2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
6 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
8 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
12 * This handles all read/write requests to block devices
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/backing-dev.h>
17 #include <linux/bio.h>
18 #include <linux/blkdev.h>
19 #include <linux/highmem.h>
21 #include <linux/kernel_stat.h>
22 #include <linux/string.h>
23 #include <linux/init.h>
24 #include <linux/completion.h>
25 #include <linux/slab.h>
26 #include <linux/swap.h>
27 #include <linux/writeback.h>
28 #include <linux/task_io_accounting_ops.h>
29 #include <linux/blktrace_api.h>
30 #include <linux/fault-inject.h>
34 static int __make_request(struct request_queue
*q
, struct bio
*bio
);
37 * For the allocated request tables
39 static struct kmem_cache
*request_cachep
;
42 * For queue allocation
44 struct kmem_cache
*blk_requestq_cachep
;
47 * Controlling structure to kblockd
49 static struct workqueue_struct
*kblockd_workqueue
;
51 static void drive_stat_acct(struct request
*rq
, int new_io
)
53 struct hd_struct
*part
;
54 int rw
= rq_data_dir(rq
);
57 if (!blk_fs_request(rq
) || !rq
->rq_disk
)
60 cpu
= part_stat_lock();
61 part
= disk_map_sector_rcu(rq
->rq_disk
, rq
->sector
);
64 part_stat_inc(cpu
, part
, merges
[rw
]);
66 part_round_stats(cpu
, part
);
67 part_inc_in_flight(part
);
73 void blk_queue_congestion_threshold(struct request_queue
*q
)
77 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) + 1;
78 if (nr
> q
->nr_requests
)
80 q
->nr_congestion_on
= nr
;
82 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) - (q
->nr_requests
/ 16) - 1;
85 q
->nr_congestion_off
= nr
;
89 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
92 * Locates the passed device's request queue and returns the address of its
95 * Will return NULL if the request queue cannot be located.
97 struct backing_dev_info
*blk_get_backing_dev_info(struct block_device
*bdev
)
99 struct backing_dev_info
*ret
= NULL
;
100 struct request_queue
*q
= bdev_get_queue(bdev
);
103 ret
= &q
->backing_dev_info
;
106 EXPORT_SYMBOL(blk_get_backing_dev_info
);
108 void blk_rq_init(struct request_queue
*q
, struct request
*rq
)
110 memset(rq
, 0, sizeof(*rq
));
112 INIT_LIST_HEAD(&rq
->queuelist
);
113 INIT_LIST_HEAD(&rq
->timeout_list
);
116 rq
->sector
= rq
->hard_sector
= (sector_t
) -1;
117 INIT_HLIST_NODE(&rq
->hash
);
118 RB_CLEAR_NODE(&rq
->rb_node
);
123 EXPORT_SYMBOL(blk_rq_init
);
125 static void req_bio_endio(struct request
*rq
, struct bio
*bio
,
126 unsigned int nbytes
, int error
)
128 struct request_queue
*q
= rq
->q
;
130 if (&q
->bar_rq
!= rq
) {
132 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
133 else if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
136 if (unlikely(nbytes
> bio
->bi_size
)) {
137 printk(KERN_ERR
"%s: want %u bytes done, %u left\n",
138 __func__
, nbytes
, bio
->bi_size
);
139 nbytes
= bio
->bi_size
;
142 bio
->bi_size
-= nbytes
;
143 bio
->bi_sector
+= (nbytes
>> 9);
145 if (bio_integrity(bio
))
146 bio_integrity_advance(bio
, nbytes
);
148 if (bio
->bi_size
== 0)
149 bio_endio(bio
, error
);
153 * Okay, this is the barrier request in progress, just
156 if (error
&& !q
->orderr
)
161 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
165 printk(KERN_INFO
"%s: dev %s: type=%x, flags=%x\n", msg
,
166 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->cmd_type
,
169 printk(KERN_INFO
" sector %llu, nr/cnr %lu/%u\n",
170 (unsigned long long)rq
->sector
,
172 rq
->current_nr_sectors
);
173 printk(KERN_INFO
" bio %p, biotail %p, buffer %p, data %p, len %u\n",
174 rq
->bio
, rq
->biotail
,
175 rq
->buffer
, rq
->data
,
178 if (blk_pc_request(rq
)) {
179 printk(KERN_INFO
" cdb: ");
180 for (bit
= 0; bit
< BLK_MAX_CDB
; bit
++)
181 printk("%02x ", rq
->cmd
[bit
]);
185 EXPORT_SYMBOL(blk_dump_rq_flags
);
188 * "plug" the device if there are no outstanding requests: this will
189 * force the transfer to start only after we have put all the requests
192 * This is called with interrupts off and no requests on the queue and
193 * with the queue lock held.
195 void blk_plug_device(struct request_queue
*q
)
197 WARN_ON(!irqs_disabled());
200 * don't plug a stopped queue, it must be paired with blk_start_queue()
201 * which will restart the queueing
203 if (blk_queue_stopped(q
))
206 if (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED
, q
)) {
207 mod_timer(&q
->unplug_timer
, jiffies
+ q
->unplug_delay
);
208 blk_add_trace_generic(q
, NULL
, 0, BLK_TA_PLUG
);
211 EXPORT_SYMBOL(blk_plug_device
);
214 * blk_plug_device_unlocked - plug a device without queue lock held
215 * @q: The &struct request_queue to plug
218 * Like @blk_plug_device(), but grabs the queue lock and disables
221 void blk_plug_device_unlocked(struct request_queue
*q
)
225 spin_lock_irqsave(q
->queue_lock
, flags
);
227 spin_unlock_irqrestore(q
->queue_lock
, flags
);
229 EXPORT_SYMBOL(blk_plug_device_unlocked
);
232 * remove the queue from the plugged list, if present. called with
233 * queue lock held and interrupts disabled.
235 int blk_remove_plug(struct request_queue
*q
)
237 WARN_ON(!irqs_disabled());
239 if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED
, q
))
242 del_timer(&q
->unplug_timer
);
245 EXPORT_SYMBOL(blk_remove_plug
);
248 * remove the plug and let it rip..
250 void __generic_unplug_device(struct request_queue
*q
)
252 if (unlikely(blk_queue_stopped(q
)))
255 if (!blk_remove_plug(q
))
260 EXPORT_SYMBOL(__generic_unplug_device
);
263 * generic_unplug_device - fire a request queue
264 * @q: The &struct request_queue in question
267 * Linux uses plugging to build bigger requests queues before letting
268 * the device have at them. If a queue is plugged, the I/O scheduler
269 * is still adding and merging requests on the queue. Once the queue
270 * gets unplugged, the request_fn defined for the queue is invoked and
273 void generic_unplug_device(struct request_queue
*q
)
275 if (blk_queue_plugged(q
)) {
276 spin_lock_irq(q
->queue_lock
);
277 __generic_unplug_device(q
);
278 spin_unlock_irq(q
->queue_lock
);
281 EXPORT_SYMBOL(generic_unplug_device
);
283 static void blk_backing_dev_unplug(struct backing_dev_info
*bdi
,
286 struct request_queue
*q
= bdi
->unplug_io_data
;
291 void blk_unplug_work(struct work_struct
*work
)
293 struct request_queue
*q
=
294 container_of(work
, struct request_queue
, unplug_work
);
296 blk_add_trace_pdu_int(q
, BLK_TA_UNPLUG_IO
, NULL
,
297 q
->rq
.count
[READ
] + q
->rq
.count
[WRITE
]);
302 void blk_unplug_timeout(unsigned long data
)
304 struct request_queue
*q
= (struct request_queue
*)data
;
306 blk_add_trace_pdu_int(q
, BLK_TA_UNPLUG_TIMER
, NULL
,
307 q
->rq
.count
[READ
] + q
->rq
.count
[WRITE
]);
309 kblockd_schedule_work(q
, &q
->unplug_work
);
312 void blk_unplug(struct request_queue
*q
)
315 * devices don't necessarily have an ->unplug_fn defined
318 blk_add_trace_pdu_int(q
, BLK_TA_UNPLUG_IO
, NULL
,
319 q
->rq
.count
[READ
] + q
->rq
.count
[WRITE
]);
324 EXPORT_SYMBOL(blk_unplug
);
326 static void blk_invoke_request_fn(struct request_queue
*q
)
329 * one level of recursion is ok and is much faster than kicking
330 * the unplug handling
332 if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER
, q
)) {
334 queue_flag_clear(QUEUE_FLAG_REENTER
, q
);
336 queue_flag_set(QUEUE_FLAG_PLUGGED
, q
);
337 kblockd_schedule_work(q
, &q
->unplug_work
);
342 * blk_start_queue - restart a previously stopped queue
343 * @q: The &struct request_queue in question
346 * blk_start_queue() will clear the stop flag on the queue, and call
347 * the request_fn for the queue if it was in a stopped state when
348 * entered. Also see blk_stop_queue(). Queue lock must be held.
350 void blk_start_queue(struct request_queue
*q
)
352 WARN_ON(!irqs_disabled());
354 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
355 blk_invoke_request_fn(q
);
357 EXPORT_SYMBOL(blk_start_queue
);
360 * blk_stop_queue - stop a queue
361 * @q: The &struct request_queue in question
364 * The Linux block layer assumes that a block driver will consume all
365 * entries on the request queue when the request_fn strategy is called.
366 * Often this will not happen, because of hardware limitations (queue
367 * depth settings). If a device driver gets a 'queue full' response,
368 * or if it simply chooses not to queue more I/O at one point, it can
369 * call this function to prevent the request_fn from being called until
370 * the driver has signalled it's ready to go again. This happens by calling
371 * blk_start_queue() to restart queue operations. Queue lock must be held.
373 void blk_stop_queue(struct request_queue
*q
)
376 queue_flag_set(QUEUE_FLAG_STOPPED
, q
);
378 EXPORT_SYMBOL(blk_stop_queue
);
381 * blk_sync_queue - cancel any pending callbacks on a queue
385 * The block layer may perform asynchronous callback activity
386 * on a queue, such as calling the unplug function after a timeout.
387 * A block device may call blk_sync_queue to ensure that any
388 * such activity is cancelled, thus allowing it to release resources
389 * that the callbacks might use. The caller must already have made sure
390 * that its ->make_request_fn will not re-add plugging prior to calling
394 void blk_sync_queue(struct request_queue
*q
)
396 del_timer_sync(&q
->unplug_timer
);
397 kblockd_flush_work(&q
->unplug_work
);
399 EXPORT_SYMBOL(blk_sync_queue
);
402 * blk_run_queue - run a single device queue
403 * @q: The queue to run
405 void __blk_run_queue(struct request_queue
*q
)
410 * Only recurse once to avoid overrunning the stack, let the unplug
411 * handling reinvoke the handler shortly if we already got there.
413 if (!elv_queue_empty(q
))
414 blk_invoke_request_fn(q
);
416 EXPORT_SYMBOL(__blk_run_queue
);
419 * blk_run_queue - run a single device queue
420 * @q: The queue to run
422 void blk_run_queue(struct request_queue
*q
)
426 spin_lock_irqsave(q
->queue_lock
, flags
);
428 spin_unlock_irqrestore(q
->queue_lock
, flags
);
430 EXPORT_SYMBOL(blk_run_queue
);
432 void blk_put_queue(struct request_queue
*q
)
434 kobject_put(&q
->kobj
);
437 void blk_cleanup_queue(struct request_queue
*q
)
439 mutex_lock(&q
->sysfs_lock
);
440 queue_flag_set_unlocked(QUEUE_FLAG_DEAD
, q
);
441 mutex_unlock(&q
->sysfs_lock
);
444 elevator_exit(q
->elevator
);
448 EXPORT_SYMBOL(blk_cleanup_queue
);
450 static int blk_init_free_list(struct request_queue
*q
)
452 struct request_list
*rl
= &q
->rq
;
454 rl
->count
[READ
] = rl
->count
[WRITE
] = 0;
455 rl
->starved
[READ
] = rl
->starved
[WRITE
] = 0;
457 init_waitqueue_head(&rl
->wait
[READ
]);
458 init_waitqueue_head(&rl
->wait
[WRITE
]);
460 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, mempool_alloc_slab
,
461 mempool_free_slab
, request_cachep
, q
->node
);
469 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
471 return blk_alloc_queue_node(gfp_mask
, -1);
473 EXPORT_SYMBOL(blk_alloc_queue
);
475 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
477 struct request_queue
*q
;
480 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
481 gfp_mask
| __GFP_ZERO
, node_id
);
485 q
->backing_dev_info
.unplug_io_fn
= blk_backing_dev_unplug
;
486 q
->backing_dev_info
.unplug_io_data
= q
;
487 err
= bdi_init(&q
->backing_dev_info
);
489 kmem_cache_free(blk_requestq_cachep
, q
);
493 init_timer(&q
->unplug_timer
);
494 setup_timer(&q
->timeout
, blk_rq_timed_out_timer
, (unsigned long) q
);
495 INIT_LIST_HEAD(&q
->timeout_list
);
497 kobject_init(&q
->kobj
, &blk_queue_ktype
);
499 mutex_init(&q
->sysfs_lock
);
500 spin_lock_init(&q
->__queue_lock
);
504 EXPORT_SYMBOL(blk_alloc_queue_node
);
507 * blk_init_queue - prepare a request queue for use with a block device
508 * @rfn: The function to be called to process requests that have been
509 * placed on the queue.
510 * @lock: Request queue spin lock
513 * If a block device wishes to use the standard request handling procedures,
514 * which sorts requests and coalesces adjacent requests, then it must
515 * call blk_init_queue(). The function @rfn will be called when there
516 * are requests on the queue that need to be processed. If the device
517 * supports plugging, then @rfn may not be called immediately when requests
518 * are available on the queue, but may be called at some time later instead.
519 * Plugged queues are generally unplugged when a buffer belonging to one
520 * of the requests on the queue is needed, or due to memory pressure.
522 * @rfn is not required, or even expected, to remove all requests off the
523 * queue, but only as many as it can handle at a time. If it does leave
524 * requests on the queue, it is responsible for arranging that the requests
525 * get dealt with eventually.
527 * The queue spin lock must be held while manipulating the requests on the
528 * request queue; this lock will be taken also from interrupt context, so irq
529 * disabling is needed for it.
531 * Function returns a pointer to the initialized request queue, or %NULL if
535 * blk_init_queue() must be paired with a blk_cleanup_queue() call
536 * when the block device is deactivated (such as at module unload).
539 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
541 return blk_init_queue_node(rfn
, lock
, -1);
543 EXPORT_SYMBOL(blk_init_queue
);
545 struct request_queue
*
546 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
548 struct request_queue
*q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
554 if (blk_init_free_list(q
)) {
555 kmem_cache_free(blk_requestq_cachep
, q
);
560 * if caller didn't supply a lock, they get per-queue locking with
564 lock
= &q
->__queue_lock
;
567 q
->prep_rq_fn
= NULL
;
568 q
->unplug_fn
= generic_unplug_device
;
569 q
->queue_flags
= (1 << QUEUE_FLAG_CLUSTER
);
570 q
->queue_lock
= lock
;
572 blk_queue_segment_boundary(q
, 0xffffffff);
574 blk_queue_make_request(q
, __make_request
);
575 blk_queue_max_segment_size(q
, MAX_SEGMENT_SIZE
);
577 blk_queue_max_hw_segments(q
, MAX_HW_SEGMENTS
);
578 blk_queue_max_phys_segments(q
, MAX_PHYS_SEGMENTS
);
580 q
->sg_reserved_size
= INT_MAX
;
582 blk_set_cmd_filter_defaults(&q
->cmd_filter
);
587 if (!elevator_init(q
, NULL
)) {
588 blk_queue_congestion_threshold(q
);
595 EXPORT_SYMBOL(blk_init_queue_node
);
597 int blk_get_queue(struct request_queue
*q
)
599 if (likely(!test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
))) {
600 kobject_get(&q
->kobj
);
607 static inline void blk_free_request(struct request_queue
*q
, struct request
*rq
)
609 if (rq
->cmd_flags
& REQ_ELVPRIV
)
610 elv_put_request(q
, rq
);
611 mempool_free(rq
, q
->rq
.rq_pool
);
614 static struct request
*
615 blk_alloc_request(struct request_queue
*q
, int rw
, int priv
, gfp_t gfp_mask
)
617 struct request
*rq
= mempool_alloc(q
->rq
.rq_pool
, gfp_mask
);
624 rq
->cmd_flags
= rw
| REQ_ALLOCED
;
627 if (unlikely(elv_set_request(q
, rq
, gfp_mask
))) {
628 mempool_free(rq
, q
->rq
.rq_pool
);
631 rq
->cmd_flags
|= REQ_ELVPRIV
;
638 * ioc_batching returns true if the ioc is a valid batching request and
639 * should be given priority access to a request.
641 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
647 * Make sure the process is able to allocate at least 1 request
648 * even if the batch times out, otherwise we could theoretically
651 return ioc
->nr_batch_requests
== q
->nr_batching
||
652 (ioc
->nr_batch_requests
> 0
653 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
657 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
658 * will cause the process to be a "batcher" on all queues in the system. This
659 * is the behaviour we want though - once it gets a wakeup it should be given
662 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
664 if (!ioc
|| ioc_batching(q
, ioc
))
667 ioc
->nr_batch_requests
= q
->nr_batching
;
668 ioc
->last_waited
= jiffies
;
671 static void __freed_request(struct request_queue
*q
, int rw
)
673 struct request_list
*rl
= &q
->rq
;
675 if (rl
->count
[rw
] < queue_congestion_off_threshold(q
))
676 blk_clear_queue_congested(q
, rw
);
678 if (rl
->count
[rw
] + 1 <= q
->nr_requests
) {
679 if (waitqueue_active(&rl
->wait
[rw
]))
680 wake_up(&rl
->wait
[rw
]);
682 blk_clear_queue_full(q
, rw
);
687 * A request has just been released. Account for it, update the full and
688 * congestion status, wake up any waiters. Called under q->queue_lock.
690 static void freed_request(struct request_queue
*q
, int rw
, int priv
)
692 struct request_list
*rl
= &q
->rq
;
698 __freed_request(q
, rw
);
700 if (unlikely(rl
->starved
[rw
^ 1]))
701 __freed_request(q
, rw
^ 1);
704 #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
706 * Get a free request, queue_lock must be held.
707 * Returns NULL on failure, with queue_lock held.
708 * Returns !NULL on success, with queue_lock *not held*.
710 static struct request
*get_request(struct request_queue
*q
, int rw_flags
,
711 struct bio
*bio
, gfp_t gfp_mask
)
713 struct request
*rq
= NULL
;
714 struct request_list
*rl
= &q
->rq
;
715 struct io_context
*ioc
= NULL
;
716 const int rw
= rw_flags
& 0x01;
719 may_queue
= elv_may_queue(q
, rw_flags
);
720 if (may_queue
== ELV_MQUEUE_NO
)
723 if (rl
->count
[rw
]+1 >= queue_congestion_on_threshold(q
)) {
724 if (rl
->count
[rw
]+1 >= q
->nr_requests
) {
725 ioc
= current_io_context(GFP_ATOMIC
, q
->node
);
727 * The queue will fill after this allocation, so set
728 * it as full, and mark this process as "batching".
729 * This process will be allowed to complete a batch of
730 * requests, others will be blocked.
732 if (!blk_queue_full(q
, rw
)) {
733 ioc_set_batching(q
, ioc
);
734 blk_set_queue_full(q
, rw
);
736 if (may_queue
!= ELV_MQUEUE_MUST
737 && !ioc_batching(q
, ioc
)) {
739 * The queue is full and the allocating
740 * process is not a "batcher", and not
741 * exempted by the IO scheduler
747 blk_set_queue_congested(q
, rw
);
751 * Only allow batching queuers to allocate up to 50% over the defined
752 * limit of requests, otherwise we could have thousands of requests
753 * allocated with any setting of ->nr_requests
755 if (rl
->count
[rw
] >= (3 * q
->nr_requests
/ 2))
761 priv
= !test_bit(QUEUE_FLAG_ELVSWITCH
, &q
->queue_flags
);
765 spin_unlock_irq(q
->queue_lock
);
767 rq
= blk_alloc_request(q
, rw_flags
, priv
, gfp_mask
);
770 * Allocation failed presumably due to memory. Undo anything
771 * we might have messed up.
773 * Allocating task should really be put onto the front of the
774 * wait queue, but this is pretty rare.
776 spin_lock_irq(q
->queue_lock
);
777 freed_request(q
, rw
, priv
);
780 * in the very unlikely event that allocation failed and no
781 * requests for this direction was pending, mark us starved
782 * so that freeing of a request in the other direction will
783 * notice us. another possible fix would be to split the
784 * rq mempool into READ and WRITE
787 if (unlikely(rl
->count
[rw
] == 0))
794 * ioc may be NULL here, and ioc_batching will be false. That's
795 * OK, if the queue is under the request limit then requests need
796 * not count toward the nr_batch_requests limit. There will always
797 * be some limit enforced by BLK_BATCH_TIME.
799 if (ioc_batching(q
, ioc
))
800 ioc
->nr_batch_requests
--;
802 blk_add_trace_generic(q
, bio
, rw
, BLK_TA_GETRQ
);
808 * No available requests for this queue, unplug the device and wait for some
809 * requests to become available.
811 * Called with q->queue_lock held, and returns with it unlocked.
813 static struct request
*get_request_wait(struct request_queue
*q
, int rw_flags
,
816 const int rw
= rw_flags
& 0x01;
819 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
822 struct io_context
*ioc
;
823 struct request_list
*rl
= &q
->rq
;
825 prepare_to_wait_exclusive(&rl
->wait
[rw
], &wait
,
826 TASK_UNINTERRUPTIBLE
);
828 blk_add_trace_generic(q
, bio
, rw
, BLK_TA_SLEEPRQ
);
830 __generic_unplug_device(q
);
831 spin_unlock_irq(q
->queue_lock
);
835 * After sleeping, we become a "batching" process and
836 * will be able to allocate at least one request, and
837 * up to a big batch of them for a small period time.
838 * See ioc_batching, ioc_set_batching
840 ioc
= current_io_context(GFP_NOIO
, q
->node
);
841 ioc_set_batching(q
, ioc
);
843 spin_lock_irq(q
->queue_lock
);
844 finish_wait(&rl
->wait
[rw
], &wait
);
846 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
852 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
856 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
858 spin_lock_irq(q
->queue_lock
);
859 if (gfp_mask
& __GFP_WAIT
) {
860 rq
= get_request_wait(q
, rw
, NULL
);
862 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
864 spin_unlock_irq(q
->queue_lock
);
866 /* q->queue_lock is unlocked at this point */
870 EXPORT_SYMBOL(blk_get_request
);
873 * blk_start_queueing - initiate dispatch of requests to device
874 * @q: request queue to kick into gear
876 * This is basically a helper to remove the need to know whether a queue
877 * is plugged or not if someone just wants to initiate dispatch of requests
880 * The queue lock must be held with interrupts disabled.
882 void blk_start_queueing(struct request_queue
*q
)
884 if (!blk_queue_plugged(q
))
887 __generic_unplug_device(q
);
889 EXPORT_SYMBOL(blk_start_queueing
);
892 * blk_requeue_request - put a request back on queue
893 * @q: request queue where request should be inserted
894 * @rq: request to be inserted
897 * Drivers often keep queueing requests until the hardware cannot accept
898 * more, when that condition happens we need to put the request back
899 * on the queue. Must be called with queue lock held.
901 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
903 blk_delete_timer(rq
);
904 blk_clear_rq_complete(rq
);
905 blk_add_trace_rq(q
, rq
, BLK_TA_REQUEUE
);
907 if (blk_rq_tagged(rq
))
908 blk_queue_end_tag(q
, rq
);
910 elv_requeue_request(q
, rq
);
912 EXPORT_SYMBOL(blk_requeue_request
);
915 * blk_insert_request - insert a special request into a request queue
916 * @q: request queue where request should be inserted
917 * @rq: request to be inserted
918 * @at_head: insert request at head or tail of queue
919 * @data: private data
922 * Many block devices need to execute commands asynchronously, so they don't
923 * block the whole kernel from preemption during request execution. This is
924 * accomplished normally by inserting aritficial requests tagged as
925 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
926 * be scheduled for actual execution by the request queue.
928 * We have the option of inserting the head or the tail of the queue.
929 * Typically we use the tail for new ioctls and so forth. We use the head
930 * of the queue for things like a QUEUE_FULL message from a device, or a
931 * host that is unable to accept a particular command.
933 void blk_insert_request(struct request_queue
*q
, struct request
*rq
,
934 int at_head
, void *data
)
936 int where
= at_head
? ELEVATOR_INSERT_FRONT
: ELEVATOR_INSERT_BACK
;
940 * tell I/O scheduler that this isn't a regular read/write (ie it
941 * must not attempt merges on this) and that it acts as a soft
944 rq
->cmd_type
= REQ_TYPE_SPECIAL
;
945 rq
->cmd_flags
|= REQ_SOFTBARRIER
;
949 spin_lock_irqsave(q
->queue_lock
, flags
);
952 * If command is tagged, release the tag
954 if (blk_rq_tagged(rq
))
955 blk_queue_end_tag(q
, rq
);
957 drive_stat_acct(rq
, 1);
958 __elv_add_request(q
, rq
, where
, 0);
959 blk_start_queueing(q
);
960 spin_unlock_irqrestore(q
->queue_lock
, flags
);
962 EXPORT_SYMBOL(blk_insert_request
);
965 * add-request adds a request to the linked list.
966 * queue lock is held and interrupts disabled, as we muck with the
967 * request queue list.
969 static inline void add_request(struct request_queue
*q
, struct request
*req
)
971 drive_stat_acct(req
, 1);
974 * elevator indicated where it wants this request to be
975 * inserted at elevator_merge time
977 __elv_add_request(q
, req
, ELEVATOR_INSERT_SORT
, 0);
980 static void part_round_stats_single(int cpu
, struct hd_struct
*part
,
983 if (now
== part
->stamp
)
986 if (part
->in_flight
) {
987 __part_stat_add(cpu
, part
, time_in_queue
,
988 part
->in_flight
* (now
- part
->stamp
));
989 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
995 * part_round_stats() - Round off the performance stats on a struct
998 * The average IO queue length and utilisation statistics are maintained
999 * by observing the current state of the queue length and the amount of
1000 * time it has been in this state for.
1002 * Normally, that accounting is done on IO completion, but that can result
1003 * in more than a second's worth of IO being accounted for within any one
1004 * second, leading to >100% utilisation. To deal with that, we call this
1005 * function to do a round-off before returning the results when reading
1006 * /proc/diskstats. This accounts immediately for all queue usage up to
1007 * the current jiffies and restarts the counters again.
1009 void part_round_stats(int cpu
, struct hd_struct
*part
)
1011 unsigned long now
= jiffies
;
1014 part_round_stats_single(cpu
, &part_to_disk(part
)->part0
, now
);
1015 part_round_stats_single(cpu
, part
, now
);
1017 EXPORT_SYMBOL_GPL(part_round_stats
);
1020 * queue lock must be held
1022 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1026 if (unlikely(--req
->ref_count
))
1029 elv_completed_request(q
, req
);
1032 * Request may not have originated from ll_rw_blk. if not,
1033 * it didn't come out of our reserved rq pools
1035 if (req
->cmd_flags
& REQ_ALLOCED
) {
1036 int rw
= rq_data_dir(req
);
1037 int priv
= req
->cmd_flags
& REQ_ELVPRIV
;
1039 BUG_ON(!list_empty(&req
->queuelist
));
1040 BUG_ON(!hlist_unhashed(&req
->hash
));
1042 blk_free_request(q
, req
);
1043 freed_request(q
, rw
, priv
);
1046 EXPORT_SYMBOL_GPL(__blk_put_request
);
1048 void blk_put_request(struct request
*req
)
1050 unsigned long flags
;
1051 struct request_queue
*q
= req
->q
;
1053 spin_lock_irqsave(q
->queue_lock
, flags
);
1054 __blk_put_request(q
, req
);
1055 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1057 EXPORT_SYMBOL(blk_put_request
);
1059 void init_request_from_bio(struct request
*req
, struct bio
*bio
)
1061 req
->cpu
= bio
->bi_comp_cpu
;
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_discard(bio
))) {
1074 req
->cmd_flags
|= REQ_DISCARD
;
1075 if (bio_barrier(bio
))
1076 req
->cmd_flags
|= REQ_SOFTBARRIER
;
1077 req
->q
->prepare_discard_fn(req
->q
, req
);
1078 } else if (unlikely(bio_barrier(bio
)))
1079 req
->cmd_flags
|= (REQ_HARDBARRIER
| REQ_NOMERGE
);
1082 req
->cmd_flags
|= REQ_RW_SYNC
;
1083 if (bio_rw_meta(bio
))
1084 req
->cmd_flags
|= REQ_RW_META
;
1087 req
->hard_sector
= req
->sector
= bio
->bi_sector
;
1088 req
->ioprio
= bio_prio(bio
);
1089 req
->start_time
= jiffies
;
1090 blk_rq_bio_prep(req
->q
, req
, bio
);
1093 static int __make_request(struct request_queue
*q
, struct bio
*bio
)
1095 struct request
*req
;
1096 int el_ret
, nr_sectors
, barrier
, discard
, err
;
1097 const unsigned short prio
= bio_prio(bio
);
1098 const int sync
= bio_sync(bio
);
1101 nr_sectors
= bio_sectors(bio
);
1104 * low level driver can indicate that it wants pages above a
1105 * certain limit bounced to low memory (ie for highmem, or even
1106 * ISA dma in theory)
1108 blk_queue_bounce(q
, &bio
);
1110 barrier
= bio_barrier(bio
);
1111 if (unlikely(barrier
) && bio_has_data(bio
) &&
1112 (q
->next_ordered
== QUEUE_ORDERED_NONE
)) {
1117 discard
= bio_discard(bio
);
1118 if (unlikely(discard
) && !q
->prepare_discard_fn
) {
1123 spin_lock_irq(q
->queue_lock
);
1125 if (unlikely(barrier
) || elv_queue_empty(q
))
1128 el_ret
= elv_merge(q
, &req
, bio
);
1130 case ELEVATOR_BACK_MERGE
:
1131 BUG_ON(!rq_mergeable(req
));
1133 if (!ll_back_merge_fn(q
, req
, bio
))
1136 blk_add_trace_bio(q
, bio
, BLK_TA_BACKMERGE
);
1138 req
->biotail
->bi_next
= bio
;
1140 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
1141 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
1142 if (!blk_rq_cpu_valid(req
))
1143 req
->cpu
= bio
->bi_comp_cpu
;
1144 drive_stat_acct(req
, 0);
1145 if (!attempt_back_merge(q
, req
))
1146 elv_merged_request(q
, req
, el_ret
);
1149 case ELEVATOR_FRONT_MERGE
:
1150 BUG_ON(!rq_mergeable(req
));
1152 if (!ll_front_merge_fn(q
, req
, bio
))
1155 blk_add_trace_bio(q
, bio
, BLK_TA_FRONTMERGE
);
1157 bio
->bi_next
= req
->bio
;
1161 * may not be valid. if the low level driver said
1162 * it didn't need a bounce buffer then it better
1163 * not touch req->buffer either...
1165 req
->buffer
= bio_data(bio
);
1166 req
->current_nr_sectors
= bio_cur_sectors(bio
);
1167 req
->hard_cur_sectors
= req
->current_nr_sectors
;
1168 req
->sector
= req
->hard_sector
= bio
->bi_sector
;
1169 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
1170 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
1171 if (!blk_rq_cpu_valid(req
))
1172 req
->cpu
= bio
->bi_comp_cpu
;
1173 drive_stat_acct(req
, 0);
1174 if (!attempt_front_merge(q
, req
))
1175 elv_merged_request(q
, req
, el_ret
);
1178 /* ELV_NO_MERGE: elevator says don't/can't merge. */
1185 * This sync check and mask will be re-done in init_request_from_bio(),
1186 * but we need to set it earlier to expose the sync flag to the
1187 * rq allocator and io schedulers.
1189 rw_flags
= bio_data_dir(bio
);
1191 rw_flags
|= REQ_RW_SYNC
;
1194 * Grab a free request. This is might sleep but can not fail.
1195 * Returns with the queue unlocked.
1197 req
= get_request_wait(q
, rw_flags
, bio
);
1200 * After dropping the lock and possibly sleeping here, our request
1201 * may now be mergeable after it had proven unmergeable (above).
1202 * We don't worry about that case for efficiency. It won't happen
1203 * often, and the elevators are able to handle it.
1205 init_request_from_bio(req
, bio
);
1207 spin_lock_irq(q
->queue_lock
);
1208 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
) ||
1209 bio_flagged(bio
, BIO_CPU_AFFINE
))
1210 req
->cpu
= blk_cpu_to_group(smp_processor_id());
1211 if (elv_queue_empty(q
))
1213 add_request(q
, req
);
1216 __generic_unplug_device(q
);
1217 spin_unlock_irq(q
->queue_lock
);
1221 bio_endio(bio
, err
);
1226 * If bio->bi_dev is a partition, remap the location
1228 static inline void blk_partition_remap(struct bio
*bio
)
1230 struct block_device
*bdev
= bio
->bi_bdev
;
1232 if (bio_sectors(bio
) && bdev
!= bdev
->bd_contains
) {
1233 struct hd_struct
*p
= bdev
->bd_part
;
1235 bio
->bi_sector
+= p
->start_sect
;
1236 bio
->bi_bdev
= bdev
->bd_contains
;
1238 blk_add_trace_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
1239 bdev
->bd_dev
, bio
->bi_sector
,
1240 bio
->bi_sector
- p
->start_sect
);
1244 static void handle_bad_sector(struct bio
*bio
)
1246 char b
[BDEVNAME_SIZE
];
1248 printk(KERN_INFO
"attempt to access beyond end of device\n");
1249 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
1250 bdevname(bio
->bi_bdev
, b
),
1252 (unsigned long long)bio
->bi_sector
+ bio_sectors(bio
),
1253 (long long)(bio
->bi_bdev
->bd_inode
->i_size
>> 9));
1255 set_bit(BIO_EOF
, &bio
->bi_flags
);
1258 #ifdef CONFIG_FAIL_MAKE_REQUEST
1260 static DECLARE_FAULT_ATTR(fail_make_request
);
1262 static int __init
setup_fail_make_request(char *str
)
1264 return setup_fault_attr(&fail_make_request
, str
);
1266 __setup("fail_make_request=", setup_fail_make_request
);
1268 static int should_fail_request(struct bio
*bio
)
1270 struct hd_struct
*part
= bio
->bi_bdev
->bd_part
;
1272 if (part_to_disk(part
)->part0
.make_it_fail
|| part
->make_it_fail
)
1273 return should_fail(&fail_make_request
, bio
->bi_size
);
1278 static int __init
fail_make_request_debugfs(void)
1280 return init_fault_attr_dentries(&fail_make_request
,
1281 "fail_make_request");
1284 late_initcall(fail_make_request_debugfs
);
1286 #else /* CONFIG_FAIL_MAKE_REQUEST */
1288 static inline int should_fail_request(struct bio
*bio
)
1293 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1296 * Check whether this bio extends beyond the end of the device.
1298 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
1305 /* Test device or partition size, when known. */
1306 maxsector
= bio
->bi_bdev
->bd_inode
->i_size
>> 9;
1308 sector_t sector
= bio
->bi_sector
;
1310 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
1312 * This may well happen - the kernel calls bread()
1313 * without checking the size of the device, e.g., when
1314 * mounting a device.
1316 handle_bad_sector(bio
);
1325 * generic_make_request - hand a buffer to its device driver for I/O
1326 * @bio: The bio describing the location in memory and on the device.
1328 * generic_make_request() is used to make I/O requests of block
1329 * devices. It is passed a &struct bio, which describes the I/O that needs
1332 * generic_make_request() does not return any status. The
1333 * success/failure status of the request, along with notification of
1334 * completion, is delivered asynchronously through the bio->bi_end_io
1335 * function described (one day) else where.
1337 * The caller of generic_make_request must make sure that bi_io_vec
1338 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1339 * set to describe the device address, and the
1340 * bi_end_io and optionally bi_private are set to describe how
1341 * completion notification should be signaled.
1343 * generic_make_request and the drivers it calls may use bi_next if this
1344 * bio happens to be merged with someone else, and may change bi_dev and
1345 * bi_sector for remaps as it sees fit. So the values of these fields
1346 * should NOT be depended on after the call to generic_make_request.
1348 static inline void __generic_make_request(struct bio
*bio
)
1350 struct request_queue
*q
;
1351 sector_t old_sector
;
1352 int ret
, nr_sectors
= bio_sectors(bio
);
1358 if (bio_check_eod(bio
, nr_sectors
))
1362 * Resolve the mapping until finished. (drivers are
1363 * still free to implement/resolve their own stacking
1364 * by explicitly returning 0)
1366 * NOTE: we don't repeat the blk_size check for each new device.
1367 * Stacking drivers are expected to know what they are doing.
1372 char b
[BDEVNAME_SIZE
];
1374 q
= bdev_get_queue(bio
->bi_bdev
);
1377 "generic_make_request: Trying to access "
1378 "nonexistent block-device %s (%Lu)\n",
1379 bdevname(bio
->bi_bdev
, b
),
1380 (long long) bio
->bi_sector
);
1382 bio_endio(bio
, err
);
1386 if (unlikely(nr_sectors
> q
->max_hw_sectors
)) {
1387 printk(KERN_ERR
"bio too big device %s (%u > %u)\n",
1388 bdevname(bio
->bi_bdev
, b
),
1394 if (unlikely(test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)))
1397 if (should_fail_request(bio
))
1401 * If this device has partitions, remap block n
1402 * of partition p to block n+start(p) of the disk.
1404 blk_partition_remap(bio
);
1406 if (bio_integrity_enabled(bio
) && bio_integrity_prep(bio
))
1409 if (old_sector
!= -1)
1410 blk_add_trace_remap(q
, bio
, old_dev
, bio
->bi_sector
,
1413 blk_add_trace_bio(q
, bio
, BLK_TA_QUEUE
);
1415 old_sector
= bio
->bi_sector
;
1416 old_dev
= bio
->bi_bdev
->bd_dev
;
1418 if (bio_check_eod(bio
, nr_sectors
))
1420 if ((bio_empty_barrier(bio
) && !q
->prepare_flush_fn
) ||
1421 (bio_discard(bio
) && !q
->prepare_discard_fn
)) {
1426 ret
= q
->make_request_fn(q
, bio
);
1431 * We only want one ->make_request_fn to be active at a time,
1432 * else stack usage with stacked devices could be a problem.
1433 * So use current->bio_{list,tail} to keep a list of requests
1434 * submited by a make_request_fn function.
1435 * current->bio_tail is also used as a flag to say if
1436 * generic_make_request is currently active in this task or not.
1437 * If it is NULL, then no make_request is active. If it is non-NULL,
1438 * then a make_request is active, and new requests should be added
1441 void generic_make_request(struct bio
*bio
)
1443 if (current
->bio_tail
) {
1444 /* make_request is active */
1445 *(current
->bio_tail
) = bio
;
1446 bio
->bi_next
= NULL
;
1447 current
->bio_tail
= &bio
->bi_next
;
1450 /* following loop may be a bit non-obvious, and so deserves some
1452 * Before entering the loop, bio->bi_next is NULL (as all callers
1453 * ensure that) so we have a list with a single bio.
1454 * We pretend that we have just taken it off a longer list, so
1455 * we assign bio_list to the next (which is NULL) and bio_tail
1456 * to &bio_list, thus initialising the bio_list of new bios to be
1457 * added. __generic_make_request may indeed add some more bios
1458 * through a recursive call to generic_make_request. If it
1459 * did, we find a non-NULL value in bio_list and re-enter the loop
1460 * from the top. In this case we really did just take the bio
1461 * of the top of the list (no pretending) and so fixup bio_list and
1462 * bio_tail or bi_next, and call into __generic_make_request again.
1464 * The loop was structured like this to make only one call to
1465 * __generic_make_request (which is important as it is large and
1466 * inlined) and to keep the structure simple.
1468 BUG_ON(bio
->bi_next
);
1470 current
->bio_list
= bio
->bi_next
;
1471 if (bio
->bi_next
== NULL
)
1472 current
->bio_tail
= ¤t
->bio_list
;
1474 bio
->bi_next
= NULL
;
1475 __generic_make_request(bio
);
1476 bio
= current
->bio_list
;
1478 current
->bio_tail
= NULL
; /* deactivate */
1480 EXPORT_SYMBOL(generic_make_request
);
1483 * submit_bio - submit a bio to the block device layer for I/O
1484 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1485 * @bio: The &struct bio which describes the I/O
1487 * submit_bio() is very similar in purpose to generic_make_request(), and
1488 * uses that function to do most of the work. Both are fairly rough
1489 * interfaces; @bio must be presetup and ready for I/O.
1492 void submit_bio(int rw
, struct bio
*bio
)
1494 int count
= bio_sectors(bio
);
1499 * If it's a regular read/write or a barrier with data attached,
1500 * go through the normal accounting stuff before submission.
1502 if (bio_has_data(bio
)) {
1504 count_vm_events(PGPGOUT
, count
);
1506 task_io_account_read(bio
->bi_size
);
1507 count_vm_events(PGPGIN
, count
);
1510 if (unlikely(block_dump
)) {
1511 char b
[BDEVNAME_SIZE
];
1512 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s\n",
1513 current
->comm
, task_pid_nr(current
),
1514 (rw
& WRITE
) ? "WRITE" : "READ",
1515 (unsigned long long)bio
->bi_sector
,
1516 bdevname(bio
->bi_bdev
, b
));
1520 generic_make_request(bio
);
1522 EXPORT_SYMBOL(submit_bio
);
1525 * __end_that_request_first - end I/O on a request
1526 * @req: the request being processed
1527 * @error: %0 for success, < %0 for error
1528 * @nr_bytes: number of bytes to complete
1531 * Ends I/O on a number of bytes attached to @req, and sets it up
1532 * for the next range of segments (if any) in the cluster.
1535 * %0 - we are done with this request, call end_that_request_last()
1536 * %1 - still buffers pending for this request
1538 static int __end_that_request_first(struct request
*req
, int error
,
1541 int total_bytes
, bio_nbytes
, next_idx
= 0;
1544 blk_add_trace_rq(req
->q
, req
, BLK_TA_COMPLETE
);
1547 * for a REQ_TYPE_BLOCK_PC request, we want to carry any eventual
1548 * sense key with us all the way through
1550 if (!blk_pc_request(req
))
1553 if (error
&& (blk_fs_request(req
) && !(req
->cmd_flags
& REQ_QUIET
))) {
1554 printk(KERN_ERR
"end_request: I/O error, dev %s, sector %llu\n",
1555 req
->rq_disk
? req
->rq_disk
->disk_name
: "?",
1556 (unsigned long long)req
->sector
);
1559 if (blk_fs_request(req
) && req
->rq_disk
) {
1560 const int rw
= rq_data_dir(req
);
1561 struct hd_struct
*part
;
1564 cpu
= part_stat_lock();
1565 part
= disk_map_sector_rcu(req
->rq_disk
, req
->sector
);
1566 part_stat_add(cpu
, part
, sectors
[rw
], nr_bytes
>> 9);
1570 total_bytes
= bio_nbytes
= 0;
1571 while ((bio
= req
->bio
) != NULL
) {
1575 * For an empty barrier request, the low level driver must
1576 * store a potential error location in ->sector. We pass
1577 * that back up in ->bi_sector.
1579 if (blk_empty_barrier(req
))
1580 bio
->bi_sector
= req
->sector
;
1582 if (nr_bytes
>= bio
->bi_size
) {
1583 req
->bio
= bio
->bi_next
;
1584 nbytes
= bio
->bi_size
;
1585 req_bio_endio(req
, bio
, nbytes
, error
);
1589 int idx
= bio
->bi_idx
+ next_idx
;
1591 if (unlikely(bio
->bi_idx
>= bio
->bi_vcnt
)) {
1592 blk_dump_rq_flags(req
, "__end_that");
1593 printk(KERN_ERR
"%s: bio idx %d >= vcnt %d\n",
1594 __func__
, bio
->bi_idx
, bio
->bi_vcnt
);
1598 nbytes
= bio_iovec_idx(bio
, idx
)->bv_len
;
1599 BIO_BUG_ON(nbytes
> bio
->bi_size
);
1602 * not a complete bvec done
1604 if (unlikely(nbytes
> nr_bytes
)) {
1605 bio_nbytes
+= nr_bytes
;
1606 total_bytes
+= nr_bytes
;
1611 * advance to the next vector
1614 bio_nbytes
+= nbytes
;
1617 total_bytes
+= nbytes
;
1623 * end more in this run, or just return 'not-done'
1625 if (unlikely(nr_bytes
<= 0))
1637 * if the request wasn't completed, update state
1640 req_bio_endio(req
, bio
, bio_nbytes
, error
);
1641 bio
->bi_idx
+= next_idx
;
1642 bio_iovec(bio
)->bv_offset
+= nr_bytes
;
1643 bio_iovec(bio
)->bv_len
-= nr_bytes
;
1646 blk_recalc_rq_sectors(req
, total_bytes
>> 9);
1647 blk_recalc_rq_segments(req
);
1652 * queue lock must be held
1654 static void end_that_request_last(struct request
*req
, int error
)
1656 struct gendisk
*disk
= req
->rq_disk
;
1658 blk_delete_timer(req
);
1660 if (blk_rq_tagged(req
))
1661 blk_queue_end_tag(req
->q
, req
);
1663 if (blk_queued_rq(req
))
1664 blkdev_dequeue_request(req
);
1666 if (unlikely(laptop_mode
) && blk_fs_request(req
))
1667 laptop_io_completion();
1670 * Account IO completion. bar_rq isn't accounted as a normal
1671 * IO on queueing nor completion. Accounting the containing
1672 * request is enough.
1674 if (disk
&& blk_fs_request(req
) && req
!= &req
->q
->bar_rq
) {
1675 unsigned long duration
= jiffies
- req
->start_time
;
1676 const int rw
= rq_data_dir(req
);
1677 struct hd_struct
*part
;
1680 cpu
= part_stat_lock();
1681 part
= disk_map_sector_rcu(disk
, req
->sector
);
1683 part_stat_inc(cpu
, part
, ios
[rw
]);
1684 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
1685 part_round_stats(cpu
, part
);
1686 part_dec_in_flight(part
);
1692 req
->end_io(req
, error
);
1694 if (blk_bidi_rq(req
))
1695 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
1697 __blk_put_request(req
->q
, req
);
1701 static inline void __end_request(struct request
*rq
, int uptodate
,
1702 unsigned int nr_bytes
)
1707 error
= uptodate
? uptodate
: -EIO
;
1709 __blk_end_request(rq
, error
, nr_bytes
);
1713 * blk_rq_bytes - Returns bytes left to complete in the entire request
1714 * @rq: the request being processed
1716 unsigned int blk_rq_bytes(struct request
*rq
)
1718 if (blk_fs_request(rq
))
1719 return rq
->hard_nr_sectors
<< 9;
1721 return rq
->data_len
;
1723 EXPORT_SYMBOL_GPL(blk_rq_bytes
);
1726 * blk_rq_cur_bytes - Returns bytes left to complete in the current segment
1727 * @rq: the request being processed
1729 unsigned int blk_rq_cur_bytes(struct request
*rq
)
1731 if (blk_fs_request(rq
))
1732 return rq
->current_nr_sectors
<< 9;
1735 return rq
->bio
->bi_size
;
1737 return rq
->data_len
;
1739 EXPORT_SYMBOL_GPL(blk_rq_cur_bytes
);
1742 * end_queued_request - end all I/O on a queued request
1743 * @rq: the request being processed
1744 * @uptodate: error value or %0/%1 uptodate flag
1747 * Ends all I/O on a request, and removes it from the block layer queues.
1748 * Not suitable for normal I/O completion, unless the driver still has
1749 * the request attached to the block layer.
1752 void end_queued_request(struct request
*rq
, int uptodate
)
1754 __end_request(rq
, uptodate
, blk_rq_bytes(rq
));
1756 EXPORT_SYMBOL(end_queued_request
);
1759 * end_dequeued_request - end all I/O on a dequeued request
1760 * @rq: the request being processed
1761 * @uptodate: error value or %0/%1 uptodate flag
1764 * Ends all I/O on a request. The request must already have been
1765 * dequeued using blkdev_dequeue_request(), as is normally the case
1769 void end_dequeued_request(struct request
*rq
, int uptodate
)
1771 __end_request(rq
, uptodate
, blk_rq_bytes(rq
));
1773 EXPORT_SYMBOL(end_dequeued_request
);
1777 * end_request - end I/O on the current segment of the request
1778 * @req: the request being processed
1779 * @uptodate: error value or %0/%1 uptodate flag
1782 * Ends I/O on the current segment of a request. If that is the only
1783 * remaining segment, the request is also completed and freed.
1785 * This is a remnant of how older block drivers handled I/O completions.
1786 * Modern drivers typically end I/O on the full request in one go, unless
1787 * they have a residual value to account for. For that case this function
1788 * isn't really useful, unless the residual just happens to be the
1789 * full current segment. In other words, don't use this function in new
1790 * code. Use blk_end_request() or __blk_end_request() to end partial parts
1791 * of a request, or end_dequeued_request() and end_queued_request() to
1792 * completely end IO on a dequeued/queued request.
1795 void end_request(struct request
*req
, int uptodate
)
1797 __end_request(req
, uptodate
, req
->hard_cur_sectors
<< 9);
1799 EXPORT_SYMBOL(end_request
);
1802 * blk_end_io - Generic end_io function to complete a request.
1803 * @rq: the request being processed
1804 * @error: %0 for success, < %0 for error
1805 * @nr_bytes: number of bytes to complete @rq
1806 * @bidi_bytes: number of bytes to complete @rq->next_rq
1807 * @drv_callback: function called between completion of bios in the request
1808 * and completion of the request.
1809 * If the callback returns non %0, this helper returns without
1810 * completion of the request.
1813 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
1814 * If @rq has leftover, sets it up for the next range of segments.
1817 * %0 - we are done with this request
1818 * %1 - this request is not freed yet, it still has pending buffers.
1820 static int blk_end_io(struct request
*rq
, int error
, unsigned int nr_bytes
,
1821 unsigned int bidi_bytes
,
1822 int (drv_callback
)(struct request
*))
1824 struct request_queue
*q
= rq
->q
;
1825 unsigned long flags
= 0UL;
1828 if (__end_that_request_first(rq
, error
, nr_bytes
))
1831 /* Bidi request must be completed as a whole */
1832 if (blk_bidi_rq(rq
) &&
1833 __end_that_request_first(rq
->next_rq
, error
, bidi_bytes
))
1837 /* Special feature for tricky drivers */
1838 if (drv_callback
&& drv_callback(rq
))
1841 add_disk_randomness(rq
->rq_disk
);
1843 spin_lock_irqsave(q
->queue_lock
, flags
);
1844 end_that_request_last(rq
, error
);
1845 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1851 * blk_end_request - Helper function for drivers to complete the request.
1852 * @rq: the request being processed
1853 * @error: %0 for success, < %0 for error
1854 * @nr_bytes: number of bytes to complete
1857 * Ends I/O on a number of bytes attached to @rq.
1858 * If @rq has leftover, sets it up for the next range of segments.
1861 * %0 - we are done with this request
1862 * %1 - still buffers pending for this request
1864 int blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
1866 return blk_end_io(rq
, error
, nr_bytes
, 0, NULL
);
1868 EXPORT_SYMBOL_GPL(blk_end_request
);
1871 * __blk_end_request - Helper function for drivers to complete the request.
1872 * @rq: the request being processed
1873 * @error: %0 for success, < %0 for error
1874 * @nr_bytes: number of bytes to complete
1877 * Must be called with queue lock held unlike blk_end_request().
1880 * %0 - we are done with this request
1881 * %1 - still buffers pending for this request
1883 int __blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
1885 if (rq
->bio
&& __end_that_request_first(rq
, error
, nr_bytes
))
1888 add_disk_randomness(rq
->rq_disk
);
1890 end_that_request_last(rq
, error
);
1894 EXPORT_SYMBOL_GPL(__blk_end_request
);
1897 * blk_end_bidi_request - Helper function for drivers to complete bidi request.
1898 * @rq: the bidi request being processed
1899 * @error: %0 for success, < %0 for error
1900 * @nr_bytes: number of bytes to complete @rq
1901 * @bidi_bytes: number of bytes to complete @rq->next_rq
1904 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
1907 * %0 - we are done with this request
1908 * %1 - still buffers pending for this request
1910 int blk_end_bidi_request(struct request
*rq
, int error
, unsigned int nr_bytes
,
1911 unsigned int bidi_bytes
)
1913 return blk_end_io(rq
, error
, nr_bytes
, bidi_bytes
, NULL
);
1915 EXPORT_SYMBOL_GPL(blk_end_bidi_request
);
1918 * blk_end_request_callback - Special helper function for tricky drivers
1919 * @rq: the request being processed
1920 * @error: %0 for success, < %0 for error
1921 * @nr_bytes: number of bytes to complete
1922 * @drv_callback: function called between completion of bios in the request
1923 * and completion of the request.
1924 * If the callback returns non %0, this helper returns without
1925 * completion of the request.
1928 * Ends I/O on a number of bytes attached to @rq.
1929 * If @rq has leftover, sets it up for the next range of segments.
1931 * This special helper function is used only for existing tricky drivers.
1932 * (e.g. cdrom_newpc_intr() of ide-cd)
1933 * This interface will be removed when such drivers are rewritten.
1934 * Don't use this interface in other places anymore.
1937 * %0 - we are done with this request
1938 * %1 - this request is not freed yet.
1939 * this request still has pending buffers or
1940 * the driver doesn't want to finish this request yet.
1942 int blk_end_request_callback(struct request
*rq
, int error
,
1943 unsigned int nr_bytes
,
1944 int (drv_callback
)(struct request
*))
1946 return blk_end_io(rq
, error
, nr_bytes
, 0, drv_callback
);
1948 EXPORT_SYMBOL_GPL(blk_end_request_callback
);
1950 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
1953 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw, and
1954 we want BIO_RW_AHEAD (bit 1) to imply REQ_FAILFAST (bit 1). */
1955 rq
->cmd_flags
|= (bio
->bi_rw
& 3);
1957 if (bio_has_data(bio
)) {
1958 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
1959 rq
->buffer
= bio_data(bio
);
1961 rq
->current_nr_sectors
= bio_cur_sectors(bio
);
1962 rq
->hard_cur_sectors
= rq
->current_nr_sectors
;
1963 rq
->hard_nr_sectors
= rq
->nr_sectors
= bio_sectors(bio
);
1964 rq
->data_len
= bio
->bi_size
;
1966 rq
->bio
= rq
->biotail
= bio
;
1969 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
1972 int kblockd_schedule_work(struct request_queue
*q
, struct work_struct
*work
)
1974 return queue_work(kblockd_workqueue
, work
);
1976 EXPORT_SYMBOL(kblockd_schedule_work
);
1978 void kblockd_flush_work(struct work_struct
*work
)
1980 cancel_work_sync(work
);
1982 EXPORT_SYMBOL(kblockd_flush_work
);
1984 int __init
blk_dev_init(void)
1986 kblockd_workqueue
= create_workqueue("kblockd");
1987 if (!kblockd_workqueue
)
1988 panic("Failed to create kblockd\n");
1990 request_cachep
= kmem_cache_create("blkdev_requests",
1991 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
1993 blk_requestq_cachep
= kmem_cache_create("blkdev_queue",
1994 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);