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
) || !rq
->rq_disk
)
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
);
137 EXPORT_SYMBOL(blk_rq_init
);
139 static void req_bio_endio(struct request
*rq
, struct bio
*bio
,
140 unsigned int nbytes
, int error
)
142 struct request_queue
*q
= rq
->q
;
144 if (&q
->bar_rq
!= rq
) {
146 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
147 else if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
150 if (unlikely(nbytes
> bio
->bi_size
)) {
151 printk(KERN_ERR
"%s: want %u bytes done, %u left\n",
152 __func__
, nbytes
, bio
->bi_size
);
153 nbytes
= bio
->bi_size
;
156 if (unlikely(rq
->cmd_flags
& REQ_QUIET
))
157 set_bit(BIO_QUIET
, &bio
->bi_flags
);
159 bio
->bi_size
-= nbytes
;
160 bio
->bi_sector
+= (nbytes
>> 9);
162 if (bio_integrity(bio
))
163 bio_integrity_advance(bio
, nbytes
);
165 if (bio
->bi_size
== 0)
166 bio_endio(bio
, error
);
170 * Okay, this is the barrier request in progress, just
173 if (error
&& !q
->orderr
)
178 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
182 printk(KERN_INFO
"%s: dev %s: type=%x, flags=%x\n", msg
,
183 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->cmd_type
,
186 printk(KERN_INFO
" sector %llu, nr/cnr %lu/%u\n",
187 (unsigned long long)rq
->sector
,
189 rq
->current_nr_sectors
);
190 printk(KERN_INFO
" bio %p, biotail %p, buffer %p, data %p, len %u\n",
191 rq
->bio
, rq
->biotail
,
192 rq
->buffer
, rq
->data
,
195 if (blk_pc_request(rq
)) {
196 printk(KERN_INFO
" cdb: ");
197 for (bit
= 0; bit
< BLK_MAX_CDB
; bit
++)
198 printk("%02x ", rq
->cmd
[bit
]);
202 EXPORT_SYMBOL(blk_dump_rq_flags
);
205 * "plug" the device if there are no outstanding requests: this will
206 * force the transfer to start only after we have put all the requests
209 * This is called with interrupts off and no requests on the queue and
210 * with the queue lock held.
212 void blk_plug_device(struct request_queue
*q
)
214 WARN_ON(!irqs_disabled());
217 * don't plug a stopped queue, it must be paired with blk_start_queue()
218 * which will restart the queueing
220 if (blk_queue_stopped(q
))
223 if (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED
, q
)) {
224 mod_timer(&q
->unplug_timer
, jiffies
+ q
->unplug_delay
);
228 EXPORT_SYMBOL(blk_plug_device
);
231 * blk_plug_device_unlocked - plug a device without queue lock held
232 * @q: The &struct request_queue to plug
235 * Like @blk_plug_device(), but grabs the queue lock and disables
238 void blk_plug_device_unlocked(struct request_queue
*q
)
242 spin_lock_irqsave(q
->queue_lock
, flags
);
244 spin_unlock_irqrestore(q
->queue_lock
, flags
);
246 EXPORT_SYMBOL(blk_plug_device_unlocked
);
249 * remove the queue from the plugged list, if present. called with
250 * queue lock held and interrupts disabled.
252 int blk_remove_plug(struct request_queue
*q
)
254 WARN_ON(!irqs_disabled());
256 if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED
, q
))
259 del_timer(&q
->unplug_timer
);
262 EXPORT_SYMBOL(blk_remove_plug
);
265 * remove the plug and let it rip..
267 void __generic_unplug_device(struct request_queue
*q
)
269 if (unlikely(blk_queue_stopped(q
)))
272 if (!blk_remove_plug(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
);
336 static void blk_invoke_request_fn(struct request_queue
*q
)
338 if (unlikely(blk_queue_stopped(q
)))
342 * one level of recursion is ok and is much faster than kicking
343 * the unplug handling
345 if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER
, q
)) {
347 queue_flag_clear(QUEUE_FLAG_REENTER
, q
);
349 queue_flag_set(QUEUE_FLAG_PLUGGED
, q
);
350 kblockd_schedule_work(q
, &q
->unplug_work
);
355 * blk_start_queue - restart a previously stopped queue
356 * @q: The &struct request_queue in question
359 * blk_start_queue() will clear the stop flag on the queue, and call
360 * the request_fn for the queue if it was in a stopped state when
361 * entered. Also see blk_stop_queue(). Queue lock must be held.
363 void blk_start_queue(struct request_queue
*q
)
365 WARN_ON(!irqs_disabled());
367 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
368 blk_invoke_request_fn(q
);
370 EXPORT_SYMBOL(blk_start_queue
);
373 * blk_stop_queue - stop a queue
374 * @q: The &struct request_queue in question
377 * The Linux block layer assumes that a block driver will consume all
378 * entries on the request queue when the request_fn strategy is called.
379 * Often this will not happen, because of hardware limitations (queue
380 * depth settings). If a device driver gets a 'queue full' response,
381 * or if it simply chooses not to queue more I/O at one point, it can
382 * call this function to prevent the request_fn from being called until
383 * the driver has signalled it's ready to go again. This happens by calling
384 * blk_start_queue() to restart queue operations. Queue lock must be held.
386 void blk_stop_queue(struct request_queue
*q
)
389 queue_flag_set(QUEUE_FLAG_STOPPED
, q
);
391 EXPORT_SYMBOL(blk_stop_queue
);
394 * blk_sync_queue - cancel any pending callbacks on a queue
398 * The block layer may perform asynchronous callback activity
399 * on a queue, such as calling the unplug function after a timeout.
400 * A block device may call blk_sync_queue to ensure that any
401 * such activity is cancelled, thus allowing it to release resources
402 * that the callbacks might use. The caller must already have made sure
403 * that its ->make_request_fn will not re-add plugging prior to calling
407 void blk_sync_queue(struct request_queue
*q
)
409 del_timer_sync(&q
->unplug_timer
);
410 del_timer_sync(&q
->timeout
);
411 kblockd_flush_work(&q
->unplug_work
);
413 EXPORT_SYMBOL(blk_sync_queue
);
416 * __blk_run_queue - run a single device queue
417 * @q: The queue to run
420 * See @blk_run_queue. This variant must be called with the queue lock
421 * held and interrupts disabled.
424 void __blk_run_queue(struct request_queue
*q
)
429 * Only recurse once to avoid overrunning the stack, let the unplug
430 * handling reinvoke the handler shortly if we already got there.
432 if (!elv_queue_empty(q
))
433 blk_invoke_request_fn(q
);
435 EXPORT_SYMBOL(__blk_run_queue
);
438 * blk_run_queue - run a single device queue
439 * @q: The queue to run
442 * Invoke request handling on this queue, if it has pending work to do.
443 * May be used to restart queueing when a request has completed. Also
444 * See @blk_start_queueing.
447 void blk_run_queue(struct request_queue
*q
)
451 spin_lock_irqsave(q
->queue_lock
, flags
);
453 spin_unlock_irqrestore(q
->queue_lock
, flags
);
455 EXPORT_SYMBOL(blk_run_queue
);
457 void blk_put_queue(struct request_queue
*q
)
459 kobject_put(&q
->kobj
);
462 void blk_cleanup_queue(struct request_queue
*q
)
465 * We know we have process context here, so we can be a little
466 * cautious and ensure that pending block actions on this device
467 * are done before moving on. Going into this function, we should
468 * not have processes doing IO to this device.
472 mutex_lock(&q
->sysfs_lock
);
473 queue_flag_set_unlocked(QUEUE_FLAG_DEAD
, q
);
474 mutex_unlock(&q
->sysfs_lock
);
477 elevator_exit(q
->elevator
);
481 EXPORT_SYMBOL(blk_cleanup_queue
);
483 static int blk_init_free_list(struct request_queue
*q
)
485 struct request_list
*rl
= &q
->rq
;
487 rl
->count
[READ
] = rl
->count
[WRITE
] = 0;
488 rl
->starved
[READ
] = rl
->starved
[WRITE
] = 0;
490 init_waitqueue_head(&rl
->wait
[READ
]);
491 init_waitqueue_head(&rl
->wait
[WRITE
]);
493 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, mempool_alloc_slab
,
494 mempool_free_slab
, request_cachep
, q
->node
);
502 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
504 return blk_alloc_queue_node(gfp_mask
, -1);
506 EXPORT_SYMBOL(blk_alloc_queue
);
508 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
510 struct request_queue
*q
;
513 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
514 gfp_mask
| __GFP_ZERO
, node_id
);
518 q
->backing_dev_info
.unplug_io_fn
= blk_backing_dev_unplug
;
519 q
->backing_dev_info
.unplug_io_data
= q
;
520 err
= bdi_init(&q
->backing_dev_info
);
522 kmem_cache_free(blk_requestq_cachep
, q
);
526 init_timer(&q
->unplug_timer
);
527 setup_timer(&q
->timeout
, blk_rq_timed_out_timer
, (unsigned long) q
);
528 INIT_LIST_HEAD(&q
->timeout_list
);
529 INIT_WORK(&q
->unplug_work
, blk_unplug_work
);
531 kobject_init(&q
->kobj
, &blk_queue_ktype
);
533 mutex_init(&q
->sysfs_lock
);
534 spin_lock_init(&q
->__queue_lock
);
538 EXPORT_SYMBOL(blk_alloc_queue_node
);
541 * blk_init_queue - prepare a request queue for use with a block device
542 * @rfn: The function to be called to process requests that have been
543 * placed on the queue.
544 * @lock: Request queue spin lock
547 * If a block device wishes to use the standard request handling procedures,
548 * which sorts requests and coalesces adjacent requests, then it must
549 * call blk_init_queue(). The function @rfn will be called when there
550 * are requests on the queue that need to be processed. If the device
551 * supports plugging, then @rfn may not be called immediately when requests
552 * are available on the queue, but may be called at some time later instead.
553 * Plugged queues are generally unplugged when a buffer belonging to one
554 * of the requests on the queue is needed, or due to memory pressure.
556 * @rfn is not required, or even expected, to remove all requests off the
557 * queue, but only as many as it can handle at a time. If it does leave
558 * requests on the queue, it is responsible for arranging that the requests
559 * get dealt with eventually.
561 * The queue spin lock must be held while manipulating the requests on the
562 * request queue; this lock will be taken also from interrupt context, so irq
563 * disabling is needed for it.
565 * Function returns a pointer to the initialized request queue, or %NULL if
569 * blk_init_queue() must be paired with a blk_cleanup_queue() call
570 * when the block device is deactivated (such as at module unload).
573 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
575 return blk_init_queue_node(rfn
, lock
, -1);
577 EXPORT_SYMBOL(blk_init_queue
);
579 struct request_queue
*
580 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
582 struct request_queue
*q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
588 if (blk_init_free_list(q
)) {
589 kmem_cache_free(blk_requestq_cachep
, q
);
594 * if caller didn't supply a lock, they get per-queue locking with
598 lock
= &q
->__queue_lock
;
601 q
->prep_rq_fn
= NULL
;
602 q
->unplug_fn
= generic_unplug_device
;
603 q
->queue_flags
= (1 << QUEUE_FLAG_CLUSTER
|
604 1 << QUEUE_FLAG_STACKABLE
);
605 q
->queue_lock
= lock
;
607 blk_queue_segment_boundary(q
, BLK_SEG_BOUNDARY_MASK
);
609 blk_queue_make_request(q
, __make_request
);
610 blk_queue_max_segment_size(q
, MAX_SEGMENT_SIZE
);
612 blk_queue_max_hw_segments(q
, MAX_HW_SEGMENTS
);
613 blk_queue_max_phys_segments(q
, MAX_PHYS_SEGMENTS
);
615 q
->sg_reserved_size
= INT_MAX
;
617 blk_set_cmd_filter_defaults(&q
->cmd_filter
);
622 if (!elevator_init(q
, NULL
)) {
623 blk_queue_congestion_threshold(q
);
630 EXPORT_SYMBOL(blk_init_queue_node
);
632 int blk_get_queue(struct request_queue
*q
)
634 if (likely(!test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
))) {
635 kobject_get(&q
->kobj
);
642 static inline void blk_free_request(struct request_queue
*q
, struct request
*rq
)
644 if (rq
->cmd_flags
& REQ_ELVPRIV
)
645 elv_put_request(q
, rq
);
646 mempool_free(rq
, q
->rq
.rq_pool
);
649 static struct request
*
650 blk_alloc_request(struct request_queue
*q
, int rw
, int priv
, gfp_t gfp_mask
)
652 struct request
*rq
= mempool_alloc(q
->rq
.rq_pool
, gfp_mask
);
659 rq
->cmd_flags
= rw
| REQ_ALLOCED
;
662 if (unlikely(elv_set_request(q
, rq
, gfp_mask
))) {
663 mempool_free(rq
, q
->rq
.rq_pool
);
666 rq
->cmd_flags
|= REQ_ELVPRIV
;
673 * ioc_batching returns true if the ioc is a valid batching request and
674 * should be given priority access to a request.
676 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
682 * Make sure the process is able to allocate at least 1 request
683 * even if the batch times out, otherwise we could theoretically
686 return ioc
->nr_batch_requests
== q
->nr_batching
||
687 (ioc
->nr_batch_requests
> 0
688 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
692 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
693 * will cause the process to be a "batcher" on all queues in the system. This
694 * is the behaviour we want though - once it gets a wakeup it should be given
697 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
699 if (!ioc
|| ioc_batching(q
, ioc
))
702 ioc
->nr_batch_requests
= q
->nr_batching
;
703 ioc
->last_waited
= jiffies
;
706 static void __freed_request(struct request_queue
*q
, int rw
)
708 struct request_list
*rl
= &q
->rq
;
710 if (rl
->count
[rw
] < queue_congestion_off_threshold(q
))
711 blk_clear_queue_congested(q
, rw
);
713 if (rl
->count
[rw
] + 1 <= q
->nr_requests
) {
714 if (waitqueue_active(&rl
->wait
[rw
]))
715 wake_up(&rl
->wait
[rw
]);
717 blk_clear_queue_full(q
, rw
);
722 * A request has just been released. Account for it, update the full and
723 * congestion status, wake up any waiters. Called under q->queue_lock.
725 static void freed_request(struct request_queue
*q
, int rw
, int priv
)
727 struct request_list
*rl
= &q
->rq
;
733 __freed_request(q
, rw
);
735 if (unlikely(rl
->starved
[rw
^ 1]))
736 __freed_request(q
, rw
^ 1);
739 #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
741 * Get a free request, queue_lock must be held.
742 * Returns NULL on failure, with queue_lock held.
743 * Returns !NULL on success, with queue_lock *not held*.
745 static struct request
*get_request(struct request_queue
*q
, int rw_flags
,
746 struct bio
*bio
, gfp_t gfp_mask
)
748 struct request
*rq
= NULL
;
749 struct request_list
*rl
= &q
->rq
;
750 struct io_context
*ioc
= NULL
;
751 const int rw
= rw_flags
& 0x01;
754 may_queue
= elv_may_queue(q
, rw_flags
);
755 if (may_queue
== ELV_MQUEUE_NO
)
758 if (rl
->count
[rw
]+1 >= queue_congestion_on_threshold(q
)) {
759 if (rl
->count
[rw
]+1 >= q
->nr_requests
) {
760 ioc
= current_io_context(GFP_ATOMIC
, q
->node
);
762 * The queue will fill after this allocation, so set
763 * it as full, and mark this process as "batching".
764 * This process will be allowed to complete a batch of
765 * requests, others will be blocked.
767 if (!blk_queue_full(q
, rw
)) {
768 ioc_set_batching(q
, ioc
);
769 blk_set_queue_full(q
, rw
);
771 if (may_queue
!= ELV_MQUEUE_MUST
772 && !ioc_batching(q
, ioc
)) {
774 * The queue is full and the allocating
775 * process is not a "batcher", and not
776 * exempted by the IO scheduler
782 blk_set_queue_congested(q
, rw
);
786 * Only allow batching queuers to allocate up to 50% over the defined
787 * limit of requests, otherwise we could have thousands of requests
788 * allocated with any setting of ->nr_requests
790 if (rl
->count
[rw
] >= (3 * q
->nr_requests
/ 2))
796 priv
= !test_bit(QUEUE_FLAG_ELVSWITCH
, &q
->queue_flags
);
800 spin_unlock_irq(q
->queue_lock
);
802 rq
= blk_alloc_request(q
, rw_flags
, priv
, gfp_mask
);
805 * Allocation failed presumably due to memory. Undo anything
806 * we might have messed up.
808 * Allocating task should really be put onto the front of the
809 * wait queue, but this is pretty rare.
811 spin_lock_irq(q
->queue_lock
);
812 freed_request(q
, rw
, priv
);
815 * in the very unlikely event that allocation failed and no
816 * requests for this direction was pending, mark us starved
817 * so that freeing of a request in the other direction will
818 * notice us. another possible fix would be to split the
819 * rq mempool into READ and WRITE
822 if (unlikely(rl
->count
[rw
] == 0))
829 * ioc may be NULL here, and ioc_batching will be false. That's
830 * OK, if the queue is under the request limit then requests need
831 * not count toward the nr_batch_requests limit. There will always
832 * be some limit enforced by BLK_BATCH_TIME.
834 if (ioc_batching(q
, ioc
))
835 ioc
->nr_batch_requests
--;
837 trace_block_getrq(q
, bio
, rw
);
843 * No available requests for this queue, unplug the device and wait for some
844 * requests to become available.
846 * Called with q->queue_lock held, and returns with it unlocked.
848 static struct request
*get_request_wait(struct request_queue
*q
, int rw_flags
,
851 const int rw
= rw_flags
& 0x01;
854 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
857 struct io_context
*ioc
;
858 struct request_list
*rl
= &q
->rq
;
860 prepare_to_wait_exclusive(&rl
->wait
[rw
], &wait
,
861 TASK_UNINTERRUPTIBLE
);
863 trace_block_sleeprq(q
, bio
, rw
);
865 __generic_unplug_device(q
);
866 spin_unlock_irq(q
->queue_lock
);
870 * After sleeping, we become a "batching" process and
871 * will be able to allocate at least one request, and
872 * up to a big batch of them for a small period time.
873 * See ioc_batching, ioc_set_batching
875 ioc
= current_io_context(GFP_NOIO
, q
->node
);
876 ioc_set_batching(q
, ioc
);
878 spin_lock_irq(q
->queue_lock
);
879 finish_wait(&rl
->wait
[rw
], &wait
);
881 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
887 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
891 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
893 spin_lock_irq(q
->queue_lock
);
894 if (gfp_mask
& __GFP_WAIT
) {
895 rq
= get_request_wait(q
, rw
, NULL
);
897 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
899 spin_unlock_irq(q
->queue_lock
);
901 /* q->queue_lock is unlocked at this point */
905 EXPORT_SYMBOL(blk_get_request
);
908 * blk_start_queueing - initiate dispatch of requests to device
909 * @q: request queue to kick into gear
911 * This is basically a helper to remove the need to know whether a queue
912 * is plugged or not if someone just wants to initiate dispatch of requests
913 * for this queue. Should be used to start queueing on a device outside
914 * of ->request_fn() context. Also see @blk_run_queue.
916 * The queue lock must be held with interrupts disabled.
918 void blk_start_queueing(struct request_queue
*q
)
920 if (!blk_queue_plugged(q
)) {
921 if (unlikely(blk_queue_stopped(q
)))
925 __generic_unplug_device(q
);
927 EXPORT_SYMBOL(blk_start_queueing
);
930 * blk_requeue_request - put a request back on queue
931 * @q: request queue where request should be inserted
932 * @rq: request to be inserted
935 * Drivers often keep queueing requests until the hardware cannot accept
936 * more, when that condition happens we need to put the request back
937 * on the queue. Must be called with queue lock held.
939 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
941 blk_delete_timer(rq
);
942 blk_clear_rq_complete(rq
);
943 trace_block_rq_requeue(q
, rq
);
945 if (blk_rq_tagged(rq
))
946 blk_queue_end_tag(q
, rq
);
948 elv_requeue_request(q
, rq
);
950 EXPORT_SYMBOL(blk_requeue_request
);
953 * blk_insert_request - insert a special request into a request queue
954 * @q: request queue where request should be inserted
955 * @rq: request to be inserted
956 * @at_head: insert request at head or tail of queue
957 * @data: private data
960 * Many block devices need to execute commands asynchronously, so they don't
961 * block the whole kernel from preemption during request execution. This is
962 * accomplished normally by inserting aritficial requests tagged as
963 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
964 * be scheduled for actual execution by the request queue.
966 * We have the option of inserting the head or the tail of the queue.
967 * Typically we use the tail for new ioctls and so forth. We use the head
968 * of the queue for things like a QUEUE_FULL message from a device, or a
969 * host that is unable to accept a particular command.
971 void blk_insert_request(struct request_queue
*q
, struct request
*rq
,
972 int at_head
, void *data
)
974 int where
= at_head
? ELEVATOR_INSERT_FRONT
: ELEVATOR_INSERT_BACK
;
978 * tell I/O scheduler that this isn't a regular read/write (ie it
979 * must not attempt merges on this) and that it acts as a soft
982 rq
->cmd_type
= REQ_TYPE_SPECIAL
;
983 rq
->cmd_flags
|= REQ_SOFTBARRIER
;
987 spin_lock_irqsave(q
->queue_lock
, flags
);
990 * If command is tagged, release the tag
992 if (blk_rq_tagged(rq
))
993 blk_queue_end_tag(q
, rq
);
995 drive_stat_acct(rq
, 1);
996 __elv_add_request(q
, rq
, where
, 0);
997 blk_start_queueing(q
);
998 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1000 EXPORT_SYMBOL(blk_insert_request
);
1003 * add-request adds a request to the linked list.
1004 * queue lock is held and interrupts disabled, as we muck with the
1005 * request queue list.
1007 static inline void add_request(struct request_queue
*q
, struct request
*req
)
1009 drive_stat_acct(req
, 1);
1012 * elevator indicated where it wants this request to be
1013 * inserted at elevator_merge time
1015 __elv_add_request(q
, req
, ELEVATOR_INSERT_SORT
, 0);
1018 static void part_round_stats_single(int cpu
, struct hd_struct
*part
,
1021 if (now
== part
->stamp
)
1024 if (part
->in_flight
) {
1025 __part_stat_add(cpu
, part
, time_in_queue
,
1026 part
->in_flight
* (now
- part
->stamp
));
1027 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
1033 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1034 * @cpu: cpu number for stats access
1035 * @part: target partition
1037 * The average IO queue length and utilisation statistics are maintained
1038 * by observing the current state of the queue length and the amount of
1039 * time it has been in this state for.
1041 * Normally, that accounting is done on IO completion, but that can result
1042 * in more than a second's worth of IO being accounted for within any one
1043 * second, leading to >100% utilisation. To deal with that, we call this
1044 * function to do a round-off before returning the results when reading
1045 * /proc/diskstats. This accounts immediately for all queue usage up to
1046 * the current jiffies and restarts the counters again.
1048 void part_round_stats(int cpu
, struct hd_struct
*part
)
1050 unsigned long now
= jiffies
;
1053 part_round_stats_single(cpu
, &part_to_disk(part
)->part0
, now
);
1054 part_round_stats_single(cpu
, part
, now
);
1056 EXPORT_SYMBOL_GPL(part_round_stats
);
1059 * queue lock must be held
1061 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1065 if (unlikely(--req
->ref_count
))
1068 elv_completed_request(q
, req
);
1071 * Request may not have originated from ll_rw_blk. if not,
1072 * it didn't come out of our reserved rq pools
1074 if (req
->cmd_flags
& REQ_ALLOCED
) {
1075 int rw
= rq_data_dir(req
);
1076 int priv
= req
->cmd_flags
& REQ_ELVPRIV
;
1078 BUG_ON(!list_empty(&req
->queuelist
));
1079 BUG_ON(!hlist_unhashed(&req
->hash
));
1081 blk_free_request(q
, req
);
1082 freed_request(q
, rw
, priv
);
1085 EXPORT_SYMBOL_GPL(__blk_put_request
);
1087 void blk_put_request(struct request
*req
)
1089 unsigned long flags
;
1090 struct request_queue
*q
= req
->q
;
1092 spin_lock_irqsave(q
->queue_lock
, flags
);
1093 __blk_put_request(q
, req
);
1094 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1096 EXPORT_SYMBOL(blk_put_request
);
1098 void init_request_from_bio(struct request
*req
, struct bio
*bio
)
1100 req
->cpu
= bio
->bi_comp_cpu
;
1101 req
->cmd_type
= REQ_TYPE_FS
;
1104 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
1106 if (bio_rw_ahead(bio
))
1107 req
->cmd_flags
|= (REQ_FAILFAST_DEV
| REQ_FAILFAST_TRANSPORT
|
1108 REQ_FAILFAST_DRIVER
);
1109 if (bio_failfast_dev(bio
))
1110 req
->cmd_flags
|= REQ_FAILFAST_DEV
;
1111 if (bio_failfast_transport(bio
))
1112 req
->cmd_flags
|= REQ_FAILFAST_TRANSPORT
;
1113 if (bio_failfast_driver(bio
))
1114 req
->cmd_flags
|= REQ_FAILFAST_DRIVER
;
1117 * REQ_BARRIER implies no merging, but lets make it explicit
1119 if (unlikely(bio_discard(bio
))) {
1120 req
->cmd_flags
|= REQ_DISCARD
;
1121 if (bio_barrier(bio
))
1122 req
->cmd_flags
|= REQ_SOFTBARRIER
;
1123 req
->q
->prepare_discard_fn(req
->q
, req
);
1124 } else if (unlikely(bio_barrier(bio
)))
1125 req
->cmd_flags
|= (REQ_HARDBARRIER
| REQ_NOMERGE
);
1128 req
->cmd_flags
|= REQ_RW_SYNC
;
1129 if (bio_rw_meta(bio
))
1130 req
->cmd_flags
|= REQ_RW_META
;
1133 req
->hard_sector
= req
->sector
= bio
->bi_sector
;
1134 req
->ioprio
= bio_prio(bio
);
1135 req
->start_time
= jiffies
;
1136 blk_rq_bio_prep(req
->q
, req
, bio
);
1139 static int __make_request(struct request_queue
*q
, struct bio
*bio
)
1141 struct request
*req
;
1142 int el_ret
, nr_sectors
, barrier
, discard
, err
;
1143 const unsigned short prio
= bio_prio(bio
);
1144 const int sync
= bio_sync(bio
);
1147 nr_sectors
= bio_sectors(bio
);
1150 * low level driver can indicate that it wants pages above a
1151 * certain limit bounced to low memory (ie for highmem, or even
1152 * ISA dma in theory)
1154 blk_queue_bounce(q
, &bio
);
1156 barrier
= bio_barrier(bio
);
1157 if (unlikely(barrier
) && bio_has_data(bio
) &&
1158 (q
->next_ordered
== QUEUE_ORDERED_NONE
)) {
1163 discard
= bio_discard(bio
);
1164 if (unlikely(discard
) && !q
->prepare_discard_fn
) {
1169 spin_lock_irq(q
->queue_lock
);
1171 if (unlikely(barrier
) || elv_queue_empty(q
))
1174 el_ret
= elv_merge(q
, &req
, bio
);
1176 case ELEVATOR_BACK_MERGE
:
1177 BUG_ON(!rq_mergeable(req
));
1179 if (!ll_back_merge_fn(q
, req
, bio
))
1182 trace_block_bio_backmerge(q
, bio
);
1184 req
->biotail
->bi_next
= bio
;
1186 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
1187 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
1188 if (!blk_rq_cpu_valid(req
))
1189 req
->cpu
= bio
->bi_comp_cpu
;
1190 drive_stat_acct(req
, 0);
1191 if (!attempt_back_merge(q
, req
))
1192 elv_merged_request(q
, req
, el_ret
);
1195 case ELEVATOR_FRONT_MERGE
:
1196 BUG_ON(!rq_mergeable(req
));
1198 if (!ll_front_merge_fn(q
, req
, bio
))
1201 trace_block_bio_frontmerge(q
, bio
);
1203 bio
->bi_next
= req
->bio
;
1207 * may not be valid. if the low level driver said
1208 * it didn't need a bounce buffer then it better
1209 * not touch req->buffer either...
1211 req
->buffer
= bio_data(bio
);
1212 req
->current_nr_sectors
= bio_cur_sectors(bio
);
1213 req
->hard_cur_sectors
= req
->current_nr_sectors
;
1214 req
->sector
= req
->hard_sector
= bio
->bi_sector
;
1215 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
1216 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
1217 if (!blk_rq_cpu_valid(req
))
1218 req
->cpu
= bio
->bi_comp_cpu
;
1219 drive_stat_acct(req
, 0);
1220 if (!attempt_front_merge(q
, req
))
1221 elv_merged_request(q
, req
, el_ret
);
1224 /* ELV_NO_MERGE: elevator says don't/can't merge. */
1231 * This sync check and mask will be re-done in init_request_from_bio(),
1232 * but we need to set it earlier to expose the sync flag to the
1233 * rq allocator and io schedulers.
1235 rw_flags
= bio_data_dir(bio
);
1237 rw_flags
|= REQ_RW_SYNC
;
1240 * Grab a free request. This is might sleep but can not fail.
1241 * Returns with the queue unlocked.
1243 req
= get_request_wait(q
, rw_flags
, bio
);
1246 * After dropping the lock and possibly sleeping here, our request
1247 * may now be mergeable after it had proven unmergeable (above).
1248 * We don't worry about that case for efficiency. It won't happen
1249 * often, and the elevators are able to handle it.
1251 init_request_from_bio(req
, bio
);
1253 spin_lock_irq(q
->queue_lock
);
1254 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
) ||
1255 bio_flagged(bio
, BIO_CPU_AFFINE
))
1256 req
->cpu
= blk_cpu_to_group(smp_processor_id());
1257 if (elv_queue_empty(q
))
1259 add_request(q
, req
);
1262 __generic_unplug_device(q
);
1263 spin_unlock_irq(q
->queue_lock
);
1267 bio_endio(bio
, err
);
1272 * If bio->bi_dev is a partition, remap the location
1274 static inline void blk_partition_remap(struct bio
*bio
)
1276 struct block_device
*bdev
= bio
->bi_bdev
;
1278 if (bio_sectors(bio
) && bdev
!= bdev
->bd_contains
) {
1279 struct hd_struct
*p
= bdev
->bd_part
;
1281 bio
->bi_sector
+= p
->start_sect
;
1282 bio
->bi_bdev
= bdev
->bd_contains
;
1284 trace_block_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
1285 bdev
->bd_dev
, bio
->bi_sector
,
1286 bio
->bi_sector
- p
->start_sect
);
1290 static void handle_bad_sector(struct bio
*bio
)
1292 char b
[BDEVNAME_SIZE
];
1294 printk(KERN_INFO
"attempt to access beyond end of device\n");
1295 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
1296 bdevname(bio
->bi_bdev
, b
),
1298 (unsigned long long)bio
->bi_sector
+ bio_sectors(bio
),
1299 (long long)(bio
->bi_bdev
->bd_inode
->i_size
>> 9));
1301 set_bit(BIO_EOF
, &bio
->bi_flags
);
1304 #ifdef CONFIG_FAIL_MAKE_REQUEST
1306 static DECLARE_FAULT_ATTR(fail_make_request
);
1308 static int __init
setup_fail_make_request(char *str
)
1310 return setup_fault_attr(&fail_make_request
, str
);
1312 __setup("fail_make_request=", setup_fail_make_request
);
1314 static int should_fail_request(struct bio
*bio
)
1316 struct hd_struct
*part
= bio
->bi_bdev
->bd_part
;
1318 if (part_to_disk(part
)->part0
.make_it_fail
|| part
->make_it_fail
)
1319 return should_fail(&fail_make_request
, bio
->bi_size
);
1324 static int __init
fail_make_request_debugfs(void)
1326 return init_fault_attr_dentries(&fail_make_request
,
1327 "fail_make_request");
1330 late_initcall(fail_make_request_debugfs
);
1332 #else /* CONFIG_FAIL_MAKE_REQUEST */
1334 static inline int should_fail_request(struct bio
*bio
)
1339 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1342 * Check whether this bio extends beyond the end of the device.
1344 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
1351 /* Test device or partition size, when known. */
1352 maxsector
= bio
->bi_bdev
->bd_inode
->i_size
>> 9;
1354 sector_t sector
= bio
->bi_sector
;
1356 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
1358 * This may well happen - the kernel calls bread()
1359 * without checking the size of the device, e.g., when
1360 * mounting a device.
1362 handle_bad_sector(bio
);
1371 * generic_make_request - hand a buffer to its device driver for I/O
1372 * @bio: The bio describing the location in memory and on the device.
1374 * generic_make_request() is used to make I/O requests of block
1375 * devices. It is passed a &struct bio, which describes the I/O that needs
1378 * generic_make_request() does not return any status. The
1379 * success/failure status of the request, along with notification of
1380 * completion, is delivered asynchronously through the bio->bi_end_io
1381 * function described (one day) else where.
1383 * The caller of generic_make_request must make sure that bi_io_vec
1384 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1385 * set to describe the device address, and the
1386 * bi_end_io and optionally bi_private are set to describe how
1387 * completion notification should be signaled.
1389 * generic_make_request and the drivers it calls may use bi_next if this
1390 * bio happens to be merged with someone else, and may change bi_dev and
1391 * bi_sector for remaps as it sees fit. So the values of these fields
1392 * should NOT be depended on after the call to generic_make_request.
1394 static inline void __generic_make_request(struct bio
*bio
)
1396 struct request_queue
*q
;
1397 sector_t old_sector
;
1398 int ret
, nr_sectors
= bio_sectors(bio
);
1404 if (bio_check_eod(bio
, nr_sectors
))
1408 * Resolve the mapping until finished. (drivers are
1409 * still free to implement/resolve their own stacking
1410 * by explicitly returning 0)
1412 * NOTE: we don't repeat the blk_size check for each new device.
1413 * Stacking drivers are expected to know what they are doing.
1418 char b
[BDEVNAME_SIZE
];
1420 q
= bdev_get_queue(bio
->bi_bdev
);
1423 "generic_make_request: Trying to access "
1424 "nonexistent block-device %s (%Lu)\n",
1425 bdevname(bio
->bi_bdev
, b
),
1426 (long long) bio
->bi_sector
);
1428 bio_endio(bio
, err
);
1432 if (unlikely(nr_sectors
> q
->max_hw_sectors
)) {
1433 printk(KERN_ERR
"bio too big device %s (%u > %u)\n",
1434 bdevname(bio
->bi_bdev
, b
),
1440 if (unlikely(test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)))
1443 if (should_fail_request(bio
))
1447 * If this device has partitions, remap block n
1448 * of partition p to block n+start(p) of the disk.
1450 blk_partition_remap(bio
);
1452 if (bio_integrity_enabled(bio
) && bio_integrity_prep(bio
))
1455 if (old_sector
!= -1)
1456 trace_block_remap(q
, bio
, old_dev
, bio
->bi_sector
,
1459 trace_block_bio_queue(q
, bio
);
1461 old_sector
= bio
->bi_sector
;
1462 old_dev
= bio
->bi_bdev
->bd_dev
;
1464 if (bio_check_eod(bio
, nr_sectors
))
1466 if ((bio_empty_barrier(bio
) && !q
->prepare_flush_fn
) ||
1467 (bio_discard(bio
) && !q
->prepare_discard_fn
)) {
1472 ret
= q
->make_request_fn(q
, bio
);
1477 * We only want one ->make_request_fn to be active at a time,
1478 * else stack usage with stacked devices could be a problem.
1479 * So use current->bio_{list,tail} to keep a list of requests
1480 * submited by a make_request_fn function.
1481 * current->bio_tail is also used as a flag to say if
1482 * generic_make_request is currently active in this task or not.
1483 * If it is NULL, then no make_request is active. If it is non-NULL,
1484 * then a make_request is active, and new requests should be added
1487 void generic_make_request(struct bio
*bio
)
1489 if (current
->bio_tail
) {
1490 /* make_request is active */
1491 *(current
->bio_tail
) = bio
;
1492 bio
->bi_next
= NULL
;
1493 current
->bio_tail
= &bio
->bi_next
;
1496 /* following loop may be a bit non-obvious, and so deserves some
1498 * Before entering the loop, bio->bi_next is NULL (as all callers
1499 * ensure that) so we have a list with a single bio.
1500 * We pretend that we have just taken it off a longer list, so
1501 * we assign bio_list to the next (which is NULL) and bio_tail
1502 * to &bio_list, thus initialising the bio_list of new bios to be
1503 * added. __generic_make_request may indeed add some more bios
1504 * through a recursive call to generic_make_request. If it
1505 * did, we find a non-NULL value in bio_list and re-enter the loop
1506 * from the top. In this case we really did just take the bio
1507 * of the top of the list (no pretending) and so fixup bio_list and
1508 * bio_tail or bi_next, and call into __generic_make_request again.
1510 * The loop was structured like this to make only one call to
1511 * __generic_make_request (which is important as it is large and
1512 * inlined) and to keep the structure simple.
1514 BUG_ON(bio
->bi_next
);
1516 current
->bio_list
= bio
->bi_next
;
1517 if (bio
->bi_next
== NULL
)
1518 current
->bio_tail
= ¤t
->bio_list
;
1520 bio
->bi_next
= NULL
;
1521 __generic_make_request(bio
);
1522 bio
= current
->bio_list
;
1524 current
->bio_tail
= NULL
; /* deactivate */
1526 EXPORT_SYMBOL(generic_make_request
);
1529 * submit_bio - submit a bio to the block device layer for I/O
1530 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1531 * @bio: The &struct bio which describes the I/O
1533 * submit_bio() is very similar in purpose to generic_make_request(), and
1534 * uses that function to do most of the work. Both are fairly rough
1535 * interfaces; @bio must be presetup and ready for I/O.
1538 void submit_bio(int rw
, struct bio
*bio
)
1540 int count
= bio_sectors(bio
);
1545 * If it's a regular read/write or a barrier with data attached,
1546 * go through the normal accounting stuff before submission.
1548 if (bio_has_data(bio
)) {
1550 count_vm_events(PGPGOUT
, count
);
1552 task_io_account_read(bio
->bi_size
);
1553 count_vm_events(PGPGIN
, count
);
1556 if (unlikely(block_dump
)) {
1557 char b
[BDEVNAME_SIZE
];
1558 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s\n",
1559 current
->comm
, task_pid_nr(current
),
1560 (rw
& WRITE
) ? "WRITE" : "READ",
1561 (unsigned long long)bio
->bi_sector
,
1562 bdevname(bio
->bi_bdev
, b
));
1566 generic_make_request(bio
);
1568 EXPORT_SYMBOL(submit_bio
);
1571 * blk_rq_check_limits - Helper function to check a request for the queue limit
1573 * @rq: the request being checked
1576 * @rq may have been made based on weaker limitations of upper-level queues
1577 * in request stacking drivers, and it may violate the limitation of @q.
1578 * Since the block layer and the underlying device driver trust @rq
1579 * after it is inserted to @q, it should be checked against @q before
1580 * the insertion using this generic function.
1582 * This function should also be useful for request stacking drivers
1583 * in some cases below, so export this fuction.
1584 * Request stacking drivers like request-based dm may change the queue
1585 * limits while requests are in the queue (e.g. dm's table swapping).
1586 * Such request stacking drivers should check those requests agaist
1587 * the new queue limits again when they dispatch those requests,
1588 * although such checkings are also done against the old queue limits
1589 * when submitting requests.
1591 int blk_rq_check_limits(struct request_queue
*q
, struct request
*rq
)
1593 if (rq
->nr_sectors
> q
->max_sectors
||
1594 rq
->data_len
> q
->max_hw_sectors
<< 9) {
1595 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
1600 * queue's settings related to segment counting like q->bounce_pfn
1601 * may differ from that of other stacking queues.
1602 * Recalculate it to check the request correctly on this queue's
1605 blk_recalc_rq_segments(rq
);
1606 if (rq
->nr_phys_segments
> q
->max_phys_segments
||
1607 rq
->nr_phys_segments
> q
->max_hw_segments
) {
1608 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
1614 EXPORT_SYMBOL_GPL(blk_rq_check_limits
);
1617 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1618 * @q: the queue to submit the request
1619 * @rq: the request being queued
1621 int blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
1623 unsigned long flags
;
1625 if (blk_rq_check_limits(q
, rq
))
1628 #ifdef CONFIG_FAIL_MAKE_REQUEST
1629 if (rq
->rq_disk
&& rq
->rq_disk
->part0
.make_it_fail
&&
1630 should_fail(&fail_make_request
, blk_rq_bytes(rq
)))
1634 spin_lock_irqsave(q
->queue_lock
, flags
);
1637 * Submitting request must be dequeued before calling this function
1638 * because it will be linked to another request_queue
1640 BUG_ON(blk_queued_rq(rq
));
1642 drive_stat_acct(rq
, 1);
1643 __elv_add_request(q
, rq
, ELEVATOR_INSERT_BACK
, 0);
1645 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1649 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
1652 * blkdev_dequeue_request - dequeue request and start timeout timer
1653 * @req: request to dequeue
1655 * Dequeue @req and start timeout timer on it. This hands off the
1656 * request to the driver.
1658 * Block internal functions which don't want to start timer should
1659 * call elv_dequeue_request().
1661 void blkdev_dequeue_request(struct request
*req
)
1663 elv_dequeue_request(req
->q
, req
);
1666 * We are now handing the request to the hardware, add the
1671 EXPORT_SYMBOL(blkdev_dequeue_request
);
1674 * __end_that_request_first - end I/O on a request
1675 * @req: the request being processed
1676 * @error: %0 for success, < %0 for error
1677 * @nr_bytes: number of bytes to complete
1680 * Ends I/O on a number of bytes attached to @req, and sets it up
1681 * for the next range of segments (if any) in the cluster.
1684 * %0 - we are done with this request, call end_that_request_last()
1685 * %1 - still buffers pending for this request
1687 static int __end_that_request_first(struct request
*req
, int error
,
1690 int total_bytes
, bio_nbytes
, next_idx
= 0;
1693 trace_block_rq_complete(req
->q
, req
);
1696 * for a REQ_TYPE_BLOCK_PC request, we want to carry any eventual
1697 * sense key with us all the way through
1699 if (!blk_pc_request(req
))
1702 if (error
&& (blk_fs_request(req
) && !(req
->cmd_flags
& REQ_QUIET
))) {
1703 printk(KERN_ERR
"end_request: I/O error, dev %s, sector %llu\n",
1704 req
->rq_disk
? req
->rq_disk
->disk_name
: "?",
1705 (unsigned long long)req
->sector
);
1708 if (blk_fs_request(req
) && req
->rq_disk
) {
1709 const int rw
= rq_data_dir(req
);
1710 struct hd_struct
*part
;
1713 cpu
= part_stat_lock();
1714 part
= disk_map_sector_rcu(req
->rq_disk
, req
->sector
);
1715 part_stat_add(cpu
, part
, sectors
[rw
], nr_bytes
>> 9);
1719 total_bytes
= bio_nbytes
= 0;
1720 while ((bio
= req
->bio
) != NULL
) {
1724 * For an empty barrier request, the low level driver must
1725 * store a potential error location in ->sector. We pass
1726 * that back up in ->bi_sector.
1728 if (blk_empty_barrier(req
))
1729 bio
->bi_sector
= req
->sector
;
1731 if (nr_bytes
>= bio
->bi_size
) {
1732 req
->bio
= bio
->bi_next
;
1733 nbytes
= bio
->bi_size
;
1734 req_bio_endio(req
, bio
, nbytes
, error
);
1738 int idx
= bio
->bi_idx
+ next_idx
;
1740 if (unlikely(bio
->bi_idx
>= bio
->bi_vcnt
)) {
1741 blk_dump_rq_flags(req
, "__end_that");
1742 printk(KERN_ERR
"%s: bio idx %d >= vcnt %d\n",
1743 __func__
, bio
->bi_idx
, bio
->bi_vcnt
);
1747 nbytes
= bio_iovec_idx(bio
, idx
)->bv_len
;
1748 BIO_BUG_ON(nbytes
> bio
->bi_size
);
1751 * not a complete bvec done
1753 if (unlikely(nbytes
> nr_bytes
)) {
1754 bio_nbytes
+= nr_bytes
;
1755 total_bytes
+= nr_bytes
;
1760 * advance to the next vector
1763 bio_nbytes
+= nbytes
;
1766 total_bytes
+= nbytes
;
1772 * end more in this run, or just return 'not-done'
1774 if (unlikely(nr_bytes
<= 0))
1786 * if the request wasn't completed, update state
1789 req_bio_endio(req
, bio
, bio_nbytes
, error
);
1790 bio
->bi_idx
+= next_idx
;
1791 bio_iovec(bio
)->bv_offset
+= nr_bytes
;
1792 bio_iovec(bio
)->bv_len
-= nr_bytes
;
1795 blk_recalc_rq_sectors(req
, total_bytes
>> 9);
1796 blk_recalc_rq_segments(req
);
1801 * queue lock must be held
1803 static void end_that_request_last(struct request
*req
, int error
)
1805 struct gendisk
*disk
= req
->rq_disk
;
1807 if (blk_rq_tagged(req
))
1808 blk_queue_end_tag(req
->q
, req
);
1810 if (blk_queued_rq(req
))
1811 elv_dequeue_request(req
->q
, req
);
1813 if (unlikely(laptop_mode
) && blk_fs_request(req
))
1814 laptop_io_completion();
1816 blk_delete_timer(req
);
1819 * Account IO completion. bar_rq isn't accounted as a normal
1820 * IO on queueing nor completion. Accounting the containing
1821 * request is enough.
1823 if (disk
&& blk_fs_request(req
) && req
!= &req
->q
->bar_rq
) {
1824 unsigned long duration
= jiffies
- req
->start_time
;
1825 const int rw
= rq_data_dir(req
);
1826 struct hd_struct
*part
;
1829 cpu
= part_stat_lock();
1830 part
= disk_map_sector_rcu(disk
, req
->sector
);
1832 part_stat_inc(cpu
, part
, ios
[rw
]);
1833 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
1834 part_round_stats(cpu
, part
);
1835 part_dec_in_flight(part
);
1841 req
->end_io(req
, error
);
1843 if (blk_bidi_rq(req
))
1844 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
1846 __blk_put_request(req
->q
, req
);
1851 * blk_rq_bytes - Returns bytes left to complete in the entire request
1852 * @rq: the request being processed
1854 unsigned int blk_rq_bytes(struct request
*rq
)
1856 if (blk_fs_request(rq
))
1857 return rq
->hard_nr_sectors
<< 9;
1859 return rq
->data_len
;
1861 EXPORT_SYMBOL_GPL(blk_rq_bytes
);
1864 * blk_rq_cur_bytes - Returns bytes left to complete in the current segment
1865 * @rq: the request being processed
1867 unsigned int blk_rq_cur_bytes(struct request
*rq
)
1869 if (blk_fs_request(rq
))
1870 return rq
->current_nr_sectors
<< 9;
1873 return rq
->bio
->bi_size
;
1875 return rq
->data_len
;
1877 EXPORT_SYMBOL_GPL(blk_rq_cur_bytes
);
1880 * end_request - end I/O on the current segment of the request
1881 * @req: the request being processed
1882 * @uptodate: error value or %0/%1 uptodate flag
1885 * Ends I/O on the current segment of a request. If that is the only
1886 * remaining segment, the request is also completed and freed.
1888 * This is a remnant of how older block drivers handled I/O completions.
1889 * Modern drivers typically end I/O on the full request in one go, unless
1890 * they have a residual value to account for. For that case this function
1891 * isn't really useful, unless the residual just happens to be the
1892 * full current segment. In other words, don't use this function in new
1893 * code. Use blk_end_request() or __blk_end_request() to end a request.
1895 void end_request(struct request
*req
, int uptodate
)
1900 error
= uptodate
? uptodate
: -EIO
;
1902 __blk_end_request(req
, error
, req
->hard_cur_sectors
<< 9);
1904 EXPORT_SYMBOL(end_request
);
1906 static int end_that_request_data(struct request
*rq
, int error
,
1907 unsigned int nr_bytes
, unsigned int bidi_bytes
)
1910 if (__end_that_request_first(rq
, error
, nr_bytes
))
1913 /* Bidi request must be completed as a whole */
1914 if (blk_bidi_rq(rq
) &&
1915 __end_that_request_first(rq
->next_rq
, error
, bidi_bytes
))
1923 * blk_end_io - Generic end_io function to complete a request.
1924 * @rq: the request being processed
1925 * @error: %0 for success, < %0 for error
1926 * @nr_bytes: number of bytes to complete @rq
1927 * @bidi_bytes: number of bytes to complete @rq->next_rq
1928 * @drv_callback: function called between completion of bios in the request
1929 * and completion of the request.
1930 * If the callback returns non %0, this helper returns without
1931 * completion of the request.
1934 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
1935 * If @rq has leftover, sets it up for the next range of segments.
1938 * %0 - we are done with this request
1939 * %1 - this request is not freed yet, it still has pending buffers.
1941 static int blk_end_io(struct request
*rq
, int error
, unsigned int nr_bytes
,
1942 unsigned int bidi_bytes
,
1943 int (drv_callback
)(struct request
*))
1945 struct request_queue
*q
= rq
->q
;
1946 unsigned long flags
= 0UL;
1948 if (end_that_request_data(rq
, error
, nr_bytes
, bidi_bytes
))
1951 /* Special feature for tricky drivers */
1952 if (drv_callback
&& drv_callback(rq
))
1955 add_disk_randomness(rq
->rq_disk
);
1957 spin_lock_irqsave(q
->queue_lock
, flags
);
1958 end_that_request_last(rq
, error
);
1959 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1965 * blk_end_request - Helper function for drivers to complete the request.
1966 * @rq: the request being processed
1967 * @error: %0 for success, < %0 for error
1968 * @nr_bytes: number of bytes to complete
1971 * Ends I/O on a number of bytes attached to @rq.
1972 * If @rq has leftover, sets it up for the next range of segments.
1975 * %0 - we are done with this request
1976 * %1 - still buffers pending for this request
1978 int blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
1980 return blk_end_io(rq
, error
, nr_bytes
, 0, NULL
);
1982 EXPORT_SYMBOL_GPL(blk_end_request
);
1985 * __blk_end_request - Helper function for drivers to complete the request.
1986 * @rq: the request being processed
1987 * @error: %0 for success, < %0 for error
1988 * @nr_bytes: number of bytes to complete
1991 * Must be called with queue lock held unlike blk_end_request().
1994 * %0 - we are done with this request
1995 * %1 - still buffers pending for this request
1997 int __blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
1999 if (rq
->bio
&& __end_that_request_first(rq
, error
, nr_bytes
))
2002 add_disk_randomness(rq
->rq_disk
);
2004 end_that_request_last(rq
, error
);
2008 EXPORT_SYMBOL_GPL(__blk_end_request
);
2011 * blk_end_bidi_request - Helper function for drivers to complete bidi request.
2012 * @rq: the bidi request being processed
2013 * @error: %0 for success, < %0 for error
2014 * @nr_bytes: number of bytes to complete @rq
2015 * @bidi_bytes: number of bytes to complete @rq->next_rq
2018 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2021 * %0 - we are done with this request
2022 * %1 - still buffers pending for this request
2024 int blk_end_bidi_request(struct request
*rq
, int error
, unsigned int nr_bytes
,
2025 unsigned int bidi_bytes
)
2027 return blk_end_io(rq
, error
, nr_bytes
, bidi_bytes
, NULL
);
2029 EXPORT_SYMBOL_GPL(blk_end_bidi_request
);
2032 * blk_update_request - Special helper function for request stacking drivers
2033 * @rq: the request being processed
2034 * @error: %0 for success, < %0 for error
2035 * @nr_bytes: number of bytes to complete @rq
2038 * Ends I/O on a number of bytes attached to @rq, but doesn't complete
2039 * the request structure even if @rq doesn't have leftover.
2040 * If @rq has leftover, sets it up for the next range of segments.
2042 * This special helper function is only for request stacking drivers
2043 * (e.g. request-based dm) so that they can handle partial completion.
2044 * Actual device drivers should use blk_end_request instead.
2046 void blk_update_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2048 if (!end_that_request_data(rq
, error
, nr_bytes
, 0)) {
2050 * These members are not updated in end_that_request_data()
2051 * when all bios are completed.
2052 * Update them so that the request stacking driver can find
2053 * how many bytes remain in the request later.
2055 rq
->nr_sectors
= rq
->hard_nr_sectors
= 0;
2056 rq
->current_nr_sectors
= rq
->hard_cur_sectors
= 0;
2059 EXPORT_SYMBOL_GPL(blk_update_request
);
2062 * blk_end_request_callback - Special helper function for tricky drivers
2063 * @rq: the request being processed
2064 * @error: %0 for success, < %0 for error
2065 * @nr_bytes: number of bytes to complete
2066 * @drv_callback: function called between completion of bios in the request
2067 * and completion of the request.
2068 * If the callback returns non %0, this helper returns without
2069 * completion of the request.
2072 * Ends I/O on a number of bytes attached to @rq.
2073 * If @rq has leftover, sets it up for the next range of segments.
2075 * This special helper function is used only for existing tricky drivers.
2076 * (e.g. cdrom_newpc_intr() of ide-cd)
2077 * This interface will be removed when such drivers are rewritten.
2078 * Don't use this interface in other places anymore.
2081 * %0 - we are done with this request
2082 * %1 - this request is not freed yet.
2083 * this request still has pending buffers or
2084 * the driver doesn't want to finish this request yet.
2086 int blk_end_request_callback(struct request
*rq
, int error
,
2087 unsigned int nr_bytes
,
2088 int (drv_callback
)(struct request
*))
2090 return blk_end_io(rq
, error
, nr_bytes
, 0, drv_callback
);
2092 EXPORT_SYMBOL_GPL(blk_end_request_callback
);
2094 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
2097 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw, and
2098 we want BIO_RW_AHEAD (bit 1) to imply REQ_FAILFAST (bit 1). */
2099 rq
->cmd_flags
|= (bio
->bi_rw
& 3);
2101 if (bio_has_data(bio
)) {
2102 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2103 rq
->buffer
= bio_data(bio
);
2105 rq
->current_nr_sectors
= bio_cur_sectors(bio
);
2106 rq
->hard_cur_sectors
= rq
->current_nr_sectors
;
2107 rq
->hard_nr_sectors
= rq
->nr_sectors
= bio_sectors(bio
);
2108 rq
->data_len
= bio
->bi_size
;
2110 rq
->bio
= rq
->biotail
= bio
;
2113 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2117 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2118 * @q : the queue of the device being checked
2121 * Check if underlying low-level drivers of a device are busy.
2122 * If the drivers want to export their busy state, they must set own
2123 * exporting function using blk_queue_lld_busy() first.
2125 * Basically, this function is used only by request stacking drivers
2126 * to stop dispatching requests to underlying devices when underlying
2127 * devices are busy. This behavior helps more I/O merging on the queue
2128 * of the request stacking driver and prevents I/O throughput regression
2129 * on burst I/O load.
2132 * 0 - Not busy (The request stacking driver should dispatch request)
2133 * 1 - Busy (The request stacking driver should stop dispatching request)
2135 int blk_lld_busy(struct request_queue
*q
)
2138 return q
->lld_busy_fn(q
);
2142 EXPORT_SYMBOL_GPL(blk_lld_busy
);
2144 int kblockd_schedule_work(struct request_queue
*q
, struct work_struct
*work
)
2146 return queue_work(kblockd_workqueue
, work
);
2148 EXPORT_SYMBOL(kblockd_schedule_work
);
2150 void kblockd_flush_work(struct work_struct
*work
)
2152 cancel_work_sync(work
);
2154 EXPORT_SYMBOL(kblockd_flush_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
);