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
);
42 * For the allocated request tables
44 static struct kmem_cache
*request_cachep
;
47 * For queue allocation
49 struct kmem_cache
*blk_requestq_cachep
;
52 * Controlling structure to kblockd
54 static struct workqueue_struct
*kblockd_workqueue
;
56 static void drive_stat_acct(struct request
*rq
, int new_io
)
58 struct hd_struct
*part
;
59 int rw
= rq_data_dir(rq
);
62 if (!blk_do_io_stat(rq
))
65 cpu
= part_stat_lock();
69 part_stat_inc(cpu
, part
, merges
[rw
]);
71 part
= disk_map_sector_rcu(rq
->rq_disk
, blk_rq_pos(rq
));
72 if (!hd_struct_try_get(part
)) {
74 * The partition is already being removed,
75 * the request will be accounted on the disk only
77 * We take a reference on disk->part0 although that
78 * partition will never be deleted, so we can treat
79 * it as any other partition.
81 part
= &rq
->rq_disk
->part0
;
84 part_round_stats(cpu
, part
);
85 part_inc_in_flight(part
, rw
);
92 void blk_queue_congestion_threshold(struct request_queue
*q
)
96 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) + 1;
97 if (nr
> q
->nr_requests
)
99 q
->nr_congestion_on
= nr
;
101 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) - (q
->nr_requests
/ 16) - 1;
104 q
->nr_congestion_off
= nr
;
108 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
111 * Locates the passed device's request queue and returns the address of its
114 * Will return NULL if the request queue cannot be located.
116 struct backing_dev_info
*blk_get_backing_dev_info(struct block_device
*bdev
)
118 struct backing_dev_info
*ret
= NULL
;
119 struct request_queue
*q
= bdev_get_queue(bdev
);
122 ret
= &q
->backing_dev_info
;
125 EXPORT_SYMBOL(blk_get_backing_dev_info
);
127 void blk_rq_init(struct request_queue
*q
, struct request
*rq
)
129 memset(rq
, 0, sizeof(*rq
));
131 INIT_LIST_HEAD(&rq
->queuelist
);
132 INIT_LIST_HEAD(&rq
->timeout_list
);
135 rq
->__sector
= (sector_t
) -1;
136 INIT_HLIST_NODE(&rq
->hash
);
137 RB_CLEAR_NODE(&rq
->rb_node
);
139 rq
->cmd_len
= BLK_MAX_CDB
;
142 rq
->start_time
= jiffies
;
143 set_start_time_ns(rq
);
146 EXPORT_SYMBOL(blk_rq_init
);
148 static void req_bio_endio(struct request
*rq
, struct bio
*bio
,
149 unsigned int nbytes
, int error
)
152 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
153 else if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
156 if (unlikely(nbytes
> bio
->bi_size
)) {
157 printk(KERN_ERR
"%s: want %u bytes done, %u left\n",
158 __func__
, nbytes
, bio
->bi_size
);
159 nbytes
= bio
->bi_size
;
162 if (unlikely(rq
->cmd_flags
& REQ_QUIET
))
163 set_bit(BIO_QUIET
, &bio
->bi_flags
);
165 bio
->bi_size
-= nbytes
;
166 bio
->bi_sector
+= (nbytes
>> 9);
168 if (bio_integrity(bio
))
169 bio_integrity_advance(bio
, nbytes
);
171 /* don't actually finish bio if it's part of flush sequence */
172 if (bio
->bi_size
== 0 && !(rq
->cmd_flags
& REQ_FLUSH_SEQ
))
173 bio_endio(bio
, error
);
176 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
180 printk(KERN_INFO
"%s: dev %s: type=%x, flags=%x\n", msg
,
181 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->cmd_type
,
184 printk(KERN_INFO
" sector %llu, nr/cnr %u/%u\n",
185 (unsigned long long)blk_rq_pos(rq
),
186 blk_rq_sectors(rq
), blk_rq_cur_sectors(rq
));
187 printk(KERN_INFO
" bio %p, biotail %p, buffer %p, len %u\n",
188 rq
->bio
, rq
->biotail
, rq
->buffer
, blk_rq_bytes(rq
));
190 if (rq
->cmd_type
== REQ_TYPE_BLOCK_PC
) {
191 printk(KERN_INFO
" cdb: ");
192 for (bit
= 0; bit
< BLK_MAX_CDB
; bit
++)
193 printk("%02x ", rq
->cmd
[bit
]);
197 EXPORT_SYMBOL(blk_dump_rq_flags
);
199 static void blk_delay_work(struct work_struct
*work
)
201 struct request_queue
*q
;
203 q
= container_of(work
, struct request_queue
, delay_work
.work
);
204 spin_lock_irq(q
->queue_lock
);
206 spin_unlock_irq(q
->queue_lock
);
210 * blk_delay_queue - restart queueing after defined interval
211 * @q: The &struct request_queue in question
212 * @msecs: Delay in msecs
215 * Sometimes queueing needs to be postponed for a little while, to allow
216 * resources to come back. This function will make sure that queueing is
217 * restarted around the specified time.
219 void blk_delay_queue(struct request_queue
*q
, unsigned long msecs
)
221 queue_delayed_work(kblockd_workqueue
, &q
->delay_work
,
222 msecs_to_jiffies(msecs
));
224 EXPORT_SYMBOL(blk_delay_queue
);
227 * blk_start_queue - restart a previously stopped queue
228 * @q: The &struct request_queue in question
231 * blk_start_queue() will clear the stop flag on the queue, and call
232 * the request_fn for the queue if it was in a stopped state when
233 * entered. Also see blk_stop_queue(). Queue lock must be held.
235 void blk_start_queue(struct request_queue
*q
)
237 WARN_ON(!irqs_disabled());
239 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
242 EXPORT_SYMBOL(blk_start_queue
);
245 * blk_stop_queue - stop a queue
246 * @q: The &struct request_queue in question
249 * The Linux block layer assumes that a block driver will consume all
250 * entries on the request queue when the request_fn strategy is called.
251 * Often this will not happen, because of hardware limitations (queue
252 * depth settings). If a device driver gets a 'queue full' response,
253 * or if it simply chooses not to queue more I/O at one point, it can
254 * call this function to prevent the request_fn from being called until
255 * the driver has signalled it's ready to go again. This happens by calling
256 * blk_start_queue() to restart queue operations. Queue lock must be held.
258 void blk_stop_queue(struct request_queue
*q
)
260 __cancel_delayed_work(&q
->delay_work
);
261 queue_flag_set(QUEUE_FLAG_STOPPED
, q
);
263 EXPORT_SYMBOL(blk_stop_queue
);
266 * blk_sync_queue - cancel any pending callbacks on a queue
270 * The block layer may perform asynchronous callback activity
271 * on a queue, such as calling the unplug function after a timeout.
272 * A block device may call blk_sync_queue to ensure that any
273 * such activity is cancelled, thus allowing it to release resources
274 * that the callbacks might use. The caller must already have made sure
275 * that its ->make_request_fn will not re-add plugging prior to calling
278 * This function does not cancel any asynchronous activity arising
279 * out of elevator or throttling code. That would require elevaotor_exit()
280 * and blk_throtl_exit() to be called with queue lock initialized.
283 void blk_sync_queue(struct request_queue
*q
)
285 del_timer_sync(&q
->timeout
);
286 cancel_delayed_work_sync(&q
->delay_work
);
288 EXPORT_SYMBOL(blk_sync_queue
);
291 * __blk_run_queue - run a single device queue
292 * @q: The queue to run
295 * See @blk_run_queue. This variant must be called with the queue lock
296 * held and interrupts disabled.
298 void __blk_run_queue(struct request_queue
*q
)
300 if (unlikely(blk_queue_stopped(q
)))
305 EXPORT_SYMBOL(__blk_run_queue
);
308 * blk_run_queue_async - run a single device queue in workqueue context
309 * @q: The queue to run
312 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
315 void blk_run_queue_async(struct request_queue
*q
)
317 if (likely(!blk_queue_stopped(q
))) {
318 __cancel_delayed_work(&q
->delay_work
);
319 queue_delayed_work(kblockd_workqueue
, &q
->delay_work
, 0);
322 EXPORT_SYMBOL(blk_run_queue_async
);
325 * blk_run_queue - run a single device queue
326 * @q: The queue to run
329 * Invoke request handling on this queue, if it has pending work to do.
330 * May be used to restart queueing when a request has completed.
332 void blk_run_queue(struct request_queue
*q
)
336 spin_lock_irqsave(q
->queue_lock
, flags
);
338 spin_unlock_irqrestore(q
->queue_lock
, flags
);
340 EXPORT_SYMBOL(blk_run_queue
);
342 void blk_put_queue(struct request_queue
*q
)
344 kobject_put(&q
->kobj
);
346 EXPORT_SYMBOL(blk_put_queue
);
349 * Note: If a driver supplied the queue lock, it should not zap that lock
350 * unexpectedly as some queue cleanup components like elevator_exit() and
351 * blk_throtl_exit() need queue lock.
353 void blk_cleanup_queue(struct request_queue
*q
)
356 * We know we have process context here, so we can be a little
357 * cautious and ensure that pending block actions on this device
358 * are done before moving on. Going into this function, we should
359 * not have processes doing IO to this device.
363 del_timer_sync(&q
->backing_dev_info
.laptop_mode_wb_timer
);
364 mutex_lock(&q
->sysfs_lock
);
365 queue_flag_set_unlocked(QUEUE_FLAG_DEAD
, q
);
366 mutex_unlock(&q
->sysfs_lock
);
369 elevator_exit(q
->elevator
);
375 EXPORT_SYMBOL(blk_cleanup_queue
);
377 static int blk_init_free_list(struct request_queue
*q
)
379 struct request_list
*rl
= &q
->rq
;
381 if (unlikely(rl
->rq_pool
))
384 rl
->count
[BLK_RW_SYNC
] = rl
->count
[BLK_RW_ASYNC
] = 0;
385 rl
->starved
[BLK_RW_SYNC
] = rl
->starved
[BLK_RW_ASYNC
] = 0;
387 init_waitqueue_head(&rl
->wait
[BLK_RW_SYNC
]);
388 init_waitqueue_head(&rl
->wait
[BLK_RW_ASYNC
]);
390 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, mempool_alloc_slab
,
391 mempool_free_slab
, request_cachep
, q
->node
);
399 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
401 return blk_alloc_queue_node(gfp_mask
, -1);
403 EXPORT_SYMBOL(blk_alloc_queue
);
405 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
407 struct request_queue
*q
;
410 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
411 gfp_mask
| __GFP_ZERO
, node_id
);
415 q
->backing_dev_info
.ra_pages
=
416 (VM_MAX_READAHEAD
* 1024) / PAGE_CACHE_SIZE
;
417 q
->backing_dev_info
.state
= 0;
418 q
->backing_dev_info
.capabilities
= BDI_CAP_MAP_COPY
;
419 q
->backing_dev_info
.name
= "block";
421 err
= bdi_init(&q
->backing_dev_info
);
423 kmem_cache_free(blk_requestq_cachep
, q
);
427 if (blk_throtl_init(q
)) {
428 kmem_cache_free(blk_requestq_cachep
, q
);
432 setup_timer(&q
->backing_dev_info
.laptop_mode_wb_timer
,
433 laptop_mode_timer_fn
, (unsigned long) q
);
434 setup_timer(&q
->timeout
, blk_rq_timed_out_timer
, (unsigned long) q
);
435 INIT_LIST_HEAD(&q
->timeout_list
);
436 INIT_LIST_HEAD(&q
->flush_queue
[0]);
437 INIT_LIST_HEAD(&q
->flush_queue
[1]);
438 INIT_LIST_HEAD(&q
->flush_data_in_flight
);
439 INIT_DELAYED_WORK(&q
->delay_work
, blk_delay_work
);
441 kobject_init(&q
->kobj
, &blk_queue_ktype
);
443 mutex_init(&q
->sysfs_lock
);
444 spin_lock_init(&q
->__queue_lock
);
447 * By default initialize queue_lock to internal lock and driver can
448 * override it later if need be.
450 q
->queue_lock
= &q
->__queue_lock
;
454 EXPORT_SYMBOL(blk_alloc_queue_node
);
457 * blk_init_queue - prepare a request queue for use with a block device
458 * @rfn: The function to be called to process requests that have been
459 * placed on the queue.
460 * @lock: Request queue spin lock
463 * If a block device wishes to use the standard request handling procedures,
464 * which sorts requests and coalesces adjacent requests, then it must
465 * call blk_init_queue(). The function @rfn will be called when there
466 * are requests on the queue that need to be processed. If the device
467 * supports plugging, then @rfn may not be called immediately when requests
468 * are available on the queue, but may be called at some time later instead.
469 * Plugged queues are generally unplugged when a buffer belonging to one
470 * of the requests on the queue is needed, or due to memory pressure.
472 * @rfn is not required, or even expected, to remove all requests off the
473 * queue, but only as many as it can handle at a time. If it does leave
474 * requests on the queue, it is responsible for arranging that the requests
475 * get dealt with eventually.
477 * The queue spin lock must be held while manipulating the requests on the
478 * request queue; this lock will be taken also from interrupt context, so irq
479 * disabling is needed for it.
481 * Function returns a pointer to the initialized request queue, or %NULL if
485 * blk_init_queue() must be paired with a blk_cleanup_queue() call
486 * when the block device is deactivated (such as at module unload).
489 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
491 return blk_init_queue_node(rfn
, lock
, -1);
493 EXPORT_SYMBOL(blk_init_queue
);
495 struct request_queue
*
496 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
498 struct request_queue
*uninit_q
, *q
;
500 uninit_q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
504 q
= blk_init_allocated_queue_node(uninit_q
, rfn
, lock
, node_id
);
506 blk_cleanup_queue(uninit_q
);
510 EXPORT_SYMBOL(blk_init_queue_node
);
512 struct request_queue
*
513 blk_init_allocated_queue(struct request_queue
*q
, request_fn_proc
*rfn
,
516 return blk_init_allocated_queue_node(q
, rfn
, lock
, -1);
518 EXPORT_SYMBOL(blk_init_allocated_queue
);
520 struct request_queue
*
521 blk_init_allocated_queue_node(struct request_queue
*q
, request_fn_proc
*rfn
,
522 spinlock_t
*lock
, int node_id
)
528 if (blk_init_free_list(q
))
532 q
->prep_rq_fn
= NULL
;
533 q
->unprep_rq_fn
= NULL
;
534 q
->queue_flags
= QUEUE_FLAG_DEFAULT
;
536 /* Override internal queue lock with supplied lock pointer */
538 q
->queue_lock
= lock
;
541 * This also sets hw/phys segments, boundary and size
543 blk_queue_make_request(q
, blk_queue_bio
);
545 q
->sg_reserved_size
= INT_MAX
;
550 if (!elevator_init(q
, NULL
)) {
551 blk_queue_congestion_threshold(q
);
557 EXPORT_SYMBOL(blk_init_allocated_queue_node
);
559 int blk_get_queue(struct request_queue
*q
)
561 if (likely(!test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
))) {
562 kobject_get(&q
->kobj
);
568 EXPORT_SYMBOL(blk_get_queue
);
570 static inline void blk_free_request(struct request_queue
*q
, struct request
*rq
)
572 if (rq
->cmd_flags
& REQ_ELVPRIV
)
573 elv_put_request(q
, rq
);
574 mempool_free(rq
, q
->rq
.rq_pool
);
577 static struct request
*
578 blk_alloc_request(struct request_queue
*q
, int flags
, int priv
, gfp_t gfp_mask
)
580 struct request
*rq
= mempool_alloc(q
->rq
.rq_pool
, gfp_mask
);
587 rq
->cmd_flags
= flags
| REQ_ALLOCED
;
590 if (unlikely(elv_set_request(q
, rq
, gfp_mask
))) {
591 mempool_free(rq
, q
->rq
.rq_pool
);
594 rq
->cmd_flags
|= REQ_ELVPRIV
;
601 * ioc_batching returns true if the ioc is a valid batching request and
602 * should be given priority access to a request.
604 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
610 * Make sure the process is able to allocate at least 1 request
611 * even if the batch times out, otherwise we could theoretically
614 return ioc
->nr_batch_requests
== q
->nr_batching
||
615 (ioc
->nr_batch_requests
> 0
616 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
620 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
621 * will cause the process to be a "batcher" on all queues in the system. This
622 * is the behaviour we want though - once it gets a wakeup it should be given
625 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
627 if (!ioc
|| ioc_batching(q
, ioc
))
630 ioc
->nr_batch_requests
= q
->nr_batching
;
631 ioc
->last_waited
= jiffies
;
634 static void __freed_request(struct request_queue
*q
, int sync
)
636 struct request_list
*rl
= &q
->rq
;
638 if (rl
->count
[sync
] < queue_congestion_off_threshold(q
))
639 blk_clear_queue_congested(q
, sync
);
641 if (rl
->count
[sync
] + 1 <= q
->nr_requests
) {
642 if (waitqueue_active(&rl
->wait
[sync
]))
643 wake_up(&rl
->wait
[sync
]);
645 blk_clear_queue_full(q
, sync
);
650 * A request has just been released. Account for it, update the full and
651 * congestion status, wake up any waiters. Called under q->queue_lock.
653 static void freed_request(struct request_queue
*q
, int sync
, int priv
)
655 struct request_list
*rl
= &q
->rq
;
661 __freed_request(q
, sync
);
663 if (unlikely(rl
->starved
[sync
^ 1]))
664 __freed_request(q
, sync
^ 1);
668 * Determine if elevator data should be initialized when allocating the
669 * request associated with @bio.
671 static bool blk_rq_should_init_elevator(struct bio
*bio
)
677 * Flush requests do not use the elevator so skip initialization.
678 * This allows a request to share the flush and elevator data.
680 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
))
687 * Get a free request, queue_lock must be held.
688 * Returns NULL on failure, with queue_lock held.
689 * Returns !NULL on success, with queue_lock *not held*.
691 static struct request
*get_request(struct request_queue
*q
, int rw_flags
,
692 struct bio
*bio
, gfp_t gfp_mask
)
694 struct request
*rq
= NULL
;
695 struct request_list
*rl
= &q
->rq
;
696 struct io_context
*ioc
= NULL
;
697 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
698 int may_queue
, priv
= 0;
700 may_queue
= elv_may_queue(q
, rw_flags
);
701 if (may_queue
== ELV_MQUEUE_NO
)
704 if (rl
->count
[is_sync
]+1 >= queue_congestion_on_threshold(q
)) {
705 if (rl
->count
[is_sync
]+1 >= q
->nr_requests
) {
706 ioc
= current_io_context(GFP_ATOMIC
, q
->node
);
708 * The queue will fill after this allocation, so set
709 * it as full, and mark this process as "batching".
710 * This process will be allowed to complete a batch of
711 * requests, others will be blocked.
713 if (!blk_queue_full(q
, is_sync
)) {
714 ioc_set_batching(q
, ioc
);
715 blk_set_queue_full(q
, is_sync
);
717 if (may_queue
!= ELV_MQUEUE_MUST
718 && !ioc_batching(q
, ioc
)) {
720 * The queue is full and the allocating
721 * process is not a "batcher", and not
722 * exempted by the IO scheduler
728 blk_set_queue_congested(q
, is_sync
);
732 * Only allow batching queuers to allocate up to 50% over the defined
733 * limit of requests, otherwise we could have thousands of requests
734 * allocated with any setting of ->nr_requests
736 if (rl
->count
[is_sync
] >= (3 * q
->nr_requests
/ 2))
739 rl
->count
[is_sync
]++;
740 rl
->starved
[is_sync
] = 0;
742 if (blk_rq_should_init_elevator(bio
)) {
743 priv
= !test_bit(QUEUE_FLAG_ELVSWITCH
, &q
->queue_flags
);
748 if (blk_queue_io_stat(q
))
749 rw_flags
|= REQ_IO_STAT
;
750 spin_unlock_irq(q
->queue_lock
);
752 rq
= blk_alloc_request(q
, rw_flags
, priv
, gfp_mask
);
755 * Allocation failed presumably due to memory. Undo anything
756 * we might have messed up.
758 * Allocating task should really be put onto the front of the
759 * wait queue, but this is pretty rare.
761 spin_lock_irq(q
->queue_lock
);
762 freed_request(q
, is_sync
, priv
);
765 * in the very unlikely event that allocation failed and no
766 * requests for this direction was pending, mark us starved
767 * so that freeing of a request in the other direction will
768 * notice us. another possible fix would be to split the
769 * rq mempool into READ and WRITE
772 if (unlikely(rl
->count
[is_sync
] == 0))
773 rl
->starved
[is_sync
] = 1;
779 * ioc may be NULL here, and ioc_batching will be false. That's
780 * OK, if the queue is under the request limit then requests need
781 * not count toward the nr_batch_requests limit. There will always
782 * be some limit enforced by BLK_BATCH_TIME.
784 if (ioc_batching(q
, ioc
))
785 ioc
->nr_batch_requests
--;
787 trace_block_getrq(q
, bio
, rw_flags
& 1);
793 * No available requests for this queue, wait for some requests to become
796 * Called with q->queue_lock held, and returns with it unlocked.
798 static struct request
*get_request_wait(struct request_queue
*q
, int rw_flags
,
801 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
804 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
807 struct io_context
*ioc
;
808 struct request_list
*rl
= &q
->rq
;
810 prepare_to_wait_exclusive(&rl
->wait
[is_sync
], &wait
,
811 TASK_UNINTERRUPTIBLE
);
813 trace_block_sleeprq(q
, bio
, rw_flags
& 1);
815 spin_unlock_irq(q
->queue_lock
);
819 * After sleeping, we become a "batching" process and
820 * will be able to allocate at least one request, and
821 * up to a big batch of them for a small period time.
822 * See ioc_batching, ioc_set_batching
824 ioc
= current_io_context(GFP_NOIO
, q
->node
);
825 ioc_set_batching(q
, ioc
);
827 spin_lock_irq(q
->queue_lock
);
828 finish_wait(&rl
->wait
[is_sync
], &wait
);
830 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
836 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
840 if (unlikely(test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)))
843 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
845 spin_lock_irq(q
->queue_lock
);
846 if (gfp_mask
& __GFP_WAIT
) {
847 rq
= get_request_wait(q
, rw
, NULL
);
849 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
851 spin_unlock_irq(q
->queue_lock
);
853 /* q->queue_lock is unlocked at this point */
857 EXPORT_SYMBOL(blk_get_request
);
860 * blk_make_request - given a bio, allocate a corresponding struct request.
861 * @q: target request queue
862 * @bio: The bio describing the memory mappings that will be submitted for IO.
863 * It may be a chained-bio properly constructed by block/bio layer.
864 * @gfp_mask: gfp flags to be used for memory allocation
866 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
867 * type commands. Where the struct request needs to be farther initialized by
868 * the caller. It is passed a &struct bio, which describes the memory info of
871 * The caller of blk_make_request must make sure that bi_io_vec
872 * are set to describe the memory buffers. That bio_data_dir() will return
873 * the needed direction of the request. (And all bio's in the passed bio-chain
874 * are properly set accordingly)
876 * If called under none-sleepable conditions, mapped bio buffers must not
877 * need bouncing, by calling the appropriate masked or flagged allocator,
878 * suitable for the target device. Otherwise the call to blk_queue_bounce will
881 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
882 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
883 * anything but the first bio in the chain. Otherwise you risk waiting for IO
884 * completion of a bio that hasn't been submitted yet, thus resulting in a
885 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
886 * of bio_alloc(), as that avoids the mempool deadlock.
887 * If possible a big IO should be split into smaller parts when allocation
888 * fails. Partial allocation should not be an error, or you risk a live-lock.
890 struct request
*blk_make_request(struct request_queue
*q
, struct bio
*bio
,
893 struct request
*rq
= blk_get_request(q
, bio_data_dir(bio
), gfp_mask
);
896 return ERR_PTR(-ENOMEM
);
899 struct bio
*bounce_bio
= bio
;
902 blk_queue_bounce(q
, &bounce_bio
);
903 ret
= blk_rq_append_bio(q
, rq
, bounce_bio
);
912 EXPORT_SYMBOL(blk_make_request
);
915 * blk_requeue_request - put a request back on queue
916 * @q: request queue where request should be inserted
917 * @rq: request to be inserted
920 * Drivers often keep queueing requests until the hardware cannot accept
921 * more, when that condition happens we need to put the request back
922 * on the queue. Must be called with queue lock held.
924 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
926 blk_delete_timer(rq
);
927 blk_clear_rq_complete(rq
);
928 trace_block_rq_requeue(q
, rq
);
930 if (blk_rq_tagged(rq
))
931 blk_queue_end_tag(q
, rq
);
933 BUG_ON(blk_queued_rq(rq
));
935 elv_requeue_request(q
, rq
);
937 EXPORT_SYMBOL(blk_requeue_request
);
939 static void add_acct_request(struct request_queue
*q
, struct request
*rq
,
942 drive_stat_acct(rq
, 1);
943 __elv_add_request(q
, rq
, where
);
947 * blk_insert_request - insert a special request into a request queue
948 * @q: request queue where request should be inserted
949 * @rq: request to be inserted
950 * @at_head: insert request at head or tail of queue
951 * @data: private data
954 * Many block devices need to execute commands asynchronously, so they don't
955 * block the whole kernel from preemption during request execution. This is
956 * accomplished normally by inserting aritficial requests tagged as
957 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
958 * be scheduled for actual execution by the request queue.
960 * We have the option of inserting the head or the tail of the queue.
961 * Typically we use the tail for new ioctls and so forth. We use the head
962 * of the queue for things like a QUEUE_FULL message from a device, or a
963 * host that is unable to accept a particular command.
965 void blk_insert_request(struct request_queue
*q
, struct request
*rq
,
966 int at_head
, void *data
)
968 int where
= at_head
? ELEVATOR_INSERT_FRONT
: ELEVATOR_INSERT_BACK
;
972 * tell I/O scheduler that this isn't a regular read/write (ie it
973 * must not attempt merges on this) and that it acts as a soft
976 rq
->cmd_type
= REQ_TYPE_SPECIAL
;
980 spin_lock_irqsave(q
->queue_lock
, flags
);
983 * If command is tagged, release the tag
985 if (blk_rq_tagged(rq
))
986 blk_queue_end_tag(q
, rq
);
988 add_acct_request(q
, rq
, where
);
990 spin_unlock_irqrestore(q
->queue_lock
, flags
);
992 EXPORT_SYMBOL(blk_insert_request
);
994 static void part_round_stats_single(int cpu
, struct hd_struct
*part
,
997 if (now
== part
->stamp
)
1000 if (part_in_flight(part
)) {
1001 __part_stat_add(cpu
, part
, time_in_queue
,
1002 part_in_flight(part
) * (now
- part
->stamp
));
1003 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
1009 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1010 * @cpu: cpu number for stats access
1011 * @part: target partition
1013 * The average IO queue length and utilisation statistics are maintained
1014 * by observing the current state of the queue length and the amount of
1015 * time it has been in this state for.
1017 * Normally, that accounting is done on IO completion, but that can result
1018 * in more than a second's worth of IO being accounted for within any one
1019 * second, leading to >100% utilisation. To deal with that, we call this
1020 * function to do a round-off before returning the results when reading
1021 * /proc/diskstats. This accounts immediately for all queue usage up to
1022 * the current jiffies and restarts the counters again.
1024 void part_round_stats(int cpu
, struct hd_struct
*part
)
1026 unsigned long now
= jiffies
;
1029 part_round_stats_single(cpu
, &part_to_disk(part
)->part0
, now
);
1030 part_round_stats_single(cpu
, part
, now
);
1032 EXPORT_SYMBOL_GPL(part_round_stats
);
1035 * queue lock must be held
1037 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1041 if (unlikely(--req
->ref_count
))
1044 elv_completed_request(q
, req
);
1046 /* this is a bio leak */
1047 WARN_ON(req
->bio
!= NULL
);
1050 * Request may not have originated from ll_rw_blk. if not,
1051 * it didn't come out of our reserved rq pools
1053 if (req
->cmd_flags
& REQ_ALLOCED
) {
1054 int is_sync
= rq_is_sync(req
) != 0;
1055 int priv
= req
->cmd_flags
& REQ_ELVPRIV
;
1057 BUG_ON(!list_empty(&req
->queuelist
));
1058 BUG_ON(!hlist_unhashed(&req
->hash
));
1060 blk_free_request(q
, req
);
1061 freed_request(q
, is_sync
, priv
);
1064 EXPORT_SYMBOL_GPL(__blk_put_request
);
1066 void blk_put_request(struct request
*req
)
1068 unsigned long flags
;
1069 struct request_queue
*q
= req
->q
;
1071 spin_lock_irqsave(q
->queue_lock
, flags
);
1072 __blk_put_request(q
, req
);
1073 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1075 EXPORT_SYMBOL(blk_put_request
);
1078 * blk_add_request_payload - add a payload to a request
1079 * @rq: request to update
1080 * @page: page backing the payload
1081 * @len: length of the payload.
1083 * This allows to later add a payload to an already submitted request by
1084 * a block driver. The driver needs to take care of freeing the payload
1087 * Note that this is a quite horrible hack and nothing but handling of
1088 * discard requests should ever use it.
1090 void blk_add_request_payload(struct request
*rq
, struct page
*page
,
1093 struct bio
*bio
= rq
->bio
;
1095 bio
->bi_io_vec
->bv_page
= page
;
1096 bio
->bi_io_vec
->bv_offset
= 0;
1097 bio
->bi_io_vec
->bv_len
= len
;
1101 bio
->bi_phys_segments
= 1;
1103 rq
->__data_len
= rq
->resid_len
= len
;
1104 rq
->nr_phys_segments
= 1;
1105 rq
->buffer
= bio_data(bio
);
1107 EXPORT_SYMBOL_GPL(blk_add_request_payload
);
1109 static bool bio_attempt_back_merge(struct request_queue
*q
, struct request
*req
,
1112 const int ff
= bio
->bi_rw
& REQ_FAILFAST_MASK
;
1114 if (!ll_back_merge_fn(q
, req
, bio
))
1117 trace_block_bio_backmerge(q
, bio
);
1119 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1120 blk_rq_set_mixed_merge(req
);
1122 req
->biotail
->bi_next
= bio
;
1124 req
->__data_len
+= bio
->bi_size
;
1125 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1127 drive_stat_acct(req
, 0);
1128 elv_bio_merged(q
, req
, bio
);
1132 static bool bio_attempt_front_merge(struct request_queue
*q
,
1133 struct request
*req
, struct bio
*bio
)
1135 const int ff
= bio
->bi_rw
& REQ_FAILFAST_MASK
;
1137 if (!ll_front_merge_fn(q
, req
, bio
))
1140 trace_block_bio_frontmerge(q
, bio
);
1142 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1143 blk_rq_set_mixed_merge(req
);
1145 bio
->bi_next
= req
->bio
;
1149 * may not be valid. if the low level driver said
1150 * it didn't need a bounce buffer then it better
1151 * not touch req->buffer either...
1153 req
->buffer
= bio_data(bio
);
1154 req
->__sector
= bio
->bi_sector
;
1155 req
->__data_len
+= bio
->bi_size
;
1156 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1158 drive_stat_acct(req
, 0);
1159 elv_bio_merged(q
, req
, bio
);
1164 * Attempts to merge with the plugged list in the current process. Returns
1165 * true if merge was successful, otherwise false.
1167 static bool attempt_plug_merge(struct task_struct
*tsk
, struct request_queue
*q
,
1170 struct blk_plug
*plug
;
1178 list_for_each_entry_reverse(rq
, &plug
->list
, queuelist
) {
1184 el_ret
= elv_try_merge(rq
, bio
);
1185 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1186 ret
= bio_attempt_back_merge(q
, rq
, bio
);
1189 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1190 ret
= bio_attempt_front_merge(q
, rq
, bio
);
1199 void init_request_from_bio(struct request
*req
, struct bio
*bio
)
1201 req
->cpu
= bio
->bi_comp_cpu
;
1202 req
->cmd_type
= REQ_TYPE_FS
;
1204 req
->cmd_flags
|= bio
->bi_rw
& REQ_COMMON_MASK
;
1205 if (bio
->bi_rw
& REQ_RAHEAD
)
1206 req
->cmd_flags
|= REQ_FAILFAST_MASK
;
1209 req
->__sector
= bio
->bi_sector
;
1210 req
->ioprio
= bio_prio(bio
);
1211 blk_rq_bio_prep(req
->q
, req
, bio
);
1214 void blk_queue_bio(struct request_queue
*q
, struct bio
*bio
)
1216 const bool sync
= !!(bio
->bi_rw
& REQ_SYNC
);
1217 struct blk_plug
*plug
;
1218 int el_ret
, rw_flags
, where
= ELEVATOR_INSERT_SORT
;
1219 struct request
*req
;
1222 * low level driver can indicate that it wants pages above a
1223 * certain limit bounced to low memory (ie for highmem, or even
1224 * ISA dma in theory)
1226 blk_queue_bounce(q
, &bio
);
1228 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) {
1229 spin_lock_irq(q
->queue_lock
);
1230 where
= ELEVATOR_INSERT_FLUSH
;
1235 * Check if we can merge with the plugged list before grabbing
1238 if (attempt_plug_merge(current
, q
, bio
))
1241 spin_lock_irq(q
->queue_lock
);
1243 el_ret
= elv_merge(q
, &req
, bio
);
1244 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1245 if (bio_attempt_back_merge(q
, req
, bio
)) {
1246 if (!attempt_back_merge(q
, req
))
1247 elv_merged_request(q
, req
, el_ret
);
1250 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1251 if (bio_attempt_front_merge(q
, req
, bio
)) {
1252 if (!attempt_front_merge(q
, req
))
1253 elv_merged_request(q
, req
, el_ret
);
1260 * This sync check and mask will be re-done in init_request_from_bio(),
1261 * but we need to set it earlier to expose the sync flag to the
1262 * rq allocator and io schedulers.
1264 rw_flags
= bio_data_dir(bio
);
1266 rw_flags
|= REQ_SYNC
;
1269 * Grab a free request. This is might sleep but can not fail.
1270 * Returns with the queue unlocked.
1272 req
= get_request_wait(q
, rw_flags
, bio
);
1275 * After dropping the lock and possibly sleeping here, our request
1276 * may now be mergeable after it had proven unmergeable (above).
1277 * We don't worry about that case for efficiency. It won't happen
1278 * often, and the elevators are able to handle it.
1280 init_request_from_bio(req
, bio
);
1282 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
) ||
1283 bio_flagged(bio
, BIO_CPU_AFFINE
))
1284 req
->cpu
= raw_smp_processor_id();
1286 plug
= current
->plug
;
1289 * If this is the first request added after a plug, fire
1290 * of a plug trace. If others have been added before, check
1291 * if we have multiple devices in this plug. If so, make a
1292 * note to sort the list before dispatch.
1294 if (list_empty(&plug
->list
))
1295 trace_block_plug(q
);
1296 else if (!plug
->should_sort
) {
1297 struct request
*__rq
;
1299 __rq
= list_entry_rq(plug
->list
.prev
);
1301 plug
->should_sort
= 1;
1303 list_add_tail(&req
->queuelist
, &plug
->list
);
1305 drive_stat_acct(req
, 1);
1306 if (plug
->count
>= BLK_MAX_REQUEST_COUNT
)
1307 blk_flush_plug_list(plug
, false);
1309 spin_lock_irq(q
->queue_lock
);
1310 add_acct_request(q
, req
, where
);
1313 spin_unlock_irq(q
->queue_lock
);
1316 EXPORT_SYMBOL_GPL(blk_queue_bio
); /* for device mapper only */
1319 * If bio->bi_dev is a partition, remap the location
1321 static inline void blk_partition_remap(struct bio
*bio
)
1323 struct block_device
*bdev
= bio
->bi_bdev
;
1325 if (bio_sectors(bio
) && bdev
!= bdev
->bd_contains
) {
1326 struct hd_struct
*p
= bdev
->bd_part
;
1328 bio
->bi_sector
+= p
->start_sect
;
1329 bio
->bi_bdev
= bdev
->bd_contains
;
1331 trace_block_bio_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
1333 bio
->bi_sector
- p
->start_sect
);
1337 static void handle_bad_sector(struct bio
*bio
)
1339 char b
[BDEVNAME_SIZE
];
1341 printk(KERN_INFO
"attempt to access beyond end of device\n");
1342 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
1343 bdevname(bio
->bi_bdev
, b
),
1345 (unsigned long long)bio
->bi_sector
+ bio_sectors(bio
),
1346 (long long)(i_size_read(bio
->bi_bdev
->bd_inode
) >> 9));
1348 set_bit(BIO_EOF
, &bio
->bi_flags
);
1351 #ifdef CONFIG_FAIL_MAKE_REQUEST
1353 static DECLARE_FAULT_ATTR(fail_make_request
);
1355 static int __init
setup_fail_make_request(char *str
)
1357 return setup_fault_attr(&fail_make_request
, str
);
1359 __setup("fail_make_request=", setup_fail_make_request
);
1361 static bool should_fail_request(struct hd_struct
*part
, unsigned int bytes
)
1363 return part
->make_it_fail
&& should_fail(&fail_make_request
, bytes
);
1366 static int __init
fail_make_request_debugfs(void)
1368 struct dentry
*dir
= fault_create_debugfs_attr("fail_make_request",
1369 NULL
, &fail_make_request
);
1371 return IS_ERR(dir
) ? PTR_ERR(dir
) : 0;
1374 late_initcall(fail_make_request_debugfs
);
1376 #else /* CONFIG_FAIL_MAKE_REQUEST */
1378 static inline bool should_fail_request(struct hd_struct
*part
,
1384 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1387 * Check whether this bio extends beyond the end of the device.
1389 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
1396 /* Test device or partition size, when known. */
1397 maxsector
= i_size_read(bio
->bi_bdev
->bd_inode
) >> 9;
1399 sector_t sector
= bio
->bi_sector
;
1401 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
1403 * This may well happen - the kernel calls bread()
1404 * without checking the size of the device, e.g., when
1405 * mounting a device.
1407 handle_bad_sector(bio
);
1415 static noinline_for_stack
bool
1416 generic_make_request_checks(struct bio
*bio
)
1418 struct request_queue
*q
;
1419 int nr_sectors
= bio_sectors(bio
);
1421 char b
[BDEVNAME_SIZE
];
1422 struct hd_struct
*part
;
1426 if (bio_check_eod(bio
, nr_sectors
))
1429 q
= bdev_get_queue(bio
->bi_bdev
);
1432 "generic_make_request: Trying to access "
1433 "nonexistent block-device %s (%Lu)\n",
1434 bdevname(bio
->bi_bdev
, b
),
1435 (long long) bio
->bi_sector
);
1439 if (unlikely(!(bio
->bi_rw
& REQ_DISCARD
) &&
1440 nr_sectors
> queue_max_hw_sectors(q
))) {
1441 printk(KERN_ERR
"bio too big device %s (%u > %u)\n",
1442 bdevname(bio
->bi_bdev
, b
),
1444 queue_max_hw_sectors(q
));
1448 if (unlikely(test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)))
1451 part
= bio
->bi_bdev
->bd_part
;
1452 if (should_fail_request(part
, bio
->bi_size
) ||
1453 should_fail_request(&part_to_disk(part
)->part0
,
1458 * If this device has partitions, remap block n
1459 * of partition p to block n+start(p) of the disk.
1461 blk_partition_remap(bio
);
1463 if (bio_integrity_enabled(bio
) && bio_integrity_prep(bio
))
1466 if (bio_check_eod(bio
, nr_sectors
))
1470 * Filter flush bio's early so that make_request based
1471 * drivers without flush support don't have to worry
1474 if ((bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) && !q
->flush_flags
) {
1475 bio
->bi_rw
&= ~(REQ_FLUSH
| REQ_FUA
);
1482 if ((bio
->bi_rw
& REQ_DISCARD
) &&
1483 (!blk_queue_discard(q
) ||
1484 ((bio
->bi_rw
& REQ_SECURE
) &&
1485 !blk_queue_secdiscard(q
)))) {
1490 if (blk_throtl_bio(q
, &bio
))
1493 /* if bio = NULL, bio has been throttled and will be submitted later. */
1497 trace_block_bio_queue(q
, bio
);
1501 bio_endio(bio
, err
);
1506 * generic_make_request - hand a buffer to its device driver for I/O
1507 * @bio: The bio describing the location in memory and on the device.
1509 * generic_make_request() is used to make I/O requests of block
1510 * devices. It is passed a &struct bio, which describes the I/O that needs
1513 * generic_make_request() does not return any status. The
1514 * success/failure status of the request, along with notification of
1515 * completion, is delivered asynchronously through the bio->bi_end_io
1516 * function described (one day) else where.
1518 * The caller of generic_make_request must make sure that bi_io_vec
1519 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1520 * set to describe the device address, and the
1521 * bi_end_io and optionally bi_private are set to describe how
1522 * completion notification should be signaled.
1524 * generic_make_request and the drivers it calls may use bi_next if this
1525 * bio happens to be merged with someone else, and may resubmit the bio to
1526 * a lower device by calling into generic_make_request recursively, which
1527 * means the bio should NOT be touched after the call to ->make_request_fn.
1529 void generic_make_request(struct bio
*bio
)
1531 struct bio_list bio_list_on_stack
;
1533 if (!generic_make_request_checks(bio
))
1537 * We only want one ->make_request_fn to be active at a time, else
1538 * stack usage with stacked devices could be a problem. So use
1539 * current->bio_list to keep a list of requests submited by a
1540 * make_request_fn function. current->bio_list is also used as a
1541 * flag to say if generic_make_request is currently active in this
1542 * task or not. If it is NULL, then no make_request is active. If
1543 * it is non-NULL, then a make_request is active, and new requests
1544 * should be added at the tail
1546 if (current
->bio_list
) {
1547 bio_list_add(current
->bio_list
, bio
);
1551 /* following loop may be a bit non-obvious, and so deserves some
1553 * Before entering the loop, bio->bi_next is NULL (as all callers
1554 * ensure that) so we have a list with a single bio.
1555 * We pretend that we have just taken it off a longer list, so
1556 * we assign bio_list to a pointer to the bio_list_on_stack,
1557 * thus initialising the bio_list of new bios to be
1558 * added. ->make_request() may indeed add some more bios
1559 * through a recursive call to generic_make_request. If it
1560 * did, we find a non-NULL value in bio_list and re-enter the loop
1561 * from the top. In this case we really did just take the bio
1562 * of the top of the list (no pretending) and so remove it from
1563 * bio_list, and call into ->make_request() again.
1565 BUG_ON(bio
->bi_next
);
1566 bio_list_init(&bio_list_on_stack
);
1567 current
->bio_list
= &bio_list_on_stack
;
1569 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
1571 q
->make_request_fn(q
, bio
);
1573 bio
= bio_list_pop(current
->bio_list
);
1575 current
->bio_list
= NULL
; /* deactivate */
1577 EXPORT_SYMBOL(generic_make_request
);
1580 * submit_bio - submit a bio to the block device layer for I/O
1581 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1582 * @bio: The &struct bio which describes the I/O
1584 * submit_bio() is very similar in purpose to generic_make_request(), and
1585 * uses that function to do most of the work. Both are fairly rough
1586 * interfaces; @bio must be presetup and ready for I/O.
1589 void submit_bio(int rw
, struct bio
*bio
)
1591 int count
= bio_sectors(bio
);
1596 * If it's a regular read/write or a barrier with data attached,
1597 * go through the normal accounting stuff before submission.
1599 if (bio_has_data(bio
) && !(rw
& REQ_DISCARD
)) {
1601 count_vm_events(PGPGOUT
, count
);
1603 task_io_account_read(bio
->bi_size
);
1604 count_vm_events(PGPGIN
, count
);
1607 if (unlikely(block_dump
)) {
1608 char b
[BDEVNAME_SIZE
];
1609 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s (%u sectors)\n",
1610 current
->comm
, task_pid_nr(current
),
1611 (rw
& WRITE
) ? "WRITE" : "READ",
1612 (unsigned long long)bio
->bi_sector
,
1613 bdevname(bio
->bi_bdev
, b
),
1618 generic_make_request(bio
);
1620 EXPORT_SYMBOL(submit_bio
);
1623 * blk_rq_check_limits - Helper function to check a request for the queue limit
1625 * @rq: the request being checked
1628 * @rq may have been made based on weaker limitations of upper-level queues
1629 * in request stacking drivers, and it may violate the limitation of @q.
1630 * Since the block layer and the underlying device driver trust @rq
1631 * after it is inserted to @q, it should be checked against @q before
1632 * the insertion using this generic function.
1634 * This function should also be useful for request stacking drivers
1635 * in some cases below, so export this function.
1636 * Request stacking drivers like request-based dm may change the queue
1637 * limits while requests are in the queue (e.g. dm's table swapping).
1638 * Such request stacking drivers should check those requests agaist
1639 * the new queue limits again when they dispatch those requests,
1640 * although such checkings are also done against the old queue limits
1641 * when submitting requests.
1643 int blk_rq_check_limits(struct request_queue
*q
, struct request
*rq
)
1645 if (rq
->cmd_flags
& REQ_DISCARD
)
1648 if (blk_rq_sectors(rq
) > queue_max_sectors(q
) ||
1649 blk_rq_bytes(rq
) > queue_max_hw_sectors(q
) << 9) {
1650 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
1655 * queue's settings related to segment counting like q->bounce_pfn
1656 * may differ from that of other stacking queues.
1657 * Recalculate it to check the request correctly on this queue's
1660 blk_recalc_rq_segments(rq
);
1661 if (rq
->nr_phys_segments
> queue_max_segments(q
)) {
1662 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
1668 EXPORT_SYMBOL_GPL(blk_rq_check_limits
);
1671 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1672 * @q: the queue to submit the request
1673 * @rq: the request being queued
1675 int blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
1677 unsigned long flags
;
1679 if (blk_rq_check_limits(q
, rq
))
1683 should_fail_request(&rq
->rq_disk
->part0
, blk_rq_bytes(rq
)))
1686 spin_lock_irqsave(q
->queue_lock
, flags
);
1689 * Submitting request must be dequeued before calling this function
1690 * because it will be linked to another request_queue
1692 BUG_ON(blk_queued_rq(rq
));
1694 add_acct_request(q
, rq
, ELEVATOR_INSERT_BACK
);
1695 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1699 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
1702 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1703 * @rq: request to examine
1706 * A request could be merge of IOs which require different failure
1707 * handling. This function determines the number of bytes which
1708 * can be failed from the beginning of the request without
1709 * crossing into area which need to be retried further.
1712 * The number of bytes to fail.
1715 * queue_lock must be held.
1717 unsigned int blk_rq_err_bytes(const struct request
*rq
)
1719 unsigned int ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
1720 unsigned int bytes
= 0;
1723 if (!(rq
->cmd_flags
& REQ_MIXED_MERGE
))
1724 return blk_rq_bytes(rq
);
1727 * Currently the only 'mixing' which can happen is between
1728 * different fastfail types. We can safely fail portions
1729 * which have all the failfast bits that the first one has -
1730 * the ones which are at least as eager to fail as the first
1733 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
1734 if ((bio
->bi_rw
& ff
) != ff
)
1736 bytes
+= bio
->bi_size
;
1739 /* this could lead to infinite loop */
1740 BUG_ON(blk_rq_bytes(rq
) && !bytes
);
1743 EXPORT_SYMBOL_GPL(blk_rq_err_bytes
);
1745 static void blk_account_io_completion(struct request
*req
, unsigned int bytes
)
1747 if (blk_do_io_stat(req
)) {
1748 const int rw
= rq_data_dir(req
);
1749 struct hd_struct
*part
;
1752 cpu
= part_stat_lock();
1754 part_stat_add(cpu
, part
, sectors
[rw
], bytes
>> 9);
1759 static void blk_account_io_done(struct request
*req
)
1762 * Account IO completion. flush_rq isn't accounted as a
1763 * normal IO on queueing nor completion. Accounting the
1764 * containing request is enough.
1766 if (blk_do_io_stat(req
) && !(req
->cmd_flags
& REQ_FLUSH_SEQ
)) {
1767 unsigned long duration
= jiffies
- req
->start_time
;
1768 const int rw
= rq_data_dir(req
);
1769 struct hd_struct
*part
;
1772 cpu
= part_stat_lock();
1775 part_stat_inc(cpu
, part
, ios
[rw
]);
1776 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
1777 part_round_stats(cpu
, part
);
1778 part_dec_in_flight(part
, rw
);
1780 hd_struct_put(part
);
1786 * blk_peek_request - peek at the top of a request queue
1787 * @q: request queue to peek at
1790 * Return the request at the top of @q. The returned request
1791 * should be started using blk_start_request() before LLD starts
1795 * Pointer to the request at the top of @q if available. Null
1799 * queue_lock must be held.
1801 struct request
*blk_peek_request(struct request_queue
*q
)
1806 while ((rq
= __elv_next_request(q
)) != NULL
) {
1807 if (!(rq
->cmd_flags
& REQ_STARTED
)) {
1809 * This is the first time the device driver
1810 * sees this request (possibly after
1811 * requeueing). Notify IO scheduler.
1813 if (rq
->cmd_flags
& REQ_SORTED
)
1814 elv_activate_rq(q
, rq
);
1817 * just mark as started even if we don't start
1818 * it, a request that has been delayed should
1819 * not be passed by new incoming requests
1821 rq
->cmd_flags
|= REQ_STARTED
;
1822 trace_block_rq_issue(q
, rq
);
1825 if (!q
->boundary_rq
|| q
->boundary_rq
== rq
) {
1826 q
->end_sector
= rq_end_sector(rq
);
1827 q
->boundary_rq
= NULL
;
1830 if (rq
->cmd_flags
& REQ_DONTPREP
)
1833 if (q
->dma_drain_size
&& blk_rq_bytes(rq
)) {
1835 * make sure space for the drain appears we
1836 * know we can do this because max_hw_segments
1837 * has been adjusted to be one fewer than the
1840 rq
->nr_phys_segments
++;
1846 ret
= q
->prep_rq_fn(q
, rq
);
1847 if (ret
== BLKPREP_OK
) {
1849 } else if (ret
== BLKPREP_DEFER
) {
1851 * the request may have been (partially) prepped.
1852 * we need to keep this request in the front to
1853 * avoid resource deadlock. REQ_STARTED will
1854 * prevent other fs requests from passing this one.
1856 if (q
->dma_drain_size
&& blk_rq_bytes(rq
) &&
1857 !(rq
->cmd_flags
& REQ_DONTPREP
)) {
1859 * remove the space for the drain we added
1860 * so that we don't add it again
1862 --rq
->nr_phys_segments
;
1867 } else if (ret
== BLKPREP_KILL
) {
1868 rq
->cmd_flags
|= REQ_QUIET
;
1870 * Mark this request as started so we don't trigger
1871 * any debug logic in the end I/O path.
1873 blk_start_request(rq
);
1874 __blk_end_request_all(rq
, -EIO
);
1876 printk(KERN_ERR
"%s: bad return=%d\n", __func__
, ret
);
1883 EXPORT_SYMBOL(blk_peek_request
);
1885 void blk_dequeue_request(struct request
*rq
)
1887 struct request_queue
*q
= rq
->q
;
1889 BUG_ON(list_empty(&rq
->queuelist
));
1890 BUG_ON(ELV_ON_HASH(rq
));
1892 list_del_init(&rq
->queuelist
);
1895 * the time frame between a request being removed from the lists
1896 * and to it is freed is accounted as io that is in progress at
1899 if (blk_account_rq(rq
)) {
1900 q
->in_flight
[rq_is_sync(rq
)]++;
1901 set_io_start_time_ns(rq
);
1906 * blk_start_request - start request processing on the driver
1907 * @req: request to dequeue
1910 * Dequeue @req and start timeout timer on it. This hands off the
1911 * request to the driver.
1913 * Block internal functions which don't want to start timer should
1914 * call blk_dequeue_request().
1917 * queue_lock must be held.
1919 void blk_start_request(struct request
*req
)
1921 blk_dequeue_request(req
);
1924 * We are now handing the request to the hardware, initialize
1925 * resid_len to full count and add the timeout handler.
1927 req
->resid_len
= blk_rq_bytes(req
);
1928 if (unlikely(blk_bidi_rq(req
)))
1929 req
->next_rq
->resid_len
= blk_rq_bytes(req
->next_rq
);
1933 EXPORT_SYMBOL(blk_start_request
);
1936 * blk_fetch_request - fetch a request from a request queue
1937 * @q: request queue to fetch a request from
1940 * Return the request at the top of @q. The request is started on
1941 * return and LLD can start processing it immediately.
1944 * Pointer to the request at the top of @q if available. Null
1948 * queue_lock must be held.
1950 struct request
*blk_fetch_request(struct request_queue
*q
)
1954 rq
= blk_peek_request(q
);
1956 blk_start_request(rq
);
1959 EXPORT_SYMBOL(blk_fetch_request
);
1962 * blk_update_request - Special helper function for request stacking drivers
1963 * @req: the request being processed
1964 * @error: %0 for success, < %0 for error
1965 * @nr_bytes: number of bytes to complete @req
1968 * Ends I/O on a number of bytes attached to @req, but doesn't complete
1969 * the request structure even if @req doesn't have leftover.
1970 * If @req has leftover, sets it up for the next range of segments.
1972 * This special helper function is only for request stacking drivers
1973 * (e.g. request-based dm) so that they can handle partial completion.
1974 * Actual device drivers should use blk_end_request instead.
1976 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
1977 * %false return from this function.
1980 * %false - this request doesn't have any more data
1981 * %true - this request has more data
1983 bool blk_update_request(struct request
*req
, int error
, unsigned int nr_bytes
)
1985 int total_bytes
, bio_nbytes
, next_idx
= 0;
1991 trace_block_rq_complete(req
->q
, req
);
1994 * For fs requests, rq is just carrier of independent bio's
1995 * and each partial completion should be handled separately.
1996 * Reset per-request error on each partial completion.
1998 * TODO: tj: This is too subtle. It would be better to let
1999 * low level drivers do what they see fit.
2001 if (req
->cmd_type
== REQ_TYPE_FS
)
2004 if (error
&& req
->cmd_type
== REQ_TYPE_FS
&&
2005 !(req
->cmd_flags
& REQ_QUIET
)) {
2010 error_type
= "recoverable transport";
2013 error_type
= "critical target";
2016 error_type
= "critical nexus";
2023 printk(KERN_ERR
"end_request: %s error, dev %s, sector %llu\n",
2024 error_type
, req
->rq_disk
? req
->rq_disk
->disk_name
: "?",
2025 (unsigned long long)blk_rq_pos(req
));
2028 blk_account_io_completion(req
, nr_bytes
);
2030 total_bytes
= bio_nbytes
= 0;
2031 while ((bio
= req
->bio
) != NULL
) {
2034 if (nr_bytes
>= bio
->bi_size
) {
2035 req
->bio
= bio
->bi_next
;
2036 nbytes
= bio
->bi_size
;
2037 req_bio_endio(req
, bio
, nbytes
, error
);
2041 int idx
= bio
->bi_idx
+ next_idx
;
2043 if (unlikely(idx
>= bio
->bi_vcnt
)) {
2044 blk_dump_rq_flags(req
, "__end_that");
2045 printk(KERN_ERR
"%s: bio idx %d >= vcnt %d\n",
2046 __func__
, idx
, bio
->bi_vcnt
);
2050 nbytes
= bio_iovec_idx(bio
, idx
)->bv_len
;
2051 BIO_BUG_ON(nbytes
> bio
->bi_size
);
2054 * not a complete bvec done
2056 if (unlikely(nbytes
> nr_bytes
)) {
2057 bio_nbytes
+= nr_bytes
;
2058 total_bytes
+= nr_bytes
;
2063 * advance to the next vector
2066 bio_nbytes
+= nbytes
;
2069 total_bytes
+= nbytes
;
2075 * end more in this run, or just return 'not-done'
2077 if (unlikely(nr_bytes
<= 0))
2087 * Reset counters so that the request stacking driver
2088 * can find how many bytes remain in the request
2091 req
->__data_len
= 0;
2096 * if the request wasn't completed, update state
2099 req_bio_endio(req
, bio
, bio_nbytes
, error
);
2100 bio
->bi_idx
+= next_idx
;
2101 bio_iovec(bio
)->bv_offset
+= nr_bytes
;
2102 bio_iovec(bio
)->bv_len
-= nr_bytes
;
2105 req
->__data_len
-= total_bytes
;
2106 req
->buffer
= bio_data(req
->bio
);
2108 /* update sector only for requests with clear definition of sector */
2109 if (req
->cmd_type
== REQ_TYPE_FS
|| (req
->cmd_flags
& REQ_DISCARD
))
2110 req
->__sector
+= total_bytes
>> 9;
2112 /* mixed attributes always follow the first bio */
2113 if (req
->cmd_flags
& REQ_MIXED_MERGE
) {
2114 req
->cmd_flags
&= ~REQ_FAILFAST_MASK
;
2115 req
->cmd_flags
|= req
->bio
->bi_rw
& REQ_FAILFAST_MASK
;
2119 * If total number of sectors is less than the first segment
2120 * size, something has gone terribly wrong.
2122 if (blk_rq_bytes(req
) < blk_rq_cur_bytes(req
)) {
2123 blk_dump_rq_flags(req
, "request botched");
2124 req
->__data_len
= blk_rq_cur_bytes(req
);
2127 /* recalculate the number of segments */
2128 blk_recalc_rq_segments(req
);
2132 EXPORT_SYMBOL_GPL(blk_update_request
);
2134 static bool blk_update_bidi_request(struct request
*rq
, int error
,
2135 unsigned int nr_bytes
,
2136 unsigned int bidi_bytes
)
2138 if (blk_update_request(rq
, error
, nr_bytes
))
2141 /* Bidi request must be completed as a whole */
2142 if (unlikely(blk_bidi_rq(rq
)) &&
2143 blk_update_request(rq
->next_rq
, error
, bidi_bytes
))
2146 if (blk_queue_add_random(rq
->q
))
2147 add_disk_randomness(rq
->rq_disk
);
2153 * blk_unprep_request - unprepare a request
2156 * This function makes a request ready for complete resubmission (or
2157 * completion). It happens only after all error handling is complete,
2158 * so represents the appropriate moment to deallocate any resources
2159 * that were allocated to the request in the prep_rq_fn. The queue
2160 * lock is held when calling this.
2162 void blk_unprep_request(struct request
*req
)
2164 struct request_queue
*q
= req
->q
;
2166 req
->cmd_flags
&= ~REQ_DONTPREP
;
2167 if (q
->unprep_rq_fn
)
2168 q
->unprep_rq_fn(q
, req
);
2170 EXPORT_SYMBOL_GPL(blk_unprep_request
);
2173 * queue lock must be held
2175 static void blk_finish_request(struct request
*req
, int error
)
2177 if (blk_rq_tagged(req
))
2178 blk_queue_end_tag(req
->q
, req
);
2180 BUG_ON(blk_queued_rq(req
));
2182 if (unlikely(laptop_mode
) && req
->cmd_type
== REQ_TYPE_FS
)
2183 laptop_io_completion(&req
->q
->backing_dev_info
);
2185 blk_delete_timer(req
);
2187 if (req
->cmd_flags
& REQ_DONTPREP
)
2188 blk_unprep_request(req
);
2191 blk_account_io_done(req
);
2194 req
->end_io(req
, error
);
2196 if (blk_bidi_rq(req
))
2197 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
2199 __blk_put_request(req
->q
, req
);
2204 * blk_end_bidi_request - Complete a bidi request
2205 * @rq: the request to complete
2206 * @error: %0 for success, < %0 for error
2207 * @nr_bytes: number of bytes to complete @rq
2208 * @bidi_bytes: number of bytes to complete @rq->next_rq
2211 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2212 * Drivers that supports bidi can safely call this member for any
2213 * type of request, bidi or uni. In the later case @bidi_bytes is
2217 * %false - we are done with this request
2218 * %true - still buffers pending for this request
2220 static bool blk_end_bidi_request(struct request
*rq
, int error
,
2221 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2223 struct request_queue
*q
= rq
->q
;
2224 unsigned long flags
;
2226 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2229 spin_lock_irqsave(q
->queue_lock
, flags
);
2230 blk_finish_request(rq
, error
);
2231 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2237 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2238 * @rq: the request to complete
2239 * @error: %0 for success, < %0 for error
2240 * @nr_bytes: number of bytes to complete @rq
2241 * @bidi_bytes: number of bytes to complete @rq->next_rq
2244 * Identical to blk_end_bidi_request() except that queue lock is
2245 * assumed to be locked on entry and remains so on return.
2248 * %false - we are done with this request
2249 * %true - still buffers pending for this request
2251 static bool __blk_end_bidi_request(struct request
*rq
, int error
,
2252 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2254 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2257 blk_finish_request(rq
, error
);
2263 * blk_end_request - Helper function for drivers to complete the request.
2264 * @rq: the request being processed
2265 * @error: %0 for success, < %0 for error
2266 * @nr_bytes: number of bytes to complete
2269 * Ends I/O on a number of bytes attached to @rq.
2270 * If @rq has leftover, sets it up for the next range of segments.
2273 * %false - we are done with this request
2274 * %true - still buffers pending for this request
2276 bool blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2278 return blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2280 EXPORT_SYMBOL(blk_end_request
);
2283 * blk_end_request_all - Helper function for drives to finish the request.
2284 * @rq: the request to finish
2285 * @error: %0 for success, < %0 for error
2288 * Completely finish @rq.
2290 void blk_end_request_all(struct request
*rq
, int error
)
2293 unsigned int bidi_bytes
= 0;
2295 if (unlikely(blk_bidi_rq(rq
)))
2296 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2298 pending
= blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2301 EXPORT_SYMBOL(blk_end_request_all
);
2304 * blk_end_request_cur - Helper function to finish the current request chunk.
2305 * @rq: the request to finish the current chunk for
2306 * @error: %0 for success, < %0 for error
2309 * Complete the current consecutively mapped chunk from @rq.
2312 * %false - we are done with this request
2313 * %true - still buffers pending for this request
2315 bool blk_end_request_cur(struct request
*rq
, int error
)
2317 return blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2319 EXPORT_SYMBOL(blk_end_request_cur
);
2322 * blk_end_request_err - Finish a request till the next failure boundary.
2323 * @rq: the request to finish till the next failure boundary for
2324 * @error: must be negative errno
2327 * Complete @rq till the next failure boundary.
2330 * %false - we are done with this request
2331 * %true - still buffers pending for this request
2333 bool blk_end_request_err(struct request
*rq
, int error
)
2335 WARN_ON(error
>= 0);
2336 return blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2338 EXPORT_SYMBOL_GPL(blk_end_request_err
);
2341 * __blk_end_request - Helper function for drivers to complete the request.
2342 * @rq: the request being processed
2343 * @error: %0 for success, < %0 for error
2344 * @nr_bytes: number of bytes to complete
2347 * Must be called with queue lock held unlike blk_end_request().
2350 * %false - we are done with this request
2351 * %true - still buffers pending for this request
2353 bool __blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2355 return __blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2357 EXPORT_SYMBOL(__blk_end_request
);
2360 * __blk_end_request_all - Helper function for drives to finish the request.
2361 * @rq: the request to finish
2362 * @error: %0 for success, < %0 for error
2365 * Completely finish @rq. Must be called with queue lock held.
2367 void __blk_end_request_all(struct request
*rq
, int error
)
2370 unsigned int bidi_bytes
= 0;
2372 if (unlikely(blk_bidi_rq(rq
)))
2373 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2375 pending
= __blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2378 EXPORT_SYMBOL(__blk_end_request_all
);
2381 * __blk_end_request_cur - Helper function to finish the current request chunk.
2382 * @rq: the request to finish the current chunk for
2383 * @error: %0 for success, < %0 for error
2386 * Complete the current consecutively mapped chunk from @rq. Must
2387 * be called with queue lock held.
2390 * %false - we are done with this request
2391 * %true - still buffers pending for this request
2393 bool __blk_end_request_cur(struct request
*rq
, int error
)
2395 return __blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2397 EXPORT_SYMBOL(__blk_end_request_cur
);
2400 * __blk_end_request_err - Finish a request till the next failure boundary.
2401 * @rq: the request to finish till the next failure boundary for
2402 * @error: must be negative errno
2405 * Complete @rq till the next failure boundary. Must be called
2406 * with queue lock held.
2409 * %false - we are done with this request
2410 * %true - still buffers pending for this request
2412 bool __blk_end_request_err(struct request
*rq
, int error
)
2414 WARN_ON(error
>= 0);
2415 return __blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2417 EXPORT_SYMBOL_GPL(__blk_end_request_err
);
2419 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
2422 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2423 rq
->cmd_flags
|= bio
->bi_rw
& REQ_WRITE
;
2425 if (bio_has_data(bio
)) {
2426 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2427 rq
->buffer
= bio_data(bio
);
2429 rq
->__data_len
= bio
->bi_size
;
2430 rq
->bio
= rq
->biotail
= bio
;
2433 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2436 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2438 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2439 * @rq: the request to be flushed
2442 * Flush all pages in @rq.
2444 void rq_flush_dcache_pages(struct request
*rq
)
2446 struct req_iterator iter
;
2447 struct bio_vec
*bvec
;
2449 rq_for_each_segment(bvec
, rq
, iter
)
2450 flush_dcache_page(bvec
->bv_page
);
2452 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages
);
2456 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2457 * @q : the queue of the device being checked
2460 * Check if underlying low-level drivers of a device are busy.
2461 * If the drivers want to export their busy state, they must set own
2462 * exporting function using blk_queue_lld_busy() first.
2464 * Basically, this function is used only by request stacking drivers
2465 * to stop dispatching requests to underlying devices when underlying
2466 * devices are busy. This behavior helps more I/O merging on the queue
2467 * of the request stacking driver and prevents I/O throughput regression
2468 * on burst I/O load.
2471 * 0 - Not busy (The request stacking driver should dispatch request)
2472 * 1 - Busy (The request stacking driver should stop dispatching request)
2474 int blk_lld_busy(struct request_queue
*q
)
2477 return q
->lld_busy_fn(q
);
2481 EXPORT_SYMBOL_GPL(blk_lld_busy
);
2484 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2485 * @rq: the clone request to be cleaned up
2488 * Free all bios in @rq for a cloned request.
2490 void blk_rq_unprep_clone(struct request
*rq
)
2494 while ((bio
= rq
->bio
) != NULL
) {
2495 rq
->bio
= bio
->bi_next
;
2500 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone
);
2503 * Copy attributes of the original request to the clone request.
2504 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2506 static void __blk_rq_prep_clone(struct request
*dst
, struct request
*src
)
2508 dst
->cpu
= src
->cpu
;
2509 dst
->cmd_flags
= (src
->cmd_flags
& REQ_CLONE_MASK
) | REQ_NOMERGE
;
2510 dst
->cmd_type
= src
->cmd_type
;
2511 dst
->__sector
= blk_rq_pos(src
);
2512 dst
->__data_len
= blk_rq_bytes(src
);
2513 dst
->nr_phys_segments
= src
->nr_phys_segments
;
2514 dst
->ioprio
= src
->ioprio
;
2515 dst
->extra_len
= src
->extra_len
;
2519 * blk_rq_prep_clone - Helper function to setup clone request
2520 * @rq: the request to be setup
2521 * @rq_src: original request to be cloned
2522 * @bs: bio_set that bios for clone are allocated from
2523 * @gfp_mask: memory allocation mask for bio
2524 * @bio_ctr: setup function to be called for each clone bio.
2525 * Returns %0 for success, non %0 for failure.
2526 * @data: private data to be passed to @bio_ctr
2529 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2530 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2531 * are not copied, and copying such parts is the caller's responsibility.
2532 * Also, pages which the original bios are pointing to are not copied
2533 * and the cloned bios just point same pages.
2534 * So cloned bios must be completed before original bios, which means
2535 * the caller must complete @rq before @rq_src.
2537 int blk_rq_prep_clone(struct request
*rq
, struct request
*rq_src
,
2538 struct bio_set
*bs
, gfp_t gfp_mask
,
2539 int (*bio_ctr
)(struct bio
*, struct bio
*, void *),
2542 struct bio
*bio
, *bio_src
;
2547 blk_rq_init(NULL
, rq
);
2549 __rq_for_each_bio(bio_src
, rq_src
) {
2550 bio
= bio_alloc_bioset(gfp_mask
, bio_src
->bi_max_vecs
, bs
);
2554 __bio_clone(bio
, bio_src
);
2556 if (bio_integrity(bio_src
) &&
2557 bio_integrity_clone(bio
, bio_src
, gfp_mask
, bs
))
2560 if (bio_ctr
&& bio_ctr(bio
, bio_src
, data
))
2564 rq
->biotail
->bi_next
= bio
;
2567 rq
->bio
= rq
->biotail
= bio
;
2570 __blk_rq_prep_clone(rq
, rq_src
);
2577 blk_rq_unprep_clone(rq
);
2581 EXPORT_SYMBOL_GPL(blk_rq_prep_clone
);
2583 int kblockd_schedule_work(struct request_queue
*q
, struct work_struct
*work
)
2585 return queue_work(kblockd_workqueue
, work
);
2587 EXPORT_SYMBOL(kblockd_schedule_work
);
2589 int kblockd_schedule_delayed_work(struct request_queue
*q
,
2590 struct delayed_work
*dwork
, unsigned long delay
)
2592 return queue_delayed_work(kblockd_workqueue
, dwork
, delay
);
2594 EXPORT_SYMBOL(kblockd_schedule_delayed_work
);
2596 #define PLUG_MAGIC 0x91827364
2599 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2600 * @plug: The &struct blk_plug that needs to be initialized
2603 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2604 * pending I/O should the task end up blocking between blk_start_plug() and
2605 * blk_finish_plug(). This is important from a performance perspective, but
2606 * also ensures that we don't deadlock. For instance, if the task is blocking
2607 * for a memory allocation, memory reclaim could end up wanting to free a
2608 * page belonging to that request that is currently residing in our private
2609 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2610 * this kind of deadlock.
2612 void blk_start_plug(struct blk_plug
*plug
)
2614 struct task_struct
*tsk
= current
;
2616 plug
->magic
= PLUG_MAGIC
;
2617 INIT_LIST_HEAD(&plug
->list
);
2618 INIT_LIST_HEAD(&plug
->cb_list
);
2619 plug
->should_sort
= 0;
2623 * If this is a nested plug, don't actually assign it. It will be
2624 * flushed on its own.
2628 * Store ordering should not be needed here, since a potential
2629 * preempt will imply a full memory barrier
2634 EXPORT_SYMBOL(blk_start_plug
);
2636 static int plug_rq_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
2638 struct request
*rqa
= container_of(a
, struct request
, queuelist
);
2639 struct request
*rqb
= container_of(b
, struct request
, queuelist
);
2641 return !(rqa
->q
<= rqb
->q
);
2645 * If 'from_schedule' is true, then postpone the dispatch of requests
2646 * until a safe kblockd context. We due this to avoid accidental big
2647 * additional stack usage in driver dispatch, in places where the originally
2648 * plugger did not intend it.
2650 static void queue_unplugged(struct request_queue
*q
, unsigned int depth
,
2652 __releases(q
->queue_lock
)
2654 trace_block_unplug(q
, depth
, !from_schedule
);
2657 * If we are punting this to kblockd, then we can safely drop
2658 * the queue_lock before waking kblockd (which needs to take
2661 if (from_schedule
) {
2662 spin_unlock(q
->queue_lock
);
2663 blk_run_queue_async(q
);
2666 spin_unlock(q
->queue_lock
);
2671 static void flush_plug_callbacks(struct blk_plug
*plug
)
2673 LIST_HEAD(callbacks
);
2675 if (list_empty(&plug
->cb_list
))
2678 list_splice_init(&plug
->cb_list
, &callbacks
);
2680 while (!list_empty(&callbacks
)) {
2681 struct blk_plug_cb
*cb
= list_first_entry(&callbacks
,
2684 list_del(&cb
->list
);
2689 void blk_flush_plug_list(struct blk_plug
*plug
, bool from_schedule
)
2691 struct request_queue
*q
;
2692 unsigned long flags
;
2697 BUG_ON(plug
->magic
!= PLUG_MAGIC
);
2699 flush_plug_callbacks(plug
);
2700 if (list_empty(&plug
->list
))
2703 list_splice_init(&plug
->list
, &list
);
2706 if (plug
->should_sort
) {
2707 list_sort(NULL
, &list
, plug_rq_cmp
);
2708 plug
->should_sort
= 0;
2715 * Save and disable interrupts here, to avoid doing it for every
2716 * queue lock we have to take.
2718 local_irq_save(flags
);
2719 while (!list_empty(&list
)) {
2720 rq
= list_entry_rq(list
.next
);
2721 list_del_init(&rq
->queuelist
);
2725 * This drops the queue lock
2728 queue_unplugged(q
, depth
, from_schedule
);
2731 spin_lock(q
->queue_lock
);
2734 * rq is already accounted, so use raw insert
2736 if (rq
->cmd_flags
& (REQ_FLUSH
| REQ_FUA
))
2737 __elv_add_request(q
, rq
, ELEVATOR_INSERT_FLUSH
);
2739 __elv_add_request(q
, rq
, ELEVATOR_INSERT_SORT_MERGE
);
2745 * This drops the queue lock
2748 queue_unplugged(q
, depth
, from_schedule
);
2750 local_irq_restore(flags
);
2753 void blk_finish_plug(struct blk_plug
*plug
)
2755 blk_flush_plug_list(plug
, false);
2757 if (plug
== current
->plug
)
2758 current
->plug
= NULL
;
2760 EXPORT_SYMBOL(blk_finish_plug
);
2762 int __init
blk_dev_init(void)
2764 BUILD_BUG_ON(__REQ_NR_BITS
> 8 *
2765 sizeof(((struct request
*)0)->cmd_flags
));
2767 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
2768 kblockd_workqueue
= alloc_workqueue("kblockd",
2769 WQ_MEM_RECLAIM
| WQ_HIGHPRI
, 0);
2770 if (!kblockd_workqueue
)
2771 panic("Failed to create kblockd\n");
2773 request_cachep
= kmem_cache_create("blkdev_requests",
2774 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
2776 blk_requestq_cachep
= kmem_cache_create("blkdev_queue",
2777 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);