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/interrupt.h>
30 #include <linux/cpu.h>
31 #include <linux/blktrace_api.h>
32 #include <linux/fault-inject.h>
36 static int __make_request(struct request_queue
*q
, struct bio
*bio
);
39 * For the allocated request tables
41 static struct kmem_cache
*request_cachep
;
44 * For queue allocation
46 struct kmem_cache
*blk_requestq_cachep
;
49 * Controlling structure to kblockd
51 static struct workqueue_struct
*kblockd_workqueue
;
53 static DEFINE_PER_CPU(struct list_head
, blk_cpu_done
);
55 static void drive_stat_acct(struct request
*rq
, int new_io
)
57 int rw
= rq_data_dir(rq
);
59 if (!blk_fs_request(rq
) || !rq
->rq_disk
)
63 __all_stat_inc(rq
->rq_disk
, merges
[rw
], rq
->sector
);
65 struct hd_struct
*part
= get_part(rq
->rq_disk
, rq
->sector
);
66 disk_round_stats(rq
->rq_disk
);
67 rq
->rq_disk
->in_flight
++;
69 part_round_stats(part
);
75 void blk_queue_congestion_threshold(struct request_queue
*q
)
79 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) + 1;
80 if (nr
> q
->nr_requests
)
82 q
->nr_congestion_on
= nr
;
84 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) - (q
->nr_requests
/ 16) - 1;
87 q
->nr_congestion_off
= nr
;
91 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
94 * Locates the passed device's request queue and returns the address of its
97 * Will return NULL if the request queue cannot be located.
99 struct backing_dev_info
*blk_get_backing_dev_info(struct block_device
*bdev
)
101 struct backing_dev_info
*ret
= NULL
;
102 struct request_queue
*q
= bdev_get_queue(bdev
);
105 ret
= &q
->backing_dev_info
;
108 EXPORT_SYMBOL(blk_get_backing_dev_info
);
110 void blk_rq_init(struct request_queue
*q
, struct request
*rq
)
112 memset(rq
, 0, sizeof(*rq
));
114 INIT_LIST_HEAD(&rq
->queuelist
);
115 INIT_LIST_HEAD(&rq
->donelist
);
117 rq
->sector
= rq
->hard_sector
= (sector_t
) -1;
118 INIT_HLIST_NODE(&rq
->hash
);
119 RB_CLEAR_NODE(&rq
->rb_node
);
124 EXPORT_SYMBOL(blk_rq_init
);
126 static void req_bio_endio(struct request
*rq
, struct bio
*bio
,
127 unsigned int nbytes
, int error
)
129 struct request_queue
*q
= rq
->q
;
131 if (&q
->bar_rq
!= rq
) {
133 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
134 else if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
137 if (unlikely(nbytes
> bio
->bi_size
)) {
138 printk(KERN_ERR
"%s: want %u bytes done, %u left\n",
139 __func__
, nbytes
, bio
->bi_size
);
140 nbytes
= bio
->bi_size
;
143 bio
->bi_size
-= nbytes
;
144 bio
->bi_sector
+= (nbytes
>> 9);
145 if (bio
->bi_size
== 0)
146 bio_endio(bio
, error
);
150 * Okay, this is the barrier request in progress, just
153 if (error
&& !q
->orderr
)
158 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
162 printk(KERN_INFO
"%s: dev %s: type=%x, flags=%x\n", msg
,
163 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->cmd_type
,
166 printk(KERN_INFO
" sector %llu, nr/cnr %lu/%u\n",
167 (unsigned long long)rq
->sector
,
169 rq
->current_nr_sectors
);
170 printk(KERN_INFO
" bio %p, biotail %p, buffer %p, data %p, len %u\n",
171 rq
->bio
, rq
->biotail
,
172 rq
->buffer
, rq
->data
,
175 if (blk_pc_request(rq
)) {
176 printk(KERN_INFO
" cdb: ");
177 for (bit
= 0; bit
< BLK_MAX_CDB
; bit
++)
178 printk("%02x ", rq
->cmd
[bit
]);
182 EXPORT_SYMBOL(blk_dump_rq_flags
);
185 * "plug" the device if there are no outstanding requests: this will
186 * force the transfer to start only after we have put all the requests
189 * This is called with interrupts off and no requests on the queue and
190 * with the queue lock held.
192 void blk_plug_device(struct request_queue
*q
)
194 WARN_ON(!irqs_disabled());
197 * don't plug a stopped queue, it must be paired with blk_start_queue()
198 * which will restart the queueing
200 if (blk_queue_stopped(q
))
203 if (!test_bit(QUEUE_FLAG_PLUGGED
, &q
->queue_flags
)) {
204 __set_bit(QUEUE_FLAG_PLUGGED
, &q
->queue_flags
);
205 mod_timer(&q
->unplug_timer
, jiffies
+ q
->unplug_delay
);
206 blk_add_trace_generic(q
, NULL
, 0, BLK_TA_PLUG
);
209 EXPORT_SYMBOL(blk_plug_device
);
212 * remove the queue from the plugged list, if present. called with
213 * queue lock held and interrupts disabled.
215 int blk_remove_plug(struct request_queue
*q
)
217 WARN_ON(!irqs_disabled());
219 if (!test_bit(QUEUE_FLAG_PLUGGED
, &q
->queue_flags
))
222 queue_flag_clear(QUEUE_FLAG_PLUGGED
, q
);
223 del_timer(&q
->unplug_timer
);
226 EXPORT_SYMBOL(blk_remove_plug
);
229 * remove the plug and let it rip..
231 void __generic_unplug_device(struct request_queue
*q
)
233 if (unlikely(blk_queue_stopped(q
)))
236 if (!blk_remove_plug(q
))
241 EXPORT_SYMBOL(__generic_unplug_device
);
244 * generic_unplug_device - fire a request queue
245 * @q: The &struct request_queue in question
248 * Linux uses plugging to build bigger requests queues before letting
249 * the device have at them. If a queue is plugged, the I/O scheduler
250 * is still adding and merging requests on the queue. Once the queue
251 * gets unplugged, the request_fn defined for the queue is invoked and
254 void generic_unplug_device(struct request_queue
*q
)
256 if (blk_queue_plugged(q
)) {
257 spin_lock_irq(q
->queue_lock
);
258 __generic_unplug_device(q
);
259 spin_unlock_irq(q
->queue_lock
);
262 EXPORT_SYMBOL(generic_unplug_device
);
264 static void blk_backing_dev_unplug(struct backing_dev_info
*bdi
,
267 struct request_queue
*q
= bdi
->unplug_io_data
;
272 void blk_unplug_work(struct work_struct
*work
)
274 struct request_queue
*q
=
275 container_of(work
, struct request_queue
, unplug_work
);
277 blk_add_trace_pdu_int(q
, BLK_TA_UNPLUG_IO
, NULL
,
278 q
->rq
.count
[READ
] + q
->rq
.count
[WRITE
]);
283 void blk_unplug_timeout(unsigned long data
)
285 struct request_queue
*q
= (struct request_queue
*)data
;
287 blk_add_trace_pdu_int(q
, BLK_TA_UNPLUG_TIMER
, NULL
,
288 q
->rq
.count
[READ
] + q
->rq
.count
[WRITE
]);
290 kblockd_schedule_work(&q
->unplug_work
);
293 void blk_unplug(struct request_queue
*q
)
296 * devices don't necessarily have an ->unplug_fn defined
299 blk_add_trace_pdu_int(q
, BLK_TA_UNPLUG_IO
, NULL
,
300 q
->rq
.count
[READ
] + q
->rq
.count
[WRITE
]);
305 EXPORT_SYMBOL(blk_unplug
);
308 * blk_start_queue - restart a previously stopped queue
309 * @q: The &struct request_queue in question
312 * blk_start_queue() will clear the stop flag on the queue, and call
313 * the request_fn for the queue if it was in a stopped state when
314 * entered. Also see blk_stop_queue(). Queue lock must be held.
316 void blk_start_queue(struct request_queue
*q
)
318 WARN_ON(!irqs_disabled());
320 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
323 * one level of recursion is ok and is much faster than kicking
324 * the unplug handling
326 if (!test_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
)) {
327 queue_flag_set(QUEUE_FLAG_REENTER
, q
);
329 queue_flag_clear(QUEUE_FLAG_REENTER
, q
);
332 kblockd_schedule_work(&q
->unplug_work
);
335 EXPORT_SYMBOL(blk_start_queue
);
338 * blk_stop_queue - stop a queue
339 * @q: The &struct request_queue in question
342 * The Linux block layer assumes that a block driver will consume all
343 * entries on the request queue when the request_fn strategy is called.
344 * Often this will not happen, because of hardware limitations (queue
345 * depth settings). If a device driver gets a 'queue full' response,
346 * or if it simply chooses not to queue more I/O at one point, it can
347 * call this function to prevent the request_fn from being called until
348 * the driver has signalled it's ready to go again. This happens by calling
349 * blk_start_queue() to restart queue operations. Queue lock must be held.
351 void blk_stop_queue(struct request_queue
*q
)
354 queue_flag_set(QUEUE_FLAG_STOPPED
, q
);
356 EXPORT_SYMBOL(blk_stop_queue
);
359 * blk_sync_queue - cancel any pending callbacks on a queue
363 * The block layer may perform asynchronous callback activity
364 * on a queue, such as calling the unplug function after a timeout.
365 * A block device may call blk_sync_queue to ensure that any
366 * such activity is cancelled, thus allowing it to release resources
367 * that the callbacks might use. The caller must already have made sure
368 * that its ->make_request_fn will not re-add plugging prior to calling
372 void blk_sync_queue(struct request_queue
*q
)
374 del_timer_sync(&q
->unplug_timer
);
375 kblockd_flush_work(&q
->unplug_work
);
377 EXPORT_SYMBOL(blk_sync_queue
);
380 * blk_run_queue - run a single device queue
381 * @q: The queue to run
383 void __blk_run_queue(struct request_queue
*q
)
388 * Only recurse once to avoid overrunning the stack, let the unplug
389 * handling reinvoke the handler shortly if we already got there.
391 if (!elv_queue_empty(q
)) {
392 if (!test_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
)) {
393 queue_flag_set(QUEUE_FLAG_REENTER
, q
);
395 queue_flag_clear(QUEUE_FLAG_REENTER
, q
);
398 kblockd_schedule_work(&q
->unplug_work
);
402 EXPORT_SYMBOL(__blk_run_queue
);
405 * blk_run_queue - run a single device queue
406 * @q: The queue to run
408 void blk_run_queue(struct request_queue
*q
)
412 spin_lock_irqsave(q
->queue_lock
, flags
);
414 spin_unlock_irqrestore(q
->queue_lock
, flags
);
416 EXPORT_SYMBOL(blk_run_queue
);
418 void blk_put_queue(struct request_queue
*q
)
420 kobject_put(&q
->kobj
);
423 void blk_cleanup_queue(struct request_queue
*q
)
425 mutex_lock(&q
->sysfs_lock
);
426 queue_flag_set_unlocked(QUEUE_FLAG_DEAD
, q
);
427 mutex_unlock(&q
->sysfs_lock
);
430 elevator_exit(q
->elevator
);
434 EXPORT_SYMBOL(blk_cleanup_queue
);
436 static int blk_init_free_list(struct request_queue
*q
)
438 struct request_list
*rl
= &q
->rq
;
440 rl
->count
[READ
] = rl
->count
[WRITE
] = 0;
441 rl
->starved
[READ
] = rl
->starved
[WRITE
] = 0;
443 init_waitqueue_head(&rl
->wait
[READ
]);
444 init_waitqueue_head(&rl
->wait
[WRITE
]);
446 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, mempool_alloc_slab
,
447 mempool_free_slab
, request_cachep
, q
->node
);
455 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
457 return blk_alloc_queue_node(gfp_mask
, -1);
459 EXPORT_SYMBOL(blk_alloc_queue
);
461 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
463 struct request_queue
*q
;
466 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
467 gfp_mask
| __GFP_ZERO
, node_id
);
471 q
->backing_dev_info
.unplug_io_fn
= blk_backing_dev_unplug
;
472 q
->backing_dev_info
.unplug_io_data
= q
;
473 err
= bdi_init(&q
->backing_dev_info
);
475 kmem_cache_free(blk_requestq_cachep
, q
);
479 init_timer(&q
->unplug_timer
);
481 kobject_init(&q
->kobj
, &blk_queue_ktype
);
483 mutex_init(&q
->sysfs_lock
);
487 EXPORT_SYMBOL(blk_alloc_queue_node
);
490 * blk_init_queue - prepare a request queue for use with a block device
491 * @rfn: The function to be called to process requests that have been
492 * placed on the queue.
493 * @lock: Request queue spin lock
496 * If a block device wishes to use the standard request handling procedures,
497 * which sorts requests and coalesces adjacent requests, then it must
498 * call blk_init_queue(). The function @rfn will be called when there
499 * are requests on the queue that need to be processed. If the device
500 * supports plugging, then @rfn may not be called immediately when requests
501 * are available on the queue, but may be called at some time later instead.
502 * Plugged queues are generally unplugged when a buffer belonging to one
503 * of the requests on the queue is needed, or due to memory pressure.
505 * @rfn is not required, or even expected, to remove all requests off the
506 * queue, but only as many as it can handle at a time. If it does leave
507 * requests on the queue, it is responsible for arranging that the requests
508 * get dealt with eventually.
510 * The queue spin lock must be held while manipulating the requests on the
511 * request queue; this lock will be taken also from interrupt context, so irq
512 * disabling is needed for it.
514 * Function returns a pointer to the initialized request queue, or NULL if
518 * blk_init_queue() must be paired with a blk_cleanup_queue() call
519 * when the block device is deactivated (such as at module unload).
522 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
524 return blk_init_queue_node(rfn
, lock
, -1);
526 EXPORT_SYMBOL(blk_init_queue
);
528 struct request_queue
*
529 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
531 struct request_queue
*q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
537 if (blk_init_free_list(q
)) {
538 kmem_cache_free(blk_requestq_cachep
, q
);
543 * if caller didn't supply a lock, they get per-queue locking with
547 spin_lock_init(&q
->__queue_lock
);
548 lock
= &q
->__queue_lock
;
552 q
->prep_rq_fn
= NULL
;
553 q
->unplug_fn
= generic_unplug_device
;
554 q
->queue_flags
= (1 << QUEUE_FLAG_CLUSTER
);
555 q
->queue_lock
= lock
;
557 blk_queue_segment_boundary(q
, 0xffffffff);
559 blk_queue_make_request(q
, __make_request
);
560 blk_queue_max_segment_size(q
, MAX_SEGMENT_SIZE
);
562 blk_queue_max_hw_segments(q
, MAX_HW_SEGMENTS
);
563 blk_queue_max_phys_segments(q
, MAX_PHYS_SEGMENTS
);
565 q
->sg_reserved_size
= INT_MAX
;
570 if (!elevator_init(q
, NULL
)) {
571 blk_queue_congestion_threshold(q
);
578 EXPORT_SYMBOL(blk_init_queue_node
);
580 int blk_get_queue(struct request_queue
*q
)
582 if (likely(!test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
))) {
583 kobject_get(&q
->kobj
);
590 static inline void blk_free_request(struct request_queue
*q
, struct request
*rq
)
592 if (rq
->cmd_flags
& REQ_ELVPRIV
)
593 elv_put_request(q
, rq
);
594 mempool_free(rq
, q
->rq
.rq_pool
);
597 static struct request
*
598 blk_alloc_request(struct request_queue
*q
, int rw
, int priv
, gfp_t gfp_mask
)
600 struct request
*rq
= mempool_alloc(q
->rq
.rq_pool
, gfp_mask
);
608 * first three bits are identical in rq->cmd_flags and bio->bi_rw,
609 * see bio.h and blkdev.h
611 rq
->cmd_flags
= rw
| REQ_ALLOCED
;
614 if (unlikely(elv_set_request(q
, rq
, gfp_mask
))) {
615 mempool_free(rq
, q
->rq
.rq_pool
);
618 rq
->cmd_flags
|= REQ_ELVPRIV
;
625 * ioc_batching returns true if the ioc is a valid batching request and
626 * should be given priority access to a request.
628 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
634 * Make sure the process is able to allocate at least 1 request
635 * even if the batch times out, otherwise we could theoretically
638 return ioc
->nr_batch_requests
== q
->nr_batching
||
639 (ioc
->nr_batch_requests
> 0
640 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
644 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
645 * will cause the process to be a "batcher" on all queues in the system. This
646 * is the behaviour we want though - once it gets a wakeup it should be given
649 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
651 if (!ioc
|| ioc_batching(q
, ioc
))
654 ioc
->nr_batch_requests
= q
->nr_batching
;
655 ioc
->last_waited
= jiffies
;
658 static void __freed_request(struct request_queue
*q
, int rw
)
660 struct request_list
*rl
= &q
->rq
;
662 if (rl
->count
[rw
] < queue_congestion_off_threshold(q
))
663 blk_clear_queue_congested(q
, rw
);
665 if (rl
->count
[rw
] + 1 <= q
->nr_requests
) {
666 if (waitqueue_active(&rl
->wait
[rw
]))
667 wake_up(&rl
->wait
[rw
]);
669 blk_clear_queue_full(q
, rw
);
674 * A request has just been released. Account for it, update the full and
675 * congestion status, wake up any waiters. Called under q->queue_lock.
677 static void freed_request(struct request_queue
*q
, int rw
, int priv
)
679 struct request_list
*rl
= &q
->rq
;
685 __freed_request(q
, rw
);
687 if (unlikely(rl
->starved
[rw
^ 1]))
688 __freed_request(q
, rw
^ 1);
691 #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
693 * Get a free request, queue_lock must be held.
694 * Returns NULL on failure, with queue_lock held.
695 * Returns !NULL on success, with queue_lock *not held*.
697 static struct request
*get_request(struct request_queue
*q
, int rw_flags
,
698 struct bio
*bio
, gfp_t gfp_mask
)
700 struct request
*rq
= NULL
;
701 struct request_list
*rl
= &q
->rq
;
702 struct io_context
*ioc
= NULL
;
703 const int rw
= rw_flags
& 0x01;
706 may_queue
= elv_may_queue(q
, rw_flags
);
707 if (may_queue
== ELV_MQUEUE_NO
)
710 if (rl
->count
[rw
]+1 >= queue_congestion_on_threshold(q
)) {
711 if (rl
->count
[rw
]+1 >= q
->nr_requests
) {
712 ioc
= current_io_context(GFP_ATOMIC
, q
->node
);
714 * The queue will fill after this allocation, so set
715 * it as full, and mark this process as "batching".
716 * This process will be allowed to complete a batch of
717 * requests, others will be blocked.
719 if (!blk_queue_full(q
, rw
)) {
720 ioc_set_batching(q
, ioc
);
721 blk_set_queue_full(q
, rw
);
723 if (may_queue
!= ELV_MQUEUE_MUST
724 && !ioc_batching(q
, ioc
)) {
726 * The queue is full and the allocating
727 * process is not a "batcher", and not
728 * exempted by the IO scheduler
734 blk_set_queue_congested(q
, rw
);
738 * Only allow batching queuers to allocate up to 50% over the defined
739 * limit of requests, otherwise we could have thousands of requests
740 * allocated with any setting of ->nr_requests
742 if (rl
->count
[rw
] >= (3 * q
->nr_requests
/ 2))
748 priv
= !test_bit(QUEUE_FLAG_ELVSWITCH
, &q
->queue_flags
);
752 spin_unlock_irq(q
->queue_lock
);
754 rq
= blk_alloc_request(q
, rw_flags
, priv
, gfp_mask
);
757 * Allocation failed presumably due to memory. Undo anything
758 * we might have messed up.
760 * Allocating task should really be put onto the front of the
761 * wait queue, but this is pretty rare.
763 spin_lock_irq(q
->queue_lock
);
764 freed_request(q
, rw
, priv
);
767 * in the very unlikely event that allocation failed and no
768 * requests for this direction was pending, mark us starved
769 * so that freeing of a request in the other direction will
770 * notice us. another possible fix would be to split the
771 * rq mempool into READ and WRITE
774 if (unlikely(rl
->count
[rw
] == 0))
781 * ioc may be NULL here, and ioc_batching will be false. That's
782 * OK, if the queue is under the request limit then requests need
783 * not count toward the nr_batch_requests limit. There will always
784 * be some limit enforced by BLK_BATCH_TIME.
786 if (ioc_batching(q
, ioc
))
787 ioc
->nr_batch_requests
--;
789 blk_add_trace_generic(q
, bio
, rw
, BLK_TA_GETRQ
);
795 * No available requests for this queue, unplug the device and wait for some
796 * requests to become available.
798 * Called with q->queue_lock held, and returns with it unlocked.
800 static struct request
*get_request_wait(struct request_queue
*q
, int rw_flags
,
803 const int rw
= rw_flags
& 0x01;
806 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
809 struct request_list
*rl
= &q
->rq
;
811 prepare_to_wait_exclusive(&rl
->wait
[rw
], &wait
,
812 TASK_UNINTERRUPTIBLE
);
814 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
817 struct io_context
*ioc
;
819 blk_add_trace_generic(q
, bio
, rw
, BLK_TA_SLEEPRQ
);
821 __generic_unplug_device(q
);
822 spin_unlock_irq(q
->queue_lock
);
826 * After sleeping, we become a "batching" process and
827 * will be able to allocate at least one request, and
828 * up to a big batch of them for a small period time.
829 * See ioc_batching, ioc_set_batching
831 ioc
= current_io_context(GFP_NOIO
, q
->node
);
832 ioc_set_batching(q
, ioc
);
834 spin_lock_irq(q
->queue_lock
);
836 finish_wait(&rl
->wait
[rw
], &wait
);
842 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
846 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
848 spin_lock_irq(q
->queue_lock
);
849 if (gfp_mask
& __GFP_WAIT
) {
850 rq
= get_request_wait(q
, rw
, NULL
);
852 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
854 spin_unlock_irq(q
->queue_lock
);
856 /* q->queue_lock is unlocked at this point */
860 EXPORT_SYMBOL(blk_get_request
);
863 * blk_start_queueing - initiate dispatch of requests to device
864 * @q: request queue to kick into gear
866 * This is basically a helper to remove the need to know whether a queue
867 * is plugged or not if someone just wants to initiate dispatch of requests
870 * The queue lock must be held with interrupts disabled.
872 void blk_start_queueing(struct request_queue
*q
)
874 if (!blk_queue_plugged(q
))
877 __generic_unplug_device(q
);
879 EXPORT_SYMBOL(blk_start_queueing
);
882 * blk_requeue_request - put a request back on queue
883 * @q: request queue where request should be inserted
884 * @rq: request to be inserted
887 * Drivers often keep queueing requests until the hardware cannot accept
888 * more, when that condition happens we need to put the request back
889 * on the queue. Must be called with queue lock held.
891 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
893 blk_add_trace_rq(q
, rq
, BLK_TA_REQUEUE
);
895 if (blk_rq_tagged(rq
))
896 blk_queue_end_tag(q
, rq
);
898 elv_requeue_request(q
, rq
);
900 EXPORT_SYMBOL(blk_requeue_request
);
903 * blk_insert_request - insert a special request in to a request queue
904 * @q: request queue where request should be inserted
905 * @rq: request to be inserted
906 * @at_head: insert request at head or tail of queue
907 * @data: private data
910 * Many block devices need to execute commands asynchronously, so they don't
911 * block the whole kernel from preemption during request execution. This is
912 * accomplished normally by inserting aritficial requests tagged as
913 * REQ_SPECIAL in to the corresponding request queue, and letting them be
914 * scheduled for actual execution by the request queue.
916 * We have the option of inserting the head or the tail of the queue.
917 * Typically we use the tail for new ioctls and so forth. We use the head
918 * of the queue for things like a QUEUE_FULL message from a device, or a
919 * host that is unable to accept a particular command.
921 void blk_insert_request(struct request_queue
*q
, struct request
*rq
,
922 int at_head
, void *data
)
924 int where
= at_head
? ELEVATOR_INSERT_FRONT
: ELEVATOR_INSERT_BACK
;
928 * tell I/O scheduler that this isn't a regular read/write (ie it
929 * must not attempt merges on this) and that it acts as a soft
932 rq
->cmd_type
= REQ_TYPE_SPECIAL
;
933 rq
->cmd_flags
|= REQ_SOFTBARRIER
;
937 spin_lock_irqsave(q
->queue_lock
, flags
);
940 * If command is tagged, release the tag
942 if (blk_rq_tagged(rq
))
943 blk_queue_end_tag(q
, rq
);
945 drive_stat_acct(rq
, 1);
946 __elv_add_request(q
, rq
, where
, 0);
947 blk_start_queueing(q
);
948 spin_unlock_irqrestore(q
->queue_lock
, flags
);
950 EXPORT_SYMBOL(blk_insert_request
);
953 * add-request adds a request to the linked list.
954 * queue lock is held and interrupts disabled, as we muck with the
955 * request queue list.
957 static inline void add_request(struct request_queue
*q
, struct request
*req
)
959 drive_stat_acct(req
, 1);
962 * elevator indicated where it wants this request to be
963 * inserted at elevator_merge time
965 __elv_add_request(q
, req
, ELEVATOR_INSERT_SORT
, 0);
969 * disk_round_stats() - Round off the performance stats on a struct
972 * The average IO queue length and utilisation statistics are maintained
973 * by observing the current state of the queue length and the amount of
974 * time it has been in this state for.
976 * Normally, that accounting is done on IO completion, but that can result
977 * in more than a second's worth of IO being accounted for within any one
978 * second, leading to >100% utilisation. To deal with that, we call this
979 * function to do a round-off before returning the results when reading
980 * /proc/diskstats. This accounts immediately for all queue usage up to
981 * the current jiffies and restarts the counters again.
983 void disk_round_stats(struct gendisk
*disk
)
985 unsigned long now
= jiffies
;
987 if (now
== disk
->stamp
)
990 if (disk
->in_flight
) {
991 __disk_stat_add(disk
, time_in_queue
,
992 disk
->in_flight
* (now
- disk
->stamp
));
993 __disk_stat_add(disk
, io_ticks
, (now
- disk
->stamp
));
997 EXPORT_SYMBOL_GPL(disk_round_stats
);
999 void part_round_stats(struct hd_struct
*part
)
1001 unsigned long now
= jiffies
;
1003 if (now
== part
->stamp
)
1006 if (part
->in_flight
) {
1007 __part_stat_add(part
, time_in_queue
,
1008 part
->in_flight
* (now
- part
->stamp
));
1009 __part_stat_add(part
, io_ticks
, (now
- part
->stamp
));
1015 * queue lock must be held
1017 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1021 if (unlikely(--req
->ref_count
))
1024 elv_completed_request(q
, req
);
1027 * Request may not have originated from ll_rw_blk. if not,
1028 * it didn't come out of our reserved rq pools
1030 if (req
->cmd_flags
& REQ_ALLOCED
) {
1031 int rw
= rq_data_dir(req
);
1032 int priv
= req
->cmd_flags
& REQ_ELVPRIV
;
1034 BUG_ON(!list_empty(&req
->queuelist
));
1035 BUG_ON(!hlist_unhashed(&req
->hash
));
1037 blk_free_request(q
, req
);
1038 freed_request(q
, rw
, priv
);
1041 EXPORT_SYMBOL_GPL(__blk_put_request
);
1043 void blk_put_request(struct request
*req
)
1045 unsigned long flags
;
1046 struct request_queue
*q
= req
->q
;
1049 * Gee, IDE calls in w/ NULL q. Fix IDE and remove the
1050 * following if (q) test.
1053 spin_lock_irqsave(q
->queue_lock
, flags
);
1054 __blk_put_request(q
, req
);
1055 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1058 EXPORT_SYMBOL(blk_put_request
);
1060 void init_request_from_bio(struct request
*req
, struct bio
*bio
)
1062 req
->cmd_type
= REQ_TYPE_FS
;
1065 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
1067 if (bio_rw_ahead(bio
) || bio_failfast(bio
))
1068 req
->cmd_flags
|= REQ_FAILFAST
;
1071 * REQ_BARRIER implies no merging, but lets make it explicit
1073 if (unlikely(bio_barrier(bio
)))
1074 req
->cmd_flags
|= (REQ_HARDBARRIER
| REQ_NOMERGE
);
1077 req
->cmd_flags
|= REQ_RW_SYNC
;
1078 if (bio_rw_meta(bio
))
1079 req
->cmd_flags
|= REQ_RW_META
;
1082 req
->hard_sector
= req
->sector
= bio
->bi_sector
;
1083 req
->ioprio
= bio_prio(bio
);
1084 req
->start_time
= jiffies
;
1085 blk_rq_bio_prep(req
->q
, req
, bio
);
1088 static int __make_request(struct request_queue
*q
, struct bio
*bio
)
1090 struct request
*req
;
1091 int el_ret
, nr_sectors
, barrier
, err
;
1092 const unsigned short prio
= bio_prio(bio
);
1093 const int sync
= bio_sync(bio
);
1096 nr_sectors
= bio_sectors(bio
);
1099 * low level driver can indicate that it wants pages above a
1100 * certain limit bounced to low memory (ie for highmem, or even
1101 * ISA dma in theory)
1103 blk_queue_bounce(q
, &bio
);
1105 barrier
= bio_barrier(bio
);
1106 if (unlikely(barrier
) && (q
->next_ordered
== QUEUE_ORDERED_NONE
)) {
1111 spin_lock_irq(q
->queue_lock
);
1113 if (unlikely(barrier
) || elv_queue_empty(q
))
1116 el_ret
= elv_merge(q
, &req
, bio
);
1118 case ELEVATOR_BACK_MERGE
:
1119 BUG_ON(!rq_mergeable(req
));
1121 if (!ll_back_merge_fn(q
, req
, bio
))
1124 blk_add_trace_bio(q
, bio
, BLK_TA_BACKMERGE
);
1126 req
->biotail
->bi_next
= bio
;
1128 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
1129 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
1130 drive_stat_acct(req
, 0);
1131 if (!attempt_back_merge(q
, req
))
1132 elv_merged_request(q
, req
, el_ret
);
1135 case ELEVATOR_FRONT_MERGE
:
1136 BUG_ON(!rq_mergeable(req
));
1138 if (!ll_front_merge_fn(q
, req
, bio
))
1141 blk_add_trace_bio(q
, bio
, BLK_TA_FRONTMERGE
);
1143 bio
->bi_next
= req
->bio
;
1147 * may not be valid. if the low level driver said
1148 * it didn't need a bounce buffer then it better
1149 * not touch req->buffer either...
1151 req
->buffer
= bio_data(bio
);
1152 req
->current_nr_sectors
= bio_cur_sectors(bio
);
1153 req
->hard_cur_sectors
= req
->current_nr_sectors
;
1154 req
->sector
= req
->hard_sector
= bio
->bi_sector
;
1155 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
1156 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
1157 drive_stat_acct(req
, 0);
1158 if (!attempt_front_merge(q
, req
))
1159 elv_merged_request(q
, req
, el_ret
);
1162 /* ELV_NO_MERGE: elevator says don't/can't merge. */
1169 * This sync check and mask will be re-done in init_request_from_bio(),
1170 * but we need to set it earlier to expose the sync flag to the
1171 * rq allocator and io schedulers.
1173 rw_flags
= bio_data_dir(bio
);
1175 rw_flags
|= REQ_RW_SYNC
;
1178 * Grab a free request. This is might sleep but can not fail.
1179 * Returns with the queue unlocked.
1181 req
= get_request_wait(q
, rw_flags
, bio
);
1184 * After dropping the lock and possibly sleeping here, our request
1185 * may now be mergeable after it had proven unmergeable (above).
1186 * We don't worry about that case for efficiency. It won't happen
1187 * often, and the elevators are able to handle it.
1189 init_request_from_bio(req
, bio
);
1191 spin_lock_irq(q
->queue_lock
);
1192 if (elv_queue_empty(q
))
1194 add_request(q
, req
);
1197 __generic_unplug_device(q
);
1199 spin_unlock_irq(q
->queue_lock
);
1203 bio_endio(bio
, err
);
1208 * If bio->bi_dev is a partition, remap the location
1210 static inline void blk_partition_remap(struct bio
*bio
)
1212 struct block_device
*bdev
= bio
->bi_bdev
;
1214 if (bio_sectors(bio
) && bdev
!= bdev
->bd_contains
) {
1215 struct hd_struct
*p
= bdev
->bd_part
;
1217 bio
->bi_sector
+= p
->start_sect
;
1218 bio
->bi_bdev
= bdev
->bd_contains
;
1220 blk_add_trace_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
1221 bdev
->bd_dev
, bio
->bi_sector
,
1222 bio
->bi_sector
- p
->start_sect
);
1226 static void handle_bad_sector(struct bio
*bio
)
1228 char b
[BDEVNAME_SIZE
];
1230 printk(KERN_INFO
"attempt to access beyond end of device\n");
1231 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
1232 bdevname(bio
->bi_bdev
, b
),
1234 (unsigned long long)bio
->bi_sector
+ bio_sectors(bio
),
1235 (long long)(bio
->bi_bdev
->bd_inode
->i_size
>> 9));
1237 set_bit(BIO_EOF
, &bio
->bi_flags
);
1240 #ifdef CONFIG_FAIL_MAKE_REQUEST
1242 static DECLARE_FAULT_ATTR(fail_make_request
);
1244 static int __init
setup_fail_make_request(char *str
)
1246 return setup_fault_attr(&fail_make_request
, str
);
1248 __setup("fail_make_request=", setup_fail_make_request
);
1250 static int should_fail_request(struct bio
*bio
)
1252 if ((bio
->bi_bdev
->bd_disk
->flags
& GENHD_FL_FAIL
) ||
1253 (bio
->bi_bdev
->bd_part
&& bio
->bi_bdev
->bd_part
->make_it_fail
))
1254 return should_fail(&fail_make_request
, bio
->bi_size
);
1259 static int __init
fail_make_request_debugfs(void)
1261 return init_fault_attr_dentries(&fail_make_request
,
1262 "fail_make_request");
1265 late_initcall(fail_make_request_debugfs
);
1267 #else /* CONFIG_FAIL_MAKE_REQUEST */
1269 static inline int should_fail_request(struct bio
*bio
)
1274 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1277 * Check whether this bio extends beyond the end of the device.
1279 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
1286 /* Test device or partition size, when known. */
1287 maxsector
= bio
->bi_bdev
->bd_inode
->i_size
>> 9;
1289 sector_t sector
= bio
->bi_sector
;
1291 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
1293 * This may well happen - the kernel calls bread()
1294 * without checking the size of the device, e.g., when
1295 * mounting a device.
1297 handle_bad_sector(bio
);
1306 * generic_make_request: hand a buffer to its device driver for I/O
1307 * @bio: The bio describing the location in memory and on the device.
1309 * generic_make_request() is used to make I/O requests of block
1310 * devices. It is passed a &struct bio, which describes the I/O that needs
1313 * generic_make_request() does not return any status. The
1314 * success/failure status of the request, along with notification of
1315 * completion, is delivered asynchronously through the bio->bi_end_io
1316 * function described (one day) else where.
1318 * The caller of generic_make_request must make sure that bi_io_vec
1319 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1320 * set to describe the device address, and the
1321 * bi_end_io and optionally bi_private are set to describe how
1322 * completion notification should be signaled.
1324 * generic_make_request and the drivers it calls may use bi_next if this
1325 * bio happens to be merged with someone else, and may change bi_dev and
1326 * bi_sector for remaps as it sees fit. So the values of these fields
1327 * should NOT be depended on after the call to generic_make_request.
1329 static inline void __generic_make_request(struct bio
*bio
)
1331 struct request_queue
*q
;
1332 sector_t old_sector
;
1333 int ret
, nr_sectors
= bio_sectors(bio
);
1339 if (bio_check_eod(bio
, nr_sectors
))
1343 * Resolve the mapping until finished. (drivers are
1344 * still free to implement/resolve their own stacking
1345 * by explicitly returning 0)
1347 * NOTE: we don't repeat the blk_size check for each new device.
1348 * Stacking drivers are expected to know what they are doing.
1353 char b
[BDEVNAME_SIZE
];
1355 q
= bdev_get_queue(bio
->bi_bdev
);
1358 "generic_make_request: Trying to access "
1359 "nonexistent block-device %s (%Lu)\n",
1360 bdevname(bio
->bi_bdev
, b
),
1361 (long long) bio
->bi_sector
);
1363 bio_endio(bio
, err
);
1367 if (unlikely(nr_sectors
> q
->max_hw_sectors
)) {
1368 printk(KERN_ERR
"bio too big device %s (%u > %u)\n",
1369 bdevname(bio
->bi_bdev
, b
),
1375 if (unlikely(test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)))
1378 if (should_fail_request(bio
))
1382 * If this device has partitions, remap block n
1383 * of partition p to block n+start(p) of the disk.
1385 blk_partition_remap(bio
);
1387 if (old_sector
!= -1)
1388 blk_add_trace_remap(q
, bio
, old_dev
, bio
->bi_sector
,
1391 blk_add_trace_bio(q
, bio
, BLK_TA_QUEUE
);
1393 old_sector
= bio
->bi_sector
;
1394 old_dev
= bio
->bi_bdev
->bd_dev
;
1396 if (bio_check_eod(bio
, nr_sectors
))
1398 if (bio_empty_barrier(bio
) && !q
->prepare_flush_fn
) {
1403 ret
= q
->make_request_fn(q
, bio
);
1408 * We only want one ->make_request_fn to be active at a time,
1409 * else stack usage with stacked devices could be a problem.
1410 * So use current->bio_{list,tail} to keep a list of requests
1411 * submited by a make_request_fn function.
1412 * current->bio_tail is also used as a flag to say if
1413 * generic_make_request is currently active in this task or not.
1414 * If it is NULL, then no make_request is active. If it is non-NULL,
1415 * then a make_request is active, and new requests should be added
1418 void generic_make_request(struct bio
*bio
)
1420 if (current
->bio_tail
) {
1421 /* make_request is active */
1422 *(current
->bio_tail
) = bio
;
1423 bio
->bi_next
= NULL
;
1424 current
->bio_tail
= &bio
->bi_next
;
1427 /* following loop may be a bit non-obvious, and so deserves some
1429 * Before entering the loop, bio->bi_next is NULL (as all callers
1430 * ensure that) so we have a list with a single bio.
1431 * We pretend that we have just taken it off a longer list, so
1432 * we assign bio_list to the next (which is NULL) and bio_tail
1433 * to &bio_list, thus initialising the bio_list of new bios to be
1434 * added. __generic_make_request may indeed add some more bios
1435 * through a recursive call to generic_make_request. If it
1436 * did, we find a non-NULL value in bio_list and re-enter the loop
1437 * from the top. In this case we really did just take the bio
1438 * of the top of the list (no pretending) and so fixup bio_list and
1439 * bio_tail or bi_next, and call into __generic_make_request again.
1441 * The loop was structured like this to make only one call to
1442 * __generic_make_request (which is important as it is large and
1443 * inlined) and to keep the structure simple.
1445 BUG_ON(bio
->bi_next
);
1447 current
->bio_list
= bio
->bi_next
;
1448 if (bio
->bi_next
== NULL
)
1449 current
->bio_tail
= ¤t
->bio_list
;
1451 bio
->bi_next
= NULL
;
1452 __generic_make_request(bio
);
1453 bio
= current
->bio_list
;
1455 current
->bio_tail
= NULL
; /* deactivate */
1457 EXPORT_SYMBOL(generic_make_request
);
1460 * submit_bio: submit a bio to the block device layer for I/O
1461 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1462 * @bio: The &struct bio which describes the I/O
1464 * submit_bio() is very similar in purpose to generic_make_request(), and
1465 * uses that function to do most of the work. Both are fairly rough
1466 * interfaces, @bio must be presetup and ready for I/O.
1469 void submit_bio(int rw
, struct bio
*bio
)
1471 int count
= bio_sectors(bio
);
1476 * If it's a regular read/write or a barrier with data attached,
1477 * go through the normal accounting stuff before submission.
1479 if (!bio_empty_barrier(bio
)) {
1481 BIO_BUG_ON(!bio
->bi_size
);
1482 BIO_BUG_ON(!bio
->bi_io_vec
);
1485 count_vm_events(PGPGOUT
, count
);
1487 task_io_account_read(bio
->bi_size
);
1488 count_vm_events(PGPGIN
, count
);
1491 if (unlikely(block_dump
)) {
1492 char b
[BDEVNAME_SIZE
];
1493 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s\n",
1494 current
->comm
, task_pid_nr(current
),
1495 (rw
& WRITE
) ? "WRITE" : "READ",
1496 (unsigned long long)bio
->bi_sector
,
1497 bdevname(bio
->bi_bdev
, b
));
1501 generic_make_request(bio
);
1503 EXPORT_SYMBOL(submit_bio
);
1506 * __end_that_request_first - end I/O on a request
1507 * @req: the request being processed
1508 * @error: 0 for success, < 0 for error
1509 * @nr_bytes: number of bytes to complete
1512 * Ends I/O on a number of bytes attached to @req, and sets it up
1513 * for the next range of segments (if any) in the cluster.
1516 * 0 - we are done with this request, call end_that_request_last()
1517 * 1 - still buffers pending for this request
1519 static int __end_that_request_first(struct request
*req
, int error
,
1522 int total_bytes
, bio_nbytes
, next_idx
= 0;
1525 blk_add_trace_rq(req
->q
, req
, BLK_TA_COMPLETE
);
1528 * for a REQ_BLOCK_PC request, we want to carry any eventual
1529 * sense key with us all the way through
1531 if (!blk_pc_request(req
))
1534 if (error
&& (blk_fs_request(req
) && !(req
->cmd_flags
& REQ_QUIET
))) {
1535 printk(KERN_ERR
"end_request: I/O error, dev %s, sector %llu\n",
1536 req
->rq_disk
? req
->rq_disk
->disk_name
: "?",
1537 (unsigned long long)req
->sector
);
1540 if (blk_fs_request(req
) && req
->rq_disk
) {
1541 const int rw
= rq_data_dir(req
);
1543 all_stat_add(req
->rq_disk
, sectors
[rw
],
1544 nr_bytes
>> 9, req
->sector
);
1547 total_bytes
= bio_nbytes
= 0;
1548 while ((bio
= req
->bio
) != NULL
) {
1552 * For an empty barrier request, the low level driver must
1553 * store a potential error location in ->sector. We pass
1554 * that back up in ->bi_sector.
1556 if (blk_empty_barrier(req
))
1557 bio
->bi_sector
= req
->sector
;
1559 if (nr_bytes
>= bio
->bi_size
) {
1560 req
->bio
= bio
->bi_next
;
1561 nbytes
= bio
->bi_size
;
1562 req_bio_endio(req
, bio
, nbytes
, error
);
1566 int idx
= bio
->bi_idx
+ next_idx
;
1568 if (unlikely(bio
->bi_idx
>= bio
->bi_vcnt
)) {
1569 blk_dump_rq_flags(req
, "__end_that");
1570 printk(KERN_ERR
"%s: bio idx %d >= vcnt %d\n",
1571 __func__
, bio
->bi_idx
, bio
->bi_vcnt
);
1575 nbytes
= bio_iovec_idx(bio
, idx
)->bv_len
;
1576 BIO_BUG_ON(nbytes
> bio
->bi_size
);
1579 * not a complete bvec done
1581 if (unlikely(nbytes
> nr_bytes
)) {
1582 bio_nbytes
+= nr_bytes
;
1583 total_bytes
+= nr_bytes
;
1588 * advance to the next vector
1591 bio_nbytes
+= nbytes
;
1594 total_bytes
+= nbytes
;
1600 * end more in this run, or just return 'not-done'
1602 if (unlikely(nr_bytes
<= 0))
1614 * if the request wasn't completed, update state
1617 req_bio_endio(req
, bio
, bio_nbytes
, error
);
1618 bio
->bi_idx
+= next_idx
;
1619 bio_iovec(bio
)->bv_offset
+= nr_bytes
;
1620 bio_iovec(bio
)->bv_len
-= nr_bytes
;
1623 blk_recalc_rq_sectors(req
, total_bytes
>> 9);
1624 blk_recalc_rq_segments(req
);
1629 * splice the completion data to a local structure and hand off to
1630 * process_completion_queue() to complete the requests
1632 static void blk_done_softirq(struct softirq_action
*h
)
1634 struct list_head
*cpu_list
, local_list
;
1636 local_irq_disable();
1637 cpu_list
= &__get_cpu_var(blk_cpu_done
);
1638 list_replace_init(cpu_list
, &local_list
);
1641 while (!list_empty(&local_list
)) {
1644 rq
= list_entry(local_list
.next
, struct request
, donelist
);
1645 list_del_init(&rq
->donelist
);
1646 rq
->q
->softirq_done_fn(rq
);
1650 static int __cpuinit
blk_cpu_notify(struct notifier_block
*self
,
1651 unsigned long action
, void *hcpu
)
1654 * If a CPU goes away, splice its entries to the current CPU
1655 * and trigger a run of the softirq
1657 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
1658 int cpu
= (unsigned long) hcpu
;
1660 local_irq_disable();
1661 list_splice_init(&per_cpu(blk_cpu_done
, cpu
),
1662 &__get_cpu_var(blk_cpu_done
));
1663 raise_softirq_irqoff(BLOCK_SOFTIRQ
);
1671 static struct notifier_block blk_cpu_notifier __cpuinitdata
= {
1672 .notifier_call
= blk_cpu_notify
,
1676 * blk_complete_request - end I/O on a request
1677 * @req: the request being processed
1680 * Ends all I/O on a request. It does not handle partial completions,
1681 * unless the driver actually implements this in its completion callback
1682 * through requeueing. The actual completion happens out-of-order,
1683 * through a softirq handler. The user must have registered a completion
1684 * callback through blk_queue_softirq_done().
1687 void blk_complete_request(struct request
*req
)
1689 struct list_head
*cpu_list
;
1690 unsigned long flags
;
1692 BUG_ON(!req
->q
->softirq_done_fn
);
1694 local_irq_save(flags
);
1696 cpu_list
= &__get_cpu_var(blk_cpu_done
);
1697 list_add_tail(&req
->donelist
, cpu_list
);
1698 raise_softirq_irqoff(BLOCK_SOFTIRQ
);
1700 local_irq_restore(flags
);
1702 EXPORT_SYMBOL(blk_complete_request
);
1705 * queue lock must be held
1707 static void end_that_request_last(struct request
*req
, int error
)
1709 struct gendisk
*disk
= req
->rq_disk
;
1711 if (blk_rq_tagged(req
))
1712 blk_queue_end_tag(req
->q
, req
);
1714 if (blk_queued_rq(req
))
1715 blkdev_dequeue_request(req
);
1717 if (unlikely(laptop_mode
) && blk_fs_request(req
))
1718 laptop_io_completion();
1721 * Account IO completion. bar_rq isn't accounted as a normal
1722 * IO on queueing nor completion. Accounting the containing
1723 * request is enough.
1725 if (disk
&& blk_fs_request(req
) && req
!= &req
->q
->bar_rq
) {
1726 unsigned long duration
= jiffies
- req
->start_time
;
1727 const int rw
= rq_data_dir(req
);
1728 struct hd_struct
*part
= get_part(disk
, req
->sector
);
1730 __all_stat_inc(disk
, ios
[rw
], req
->sector
);
1731 __all_stat_add(disk
, ticks
[rw
], duration
, req
->sector
);
1732 disk_round_stats(disk
);
1735 part_round_stats(part
);
1741 req
->end_io(req
, error
);
1743 if (blk_bidi_rq(req
))
1744 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
1746 __blk_put_request(req
->q
, req
);
1750 static inline void __end_request(struct request
*rq
, int uptodate
,
1751 unsigned int nr_bytes
)
1756 error
= uptodate
? uptodate
: -EIO
;
1758 __blk_end_request(rq
, error
, nr_bytes
);
1762 * blk_rq_bytes - Returns bytes left to complete in the entire request
1763 * @rq: the request being processed
1765 unsigned int blk_rq_bytes(struct request
*rq
)
1767 if (blk_fs_request(rq
))
1768 return rq
->hard_nr_sectors
<< 9;
1770 return rq
->data_len
;
1772 EXPORT_SYMBOL_GPL(blk_rq_bytes
);
1775 * blk_rq_cur_bytes - Returns bytes left to complete in the current segment
1776 * @rq: the request being processed
1778 unsigned int blk_rq_cur_bytes(struct request
*rq
)
1780 if (blk_fs_request(rq
))
1781 return rq
->current_nr_sectors
<< 9;
1784 return rq
->bio
->bi_size
;
1786 return rq
->data_len
;
1788 EXPORT_SYMBOL_GPL(blk_rq_cur_bytes
);
1791 * end_queued_request - end all I/O on a queued request
1792 * @rq: the request being processed
1793 * @uptodate: error value or 0/1 uptodate flag
1796 * Ends all I/O on a request, and removes it from the block layer queues.
1797 * Not suitable for normal IO completion, unless the driver still has
1798 * the request attached to the block layer.
1801 void end_queued_request(struct request
*rq
, int uptodate
)
1803 __end_request(rq
, uptodate
, blk_rq_bytes(rq
));
1805 EXPORT_SYMBOL(end_queued_request
);
1808 * end_dequeued_request - end all I/O on a dequeued request
1809 * @rq: the request being processed
1810 * @uptodate: error value or 0/1 uptodate flag
1813 * Ends all I/O on a request. The request must already have been
1814 * dequeued using blkdev_dequeue_request(), as is normally the case
1818 void end_dequeued_request(struct request
*rq
, int uptodate
)
1820 __end_request(rq
, uptodate
, blk_rq_bytes(rq
));
1822 EXPORT_SYMBOL(end_dequeued_request
);
1826 * end_request - end I/O on the current segment of the request
1827 * @req: the request being processed
1828 * @uptodate: error value or 0/1 uptodate flag
1831 * Ends I/O on the current segment of a request. If that is the only
1832 * remaining segment, the request is also completed and freed.
1834 * This is a remnant of how older block drivers handled IO completions.
1835 * Modern drivers typically end IO on the full request in one go, unless
1836 * they have a residual value to account for. For that case this function
1837 * isn't really useful, unless the residual just happens to be the
1838 * full current segment. In other words, don't use this function in new
1839 * code. Either use end_request_completely(), or the
1840 * end_that_request_chunk() (along with end_that_request_last()) for
1841 * partial completions.
1844 void end_request(struct request
*req
, int uptodate
)
1846 __end_request(req
, uptodate
, req
->hard_cur_sectors
<< 9);
1848 EXPORT_SYMBOL(end_request
);
1851 * blk_end_io - Generic end_io function to complete a request.
1852 * @rq: the request being processed
1853 * @error: 0 for success, < 0 for error
1854 * @nr_bytes: number of bytes to complete @rq
1855 * @bidi_bytes: number of bytes to complete @rq->next_rq
1856 * @drv_callback: function called between completion of bios in the request
1857 * and completion of the request.
1858 * If the callback returns non 0, this helper returns without
1859 * completion of the request.
1862 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
1863 * If @rq has leftover, sets it up for the next range of segments.
1866 * 0 - we are done with this request
1867 * 1 - this request is not freed yet, it still has pending buffers.
1869 static int blk_end_io(struct request
*rq
, int error
, unsigned int nr_bytes
,
1870 unsigned int bidi_bytes
,
1871 int (drv_callback
)(struct request
*))
1873 struct request_queue
*q
= rq
->q
;
1874 unsigned long flags
= 0UL;
1876 if (blk_fs_request(rq
) || blk_pc_request(rq
)) {
1877 if (__end_that_request_first(rq
, error
, nr_bytes
))
1880 /* Bidi request must be completed as a whole */
1881 if (blk_bidi_rq(rq
) &&
1882 __end_that_request_first(rq
->next_rq
, error
, bidi_bytes
))
1886 /* Special feature for tricky drivers */
1887 if (drv_callback
&& drv_callback(rq
))
1890 add_disk_randomness(rq
->rq_disk
);
1892 spin_lock_irqsave(q
->queue_lock
, flags
);
1893 end_that_request_last(rq
, error
);
1894 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1900 * blk_end_request - Helper function for drivers to complete the request.
1901 * @rq: the request being processed
1902 * @error: 0 for success, < 0 for error
1903 * @nr_bytes: number of bytes to complete
1906 * Ends I/O on a number of bytes attached to @rq.
1907 * If @rq has leftover, sets it up for the next range of segments.
1910 * 0 - we are done with this request
1911 * 1 - still buffers pending for this request
1913 int blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
1915 return blk_end_io(rq
, error
, nr_bytes
, 0, NULL
);
1917 EXPORT_SYMBOL_GPL(blk_end_request
);
1920 * __blk_end_request - Helper function for drivers to complete the request.
1921 * @rq: the request being processed
1922 * @error: 0 for success, < 0 for error
1923 * @nr_bytes: number of bytes to complete
1926 * Must be called with queue lock held unlike blk_end_request().
1929 * 0 - we are done with this request
1930 * 1 - still buffers pending for this request
1932 int __blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
1934 if (blk_fs_request(rq
) || blk_pc_request(rq
)) {
1935 if (__end_that_request_first(rq
, error
, nr_bytes
))
1939 add_disk_randomness(rq
->rq_disk
);
1941 end_that_request_last(rq
, error
);
1945 EXPORT_SYMBOL_GPL(__blk_end_request
);
1948 * blk_end_bidi_request - Helper function for drivers to complete bidi request.
1949 * @rq: the bidi request being processed
1950 * @error: 0 for success, < 0 for error
1951 * @nr_bytes: number of bytes to complete @rq
1952 * @bidi_bytes: number of bytes to complete @rq->next_rq
1955 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
1958 * 0 - we are done with this request
1959 * 1 - still buffers pending for this request
1961 int blk_end_bidi_request(struct request
*rq
, int error
, unsigned int nr_bytes
,
1962 unsigned int bidi_bytes
)
1964 return blk_end_io(rq
, error
, nr_bytes
, bidi_bytes
, NULL
);
1966 EXPORT_SYMBOL_GPL(blk_end_bidi_request
);
1969 * blk_end_request_callback - Special helper function for tricky drivers
1970 * @rq: the request being processed
1971 * @error: 0 for success, < 0 for error
1972 * @nr_bytes: number of bytes to complete
1973 * @drv_callback: function called between completion of bios in the request
1974 * and completion of the request.
1975 * If the callback returns non 0, this helper returns without
1976 * completion of the request.
1979 * Ends I/O on a number of bytes attached to @rq.
1980 * If @rq has leftover, sets it up for the next range of segments.
1982 * This special helper function is used only for existing tricky drivers.
1983 * (e.g. cdrom_newpc_intr() of ide-cd)
1984 * This interface will be removed when such drivers are rewritten.
1985 * Don't use this interface in other places anymore.
1988 * 0 - we are done with this request
1989 * 1 - this request is not freed yet.
1990 * this request still has pending buffers or
1991 * the driver doesn't want to finish this request yet.
1993 int blk_end_request_callback(struct request
*rq
, int error
,
1994 unsigned int nr_bytes
,
1995 int (drv_callback
)(struct request
*))
1997 return blk_end_io(rq
, error
, nr_bytes
, 0, drv_callback
);
1999 EXPORT_SYMBOL_GPL(blk_end_request_callback
);
2001 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
2004 /* first two bits are identical in rq->cmd_flags and bio->bi_rw */
2005 rq
->cmd_flags
|= (bio
->bi_rw
& 3);
2007 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2008 rq
->nr_hw_segments
= bio_hw_segments(q
, bio
);
2009 rq
->current_nr_sectors
= bio_cur_sectors(bio
);
2010 rq
->hard_cur_sectors
= rq
->current_nr_sectors
;
2011 rq
->hard_nr_sectors
= rq
->nr_sectors
= bio_sectors(bio
);
2012 rq
->buffer
= bio_data(bio
);
2013 rq
->data_len
= bio
->bi_size
;
2015 rq
->bio
= rq
->biotail
= bio
;
2018 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2021 int kblockd_schedule_work(struct work_struct
*work
)
2023 return queue_work(kblockd_workqueue
, work
);
2025 EXPORT_SYMBOL(kblockd_schedule_work
);
2027 void kblockd_flush_work(struct work_struct
*work
)
2029 cancel_work_sync(work
);
2031 EXPORT_SYMBOL(kblockd_flush_work
);
2033 int __init
blk_dev_init(void)
2037 kblockd_workqueue
= create_workqueue("kblockd");
2038 if (!kblockd_workqueue
)
2039 panic("Failed to create kblockd\n");
2041 request_cachep
= kmem_cache_create("blkdev_requests",
2042 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
2044 blk_requestq_cachep
= kmem_cache_create("blkdev_queue",
2045 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);
2047 for_each_possible_cpu(i
)
2048 INIT_LIST_HEAD(&per_cpu(blk_cpu_done
, i
));
2050 open_softirq(BLOCK_SOFTIRQ
, blk_done_softirq
, NULL
);
2051 register_hotcpu_notifier(&blk_cpu_notifier
);