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
)
363 spin_lock_irq(q
->queue_lock
);
365 elv_drain_elevator(q
);
372 nr_rqs
= q
->rq
.count
[0] + q
->rq
.count
[1];
374 nr_rqs
= q
->rq
.elvpriv
;
376 spin_unlock_irq(q
->queue_lock
);
385 * blk_cleanup_queue - shutdown a request queue
386 * @q: request queue to shutdown
388 * Mark @q DEAD, drain all pending requests, destroy and put it. All
389 * future requests will be failed immediately with -ENODEV.
391 void blk_cleanup_queue(struct request_queue
*q
)
393 spinlock_t
*lock
= q
->queue_lock
;
395 /* mark @q DEAD, no new request or merges will be allowed afterwards */
396 mutex_lock(&q
->sysfs_lock
);
397 queue_flag_set_unlocked(QUEUE_FLAG_DEAD
, q
);
400 queue_flag_set(QUEUE_FLAG_NOMERGES
, q
);
401 queue_flag_set(QUEUE_FLAG_NOXMERGES
, q
);
402 queue_flag_set(QUEUE_FLAG_DEAD
, q
);
404 if (q
->queue_lock
!= &q
->__queue_lock
)
405 q
->queue_lock
= &q
->__queue_lock
;
407 spin_unlock_irq(lock
);
408 mutex_unlock(&q
->sysfs_lock
);
411 * Drain all requests queued before DEAD marking. The caller might
412 * be trying to tear down @q before its elevator is initialized, in
413 * which case we don't want to call into draining.
416 blk_drain_queue(q
, true);
418 /* @q won't process any more request, flush async actions */
419 del_timer_sync(&q
->backing_dev_info
.laptop_mode_wb_timer
);
422 /* @q is and will stay empty, shutdown and put */
425 EXPORT_SYMBOL(blk_cleanup_queue
);
427 static int blk_init_free_list(struct request_queue
*q
)
429 struct request_list
*rl
= &q
->rq
;
431 if (unlikely(rl
->rq_pool
))
434 rl
->count
[BLK_RW_SYNC
] = rl
->count
[BLK_RW_ASYNC
] = 0;
435 rl
->starved
[BLK_RW_SYNC
] = rl
->starved
[BLK_RW_ASYNC
] = 0;
437 init_waitqueue_head(&rl
->wait
[BLK_RW_SYNC
]);
438 init_waitqueue_head(&rl
->wait
[BLK_RW_ASYNC
]);
440 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, mempool_alloc_slab
,
441 mempool_free_slab
, request_cachep
, q
->node
);
449 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
451 return blk_alloc_queue_node(gfp_mask
, -1);
453 EXPORT_SYMBOL(blk_alloc_queue
);
455 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
457 struct request_queue
*q
;
460 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
461 gfp_mask
| __GFP_ZERO
, node_id
);
465 q
->backing_dev_info
.ra_pages
=
466 (VM_MAX_READAHEAD
* 1024) / PAGE_CACHE_SIZE
;
467 q
->backing_dev_info
.state
= 0;
468 q
->backing_dev_info
.capabilities
= BDI_CAP_MAP_COPY
;
469 q
->backing_dev_info
.name
= "block";
471 err
= bdi_init(&q
->backing_dev_info
);
473 kmem_cache_free(blk_requestq_cachep
, q
);
477 if (blk_throtl_init(q
)) {
478 kmem_cache_free(blk_requestq_cachep
, q
);
482 setup_timer(&q
->backing_dev_info
.laptop_mode_wb_timer
,
483 laptop_mode_timer_fn
, (unsigned long) q
);
484 setup_timer(&q
->timeout
, blk_rq_timed_out_timer
, (unsigned long) q
);
485 INIT_LIST_HEAD(&q
->timeout_list
);
486 INIT_LIST_HEAD(&q
->flush_queue
[0]);
487 INIT_LIST_HEAD(&q
->flush_queue
[1]);
488 INIT_LIST_HEAD(&q
->flush_data_in_flight
);
489 INIT_DELAYED_WORK(&q
->delay_work
, blk_delay_work
);
491 kobject_init(&q
->kobj
, &blk_queue_ktype
);
493 mutex_init(&q
->sysfs_lock
);
494 spin_lock_init(&q
->__queue_lock
);
497 * By default initialize queue_lock to internal lock and driver can
498 * override it later if need be.
500 q
->queue_lock
= &q
->__queue_lock
;
504 EXPORT_SYMBOL(blk_alloc_queue_node
);
507 * blk_init_queue - prepare a request queue for use with a block device
508 * @rfn: The function to be called to process requests that have been
509 * placed on the queue.
510 * @lock: Request queue spin lock
513 * If a block device wishes to use the standard request handling procedures,
514 * which sorts requests and coalesces adjacent requests, then it must
515 * call blk_init_queue(). The function @rfn will be called when there
516 * are requests on the queue that need to be processed. If the device
517 * supports plugging, then @rfn may not be called immediately when requests
518 * are available on the queue, but may be called at some time later instead.
519 * Plugged queues are generally unplugged when a buffer belonging to one
520 * of the requests on the queue is needed, or due to memory pressure.
522 * @rfn is not required, or even expected, to remove all requests off the
523 * queue, but only as many as it can handle at a time. If it does leave
524 * requests on the queue, it is responsible for arranging that the requests
525 * get dealt with eventually.
527 * The queue spin lock must be held while manipulating the requests on the
528 * request queue; this lock will be taken also from interrupt context, so irq
529 * disabling is needed for it.
531 * Function returns a pointer to the initialized request queue, or %NULL if
535 * blk_init_queue() must be paired with a blk_cleanup_queue() call
536 * when the block device is deactivated (such as at module unload).
539 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
541 return blk_init_queue_node(rfn
, lock
, -1);
543 EXPORT_SYMBOL(blk_init_queue
);
545 struct request_queue
*
546 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
548 struct request_queue
*uninit_q
, *q
;
550 uninit_q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
554 q
= blk_init_allocated_queue_node(uninit_q
, rfn
, lock
, node_id
);
556 blk_cleanup_queue(uninit_q
);
560 EXPORT_SYMBOL(blk_init_queue_node
);
562 struct request_queue
*
563 blk_init_allocated_queue(struct request_queue
*q
, request_fn_proc
*rfn
,
566 return blk_init_allocated_queue_node(q
, rfn
, lock
, -1);
568 EXPORT_SYMBOL(blk_init_allocated_queue
);
570 struct request_queue
*
571 blk_init_allocated_queue_node(struct request_queue
*q
, request_fn_proc
*rfn
,
572 spinlock_t
*lock
, int node_id
)
578 if (blk_init_free_list(q
))
582 q
->prep_rq_fn
= NULL
;
583 q
->unprep_rq_fn
= NULL
;
584 q
->queue_flags
= QUEUE_FLAG_DEFAULT
;
586 /* Override internal queue lock with supplied lock pointer */
588 q
->queue_lock
= lock
;
591 * This also sets hw/phys segments, boundary and size
593 blk_queue_make_request(q
, blk_queue_bio
);
595 q
->sg_reserved_size
= INT_MAX
;
600 if (!elevator_init(q
, NULL
)) {
601 blk_queue_congestion_threshold(q
);
607 EXPORT_SYMBOL(blk_init_allocated_queue_node
);
609 int blk_get_queue(struct request_queue
*q
)
611 if (likely(!test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
))) {
612 kobject_get(&q
->kobj
);
618 EXPORT_SYMBOL(blk_get_queue
);
620 static inline void blk_free_request(struct request_queue
*q
, struct request
*rq
)
622 if (rq
->cmd_flags
& REQ_ELVPRIV
)
623 elv_put_request(q
, rq
);
624 mempool_free(rq
, q
->rq
.rq_pool
);
627 static struct request
*
628 blk_alloc_request(struct request_queue
*q
, unsigned int flags
, gfp_t gfp_mask
)
630 struct request
*rq
= mempool_alloc(q
->rq
.rq_pool
, gfp_mask
);
637 rq
->cmd_flags
= flags
| REQ_ALLOCED
;
639 if ((flags
& REQ_ELVPRIV
) &&
640 unlikely(elv_set_request(q
, rq
, gfp_mask
))) {
641 mempool_free(rq
, q
->rq
.rq_pool
);
649 * ioc_batching returns true if the ioc is a valid batching request and
650 * should be given priority access to a request.
652 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
658 * Make sure the process is able to allocate at least 1 request
659 * even if the batch times out, otherwise we could theoretically
662 return ioc
->nr_batch_requests
== q
->nr_batching
||
663 (ioc
->nr_batch_requests
> 0
664 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
668 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
669 * will cause the process to be a "batcher" on all queues in the system. This
670 * is the behaviour we want though - once it gets a wakeup it should be given
673 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
675 if (!ioc
|| ioc_batching(q
, ioc
))
678 ioc
->nr_batch_requests
= q
->nr_batching
;
679 ioc
->last_waited
= jiffies
;
682 static void __freed_request(struct request_queue
*q
, int sync
)
684 struct request_list
*rl
= &q
->rq
;
686 if (rl
->count
[sync
] < queue_congestion_off_threshold(q
))
687 blk_clear_queue_congested(q
, sync
);
689 if (rl
->count
[sync
] + 1 <= q
->nr_requests
) {
690 if (waitqueue_active(&rl
->wait
[sync
]))
691 wake_up(&rl
->wait
[sync
]);
693 blk_clear_queue_full(q
, sync
);
698 * A request has just been released. Account for it, update the full and
699 * congestion status, wake up any waiters. Called under q->queue_lock.
701 static void freed_request(struct request_queue
*q
, unsigned int flags
)
703 struct request_list
*rl
= &q
->rq
;
704 int sync
= rw_is_sync(flags
);
707 if (flags
& REQ_ELVPRIV
)
710 __freed_request(q
, sync
);
712 if (unlikely(rl
->starved
[sync
^ 1]))
713 __freed_request(q
, sync
^ 1);
717 * Determine if elevator data should be initialized when allocating the
718 * request associated with @bio.
720 static bool blk_rq_should_init_elevator(struct bio
*bio
)
726 * Flush requests do not use the elevator so skip initialization.
727 * This allows a request to share the flush and elevator data.
729 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
))
736 * get_request - get a free request
737 * @q: request_queue to allocate request from
738 * @rw_flags: RW and SYNC flags
739 * @bio: bio to allocate request for (can be %NULL)
740 * @gfp_mask: allocation mask
742 * Get a free request from @q. This function may fail under memory
743 * pressure or if @q is dead.
745 * Must be callled with @q->queue_lock held and,
746 * Returns %NULL on failure, with @q->queue_lock held.
747 * Returns !%NULL on success, with @q->queue_lock *not held*.
749 static struct request
*get_request(struct request_queue
*q
, int rw_flags
,
750 struct bio
*bio
, gfp_t gfp_mask
)
752 struct request
*rq
= NULL
;
753 struct request_list
*rl
= &q
->rq
;
754 struct io_context
*ioc
= NULL
;
755 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
758 if (unlikely(test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)))
761 may_queue
= elv_may_queue(q
, rw_flags
);
762 if (may_queue
== ELV_MQUEUE_NO
)
765 if (rl
->count
[is_sync
]+1 >= queue_congestion_on_threshold(q
)) {
766 if (rl
->count
[is_sync
]+1 >= q
->nr_requests
) {
767 ioc
= current_io_context(GFP_ATOMIC
, q
->node
);
769 * The queue will fill after this allocation, so set
770 * it as full, and mark this process as "batching".
771 * This process will be allowed to complete a batch of
772 * requests, others will be blocked.
774 if (!blk_queue_full(q
, is_sync
)) {
775 ioc_set_batching(q
, ioc
);
776 blk_set_queue_full(q
, is_sync
);
778 if (may_queue
!= ELV_MQUEUE_MUST
779 && !ioc_batching(q
, ioc
)) {
781 * The queue is full and the allocating
782 * process is not a "batcher", and not
783 * exempted by the IO scheduler
789 blk_set_queue_congested(q
, is_sync
);
793 * Only allow batching queuers to allocate up to 50% over the defined
794 * limit of requests, otherwise we could have thousands of requests
795 * allocated with any setting of ->nr_requests
797 if (rl
->count
[is_sync
] >= (3 * q
->nr_requests
/ 2))
800 rl
->count
[is_sync
]++;
801 rl
->starved
[is_sync
] = 0;
803 if (blk_rq_should_init_elevator(bio
) &&
804 !test_bit(QUEUE_FLAG_ELVSWITCH
, &q
->queue_flags
)) {
805 rw_flags
|= REQ_ELVPRIV
;
809 if (blk_queue_io_stat(q
))
810 rw_flags
|= REQ_IO_STAT
;
811 spin_unlock_irq(q
->queue_lock
);
813 rq
= blk_alloc_request(q
, rw_flags
, gfp_mask
);
816 * Allocation failed presumably due to memory. Undo anything
817 * we might have messed up.
819 * Allocating task should really be put onto the front of the
820 * wait queue, but this is pretty rare.
822 spin_lock_irq(q
->queue_lock
);
823 freed_request(q
, rw_flags
);
826 * in the very unlikely event that allocation failed and no
827 * requests for this direction was pending, mark us starved
828 * so that freeing of a request in the other direction will
829 * notice us. another possible fix would be to split the
830 * rq mempool into READ and WRITE
833 if (unlikely(rl
->count
[is_sync
] == 0))
834 rl
->starved
[is_sync
] = 1;
840 * ioc may be NULL here, and ioc_batching will be false. That's
841 * OK, if the queue is under the request limit then requests need
842 * not count toward the nr_batch_requests limit. There will always
843 * be some limit enforced by BLK_BATCH_TIME.
845 if (ioc_batching(q
, ioc
))
846 ioc
->nr_batch_requests
--;
848 trace_block_getrq(q
, bio
, rw_flags
& 1);
854 * get_request_wait - get a free request with retry
855 * @q: request_queue to allocate request from
856 * @rw_flags: RW and SYNC flags
857 * @bio: bio to allocate request for (can be %NULL)
859 * Get a free request from @q. This function keeps retrying under memory
860 * pressure and fails iff @q is dead.
862 * Must be callled with @q->queue_lock held and,
863 * Returns %NULL on failure, with @q->queue_lock held.
864 * Returns !%NULL on success, with @q->queue_lock *not held*.
866 static struct request
*get_request_wait(struct request_queue
*q
, int rw_flags
,
869 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
872 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
875 struct io_context
*ioc
;
876 struct request_list
*rl
= &q
->rq
;
878 if (unlikely(test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)))
881 prepare_to_wait_exclusive(&rl
->wait
[is_sync
], &wait
,
882 TASK_UNINTERRUPTIBLE
);
884 trace_block_sleeprq(q
, bio
, rw_flags
& 1);
886 spin_unlock_irq(q
->queue_lock
);
890 * After sleeping, we become a "batching" process and
891 * will be able to allocate at least one request, and
892 * up to a big batch of them for a small period time.
893 * See ioc_batching, ioc_set_batching
895 ioc
= current_io_context(GFP_NOIO
, q
->node
);
896 ioc_set_batching(q
, ioc
);
898 spin_lock_irq(q
->queue_lock
);
899 finish_wait(&rl
->wait
[is_sync
], &wait
);
901 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
907 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
911 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
913 spin_lock_irq(q
->queue_lock
);
914 if (gfp_mask
& __GFP_WAIT
)
915 rq
= get_request_wait(q
, rw
, NULL
);
917 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
919 spin_unlock_irq(q
->queue_lock
);
920 /* q->queue_lock is unlocked at this point */
924 EXPORT_SYMBOL(blk_get_request
);
927 * blk_make_request - given a bio, allocate a corresponding struct request.
928 * @q: target request queue
929 * @bio: The bio describing the memory mappings that will be submitted for IO.
930 * It may be a chained-bio properly constructed by block/bio layer.
931 * @gfp_mask: gfp flags to be used for memory allocation
933 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
934 * type commands. Where the struct request needs to be farther initialized by
935 * the caller. It is passed a &struct bio, which describes the memory info of
938 * The caller of blk_make_request must make sure that bi_io_vec
939 * are set to describe the memory buffers. That bio_data_dir() will return
940 * the needed direction of the request. (And all bio's in the passed bio-chain
941 * are properly set accordingly)
943 * If called under none-sleepable conditions, mapped bio buffers must not
944 * need bouncing, by calling the appropriate masked or flagged allocator,
945 * suitable for the target device. Otherwise the call to blk_queue_bounce will
948 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
949 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
950 * anything but the first bio in the chain. Otherwise you risk waiting for IO
951 * completion of a bio that hasn't been submitted yet, thus resulting in a
952 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
953 * of bio_alloc(), as that avoids the mempool deadlock.
954 * If possible a big IO should be split into smaller parts when allocation
955 * fails. Partial allocation should not be an error, or you risk a live-lock.
957 struct request
*blk_make_request(struct request_queue
*q
, struct bio
*bio
,
960 struct request
*rq
= blk_get_request(q
, bio_data_dir(bio
), gfp_mask
);
963 return ERR_PTR(-ENOMEM
);
966 struct bio
*bounce_bio
= bio
;
969 blk_queue_bounce(q
, &bounce_bio
);
970 ret
= blk_rq_append_bio(q
, rq
, bounce_bio
);
979 EXPORT_SYMBOL(blk_make_request
);
982 * blk_requeue_request - put a request back on queue
983 * @q: request queue where request should be inserted
984 * @rq: request to be inserted
987 * Drivers often keep queueing requests until the hardware cannot accept
988 * more, when that condition happens we need to put the request back
989 * on the queue. Must be called with queue lock held.
991 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
993 blk_delete_timer(rq
);
994 blk_clear_rq_complete(rq
);
995 trace_block_rq_requeue(q
, rq
);
997 if (blk_rq_tagged(rq
))
998 blk_queue_end_tag(q
, rq
);
1000 BUG_ON(blk_queued_rq(rq
));
1002 elv_requeue_request(q
, rq
);
1004 EXPORT_SYMBOL(blk_requeue_request
);
1006 static void add_acct_request(struct request_queue
*q
, struct request
*rq
,
1009 drive_stat_acct(rq
, 1);
1010 __elv_add_request(q
, rq
, where
);
1014 * blk_insert_request - insert a special request into a request queue
1015 * @q: request queue where request should be inserted
1016 * @rq: request to be inserted
1017 * @at_head: insert request at head or tail of queue
1018 * @data: private data
1021 * Many block devices need to execute commands asynchronously, so they don't
1022 * block the whole kernel from preemption during request execution. This is
1023 * accomplished normally by inserting aritficial requests tagged as
1024 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
1025 * be scheduled for actual execution by the request queue.
1027 * We have the option of inserting the head or the tail of the queue.
1028 * Typically we use the tail for new ioctls and so forth. We use the head
1029 * of the queue for things like a QUEUE_FULL message from a device, or a
1030 * host that is unable to accept a particular command.
1032 void blk_insert_request(struct request_queue
*q
, struct request
*rq
,
1033 int at_head
, void *data
)
1035 int where
= at_head
? ELEVATOR_INSERT_FRONT
: ELEVATOR_INSERT_BACK
;
1036 unsigned long flags
;
1039 * tell I/O scheduler that this isn't a regular read/write (ie it
1040 * must not attempt merges on this) and that it acts as a soft
1043 rq
->cmd_type
= REQ_TYPE_SPECIAL
;
1047 spin_lock_irqsave(q
->queue_lock
, flags
);
1050 * If command is tagged, release the tag
1052 if (blk_rq_tagged(rq
))
1053 blk_queue_end_tag(q
, rq
);
1055 add_acct_request(q
, rq
, where
);
1057 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1059 EXPORT_SYMBOL(blk_insert_request
);
1061 static void part_round_stats_single(int cpu
, struct hd_struct
*part
,
1064 if (now
== part
->stamp
)
1067 if (part_in_flight(part
)) {
1068 __part_stat_add(cpu
, part
, time_in_queue
,
1069 part_in_flight(part
) * (now
- part
->stamp
));
1070 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
1076 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1077 * @cpu: cpu number for stats access
1078 * @part: target partition
1080 * The average IO queue length and utilisation statistics are maintained
1081 * by observing the current state of the queue length and the amount of
1082 * time it has been in this state for.
1084 * Normally, that accounting is done on IO completion, but that can result
1085 * in more than a second's worth of IO being accounted for within any one
1086 * second, leading to >100% utilisation. To deal with that, we call this
1087 * function to do a round-off before returning the results when reading
1088 * /proc/diskstats. This accounts immediately for all queue usage up to
1089 * the current jiffies and restarts the counters again.
1091 void part_round_stats(int cpu
, struct hd_struct
*part
)
1093 unsigned long now
= jiffies
;
1096 part_round_stats_single(cpu
, &part_to_disk(part
)->part0
, now
);
1097 part_round_stats_single(cpu
, part
, now
);
1099 EXPORT_SYMBOL_GPL(part_round_stats
);
1102 * queue lock must be held
1104 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1108 if (unlikely(--req
->ref_count
))
1111 elv_completed_request(q
, req
);
1113 /* this is a bio leak */
1114 WARN_ON(req
->bio
!= NULL
);
1117 * Request may not have originated from ll_rw_blk. if not,
1118 * it didn't come out of our reserved rq pools
1120 if (req
->cmd_flags
& REQ_ALLOCED
) {
1121 unsigned int flags
= req
->cmd_flags
;
1123 BUG_ON(!list_empty(&req
->queuelist
));
1124 BUG_ON(!hlist_unhashed(&req
->hash
));
1126 blk_free_request(q
, req
);
1127 freed_request(q
, flags
);
1130 EXPORT_SYMBOL_GPL(__blk_put_request
);
1132 void blk_put_request(struct request
*req
)
1134 unsigned long flags
;
1135 struct request_queue
*q
= req
->q
;
1137 spin_lock_irqsave(q
->queue_lock
, flags
);
1138 __blk_put_request(q
, req
);
1139 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1141 EXPORT_SYMBOL(blk_put_request
);
1144 * blk_add_request_payload - add a payload to a request
1145 * @rq: request to update
1146 * @page: page backing the payload
1147 * @len: length of the payload.
1149 * This allows to later add a payload to an already submitted request by
1150 * a block driver. The driver needs to take care of freeing the payload
1153 * Note that this is a quite horrible hack and nothing but handling of
1154 * discard requests should ever use it.
1156 void blk_add_request_payload(struct request
*rq
, struct page
*page
,
1159 struct bio
*bio
= rq
->bio
;
1161 bio
->bi_io_vec
->bv_page
= page
;
1162 bio
->bi_io_vec
->bv_offset
= 0;
1163 bio
->bi_io_vec
->bv_len
= len
;
1167 bio
->bi_phys_segments
= 1;
1169 rq
->__data_len
= rq
->resid_len
= len
;
1170 rq
->nr_phys_segments
= 1;
1171 rq
->buffer
= bio_data(bio
);
1173 EXPORT_SYMBOL_GPL(blk_add_request_payload
);
1175 static bool bio_attempt_back_merge(struct request_queue
*q
, struct request
*req
,
1178 const int ff
= bio
->bi_rw
& REQ_FAILFAST_MASK
;
1180 if (!ll_back_merge_fn(q
, req
, bio
))
1183 trace_block_bio_backmerge(q
, bio
);
1185 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1186 blk_rq_set_mixed_merge(req
);
1188 req
->biotail
->bi_next
= bio
;
1190 req
->__data_len
+= bio
->bi_size
;
1191 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1193 drive_stat_acct(req
, 0);
1194 elv_bio_merged(q
, req
, bio
);
1198 static bool bio_attempt_front_merge(struct request_queue
*q
,
1199 struct request
*req
, struct bio
*bio
)
1201 const int ff
= bio
->bi_rw
& REQ_FAILFAST_MASK
;
1203 if (!ll_front_merge_fn(q
, req
, bio
))
1206 trace_block_bio_frontmerge(q
, bio
);
1208 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1209 blk_rq_set_mixed_merge(req
);
1211 bio
->bi_next
= req
->bio
;
1215 * may not be valid. if the low level driver said
1216 * it didn't need a bounce buffer then it better
1217 * not touch req->buffer either...
1219 req
->buffer
= bio_data(bio
);
1220 req
->__sector
= bio
->bi_sector
;
1221 req
->__data_len
+= bio
->bi_size
;
1222 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1224 drive_stat_acct(req
, 0);
1225 elv_bio_merged(q
, req
, bio
);
1230 * attempt_plug_merge - try to merge with %current's plugged list
1231 * @q: request_queue new bio is being queued at
1232 * @bio: new bio being queued
1233 * @request_count: out parameter for number of traversed plugged requests
1235 * Determine whether @bio being queued on @q can be merged with a request
1236 * on %current's plugged list. Returns %true if merge was successful,
1239 * This function is called without @q->queue_lock; however, elevator is
1240 * accessed iff there already are requests on the plugged list which in
1241 * turn guarantees validity of the elevator.
1243 * Note that, on successful merge, elevator operation
1244 * elevator_bio_merged_fn() will be called without queue lock. Elevator
1245 * must be ready for this.
1247 static bool attempt_plug_merge(struct request_queue
*q
, struct bio
*bio
,
1248 unsigned int *request_count
)
1250 struct blk_plug
*plug
;
1254 plug
= current
->plug
;
1259 list_for_each_entry_reverse(rq
, &plug
->list
, queuelist
) {
1267 el_ret
= elv_try_merge(rq
, bio
);
1268 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1269 ret
= bio_attempt_back_merge(q
, rq
, bio
);
1272 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1273 ret
= bio_attempt_front_merge(q
, rq
, bio
);
1282 void init_request_from_bio(struct request
*req
, struct bio
*bio
)
1284 req
->cmd_type
= REQ_TYPE_FS
;
1286 req
->cmd_flags
|= bio
->bi_rw
& REQ_COMMON_MASK
;
1287 if (bio
->bi_rw
& REQ_RAHEAD
)
1288 req
->cmd_flags
|= REQ_FAILFAST_MASK
;
1291 req
->__sector
= bio
->bi_sector
;
1292 req
->ioprio
= bio_prio(bio
);
1293 blk_rq_bio_prep(req
->q
, req
, bio
);
1296 void blk_queue_bio(struct request_queue
*q
, struct bio
*bio
)
1298 const bool sync
= !!(bio
->bi_rw
& REQ_SYNC
);
1299 struct blk_plug
*plug
;
1300 int el_ret
, rw_flags
, where
= ELEVATOR_INSERT_SORT
;
1301 struct request
*req
;
1302 unsigned int request_count
= 0;
1305 * low level driver can indicate that it wants pages above a
1306 * certain limit bounced to low memory (ie for highmem, or even
1307 * ISA dma in theory)
1309 blk_queue_bounce(q
, &bio
);
1311 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) {
1312 spin_lock_irq(q
->queue_lock
);
1313 where
= ELEVATOR_INSERT_FLUSH
;
1318 * Check if we can merge with the plugged list before grabbing
1321 if (attempt_plug_merge(q
, bio
, &request_count
))
1324 spin_lock_irq(q
->queue_lock
);
1326 el_ret
= elv_merge(q
, &req
, bio
);
1327 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1328 if (bio_attempt_back_merge(q
, req
, bio
)) {
1329 if (!attempt_back_merge(q
, req
))
1330 elv_merged_request(q
, req
, el_ret
);
1333 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1334 if (bio_attempt_front_merge(q
, req
, bio
)) {
1335 if (!attempt_front_merge(q
, req
))
1336 elv_merged_request(q
, req
, el_ret
);
1343 * This sync check and mask will be re-done in init_request_from_bio(),
1344 * but we need to set it earlier to expose the sync flag to the
1345 * rq allocator and io schedulers.
1347 rw_flags
= bio_data_dir(bio
);
1349 rw_flags
|= REQ_SYNC
;
1352 * Grab a free request. This is might sleep but can not fail.
1353 * Returns with the queue unlocked.
1355 req
= get_request_wait(q
, rw_flags
, bio
);
1356 if (unlikely(!req
)) {
1357 bio_endio(bio
, -ENODEV
); /* @q is dead */
1362 * After dropping the lock and possibly sleeping here, our request
1363 * may now be mergeable after it had proven unmergeable (above).
1364 * We don't worry about that case for efficiency. It won't happen
1365 * often, and the elevators are able to handle it.
1367 init_request_from_bio(req
, bio
);
1369 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
))
1370 req
->cpu
= raw_smp_processor_id();
1372 plug
= current
->plug
;
1375 * If this is the first request added after a plug, fire
1376 * of a plug trace. If others have been added before, check
1377 * if we have multiple devices in this plug. If so, make a
1378 * note to sort the list before dispatch.
1380 if (list_empty(&plug
->list
))
1381 trace_block_plug(q
);
1383 if (!plug
->should_sort
) {
1384 struct request
*__rq
;
1386 __rq
= list_entry_rq(plug
->list
.prev
);
1388 plug
->should_sort
= 1;
1390 if (request_count
>= BLK_MAX_REQUEST_COUNT
) {
1391 blk_flush_plug_list(plug
, false);
1392 trace_block_plug(q
);
1395 list_add_tail(&req
->queuelist
, &plug
->list
);
1396 drive_stat_acct(req
, 1);
1398 spin_lock_irq(q
->queue_lock
);
1399 add_acct_request(q
, req
, where
);
1402 spin_unlock_irq(q
->queue_lock
);
1405 EXPORT_SYMBOL_GPL(blk_queue_bio
); /* for device mapper only */
1408 * If bio->bi_dev is a partition, remap the location
1410 static inline void blk_partition_remap(struct bio
*bio
)
1412 struct block_device
*bdev
= bio
->bi_bdev
;
1414 if (bio_sectors(bio
) && bdev
!= bdev
->bd_contains
) {
1415 struct hd_struct
*p
= bdev
->bd_part
;
1417 bio
->bi_sector
+= p
->start_sect
;
1418 bio
->bi_bdev
= bdev
->bd_contains
;
1420 trace_block_bio_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
1422 bio
->bi_sector
- p
->start_sect
);
1426 static void handle_bad_sector(struct bio
*bio
)
1428 char b
[BDEVNAME_SIZE
];
1430 printk(KERN_INFO
"attempt to access beyond end of device\n");
1431 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
1432 bdevname(bio
->bi_bdev
, b
),
1434 (unsigned long long)bio
->bi_sector
+ bio_sectors(bio
),
1435 (long long)(i_size_read(bio
->bi_bdev
->bd_inode
) >> 9));
1437 set_bit(BIO_EOF
, &bio
->bi_flags
);
1440 #ifdef CONFIG_FAIL_MAKE_REQUEST
1442 static DECLARE_FAULT_ATTR(fail_make_request
);
1444 static int __init
setup_fail_make_request(char *str
)
1446 return setup_fault_attr(&fail_make_request
, str
);
1448 __setup("fail_make_request=", setup_fail_make_request
);
1450 static bool should_fail_request(struct hd_struct
*part
, unsigned int bytes
)
1452 return part
->make_it_fail
&& should_fail(&fail_make_request
, bytes
);
1455 static int __init
fail_make_request_debugfs(void)
1457 struct dentry
*dir
= fault_create_debugfs_attr("fail_make_request",
1458 NULL
, &fail_make_request
);
1460 return IS_ERR(dir
) ? PTR_ERR(dir
) : 0;
1463 late_initcall(fail_make_request_debugfs
);
1465 #else /* CONFIG_FAIL_MAKE_REQUEST */
1467 static inline bool should_fail_request(struct hd_struct
*part
,
1473 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1476 * Check whether this bio extends beyond the end of the device.
1478 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
1485 /* Test device or partition size, when known. */
1486 maxsector
= i_size_read(bio
->bi_bdev
->bd_inode
) >> 9;
1488 sector_t sector
= bio
->bi_sector
;
1490 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
1492 * This may well happen - the kernel calls bread()
1493 * without checking the size of the device, e.g., when
1494 * mounting a device.
1496 handle_bad_sector(bio
);
1504 static noinline_for_stack
bool
1505 generic_make_request_checks(struct bio
*bio
)
1507 struct request_queue
*q
;
1508 int nr_sectors
= bio_sectors(bio
);
1510 char b
[BDEVNAME_SIZE
];
1511 struct hd_struct
*part
;
1515 if (bio_check_eod(bio
, nr_sectors
))
1518 q
= bdev_get_queue(bio
->bi_bdev
);
1521 "generic_make_request: Trying to access "
1522 "nonexistent block-device %s (%Lu)\n",
1523 bdevname(bio
->bi_bdev
, b
),
1524 (long long) bio
->bi_sector
);
1528 if (unlikely(!(bio
->bi_rw
& REQ_DISCARD
) &&
1529 nr_sectors
> queue_max_hw_sectors(q
))) {
1530 printk(KERN_ERR
"bio too big device %s (%u > %u)\n",
1531 bdevname(bio
->bi_bdev
, b
),
1533 queue_max_hw_sectors(q
));
1537 part
= bio
->bi_bdev
->bd_part
;
1538 if (should_fail_request(part
, bio
->bi_size
) ||
1539 should_fail_request(&part_to_disk(part
)->part0
,
1544 * If this device has partitions, remap block n
1545 * of partition p to block n+start(p) of the disk.
1547 blk_partition_remap(bio
);
1549 if (bio_integrity_enabled(bio
) && bio_integrity_prep(bio
))
1552 if (bio_check_eod(bio
, nr_sectors
))
1556 * Filter flush bio's early so that make_request based
1557 * drivers without flush support don't have to worry
1560 if ((bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) && !q
->flush_flags
) {
1561 bio
->bi_rw
&= ~(REQ_FLUSH
| REQ_FUA
);
1568 if ((bio
->bi_rw
& REQ_DISCARD
) &&
1569 (!blk_queue_discard(q
) ||
1570 ((bio
->bi_rw
& REQ_SECURE
) &&
1571 !blk_queue_secdiscard(q
)))) {
1576 if (blk_throtl_bio(q
, bio
))
1577 return false; /* throttled, will be resubmitted later */
1579 trace_block_bio_queue(q
, bio
);
1583 bio_endio(bio
, err
);
1588 * generic_make_request - hand a buffer to its device driver for I/O
1589 * @bio: The bio describing the location in memory and on the device.
1591 * generic_make_request() is used to make I/O requests of block
1592 * devices. It is passed a &struct bio, which describes the I/O that needs
1595 * generic_make_request() does not return any status. The
1596 * success/failure status of the request, along with notification of
1597 * completion, is delivered asynchronously through the bio->bi_end_io
1598 * function described (one day) else where.
1600 * The caller of generic_make_request must make sure that bi_io_vec
1601 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1602 * set to describe the device address, and the
1603 * bi_end_io and optionally bi_private are set to describe how
1604 * completion notification should be signaled.
1606 * generic_make_request and the drivers it calls may use bi_next if this
1607 * bio happens to be merged with someone else, and may resubmit the bio to
1608 * a lower device by calling into generic_make_request recursively, which
1609 * means the bio should NOT be touched after the call to ->make_request_fn.
1611 void generic_make_request(struct bio
*bio
)
1613 struct bio_list bio_list_on_stack
;
1615 if (!generic_make_request_checks(bio
))
1619 * We only want one ->make_request_fn to be active at a time, else
1620 * stack usage with stacked devices could be a problem. So use
1621 * current->bio_list to keep a list of requests submited by a
1622 * make_request_fn function. current->bio_list is also used as a
1623 * flag to say if generic_make_request is currently active in this
1624 * task or not. If it is NULL, then no make_request is active. If
1625 * it is non-NULL, then a make_request is active, and new requests
1626 * should be added at the tail
1628 if (current
->bio_list
) {
1629 bio_list_add(current
->bio_list
, bio
);
1633 /* following loop may be a bit non-obvious, and so deserves some
1635 * Before entering the loop, bio->bi_next is NULL (as all callers
1636 * ensure that) so we have a list with a single bio.
1637 * We pretend that we have just taken it off a longer list, so
1638 * we assign bio_list to a pointer to the bio_list_on_stack,
1639 * thus initialising the bio_list of new bios to be
1640 * added. ->make_request() may indeed add some more bios
1641 * through a recursive call to generic_make_request. If it
1642 * did, we find a non-NULL value in bio_list and re-enter the loop
1643 * from the top. In this case we really did just take the bio
1644 * of the top of the list (no pretending) and so remove it from
1645 * bio_list, and call into ->make_request() again.
1647 BUG_ON(bio
->bi_next
);
1648 bio_list_init(&bio_list_on_stack
);
1649 current
->bio_list
= &bio_list_on_stack
;
1651 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
1653 q
->make_request_fn(q
, bio
);
1655 bio
= bio_list_pop(current
->bio_list
);
1657 current
->bio_list
= NULL
; /* deactivate */
1659 EXPORT_SYMBOL(generic_make_request
);
1662 * submit_bio - submit a bio to the block device layer for I/O
1663 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1664 * @bio: The &struct bio which describes the I/O
1666 * submit_bio() is very similar in purpose to generic_make_request(), and
1667 * uses that function to do most of the work. Both are fairly rough
1668 * interfaces; @bio must be presetup and ready for I/O.
1671 void submit_bio(int rw
, struct bio
*bio
)
1673 int count
= bio_sectors(bio
);
1678 * If it's a regular read/write or a barrier with data attached,
1679 * go through the normal accounting stuff before submission.
1681 if (bio_has_data(bio
) && !(rw
& REQ_DISCARD
)) {
1683 count_vm_events(PGPGOUT
, count
);
1685 task_io_account_read(bio
->bi_size
);
1686 count_vm_events(PGPGIN
, count
);
1689 if (unlikely(block_dump
)) {
1690 char b
[BDEVNAME_SIZE
];
1691 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s (%u sectors)\n",
1692 current
->comm
, task_pid_nr(current
),
1693 (rw
& WRITE
) ? "WRITE" : "READ",
1694 (unsigned long long)bio
->bi_sector
,
1695 bdevname(bio
->bi_bdev
, b
),
1700 generic_make_request(bio
);
1702 EXPORT_SYMBOL(submit_bio
);
1705 * blk_rq_check_limits - Helper function to check a request for the queue limit
1707 * @rq: the request being checked
1710 * @rq may have been made based on weaker limitations of upper-level queues
1711 * in request stacking drivers, and it may violate the limitation of @q.
1712 * Since the block layer and the underlying device driver trust @rq
1713 * after it is inserted to @q, it should be checked against @q before
1714 * the insertion using this generic function.
1716 * This function should also be useful for request stacking drivers
1717 * in some cases below, so export this function.
1718 * Request stacking drivers like request-based dm may change the queue
1719 * limits while requests are in the queue (e.g. dm's table swapping).
1720 * Such request stacking drivers should check those requests agaist
1721 * the new queue limits again when they dispatch those requests,
1722 * although such checkings are also done against the old queue limits
1723 * when submitting requests.
1725 int blk_rq_check_limits(struct request_queue
*q
, struct request
*rq
)
1727 if (rq
->cmd_flags
& REQ_DISCARD
)
1730 if (blk_rq_sectors(rq
) > queue_max_sectors(q
) ||
1731 blk_rq_bytes(rq
) > queue_max_hw_sectors(q
) << 9) {
1732 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
1737 * queue's settings related to segment counting like q->bounce_pfn
1738 * may differ from that of other stacking queues.
1739 * Recalculate it to check the request correctly on this queue's
1742 blk_recalc_rq_segments(rq
);
1743 if (rq
->nr_phys_segments
> queue_max_segments(q
)) {
1744 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
1750 EXPORT_SYMBOL_GPL(blk_rq_check_limits
);
1753 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1754 * @q: the queue to submit the request
1755 * @rq: the request being queued
1757 int blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
1759 unsigned long flags
;
1760 int where
= ELEVATOR_INSERT_BACK
;
1762 if (blk_rq_check_limits(q
, rq
))
1766 should_fail_request(&rq
->rq_disk
->part0
, blk_rq_bytes(rq
)))
1769 spin_lock_irqsave(q
->queue_lock
, flags
);
1772 * Submitting request must be dequeued before calling this function
1773 * because it will be linked to another request_queue
1775 BUG_ON(blk_queued_rq(rq
));
1777 if (rq
->cmd_flags
& (REQ_FLUSH
|REQ_FUA
))
1778 where
= ELEVATOR_INSERT_FLUSH
;
1780 add_acct_request(q
, rq
, where
);
1781 if (where
== ELEVATOR_INSERT_FLUSH
)
1783 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1787 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
1790 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1791 * @rq: request to examine
1794 * A request could be merge of IOs which require different failure
1795 * handling. This function determines the number of bytes which
1796 * can be failed from the beginning of the request without
1797 * crossing into area which need to be retried further.
1800 * The number of bytes to fail.
1803 * queue_lock must be held.
1805 unsigned int blk_rq_err_bytes(const struct request
*rq
)
1807 unsigned int ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
1808 unsigned int bytes
= 0;
1811 if (!(rq
->cmd_flags
& REQ_MIXED_MERGE
))
1812 return blk_rq_bytes(rq
);
1815 * Currently the only 'mixing' which can happen is between
1816 * different fastfail types. We can safely fail portions
1817 * which have all the failfast bits that the first one has -
1818 * the ones which are at least as eager to fail as the first
1821 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
1822 if ((bio
->bi_rw
& ff
) != ff
)
1824 bytes
+= bio
->bi_size
;
1827 /* this could lead to infinite loop */
1828 BUG_ON(blk_rq_bytes(rq
) && !bytes
);
1831 EXPORT_SYMBOL_GPL(blk_rq_err_bytes
);
1833 static void blk_account_io_completion(struct request
*req
, unsigned int bytes
)
1835 if (blk_do_io_stat(req
)) {
1836 const int rw
= rq_data_dir(req
);
1837 struct hd_struct
*part
;
1840 cpu
= part_stat_lock();
1842 part_stat_add(cpu
, part
, sectors
[rw
], bytes
>> 9);
1847 static void blk_account_io_done(struct request
*req
)
1850 * Account IO completion. flush_rq isn't accounted as a
1851 * normal IO on queueing nor completion. Accounting the
1852 * containing request is enough.
1854 if (blk_do_io_stat(req
) && !(req
->cmd_flags
& REQ_FLUSH_SEQ
)) {
1855 unsigned long duration
= jiffies
- req
->start_time
;
1856 const int rw
= rq_data_dir(req
);
1857 struct hd_struct
*part
;
1860 cpu
= part_stat_lock();
1863 part_stat_inc(cpu
, part
, ios
[rw
]);
1864 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
1865 part_round_stats(cpu
, part
);
1866 part_dec_in_flight(part
, rw
);
1868 hd_struct_put(part
);
1874 * blk_peek_request - peek at the top of a request queue
1875 * @q: request queue to peek at
1878 * Return the request at the top of @q. The returned request
1879 * should be started using blk_start_request() before LLD starts
1883 * Pointer to the request at the top of @q if available. Null
1887 * queue_lock must be held.
1889 struct request
*blk_peek_request(struct request_queue
*q
)
1894 while ((rq
= __elv_next_request(q
)) != NULL
) {
1895 if (!(rq
->cmd_flags
& REQ_STARTED
)) {
1897 * This is the first time the device driver
1898 * sees this request (possibly after
1899 * requeueing). Notify IO scheduler.
1901 if (rq
->cmd_flags
& REQ_SORTED
)
1902 elv_activate_rq(q
, rq
);
1905 * just mark as started even if we don't start
1906 * it, a request that has been delayed should
1907 * not be passed by new incoming requests
1909 rq
->cmd_flags
|= REQ_STARTED
;
1910 trace_block_rq_issue(q
, rq
);
1913 if (!q
->boundary_rq
|| q
->boundary_rq
== rq
) {
1914 q
->end_sector
= rq_end_sector(rq
);
1915 q
->boundary_rq
= NULL
;
1918 if (rq
->cmd_flags
& REQ_DONTPREP
)
1921 if (q
->dma_drain_size
&& blk_rq_bytes(rq
)) {
1923 * make sure space for the drain appears we
1924 * know we can do this because max_hw_segments
1925 * has been adjusted to be one fewer than the
1928 rq
->nr_phys_segments
++;
1934 ret
= q
->prep_rq_fn(q
, rq
);
1935 if (ret
== BLKPREP_OK
) {
1937 } else if (ret
== BLKPREP_DEFER
) {
1939 * the request may have been (partially) prepped.
1940 * we need to keep this request in the front to
1941 * avoid resource deadlock. REQ_STARTED will
1942 * prevent other fs requests from passing this one.
1944 if (q
->dma_drain_size
&& blk_rq_bytes(rq
) &&
1945 !(rq
->cmd_flags
& REQ_DONTPREP
)) {
1947 * remove the space for the drain we added
1948 * so that we don't add it again
1950 --rq
->nr_phys_segments
;
1955 } else if (ret
== BLKPREP_KILL
) {
1956 rq
->cmd_flags
|= REQ_QUIET
;
1958 * Mark this request as started so we don't trigger
1959 * any debug logic in the end I/O path.
1961 blk_start_request(rq
);
1962 __blk_end_request_all(rq
, -EIO
);
1964 printk(KERN_ERR
"%s: bad return=%d\n", __func__
, ret
);
1971 EXPORT_SYMBOL(blk_peek_request
);
1973 void blk_dequeue_request(struct request
*rq
)
1975 struct request_queue
*q
= rq
->q
;
1977 BUG_ON(list_empty(&rq
->queuelist
));
1978 BUG_ON(ELV_ON_HASH(rq
));
1980 list_del_init(&rq
->queuelist
);
1983 * the time frame between a request being removed from the lists
1984 * and to it is freed is accounted as io that is in progress at
1987 if (blk_account_rq(rq
)) {
1988 q
->in_flight
[rq_is_sync(rq
)]++;
1989 set_io_start_time_ns(rq
);
1994 * blk_start_request - start request processing on the driver
1995 * @req: request to dequeue
1998 * Dequeue @req and start timeout timer on it. This hands off the
1999 * request to the driver.
2001 * Block internal functions which don't want to start timer should
2002 * call blk_dequeue_request().
2005 * queue_lock must be held.
2007 void blk_start_request(struct request
*req
)
2009 blk_dequeue_request(req
);
2012 * We are now handing the request to the hardware, initialize
2013 * resid_len to full count and add the timeout handler.
2015 req
->resid_len
= blk_rq_bytes(req
);
2016 if (unlikely(blk_bidi_rq(req
)))
2017 req
->next_rq
->resid_len
= blk_rq_bytes(req
->next_rq
);
2021 EXPORT_SYMBOL(blk_start_request
);
2024 * blk_fetch_request - fetch a request from a request queue
2025 * @q: request queue to fetch a request from
2028 * Return the request at the top of @q. The request is started on
2029 * return and LLD can start processing it immediately.
2032 * Pointer to the request at the top of @q if available. Null
2036 * queue_lock must be held.
2038 struct request
*blk_fetch_request(struct request_queue
*q
)
2042 rq
= blk_peek_request(q
);
2044 blk_start_request(rq
);
2047 EXPORT_SYMBOL(blk_fetch_request
);
2050 * blk_update_request - Special helper function for request stacking drivers
2051 * @req: the request being processed
2052 * @error: %0 for success, < %0 for error
2053 * @nr_bytes: number of bytes to complete @req
2056 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2057 * the request structure even if @req doesn't have leftover.
2058 * If @req has leftover, sets it up for the next range of segments.
2060 * This special helper function is only for request stacking drivers
2061 * (e.g. request-based dm) so that they can handle partial completion.
2062 * Actual device drivers should use blk_end_request instead.
2064 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2065 * %false return from this function.
2068 * %false - this request doesn't have any more data
2069 * %true - this request has more data
2071 bool blk_update_request(struct request
*req
, int error
, unsigned int nr_bytes
)
2073 int total_bytes
, bio_nbytes
, next_idx
= 0;
2079 trace_block_rq_complete(req
->q
, req
);
2082 * For fs requests, rq is just carrier of independent bio's
2083 * and each partial completion should be handled separately.
2084 * Reset per-request error on each partial completion.
2086 * TODO: tj: This is too subtle. It would be better to let
2087 * low level drivers do what they see fit.
2089 if (req
->cmd_type
== REQ_TYPE_FS
)
2092 if (error
&& req
->cmd_type
== REQ_TYPE_FS
&&
2093 !(req
->cmd_flags
& REQ_QUIET
)) {
2098 error_type
= "recoverable transport";
2101 error_type
= "critical target";
2104 error_type
= "critical nexus";
2111 printk(KERN_ERR
"end_request: %s error, dev %s, sector %llu\n",
2112 error_type
, req
->rq_disk
? req
->rq_disk
->disk_name
: "?",
2113 (unsigned long long)blk_rq_pos(req
));
2116 blk_account_io_completion(req
, nr_bytes
);
2118 total_bytes
= bio_nbytes
= 0;
2119 while ((bio
= req
->bio
) != NULL
) {
2122 if (nr_bytes
>= bio
->bi_size
) {
2123 req
->bio
= bio
->bi_next
;
2124 nbytes
= bio
->bi_size
;
2125 req_bio_endio(req
, bio
, nbytes
, error
);
2129 int idx
= bio
->bi_idx
+ next_idx
;
2131 if (unlikely(idx
>= bio
->bi_vcnt
)) {
2132 blk_dump_rq_flags(req
, "__end_that");
2133 printk(KERN_ERR
"%s: bio idx %d >= vcnt %d\n",
2134 __func__
, idx
, bio
->bi_vcnt
);
2138 nbytes
= bio_iovec_idx(bio
, idx
)->bv_len
;
2139 BIO_BUG_ON(nbytes
> bio
->bi_size
);
2142 * not a complete bvec done
2144 if (unlikely(nbytes
> nr_bytes
)) {
2145 bio_nbytes
+= nr_bytes
;
2146 total_bytes
+= nr_bytes
;
2151 * advance to the next vector
2154 bio_nbytes
+= nbytes
;
2157 total_bytes
+= nbytes
;
2163 * end more in this run, or just return 'not-done'
2165 if (unlikely(nr_bytes
<= 0))
2175 * Reset counters so that the request stacking driver
2176 * can find how many bytes remain in the request
2179 req
->__data_len
= 0;
2184 * if the request wasn't completed, update state
2187 req_bio_endio(req
, bio
, bio_nbytes
, error
);
2188 bio
->bi_idx
+= next_idx
;
2189 bio_iovec(bio
)->bv_offset
+= nr_bytes
;
2190 bio_iovec(bio
)->bv_len
-= nr_bytes
;
2193 req
->__data_len
-= total_bytes
;
2194 req
->buffer
= bio_data(req
->bio
);
2196 /* update sector only for requests with clear definition of sector */
2197 if (req
->cmd_type
== REQ_TYPE_FS
|| (req
->cmd_flags
& REQ_DISCARD
))
2198 req
->__sector
+= total_bytes
>> 9;
2200 /* mixed attributes always follow the first bio */
2201 if (req
->cmd_flags
& REQ_MIXED_MERGE
) {
2202 req
->cmd_flags
&= ~REQ_FAILFAST_MASK
;
2203 req
->cmd_flags
|= req
->bio
->bi_rw
& REQ_FAILFAST_MASK
;
2207 * If total number of sectors is less than the first segment
2208 * size, something has gone terribly wrong.
2210 if (blk_rq_bytes(req
) < blk_rq_cur_bytes(req
)) {
2211 blk_dump_rq_flags(req
, "request botched");
2212 req
->__data_len
= blk_rq_cur_bytes(req
);
2215 /* recalculate the number of segments */
2216 blk_recalc_rq_segments(req
);
2220 EXPORT_SYMBOL_GPL(blk_update_request
);
2222 static bool blk_update_bidi_request(struct request
*rq
, int error
,
2223 unsigned int nr_bytes
,
2224 unsigned int bidi_bytes
)
2226 if (blk_update_request(rq
, error
, nr_bytes
))
2229 /* Bidi request must be completed as a whole */
2230 if (unlikely(blk_bidi_rq(rq
)) &&
2231 blk_update_request(rq
->next_rq
, error
, bidi_bytes
))
2234 if (blk_queue_add_random(rq
->q
))
2235 add_disk_randomness(rq
->rq_disk
);
2241 * blk_unprep_request - unprepare a request
2244 * This function makes a request ready for complete resubmission (or
2245 * completion). It happens only after all error handling is complete,
2246 * so represents the appropriate moment to deallocate any resources
2247 * that were allocated to the request in the prep_rq_fn. The queue
2248 * lock is held when calling this.
2250 void blk_unprep_request(struct request
*req
)
2252 struct request_queue
*q
= req
->q
;
2254 req
->cmd_flags
&= ~REQ_DONTPREP
;
2255 if (q
->unprep_rq_fn
)
2256 q
->unprep_rq_fn(q
, req
);
2258 EXPORT_SYMBOL_GPL(blk_unprep_request
);
2261 * queue lock must be held
2263 static void blk_finish_request(struct request
*req
, int error
)
2265 if (blk_rq_tagged(req
))
2266 blk_queue_end_tag(req
->q
, req
);
2268 BUG_ON(blk_queued_rq(req
));
2270 if (unlikely(laptop_mode
) && req
->cmd_type
== REQ_TYPE_FS
)
2271 laptop_io_completion(&req
->q
->backing_dev_info
);
2273 blk_delete_timer(req
);
2275 if (req
->cmd_flags
& REQ_DONTPREP
)
2276 blk_unprep_request(req
);
2279 blk_account_io_done(req
);
2282 req
->end_io(req
, error
);
2284 if (blk_bidi_rq(req
))
2285 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
2287 __blk_put_request(req
->q
, req
);
2292 * blk_end_bidi_request - Complete a bidi request
2293 * @rq: the request to complete
2294 * @error: %0 for success, < %0 for error
2295 * @nr_bytes: number of bytes to complete @rq
2296 * @bidi_bytes: number of bytes to complete @rq->next_rq
2299 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2300 * Drivers that supports bidi can safely call this member for any
2301 * type of request, bidi or uni. In the later case @bidi_bytes is
2305 * %false - we are done with this request
2306 * %true - still buffers pending for this request
2308 static bool blk_end_bidi_request(struct request
*rq
, int error
,
2309 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2311 struct request_queue
*q
= rq
->q
;
2312 unsigned long flags
;
2314 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2317 spin_lock_irqsave(q
->queue_lock
, flags
);
2318 blk_finish_request(rq
, error
);
2319 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2325 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2326 * @rq: the request to complete
2327 * @error: %0 for success, < %0 for error
2328 * @nr_bytes: number of bytes to complete @rq
2329 * @bidi_bytes: number of bytes to complete @rq->next_rq
2332 * Identical to blk_end_bidi_request() except that queue lock is
2333 * assumed to be locked on entry and remains so on return.
2336 * %false - we are done with this request
2337 * %true - still buffers pending for this request
2339 bool __blk_end_bidi_request(struct request
*rq
, int error
,
2340 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2342 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2345 blk_finish_request(rq
, error
);
2351 * blk_end_request - Helper function for drivers to complete the request.
2352 * @rq: the request being processed
2353 * @error: %0 for success, < %0 for error
2354 * @nr_bytes: number of bytes to complete
2357 * Ends I/O on a number of bytes attached to @rq.
2358 * If @rq has leftover, sets it up for the next range of segments.
2361 * %false - we are done with this request
2362 * %true - still buffers pending for this request
2364 bool blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2366 return blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2368 EXPORT_SYMBOL(blk_end_request
);
2371 * blk_end_request_all - Helper function for drives to finish the request.
2372 * @rq: the request to finish
2373 * @error: %0 for success, < %0 for error
2376 * Completely finish @rq.
2378 void blk_end_request_all(struct request
*rq
, int error
)
2381 unsigned int bidi_bytes
= 0;
2383 if (unlikely(blk_bidi_rq(rq
)))
2384 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2386 pending
= blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2389 EXPORT_SYMBOL(blk_end_request_all
);
2392 * blk_end_request_cur - Helper function to finish the current request chunk.
2393 * @rq: the request to finish the current chunk for
2394 * @error: %0 for success, < %0 for error
2397 * Complete the current consecutively mapped chunk from @rq.
2400 * %false - we are done with this request
2401 * %true - still buffers pending for this request
2403 bool blk_end_request_cur(struct request
*rq
, int error
)
2405 return blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2407 EXPORT_SYMBOL(blk_end_request_cur
);
2410 * blk_end_request_err - Finish a request till the next failure boundary.
2411 * @rq: the request to finish till the next failure boundary for
2412 * @error: must be negative errno
2415 * Complete @rq till the next failure boundary.
2418 * %false - we are done with this request
2419 * %true - still buffers pending for this request
2421 bool blk_end_request_err(struct request
*rq
, int error
)
2423 WARN_ON(error
>= 0);
2424 return blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2426 EXPORT_SYMBOL_GPL(blk_end_request_err
);
2429 * __blk_end_request - Helper function for drivers to complete the request.
2430 * @rq: the request being processed
2431 * @error: %0 for success, < %0 for error
2432 * @nr_bytes: number of bytes to complete
2435 * Must be called with queue lock held unlike blk_end_request().
2438 * %false - we are done with this request
2439 * %true - still buffers pending for this request
2441 bool __blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2443 return __blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2445 EXPORT_SYMBOL(__blk_end_request
);
2448 * __blk_end_request_all - Helper function for drives to finish the request.
2449 * @rq: the request to finish
2450 * @error: %0 for success, < %0 for error
2453 * Completely finish @rq. Must be called with queue lock held.
2455 void __blk_end_request_all(struct request
*rq
, int error
)
2458 unsigned int bidi_bytes
= 0;
2460 if (unlikely(blk_bidi_rq(rq
)))
2461 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2463 pending
= __blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2466 EXPORT_SYMBOL(__blk_end_request_all
);
2469 * __blk_end_request_cur - Helper function to finish the current request chunk.
2470 * @rq: the request to finish the current chunk for
2471 * @error: %0 for success, < %0 for error
2474 * Complete the current consecutively mapped chunk from @rq. Must
2475 * be called with queue lock held.
2478 * %false - we are done with this request
2479 * %true - still buffers pending for this request
2481 bool __blk_end_request_cur(struct request
*rq
, int error
)
2483 return __blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2485 EXPORT_SYMBOL(__blk_end_request_cur
);
2488 * __blk_end_request_err - Finish a request till the next failure boundary.
2489 * @rq: the request to finish till the next failure boundary for
2490 * @error: must be negative errno
2493 * Complete @rq till the next failure boundary. Must be called
2494 * with queue lock held.
2497 * %false - we are done with this request
2498 * %true - still buffers pending for this request
2500 bool __blk_end_request_err(struct request
*rq
, int error
)
2502 WARN_ON(error
>= 0);
2503 return __blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2505 EXPORT_SYMBOL_GPL(__blk_end_request_err
);
2507 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
2510 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2511 rq
->cmd_flags
|= bio
->bi_rw
& REQ_WRITE
;
2513 if (bio_has_data(bio
)) {
2514 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2515 rq
->buffer
= bio_data(bio
);
2517 rq
->__data_len
= bio
->bi_size
;
2518 rq
->bio
= rq
->biotail
= bio
;
2521 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2524 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2526 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2527 * @rq: the request to be flushed
2530 * Flush all pages in @rq.
2532 void rq_flush_dcache_pages(struct request
*rq
)
2534 struct req_iterator iter
;
2535 struct bio_vec
*bvec
;
2537 rq_for_each_segment(bvec
, rq
, iter
)
2538 flush_dcache_page(bvec
->bv_page
);
2540 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages
);
2544 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2545 * @q : the queue of the device being checked
2548 * Check if underlying low-level drivers of a device are busy.
2549 * If the drivers want to export their busy state, they must set own
2550 * exporting function using blk_queue_lld_busy() first.
2552 * Basically, this function is used only by request stacking drivers
2553 * to stop dispatching requests to underlying devices when underlying
2554 * devices are busy. This behavior helps more I/O merging on the queue
2555 * of the request stacking driver and prevents I/O throughput regression
2556 * on burst I/O load.
2559 * 0 - Not busy (The request stacking driver should dispatch request)
2560 * 1 - Busy (The request stacking driver should stop dispatching request)
2562 int blk_lld_busy(struct request_queue
*q
)
2565 return q
->lld_busy_fn(q
);
2569 EXPORT_SYMBOL_GPL(blk_lld_busy
);
2572 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2573 * @rq: the clone request to be cleaned up
2576 * Free all bios in @rq for a cloned request.
2578 void blk_rq_unprep_clone(struct request
*rq
)
2582 while ((bio
= rq
->bio
) != NULL
) {
2583 rq
->bio
= bio
->bi_next
;
2588 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone
);
2591 * Copy attributes of the original request to the clone request.
2592 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2594 static void __blk_rq_prep_clone(struct request
*dst
, struct request
*src
)
2596 dst
->cpu
= src
->cpu
;
2597 dst
->cmd_flags
= (src
->cmd_flags
& REQ_CLONE_MASK
) | REQ_NOMERGE
;
2598 dst
->cmd_type
= src
->cmd_type
;
2599 dst
->__sector
= blk_rq_pos(src
);
2600 dst
->__data_len
= blk_rq_bytes(src
);
2601 dst
->nr_phys_segments
= src
->nr_phys_segments
;
2602 dst
->ioprio
= src
->ioprio
;
2603 dst
->extra_len
= src
->extra_len
;
2607 * blk_rq_prep_clone - Helper function to setup clone request
2608 * @rq: the request to be setup
2609 * @rq_src: original request to be cloned
2610 * @bs: bio_set that bios for clone are allocated from
2611 * @gfp_mask: memory allocation mask for bio
2612 * @bio_ctr: setup function to be called for each clone bio.
2613 * Returns %0 for success, non %0 for failure.
2614 * @data: private data to be passed to @bio_ctr
2617 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2618 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2619 * are not copied, and copying such parts is the caller's responsibility.
2620 * Also, pages which the original bios are pointing to are not copied
2621 * and the cloned bios just point same pages.
2622 * So cloned bios must be completed before original bios, which means
2623 * the caller must complete @rq before @rq_src.
2625 int blk_rq_prep_clone(struct request
*rq
, struct request
*rq_src
,
2626 struct bio_set
*bs
, gfp_t gfp_mask
,
2627 int (*bio_ctr
)(struct bio
*, struct bio
*, void *),
2630 struct bio
*bio
, *bio_src
;
2635 blk_rq_init(NULL
, rq
);
2637 __rq_for_each_bio(bio_src
, rq_src
) {
2638 bio
= bio_alloc_bioset(gfp_mask
, bio_src
->bi_max_vecs
, bs
);
2642 __bio_clone(bio
, bio_src
);
2644 if (bio_integrity(bio_src
) &&
2645 bio_integrity_clone(bio
, bio_src
, gfp_mask
, bs
))
2648 if (bio_ctr
&& bio_ctr(bio
, bio_src
, data
))
2652 rq
->biotail
->bi_next
= bio
;
2655 rq
->bio
= rq
->biotail
= bio
;
2658 __blk_rq_prep_clone(rq
, rq_src
);
2665 blk_rq_unprep_clone(rq
);
2669 EXPORT_SYMBOL_GPL(blk_rq_prep_clone
);
2671 int kblockd_schedule_work(struct request_queue
*q
, struct work_struct
*work
)
2673 return queue_work(kblockd_workqueue
, work
);
2675 EXPORT_SYMBOL(kblockd_schedule_work
);
2677 int kblockd_schedule_delayed_work(struct request_queue
*q
,
2678 struct delayed_work
*dwork
, unsigned long delay
)
2680 return queue_delayed_work(kblockd_workqueue
, dwork
, delay
);
2682 EXPORT_SYMBOL(kblockd_schedule_delayed_work
);
2684 #define PLUG_MAGIC 0x91827364
2687 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2688 * @plug: The &struct blk_plug that needs to be initialized
2691 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2692 * pending I/O should the task end up blocking between blk_start_plug() and
2693 * blk_finish_plug(). This is important from a performance perspective, but
2694 * also ensures that we don't deadlock. For instance, if the task is blocking
2695 * for a memory allocation, memory reclaim could end up wanting to free a
2696 * page belonging to that request that is currently residing in our private
2697 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2698 * this kind of deadlock.
2700 void blk_start_plug(struct blk_plug
*plug
)
2702 struct task_struct
*tsk
= current
;
2704 plug
->magic
= PLUG_MAGIC
;
2705 INIT_LIST_HEAD(&plug
->list
);
2706 INIT_LIST_HEAD(&plug
->cb_list
);
2707 plug
->should_sort
= 0;
2710 * If this is a nested plug, don't actually assign it. It will be
2711 * flushed on its own.
2715 * Store ordering should not be needed here, since a potential
2716 * preempt will imply a full memory barrier
2721 EXPORT_SYMBOL(blk_start_plug
);
2723 static int plug_rq_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
2725 struct request
*rqa
= container_of(a
, struct request
, queuelist
);
2726 struct request
*rqb
= container_of(b
, struct request
, queuelist
);
2728 return !(rqa
->q
<= rqb
->q
);
2732 * If 'from_schedule' is true, then postpone the dispatch of requests
2733 * until a safe kblockd context. We due this to avoid accidental big
2734 * additional stack usage in driver dispatch, in places where the originally
2735 * plugger did not intend it.
2737 static void queue_unplugged(struct request_queue
*q
, unsigned int depth
,
2739 __releases(q
->queue_lock
)
2741 trace_block_unplug(q
, depth
, !from_schedule
);
2744 * If we are punting this to kblockd, then we can safely drop
2745 * the queue_lock before waking kblockd (which needs to take
2748 if (from_schedule
) {
2749 spin_unlock(q
->queue_lock
);
2750 blk_run_queue_async(q
);
2753 spin_unlock(q
->queue_lock
);
2758 static void flush_plug_callbacks(struct blk_plug
*plug
)
2760 LIST_HEAD(callbacks
);
2762 if (list_empty(&plug
->cb_list
))
2765 list_splice_init(&plug
->cb_list
, &callbacks
);
2767 while (!list_empty(&callbacks
)) {
2768 struct blk_plug_cb
*cb
= list_first_entry(&callbacks
,
2771 list_del(&cb
->list
);
2776 void blk_flush_plug_list(struct blk_plug
*plug
, bool from_schedule
)
2778 struct request_queue
*q
;
2779 unsigned long flags
;
2784 BUG_ON(plug
->magic
!= PLUG_MAGIC
);
2786 flush_plug_callbacks(plug
);
2787 if (list_empty(&plug
->list
))
2790 list_splice_init(&plug
->list
, &list
);
2792 if (plug
->should_sort
) {
2793 list_sort(NULL
, &list
, plug_rq_cmp
);
2794 plug
->should_sort
= 0;
2801 * Save and disable interrupts here, to avoid doing it for every
2802 * queue lock we have to take.
2804 local_irq_save(flags
);
2805 while (!list_empty(&list
)) {
2806 rq
= list_entry_rq(list
.next
);
2807 list_del_init(&rq
->queuelist
);
2811 * This drops the queue lock
2814 queue_unplugged(q
, depth
, from_schedule
);
2817 spin_lock(q
->queue_lock
);
2820 * rq is already accounted, so use raw insert
2822 if (rq
->cmd_flags
& (REQ_FLUSH
| REQ_FUA
))
2823 __elv_add_request(q
, rq
, ELEVATOR_INSERT_FLUSH
);
2825 __elv_add_request(q
, rq
, ELEVATOR_INSERT_SORT_MERGE
);
2831 * This drops the queue lock
2834 queue_unplugged(q
, depth
, from_schedule
);
2836 local_irq_restore(flags
);
2839 void blk_finish_plug(struct blk_plug
*plug
)
2841 blk_flush_plug_list(plug
, false);
2843 if (plug
== current
->plug
)
2844 current
->plug
= NULL
;
2846 EXPORT_SYMBOL(blk_finish_plug
);
2848 int __init
blk_dev_init(void)
2850 BUILD_BUG_ON(__REQ_NR_BITS
> 8 *
2851 sizeof(((struct request
*)0)->cmd_flags
));
2853 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
2854 kblockd_workqueue
= alloc_workqueue("kblockd",
2855 WQ_MEM_RECLAIM
| WQ_HIGHPRI
, 0);
2856 if (!kblockd_workqueue
)
2857 panic("Failed to create kblockd\n");
2859 request_cachep
= kmem_cache_create("blkdev_requests",
2860 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
2862 blk_requestq_cachep
= kmem_cache_create("blkdev_queue",
2863 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);