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/highmem.h>
21 #include <linux/kernel_stat.h>
22 #include <linux/string.h>
23 #include <linux/init.h>
24 #include <linux/completion.h>
25 #include <linux/slab.h>
26 #include <linux/swap.h>
27 #include <linux/writeback.h>
28 #include <linux/task_io_accounting_ops.h>
29 #include <linux/fault-inject.h>
30 #include <linux/list_sort.h>
31 #include <linux/delay.h>
32 #include <linux/ratelimit.h>
34 #define CREATE_TRACE_POINTS
35 #include <trace/events/block.h>
38 #include "blk-cgroup.h"
40 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap
);
41 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap
);
42 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete
);
44 DEFINE_IDA(blk_queue_ida
);
47 * For the allocated request tables
49 static struct kmem_cache
*request_cachep
;
52 * For queue allocation
54 struct kmem_cache
*blk_requestq_cachep
;
57 * Controlling structure to kblockd
59 static struct workqueue_struct
*kblockd_workqueue
;
61 static void drive_stat_acct(struct request
*rq
, int new_io
)
63 struct hd_struct
*part
;
64 int rw
= rq_data_dir(rq
);
67 if (!blk_do_io_stat(rq
))
70 cpu
= part_stat_lock();
74 part_stat_inc(cpu
, part
, merges
[rw
]);
76 part
= disk_map_sector_rcu(rq
->rq_disk
, blk_rq_pos(rq
));
77 if (!hd_struct_try_get(part
)) {
79 * The partition is already being removed,
80 * the request will be accounted on the disk only
82 * We take a reference on disk->part0 although that
83 * partition will never be deleted, so we can treat
84 * it as any other partition.
86 part
= &rq
->rq_disk
->part0
;
89 part_round_stats(cpu
, part
);
90 part_inc_in_flight(part
, rw
);
97 void blk_queue_congestion_threshold(struct request_queue
*q
)
101 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) + 1;
102 if (nr
> q
->nr_requests
)
104 q
->nr_congestion_on
= nr
;
106 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) - (q
->nr_requests
/ 16) - 1;
109 q
->nr_congestion_off
= nr
;
113 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
116 * Locates the passed device's request queue and returns the address of its
119 * Will return NULL if the request queue cannot be located.
121 struct backing_dev_info
*blk_get_backing_dev_info(struct block_device
*bdev
)
123 struct backing_dev_info
*ret
= NULL
;
124 struct request_queue
*q
= bdev_get_queue(bdev
);
127 ret
= &q
->backing_dev_info
;
130 EXPORT_SYMBOL(blk_get_backing_dev_info
);
132 void blk_rq_init(struct request_queue
*q
, struct request
*rq
)
134 memset(rq
, 0, sizeof(*rq
));
136 INIT_LIST_HEAD(&rq
->queuelist
);
137 INIT_LIST_HEAD(&rq
->timeout_list
);
140 rq
->__sector
= (sector_t
) -1;
141 INIT_HLIST_NODE(&rq
->hash
);
142 RB_CLEAR_NODE(&rq
->rb_node
);
144 rq
->cmd_len
= BLK_MAX_CDB
;
147 rq
->start_time
= jiffies
;
148 set_start_time_ns(rq
);
151 EXPORT_SYMBOL(blk_rq_init
);
153 static void req_bio_endio(struct request
*rq
, struct bio
*bio
,
154 unsigned int nbytes
, int error
)
157 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
158 else if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
161 if (unlikely(nbytes
> bio
->bi_size
)) {
162 printk(KERN_ERR
"%s: want %u bytes done, %u left\n",
163 __func__
, nbytes
, bio
->bi_size
);
164 nbytes
= bio
->bi_size
;
167 if (unlikely(rq
->cmd_flags
& REQ_QUIET
))
168 set_bit(BIO_QUIET
, &bio
->bi_flags
);
170 bio
->bi_size
-= nbytes
;
171 bio
->bi_sector
+= (nbytes
>> 9);
173 if (bio_integrity(bio
))
174 bio_integrity_advance(bio
, nbytes
);
176 /* don't actually finish bio if it's part of flush sequence */
177 if (bio
->bi_size
== 0 && !(rq
->cmd_flags
& REQ_FLUSH_SEQ
))
178 bio_endio(bio
, error
);
181 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
185 printk(KERN_INFO
"%s: dev %s: type=%x, flags=%x\n", msg
,
186 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->cmd_type
,
189 printk(KERN_INFO
" sector %llu, nr/cnr %u/%u\n",
190 (unsigned long long)blk_rq_pos(rq
),
191 blk_rq_sectors(rq
), blk_rq_cur_sectors(rq
));
192 printk(KERN_INFO
" bio %p, biotail %p, buffer %p, len %u\n",
193 rq
->bio
, rq
->biotail
, rq
->buffer
, blk_rq_bytes(rq
));
195 if (rq
->cmd_type
== REQ_TYPE_BLOCK_PC
) {
196 printk(KERN_INFO
" cdb: ");
197 for (bit
= 0; bit
< BLK_MAX_CDB
; bit
++)
198 printk("%02x ", rq
->cmd
[bit
]);
202 EXPORT_SYMBOL(blk_dump_rq_flags
);
204 static void blk_delay_work(struct work_struct
*work
)
206 struct request_queue
*q
;
208 q
= container_of(work
, struct request_queue
, delay_work
.work
);
209 spin_lock_irq(q
->queue_lock
);
211 spin_unlock_irq(q
->queue_lock
);
215 * blk_delay_queue - restart queueing after defined interval
216 * @q: The &struct request_queue in question
217 * @msecs: Delay in msecs
220 * Sometimes queueing needs to be postponed for a little while, to allow
221 * resources to come back. This function will make sure that queueing is
222 * restarted around the specified time.
224 void blk_delay_queue(struct request_queue
*q
, unsigned long msecs
)
226 queue_delayed_work(kblockd_workqueue
, &q
->delay_work
,
227 msecs_to_jiffies(msecs
));
229 EXPORT_SYMBOL(blk_delay_queue
);
232 * blk_start_queue - restart a previously stopped queue
233 * @q: The &struct request_queue in question
236 * blk_start_queue() will clear the stop flag on the queue, and call
237 * the request_fn for the queue if it was in a stopped state when
238 * entered. Also see blk_stop_queue(). Queue lock must be held.
240 void blk_start_queue(struct request_queue
*q
)
242 WARN_ON(!irqs_disabled());
244 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
247 EXPORT_SYMBOL(blk_start_queue
);
250 * blk_stop_queue - stop a queue
251 * @q: The &struct request_queue in question
254 * The Linux block layer assumes that a block driver will consume all
255 * entries on the request queue when the request_fn strategy is called.
256 * Often this will not happen, because of hardware limitations (queue
257 * depth settings). If a device driver gets a 'queue full' response,
258 * or if it simply chooses not to queue more I/O at one point, it can
259 * call this function to prevent the request_fn from being called until
260 * the driver has signalled it's ready to go again. This happens by calling
261 * blk_start_queue() to restart queue operations. Queue lock must be held.
263 void blk_stop_queue(struct request_queue
*q
)
265 cancel_delayed_work(&q
->delay_work
);
266 queue_flag_set(QUEUE_FLAG_STOPPED
, q
);
268 EXPORT_SYMBOL(blk_stop_queue
);
271 * blk_sync_queue - cancel any pending callbacks on a queue
275 * The block layer may perform asynchronous callback activity
276 * on a queue, such as calling the unplug function after a timeout.
277 * A block device may call blk_sync_queue to ensure that any
278 * such activity is cancelled, thus allowing it to release resources
279 * that the callbacks might use. The caller must already have made sure
280 * that its ->make_request_fn will not re-add plugging prior to calling
283 * This function does not cancel any asynchronous activity arising
284 * out of elevator or throttling code. That would require elevaotor_exit()
285 * and blkcg_exit_queue() to be called with queue lock initialized.
288 void blk_sync_queue(struct request_queue
*q
)
290 del_timer_sync(&q
->timeout
);
291 cancel_delayed_work_sync(&q
->delay_work
);
293 EXPORT_SYMBOL(blk_sync_queue
);
296 * __blk_run_queue - run a single device queue
297 * @q: The queue to run
300 * See @blk_run_queue. This variant must be called with the queue lock
301 * held and interrupts disabled.
303 void __blk_run_queue(struct request_queue
*q
)
305 if (unlikely(blk_queue_stopped(q
)))
310 EXPORT_SYMBOL(__blk_run_queue
);
313 * blk_run_queue_async - run a single device queue in workqueue context
314 * @q: The queue to run
317 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
320 void blk_run_queue_async(struct request_queue
*q
)
322 if (likely(!blk_queue_stopped(q
)))
323 mod_delayed_work(kblockd_workqueue
, &q
->delay_work
, 0);
325 EXPORT_SYMBOL(blk_run_queue_async
);
328 * blk_run_queue - run a single device queue
329 * @q: The queue to run
332 * Invoke request handling on this queue, if it has pending work to do.
333 * May be used to restart queueing when a request has completed.
335 void blk_run_queue(struct request_queue
*q
)
339 spin_lock_irqsave(q
->queue_lock
, flags
);
341 spin_unlock_irqrestore(q
->queue_lock
, flags
);
343 EXPORT_SYMBOL(blk_run_queue
);
345 void blk_put_queue(struct request_queue
*q
)
347 kobject_put(&q
->kobj
);
349 EXPORT_SYMBOL(blk_put_queue
);
352 * blk_drain_queue - drain requests from request_queue
354 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
356 * Drain requests from @q. If @drain_all is set, all requests are drained.
357 * If not, only ELVPRIV requests are drained. The caller is responsible
358 * for ensuring that no new requests which need to be drained are queued.
360 void blk_drain_queue(struct request_queue
*q
, bool drain_all
)
367 spin_lock_irq(q
->queue_lock
);
370 * The caller might be trying to drain @q before its
371 * elevator is initialized.
374 elv_drain_elevator(q
);
376 blkcg_drain_queue(q
);
379 * This function might be called on a queue which failed
380 * driver init after queue creation or is not yet fully
381 * active yet. Some drivers (e.g. fd and loop) get unhappy
382 * in such cases. Kick queue iff dispatch queue has
383 * something on it and @q has request_fn set.
385 if (!list_empty(&q
->queue_head
) && q
->request_fn
)
388 drain
|= q
->nr_rqs_elvpriv
;
391 * Unfortunately, requests are queued at and tracked from
392 * multiple places and there's no single counter which can
393 * be drained. Check all the queues and counters.
396 drain
|= !list_empty(&q
->queue_head
);
397 for (i
= 0; i
< 2; i
++) {
398 drain
|= q
->nr_rqs
[i
];
399 drain
|= q
->in_flight
[i
];
400 drain
|= !list_empty(&q
->flush_queue
[i
]);
404 spin_unlock_irq(q
->queue_lock
);
412 * With queue marked dead, any woken up waiter will fail the
413 * allocation path, so the wakeup chaining is lost and we're
414 * left with hung waiters. We need to wake up those waiters.
417 struct request_list
*rl
;
419 spin_lock_irq(q
->queue_lock
);
421 blk_queue_for_each_rl(rl
, q
)
422 for (i
= 0; i
< ARRAY_SIZE(rl
->wait
); i
++)
423 wake_up_all(&rl
->wait
[i
]);
425 spin_unlock_irq(q
->queue_lock
);
430 * blk_queue_bypass_start - enter queue bypass mode
431 * @q: queue of interest
433 * In bypass mode, only the dispatch FIFO queue of @q is used. This
434 * function makes @q enter bypass mode and drains all requests which were
435 * throttled or issued before. On return, it's guaranteed that no request
436 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
437 * inside queue or RCU read lock.
439 void blk_queue_bypass_start(struct request_queue
*q
)
443 spin_lock_irq(q
->queue_lock
);
444 drain
= !q
->bypass_depth
++;
445 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
446 spin_unlock_irq(q
->queue_lock
);
449 blk_drain_queue(q
, false);
450 /* ensure blk_queue_bypass() is %true inside RCU read lock */
454 EXPORT_SYMBOL_GPL(blk_queue_bypass_start
);
457 * blk_queue_bypass_end - leave queue bypass mode
458 * @q: queue of interest
460 * Leave bypass mode and restore the normal queueing behavior.
462 void blk_queue_bypass_end(struct request_queue
*q
)
464 spin_lock_irq(q
->queue_lock
);
465 if (!--q
->bypass_depth
)
466 queue_flag_clear(QUEUE_FLAG_BYPASS
, q
);
467 WARN_ON_ONCE(q
->bypass_depth
< 0);
468 spin_unlock_irq(q
->queue_lock
);
470 EXPORT_SYMBOL_GPL(blk_queue_bypass_end
);
473 * blk_cleanup_queue - shutdown a request queue
474 * @q: request queue to shutdown
476 * Mark @q DEAD, drain all pending requests, destroy and put it. All
477 * future requests will be failed immediately with -ENODEV.
479 void blk_cleanup_queue(struct request_queue
*q
)
481 spinlock_t
*lock
= q
->queue_lock
;
483 /* mark @q DEAD, no new request or merges will be allowed afterwards */
484 mutex_lock(&q
->sysfs_lock
);
485 queue_flag_set_unlocked(QUEUE_FLAG_DEAD
, q
);
489 * Dead queue is permanently in bypass mode till released. Note
490 * that, unlike blk_queue_bypass_start(), we aren't performing
491 * synchronize_rcu() after entering bypass mode to avoid the delay
492 * as some drivers create and destroy a lot of queues while
493 * probing. This is still safe because blk_release_queue() will be
494 * called only after the queue refcnt drops to zero and nothing,
495 * RCU or not, would be traversing the queue by then.
498 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
500 queue_flag_set(QUEUE_FLAG_NOMERGES
, q
);
501 queue_flag_set(QUEUE_FLAG_NOXMERGES
, q
);
502 queue_flag_set(QUEUE_FLAG_DEAD
, q
);
503 spin_unlock_irq(lock
);
504 mutex_unlock(&q
->sysfs_lock
);
506 /* drain all requests queued before DEAD marking */
507 blk_drain_queue(q
, true);
509 /* @q won't process any more request, flush async actions */
510 del_timer_sync(&q
->backing_dev_info
.laptop_mode_wb_timer
);
514 if (q
->queue_lock
!= &q
->__queue_lock
)
515 q
->queue_lock
= &q
->__queue_lock
;
516 spin_unlock_irq(lock
);
518 /* @q is and will stay empty, shutdown and put */
521 EXPORT_SYMBOL(blk_cleanup_queue
);
523 int blk_init_rl(struct request_list
*rl
, struct request_queue
*q
,
526 if (unlikely(rl
->rq_pool
))
530 rl
->count
[BLK_RW_SYNC
] = rl
->count
[BLK_RW_ASYNC
] = 0;
531 rl
->starved
[BLK_RW_SYNC
] = rl
->starved
[BLK_RW_ASYNC
] = 0;
532 init_waitqueue_head(&rl
->wait
[BLK_RW_SYNC
]);
533 init_waitqueue_head(&rl
->wait
[BLK_RW_ASYNC
]);
535 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, mempool_alloc_slab
,
536 mempool_free_slab
, request_cachep
,
544 void blk_exit_rl(struct request_list
*rl
)
547 mempool_destroy(rl
->rq_pool
);
550 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
552 return blk_alloc_queue_node(gfp_mask
, -1);
554 EXPORT_SYMBOL(blk_alloc_queue
);
556 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
558 struct request_queue
*q
;
561 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
562 gfp_mask
| __GFP_ZERO
, node_id
);
566 q
->id
= ida_simple_get(&blk_queue_ida
, 0, 0, gfp_mask
);
570 q
->backing_dev_info
.ra_pages
=
571 (VM_MAX_READAHEAD
* 1024) / PAGE_CACHE_SIZE
;
572 q
->backing_dev_info
.state
= 0;
573 q
->backing_dev_info
.capabilities
= BDI_CAP_MAP_COPY
;
574 q
->backing_dev_info
.name
= "block";
577 err
= bdi_init(&q
->backing_dev_info
);
581 setup_timer(&q
->backing_dev_info
.laptop_mode_wb_timer
,
582 laptop_mode_timer_fn
, (unsigned long) q
);
583 setup_timer(&q
->timeout
, blk_rq_timed_out_timer
, (unsigned long) q
);
584 INIT_LIST_HEAD(&q
->queue_head
);
585 INIT_LIST_HEAD(&q
->timeout_list
);
586 INIT_LIST_HEAD(&q
->icq_list
);
587 #ifdef CONFIG_BLK_CGROUP
588 INIT_LIST_HEAD(&q
->blkg_list
);
590 INIT_LIST_HEAD(&q
->flush_queue
[0]);
591 INIT_LIST_HEAD(&q
->flush_queue
[1]);
592 INIT_LIST_HEAD(&q
->flush_data_in_flight
);
593 INIT_DELAYED_WORK(&q
->delay_work
, blk_delay_work
);
595 kobject_init(&q
->kobj
, &blk_queue_ktype
);
597 mutex_init(&q
->sysfs_lock
);
598 spin_lock_init(&q
->__queue_lock
);
601 * By default initialize queue_lock to internal lock and driver can
602 * override it later if need be.
604 q
->queue_lock
= &q
->__queue_lock
;
607 * A queue starts its life with bypass turned on to avoid
608 * unnecessary bypass on/off overhead and nasty surprises during
609 * init. The initial bypass will be finished at the end of
610 * blk_init_allocated_queue().
613 __set_bit(QUEUE_FLAG_BYPASS
, &q
->queue_flags
);
615 if (blkcg_init_queue(q
))
621 ida_simple_remove(&blk_queue_ida
, q
->id
);
623 kmem_cache_free(blk_requestq_cachep
, q
);
626 EXPORT_SYMBOL(blk_alloc_queue_node
);
629 * blk_init_queue - prepare a request queue for use with a block device
630 * @rfn: The function to be called to process requests that have been
631 * placed on the queue.
632 * @lock: Request queue spin lock
635 * If a block device wishes to use the standard request handling procedures,
636 * which sorts requests and coalesces adjacent requests, then it must
637 * call blk_init_queue(). The function @rfn will be called when there
638 * are requests on the queue that need to be processed. If the device
639 * supports plugging, then @rfn may not be called immediately when requests
640 * are available on the queue, but may be called at some time later instead.
641 * Plugged queues are generally unplugged when a buffer belonging to one
642 * of the requests on the queue is needed, or due to memory pressure.
644 * @rfn is not required, or even expected, to remove all requests off the
645 * queue, but only as many as it can handle at a time. If it does leave
646 * requests on the queue, it is responsible for arranging that the requests
647 * get dealt with eventually.
649 * The queue spin lock must be held while manipulating the requests on the
650 * request queue; this lock will be taken also from interrupt context, so irq
651 * disabling is needed for it.
653 * Function returns a pointer to the initialized request queue, or %NULL if
657 * blk_init_queue() must be paired with a blk_cleanup_queue() call
658 * when the block device is deactivated (such as at module unload).
661 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
663 return blk_init_queue_node(rfn
, lock
, -1);
665 EXPORT_SYMBOL(blk_init_queue
);
667 struct request_queue
*
668 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
670 struct request_queue
*uninit_q
, *q
;
672 uninit_q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
676 q
= blk_init_allocated_queue(uninit_q
, rfn
, lock
);
678 blk_cleanup_queue(uninit_q
);
682 EXPORT_SYMBOL(blk_init_queue_node
);
684 struct request_queue
*
685 blk_init_allocated_queue(struct request_queue
*q
, request_fn_proc
*rfn
,
691 if (blk_init_rl(&q
->root_rl
, q
, GFP_KERNEL
))
695 q
->prep_rq_fn
= NULL
;
696 q
->unprep_rq_fn
= NULL
;
697 q
->queue_flags
= QUEUE_FLAG_DEFAULT
;
699 /* Override internal queue lock with supplied lock pointer */
701 q
->queue_lock
= lock
;
704 * This also sets hw/phys segments, boundary and size
706 blk_queue_make_request(q
, blk_queue_bio
);
708 q
->sg_reserved_size
= INT_MAX
;
711 if (elevator_init(q
, NULL
))
714 blk_queue_congestion_threshold(q
);
716 /* all done, end the initial bypass */
717 blk_queue_bypass_end(q
);
720 EXPORT_SYMBOL(blk_init_allocated_queue
);
722 bool blk_get_queue(struct request_queue
*q
)
724 if (likely(!blk_queue_dead(q
))) {
731 EXPORT_SYMBOL(blk_get_queue
);
733 static inline void blk_free_request(struct request_list
*rl
, struct request
*rq
)
735 if (rq
->cmd_flags
& REQ_ELVPRIV
) {
736 elv_put_request(rl
->q
, rq
);
738 put_io_context(rq
->elv
.icq
->ioc
);
741 mempool_free(rq
, rl
->rq_pool
);
745 * ioc_batching returns true if the ioc is a valid batching request and
746 * should be given priority access to a request.
748 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
754 * Make sure the process is able to allocate at least 1 request
755 * even if the batch times out, otherwise we could theoretically
758 return ioc
->nr_batch_requests
== q
->nr_batching
||
759 (ioc
->nr_batch_requests
> 0
760 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
764 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
765 * will cause the process to be a "batcher" on all queues in the system. This
766 * is the behaviour we want though - once it gets a wakeup it should be given
769 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
771 if (!ioc
|| ioc_batching(q
, ioc
))
774 ioc
->nr_batch_requests
= q
->nr_batching
;
775 ioc
->last_waited
= jiffies
;
778 static void __freed_request(struct request_list
*rl
, int sync
)
780 struct request_queue
*q
= rl
->q
;
783 * bdi isn't aware of blkcg yet. As all async IOs end up root
784 * blkcg anyway, just use root blkcg state.
786 if (rl
== &q
->root_rl
&&
787 rl
->count
[sync
] < queue_congestion_off_threshold(q
))
788 blk_clear_queue_congested(q
, sync
);
790 if (rl
->count
[sync
] + 1 <= q
->nr_requests
) {
791 if (waitqueue_active(&rl
->wait
[sync
]))
792 wake_up(&rl
->wait
[sync
]);
794 blk_clear_rl_full(rl
, sync
);
799 * A request has just been released. Account for it, update the full and
800 * congestion status, wake up any waiters. Called under q->queue_lock.
802 static void freed_request(struct request_list
*rl
, unsigned int flags
)
804 struct request_queue
*q
= rl
->q
;
805 int sync
= rw_is_sync(flags
);
809 if (flags
& REQ_ELVPRIV
)
812 __freed_request(rl
, sync
);
814 if (unlikely(rl
->starved
[sync
^ 1]))
815 __freed_request(rl
, sync
^ 1);
819 * Determine if elevator data should be initialized when allocating the
820 * request associated with @bio.
822 static bool blk_rq_should_init_elevator(struct bio
*bio
)
828 * Flush requests do not use the elevator so skip initialization.
829 * This allows a request to share the flush and elevator data.
831 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
))
838 * rq_ioc - determine io_context for request allocation
839 * @bio: request being allocated is for this bio (can be %NULL)
841 * Determine io_context to use for request allocation for @bio. May return
842 * %NULL if %current->io_context doesn't exist.
844 static struct io_context
*rq_ioc(struct bio
*bio
)
846 #ifdef CONFIG_BLK_CGROUP
847 if (bio
&& bio
->bi_ioc
)
850 return current
->io_context
;
854 * __get_request - get a free request
855 * @rl: request list to allocate from
856 * @rw_flags: RW and SYNC flags
857 * @bio: bio to allocate request for (can be %NULL)
858 * @gfp_mask: allocation mask
860 * Get a free request from @q. This function may fail under memory
861 * pressure or if @q is dead.
863 * Must be callled with @q->queue_lock held and,
864 * Returns %NULL on failure, with @q->queue_lock held.
865 * Returns !%NULL on success, with @q->queue_lock *not held*.
867 static struct request
*__get_request(struct request_list
*rl
, int rw_flags
,
868 struct bio
*bio
, gfp_t gfp_mask
)
870 struct request_queue
*q
= rl
->q
;
872 struct elevator_type
*et
= q
->elevator
->type
;
873 struct io_context
*ioc
= rq_ioc(bio
);
874 struct io_cq
*icq
= NULL
;
875 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
878 if (unlikely(blk_queue_dead(q
)))
881 may_queue
= elv_may_queue(q
, rw_flags
);
882 if (may_queue
== ELV_MQUEUE_NO
)
885 if (rl
->count
[is_sync
]+1 >= queue_congestion_on_threshold(q
)) {
886 if (rl
->count
[is_sync
]+1 >= q
->nr_requests
) {
888 * The queue will fill after this allocation, so set
889 * it as full, and mark this process as "batching".
890 * This process will be allowed to complete a batch of
891 * requests, others will be blocked.
893 if (!blk_rl_full(rl
, is_sync
)) {
894 ioc_set_batching(q
, ioc
);
895 blk_set_rl_full(rl
, is_sync
);
897 if (may_queue
!= ELV_MQUEUE_MUST
898 && !ioc_batching(q
, ioc
)) {
900 * The queue is full and the allocating
901 * process is not a "batcher", and not
902 * exempted by the IO scheduler
909 * bdi isn't aware of blkcg yet. As all async IOs end up
910 * root blkcg anyway, just use root blkcg state.
912 if (rl
== &q
->root_rl
)
913 blk_set_queue_congested(q
, is_sync
);
917 * Only allow batching queuers to allocate up to 50% over the defined
918 * limit of requests, otherwise we could have thousands of requests
919 * allocated with any setting of ->nr_requests
921 if (rl
->count
[is_sync
] >= (3 * q
->nr_requests
/ 2))
924 q
->nr_rqs
[is_sync
]++;
925 rl
->count
[is_sync
]++;
926 rl
->starved
[is_sync
] = 0;
929 * Decide whether the new request will be managed by elevator. If
930 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
931 * prevent the current elevator from being destroyed until the new
932 * request is freed. This guarantees icq's won't be destroyed and
933 * makes creating new ones safe.
935 * Also, lookup icq while holding queue_lock. If it doesn't exist,
936 * it will be created after releasing queue_lock.
938 if (blk_rq_should_init_elevator(bio
) && !blk_queue_bypass(q
)) {
939 rw_flags
|= REQ_ELVPRIV
;
941 if (et
->icq_cache
&& ioc
)
942 icq
= ioc_lookup_icq(ioc
, q
);
945 if (blk_queue_io_stat(q
))
946 rw_flags
|= REQ_IO_STAT
;
947 spin_unlock_irq(q
->queue_lock
);
949 /* allocate and init request */
950 rq
= mempool_alloc(rl
->rq_pool
, gfp_mask
);
955 blk_rq_set_rl(rq
, rl
);
956 rq
->cmd_flags
= rw_flags
| REQ_ALLOCED
;
959 if (rw_flags
& REQ_ELVPRIV
) {
960 if (unlikely(et
->icq_cache
&& !icq
)) {
962 icq
= ioc_create_icq(ioc
, q
, gfp_mask
);
968 if (unlikely(elv_set_request(q
, rq
, bio
, gfp_mask
)))
971 /* @rq->elv.icq holds io_context until @rq is freed */
973 get_io_context(icq
->ioc
);
977 * ioc may be NULL here, and ioc_batching will be false. That's
978 * OK, if the queue is under the request limit then requests need
979 * not count toward the nr_batch_requests limit. There will always
980 * be some limit enforced by BLK_BATCH_TIME.
982 if (ioc_batching(q
, ioc
))
983 ioc
->nr_batch_requests
--;
985 trace_block_getrq(q
, bio
, rw_flags
& 1);
990 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
991 * and may fail indefinitely under memory pressure and thus
992 * shouldn't stall IO. Treat this request as !elvpriv. This will
993 * disturb iosched and blkcg but weird is bettern than dead.
995 printk_ratelimited(KERN_WARNING
"%s: request aux data allocation failed, iosched may be disturbed\n",
996 dev_name(q
->backing_dev_info
.dev
));
998 rq
->cmd_flags
&= ~REQ_ELVPRIV
;
1001 spin_lock_irq(q
->queue_lock
);
1002 q
->nr_rqs_elvpriv
--;
1003 spin_unlock_irq(q
->queue_lock
);
1008 * Allocation failed presumably due to memory. Undo anything we
1009 * might have messed up.
1011 * Allocating task should really be put onto the front of the wait
1012 * queue, but this is pretty rare.
1014 spin_lock_irq(q
->queue_lock
);
1015 freed_request(rl
, rw_flags
);
1018 * in the very unlikely event that allocation failed and no
1019 * requests for this direction was pending, mark us starved so that
1020 * freeing of a request in the other direction will notice
1021 * us. another possible fix would be to split the rq mempool into
1025 if (unlikely(rl
->count
[is_sync
] == 0))
1026 rl
->starved
[is_sync
] = 1;
1031 * get_request - get a free request
1032 * @q: request_queue to allocate request from
1033 * @rw_flags: RW and SYNC flags
1034 * @bio: bio to allocate request for (can be %NULL)
1035 * @gfp_mask: allocation mask
1037 * Get a free request from @q. If %__GFP_WAIT is set in @gfp_mask, this
1038 * function keeps retrying under memory pressure and fails iff @q is dead.
1040 * Must be callled with @q->queue_lock held and,
1041 * Returns %NULL on failure, with @q->queue_lock held.
1042 * Returns !%NULL on success, with @q->queue_lock *not held*.
1044 static struct request
*get_request(struct request_queue
*q
, int rw_flags
,
1045 struct bio
*bio
, gfp_t gfp_mask
)
1047 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
1049 struct request_list
*rl
;
1052 rl
= blk_get_rl(q
, bio
); /* transferred to @rq on success */
1054 rq
= __get_request(rl
, rw_flags
, bio
, gfp_mask
);
1058 if (!(gfp_mask
& __GFP_WAIT
) || unlikely(blk_queue_dead(q
))) {
1063 /* wait on @rl and retry */
1064 prepare_to_wait_exclusive(&rl
->wait
[is_sync
], &wait
,
1065 TASK_UNINTERRUPTIBLE
);
1067 trace_block_sleeprq(q
, bio
, rw_flags
& 1);
1069 spin_unlock_irq(q
->queue_lock
);
1073 * After sleeping, we become a "batching" process and will be able
1074 * to allocate at least one request, and up to a big batch of them
1075 * for a small period time. See ioc_batching, ioc_set_batching
1077 ioc_set_batching(q
, current
->io_context
);
1079 spin_lock_irq(q
->queue_lock
);
1080 finish_wait(&rl
->wait
[is_sync
], &wait
);
1085 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
1089 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
1091 /* create ioc upfront */
1092 create_io_context(gfp_mask
, q
->node
);
1094 spin_lock_irq(q
->queue_lock
);
1095 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
1097 spin_unlock_irq(q
->queue_lock
);
1098 /* q->queue_lock is unlocked at this point */
1102 EXPORT_SYMBOL(blk_get_request
);
1105 * blk_make_request - given a bio, allocate a corresponding struct request.
1106 * @q: target request queue
1107 * @bio: The bio describing the memory mappings that will be submitted for IO.
1108 * It may be a chained-bio properly constructed by block/bio layer.
1109 * @gfp_mask: gfp flags to be used for memory allocation
1111 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1112 * type commands. Where the struct request needs to be farther initialized by
1113 * the caller. It is passed a &struct bio, which describes the memory info of
1116 * The caller of blk_make_request must make sure that bi_io_vec
1117 * are set to describe the memory buffers. That bio_data_dir() will return
1118 * the needed direction of the request. (And all bio's in the passed bio-chain
1119 * are properly set accordingly)
1121 * If called under none-sleepable conditions, mapped bio buffers must not
1122 * need bouncing, by calling the appropriate masked or flagged allocator,
1123 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1126 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1127 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1128 * anything but the first bio in the chain. Otherwise you risk waiting for IO
1129 * completion of a bio that hasn't been submitted yet, thus resulting in a
1130 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1131 * of bio_alloc(), as that avoids the mempool deadlock.
1132 * If possible a big IO should be split into smaller parts when allocation
1133 * fails. Partial allocation should not be an error, or you risk a live-lock.
1135 struct request
*blk_make_request(struct request_queue
*q
, struct bio
*bio
,
1138 struct request
*rq
= blk_get_request(q
, bio_data_dir(bio
), gfp_mask
);
1141 return ERR_PTR(-ENOMEM
);
1144 struct bio
*bounce_bio
= bio
;
1147 blk_queue_bounce(q
, &bounce_bio
);
1148 ret
= blk_rq_append_bio(q
, rq
, bounce_bio
);
1149 if (unlikely(ret
)) {
1150 blk_put_request(rq
);
1151 return ERR_PTR(ret
);
1157 EXPORT_SYMBOL(blk_make_request
);
1160 * blk_requeue_request - put a request back on queue
1161 * @q: request queue where request should be inserted
1162 * @rq: request to be inserted
1165 * Drivers often keep queueing requests until the hardware cannot accept
1166 * more, when that condition happens we need to put the request back
1167 * on the queue. Must be called with queue lock held.
1169 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
1171 blk_delete_timer(rq
);
1172 blk_clear_rq_complete(rq
);
1173 trace_block_rq_requeue(q
, rq
);
1175 if (blk_rq_tagged(rq
))
1176 blk_queue_end_tag(q
, rq
);
1178 BUG_ON(blk_queued_rq(rq
));
1180 elv_requeue_request(q
, rq
);
1182 EXPORT_SYMBOL(blk_requeue_request
);
1184 static void add_acct_request(struct request_queue
*q
, struct request
*rq
,
1187 drive_stat_acct(rq
, 1);
1188 __elv_add_request(q
, rq
, where
);
1191 static void part_round_stats_single(int cpu
, struct hd_struct
*part
,
1194 if (now
== part
->stamp
)
1197 if (part_in_flight(part
)) {
1198 __part_stat_add(cpu
, part
, time_in_queue
,
1199 part_in_flight(part
) * (now
- part
->stamp
));
1200 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
1206 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1207 * @cpu: cpu number for stats access
1208 * @part: target partition
1210 * The average IO queue length and utilisation statistics are maintained
1211 * by observing the current state of the queue length and the amount of
1212 * time it has been in this state for.
1214 * Normally, that accounting is done on IO completion, but that can result
1215 * in more than a second's worth of IO being accounted for within any one
1216 * second, leading to >100% utilisation. To deal with that, we call this
1217 * function to do a round-off before returning the results when reading
1218 * /proc/diskstats. This accounts immediately for all queue usage up to
1219 * the current jiffies and restarts the counters again.
1221 void part_round_stats(int cpu
, struct hd_struct
*part
)
1223 unsigned long now
= jiffies
;
1226 part_round_stats_single(cpu
, &part_to_disk(part
)->part0
, now
);
1227 part_round_stats_single(cpu
, part
, now
);
1229 EXPORT_SYMBOL_GPL(part_round_stats
);
1232 * queue lock must be held
1234 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1238 if (unlikely(--req
->ref_count
))
1241 elv_completed_request(q
, req
);
1243 /* this is a bio leak */
1244 WARN_ON(req
->bio
!= NULL
);
1247 * Request may not have originated from ll_rw_blk. if not,
1248 * it didn't come out of our reserved rq pools
1250 if (req
->cmd_flags
& REQ_ALLOCED
) {
1251 unsigned int flags
= req
->cmd_flags
;
1252 struct request_list
*rl
= blk_rq_rl(req
);
1254 BUG_ON(!list_empty(&req
->queuelist
));
1255 BUG_ON(!hlist_unhashed(&req
->hash
));
1257 blk_free_request(rl
, req
);
1258 freed_request(rl
, flags
);
1262 EXPORT_SYMBOL_GPL(__blk_put_request
);
1264 void blk_put_request(struct request
*req
)
1266 unsigned long flags
;
1267 struct request_queue
*q
= req
->q
;
1269 spin_lock_irqsave(q
->queue_lock
, flags
);
1270 __blk_put_request(q
, req
);
1271 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1273 EXPORT_SYMBOL(blk_put_request
);
1276 * blk_add_request_payload - add a payload to a request
1277 * @rq: request to update
1278 * @page: page backing the payload
1279 * @len: length of the payload.
1281 * This allows to later add a payload to an already submitted request by
1282 * a block driver. The driver needs to take care of freeing the payload
1285 * Note that this is a quite horrible hack and nothing but handling of
1286 * discard requests should ever use it.
1288 void blk_add_request_payload(struct request
*rq
, struct page
*page
,
1291 struct bio
*bio
= rq
->bio
;
1293 bio
->bi_io_vec
->bv_page
= page
;
1294 bio
->bi_io_vec
->bv_offset
= 0;
1295 bio
->bi_io_vec
->bv_len
= len
;
1299 bio
->bi_phys_segments
= 1;
1301 rq
->__data_len
= rq
->resid_len
= len
;
1302 rq
->nr_phys_segments
= 1;
1303 rq
->buffer
= bio_data(bio
);
1305 EXPORT_SYMBOL_GPL(blk_add_request_payload
);
1307 static bool bio_attempt_back_merge(struct request_queue
*q
, struct request
*req
,
1310 const int ff
= bio
->bi_rw
& REQ_FAILFAST_MASK
;
1312 if (!ll_back_merge_fn(q
, req
, bio
))
1315 trace_block_bio_backmerge(q
, bio
);
1317 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1318 blk_rq_set_mixed_merge(req
);
1320 req
->biotail
->bi_next
= bio
;
1322 req
->__data_len
+= bio
->bi_size
;
1323 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1325 drive_stat_acct(req
, 0);
1329 static bool bio_attempt_front_merge(struct request_queue
*q
,
1330 struct request
*req
, struct bio
*bio
)
1332 const int ff
= bio
->bi_rw
& REQ_FAILFAST_MASK
;
1334 if (!ll_front_merge_fn(q
, req
, bio
))
1337 trace_block_bio_frontmerge(q
, bio
);
1339 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1340 blk_rq_set_mixed_merge(req
);
1342 bio
->bi_next
= req
->bio
;
1346 * may not be valid. if the low level driver said
1347 * it didn't need a bounce buffer then it better
1348 * not touch req->buffer either...
1350 req
->buffer
= bio_data(bio
);
1351 req
->__sector
= bio
->bi_sector
;
1352 req
->__data_len
+= bio
->bi_size
;
1353 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1355 drive_stat_acct(req
, 0);
1360 * attempt_plug_merge - try to merge with %current's plugged list
1361 * @q: request_queue new bio is being queued at
1362 * @bio: new bio being queued
1363 * @request_count: out parameter for number of traversed plugged requests
1365 * Determine whether @bio being queued on @q can be merged with a request
1366 * on %current's plugged list. Returns %true if merge was successful,
1369 * Plugging coalesces IOs from the same issuer for the same purpose without
1370 * going through @q->queue_lock. As such it's more of an issuing mechanism
1371 * than scheduling, and the request, while may have elvpriv data, is not
1372 * added on the elevator at this point. In addition, we don't have
1373 * reliable access to the elevator outside queue lock. Only check basic
1374 * merging parameters without querying the elevator.
1376 static bool attempt_plug_merge(struct request_queue
*q
, struct bio
*bio
,
1377 unsigned int *request_count
)
1379 struct blk_plug
*plug
;
1383 plug
= current
->plug
;
1388 list_for_each_entry_reverse(rq
, &plug
->list
, queuelist
) {
1394 if (rq
->q
!= q
|| !blk_rq_merge_ok(rq
, bio
))
1397 el_ret
= blk_try_merge(rq
, bio
);
1398 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1399 ret
= bio_attempt_back_merge(q
, rq
, bio
);
1402 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1403 ret
= bio_attempt_front_merge(q
, rq
, bio
);
1412 void init_request_from_bio(struct request
*req
, struct bio
*bio
)
1414 req
->cmd_type
= REQ_TYPE_FS
;
1416 req
->cmd_flags
|= bio
->bi_rw
& REQ_COMMON_MASK
;
1417 if (bio
->bi_rw
& REQ_RAHEAD
)
1418 req
->cmd_flags
|= REQ_FAILFAST_MASK
;
1421 req
->__sector
= bio
->bi_sector
;
1422 req
->ioprio
= bio_prio(bio
);
1423 blk_rq_bio_prep(req
->q
, req
, bio
);
1426 void blk_queue_bio(struct request_queue
*q
, struct bio
*bio
)
1428 const bool sync
= !!(bio
->bi_rw
& REQ_SYNC
);
1429 struct blk_plug
*plug
;
1430 int el_ret
, rw_flags
, where
= ELEVATOR_INSERT_SORT
;
1431 struct request
*req
;
1432 unsigned int request_count
= 0;
1435 * low level driver can indicate that it wants pages above a
1436 * certain limit bounced to low memory (ie for highmem, or even
1437 * ISA dma in theory)
1439 blk_queue_bounce(q
, &bio
);
1441 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) {
1442 spin_lock_irq(q
->queue_lock
);
1443 where
= ELEVATOR_INSERT_FLUSH
;
1448 * Check if we can merge with the plugged list before grabbing
1451 if (attempt_plug_merge(q
, bio
, &request_count
))
1454 spin_lock_irq(q
->queue_lock
);
1456 el_ret
= elv_merge(q
, &req
, bio
);
1457 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1458 if (bio_attempt_back_merge(q
, req
, bio
)) {
1459 elv_bio_merged(q
, req
, bio
);
1460 if (!attempt_back_merge(q
, req
))
1461 elv_merged_request(q
, req
, el_ret
);
1464 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1465 if (bio_attempt_front_merge(q
, req
, bio
)) {
1466 elv_bio_merged(q
, req
, bio
);
1467 if (!attempt_front_merge(q
, req
))
1468 elv_merged_request(q
, req
, el_ret
);
1475 * This sync check and mask will be re-done in init_request_from_bio(),
1476 * but we need to set it earlier to expose the sync flag to the
1477 * rq allocator and io schedulers.
1479 rw_flags
= bio_data_dir(bio
);
1481 rw_flags
|= REQ_SYNC
;
1484 * Grab a free request. This is might sleep but can not fail.
1485 * Returns with the queue unlocked.
1487 req
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
1488 if (unlikely(!req
)) {
1489 bio_endio(bio
, -ENODEV
); /* @q is dead */
1494 * After dropping the lock and possibly sleeping here, our request
1495 * may now be mergeable after it had proven unmergeable (above).
1496 * We don't worry about that case for efficiency. It won't happen
1497 * often, and the elevators are able to handle it.
1499 init_request_from_bio(req
, bio
);
1501 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
))
1502 req
->cpu
= raw_smp_processor_id();
1504 plug
= current
->plug
;
1507 * If this is the first request added after a plug, fire
1508 * of a plug trace. If others have been added before, check
1509 * if we have multiple devices in this plug. If so, make a
1510 * note to sort the list before dispatch.
1512 if (list_empty(&plug
->list
))
1513 trace_block_plug(q
);
1515 if (!plug
->should_sort
) {
1516 struct request
*__rq
;
1518 __rq
= list_entry_rq(plug
->list
.prev
);
1520 plug
->should_sort
= 1;
1522 if (request_count
>= BLK_MAX_REQUEST_COUNT
) {
1523 blk_flush_plug_list(plug
, false);
1524 trace_block_plug(q
);
1527 list_add_tail(&req
->queuelist
, &plug
->list
);
1528 drive_stat_acct(req
, 1);
1530 spin_lock_irq(q
->queue_lock
);
1531 add_acct_request(q
, req
, where
);
1534 spin_unlock_irq(q
->queue_lock
);
1537 EXPORT_SYMBOL_GPL(blk_queue_bio
); /* for device mapper only */
1540 * If bio->bi_dev is a partition, remap the location
1542 static inline void blk_partition_remap(struct bio
*bio
)
1544 struct block_device
*bdev
= bio
->bi_bdev
;
1546 if (bio_sectors(bio
) && bdev
!= bdev
->bd_contains
) {
1547 struct hd_struct
*p
= bdev
->bd_part
;
1549 bio
->bi_sector
+= p
->start_sect
;
1550 bio
->bi_bdev
= bdev
->bd_contains
;
1552 trace_block_bio_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
1554 bio
->bi_sector
- p
->start_sect
);
1558 static void handle_bad_sector(struct bio
*bio
)
1560 char b
[BDEVNAME_SIZE
];
1562 printk(KERN_INFO
"attempt to access beyond end of device\n");
1563 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
1564 bdevname(bio
->bi_bdev
, b
),
1566 (unsigned long long)bio
->bi_sector
+ bio_sectors(bio
),
1567 (long long)(i_size_read(bio
->bi_bdev
->bd_inode
) >> 9));
1569 set_bit(BIO_EOF
, &bio
->bi_flags
);
1572 #ifdef CONFIG_FAIL_MAKE_REQUEST
1574 static DECLARE_FAULT_ATTR(fail_make_request
);
1576 static int __init
setup_fail_make_request(char *str
)
1578 return setup_fault_attr(&fail_make_request
, str
);
1580 __setup("fail_make_request=", setup_fail_make_request
);
1582 static bool should_fail_request(struct hd_struct
*part
, unsigned int bytes
)
1584 return part
->make_it_fail
&& should_fail(&fail_make_request
, bytes
);
1587 static int __init
fail_make_request_debugfs(void)
1589 struct dentry
*dir
= fault_create_debugfs_attr("fail_make_request",
1590 NULL
, &fail_make_request
);
1592 return IS_ERR(dir
) ? PTR_ERR(dir
) : 0;
1595 late_initcall(fail_make_request_debugfs
);
1597 #else /* CONFIG_FAIL_MAKE_REQUEST */
1599 static inline bool should_fail_request(struct hd_struct
*part
,
1605 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1608 * Check whether this bio extends beyond the end of the device.
1610 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
1617 /* Test device or partition size, when known. */
1618 maxsector
= i_size_read(bio
->bi_bdev
->bd_inode
) >> 9;
1620 sector_t sector
= bio
->bi_sector
;
1622 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
1624 * This may well happen - the kernel calls bread()
1625 * without checking the size of the device, e.g., when
1626 * mounting a device.
1628 handle_bad_sector(bio
);
1636 static noinline_for_stack
bool
1637 generic_make_request_checks(struct bio
*bio
)
1639 struct request_queue
*q
;
1640 int nr_sectors
= bio_sectors(bio
);
1642 char b
[BDEVNAME_SIZE
];
1643 struct hd_struct
*part
;
1647 if (bio_check_eod(bio
, nr_sectors
))
1650 q
= bdev_get_queue(bio
->bi_bdev
);
1653 "generic_make_request: Trying to access "
1654 "nonexistent block-device %s (%Lu)\n",
1655 bdevname(bio
->bi_bdev
, b
),
1656 (long long) bio
->bi_sector
);
1660 if (unlikely(!(bio
->bi_rw
& REQ_DISCARD
) &&
1661 nr_sectors
> queue_max_hw_sectors(q
))) {
1662 printk(KERN_ERR
"bio too big device %s (%u > %u)\n",
1663 bdevname(bio
->bi_bdev
, b
),
1665 queue_max_hw_sectors(q
));
1669 part
= bio
->bi_bdev
->bd_part
;
1670 if (should_fail_request(part
, bio
->bi_size
) ||
1671 should_fail_request(&part_to_disk(part
)->part0
,
1676 * If this device has partitions, remap block n
1677 * of partition p to block n+start(p) of the disk.
1679 blk_partition_remap(bio
);
1681 if (bio_integrity_enabled(bio
) && bio_integrity_prep(bio
))
1684 if (bio_check_eod(bio
, nr_sectors
))
1688 * Filter flush bio's early so that make_request based
1689 * drivers without flush support don't have to worry
1692 if ((bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) && !q
->flush_flags
) {
1693 bio
->bi_rw
&= ~(REQ_FLUSH
| REQ_FUA
);
1700 if ((bio
->bi_rw
& REQ_DISCARD
) &&
1701 (!blk_queue_discard(q
) ||
1702 ((bio
->bi_rw
& REQ_SECURE
) &&
1703 !blk_queue_secdiscard(q
)))) {
1709 * Various block parts want %current->io_context and lazy ioc
1710 * allocation ends up trading a lot of pain for a small amount of
1711 * memory. Just allocate it upfront. This may fail and block
1712 * layer knows how to live with it.
1714 create_io_context(GFP_ATOMIC
, q
->node
);
1716 if (blk_throtl_bio(q
, bio
))
1717 return false; /* throttled, will be resubmitted later */
1719 trace_block_bio_queue(q
, bio
);
1723 bio_endio(bio
, err
);
1728 * generic_make_request - hand a buffer to its device driver for I/O
1729 * @bio: The bio describing the location in memory and on the device.
1731 * generic_make_request() is used to make I/O requests of block
1732 * devices. It is passed a &struct bio, which describes the I/O that needs
1735 * generic_make_request() does not return any status. The
1736 * success/failure status of the request, along with notification of
1737 * completion, is delivered asynchronously through the bio->bi_end_io
1738 * function described (one day) else where.
1740 * The caller of generic_make_request must make sure that bi_io_vec
1741 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1742 * set to describe the device address, and the
1743 * bi_end_io and optionally bi_private are set to describe how
1744 * completion notification should be signaled.
1746 * generic_make_request and the drivers it calls may use bi_next if this
1747 * bio happens to be merged with someone else, and may resubmit the bio to
1748 * a lower device by calling into generic_make_request recursively, which
1749 * means the bio should NOT be touched after the call to ->make_request_fn.
1751 void generic_make_request(struct bio
*bio
)
1753 struct bio_list bio_list_on_stack
;
1755 if (!generic_make_request_checks(bio
))
1759 * We only want one ->make_request_fn to be active at a time, else
1760 * stack usage with stacked devices could be a problem. So use
1761 * current->bio_list to keep a list of requests submited by a
1762 * make_request_fn function. current->bio_list is also used as a
1763 * flag to say if generic_make_request is currently active in this
1764 * task or not. If it is NULL, then no make_request is active. If
1765 * it is non-NULL, then a make_request is active, and new requests
1766 * should be added at the tail
1768 if (current
->bio_list
) {
1769 bio_list_add(current
->bio_list
, bio
);
1773 /* following loop may be a bit non-obvious, and so deserves some
1775 * Before entering the loop, bio->bi_next is NULL (as all callers
1776 * ensure that) so we have a list with a single bio.
1777 * We pretend that we have just taken it off a longer list, so
1778 * we assign bio_list to a pointer to the bio_list_on_stack,
1779 * thus initialising the bio_list of new bios to be
1780 * added. ->make_request() may indeed add some more bios
1781 * through a recursive call to generic_make_request. If it
1782 * did, we find a non-NULL value in bio_list and re-enter the loop
1783 * from the top. In this case we really did just take the bio
1784 * of the top of the list (no pretending) and so remove it from
1785 * bio_list, and call into ->make_request() again.
1787 BUG_ON(bio
->bi_next
);
1788 bio_list_init(&bio_list_on_stack
);
1789 current
->bio_list
= &bio_list_on_stack
;
1791 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
1793 q
->make_request_fn(q
, bio
);
1795 bio
= bio_list_pop(current
->bio_list
);
1797 current
->bio_list
= NULL
; /* deactivate */
1799 EXPORT_SYMBOL(generic_make_request
);
1802 * submit_bio - submit a bio to the block device layer for I/O
1803 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1804 * @bio: The &struct bio which describes the I/O
1806 * submit_bio() is very similar in purpose to generic_make_request(), and
1807 * uses that function to do most of the work. Both are fairly rough
1808 * interfaces; @bio must be presetup and ready for I/O.
1811 void submit_bio(int rw
, struct bio
*bio
)
1813 int count
= bio_sectors(bio
);
1818 * If it's a regular read/write or a barrier with data attached,
1819 * go through the normal accounting stuff before submission.
1821 if (bio_has_data(bio
) && !(rw
& REQ_DISCARD
)) {
1823 count_vm_events(PGPGOUT
, count
);
1825 task_io_account_read(bio
->bi_size
);
1826 count_vm_events(PGPGIN
, count
);
1829 if (unlikely(block_dump
)) {
1830 char b
[BDEVNAME_SIZE
];
1831 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s (%u sectors)\n",
1832 current
->comm
, task_pid_nr(current
),
1833 (rw
& WRITE
) ? "WRITE" : "READ",
1834 (unsigned long long)bio
->bi_sector
,
1835 bdevname(bio
->bi_bdev
, b
),
1840 generic_make_request(bio
);
1842 EXPORT_SYMBOL(submit_bio
);
1845 * blk_rq_check_limits - Helper function to check a request for the queue limit
1847 * @rq: the request being checked
1850 * @rq may have been made based on weaker limitations of upper-level queues
1851 * in request stacking drivers, and it may violate the limitation of @q.
1852 * Since the block layer and the underlying device driver trust @rq
1853 * after it is inserted to @q, it should be checked against @q before
1854 * the insertion using this generic function.
1856 * This function should also be useful for request stacking drivers
1857 * in some cases below, so export this function.
1858 * Request stacking drivers like request-based dm may change the queue
1859 * limits while requests are in the queue (e.g. dm's table swapping).
1860 * Such request stacking drivers should check those requests agaist
1861 * the new queue limits again when they dispatch those requests,
1862 * although such checkings are also done against the old queue limits
1863 * when submitting requests.
1865 int blk_rq_check_limits(struct request_queue
*q
, struct request
*rq
)
1867 if (rq
->cmd_flags
& REQ_DISCARD
)
1870 if (blk_rq_sectors(rq
) > queue_max_sectors(q
) ||
1871 blk_rq_bytes(rq
) > queue_max_hw_sectors(q
) << 9) {
1872 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
1877 * queue's settings related to segment counting like q->bounce_pfn
1878 * may differ from that of other stacking queues.
1879 * Recalculate it to check the request correctly on this queue's
1882 blk_recalc_rq_segments(rq
);
1883 if (rq
->nr_phys_segments
> queue_max_segments(q
)) {
1884 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
1890 EXPORT_SYMBOL_GPL(blk_rq_check_limits
);
1893 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1894 * @q: the queue to submit the request
1895 * @rq: the request being queued
1897 int blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
1899 unsigned long flags
;
1900 int where
= ELEVATOR_INSERT_BACK
;
1902 if (blk_rq_check_limits(q
, rq
))
1906 should_fail_request(&rq
->rq_disk
->part0
, blk_rq_bytes(rq
)))
1909 spin_lock_irqsave(q
->queue_lock
, flags
);
1910 if (unlikely(blk_queue_dead(q
))) {
1911 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1916 * Submitting request must be dequeued before calling this function
1917 * because it will be linked to another request_queue
1919 BUG_ON(blk_queued_rq(rq
));
1921 if (rq
->cmd_flags
& (REQ_FLUSH
|REQ_FUA
))
1922 where
= ELEVATOR_INSERT_FLUSH
;
1924 add_acct_request(q
, rq
, where
);
1925 if (where
== ELEVATOR_INSERT_FLUSH
)
1927 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1931 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
1934 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1935 * @rq: request to examine
1938 * A request could be merge of IOs which require different failure
1939 * handling. This function determines the number of bytes which
1940 * can be failed from the beginning of the request without
1941 * crossing into area which need to be retried further.
1944 * The number of bytes to fail.
1947 * queue_lock must be held.
1949 unsigned int blk_rq_err_bytes(const struct request
*rq
)
1951 unsigned int ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
1952 unsigned int bytes
= 0;
1955 if (!(rq
->cmd_flags
& REQ_MIXED_MERGE
))
1956 return blk_rq_bytes(rq
);
1959 * Currently the only 'mixing' which can happen is between
1960 * different fastfail types. We can safely fail portions
1961 * which have all the failfast bits that the first one has -
1962 * the ones which are at least as eager to fail as the first
1965 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
1966 if ((bio
->bi_rw
& ff
) != ff
)
1968 bytes
+= bio
->bi_size
;
1971 /* this could lead to infinite loop */
1972 BUG_ON(blk_rq_bytes(rq
) && !bytes
);
1975 EXPORT_SYMBOL_GPL(blk_rq_err_bytes
);
1977 static void blk_account_io_completion(struct request
*req
, unsigned int bytes
)
1979 if (blk_do_io_stat(req
)) {
1980 const int rw
= rq_data_dir(req
);
1981 struct hd_struct
*part
;
1984 cpu
= part_stat_lock();
1986 part_stat_add(cpu
, part
, sectors
[rw
], bytes
>> 9);
1991 static void blk_account_io_done(struct request
*req
)
1994 * Account IO completion. flush_rq isn't accounted as a
1995 * normal IO on queueing nor completion. Accounting the
1996 * containing request is enough.
1998 if (blk_do_io_stat(req
) && !(req
->cmd_flags
& REQ_FLUSH_SEQ
)) {
1999 unsigned long duration
= jiffies
- req
->start_time
;
2000 const int rw
= rq_data_dir(req
);
2001 struct hd_struct
*part
;
2004 cpu
= part_stat_lock();
2007 part_stat_inc(cpu
, part
, ios
[rw
]);
2008 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
2009 part_round_stats(cpu
, part
);
2010 part_dec_in_flight(part
, rw
);
2012 hd_struct_put(part
);
2018 * blk_peek_request - peek at the top of a request queue
2019 * @q: request queue to peek at
2022 * Return the request at the top of @q. The returned request
2023 * should be started using blk_start_request() before LLD starts
2027 * Pointer to the request at the top of @q if available. Null
2031 * queue_lock must be held.
2033 struct request
*blk_peek_request(struct request_queue
*q
)
2038 while ((rq
= __elv_next_request(q
)) != NULL
) {
2039 if (!(rq
->cmd_flags
& REQ_STARTED
)) {
2041 * This is the first time the device driver
2042 * sees this request (possibly after
2043 * requeueing). Notify IO scheduler.
2045 if (rq
->cmd_flags
& REQ_SORTED
)
2046 elv_activate_rq(q
, rq
);
2049 * just mark as started even if we don't start
2050 * it, a request that has been delayed should
2051 * not be passed by new incoming requests
2053 rq
->cmd_flags
|= REQ_STARTED
;
2054 trace_block_rq_issue(q
, rq
);
2057 if (!q
->boundary_rq
|| q
->boundary_rq
== rq
) {
2058 q
->end_sector
= rq_end_sector(rq
);
2059 q
->boundary_rq
= NULL
;
2062 if (rq
->cmd_flags
& REQ_DONTPREP
)
2065 if (q
->dma_drain_size
&& blk_rq_bytes(rq
)) {
2067 * make sure space for the drain appears we
2068 * know we can do this because max_hw_segments
2069 * has been adjusted to be one fewer than the
2072 rq
->nr_phys_segments
++;
2078 ret
= q
->prep_rq_fn(q
, rq
);
2079 if (ret
== BLKPREP_OK
) {
2081 } else if (ret
== BLKPREP_DEFER
) {
2083 * the request may have been (partially) prepped.
2084 * we need to keep this request in the front to
2085 * avoid resource deadlock. REQ_STARTED will
2086 * prevent other fs requests from passing this one.
2088 if (q
->dma_drain_size
&& blk_rq_bytes(rq
) &&
2089 !(rq
->cmd_flags
& REQ_DONTPREP
)) {
2091 * remove the space for the drain we added
2092 * so that we don't add it again
2094 --rq
->nr_phys_segments
;
2099 } else if (ret
== BLKPREP_KILL
) {
2100 rq
->cmd_flags
|= REQ_QUIET
;
2102 * Mark this request as started so we don't trigger
2103 * any debug logic in the end I/O path.
2105 blk_start_request(rq
);
2106 __blk_end_request_all(rq
, -EIO
);
2108 printk(KERN_ERR
"%s: bad return=%d\n", __func__
, ret
);
2115 EXPORT_SYMBOL(blk_peek_request
);
2117 void blk_dequeue_request(struct request
*rq
)
2119 struct request_queue
*q
= rq
->q
;
2121 BUG_ON(list_empty(&rq
->queuelist
));
2122 BUG_ON(ELV_ON_HASH(rq
));
2124 list_del_init(&rq
->queuelist
);
2127 * the time frame between a request being removed from the lists
2128 * and to it is freed is accounted as io that is in progress at
2131 if (blk_account_rq(rq
)) {
2132 q
->in_flight
[rq_is_sync(rq
)]++;
2133 set_io_start_time_ns(rq
);
2138 * blk_start_request - start request processing on the driver
2139 * @req: request to dequeue
2142 * Dequeue @req and start timeout timer on it. This hands off the
2143 * request to the driver.
2145 * Block internal functions which don't want to start timer should
2146 * call blk_dequeue_request().
2149 * queue_lock must be held.
2151 void blk_start_request(struct request
*req
)
2153 blk_dequeue_request(req
);
2156 * We are now handing the request to the hardware, initialize
2157 * resid_len to full count and add the timeout handler.
2159 req
->resid_len
= blk_rq_bytes(req
);
2160 if (unlikely(blk_bidi_rq(req
)))
2161 req
->next_rq
->resid_len
= blk_rq_bytes(req
->next_rq
);
2165 EXPORT_SYMBOL(blk_start_request
);
2168 * blk_fetch_request - fetch a request from a request queue
2169 * @q: request queue to fetch a request from
2172 * Return the request at the top of @q. The request is started on
2173 * return and LLD can start processing it immediately.
2176 * Pointer to the request at the top of @q if available. Null
2180 * queue_lock must be held.
2182 struct request
*blk_fetch_request(struct request_queue
*q
)
2186 rq
= blk_peek_request(q
);
2188 blk_start_request(rq
);
2191 EXPORT_SYMBOL(blk_fetch_request
);
2194 * blk_update_request - Special helper function for request stacking drivers
2195 * @req: the request being processed
2196 * @error: %0 for success, < %0 for error
2197 * @nr_bytes: number of bytes to complete @req
2200 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2201 * the request structure even if @req doesn't have leftover.
2202 * If @req has leftover, sets it up for the next range of segments.
2204 * This special helper function is only for request stacking drivers
2205 * (e.g. request-based dm) so that they can handle partial completion.
2206 * Actual device drivers should use blk_end_request instead.
2208 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2209 * %false return from this function.
2212 * %false - this request doesn't have any more data
2213 * %true - this request has more data
2215 bool blk_update_request(struct request
*req
, int error
, unsigned int nr_bytes
)
2217 int total_bytes
, bio_nbytes
, next_idx
= 0;
2223 trace_block_rq_complete(req
->q
, req
);
2226 * For fs requests, rq is just carrier of independent bio's
2227 * and each partial completion should be handled separately.
2228 * Reset per-request error on each partial completion.
2230 * TODO: tj: This is too subtle. It would be better to let
2231 * low level drivers do what they see fit.
2233 if (req
->cmd_type
== REQ_TYPE_FS
)
2236 if (error
&& req
->cmd_type
== REQ_TYPE_FS
&&
2237 !(req
->cmd_flags
& REQ_QUIET
)) {
2242 error_type
= "recoverable transport";
2245 error_type
= "critical target";
2248 error_type
= "critical nexus";
2255 printk_ratelimited(KERN_ERR
"end_request: %s error, dev %s, sector %llu\n",
2256 error_type
, req
->rq_disk
?
2257 req
->rq_disk
->disk_name
: "?",
2258 (unsigned long long)blk_rq_pos(req
));
2262 blk_account_io_completion(req
, nr_bytes
);
2264 total_bytes
= bio_nbytes
= 0;
2265 while ((bio
= req
->bio
) != NULL
) {
2268 if (nr_bytes
>= bio
->bi_size
) {
2269 req
->bio
= bio
->bi_next
;
2270 nbytes
= bio
->bi_size
;
2271 req_bio_endio(req
, bio
, nbytes
, error
);
2275 int idx
= bio
->bi_idx
+ next_idx
;
2277 if (unlikely(idx
>= bio
->bi_vcnt
)) {
2278 blk_dump_rq_flags(req
, "__end_that");
2279 printk(KERN_ERR
"%s: bio idx %d >= vcnt %d\n",
2280 __func__
, idx
, bio
->bi_vcnt
);
2284 nbytes
= bio_iovec_idx(bio
, idx
)->bv_len
;
2285 BIO_BUG_ON(nbytes
> bio
->bi_size
);
2288 * not a complete bvec done
2290 if (unlikely(nbytes
> nr_bytes
)) {
2291 bio_nbytes
+= nr_bytes
;
2292 total_bytes
+= nr_bytes
;
2297 * advance to the next vector
2300 bio_nbytes
+= nbytes
;
2303 total_bytes
+= nbytes
;
2309 * end more in this run, or just return 'not-done'
2311 if (unlikely(nr_bytes
<= 0))
2321 * Reset counters so that the request stacking driver
2322 * can find how many bytes remain in the request
2325 req
->__data_len
= 0;
2330 * if the request wasn't completed, update state
2333 req_bio_endio(req
, bio
, bio_nbytes
, error
);
2334 bio
->bi_idx
+= next_idx
;
2335 bio_iovec(bio
)->bv_offset
+= nr_bytes
;
2336 bio_iovec(bio
)->bv_len
-= nr_bytes
;
2339 req
->__data_len
-= total_bytes
;
2340 req
->buffer
= bio_data(req
->bio
);
2342 /* update sector only for requests with clear definition of sector */
2343 if (req
->cmd_type
== REQ_TYPE_FS
|| (req
->cmd_flags
& REQ_DISCARD
))
2344 req
->__sector
+= total_bytes
>> 9;
2346 /* mixed attributes always follow the first bio */
2347 if (req
->cmd_flags
& REQ_MIXED_MERGE
) {
2348 req
->cmd_flags
&= ~REQ_FAILFAST_MASK
;
2349 req
->cmd_flags
|= req
->bio
->bi_rw
& REQ_FAILFAST_MASK
;
2353 * If total number of sectors is less than the first segment
2354 * size, something has gone terribly wrong.
2356 if (blk_rq_bytes(req
) < blk_rq_cur_bytes(req
)) {
2357 blk_dump_rq_flags(req
, "request botched");
2358 req
->__data_len
= blk_rq_cur_bytes(req
);
2361 /* recalculate the number of segments */
2362 blk_recalc_rq_segments(req
);
2366 EXPORT_SYMBOL_GPL(blk_update_request
);
2368 static bool blk_update_bidi_request(struct request
*rq
, int error
,
2369 unsigned int nr_bytes
,
2370 unsigned int bidi_bytes
)
2372 if (blk_update_request(rq
, error
, nr_bytes
))
2375 /* Bidi request must be completed as a whole */
2376 if (unlikely(blk_bidi_rq(rq
)) &&
2377 blk_update_request(rq
->next_rq
, error
, bidi_bytes
))
2380 if (blk_queue_add_random(rq
->q
))
2381 add_disk_randomness(rq
->rq_disk
);
2387 * blk_unprep_request - unprepare a request
2390 * This function makes a request ready for complete resubmission (or
2391 * completion). It happens only after all error handling is complete,
2392 * so represents the appropriate moment to deallocate any resources
2393 * that were allocated to the request in the prep_rq_fn. The queue
2394 * lock is held when calling this.
2396 void blk_unprep_request(struct request
*req
)
2398 struct request_queue
*q
= req
->q
;
2400 req
->cmd_flags
&= ~REQ_DONTPREP
;
2401 if (q
->unprep_rq_fn
)
2402 q
->unprep_rq_fn(q
, req
);
2404 EXPORT_SYMBOL_GPL(blk_unprep_request
);
2407 * queue lock must be held
2409 static void blk_finish_request(struct request
*req
, int error
)
2411 if (blk_rq_tagged(req
))
2412 blk_queue_end_tag(req
->q
, req
);
2414 BUG_ON(blk_queued_rq(req
));
2416 if (unlikely(laptop_mode
) && req
->cmd_type
== REQ_TYPE_FS
)
2417 laptop_io_completion(&req
->q
->backing_dev_info
);
2419 blk_delete_timer(req
);
2421 if (req
->cmd_flags
& REQ_DONTPREP
)
2422 blk_unprep_request(req
);
2425 blk_account_io_done(req
);
2428 req
->end_io(req
, error
);
2430 if (blk_bidi_rq(req
))
2431 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
2433 __blk_put_request(req
->q
, req
);
2438 * blk_end_bidi_request - Complete a bidi request
2439 * @rq: the request to complete
2440 * @error: %0 for success, < %0 for error
2441 * @nr_bytes: number of bytes to complete @rq
2442 * @bidi_bytes: number of bytes to complete @rq->next_rq
2445 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2446 * Drivers that supports bidi can safely call this member for any
2447 * type of request, bidi or uni. In the later case @bidi_bytes is
2451 * %false - we are done with this request
2452 * %true - still buffers pending for this request
2454 static bool blk_end_bidi_request(struct request
*rq
, int error
,
2455 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2457 struct request_queue
*q
= rq
->q
;
2458 unsigned long flags
;
2460 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2463 spin_lock_irqsave(q
->queue_lock
, flags
);
2464 blk_finish_request(rq
, error
);
2465 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2471 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2472 * @rq: the request to complete
2473 * @error: %0 for success, < %0 for error
2474 * @nr_bytes: number of bytes to complete @rq
2475 * @bidi_bytes: number of bytes to complete @rq->next_rq
2478 * Identical to blk_end_bidi_request() except that queue lock is
2479 * assumed to be locked on entry and remains so on return.
2482 * %false - we are done with this request
2483 * %true - still buffers pending for this request
2485 bool __blk_end_bidi_request(struct request
*rq
, int error
,
2486 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2488 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2491 blk_finish_request(rq
, error
);
2497 * blk_end_request - Helper function for drivers to complete the request.
2498 * @rq: the request being processed
2499 * @error: %0 for success, < %0 for error
2500 * @nr_bytes: number of bytes to complete
2503 * Ends I/O on a number of bytes attached to @rq.
2504 * If @rq has leftover, sets it up for the next range of segments.
2507 * %false - we are done with this request
2508 * %true - still buffers pending for this request
2510 bool blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2512 return blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2514 EXPORT_SYMBOL(blk_end_request
);
2517 * blk_end_request_all - Helper function for drives to finish the request.
2518 * @rq: the request to finish
2519 * @error: %0 for success, < %0 for error
2522 * Completely finish @rq.
2524 void blk_end_request_all(struct request
*rq
, int error
)
2527 unsigned int bidi_bytes
= 0;
2529 if (unlikely(blk_bidi_rq(rq
)))
2530 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2532 pending
= blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2535 EXPORT_SYMBOL(blk_end_request_all
);
2538 * blk_end_request_cur - Helper function to finish the current request chunk.
2539 * @rq: the request to finish the current chunk for
2540 * @error: %0 for success, < %0 for error
2543 * Complete the current consecutively mapped chunk from @rq.
2546 * %false - we are done with this request
2547 * %true - still buffers pending for this request
2549 bool blk_end_request_cur(struct request
*rq
, int error
)
2551 return blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2553 EXPORT_SYMBOL(blk_end_request_cur
);
2556 * blk_end_request_err - Finish a request till the next failure boundary.
2557 * @rq: the request to finish till the next failure boundary for
2558 * @error: must be negative errno
2561 * Complete @rq till the next failure boundary.
2564 * %false - we are done with this request
2565 * %true - still buffers pending for this request
2567 bool blk_end_request_err(struct request
*rq
, int error
)
2569 WARN_ON(error
>= 0);
2570 return blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2572 EXPORT_SYMBOL_GPL(blk_end_request_err
);
2575 * __blk_end_request - Helper function for drivers to complete the request.
2576 * @rq: the request being processed
2577 * @error: %0 for success, < %0 for error
2578 * @nr_bytes: number of bytes to complete
2581 * Must be called with queue lock held unlike blk_end_request().
2584 * %false - we are done with this request
2585 * %true - still buffers pending for this request
2587 bool __blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2589 return __blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2591 EXPORT_SYMBOL(__blk_end_request
);
2594 * __blk_end_request_all - Helper function for drives to finish the request.
2595 * @rq: the request to finish
2596 * @error: %0 for success, < %0 for error
2599 * Completely finish @rq. Must be called with queue lock held.
2601 void __blk_end_request_all(struct request
*rq
, int error
)
2604 unsigned int bidi_bytes
= 0;
2606 if (unlikely(blk_bidi_rq(rq
)))
2607 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2609 pending
= __blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2612 EXPORT_SYMBOL(__blk_end_request_all
);
2615 * __blk_end_request_cur - Helper function to finish the current request chunk.
2616 * @rq: the request to finish the current chunk for
2617 * @error: %0 for success, < %0 for error
2620 * Complete the current consecutively mapped chunk from @rq. Must
2621 * be called with queue lock held.
2624 * %false - we are done with this request
2625 * %true - still buffers pending for this request
2627 bool __blk_end_request_cur(struct request
*rq
, int error
)
2629 return __blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2631 EXPORT_SYMBOL(__blk_end_request_cur
);
2634 * __blk_end_request_err - Finish a request till the next failure boundary.
2635 * @rq: the request to finish till the next failure boundary for
2636 * @error: must be negative errno
2639 * Complete @rq till the next failure boundary. Must be called
2640 * with queue lock held.
2643 * %false - we are done with this request
2644 * %true - still buffers pending for this request
2646 bool __blk_end_request_err(struct request
*rq
, int error
)
2648 WARN_ON(error
>= 0);
2649 return __blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2651 EXPORT_SYMBOL_GPL(__blk_end_request_err
);
2653 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
2656 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2657 rq
->cmd_flags
|= bio
->bi_rw
& REQ_WRITE
;
2659 if (bio_has_data(bio
)) {
2660 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2661 rq
->buffer
= bio_data(bio
);
2663 rq
->__data_len
= bio
->bi_size
;
2664 rq
->bio
= rq
->biotail
= bio
;
2667 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2670 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2672 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2673 * @rq: the request to be flushed
2676 * Flush all pages in @rq.
2678 void rq_flush_dcache_pages(struct request
*rq
)
2680 struct req_iterator iter
;
2681 struct bio_vec
*bvec
;
2683 rq_for_each_segment(bvec
, rq
, iter
)
2684 flush_dcache_page(bvec
->bv_page
);
2686 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages
);
2690 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2691 * @q : the queue of the device being checked
2694 * Check if underlying low-level drivers of a device are busy.
2695 * If the drivers want to export their busy state, they must set own
2696 * exporting function using blk_queue_lld_busy() first.
2698 * Basically, this function is used only by request stacking drivers
2699 * to stop dispatching requests to underlying devices when underlying
2700 * devices are busy. This behavior helps more I/O merging on the queue
2701 * of the request stacking driver and prevents I/O throughput regression
2702 * on burst I/O load.
2705 * 0 - Not busy (The request stacking driver should dispatch request)
2706 * 1 - Busy (The request stacking driver should stop dispatching request)
2708 int blk_lld_busy(struct request_queue
*q
)
2711 return q
->lld_busy_fn(q
);
2715 EXPORT_SYMBOL_GPL(blk_lld_busy
);
2718 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2719 * @rq: the clone request to be cleaned up
2722 * Free all bios in @rq for a cloned request.
2724 void blk_rq_unprep_clone(struct request
*rq
)
2728 while ((bio
= rq
->bio
) != NULL
) {
2729 rq
->bio
= bio
->bi_next
;
2734 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone
);
2737 * Copy attributes of the original request to the clone request.
2738 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2740 static void __blk_rq_prep_clone(struct request
*dst
, struct request
*src
)
2742 dst
->cpu
= src
->cpu
;
2743 dst
->cmd_flags
= (src
->cmd_flags
& REQ_CLONE_MASK
) | REQ_NOMERGE
;
2744 dst
->cmd_type
= src
->cmd_type
;
2745 dst
->__sector
= blk_rq_pos(src
);
2746 dst
->__data_len
= blk_rq_bytes(src
);
2747 dst
->nr_phys_segments
= src
->nr_phys_segments
;
2748 dst
->ioprio
= src
->ioprio
;
2749 dst
->extra_len
= src
->extra_len
;
2753 * blk_rq_prep_clone - Helper function to setup clone request
2754 * @rq: the request to be setup
2755 * @rq_src: original request to be cloned
2756 * @bs: bio_set that bios for clone are allocated from
2757 * @gfp_mask: memory allocation mask for bio
2758 * @bio_ctr: setup function to be called for each clone bio.
2759 * Returns %0 for success, non %0 for failure.
2760 * @data: private data to be passed to @bio_ctr
2763 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2764 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2765 * are not copied, and copying such parts is the caller's responsibility.
2766 * Also, pages which the original bios are pointing to are not copied
2767 * and the cloned bios just point same pages.
2768 * So cloned bios must be completed before original bios, which means
2769 * the caller must complete @rq before @rq_src.
2771 int blk_rq_prep_clone(struct request
*rq
, struct request
*rq_src
,
2772 struct bio_set
*bs
, gfp_t gfp_mask
,
2773 int (*bio_ctr
)(struct bio
*, struct bio
*, void *),
2776 struct bio
*bio
, *bio_src
;
2781 blk_rq_init(NULL
, rq
);
2783 __rq_for_each_bio(bio_src
, rq_src
) {
2784 bio
= bio_alloc_bioset(gfp_mask
, bio_src
->bi_max_vecs
, bs
);
2788 __bio_clone(bio
, bio_src
);
2790 if (bio_integrity(bio_src
) &&
2791 bio_integrity_clone(bio
, bio_src
, gfp_mask
, bs
))
2794 if (bio_ctr
&& bio_ctr(bio
, bio_src
, data
))
2798 rq
->biotail
->bi_next
= bio
;
2801 rq
->bio
= rq
->biotail
= bio
;
2804 __blk_rq_prep_clone(rq
, rq_src
);
2811 blk_rq_unprep_clone(rq
);
2815 EXPORT_SYMBOL_GPL(blk_rq_prep_clone
);
2817 int kblockd_schedule_work(struct request_queue
*q
, struct work_struct
*work
)
2819 return queue_work(kblockd_workqueue
, work
);
2821 EXPORT_SYMBOL(kblockd_schedule_work
);
2823 int kblockd_schedule_delayed_work(struct request_queue
*q
,
2824 struct delayed_work
*dwork
, unsigned long delay
)
2826 return queue_delayed_work(kblockd_workqueue
, dwork
, delay
);
2828 EXPORT_SYMBOL(kblockd_schedule_delayed_work
);
2830 #define PLUG_MAGIC 0x91827364
2833 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2834 * @plug: The &struct blk_plug that needs to be initialized
2837 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2838 * pending I/O should the task end up blocking between blk_start_plug() and
2839 * blk_finish_plug(). This is important from a performance perspective, but
2840 * also ensures that we don't deadlock. For instance, if the task is blocking
2841 * for a memory allocation, memory reclaim could end up wanting to free a
2842 * page belonging to that request that is currently residing in our private
2843 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2844 * this kind of deadlock.
2846 void blk_start_plug(struct blk_plug
*plug
)
2848 struct task_struct
*tsk
= current
;
2850 plug
->magic
= PLUG_MAGIC
;
2851 INIT_LIST_HEAD(&plug
->list
);
2852 INIT_LIST_HEAD(&plug
->cb_list
);
2853 plug
->should_sort
= 0;
2856 * If this is a nested plug, don't actually assign it. It will be
2857 * flushed on its own.
2861 * Store ordering should not be needed here, since a potential
2862 * preempt will imply a full memory barrier
2867 EXPORT_SYMBOL(blk_start_plug
);
2869 static int plug_rq_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
2871 struct request
*rqa
= container_of(a
, struct request
, queuelist
);
2872 struct request
*rqb
= container_of(b
, struct request
, queuelist
);
2874 return !(rqa
->q
<= rqb
->q
);
2878 * If 'from_schedule' is true, then postpone the dispatch of requests
2879 * until a safe kblockd context. We due this to avoid accidental big
2880 * additional stack usage in driver dispatch, in places where the originally
2881 * plugger did not intend it.
2883 static void queue_unplugged(struct request_queue
*q
, unsigned int depth
,
2885 __releases(q
->queue_lock
)
2887 trace_block_unplug(q
, depth
, !from_schedule
);
2890 * Don't mess with dead queue.
2892 if (unlikely(blk_queue_dead(q
))) {
2893 spin_unlock(q
->queue_lock
);
2898 * If we are punting this to kblockd, then we can safely drop
2899 * the queue_lock before waking kblockd (which needs to take
2902 if (from_schedule
) {
2903 spin_unlock(q
->queue_lock
);
2904 blk_run_queue_async(q
);
2907 spin_unlock(q
->queue_lock
);
2912 static void flush_plug_callbacks(struct blk_plug
*plug
, bool from_schedule
)
2914 LIST_HEAD(callbacks
);
2916 while (!list_empty(&plug
->cb_list
)) {
2917 list_splice_init(&plug
->cb_list
, &callbacks
);
2919 while (!list_empty(&callbacks
)) {
2920 struct blk_plug_cb
*cb
= list_first_entry(&callbacks
,
2923 list_del(&cb
->list
);
2924 cb
->callback(cb
, from_schedule
);
2929 struct blk_plug_cb
*blk_check_plugged(blk_plug_cb_fn unplug
, void *data
,
2932 struct blk_plug
*plug
= current
->plug
;
2933 struct blk_plug_cb
*cb
;
2938 list_for_each_entry(cb
, &plug
->cb_list
, list
)
2939 if (cb
->callback
== unplug
&& cb
->data
== data
)
2942 /* Not currently on the callback list */
2943 BUG_ON(size
< sizeof(*cb
));
2944 cb
= kzalloc(size
, GFP_ATOMIC
);
2947 cb
->callback
= unplug
;
2948 list_add(&cb
->list
, &plug
->cb_list
);
2952 EXPORT_SYMBOL(blk_check_plugged
);
2954 void blk_flush_plug_list(struct blk_plug
*plug
, bool from_schedule
)
2956 struct request_queue
*q
;
2957 unsigned long flags
;
2962 BUG_ON(plug
->magic
!= PLUG_MAGIC
);
2964 flush_plug_callbacks(plug
, from_schedule
);
2965 if (list_empty(&plug
->list
))
2968 list_splice_init(&plug
->list
, &list
);
2970 if (plug
->should_sort
) {
2971 list_sort(NULL
, &list
, plug_rq_cmp
);
2972 plug
->should_sort
= 0;
2979 * Save and disable interrupts here, to avoid doing it for every
2980 * queue lock we have to take.
2982 local_irq_save(flags
);
2983 while (!list_empty(&list
)) {
2984 rq
= list_entry_rq(list
.next
);
2985 list_del_init(&rq
->queuelist
);
2989 * This drops the queue lock
2992 queue_unplugged(q
, depth
, from_schedule
);
2995 spin_lock(q
->queue_lock
);
2999 * Short-circuit if @q is dead
3001 if (unlikely(blk_queue_dead(q
))) {
3002 __blk_end_request_all(rq
, -ENODEV
);
3007 * rq is already accounted, so use raw insert
3009 if (rq
->cmd_flags
& (REQ_FLUSH
| REQ_FUA
))
3010 __elv_add_request(q
, rq
, ELEVATOR_INSERT_FLUSH
);
3012 __elv_add_request(q
, rq
, ELEVATOR_INSERT_SORT_MERGE
);
3018 * This drops the queue lock
3021 queue_unplugged(q
, depth
, from_schedule
);
3023 local_irq_restore(flags
);
3026 void blk_finish_plug(struct blk_plug
*plug
)
3028 blk_flush_plug_list(plug
, false);
3030 if (plug
== current
->plug
)
3031 current
->plug
= NULL
;
3033 EXPORT_SYMBOL(blk_finish_plug
);
3035 int __init
blk_dev_init(void)
3037 BUILD_BUG_ON(__REQ_NR_BITS
> 8 *
3038 sizeof(((struct request
*)0)->cmd_flags
));
3040 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3041 kblockd_workqueue
= alloc_workqueue("kblockd",
3042 WQ_MEM_RECLAIM
| WQ_HIGHPRI
, 0);
3043 if (!kblockd_workqueue
)
3044 panic("Failed to create kblockd\n");
3046 request_cachep
= kmem_cache_create("blkdev_requests",
3047 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
3049 blk_requestq_cachep
= kmem_cache_create("blkdev_queue",
3050 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);