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
->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_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_delayed_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 queue_flag_set_unlocked(QUEUE_FLAG_DYING
, q
);
521 blk_mq_wake_waiters(q
);
523 struct request_list
*rl
;
525 blk_queue_for_each_rl(rl
, q
) {
527 wake_up(&rl
->wait
[BLK_RW_SYNC
]);
528 wake_up(&rl
->wait
[BLK_RW_ASYNC
]);
533 EXPORT_SYMBOL_GPL(blk_set_queue_dying
);
536 * blk_cleanup_queue - shutdown a request queue
537 * @q: request queue to shutdown
539 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
540 * put it. All future requests will be failed immediately with -ENODEV.
542 void blk_cleanup_queue(struct request_queue
*q
)
544 spinlock_t
*lock
= q
->queue_lock
;
546 /* mark @q DYING, no new request or merges will be allowed afterwards */
547 mutex_lock(&q
->sysfs_lock
);
548 blk_set_queue_dying(q
);
552 * A dying queue is permanently in bypass mode till released. Note
553 * that, unlike blk_queue_bypass_start(), we aren't performing
554 * synchronize_rcu() after entering bypass mode to avoid the delay
555 * as some drivers create and destroy a lot of queues while
556 * probing. This is still safe because blk_release_queue() will be
557 * called only after the queue refcnt drops to zero and nothing,
558 * RCU or not, would be traversing the queue by then.
561 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
563 queue_flag_set(QUEUE_FLAG_NOMERGES
, q
);
564 queue_flag_set(QUEUE_FLAG_NOXMERGES
, q
);
565 queue_flag_set(QUEUE_FLAG_DYING
, q
);
566 spin_unlock_irq(lock
);
567 mutex_unlock(&q
->sysfs_lock
);
570 * Drain all requests queued before DYING marking. Set DEAD flag to
571 * prevent that q->request_fn() gets invoked after draining finished.
576 __blk_drain_queue(q
, true);
577 queue_flag_set(QUEUE_FLAG_DEAD
, q
);
578 spin_unlock_irq(lock
);
580 /* for synchronous bio-based driver finish in-flight integrity i/o */
581 blk_flush_integrity();
583 /* @q won't process any more request, flush async actions */
584 del_timer_sync(&q
->backing_dev_info
.laptop_mode_wb_timer
);
588 blk_mq_free_queue(q
);
589 percpu_ref_exit(&q
->q_usage_counter
);
592 if (q
->queue_lock
!= &q
->__queue_lock
)
593 q
->queue_lock
= &q
->__queue_lock
;
594 spin_unlock_irq(lock
);
596 bdi_unregister(&q
->backing_dev_info
);
598 /* @q is and will stay empty, shutdown and put */
601 EXPORT_SYMBOL(blk_cleanup_queue
);
603 /* Allocate memory local to the request queue */
604 static void *alloc_request_struct(gfp_t gfp_mask
, void *data
)
606 int nid
= (int)(long)data
;
607 return kmem_cache_alloc_node(request_cachep
, gfp_mask
, nid
);
610 static void free_request_struct(void *element
, void *unused
)
612 kmem_cache_free(request_cachep
, element
);
615 int blk_init_rl(struct request_list
*rl
, struct request_queue
*q
,
618 if (unlikely(rl
->rq_pool
))
622 rl
->count
[BLK_RW_SYNC
] = rl
->count
[BLK_RW_ASYNC
] = 0;
623 rl
->starved
[BLK_RW_SYNC
] = rl
->starved
[BLK_RW_ASYNC
] = 0;
624 init_waitqueue_head(&rl
->wait
[BLK_RW_SYNC
]);
625 init_waitqueue_head(&rl
->wait
[BLK_RW_ASYNC
]);
627 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, alloc_request_struct
,
629 (void *)(long)q
->node
, gfp_mask
,
637 void blk_exit_rl(struct request_list
*rl
)
640 mempool_destroy(rl
->rq_pool
);
643 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
645 return blk_alloc_queue_node(gfp_mask
, NUMA_NO_NODE
);
647 EXPORT_SYMBOL(blk_alloc_queue
);
649 int blk_queue_enter(struct request_queue
*q
, bool nowait
)
654 if (percpu_ref_tryget_live(&q
->q_usage_counter
))
660 ret
= wait_event_interruptible(q
->mq_freeze_wq
,
661 !atomic_read(&q
->mq_freeze_depth
) ||
663 if (blk_queue_dying(q
))
670 void blk_queue_exit(struct request_queue
*q
)
672 percpu_ref_put(&q
->q_usage_counter
);
675 static void blk_queue_usage_counter_release(struct percpu_ref
*ref
)
677 struct request_queue
*q
=
678 container_of(ref
, struct request_queue
, q_usage_counter
);
680 wake_up_all(&q
->mq_freeze_wq
);
683 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
685 struct request_queue
*q
;
688 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
689 gfp_mask
| __GFP_ZERO
, node_id
);
693 q
->id
= ida_simple_get(&blk_queue_ida
, 0, 0, gfp_mask
);
697 q
->bio_split
= bioset_create(BIO_POOL_SIZE
, 0);
701 q
->backing_dev_info
.ra_pages
=
702 (VM_MAX_READAHEAD
* 1024) / PAGE_CACHE_SIZE
;
703 q
->backing_dev_info
.capabilities
= BDI_CAP_CGROUP_WRITEBACK
;
704 q
->backing_dev_info
.name
= "block";
707 err
= bdi_init(&q
->backing_dev_info
);
711 setup_timer(&q
->backing_dev_info
.laptop_mode_wb_timer
,
712 laptop_mode_timer_fn
, (unsigned long) q
);
713 setup_timer(&q
->timeout
, blk_rq_timed_out_timer
, (unsigned long) q
);
714 INIT_LIST_HEAD(&q
->queue_head
);
715 INIT_LIST_HEAD(&q
->timeout_list
);
716 INIT_LIST_HEAD(&q
->icq_list
);
717 #ifdef CONFIG_BLK_CGROUP
718 INIT_LIST_HEAD(&q
->blkg_list
);
720 INIT_DELAYED_WORK(&q
->delay_work
, blk_delay_work
);
722 kobject_init(&q
->kobj
, &blk_queue_ktype
);
724 mutex_init(&q
->sysfs_lock
);
725 spin_lock_init(&q
->__queue_lock
);
728 * By default initialize queue_lock to internal lock and driver can
729 * override it later if need be.
731 q
->queue_lock
= &q
->__queue_lock
;
734 * A queue starts its life with bypass turned on to avoid
735 * unnecessary bypass on/off overhead and nasty surprises during
736 * init. The initial bypass will be finished when the queue is
737 * registered by blk_register_queue().
740 __set_bit(QUEUE_FLAG_BYPASS
, &q
->queue_flags
);
742 init_waitqueue_head(&q
->mq_freeze_wq
);
745 * Init percpu_ref in atomic mode so that it's faster to shutdown.
746 * See blk_register_queue() for details.
748 if (percpu_ref_init(&q
->q_usage_counter
,
749 blk_queue_usage_counter_release
,
750 PERCPU_REF_INIT_ATOMIC
, GFP_KERNEL
))
753 if (blkcg_init_queue(q
))
759 percpu_ref_exit(&q
->q_usage_counter
);
761 bdi_destroy(&q
->backing_dev_info
);
763 bioset_free(q
->bio_split
);
765 ida_simple_remove(&blk_queue_ida
, q
->id
);
767 kmem_cache_free(blk_requestq_cachep
, q
);
770 EXPORT_SYMBOL(blk_alloc_queue_node
);
773 * blk_init_queue - prepare a request queue for use with a block device
774 * @rfn: The function to be called to process requests that have been
775 * placed on the queue.
776 * @lock: Request queue spin lock
779 * If a block device wishes to use the standard request handling procedures,
780 * which sorts requests and coalesces adjacent requests, then it must
781 * call blk_init_queue(). The function @rfn will be called when there
782 * are requests on the queue that need to be processed. If the device
783 * supports plugging, then @rfn may not be called immediately when requests
784 * are available on the queue, but may be called at some time later instead.
785 * Plugged queues are generally unplugged when a buffer belonging to one
786 * of the requests on the queue is needed, or due to memory pressure.
788 * @rfn is not required, or even expected, to remove all requests off the
789 * queue, but only as many as it can handle at a time. If it does leave
790 * requests on the queue, it is responsible for arranging that the requests
791 * get dealt with eventually.
793 * The queue spin lock must be held while manipulating the requests on the
794 * request queue; this lock will be taken also from interrupt context, so irq
795 * disabling is needed for it.
797 * Function returns a pointer to the initialized request queue, or %NULL if
801 * blk_init_queue() must be paired with a blk_cleanup_queue() call
802 * when the block device is deactivated (such as at module unload).
805 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
807 return blk_init_queue_node(rfn
, lock
, NUMA_NO_NODE
);
809 EXPORT_SYMBOL(blk_init_queue
);
811 struct request_queue
*
812 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
814 struct request_queue
*uninit_q
, *q
;
816 uninit_q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
820 q
= blk_init_allocated_queue(uninit_q
, rfn
, lock
);
822 blk_cleanup_queue(uninit_q
);
826 EXPORT_SYMBOL(blk_init_queue_node
);
828 static blk_qc_t
blk_queue_bio(struct request_queue
*q
, struct bio
*bio
);
830 struct request_queue
*
831 blk_init_allocated_queue(struct request_queue
*q
, request_fn_proc
*rfn
,
837 q
->fq
= blk_alloc_flush_queue(q
, NUMA_NO_NODE
, 0);
841 if (blk_init_rl(&q
->root_rl
, q
, GFP_KERNEL
))
845 q
->prep_rq_fn
= NULL
;
846 q
->unprep_rq_fn
= NULL
;
847 q
->queue_flags
|= QUEUE_FLAG_DEFAULT
;
849 /* Override internal queue lock with supplied lock pointer */
851 q
->queue_lock
= lock
;
854 * This also sets hw/phys segments, boundary and size
856 blk_queue_make_request(q
, blk_queue_bio
);
858 q
->sg_reserved_size
= INT_MAX
;
860 /* Protect q->elevator from elevator_change */
861 mutex_lock(&q
->sysfs_lock
);
864 if (elevator_init(q
, NULL
)) {
865 mutex_unlock(&q
->sysfs_lock
);
869 mutex_unlock(&q
->sysfs_lock
);
874 blk_free_flush_queue(q
->fq
);
877 EXPORT_SYMBOL(blk_init_allocated_queue
);
879 bool blk_get_queue(struct request_queue
*q
)
881 if (likely(!blk_queue_dying(q
))) {
888 EXPORT_SYMBOL(blk_get_queue
);
890 static inline void blk_free_request(struct request_list
*rl
, struct request
*rq
)
892 if (rq
->cmd_flags
& REQ_ELVPRIV
) {
893 elv_put_request(rl
->q
, rq
);
895 put_io_context(rq
->elv
.icq
->ioc
);
898 mempool_free(rq
, rl
->rq_pool
);
902 * ioc_batching returns true if the ioc is a valid batching request and
903 * should be given priority access to a request.
905 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
911 * Make sure the process is able to allocate at least 1 request
912 * even if the batch times out, otherwise we could theoretically
915 return ioc
->nr_batch_requests
== q
->nr_batching
||
916 (ioc
->nr_batch_requests
> 0
917 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
921 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
922 * will cause the process to be a "batcher" on all queues in the system. This
923 * is the behaviour we want though - once it gets a wakeup it should be given
926 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
928 if (!ioc
|| ioc_batching(q
, ioc
))
931 ioc
->nr_batch_requests
= q
->nr_batching
;
932 ioc
->last_waited
= jiffies
;
935 static void __freed_request(struct request_list
*rl
, int sync
)
937 struct request_queue
*q
= rl
->q
;
939 if (rl
->count
[sync
] < queue_congestion_off_threshold(q
))
940 blk_clear_congested(rl
, sync
);
942 if (rl
->count
[sync
] + 1 <= q
->nr_requests
) {
943 if (waitqueue_active(&rl
->wait
[sync
]))
944 wake_up(&rl
->wait
[sync
]);
946 blk_clear_rl_full(rl
, sync
);
951 * A request has just been released. Account for it, update the full and
952 * congestion status, wake up any waiters. Called under q->queue_lock.
954 static void freed_request(struct request_list
*rl
, unsigned int flags
)
956 struct request_queue
*q
= rl
->q
;
957 int sync
= rw_is_sync(flags
);
961 if (flags
& REQ_ELVPRIV
)
964 __freed_request(rl
, sync
);
966 if (unlikely(rl
->starved
[sync
^ 1]))
967 __freed_request(rl
, sync
^ 1);
970 int blk_update_nr_requests(struct request_queue
*q
, unsigned int nr
)
972 struct request_list
*rl
;
973 int on_thresh
, off_thresh
;
975 spin_lock_irq(q
->queue_lock
);
977 blk_queue_congestion_threshold(q
);
978 on_thresh
= queue_congestion_on_threshold(q
);
979 off_thresh
= queue_congestion_off_threshold(q
);
981 blk_queue_for_each_rl(rl
, q
) {
982 if (rl
->count
[BLK_RW_SYNC
] >= on_thresh
)
983 blk_set_congested(rl
, BLK_RW_SYNC
);
984 else if (rl
->count
[BLK_RW_SYNC
] < off_thresh
)
985 blk_clear_congested(rl
, BLK_RW_SYNC
);
987 if (rl
->count
[BLK_RW_ASYNC
] >= on_thresh
)
988 blk_set_congested(rl
, BLK_RW_ASYNC
);
989 else if (rl
->count
[BLK_RW_ASYNC
] < off_thresh
)
990 blk_clear_congested(rl
, BLK_RW_ASYNC
);
992 if (rl
->count
[BLK_RW_SYNC
] >= q
->nr_requests
) {
993 blk_set_rl_full(rl
, BLK_RW_SYNC
);
995 blk_clear_rl_full(rl
, BLK_RW_SYNC
);
996 wake_up(&rl
->wait
[BLK_RW_SYNC
]);
999 if (rl
->count
[BLK_RW_ASYNC
] >= q
->nr_requests
) {
1000 blk_set_rl_full(rl
, BLK_RW_ASYNC
);
1002 blk_clear_rl_full(rl
, BLK_RW_ASYNC
);
1003 wake_up(&rl
->wait
[BLK_RW_ASYNC
]);
1007 spin_unlock_irq(q
->queue_lock
);
1012 * Determine if elevator data should be initialized when allocating the
1013 * request associated with @bio.
1015 static bool blk_rq_should_init_elevator(struct bio
*bio
)
1021 * Flush requests do not use the elevator so skip initialization.
1022 * This allows a request to share the flush and elevator data.
1024 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
))
1031 * rq_ioc - determine io_context for request allocation
1032 * @bio: request being allocated is for this bio (can be %NULL)
1034 * Determine io_context to use for request allocation for @bio. May return
1035 * %NULL if %current->io_context doesn't exist.
1037 static struct io_context
*rq_ioc(struct bio
*bio
)
1039 #ifdef CONFIG_BLK_CGROUP
1040 if (bio
&& bio
->bi_ioc
)
1043 return current
->io_context
;
1047 * __get_request - get a free request
1048 * @rl: request list to allocate from
1049 * @rw_flags: RW and SYNC flags
1050 * @bio: bio to allocate request for (can be %NULL)
1051 * @gfp_mask: allocation mask
1053 * Get a free request from @q. This function may fail under memory
1054 * pressure or if @q is dead.
1056 * Must be called with @q->queue_lock held and,
1057 * Returns ERR_PTR on failure, with @q->queue_lock held.
1058 * Returns request pointer on success, with @q->queue_lock *not held*.
1060 static struct request
*__get_request(struct request_list
*rl
, int rw_flags
,
1061 struct bio
*bio
, gfp_t gfp_mask
)
1063 struct request_queue
*q
= rl
->q
;
1065 struct elevator_type
*et
= q
->elevator
->type
;
1066 struct io_context
*ioc
= rq_ioc(bio
);
1067 struct io_cq
*icq
= NULL
;
1068 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
1071 if (unlikely(blk_queue_dying(q
)))
1072 return ERR_PTR(-ENODEV
);
1074 may_queue
= elv_may_queue(q
, rw_flags
);
1075 if (may_queue
== ELV_MQUEUE_NO
)
1078 if (rl
->count
[is_sync
]+1 >= queue_congestion_on_threshold(q
)) {
1079 if (rl
->count
[is_sync
]+1 >= q
->nr_requests
) {
1081 * The queue will fill after this allocation, so set
1082 * it as full, and mark this process as "batching".
1083 * This process will be allowed to complete a batch of
1084 * requests, others will be blocked.
1086 if (!blk_rl_full(rl
, is_sync
)) {
1087 ioc_set_batching(q
, ioc
);
1088 blk_set_rl_full(rl
, is_sync
);
1090 if (may_queue
!= ELV_MQUEUE_MUST
1091 && !ioc_batching(q
, ioc
)) {
1093 * The queue is full and the allocating
1094 * process is not a "batcher", and not
1095 * exempted by the IO scheduler
1097 return ERR_PTR(-ENOMEM
);
1101 blk_set_congested(rl
, is_sync
);
1105 * Only allow batching queuers to allocate up to 50% over the defined
1106 * limit of requests, otherwise we could have thousands of requests
1107 * allocated with any setting of ->nr_requests
1109 if (rl
->count
[is_sync
] >= (3 * q
->nr_requests
/ 2))
1110 return ERR_PTR(-ENOMEM
);
1112 q
->nr_rqs
[is_sync
]++;
1113 rl
->count
[is_sync
]++;
1114 rl
->starved
[is_sync
] = 0;
1117 * Decide whether the new request will be managed by elevator. If
1118 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
1119 * prevent the current elevator from being destroyed until the new
1120 * request is freed. This guarantees icq's won't be destroyed and
1121 * makes creating new ones safe.
1123 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1124 * it will be created after releasing queue_lock.
1126 if (blk_rq_should_init_elevator(bio
) && !blk_queue_bypass(q
)) {
1127 rw_flags
|= REQ_ELVPRIV
;
1128 q
->nr_rqs_elvpriv
++;
1129 if (et
->icq_cache
&& ioc
)
1130 icq
= ioc_lookup_icq(ioc
, q
);
1133 if (blk_queue_io_stat(q
))
1134 rw_flags
|= REQ_IO_STAT
;
1135 spin_unlock_irq(q
->queue_lock
);
1137 /* allocate and init request */
1138 rq
= mempool_alloc(rl
->rq_pool
, gfp_mask
);
1143 blk_rq_set_rl(rq
, rl
);
1144 rq
->cmd_flags
= rw_flags
| REQ_ALLOCED
;
1147 if (rw_flags
& REQ_ELVPRIV
) {
1148 if (unlikely(et
->icq_cache
&& !icq
)) {
1150 icq
= ioc_create_icq(ioc
, q
, gfp_mask
);
1156 if (unlikely(elv_set_request(q
, rq
, bio
, gfp_mask
)))
1159 /* @rq->elv.icq holds io_context until @rq is freed */
1161 get_io_context(icq
->ioc
);
1165 * ioc may be NULL here, and ioc_batching will be false. That's
1166 * OK, if the queue is under the request limit then requests need
1167 * not count toward the nr_batch_requests limit. There will always
1168 * be some limit enforced by BLK_BATCH_TIME.
1170 if (ioc_batching(q
, ioc
))
1171 ioc
->nr_batch_requests
--;
1173 trace_block_getrq(q
, bio
, rw_flags
& 1);
1178 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1179 * and may fail indefinitely under memory pressure and thus
1180 * shouldn't stall IO. Treat this request as !elvpriv. This will
1181 * disturb iosched and blkcg but weird is bettern than dead.
1183 printk_ratelimited(KERN_WARNING
"%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1184 __func__
, dev_name(q
->backing_dev_info
.dev
));
1186 rq
->cmd_flags
&= ~REQ_ELVPRIV
;
1189 spin_lock_irq(q
->queue_lock
);
1190 q
->nr_rqs_elvpriv
--;
1191 spin_unlock_irq(q
->queue_lock
);
1196 * Allocation failed presumably due to memory. Undo anything we
1197 * might have messed up.
1199 * Allocating task should really be put onto the front of the wait
1200 * queue, but this is pretty rare.
1202 spin_lock_irq(q
->queue_lock
);
1203 freed_request(rl
, rw_flags
);
1206 * in the very unlikely event that allocation failed and no
1207 * requests for this direction was pending, mark us starved so that
1208 * freeing of a request in the other direction will notice
1209 * us. another possible fix would be to split the rq mempool into
1213 if (unlikely(rl
->count
[is_sync
] == 0))
1214 rl
->starved
[is_sync
] = 1;
1215 return ERR_PTR(-ENOMEM
);
1219 * get_request - get a free request
1220 * @q: request_queue to allocate request from
1221 * @rw_flags: RW and SYNC flags
1222 * @bio: bio to allocate request for (can be %NULL)
1223 * @gfp_mask: allocation mask
1225 * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask,
1226 * this function keeps retrying under memory pressure and fails iff @q is dead.
1228 * Must be called with @q->queue_lock held and,
1229 * Returns ERR_PTR on failure, with @q->queue_lock held.
1230 * Returns request pointer on success, with @q->queue_lock *not held*.
1232 static struct request
*get_request(struct request_queue
*q
, int rw_flags
,
1233 struct bio
*bio
, gfp_t gfp_mask
)
1235 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
1237 struct request_list
*rl
;
1240 rl
= blk_get_rl(q
, bio
); /* transferred to @rq on success */
1242 rq
= __get_request(rl
, rw_flags
, bio
, gfp_mask
);
1246 if (!gfpflags_allow_blocking(gfp_mask
) || unlikely(blk_queue_dying(q
))) {
1251 /* wait on @rl and retry */
1252 prepare_to_wait_exclusive(&rl
->wait
[is_sync
], &wait
,
1253 TASK_UNINTERRUPTIBLE
);
1255 trace_block_sleeprq(q
, bio
, rw_flags
& 1);
1257 spin_unlock_irq(q
->queue_lock
);
1261 * After sleeping, we become a "batching" process and will be able
1262 * to allocate at least one request, and up to a big batch of them
1263 * for a small period time. See ioc_batching, ioc_set_batching
1265 ioc_set_batching(q
, current
->io_context
);
1267 spin_lock_irq(q
->queue_lock
);
1268 finish_wait(&rl
->wait
[is_sync
], &wait
);
1273 static struct request
*blk_old_get_request(struct request_queue
*q
, int rw
,
1278 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
1280 /* create ioc upfront */
1281 create_io_context(gfp_mask
, q
->node
);
1283 spin_lock_irq(q
->queue_lock
);
1284 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
1286 spin_unlock_irq(q
->queue_lock
);
1287 /* q->queue_lock is unlocked at this point */
1292 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
1295 return blk_mq_alloc_request(q
, rw
,
1296 (gfp_mask
& __GFP_DIRECT_RECLAIM
) ?
1297 0 : BLK_MQ_REQ_NOWAIT
);
1299 return blk_old_get_request(q
, rw
, gfp_mask
);
1301 EXPORT_SYMBOL(blk_get_request
);
1304 * blk_make_request - given a bio, allocate a corresponding struct request.
1305 * @q: target request queue
1306 * @bio: The bio describing the memory mappings that will be submitted for IO.
1307 * It may be a chained-bio properly constructed by block/bio layer.
1308 * @gfp_mask: gfp flags to be used for memory allocation
1310 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1311 * type commands. Where the struct request needs to be farther initialized by
1312 * the caller. It is passed a &struct bio, which describes the memory info of
1315 * The caller of blk_make_request must make sure that bi_io_vec
1316 * are set to describe the memory buffers. That bio_data_dir() will return
1317 * the needed direction of the request. (And all bio's in the passed bio-chain
1318 * are properly set accordingly)
1320 * If called under none-sleepable conditions, mapped bio buffers must not
1321 * need bouncing, by calling the appropriate masked or flagged allocator,
1322 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1325 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1326 * given to how you allocate bios. In particular, you cannot use
1327 * __GFP_DIRECT_RECLAIM for anything but the first bio in the chain. Otherwise
1328 * you risk waiting for IO completion of a bio that hasn't been submitted yet,
1329 * thus resulting in a deadlock. Alternatively bios should be allocated using
1330 * bio_kmalloc() instead of bio_alloc(), as that avoids the mempool deadlock.
1331 * If possible a big IO should be split into smaller parts when allocation
1332 * fails. Partial allocation should not be an error, or you risk a live-lock.
1334 struct request
*blk_make_request(struct request_queue
*q
, struct bio
*bio
,
1337 struct request
*rq
= blk_get_request(q
, bio_data_dir(bio
), gfp_mask
);
1342 blk_rq_set_block_pc(rq
);
1345 struct bio
*bounce_bio
= bio
;
1348 blk_queue_bounce(q
, &bounce_bio
);
1349 ret
= blk_rq_append_bio(q
, rq
, bounce_bio
);
1350 if (unlikely(ret
)) {
1351 blk_put_request(rq
);
1352 return ERR_PTR(ret
);
1358 EXPORT_SYMBOL(blk_make_request
);
1361 * blk_rq_set_block_pc - initialize a request to type BLOCK_PC
1362 * @rq: request to be initialized
1365 void blk_rq_set_block_pc(struct request
*rq
)
1367 rq
->cmd_type
= REQ_TYPE_BLOCK_PC
;
1369 rq
->__sector
= (sector_t
) -1;
1370 rq
->bio
= rq
->biotail
= NULL
;
1371 memset(rq
->__cmd
, 0, sizeof(rq
->__cmd
));
1373 EXPORT_SYMBOL(blk_rq_set_block_pc
);
1376 * blk_requeue_request - put a request back on queue
1377 * @q: request queue where request should be inserted
1378 * @rq: request to be inserted
1381 * Drivers often keep queueing requests until the hardware cannot accept
1382 * more, when that condition happens we need to put the request back
1383 * on the queue. Must be called with queue lock held.
1385 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
1387 blk_delete_timer(rq
);
1388 blk_clear_rq_complete(rq
);
1389 trace_block_rq_requeue(q
, rq
);
1391 if (rq
->cmd_flags
& REQ_QUEUED
)
1392 blk_queue_end_tag(q
, rq
);
1394 BUG_ON(blk_queued_rq(rq
));
1396 elv_requeue_request(q
, rq
);
1398 EXPORT_SYMBOL(blk_requeue_request
);
1400 static void add_acct_request(struct request_queue
*q
, struct request
*rq
,
1403 blk_account_io_start(rq
, true);
1404 __elv_add_request(q
, rq
, where
);
1407 static void part_round_stats_single(int cpu
, struct hd_struct
*part
,
1412 if (now
== part
->stamp
)
1415 inflight
= part_in_flight(part
);
1417 __part_stat_add(cpu
, part
, time_in_queue
,
1418 inflight
* (now
- part
->stamp
));
1419 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
1425 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1426 * @cpu: cpu number for stats access
1427 * @part: target partition
1429 * The average IO queue length and utilisation statistics are maintained
1430 * by observing the current state of the queue length and the amount of
1431 * time it has been in this state for.
1433 * Normally, that accounting is done on IO completion, but that can result
1434 * in more than a second's worth of IO being accounted for within any one
1435 * second, leading to >100% utilisation. To deal with that, we call this
1436 * function to do a round-off before returning the results when reading
1437 * /proc/diskstats. This accounts immediately for all queue usage up to
1438 * the current jiffies and restarts the counters again.
1440 void part_round_stats(int cpu
, struct hd_struct
*part
)
1442 unsigned long now
= jiffies
;
1445 part_round_stats_single(cpu
, &part_to_disk(part
)->part0
, now
);
1446 part_round_stats_single(cpu
, part
, now
);
1448 EXPORT_SYMBOL_GPL(part_round_stats
);
1451 static void blk_pm_put_request(struct request
*rq
)
1453 if (rq
->q
->dev
&& !(rq
->cmd_flags
& REQ_PM
) && !--rq
->q
->nr_pending
)
1454 pm_runtime_mark_last_busy(rq
->q
->dev
);
1457 static inline void blk_pm_put_request(struct request
*rq
) {}
1461 * queue lock must be held
1463 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1469 blk_mq_free_request(req
);
1473 blk_pm_put_request(req
);
1475 elv_completed_request(q
, req
);
1477 /* this is a bio leak */
1478 WARN_ON(req
->bio
!= NULL
);
1481 * Request may not have originated from ll_rw_blk. if not,
1482 * it didn't come out of our reserved rq pools
1484 if (req
->cmd_flags
& REQ_ALLOCED
) {
1485 unsigned int flags
= req
->cmd_flags
;
1486 struct request_list
*rl
= blk_rq_rl(req
);
1488 BUG_ON(!list_empty(&req
->queuelist
));
1489 BUG_ON(ELV_ON_HASH(req
));
1491 blk_free_request(rl
, req
);
1492 freed_request(rl
, flags
);
1496 EXPORT_SYMBOL_GPL(__blk_put_request
);
1498 void blk_put_request(struct request
*req
)
1500 struct request_queue
*q
= req
->q
;
1503 blk_mq_free_request(req
);
1505 unsigned long flags
;
1507 spin_lock_irqsave(q
->queue_lock
, flags
);
1508 __blk_put_request(q
, req
);
1509 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1512 EXPORT_SYMBOL(blk_put_request
);
1515 * blk_add_request_payload - add a payload to a request
1516 * @rq: request to update
1517 * @page: page backing the payload
1518 * @len: length of the payload.
1520 * This allows to later add a payload to an already submitted request by
1521 * a block driver. The driver needs to take care of freeing the payload
1524 * Note that this is a quite horrible hack and nothing but handling of
1525 * discard requests should ever use it.
1527 void blk_add_request_payload(struct request
*rq
, struct page
*page
,
1530 struct bio
*bio
= rq
->bio
;
1532 bio
->bi_io_vec
->bv_page
= page
;
1533 bio
->bi_io_vec
->bv_offset
= 0;
1534 bio
->bi_io_vec
->bv_len
= len
;
1536 bio
->bi_iter
.bi_size
= len
;
1538 bio
->bi_phys_segments
= 1;
1540 rq
->__data_len
= rq
->resid_len
= len
;
1541 rq
->nr_phys_segments
= 1;
1543 EXPORT_SYMBOL_GPL(blk_add_request_payload
);
1545 bool bio_attempt_back_merge(struct request_queue
*q
, struct request
*req
,
1548 const int ff
= bio
->bi_rw
& REQ_FAILFAST_MASK
;
1550 if (!ll_back_merge_fn(q
, req
, bio
))
1553 trace_block_bio_backmerge(q
, req
, bio
);
1555 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1556 blk_rq_set_mixed_merge(req
);
1558 req
->biotail
->bi_next
= bio
;
1560 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1561 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1563 blk_account_io_start(req
, false);
1567 bool bio_attempt_front_merge(struct request_queue
*q
, struct request
*req
,
1570 const int ff
= bio
->bi_rw
& REQ_FAILFAST_MASK
;
1572 if (!ll_front_merge_fn(q
, req
, bio
))
1575 trace_block_bio_frontmerge(q
, req
, bio
);
1577 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1578 blk_rq_set_mixed_merge(req
);
1580 bio
->bi_next
= req
->bio
;
1583 req
->__sector
= bio
->bi_iter
.bi_sector
;
1584 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1585 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1587 blk_account_io_start(req
, false);
1592 * blk_attempt_plug_merge - try to merge with %current's plugged list
1593 * @q: request_queue new bio is being queued at
1594 * @bio: new bio being queued
1595 * @request_count: out parameter for number of traversed plugged requests
1596 * @same_queue_rq: pointer to &struct request that gets filled in when
1597 * another request associated with @q is found on the plug list
1598 * (optional, may be %NULL)
1600 * Determine whether @bio being queued on @q can be merged with a request
1601 * on %current's plugged list. Returns %true if merge was successful,
1604 * Plugging coalesces IOs from the same issuer for the same purpose without
1605 * going through @q->queue_lock. As such it's more of an issuing mechanism
1606 * than scheduling, and the request, while may have elvpriv data, is not
1607 * added on the elevator at this point. In addition, we don't have
1608 * reliable access to the elevator outside queue lock. Only check basic
1609 * merging parameters without querying the elevator.
1611 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1613 bool blk_attempt_plug_merge(struct request_queue
*q
, struct bio
*bio
,
1614 unsigned int *request_count
,
1615 struct request
**same_queue_rq
)
1617 struct blk_plug
*plug
;
1620 struct list_head
*plug_list
;
1622 plug
= current
->plug
;
1628 plug_list
= &plug
->mq_list
;
1630 plug_list
= &plug
->list
;
1632 list_for_each_entry_reverse(rq
, plug_list
, queuelist
) {
1638 * Only blk-mq multiple hardware queues case checks the
1639 * rq in the same queue, there should be only one such
1643 *same_queue_rq
= rq
;
1646 if (rq
->q
!= q
|| !blk_rq_merge_ok(rq
, bio
))
1649 el_ret
= blk_try_merge(rq
, bio
);
1650 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1651 ret
= bio_attempt_back_merge(q
, rq
, bio
);
1654 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1655 ret
= bio_attempt_front_merge(q
, rq
, bio
);
1664 unsigned int blk_plug_queued_count(struct request_queue
*q
)
1666 struct blk_plug
*plug
;
1668 struct list_head
*plug_list
;
1669 unsigned int ret
= 0;
1671 plug
= current
->plug
;
1676 plug_list
= &plug
->mq_list
;
1678 plug_list
= &plug
->list
;
1680 list_for_each_entry(rq
, plug_list
, queuelist
) {
1688 void init_request_from_bio(struct request
*req
, struct bio
*bio
)
1690 req
->cmd_type
= REQ_TYPE_FS
;
1692 req
->cmd_flags
|= bio
->bi_rw
& REQ_COMMON_MASK
;
1693 if (bio
->bi_rw
& REQ_RAHEAD
)
1694 req
->cmd_flags
|= REQ_FAILFAST_MASK
;
1697 req
->__sector
= bio
->bi_iter
.bi_sector
;
1698 req
->ioprio
= bio_prio(bio
);
1699 blk_rq_bio_prep(req
->q
, req
, bio
);
1702 static blk_qc_t
blk_queue_bio(struct request_queue
*q
, struct bio
*bio
)
1704 const bool sync
= !!(bio
->bi_rw
& REQ_SYNC
);
1705 struct blk_plug
*plug
;
1706 int el_ret
, rw_flags
, where
= ELEVATOR_INSERT_SORT
;
1707 struct request
*req
;
1708 unsigned int request_count
= 0;
1711 * low level driver can indicate that it wants pages above a
1712 * certain limit bounced to low memory (ie for highmem, or even
1713 * ISA dma in theory)
1715 blk_queue_bounce(q
, &bio
);
1717 blk_queue_split(q
, &bio
, q
->bio_split
);
1719 if (bio_integrity_enabled(bio
) && bio_integrity_prep(bio
)) {
1720 bio
->bi_error
= -EIO
;
1722 return BLK_QC_T_NONE
;
1725 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) {
1726 spin_lock_irq(q
->queue_lock
);
1727 where
= ELEVATOR_INSERT_FLUSH
;
1732 * Check if we can merge with the plugged list before grabbing
1735 if (!blk_queue_nomerges(q
)) {
1736 if (blk_attempt_plug_merge(q
, bio
, &request_count
, NULL
))
1737 return BLK_QC_T_NONE
;
1739 request_count
= blk_plug_queued_count(q
);
1741 spin_lock_irq(q
->queue_lock
);
1743 el_ret
= elv_merge(q
, &req
, bio
);
1744 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1745 if (bio_attempt_back_merge(q
, req
, bio
)) {
1746 elv_bio_merged(q
, req
, bio
);
1747 if (!attempt_back_merge(q
, req
))
1748 elv_merged_request(q
, req
, el_ret
);
1751 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1752 if (bio_attempt_front_merge(q
, req
, bio
)) {
1753 elv_bio_merged(q
, req
, bio
);
1754 if (!attempt_front_merge(q
, req
))
1755 elv_merged_request(q
, req
, el_ret
);
1762 * This sync check and mask will be re-done in init_request_from_bio(),
1763 * but we need to set it earlier to expose the sync flag to the
1764 * rq allocator and io schedulers.
1766 rw_flags
= bio_data_dir(bio
);
1768 rw_flags
|= REQ_SYNC
;
1771 * Grab a free request. This is might sleep but can not fail.
1772 * Returns with the queue unlocked.
1774 req
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
1776 bio
->bi_error
= PTR_ERR(req
);
1782 * After dropping the lock and possibly sleeping here, our request
1783 * may now be mergeable after it had proven unmergeable (above).
1784 * We don't worry about that case for efficiency. It won't happen
1785 * often, and the elevators are able to handle it.
1787 init_request_from_bio(req
, bio
);
1789 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
))
1790 req
->cpu
= raw_smp_processor_id();
1792 plug
= current
->plug
;
1795 * If this is the first request added after a plug, fire
1799 trace_block_plug(q
);
1801 if (request_count
>= BLK_MAX_REQUEST_COUNT
) {
1802 blk_flush_plug_list(plug
, false);
1803 trace_block_plug(q
);
1806 list_add_tail(&req
->queuelist
, &plug
->list
);
1807 blk_account_io_start(req
, true);
1809 spin_lock_irq(q
->queue_lock
);
1810 add_acct_request(q
, req
, where
);
1813 spin_unlock_irq(q
->queue_lock
);
1816 return BLK_QC_T_NONE
;
1820 * If bio->bi_dev is a partition, remap the location
1822 static inline void blk_partition_remap(struct bio
*bio
)
1824 struct block_device
*bdev
= bio
->bi_bdev
;
1826 if (bio_sectors(bio
) && bdev
!= bdev
->bd_contains
) {
1827 struct hd_struct
*p
= bdev
->bd_part
;
1829 bio
->bi_iter
.bi_sector
+= p
->start_sect
;
1830 bio
->bi_bdev
= bdev
->bd_contains
;
1832 trace_block_bio_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
1834 bio
->bi_iter
.bi_sector
- p
->start_sect
);
1838 static void handle_bad_sector(struct bio
*bio
)
1840 char b
[BDEVNAME_SIZE
];
1842 printk(KERN_INFO
"attempt to access beyond end of device\n");
1843 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
1844 bdevname(bio
->bi_bdev
, b
),
1846 (unsigned long long)bio_end_sector(bio
),
1847 (long long)(i_size_read(bio
->bi_bdev
->bd_inode
) >> 9));
1850 #ifdef CONFIG_FAIL_MAKE_REQUEST
1852 static DECLARE_FAULT_ATTR(fail_make_request
);
1854 static int __init
setup_fail_make_request(char *str
)
1856 return setup_fault_attr(&fail_make_request
, str
);
1858 __setup("fail_make_request=", setup_fail_make_request
);
1860 static bool should_fail_request(struct hd_struct
*part
, unsigned int bytes
)
1862 return part
->make_it_fail
&& should_fail(&fail_make_request
, bytes
);
1865 static int __init
fail_make_request_debugfs(void)
1867 struct dentry
*dir
= fault_create_debugfs_attr("fail_make_request",
1868 NULL
, &fail_make_request
);
1870 return PTR_ERR_OR_ZERO(dir
);
1873 late_initcall(fail_make_request_debugfs
);
1875 #else /* CONFIG_FAIL_MAKE_REQUEST */
1877 static inline bool should_fail_request(struct hd_struct
*part
,
1883 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1886 * Check whether this bio extends beyond the end of the device.
1888 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
1895 /* Test device or partition size, when known. */
1896 maxsector
= i_size_read(bio
->bi_bdev
->bd_inode
) >> 9;
1898 sector_t sector
= bio
->bi_iter
.bi_sector
;
1900 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
1902 * This may well happen - the kernel calls bread()
1903 * without checking the size of the device, e.g., when
1904 * mounting a device.
1906 handle_bad_sector(bio
);
1914 static noinline_for_stack
bool
1915 generic_make_request_checks(struct bio
*bio
)
1917 struct request_queue
*q
;
1918 int nr_sectors
= bio_sectors(bio
);
1920 char b
[BDEVNAME_SIZE
];
1921 struct hd_struct
*part
;
1925 if (bio_check_eod(bio
, nr_sectors
))
1928 q
= bdev_get_queue(bio
->bi_bdev
);
1931 "generic_make_request: Trying to access "
1932 "nonexistent block-device %s (%Lu)\n",
1933 bdevname(bio
->bi_bdev
, b
),
1934 (long long) bio
->bi_iter
.bi_sector
);
1938 part
= bio
->bi_bdev
->bd_part
;
1939 if (should_fail_request(part
, bio
->bi_iter
.bi_size
) ||
1940 should_fail_request(&part_to_disk(part
)->part0
,
1941 bio
->bi_iter
.bi_size
))
1945 * If this device has partitions, remap block n
1946 * of partition p to block n+start(p) of the disk.
1948 blk_partition_remap(bio
);
1950 if (bio_check_eod(bio
, nr_sectors
))
1954 * Filter flush bio's early so that make_request based
1955 * drivers without flush support don't have to worry
1958 if ((bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) && !q
->flush_flags
) {
1959 bio
->bi_rw
&= ~(REQ_FLUSH
| REQ_FUA
);
1966 if ((bio
->bi_rw
& REQ_DISCARD
) &&
1967 (!blk_queue_discard(q
) ||
1968 ((bio
->bi_rw
& REQ_SECURE
) && !blk_queue_secdiscard(q
)))) {
1973 if (bio
->bi_rw
& REQ_WRITE_SAME
&& !bdev_write_same(bio
->bi_bdev
)) {
1979 * Various block parts want %current->io_context and lazy ioc
1980 * allocation ends up trading a lot of pain for a small amount of
1981 * memory. Just allocate it upfront. This may fail and block
1982 * layer knows how to live with it.
1984 create_io_context(GFP_ATOMIC
, q
->node
);
1986 if (!blkcg_bio_issue_check(q
, bio
))
1989 trace_block_bio_queue(q
, bio
);
1993 bio
->bi_error
= err
;
1999 * generic_make_request - hand a buffer to its device driver for I/O
2000 * @bio: The bio describing the location in memory and on the device.
2002 * generic_make_request() is used to make I/O requests of block
2003 * devices. It is passed a &struct bio, which describes the I/O that needs
2006 * generic_make_request() does not return any status. The
2007 * success/failure status of the request, along with notification of
2008 * completion, is delivered asynchronously through the bio->bi_end_io
2009 * function described (one day) else where.
2011 * The caller of generic_make_request must make sure that bi_io_vec
2012 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2013 * set to describe the device address, and the
2014 * bi_end_io and optionally bi_private are set to describe how
2015 * completion notification should be signaled.
2017 * generic_make_request and the drivers it calls may use bi_next if this
2018 * bio happens to be merged with someone else, and may resubmit the bio to
2019 * a lower device by calling into generic_make_request recursively, which
2020 * means the bio should NOT be touched after the call to ->make_request_fn.
2022 blk_qc_t
generic_make_request(struct bio
*bio
)
2024 struct bio_list bio_list_on_stack
;
2025 blk_qc_t ret
= BLK_QC_T_NONE
;
2027 if (!generic_make_request_checks(bio
))
2031 * We only want one ->make_request_fn to be active at a time, else
2032 * stack usage with stacked devices could be a problem. So use
2033 * current->bio_list to keep a list of requests submited by a
2034 * make_request_fn function. current->bio_list is also used as a
2035 * flag to say if generic_make_request is currently active in this
2036 * task or not. If it is NULL, then no make_request is active. If
2037 * it is non-NULL, then a make_request is active, and new requests
2038 * should be added at the tail
2040 if (current
->bio_list
) {
2041 bio_list_add(current
->bio_list
, bio
);
2045 /* following loop may be a bit non-obvious, and so deserves some
2047 * Before entering the loop, bio->bi_next is NULL (as all callers
2048 * ensure that) so we have a list with a single bio.
2049 * We pretend that we have just taken it off a longer list, so
2050 * we assign bio_list to a pointer to the bio_list_on_stack,
2051 * thus initialising the bio_list of new bios to be
2052 * added. ->make_request() may indeed add some more bios
2053 * through a recursive call to generic_make_request. If it
2054 * did, we find a non-NULL value in bio_list and re-enter the loop
2055 * from the top. In this case we really did just take the bio
2056 * of the top of the list (no pretending) and so remove it from
2057 * bio_list, and call into ->make_request() again.
2059 BUG_ON(bio
->bi_next
);
2060 bio_list_init(&bio_list_on_stack
);
2061 current
->bio_list
= &bio_list_on_stack
;
2063 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
2065 if (likely(blk_queue_enter(q
, false) == 0)) {
2066 ret
= q
->make_request_fn(q
, bio
);
2070 bio
= bio_list_pop(current
->bio_list
);
2072 struct bio
*bio_next
= bio_list_pop(current
->bio_list
);
2078 current
->bio_list
= NULL
; /* deactivate */
2083 EXPORT_SYMBOL(generic_make_request
);
2086 * submit_bio - submit a bio to the block device layer for I/O
2087 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
2088 * @bio: The &struct bio which describes the I/O
2090 * submit_bio() is very similar in purpose to generic_make_request(), and
2091 * uses that function to do most of the work. Both are fairly rough
2092 * interfaces; @bio must be presetup and ready for I/O.
2095 blk_qc_t
submit_bio(int rw
, struct bio
*bio
)
2100 * If it's a regular read/write or a barrier with data attached,
2101 * go through the normal accounting stuff before submission.
2103 if (bio_has_data(bio
)) {
2106 if (unlikely(rw
& REQ_WRITE_SAME
))
2107 count
= bdev_logical_block_size(bio
->bi_bdev
) >> 9;
2109 count
= bio_sectors(bio
);
2112 count_vm_events(PGPGOUT
, count
);
2114 task_io_account_read(bio
->bi_iter
.bi_size
);
2115 count_vm_events(PGPGIN
, count
);
2118 if (unlikely(block_dump
)) {
2119 char b
[BDEVNAME_SIZE
];
2120 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s (%u sectors)\n",
2121 current
->comm
, task_pid_nr(current
),
2122 (rw
& WRITE
) ? "WRITE" : "READ",
2123 (unsigned long long)bio
->bi_iter
.bi_sector
,
2124 bdevname(bio
->bi_bdev
, b
),
2129 return generic_make_request(bio
);
2131 EXPORT_SYMBOL(submit_bio
);
2134 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2135 * for new the queue limits
2137 * @rq: the request being checked
2140 * @rq may have been made based on weaker limitations of upper-level queues
2141 * in request stacking drivers, and it may violate the limitation of @q.
2142 * Since the block layer and the underlying device driver trust @rq
2143 * after it is inserted to @q, it should be checked against @q before
2144 * the insertion using this generic function.
2146 * Request stacking drivers like request-based dm may change the queue
2147 * limits when retrying requests on other queues. Those requests need
2148 * to be checked against the new queue limits again during dispatch.
2150 static int blk_cloned_rq_check_limits(struct request_queue
*q
,
2153 if (blk_rq_sectors(rq
) > blk_queue_get_max_sectors(q
, rq
->cmd_flags
)) {
2154 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
2159 * queue's settings related to segment counting like q->bounce_pfn
2160 * may differ from that of other stacking queues.
2161 * Recalculate it to check the request correctly on this queue's
2164 blk_recalc_rq_segments(rq
);
2165 if (rq
->nr_phys_segments
> queue_max_segments(q
)) {
2166 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
2174 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2175 * @q: the queue to submit the request
2176 * @rq: the request being queued
2178 int blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
2180 unsigned long flags
;
2181 int where
= ELEVATOR_INSERT_BACK
;
2183 if (blk_cloned_rq_check_limits(q
, rq
))
2187 should_fail_request(&rq
->rq_disk
->part0
, blk_rq_bytes(rq
)))
2191 if (blk_queue_io_stat(q
))
2192 blk_account_io_start(rq
, true);
2193 blk_mq_insert_request(rq
, false, true, true);
2197 spin_lock_irqsave(q
->queue_lock
, flags
);
2198 if (unlikely(blk_queue_dying(q
))) {
2199 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2204 * Submitting request must be dequeued before calling this function
2205 * because it will be linked to another request_queue
2207 BUG_ON(blk_queued_rq(rq
));
2209 if (rq
->cmd_flags
& (REQ_FLUSH
|REQ_FUA
))
2210 where
= ELEVATOR_INSERT_FLUSH
;
2212 add_acct_request(q
, rq
, where
);
2213 if (where
== ELEVATOR_INSERT_FLUSH
)
2215 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2219 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
2222 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2223 * @rq: request to examine
2226 * A request could be merge of IOs which require different failure
2227 * handling. This function determines the number of bytes which
2228 * can be failed from the beginning of the request without
2229 * crossing into area which need to be retried further.
2232 * The number of bytes to fail.
2235 * queue_lock must be held.
2237 unsigned int blk_rq_err_bytes(const struct request
*rq
)
2239 unsigned int ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
2240 unsigned int bytes
= 0;
2243 if (!(rq
->cmd_flags
& REQ_MIXED_MERGE
))
2244 return blk_rq_bytes(rq
);
2247 * Currently the only 'mixing' which can happen is between
2248 * different fastfail types. We can safely fail portions
2249 * which have all the failfast bits that the first one has -
2250 * the ones which are at least as eager to fail as the first
2253 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
2254 if ((bio
->bi_rw
& ff
) != ff
)
2256 bytes
+= bio
->bi_iter
.bi_size
;
2259 /* this could lead to infinite loop */
2260 BUG_ON(blk_rq_bytes(rq
) && !bytes
);
2263 EXPORT_SYMBOL_GPL(blk_rq_err_bytes
);
2265 void blk_account_io_completion(struct request
*req
, unsigned int bytes
)
2267 if (blk_do_io_stat(req
)) {
2268 const int rw
= rq_data_dir(req
);
2269 struct hd_struct
*part
;
2272 cpu
= part_stat_lock();
2274 part_stat_add(cpu
, part
, sectors
[rw
], bytes
>> 9);
2279 void blk_account_io_done(struct request
*req
)
2282 * Account IO completion. flush_rq isn't accounted as a
2283 * normal IO on queueing nor completion. Accounting the
2284 * containing request is enough.
2286 if (blk_do_io_stat(req
) && !(req
->cmd_flags
& REQ_FLUSH_SEQ
)) {
2287 unsigned long duration
= jiffies
- req
->start_time
;
2288 const int rw
= rq_data_dir(req
);
2289 struct hd_struct
*part
;
2292 cpu
= part_stat_lock();
2295 part_stat_inc(cpu
, part
, ios
[rw
]);
2296 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
2297 part_round_stats(cpu
, part
);
2298 part_dec_in_flight(part
, rw
);
2300 hd_struct_put(part
);
2307 * Don't process normal requests when queue is suspended
2308 * or in the process of suspending/resuming
2310 static struct request
*blk_pm_peek_request(struct request_queue
*q
,
2313 if (q
->dev
&& (q
->rpm_status
== RPM_SUSPENDED
||
2314 (q
->rpm_status
!= RPM_ACTIVE
&& !(rq
->cmd_flags
& REQ_PM
))))
2320 static inline struct request
*blk_pm_peek_request(struct request_queue
*q
,
2327 void blk_account_io_start(struct request
*rq
, bool new_io
)
2329 struct hd_struct
*part
;
2330 int rw
= rq_data_dir(rq
);
2333 if (!blk_do_io_stat(rq
))
2336 cpu
= part_stat_lock();
2340 part_stat_inc(cpu
, part
, merges
[rw
]);
2342 part
= disk_map_sector_rcu(rq
->rq_disk
, blk_rq_pos(rq
));
2343 if (!hd_struct_try_get(part
)) {
2345 * The partition is already being removed,
2346 * the request will be accounted on the disk only
2348 * We take a reference on disk->part0 although that
2349 * partition will never be deleted, so we can treat
2350 * it as any other partition.
2352 part
= &rq
->rq_disk
->part0
;
2353 hd_struct_get(part
);
2355 part_round_stats(cpu
, part
);
2356 part_inc_in_flight(part
, rw
);
2364 * blk_peek_request - peek at the top of a request queue
2365 * @q: request queue to peek at
2368 * Return the request at the top of @q. The returned request
2369 * should be started using blk_start_request() before LLD starts
2373 * Pointer to the request at the top of @q if available. Null
2377 * queue_lock must be held.
2379 struct request
*blk_peek_request(struct request_queue
*q
)
2384 while ((rq
= __elv_next_request(q
)) != NULL
) {
2386 rq
= blk_pm_peek_request(q
, rq
);
2390 if (!(rq
->cmd_flags
& REQ_STARTED
)) {
2392 * This is the first time the device driver
2393 * sees this request (possibly after
2394 * requeueing). Notify IO scheduler.
2396 if (rq
->cmd_flags
& REQ_SORTED
)
2397 elv_activate_rq(q
, rq
);
2400 * just mark as started even if we don't start
2401 * it, a request that has been delayed should
2402 * not be passed by new incoming requests
2404 rq
->cmd_flags
|= REQ_STARTED
;
2405 trace_block_rq_issue(q
, rq
);
2408 if (!q
->boundary_rq
|| q
->boundary_rq
== rq
) {
2409 q
->end_sector
= rq_end_sector(rq
);
2410 q
->boundary_rq
= NULL
;
2413 if (rq
->cmd_flags
& REQ_DONTPREP
)
2416 if (q
->dma_drain_size
&& blk_rq_bytes(rq
)) {
2418 * make sure space for the drain appears we
2419 * know we can do this because max_hw_segments
2420 * has been adjusted to be one fewer than the
2423 rq
->nr_phys_segments
++;
2429 ret
= q
->prep_rq_fn(q
, rq
);
2430 if (ret
== BLKPREP_OK
) {
2432 } else if (ret
== BLKPREP_DEFER
) {
2434 * the request may have been (partially) prepped.
2435 * we need to keep this request in the front to
2436 * avoid resource deadlock. REQ_STARTED will
2437 * prevent other fs requests from passing this one.
2439 if (q
->dma_drain_size
&& blk_rq_bytes(rq
) &&
2440 !(rq
->cmd_flags
& REQ_DONTPREP
)) {
2442 * remove the space for the drain we added
2443 * so that we don't add it again
2445 --rq
->nr_phys_segments
;
2450 } else if (ret
== BLKPREP_KILL
) {
2451 rq
->cmd_flags
|= REQ_QUIET
;
2453 * Mark this request as started so we don't trigger
2454 * any debug logic in the end I/O path.
2456 blk_start_request(rq
);
2457 __blk_end_request_all(rq
, -EIO
);
2459 printk(KERN_ERR
"%s: bad return=%d\n", __func__
, ret
);
2466 EXPORT_SYMBOL(blk_peek_request
);
2468 void blk_dequeue_request(struct request
*rq
)
2470 struct request_queue
*q
= rq
->q
;
2472 BUG_ON(list_empty(&rq
->queuelist
));
2473 BUG_ON(ELV_ON_HASH(rq
));
2475 list_del_init(&rq
->queuelist
);
2478 * the time frame between a request being removed from the lists
2479 * and to it is freed is accounted as io that is in progress at
2482 if (blk_account_rq(rq
)) {
2483 q
->in_flight
[rq_is_sync(rq
)]++;
2484 set_io_start_time_ns(rq
);
2489 * blk_start_request - start request processing on the driver
2490 * @req: request to dequeue
2493 * Dequeue @req and start timeout timer on it. This hands off the
2494 * request to the driver.
2496 * Block internal functions which don't want to start timer should
2497 * call blk_dequeue_request().
2500 * queue_lock must be held.
2502 void blk_start_request(struct request
*req
)
2504 blk_dequeue_request(req
);
2507 * We are now handing the request to the hardware, initialize
2508 * resid_len to full count and add the timeout handler.
2510 req
->resid_len
= blk_rq_bytes(req
);
2511 if (unlikely(blk_bidi_rq(req
)))
2512 req
->next_rq
->resid_len
= blk_rq_bytes(req
->next_rq
);
2514 BUG_ON(test_bit(REQ_ATOM_COMPLETE
, &req
->atomic_flags
));
2517 EXPORT_SYMBOL(blk_start_request
);
2520 * blk_fetch_request - fetch a request from a request queue
2521 * @q: request queue to fetch a request from
2524 * Return the request at the top of @q. The request is started on
2525 * return and LLD can start processing it immediately.
2528 * Pointer to the request at the top of @q if available. Null
2532 * queue_lock must be held.
2534 struct request
*blk_fetch_request(struct request_queue
*q
)
2538 rq
= blk_peek_request(q
);
2540 blk_start_request(rq
);
2543 EXPORT_SYMBOL(blk_fetch_request
);
2546 * blk_update_request - Special helper function for request stacking drivers
2547 * @req: the request being processed
2548 * @error: %0 for success, < %0 for error
2549 * @nr_bytes: number of bytes to complete @req
2552 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2553 * the request structure even if @req doesn't have leftover.
2554 * If @req has leftover, sets it up for the next range of segments.
2556 * This special helper function is only for request stacking drivers
2557 * (e.g. request-based dm) so that they can handle partial completion.
2558 * Actual device drivers should use blk_end_request instead.
2560 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2561 * %false return from this function.
2564 * %false - this request doesn't have any more data
2565 * %true - this request has more data
2567 bool blk_update_request(struct request
*req
, int error
, unsigned int nr_bytes
)
2571 trace_block_rq_complete(req
->q
, req
, nr_bytes
);
2577 * For fs requests, rq is just carrier of independent bio's
2578 * and each partial completion should be handled separately.
2579 * Reset per-request error on each partial completion.
2581 * TODO: tj: This is too subtle. It would be better to let
2582 * low level drivers do what they see fit.
2584 if (req
->cmd_type
== REQ_TYPE_FS
)
2587 if (error
&& req
->cmd_type
== REQ_TYPE_FS
&&
2588 !(req
->cmd_flags
& REQ_QUIET
)) {
2593 error_type
= "recoverable transport";
2596 error_type
= "critical target";
2599 error_type
= "critical nexus";
2602 error_type
= "timeout";
2605 error_type
= "critical space allocation";
2608 error_type
= "critical medium";
2615 printk_ratelimited(KERN_ERR
"%s: %s error, dev %s, sector %llu\n",
2616 __func__
, error_type
, req
->rq_disk
?
2617 req
->rq_disk
->disk_name
: "?",
2618 (unsigned long long)blk_rq_pos(req
));
2622 blk_account_io_completion(req
, nr_bytes
);
2626 struct bio
*bio
= req
->bio
;
2627 unsigned bio_bytes
= min(bio
->bi_iter
.bi_size
, nr_bytes
);
2629 if (bio_bytes
== bio
->bi_iter
.bi_size
)
2630 req
->bio
= bio
->bi_next
;
2632 req_bio_endio(req
, bio
, bio_bytes
, error
);
2634 total_bytes
+= bio_bytes
;
2635 nr_bytes
-= bio_bytes
;
2646 * Reset counters so that the request stacking driver
2647 * can find how many bytes remain in the request
2650 req
->__data_len
= 0;
2654 req
->__data_len
-= total_bytes
;
2656 /* update sector only for requests with clear definition of sector */
2657 if (req
->cmd_type
== REQ_TYPE_FS
)
2658 req
->__sector
+= total_bytes
>> 9;
2660 /* mixed attributes always follow the first bio */
2661 if (req
->cmd_flags
& REQ_MIXED_MERGE
) {
2662 req
->cmd_flags
&= ~REQ_FAILFAST_MASK
;
2663 req
->cmd_flags
|= req
->bio
->bi_rw
& REQ_FAILFAST_MASK
;
2667 * If total number of sectors is less than the first segment
2668 * size, something has gone terribly wrong.
2670 if (blk_rq_bytes(req
) < blk_rq_cur_bytes(req
)) {
2671 blk_dump_rq_flags(req
, "request botched");
2672 req
->__data_len
= blk_rq_cur_bytes(req
);
2675 /* recalculate the number of segments */
2676 blk_recalc_rq_segments(req
);
2680 EXPORT_SYMBOL_GPL(blk_update_request
);
2682 static bool blk_update_bidi_request(struct request
*rq
, int error
,
2683 unsigned int nr_bytes
,
2684 unsigned int bidi_bytes
)
2686 if (blk_update_request(rq
, error
, nr_bytes
))
2689 /* Bidi request must be completed as a whole */
2690 if (unlikely(blk_bidi_rq(rq
)) &&
2691 blk_update_request(rq
->next_rq
, error
, bidi_bytes
))
2694 if (blk_queue_add_random(rq
->q
))
2695 add_disk_randomness(rq
->rq_disk
);
2701 * blk_unprep_request - unprepare a request
2704 * This function makes a request ready for complete resubmission (or
2705 * completion). It happens only after all error handling is complete,
2706 * so represents the appropriate moment to deallocate any resources
2707 * that were allocated to the request in the prep_rq_fn. The queue
2708 * lock is held when calling this.
2710 void blk_unprep_request(struct request
*req
)
2712 struct request_queue
*q
= req
->q
;
2714 req
->cmd_flags
&= ~REQ_DONTPREP
;
2715 if (q
->unprep_rq_fn
)
2716 q
->unprep_rq_fn(q
, req
);
2718 EXPORT_SYMBOL_GPL(blk_unprep_request
);
2721 * queue lock must be held
2723 void blk_finish_request(struct request
*req
, int error
)
2725 if (req
->cmd_flags
& REQ_QUEUED
)
2726 blk_queue_end_tag(req
->q
, req
);
2728 BUG_ON(blk_queued_rq(req
));
2730 if (unlikely(laptop_mode
) && req
->cmd_type
== REQ_TYPE_FS
)
2731 laptop_io_completion(&req
->q
->backing_dev_info
);
2733 blk_delete_timer(req
);
2735 if (req
->cmd_flags
& REQ_DONTPREP
)
2736 blk_unprep_request(req
);
2738 blk_account_io_done(req
);
2741 req
->end_io(req
, error
);
2743 if (blk_bidi_rq(req
))
2744 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
2746 __blk_put_request(req
->q
, req
);
2749 EXPORT_SYMBOL(blk_finish_request
);
2752 * blk_end_bidi_request - Complete a bidi request
2753 * @rq: the request to complete
2754 * @error: %0 for success, < %0 for error
2755 * @nr_bytes: number of bytes to complete @rq
2756 * @bidi_bytes: number of bytes to complete @rq->next_rq
2759 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2760 * Drivers that supports bidi can safely call this member for any
2761 * type of request, bidi or uni. In the later case @bidi_bytes is
2765 * %false - we are done with this request
2766 * %true - still buffers pending for this request
2768 static bool blk_end_bidi_request(struct request
*rq
, int error
,
2769 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2771 struct request_queue
*q
= rq
->q
;
2772 unsigned long flags
;
2774 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2777 spin_lock_irqsave(q
->queue_lock
, flags
);
2778 blk_finish_request(rq
, error
);
2779 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2785 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2786 * @rq: the request to complete
2787 * @error: %0 for success, < %0 for error
2788 * @nr_bytes: number of bytes to complete @rq
2789 * @bidi_bytes: number of bytes to complete @rq->next_rq
2792 * Identical to blk_end_bidi_request() except that queue lock is
2793 * assumed to be locked on entry and remains so on return.
2796 * %false - we are done with this request
2797 * %true - still buffers pending for this request
2799 bool __blk_end_bidi_request(struct request
*rq
, int error
,
2800 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2802 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2805 blk_finish_request(rq
, error
);
2811 * blk_end_request - Helper function for drivers to complete the request.
2812 * @rq: the request being processed
2813 * @error: %0 for success, < %0 for error
2814 * @nr_bytes: number of bytes to complete
2817 * Ends I/O on a number of bytes attached to @rq.
2818 * If @rq has leftover, sets it up for the next range of segments.
2821 * %false - we are done with this request
2822 * %true - still buffers pending for this request
2824 bool blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2826 return blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2828 EXPORT_SYMBOL(blk_end_request
);
2831 * blk_end_request_all - Helper function for drives to finish the request.
2832 * @rq: the request to finish
2833 * @error: %0 for success, < %0 for error
2836 * Completely finish @rq.
2838 void blk_end_request_all(struct request
*rq
, int error
)
2841 unsigned int bidi_bytes
= 0;
2843 if (unlikely(blk_bidi_rq(rq
)))
2844 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2846 pending
= blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2849 EXPORT_SYMBOL(blk_end_request_all
);
2852 * blk_end_request_cur - Helper function to finish the current request chunk.
2853 * @rq: the request to finish the current chunk for
2854 * @error: %0 for success, < %0 for error
2857 * Complete the current consecutively mapped chunk from @rq.
2860 * %false - we are done with this request
2861 * %true - still buffers pending for this request
2863 bool blk_end_request_cur(struct request
*rq
, int error
)
2865 return blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2867 EXPORT_SYMBOL(blk_end_request_cur
);
2870 * blk_end_request_err - Finish a request till the next failure boundary.
2871 * @rq: the request to finish till the next failure boundary for
2872 * @error: must be negative errno
2875 * Complete @rq till the next failure boundary.
2878 * %false - we are done with this request
2879 * %true - still buffers pending for this request
2881 bool blk_end_request_err(struct request
*rq
, int error
)
2883 WARN_ON(error
>= 0);
2884 return blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2886 EXPORT_SYMBOL_GPL(blk_end_request_err
);
2889 * __blk_end_request - Helper function for drivers to complete the request.
2890 * @rq: the request being processed
2891 * @error: %0 for success, < %0 for error
2892 * @nr_bytes: number of bytes to complete
2895 * Must be called with queue lock held unlike blk_end_request().
2898 * %false - we are done with this request
2899 * %true - still buffers pending for this request
2901 bool __blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2903 return __blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2905 EXPORT_SYMBOL(__blk_end_request
);
2908 * __blk_end_request_all - Helper function for drives to finish the request.
2909 * @rq: the request to finish
2910 * @error: %0 for success, < %0 for error
2913 * Completely finish @rq. Must be called with queue lock held.
2915 void __blk_end_request_all(struct request
*rq
, int error
)
2918 unsigned int bidi_bytes
= 0;
2920 if (unlikely(blk_bidi_rq(rq
)))
2921 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2923 pending
= __blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2926 EXPORT_SYMBOL(__blk_end_request_all
);
2929 * __blk_end_request_cur - Helper function to finish the current request chunk.
2930 * @rq: the request to finish the current chunk for
2931 * @error: %0 for success, < %0 for error
2934 * Complete the current consecutively mapped chunk from @rq. Must
2935 * be called with queue lock held.
2938 * %false - we are done with this request
2939 * %true - still buffers pending for this request
2941 bool __blk_end_request_cur(struct request
*rq
, int error
)
2943 return __blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2945 EXPORT_SYMBOL(__blk_end_request_cur
);
2948 * __blk_end_request_err - Finish a request till the next failure boundary.
2949 * @rq: the request to finish till the next failure boundary for
2950 * @error: must be negative errno
2953 * Complete @rq till the next failure boundary. Must be called
2954 * with queue lock held.
2957 * %false - we are done with this request
2958 * %true - still buffers pending for this request
2960 bool __blk_end_request_err(struct request
*rq
, int error
)
2962 WARN_ON(error
>= 0);
2963 return __blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2965 EXPORT_SYMBOL_GPL(__blk_end_request_err
);
2967 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
2970 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2971 rq
->cmd_flags
|= bio
->bi_rw
& REQ_WRITE
;
2973 if (bio_has_data(bio
))
2974 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2976 rq
->__data_len
= bio
->bi_iter
.bi_size
;
2977 rq
->bio
= rq
->biotail
= bio
;
2980 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2983 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2985 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2986 * @rq: the request to be flushed
2989 * Flush all pages in @rq.
2991 void rq_flush_dcache_pages(struct request
*rq
)
2993 struct req_iterator iter
;
2994 struct bio_vec bvec
;
2996 rq_for_each_segment(bvec
, rq
, iter
)
2997 flush_dcache_page(bvec
.bv_page
);
2999 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages
);
3003 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
3004 * @q : the queue of the device being checked
3007 * Check if underlying low-level drivers of a device are busy.
3008 * If the drivers want to export their busy state, they must set own
3009 * exporting function using blk_queue_lld_busy() first.
3011 * Basically, this function is used only by request stacking drivers
3012 * to stop dispatching requests to underlying devices when underlying
3013 * devices are busy. This behavior helps more I/O merging on the queue
3014 * of the request stacking driver and prevents I/O throughput regression
3015 * on burst I/O load.
3018 * 0 - Not busy (The request stacking driver should dispatch request)
3019 * 1 - Busy (The request stacking driver should stop dispatching request)
3021 int blk_lld_busy(struct request_queue
*q
)
3024 return q
->lld_busy_fn(q
);
3028 EXPORT_SYMBOL_GPL(blk_lld_busy
);
3031 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
3032 * @rq: the clone request to be cleaned up
3035 * Free all bios in @rq for a cloned request.
3037 void blk_rq_unprep_clone(struct request
*rq
)
3041 while ((bio
= rq
->bio
) != NULL
) {
3042 rq
->bio
= bio
->bi_next
;
3047 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone
);
3050 * Copy attributes of the original request to the clone request.
3051 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3053 static void __blk_rq_prep_clone(struct request
*dst
, struct request
*src
)
3055 dst
->cpu
= src
->cpu
;
3056 dst
->cmd_flags
|= (src
->cmd_flags
& REQ_CLONE_MASK
) | REQ_NOMERGE
;
3057 dst
->cmd_type
= src
->cmd_type
;
3058 dst
->__sector
= blk_rq_pos(src
);
3059 dst
->__data_len
= blk_rq_bytes(src
);
3060 dst
->nr_phys_segments
= src
->nr_phys_segments
;
3061 dst
->ioprio
= src
->ioprio
;
3062 dst
->extra_len
= src
->extra_len
;
3066 * blk_rq_prep_clone - Helper function to setup clone request
3067 * @rq: the request to be setup
3068 * @rq_src: original request to be cloned
3069 * @bs: bio_set that bios for clone are allocated from
3070 * @gfp_mask: memory allocation mask for bio
3071 * @bio_ctr: setup function to be called for each clone bio.
3072 * Returns %0 for success, non %0 for failure.
3073 * @data: private data to be passed to @bio_ctr
3076 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3077 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3078 * are not copied, and copying such parts is the caller's responsibility.
3079 * Also, pages which the original bios are pointing to are not copied
3080 * and the cloned bios just point same pages.
3081 * So cloned bios must be completed before original bios, which means
3082 * the caller must complete @rq before @rq_src.
3084 int blk_rq_prep_clone(struct request
*rq
, struct request
*rq_src
,
3085 struct bio_set
*bs
, gfp_t gfp_mask
,
3086 int (*bio_ctr
)(struct bio
*, struct bio
*, void *),
3089 struct bio
*bio
, *bio_src
;
3094 __rq_for_each_bio(bio_src
, rq_src
) {
3095 bio
= bio_clone_fast(bio_src
, gfp_mask
, bs
);
3099 if (bio_ctr
&& bio_ctr(bio
, bio_src
, data
))
3103 rq
->biotail
->bi_next
= bio
;
3106 rq
->bio
= rq
->biotail
= bio
;
3109 __blk_rq_prep_clone(rq
, rq_src
);
3116 blk_rq_unprep_clone(rq
);
3120 EXPORT_SYMBOL_GPL(blk_rq_prep_clone
);
3122 int kblockd_schedule_work(struct work_struct
*work
)
3124 return queue_work(kblockd_workqueue
, work
);
3126 EXPORT_SYMBOL(kblockd_schedule_work
);
3128 int kblockd_schedule_delayed_work(struct delayed_work
*dwork
,
3129 unsigned long delay
)
3131 return queue_delayed_work(kblockd_workqueue
, dwork
, delay
);
3133 EXPORT_SYMBOL(kblockd_schedule_delayed_work
);
3135 int kblockd_schedule_delayed_work_on(int cpu
, struct delayed_work
*dwork
,
3136 unsigned long delay
)
3138 return queue_delayed_work_on(cpu
, kblockd_workqueue
, dwork
, delay
);
3140 EXPORT_SYMBOL(kblockd_schedule_delayed_work_on
);
3143 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3144 * @plug: The &struct blk_plug that needs to be initialized
3147 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3148 * pending I/O should the task end up blocking between blk_start_plug() and
3149 * blk_finish_plug(). This is important from a performance perspective, but
3150 * also ensures that we don't deadlock. For instance, if the task is blocking
3151 * for a memory allocation, memory reclaim could end up wanting to free a
3152 * page belonging to that request that is currently residing in our private
3153 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3154 * this kind of deadlock.
3156 void blk_start_plug(struct blk_plug
*plug
)
3158 struct task_struct
*tsk
= current
;
3161 * If this is a nested plug, don't actually assign it.
3166 INIT_LIST_HEAD(&plug
->list
);
3167 INIT_LIST_HEAD(&plug
->mq_list
);
3168 INIT_LIST_HEAD(&plug
->cb_list
);
3170 * Store ordering should not be needed here, since a potential
3171 * preempt will imply a full memory barrier
3175 EXPORT_SYMBOL(blk_start_plug
);
3177 static int plug_rq_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
3179 struct request
*rqa
= container_of(a
, struct request
, queuelist
);
3180 struct request
*rqb
= container_of(b
, struct request
, queuelist
);
3182 return !(rqa
->q
< rqb
->q
||
3183 (rqa
->q
== rqb
->q
&& blk_rq_pos(rqa
) < blk_rq_pos(rqb
)));
3187 * If 'from_schedule' is true, then postpone the dispatch of requests
3188 * until a safe kblockd context. We due this to avoid accidental big
3189 * additional stack usage in driver dispatch, in places where the originally
3190 * plugger did not intend it.
3192 static void queue_unplugged(struct request_queue
*q
, unsigned int depth
,
3194 __releases(q
->queue_lock
)
3196 trace_block_unplug(q
, depth
, !from_schedule
);
3199 blk_run_queue_async(q
);
3202 spin_unlock(q
->queue_lock
);
3205 static void flush_plug_callbacks(struct blk_plug
*plug
, bool from_schedule
)
3207 LIST_HEAD(callbacks
);
3209 while (!list_empty(&plug
->cb_list
)) {
3210 list_splice_init(&plug
->cb_list
, &callbacks
);
3212 while (!list_empty(&callbacks
)) {
3213 struct blk_plug_cb
*cb
= list_first_entry(&callbacks
,
3216 list_del(&cb
->list
);
3217 cb
->callback(cb
, from_schedule
);
3222 struct blk_plug_cb
*blk_check_plugged(blk_plug_cb_fn unplug
, void *data
,
3225 struct blk_plug
*plug
= current
->plug
;
3226 struct blk_plug_cb
*cb
;
3231 list_for_each_entry(cb
, &plug
->cb_list
, list
)
3232 if (cb
->callback
== unplug
&& cb
->data
== data
)
3235 /* Not currently on the callback list */
3236 BUG_ON(size
< sizeof(*cb
));
3237 cb
= kzalloc(size
, GFP_ATOMIC
);
3240 cb
->callback
= unplug
;
3241 list_add(&cb
->list
, &plug
->cb_list
);
3245 EXPORT_SYMBOL(blk_check_plugged
);
3247 void blk_flush_plug_list(struct blk_plug
*plug
, bool from_schedule
)
3249 struct request_queue
*q
;
3250 unsigned long flags
;
3255 flush_plug_callbacks(plug
, from_schedule
);
3257 if (!list_empty(&plug
->mq_list
))
3258 blk_mq_flush_plug_list(plug
, from_schedule
);
3260 if (list_empty(&plug
->list
))
3263 list_splice_init(&plug
->list
, &list
);
3265 list_sort(NULL
, &list
, plug_rq_cmp
);
3271 * Save and disable interrupts here, to avoid doing it for every
3272 * queue lock we have to take.
3274 local_irq_save(flags
);
3275 while (!list_empty(&list
)) {
3276 rq
= list_entry_rq(list
.next
);
3277 list_del_init(&rq
->queuelist
);
3281 * This drops the queue lock
3284 queue_unplugged(q
, depth
, from_schedule
);
3287 spin_lock(q
->queue_lock
);
3291 * Short-circuit if @q is dead
3293 if (unlikely(blk_queue_dying(q
))) {
3294 __blk_end_request_all(rq
, -ENODEV
);
3299 * rq is already accounted, so use raw insert
3301 if (rq
->cmd_flags
& (REQ_FLUSH
| REQ_FUA
))
3302 __elv_add_request(q
, rq
, ELEVATOR_INSERT_FLUSH
);
3304 __elv_add_request(q
, rq
, ELEVATOR_INSERT_SORT_MERGE
);
3310 * This drops the queue lock
3313 queue_unplugged(q
, depth
, from_schedule
);
3315 local_irq_restore(flags
);
3318 void blk_finish_plug(struct blk_plug
*plug
)
3320 if (plug
!= current
->plug
)
3322 blk_flush_plug_list(plug
, false);
3324 current
->plug
= NULL
;
3326 EXPORT_SYMBOL(blk_finish_plug
);
3328 bool blk_poll(struct request_queue
*q
, blk_qc_t cookie
)
3330 struct blk_plug
*plug
;
3333 if (!q
->mq_ops
|| !q
->mq_ops
->poll
|| !blk_qc_t_valid(cookie
) ||
3334 !test_bit(QUEUE_FLAG_POLL
, &q
->queue_flags
))
3337 plug
= current
->plug
;
3339 blk_flush_plug_list(plug
, false);
3341 state
= current
->state
;
3342 while (!need_resched()) {
3343 unsigned int queue_num
= blk_qc_t_to_queue_num(cookie
);
3344 struct blk_mq_hw_ctx
*hctx
= q
->queue_hw_ctx
[queue_num
];
3347 hctx
->poll_invoked
++;
3349 ret
= q
->mq_ops
->poll(hctx
, blk_qc_t_to_tag(cookie
));
3351 hctx
->poll_success
++;
3352 set_current_state(TASK_RUNNING
);
3356 if (signal_pending_state(state
, current
))
3357 set_current_state(TASK_RUNNING
);
3359 if (current
->state
== TASK_RUNNING
)
3371 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3372 * @q: the queue of the device
3373 * @dev: the device the queue belongs to
3376 * Initialize runtime-PM-related fields for @q and start auto suspend for
3377 * @dev. Drivers that want to take advantage of request-based runtime PM
3378 * should call this function after @dev has been initialized, and its
3379 * request queue @q has been allocated, and runtime PM for it can not happen
3380 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3381 * cases, driver should call this function before any I/O has taken place.
3383 * This function takes care of setting up using auto suspend for the device,
3384 * the autosuspend delay is set to -1 to make runtime suspend impossible
3385 * until an updated value is either set by user or by driver. Drivers do
3386 * not need to touch other autosuspend settings.
3388 * The block layer runtime PM is request based, so only works for drivers
3389 * that use request as their IO unit instead of those directly use bio's.
3391 void blk_pm_runtime_init(struct request_queue
*q
, struct device
*dev
)
3394 q
->rpm_status
= RPM_ACTIVE
;
3395 pm_runtime_set_autosuspend_delay(q
->dev
, -1);
3396 pm_runtime_use_autosuspend(q
->dev
);
3398 EXPORT_SYMBOL(blk_pm_runtime_init
);
3401 * blk_pre_runtime_suspend - Pre runtime suspend check
3402 * @q: the queue of the device
3405 * This function will check if runtime suspend is allowed for the device
3406 * by examining if there are any requests pending in the queue. If there
3407 * are requests pending, the device can not be runtime suspended; otherwise,
3408 * the queue's status will be updated to SUSPENDING and the driver can
3409 * proceed to suspend the device.
3411 * For the not allowed case, we mark last busy for the device so that
3412 * runtime PM core will try to autosuspend it some time later.
3414 * This function should be called near the start of the device's
3415 * runtime_suspend callback.
3418 * 0 - OK to runtime suspend the device
3419 * -EBUSY - Device should not be runtime suspended
3421 int blk_pre_runtime_suspend(struct request_queue
*q
)
3428 spin_lock_irq(q
->queue_lock
);
3429 if (q
->nr_pending
) {
3431 pm_runtime_mark_last_busy(q
->dev
);
3433 q
->rpm_status
= RPM_SUSPENDING
;
3435 spin_unlock_irq(q
->queue_lock
);
3438 EXPORT_SYMBOL(blk_pre_runtime_suspend
);
3441 * blk_post_runtime_suspend - Post runtime suspend processing
3442 * @q: the queue of the device
3443 * @err: return value of the device's runtime_suspend function
3446 * Update the queue's runtime status according to the return value of the
3447 * device's runtime suspend function and mark last busy for the device so
3448 * that PM core will try to auto suspend the device at a later time.
3450 * This function should be called near the end of the device's
3451 * runtime_suspend callback.
3453 void blk_post_runtime_suspend(struct request_queue
*q
, int err
)
3458 spin_lock_irq(q
->queue_lock
);
3460 q
->rpm_status
= RPM_SUSPENDED
;
3462 q
->rpm_status
= RPM_ACTIVE
;
3463 pm_runtime_mark_last_busy(q
->dev
);
3465 spin_unlock_irq(q
->queue_lock
);
3467 EXPORT_SYMBOL(blk_post_runtime_suspend
);
3470 * blk_pre_runtime_resume - Pre runtime resume processing
3471 * @q: the queue of the device
3474 * Update the queue's runtime status to RESUMING in preparation for the
3475 * runtime resume of the device.
3477 * This function should be called near the start of the device's
3478 * runtime_resume callback.
3480 void blk_pre_runtime_resume(struct request_queue
*q
)
3485 spin_lock_irq(q
->queue_lock
);
3486 q
->rpm_status
= RPM_RESUMING
;
3487 spin_unlock_irq(q
->queue_lock
);
3489 EXPORT_SYMBOL(blk_pre_runtime_resume
);
3492 * blk_post_runtime_resume - Post runtime resume processing
3493 * @q: the queue of the device
3494 * @err: return value of the device's runtime_resume function
3497 * Update the queue's runtime status according to the return value of the
3498 * device's runtime_resume function. If it is successfully resumed, process
3499 * the requests that are queued into the device's queue when it is resuming
3500 * and then mark last busy and initiate autosuspend for it.
3502 * This function should be called near the end of the device's
3503 * runtime_resume callback.
3505 void blk_post_runtime_resume(struct request_queue
*q
, int err
)
3510 spin_lock_irq(q
->queue_lock
);
3512 q
->rpm_status
= RPM_ACTIVE
;
3514 pm_runtime_mark_last_busy(q
->dev
);
3515 pm_request_autosuspend(q
->dev
);
3517 q
->rpm_status
= RPM_SUSPENDED
;
3519 spin_unlock_irq(q
->queue_lock
);
3521 EXPORT_SYMBOL(blk_post_runtime_resume
);
3524 int __init
blk_dev_init(void)
3526 BUILD_BUG_ON(__REQ_NR_BITS
> 8 *
3527 FIELD_SIZEOF(struct request
, cmd_flags
));
3529 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3530 kblockd_workqueue
= alloc_workqueue("kblockd",
3531 WQ_MEM_RECLAIM
| WQ_HIGHPRI
, 0);
3532 if (!kblockd_workqueue
)
3533 panic("Failed to create kblockd\n");
3535 request_cachep
= kmem_cache_create("blkdev_requests",
3536 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
3538 blk_requestq_cachep
= kmem_cache_create("request_queue",
3539 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);