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
);
368 elv_drain_elevator(q
);
374 drain
|= q
->rq
.elvpriv
;
377 * Unfortunately, requests are queued at and tracked from
378 * multiple places and there's no single counter which can
379 * be drained. Check all the queues and counters.
382 drain
|= !list_empty(&q
->queue_head
);
383 for (i
= 0; i
< 2; i
++) {
384 drain
|= q
->rq
.count
[i
];
385 drain
|= q
->in_flight
[i
];
386 drain
|= !list_empty(&q
->flush_queue
[i
]);
390 spin_unlock_irq(q
->queue_lock
);
399 * blk_cleanup_queue - shutdown a request queue
400 * @q: request queue to shutdown
402 * Mark @q DEAD, drain all pending requests, destroy and put it. All
403 * future requests will be failed immediately with -ENODEV.
405 void blk_cleanup_queue(struct request_queue
*q
)
407 spinlock_t
*lock
= q
->queue_lock
;
409 /* mark @q DEAD, no new request or merges will be allowed afterwards */
410 mutex_lock(&q
->sysfs_lock
);
411 queue_flag_set_unlocked(QUEUE_FLAG_DEAD
, q
);
414 queue_flag_set(QUEUE_FLAG_NOMERGES
, q
);
415 queue_flag_set(QUEUE_FLAG_NOXMERGES
, q
);
416 queue_flag_set(QUEUE_FLAG_DEAD
, q
);
418 if (q
->queue_lock
!= &q
->__queue_lock
)
419 q
->queue_lock
= &q
->__queue_lock
;
421 spin_unlock_irq(lock
);
422 mutex_unlock(&q
->sysfs_lock
);
425 * Drain all requests queued before DEAD marking. The caller might
426 * be trying to tear down @q before its elevator is initialized, in
427 * which case we don't want to call into draining.
430 blk_drain_queue(q
, true);
432 /* @q won't process any more request, flush async actions */
433 del_timer_sync(&q
->backing_dev_info
.laptop_mode_wb_timer
);
436 /* @q is and will stay empty, shutdown and put */
439 EXPORT_SYMBOL(blk_cleanup_queue
);
441 static int blk_init_free_list(struct request_queue
*q
)
443 struct request_list
*rl
= &q
->rq
;
445 if (unlikely(rl
->rq_pool
))
448 rl
->count
[BLK_RW_SYNC
] = rl
->count
[BLK_RW_ASYNC
] = 0;
449 rl
->starved
[BLK_RW_SYNC
] = rl
->starved
[BLK_RW_ASYNC
] = 0;
451 init_waitqueue_head(&rl
->wait
[BLK_RW_SYNC
]);
452 init_waitqueue_head(&rl
->wait
[BLK_RW_ASYNC
]);
454 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, mempool_alloc_slab
,
455 mempool_free_slab
, request_cachep
, q
->node
);
463 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
465 return blk_alloc_queue_node(gfp_mask
, -1);
467 EXPORT_SYMBOL(blk_alloc_queue
);
469 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
471 struct request_queue
*q
;
474 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
475 gfp_mask
| __GFP_ZERO
, node_id
);
479 q
->id
= ida_simple_get(&blk_queue_ida
, 0, 0, GFP_KERNEL
);
483 q
->backing_dev_info
.ra_pages
=
484 (VM_MAX_READAHEAD
* 1024) / PAGE_CACHE_SIZE
;
485 q
->backing_dev_info
.state
= 0;
486 q
->backing_dev_info
.capabilities
= BDI_CAP_MAP_COPY
;
487 q
->backing_dev_info
.name
= "block";
489 err
= bdi_init(&q
->backing_dev_info
);
493 if (blk_throtl_init(q
))
496 setup_timer(&q
->backing_dev_info
.laptop_mode_wb_timer
,
497 laptop_mode_timer_fn
, (unsigned long) q
);
498 setup_timer(&q
->timeout
, blk_rq_timed_out_timer
, (unsigned long) q
);
499 INIT_LIST_HEAD(&q
->timeout_list
);
500 INIT_LIST_HEAD(&q
->flush_queue
[0]);
501 INIT_LIST_HEAD(&q
->flush_queue
[1]);
502 INIT_LIST_HEAD(&q
->flush_data_in_flight
);
503 INIT_DELAYED_WORK(&q
->delay_work
, blk_delay_work
);
505 kobject_init(&q
->kobj
, &blk_queue_ktype
);
507 mutex_init(&q
->sysfs_lock
);
508 spin_lock_init(&q
->__queue_lock
);
511 * By default initialize queue_lock to internal lock and driver can
512 * override it later if need be.
514 q
->queue_lock
= &q
->__queue_lock
;
519 ida_simple_remove(&blk_queue_ida
, q
->id
);
521 kmem_cache_free(blk_requestq_cachep
, q
);
524 EXPORT_SYMBOL(blk_alloc_queue_node
);
527 * blk_init_queue - prepare a request queue for use with a block device
528 * @rfn: The function to be called to process requests that have been
529 * placed on the queue.
530 * @lock: Request queue spin lock
533 * If a block device wishes to use the standard request handling procedures,
534 * which sorts requests and coalesces adjacent requests, then it must
535 * call blk_init_queue(). The function @rfn will be called when there
536 * are requests on the queue that need to be processed. If the device
537 * supports plugging, then @rfn may not be called immediately when requests
538 * are available on the queue, but may be called at some time later instead.
539 * Plugged queues are generally unplugged when a buffer belonging to one
540 * of the requests on the queue is needed, or due to memory pressure.
542 * @rfn is not required, or even expected, to remove all requests off the
543 * queue, but only as many as it can handle at a time. If it does leave
544 * requests on the queue, it is responsible for arranging that the requests
545 * get dealt with eventually.
547 * The queue spin lock must be held while manipulating the requests on the
548 * request queue; this lock will be taken also from interrupt context, so irq
549 * disabling is needed for it.
551 * Function returns a pointer to the initialized request queue, or %NULL if
555 * blk_init_queue() must be paired with a blk_cleanup_queue() call
556 * when the block device is deactivated (such as at module unload).
559 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
561 return blk_init_queue_node(rfn
, lock
, -1);
563 EXPORT_SYMBOL(blk_init_queue
);
565 struct request_queue
*
566 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
568 struct request_queue
*uninit_q
, *q
;
570 uninit_q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
574 q
= blk_init_allocated_queue_node(uninit_q
, rfn
, lock
, node_id
);
576 blk_cleanup_queue(uninit_q
);
580 EXPORT_SYMBOL(blk_init_queue_node
);
582 struct request_queue
*
583 blk_init_allocated_queue(struct request_queue
*q
, request_fn_proc
*rfn
,
586 return blk_init_allocated_queue_node(q
, rfn
, lock
, -1);
588 EXPORT_SYMBOL(blk_init_allocated_queue
);
590 struct request_queue
*
591 blk_init_allocated_queue_node(struct request_queue
*q
, request_fn_proc
*rfn
,
592 spinlock_t
*lock
, int node_id
)
598 if (blk_init_free_list(q
))
602 q
->prep_rq_fn
= NULL
;
603 q
->unprep_rq_fn
= NULL
;
604 q
->queue_flags
= QUEUE_FLAG_DEFAULT
;
606 /* Override internal queue lock with supplied lock pointer */
608 q
->queue_lock
= lock
;
611 * This also sets hw/phys segments, boundary and size
613 blk_queue_make_request(q
, blk_queue_bio
);
615 q
->sg_reserved_size
= INT_MAX
;
620 if (!elevator_init(q
, NULL
)) {
621 blk_queue_congestion_threshold(q
);
627 EXPORT_SYMBOL(blk_init_allocated_queue_node
);
629 bool blk_get_queue(struct request_queue
*q
)
631 if (likely(!blk_queue_dead(q
))) {
638 EXPORT_SYMBOL(blk_get_queue
);
640 static inline void blk_free_request(struct request_queue
*q
, struct request
*rq
)
642 if (rq
->cmd_flags
& REQ_ELVPRIV
)
643 elv_put_request(q
, rq
);
644 mempool_free(rq
, q
->rq
.rq_pool
);
647 static struct request
*
648 blk_alloc_request(struct request_queue
*q
, unsigned int flags
, gfp_t gfp_mask
)
650 struct request
*rq
= mempool_alloc(q
->rq
.rq_pool
, gfp_mask
);
657 rq
->cmd_flags
= flags
| REQ_ALLOCED
;
659 if ((flags
& REQ_ELVPRIV
) &&
660 unlikely(elv_set_request(q
, rq
, gfp_mask
))) {
661 mempool_free(rq
, q
->rq
.rq_pool
);
669 * ioc_batching returns true if the ioc is a valid batching request and
670 * should be given priority access to a request.
672 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
678 * Make sure the process is able to allocate at least 1 request
679 * even if the batch times out, otherwise we could theoretically
682 return ioc
->nr_batch_requests
== q
->nr_batching
||
683 (ioc
->nr_batch_requests
> 0
684 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
688 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
689 * will cause the process to be a "batcher" on all queues in the system. This
690 * is the behaviour we want though - once it gets a wakeup it should be given
693 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
695 if (!ioc
|| ioc_batching(q
, ioc
))
698 ioc
->nr_batch_requests
= q
->nr_batching
;
699 ioc
->last_waited
= jiffies
;
702 static void __freed_request(struct request_queue
*q
, int sync
)
704 struct request_list
*rl
= &q
->rq
;
706 if (rl
->count
[sync
] < queue_congestion_off_threshold(q
))
707 blk_clear_queue_congested(q
, sync
);
709 if (rl
->count
[sync
] + 1 <= q
->nr_requests
) {
710 if (waitqueue_active(&rl
->wait
[sync
]))
711 wake_up(&rl
->wait
[sync
]);
713 blk_clear_queue_full(q
, sync
);
718 * A request has just been released. Account for it, update the full and
719 * congestion status, wake up any waiters. Called under q->queue_lock.
721 static void freed_request(struct request_queue
*q
, unsigned int flags
)
723 struct request_list
*rl
= &q
->rq
;
724 int sync
= rw_is_sync(flags
);
727 if (flags
& REQ_ELVPRIV
)
730 __freed_request(q
, sync
);
732 if (unlikely(rl
->starved
[sync
^ 1]))
733 __freed_request(q
, sync
^ 1);
737 * Determine if elevator data should be initialized when allocating the
738 * request associated with @bio.
740 static bool blk_rq_should_init_elevator(struct bio
*bio
)
746 * Flush requests do not use the elevator so skip initialization.
747 * This allows a request to share the flush and elevator data.
749 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
))
756 * get_request - get a free request
757 * @q: request_queue to allocate request from
758 * @rw_flags: RW and SYNC flags
759 * @bio: bio to allocate request for (can be %NULL)
760 * @gfp_mask: allocation mask
762 * Get a free request from @q. This function may fail under memory
763 * pressure or if @q is dead.
765 * Must be callled with @q->queue_lock held and,
766 * Returns %NULL on failure, with @q->queue_lock held.
767 * Returns !%NULL on success, with @q->queue_lock *not held*.
769 static struct request
*get_request(struct request_queue
*q
, int rw_flags
,
770 struct bio
*bio
, gfp_t gfp_mask
)
772 struct request
*rq
= NULL
;
773 struct request_list
*rl
= &q
->rq
;
774 struct io_context
*ioc
= NULL
;
775 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
778 if (unlikely(blk_queue_dead(q
)))
781 may_queue
= elv_may_queue(q
, rw_flags
);
782 if (may_queue
== ELV_MQUEUE_NO
)
785 if (rl
->count
[is_sync
]+1 >= queue_congestion_on_threshold(q
)) {
786 if (rl
->count
[is_sync
]+1 >= q
->nr_requests
) {
787 ioc
= current_io_context(GFP_ATOMIC
, q
->node
);
789 * The queue will fill after this allocation, so set
790 * it as full, and mark this process as "batching".
791 * This process will be allowed to complete a batch of
792 * requests, others will be blocked.
794 if (!blk_queue_full(q
, is_sync
)) {
795 ioc_set_batching(q
, ioc
);
796 blk_set_queue_full(q
, is_sync
);
798 if (may_queue
!= ELV_MQUEUE_MUST
799 && !ioc_batching(q
, ioc
)) {
801 * The queue is full and the allocating
802 * process is not a "batcher", and not
803 * exempted by the IO scheduler
809 blk_set_queue_congested(q
, is_sync
);
813 * Only allow batching queuers to allocate up to 50% over the defined
814 * limit of requests, otherwise we could have thousands of requests
815 * allocated with any setting of ->nr_requests
817 if (rl
->count
[is_sync
] >= (3 * q
->nr_requests
/ 2))
820 rl
->count
[is_sync
]++;
821 rl
->starved
[is_sync
] = 0;
823 if (blk_rq_should_init_elevator(bio
) &&
824 !test_bit(QUEUE_FLAG_ELVSWITCH
, &q
->queue_flags
)) {
825 rw_flags
|= REQ_ELVPRIV
;
829 if (blk_queue_io_stat(q
))
830 rw_flags
|= REQ_IO_STAT
;
831 spin_unlock_irq(q
->queue_lock
);
833 rq
= blk_alloc_request(q
, rw_flags
, gfp_mask
);
836 * Allocation failed presumably due to memory. Undo anything
837 * we might have messed up.
839 * Allocating task should really be put onto the front of the
840 * wait queue, but this is pretty rare.
842 spin_lock_irq(q
->queue_lock
);
843 freed_request(q
, rw_flags
);
846 * in the very unlikely event that allocation failed and no
847 * requests for this direction was pending, mark us starved
848 * so that freeing of a request in the other direction will
849 * notice us. another possible fix would be to split the
850 * rq mempool into READ and WRITE
853 if (unlikely(rl
->count
[is_sync
] == 0))
854 rl
->starved
[is_sync
] = 1;
860 * ioc may be NULL here, and ioc_batching will be false. That's
861 * OK, if the queue is under the request limit then requests need
862 * not count toward the nr_batch_requests limit. There will always
863 * be some limit enforced by BLK_BATCH_TIME.
865 if (ioc_batching(q
, ioc
))
866 ioc
->nr_batch_requests
--;
868 trace_block_getrq(q
, bio
, rw_flags
& 1);
874 * get_request_wait - get a free request with retry
875 * @q: request_queue to allocate request from
876 * @rw_flags: RW and SYNC flags
877 * @bio: bio to allocate request for (can be %NULL)
879 * Get a free request from @q. This function keeps retrying under memory
880 * pressure and fails iff @q is dead.
882 * Must be callled with @q->queue_lock held and,
883 * Returns %NULL on failure, with @q->queue_lock held.
884 * Returns !%NULL on success, with @q->queue_lock *not held*.
886 static struct request
*get_request_wait(struct request_queue
*q
, int rw_flags
,
889 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
892 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
895 struct io_context
*ioc
;
896 struct request_list
*rl
= &q
->rq
;
898 if (unlikely(blk_queue_dead(q
)))
901 prepare_to_wait_exclusive(&rl
->wait
[is_sync
], &wait
,
902 TASK_UNINTERRUPTIBLE
);
904 trace_block_sleeprq(q
, bio
, rw_flags
& 1);
906 spin_unlock_irq(q
->queue_lock
);
910 * After sleeping, we become a "batching" process and
911 * will be able to allocate at least one request, and
912 * up to a big batch of them for a small period time.
913 * See ioc_batching, ioc_set_batching
915 ioc
= current_io_context(GFP_NOIO
, q
->node
);
916 ioc_set_batching(q
, ioc
);
918 spin_lock_irq(q
->queue_lock
);
919 finish_wait(&rl
->wait
[is_sync
], &wait
);
921 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
927 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
931 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
933 spin_lock_irq(q
->queue_lock
);
934 if (gfp_mask
& __GFP_WAIT
)
935 rq
= get_request_wait(q
, rw
, NULL
);
937 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
939 spin_unlock_irq(q
->queue_lock
);
940 /* q->queue_lock is unlocked at this point */
944 EXPORT_SYMBOL(blk_get_request
);
947 * blk_make_request - given a bio, allocate a corresponding struct request.
948 * @q: target request queue
949 * @bio: The bio describing the memory mappings that will be submitted for IO.
950 * It may be a chained-bio properly constructed by block/bio layer.
951 * @gfp_mask: gfp flags to be used for memory allocation
953 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
954 * type commands. Where the struct request needs to be farther initialized by
955 * the caller. It is passed a &struct bio, which describes the memory info of
958 * The caller of blk_make_request must make sure that bi_io_vec
959 * are set to describe the memory buffers. That bio_data_dir() will return
960 * the needed direction of the request. (And all bio's in the passed bio-chain
961 * are properly set accordingly)
963 * If called under none-sleepable conditions, mapped bio buffers must not
964 * need bouncing, by calling the appropriate masked or flagged allocator,
965 * suitable for the target device. Otherwise the call to blk_queue_bounce will
968 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
969 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
970 * anything but the first bio in the chain. Otherwise you risk waiting for IO
971 * completion of a bio that hasn't been submitted yet, thus resulting in a
972 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
973 * of bio_alloc(), as that avoids the mempool deadlock.
974 * If possible a big IO should be split into smaller parts when allocation
975 * fails. Partial allocation should not be an error, or you risk a live-lock.
977 struct request
*blk_make_request(struct request_queue
*q
, struct bio
*bio
,
980 struct request
*rq
= blk_get_request(q
, bio_data_dir(bio
), gfp_mask
);
983 return ERR_PTR(-ENOMEM
);
986 struct bio
*bounce_bio
= bio
;
989 blk_queue_bounce(q
, &bounce_bio
);
990 ret
= blk_rq_append_bio(q
, rq
, bounce_bio
);
999 EXPORT_SYMBOL(blk_make_request
);
1002 * blk_requeue_request - put a request back on queue
1003 * @q: request queue where request should be inserted
1004 * @rq: request to be inserted
1007 * Drivers often keep queueing requests until the hardware cannot accept
1008 * more, when that condition happens we need to put the request back
1009 * on the queue. Must be called with queue lock held.
1011 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
1013 blk_delete_timer(rq
);
1014 blk_clear_rq_complete(rq
);
1015 trace_block_rq_requeue(q
, rq
);
1017 if (blk_rq_tagged(rq
))
1018 blk_queue_end_tag(q
, rq
);
1020 BUG_ON(blk_queued_rq(rq
));
1022 elv_requeue_request(q
, rq
);
1024 EXPORT_SYMBOL(blk_requeue_request
);
1026 static void add_acct_request(struct request_queue
*q
, struct request
*rq
,
1029 drive_stat_acct(rq
, 1);
1030 __elv_add_request(q
, rq
, where
);
1033 static void part_round_stats_single(int cpu
, struct hd_struct
*part
,
1036 if (now
== part
->stamp
)
1039 if (part_in_flight(part
)) {
1040 __part_stat_add(cpu
, part
, time_in_queue
,
1041 part_in_flight(part
) * (now
- part
->stamp
));
1042 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
1048 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1049 * @cpu: cpu number for stats access
1050 * @part: target partition
1052 * The average IO queue length and utilisation statistics are maintained
1053 * by observing the current state of the queue length and the amount of
1054 * time it has been in this state for.
1056 * Normally, that accounting is done on IO completion, but that can result
1057 * in more than a second's worth of IO being accounted for within any one
1058 * second, leading to >100% utilisation. To deal with that, we call this
1059 * function to do a round-off before returning the results when reading
1060 * /proc/diskstats. This accounts immediately for all queue usage up to
1061 * the current jiffies and restarts the counters again.
1063 void part_round_stats(int cpu
, struct hd_struct
*part
)
1065 unsigned long now
= jiffies
;
1068 part_round_stats_single(cpu
, &part_to_disk(part
)->part0
, now
);
1069 part_round_stats_single(cpu
, part
, now
);
1071 EXPORT_SYMBOL_GPL(part_round_stats
);
1074 * queue lock must be held
1076 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1080 if (unlikely(--req
->ref_count
))
1083 elv_completed_request(q
, req
);
1085 /* this is a bio leak */
1086 WARN_ON(req
->bio
!= NULL
);
1089 * Request may not have originated from ll_rw_blk. if not,
1090 * it didn't come out of our reserved rq pools
1092 if (req
->cmd_flags
& REQ_ALLOCED
) {
1093 unsigned int flags
= req
->cmd_flags
;
1095 BUG_ON(!list_empty(&req
->queuelist
));
1096 BUG_ON(!hlist_unhashed(&req
->hash
));
1098 blk_free_request(q
, req
);
1099 freed_request(q
, flags
);
1102 EXPORT_SYMBOL_GPL(__blk_put_request
);
1104 void blk_put_request(struct request
*req
)
1106 unsigned long flags
;
1107 struct request_queue
*q
= req
->q
;
1109 spin_lock_irqsave(q
->queue_lock
, flags
);
1110 __blk_put_request(q
, req
);
1111 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1113 EXPORT_SYMBOL(blk_put_request
);
1116 * blk_add_request_payload - add a payload to a request
1117 * @rq: request to update
1118 * @page: page backing the payload
1119 * @len: length of the payload.
1121 * This allows to later add a payload to an already submitted request by
1122 * a block driver. The driver needs to take care of freeing the payload
1125 * Note that this is a quite horrible hack and nothing but handling of
1126 * discard requests should ever use it.
1128 void blk_add_request_payload(struct request
*rq
, struct page
*page
,
1131 struct bio
*bio
= rq
->bio
;
1133 bio
->bi_io_vec
->bv_page
= page
;
1134 bio
->bi_io_vec
->bv_offset
= 0;
1135 bio
->bi_io_vec
->bv_len
= len
;
1139 bio
->bi_phys_segments
= 1;
1141 rq
->__data_len
= rq
->resid_len
= len
;
1142 rq
->nr_phys_segments
= 1;
1143 rq
->buffer
= bio_data(bio
);
1145 EXPORT_SYMBOL_GPL(blk_add_request_payload
);
1147 static bool bio_attempt_back_merge(struct request_queue
*q
, struct request
*req
,
1150 const int ff
= bio
->bi_rw
& REQ_FAILFAST_MASK
;
1152 if (!ll_back_merge_fn(q
, req
, bio
))
1155 trace_block_bio_backmerge(q
, bio
);
1157 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1158 blk_rq_set_mixed_merge(req
);
1160 req
->biotail
->bi_next
= bio
;
1162 req
->__data_len
+= bio
->bi_size
;
1163 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1165 drive_stat_acct(req
, 0);
1166 elv_bio_merged(q
, req
, bio
);
1170 static bool bio_attempt_front_merge(struct request_queue
*q
,
1171 struct request
*req
, struct bio
*bio
)
1173 const int ff
= bio
->bi_rw
& REQ_FAILFAST_MASK
;
1175 if (!ll_front_merge_fn(q
, req
, bio
))
1178 trace_block_bio_frontmerge(q
, bio
);
1180 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1181 blk_rq_set_mixed_merge(req
);
1183 bio
->bi_next
= req
->bio
;
1187 * may not be valid. if the low level driver said
1188 * it didn't need a bounce buffer then it better
1189 * not touch req->buffer either...
1191 req
->buffer
= bio_data(bio
);
1192 req
->__sector
= bio
->bi_sector
;
1193 req
->__data_len
+= bio
->bi_size
;
1194 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1196 drive_stat_acct(req
, 0);
1197 elv_bio_merged(q
, req
, bio
);
1202 * attempt_plug_merge - try to merge with %current's plugged list
1203 * @q: request_queue new bio is being queued at
1204 * @bio: new bio being queued
1205 * @request_count: out parameter for number of traversed plugged requests
1207 * Determine whether @bio being queued on @q can be merged with a request
1208 * on %current's plugged list. Returns %true if merge was successful,
1211 * This function is called without @q->queue_lock; however, elevator is
1212 * accessed iff there already are requests on the plugged list which in
1213 * turn guarantees validity of the elevator.
1215 * Note that, on successful merge, elevator operation
1216 * elevator_bio_merged_fn() will be called without queue lock. Elevator
1217 * must be ready for this.
1219 static bool attempt_plug_merge(struct request_queue
*q
, struct bio
*bio
,
1220 unsigned int *request_count
)
1222 struct blk_plug
*plug
;
1226 plug
= current
->plug
;
1231 list_for_each_entry_reverse(rq
, &plug
->list
, queuelist
) {
1239 el_ret
= elv_try_merge(rq
, bio
);
1240 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1241 ret
= bio_attempt_back_merge(q
, rq
, bio
);
1244 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1245 ret
= bio_attempt_front_merge(q
, rq
, bio
);
1254 void init_request_from_bio(struct request
*req
, struct bio
*bio
)
1256 req
->cmd_type
= REQ_TYPE_FS
;
1258 req
->cmd_flags
|= bio
->bi_rw
& REQ_COMMON_MASK
;
1259 if (bio
->bi_rw
& REQ_RAHEAD
)
1260 req
->cmd_flags
|= REQ_FAILFAST_MASK
;
1263 req
->__sector
= bio
->bi_sector
;
1264 req
->ioprio
= bio_prio(bio
);
1265 blk_rq_bio_prep(req
->q
, req
, bio
);
1268 void blk_queue_bio(struct request_queue
*q
, struct bio
*bio
)
1270 const bool sync
= !!(bio
->bi_rw
& REQ_SYNC
);
1271 struct blk_plug
*plug
;
1272 int el_ret
, rw_flags
, where
= ELEVATOR_INSERT_SORT
;
1273 struct request
*req
;
1274 unsigned int request_count
= 0;
1277 * low level driver can indicate that it wants pages above a
1278 * certain limit bounced to low memory (ie for highmem, or even
1279 * ISA dma in theory)
1281 blk_queue_bounce(q
, &bio
);
1283 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) {
1284 spin_lock_irq(q
->queue_lock
);
1285 where
= ELEVATOR_INSERT_FLUSH
;
1290 * Check if we can merge with the plugged list before grabbing
1293 if (attempt_plug_merge(q
, bio
, &request_count
))
1296 spin_lock_irq(q
->queue_lock
);
1298 el_ret
= elv_merge(q
, &req
, bio
);
1299 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1300 if (bio_attempt_back_merge(q
, req
, bio
)) {
1301 if (!attempt_back_merge(q
, req
))
1302 elv_merged_request(q
, req
, el_ret
);
1305 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1306 if (bio_attempt_front_merge(q
, req
, bio
)) {
1307 if (!attempt_front_merge(q
, req
))
1308 elv_merged_request(q
, req
, el_ret
);
1315 * This sync check and mask will be re-done in init_request_from_bio(),
1316 * but we need to set it earlier to expose the sync flag to the
1317 * rq allocator and io schedulers.
1319 rw_flags
= bio_data_dir(bio
);
1321 rw_flags
|= REQ_SYNC
;
1324 * Grab a free request. This is might sleep but can not fail.
1325 * Returns with the queue unlocked.
1327 req
= get_request_wait(q
, rw_flags
, bio
);
1328 if (unlikely(!req
)) {
1329 bio_endio(bio
, -ENODEV
); /* @q is dead */
1334 * After dropping the lock and possibly sleeping here, our request
1335 * may now be mergeable after it had proven unmergeable (above).
1336 * We don't worry about that case for efficiency. It won't happen
1337 * often, and the elevators are able to handle it.
1339 init_request_from_bio(req
, bio
);
1341 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
))
1342 req
->cpu
= raw_smp_processor_id();
1344 plug
= current
->plug
;
1347 * If this is the first request added after a plug, fire
1348 * of a plug trace. If others have been added before, check
1349 * if we have multiple devices in this plug. If so, make a
1350 * note to sort the list before dispatch.
1352 if (list_empty(&plug
->list
))
1353 trace_block_plug(q
);
1355 if (!plug
->should_sort
) {
1356 struct request
*__rq
;
1358 __rq
= list_entry_rq(plug
->list
.prev
);
1360 plug
->should_sort
= 1;
1362 if (request_count
>= BLK_MAX_REQUEST_COUNT
) {
1363 blk_flush_plug_list(plug
, false);
1364 trace_block_plug(q
);
1367 list_add_tail(&req
->queuelist
, &plug
->list
);
1368 drive_stat_acct(req
, 1);
1370 spin_lock_irq(q
->queue_lock
);
1371 add_acct_request(q
, req
, where
);
1374 spin_unlock_irq(q
->queue_lock
);
1377 EXPORT_SYMBOL_GPL(blk_queue_bio
); /* for device mapper only */
1380 * If bio->bi_dev is a partition, remap the location
1382 static inline void blk_partition_remap(struct bio
*bio
)
1384 struct block_device
*bdev
= bio
->bi_bdev
;
1386 if (bio_sectors(bio
) && bdev
!= bdev
->bd_contains
) {
1387 struct hd_struct
*p
= bdev
->bd_part
;
1389 bio
->bi_sector
+= p
->start_sect
;
1390 bio
->bi_bdev
= bdev
->bd_contains
;
1392 trace_block_bio_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
1394 bio
->bi_sector
- p
->start_sect
);
1398 static void handle_bad_sector(struct bio
*bio
)
1400 char b
[BDEVNAME_SIZE
];
1402 printk(KERN_INFO
"attempt to access beyond end of device\n");
1403 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
1404 bdevname(bio
->bi_bdev
, b
),
1406 (unsigned long long)bio
->bi_sector
+ bio_sectors(bio
),
1407 (long long)(i_size_read(bio
->bi_bdev
->bd_inode
) >> 9));
1409 set_bit(BIO_EOF
, &bio
->bi_flags
);
1412 #ifdef CONFIG_FAIL_MAKE_REQUEST
1414 static DECLARE_FAULT_ATTR(fail_make_request
);
1416 static int __init
setup_fail_make_request(char *str
)
1418 return setup_fault_attr(&fail_make_request
, str
);
1420 __setup("fail_make_request=", setup_fail_make_request
);
1422 static bool should_fail_request(struct hd_struct
*part
, unsigned int bytes
)
1424 return part
->make_it_fail
&& should_fail(&fail_make_request
, bytes
);
1427 static int __init
fail_make_request_debugfs(void)
1429 struct dentry
*dir
= fault_create_debugfs_attr("fail_make_request",
1430 NULL
, &fail_make_request
);
1432 return IS_ERR(dir
) ? PTR_ERR(dir
) : 0;
1435 late_initcall(fail_make_request_debugfs
);
1437 #else /* CONFIG_FAIL_MAKE_REQUEST */
1439 static inline bool should_fail_request(struct hd_struct
*part
,
1445 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1448 * Check whether this bio extends beyond the end of the device.
1450 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
1457 /* Test device or partition size, when known. */
1458 maxsector
= i_size_read(bio
->bi_bdev
->bd_inode
) >> 9;
1460 sector_t sector
= bio
->bi_sector
;
1462 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
1464 * This may well happen - the kernel calls bread()
1465 * without checking the size of the device, e.g., when
1466 * mounting a device.
1468 handle_bad_sector(bio
);
1476 static noinline_for_stack
bool
1477 generic_make_request_checks(struct bio
*bio
)
1479 struct request_queue
*q
;
1480 int nr_sectors
= bio_sectors(bio
);
1482 char b
[BDEVNAME_SIZE
];
1483 struct hd_struct
*part
;
1487 if (bio_check_eod(bio
, nr_sectors
))
1490 q
= bdev_get_queue(bio
->bi_bdev
);
1493 "generic_make_request: Trying to access "
1494 "nonexistent block-device %s (%Lu)\n",
1495 bdevname(bio
->bi_bdev
, b
),
1496 (long long) bio
->bi_sector
);
1500 if (unlikely(!(bio
->bi_rw
& REQ_DISCARD
) &&
1501 nr_sectors
> queue_max_hw_sectors(q
))) {
1502 printk(KERN_ERR
"bio too big device %s (%u > %u)\n",
1503 bdevname(bio
->bi_bdev
, b
),
1505 queue_max_hw_sectors(q
));
1509 part
= bio
->bi_bdev
->bd_part
;
1510 if (should_fail_request(part
, bio
->bi_size
) ||
1511 should_fail_request(&part_to_disk(part
)->part0
,
1516 * If this device has partitions, remap block n
1517 * of partition p to block n+start(p) of the disk.
1519 blk_partition_remap(bio
);
1521 if (bio_integrity_enabled(bio
) && bio_integrity_prep(bio
))
1524 if (bio_check_eod(bio
, nr_sectors
))
1528 * Filter flush bio's early so that make_request based
1529 * drivers without flush support don't have to worry
1532 if ((bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) && !q
->flush_flags
) {
1533 bio
->bi_rw
&= ~(REQ_FLUSH
| REQ_FUA
);
1540 if ((bio
->bi_rw
& REQ_DISCARD
) &&
1541 (!blk_queue_discard(q
) ||
1542 ((bio
->bi_rw
& REQ_SECURE
) &&
1543 !blk_queue_secdiscard(q
)))) {
1548 if (blk_throtl_bio(q
, bio
))
1549 return false; /* throttled, will be resubmitted later */
1551 trace_block_bio_queue(q
, bio
);
1555 bio_endio(bio
, err
);
1560 * generic_make_request - hand a buffer to its device driver for I/O
1561 * @bio: The bio describing the location in memory and on the device.
1563 * generic_make_request() is used to make I/O requests of block
1564 * devices. It is passed a &struct bio, which describes the I/O that needs
1567 * generic_make_request() does not return any status. The
1568 * success/failure status of the request, along with notification of
1569 * completion, is delivered asynchronously through the bio->bi_end_io
1570 * function described (one day) else where.
1572 * The caller of generic_make_request must make sure that bi_io_vec
1573 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1574 * set to describe the device address, and the
1575 * bi_end_io and optionally bi_private are set to describe how
1576 * completion notification should be signaled.
1578 * generic_make_request and the drivers it calls may use bi_next if this
1579 * bio happens to be merged with someone else, and may resubmit the bio to
1580 * a lower device by calling into generic_make_request recursively, which
1581 * means the bio should NOT be touched after the call to ->make_request_fn.
1583 void generic_make_request(struct bio
*bio
)
1585 struct bio_list bio_list_on_stack
;
1587 if (!generic_make_request_checks(bio
))
1591 * We only want one ->make_request_fn to be active at a time, else
1592 * stack usage with stacked devices could be a problem. So use
1593 * current->bio_list to keep a list of requests submited by a
1594 * make_request_fn function. current->bio_list is also used as a
1595 * flag to say if generic_make_request is currently active in this
1596 * task or not. If it is NULL, then no make_request is active. If
1597 * it is non-NULL, then a make_request is active, and new requests
1598 * should be added at the tail
1600 if (current
->bio_list
) {
1601 bio_list_add(current
->bio_list
, bio
);
1605 /* following loop may be a bit non-obvious, and so deserves some
1607 * Before entering the loop, bio->bi_next is NULL (as all callers
1608 * ensure that) so we have a list with a single bio.
1609 * We pretend that we have just taken it off a longer list, so
1610 * we assign bio_list to a pointer to the bio_list_on_stack,
1611 * thus initialising the bio_list of new bios to be
1612 * added. ->make_request() may indeed add some more bios
1613 * through a recursive call to generic_make_request. If it
1614 * did, we find a non-NULL value in bio_list and re-enter the loop
1615 * from the top. In this case we really did just take the bio
1616 * of the top of the list (no pretending) and so remove it from
1617 * bio_list, and call into ->make_request() again.
1619 BUG_ON(bio
->bi_next
);
1620 bio_list_init(&bio_list_on_stack
);
1621 current
->bio_list
= &bio_list_on_stack
;
1623 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
1625 q
->make_request_fn(q
, bio
);
1627 bio
= bio_list_pop(current
->bio_list
);
1629 current
->bio_list
= NULL
; /* deactivate */
1631 EXPORT_SYMBOL(generic_make_request
);
1634 * submit_bio - submit a bio to the block device layer for I/O
1635 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1636 * @bio: The &struct bio which describes the I/O
1638 * submit_bio() is very similar in purpose to generic_make_request(), and
1639 * uses that function to do most of the work. Both are fairly rough
1640 * interfaces; @bio must be presetup and ready for I/O.
1643 void submit_bio(int rw
, struct bio
*bio
)
1645 int count
= bio_sectors(bio
);
1650 * If it's a regular read/write or a barrier with data attached,
1651 * go through the normal accounting stuff before submission.
1653 if (bio_has_data(bio
) && !(rw
& REQ_DISCARD
)) {
1655 count_vm_events(PGPGOUT
, count
);
1657 task_io_account_read(bio
->bi_size
);
1658 count_vm_events(PGPGIN
, count
);
1661 if (unlikely(block_dump
)) {
1662 char b
[BDEVNAME_SIZE
];
1663 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s (%u sectors)\n",
1664 current
->comm
, task_pid_nr(current
),
1665 (rw
& WRITE
) ? "WRITE" : "READ",
1666 (unsigned long long)bio
->bi_sector
,
1667 bdevname(bio
->bi_bdev
, b
),
1672 generic_make_request(bio
);
1674 EXPORT_SYMBOL(submit_bio
);
1677 * blk_rq_check_limits - Helper function to check a request for the queue limit
1679 * @rq: the request being checked
1682 * @rq may have been made based on weaker limitations of upper-level queues
1683 * in request stacking drivers, and it may violate the limitation of @q.
1684 * Since the block layer and the underlying device driver trust @rq
1685 * after it is inserted to @q, it should be checked against @q before
1686 * the insertion using this generic function.
1688 * This function should also be useful for request stacking drivers
1689 * in some cases below, so export this function.
1690 * Request stacking drivers like request-based dm may change the queue
1691 * limits while requests are in the queue (e.g. dm's table swapping).
1692 * Such request stacking drivers should check those requests agaist
1693 * the new queue limits again when they dispatch those requests,
1694 * although such checkings are also done against the old queue limits
1695 * when submitting requests.
1697 int blk_rq_check_limits(struct request_queue
*q
, struct request
*rq
)
1699 if (rq
->cmd_flags
& REQ_DISCARD
)
1702 if (blk_rq_sectors(rq
) > queue_max_sectors(q
) ||
1703 blk_rq_bytes(rq
) > queue_max_hw_sectors(q
) << 9) {
1704 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
1709 * queue's settings related to segment counting like q->bounce_pfn
1710 * may differ from that of other stacking queues.
1711 * Recalculate it to check the request correctly on this queue's
1714 blk_recalc_rq_segments(rq
);
1715 if (rq
->nr_phys_segments
> queue_max_segments(q
)) {
1716 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
1722 EXPORT_SYMBOL_GPL(blk_rq_check_limits
);
1725 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1726 * @q: the queue to submit the request
1727 * @rq: the request being queued
1729 int blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
1731 unsigned long flags
;
1732 int where
= ELEVATOR_INSERT_BACK
;
1734 if (blk_rq_check_limits(q
, rq
))
1738 should_fail_request(&rq
->rq_disk
->part0
, blk_rq_bytes(rq
)))
1741 spin_lock_irqsave(q
->queue_lock
, flags
);
1742 if (unlikely(blk_queue_dead(q
))) {
1743 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1748 * Submitting request must be dequeued before calling this function
1749 * because it will be linked to another request_queue
1751 BUG_ON(blk_queued_rq(rq
));
1753 if (rq
->cmd_flags
& (REQ_FLUSH
|REQ_FUA
))
1754 where
= ELEVATOR_INSERT_FLUSH
;
1756 add_acct_request(q
, rq
, where
);
1757 if (where
== ELEVATOR_INSERT_FLUSH
)
1759 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1763 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
1766 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1767 * @rq: request to examine
1770 * A request could be merge of IOs which require different failure
1771 * handling. This function determines the number of bytes which
1772 * can be failed from the beginning of the request without
1773 * crossing into area which need to be retried further.
1776 * The number of bytes to fail.
1779 * queue_lock must be held.
1781 unsigned int blk_rq_err_bytes(const struct request
*rq
)
1783 unsigned int ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
1784 unsigned int bytes
= 0;
1787 if (!(rq
->cmd_flags
& REQ_MIXED_MERGE
))
1788 return blk_rq_bytes(rq
);
1791 * Currently the only 'mixing' which can happen is between
1792 * different fastfail types. We can safely fail portions
1793 * which have all the failfast bits that the first one has -
1794 * the ones which are at least as eager to fail as the first
1797 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
1798 if ((bio
->bi_rw
& ff
) != ff
)
1800 bytes
+= bio
->bi_size
;
1803 /* this could lead to infinite loop */
1804 BUG_ON(blk_rq_bytes(rq
) && !bytes
);
1807 EXPORT_SYMBOL_GPL(blk_rq_err_bytes
);
1809 static void blk_account_io_completion(struct request
*req
, unsigned int bytes
)
1811 if (blk_do_io_stat(req
)) {
1812 const int rw
= rq_data_dir(req
);
1813 struct hd_struct
*part
;
1816 cpu
= part_stat_lock();
1818 part_stat_add(cpu
, part
, sectors
[rw
], bytes
>> 9);
1823 static void blk_account_io_done(struct request
*req
)
1826 * Account IO completion. flush_rq isn't accounted as a
1827 * normal IO on queueing nor completion. Accounting the
1828 * containing request is enough.
1830 if (blk_do_io_stat(req
) && !(req
->cmd_flags
& REQ_FLUSH_SEQ
)) {
1831 unsigned long duration
= jiffies
- req
->start_time
;
1832 const int rw
= rq_data_dir(req
);
1833 struct hd_struct
*part
;
1836 cpu
= part_stat_lock();
1839 part_stat_inc(cpu
, part
, ios
[rw
]);
1840 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
1841 part_round_stats(cpu
, part
);
1842 part_dec_in_flight(part
, rw
);
1844 hd_struct_put(part
);
1850 * blk_peek_request - peek at the top of a request queue
1851 * @q: request queue to peek at
1854 * Return the request at the top of @q. The returned request
1855 * should be started using blk_start_request() before LLD starts
1859 * Pointer to the request at the top of @q if available. Null
1863 * queue_lock must be held.
1865 struct request
*blk_peek_request(struct request_queue
*q
)
1870 while ((rq
= __elv_next_request(q
)) != NULL
) {
1871 if (!(rq
->cmd_flags
& REQ_STARTED
)) {
1873 * This is the first time the device driver
1874 * sees this request (possibly after
1875 * requeueing). Notify IO scheduler.
1877 if (rq
->cmd_flags
& REQ_SORTED
)
1878 elv_activate_rq(q
, rq
);
1881 * just mark as started even if we don't start
1882 * it, a request that has been delayed should
1883 * not be passed by new incoming requests
1885 rq
->cmd_flags
|= REQ_STARTED
;
1886 trace_block_rq_issue(q
, rq
);
1889 if (!q
->boundary_rq
|| q
->boundary_rq
== rq
) {
1890 q
->end_sector
= rq_end_sector(rq
);
1891 q
->boundary_rq
= NULL
;
1894 if (rq
->cmd_flags
& REQ_DONTPREP
)
1897 if (q
->dma_drain_size
&& blk_rq_bytes(rq
)) {
1899 * make sure space for the drain appears we
1900 * know we can do this because max_hw_segments
1901 * has been adjusted to be one fewer than the
1904 rq
->nr_phys_segments
++;
1910 ret
= q
->prep_rq_fn(q
, rq
);
1911 if (ret
== BLKPREP_OK
) {
1913 } else if (ret
== BLKPREP_DEFER
) {
1915 * the request may have been (partially) prepped.
1916 * we need to keep this request in the front to
1917 * avoid resource deadlock. REQ_STARTED will
1918 * prevent other fs requests from passing this one.
1920 if (q
->dma_drain_size
&& blk_rq_bytes(rq
) &&
1921 !(rq
->cmd_flags
& REQ_DONTPREP
)) {
1923 * remove the space for the drain we added
1924 * so that we don't add it again
1926 --rq
->nr_phys_segments
;
1931 } else if (ret
== BLKPREP_KILL
) {
1932 rq
->cmd_flags
|= REQ_QUIET
;
1934 * Mark this request as started so we don't trigger
1935 * any debug logic in the end I/O path.
1937 blk_start_request(rq
);
1938 __blk_end_request_all(rq
, -EIO
);
1940 printk(KERN_ERR
"%s: bad return=%d\n", __func__
, ret
);
1947 EXPORT_SYMBOL(blk_peek_request
);
1949 void blk_dequeue_request(struct request
*rq
)
1951 struct request_queue
*q
= rq
->q
;
1953 BUG_ON(list_empty(&rq
->queuelist
));
1954 BUG_ON(ELV_ON_HASH(rq
));
1956 list_del_init(&rq
->queuelist
);
1959 * the time frame between a request being removed from the lists
1960 * and to it is freed is accounted as io that is in progress at
1963 if (blk_account_rq(rq
)) {
1964 q
->in_flight
[rq_is_sync(rq
)]++;
1965 set_io_start_time_ns(rq
);
1970 * blk_start_request - start request processing on the driver
1971 * @req: request to dequeue
1974 * Dequeue @req and start timeout timer on it. This hands off the
1975 * request to the driver.
1977 * Block internal functions which don't want to start timer should
1978 * call blk_dequeue_request().
1981 * queue_lock must be held.
1983 void blk_start_request(struct request
*req
)
1985 blk_dequeue_request(req
);
1988 * We are now handing the request to the hardware, initialize
1989 * resid_len to full count and add the timeout handler.
1991 req
->resid_len
= blk_rq_bytes(req
);
1992 if (unlikely(blk_bidi_rq(req
)))
1993 req
->next_rq
->resid_len
= blk_rq_bytes(req
->next_rq
);
1997 EXPORT_SYMBOL(blk_start_request
);
2000 * blk_fetch_request - fetch a request from a request queue
2001 * @q: request queue to fetch a request from
2004 * Return the request at the top of @q. The request is started on
2005 * return and LLD can start processing it immediately.
2008 * Pointer to the request at the top of @q if available. Null
2012 * queue_lock must be held.
2014 struct request
*blk_fetch_request(struct request_queue
*q
)
2018 rq
= blk_peek_request(q
);
2020 blk_start_request(rq
);
2023 EXPORT_SYMBOL(blk_fetch_request
);
2026 * blk_update_request - Special helper function for request stacking drivers
2027 * @req: the request being processed
2028 * @error: %0 for success, < %0 for error
2029 * @nr_bytes: number of bytes to complete @req
2032 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2033 * the request structure even if @req doesn't have leftover.
2034 * If @req has leftover, sets it up for the next range of segments.
2036 * This special helper function is only for request stacking drivers
2037 * (e.g. request-based dm) so that they can handle partial completion.
2038 * Actual device drivers should use blk_end_request instead.
2040 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2041 * %false return from this function.
2044 * %false - this request doesn't have any more data
2045 * %true - this request has more data
2047 bool blk_update_request(struct request
*req
, int error
, unsigned int nr_bytes
)
2049 int total_bytes
, bio_nbytes
, next_idx
= 0;
2055 trace_block_rq_complete(req
->q
, req
);
2058 * For fs requests, rq is just carrier of independent bio's
2059 * and each partial completion should be handled separately.
2060 * Reset per-request error on each partial completion.
2062 * TODO: tj: This is too subtle. It would be better to let
2063 * low level drivers do what they see fit.
2065 if (req
->cmd_type
== REQ_TYPE_FS
)
2068 if (error
&& req
->cmd_type
== REQ_TYPE_FS
&&
2069 !(req
->cmd_flags
& REQ_QUIET
)) {
2074 error_type
= "recoverable transport";
2077 error_type
= "critical target";
2080 error_type
= "critical nexus";
2087 printk(KERN_ERR
"end_request: %s error, dev %s, sector %llu\n",
2088 error_type
, req
->rq_disk
? req
->rq_disk
->disk_name
: "?",
2089 (unsigned long long)blk_rq_pos(req
));
2092 blk_account_io_completion(req
, nr_bytes
);
2094 total_bytes
= bio_nbytes
= 0;
2095 while ((bio
= req
->bio
) != NULL
) {
2098 if (nr_bytes
>= bio
->bi_size
) {
2099 req
->bio
= bio
->bi_next
;
2100 nbytes
= bio
->bi_size
;
2101 req_bio_endio(req
, bio
, nbytes
, error
);
2105 int idx
= bio
->bi_idx
+ next_idx
;
2107 if (unlikely(idx
>= bio
->bi_vcnt
)) {
2108 blk_dump_rq_flags(req
, "__end_that");
2109 printk(KERN_ERR
"%s: bio idx %d >= vcnt %d\n",
2110 __func__
, idx
, bio
->bi_vcnt
);
2114 nbytes
= bio_iovec_idx(bio
, idx
)->bv_len
;
2115 BIO_BUG_ON(nbytes
> bio
->bi_size
);
2118 * not a complete bvec done
2120 if (unlikely(nbytes
> nr_bytes
)) {
2121 bio_nbytes
+= nr_bytes
;
2122 total_bytes
+= nr_bytes
;
2127 * advance to the next vector
2130 bio_nbytes
+= nbytes
;
2133 total_bytes
+= nbytes
;
2139 * end more in this run, or just return 'not-done'
2141 if (unlikely(nr_bytes
<= 0))
2151 * Reset counters so that the request stacking driver
2152 * can find how many bytes remain in the request
2155 req
->__data_len
= 0;
2160 * if the request wasn't completed, update state
2163 req_bio_endio(req
, bio
, bio_nbytes
, error
);
2164 bio
->bi_idx
+= next_idx
;
2165 bio_iovec(bio
)->bv_offset
+= nr_bytes
;
2166 bio_iovec(bio
)->bv_len
-= nr_bytes
;
2169 req
->__data_len
-= total_bytes
;
2170 req
->buffer
= bio_data(req
->bio
);
2172 /* update sector only for requests with clear definition of sector */
2173 if (req
->cmd_type
== REQ_TYPE_FS
|| (req
->cmd_flags
& REQ_DISCARD
))
2174 req
->__sector
+= total_bytes
>> 9;
2176 /* mixed attributes always follow the first bio */
2177 if (req
->cmd_flags
& REQ_MIXED_MERGE
) {
2178 req
->cmd_flags
&= ~REQ_FAILFAST_MASK
;
2179 req
->cmd_flags
|= req
->bio
->bi_rw
& REQ_FAILFAST_MASK
;
2183 * If total number of sectors is less than the first segment
2184 * size, something has gone terribly wrong.
2186 if (blk_rq_bytes(req
) < blk_rq_cur_bytes(req
)) {
2187 blk_dump_rq_flags(req
, "request botched");
2188 req
->__data_len
= blk_rq_cur_bytes(req
);
2191 /* recalculate the number of segments */
2192 blk_recalc_rq_segments(req
);
2196 EXPORT_SYMBOL_GPL(blk_update_request
);
2198 static bool blk_update_bidi_request(struct request
*rq
, int error
,
2199 unsigned int nr_bytes
,
2200 unsigned int bidi_bytes
)
2202 if (blk_update_request(rq
, error
, nr_bytes
))
2205 /* Bidi request must be completed as a whole */
2206 if (unlikely(blk_bidi_rq(rq
)) &&
2207 blk_update_request(rq
->next_rq
, error
, bidi_bytes
))
2210 if (blk_queue_add_random(rq
->q
))
2211 add_disk_randomness(rq
->rq_disk
);
2217 * blk_unprep_request - unprepare a request
2220 * This function makes a request ready for complete resubmission (or
2221 * completion). It happens only after all error handling is complete,
2222 * so represents the appropriate moment to deallocate any resources
2223 * that were allocated to the request in the prep_rq_fn. The queue
2224 * lock is held when calling this.
2226 void blk_unprep_request(struct request
*req
)
2228 struct request_queue
*q
= req
->q
;
2230 req
->cmd_flags
&= ~REQ_DONTPREP
;
2231 if (q
->unprep_rq_fn
)
2232 q
->unprep_rq_fn(q
, req
);
2234 EXPORT_SYMBOL_GPL(blk_unprep_request
);
2237 * queue lock must be held
2239 static void blk_finish_request(struct request
*req
, int error
)
2241 if (blk_rq_tagged(req
))
2242 blk_queue_end_tag(req
->q
, req
);
2244 BUG_ON(blk_queued_rq(req
));
2246 if (unlikely(laptop_mode
) && req
->cmd_type
== REQ_TYPE_FS
)
2247 laptop_io_completion(&req
->q
->backing_dev_info
);
2249 blk_delete_timer(req
);
2251 if (req
->cmd_flags
& REQ_DONTPREP
)
2252 blk_unprep_request(req
);
2255 blk_account_io_done(req
);
2258 req
->end_io(req
, error
);
2260 if (blk_bidi_rq(req
))
2261 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
2263 __blk_put_request(req
->q
, req
);
2268 * blk_end_bidi_request - Complete a bidi request
2269 * @rq: the request to complete
2270 * @error: %0 for success, < %0 for error
2271 * @nr_bytes: number of bytes to complete @rq
2272 * @bidi_bytes: number of bytes to complete @rq->next_rq
2275 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2276 * Drivers that supports bidi can safely call this member for any
2277 * type of request, bidi or uni. In the later case @bidi_bytes is
2281 * %false - we are done with this request
2282 * %true - still buffers pending for this request
2284 static bool blk_end_bidi_request(struct request
*rq
, int error
,
2285 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2287 struct request_queue
*q
= rq
->q
;
2288 unsigned long flags
;
2290 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2293 spin_lock_irqsave(q
->queue_lock
, flags
);
2294 blk_finish_request(rq
, error
);
2295 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2301 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2302 * @rq: the request to complete
2303 * @error: %0 for success, < %0 for error
2304 * @nr_bytes: number of bytes to complete @rq
2305 * @bidi_bytes: number of bytes to complete @rq->next_rq
2308 * Identical to blk_end_bidi_request() except that queue lock is
2309 * assumed to be locked on entry and remains so on return.
2312 * %false - we are done with this request
2313 * %true - still buffers pending for this request
2315 bool __blk_end_bidi_request(struct request
*rq
, int error
,
2316 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2318 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2321 blk_finish_request(rq
, error
);
2327 * blk_end_request - Helper function for drivers to complete the request.
2328 * @rq: the request being processed
2329 * @error: %0 for success, < %0 for error
2330 * @nr_bytes: number of bytes to complete
2333 * Ends I/O on a number of bytes attached to @rq.
2334 * If @rq has leftover, sets it up for the next range of segments.
2337 * %false - we are done with this request
2338 * %true - still buffers pending for this request
2340 bool blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2342 return blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2344 EXPORT_SYMBOL(blk_end_request
);
2347 * blk_end_request_all - Helper function for drives to finish the request.
2348 * @rq: the request to finish
2349 * @error: %0 for success, < %0 for error
2352 * Completely finish @rq.
2354 void blk_end_request_all(struct request
*rq
, int error
)
2357 unsigned int bidi_bytes
= 0;
2359 if (unlikely(blk_bidi_rq(rq
)))
2360 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2362 pending
= blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2365 EXPORT_SYMBOL(blk_end_request_all
);
2368 * blk_end_request_cur - Helper function to finish the current request chunk.
2369 * @rq: the request to finish the current chunk for
2370 * @error: %0 for success, < %0 for error
2373 * Complete the current consecutively mapped chunk from @rq.
2376 * %false - we are done with this request
2377 * %true - still buffers pending for this request
2379 bool blk_end_request_cur(struct request
*rq
, int error
)
2381 return blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2383 EXPORT_SYMBOL(blk_end_request_cur
);
2386 * blk_end_request_err - Finish a request till the next failure boundary.
2387 * @rq: the request to finish till the next failure boundary for
2388 * @error: must be negative errno
2391 * Complete @rq till the next failure boundary.
2394 * %false - we are done with this request
2395 * %true - still buffers pending for this request
2397 bool blk_end_request_err(struct request
*rq
, int error
)
2399 WARN_ON(error
>= 0);
2400 return blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2402 EXPORT_SYMBOL_GPL(blk_end_request_err
);
2405 * __blk_end_request - Helper function for drivers to complete the request.
2406 * @rq: the request being processed
2407 * @error: %0 for success, < %0 for error
2408 * @nr_bytes: number of bytes to complete
2411 * Must be called with queue lock held unlike blk_end_request().
2414 * %false - we are done with this request
2415 * %true - still buffers pending for this request
2417 bool __blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2419 return __blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2421 EXPORT_SYMBOL(__blk_end_request
);
2424 * __blk_end_request_all - Helper function for drives to finish the request.
2425 * @rq: the request to finish
2426 * @error: %0 for success, < %0 for error
2429 * Completely finish @rq. Must be called with queue lock held.
2431 void __blk_end_request_all(struct request
*rq
, int error
)
2434 unsigned int bidi_bytes
= 0;
2436 if (unlikely(blk_bidi_rq(rq
)))
2437 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2439 pending
= __blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2442 EXPORT_SYMBOL(__blk_end_request_all
);
2445 * __blk_end_request_cur - Helper function to finish the current request chunk.
2446 * @rq: the request to finish the current chunk for
2447 * @error: %0 for success, < %0 for error
2450 * Complete the current consecutively mapped chunk from @rq. Must
2451 * be called with queue lock held.
2454 * %false - we are done with this request
2455 * %true - still buffers pending for this request
2457 bool __blk_end_request_cur(struct request
*rq
, int error
)
2459 return __blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2461 EXPORT_SYMBOL(__blk_end_request_cur
);
2464 * __blk_end_request_err - Finish a request till the next failure boundary.
2465 * @rq: the request to finish till the next failure boundary for
2466 * @error: must be negative errno
2469 * Complete @rq till the next failure boundary. Must be called
2470 * with queue lock held.
2473 * %false - we are done with this request
2474 * %true - still buffers pending for this request
2476 bool __blk_end_request_err(struct request
*rq
, int error
)
2478 WARN_ON(error
>= 0);
2479 return __blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2481 EXPORT_SYMBOL_GPL(__blk_end_request_err
);
2483 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
2486 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2487 rq
->cmd_flags
|= bio
->bi_rw
& REQ_WRITE
;
2489 if (bio_has_data(bio
)) {
2490 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2491 rq
->buffer
= bio_data(bio
);
2493 rq
->__data_len
= bio
->bi_size
;
2494 rq
->bio
= rq
->biotail
= bio
;
2497 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2500 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2502 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2503 * @rq: the request to be flushed
2506 * Flush all pages in @rq.
2508 void rq_flush_dcache_pages(struct request
*rq
)
2510 struct req_iterator iter
;
2511 struct bio_vec
*bvec
;
2513 rq_for_each_segment(bvec
, rq
, iter
)
2514 flush_dcache_page(bvec
->bv_page
);
2516 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages
);
2520 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2521 * @q : the queue of the device being checked
2524 * Check if underlying low-level drivers of a device are busy.
2525 * If the drivers want to export their busy state, they must set own
2526 * exporting function using blk_queue_lld_busy() first.
2528 * Basically, this function is used only by request stacking drivers
2529 * to stop dispatching requests to underlying devices when underlying
2530 * devices are busy. This behavior helps more I/O merging on the queue
2531 * of the request stacking driver and prevents I/O throughput regression
2532 * on burst I/O load.
2535 * 0 - Not busy (The request stacking driver should dispatch request)
2536 * 1 - Busy (The request stacking driver should stop dispatching request)
2538 int blk_lld_busy(struct request_queue
*q
)
2541 return q
->lld_busy_fn(q
);
2545 EXPORT_SYMBOL_GPL(blk_lld_busy
);
2548 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2549 * @rq: the clone request to be cleaned up
2552 * Free all bios in @rq for a cloned request.
2554 void blk_rq_unprep_clone(struct request
*rq
)
2558 while ((bio
= rq
->bio
) != NULL
) {
2559 rq
->bio
= bio
->bi_next
;
2564 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone
);
2567 * Copy attributes of the original request to the clone request.
2568 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2570 static void __blk_rq_prep_clone(struct request
*dst
, struct request
*src
)
2572 dst
->cpu
= src
->cpu
;
2573 dst
->cmd_flags
= (src
->cmd_flags
& REQ_CLONE_MASK
) | REQ_NOMERGE
;
2574 dst
->cmd_type
= src
->cmd_type
;
2575 dst
->__sector
= blk_rq_pos(src
);
2576 dst
->__data_len
= blk_rq_bytes(src
);
2577 dst
->nr_phys_segments
= src
->nr_phys_segments
;
2578 dst
->ioprio
= src
->ioprio
;
2579 dst
->extra_len
= src
->extra_len
;
2583 * blk_rq_prep_clone - Helper function to setup clone request
2584 * @rq: the request to be setup
2585 * @rq_src: original request to be cloned
2586 * @bs: bio_set that bios for clone are allocated from
2587 * @gfp_mask: memory allocation mask for bio
2588 * @bio_ctr: setup function to be called for each clone bio.
2589 * Returns %0 for success, non %0 for failure.
2590 * @data: private data to be passed to @bio_ctr
2593 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2594 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2595 * are not copied, and copying such parts is the caller's responsibility.
2596 * Also, pages which the original bios are pointing to are not copied
2597 * and the cloned bios just point same pages.
2598 * So cloned bios must be completed before original bios, which means
2599 * the caller must complete @rq before @rq_src.
2601 int blk_rq_prep_clone(struct request
*rq
, struct request
*rq_src
,
2602 struct bio_set
*bs
, gfp_t gfp_mask
,
2603 int (*bio_ctr
)(struct bio
*, struct bio
*, void *),
2606 struct bio
*bio
, *bio_src
;
2611 blk_rq_init(NULL
, rq
);
2613 __rq_for_each_bio(bio_src
, rq_src
) {
2614 bio
= bio_alloc_bioset(gfp_mask
, bio_src
->bi_max_vecs
, bs
);
2618 __bio_clone(bio
, bio_src
);
2620 if (bio_integrity(bio_src
) &&
2621 bio_integrity_clone(bio
, bio_src
, gfp_mask
, bs
))
2624 if (bio_ctr
&& bio_ctr(bio
, bio_src
, data
))
2628 rq
->biotail
->bi_next
= bio
;
2631 rq
->bio
= rq
->biotail
= bio
;
2634 __blk_rq_prep_clone(rq
, rq_src
);
2641 blk_rq_unprep_clone(rq
);
2645 EXPORT_SYMBOL_GPL(blk_rq_prep_clone
);
2647 int kblockd_schedule_work(struct request_queue
*q
, struct work_struct
*work
)
2649 return queue_work(kblockd_workqueue
, work
);
2651 EXPORT_SYMBOL(kblockd_schedule_work
);
2653 int kblockd_schedule_delayed_work(struct request_queue
*q
,
2654 struct delayed_work
*dwork
, unsigned long delay
)
2656 return queue_delayed_work(kblockd_workqueue
, dwork
, delay
);
2658 EXPORT_SYMBOL(kblockd_schedule_delayed_work
);
2660 #define PLUG_MAGIC 0x91827364
2663 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2664 * @plug: The &struct blk_plug that needs to be initialized
2667 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2668 * pending I/O should the task end up blocking between blk_start_plug() and
2669 * blk_finish_plug(). This is important from a performance perspective, but
2670 * also ensures that we don't deadlock. For instance, if the task is blocking
2671 * for a memory allocation, memory reclaim could end up wanting to free a
2672 * page belonging to that request that is currently residing in our private
2673 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2674 * this kind of deadlock.
2676 void blk_start_plug(struct blk_plug
*plug
)
2678 struct task_struct
*tsk
= current
;
2680 plug
->magic
= PLUG_MAGIC
;
2681 INIT_LIST_HEAD(&plug
->list
);
2682 INIT_LIST_HEAD(&plug
->cb_list
);
2683 plug
->should_sort
= 0;
2686 * If this is a nested plug, don't actually assign it. It will be
2687 * flushed on its own.
2691 * Store ordering should not be needed here, since a potential
2692 * preempt will imply a full memory barrier
2697 EXPORT_SYMBOL(blk_start_plug
);
2699 static int plug_rq_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
2701 struct request
*rqa
= container_of(a
, struct request
, queuelist
);
2702 struct request
*rqb
= container_of(b
, struct request
, queuelist
);
2704 return !(rqa
->q
<= rqb
->q
);
2708 * If 'from_schedule' is true, then postpone the dispatch of requests
2709 * until a safe kblockd context. We due this to avoid accidental big
2710 * additional stack usage in driver dispatch, in places where the originally
2711 * plugger did not intend it.
2713 static void queue_unplugged(struct request_queue
*q
, unsigned int depth
,
2715 __releases(q
->queue_lock
)
2717 trace_block_unplug(q
, depth
, !from_schedule
);
2720 * Don't mess with dead queue.
2722 if (unlikely(blk_queue_dead(q
))) {
2723 spin_unlock(q
->queue_lock
);
2728 * If we are punting this to kblockd, then we can safely drop
2729 * the queue_lock before waking kblockd (which needs to take
2732 if (from_schedule
) {
2733 spin_unlock(q
->queue_lock
);
2734 blk_run_queue_async(q
);
2737 spin_unlock(q
->queue_lock
);
2742 static void flush_plug_callbacks(struct blk_plug
*plug
)
2744 LIST_HEAD(callbacks
);
2746 if (list_empty(&plug
->cb_list
))
2749 list_splice_init(&plug
->cb_list
, &callbacks
);
2751 while (!list_empty(&callbacks
)) {
2752 struct blk_plug_cb
*cb
= list_first_entry(&callbacks
,
2755 list_del(&cb
->list
);
2760 void blk_flush_plug_list(struct blk_plug
*plug
, bool from_schedule
)
2762 struct request_queue
*q
;
2763 unsigned long flags
;
2768 BUG_ON(plug
->magic
!= PLUG_MAGIC
);
2770 flush_plug_callbacks(plug
);
2771 if (list_empty(&plug
->list
))
2774 list_splice_init(&plug
->list
, &list
);
2776 if (plug
->should_sort
) {
2777 list_sort(NULL
, &list
, plug_rq_cmp
);
2778 plug
->should_sort
= 0;
2785 * Save and disable interrupts here, to avoid doing it for every
2786 * queue lock we have to take.
2788 local_irq_save(flags
);
2789 while (!list_empty(&list
)) {
2790 rq
= list_entry_rq(list
.next
);
2791 list_del_init(&rq
->queuelist
);
2795 * This drops the queue lock
2798 queue_unplugged(q
, depth
, from_schedule
);
2801 spin_lock(q
->queue_lock
);
2805 * Short-circuit if @q is dead
2807 if (unlikely(blk_queue_dead(q
))) {
2808 __blk_end_request_all(rq
, -ENODEV
);
2813 * rq is already accounted, so use raw insert
2815 if (rq
->cmd_flags
& (REQ_FLUSH
| REQ_FUA
))
2816 __elv_add_request(q
, rq
, ELEVATOR_INSERT_FLUSH
);
2818 __elv_add_request(q
, rq
, ELEVATOR_INSERT_SORT_MERGE
);
2824 * This drops the queue lock
2827 queue_unplugged(q
, depth
, from_schedule
);
2829 local_irq_restore(flags
);
2832 void blk_finish_plug(struct blk_plug
*plug
)
2834 blk_flush_plug_list(plug
, false);
2836 if (plug
== current
->plug
)
2837 current
->plug
= NULL
;
2839 EXPORT_SYMBOL(blk_finish_plug
);
2841 int __init
blk_dev_init(void)
2843 BUILD_BUG_ON(__REQ_NR_BITS
> 8 *
2844 sizeof(((struct request
*)0)->cmd_flags
));
2846 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
2847 kblockd_workqueue
= alloc_workqueue("kblockd",
2848 WQ_MEM_RECLAIM
| WQ_HIGHPRI
, 0);
2849 if (!kblockd_workqueue
)
2850 panic("Failed to create kblockd\n");
2852 request_cachep
= kmem_cache_create("blkdev_requests",
2853 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
2855 blk_requestq_cachep
= kmem_cache_create("blkdev_queue",
2856 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);