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/fault-inject.h>
30 #include <linux/list_sort.h>
32 #define CREATE_TRACE_POINTS
33 #include <trace/events/block.h>
37 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap
);
38 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap
);
39 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete
);
41 static int __make_request(struct request_queue
*q
, struct bio
*bio
);
44 * For the allocated request tables
46 static struct kmem_cache
*request_cachep
;
49 * For queue allocation
51 struct kmem_cache
*blk_requestq_cachep
;
54 * Controlling structure to kblockd
56 static struct workqueue_struct
*kblockd_workqueue
;
58 static void drive_stat_acct(struct request
*rq
, int new_io
)
60 struct hd_struct
*part
;
61 int rw
= rq_data_dir(rq
);
64 if (!blk_do_io_stat(rq
))
67 cpu
= part_stat_lock();
71 part_stat_inc(cpu
, part
, merges
[rw
]);
73 part
= disk_map_sector_rcu(rq
->rq_disk
, blk_rq_pos(rq
));
74 if (!hd_struct_try_get(part
)) {
76 * The partition is already being removed,
77 * the request will be accounted on the disk only
79 * We take a reference on disk->part0 although that
80 * partition will never be deleted, so we can treat
81 * it as any other partition.
83 part
= &rq
->rq_disk
->part0
;
86 part_round_stats(cpu
, part
);
87 part_inc_in_flight(part
, rw
);
94 void blk_queue_congestion_threshold(struct request_queue
*q
)
98 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) + 1;
99 if (nr
> q
->nr_requests
)
101 q
->nr_congestion_on
= nr
;
103 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) - (q
->nr_requests
/ 16) - 1;
106 q
->nr_congestion_off
= nr
;
110 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
113 * Locates the passed device's request queue and returns the address of its
116 * Will return NULL if the request queue cannot be located.
118 struct backing_dev_info
*blk_get_backing_dev_info(struct block_device
*bdev
)
120 struct backing_dev_info
*ret
= NULL
;
121 struct request_queue
*q
= bdev_get_queue(bdev
);
124 ret
= &q
->backing_dev_info
;
127 EXPORT_SYMBOL(blk_get_backing_dev_info
);
129 void blk_rq_init(struct request_queue
*q
, struct request
*rq
)
131 memset(rq
, 0, sizeof(*rq
));
133 INIT_LIST_HEAD(&rq
->queuelist
);
134 INIT_LIST_HEAD(&rq
->timeout_list
);
137 rq
->__sector
= (sector_t
) -1;
138 INIT_HLIST_NODE(&rq
->hash
);
139 RB_CLEAR_NODE(&rq
->rb_node
);
141 rq
->cmd_len
= BLK_MAX_CDB
;
144 rq
->start_time
= jiffies
;
145 set_start_time_ns(rq
);
148 EXPORT_SYMBOL(blk_rq_init
);
150 static void req_bio_endio(struct request
*rq
, struct bio
*bio
,
151 unsigned int nbytes
, int error
)
154 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
155 else if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
158 if (unlikely(nbytes
> bio
->bi_size
)) {
159 printk(KERN_ERR
"%s: want %u bytes done, %u left\n",
160 __func__
, nbytes
, bio
->bi_size
);
161 nbytes
= bio
->bi_size
;
164 if (unlikely(rq
->cmd_flags
& REQ_QUIET
))
165 set_bit(BIO_QUIET
, &bio
->bi_flags
);
167 bio
->bi_size
-= nbytes
;
168 bio
->bi_sector
+= (nbytes
>> 9);
170 if (bio_integrity(bio
))
171 bio_integrity_advance(bio
, nbytes
);
173 /* don't actually finish bio if it's part of flush sequence */
174 if (bio
->bi_size
== 0 && !(rq
->cmd_flags
& REQ_FLUSH_SEQ
))
175 bio_endio(bio
, error
);
178 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
182 printk(KERN_INFO
"%s: dev %s: type=%x, flags=%x\n", msg
,
183 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->cmd_type
,
186 printk(KERN_INFO
" sector %llu, nr/cnr %u/%u\n",
187 (unsigned long long)blk_rq_pos(rq
),
188 blk_rq_sectors(rq
), blk_rq_cur_sectors(rq
));
189 printk(KERN_INFO
" bio %p, biotail %p, buffer %p, len %u\n",
190 rq
->bio
, rq
->biotail
, rq
->buffer
, blk_rq_bytes(rq
));
192 if (rq
->cmd_type
== REQ_TYPE_BLOCK_PC
) {
193 printk(KERN_INFO
" cdb: ");
194 for (bit
= 0; bit
< BLK_MAX_CDB
; bit
++)
195 printk("%02x ", rq
->cmd
[bit
]);
199 EXPORT_SYMBOL(blk_dump_rq_flags
);
201 static void blk_delay_work(struct work_struct
*work
)
203 struct request_queue
*q
;
205 q
= container_of(work
, struct request_queue
, delay_work
.work
);
206 spin_lock_irq(q
->queue_lock
);
208 spin_unlock_irq(q
->queue_lock
);
212 * blk_delay_queue - restart queueing after defined interval
213 * @q: The &struct request_queue in question
214 * @msecs: Delay in msecs
217 * Sometimes queueing needs to be postponed for a little while, to allow
218 * resources to come back. This function will make sure that queueing is
219 * restarted around the specified time.
221 void blk_delay_queue(struct request_queue
*q
, unsigned long msecs
)
223 queue_delayed_work(kblockd_workqueue
, &q
->delay_work
,
224 msecs_to_jiffies(msecs
));
226 EXPORT_SYMBOL(blk_delay_queue
);
229 * blk_start_queue - restart a previously stopped queue
230 * @q: The &struct request_queue in question
233 * blk_start_queue() will clear the stop flag on the queue, and call
234 * the request_fn for the queue if it was in a stopped state when
235 * entered. Also see blk_stop_queue(). Queue lock must be held.
237 void blk_start_queue(struct request_queue
*q
)
239 WARN_ON(!irqs_disabled());
241 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
244 EXPORT_SYMBOL(blk_start_queue
);
247 * blk_stop_queue - stop a queue
248 * @q: The &struct request_queue in question
251 * The Linux block layer assumes that a block driver will consume all
252 * entries on the request queue when the request_fn strategy is called.
253 * Often this will not happen, because of hardware limitations (queue
254 * depth settings). If a device driver gets a 'queue full' response,
255 * or if it simply chooses not to queue more I/O at one point, it can
256 * call this function to prevent the request_fn from being called until
257 * the driver has signalled it's ready to go again. This happens by calling
258 * blk_start_queue() to restart queue operations. Queue lock must be held.
260 void blk_stop_queue(struct request_queue
*q
)
262 __cancel_delayed_work(&q
->delay_work
);
263 queue_flag_set(QUEUE_FLAG_STOPPED
, q
);
265 EXPORT_SYMBOL(blk_stop_queue
);
268 * blk_sync_queue - cancel any pending callbacks on a queue
272 * The block layer may perform asynchronous callback activity
273 * on a queue, such as calling the unplug function after a timeout.
274 * A block device may call blk_sync_queue to ensure that any
275 * such activity is cancelled, thus allowing it to release resources
276 * that the callbacks might use. The caller must already have made sure
277 * that its ->make_request_fn will not re-add plugging prior to calling
280 * This function does not cancel any asynchronous activity arising
281 * out of elevator or throttling code. That would require elevaotor_exit()
282 * and blk_throtl_exit() to be called with queue lock initialized.
285 void blk_sync_queue(struct request_queue
*q
)
287 del_timer_sync(&q
->timeout
);
288 cancel_delayed_work_sync(&q
->delay_work
);
290 EXPORT_SYMBOL(blk_sync_queue
);
293 * __blk_run_queue - run a single device queue
294 * @q: The queue to run
297 * See @blk_run_queue. This variant must be called with the queue lock
298 * held and interrupts disabled.
300 void __blk_run_queue(struct request_queue
*q
)
302 if (unlikely(blk_queue_stopped(q
)))
307 EXPORT_SYMBOL(__blk_run_queue
);
310 * blk_run_queue_async - run a single device queue in workqueue context
311 * @q: The queue to run
314 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
317 void blk_run_queue_async(struct request_queue
*q
)
319 if (likely(!blk_queue_stopped(q
))) {
320 __cancel_delayed_work(&q
->delay_work
);
321 queue_delayed_work(kblockd_workqueue
, &q
->delay_work
, 0);
324 EXPORT_SYMBOL(blk_run_queue_async
);
327 * blk_run_queue - run a single device queue
328 * @q: The queue to run
331 * Invoke request handling on this queue, if it has pending work to do.
332 * May be used to restart queueing when a request has completed.
334 void blk_run_queue(struct request_queue
*q
)
338 spin_lock_irqsave(q
->queue_lock
, flags
);
340 spin_unlock_irqrestore(q
->queue_lock
, flags
);
342 EXPORT_SYMBOL(blk_run_queue
);
344 void blk_put_queue(struct request_queue
*q
)
346 kobject_put(&q
->kobj
);
348 EXPORT_SYMBOL(blk_put_queue
);
351 * Note: If a driver supplied the queue lock, it is disconnected
352 * by this function. The actual state of the lock doesn't matter
353 * here as the request_queue isn't accessible after this point
354 * (QUEUE_FLAG_DEAD is set) and no other requests will be queued.
356 void blk_cleanup_queue(struct request_queue
*q
)
359 * We know we have process context here, so we can be a little
360 * cautious and ensure that pending block actions on this device
361 * are done before moving on. Going into this function, we should
362 * not have processes doing IO to this device.
366 del_timer_sync(&q
->backing_dev_info
.laptop_mode_wb_timer
);
367 mutex_lock(&q
->sysfs_lock
);
368 queue_flag_set_unlocked(QUEUE_FLAG_DEAD
, q
);
369 mutex_unlock(&q
->sysfs_lock
);
371 if (q
->queue_lock
!= &q
->__queue_lock
)
372 q
->queue_lock
= &q
->__queue_lock
;
376 EXPORT_SYMBOL(blk_cleanup_queue
);
378 static int blk_init_free_list(struct request_queue
*q
)
380 struct request_list
*rl
= &q
->rq
;
382 if (unlikely(rl
->rq_pool
))
385 rl
->count
[BLK_RW_SYNC
] = rl
->count
[BLK_RW_ASYNC
] = 0;
386 rl
->starved
[BLK_RW_SYNC
] = rl
->starved
[BLK_RW_ASYNC
] = 0;
388 init_waitqueue_head(&rl
->wait
[BLK_RW_SYNC
]);
389 init_waitqueue_head(&rl
->wait
[BLK_RW_ASYNC
]);
391 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, mempool_alloc_slab
,
392 mempool_free_slab
, request_cachep
, q
->node
);
400 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
402 return blk_alloc_queue_node(gfp_mask
, -1);
404 EXPORT_SYMBOL(blk_alloc_queue
);
406 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
408 struct request_queue
*q
;
411 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
412 gfp_mask
| __GFP_ZERO
, node_id
);
416 q
->backing_dev_info
.ra_pages
=
417 (VM_MAX_READAHEAD
* 1024) / PAGE_CACHE_SIZE
;
418 q
->backing_dev_info
.state
= 0;
419 q
->backing_dev_info
.capabilities
= BDI_CAP_MAP_COPY
;
420 q
->backing_dev_info
.name
= "block";
423 err
= bdi_init(&q
->backing_dev_info
);
425 kmem_cache_free(blk_requestq_cachep
, q
);
429 if (blk_throtl_init(q
)) {
430 kmem_cache_free(blk_requestq_cachep
, q
);
434 setup_timer(&q
->backing_dev_info
.laptop_mode_wb_timer
,
435 laptop_mode_timer_fn
, (unsigned long) q
);
436 setup_timer(&q
->timeout
, blk_rq_timed_out_timer
, (unsigned long) q
);
437 INIT_LIST_HEAD(&q
->timeout_list
);
438 INIT_LIST_HEAD(&q
->flush_queue
[0]);
439 INIT_LIST_HEAD(&q
->flush_queue
[1]);
440 INIT_LIST_HEAD(&q
->flush_data_in_flight
);
441 INIT_DELAYED_WORK(&q
->delay_work
, blk_delay_work
);
443 kobject_init(&q
->kobj
, &blk_queue_ktype
);
445 mutex_init(&q
->sysfs_lock
);
446 spin_lock_init(&q
->__queue_lock
);
449 * By default initialize queue_lock to internal lock and driver can
450 * override it later if need be.
452 q
->queue_lock
= &q
->__queue_lock
;
456 EXPORT_SYMBOL(blk_alloc_queue_node
);
459 * blk_init_queue - prepare a request queue for use with a block device
460 * @rfn: The function to be called to process requests that have been
461 * placed on the queue.
462 * @lock: Request queue spin lock
465 * If a block device wishes to use the standard request handling procedures,
466 * which sorts requests and coalesces adjacent requests, then it must
467 * call blk_init_queue(). The function @rfn will be called when there
468 * are requests on the queue that need to be processed. If the device
469 * supports plugging, then @rfn may not be called immediately when requests
470 * are available on the queue, but may be called at some time later instead.
471 * Plugged queues are generally unplugged when a buffer belonging to one
472 * of the requests on the queue is needed, or due to memory pressure.
474 * @rfn is not required, or even expected, to remove all requests off the
475 * queue, but only as many as it can handle at a time. If it does leave
476 * requests on the queue, it is responsible for arranging that the requests
477 * get dealt with eventually.
479 * The queue spin lock must be held while manipulating the requests on the
480 * request queue; this lock will be taken also from interrupt context, so irq
481 * disabling is needed for it.
483 * Function returns a pointer to the initialized request queue, or %NULL if
487 * blk_init_queue() must be paired with a blk_cleanup_queue() call
488 * when the block device is deactivated (such as at module unload).
491 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
493 return blk_init_queue_node(rfn
, lock
, -1);
495 EXPORT_SYMBOL(blk_init_queue
);
497 struct request_queue
*
498 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
500 struct request_queue
*uninit_q
, *q
;
502 uninit_q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
506 q
= blk_init_allocated_queue(uninit_q
, rfn
, lock
);
508 blk_cleanup_queue(uninit_q
);
512 EXPORT_SYMBOL(blk_init_queue_node
);
514 struct request_queue
*
515 blk_init_allocated_queue(struct request_queue
*q
, request_fn_proc
*rfn
,
521 if (blk_init_free_list(q
))
525 q
->prep_rq_fn
= NULL
;
526 q
->unprep_rq_fn
= NULL
;
527 q
->queue_flags
= QUEUE_FLAG_DEFAULT
;
529 /* Override internal queue lock with supplied lock pointer */
531 q
->queue_lock
= lock
;
534 * This also sets hw/phys segments, boundary and size
536 blk_queue_make_request(q
, __make_request
);
538 q
->sg_reserved_size
= INT_MAX
;
543 if (!elevator_init(q
, NULL
)) {
544 blk_queue_congestion_threshold(q
);
550 EXPORT_SYMBOL(blk_init_allocated_queue
);
552 int blk_get_queue(struct request_queue
*q
)
554 if (likely(!test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
))) {
555 kobject_get(&q
->kobj
);
561 EXPORT_SYMBOL(blk_get_queue
);
563 static inline void blk_free_request(struct request_queue
*q
, struct request
*rq
)
565 if (rq
->cmd_flags
& REQ_ELVPRIV
)
566 elv_put_request(q
, rq
);
567 mempool_free(rq
, q
->rq
.rq_pool
);
570 static struct request
*
571 blk_alloc_request(struct request_queue
*q
, int flags
, int priv
, gfp_t gfp_mask
)
573 struct request
*rq
= mempool_alloc(q
->rq
.rq_pool
, gfp_mask
);
580 rq
->cmd_flags
= flags
| REQ_ALLOCED
;
583 if (unlikely(elv_set_request(q
, rq
, gfp_mask
))) {
584 mempool_free(rq
, q
->rq
.rq_pool
);
587 rq
->cmd_flags
|= REQ_ELVPRIV
;
594 * ioc_batching returns true if the ioc is a valid batching request and
595 * should be given priority access to a request.
597 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
603 * Make sure the process is able to allocate at least 1 request
604 * even if the batch times out, otherwise we could theoretically
607 return ioc
->nr_batch_requests
== q
->nr_batching
||
608 (ioc
->nr_batch_requests
> 0
609 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
613 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
614 * will cause the process to be a "batcher" on all queues in the system. This
615 * is the behaviour we want though - once it gets a wakeup it should be given
618 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
620 if (!ioc
|| ioc_batching(q
, ioc
))
623 ioc
->nr_batch_requests
= q
->nr_batching
;
624 ioc
->last_waited
= jiffies
;
627 static void __freed_request(struct request_queue
*q
, int sync
)
629 struct request_list
*rl
= &q
->rq
;
631 if (rl
->count
[sync
] < queue_congestion_off_threshold(q
))
632 blk_clear_queue_congested(q
, sync
);
634 if (rl
->count
[sync
] + 1 <= q
->nr_requests
) {
635 if (waitqueue_active(&rl
->wait
[sync
]))
636 wake_up(&rl
->wait
[sync
]);
638 blk_clear_queue_full(q
, sync
);
643 * A request has just been released. Account for it, update the full and
644 * congestion status, wake up any waiters. Called under q->queue_lock.
646 static void freed_request(struct request_queue
*q
, int sync
, int priv
)
648 struct request_list
*rl
= &q
->rq
;
654 __freed_request(q
, sync
);
656 if (unlikely(rl
->starved
[sync
^ 1]))
657 __freed_request(q
, sync
^ 1);
661 * Determine if elevator data should be initialized when allocating the
662 * request associated with @bio.
664 static bool blk_rq_should_init_elevator(struct bio
*bio
)
670 * Flush requests do not use the elevator so skip initialization.
671 * This allows a request to share the flush and elevator data.
673 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
))
680 * Get a free request, queue_lock must be held.
681 * Returns NULL on failure, with queue_lock held.
682 * Returns !NULL on success, with queue_lock *not held*.
684 static struct request
*get_request(struct request_queue
*q
, int rw_flags
,
685 struct bio
*bio
, gfp_t gfp_mask
)
687 struct request
*rq
= NULL
;
688 struct request_list
*rl
= &q
->rq
;
689 struct io_context
*ioc
= NULL
;
690 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
691 int may_queue
, priv
= 0;
693 may_queue
= elv_may_queue(q
, rw_flags
);
694 if (may_queue
== ELV_MQUEUE_NO
)
697 if (rl
->count
[is_sync
]+1 >= queue_congestion_on_threshold(q
)) {
698 if (rl
->count
[is_sync
]+1 >= q
->nr_requests
) {
699 ioc
= current_io_context(GFP_ATOMIC
, q
->node
);
701 * The queue will fill after this allocation, so set
702 * it as full, and mark this process as "batching".
703 * This process will be allowed to complete a batch of
704 * requests, others will be blocked.
706 if (!blk_queue_full(q
, is_sync
)) {
707 ioc_set_batching(q
, ioc
);
708 blk_set_queue_full(q
, is_sync
);
710 if (may_queue
!= ELV_MQUEUE_MUST
711 && !ioc_batching(q
, ioc
)) {
713 * The queue is full and the allocating
714 * process is not a "batcher", and not
715 * exempted by the IO scheduler
721 blk_set_queue_congested(q
, is_sync
);
725 * Only allow batching queuers to allocate up to 50% over the defined
726 * limit of requests, otherwise we could have thousands of requests
727 * allocated with any setting of ->nr_requests
729 if (rl
->count
[is_sync
] >= (3 * q
->nr_requests
/ 2))
732 rl
->count
[is_sync
]++;
733 rl
->starved
[is_sync
] = 0;
735 if (blk_rq_should_init_elevator(bio
)) {
736 priv
= !test_bit(QUEUE_FLAG_ELVSWITCH
, &q
->queue_flags
);
741 if (blk_queue_io_stat(q
))
742 rw_flags
|= REQ_IO_STAT
;
743 spin_unlock_irq(q
->queue_lock
);
745 rq
= blk_alloc_request(q
, rw_flags
, priv
, gfp_mask
);
748 * Allocation failed presumably due to memory. Undo anything
749 * we might have messed up.
751 * Allocating task should really be put onto the front of the
752 * wait queue, but this is pretty rare.
754 spin_lock_irq(q
->queue_lock
);
755 freed_request(q
, is_sync
, priv
);
758 * in the very unlikely event that allocation failed and no
759 * requests for this direction was pending, mark us starved
760 * so that freeing of a request in the other direction will
761 * notice us. another possible fix would be to split the
762 * rq mempool into READ and WRITE
765 if (unlikely(rl
->count
[is_sync
] == 0))
766 rl
->starved
[is_sync
] = 1;
772 * ioc may be NULL here, and ioc_batching will be false. That's
773 * OK, if the queue is under the request limit then requests need
774 * not count toward the nr_batch_requests limit. There will always
775 * be some limit enforced by BLK_BATCH_TIME.
777 if (ioc_batching(q
, ioc
))
778 ioc
->nr_batch_requests
--;
780 trace_block_getrq(q
, bio
, rw_flags
& 1);
786 * No available requests for this queue, wait for some requests to become
789 * Called with q->queue_lock held, and returns with it unlocked.
791 static struct request
*get_request_wait(struct request_queue
*q
, int rw_flags
,
794 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
797 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
800 struct io_context
*ioc
;
801 struct request_list
*rl
= &q
->rq
;
803 prepare_to_wait_exclusive(&rl
->wait
[is_sync
], &wait
,
804 TASK_UNINTERRUPTIBLE
);
806 trace_block_sleeprq(q
, bio
, rw_flags
& 1);
808 spin_unlock_irq(q
->queue_lock
);
812 * After sleeping, we become a "batching" process and
813 * will be able to allocate at least one request, and
814 * up to a big batch of them for a small period time.
815 * See ioc_batching, ioc_set_batching
817 ioc
= current_io_context(GFP_NOIO
, q
->node
);
818 ioc_set_batching(q
, ioc
);
820 spin_lock_irq(q
->queue_lock
);
821 finish_wait(&rl
->wait
[is_sync
], &wait
);
823 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
829 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
833 if (unlikely(test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)))
836 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
838 spin_lock_irq(q
->queue_lock
);
839 if (gfp_mask
& __GFP_WAIT
) {
840 rq
= get_request_wait(q
, rw
, NULL
);
842 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
844 spin_unlock_irq(q
->queue_lock
);
846 /* q->queue_lock is unlocked at this point */
850 EXPORT_SYMBOL(blk_get_request
);
853 * blk_make_request - given a bio, allocate a corresponding struct request.
854 * @q: target request queue
855 * @bio: The bio describing the memory mappings that will be submitted for IO.
856 * It may be a chained-bio properly constructed by block/bio layer.
857 * @gfp_mask: gfp flags to be used for memory allocation
859 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
860 * type commands. Where the struct request needs to be farther initialized by
861 * the caller. It is passed a &struct bio, which describes the memory info of
864 * The caller of blk_make_request must make sure that bi_io_vec
865 * are set to describe the memory buffers. That bio_data_dir() will return
866 * the needed direction of the request. (And all bio's in the passed bio-chain
867 * are properly set accordingly)
869 * If called under none-sleepable conditions, mapped bio buffers must not
870 * need bouncing, by calling the appropriate masked or flagged allocator,
871 * suitable for the target device. Otherwise the call to blk_queue_bounce will
874 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
875 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
876 * anything but the first bio in the chain. Otherwise you risk waiting for IO
877 * completion of a bio that hasn't been submitted yet, thus resulting in a
878 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
879 * of bio_alloc(), as that avoids the mempool deadlock.
880 * If possible a big IO should be split into smaller parts when allocation
881 * fails. Partial allocation should not be an error, or you risk a live-lock.
883 struct request
*blk_make_request(struct request_queue
*q
, struct bio
*bio
,
886 struct request
*rq
= blk_get_request(q
, bio_data_dir(bio
), gfp_mask
);
889 return ERR_PTR(-ENOMEM
);
892 struct bio
*bounce_bio
= bio
;
895 blk_queue_bounce(q
, &bounce_bio
);
896 ret
= blk_rq_append_bio(q
, rq
, bounce_bio
);
905 EXPORT_SYMBOL(blk_make_request
);
908 * blk_requeue_request - put a request back on queue
909 * @q: request queue where request should be inserted
910 * @rq: request to be inserted
913 * Drivers often keep queueing requests until the hardware cannot accept
914 * more, when that condition happens we need to put the request back
915 * on the queue. Must be called with queue lock held.
917 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
919 blk_delete_timer(rq
);
920 blk_clear_rq_complete(rq
);
921 trace_block_rq_requeue(q
, rq
);
923 if (blk_rq_tagged(rq
))
924 blk_queue_end_tag(q
, rq
);
926 BUG_ON(blk_queued_rq(rq
));
928 elv_requeue_request(q
, rq
);
930 EXPORT_SYMBOL(blk_requeue_request
);
932 static void add_acct_request(struct request_queue
*q
, struct request
*rq
,
935 drive_stat_acct(rq
, 1);
936 __elv_add_request(q
, rq
, where
);
940 * blk_insert_request - insert a special request into a request queue
941 * @q: request queue where request should be inserted
942 * @rq: request to be inserted
943 * @at_head: insert request at head or tail of queue
944 * @data: private data
947 * Many block devices need to execute commands asynchronously, so they don't
948 * block the whole kernel from preemption during request execution. This is
949 * accomplished normally by inserting aritficial requests tagged as
950 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
951 * be scheduled for actual execution by the request queue.
953 * We have the option of inserting the head or the tail of the queue.
954 * Typically we use the tail for new ioctls and so forth. We use the head
955 * of the queue for things like a QUEUE_FULL message from a device, or a
956 * host that is unable to accept a particular command.
958 void blk_insert_request(struct request_queue
*q
, struct request
*rq
,
959 int at_head
, void *data
)
961 int where
= at_head
? ELEVATOR_INSERT_FRONT
: ELEVATOR_INSERT_BACK
;
965 * tell I/O scheduler that this isn't a regular read/write (ie it
966 * must not attempt merges on this) and that it acts as a soft
969 rq
->cmd_type
= REQ_TYPE_SPECIAL
;
973 spin_lock_irqsave(q
->queue_lock
, flags
);
976 * If command is tagged, release the tag
978 if (blk_rq_tagged(rq
))
979 blk_queue_end_tag(q
, rq
);
981 add_acct_request(q
, rq
, where
);
983 spin_unlock_irqrestore(q
->queue_lock
, flags
);
985 EXPORT_SYMBOL(blk_insert_request
);
987 static void part_round_stats_single(int cpu
, struct hd_struct
*part
,
990 if (now
== part
->stamp
)
993 if (part_in_flight(part
)) {
994 __part_stat_add(cpu
, part
, time_in_queue
,
995 part_in_flight(part
) * (now
- part
->stamp
));
996 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
1002 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1003 * @cpu: cpu number for stats access
1004 * @part: target partition
1006 * The average IO queue length and utilisation statistics are maintained
1007 * by observing the current state of the queue length and the amount of
1008 * time it has been in this state for.
1010 * Normally, that accounting is done on IO completion, but that can result
1011 * in more than a second's worth of IO being accounted for within any one
1012 * second, leading to >100% utilisation. To deal with that, we call this
1013 * function to do a round-off before returning the results when reading
1014 * /proc/diskstats. This accounts immediately for all queue usage up to
1015 * the current jiffies and restarts the counters again.
1017 void part_round_stats(int cpu
, struct hd_struct
*part
)
1019 unsigned long now
= jiffies
;
1022 part_round_stats_single(cpu
, &part_to_disk(part
)->part0
, now
);
1023 part_round_stats_single(cpu
, part
, now
);
1025 EXPORT_SYMBOL_GPL(part_round_stats
);
1028 * queue lock must be held
1030 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1034 if (unlikely(--req
->ref_count
))
1037 elv_completed_request(q
, req
);
1039 /* this is a bio leak */
1040 WARN_ON(req
->bio
!= NULL
);
1043 * Request may not have originated from ll_rw_blk. if not,
1044 * it didn't come out of our reserved rq pools
1046 if (req
->cmd_flags
& REQ_ALLOCED
) {
1047 int is_sync
= rq_is_sync(req
) != 0;
1048 int priv
= req
->cmd_flags
& REQ_ELVPRIV
;
1050 BUG_ON(!list_empty(&req
->queuelist
));
1051 BUG_ON(!hlist_unhashed(&req
->hash
));
1053 blk_free_request(q
, req
);
1054 freed_request(q
, is_sync
, priv
);
1057 EXPORT_SYMBOL_GPL(__blk_put_request
);
1059 void blk_put_request(struct request
*req
)
1061 unsigned long flags
;
1062 struct request_queue
*q
= req
->q
;
1064 spin_lock_irqsave(q
->queue_lock
, flags
);
1065 __blk_put_request(q
, req
);
1066 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1068 EXPORT_SYMBOL(blk_put_request
);
1071 * blk_add_request_payload - add a payload to a request
1072 * @rq: request to update
1073 * @page: page backing the payload
1074 * @len: length of the payload.
1076 * This allows to later add a payload to an already submitted request by
1077 * a block driver. The driver needs to take care of freeing the payload
1080 * Note that this is a quite horrible hack and nothing but handling of
1081 * discard requests should ever use it.
1083 void blk_add_request_payload(struct request
*rq
, struct page
*page
,
1086 struct bio
*bio
= rq
->bio
;
1088 bio
->bi_io_vec
->bv_page
= page
;
1089 bio
->bi_io_vec
->bv_offset
= 0;
1090 bio
->bi_io_vec
->bv_len
= len
;
1094 bio
->bi_phys_segments
= 1;
1096 rq
->__data_len
= rq
->resid_len
= len
;
1097 rq
->nr_phys_segments
= 1;
1098 rq
->buffer
= bio_data(bio
);
1100 EXPORT_SYMBOL_GPL(blk_add_request_payload
);
1102 static bool bio_attempt_back_merge(struct request_queue
*q
, struct request
*req
,
1105 const int ff
= bio
->bi_rw
& REQ_FAILFAST_MASK
;
1107 if (!ll_back_merge_fn(q
, req
, bio
))
1110 trace_block_bio_backmerge(q
, bio
);
1112 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1113 blk_rq_set_mixed_merge(req
);
1115 req
->biotail
->bi_next
= bio
;
1117 req
->__data_len
+= bio
->bi_size
;
1118 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1120 drive_stat_acct(req
, 0);
1121 elv_bio_merged(q
, req
, bio
);
1125 static bool bio_attempt_front_merge(struct request_queue
*q
,
1126 struct request
*req
, struct bio
*bio
)
1128 const int ff
= bio
->bi_rw
& REQ_FAILFAST_MASK
;
1130 if (!ll_front_merge_fn(q
, req
, bio
))
1133 trace_block_bio_frontmerge(q
, bio
);
1135 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1136 blk_rq_set_mixed_merge(req
);
1138 bio
->bi_next
= req
->bio
;
1142 * may not be valid. if the low level driver said
1143 * it didn't need a bounce buffer then it better
1144 * not touch req->buffer either...
1146 req
->buffer
= bio_data(bio
);
1147 req
->__sector
= bio
->bi_sector
;
1148 req
->__data_len
+= bio
->bi_size
;
1149 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1151 drive_stat_acct(req
, 0);
1152 elv_bio_merged(q
, req
, bio
);
1157 * Attempts to merge with the plugged list in the current process. Returns
1158 * true if merge was successful, otherwise false.
1160 static bool attempt_plug_merge(struct task_struct
*tsk
, struct request_queue
*q
,
1163 struct blk_plug
*plug
;
1171 list_for_each_entry_reverse(rq
, &plug
->list
, queuelist
) {
1177 el_ret
= elv_try_merge(rq
, bio
);
1178 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1179 ret
= bio_attempt_back_merge(q
, rq
, bio
);
1182 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1183 ret
= bio_attempt_front_merge(q
, rq
, bio
);
1192 void init_request_from_bio(struct request
*req
, struct bio
*bio
)
1194 req
->cpu
= bio
->bi_comp_cpu
;
1195 req
->cmd_type
= REQ_TYPE_FS
;
1197 req
->cmd_flags
|= bio
->bi_rw
& REQ_COMMON_MASK
;
1198 if (bio
->bi_rw
& REQ_RAHEAD
)
1199 req
->cmd_flags
|= REQ_FAILFAST_MASK
;
1202 req
->__sector
= bio
->bi_sector
;
1203 req
->ioprio
= bio_prio(bio
);
1204 blk_rq_bio_prep(req
->q
, req
, bio
);
1207 static int __make_request(struct request_queue
*q
, struct bio
*bio
)
1209 const bool sync
= !!(bio
->bi_rw
& REQ_SYNC
);
1210 struct blk_plug
*plug
;
1211 int el_ret
, rw_flags
, where
= ELEVATOR_INSERT_SORT
;
1212 struct request
*req
;
1215 * low level driver can indicate that it wants pages above a
1216 * certain limit bounced to low memory (ie for highmem, or even
1217 * ISA dma in theory)
1219 blk_queue_bounce(q
, &bio
);
1221 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) {
1222 spin_lock_irq(q
->queue_lock
);
1223 where
= ELEVATOR_INSERT_FLUSH
;
1228 * Check if we can merge with the plugged list before grabbing
1231 if (attempt_plug_merge(current
, q
, bio
))
1234 spin_lock_irq(q
->queue_lock
);
1236 el_ret
= elv_merge(q
, &req
, bio
);
1237 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1238 if (bio_attempt_back_merge(q
, req
, bio
)) {
1239 if (!attempt_back_merge(q
, req
))
1240 elv_merged_request(q
, req
, el_ret
);
1243 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1244 if (bio_attempt_front_merge(q
, req
, bio
)) {
1245 if (!attempt_front_merge(q
, req
))
1246 elv_merged_request(q
, req
, el_ret
);
1253 * This sync check and mask will be re-done in init_request_from_bio(),
1254 * but we need to set it earlier to expose the sync flag to the
1255 * rq allocator and io schedulers.
1257 rw_flags
= bio_data_dir(bio
);
1259 rw_flags
|= REQ_SYNC
;
1262 * Grab a free request. This is might sleep but can not fail.
1263 * Returns with the queue unlocked.
1265 req
= get_request_wait(q
, rw_flags
, bio
);
1268 * After dropping the lock and possibly sleeping here, our request
1269 * may now be mergeable after it had proven unmergeable (above).
1270 * We don't worry about that case for efficiency. It won't happen
1271 * often, and the elevators are able to handle it.
1273 init_request_from_bio(req
, bio
);
1275 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
) ||
1276 bio_flagged(bio
, BIO_CPU_AFFINE
)) {
1277 req
->cpu
= blk_cpu_to_group(get_cpu());
1281 plug
= current
->plug
;
1284 * If this is the first request added after a plug, fire
1285 * of a plug trace. If others have been added before, check
1286 * if we have multiple devices in this plug. If so, make a
1287 * note to sort the list before dispatch.
1289 if (list_empty(&plug
->list
))
1290 trace_block_plug(q
);
1291 else if (!plug
->should_sort
) {
1292 struct request
*__rq
;
1294 __rq
= list_entry_rq(plug
->list
.prev
);
1296 plug
->should_sort
= 1;
1298 list_add_tail(&req
->queuelist
, &plug
->list
);
1299 drive_stat_acct(req
, 1);
1301 spin_lock_irq(q
->queue_lock
);
1302 add_acct_request(q
, req
, where
);
1305 spin_unlock_irq(q
->queue_lock
);
1312 * If bio->bi_dev is a partition, remap the location
1314 static inline void blk_partition_remap(struct bio
*bio
)
1316 struct block_device
*bdev
= bio
->bi_bdev
;
1318 if (bio_sectors(bio
) && bdev
!= bdev
->bd_contains
) {
1319 struct hd_struct
*p
= bdev
->bd_part
;
1321 bio
->bi_sector
+= p
->start_sect
;
1322 bio
->bi_bdev
= bdev
->bd_contains
;
1324 trace_block_bio_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
1326 bio
->bi_sector
- p
->start_sect
);
1330 static void handle_bad_sector(struct bio
*bio
)
1332 char b
[BDEVNAME_SIZE
];
1334 printk(KERN_INFO
"attempt to access beyond end of device\n");
1335 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
1336 bdevname(bio
->bi_bdev
, b
),
1338 (unsigned long long)bio
->bi_sector
+ bio_sectors(bio
),
1339 (long long)(i_size_read(bio
->bi_bdev
->bd_inode
) >> 9));
1341 set_bit(BIO_EOF
, &bio
->bi_flags
);
1344 #ifdef CONFIG_FAIL_MAKE_REQUEST
1346 static DECLARE_FAULT_ATTR(fail_make_request
);
1348 static int __init
setup_fail_make_request(char *str
)
1350 return setup_fault_attr(&fail_make_request
, str
);
1352 __setup("fail_make_request=", setup_fail_make_request
);
1354 static int should_fail_request(struct bio
*bio
)
1356 struct hd_struct
*part
= bio
->bi_bdev
->bd_part
;
1358 if (part_to_disk(part
)->part0
.make_it_fail
|| part
->make_it_fail
)
1359 return should_fail(&fail_make_request
, bio
->bi_size
);
1364 static int __init
fail_make_request_debugfs(void)
1366 return init_fault_attr_dentries(&fail_make_request
,
1367 "fail_make_request");
1370 late_initcall(fail_make_request_debugfs
);
1372 #else /* CONFIG_FAIL_MAKE_REQUEST */
1374 static inline int should_fail_request(struct bio
*bio
)
1379 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1382 * Check whether this bio extends beyond the end of the device.
1384 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
1391 /* Test device or partition size, when known. */
1392 maxsector
= i_size_read(bio
->bi_bdev
->bd_inode
) >> 9;
1394 sector_t sector
= bio
->bi_sector
;
1396 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
1398 * This may well happen - the kernel calls bread()
1399 * without checking the size of the device, e.g., when
1400 * mounting a device.
1402 handle_bad_sector(bio
);
1411 * generic_make_request - hand a buffer to its device driver for I/O
1412 * @bio: The bio describing the location in memory and on the device.
1414 * generic_make_request() is used to make I/O requests of block
1415 * devices. It is passed a &struct bio, which describes the I/O that needs
1418 * generic_make_request() does not return any status. The
1419 * success/failure status of the request, along with notification of
1420 * completion, is delivered asynchronously through the bio->bi_end_io
1421 * function described (one day) else where.
1423 * The caller of generic_make_request must make sure that bi_io_vec
1424 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1425 * set to describe the device address, and the
1426 * bi_end_io and optionally bi_private are set to describe how
1427 * completion notification should be signaled.
1429 * generic_make_request and the drivers it calls may use bi_next if this
1430 * bio happens to be merged with someone else, and may change bi_dev and
1431 * bi_sector for remaps as it sees fit. So the values of these fields
1432 * should NOT be depended on after the call to generic_make_request.
1434 static inline void __generic_make_request(struct bio
*bio
)
1436 struct request_queue
*q
;
1437 sector_t old_sector
;
1438 int ret
, nr_sectors
= bio_sectors(bio
);
1444 if (bio_check_eod(bio
, nr_sectors
))
1448 * Resolve the mapping until finished. (drivers are
1449 * still free to implement/resolve their own stacking
1450 * by explicitly returning 0)
1452 * NOTE: we don't repeat the blk_size check for each new device.
1453 * Stacking drivers are expected to know what they are doing.
1458 char b
[BDEVNAME_SIZE
];
1460 q
= bdev_get_queue(bio
->bi_bdev
);
1463 "generic_make_request: Trying to access "
1464 "nonexistent block-device %s (%Lu)\n",
1465 bdevname(bio
->bi_bdev
, b
),
1466 (long long) bio
->bi_sector
);
1470 if (unlikely(!(bio
->bi_rw
& REQ_DISCARD
) &&
1471 nr_sectors
> queue_max_hw_sectors(q
))) {
1472 printk(KERN_ERR
"bio too big device %s (%u > %u)\n",
1473 bdevname(bio
->bi_bdev
, b
),
1475 queue_max_hw_sectors(q
));
1479 if (unlikely(test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)))
1482 if (should_fail_request(bio
))
1486 * If this device has partitions, remap block n
1487 * of partition p to block n+start(p) of the disk.
1489 blk_partition_remap(bio
);
1491 if (bio_integrity_enabled(bio
) && bio_integrity_prep(bio
))
1494 if (old_sector
!= -1)
1495 trace_block_bio_remap(q
, bio
, old_dev
, old_sector
);
1497 old_sector
= bio
->bi_sector
;
1498 old_dev
= bio
->bi_bdev
->bd_dev
;
1500 if (bio_check_eod(bio
, nr_sectors
))
1504 * Filter flush bio's early so that make_request based
1505 * drivers without flush support don't have to worry
1508 if ((bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) && !q
->flush_flags
) {
1509 bio
->bi_rw
&= ~(REQ_FLUSH
| REQ_FUA
);
1516 if ((bio
->bi_rw
& REQ_DISCARD
) &&
1517 (!blk_queue_discard(q
) ||
1518 ((bio
->bi_rw
& REQ_SECURE
) &&
1519 !blk_queue_secdiscard(q
)))) {
1524 if (blk_throtl_bio(q
, &bio
))
1528 * If bio = NULL, bio has been throttled and will be submitted
1534 trace_block_bio_queue(q
, bio
);
1536 ret
= q
->make_request_fn(q
, bio
);
1542 bio_endio(bio
, err
);
1546 * We only want one ->make_request_fn to be active at a time,
1547 * else stack usage with stacked devices could be a problem.
1548 * So use current->bio_list to keep a list of requests
1549 * submited by a make_request_fn function.
1550 * current->bio_list is also used as a flag to say if
1551 * generic_make_request is currently active in this task or not.
1552 * If it is NULL, then no make_request is active. If it is non-NULL,
1553 * then a make_request is active, and new requests should be added
1556 void generic_make_request(struct bio
*bio
)
1558 struct bio_list bio_list_on_stack
;
1560 if (current
->bio_list
) {
1561 /* make_request is active */
1562 bio_list_add(current
->bio_list
, bio
);
1565 /* following loop may be a bit non-obvious, and so deserves some
1567 * Before entering the loop, bio->bi_next is NULL (as all callers
1568 * ensure that) so we have a list with a single bio.
1569 * We pretend that we have just taken it off a longer list, so
1570 * we assign bio_list to a pointer to the bio_list_on_stack,
1571 * thus initialising the bio_list of new bios to be
1572 * added. __generic_make_request may indeed add some more bios
1573 * through a recursive call to generic_make_request. If it
1574 * did, we find a non-NULL value in bio_list and re-enter the loop
1575 * from the top. In this case we really did just take the bio
1576 * of the top of the list (no pretending) and so remove it from
1577 * bio_list, and call into __generic_make_request again.
1579 * The loop was structured like this to make only one call to
1580 * __generic_make_request (which is important as it is large and
1581 * inlined) and to keep the structure simple.
1583 BUG_ON(bio
->bi_next
);
1584 bio_list_init(&bio_list_on_stack
);
1585 current
->bio_list
= &bio_list_on_stack
;
1587 __generic_make_request(bio
);
1588 bio
= bio_list_pop(current
->bio_list
);
1590 current
->bio_list
= NULL
; /* deactivate */
1592 EXPORT_SYMBOL(generic_make_request
);
1595 * submit_bio - submit a bio to the block device layer for I/O
1596 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1597 * @bio: The &struct bio which describes the I/O
1599 * submit_bio() is very similar in purpose to generic_make_request(), and
1600 * uses that function to do most of the work. Both are fairly rough
1601 * interfaces; @bio must be presetup and ready for I/O.
1604 void submit_bio(int rw
, struct bio
*bio
)
1606 int count
= bio_sectors(bio
);
1611 * If it's a regular read/write or a barrier with data attached,
1612 * go through the normal accounting stuff before submission.
1614 if (bio_has_data(bio
) && !(rw
& REQ_DISCARD
)) {
1616 count_vm_events(PGPGOUT
, count
);
1618 task_io_account_read(bio
->bi_size
);
1619 count_vm_events(PGPGIN
, count
);
1622 if (unlikely(block_dump
)) {
1623 char b
[BDEVNAME_SIZE
];
1624 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s (%u sectors)\n",
1625 current
->comm
, task_pid_nr(current
),
1626 (rw
& WRITE
) ? "WRITE" : "READ",
1627 (unsigned long long)bio
->bi_sector
,
1628 bdevname(bio
->bi_bdev
, b
),
1633 generic_make_request(bio
);
1635 EXPORT_SYMBOL(submit_bio
);
1638 * blk_rq_check_limits - Helper function to check a request for the queue limit
1640 * @rq: the request being checked
1643 * @rq may have been made based on weaker limitations of upper-level queues
1644 * in request stacking drivers, and it may violate the limitation of @q.
1645 * Since the block layer and the underlying device driver trust @rq
1646 * after it is inserted to @q, it should be checked against @q before
1647 * the insertion using this generic function.
1649 * This function should also be useful for request stacking drivers
1650 * in some cases below, so export this function.
1651 * Request stacking drivers like request-based dm may change the queue
1652 * limits while requests are in the queue (e.g. dm's table swapping).
1653 * Such request stacking drivers should check those requests agaist
1654 * the new queue limits again when they dispatch those requests,
1655 * although such checkings are also done against the old queue limits
1656 * when submitting requests.
1658 int blk_rq_check_limits(struct request_queue
*q
, struct request
*rq
)
1660 if (rq
->cmd_flags
& REQ_DISCARD
)
1663 if (blk_rq_sectors(rq
) > queue_max_sectors(q
) ||
1664 blk_rq_bytes(rq
) > queue_max_hw_sectors(q
) << 9) {
1665 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
1670 * queue's settings related to segment counting like q->bounce_pfn
1671 * may differ from that of other stacking queues.
1672 * Recalculate it to check the request correctly on this queue's
1675 blk_recalc_rq_segments(rq
);
1676 if (rq
->nr_phys_segments
> queue_max_segments(q
)) {
1677 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
1683 EXPORT_SYMBOL_GPL(blk_rq_check_limits
);
1686 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1687 * @q: the queue to submit the request
1688 * @rq: the request being queued
1690 int blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
1692 unsigned long flags
;
1694 if (blk_rq_check_limits(q
, rq
))
1697 #ifdef CONFIG_FAIL_MAKE_REQUEST
1698 if (rq
->rq_disk
&& rq
->rq_disk
->part0
.make_it_fail
&&
1699 should_fail(&fail_make_request
, blk_rq_bytes(rq
)))
1703 spin_lock_irqsave(q
->queue_lock
, flags
);
1706 * Submitting request must be dequeued before calling this function
1707 * because it will be linked to another request_queue
1709 BUG_ON(blk_queued_rq(rq
));
1711 add_acct_request(q
, rq
, ELEVATOR_INSERT_BACK
);
1712 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1716 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
1719 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1720 * @rq: request to examine
1723 * A request could be merge of IOs which require different failure
1724 * handling. This function determines the number of bytes which
1725 * can be failed from the beginning of the request without
1726 * crossing into area which need to be retried further.
1729 * The number of bytes to fail.
1732 * queue_lock must be held.
1734 unsigned int blk_rq_err_bytes(const struct request
*rq
)
1736 unsigned int ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
1737 unsigned int bytes
= 0;
1740 if (!(rq
->cmd_flags
& REQ_MIXED_MERGE
))
1741 return blk_rq_bytes(rq
);
1744 * Currently the only 'mixing' which can happen is between
1745 * different fastfail types. We can safely fail portions
1746 * which have all the failfast bits that the first one has -
1747 * the ones which are at least as eager to fail as the first
1750 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
1751 if ((bio
->bi_rw
& ff
) != ff
)
1753 bytes
+= bio
->bi_size
;
1756 /* this could lead to infinite loop */
1757 BUG_ON(blk_rq_bytes(rq
) && !bytes
);
1760 EXPORT_SYMBOL_GPL(blk_rq_err_bytes
);
1762 static void blk_account_io_completion(struct request
*req
, unsigned int bytes
)
1764 if (blk_do_io_stat(req
)) {
1765 const int rw
= rq_data_dir(req
);
1766 struct hd_struct
*part
;
1769 cpu
= part_stat_lock();
1771 part_stat_add(cpu
, part
, sectors
[rw
], bytes
>> 9);
1776 static void blk_account_io_done(struct request
*req
)
1779 * Account IO completion. flush_rq isn't accounted as a
1780 * normal IO on queueing nor completion. Accounting the
1781 * containing request is enough.
1783 if (blk_do_io_stat(req
) && !(req
->cmd_flags
& REQ_FLUSH_SEQ
)) {
1784 unsigned long duration
= jiffies
- req
->start_time
;
1785 const int rw
= rq_data_dir(req
);
1786 struct hd_struct
*part
;
1789 cpu
= part_stat_lock();
1792 part_stat_inc(cpu
, part
, ios
[rw
]);
1793 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
1794 part_round_stats(cpu
, part
);
1795 part_dec_in_flight(part
, rw
);
1797 hd_struct_put(part
);
1803 * blk_peek_request - peek at the top of a request queue
1804 * @q: request queue to peek at
1807 * Return the request at the top of @q. The returned request
1808 * should be started using blk_start_request() before LLD starts
1812 * Pointer to the request at the top of @q if available. Null
1816 * queue_lock must be held.
1818 struct request
*blk_peek_request(struct request_queue
*q
)
1823 while ((rq
= __elv_next_request(q
)) != NULL
) {
1824 if (!(rq
->cmd_flags
& REQ_STARTED
)) {
1826 * This is the first time the device driver
1827 * sees this request (possibly after
1828 * requeueing). Notify IO scheduler.
1830 if (rq
->cmd_flags
& REQ_SORTED
)
1831 elv_activate_rq(q
, rq
);
1834 * just mark as started even if we don't start
1835 * it, a request that has been delayed should
1836 * not be passed by new incoming requests
1838 rq
->cmd_flags
|= REQ_STARTED
;
1839 trace_block_rq_issue(q
, rq
);
1842 if (!q
->boundary_rq
|| q
->boundary_rq
== rq
) {
1843 q
->end_sector
= rq_end_sector(rq
);
1844 q
->boundary_rq
= NULL
;
1847 if (rq
->cmd_flags
& REQ_DONTPREP
)
1850 if (q
->dma_drain_size
&& blk_rq_bytes(rq
)) {
1852 * make sure space for the drain appears we
1853 * know we can do this because max_hw_segments
1854 * has been adjusted to be one fewer than the
1857 rq
->nr_phys_segments
++;
1863 ret
= q
->prep_rq_fn(q
, rq
);
1864 if (ret
== BLKPREP_OK
) {
1866 } else if (ret
== BLKPREP_DEFER
) {
1868 * the request may have been (partially) prepped.
1869 * we need to keep this request in the front to
1870 * avoid resource deadlock. REQ_STARTED will
1871 * prevent other fs requests from passing this one.
1873 if (q
->dma_drain_size
&& blk_rq_bytes(rq
) &&
1874 !(rq
->cmd_flags
& REQ_DONTPREP
)) {
1876 * remove the space for the drain we added
1877 * so that we don't add it again
1879 --rq
->nr_phys_segments
;
1884 } else if (ret
== BLKPREP_KILL
) {
1885 rq
->cmd_flags
|= REQ_QUIET
;
1887 * Mark this request as started so we don't trigger
1888 * any debug logic in the end I/O path.
1890 blk_start_request(rq
);
1891 __blk_end_request_all(rq
, -EIO
);
1893 printk(KERN_ERR
"%s: bad return=%d\n", __func__
, ret
);
1900 EXPORT_SYMBOL(blk_peek_request
);
1902 void blk_dequeue_request(struct request
*rq
)
1904 struct request_queue
*q
= rq
->q
;
1906 BUG_ON(list_empty(&rq
->queuelist
));
1907 BUG_ON(ELV_ON_HASH(rq
));
1909 list_del_init(&rq
->queuelist
);
1912 * the time frame between a request being removed from the lists
1913 * and to it is freed is accounted as io that is in progress at
1916 if (blk_account_rq(rq
)) {
1917 q
->in_flight
[rq_is_sync(rq
)]++;
1918 set_io_start_time_ns(rq
);
1923 * blk_start_request - start request processing on the driver
1924 * @req: request to dequeue
1927 * Dequeue @req and start timeout timer on it. This hands off the
1928 * request to the driver.
1930 * Block internal functions which don't want to start timer should
1931 * call blk_dequeue_request().
1934 * queue_lock must be held.
1936 void blk_start_request(struct request
*req
)
1938 blk_dequeue_request(req
);
1941 * We are now handing the request to the hardware, initialize
1942 * resid_len to full count and add the timeout handler.
1944 req
->resid_len
= blk_rq_bytes(req
);
1945 if (unlikely(blk_bidi_rq(req
)))
1946 req
->next_rq
->resid_len
= blk_rq_bytes(req
->next_rq
);
1950 EXPORT_SYMBOL(blk_start_request
);
1953 * blk_fetch_request - fetch a request from a request queue
1954 * @q: request queue to fetch a request from
1957 * Return the request at the top of @q. The request is started on
1958 * return and LLD can start processing it immediately.
1961 * Pointer to the request at the top of @q if available. Null
1965 * queue_lock must be held.
1967 struct request
*blk_fetch_request(struct request_queue
*q
)
1971 rq
= blk_peek_request(q
);
1973 blk_start_request(rq
);
1976 EXPORT_SYMBOL(blk_fetch_request
);
1979 * blk_update_request - Special helper function for request stacking drivers
1980 * @req: the request being processed
1981 * @error: %0 for success, < %0 for error
1982 * @nr_bytes: number of bytes to complete @req
1985 * Ends I/O on a number of bytes attached to @req, but doesn't complete
1986 * the request structure even if @req doesn't have leftover.
1987 * If @req has leftover, sets it up for the next range of segments.
1989 * This special helper function is only for request stacking drivers
1990 * (e.g. request-based dm) so that they can handle partial completion.
1991 * Actual device drivers should use blk_end_request instead.
1993 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
1994 * %false return from this function.
1997 * %false - this request doesn't have any more data
1998 * %true - this request has more data
2000 bool blk_update_request(struct request
*req
, int error
, unsigned int nr_bytes
)
2002 int total_bytes
, bio_nbytes
, next_idx
= 0;
2008 trace_block_rq_complete(req
->q
, req
);
2011 * For fs requests, rq is just carrier of independent bio's
2012 * and each partial completion should be handled separately.
2013 * Reset per-request error on each partial completion.
2015 * TODO: tj: This is too subtle. It would be better to let
2016 * low level drivers do what they see fit.
2018 if (req
->cmd_type
== REQ_TYPE_FS
)
2021 if (error
&& req
->cmd_type
== REQ_TYPE_FS
&&
2022 !(req
->cmd_flags
& REQ_QUIET
)) {
2027 error_type
= "recoverable transport";
2030 error_type
= "critical target";
2033 error_type
= "critical nexus";
2040 printk(KERN_ERR
"end_request: %s error, dev %s, sector %llu\n",
2041 error_type
, req
->rq_disk
? req
->rq_disk
->disk_name
: "?",
2042 (unsigned long long)blk_rq_pos(req
));
2045 blk_account_io_completion(req
, nr_bytes
);
2047 total_bytes
= bio_nbytes
= 0;
2048 while ((bio
= req
->bio
) != NULL
) {
2051 if (nr_bytes
>= bio
->bi_size
) {
2052 req
->bio
= bio
->bi_next
;
2053 nbytes
= bio
->bi_size
;
2054 req_bio_endio(req
, bio
, nbytes
, error
);
2058 int idx
= bio
->bi_idx
+ next_idx
;
2060 if (unlikely(idx
>= bio
->bi_vcnt
)) {
2061 blk_dump_rq_flags(req
, "__end_that");
2062 printk(KERN_ERR
"%s: bio idx %d >= vcnt %d\n",
2063 __func__
, idx
, bio
->bi_vcnt
);
2067 nbytes
= bio_iovec_idx(bio
, idx
)->bv_len
;
2068 BIO_BUG_ON(nbytes
> bio
->bi_size
);
2071 * not a complete bvec done
2073 if (unlikely(nbytes
> nr_bytes
)) {
2074 bio_nbytes
+= nr_bytes
;
2075 total_bytes
+= nr_bytes
;
2080 * advance to the next vector
2083 bio_nbytes
+= nbytes
;
2086 total_bytes
+= nbytes
;
2092 * end more in this run, or just return 'not-done'
2094 if (unlikely(nr_bytes
<= 0))
2104 * Reset counters so that the request stacking driver
2105 * can find how many bytes remain in the request
2108 req
->__data_len
= 0;
2113 * if the request wasn't completed, update state
2116 req_bio_endio(req
, bio
, bio_nbytes
, error
);
2117 bio
->bi_idx
+= next_idx
;
2118 bio_iovec(bio
)->bv_offset
+= nr_bytes
;
2119 bio_iovec(bio
)->bv_len
-= nr_bytes
;
2122 req
->__data_len
-= total_bytes
;
2123 req
->buffer
= bio_data(req
->bio
);
2125 /* update sector only for requests with clear definition of sector */
2126 if (req
->cmd_type
== REQ_TYPE_FS
|| (req
->cmd_flags
& REQ_DISCARD
))
2127 req
->__sector
+= total_bytes
>> 9;
2129 /* mixed attributes always follow the first bio */
2130 if (req
->cmd_flags
& REQ_MIXED_MERGE
) {
2131 req
->cmd_flags
&= ~REQ_FAILFAST_MASK
;
2132 req
->cmd_flags
|= req
->bio
->bi_rw
& REQ_FAILFAST_MASK
;
2136 * If total number of sectors is less than the first segment
2137 * size, something has gone terribly wrong.
2139 if (blk_rq_bytes(req
) < blk_rq_cur_bytes(req
)) {
2140 blk_dump_rq_flags(req
, "request botched");
2141 req
->__data_len
= blk_rq_cur_bytes(req
);
2144 /* recalculate the number of segments */
2145 blk_recalc_rq_segments(req
);
2149 EXPORT_SYMBOL_GPL(blk_update_request
);
2151 static bool blk_update_bidi_request(struct request
*rq
, int error
,
2152 unsigned int nr_bytes
,
2153 unsigned int bidi_bytes
)
2155 if (blk_update_request(rq
, error
, nr_bytes
))
2158 /* Bidi request must be completed as a whole */
2159 if (unlikely(blk_bidi_rq(rq
)) &&
2160 blk_update_request(rq
->next_rq
, error
, bidi_bytes
))
2163 if (blk_queue_add_random(rq
->q
))
2164 add_disk_randomness(rq
->rq_disk
);
2170 * blk_unprep_request - unprepare a request
2173 * This function makes a request ready for complete resubmission (or
2174 * completion). It happens only after all error handling is complete,
2175 * so represents the appropriate moment to deallocate any resources
2176 * that were allocated to the request in the prep_rq_fn. The queue
2177 * lock is held when calling this.
2179 void blk_unprep_request(struct request
*req
)
2181 struct request_queue
*q
= req
->q
;
2183 req
->cmd_flags
&= ~REQ_DONTPREP
;
2184 if (q
->unprep_rq_fn
)
2185 q
->unprep_rq_fn(q
, req
);
2187 EXPORT_SYMBOL_GPL(blk_unprep_request
);
2190 * queue lock must be held
2192 static void blk_finish_request(struct request
*req
, int error
)
2194 if (blk_rq_tagged(req
))
2195 blk_queue_end_tag(req
->q
, req
);
2197 BUG_ON(blk_queued_rq(req
));
2199 if (unlikely(laptop_mode
) && req
->cmd_type
== REQ_TYPE_FS
)
2200 laptop_io_completion(&req
->q
->backing_dev_info
);
2202 blk_delete_timer(req
);
2204 if (req
->cmd_flags
& REQ_DONTPREP
)
2205 blk_unprep_request(req
);
2208 blk_account_io_done(req
);
2211 req
->end_io(req
, error
);
2213 if (blk_bidi_rq(req
))
2214 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
2216 __blk_put_request(req
->q
, req
);
2221 * blk_end_bidi_request - Complete a bidi request
2222 * @rq: the request to complete
2223 * @error: %0 for success, < %0 for error
2224 * @nr_bytes: number of bytes to complete @rq
2225 * @bidi_bytes: number of bytes to complete @rq->next_rq
2228 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2229 * Drivers that supports bidi can safely call this member for any
2230 * type of request, bidi or uni. In the later case @bidi_bytes is
2234 * %false - we are done with this request
2235 * %true - still buffers pending for this request
2237 static bool blk_end_bidi_request(struct request
*rq
, int error
,
2238 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2240 struct request_queue
*q
= rq
->q
;
2241 unsigned long flags
;
2243 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2246 spin_lock_irqsave(q
->queue_lock
, flags
);
2247 blk_finish_request(rq
, error
);
2248 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2254 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2255 * @rq: the request to complete
2256 * @error: %0 for success, < %0 for error
2257 * @nr_bytes: number of bytes to complete @rq
2258 * @bidi_bytes: number of bytes to complete @rq->next_rq
2261 * Identical to blk_end_bidi_request() except that queue lock is
2262 * assumed to be locked on entry and remains so on return.
2265 * %false - we are done with this request
2266 * %true - still buffers pending for this request
2268 static bool __blk_end_bidi_request(struct request
*rq
, int error
,
2269 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2271 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2274 blk_finish_request(rq
, error
);
2280 * blk_end_request - Helper function for drivers to complete the request.
2281 * @rq: the request being processed
2282 * @error: %0 for success, < %0 for error
2283 * @nr_bytes: number of bytes to complete
2286 * Ends I/O on a number of bytes attached to @rq.
2287 * If @rq has leftover, sets it up for the next range of segments.
2290 * %false - we are done with this request
2291 * %true - still buffers pending for this request
2293 bool blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2295 return blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2297 EXPORT_SYMBOL(blk_end_request
);
2300 * blk_end_request_all - Helper function for drives to finish the request.
2301 * @rq: the request to finish
2302 * @error: %0 for success, < %0 for error
2305 * Completely finish @rq.
2307 void blk_end_request_all(struct request
*rq
, int error
)
2310 unsigned int bidi_bytes
= 0;
2312 if (unlikely(blk_bidi_rq(rq
)))
2313 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2315 pending
= blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2318 EXPORT_SYMBOL(blk_end_request_all
);
2321 * blk_end_request_cur - Helper function to finish the current request chunk.
2322 * @rq: the request to finish the current chunk for
2323 * @error: %0 for success, < %0 for error
2326 * Complete the current consecutively mapped chunk from @rq.
2329 * %false - we are done with this request
2330 * %true - still buffers pending for this request
2332 bool blk_end_request_cur(struct request
*rq
, int error
)
2334 return blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2336 EXPORT_SYMBOL(blk_end_request_cur
);
2339 * blk_end_request_err - Finish a request till the next failure boundary.
2340 * @rq: the request to finish till the next failure boundary for
2341 * @error: must be negative errno
2344 * Complete @rq till the next failure boundary.
2347 * %false - we are done with this request
2348 * %true - still buffers pending for this request
2350 bool blk_end_request_err(struct request
*rq
, int error
)
2352 WARN_ON(error
>= 0);
2353 return blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2355 EXPORT_SYMBOL_GPL(blk_end_request_err
);
2358 * __blk_end_request - Helper function for drivers to complete the request.
2359 * @rq: the request being processed
2360 * @error: %0 for success, < %0 for error
2361 * @nr_bytes: number of bytes to complete
2364 * Must be called with queue lock held unlike blk_end_request().
2367 * %false - we are done with this request
2368 * %true - still buffers pending for this request
2370 bool __blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2372 return __blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2374 EXPORT_SYMBOL(__blk_end_request
);
2377 * __blk_end_request_all - Helper function for drives to finish the request.
2378 * @rq: the request to finish
2379 * @error: %0 for success, < %0 for error
2382 * Completely finish @rq. Must be called with queue lock held.
2384 void __blk_end_request_all(struct request
*rq
, int error
)
2387 unsigned int bidi_bytes
= 0;
2389 if (unlikely(blk_bidi_rq(rq
)))
2390 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2392 pending
= __blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2395 EXPORT_SYMBOL(__blk_end_request_all
);
2398 * __blk_end_request_cur - Helper function to finish the current request chunk.
2399 * @rq: the request to finish the current chunk for
2400 * @error: %0 for success, < %0 for error
2403 * Complete the current consecutively mapped chunk from @rq. Must
2404 * be called with queue lock held.
2407 * %false - we are done with this request
2408 * %true - still buffers pending for this request
2410 bool __blk_end_request_cur(struct request
*rq
, int error
)
2412 return __blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2414 EXPORT_SYMBOL(__blk_end_request_cur
);
2417 * __blk_end_request_err - Finish a request till the next failure boundary.
2418 * @rq: the request to finish till the next failure boundary for
2419 * @error: must be negative errno
2422 * Complete @rq till the next failure boundary. Must be called
2423 * with queue lock held.
2426 * %false - we are done with this request
2427 * %true - still buffers pending for this request
2429 bool __blk_end_request_err(struct request
*rq
, int error
)
2431 WARN_ON(error
>= 0);
2432 return __blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2434 EXPORT_SYMBOL_GPL(__blk_end_request_err
);
2436 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
2439 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2440 rq
->cmd_flags
|= bio
->bi_rw
& REQ_WRITE
;
2442 if (bio_has_data(bio
)) {
2443 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2444 rq
->buffer
= bio_data(bio
);
2446 rq
->__data_len
= bio
->bi_size
;
2447 rq
->bio
= rq
->biotail
= bio
;
2450 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2453 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2455 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2456 * @rq: the request to be flushed
2459 * Flush all pages in @rq.
2461 void rq_flush_dcache_pages(struct request
*rq
)
2463 struct req_iterator iter
;
2464 struct bio_vec
*bvec
;
2466 rq_for_each_segment(bvec
, rq
, iter
)
2467 flush_dcache_page(bvec
->bv_page
);
2469 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages
);
2473 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2474 * @q : the queue of the device being checked
2477 * Check if underlying low-level drivers of a device are busy.
2478 * If the drivers want to export their busy state, they must set own
2479 * exporting function using blk_queue_lld_busy() first.
2481 * Basically, this function is used only by request stacking drivers
2482 * to stop dispatching requests to underlying devices when underlying
2483 * devices are busy. This behavior helps more I/O merging on the queue
2484 * of the request stacking driver and prevents I/O throughput regression
2485 * on burst I/O load.
2488 * 0 - Not busy (The request stacking driver should dispatch request)
2489 * 1 - Busy (The request stacking driver should stop dispatching request)
2491 int blk_lld_busy(struct request_queue
*q
)
2494 return q
->lld_busy_fn(q
);
2498 EXPORT_SYMBOL_GPL(blk_lld_busy
);
2501 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2502 * @rq: the clone request to be cleaned up
2505 * Free all bios in @rq for a cloned request.
2507 void blk_rq_unprep_clone(struct request
*rq
)
2511 while ((bio
= rq
->bio
) != NULL
) {
2512 rq
->bio
= bio
->bi_next
;
2517 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone
);
2520 * Copy attributes of the original request to the clone request.
2521 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2523 static void __blk_rq_prep_clone(struct request
*dst
, struct request
*src
)
2525 dst
->cpu
= src
->cpu
;
2526 dst
->cmd_flags
= (src
->cmd_flags
& REQ_CLONE_MASK
) | REQ_NOMERGE
;
2527 dst
->cmd_type
= src
->cmd_type
;
2528 dst
->__sector
= blk_rq_pos(src
);
2529 dst
->__data_len
= blk_rq_bytes(src
);
2530 dst
->nr_phys_segments
= src
->nr_phys_segments
;
2531 dst
->ioprio
= src
->ioprio
;
2532 dst
->extra_len
= src
->extra_len
;
2536 * blk_rq_prep_clone - Helper function to setup clone request
2537 * @rq: the request to be setup
2538 * @rq_src: original request to be cloned
2539 * @bs: bio_set that bios for clone are allocated from
2540 * @gfp_mask: memory allocation mask for bio
2541 * @bio_ctr: setup function to be called for each clone bio.
2542 * Returns %0 for success, non %0 for failure.
2543 * @data: private data to be passed to @bio_ctr
2546 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2547 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2548 * are not copied, and copying such parts is the caller's responsibility.
2549 * Also, pages which the original bios are pointing to are not copied
2550 * and the cloned bios just point same pages.
2551 * So cloned bios must be completed before original bios, which means
2552 * the caller must complete @rq before @rq_src.
2554 int blk_rq_prep_clone(struct request
*rq
, struct request
*rq_src
,
2555 struct bio_set
*bs
, gfp_t gfp_mask
,
2556 int (*bio_ctr
)(struct bio
*, struct bio
*, void *),
2559 struct bio
*bio
, *bio_src
;
2564 blk_rq_init(NULL
, rq
);
2566 __rq_for_each_bio(bio_src
, rq_src
) {
2567 bio
= bio_alloc_bioset(gfp_mask
, bio_src
->bi_max_vecs
, bs
);
2571 __bio_clone(bio
, bio_src
);
2573 if (bio_integrity(bio_src
) &&
2574 bio_integrity_clone(bio
, bio_src
, gfp_mask
, bs
))
2577 if (bio_ctr
&& bio_ctr(bio
, bio_src
, data
))
2581 rq
->biotail
->bi_next
= bio
;
2584 rq
->bio
= rq
->biotail
= bio
;
2587 __blk_rq_prep_clone(rq
, rq_src
);
2594 blk_rq_unprep_clone(rq
);
2598 EXPORT_SYMBOL_GPL(blk_rq_prep_clone
);
2600 int kblockd_schedule_work(struct request_queue
*q
, struct work_struct
*work
)
2602 return queue_work(kblockd_workqueue
, work
);
2604 EXPORT_SYMBOL(kblockd_schedule_work
);
2606 int kblockd_schedule_delayed_work(struct request_queue
*q
,
2607 struct delayed_work
*dwork
, unsigned long delay
)
2609 return queue_delayed_work(kblockd_workqueue
, dwork
, delay
);
2611 EXPORT_SYMBOL(kblockd_schedule_delayed_work
);
2613 #define PLUG_MAGIC 0x91827364
2615 void blk_start_plug(struct blk_plug
*plug
)
2617 struct task_struct
*tsk
= current
;
2619 plug
->magic
= PLUG_MAGIC
;
2620 INIT_LIST_HEAD(&plug
->list
);
2621 INIT_LIST_HEAD(&plug
->cb_list
);
2622 plug
->should_sort
= 0;
2625 * If this is a nested plug, don't actually assign it. It will be
2626 * flushed on its own.
2630 * Store ordering should not be needed here, since a potential
2631 * preempt will imply a full memory barrier
2636 EXPORT_SYMBOL(blk_start_plug
);
2638 static int plug_rq_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
2640 struct request
*rqa
= container_of(a
, struct request
, queuelist
);
2641 struct request
*rqb
= container_of(b
, struct request
, queuelist
);
2643 return !(rqa
->q
<= rqb
->q
);
2647 * If 'from_schedule' is true, then postpone the dispatch of requests
2648 * until a safe kblockd context. We due this to avoid accidental big
2649 * additional stack usage in driver dispatch, in places where the originally
2650 * plugger did not intend it.
2652 static void queue_unplugged(struct request_queue
*q
, unsigned int depth
,
2654 __releases(q
->queue_lock
)
2656 trace_block_unplug(q
, depth
, !from_schedule
);
2659 * If we are punting this to kblockd, then we can safely drop
2660 * the queue_lock before waking kblockd (which needs to take
2663 if (from_schedule
) {
2664 spin_unlock(q
->queue_lock
);
2665 blk_run_queue_async(q
);
2668 spin_unlock(q
->queue_lock
);
2673 static void flush_plug_callbacks(struct blk_plug
*plug
)
2675 LIST_HEAD(callbacks
);
2677 if (list_empty(&plug
->cb_list
))
2680 list_splice_init(&plug
->cb_list
, &callbacks
);
2682 while (!list_empty(&callbacks
)) {
2683 struct blk_plug_cb
*cb
= list_first_entry(&callbacks
,
2686 list_del(&cb
->list
);
2691 void blk_flush_plug_list(struct blk_plug
*plug
, bool from_schedule
)
2693 struct request_queue
*q
;
2694 unsigned long flags
;
2699 BUG_ON(plug
->magic
!= PLUG_MAGIC
);
2701 flush_plug_callbacks(plug
);
2702 if (list_empty(&plug
->list
))
2705 list_splice_init(&plug
->list
, &list
);
2707 if (plug
->should_sort
) {
2708 list_sort(NULL
, &list
, plug_rq_cmp
);
2709 plug
->should_sort
= 0;
2716 * Save and disable interrupts here, to avoid doing it for every
2717 * queue lock we have to take.
2719 local_irq_save(flags
);
2720 while (!list_empty(&list
)) {
2721 rq
= list_entry_rq(list
.next
);
2722 list_del_init(&rq
->queuelist
);
2726 * This drops the queue lock
2729 queue_unplugged(q
, depth
, from_schedule
);
2732 spin_lock(q
->queue_lock
);
2735 * rq is already accounted, so use raw insert
2737 if (rq
->cmd_flags
& (REQ_FLUSH
| REQ_FUA
))
2738 __elv_add_request(q
, rq
, ELEVATOR_INSERT_FLUSH
);
2740 __elv_add_request(q
, rq
, ELEVATOR_INSERT_SORT_MERGE
);
2746 * This drops the queue lock
2749 queue_unplugged(q
, depth
, from_schedule
);
2751 local_irq_restore(flags
);
2754 void blk_finish_plug(struct blk_plug
*plug
)
2756 blk_flush_plug_list(plug
, false);
2758 if (plug
== current
->plug
)
2759 current
->plug
= NULL
;
2761 EXPORT_SYMBOL(blk_finish_plug
);
2763 int __init
blk_dev_init(void)
2765 BUILD_BUG_ON(__REQ_NR_BITS
> 8 *
2766 sizeof(((struct request
*)0)->cmd_flags
));
2768 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
2769 kblockd_workqueue
= alloc_workqueue("kblockd",
2770 WQ_MEM_RECLAIM
| WQ_HIGHPRI
, 0);
2771 if (!kblockd_workqueue
)
2772 panic("Failed to create kblockd\n");
2774 request_cachep
= kmem_cache_create("blkdev_requests",
2775 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
2777 blk_requestq_cachep
= kmem_cache_create("blkdev_queue",
2778 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);