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
);
42 DEFINE_IDA(blk_queue_ida
);
45 * For the allocated request tables
47 static struct kmem_cache
*request_cachep
;
50 * For queue allocation
52 struct kmem_cache
*blk_requestq_cachep
;
55 * Controlling structure to kblockd
57 static struct workqueue_struct
*kblockd_workqueue
;
59 static void drive_stat_acct(struct request
*rq
, int new_io
)
61 struct hd_struct
*part
;
62 int rw
= rq_data_dir(rq
);
65 if (!blk_do_io_stat(rq
))
68 cpu
= part_stat_lock();
72 part_stat_inc(cpu
, part
, merges
[rw
]);
74 part
= disk_map_sector_rcu(rq
->rq_disk
, blk_rq_pos(rq
));
75 if (!hd_struct_try_get(part
)) {
77 * The partition is already being removed,
78 * the request will be accounted on the disk only
80 * We take a reference on disk->part0 although that
81 * partition will never be deleted, so we can treat
82 * it as any other partition.
84 part
= &rq
->rq_disk
->part0
;
87 part_round_stats(cpu
, part
);
88 part_inc_in_flight(part
, rw
);
95 void blk_queue_congestion_threshold(struct request_queue
*q
)
99 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) + 1;
100 if (nr
> q
->nr_requests
)
102 q
->nr_congestion_on
= nr
;
104 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) - (q
->nr_requests
/ 16) - 1;
107 q
->nr_congestion_off
= nr
;
111 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
114 * Locates the passed device's request queue and returns the address of its
117 * Will return NULL if the request queue cannot be located.
119 struct backing_dev_info
*blk_get_backing_dev_info(struct block_device
*bdev
)
121 struct backing_dev_info
*ret
= NULL
;
122 struct request_queue
*q
= bdev_get_queue(bdev
);
125 ret
= &q
->backing_dev_info
;
128 EXPORT_SYMBOL(blk_get_backing_dev_info
);
130 void blk_rq_init(struct request_queue
*q
, struct request
*rq
)
132 memset(rq
, 0, sizeof(*rq
));
134 INIT_LIST_HEAD(&rq
->queuelist
);
135 INIT_LIST_HEAD(&rq
->timeout_list
);
138 rq
->__sector
= (sector_t
) -1;
139 INIT_HLIST_NODE(&rq
->hash
);
140 RB_CLEAR_NODE(&rq
->rb_node
);
142 rq
->cmd_len
= BLK_MAX_CDB
;
145 rq
->start_time
= jiffies
;
146 set_start_time_ns(rq
);
149 EXPORT_SYMBOL(blk_rq_init
);
151 static void req_bio_endio(struct request
*rq
, struct bio
*bio
,
152 unsigned int nbytes
, int error
)
155 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
156 else if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
159 if (unlikely(nbytes
> bio
->bi_size
)) {
160 printk(KERN_ERR
"%s: want %u bytes done, %u left\n",
161 __func__
, nbytes
, bio
->bi_size
);
162 nbytes
= bio
->bi_size
;
165 if (unlikely(rq
->cmd_flags
& REQ_QUIET
))
166 set_bit(BIO_QUIET
, &bio
->bi_flags
);
168 bio
->bi_size
-= nbytes
;
169 bio
->bi_sector
+= (nbytes
>> 9);
171 if (bio_integrity(bio
))
172 bio_integrity_advance(bio
, nbytes
);
174 /* don't actually finish bio if it's part of flush sequence */
175 if (bio
->bi_size
== 0 && !(rq
->cmd_flags
& REQ_FLUSH_SEQ
))
176 bio_endio(bio
, error
);
179 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
183 printk(KERN_INFO
"%s: dev %s: type=%x, flags=%x\n", msg
,
184 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->cmd_type
,
187 printk(KERN_INFO
" sector %llu, nr/cnr %u/%u\n",
188 (unsigned long long)blk_rq_pos(rq
),
189 blk_rq_sectors(rq
), blk_rq_cur_sectors(rq
));
190 printk(KERN_INFO
" bio %p, biotail %p, buffer %p, len %u\n",
191 rq
->bio
, rq
->biotail
, rq
->buffer
, blk_rq_bytes(rq
));
193 if (rq
->cmd_type
== REQ_TYPE_BLOCK_PC
) {
194 printk(KERN_INFO
" cdb: ");
195 for (bit
= 0; bit
< BLK_MAX_CDB
; bit
++)
196 printk("%02x ", rq
->cmd
[bit
]);
200 EXPORT_SYMBOL(blk_dump_rq_flags
);
202 static void blk_delay_work(struct work_struct
*work
)
204 struct request_queue
*q
;
206 q
= container_of(work
, struct request_queue
, delay_work
.work
);
207 spin_lock_irq(q
->queue_lock
);
209 spin_unlock_irq(q
->queue_lock
);
213 * blk_delay_queue - restart queueing after defined interval
214 * @q: The &struct request_queue in question
215 * @msecs: Delay in msecs
218 * Sometimes queueing needs to be postponed for a little while, to allow
219 * resources to come back. This function will make sure that queueing is
220 * restarted around the specified time.
222 void blk_delay_queue(struct request_queue
*q
, unsigned long msecs
)
224 queue_delayed_work(kblockd_workqueue
, &q
->delay_work
,
225 msecs_to_jiffies(msecs
));
227 EXPORT_SYMBOL(blk_delay_queue
);
230 * blk_start_queue - restart a previously stopped queue
231 * @q: The &struct request_queue in question
234 * blk_start_queue() will clear the stop flag on the queue, and call
235 * the request_fn for the queue if it was in a stopped state when
236 * entered. Also see blk_stop_queue(). Queue lock must be held.
238 void blk_start_queue(struct request_queue
*q
)
240 WARN_ON(!irqs_disabled());
242 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
245 EXPORT_SYMBOL(blk_start_queue
);
248 * blk_stop_queue - stop a queue
249 * @q: The &struct request_queue in question
252 * The Linux block layer assumes that a block driver will consume all
253 * entries on the request queue when the request_fn strategy is called.
254 * Often this will not happen, because of hardware limitations (queue
255 * depth settings). If a device driver gets a 'queue full' response,
256 * or if it simply chooses not to queue more I/O at one point, it can
257 * call this function to prevent the request_fn from being called until
258 * the driver has signalled it's ready to go again. This happens by calling
259 * blk_start_queue() to restart queue operations. Queue lock must be held.
261 void blk_stop_queue(struct request_queue
*q
)
263 __cancel_delayed_work(&q
->delay_work
);
264 queue_flag_set(QUEUE_FLAG_STOPPED
, q
);
266 EXPORT_SYMBOL(blk_stop_queue
);
269 * blk_sync_queue - cancel any pending callbacks on a queue
273 * The block layer may perform asynchronous callback activity
274 * on a queue, such as calling the unplug function after a timeout.
275 * A block device may call blk_sync_queue to ensure that any
276 * such activity is cancelled, thus allowing it to release resources
277 * that the callbacks might use. The caller must already have made sure
278 * that its ->make_request_fn will not re-add plugging prior to calling
281 * This function does not cancel any asynchronous activity arising
282 * out of elevator or throttling code. That would require elevaotor_exit()
283 * and blk_throtl_exit() to be called with queue lock initialized.
286 void blk_sync_queue(struct request_queue
*q
)
288 del_timer_sync(&q
->timeout
);
289 cancel_delayed_work_sync(&q
->delay_work
);
291 EXPORT_SYMBOL(blk_sync_queue
);
294 * __blk_run_queue - run a single device queue
295 * @q: The queue to run
298 * See @blk_run_queue. This variant must be called with the queue lock
299 * held and interrupts disabled.
301 void __blk_run_queue(struct request_queue
*q
)
303 if (unlikely(blk_queue_stopped(q
)))
308 EXPORT_SYMBOL(__blk_run_queue
);
311 * blk_run_queue_async - run a single device queue in workqueue context
312 * @q: The queue to run
315 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
318 void blk_run_queue_async(struct request_queue
*q
)
320 if (likely(!blk_queue_stopped(q
))) {
321 __cancel_delayed_work(&q
->delay_work
);
322 queue_delayed_work(kblockd_workqueue
, &q
->delay_work
, 0);
325 EXPORT_SYMBOL(blk_run_queue_async
);
328 * blk_run_queue - run a single device queue
329 * @q: The queue to run
332 * Invoke request handling on this queue, if it has pending work to do.
333 * May be used to restart queueing when a request has completed.
335 void blk_run_queue(struct request_queue
*q
)
339 spin_lock_irqsave(q
->queue_lock
, flags
);
341 spin_unlock_irqrestore(q
->queue_lock
, flags
);
343 EXPORT_SYMBOL(blk_run_queue
);
345 void blk_put_queue(struct request_queue
*q
)
347 kobject_put(&q
->kobj
);
349 EXPORT_SYMBOL(blk_put_queue
);
352 * blk_drain_queue - drain requests from request_queue
354 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
356 * Drain requests from @q. If @drain_all is set, all requests are drained.
357 * If not, only ELVPRIV requests are drained. The caller is responsible
358 * for ensuring that no new requests which need to be drained are queued.
360 void blk_drain_queue(struct request_queue
*q
, bool drain_all
)
366 spin_lock_irq(q
->queue_lock
);
369 * The caller might be trying to drain @q before its
370 * elevator is initialized.
373 elv_drain_elevator(q
);
379 * This function might be called on a queue which failed
380 * driver init after queue creation or is not yet fully
381 * active yet. Some drivers (e.g. fd and loop) get unhappy
382 * in such cases. Kick queue iff dispatch queue has
383 * something on it and @q has request_fn set.
385 if (!list_empty(&q
->queue_head
) && q
->request_fn
)
388 drain
|= q
->rq
.elvpriv
;
391 * Unfortunately, requests are queued at and tracked from
392 * multiple places and there's no single counter which can
393 * be drained. Check all the queues and counters.
396 drain
|= !list_empty(&q
->queue_head
);
397 for (i
= 0; i
< 2; i
++) {
398 drain
|= q
->rq
.count
[i
];
399 drain
|= q
->in_flight
[i
];
400 drain
|= !list_empty(&q
->flush_queue
[i
]);
404 spin_unlock_irq(q
->queue_lock
);
413 * blk_queue_bypass_start - enter queue bypass mode
414 * @q: queue of interest
416 * In bypass mode, only the dispatch FIFO queue of @q is used. This
417 * function makes @q enter bypass mode and drains all requests which were
418 * issued before. On return, it's guaranteed that no request has ELVPRIV
421 void blk_queue_bypass_start(struct request_queue
*q
)
423 spin_lock_irq(q
->queue_lock
);
425 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
426 spin_unlock_irq(q
->queue_lock
);
428 blk_drain_queue(q
, false);
430 EXPORT_SYMBOL_GPL(blk_queue_bypass_start
);
433 * blk_queue_bypass_end - leave queue bypass mode
434 * @q: queue of interest
436 * Leave bypass mode and restore the normal queueing behavior.
438 void blk_queue_bypass_end(struct request_queue
*q
)
440 spin_lock_irq(q
->queue_lock
);
441 if (!--q
->bypass_depth
)
442 queue_flag_clear(QUEUE_FLAG_BYPASS
, q
);
443 WARN_ON_ONCE(q
->bypass_depth
< 0);
444 spin_unlock_irq(q
->queue_lock
);
446 EXPORT_SYMBOL_GPL(blk_queue_bypass_end
);
449 * blk_cleanup_queue - shutdown a request queue
450 * @q: request queue to shutdown
452 * Mark @q DEAD, drain all pending requests, destroy and put it. All
453 * future requests will be failed immediately with -ENODEV.
455 void blk_cleanup_queue(struct request_queue
*q
)
457 spinlock_t
*lock
= q
->queue_lock
;
459 /* mark @q DEAD, no new request or merges will be allowed afterwards */
460 mutex_lock(&q
->sysfs_lock
);
461 queue_flag_set_unlocked(QUEUE_FLAG_DEAD
, q
);
464 queue_flag_set(QUEUE_FLAG_NOMERGES
, q
);
465 queue_flag_set(QUEUE_FLAG_NOXMERGES
, q
);
466 queue_flag_set(QUEUE_FLAG_DEAD
, q
);
468 if (q
->queue_lock
!= &q
->__queue_lock
)
469 q
->queue_lock
= &q
->__queue_lock
;
471 spin_unlock_irq(lock
);
472 mutex_unlock(&q
->sysfs_lock
);
474 /* drain all requests queued before DEAD marking */
475 blk_drain_queue(q
, true);
477 /* @q won't process any more request, flush async actions */
478 del_timer_sync(&q
->backing_dev_info
.laptop_mode_wb_timer
);
481 /* @q is and will stay empty, shutdown and put */
484 EXPORT_SYMBOL(blk_cleanup_queue
);
486 static int blk_init_free_list(struct request_queue
*q
)
488 struct request_list
*rl
= &q
->rq
;
490 if (unlikely(rl
->rq_pool
))
493 rl
->count
[BLK_RW_SYNC
] = rl
->count
[BLK_RW_ASYNC
] = 0;
494 rl
->starved
[BLK_RW_SYNC
] = rl
->starved
[BLK_RW_ASYNC
] = 0;
496 init_waitqueue_head(&rl
->wait
[BLK_RW_SYNC
]);
497 init_waitqueue_head(&rl
->wait
[BLK_RW_ASYNC
]);
499 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, mempool_alloc_slab
,
500 mempool_free_slab
, request_cachep
, q
->node
);
508 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
510 return blk_alloc_queue_node(gfp_mask
, -1);
512 EXPORT_SYMBOL(blk_alloc_queue
);
514 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
516 struct request_queue
*q
;
519 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
520 gfp_mask
| __GFP_ZERO
, node_id
);
524 q
->id
= ida_simple_get(&blk_queue_ida
, 0, 0, GFP_KERNEL
);
528 q
->backing_dev_info
.ra_pages
=
529 (VM_MAX_READAHEAD
* 1024) / PAGE_CACHE_SIZE
;
530 q
->backing_dev_info
.state
= 0;
531 q
->backing_dev_info
.capabilities
= BDI_CAP_MAP_COPY
;
532 q
->backing_dev_info
.name
= "block";
535 err
= bdi_init(&q
->backing_dev_info
);
539 if (blk_throtl_init(q
))
542 setup_timer(&q
->backing_dev_info
.laptop_mode_wb_timer
,
543 laptop_mode_timer_fn
, (unsigned long) q
);
544 setup_timer(&q
->timeout
, blk_rq_timed_out_timer
, (unsigned long) q
);
545 INIT_LIST_HEAD(&q
->queue_head
);
546 INIT_LIST_HEAD(&q
->timeout_list
);
547 INIT_LIST_HEAD(&q
->icq_list
);
548 INIT_LIST_HEAD(&q
->flush_queue
[0]);
549 INIT_LIST_HEAD(&q
->flush_queue
[1]);
550 INIT_LIST_HEAD(&q
->flush_data_in_flight
);
551 INIT_DELAYED_WORK(&q
->delay_work
, blk_delay_work
);
553 kobject_init(&q
->kobj
, &blk_queue_ktype
);
555 mutex_init(&q
->sysfs_lock
);
556 spin_lock_init(&q
->__queue_lock
);
559 * By default initialize queue_lock to internal lock and driver can
560 * override it later if need be.
562 q
->queue_lock
= &q
->__queue_lock
;
567 ida_simple_remove(&blk_queue_ida
, q
->id
);
569 kmem_cache_free(blk_requestq_cachep
, q
);
572 EXPORT_SYMBOL(blk_alloc_queue_node
);
575 * blk_init_queue - prepare a request queue for use with a block device
576 * @rfn: The function to be called to process requests that have been
577 * placed on the queue.
578 * @lock: Request queue spin lock
581 * If a block device wishes to use the standard request handling procedures,
582 * which sorts requests and coalesces adjacent requests, then it must
583 * call blk_init_queue(). The function @rfn will be called when there
584 * are requests on the queue that need to be processed. If the device
585 * supports plugging, then @rfn may not be called immediately when requests
586 * are available on the queue, but may be called at some time later instead.
587 * Plugged queues are generally unplugged when a buffer belonging to one
588 * of the requests on the queue is needed, or due to memory pressure.
590 * @rfn is not required, or even expected, to remove all requests off the
591 * queue, but only as many as it can handle at a time. If it does leave
592 * requests on the queue, it is responsible for arranging that the requests
593 * get dealt with eventually.
595 * The queue spin lock must be held while manipulating the requests on the
596 * request queue; this lock will be taken also from interrupt context, so irq
597 * disabling is needed for it.
599 * Function returns a pointer to the initialized request queue, or %NULL if
603 * blk_init_queue() must be paired with a blk_cleanup_queue() call
604 * when the block device is deactivated (such as at module unload).
607 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
609 return blk_init_queue_node(rfn
, lock
, -1);
611 EXPORT_SYMBOL(blk_init_queue
);
613 struct request_queue
*
614 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
616 struct request_queue
*uninit_q
, *q
;
618 uninit_q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
622 q
= blk_init_allocated_queue(uninit_q
, rfn
, lock
);
624 blk_cleanup_queue(uninit_q
);
628 EXPORT_SYMBOL(blk_init_queue_node
);
630 struct request_queue
*
631 blk_init_allocated_queue(struct request_queue
*q
, request_fn_proc
*rfn
,
637 if (blk_init_free_list(q
))
641 q
->prep_rq_fn
= NULL
;
642 q
->unprep_rq_fn
= NULL
;
643 q
->queue_flags
= QUEUE_FLAG_DEFAULT
;
645 /* Override internal queue lock with supplied lock pointer */
647 q
->queue_lock
= lock
;
650 * This also sets hw/phys segments, boundary and size
652 blk_queue_make_request(q
, blk_queue_bio
);
654 q
->sg_reserved_size
= INT_MAX
;
659 if (!elevator_init(q
, NULL
)) {
660 blk_queue_congestion_threshold(q
);
666 EXPORT_SYMBOL(blk_init_allocated_queue
);
668 bool blk_get_queue(struct request_queue
*q
)
670 if (likely(!blk_queue_dead(q
))) {
677 EXPORT_SYMBOL(blk_get_queue
);
679 static inline void blk_free_request(struct request_queue
*q
, struct request
*rq
)
681 if (rq
->cmd_flags
& REQ_ELVPRIV
) {
682 elv_put_request(q
, rq
);
684 put_io_context(rq
->elv
.icq
->ioc
);
687 mempool_free(rq
, q
->rq
.rq_pool
);
690 static struct request
*
691 blk_alloc_request(struct request_queue
*q
, struct io_cq
*icq
,
692 unsigned int flags
, gfp_t gfp_mask
)
694 struct request
*rq
= mempool_alloc(q
->rq
.rq_pool
, gfp_mask
);
701 rq
->cmd_flags
= flags
| REQ_ALLOCED
;
703 if (flags
& REQ_ELVPRIV
) {
705 if (unlikely(elv_set_request(q
, rq
, gfp_mask
))) {
706 mempool_free(rq
, q
->rq
.rq_pool
);
709 /* @rq->elv.icq holds on to io_context until @rq is freed */
711 get_io_context(icq
->ioc
);
718 * ioc_batching returns true if the ioc is a valid batching request and
719 * should be given priority access to a request.
721 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
727 * Make sure the process is able to allocate at least 1 request
728 * even if the batch times out, otherwise we could theoretically
731 return ioc
->nr_batch_requests
== q
->nr_batching
||
732 (ioc
->nr_batch_requests
> 0
733 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
737 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
738 * will cause the process to be a "batcher" on all queues in the system. This
739 * is the behaviour we want though - once it gets a wakeup it should be given
742 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
744 if (!ioc
|| ioc_batching(q
, ioc
))
747 ioc
->nr_batch_requests
= q
->nr_batching
;
748 ioc
->last_waited
= jiffies
;
751 static void __freed_request(struct request_queue
*q
, int sync
)
753 struct request_list
*rl
= &q
->rq
;
755 if (rl
->count
[sync
] < queue_congestion_off_threshold(q
))
756 blk_clear_queue_congested(q
, sync
);
758 if (rl
->count
[sync
] + 1 <= q
->nr_requests
) {
759 if (waitqueue_active(&rl
->wait
[sync
]))
760 wake_up(&rl
->wait
[sync
]);
762 blk_clear_queue_full(q
, sync
);
767 * A request has just been released. Account for it, update the full and
768 * congestion status, wake up any waiters. Called under q->queue_lock.
770 static void freed_request(struct request_queue
*q
, unsigned int flags
)
772 struct request_list
*rl
= &q
->rq
;
773 int sync
= rw_is_sync(flags
);
776 if (flags
& REQ_ELVPRIV
)
779 __freed_request(q
, sync
);
781 if (unlikely(rl
->starved
[sync
^ 1]))
782 __freed_request(q
, sync
^ 1);
786 * Determine if elevator data should be initialized when allocating the
787 * request associated with @bio.
789 static bool blk_rq_should_init_elevator(struct bio
*bio
)
795 * Flush requests do not use the elevator so skip initialization.
796 * This allows a request to share the flush and elevator data.
798 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
))
805 * get_request - get a free request
806 * @q: request_queue to allocate request from
807 * @rw_flags: RW and SYNC flags
808 * @bio: bio to allocate request for (can be %NULL)
809 * @gfp_mask: allocation mask
811 * Get a free request from @q. This function may fail under memory
812 * pressure or if @q is dead.
814 * Must be callled with @q->queue_lock held and,
815 * Returns %NULL on failure, with @q->queue_lock held.
816 * Returns !%NULL on success, with @q->queue_lock *not held*.
818 static struct request
*get_request(struct request_queue
*q
, int rw_flags
,
819 struct bio
*bio
, gfp_t gfp_mask
)
821 struct request
*rq
= NULL
;
822 struct request_list
*rl
= &q
->rq
;
823 struct elevator_type
*et
;
824 struct io_context
*ioc
;
825 struct io_cq
*icq
= NULL
;
826 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
827 bool retried
= false;
830 et
= q
->elevator
->type
;
831 ioc
= current
->io_context
;
833 if (unlikely(blk_queue_dead(q
)))
836 may_queue
= elv_may_queue(q
, rw_flags
);
837 if (may_queue
== ELV_MQUEUE_NO
)
840 if (rl
->count
[is_sync
]+1 >= queue_congestion_on_threshold(q
)) {
841 if (rl
->count
[is_sync
]+1 >= q
->nr_requests
) {
843 * We want ioc to record batching state. If it's
844 * not already there, creating a new one requires
845 * dropping queue_lock, which in turn requires
846 * retesting conditions to avoid queue hang.
848 if (!ioc
&& !retried
) {
849 spin_unlock_irq(q
->queue_lock
);
850 create_io_context(current
, gfp_mask
, q
->node
);
851 spin_lock_irq(q
->queue_lock
);
857 * The queue will fill after this allocation, so set
858 * it as full, and mark this process as "batching".
859 * This process will be allowed to complete a batch of
860 * requests, others will be blocked.
862 if (!blk_queue_full(q
, is_sync
)) {
863 ioc_set_batching(q
, ioc
);
864 blk_set_queue_full(q
, is_sync
);
866 if (may_queue
!= ELV_MQUEUE_MUST
867 && !ioc_batching(q
, ioc
)) {
869 * The queue is full and the allocating
870 * process is not a "batcher", and not
871 * exempted by the IO scheduler
877 blk_set_queue_congested(q
, is_sync
);
881 * Only allow batching queuers to allocate up to 50% over the defined
882 * limit of requests, otherwise we could have thousands of requests
883 * allocated with any setting of ->nr_requests
885 if (rl
->count
[is_sync
] >= (3 * q
->nr_requests
/ 2))
888 rl
->count
[is_sync
]++;
889 rl
->starved
[is_sync
] = 0;
892 * Decide whether the new request will be managed by elevator. If
893 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
894 * prevent the current elevator from being destroyed until the new
895 * request is freed. This guarantees icq's won't be destroyed and
896 * makes creating new ones safe.
898 * Also, lookup icq while holding queue_lock. If it doesn't exist,
899 * it will be created after releasing queue_lock.
901 if (blk_rq_should_init_elevator(bio
) && !blk_queue_bypass(q
)) {
902 rw_flags
|= REQ_ELVPRIV
;
904 if (et
->icq_cache
&& ioc
)
905 icq
= ioc_lookup_icq(ioc
, q
);
908 if (blk_queue_io_stat(q
))
909 rw_flags
|= REQ_IO_STAT
;
910 spin_unlock_irq(q
->queue_lock
);
912 /* create icq if missing */
913 if ((rw_flags
& REQ_ELVPRIV
) && unlikely(et
->icq_cache
&& !icq
)) {
914 icq
= ioc_create_icq(q
, gfp_mask
);
919 rq
= blk_alloc_request(q
, icq
, rw_flags
, gfp_mask
);
924 * Allocation failed presumably due to memory. Undo anything
925 * we might have messed up.
927 * Allocating task should really be put onto the front of the
928 * wait queue, but this is pretty rare.
930 spin_lock_irq(q
->queue_lock
);
931 freed_request(q
, rw_flags
);
934 * in the very unlikely event that allocation failed and no
935 * requests for this direction was pending, mark us starved
936 * so that freeing of a request in the other direction will
937 * notice us. another possible fix would be to split the
938 * rq mempool into READ and WRITE
941 if (unlikely(rl
->count
[is_sync
] == 0))
942 rl
->starved
[is_sync
] = 1;
948 * ioc may be NULL here, and ioc_batching will be false. That's
949 * OK, if the queue is under the request limit then requests need
950 * not count toward the nr_batch_requests limit. There will always
951 * be some limit enforced by BLK_BATCH_TIME.
953 if (ioc_batching(q
, ioc
))
954 ioc
->nr_batch_requests
--;
956 trace_block_getrq(q
, bio
, rw_flags
& 1);
962 * get_request_wait - get a free request with retry
963 * @q: request_queue to allocate request from
964 * @rw_flags: RW and SYNC flags
965 * @bio: bio to allocate request for (can be %NULL)
967 * Get a free request from @q. This function keeps retrying under memory
968 * pressure and fails iff @q is dead.
970 * Must be callled with @q->queue_lock held and,
971 * Returns %NULL on failure, with @q->queue_lock held.
972 * Returns !%NULL on success, with @q->queue_lock *not held*.
974 static struct request
*get_request_wait(struct request_queue
*q
, int rw_flags
,
977 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
980 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
983 struct request_list
*rl
= &q
->rq
;
985 if (unlikely(blk_queue_dead(q
)))
988 prepare_to_wait_exclusive(&rl
->wait
[is_sync
], &wait
,
989 TASK_UNINTERRUPTIBLE
);
991 trace_block_sleeprq(q
, bio
, rw_flags
& 1);
993 spin_unlock_irq(q
->queue_lock
);
997 * After sleeping, we become a "batching" process and
998 * will be able to allocate at least one request, and
999 * up to a big batch of them for a small period time.
1000 * See ioc_batching, ioc_set_batching
1002 create_io_context(current
, GFP_NOIO
, q
->node
);
1003 ioc_set_batching(q
, current
->io_context
);
1005 spin_lock_irq(q
->queue_lock
);
1006 finish_wait(&rl
->wait
[is_sync
], &wait
);
1008 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
1014 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
1018 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
1020 spin_lock_irq(q
->queue_lock
);
1021 if (gfp_mask
& __GFP_WAIT
)
1022 rq
= get_request_wait(q
, rw
, NULL
);
1024 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
1026 spin_unlock_irq(q
->queue_lock
);
1027 /* q->queue_lock is unlocked at this point */
1031 EXPORT_SYMBOL(blk_get_request
);
1034 * blk_make_request - given a bio, allocate a corresponding struct request.
1035 * @q: target request queue
1036 * @bio: The bio describing the memory mappings that will be submitted for IO.
1037 * It may be a chained-bio properly constructed by block/bio layer.
1038 * @gfp_mask: gfp flags to be used for memory allocation
1040 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1041 * type commands. Where the struct request needs to be farther initialized by
1042 * the caller. It is passed a &struct bio, which describes the memory info of
1045 * The caller of blk_make_request must make sure that bi_io_vec
1046 * are set to describe the memory buffers. That bio_data_dir() will return
1047 * the needed direction of the request. (And all bio's in the passed bio-chain
1048 * are properly set accordingly)
1050 * If called under none-sleepable conditions, mapped bio buffers must not
1051 * need bouncing, by calling the appropriate masked or flagged allocator,
1052 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1055 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1056 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1057 * anything but the first bio in the chain. Otherwise you risk waiting for IO
1058 * completion of a bio that hasn't been submitted yet, thus resulting in a
1059 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1060 * of bio_alloc(), as that avoids the mempool deadlock.
1061 * If possible a big IO should be split into smaller parts when allocation
1062 * fails. Partial allocation should not be an error, or you risk a live-lock.
1064 struct request
*blk_make_request(struct request_queue
*q
, struct bio
*bio
,
1067 struct request
*rq
= blk_get_request(q
, bio_data_dir(bio
), gfp_mask
);
1070 return ERR_PTR(-ENOMEM
);
1073 struct bio
*bounce_bio
= bio
;
1076 blk_queue_bounce(q
, &bounce_bio
);
1077 ret
= blk_rq_append_bio(q
, rq
, bounce_bio
);
1078 if (unlikely(ret
)) {
1079 blk_put_request(rq
);
1080 return ERR_PTR(ret
);
1086 EXPORT_SYMBOL(blk_make_request
);
1089 * blk_requeue_request - put a request back on queue
1090 * @q: request queue where request should be inserted
1091 * @rq: request to be inserted
1094 * Drivers often keep queueing requests until the hardware cannot accept
1095 * more, when that condition happens we need to put the request back
1096 * on the queue. Must be called with queue lock held.
1098 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
1100 blk_delete_timer(rq
);
1101 blk_clear_rq_complete(rq
);
1102 trace_block_rq_requeue(q
, rq
);
1104 if (blk_rq_tagged(rq
))
1105 blk_queue_end_tag(q
, rq
);
1107 BUG_ON(blk_queued_rq(rq
));
1109 elv_requeue_request(q
, rq
);
1111 EXPORT_SYMBOL(blk_requeue_request
);
1113 static void add_acct_request(struct request_queue
*q
, struct request
*rq
,
1116 drive_stat_acct(rq
, 1);
1117 __elv_add_request(q
, rq
, where
);
1120 static void part_round_stats_single(int cpu
, struct hd_struct
*part
,
1123 if (now
== part
->stamp
)
1126 if (part_in_flight(part
)) {
1127 __part_stat_add(cpu
, part
, time_in_queue
,
1128 part_in_flight(part
) * (now
- part
->stamp
));
1129 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
1135 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1136 * @cpu: cpu number for stats access
1137 * @part: target partition
1139 * The average IO queue length and utilisation statistics are maintained
1140 * by observing the current state of the queue length and the amount of
1141 * time it has been in this state for.
1143 * Normally, that accounting is done on IO completion, but that can result
1144 * in more than a second's worth of IO being accounted for within any one
1145 * second, leading to >100% utilisation. To deal with that, we call this
1146 * function to do a round-off before returning the results when reading
1147 * /proc/diskstats. This accounts immediately for all queue usage up to
1148 * the current jiffies and restarts the counters again.
1150 void part_round_stats(int cpu
, struct hd_struct
*part
)
1152 unsigned long now
= jiffies
;
1155 part_round_stats_single(cpu
, &part_to_disk(part
)->part0
, now
);
1156 part_round_stats_single(cpu
, part
, now
);
1158 EXPORT_SYMBOL_GPL(part_round_stats
);
1161 * queue lock must be held
1163 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1167 if (unlikely(--req
->ref_count
))
1170 elv_completed_request(q
, req
);
1172 /* this is a bio leak */
1173 WARN_ON(req
->bio
!= NULL
);
1176 * Request may not have originated from ll_rw_blk. if not,
1177 * it didn't come out of our reserved rq pools
1179 if (req
->cmd_flags
& REQ_ALLOCED
) {
1180 unsigned int flags
= req
->cmd_flags
;
1182 BUG_ON(!list_empty(&req
->queuelist
));
1183 BUG_ON(!hlist_unhashed(&req
->hash
));
1185 blk_free_request(q
, req
);
1186 freed_request(q
, flags
);
1189 EXPORT_SYMBOL_GPL(__blk_put_request
);
1191 void blk_put_request(struct request
*req
)
1193 unsigned long flags
;
1194 struct request_queue
*q
= req
->q
;
1196 spin_lock_irqsave(q
->queue_lock
, flags
);
1197 __blk_put_request(q
, req
);
1198 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1200 EXPORT_SYMBOL(blk_put_request
);
1203 * blk_add_request_payload - add a payload to a request
1204 * @rq: request to update
1205 * @page: page backing the payload
1206 * @len: length of the payload.
1208 * This allows to later add a payload to an already submitted request by
1209 * a block driver. The driver needs to take care of freeing the payload
1212 * Note that this is a quite horrible hack and nothing but handling of
1213 * discard requests should ever use it.
1215 void blk_add_request_payload(struct request
*rq
, struct page
*page
,
1218 struct bio
*bio
= rq
->bio
;
1220 bio
->bi_io_vec
->bv_page
= page
;
1221 bio
->bi_io_vec
->bv_offset
= 0;
1222 bio
->bi_io_vec
->bv_len
= len
;
1226 bio
->bi_phys_segments
= 1;
1228 rq
->__data_len
= rq
->resid_len
= len
;
1229 rq
->nr_phys_segments
= 1;
1230 rq
->buffer
= bio_data(bio
);
1232 EXPORT_SYMBOL_GPL(blk_add_request_payload
);
1234 static bool bio_attempt_back_merge(struct request_queue
*q
, struct request
*req
,
1237 const int ff
= bio
->bi_rw
& REQ_FAILFAST_MASK
;
1239 if (!ll_back_merge_fn(q
, req
, bio
))
1242 trace_block_bio_backmerge(q
, bio
);
1244 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1245 blk_rq_set_mixed_merge(req
);
1247 req
->biotail
->bi_next
= bio
;
1249 req
->__data_len
+= bio
->bi_size
;
1250 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1252 drive_stat_acct(req
, 0);
1256 static bool bio_attempt_front_merge(struct request_queue
*q
,
1257 struct request
*req
, struct bio
*bio
)
1259 const int ff
= bio
->bi_rw
& REQ_FAILFAST_MASK
;
1261 if (!ll_front_merge_fn(q
, req
, bio
))
1264 trace_block_bio_frontmerge(q
, bio
);
1266 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1267 blk_rq_set_mixed_merge(req
);
1269 bio
->bi_next
= req
->bio
;
1273 * may not be valid. if the low level driver said
1274 * it didn't need a bounce buffer then it better
1275 * not touch req->buffer either...
1277 req
->buffer
= bio_data(bio
);
1278 req
->__sector
= bio
->bi_sector
;
1279 req
->__data_len
+= bio
->bi_size
;
1280 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1282 drive_stat_acct(req
, 0);
1287 * attempt_plug_merge - try to merge with %current's plugged list
1288 * @q: request_queue new bio is being queued at
1289 * @bio: new bio being queued
1290 * @request_count: out parameter for number of traversed plugged requests
1292 * Determine whether @bio being queued on @q can be merged with a request
1293 * on %current's plugged list. Returns %true if merge was successful,
1296 * Plugging coalesces IOs from the same issuer for the same purpose without
1297 * going through @q->queue_lock. As such it's more of an issuing mechanism
1298 * than scheduling, and the request, while may have elvpriv data, is not
1299 * added on the elevator at this point. In addition, we don't have
1300 * reliable access to the elevator outside queue lock. Only check basic
1301 * merging parameters without querying the elevator.
1303 static bool attempt_plug_merge(struct request_queue
*q
, struct bio
*bio
,
1304 unsigned int *request_count
)
1306 struct blk_plug
*plug
;
1310 plug
= current
->plug
;
1315 list_for_each_entry_reverse(rq
, &plug
->list
, queuelist
) {
1320 if (rq
->q
!= q
|| !blk_rq_merge_ok(rq
, bio
))
1323 el_ret
= blk_try_merge(rq
, bio
);
1324 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1325 ret
= bio_attempt_back_merge(q
, rq
, bio
);
1328 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1329 ret
= bio_attempt_front_merge(q
, rq
, bio
);
1338 void init_request_from_bio(struct request
*req
, struct bio
*bio
)
1340 req
->cmd_type
= REQ_TYPE_FS
;
1342 req
->cmd_flags
|= bio
->bi_rw
& REQ_COMMON_MASK
;
1343 if (bio
->bi_rw
& REQ_RAHEAD
)
1344 req
->cmd_flags
|= REQ_FAILFAST_MASK
;
1347 req
->__sector
= bio
->bi_sector
;
1348 req
->ioprio
= bio_prio(bio
);
1349 blk_rq_bio_prep(req
->q
, req
, bio
);
1352 void blk_queue_bio(struct request_queue
*q
, struct bio
*bio
)
1354 const bool sync
= !!(bio
->bi_rw
& REQ_SYNC
);
1355 struct blk_plug
*plug
;
1356 int el_ret
, rw_flags
, where
= ELEVATOR_INSERT_SORT
;
1357 struct request
*req
;
1358 unsigned int request_count
= 0;
1361 * low level driver can indicate that it wants pages above a
1362 * certain limit bounced to low memory (ie for highmem, or even
1363 * ISA dma in theory)
1365 blk_queue_bounce(q
, &bio
);
1367 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) {
1368 spin_lock_irq(q
->queue_lock
);
1369 where
= ELEVATOR_INSERT_FLUSH
;
1374 * Check if we can merge with the plugged list before grabbing
1377 if (attempt_plug_merge(q
, bio
, &request_count
))
1380 spin_lock_irq(q
->queue_lock
);
1382 el_ret
= elv_merge(q
, &req
, bio
);
1383 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1384 if (bio_attempt_back_merge(q
, req
, bio
)) {
1385 elv_bio_merged(q
, req
, bio
);
1386 if (!attempt_back_merge(q
, req
))
1387 elv_merged_request(q
, req
, el_ret
);
1390 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1391 if (bio_attempt_front_merge(q
, req
, bio
)) {
1392 elv_bio_merged(q
, req
, bio
);
1393 if (!attempt_front_merge(q
, req
))
1394 elv_merged_request(q
, req
, el_ret
);
1401 * This sync check and mask will be re-done in init_request_from_bio(),
1402 * but we need to set it earlier to expose the sync flag to the
1403 * rq allocator and io schedulers.
1405 rw_flags
= bio_data_dir(bio
);
1407 rw_flags
|= REQ_SYNC
;
1410 * Grab a free request. This is might sleep but can not fail.
1411 * Returns with the queue unlocked.
1413 req
= get_request_wait(q
, rw_flags
, bio
);
1414 if (unlikely(!req
)) {
1415 bio_endio(bio
, -ENODEV
); /* @q is dead */
1420 * After dropping the lock and possibly sleeping here, our request
1421 * may now be mergeable after it had proven unmergeable (above).
1422 * We don't worry about that case for efficiency. It won't happen
1423 * often, and the elevators are able to handle it.
1425 init_request_from_bio(req
, bio
);
1427 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
))
1428 req
->cpu
= raw_smp_processor_id();
1430 plug
= current
->plug
;
1433 * If this is the first request added after a plug, fire
1434 * of a plug trace. If others have been added before, check
1435 * if we have multiple devices in this plug. If so, make a
1436 * note to sort the list before dispatch.
1438 if (list_empty(&plug
->list
))
1439 trace_block_plug(q
);
1441 if (!plug
->should_sort
) {
1442 struct request
*__rq
;
1444 __rq
= list_entry_rq(plug
->list
.prev
);
1446 plug
->should_sort
= 1;
1448 if (request_count
>= BLK_MAX_REQUEST_COUNT
) {
1449 blk_flush_plug_list(plug
, false);
1450 trace_block_plug(q
);
1453 list_add_tail(&req
->queuelist
, &plug
->list
);
1454 drive_stat_acct(req
, 1);
1456 spin_lock_irq(q
->queue_lock
);
1457 add_acct_request(q
, req
, where
);
1460 spin_unlock_irq(q
->queue_lock
);
1463 EXPORT_SYMBOL_GPL(blk_queue_bio
); /* for device mapper only */
1466 * If bio->bi_dev is a partition, remap the location
1468 static inline void blk_partition_remap(struct bio
*bio
)
1470 struct block_device
*bdev
= bio
->bi_bdev
;
1472 if (bio_sectors(bio
) && bdev
!= bdev
->bd_contains
) {
1473 struct hd_struct
*p
= bdev
->bd_part
;
1475 bio
->bi_sector
+= p
->start_sect
;
1476 bio
->bi_bdev
= bdev
->bd_contains
;
1478 trace_block_bio_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
1480 bio
->bi_sector
- p
->start_sect
);
1484 static void handle_bad_sector(struct bio
*bio
)
1486 char b
[BDEVNAME_SIZE
];
1488 printk(KERN_INFO
"attempt to access beyond end of device\n");
1489 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
1490 bdevname(bio
->bi_bdev
, b
),
1492 (unsigned long long)bio
->bi_sector
+ bio_sectors(bio
),
1493 (long long)(i_size_read(bio
->bi_bdev
->bd_inode
) >> 9));
1495 set_bit(BIO_EOF
, &bio
->bi_flags
);
1498 #ifdef CONFIG_FAIL_MAKE_REQUEST
1500 static DECLARE_FAULT_ATTR(fail_make_request
);
1502 static int __init
setup_fail_make_request(char *str
)
1504 return setup_fault_attr(&fail_make_request
, str
);
1506 __setup("fail_make_request=", setup_fail_make_request
);
1508 static bool should_fail_request(struct hd_struct
*part
, unsigned int bytes
)
1510 return part
->make_it_fail
&& should_fail(&fail_make_request
, bytes
);
1513 static int __init
fail_make_request_debugfs(void)
1515 struct dentry
*dir
= fault_create_debugfs_attr("fail_make_request",
1516 NULL
, &fail_make_request
);
1518 return IS_ERR(dir
) ? PTR_ERR(dir
) : 0;
1521 late_initcall(fail_make_request_debugfs
);
1523 #else /* CONFIG_FAIL_MAKE_REQUEST */
1525 static inline bool should_fail_request(struct hd_struct
*part
,
1531 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1534 * Check whether this bio extends beyond the end of the device.
1536 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
1543 /* Test device or partition size, when known. */
1544 maxsector
= i_size_read(bio
->bi_bdev
->bd_inode
) >> 9;
1546 sector_t sector
= bio
->bi_sector
;
1548 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
1550 * This may well happen - the kernel calls bread()
1551 * without checking the size of the device, e.g., when
1552 * mounting a device.
1554 handle_bad_sector(bio
);
1562 static noinline_for_stack
bool
1563 generic_make_request_checks(struct bio
*bio
)
1565 struct request_queue
*q
;
1566 int nr_sectors
= bio_sectors(bio
);
1568 char b
[BDEVNAME_SIZE
];
1569 struct hd_struct
*part
;
1573 if (bio_check_eod(bio
, nr_sectors
))
1576 q
= bdev_get_queue(bio
->bi_bdev
);
1579 "generic_make_request: Trying to access "
1580 "nonexistent block-device %s (%Lu)\n",
1581 bdevname(bio
->bi_bdev
, b
),
1582 (long long) bio
->bi_sector
);
1586 if (unlikely(!(bio
->bi_rw
& REQ_DISCARD
) &&
1587 nr_sectors
> queue_max_hw_sectors(q
))) {
1588 printk(KERN_ERR
"bio too big device %s (%u > %u)\n",
1589 bdevname(bio
->bi_bdev
, b
),
1591 queue_max_hw_sectors(q
));
1595 part
= bio
->bi_bdev
->bd_part
;
1596 if (should_fail_request(part
, bio
->bi_size
) ||
1597 should_fail_request(&part_to_disk(part
)->part0
,
1602 * If this device has partitions, remap block n
1603 * of partition p to block n+start(p) of the disk.
1605 blk_partition_remap(bio
);
1607 if (bio_integrity_enabled(bio
) && bio_integrity_prep(bio
))
1610 if (bio_check_eod(bio
, nr_sectors
))
1614 * Filter flush bio's early so that make_request based
1615 * drivers without flush support don't have to worry
1618 if ((bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) && !q
->flush_flags
) {
1619 bio
->bi_rw
&= ~(REQ_FLUSH
| REQ_FUA
);
1626 if ((bio
->bi_rw
& REQ_DISCARD
) &&
1627 (!blk_queue_discard(q
) ||
1628 ((bio
->bi_rw
& REQ_SECURE
) &&
1629 !blk_queue_secdiscard(q
)))) {
1634 if (blk_throtl_bio(q
, bio
))
1635 return false; /* throttled, will be resubmitted later */
1637 trace_block_bio_queue(q
, bio
);
1641 bio_endio(bio
, err
);
1646 * generic_make_request - hand a buffer to its device driver for I/O
1647 * @bio: The bio describing the location in memory and on the device.
1649 * generic_make_request() is used to make I/O requests of block
1650 * devices. It is passed a &struct bio, which describes the I/O that needs
1653 * generic_make_request() does not return any status. The
1654 * success/failure status of the request, along with notification of
1655 * completion, is delivered asynchronously through the bio->bi_end_io
1656 * function described (one day) else where.
1658 * The caller of generic_make_request must make sure that bi_io_vec
1659 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1660 * set to describe the device address, and the
1661 * bi_end_io and optionally bi_private are set to describe how
1662 * completion notification should be signaled.
1664 * generic_make_request and the drivers it calls may use bi_next if this
1665 * bio happens to be merged with someone else, and may resubmit the bio to
1666 * a lower device by calling into generic_make_request recursively, which
1667 * means the bio should NOT be touched after the call to ->make_request_fn.
1669 void generic_make_request(struct bio
*bio
)
1671 struct bio_list bio_list_on_stack
;
1673 if (!generic_make_request_checks(bio
))
1677 * We only want one ->make_request_fn to be active at a time, else
1678 * stack usage with stacked devices could be a problem. So use
1679 * current->bio_list to keep a list of requests submited by a
1680 * make_request_fn function. current->bio_list is also used as a
1681 * flag to say if generic_make_request is currently active in this
1682 * task or not. If it is NULL, then no make_request is active. If
1683 * it is non-NULL, then a make_request is active, and new requests
1684 * should be added at the tail
1686 if (current
->bio_list
) {
1687 bio_list_add(current
->bio_list
, bio
);
1691 /* following loop may be a bit non-obvious, and so deserves some
1693 * Before entering the loop, bio->bi_next is NULL (as all callers
1694 * ensure that) so we have a list with a single bio.
1695 * We pretend that we have just taken it off a longer list, so
1696 * we assign bio_list to a pointer to the bio_list_on_stack,
1697 * thus initialising the bio_list of new bios to be
1698 * added. ->make_request() may indeed add some more bios
1699 * through a recursive call to generic_make_request. If it
1700 * did, we find a non-NULL value in bio_list and re-enter the loop
1701 * from the top. In this case we really did just take the bio
1702 * of the top of the list (no pretending) and so remove it from
1703 * bio_list, and call into ->make_request() again.
1705 BUG_ON(bio
->bi_next
);
1706 bio_list_init(&bio_list_on_stack
);
1707 current
->bio_list
= &bio_list_on_stack
;
1709 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
1711 q
->make_request_fn(q
, bio
);
1713 bio
= bio_list_pop(current
->bio_list
);
1715 current
->bio_list
= NULL
; /* deactivate */
1717 EXPORT_SYMBOL(generic_make_request
);
1720 * submit_bio - submit a bio to the block device layer for I/O
1721 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1722 * @bio: The &struct bio which describes the I/O
1724 * submit_bio() is very similar in purpose to generic_make_request(), and
1725 * uses that function to do most of the work. Both are fairly rough
1726 * interfaces; @bio must be presetup and ready for I/O.
1729 void submit_bio(int rw
, struct bio
*bio
)
1731 int count
= bio_sectors(bio
);
1736 * If it's a regular read/write or a barrier with data attached,
1737 * go through the normal accounting stuff before submission.
1739 if (bio_has_data(bio
) && !(rw
& REQ_DISCARD
)) {
1741 count_vm_events(PGPGOUT
, count
);
1743 task_io_account_read(bio
->bi_size
);
1744 count_vm_events(PGPGIN
, count
);
1747 if (unlikely(block_dump
)) {
1748 char b
[BDEVNAME_SIZE
];
1749 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s (%u sectors)\n",
1750 current
->comm
, task_pid_nr(current
),
1751 (rw
& WRITE
) ? "WRITE" : "READ",
1752 (unsigned long long)bio
->bi_sector
,
1753 bdevname(bio
->bi_bdev
, b
),
1758 generic_make_request(bio
);
1760 EXPORT_SYMBOL(submit_bio
);
1763 * blk_rq_check_limits - Helper function to check a request for the queue limit
1765 * @rq: the request being checked
1768 * @rq may have been made based on weaker limitations of upper-level queues
1769 * in request stacking drivers, and it may violate the limitation of @q.
1770 * Since the block layer and the underlying device driver trust @rq
1771 * after it is inserted to @q, it should be checked against @q before
1772 * the insertion using this generic function.
1774 * This function should also be useful for request stacking drivers
1775 * in some cases below, so export this function.
1776 * Request stacking drivers like request-based dm may change the queue
1777 * limits while requests are in the queue (e.g. dm's table swapping).
1778 * Such request stacking drivers should check those requests agaist
1779 * the new queue limits again when they dispatch those requests,
1780 * although such checkings are also done against the old queue limits
1781 * when submitting requests.
1783 int blk_rq_check_limits(struct request_queue
*q
, struct request
*rq
)
1785 if (rq
->cmd_flags
& REQ_DISCARD
)
1788 if (blk_rq_sectors(rq
) > queue_max_sectors(q
) ||
1789 blk_rq_bytes(rq
) > queue_max_hw_sectors(q
) << 9) {
1790 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
1795 * queue's settings related to segment counting like q->bounce_pfn
1796 * may differ from that of other stacking queues.
1797 * Recalculate it to check the request correctly on this queue's
1800 blk_recalc_rq_segments(rq
);
1801 if (rq
->nr_phys_segments
> queue_max_segments(q
)) {
1802 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
1808 EXPORT_SYMBOL_GPL(blk_rq_check_limits
);
1811 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1812 * @q: the queue to submit the request
1813 * @rq: the request being queued
1815 int blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
1817 unsigned long flags
;
1818 int where
= ELEVATOR_INSERT_BACK
;
1820 if (blk_rq_check_limits(q
, rq
))
1824 should_fail_request(&rq
->rq_disk
->part0
, blk_rq_bytes(rq
)))
1827 spin_lock_irqsave(q
->queue_lock
, flags
);
1828 if (unlikely(blk_queue_dead(q
))) {
1829 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1834 * Submitting request must be dequeued before calling this function
1835 * because it will be linked to another request_queue
1837 BUG_ON(blk_queued_rq(rq
));
1839 if (rq
->cmd_flags
& (REQ_FLUSH
|REQ_FUA
))
1840 where
= ELEVATOR_INSERT_FLUSH
;
1842 add_acct_request(q
, rq
, where
);
1843 if (where
== ELEVATOR_INSERT_FLUSH
)
1845 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1849 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
1852 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1853 * @rq: request to examine
1856 * A request could be merge of IOs which require different failure
1857 * handling. This function determines the number of bytes which
1858 * can be failed from the beginning of the request without
1859 * crossing into area which need to be retried further.
1862 * The number of bytes to fail.
1865 * queue_lock must be held.
1867 unsigned int blk_rq_err_bytes(const struct request
*rq
)
1869 unsigned int ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
1870 unsigned int bytes
= 0;
1873 if (!(rq
->cmd_flags
& REQ_MIXED_MERGE
))
1874 return blk_rq_bytes(rq
);
1877 * Currently the only 'mixing' which can happen is between
1878 * different fastfail types. We can safely fail portions
1879 * which have all the failfast bits that the first one has -
1880 * the ones which are at least as eager to fail as the first
1883 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
1884 if ((bio
->bi_rw
& ff
) != ff
)
1886 bytes
+= bio
->bi_size
;
1889 /* this could lead to infinite loop */
1890 BUG_ON(blk_rq_bytes(rq
) && !bytes
);
1893 EXPORT_SYMBOL_GPL(blk_rq_err_bytes
);
1895 static void blk_account_io_completion(struct request
*req
, unsigned int bytes
)
1897 if (blk_do_io_stat(req
)) {
1898 const int rw
= rq_data_dir(req
);
1899 struct hd_struct
*part
;
1902 cpu
= part_stat_lock();
1904 part_stat_add(cpu
, part
, sectors
[rw
], bytes
>> 9);
1909 static void blk_account_io_done(struct request
*req
)
1912 * Account IO completion. flush_rq isn't accounted as a
1913 * normal IO on queueing nor completion. Accounting the
1914 * containing request is enough.
1916 if (blk_do_io_stat(req
) && !(req
->cmd_flags
& REQ_FLUSH_SEQ
)) {
1917 unsigned long duration
= jiffies
- req
->start_time
;
1918 const int rw
= rq_data_dir(req
);
1919 struct hd_struct
*part
;
1922 cpu
= part_stat_lock();
1925 part_stat_inc(cpu
, part
, ios
[rw
]);
1926 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
1927 part_round_stats(cpu
, part
);
1928 part_dec_in_flight(part
, rw
);
1930 hd_struct_put(part
);
1936 * blk_peek_request - peek at the top of a request queue
1937 * @q: request queue to peek at
1940 * Return the request at the top of @q. The returned request
1941 * should be started using blk_start_request() before LLD starts
1945 * Pointer to the request at the top of @q if available. Null
1949 * queue_lock must be held.
1951 struct request
*blk_peek_request(struct request_queue
*q
)
1956 while ((rq
= __elv_next_request(q
)) != NULL
) {
1957 if (!(rq
->cmd_flags
& REQ_STARTED
)) {
1959 * This is the first time the device driver
1960 * sees this request (possibly after
1961 * requeueing). Notify IO scheduler.
1963 if (rq
->cmd_flags
& REQ_SORTED
)
1964 elv_activate_rq(q
, rq
);
1967 * just mark as started even if we don't start
1968 * it, a request that has been delayed should
1969 * not be passed by new incoming requests
1971 rq
->cmd_flags
|= REQ_STARTED
;
1972 trace_block_rq_issue(q
, rq
);
1975 if (!q
->boundary_rq
|| q
->boundary_rq
== rq
) {
1976 q
->end_sector
= rq_end_sector(rq
);
1977 q
->boundary_rq
= NULL
;
1980 if (rq
->cmd_flags
& REQ_DONTPREP
)
1983 if (q
->dma_drain_size
&& blk_rq_bytes(rq
)) {
1985 * make sure space for the drain appears we
1986 * know we can do this because max_hw_segments
1987 * has been adjusted to be one fewer than the
1990 rq
->nr_phys_segments
++;
1996 ret
= q
->prep_rq_fn(q
, rq
);
1997 if (ret
== BLKPREP_OK
) {
1999 } else if (ret
== BLKPREP_DEFER
) {
2001 * the request may have been (partially) prepped.
2002 * we need to keep this request in the front to
2003 * avoid resource deadlock. REQ_STARTED will
2004 * prevent other fs requests from passing this one.
2006 if (q
->dma_drain_size
&& blk_rq_bytes(rq
) &&
2007 !(rq
->cmd_flags
& REQ_DONTPREP
)) {
2009 * remove the space for the drain we added
2010 * so that we don't add it again
2012 --rq
->nr_phys_segments
;
2017 } else if (ret
== BLKPREP_KILL
) {
2018 rq
->cmd_flags
|= REQ_QUIET
;
2020 * Mark this request as started so we don't trigger
2021 * any debug logic in the end I/O path.
2023 blk_start_request(rq
);
2024 __blk_end_request_all(rq
, -EIO
);
2026 printk(KERN_ERR
"%s: bad return=%d\n", __func__
, ret
);
2033 EXPORT_SYMBOL(blk_peek_request
);
2035 void blk_dequeue_request(struct request
*rq
)
2037 struct request_queue
*q
= rq
->q
;
2039 BUG_ON(list_empty(&rq
->queuelist
));
2040 BUG_ON(ELV_ON_HASH(rq
));
2042 list_del_init(&rq
->queuelist
);
2045 * the time frame between a request being removed from the lists
2046 * and to it is freed is accounted as io that is in progress at
2049 if (blk_account_rq(rq
)) {
2050 q
->in_flight
[rq_is_sync(rq
)]++;
2051 set_io_start_time_ns(rq
);
2056 * blk_start_request - start request processing on the driver
2057 * @req: request to dequeue
2060 * Dequeue @req and start timeout timer on it. This hands off the
2061 * request to the driver.
2063 * Block internal functions which don't want to start timer should
2064 * call blk_dequeue_request().
2067 * queue_lock must be held.
2069 void blk_start_request(struct request
*req
)
2071 blk_dequeue_request(req
);
2074 * We are now handing the request to the hardware, initialize
2075 * resid_len to full count and add the timeout handler.
2077 req
->resid_len
= blk_rq_bytes(req
);
2078 if (unlikely(blk_bidi_rq(req
)))
2079 req
->next_rq
->resid_len
= blk_rq_bytes(req
->next_rq
);
2083 EXPORT_SYMBOL(blk_start_request
);
2086 * blk_fetch_request - fetch a request from a request queue
2087 * @q: request queue to fetch a request from
2090 * Return the request at the top of @q. The request is started on
2091 * return and LLD can start processing it immediately.
2094 * Pointer to the request at the top of @q if available. Null
2098 * queue_lock must be held.
2100 struct request
*blk_fetch_request(struct request_queue
*q
)
2104 rq
= blk_peek_request(q
);
2106 blk_start_request(rq
);
2109 EXPORT_SYMBOL(blk_fetch_request
);
2112 * blk_update_request - Special helper function for request stacking drivers
2113 * @req: the request being processed
2114 * @error: %0 for success, < %0 for error
2115 * @nr_bytes: number of bytes to complete @req
2118 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2119 * the request structure even if @req doesn't have leftover.
2120 * If @req has leftover, sets it up for the next range of segments.
2122 * This special helper function is only for request stacking drivers
2123 * (e.g. request-based dm) so that they can handle partial completion.
2124 * Actual device drivers should use blk_end_request instead.
2126 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2127 * %false return from this function.
2130 * %false - this request doesn't have any more data
2131 * %true - this request has more data
2133 bool blk_update_request(struct request
*req
, int error
, unsigned int nr_bytes
)
2135 int total_bytes
, bio_nbytes
, next_idx
= 0;
2141 trace_block_rq_complete(req
->q
, req
);
2144 * For fs requests, rq is just carrier of independent bio's
2145 * and each partial completion should be handled separately.
2146 * Reset per-request error on each partial completion.
2148 * TODO: tj: This is too subtle. It would be better to let
2149 * low level drivers do what they see fit.
2151 if (req
->cmd_type
== REQ_TYPE_FS
)
2154 if (error
&& req
->cmd_type
== REQ_TYPE_FS
&&
2155 !(req
->cmd_flags
& REQ_QUIET
)) {
2160 error_type
= "recoverable transport";
2163 error_type
= "critical target";
2166 error_type
= "critical nexus";
2173 printk(KERN_ERR
"end_request: %s error, dev %s, sector %llu\n",
2174 error_type
, req
->rq_disk
? req
->rq_disk
->disk_name
: "?",
2175 (unsigned long long)blk_rq_pos(req
));
2178 blk_account_io_completion(req
, nr_bytes
);
2180 total_bytes
= bio_nbytes
= 0;
2181 while ((bio
= req
->bio
) != NULL
) {
2184 if (nr_bytes
>= bio
->bi_size
) {
2185 req
->bio
= bio
->bi_next
;
2186 nbytes
= bio
->bi_size
;
2187 req_bio_endio(req
, bio
, nbytes
, error
);
2191 int idx
= bio
->bi_idx
+ next_idx
;
2193 if (unlikely(idx
>= bio
->bi_vcnt
)) {
2194 blk_dump_rq_flags(req
, "__end_that");
2195 printk(KERN_ERR
"%s: bio idx %d >= vcnt %d\n",
2196 __func__
, idx
, bio
->bi_vcnt
);
2200 nbytes
= bio_iovec_idx(bio
, idx
)->bv_len
;
2201 BIO_BUG_ON(nbytes
> bio
->bi_size
);
2204 * not a complete bvec done
2206 if (unlikely(nbytes
> nr_bytes
)) {
2207 bio_nbytes
+= nr_bytes
;
2208 total_bytes
+= nr_bytes
;
2213 * advance to the next vector
2216 bio_nbytes
+= nbytes
;
2219 total_bytes
+= nbytes
;
2225 * end more in this run, or just return 'not-done'
2227 if (unlikely(nr_bytes
<= 0))
2237 * Reset counters so that the request stacking driver
2238 * can find how many bytes remain in the request
2241 req
->__data_len
= 0;
2246 * if the request wasn't completed, update state
2249 req_bio_endio(req
, bio
, bio_nbytes
, error
);
2250 bio
->bi_idx
+= next_idx
;
2251 bio_iovec(bio
)->bv_offset
+= nr_bytes
;
2252 bio_iovec(bio
)->bv_len
-= nr_bytes
;
2255 req
->__data_len
-= total_bytes
;
2256 req
->buffer
= bio_data(req
->bio
);
2258 /* update sector only for requests with clear definition of sector */
2259 if (req
->cmd_type
== REQ_TYPE_FS
|| (req
->cmd_flags
& REQ_DISCARD
))
2260 req
->__sector
+= total_bytes
>> 9;
2262 /* mixed attributes always follow the first bio */
2263 if (req
->cmd_flags
& REQ_MIXED_MERGE
) {
2264 req
->cmd_flags
&= ~REQ_FAILFAST_MASK
;
2265 req
->cmd_flags
|= req
->bio
->bi_rw
& REQ_FAILFAST_MASK
;
2269 * If total number of sectors is less than the first segment
2270 * size, something has gone terribly wrong.
2272 if (blk_rq_bytes(req
) < blk_rq_cur_bytes(req
)) {
2273 blk_dump_rq_flags(req
, "request botched");
2274 req
->__data_len
= blk_rq_cur_bytes(req
);
2277 /* recalculate the number of segments */
2278 blk_recalc_rq_segments(req
);
2282 EXPORT_SYMBOL_GPL(blk_update_request
);
2284 static bool blk_update_bidi_request(struct request
*rq
, int error
,
2285 unsigned int nr_bytes
,
2286 unsigned int bidi_bytes
)
2288 if (blk_update_request(rq
, error
, nr_bytes
))
2291 /* Bidi request must be completed as a whole */
2292 if (unlikely(blk_bidi_rq(rq
)) &&
2293 blk_update_request(rq
->next_rq
, error
, bidi_bytes
))
2296 if (blk_queue_add_random(rq
->q
))
2297 add_disk_randomness(rq
->rq_disk
);
2303 * blk_unprep_request - unprepare a request
2306 * This function makes a request ready for complete resubmission (or
2307 * completion). It happens only after all error handling is complete,
2308 * so represents the appropriate moment to deallocate any resources
2309 * that were allocated to the request in the prep_rq_fn. The queue
2310 * lock is held when calling this.
2312 void blk_unprep_request(struct request
*req
)
2314 struct request_queue
*q
= req
->q
;
2316 req
->cmd_flags
&= ~REQ_DONTPREP
;
2317 if (q
->unprep_rq_fn
)
2318 q
->unprep_rq_fn(q
, req
);
2320 EXPORT_SYMBOL_GPL(blk_unprep_request
);
2323 * queue lock must be held
2325 static void blk_finish_request(struct request
*req
, int error
)
2327 if (blk_rq_tagged(req
))
2328 blk_queue_end_tag(req
->q
, req
);
2330 BUG_ON(blk_queued_rq(req
));
2332 if (unlikely(laptop_mode
) && req
->cmd_type
== REQ_TYPE_FS
)
2333 laptop_io_completion(&req
->q
->backing_dev_info
);
2335 blk_delete_timer(req
);
2337 if (req
->cmd_flags
& REQ_DONTPREP
)
2338 blk_unprep_request(req
);
2341 blk_account_io_done(req
);
2344 req
->end_io(req
, error
);
2346 if (blk_bidi_rq(req
))
2347 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
2349 __blk_put_request(req
->q
, req
);
2354 * blk_end_bidi_request - Complete a bidi request
2355 * @rq: the request to complete
2356 * @error: %0 for success, < %0 for error
2357 * @nr_bytes: number of bytes to complete @rq
2358 * @bidi_bytes: number of bytes to complete @rq->next_rq
2361 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2362 * Drivers that supports bidi can safely call this member for any
2363 * type of request, bidi or uni. In the later case @bidi_bytes is
2367 * %false - we are done with this request
2368 * %true - still buffers pending for this request
2370 static bool blk_end_bidi_request(struct request
*rq
, int error
,
2371 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2373 struct request_queue
*q
= rq
->q
;
2374 unsigned long flags
;
2376 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2379 spin_lock_irqsave(q
->queue_lock
, flags
);
2380 blk_finish_request(rq
, error
);
2381 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2387 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2388 * @rq: the request to complete
2389 * @error: %0 for success, < %0 for error
2390 * @nr_bytes: number of bytes to complete @rq
2391 * @bidi_bytes: number of bytes to complete @rq->next_rq
2394 * Identical to blk_end_bidi_request() except that queue lock is
2395 * assumed to be locked on entry and remains so on return.
2398 * %false - we are done with this request
2399 * %true - still buffers pending for this request
2401 bool __blk_end_bidi_request(struct request
*rq
, int error
,
2402 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2404 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2407 blk_finish_request(rq
, error
);
2413 * blk_end_request - Helper function for drivers to complete the request.
2414 * @rq: the request being processed
2415 * @error: %0 for success, < %0 for error
2416 * @nr_bytes: number of bytes to complete
2419 * Ends I/O on a number of bytes attached to @rq.
2420 * If @rq has leftover, sets it up for the next range of segments.
2423 * %false - we are done with this request
2424 * %true - still buffers pending for this request
2426 bool blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2428 return blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2430 EXPORT_SYMBOL(blk_end_request
);
2433 * blk_end_request_all - Helper function for drives to finish the request.
2434 * @rq: the request to finish
2435 * @error: %0 for success, < %0 for error
2438 * Completely finish @rq.
2440 void blk_end_request_all(struct request
*rq
, int error
)
2443 unsigned int bidi_bytes
= 0;
2445 if (unlikely(blk_bidi_rq(rq
)))
2446 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2448 pending
= blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2451 EXPORT_SYMBOL(blk_end_request_all
);
2454 * blk_end_request_cur - Helper function to finish the current request chunk.
2455 * @rq: the request to finish the current chunk for
2456 * @error: %0 for success, < %0 for error
2459 * Complete the current consecutively mapped chunk from @rq.
2462 * %false - we are done with this request
2463 * %true - still buffers pending for this request
2465 bool blk_end_request_cur(struct request
*rq
, int error
)
2467 return blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2469 EXPORT_SYMBOL(blk_end_request_cur
);
2472 * blk_end_request_err - Finish a request till the next failure boundary.
2473 * @rq: the request to finish till the next failure boundary for
2474 * @error: must be negative errno
2477 * Complete @rq till the next failure boundary.
2480 * %false - we are done with this request
2481 * %true - still buffers pending for this request
2483 bool blk_end_request_err(struct request
*rq
, int error
)
2485 WARN_ON(error
>= 0);
2486 return blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2488 EXPORT_SYMBOL_GPL(blk_end_request_err
);
2491 * __blk_end_request - Helper function for drivers to complete the request.
2492 * @rq: the request being processed
2493 * @error: %0 for success, < %0 for error
2494 * @nr_bytes: number of bytes to complete
2497 * Must be called with queue lock held unlike blk_end_request().
2500 * %false - we are done with this request
2501 * %true - still buffers pending for this request
2503 bool __blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2505 return __blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2507 EXPORT_SYMBOL(__blk_end_request
);
2510 * __blk_end_request_all - Helper function for drives to finish the request.
2511 * @rq: the request to finish
2512 * @error: %0 for success, < %0 for error
2515 * Completely finish @rq. Must be called with queue lock held.
2517 void __blk_end_request_all(struct request
*rq
, int error
)
2520 unsigned int bidi_bytes
= 0;
2522 if (unlikely(blk_bidi_rq(rq
)))
2523 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2525 pending
= __blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2528 EXPORT_SYMBOL(__blk_end_request_all
);
2531 * __blk_end_request_cur - Helper function to finish the current request chunk.
2532 * @rq: the request to finish the current chunk for
2533 * @error: %0 for success, < %0 for error
2536 * Complete the current consecutively mapped chunk from @rq. Must
2537 * be called with queue lock held.
2540 * %false - we are done with this request
2541 * %true - still buffers pending for this request
2543 bool __blk_end_request_cur(struct request
*rq
, int error
)
2545 return __blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2547 EXPORT_SYMBOL(__blk_end_request_cur
);
2550 * __blk_end_request_err - Finish a request till the next failure boundary.
2551 * @rq: the request to finish till the next failure boundary for
2552 * @error: must be negative errno
2555 * Complete @rq till the next failure boundary. Must be called
2556 * with queue lock held.
2559 * %false - we are done with this request
2560 * %true - still buffers pending for this request
2562 bool __blk_end_request_err(struct request
*rq
, int error
)
2564 WARN_ON(error
>= 0);
2565 return __blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2567 EXPORT_SYMBOL_GPL(__blk_end_request_err
);
2569 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
2572 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2573 rq
->cmd_flags
|= bio
->bi_rw
& REQ_WRITE
;
2575 if (bio_has_data(bio
)) {
2576 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2577 rq
->buffer
= bio_data(bio
);
2579 rq
->__data_len
= bio
->bi_size
;
2580 rq
->bio
= rq
->biotail
= bio
;
2583 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2586 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2588 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2589 * @rq: the request to be flushed
2592 * Flush all pages in @rq.
2594 void rq_flush_dcache_pages(struct request
*rq
)
2596 struct req_iterator iter
;
2597 struct bio_vec
*bvec
;
2599 rq_for_each_segment(bvec
, rq
, iter
)
2600 flush_dcache_page(bvec
->bv_page
);
2602 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages
);
2606 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2607 * @q : the queue of the device being checked
2610 * Check if underlying low-level drivers of a device are busy.
2611 * If the drivers want to export their busy state, they must set own
2612 * exporting function using blk_queue_lld_busy() first.
2614 * Basically, this function is used only by request stacking drivers
2615 * to stop dispatching requests to underlying devices when underlying
2616 * devices are busy. This behavior helps more I/O merging on the queue
2617 * of the request stacking driver and prevents I/O throughput regression
2618 * on burst I/O load.
2621 * 0 - Not busy (The request stacking driver should dispatch request)
2622 * 1 - Busy (The request stacking driver should stop dispatching request)
2624 int blk_lld_busy(struct request_queue
*q
)
2627 return q
->lld_busy_fn(q
);
2631 EXPORT_SYMBOL_GPL(blk_lld_busy
);
2634 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2635 * @rq: the clone request to be cleaned up
2638 * Free all bios in @rq for a cloned request.
2640 void blk_rq_unprep_clone(struct request
*rq
)
2644 while ((bio
= rq
->bio
) != NULL
) {
2645 rq
->bio
= bio
->bi_next
;
2650 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone
);
2653 * Copy attributes of the original request to the clone request.
2654 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2656 static void __blk_rq_prep_clone(struct request
*dst
, struct request
*src
)
2658 dst
->cpu
= src
->cpu
;
2659 dst
->cmd_flags
= (src
->cmd_flags
& REQ_CLONE_MASK
) | REQ_NOMERGE
;
2660 dst
->cmd_type
= src
->cmd_type
;
2661 dst
->__sector
= blk_rq_pos(src
);
2662 dst
->__data_len
= blk_rq_bytes(src
);
2663 dst
->nr_phys_segments
= src
->nr_phys_segments
;
2664 dst
->ioprio
= src
->ioprio
;
2665 dst
->extra_len
= src
->extra_len
;
2669 * blk_rq_prep_clone - Helper function to setup clone request
2670 * @rq: the request to be setup
2671 * @rq_src: original request to be cloned
2672 * @bs: bio_set that bios for clone are allocated from
2673 * @gfp_mask: memory allocation mask for bio
2674 * @bio_ctr: setup function to be called for each clone bio.
2675 * Returns %0 for success, non %0 for failure.
2676 * @data: private data to be passed to @bio_ctr
2679 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2680 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2681 * are not copied, and copying such parts is the caller's responsibility.
2682 * Also, pages which the original bios are pointing to are not copied
2683 * and the cloned bios just point same pages.
2684 * So cloned bios must be completed before original bios, which means
2685 * the caller must complete @rq before @rq_src.
2687 int blk_rq_prep_clone(struct request
*rq
, struct request
*rq_src
,
2688 struct bio_set
*bs
, gfp_t gfp_mask
,
2689 int (*bio_ctr
)(struct bio
*, struct bio
*, void *),
2692 struct bio
*bio
, *bio_src
;
2697 blk_rq_init(NULL
, rq
);
2699 __rq_for_each_bio(bio_src
, rq_src
) {
2700 bio
= bio_alloc_bioset(gfp_mask
, bio_src
->bi_max_vecs
, bs
);
2704 __bio_clone(bio
, bio_src
);
2706 if (bio_integrity(bio_src
) &&
2707 bio_integrity_clone(bio
, bio_src
, gfp_mask
, bs
))
2710 if (bio_ctr
&& bio_ctr(bio
, bio_src
, data
))
2714 rq
->biotail
->bi_next
= bio
;
2717 rq
->bio
= rq
->biotail
= bio
;
2720 __blk_rq_prep_clone(rq
, rq_src
);
2727 blk_rq_unprep_clone(rq
);
2731 EXPORT_SYMBOL_GPL(blk_rq_prep_clone
);
2733 int kblockd_schedule_work(struct request_queue
*q
, struct work_struct
*work
)
2735 return queue_work(kblockd_workqueue
, work
);
2737 EXPORT_SYMBOL(kblockd_schedule_work
);
2739 int kblockd_schedule_delayed_work(struct request_queue
*q
,
2740 struct delayed_work
*dwork
, unsigned long delay
)
2742 return queue_delayed_work(kblockd_workqueue
, dwork
, delay
);
2744 EXPORT_SYMBOL(kblockd_schedule_delayed_work
);
2746 #define PLUG_MAGIC 0x91827364
2749 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2750 * @plug: The &struct blk_plug that needs to be initialized
2753 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2754 * pending I/O should the task end up blocking between blk_start_plug() and
2755 * blk_finish_plug(). This is important from a performance perspective, but
2756 * also ensures that we don't deadlock. For instance, if the task is blocking
2757 * for a memory allocation, memory reclaim could end up wanting to free a
2758 * page belonging to that request that is currently residing in our private
2759 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2760 * this kind of deadlock.
2762 void blk_start_plug(struct blk_plug
*plug
)
2764 struct task_struct
*tsk
= current
;
2766 plug
->magic
= PLUG_MAGIC
;
2767 INIT_LIST_HEAD(&plug
->list
);
2768 INIT_LIST_HEAD(&plug
->cb_list
);
2769 plug
->should_sort
= 0;
2772 * If this is a nested plug, don't actually assign it. It will be
2773 * flushed on its own.
2777 * Store ordering should not be needed here, since a potential
2778 * preempt will imply a full memory barrier
2783 EXPORT_SYMBOL(blk_start_plug
);
2785 static int plug_rq_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
2787 struct request
*rqa
= container_of(a
, struct request
, queuelist
);
2788 struct request
*rqb
= container_of(b
, struct request
, queuelist
);
2790 return !(rqa
->q
<= rqb
->q
);
2794 * If 'from_schedule' is true, then postpone the dispatch of requests
2795 * until a safe kblockd context. We due this to avoid accidental big
2796 * additional stack usage in driver dispatch, in places where the originally
2797 * plugger did not intend it.
2799 static void queue_unplugged(struct request_queue
*q
, unsigned int depth
,
2801 __releases(q
->queue_lock
)
2803 trace_block_unplug(q
, depth
, !from_schedule
);
2806 * Don't mess with dead queue.
2808 if (unlikely(blk_queue_dead(q
))) {
2809 spin_unlock(q
->queue_lock
);
2814 * If we are punting this to kblockd, then we can safely drop
2815 * the queue_lock before waking kblockd (which needs to take
2818 if (from_schedule
) {
2819 spin_unlock(q
->queue_lock
);
2820 blk_run_queue_async(q
);
2823 spin_unlock(q
->queue_lock
);
2828 static void flush_plug_callbacks(struct blk_plug
*plug
)
2830 LIST_HEAD(callbacks
);
2832 if (list_empty(&plug
->cb_list
))
2835 list_splice_init(&plug
->cb_list
, &callbacks
);
2837 while (!list_empty(&callbacks
)) {
2838 struct blk_plug_cb
*cb
= list_first_entry(&callbacks
,
2841 list_del(&cb
->list
);
2846 void blk_flush_plug_list(struct blk_plug
*plug
, bool from_schedule
)
2848 struct request_queue
*q
;
2849 unsigned long flags
;
2854 BUG_ON(plug
->magic
!= PLUG_MAGIC
);
2856 flush_plug_callbacks(plug
);
2857 if (list_empty(&plug
->list
))
2860 list_splice_init(&plug
->list
, &list
);
2862 if (plug
->should_sort
) {
2863 list_sort(NULL
, &list
, plug_rq_cmp
);
2864 plug
->should_sort
= 0;
2871 * Save and disable interrupts here, to avoid doing it for every
2872 * queue lock we have to take.
2874 local_irq_save(flags
);
2875 while (!list_empty(&list
)) {
2876 rq
= list_entry_rq(list
.next
);
2877 list_del_init(&rq
->queuelist
);
2881 * This drops the queue lock
2884 queue_unplugged(q
, depth
, from_schedule
);
2887 spin_lock(q
->queue_lock
);
2891 * Short-circuit if @q is dead
2893 if (unlikely(blk_queue_dead(q
))) {
2894 __blk_end_request_all(rq
, -ENODEV
);
2899 * rq is already accounted, so use raw insert
2901 if (rq
->cmd_flags
& (REQ_FLUSH
| REQ_FUA
))
2902 __elv_add_request(q
, rq
, ELEVATOR_INSERT_FLUSH
);
2904 __elv_add_request(q
, rq
, ELEVATOR_INSERT_SORT_MERGE
);
2910 * This drops the queue lock
2913 queue_unplugged(q
, depth
, from_schedule
);
2915 local_irq_restore(flags
);
2918 void blk_finish_plug(struct blk_plug
*plug
)
2920 blk_flush_plug_list(plug
, false);
2922 if (plug
== current
->plug
)
2923 current
->plug
= NULL
;
2925 EXPORT_SYMBOL(blk_finish_plug
);
2927 int __init
blk_dev_init(void)
2929 BUILD_BUG_ON(__REQ_NR_BITS
> 8 *
2930 sizeof(((struct request
*)0)->cmd_flags
));
2932 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
2933 kblockd_workqueue
= alloc_workqueue("kblockd",
2934 WQ_MEM_RECLAIM
| WQ_HIGHPRI
, 0);
2935 if (!kblockd_workqueue
)
2936 panic("Failed to create kblockd\n");
2938 request_cachep
= kmem_cache_create("blkdev_requests",
2939 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
2941 blk_requestq_cachep
= kmem_cache_create("blkdev_queue",
2942 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);