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 <<<<<<< HEAD
:block
/blk
-core
.c
42 struct kmem_cache
*request_cachep
;
44 static struct kmem_cache
*request_cachep
;
45 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a
:block
/blk
-core
.c
48 * For queue allocation
50 struct kmem_cache
*blk_requestq_cachep
;
53 * Controlling structure to kblockd
55 static struct workqueue_struct
*kblockd_workqueue
;
57 static DEFINE_PER_CPU(struct list_head
, blk_cpu_done
);
59 static void drive_stat_acct(struct request
*rq
, int new_io
)
61 int rw
= rq_data_dir(rq
);
63 if (!blk_fs_request(rq
) || !rq
->rq_disk
)
67 __all_stat_inc(rq
->rq_disk
, merges
[rw
], rq
->sector
);
69 struct hd_struct
*part
= get_part(rq
->rq_disk
, rq
->sector
);
70 disk_round_stats(rq
->rq_disk
);
71 rq
->rq_disk
->in_flight
++;
73 part_round_stats(part
);
79 void blk_queue_congestion_threshold(struct request_queue
*q
)
83 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) + 1;
84 if (nr
> q
->nr_requests
)
86 q
->nr_congestion_on
= nr
;
88 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) - (q
->nr_requests
/ 16) - 1;
91 q
->nr_congestion_off
= nr
;
95 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
98 * Locates the passed device's request queue and returns the address of its
101 * Will return NULL if the request queue cannot be located.
103 struct backing_dev_info
*blk_get_backing_dev_info(struct block_device
*bdev
)
105 struct backing_dev_info
*ret
= NULL
;
106 struct request_queue
*q
= bdev_get_queue(bdev
);
109 ret
= &q
->backing_dev_info
;
112 EXPORT_SYMBOL(blk_get_backing_dev_info
);
115 * We can't just memset() the structure, since the allocation path
116 * already stored some information in the request.
118 void rq_init(struct request_queue
*q
, struct request
*rq
)
120 INIT_LIST_HEAD(&rq
->queuelist
);
121 INIT_LIST_HEAD(&rq
->donelist
);
123 rq
->sector
= rq
->hard_sector
= (sector_t
) -1;
124 rq
->nr_sectors
= rq
->hard_nr_sectors
= 0;
125 rq
->current_nr_sectors
= rq
->hard_cur_sectors
= 0;
126 rq
->bio
= rq
->biotail
= NULL
;
127 INIT_HLIST_NODE(&rq
->hash
);
128 RB_CLEAR_NODE(&rq
->rb_node
);
130 rq
->nr_phys_segments
= 0;
131 rq
->nr_hw_segments
= 0;
139 memset(rq
->cmd
, 0, sizeof(rq
->cmd
));
141 <<<<<<< HEAD
:block
/blk
-core
.c
144 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a
:block
/blk
-core
.c
149 rq
->end_io_data
= NULL
;
153 static void req_bio_endio(struct request
*rq
, struct bio
*bio
,
154 unsigned int nbytes
, int error
)
156 struct request_queue
*q
= rq
->q
;
158 if (&q
->bar_rq
!= rq
) {
160 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
161 else if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
164 if (unlikely(nbytes
> bio
->bi_size
)) {
165 printk(KERN_ERR
"%s: want %u bytes done, %u left\n",
166 __FUNCTION__
, nbytes
, bio
->bi_size
);
167 nbytes
= bio
->bi_size
;
170 bio
->bi_size
-= nbytes
;
171 bio
->bi_sector
+= (nbytes
>> 9);
172 if (bio
->bi_size
== 0)
173 bio_endio(bio
, error
);
177 * Okay, this is the barrier request in progress, just
180 if (error
&& !q
->orderr
)
185 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
189 printk(KERN_INFO
"%s: dev %s: type=%x, flags=%x\n", msg
,
190 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->cmd_type
,
193 printk(KERN_INFO
" sector %llu, nr/cnr %lu/%u\n",
194 (unsigned long long)rq
->sector
,
196 rq
->current_nr_sectors
);
197 printk(KERN_INFO
" bio %p, biotail %p, buffer %p, data %p, len %u\n",
198 rq
->bio
, rq
->biotail
,
199 rq
->buffer
, rq
->data
,
202 if (blk_pc_request(rq
)) {
203 printk(KERN_INFO
" cdb: ");
204 for (bit
= 0; bit
< sizeof(rq
->cmd
); bit
++)
205 printk("%02x ", rq
->cmd
[bit
]);
209 EXPORT_SYMBOL(blk_dump_rq_flags
);
212 * "plug" the device if there are no outstanding requests: this will
213 * force the transfer to start only after we have put all the requests
216 * This is called with interrupts off and no requests on the queue and
217 * with the queue lock held.
219 void blk_plug_device(struct request_queue
*q
)
221 WARN_ON(!irqs_disabled());
224 * don't plug a stopped queue, it must be paired with blk_start_queue()
225 * which will restart the queueing
227 if (blk_queue_stopped(q
))
230 if (!test_and_set_bit(QUEUE_FLAG_PLUGGED
, &q
->queue_flags
)) {
231 mod_timer(&q
->unplug_timer
, jiffies
+ q
->unplug_delay
);
232 blk_add_trace_generic(q
, NULL
, 0, BLK_TA_PLUG
);
235 EXPORT_SYMBOL(blk_plug_device
);
238 * remove the queue from the plugged list, if present. called with
239 * queue lock held and interrupts disabled.
241 int blk_remove_plug(struct request_queue
*q
)
243 WARN_ON(!irqs_disabled());
245 if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED
, &q
->queue_flags
))
248 del_timer(&q
->unplug_timer
);
251 EXPORT_SYMBOL(blk_remove_plug
);
254 * remove the plug and let it rip..
256 void __generic_unplug_device(struct request_queue
*q
)
258 if (unlikely(blk_queue_stopped(q
)))
261 if (!blk_remove_plug(q
))
266 EXPORT_SYMBOL(__generic_unplug_device
);
269 * generic_unplug_device - fire a request queue
270 * @q: The &struct request_queue in question
273 * Linux uses plugging to build bigger requests queues before letting
274 * the device have at them. If a queue is plugged, the I/O scheduler
275 * is still adding and merging requests on the queue. Once the queue
276 * gets unplugged, the request_fn defined for the queue is invoked and
279 void generic_unplug_device(struct request_queue
*q
)
281 spin_lock_irq(q
->queue_lock
);
282 __generic_unplug_device(q
);
283 spin_unlock_irq(q
->queue_lock
);
285 EXPORT_SYMBOL(generic_unplug_device
);
287 static void blk_backing_dev_unplug(struct backing_dev_info
*bdi
,
290 struct request_queue
*q
= bdi
->unplug_io_data
;
295 void blk_unplug_work(struct work_struct
*work
)
297 struct request_queue
*q
=
298 container_of(work
, struct request_queue
, unplug_work
);
300 blk_add_trace_pdu_int(q
, BLK_TA_UNPLUG_IO
, NULL
,
301 q
->rq
.count
[READ
] + q
->rq
.count
[WRITE
]);
306 void blk_unplug_timeout(unsigned long data
)
308 struct request_queue
*q
= (struct request_queue
*)data
;
310 blk_add_trace_pdu_int(q
, BLK_TA_UNPLUG_TIMER
, NULL
,
311 q
->rq
.count
[READ
] + q
->rq
.count
[WRITE
]);
313 kblockd_schedule_work(&q
->unplug_work
);
316 void blk_unplug(struct request_queue
*q
)
319 * devices don't necessarily have an ->unplug_fn defined
322 blk_add_trace_pdu_int(q
, BLK_TA_UNPLUG_IO
, NULL
,
323 q
->rq
.count
[READ
] + q
->rq
.count
[WRITE
]);
328 EXPORT_SYMBOL(blk_unplug
);
331 * blk_start_queue - restart a previously stopped queue
332 * @q: The &struct request_queue in question
335 * blk_start_queue() will clear the stop flag on the queue, and call
336 * the request_fn for the queue if it was in a stopped state when
337 * entered. Also see blk_stop_queue(). Queue lock must be held.
339 void blk_start_queue(struct request_queue
*q
)
341 WARN_ON(!irqs_disabled());
343 clear_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
);
346 * one level of recursion is ok and is much faster than kicking
347 * the unplug handling
349 if (!test_and_set_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
)) {
351 clear_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
);
354 kblockd_schedule_work(&q
->unplug_work
);
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 set_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
);
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
)
409 spin_lock_irqsave(q
->queue_lock
, flags
);
413 * Only recurse once to avoid overrunning the stack, let the unplug
414 * handling reinvoke the handler shortly if we already got there.
416 if (!elv_queue_empty(q
)) {
417 if (!test_and_set_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
)) {
419 clear_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
);
422 kblockd_schedule_work(&q
->unplug_work
);
426 spin_unlock_irqrestore(q
->queue_lock
, flags
);
428 EXPORT_SYMBOL(blk_run_queue
);
430 void blk_put_queue(struct request_queue
*q
)
432 kobject_put(&q
->kobj
);
434 <<<<<<< HEAD
:block
/blk
-core
.c
435 EXPORT_SYMBOL(blk_put_queue
);
437 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a
:block
/blk
-core
.c
439 void blk_cleanup_queue(struct request_queue
*q
)
441 mutex_lock(&q
->sysfs_lock
);
442 set_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
);
443 mutex_unlock(&q
->sysfs_lock
);
446 elevator_exit(q
->elevator
);
450 EXPORT_SYMBOL(blk_cleanup_queue
);
452 static int blk_init_free_list(struct request_queue
*q
)
454 struct request_list
*rl
= &q
->rq
;
456 rl
->count
[READ
] = rl
->count
[WRITE
] = 0;
457 rl
->starved
[READ
] = rl
->starved
[WRITE
] = 0;
459 init_waitqueue_head(&rl
->wait
[READ
]);
460 init_waitqueue_head(&rl
->wait
[WRITE
]);
462 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, mempool_alloc_slab
,
463 mempool_free_slab
, request_cachep
, q
->node
);
471 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
473 return blk_alloc_queue_node(gfp_mask
, -1);
475 EXPORT_SYMBOL(blk_alloc_queue
);
477 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
479 struct request_queue
*q
;
482 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
483 gfp_mask
| __GFP_ZERO
, node_id
);
487 q
->backing_dev_info
.unplug_io_fn
= blk_backing_dev_unplug
;
488 q
->backing_dev_info
.unplug_io_data
= q
;
489 err
= bdi_init(&q
->backing_dev_info
);
491 kmem_cache_free(blk_requestq_cachep
, q
);
495 init_timer(&q
->unplug_timer
);
497 kobject_init(&q
->kobj
, &blk_queue_ktype
);
499 mutex_init(&q
->sysfs_lock
);
503 EXPORT_SYMBOL(blk_alloc_queue_node
);
506 * blk_init_queue - prepare a request queue for use with a block device
507 * @rfn: The function to be called to process requests that have been
508 * placed on the queue.
509 * @lock: Request queue spin lock
512 * If a block device wishes to use the standard request handling procedures,
513 * which sorts requests and coalesces adjacent requests, then it must
514 * call blk_init_queue(). The function @rfn will be called when there
515 * are requests on the queue that need to be processed. If the device
516 * supports plugging, then @rfn may not be called immediately when requests
517 * are available on the queue, but may be called at some time later instead.
518 * Plugged queues are generally unplugged when a buffer belonging to one
519 * of the requests on the queue is needed, or due to memory pressure.
521 * @rfn is not required, or even expected, to remove all requests off the
522 * queue, but only as many as it can handle at a time. If it does leave
523 * requests on the queue, it is responsible for arranging that the requests
524 * get dealt with eventually.
526 * The queue spin lock must be held while manipulating the requests on the
527 * request queue; this lock will be taken also from interrupt context, so irq
528 * disabling is needed for it.
530 * Function returns a pointer to the initialized request queue, or NULL if
534 * blk_init_queue() must be paired with a blk_cleanup_queue() call
535 * when the block device is deactivated (such as at module unload).
538 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
540 return blk_init_queue_node(rfn
, lock
, -1);
542 EXPORT_SYMBOL(blk_init_queue
);
544 struct request_queue
*
545 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
547 struct request_queue
*q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
553 if (blk_init_free_list(q
)) {
554 kmem_cache_free(blk_requestq_cachep
, q
);
559 * if caller didn't supply a lock, they get per-queue locking with
563 spin_lock_init(&q
->__queue_lock
);
564 lock
= &q
->__queue_lock
;
568 q
->prep_rq_fn
= NULL
;
569 q
->unplug_fn
= generic_unplug_device
;
570 q
->queue_flags
= (1 << QUEUE_FLAG_CLUSTER
);
571 q
->queue_lock
= lock
;
573 blk_queue_segment_boundary(q
, 0xffffffff);
575 blk_queue_make_request(q
, __make_request
);
576 blk_queue_max_segment_size(q
, MAX_SEGMENT_SIZE
);
578 blk_queue_max_hw_segments(q
, MAX_HW_SEGMENTS
);
579 blk_queue_max_phys_segments(q
, MAX_PHYS_SEGMENTS
);
581 q
->sg_reserved_size
= INT_MAX
;
586 if (!elevator_init(q
, NULL
)) {
587 blk_queue_congestion_threshold(q
);
594 EXPORT_SYMBOL(blk_init_queue_node
);
596 int blk_get_queue(struct request_queue
*q
)
598 if (likely(!test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
))) {
599 kobject_get(&q
->kobj
);
605 <<<<<<< HEAD
:block
/blk
-core
.c
606 EXPORT_SYMBOL(blk_get_queue
);
608 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a
:block
/blk
-core
.c
610 static inline void blk_free_request(struct request_queue
*q
, struct request
*rq
)
612 if (rq
->cmd_flags
& REQ_ELVPRIV
)
613 elv_put_request(q
, rq
);
614 mempool_free(rq
, q
->rq
.rq_pool
);
617 static struct request
*
618 blk_alloc_request(struct request_queue
*q
, int rw
, int priv
, gfp_t gfp_mask
)
620 struct request
*rq
= mempool_alloc(q
->rq
.rq_pool
, gfp_mask
);
626 * first three bits are identical in rq->cmd_flags and bio->bi_rw,
627 * see bio.h and blkdev.h
629 rq
->cmd_flags
= rw
| REQ_ALLOCED
;
632 if (unlikely(elv_set_request(q
, rq
, gfp_mask
))) {
633 mempool_free(rq
, q
->rq
.rq_pool
);
636 rq
->cmd_flags
|= REQ_ELVPRIV
;
643 * ioc_batching returns true if the ioc is a valid batching request and
644 * should be given priority access to a request.
646 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
652 * Make sure the process is able to allocate at least 1 request
653 * even if the batch times out, otherwise we could theoretically
656 return ioc
->nr_batch_requests
== q
->nr_batching
||
657 (ioc
->nr_batch_requests
> 0
658 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
662 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
663 * will cause the process to be a "batcher" on all queues in the system. This
664 * is the behaviour we want though - once it gets a wakeup it should be given
667 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
669 if (!ioc
|| ioc_batching(q
, ioc
))
672 ioc
->nr_batch_requests
= q
->nr_batching
;
673 ioc
->last_waited
= jiffies
;
676 static void __freed_request(struct request_queue
*q
, int rw
)
678 struct request_list
*rl
= &q
->rq
;
680 if (rl
->count
[rw
] < queue_congestion_off_threshold(q
))
681 blk_clear_queue_congested(q
, rw
);
683 if (rl
->count
[rw
] + 1 <= q
->nr_requests
) {
684 if (waitqueue_active(&rl
->wait
[rw
]))
685 wake_up(&rl
->wait
[rw
]);
687 blk_clear_queue_full(q
, rw
);
692 * A request has just been released. Account for it, update the full and
693 * congestion status, wake up any waiters. Called under q->queue_lock.
695 static void freed_request(struct request_queue
*q
, int rw
, int priv
)
697 struct request_list
*rl
= &q
->rq
;
703 __freed_request(q
, rw
);
705 if (unlikely(rl
->starved
[rw
^ 1]))
706 __freed_request(q
, rw
^ 1);
709 #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
711 * Get a free request, queue_lock must be held.
712 * Returns NULL on failure, with queue_lock held.
713 * Returns !NULL on success, with queue_lock *not held*.
715 static struct request
*get_request(struct request_queue
*q
, int rw_flags
,
716 struct bio
*bio
, gfp_t gfp_mask
)
718 struct request
*rq
= NULL
;
719 struct request_list
*rl
= &q
->rq
;
720 struct io_context
*ioc
= NULL
;
721 const int rw
= rw_flags
& 0x01;
724 may_queue
= elv_may_queue(q
, rw_flags
);
725 if (may_queue
== ELV_MQUEUE_NO
)
728 if (rl
->count
[rw
]+1 >= queue_congestion_on_threshold(q
)) {
729 if (rl
->count
[rw
]+1 >= q
->nr_requests
) {
730 ioc
= current_io_context(GFP_ATOMIC
, q
->node
);
732 * The queue will fill after this allocation, so set
733 * it as full, and mark this process as "batching".
734 * This process will be allowed to complete a batch of
735 * requests, others will be blocked.
737 if (!blk_queue_full(q
, rw
)) {
738 ioc_set_batching(q
, ioc
);
739 blk_set_queue_full(q
, rw
);
741 if (may_queue
!= ELV_MQUEUE_MUST
742 && !ioc_batching(q
, ioc
)) {
744 * The queue is full and the allocating
745 * process is not a "batcher", and not
746 * exempted by the IO scheduler
752 blk_set_queue_congested(q
, rw
);
756 * Only allow batching queuers to allocate up to 50% over the defined
757 * limit of requests, otherwise we could have thousands of requests
758 * allocated with any setting of ->nr_requests
760 if (rl
->count
[rw
] >= (3 * q
->nr_requests
/ 2))
766 priv
= !test_bit(QUEUE_FLAG_ELVSWITCH
, &q
->queue_flags
);
770 spin_unlock_irq(q
->queue_lock
);
772 rq
= blk_alloc_request(q
, rw_flags
, priv
, gfp_mask
);
775 * Allocation failed presumably due to memory. Undo anything
776 * we might have messed up.
778 * Allocating task should really be put onto the front of the
779 * wait queue, but this is pretty rare.
781 spin_lock_irq(q
->queue_lock
);
782 freed_request(q
, rw
, priv
);
785 * in the very unlikely event that allocation failed and no
786 * requests for this direction was pending, mark us starved
787 * so that freeing of a request in the other direction will
788 * notice us. another possible fix would be to split the
789 * rq mempool into READ and WRITE
792 if (unlikely(rl
->count
[rw
] == 0))
799 * ioc may be NULL here, and ioc_batching will be false. That's
800 * OK, if the queue is under the request limit then requests need
801 * not count toward the nr_batch_requests limit. There will always
802 * be some limit enforced by BLK_BATCH_TIME.
804 if (ioc_batching(q
, ioc
))
805 ioc
->nr_batch_requests
--;
809 blk_add_trace_generic(q
, bio
, rw
, BLK_TA_GETRQ
);
815 * No available requests for this queue, unplug the device and wait for some
816 * requests to become available.
818 * Called with q->queue_lock held, and returns with it unlocked.
820 static struct request
*get_request_wait(struct request_queue
*q
, int rw_flags
,
823 const int rw
= rw_flags
& 0x01;
826 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
829 struct request_list
*rl
= &q
->rq
;
831 prepare_to_wait_exclusive(&rl
->wait
[rw
], &wait
,
832 TASK_UNINTERRUPTIBLE
);
834 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
837 struct io_context
*ioc
;
839 blk_add_trace_generic(q
, bio
, rw
, BLK_TA_SLEEPRQ
);
841 __generic_unplug_device(q
);
842 spin_unlock_irq(q
->queue_lock
);
846 * After sleeping, we become a "batching" process and
847 * will be able to allocate at least one request, and
848 * up to a big batch of them for a small period time.
849 * See ioc_batching, ioc_set_batching
851 ioc
= current_io_context(GFP_NOIO
, q
->node
);
852 ioc_set_batching(q
, ioc
);
854 spin_lock_irq(q
->queue_lock
);
856 finish_wait(&rl
->wait
[rw
], &wait
);
862 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
866 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
868 spin_lock_irq(q
->queue_lock
);
869 if (gfp_mask
& __GFP_WAIT
) {
870 rq
= get_request_wait(q
, rw
, NULL
);
872 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
874 spin_unlock_irq(q
->queue_lock
);
876 /* q->queue_lock is unlocked at this point */
880 EXPORT_SYMBOL(blk_get_request
);
883 * blk_start_queueing - initiate dispatch of requests to device
884 * @q: request queue to kick into gear
886 * This is basically a helper to remove the need to know whether a queue
887 * is plugged or not if someone just wants to initiate dispatch of requests
890 * The queue lock must be held with interrupts disabled.
892 void blk_start_queueing(struct request_queue
*q
)
894 if (!blk_queue_plugged(q
))
897 __generic_unplug_device(q
);
899 EXPORT_SYMBOL(blk_start_queueing
);
902 * blk_requeue_request - put a request back on queue
903 * @q: request queue where request should be inserted
904 * @rq: request to be inserted
907 * Drivers often keep queueing requests until the hardware cannot accept
908 * more, when that condition happens we need to put the request back
909 * on the queue. Must be called with queue lock held.
911 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
913 blk_add_trace_rq(q
, rq
, BLK_TA_REQUEUE
);
915 if (blk_rq_tagged(rq
))
916 blk_queue_end_tag(q
, rq
);
918 elv_requeue_request(q
, rq
);
920 EXPORT_SYMBOL(blk_requeue_request
);
923 * blk_insert_request - insert a special request in to a request queue
924 * @q: request queue where request should be inserted
925 * @rq: request to be inserted
926 * @at_head: insert request at head or tail of queue
927 * @data: private data
930 * Many block devices need to execute commands asynchronously, so they don't
931 * block the whole kernel from preemption during request execution. This is
932 * accomplished normally by inserting aritficial requests tagged as
933 * REQ_SPECIAL in to the corresponding request queue, and letting them be
934 * scheduled for actual execution by the request queue.
936 * We have the option of inserting the head or the tail of the queue.
937 * Typically we use the tail for new ioctls and so forth. We use the head
938 * of the queue for things like a QUEUE_FULL message from a device, or a
939 * host that is unable to accept a particular command.
941 void blk_insert_request(struct request_queue
*q
, struct request
*rq
,
942 int at_head
, void *data
)
944 int where
= at_head
? ELEVATOR_INSERT_FRONT
: ELEVATOR_INSERT_BACK
;
948 * tell I/O scheduler that this isn't a regular read/write (ie it
949 * must not attempt merges on this) and that it acts as a soft
952 rq
->cmd_type
= REQ_TYPE_SPECIAL
;
953 rq
->cmd_flags
|= REQ_SOFTBARRIER
;
957 spin_lock_irqsave(q
->queue_lock
, flags
);
960 * If command is tagged, release the tag
962 if (blk_rq_tagged(rq
))
963 blk_queue_end_tag(q
, rq
);
965 drive_stat_acct(rq
, 1);
966 __elv_add_request(q
, rq
, where
, 0);
967 blk_start_queueing(q
);
968 spin_unlock_irqrestore(q
->queue_lock
, flags
);
970 EXPORT_SYMBOL(blk_insert_request
);
973 * add-request adds a request to the linked list.
974 * queue lock is held and interrupts disabled, as we muck with the
975 * request queue list.
977 static inline void add_request(struct request_queue
*q
, struct request
*req
)
979 drive_stat_acct(req
, 1);
982 * elevator indicated where it wants this request to be
983 * inserted at elevator_merge time
985 __elv_add_request(q
, req
, ELEVATOR_INSERT_SORT
, 0);
989 * disk_round_stats() - Round off the performance stats on a struct
992 * The average IO queue length and utilisation statistics are maintained
993 * by observing the current state of the queue length and the amount of
994 * time it has been in this state for.
996 * Normally, that accounting is done on IO completion, but that can result
997 * in more than a second's worth of IO being accounted for within any one
998 * second, leading to >100% utilisation. To deal with that, we call this
999 * function to do a round-off before returning the results when reading
1000 * /proc/diskstats. This accounts immediately for all queue usage up to
1001 * the current jiffies and restarts the counters again.
1003 void disk_round_stats(struct gendisk
*disk
)
1005 unsigned long now
= jiffies
;
1007 if (now
== disk
->stamp
)
1010 if (disk
->in_flight
) {
1011 __disk_stat_add(disk
, time_in_queue
,
1012 disk
->in_flight
* (now
- disk
->stamp
));
1013 __disk_stat_add(disk
, io_ticks
, (now
- disk
->stamp
));
1017 EXPORT_SYMBOL_GPL(disk_round_stats
);
1019 void part_round_stats(struct hd_struct
*part
)
1021 unsigned long now
= jiffies
;
1023 if (now
== part
->stamp
)
1026 if (part
->in_flight
) {
1027 __part_stat_add(part
, time_in_queue
,
1028 part
->in_flight
* (now
- part
->stamp
));
1029 __part_stat_add(part
, io_ticks
, (now
- part
->stamp
));
1035 * queue lock must be held
1037 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1041 if (unlikely(--req
->ref_count
))
1044 elv_completed_request(q
, req
);
1047 * Request may not have originated from ll_rw_blk. if not,
1048 * it didn't come out of our reserved rq pools
1050 if (req
->cmd_flags
& REQ_ALLOCED
) {
1051 int rw
= rq_data_dir(req
);
1052 int priv
= req
->cmd_flags
& REQ_ELVPRIV
;
1054 BUG_ON(!list_empty(&req
->queuelist
));
1055 BUG_ON(!hlist_unhashed(&req
->hash
));
1057 blk_free_request(q
, req
);
1058 freed_request(q
, rw
, priv
);
1061 EXPORT_SYMBOL_GPL(__blk_put_request
);
1063 void blk_put_request(struct request
*req
)
1065 unsigned long flags
;
1066 struct request_queue
*q
= req
->q
;
1069 * Gee, IDE calls in w/ NULL q. Fix IDE and remove the
1070 * following if (q) test.
1073 spin_lock_irqsave(q
->queue_lock
, flags
);
1074 __blk_put_request(q
, req
);
1075 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1078 EXPORT_SYMBOL(blk_put_request
);
1080 void init_request_from_bio(struct request
*req
, struct bio
*bio
)
1082 req
->cmd_type
= REQ_TYPE_FS
;
1085 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
1087 if (bio_rw_ahead(bio
) || bio_failfast(bio
))
1088 req
->cmd_flags
|= REQ_FAILFAST
;
1091 * REQ_BARRIER implies no merging, but lets make it explicit
1093 if (unlikely(bio_barrier(bio
)))
1094 req
->cmd_flags
|= (REQ_HARDBARRIER
| REQ_NOMERGE
);
1097 req
->cmd_flags
|= REQ_RW_SYNC
;
1098 if (bio_rw_meta(bio
))
1099 req
->cmd_flags
|= REQ_RW_META
;
1102 req
->hard_sector
= req
->sector
= bio
->bi_sector
;
1103 req
->ioprio
= bio_prio(bio
);
1104 req
->start_time
= jiffies
;
1105 blk_rq_bio_prep(req
->q
, req
, bio
);
1108 static int __make_request(struct request_queue
*q
, struct bio
*bio
)
1110 struct request
*req
;
1111 int el_ret
, nr_sectors
, barrier
, err
;
1112 const unsigned short prio
= bio_prio(bio
);
1113 const int sync
= bio_sync(bio
);
1116 nr_sectors
= bio_sectors(bio
);
1119 * low level driver can indicate that it wants pages above a
1120 * certain limit bounced to low memory (ie for highmem, or even
1121 * ISA dma in theory)
1123 blk_queue_bounce(q
, &bio
);
1125 barrier
= bio_barrier(bio
);
1126 if (unlikely(barrier
) && (q
->next_ordered
== QUEUE_ORDERED_NONE
)) {
1131 spin_lock_irq(q
->queue_lock
);
1133 if (unlikely(barrier
) || elv_queue_empty(q
))
1136 el_ret
= elv_merge(q
, &req
, bio
);
1138 case ELEVATOR_BACK_MERGE
:
1139 BUG_ON(!rq_mergeable(req
));
1141 if (!ll_back_merge_fn(q
, req
, bio
))
1144 blk_add_trace_bio(q
, bio
, BLK_TA_BACKMERGE
);
1146 req
->biotail
->bi_next
= bio
;
1148 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
1149 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
1150 drive_stat_acct(req
, 0);
1151 if (!attempt_back_merge(q
, req
))
1152 elv_merged_request(q
, req
, el_ret
);
1155 case ELEVATOR_FRONT_MERGE
:
1156 BUG_ON(!rq_mergeable(req
));
1158 if (!ll_front_merge_fn(q
, req
, bio
))
1161 blk_add_trace_bio(q
, bio
, BLK_TA_FRONTMERGE
);
1163 bio
->bi_next
= req
->bio
;
1167 * may not be valid. if the low level driver said
1168 * it didn't need a bounce buffer then it better
1169 * not touch req->buffer either...
1171 req
->buffer
= bio_data(bio
);
1172 req
->current_nr_sectors
= bio_cur_sectors(bio
);
1173 req
->hard_cur_sectors
= req
->current_nr_sectors
;
1174 req
->sector
= req
->hard_sector
= bio
->bi_sector
;
1175 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
1176 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
1177 drive_stat_acct(req
, 0);
1178 if (!attempt_front_merge(q
, req
))
1179 elv_merged_request(q
, req
, el_ret
);
1182 /* ELV_NO_MERGE: elevator says don't/can't merge. */
1189 * This sync check and mask will be re-done in init_request_from_bio(),
1190 * but we need to set it earlier to expose the sync flag to the
1191 * rq allocator and io schedulers.
1193 rw_flags
= bio_data_dir(bio
);
1195 rw_flags
|= REQ_RW_SYNC
;
1198 * Grab a free request. This is might sleep but can not fail.
1199 * Returns with the queue unlocked.
1201 req
= get_request_wait(q
, rw_flags
, bio
);
1204 * After dropping the lock and possibly sleeping here, our request
1205 * may now be mergeable after it had proven unmergeable (above).
1206 * We don't worry about that case for efficiency. It won't happen
1207 * often, and the elevators are able to handle it.
1209 init_request_from_bio(req
, bio
);
1211 spin_lock_irq(q
->queue_lock
);
1212 if (elv_queue_empty(q
))
1214 add_request(q
, req
);
1217 __generic_unplug_device(q
);
1219 spin_unlock_irq(q
->queue_lock
);
1223 bio_endio(bio
, err
);
1228 * If bio->bi_dev is a partition, remap the location
1230 static inline void blk_partition_remap(struct bio
*bio
)
1232 struct block_device
*bdev
= bio
->bi_bdev
;
1234 if (bio_sectors(bio
) && bdev
!= bdev
->bd_contains
) {
1235 struct hd_struct
*p
= bdev
->bd_part
;
1237 bio
->bi_sector
+= p
->start_sect
;
1238 bio
->bi_bdev
= bdev
->bd_contains
;
1240 blk_add_trace_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
1241 bdev
->bd_dev
, bio
->bi_sector
,
1242 bio
->bi_sector
- p
->start_sect
);
1246 static void handle_bad_sector(struct bio
*bio
)
1248 char b
[BDEVNAME_SIZE
];
1250 printk(KERN_INFO
"attempt to access beyond end of device\n");
1251 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
1252 bdevname(bio
->bi_bdev
, b
),
1254 (unsigned long long)bio
->bi_sector
+ bio_sectors(bio
),
1255 (long long)(bio
->bi_bdev
->bd_inode
->i_size
>> 9));
1257 set_bit(BIO_EOF
, &bio
->bi_flags
);
1260 #ifdef CONFIG_FAIL_MAKE_REQUEST
1262 static DECLARE_FAULT_ATTR(fail_make_request
);
1264 static int __init
setup_fail_make_request(char *str
)
1266 return setup_fault_attr(&fail_make_request
, str
);
1268 __setup("fail_make_request=", setup_fail_make_request
);
1270 static int should_fail_request(struct bio
*bio
)
1272 if ((bio
->bi_bdev
->bd_disk
->flags
& GENHD_FL_FAIL
) ||
1273 (bio
->bi_bdev
->bd_part
&& bio
->bi_bdev
->bd_part
->make_it_fail
))
1274 return should_fail(&fail_make_request
, bio
->bi_size
);
1279 static int __init
fail_make_request_debugfs(void)
1281 return init_fault_attr_dentries(&fail_make_request
,
1282 "fail_make_request");
1285 late_initcall(fail_make_request_debugfs
);
1287 #else /* CONFIG_FAIL_MAKE_REQUEST */
1289 static inline int should_fail_request(struct bio
*bio
)
1294 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1297 * Check whether this bio extends beyond the end of the device.
1299 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
1306 /* Test device or partition size, when known. */
1307 maxsector
= bio
->bi_bdev
->bd_inode
->i_size
>> 9;
1309 sector_t sector
= bio
->bi_sector
;
1311 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
1313 * This may well happen - the kernel calls bread()
1314 * without checking the size of the device, e.g., when
1315 * mounting a device.
1317 handle_bad_sector(bio
);
1326 * generic_make_request: hand a buffer to its device driver for I/O
1327 * @bio: The bio describing the location in memory and on the device.
1329 * generic_make_request() is used to make I/O requests of block
1330 * devices. It is passed a &struct bio, which describes the I/O that needs
1333 * generic_make_request() does not return any status. The
1334 * success/failure status of the request, along with notification of
1335 * completion, is delivered asynchronously through the bio->bi_end_io
1336 * function described (one day) else where.
1338 * The caller of generic_make_request must make sure that bi_io_vec
1339 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1340 * set to describe the device address, and the
1341 * bi_end_io and optionally bi_private are set to describe how
1342 * completion notification should be signaled.
1344 * generic_make_request and the drivers it calls may use bi_next if this
1345 * bio happens to be merged with someone else, and may change bi_dev and
1346 * bi_sector for remaps as it sees fit. So the values of these fields
1347 * should NOT be depended on after the call to generic_make_request.
1349 static inline void __generic_make_request(struct bio
*bio
)
1351 struct request_queue
*q
;
1352 sector_t old_sector
;
1353 int ret
, nr_sectors
= bio_sectors(bio
);
1359 if (bio_check_eod(bio
, nr_sectors
))
1363 * Resolve the mapping until finished. (drivers are
1364 * still free to implement/resolve their own stacking
1365 * by explicitly returning 0)
1367 * NOTE: we don't repeat the blk_size check for each new device.
1368 * Stacking drivers are expected to know what they are doing.
1373 char b
[BDEVNAME_SIZE
];
1375 q
= bdev_get_queue(bio
->bi_bdev
);
1378 "generic_make_request: Trying to access "
1379 "nonexistent block-device %s (%Lu)\n",
1380 bdevname(bio
->bi_bdev
, b
),
1381 (long long) bio
->bi_sector
);
1383 bio_endio(bio
, err
);
1387 if (unlikely(nr_sectors
> q
->max_hw_sectors
)) {
1388 printk(KERN_ERR
"bio too big device %s (%u > %u)\n",
1389 bdevname(bio
->bi_bdev
, b
),
1395 if (unlikely(test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)))
1398 if (should_fail_request(bio
))
1402 * If this device has partitions, remap block n
1403 * of partition p to block n+start(p) of the disk.
1405 blk_partition_remap(bio
);
1407 if (old_sector
!= -1)
1408 blk_add_trace_remap(q
, bio
, old_dev
, bio
->bi_sector
,
1411 blk_add_trace_bio(q
, bio
, BLK_TA_QUEUE
);
1413 old_sector
= bio
->bi_sector
;
1414 old_dev
= bio
->bi_bdev
->bd_dev
;
1416 if (bio_check_eod(bio
, nr_sectors
))
1418 if (bio_empty_barrier(bio
) && !q
->prepare_flush_fn
) {
1423 ret
= q
->make_request_fn(q
, bio
);
1428 * We only want one ->make_request_fn to be active at a time,
1429 * else stack usage with stacked devices could be a problem.
1430 * So use current->bio_{list,tail} to keep a list of requests
1431 * submited by a make_request_fn function.
1432 * current->bio_tail is also used as a flag to say if
1433 * generic_make_request is currently active in this task or not.
1434 * If it is NULL, then no make_request is active. If it is non-NULL,
1435 * then a make_request is active, and new requests should be added
1438 void generic_make_request(struct bio
*bio
)
1440 if (current
->bio_tail
) {
1441 /* make_request is active */
1442 *(current
->bio_tail
) = bio
;
1443 bio
->bi_next
= NULL
;
1444 current
->bio_tail
= &bio
->bi_next
;
1447 /* following loop may be a bit non-obvious, and so deserves some
1449 * Before entering the loop, bio->bi_next is NULL (as all callers
1450 * ensure that) so we have a list with a single bio.
1451 * We pretend that we have just taken it off a longer list, so
1452 * we assign bio_list to the next (which is NULL) and bio_tail
1453 * to &bio_list, thus initialising the bio_list of new bios to be
1454 * added. __generic_make_request may indeed add some more bios
1455 * through a recursive call to generic_make_request. If it
1456 * did, we find a non-NULL value in bio_list and re-enter the loop
1457 * from the top. In this case we really did just take the bio
1458 * of the top of the list (no pretending) and so fixup bio_list and
1459 * bio_tail or bi_next, and call into __generic_make_request again.
1461 * The loop was structured like this to make only one call to
1462 * __generic_make_request (which is important as it is large and
1463 * inlined) and to keep the structure simple.
1465 BUG_ON(bio
->bi_next
);
1467 current
->bio_list
= bio
->bi_next
;
1468 if (bio
->bi_next
== NULL
)
1469 current
->bio_tail
= ¤t
->bio_list
;
1471 bio
->bi_next
= NULL
;
1472 __generic_make_request(bio
);
1473 bio
= current
->bio_list
;
1475 current
->bio_tail
= NULL
; /* deactivate */
1477 EXPORT_SYMBOL(generic_make_request
);
1480 * submit_bio: submit a bio to the block device layer for I/O
1481 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1482 * @bio: The &struct bio which describes the I/O
1484 * submit_bio() is very similar in purpose to generic_make_request(), and
1485 * uses that function to do most of the work. Both are fairly rough
1486 * interfaces, @bio must be presetup and ready for I/O.
1489 void submit_bio(int rw
, struct bio
*bio
)
1491 int count
= bio_sectors(bio
);
1496 * If it's a regular read/write or a barrier with data attached,
1497 * go through the normal accounting stuff before submission.
1499 if (!bio_empty_barrier(bio
)) {
1501 BIO_BUG_ON(!bio
->bi_size
);
1502 BIO_BUG_ON(!bio
->bi_io_vec
);
1505 count_vm_events(PGPGOUT
, count
);
1507 task_io_account_read(bio
->bi_size
);
1508 count_vm_events(PGPGIN
, count
);
1511 if (unlikely(block_dump
)) {
1512 char b
[BDEVNAME_SIZE
];
1513 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s\n",
1514 current
->comm
, task_pid_nr(current
),
1515 (rw
& WRITE
) ? "WRITE" : "READ",
1516 (unsigned long long)bio
->bi_sector
,
1517 bdevname(bio
->bi_bdev
, b
));
1521 generic_make_request(bio
);
1523 EXPORT_SYMBOL(submit_bio
);
1526 * __end_that_request_first - end I/O on a request
1527 * @req: the request being processed
1528 * @error: 0 for success, < 0 for error
1529 * @nr_bytes: number of bytes to complete
1532 * Ends I/O on a number of bytes attached to @req, and sets it up
1533 * for the next range of segments (if any) in the cluster.
1536 * 0 - we are done with this request, call end_that_request_last()
1537 * 1 - still buffers pending for this request
1539 static int __end_that_request_first(struct request
*req
, int error
,
1542 int total_bytes
, bio_nbytes
, next_idx
= 0;
1545 blk_add_trace_rq(req
->q
, req
, BLK_TA_COMPLETE
);
1548 * for a REQ_BLOCK_PC request, we want to carry any eventual
1549 * sense key with us all the way through
1551 if (!blk_pc_request(req
))
1554 if (error
&& (blk_fs_request(req
) && !(req
->cmd_flags
& REQ_QUIET
))) {
1555 printk(KERN_ERR
"end_request: I/O error, dev %s, sector %llu\n",
1556 req
->rq_disk
? req
->rq_disk
->disk_name
: "?",
1557 (unsigned long long)req
->sector
);
1560 if (blk_fs_request(req
) && req
->rq_disk
) {
1561 const int rw
= rq_data_dir(req
);
1563 all_stat_add(req
->rq_disk
, sectors
[rw
],
1564 nr_bytes
>> 9, req
->sector
);
1567 total_bytes
= bio_nbytes
= 0;
1568 while ((bio
= req
->bio
) != NULL
) {
1572 * For an empty barrier request, the low level driver must
1573 * store a potential error location in ->sector. We pass
1574 * that back up in ->bi_sector.
1576 if (blk_empty_barrier(req
))
1577 bio
->bi_sector
= req
->sector
;
1579 if (nr_bytes
>= bio
->bi_size
) {
1580 req
->bio
= bio
->bi_next
;
1581 nbytes
= bio
->bi_size
;
1582 req_bio_endio(req
, bio
, nbytes
, error
);
1586 int idx
= bio
->bi_idx
+ next_idx
;
1588 if (unlikely(bio
->bi_idx
>= bio
->bi_vcnt
)) {
1589 blk_dump_rq_flags(req
, "__end_that");
1590 printk(KERN_ERR
"%s: bio idx %d >= vcnt %d\n",
1591 __FUNCTION__
, bio
->bi_idx
,
1596 nbytes
= bio_iovec_idx(bio
, idx
)->bv_len
;
1597 BIO_BUG_ON(nbytes
> bio
->bi_size
);
1600 * not a complete bvec done
1602 if (unlikely(nbytes
> nr_bytes
)) {
1603 bio_nbytes
+= nr_bytes
;
1604 total_bytes
+= nr_bytes
;
1609 * advance to the next vector
1612 bio_nbytes
+= nbytes
;
1615 total_bytes
+= nbytes
;
1621 * end more in this run, or just return 'not-done'
1623 if (unlikely(nr_bytes
<= 0))
1635 * if the request wasn't completed, update state
1638 req_bio_endio(req
, bio
, bio_nbytes
, error
);
1639 bio
->bi_idx
+= next_idx
;
1640 bio_iovec(bio
)->bv_offset
+= nr_bytes
;
1641 bio_iovec(bio
)->bv_len
-= nr_bytes
;
1644 blk_recalc_rq_sectors(req
, total_bytes
>> 9);
1645 blk_recalc_rq_segments(req
);
1650 * splice the completion data to a local structure and hand off to
1651 * process_completion_queue() to complete the requests
1653 static void blk_done_softirq(struct softirq_action
*h
)
1655 struct list_head
*cpu_list
, local_list
;
1657 local_irq_disable();
1658 cpu_list
= &__get_cpu_var(blk_cpu_done
);
1659 list_replace_init(cpu_list
, &local_list
);
1662 while (!list_empty(&local_list
)) {
1665 rq
= list_entry(local_list
.next
, struct request
, donelist
);
1666 list_del_init(&rq
->donelist
);
1667 rq
->q
->softirq_done_fn(rq
);
1671 static int __cpuinit
blk_cpu_notify(struct notifier_block
*self
,
1672 unsigned long action
, void *hcpu
)
1675 * If a CPU goes away, splice its entries to the current CPU
1676 * and trigger a run of the softirq
1678 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
1679 int cpu
= (unsigned long) hcpu
;
1681 local_irq_disable();
1682 list_splice_init(&per_cpu(blk_cpu_done
, cpu
),
1683 &__get_cpu_var(blk_cpu_done
));
1684 raise_softirq_irqoff(BLOCK_SOFTIRQ
);
1692 static struct notifier_block blk_cpu_notifier __cpuinitdata
= {
1693 .notifier_call
= blk_cpu_notify
,
1697 * blk_complete_request - end I/O on a request
1698 * @req: the request being processed
1701 * Ends all I/O on a request. It does not handle partial completions,
1702 * unless the driver actually implements this in its completion callback
1703 * through requeueing. The actual completion happens out-of-order,
1704 * through a softirq handler. The user must have registered a completion
1705 * callback through blk_queue_softirq_done().
1708 void blk_complete_request(struct request
*req
)
1710 struct list_head
*cpu_list
;
1711 unsigned long flags
;
1713 BUG_ON(!req
->q
->softirq_done_fn
);
1715 local_irq_save(flags
);
1717 cpu_list
= &__get_cpu_var(blk_cpu_done
);
1718 list_add_tail(&req
->donelist
, cpu_list
);
1719 raise_softirq_irqoff(BLOCK_SOFTIRQ
);
1721 local_irq_restore(flags
);
1723 EXPORT_SYMBOL(blk_complete_request
);
1726 * queue lock must be held
1728 static void end_that_request_last(struct request
*req
, int error
)
1730 struct gendisk
*disk
= req
->rq_disk
;
1732 if (blk_rq_tagged(req
))
1733 blk_queue_end_tag(req
->q
, req
);
1735 if (blk_queued_rq(req
))
1736 blkdev_dequeue_request(req
);
1738 if (unlikely(laptop_mode
) && blk_fs_request(req
))
1739 laptop_io_completion();
1742 * Account IO completion. bar_rq isn't accounted as a normal
1743 * IO on queueing nor completion. Accounting the containing
1744 * request is enough.
1746 if (disk
&& blk_fs_request(req
) && req
!= &req
->q
->bar_rq
) {
1747 unsigned long duration
= jiffies
- req
->start_time
;
1748 const int rw
= rq_data_dir(req
);
1749 struct hd_struct
*part
= get_part(disk
, req
->sector
);
1751 __all_stat_inc(disk
, ios
[rw
], req
->sector
);
1752 __all_stat_add(disk
, ticks
[rw
], duration
, req
->sector
);
1753 disk_round_stats(disk
);
1756 part_round_stats(part
);
1762 req
->end_io(req
, error
);
1764 if (blk_bidi_rq(req
))
1765 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
1767 __blk_put_request(req
->q
, req
);
1771 static inline void __end_request(struct request
*rq
, int uptodate
,
1772 unsigned int nr_bytes
)
1777 error
= uptodate
? uptodate
: -EIO
;
1779 __blk_end_request(rq
, error
, nr_bytes
);
1783 * blk_rq_bytes - Returns bytes left to complete in the entire request
1784 <<<<<<< HEAD:block/blk-core.c
1786 * @rq: the request being processed
1787 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a:block/blk-core.c
1789 unsigned int blk_rq_bytes(struct request
*rq
)
1791 if (blk_fs_request(rq
))
1792 return rq
->hard_nr_sectors
<< 9;
1794 return rq
->data_len
;
1796 EXPORT_SYMBOL_GPL(blk_rq_bytes
);
1799 * blk_rq_cur_bytes - Returns bytes left to complete in the current segment
1800 <<<<<<< HEAD:block/blk-core.c
1802 * @rq: the request being processed
1803 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a:block/blk-core.c
1805 unsigned int blk_rq_cur_bytes(struct request
*rq
)
1807 if (blk_fs_request(rq
))
1808 return rq
->current_nr_sectors
<< 9;
1811 return rq
->bio
->bi_size
;
1813 return rq
->data_len
;
1815 EXPORT_SYMBOL_GPL(blk_rq_cur_bytes
);
1818 * end_queued_request - end all I/O on a queued request
1819 * @rq: the request being processed
1820 * @uptodate: error value or 0/1 uptodate flag
1823 * Ends all I/O on a request, and removes it from the block layer queues.
1824 * Not suitable for normal IO completion, unless the driver still has
1825 * the request attached to the block layer.
1828 void end_queued_request(struct request
*rq
, int uptodate
)
1830 __end_request(rq
, uptodate
, blk_rq_bytes(rq
));
1832 EXPORT_SYMBOL(end_queued_request
);
1835 * end_dequeued_request - end all I/O on a dequeued request
1836 * @rq: the request being processed
1837 * @uptodate: error value or 0/1 uptodate flag
1840 * Ends all I/O on a request. The request must already have been
1841 * dequeued using blkdev_dequeue_request(), as is normally the case
1845 void end_dequeued_request(struct request
*rq
, int uptodate
)
1847 __end_request(rq
, uptodate
, blk_rq_bytes(rq
));
1849 EXPORT_SYMBOL(end_dequeued_request
);
1853 * end_request - end I/O on the current segment of the request
1854 * @req: the request being processed
1855 * @uptodate: error value or 0/1 uptodate flag
1858 * Ends I/O on the current segment of a request. If that is the only
1859 * remaining segment, the request is also completed and freed.
1861 * This is a remnant of how older block drivers handled IO completions.
1862 * Modern drivers typically end IO on the full request in one go, unless
1863 * they have a residual value to account for. For that case this function
1864 * isn't really useful, unless the residual just happens to be the
1865 * full current segment. In other words, don't use this function in new
1866 * code. Either use end_request_completely(), or the
1867 * end_that_request_chunk() (along with end_that_request_last()) for
1868 * partial completions.
1871 void end_request(struct request
*req
, int uptodate
)
1873 __end_request(req
, uptodate
, req
->hard_cur_sectors
<< 9);
1875 EXPORT_SYMBOL(end_request
);
1878 * blk_end_io - Generic end_io function to complete a request.
1879 * @rq: the request being processed
1880 * @error: 0 for success, < 0 for error
1881 * @nr_bytes: number of bytes to complete @rq
1882 * @bidi_bytes: number of bytes to complete @rq->next_rq
1883 * @drv_callback: function called between completion of bios in the request
1884 * and completion of the request.
1885 * If the callback returns non 0, this helper returns without
1886 * completion of the request.
1889 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
1890 * If @rq has leftover, sets it up for the next range of segments.
1893 * 0 - we are done with this request
1894 * 1 - this request is not freed yet, it still has pending buffers.
1896 static int blk_end_io(struct request
*rq
, int error
, unsigned int nr_bytes
,
1897 unsigned int bidi_bytes
,
1898 int (drv_callback
)(struct request
*))
1900 struct request_queue
*q
= rq
->q
;
1901 unsigned long flags
= 0UL;
1903 if (blk_fs_request(rq
) || blk_pc_request(rq
)) {
1904 if (__end_that_request_first(rq
, error
, nr_bytes
))
1907 /* Bidi request must be completed as a whole */
1908 if (blk_bidi_rq(rq
) &&
1909 __end_that_request_first(rq
->next_rq
, error
, bidi_bytes
))
1913 /* Special feature for tricky drivers */
1914 if (drv_callback
&& drv_callback(rq
))
1917 add_disk_randomness(rq
->rq_disk
);
1919 spin_lock_irqsave(q
->queue_lock
, flags
);
1920 end_that_request_last(rq
, error
);
1921 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1927 * blk_end_request - Helper function for drivers to complete the request.
1928 * @rq: the request being processed
1929 * @error: 0 for success, < 0 for error
1930 * @nr_bytes: number of bytes to complete
1933 * Ends I/O on a number of bytes attached to @rq.
1934 * If @rq has leftover, sets it up for the next range of segments.
1937 * 0 - we are done with this request
1938 * 1 - still buffers pending for this request
1940 int blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
1942 return blk_end_io(rq
, error
, nr_bytes
, 0, NULL
);
1944 EXPORT_SYMBOL_GPL(blk_end_request
);
1947 * __blk_end_request - Helper function for drivers to complete the request.
1948 * @rq: the request being processed
1949 * @error: 0 for success, < 0 for error
1950 * @nr_bytes: number of bytes to complete
1953 * Must be called with queue lock held unlike blk_end_request().
1956 * 0 - we are done with this request
1957 * 1 - still buffers pending for this request
1959 int __blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
1961 if (blk_fs_request(rq
) || blk_pc_request(rq
)) {
1962 if (__end_that_request_first(rq
, error
, nr_bytes
))
1966 add_disk_randomness(rq
->rq_disk
);
1968 end_that_request_last(rq
, error
);
1972 EXPORT_SYMBOL_GPL(__blk_end_request
);
1975 * blk_end_bidi_request - Helper function for drivers to complete bidi request.
1976 * @rq: the bidi request being processed
1977 * @error: 0 for success, < 0 for error
1978 * @nr_bytes: number of bytes to complete @rq
1979 * @bidi_bytes: number of bytes to complete @rq->next_rq
1982 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
1985 * 0 - we are done with this request
1986 * 1 - still buffers pending for this request
1988 int blk_end_bidi_request(struct request
*rq
, int error
, unsigned int nr_bytes
,
1989 unsigned int bidi_bytes
)
1991 return blk_end_io(rq
, error
, nr_bytes
, bidi_bytes
, NULL
);
1993 EXPORT_SYMBOL_GPL(blk_end_bidi_request
);
1996 * blk_end_request_callback - Special helper function for tricky drivers
1997 * @rq: the request being processed
1998 * @error: 0 for success, < 0 for error
1999 * @nr_bytes: number of bytes to complete
2000 * @drv_callback: function called between completion of bios in the request
2001 * and completion of the request.
2002 * If the callback returns non 0, this helper returns without
2003 * completion of the request.
2006 * Ends I/O on a number of bytes attached to @rq.
2007 * If @rq has leftover, sets it up for the next range of segments.
2009 * This special helper function is used only for existing tricky drivers.
2010 * (e.g. cdrom_newpc_intr() of ide-cd)
2011 * This interface will be removed when such drivers are rewritten.
2012 * Don't use this interface in other places anymore.
2015 * 0 - we are done with this request
2016 * 1 - this request is not freed yet.
2017 * this request still has pending buffers or
2018 * the driver doesn't want to finish this request yet.
2020 int blk_end_request_callback(struct request
*rq
, int error
,
2021 unsigned int nr_bytes
,
2022 int (drv_callback
)(struct request
*))
2024 return blk_end_io(rq
, error
, nr_bytes
, 0, drv_callback
);
2026 EXPORT_SYMBOL_GPL(blk_end_request_callback
);
2028 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
2031 /* first two bits are identical in rq->cmd_flags and bio->bi_rw */
2032 rq
->cmd_flags
|= (bio
->bi_rw
& 3);
2034 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2035 rq
->nr_hw_segments
= bio_hw_segments(q
, bio
);
2036 rq
->current_nr_sectors
= bio_cur_sectors(bio
);
2037 rq
->hard_cur_sectors
= rq
->current_nr_sectors
;
2038 rq
->hard_nr_sectors
= rq
->nr_sectors
= bio_sectors(bio
);
2039 rq
->buffer
= bio_data(bio
);
2040 rq
->data_len
= bio
->bi_size
;
2042 rq
->bio
= rq
->biotail
= bio
;
2045 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2048 int kblockd_schedule_work(struct work_struct
*work
)
2050 return queue_work(kblockd_workqueue
, work
);
2052 EXPORT_SYMBOL(kblockd_schedule_work
);
2054 void kblockd_flush_work(struct work_struct
*work
)
2056 cancel_work_sync(work
);
2058 EXPORT_SYMBOL(kblockd_flush_work
);
2060 int __init
blk_dev_init(void)
2064 kblockd_workqueue
= create_workqueue("kblockd");
2065 if (!kblockd_workqueue
)
2066 panic("Failed to create kblockd\n");
2068 request_cachep
= kmem_cache_create("blkdev_requests",
2069 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
2071 blk_requestq_cachep
= kmem_cache_create("blkdev_queue",
2072 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);
2074 for_each_possible_cpu(i
)
2075 INIT_LIST_HEAD(&per_cpu(blk_cpu_done
, i
));
2077 open_softirq(BLOCK_SOFTIRQ
, blk_done_softirq
, NULL
);
2078 register_hotcpu_notifier(&blk_cpu_notifier
);