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/blktrace_api.h>
30 #include <linux/fault-inject.h>
31 #include <trace/block.h>
35 DEFINE_TRACE(block_plug
);
36 DEFINE_TRACE(block_unplug_io
);
37 DEFINE_TRACE(block_unplug_timer
);
38 DEFINE_TRACE(block_getrq
);
39 DEFINE_TRACE(block_sleeprq
);
40 DEFINE_TRACE(block_rq_requeue
);
41 DEFINE_TRACE(block_bio_backmerge
);
42 DEFINE_TRACE(block_bio_frontmerge
);
43 DEFINE_TRACE(block_bio_queue
);
44 DEFINE_TRACE(block_rq_complete
);
45 DEFINE_TRACE(block_remap
); /* Also used in drivers/md/dm.c */
46 EXPORT_TRACEPOINT_SYMBOL_GPL(block_remap
);
48 static int __make_request(struct request_queue
*q
, struct bio
*bio
);
51 * For the allocated request tables
53 static struct kmem_cache
*request_cachep
;
56 * For queue allocation
58 struct kmem_cache
*blk_requestq_cachep
;
61 * Controlling structure to kblockd
63 static struct workqueue_struct
*kblockd_workqueue
;
65 static void drive_stat_acct(struct request
*rq
, int new_io
)
67 struct hd_struct
*part
;
68 int rw
= rq_data_dir(rq
);
71 if (!blk_do_io_stat(rq
))
74 cpu
= part_stat_lock();
75 part
= disk_map_sector_rcu(rq
->rq_disk
, blk_rq_pos(rq
));
78 part_stat_inc(cpu
, part
, merges
[rw
]);
80 part_round_stats(cpu
, part
);
81 part_inc_in_flight(part
);
87 void blk_queue_congestion_threshold(struct request_queue
*q
)
91 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) + 1;
92 if (nr
> q
->nr_requests
)
94 q
->nr_congestion_on
= nr
;
96 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) - (q
->nr_requests
/ 16) - 1;
99 q
->nr_congestion_off
= nr
;
103 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
106 * Locates the passed device's request queue and returns the address of its
109 * Will return NULL if the request queue cannot be located.
111 struct backing_dev_info
*blk_get_backing_dev_info(struct block_device
*bdev
)
113 struct backing_dev_info
*ret
= NULL
;
114 struct request_queue
*q
= bdev_get_queue(bdev
);
117 ret
= &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
= rq
->hard_sector
= (sector_t
) -1;
131 INIT_HLIST_NODE(&rq
->hash
);
132 RB_CLEAR_NODE(&rq
->rb_node
);
134 rq
->cmd_len
= BLK_MAX_CDB
;
137 rq
->start_time
= jiffies
;
139 EXPORT_SYMBOL(blk_rq_init
);
141 static void req_bio_endio(struct request
*rq
, struct bio
*bio
,
142 unsigned int nbytes
, int error
)
144 struct request_queue
*q
= rq
->q
;
146 if (&q
->bar_rq
!= rq
) {
148 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
149 else if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
152 if (unlikely(nbytes
> bio
->bi_size
)) {
153 printk(KERN_ERR
"%s: want %u bytes done, %u left\n",
154 __func__
, nbytes
, bio
->bi_size
);
155 nbytes
= bio
->bi_size
;
158 if (unlikely(rq
->cmd_flags
& REQ_QUIET
))
159 set_bit(BIO_QUIET
, &bio
->bi_flags
);
161 bio
->bi_size
-= nbytes
;
162 bio
->bi_sector
+= (nbytes
>> 9);
164 if (bio_integrity(bio
))
165 bio_integrity_advance(bio
, nbytes
);
167 if (bio
->bi_size
== 0)
168 bio_endio(bio
, error
);
172 * Okay, this is the barrier request in progress, just
175 if (error
&& !q
->orderr
)
180 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
184 printk(KERN_INFO
"%s: dev %s: type=%x, flags=%x\n", msg
,
185 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->cmd_type
,
188 printk(KERN_INFO
" sector %llu, nr/cnr %u/%u\n",
189 (unsigned long long)blk_rq_pos(rq
),
190 blk_rq_sectors(rq
), blk_rq_cur_sectors(rq
));
191 printk(KERN_INFO
" bio %p, biotail %p, buffer %p, len %u\n",
192 rq
->bio
, rq
->biotail
,
193 rq
->buffer
, rq
->data_len
);
195 if (blk_pc_request(rq
)) {
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
);
205 * "plug" the device if there are no outstanding requests: this will
206 * force the transfer to start only after we have put all the requests
209 * This is called with interrupts off and no requests on the queue and
210 * with the queue lock held.
212 void blk_plug_device(struct request_queue
*q
)
214 WARN_ON(!irqs_disabled());
217 * don't plug a stopped queue, it must be paired with blk_start_queue()
218 * which will restart the queueing
220 if (blk_queue_stopped(q
))
223 if (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED
, q
)) {
224 mod_timer(&q
->unplug_timer
, jiffies
+ q
->unplug_delay
);
228 EXPORT_SYMBOL(blk_plug_device
);
231 * blk_plug_device_unlocked - plug a device without queue lock held
232 * @q: The &struct request_queue to plug
235 * Like @blk_plug_device(), but grabs the queue lock and disables
238 void blk_plug_device_unlocked(struct request_queue
*q
)
242 spin_lock_irqsave(q
->queue_lock
, flags
);
244 spin_unlock_irqrestore(q
->queue_lock
, flags
);
246 EXPORT_SYMBOL(blk_plug_device_unlocked
);
249 * remove the queue from the plugged list, if present. called with
250 * queue lock held and interrupts disabled.
252 int blk_remove_plug(struct request_queue
*q
)
254 WARN_ON(!irqs_disabled());
256 if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED
, q
))
259 del_timer(&q
->unplug_timer
);
262 EXPORT_SYMBOL(blk_remove_plug
);
265 * remove the plug and let it rip..
267 void __generic_unplug_device(struct request_queue
*q
)
269 if (unlikely(blk_queue_stopped(q
)))
271 if (!blk_remove_plug(q
) && !blk_queue_nonrot(q
))
278 * generic_unplug_device - fire a request queue
279 * @q: The &struct request_queue in question
282 * Linux uses plugging to build bigger requests queues before letting
283 * the device have at them. If a queue is plugged, the I/O scheduler
284 * is still adding and merging requests on the queue. Once the queue
285 * gets unplugged, the request_fn defined for the queue is invoked and
288 void generic_unplug_device(struct request_queue
*q
)
290 if (blk_queue_plugged(q
)) {
291 spin_lock_irq(q
->queue_lock
);
292 __generic_unplug_device(q
);
293 spin_unlock_irq(q
->queue_lock
);
296 EXPORT_SYMBOL(generic_unplug_device
);
298 static void blk_backing_dev_unplug(struct backing_dev_info
*bdi
,
301 struct request_queue
*q
= bdi
->unplug_io_data
;
306 void blk_unplug_work(struct work_struct
*work
)
308 struct request_queue
*q
=
309 container_of(work
, struct request_queue
, unplug_work
);
311 trace_block_unplug_io(q
);
315 void blk_unplug_timeout(unsigned long data
)
317 struct request_queue
*q
= (struct request_queue
*)data
;
319 trace_block_unplug_timer(q
);
320 kblockd_schedule_work(q
, &q
->unplug_work
);
323 void blk_unplug(struct request_queue
*q
)
326 * devices don't necessarily have an ->unplug_fn defined
329 trace_block_unplug_io(q
);
333 EXPORT_SYMBOL(blk_unplug
);
336 * blk_start_queue - restart a previously stopped queue
337 * @q: The &struct request_queue in question
340 * blk_start_queue() will clear the stop flag on the queue, and call
341 * the request_fn for the queue if it was in a stopped state when
342 * entered. Also see blk_stop_queue(). Queue lock must be held.
344 void blk_start_queue(struct request_queue
*q
)
346 WARN_ON(!irqs_disabled());
348 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
351 EXPORT_SYMBOL(blk_start_queue
);
354 * blk_stop_queue - stop a queue
355 * @q: The &struct request_queue in question
358 * The Linux block layer assumes that a block driver will consume all
359 * entries on the request queue when the request_fn strategy is called.
360 * Often this will not happen, because of hardware limitations (queue
361 * depth settings). If a device driver gets a 'queue full' response,
362 * or if it simply chooses not to queue more I/O at one point, it can
363 * call this function to prevent the request_fn from being called until
364 * the driver has signalled it's ready to go again. This happens by calling
365 * blk_start_queue() to restart queue operations. Queue lock must be held.
367 void blk_stop_queue(struct request_queue
*q
)
370 queue_flag_set(QUEUE_FLAG_STOPPED
, q
);
372 EXPORT_SYMBOL(blk_stop_queue
);
375 * blk_sync_queue - cancel any pending callbacks on a queue
379 * The block layer may perform asynchronous callback activity
380 * on a queue, such as calling the unplug function after a timeout.
381 * A block device may call blk_sync_queue to ensure that any
382 * such activity is cancelled, thus allowing it to release resources
383 * that the callbacks might use. The caller must already have made sure
384 * that its ->make_request_fn will not re-add plugging prior to calling
388 void blk_sync_queue(struct request_queue
*q
)
390 del_timer_sync(&q
->unplug_timer
);
391 del_timer_sync(&q
->timeout
);
392 cancel_work_sync(&q
->unplug_work
);
394 EXPORT_SYMBOL(blk_sync_queue
);
397 * __blk_run_queue - run a single device queue
398 * @q: The queue to run
401 * See @blk_run_queue. This variant must be called with the queue lock
402 * held and interrupts disabled.
405 void __blk_run_queue(struct request_queue
*q
)
409 if (unlikely(blk_queue_stopped(q
)))
412 if (elv_queue_empty(q
))
416 * Only recurse once to avoid overrunning the stack, let the unplug
417 * handling reinvoke the handler shortly if we already got there.
419 if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER
, q
)) {
421 queue_flag_clear(QUEUE_FLAG_REENTER
, q
);
423 queue_flag_set(QUEUE_FLAG_PLUGGED
, q
);
424 kblockd_schedule_work(q
, &q
->unplug_work
);
427 EXPORT_SYMBOL(__blk_run_queue
);
430 * blk_run_queue - run a single device queue
431 * @q: The queue to run
434 * Invoke request handling on this queue, if it has pending work to do.
435 * May be used to restart queueing when a request has completed.
437 void blk_run_queue(struct request_queue
*q
)
441 spin_lock_irqsave(q
->queue_lock
, flags
);
443 spin_unlock_irqrestore(q
->queue_lock
, flags
);
445 EXPORT_SYMBOL(blk_run_queue
);
447 void blk_put_queue(struct request_queue
*q
)
449 kobject_put(&q
->kobj
);
452 void blk_cleanup_queue(struct request_queue
*q
)
455 * We know we have process context here, so we can be a little
456 * cautious and ensure that pending block actions on this device
457 * are done before moving on. Going into this function, we should
458 * not have processes doing IO to this device.
462 mutex_lock(&q
->sysfs_lock
);
463 queue_flag_set_unlocked(QUEUE_FLAG_DEAD
, q
);
464 mutex_unlock(&q
->sysfs_lock
);
467 elevator_exit(q
->elevator
);
471 EXPORT_SYMBOL(blk_cleanup_queue
);
473 static int blk_init_free_list(struct request_queue
*q
)
475 struct request_list
*rl
= &q
->rq
;
477 rl
->count
[BLK_RW_SYNC
] = rl
->count
[BLK_RW_ASYNC
] = 0;
478 rl
->starved
[BLK_RW_SYNC
] = rl
->starved
[BLK_RW_ASYNC
] = 0;
480 init_waitqueue_head(&rl
->wait
[BLK_RW_SYNC
]);
481 init_waitqueue_head(&rl
->wait
[BLK_RW_ASYNC
]);
483 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, mempool_alloc_slab
,
484 mempool_free_slab
, request_cachep
, q
->node
);
492 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
494 return blk_alloc_queue_node(gfp_mask
, -1);
496 EXPORT_SYMBOL(blk_alloc_queue
);
498 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
500 struct request_queue
*q
;
503 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
504 gfp_mask
| __GFP_ZERO
, node_id
);
508 q
->backing_dev_info
.unplug_io_fn
= blk_backing_dev_unplug
;
509 q
->backing_dev_info
.unplug_io_data
= q
;
510 err
= bdi_init(&q
->backing_dev_info
);
512 kmem_cache_free(blk_requestq_cachep
, q
);
516 init_timer(&q
->unplug_timer
);
517 setup_timer(&q
->timeout
, blk_rq_timed_out_timer
, (unsigned long) q
);
518 INIT_LIST_HEAD(&q
->timeout_list
);
519 INIT_WORK(&q
->unplug_work
, blk_unplug_work
);
521 kobject_init(&q
->kobj
, &blk_queue_ktype
);
523 mutex_init(&q
->sysfs_lock
);
524 spin_lock_init(&q
->__queue_lock
);
528 EXPORT_SYMBOL(blk_alloc_queue_node
);
531 * blk_init_queue - prepare a request queue for use with a block device
532 * @rfn: The function to be called to process requests that have been
533 * placed on the queue.
534 * @lock: Request queue spin lock
537 * If a block device wishes to use the standard request handling procedures,
538 * which sorts requests and coalesces adjacent requests, then it must
539 * call blk_init_queue(). The function @rfn will be called when there
540 * are requests on the queue that need to be processed. If the device
541 * supports plugging, then @rfn may not be called immediately when requests
542 * are available on the queue, but may be called at some time later instead.
543 * Plugged queues are generally unplugged when a buffer belonging to one
544 * of the requests on the queue is needed, or due to memory pressure.
546 * @rfn is not required, or even expected, to remove all requests off the
547 * queue, but only as many as it can handle at a time. If it does leave
548 * requests on the queue, it is responsible for arranging that the requests
549 * get dealt with eventually.
551 * The queue spin lock must be held while manipulating the requests on the
552 * request queue; this lock will be taken also from interrupt context, so irq
553 * disabling is needed for it.
555 * Function returns a pointer to the initialized request queue, or %NULL if
559 * blk_init_queue() must be paired with a blk_cleanup_queue() call
560 * when the block device is deactivated (such as at module unload).
563 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
565 return blk_init_queue_node(rfn
, lock
, -1);
567 EXPORT_SYMBOL(blk_init_queue
);
569 struct request_queue
*
570 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
572 struct request_queue
*q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
578 if (blk_init_free_list(q
)) {
579 kmem_cache_free(blk_requestq_cachep
, q
);
584 * if caller didn't supply a lock, they get per-queue locking with
588 lock
= &q
->__queue_lock
;
591 q
->prep_rq_fn
= NULL
;
592 q
->unplug_fn
= generic_unplug_device
;
593 q
->queue_flags
= QUEUE_FLAG_DEFAULT
;
594 q
->queue_lock
= lock
;
597 * This also sets hw/phys segments, boundary and size
599 blk_queue_make_request(q
, __make_request
);
601 q
->sg_reserved_size
= INT_MAX
;
603 blk_set_cmd_filter_defaults(&q
->cmd_filter
);
608 if (!elevator_init(q
, NULL
)) {
609 blk_queue_congestion_threshold(q
);
616 EXPORT_SYMBOL(blk_init_queue_node
);
618 int blk_get_queue(struct request_queue
*q
)
620 if (likely(!test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
))) {
621 kobject_get(&q
->kobj
);
628 static inline void blk_free_request(struct request_queue
*q
, struct request
*rq
)
630 if (rq
->cmd_flags
& REQ_ELVPRIV
)
631 elv_put_request(q
, rq
);
632 mempool_free(rq
, q
->rq
.rq_pool
);
635 static struct request
*
636 blk_alloc_request(struct request_queue
*q
, int flags
, int priv
, gfp_t gfp_mask
)
638 struct request
*rq
= mempool_alloc(q
->rq
.rq_pool
, gfp_mask
);
645 rq
->cmd_flags
= flags
| REQ_ALLOCED
;
648 if (unlikely(elv_set_request(q
, rq
, gfp_mask
))) {
649 mempool_free(rq
, q
->rq
.rq_pool
);
652 rq
->cmd_flags
|= REQ_ELVPRIV
;
659 * ioc_batching returns true if the ioc is a valid batching request and
660 * should be given priority access to a request.
662 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
668 * Make sure the process is able to allocate at least 1 request
669 * even if the batch times out, otherwise we could theoretically
672 return ioc
->nr_batch_requests
== q
->nr_batching
||
673 (ioc
->nr_batch_requests
> 0
674 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
678 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
679 * will cause the process to be a "batcher" on all queues in the system. This
680 * is the behaviour we want though - once it gets a wakeup it should be given
683 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
685 if (!ioc
|| ioc_batching(q
, ioc
))
688 ioc
->nr_batch_requests
= q
->nr_batching
;
689 ioc
->last_waited
= jiffies
;
692 static void __freed_request(struct request_queue
*q
, int sync
)
694 struct request_list
*rl
= &q
->rq
;
696 if (rl
->count
[sync
] < queue_congestion_off_threshold(q
))
697 blk_clear_queue_congested(q
, sync
);
699 if (rl
->count
[sync
] + 1 <= q
->nr_requests
) {
700 if (waitqueue_active(&rl
->wait
[sync
]))
701 wake_up(&rl
->wait
[sync
]);
703 blk_clear_queue_full(q
, sync
);
708 * A request has just been released. Account for it, update the full and
709 * congestion status, wake up any waiters. Called under q->queue_lock.
711 static void freed_request(struct request_queue
*q
, int sync
, int priv
)
713 struct request_list
*rl
= &q
->rq
;
719 __freed_request(q
, sync
);
721 if (unlikely(rl
->starved
[sync
^ 1]))
722 __freed_request(q
, sync
^ 1);
726 * Get a free request, queue_lock must be held.
727 * Returns NULL on failure, with queue_lock held.
728 * Returns !NULL on success, with queue_lock *not held*.
730 static struct request
*get_request(struct request_queue
*q
, int rw_flags
,
731 struct bio
*bio
, gfp_t gfp_mask
)
733 struct request
*rq
= NULL
;
734 struct request_list
*rl
= &q
->rq
;
735 struct io_context
*ioc
= NULL
;
736 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
739 may_queue
= elv_may_queue(q
, rw_flags
);
740 if (may_queue
== ELV_MQUEUE_NO
)
743 if (rl
->count
[is_sync
]+1 >= queue_congestion_on_threshold(q
)) {
744 if (rl
->count
[is_sync
]+1 >= q
->nr_requests
) {
745 ioc
= current_io_context(GFP_ATOMIC
, q
->node
);
747 * The queue will fill after this allocation, so set
748 * it as full, and mark this process as "batching".
749 * This process will be allowed to complete a batch of
750 * requests, others will be blocked.
752 if (!blk_queue_full(q
, is_sync
)) {
753 ioc_set_batching(q
, ioc
);
754 blk_set_queue_full(q
, is_sync
);
756 if (may_queue
!= ELV_MQUEUE_MUST
757 && !ioc_batching(q
, ioc
)) {
759 * The queue is full and the allocating
760 * process is not a "batcher", and not
761 * exempted by the IO scheduler
767 blk_set_queue_congested(q
, is_sync
);
771 * Only allow batching queuers to allocate up to 50% over the defined
772 * limit of requests, otherwise we could have thousands of requests
773 * allocated with any setting of ->nr_requests
775 if (rl
->count
[is_sync
] >= (3 * q
->nr_requests
/ 2))
778 rl
->count
[is_sync
]++;
779 rl
->starved
[is_sync
] = 0;
781 priv
= !test_bit(QUEUE_FLAG_ELVSWITCH
, &q
->queue_flags
);
785 if (blk_queue_io_stat(q
))
786 rw_flags
|= REQ_IO_STAT
;
787 spin_unlock_irq(q
->queue_lock
);
789 rq
= blk_alloc_request(q
, rw_flags
, priv
, gfp_mask
);
792 * Allocation failed presumably due to memory. Undo anything
793 * we might have messed up.
795 * Allocating task should really be put onto the front of the
796 * wait queue, but this is pretty rare.
798 spin_lock_irq(q
->queue_lock
);
799 freed_request(q
, is_sync
, priv
);
802 * in the very unlikely event that allocation failed and no
803 * requests for this direction was pending, mark us starved
804 * so that freeing of a request in the other direction will
805 * notice us. another possible fix would be to split the
806 * rq mempool into READ and WRITE
809 if (unlikely(rl
->count
[is_sync
] == 0))
810 rl
->starved
[is_sync
] = 1;
816 * ioc may be NULL here, and ioc_batching will be false. That's
817 * OK, if the queue is under the request limit then requests need
818 * not count toward the nr_batch_requests limit. There will always
819 * be some limit enforced by BLK_BATCH_TIME.
821 if (ioc_batching(q
, ioc
))
822 ioc
->nr_batch_requests
--;
824 trace_block_getrq(q
, bio
, rw_flags
& 1);
830 * No available requests for this queue, unplug the device and wait for some
831 * requests to become available.
833 * Called with q->queue_lock held, and returns with it unlocked.
835 static struct request
*get_request_wait(struct request_queue
*q
, int rw_flags
,
838 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
841 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
844 struct io_context
*ioc
;
845 struct request_list
*rl
= &q
->rq
;
847 prepare_to_wait_exclusive(&rl
->wait
[is_sync
], &wait
,
848 TASK_UNINTERRUPTIBLE
);
850 trace_block_sleeprq(q
, bio
, rw_flags
& 1);
852 __generic_unplug_device(q
);
853 spin_unlock_irq(q
->queue_lock
);
857 * After sleeping, we become a "batching" process and
858 * will be able to allocate at least one request, and
859 * up to a big batch of them for a small period time.
860 * See ioc_batching, ioc_set_batching
862 ioc
= current_io_context(GFP_NOIO
, q
->node
);
863 ioc_set_batching(q
, ioc
);
865 spin_lock_irq(q
->queue_lock
);
866 finish_wait(&rl
->wait
[is_sync
], &wait
);
868 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
874 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
878 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
880 spin_lock_irq(q
->queue_lock
);
881 if (gfp_mask
& __GFP_WAIT
) {
882 rq
= get_request_wait(q
, rw
, NULL
);
884 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
886 spin_unlock_irq(q
->queue_lock
);
888 /* q->queue_lock is unlocked at this point */
892 EXPORT_SYMBOL(blk_get_request
);
895 * blk_requeue_request - put a request back on queue
896 * @q: request queue where request should be inserted
897 * @rq: request to be inserted
900 * Drivers often keep queueing requests until the hardware cannot accept
901 * more, when that condition happens we need to put the request back
902 * on the queue. Must be called with queue lock held.
904 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
906 blk_delete_timer(rq
);
907 blk_clear_rq_complete(rq
);
908 trace_block_rq_requeue(q
, rq
);
910 if (blk_rq_tagged(rq
))
911 blk_queue_end_tag(q
, rq
);
913 elv_requeue_request(q
, rq
);
915 EXPORT_SYMBOL(blk_requeue_request
);
918 * blk_insert_request - insert a special request into a request queue
919 * @q: request queue where request should be inserted
920 * @rq: request to be inserted
921 * @at_head: insert request at head or tail of queue
922 * @data: private data
925 * Many block devices need to execute commands asynchronously, so they don't
926 * block the whole kernel from preemption during request execution. This is
927 * accomplished normally by inserting aritficial requests tagged as
928 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
929 * be scheduled for actual execution by the request queue.
931 * We have the option of inserting the head or the tail of the queue.
932 * Typically we use the tail for new ioctls and so forth. We use the head
933 * of the queue for things like a QUEUE_FULL message from a device, or a
934 * host that is unable to accept a particular command.
936 void blk_insert_request(struct request_queue
*q
, struct request
*rq
,
937 int at_head
, void *data
)
939 int where
= at_head
? ELEVATOR_INSERT_FRONT
: ELEVATOR_INSERT_BACK
;
943 * tell I/O scheduler that this isn't a regular read/write (ie it
944 * must not attempt merges on this) and that it acts as a soft
947 rq
->cmd_type
= REQ_TYPE_SPECIAL
;
951 spin_lock_irqsave(q
->queue_lock
, flags
);
954 * If command is tagged, release the tag
956 if (blk_rq_tagged(rq
))
957 blk_queue_end_tag(q
, rq
);
959 drive_stat_acct(rq
, 1);
960 __elv_add_request(q
, rq
, where
, 0);
962 spin_unlock_irqrestore(q
->queue_lock
, flags
);
964 EXPORT_SYMBOL(blk_insert_request
);
967 * add-request adds a request to the linked list.
968 * queue lock is held and interrupts disabled, as we muck with the
969 * request queue list.
971 static inline void add_request(struct request_queue
*q
, struct request
*req
)
973 drive_stat_acct(req
, 1);
976 * elevator indicated where it wants this request to be
977 * inserted at elevator_merge time
979 __elv_add_request(q
, req
, ELEVATOR_INSERT_SORT
, 0);
982 static void part_round_stats_single(int cpu
, struct hd_struct
*part
,
985 if (now
== part
->stamp
)
988 if (part
->in_flight
) {
989 __part_stat_add(cpu
, part
, time_in_queue
,
990 part
->in_flight
* (now
- part
->stamp
));
991 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
997 * part_round_stats() - Round off the performance stats on a struct disk_stats.
998 * @cpu: cpu number for stats access
999 * @part: target partition
1001 * The average IO queue length and utilisation statistics are maintained
1002 * by observing the current state of the queue length and the amount of
1003 * time it has been in this state for.
1005 * Normally, that accounting is done on IO completion, but that can result
1006 * in more than a second's worth of IO being accounted for within any one
1007 * second, leading to >100% utilisation. To deal with that, we call this
1008 * function to do a round-off before returning the results when reading
1009 * /proc/diskstats. This accounts immediately for all queue usage up to
1010 * the current jiffies and restarts the counters again.
1012 void part_round_stats(int cpu
, struct hd_struct
*part
)
1014 unsigned long now
= jiffies
;
1017 part_round_stats_single(cpu
, &part_to_disk(part
)->part0
, now
);
1018 part_round_stats_single(cpu
, part
, now
);
1020 EXPORT_SYMBOL_GPL(part_round_stats
);
1023 * queue lock must be held
1025 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1029 if (unlikely(--req
->ref_count
))
1032 elv_completed_request(q
, req
);
1034 /* this is a bio leak */
1035 WARN_ON(req
->bio
!= NULL
);
1038 * Request may not have originated from ll_rw_blk. if not,
1039 * it didn't come out of our reserved rq pools
1041 if (req
->cmd_flags
& REQ_ALLOCED
) {
1042 int is_sync
= rq_is_sync(req
) != 0;
1043 int priv
= req
->cmd_flags
& REQ_ELVPRIV
;
1045 BUG_ON(!list_empty(&req
->queuelist
));
1046 BUG_ON(!hlist_unhashed(&req
->hash
));
1048 blk_free_request(q
, req
);
1049 freed_request(q
, is_sync
, priv
);
1052 EXPORT_SYMBOL_GPL(__blk_put_request
);
1054 void blk_put_request(struct request
*req
)
1056 unsigned long flags
;
1057 struct request_queue
*q
= req
->q
;
1059 spin_lock_irqsave(q
->queue_lock
, flags
);
1060 __blk_put_request(q
, req
);
1061 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1063 EXPORT_SYMBOL(blk_put_request
);
1065 void init_request_from_bio(struct request
*req
, struct bio
*bio
)
1067 req
->cpu
= bio
->bi_comp_cpu
;
1068 req
->cmd_type
= REQ_TYPE_FS
;
1071 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
1073 if (bio_rw_ahead(bio
))
1074 req
->cmd_flags
|= (REQ_FAILFAST_DEV
| REQ_FAILFAST_TRANSPORT
|
1075 REQ_FAILFAST_DRIVER
);
1076 if (bio_failfast_dev(bio
))
1077 req
->cmd_flags
|= REQ_FAILFAST_DEV
;
1078 if (bio_failfast_transport(bio
))
1079 req
->cmd_flags
|= REQ_FAILFAST_TRANSPORT
;
1080 if (bio_failfast_driver(bio
))
1081 req
->cmd_flags
|= REQ_FAILFAST_DRIVER
;
1083 if (unlikely(bio_discard(bio
))) {
1084 req
->cmd_flags
|= REQ_DISCARD
;
1085 if (bio_barrier(bio
))
1086 req
->cmd_flags
|= REQ_SOFTBARRIER
;
1087 req
->q
->prepare_discard_fn(req
->q
, req
);
1088 } else if (unlikely(bio_barrier(bio
)))
1089 req
->cmd_flags
|= REQ_HARDBARRIER
;
1092 req
->cmd_flags
|= REQ_RW_SYNC
;
1093 if (bio_rw_meta(bio
))
1094 req
->cmd_flags
|= REQ_RW_META
;
1095 if (bio_noidle(bio
))
1096 req
->cmd_flags
|= REQ_NOIDLE
;
1099 req
->hard_sector
= req
->sector
= bio
->bi_sector
;
1100 req
->ioprio
= bio_prio(bio
);
1101 blk_rq_bio_prep(req
->q
, req
, bio
);
1105 * Only disabling plugging for non-rotational devices if it does tagging
1106 * as well, otherwise we do need the proper merging
1108 static inline bool queue_should_plug(struct request_queue
*q
)
1110 return !(blk_queue_nonrot(q
) && blk_queue_tagged(q
));
1113 static int __make_request(struct request_queue
*q
, struct bio
*bio
)
1115 struct request
*req
;
1116 int el_ret
, nr_sectors
;
1117 const unsigned short prio
= bio_prio(bio
);
1118 const int sync
= bio_sync(bio
);
1119 const int unplug
= bio_unplug(bio
);
1122 nr_sectors
= bio_sectors(bio
);
1125 * low level driver can indicate that it wants pages above a
1126 * certain limit bounced to low memory (ie for highmem, or even
1127 * ISA dma in theory)
1129 blk_queue_bounce(q
, &bio
);
1131 spin_lock_irq(q
->queue_lock
);
1133 if (unlikely(bio_barrier(bio
)) || elv_queue_empty(q
))
1136 el_ret
= elv_merge(q
, &req
, bio
);
1138 case ELEVATOR_BACK_MERGE
:
1139 BUG_ON(!rq_mergeable(req
));
1141 if (!ll_back_merge_fn(q
, req
, bio
))
1144 trace_block_bio_backmerge(q
, bio
);
1146 req
->biotail
->bi_next
= bio
;
1148 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
1149 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
1150 if (!blk_rq_cpu_valid(req
))
1151 req
->cpu
= bio
->bi_comp_cpu
;
1152 drive_stat_acct(req
, 0);
1153 if (!attempt_back_merge(q
, req
))
1154 elv_merged_request(q
, req
, el_ret
);
1157 case ELEVATOR_FRONT_MERGE
:
1158 BUG_ON(!rq_mergeable(req
));
1160 if (!ll_front_merge_fn(q
, req
, bio
))
1163 trace_block_bio_frontmerge(q
, bio
);
1165 bio
->bi_next
= req
->bio
;
1169 * may not be valid. if the low level driver said
1170 * it didn't need a bounce buffer then it better
1171 * not touch req->buffer either...
1173 req
->buffer
= bio_data(bio
);
1174 req
->current_nr_sectors
= bio_cur_sectors(bio
);
1175 req
->hard_cur_sectors
= req
->current_nr_sectors
;
1176 req
->sector
= req
->hard_sector
= bio
->bi_sector
;
1177 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
1178 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
1179 if (!blk_rq_cpu_valid(req
))
1180 req
->cpu
= bio
->bi_comp_cpu
;
1181 drive_stat_acct(req
, 0);
1182 if (!attempt_front_merge(q
, req
))
1183 elv_merged_request(q
, req
, el_ret
);
1186 /* ELV_NO_MERGE: elevator says don't/can't merge. */
1193 * This sync check and mask will be re-done in init_request_from_bio(),
1194 * but we need to set it earlier to expose the sync flag to the
1195 * rq allocator and io schedulers.
1197 rw_flags
= bio_data_dir(bio
);
1199 rw_flags
|= REQ_RW_SYNC
;
1202 * Grab a free request. This is might sleep but can not fail.
1203 * Returns with the queue unlocked.
1205 req
= get_request_wait(q
, rw_flags
, bio
);
1208 * After dropping the lock and possibly sleeping here, our request
1209 * may now be mergeable after it had proven unmergeable (above).
1210 * We don't worry about that case for efficiency. It won't happen
1211 * often, and the elevators are able to handle it.
1213 init_request_from_bio(req
, bio
);
1215 spin_lock_irq(q
->queue_lock
);
1216 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
) ||
1217 bio_flagged(bio
, BIO_CPU_AFFINE
))
1218 req
->cpu
= blk_cpu_to_group(smp_processor_id());
1219 if (queue_should_plug(q
) && elv_queue_empty(q
))
1221 add_request(q
, req
);
1223 if (unplug
|| !queue_should_plug(q
))
1224 __generic_unplug_device(q
);
1225 spin_unlock_irq(q
->queue_lock
);
1230 * If bio->bi_dev is a partition, remap the location
1232 static inline void blk_partition_remap(struct bio
*bio
)
1234 struct block_device
*bdev
= bio
->bi_bdev
;
1236 if (bio_sectors(bio
) && bdev
!= bdev
->bd_contains
) {
1237 struct hd_struct
*p
= bdev
->bd_part
;
1239 bio
->bi_sector
+= p
->start_sect
;
1240 bio
->bi_bdev
= bdev
->bd_contains
;
1242 trace_block_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
1243 bdev
->bd_dev
, bio
->bi_sector
,
1244 bio
->bi_sector
- p
->start_sect
);
1248 static void handle_bad_sector(struct bio
*bio
)
1250 char b
[BDEVNAME_SIZE
];
1252 printk(KERN_INFO
"attempt to access beyond end of device\n");
1253 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
1254 bdevname(bio
->bi_bdev
, b
),
1256 (unsigned long long)bio
->bi_sector
+ bio_sectors(bio
),
1257 (long long)(bio
->bi_bdev
->bd_inode
->i_size
>> 9));
1259 set_bit(BIO_EOF
, &bio
->bi_flags
);
1262 #ifdef CONFIG_FAIL_MAKE_REQUEST
1264 static DECLARE_FAULT_ATTR(fail_make_request
);
1266 static int __init
setup_fail_make_request(char *str
)
1268 return setup_fault_attr(&fail_make_request
, str
);
1270 __setup("fail_make_request=", setup_fail_make_request
);
1272 static int should_fail_request(struct bio
*bio
)
1274 struct hd_struct
*part
= bio
->bi_bdev
->bd_part
;
1276 if (part_to_disk(part
)->part0
.make_it_fail
|| part
->make_it_fail
)
1277 return should_fail(&fail_make_request
, bio
->bi_size
);
1282 static int __init
fail_make_request_debugfs(void)
1284 return init_fault_attr_dentries(&fail_make_request
,
1285 "fail_make_request");
1288 late_initcall(fail_make_request_debugfs
);
1290 #else /* CONFIG_FAIL_MAKE_REQUEST */
1292 static inline int should_fail_request(struct bio
*bio
)
1297 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1300 * Check whether this bio extends beyond the end of the device.
1302 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
1309 /* Test device or partition size, when known. */
1310 maxsector
= bio
->bi_bdev
->bd_inode
->i_size
>> 9;
1312 sector_t sector
= bio
->bi_sector
;
1314 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
1316 * This may well happen - the kernel calls bread()
1317 * without checking the size of the device, e.g., when
1318 * mounting a device.
1320 handle_bad_sector(bio
);
1329 * generic_make_request - hand a buffer to its device driver for I/O
1330 * @bio: The bio describing the location in memory and on the device.
1332 * generic_make_request() is used to make I/O requests of block
1333 * devices. It is passed a &struct bio, which describes the I/O that needs
1336 * generic_make_request() does not return any status. The
1337 * success/failure status of the request, along with notification of
1338 * completion, is delivered asynchronously through the bio->bi_end_io
1339 * function described (one day) else where.
1341 * The caller of generic_make_request must make sure that bi_io_vec
1342 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1343 * set to describe the device address, and the
1344 * bi_end_io and optionally bi_private are set to describe how
1345 * completion notification should be signaled.
1347 * generic_make_request and the drivers it calls may use bi_next if this
1348 * bio happens to be merged with someone else, and may change bi_dev and
1349 * bi_sector for remaps as it sees fit. So the values of these fields
1350 * should NOT be depended on after the call to generic_make_request.
1352 static inline void __generic_make_request(struct bio
*bio
)
1354 struct request_queue
*q
;
1355 sector_t old_sector
;
1356 int ret
, nr_sectors
= bio_sectors(bio
);
1362 if (bio_check_eod(bio
, nr_sectors
))
1366 * Resolve the mapping until finished. (drivers are
1367 * still free to implement/resolve their own stacking
1368 * by explicitly returning 0)
1370 * NOTE: we don't repeat the blk_size check for each new device.
1371 * Stacking drivers are expected to know what they are doing.
1376 char b
[BDEVNAME_SIZE
];
1378 q
= bdev_get_queue(bio
->bi_bdev
);
1381 "generic_make_request: Trying to access "
1382 "nonexistent block-device %s (%Lu)\n",
1383 bdevname(bio
->bi_bdev
, b
),
1384 (long long) bio
->bi_sector
);
1388 if (unlikely(nr_sectors
> q
->max_hw_sectors
)) {
1389 printk(KERN_ERR
"bio too big device %s (%u > %u)\n",
1390 bdevname(bio
->bi_bdev
, b
),
1396 if (unlikely(test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)))
1399 if (should_fail_request(bio
))
1403 * If this device has partitions, remap block n
1404 * of partition p to block n+start(p) of the disk.
1406 blk_partition_remap(bio
);
1408 if (bio_integrity_enabled(bio
) && bio_integrity_prep(bio
))
1411 if (old_sector
!= -1)
1412 trace_block_remap(q
, bio
, old_dev
, bio
->bi_sector
,
1415 trace_block_bio_queue(q
, bio
);
1417 old_sector
= bio
->bi_sector
;
1418 old_dev
= bio
->bi_bdev
->bd_dev
;
1420 if (bio_check_eod(bio
, nr_sectors
))
1423 if (bio_discard(bio
) && !q
->prepare_discard_fn
) {
1427 if (bio_barrier(bio
) && bio_has_data(bio
) &&
1428 (q
->next_ordered
== QUEUE_ORDERED_NONE
)) {
1433 ret
= q
->make_request_fn(q
, bio
);
1439 bio_endio(bio
, err
);
1443 * We only want one ->make_request_fn to be active at a time,
1444 * else stack usage with stacked devices could be a problem.
1445 * So use current->bio_{list,tail} to keep a list of requests
1446 * submited by a make_request_fn function.
1447 * current->bio_tail is also used as a flag to say if
1448 * generic_make_request is currently active in this task or not.
1449 * If it is NULL, then no make_request is active. If it is non-NULL,
1450 * then a make_request is active, and new requests should be added
1453 void generic_make_request(struct bio
*bio
)
1455 if (current
->bio_tail
) {
1456 /* make_request is active */
1457 *(current
->bio_tail
) = bio
;
1458 bio
->bi_next
= NULL
;
1459 current
->bio_tail
= &bio
->bi_next
;
1462 /* following loop may be a bit non-obvious, and so deserves some
1464 * Before entering the loop, bio->bi_next is NULL (as all callers
1465 * ensure that) so we have a list with a single bio.
1466 * We pretend that we have just taken it off a longer list, so
1467 * we assign bio_list to the next (which is NULL) and bio_tail
1468 * to &bio_list, thus initialising the bio_list of new bios to be
1469 * added. __generic_make_request may indeed add some more bios
1470 * through a recursive call to generic_make_request. If it
1471 * did, we find a non-NULL value in bio_list and re-enter the loop
1472 * from the top. In this case we really did just take the bio
1473 * of the top of the list (no pretending) and so fixup bio_list and
1474 * bio_tail or bi_next, and call into __generic_make_request again.
1476 * The loop was structured like this to make only one call to
1477 * __generic_make_request (which is important as it is large and
1478 * inlined) and to keep the structure simple.
1480 BUG_ON(bio
->bi_next
);
1482 current
->bio_list
= bio
->bi_next
;
1483 if (bio
->bi_next
== NULL
)
1484 current
->bio_tail
= ¤t
->bio_list
;
1486 bio
->bi_next
= NULL
;
1487 __generic_make_request(bio
);
1488 bio
= current
->bio_list
;
1490 current
->bio_tail
= NULL
; /* deactivate */
1492 EXPORT_SYMBOL(generic_make_request
);
1495 * submit_bio - submit a bio to the block device layer for I/O
1496 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1497 * @bio: The &struct bio which describes the I/O
1499 * submit_bio() is very similar in purpose to generic_make_request(), and
1500 * uses that function to do most of the work. Both are fairly rough
1501 * interfaces; @bio must be presetup and ready for I/O.
1504 void submit_bio(int rw
, struct bio
*bio
)
1506 int count
= bio_sectors(bio
);
1511 * If it's a regular read/write or a barrier with data attached,
1512 * go through the normal accounting stuff before submission.
1514 if (bio_has_data(bio
)) {
1516 count_vm_events(PGPGOUT
, count
);
1518 task_io_account_read(bio
->bi_size
);
1519 count_vm_events(PGPGIN
, count
);
1522 if (unlikely(block_dump
)) {
1523 char b
[BDEVNAME_SIZE
];
1524 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s\n",
1525 current
->comm
, task_pid_nr(current
),
1526 (rw
& WRITE
) ? "WRITE" : "READ",
1527 (unsigned long long)bio
->bi_sector
,
1528 bdevname(bio
->bi_bdev
, b
));
1532 generic_make_request(bio
);
1534 EXPORT_SYMBOL(submit_bio
);
1537 * blk_rq_check_limits - Helper function to check a request for the queue limit
1539 * @rq: the request being checked
1542 * @rq may have been made based on weaker limitations of upper-level queues
1543 * in request stacking drivers, and it may violate the limitation of @q.
1544 * Since the block layer and the underlying device driver trust @rq
1545 * after it is inserted to @q, it should be checked against @q before
1546 * the insertion using this generic function.
1548 * This function should also be useful for request stacking drivers
1549 * in some cases below, so export this fuction.
1550 * Request stacking drivers like request-based dm may change the queue
1551 * limits while requests are in the queue (e.g. dm's table swapping).
1552 * Such request stacking drivers should check those requests agaist
1553 * the new queue limits again when they dispatch those requests,
1554 * although such checkings are also done against the old queue limits
1555 * when submitting requests.
1557 int blk_rq_check_limits(struct request_queue
*q
, struct request
*rq
)
1559 if (blk_rq_sectors(rq
) > q
->max_sectors
||
1560 rq
->data_len
> q
->max_hw_sectors
<< 9) {
1561 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
1566 * queue's settings related to segment counting like q->bounce_pfn
1567 * may differ from that of other stacking queues.
1568 * Recalculate it to check the request correctly on this queue's
1571 blk_recalc_rq_segments(rq
);
1572 if (rq
->nr_phys_segments
> q
->max_phys_segments
||
1573 rq
->nr_phys_segments
> q
->max_hw_segments
) {
1574 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
1580 EXPORT_SYMBOL_GPL(blk_rq_check_limits
);
1583 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1584 * @q: the queue to submit the request
1585 * @rq: the request being queued
1587 int blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
1589 unsigned long flags
;
1591 if (blk_rq_check_limits(q
, rq
))
1594 #ifdef CONFIG_FAIL_MAKE_REQUEST
1595 if (rq
->rq_disk
&& rq
->rq_disk
->part0
.make_it_fail
&&
1596 should_fail(&fail_make_request
, blk_rq_bytes(rq
)))
1600 spin_lock_irqsave(q
->queue_lock
, flags
);
1603 * Submitting request must be dequeued before calling this function
1604 * because it will be linked to another request_queue
1606 BUG_ON(blk_queued_rq(rq
));
1608 drive_stat_acct(rq
, 1);
1609 __elv_add_request(q
, rq
, ELEVATOR_INSERT_BACK
, 0);
1611 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1615 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
1618 * blkdev_dequeue_request - dequeue request and start timeout timer
1619 * @req: request to dequeue
1621 * Dequeue @req and start timeout timer on it. This hands off the
1622 * request to the driver.
1624 * Block internal functions which don't want to start timer should
1625 * call elv_dequeue_request().
1627 void blkdev_dequeue_request(struct request
*req
)
1629 elv_dequeue_request(req
->q
, req
);
1632 * We are now handing the request to the hardware, add the
1637 EXPORT_SYMBOL(blkdev_dequeue_request
);
1639 static void blk_account_io_completion(struct request
*req
, unsigned int bytes
)
1641 if (blk_do_io_stat(req
)) {
1642 const int rw
= rq_data_dir(req
);
1643 struct hd_struct
*part
;
1646 cpu
= part_stat_lock();
1647 part
= disk_map_sector_rcu(req
->rq_disk
, blk_rq_pos(req
));
1648 part_stat_add(cpu
, part
, sectors
[rw
], bytes
>> 9);
1653 static void blk_account_io_done(struct request
*req
)
1656 * Account IO completion. bar_rq isn't accounted as a normal
1657 * IO on queueing nor completion. Accounting the containing
1658 * request is enough.
1660 if (blk_do_io_stat(req
) && req
!= &req
->q
->bar_rq
) {
1661 unsigned long duration
= jiffies
- req
->start_time
;
1662 const int rw
= rq_data_dir(req
);
1663 struct hd_struct
*part
;
1666 cpu
= part_stat_lock();
1667 part
= disk_map_sector_rcu(req
->rq_disk
, blk_rq_pos(req
));
1669 part_stat_inc(cpu
, part
, ios
[rw
]);
1670 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
1671 part_round_stats(cpu
, part
);
1672 part_dec_in_flight(part
);
1679 * blk_rq_bytes - Returns bytes left to complete in the entire request
1680 * @rq: the request being processed
1682 unsigned int blk_rq_bytes(struct request
*rq
)
1684 if (blk_fs_request(rq
))
1685 return blk_rq_sectors(rq
) << 9;
1687 return rq
->data_len
;
1689 EXPORT_SYMBOL_GPL(blk_rq_bytes
);
1692 * blk_rq_cur_bytes - Returns bytes left to complete in the current segment
1693 * @rq: the request being processed
1695 unsigned int blk_rq_cur_bytes(struct request
*rq
)
1697 if (blk_fs_request(rq
))
1698 return rq
->current_nr_sectors
<< 9;
1701 return rq
->bio
->bi_size
;
1703 return rq
->data_len
;
1705 EXPORT_SYMBOL_GPL(blk_rq_cur_bytes
);
1707 struct request
*elv_next_request(struct request_queue
*q
)
1712 while ((rq
= __elv_next_request(q
)) != NULL
) {
1713 if (!(rq
->cmd_flags
& REQ_STARTED
)) {
1715 * This is the first time the device driver
1716 * sees this request (possibly after
1717 * requeueing). Notify IO scheduler.
1719 if (blk_sorted_rq(rq
))
1720 elv_activate_rq(q
, rq
);
1723 * just mark as started even if we don't start
1724 * it, a request that has been delayed should
1725 * not be passed by new incoming requests
1727 rq
->cmd_flags
|= REQ_STARTED
;
1728 trace_block_rq_issue(q
, rq
);
1731 if (!q
->boundary_rq
|| q
->boundary_rq
== rq
) {
1732 q
->end_sector
= rq_end_sector(rq
);
1733 q
->boundary_rq
= NULL
;
1736 if (rq
->cmd_flags
& REQ_DONTPREP
)
1739 if (q
->dma_drain_size
&& rq
->data_len
) {
1741 * make sure space for the drain appears we
1742 * know we can do this because max_hw_segments
1743 * has been adjusted to be one fewer than the
1746 rq
->nr_phys_segments
++;
1752 ret
= q
->prep_rq_fn(q
, rq
);
1753 if (ret
== BLKPREP_OK
) {
1755 } else if (ret
== BLKPREP_DEFER
) {
1757 * the request may have been (partially) prepped.
1758 * we need to keep this request in the front to
1759 * avoid resource deadlock. REQ_STARTED will
1760 * prevent other fs requests from passing this one.
1762 if (q
->dma_drain_size
&& rq
->data_len
&&
1763 !(rq
->cmd_flags
& REQ_DONTPREP
)) {
1765 * remove the space for the drain we added
1766 * so that we don't add it again
1768 --rq
->nr_phys_segments
;
1773 } else if (ret
== BLKPREP_KILL
) {
1774 rq
->cmd_flags
|= REQ_QUIET
;
1775 __blk_end_request_all(rq
, -EIO
);
1777 printk(KERN_ERR
"%s: bad return=%d\n", __func__
, ret
);
1784 EXPORT_SYMBOL(elv_next_request
);
1786 void elv_dequeue_request(struct request_queue
*q
, struct request
*rq
)
1788 BUG_ON(list_empty(&rq
->queuelist
));
1789 BUG_ON(ELV_ON_HASH(rq
));
1791 list_del_init(&rq
->queuelist
);
1794 * the time frame between a request being removed from the lists
1795 * and to it is freed is accounted as io that is in progress at
1798 if (blk_account_rq(rq
))
1803 * blk_update_request - Special helper function for request stacking drivers
1804 * @rq: the request being processed
1805 * @error: %0 for success, < %0 for error
1806 * @nr_bytes: number of bytes to complete @rq
1809 * Ends I/O on a number of bytes attached to @rq, but doesn't complete
1810 * the request structure even if @rq doesn't have leftover.
1811 * If @rq has leftover, sets it up for the next range of segments.
1813 * This special helper function is only for request stacking drivers
1814 * (e.g. request-based dm) so that they can handle partial completion.
1815 * Actual device drivers should use blk_end_request instead.
1817 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
1818 * %false return from this function.
1821 * %false - this request doesn't have any more data
1822 * %true - this request has more data
1824 bool blk_update_request(struct request
*req
, int error
, unsigned int nr_bytes
)
1826 int total_bytes
, bio_nbytes
, next_idx
= 0;
1832 trace_block_rq_complete(req
->q
, req
);
1835 * For fs requests, rq is just carrier of independent bio's
1836 * and each partial completion should be handled separately.
1837 * Reset per-request error on each partial completion.
1839 * TODO: tj: This is too subtle. It would be better to let
1840 * low level drivers do what they see fit.
1842 if (blk_fs_request(req
))
1845 if (error
&& (blk_fs_request(req
) && !(req
->cmd_flags
& REQ_QUIET
))) {
1846 printk(KERN_ERR
"end_request: I/O error, dev %s, sector %llu\n",
1847 req
->rq_disk
? req
->rq_disk
->disk_name
: "?",
1848 (unsigned long long)blk_rq_pos(req
));
1851 blk_account_io_completion(req
, nr_bytes
);
1853 total_bytes
= bio_nbytes
= 0;
1854 while ((bio
= req
->bio
) != NULL
) {
1857 if (nr_bytes
>= bio
->bi_size
) {
1858 req
->bio
= bio
->bi_next
;
1859 nbytes
= bio
->bi_size
;
1860 req_bio_endio(req
, bio
, nbytes
, error
);
1864 int idx
= bio
->bi_idx
+ next_idx
;
1866 if (unlikely(bio
->bi_idx
>= bio
->bi_vcnt
)) {
1867 blk_dump_rq_flags(req
, "__end_that");
1868 printk(KERN_ERR
"%s: bio idx %d >= vcnt %d\n",
1869 __func__
, bio
->bi_idx
, bio
->bi_vcnt
);
1873 nbytes
= bio_iovec_idx(bio
, idx
)->bv_len
;
1874 BIO_BUG_ON(nbytes
> bio
->bi_size
);
1877 * not a complete bvec done
1879 if (unlikely(nbytes
> nr_bytes
)) {
1880 bio_nbytes
+= nr_bytes
;
1881 total_bytes
+= nr_bytes
;
1886 * advance to the next vector
1889 bio_nbytes
+= nbytes
;
1892 total_bytes
+= nbytes
;
1898 * end more in this run, or just return 'not-done'
1900 if (unlikely(nr_bytes
<= 0))
1910 * Reset counters so that the request stacking driver
1911 * can find how many bytes remain in the request
1914 req
->nr_sectors
= req
->hard_nr_sectors
= 0;
1915 req
->current_nr_sectors
= req
->hard_cur_sectors
= 0;
1920 * if the request wasn't completed, update state
1923 req_bio_endio(req
, bio
, bio_nbytes
, error
);
1924 bio
->bi_idx
+= next_idx
;
1925 bio_iovec(bio
)->bv_offset
+= nr_bytes
;
1926 bio_iovec(bio
)->bv_len
-= nr_bytes
;
1929 blk_recalc_rq_sectors(req
, total_bytes
>> 9);
1930 blk_recalc_rq_segments(req
);
1933 EXPORT_SYMBOL_GPL(blk_update_request
);
1935 static bool blk_update_bidi_request(struct request
*rq
, int error
,
1936 unsigned int nr_bytes
,
1937 unsigned int bidi_bytes
)
1939 if (blk_update_request(rq
, error
, nr_bytes
))
1942 /* Bidi request must be completed as a whole */
1943 if (unlikely(blk_bidi_rq(rq
)) &&
1944 blk_update_request(rq
->next_rq
, error
, bidi_bytes
))
1947 add_disk_randomness(rq
->rq_disk
);
1953 * queue lock must be held
1955 static void blk_finish_request(struct request
*req
, int error
)
1957 if (blk_rq_tagged(req
))
1958 blk_queue_end_tag(req
->q
, req
);
1960 if (blk_queued_rq(req
))
1961 elv_dequeue_request(req
->q
, req
);
1963 if (unlikely(laptop_mode
) && blk_fs_request(req
))
1964 laptop_io_completion();
1966 blk_delete_timer(req
);
1968 blk_account_io_done(req
);
1971 req
->end_io(req
, error
);
1973 if (blk_bidi_rq(req
))
1974 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
1976 __blk_put_request(req
->q
, req
);
1981 * blk_end_bidi_request - Complete a bidi request
1982 * @rq: the request to complete
1983 * @error: %0 for success, < %0 for error
1984 * @nr_bytes: number of bytes to complete @rq
1985 * @bidi_bytes: number of bytes to complete @rq->next_rq
1988 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
1989 * Drivers that supports bidi can safely call this member for any
1990 * type of request, bidi or uni. In the later case @bidi_bytes is
1994 * %false - we are done with this request
1995 * %true - still buffers pending for this request
1997 bool blk_end_bidi_request(struct request
*rq
, int error
,
1998 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2000 struct request_queue
*q
= rq
->q
;
2001 unsigned long flags
;
2003 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2006 spin_lock_irqsave(q
->queue_lock
, flags
);
2007 blk_finish_request(rq
, error
);
2008 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2012 EXPORT_SYMBOL_GPL(blk_end_bidi_request
);
2015 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2016 * @rq: the request to complete
2017 * @error: %0 for success, < %0 for error
2018 * @nr_bytes: number of bytes to complete @rq
2019 * @bidi_bytes: number of bytes to complete @rq->next_rq
2022 * Identical to blk_end_bidi_request() except that queue lock is
2023 * assumed to be locked on entry and remains so on return.
2026 * %false - we are done with this request
2027 * %true - still buffers pending for this request
2029 bool __blk_end_bidi_request(struct request
*rq
, int error
,
2030 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2032 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2035 blk_finish_request(rq
, error
);
2039 EXPORT_SYMBOL_GPL(__blk_end_bidi_request
);
2041 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
2044 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw, and
2045 we want BIO_RW_AHEAD (bit 1) to imply REQ_FAILFAST (bit 1). */
2046 rq
->cmd_flags
|= (bio
->bi_rw
& 3);
2048 if (bio_has_data(bio
)) {
2049 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2050 rq
->buffer
= bio_data(bio
);
2052 rq
->current_nr_sectors
= bio_cur_sectors(bio
);
2053 rq
->hard_cur_sectors
= rq
->current_nr_sectors
;
2054 rq
->hard_nr_sectors
= rq
->nr_sectors
= bio_sectors(bio
);
2055 rq
->data_len
= bio
->bi_size
;
2057 rq
->bio
= rq
->biotail
= bio
;
2060 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2064 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2065 * @q : the queue of the device being checked
2068 * Check if underlying low-level drivers of a device are busy.
2069 * If the drivers want to export their busy state, they must set own
2070 * exporting function using blk_queue_lld_busy() first.
2072 * Basically, this function is used only by request stacking drivers
2073 * to stop dispatching requests to underlying devices when underlying
2074 * devices are busy. This behavior helps more I/O merging on the queue
2075 * of the request stacking driver and prevents I/O throughput regression
2076 * on burst I/O load.
2079 * 0 - Not busy (The request stacking driver should dispatch request)
2080 * 1 - Busy (The request stacking driver should stop dispatching request)
2082 int blk_lld_busy(struct request_queue
*q
)
2085 return q
->lld_busy_fn(q
);
2089 EXPORT_SYMBOL_GPL(blk_lld_busy
);
2091 int kblockd_schedule_work(struct request_queue
*q
, struct work_struct
*work
)
2093 return queue_work(kblockd_workqueue
, work
);
2095 EXPORT_SYMBOL(kblockd_schedule_work
);
2097 int __init
blk_dev_init(void)
2099 BUILD_BUG_ON(__REQ_NR_BITS
> 8 *
2100 sizeof(((struct request
*)0)->cmd_flags
));
2102 kblockd_workqueue
= create_workqueue("kblockd");
2103 if (!kblockd_workqueue
)
2104 panic("Failed to create kblockd\n");
2106 request_cachep
= kmem_cache_create("blkdev_requests",
2107 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
2109 blk_requestq_cachep
= kmem_cache_create("blkdev_queue",
2110 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);