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_fs_request(rq
) || !blk_do_io_stat(rq
))
74 cpu
= part_stat_lock();
75 part
= disk_map_sector_rcu(rq
->rq_disk
, rq
->sector
);
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
;
138 EXPORT_SYMBOL(blk_rq_init
);
140 static void req_bio_endio(struct request
*rq
, struct bio
*bio
,
141 unsigned int nbytes
, int error
)
143 struct request_queue
*q
= rq
->q
;
145 if (&q
->bar_rq
!= rq
) {
147 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
148 else if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
151 if (unlikely(nbytes
> bio
->bi_size
)) {
152 printk(KERN_ERR
"%s: want %u bytes done, %u left\n",
153 __func__
, nbytes
, bio
->bi_size
);
154 nbytes
= bio
->bi_size
;
157 if (unlikely(rq
->cmd_flags
& REQ_QUIET
))
158 set_bit(BIO_QUIET
, &bio
->bi_flags
);
160 bio
->bi_size
-= nbytes
;
161 bio
->bi_sector
+= (nbytes
>> 9);
163 if (bio_integrity(bio
))
164 bio_integrity_advance(bio
, nbytes
);
166 if (bio
->bi_size
== 0)
167 bio_endio(bio
, error
);
171 * Okay, this is the barrier request in progress, just
174 if (error
&& !q
->orderr
)
179 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
183 printk(KERN_INFO
"%s: dev %s: type=%x, flags=%x\n", msg
,
184 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->cmd_type
,
187 printk(KERN_INFO
" sector %llu, nr/cnr %lu/%u\n",
188 (unsigned long long)rq
->sector
,
190 rq
->current_nr_sectors
);
191 printk(KERN_INFO
" bio %p, biotail %p, buffer %p, data %p, len %u\n",
192 rq
->bio
, rq
->biotail
,
193 rq
->buffer
, rq
->data
,
196 if (blk_pc_request(rq
)) {
197 printk(KERN_INFO
" cdb: ");
198 for (bit
= 0; bit
< BLK_MAX_CDB
; bit
++)
199 printk("%02x ", rq
->cmd
[bit
]);
203 EXPORT_SYMBOL(blk_dump_rq_flags
);
206 * "plug" the device if there are no outstanding requests: this will
207 * force the transfer to start only after we have put all the requests
210 * This is called with interrupts off and no requests on the queue and
211 * with the queue lock held.
213 void blk_plug_device(struct request_queue
*q
)
215 WARN_ON(!irqs_disabled());
218 * don't plug a stopped queue, it must be paired with blk_start_queue()
219 * which will restart the queueing
221 if (blk_queue_stopped(q
))
224 if (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED
, q
)) {
225 mod_timer(&q
->unplug_timer
, jiffies
+ q
->unplug_delay
);
229 EXPORT_SYMBOL(blk_plug_device
);
232 * blk_plug_device_unlocked - plug a device without queue lock held
233 * @q: The &struct request_queue to plug
236 * Like @blk_plug_device(), but grabs the queue lock and disables
239 void blk_plug_device_unlocked(struct request_queue
*q
)
243 spin_lock_irqsave(q
->queue_lock
, flags
);
245 spin_unlock_irqrestore(q
->queue_lock
, flags
);
247 EXPORT_SYMBOL(blk_plug_device_unlocked
);
250 * remove the queue from the plugged list, if present. called with
251 * queue lock held and interrupts disabled.
253 int blk_remove_plug(struct request_queue
*q
)
255 WARN_ON(!irqs_disabled());
257 if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED
, q
))
260 del_timer(&q
->unplug_timer
);
263 EXPORT_SYMBOL(blk_remove_plug
);
266 * remove the plug and let it rip..
268 void __generic_unplug_device(struct request_queue
*q
)
270 if (unlikely(blk_queue_stopped(q
)))
272 if (!blk_remove_plug(q
) && !blk_queue_nonrot(q
))
279 * generic_unplug_device - fire a request queue
280 * @q: The &struct request_queue in question
283 * Linux uses plugging to build bigger requests queues before letting
284 * the device have at them. If a queue is plugged, the I/O scheduler
285 * is still adding and merging requests on the queue. Once the queue
286 * gets unplugged, the request_fn defined for the queue is invoked and
289 void generic_unplug_device(struct request_queue
*q
)
291 if (blk_queue_plugged(q
)) {
292 spin_lock_irq(q
->queue_lock
);
293 __generic_unplug_device(q
);
294 spin_unlock_irq(q
->queue_lock
);
297 EXPORT_SYMBOL(generic_unplug_device
);
299 static void blk_backing_dev_unplug(struct backing_dev_info
*bdi
,
302 struct request_queue
*q
= bdi
->unplug_io_data
;
307 void blk_unplug_work(struct work_struct
*work
)
309 struct request_queue
*q
=
310 container_of(work
, struct request_queue
, unplug_work
);
312 trace_block_unplug_io(q
);
316 void blk_unplug_timeout(unsigned long data
)
318 struct request_queue
*q
= (struct request_queue
*)data
;
320 trace_block_unplug_timer(q
);
321 kblockd_schedule_work(q
, &q
->unplug_work
);
324 void blk_unplug(struct request_queue
*q
)
327 * devices don't necessarily have an ->unplug_fn defined
330 trace_block_unplug_io(q
);
334 EXPORT_SYMBOL(blk_unplug
);
337 * blk_start_queue - restart a previously stopped queue
338 * @q: The &struct request_queue in question
341 * blk_start_queue() will clear the stop flag on the queue, and call
342 * the request_fn for the queue if it was in a stopped state when
343 * entered. Also see blk_stop_queue(). Queue lock must be held.
345 void blk_start_queue(struct request_queue
*q
)
347 WARN_ON(!irqs_disabled());
349 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
352 EXPORT_SYMBOL(blk_start_queue
);
355 * blk_stop_queue - stop a queue
356 * @q: The &struct request_queue in question
359 * The Linux block layer assumes that a block driver will consume all
360 * entries on the request queue when the request_fn strategy is called.
361 * Often this will not happen, because of hardware limitations (queue
362 * depth settings). If a device driver gets a 'queue full' response,
363 * or if it simply chooses not to queue more I/O at one point, it can
364 * call this function to prevent the request_fn from being called until
365 * the driver has signalled it's ready to go again. This happens by calling
366 * blk_start_queue() to restart queue operations. Queue lock must be held.
368 void blk_stop_queue(struct request_queue
*q
)
371 queue_flag_set(QUEUE_FLAG_STOPPED
, q
);
373 EXPORT_SYMBOL(blk_stop_queue
);
376 * blk_sync_queue - cancel any pending callbacks on a queue
380 * The block layer may perform asynchronous callback activity
381 * on a queue, such as calling the unplug function after a timeout.
382 * A block device may call blk_sync_queue to ensure that any
383 * such activity is cancelled, thus allowing it to release resources
384 * that the callbacks might use. The caller must already have made sure
385 * that its ->make_request_fn will not re-add plugging prior to calling
389 void blk_sync_queue(struct request_queue
*q
)
391 del_timer_sync(&q
->unplug_timer
);
392 del_timer_sync(&q
->timeout
);
393 cancel_work_sync(&q
->unplug_work
);
395 EXPORT_SYMBOL(blk_sync_queue
);
398 * __blk_run_queue - run a single device queue
399 * @q: The queue to run
402 * See @blk_run_queue. This variant must be called with the queue lock
403 * held and interrupts disabled.
406 void __blk_run_queue(struct request_queue
*q
)
410 if (unlikely(blk_queue_stopped(q
)))
413 if (elv_queue_empty(q
))
417 * Only recurse once to avoid overrunning the stack, let the unplug
418 * handling reinvoke the handler shortly if we already got there.
420 if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER
, q
)) {
422 queue_flag_clear(QUEUE_FLAG_REENTER
, q
);
424 queue_flag_set(QUEUE_FLAG_PLUGGED
, q
);
425 kblockd_schedule_work(q
, &q
->unplug_work
);
428 EXPORT_SYMBOL(__blk_run_queue
);
431 * blk_run_queue - run a single device queue
432 * @q: The queue to run
435 * Invoke request handling on this queue, if it has pending work to do.
436 * May be used to restart queueing when a request has completed. Also
437 * See @blk_start_queueing.
440 void blk_run_queue(struct request_queue
*q
)
444 spin_lock_irqsave(q
->queue_lock
, flags
);
446 spin_unlock_irqrestore(q
->queue_lock
, flags
);
448 EXPORT_SYMBOL(blk_run_queue
);
450 void blk_put_queue(struct request_queue
*q
)
452 kobject_put(&q
->kobj
);
455 void blk_cleanup_queue(struct request_queue
*q
)
458 * We know we have process context here, so we can be a little
459 * cautious and ensure that pending block actions on this device
460 * are done before moving on. Going into this function, we should
461 * not have processes doing IO to this device.
465 mutex_lock(&q
->sysfs_lock
);
466 queue_flag_set_unlocked(QUEUE_FLAG_DEAD
, q
);
467 mutex_unlock(&q
->sysfs_lock
);
470 elevator_exit(q
->elevator
);
474 EXPORT_SYMBOL(blk_cleanup_queue
);
476 static int blk_init_free_list(struct request_queue
*q
)
478 struct request_list
*rl
= &q
->rq
;
480 rl
->count
[BLK_RW_SYNC
] = rl
->count
[BLK_RW_ASYNC
] = 0;
481 rl
->starved
[BLK_RW_SYNC
] = rl
->starved
[BLK_RW_ASYNC
] = 0;
483 init_waitqueue_head(&rl
->wait
[BLK_RW_SYNC
]);
484 init_waitqueue_head(&rl
->wait
[BLK_RW_ASYNC
]);
486 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, mempool_alloc_slab
,
487 mempool_free_slab
, request_cachep
, q
->node
);
495 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
497 return blk_alloc_queue_node(gfp_mask
, -1);
499 EXPORT_SYMBOL(blk_alloc_queue
);
501 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
503 struct request_queue
*q
;
506 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
507 gfp_mask
| __GFP_ZERO
, node_id
);
511 q
->backing_dev_info
.unplug_io_fn
= blk_backing_dev_unplug
;
512 q
->backing_dev_info
.unplug_io_data
= q
;
513 err
= bdi_init(&q
->backing_dev_info
);
515 kmem_cache_free(blk_requestq_cachep
, q
);
519 init_timer(&q
->unplug_timer
);
520 setup_timer(&q
->timeout
, blk_rq_timed_out_timer
, (unsigned long) q
);
521 INIT_LIST_HEAD(&q
->timeout_list
);
522 INIT_WORK(&q
->unplug_work
, blk_unplug_work
);
524 kobject_init(&q
->kobj
, &blk_queue_ktype
);
526 mutex_init(&q
->sysfs_lock
);
527 spin_lock_init(&q
->__queue_lock
);
531 EXPORT_SYMBOL(blk_alloc_queue_node
);
534 * blk_init_queue - prepare a request queue for use with a block device
535 * @rfn: The function to be called to process requests that have been
536 * placed on the queue.
537 * @lock: Request queue spin lock
540 * If a block device wishes to use the standard request handling procedures,
541 * which sorts requests and coalesces adjacent requests, then it must
542 * call blk_init_queue(). The function @rfn will be called when there
543 * are requests on the queue that need to be processed. If the device
544 * supports plugging, then @rfn may not be called immediately when requests
545 * are available on the queue, but may be called at some time later instead.
546 * Plugged queues are generally unplugged when a buffer belonging to one
547 * of the requests on the queue is needed, or due to memory pressure.
549 * @rfn is not required, or even expected, to remove all requests off the
550 * queue, but only as many as it can handle at a time. If it does leave
551 * requests on the queue, it is responsible for arranging that the requests
552 * get dealt with eventually.
554 * The queue spin lock must be held while manipulating the requests on the
555 * request queue; this lock will be taken also from interrupt context, so irq
556 * disabling is needed for it.
558 * Function returns a pointer to the initialized request queue, or %NULL if
562 * blk_init_queue() must be paired with a blk_cleanup_queue() call
563 * when the block device is deactivated (such as at module unload).
566 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
568 return blk_init_queue_node(rfn
, lock
, -1);
570 EXPORT_SYMBOL(blk_init_queue
);
572 struct request_queue
*
573 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
575 struct request_queue
*q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
581 if (blk_init_free_list(q
)) {
582 kmem_cache_free(blk_requestq_cachep
, q
);
587 * if caller didn't supply a lock, they get per-queue locking with
591 lock
= &q
->__queue_lock
;
594 q
->prep_rq_fn
= NULL
;
595 q
->unplug_fn
= generic_unplug_device
;
596 q
->queue_flags
= QUEUE_FLAG_DEFAULT
;
597 q
->queue_lock
= lock
;
600 * This also sets hw/phys segments, boundary and size
602 blk_queue_make_request(q
, __make_request
);
604 q
->sg_reserved_size
= INT_MAX
;
606 blk_set_cmd_filter_defaults(&q
->cmd_filter
);
611 if (!elevator_init(q
, NULL
)) {
612 blk_queue_congestion_threshold(q
);
619 EXPORT_SYMBOL(blk_init_queue_node
);
621 int blk_get_queue(struct request_queue
*q
)
623 if (likely(!test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
))) {
624 kobject_get(&q
->kobj
);
631 static inline void blk_free_request(struct request_queue
*q
, struct request
*rq
)
633 if (rq
->cmd_flags
& REQ_ELVPRIV
)
634 elv_put_request(q
, rq
);
635 mempool_free(rq
, q
->rq
.rq_pool
);
638 static struct request
*
639 blk_alloc_request(struct request_queue
*q
, int flags
, int priv
, gfp_t gfp_mask
)
641 struct request
*rq
= mempool_alloc(q
->rq
.rq_pool
, gfp_mask
);
648 rq
->cmd_flags
= flags
| REQ_ALLOCED
;
651 if (unlikely(elv_set_request(q
, rq
, gfp_mask
))) {
652 mempool_free(rq
, q
->rq
.rq_pool
);
655 rq
->cmd_flags
|= REQ_ELVPRIV
;
662 * ioc_batching returns true if the ioc is a valid batching request and
663 * should be given priority access to a request.
665 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
671 * Make sure the process is able to allocate at least 1 request
672 * even if the batch times out, otherwise we could theoretically
675 return ioc
->nr_batch_requests
== q
->nr_batching
||
676 (ioc
->nr_batch_requests
> 0
677 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
681 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
682 * will cause the process to be a "batcher" on all queues in the system. This
683 * is the behaviour we want though - once it gets a wakeup it should be given
686 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
688 if (!ioc
|| ioc_batching(q
, ioc
))
691 ioc
->nr_batch_requests
= q
->nr_batching
;
692 ioc
->last_waited
= jiffies
;
695 static void __freed_request(struct request_queue
*q
, int sync
)
697 struct request_list
*rl
= &q
->rq
;
699 if (rl
->count
[sync
] < queue_congestion_off_threshold(q
))
700 blk_clear_queue_congested(q
, sync
);
702 if (rl
->count
[sync
] + 1 <= q
->nr_requests
) {
703 if (waitqueue_active(&rl
->wait
[sync
]))
704 wake_up(&rl
->wait
[sync
]);
706 blk_clear_queue_full(q
, sync
);
711 * A request has just been released. Account for it, update the full and
712 * congestion status, wake up any waiters. Called under q->queue_lock.
714 static void freed_request(struct request_queue
*q
, int sync
, int priv
)
716 struct request_list
*rl
= &q
->rq
;
722 __freed_request(q
, sync
);
724 if (unlikely(rl
->starved
[sync
^ 1]))
725 __freed_request(q
, sync
^ 1);
729 * Get a free request, queue_lock must be held.
730 * Returns NULL on failure, with queue_lock held.
731 * Returns !NULL on success, with queue_lock *not held*.
733 static struct request
*get_request(struct request_queue
*q
, int rw_flags
,
734 struct bio
*bio
, gfp_t gfp_mask
)
736 struct request
*rq
= NULL
;
737 struct request_list
*rl
= &q
->rq
;
738 struct io_context
*ioc
= NULL
;
739 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
742 may_queue
= elv_may_queue(q
, rw_flags
);
743 if (may_queue
== ELV_MQUEUE_NO
)
746 if (rl
->count
[is_sync
]+1 >= queue_congestion_on_threshold(q
)) {
747 if (rl
->count
[is_sync
]+1 >= q
->nr_requests
) {
748 ioc
= current_io_context(GFP_ATOMIC
, q
->node
);
750 * The queue will fill after this allocation, so set
751 * it as full, and mark this process as "batching".
752 * This process will be allowed to complete a batch of
753 * requests, others will be blocked.
755 if (!blk_queue_full(q
, is_sync
)) {
756 ioc_set_batching(q
, ioc
);
757 blk_set_queue_full(q
, is_sync
);
759 if (may_queue
!= ELV_MQUEUE_MUST
760 && !ioc_batching(q
, ioc
)) {
762 * The queue is full and the allocating
763 * process is not a "batcher", and not
764 * exempted by the IO scheduler
770 blk_set_queue_congested(q
, is_sync
);
774 * Only allow batching queuers to allocate up to 50% over the defined
775 * limit of requests, otherwise we could have thousands of requests
776 * allocated with any setting of ->nr_requests
778 if (rl
->count
[is_sync
] >= (3 * q
->nr_requests
/ 2))
781 rl
->count
[is_sync
]++;
782 rl
->starved
[is_sync
] = 0;
784 priv
= !test_bit(QUEUE_FLAG_ELVSWITCH
, &q
->queue_flags
);
788 if (blk_queue_io_stat(q
))
789 rw_flags
|= REQ_IO_STAT
;
790 spin_unlock_irq(q
->queue_lock
);
792 rq
= blk_alloc_request(q
, rw_flags
, priv
, gfp_mask
);
795 * Allocation failed presumably due to memory. Undo anything
796 * we might have messed up.
798 * Allocating task should really be put onto the front of the
799 * wait queue, but this is pretty rare.
801 spin_lock_irq(q
->queue_lock
);
802 freed_request(q
, is_sync
, priv
);
805 * in the very unlikely event that allocation failed and no
806 * requests for this direction was pending, mark us starved
807 * so that freeing of a request in the other direction will
808 * notice us. another possible fix would be to split the
809 * rq mempool into READ and WRITE
812 if (unlikely(rl
->count
[is_sync
] == 0))
813 rl
->starved
[is_sync
] = 1;
819 * ioc may be NULL here, and ioc_batching will be false. That's
820 * OK, if the queue is under the request limit then requests need
821 * not count toward the nr_batch_requests limit. There will always
822 * be some limit enforced by BLK_BATCH_TIME.
824 if (ioc_batching(q
, ioc
))
825 ioc
->nr_batch_requests
--;
827 trace_block_getrq(q
, bio
, rw_flags
& 1);
833 * No available requests for this queue, unplug the device and wait for some
834 * requests to become available.
836 * Called with q->queue_lock held, and returns with it unlocked.
838 static struct request
*get_request_wait(struct request_queue
*q
, int rw_flags
,
841 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
844 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
847 struct io_context
*ioc
;
848 struct request_list
*rl
= &q
->rq
;
850 prepare_to_wait_exclusive(&rl
->wait
[is_sync
], &wait
,
851 TASK_UNINTERRUPTIBLE
);
853 trace_block_sleeprq(q
, bio
, rw_flags
& 1);
855 __generic_unplug_device(q
);
856 spin_unlock_irq(q
->queue_lock
);
860 * After sleeping, we become a "batching" process and
861 * will be able to allocate at least one request, and
862 * up to a big batch of them for a small period time.
863 * See ioc_batching, ioc_set_batching
865 ioc
= current_io_context(GFP_NOIO
, q
->node
);
866 ioc_set_batching(q
, ioc
);
868 spin_lock_irq(q
->queue_lock
);
869 finish_wait(&rl
->wait
[is_sync
], &wait
);
871 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
877 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
881 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
883 spin_lock_irq(q
->queue_lock
);
884 if (gfp_mask
& __GFP_WAIT
) {
885 rq
= get_request_wait(q
, rw
, NULL
);
887 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
889 spin_unlock_irq(q
->queue_lock
);
891 /* q->queue_lock is unlocked at this point */
895 EXPORT_SYMBOL(blk_get_request
);
898 * blk_start_queueing - initiate dispatch of requests to device
899 * @q: request queue to kick into gear
901 * This is basically a helper to remove the need to know whether a queue
902 * is plugged or not if someone just wants to initiate dispatch of requests
903 * for this queue. Should be used to start queueing on a device outside
904 * of ->request_fn() context. Also see @blk_run_queue.
906 * The queue lock must be held with interrupts disabled.
908 void blk_start_queueing(struct request_queue
*q
)
910 if (!blk_queue_plugged(q
)) {
911 if (unlikely(blk_queue_stopped(q
)))
915 __generic_unplug_device(q
);
917 EXPORT_SYMBOL(blk_start_queueing
);
920 * blk_requeue_request - put a request back on queue
921 * @q: request queue where request should be inserted
922 * @rq: request to be inserted
925 * Drivers often keep queueing requests until the hardware cannot accept
926 * more, when that condition happens we need to put the request back
927 * on the queue. Must be called with queue lock held.
929 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
931 blk_delete_timer(rq
);
932 blk_clear_rq_complete(rq
);
933 trace_block_rq_requeue(q
, rq
);
935 if (blk_rq_tagged(rq
))
936 blk_queue_end_tag(q
, rq
);
938 elv_requeue_request(q
, rq
);
940 EXPORT_SYMBOL(blk_requeue_request
);
943 * blk_insert_request - insert a special request into a request queue
944 * @q: request queue where request should be inserted
945 * @rq: request to be inserted
946 * @at_head: insert request at head or tail of queue
947 * @data: private data
950 * Many block devices need to execute commands asynchronously, so they don't
951 * block the whole kernel from preemption during request execution. This is
952 * accomplished normally by inserting aritficial requests tagged as
953 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
954 * be scheduled for actual execution by the request queue.
956 * We have the option of inserting the head or the tail of the queue.
957 * Typically we use the tail for new ioctls and so forth. We use the head
958 * of the queue for things like a QUEUE_FULL message from a device, or a
959 * host that is unable to accept a particular command.
961 void blk_insert_request(struct request_queue
*q
, struct request
*rq
,
962 int at_head
, void *data
)
964 int where
= at_head
? ELEVATOR_INSERT_FRONT
: ELEVATOR_INSERT_BACK
;
968 * tell I/O scheduler that this isn't a regular read/write (ie it
969 * must not attempt merges on this) and that it acts as a soft
972 rq
->cmd_type
= REQ_TYPE_SPECIAL
;
973 rq
->cmd_flags
|= REQ_SOFTBARRIER
;
977 spin_lock_irqsave(q
->queue_lock
, flags
);
980 * If command is tagged, release the tag
982 if (blk_rq_tagged(rq
))
983 blk_queue_end_tag(q
, rq
);
985 drive_stat_acct(rq
, 1);
986 __elv_add_request(q
, rq
, where
, 0);
987 blk_start_queueing(q
);
988 spin_unlock_irqrestore(q
->queue_lock
, flags
);
990 EXPORT_SYMBOL(blk_insert_request
);
993 * add-request adds a request to the linked list.
994 * queue lock is held and interrupts disabled, as we muck with the
995 * request queue list.
997 static inline void add_request(struct request_queue
*q
, struct request
*req
)
999 drive_stat_acct(req
, 1);
1002 * elevator indicated where it wants this request to be
1003 * inserted at elevator_merge time
1005 __elv_add_request(q
, req
, ELEVATOR_INSERT_SORT
, 0);
1008 static void part_round_stats_single(int cpu
, struct hd_struct
*part
,
1011 if (now
== part
->stamp
)
1014 if (part
->in_flight
) {
1015 __part_stat_add(cpu
, part
, time_in_queue
,
1016 part
->in_flight
* (now
- part
->stamp
));
1017 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
1023 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1024 * @cpu: cpu number for stats access
1025 * @part: target partition
1027 * The average IO queue length and utilisation statistics are maintained
1028 * by observing the current state of the queue length and the amount of
1029 * time it has been in this state for.
1031 * Normally, that accounting is done on IO completion, but that can result
1032 * in more than a second's worth of IO being accounted for within any one
1033 * second, leading to >100% utilisation. To deal with that, we call this
1034 * function to do a round-off before returning the results when reading
1035 * /proc/diskstats. This accounts immediately for all queue usage up to
1036 * the current jiffies and restarts the counters again.
1038 void part_round_stats(int cpu
, struct hd_struct
*part
)
1040 unsigned long now
= jiffies
;
1043 part_round_stats_single(cpu
, &part_to_disk(part
)->part0
, now
);
1044 part_round_stats_single(cpu
, part
, now
);
1046 EXPORT_SYMBOL_GPL(part_round_stats
);
1049 * queue lock must be held
1051 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1055 if (unlikely(--req
->ref_count
))
1058 elv_completed_request(q
, req
);
1060 /* this is a bio leak */
1061 WARN_ON(req
->bio
!= NULL
);
1064 * Request may not have originated from ll_rw_blk. if not,
1065 * it didn't come out of our reserved rq pools
1067 if (req
->cmd_flags
& REQ_ALLOCED
) {
1068 int is_sync
= rq_is_sync(req
) != 0;
1069 int priv
= req
->cmd_flags
& REQ_ELVPRIV
;
1071 BUG_ON(!list_empty(&req
->queuelist
));
1072 BUG_ON(!hlist_unhashed(&req
->hash
));
1074 blk_free_request(q
, req
);
1075 freed_request(q
, is_sync
, priv
);
1078 EXPORT_SYMBOL_GPL(__blk_put_request
);
1080 void blk_put_request(struct request
*req
)
1082 unsigned long flags
;
1083 struct request_queue
*q
= req
->q
;
1085 spin_lock_irqsave(q
->queue_lock
, flags
);
1086 __blk_put_request(q
, req
);
1087 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1089 EXPORT_SYMBOL(blk_put_request
);
1091 void init_request_from_bio(struct request
*req
, struct bio
*bio
)
1093 req
->cpu
= bio
->bi_comp_cpu
;
1094 req
->cmd_type
= REQ_TYPE_FS
;
1097 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
1099 if (bio_rw_ahead(bio
))
1100 req
->cmd_flags
|= (REQ_FAILFAST_DEV
| REQ_FAILFAST_TRANSPORT
|
1101 REQ_FAILFAST_DRIVER
);
1102 if (bio_failfast_dev(bio
))
1103 req
->cmd_flags
|= REQ_FAILFAST_DEV
;
1104 if (bio_failfast_transport(bio
))
1105 req
->cmd_flags
|= REQ_FAILFAST_TRANSPORT
;
1106 if (bio_failfast_driver(bio
))
1107 req
->cmd_flags
|= REQ_FAILFAST_DRIVER
;
1110 * REQ_BARRIER implies no merging, but lets make it explicit
1112 if (unlikely(bio_discard(bio
))) {
1113 req
->cmd_flags
|= REQ_DISCARD
;
1114 if (bio_barrier(bio
))
1115 req
->cmd_flags
|= REQ_SOFTBARRIER
;
1116 req
->q
->prepare_discard_fn(req
->q
, req
);
1117 } else if (unlikely(bio_barrier(bio
)))
1118 req
->cmd_flags
|= (REQ_HARDBARRIER
| REQ_NOMERGE
);
1121 req
->cmd_flags
|= REQ_RW_SYNC
;
1122 if (bio_rw_meta(bio
))
1123 req
->cmd_flags
|= REQ_RW_META
;
1124 if (bio_noidle(bio
))
1125 req
->cmd_flags
|= REQ_NOIDLE
;
1128 req
->hard_sector
= req
->sector
= bio
->bi_sector
;
1129 req
->ioprio
= bio_prio(bio
);
1130 req
->start_time
= jiffies
;
1131 blk_rq_bio_prep(req
->q
, req
, bio
);
1135 * Only disabling plugging for non-rotational devices if it does tagging
1136 * as well, otherwise we do need the proper merging
1138 static inline bool queue_should_plug(struct request_queue
*q
)
1140 return !(blk_queue_nonrot(q
) && blk_queue_tagged(q
));
1143 static int __make_request(struct request_queue
*q
, struct bio
*bio
)
1145 struct request
*req
;
1146 int el_ret
, nr_sectors
;
1147 const unsigned short prio
= bio_prio(bio
);
1148 const int sync
= bio_sync(bio
);
1149 const int unplug
= bio_unplug(bio
);
1152 nr_sectors
= bio_sectors(bio
);
1155 * low level driver can indicate that it wants pages above a
1156 * certain limit bounced to low memory (ie for highmem, or even
1157 * ISA dma in theory)
1159 blk_queue_bounce(q
, &bio
);
1161 spin_lock_irq(q
->queue_lock
);
1163 if (unlikely(bio_barrier(bio
)) || elv_queue_empty(q
))
1166 el_ret
= elv_merge(q
, &req
, bio
);
1168 case ELEVATOR_BACK_MERGE
:
1169 BUG_ON(!rq_mergeable(req
));
1171 if (!ll_back_merge_fn(q
, req
, bio
))
1174 trace_block_bio_backmerge(q
, bio
);
1176 req
->biotail
->bi_next
= bio
;
1178 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
1179 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
1180 if (!blk_rq_cpu_valid(req
))
1181 req
->cpu
= bio
->bi_comp_cpu
;
1182 drive_stat_acct(req
, 0);
1183 if (!attempt_back_merge(q
, req
))
1184 elv_merged_request(q
, req
, el_ret
);
1187 case ELEVATOR_FRONT_MERGE
:
1188 BUG_ON(!rq_mergeable(req
));
1190 if (!ll_front_merge_fn(q
, req
, bio
))
1193 trace_block_bio_frontmerge(q
, bio
);
1195 bio
->bi_next
= req
->bio
;
1199 * may not be valid. if the low level driver said
1200 * it didn't need a bounce buffer then it better
1201 * not touch req->buffer either...
1203 req
->buffer
= bio_data(bio
);
1204 req
->current_nr_sectors
= bio_cur_sectors(bio
);
1205 req
->hard_cur_sectors
= req
->current_nr_sectors
;
1206 req
->sector
= req
->hard_sector
= bio
->bi_sector
;
1207 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
1208 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
1209 if (!blk_rq_cpu_valid(req
))
1210 req
->cpu
= bio
->bi_comp_cpu
;
1211 drive_stat_acct(req
, 0);
1212 if (!attempt_front_merge(q
, req
))
1213 elv_merged_request(q
, req
, el_ret
);
1216 /* ELV_NO_MERGE: elevator says don't/can't merge. */
1223 * This sync check and mask will be re-done in init_request_from_bio(),
1224 * but we need to set it earlier to expose the sync flag to the
1225 * rq allocator and io schedulers.
1227 rw_flags
= bio_data_dir(bio
);
1229 rw_flags
|= REQ_RW_SYNC
;
1232 * Grab a free request. This is might sleep but can not fail.
1233 * Returns with the queue unlocked.
1235 req
= get_request_wait(q
, rw_flags
, bio
);
1238 * After dropping the lock and possibly sleeping here, our request
1239 * may now be mergeable after it had proven unmergeable (above).
1240 * We don't worry about that case for efficiency. It won't happen
1241 * often, and the elevators are able to handle it.
1243 init_request_from_bio(req
, bio
);
1245 spin_lock_irq(q
->queue_lock
);
1246 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
) ||
1247 bio_flagged(bio
, BIO_CPU_AFFINE
))
1248 req
->cpu
= blk_cpu_to_group(smp_processor_id());
1249 if (queue_should_plug(q
) && elv_queue_empty(q
))
1251 add_request(q
, req
);
1253 if (unplug
|| !queue_should_plug(q
))
1254 __generic_unplug_device(q
);
1255 spin_unlock_irq(q
->queue_lock
);
1260 * If bio->bi_dev is a partition, remap the location
1262 static inline void blk_partition_remap(struct bio
*bio
)
1264 struct block_device
*bdev
= bio
->bi_bdev
;
1266 if (bio_sectors(bio
) && bdev
!= bdev
->bd_contains
) {
1267 struct hd_struct
*p
= bdev
->bd_part
;
1269 bio
->bi_sector
+= p
->start_sect
;
1270 bio
->bi_bdev
= bdev
->bd_contains
;
1272 trace_block_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
1273 bdev
->bd_dev
, bio
->bi_sector
,
1274 bio
->bi_sector
- p
->start_sect
);
1278 static void handle_bad_sector(struct bio
*bio
)
1280 char b
[BDEVNAME_SIZE
];
1282 printk(KERN_INFO
"attempt to access beyond end of device\n");
1283 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
1284 bdevname(bio
->bi_bdev
, b
),
1286 (unsigned long long)bio
->bi_sector
+ bio_sectors(bio
),
1287 (long long)(bio
->bi_bdev
->bd_inode
->i_size
>> 9));
1289 set_bit(BIO_EOF
, &bio
->bi_flags
);
1292 #ifdef CONFIG_FAIL_MAKE_REQUEST
1294 static DECLARE_FAULT_ATTR(fail_make_request
);
1296 static int __init
setup_fail_make_request(char *str
)
1298 return setup_fault_attr(&fail_make_request
, str
);
1300 __setup("fail_make_request=", setup_fail_make_request
);
1302 static int should_fail_request(struct bio
*bio
)
1304 struct hd_struct
*part
= bio
->bi_bdev
->bd_part
;
1306 if (part_to_disk(part
)->part0
.make_it_fail
|| part
->make_it_fail
)
1307 return should_fail(&fail_make_request
, bio
->bi_size
);
1312 static int __init
fail_make_request_debugfs(void)
1314 return init_fault_attr_dentries(&fail_make_request
,
1315 "fail_make_request");
1318 late_initcall(fail_make_request_debugfs
);
1320 #else /* CONFIG_FAIL_MAKE_REQUEST */
1322 static inline int should_fail_request(struct bio
*bio
)
1327 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1330 * Check whether this bio extends beyond the end of the device.
1332 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
1339 /* Test device or partition size, when known. */
1340 maxsector
= bio
->bi_bdev
->bd_inode
->i_size
>> 9;
1342 sector_t sector
= bio
->bi_sector
;
1344 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
1346 * This may well happen - the kernel calls bread()
1347 * without checking the size of the device, e.g., when
1348 * mounting a device.
1350 handle_bad_sector(bio
);
1359 * generic_make_request - hand a buffer to its device driver for I/O
1360 * @bio: The bio describing the location in memory and on the device.
1362 * generic_make_request() is used to make I/O requests of block
1363 * devices. It is passed a &struct bio, which describes the I/O that needs
1366 * generic_make_request() does not return any status. The
1367 * success/failure status of the request, along with notification of
1368 * completion, is delivered asynchronously through the bio->bi_end_io
1369 * function described (one day) else where.
1371 * The caller of generic_make_request must make sure that bi_io_vec
1372 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1373 * set to describe the device address, and the
1374 * bi_end_io and optionally bi_private are set to describe how
1375 * completion notification should be signaled.
1377 * generic_make_request and the drivers it calls may use bi_next if this
1378 * bio happens to be merged with someone else, and may change bi_dev and
1379 * bi_sector for remaps as it sees fit. So the values of these fields
1380 * should NOT be depended on after the call to generic_make_request.
1382 static inline void __generic_make_request(struct bio
*bio
)
1384 struct request_queue
*q
;
1385 sector_t old_sector
;
1386 int ret
, nr_sectors
= bio_sectors(bio
);
1392 if (bio_check_eod(bio
, nr_sectors
))
1396 * Resolve the mapping until finished. (drivers are
1397 * still free to implement/resolve their own stacking
1398 * by explicitly returning 0)
1400 * NOTE: we don't repeat the blk_size check for each new device.
1401 * Stacking drivers are expected to know what they are doing.
1406 char b
[BDEVNAME_SIZE
];
1408 q
= bdev_get_queue(bio
->bi_bdev
);
1411 "generic_make_request: Trying to access "
1412 "nonexistent block-device %s (%Lu)\n",
1413 bdevname(bio
->bi_bdev
, b
),
1414 (long long) bio
->bi_sector
);
1418 if (unlikely(nr_sectors
> q
->max_hw_sectors
)) {
1419 printk(KERN_ERR
"bio too big device %s (%u > %u)\n",
1420 bdevname(bio
->bi_bdev
, b
),
1426 if (unlikely(test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)))
1429 if (should_fail_request(bio
))
1433 * If this device has partitions, remap block n
1434 * of partition p to block n+start(p) of the disk.
1436 blk_partition_remap(bio
);
1438 if (bio_integrity_enabled(bio
) && bio_integrity_prep(bio
))
1441 if (old_sector
!= -1)
1442 trace_block_remap(q
, bio
, old_dev
, bio
->bi_sector
,
1445 trace_block_bio_queue(q
, bio
);
1447 old_sector
= bio
->bi_sector
;
1448 old_dev
= bio
->bi_bdev
->bd_dev
;
1450 if (bio_check_eod(bio
, nr_sectors
))
1453 if (bio_discard(bio
) && !q
->prepare_discard_fn
) {
1457 if (bio_barrier(bio
) && bio_has_data(bio
) &&
1458 (q
->next_ordered
== QUEUE_ORDERED_NONE
)) {
1463 ret
= q
->make_request_fn(q
, bio
);
1469 bio_endio(bio
, err
);
1473 * We only want one ->make_request_fn to be active at a time,
1474 * else stack usage with stacked devices could be a problem.
1475 * So use current->bio_{list,tail} to keep a list of requests
1476 * submited by a make_request_fn function.
1477 * current->bio_tail is also used as a flag to say if
1478 * generic_make_request is currently active in this task or not.
1479 * If it is NULL, then no make_request is active. If it is non-NULL,
1480 * then a make_request is active, and new requests should be added
1483 void generic_make_request(struct bio
*bio
)
1485 if (current
->bio_tail
) {
1486 /* make_request is active */
1487 *(current
->bio_tail
) = bio
;
1488 bio
->bi_next
= NULL
;
1489 current
->bio_tail
= &bio
->bi_next
;
1492 /* following loop may be a bit non-obvious, and so deserves some
1494 * Before entering the loop, bio->bi_next is NULL (as all callers
1495 * ensure that) so we have a list with a single bio.
1496 * We pretend that we have just taken it off a longer list, so
1497 * we assign bio_list to the next (which is NULL) and bio_tail
1498 * to &bio_list, thus initialising the bio_list of new bios to be
1499 * added. __generic_make_request may indeed add some more bios
1500 * through a recursive call to generic_make_request. If it
1501 * did, we find a non-NULL value in bio_list and re-enter the loop
1502 * from the top. In this case we really did just take the bio
1503 * of the top of the list (no pretending) and so fixup bio_list and
1504 * bio_tail or bi_next, and call into __generic_make_request again.
1506 * The loop was structured like this to make only one call to
1507 * __generic_make_request (which is important as it is large and
1508 * inlined) and to keep the structure simple.
1510 BUG_ON(bio
->bi_next
);
1512 current
->bio_list
= bio
->bi_next
;
1513 if (bio
->bi_next
== NULL
)
1514 current
->bio_tail
= ¤t
->bio_list
;
1516 bio
->bi_next
= NULL
;
1517 __generic_make_request(bio
);
1518 bio
= current
->bio_list
;
1520 current
->bio_tail
= NULL
; /* deactivate */
1522 EXPORT_SYMBOL(generic_make_request
);
1525 * submit_bio - submit a bio to the block device layer for I/O
1526 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1527 * @bio: The &struct bio which describes the I/O
1529 * submit_bio() is very similar in purpose to generic_make_request(), and
1530 * uses that function to do most of the work. Both are fairly rough
1531 * interfaces; @bio must be presetup and ready for I/O.
1534 void submit_bio(int rw
, struct bio
*bio
)
1536 int count
= bio_sectors(bio
);
1541 * If it's a regular read/write or a barrier with data attached,
1542 * go through the normal accounting stuff before submission.
1544 if (bio_has_data(bio
)) {
1546 count_vm_events(PGPGOUT
, count
);
1548 task_io_account_read(bio
->bi_size
);
1549 count_vm_events(PGPGIN
, count
);
1552 if (unlikely(block_dump
)) {
1553 char b
[BDEVNAME_SIZE
];
1554 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s\n",
1555 current
->comm
, task_pid_nr(current
),
1556 (rw
& WRITE
) ? "WRITE" : "READ",
1557 (unsigned long long)bio
->bi_sector
,
1558 bdevname(bio
->bi_bdev
, b
));
1562 generic_make_request(bio
);
1564 EXPORT_SYMBOL(submit_bio
);
1567 * blk_rq_check_limits - Helper function to check a request for the queue limit
1569 * @rq: the request being checked
1572 * @rq may have been made based on weaker limitations of upper-level queues
1573 * in request stacking drivers, and it may violate the limitation of @q.
1574 * Since the block layer and the underlying device driver trust @rq
1575 * after it is inserted to @q, it should be checked against @q before
1576 * the insertion using this generic function.
1578 * This function should also be useful for request stacking drivers
1579 * in some cases below, so export this fuction.
1580 * Request stacking drivers like request-based dm may change the queue
1581 * limits while requests are in the queue (e.g. dm's table swapping).
1582 * Such request stacking drivers should check those requests agaist
1583 * the new queue limits again when they dispatch those requests,
1584 * although such checkings are also done against the old queue limits
1585 * when submitting requests.
1587 int blk_rq_check_limits(struct request_queue
*q
, struct request
*rq
)
1589 if (rq
->nr_sectors
> q
->max_sectors
||
1590 rq
->data_len
> q
->max_hw_sectors
<< 9) {
1591 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
1596 * queue's settings related to segment counting like q->bounce_pfn
1597 * may differ from that of other stacking queues.
1598 * Recalculate it to check the request correctly on this queue's
1601 blk_recalc_rq_segments(rq
);
1602 if (rq
->nr_phys_segments
> q
->max_phys_segments
||
1603 rq
->nr_phys_segments
> q
->max_hw_segments
) {
1604 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
1610 EXPORT_SYMBOL_GPL(blk_rq_check_limits
);
1613 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1614 * @q: the queue to submit the request
1615 * @rq: the request being queued
1617 int blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
1619 unsigned long flags
;
1621 if (blk_rq_check_limits(q
, rq
))
1624 #ifdef CONFIG_FAIL_MAKE_REQUEST
1625 if (rq
->rq_disk
&& rq
->rq_disk
->part0
.make_it_fail
&&
1626 should_fail(&fail_make_request
, blk_rq_bytes(rq
)))
1630 spin_lock_irqsave(q
->queue_lock
, flags
);
1633 * Submitting request must be dequeued before calling this function
1634 * because it will be linked to another request_queue
1636 BUG_ON(blk_queued_rq(rq
));
1638 drive_stat_acct(rq
, 1);
1639 __elv_add_request(q
, rq
, ELEVATOR_INSERT_BACK
, 0);
1641 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1645 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
1648 * blkdev_dequeue_request - dequeue request and start timeout timer
1649 * @req: request to dequeue
1651 * Dequeue @req and start timeout timer on it. This hands off the
1652 * request to the driver.
1654 * Block internal functions which don't want to start timer should
1655 * call elv_dequeue_request().
1657 void blkdev_dequeue_request(struct request
*req
)
1659 elv_dequeue_request(req
->q
, req
);
1662 * We are now handing the request to the hardware, add the
1667 EXPORT_SYMBOL(blkdev_dequeue_request
);
1669 static void blk_account_io_completion(struct request
*req
, unsigned int bytes
)
1671 if (!blk_do_io_stat(req
))
1674 if (blk_fs_request(req
)) {
1675 const int rw
= rq_data_dir(req
);
1676 struct hd_struct
*part
;
1679 cpu
= part_stat_lock();
1680 part
= disk_map_sector_rcu(req
->rq_disk
, req
->sector
);
1681 part_stat_add(cpu
, part
, sectors
[rw
], bytes
>> 9);
1686 static void blk_account_io_done(struct request
*req
)
1688 if (!blk_do_io_stat(req
))
1692 * Account IO completion. bar_rq isn't accounted as a normal
1693 * IO on queueing nor completion. Accounting the containing
1694 * request is enough.
1696 if (blk_fs_request(req
) && req
!= &req
->q
->bar_rq
) {
1697 unsigned long duration
= jiffies
- req
->start_time
;
1698 const int rw
= rq_data_dir(req
);
1699 struct hd_struct
*part
;
1702 cpu
= part_stat_lock();
1703 part
= disk_map_sector_rcu(req
->rq_disk
, req
->sector
);
1705 part_stat_inc(cpu
, part
, ios
[rw
]);
1706 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
1707 part_round_stats(cpu
, part
);
1708 part_dec_in_flight(part
);
1715 * __end_that_request_first - end I/O on a request
1716 * @req: the request being processed
1717 * @error: %0 for success, < %0 for error
1718 * @nr_bytes: number of bytes to complete
1721 * Ends I/O on a number of bytes attached to @req, and sets it up
1722 * for the next range of segments (if any) in the cluster.
1725 * %0 - we are done with this request, call end_that_request_last()
1726 * %1 - still buffers pending for this request
1728 static int __end_that_request_first(struct request
*req
, int error
,
1731 int total_bytes
, bio_nbytes
, next_idx
= 0;
1734 trace_block_rq_complete(req
->q
, req
);
1737 * For fs requests, rq is just carrier of independent bio's
1738 * and each partial completion should be handled separately.
1739 * Reset per-request error on each partial completion.
1741 * TODO: tj: This is too subtle. It would be better to let
1742 * low level drivers do what they see fit.
1744 if (blk_fs_request(req
))
1747 if (error
&& (blk_fs_request(req
) && !(req
->cmd_flags
& REQ_QUIET
))) {
1748 printk(KERN_ERR
"end_request: I/O error, dev %s, sector %llu\n",
1749 req
->rq_disk
? req
->rq_disk
->disk_name
: "?",
1750 (unsigned long long)req
->sector
);
1753 blk_account_io_completion(req
, nr_bytes
);
1755 total_bytes
= bio_nbytes
= 0;
1756 while ((bio
= req
->bio
) != NULL
) {
1759 if (nr_bytes
>= bio
->bi_size
) {
1760 req
->bio
= bio
->bi_next
;
1761 nbytes
= bio
->bi_size
;
1762 req_bio_endio(req
, bio
, nbytes
, error
);
1766 int idx
= bio
->bi_idx
+ next_idx
;
1768 if (unlikely(bio
->bi_idx
>= bio
->bi_vcnt
)) {
1769 blk_dump_rq_flags(req
, "__end_that");
1770 printk(KERN_ERR
"%s: bio idx %d >= vcnt %d\n",
1771 __func__
, bio
->bi_idx
, bio
->bi_vcnt
);
1775 nbytes
= bio_iovec_idx(bio
, idx
)->bv_len
;
1776 BIO_BUG_ON(nbytes
> bio
->bi_size
);
1779 * not a complete bvec done
1781 if (unlikely(nbytes
> nr_bytes
)) {
1782 bio_nbytes
+= nr_bytes
;
1783 total_bytes
+= nr_bytes
;
1788 * advance to the next vector
1791 bio_nbytes
+= nbytes
;
1794 total_bytes
+= nbytes
;
1800 * end more in this run, or just return 'not-done'
1802 if (unlikely(nr_bytes
<= 0))
1814 * if the request wasn't completed, update state
1817 req_bio_endio(req
, bio
, bio_nbytes
, error
);
1818 bio
->bi_idx
+= next_idx
;
1819 bio_iovec(bio
)->bv_offset
+= nr_bytes
;
1820 bio_iovec(bio
)->bv_len
-= nr_bytes
;
1823 blk_recalc_rq_sectors(req
, total_bytes
>> 9);
1824 blk_recalc_rq_segments(req
);
1829 * queue lock must be held
1831 static void end_that_request_last(struct request
*req
, int error
)
1833 if (blk_rq_tagged(req
))
1834 blk_queue_end_tag(req
->q
, req
);
1836 if (blk_queued_rq(req
))
1837 elv_dequeue_request(req
->q
, req
);
1839 if (unlikely(laptop_mode
) && blk_fs_request(req
))
1840 laptop_io_completion();
1842 blk_delete_timer(req
);
1844 blk_account_io_done(req
);
1847 req
->end_io(req
, error
);
1849 if (blk_bidi_rq(req
))
1850 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
1852 __blk_put_request(req
->q
, req
);
1857 * blk_rq_bytes - Returns bytes left to complete in the entire request
1858 * @rq: the request being processed
1860 unsigned int blk_rq_bytes(struct request
*rq
)
1862 if (blk_fs_request(rq
))
1863 return rq
->hard_nr_sectors
<< 9;
1865 return rq
->data_len
;
1867 EXPORT_SYMBOL_GPL(blk_rq_bytes
);
1870 * blk_rq_cur_bytes - Returns bytes left to complete in the current segment
1871 * @rq: the request being processed
1873 unsigned int blk_rq_cur_bytes(struct request
*rq
)
1875 if (blk_fs_request(rq
))
1876 return rq
->current_nr_sectors
<< 9;
1879 return rq
->bio
->bi_size
;
1881 return rq
->data_len
;
1883 EXPORT_SYMBOL_GPL(blk_rq_cur_bytes
);
1886 * end_request - end I/O on the current segment of the request
1887 * @req: the request being processed
1888 * @uptodate: error value or %0/%1 uptodate flag
1891 * Ends I/O on the current segment of a request. If that is the only
1892 * remaining segment, the request is also completed and freed.
1894 * This is a remnant of how older block drivers handled I/O completions.
1895 * Modern drivers typically end I/O on the full request in one go, unless
1896 * they have a residual value to account for. For that case this function
1897 * isn't really useful, unless the residual just happens to be the
1898 * full current segment. In other words, don't use this function in new
1899 * code. Use blk_end_request() or __blk_end_request() to end a request.
1901 void end_request(struct request
*req
, int uptodate
)
1906 error
= uptodate
? uptodate
: -EIO
;
1908 __blk_end_request(req
, error
, req
->hard_cur_sectors
<< 9);
1910 EXPORT_SYMBOL(end_request
);
1912 static int end_that_request_data(struct request
*rq
, int error
,
1913 unsigned int nr_bytes
, unsigned int bidi_bytes
)
1916 if (__end_that_request_first(rq
, error
, nr_bytes
))
1919 /* Bidi request must be completed as a whole */
1920 if (blk_bidi_rq(rq
) &&
1921 __end_that_request_first(rq
->next_rq
, error
, bidi_bytes
))
1929 * blk_end_io - Generic end_io function to complete a request.
1930 * @rq: the request being processed
1931 * @error: %0 for success, < %0 for error
1932 * @nr_bytes: number of bytes to complete @rq
1933 * @bidi_bytes: number of bytes to complete @rq->next_rq
1934 * @drv_callback: function called between completion of bios in the request
1935 * and completion of the request.
1936 * If the callback returns non %0, this helper returns without
1937 * completion of the request.
1940 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
1941 * If @rq has leftover, sets it up for the next range of segments.
1944 * %0 - we are done with this request
1945 * %1 - this request is not freed yet, it still has pending buffers.
1947 static int blk_end_io(struct request
*rq
, int error
, unsigned int nr_bytes
,
1948 unsigned int bidi_bytes
,
1949 int (drv_callback
)(struct request
*))
1951 struct request_queue
*q
= rq
->q
;
1952 unsigned long flags
= 0UL;
1954 if (end_that_request_data(rq
, error
, nr_bytes
, bidi_bytes
))
1957 /* Special feature for tricky drivers */
1958 if (drv_callback
&& drv_callback(rq
))
1961 add_disk_randomness(rq
->rq_disk
);
1963 spin_lock_irqsave(q
->queue_lock
, flags
);
1964 end_that_request_last(rq
, error
);
1965 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1971 * blk_end_request - Helper function for drivers to complete the request.
1972 * @rq: the request being processed
1973 * @error: %0 for success, < %0 for error
1974 * @nr_bytes: number of bytes to complete
1977 * Ends I/O on a number of bytes attached to @rq.
1978 * If @rq has leftover, sets it up for the next range of segments.
1981 * %0 - we are done with this request
1982 * %1 - still buffers pending for this request
1984 int blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
1986 return blk_end_io(rq
, error
, nr_bytes
, 0, NULL
);
1988 EXPORT_SYMBOL_GPL(blk_end_request
);
1991 * __blk_end_request - Helper function for drivers to complete the request.
1992 * @rq: the request being processed
1993 * @error: %0 for success, < %0 for error
1994 * @nr_bytes: number of bytes to complete
1997 * Must be called with queue lock held unlike blk_end_request().
2000 * %0 - we are done with this request
2001 * %1 - still buffers pending for this request
2003 int __blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2005 if (rq
->bio
&& __end_that_request_first(rq
, error
, nr_bytes
))
2008 add_disk_randomness(rq
->rq_disk
);
2010 end_that_request_last(rq
, error
);
2014 EXPORT_SYMBOL_GPL(__blk_end_request
);
2017 * blk_end_bidi_request - Helper function for drivers to complete bidi request.
2018 * @rq: the bidi request being processed
2019 * @error: %0 for success, < %0 for error
2020 * @nr_bytes: number of bytes to complete @rq
2021 * @bidi_bytes: number of bytes to complete @rq->next_rq
2024 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2027 * %0 - we are done with this request
2028 * %1 - still buffers pending for this request
2030 int blk_end_bidi_request(struct request
*rq
, int error
, unsigned int nr_bytes
,
2031 unsigned int bidi_bytes
)
2033 return blk_end_io(rq
, error
, nr_bytes
, bidi_bytes
, NULL
);
2035 EXPORT_SYMBOL_GPL(blk_end_bidi_request
);
2038 * blk_update_request - Special helper function for request stacking drivers
2039 * @rq: the request being processed
2040 * @error: %0 for success, < %0 for error
2041 * @nr_bytes: number of bytes to complete @rq
2044 * Ends I/O on a number of bytes attached to @rq, but doesn't complete
2045 * the request structure even if @rq doesn't have leftover.
2046 * If @rq has leftover, sets it up for the next range of segments.
2048 * This special helper function is only for request stacking drivers
2049 * (e.g. request-based dm) so that they can handle partial completion.
2050 * Actual device drivers should use blk_end_request instead.
2052 void blk_update_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2054 if (!end_that_request_data(rq
, error
, nr_bytes
, 0)) {
2056 * These members are not updated in end_that_request_data()
2057 * when all bios are completed.
2058 * Update them so that the request stacking driver can find
2059 * how many bytes remain in the request later.
2061 rq
->nr_sectors
= rq
->hard_nr_sectors
= 0;
2062 rq
->current_nr_sectors
= rq
->hard_cur_sectors
= 0;
2065 EXPORT_SYMBOL_GPL(blk_update_request
);
2068 * blk_end_request_callback - Special helper function for tricky drivers
2069 * @rq: the request being processed
2070 * @error: %0 for success, < %0 for error
2071 * @nr_bytes: number of bytes to complete
2072 * @drv_callback: function called between completion of bios in the request
2073 * and completion of the request.
2074 * If the callback returns non %0, this helper returns without
2075 * completion of the request.
2078 * Ends I/O on a number of bytes attached to @rq.
2079 * If @rq has leftover, sets it up for the next range of segments.
2081 * This special helper function is used only for existing tricky drivers.
2082 * (e.g. cdrom_newpc_intr() of ide-cd)
2083 * This interface will be removed when such drivers are rewritten.
2084 * Don't use this interface in other places anymore.
2087 * %0 - we are done with this request
2088 * %1 - this request is not freed yet.
2089 * this request still has pending buffers or
2090 * the driver doesn't want to finish this request yet.
2092 int blk_end_request_callback(struct request
*rq
, int error
,
2093 unsigned int nr_bytes
,
2094 int (drv_callback
)(struct request
*))
2096 return blk_end_io(rq
, error
, nr_bytes
, 0, drv_callback
);
2098 EXPORT_SYMBOL_GPL(blk_end_request_callback
);
2100 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
2103 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw, and
2104 we want BIO_RW_AHEAD (bit 1) to imply REQ_FAILFAST (bit 1). */
2105 rq
->cmd_flags
|= (bio
->bi_rw
& 3);
2107 if (bio_has_data(bio
)) {
2108 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2109 rq
->buffer
= bio_data(bio
);
2111 rq
->current_nr_sectors
= bio_cur_sectors(bio
);
2112 rq
->hard_cur_sectors
= rq
->current_nr_sectors
;
2113 rq
->hard_nr_sectors
= rq
->nr_sectors
= bio_sectors(bio
);
2114 rq
->data_len
= bio
->bi_size
;
2116 rq
->bio
= rq
->biotail
= bio
;
2119 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2123 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2124 * @q : the queue of the device being checked
2127 * Check if underlying low-level drivers of a device are busy.
2128 * If the drivers want to export their busy state, they must set own
2129 * exporting function using blk_queue_lld_busy() first.
2131 * Basically, this function is used only by request stacking drivers
2132 * to stop dispatching requests to underlying devices when underlying
2133 * devices are busy. This behavior helps more I/O merging on the queue
2134 * of the request stacking driver and prevents I/O throughput regression
2135 * on burst I/O load.
2138 * 0 - Not busy (The request stacking driver should dispatch request)
2139 * 1 - Busy (The request stacking driver should stop dispatching request)
2141 int blk_lld_busy(struct request_queue
*q
)
2144 return q
->lld_busy_fn(q
);
2148 EXPORT_SYMBOL_GPL(blk_lld_busy
);
2150 int kblockd_schedule_work(struct request_queue
*q
, struct work_struct
*work
)
2152 return queue_work(kblockd_workqueue
, work
);
2154 EXPORT_SYMBOL(kblockd_schedule_work
);
2156 int __init
blk_dev_init(void)
2158 kblockd_workqueue
= create_workqueue("kblockd");
2159 if (!kblockd_workqueue
)
2160 panic("Failed to create kblockd\n");
2162 request_cachep
= kmem_cache_create("blkdev_requests",
2163 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
2165 blk_requestq_cachep
= kmem_cache_create("blkdev_queue",
2166 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);