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
);
113 INIT_LIST_HEAD(&rq
->timeout_list
);
116 rq
->sector
= rq
->hard_sector
= (sector_t
) -1;
117 INIT_HLIST_NODE(&rq
->hash
);
118 RB_CLEAR_NODE(&rq
->rb_node
);
123 EXPORT_SYMBOL(blk_rq_init
);
125 static void req_bio_endio(struct request
*rq
, struct bio
*bio
,
126 unsigned int nbytes
, int error
)
128 struct request_queue
*q
= rq
->q
;
130 if (&q
->bar_rq
!= rq
) {
132 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
133 else if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
136 if (unlikely(nbytes
> bio
->bi_size
)) {
137 printk(KERN_ERR
"%s: want %u bytes done, %u left\n",
138 __func__
, nbytes
, bio
->bi_size
);
139 nbytes
= bio
->bi_size
;
142 bio
->bi_size
-= nbytes
;
143 bio
->bi_sector
+= (nbytes
>> 9);
145 if (bio_integrity(bio
))
146 bio_integrity_advance(bio
, nbytes
);
148 if (bio
->bi_size
== 0)
149 bio_endio(bio
, error
);
153 * Okay, this is the barrier request in progress, just
156 if (error
&& !q
->orderr
)
161 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
165 printk(KERN_INFO
"%s: dev %s: type=%x, flags=%x\n", msg
,
166 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->cmd_type
,
169 printk(KERN_INFO
" sector %llu, nr/cnr %lu/%u\n",
170 (unsigned long long)rq
->sector
,
172 rq
->current_nr_sectors
);
173 printk(KERN_INFO
" bio %p, biotail %p, buffer %p, data %p, len %u\n",
174 rq
->bio
, rq
->biotail
,
175 rq
->buffer
, rq
->data
,
178 if (blk_pc_request(rq
)) {
179 printk(KERN_INFO
" cdb: ");
180 for (bit
= 0; bit
< BLK_MAX_CDB
; bit
++)
181 printk("%02x ", rq
->cmd
[bit
]);
185 EXPORT_SYMBOL(blk_dump_rq_flags
);
188 * "plug" the device if there are no outstanding requests: this will
189 * force the transfer to start only after we have put all the requests
192 * This is called with interrupts off and no requests on the queue and
193 * with the queue lock held.
195 void blk_plug_device(struct request_queue
*q
)
197 WARN_ON(!irqs_disabled());
200 * don't plug a stopped queue, it must be paired with blk_start_queue()
201 * which will restart the queueing
203 if (blk_queue_stopped(q
))
206 if (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED
, q
)) {
207 mod_timer(&q
->unplug_timer
, jiffies
+ q
->unplug_delay
);
208 blk_add_trace_generic(q
, NULL
, 0, BLK_TA_PLUG
);
211 EXPORT_SYMBOL(blk_plug_device
);
214 * blk_plug_device_unlocked - plug a device without queue lock held
215 * @q: The &struct request_queue to plug
218 * Like @blk_plug_device(), but grabs the queue lock and disables
221 void blk_plug_device_unlocked(struct request_queue
*q
)
225 spin_lock_irqsave(q
->queue_lock
, flags
);
227 spin_unlock_irqrestore(q
->queue_lock
, flags
);
229 EXPORT_SYMBOL(blk_plug_device_unlocked
);
232 * remove the queue from the plugged list, if present. called with
233 * queue lock held and interrupts disabled.
235 int blk_remove_plug(struct request_queue
*q
)
237 WARN_ON(!irqs_disabled());
239 if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED
, q
))
242 del_timer(&q
->unplug_timer
);
245 EXPORT_SYMBOL(blk_remove_plug
);
248 * remove the plug and let it rip..
250 void __generic_unplug_device(struct request_queue
*q
)
252 if (unlikely(blk_queue_stopped(q
)))
255 if (!blk_remove_plug(q
))
260 EXPORT_SYMBOL(__generic_unplug_device
);
263 * generic_unplug_device - fire a request queue
264 * @q: The &struct request_queue in question
267 * Linux uses plugging to build bigger requests queues before letting
268 * the device have at them. If a queue is plugged, the I/O scheduler
269 * is still adding and merging requests on the queue. Once the queue
270 * gets unplugged, the request_fn defined for the queue is invoked and
273 void generic_unplug_device(struct request_queue
*q
)
275 if (blk_queue_plugged(q
)) {
276 spin_lock_irq(q
->queue_lock
);
277 __generic_unplug_device(q
);
278 spin_unlock_irq(q
->queue_lock
);
281 EXPORT_SYMBOL(generic_unplug_device
);
283 static void blk_backing_dev_unplug(struct backing_dev_info
*bdi
,
286 struct request_queue
*q
= bdi
->unplug_io_data
;
291 void blk_unplug_work(struct work_struct
*work
)
293 struct request_queue
*q
=
294 container_of(work
, struct request_queue
, unplug_work
);
296 blk_add_trace_pdu_int(q
, BLK_TA_UNPLUG_IO
, NULL
,
297 q
->rq
.count
[READ
] + q
->rq
.count
[WRITE
]);
302 void blk_unplug_timeout(unsigned long data
)
304 struct request_queue
*q
= (struct request_queue
*)data
;
306 blk_add_trace_pdu_int(q
, BLK_TA_UNPLUG_TIMER
, NULL
,
307 q
->rq
.count
[READ
] + q
->rq
.count
[WRITE
]);
309 kblockd_schedule_work(q
, &q
->unplug_work
);
312 void blk_unplug(struct request_queue
*q
)
315 * devices don't necessarily have an ->unplug_fn defined
318 blk_add_trace_pdu_int(q
, BLK_TA_UNPLUG_IO
, NULL
,
319 q
->rq
.count
[READ
] + q
->rq
.count
[WRITE
]);
324 EXPORT_SYMBOL(blk_unplug
);
326 static void blk_invoke_request_fn(struct request_queue
*q
)
328 if (unlikely(blk_queue_stopped(q
)))
332 * one level of recursion is ok and is much faster than kicking
333 * the unplug handling
335 if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER
, q
)) {
337 queue_flag_clear(QUEUE_FLAG_REENTER
, q
);
339 queue_flag_set(QUEUE_FLAG_PLUGGED
, q
);
340 kblockd_schedule_work(q
, &q
->unplug_work
);
345 * blk_start_queue - restart a previously stopped queue
346 * @q: The &struct request_queue in question
349 * blk_start_queue() will clear the stop flag on the queue, and call
350 * the request_fn for the queue if it was in a stopped state when
351 * entered. Also see blk_stop_queue(). Queue lock must be held.
353 void blk_start_queue(struct request_queue
*q
)
355 WARN_ON(!irqs_disabled());
357 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
358 blk_invoke_request_fn(q
);
360 EXPORT_SYMBOL(blk_start_queue
);
363 * blk_stop_queue - stop a queue
364 * @q: The &struct request_queue in question
367 * The Linux block layer assumes that a block driver will consume all
368 * entries on the request queue when the request_fn strategy is called.
369 * Often this will not happen, because of hardware limitations (queue
370 * depth settings). If a device driver gets a 'queue full' response,
371 * or if it simply chooses not to queue more I/O at one point, it can
372 * call this function to prevent the request_fn from being called until
373 * the driver has signalled it's ready to go again. This happens by calling
374 * blk_start_queue() to restart queue operations. Queue lock must be held.
376 void blk_stop_queue(struct request_queue
*q
)
379 queue_flag_set(QUEUE_FLAG_STOPPED
, q
);
381 EXPORT_SYMBOL(blk_stop_queue
);
384 * blk_sync_queue - cancel any pending callbacks on a queue
388 * The block layer may perform asynchronous callback activity
389 * on a queue, such as calling the unplug function after a timeout.
390 * A block device may call blk_sync_queue to ensure that any
391 * such activity is cancelled, thus allowing it to release resources
392 * that the callbacks might use. The caller must already have made sure
393 * that its ->make_request_fn will not re-add plugging prior to calling
397 void blk_sync_queue(struct request_queue
*q
)
399 del_timer_sync(&q
->unplug_timer
);
400 kblockd_flush_work(&q
->unplug_work
);
402 EXPORT_SYMBOL(blk_sync_queue
);
405 * __blk_run_queue - run a single device queue
406 * @q: The queue to run
409 * See @blk_run_queue. This variant must be called with the queue lock
410 * held and interrupts disabled.
413 void __blk_run_queue(struct request_queue
*q
)
418 * Only recurse once to avoid overrunning the stack, let the unplug
419 * handling reinvoke the handler shortly if we already got there.
421 if (!elv_queue_empty(q
))
422 blk_invoke_request_fn(q
);
424 EXPORT_SYMBOL(__blk_run_queue
);
427 * blk_run_queue - run a single device queue
428 * @q: The queue to run
431 * Invoke request handling on this queue, if it has pending work to do.
432 * May be used to restart queueing when a request has completed. Also
433 * See @blk_start_queueing.
436 void blk_run_queue(struct request_queue
*q
)
440 spin_lock_irqsave(q
->queue_lock
, flags
);
442 spin_unlock_irqrestore(q
->queue_lock
, flags
);
444 EXPORT_SYMBOL(blk_run_queue
);
446 void blk_put_queue(struct request_queue
*q
)
448 kobject_put(&q
->kobj
);
451 void blk_cleanup_queue(struct request_queue
*q
)
454 * We know we have process context here, so we can be a little
455 * cautious and ensure that pending block actions on this device
456 * are done before moving on. Going into this function, we should
457 * not have processes doing IO to this device.
461 mutex_lock(&q
->sysfs_lock
);
462 queue_flag_set_unlocked(QUEUE_FLAG_DEAD
, q
);
463 mutex_unlock(&q
->sysfs_lock
);
466 elevator_exit(q
->elevator
);
470 EXPORT_SYMBOL(blk_cleanup_queue
);
472 static int blk_init_free_list(struct request_queue
*q
)
474 struct request_list
*rl
= &q
->rq
;
476 rl
->count
[READ
] = rl
->count
[WRITE
] = 0;
477 rl
->starved
[READ
] = rl
->starved
[WRITE
] = 0;
479 init_waitqueue_head(&rl
->wait
[READ
]);
480 init_waitqueue_head(&rl
->wait
[WRITE
]);
482 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, mempool_alloc_slab
,
483 mempool_free_slab
, request_cachep
, q
->node
);
491 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
493 return blk_alloc_queue_node(gfp_mask
, -1);
495 EXPORT_SYMBOL(blk_alloc_queue
);
497 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
499 struct request_queue
*q
;
502 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
503 gfp_mask
| __GFP_ZERO
, node_id
);
507 q
->backing_dev_info
.unplug_io_fn
= blk_backing_dev_unplug
;
508 q
->backing_dev_info
.unplug_io_data
= q
;
509 err
= bdi_init(&q
->backing_dev_info
);
511 kmem_cache_free(blk_requestq_cachep
, q
);
515 init_timer(&q
->unplug_timer
);
516 setup_timer(&q
->timeout
, blk_rq_timed_out_timer
, (unsigned long) q
);
517 INIT_LIST_HEAD(&q
->timeout_list
);
519 kobject_init(&q
->kobj
, &blk_queue_ktype
);
521 mutex_init(&q
->sysfs_lock
);
522 spin_lock_init(&q
->__queue_lock
);
526 EXPORT_SYMBOL(blk_alloc_queue_node
);
529 * blk_init_queue - prepare a request queue for use with a block device
530 * @rfn: The function to be called to process requests that have been
531 * placed on the queue.
532 * @lock: Request queue spin lock
535 * If a block device wishes to use the standard request handling procedures,
536 * which sorts requests and coalesces adjacent requests, then it must
537 * call blk_init_queue(). The function @rfn will be called when there
538 * are requests on the queue that need to be processed. If the device
539 * supports plugging, then @rfn may not be called immediately when requests
540 * are available on the queue, but may be called at some time later instead.
541 * Plugged queues are generally unplugged when a buffer belonging to one
542 * of the requests on the queue is needed, or due to memory pressure.
544 * @rfn is not required, or even expected, to remove all requests off the
545 * queue, but only as many as it can handle at a time. If it does leave
546 * requests on the queue, it is responsible for arranging that the requests
547 * get dealt with eventually.
549 * The queue spin lock must be held while manipulating the requests on the
550 * request queue; this lock will be taken also from interrupt context, so irq
551 * disabling is needed for it.
553 * Function returns a pointer to the initialized request queue, or %NULL if
557 * blk_init_queue() must be paired with a blk_cleanup_queue() call
558 * when the block device is deactivated (such as at module unload).
561 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
563 return blk_init_queue_node(rfn
, lock
, -1);
565 EXPORT_SYMBOL(blk_init_queue
);
567 struct request_queue
*
568 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
570 struct request_queue
*q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
576 if (blk_init_free_list(q
)) {
577 kmem_cache_free(blk_requestq_cachep
, q
);
582 * if caller didn't supply a lock, they get per-queue locking with
586 lock
= &q
->__queue_lock
;
589 q
->prep_rq_fn
= NULL
;
590 q
->unplug_fn
= generic_unplug_device
;
591 q
->queue_flags
= (1 << QUEUE_FLAG_CLUSTER
|
592 1 << QUEUE_FLAG_STACKABLE
);
593 q
->queue_lock
= lock
;
595 blk_queue_segment_boundary(q
, 0xffffffff);
597 blk_queue_make_request(q
, __make_request
);
598 blk_queue_max_segment_size(q
, MAX_SEGMENT_SIZE
);
600 blk_queue_max_hw_segments(q
, MAX_HW_SEGMENTS
);
601 blk_queue_max_phys_segments(q
, MAX_PHYS_SEGMENTS
);
603 q
->sg_reserved_size
= INT_MAX
;
605 blk_set_cmd_filter_defaults(&q
->cmd_filter
);
610 if (!elevator_init(q
, NULL
)) {
611 blk_queue_congestion_threshold(q
);
618 EXPORT_SYMBOL(blk_init_queue_node
);
620 int blk_get_queue(struct request_queue
*q
)
622 if (likely(!test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
))) {
623 kobject_get(&q
->kobj
);
630 static inline void blk_free_request(struct request_queue
*q
, struct request
*rq
)
632 if (rq
->cmd_flags
& REQ_ELVPRIV
)
633 elv_put_request(q
, rq
);
634 mempool_free(rq
, q
->rq
.rq_pool
);
637 static struct request
*
638 blk_alloc_request(struct request_queue
*q
, int rw
, int priv
, gfp_t gfp_mask
)
640 struct request
*rq
= mempool_alloc(q
->rq
.rq_pool
, gfp_mask
);
647 rq
->cmd_flags
= rw
| REQ_ALLOCED
;
650 if (unlikely(elv_set_request(q
, rq
, gfp_mask
))) {
651 mempool_free(rq
, q
->rq
.rq_pool
);
654 rq
->cmd_flags
|= REQ_ELVPRIV
;
661 * ioc_batching returns true if the ioc is a valid batching request and
662 * should be given priority access to a request.
664 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
670 * Make sure the process is able to allocate at least 1 request
671 * even if the batch times out, otherwise we could theoretically
674 return ioc
->nr_batch_requests
== q
->nr_batching
||
675 (ioc
->nr_batch_requests
> 0
676 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
680 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
681 * will cause the process to be a "batcher" on all queues in the system. This
682 * is the behaviour we want though - once it gets a wakeup it should be given
685 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
687 if (!ioc
|| ioc_batching(q
, ioc
))
690 ioc
->nr_batch_requests
= q
->nr_batching
;
691 ioc
->last_waited
= jiffies
;
694 static void __freed_request(struct request_queue
*q
, int rw
)
696 struct request_list
*rl
= &q
->rq
;
698 if (rl
->count
[rw
] < queue_congestion_off_threshold(q
))
699 blk_clear_queue_congested(q
, rw
);
701 if (rl
->count
[rw
] + 1 <= q
->nr_requests
) {
702 if (waitqueue_active(&rl
->wait
[rw
]))
703 wake_up(&rl
->wait
[rw
]);
705 blk_clear_queue_full(q
, rw
);
710 * A request has just been released. Account for it, update the full and
711 * congestion status, wake up any waiters. Called under q->queue_lock.
713 static void freed_request(struct request_queue
*q
, int rw
, int priv
)
715 struct request_list
*rl
= &q
->rq
;
721 __freed_request(q
, rw
);
723 if (unlikely(rl
->starved
[rw
^ 1]))
724 __freed_request(q
, rw
^ 1);
727 #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
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 int rw
= rw_flags
& 0x01;
742 may_queue
= elv_may_queue(q
, rw_flags
);
743 if (may_queue
== ELV_MQUEUE_NO
)
746 if (rl
->count
[rw
]+1 >= queue_congestion_on_threshold(q
)) {
747 if (rl
->count
[rw
]+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
, rw
)) {
756 ioc_set_batching(q
, ioc
);
757 blk_set_queue_full(q
, rw
);
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
, rw
);
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
[rw
] >= (3 * q
->nr_requests
/ 2))
784 priv
= !test_bit(QUEUE_FLAG_ELVSWITCH
, &q
->queue_flags
);
788 spin_unlock_irq(q
->queue_lock
);
790 rq
= blk_alloc_request(q
, rw_flags
, priv
, gfp_mask
);
793 * Allocation failed presumably due to memory. Undo anything
794 * we might have messed up.
796 * Allocating task should really be put onto the front of the
797 * wait queue, but this is pretty rare.
799 spin_lock_irq(q
->queue_lock
);
800 freed_request(q
, rw
, priv
);
803 * in the very unlikely event that allocation failed and no
804 * requests for this direction was pending, mark us starved
805 * so that freeing of a request in the other direction will
806 * notice us. another possible fix would be to split the
807 * rq mempool into READ and WRITE
810 if (unlikely(rl
->count
[rw
] == 0))
817 * ioc may be NULL here, and ioc_batching will be false. That's
818 * OK, if the queue is under the request limit then requests need
819 * not count toward the nr_batch_requests limit. There will always
820 * be some limit enforced by BLK_BATCH_TIME.
822 if (ioc_batching(q
, ioc
))
823 ioc
->nr_batch_requests
--;
825 blk_add_trace_generic(q
, bio
, rw
, BLK_TA_GETRQ
);
831 * No available requests for this queue, unplug the device and wait for some
832 * requests to become available.
834 * Called with q->queue_lock held, and returns with it unlocked.
836 static struct request
*get_request_wait(struct request_queue
*q
, int rw_flags
,
839 const int rw
= rw_flags
& 0x01;
842 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
845 struct io_context
*ioc
;
846 struct request_list
*rl
= &q
->rq
;
848 prepare_to_wait_exclusive(&rl
->wait
[rw
], &wait
,
849 TASK_UNINTERRUPTIBLE
);
851 blk_add_trace_generic(q
, bio
, rw
, BLK_TA_SLEEPRQ
);
853 __generic_unplug_device(q
);
854 spin_unlock_irq(q
->queue_lock
);
858 * After sleeping, we become a "batching" process and
859 * will be able to allocate at least one request, and
860 * up to a big batch of them for a small period time.
861 * See ioc_batching, ioc_set_batching
863 ioc
= current_io_context(GFP_NOIO
, q
->node
);
864 ioc_set_batching(q
, ioc
);
866 spin_lock_irq(q
->queue_lock
);
867 finish_wait(&rl
->wait
[rw
], &wait
);
869 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
875 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
879 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
881 spin_lock_irq(q
->queue_lock
);
882 if (gfp_mask
& __GFP_WAIT
) {
883 rq
= get_request_wait(q
, rw
, NULL
);
885 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
887 spin_unlock_irq(q
->queue_lock
);
889 /* q->queue_lock is unlocked at this point */
893 EXPORT_SYMBOL(blk_get_request
);
896 * blk_start_queueing - initiate dispatch of requests to device
897 * @q: request queue to kick into gear
899 * This is basically a helper to remove the need to know whether a queue
900 * is plugged or not if someone just wants to initiate dispatch of requests
901 * for this queue. Should be used to start queueing on a device outside
902 * of ->request_fn() context. Also see @blk_run_queue.
904 * The queue lock must be held with interrupts disabled.
906 void blk_start_queueing(struct request_queue
*q
)
908 if (!blk_queue_plugged(q
)) {
909 if (unlikely(blk_queue_stopped(q
)))
913 __generic_unplug_device(q
);
915 EXPORT_SYMBOL(blk_start_queueing
);
918 * blk_requeue_request - put a request back on queue
919 * @q: request queue where request should be inserted
920 * @rq: request to be inserted
923 * Drivers often keep queueing requests until the hardware cannot accept
924 * more, when that condition happens we need to put the request back
925 * on the queue. Must be called with queue lock held.
927 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
929 blk_delete_timer(rq
);
930 blk_clear_rq_complete(rq
);
931 blk_add_trace_rq(q
, rq
, BLK_TA_REQUEUE
);
933 if (blk_rq_tagged(rq
))
934 blk_queue_end_tag(q
, rq
);
936 elv_requeue_request(q
, rq
);
938 EXPORT_SYMBOL(blk_requeue_request
);
941 * blk_insert_request - insert a special request into a request queue
942 * @q: request queue where request should be inserted
943 * @rq: request to be inserted
944 * @at_head: insert request at head or tail of queue
945 * @data: private data
948 * Many block devices need to execute commands asynchronously, so they don't
949 * block the whole kernel from preemption during request execution. This is
950 * accomplished normally by inserting aritficial requests tagged as
951 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
952 * be scheduled for actual execution by the request queue.
954 * We have the option of inserting the head or the tail of the queue.
955 * Typically we use the tail for new ioctls and so forth. We use the head
956 * of the queue for things like a QUEUE_FULL message from a device, or a
957 * host that is unable to accept a particular command.
959 void blk_insert_request(struct request_queue
*q
, struct request
*rq
,
960 int at_head
, void *data
)
962 int where
= at_head
? ELEVATOR_INSERT_FRONT
: ELEVATOR_INSERT_BACK
;
966 * tell I/O scheduler that this isn't a regular read/write (ie it
967 * must not attempt merges on this) and that it acts as a soft
970 rq
->cmd_type
= REQ_TYPE_SPECIAL
;
971 rq
->cmd_flags
|= REQ_SOFTBARRIER
;
975 spin_lock_irqsave(q
->queue_lock
, flags
);
978 * If command is tagged, release the tag
980 if (blk_rq_tagged(rq
))
981 blk_queue_end_tag(q
, rq
);
983 drive_stat_acct(rq
, 1);
984 __elv_add_request(q
, rq
, where
, 0);
985 blk_start_queueing(q
);
986 spin_unlock_irqrestore(q
->queue_lock
, flags
);
988 EXPORT_SYMBOL(blk_insert_request
);
991 * add-request adds a request to the linked list.
992 * queue lock is held and interrupts disabled, as we muck with the
993 * request queue list.
995 static inline void add_request(struct request_queue
*q
, struct request
*req
)
997 drive_stat_acct(req
, 1);
1000 * elevator indicated where it wants this request to be
1001 * inserted at elevator_merge time
1003 __elv_add_request(q
, req
, ELEVATOR_INSERT_SORT
, 0);
1006 static void part_round_stats_single(int cpu
, struct hd_struct
*part
,
1009 if (now
== part
->stamp
)
1012 if (part
->in_flight
) {
1013 __part_stat_add(cpu
, part
, time_in_queue
,
1014 part
->in_flight
* (now
- part
->stamp
));
1015 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
1021 * part_round_stats() - Round off the performance stats on a struct
1024 * The average IO queue length and utilisation statistics are maintained
1025 * by observing the current state of the queue length and the amount of
1026 * time it has been in this state for.
1028 * Normally, that accounting is done on IO completion, but that can result
1029 * in more than a second's worth of IO being accounted for within any one
1030 * second, leading to >100% utilisation. To deal with that, we call this
1031 * function to do a round-off before returning the results when reading
1032 * /proc/diskstats. This accounts immediately for all queue usage up to
1033 * the current jiffies and restarts the counters again.
1035 void part_round_stats(int cpu
, struct hd_struct
*part
)
1037 unsigned long now
= jiffies
;
1040 part_round_stats_single(cpu
, &part_to_disk(part
)->part0
, now
);
1041 part_round_stats_single(cpu
, part
, now
);
1043 EXPORT_SYMBOL_GPL(part_round_stats
);
1046 * queue lock must be held
1048 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1052 if (unlikely(--req
->ref_count
))
1055 elv_completed_request(q
, req
);
1058 * Request may not have originated from ll_rw_blk. if not,
1059 * it didn't come out of our reserved rq pools
1061 if (req
->cmd_flags
& REQ_ALLOCED
) {
1062 int rw
= rq_data_dir(req
);
1063 int priv
= req
->cmd_flags
& REQ_ELVPRIV
;
1065 BUG_ON(!list_empty(&req
->queuelist
));
1066 BUG_ON(!hlist_unhashed(&req
->hash
));
1068 blk_free_request(q
, req
);
1069 freed_request(q
, rw
, priv
);
1072 EXPORT_SYMBOL_GPL(__blk_put_request
);
1074 void blk_put_request(struct request
*req
)
1076 unsigned long flags
;
1077 struct request_queue
*q
= req
->q
;
1079 spin_lock_irqsave(q
->queue_lock
, flags
);
1080 __blk_put_request(q
, req
);
1081 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1083 EXPORT_SYMBOL(blk_put_request
);
1085 void init_request_from_bio(struct request
*req
, struct bio
*bio
)
1087 req
->cpu
= bio
->bi_comp_cpu
;
1088 req
->cmd_type
= REQ_TYPE_FS
;
1091 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
1093 if (bio_rw_ahead(bio
) || bio_failfast(bio
))
1094 req
->cmd_flags
|= REQ_FAILFAST
;
1097 * REQ_BARRIER implies no merging, but lets make it explicit
1099 if (unlikely(bio_discard(bio
))) {
1100 req
->cmd_flags
|= REQ_DISCARD
;
1101 if (bio_barrier(bio
))
1102 req
->cmd_flags
|= REQ_SOFTBARRIER
;
1103 req
->q
->prepare_discard_fn(req
->q
, req
);
1104 } else if (unlikely(bio_barrier(bio
)))
1105 req
->cmd_flags
|= (REQ_HARDBARRIER
| REQ_NOMERGE
);
1108 req
->cmd_flags
|= REQ_RW_SYNC
;
1109 if (bio_rw_meta(bio
))
1110 req
->cmd_flags
|= REQ_RW_META
;
1113 req
->hard_sector
= req
->sector
= bio
->bi_sector
;
1114 req
->ioprio
= bio_prio(bio
);
1115 req
->start_time
= jiffies
;
1116 blk_rq_bio_prep(req
->q
, req
, bio
);
1119 static int __make_request(struct request_queue
*q
, struct bio
*bio
)
1121 struct request
*req
;
1122 int el_ret
, nr_sectors
, barrier
, discard
, err
;
1123 const unsigned short prio
= bio_prio(bio
);
1124 const int sync
= bio_sync(bio
);
1127 nr_sectors
= bio_sectors(bio
);
1130 * low level driver can indicate that it wants pages above a
1131 * certain limit bounced to low memory (ie for highmem, or even
1132 * ISA dma in theory)
1134 blk_queue_bounce(q
, &bio
);
1136 barrier
= bio_barrier(bio
);
1137 if (unlikely(barrier
) && bio_has_data(bio
) &&
1138 (q
->next_ordered
== QUEUE_ORDERED_NONE
)) {
1143 discard
= bio_discard(bio
);
1144 if (unlikely(discard
) && !q
->prepare_discard_fn
) {
1149 spin_lock_irq(q
->queue_lock
);
1151 if (unlikely(barrier
) || elv_queue_empty(q
))
1154 el_ret
= elv_merge(q
, &req
, bio
);
1156 case ELEVATOR_BACK_MERGE
:
1157 BUG_ON(!rq_mergeable(req
));
1159 if (!ll_back_merge_fn(q
, req
, bio
))
1162 blk_add_trace_bio(q
, bio
, BLK_TA_BACKMERGE
);
1164 req
->biotail
->bi_next
= bio
;
1166 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
1167 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
1168 if (!blk_rq_cpu_valid(req
))
1169 req
->cpu
= bio
->bi_comp_cpu
;
1170 drive_stat_acct(req
, 0);
1171 if (!attempt_back_merge(q
, req
))
1172 elv_merged_request(q
, req
, el_ret
);
1175 case ELEVATOR_FRONT_MERGE
:
1176 BUG_ON(!rq_mergeable(req
));
1178 if (!ll_front_merge_fn(q
, req
, bio
))
1181 blk_add_trace_bio(q
, bio
, BLK_TA_FRONTMERGE
);
1183 bio
->bi_next
= req
->bio
;
1187 * may not be valid. if the low level driver said
1188 * it didn't need a bounce buffer then it better
1189 * not touch req->buffer either...
1191 req
->buffer
= bio_data(bio
);
1192 req
->current_nr_sectors
= bio_cur_sectors(bio
);
1193 req
->hard_cur_sectors
= req
->current_nr_sectors
;
1194 req
->sector
= req
->hard_sector
= bio
->bi_sector
;
1195 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
1196 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
1197 if (!blk_rq_cpu_valid(req
))
1198 req
->cpu
= bio
->bi_comp_cpu
;
1199 drive_stat_acct(req
, 0);
1200 if (!attempt_front_merge(q
, req
))
1201 elv_merged_request(q
, req
, el_ret
);
1204 /* ELV_NO_MERGE: elevator says don't/can't merge. */
1211 * This sync check and mask will be re-done in init_request_from_bio(),
1212 * but we need to set it earlier to expose the sync flag to the
1213 * rq allocator and io schedulers.
1215 rw_flags
= bio_data_dir(bio
);
1217 rw_flags
|= REQ_RW_SYNC
;
1220 * Grab a free request. This is might sleep but can not fail.
1221 * Returns with the queue unlocked.
1223 req
= get_request_wait(q
, rw_flags
, bio
);
1226 * After dropping the lock and possibly sleeping here, our request
1227 * may now be mergeable after it had proven unmergeable (above).
1228 * We don't worry about that case for efficiency. It won't happen
1229 * often, and the elevators are able to handle it.
1231 init_request_from_bio(req
, bio
);
1233 spin_lock_irq(q
->queue_lock
);
1234 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
) ||
1235 bio_flagged(bio
, BIO_CPU_AFFINE
))
1236 req
->cpu
= blk_cpu_to_group(smp_processor_id());
1237 if (elv_queue_empty(q
))
1239 add_request(q
, req
);
1242 __generic_unplug_device(q
);
1243 spin_unlock_irq(q
->queue_lock
);
1247 bio_endio(bio
, err
);
1252 * If bio->bi_dev is a partition, remap the location
1254 static inline void blk_partition_remap(struct bio
*bio
)
1256 struct block_device
*bdev
= bio
->bi_bdev
;
1258 if (bio_sectors(bio
) && bdev
!= bdev
->bd_contains
) {
1259 struct hd_struct
*p
= bdev
->bd_part
;
1261 bio
->bi_sector
+= p
->start_sect
;
1262 bio
->bi_bdev
= bdev
->bd_contains
;
1264 blk_add_trace_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
1265 bdev
->bd_dev
, bio
->bi_sector
,
1266 bio
->bi_sector
- p
->start_sect
);
1270 static void handle_bad_sector(struct bio
*bio
)
1272 char b
[BDEVNAME_SIZE
];
1274 printk(KERN_INFO
"attempt to access beyond end of device\n");
1275 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
1276 bdevname(bio
->bi_bdev
, b
),
1278 (unsigned long long)bio
->bi_sector
+ bio_sectors(bio
),
1279 (long long)(bio
->bi_bdev
->bd_inode
->i_size
>> 9));
1281 set_bit(BIO_EOF
, &bio
->bi_flags
);
1284 #ifdef CONFIG_FAIL_MAKE_REQUEST
1286 static DECLARE_FAULT_ATTR(fail_make_request
);
1288 static int __init
setup_fail_make_request(char *str
)
1290 return setup_fault_attr(&fail_make_request
, str
);
1292 __setup("fail_make_request=", setup_fail_make_request
);
1294 static int should_fail_request(struct bio
*bio
)
1296 struct hd_struct
*part
= bio
->bi_bdev
->bd_part
;
1298 if (part_to_disk(part
)->part0
.make_it_fail
|| part
->make_it_fail
)
1299 return should_fail(&fail_make_request
, bio
->bi_size
);
1304 static int __init
fail_make_request_debugfs(void)
1306 return init_fault_attr_dentries(&fail_make_request
,
1307 "fail_make_request");
1310 late_initcall(fail_make_request_debugfs
);
1312 #else /* CONFIG_FAIL_MAKE_REQUEST */
1314 static inline int should_fail_request(struct bio
*bio
)
1319 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1322 * Check whether this bio extends beyond the end of the device.
1324 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
1331 /* Test device or partition size, when known. */
1332 maxsector
= bio
->bi_bdev
->bd_inode
->i_size
>> 9;
1334 sector_t sector
= bio
->bi_sector
;
1336 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
1338 * This may well happen - the kernel calls bread()
1339 * without checking the size of the device, e.g., when
1340 * mounting a device.
1342 handle_bad_sector(bio
);
1351 * generic_make_request - hand a buffer to its device driver for I/O
1352 * @bio: The bio describing the location in memory and on the device.
1354 * generic_make_request() is used to make I/O requests of block
1355 * devices. It is passed a &struct bio, which describes the I/O that needs
1358 * generic_make_request() does not return any status. The
1359 * success/failure status of the request, along with notification of
1360 * completion, is delivered asynchronously through the bio->bi_end_io
1361 * function described (one day) else where.
1363 * The caller of generic_make_request must make sure that bi_io_vec
1364 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1365 * set to describe the device address, and the
1366 * bi_end_io and optionally bi_private are set to describe how
1367 * completion notification should be signaled.
1369 * generic_make_request and the drivers it calls may use bi_next if this
1370 * bio happens to be merged with someone else, and may change bi_dev and
1371 * bi_sector for remaps as it sees fit. So the values of these fields
1372 * should NOT be depended on after the call to generic_make_request.
1374 static inline void __generic_make_request(struct bio
*bio
)
1376 struct request_queue
*q
;
1377 sector_t old_sector
;
1378 int ret
, nr_sectors
= bio_sectors(bio
);
1384 if (bio_check_eod(bio
, nr_sectors
))
1388 * Resolve the mapping until finished. (drivers are
1389 * still free to implement/resolve their own stacking
1390 * by explicitly returning 0)
1392 * NOTE: we don't repeat the blk_size check for each new device.
1393 * Stacking drivers are expected to know what they are doing.
1398 char b
[BDEVNAME_SIZE
];
1400 q
= bdev_get_queue(bio
->bi_bdev
);
1403 "generic_make_request: Trying to access "
1404 "nonexistent block-device %s (%Lu)\n",
1405 bdevname(bio
->bi_bdev
, b
),
1406 (long long) bio
->bi_sector
);
1408 bio_endio(bio
, err
);
1412 if (unlikely(nr_sectors
> q
->max_hw_sectors
)) {
1413 printk(KERN_ERR
"bio too big device %s (%u > %u)\n",
1414 bdevname(bio
->bi_bdev
, b
),
1420 if (unlikely(test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)))
1423 if (should_fail_request(bio
))
1427 * If this device has partitions, remap block n
1428 * of partition p to block n+start(p) of the disk.
1430 blk_partition_remap(bio
);
1432 if (bio_integrity_enabled(bio
) && bio_integrity_prep(bio
))
1435 if (old_sector
!= -1)
1436 blk_add_trace_remap(q
, bio
, old_dev
, bio
->bi_sector
,
1439 blk_add_trace_bio(q
, bio
, BLK_TA_QUEUE
);
1441 old_sector
= bio
->bi_sector
;
1442 old_dev
= bio
->bi_bdev
->bd_dev
;
1444 if (bio_check_eod(bio
, nr_sectors
))
1446 if ((bio_empty_barrier(bio
) && !q
->prepare_flush_fn
) ||
1447 (bio_discard(bio
) && !q
->prepare_discard_fn
)) {
1452 ret
= q
->make_request_fn(q
, bio
);
1457 * We only want one ->make_request_fn to be active at a time,
1458 * else stack usage with stacked devices could be a problem.
1459 * So use current->bio_{list,tail} to keep a list of requests
1460 * submited by a make_request_fn function.
1461 * current->bio_tail is also used as a flag to say if
1462 * generic_make_request is currently active in this task or not.
1463 * If it is NULL, then no make_request is active. If it is non-NULL,
1464 * then a make_request is active, and new requests should be added
1467 void generic_make_request(struct bio
*bio
)
1469 if (current
->bio_tail
) {
1470 /* make_request is active */
1471 *(current
->bio_tail
) = bio
;
1472 bio
->bi_next
= NULL
;
1473 current
->bio_tail
= &bio
->bi_next
;
1476 /* following loop may be a bit non-obvious, and so deserves some
1478 * Before entering the loop, bio->bi_next is NULL (as all callers
1479 * ensure that) so we have a list with a single bio.
1480 * We pretend that we have just taken it off a longer list, so
1481 * we assign bio_list to the next (which is NULL) and bio_tail
1482 * to &bio_list, thus initialising the bio_list of new bios to be
1483 * added. __generic_make_request may indeed add some more bios
1484 * through a recursive call to generic_make_request. If it
1485 * did, we find a non-NULL value in bio_list and re-enter the loop
1486 * from the top. In this case we really did just take the bio
1487 * of the top of the list (no pretending) and so fixup bio_list and
1488 * bio_tail or bi_next, and call into __generic_make_request again.
1490 * The loop was structured like this to make only one call to
1491 * __generic_make_request (which is important as it is large and
1492 * inlined) and to keep the structure simple.
1494 BUG_ON(bio
->bi_next
);
1496 current
->bio_list
= bio
->bi_next
;
1497 if (bio
->bi_next
== NULL
)
1498 current
->bio_tail
= ¤t
->bio_list
;
1500 bio
->bi_next
= NULL
;
1501 __generic_make_request(bio
);
1502 bio
= current
->bio_list
;
1504 current
->bio_tail
= NULL
; /* deactivate */
1506 EXPORT_SYMBOL(generic_make_request
);
1509 * submit_bio - submit a bio to the block device layer for I/O
1510 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1511 * @bio: The &struct bio which describes the I/O
1513 * submit_bio() is very similar in purpose to generic_make_request(), and
1514 * uses that function to do most of the work. Both are fairly rough
1515 * interfaces; @bio must be presetup and ready for I/O.
1518 void submit_bio(int rw
, struct bio
*bio
)
1520 int count
= bio_sectors(bio
);
1525 * If it's a regular read/write or a barrier with data attached,
1526 * go through the normal accounting stuff before submission.
1528 if (bio_has_data(bio
)) {
1530 count_vm_events(PGPGOUT
, count
);
1532 task_io_account_read(bio
->bi_size
);
1533 count_vm_events(PGPGIN
, count
);
1536 if (unlikely(block_dump
)) {
1537 char b
[BDEVNAME_SIZE
];
1538 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s\n",
1539 current
->comm
, task_pid_nr(current
),
1540 (rw
& WRITE
) ? "WRITE" : "READ",
1541 (unsigned long long)bio
->bi_sector
,
1542 bdevname(bio
->bi_bdev
, b
));
1546 generic_make_request(bio
);
1548 EXPORT_SYMBOL(submit_bio
);
1551 * blk_rq_check_limits - Helper function to check a request for the queue limit
1553 * @rq: the request being checked
1556 * @rq may have been made based on weaker limitations of upper-level queues
1557 * in request stacking drivers, and it may violate the limitation of @q.
1558 * Since the block layer and the underlying device driver trust @rq
1559 * after it is inserted to @q, it should be checked against @q before
1560 * the insertion using this generic function.
1562 * This function should also be useful for request stacking drivers
1563 * in some cases below, so export this fuction.
1564 * Request stacking drivers like request-based dm may change the queue
1565 * limits while requests are in the queue (e.g. dm's table swapping).
1566 * Such request stacking drivers should check those requests agaist
1567 * the new queue limits again when they dispatch those requests,
1568 * although such checkings are also done against the old queue limits
1569 * when submitting requests.
1571 int blk_rq_check_limits(struct request_queue
*q
, struct request
*rq
)
1573 if (rq
->nr_sectors
> q
->max_sectors
||
1574 rq
->data_len
> q
->max_hw_sectors
<< 9) {
1575 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
1580 * queue's settings related to segment counting like q->bounce_pfn
1581 * may differ from that of other stacking queues.
1582 * Recalculate it to check the request correctly on this queue's
1585 blk_recalc_rq_segments(rq
);
1586 if (rq
->nr_phys_segments
> q
->max_phys_segments
||
1587 rq
->nr_phys_segments
> q
->max_hw_segments
) {
1588 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
1594 EXPORT_SYMBOL_GPL(blk_rq_check_limits
);
1597 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1598 * @q: the queue to submit the request
1599 * @rq: the request being queued
1601 int blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
1603 unsigned long flags
;
1605 if (blk_rq_check_limits(q
, rq
))
1608 #ifdef CONFIG_FAIL_MAKE_REQUEST
1609 if (rq
->rq_disk
&& rq
->rq_disk
->part0
.make_it_fail
&&
1610 should_fail(&fail_make_request
, blk_rq_bytes(rq
)))
1614 spin_lock_irqsave(q
->queue_lock
, flags
);
1617 * Submitting request must be dequeued before calling this function
1618 * because it will be linked to another request_queue
1620 BUG_ON(blk_queued_rq(rq
));
1622 drive_stat_acct(rq
, 1);
1623 __elv_add_request(q
, rq
, ELEVATOR_INSERT_BACK
, 0);
1625 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1629 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
1632 * __end_that_request_first - end I/O on a request
1633 * @req: the request being processed
1634 * @error: %0 for success, < %0 for error
1635 * @nr_bytes: number of bytes to complete
1638 * Ends I/O on a number of bytes attached to @req, and sets it up
1639 * for the next range of segments (if any) in the cluster.
1642 * %0 - we are done with this request, call end_that_request_last()
1643 * %1 - still buffers pending for this request
1645 static int __end_that_request_first(struct request
*req
, int error
,
1648 int total_bytes
, bio_nbytes
, next_idx
= 0;
1651 blk_add_trace_rq(req
->q
, req
, BLK_TA_COMPLETE
);
1654 * for a REQ_TYPE_BLOCK_PC request, we want to carry any eventual
1655 * sense key with us all the way through
1657 if (!blk_pc_request(req
))
1660 if (error
&& (blk_fs_request(req
) && !(req
->cmd_flags
& REQ_QUIET
))) {
1661 printk(KERN_ERR
"end_request: I/O error, dev %s, sector %llu\n",
1662 req
->rq_disk
? req
->rq_disk
->disk_name
: "?",
1663 (unsigned long long)req
->sector
);
1666 if (blk_fs_request(req
) && req
->rq_disk
) {
1667 const int rw
= rq_data_dir(req
);
1668 struct hd_struct
*part
;
1671 cpu
= part_stat_lock();
1672 part
= disk_map_sector_rcu(req
->rq_disk
, req
->sector
);
1673 part_stat_add(cpu
, part
, sectors
[rw
], nr_bytes
>> 9);
1677 total_bytes
= bio_nbytes
= 0;
1678 while ((bio
= req
->bio
) != NULL
) {
1682 * For an empty barrier request, the low level driver must
1683 * store a potential error location in ->sector. We pass
1684 * that back up in ->bi_sector.
1686 if (blk_empty_barrier(req
))
1687 bio
->bi_sector
= req
->sector
;
1689 if (nr_bytes
>= bio
->bi_size
) {
1690 req
->bio
= bio
->bi_next
;
1691 nbytes
= bio
->bi_size
;
1692 req_bio_endio(req
, bio
, nbytes
, error
);
1696 int idx
= bio
->bi_idx
+ next_idx
;
1698 if (unlikely(bio
->bi_idx
>= bio
->bi_vcnt
)) {
1699 blk_dump_rq_flags(req
, "__end_that");
1700 printk(KERN_ERR
"%s: bio idx %d >= vcnt %d\n",
1701 __func__
, bio
->bi_idx
, bio
->bi_vcnt
);
1705 nbytes
= bio_iovec_idx(bio
, idx
)->bv_len
;
1706 BIO_BUG_ON(nbytes
> bio
->bi_size
);
1709 * not a complete bvec done
1711 if (unlikely(nbytes
> nr_bytes
)) {
1712 bio_nbytes
+= nr_bytes
;
1713 total_bytes
+= nr_bytes
;
1718 * advance to the next vector
1721 bio_nbytes
+= nbytes
;
1724 total_bytes
+= nbytes
;
1730 * end more in this run, or just return 'not-done'
1732 if (unlikely(nr_bytes
<= 0))
1744 * if the request wasn't completed, update state
1747 req_bio_endio(req
, bio
, bio_nbytes
, error
);
1748 bio
->bi_idx
+= next_idx
;
1749 bio_iovec(bio
)->bv_offset
+= nr_bytes
;
1750 bio_iovec(bio
)->bv_len
-= nr_bytes
;
1753 blk_recalc_rq_sectors(req
, total_bytes
>> 9);
1754 blk_recalc_rq_segments(req
);
1759 * queue lock must be held
1761 static void end_that_request_last(struct request
*req
, int error
)
1763 struct gendisk
*disk
= req
->rq_disk
;
1765 blk_delete_timer(req
);
1767 if (blk_rq_tagged(req
))
1768 blk_queue_end_tag(req
->q
, req
);
1770 if (blk_queued_rq(req
))
1771 blkdev_dequeue_request(req
);
1773 if (unlikely(laptop_mode
) && blk_fs_request(req
))
1774 laptop_io_completion();
1777 * Account IO completion. bar_rq isn't accounted as a normal
1778 * IO on queueing nor completion. Accounting the containing
1779 * request is enough.
1781 if (disk
&& blk_fs_request(req
) && req
!= &req
->q
->bar_rq
) {
1782 unsigned long duration
= jiffies
- req
->start_time
;
1783 const int rw
= rq_data_dir(req
);
1784 struct hd_struct
*part
;
1787 cpu
= part_stat_lock();
1788 part
= disk_map_sector_rcu(disk
, req
->sector
);
1790 part_stat_inc(cpu
, part
, ios
[rw
]);
1791 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
1792 part_round_stats(cpu
, part
);
1793 part_dec_in_flight(part
);
1799 req
->end_io(req
, error
);
1801 if (blk_bidi_rq(req
))
1802 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
1804 __blk_put_request(req
->q
, req
);
1809 * blk_rq_bytes - Returns bytes left to complete in the entire request
1810 * @rq: the request being processed
1812 unsigned int blk_rq_bytes(struct request
*rq
)
1814 if (blk_fs_request(rq
))
1815 return rq
->hard_nr_sectors
<< 9;
1817 return rq
->data_len
;
1819 EXPORT_SYMBOL_GPL(blk_rq_bytes
);
1822 * blk_rq_cur_bytes - Returns bytes left to complete in the current segment
1823 * @rq: the request being processed
1825 unsigned int blk_rq_cur_bytes(struct request
*rq
)
1827 if (blk_fs_request(rq
))
1828 return rq
->current_nr_sectors
<< 9;
1831 return rq
->bio
->bi_size
;
1833 return rq
->data_len
;
1835 EXPORT_SYMBOL_GPL(blk_rq_cur_bytes
);
1838 * end_request - end I/O on the current segment of the request
1839 * @req: the request being processed
1840 * @uptodate: error value or %0/%1 uptodate flag
1843 * Ends I/O on the current segment of a request. If that is the only
1844 * remaining segment, the request is also completed and freed.
1846 * This is a remnant of how older block drivers handled I/O completions.
1847 * Modern drivers typically end I/O on the full request in one go, unless
1848 * they have a residual value to account for. For that case this function
1849 * isn't really useful, unless the residual just happens to be the
1850 * full current segment. In other words, don't use this function in new
1851 * code. Use blk_end_request() or __blk_end_request() to end a request.
1853 void end_request(struct request
*req
, int uptodate
)
1858 error
= uptodate
? uptodate
: -EIO
;
1860 __blk_end_request(req
, error
, req
->hard_cur_sectors
<< 9);
1862 EXPORT_SYMBOL(end_request
);
1864 static int end_that_request_data(struct request
*rq
, int error
,
1865 unsigned int nr_bytes
, unsigned int bidi_bytes
)
1868 if (__end_that_request_first(rq
, error
, nr_bytes
))
1871 /* Bidi request must be completed as a whole */
1872 if (blk_bidi_rq(rq
) &&
1873 __end_that_request_first(rq
->next_rq
, error
, bidi_bytes
))
1881 * blk_end_io - Generic end_io function to complete a request.
1882 * @rq: the request being processed
1883 * @error: %0 for success, < %0 for error
1884 * @nr_bytes: number of bytes to complete @rq
1885 * @bidi_bytes: number of bytes to complete @rq->next_rq
1886 * @drv_callback: function called between completion of bios in the request
1887 * and completion of the request.
1888 * If the callback returns non %0, this helper returns without
1889 * completion of the request.
1892 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
1893 * If @rq has leftover, sets it up for the next range of segments.
1896 * %0 - we are done with this request
1897 * %1 - this request is not freed yet, it still has pending buffers.
1899 static int blk_end_io(struct request
*rq
, int error
, unsigned int nr_bytes
,
1900 unsigned int bidi_bytes
,
1901 int (drv_callback
)(struct request
*))
1903 struct request_queue
*q
= rq
->q
;
1904 unsigned long flags
= 0UL;
1906 if (end_that_request_data(rq
, error
, nr_bytes
, bidi_bytes
))
1909 /* Special feature for tricky drivers */
1910 if (drv_callback
&& drv_callback(rq
))
1913 add_disk_randomness(rq
->rq_disk
);
1915 spin_lock_irqsave(q
->queue_lock
, flags
);
1916 end_that_request_last(rq
, error
);
1917 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1923 * blk_end_request - Helper function for drivers to complete the request.
1924 * @rq: the request being processed
1925 * @error: %0 for success, < %0 for error
1926 * @nr_bytes: number of bytes to complete
1929 * Ends I/O on a number of bytes attached to @rq.
1930 * If @rq has leftover, sets it up for the next range of segments.
1933 * %0 - we are done with this request
1934 * %1 - still buffers pending for this request
1936 int blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
1938 return blk_end_io(rq
, error
, nr_bytes
, 0, NULL
);
1940 EXPORT_SYMBOL_GPL(blk_end_request
);
1943 * __blk_end_request - Helper function for drivers to complete the request.
1944 * @rq: the request being processed
1945 * @error: %0 for success, < %0 for error
1946 * @nr_bytes: number of bytes to complete
1949 * Must be called with queue lock held unlike blk_end_request().
1952 * %0 - we are done with this request
1953 * %1 - still buffers pending for this request
1955 int __blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
1957 if (rq
->bio
&& __end_that_request_first(rq
, error
, nr_bytes
))
1960 add_disk_randomness(rq
->rq_disk
);
1962 end_that_request_last(rq
, error
);
1966 EXPORT_SYMBOL_GPL(__blk_end_request
);
1969 * blk_end_bidi_request - Helper function for drivers to complete bidi request.
1970 * @rq: the bidi request being processed
1971 * @error: %0 for success, < %0 for error
1972 * @nr_bytes: number of bytes to complete @rq
1973 * @bidi_bytes: number of bytes to complete @rq->next_rq
1976 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
1979 * %0 - we are done with this request
1980 * %1 - still buffers pending for this request
1982 int blk_end_bidi_request(struct request
*rq
, int error
, unsigned int nr_bytes
,
1983 unsigned int bidi_bytes
)
1985 return blk_end_io(rq
, error
, nr_bytes
, bidi_bytes
, NULL
);
1987 EXPORT_SYMBOL_GPL(blk_end_bidi_request
);
1990 * blk_update_request - Special helper function for request stacking drivers
1991 * @rq: the request being processed
1992 * @error: %0 for success, < %0 for error
1993 * @nr_bytes: number of bytes to complete @rq
1996 * Ends I/O on a number of bytes attached to @rq, but doesn't complete
1997 * the request structure even if @rq doesn't have leftover.
1998 * If @rq has leftover, sets it up for the next range of segments.
2000 * This special helper function is only for request stacking drivers
2001 * (e.g. request-based dm) so that they can handle partial completion.
2002 * Actual device drivers should use blk_end_request instead.
2004 void blk_update_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2006 if (!end_that_request_data(rq
, error
, nr_bytes
, 0)) {
2008 * These members are not updated in end_that_request_data()
2009 * when all bios are completed.
2010 * Update them so that the request stacking driver can find
2011 * how many bytes remain in the request later.
2013 rq
->nr_sectors
= rq
->hard_nr_sectors
= 0;
2014 rq
->current_nr_sectors
= rq
->hard_cur_sectors
= 0;
2017 EXPORT_SYMBOL_GPL(blk_update_request
);
2020 * blk_end_request_callback - Special helper function for tricky drivers
2021 * @rq: the request being processed
2022 * @error: %0 for success, < %0 for error
2023 * @nr_bytes: number of bytes to complete
2024 * @drv_callback: function called between completion of bios in the request
2025 * and completion of the request.
2026 * If the callback returns non %0, this helper returns without
2027 * completion of the request.
2030 * Ends I/O on a number of bytes attached to @rq.
2031 * If @rq has leftover, sets it up for the next range of segments.
2033 * This special helper function is used only for existing tricky drivers.
2034 * (e.g. cdrom_newpc_intr() of ide-cd)
2035 * This interface will be removed when such drivers are rewritten.
2036 * Don't use this interface in other places anymore.
2039 * %0 - we are done with this request
2040 * %1 - this request is not freed yet.
2041 * this request still has pending buffers or
2042 * the driver doesn't want to finish this request yet.
2044 int blk_end_request_callback(struct request
*rq
, int error
,
2045 unsigned int nr_bytes
,
2046 int (drv_callback
)(struct request
*))
2048 return blk_end_io(rq
, error
, nr_bytes
, 0, drv_callback
);
2050 EXPORT_SYMBOL_GPL(blk_end_request_callback
);
2052 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
2055 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw, and
2056 we want BIO_RW_AHEAD (bit 1) to imply REQ_FAILFAST (bit 1). */
2057 rq
->cmd_flags
|= (bio
->bi_rw
& 3);
2059 if (bio_has_data(bio
)) {
2060 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2061 rq
->buffer
= bio_data(bio
);
2063 rq
->current_nr_sectors
= bio_cur_sectors(bio
);
2064 rq
->hard_cur_sectors
= rq
->current_nr_sectors
;
2065 rq
->hard_nr_sectors
= rq
->nr_sectors
= bio_sectors(bio
);
2066 rq
->data_len
= bio
->bi_size
;
2068 rq
->bio
= rq
->biotail
= bio
;
2071 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2075 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2076 * @q : the queue of the device being checked
2079 * Check if underlying low-level drivers of a device are busy.
2080 * If the drivers want to export their busy state, they must set own
2081 * exporting function using blk_queue_lld_busy() first.
2083 * Basically, this function is used only by request stacking drivers
2084 * to stop dispatching requests to underlying devices when underlying
2085 * devices are busy. This behavior helps more I/O merging on the queue
2086 * of the request stacking driver and prevents I/O throughput regression
2087 * on burst I/O load.
2090 * 0 - Not busy (The request stacking driver should dispatch request)
2091 * 1 - Busy (The request stacking driver should stop dispatching request)
2093 int blk_lld_busy(struct request_queue
*q
)
2096 return q
->lld_busy_fn(q
);
2100 EXPORT_SYMBOL_GPL(blk_lld_busy
);
2102 int kblockd_schedule_work(struct request_queue
*q
, struct work_struct
*work
)
2104 return queue_work(kblockd_workqueue
, work
);
2106 EXPORT_SYMBOL(kblockd_schedule_work
);
2108 void kblockd_flush_work(struct work_struct
*work
)
2110 cancel_work_sync(work
);
2112 EXPORT_SYMBOL(kblockd_flush_work
);
2114 int __init
blk_dev_init(void)
2116 kblockd_workqueue
= create_workqueue("kblockd");
2117 if (!kblockd_workqueue
)
2118 panic("Failed to create kblockd\n");
2120 request_cachep
= kmem_cache_create("blkdev_requests",
2121 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
2123 blk_requestq_cachep
= kmem_cache_create("blkdev_queue",
2124 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);