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
;
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
->rq_flags
& RQF_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
->rq_flags
& RQF_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_async - asynchronously restart a previously stopped queue
211 * @q: The &struct request_queue in question
214 * blk_start_queue_async() will clear the stop flag on the queue, and
215 * ensure that the request_fn for the queue is run from an async
218 void blk_start_queue_async(struct request_queue
*q
)
220 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
221 blk_run_queue_async(q
);
223 EXPORT_SYMBOL(blk_start_queue_async
);
226 * blk_start_queue - restart a previously stopped queue
227 * @q: The &struct request_queue in question
230 * blk_start_queue() will clear the stop flag on the queue, and call
231 * the request_fn for the queue if it was in a stopped state when
232 * entered. Also see blk_stop_queue(). Queue lock must be held.
234 void blk_start_queue(struct request_queue
*q
)
236 WARN_ON(!irqs_disabled());
238 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
241 EXPORT_SYMBOL(blk_start_queue
);
244 * blk_stop_queue - stop a queue
245 * @q: The &struct request_queue in question
248 * The Linux block layer assumes that a block driver will consume all
249 * entries on the request queue when the request_fn strategy is called.
250 * Often this will not happen, because of hardware limitations (queue
251 * depth settings). If a device driver gets a 'queue full' response,
252 * or if it simply chooses not to queue more I/O at one point, it can
253 * call this function to prevent the request_fn from being called until
254 * the driver has signalled it's ready to go again. This happens by calling
255 * blk_start_queue() to restart queue operations. Queue lock must be held.
257 void blk_stop_queue(struct request_queue
*q
)
259 cancel_delayed_work(&q
->delay_work
);
260 queue_flag_set(QUEUE_FLAG_STOPPED
, q
);
262 EXPORT_SYMBOL(blk_stop_queue
);
265 * blk_sync_queue - cancel any pending callbacks on a queue
269 * The block layer may perform asynchronous callback activity
270 * on a queue, such as calling the unplug function after a timeout.
271 * A block device may call blk_sync_queue to ensure that any
272 * such activity is cancelled, thus allowing it to release resources
273 * that the callbacks might use. The caller must already have made sure
274 * that its ->make_request_fn will not re-add plugging prior to calling
277 * This function does not cancel any asynchronous activity arising
278 * out of elevator or throttling code. That would require elevator_exit()
279 * and blkcg_exit_queue() to be called with queue lock initialized.
282 void blk_sync_queue(struct request_queue
*q
)
284 del_timer_sync(&q
->timeout
);
287 struct blk_mq_hw_ctx
*hctx
;
290 queue_for_each_hw_ctx(q
, hctx
, i
) {
291 cancel_work_sync(&hctx
->run_work
);
292 cancel_delayed_work_sync(&hctx
->delay_work
);
295 cancel_delayed_work_sync(&q
->delay_work
);
298 EXPORT_SYMBOL(blk_sync_queue
);
301 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
302 * @q: The queue to run
305 * Invoke request handling on a queue if there are any pending requests.
306 * May be used to restart request handling after a request has completed.
307 * This variant runs the queue whether or not the queue has been
308 * stopped. Must be called with the queue lock held and interrupts
309 * disabled. See also @blk_run_queue.
311 inline void __blk_run_queue_uncond(struct request_queue
*q
)
313 if (unlikely(blk_queue_dead(q
)))
317 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
318 * the queue lock internally. As a result multiple threads may be
319 * running such a request function concurrently. Keep track of the
320 * number of active request_fn invocations such that blk_drain_queue()
321 * can wait until all these request_fn calls have finished.
323 q
->request_fn_active
++;
325 q
->request_fn_active
--;
327 EXPORT_SYMBOL_GPL(__blk_run_queue_uncond
);
330 * __blk_run_queue - run a single device queue
331 * @q: The queue to run
334 * See @blk_run_queue. This variant must be called with the queue lock
335 * held and interrupts disabled.
337 void __blk_run_queue(struct request_queue
*q
)
339 if (unlikely(blk_queue_stopped(q
)))
342 __blk_run_queue_uncond(q
);
344 EXPORT_SYMBOL(__blk_run_queue
);
347 * blk_run_queue_async - run a single device queue in workqueue context
348 * @q: The queue to run
351 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
352 * of us. The caller must hold the queue lock.
354 void blk_run_queue_async(struct request_queue
*q
)
356 if (likely(!blk_queue_stopped(q
) && !blk_queue_dead(q
)))
357 mod_delayed_work(kblockd_workqueue
, &q
->delay_work
, 0);
359 EXPORT_SYMBOL(blk_run_queue_async
);
362 * blk_run_queue - run a single device queue
363 * @q: The queue to run
366 * Invoke request handling on this queue, if it has pending work to do.
367 * May be used to restart queueing when a request has completed.
369 void blk_run_queue(struct request_queue
*q
)
373 spin_lock_irqsave(q
->queue_lock
, flags
);
375 spin_unlock_irqrestore(q
->queue_lock
, flags
);
377 EXPORT_SYMBOL(blk_run_queue
);
379 void blk_put_queue(struct request_queue
*q
)
381 kobject_put(&q
->kobj
);
383 EXPORT_SYMBOL(blk_put_queue
);
386 * __blk_drain_queue - drain requests from request_queue
388 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
390 * Drain requests from @q. If @drain_all is set, all requests are drained.
391 * If not, only ELVPRIV requests are drained. The caller is responsible
392 * for ensuring that no new requests which need to be drained are queued.
394 static void __blk_drain_queue(struct request_queue
*q
, bool drain_all
)
395 __releases(q
->queue_lock
)
396 __acquires(q
->queue_lock
)
400 lockdep_assert_held(q
->queue_lock
);
406 * The caller might be trying to drain @q before its
407 * elevator is initialized.
410 elv_drain_elevator(q
);
412 blkcg_drain_queue(q
);
415 * This function might be called on a queue which failed
416 * driver init after queue creation or is not yet fully
417 * active yet. Some drivers (e.g. fd and loop) get unhappy
418 * in such cases. Kick queue iff dispatch queue has
419 * something on it and @q has request_fn set.
421 if (!list_empty(&q
->queue_head
) && q
->request_fn
)
424 drain
|= q
->nr_rqs_elvpriv
;
425 drain
|= q
->request_fn_active
;
428 * Unfortunately, requests are queued at and tracked from
429 * multiple places and there's no single counter which can
430 * be drained. Check all the queues and counters.
433 struct blk_flush_queue
*fq
= blk_get_flush_queue(q
, NULL
);
434 drain
|= !list_empty(&q
->queue_head
);
435 for (i
= 0; i
< 2; i
++) {
436 drain
|= q
->nr_rqs
[i
];
437 drain
|= q
->in_flight
[i
];
439 drain
|= !list_empty(&fq
->flush_queue
[i
]);
446 spin_unlock_irq(q
->queue_lock
);
450 spin_lock_irq(q
->queue_lock
);
454 * With queue marked dead, any woken up waiter will fail the
455 * allocation path, so the wakeup chaining is lost and we're
456 * left with hung waiters. We need to wake up those waiters.
459 struct request_list
*rl
;
461 blk_queue_for_each_rl(rl
, q
)
462 for (i
= 0; i
< ARRAY_SIZE(rl
->wait
); i
++)
463 wake_up_all(&rl
->wait
[i
]);
468 * blk_queue_bypass_start - enter queue bypass mode
469 * @q: queue of interest
471 * In bypass mode, only the dispatch FIFO queue of @q is used. This
472 * function makes @q enter bypass mode and drains all requests which were
473 * throttled or issued before. On return, it's guaranteed that no request
474 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
475 * inside queue or RCU read lock.
477 void blk_queue_bypass_start(struct request_queue
*q
)
479 spin_lock_irq(q
->queue_lock
);
481 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
482 spin_unlock_irq(q
->queue_lock
);
485 * Queues start drained. Skip actual draining till init is
486 * complete. This avoids lenghty delays during queue init which
487 * can happen many times during boot.
489 if (blk_queue_init_done(q
)) {
490 spin_lock_irq(q
->queue_lock
);
491 __blk_drain_queue(q
, false);
492 spin_unlock_irq(q
->queue_lock
);
494 /* ensure blk_queue_bypass() is %true inside RCU read lock */
498 EXPORT_SYMBOL_GPL(blk_queue_bypass_start
);
501 * blk_queue_bypass_end - leave queue bypass mode
502 * @q: queue of interest
504 * Leave bypass mode and restore the normal queueing behavior.
506 void blk_queue_bypass_end(struct request_queue
*q
)
508 spin_lock_irq(q
->queue_lock
);
509 if (!--q
->bypass_depth
)
510 queue_flag_clear(QUEUE_FLAG_BYPASS
, q
);
511 WARN_ON_ONCE(q
->bypass_depth
< 0);
512 spin_unlock_irq(q
->queue_lock
);
514 EXPORT_SYMBOL_GPL(blk_queue_bypass_end
);
516 void blk_set_queue_dying(struct request_queue
*q
)
518 spin_lock_irq(q
->queue_lock
);
519 queue_flag_set(QUEUE_FLAG_DYING
, q
);
520 spin_unlock_irq(q
->queue_lock
);
523 blk_mq_wake_waiters(q
);
525 struct request_list
*rl
;
527 blk_queue_for_each_rl(rl
, q
) {
529 wake_up(&rl
->wait
[BLK_RW_SYNC
]);
530 wake_up(&rl
->wait
[BLK_RW_ASYNC
]);
535 EXPORT_SYMBOL_GPL(blk_set_queue_dying
);
538 * blk_cleanup_queue - shutdown a request queue
539 * @q: request queue to shutdown
541 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
542 * put it. All future requests will be failed immediately with -ENODEV.
544 void blk_cleanup_queue(struct request_queue
*q
)
546 spinlock_t
*lock
= q
->queue_lock
;
548 /* mark @q DYING, no new request or merges will be allowed afterwards */
549 mutex_lock(&q
->sysfs_lock
);
550 blk_set_queue_dying(q
);
554 * A dying queue is permanently in bypass mode till released. Note
555 * that, unlike blk_queue_bypass_start(), we aren't performing
556 * synchronize_rcu() after entering bypass mode to avoid the delay
557 * as some drivers create and destroy a lot of queues while
558 * probing. This is still safe because blk_release_queue() will be
559 * called only after the queue refcnt drops to zero and nothing,
560 * RCU or not, would be traversing the queue by then.
563 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
565 queue_flag_set(QUEUE_FLAG_NOMERGES
, q
);
566 queue_flag_set(QUEUE_FLAG_NOXMERGES
, q
);
567 queue_flag_set(QUEUE_FLAG_DYING
, q
);
568 spin_unlock_irq(lock
);
569 mutex_unlock(&q
->sysfs_lock
);
572 * Drain all requests queued before DYING marking. Set DEAD flag to
573 * prevent that q->request_fn() gets invoked after draining finished.
578 __blk_drain_queue(q
, true);
579 queue_flag_set(QUEUE_FLAG_DEAD
, q
);
580 spin_unlock_irq(lock
);
582 /* for synchronous bio-based driver finish in-flight integrity i/o */
583 blk_flush_integrity();
585 /* @q won't process any more request, flush async actions */
586 del_timer_sync(&q
->backing_dev_info
.laptop_mode_wb_timer
);
590 blk_mq_free_queue(q
);
591 percpu_ref_exit(&q
->q_usage_counter
);
594 if (q
->queue_lock
!= &q
->__queue_lock
)
595 q
->queue_lock
= &q
->__queue_lock
;
596 spin_unlock_irq(lock
);
598 bdi_unregister(&q
->backing_dev_info
);
600 /* @q is and will stay empty, shutdown and put */
603 EXPORT_SYMBOL(blk_cleanup_queue
);
605 /* Allocate memory local to the request queue */
606 static void *alloc_request_struct(gfp_t gfp_mask
, void *data
)
608 int nid
= (int)(long)data
;
609 return kmem_cache_alloc_node(request_cachep
, gfp_mask
, nid
);
612 static void free_request_struct(void *element
, void *unused
)
614 kmem_cache_free(request_cachep
, element
);
617 int blk_init_rl(struct request_list
*rl
, struct request_queue
*q
,
620 if (unlikely(rl
->rq_pool
))
624 rl
->count
[BLK_RW_SYNC
] = rl
->count
[BLK_RW_ASYNC
] = 0;
625 rl
->starved
[BLK_RW_SYNC
] = rl
->starved
[BLK_RW_ASYNC
] = 0;
626 init_waitqueue_head(&rl
->wait
[BLK_RW_SYNC
]);
627 init_waitqueue_head(&rl
->wait
[BLK_RW_ASYNC
]);
629 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, alloc_request_struct
,
631 (void *)(long)q
->node
, gfp_mask
,
639 void blk_exit_rl(struct request_list
*rl
)
642 mempool_destroy(rl
->rq_pool
);
645 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
647 return blk_alloc_queue_node(gfp_mask
, NUMA_NO_NODE
);
649 EXPORT_SYMBOL(blk_alloc_queue
);
651 int blk_queue_enter(struct request_queue
*q
, bool nowait
)
656 if (percpu_ref_tryget_live(&q
->q_usage_counter
))
662 ret
= wait_event_interruptible(q
->mq_freeze_wq
,
663 !atomic_read(&q
->mq_freeze_depth
) ||
665 if (blk_queue_dying(q
))
672 void blk_queue_exit(struct request_queue
*q
)
674 percpu_ref_put(&q
->q_usage_counter
);
677 static void blk_queue_usage_counter_release(struct percpu_ref
*ref
)
679 struct request_queue
*q
=
680 container_of(ref
, struct request_queue
, q_usage_counter
);
682 wake_up_all(&q
->mq_freeze_wq
);
685 static void blk_rq_timed_out_timer(unsigned long data
)
687 struct request_queue
*q
= (struct request_queue
*)data
;
689 kblockd_schedule_work(&q
->timeout_work
);
692 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
694 struct request_queue
*q
;
697 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
698 gfp_mask
| __GFP_ZERO
, node_id
);
702 q
->id
= ida_simple_get(&blk_queue_ida
, 0, 0, gfp_mask
);
706 q
->bio_split
= bioset_create(BIO_POOL_SIZE
, 0);
710 q
->backing_dev_info
.ra_pages
=
711 (VM_MAX_READAHEAD
* 1024) / PAGE_SIZE
;
712 q
->backing_dev_info
.capabilities
= BDI_CAP_CGROUP_WRITEBACK
;
713 q
->backing_dev_info
.name
= "block";
716 err
= bdi_init(&q
->backing_dev_info
);
720 setup_timer(&q
->backing_dev_info
.laptop_mode_wb_timer
,
721 laptop_mode_timer_fn
, (unsigned long) q
);
722 setup_timer(&q
->timeout
, blk_rq_timed_out_timer
, (unsigned long) q
);
723 INIT_LIST_HEAD(&q
->queue_head
);
724 INIT_LIST_HEAD(&q
->timeout_list
);
725 INIT_LIST_HEAD(&q
->icq_list
);
726 #ifdef CONFIG_BLK_CGROUP
727 INIT_LIST_HEAD(&q
->blkg_list
);
729 INIT_DELAYED_WORK(&q
->delay_work
, blk_delay_work
);
731 kobject_init(&q
->kobj
, &blk_queue_ktype
);
733 mutex_init(&q
->sysfs_lock
);
734 spin_lock_init(&q
->__queue_lock
);
737 * By default initialize queue_lock to internal lock and driver can
738 * override it later if need be.
740 q
->queue_lock
= &q
->__queue_lock
;
743 * A queue starts its life with bypass turned on to avoid
744 * unnecessary bypass on/off overhead and nasty surprises during
745 * init. The initial bypass will be finished when the queue is
746 * registered by blk_register_queue().
749 __set_bit(QUEUE_FLAG_BYPASS
, &q
->queue_flags
);
751 init_waitqueue_head(&q
->mq_freeze_wq
);
754 * Init percpu_ref in atomic mode so that it's faster to shutdown.
755 * See blk_register_queue() for details.
757 if (percpu_ref_init(&q
->q_usage_counter
,
758 blk_queue_usage_counter_release
,
759 PERCPU_REF_INIT_ATOMIC
, GFP_KERNEL
))
762 if (blkcg_init_queue(q
))
768 percpu_ref_exit(&q
->q_usage_counter
);
770 bdi_destroy(&q
->backing_dev_info
);
772 bioset_free(q
->bio_split
);
774 ida_simple_remove(&blk_queue_ida
, q
->id
);
776 kmem_cache_free(blk_requestq_cachep
, q
);
779 EXPORT_SYMBOL(blk_alloc_queue_node
);
782 * blk_init_queue - prepare a request queue for use with a block device
783 * @rfn: The function to be called to process requests that have been
784 * placed on the queue.
785 * @lock: Request queue spin lock
788 * If a block device wishes to use the standard request handling procedures,
789 * which sorts requests and coalesces adjacent requests, then it must
790 * call blk_init_queue(). The function @rfn will be called when there
791 * are requests on the queue that need to be processed. If the device
792 * supports plugging, then @rfn may not be called immediately when requests
793 * are available on the queue, but may be called at some time later instead.
794 * Plugged queues are generally unplugged when a buffer belonging to one
795 * of the requests on the queue is needed, or due to memory pressure.
797 * @rfn is not required, or even expected, to remove all requests off the
798 * queue, but only as many as it can handle at a time. If it does leave
799 * requests on the queue, it is responsible for arranging that the requests
800 * get dealt with eventually.
802 * The queue spin lock must be held while manipulating the requests on the
803 * request queue; this lock will be taken also from interrupt context, so irq
804 * disabling is needed for it.
806 * Function returns a pointer to the initialized request queue, or %NULL if
810 * blk_init_queue() must be paired with a blk_cleanup_queue() call
811 * when the block device is deactivated (such as at module unload).
814 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
816 return blk_init_queue_node(rfn
, lock
, NUMA_NO_NODE
);
818 EXPORT_SYMBOL(blk_init_queue
);
820 struct request_queue
*
821 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
823 struct request_queue
*uninit_q
, *q
;
825 uninit_q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
829 q
= blk_init_allocated_queue(uninit_q
, rfn
, lock
);
831 blk_cleanup_queue(uninit_q
);
835 EXPORT_SYMBOL(blk_init_queue_node
);
837 static blk_qc_t
blk_queue_bio(struct request_queue
*q
, struct bio
*bio
);
839 struct request_queue
*
840 blk_init_allocated_queue(struct request_queue
*q
, request_fn_proc
*rfn
,
846 q
->fq
= blk_alloc_flush_queue(q
, NUMA_NO_NODE
, 0);
850 if (blk_init_rl(&q
->root_rl
, q
, GFP_KERNEL
))
853 INIT_WORK(&q
->timeout_work
, blk_timeout_work
);
855 q
->prep_rq_fn
= NULL
;
856 q
->unprep_rq_fn
= NULL
;
857 q
->queue_flags
|= QUEUE_FLAG_DEFAULT
;
859 /* Override internal queue lock with supplied lock pointer */
861 q
->queue_lock
= lock
;
864 * This also sets hw/phys segments, boundary and size
866 blk_queue_make_request(q
, blk_queue_bio
);
868 q
->sg_reserved_size
= INT_MAX
;
870 /* Protect q->elevator from elevator_change */
871 mutex_lock(&q
->sysfs_lock
);
874 if (elevator_init(q
, NULL
)) {
875 mutex_unlock(&q
->sysfs_lock
);
879 mutex_unlock(&q
->sysfs_lock
);
884 blk_free_flush_queue(q
->fq
);
887 EXPORT_SYMBOL(blk_init_allocated_queue
);
889 bool blk_get_queue(struct request_queue
*q
)
891 if (likely(!blk_queue_dying(q
))) {
898 EXPORT_SYMBOL(blk_get_queue
);
900 static inline void blk_free_request(struct request_list
*rl
, struct request
*rq
)
902 if (rq
->rq_flags
& RQF_ELVPRIV
) {
903 elv_put_request(rl
->q
, rq
);
905 put_io_context(rq
->elv
.icq
->ioc
);
908 mempool_free(rq
, rl
->rq_pool
);
912 * ioc_batching returns true if the ioc is a valid batching request and
913 * should be given priority access to a request.
915 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
921 * Make sure the process is able to allocate at least 1 request
922 * even if the batch times out, otherwise we could theoretically
925 return ioc
->nr_batch_requests
== q
->nr_batching
||
926 (ioc
->nr_batch_requests
> 0
927 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
931 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
932 * will cause the process to be a "batcher" on all queues in the system. This
933 * is the behaviour we want though - once it gets a wakeup it should be given
936 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
938 if (!ioc
|| ioc_batching(q
, ioc
))
941 ioc
->nr_batch_requests
= q
->nr_batching
;
942 ioc
->last_waited
= jiffies
;
945 static void __freed_request(struct request_list
*rl
, int sync
)
947 struct request_queue
*q
= rl
->q
;
949 if (rl
->count
[sync
] < queue_congestion_off_threshold(q
))
950 blk_clear_congested(rl
, sync
);
952 if (rl
->count
[sync
] + 1 <= q
->nr_requests
) {
953 if (waitqueue_active(&rl
->wait
[sync
]))
954 wake_up(&rl
->wait
[sync
]);
956 blk_clear_rl_full(rl
, sync
);
961 * A request has just been released. Account for it, update the full and
962 * congestion status, wake up any waiters. Called under q->queue_lock.
964 static void freed_request(struct request_list
*rl
, bool sync
,
965 req_flags_t rq_flags
)
967 struct request_queue
*q
= rl
->q
;
971 if (rq_flags
& RQF_ELVPRIV
)
974 __freed_request(rl
, sync
);
976 if (unlikely(rl
->starved
[sync
^ 1]))
977 __freed_request(rl
, sync
^ 1);
980 int blk_update_nr_requests(struct request_queue
*q
, unsigned int nr
)
982 struct request_list
*rl
;
983 int on_thresh
, off_thresh
;
985 spin_lock_irq(q
->queue_lock
);
987 blk_queue_congestion_threshold(q
);
988 on_thresh
= queue_congestion_on_threshold(q
);
989 off_thresh
= queue_congestion_off_threshold(q
);
991 blk_queue_for_each_rl(rl
, q
) {
992 if (rl
->count
[BLK_RW_SYNC
] >= on_thresh
)
993 blk_set_congested(rl
, BLK_RW_SYNC
);
994 else if (rl
->count
[BLK_RW_SYNC
] < off_thresh
)
995 blk_clear_congested(rl
, BLK_RW_SYNC
);
997 if (rl
->count
[BLK_RW_ASYNC
] >= on_thresh
)
998 blk_set_congested(rl
, BLK_RW_ASYNC
);
999 else if (rl
->count
[BLK_RW_ASYNC
] < off_thresh
)
1000 blk_clear_congested(rl
, BLK_RW_ASYNC
);
1002 if (rl
->count
[BLK_RW_SYNC
] >= q
->nr_requests
) {
1003 blk_set_rl_full(rl
, BLK_RW_SYNC
);
1005 blk_clear_rl_full(rl
, BLK_RW_SYNC
);
1006 wake_up(&rl
->wait
[BLK_RW_SYNC
]);
1009 if (rl
->count
[BLK_RW_ASYNC
] >= q
->nr_requests
) {
1010 blk_set_rl_full(rl
, BLK_RW_ASYNC
);
1012 blk_clear_rl_full(rl
, BLK_RW_ASYNC
);
1013 wake_up(&rl
->wait
[BLK_RW_ASYNC
]);
1017 spin_unlock_irq(q
->queue_lock
);
1022 * Determine if elevator data should be initialized when allocating the
1023 * request associated with @bio.
1025 static bool blk_rq_should_init_elevator(struct bio
*bio
)
1031 * Flush requests do not use the elevator so skip initialization.
1032 * This allows a request to share the flush and elevator data.
1034 if (bio
->bi_opf
& (REQ_PREFLUSH
| REQ_FUA
))
1041 * rq_ioc - determine io_context for request allocation
1042 * @bio: request being allocated is for this bio (can be %NULL)
1044 * Determine io_context to use for request allocation for @bio. May return
1045 * %NULL if %current->io_context doesn't exist.
1047 static struct io_context
*rq_ioc(struct bio
*bio
)
1049 #ifdef CONFIG_BLK_CGROUP
1050 if (bio
&& bio
->bi_ioc
)
1053 return current
->io_context
;
1057 * __get_request - get a free request
1058 * @rl: request list to allocate from
1059 * @op: operation and flags
1060 * @bio: bio to allocate request for (can be %NULL)
1061 * @gfp_mask: allocation mask
1063 * Get a free request from @q. This function may fail under memory
1064 * pressure or if @q is dead.
1066 * Must be called with @q->queue_lock held and,
1067 * Returns ERR_PTR on failure, with @q->queue_lock held.
1068 * Returns request pointer on success, with @q->queue_lock *not held*.
1070 static struct request
*__get_request(struct request_list
*rl
, unsigned int op
,
1071 struct bio
*bio
, gfp_t gfp_mask
)
1073 struct request_queue
*q
= rl
->q
;
1075 struct elevator_type
*et
= q
->elevator
->type
;
1076 struct io_context
*ioc
= rq_ioc(bio
);
1077 struct io_cq
*icq
= NULL
;
1078 const bool is_sync
= op_is_sync(op
);
1080 req_flags_t rq_flags
= RQF_ALLOCED
;
1082 if (unlikely(blk_queue_dying(q
)))
1083 return ERR_PTR(-ENODEV
);
1085 may_queue
= elv_may_queue(q
, op
);
1086 if (may_queue
== ELV_MQUEUE_NO
)
1089 if (rl
->count
[is_sync
]+1 >= queue_congestion_on_threshold(q
)) {
1090 if (rl
->count
[is_sync
]+1 >= q
->nr_requests
) {
1092 * The queue will fill after this allocation, so set
1093 * it as full, and mark this process as "batching".
1094 * This process will be allowed to complete a batch of
1095 * requests, others will be blocked.
1097 if (!blk_rl_full(rl
, is_sync
)) {
1098 ioc_set_batching(q
, ioc
);
1099 blk_set_rl_full(rl
, is_sync
);
1101 if (may_queue
!= ELV_MQUEUE_MUST
1102 && !ioc_batching(q
, ioc
)) {
1104 * The queue is full and the allocating
1105 * process is not a "batcher", and not
1106 * exempted by the IO scheduler
1108 return ERR_PTR(-ENOMEM
);
1112 blk_set_congested(rl
, is_sync
);
1116 * Only allow batching queuers to allocate up to 50% over the defined
1117 * limit of requests, otherwise we could have thousands of requests
1118 * allocated with any setting of ->nr_requests
1120 if (rl
->count
[is_sync
] >= (3 * q
->nr_requests
/ 2))
1121 return ERR_PTR(-ENOMEM
);
1123 q
->nr_rqs
[is_sync
]++;
1124 rl
->count
[is_sync
]++;
1125 rl
->starved
[is_sync
] = 0;
1128 * Decide whether the new request will be managed by elevator. If
1129 * so, mark @rq_flags and increment elvpriv. Non-zero elvpriv will
1130 * prevent the current elevator from being destroyed until the new
1131 * request is freed. This guarantees icq's won't be destroyed and
1132 * makes creating new ones safe.
1134 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1135 * it will be created after releasing queue_lock.
1137 if (blk_rq_should_init_elevator(bio
) && !blk_queue_bypass(q
)) {
1138 rq_flags
|= RQF_ELVPRIV
;
1139 q
->nr_rqs_elvpriv
++;
1140 if (et
->icq_cache
&& ioc
)
1141 icq
= ioc_lookup_icq(ioc
, q
);
1144 if (blk_queue_io_stat(q
))
1145 rq_flags
|= RQF_IO_STAT
;
1146 spin_unlock_irq(q
->queue_lock
);
1148 /* allocate and init request */
1149 rq
= mempool_alloc(rl
->rq_pool
, gfp_mask
);
1154 blk_rq_set_rl(rq
, rl
);
1156 rq
->rq_flags
= rq_flags
;
1159 if (rq_flags
& RQF_ELVPRIV
) {
1160 if (unlikely(et
->icq_cache
&& !icq
)) {
1162 icq
= ioc_create_icq(ioc
, q
, gfp_mask
);
1168 if (unlikely(elv_set_request(q
, rq
, bio
, gfp_mask
)))
1171 /* @rq->elv.icq holds io_context until @rq is freed */
1173 get_io_context(icq
->ioc
);
1177 * ioc may be NULL here, and ioc_batching will be false. That's
1178 * OK, if the queue is under the request limit then requests need
1179 * not count toward the nr_batch_requests limit. There will always
1180 * be some limit enforced by BLK_BATCH_TIME.
1182 if (ioc_batching(q
, ioc
))
1183 ioc
->nr_batch_requests
--;
1185 trace_block_getrq(q
, bio
, op
);
1190 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1191 * and may fail indefinitely under memory pressure and thus
1192 * shouldn't stall IO. Treat this request as !elvpriv. This will
1193 * disturb iosched and blkcg but weird is bettern than dead.
1195 printk_ratelimited(KERN_WARNING
"%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1196 __func__
, dev_name(q
->backing_dev_info
.dev
));
1198 rq
->rq_flags
&= ~RQF_ELVPRIV
;
1201 spin_lock_irq(q
->queue_lock
);
1202 q
->nr_rqs_elvpriv
--;
1203 spin_unlock_irq(q
->queue_lock
);
1208 * Allocation failed presumably due to memory. Undo anything we
1209 * might have messed up.
1211 * Allocating task should really be put onto the front of the wait
1212 * queue, but this is pretty rare.
1214 spin_lock_irq(q
->queue_lock
);
1215 freed_request(rl
, is_sync
, rq_flags
);
1218 * in the very unlikely event that allocation failed and no
1219 * requests for this direction was pending, mark us starved so that
1220 * freeing of a request in the other direction will notice
1221 * us. another possible fix would be to split the rq mempool into
1225 if (unlikely(rl
->count
[is_sync
] == 0))
1226 rl
->starved
[is_sync
] = 1;
1227 return ERR_PTR(-ENOMEM
);
1231 * get_request - get a free request
1232 * @q: request_queue to allocate request from
1233 * @op: operation and flags
1234 * @bio: bio to allocate request for (can be %NULL)
1235 * @gfp_mask: allocation mask
1237 * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask,
1238 * this function keeps retrying under memory pressure and fails iff @q is dead.
1240 * Must be called with @q->queue_lock held and,
1241 * Returns ERR_PTR on failure, with @q->queue_lock held.
1242 * Returns request pointer on success, with @q->queue_lock *not held*.
1244 static struct request
*get_request(struct request_queue
*q
, unsigned int op
,
1245 struct bio
*bio
, gfp_t gfp_mask
)
1247 const bool is_sync
= op_is_sync(op
);
1249 struct request_list
*rl
;
1252 rl
= blk_get_rl(q
, bio
); /* transferred to @rq on success */
1254 rq
= __get_request(rl
, op
, bio
, gfp_mask
);
1258 if (!gfpflags_allow_blocking(gfp_mask
) || unlikely(blk_queue_dying(q
))) {
1263 /* wait on @rl and retry */
1264 prepare_to_wait_exclusive(&rl
->wait
[is_sync
], &wait
,
1265 TASK_UNINTERRUPTIBLE
);
1267 trace_block_sleeprq(q
, bio
, op
);
1269 spin_unlock_irq(q
->queue_lock
);
1273 * After sleeping, we become a "batching" process and will be able
1274 * to allocate at least one request, and up to a big batch of them
1275 * for a small period time. See ioc_batching, ioc_set_batching
1277 ioc_set_batching(q
, current
->io_context
);
1279 spin_lock_irq(q
->queue_lock
);
1280 finish_wait(&rl
->wait
[is_sync
], &wait
);
1285 static struct request
*blk_old_get_request(struct request_queue
*q
, int rw
,
1290 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
1292 /* create ioc upfront */
1293 create_io_context(gfp_mask
, q
->node
);
1295 spin_lock_irq(q
->queue_lock
);
1296 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
1298 spin_unlock_irq(q
->queue_lock
);
1302 /* q->queue_lock is unlocked at this point */
1304 rq
->__sector
= (sector_t
) -1;
1305 rq
->bio
= rq
->biotail
= NULL
;
1309 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
1312 return blk_mq_alloc_request(q
, rw
,
1313 (gfp_mask
& __GFP_DIRECT_RECLAIM
) ?
1314 0 : BLK_MQ_REQ_NOWAIT
);
1316 return blk_old_get_request(q
, rw
, gfp_mask
);
1318 EXPORT_SYMBOL(blk_get_request
);
1321 * blk_rq_set_block_pc - initialize a request to type BLOCK_PC
1322 * @rq: request to be initialized
1325 void blk_rq_set_block_pc(struct request
*rq
)
1327 rq
->cmd_type
= REQ_TYPE_BLOCK_PC
;
1328 memset(rq
->__cmd
, 0, sizeof(rq
->__cmd
));
1330 EXPORT_SYMBOL(blk_rq_set_block_pc
);
1333 * blk_requeue_request - put a request back on queue
1334 * @q: request queue where request should be inserted
1335 * @rq: request to be inserted
1338 * Drivers often keep queueing requests until the hardware cannot accept
1339 * more, when that condition happens we need to put the request back
1340 * on the queue. Must be called with queue lock held.
1342 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
1344 blk_delete_timer(rq
);
1345 blk_clear_rq_complete(rq
);
1346 trace_block_rq_requeue(q
, rq
);
1348 if (rq
->rq_flags
& RQF_QUEUED
)
1349 blk_queue_end_tag(q
, rq
);
1351 BUG_ON(blk_queued_rq(rq
));
1353 elv_requeue_request(q
, rq
);
1355 EXPORT_SYMBOL(blk_requeue_request
);
1357 static void add_acct_request(struct request_queue
*q
, struct request
*rq
,
1360 blk_account_io_start(rq
, true);
1361 __elv_add_request(q
, rq
, where
);
1364 static void part_round_stats_single(int cpu
, struct hd_struct
*part
,
1369 if (now
== part
->stamp
)
1372 inflight
= part_in_flight(part
);
1374 __part_stat_add(cpu
, part
, time_in_queue
,
1375 inflight
* (now
- part
->stamp
));
1376 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
1382 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1383 * @cpu: cpu number for stats access
1384 * @part: target partition
1386 * The average IO queue length and utilisation statistics are maintained
1387 * by observing the current state of the queue length and the amount of
1388 * time it has been in this state for.
1390 * Normally, that accounting is done on IO completion, but that can result
1391 * in more than a second's worth of IO being accounted for within any one
1392 * second, leading to >100% utilisation. To deal with that, we call this
1393 * function to do a round-off before returning the results when reading
1394 * /proc/diskstats. This accounts immediately for all queue usage up to
1395 * the current jiffies and restarts the counters again.
1397 void part_round_stats(int cpu
, struct hd_struct
*part
)
1399 unsigned long now
= jiffies
;
1402 part_round_stats_single(cpu
, &part_to_disk(part
)->part0
, now
);
1403 part_round_stats_single(cpu
, part
, now
);
1405 EXPORT_SYMBOL_GPL(part_round_stats
);
1408 static void blk_pm_put_request(struct request
*rq
)
1410 if (rq
->q
->dev
&& !(rq
->rq_flags
& RQF_PM
) && !--rq
->q
->nr_pending
)
1411 pm_runtime_mark_last_busy(rq
->q
->dev
);
1414 static inline void blk_pm_put_request(struct request
*rq
) {}
1418 * queue lock must be held
1420 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1422 req_flags_t rq_flags
= req
->rq_flags
;
1428 blk_mq_free_request(req
);
1432 blk_pm_put_request(req
);
1434 elv_completed_request(q
, req
);
1436 /* this is a bio leak */
1437 WARN_ON(req
->bio
!= NULL
);
1440 * Request may not have originated from ll_rw_blk. if not,
1441 * it didn't come out of our reserved rq pools
1443 if (rq_flags
& RQF_ALLOCED
) {
1444 struct request_list
*rl
= blk_rq_rl(req
);
1445 bool sync
= op_is_sync(req
->cmd_flags
);
1447 BUG_ON(!list_empty(&req
->queuelist
));
1448 BUG_ON(ELV_ON_HASH(req
));
1450 blk_free_request(rl
, req
);
1451 freed_request(rl
, sync
, rq_flags
);
1455 EXPORT_SYMBOL_GPL(__blk_put_request
);
1457 void blk_put_request(struct request
*req
)
1459 struct request_queue
*q
= req
->q
;
1462 blk_mq_free_request(req
);
1464 unsigned long flags
;
1466 spin_lock_irqsave(q
->queue_lock
, flags
);
1467 __blk_put_request(q
, req
);
1468 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1471 EXPORT_SYMBOL(blk_put_request
);
1474 * blk_add_request_payload - add a payload to a request
1475 * @rq: request to update
1476 * @page: page backing the payload
1477 * @offset: offset in page
1478 * @len: length of the payload.
1480 * This allows to later add a payload to an already submitted request by
1481 * a block driver. The driver needs to take care of freeing the payload
1484 * Note that this is a quite horrible hack and nothing but handling of
1485 * discard requests should ever use it.
1487 void blk_add_request_payload(struct request
*rq
, struct page
*page
,
1488 int offset
, unsigned int len
)
1490 struct bio
*bio
= rq
->bio
;
1492 bio
->bi_io_vec
->bv_page
= page
;
1493 bio
->bi_io_vec
->bv_offset
= offset
;
1494 bio
->bi_io_vec
->bv_len
= len
;
1496 bio
->bi_iter
.bi_size
= len
;
1498 bio
->bi_phys_segments
= 1;
1500 rq
->__data_len
= rq
->resid_len
= len
;
1501 rq
->nr_phys_segments
= 1;
1503 EXPORT_SYMBOL_GPL(blk_add_request_payload
);
1505 bool bio_attempt_back_merge(struct request_queue
*q
, struct request
*req
,
1508 const int ff
= bio
->bi_opf
& REQ_FAILFAST_MASK
;
1510 if (!ll_back_merge_fn(q
, req
, bio
))
1513 trace_block_bio_backmerge(q
, req
, bio
);
1515 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1516 blk_rq_set_mixed_merge(req
);
1518 req
->biotail
->bi_next
= bio
;
1520 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1521 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1523 blk_account_io_start(req
, false);
1527 bool bio_attempt_front_merge(struct request_queue
*q
, struct request
*req
,
1530 const int ff
= bio
->bi_opf
& REQ_FAILFAST_MASK
;
1532 if (!ll_front_merge_fn(q
, req
, bio
))
1535 trace_block_bio_frontmerge(q
, req
, bio
);
1537 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1538 blk_rq_set_mixed_merge(req
);
1540 bio
->bi_next
= req
->bio
;
1543 req
->__sector
= bio
->bi_iter
.bi_sector
;
1544 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1545 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1547 blk_account_io_start(req
, false);
1552 * blk_attempt_plug_merge - try to merge with %current's plugged list
1553 * @q: request_queue new bio is being queued at
1554 * @bio: new bio being queued
1555 * @request_count: out parameter for number of traversed plugged requests
1556 * @same_queue_rq: pointer to &struct request that gets filled in when
1557 * another request associated with @q is found on the plug list
1558 * (optional, may be %NULL)
1560 * Determine whether @bio being queued on @q can be merged with a request
1561 * on %current's plugged list. Returns %true if merge was successful,
1564 * Plugging coalesces IOs from the same issuer for the same purpose without
1565 * going through @q->queue_lock. As such it's more of an issuing mechanism
1566 * than scheduling, and the request, while may have elvpriv data, is not
1567 * added on the elevator at this point. In addition, we don't have
1568 * reliable access to the elevator outside queue lock. Only check basic
1569 * merging parameters without querying the elevator.
1571 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1573 bool blk_attempt_plug_merge(struct request_queue
*q
, struct bio
*bio
,
1574 unsigned int *request_count
,
1575 struct request
**same_queue_rq
)
1577 struct blk_plug
*plug
;
1580 struct list_head
*plug_list
;
1582 plug
= current
->plug
;
1588 plug_list
= &plug
->mq_list
;
1590 plug_list
= &plug
->list
;
1592 list_for_each_entry_reverse(rq
, plug_list
, queuelist
) {
1598 * Only blk-mq multiple hardware queues case checks the
1599 * rq in the same queue, there should be only one such
1603 *same_queue_rq
= rq
;
1606 if (rq
->q
!= q
|| !blk_rq_merge_ok(rq
, bio
))
1609 el_ret
= blk_try_merge(rq
, bio
);
1610 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1611 ret
= bio_attempt_back_merge(q
, rq
, bio
);
1614 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1615 ret
= bio_attempt_front_merge(q
, rq
, bio
);
1624 unsigned int blk_plug_queued_count(struct request_queue
*q
)
1626 struct blk_plug
*plug
;
1628 struct list_head
*plug_list
;
1629 unsigned int ret
= 0;
1631 plug
= current
->plug
;
1636 plug_list
= &plug
->mq_list
;
1638 plug_list
= &plug
->list
;
1640 list_for_each_entry(rq
, plug_list
, queuelist
) {
1648 void init_request_from_bio(struct request
*req
, struct bio
*bio
)
1650 req
->cmd_type
= REQ_TYPE_FS
;
1651 if (bio
->bi_opf
& REQ_RAHEAD
)
1652 req
->cmd_flags
|= REQ_FAILFAST_MASK
;
1655 req
->__sector
= bio
->bi_iter
.bi_sector
;
1656 req
->ioprio
= bio_prio(bio
);
1657 blk_rq_bio_prep(req
->q
, req
, bio
);
1660 static blk_qc_t
blk_queue_bio(struct request_queue
*q
, struct bio
*bio
)
1662 struct blk_plug
*plug
;
1663 int el_ret
, where
= ELEVATOR_INSERT_SORT
;
1664 struct request
*req
;
1665 unsigned int request_count
= 0;
1668 * low level driver can indicate that it wants pages above a
1669 * certain limit bounced to low memory (ie for highmem, or even
1670 * ISA dma in theory)
1672 blk_queue_bounce(q
, &bio
);
1674 blk_queue_split(q
, &bio
, q
->bio_split
);
1676 if (bio_integrity_enabled(bio
) && bio_integrity_prep(bio
)) {
1677 bio
->bi_error
= -EIO
;
1679 return BLK_QC_T_NONE
;
1682 if (bio
->bi_opf
& (REQ_PREFLUSH
| REQ_FUA
)) {
1683 spin_lock_irq(q
->queue_lock
);
1684 where
= ELEVATOR_INSERT_FLUSH
;
1689 * Check if we can merge with the plugged list before grabbing
1692 if (!blk_queue_nomerges(q
)) {
1693 if (blk_attempt_plug_merge(q
, bio
, &request_count
, NULL
))
1694 return BLK_QC_T_NONE
;
1696 request_count
= blk_plug_queued_count(q
);
1698 spin_lock_irq(q
->queue_lock
);
1700 el_ret
= elv_merge(q
, &req
, bio
);
1701 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1702 if (bio_attempt_back_merge(q
, req
, bio
)) {
1703 elv_bio_merged(q
, req
, bio
);
1704 if (!attempt_back_merge(q
, req
))
1705 elv_merged_request(q
, req
, el_ret
);
1708 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1709 if (bio_attempt_front_merge(q
, req
, bio
)) {
1710 elv_bio_merged(q
, req
, bio
);
1711 if (!attempt_front_merge(q
, req
))
1712 elv_merged_request(q
, req
, el_ret
);
1719 * Grab a free request. This is might sleep but can not fail.
1720 * Returns with the queue unlocked.
1722 req
= get_request(q
, bio
->bi_opf
, bio
, GFP_NOIO
);
1724 bio
->bi_error
= PTR_ERR(req
);
1730 * After dropping the lock and possibly sleeping here, our request
1731 * may now be mergeable after it had proven unmergeable (above).
1732 * We don't worry about that case for efficiency. It won't happen
1733 * often, and the elevators are able to handle it.
1735 init_request_from_bio(req
, bio
);
1737 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
))
1738 req
->cpu
= raw_smp_processor_id();
1740 plug
= current
->plug
;
1743 * If this is the first request added after a plug, fire
1747 trace_block_plug(q
);
1749 struct request
*last
= list_entry_rq(plug
->list
.prev
);
1750 if (request_count
>= BLK_MAX_REQUEST_COUNT
||
1751 blk_rq_bytes(last
) >= BLK_PLUG_FLUSH_SIZE
) {
1752 blk_flush_plug_list(plug
, false);
1753 trace_block_plug(q
);
1756 list_add_tail(&req
->queuelist
, &plug
->list
);
1757 blk_account_io_start(req
, true);
1759 spin_lock_irq(q
->queue_lock
);
1760 add_acct_request(q
, req
, where
);
1763 spin_unlock_irq(q
->queue_lock
);
1766 return BLK_QC_T_NONE
;
1770 * If bio->bi_dev is a partition, remap the location
1772 static inline void blk_partition_remap(struct bio
*bio
)
1774 struct block_device
*bdev
= bio
->bi_bdev
;
1776 if (bio_sectors(bio
) && bdev
!= bdev
->bd_contains
) {
1777 struct hd_struct
*p
= bdev
->bd_part
;
1779 bio
->bi_iter
.bi_sector
+= p
->start_sect
;
1780 bio
->bi_bdev
= bdev
->bd_contains
;
1782 trace_block_bio_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
1784 bio
->bi_iter
.bi_sector
- p
->start_sect
);
1788 static void handle_bad_sector(struct bio
*bio
)
1790 char b
[BDEVNAME_SIZE
];
1792 printk(KERN_INFO
"attempt to access beyond end of device\n");
1793 printk(KERN_INFO
"%s: rw=%d, want=%Lu, limit=%Lu\n",
1794 bdevname(bio
->bi_bdev
, b
),
1796 (unsigned long long)bio_end_sector(bio
),
1797 (long long)(i_size_read(bio
->bi_bdev
->bd_inode
) >> 9));
1800 #ifdef CONFIG_FAIL_MAKE_REQUEST
1802 static DECLARE_FAULT_ATTR(fail_make_request
);
1804 static int __init
setup_fail_make_request(char *str
)
1806 return setup_fault_attr(&fail_make_request
, str
);
1808 __setup("fail_make_request=", setup_fail_make_request
);
1810 static bool should_fail_request(struct hd_struct
*part
, unsigned int bytes
)
1812 return part
->make_it_fail
&& should_fail(&fail_make_request
, bytes
);
1815 static int __init
fail_make_request_debugfs(void)
1817 struct dentry
*dir
= fault_create_debugfs_attr("fail_make_request",
1818 NULL
, &fail_make_request
);
1820 return PTR_ERR_OR_ZERO(dir
);
1823 late_initcall(fail_make_request_debugfs
);
1825 #else /* CONFIG_FAIL_MAKE_REQUEST */
1827 static inline bool should_fail_request(struct hd_struct
*part
,
1833 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1836 * Check whether this bio extends beyond the end of the device.
1838 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
1845 /* Test device or partition size, when known. */
1846 maxsector
= i_size_read(bio
->bi_bdev
->bd_inode
) >> 9;
1848 sector_t sector
= bio
->bi_iter
.bi_sector
;
1850 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
1852 * This may well happen - the kernel calls bread()
1853 * without checking the size of the device, e.g., when
1854 * mounting a device.
1856 handle_bad_sector(bio
);
1864 static noinline_for_stack
bool
1865 generic_make_request_checks(struct bio
*bio
)
1867 struct request_queue
*q
;
1868 int nr_sectors
= bio_sectors(bio
);
1870 char b
[BDEVNAME_SIZE
];
1871 struct hd_struct
*part
;
1875 if (bio_check_eod(bio
, nr_sectors
))
1878 q
= bdev_get_queue(bio
->bi_bdev
);
1881 "generic_make_request: Trying to access "
1882 "nonexistent block-device %s (%Lu)\n",
1883 bdevname(bio
->bi_bdev
, b
),
1884 (long long) bio
->bi_iter
.bi_sector
);
1888 part
= bio
->bi_bdev
->bd_part
;
1889 if (should_fail_request(part
, bio
->bi_iter
.bi_size
) ||
1890 should_fail_request(&part_to_disk(part
)->part0
,
1891 bio
->bi_iter
.bi_size
))
1895 * If this device has partitions, remap block n
1896 * of partition p to block n+start(p) of the disk.
1898 blk_partition_remap(bio
);
1900 if (bio_check_eod(bio
, nr_sectors
))
1904 * Filter flush bio's early so that make_request based
1905 * drivers without flush support don't have to worry
1908 if ((bio
->bi_opf
& (REQ_PREFLUSH
| REQ_FUA
)) &&
1909 !test_bit(QUEUE_FLAG_WC
, &q
->queue_flags
)) {
1910 bio
->bi_opf
&= ~(REQ_PREFLUSH
| REQ_FUA
);
1917 switch (bio_op(bio
)) {
1918 case REQ_OP_DISCARD
:
1919 if (!blk_queue_discard(q
))
1922 case REQ_OP_SECURE_ERASE
:
1923 if (!blk_queue_secure_erase(q
))
1926 case REQ_OP_WRITE_SAME
:
1927 if (!bdev_write_same(bio
->bi_bdev
))
1929 case REQ_OP_ZONE_REPORT
:
1930 case REQ_OP_ZONE_RESET
:
1931 if (!bdev_is_zoned(bio
->bi_bdev
))
1939 * Various block parts want %current->io_context and lazy ioc
1940 * allocation ends up trading a lot of pain for a small amount of
1941 * memory. Just allocate it upfront. This may fail and block
1942 * layer knows how to live with it.
1944 create_io_context(GFP_ATOMIC
, q
->node
);
1946 if (!blkcg_bio_issue_check(q
, bio
))
1949 trace_block_bio_queue(q
, bio
);
1955 bio
->bi_error
= err
;
1961 * generic_make_request - hand a buffer to its device driver for I/O
1962 * @bio: The bio describing the location in memory and on the device.
1964 * generic_make_request() is used to make I/O requests of block
1965 * devices. It is passed a &struct bio, which describes the I/O that needs
1968 * generic_make_request() does not return any status. The
1969 * success/failure status of the request, along with notification of
1970 * completion, is delivered asynchronously through the bio->bi_end_io
1971 * function described (one day) else where.
1973 * The caller of generic_make_request must make sure that bi_io_vec
1974 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1975 * set to describe the device address, and the
1976 * bi_end_io and optionally bi_private are set to describe how
1977 * completion notification should be signaled.
1979 * generic_make_request and the drivers it calls may use bi_next if this
1980 * bio happens to be merged with someone else, and may resubmit the bio to
1981 * a lower device by calling into generic_make_request recursively, which
1982 * means the bio should NOT be touched after the call to ->make_request_fn.
1984 blk_qc_t
generic_make_request(struct bio
*bio
)
1986 struct bio_list bio_list_on_stack
;
1987 blk_qc_t ret
= BLK_QC_T_NONE
;
1989 if (!generic_make_request_checks(bio
))
1993 * We only want one ->make_request_fn to be active at a time, else
1994 * stack usage with stacked devices could be a problem. So use
1995 * current->bio_list to keep a list of requests submited by a
1996 * make_request_fn function. current->bio_list is also used as a
1997 * flag to say if generic_make_request is currently active in this
1998 * task or not. If it is NULL, then no make_request is active. If
1999 * it is non-NULL, then a make_request is active, and new requests
2000 * should be added at the tail
2002 if (current
->bio_list
) {
2003 bio_list_add(current
->bio_list
, bio
);
2007 /* following loop may be a bit non-obvious, and so deserves some
2009 * Before entering the loop, bio->bi_next is NULL (as all callers
2010 * ensure that) so we have a list with a single bio.
2011 * We pretend that we have just taken it off a longer list, so
2012 * we assign bio_list to a pointer to the bio_list_on_stack,
2013 * thus initialising the bio_list of new bios to be
2014 * added. ->make_request() may indeed add some more bios
2015 * through a recursive call to generic_make_request. If it
2016 * did, we find a non-NULL value in bio_list and re-enter the loop
2017 * from the top. In this case we really did just take the bio
2018 * of the top of the list (no pretending) and so remove it from
2019 * bio_list, and call into ->make_request() again.
2021 BUG_ON(bio
->bi_next
);
2022 bio_list_init(&bio_list_on_stack
);
2023 current
->bio_list
= &bio_list_on_stack
;
2025 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
2027 if (likely(blk_queue_enter(q
, false) == 0)) {
2028 ret
= q
->make_request_fn(q
, bio
);
2032 bio
= bio_list_pop(current
->bio_list
);
2034 struct bio
*bio_next
= bio_list_pop(current
->bio_list
);
2040 current
->bio_list
= NULL
; /* deactivate */
2045 EXPORT_SYMBOL(generic_make_request
);
2048 * submit_bio - submit a bio to the block device layer for I/O
2049 * @bio: The &struct bio which describes the I/O
2051 * submit_bio() is very similar in purpose to generic_make_request(), and
2052 * uses that function to do most of the work. Both are fairly rough
2053 * interfaces; @bio must be presetup and ready for I/O.
2056 blk_qc_t
submit_bio(struct bio
*bio
)
2059 * If it's a regular read/write or a barrier with data attached,
2060 * go through the normal accounting stuff before submission.
2062 if (bio_has_data(bio
)) {
2065 if (unlikely(bio_op(bio
) == REQ_OP_WRITE_SAME
))
2066 count
= bdev_logical_block_size(bio
->bi_bdev
) >> 9;
2068 count
= bio_sectors(bio
);
2070 if (op_is_write(bio_op(bio
))) {
2071 count_vm_events(PGPGOUT
, count
);
2073 task_io_account_read(bio
->bi_iter
.bi_size
);
2074 count_vm_events(PGPGIN
, count
);
2077 if (unlikely(block_dump
)) {
2078 char b
[BDEVNAME_SIZE
];
2079 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s (%u sectors)\n",
2080 current
->comm
, task_pid_nr(current
),
2081 op_is_write(bio_op(bio
)) ? "WRITE" : "READ",
2082 (unsigned long long)bio
->bi_iter
.bi_sector
,
2083 bdevname(bio
->bi_bdev
, b
),
2088 return generic_make_request(bio
);
2090 EXPORT_SYMBOL(submit_bio
);
2093 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2094 * for new the queue limits
2096 * @rq: the request being checked
2099 * @rq may have been made based on weaker limitations of upper-level queues
2100 * in request stacking drivers, and it may violate the limitation of @q.
2101 * Since the block layer and the underlying device driver trust @rq
2102 * after it is inserted to @q, it should be checked against @q before
2103 * the insertion using this generic function.
2105 * Request stacking drivers like request-based dm may change the queue
2106 * limits when retrying requests on other queues. Those requests need
2107 * to be checked against the new queue limits again during dispatch.
2109 static int blk_cloned_rq_check_limits(struct request_queue
*q
,
2112 if (blk_rq_sectors(rq
) > blk_queue_get_max_sectors(q
, req_op(rq
))) {
2113 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
2118 * queue's settings related to segment counting like q->bounce_pfn
2119 * may differ from that of other stacking queues.
2120 * Recalculate it to check the request correctly on this queue's
2123 blk_recalc_rq_segments(rq
);
2124 if (rq
->nr_phys_segments
> queue_max_segments(q
)) {
2125 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
2133 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2134 * @q: the queue to submit the request
2135 * @rq: the request being queued
2137 int blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
2139 unsigned long flags
;
2140 int where
= ELEVATOR_INSERT_BACK
;
2142 if (blk_cloned_rq_check_limits(q
, rq
))
2146 should_fail_request(&rq
->rq_disk
->part0
, blk_rq_bytes(rq
)))
2150 if (blk_queue_io_stat(q
))
2151 blk_account_io_start(rq
, true);
2152 blk_mq_insert_request(rq
, false, true, false);
2156 spin_lock_irqsave(q
->queue_lock
, flags
);
2157 if (unlikely(blk_queue_dying(q
))) {
2158 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2163 * Submitting request must be dequeued before calling this function
2164 * because it will be linked to another request_queue
2166 BUG_ON(blk_queued_rq(rq
));
2168 if (rq
->cmd_flags
& (REQ_PREFLUSH
| REQ_FUA
))
2169 where
= ELEVATOR_INSERT_FLUSH
;
2171 add_acct_request(q
, rq
, where
);
2172 if (where
== ELEVATOR_INSERT_FLUSH
)
2174 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2178 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
2181 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2182 * @rq: request to examine
2185 * A request could be merge of IOs which require different failure
2186 * handling. This function determines the number of bytes which
2187 * can be failed from the beginning of the request without
2188 * crossing into area which need to be retried further.
2191 * The number of bytes to fail.
2194 * queue_lock must be held.
2196 unsigned int blk_rq_err_bytes(const struct request
*rq
)
2198 unsigned int ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
2199 unsigned int bytes
= 0;
2202 if (!(rq
->rq_flags
& RQF_MIXED_MERGE
))
2203 return blk_rq_bytes(rq
);
2206 * Currently the only 'mixing' which can happen is between
2207 * different fastfail types. We can safely fail portions
2208 * which have all the failfast bits that the first one has -
2209 * the ones which are at least as eager to fail as the first
2212 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
2213 if ((bio
->bi_opf
& ff
) != ff
)
2215 bytes
+= bio
->bi_iter
.bi_size
;
2218 /* this could lead to infinite loop */
2219 BUG_ON(blk_rq_bytes(rq
) && !bytes
);
2222 EXPORT_SYMBOL_GPL(blk_rq_err_bytes
);
2224 void blk_account_io_completion(struct request
*req
, unsigned int bytes
)
2226 if (blk_do_io_stat(req
)) {
2227 const int rw
= rq_data_dir(req
);
2228 struct hd_struct
*part
;
2231 cpu
= part_stat_lock();
2233 part_stat_add(cpu
, part
, sectors
[rw
], bytes
>> 9);
2238 void blk_account_io_done(struct request
*req
)
2241 * Account IO completion. flush_rq isn't accounted as a
2242 * normal IO on queueing nor completion. Accounting the
2243 * containing request is enough.
2245 if (blk_do_io_stat(req
) && !(req
->rq_flags
& RQF_FLUSH_SEQ
)) {
2246 unsigned long duration
= jiffies
- req
->start_time
;
2247 const int rw
= rq_data_dir(req
);
2248 struct hd_struct
*part
;
2251 cpu
= part_stat_lock();
2254 part_stat_inc(cpu
, part
, ios
[rw
]);
2255 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
2256 part_round_stats(cpu
, part
);
2257 part_dec_in_flight(part
, rw
);
2259 hd_struct_put(part
);
2266 * Don't process normal requests when queue is suspended
2267 * or in the process of suspending/resuming
2269 static struct request
*blk_pm_peek_request(struct request_queue
*q
,
2272 if (q
->dev
&& (q
->rpm_status
== RPM_SUSPENDED
||
2273 (q
->rpm_status
!= RPM_ACTIVE
&& !(rq
->rq_flags
& RQF_PM
))))
2279 static inline struct request
*blk_pm_peek_request(struct request_queue
*q
,
2286 void blk_account_io_start(struct request
*rq
, bool new_io
)
2288 struct hd_struct
*part
;
2289 int rw
= rq_data_dir(rq
);
2292 if (!blk_do_io_stat(rq
))
2295 cpu
= part_stat_lock();
2299 part_stat_inc(cpu
, part
, merges
[rw
]);
2301 part
= disk_map_sector_rcu(rq
->rq_disk
, blk_rq_pos(rq
));
2302 if (!hd_struct_try_get(part
)) {
2304 * The partition is already being removed,
2305 * the request will be accounted on the disk only
2307 * We take a reference on disk->part0 although that
2308 * partition will never be deleted, so we can treat
2309 * it as any other partition.
2311 part
= &rq
->rq_disk
->part0
;
2312 hd_struct_get(part
);
2314 part_round_stats(cpu
, part
);
2315 part_inc_in_flight(part
, rw
);
2323 * blk_peek_request - peek at the top of a request queue
2324 * @q: request queue to peek at
2327 * Return the request at the top of @q. The returned request
2328 * should be started using blk_start_request() before LLD starts
2332 * Pointer to the request at the top of @q if available. Null
2336 * queue_lock must be held.
2338 struct request
*blk_peek_request(struct request_queue
*q
)
2343 while ((rq
= __elv_next_request(q
)) != NULL
) {
2345 rq
= blk_pm_peek_request(q
, rq
);
2349 if (!(rq
->rq_flags
& RQF_STARTED
)) {
2351 * This is the first time the device driver
2352 * sees this request (possibly after
2353 * requeueing). Notify IO scheduler.
2355 if (rq
->rq_flags
& RQF_SORTED
)
2356 elv_activate_rq(q
, rq
);
2359 * just mark as started even if we don't start
2360 * it, a request that has been delayed should
2361 * not be passed by new incoming requests
2363 rq
->rq_flags
|= RQF_STARTED
;
2364 trace_block_rq_issue(q
, rq
);
2367 if (!q
->boundary_rq
|| q
->boundary_rq
== rq
) {
2368 q
->end_sector
= rq_end_sector(rq
);
2369 q
->boundary_rq
= NULL
;
2372 if (rq
->rq_flags
& RQF_DONTPREP
)
2375 if (q
->dma_drain_size
&& blk_rq_bytes(rq
)) {
2377 * make sure space for the drain appears we
2378 * know we can do this because max_hw_segments
2379 * has been adjusted to be one fewer than the
2382 rq
->nr_phys_segments
++;
2388 ret
= q
->prep_rq_fn(q
, rq
);
2389 if (ret
== BLKPREP_OK
) {
2391 } else if (ret
== BLKPREP_DEFER
) {
2393 * the request may have been (partially) prepped.
2394 * we need to keep this request in the front to
2395 * avoid resource deadlock. RQF_STARTED will
2396 * prevent other fs requests from passing this one.
2398 if (q
->dma_drain_size
&& blk_rq_bytes(rq
) &&
2399 !(rq
->rq_flags
& RQF_DONTPREP
)) {
2401 * remove the space for the drain we added
2402 * so that we don't add it again
2404 --rq
->nr_phys_segments
;
2409 } else if (ret
== BLKPREP_KILL
|| ret
== BLKPREP_INVALID
) {
2410 int err
= (ret
== BLKPREP_INVALID
) ? -EREMOTEIO
: -EIO
;
2412 rq
->rq_flags
|= RQF_QUIET
;
2414 * Mark this request as started so we don't trigger
2415 * any debug logic in the end I/O path.
2417 blk_start_request(rq
);
2418 __blk_end_request_all(rq
, err
);
2420 printk(KERN_ERR
"%s: bad return=%d\n", __func__
, ret
);
2427 EXPORT_SYMBOL(blk_peek_request
);
2429 void blk_dequeue_request(struct request
*rq
)
2431 struct request_queue
*q
= rq
->q
;
2433 BUG_ON(list_empty(&rq
->queuelist
));
2434 BUG_ON(ELV_ON_HASH(rq
));
2436 list_del_init(&rq
->queuelist
);
2439 * the time frame between a request being removed from the lists
2440 * and to it is freed is accounted as io that is in progress at
2443 if (blk_account_rq(rq
)) {
2444 q
->in_flight
[rq_is_sync(rq
)]++;
2445 set_io_start_time_ns(rq
);
2450 * blk_start_request - start request processing on the driver
2451 * @req: request to dequeue
2454 * Dequeue @req and start timeout timer on it. This hands off the
2455 * request to the driver.
2457 * Block internal functions which don't want to start timer should
2458 * call blk_dequeue_request().
2461 * queue_lock must be held.
2463 void blk_start_request(struct request
*req
)
2465 blk_dequeue_request(req
);
2467 if (test_bit(QUEUE_FLAG_STATS
, &req
->q
->queue_flags
)) {
2468 blk_stat_set_issue_time(&req
->issue_stat
);
2469 req
->rq_flags
|= RQF_STATS
;
2473 * We are now handing the request to the hardware, initialize
2474 * resid_len to full count and add the timeout handler.
2476 req
->resid_len
= blk_rq_bytes(req
);
2477 if (unlikely(blk_bidi_rq(req
)))
2478 req
->next_rq
->resid_len
= blk_rq_bytes(req
->next_rq
);
2480 BUG_ON(test_bit(REQ_ATOM_COMPLETE
, &req
->atomic_flags
));
2483 EXPORT_SYMBOL(blk_start_request
);
2486 * blk_fetch_request - fetch a request from a request queue
2487 * @q: request queue to fetch a request from
2490 * Return the request at the top of @q. The request is started on
2491 * return and LLD can start processing it immediately.
2494 * Pointer to the request at the top of @q if available. Null
2498 * queue_lock must be held.
2500 struct request
*blk_fetch_request(struct request_queue
*q
)
2504 rq
= blk_peek_request(q
);
2506 blk_start_request(rq
);
2509 EXPORT_SYMBOL(blk_fetch_request
);
2512 * blk_update_request - Special helper function for request stacking drivers
2513 * @req: the request being processed
2514 * @error: %0 for success, < %0 for error
2515 * @nr_bytes: number of bytes to complete @req
2518 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2519 * the request structure even if @req doesn't have leftover.
2520 * If @req has leftover, sets it up for the next range of segments.
2522 * This special helper function is only for request stacking drivers
2523 * (e.g. request-based dm) so that they can handle partial completion.
2524 * Actual device drivers should use blk_end_request instead.
2526 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2527 * %false return from this function.
2530 * %false - this request doesn't have any more data
2531 * %true - this request has more data
2533 bool blk_update_request(struct request
*req
, int error
, unsigned int nr_bytes
)
2537 trace_block_rq_complete(req
->q
, req
, nr_bytes
);
2543 * For fs requests, rq is just carrier of independent bio's
2544 * and each partial completion should be handled separately.
2545 * Reset per-request error on each partial completion.
2547 * TODO: tj: This is too subtle. It would be better to let
2548 * low level drivers do what they see fit.
2550 if (req
->cmd_type
== REQ_TYPE_FS
)
2553 if (error
&& req
->cmd_type
== REQ_TYPE_FS
&&
2554 !(req
->rq_flags
& RQF_QUIET
)) {
2559 error_type
= "recoverable transport";
2562 error_type
= "critical target";
2565 error_type
= "critical nexus";
2568 error_type
= "timeout";
2571 error_type
= "critical space allocation";
2574 error_type
= "critical medium";
2581 printk_ratelimited(KERN_ERR
"%s: %s error, dev %s, sector %llu\n",
2582 __func__
, error_type
, req
->rq_disk
?
2583 req
->rq_disk
->disk_name
: "?",
2584 (unsigned long long)blk_rq_pos(req
));
2588 blk_account_io_completion(req
, nr_bytes
);
2592 struct bio
*bio
= req
->bio
;
2593 unsigned bio_bytes
= min(bio
->bi_iter
.bi_size
, nr_bytes
);
2595 if (bio_bytes
== bio
->bi_iter
.bi_size
)
2596 req
->bio
= bio
->bi_next
;
2598 req_bio_endio(req
, bio
, bio_bytes
, error
);
2600 total_bytes
+= bio_bytes
;
2601 nr_bytes
-= bio_bytes
;
2612 * Reset counters so that the request stacking driver
2613 * can find how many bytes remain in the request
2616 req
->__data_len
= 0;
2620 req
->__data_len
-= total_bytes
;
2622 /* update sector only for requests with clear definition of sector */
2623 if (req
->cmd_type
== REQ_TYPE_FS
)
2624 req
->__sector
+= total_bytes
>> 9;
2626 /* mixed attributes always follow the first bio */
2627 if (req
->rq_flags
& RQF_MIXED_MERGE
) {
2628 req
->cmd_flags
&= ~REQ_FAILFAST_MASK
;
2629 req
->cmd_flags
|= req
->bio
->bi_opf
& REQ_FAILFAST_MASK
;
2633 * If total number of sectors is less than the first segment
2634 * size, something has gone terribly wrong.
2636 if (blk_rq_bytes(req
) < blk_rq_cur_bytes(req
)) {
2637 blk_dump_rq_flags(req
, "request botched");
2638 req
->__data_len
= blk_rq_cur_bytes(req
);
2641 /* recalculate the number of segments */
2642 blk_recalc_rq_segments(req
);
2646 EXPORT_SYMBOL_GPL(blk_update_request
);
2648 static bool blk_update_bidi_request(struct request
*rq
, int error
,
2649 unsigned int nr_bytes
,
2650 unsigned int bidi_bytes
)
2652 if (blk_update_request(rq
, error
, nr_bytes
))
2655 /* Bidi request must be completed as a whole */
2656 if (unlikely(blk_bidi_rq(rq
)) &&
2657 blk_update_request(rq
->next_rq
, error
, bidi_bytes
))
2660 if (blk_queue_add_random(rq
->q
))
2661 add_disk_randomness(rq
->rq_disk
);
2667 * blk_unprep_request - unprepare a request
2670 * This function makes a request ready for complete resubmission (or
2671 * completion). It happens only after all error handling is complete,
2672 * so represents the appropriate moment to deallocate any resources
2673 * that were allocated to the request in the prep_rq_fn. The queue
2674 * lock is held when calling this.
2676 void blk_unprep_request(struct request
*req
)
2678 struct request_queue
*q
= req
->q
;
2680 req
->rq_flags
&= ~RQF_DONTPREP
;
2681 if (q
->unprep_rq_fn
)
2682 q
->unprep_rq_fn(q
, req
);
2684 EXPORT_SYMBOL_GPL(blk_unprep_request
);
2687 * queue lock must be held
2689 void blk_finish_request(struct request
*req
, int error
)
2691 struct request_queue
*q
= req
->q
;
2693 if (req
->rq_flags
& RQF_STATS
)
2694 blk_stat_add(&q
->rq_stats
[rq_data_dir(req
)], req
);
2696 if (req
->rq_flags
& RQF_QUEUED
)
2697 blk_queue_end_tag(q
, req
);
2699 BUG_ON(blk_queued_rq(req
));
2701 if (unlikely(laptop_mode
) && req
->cmd_type
== REQ_TYPE_FS
)
2702 laptop_io_completion(&req
->q
->backing_dev_info
);
2704 blk_delete_timer(req
);
2706 if (req
->rq_flags
& RQF_DONTPREP
)
2707 blk_unprep_request(req
);
2709 blk_account_io_done(req
);
2712 req
->end_io(req
, error
);
2714 if (blk_bidi_rq(req
))
2715 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
2717 __blk_put_request(q
, req
);
2720 EXPORT_SYMBOL(blk_finish_request
);
2723 * blk_end_bidi_request - Complete a bidi request
2724 * @rq: the request to complete
2725 * @error: %0 for success, < %0 for error
2726 * @nr_bytes: number of bytes to complete @rq
2727 * @bidi_bytes: number of bytes to complete @rq->next_rq
2730 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2731 * Drivers that supports bidi can safely call this member for any
2732 * type of request, bidi or uni. In the later case @bidi_bytes is
2736 * %false - we are done with this request
2737 * %true - still buffers pending for this request
2739 static bool blk_end_bidi_request(struct request
*rq
, int error
,
2740 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2742 struct request_queue
*q
= rq
->q
;
2743 unsigned long flags
;
2745 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2748 spin_lock_irqsave(q
->queue_lock
, flags
);
2749 blk_finish_request(rq
, error
);
2750 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2756 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2757 * @rq: the request to complete
2758 * @error: %0 for success, < %0 for error
2759 * @nr_bytes: number of bytes to complete @rq
2760 * @bidi_bytes: number of bytes to complete @rq->next_rq
2763 * Identical to blk_end_bidi_request() except that queue lock is
2764 * assumed to be locked on entry and remains so on return.
2767 * %false - we are done with this request
2768 * %true - still buffers pending for this request
2770 bool __blk_end_bidi_request(struct request
*rq
, int error
,
2771 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2773 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2776 blk_finish_request(rq
, error
);
2782 * blk_end_request - Helper function for drivers to complete the request.
2783 * @rq: the request being processed
2784 * @error: %0 for success, < %0 for error
2785 * @nr_bytes: number of bytes to complete
2788 * Ends I/O on a number of bytes attached to @rq.
2789 * If @rq has leftover, sets it up for the next range of segments.
2792 * %false - we are done with this request
2793 * %true - still buffers pending for this request
2795 bool blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2797 return blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2799 EXPORT_SYMBOL(blk_end_request
);
2802 * blk_end_request_all - Helper function for drives to finish the request.
2803 * @rq: the request to finish
2804 * @error: %0 for success, < %0 for error
2807 * Completely finish @rq.
2809 void blk_end_request_all(struct request
*rq
, int error
)
2812 unsigned int bidi_bytes
= 0;
2814 if (unlikely(blk_bidi_rq(rq
)))
2815 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2817 pending
= blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2820 EXPORT_SYMBOL(blk_end_request_all
);
2823 * blk_end_request_cur - Helper function to finish the current request chunk.
2824 * @rq: the request to finish the current chunk for
2825 * @error: %0 for success, < %0 for error
2828 * Complete the current consecutively mapped chunk from @rq.
2831 * %false - we are done with this request
2832 * %true - still buffers pending for this request
2834 bool blk_end_request_cur(struct request
*rq
, int error
)
2836 return blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2838 EXPORT_SYMBOL(blk_end_request_cur
);
2841 * blk_end_request_err - Finish a request till the next failure boundary.
2842 * @rq: the request to finish till the next failure boundary for
2843 * @error: must be negative errno
2846 * Complete @rq till the next failure boundary.
2849 * %false - we are done with this request
2850 * %true - still buffers pending for this request
2852 bool blk_end_request_err(struct request
*rq
, int error
)
2854 WARN_ON(error
>= 0);
2855 return blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2857 EXPORT_SYMBOL_GPL(blk_end_request_err
);
2860 * __blk_end_request - Helper function for drivers to complete the request.
2861 * @rq: the request being processed
2862 * @error: %0 for success, < %0 for error
2863 * @nr_bytes: number of bytes to complete
2866 * Must be called with queue lock held unlike blk_end_request().
2869 * %false - we are done with this request
2870 * %true - still buffers pending for this request
2872 bool __blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2874 return __blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2876 EXPORT_SYMBOL(__blk_end_request
);
2879 * __blk_end_request_all - Helper function for drives to finish the request.
2880 * @rq: the request to finish
2881 * @error: %0 for success, < %0 for error
2884 * Completely finish @rq. Must be called with queue lock held.
2886 void __blk_end_request_all(struct request
*rq
, int error
)
2889 unsigned int bidi_bytes
= 0;
2891 if (unlikely(blk_bidi_rq(rq
)))
2892 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2894 pending
= __blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2897 EXPORT_SYMBOL(__blk_end_request_all
);
2900 * __blk_end_request_cur - Helper function to finish the current request chunk.
2901 * @rq: the request to finish the current chunk for
2902 * @error: %0 for success, < %0 for error
2905 * Complete the current consecutively mapped chunk from @rq. Must
2906 * be called with queue lock held.
2909 * %false - we are done with this request
2910 * %true - still buffers pending for this request
2912 bool __blk_end_request_cur(struct request
*rq
, int error
)
2914 return __blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2916 EXPORT_SYMBOL(__blk_end_request_cur
);
2919 * __blk_end_request_err - Finish a request till the next failure boundary.
2920 * @rq: the request to finish till the next failure boundary for
2921 * @error: must be negative errno
2924 * Complete @rq till the next failure boundary. Must be called
2925 * with queue lock held.
2928 * %false - we are done with this request
2929 * %true - still buffers pending for this request
2931 bool __blk_end_request_err(struct request
*rq
, int error
)
2933 WARN_ON(error
>= 0);
2934 return __blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2936 EXPORT_SYMBOL_GPL(__blk_end_request_err
);
2938 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
2941 if (bio_has_data(bio
))
2942 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2944 rq
->__data_len
= bio
->bi_iter
.bi_size
;
2945 rq
->bio
= rq
->biotail
= bio
;
2948 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2951 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2953 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2954 * @rq: the request to be flushed
2957 * Flush all pages in @rq.
2959 void rq_flush_dcache_pages(struct request
*rq
)
2961 struct req_iterator iter
;
2962 struct bio_vec bvec
;
2964 rq_for_each_segment(bvec
, rq
, iter
)
2965 flush_dcache_page(bvec
.bv_page
);
2967 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages
);
2971 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2972 * @q : the queue of the device being checked
2975 * Check if underlying low-level drivers of a device are busy.
2976 * If the drivers want to export their busy state, they must set own
2977 * exporting function using blk_queue_lld_busy() first.
2979 * Basically, this function is used only by request stacking drivers
2980 * to stop dispatching requests to underlying devices when underlying
2981 * devices are busy. This behavior helps more I/O merging on the queue
2982 * of the request stacking driver and prevents I/O throughput regression
2983 * on burst I/O load.
2986 * 0 - Not busy (The request stacking driver should dispatch request)
2987 * 1 - Busy (The request stacking driver should stop dispatching request)
2989 int blk_lld_busy(struct request_queue
*q
)
2992 return q
->lld_busy_fn(q
);
2996 EXPORT_SYMBOL_GPL(blk_lld_busy
);
2999 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
3000 * @rq: the clone request to be cleaned up
3003 * Free all bios in @rq for a cloned request.
3005 void blk_rq_unprep_clone(struct request
*rq
)
3009 while ((bio
= rq
->bio
) != NULL
) {
3010 rq
->bio
= bio
->bi_next
;
3015 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone
);
3018 * Copy attributes of the original request to the clone request.
3019 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3021 static void __blk_rq_prep_clone(struct request
*dst
, struct request
*src
)
3023 dst
->cpu
= src
->cpu
;
3024 dst
->cmd_flags
= src
->cmd_flags
| REQ_NOMERGE
;
3025 dst
->cmd_type
= src
->cmd_type
;
3026 dst
->__sector
= blk_rq_pos(src
);
3027 dst
->__data_len
= blk_rq_bytes(src
);
3028 dst
->nr_phys_segments
= src
->nr_phys_segments
;
3029 dst
->ioprio
= src
->ioprio
;
3030 dst
->extra_len
= src
->extra_len
;
3034 * blk_rq_prep_clone - Helper function to setup clone request
3035 * @rq: the request to be setup
3036 * @rq_src: original request to be cloned
3037 * @bs: bio_set that bios for clone are allocated from
3038 * @gfp_mask: memory allocation mask for bio
3039 * @bio_ctr: setup function to be called for each clone bio.
3040 * Returns %0 for success, non %0 for failure.
3041 * @data: private data to be passed to @bio_ctr
3044 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3045 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3046 * are not copied, and copying such parts is the caller's responsibility.
3047 * Also, pages which the original bios are pointing to are not copied
3048 * and the cloned bios just point same pages.
3049 * So cloned bios must be completed before original bios, which means
3050 * the caller must complete @rq before @rq_src.
3052 int blk_rq_prep_clone(struct request
*rq
, struct request
*rq_src
,
3053 struct bio_set
*bs
, gfp_t gfp_mask
,
3054 int (*bio_ctr
)(struct bio
*, struct bio
*, void *),
3057 struct bio
*bio
, *bio_src
;
3062 __rq_for_each_bio(bio_src
, rq_src
) {
3063 bio
= bio_clone_fast(bio_src
, gfp_mask
, bs
);
3067 if (bio_ctr
&& bio_ctr(bio
, bio_src
, data
))
3071 rq
->biotail
->bi_next
= bio
;
3074 rq
->bio
= rq
->biotail
= bio
;
3077 __blk_rq_prep_clone(rq
, rq_src
);
3084 blk_rq_unprep_clone(rq
);
3088 EXPORT_SYMBOL_GPL(blk_rq_prep_clone
);
3090 int kblockd_schedule_work(struct work_struct
*work
)
3092 return queue_work(kblockd_workqueue
, work
);
3094 EXPORT_SYMBOL(kblockd_schedule_work
);
3096 int kblockd_schedule_work_on(int cpu
, struct work_struct
*work
)
3098 return queue_work_on(cpu
, kblockd_workqueue
, work
);
3100 EXPORT_SYMBOL(kblockd_schedule_work_on
);
3102 int kblockd_schedule_delayed_work(struct delayed_work
*dwork
,
3103 unsigned long delay
)
3105 return queue_delayed_work(kblockd_workqueue
, dwork
, delay
);
3107 EXPORT_SYMBOL(kblockd_schedule_delayed_work
);
3109 int kblockd_schedule_delayed_work_on(int cpu
, struct delayed_work
*dwork
,
3110 unsigned long delay
)
3112 return queue_delayed_work_on(cpu
, kblockd_workqueue
, dwork
, delay
);
3114 EXPORT_SYMBOL(kblockd_schedule_delayed_work_on
);
3117 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3118 * @plug: The &struct blk_plug that needs to be initialized
3121 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3122 * pending I/O should the task end up blocking between blk_start_plug() and
3123 * blk_finish_plug(). This is important from a performance perspective, but
3124 * also ensures that we don't deadlock. For instance, if the task is blocking
3125 * for a memory allocation, memory reclaim could end up wanting to free a
3126 * page belonging to that request that is currently residing in our private
3127 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3128 * this kind of deadlock.
3130 void blk_start_plug(struct blk_plug
*plug
)
3132 struct task_struct
*tsk
= current
;
3135 * If this is a nested plug, don't actually assign it.
3140 INIT_LIST_HEAD(&plug
->list
);
3141 INIT_LIST_HEAD(&plug
->mq_list
);
3142 INIT_LIST_HEAD(&plug
->cb_list
);
3144 * Store ordering should not be needed here, since a potential
3145 * preempt will imply a full memory barrier
3149 EXPORT_SYMBOL(blk_start_plug
);
3151 static int plug_rq_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
3153 struct request
*rqa
= container_of(a
, struct request
, queuelist
);
3154 struct request
*rqb
= container_of(b
, struct request
, queuelist
);
3156 return !(rqa
->q
< rqb
->q
||
3157 (rqa
->q
== rqb
->q
&& blk_rq_pos(rqa
) < blk_rq_pos(rqb
)));
3161 * If 'from_schedule' is true, then postpone the dispatch of requests
3162 * until a safe kblockd context. We due this to avoid accidental big
3163 * additional stack usage in driver dispatch, in places where the originally
3164 * plugger did not intend it.
3166 static void queue_unplugged(struct request_queue
*q
, unsigned int depth
,
3168 __releases(q
->queue_lock
)
3170 trace_block_unplug(q
, depth
, !from_schedule
);
3173 blk_run_queue_async(q
);
3176 spin_unlock(q
->queue_lock
);
3179 static void flush_plug_callbacks(struct blk_plug
*plug
, bool from_schedule
)
3181 LIST_HEAD(callbacks
);
3183 while (!list_empty(&plug
->cb_list
)) {
3184 list_splice_init(&plug
->cb_list
, &callbacks
);
3186 while (!list_empty(&callbacks
)) {
3187 struct blk_plug_cb
*cb
= list_first_entry(&callbacks
,
3190 list_del(&cb
->list
);
3191 cb
->callback(cb
, from_schedule
);
3196 struct blk_plug_cb
*blk_check_plugged(blk_plug_cb_fn unplug
, void *data
,
3199 struct blk_plug
*plug
= current
->plug
;
3200 struct blk_plug_cb
*cb
;
3205 list_for_each_entry(cb
, &plug
->cb_list
, list
)
3206 if (cb
->callback
== unplug
&& cb
->data
== data
)
3209 /* Not currently on the callback list */
3210 BUG_ON(size
< sizeof(*cb
));
3211 cb
= kzalloc(size
, GFP_ATOMIC
);
3214 cb
->callback
= unplug
;
3215 list_add(&cb
->list
, &plug
->cb_list
);
3219 EXPORT_SYMBOL(blk_check_plugged
);
3221 void blk_flush_plug_list(struct blk_plug
*plug
, bool from_schedule
)
3223 struct request_queue
*q
;
3224 unsigned long flags
;
3229 flush_plug_callbacks(plug
, from_schedule
);
3231 if (!list_empty(&plug
->mq_list
))
3232 blk_mq_flush_plug_list(plug
, from_schedule
);
3234 if (list_empty(&plug
->list
))
3237 list_splice_init(&plug
->list
, &list
);
3239 list_sort(NULL
, &list
, plug_rq_cmp
);
3245 * Save and disable interrupts here, to avoid doing it for every
3246 * queue lock we have to take.
3248 local_irq_save(flags
);
3249 while (!list_empty(&list
)) {
3250 rq
= list_entry_rq(list
.next
);
3251 list_del_init(&rq
->queuelist
);
3255 * This drops the queue lock
3258 queue_unplugged(q
, depth
, from_schedule
);
3261 spin_lock(q
->queue_lock
);
3265 * Short-circuit if @q is dead
3267 if (unlikely(blk_queue_dying(q
))) {
3268 __blk_end_request_all(rq
, -ENODEV
);
3273 * rq is already accounted, so use raw insert
3275 if (rq
->cmd_flags
& (REQ_PREFLUSH
| REQ_FUA
))
3276 __elv_add_request(q
, rq
, ELEVATOR_INSERT_FLUSH
);
3278 __elv_add_request(q
, rq
, ELEVATOR_INSERT_SORT_MERGE
);
3284 * This drops the queue lock
3287 queue_unplugged(q
, depth
, from_schedule
);
3289 local_irq_restore(flags
);
3292 void blk_finish_plug(struct blk_plug
*plug
)
3294 if (plug
!= current
->plug
)
3296 blk_flush_plug_list(plug
, false);
3298 current
->plug
= NULL
;
3300 EXPORT_SYMBOL(blk_finish_plug
);
3302 bool blk_poll(struct request_queue
*q
, blk_qc_t cookie
)
3304 struct blk_plug
*plug
;
3306 unsigned int queue_num
;
3307 struct blk_mq_hw_ctx
*hctx
;
3309 if (!q
->mq_ops
|| !q
->mq_ops
->poll
|| !blk_qc_t_valid(cookie
) ||
3310 !test_bit(QUEUE_FLAG_POLL
, &q
->queue_flags
))
3313 queue_num
= blk_qc_t_to_queue_num(cookie
);
3314 hctx
= q
->queue_hw_ctx
[queue_num
];
3315 hctx
->poll_considered
++;
3317 plug
= current
->plug
;
3319 blk_flush_plug_list(plug
, false);
3321 state
= current
->state
;
3322 while (!need_resched()) {
3325 hctx
->poll_invoked
++;
3327 ret
= q
->mq_ops
->poll(hctx
, blk_qc_t_to_tag(cookie
));
3329 hctx
->poll_success
++;
3330 set_current_state(TASK_RUNNING
);
3334 if (signal_pending_state(state
, current
))
3335 set_current_state(TASK_RUNNING
);
3337 if (current
->state
== TASK_RUNNING
)
3346 EXPORT_SYMBOL_GPL(blk_poll
);
3350 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3351 * @q: the queue of the device
3352 * @dev: the device the queue belongs to
3355 * Initialize runtime-PM-related fields for @q and start auto suspend for
3356 * @dev. Drivers that want to take advantage of request-based runtime PM
3357 * should call this function after @dev has been initialized, and its
3358 * request queue @q has been allocated, and runtime PM for it can not happen
3359 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3360 * cases, driver should call this function before any I/O has taken place.
3362 * This function takes care of setting up using auto suspend for the device,
3363 * the autosuspend delay is set to -1 to make runtime suspend impossible
3364 * until an updated value is either set by user or by driver. Drivers do
3365 * not need to touch other autosuspend settings.
3367 * The block layer runtime PM is request based, so only works for drivers
3368 * that use request as their IO unit instead of those directly use bio's.
3370 void blk_pm_runtime_init(struct request_queue
*q
, struct device
*dev
)
3373 q
->rpm_status
= RPM_ACTIVE
;
3374 pm_runtime_set_autosuspend_delay(q
->dev
, -1);
3375 pm_runtime_use_autosuspend(q
->dev
);
3377 EXPORT_SYMBOL(blk_pm_runtime_init
);
3380 * blk_pre_runtime_suspend - Pre runtime suspend check
3381 * @q: the queue of the device
3384 * This function will check if runtime suspend is allowed for the device
3385 * by examining if there are any requests pending in the queue. If there
3386 * are requests pending, the device can not be runtime suspended; otherwise,
3387 * the queue's status will be updated to SUSPENDING and the driver can
3388 * proceed to suspend the device.
3390 * For the not allowed case, we mark last busy for the device so that
3391 * runtime PM core will try to autosuspend it some time later.
3393 * This function should be called near the start of the device's
3394 * runtime_suspend callback.
3397 * 0 - OK to runtime suspend the device
3398 * -EBUSY - Device should not be runtime suspended
3400 int blk_pre_runtime_suspend(struct request_queue
*q
)
3407 spin_lock_irq(q
->queue_lock
);
3408 if (q
->nr_pending
) {
3410 pm_runtime_mark_last_busy(q
->dev
);
3412 q
->rpm_status
= RPM_SUSPENDING
;
3414 spin_unlock_irq(q
->queue_lock
);
3417 EXPORT_SYMBOL(blk_pre_runtime_suspend
);
3420 * blk_post_runtime_suspend - Post runtime suspend processing
3421 * @q: the queue of the device
3422 * @err: return value of the device's runtime_suspend function
3425 * Update the queue's runtime status according to the return value of the
3426 * device's runtime suspend function and mark last busy for the device so
3427 * that PM core will try to auto suspend the device at a later time.
3429 * This function should be called near the end of the device's
3430 * runtime_suspend callback.
3432 void blk_post_runtime_suspend(struct request_queue
*q
, int err
)
3437 spin_lock_irq(q
->queue_lock
);
3439 q
->rpm_status
= RPM_SUSPENDED
;
3441 q
->rpm_status
= RPM_ACTIVE
;
3442 pm_runtime_mark_last_busy(q
->dev
);
3444 spin_unlock_irq(q
->queue_lock
);
3446 EXPORT_SYMBOL(blk_post_runtime_suspend
);
3449 * blk_pre_runtime_resume - Pre runtime resume processing
3450 * @q: the queue of the device
3453 * Update the queue's runtime status to RESUMING in preparation for the
3454 * runtime resume of the device.
3456 * This function should be called near the start of the device's
3457 * runtime_resume callback.
3459 void blk_pre_runtime_resume(struct request_queue
*q
)
3464 spin_lock_irq(q
->queue_lock
);
3465 q
->rpm_status
= RPM_RESUMING
;
3466 spin_unlock_irq(q
->queue_lock
);
3468 EXPORT_SYMBOL(blk_pre_runtime_resume
);
3471 * blk_post_runtime_resume - Post runtime resume processing
3472 * @q: the queue of the device
3473 * @err: return value of the device's runtime_resume function
3476 * Update the queue's runtime status according to the return value of the
3477 * device's runtime_resume function. If it is successfully resumed, process
3478 * the requests that are queued into the device's queue when it is resuming
3479 * and then mark last busy and initiate autosuspend for it.
3481 * This function should be called near the end of the device's
3482 * runtime_resume callback.
3484 void blk_post_runtime_resume(struct request_queue
*q
, int err
)
3489 spin_lock_irq(q
->queue_lock
);
3491 q
->rpm_status
= RPM_ACTIVE
;
3493 pm_runtime_mark_last_busy(q
->dev
);
3494 pm_request_autosuspend(q
->dev
);
3496 q
->rpm_status
= RPM_SUSPENDED
;
3498 spin_unlock_irq(q
->queue_lock
);
3500 EXPORT_SYMBOL(blk_post_runtime_resume
);
3503 * blk_set_runtime_active - Force runtime status of the queue to be active
3504 * @q: the queue of the device
3506 * If the device is left runtime suspended during system suspend the resume
3507 * hook typically resumes the device and corrects runtime status
3508 * accordingly. However, that does not affect the queue runtime PM status
3509 * which is still "suspended". This prevents processing requests from the
3512 * This function can be used in driver's resume hook to correct queue
3513 * runtime PM status and re-enable peeking requests from the queue. It
3514 * should be called before first request is added to the queue.
3516 void blk_set_runtime_active(struct request_queue
*q
)
3518 spin_lock_irq(q
->queue_lock
);
3519 q
->rpm_status
= RPM_ACTIVE
;
3520 pm_runtime_mark_last_busy(q
->dev
);
3521 pm_request_autosuspend(q
->dev
);
3522 spin_unlock_irq(q
->queue_lock
);
3524 EXPORT_SYMBOL(blk_set_runtime_active
);
3527 int __init
blk_dev_init(void)
3529 BUILD_BUG_ON(REQ_OP_LAST
>= (1 << REQ_OP_BITS
));
3530 BUILD_BUG_ON(REQ_OP_BITS
+ REQ_FLAG_BITS
> 8 *
3531 FIELD_SIZEOF(struct request
, cmd_flags
));
3532 BUILD_BUG_ON(REQ_OP_BITS
+ REQ_FLAG_BITS
> 8 *
3533 FIELD_SIZEOF(struct bio
, bi_opf
));
3535 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3536 kblockd_workqueue
= alloc_workqueue("kblockd",
3537 WQ_MEM_RECLAIM
| WQ_HIGHPRI
, 0);
3538 if (!kblockd_workqueue
)
3539 panic("Failed to create kblockd\n");
3541 request_cachep
= kmem_cache_create("blkdev_requests",
3542 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
3544 blk_requestq_cachep
= kmem_cache_create("request_queue",
3545 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);