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>
31 #include <linux/delay.h>
33 #define CREATE_TRACE_POINTS
34 #include <trace/events/block.h>
38 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap
);
39 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap
);
40 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete
);
43 * For the allocated request tables
45 static struct kmem_cache
*request_cachep
;
48 * For queue allocation
50 struct kmem_cache
*blk_requestq_cachep
;
53 * Controlling structure to kblockd
55 static struct workqueue_struct
*kblockd_workqueue
;
57 static void drive_stat_acct(struct request
*rq
, int new_io
)
59 struct hd_struct
*part
;
60 int rw
= rq_data_dir(rq
);
63 if (!blk_do_io_stat(rq
))
66 cpu
= part_stat_lock();
70 part_stat_inc(cpu
, part
, merges
[rw
]);
72 part
= disk_map_sector_rcu(rq
->rq_disk
, blk_rq_pos(rq
));
73 if (!hd_struct_try_get(part
)) {
75 * The partition is already being removed,
76 * the request will be accounted on the disk only
78 * We take a reference on disk->part0 although that
79 * partition will never be deleted, so we can treat
80 * it as any other partition.
82 part
= &rq
->rq_disk
->part0
;
85 part_round_stats(cpu
, part
);
86 part_inc_in_flight(part
, rw
);
93 void blk_queue_congestion_threshold(struct request_queue
*q
)
97 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) + 1;
98 if (nr
> q
->nr_requests
)
100 q
->nr_congestion_on
= nr
;
102 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) - (q
->nr_requests
/ 16) - 1;
105 q
->nr_congestion_off
= nr
;
109 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
112 * Locates the passed device's request queue and returns the address of its
115 * Will return NULL if the request queue cannot be located.
117 struct backing_dev_info
*blk_get_backing_dev_info(struct block_device
*bdev
)
119 struct backing_dev_info
*ret
= NULL
;
120 struct request_queue
*q
= bdev_get_queue(bdev
);
123 ret
= &q
->backing_dev_info
;
126 EXPORT_SYMBOL(blk_get_backing_dev_info
);
128 void blk_rq_init(struct request_queue
*q
, struct request
*rq
)
130 memset(rq
, 0, sizeof(*rq
));
132 INIT_LIST_HEAD(&rq
->queuelist
);
133 INIT_LIST_HEAD(&rq
->timeout_list
);
136 rq
->__sector
= (sector_t
) -1;
137 INIT_HLIST_NODE(&rq
->hash
);
138 RB_CLEAR_NODE(&rq
->rb_node
);
140 rq
->cmd_len
= BLK_MAX_CDB
;
143 rq
->start_time
= jiffies
;
144 set_start_time_ns(rq
);
147 EXPORT_SYMBOL(blk_rq_init
);
149 static void req_bio_endio(struct request
*rq
, struct bio
*bio
,
150 unsigned int nbytes
, int error
)
153 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
154 else if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
157 if (unlikely(nbytes
> bio
->bi_size
)) {
158 printk(KERN_ERR
"%s: want %u bytes done, %u left\n",
159 __func__
, nbytes
, bio
->bi_size
);
160 nbytes
= bio
->bi_size
;
163 if (unlikely(rq
->cmd_flags
& REQ_QUIET
))
164 set_bit(BIO_QUIET
, &bio
->bi_flags
);
166 bio
->bi_size
-= nbytes
;
167 bio
->bi_sector
+= (nbytes
>> 9);
169 if (bio_integrity(bio
))
170 bio_integrity_advance(bio
, nbytes
);
172 /* don't actually finish bio if it's part of flush sequence */
173 if (bio
->bi_size
== 0 && !(rq
->cmd_flags
& REQ_FLUSH_SEQ
))
174 bio_endio(bio
, error
);
177 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
181 printk(KERN_INFO
"%s: dev %s: type=%x, flags=%x\n", msg
,
182 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->cmd_type
,
185 printk(KERN_INFO
" sector %llu, nr/cnr %u/%u\n",
186 (unsigned long long)blk_rq_pos(rq
),
187 blk_rq_sectors(rq
), blk_rq_cur_sectors(rq
));
188 printk(KERN_INFO
" bio %p, biotail %p, buffer %p, len %u\n",
189 rq
->bio
, rq
->biotail
, rq
->buffer
, blk_rq_bytes(rq
));
191 if (rq
->cmd_type
== REQ_TYPE_BLOCK_PC
) {
192 printk(KERN_INFO
" cdb: ");
193 for (bit
= 0; bit
< BLK_MAX_CDB
; bit
++)
194 printk("%02x ", rq
->cmd
[bit
]);
198 EXPORT_SYMBOL(blk_dump_rq_flags
);
200 static void blk_delay_work(struct work_struct
*work
)
202 struct request_queue
*q
;
204 q
= container_of(work
, struct request_queue
, delay_work
.work
);
205 spin_lock_irq(q
->queue_lock
);
207 spin_unlock_irq(q
->queue_lock
);
211 * blk_delay_queue - restart queueing after defined interval
212 * @q: The &struct request_queue in question
213 * @msecs: Delay in msecs
216 * Sometimes queueing needs to be postponed for a little while, to allow
217 * resources to come back. This function will make sure that queueing is
218 * restarted around the specified time.
220 void blk_delay_queue(struct request_queue
*q
, unsigned long msecs
)
222 queue_delayed_work(kblockd_workqueue
, &q
->delay_work
,
223 msecs_to_jiffies(msecs
));
225 EXPORT_SYMBOL(blk_delay_queue
);
228 * blk_start_queue - restart a previously stopped queue
229 * @q: The &struct request_queue in question
232 * blk_start_queue() will clear the stop flag on the queue, and call
233 * the request_fn for the queue if it was in a stopped state when
234 * entered. Also see blk_stop_queue(). Queue lock must be held.
236 void blk_start_queue(struct request_queue
*q
)
238 WARN_ON(!irqs_disabled());
240 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
243 EXPORT_SYMBOL(blk_start_queue
);
246 * blk_stop_queue - stop a queue
247 * @q: The &struct request_queue in question
250 * The Linux block layer assumes that a block driver will consume all
251 * entries on the request queue when the request_fn strategy is called.
252 * Often this will not happen, because of hardware limitations (queue
253 * depth settings). If a device driver gets a 'queue full' response,
254 * or if it simply chooses not to queue more I/O at one point, it can
255 * call this function to prevent the request_fn from being called until
256 * the driver has signalled it's ready to go again. This happens by calling
257 * blk_start_queue() to restart queue operations. Queue lock must be held.
259 void blk_stop_queue(struct request_queue
*q
)
261 __cancel_delayed_work(&q
->delay_work
);
262 queue_flag_set(QUEUE_FLAG_STOPPED
, q
);
264 EXPORT_SYMBOL(blk_stop_queue
);
267 * blk_sync_queue - cancel any pending callbacks on a queue
271 * The block layer may perform asynchronous callback activity
272 * on a queue, such as calling the unplug function after a timeout.
273 * A block device may call blk_sync_queue to ensure that any
274 * such activity is cancelled, thus allowing it to release resources
275 * that the callbacks might use. The caller must already have made sure
276 * that its ->make_request_fn will not re-add plugging prior to calling
279 * This function does not cancel any asynchronous activity arising
280 * out of elevator or throttling code. That would require elevaotor_exit()
281 * and blk_throtl_exit() to be called with queue lock initialized.
284 void blk_sync_queue(struct request_queue
*q
)
286 del_timer_sync(&q
->timeout
);
287 cancel_delayed_work_sync(&q
->delay_work
);
289 EXPORT_SYMBOL(blk_sync_queue
);
292 * __blk_run_queue - run a single device queue
293 * @q: The queue to run
296 * See @blk_run_queue. This variant must be called with the queue lock
297 * held and interrupts disabled.
299 void __blk_run_queue(struct request_queue
*q
)
301 if (unlikely(blk_queue_stopped(q
)))
306 EXPORT_SYMBOL(__blk_run_queue
);
309 * blk_run_queue_async - run a single device queue in workqueue context
310 * @q: The queue to run
313 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
316 void blk_run_queue_async(struct request_queue
*q
)
318 if (likely(!blk_queue_stopped(q
))) {
319 __cancel_delayed_work(&q
->delay_work
);
320 queue_delayed_work(kblockd_workqueue
, &q
->delay_work
, 0);
323 EXPORT_SYMBOL(blk_run_queue_async
);
326 * blk_run_queue - run a single device queue
327 * @q: The queue to run
330 * Invoke request handling on this queue, if it has pending work to do.
331 * May be used to restart queueing when a request has completed.
333 void blk_run_queue(struct request_queue
*q
)
337 spin_lock_irqsave(q
->queue_lock
, flags
);
339 spin_unlock_irqrestore(q
->queue_lock
, flags
);
341 EXPORT_SYMBOL(blk_run_queue
);
343 void blk_put_queue(struct request_queue
*q
)
345 kobject_put(&q
->kobj
);
347 EXPORT_SYMBOL(blk_put_queue
);
350 * blk_drain_queue - drain requests from request_queue
352 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
354 * Drain requests from @q. If @drain_all is set, all requests are drained.
355 * If not, only ELVPRIV requests are drained. The caller is responsible
356 * for ensuring that no new requests which need to be drained are queued.
358 void blk_drain_queue(struct request_queue
*q
, bool drain_all
)
364 spin_lock_irq(q
->queue_lock
);
366 elv_drain_elevator(q
);
372 drain
|= q
->rq
.elvpriv
;
375 * Unfortunately, requests are queued at and tracked from
376 * multiple places and there's no single counter which can
377 * be drained. Check all the queues and counters.
380 drain
|= !list_empty(&q
->queue_head
);
381 for (i
= 0; i
< 2; i
++) {
382 drain
|= q
->rq
.count
[i
];
383 drain
|= q
->in_flight
[i
];
384 drain
|= !list_empty(&q
->flush_queue
[i
]);
388 spin_unlock_irq(q
->queue_lock
);
397 * blk_cleanup_queue - shutdown a request queue
398 * @q: request queue to shutdown
400 * Mark @q DEAD, drain all pending requests, destroy and put it. All
401 * future requests will be failed immediately with -ENODEV.
403 void blk_cleanup_queue(struct request_queue
*q
)
405 spinlock_t
*lock
= q
->queue_lock
;
407 /* mark @q DEAD, no new request or merges will be allowed afterwards */
408 mutex_lock(&q
->sysfs_lock
);
409 queue_flag_set_unlocked(QUEUE_FLAG_DEAD
, q
);
412 queue_flag_set(QUEUE_FLAG_NOMERGES
, q
);
413 queue_flag_set(QUEUE_FLAG_NOXMERGES
, q
);
414 queue_flag_set(QUEUE_FLAG_DEAD
, q
);
416 if (q
->queue_lock
!= &q
->__queue_lock
)
417 q
->queue_lock
= &q
->__queue_lock
;
419 spin_unlock_irq(lock
);
420 mutex_unlock(&q
->sysfs_lock
);
423 * Drain all requests queued before DEAD marking. The caller might
424 * be trying to tear down @q before its elevator is initialized, in
425 * which case we don't want to call into draining.
428 blk_drain_queue(q
, true);
430 /* @q won't process any more request, flush async actions */
431 del_timer_sync(&q
->backing_dev_info
.laptop_mode_wb_timer
);
434 /* @q is and will stay empty, shutdown and put */
437 EXPORT_SYMBOL(blk_cleanup_queue
);
439 static int blk_init_free_list(struct request_queue
*q
)
441 struct request_list
*rl
= &q
->rq
;
443 if (unlikely(rl
->rq_pool
))
446 rl
->count
[BLK_RW_SYNC
] = rl
->count
[BLK_RW_ASYNC
] = 0;
447 rl
->starved
[BLK_RW_SYNC
] = rl
->starved
[BLK_RW_ASYNC
] = 0;
449 init_waitqueue_head(&rl
->wait
[BLK_RW_SYNC
]);
450 init_waitqueue_head(&rl
->wait
[BLK_RW_ASYNC
]);
452 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, mempool_alloc_slab
,
453 mempool_free_slab
, request_cachep
, q
->node
);
461 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
463 return blk_alloc_queue_node(gfp_mask
, -1);
465 EXPORT_SYMBOL(blk_alloc_queue
);
467 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
469 struct request_queue
*q
;
472 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
473 gfp_mask
| __GFP_ZERO
, node_id
);
477 q
->backing_dev_info
.ra_pages
=
478 (VM_MAX_READAHEAD
* 1024) / PAGE_CACHE_SIZE
;
479 q
->backing_dev_info
.state
= 0;
480 q
->backing_dev_info
.capabilities
= BDI_CAP_MAP_COPY
;
481 q
->backing_dev_info
.name
= "block";
483 err
= bdi_init(&q
->backing_dev_info
);
485 kmem_cache_free(blk_requestq_cachep
, q
);
489 if (blk_throtl_init(q
)) {
490 kmem_cache_free(blk_requestq_cachep
, q
);
494 setup_timer(&q
->backing_dev_info
.laptop_mode_wb_timer
,
495 laptop_mode_timer_fn
, (unsigned long) q
);
496 setup_timer(&q
->timeout
, blk_rq_timed_out_timer
, (unsigned long) q
);
497 INIT_LIST_HEAD(&q
->timeout_list
);
498 INIT_LIST_HEAD(&q
->flush_queue
[0]);
499 INIT_LIST_HEAD(&q
->flush_queue
[1]);
500 INIT_LIST_HEAD(&q
->flush_data_in_flight
);
501 INIT_DELAYED_WORK(&q
->delay_work
, blk_delay_work
);
503 kobject_init(&q
->kobj
, &blk_queue_ktype
);
505 mutex_init(&q
->sysfs_lock
);
506 spin_lock_init(&q
->__queue_lock
);
509 * By default initialize queue_lock to internal lock and driver can
510 * override it later if need be.
512 q
->queue_lock
= &q
->__queue_lock
;
516 EXPORT_SYMBOL(blk_alloc_queue_node
);
519 * blk_init_queue - prepare a request queue for use with a block device
520 * @rfn: The function to be called to process requests that have been
521 * placed on the queue.
522 * @lock: Request queue spin lock
525 * If a block device wishes to use the standard request handling procedures,
526 * which sorts requests and coalesces adjacent requests, then it must
527 * call blk_init_queue(). The function @rfn will be called when there
528 * are requests on the queue that need to be processed. If the device
529 * supports plugging, then @rfn may not be called immediately when requests
530 * are available on the queue, but may be called at some time later instead.
531 * Plugged queues are generally unplugged when a buffer belonging to one
532 * of the requests on the queue is needed, or due to memory pressure.
534 * @rfn is not required, or even expected, to remove all requests off the
535 * queue, but only as many as it can handle at a time. If it does leave
536 * requests on the queue, it is responsible for arranging that the requests
537 * get dealt with eventually.
539 * The queue spin lock must be held while manipulating the requests on the
540 * request queue; this lock will be taken also from interrupt context, so irq
541 * disabling is needed for it.
543 * Function returns a pointer to the initialized request queue, or %NULL if
547 * blk_init_queue() must be paired with a blk_cleanup_queue() call
548 * when the block device is deactivated (such as at module unload).
551 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
553 return blk_init_queue_node(rfn
, lock
, -1);
555 EXPORT_SYMBOL(blk_init_queue
);
557 struct request_queue
*
558 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
560 struct request_queue
*uninit_q
, *q
;
562 uninit_q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
566 q
= blk_init_allocated_queue_node(uninit_q
, rfn
, lock
, node_id
);
568 blk_cleanup_queue(uninit_q
);
572 EXPORT_SYMBOL(blk_init_queue_node
);
574 struct request_queue
*
575 blk_init_allocated_queue(struct request_queue
*q
, request_fn_proc
*rfn
,
578 return blk_init_allocated_queue_node(q
, rfn
, lock
, -1);
580 EXPORT_SYMBOL(blk_init_allocated_queue
);
582 struct request_queue
*
583 blk_init_allocated_queue_node(struct request_queue
*q
, request_fn_proc
*rfn
,
584 spinlock_t
*lock
, int node_id
)
590 if (blk_init_free_list(q
))
594 q
->prep_rq_fn
= NULL
;
595 q
->unprep_rq_fn
= NULL
;
596 q
->queue_flags
= QUEUE_FLAG_DEFAULT
;
598 /* Override internal queue lock with supplied lock pointer */
600 q
->queue_lock
= lock
;
603 * This also sets hw/phys segments, boundary and size
605 blk_queue_make_request(q
, blk_queue_bio
);
607 q
->sg_reserved_size
= INT_MAX
;
612 if (!elevator_init(q
, NULL
)) {
613 blk_queue_congestion_threshold(q
);
619 EXPORT_SYMBOL(blk_init_allocated_queue_node
);
621 int blk_get_queue(struct request_queue
*q
)
623 if (likely(!blk_queue_dead(q
))) {
624 kobject_get(&q
->kobj
);
630 EXPORT_SYMBOL(blk_get_queue
);
632 static inline void blk_free_request(struct request_queue
*q
, struct request
*rq
)
634 if (rq
->cmd_flags
& REQ_ELVPRIV
)
635 elv_put_request(q
, rq
);
636 mempool_free(rq
, q
->rq
.rq_pool
);
639 static struct request
*
640 blk_alloc_request(struct request_queue
*q
, unsigned int flags
, gfp_t gfp_mask
)
642 struct request
*rq
= mempool_alloc(q
->rq
.rq_pool
, gfp_mask
);
649 rq
->cmd_flags
= flags
| REQ_ALLOCED
;
651 if ((flags
& REQ_ELVPRIV
) &&
652 unlikely(elv_set_request(q
, rq
, gfp_mask
))) {
653 mempool_free(rq
, q
->rq
.rq_pool
);
661 * ioc_batching returns true if the ioc is a valid batching request and
662 * should be given priority access to a request.
664 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
670 * Make sure the process is able to allocate at least 1 request
671 * even if the batch times out, otherwise we could theoretically
674 return ioc
->nr_batch_requests
== q
->nr_batching
||
675 (ioc
->nr_batch_requests
> 0
676 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
680 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
681 * will cause the process to be a "batcher" on all queues in the system. This
682 * is the behaviour we want though - once it gets a wakeup it should be given
685 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
687 if (!ioc
|| ioc_batching(q
, ioc
))
690 ioc
->nr_batch_requests
= q
->nr_batching
;
691 ioc
->last_waited
= jiffies
;
694 static void __freed_request(struct request_queue
*q
, int sync
)
696 struct request_list
*rl
= &q
->rq
;
698 if (rl
->count
[sync
] < queue_congestion_off_threshold(q
))
699 blk_clear_queue_congested(q
, sync
);
701 if (rl
->count
[sync
] + 1 <= q
->nr_requests
) {
702 if (waitqueue_active(&rl
->wait
[sync
]))
703 wake_up(&rl
->wait
[sync
]);
705 blk_clear_queue_full(q
, sync
);
710 * A request has just been released. Account for it, update the full and
711 * congestion status, wake up any waiters. Called under q->queue_lock.
713 static void freed_request(struct request_queue
*q
, unsigned int flags
)
715 struct request_list
*rl
= &q
->rq
;
716 int sync
= rw_is_sync(flags
);
719 if (flags
& REQ_ELVPRIV
)
722 __freed_request(q
, sync
);
724 if (unlikely(rl
->starved
[sync
^ 1]))
725 __freed_request(q
, sync
^ 1);
729 * Determine if elevator data should be initialized when allocating the
730 * request associated with @bio.
732 static bool blk_rq_should_init_elevator(struct bio
*bio
)
738 * Flush requests do not use the elevator so skip initialization.
739 * This allows a request to share the flush and elevator data.
741 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
))
748 * get_request - get a free request
749 * @q: request_queue to allocate request from
750 * @rw_flags: RW and SYNC flags
751 * @bio: bio to allocate request for (can be %NULL)
752 * @gfp_mask: allocation mask
754 * Get a free request from @q. This function may fail under memory
755 * pressure or if @q is dead.
757 * Must be callled with @q->queue_lock held and,
758 * Returns %NULL on failure, with @q->queue_lock held.
759 * Returns !%NULL on success, with @q->queue_lock *not held*.
761 static struct request
*get_request(struct request_queue
*q
, int rw_flags
,
762 struct bio
*bio
, gfp_t gfp_mask
)
764 struct request
*rq
= NULL
;
765 struct request_list
*rl
= &q
->rq
;
766 struct io_context
*ioc
= NULL
;
767 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
770 if (unlikely(blk_queue_dead(q
)))
773 may_queue
= elv_may_queue(q
, rw_flags
);
774 if (may_queue
== ELV_MQUEUE_NO
)
777 if (rl
->count
[is_sync
]+1 >= queue_congestion_on_threshold(q
)) {
778 if (rl
->count
[is_sync
]+1 >= q
->nr_requests
) {
779 ioc
= current_io_context(GFP_ATOMIC
, q
->node
);
781 * The queue will fill after this allocation, so set
782 * it as full, and mark this process as "batching".
783 * This process will be allowed to complete a batch of
784 * requests, others will be blocked.
786 if (!blk_queue_full(q
, is_sync
)) {
787 ioc_set_batching(q
, ioc
);
788 blk_set_queue_full(q
, is_sync
);
790 if (may_queue
!= ELV_MQUEUE_MUST
791 && !ioc_batching(q
, ioc
)) {
793 * The queue is full and the allocating
794 * process is not a "batcher", and not
795 * exempted by the IO scheduler
801 blk_set_queue_congested(q
, is_sync
);
805 * Only allow batching queuers to allocate up to 50% over the defined
806 * limit of requests, otherwise we could have thousands of requests
807 * allocated with any setting of ->nr_requests
809 if (rl
->count
[is_sync
] >= (3 * q
->nr_requests
/ 2))
812 rl
->count
[is_sync
]++;
813 rl
->starved
[is_sync
] = 0;
815 if (blk_rq_should_init_elevator(bio
) &&
816 !test_bit(QUEUE_FLAG_ELVSWITCH
, &q
->queue_flags
)) {
817 rw_flags
|= REQ_ELVPRIV
;
821 if (blk_queue_io_stat(q
))
822 rw_flags
|= REQ_IO_STAT
;
823 spin_unlock_irq(q
->queue_lock
);
825 rq
= blk_alloc_request(q
, rw_flags
, gfp_mask
);
828 * Allocation failed presumably due to memory. Undo anything
829 * we might have messed up.
831 * Allocating task should really be put onto the front of the
832 * wait queue, but this is pretty rare.
834 spin_lock_irq(q
->queue_lock
);
835 freed_request(q
, rw_flags
);
838 * in the very unlikely event that allocation failed and no
839 * requests for this direction was pending, mark us starved
840 * so that freeing of a request in the other direction will
841 * notice us. another possible fix would be to split the
842 * rq mempool into READ and WRITE
845 if (unlikely(rl
->count
[is_sync
] == 0))
846 rl
->starved
[is_sync
] = 1;
852 * ioc may be NULL here, and ioc_batching will be false. That's
853 * OK, if the queue is under the request limit then requests need
854 * not count toward the nr_batch_requests limit. There will always
855 * be some limit enforced by BLK_BATCH_TIME.
857 if (ioc_batching(q
, ioc
))
858 ioc
->nr_batch_requests
--;
860 trace_block_getrq(q
, bio
, rw_flags
& 1);
866 * get_request_wait - get a free request with retry
867 * @q: request_queue to allocate request from
868 * @rw_flags: RW and SYNC flags
869 * @bio: bio to allocate request for (can be %NULL)
871 * Get a free request from @q. This function keeps retrying under memory
872 * pressure and fails iff @q is dead.
874 * Must be callled with @q->queue_lock held and,
875 * Returns %NULL on failure, with @q->queue_lock held.
876 * Returns !%NULL on success, with @q->queue_lock *not held*.
878 static struct request
*get_request_wait(struct request_queue
*q
, int rw_flags
,
881 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
884 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
887 struct io_context
*ioc
;
888 struct request_list
*rl
= &q
->rq
;
890 if (unlikely(blk_queue_dead(q
)))
893 prepare_to_wait_exclusive(&rl
->wait
[is_sync
], &wait
,
894 TASK_UNINTERRUPTIBLE
);
896 trace_block_sleeprq(q
, bio
, rw_flags
& 1);
898 spin_unlock_irq(q
->queue_lock
);
902 * After sleeping, we become a "batching" process and
903 * will be able to allocate at least one request, and
904 * up to a big batch of them for a small period time.
905 * See ioc_batching, ioc_set_batching
907 ioc
= current_io_context(GFP_NOIO
, q
->node
);
908 ioc_set_batching(q
, ioc
);
910 spin_lock_irq(q
->queue_lock
);
911 finish_wait(&rl
->wait
[is_sync
], &wait
);
913 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
919 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
923 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
925 spin_lock_irq(q
->queue_lock
);
926 if (gfp_mask
& __GFP_WAIT
)
927 rq
= get_request_wait(q
, rw
, NULL
);
929 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
931 spin_unlock_irq(q
->queue_lock
);
932 /* q->queue_lock is unlocked at this point */
936 EXPORT_SYMBOL(blk_get_request
);
939 * blk_make_request - given a bio, allocate a corresponding struct request.
940 * @q: target request queue
941 * @bio: The bio describing the memory mappings that will be submitted for IO.
942 * It may be a chained-bio properly constructed by block/bio layer.
943 * @gfp_mask: gfp flags to be used for memory allocation
945 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
946 * type commands. Where the struct request needs to be farther initialized by
947 * the caller. It is passed a &struct bio, which describes the memory info of
950 * The caller of blk_make_request must make sure that bi_io_vec
951 * are set to describe the memory buffers. That bio_data_dir() will return
952 * the needed direction of the request. (And all bio's in the passed bio-chain
953 * are properly set accordingly)
955 * If called under none-sleepable conditions, mapped bio buffers must not
956 * need bouncing, by calling the appropriate masked or flagged allocator,
957 * suitable for the target device. Otherwise the call to blk_queue_bounce will
960 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
961 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
962 * anything but the first bio in the chain. Otherwise you risk waiting for IO
963 * completion of a bio that hasn't been submitted yet, thus resulting in a
964 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
965 * of bio_alloc(), as that avoids the mempool deadlock.
966 * If possible a big IO should be split into smaller parts when allocation
967 * fails. Partial allocation should not be an error, or you risk a live-lock.
969 struct request
*blk_make_request(struct request_queue
*q
, struct bio
*bio
,
972 struct request
*rq
= blk_get_request(q
, bio_data_dir(bio
), gfp_mask
);
975 return ERR_PTR(-ENOMEM
);
978 struct bio
*bounce_bio
= bio
;
981 blk_queue_bounce(q
, &bounce_bio
);
982 ret
= blk_rq_append_bio(q
, rq
, bounce_bio
);
991 EXPORT_SYMBOL(blk_make_request
);
994 * blk_requeue_request - put a request back on queue
995 * @q: request queue where request should be inserted
996 * @rq: request to be inserted
999 * Drivers often keep queueing requests until the hardware cannot accept
1000 * more, when that condition happens we need to put the request back
1001 * on the queue. Must be called with queue lock held.
1003 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
1005 blk_delete_timer(rq
);
1006 blk_clear_rq_complete(rq
);
1007 trace_block_rq_requeue(q
, rq
);
1009 if (blk_rq_tagged(rq
))
1010 blk_queue_end_tag(q
, rq
);
1012 BUG_ON(blk_queued_rq(rq
));
1014 elv_requeue_request(q
, rq
);
1016 EXPORT_SYMBOL(blk_requeue_request
);
1018 static void add_acct_request(struct request_queue
*q
, struct request
*rq
,
1021 drive_stat_acct(rq
, 1);
1022 __elv_add_request(q
, rq
, where
);
1025 static void part_round_stats_single(int cpu
, struct hd_struct
*part
,
1028 if (now
== part
->stamp
)
1031 if (part_in_flight(part
)) {
1032 __part_stat_add(cpu
, part
, time_in_queue
,
1033 part_in_flight(part
) * (now
- part
->stamp
));
1034 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
1040 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1041 * @cpu: cpu number for stats access
1042 * @part: target partition
1044 * The average IO queue length and utilisation statistics are maintained
1045 * by observing the current state of the queue length and the amount of
1046 * time it has been in this state for.
1048 * Normally, that accounting is done on IO completion, but that can result
1049 * in more than a second's worth of IO being accounted for within any one
1050 * second, leading to >100% utilisation. To deal with that, we call this
1051 * function to do a round-off before returning the results when reading
1052 * /proc/diskstats. This accounts immediately for all queue usage up to
1053 * the current jiffies and restarts the counters again.
1055 void part_round_stats(int cpu
, struct hd_struct
*part
)
1057 unsigned long now
= jiffies
;
1060 part_round_stats_single(cpu
, &part_to_disk(part
)->part0
, now
);
1061 part_round_stats_single(cpu
, part
, now
);
1063 EXPORT_SYMBOL_GPL(part_round_stats
);
1066 * queue lock must be held
1068 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1072 if (unlikely(--req
->ref_count
))
1075 elv_completed_request(q
, req
);
1077 /* this is a bio leak */
1078 WARN_ON(req
->bio
!= NULL
);
1081 * Request may not have originated from ll_rw_blk. if not,
1082 * it didn't come out of our reserved rq pools
1084 if (req
->cmd_flags
& REQ_ALLOCED
) {
1085 unsigned int flags
= req
->cmd_flags
;
1087 BUG_ON(!list_empty(&req
->queuelist
));
1088 BUG_ON(!hlist_unhashed(&req
->hash
));
1090 blk_free_request(q
, req
);
1091 freed_request(q
, flags
);
1094 EXPORT_SYMBOL_GPL(__blk_put_request
);
1096 void blk_put_request(struct request
*req
)
1098 unsigned long flags
;
1099 struct request_queue
*q
= req
->q
;
1101 spin_lock_irqsave(q
->queue_lock
, flags
);
1102 __blk_put_request(q
, req
);
1103 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1105 EXPORT_SYMBOL(blk_put_request
);
1108 * blk_add_request_payload - add a payload to a request
1109 * @rq: request to update
1110 * @page: page backing the payload
1111 * @len: length of the payload.
1113 * This allows to later add a payload to an already submitted request by
1114 * a block driver. The driver needs to take care of freeing the payload
1117 * Note that this is a quite horrible hack and nothing but handling of
1118 * discard requests should ever use it.
1120 void blk_add_request_payload(struct request
*rq
, struct page
*page
,
1123 struct bio
*bio
= rq
->bio
;
1125 bio
->bi_io_vec
->bv_page
= page
;
1126 bio
->bi_io_vec
->bv_offset
= 0;
1127 bio
->bi_io_vec
->bv_len
= len
;
1131 bio
->bi_phys_segments
= 1;
1133 rq
->__data_len
= rq
->resid_len
= len
;
1134 rq
->nr_phys_segments
= 1;
1135 rq
->buffer
= bio_data(bio
);
1137 EXPORT_SYMBOL_GPL(blk_add_request_payload
);
1139 static bool bio_attempt_back_merge(struct request_queue
*q
, struct request
*req
,
1142 const int ff
= bio
->bi_rw
& REQ_FAILFAST_MASK
;
1144 if (!ll_back_merge_fn(q
, req
, bio
))
1147 trace_block_bio_backmerge(q
, bio
);
1149 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1150 blk_rq_set_mixed_merge(req
);
1152 req
->biotail
->bi_next
= bio
;
1154 req
->__data_len
+= bio
->bi_size
;
1155 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1157 drive_stat_acct(req
, 0);
1158 elv_bio_merged(q
, req
, bio
);
1162 static bool bio_attempt_front_merge(struct request_queue
*q
,
1163 struct request
*req
, struct bio
*bio
)
1165 const int ff
= bio
->bi_rw
& REQ_FAILFAST_MASK
;
1167 if (!ll_front_merge_fn(q
, req
, bio
))
1170 trace_block_bio_frontmerge(q
, bio
);
1172 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1173 blk_rq_set_mixed_merge(req
);
1175 bio
->bi_next
= req
->bio
;
1179 * may not be valid. if the low level driver said
1180 * it didn't need a bounce buffer then it better
1181 * not touch req->buffer either...
1183 req
->buffer
= bio_data(bio
);
1184 req
->__sector
= bio
->bi_sector
;
1185 req
->__data_len
+= bio
->bi_size
;
1186 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1188 drive_stat_acct(req
, 0);
1189 elv_bio_merged(q
, req
, bio
);
1194 * attempt_plug_merge - try to merge with %current's plugged list
1195 * @q: request_queue new bio is being queued at
1196 * @bio: new bio being queued
1197 * @request_count: out parameter for number of traversed plugged requests
1199 * Determine whether @bio being queued on @q can be merged with a request
1200 * on %current's plugged list. Returns %true if merge was successful,
1203 * This function is called without @q->queue_lock; however, elevator is
1204 * accessed iff there already are requests on the plugged list which in
1205 * turn guarantees validity of the elevator.
1207 * Note that, on successful merge, elevator operation
1208 * elevator_bio_merged_fn() will be called without queue lock. Elevator
1209 * must be ready for this.
1211 static bool attempt_plug_merge(struct request_queue
*q
, struct bio
*bio
,
1212 unsigned int *request_count
)
1214 struct blk_plug
*plug
;
1218 plug
= current
->plug
;
1223 list_for_each_entry_reverse(rq
, &plug
->list
, queuelist
) {
1231 el_ret
= elv_try_merge(rq
, bio
);
1232 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1233 ret
= bio_attempt_back_merge(q
, rq
, bio
);
1236 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1237 ret
= bio_attempt_front_merge(q
, rq
, bio
);
1246 void init_request_from_bio(struct request
*req
, struct bio
*bio
)
1248 req
->cmd_type
= REQ_TYPE_FS
;
1250 req
->cmd_flags
|= bio
->bi_rw
& REQ_COMMON_MASK
;
1251 if (bio
->bi_rw
& REQ_RAHEAD
)
1252 req
->cmd_flags
|= REQ_FAILFAST_MASK
;
1255 req
->__sector
= bio
->bi_sector
;
1256 req
->ioprio
= bio_prio(bio
);
1257 blk_rq_bio_prep(req
->q
, req
, bio
);
1260 void blk_queue_bio(struct request_queue
*q
, struct bio
*bio
)
1262 const bool sync
= !!(bio
->bi_rw
& REQ_SYNC
);
1263 struct blk_plug
*plug
;
1264 int el_ret
, rw_flags
, where
= ELEVATOR_INSERT_SORT
;
1265 struct request
*req
;
1266 unsigned int request_count
= 0;
1269 * low level driver can indicate that it wants pages above a
1270 * certain limit bounced to low memory (ie for highmem, or even
1271 * ISA dma in theory)
1273 blk_queue_bounce(q
, &bio
);
1275 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) {
1276 spin_lock_irq(q
->queue_lock
);
1277 where
= ELEVATOR_INSERT_FLUSH
;
1282 * Check if we can merge with the plugged list before grabbing
1285 if (attempt_plug_merge(q
, bio
, &request_count
))
1288 spin_lock_irq(q
->queue_lock
);
1290 el_ret
= elv_merge(q
, &req
, bio
);
1291 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1292 if (bio_attempt_back_merge(q
, req
, bio
)) {
1293 if (!attempt_back_merge(q
, req
))
1294 elv_merged_request(q
, req
, el_ret
);
1297 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1298 if (bio_attempt_front_merge(q
, req
, bio
)) {
1299 if (!attempt_front_merge(q
, req
))
1300 elv_merged_request(q
, req
, el_ret
);
1307 * This sync check and mask will be re-done in init_request_from_bio(),
1308 * but we need to set it earlier to expose the sync flag to the
1309 * rq allocator and io schedulers.
1311 rw_flags
= bio_data_dir(bio
);
1313 rw_flags
|= REQ_SYNC
;
1316 * Grab a free request. This is might sleep but can not fail.
1317 * Returns with the queue unlocked.
1319 req
= get_request_wait(q
, rw_flags
, bio
);
1320 if (unlikely(!req
)) {
1321 bio_endio(bio
, -ENODEV
); /* @q is dead */
1326 * After dropping the lock and possibly sleeping here, our request
1327 * may now be mergeable after it had proven unmergeable (above).
1328 * We don't worry about that case for efficiency. It won't happen
1329 * often, and the elevators are able to handle it.
1331 init_request_from_bio(req
, bio
);
1333 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
))
1334 req
->cpu
= raw_smp_processor_id();
1336 plug
= current
->plug
;
1339 * If this is the first request added after a plug, fire
1340 * of a plug trace. If others have been added before, check
1341 * if we have multiple devices in this plug. If so, make a
1342 * note to sort the list before dispatch.
1344 if (list_empty(&plug
->list
))
1345 trace_block_plug(q
);
1347 if (!plug
->should_sort
) {
1348 struct request
*__rq
;
1350 __rq
= list_entry_rq(plug
->list
.prev
);
1352 plug
->should_sort
= 1;
1354 if (request_count
>= BLK_MAX_REQUEST_COUNT
) {
1355 blk_flush_plug_list(plug
, false);
1356 trace_block_plug(q
);
1359 list_add_tail(&req
->queuelist
, &plug
->list
);
1360 drive_stat_acct(req
, 1);
1362 spin_lock_irq(q
->queue_lock
);
1363 add_acct_request(q
, req
, where
);
1366 spin_unlock_irq(q
->queue_lock
);
1369 EXPORT_SYMBOL_GPL(blk_queue_bio
); /* for device mapper only */
1372 * If bio->bi_dev is a partition, remap the location
1374 static inline void blk_partition_remap(struct bio
*bio
)
1376 struct block_device
*bdev
= bio
->bi_bdev
;
1378 if (bio_sectors(bio
) && bdev
!= bdev
->bd_contains
) {
1379 struct hd_struct
*p
= bdev
->bd_part
;
1381 bio
->bi_sector
+= p
->start_sect
;
1382 bio
->bi_bdev
= bdev
->bd_contains
;
1384 trace_block_bio_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
1386 bio
->bi_sector
- p
->start_sect
);
1390 static void handle_bad_sector(struct bio
*bio
)
1392 char b
[BDEVNAME_SIZE
];
1394 printk(KERN_INFO
"attempt to access beyond end of device\n");
1395 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
1396 bdevname(bio
->bi_bdev
, b
),
1398 (unsigned long long)bio
->bi_sector
+ bio_sectors(bio
),
1399 (long long)(i_size_read(bio
->bi_bdev
->bd_inode
) >> 9));
1401 set_bit(BIO_EOF
, &bio
->bi_flags
);
1404 #ifdef CONFIG_FAIL_MAKE_REQUEST
1406 static DECLARE_FAULT_ATTR(fail_make_request
);
1408 static int __init
setup_fail_make_request(char *str
)
1410 return setup_fault_attr(&fail_make_request
, str
);
1412 __setup("fail_make_request=", setup_fail_make_request
);
1414 static bool should_fail_request(struct hd_struct
*part
, unsigned int bytes
)
1416 return part
->make_it_fail
&& should_fail(&fail_make_request
, bytes
);
1419 static int __init
fail_make_request_debugfs(void)
1421 struct dentry
*dir
= fault_create_debugfs_attr("fail_make_request",
1422 NULL
, &fail_make_request
);
1424 return IS_ERR(dir
) ? PTR_ERR(dir
) : 0;
1427 late_initcall(fail_make_request_debugfs
);
1429 #else /* CONFIG_FAIL_MAKE_REQUEST */
1431 static inline bool should_fail_request(struct hd_struct
*part
,
1437 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1440 * Check whether this bio extends beyond the end of the device.
1442 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
1449 /* Test device or partition size, when known. */
1450 maxsector
= i_size_read(bio
->bi_bdev
->bd_inode
) >> 9;
1452 sector_t sector
= bio
->bi_sector
;
1454 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
1456 * This may well happen - the kernel calls bread()
1457 * without checking the size of the device, e.g., when
1458 * mounting a device.
1460 handle_bad_sector(bio
);
1468 static noinline_for_stack
bool
1469 generic_make_request_checks(struct bio
*bio
)
1471 struct request_queue
*q
;
1472 int nr_sectors
= bio_sectors(bio
);
1474 char b
[BDEVNAME_SIZE
];
1475 struct hd_struct
*part
;
1479 if (bio_check_eod(bio
, nr_sectors
))
1482 q
= bdev_get_queue(bio
->bi_bdev
);
1485 "generic_make_request: Trying to access "
1486 "nonexistent block-device %s (%Lu)\n",
1487 bdevname(bio
->bi_bdev
, b
),
1488 (long long) bio
->bi_sector
);
1492 if (unlikely(!(bio
->bi_rw
& REQ_DISCARD
) &&
1493 nr_sectors
> queue_max_hw_sectors(q
))) {
1494 printk(KERN_ERR
"bio too big device %s (%u > %u)\n",
1495 bdevname(bio
->bi_bdev
, b
),
1497 queue_max_hw_sectors(q
));
1501 part
= bio
->bi_bdev
->bd_part
;
1502 if (should_fail_request(part
, bio
->bi_size
) ||
1503 should_fail_request(&part_to_disk(part
)->part0
,
1508 * If this device has partitions, remap block n
1509 * of partition p to block n+start(p) of the disk.
1511 blk_partition_remap(bio
);
1513 if (bio_integrity_enabled(bio
) && bio_integrity_prep(bio
))
1516 if (bio_check_eod(bio
, nr_sectors
))
1520 * Filter flush bio's early so that make_request based
1521 * drivers without flush support don't have to worry
1524 if ((bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) && !q
->flush_flags
) {
1525 bio
->bi_rw
&= ~(REQ_FLUSH
| REQ_FUA
);
1532 if ((bio
->bi_rw
& REQ_DISCARD
) &&
1533 (!blk_queue_discard(q
) ||
1534 ((bio
->bi_rw
& REQ_SECURE
) &&
1535 !blk_queue_secdiscard(q
)))) {
1540 if (blk_throtl_bio(q
, bio
))
1541 return false; /* throttled, will be resubmitted later */
1543 trace_block_bio_queue(q
, bio
);
1547 bio_endio(bio
, err
);
1552 * generic_make_request - hand a buffer to its device driver for I/O
1553 * @bio: The bio describing the location in memory and on the device.
1555 * generic_make_request() is used to make I/O requests of block
1556 * devices. It is passed a &struct bio, which describes the I/O that needs
1559 * generic_make_request() does not return any status. The
1560 * success/failure status of the request, along with notification of
1561 * completion, is delivered asynchronously through the bio->bi_end_io
1562 * function described (one day) else where.
1564 * The caller of generic_make_request must make sure that bi_io_vec
1565 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1566 * set to describe the device address, and the
1567 * bi_end_io and optionally bi_private are set to describe how
1568 * completion notification should be signaled.
1570 * generic_make_request and the drivers it calls may use bi_next if this
1571 * bio happens to be merged with someone else, and may resubmit the bio to
1572 * a lower device by calling into generic_make_request recursively, which
1573 * means the bio should NOT be touched after the call to ->make_request_fn.
1575 void generic_make_request(struct bio
*bio
)
1577 struct bio_list bio_list_on_stack
;
1579 if (!generic_make_request_checks(bio
))
1583 * We only want one ->make_request_fn to be active at a time, else
1584 * stack usage with stacked devices could be a problem. So use
1585 * current->bio_list to keep a list of requests submited by a
1586 * make_request_fn function. current->bio_list is also used as a
1587 * flag to say if generic_make_request is currently active in this
1588 * task or not. If it is NULL, then no make_request is active. If
1589 * it is non-NULL, then a make_request is active, and new requests
1590 * should be added at the tail
1592 if (current
->bio_list
) {
1593 bio_list_add(current
->bio_list
, bio
);
1597 /* following loop may be a bit non-obvious, and so deserves some
1599 * Before entering the loop, bio->bi_next is NULL (as all callers
1600 * ensure that) so we have a list with a single bio.
1601 * We pretend that we have just taken it off a longer list, so
1602 * we assign bio_list to a pointer to the bio_list_on_stack,
1603 * thus initialising the bio_list of new bios to be
1604 * added. ->make_request() may indeed add some more bios
1605 * through a recursive call to generic_make_request. If it
1606 * did, we find a non-NULL value in bio_list and re-enter the loop
1607 * from the top. In this case we really did just take the bio
1608 * of the top of the list (no pretending) and so remove it from
1609 * bio_list, and call into ->make_request() again.
1611 BUG_ON(bio
->bi_next
);
1612 bio_list_init(&bio_list_on_stack
);
1613 current
->bio_list
= &bio_list_on_stack
;
1615 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
1617 q
->make_request_fn(q
, bio
);
1619 bio
= bio_list_pop(current
->bio_list
);
1621 current
->bio_list
= NULL
; /* deactivate */
1623 EXPORT_SYMBOL(generic_make_request
);
1626 * submit_bio - submit a bio to the block device layer for I/O
1627 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1628 * @bio: The &struct bio which describes the I/O
1630 * submit_bio() is very similar in purpose to generic_make_request(), and
1631 * uses that function to do most of the work. Both are fairly rough
1632 * interfaces; @bio must be presetup and ready for I/O.
1635 void submit_bio(int rw
, struct bio
*bio
)
1637 int count
= bio_sectors(bio
);
1642 * If it's a regular read/write or a barrier with data attached,
1643 * go through the normal accounting stuff before submission.
1645 if (bio_has_data(bio
) && !(rw
& REQ_DISCARD
)) {
1647 count_vm_events(PGPGOUT
, count
);
1649 task_io_account_read(bio
->bi_size
);
1650 count_vm_events(PGPGIN
, count
);
1653 if (unlikely(block_dump
)) {
1654 char b
[BDEVNAME_SIZE
];
1655 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s (%u sectors)\n",
1656 current
->comm
, task_pid_nr(current
),
1657 (rw
& WRITE
) ? "WRITE" : "READ",
1658 (unsigned long long)bio
->bi_sector
,
1659 bdevname(bio
->bi_bdev
, b
),
1664 generic_make_request(bio
);
1666 EXPORT_SYMBOL(submit_bio
);
1669 * blk_rq_check_limits - Helper function to check a request for the queue limit
1671 * @rq: the request being checked
1674 * @rq may have been made based on weaker limitations of upper-level queues
1675 * in request stacking drivers, and it may violate the limitation of @q.
1676 * Since the block layer and the underlying device driver trust @rq
1677 * after it is inserted to @q, it should be checked against @q before
1678 * the insertion using this generic function.
1680 * This function should also be useful for request stacking drivers
1681 * in some cases below, so export this function.
1682 * Request stacking drivers like request-based dm may change the queue
1683 * limits while requests are in the queue (e.g. dm's table swapping).
1684 * Such request stacking drivers should check those requests agaist
1685 * the new queue limits again when they dispatch those requests,
1686 * although such checkings are also done against the old queue limits
1687 * when submitting requests.
1689 int blk_rq_check_limits(struct request_queue
*q
, struct request
*rq
)
1691 if (rq
->cmd_flags
& REQ_DISCARD
)
1694 if (blk_rq_sectors(rq
) > queue_max_sectors(q
) ||
1695 blk_rq_bytes(rq
) > queue_max_hw_sectors(q
) << 9) {
1696 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
1701 * queue's settings related to segment counting like q->bounce_pfn
1702 * may differ from that of other stacking queues.
1703 * Recalculate it to check the request correctly on this queue's
1706 blk_recalc_rq_segments(rq
);
1707 if (rq
->nr_phys_segments
> queue_max_segments(q
)) {
1708 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
1714 EXPORT_SYMBOL_GPL(blk_rq_check_limits
);
1717 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1718 * @q: the queue to submit the request
1719 * @rq: the request being queued
1721 int blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
1723 unsigned long flags
;
1724 int where
= ELEVATOR_INSERT_BACK
;
1726 if (blk_rq_check_limits(q
, rq
))
1730 should_fail_request(&rq
->rq_disk
->part0
, blk_rq_bytes(rq
)))
1733 spin_lock_irqsave(q
->queue_lock
, flags
);
1734 if (unlikely(blk_queue_dead(q
))) {
1735 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1740 * Submitting request must be dequeued before calling this function
1741 * because it will be linked to another request_queue
1743 BUG_ON(blk_queued_rq(rq
));
1745 if (rq
->cmd_flags
& (REQ_FLUSH
|REQ_FUA
))
1746 where
= ELEVATOR_INSERT_FLUSH
;
1748 add_acct_request(q
, rq
, where
);
1749 if (where
== ELEVATOR_INSERT_FLUSH
)
1751 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1755 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
1758 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1759 * @rq: request to examine
1762 * A request could be merge of IOs which require different failure
1763 * handling. This function determines the number of bytes which
1764 * can be failed from the beginning of the request without
1765 * crossing into area which need to be retried further.
1768 * The number of bytes to fail.
1771 * queue_lock must be held.
1773 unsigned int blk_rq_err_bytes(const struct request
*rq
)
1775 unsigned int ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
1776 unsigned int bytes
= 0;
1779 if (!(rq
->cmd_flags
& REQ_MIXED_MERGE
))
1780 return blk_rq_bytes(rq
);
1783 * Currently the only 'mixing' which can happen is between
1784 * different fastfail types. We can safely fail portions
1785 * which have all the failfast bits that the first one has -
1786 * the ones which are at least as eager to fail as the first
1789 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
1790 if ((bio
->bi_rw
& ff
) != ff
)
1792 bytes
+= bio
->bi_size
;
1795 /* this could lead to infinite loop */
1796 BUG_ON(blk_rq_bytes(rq
) && !bytes
);
1799 EXPORT_SYMBOL_GPL(blk_rq_err_bytes
);
1801 static void blk_account_io_completion(struct request
*req
, unsigned int bytes
)
1803 if (blk_do_io_stat(req
)) {
1804 const int rw
= rq_data_dir(req
);
1805 struct hd_struct
*part
;
1808 cpu
= part_stat_lock();
1810 part_stat_add(cpu
, part
, sectors
[rw
], bytes
>> 9);
1815 static void blk_account_io_done(struct request
*req
)
1818 * Account IO completion. flush_rq isn't accounted as a
1819 * normal IO on queueing nor completion. Accounting the
1820 * containing request is enough.
1822 if (blk_do_io_stat(req
) && !(req
->cmd_flags
& REQ_FLUSH_SEQ
)) {
1823 unsigned long duration
= jiffies
- req
->start_time
;
1824 const int rw
= rq_data_dir(req
);
1825 struct hd_struct
*part
;
1828 cpu
= part_stat_lock();
1831 part_stat_inc(cpu
, part
, ios
[rw
]);
1832 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
1833 part_round_stats(cpu
, part
);
1834 part_dec_in_flight(part
, rw
);
1836 hd_struct_put(part
);
1842 * blk_peek_request - peek at the top of a request queue
1843 * @q: request queue to peek at
1846 * Return the request at the top of @q. The returned request
1847 * should be started using blk_start_request() before LLD starts
1851 * Pointer to the request at the top of @q if available. Null
1855 * queue_lock must be held.
1857 struct request
*blk_peek_request(struct request_queue
*q
)
1862 while ((rq
= __elv_next_request(q
)) != NULL
) {
1863 if (!(rq
->cmd_flags
& REQ_STARTED
)) {
1865 * This is the first time the device driver
1866 * sees this request (possibly after
1867 * requeueing). Notify IO scheduler.
1869 if (rq
->cmd_flags
& REQ_SORTED
)
1870 elv_activate_rq(q
, rq
);
1873 * just mark as started even if we don't start
1874 * it, a request that has been delayed should
1875 * not be passed by new incoming requests
1877 rq
->cmd_flags
|= REQ_STARTED
;
1878 trace_block_rq_issue(q
, rq
);
1881 if (!q
->boundary_rq
|| q
->boundary_rq
== rq
) {
1882 q
->end_sector
= rq_end_sector(rq
);
1883 q
->boundary_rq
= NULL
;
1886 if (rq
->cmd_flags
& REQ_DONTPREP
)
1889 if (q
->dma_drain_size
&& blk_rq_bytes(rq
)) {
1891 * make sure space for the drain appears we
1892 * know we can do this because max_hw_segments
1893 * has been adjusted to be one fewer than the
1896 rq
->nr_phys_segments
++;
1902 ret
= q
->prep_rq_fn(q
, rq
);
1903 if (ret
== BLKPREP_OK
) {
1905 } else if (ret
== BLKPREP_DEFER
) {
1907 * the request may have been (partially) prepped.
1908 * we need to keep this request in the front to
1909 * avoid resource deadlock. REQ_STARTED will
1910 * prevent other fs requests from passing this one.
1912 if (q
->dma_drain_size
&& blk_rq_bytes(rq
) &&
1913 !(rq
->cmd_flags
& REQ_DONTPREP
)) {
1915 * remove the space for the drain we added
1916 * so that we don't add it again
1918 --rq
->nr_phys_segments
;
1923 } else if (ret
== BLKPREP_KILL
) {
1924 rq
->cmd_flags
|= REQ_QUIET
;
1926 * Mark this request as started so we don't trigger
1927 * any debug logic in the end I/O path.
1929 blk_start_request(rq
);
1930 __blk_end_request_all(rq
, -EIO
);
1932 printk(KERN_ERR
"%s: bad return=%d\n", __func__
, ret
);
1939 EXPORT_SYMBOL(blk_peek_request
);
1941 void blk_dequeue_request(struct request
*rq
)
1943 struct request_queue
*q
= rq
->q
;
1945 BUG_ON(list_empty(&rq
->queuelist
));
1946 BUG_ON(ELV_ON_HASH(rq
));
1948 list_del_init(&rq
->queuelist
);
1951 * the time frame between a request being removed from the lists
1952 * and to it is freed is accounted as io that is in progress at
1955 if (blk_account_rq(rq
)) {
1956 q
->in_flight
[rq_is_sync(rq
)]++;
1957 set_io_start_time_ns(rq
);
1962 * blk_start_request - start request processing on the driver
1963 * @req: request to dequeue
1966 * Dequeue @req and start timeout timer on it. This hands off the
1967 * request to the driver.
1969 * Block internal functions which don't want to start timer should
1970 * call blk_dequeue_request().
1973 * queue_lock must be held.
1975 void blk_start_request(struct request
*req
)
1977 blk_dequeue_request(req
);
1980 * We are now handing the request to the hardware, initialize
1981 * resid_len to full count and add the timeout handler.
1983 req
->resid_len
= blk_rq_bytes(req
);
1984 if (unlikely(blk_bidi_rq(req
)))
1985 req
->next_rq
->resid_len
= blk_rq_bytes(req
->next_rq
);
1989 EXPORT_SYMBOL(blk_start_request
);
1992 * blk_fetch_request - fetch a request from a request queue
1993 * @q: request queue to fetch a request from
1996 * Return the request at the top of @q. The request is started on
1997 * return and LLD can start processing it immediately.
2000 * Pointer to the request at the top of @q if available. Null
2004 * queue_lock must be held.
2006 struct request
*blk_fetch_request(struct request_queue
*q
)
2010 rq
= blk_peek_request(q
);
2012 blk_start_request(rq
);
2015 EXPORT_SYMBOL(blk_fetch_request
);
2018 * blk_update_request - Special helper function for request stacking drivers
2019 * @req: the request being processed
2020 * @error: %0 for success, < %0 for error
2021 * @nr_bytes: number of bytes to complete @req
2024 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2025 * the request structure even if @req doesn't have leftover.
2026 * If @req has leftover, sets it up for the next range of segments.
2028 * This special helper function is only for request stacking drivers
2029 * (e.g. request-based dm) so that they can handle partial completion.
2030 * Actual device drivers should use blk_end_request instead.
2032 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2033 * %false return from this function.
2036 * %false - this request doesn't have any more data
2037 * %true - this request has more data
2039 bool blk_update_request(struct request
*req
, int error
, unsigned int nr_bytes
)
2041 int total_bytes
, bio_nbytes
, next_idx
= 0;
2047 trace_block_rq_complete(req
->q
, req
);
2050 * For fs requests, rq is just carrier of independent bio's
2051 * and each partial completion should be handled separately.
2052 * Reset per-request error on each partial completion.
2054 * TODO: tj: This is too subtle. It would be better to let
2055 * low level drivers do what they see fit.
2057 if (req
->cmd_type
== REQ_TYPE_FS
)
2060 if (error
&& req
->cmd_type
== REQ_TYPE_FS
&&
2061 !(req
->cmd_flags
& REQ_QUIET
)) {
2066 error_type
= "recoverable transport";
2069 error_type
= "critical target";
2072 error_type
= "critical nexus";
2079 printk(KERN_ERR
"end_request: %s error, dev %s, sector %llu\n",
2080 error_type
, req
->rq_disk
? req
->rq_disk
->disk_name
: "?",
2081 (unsigned long long)blk_rq_pos(req
));
2084 blk_account_io_completion(req
, nr_bytes
);
2086 total_bytes
= bio_nbytes
= 0;
2087 while ((bio
= req
->bio
) != NULL
) {
2090 if (nr_bytes
>= bio
->bi_size
) {
2091 req
->bio
= bio
->bi_next
;
2092 nbytes
= bio
->bi_size
;
2093 req_bio_endio(req
, bio
, nbytes
, error
);
2097 int idx
= bio
->bi_idx
+ next_idx
;
2099 if (unlikely(idx
>= bio
->bi_vcnt
)) {
2100 blk_dump_rq_flags(req
, "__end_that");
2101 printk(KERN_ERR
"%s: bio idx %d >= vcnt %d\n",
2102 __func__
, idx
, bio
->bi_vcnt
);
2106 nbytes
= bio_iovec_idx(bio
, idx
)->bv_len
;
2107 BIO_BUG_ON(nbytes
> bio
->bi_size
);
2110 * not a complete bvec done
2112 if (unlikely(nbytes
> nr_bytes
)) {
2113 bio_nbytes
+= nr_bytes
;
2114 total_bytes
+= nr_bytes
;
2119 * advance to the next vector
2122 bio_nbytes
+= nbytes
;
2125 total_bytes
+= nbytes
;
2131 * end more in this run, or just return 'not-done'
2133 if (unlikely(nr_bytes
<= 0))
2143 * Reset counters so that the request stacking driver
2144 * can find how many bytes remain in the request
2147 req
->__data_len
= 0;
2152 * if the request wasn't completed, update state
2155 req_bio_endio(req
, bio
, bio_nbytes
, error
);
2156 bio
->bi_idx
+= next_idx
;
2157 bio_iovec(bio
)->bv_offset
+= nr_bytes
;
2158 bio_iovec(bio
)->bv_len
-= nr_bytes
;
2161 req
->__data_len
-= total_bytes
;
2162 req
->buffer
= bio_data(req
->bio
);
2164 /* update sector only for requests with clear definition of sector */
2165 if (req
->cmd_type
== REQ_TYPE_FS
|| (req
->cmd_flags
& REQ_DISCARD
))
2166 req
->__sector
+= total_bytes
>> 9;
2168 /* mixed attributes always follow the first bio */
2169 if (req
->cmd_flags
& REQ_MIXED_MERGE
) {
2170 req
->cmd_flags
&= ~REQ_FAILFAST_MASK
;
2171 req
->cmd_flags
|= req
->bio
->bi_rw
& REQ_FAILFAST_MASK
;
2175 * If total number of sectors is less than the first segment
2176 * size, something has gone terribly wrong.
2178 if (blk_rq_bytes(req
) < blk_rq_cur_bytes(req
)) {
2179 blk_dump_rq_flags(req
, "request botched");
2180 req
->__data_len
= blk_rq_cur_bytes(req
);
2183 /* recalculate the number of segments */
2184 blk_recalc_rq_segments(req
);
2188 EXPORT_SYMBOL_GPL(blk_update_request
);
2190 static bool blk_update_bidi_request(struct request
*rq
, int error
,
2191 unsigned int nr_bytes
,
2192 unsigned int bidi_bytes
)
2194 if (blk_update_request(rq
, error
, nr_bytes
))
2197 /* Bidi request must be completed as a whole */
2198 if (unlikely(blk_bidi_rq(rq
)) &&
2199 blk_update_request(rq
->next_rq
, error
, bidi_bytes
))
2202 if (blk_queue_add_random(rq
->q
))
2203 add_disk_randomness(rq
->rq_disk
);
2209 * blk_unprep_request - unprepare a request
2212 * This function makes a request ready for complete resubmission (or
2213 * completion). It happens only after all error handling is complete,
2214 * so represents the appropriate moment to deallocate any resources
2215 * that were allocated to the request in the prep_rq_fn. The queue
2216 * lock is held when calling this.
2218 void blk_unprep_request(struct request
*req
)
2220 struct request_queue
*q
= req
->q
;
2222 req
->cmd_flags
&= ~REQ_DONTPREP
;
2223 if (q
->unprep_rq_fn
)
2224 q
->unprep_rq_fn(q
, req
);
2226 EXPORT_SYMBOL_GPL(blk_unprep_request
);
2229 * queue lock must be held
2231 static void blk_finish_request(struct request
*req
, int error
)
2233 if (blk_rq_tagged(req
))
2234 blk_queue_end_tag(req
->q
, req
);
2236 BUG_ON(blk_queued_rq(req
));
2238 if (unlikely(laptop_mode
) && req
->cmd_type
== REQ_TYPE_FS
)
2239 laptop_io_completion(&req
->q
->backing_dev_info
);
2241 blk_delete_timer(req
);
2243 if (req
->cmd_flags
& REQ_DONTPREP
)
2244 blk_unprep_request(req
);
2247 blk_account_io_done(req
);
2250 req
->end_io(req
, error
);
2252 if (blk_bidi_rq(req
))
2253 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
2255 __blk_put_request(req
->q
, req
);
2260 * blk_end_bidi_request - Complete a bidi request
2261 * @rq: the request to complete
2262 * @error: %0 for success, < %0 for error
2263 * @nr_bytes: number of bytes to complete @rq
2264 * @bidi_bytes: number of bytes to complete @rq->next_rq
2267 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2268 * Drivers that supports bidi can safely call this member for any
2269 * type of request, bidi or uni. In the later case @bidi_bytes is
2273 * %false - we are done with this request
2274 * %true - still buffers pending for this request
2276 static bool blk_end_bidi_request(struct request
*rq
, int error
,
2277 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2279 struct request_queue
*q
= rq
->q
;
2280 unsigned long flags
;
2282 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2285 spin_lock_irqsave(q
->queue_lock
, flags
);
2286 blk_finish_request(rq
, error
);
2287 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2293 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2294 * @rq: the request to complete
2295 * @error: %0 for success, < %0 for error
2296 * @nr_bytes: number of bytes to complete @rq
2297 * @bidi_bytes: number of bytes to complete @rq->next_rq
2300 * Identical to blk_end_bidi_request() except that queue lock is
2301 * assumed to be locked on entry and remains so on return.
2304 * %false - we are done with this request
2305 * %true - still buffers pending for this request
2307 bool __blk_end_bidi_request(struct request
*rq
, int error
,
2308 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2310 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2313 blk_finish_request(rq
, error
);
2319 * blk_end_request - Helper function for drivers to complete the request.
2320 * @rq: the request being processed
2321 * @error: %0 for success, < %0 for error
2322 * @nr_bytes: number of bytes to complete
2325 * Ends I/O on a number of bytes attached to @rq.
2326 * If @rq has leftover, sets it up for the next range of segments.
2329 * %false - we are done with this request
2330 * %true - still buffers pending for this request
2332 bool blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2334 return blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2336 EXPORT_SYMBOL(blk_end_request
);
2339 * blk_end_request_all - Helper function for drives to finish the request.
2340 * @rq: the request to finish
2341 * @error: %0 for success, < %0 for error
2344 * Completely finish @rq.
2346 void blk_end_request_all(struct request
*rq
, int error
)
2349 unsigned int bidi_bytes
= 0;
2351 if (unlikely(blk_bidi_rq(rq
)))
2352 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2354 pending
= blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2357 EXPORT_SYMBOL(blk_end_request_all
);
2360 * blk_end_request_cur - Helper function to finish the current request chunk.
2361 * @rq: the request to finish the current chunk for
2362 * @error: %0 for success, < %0 for error
2365 * Complete the current consecutively mapped chunk from @rq.
2368 * %false - we are done with this request
2369 * %true - still buffers pending for this request
2371 bool blk_end_request_cur(struct request
*rq
, int error
)
2373 return blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2375 EXPORT_SYMBOL(blk_end_request_cur
);
2378 * blk_end_request_err - Finish a request till the next failure boundary.
2379 * @rq: the request to finish till the next failure boundary for
2380 * @error: must be negative errno
2383 * Complete @rq till the next failure boundary.
2386 * %false - we are done with this request
2387 * %true - still buffers pending for this request
2389 bool blk_end_request_err(struct request
*rq
, int error
)
2391 WARN_ON(error
>= 0);
2392 return blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2394 EXPORT_SYMBOL_GPL(blk_end_request_err
);
2397 * __blk_end_request - Helper function for drivers to complete the request.
2398 * @rq: the request being processed
2399 * @error: %0 for success, < %0 for error
2400 * @nr_bytes: number of bytes to complete
2403 * Must be called with queue lock held unlike blk_end_request().
2406 * %false - we are done with this request
2407 * %true - still buffers pending for this request
2409 bool __blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2411 return __blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2413 EXPORT_SYMBOL(__blk_end_request
);
2416 * __blk_end_request_all - Helper function for drives to finish the request.
2417 * @rq: the request to finish
2418 * @error: %0 for success, < %0 for error
2421 * Completely finish @rq. Must be called with queue lock held.
2423 void __blk_end_request_all(struct request
*rq
, int error
)
2426 unsigned int bidi_bytes
= 0;
2428 if (unlikely(blk_bidi_rq(rq
)))
2429 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2431 pending
= __blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2434 EXPORT_SYMBOL(__blk_end_request_all
);
2437 * __blk_end_request_cur - Helper function to finish the current request chunk.
2438 * @rq: the request to finish the current chunk for
2439 * @error: %0 for success, < %0 for error
2442 * Complete the current consecutively mapped chunk from @rq. Must
2443 * be called with queue lock held.
2446 * %false - we are done with this request
2447 * %true - still buffers pending for this request
2449 bool __blk_end_request_cur(struct request
*rq
, int error
)
2451 return __blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2453 EXPORT_SYMBOL(__blk_end_request_cur
);
2456 * __blk_end_request_err - Finish a request till the next failure boundary.
2457 * @rq: the request to finish till the next failure boundary for
2458 * @error: must be negative errno
2461 * Complete @rq till the next failure boundary. Must be called
2462 * with queue lock held.
2465 * %false - we are done with this request
2466 * %true - still buffers pending for this request
2468 bool __blk_end_request_err(struct request
*rq
, int error
)
2470 WARN_ON(error
>= 0);
2471 return __blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2473 EXPORT_SYMBOL_GPL(__blk_end_request_err
);
2475 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
2478 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2479 rq
->cmd_flags
|= bio
->bi_rw
& REQ_WRITE
;
2481 if (bio_has_data(bio
)) {
2482 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2483 rq
->buffer
= bio_data(bio
);
2485 rq
->__data_len
= bio
->bi_size
;
2486 rq
->bio
= rq
->biotail
= bio
;
2489 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2492 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2494 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2495 * @rq: the request to be flushed
2498 * Flush all pages in @rq.
2500 void rq_flush_dcache_pages(struct request
*rq
)
2502 struct req_iterator iter
;
2503 struct bio_vec
*bvec
;
2505 rq_for_each_segment(bvec
, rq
, iter
)
2506 flush_dcache_page(bvec
->bv_page
);
2508 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages
);
2512 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2513 * @q : the queue of the device being checked
2516 * Check if underlying low-level drivers of a device are busy.
2517 * If the drivers want to export their busy state, they must set own
2518 * exporting function using blk_queue_lld_busy() first.
2520 * Basically, this function is used only by request stacking drivers
2521 * to stop dispatching requests to underlying devices when underlying
2522 * devices are busy. This behavior helps more I/O merging on the queue
2523 * of the request stacking driver and prevents I/O throughput regression
2524 * on burst I/O load.
2527 * 0 - Not busy (The request stacking driver should dispatch request)
2528 * 1 - Busy (The request stacking driver should stop dispatching request)
2530 int blk_lld_busy(struct request_queue
*q
)
2533 return q
->lld_busy_fn(q
);
2537 EXPORT_SYMBOL_GPL(blk_lld_busy
);
2540 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2541 * @rq: the clone request to be cleaned up
2544 * Free all bios in @rq for a cloned request.
2546 void blk_rq_unprep_clone(struct request
*rq
)
2550 while ((bio
= rq
->bio
) != NULL
) {
2551 rq
->bio
= bio
->bi_next
;
2556 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone
);
2559 * Copy attributes of the original request to the clone request.
2560 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2562 static void __blk_rq_prep_clone(struct request
*dst
, struct request
*src
)
2564 dst
->cpu
= src
->cpu
;
2565 dst
->cmd_flags
= (src
->cmd_flags
& REQ_CLONE_MASK
) | REQ_NOMERGE
;
2566 dst
->cmd_type
= src
->cmd_type
;
2567 dst
->__sector
= blk_rq_pos(src
);
2568 dst
->__data_len
= blk_rq_bytes(src
);
2569 dst
->nr_phys_segments
= src
->nr_phys_segments
;
2570 dst
->ioprio
= src
->ioprio
;
2571 dst
->extra_len
= src
->extra_len
;
2575 * blk_rq_prep_clone - Helper function to setup clone request
2576 * @rq: the request to be setup
2577 * @rq_src: original request to be cloned
2578 * @bs: bio_set that bios for clone are allocated from
2579 * @gfp_mask: memory allocation mask for bio
2580 * @bio_ctr: setup function to be called for each clone bio.
2581 * Returns %0 for success, non %0 for failure.
2582 * @data: private data to be passed to @bio_ctr
2585 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2586 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2587 * are not copied, and copying such parts is the caller's responsibility.
2588 * Also, pages which the original bios are pointing to are not copied
2589 * and the cloned bios just point same pages.
2590 * So cloned bios must be completed before original bios, which means
2591 * the caller must complete @rq before @rq_src.
2593 int blk_rq_prep_clone(struct request
*rq
, struct request
*rq_src
,
2594 struct bio_set
*bs
, gfp_t gfp_mask
,
2595 int (*bio_ctr
)(struct bio
*, struct bio
*, void *),
2598 struct bio
*bio
, *bio_src
;
2603 blk_rq_init(NULL
, rq
);
2605 __rq_for_each_bio(bio_src
, rq_src
) {
2606 bio
= bio_alloc_bioset(gfp_mask
, bio_src
->bi_max_vecs
, bs
);
2610 __bio_clone(bio
, bio_src
);
2612 if (bio_integrity(bio_src
) &&
2613 bio_integrity_clone(bio
, bio_src
, gfp_mask
, bs
))
2616 if (bio_ctr
&& bio_ctr(bio
, bio_src
, data
))
2620 rq
->biotail
->bi_next
= bio
;
2623 rq
->bio
= rq
->biotail
= bio
;
2626 __blk_rq_prep_clone(rq
, rq_src
);
2633 blk_rq_unprep_clone(rq
);
2637 EXPORT_SYMBOL_GPL(blk_rq_prep_clone
);
2639 int kblockd_schedule_work(struct request_queue
*q
, struct work_struct
*work
)
2641 return queue_work(kblockd_workqueue
, work
);
2643 EXPORT_SYMBOL(kblockd_schedule_work
);
2645 int kblockd_schedule_delayed_work(struct request_queue
*q
,
2646 struct delayed_work
*dwork
, unsigned long delay
)
2648 return queue_delayed_work(kblockd_workqueue
, dwork
, delay
);
2650 EXPORT_SYMBOL(kblockd_schedule_delayed_work
);
2652 #define PLUG_MAGIC 0x91827364
2655 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2656 * @plug: The &struct blk_plug that needs to be initialized
2659 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2660 * pending I/O should the task end up blocking between blk_start_plug() and
2661 * blk_finish_plug(). This is important from a performance perspective, but
2662 * also ensures that we don't deadlock. For instance, if the task is blocking
2663 * for a memory allocation, memory reclaim could end up wanting to free a
2664 * page belonging to that request that is currently residing in our private
2665 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2666 * this kind of deadlock.
2668 void blk_start_plug(struct blk_plug
*plug
)
2670 struct task_struct
*tsk
= current
;
2672 plug
->magic
= PLUG_MAGIC
;
2673 INIT_LIST_HEAD(&plug
->list
);
2674 INIT_LIST_HEAD(&plug
->cb_list
);
2675 plug
->should_sort
= 0;
2678 * If this is a nested plug, don't actually assign it. It will be
2679 * flushed on its own.
2683 * Store ordering should not be needed here, since a potential
2684 * preempt will imply a full memory barrier
2689 EXPORT_SYMBOL(blk_start_plug
);
2691 static int plug_rq_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
2693 struct request
*rqa
= container_of(a
, struct request
, queuelist
);
2694 struct request
*rqb
= container_of(b
, struct request
, queuelist
);
2696 return !(rqa
->q
<= rqb
->q
);
2700 * If 'from_schedule' is true, then postpone the dispatch of requests
2701 * until a safe kblockd context. We due this to avoid accidental big
2702 * additional stack usage in driver dispatch, in places where the originally
2703 * plugger did not intend it.
2705 static void queue_unplugged(struct request_queue
*q
, unsigned int depth
,
2707 __releases(q
->queue_lock
)
2709 trace_block_unplug(q
, depth
, !from_schedule
);
2712 * Don't mess with dead queue.
2714 if (unlikely(blk_queue_dead(q
))) {
2715 spin_unlock(q
->queue_lock
);
2720 * If we are punting this to kblockd, then we can safely drop
2721 * the queue_lock before waking kblockd (which needs to take
2724 if (from_schedule
) {
2725 spin_unlock(q
->queue_lock
);
2726 blk_run_queue_async(q
);
2729 spin_unlock(q
->queue_lock
);
2734 static void flush_plug_callbacks(struct blk_plug
*plug
)
2736 LIST_HEAD(callbacks
);
2738 if (list_empty(&plug
->cb_list
))
2741 list_splice_init(&plug
->cb_list
, &callbacks
);
2743 while (!list_empty(&callbacks
)) {
2744 struct blk_plug_cb
*cb
= list_first_entry(&callbacks
,
2747 list_del(&cb
->list
);
2752 void blk_flush_plug_list(struct blk_plug
*plug
, bool from_schedule
)
2754 struct request_queue
*q
;
2755 unsigned long flags
;
2760 BUG_ON(plug
->magic
!= PLUG_MAGIC
);
2762 flush_plug_callbacks(plug
);
2763 if (list_empty(&plug
->list
))
2766 list_splice_init(&plug
->list
, &list
);
2768 if (plug
->should_sort
) {
2769 list_sort(NULL
, &list
, plug_rq_cmp
);
2770 plug
->should_sort
= 0;
2777 * Save and disable interrupts here, to avoid doing it for every
2778 * queue lock we have to take.
2780 local_irq_save(flags
);
2781 while (!list_empty(&list
)) {
2782 rq
= list_entry_rq(list
.next
);
2783 list_del_init(&rq
->queuelist
);
2787 * This drops the queue lock
2790 queue_unplugged(q
, depth
, from_schedule
);
2793 spin_lock(q
->queue_lock
);
2797 * Short-circuit if @q is dead
2799 if (unlikely(blk_queue_dead(q
))) {
2800 __blk_end_request_all(rq
, -ENODEV
);
2805 * rq is already accounted, so use raw insert
2807 if (rq
->cmd_flags
& (REQ_FLUSH
| REQ_FUA
))
2808 __elv_add_request(q
, rq
, ELEVATOR_INSERT_FLUSH
);
2810 __elv_add_request(q
, rq
, ELEVATOR_INSERT_SORT_MERGE
);
2816 * This drops the queue lock
2819 queue_unplugged(q
, depth
, from_schedule
);
2821 local_irq_restore(flags
);
2824 void blk_finish_plug(struct blk_plug
*plug
)
2826 blk_flush_plug_list(plug
, false);
2828 if (plug
== current
->plug
)
2829 current
->plug
= NULL
;
2831 EXPORT_SYMBOL(blk_finish_plug
);
2833 int __init
blk_dev_init(void)
2835 BUILD_BUG_ON(__REQ_NR_BITS
> 8 *
2836 sizeof(((struct request
*)0)->cmd_flags
));
2838 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
2839 kblockd_workqueue
= alloc_workqueue("kblockd",
2840 WQ_MEM_RECLAIM
| WQ_HIGHPRI
, 0);
2841 if (!kblockd_workqueue
)
2842 panic("Failed to create kblockd\n");
2844 request_cachep
= kmem_cache_create("blkdev_requests",
2845 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
2847 blk_requestq_cachep
= kmem_cache_create("blkdev_queue",
2848 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);