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
)
329 * one level of recursion is ok and is much faster than kicking
330 * the unplug handling
332 if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER
, q
)) {
334 queue_flag_clear(QUEUE_FLAG_REENTER
, q
);
336 queue_flag_set(QUEUE_FLAG_PLUGGED
, q
);
337 kblockd_schedule_work(q
, &q
->unplug_work
);
342 * blk_start_queue - restart a previously stopped queue
343 * @q: The &struct request_queue in question
346 * blk_start_queue() will clear the stop flag on the queue, and call
347 * the request_fn for the queue if it was in a stopped state when
348 * entered. Also see blk_stop_queue(). Queue lock must be held.
350 void blk_start_queue(struct request_queue
*q
)
352 WARN_ON(!irqs_disabled());
354 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
355 blk_invoke_request_fn(q
);
357 EXPORT_SYMBOL(blk_start_queue
);
360 * blk_stop_queue - stop a queue
361 * @q: The &struct request_queue in question
364 * The Linux block layer assumes that a block driver will consume all
365 * entries on the request queue when the request_fn strategy is called.
366 * Often this will not happen, because of hardware limitations (queue
367 * depth settings). If a device driver gets a 'queue full' response,
368 * or if it simply chooses not to queue more I/O at one point, it can
369 * call this function to prevent the request_fn from being called until
370 * the driver has signalled it's ready to go again. This happens by calling
371 * blk_start_queue() to restart queue operations. Queue lock must be held.
373 void blk_stop_queue(struct request_queue
*q
)
376 queue_flag_set(QUEUE_FLAG_STOPPED
, q
);
378 EXPORT_SYMBOL(blk_stop_queue
);
381 * blk_sync_queue - cancel any pending callbacks on a queue
385 * The block layer may perform asynchronous callback activity
386 * on a queue, such as calling the unplug function after a timeout.
387 * A block device may call blk_sync_queue to ensure that any
388 * such activity is cancelled, thus allowing it to release resources
389 * that the callbacks might use. The caller must already have made sure
390 * that its ->make_request_fn will not re-add plugging prior to calling
394 void blk_sync_queue(struct request_queue
*q
)
396 del_timer_sync(&q
->unplug_timer
);
397 kblockd_flush_work(&q
->unplug_work
);
399 EXPORT_SYMBOL(blk_sync_queue
);
402 * blk_run_queue - run a single device queue
403 * @q: The queue to run
405 void __blk_run_queue(struct request_queue
*q
)
410 * Only recurse once to avoid overrunning the stack, let the unplug
411 * handling reinvoke the handler shortly if we already got there.
413 if (!elv_queue_empty(q
))
414 blk_invoke_request_fn(q
);
416 EXPORT_SYMBOL(__blk_run_queue
);
419 * blk_run_queue - run a single device queue
420 * @q: The queue to run
422 void blk_run_queue(struct request_queue
*q
)
426 spin_lock_irqsave(q
->queue_lock
, flags
);
428 spin_unlock_irqrestore(q
->queue_lock
, flags
);
430 EXPORT_SYMBOL(blk_run_queue
);
432 void blk_put_queue(struct request_queue
*q
)
434 kobject_put(&q
->kobj
);
437 void blk_cleanup_queue(struct request_queue
*q
)
440 * We know we have process context here, so we can be a little
441 * cautious and ensure that pending block actions on this device
442 * are done before moving on. Going into this function, we should
443 * not have processes doing IO to this device.
447 mutex_lock(&q
->sysfs_lock
);
448 queue_flag_set_unlocked(QUEUE_FLAG_DEAD
, q
);
449 mutex_unlock(&q
->sysfs_lock
);
452 elevator_exit(q
->elevator
);
456 EXPORT_SYMBOL(blk_cleanup_queue
);
458 static int blk_init_free_list(struct request_queue
*q
)
460 struct request_list
*rl
= &q
->rq
;
462 rl
->count
[READ
] = rl
->count
[WRITE
] = 0;
463 rl
->starved
[READ
] = rl
->starved
[WRITE
] = 0;
465 init_waitqueue_head(&rl
->wait
[READ
]);
466 init_waitqueue_head(&rl
->wait
[WRITE
]);
468 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, mempool_alloc_slab
,
469 mempool_free_slab
, request_cachep
, q
->node
);
477 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
479 return blk_alloc_queue_node(gfp_mask
, -1);
481 EXPORT_SYMBOL(blk_alloc_queue
);
483 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
485 struct request_queue
*q
;
488 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
489 gfp_mask
| __GFP_ZERO
, node_id
);
493 q
->backing_dev_info
.unplug_io_fn
= blk_backing_dev_unplug
;
494 q
->backing_dev_info
.unplug_io_data
= q
;
495 err
= bdi_init(&q
->backing_dev_info
);
497 kmem_cache_free(blk_requestq_cachep
, q
);
501 init_timer(&q
->unplug_timer
);
502 setup_timer(&q
->timeout
, blk_rq_timed_out_timer
, (unsigned long) q
);
503 INIT_LIST_HEAD(&q
->timeout_list
);
505 kobject_init(&q
->kobj
, &blk_queue_ktype
);
507 mutex_init(&q
->sysfs_lock
);
508 spin_lock_init(&q
->__queue_lock
);
512 EXPORT_SYMBOL(blk_alloc_queue_node
);
515 * blk_init_queue - prepare a request queue for use with a block device
516 * @rfn: The function to be called to process requests that have been
517 * placed on the queue.
518 * @lock: Request queue spin lock
521 * If a block device wishes to use the standard request handling procedures,
522 * which sorts requests and coalesces adjacent requests, then it must
523 * call blk_init_queue(). The function @rfn will be called when there
524 * are requests on the queue that need to be processed. If the device
525 * supports plugging, then @rfn may not be called immediately when requests
526 * are available on the queue, but may be called at some time later instead.
527 * Plugged queues are generally unplugged when a buffer belonging to one
528 * of the requests on the queue is needed, or due to memory pressure.
530 * @rfn is not required, or even expected, to remove all requests off the
531 * queue, but only as many as it can handle at a time. If it does leave
532 * requests on the queue, it is responsible for arranging that the requests
533 * get dealt with eventually.
535 * The queue spin lock must be held while manipulating the requests on the
536 * request queue; this lock will be taken also from interrupt context, so irq
537 * disabling is needed for it.
539 * Function returns a pointer to the initialized request queue, or %NULL if
543 * blk_init_queue() must be paired with a blk_cleanup_queue() call
544 * when the block device is deactivated (such as at module unload).
547 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
549 return blk_init_queue_node(rfn
, lock
, -1);
551 EXPORT_SYMBOL(blk_init_queue
);
553 struct request_queue
*
554 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
556 struct request_queue
*q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
562 if (blk_init_free_list(q
)) {
563 kmem_cache_free(blk_requestq_cachep
, q
);
568 * if caller didn't supply a lock, they get per-queue locking with
572 lock
= &q
->__queue_lock
;
575 q
->prep_rq_fn
= NULL
;
576 q
->unplug_fn
= generic_unplug_device
;
577 q
->queue_flags
= (1 << QUEUE_FLAG_CLUSTER
|
578 1 << QUEUE_FLAG_STACKABLE
);
579 q
->queue_lock
= lock
;
581 blk_queue_segment_boundary(q
, 0xffffffff);
583 blk_queue_make_request(q
, __make_request
);
584 blk_queue_max_segment_size(q
, MAX_SEGMENT_SIZE
);
586 blk_queue_max_hw_segments(q
, MAX_HW_SEGMENTS
);
587 blk_queue_max_phys_segments(q
, MAX_PHYS_SEGMENTS
);
589 q
->sg_reserved_size
= INT_MAX
;
591 blk_set_cmd_filter_defaults(&q
->cmd_filter
);
596 if (!elevator_init(q
, NULL
)) {
597 blk_queue_congestion_threshold(q
);
604 EXPORT_SYMBOL(blk_init_queue_node
);
606 int blk_get_queue(struct request_queue
*q
)
608 if (likely(!test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
))) {
609 kobject_get(&q
->kobj
);
616 static inline void blk_free_request(struct request_queue
*q
, struct request
*rq
)
618 if (rq
->cmd_flags
& REQ_ELVPRIV
)
619 elv_put_request(q
, rq
);
620 mempool_free(rq
, q
->rq
.rq_pool
);
623 static struct request
*
624 blk_alloc_request(struct request_queue
*q
, int rw
, int priv
, gfp_t gfp_mask
)
626 struct request
*rq
= mempool_alloc(q
->rq
.rq_pool
, gfp_mask
);
633 rq
->cmd_flags
= rw
| REQ_ALLOCED
;
636 if (unlikely(elv_set_request(q
, rq
, gfp_mask
))) {
637 mempool_free(rq
, q
->rq
.rq_pool
);
640 rq
->cmd_flags
|= REQ_ELVPRIV
;
647 * ioc_batching returns true if the ioc is a valid batching request and
648 * should be given priority access to a request.
650 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
656 * Make sure the process is able to allocate at least 1 request
657 * even if the batch times out, otherwise we could theoretically
660 return ioc
->nr_batch_requests
== q
->nr_batching
||
661 (ioc
->nr_batch_requests
> 0
662 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
666 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
667 * will cause the process to be a "batcher" on all queues in the system. This
668 * is the behaviour we want though - once it gets a wakeup it should be given
671 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
673 if (!ioc
|| ioc_batching(q
, ioc
))
676 ioc
->nr_batch_requests
= q
->nr_batching
;
677 ioc
->last_waited
= jiffies
;
680 static void __freed_request(struct request_queue
*q
, int rw
)
682 struct request_list
*rl
= &q
->rq
;
684 if (rl
->count
[rw
] < queue_congestion_off_threshold(q
))
685 blk_clear_queue_congested(q
, rw
);
687 if (rl
->count
[rw
] + 1 <= q
->nr_requests
) {
688 if (waitqueue_active(&rl
->wait
[rw
]))
689 wake_up(&rl
->wait
[rw
]);
691 blk_clear_queue_full(q
, rw
);
696 * A request has just been released. Account for it, update the full and
697 * congestion status, wake up any waiters. Called under q->queue_lock.
699 static void freed_request(struct request_queue
*q
, int rw
, int priv
)
701 struct request_list
*rl
= &q
->rq
;
707 __freed_request(q
, rw
);
709 if (unlikely(rl
->starved
[rw
^ 1]))
710 __freed_request(q
, rw
^ 1);
713 #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
715 * Get a free request, queue_lock must be held.
716 * Returns NULL on failure, with queue_lock held.
717 * Returns !NULL on success, with queue_lock *not held*.
719 static struct request
*get_request(struct request_queue
*q
, int rw_flags
,
720 struct bio
*bio
, gfp_t gfp_mask
)
722 struct request
*rq
= NULL
;
723 struct request_list
*rl
= &q
->rq
;
724 struct io_context
*ioc
= NULL
;
725 const int rw
= rw_flags
& 0x01;
728 may_queue
= elv_may_queue(q
, rw_flags
);
729 if (may_queue
== ELV_MQUEUE_NO
)
732 if (rl
->count
[rw
]+1 >= queue_congestion_on_threshold(q
)) {
733 if (rl
->count
[rw
]+1 >= q
->nr_requests
) {
734 ioc
= current_io_context(GFP_ATOMIC
, q
->node
);
736 * The queue will fill after this allocation, so set
737 * it as full, and mark this process as "batching".
738 * This process will be allowed to complete a batch of
739 * requests, others will be blocked.
741 if (!blk_queue_full(q
, rw
)) {
742 ioc_set_batching(q
, ioc
);
743 blk_set_queue_full(q
, rw
);
745 if (may_queue
!= ELV_MQUEUE_MUST
746 && !ioc_batching(q
, ioc
)) {
748 * The queue is full and the allocating
749 * process is not a "batcher", and not
750 * exempted by the IO scheduler
756 blk_set_queue_congested(q
, rw
);
760 * Only allow batching queuers to allocate up to 50% over the defined
761 * limit of requests, otherwise we could have thousands of requests
762 * allocated with any setting of ->nr_requests
764 if (rl
->count
[rw
] >= (3 * q
->nr_requests
/ 2))
770 priv
= !test_bit(QUEUE_FLAG_ELVSWITCH
, &q
->queue_flags
);
774 spin_unlock_irq(q
->queue_lock
);
776 rq
= blk_alloc_request(q
, rw_flags
, priv
, gfp_mask
);
779 * Allocation failed presumably due to memory. Undo anything
780 * we might have messed up.
782 * Allocating task should really be put onto the front of the
783 * wait queue, but this is pretty rare.
785 spin_lock_irq(q
->queue_lock
);
786 freed_request(q
, rw
, priv
);
789 * in the very unlikely event that allocation failed and no
790 * requests for this direction was pending, mark us starved
791 * so that freeing of a request in the other direction will
792 * notice us. another possible fix would be to split the
793 * rq mempool into READ and WRITE
796 if (unlikely(rl
->count
[rw
] == 0))
803 * ioc may be NULL here, and ioc_batching will be false. That's
804 * OK, if the queue is under the request limit then requests need
805 * not count toward the nr_batch_requests limit. There will always
806 * be some limit enforced by BLK_BATCH_TIME.
808 if (ioc_batching(q
, ioc
))
809 ioc
->nr_batch_requests
--;
811 blk_add_trace_generic(q
, bio
, rw
, BLK_TA_GETRQ
);
817 * No available requests for this queue, unplug the device and wait for some
818 * requests to become available.
820 * Called with q->queue_lock held, and returns with it unlocked.
822 static struct request
*get_request_wait(struct request_queue
*q
, int rw_flags
,
825 const int rw
= rw_flags
& 0x01;
828 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
831 struct io_context
*ioc
;
832 struct request_list
*rl
= &q
->rq
;
834 prepare_to_wait_exclusive(&rl
->wait
[rw
], &wait
,
835 TASK_UNINTERRUPTIBLE
);
837 blk_add_trace_generic(q
, bio
, rw
, BLK_TA_SLEEPRQ
);
839 __generic_unplug_device(q
);
840 spin_unlock_irq(q
->queue_lock
);
844 * After sleeping, we become a "batching" process and
845 * will be able to allocate at least one request, and
846 * up to a big batch of them for a small period time.
847 * See ioc_batching, ioc_set_batching
849 ioc
= current_io_context(GFP_NOIO
, q
->node
);
850 ioc_set_batching(q
, ioc
);
852 spin_lock_irq(q
->queue_lock
);
853 finish_wait(&rl
->wait
[rw
], &wait
);
855 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
861 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
865 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
867 spin_lock_irq(q
->queue_lock
);
868 if (gfp_mask
& __GFP_WAIT
) {
869 rq
= get_request_wait(q
, rw
, NULL
);
871 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
873 spin_unlock_irq(q
->queue_lock
);
875 /* q->queue_lock is unlocked at this point */
879 EXPORT_SYMBOL(blk_get_request
);
882 * blk_start_queueing - initiate dispatch of requests to device
883 * @q: request queue to kick into gear
885 * This is basically a helper to remove the need to know whether a queue
886 * is plugged or not if someone just wants to initiate dispatch of requests
889 * The queue lock must be held with interrupts disabled.
891 void blk_start_queueing(struct request_queue
*q
)
893 if (!blk_queue_plugged(q
))
896 __generic_unplug_device(q
);
898 EXPORT_SYMBOL(blk_start_queueing
);
901 * blk_requeue_request - put a request back on queue
902 * @q: request queue where request should be inserted
903 * @rq: request to be inserted
906 * Drivers often keep queueing requests until the hardware cannot accept
907 * more, when that condition happens we need to put the request back
908 * on the queue. Must be called with queue lock held.
910 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
912 blk_delete_timer(rq
);
913 blk_clear_rq_complete(rq
);
914 blk_add_trace_rq(q
, rq
, BLK_TA_REQUEUE
);
916 if (blk_rq_tagged(rq
))
917 blk_queue_end_tag(q
, rq
);
919 elv_requeue_request(q
, rq
);
921 EXPORT_SYMBOL(blk_requeue_request
);
924 * blk_insert_request - insert a special request into a request queue
925 * @q: request queue where request should be inserted
926 * @rq: request to be inserted
927 * @at_head: insert request at head or tail of queue
928 * @data: private data
931 * Many block devices need to execute commands asynchronously, so they don't
932 * block the whole kernel from preemption during request execution. This is
933 * accomplished normally by inserting aritficial requests tagged as
934 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
935 * be scheduled for actual execution by the request queue.
937 * We have the option of inserting the head or the tail of the queue.
938 * Typically we use the tail for new ioctls and so forth. We use the head
939 * of the queue for things like a QUEUE_FULL message from a device, or a
940 * host that is unable to accept a particular command.
942 void blk_insert_request(struct request_queue
*q
, struct request
*rq
,
943 int at_head
, void *data
)
945 int where
= at_head
? ELEVATOR_INSERT_FRONT
: ELEVATOR_INSERT_BACK
;
949 * tell I/O scheduler that this isn't a regular read/write (ie it
950 * must not attempt merges on this) and that it acts as a soft
953 rq
->cmd_type
= REQ_TYPE_SPECIAL
;
954 rq
->cmd_flags
|= REQ_SOFTBARRIER
;
958 spin_lock_irqsave(q
->queue_lock
, flags
);
961 * If command is tagged, release the tag
963 if (blk_rq_tagged(rq
))
964 blk_queue_end_tag(q
, rq
);
966 drive_stat_acct(rq
, 1);
967 __elv_add_request(q
, rq
, where
, 0);
968 blk_start_queueing(q
);
969 spin_unlock_irqrestore(q
->queue_lock
, flags
);
971 EXPORT_SYMBOL(blk_insert_request
);
974 * add-request adds a request to the linked list.
975 * queue lock is held and interrupts disabled, as we muck with the
976 * request queue list.
978 static inline void add_request(struct request_queue
*q
, struct request
*req
)
980 drive_stat_acct(req
, 1);
983 * elevator indicated where it wants this request to be
984 * inserted at elevator_merge time
986 __elv_add_request(q
, req
, ELEVATOR_INSERT_SORT
, 0);
989 static void part_round_stats_single(int cpu
, struct hd_struct
*part
,
992 if (now
== part
->stamp
)
995 if (part
->in_flight
) {
996 __part_stat_add(cpu
, part
, time_in_queue
,
997 part
->in_flight
* (now
- part
->stamp
));
998 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
1004 * part_round_stats() - Round off the performance stats on a struct
1007 * The average IO queue length and utilisation statistics are maintained
1008 * by observing the current state of the queue length and the amount of
1009 * time it has been in this state for.
1011 * Normally, that accounting is done on IO completion, but that can result
1012 * in more than a second's worth of IO being accounted for within any one
1013 * second, leading to >100% utilisation. To deal with that, we call this
1014 * function to do a round-off before returning the results when reading
1015 * /proc/diskstats. This accounts immediately for all queue usage up to
1016 * the current jiffies and restarts the counters again.
1018 void part_round_stats(int cpu
, struct hd_struct
*part
)
1020 unsigned long now
= jiffies
;
1023 part_round_stats_single(cpu
, &part_to_disk(part
)->part0
, now
);
1024 part_round_stats_single(cpu
, part
, now
);
1026 EXPORT_SYMBOL_GPL(part_round_stats
);
1029 * queue lock must be held
1031 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1035 if (unlikely(--req
->ref_count
))
1038 elv_completed_request(q
, req
);
1041 * Request may not have originated from ll_rw_blk. if not,
1042 * it didn't come out of our reserved rq pools
1044 if (req
->cmd_flags
& REQ_ALLOCED
) {
1045 int rw
= rq_data_dir(req
);
1046 int priv
= req
->cmd_flags
& REQ_ELVPRIV
;
1048 BUG_ON(!list_empty(&req
->queuelist
));
1049 BUG_ON(!hlist_unhashed(&req
->hash
));
1051 blk_free_request(q
, req
);
1052 freed_request(q
, rw
, priv
);
1055 EXPORT_SYMBOL_GPL(__blk_put_request
);
1057 void blk_put_request(struct request
*req
)
1059 unsigned long flags
;
1060 struct request_queue
*q
= req
->q
;
1062 spin_lock_irqsave(q
->queue_lock
, flags
);
1063 __blk_put_request(q
, req
);
1064 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1066 EXPORT_SYMBOL(blk_put_request
);
1068 void init_request_from_bio(struct request
*req
, struct bio
*bio
)
1070 req
->cpu
= bio
->bi_comp_cpu
;
1071 req
->cmd_type
= REQ_TYPE_FS
;
1074 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
1076 if (bio_rw_ahead(bio
) || bio_failfast(bio
))
1077 req
->cmd_flags
|= REQ_FAILFAST
;
1080 * REQ_BARRIER implies no merging, but lets make it explicit
1082 if (unlikely(bio_discard(bio
))) {
1083 req
->cmd_flags
|= REQ_DISCARD
;
1084 if (bio_barrier(bio
))
1085 req
->cmd_flags
|= REQ_SOFTBARRIER
;
1086 req
->q
->prepare_discard_fn(req
->q
, req
);
1087 } else if (unlikely(bio_barrier(bio
)))
1088 req
->cmd_flags
|= (REQ_HARDBARRIER
| REQ_NOMERGE
);
1091 req
->cmd_flags
|= REQ_RW_SYNC
;
1092 if (bio_rw_meta(bio
))
1093 req
->cmd_flags
|= REQ_RW_META
;
1096 req
->hard_sector
= req
->sector
= bio
->bi_sector
;
1097 req
->ioprio
= bio_prio(bio
);
1098 req
->start_time
= jiffies
;
1099 blk_rq_bio_prep(req
->q
, req
, bio
);
1102 static int __make_request(struct request_queue
*q
, struct bio
*bio
)
1104 struct request
*req
;
1105 int el_ret
, nr_sectors
, barrier
, discard
, err
;
1106 const unsigned short prio
= bio_prio(bio
);
1107 const int sync
= bio_sync(bio
);
1110 nr_sectors
= bio_sectors(bio
);
1113 * low level driver can indicate that it wants pages above a
1114 * certain limit bounced to low memory (ie for highmem, or even
1115 * ISA dma in theory)
1117 blk_queue_bounce(q
, &bio
);
1119 barrier
= bio_barrier(bio
);
1120 if (unlikely(barrier
) && bio_has_data(bio
) &&
1121 (q
->next_ordered
== QUEUE_ORDERED_NONE
)) {
1126 discard
= bio_discard(bio
);
1127 if (unlikely(discard
) && !q
->prepare_discard_fn
) {
1132 spin_lock_irq(q
->queue_lock
);
1134 if (unlikely(barrier
) || elv_queue_empty(q
))
1137 el_ret
= elv_merge(q
, &req
, bio
);
1139 case ELEVATOR_BACK_MERGE
:
1140 BUG_ON(!rq_mergeable(req
));
1142 if (!ll_back_merge_fn(q
, req
, bio
))
1145 blk_add_trace_bio(q
, bio
, BLK_TA_BACKMERGE
);
1147 req
->biotail
->bi_next
= bio
;
1149 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
1150 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
1151 if (!blk_rq_cpu_valid(req
))
1152 req
->cpu
= bio
->bi_comp_cpu
;
1153 drive_stat_acct(req
, 0);
1154 if (!attempt_back_merge(q
, req
))
1155 elv_merged_request(q
, req
, el_ret
);
1158 case ELEVATOR_FRONT_MERGE
:
1159 BUG_ON(!rq_mergeable(req
));
1161 if (!ll_front_merge_fn(q
, req
, bio
))
1164 blk_add_trace_bio(q
, bio
, BLK_TA_FRONTMERGE
);
1166 bio
->bi_next
= req
->bio
;
1170 * may not be valid. if the low level driver said
1171 * it didn't need a bounce buffer then it better
1172 * not touch req->buffer either...
1174 req
->buffer
= bio_data(bio
);
1175 req
->current_nr_sectors
= bio_cur_sectors(bio
);
1176 req
->hard_cur_sectors
= req
->current_nr_sectors
;
1177 req
->sector
= req
->hard_sector
= bio
->bi_sector
;
1178 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
1179 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
1180 if (!blk_rq_cpu_valid(req
))
1181 req
->cpu
= bio
->bi_comp_cpu
;
1182 drive_stat_acct(req
, 0);
1183 if (!attempt_front_merge(q
, req
))
1184 elv_merged_request(q
, req
, el_ret
);
1187 /* ELV_NO_MERGE: elevator says don't/can't merge. */
1194 * This sync check and mask will be re-done in init_request_from_bio(),
1195 * but we need to set it earlier to expose the sync flag to the
1196 * rq allocator and io schedulers.
1198 rw_flags
= bio_data_dir(bio
);
1200 rw_flags
|= REQ_RW_SYNC
;
1203 * Grab a free request. This is might sleep but can not fail.
1204 * Returns with the queue unlocked.
1206 req
= get_request_wait(q
, rw_flags
, bio
);
1209 * After dropping the lock and possibly sleeping here, our request
1210 * may now be mergeable after it had proven unmergeable (above).
1211 * We don't worry about that case for efficiency. It won't happen
1212 * often, and the elevators are able to handle it.
1214 init_request_from_bio(req
, bio
);
1216 spin_lock_irq(q
->queue_lock
);
1217 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
) ||
1218 bio_flagged(bio
, BIO_CPU_AFFINE
))
1219 req
->cpu
= blk_cpu_to_group(smp_processor_id());
1220 if (elv_queue_empty(q
))
1222 add_request(q
, req
);
1225 __generic_unplug_device(q
);
1226 spin_unlock_irq(q
->queue_lock
);
1230 bio_endio(bio
, err
);
1235 * If bio->bi_dev is a partition, remap the location
1237 static inline void blk_partition_remap(struct bio
*bio
)
1239 struct block_device
*bdev
= bio
->bi_bdev
;
1241 if (bio_sectors(bio
) && bdev
!= bdev
->bd_contains
) {
1242 struct hd_struct
*p
= bdev
->bd_part
;
1244 bio
->bi_sector
+= p
->start_sect
;
1245 bio
->bi_bdev
= bdev
->bd_contains
;
1247 blk_add_trace_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
1248 bdev
->bd_dev
, bio
->bi_sector
,
1249 bio
->bi_sector
- p
->start_sect
);
1253 static void handle_bad_sector(struct bio
*bio
)
1255 char b
[BDEVNAME_SIZE
];
1257 printk(KERN_INFO
"attempt to access beyond end of device\n");
1258 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
1259 bdevname(bio
->bi_bdev
, b
),
1261 (unsigned long long)bio
->bi_sector
+ bio_sectors(bio
),
1262 (long long)(bio
->bi_bdev
->bd_inode
->i_size
>> 9));
1264 set_bit(BIO_EOF
, &bio
->bi_flags
);
1267 #ifdef CONFIG_FAIL_MAKE_REQUEST
1269 static DECLARE_FAULT_ATTR(fail_make_request
);
1271 static int __init
setup_fail_make_request(char *str
)
1273 return setup_fault_attr(&fail_make_request
, str
);
1275 __setup("fail_make_request=", setup_fail_make_request
);
1277 static int should_fail_request(struct bio
*bio
)
1279 struct hd_struct
*part
= bio
->bi_bdev
->bd_part
;
1281 if (part_to_disk(part
)->part0
.make_it_fail
|| part
->make_it_fail
)
1282 return should_fail(&fail_make_request
, bio
->bi_size
);
1287 static int __init
fail_make_request_debugfs(void)
1289 return init_fault_attr_dentries(&fail_make_request
,
1290 "fail_make_request");
1293 late_initcall(fail_make_request_debugfs
);
1295 #else /* CONFIG_FAIL_MAKE_REQUEST */
1297 static inline int should_fail_request(struct bio
*bio
)
1302 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1305 * Check whether this bio extends beyond the end of the device.
1307 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
1314 /* Test device or partition size, when known. */
1315 maxsector
= bio
->bi_bdev
->bd_inode
->i_size
>> 9;
1317 sector_t sector
= bio
->bi_sector
;
1319 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
1321 * This may well happen - the kernel calls bread()
1322 * without checking the size of the device, e.g., when
1323 * mounting a device.
1325 handle_bad_sector(bio
);
1334 * generic_make_request - hand a buffer to its device driver for I/O
1335 * @bio: The bio describing the location in memory and on the device.
1337 * generic_make_request() is used to make I/O requests of block
1338 * devices. It is passed a &struct bio, which describes the I/O that needs
1341 * generic_make_request() does not return any status. The
1342 * success/failure status of the request, along with notification of
1343 * completion, is delivered asynchronously through the bio->bi_end_io
1344 * function described (one day) else where.
1346 * The caller of generic_make_request must make sure that bi_io_vec
1347 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1348 * set to describe the device address, and the
1349 * bi_end_io and optionally bi_private are set to describe how
1350 * completion notification should be signaled.
1352 * generic_make_request and the drivers it calls may use bi_next if this
1353 * bio happens to be merged with someone else, and may change bi_dev and
1354 * bi_sector for remaps as it sees fit. So the values of these fields
1355 * should NOT be depended on after the call to generic_make_request.
1357 static inline void __generic_make_request(struct bio
*bio
)
1359 struct request_queue
*q
;
1360 sector_t old_sector
;
1361 int ret
, nr_sectors
= bio_sectors(bio
);
1367 if (bio_check_eod(bio
, nr_sectors
))
1371 * Resolve the mapping until finished. (drivers are
1372 * still free to implement/resolve their own stacking
1373 * by explicitly returning 0)
1375 * NOTE: we don't repeat the blk_size check for each new device.
1376 * Stacking drivers are expected to know what they are doing.
1381 char b
[BDEVNAME_SIZE
];
1383 q
= bdev_get_queue(bio
->bi_bdev
);
1386 "generic_make_request: Trying to access "
1387 "nonexistent block-device %s (%Lu)\n",
1388 bdevname(bio
->bi_bdev
, b
),
1389 (long long) bio
->bi_sector
);
1391 bio_endio(bio
, err
);
1395 if (unlikely(nr_sectors
> q
->max_hw_sectors
)) {
1396 printk(KERN_ERR
"bio too big device %s (%u > %u)\n",
1397 bdevname(bio
->bi_bdev
, b
),
1403 if (unlikely(test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)))
1406 if (should_fail_request(bio
))
1410 * If this device has partitions, remap block n
1411 * of partition p to block n+start(p) of the disk.
1413 blk_partition_remap(bio
);
1415 if (bio_integrity_enabled(bio
) && bio_integrity_prep(bio
))
1418 if (old_sector
!= -1)
1419 blk_add_trace_remap(q
, bio
, old_dev
, bio
->bi_sector
,
1422 blk_add_trace_bio(q
, bio
, BLK_TA_QUEUE
);
1424 old_sector
= bio
->bi_sector
;
1425 old_dev
= bio
->bi_bdev
->bd_dev
;
1427 if (bio_check_eod(bio
, nr_sectors
))
1429 if ((bio_empty_barrier(bio
) && !q
->prepare_flush_fn
) ||
1430 (bio_discard(bio
) && !q
->prepare_discard_fn
)) {
1435 ret
= q
->make_request_fn(q
, bio
);
1440 * We only want one ->make_request_fn to be active at a time,
1441 * else stack usage with stacked devices could be a problem.
1442 * So use current->bio_{list,tail} to keep a list of requests
1443 * submited by a make_request_fn function.
1444 * current->bio_tail is also used as a flag to say if
1445 * generic_make_request is currently active in this task or not.
1446 * If it is NULL, then no make_request is active. If it is non-NULL,
1447 * then a make_request is active, and new requests should be added
1450 void generic_make_request(struct bio
*bio
)
1452 if (current
->bio_tail
) {
1453 /* make_request is active */
1454 *(current
->bio_tail
) = bio
;
1455 bio
->bi_next
= NULL
;
1456 current
->bio_tail
= &bio
->bi_next
;
1459 /* following loop may be a bit non-obvious, and so deserves some
1461 * Before entering the loop, bio->bi_next is NULL (as all callers
1462 * ensure that) so we have a list with a single bio.
1463 * We pretend that we have just taken it off a longer list, so
1464 * we assign bio_list to the next (which is NULL) and bio_tail
1465 * to &bio_list, thus initialising the bio_list of new bios to be
1466 * added. __generic_make_request may indeed add some more bios
1467 * through a recursive call to generic_make_request. If it
1468 * did, we find a non-NULL value in bio_list and re-enter the loop
1469 * from the top. In this case we really did just take the bio
1470 * of the top of the list (no pretending) and so fixup bio_list and
1471 * bio_tail or bi_next, and call into __generic_make_request again.
1473 * The loop was structured like this to make only one call to
1474 * __generic_make_request (which is important as it is large and
1475 * inlined) and to keep the structure simple.
1477 BUG_ON(bio
->bi_next
);
1479 current
->bio_list
= bio
->bi_next
;
1480 if (bio
->bi_next
== NULL
)
1481 current
->bio_tail
= ¤t
->bio_list
;
1483 bio
->bi_next
= NULL
;
1484 __generic_make_request(bio
);
1485 bio
= current
->bio_list
;
1487 current
->bio_tail
= NULL
; /* deactivate */
1489 EXPORT_SYMBOL(generic_make_request
);
1492 * submit_bio - submit a bio to the block device layer for I/O
1493 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1494 * @bio: The &struct bio which describes the I/O
1496 * submit_bio() is very similar in purpose to generic_make_request(), and
1497 * uses that function to do most of the work. Both are fairly rough
1498 * interfaces; @bio must be presetup and ready for I/O.
1501 void submit_bio(int rw
, struct bio
*bio
)
1503 int count
= bio_sectors(bio
);
1508 * If it's a regular read/write or a barrier with data attached,
1509 * go through the normal accounting stuff before submission.
1511 if (bio_has_data(bio
)) {
1513 count_vm_events(PGPGOUT
, count
);
1515 task_io_account_read(bio
->bi_size
);
1516 count_vm_events(PGPGIN
, count
);
1519 if (unlikely(block_dump
)) {
1520 char b
[BDEVNAME_SIZE
];
1521 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s\n",
1522 current
->comm
, task_pid_nr(current
),
1523 (rw
& WRITE
) ? "WRITE" : "READ",
1524 (unsigned long long)bio
->bi_sector
,
1525 bdevname(bio
->bi_bdev
, b
));
1529 generic_make_request(bio
);
1531 EXPORT_SYMBOL(submit_bio
);
1534 * blk_rq_check_limits - Helper function to check a request for the queue limit
1536 * @rq: the request being checked
1539 * @rq may have been made based on weaker limitations of upper-level queues
1540 * in request stacking drivers, and it may violate the limitation of @q.
1541 * Since the block layer and the underlying device driver trust @rq
1542 * after it is inserted to @q, it should be checked against @q before
1543 * the insertion using this generic function.
1545 * This function should also be useful for request stacking drivers
1546 * in some cases below, so export this fuction.
1547 * Request stacking drivers like request-based dm may change the queue
1548 * limits while requests are in the queue (e.g. dm's table swapping).
1549 * Such request stacking drivers should check those requests agaist
1550 * the new queue limits again when they dispatch those requests,
1551 * although such checkings are also done against the old queue limits
1552 * when submitting requests.
1554 int blk_rq_check_limits(struct request_queue
*q
, struct request
*rq
)
1556 if (rq
->nr_sectors
> q
->max_sectors
||
1557 rq
->data_len
> q
->max_hw_sectors
<< 9) {
1558 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
1563 * queue's settings related to segment counting like q->bounce_pfn
1564 * may differ from that of other stacking queues.
1565 * Recalculate it to check the request correctly on this queue's
1568 blk_recalc_rq_segments(rq
);
1569 if (rq
->nr_phys_segments
> q
->max_phys_segments
||
1570 rq
->nr_phys_segments
> q
->max_hw_segments
) {
1571 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
1577 EXPORT_SYMBOL_GPL(blk_rq_check_limits
);
1580 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1581 * @q: the queue to submit the request
1582 * @rq: the request being queued
1584 int blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
1586 unsigned long flags
;
1588 if (blk_rq_check_limits(q
, rq
))
1591 #ifdef CONFIG_FAIL_MAKE_REQUEST
1592 if (rq
->rq_disk
&& rq
->rq_disk
->part0
.make_it_fail
&&
1593 should_fail(&fail_make_request
, blk_rq_bytes(rq
)))
1597 spin_lock_irqsave(q
->queue_lock
, flags
);
1600 * Submitting request must be dequeued before calling this function
1601 * because it will be linked to another request_queue
1603 BUG_ON(blk_queued_rq(rq
));
1605 drive_stat_acct(rq
, 1);
1606 __elv_add_request(q
, rq
, ELEVATOR_INSERT_BACK
, 0);
1608 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1612 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
1615 * __end_that_request_first - end I/O on a request
1616 * @req: the request being processed
1617 * @error: %0 for success, < %0 for error
1618 * @nr_bytes: number of bytes to complete
1621 * Ends I/O on a number of bytes attached to @req, and sets it up
1622 * for the next range of segments (if any) in the cluster.
1625 * %0 - we are done with this request, call end_that_request_last()
1626 * %1 - still buffers pending for this request
1628 static int __end_that_request_first(struct request
*req
, int error
,
1631 int total_bytes
, bio_nbytes
, next_idx
= 0;
1634 blk_add_trace_rq(req
->q
, req
, BLK_TA_COMPLETE
);
1637 * for a REQ_TYPE_BLOCK_PC request, we want to carry any eventual
1638 * sense key with us all the way through
1640 if (!blk_pc_request(req
))
1643 if (error
&& (blk_fs_request(req
) && !(req
->cmd_flags
& REQ_QUIET
))) {
1644 printk(KERN_ERR
"end_request: I/O error, dev %s, sector %llu\n",
1645 req
->rq_disk
? req
->rq_disk
->disk_name
: "?",
1646 (unsigned long long)req
->sector
);
1649 if (blk_fs_request(req
) && req
->rq_disk
) {
1650 const int rw
= rq_data_dir(req
);
1651 struct hd_struct
*part
;
1654 cpu
= part_stat_lock();
1655 part
= disk_map_sector_rcu(req
->rq_disk
, req
->sector
);
1656 part_stat_add(cpu
, part
, sectors
[rw
], nr_bytes
>> 9);
1660 total_bytes
= bio_nbytes
= 0;
1661 while ((bio
= req
->bio
) != NULL
) {
1665 * For an empty barrier request, the low level driver must
1666 * store a potential error location in ->sector. We pass
1667 * that back up in ->bi_sector.
1669 if (blk_empty_barrier(req
))
1670 bio
->bi_sector
= req
->sector
;
1672 if (nr_bytes
>= bio
->bi_size
) {
1673 req
->bio
= bio
->bi_next
;
1674 nbytes
= bio
->bi_size
;
1675 req_bio_endio(req
, bio
, nbytes
, error
);
1679 int idx
= bio
->bi_idx
+ next_idx
;
1681 if (unlikely(bio
->bi_idx
>= bio
->bi_vcnt
)) {
1682 blk_dump_rq_flags(req
, "__end_that");
1683 printk(KERN_ERR
"%s: bio idx %d >= vcnt %d\n",
1684 __func__
, bio
->bi_idx
, bio
->bi_vcnt
);
1688 nbytes
= bio_iovec_idx(bio
, idx
)->bv_len
;
1689 BIO_BUG_ON(nbytes
> bio
->bi_size
);
1692 * not a complete bvec done
1694 if (unlikely(nbytes
> nr_bytes
)) {
1695 bio_nbytes
+= nr_bytes
;
1696 total_bytes
+= nr_bytes
;
1701 * advance to the next vector
1704 bio_nbytes
+= nbytes
;
1707 total_bytes
+= nbytes
;
1713 * end more in this run, or just return 'not-done'
1715 if (unlikely(nr_bytes
<= 0))
1727 * if the request wasn't completed, update state
1730 req_bio_endio(req
, bio
, bio_nbytes
, error
);
1731 bio
->bi_idx
+= next_idx
;
1732 bio_iovec(bio
)->bv_offset
+= nr_bytes
;
1733 bio_iovec(bio
)->bv_len
-= nr_bytes
;
1736 blk_recalc_rq_sectors(req
, total_bytes
>> 9);
1737 blk_recalc_rq_segments(req
);
1742 * queue lock must be held
1744 static void end_that_request_last(struct request
*req
, int error
)
1746 struct gendisk
*disk
= req
->rq_disk
;
1748 blk_delete_timer(req
);
1750 if (blk_rq_tagged(req
))
1751 blk_queue_end_tag(req
->q
, req
);
1753 if (blk_queued_rq(req
))
1754 blkdev_dequeue_request(req
);
1756 if (unlikely(laptop_mode
) && blk_fs_request(req
))
1757 laptop_io_completion();
1760 * Account IO completion. bar_rq isn't accounted as a normal
1761 * IO on queueing nor completion. Accounting the containing
1762 * request is enough.
1764 if (disk
&& blk_fs_request(req
) && req
!= &req
->q
->bar_rq
) {
1765 unsigned long duration
= jiffies
- req
->start_time
;
1766 const int rw
= rq_data_dir(req
);
1767 struct hd_struct
*part
;
1770 cpu
= part_stat_lock();
1771 part
= disk_map_sector_rcu(disk
, req
->sector
);
1773 part_stat_inc(cpu
, part
, ios
[rw
]);
1774 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
1775 part_round_stats(cpu
, part
);
1776 part_dec_in_flight(part
);
1782 req
->end_io(req
, error
);
1784 if (blk_bidi_rq(req
))
1785 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
1787 __blk_put_request(req
->q
, req
);
1791 static inline void __end_request(struct request
*rq
, int uptodate
,
1792 unsigned int nr_bytes
)
1797 error
= uptodate
? uptodate
: -EIO
;
1799 __blk_end_request(rq
, error
, nr_bytes
);
1803 * blk_rq_bytes - Returns bytes left to complete in the entire request
1804 * @rq: the request being processed
1806 unsigned int blk_rq_bytes(struct request
*rq
)
1808 if (blk_fs_request(rq
))
1809 return rq
->hard_nr_sectors
<< 9;
1811 return rq
->data_len
;
1813 EXPORT_SYMBOL_GPL(blk_rq_bytes
);
1816 * blk_rq_cur_bytes - Returns bytes left to complete in the current segment
1817 * @rq: the request being processed
1819 unsigned int blk_rq_cur_bytes(struct request
*rq
)
1821 if (blk_fs_request(rq
))
1822 return rq
->current_nr_sectors
<< 9;
1825 return rq
->bio
->bi_size
;
1827 return rq
->data_len
;
1829 EXPORT_SYMBOL_GPL(blk_rq_cur_bytes
);
1832 * end_queued_request - end all I/O on a queued request
1833 * @rq: the request being processed
1834 * @uptodate: error value or %0/%1 uptodate flag
1837 * Ends all I/O on a request, and removes it from the block layer queues.
1838 * Not suitable for normal I/O completion, unless the driver still has
1839 * the request attached to the block layer.
1842 void end_queued_request(struct request
*rq
, int uptodate
)
1844 __end_request(rq
, uptodate
, blk_rq_bytes(rq
));
1846 EXPORT_SYMBOL(end_queued_request
);
1849 * end_dequeued_request - end all I/O on a dequeued request
1850 * @rq: the request being processed
1851 * @uptodate: error value or %0/%1 uptodate flag
1854 * Ends all I/O on a request. The request must already have been
1855 * dequeued using blkdev_dequeue_request(), as is normally the case
1859 void end_dequeued_request(struct request
*rq
, int uptodate
)
1861 __end_request(rq
, uptodate
, blk_rq_bytes(rq
));
1863 EXPORT_SYMBOL(end_dequeued_request
);
1867 * end_request - end I/O on the current segment of the request
1868 * @req: the request being processed
1869 * @uptodate: error value or %0/%1 uptodate flag
1872 * Ends I/O on the current segment of a request. If that is the only
1873 * remaining segment, the request is also completed and freed.
1875 * This is a remnant of how older block drivers handled I/O completions.
1876 * Modern drivers typically end I/O on the full request in one go, unless
1877 * they have a residual value to account for. For that case this function
1878 * isn't really useful, unless the residual just happens to be the
1879 * full current segment. In other words, don't use this function in new
1880 * code. Use blk_end_request() or __blk_end_request() to end partial parts
1881 * of a request, or end_dequeued_request() and end_queued_request() to
1882 * completely end IO on a dequeued/queued request.
1885 void end_request(struct request
*req
, int uptodate
)
1887 __end_request(req
, uptodate
, req
->hard_cur_sectors
<< 9);
1889 EXPORT_SYMBOL(end_request
);
1891 static int end_that_request_data(struct request
*rq
, int error
,
1892 unsigned int nr_bytes
, unsigned int bidi_bytes
)
1895 if (__end_that_request_first(rq
, error
, nr_bytes
))
1898 /* Bidi request must be completed as a whole */
1899 if (blk_bidi_rq(rq
) &&
1900 __end_that_request_first(rq
->next_rq
, error
, bidi_bytes
))
1908 * blk_end_io - Generic end_io function to complete a request.
1909 * @rq: the request being processed
1910 * @error: %0 for success, < %0 for error
1911 * @nr_bytes: number of bytes to complete @rq
1912 * @bidi_bytes: number of bytes to complete @rq->next_rq
1913 * @drv_callback: function called between completion of bios in the request
1914 * and completion of the request.
1915 * If the callback returns non %0, this helper returns without
1916 * completion of the request.
1919 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
1920 * If @rq has leftover, sets it up for the next range of segments.
1923 * %0 - we are done with this request
1924 * %1 - this request is not freed yet, it still has pending buffers.
1926 static int blk_end_io(struct request
*rq
, int error
, unsigned int nr_bytes
,
1927 unsigned int bidi_bytes
,
1928 int (drv_callback
)(struct request
*))
1930 struct request_queue
*q
= rq
->q
;
1931 unsigned long flags
= 0UL;
1933 if (end_that_request_data(rq
, error
, nr_bytes
, bidi_bytes
))
1936 /* Special feature for tricky drivers */
1937 if (drv_callback
&& drv_callback(rq
))
1940 add_disk_randomness(rq
->rq_disk
);
1942 spin_lock_irqsave(q
->queue_lock
, flags
);
1943 end_that_request_last(rq
, error
);
1944 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1950 * blk_end_request - Helper function for drivers to complete the request.
1951 * @rq: the request being processed
1952 * @error: %0 for success, < %0 for error
1953 * @nr_bytes: number of bytes to complete
1956 * Ends I/O on a number of bytes attached to @rq.
1957 * If @rq has leftover, sets it up for the next range of segments.
1960 * %0 - we are done with this request
1961 * %1 - still buffers pending for this request
1963 int blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
1965 return blk_end_io(rq
, error
, nr_bytes
, 0, NULL
);
1967 EXPORT_SYMBOL_GPL(blk_end_request
);
1970 * __blk_end_request - Helper function for drivers to complete the request.
1971 * @rq: the request being processed
1972 * @error: %0 for success, < %0 for error
1973 * @nr_bytes: number of bytes to complete
1976 * Must be called with queue lock held unlike blk_end_request().
1979 * %0 - we are done with this request
1980 * %1 - still buffers pending for this request
1982 int __blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
1984 if (rq
->bio
&& __end_that_request_first(rq
, error
, nr_bytes
))
1987 add_disk_randomness(rq
->rq_disk
);
1989 end_that_request_last(rq
, error
);
1993 EXPORT_SYMBOL_GPL(__blk_end_request
);
1996 * blk_end_bidi_request - Helper function for drivers to complete bidi request.
1997 * @rq: the bidi request being processed
1998 * @error: %0 for success, < %0 for error
1999 * @nr_bytes: number of bytes to complete @rq
2000 * @bidi_bytes: number of bytes to complete @rq->next_rq
2003 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2006 * %0 - we are done with this request
2007 * %1 - still buffers pending for this request
2009 int blk_end_bidi_request(struct request
*rq
, int error
, unsigned int nr_bytes
,
2010 unsigned int bidi_bytes
)
2012 return blk_end_io(rq
, error
, nr_bytes
, bidi_bytes
, NULL
);
2014 EXPORT_SYMBOL_GPL(blk_end_bidi_request
);
2017 * blk_update_request - Special helper function for request stacking drivers
2018 * @rq: the request being processed
2019 * @error: %0 for success, < %0 for error
2020 * @nr_bytes: number of bytes to complete @rq
2023 * Ends I/O on a number of bytes attached to @rq, but doesn't complete
2024 * the request structure even if @rq doesn't have leftover.
2025 * If @rq has leftover, sets it up for the next range of segments.
2027 * This special helper function is only for request stacking drivers
2028 * (e.g. request-based dm) so that they can handle partial completion.
2029 * Actual device drivers should use blk_end_request instead.
2031 void blk_update_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2033 if (!end_that_request_data(rq
, error
, nr_bytes
, 0)) {
2035 * These members are not updated in end_that_request_data()
2036 * when all bios are completed.
2037 * Update them so that the request stacking driver can find
2038 * how many bytes remain in the request later.
2040 rq
->nr_sectors
= rq
->hard_nr_sectors
= 0;
2041 rq
->current_nr_sectors
= rq
->hard_cur_sectors
= 0;
2044 EXPORT_SYMBOL_GPL(blk_update_request
);
2047 * blk_end_request_callback - Special helper function for tricky drivers
2048 * @rq: the request being processed
2049 * @error: %0 for success, < %0 for error
2050 * @nr_bytes: number of bytes to complete
2051 * @drv_callback: function called between completion of bios in the request
2052 * and completion of the request.
2053 * If the callback returns non %0, this helper returns without
2054 * completion of the request.
2057 * Ends I/O on a number of bytes attached to @rq.
2058 * If @rq has leftover, sets it up for the next range of segments.
2060 * This special helper function is used only for existing tricky drivers.
2061 * (e.g. cdrom_newpc_intr() of ide-cd)
2062 * This interface will be removed when such drivers are rewritten.
2063 * Don't use this interface in other places anymore.
2066 * %0 - we are done with this request
2067 * %1 - this request is not freed yet.
2068 * this request still has pending buffers or
2069 * the driver doesn't want to finish this request yet.
2071 int blk_end_request_callback(struct request
*rq
, int error
,
2072 unsigned int nr_bytes
,
2073 int (drv_callback
)(struct request
*))
2075 return blk_end_io(rq
, error
, nr_bytes
, 0, drv_callback
);
2077 EXPORT_SYMBOL_GPL(blk_end_request_callback
);
2079 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
2082 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw, and
2083 we want BIO_RW_AHEAD (bit 1) to imply REQ_FAILFAST (bit 1). */
2084 rq
->cmd_flags
|= (bio
->bi_rw
& 3);
2086 if (bio_has_data(bio
)) {
2087 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2088 rq
->buffer
= bio_data(bio
);
2090 rq
->current_nr_sectors
= bio_cur_sectors(bio
);
2091 rq
->hard_cur_sectors
= rq
->current_nr_sectors
;
2092 rq
->hard_nr_sectors
= rq
->nr_sectors
= bio_sectors(bio
);
2093 rq
->data_len
= bio
->bi_size
;
2095 rq
->bio
= rq
->biotail
= bio
;
2098 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2101 int kblockd_schedule_work(struct request_queue
*q
, struct work_struct
*work
)
2103 return queue_work(kblockd_workqueue
, work
);
2105 EXPORT_SYMBOL(kblockd_schedule_work
);
2107 void kblockd_flush_work(struct work_struct
*work
)
2109 cancel_work_sync(work
);
2111 EXPORT_SYMBOL(kblockd_flush_work
);
2113 int __init
blk_dev_init(void)
2115 kblockd_workqueue
= create_workqueue("kblockd");
2116 if (!kblockd_workqueue
)
2117 panic("Failed to create kblockd\n");
2119 request_cachep
= kmem_cache_create("blkdev_requests",
2120 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
2122 blk_requestq_cachep
= kmem_cache_create("blkdev_queue",
2123 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);