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>
34 static int __make_request(struct request_queue
*q
, struct bio
*bio
);
37 * For the allocated request tables
39 static struct kmem_cache
*request_cachep
;
42 * For queue allocation
44 struct kmem_cache
*blk_requestq_cachep
;
47 * Controlling structure to kblockd
49 static struct workqueue_struct
*kblockd_workqueue
;
51 static void drive_stat_acct(struct request
*rq
, int new_io
)
53 struct hd_struct
*part
;
54 int rw
= rq_data_dir(rq
);
57 if (!blk_fs_request(rq
) || !rq
->rq_disk
)
60 cpu
= part_stat_lock();
61 part
= disk_map_sector_rcu(rq
->rq_disk
, rq
->sector
);
64 part_stat_inc(cpu
, part
, merges
[rw
]);
66 part_round_stats(cpu
, part
);
67 part_inc_in_flight(part
);
73 void blk_queue_congestion_threshold(struct request_queue
*q
)
77 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) + 1;
78 if (nr
> q
->nr_requests
)
80 q
->nr_congestion_on
= nr
;
82 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) - (q
->nr_requests
/ 16) - 1;
85 q
->nr_congestion_off
= nr
;
89 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
92 * Locates the passed device's request queue and returns the address of its
95 * Will return NULL if the request queue cannot be located.
97 struct backing_dev_info
*blk_get_backing_dev_info(struct block_device
*bdev
)
99 struct backing_dev_info
*ret
= NULL
;
100 struct request_queue
*q
= bdev_get_queue(bdev
);
103 ret
= &q
->backing_dev_info
;
106 EXPORT_SYMBOL(blk_get_backing_dev_info
);
108 void blk_rq_init(struct request_queue
*q
, struct request
*rq
)
110 memset(rq
, 0, sizeof(*rq
));
112 INIT_LIST_HEAD(&rq
->queuelist
);
115 rq
->sector
= rq
->hard_sector
= (sector_t
) -1;
116 INIT_HLIST_NODE(&rq
->hash
);
117 RB_CLEAR_NODE(&rq
->rb_node
);
122 EXPORT_SYMBOL(blk_rq_init
);
124 static void req_bio_endio(struct request
*rq
, struct bio
*bio
,
125 unsigned int nbytes
, int error
)
127 struct request_queue
*q
= rq
->q
;
129 if (&q
->bar_rq
!= rq
) {
131 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
132 else if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
135 if (unlikely(nbytes
> bio
->bi_size
)) {
136 printk(KERN_ERR
"%s: want %u bytes done, %u left\n",
137 __func__
, nbytes
, bio
->bi_size
);
138 nbytes
= bio
->bi_size
;
141 bio
->bi_size
-= nbytes
;
142 bio
->bi_sector
+= (nbytes
>> 9);
144 if (bio_integrity(bio
))
145 bio_integrity_advance(bio
, nbytes
);
147 if (bio
->bi_size
== 0)
148 bio_endio(bio
, error
);
152 * Okay, this is the barrier request in progress, just
155 if (error
&& !q
->orderr
)
160 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
164 printk(KERN_INFO
"%s: dev %s: type=%x, flags=%x\n", msg
,
165 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->cmd_type
,
168 printk(KERN_INFO
" sector %llu, nr/cnr %lu/%u\n",
169 (unsigned long long)rq
->sector
,
171 rq
->current_nr_sectors
);
172 printk(KERN_INFO
" bio %p, biotail %p, buffer %p, data %p, len %u\n",
173 rq
->bio
, rq
->biotail
,
174 rq
->buffer
, rq
->data
,
177 if (blk_pc_request(rq
)) {
178 printk(KERN_INFO
" cdb: ");
179 for (bit
= 0; bit
< BLK_MAX_CDB
; bit
++)
180 printk("%02x ", rq
->cmd
[bit
]);
184 EXPORT_SYMBOL(blk_dump_rq_flags
);
187 * "plug" the device if there are no outstanding requests: this will
188 * force the transfer to start only after we have put all the requests
191 * This is called with interrupts off and no requests on the queue and
192 * with the queue lock held.
194 void blk_plug_device(struct request_queue
*q
)
196 WARN_ON(!irqs_disabled());
199 * don't plug a stopped queue, it must be paired with blk_start_queue()
200 * which will restart the queueing
202 if (blk_queue_stopped(q
))
205 if (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED
, q
)) {
206 mod_timer(&q
->unplug_timer
, jiffies
+ q
->unplug_delay
);
207 blk_add_trace_generic(q
, NULL
, 0, BLK_TA_PLUG
);
210 EXPORT_SYMBOL(blk_plug_device
);
213 * blk_plug_device_unlocked - plug a device without queue lock held
214 * @q: The &struct request_queue to plug
217 * Like @blk_plug_device(), but grabs the queue lock and disables
220 void blk_plug_device_unlocked(struct request_queue
*q
)
224 spin_lock_irqsave(q
->queue_lock
, flags
);
226 spin_unlock_irqrestore(q
->queue_lock
, flags
);
228 EXPORT_SYMBOL(blk_plug_device_unlocked
);
231 * remove the queue from the plugged list, if present. called with
232 * queue lock held and interrupts disabled.
234 int blk_remove_plug(struct request_queue
*q
)
236 WARN_ON(!irqs_disabled());
238 if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED
, q
))
241 del_timer(&q
->unplug_timer
);
244 EXPORT_SYMBOL(blk_remove_plug
);
247 * remove the plug and let it rip..
249 void __generic_unplug_device(struct request_queue
*q
)
251 if (unlikely(blk_queue_stopped(q
)))
254 if (!blk_remove_plug(q
))
259 EXPORT_SYMBOL(__generic_unplug_device
);
262 * generic_unplug_device - fire a request queue
263 * @q: The &struct request_queue in question
266 * Linux uses plugging to build bigger requests queues before letting
267 * the device have at them. If a queue is plugged, the I/O scheduler
268 * is still adding and merging requests on the queue. Once the queue
269 * gets unplugged, the request_fn defined for the queue is invoked and
272 void generic_unplug_device(struct request_queue
*q
)
274 if (blk_queue_plugged(q
)) {
275 spin_lock_irq(q
->queue_lock
);
276 __generic_unplug_device(q
);
277 spin_unlock_irq(q
->queue_lock
);
280 EXPORT_SYMBOL(generic_unplug_device
);
282 static void blk_backing_dev_unplug(struct backing_dev_info
*bdi
,
285 struct request_queue
*q
= bdi
->unplug_io_data
;
290 void blk_unplug_work(struct work_struct
*work
)
292 struct request_queue
*q
=
293 container_of(work
, struct request_queue
, unplug_work
);
295 blk_add_trace_pdu_int(q
, BLK_TA_UNPLUG_IO
, NULL
,
296 q
->rq
.count
[READ
] + q
->rq
.count
[WRITE
]);
301 void blk_unplug_timeout(unsigned long data
)
303 struct request_queue
*q
= (struct request_queue
*)data
;
305 blk_add_trace_pdu_int(q
, BLK_TA_UNPLUG_TIMER
, NULL
,
306 q
->rq
.count
[READ
] + q
->rq
.count
[WRITE
]);
308 kblockd_schedule_work(q
, &q
->unplug_work
);
311 void blk_unplug(struct request_queue
*q
)
314 * devices don't necessarily have an ->unplug_fn defined
317 blk_add_trace_pdu_int(q
, BLK_TA_UNPLUG_IO
, NULL
,
318 q
->rq
.count
[READ
] + q
->rq
.count
[WRITE
]);
323 EXPORT_SYMBOL(blk_unplug
);
325 static void blk_invoke_request_fn(struct request_queue
*q
)
328 * one level of recursion is ok and is much faster than kicking
329 * the unplug handling
331 if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER
, q
)) {
333 queue_flag_clear(QUEUE_FLAG_REENTER
, q
);
335 queue_flag_set(QUEUE_FLAG_PLUGGED
, q
);
336 kblockd_schedule_work(q
, &q
->unplug_work
);
341 * blk_start_queue - restart a previously stopped queue
342 * @q: The &struct request_queue in question
345 * blk_start_queue() will clear the stop flag on the queue, and call
346 * the request_fn for the queue if it was in a stopped state when
347 * entered. Also see blk_stop_queue(). Queue lock must be held.
349 void blk_start_queue(struct request_queue
*q
)
351 WARN_ON(!irqs_disabled());
353 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
354 blk_invoke_request_fn(q
);
356 EXPORT_SYMBOL(blk_start_queue
);
359 * blk_stop_queue - stop a queue
360 * @q: The &struct request_queue in question
363 * The Linux block layer assumes that a block driver will consume all
364 * entries on the request queue when the request_fn strategy is called.
365 * Often this will not happen, because of hardware limitations (queue
366 * depth settings). If a device driver gets a 'queue full' response,
367 * or if it simply chooses not to queue more I/O at one point, it can
368 * call this function to prevent the request_fn from being called until
369 * the driver has signalled it's ready to go again. This happens by calling
370 * blk_start_queue() to restart queue operations. Queue lock must be held.
372 void blk_stop_queue(struct request_queue
*q
)
375 queue_flag_set(QUEUE_FLAG_STOPPED
, q
);
377 EXPORT_SYMBOL(blk_stop_queue
);
380 * blk_sync_queue - cancel any pending callbacks on a queue
384 * The block layer may perform asynchronous callback activity
385 * on a queue, such as calling the unplug function after a timeout.
386 * A block device may call blk_sync_queue to ensure that any
387 * such activity is cancelled, thus allowing it to release resources
388 * that the callbacks might use. The caller must already have made sure
389 * that its ->make_request_fn will not re-add plugging prior to calling
393 void blk_sync_queue(struct request_queue
*q
)
395 del_timer_sync(&q
->unplug_timer
);
396 kblockd_flush_work(&q
->unplug_work
);
398 EXPORT_SYMBOL(blk_sync_queue
);
401 * blk_run_queue - run a single device queue
402 * @q: The queue to run
404 void __blk_run_queue(struct request_queue
*q
)
409 * Only recurse once to avoid overrunning the stack, let the unplug
410 * handling reinvoke the handler shortly if we already got there.
412 if (!elv_queue_empty(q
))
413 blk_invoke_request_fn(q
);
415 EXPORT_SYMBOL(__blk_run_queue
);
418 * blk_run_queue - run a single device queue
419 * @q: The queue to run
421 void blk_run_queue(struct request_queue
*q
)
425 spin_lock_irqsave(q
->queue_lock
, flags
);
427 spin_unlock_irqrestore(q
->queue_lock
, flags
);
429 EXPORT_SYMBOL(blk_run_queue
);
431 void blk_put_queue(struct request_queue
*q
)
433 kobject_put(&q
->kobj
);
436 void blk_cleanup_queue(struct request_queue
*q
)
438 mutex_lock(&q
->sysfs_lock
);
439 queue_flag_set_unlocked(QUEUE_FLAG_DEAD
, q
);
440 mutex_unlock(&q
->sysfs_lock
);
443 elevator_exit(q
->elevator
);
447 EXPORT_SYMBOL(blk_cleanup_queue
);
449 static int blk_init_free_list(struct request_queue
*q
)
451 struct request_list
*rl
= &q
->rq
;
453 rl
->count
[READ
] = rl
->count
[WRITE
] = 0;
454 rl
->starved
[READ
] = rl
->starved
[WRITE
] = 0;
456 init_waitqueue_head(&rl
->wait
[READ
]);
457 init_waitqueue_head(&rl
->wait
[WRITE
]);
459 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, mempool_alloc_slab
,
460 mempool_free_slab
, request_cachep
, q
->node
);
468 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
470 return blk_alloc_queue_node(gfp_mask
, -1);
472 EXPORT_SYMBOL(blk_alloc_queue
);
474 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
476 struct request_queue
*q
;
479 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
480 gfp_mask
| __GFP_ZERO
, node_id
);
484 q
->backing_dev_info
.unplug_io_fn
= blk_backing_dev_unplug
;
485 q
->backing_dev_info
.unplug_io_data
= q
;
486 err
= bdi_init(&q
->backing_dev_info
);
488 kmem_cache_free(blk_requestq_cachep
, q
);
492 init_timer(&q
->unplug_timer
);
494 kobject_init(&q
->kobj
, &blk_queue_ktype
);
496 mutex_init(&q
->sysfs_lock
);
497 spin_lock_init(&q
->__queue_lock
);
501 EXPORT_SYMBOL(blk_alloc_queue_node
);
504 * blk_init_queue - prepare a request queue for use with a block device
505 * @rfn: The function to be called to process requests that have been
506 * placed on the queue.
507 * @lock: Request queue spin lock
510 * If a block device wishes to use the standard request handling procedures,
511 * which sorts requests and coalesces adjacent requests, then it must
512 * call blk_init_queue(). The function @rfn will be called when there
513 * are requests on the queue that need to be processed. If the device
514 * supports plugging, then @rfn may not be called immediately when requests
515 * are available on the queue, but may be called at some time later instead.
516 * Plugged queues are generally unplugged when a buffer belonging to one
517 * of the requests on the queue is needed, or due to memory pressure.
519 * @rfn is not required, or even expected, to remove all requests off the
520 * queue, but only as many as it can handle at a time. If it does leave
521 * requests on the queue, it is responsible for arranging that the requests
522 * get dealt with eventually.
524 * The queue spin lock must be held while manipulating the requests on the
525 * request queue; this lock will be taken also from interrupt context, so irq
526 * disabling is needed for it.
528 * Function returns a pointer to the initialized request queue, or %NULL if
532 * blk_init_queue() must be paired with a blk_cleanup_queue() call
533 * when the block device is deactivated (such as at module unload).
536 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
538 return blk_init_queue_node(rfn
, lock
, -1);
540 EXPORT_SYMBOL(blk_init_queue
);
542 struct request_queue
*
543 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
545 struct request_queue
*q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
551 if (blk_init_free_list(q
)) {
552 kmem_cache_free(blk_requestq_cachep
, q
);
557 * if caller didn't supply a lock, they get per-queue locking with
561 lock
= &q
->__queue_lock
;
564 q
->prep_rq_fn
= NULL
;
565 q
->unplug_fn
= generic_unplug_device
;
566 q
->queue_flags
= (1 << QUEUE_FLAG_CLUSTER
);
567 q
->queue_lock
= lock
;
569 blk_queue_segment_boundary(q
, 0xffffffff);
571 blk_queue_make_request(q
, __make_request
);
572 blk_queue_max_segment_size(q
, MAX_SEGMENT_SIZE
);
574 blk_queue_max_hw_segments(q
, MAX_HW_SEGMENTS
);
575 blk_queue_max_phys_segments(q
, MAX_PHYS_SEGMENTS
);
577 q
->sg_reserved_size
= INT_MAX
;
579 blk_set_cmd_filter_defaults(&q
->cmd_filter
);
584 if (!elevator_init(q
, NULL
)) {
585 blk_queue_congestion_threshold(q
);
592 EXPORT_SYMBOL(blk_init_queue_node
);
594 int blk_get_queue(struct request_queue
*q
)
596 if (likely(!test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
))) {
597 kobject_get(&q
->kobj
);
604 static inline void blk_free_request(struct request_queue
*q
, struct request
*rq
)
606 if (rq
->cmd_flags
& REQ_ELVPRIV
)
607 elv_put_request(q
, rq
);
608 mempool_free(rq
, q
->rq
.rq_pool
);
611 static struct request
*
612 blk_alloc_request(struct request_queue
*q
, int rw
, int priv
, gfp_t gfp_mask
)
614 struct request
*rq
= mempool_alloc(q
->rq
.rq_pool
, gfp_mask
);
621 rq
->cmd_flags
= rw
| REQ_ALLOCED
;
624 if (unlikely(elv_set_request(q
, rq
, gfp_mask
))) {
625 mempool_free(rq
, q
->rq
.rq_pool
);
628 rq
->cmd_flags
|= REQ_ELVPRIV
;
635 * ioc_batching returns true if the ioc is a valid batching request and
636 * should be given priority access to a request.
638 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
644 * Make sure the process is able to allocate at least 1 request
645 * even if the batch times out, otherwise we could theoretically
648 return ioc
->nr_batch_requests
== q
->nr_batching
||
649 (ioc
->nr_batch_requests
> 0
650 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
654 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
655 * will cause the process to be a "batcher" on all queues in the system. This
656 * is the behaviour we want though - once it gets a wakeup it should be given
659 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
661 if (!ioc
|| ioc_batching(q
, ioc
))
664 ioc
->nr_batch_requests
= q
->nr_batching
;
665 ioc
->last_waited
= jiffies
;
668 static void __freed_request(struct request_queue
*q
, int rw
)
670 struct request_list
*rl
= &q
->rq
;
672 if (rl
->count
[rw
] < queue_congestion_off_threshold(q
))
673 blk_clear_queue_congested(q
, rw
);
675 if (rl
->count
[rw
] + 1 <= q
->nr_requests
) {
676 if (waitqueue_active(&rl
->wait
[rw
]))
677 wake_up(&rl
->wait
[rw
]);
679 blk_clear_queue_full(q
, rw
);
684 * A request has just been released. Account for it, update the full and
685 * congestion status, wake up any waiters. Called under q->queue_lock.
687 static void freed_request(struct request_queue
*q
, int rw
, int priv
)
689 struct request_list
*rl
= &q
->rq
;
695 __freed_request(q
, rw
);
697 if (unlikely(rl
->starved
[rw
^ 1]))
698 __freed_request(q
, rw
^ 1);
701 #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
703 * Get a free request, queue_lock must be held.
704 * Returns NULL on failure, with queue_lock held.
705 * Returns !NULL on success, with queue_lock *not held*.
707 static struct request
*get_request(struct request_queue
*q
, int rw_flags
,
708 struct bio
*bio
, gfp_t gfp_mask
)
710 struct request
*rq
= NULL
;
711 struct request_list
*rl
= &q
->rq
;
712 struct io_context
*ioc
= NULL
;
713 const int rw
= rw_flags
& 0x01;
716 may_queue
= elv_may_queue(q
, rw_flags
);
717 if (may_queue
== ELV_MQUEUE_NO
)
720 if (rl
->count
[rw
]+1 >= queue_congestion_on_threshold(q
)) {
721 if (rl
->count
[rw
]+1 >= q
->nr_requests
) {
722 ioc
= current_io_context(GFP_ATOMIC
, q
->node
);
724 * The queue will fill after this allocation, so set
725 * it as full, and mark this process as "batching".
726 * This process will be allowed to complete a batch of
727 * requests, others will be blocked.
729 if (!blk_queue_full(q
, rw
)) {
730 ioc_set_batching(q
, ioc
);
731 blk_set_queue_full(q
, rw
);
733 if (may_queue
!= ELV_MQUEUE_MUST
734 && !ioc_batching(q
, ioc
)) {
736 * The queue is full and the allocating
737 * process is not a "batcher", and not
738 * exempted by the IO scheduler
744 blk_set_queue_congested(q
, rw
);
748 * Only allow batching queuers to allocate up to 50% over the defined
749 * limit of requests, otherwise we could have thousands of requests
750 * allocated with any setting of ->nr_requests
752 if (rl
->count
[rw
] >= (3 * q
->nr_requests
/ 2))
758 priv
= !test_bit(QUEUE_FLAG_ELVSWITCH
, &q
->queue_flags
);
762 spin_unlock_irq(q
->queue_lock
);
764 rq
= blk_alloc_request(q
, rw_flags
, priv
, gfp_mask
);
767 * Allocation failed presumably due to memory. Undo anything
768 * we might have messed up.
770 * Allocating task should really be put onto the front of the
771 * wait queue, but this is pretty rare.
773 spin_lock_irq(q
->queue_lock
);
774 freed_request(q
, rw
, priv
);
777 * in the very unlikely event that allocation failed and no
778 * requests for this direction was pending, mark us starved
779 * so that freeing of a request in the other direction will
780 * notice us. another possible fix would be to split the
781 * rq mempool into READ and WRITE
784 if (unlikely(rl
->count
[rw
] == 0))
791 * ioc may be NULL here, and ioc_batching will be false. That's
792 * OK, if the queue is under the request limit then requests need
793 * not count toward the nr_batch_requests limit. There will always
794 * be some limit enforced by BLK_BATCH_TIME.
796 if (ioc_batching(q
, ioc
))
797 ioc
->nr_batch_requests
--;
799 blk_add_trace_generic(q
, bio
, rw
, BLK_TA_GETRQ
);
805 * No available requests for this queue, unplug the device and wait for some
806 * requests to become available.
808 * Called with q->queue_lock held, and returns with it unlocked.
810 static struct request
*get_request_wait(struct request_queue
*q
, int rw_flags
,
813 const int rw
= rw_flags
& 0x01;
816 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
819 struct io_context
*ioc
;
820 struct request_list
*rl
= &q
->rq
;
822 prepare_to_wait_exclusive(&rl
->wait
[rw
], &wait
,
823 TASK_UNINTERRUPTIBLE
);
825 blk_add_trace_generic(q
, bio
, rw
, BLK_TA_SLEEPRQ
);
827 __generic_unplug_device(q
);
828 spin_unlock_irq(q
->queue_lock
);
832 * After sleeping, we become a "batching" process and
833 * will be able to allocate at least one request, and
834 * up to a big batch of them for a small period time.
835 * See ioc_batching, ioc_set_batching
837 ioc
= current_io_context(GFP_NOIO
, q
->node
);
838 ioc_set_batching(q
, ioc
);
840 spin_lock_irq(q
->queue_lock
);
841 finish_wait(&rl
->wait
[rw
], &wait
);
843 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
849 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
853 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
855 spin_lock_irq(q
->queue_lock
);
856 if (gfp_mask
& __GFP_WAIT
) {
857 rq
= get_request_wait(q
, rw
, NULL
);
859 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
861 spin_unlock_irq(q
->queue_lock
);
863 /* q->queue_lock is unlocked at this point */
867 EXPORT_SYMBOL(blk_get_request
);
870 * blk_start_queueing - initiate dispatch of requests to device
871 * @q: request queue to kick into gear
873 * This is basically a helper to remove the need to know whether a queue
874 * is plugged or not if someone just wants to initiate dispatch of requests
877 * The queue lock must be held with interrupts disabled.
879 void blk_start_queueing(struct request_queue
*q
)
881 if (!blk_queue_plugged(q
))
884 __generic_unplug_device(q
);
886 EXPORT_SYMBOL(blk_start_queueing
);
889 * blk_requeue_request - put a request back on queue
890 * @q: request queue where request should be inserted
891 * @rq: request to be inserted
894 * Drivers often keep queueing requests until the hardware cannot accept
895 * more, when that condition happens we need to put the request back
896 * on the queue. Must be called with queue lock held.
898 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
900 blk_add_trace_rq(q
, rq
, BLK_TA_REQUEUE
);
902 if (blk_rq_tagged(rq
))
903 blk_queue_end_tag(q
, rq
);
905 elv_requeue_request(q
, rq
);
907 EXPORT_SYMBOL(blk_requeue_request
);
910 * blk_insert_request - insert a special request into a request queue
911 * @q: request queue where request should be inserted
912 * @rq: request to be inserted
913 * @at_head: insert request at head or tail of queue
914 * @data: private data
917 * Many block devices need to execute commands asynchronously, so they don't
918 * block the whole kernel from preemption during request execution. This is
919 * accomplished normally by inserting aritficial requests tagged as
920 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
921 * be scheduled for actual execution by the request queue.
923 * We have the option of inserting the head or the tail of the queue.
924 * Typically we use the tail for new ioctls and so forth. We use the head
925 * of the queue for things like a QUEUE_FULL message from a device, or a
926 * host that is unable to accept a particular command.
928 void blk_insert_request(struct request_queue
*q
, struct request
*rq
,
929 int at_head
, void *data
)
931 int where
= at_head
? ELEVATOR_INSERT_FRONT
: ELEVATOR_INSERT_BACK
;
935 * tell I/O scheduler that this isn't a regular read/write (ie it
936 * must not attempt merges on this) and that it acts as a soft
939 rq
->cmd_type
= REQ_TYPE_SPECIAL
;
940 rq
->cmd_flags
|= REQ_SOFTBARRIER
;
944 spin_lock_irqsave(q
->queue_lock
, flags
);
947 * If command is tagged, release the tag
949 if (blk_rq_tagged(rq
))
950 blk_queue_end_tag(q
, rq
);
952 drive_stat_acct(rq
, 1);
953 __elv_add_request(q
, rq
, where
, 0);
954 blk_start_queueing(q
);
955 spin_unlock_irqrestore(q
->queue_lock
, flags
);
957 EXPORT_SYMBOL(blk_insert_request
);
960 * add-request adds a request to the linked list.
961 * queue lock is held and interrupts disabled, as we muck with the
962 * request queue list.
964 static inline void add_request(struct request_queue
*q
, struct request
*req
)
966 drive_stat_acct(req
, 1);
969 * elevator indicated where it wants this request to be
970 * inserted at elevator_merge time
972 __elv_add_request(q
, req
, ELEVATOR_INSERT_SORT
, 0);
975 static void part_round_stats_single(int cpu
, struct hd_struct
*part
,
978 if (now
== part
->stamp
)
981 if (part
->in_flight
) {
982 __part_stat_add(cpu
, part
, time_in_queue
,
983 part
->in_flight
* (now
- part
->stamp
));
984 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
990 * part_round_stats() - Round off the performance stats on a struct
993 * The average IO queue length and utilisation statistics are maintained
994 * by observing the current state of the queue length and the amount of
995 * time it has been in this state for.
997 * Normally, that accounting is done on IO completion, but that can result
998 * in more than a second's worth of IO being accounted for within any one
999 * second, leading to >100% utilisation. To deal with that, we call this
1000 * function to do a round-off before returning the results when reading
1001 * /proc/diskstats. This accounts immediately for all queue usage up to
1002 * the current jiffies and restarts the counters again.
1004 void part_round_stats(int cpu
, struct hd_struct
*part
)
1006 unsigned long now
= jiffies
;
1009 part_round_stats_single(cpu
, &part_to_disk(part
)->part0
, now
);
1010 part_round_stats_single(cpu
, part
, now
);
1012 EXPORT_SYMBOL_GPL(part_round_stats
);
1015 * queue lock must be held
1017 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1021 if (unlikely(--req
->ref_count
))
1024 elv_completed_request(q
, req
);
1027 * Request may not have originated from ll_rw_blk. if not,
1028 * it didn't come out of our reserved rq pools
1030 if (req
->cmd_flags
& REQ_ALLOCED
) {
1031 int rw
= rq_data_dir(req
);
1032 int priv
= req
->cmd_flags
& REQ_ELVPRIV
;
1034 BUG_ON(!list_empty(&req
->queuelist
));
1035 BUG_ON(!hlist_unhashed(&req
->hash
));
1037 blk_free_request(q
, req
);
1038 freed_request(q
, rw
, priv
);
1041 EXPORT_SYMBOL_GPL(__blk_put_request
);
1043 void blk_put_request(struct request
*req
)
1045 unsigned long flags
;
1046 struct request_queue
*q
= req
->q
;
1048 spin_lock_irqsave(q
->queue_lock
, flags
);
1049 __blk_put_request(q
, req
);
1050 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1052 EXPORT_SYMBOL(blk_put_request
);
1054 void init_request_from_bio(struct request
*req
, struct bio
*bio
)
1056 req
->cpu
= bio
->bi_comp_cpu
;
1057 req
->cmd_type
= REQ_TYPE_FS
;
1060 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
1062 if (bio_rw_ahead(bio
) || bio_failfast(bio
))
1063 req
->cmd_flags
|= REQ_FAILFAST
;
1066 * REQ_BARRIER implies no merging, but lets make it explicit
1068 if (unlikely(bio_discard(bio
))) {
1069 req
->cmd_flags
|= REQ_DISCARD
;
1070 if (bio_barrier(bio
))
1071 req
->cmd_flags
|= REQ_SOFTBARRIER
;
1072 req
->q
->prepare_discard_fn(req
->q
, req
);
1073 } else if (unlikely(bio_barrier(bio
)))
1074 req
->cmd_flags
|= (REQ_HARDBARRIER
| REQ_NOMERGE
);
1077 req
->cmd_flags
|= REQ_RW_SYNC
;
1078 if (bio_rw_meta(bio
))
1079 req
->cmd_flags
|= REQ_RW_META
;
1082 req
->hard_sector
= req
->sector
= bio
->bi_sector
;
1083 req
->ioprio
= bio_prio(bio
);
1084 req
->start_time
= jiffies
;
1085 blk_rq_bio_prep(req
->q
, req
, bio
);
1088 static int __make_request(struct request_queue
*q
, struct bio
*bio
)
1090 struct request
*req
;
1091 int el_ret
, nr_sectors
, barrier
, discard
, err
;
1092 const unsigned short prio
= bio_prio(bio
);
1093 const int sync
= bio_sync(bio
);
1096 nr_sectors
= bio_sectors(bio
);
1099 * low level driver can indicate that it wants pages above a
1100 * certain limit bounced to low memory (ie for highmem, or even
1101 * ISA dma in theory)
1103 blk_queue_bounce(q
, &bio
);
1105 barrier
= bio_barrier(bio
);
1106 if (unlikely(barrier
) && bio_has_data(bio
) &&
1107 (q
->next_ordered
== QUEUE_ORDERED_NONE
)) {
1112 discard
= bio_discard(bio
);
1113 if (unlikely(discard
) && !q
->prepare_discard_fn
) {
1118 spin_lock_irq(q
->queue_lock
);
1120 if (unlikely(barrier
) || elv_queue_empty(q
))
1123 el_ret
= elv_merge(q
, &req
, bio
);
1125 case ELEVATOR_BACK_MERGE
:
1126 BUG_ON(!rq_mergeable(req
));
1128 if (!ll_back_merge_fn(q
, req
, bio
))
1131 blk_add_trace_bio(q
, bio
, BLK_TA_BACKMERGE
);
1133 req
->biotail
->bi_next
= bio
;
1135 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
1136 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
1137 drive_stat_acct(req
, 0);
1138 if (!attempt_back_merge(q
, req
))
1139 elv_merged_request(q
, req
, el_ret
);
1142 case ELEVATOR_FRONT_MERGE
:
1143 BUG_ON(!rq_mergeable(req
));
1145 if (!ll_front_merge_fn(q
, req
, bio
))
1148 blk_add_trace_bio(q
, bio
, BLK_TA_FRONTMERGE
);
1150 bio
->bi_next
= req
->bio
;
1154 * may not be valid. if the low level driver said
1155 * it didn't need a bounce buffer then it better
1156 * not touch req->buffer either...
1158 req
->buffer
= bio_data(bio
);
1159 req
->current_nr_sectors
= bio_cur_sectors(bio
);
1160 req
->hard_cur_sectors
= req
->current_nr_sectors
;
1161 req
->sector
= req
->hard_sector
= bio
->bi_sector
;
1162 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
1163 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
1164 drive_stat_acct(req
, 0);
1165 if (!attempt_front_merge(q
, req
))
1166 elv_merged_request(q
, req
, el_ret
);
1169 /* ELV_NO_MERGE: elevator says don't/can't merge. */
1176 * This sync check and mask will be re-done in init_request_from_bio(),
1177 * but we need to set it earlier to expose the sync flag to the
1178 * rq allocator and io schedulers.
1180 rw_flags
= bio_data_dir(bio
);
1182 rw_flags
|= REQ_RW_SYNC
;
1185 * Grab a free request. This is might sleep but can not fail.
1186 * Returns with the queue unlocked.
1188 req
= get_request_wait(q
, rw_flags
, bio
);
1191 * After dropping the lock and possibly sleeping here, our request
1192 * may now be mergeable after it had proven unmergeable (above).
1193 * We don't worry about that case for efficiency. It won't happen
1194 * often, and the elevators are able to handle it.
1196 init_request_from_bio(req
, bio
);
1198 spin_lock_irq(q
->queue_lock
);
1199 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
) ||
1200 bio_flagged(bio
, BIO_CPU_AFFINE
))
1201 req
->cpu
= blk_cpu_to_group(smp_processor_id());
1202 if (elv_queue_empty(q
))
1204 add_request(q
, req
);
1207 __generic_unplug_device(q
);
1208 spin_unlock_irq(q
->queue_lock
);
1212 bio_endio(bio
, err
);
1217 * If bio->bi_dev is a partition, remap the location
1219 static inline void blk_partition_remap(struct bio
*bio
)
1221 struct block_device
*bdev
= bio
->bi_bdev
;
1223 if (bio_sectors(bio
) && bdev
!= bdev
->bd_contains
) {
1224 struct hd_struct
*p
= bdev
->bd_part
;
1226 bio
->bi_sector
+= p
->start_sect
;
1227 bio
->bi_bdev
= bdev
->bd_contains
;
1229 blk_add_trace_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
1230 bdev
->bd_dev
, bio
->bi_sector
,
1231 bio
->bi_sector
- p
->start_sect
);
1235 static void handle_bad_sector(struct bio
*bio
)
1237 char b
[BDEVNAME_SIZE
];
1239 printk(KERN_INFO
"attempt to access beyond end of device\n");
1240 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
1241 bdevname(bio
->bi_bdev
, b
),
1243 (unsigned long long)bio
->bi_sector
+ bio_sectors(bio
),
1244 (long long)(bio
->bi_bdev
->bd_inode
->i_size
>> 9));
1246 set_bit(BIO_EOF
, &bio
->bi_flags
);
1249 #ifdef CONFIG_FAIL_MAKE_REQUEST
1251 static DECLARE_FAULT_ATTR(fail_make_request
);
1253 static int __init
setup_fail_make_request(char *str
)
1255 return setup_fault_attr(&fail_make_request
, str
);
1257 __setup("fail_make_request=", setup_fail_make_request
);
1259 static int should_fail_request(struct bio
*bio
)
1261 struct hd_struct
*part
= bio
->bi_bdev
->bd_part
;
1263 if (part_to_disk(part
)->part0
.make_it_fail
|| part
->make_it_fail
)
1264 return should_fail(&fail_make_request
, bio
->bi_size
);
1269 static int __init
fail_make_request_debugfs(void)
1271 return init_fault_attr_dentries(&fail_make_request
,
1272 "fail_make_request");
1275 late_initcall(fail_make_request_debugfs
);
1277 #else /* CONFIG_FAIL_MAKE_REQUEST */
1279 static inline int should_fail_request(struct bio
*bio
)
1284 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1287 * Check whether this bio extends beyond the end of the device.
1289 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
1296 /* Test device or partition size, when known. */
1297 maxsector
= bio
->bi_bdev
->bd_inode
->i_size
>> 9;
1299 sector_t sector
= bio
->bi_sector
;
1301 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
1303 * This may well happen - the kernel calls bread()
1304 * without checking the size of the device, e.g., when
1305 * mounting a device.
1307 handle_bad_sector(bio
);
1316 * generic_make_request - hand a buffer to its device driver for I/O
1317 * @bio: The bio describing the location in memory and on the device.
1319 * generic_make_request() is used to make I/O requests of block
1320 * devices. It is passed a &struct bio, which describes the I/O that needs
1323 * generic_make_request() does not return any status. The
1324 * success/failure status of the request, along with notification of
1325 * completion, is delivered asynchronously through the bio->bi_end_io
1326 * function described (one day) else where.
1328 * The caller of generic_make_request must make sure that bi_io_vec
1329 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1330 * set to describe the device address, and the
1331 * bi_end_io and optionally bi_private are set to describe how
1332 * completion notification should be signaled.
1334 * generic_make_request and the drivers it calls may use bi_next if this
1335 * bio happens to be merged with someone else, and may change bi_dev and
1336 * bi_sector for remaps as it sees fit. So the values of these fields
1337 * should NOT be depended on after the call to generic_make_request.
1339 static inline void __generic_make_request(struct bio
*bio
)
1341 struct request_queue
*q
;
1342 sector_t old_sector
;
1343 int ret
, nr_sectors
= bio_sectors(bio
);
1349 if (bio_check_eod(bio
, nr_sectors
))
1353 * Resolve the mapping until finished. (drivers are
1354 * still free to implement/resolve their own stacking
1355 * by explicitly returning 0)
1357 * NOTE: we don't repeat the blk_size check for each new device.
1358 * Stacking drivers are expected to know what they are doing.
1363 char b
[BDEVNAME_SIZE
];
1365 q
= bdev_get_queue(bio
->bi_bdev
);
1368 "generic_make_request: Trying to access "
1369 "nonexistent block-device %s (%Lu)\n",
1370 bdevname(bio
->bi_bdev
, b
),
1371 (long long) bio
->bi_sector
);
1373 bio_endio(bio
, err
);
1377 if (unlikely(nr_sectors
> q
->max_hw_sectors
)) {
1378 printk(KERN_ERR
"bio too big device %s (%u > %u)\n",
1379 bdevname(bio
->bi_bdev
, b
),
1385 if (unlikely(test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)))
1388 if (should_fail_request(bio
))
1392 * If this device has partitions, remap block n
1393 * of partition p to block n+start(p) of the disk.
1395 blk_partition_remap(bio
);
1397 if (bio_integrity_enabled(bio
) && bio_integrity_prep(bio
))
1400 if (old_sector
!= -1)
1401 blk_add_trace_remap(q
, bio
, old_dev
, bio
->bi_sector
,
1404 blk_add_trace_bio(q
, bio
, BLK_TA_QUEUE
);
1406 old_sector
= bio
->bi_sector
;
1407 old_dev
= bio
->bi_bdev
->bd_dev
;
1409 if (bio_check_eod(bio
, nr_sectors
))
1411 if ((bio_empty_barrier(bio
) && !q
->prepare_flush_fn
) ||
1412 (bio_discard(bio
) && !q
->prepare_discard_fn
)) {
1417 ret
= q
->make_request_fn(q
, bio
);
1422 * We only want one ->make_request_fn to be active at a time,
1423 * else stack usage with stacked devices could be a problem.
1424 * So use current->bio_{list,tail} to keep a list of requests
1425 * submited by a make_request_fn function.
1426 * current->bio_tail is also used as a flag to say if
1427 * generic_make_request is currently active in this task or not.
1428 * If it is NULL, then no make_request is active. If it is non-NULL,
1429 * then a make_request is active, and new requests should be added
1432 void generic_make_request(struct bio
*bio
)
1434 if (current
->bio_tail
) {
1435 /* make_request is active */
1436 *(current
->bio_tail
) = bio
;
1437 bio
->bi_next
= NULL
;
1438 current
->bio_tail
= &bio
->bi_next
;
1441 /* following loop may be a bit non-obvious, and so deserves some
1443 * Before entering the loop, bio->bi_next is NULL (as all callers
1444 * ensure that) so we have a list with a single bio.
1445 * We pretend that we have just taken it off a longer list, so
1446 * we assign bio_list to the next (which is NULL) and bio_tail
1447 * to &bio_list, thus initialising the bio_list of new bios to be
1448 * added. __generic_make_request may indeed add some more bios
1449 * through a recursive call to generic_make_request. If it
1450 * did, we find a non-NULL value in bio_list and re-enter the loop
1451 * from the top. In this case we really did just take the bio
1452 * of the top of the list (no pretending) and so fixup bio_list and
1453 * bio_tail or bi_next, and call into __generic_make_request again.
1455 * The loop was structured like this to make only one call to
1456 * __generic_make_request (which is important as it is large and
1457 * inlined) and to keep the structure simple.
1459 BUG_ON(bio
->bi_next
);
1461 current
->bio_list
= bio
->bi_next
;
1462 if (bio
->bi_next
== NULL
)
1463 current
->bio_tail
= ¤t
->bio_list
;
1465 bio
->bi_next
= NULL
;
1466 __generic_make_request(bio
);
1467 bio
= current
->bio_list
;
1469 current
->bio_tail
= NULL
; /* deactivate */
1471 EXPORT_SYMBOL(generic_make_request
);
1474 * submit_bio - submit a bio to the block device layer for I/O
1475 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1476 * @bio: The &struct bio which describes the I/O
1478 * submit_bio() is very similar in purpose to generic_make_request(), and
1479 * uses that function to do most of the work. Both are fairly rough
1480 * interfaces; @bio must be presetup and ready for I/O.
1483 void submit_bio(int rw
, struct bio
*bio
)
1485 int count
= bio_sectors(bio
);
1490 * If it's a regular read/write or a barrier with data attached,
1491 * go through the normal accounting stuff before submission.
1493 if (bio_has_data(bio
)) {
1495 count_vm_events(PGPGOUT
, count
);
1497 task_io_account_read(bio
->bi_size
);
1498 count_vm_events(PGPGIN
, count
);
1501 if (unlikely(block_dump
)) {
1502 char b
[BDEVNAME_SIZE
];
1503 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s\n",
1504 current
->comm
, task_pid_nr(current
),
1505 (rw
& WRITE
) ? "WRITE" : "READ",
1506 (unsigned long long)bio
->bi_sector
,
1507 bdevname(bio
->bi_bdev
, b
));
1511 generic_make_request(bio
);
1513 EXPORT_SYMBOL(submit_bio
);
1516 * __end_that_request_first - end I/O on a request
1517 * @req: the request being processed
1518 * @error: %0 for success, < %0 for error
1519 * @nr_bytes: number of bytes to complete
1522 * Ends I/O on a number of bytes attached to @req, and sets it up
1523 * for the next range of segments (if any) in the cluster.
1526 * %0 - we are done with this request, call end_that_request_last()
1527 * %1 - still buffers pending for this request
1529 static int __end_that_request_first(struct request
*req
, int error
,
1532 int total_bytes
, bio_nbytes
, next_idx
= 0;
1535 blk_add_trace_rq(req
->q
, req
, BLK_TA_COMPLETE
);
1538 * for a REQ_TYPE_BLOCK_PC request, we want to carry any eventual
1539 * sense key with us all the way through
1541 if (!blk_pc_request(req
))
1544 if (error
&& (blk_fs_request(req
) && !(req
->cmd_flags
& REQ_QUIET
))) {
1545 printk(KERN_ERR
"end_request: I/O error, dev %s, sector %llu\n",
1546 req
->rq_disk
? req
->rq_disk
->disk_name
: "?",
1547 (unsigned long long)req
->sector
);
1550 if (blk_fs_request(req
) && req
->rq_disk
) {
1551 const int rw
= rq_data_dir(req
);
1552 struct hd_struct
*part
;
1555 cpu
= part_stat_lock();
1556 part
= disk_map_sector_rcu(req
->rq_disk
, req
->sector
);
1557 part_stat_add(cpu
, part
, sectors
[rw
], nr_bytes
>> 9);
1561 total_bytes
= bio_nbytes
= 0;
1562 while ((bio
= req
->bio
) != NULL
) {
1566 * For an empty barrier request, the low level driver must
1567 * store a potential error location in ->sector. We pass
1568 * that back up in ->bi_sector.
1570 if (blk_empty_barrier(req
))
1571 bio
->bi_sector
= req
->sector
;
1573 if (nr_bytes
>= bio
->bi_size
) {
1574 req
->bio
= bio
->bi_next
;
1575 nbytes
= bio
->bi_size
;
1576 req_bio_endio(req
, bio
, nbytes
, error
);
1580 int idx
= bio
->bi_idx
+ next_idx
;
1582 if (unlikely(bio
->bi_idx
>= bio
->bi_vcnt
)) {
1583 blk_dump_rq_flags(req
, "__end_that");
1584 printk(KERN_ERR
"%s: bio idx %d >= vcnt %d\n",
1585 __func__
, bio
->bi_idx
, bio
->bi_vcnt
);
1589 nbytes
= bio_iovec_idx(bio
, idx
)->bv_len
;
1590 BIO_BUG_ON(nbytes
> bio
->bi_size
);
1593 * not a complete bvec done
1595 if (unlikely(nbytes
> nr_bytes
)) {
1596 bio_nbytes
+= nr_bytes
;
1597 total_bytes
+= nr_bytes
;
1602 * advance to the next vector
1605 bio_nbytes
+= nbytes
;
1608 total_bytes
+= nbytes
;
1614 * end more in this run, or just return 'not-done'
1616 if (unlikely(nr_bytes
<= 0))
1628 * if the request wasn't completed, update state
1631 req_bio_endio(req
, bio
, bio_nbytes
, error
);
1632 bio
->bi_idx
+= next_idx
;
1633 bio_iovec(bio
)->bv_offset
+= nr_bytes
;
1634 bio_iovec(bio
)->bv_len
-= nr_bytes
;
1637 blk_recalc_rq_sectors(req
, total_bytes
>> 9);
1638 blk_recalc_rq_segments(req
);
1643 * queue lock must be held
1645 static void end_that_request_last(struct request
*req
, int error
)
1647 struct gendisk
*disk
= req
->rq_disk
;
1649 if (blk_rq_tagged(req
))
1650 blk_queue_end_tag(req
->q
, req
);
1652 if (blk_queued_rq(req
))
1653 blkdev_dequeue_request(req
);
1655 if (unlikely(laptop_mode
) && blk_fs_request(req
))
1656 laptop_io_completion();
1659 * Account IO completion. bar_rq isn't accounted as a normal
1660 * IO on queueing nor completion. Accounting the containing
1661 * request is enough.
1663 if (disk
&& blk_fs_request(req
) && req
!= &req
->q
->bar_rq
) {
1664 unsigned long duration
= jiffies
- req
->start_time
;
1665 const int rw
= rq_data_dir(req
);
1666 struct hd_struct
*part
;
1669 cpu
= part_stat_lock();
1670 part
= disk_map_sector_rcu(disk
, req
->sector
);
1672 part_stat_inc(cpu
, part
, ios
[rw
]);
1673 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
1674 part_round_stats(cpu
, part
);
1675 part_dec_in_flight(part
);
1681 req
->end_io(req
, error
);
1683 if (blk_bidi_rq(req
))
1684 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
1686 __blk_put_request(req
->q
, req
);
1690 static inline void __end_request(struct request
*rq
, int uptodate
,
1691 unsigned int nr_bytes
)
1696 error
= uptodate
? uptodate
: -EIO
;
1698 __blk_end_request(rq
, error
, nr_bytes
);
1702 * blk_rq_bytes - Returns bytes left to complete in the entire request
1703 * @rq: the request being processed
1705 unsigned int blk_rq_bytes(struct request
*rq
)
1707 if (blk_fs_request(rq
))
1708 return rq
->hard_nr_sectors
<< 9;
1710 return rq
->data_len
;
1712 EXPORT_SYMBOL_GPL(blk_rq_bytes
);
1715 * blk_rq_cur_bytes - Returns bytes left to complete in the current segment
1716 * @rq: the request being processed
1718 unsigned int blk_rq_cur_bytes(struct request
*rq
)
1720 if (blk_fs_request(rq
))
1721 return rq
->current_nr_sectors
<< 9;
1724 return rq
->bio
->bi_size
;
1726 return rq
->data_len
;
1728 EXPORT_SYMBOL_GPL(blk_rq_cur_bytes
);
1731 * end_queued_request - end all I/O on a queued request
1732 * @rq: the request being processed
1733 * @uptodate: error value or %0/%1 uptodate flag
1736 * Ends all I/O on a request, and removes it from the block layer queues.
1737 * Not suitable for normal I/O completion, unless the driver still has
1738 * the request attached to the block layer.
1741 void end_queued_request(struct request
*rq
, int uptodate
)
1743 __end_request(rq
, uptodate
, blk_rq_bytes(rq
));
1745 EXPORT_SYMBOL(end_queued_request
);
1748 * end_dequeued_request - end all I/O on a dequeued request
1749 * @rq: the request being processed
1750 * @uptodate: error value or %0/%1 uptodate flag
1753 * Ends all I/O on a request. The request must already have been
1754 * dequeued using blkdev_dequeue_request(), as is normally the case
1758 void end_dequeued_request(struct request
*rq
, int uptodate
)
1760 __end_request(rq
, uptodate
, blk_rq_bytes(rq
));
1762 EXPORT_SYMBOL(end_dequeued_request
);
1766 * end_request - end I/O on the current segment of the request
1767 * @req: the request being processed
1768 * @uptodate: error value or %0/%1 uptodate flag
1771 * Ends I/O on the current segment of a request. If that is the only
1772 * remaining segment, the request is also completed and freed.
1774 * This is a remnant of how older block drivers handled I/O completions.
1775 * Modern drivers typically end I/O on the full request in one go, unless
1776 * they have a residual value to account for. For that case this function
1777 * isn't really useful, unless the residual just happens to be the
1778 * full current segment. In other words, don't use this function in new
1779 * code. Either use end_request_completely(), or the
1780 * end_that_request_chunk() (along with end_that_request_last()) for
1781 * partial completions.
1784 void end_request(struct request
*req
, int uptodate
)
1786 __end_request(req
, uptodate
, req
->hard_cur_sectors
<< 9);
1788 EXPORT_SYMBOL(end_request
);
1791 * blk_end_io - Generic end_io function to complete a request.
1792 * @rq: the request being processed
1793 * @error: %0 for success, < %0 for error
1794 * @nr_bytes: number of bytes to complete @rq
1795 * @bidi_bytes: number of bytes to complete @rq->next_rq
1796 * @drv_callback: function called between completion of bios in the request
1797 * and completion of the request.
1798 * If the callback returns non %0, this helper returns without
1799 * completion of the request.
1802 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
1803 * If @rq has leftover, sets it up for the next range of segments.
1806 * %0 - we are done with this request
1807 * %1 - this request is not freed yet, it still has pending buffers.
1809 static int blk_end_io(struct request
*rq
, int error
, unsigned int nr_bytes
,
1810 unsigned int bidi_bytes
,
1811 int (drv_callback
)(struct request
*))
1813 struct request_queue
*q
= rq
->q
;
1814 unsigned long flags
= 0UL;
1816 if (bio_has_data(rq
->bio
) || blk_discard_rq(rq
)) {
1817 if (__end_that_request_first(rq
, error
, nr_bytes
))
1820 /* Bidi request must be completed as a whole */
1821 if (blk_bidi_rq(rq
) &&
1822 __end_that_request_first(rq
->next_rq
, error
, bidi_bytes
))
1826 /* Special feature for tricky drivers */
1827 if (drv_callback
&& drv_callback(rq
))
1830 add_disk_randomness(rq
->rq_disk
);
1832 spin_lock_irqsave(q
->queue_lock
, flags
);
1833 end_that_request_last(rq
, error
);
1834 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1840 * blk_end_request - Helper function for drivers to complete the request.
1841 * @rq: the request being processed
1842 * @error: %0 for success, < %0 for error
1843 * @nr_bytes: number of bytes to complete
1846 * Ends I/O on a number of bytes attached to @rq.
1847 * If @rq has leftover, sets it up for the next range of segments.
1850 * %0 - we are done with this request
1851 * %1 - still buffers pending for this request
1853 int blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
1855 return blk_end_io(rq
, error
, nr_bytes
, 0, NULL
);
1857 EXPORT_SYMBOL_GPL(blk_end_request
);
1860 * __blk_end_request - Helper function for drivers to complete the request.
1861 * @rq: the request being processed
1862 * @error: %0 for success, < %0 for error
1863 * @nr_bytes: number of bytes to complete
1866 * Must be called with queue lock held unlike blk_end_request().
1869 * %0 - we are done with this request
1870 * %1 - still buffers pending for this request
1872 int __blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
1874 if ((bio_has_data(rq
->bio
) || blk_discard_rq(rq
)) &&
1875 __end_that_request_first(rq
, error
, nr_bytes
))
1878 add_disk_randomness(rq
->rq_disk
);
1880 end_that_request_last(rq
, error
);
1884 EXPORT_SYMBOL_GPL(__blk_end_request
);
1887 * blk_end_bidi_request - Helper function for drivers to complete bidi request.
1888 * @rq: the bidi request being processed
1889 * @error: %0 for success, < %0 for error
1890 * @nr_bytes: number of bytes to complete @rq
1891 * @bidi_bytes: number of bytes to complete @rq->next_rq
1894 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
1897 * %0 - we are done with this request
1898 * %1 - still buffers pending for this request
1900 int blk_end_bidi_request(struct request
*rq
, int error
, unsigned int nr_bytes
,
1901 unsigned int bidi_bytes
)
1903 return blk_end_io(rq
, error
, nr_bytes
, bidi_bytes
, NULL
);
1905 EXPORT_SYMBOL_GPL(blk_end_bidi_request
);
1908 * blk_end_request_callback - Special helper function for tricky drivers
1909 * @rq: the request being processed
1910 * @error: %0 for success, < %0 for error
1911 * @nr_bytes: number of bytes to complete
1912 * @drv_callback: function called between completion of bios in the request
1913 * and completion of the request.
1914 * If the callback returns non %0, this helper returns without
1915 * completion of the request.
1918 * Ends I/O on a number of bytes attached to @rq.
1919 * If @rq has leftover, sets it up for the next range of segments.
1921 * This special helper function is used only for existing tricky drivers.
1922 * (e.g. cdrom_newpc_intr() of ide-cd)
1923 * This interface will be removed when such drivers are rewritten.
1924 * Don't use this interface in other places anymore.
1927 * %0 - we are done with this request
1928 * %1 - this request is not freed yet.
1929 * this request still has pending buffers or
1930 * the driver doesn't want to finish this request yet.
1932 int blk_end_request_callback(struct request
*rq
, int error
,
1933 unsigned int nr_bytes
,
1934 int (drv_callback
)(struct request
*))
1936 return blk_end_io(rq
, error
, nr_bytes
, 0, drv_callback
);
1938 EXPORT_SYMBOL_GPL(blk_end_request_callback
);
1940 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
1943 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw, and
1944 we want BIO_RW_AHEAD (bit 1) to imply REQ_FAILFAST (bit 1). */
1945 rq
->cmd_flags
|= (bio
->bi_rw
& 3);
1947 if (bio_has_data(bio
)) {
1948 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
1949 rq
->buffer
= bio_data(bio
);
1951 rq
->current_nr_sectors
= bio_cur_sectors(bio
);
1952 rq
->hard_cur_sectors
= rq
->current_nr_sectors
;
1953 rq
->hard_nr_sectors
= rq
->nr_sectors
= bio_sectors(bio
);
1954 rq
->data_len
= bio
->bi_size
;
1956 rq
->bio
= rq
->biotail
= bio
;
1959 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
1962 int kblockd_schedule_work(struct request_queue
*q
, struct work_struct
*work
)
1964 return queue_work(kblockd_workqueue
, work
);
1966 EXPORT_SYMBOL(kblockd_schedule_work
);
1968 void kblockd_flush_work(struct work_struct
*work
)
1970 cancel_work_sync(work
);
1972 EXPORT_SYMBOL(kblockd_flush_work
);
1974 int __init
blk_dev_init(void)
1976 kblockd_workqueue
= create_workqueue("kblockd");
1977 if (!kblockd_workqueue
)
1978 panic("Failed to create kblockd\n");
1980 request_cachep
= kmem_cache_create("blkdev_requests",
1981 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
1983 blk_requestq_cachep
= kmem_cache_create("blkdev_queue",
1984 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);