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
)))
271 if (!blk_remove_plug(q
) && !blk_queue_nonrot(q
))
278 * generic_unplug_device - fire a request queue
279 * @q: The &struct request_queue in question
282 * Linux uses plugging to build bigger requests queues before letting
283 * the device have at them. If a queue is plugged, the I/O scheduler
284 * is still adding and merging requests on the queue. Once the queue
285 * gets unplugged, the request_fn defined for the queue is invoked and
288 void generic_unplug_device(struct request_queue
*q
)
290 if (blk_queue_plugged(q
)) {
291 spin_lock_irq(q
->queue_lock
);
292 __generic_unplug_device(q
);
293 spin_unlock_irq(q
->queue_lock
);
296 EXPORT_SYMBOL(generic_unplug_device
);
298 static void blk_backing_dev_unplug(struct backing_dev_info
*bdi
,
301 struct request_queue
*q
= bdi
->unplug_io_data
;
306 void blk_unplug_work(struct work_struct
*work
)
308 struct request_queue
*q
=
309 container_of(work
, struct request_queue
, unplug_work
);
311 trace_block_unplug_io(q
);
315 void blk_unplug_timeout(unsigned long data
)
317 struct request_queue
*q
= (struct request_queue
*)data
;
319 trace_block_unplug_timer(q
);
320 kblockd_schedule_work(q
, &q
->unplug_work
);
323 void blk_unplug(struct request_queue
*q
)
326 * devices don't necessarily have an ->unplug_fn defined
329 trace_block_unplug_io(q
);
333 EXPORT_SYMBOL(blk_unplug
);
335 static void blk_invoke_request_fn(struct request_queue
*q
)
337 if (unlikely(blk_queue_stopped(q
)))
341 * one level of recursion is ok and is much faster than kicking
342 * the unplug handling
344 if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER
, q
)) {
346 queue_flag_clear(QUEUE_FLAG_REENTER
, q
);
348 queue_flag_set(QUEUE_FLAG_PLUGGED
, q
);
349 kblockd_schedule_work(q
, &q
->unplug_work
);
354 * blk_start_queue - restart a previously stopped queue
355 * @q: The &struct request_queue in question
358 * blk_start_queue() will clear the stop flag on the queue, and call
359 * the request_fn for the queue if it was in a stopped state when
360 * entered. Also see blk_stop_queue(). Queue lock must be held.
362 void blk_start_queue(struct request_queue
*q
)
364 WARN_ON(!irqs_disabled());
366 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
367 blk_invoke_request_fn(q
);
369 EXPORT_SYMBOL(blk_start_queue
);
372 * blk_stop_queue - stop a queue
373 * @q: The &struct request_queue in question
376 * The Linux block layer assumes that a block driver will consume all
377 * entries on the request queue when the request_fn strategy is called.
378 * Often this will not happen, because of hardware limitations (queue
379 * depth settings). If a device driver gets a 'queue full' response,
380 * or if it simply chooses not to queue more I/O at one point, it can
381 * call this function to prevent the request_fn from being called until
382 * the driver has signalled it's ready to go again. This happens by calling
383 * blk_start_queue() to restart queue operations. Queue lock must be held.
385 void blk_stop_queue(struct request_queue
*q
)
388 queue_flag_set(QUEUE_FLAG_STOPPED
, q
);
390 EXPORT_SYMBOL(blk_stop_queue
);
393 * blk_sync_queue - cancel any pending callbacks on a queue
397 * The block layer may perform asynchronous callback activity
398 * on a queue, such as calling the unplug function after a timeout.
399 * A block device may call blk_sync_queue to ensure that any
400 * such activity is cancelled, thus allowing it to release resources
401 * that the callbacks might use. The caller must already have made sure
402 * that its ->make_request_fn will not re-add plugging prior to calling
406 void blk_sync_queue(struct request_queue
*q
)
408 del_timer_sync(&q
->unplug_timer
);
409 del_timer_sync(&q
->timeout
);
410 cancel_work_sync(&q
->unplug_work
);
412 EXPORT_SYMBOL(blk_sync_queue
);
415 * __blk_run_queue - run a single device queue
416 * @q: The queue to run
419 * See @blk_run_queue. This variant must be called with the queue lock
420 * held and interrupts disabled.
423 void __blk_run_queue(struct request_queue
*q
)
428 * Only recurse once to avoid overrunning the stack, let the unplug
429 * handling reinvoke the handler shortly if we already got there.
431 if (!elv_queue_empty(q
))
432 blk_invoke_request_fn(q
);
434 EXPORT_SYMBOL(__blk_run_queue
);
437 * blk_run_queue - run a single device queue
438 * @q: The queue to run
441 * Invoke request handling on this queue, if it has pending work to do.
442 * May be used to restart queueing when a request has completed. Also
443 * See @blk_start_queueing.
446 void blk_run_queue(struct request_queue
*q
)
450 spin_lock_irqsave(q
->queue_lock
, flags
);
452 spin_unlock_irqrestore(q
->queue_lock
, flags
);
454 EXPORT_SYMBOL(blk_run_queue
);
456 void blk_put_queue(struct request_queue
*q
)
458 kobject_put(&q
->kobj
);
461 void blk_cleanup_queue(struct request_queue
*q
)
464 * We know we have process context here, so we can be a little
465 * cautious and ensure that pending block actions on this device
466 * are done before moving on. Going into this function, we should
467 * not have processes doing IO to this device.
471 mutex_lock(&q
->sysfs_lock
);
472 queue_flag_set_unlocked(QUEUE_FLAG_DEAD
, q
);
473 mutex_unlock(&q
->sysfs_lock
);
476 elevator_exit(q
->elevator
);
480 EXPORT_SYMBOL(blk_cleanup_queue
);
482 static int blk_init_free_list(struct request_queue
*q
)
484 struct request_list
*rl
= &q
->rq
;
486 rl
->count
[READ
] = rl
->count
[WRITE
] = 0;
487 rl
->starved
[READ
] = rl
->starved
[WRITE
] = 0;
489 init_waitqueue_head(&rl
->wait
[READ
]);
490 init_waitqueue_head(&rl
->wait
[WRITE
]);
492 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, mempool_alloc_slab
,
493 mempool_free_slab
, request_cachep
, q
->node
);
501 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
503 return blk_alloc_queue_node(gfp_mask
, -1);
505 EXPORT_SYMBOL(blk_alloc_queue
);
507 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
509 struct request_queue
*q
;
512 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
513 gfp_mask
| __GFP_ZERO
, node_id
);
517 q
->backing_dev_info
.unplug_io_fn
= blk_backing_dev_unplug
;
518 q
->backing_dev_info
.unplug_io_data
= q
;
519 err
= bdi_init(&q
->backing_dev_info
);
521 kmem_cache_free(blk_requestq_cachep
, q
);
525 init_timer(&q
->unplug_timer
);
526 setup_timer(&q
->timeout
, blk_rq_timed_out_timer
, (unsigned long) q
);
527 INIT_LIST_HEAD(&q
->timeout_list
);
528 INIT_WORK(&q
->unplug_work
, blk_unplug_work
);
530 kobject_init(&q
->kobj
, &blk_queue_ktype
);
532 mutex_init(&q
->sysfs_lock
);
533 spin_lock_init(&q
->__queue_lock
);
537 EXPORT_SYMBOL(blk_alloc_queue_node
);
540 * blk_init_queue - prepare a request queue for use with a block device
541 * @rfn: The function to be called to process requests that have been
542 * placed on the queue.
543 * @lock: Request queue spin lock
546 * If a block device wishes to use the standard request handling procedures,
547 * which sorts requests and coalesces adjacent requests, then it must
548 * call blk_init_queue(). The function @rfn will be called when there
549 * are requests on the queue that need to be processed. If the device
550 * supports plugging, then @rfn may not be called immediately when requests
551 * are available on the queue, but may be called at some time later instead.
552 * Plugged queues are generally unplugged when a buffer belonging to one
553 * of the requests on the queue is needed, or due to memory pressure.
555 * @rfn is not required, or even expected, to remove all requests off the
556 * queue, but only as many as it can handle at a time. If it does leave
557 * requests on the queue, it is responsible for arranging that the requests
558 * get dealt with eventually.
560 * The queue spin lock must be held while manipulating the requests on the
561 * request queue; this lock will be taken also from interrupt context, so irq
562 * disabling is needed for it.
564 * Function returns a pointer to the initialized request queue, or %NULL if
568 * blk_init_queue() must be paired with a blk_cleanup_queue() call
569 * when the block device is deactivated (such as at module unload).
572 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
574 return blk_init_queue_node(rfn
, lock
, -1);
576 EXPORT_SYMBOL(blk_init_queue
);
578 struct request_queue
*
579 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
581 struct request_queue
*q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
587 if (blk_init_free_list(q
)) {
588 kmem_cache_free(blk_requestq_cachep
, q
);
593 * if caller didn't supply a lock, they get per-queue locking with
597 lock
= &q
->__queue_lock
;
600 q
->prep_rq_fn
= NULL
;
601 q
->unplug_fn
= generic_unplug_device
;
602 q
->queue_flags
= (1 << QUEUE_FLAG_CLUSTER
|
603 1 << QUEUE_FLAG_STACKABLE
);
604 q
->queue_lock
= lock
;
606 blk_queue_segment_boundary(q
, BLK_SEG_BOUNDARY_MASK
);
608 blk_queue_make_request(q
, __make_request
);
609 blk_queue_max_segment_size(q
, MAX_SEGMENT_SIZE
);
611 blk_queue_max_hw_segments(q
, MAX_HW_SEGMENTS
);
612 blk_queue_max_phys_segments(q
, MAX_PHYS_SEGMENTS
);
614 q
->sg_reserved_size
= INT_MAX
;
616 blk_set_cmd_filter_defaults(&q
->cmd_filter
);
621 if (!elevator_init(q
, NULL
)) {
622 blk_queue_congestion_threshold(q
);
629 EXPORT_SYMBOL(blk_init_queue_node
);
631 int blk_get_queue(struct request_queue
*q
)
633 if (likely(!test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
))) {
634 kobject_get(&q
->kobj
);
641 static inline void blk_free_request(struct request_queue
*q
, struct request
*rq
)
643 if (rq
->cmd_flags
& REQ_ELVPRIV
)
644 elv_put_request(q
, rq
);
645 mempool_free(rq
, q
->rq
.rq_pool
);
648 static struct request
*
649 blk_alloc_request(struct request_queue
*q
, int rw
, int priv
, gfp_t gfp_mask
)
651 struct request
*rq
= mempool_alloc(q
->rq
.rq_pool
, gfp_mask
);
658 rq
->cmd_flags
= rw
| REQ_ALLOCED
;
661 if (unlikely(elv_set_request(q
, rq
, gfp_mask
))) {
662 mempool_free(rq
, q
->rq
.rq_pool
);
665 rq
->cmd_flags
|= REQ_ELVPRIV
;
672 * ioc_batching returns true if the ioc is a valid batching request and
673 * should be given priority access to a request.
675 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
681 * Make sure the process is able to allocate at least 1 request
682 * even if the batch times out, otherwise we could theoretically
685 return ioc
->nr_batch_requests
== q
->nr_batching
||
686 (ioc
->nr_batch_requests
> 0
687 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
691 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
692 * will cause the process to be a "batcher" on all queues in the system. This
693 * is the behaviour we want though - once it gets a wakeup it should be given
696 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
698 if (!ioc
|| ioc_batching(q
, ioc
))
701 ioc
->nr_batch_requests
= q
->nr_batching
;
702 ioc
->last_waited
= jiffies
;
705 static void __freed_request(struct request_queue
*q
, int rw
)
707 struct request_list
*rl
= &q
->rq
;
709 if (rl
->count
[rw
] < queue_congestion_off_threshold(q
))
710 blk_clear_queue_congested(q
, rw
);
712 if (rl
->count
[rw
] + 1 <= q
->nr_requests
) {
713 if (waitqueue_active(&rl
->wait
[rw
]))
714 wake_up(&rl
->wait
[rw
]);
716 blk_clear_queue_full(q
, rw
);
721 * A request has just been released. Account for it, update the full and
722 * congestion status, wake up any waiters. Called under q->queue_lock.
724 static void freed_request(struct request_queue
*q
, int rw
, int priv
)
726 struct request_list
*rl
= &q
->rq
;
732 __freed_request(q
, rw
);
734 if (unlikely(rl
->starved
[rw
^ 1]))
735 __freed_request(q
, rw
^ 1);
738 #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
740 * Get a free request, queue_lock must be held.
741 * Returns NULL on failure, with queue_lock held.
742 * Returns !NULL on success, with queue_lock *not held*.
744 static struct request
*get_request(struct request_queue
*q
, int rw_flags
,
745 struct bio
*bio
, gfp_t gfp_mask
)
747 struct request
*rq
= NULL
;
748 struct request_list
*rl
= &q
->rq
;
749 struct io_context
*ioc
= NULL
;
750 const int rw
= rw_flags
& 0x01;
753 may_queue
= elv_may_queue(q
, rw_flags
);
754 if (may_queue
== ELV_MQUEUE_NO
)
757 if (rl
->count
[rw
]+1 >= queue_congestion_on_threshold(q
)) {
758 if (rl
->count
[rw
]+1 >= q
->nr_requests
) {
759 ioc
= current_io_context(GFP_ATOMIC
, q
->node
);
761 * The queue will fill after this allocation, so set
762 * it as full, and mark this process as "batching".
763 * This process will be allowed to complete a batch of
764 * requests, others will be blocked.
766 if (!blk_queue_full(q
, rw
)) {
767 ioc_set_batching(q
, ioc
);
768 blk_set_queue_full(q
, rw
);
770 if (may_queue
!= ELV_MQUEUE_MUST
771 && !ioc_batching(q
, ioc
)) {
773 * The queue is full and the allocating
774 * process is not a "batcher", and not
775 * exempted by the IO scheduler
781 blk_set_queue_congested(q
, rw
);
785 * Only allow batching queuers to allocate up to 50% over the defined
786 * limit of requests, otherwise we could have thousands of requests
787 * allocated with any setting of ->nr_requests
789 if (rl
->count
[rw
] >= (3 * q
->nr_requests
/ 2))
795 priv
= !test_bit(QUEUE_FLAG_ELVSWITCH
, &q
->queue_flags
);
799 spin_unlock_irq(q
->queue_lock
);
801 rq
= blk_alloc_request(q
, rw_flags
, priv
, gfp_mask
);
804 * Allocation failed presumably due to memory. Undo anything
805 * we might have messed up.
807 * Allocating task should really be put onto the front of the
808 * wait queue, but this is pretty rare.
810 spin_lock_irq(q
->queue_lock
);
811 freed_request(q
, rw
, priv
);
814 * in the very unlikely event that allocation failed and no
815 * requests for this direction was pending, mark us starved
816 * so that freeing of a request in the other direction will
817 * notice us. another possible fix would be to split the
818 * rq mempool into READ and WRITE
821 if (unlikely(rl
->count
[rw
] == 0))
828 * ioc may be NULL here, and ioc_batching will be false. That's
829 * OK, if the queue is under the request limit then requests need
830 * not count toward the nr_batch_requests limit. There will always
831 * be some limit enforced by BLK_BATCH_TIME.
833 if (ioc_batching(q
, ioc
))
834 ioc
->nr_batch_requests
--;
836 trace_block_getrq(q
, bio
, rw
);
842 * No available requests for this queue, unplug the device and wait for some
843 * requests to become available.
845 * Called with q->queue_lock held, and returns with it unlocked.
847 static struct request
*get_request_wait(struct request_queue
*q
, int rw_flags
,
850 const int rw
= rw_flags
& 0x01;
853 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
856 struct io_context
*ioc
;
857 struct request_list
*rl
= &q
->rq
;
859 prepare_to_wait_exclusive(&rl
->wait
[rw
], &wait
,
860 TASK_UNINTERRUPTIBLE
);
862 trace_block_sleeprq(q
, bio
, rw
);
864 __generic_unplug_device(q
);
865 spin_unlock_irq(q
->queue_lock
);
869 * After sleeping, we become a "batching" process and
870 * will be able to allocate at least one request, and
871 * up to a big batch of them for a small period time.
872 * See ioc_batching, ioc_set_batching
874 ioc
= current_io_context(GFP_NOIO
, q
->node
);
875 ioc_set_batching(q
, ioc
);
877 spin_lock_irq(q
->queue_lock
);
878 finish_wait(&rl
->wait
[rw
], &wait
);
880 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
886 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
890 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
892 spin_lock_irq(q
->queue_lock
);
893 if (gfp_mask
& __GFP_WAIT
) {
894 rq
= get_request_wait(q
, rw
, NULL
);
896 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
898 spin_unlock_irq(q
->queue_lock
);
900 /* q->queue_lock is unlocked at this point */
904 EXPORT_SYMBOL(blk_get_request
);
907 * blk_start_queueing - initiate dispatch of requests to device
908 * @q: request queue to kick into gear
910 * This is basically a helper to remove the need to know whether a queue
911 * is plugged or not if someone just wants to initiate dispatch of requests
912 * for this queue. Should be used to start queueing on a device outside
913 * of ->request_fn() context. Also see @blk_run_queue.
915 * The queue lock must be held with interrupts disabled.
917 void blk_start_queueing(struct request_queue
*q
)
919 if (!blk_queue_plugged(q
)) {
920 if (unlikely(blk_queue_stopped(q
)))
924 __generic_unplug_device(q
);
926 EXPORT_SYMBOL(blk_start_queueing
);
929 * blk_requeue_request - put a request back on queue
930 * @q: request queue where request should be inserted
931 * @rq: request to be inserted
934 * Drivers often keep queueing requests until the hardware cannot accept
935 * more, when that condition happens we need to put the request back
936 * on the queue. Must be called with queue lock held.
938 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
940 blk_delete_timer(rq
);
941 blk_clear_rq_complete(rq
);
942 trace_block_rq_requeue(q
, rq
);
944 if (blk_rq_tagged(rq
))
945 blk_queue_end_tag(q
, rq
);
947 elv_requeue_request(q
, rq
);
949 EXPORT_SYMBOL(blk_requeue_request
);
952 * blk_insert_request - insert a special request into a request queue
953 * @q: request queue where request should be inserted
954 * @rq: request to be inserted
955 * @at_head: insert request at head or tail of queue
956 * @data: private data
959 * Many block devices need to execute commands asynchronously, so they don't
960 * block the whole kernel from preemption during request execution. This is
961 * accomplished normally by inserting aritficial requests tagged as
962 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
963 * be scheduled for actual execution by the request queue.
965 * We have the option of inserting the head or the tail of the queue.
966 * Typically we use the tail for new ioctls and so forth. We use the head
967 * of the queue for things like a QUEUE_FULL message from a device, or a
968 * host that is unable to accept a particular command.
970 void blk_insert_request(struct request_queue
*q
, struct request
*rq
,
971 int at_head
, void *data
)
973 int where
= at_head
? ELEVATOR_INSERT_FRONT
: ELEVATOR_INSERT_BACK
;
977 * tell I/O scheduler that this isn't a regular read/write (ie it
978 * must not attempt merges on this) and that it acts as a soft
981 rq
->cmd_type
= REQ_TYPE_SPECIAL
;
982 rq
->cmd_flags
|= REQ_SOFTBARRIER
;
986 spin_lock_irqsave(q
->queue_lock
, flags
);
989 * If command is tagged, release the tag
991 if (blk_rq_tagged(rq
))
992 blk_queue_end_tag(q
, rq
);
994 drive_stat_acct(rq
, 1);
995 __elv_add_request(q
, rq
, where
, 0);
996 blk_start_queueing(q
);
997 spin_unlock_irqrestore(q
->queue_lock
, flags
);
999 EXPORT_SYMBOL(blk_insert_request
);
1002 * add-request adds a request to the linked list.
1003 * queue lock is held and interrupts disabled, as we muck with the
1004 * request queue list.
1006 static inline void add_request(struct request_queue
*q
, struct request
*req
)
1008 drive_stat_acct(req
, 1);
1011 * elevator indicated where it wants this request to be
1012 * inserted at elevator_merge time
1014 __elv_add_request(q
, req
, ELEVATOR_INSERT_SORT
, 0);
1017 static void part_round_stats_single(int cpu
, struct hd_struct
*part
,
1020 if (now
== part
->stamp
)
1023 if (part
->in_flight
) {
1024 __part_stat_add(cpu
, part
, time_in_queue
,
1025 part
->in_flight
* (now
- part
->stamp
));
1026 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
1032 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1033 * @cpu: cpu number for stats access
1034 * @part: target partition
1036 * The average IO queue length and utilisation statistics are maintained
1037 * by observing the current state of the queue length and the amount of
1038 * time it has been in this state for.
1040 * Normally, that accounting is done on IO completion, but that can result
1041 * in more than a second's worth of IO being accounted for within any one
1042 * second, leading to >100% utilisation. To deal with that, we call this
1043 * function to do a round-off before returning the results when reading
1044 * /proc/diskstats. This accounts immediately for all queue usage up to
1045 * the current jiffies and restarts the counters again.
1047 void part_round_stats(int cpu
, struct hd_struct
*part
)
1049 unsigned long now
= jiffies
;
1052 part_round_stats_single(cpu
, &part_to_disk(part
)->part0
, now
);
1053 part_round_stats_single(cpu
, part
, now
);
1055 EXPORT_SYMBOL_GPL(part_round_stats
);
1058 * queue lock must be held
1060 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1064 if (unlikely(--req
->ref_count
))
1067 elv_completed_request(q
, req
);
1070 * Request may not have originated from ll_rw_blk. if not,
1071 * it didn't come out of our reserved rq pools
1073 if (req
->cmd_flags
& REQ_ALLOCED
) {
1074 int rw
= rq_data_dir(req
);
1075 int priv
= req
->cmd_flags
& REQ_ELVPRIV
;
1077 BUG_ON(!list_empty(&req
->queuelist
));
1078 BUG_ON(!hlist_unhashed(&req
->hash
));
1080 blk_free_request(q
, req
);
1081 freed_request(q
, rw
, priv
);
1084 EXPORT_SYMBOL_GPL(__blk_put_request
);
1086 void blk_put_request(struct request
*req
)
1088 unsigned long flags
;
1089 struct request_queue
*q
= req
->q
;
1091 spin_lock_irqsave(q
->queue_lock
, flags
);
1092 __blk_put_request(q
, req
);
1093 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1095 EXPORT_SYMBOL(blk_put_request
);
1097 void init_request_from_bio(struct request
*req
, struct bio
*bio
)
1099 req
->cpu
= bio
->bi_comp_cpu
;
1100 req
->cmd_type
= REQ_TYPE_FS
;
1103 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
1105 if (bio_rw_ahead(bio
))
1106 req
->cmd_flags
|= (REQ_FAILFAST_DEV
| REQ_FAILFAST_TRANSPORT
|
1107 REQ_FAILFAST_DRIVER
);
1108 if (bio_failfast_dev(bio
))
1109 req
->cmd_flags
|= REQ_FAILFAST_DEV
;
1110 if (bio_failfast_transport(bio
))
1111 req
->cmd_flags
|= REQ_FAILFAST_TRANSPORT
;
1112 if (bio_failfast_driver(bio
))
1113 req
->cmd_flags
|= REQ_FAILFAST_DRIVER
;
1116 * REQ_BARRIER implies no merging, but lets make it explicit
1118 if (unlikely(bio_discard(bio
))) {
1119 req
->cmd_flags
|= REQ_DISCARD
;
1120 if (bio_barrier(bio
))
1121 req
->cmd_flags
|= REQ_SOFTBARRIER
;
1122 req
->q
->prepare_discard_fn(req
->q
, req
);
1123 } else if (unlikely(bio_barrier(bio
)))
1124 req
->cmd_flags
|= (REQ_HARDBARRIER
| REQ_NOMERGE
);
1127 req
->cmd_flags
|= REQ_RW_SYNC
;
1128 if (bio_unplug(bio
))
1129 req
->cmd_flags
|= REQ_UNPLUG
;
1130 if (bio_rw_meta(bio
))
1131 req
->cmd_flags
|= REQ_RW_META
;
1134 req
->hard_sector
= req
->sector
= bio
->bi_sector
;
1135 req
->ioprio
= bio_prio(bio
);
1136 req
->start_time
= jiffies
;
1137 blk_rq_bio_prep(req
->q
, req
, bio
);
1140 static int __make_request(struct request_queue
*q
, struct bio
*bio
)
1142 struct request
*req
;
1143 int el_ret
, nr_sectors
;
1144 const unsigned short prio
= bio_prio(bio
);
1145 const int sync
= bio_sync(bio
);
1146 const int unplug
= bio_unplug(bio
);
1149 nr_sectors
= bio_sectors(bio
);
1152 * low level driver can indicate that it wants pages above a
1153 * certain limit bounced to low memory (ie for highmem, or even
1154 * ISA dma in theory)
1156 blk_queue_bounce(q
, &bio
);
1158 spin_lock_irq(q
->queue_lock
);
1160 if (unlikely(bio_barrier(bio
)) || elv_queue_empty(q
))
1163 el_ret
= elv_merge(q
, &req
, bio
);
1165 case ELEVATOR_BACK_MERGE
:
1166 BUG_ON(!rq_mergeable(req
));
1168 if (!ll_back_merge_fn(q
, req
, bio
))
1171 trace_block_bio_backmerge(q
, bio
);
1173 req
->biotail
->bi_next
= bio
;
1175 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
1176 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
1177 if (!blk_rq_cpu_valid(req
))
1178 req
->cpu
= bio
->bi_comp_cpu
;
1179 drive_stat_acct(req
, 0);
1180 if (!attempt_back_merge(q
, req
))
1181 elv_merged_request(q
, req
, el_ret
);
1184 case ELEVATOR_FRONT_MERGE
:
1185 BUG_ON(!rq_mergeable(req
));
1187 if (!ll_front_merge_fn(q
, req
, bio
))
1190 trace_block_bio_frontmerge(q
, bio
);
1192 bio
->bi_next
= req
->bio
;
1196 * may not be valid. if the low level driver said
1197 * it didn't need a bounce buffer then it better
1198 * not touch req->buffer either...
1200 req
->buffer
= bio_data(bio
);
1201 req
->current_nr_sectors
= bio_cur_sectors(bio
);
1202 req
->hard_cur_sectors
= req
->current_nr_sectors
;
1203 req
->sector
= req
->hard_sector
= bio
->bi_sector
;
1204 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
1205 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
1206 if (!blk_rq_cpu_valid(req
))
1207 req
->cpu
= bio
->bi_comp_cpu
;
1208 drive_stat_acct(req
, 0);
1209 if (!attempt_front_merge(q
, req
))
1210 elv_merged_request(q
, req
, el_ret
);
1213 /* ELV_NO_MERGE: elevator says don't/can't merge. */
1220 * This sync check and mask will be re-done in init_request_from_bio(),
1221 * but we need to set it earlier to expose the sync flag to the
1222 * rq allocator and io schedulers.
1224 rw_flags
= bio_data_dir(bio
);
1226 rw_flags
|= REQ_RW_SYNC
;
1229 * Grab a free request. This is might sleep but can not fail.
1230 * Returns with the queue unlocked.
1232 req
= get_request_wait(q
, rw_flags
, bio
);
1235 * After dropping the lock and possibly sleeping here, our request
1236 * may now be mergeable after it had proven unmergeable (above).
1237 * We don't worry about that case for efficiency. It won't happen
1238 * often, and the elevators are able to handle it.
1240 init_request_from_bio(req
, bio
);
1242 spin_lock_irq(q
->queue_lock
);
1243 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
) ||
1244 bio_flagged(bio
, BIO_CPU_AFFINE
))
1245 req
->cpu
= blk_cpu_to_group(smp_processor_id());
1246 if (!blk_queue_nonrot(q
) && elv_queue_empty(q
))
1248 add_request(q
, req
);
1250 if (unplug
|| blk_queue_nonrot(q
))
1251 __generic_unplug_device(q
);
1252 spin_unlock_irq(q
->queue_lock
);
1257 * If bio->bi_dev is a partition, remap the location
1259 static inline void blk_partition_remap(struct bio
*bio
)
1261 struct block_device
*bdev
= bio
->bi_bdev
;
1263 if (bio_sectors(bio
) && bdev
!= bdev
->bd_contains
) {
1264 struct hd_struct
*p
= bdev
->bd_part
;
1266 bio
->bi_sector
+= p
->start_sect
;
1267 bio
->bi_bdev
= bdev
->bd_contains
;
1269 trace_block_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
1270 bdev
->bd_dev
, bio
->bi_sector
,
1271 bio
->bi_sector
- p
->start_sect
);
1275 static void handle_bad_sector(struct bio
*bio
)
1277 char b
[BDEVNAME_SIZE
];
1279 printk(KERN_INFO
"attempt to access beyond end of device\n");
1280 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
1281 bdevname(bio
->bi_bdev
, b
),
1283 (unsigned long long)bio
->bi_sector
+ bio_sectors(bio
),
1284 (long long)(bio
->bi_bdev
->bd_inode
->i_size
>> 9));
1286 set_bit(BIO_EOF
, &bio
->bi_flags
);
1289 #ifdef CONFIG_FAIL_MAKE_REQUEST
1291 static DECLARE_FAULT_ATTR(fail_make_request
);
1293 static int __init
setup_fail_make_request(char *str
)
1295 return setup_fault_attr(&fail_make_request
, str
);
1297 __setup("fail_make_request=", setup_fail_make_request
);
1299 static int should_fail_request(struct bio
*bio
)
1301 struct hd_struct
*part
= bio
->bi_bdev
->bd_part
;
1303 if (part_to_disk(part
)->part0
.make_it_fail
|| part
->make_it_fail
)
1304 return should_fail(&fail_make_request
, bio
->bi_size
);
1309 static int __init
fail_make_request_debugfs(void)
1311 return init_fault_attr_dentries(&fail_make_request
,
1312 "fail_make_request");
1315 late_initcall(fail_make_request_debugfs
);
1317 #else /* CONFIG_FAIL_MAKE_REQUEST */
1319 static inline int should_fail_request(struct bio
*bio
)
1324 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1327 * Check whether this bio extends beyond the end of the device.
1329 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
1336 /* Test device or partition size, when known. */
1337 maxsector
= bio
->bi_bdev
->bd_inode
->i_size
>> 9;
1339 sector_t sector
= bio
->bi_sector
;
1341 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
1343 * This may well happen - the kernel calls bread()
1344 * without checking the size of the device, e.g., when
1345 * mounting a device.
1347 handle_bad_sector(bio
);
1356 * generic_make_request - hand a buffer to its device driver for I/O
1357 * @bio: The bio describing the location in memory and on the device.
1359 * generic_make_request() is used to make I/O requests of block
1360 * devices. It is passed a &struct bio, which describes the I/O that needs
1363 * generic_make_request() does not return any status. The
1364 * success/failure status of the request, along with notification of
1365 * completion, is delivered asynchronously through the bio->bi_end_io
1366 * function described (one day) else where.
1368 * The caller of generic_make_request must make sure that bi_io_vec
1369 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1370 * set to describe the device address, and the
1371 * bi_end_io and optionally bi_private are set to describe how
1372 * completion notification should be signaled.
1374 * generic_make_request and the drivers it calls may use bi_next if this
1375 * bio happens to be merged with someone else, and may change bi_dev and
1376 * bi_sector for remaps as it sees fit. So the values of these fields
1377 * should NOT be depended on after the call to generic_make_request.
1379 static inline void __generic_make_request(struct bio
*bio
)
1381 struct request_queue
*q
;
1382 sector_t old_sector
;
1383 int ret
, nr_sectors
= bio_sectors(bio
);
1389 if (bio_check_eod(bio
, nr_sectors
))
1393 * Resolve the mapping until finished. (drivers are
1394 * still free to implement/resolve their own stacking
1395 * by explicitly returning 0)
1397 * NOTE: we don't repeat the blk_size check for each new device.
1398 * Stacking drivers are expected to know what they are doing.
1403 char b
[BDEVNAME_SIZE
];
1405 q
= bdev_get_queue(bio
->bi_bdev
);
1408 "generic_make_request: Trying to access "
1409 "nonexistent block-device %s (%Lu)\n",
1410 bdevname(bio
->bi_bdev
, b
),
1411 (long long) bio
->bi_sector
);
1415 if (unlikely(nr_sectors
> q
->max_hw_sectors
)) {
1416 printk(KERN_ERR
"bio too big device %s (%u > %u)\n",
1417 bdevname(bio
->bi_bdev
, b
),
1423 if (unlikely(test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)))
1426 if (should_fail_request(bio
))
1430 * If this device has partitions, remap block n
1431 * of partition p to block n+start(p) of the disk.
1433 blk_partition_remap(bio
);
1435 if (bio_integrity_enabled(bio
) && bio_integrity_prep(bio
))
1438 if (old_sector
!= -1)
1439 trace_block_remap(q
, bio
, old_dev
, bio
->bi_sector
,
1442 trace_block_bio_queue(q
, bio
);
1444 old_sector
= bio
->bi_sector
;
1445 old_dev
= bio
->bi_bdev
->bd_dev
;
1447 if (bio_check_eod(bio
, nr_sectors
))
1450 if (bio_discard(bio
) && !q
->prepare_discard_fn
) {
1455 ret
= q
->make_request_fn(q
, bio
);
1461 bio_endio(bio
, err
);
1465 * We only want one ->make_request_fn to be active at a time,
1466 * else stack usage with stacked devices could be a problem.
1467 * So use current->bio_{list,tail} to keep a list of requests
1468 * submited by a make_request_fn function.
1469 * current->bio_tail is also used as a flag to say if
1470 * generic_make_request is currently active in this task or not.
1471 * If it is NULL, then no make_request is active. If it is non-NULL,
1472 * then a make_request is active, and new requests should be added
1475 void generic_make_request(struct bio
*bio
)
1477 if (current
->bio_tail
) {
1478 /* make_request is active */
1479 *(current
->bio_tail
) = bio
;
1480 bio
->bi_next
= NULL
;
1481 current
->bio_tail
= &bio
->bi_next
;
1484 /* following loop may be a bit non-obvious, and so deserves some
1486 * Before entering the loop, bio->bi_next is NULL (as all callers
1487 * ensure that) so we have a list with a single bio.
1488 * We pretend that we have just taken it off a longer list, so
1489 * we assign bio_list to the next (which is NULL) and bio_tail
1490 * to &bio_list, thus initialising the bio_list of new bios to be
1491 * added. __generic_make_request may indeed add some more bios
1492 * through a recursive call to generic_make_request. If it
1493 * did, we find a non-NULL value in bio_list and re-enter the loop
1494 * from the top. In this case we really did just take the bio
1495 * of the top of the list (no pretending) and so fixup bio_list and
1496 * bio_tail or bi_next, and call into __generic_make_request again.
1498 * The loop was structured like this to make only one call to
1499 * __generic_make_request (which is important as it is large and
1500 * inlined) and to keep the structure simple.
1502 BUG_ON(bio
->bi_next
);
1504 current
->bio_list
= bio
->bi_next
;
1505 if (bio
->bi_next
== NULL
)
1506 current
->bio_tail
= ¤t
->bio_list
;
1508 bio
->bi_next
= NULL
;
1509 __generic_make_request(bio
);
1510 bio
= current
->bio_list
;
1512 current
->bio_tail
= NULL
; /* deactivate */
1514 EXPORT_SYMBOL(generic_make_request
);
1517 * submit_bio - submit a bio to the block device layer for I/O
1518 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1519 * @bio: The &struct bio which describes the I/O
1521 * submit_bio() is very similar in purpose to generic_make_request(), and
1522 * uses that function to do most of the work. Both are fairly rough
1523 * interfaces; @bio must be presetup and ready for I/O.
1526 void submit_bio(int rw
, struct bio
*bio
)
1528 int count
= bio_sectors(bio
);
1533 * If it's a regular read/write or a barrier with data attached,
1534 * go through the normal accounting stuff before submission.
1536 if (bio_has_data(bio
)) {
1538 count_vm_events(PGPGOUT
, count
);
1540 task_io_account_read(bio
->bi_size
);
1541 count_vm_events(PGPGIN
, count
);
1544 if (unlikely(block_dump
)) {
1545 char b
[BDEVNAME_SIZE
];
1546 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s\n",
1547 current
->comm
, task_pid_nr(current
),
1548 (rw
& WRITE
) ? "WRITE" : "READ",
1549 (unsigned long long)bio
->bi_sector
,
1550 bdevname(bio
->bi_bdev
, b
));
1554 generic_make_request(bio
);
1556 EXPORT_SYMBOL(submit_bio
);
1559 * blk_rq_check_limits - Helper function to check a request for the queue limit
1561 * @rq: the request being checked
1564 * @rq may have been made based on weaker limitations of upper-level queues
1565 * in request stacking drivers, and it may violate the limitation of @q.
1566 * Since the block layer and the underlying device driver trust @rq
1567 * after it is inserted to @q, it should be checked against @q before
1568 * the insertion using this generic function.
1570 * This function should also be useful for request stacking drivers
1571 * in some cases below, so export this fuction.
1572 * Request stacking drivers like request-based dm may change the queue
1573 * limits while requests are in the queue (e.g. dm's table swapping).
1574 * Such request stacking drivers should check those requests agaist
1575 * the new queue limits again when they dispatch those requests,
1576 * although such checkings are also done against the old queue limits
1577 * when submitting requests.
1579 int blk_rq_check_limits(struct request_queue
*q
, struct request
*rq
)
1581 if (rq
->nr_sectors
> q
->max_sectors
||
1582 rq
->data_len
> q
->max_hw_sectors
<< 9) {
1583 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
1588 * queue's settings related to segment counting like q->bounce_pfn
1589 * may differ from that of other stacking queues.
1590 * Recalculate it to check the request correctly on this queue's
1593 blk_recalc_rq_segments(rq
);
1594 if (rq
->nr_phys_segments
> q
->max_phys_segments
||
1595 rq
->nr_phys_segments
> q
->max_hw_segments
) {
1596 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
1602 EXPORT_SYMBOL_GPL(blk_rq_check_limits
);
1605 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1606 * @q: the queue to submit the request
1607 * @rq: the request being queued
1609 int blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
1611 unsigned long flags
;
1613 if (blk_rq_check_limits(q
, rq
))
1616 #ifdef CONFIG_FAIL_MAKE_REQUEST
1617 if (rq
->rq_disk
&& rq
->rq_disk
->part0
.make_it_fail
&&
1618 should_fail(&fail_make_request
, blk_rq_bytes(rq
)))
1622 spin_lock_irqsave(q
->queue_lock
, flags
);
1625 * Submitting request must be dequeued before calling this function
1626 * because it will be linked to another request_queue
1628 BUG_ON(blk_queued_rq(rq
));
1630 drive_stat_acct(rq
, 1);
1631 __elv_add_request(q
, rq
, ELEVATOR_INSERT_BACK
, 0);
1633 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1637 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
1640 * blkdev_dequeue_request - dequeue request and start timeout timer
1641 * @req: request to dequeue
1643 * Dequeue @req and start timeout timer on it. This hands off the
1644 * request to the driver.
1646 * Block internal functions which don't want to start timer should
1647 * call elv_dequeue_request().
1649 void blkdev_dequeue_request(struct request
*req
)
1651 elv_dequeue_request(req
->q
, req
);
1654 * We are now handing the request to the hardware, add the
1659 EXPORT_SYMBOL(blkdev_dequeue_request
);
1662 * __end_that_request_first - end I/O on a request
1663 * @req: the request being processed
1664 * @error: %0 for success, < %0 for error
1665 * @nr_bytes: number of bytes to complete
1668 * Ends I/O on a number of bytes attached to @req, and sets it up
1669 * for the next range of segments (if any) in the cluster.
1672 * %0 - we are done with this request, call end_that_request_last()
1673 * %1 - still buffers pending for this request
1675 static int __end_that_request_first(struct request
*req
, int error
,
1678 int total_bytes
, bio_nbytes
, next_idx
= 0;
1681 trace_block_rq_complete(req
->q
, req
);
1684 * for a REQ_TYPE_BLOCK_PC request, we want to carry any eventual
1685 * sense key with us all the way through
1687 if (!blk_pc_request(req
))
1690 if (error
&& (blk_fs_request(req
) && !(req
->cmd_flags
& REQ_QUIET
))) {
1691 printk(KERN_ERR
"end_request: I/O error, dev %s, sector %llu\n",
1692 req
->rq_disk
? req
->rq_disk
->disk_name
: "?",
1693 (unsigned long long)req
->sector
);
1696 if (blk_fs_request(req
) && req
->rq_disk
) {
1697 const int rw
= rq_data_dir(req
);
1698 struct hd_struct
*part
;
1701 cpu
= part_stat_lock();
1702 part
= disk_map_sector_rcu(req
->rq_disk
, req
->sector
);
1703 part_stat_add(cpu
, part
, sectors
[rw
], nr_bytes
>> 9);
1707 total_bytes
= bio_nbytes
= 0;
1708 while ((bio
= req
->bio
) != NULL
) {
1711 if (nr_bytes
>= bio
->bi_size
) {
1712 req
->bio
= bio
->bi_next
;
1713 nbytes
= bio
->bi_size
;
1714 req_bio_endio(req
, bio
, nbytes
, error
);
1718 int idx
= bio
->bi_idx
+ next_idx
;
1720 if (unlikely(bio
->bi_idx
>= bio
->bi_vcnt
)) {
1721 blk_dump_rq_flags(req
, "__end_that");
1722 printk(KERN_ERR
"%s: bio idx %d >= vcnt %d\n",
1723 __func__
, bio
->bi_idx
, bio
->bi_vcnt
);
1727 nbytes
= bio_iovec_idx(bio
, idx
)->bv_len
;
1728 BIO_BUG_ON(nbytes
> bio
->bi_size
);
1731 * not a complete bvec done
1733 if (unlikely(nbytes
> nr_bytes
)) {
1734 bio_nbytes
+= nr_bytes
;
1735 total_bytes
+= nr_bytes
;
1740 * advance to the next vector
1743 bio_nbytes
+= nbytes
;
1746 total_bytes
+= nbytes
;
1752 * end more in this run, or just return 'not-done'
1754 if (unlikely(nr_bytes
<= 0))
1766 * if the request wasn't completed, update state
1769 req_bio_endio(req
, bio
, bio_nbytes
, error
);
1770 bio
->bi_idx
+= next_idx
;
1771 bio_iovec(bio
)->bv_offset
+= nr_bytes
;
1772 bio_iovec(bio
)->bv_len
-= nr_bytes
;
1775 blk_recalc_rq_sectors(req
, total_bytes
>> 9);
1776 blk_recalc_rq_segments(req
);
1781 * queue lock must be held
1783 static void end_that_request_last(struct request
*req
, int error
)
1785 struct gendisk
*disk
= req
->rq_disk
;
1787 if (blk_rq_tagged(req
))
1788 blk_queue_end_tag(req
->q
, req
);
1790 if (blk_queued_rq(req
))
1791 elv_dequeue_request(req
->q
, req
);
1793 if (unlikely(laptop_mode
) && blk_fs_request(req
))
1794 laptop_io_completion();
1796 blk_delete_timer(req
);
1799 * Account IO completion. bar_rq isn't accounted as a normal
1800 * IO on queueing nor completion. Accounting the containing
1801 * request is enough.
1803 if (disk
&& blk_fs_request(req
) && req
!= &req
->q
->bar_rq
) {
1804 unsigned long duration
= jiffies
- req
->start_time
;
1805 const int rw
= rq_data_dir(req
);
1806 struct hd_struct
*part
;
1809 cpu
= part_stat_lock();
1810 part
= disk_map_sector_rcu(disk
, req
->sector
);
1812 part_stat_inc(cpu
, part
, ios
[rw
]);
1813 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
1814 part_round_stats(cpu
, part
);
1815 part_dec_in_flight(part
);
1821 req
->end_io(req
, error
);
1823 if (blk_bidi_rq(req
))
1824 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
1826 __blk_put_request(req
->q
, req
);
1831 * blk_rq_bytes - Returns bytes left to complete in the entire request
1832 * @rq: the request being processed
1834 unsigned int blk_rq_bytes(struct request
*rq
)
1836 if (blk_fs_request(rq
))
1837 return rq
->hard_nr_sectors
<< 9;
1839 return rq
->data_len
;
1841 EXPORT_SYMBOL_GPL(blk_rq_bytes
);
1844 * blk_rq_cur_bytes - Returns bytes left to complete in the current segment
1845 * @rq: the request being processed
1847 unsigned int blk_rq_cur_bytes(struct request
*rq
)
1849 if (blk_fs_request(rq
))
1850 return rq
->current_nr_sectors
<< 9;
1853 return rq
->bio
->bi_size
;
1855 return rq
->data_len
;
1857 EXPORT_SYMBOL_GPL(blk_rq_cur_bytes
);
1860 * end_request - end I/O on the current segment of the request
1861 * @req: the request being processed
1862 * @uptodate: error value or %0/%1 uptodate flag
1865 * Ends I/O on the current segment of a request. If that is the only
1866 * remaining segment, the request is also completed and freed.
1868 * This is a remnant of how older block drivers handled I/O completions.
1869 * Modern drivers typically end I/O on the full request in one go, unless
1870 * they have a residual value to account for. For that case this function
1871 * isn't really useful, unless the residual just happens to be the
1872 * full current segment. In other words, don't use this function in new
1873 * code. Use blk_end_request() or __blk_end_request() to end a request.
1875 void end_request(struct request
*req
, int uptodate
)
1880 error
= uptodate
? uptodate
: -EIO
;
1882 __blk_end_request(req
, error
, req
->hard_cur_sectors
<< 9);
1884 EXPORT_SYMBOL(end_request
);
1886 static int end_that_request_data(struct request
*rq
, int error
,
1887 unsigned int nr_bytes
, unsigned int bidi_bytes
)
1890 if (__end_that_request_first(rq
, error
, nr_bytes
))
1893 /* Bidi request must be completed as a whole */
1894 if (blk_bidi_rq(rq
) &&
1895 __end_that_request_first(rq
->next_rq
, error
, bidi_bytes
))
1903 * blk_end_io - Generic end_io function to complete a request.
1904 * @rq: the request being processed
1905 * @error: %0 for success, < %0 for error
1906 * @nr_bytes: number of bytes to complete @rq
1907 * @bidi_bytes: number of bytes to complete @rq->next_rq
1908 * @drv_callback: function called between completion of bios in the request
1909 * and completion of the request.
1910 * If the callback returns non %0, this helper returns without
1911 * completion of the request.
1914 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
1915 * If @rq has leftover, sets it up for the next range of segments.
1918 * %0 - we are done with this request
1919 * %1 - this request is not freed yet, it still has pending buffers.
1921 static int blk_end_io(struct request
*rq
, int error
, unsigned int nr_bytes
,
1922 unsigned int bidi_bytes
,
1923 int (drv_callback
)(struct request
*))
1925 struct request_queue
*q
= rq
->q
;
1926 unsigned long flags
= 0UL;
1928 if (end_that_request_data(rq
, error
, nr_bytes
, bidi_bytes
))
1931 /* Special feature for tricky drivers */
1932 if (drv_callback
&& drv_callback(rq
))
1935 add_disk_randomness(rq
->rq_disk
);
1937 spin_lock_irqsave(q
->queue_lock
, flags
);
1938 end_that_request_last(rq
, error
);
1939 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1945 * blk_end_request - Helper function for drivers to complete the request.
1946 * @rq: the request being processed
1947 * @error: %0 for success, < %0 for error
1948 * @nr_bytes: number of bytes to complete
1951 * Ends I/O on a number of bytes attached to @rq.
1952 * If @rq has leftover, sets it up for the next range of segments.
1955 * %0 - we are done with this request
1956 * %1 - still buffers pending for this request
1958 int blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
1960 return blk_end_io(rq
, error
, nr_bytes
, 0, NULL
);
1962 EXPORT_SYMBOL_GPL(blk_end_request
);
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 * Must be called with queue lock held unlike blk_end_request().
1974 * %0 - we are done with this request
1975 * %1 - still buffers pending for this request
1977 int __blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
1979 if (rq
->bio
&& __end_that_request_first(rq
, error
, nr_bytes
))
1982 add_disk_randomness(rq
->rq_disk
);
1984 end_that_request_last(rq
, error
);
1988 EXPORT_SYMBOL_GPL(__blk_end_request
);
1991 * blk_end_bidi_request - Helper function for drivers to complete bidi request.
1992 * @rq: the bidi request being processed
1993 * @error: %0 for success, < %0 for error
1994 * @nr_bytes: number of bytes to complete @rq
1995 * @bidi_bytes: number of bytes to complete @rq->next_rq
1998 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2001 * %0 - we are done with this request
2002 * %1 - still buffers pending for this request
2004 int blk_end_bidi_request(struct request
*rq
, int error
, unsigned int nr_bytes
,
2005 unsigned int bidi_bytes
)
2007 return blk_end_io(rq
, error
, nr_bytes
, bidi_bytes
, NULL
);
2009 EXPORT_SYMBOL_GPL(blk_end_bidi_request
);
2012 * blk_update_request - Special helper function for request stacking drivers
2013 * @rq: the request being processed
2014 * @error: %0 for success, < %0 for error
2015 * @nr_bytes: number of bytes to complete @rq
2018 * Ends I/O on a number of bytes attached to @rq, but doesn't complete
2019 * the request structure even if @rq doesn't have leftover.
2020 * If @rq has leftover, sets it up for the next range of segments.
2022 * This special helper function is only for request stacking drivers
2023 * (e.g. request-based dm) so that they can handle partial completion.
2024 * Actual device drivers should use blk_end_request instead.
2026 void blk_update_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2028 if (!end_that_request_data(rq
, error
, nr_bytes
, 0)) {
2030 * These members are not updated in end_that_request_data()
2031 * when all bios are completed.
2032 * Update them so that the request stacking driver can find
2033 * how many bytes remain in the request later.
2035 rq
->nr_sectors
= rq
->hard_nr_sectors
= 0;
2036 rq
->current_nr_sectors
= rq
->hard_cur_sectors
= 0;
2039 EXPORT_SYMBOL_GPL(blk_update_request
);
2042 * blk_end_request_callback - Special helper function for tricky drivers
2043 * @rq: the request being processed
2044 * @error: %0 for success, < %0 for error
2045 * @nr_bytes: number of bytes to complete
2046 * @drv_callback: function called between completion of bios in the request
2047 * and completion of the request.
2048 * If the callback returns non %0, this helper returns without
2049 * completion of the request.
2052 * Ends I/O on a number of bytes attached to @rq.
2053 * If @rq has leftover, sets it up for the next range of segments.
2055 * This special helper function is used only for existing tricky drivers.
2056 * (e.g. cdrom_newpc_intr() of ide-cd)
2057 * This interface will be removed when such drivers are rewritten.
2058 * Don't use this interface in other places anymore.
2061 * %0 - we are done with this request
2062 * %1 - this request is not freed yet.
2063 * this request still has pending buffers or
2064 * the driver doesn't want to finish this request yet.
2066 int blk_end_request_callback(struct request
*rq
, int error
,
2067 unsigned int nr_bytes
,
2068 int (drv_callback
)(struct request
*))
2070 return blk_end_io(rq
, error
, nr_bytes
, 0, drv_callback
);
2072 EXPORT_SYMBOL_GPL(blk_end_request_callback
);
2074 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
2077 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw, and
2078 we want BIO_RW_AHEAD (bit 1) to imply REQ_FAILFAST (bit 1). */
2079 rq
->cmd_flags
|= (bio
->bi_rw
& 3);
2081 if (bio_has_data(bio
)) {
2082 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2083 rq
->buffer
= bio_data(bio
);
2085 rq
->current_nr_sectors
= bio_cur_sectors(bio
);
2086 rq
->hard_cur_sectors
= rq
->current_nr_sectors
;
2087 rq
->hard_nr_sectors
= rq
->nr_sectors
= bio_sectors(bio
);
2088 rq
->data_len
= bio
->bi_size
;
2090 rq
->bio
= rq
->biotail
= bio
;
2093 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2097 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2098 * @q : the queue of the device being checked
2101 * Check if underlying low-level drivers of a device are busy.
2102 * If the drivers want to export their busy state, they must set own
2103 * exporting function using blk_queue_lld_busy() first.
2105 * Basically, this function is used only by request stacking drivers
2106 * to stop dispatching requests to underlying devices when underlying
2107 * devices are busy. This behavior helps more I/O merging on the queue
2108 * of the request stacking driver and prevents I/O throughput regression
2109 * on burst I/O load.
2112 * 0 - Not busy (The request stacking driver should dispatch request)
2113 * 1 - Busy (The request stacking driver should stop dispatching request)
2115 int blk_lld_busy(struct request_queue
*q
)
2118 return q
->lld_busy_fn(q
);
2122 EXPORT_SYMBOL_GPL(blk_lld_busy
);
2124 int kblockd_schedule_work(struct request_queue
*q
, struct work_struct
*work
)
2126 return queue_work(kblockd_workqueue
, work
);
2128 EXPORT_SYMBOL(kblockd_schedule_work
);
2130 int __init
blk_dev_init(void)
2132 kblockd_workqueue
= create_workqueue("kblockd");
2133 if (!kblockd_workqueue
)
2134 panic("Failed to create kblockd\n");
2136 request_cachep
= kmem_cache_create("blkdev_requests",
2137 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
2139 blk_requestq_cachep
= kmem_cache_create("blkdev_queue",
2140 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);