2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
6 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
8 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
12 * This handles all read/write requests to block devices
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/backing-dev.h>
17 #include <linux/bio.h>
18 #include <linux/blkdev.h>
19 #include <linux/highmem.h>
21 #include <linux/kernel_stat.h>
22 #include <linux/string.h>
23 #include <linux/init.h>
24 #include <linux/completion.h>
25 #include <linux/slab.h>
26 #include <linux/swap.h>
27 #include <linux/writeback.h>
28 #include <linux/task_io_accounting_ops.h>
29 #include <linux/blktrace_api.h>
30 #include <linux/fault-inject.h>
31 #include <trace/block.h>
35 DEFINE_TRACE(block_plug
);
36 DEFINE_TRACE(block_unplug_io
);
37 DEFINE_TRACE(block_unplug_timer
);
38 DEFINE_TRACE(block_getrq
);
39 DEFINE_TRACE(block_sleeprq
);
40 DEFINE_TRACE(block_rq_requeue
);
41 DEFINE_TRACE(block_bio_backmerge
);
42 DEFINE_TRACE(block_bio_frontmerge
);
43 DEFINE_TRACE(block_bio_queue
);
44 DEFINE_TRACE(block_rq_complete
);
45 DEFINE_TRACE(block_remap
); /* Also used in drivers/md/dm.c */
46 EXPORT_TRACEPOINT_SYMBOL_GPL(block_remap
);
48 static int __make_request(struct request_queue
*q
, struct bio
*bio
);
51 * For the allocated request tables
53 static struct kmem_cache
*request_cachep
;
56 * For queue allocation
58 struct kmem_cache
*blk_requestq_cachep
;
61 * Controlling structure to kblockd
63 static struct workqueue_struct
*kblockd_workqueue
;
65 static void drive_stat_acct(struct request
*rq
, int new_io
)
67 struct gendisk
*disk
= rq
->rq_disk
;
68 struct hd_struct
*part
;
69 int rw
= rq_data_dir(rq
);
72 if (!blk_fs_request(rq
) || !disk
|| !blk_do_io_stat(disk
->queue
))
75 cpu
= part_stat_lock();
76 part
= disk_map_sector_rcu(rq
->rq_disk
, rq
->sector
);
79 part_stat_inc(cpu
, part
, merges
[rw
]);
81 part_round_stats(cpu
, part
);
82 part_inc_in_flight(part
);
88 void blk_queue_congestion_threshold(struct request_queue
*q
)
92 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) + 1;
93 if (nr
> q
->nr_requests
)
95 q
->nr_congestion_on
= nr
;
97 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) - (q
->nr_requests
/ 16) - 1;
100 q
->nr_congestion_off
= nr
;
104 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
107 * Locates the passed device's request queue and returns the address of its
110 * Will return NULL if the request queue cannot be located.
112 struct backing_dev_info
*blk_get_backing_dev_info(struct block_device
*bdev
)
114 struct backing_dev_info
*ret
= NULL
;
115 struct request_queue
*q
= bdev_get_queue(bdev
);
118 ret
= &q
->backing_dev_info
;
121 EXPORT_SYMBOL(blk_get_backing_dev_info
);
123 void blk_rq_init(struct request_queue
*q
, struct request
*rq
)
125 memset(rq
, 0, sizeof(*rq
));
127 INIT_LIST_HEAD(&rq
->queuelist
);
128 INIT_LIST_HEAD(&rq
->timeout_list
);
131 rq
->sector
= rq
->hard_sector
= (sector_t
) -1;
132 INIT_HLIST_NODE(&rq
->hash
);
133 RB_CLEAR_NODE(&rq
->rb_node
);
135 rq
->cmd_len
= BLK_MAX_CDB
;
139 EXPORT_SYMBOL(blk_rq_init
);
141 static void req_bio_endio(struct request
*rq
, struct bio
*bio
,
142 unsigned int nbytes
, int error
)
144 struct request_queue
*q
= rq
->q
;
146 if (&q
->bar_rq
!= rq
) {
148 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
149 else if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
152 if (unlikely(nbytes
> bio
->bi_size
)) {
153 printk(KERN_ERR
"%s: want %u bytes done, %u left\n",
154 __func__
, nbytes
, bio
->bi_size
);
155 nbytes
= bio
->bi_size
;
158 if (unlikely(rq
->cmd_flags
& REQ_QUIET
))
159 set_bit(BIO_QUIET
, &bio
->bi_flags
);
161 bio
->bi_size
-= nbytes
;
162 bio
->bi_sector
+= (nbytes
>> 9);
164 if (bio_integrity(bio
))
165 bio_integrity_advance(bio
, nbytes
);
167 if (bio
->bi_size
== 0)
168 bio_endio(bio
, error
);
172 * Okay, this is the barrier request in progress, just
175 if (error
&& !q
->orderr
)
180 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
184 printk(KERN_INFO
"%s: dev %s: type=%x, flags=%x\n", msg
,
185 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->cmd_type
,
188 printk(KERN_INFO
" sector %llu, nr/cnr %lu/%u\n",
189 (unsigned long long)rq
->sector
,
191 rq
->current_nr_sectors
);
192 printk(KERN_INFO
" bio %p, biotail %p, buffer %p, data %p, len %u\n",
193 rq
->bio
, rq
->biotail
,
194 rq
->buffer
, rq
->data
,
197 if (blk_pc_request(rq
)) {
198 printk(KERN_INFO
" cdb: ");
199 for (bit
= 0; bit
< BLK_MAX_CDB
; bit
++)
200 printk("%02x ", rq
->cmd
[bit
]);
204 EXPORT_SYMBOL(blk_dump_rq_flags
);
207 * "plug" the device if there are no outstanding requests: this will
208 * force the transfer to start only after we have put all the requests
211 * This is called with interrupts off and no requests on the queue and
212 * with the queue lock held.
214 void blk_plug_device(struct request_queue
*q
)
216 WARN_ON(!irqs_disabled());
219 * don't plug a stopped queue, it must be paired with blk_start_queue()
220 * which will restart the queueing
222 if (blk_queue_stopped(q
))
225 if (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED
, q
)) {
226 mod_timer(&q
->unplug_timer
, jiffies
+ q
->unplug_delay
);
230 EXPORT_SYMBOL(blk_plug_device
);
233 * blk_plug_device_unlocked - plug a device without queue lock held
234 * @q: The &struct request_queue to plug
237 * Like @blk_plug_device(), but grabs the queue lock and disables
240 void blk_plug_device_unlocked(struct request_queue
*q
)
244 spin_lock_irqsave(q
->queue_lock
, flags
);
246 spin_unlock_irqrestore(q
->queue_lock
, flags
);
248 EXPORT_SYMBOL(blk_plug_device_unlocked
);
251 * remove the queue from the plugged list, if present. called with
252 * queue lock held and interrupts disabled.
254 int blk_remove_plug(struct request_queue
*q
)
256 WARN_ON(!irqs_disabled());
258 if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED
, q
))
261 del_timer(&q
->unplug_timer
);
264 EXPORT_SYMBOL(blk_remove_plug
);
267 * remove the plug and let it rip..
269 void __generic_unplug_device(struct request_queue
*q
)
271 if (unlikely(blk_queue_stopped(q
)))
273 if (!blk_remove_plug(q
) && !blk_queue_nonrot(q
))
280 * generic_unplug_device - fire a request queue
281 * @q: The &struct request_queue in question
284 * Linux uses plugging to build bigger requests queues before letting
285 * the device have at them. If a queue is plugged, the I/O scheduler
286 * is still adding and merging requests on the queue. Once the queue
287 * gets unplugged, the request_fn defined for the queue is invoked and
290 void generic_unplug_device(struct request_queue
*q
)
292 if (blk_queue_plugged(q
)) {
293 spin_lock_irq(q
->queue_lock
);
294 __generic_unplug_device(q
);
295 spin_unlock_irq(q
->queue_lock
);
298 EXPORT_SYMBOL(generic_unplug_device
);
300 static void blk_backing_dev_unplug(struct backing_dev_info
*bdi
,
303 struct request_queue
*q
= bdi
->unplug_io_data
;
308 void blk_unplug_work(struct work_struct
*work
)
310 struct request_queue
*q
=
311 container_of(work
, struct request_queue
, unplug_work
);
313 trace_block_unplug_io(q
);
317 void blk_unplug_timeout(unsigned long data
)
319 struct request_queue
*q
= (struct request_queue
*)data
;
321 trace_block_unplug_timer(q
);
322 kblockd_schedule_work(q
, &q
->unplug_work
);
325 void blk_unplug(struct request_queue
*q
)
328 * devices don't necessarily have an ->unplug_fn defined
331 trace_block_unplug_io(q
);
335 EXPORT_SYMBOL(blk_unplug
);
337 static void blk_invoke_request_fn(struct request_queue
*q
)
339 if (unlikely(blk_queue_stopped(q
)))
343 * one level of recursion is ok and is much faster than kicking
344 * the unplug handling
346 if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER
, q
)) {
348 queue_flag_clear(QUEUE_FLAG_REENTER
, q
);
350 queue_flag_set(QUEUE_FLAG_PLUGGED
, q
);
351 kblockd_schedule_work(q
, &q
->unplug_work
);
356 * blk_start_queue - restart a previously stopped queue
357 * @q: The &struct request_queue in question
360 * blk_start_queue() will clear the stop flag on the queue, and call
361 * the request_fn for the queue if it was in a stopped state when
362 * entered. Also see blk_stop_queue(). Queue lock must be held.
364 void blk_start_queue(struct request_queue
*q
)
366 WARN_ON(!irqs_disabled());
368 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
369 blk_invoke_request_fn(q
);
371 EXPORT_SYMBOL(blk_start_queue
);
374 * blk_stop_queue - stop a queue
375 * @q: The &struct request_queue in question
378 * The Linux block layer assumes that a block driver will consume all
379 * entries on the request queue when the request_fn strategy is called.
380 * Often this will not happen, because of hardware limitations (queue
381 * depth settings). If a device driver gets a 'queue full' response,
382 * or if it simply chooses not to queue more I/O at one point, it can
383 * call this function to prevent the request_fn from being called until
384 * the driver has signalled it's ready to go again. This happens by calling
385 * blk_start_queue() to restart queue operations. Queue lock must be held.
387 void blk_stop_queue(struct request_queue
*q
)
390 queue_flag_set(QUEUE_FLAG_STOPPED
, q
);
392 EXPORT_SYMBOL(blk_stop_queue
);
395 * blk_sync_queue - cancel any pending callbacks on a queue
399 * The block layer may perform asynchronous callback activity
400 * on a queue, such as calling the unplug function after a timeout.
401 * A block device may call blk_sync_queue to ensure that any
402 * such activity is cancelled, thus allowing it to release resources
403 * that the callbacks might use. The caller must already have made sure
404 * that its ->make_request_fn will not re-add plugging prior to calling
408 void blk_sync_queue(struct request_queue
*q
)
410 del_timer_sync(&q
->unplug_timer
);
411 del_timer_sync(&q
->timeout
);
412 cancel_work_sync(&q
->unplug_work
);
414 EXPORT_SYMBOL(blk_sync_queue
);
417 * __blk_run_queue - run a single device queue
418 * @q: The queue to run
421 * See @blk_run_queue. This variant must be called with the queue lock
422 * held and interrupts disabled.
425 void __blk_run_queue(struct request_queue
*q
)
430 * Only recurse once to avoid overrunning the stack, let the unplug
431 * handling reinvoke the handler shortly if we already got there.
433 if (!elv_queue_empty(q
))
434 blk_invoke_request_fn(q
);
436 EXPORT_SYMBOL(__blk_run_queue
);
439 * blk_run_queue - run a single device queue
440 * @q: The queue to run
443 * Invoke request handling on this queue, if it has pending work to do.
444 * May be used to restart queueing when a request has completed. Also
445 * See @blk_start_queueing.
448 void blk_run_queue(struct request_queue
*q
)
452 spin_lock_irqsave(q
->queue_lock
, flags
);
454 spin_unlock_irqrestore(q
->queue_lock
, flags
);
456 EXPORT_SYMBOL(blk_run_queue
);
458 void blk_put_queue(struct request_queue
*q
)
460 kobject_put(&q
->kobj
);
463 void blk_cleanup_queue(struct request_queue
*q
)
466 * We know we have process context here, so we can be a little
467 * cautious and ensure that pending block actions on this device
468 * are done before moving on. Going into this function, we should
469 * not have processes doing IO to this device.
473 mutex_lock(&q
->sysfs_lock
);
474 queue_flag_set_unlocked(QUEUE_FLAG_DEAD
, q
);
475 mutex_unlock(&q
->sysfs_lock
);
478 elevator_exit(q
->elevator
);
482 EXPORT_SYMBOL(blk_cleanup_queue
);
484 static int blk_init_free_list(struct request_queue
*q
)
486 struct request_list
*rl
= &q
->rq
;
488 rl
->count
[READ
] = rl
->count
[WRITE
] = 0;
489 rl
->starved
[READ
] = rl
->starved
[WRITE
] = 0;
491 init_waitqueue_head(&rl
->wait
[READ
]);
492 init_waitqueue_head(&rl
->wait
[WRITE
]);
494 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, mempool_alloc_slab
,
495 mempool_free_slab
, request_cachep
, q
->node
);
503 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
505 return blk_alloc_queue_node(gfp_mask
, -1);
507 EXPORT_SYMBOL(blk_alloc_queue
);
509 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
511 struct request_queue
*q
;
514 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
515 gfp_mask
| __GFP_ZERO
, node_id
);
519 q
->backing_dev_info
.unplug_io_fn
= blk_backing_dev_unplug
;
520 q
->backing_dev_info
.unplug_io_data
= q
;
521 err
= bdi_init(&q
->backing_dev_info
);
523 kmem_cache_free(blk_requestq_cachep
, q
);
527 init_timer(&q
->unplug_timer
);
528 setup_timer(&q
->timeout
, blk_rq_timed_out_timer
, (unsigned long) q
);
529 INIT_LIST_HEAD(&q
->timeout_list
);
530 INIT_WORK(&q
->unplug_work
, blk_unplug_work
);
532 kobject_init(&q
->kobj
, &blk_queue_ktype
);
534 mutex_init(&q
->sysfs_lock
);
535 spin_lock_init(&q
->__queue_lock
);
539 EXPORT_SYMBOL(blk_alloc_queue_node
);
542 * blk_init_queue - prepare a request queue for use with a block device
543 * @rfn: The function to be called to process requests that have been
544 * placed on the queue.
545 * @lock: Request queue spin lock
548 * If a block device wishes to use the standard request handling procedures,
549 * which sorts requests and coalesces adjacent requests, then it must
550 * call blk_init_queue(). The function @rfn will be called when there
551 * are requests on the queue that need to be processed. If the device
552 * supports plugging, then @rfn may not be called immediately when requests
553 * are available on the queue, but may be called at some time later instead.
554 * Plugged queues are generally unplugged when a buffer belonging to one
555 * of the requests on the queue is needed, or due to memory pressure.
557 * @rfn is not required, or even expected, to remove all requests off the
558 * queue, but only as many as it can handle at a time. If it does leave
559 * requests on the queue, it is responsible for arranging that the requests
560 * get dealt with eventually.
562 * The queue spin lock must be held while manipulating the requests on the
563 * request queue; this lock will be taken also from interrupt context, so irq
564 * disabling is needed for it.
566 * Function returns a pointer to the initialized request queue, or %NULL if
570 * blk_init_queue() must be paired with a blk_cleanup_queue() call
571 * when the block device is deactivated (such as at module unload).
574 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
576 return blk_init_queue_node(rfn
, lock
, -1);
578 EXPORT_SYMBOL(blk_init_queue
);
580 struct request_queue
*
581 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
583 struct request_queue
*q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
589 if (blk_init_free_list(q
)) {
590 kmem_cache_free(blk_requestq_cachep
, q
);
595 * if caller didn't supply a lock, they get per-queue locking with
599 lock
= &q
->__queue_lock
;
602 q
->prep_rq_fn
= NULL
;
603 q
->unplug_fn
= generic_unplug_device
;
604 q
->queue_flags
= QUEUE_FLAG_DEFAULT
;
605 q
->queue_lock
= lock
;
607 blk_queue_segment_boundary(q
, BLK_SEG_BOUNDARY_MASK
);
609 blk_queue_make_request(q
, __make_request
);
610 blk_queue_max_segment_size(q
, MAX_SEGMENT_SIZE
);
612 blk_queue_max_hw_segments(q
, MAX_HW_SEGMENTS
);
613 blk_queue_max_phys_segments(q
, MAX_PHYS_SEGMENTS
);
615 q
->sg_reserved_size
= INT_MAX
;
617 blk_set_cmd_filter_defaults(&q
->cmd_filter
);
622 if (!elevator_init(q
, NULL
)) {
623 blk_queue_congestion_threshold(q
);
630 EXPORT_SYMBOL(blk_init_queue_node
);
632 int blk_get_queue(struct request_queue
*q
)
634 if (likely(!test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
))) {
635 kobject_get(&q
->kobj
);
642 static inline void blk_free_request(struct request_queue
*q
, struct request
*rq
)
644 if (rq
->cmd_flags
& REQ_ELVPRIV
)
645 elv_put_request(q
, rq
);
646 mempool_free(rq
, q
->rq
.rq_pool
);
649 static struct request
*
650 blk_alloc_request(struct request_queue
*q
, int rw
, int priv
, gfp_t gfp_mask
)
652 struct request
*rq
= mempool_alloc(q
->rq
.rq_pool
, gfp_mask
);
659 rq
->cmd_flags
= rw
| REQ_ALLOCED
;
662 if (unlikely(elv_set_request(q
, rq
, gfp_mask
))) {
663 mempool_free(rq
, q
->rq
.rq_pool
);
666 rq
->cmd_flags
|= REQ_ELVPRIV
;
673 * ioc_batching returns true if the ioc is a valid batching request and
674 * should be given priority access to a request.
676 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
682 * Make sure the process is able to allocate at least 1 request
683 * even if the batch times out, otherwise we could theoretically
686 return ioc
->nr_batch_requests
== q
->nr_batching
||
687 (ioc
->nr_batch_requests
> 0
688 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
692 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
693 * will cause the process to be a "batcher" on all queues in the system. This
694 * is the behaviour we want though - once it gets a wakeup it should be given
697 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
699 if (!ioc
|| ioc_batching(q
, ioc
))
702 ioc
->nr_batch_requests
= q
->nr_batching
;
703 ioc
->last_waited
= jiffies
;
706 static void __freed_request(struct request_queue
*q
, int rw
)
708 struct request_list
*rl
= &q
->rq
;
710 if (rl
->count
[rw
] < queue_congestion_off_threshold(q
))
711 blk_clear_queue_congested(q
, rw
);
713 if (rl
->count
[rw
] + 1 <= q
->nr_requests
) {
714 if (waitqueue_active(&rl
->wait
[rw
]))
715 wake_up(&rl
->wait
[rw
]);
717 blk_clear_queue_full(q
, rw
);
722 * A request has just been released. Account for it, update the full and
723 * congestion status, wake up any waiters. Called under q->queue_lock.
725 static void freed_request(struct request_queue
*q
, int rw
, int priv
)
727 struct request_list
*rl
= &q
->rq
;
733 __freed_request(q
, rw
);
735 if (unlikely(rl
->starved
[rw
^ 1]))
736 __freed_request(q
, rw
^ 1);
739 #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
741 * Get a free request, queue_lock must be held.
742 * Returns NULL on failure, with queue_lock held.
743 * Returns !NULL on success, with queue_lock *not held*.
745 static struct request
*get_request(struct request_queue
*q
, int rw_flags
,
746 struct bio
*bio
, gfp_t gfp_mask
)
748 struct request
*rq
= NULL
;
749 struct request_list
*rl
= &q
->rq
;
750 struct io_context
*ioc
= NULL
;
751 const int rw
= rw_flags
& 0x01;
754 may_queue
= elv_may_queue(q
, rw_flags
);
755 if (may_queue
== ELV_MQUEUE_NO
)
758 if (rl
->count
[rw
]+1 >= queue_congestion_on_threshold(q
)) {
759 if (rl
->count
[rw
]+1 >= q
->nr_requests
) {
760 ioc
= current_io_context(GFP_ATOMIC
, q
->node
);
762 * The queue will fill after this allocation, so set
763 * it as full, and mark this process as "batching".
764 * This process will be allowed to complete a batch of
765 * requests, others will be blocked.
767 if (!blk_queue_full(q
, rw
)) {
768 ioc_set_batching(q
, ioc
);
769 blk_set_queue_full(q
, rw
);
771 if (may_queue
!= ELV_MQUEUE_MUST
772 && !ioc_batching(q
, ioc
)) {
774 * The queue is full and the allocating
775 * process is not a "batcher", and not
776 * exempted by the IO scheduler
782 blk_set_queue_congested(q
, rw
);
786 * Only allow batching queuers to allocate up to 50% over the defined
787 * limit of requests, otherwise we could have thousands of requests
788 * allocated with any setting of ->nr_requests
790 if (rl
->count
[rw
] >= (3 * q
->nr_requests
/ 2))
796 priv
= !test_bit(QUEUE_FLAG_ELVSWITCH
, &q
->queue_flags
);
800 spin_unlock_irq(q
->queue_lock
);
802 rq
= blk_alloc_request(q
, rw_flags
, priv
, gfp_mask
);
805 * Allocation failed presumably due to memory. Undo anything
806 * we might have messed up.
808 * Allocating task should really be put onto the front of the
809 * wait queue, but this is pretty rare.
811 spin_lock_irq(q
->queue_lock
);
812 freed_request(q
, rw
, priv
);
815 * in the very unlikely event that allocation failed and no
816 * requests for this direction was pending, mark us starved
817 * so that freeing of a request in the other direction will
818 * notice us. another possible fix would be to split the
819 * rq mempool into READ and WRITE
822 if (unlikely(rl
->count
[rw
] == 0))
829 * ioc may be NULL here, and ioc_batching will be false. That's
830 * OK, if the queue is under the request limit then requests need
831 * not count toward the nr_batch_requests limit. There will always
832 * be some limit enforced by BLK_BATCH_TIME.
834 if (ioc_batching(q
, ioc
))
835 ioc
->nr_batch_requests
--;
837 trace_block_getrq(q
, bio
, rw
);
843 * No available requests for this queue, unplug the device and wait for some
844 * requests to become available.
846 * Called with q->queue_lock held, and returns with it unlocked.
848 static struct request
*get_request_wait(struct request_queue
*q
, int rw_flags
,
851 const int rw
= rw_flags
& 0x01;
854 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
857 struct io_context
*ioc
;
858 struct request_list
*rl
= &q
->rq
;
860 prepare_to_wait_exclusive(&rl
->wait
[rw
], &wait
,
861 TASK_UNINTERRUPTIBLE
);
863 trace_block_sleeprq(q
, bio
, rw
);
865 __generic_unplug_device(q
);
866 spin_unlock_irq(q
->queue_lock
);
870 * After sleeping, we become a "batching" process and
871 * will be able to allocate at least one request, and
872 * up to a big batch of them for a small period time.
873 * See ioc_batching, ioc_set_batching
875 ioc
= current_io_context(GFP_NOIO
, q
->node
);
876 ioc_set_batching(q
, ioc
);
878 spin_lock_irq(q
->queue_lock
);
879 finish_wait(&rl
->wait
[rw
], &wait
);
881 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
887 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
891 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
893 spin_lock_irq(q
->queue_lock
);
894 if (gfp_mask
& __GFP_WAIT
) {
895 rq
= get_request_wait(q
, rw
, NULL
);
897 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
899 spin_unlock_irq(q
->queue_lock
);
901 /* q->queue_lock is unlocked at this point */
905 EXPORT_SYMBOL(blk_get_request
);
908 * blk_start_queueing - initiate dispatch of requests to device
909 * @q: request queue to kick into gear
911 * This is basically a helper to remove the need to know whether a queue
912 * is plugged or not if someone just wants to initiate dispatch of requests
913 * for this queue. Should be used to start queueing on a device outside
914 * of ->request_fn() context. Also see @blk_run_queue.
916 * The queue lock must be held with interrupts disabled.
918 void blk_start_queueing(struct request_queue
*q
)
920 if (!blk_queue_plugged(q
)) {
921 if (unlikely(blk_queue_stopped(q
)))
925 __generic_unplug_device(q
);
927 EXPORT_SYMBOL(blk_start_queueing
);
930 * blk_requeue_request - put a request back on queue
931 * @q: request queue where request should be inserted
932 * @rq: request to be inserted
935 * Drivers often keep queueing requests until the hardware cannot accept
936 * more, when that condition happens we need to put the request back
937 * on the queue. Must be called with queue lock held.
939 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
941 blk_delete_timer(rq
);
942 blk_clear_rq_complete(rq
);
943 trace_block_rq_requeue(q
, rq
);
945 if (blk_rq_tagged(rq
))
946 blk_queue_end_tag(q
, rq
);
948 elv_requeue_request(q
, rq
);
950 EXPORT_SYMBOL(blk_requeue_request
);
953 * blk_insert_request - insert a special request into a request queue
954 * @q: request queue where request should be inserted
955 * @rq: request to be inserted
956 * @at_head: insert request at head or tail of queue
957 * @data: private data
960 * Many block devices need to execute commands asynchronously, so they don't
961 * block the whole kernel from preemption during request execution. This is
962 * accomplished normally by inserting aritficial requests tagged as
963 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
964 * be scheduled for actual execution by the request queue.
966 * We have the option of inserting the head or the tail of the queue.
967 * Typically we use the tail for new ioctls and so forth. We use the head
968 * of the queue for things like a QUEUE_FULL message from a device, or a
969 * host that is unable to accept a particular command.
971 void blk_insert_request(struct request_queue
*q
, struct request
*rq
,
972 int at_head
, void *data
)
974 int where
= at_head
? ELEVATOR_INSERT_FRONT
: ELEVATOR_INSERT_BACK
;
978 * tell I/O scheduler that this isn't a regular read/write (ie it
979 * must not attempt merges on this) and that it acts as a soft
982 rq
->cmd_type
= REQ_TYPE_SPECIAL
;
983 rq
->cmd_flags
|= REQ_SOFTBARRIER
;
987 spin_lock_irqsave(q
->queue_lock
, flags
);
990 * If command is tagged, release the tag
992 if (blk_rq_tagged(rq
))
993 blk_queue_end_tag(q
, rq
);
995 drive_stat_acct(rq
, 1);
996 __elv_add_request(q
, rq
, where
, 0);
997 blk_start_queueing(q
);
998 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1000 EXPORT_SYMBOL(blk_insert_request
);
1003 * add-request adds a request to the linked list.
1004 * queue lock is held and interrupts disabled, as we muck with the
1005 * request queue list.
1007 static inline void add_request(struct request_queue
*q
, struct request
*req
)
1009 drive_stat_acct(req
, 1);
1012 * elevator indicated where it wants this request to be
1013 * inserted at elevator_merge time
1015 __elv_add_request(q
, req
, ELEVATOR_INSERT_SORT
, 0);
1018 static void part_round_stats_single(int cpu
, struct hd_struct
*part
,
1021 if (now
== part
->stamp
)
1024 if (part
->in_flight
) {
1025 __part_stat_add(cpu
, part
, time_in_queue
,
1026 part
->in_flight
* (now
- part
->stamp
));
1027 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
1033 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1034 * @cpu: cpu number for stats access
1035 * @part: target partition
1037 * The average IO queue length and utilisation statistics are maintained
1038 * by observing the current state of the queue length and the amount of
1039 * time it has been in this state for.
1041 * Normally, that accounting is done on IO completion, but that can result
1042 * in more than a second's worth of IO being accounted for within any one
1043 * second, leading to >100% utilisation. To deal with that, we call this
1044 * function to do a round-off before returning the results when reading
1045 * /proc/diskstats. This accounts immediately for all queue usage up to
1046 * the current jiffies and restarts the counters again.
1048 void part_round_stats(int cpu
, struct hd_struct
*part
)
1050 unsigned long now
= jiffies
;
1053 part_round_stats_single(cpu
, &part_to_disk(part
)->part0
, now
);
1054 part_round_stats_single(cpu
, part
, now
);
1056 EXPORT_SYMBOL_GPL(part_round_stats
);
1059 * queue lock must be held
1061 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1065 if (unlikely(--req
->ref_count
))
1068 elv_completed_request(q
, req
);
1071 * Request may not have originated from ll_rw_blk. if not,
1072 * it didn't come out of our reserved rq pools
1074 if (req
->cmd_flags
& REQ_ALLOCED
) {
1075 int rw
= rq_data_dir(req
);
1076 int priv
= req
->cmd_flags
& REQ_ELVPRIV
;
1078 BUG_ON(!list_empty(&req
->queuelist
));
1079 BUG_ON(!hlist_unhashed(&req
->hash
));
1081 blk_free_request(q
, req
);
1082 freed_request(q
, rw
, priv
);
1085 EXPORT_SYMBOL_GPL(__blk_put_request
);
1087 void blk_put_request(struct request
*req
)
1089 unsigned long flags
;
1090 struct request_queue
*q
= req
->q
;
1092 spin_lock_irqsave(q
->queue_lock
, flags
);
1093 __blk_put_request(q
, req
);
1094 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1096 EXPORT_SYMBOL(blk_put_request
);
1098 void init_request_from_bio(struct request
*req
, struct bio
*bio
)
1100 req
->cpu
= bio
->bi_comp_cpu
;
1101 req
->cmd_type
= REQ_TYPE_FS
;
1104 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
1106 if (bio_rw_ahead(bio
))
1107 req
->cmd_flags
|= (REQ_FAILFAST_DEV
| REQ_FAILFAST_TRANSPORT
|
1108 REQ_FAILFAST_DRIVER
);
1109 if (bio_failfast_dev(bio
))
1110 req
->cmd_flags
|= REQ_FAILFAST_DEV
;
1111 if (bio_failfast_transport(bio
))
1112 req
->cmd_flags
|= REQ_FAILFAST_TRANSPORT
;
1113 if (bio_failfast_driver(bio
))
1114 req
->cmd_flags
|= REQ_FAILFAST_DRIVER
;
1117 * REQ_BARRIER implies no merging, but lets make it explicit
1119 if (unlikely(bio_discard(bio
))) {
1120 req
->cmd_flags
|= REQ_DISCARD
;
1121 if (bio_barrier(bio
))
1122 req
->cmd_flags
|= REQ_SOFTBARRIER
;
1123 req
->q
->prepare_discard_fn(req
->q
, req
);
1124 } else if (unlikely(bio_barrier(bio
)))
1125 req
->cmd_flags
|= (REQ_HARDBARRIER
| REQ_NOMERGE
);
1128 req
->cmd_flags
|= REQ_RW_SYNC
;
1129 if (bio_unplug(bio
))
1130 req
->cmd_flags
|= REQ_UNPLUG
;
1131 if (bio_rw_meta(bio
))
1132 req
->cmd_flags
|= REQ_RW_META
;
1135 req
->hard_sector
= req
->sector
= bio
->bi_sector
;
1136 req
->ioprio
= bio_prio(bio
);
1137 req
->start_time
= jiffies
;
1138 blk_rq_bio_prep(req
->q
, req
, bio
);
1141 static int __make_request(struct request_queue
*q
, struct bio
*bio
)
1143 struct request
*req
;
1144 int el_ret
, nr_sectors
;
1145 const unsigned short prio
= bio_prio(bio
);
1146 const int sync
= bio_sync(bio
);
1147 const int unplug
= bio_unplug(bio
);
1150 nr_sectors
= bio_sectors(bio
);
1153 * low level driver can indicate that it wants pages above a
1154 * certain limit bounced to low memory (ie for highmem, or even
1155 * ISA dma in theory)
1157 blk_queue_bounce(q
, &bio
);
1159 spin_lock_irq(q
->queue_lock
);
1161 if (unlikely(bio_barrier(bio
)) || elv_queue_empty(q
))
1164 el_ret
= elv_merge(q
, &req
, bio
);
1166 case ELEVATOR_BACK_MERGE
:
1167 BUG_ON(!rq_mergeable(req
));
1169 if (!ll_back_merge_fn(q
, req
, bio
))
1172 trace_block_bio_backmerge(q
, bio
);
1174 req
->biotail
->bi_next
= bio
;
1176 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
1177 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
1178 if (!blk_rq_cpu_valid(req
))
1179 req
->cpu
= bio
->bi_comp_cpu
;
1180 drive_stat_acct(req
, 0);
1181 if (!attempt_back_merge(q
, req
))
1182 elv_merged_request(q
, req
, el_ret
);
1185 case ELEVATOR_FRONT_MERGE
:
1186 BUG_ON(!rq_mergeable(req
));
1188 if (!ll_front_merge_fn(q
, req
, bio
))
1191 trace_block_bio_frontmerge(q
, bio
);
1193 bio
->bi_next
= req
->bio
;
1197 * may not be valid. if the low level driver said
1198 * it didn't need a bounce buffer then it better
1199 * not touch req->buffer either...
1201 req
->buffer
= bio_data(bio
);
1202 req
->current_nr_sectors
= bio_cur_sectors(bio
);
1203 req
->hard_cur_sectors
= req
->current_nr_sectors
;
1204 req
->sector
= req
->hard_sector
= bio
->bi_sector
;
1205 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
1206 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
1207 if (!blk_rq_cpu_valid(req
))
1208 req
->cpu
= bio
->bi_comp_cpu
;
1209 drive_stat_acct(req
, 0);
1210 if (!attempt_front_merge(q
, req
))
1211 elv_merged_request(q
, req
, el_ret
);
1214 /* ELV_NO_MERGE: elevator says don't/can't merge. */
1221 * This sync check and mask will be re-done in init_request_from_bio(),
1222 * but we need to set it earlier to expose the sync flag to the
1223 * rq allocator and io schedulers.
1225 rw_flags
= bio_data_dir(bio
);
1227 rw_flags
|= REQ_RW_SYNC
;
1230 * Grab a free request. This is might sleep but can not fail.
1231 * Returns with the queue unlocked.
1233 req
= get_request_wait(q
, rw_flags
, bio
);
1236 * After dropping the lock and possibly sleeping here, our request
1237 * may now be mergeable after it had proven unmergeable (above).
1238 * We don't worry about that case for efficiency. It won't happen
1239 * often, and the elevators are able to handle it.
1241 init_request_from_bio(req
, bio
);
1243 spin_lock_irq(q
->queue_lock
);
1244 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
) ||
1245 bio_flagged(bio
, BIO_CPU_AFFINE
))
1246 req
->cpu
= blk_cpu_to_group(smp_processor_id());
1247 if (!blk_queue_nonrot(q
) && elv_queue_empty(q
))
1249 add_request(q
, req
);
1251 if (unplug
|| blk_queue_nonrot(q
))
1252 __generic_unplug_device(q
);
1253 spin_unlock_irq(q
->queue_lock
);
1258 * If bio->bi_dev is a partition, remap the location
1260 static inline void blk_partition_remap(struct bio
*bio
)
1262 struct block_device
*bdev
= bio
->bi_bdev
;
1264 if (bio_sectors(bio
) && bdev
!= bdev
->bd_contains
) {
1265 struct hd_struct
*p
= bdev
->bd_part
;
1267 bio
->bi_sector
+= p
->start_sect
;
1268 bio
->bi_bdev
= bdev
->bd_contains
;
1270 trace_block_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
1271 bdev
->bd_dev
, bio
->bi_sector
,
1272 bio
->bi_sector
- p
->start_sect
);
1276 static void handle_bad_sector(struct bio
*bio
)
1278 char b
[BDEVNAME_SIZE
];
1280 printk(KERN_INFO
"attempt to access beyond end of device\n");
1281 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
1282 bdevname(bio
->bi_bdev
, b
),
1284 (unsigned long long)bio
->bi_sector
+ bio_sectors(bio
),
1285 (long long)(bio
->bi_bdev
->bd_inode
->i_size
>> 9));
1287 set_bit(BIO_EOF
, &bio
->bi_flags
);
1290 #ifdef CONFIG_FAIL_MAKE_REQUEST
1292 static DECLARE_FAULT_ATTR(fail_make_request
);
1294 static int __init
setup_fail_make_request(char *str
)
1296 return setup_fault_attr(&fail_make_request
, str
);
1298 __setup("fail_make_request=", setup_fail_make_request
);
1300 static int should_fail_request(struct bio
*bio
)
1302 struct hd_struct
*part
= bio
->bi_bdev
->bd_part
;
1304 if (part_to_disk(part
)->part0
.make_it_fail
|| part
->make_it_fail
)
1305 return should_fail(&fail_make_request
, bio
->bi_size
);
1310 static int __init
fail_make_request_debugfs(void)
1312 return init_fault_attr_dentries(&fail_make_request
,
1313 "fail_make_request");
1316 late_initcall(fail_make_request_debugfs
);
1318 #else /* CONFIG_FAIL_MAKE_REQUEST */
1320 static inline int should_fail_request(struct bio
*bio
)
1325 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1328 * Check whether this bio extends beyond the end of the device.
1330 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
1337 /* Test device or partition size, when known. */
1338 maxsector
= bio
->bi_bdev
->bd_inode
->i_size
>> 9;
1340 sector_t sector
= bio
->bi_sector
;
1342 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
1344 * This may well happen - the kernel calls bread()
1345 * without checking the size of the device, e.g., when
1346 * mounting a device.
1348 handle_bad_sector(bio
);
1357 * generic_make_request - hand a buffer to its device driver for I/O
1358 * @bio: The bio describing the location in memory and on the device.
1360 * generic_make_request() is used to make I/O requests of block
1361 * devices. It is passed a &struct bio, which describes the I/O that needs
1364 * generic_make_request() does not return any status. The
1365 * success/failure status of the request, along with notification of
1366 * completion, is delivered asynchronously through the bio->bi_end_io
1367 * function described (one day) else where.
1369 * The caller of generic_make_request must make sure that bi_io_vec
1370 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1371 * set to describe the device address, and the
1372 * bi_end_io and optionally bi_private are set to describe how
1373 * completion notification should be signaled.
1375 * generic_make_request and the drivers it calls may use bi_next if this
1376 * bio happens to be merged with someone else, and may change bi_dev and
1377 * bi_sector for remaps as it sees fit. So the values of these fields
1378 * should NOT be depended on after the call to generic_make_request.
1380 static inline void __generic_make_request(struct bio
*bio
)
1382 struct request_queue
*q
;
1383 sector_t old_sector
;
1384 int ret
, nr_sectors
= bio_sectors(bio
);
1390 if (bio_check_eod(bio
, nr_sectors
))
1394 * Resolve the mapping until finished. (drivers are
1395 * still free to implement/resolve their own stacking
1396 * by explicitly returning 0)
1398 * NOTE: we don't repeat the blk_size check for each new device.
1399 * Stacking drivers are expected to know what they are doing.
1404 char b
[BDEVNAME_SIZE
];
1406 q
= bdev_get_queue(bio
->bi_bdev
);
1409 "generic_make_request: Trying to access "
1410 "nonexistent block-device %s (%Lu)\n",
1411 bdevname(bio
->bi_bdev
, b
),
1412 (long long) bio
->bi_sector
);
1416 if (unlikely(nr_sectors
> q
->max_hw_sectors
)) {
1417 printk(KERN_ERR
"bio too big device %s (%u > %u)\n",
1418 bdevname(bio
->bi_bdev
, b
),
1424 if (unlikely(test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)))
1427 if (should_fail_request(bio
))
1431 * If this device has partitions, remap block n
1432 * of partition p to block n+start(p) of the disk.
1434 blk_partition_remap(bio
);
1436 if (bio_integrity_enabled(bio
) && bio_integrity_prep(bio
))
1439 if (old_sector
!= -1)
1440 trace_block_remap(q
, bio
, old_dev
, bio
->bi_sector
,
1443 trace_block_bio_queue(q
, bio
);
1445 old_sector
= bio
->bi_sector
;
1446 old_dev
= bio
->bi_bdev
->bd_dev
;
1448 if (bio_check_eod(bio
, nr_sectors
))
1451 if (bio_discard(bio
) && !q
->prepare_discard_fn
) {
1455 if (bio_barrier(bio
) && bio_has_data(bio
) &&
1456 (q
->next_ordered
== QUEUE_ORDERED_NONE
)) {
1461 ret
= q
->make_request_fn(q
, bio
);
1467 bio_endio(bio
, err
);
1471 * We only want one ->make_request_fn to be active at a time,
1472 * else stack usage with stacked devices could be a problem.
1473 * So use current->bio_{list,tail} to keep a list of requests
1474 * submited by a make_request_fn function.
1475 * current->bio_tail is also used as a flag to say if
1476 * generic_make_request is currently active in this task or not.
1477 * If it is NULL, then no make_request is active. If it is non-NULL,
1478 * then a make_request is active, and new requests should be added
1481 void generic_make_request(struct bio
*bio
)
1483 if (current
->bio_tail
) {
1484 /* make_request is active */
1485 *(current
->bio_tail
) = bio
;
1486 bio
->bi_next
= NULL
;
1487 current
->bio_tail
= &bio
->bi_next
;
1490 /* following loop may be a bit non-obvious, and so deserves some
1492 * Before entering the loop, bio->bi_next is NULL (as all callers
1493 * ensure that) so we have a list with a single bio.
1494 * We pretend that we have just taken it off a longer list, so
1495 * we assign bio_list to the next (which is NULL) and bio_tail
1496 * to &bio_list, thus initialising the bio_list of new bios to be
1497 * added. __generic_make_request may indeed add some more bios
1498 * through a recursive call to generic_make_request. If it
1499 * did, we find a non-NULL value in bio_list and re-enter the loop
1500 * from the top. In this case we really did just take the bio
1501 * of the top of the list (no pretending) and so fixup bio_list and
1502 * bio_tail or bi_next, and call into __generic_make_request again.
1504 * The loop was structured like this to make only one call to
1505 * __generic_make_request (which is important as it is large and
1506 * inlined) and to keep the structure simple.
1508 BUG_ON(bio
->bi_next
);
1510 current
->bio_list
= bio
->bi_next
;
1511 if (bio
->bi_next
== NULL
)
1512 current
->bio_tail
= ¤t
->bio_list
;
1514 bio
->bi_next
= NULL
;
1515 __generic_make_request(bio
);
1516 bio
= current
->bio_list
;
1518 current
->bio_tail
= NULL
; /* deactivate */
1520 EXPORT_SYMBOL(generic_make_request
);
1523 * submit_bio - submit a bio to the block device layer for I/O
1524 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1525 * @bio: The &struct bio which describes the I/O
1527 * submit_bio() is very similar in purpose to generic_make_request(), and
1528 * uses that function to do most of the work. Both are fairly rough
1529 * interfaces; @bio must be presetup and ready for I/O.
1532 void submit_bio(int rw
, struct bio
*bio
)
1534 int count
= bio_sectors(bio
);
1539 * If it's a regular read/write or a barrier with data attached,
1540 * go through the normal accounting stuff before submission.
1542 if (bio_has_data(bio
)) {
1544 count_vm_events(PGPGOUT
, count
);
1546 task_io_account_read(bio
->bi_size
);
1547 count_vm_events(PGPGIN
, count
);
1550 if (unlikely(block_dump
)) {
1551 char b
[BDEVNAME_SIZE
];
1552 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s\n",
1553 current
->comm
, task_pid_nr(current
),
1554 (rw
& WRITE
) ? "WRITE" : "READ",
1555 (unsigned long long)bio
->bi_sector
,
1556 bdevname(bio
->bi_bdev
, b
));
1560 generic_make_request(bio
);
1562 EXPORT_SYMBOL(submit_bio
);
1565 * blk_rq_check_limits - Helper function to check a request for the queue limit
1567 * @rq: the request being checked
1570 * @rq may have been made based on weaker limitations of upper-level queues
1571 * in request stacking drivers, and it may violate the limitation of @q.
1572 * Since the block layer and the underlying device driver trust @rq
1573 * after it is inserted to @q, it should be checked against @q before
1574 * the insertion using this generic function.
1576 * This function should also be useful for request stacking drivers
1577 * in some cases below, so export this fuction.
1578 * Request stacking drivers like request-based dm may change the queue
1579 * limits while requests are in the queue (e.g. dm's table swapping).
1580 * Such request stacking drivers should check those requests agaist
1581 * the new queue limits again when they dispatch those requests,
1582 * although such checkings are also done against the old queue limits
1583 * when submitting requests.
1585 int blk_rq_check_limits(struct request_queue
*q
, struct request
*rq
)
1587 if (rq
->nr_sectors
> q
->max_sectors
||
1588 rq
->data_len
> q
->max_hw_sectors
<< 9) {
1589 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
1594 * queue's settings related to segment counting like q->bounce_pfn
1595 * may differ from that of other stacking queues.
1596 * Recalculate it to check the request correctly on this queue's
1599 blk_recalc_rq_segments(rq
);
1600 if (rq
->nr_phys_segments
> q
->max_phys_segments
||
1601 rq
->nr_phys_segments
> q
->max_hw_segments
) {
1602 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
1608 EXPORT_SYMBOL_GPL(blk_rq_check_limits
);
1611 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1612 * @q: the queue to submit the request
1613 * @rq: the request being queued
1615 int blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
1617 unsigned long flags
;
1619 if (blk_rq_check_limits(q
, rq
))
1622 #ifdef CONFIG_FAIL_MAKE_REQUEST
1623 if (rq
->rq_disk
&& rq
->rq_disk
->part0
.make_it_fail
&&
1624 should_fail(&fail_make_request
, blk_rq_bytes(rq
)))
1628 spin_lock_irqsave(q
->queue_lock
, flags
);
1631 * Submitting request must be dequeued before calling this function
1632 * because it will be linked to another request_queue
1634 BUG_ON(blk_queued_rq(rq
));
1636 drive_stat_acct(rq
, 1);
1637 __elv_add_request(q
, rq
, ELEVATOR_INSERT_BACK
, 0);
1639 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1643 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
1646 * blkdev_dequeue_request - dequeue request and start timeout timer
1647 * @req: request to dequeue
1649 * Dequeue @req and start timeout timer on it. This hands off the
1650 * request to the driver.
1652 * Block internal functions which don't want to start timer should
1653 * call elv_dequeue_request().
1655 void blkdev_dequeue_request(struct request
*req
)
1657 elv_dequeue_request(req
->q
, req
);
1660 * We are now handing the request to the hardware, add the
1665 EXPORT_SYMBOL(blkdev_dequeue_request
);
1667 static void blk_account_io_completion(struct request
*req
, unsigned int bytes
)
1669 struct gendisk
*disk
= req
->rq_disk
;
1671 if (!disk
|| !blk_do_io_stat(disk
->queue
))
1674 if (blk_fs_request(req
)) {
1675 const int rw
= rq_data_dir(req
);
1676 struct hd_struct
*part
;
1679 cpu
= part_stat_lock();
1680 part
= disk_map_sector_rcu(req
->rq_disk
, req
->sector
);
1681 part_stat_add(cpu
, part
, sectors
[rw
], bytes
>> 9);
1686 static void blk_account_io_done(struct request
*req
)
1688 struct gendisk
*disk
= req
->rq_disk
;
1690 if (!disk
|| !blk_do_io_stat(disk
->queue
))
1694 * Account IO completion. bar_rq isn't accounted as a normal
1695 * IO on queueing nor completion. Accounting the containing
1696 * request is enough.
1698 if (blk_fs_request(req
) && req
!= &req
->q
->bar_rq
) {
1699 unsigned long duration
= jiffies
- req
->start_time
;
1700 const int rw
= rq_data_dir(req
);
1701 struct hd_struct
*part
;
1704 cpu
= part_stat_lock();
1705 part
= disk_map_sector_rcu(disk
, req
->sector
);
1707 part_stat_inc(cpu
, part
, ios
[rw
]);
1708 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
1709 part_round_stats(cpu
, part
);
1710 part_dec_in_flight(part
);
1717 * __end_that_request_first - end I/O on a request
1718 * @req: the request being processed
1719 * @error: %0 for success, < %0 for error
1720 * @nr_bytes: number of bytes to complete
1723 * Ends I/O on a number of bytes attached to @req, and sets it up
1724 * for the next range of segments (if any) in the cluster.
1727 * %0 - we are done with this request, call end_that_request_last()
1728 * %1 - still buffers pending for this request
1730 static int __end_that_request_first(struct request
*req
, int error
,
1733 int total_bytes
, bio_nbytes
, next_idx
= 0;
1736 trace_block_rq_complete(req
->q
, req
);
1739 * for a REQ_TYPE_BLOCK_PC request, we want to carry any eventual
1740 * sense key with us all the way through
1742 if (!blk_pc_request(req
))
1745 if (error
&& (blk_fs_request(req
) && !(req
->cmd_flags
& REQ_QUIET
))) {
1746 printk(KERN_ERR
"end_request: I/O error, dev %s, sector %llu\n",
1747 req
->rq_disk
? req
->rq_disk
->disk_name
: "?",
1748 (unsigned long long)req
->sector
);
1751 blk_account_io_completion(req
, nr_bytes
);
1753 total_bytes
= bio_nbytes
= 0;
1754 while ((bio
= req
->bio
) != NULL
) {
1757 if (nr_bytes
>= bio
->bi_size
) {
1758 req
->bio
= bio
->bi_next
;
1759 nbytes
= bio
->bi_size
;
1760 req_bio_endio(req
, bio
, nbytes
, error
);
1764 int idx
= bio
->bi_idx
+ next_idx
;
1766 if (unlikely(bio
->bi_idx
>= bio
->bi_vcnt
)) {
1767 blk_dump_rq_flags(req
, "__end_that");
1768 printk(KERN_ERR
"%s: bio idx %d >= vcnt %d\n",
1769 __func__
, bio
->bi_idx
, bio
->bi_vcnt
);
1773 nbytes
= bio_iovec_idx(bio
, idx
)->bv_len
;
1774 BIO_BUG_ON(nbytes
> bio
->bi_size
);
1777 * not a complete bvec done
1779 if (unlikely(nbytes
> nr_bytes
)) {
1780 bio_nbytes
+= nr_bytes
;
1781 total_bytes
+= nr_bytes
;
1786 * advance to the next vector
1789 bio_nbytes
+= nbytes
;
1792 total_bytes
+= nbytes
;
1798 * end more in this run, or just return 'not-done'
1800 if (unlikely(nr_bytes
<= 0))
1812 * if the request wasn't completed, update state
1815 req_bio_endio(req
, bio
, bio_nbytes
, error
);
1816 bio
->bi_idx
+= next_idx
;
1817 bio_iovec(bio
)->bv_offset
+= nr_bytes
;
1818 bio_iovec(bio
)->bv_len
-= nr_bytes
;
1821 blk_recalc_rq_sectors(req
, total_bytes
>> 9);
1822 blk_recalc_rq_segments(req
);
1827 * queue lock must be held
1829 static void end_that_request_last(struct request
*req
, int error
)
1831 if (blk_rq_tagged(req
))
1832 blk_queue_end_tag(req
->q
, req
);
1834 if (blk_queued_rq(req
))
1835 elv_dequeue_request(req
->q
, req
);
1837 if (unlikely(laptop_mode
) && blk_fs_request(req
))
1838 laptop_io_completion();
1840 blk_delete_timer(req
);
1842 blk_account_io_done(req
);
1845 req
->end_io(req
, error
);
1847 if (blk_bidi_rq(req
))
1848 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
1850 __blk_put_request(req
->q
, req
);
1855 * blk_rq_bytes - Returns bytes left to complete in the entire request
1856 * @rq: the request being processed
1858 unsigned int blk_rq_bytes(struct request
*rq
)
1860 if (blk_fs_request(rq
))
1861 return rq
->hard_nr_sectors
<< 9;
1863 return rq
->data_len
;
1865 EXPORT_SYMBOL_GPL(blk_rq_bytes
);
1868 * blk_rq_cur_bytes - Returns bytes left to complete in the current segment
1869 * @rq: the request being processed
1871 unsigned int blk_rq_cur_bytes(struct request
*rq
)
1873 if (blk_fs_request(rq
))
1874 return rq
->current_nr_sectors
<< 9;
1877 return rq
->bio
->bi_size
;
1879 return rq
->data_len
;
1881 EXPORT_SYMBOL_GPL(blk_rq_cur_bytes
);
1884 * end_request - end I/O on the current segment of the request
1885 * @req: the request being processed
1886 * @uptodate: error value or %0/%1 uptodate flag
1889 * Ends I/O on the current segment of a request. If that is the only
1890 * remaining segment, the request is also completed and freed.
1892 * This is a remnant of how older block drivers handled I/O completions.
1893 * Modern drivers typically end I/O on the full request in one go, unless
1894 * they have a residual value to account for. For that case this function
1895 * isn't really useful, unless the residual just happens to be the
1896 * full current segment. In other words, don't use this function in new
1897 * code. Use blk_end_request() or __blk_end_request() to end a request.
1899 void end_request(struct request
*req
, int uptodate
)
1904 error
= uptodate
? uptodate
: -EIO
;
1906 __blk_end_request(req
, error
, req
->hard_cur_sectors
<< 9);
1908 EXPORT_SYMBOL(end_request
);
1910 static int end_that_request_data(struct request
*rq
, int error
,
1911 unsigned int nr_bytes
, unsigned int bidi_bytes
)
1914 if (__end_that_request_first(rq
, error
, nr_bytes
))
1917 /* Bidi request must be completed as a whole */
1918 if (blk_bidi_rq(rq
) &&
1919 __end_that_request_first(rq
->next_rq
, error
, bidi_bytes
))
1927 * blk_end_io - Generic end_io function to complete a request.
1928 * @rq: the request being processed
1929 * @error: %0 for success, < %0 for error
1930 * @nr_bytes: number of bytes to complete @rq
1931 * @bidi_bytes: number of bytes to complete @rq->next_rq
1932 * @drv_callback: function called between completion of bios in the request
1933 * and completion of the request.
1934 * If the callback returns non %0, this helper returns without
1935 * completion of the request.
1938 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
1939 * If @rq has leftover, sets it up for the next range of segments.
1942 * %0 - we are done with this request
1943 * %1 - this request is not freed yet, it still has pending buffers.
1945 static int blk_end_io(struct request
*rq
, int error
, unsigned int nr_bytes
,
1946 unsigned int bidi_bytes
,
1947 int (drv_callback
)(struct request
*))
1949 struct request_queue
*q
= rq
->q
;
1950 unsigned long flags
= 0UL;
1952 if (end_that_request_data(rq
, error
, nr_bytes
, bidi_bytes
))
1955 /* Special feature for tricky drivers */
1956 if (drv_callback
&& drv_callback(rq
))
1959 add_disk_randomness(rq
->rq_disk
);
1961 spin_lock_irqsave(q
->queue_lock
, flags
);
1962 end_that_request_last(rq
, error
);
1963 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1969 * blk_end_request - Helper function for drivers to complete the request.
1970 * @rq: the request being processed
1971 * @error: %0 for success, < %0 for error
1972 * @nr_bytes: number of bytes to complete
1975 * Ends I/O on a number of bytes attached to @rq.
1976 * If @rq has leftover, sets it up for the next range of segments.
1979 * %0 - we are done with this request
1980 * %1 - still buffers pending for this request
1982 int blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
1984 return blk_end_io(rq
, error
, nr_bytes
, 0, NULL
);
1986 EXPORT_SYMBOL_GPL(blk_end_request
);
1989 * __blk_end_request - Helper function for drivers to complete the request.
1990 * @rq: the request being processed
1991 * @error: %0 for success, < %0 for error
1992 * @nr_bytes: number of bytes to complete
1995 * Must be called with queue lock held unlike blk_end_request().
1998 * %0 - we are done with this request
1999 * %1 - still buffers pending for this request
2001 int __blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2003 if (rq
->bio
&& __end_that_request_first(rq
, error
, nr_bytes
))
2006 add_disk_randomness(rq
->rq_disk
);
2008 end_that_request_last(rq
, error
);
2012 EXPORT_SYMBOL_GPL(__blk_end_request
);
2015 * blk_end_bidi_request - Helper function for drivers to complete bidi request.
2016 * @rq: the bidi request being processed
2017 * @error: %0 for success, < %0 for error
2018 * @nr_bytes: number of bytes to complete @rq
2019 * @bidi_bytes: number of bytes to complete @rq->next_rq
2022 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2025 * %0 - we are done with this request
2026 * %1 - still buffers pending for this request
2028 int blk_end_bidi_request(struct request
*rq
, int error
, unsigned int nr_bytes
,
2029 unsigned int bidi_bytes
)
2031 return blk_end_io(rq
, error
, nr_bytes
, bidi_bytes
, NULL
);
2033 EXPORT_SYMBOL_GPL(blk_end_bidi_request
);
2036 * blk_update_request - Special helper function for request stacking drivers
2037 * @rq: the request being processed
2038 * @error: %0 for success, < %0 for error
2039 * @nr_bytes: number of bytes to complete @rq
2042 * Ends I/O on a number of bytes attached to @rq, but doesn't complete
2043 * the request structure even if @rq doesn't have leftover.
2044 * If @rq has leftover, sets it up for the next range of segments.
2046 * This special helper function is only for request stacking drivers
2047 * (e.g. request-based dm) so that they can handle partial completion.
2048 * Actual device drivers should use blk_end_request instead.
2050 void blk_update_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2052 if (!end_that_request_data(rq
, error
, nr_bytes
, 0)) {
2054 * These members are not updated in end_that_request_data()
2055 * when all bios are completed.
2056 * Update them so that the request stacking driver can find
2057 * how many bytes remain in the request later.
2059 rq
->nr_sectors
= rq
->hard_nr_sectors
= 0;
2060 rq
->current_nr_sectors
= rq
->hard_cur_sectors
= 0;
2063 EXPORT_SYMBOL_GPL(blk_update_request
);
2066 * blk_end_request_callback - Special helper function for tricky drivers
2067 * @rq: the request being processed
2068 * @error: %0 for success, < %0 for error
2069 * @nr_bytes: number of bytes to complete
2070 * @drv_callback: function called between completion of bios in the request
2071 * and completion of the request.
2072 * If the callback returns non %0, this helper returns without
2073 * completion of the request.
2076 * Ends I/O on a number of bytes attached to @rq.
2077 * If @rq has leftover, sets it up for the next range of segments.
2079 * This special helper function is used only for existing tricky drivers.
2080 * (e.g. cdrom_newpc_intr() of ide-cd)
2081 * This interface will be removed when such drivers are rewritten.
2082 * Don't use this interface in other places anymore.
2085 * %0 - we are done with this request
2086 * %1 - this request is not freed yet.
2087 * this request still has pending buffers or
2088 * the driver doesn't want to finish this request yet.
2090 int blk_end_request_callback(struct request
*rq
, int error
,
2091 unsigned int nr_bytes
,
2092 int (drv_callback
)(struct request
*))
2094 return blk_end_io(rq
, error
, nr_bytes
, 0, drv_callback
);
2096 EXPORT_SYMBOL_GPL(blk_end_request_callback
);
2098 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
2101 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw, and
2102 we want BIO_RW_AHEAD (bit 1) to imply REQ_FAILFAST (bit 1). */
2103 rq
->cmd_flags
|= (bio
->bi_rw
& 3);
2105 if (bio_has_data(bio
)) {
2106 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2107 rq
->buffer
= bio_data(bio
);
2109 rq
->current_nr_sectors
= bio_cur_sectors(bio
);
2110 rq
->hard_cur_sectors
= rq
->current_nr_sectors
;
2111 rq
->hard_nr_sectors
= rq
->nr_sectors
= bio_sectors(bio
);
2112 rq
->data_len
= bio
->bi_size
;
2114 rq
->bio
= rq
->biotail
= bio
;
2117 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2121 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2122 * @q : the queue of the device being checked
2125 * Check if underlying low-level drivers of a device are busy.
2126 * If the drivers want to export their busy state, they must set own
2127 * exporting function using blk_queue_lld_busy() first.
2129 * Basically, this function is used only by request stacking drivers
2130 * to stop dispatching requests to underlying devices when underlying
2131 * devices are busy. This behavior helps more I/O merging on the queue
2132 * of the request stacking driver and prevents I/O throughput regression
2133 * on burst I/O load.
2136 * 0 - Not busy (The request stacking driver should dispatch request)
2137 * 1 - Busy (The request stacking driver should stop dispatching request)
2139 int blk_lld_busy(struct request_queue
*q
)
2142 return q
->lld_busy_fn(q
);
2146 EXPORT_SYMBOL_GPL(blk_lld_busy
);
2148 int kblockd_schedule_work(struct request_queue
*q
, struct work_struct
*work
)
2150 return queue_work(kblockd_workqueue
, work
);
2152 EXPORT_SYMBOL(kblockd_schedule_work
);
2154 int __init
blk_dev_init(void)
2156 kblockd_workqueue
= create_workqueue("kblockd");
2157 if (!kblockd_workqueue
)
2158 panic("Failed to create kblockd\n");
2160 request_cachep
= kmem_cache_create("blkdev_requests",
2161 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
2163 blk_requestq_cachep
= kmem_cache_create("blkdev_queue",
2164 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);