin_unlock_irq(q
->queue_lock
);
469 * Queues start drained. Skip actual draining till init is
470 * complete. This avoids lenghty delays during queue init which
471 * can happen many times during boot.
473 if (blk_queue_init_done(q
)) {
474 spin_lock_irq(q
->queue_lock
);
475 __blk_drain_queue(q
, false);
476 spin_unlock_irq(q
->queue_lock
);
478 /* ensure blk_queue_bypass() is %true inside RCU read lock */
482 EXPORT_SYMBOL_GPL(blk_queue_bypass_start
);
485 * blk_queue_bypass_end - leave queue bypass mode
486 * @q: queue of interest
488 * Leave bypass mode and restore the normal queueing behavior.
490 void blk_queue_bypass_end(struct request_queue
*q
)
492 spin_lock_irq(q
->queue_lock
);
493 if (!--q
->bypass_depth
)
494 queue_flag_clear(QUEUE_FLAG_BYPASS
, q
);
495 WARN_ON_ONCE(q
->bypass_depth
< 0);
496 spin_unlock_irq(q
->queue_lock
);
498 EXPORT_SYMBOL_GPL(blk_queue_bypass_end
);
500 void blk_set_queue_dying(struct request_queue
*q
)
502 queue_flag_set_unlocked(QUEUE_FLAG_DYING
, q
);
505 blk_mq_wake_waiters(q
);
507 struct request_list
*rl
;
509 blk_queue_for_each_rl(rl
, q
) {
511 wake_up(&rl
->wait
[BLK_RW_SYNC
]);
512 wake_up(&rl
->wait
[BLK_RW_ASYNC
]);
517 EXPORT_SYMBOL_GPL(blk_set_queue_dying
);
520 * blk_cleanup_queue - shutdown a request queue
521 * @q: request queue to shutdown
523 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
524 * put it. All future requests will be failed immediately with -ENODEV.
526 void blk_cleanup_queue(struct request_queue
*q
)
528 spinlock_t
*lock
= q
->queue_lock
;
530 /* mark @q DYING, no new request or merges will be allowed afterwards */
531 mutex_lock(&q
->sysfs_lock
);
532 blk_set_queue_dying(q
);
536 * A dying queue is permanently in bypass mode till released. Note
537 * that, unlike blk_queue_bypass_start(), we aren't performing
538 * synchronize_rcu() after entering bypass mode to avoid the delay
539 * as some drivers create and destroy a lot of queues while
540 * probing. This is still safe because blk_release_queue() will be
541 * called only after the queue refcnt drops to zero and nothing,
542 * RCU or not, would be traversing the queue by then.
545 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
547 queue_flag_set(QUEUE_FLAG_NOMERGES
, q
);
548 queue_flag_set(QUEUE_FLAG_NOXMERGES
, q
);
549 queue_flag_set(QUEUE_FLAG_DYING
, q
);
550 spin_unlock_irq(lock
);
551 mutex_unlock(&q
->sysfs_lock
);
554 * Drain all requests queued before DYING marking. Set DEAD flag to
555 * prevent that q->request_fn() gets invoked after draining finished.
560 __blk_drain_queue(q
, true);
561 queue_flag_set(QUEUE_FLAG_DEAD
, q
);
562 spin_unlock_irq(lock
);
564 /* for synchronous bio-based driver finish in-flight integrity i/o */
565 blk_flush_integrity();
567 /* @q won't process any more request, flush async actions */
568 del_timer_sync(&q
->backing_dev_info
.laptop_mode_wb_timer
);
572 blk_mq_free_queue(q
);
573 percpu_ref_exit(&q
->q_usage_counter
);
576 if (q
->queue_lock
!= &q
->__queue_lock
)
577 q
->queue_lock
= &q
->__queue_lock
;
578 spin_unlock_irq(lock
);
580 bdi_unregister(&q
->backing_dev_info
);
582 /* @q is and will stay empty, shutdown and put */
585 EXPORT_SYMBOL(blk_cleanup_queue
);
587 /* Allocate memory local to the request queue */
588 static void *alloc_request_struct(gfp_t gfp_mask
, void *data
)
590 int nid
= (int)(long)data
;
591 return kmem_cache_alloc_node(request_cachep
, gfp_mask
, nid
);
594 static void free_request_struct(void *element
, void *unused
)
596 kmem_cache_free(request_cachep
, element
);
599 int blk_init_rl(struct request_list
*rl
, struct request_queue
*q
,
602 if (unlikely(rl
->rq_pool
))
606 rl
->count
[BLK_RW_SYNC
] = rl
->count
[BLK_RW_ASYNC
] = 0;
607 rl
->starved
[BLK_RW_SYNC
] = rl
->starved
[BLK_RW_ASYNC
] = 0;
608 init_waitqueue_head(&rl
->wait
[BLK_RW_SYNC
]);
609 init_waitqueue_head(&rl
->wait
[BLK_RW_ASYNC
]);
611 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, alloc_request_struct
,
613 (void *)(long)q
->node
, gfp_mask
,
621 void blk_exit_rl(struct request_list
*rl
)
624 mempool_destroy(rl
->rq_pool
);
627 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
629 return blk_alloc_queue_node(gfp_mask
, NUMA_NO_NODE
);
631 EXPORT_SYMBOL(blk_alloc_queue
);
633 int blk_queue_enter(struct request_queue
*q
, gfp_t gfp
)
638 if (percpu_ref_tryget_live(&q
->q_usage_counter
))
641 if (!gfpflags_allow_blocking(gfp
))
644 ret
= wait_event_interruptible(q
->mq_freeze_wq
,
645 !atomic_read(&q
->mq_freeze_depth
) ||
647 if (blk_queue_dying(q
))
654 void blk_queue_exit(struct request_queue
*q
)
656 percpu_ref_put(&q
->q_usage_counter
);
659 static void blk_queue_usage_counter_release(struct percpu_ref
*ref
)
661 struct request_queue
*q
=
662 container_of(ref
, struct request_queue
, q_usage_counter
);
664 wake_up_all(&q
->mq_freeze_wq
);
667 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
669 struct request_queue
*q
;
672 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
673 gfp_mask
| __GFP_ZERO
, node_id
);
677 q
->id
= ida_simple_get(&blk_queue_ida
, 0, 0, gfp_mask
);
681 q
->bio_split
= bioset_create(BIO_POOL_SIZE
, 0);
685 q
->backing_dev_info
.ra_pages
=
686 (VM_MAX_READAHEAD
* 1024) / PAGE_CACHE_SIZE
;
687 q
->backing_dev_info
.capabilities
= BDI_CAP_CGROUP_WRITEBACK
;
688 q
->backing_dev_info
.name
= "block";
691 err
= bdi_init(&q
->backing_dev_info
);
695 setup_timer(&q
->backing_dev_info
.laptop_mode_wb_timer
,
696 laptop_mode_timer_fn
, (unsigned long) q
);
697 setup_timer(&q
->timeout
, blk_rq_timed_out_timer
, (unsigned long) q
);
698 INIT_LIST_HEAD(&q
->queue_head
);
699 INIT_LIST_HEAD(&q
->timeout_list
);
700 INIT_LIST_HEAD(&q
->icq_list
);
701 #ifdef CONFIG_BLK_CGROUP
702 INIT_LIST_HEAD(&q
->blkg_list
);
704 INIT_DELAYED_WORK(&q
->delay_work
, blk_delay_work
);
706 kobject_init(&q
->kobj
, &blk_queue_ktype
);
708 mutex_init(&q
->sysfs_lock
);
709 spin_lock_init(&q
->__queue_lock
);
712 * By default initialize queue_lock to internal lock and driver can
713 * override it later if need be.
715 q
->queue_lock
= &q
->__queue_lock
;
718 * A queue starts its life with bypass turned on to avoid
719 * unnecessary bypass on/off overhead and nasty surprises during
720 * init. The initial bypass will be finished when the queue is
721 * registered by blk_register_queue().
724 __set_bit(QUEUE_FLAG_BYPASS
, &q
->queue_flags
);
726 init_waitqueue_head(&q
->mq_freeze_wq
);
729 * Init percpu_ref in atomic mode so that it's faster to shutdown.
730 * See blk_register_queue() for details.
732 if (percpu_ref_init(&q
->q_usage_counter
,
733 blk_queue_usage_counter_release
,
734 PERCPU_REF_INIT_ATOMIC
, GFP_KERNEL
))
737 if (blkcg_init_queue(q
))
743 percpu_ref_exit(&q
->q_usage_counter
);
745 bdi_destroy(&q
->backing_dev_info
);
747 bioset_free(q
->bio_split
);
749 ida_simple_remove(&blk_queue_ida
, q
->id
);
751 kmem_cache_free(blk_requestq_cachep
, q
);
754 EXPORT_SYMBOL(blk_alloc_queue_node
);
757 * blk_init_queue - prepare a request queue for use with a block device
758 * @rfn: The function to be called to process requests that have been
759 * placed on the queue.
760 * @lock: Request queue spin lock
763 * If a block device wishes to use the standard request handling procedures,
764 * which sorts requests and coalesces adjacent requests, then it must
765 * call blk_init_queue(). The function @rfn will be called when there
766 * are requests on the queue that need to be processed. If the device
767 * supports plugging, then @rfn may not be called immediately when requests
768 * are available on the queue, but may be called at some time later instead.
769 * Plugged queues are generally unplugged when a buffer belonging to one
770 * of the requests on the queue is needed, or due to memory pressure.
772 * @rfn is not required, or even expected, to remove all requests off the
773 * queue, but only as many as it can handle at a time. If it does leave
774 * requests on the queue, it is responsible for arranging that the requests
775 * get dealt with eventually.
777 * The queue spin lock must be held while manipulating the requests on the
778 * request queue; this lock will be taken also from interrupt context, so irq
779 * disabling is needed for it.
781 * Function returns a pointer to the initialized request queue, or %NULL if
785 * blk_init_queue() must be paired with a blk_cleanup_queue() call
786 * when the block device is deactivated (such as at module unload).
789 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
791 return blk_init_queue_node(rfn
, lock
, NUMA_NO_NODE
);
793 EXPORT_SYMBOL(blk_init_queue
);
795 struct request_queue
*
796 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
798 struct request_queue
*uninit_q
, *q
;
800 uninit_q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
804 q
= blk_init_allocated_queue(uninit_q
, rfn
, lock
);
806 blk_cleanup_queue(uninit_q
);
810 EXPORT_SYMBOL(blk_init_queue_node
);
812 static blk_qc_t
blk_queue_bio(struct request_queue
*q
, struct bio
*bio
);
814 struct request_queue
*
815 blk_init_allocated_queue(struct request_queue
*q
, request_fn_proc
*rfn
,
821 q
->fq
= blk_alloc_flush_queue(q
, NUMA_NO_NODE
, 0);
825 if (blk_init_rl(&q
->root_rl
, q
, GFP_KERNEL
))
829 q
->prep_rq_fn
= NULL
;
830 q
->unprep_rq_fn
= NULL
;
831 q
->queue_flags
|= QUEUE_FLAG_DEFAULT
;
833 /* Override internal queue lock with supplied lock pointer */
835 q
->queue_lock
= lock
;
838 * This also sets hw/phys segments, boundary and size
840 blk_queue_make_request(q
, blk_queue_bio
);
842 q
->sg_reserved_size
= INT_MAX
;
844 /* Protect q->elevator from elevator_change */
845 mutex_lock(&q
->sysfs_lock
);
848 if (elevator_init(q
, NULL
)) {
849 mutex_unlock(&q
->sysfs_lock
);
853 mutex_unlock(&q
->sysfs_lock
);
858 blk_free_flush_queue(q
->fq
);
861 EXPORT_SYMBOL(blk_init_allocated_queue
);
863 bool blk_get_queue(struct request_queue
*q
)
865 if (likely(!blk_queue_dying(q
))) {
872 EXPORT_SYMBOL(blk_get_queue
);
874 static inline void blk_free_request(struct request_list
*rl
, struct request
*rq
)
876 if (rq
->cmd_flags
& REQ_ELVPRIV
) {
877 elv_put_request(rl
->q
, rq
);
879 put_io_context(rq
->elv
.icq
->ioc
);
882 mempool_free(rq
, rl
->rq_pool
);
886 * ioc_batching returns true if the ioc is a valid batching request and
887 * should be given priority access to a request.
889 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
895 * Make sure the process is able to allocate at least 1 request
896 * even if the batch times out, otherwise we could theoretically
899 return ioc
->nr_batch_requests
== q
->nr_batching
||
900 (ioc
->nr_batch_requests
> 0
901 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
905 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
906 * will cause the process to be a "batcher" on all queues in the system. This
907 * is the behaviour we want though - once it gets a wakeup it should be given
910 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
912 if (!ioc
|| ioc_batching(q
, ioc
))
915 ioc
->nr_batch_requests
= q
->nr_batching
;
916 ioc
->last_waited
= jiffies
;
919 static void __freed_request(struct request_list
*rl
, int sync
)
921 struct request_queue
*q
= rl
->q
;
923 if (rl
->count
[sync
] < queue_congestion_off_threshold(q
))
924 blk_clear_congested(rl
, sync
);
926 if (rl
->count
[sync
] + 1 <= q
->nr_requests
) {
927 if (waitqueue_active(&rl
->wait
[sync
]))
928 wake_up(&rl
->wait
[sync
]);
930 blk_clear_rl_full(rl
, sync
);
935 * A request has just been released. Account for it, update the full and
936 * congestion status, wake up any waiters. Called under q->queue_lock.
938 static void freed_request(struct request_list
*rl
, unsigned int flags
)
940 struct request_queue
*q
= rl
->q
;
941 int sync
= rw_is_sync(flags
);
945 if (flags
& REQ_ELVPRIV
)
948 __freed_request(rl
, sync
);
950 if (unlikely(rl
->starved
[sync
^ 1]))
951 __freed_request(rl
, sync
^ 1);
954 int blk_update_nr_requests(struct request_queue
*q
, unsigned int nr
)
956 struct request_list
*rl
;
957 int on_thresh
, off_thresh
;
959 spin_lock_irq(q
->queue_lock
);
961 blk_queue_congestion_threshold(q
);
962 on_thresh
= queue_congestion_on_threshold(q
);
963 off_thresh
= queue_congestion_off_threshold(q
);
965 blk_queue_for_each_rl(rl
, q
) {
966 if (rl
->count
[BLK_RW_SYNC
] >= on_thresh
)
967 blk_set_congested(rl
, BLK_RW_SYNC
);
968 else if (rl
->count
[BLK_RW_SYNC
] < off_thresh
)
969 blk_clear_congested(rl
, BLK_RW_SYNC
);
971 if (rl
->count
[BLK_RW_ASYNC
] >= on_thresh
)
972 blk_set_congested(rl
, BLK_RW_ASYNC
);
973 else if (rl
->count
[BLK_RW_ASYNC
] < off_thresh
)
974 blk_clear_congested(rl
, BLK_RW_ASYNC
);
976 if (rl
->count
[BLK_RW_SYNC
] >= q
->nr_requests
) {
977 blk_set_rl_full(rl
, BLK_RW_SYNC
);
979 blk_clear_rl_full(rl
, BLK_RW_SYNC
);
980 wake_up(&rl
->wait
[BLK_RW_SYNC
]);
983 if (rl
->count
[BLK_RW_ASYNC
] >= q
->nr_requests
) {
984 blk_set_rl_full(rl
, BLK_RW_ASYNC
);
986 blk_clear_rl_full(rl
, BLK_RW_ASYNC
);
987 wake_up(&rl
->wait
[BLK_RW_ASYNC
]);
991 spin_unlock_irq(q
->queue_lock
);
996 * Determine if elevator data should be initialized when allocating the
997 * request associated with @bio.
999 static bool blk_rq_should_init_elevator(struct bio
*bio
)
1005 * Flush requests do not use the elevator so skip initialization.
1006 * This allows a request to share the flush and elevator data.
1008 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
))
1015 * rq_ioc - determine io_context for request allocation
1016 * @bio: request being allocated is for this bio (can be %NULL)
1018 * Determine io_context to use for request allocation for @bio. May return
1019 * %NULL if %current->io_context doesn't exist.
1021 static struct io_context
*rq_ioc(struct bio
*bio
)
1023 #ifdef CONFIG_BLK_CGROUP
1024 if (bio
&& bio
->bi_ioc
)
1027 return current
->io_context
;
1031 * __get_request - get a free request
1032 * @rl: request list to allocate from
1033 * @rw_flags: RW and SYNC flags
1034 * @bio: bio to allocate request for (can be %NULL)
1035 * @gfp_mask: allocation mask
1037 * Get a free request from @q. This function may fail under memory
1038 * pressure or if @q is dead.
1040 * Must be called with @q->queue_lock held and,
1041 * Returns ERR_PTR on failure, with @q->queue_lock held.
1042 * Returns request pointer on success, with @q->queue_lock *not held*.
1044 static struct request
*__get_request(struct request_list
*rl
, int rw_flags
,
1045 struct bio
*bio
, gfp_t gfp_mask
)
1047 struct request_queue
*q
= rl
->q
;
1049 struct elevator_type
*et
= q
->elevator
->type
;
1050 struct io_context
*ioc
= rq_ioc(bio
);
1051 struct io_cq
*icq
= NULL
;
1052 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
1055 if (unlikely(blk_queue_dying(q
)))
1056 return ERR_PTR(-ENODEV
);
1058 may_queue
= elv_may_queue(q
, rw_flags
);
1059 if (may_queue
== ELV_MQUEUE_NO
)
1062 if (rl
->count
[is_sync
]+1 >= queue_congestion_on_threshold(q
)) {
1063 if (rl
->count
[is_sync
]+1 >= q
->nr_requests
) {
1065 * The queue will fill after this allocation, so set
1066 * it as full, and mark this process as "batching".
1067 * This process will be allowed to complete a batch of
1068 * requests, others will be blocked.
1070 if (!blk_rl_full(rl
, is_sync
)) {
1071 ioc_set_batching(q
, ioc
);
1072 blk_set_rl_full(rl
, is_sync
);
1074 if (may_queue
!= ELV_MQUEUE_MUST
1075 && !ioc_batching(q
, ioc
)) {
1077 * The queue is full and the allocating
1078 * process is not a "batcher", and not
1079 * exempted by the IO scheduler
1081 return ERR_PTR(-ENOMEM
);
1085 blk_set_congested(rl
, is_sync
);
1089 * Only allow batching queuers to allocate up to 50% over the defined
1090 * limit of requests, otherwise we could have thousands of requests
1091 * allocated with any setting of ->nr_requests
1093 if (rl
->count
[is_sync
] >= (3 * q
->nr_requests
/ 2))
1094 return ERR_PTR(-ENOMEM
);
1096 q
->nr_rqs
[is_sync
]++;
1097 rl
->count
[is_sync
]++;
1098 rl
->starved
[is_sync
] = 0;
1101 * Decide whether the new request will be managed by elevator. If
1102 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
1103 * prevent the current elevator from being destroyed until the new
1104 * request is freed. This guarantees icq's won't be destroyed and
1105 * makes creating new ones safe.
1107 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1108 * it will be created after releasing queue_lock.
1110 if (blk_rq_should_init_elevator(bio
) && !blk_queue_bypass(q
)) {
1111 rw_flags
|= REQ_ELVPRIV
;
1112 q
->nr_rqs_elvpriv
++;
1113 if (et
->icq_cache
&& ioc
)
1114 icq
= ioc_lookup_icq(ioc
, q
);
1117 if (blk_queue_io_stat(q
))
1118 rw_flags
|= REQ_IO_STAT
;
1119 spin_unlock_irq(q
->queue_lock
);
1121 /* allocate and init request */
1122 rq
= mempool_alloc(rl
->rq_pool
, gfp_mask
);
1127 blk_rq_set_rl(rq
, rl
);
1128 rq
->cmd_flags
= rw_flags
| REQ_ALLOCED
;
1131 if (rw_flags
& REQ_ELVPRIV
) {
1132 if (unlikely(et
->icq_cache
&& !icq
)) {
1134 icq
= ioc_create_icq(ioc
, q
, gfp_mask
);
1140 if (unlikely(elv_set_request(q
, rq
, bio
, gfp_mask
)))
1143 /* @rq->elv.icq holds io_context until @rq is freed */
1145 get_io_context(icq
->ioc
);
1149 * ioc may be NULL here, and ioc_batching will be false. That's
1150 * OK, if the queue is under the request limit then requests need
1151 * not count toward the nr_batch_requests limit. There will always
1152 * be some limit enforced by BLK_BATCH_TIME.
1154 if (ioc_batching(q
, ioc
))
1155 ioc
->nr_batch_requests
--;
1157 trace_block_getrq(q
, bio
, rw_flags
& 1);
1162 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1163 * and may fail indefinitely under memory pressure and thus
1164 * shouldn't stall IO. Treat this request as !elvpriv. This will
1165 * disturb iosched and blkcg but weird is bettern than dead.
1167 printk_ratelimited(KERN_WARNING
"%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1168 __func__
, dev_name(q
->backing_dev_info
.dev
));
1170 rq
->cmd_flags
&= ~REQ_ELVPRIV
;
1173 spin_lock_irq(q
->queue_lock
);
1174 q
->nr_rqs_elvpriv
--;
1175 spin_unlock_irq(q
->queue_lock
);
1180 * Allocation failed presumably due to memory. Undo anything we
1181 * might have messed up.
1183 * Allocating task should really be put onto the front of the wait
1184 * queue, but this is pretty rare.
1186 spin_lock_irq(q
->queue_lock
);
1187 freed_request(rl
, rw_flags
);
1190 * in the very unlikely event that allocation failed and no
1191 * requests for this direction was pending, mark us starved so that
1192 * freeing of a request in the other direction will notice
1193 * us. another possible fix would be to split the rq mempool into
1197 if (unlikely(rl
->count
[is_sync
] == 0))
1198 rl
->starved
[is_sync
] = 1;
1199 return ERR_PTR(-ENOMEM
);
1203 * get_request - get a free request
1204 * @q: request_queue to allocate request from
1205 * @rw_flags: RW and SYNC flags
1206 * @bio: bio to allocate request for (can be %NULL)
1207 * @gfp_mask: allocation mask
1209 * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask,
1210 * this function keeps retrying under memory pressure and fails iff @q is dead.
1212 * Must be called with @q->queue_lock held and,
1213 * Returns ERR_PTR on failure, with @q->queue_lock held.
1214 * Returns request pointer on success, with @q->queue_lock *not held*.
1216 static struct request
*get_request(struct request_queue
*q
, int rw_flags
,
1217 struct bio
*bio
, gfp_t gfp_mask
)
1219 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
1221 struct request_list
*rl
;
1224 rl
= blk_get_rl(q
, bio
); /* transferred to @rq on success */
1226 rq
= __get_request(rl
, rw_flags
, bio
, gfp_mask
);
1230 if (!gfpflags_allow_blocking(gfp_mask
) || unlikely(blk_queue_dying(q
))) {
1235 /* wait on @rl and retry */
1236 prepare_to_wait_exclusive(&rl
->wait
[is_sync
], &wait
,
1237 TASK_UNINTERRUPTIBLE
);
1239 trace_block_sleeprq(q
, bio
, rw_flags
& 1);
1241 spin_unlock_irq(q
->queue_lock
);
1245 * After sleeping, we become a "batching" process and will be able
1246 * to allocate at least one request, and up to a big batch of them
1247 * for a small period time. See ioc_batching, ioc_set_batching
1249 ioc_set_batching(q
, current
->io_context
);
1251 spin_lock_irq(q
->queue_lock
);
1252 finish_wait(&rl
->wait
[is_sync
], &wait
);
1257 static struct request
*blk_old_get_request(struct request_queue
*q
, int rw
,
1262 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
1264 /* create ioc upfront */
1265 create_io_context(gfp_mask
, q
->node
);
1267 spin_lock_irq(q
->queue_lock
);
1268 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
1270 spin_unlock_irq(q
->queue_lock
);
1271 /* q->queue_lock is unlocked at this point */
1276 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
1279 return blk_mq_alloc_request(q
, rw
, gfp_mask
, false);
1281 return blk_old_get_request(q
, rw
, gfp_mask
);
1283 EXPORT_SYMBOL(blk_get_request
);
1286 * blk_make_request - given a bio, allocate a corresponding struct request.
1287 * @q: target request queue
1288 * @bio: The bio describing the memory mappings that will be submitted for IO.
1289 * It may be a chained-bio properly constructed by block/bio layer.
1290 * @gfp_mask: gfp flags to be used for memory allocation
1292 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1293 * type commands. Where the struct request needs to be farther initialized by
1294 * the caller. It is passed a &struct bio, which describes the memory info of
1297 * The caller of blk_make_request must make sure that bi_io_vec
1298 * are set to describe the memory buffers. That bio_data_dir() will return
1299 * the needed direction of the request. (And all bio's in the passed bio-chain
1300 * are properly set accordingly)
1302 * If called under none-sleepable conditions, mapped bio buffers must not
1303 * need bouncing, by calling the appropriate masked or flagged allocator,
1304 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1307 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1308 * given to how you allocate bios. In particular, you cannot use
1309 * __GFP_DIRECT_RECLAIM for anything but the first bio in the chain. Otherwise
1310 * you risk waiting for IO completion of a bio that hasn't been submitted yet,
1311 * thus resulting in a deadlock. Alternatively bios should be allocated using
1312 * bio_kmalloc() instead of bio_alloc(), as that avoids the mempool deadlock.
1313 * If possible a big IO should be split into smaller parts when allocation
1314 * fails. Partial allocation should not be an error, or you risk a live-lock.
1316 struct request
*blk_make_request(struct request_queue
*q
, struct bio
*bio
,
1319 struct request
*rq
= blk_get_request(q
, bio_data_dir(bio
), gfp_mask
);
1324 blk_rq_set_block_pc(rq
);
1327 struct bio
*bounce_bio
= bio
;
1330 blk_queue_bounce(q
, &bounce_bio
);
1331 ret
= blk_rq_append_bio(q
, rq
, bounce_bio
);
1332 if (unlikely(ret
)) {
1333 blk_put_request(rq
);
1334 return ERR_PTR(ret
);
1340 EXPORT_SYMBOL(blk_make_request
);
1343 * blk_rq_set_block_pc - initialize a request to type BLOCK_PC
1344 * @rq: request to be initialized
1347 void blk_rq_set_block_pc(struct request
*rq
)
1349 rq
->cmd_type
= REQ_TYPE_BLOCK_PC
;
1351 rq
->__sector
= (sector_t
) -1;
1352 rq
->bio
= rq
->biotail
= NULL
;
1353 memset(rq
->__cmd
, 0, sizeof(rq
->__cmd
));
1355 EXPORT_SYMBOL(blk_rq_set_block_pc
);
1358 * blk_requeue_request - put a request back on queue
1359 * @q: request queue where request should be inserted
1360 * @rq: request to be inserted
1363 * Drivers often keep queueing requests until the hardware cannot accept
1364 * more, when that condition happens we need to put the request back
1365 * on the queue. Must be called with queue lock held.
1367 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
1369 blk_delete_timer(rq
);
1370 blk_clear_rq_complete(rq
);
1371 trace_block_rq_requeue(q
, rq
);
1373 if (rq
->cmd_flags
& REQ_QUEUED
)
1374 blk_queue_end_tag(q
, rq
);
1376 BUG_ON(blk_queued_rq(rq
));
1378 elv_requeue_request(q
, rq
);
1380 EXPORT_SYMBOL(blk_requeue_request
);
1382 static void add_acct_request(struct request_queue
*q
, struct request
*rq
,
1385 blk_account_io_start(rq
, true);
1386 __elv_add_request(q
, rq
, where
);
1389 static void part_round_stats_single(int cpu
, struct hd_struct
*part
,
1394 if (now
== part
->stamp
)
1397 inflight
= part_in_flight(part
);
1399 __part_stat_add(cpu
, part
, time_in_queue
,
1400 inflight
* (now
- part
->stamp
));
1401 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
1407 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1408 * @cpu: cpu number for stats access
1409 * @part: target partition
1411 * The average IO queue length and utilisation statistics are maintained
1412 * by observing the current state of the queue length and the amount of
1413 * time it has been in this state for.
1415 * Normally, that accounting is done on IO completion, but that can result
1416 * in more than a second's worth of IO being accounted for within any one
1417 * second, leading to >100% utilisation. To deal with that, we call this
1418 * function to do a round-off before returning the results when reading
1419 * /proc/diskstats. This accounts immediately for all queue usage up to
1420 * the current jiffies and restarts the counters again.
1422 void part_round_stats(int cpu
, struct hd_struct
*part
)
1424 unsigned long now
= jiffies
;
1427 part_round_stats_single(cpu
, &part_to_disk(part
)->part0
, now
);
1428 part_round_stats_single(cpu
, part
, now
);
1430 EXPORT_SYMBOL_GPL(part_round_stats
);
1433 static void blk_pm_put_request(struct request
*rq
)
1435 if (rq
->q
->dev
&& !(rq
->cmd_flags
& REQ_PM
) && !--rq
->q
->nr_pending
)
1436 pm_runtime_mark_last_busy(rq
->q
->dev
);
1439 static inline void blk_pm_put_request(struct request
*rq
) {}
1443 * queue lock must be held
1445 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1451 blk_mq_free_request(req
);
1455 blk_pm_put_request(req
);
1457 elv_completed_request(q
, req
);
1459 /* this is a bio leak */
1460 WARN_ON(req
->bio
!= NULL
);
1463 * Request may not have originated from ll_rw_blk. if not,
1464 * it didn't come out of our reserved rq pools
1466 if (req
->cmd_flags
& REQ_ALLOCED
) {
1467 unsigned int flags
= req
->cmd_flags
;
1468 struct request_list
*rl
= blk_rq_rl(req
);
1470 BUG_ON(!list_empty(&req
->queuelist
));
1471 BUG_ON(ELV_ON_HASH(req
));
1473 blk_free_request(rl
, req
);
1474 freed_request(rl
, flags
);
1478 EXPORT_SYMBOL_GPL(__blk_put_request
);
1480 void blk_put_request(struct request
*req
)
1482 struct request_queue
*q
= req
->q
;
1485 blk_mq_free_request(req
);
1487 unsigned long flags
;
1489 spin_lock_irqsave(q
->queue_lock
, flags
);
1490 __blk_put_request(q
, req
);
1491 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1494 EXPORT_SYMBOL(blk_put_request
);
1497 * blk_add_request_payload - add a payload to a request
1498 * @rq: request to update
1499 * @page: page backing the payload
1500 * @len: length of the payload.
1502 * This allows to later add a payload to an already submitted request by
1503 * a block driver. The driver needs to take care of freeing the payload
1506 * Note that this is a quite horrible hack and nothing but handling of
1507 * discard requests should ever use it.
1509 void blk_add_request_payload(struct request
*rq
, struct page
*page
,
1512 struct bio
*bio
= rq
->bio
;
1514 bio
->bi_io_vec
->bv_page
= page
;
1515 bio
->bi_io_vec
->bv_offset
= 0;
1516 bio
->bi_io_vec
->bv_len
= len
;
1518 bio
->bi_iter
.bi_size
= len
;
1520 bio
->bi_phys_segments
= 1;
1522 rq
->__data_len
= rq
->resid_len
= len
;
1523 rq
->nr_phys_segments
= 1;
1525 EXPORT_SYMBOL_GPL(blk_add_request_payload
);
1527 bool bio_attempt_back_merge(struct request_queue
*q
, struct request
*req
,
1530 const int ff
= bio
->bi_rw
& REQ_FAILFAST_MASK
;
1532 if (!ll_back_merge_fn(q
, req
, bio
))
1535 trace_block_bio_backmerge(q
, req
, bio
);
1537 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1538 blk_rq_set_mixed_merge(req
);
1540 req
->biotail
->bi_next
= bio
;
1542 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1543 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1545 blk_account_io_start(req
, false);
1549 bool bio_attempt_front_merge(struct request_queue
*q
, struct request
*req
,
1552 const int ff
= bio
->bi_rw
& REQ_FAILFAST_MASK
;
1554 if (!ll_front_merge_fn(q
, req
, bio
))
1557 trace_block_bio_frontmerge(q
, req
, bio
);
1559 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1560 blk_rq_set_mixed_merge(req
);
1562 bio
->bi_next
= req
->bio
;
1565 req
->__sector
= bio
->bi_iter
.bi_sector
;
1566 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1567 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1569 blk_account_io_start(req
, false);
1574 * blk_attempt_plug_merge - try to merge with %current's plugged list
1575 * @q: request_queue new bio is being queued at
1576 * @bio: new bio being queued
1577 * @request_count: out parameter for number of traversed plugged requests
1579 * Determine whether @bio being queued on @q can be merged with a request
1580 * on %current's plugged list. Returns %true if merge was successful,
1583 * Plugging coalesces IOs from the same issuer for the same purpose without
1584 * going through @q->queue_lock. As such it's more of an issuing mechanism
1585 * than scheduling, and the request, while may have elvpriv data, is not
1586 * added on the elevator at this point. In addition, we don't have
1587 * reliable access to the elevator outside queue lock. Only check basic
1588 * merging parameters without querying the elevator.
1590 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1592 bool blk_attempt_plug_merge(struct request_queue
*q
, struct bio
*bio
,
1593 unsigned int *request_count
,
1594 struct request
**same_queue_rq
)
1596 struct blk_plug
*plug
;
1599 struct list_head
*plug_list
;
1601 plug
= current
->plug
;
1607 plug_list
= &plug
->mq_list
;
1609 plug_list
= &plug
->list
;
1611 list_for_each_entry_reverse(rq
, plug_list
, queuelist
) {
1617 * Only blk-mq multiple hardware queues case checks the
1618 * rq in the same queue, there should be only one such
1622 *same_queue_rq
= rq
;
1625 if (rq
->q
!= q
|| !blk_rq_merge_ok(rq
, bio
))
1628 el_ret
= blk_try_merge(rq
, bio
);
1629 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1630 ret
= bio_attempt_back_merge(q
, rq
, bio
);
1633 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1634 ret
= bio_attempt_front_merge(q
, rq
, bio
);
1643 unsigned int blk_plug_queued_count(struct request_queue
*q
)
1645 struct blk_plug
*plug
;
1647 struct list_head
*plug_list
;
1648 unsigned int ret
= 0;
1650 plug
= current
->plug
;
1655 plug_list
= &plug
->mq_list
;
1657 plug_list
= &plug
->list
;
1659 list_for_each_entry(rq
, plug_list
, queuelist
) {
1667 void init_request_from_bio(struct request
*req
, struct bio
*bio
)
1669 req
->cmd_type
= REQ_TYPE_FS
;
1671 req
->cmd_flags
|= bio
->bi_rw
& REQ_COMMON_MASK
;
1672 if (bio
->bi_rw
& REQ_RAHEAD
)
1673 req
->cmd_flags
|= REQ_FAILFAST_MASK
;
1676 req
->__sector
= bio
->bi_iter
.bi_sector
;
1677 req
->ioprio
= bio_prio(bio
);
1678 blk_rq_bio_prep(req
->q
, req
, bio
);
1681 static blk_qc_t
blk_queue_bio(struct request_queue
*q
, struct bio
*bio
)
1683 const bool sync
= !!(bio
->bi_rw
& REQ_SYNC
);
1684 struct blk_plug
*plug
;
1685 int el_ret
, rw_flags
, where
= ELEVATOR_INSERT_SORT
;
1686 struct request
*req
;
1687 unsigned int request_count
= 0;
1689 blk_queue_split(q
, &bio
, q
->bio_split
);
1692 * low level driver can indicate that it wants pages above a
1693 * certain limit bounced to low memory (ie for highmem, or even
1694 * ISA dma in theory)
1696 blk_queue_bounce(q
, &bio
);
1698 if (bio_integrity_enabled(bio
) && bio_integrity_prep(bio
)) {
1699 bio
->bi_error
= -EIO
;
1701 return BLK_QC_T_NONE
;
1704 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) {
1705 spin_lock_irq(q
->queue_lock
);
1706 where
= ELEVATOR_INSERT_FLUSH
;
1711 * Check if we can merge with the plugged list before grabbing
1714 if (!blk_queue_nomerges(q
)) {
1715 if (blk_attempt_plug_merge(q
, bio
, &request_count
, NULL
))
1716 return BLK_QC_T_NONE
;
1718 request_count
= blk_plug_queued_count(q
);
1720 spin_lock_irq(q
->queue_lock
);
1722 el_ret
= elv_merge(q
, &req
, bio
);
1723 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1724 if (bio_attempt_back_merge(q
, req
, bio
)) {
1725 elv_bio_merged(q
, req
, bio
);
1726 if (!attempt_back_merge(q
, req
))
1727 elv_merged_request(q
, req
, el_ret
);
1730 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1731 if (bio_attempt_front_merge(q
, req
, bio
)) {
1732 elv_bio_merged(q
, req
, bio
);
1733 if (!attempt_front_merge(q
, req
))
1734 elv_merged_request(q
, req
, el_ret
);
1741 * This sync check and mask will be re-done in init_request_from_bio(),
1742 * but we need to set it earlier to expose the sync flag to the
1743 * rq allocator and io schedulers.
1745 rw_flags
= bio_data_dir(bio
);
1747 rw_flags
|= REQ_SYNC
;
1750 * Grab a free request. This is might sleep but can not fail.
1751 * Returns with the queue unlocked.
1753 req
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
1755 bio
->bi_error
= PTR_ERR(req
);
1761 * After dropping the lock and possibly sleeping here, our request
1762 * may now be mergeable after it had proven unmergeable (above).
1763 * We don't worry about that case for efficiency. It won't happen
1764 * often, and the elevators are able to handle it.
1766 init_request_from_bio(req
, bio
);
1768 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
))
1769 req
->cpu
= raw_smp_processor_id();
1771 plug
= current
->plug
;
1774 * If this is the first request added after a plug, fire
1778 trace_block_plug(q
);
1780 if (request_count
>= BLK_MAX_REQUEST_COUNT
) {
1781 blk_flush_plug_list(plug
, false);
1782 trace_block_plug(q
);
1785 list_add_tail(&req
->queuelist
, &plug
->list
);
1786 blk_account_io_start(req
, true);
1788 spin_lock_irq(q
->queue_lock
);
1789 add_acct_request(q
, req
, where
);
1792 spin_unlock_irq(q
->queue_lock
);
1795 return BLK_QC_T_NONE
;
1799 * If bio->bi_dev is a partition, remap the location
1801 static inline void blk_partition_remap(struct bio
*bio
)
1803 struct block_device
*bdev
= bio
->bi_bdev
;
1805 if (bio_sectors(bio
) && bdev
!= bdev
->bd_contains
) {
1806 struct hd_struct
*p
= bdev
->bd_part
;
1808 bio
->bi_iter
.bi_sector
+= p
->start_sect
;
1809 bio
->bi_bdev
= bdev
->bd_contains
;
1811 trace_block_bio_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
1813 bio
->bi_iter
.bi_sector
- p
->start_sect
);
1817 static void handle_bad_sector(struct bio
*bio
)
1819 char b
[BDEVNAME_SIZE
];
1821 printk(KERN_INFO
"attempt to access beyond end of device\n");
1822 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
1823 bdevname(bio
->bi_bdev
, b
),
1825 (unsigned long long)bio_end_sector(bio
),
1826 (long long)(i_size_read(bio
->bi_bdev
->bd_inode
) >> 9));
1829 #ifdef CONFIG_FAIL_MAKE_REQUEST
1831 static DECLARE_FAULT_ATTR(fail_make_request
);
1833 static int __init
setup_fail_make_request(char *str
)
1835 return setup_fault_attr(&fail_make_request
, str
);
1837 __setup("fail_make_request=", setup_fail_make_request
);
1839 static bool should_fail_request(struct hd_struct
*part
, unsigned int bytes
)
1841 return part
->make_it_fail
&& should_fail(&fail_make_request
, bytes
);
1844 static int __init
fail_make_request_debugfs(void)
1846 struct dentry
*dir
= fault_create_debugfs_attr("fail_make_request",
1847 NULL
, &fail_make_request
);
1849 return PTR_ERR_OR_ZERO(dir
);
1852 late_initcall(fail_make_request_debugfs
);
1854 #else /* CONFIG_FAIL_MAKE_REQUEST */
1856 static inline bool should_fail_request(struct hd_struct
*part
,
1862 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1865 * Check whether this bio extends beyond the end of the device.
1867 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
1874 /* Test device or partition size, when known. */
1875 maxsector
= i_size_read(bio
->bi_bdev
->bd_inode
) >> 9;
1877 sector_t sector
= bio
->bi_iter
.bi_sector
;
1879 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
1881 * This may well happen - the kernel calls bread()
1882 * without checking the size of the device, e.g., when
1883 * mounting a device.
1885 handle_bad_sector(bio
);
1893 static noinline_for_stack
bool
1894 generic_make_request_checks(struct bio
*bio
)
1896 struct request_queue
*q
;
1897 int nr_sectors
= bio_sectors(bio
);
1899 char b
[BDEVNAME_SIZE
];
1900 struct hd_struct
*part
;
1904 if (bio_check_eod(bio
, nr_sectors
))
1907 q
= bdev_get_queue(bio
->bi_bdev
);
1910 "generic_make_request: Trying to access "
1911 "nonexistent block-device %s (%Lu)\n",
1912 bdevname(bio
->bi_bdev
, b
),
1913 (long long) bio
->bi_iter
.bi_sector
);
1917 part
= bio
->bi_bdev
->bd_part
;
1918 if (should_fail_request(part
, bio
->bi_iter
.bi_size
) ||
1919 should_fail_request(&part_to_disk(part
)->part0
,
1920 bio
->bi_iter
.bi_size
))
1924 * If this device has partitions, remap block n
1925 * of partition p to block n+start(p) of the disk.
1927 blk_partition_remap(bio
);
1929 if (bio_check_eod(bio
, nr_sectors
))
1933 * Filter flush bio's early so that make_request based
1934 * drivers without flush support don't have to worry
1937 if ((bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) && !q
->flush_flags
) {
1938 bio
->bi_rw
&= ~(REQ_FLUSH
| REQ_FUA
);
1945 if ((bio
->bi_rw
& REQ_DISCARD
) &&
1946 (!blk_queue_discard(q
) ||
1947 ((bio
->bi_rw
& REQ_SECURE
) && !blk_queue_secdiscard(q
)))) {
1952 if (bio
->bi_rw
& REQ_WRITE_SAME
&& !bdev_write_same(bio
->bi_bdev
)) {
1958 * Various block parts want %current->io_context and lazy ioc
1959 * allocation ends up trading a lot of pain for a small amount of
1960 * memory. Just allocate it upfront. This may fail and block
1961 * layer knows how to live with it.
1963 create_io_context(GFP_ATOMIC
, q
->node
);
1965 if (!blkcg_bio_issue_check(q
, bio
))
1968 trace_block_bio_queue(q
, bio
);
1972 bio
->bi_error
= err
;
1978 * generic_make_request - hand a buffer to its device driver for I/O
1979 * @bio: The bio describing the location in memory and on the device.
1981 * generic_make_request() is used to make I/O requests of block
1982 * devices. It is passed a &struct bio, which describes the I/O that needs
1985 * generic_make_request() does not return any status. The
1986 * success/failure status of the request, along with notification of
1987 * completion, is delivered asynchronously through the bio->bi_end_io
1988 * function described (one day) else where.
1990 * The caller of generic_make_request must make sure that bi_io_vec
1991 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1992 * set to describe the device address, and the
1993 * bi_end_io and optionally bi_private are set to describe how
1994 * completion notification should be signaled.
1996 * generic_make_request and the drivers it calls may use bi_next if this
1997 * bio happens to be merged with someone else, and may resubmit the bio to
1998 * a lower device by calling into generic_make_request recursively, which
1999 * means the bio should NOT be touched after the call to ->make_request_fn.
2001 blk_qc_t
generic_make_request(struct bio
*bio
)
2003 struct bio_list bio_list_on_stack
;
2004 blk_qc_t ret
= BLK_QC_T_NONE
;
2006 if (!generic_make_request_checks(bio
))
2010 * We only want one ->make_request_fn to be active at a time, else
2011 * stack usage with stacked devices could be a problem. So use
2012 * current->bio_list to keep a list of requests submited by a
2013 * make_request_fn function. current->bio_list is also used as a
2014 * flag to say if generic_make_request is currently active in this
2015 * task or not. If it is NULL, then no make_request is active. If
2016 * it is non-NULL, then a make_request is active, and new requests
2017 * should be added at the tail
2019 if (current
->bio_list
) {
2020 bio_list_add(current
->bio_list
, bio
);
2024 /* following loop may be a bit non-obvious, and so deserves some
2026 * Before entering the loop, bio->bi_next is NULL (as all callers
2027 * ensure that) so we have a list with a single bio.
2028 * We pretend that we have just taken it off a longer list, so
2029 * we assign bio_list to a pointer to the bio_list_on_stack,
2030 * thus initialising the bio_list of new bios to be
2031 * added. ->make_request() may indeed add some more bios
2032 * through a recursive call to generic_make_request. If it
2033 * did, we find a non-NULL value in bio_list and re-enter the loop
2034 * from the top. In this case we really did just take the bio
2035 * of the top of the list (no pretending) and so remove it from
2036 * bio_list, and call into ->make_request() again.
2038 BUG_ON(bio
->bi_next
);
2039 bio_list_init(&bio_list_on_stack
);
2040 current
->bio_list
= &bio_list_on_stack
;
2042 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
2044 if (likely(blk_queue_enter(q
, __GFP_DIRECT_RECLAIM
) == 0)) {
2046 ret
= q
->make_request_fn(q
, bio
);
2050 bio
= bio_list_pop(current
->bio_list
);
2052 struct bio
*bio_next
= bio_list_pop(current
->bio_list
);
2058 current
->bio_list
= NULL
; /* deactivate */
2063 EXPORT_SYMBOL(generic_make_request
);
2066 * submit_bio - submit a bio to the block device layer for I/O
2067 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
2068 * @bio: The &struct bio which describes the I/O
2070 * submit_bio() is very similar in purpose to generic_make_request(), and
2071 * uses that function to do most of the work. Both are fairly rough
2072 * interfaces; @bio must be presetup and ready for I/O.
2075 blk_qc_t
submit_bio(int rw
, struct bio
*bio
)
2080 * If it's a regular read/write or a barrier with data attached,
2081 * go through the normal accounting stuff before submission.
2083 if (bio_has_data(bio
)) {
2086 if (unlikely(rw
& REQ_WRITE_SAME
))
2087 count
= bdev_logical_block_size(bio
->bi_bdev
) >> 9;
2089 count
= bio_sectors(bio
);
2092 count_vm_events(PGPGOUT
, count
);
2094 task_io_account_read(bio
->bi_iter
.bi_size
);
2095 count_vm_events(PGPGIN
, count
);
2098 if (unlikely(block_dump
)) {
2099 char b
[BDEVNAME_SIZE
];
2100 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s (%u sectors)\n",
2101 current
->comm
, task_pid_nr(current
),
2102 (rw
& WRITE
) ? "WRITE" : "READ",
2103 (unsigned long long)bio
->bi_iter
.bi_sector
,
2104 bdevname(bio
->bi_bdev
, b
),
2109 return generic_make_request(bio
);
2111 EXPORT_SYMBOL(submit_bio
);
2114 * blk_rq_check_limits - Helper function to check a request for the queue limit
2116 * @rq: the request being checked
2119 * @rq may have been made based on weaker limitations of upper-level queues
2120 * in request stacking drivers, and it may violate the limitation of @q.
2121 * Since the block layer and the underlying device driver trust @rq
2122 * after it is inserted to @q, it should be checked against @q before
2123 * the insertion using this generic function.
2125 * This function should also be useful for request stacking drivers
2126 * in some cases below, so export this function.
2127 * Request stacking drivers like request-based dm may change the queue
2128 * limits while requests are in the queue (e.g. dm's table swapping).
2129 * Such request stacking drivers should check those requests against
2130 * the new queue limits again when they dispatch those requests,
2131 * although such checkings are also done against the old queue limits
2132 * when submitting requests.
2134 int blk_rq_check_limits(struct request_queue
*q
, struct request
*rq
)
2136 if (!rq_mergeable(rq
))
2139 if (blk_rq_sectors(rq
) > blk_queue_get_max_sectors(q
, rq
->cmd_flags
)) {
2140 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
2145 * queue's settings related to segment counting like q->bounce_pfn
2146 * may differ from that of other stacking queues.
2147 * Recalculate it to check the request correctly on this queue's
2150 blk_recalc_rq_segments(rq
);
2151 if (rq
->nr_phys_segments
> queue_max_segments(q
)) {
2152 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
2158 EXPORT_SYMBOL_GPL(blk_rq_check_limits
);
2161 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2162 * @q: the queue to submit the request
2163 * @rq: the request being queued
2165 int blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
2167 unsigned long flags
;
2168 int where
= ELEVATOR_INSERT_BACK
;
2170 if (blk_rq_check_limits(q
, rq
))
2174 should_fail_request(&rq
->rq_disk
->part0
, blk_rq_bytes(rq
)))
2178 if (blk_queue_io_stat(q
))
2179 blk_account_io_start(rq
, true);
2180 blk_mq_insert_request(rq
, false, true, true);
2184 spin_lock_irqsave(q
->queue_lock
, flags
);
2185 if (unlikely(blk_queue_dying(q
))) {
2186 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2191 * Submitting request must be dequeued before calling this function
2192 * because it will be linked to another request_queue
2194 BUG_ON(blk_queued_rq(rq
));
2196 if (rq
->cmd_flags
& (REQ_FLUSH
|REQ_FUA
))
2197 where
= ELEVATOR_INSERT_FLUSH
;
2199 add_acct_request(q
, rq
, where
);
2200 if (where
== ELEVATOR_INSERT_FLUSH
)
2202 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2206 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
2209 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2210 * @rq: request to examine
2213 * A request could be merge of IOs which require different failure
2214 * handling. This function determines the number of bytes which
2215 * can be failed from the beginning of the request without
2216 * crossing into area which need to be retried further.
2219 * The number of bytes to fail.
2222 * queue_lock must be held.
2224 unsigned int blk_rq_err_bytes(const struct request
*rq
)
2226 unsigned int ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
2227 unsigned int bytes
= 0;
2230 if (!(rq
->cmd_flags
& REQ_MIXED_MERGE
))
2231 return blk_rq_bytes(rq
);
2234 * Currently the only 'mixing' which can happen is between
2235 * different fastfail types. We can safely fail portions
2236 * which have all the failfast bits that the first one has -
2237 * the ones which are at least as eager to fail as the first
2240 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
2241 if ((bio
->bi_rw
& ff
) != ff
)
2243 bytes
+= bio
->bi_iter
.bi_size
;
2246 /* this could lead to infinite loop */
2247 BUG_ON(blk_rq_bytes(rq
) && !bytes
);
2250 EXPORT_SYMBOL_GPL(blk_rq_err_bytes
);
2252 void blk_account_io_completion(struct request
*req
, unsigned int bytes
)
2254 if (blk_do_io_stat(req
)) {
2255 const int rw
= rq_data_dir(req
);
2256 struct hd_struct
*part
;
2259 cpu
= part_stat_lock();
2261 part_stat_add(cpu
, part
, sectors
[rw
], bytes
>> 9);
2266 void blk_account_io_done(struct request
*req
)
2269 * Account IO completion. flush_rq isn't accounted as a
2270 * normal IO on queueing nor completion. Accounting the
2271 * containing request is enough.
2273 if (blk_do_io_stat(req
) && !(req
->cmd_flags
& REQ_FLUSH_SEQ
)) {
2274 unsigned long duration
= jiffies
- req
->start_time
;
2275 const int rw
= rq_data_dir(req
);
2276 struct hd_struct
*part
;
2279 cpu
= part_stat_lock();
2282 part_stat_inc(cpu
, part
, ios
[rw
]);
2283 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
2284 part_round_stats(cpu
, part
);
2285 part_dec_in_flight(part
, rw
);
2287 hd_struct_put(part
);
2294 * Don't process normal requests when queue is suspended
2295 * or in the process of suspending/resuming
2297 static struct request
*blk_pm_peek_request(struct request_queue
*q
,
2300 if (q
->dev
&& (q
->rpm_status
== RPM_SUSPENDED
||
2301 (q
->rpm_status
!= RPM_ACTIVE
&& !(rq
->cmd_flags
& REQ_PM
))))
2307 static inline struct request
*blk_pm_peek_request(struct request_queue
*q
,
2314 void blk_account_io_start(struct request
*rq
, bool new_io
)
2316 struct hd_struct
*part
;
2317 int rw
= rq_data_dir(rq
);
2320 if (!blk_do_io_stat(rq
))
2323 cpu
= part_stat_lock();
2327 part_stat_inc(cpu
, part
, merges
[rw
]);
2329 part
= disk_map_sector_rcu(rq
->rq_disk
, blk_rq_pos(rq
));
2330 if (!hd_struct_try_get(part
)) {
2332 * The partition is already being removed,
2333 * the request will be accounted on the disk only
2335 * We take a reference on disk->part0 although that
2336 * partition will never be deleted, so we can treat
2337 * it as any other partition.
2339 part
= &rq
->rq_disk
->part0
;
2340 hd_struct_get(part
);
2342 part_round_stats(cpu
, part
);
2343 part_inc_in_flight(part
, rw
);
2351 * blk_peek_request - peek at the top of a request queue
2352 * @q: request queue to peek at
2355 * Return the request at the top of @q. The returned request
2356 * should be started using blk_start_request() before LLD starts
2360 * Pointer to the request at the top of @q if available. Null
2364 * queue_lock must be held.
2366 struct request
*blk_peek_request(struct request_queue
*q
)
2371 while ((rq
= __elv_next_request(q
)) != NULL
) {
2373 rq
= blk_pm_peek_request(q
, rq
);
2377 if (!(rq
->cmd_flags
& REQ_STARTED
)) {
2379 * This is the first time the device driver
2380 * sees this request (possibly after
2381 * requeueing). Notify IO scheduler.
2383 if (rq
->cmd_flags
& REQ_SORTED
)
2384 elv_activate_rq(q
, rq
);
2387 * just mark as started even if we don't start
2388 * it, a request that has been delayed should
2389 * not be passed by new incoming requests
2391 rq
->cmd_flags
|= REQ_STARTED
;
2392 trace_block_rq_issue(q
, rq
);
2395 if (!q
->boundary_rq
|| q
->boundary_rq
== rq
) {
2396 q
->end_sector
= rq_end_sector(rq
);
2397 q
->boundary_rq
= NULL
;
2400 if (rq
->cmd_flags
& REQ_DONTPREP
)
2403 if (q
->dma_drain_size
&& blk_rq_bytes(rq
)) {
2405 * make sure space for the drain appears we
2406 * know we can do this because max_hw_segments
2407 * has been adjusted to be one fewer than the
2410 rq
->nr_phys_segments
++;
2416 ret
= q
->prep_rq_fn(q
, rq
);
2417 if (ret
== BLKPREP_OK
) {
2419 } else if (ret
== BLKPREP_DEFER
) {
2421 * the request may have been (partially) prepped.
2422 * we need to keep this request in the front to
2423 * avoid resource deadlock. REQ_STARTED will
2424 * prevent other fs requests from passing this one.
2426 if (q
->dma_drain_size
&& blk_rq_bytes(rq
) &&
2427 !(rq
->cmd_flags
& REQ_DONTPREP
)) {
2429 * remove the space for the drain we added
2430 * so that we don't add it again
2432 --rq
->nr_phys_segments
;
2437 } else if (ret
== BLKPREP_KILL
) {
2438 rq
->cmd_flags
|= REQ_QUIET
;
2440 * Mark this request as started so we don't trigger
2441 * any debug logic in the end I/O path.
2443 blk_start_request(rq
);
2444 __blk_end_request_all(rq
, -EIO
);
2446 printk(KERN_ERR
"%s: bad return=%d\n", __func__
, ret
);
2453 EXPORT_SYMBOL(blk_peek_request
);
2455 void blk_dequeue_request(struct request
*rq
)
2457 struct request_queue
*q
= rq
->q
;
2459 BUG_ON(list_empty(&rq
->queuelist
));
2460 BUG_ON(ELV_ON_HASH(rq
));
2462 list_del_init(&rq
->queuelist
);
2465 * the time frame between a request being removed from the lists
2466 * and to it is freed is accounted as io that is in progress at
2469 if (blk_account_rq(rq
)) {
2470 q
->in_flight
[rq_is_sync(rq
)]++;
2471 set_io_start_time_ns(rq
);
2476 * blk_start_request - start request processing on the driver
2477 * @req: request to dequeue
2480 * Dequeue @req and start timeout timer on it. This hands off the
2481 * request to the driver.
2483 * Block internal functions which don't want to start timer should
2484 * call blk_dequeue_request().
2487 * queue_lock must be held.
2489 void blk_start_request(struct request
*req
)
2491 blk_dequeue_request(req
);
2494 * We are now handing the request to the hardware, initialize
2495 * resid_len to full count and add the timeout handler.
2497 req
->resid_len
= blk_rq_bytes(req
);
2498 if (unlikely(blk_bidi_rq(req
)))
2499 req
->next_rq
->resid_len
= blk_rq_bytes(req
->next_rq
);
2501 BUG_ON(test_bit(REQ_ATOM_COMPLETE
, &req
->atomic_flags
));
2504 EXPORT_SYMBOL(blk_start_request
);
2507 * blk_fetch_request - fetch a request from a request queue
2508 * @q: request queue to fetch a request from
2511 * Return the request at the top of @q. The request is started on
2512 * return and LLD can start processing it immediately.
2515 * Pointer to the request at the top of @q if available. Null
2519 * queue_lock must be held.
2521 struct request
*blk_fetch_request(struct request_queue
*q
)
2525 rq
= blk_peek_request(q
);
2527 blk_start_request(rq
);
2530 EXPORT_SYMBOL(blk_fetch_request
);
2533 * blk_update_request - Special helper function for request stacking drivers
2534 * @req: the request being processed
2535 * @error: %0 for success, < %0 for error
2536 * @nr_bytes: number of bytes to complete @req
2539 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2540 * the request structure even if @req doesn't have leftover.
2541 * If @req has leftover, sets it up for the next range of segments.
2543 * This special helper function is only for request stacking drivers
2544 * (e.g. request-based dm) so that they can handle partial completion.
2545 * Actual device drivers should use blk_end_request instead.
2547 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2548 * %false return from this function.
2551 * %false - this request doesn't have any more data
2552 * %true - this request has more data
2554 bool blk_update_request(struct request
*req
, int error
, unsigned int nr_bytes
)
2558 trace_block_rq_complete(req
->q
, req
, nr_bytes
);
2564 * For fs requests, rq is just carrier of independent bio's
2565 * and each partial completion should be handled separately.
2566 * Reset per-request error on each partial completion.
2568 * TODO: tj: This is too subtle. It would be better to let
2569 * low level drivers do what they see fit.
2571 if (req
->cmd_type
== REQ_TYPE_FS
)
2574 if (error
&& req
->cmd_type
== REQ_TYPE_FS
&&
2575 !(req
->cmd_flags
& REQ_QUIET
)) {
2580 error_type
= "recoverable transport";
2583 error_type
= "critical target";
2586 error_type
= "critical nexus";
2589 error_type
= "timeout";
2592 error_type
= "critical space allocation";
2595 error_type
= "critical medium";
2602 printk_ratelimited(KERN_ERR
"%s: %s error, dev %s, sector %llu\n",
2603 __func__
, error_type
, req
->rq_disk
?
2604 req
->rq_disk
->disk_name
: "?",
2605 (unsigned long long)blk_rq_pos(req
));
2609 blk_account_io_completion(req
, nr_bytes
);
2613 struct bio
*bio
= req
->bio
;
2614 unsigned bio_bytes
= min(bio
->bi_iter
.bi_size
, nr_bytes
);
2616 if (bio_bytes
== bio
->bi_iter
.bi_size
)
2617 req
->bio
= bio
->bi_next
;
2619 req_bio_endio(req
, bio
, bio_bytes
, error
);
2621 total_bytes
+= bio_bytes
;
2622 nr_bytes
-= bio_bytes
;
2633 * Reset counters so that the request stacking driver
2634 * can find how many bytes remain in the request
2637 req
->__data_len
= 0;
2641 req
->__data_len
-= total_bytes
;
2643 /* update sector only for requests with clear definition of sector */
2644 if (req
->cmd_type
== REQ_TYPE_FS
)
2645 req
->__sector
+= total_bytes
>> 9;
2647 /* mixed attributes always follow the first bio */
2648 if (req
->cmd_flags
& REQ_MIXED_MERGE
) {
2649 req
->cmd_flags
&= ~REQ_FAILFAST_MASK
;
2650 req
->cmd_flags
|= req
->bio
->bi_rw
& REQ_FAILFAST_MASK
;
2654 * If total number of sectors is less than the first segment
2655 * size, something has gone terribly wrong.
2657 if (blk_rq_bytes(req
) < blk_rq_cur_bytes(req
)) {
2658 blk_dump_rq_flags(req
, "request botched");
2659 req
->__data_len
= blk_rq_cur_bytes(req
);
2662 /* recalculate the number of segments */
2663 blk_recalc_rq_segments(req
);
2667 EXPORT_SYMBOL_GPL(blk_update_request
);
2669 static bool blk_update_bidi_request(struct request
*rq
, int error
,
2670 unsigned int nr_bytes
,
2671 unsigned int bidi_bytes
)
2673 if (blk_update_request(rq
, error
, nr_bytes
))
2676 /* Bidi request must be completed as a whole */
2677 if (unlikely(blk_bidi_rq(rq
)) &&
2678 blk_update_request(rq
->next_rq
, error
, bidi_bytes
))
2681 if (blk_queue_add_random(rq
->q
))
2682 add_disk_randomness(rq
->rq_disk
);
2688 * blk_unprep_request - unprepare a request
2691 * This function makes a request ready for complete resubmission (or
2692 * completion). It happens only after all error handling is complete,
2693 * so represents the appropriate moment to deallocate any resources
2694 * that were allocated to the request in the prep_rq_fn. The queue
2695 * lock is held when calling this.
2697 void blk_unprep_request(struct request
*req
)
2699 struct request_queue
*q
= req
->q
;
2701 req
->cmd_flags
&= ~REQ_DONTPREP
;
2702 if (q
->unprep_rq_fn
)
2703 q
->unprep_rq_fn(q
, req
);
2705 EXPORT_SYMBOL_GPL(blk_unprep_request
);
2708 * queue lock must be held
2710 void blk_finish_request(struct request
*req
, int error
)
2712 if (req
->cmd_flags
& REQ_QUEUED
)
2713 blk_queue_end_tag(req
->q
, req
);
2715 BUG_ON(blk_queued_rq(req
));
2717 if (unlikely(laptop_mode
) && req
->cmd_type
== REQ_TYPE_FS
)
2718 laptop_io_completion(&req
->q
->backing_dev_info
);
2720 blk_delete_timer(req
);
2722 if (req
->cmd_flags
& REQ_DONTPREP
)
2723 blk_unprep_request(req
);
2725 blk_account_io_done(req
);
2728 req
->end_io(req
, error
);
2730 if (blk_bidi_rq(req
))
2731 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
2733 __blk_put_request(req
->q
, req
);
2736 EXPORT_SYMBOL(blk_finish_request
);
2739 * blk_end_bidi_request - Complete a bidi request
2740 * @rq: the request to complete
2741 * @error: %0 for success, < %0 for error
2742 * @nr_bytes: number of bytes to complete @rq
2743 * @bidi_bytes: number of bytes to complete @rq->next_rq
2746 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2747 * Drivers that supports bidi can safely call this member for any
2748 * type of request, bidi or uni. In the later case @bidi_bytes is
2752 * %false - we are done with this request
2753 * %true - still buffers pending for this request
2755 static bool blk_end_bidi_request(struct request
*rq
, int error
,
2756 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2758 struct request_queue
*q
= rq
->q
;
2759 unsigned long flags
;
2761 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2764 spin_lock_irqsave(q
->queue_lock
, flags
);
2765 blk_finish_request(rq
, error
);
2766 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2772 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2773 * @rq: the request to complete
2774 * @error: %0 for success, < %0 for error
2775 * @nr_bytes: number of bytes to complete @rq
2776 * @bidi_bytes: number of bytes to complete @rq->next_rq
2779 * Identical to blk_end_bidi_request() except that queue lock is
2780 * assumed to be locked on entry and remains so on return.
2783 * %false - we are done with this request
2784 * %true - still buffers pending for this request
2786 bool __blk_end_bidi_request(struct request
*rq
, int error
,
2787 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2789 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2792 blk_finish_request(rq
, error
);
2798 * blk_end_request - Helper function for drivers to complete the request.
2799 * @rq: the request being processed
2800 * @error: %0 for success, < %0 for error
2801 * @nr_bytes: number of bytes to complete
2804 * Ends I/O on a number of bytes attached to @rq.
2805 * If @rq has leftover, sets it up for the next range of segments.
2808 * %false - we are done with this request
2809 * %true - still buffers pending for this request
2811 bool blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2813 return blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2815 EXPORT_SYMBOL(blk_end_request
);
2818 * blk_end_request_all - Helper function for drives to finish the request.
2819 * @rq: the request to finish
2820 * @error: %0 for success, < %0 for error
2823 * Completely finish @rq.
2825 void blk_end_request_all(struct request
*rq
, int error
)
2828 unsigned int bidi_bytes
= 0;
2830 if (unlikely(blk_bidi_rq(rq
)))
2831 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2833 pending
= blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2836 EXPORT_SYMBOL(blk_end_request_all
);
2839 * blk_end_request_cur - Helper function to finish the current request chunk.
2840 * @rq: the request to finish the current chunk for
2841 * @error: %0 for success, < %0 for error
2844 * Complete the current consecutively mapped chunk from @rq.
2847 * %false - we are done with this request
2848 * %true - still buffers pending for this request
2850 bool blk_end_request_cur(struct request
*rq
, int error
)
2852 return blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2854 EXPORT_SYMBOL(blk_end_request_cur
);
2857 * blk_end_request_err - Finish a request till the next failure boundary.
2858 * @rq: the request to finish till the next failure boundary for
2859 * @error: must be negative errno
2862 * Complete @rq till the next failure boundary.
2865 * %false - we are done with this request
2866 * %true - still buffers pending for this request
2868 bool blk_end_request_err(struct request
*rq
, int error
)
2870 WARN_ON(error
>= 0);
2871 return blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2873 EXPORT_SYMBOL_GPL(blk_end_request_err
);
2876 * __blk_end_request - Helper function for drivers to complete the request.
2877 * @rq: the request being processed
2878 * @error: %0 for success, < %0 for error
2879 * @nr_bytes: number of bytes to complete
2882 * Must be called with queue lock held unlike blk_end_request().
2885 * %false - we are done with this request
2886 * %true - still buffers pending for this request
2888 bool __blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2890 return __blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2892 EXPORT_SYMBOL(__blk_end_request
);
2895 * __blk_end_request_all - Helper function for drives to finish the request.
2896 * @rq: the request to finish
2897 * @error: %0 for success, < %0 for error
2900 * Completely finish @rq. Must be called with queue lock held.
2902 void __blk_end_request_all(struct request
*rq
, int error
)
2905 unsigned int bidi_bytes
= 0;
2907 if (unlikely(blk_bidi_rq(rq
)))
2908 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2910 pending
= __blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2913 EXPORT_SYMBOL(__blk_end_request_all
);
2916 * __blk_end_request_cur - Helper function to finish the current request chunk.
2917 * @rq: the request to finish the current chunk for
2918 * @error: %0 for success, < %0 for error
2921 * Complete the current consecutively mapped chunk from @rq. Must
2922 * be called with queue lock held.
2925 * %false - we are done with this request
2926 * %true - still buffers pending for this request
2928 bool __blk_end_request_cur(struct request
*rq
, int error
)
2930 return __blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2932 EXPORT_SYMBOL(__blk_end_request_cur
);
2935 * __blk_end_request_err - Finish a request till the next failure boundary.
2936 * @rq: the request to finish till the next failure boundary for
2937 * @error: must be negative errno
2940 * Complete @rq till the next failure boundary. Must be called
2941 * with queue lock held.
2944 * %false - we are done with this request
2945 * %true - still buffers pending for this request
2947 bool __blk_end_request_err(struct request
*rq
, int error
)
2949 WARN_ON(error
>= 0);
2950 return __blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2952 EXPORT_SYMBOL_GPL(__blk_end_request_err
);
2954 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
2957 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2958 rq
->cmd_flags
|= bio
->bi_rw
& REQ_WRITE
;
2960 if (bio_has_data(bio
))
2961 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2963 rq
->__data_len
= bio
->bi_iter
.bi_size
;
2964 rq
->bio
= rq
->biotail
= bio
;
2967 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2970 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2972 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2973 * @rq: the request to be flushed
2976 * Flush all pages in @rq.
2978 void rq_flush_dcache_pages(struct request
*rq
)
2980 struct req_iterator iter
;
2981 struct bio_vec bvec
;
2983 rq_for_each_segment(bvec
, rq
, iter
)
2984 flush_dcache_page(bvec
.bv_page
);
2986 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages
);
2990 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2991 * @q : the queue of the device being checked
2994 * Check if underlying low-level drivers of a device are busy.
2995 * If the drivers want to export their busy state, they must set own
2996 * exporting function using blk_queue_lld_busy() first.
2998 * Basically, this function is used only by request stacking drivers
2999 * to stop dispatching requests to underlying devices when underlying
3000 * devices are busy. This behavior helps more I/O merging on the queue
3001 * of the request stacking driver and prevents I/O throughput regression
3002 * on burst I/O load.
3005 * 0 - Not busy (The request stacking driver should dispatch request)
3006 * 1 - Busy (The request stacking driver should stop dispatching request)
3008 int blk_lld_busy(struct request_queue
*q
)
3011 return q
->lld_busy_fn(q
);
3015 EXPORT_SYMBOL_GPL(blk_lld_busy
);
3018 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
3019 * @rq: the clone request to be cleaned up
3022 * Free all bios in @rq for a cloned request.
3024 void blk_rq_unprep_clone(struct request
*rq
)
3028 while ((bio
= rq
->bio
) != NULL
) {
3029 rq
->bio
= bio
->bi_next
;
3034 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone
);
3037 * Copy attributes of the original request to the clone request.
3038 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3040 static void __blk_rq_prep_clone(struct request
*dst
, struct request
*src
)
3042 dst
->cpu
= src
->cpu
;
3043 dst
->cmd_flags
|= (src
->cmd_flags
& REQ_CLONE_MASK
) | REQ_NOMERGE
;
3044 dst
->cmd_type
= src
->cmd_type
;
3045 dst
->__sector
= blk_rq_pos(src
);
3046 dst
->__data_len
= blk_rq_bytes(src
);
3047 dst
->nr_phys_segments
= src
->nr_phys_segments
;
3048 dst
->ioprio
= src
->ioprio
;
3049 dst
->extra_len
= src
->extra_len
;
3053 * blk_rq_prep_clone - Helper function to setup clone request
3054 * @rq: the request to be setup
3055 * @rq_src: original request to be cloned
3056 * @bs: bio_set that bios for clone are allocated from
3057 * @gfp_mask: memory allocation mask for bio
3058 * @bio_ctr: setup function to be called for each clone bio.
3059 * Returns %0 for success, non %0 for failure.
3060 * @data: private data to be passed to @bio_ctr
3063 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3064 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3065 * are not copied, and copying such parts is the caller's responsibility.
3066 * Also, pages which the original bios are pointing to are not copied
3067 * and the cloned bios just point same pages.
3068 * So cloned bios must be completed before original bios, which means
3069 * the caller must complete @rq before @rq_src.
3071 int blk_rq_prep_clone(struct request
*rq
, struct request
*rq_src
,
3072 struct bio_set
*bs
, gfp_t gfp_mask
,
3073 int (*bio_ctr
)(struct bio
*, struct bio
*, void *),
3076 struct bio
*bio
, *bio_src
;
3081 __rq_for_each_bio(bio_src
, rq_src
) {
3082 bio
= bio_clone_fast(bio_src
, gfp_mask
, bs
);
3086 if (bio_ctr
&& bio_ctr(bio
, bio_src
, data
))
3090 rq
->biotail
->bi_next
= bio
;
3093 rq
->bio
= rq
->biotail
= bio
;
3096 __blk_rq_prep_clone(rq
, rq_src
);
3103 blk_rq_unprep_clone(rq
);
3107 EXPORT_SYMBOL_GPL(blk_rq_prep_clone
);
3109 int kblockd_schedule_work(struct work_struct
*work
)
3111 return queue_work(kblockd_workqueue
, work
);
3113 EXPORT_SYMBOL(kblockd_schedule_work
);
3115 int kblockd_schedule_delayed_work(struct delayed_work
*dwork
,
3116 unsigned long delay
)
3118 return queue_delayed_work(kblockd_workqueue
, dwork
, delay
);
3120 EXPORT_SYMBOL(kblockd_schedule_delayed_work
);
3122 int kblockd_schedule_delayed_work_on(int cpu
, struct delayed_work
*dwork
,
3123 unsigned long delay
)
3125 return queue_delayed_work_on(cpu
, kblockd_workqueue
, dwork
, delay
);
3127 EXPORT_SYMBOL(kblockd_schedule_delayed_work_on
);
3130 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3131 * @plug: The &struct blk_plug that needs to be initialized
3134 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3135 * pending I/O should the task end up blocking between blk_start_plug() and
3136 * blk_finish_plug(). This is important from a performance perspective, but
3137 * also ensures that we don't deadlock. For instance, if the task is blocking
3138 * for a memory allocation, memory reclaim could end up wanting to free a
3139 * page belonging to that request that is currently residing in our private
3140 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3141 * this kind of deadlock.
3143 void blk_start_plug(struct blk_plug
*plug
)
3145 struct task_struct
*tsk
= current
;
3148 * If this is a nested plug, don't actually assign it.
3153 INIT_LIST_HEAD(&plug
->list
);
3154 INIT_LIST_HEAD(&plug
->mq_list
);
3155 INIT_LIST_HEAD(&plug
->cb_list
);
3157 * Store ordering should not be needed here, since a potential
3158 * preempt will imply a full memory barrier
3162 EXPORT_SYMBOL(blk_start_plug
);
3164 static int plug_rq_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
3166 struct request
*rqa
= container_of(a
, struct request
, queuelist
);
3167 struct request
*rqb
= container_of(b
, struct request
, queuelist
);
3169 return !(rqa
->q
< rqb
->q
||
3170 (rqa
->q
== rqb
->q
&& blk_rq_pos(rqa
) < blk_rq_pos(rqb
)));
3174 * If 'from_schedule' is true, then postpone the dispatch of requests
3175 * until a safe kblockd context. We due this to avoid accidental big
3176 * additional stack usage in driver dispatch, in places where the originally
3177 * plugger did not intend it.
3179 static void queue_unplugged(struct request_queue
*q
, unsigned int depth
,
3181 __releases(q
->queue_lock
)
3183 trace_block_unplug(q
, depth
, !from_schedule
);
3186 blk_run_queue_async(q
);
3189 spin_unlock(q
->queue_lock
);
3192 static void flush_plug_callbacks(struct blk_plug
*plug
, bool from_schedule
)
3194 LIST_HEAD(callbacks
);
3196 while (!list_empty(&plug
->cb_list
)) {
3197 list_splice_init(&plug
->cb_list
, &callbacks
);
3199 while (!list_empty(&callbacks
)) {
3200 struct blk_plug_cb
*cb
= list_first_entry(&callbacks
,
3203 list_del(&cb
->list
);
3204 cb
->callback(cb
, from_schedule
);
3209 struct blk_plug_cb
*blk_check_plugged(blk_plug_cb_fn unplug
, void *data
,
3212 struct blk_plug
*plug
= current
->plug
;
3213 struct blk_plug_cb
*cb
;
3218 list_for_each_entry(cb
, &plug
->cb_list
, list
)
3219 if (cb
->callback
== unplug
&& cb
->data
== data
)
3222 /* Not currently on the callback list */
3223 BUG_ON(size
< sizeof(*cb
));
3224 cb
= kzalloc(size
, GFP_ATOMIC
);
3227 cb
->callback
= unplug
;
3228 list_add(&cb
->list
, &plug
->cb_list
);
3232 EXPORT_SYMBOL(blk_check_plugged
);
3234 void blk_flush_plug_list(struct blk_plug
*plug
, bool from_schedule
)
3236 struct request_queue
*q
;
3237 unsigned long flags
;
3242 flush_plug_callbacks(plug
, from_schedule
);
3244 if (!list_empty(&plug
->mq_list
))
3245 blk_mq_flush_plug_list(plug
, from_schedule
);
3247 if (list_empty(&plug
->list
))
3250 list_splice_init(&plug
->list
, &list
);
3252 list_sort(NULL
, &list
, plug_rq_cmp
);
3258 * Save and disable interrupts here, to avoid doing it for every
3259 * queue lock we have to take.
3261 local_irq_save(flags
);
3262 while (!list_empty(&list
)) {
3263 rq
= list_entry_rq(list
.next
);
3264 list_del_init(&rq
->queuelist
);
3268 * This drops the queue lock
3271 queue_unplugged(q
, depth
, from_schedule
);
3274 spin_lock(q
->queue_lock
);
3278 * Short-circuit if @q is dead
3280 if (unlikely(blk_queue_dying(q
))) {
3281 __blk_end_request_all(rq
, -ENODEV
);
3286 * rq is already accounted, so use raw insert
3288 if (rq
->cmd_flags
& (REQ_FLUSH
| REQ_FUA
))
3289 __elv_add_request(q
, rq
, ELEVATOR_INSERT_FLUSH
);
3291 __elv_add_request(q
, rq
, ELEVATOR_INSERT_SORT_MERGE
);
3297 * This drops the queue lock
3300 queue_unplugged(q
, depth
, from_schedule
);
3302 local_irq_restore(flags
);
3305 void blk_finish_plug(struct blk_plug
*plug
)
3307 if (plug
!= current
->plug
)
3309 blk_flush_plug_list(plug
, false);
3311 current
->plug
= NULL
;
3313 EXPORT_SYMBOL(blk_finish_plug
);
3315 bool blk_poll(struct request_queue
*q
, blk_qc_t cookie
)
3317 struct blk_plug
*plug
;
3320 if (!q
->mq_ops
|| !q
->mq_ops
->poll
|| !blk_qc_t_valid(cookie
) ||
3321 !test_bit(QUEUE_FLAG_POLL
, &q
->queue_flags
))
3324 plug
= current
->plug
;
3326 blk_flush_plug_list(plug
, false);
3328 state
= current
->state
;
3329 while (!need_resched()) {
3330 unsigned int queue_num
= blk_qc_t_to_queue_num(cookie
);
3331 struct blk_mq_hw_ctx
*hctx
= q
->queue_hw_ctx
[queue_num
];
3334 hctx
->poll_invoked
++;
3336 ret
= q
->mq_ops
->poll(hctx
, blk_qc_t_to_tag(cookie
));
3338 hctx
->poll_success
++;
3339 set_current_state(TASK_RUNNING
);
3343 if (signal_pending_state(state
, current
))
3344 set_current_state(TASK_RUNNING
);
3346 if (current
->state
== TASK_RUNNING
)
3358 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3359 * @q: the queue of the device
3360 * @dev: the device the queue belongs to
3363 * Initialize runtime-PM-related fields for @q and start auto suspend for
3364 * @dev. Drivers that want to take advantage of request-based runtime PM
3365 * should call this function after @dev has been initialized, and its
3366 * request queue @q has been allocated, and runtime PM for it can not happen
3367 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3368 * cases, driver should call this function before any I/O has taken place.
3370 * This function takes care of setting up using auto suspend for the device,
3371 * the autosuspend delay is set to -1 to make runtime suspend impossible
3372 * until an updated value is either set by user or by driver. Drivers do
3373 * not need to touch other autosuspend settings.
3375 * The block layer runtime PM is request based, so only works for drivers
3376 * that use request as their IO unit instead of those directly use bio's.
3378 void blk_pm_runtime_init(struct request_queue
*q
, struct device
*dev
)
3381 q
->rpm_status
= RPM_ACTIVE
;
3382 pm_runtime_set_autosuspend_delay(q
->dev
, -1);
3383 pm_runtime_use_autosuspend(q
->dev
);
3385 EXPORT_SYMBOL(blk_pm_runtime_init
);
3388 * blk_pre_runtime_suspend - Pre runtime suspend check
3389 * @q: the queue of the device
3392 * This function will check if runtime suspend is allowed for the device
3393 * by examining if there are any requests pending in the queue. If there
3394 * are requests pending, the device can not be runtime suspended; otherwise,
3395 * the queue's status will be updated to SUSPENDING and the driver can
3396 * proceed to suspend the device.
3398 * For the not allowed case, we mark last busy for the device so that
3399 * runtime PM core will try to autosuspend it some time later.
3401 * This function should be called near the start of the device's
3402 * runtime_suspend callback.
3405 * 0 - OK to runtime suspend the device
3406 * -EBUSY - Device should not be runtime suspended
3408 int blk_pre_runtime_suspend(struct request_queue
*q
)
3412 spin_lock_irq(q
->queue_lock
);
3413 if (q
->nr_pending
) {
3415 pm_runtime_mark_last_busy(q
->dev
);
3417 q
->rpm_status
= RPM_SUSPENDING
;
3419 spin_unlock_irq(q
->queue_lock
);
3422 EXPORT_SYMBOL(blk_pre_runtime_suspend
);
3425 * blk_post_runtime_suspend - Post runtime suspend processing
3426 * @q: the queue of the device
3427 * @err: return value of the device's runtime_suspend function
3430 * Update the queue's runtime status according to the return value of the
3431 * device's runtime suspend function and mark last busy for the device so
3432 * that PM core will try to auto suspend the device at a later time.
3434 * This function should be called near the end of the device's
3435 * runtime_suspend callback.
3437 void blk_post_runtime_suspend(struct request_queue
*q
, int err
)
3439 spin_lock_irq(q
->queue_lock
);
3441 q
->rpm_status
= RPM_SUSPENDED
;
3443 q
->rpm_status
= RPM_ACTIVE
;
3444 pm_runtime_mark_last_busy(q
->dev
);
3446 spin_unlock_irq(q
->queue_lock
);
3448 EXPORT_SYMBOL(blk_post_runtime_suspend
);
3451 * blk_pre_runtime_resume - Pre runtime resume processing
3452 * @q: the queue of the device
3455 * Update the queue's runtime status to RESUMING in preparation for the
3456 * runtime resume of the device.
3458 * This function should be called near the start of the device's
3459 * runtime_resume callback.
3461 void blk_pre_runtime_resume(struct request_queue
*q
)
3463 spin_lock_irq(q
->queue_lock
);
3464 q
->rpm_status
= RPM_RESUMING
;
3465 spin_unlock_irq(q
->queue_lock
);
3467 EXPORT_SYMBOL(blk_pre_runtime_resume
);
3470 * blk_post_runtime_resume - Post runtime resume processing
3471 * @q: the queue of the device
3472 * @err: return value of the device's runtime_resume function
3475 * Update the queue's runtime status according to the return value of the
3476 * device's runtime_resume function. If it is successfully resumed, process
3477 * the requests that are queued into the device's queue when it is resuming
3478 * and then mark last busy and initiate autosuspend for it.
3480 * This function should be called near the end of the device's
3481 * runtime_resume callback.
3483 void blk_post_runtime_resume(struct request_queue
*q
, int err
)
3485 spin_lock_irq(q
->queue_lock
);
3487 q
->rpm_status
= RPM_ACTIVE
;
3489 pm_runtime_mark_last_busy(q
->dev
);
3490 pm_request_autosuspend(q
->dev
);
3492 q
->rpm_status
= RPM_SUSPENDED
;
3494 spin_unlock_irq(q
->queue_lock
);
3496 EXPORT_SYMBOL(blk_post_runtime_resume
);
3499 int __init
blk_dev_init(void)
3501 BUILD_BUG_ON(__REQ_NR_BITS
> 8 *
3502 FIELD_SIZEOF(struct request
, cmd_flags
));
3504 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3505 kblockd_workqueue
= alloc_workqueue("kblockd",
3506 WQ_MEM_RECLAIM
| WQ_HIGHPRI
, 0);
3507 if (!kblockd_workqueue
)
3508 panic("Failed to create kblockd\n");
3510 request_cachep
= kmem_cache_create("blkdev_requests",
3511 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
3513 blk_requestq_cachep
= kmem_cache_create("blkdev_queue",
3514 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);