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
);
378 * This function might be called on a queue which failed
379 * driver init after queue creation or is not yet fully
380 * active yet. Some drivers (e.g. fd and loop) get unhappy
381 * in such cases. Kick queue iff dispatch queue has
382 * something on it and @q has request_fn set.
384 if (!list_empty(&q
->queue_head
) && q
->request_fn
)
387 drain
|= q
->rq
.elvpriv
;
390 * Unfortunately, requests are queued at and tracked from
391 * multiple places and there's no single counter which can
392 * be drained. Check all the queues and counters.
395 drain
|= !list_empty(&q
->queue_head
);
396 for (i
= 0; i
< 2; i
++) {
397 drain
|= q
->rq
.count
[i
];
398 drain
|= q
->in_flight
[i
];
399 drain
|= !list_empty(&q
->flush_queue
[i
]);
403 spin_unlock_irq(q
->queue_lock
);
412 * blk_queue_bypass_start - enter queue bypass mode
413 * @q: queue of interest
415 * In bypass mode, only the dispatch FIFO queue of @q is used. This
416 * function makes @q enter bypass mode and drains all requests which were
417 * throttled or issued before. On return, it's guaranteed that no request
418 * is being throttled or has ELVPRIV set.
420 void blk_queue_bypass_start(struct request_queue
*q
)
422 spin_lock_irq(q
->queue_lock
);
424 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
425 spin_unlock_irq(q
->queue_lock
);
427 blk_drain_queue(q
, false);
429 EXPORT_SYMBOL_GPL(blk_queue_bypass_start
);
432 * blk_queue_bypass_end - leave queue bypass mode
433 * @q: queue of interest
435 * Leave bypass mode and restore the normal queueing behavior.
437 void blk_queue_bypass_end(struct request_queue
*q
)
439 spin_lock_irq(q
->queue_lock
);
440 if (!--q
->bypass_depth
)
441 queue_flag_clear(QUEUE_FLAG_BYPASS
, q
);
442 WARN_ON_ONCE(q
->bypass_depth
< 0);
443 spin_unlock_irq(q
->queue_lock
);
445 EXPORT_SYMBOL_GPL(blk_queue_bypass_end
);
448 * blk_cleanup_queue - shutdown a request queue
449 * @q: request queue to shutdown
451 * Mark @q DEAD, drain all pending requests, destroy and put it. All
452 * future requests will be failed immediately with -ENODEV.
454 void blk_cleanup_queue(struct request_queue
*q
)
456 spinlock_t
*lock
= q
->queue_lock
;
458 /* mark @q DEAD, no new request or merges will be allowed afterwards */
459 mutex_lock(&q
->sysfs_lock
);
460 queue_flag_set_unlocked(QUEUE_FLAG_DEAD
, q
);
464 /* dead queue is permanently in bypass mode till released */
466 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
468 queue_flag_set(QUEUE_FLAG_NOMERGES
, q
);
469 queue_flag_set(QUEUE_FLAG_NOXMERGES
, q
);
470 queue_flag_set(QUEUE_FLAG_DEAD
, q
);
472 if (q
->queue_lock
!= &q
->__queue_lock
)
473 q
->queue_lock
= &q
->__queue_lock
;
475 spin_unlock_irq(lock
);
476 mutex_unlock(&q
->sysfs_lock
);
478 /* drain all requests queued before DEAD marking */
479 blk_drain_queue(q
, true);
481 /* @q won't process any more request, flush async actions */
482 del_timer_sync(&q
->backing_dev_info
.laptop_mode_wb_timer
);
485 /* @q is and will stay empty, shutdown and put */
488 EXPORT_SYMBOL(blk_cleanup_queue
);
490 static int blk_init_free_list(struct request_queue
*q
)
492 struct request_list
*rl
= &q
->rq
;
494 if (unlikely(rl
->rq_pool
))
497 rl
->count
[BLK_RW_SYNC
] = rl
->count
[BLK_RW_ASYNC
] = 0;
498 rl
->starved
[BLK_RW_SYNC
] = rl
->starved
[BLK_RW_ASYNC
] = 0;
500 init_waitqueue_head(&rl
->wait
[BLK_RW_SYNC
]);
501 init_waitqueue_head(&rl
->wait
[BLK_RW_ASYNC
]);
503 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, mempool_alloc_slab
,
504 mempool_free_slab
, request_cachep
, q
->node
);
512 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
514 return blk_alloc_queue_node(gfp_mask
, -1);
516 EXPORT_SYMBOL(blk_alloc_queue
);
518 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
520 struct request_queue
*q
;
523 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
524 gfp_mask
| __GFP_ZERO
, node_id
);
528 q
->id
= ida_simple_get(&blk_queue_ida
, 0, 0, GFP_KERNEL
);
532 q
->backing_dev_info
.ra_pages
=
533 (VM_MAX_READAHEAD
* 1024) / PAGE_CACHE_SIZE
;
534 q
->backing_dev_info
.state
= 0;
535 q
->backing_dev_info
.capabilities
= BDI_CAP_MAP_COPY
;
536 q
->backing_dev_info
.name
= "block";
539 err
= bdi_init(&q
->backing_dev_info
);
543 if (blk_throtl_init(q
))
546 setup_timer(&q
->backing_dev_info
.laptop_mode_wb_timer
,
547 laptop_mode_timer_fn
, (unsigned long) q
);
548 setup_timer(&q
->timeout
, blk_rq_timed_out_timer
, (unsigned long) q
);
549 INIT_LIST_HEAD(&q
->queue_head
);
550 INIT_LIST_HEAD(&q
->timeout_list
);
551 INIT_LIST_HEAD(&q
->icq_list
);
552 INIT_LIST_HEAD(&q
->flush_queue
[0]);
553 INIT_LIST_HEAD(&q
->flush_queue
[1]);
554 INIT_LIST_HEAD(&q
->flush_data_in_flight
);
555 INIT_DELAYED_WORK(&q
->delay_work
, blk_delay_work
);
557 kobject_init(&q
->kobj
, &blk_queue_ktype
);
559 mutex_init(&q
->sysfs_lock
);
560 spin_lock_init(&q
->__queue_lock
);
563 * By default initialize queue_lock to internal lock and driver can
564 * override it later if need be.
566 q
->queue_lock
= &q
->__queue_lock
;
571 ida_simple_remove(&blk_queue_ida
, q
->id
);
573 kmem_cache_free(blk_requestq_cachep
, q
);
576 EXPORT_SYMBOL(blk_alloc_queue_node
);
579 * blk_init_queue - prepare a request queue for use with a block device
580 * @rfn: The function to be called to process requests that have been
581 * placed on the queue.
582 * @lock: Request queue spin lock
585 * If a block device wishes to use the standard request handling procedures,
586 * which sorts requests and coalesces adjacent requests, then it must
587 * call blk_init_queue(). The function @rfn will be called when there
588 * are requests on the queue that need to be processed. If the device
589 * supports plugging, then @rfn may not be called immediately when requests
590 * are available on the queue, but may be called at some time later instead.
591 * Plugged queues are generally unplugged when a buffer belonging to one
592 * of the requests on the queue is needed, or due to memory pressure.
594 * @rfn is not required, or even expected, to remove all requests off the
595 * queue, but only as many as it can handle at a time. If it does leave
596 * requests on the queue, it is responsible for arranging that the requests
597 * get dealt with eventually.
599 * The queue spin lock must be held while manipulating the requests on the
600 * request queue; this lock will be taken also from interrupt context, so irq
601 * disabling is needed for it.
603 * Function returns a pointer to the initialized request queue, or %NULL if
607 * blk_init_queue() must be paired with a blk_cleanup_queue() call
608 * when the block device is deactivated (such as at module unload).
611 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
613 return blk_init_queue_node(rfn
, lock
, -1);
615 EXPORT_SYMBOL(blk_init_queue
);
617 struct request_queue
*
618 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
620 struct request_queue
*uninit_q
, *q
;
622 uninit_q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
626 q
= blk_init_allocated_queue(uninit_q
, rfn
, lock
);
628 blk_cleanup_queue(uninit_q
);
632 EXPORT_SYMBOL(blk_init_queue_node
);
634 struct request_queue
*
635 blk_init_allocated_queue(struct request_queue
*q
, request_fn_proc
*rfn
,
641 if (blk_init_free_list(q
))
645 q
->prep_rq_fn
= NULL
;
646 q
->unprep_rq_fn
= NULL
;
647 q
->queue_flags
= QUEUE_FLAG_DEFAULT
;
649 /* Override internal queue lock with supplied lock pointer */
651 q
->queue_lock
= lock
;
654 * This also sets hw/phys segments, boundary and size
656 blk_queue_make_request(q
, blk_queue_bio
);
658 q
->sg_reserved_size
= INT_MAX
;
663 if (!elevator_init(q
, NULL
)) {
664 blk_queue_congestion_threshold(q
);
670 EXPORT_SYMBOL(blk_init_allocated_queue
);
672 bool blk_get_queue(struct request_queue
*q
)
674 if (likely(!blk_queue_dead(q
))) {
681 EXPORT_SYMBOL(blk_get_queue
);
683 static inline void blk_free_request(struct request_queue
*q
, struct request
*rq
)
685 if (rq
->cmd_flags
& REQ_ELVPRIV
) {
686 elv_put_request(q
, rq
);
688 put_io_context(rq
->elv
.icq
->ioc
);
691 mempool_free(rq
, q
->rq
.rq_pool
);
694 static struct request
*
695 blk_alloc_request(struct request_queue
*q
, struct io_cq
*icq
,
696 unsigned int flags
, gfp_t gfp_mask
)
698 struct request
*rq
= mempool_alloc(q
->rq
.rq_pool
, gfp_mask
);
705 rq
->cmd_flags
= flags
| REQ_ALLOCED
;
707 if (flags
& REQ_ELVPRIV
) {
709 if (unlikely(elv_set_request(q
, rq
, gfp_mask
))) {
710 mempool_free(rq
, q
->rq
.rq_pool
);
713 /* @rq->elv.icq holds on to io_context until @rq is freed */
715 get_io_context(icq
->ioc
);
722 * ioc_batching returns true if the ioc is a valid batching request and
723 * should be given priority access to a request.
725 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
731 * Make sure the process is able to allocate at least 1 request
732 * even if the batch times out, otherwise we could theoretically
735 return ioc
->nr_batch_requests
== q
->nr_batching
||
736 (ioc
->nr_batch_requests
> 0
737 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
741 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
742 * will cause the process to be a "batcher" on all queues in the system. This
743 * is the behaviour we want though - once it gets a wakeup it should be given
746 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
748 if (!ioc
|| ioc_batching(q
, ioc
))
751 ioc
->nr_batch_requests
= q
->nr_batching
;
752 ioc
->last_waited
= jiffies
;
755 static void __freed_request(struct request_queue
*q
, int sync
)
757 struct request_list
*rl
= &q
->rq
;
759 if (rl
->count
[sync
] < queue_congestion_off_threshold(q
))
760 blk_clear_queue_congested(q
, sync
);
762 if (rl
->count
[sync
] + 1 <= q
->nr_requests
) {
763 if (waitqueue_active(&rl
->wait
[sync
]))
764 wake_up(&rl
->wait
[sync
]);
766 blk_clear_queue_full(q
, sync
);
771 * A request has just been released. Account for it, update the full and
772 * congestion status, wake up any waiters. Called under q->queue_lock.
774 static void freed_request(struct request_queue
*q
, unsigned int flags
)
776 struct request_list
*rl
= &q
->rq
;
777 int sync
= rw_is_sync(flags
);
780 if (flags
& REQ_ELVPRIV
)
783 __freed_request(q
, sync
);
785 if (unlikely(rl
->starved
[sync
^ 1]))
786 __freed_request(q
, sync
^ 1);
790 * Determine if elevator data should be initialized when allocating the
791 * request associated with @bio.
793 static bool blk_rq_should_init_elevator(struct bio
*bio
)
799 * Flush requests do not use the elevator so skip initialization.
800 * This allows a request to share the flush and elevator data.
802 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
))
809 * get_request - get a free request
810 * @q: request_queue to allocate request from
811 * @rw_flags: RW and SYNC flags
812 * @bio: bio to allocate request for (can be %NULL)
813 * @gfp_mask: allocation mask
815 * Get a free request from @q. This function may fail under memory
816 * pressure or if @q is dead.
818 * Must be callled with @q->queue_lock held and,
819 * Returns %NULL on failure, with @q->queue_lock held.
820 * Returns !%NULL on success, with @q->queue_lock *not held*.
822 static struct request
*get_request(struct request_queue
*q
, int rw_flags
,
823 struct bio
*bio
, gfp_t gfp_mask
)
825 struct request
*rq
= NULL
;
826 struct request_list
*rl
= &q
->rq
;
827 struct elevator_type
*et
;
828 struct io_context
*ioc
;
829 struct io_cq
*icq
= NULL
;
830 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
831 bool retried
= false;
834 et
= q
->elevator
->type
;
835 ioc
= current
->io_context
;
837 if (unlikely(blk_queue_dead(q
)))
840 may_queue
= elv_may_queue(q
, rw_flags
);
841 if (may_queue
== ELV_MQUEUE_NO
)
844 if (rl
->count
[is_sync
]+1 >= queue_congestion_on_threshold(q
)) {
845 if (rl
->count
[is_sync
]+1 >= q
->nr_requests
) {
847 * We want ioc to record batching state. If it's
848 * not already there, creating a new one requires
849 * dropping queue_lock, which in turn requires
850 * retesting conditions to avoid queue hang.
852 if (!ioc
&& !retried
) {
853 spin_unlock_irq(q
->queue_lock
);
854 create_io_context(current
, gfp_mask
, q
->node
);
855 spin_lock_irq(q
->queue_lock
);
861 * The queue will fill after this allocation, so set
862 * it as full, and mark this process as "batching".
863 * This process will be allowed to complete a batch of
864 * requests, others will be blocked.
866 if (!blk_queue_full(q
, is_sync
)) {
867 ioc_set_batching(q
, ioc
);
868 blk_set_queue_full(q
, is_sync
);
870 if (may_queue
!= ELV_MQUEUE_MUST
871 && !ioc_batching(q
, ioc
)) {
873 * The queue is full and the allocating
874 * process is not a "batcher", and not
875 * exempted by the IO scheduler
881 blk_set_queue_congested(q
, is_sync
);
885 * Only allow batching queuers to allocate up to 50% over the defined
886 * limit of requests, otherwise we could have thousands of requests
887 * allocated with any setting of ->nr_requests
889 if (rl
->count
[is_sync
] >= (3 * q
->nr_requests
/ 2))
892 rl
->count
[is_sync
]++;
893 rl
->starved
[is_sync
] = 0;
896 * Decide whether the new request will be managed by elevator. If
897 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
898 * prevent the current elevator from being destroyed until the new
899 * request is freed. This guarantees icq's won't be destroyed and
900 * makes creating new ones safe.
902 * Also, lookup icq while holding queue_lock. If it doesn't exist,
903 * it will be created after releasing queue_lock.
905 if (blk_rq_should_init_elevator(bio
) && !blk_queue_bypass(q
)) {
906 rw_flags
|= REQ_ELVPRIV
;
908 if (et
->icq_cache
&& ioc
)
909 icq
= ioc_lookup_icq(ioc
, q
);
912 if (blk_queue_io_stat(q
))
913 rw_flags
|= REQ_IO_STAT
;
914 spin_unlock_irq(q
->queue_lock
);
916 /* create icq if missing */
917 if ((rw_flags
& REQ_ELVPRIV
) && unlikely(et
->icq_cache
&& !icq
)) {
918 icq
= ioc_create_icq(q
, gfp_mask
);
923 rq
= blk_alloc_request(q
, icq
, rw_flags
, gfp_mask
);
928 * Allocation failed presumably due to memory. Undo anything
929 * we might have messed up.
931 * Allocating task should really be put onto the front of the
932 * wait queue, but this is pretty rare.
934 spin_lock_irq(q
->queue_lock
);
935 freed_request(q
, rw_flags
);
938 * in the very unlikely event that allocation failed and no
939 * requests for this direction was pending, mark us starved
940 * so that freeing of a request in the other direction will
941 * notice us. another possible fix would be to split the
942 * rq mempool into READ and WRITE
945 if (unlikely(rl
->count
[is_sync
] == 0))
946 rl
->starved
[is_sync
] = 1;
952 * ioc may be NULL here, and ioc_batching will be false. That's
953 * OK, if the queue is under the request limit then requests need
954 * not count toward the nr_batch_requests limit. There will always
955 * be some limit enforced by BLK_BATCH_TIME.
957 if (ioc_batching(q
, ioc
))
958 ioc
->nr_batch_requests
--;
960 trace_block_getrq(q
, bio
, rw_flags
& 1);
966 * get_request_wait - get a free request with retry
967 * @q: request_queue to allocate request from
968 * @rw_flags: RW and SYNC flags
969 * @bio: bio to allocate request for (can be %NULL)
971 * Get a free request from @q. This function keeps retrying under memory
972 * pressure and fails iff @q is dead.
974 * Must be callled with @q->queue_lock held and,
975 * Returns %NULL on failure, with @q->queue_lock held.
976 * Returns !%NULL on success, with @q->queue_lock *not held*.
978 static struct request
*get_request_wait(struct request_queue
*q
, int rw_flags
,
981 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
984 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
987 struct request_list
*rl
= &q
->rq
;
989 if (unlikely(blk_queue_dead(q
)))
992 prepare_to_wait_exclusive(&rl
->wait
[is_sync
], &wait
,
993 TASK_UNINTERRUPTIBLE
);
995 trace_block_sleeprq(q
, bio
, rw_flags
& 1);
997 spin_unlock_irq(q
->queue_lock
);
1001 * After sleeping, we become a "batching" process and
1002 * will be able to allocate at least one request, and
1003 * up to a big batch of them for a small period time.
1004 * See ioc_batching, ioc_set_batching
1006 create_io_context(current
, GFP_NOIO
, q
->node
);
1007 ioc_set_batching(q
, current
->io_context
);
1009 spin_lock_irq(q
->queue_lock
);
1010 finish_wait(&rl
->wait
[is_sync
], &wait
);
1012 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
1018 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
1022 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
1024 spin_lock_irq(q
->queue_lock
);
1025 if (gfp_mask
& __GFP_WAIT
)
1026 rq
= get_request_wait(q
, rw
, NULL
);
1028 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
1030 spin_unlock_irq(q
->queue_lock
);
1031 /* q->queue_lock is unlocked at this point */
1035 EXPORT_SYMBOL(blk_get_request
);
1038 * blk_make_request - given a bio, allocate a corresponding struct request.
1039 * @q: target request queue
1040 * @bio: The bio describing the memory mappings that will be submitted for IO.
1041 * It may be a chained-bio properly constructed by block/bio layer.
1042 * @gfp_mask: gfp flags to be used for memory allocation
1044 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1045 * type commands. Where the struct request needs to be farther initialized by
1046 * the caller. It is passed a &struct bio, which describes the memory info of
1049 * The caller of blk_make_request must make sure that bi_io_vec
1050 * are set to describe the memory buffers. That bio_data_dir() will return
1051 * the needed direction of the request. (And all bio's in the passed bio-chain
1052 * are properly set accordingly)
1054 * If called under none-sleepable conditions, mapped bio buffers must not
1055 * need bouncing, by calling the appropriate masked or flagged allocator,
1056 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1059 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1060 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1061 * anything but the first bio in the chain. Otherwise you risk waiting for IO
1062 * completion of a bio that hasn't been submitted yet, thus resulting in a
1063 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1064 * of bio_alloc(), as that avoids the mempool deadlock.
1065 * If possible a big IO should be split into smaller parts when allocation
1066 * fails. Partial allocation should not be an error, or you risk a live-lock.
1068 struct request
*blk_make_request(struct request_queue
*q
, struct bio
*bio
,
1071 struct request
*rq
= blk_get_request(q
, bio_data_dir(bio
), gfp_mask
);
1074 return ERR_PTR(-ENOMEM
);
1077 struct bio
*bounce_bio
= bio
;
1080 blk_queue_bounce(q
, &bounce_bio
);
1081 ret
= blk_rq_append_bio(q
, rq
, bounce_bio
);
1082 if (unlikely(ret
)) {
1083 blk_put_request(rq
);
1084 return ERR_PTR(ret
);
1090 EXPORT_SYMBOL(blk_make_request
);
1093 * blk_requeue_request - put a request back on queue
1094 * @q: request queue where request should be inserted
1095 * @rq: request to be inserted
1098 * Drivers often keep queueing requests until the hardware cannot accept
1099 * more, when that condition happens we need to put the request back
1100 * on the queue. Must be called with queue lock held.
1102 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
1104 blk_delete_timer(rq
);
1105 blk_clear_rq_complete(rq
);
1106 trace_block_rq_requeue(q
, rq
);
1108 if (blk_rq_tagged(rq
))
1109 blk_queue_end_tag(q
, rq
);
1111 BUG_ON(blk_queued_rq(rq
));
1113 elv_requeue_request(q
, rq
);
1115 EXPORT_SYMBOL(blk_requeue_request
);
1117 static void add_acct_request(struct request_queue
*q
, struct request
*rq
,
1120 drive_stat_acct(rq
, 1);
1121 __elv_add_request(q
, rq
, where
);
1124 static void part_round_stats_single(int cpu
, struct hd_struct
*part
,
1127 if (now
== part
->stamp
)
1130 if (part_in_flight(part
)) {
1131 __part_stat_add(cpu
, part
, time_in_queue
,
1132 part_in_flight(part
) * (now
- part
->stamp
));
1133 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
1139 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1140 * @cpu: cpu number for stats access
1141 * @part: target partition
1143 * The average IO queue length and utilisation statistics are maintained
1144 * by observing the current state of the queue length and the amount of
1145 * time it has been in this state for.
1147 * Normally, that accounting is done on IO completion, but that can result
1148 * in more than a second's worth of IO being accounted for within any one
1149 * second, leading to >100% utilisation. To deal with that, we call this
1150 * function to do a round-off before returning the results when reading
1151 * /proc/diskstats. This accounts immediately for all queue usage up to
1152 * the current jiffies and restarts the counters again.
1154 void part_round_stats(int cpu
, struct hd_struct
*part
)
1156 unsigned long now
= jiffies
;
1159 part_round_stats_single(cpu
, &part_to_disk(part
)->part0
, now
);
1160 part_round_stats_single(cpu
, part
, now
);
1162 EXPORT_SYMBOL_GPL(part_round_stats
);
1165 * queue lock must be held
1167 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1171 if (unlikely(--req
->ref_count
))
1174 elv_completed_request(q
, req
);
1176 /* this is a bio leak */
1177 WARN_ON(req
->bio
!= NULL
);
1180 * Request may not have originated from ll_rw_blk. if not,
1181 * it didn't come out of our reserved rq pools
1183 if (req
->cmd_flags
& REQ_ALLOCED
) {
1184 unsigned int flags
= req
->cmd_flags
;
1186 BUG_ON(!list_empty(&req
->queuelist
));
1187 BUG_ON(!hlist_unhashed(&req
->hash
));
1189 blk_free_request(q
, req
);
1190 freed_request(q
, flags
);
1193 EXPORT_SYMBOL_GPL(__blk_put_request
);
1195 void blk_put_request(struct request
*req
)
1197 unsigned long flags
;
1198 struct request_queue
*q
= req
->q
;
1200 spin_lock_irqsave(q
->queue_lock
, flags
);
1201 __blk_put_request(q
, req
);
1202 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1204 EXPORT_SYMBOL(blk_put_request
);
1207 * blk_add_request_payload - add a payload to a request
1208 * @rq: request to update
1209 * @page: page backing the payload
1210 * @len: length of the payload.
1212 * This allows to later add a payload to an already submitted request by
1213 * a block driver. The driver needs to take care of freeing the payload
1216 * Note that this is a quite horrible hack and nothing but handling of
1217 * discard requests should ever use it.
1219 void blk_add_request_payload(struct request
*rq
, struct page
*page
,
1222 struct bio
*bio
= rq
->bio
;
1224 bio
->bi_io_vec
->bv_page
= page
;
1225 bio
->bi_io_vec
->bv_offset
= 0;
1226 bio
->bi_io_vec
->bv_len
= len
;
1230 bio
->bi_phys_segments
= 1;
1232 rq
->__data_len
= rq
->resid_len
= len
;
1233 rq
->nr_phys_segments
= 1;
1234 rq
->buffer
= bio_data(bio
);
1236 EXPORT_SYMBOL_GPL(blk_add_request_payload
);
1238 static bool bio_attempt_back_merge(struct request_queue
*q
, struct request
*req
,
1241 const int ff
= bio
->bi_rw
& REQ_FAILFAST_MASK
;
1243 if (!ll_back_merge_fn(q
, req
, bio
))
1246 trace_block_bio_backmerge(q
, bio
);
1248 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1249 blk_rq_set_mixed_merge(req
);
1251 req
->biotail
->bi_next
= bio
;
1253 req
->__data_len
+= bio
->bi_size
;
1254 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1256 drive_stat_acct(req
, 0);
1260 static bool bio_attempt_front_merge(struct request_queue
*q
,
1261 struct request
*req
, struct bio
*bio
)
1263 const int ff
= bio
->bi_rw
& REQ_FAILFAST_MASK
;
1265 if (!ll_front_merge_fn(q
, req
, bio
))
1268 trace_block_bio_frontmerge(q
, bio
);
1270 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1271 blk_rq_set_mixed_merge(req
);
1273 bio
->bi_next
= req
->bio
;
1277 * may not be valid. if the low level driver said
1278 * it didn't need a bounce buffer then it better
1279 * not touch req->buffer either...
1281 req
->buffer
= bio_data(bio
);
1282 req
->__sector
= bio
->bi_sector
;
1283 req
->__data_len
+= bio
->bi_size
;
1284 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1286 drive_stat_acct(req
, 0);
1291 * attempt_plug_merge - try to merge with %current's plugged list
1292 * @q: request_queue new bio is being queued at
1293 * @bio: new bio being queued
1294 * @request_count: out parameter for number of traversed plugged requests
1296 * Determine whether @bio being queued on @q can be merged with a request
1297 * on %current's plugged list. Returns %true if merge was successful,
1300 * Plugging coalesces IOs from the same issuer for the same purpose without
1301 * going through @q->queue_lock. As such it's more of an issuing mechanism
1302 * than scheduling, and the request, while may have elvpriv data, is not
1303 * added on the elevator at this point. In addition, we don't have
1304 * reliable access to the elevator outside queue lock. Only check basic
1305 * merging parameters without querying the elevator.
1307 static bool attempt_plug_merge(struct request_queue
*q
, struct bio
*bio
,
1308 unsigned int *request_count
)
1310 struct blk_plug
*plug
;
1314 plug
= current
->plug
;
1319 list_for_each_entry_reverse(rq
, &plug
->list
, queuelist
) {
1324 if (rq
->q
!= q
|| !blk_rq_merge_ok(rq
, bio
))
1327 el_ret
= blk_try_merge(rq
, bio
);
1328 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1329 ret
= bio_attempt_back_merge(q
, rq
, bio
);
1332 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1333 ret
= bio_attempt_front_merge(q
, rq
, bio
);
1342 void init_request_from_bio(struct request
*req
, struct bio
*bio
)
1344 req
->cmd_type
= REQ_TYPE_FS
;
1346 req
->cmd_flags
|= bio
->bi_rw
& REQ_COMMON_MASK
;
1347 if (bio
->bi_rw
& REQ_RAHEAD
)
1348 req
->cmd_flags
|= REQ_FAILFAST_MASK
;
1351 req
->__sector
= bio
->bi_sector
;
1352 req
->ioprio
= bio_prio(bio
);
1353 blk_rq_bio_prep(req
->q
, req
, bio
);
1356 void blk_queue_bio(struct request_queue
*q
, struct bio
*bio
)
1358 const bool sync
= !!(bio
->bi_rw
& REQ_SYNC
);
1359 struct blk_plug
*plug
;
1360 int el_ret
, rw_flags
, where
= ELEVATOR_INSERT_SORT
;
1361 struct request
*req
;
1362 unsigned int request_count
= 0;
1365 * low level driver can indicate that it wants pages above a
1366 * certain limit bounced to low memory (ie for highmem, or even
1367 * ISA dma in theory)
1369 blk_queue_bounce(q
, &bio
);
1371 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) {
1372 spin_lock_irq(q
->queue_lock
);
1373 where
= ELEVATOR_INSERT_FLUSH
;
1378 * Check if we can merge with the plugged list before grabbing
1381 if (attempt_plug_merge(q
, bio
, &request_count
))
1384 spin_lock_irq(q
->queue_lock
);
1386 el_ret
= elv_merge(q
, &req
, bio
);
1387 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1388 if (bio_attempt_back_merge(q
, req
, bio
)) {
1389 elv_bio_merged(q
, req
, bio
);
1390 if (!attempt_back_merge(q
, req
))
1391 elv_merged_request(q
, req
, el_ret
);
1394 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1395 if (bio_attempt_front_merge(q
, req
, bio
)) {
1396 elv_bio_merged(q
, req
, bio
);
1397 if (!attempt_front_merge(q
, req
))
1398 elv_merged_request(q
, req
, el_ret
);
1405 * This sync check and mask will be re-done in init_request_from_bio(),
1406 * but we need to set it earlier to expose the sync flag to the
1407 * rq allocator and io schedulers.
1409 rw_flags
= bio_data_dir(bio
);
1411 rw_flags
|= REQ_SYNC
;
1414 * Grab a free request. This is might sleep but can not fail.
1415 * Returns with the queue unlocked.
1417 req
= get_request_wait(q
, rw_flags
, bio
);
1418 if (unlikely(!req
)) {
1419 bio_endio(bio
, -ENODEV
); /* @q is dead */
1424 * After dropping the lock and possibly sleeping here, our request
1425 * may now be mergeable after it had proven unmergeable (above).
1426 * We don't worry about that case for efficiency. It won't happen
1427 * often, and the elevators are able to handle it.
1429 init_request_from_bio(req
, bio
);
1431 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
))
1432 req
->cpu
= raw_smp_processor_id();
1434 plug
= current
->plug
;
1437 * If this is the first request added after a plug, fire
1438 * of a plug trace. If others have been added before, check
1439 * if we have multiple devices in this plug. If so, make a
1440 * note to sort the list before dispatch.
1442 if (list_empty(&plug
->list
))
1443 trace_block_plug(q
);
1445 if (!plug
->should_sort
) {
1446 struct request
*__rq
;
1448 __rq
= list_entry_rq(plug
->list
.prev
);
1450 plug
->should_sort
= 1;
1452 if (request_count
>= BLK_MAX_REQUEST_COUNT
) {
1453 blk_flush_plug_list(plug
, false);
1454 trace_block_plug(q
);
1457 list_add_tail(&req
->queuelist
, &plug
->list
);
1458 drive_stat_acct(req
, 1);
1460 spin_lock_irq(q
->queue_lock
);
1461 add_acct_request(q
, req
, where
);
1464 spin_unlock_irq(q
->queue_lock
);
1467 EXPORT_SYMBOL_GPL(blk_queue_bio
); /* for device mapper only */
1470 * If bio->bi_dev is a partition, remap the location
1472 static inline void blk_partition_remap(struct bio
*bio
)
1474 struct block_device
*bdev
= bio
->bi_bdev
;
1476 if (bio_sectors(bio
) && bdev
!= bdev
->bd_contains
) {
1477 struct hd_struct
*p
= bdev
->bd_part
;
1479 bio
->bi_sector
+= p
->start_sect
;
1480 bio
->bi_bdev
= bdev
->bd_contains
;
1482 trace_block_bio_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
1484 bio
->bi_sector
- p
->start_sect
);
1488 static void handle_bad_sector(struct bio
*bio
)
1490 char b
[BDEVNAME_SIZE
];
1492 printk(KERN_INFO
"attempt to access beyond end of device\n");
1493 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
1494 bdevname(bio
->bi_bdev
, b
),
1496 (unsigned long long)bio
->bi_sector
+ bio_sectors(bio
),
1497 (long long)(i_size_read(bio
->bi_bdev
->bd_inode
) >> 9));
1499 set_bit(BIO_EOF
, &bio
->bi_flags
);
1502 #ifdef CONFIG_FAIL_MAKE_REQUEST
1504 static DECLARE_FAULT_ATTR(fail_make_request
);
1506 static int __init
setup_fail_make_request(char *str
)
1508 return setup_fault_attr(&fail_make_request
, str
);
1510 __setup("fail_make_request=", setup_fail_make_request
);
1512 static bool should_fail_request(struct hd_struct
*part
, unsigned int bytes
)
1514 return part
->make_it_fail
&& should_fail(&fail_make_request
, bytes
);
1517 static int __init
fail_make_request_debugfs(void)
1519 struct dentry
*dir
= fault_create_debugfs_attr("fail_make_request",
1520 NULL
, &fail_make_request
);
1522 return IS_ERR(dir
) ? PTR_ERR(dir
) : 0;
1525 late_initcall(fail_make_request_debugfs
);
1527 #else /* CONFIG_FAIL_MAKE_REQUEST */
1529 static inline bool should_fail_request(struct hd_struct
*part
,
1535 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1538 * Check whether this bio extends beyond the end of the device.
1540 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
1547 /* Test device or partition size, when known. */
1548 maxsector
= i_size_read(bio
->bi_bdev
->bd_inode
) >> 9;
1550 sector_t sector
= bio
->bi_sector
;
1552 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
1554 * This may well happen - the kernel calls bread()
1555 * without checking the size of the device, e.g., when
1556 * mounting a device.
1558 handle_bad_sector(bio
);
1566 static noinline_for_stack
bool
1567 generic_make_request_checks(struct bio
*bio
)
1569 struct request_queue
*q
;
1570 int nr_sectors
= bio_sectors(bio
);
1572 char b
[BDEVNAME_SIZE
];
1573 struct hd_struct
*part
;
1577 if (bio_check_eod(bio
, nr_sectors
))
1580 q
= bdev_get_queue(bio
->bi_bdev
);
1583 "generic_make_request: Trying to access "
1584 "nonexistent block-device %s (%Lu)\n",
1585 bdevname(bio
->bi_bdev
, b
),
1586 (long long) bio
->bi_sector
);
1590 if (unlikely(!(bio
->bi_rw
& REQ_DISCARD
) &&
1591 nr_sectors
> queue_max_hw_sectors(q
))) {
1592 printk(KERN_ERR
"bio too big device %s (%u > %u)\n",
1593 bdevname(bio
->bi_bdev
, b
),
1595 queue_max_hw_sectors(q
));
1599 part
= bio
->bi_bdev
->bd_part
;
1600 if (should_fail_request(part
, bio
->bi_size
) ||
1601 should_fail_request(&part_to_disk(part
)->part0
,
1606 * If this device has partitions, remap block n
1607 * of partition p to block n+start(p) of the disk.
1609 blk_partition_remap(bio
);
1611 if (bio_integrity_enabled(bio
) && bio_integrity_prep(bio
))
1614 if (bio_check_eod(bio
, nr_sectors
))
1618 * Filter flush bio's early so that make_request based
1619 * drivers without flush support don't have to worry
1622 if ((bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) && !q
->flush_flags
) {
1623 bio
->bi_rw
&= ~(REQ_FLUSH
| REQ_FUA
);
1630 if ((bio
->bi_rw
& REQ_DISCARD
) &&
1631 (!blk_queue_discard(q
) ||
1632 ((bio
->bi_rw
& REQ_SECURE
) &&
1633 !blk_queue_secdiscard(q
)))) {
1638 if (blk_throtl_bio(q
, bio
))
1639 return false; /* throttled, will be resubmitted later */
1641 trace_block_bio_queue(q
, bio
);
1645 bio_endio(bio
, err
);
1650 * generic_make_request - hand a buffer to its device driver for I/O
1651 * @bio: The bio describing the location in memory and on the device.
1653 * generic_make_request() is used to make I/O requests of block
1654 * devices. It is passed a &struct bio, which describes the I/O that needs
1657 * generic_make_request() does not return any status. The
1658 * success/failure status of the request, along with notification of
1659 * completion, is delivered asynchronously through the bio->bi_end_io
1660 * function described (one day) else where.
1662 * The caller of generic_make_request must make sure that bi_io_vec
1663 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1664 * set to describe the device address, and the
1665 * bi_end_io and optionally bi_private are set to describe how
1666 * completion notification should be signaled.
1668 * generic_make_request and the drivers it calls may use bi_next if this
1669 * bio happens to be merged with someone else, and may resubmit the bio to
1670 * a lower device by calling into generic_make_request recursively, which
1671 * means the bio should NOT be touched after the call to ->make_request_fn.
1673 void generic_make_request(struct bio
*bio
)
1675 struct bio_list bio_list_on_stack
;
1677 if (!generic_make_request_checks(bio
))
1681 * We only want one ->make_request_fn to be active at a time, else
1682 * stack usage with stacked devices could be a problem. So use
1683 * current->bio_list to keep a list of requests submited by a
1684 * make_request_fn function. current->bio_list is also used as a
1685 * flag to say if generic_make_request is currently active in this
1686 * task or not. If it is NULL, then no make_request is active. If
1687 * it is non-NULL, then a make_request is active, and new requests
1688 * should be added at the tail
1690 if (current
->bio_list
) {
1691 bio_list_add(current
->bio_list
, bio
);
1695 /* following loop may be a bit non-obvious, and so deserves some
1697 * Before entering the loop, bio->bi_next is NULL (as all callers
1698 * ensure that) so we have a list with a single bio.
1699 * We pretend that we have just taken it off a longer list, so
1700 * we assign bio_list to a pointer to the bio_list_on_stack,
1701 * thus initialising the bio_list of new bios to be
1702 * added. ->make_request() may indeed add some more bios
1703 * through a recursive call to generic_make_request. If it
1704 * did, we find a non-NULL value in bio_list and re-enter the loop
1705 * from the top. In this case we really did just take the bio
1706 * of the top of the list (no pretending) and so remove it from
1707 * bio_list, and call into ->make_request() again.
1709 BUG_ON(bio
->bi_next
);
1710 bio_list_init(&bio_list_on_stack
);
1711 current
->bio_list
= &bio_list_on_stack
;
1713 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
1715 q
->make_request_fn(q
, bio
);
1717 bio
= bio_list_pop(current
->bio_list
);
1719 current
->bio_list
= NULL
; /* deactivate */
1721 EXPORT_SYMBOL(generic_make_request
);
1724 * submit_bio - submit a bio to the block device layer for I/O
1725 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1726 * @bio: The &struct bio which describes the I/O
1728 * submit_bio() is very similar in purpose to generic_make_request(), and
1729 * uses that function to do most of the work. Both are fairly rough
1730 * interfaces; @bio must be presetup and ready for I/O.
1733 void submit_bio(int rw
, struct bio
*bio
)
1735 int count
= bio_sectors(bio
);
1740 * If it's a regular read/write or a barrier with data attached,
1741 * go through the normal accounting stuff before submission.
1743 if (bio_has_data(bio
) && !(rw
& REQ_DISCARD
)) {
1745 count_vm_events(PGPGOUT
, count
);
1747 task_io_account_read(bio
->bi_size
);
1748 count_vm_events(PGPGIN
, count
);
1751 if (unlikely(block_dump
)) {
1752 char b
[BDEVNAME_SIZE
];
1753 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s (%u sectors)\n",
1754 current
->comm
, task_pid_nr(current
),
1755 (rw
& WRITE
) ? "WRITE" : "READ",
1756 (unsigned long long)bio
->bi_sector
,
1757 bdevname(bio
->bi_bdev
, b
),
1762 generic_make_request(bio
);
1764 EXPORT_SYMBOL(submit_bio
);
1767 * blk_rq_check_limits - Helper function to check a request for the queue limit
1769 * @rq: the request being checked
1772 * @rq may have been made based on weaker limitations of upper-level queues
1773 * in request stacking drivers, and it may violate the limitation of @q.
1774 * Since the block layer and the underlying device driver trust @rq
1775 * after it is inserted to @q, it should be checked against @q before
1776 * the insertion using this generic function.
1778 * This function should also be useful for request stacking drivers
1779 * in some cases below, so export this function.
1780 * Request stacking drivers like request-based dm may change the queue
1781 * limits while requests are in the queue (e.g. dm's table swapping).
1782 * Such request stacking drivers should check those requests agaist
1783 * the new queue limits again when they dispatch those requests,
1784 * although such checkings are also done against the old queue limits
1785 * when submitting requests.
1787 int blk_rq_check_limits(struct request_queue
*q
, struct request
*rq
)
1789 if (rq
->cmd_flags
& REQ_DISCARD
)
1792 if (blk_rq_sectors(rq
) > queue_max_sectors(q
) ||
1793 blk_rq_bytes(rq
) > queue_max_hw_sectors(q
) << 9) {
1794 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
1799 * queue's settings related to segment counting like q->bounce_pfn
1800 * may differ from that of other stacking queues.
1801 * Recalculate it to check the request correctly on this queue's
1804 blk_recalc_rq_segments(rq
);
1805 if (rq
->nr_phys_segments
> queue_max_segments(q
)) {
1806 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
1812 EXPORT_SYMBOL_GPL(blk_rq_check_limits
);
1815 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1816 * @q: the queue to submit the request
1817 * @rq: the request being queued
1819 int blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
1821 unsigned long flags
;
1822 int where
= ELEVATOR_INSERT_BACK
;
1824 if (blk_rq_check_limits(q
, rq
))
1828 should_fail_request(&rq
->rq_disk
->part0
, blk_rq_bytes(rq
)))
1831 spin_lock_irqsave(q
->queue_lock
, flags
);
1832 if (unlikely(blk_queue_dead(q
))) {
1833 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1838 * Submitting request must be dequeued before calling this function
1839 * because it will be linked to another request_queue
1841 BUG_ON(blk_queued_rq(rq
));
1843 if (rq
->cmd_flags
& (REQ_FLUSH
|REQ_FUA
))
1844 where
= ELEVATOR_INSERT_FLUSH
;
1846 add_acct_request(q
, rq
, where
);
1847 if (where
== ELEVATOR_INSERT_FLUSH
)
1849 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1853 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
1856 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1857 * @rq: request to examine
1860 * A request could be merge of IOs which require different failure
1861 * handling. This function determines the number of bytes which
1862 * can be failed from the beginning of the request without
1863 * crossing into area which need to be retried further.
1866 * The number of bytes to fail.
1869 * queue_lock must be held.
1871 unsigned int blk_rq_err_bytes(const struct request
*rq
)
1873 unsigned int ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
1874 unsigned int bytes
= 0;
1877 if (!(rq
->cmd_flags
& REQ_MIXED_MERGE
))
1878 return blk_rq_bytes(rq
);
1881 * Currently the only 'mixing' which can happen is between
1882 * different fastfail types. We can safely fail portions
1883 * which have all the failfast bits that the first one has -
1884 * the ones which are at least as eager to fail as the first
1887 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
1888 if ((bio
->bi_rw
& ff
) != ff
)
1890 bytes
+= bio
->bi_size
;
1893 /* this could lead to infinite loop */
1894 BUG_ON(blk_rq_bytes(rq
) && !bytes
);
1897 EXPORT_SYMBOL_GPL(blk_rq_err_bytes
);
1899 static void blk_account_io_completion(struct request
*req
, unsigned int bytes
)
1901 if (blk_do_io_stat(req
)) {
1902 const int rw
= rq_data_dir(req
);
1903 struct hd_struct
*part
;
1906 cpu
= part_stat_lock();
1908 part_stat_add(cpu
, part
, sectors
[rw
], bytes
>> 9);
1913 static void blk_account_io_done(struct request
*req
)
1916 * Account IO completion. flush_rq isn't accounted as a
1917 * normal IO on queueing nor completion. Accounting the
1918 * containing request is enough.
1920 if (blk_do_io_stat(req
) && !(req
->cmd_flags
& REQ_FLUSH_SEQ
)) {
1921 unsigned long duration
= jiffies
- req
->start_time
;
1922 const int rw
= rq_data_dir(req
);
1923 struct hd_struct
*part
;
1926 cpu
= part_stat_lock();
1929 part_stat_inc(cpu
, part
, ios
[rw
]);
1930 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
1931 part_round_stats(cpu
, part
);
1932 part_dec_in_flight(part
, rw
);
1934 hd_struct_put(part
);
1940 * blk_peek_request - peek at the top of a request queue
1941 * @q: request queue to peek at
1944 * Return the request at the top of @q. The returned request
1945 * should be started using blk_start_request() before LLD starts
1949 * Pointer to the request at the top of @q if available. Null
1953 * queue_lock must be held.
1955 struct request
*blk_peek_request(struct request_queue
*q
)
1960 while ((rq
= __elv_next_request(q
)) != NULL
) {
1961 if (!(rq
->cmd_flags
& REQ_STARTED
)) {
1963 * This is the first time the device driver
1964 * sees this request (possibly after
1965 * requeueing). Notify IO scheduler.
1967 if (rq
->cmd_flags
& REQ_SORTED
)
1968 elv_activate_rq(q
, rq
);
1971 * just mark as started even if we don't start
1972 * it, a request that has been delayed should
1973 * not be passed by new incoming requests
1975 rq
->cmd_flags
|= REQ_STARTED
;
1976 trace_block_rq_issue(q
, rq
);
1979 if (!q
->boundary_rq
|| q
->boundary_rq
== rq
) {
1980 q
->end_sector
= rq_end_sector(rq
);
1981 q
->boundary_rq
= NULL
;
1984 if (rq
->cmd_flags
& REQ_DONTPREP
)
1987 if (q
->dma_drain_size
&& blk_rq_bytes(rq
)) {
1989 * make sure space for the drain appears we
1990 * know we can do this because max_hw_segments
1991 * has been adjusted to be one fewer than the
1994 rq
->nr_phys_segments
++;
2000 ret
= q
->prep_rq_fn(q
, rq
);
2001 if (ret
== BLKPREP_OK
) {
2003 } else if (ret
== BLKPREP_DEFER
) {
2005 * the request may have been (partially) prepped.
2006 * we need to keep this request in the front to
2007 * avoid resource deadlock. REQ_STARTED will
2008 * prevent other fs requests from passing this one.
2010 if (q
->dma_drain_size
&& blk_rq_bytes(rq
) &&
2011 !(rq
->cmd_flags
& REQ_DONTPREP
)) {
2013 * remove the space for the drain we added
2014 * so that we don't add it again
2016 --rq
->nr_phys_segments
;
2021 } else if (ret
== BLKPREP_KILL
) {
2022 rq
->cmd_flags
|= REQ_QUIET
;
2024 * Mark this request as started so we don't trigger
2025 * any debug logic in the end I/O path.
2027 blk_start_request(rq
);
2028 __blk_end_request_all(rq
, -EIO
);
2030 printk(KERN_ERR
"%s: bad return=%d\n", __func__
, ret
);
2037 EXPORT_SYMBOL(blk_peek_request
);
2039 void blk_dequeue_request(struct request
*rq
)
2041 struct request_queue
*q
= rq
->q
;
2043 BUG_ON(list_empty(&rq
->queuelist
));
2044 BUG_ON(ELV_ON_HASH(rq
));
2046 list_del_init(&rq
->queuelist
);
2049 * the time frame between a request being removed from the lists
2050 * and to it is freed is accounted as io that is in progress at
2053 if (blk_account_rq(rq
)) {
2054 q
->in_flight
[rq_is_sync(rq
)]++;
2055 set_io_start_time_ns(rq
);
2060 * blk_start_request - start request processing on the driver
2061 * @req: request to dequeue
2064 * Dequeue @req and start timeout timer on it. This hands off the
2065 * request to the driver.
2067 * Block internal functions which don't want to start timer should
2068 * call blk_dequeue_request().
2071 * queue_lock must be held.
2073 void blk_start_request(struct request
*req
)
2075 blk_dequeue_request(req
);
2078 * We are now handing the request to the hardware, initialize
2079 * resid_len to full count and add the timeout handler.
2081 req
->resid_len
= blk_rq_bytes(req
);
2082 if (unlikely(blk_bidi_rq(req
)))
2083 req
->next_rq
->resid_len
= blk_rq_bytes(req
->next_rq
);
2087 EXPORT_SYMBOL(blk_start_request
);
2090 * blk_fetch_request - fetch a request from a request queue
2091 * @q: request queue to fetch a request from
2094 * Return the request at the top of @q. The request is started on
2095 * return and LLD can start processing it immediately.
2098 * Pointer to the request at the top of @q if available. Null
2102 * queue_lock must be held.
2104 struct request
*blk_fetch_request(struct request_queue
*q
)
2108 rq
= blk_peek_request(q
);
2110 blk_start_request(rq
);
2113 EXPORT_SYMBOL(blk_fetch_request
);
2116 * blk_update_request - Special helper function for request stacking drivers
2117 * @req: the request being processed
2118 * @error: %0 for success, < %0 for error
2119 * @nr_bytes: number of bytes to complete @req
2122 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2123 * the request structure even if @req doesn't have leftover.
2124 * If @req has leftover, sets it up for the next range of segments.
2126 * This special helper function is only for request stacking drivers
2127 * (e.g. request-based dm) so that they can handle partial completion.
2128 * Actual device drivers should use blk_end_request instead.
2130 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2131 * %false return from this function.
2134 * %false - this request doesn't have any more data
2135 * %true - this request has more data
2137 bool blk_update_request(struct request
*req
, int error
, unsigned int nr_bytes
)
2139 int total_bytes
, bio_nbytes
, next_idx
= 0;
2145 trace_block_rq_complete(req
->q
, req
);
2148 * For fs requests, rq is just carrier of independent bio's
2149 * and each partial completion should be handled separately.
2150 * Reset per-request error on each partial completion.
2152 * TODO: tj: This is too subtle. It would be better to let
2153 * low level drivers do what they see fit.
2155 if (req
->cmd_type
== REQ_TYPE_FS
)
2158 if (error
&& req
->cmd_type
== REQ_TYPE_FS
&&
2159 !(req
->cmd_flags
& REQ_QUIET
)) {
2164 error_type
= "recoverable transport";
2167 error_type
= "critical target";
2170 error_type
= "critical nexus";
2177 printk(KERN_ERR
"end_request: %s error, dev %s, sector %llu\n",
2178 error_type
, req
->rq_disk
? req
->rq_disk
->disk_name
: "?",
2179 (unsigned long long)blk_rq_pos(req
));
2182 blk_account_io_completion(req
, nr_bytes
);
2184 total_bytes
= bio_nbytes
= 0;
2185 while ((bio
= req
->bio
) != NULL
) {
2188 if (nr_bytes
>= bio
->bi_size
) {
2189 req
->bio
= bio
->bi_next
;
2190 nbytes
= bio
->bi_size
;
2191 req_bio_endio(req
, bio
, nbytes
, error
);
2195 int idx
= bio
->bi_idx
+ next_idx
;
2197 if (unlikely(idx
>= bio
->bi_vcnt
)) {
2198 blk_dump_rq_flags(req
, "__end_that");
2199 printk(KERN_ERR
"%s: bio idx %d >= vcnt %d\n",
2200 __func__
, idx
, bio
->bi_vcnt
);
2204 nbytes
= bio_iovec_idx(bio
, idx
)->bv_len
;
2205 BIO_BUG_ON(nbytes
> bio
->bi_size
);
2208 * not a complete bvec done
2210 if (unlikely(nbytes
> nr_bytes
)) {
2211 bio_nbytes
+= nr_bytes
;
2212 total_bytes
+= nr_bytes
;
2217 * advance to the next vector
2220 bio_nbytes
+= nbytes
;
2223 total_bytes
+= nbytes
;
2229 * end more in this run, or just return 'not-done'
2231 if (unlikely(nr_bytes
<= 0))
2241 * Reset counters so that the request stacking driver
2242 * can find how many bytes remain in the request
2245 req
->__data_len
= 0;
2250 * if the request wasn't completed, update state
2253 req_bio_endio(req
, bio
, bio_nbytes
, error
);
2254 bio
->bi_idx
+= next_idx
;
2255 bio_iovec(bio
)->bv_offset
+= nr_bytes
;
2256 bio_iovec(bio
)->bv_len
-= nr_bytes
;
2259 req
->__data_len
-= total_bytes
;
2260 req
->buffer
= bio_data(req
->bio
);
2262 /* update sector only for requests with clear definition of sector */
2263 if (req
->cmd_type
== REQ_TYPE_FS
|| (req
->cmd_flags
& REQ_DISCARD
))
2264 req
->__sector
+= total_bytes
>> 9;
2266 /* mixed attributes always follow the first bio */
2267 if (req
->cmd_flags
& REQ_MIXED_MERGE
) {
2268 req
->cmd_flags
&= ~REQ_FAILFAST_MASK
;
2269 req
->cmd_flags
|= req
->bio
->bi_rw
& REQ_FAILFAST_MASK
;
2273 * If total number of sectors is less than the first segment
2274 * size, something has gone terribly wrong.
2276 if (blk_rq_bytes(req
) < blk_rq_cur_bytes(req
)) {
2277 blk_dump_rq_flags(req
, "request botched");
2278 req
->__data_len
= blk_rq_cur_bytes(req
);
2281 /* recalculate the number of segments */
2282 blk_recalc_rq_segments(req
);
2286 EXPORT_SYMBOL_GPL(blk_update_request
);
2288 static bool blk_update_bidi_request(struct request
*rq
, int error
,
2289 unsigned int nr_bytes
,
2290 unsigned int bidi_bytes
)
2292 if (blk_update_request(rq
, error
, nr_bytes
))
2295 /* Bidi request must be completed as a whole */
2296 if (unlikely(blk_bidi_rq(rq
)) &&
2297 blk_update_request(rq
->next_rq
, error
, bidi_bytes
))
2300 if (blk_queue_add_random(rq
->q
))
2301 add_disk_randomness(rq
->rq_disk
);
2307 * blk_unprep_request - unprepare a request
2310 * This function makes a request ready for complete resubmission (or
2311 * completion). It happens only after all error handling is complete,
2312 * so represents the appropriate moment to deallocate any resources
2313 * that were allocated to the request in the prep_rq_fn. The queue
2314 * lock is held when calling this.
2316 void blk_unprep_request(struct request
*req
)
2318 struct request_queue
*q
= req
->q
;
2320 req
->cmd_flags
&= ~REQ_DONTPREP
;
2321 if (q
->unprep_rq_fn
)
2322 q
->unprep_rq_fn(q
, req
);
2324 EXPORT_SYMBOL_GPL(blk_unprep_request
);
2327 * queue lock must be held
2329 static void blk_finish_request(struct request
*req
, int error
)
2331 if (blk_rq_tagged(req
))
2332 blk_queue_end_tag(req
->q
, req
);
2334 BUG_ON(blk_queued_rq(req
));
2336 if (unlikely(laptop_mode
) && req
->cmd_type
== REQ_TYPE_FS
)
2337 laptop_io_completion(&req
->q
->backing_dev_info
);
2339 blk_delete_timer(req
);
2341 if (req
->cmd_flags
& REQ_DONTPREP
)
2342 blk_unprep_request(req
);
2345 blk_account_io_done(req
);
2348 req
->end_io(req
, error
);
2350 if (blk_bidi_rq(req
))
2351 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
2353 __blk_put_request(req
->q
, req
);
2358 * blk_end_bidi_request - Complete a bidi request
2359 * @rq: the request to complete
2360 * @error: %0 for success, < %0 for error
2361 * @nr_bytes: number of bytes to complete @rq
2362 * @bidi_bytes: number of bytes to complete @rq->next_rq
2365 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2366 * Drivers that supports bidi can safely call this member for any
2367 * type of request, bidi or uni. In the later case @bidi_bytes is
2371 * %false - we are done with this request
2372 * %true - still buffers pending for this request
2374 static bool blk_end_bidi_request(struct request
*rq
, int error
,
2375 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2377 struct request_queue
*q
= rq
->q
;
2378 unsigned long flags
;
2380 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2383 spin_lock_irqsave(q
->queue_lock
, flags
);
2384 blk_finish_request(rq
, error
);
2385 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2391 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2392 * @rq: the request to complete
2393 * @error: %0 for success, < %0 for error
2394 * @nr_bytes: number of bytes to complete @rq
2395 * @bidi_bytes: number of bytes to complete @rq->next_rq
2398 * Identical to blk_end_bidi_request() except that queue lock is
2399 * assumed to be locked on entry and remains so on return.
2402 * %false - we are done with this request
2403 * %true - still buffers pending for this request
2405 bool __blk_end_bidi_request(struct request
*rq
, int error
,
2406 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2408 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2411 blk_finish_request(rq
, error
);
2417 * blk_end_request - Helper function for drivers to complete the request.
2418 * @rq: the request being processed
2419 * @error: %0 for success, < %0 for error
2420 * @nr_bytes: number of bytes to complete
2423 * Ends I/O on a number of bytes attached to @rq.
2424 * If @rq has leftover, sets it up for the next range of segments.
2427 * %false - we are done with this request
2428 * %true - still buffers pending for this request
2430 bool blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2432 return blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2434 EXPORT_SYMBOL(blk_end_request
);
2437 * blk_end_request_all - Helper function for drives to finish the request.
2438 * @rq: the request to finish
2439 * @error: %0 for success, < %0 for error
2442 * Completely finish @rq.
2444 void blk_end_request_all(struct request
*rq
, int error
)
2447 unsigned int bidi_bytes
= 0;
2449 if (unlikely(blk_bidi_rq(rq
)))
2450 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2452 pending
= blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2455 EXPORT_SYMBOL(blk_end_request_all
);
2458 * blk_end_request_cur - Helper function to finish the current request chunk.
2459 * @rq: the request to finish the current chunk for
2460 * @error: %0 for success, < %0 for error
2463 * Complete the current consecutively mapped chunk from @rq.
2466 * %false - we are done with this request
2467 * %true - still buffers pending for this request
2469 bool blk_end_request_cur(struct request
*rq
, int error
)
2471 return blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2473 EXPORT_SYMBOL(blk_end_request_cur
);
2476 * blk_end_request_err - Finish a request till the next failure boundary.
2477 * @rq: the request to finish till the next failure boundary for
2478 * @error: must be negative errno
2481 * Complete @rq till the next failure boundary.
2484 * %false - we are done with this request
2485 * %true - still buffers pending for this request
2487 bool blk_end_request_err(struct request
*rq
, int error
)
2489 WARN_ON(error
>= 0);
2490 return blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2492 EXPORT_SYMBOL_GPL(blk_end_request_err
);
2495 * __blk_end_request - Helper function for drivers to complete the request.
2496 * @rq: the request being processed
2497 * @error: %0 for success, < %0 for error
2498 * @nr_bytes: number of bytes to complete
2501 * Must be called with queue lock held unlike blk_end_request().
2504 * %false - we are done with this request
2505 * %true - still buffers pending for this request
2507 bool __blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2509 return __blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2511 EXPORT_SYMBOL(__blk_end_request
);
2514 * __blk_end_request_all - Helper function for drives to finish the request.
2515 * @rq: the request to finish
2516 * @error: %0 for success, < %0 for error
2519 * Completely finish @rq. Must be called with queue lock held.
2521 void __blk_end_request_all(struct request
*rq
, int error
)
2524 unsigned int bidi_bytes
= 0;
2526 if (unlikely(blk_bidi_rq(rq
)))
2527 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2529 pending
= __blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2532 EXPORT_SYMBOL(__blk_end_request_all
);
2535 * __blk_end_request_cur - Helper function to finish the current request chunk.
2536 * @rq: the request to finish the current chunk for
2537 * @error: %0 for success, < %0 for error
2540 * Complete the current consecutively mapped chunk from @rq. Must
2541 * be called with queue lock held.
2544 * %false - we are done with this request
2545 * %true - still buffers pending for this request
2547 bool __blk_end_request_cur(struct request
*rq
, int error
)
2549 return __blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2551 EXPORT_SYMBOL(__blk_end_request_cur
);
2554 * __blk_end_request_err - Finish a request till the next failure boundary.
2555 * @rq: the request to finish till the next failure boundary for
2556 * @error: must be negative errno
2559 * Complete @rq till the next failure boundary. Must be called
2560 * with queue lock held.
2563 * %false - we are done with this request
2564 * %true - still buffers pending for this request
2566 bool __blk_end_request_err(struct request
*rq
, int error
)
2568 WARN_ON(error
>= 0);
2569 return __blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2571 EXPORT_SYMBOL_GPL(__blk_end_request_err
);
2573 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
2576 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2577 rq
->cmd_flags
|= bio
->bi_rw
& REQ_WRITE
;
2579 if (bio_has_data(bio
)) {
2580 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2581 rq
->buffer
= bio_data(bio
);
2583 rq
->__data_len
= bio
->bi_size
;
2584 rq
->bio
= rq
->biotail
= bio
;
2587 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2590 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2592 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2593 * @rq: the request to be flushed
2596 * Flush all pages in @rq.
2598 void rq_flush_dcache_pages(struct request
*rq
)
2600 struct req_iterator iter
;
2601 struct bio_vec
*bvec
;
2603 rq_for_each_segment(bvec
, rq
, iter
)
2604 flush_dcache_page(bvec
->bv_page
);
2606 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages
);
2610 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2611 * @q : the queue of the device being checked
2614 * Check if underlying low-level drivers of a device are busy.
2615 * If the drivers want to export their busy state, they must set own
2616 * exporting function using blk_queue_lld_busy() first.
2618 * Basically, this function is used only by request stacking drivers
2619 * to stop dispatching requests to underlying devices when underlying
2620 * devices are busy. This behavior helps more I/O merging on the queue
2621 * of the request stacking driver and prevents I/O throughput regression
2622 * on burst I/O load.
2625 * 0 - Not busy (The request stacking driver should dispatch request)
2626 * 1 - Busy (The request stacking driver should stop dispatching request)
2628 int blk_lld_busy(struct request_queue
*q
)
2631 return q
->lld_busy_fn(q
);
2635 EXPORT_SYMBOL_GPL(blk_lld_busy
);
2638 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2639 * @rq: the clone request to be cleaned up
2642 * Free all bios in @rq for a cloned request.
2644 void blk_rq_unprep_clone(struct request
*rq
)
2648 while ((bio
= rq
->bio
) != NULL
) {
2649 rq
->bio
= bio
->bi_next
;
2654 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone
);
2657 * Copy attributes of the original request to the clone request.
2658 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2660 static void __blk_rq_prep_clone(struct request
*dst
, struct request
*src
)
2662 dst
->cpu
= src
->cpu
;
2663 dst
->cmd_flags
= (src
->cmd_flags
& REQ_CLONE_MASK
) | REQ_NOMERGE
;
2664 dst
->cmd_type
= src
->cmd_type
;
2665 dst
->__sector
= blk_rq_pos(src
);
2666 dst
->__data_len
= blk_rq_bytes(src
);
2667 dst
->nr_phys_segments
= src
->nr_phys_segments
;
2668 dst
->ioprio
= src
->ioprio
;
2669 dst
->extra_len
= src
->extra_len
;
2673 * blk_rq_prep_clone - Helper function to setup clone request
2674 * @rq: the request to be setup
2675 * @rq_src: original request to be cloned
2676 * @bs: bio_set that bios for clone are allocated from
2677 * @gfp_mask: memory allocation mask for bio
2678 * @bio_ctr: setup function to be called for each clone bio.
2679 * Returns %0 for success, non %0 for failure.
2680 * @data: private data to be passed to @bio_ctr
2683 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2684 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2685 * are not copied, and copying such parts is the caller's responsibility.
2686 * Also, pages which the original bios are pointing to are not copied
2687 * and the cloned bios just point same pages.
2688 * So cloned bios must be completed before original bios, which means
2689 * the caller must complete @rq before @rq_src.
2691 int blk_rq_prep_clone(struct request
*rq
, struct request
*rq_src
,
2692 struct bio_set
*bs
, gfp_t gfp_mask
,
2693 int (*bio_ctr
)(struct bio
*, struct bio
*, void *),
2696 struct bio
*bio
, *bio_src
;
2701 blk_rq_init(NULL
, rq
);
2703 __rq_for_each_bio(bio_src
, rq_src
) {
2704 bio
= bio_alloc_bioset(gfp_mask
, bio_src
->bi_max_vecs
, bs
);
2708 __bio_clone(bio
, bio_src
);
2710 if (bio_integrity(bio_src
) &&
2711 bio_integrity_clone(bio
, bio_src
, gfp_mask
, bs
))
2714 if (bio_ctr
&& bio_ctr(bio
, bio_src
, data
))
2718 rq
->biotail
->bi_next
= bio
;
2721 rq
->bio
= rq
->biotail
= bio
;
2724 __blk_rq_prep_clone(rq
, rq_src
);
2731 blk_rq_unprep_clone(rq
);
2735 EXPORT_SYMBOL_GPL(blk_rq_prep_clone
);
2737 int kblockd_schedule_work(struct request_queue
*q
, struct work_struct
*work
)
2739 return queue_work(kblockd_workqueue
, work
);
2741 EXPORT_SYMBOL(kblockd_schedule_work
);
2743 int kblockd_schedule_delayed_work(struct request_queue
*q
,
2744 struct delayed_work
*dwork
, unsigned long delay
)
2746 return queue_delayed_work(kblockd_workqueue
, dwork
, delay
);
2748 EXPORT_SYMBOL(kblockd_schedule_delayed_work
);
2750 #define PLUG_MAGIC 0x91827364
2753 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2754 * @plug: The &struct blk_plug that needs to be initialized
2757 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2758 * pending I/O should the task end up blocking between blk_start_plug() and
2759 * blk_finish_plug(). This is important from a performance perspective, but
2760 * also ensures that we don't deadlock. For instance, if the task is blocking
2761 * for a memory allocation, memory reclaim could end up wanting to free a
2762 * page belonging to that request that is currently residing in our private
2763 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2764 * this kind of deadlock.
2766 void blk_start_plug(struct blk_plug
*plug
)
2768 struct task_struct
*tsk
= current
;
2770 plug
->magic
= PLUG_MAGIC
;
2771 INIT_LIST_HEAD(&plug
->list
);
2772 INIT_LIST_HEAD(&plug
->cb_list
);
2773 plug
->should_sort
= 0;
2776 * If this is a nested plug, don't actually assign it. It will be
2777 * flushed on its own.
2781 * Store ordering should not be needed here, since a potential
2782 * preempt will imply a full memory barrier
2787 EXPORT_SYMBOL(blk_start_plug
);
2789 static int plug_rq_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
2791 struct request
*rqa
= container_of(a
, struct request
, queuelist
);
2792 struct request
*rqb
= container_of(b
, struct request
, queuelist
);
2794 return !(rqa
->q
<= rqb
->q
);
2798 * If 'from_schedule' is true, then postpone the dispatch of requests
2799 * until a safe kblockd context. We due this to avoid accidental big
2800 * additional stack usage in driver dispatch, in places where the originally
2801 * plugger did not intend it.
2803 static void queue_unplugged(struct request_queue
*q
, unsigned int depth
,
2805 __releases(q
->queue_lock
)
2807 trace_block_unplug(q
, depth
, !from_schedule
);
2810 * Don't mess with dead queue.
2812 if (unlikely(blk_queue_dead(q
))) {
2813 spin_unlock(q
->queue_lock
);
2818 * If we are punting this to kblockd, then we can safely drop
2819 * the queue_lock before waking kblockd (which needs to take
2822 if (from_schedule
) {
2823 spin_unlock(q
->queue_lock
);
2824 blk_run_queue_async(q
);
2827 spin_unlock(q
->queue_lock
);
2832 static void flush_plug_callbacks(struct blk_plug
*plug
)
2834 LIST_HEAD(callbacks
);
2836 if (list_empty(&plug
->cb_list
))
2839 list_splice_init(&plug
->cb_list
, &callbacks
);
2841 while (!list_empty(&callbacks
)) {
2842 struct blk_plug_cb
*cb
= list_first_entry(&callbacks
,
2845 list_del(&cb
->list
);
2850 void blk_flush_plug_list(struct blk_plug
*plug
, bool from_schedule
)
2852 struct request_queue
*q
;
2853 unsigned long flags
;
2858 BUG_ON(plug
->magic
!= PLUG_MAGIC
);
2860 flush_plug_callbacks(plug
);
2861 if (list_empty(&plug
->list
))
2864 list_splice_init(&plug
->list
, &list
);
2866 if (plug
->should_sort
) {
2867 list_sort(NULL
, &list
, plug_rq_cmp
);
2868 plug
->should_sort
= 0;
2875 * Save and disable interrupts here, to avoid doing it for every
2876 * queue lock we have to take.
2878 local_irq_save(flags
);
2879 while (!list_empty(&list
)) {
2880 rq
= list_entry_rq(list
.next
);
2881 list_del_init(&rq
->queuelist
);
2885 * This drops the queue lock
2888 queue_unplugged(q
, depth
, from_schedule
);
2891 spin_lock(q
->queue_lock
);
2895 * Short-circuit if @q is dead
2897 if (unlikely(blk_queue_dead(q
))) {
2898 __blk_end_request_all(rq
, -ENODEV
);
2903 * rq is already accounted, so use raw insert
2905 if (rq
->cmd_flags
& (REQ_FLUSH
| REQ_FUA
))
2906 __elv_add_request(q
, rq
, ELEVATOR_INSERT_FLUSH
);
2908 __elv_add_request(q
, rq
, ELEVATOR_INSERT_SORT_MERGE
);
2914 * This drops the queue lock
2917 queue_unplugged(q
, depth
, from_schedule
);
2919 local_irq_restore(flags
);
2922 void blk_finish_plug(struct blk_plug
*plug
)
2924 blk_flush_plug_list(plug
, false);
2926 if (plug
== current
->plug
)
2927 current
->plug
= NULL
;
2929 EXPORT_SYMBOL(blk_finish_plug
);
2931 int __init
blk_dev_init(void)
2933 BUILD_BUG_ON(__REQ_NR_BITS
> 8 *
2934 sizeof(((struct request
*)0)->cmd_flags
));
2936 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
2937 kblockd_workqueue
= alloc_workqueue("kblockd",
2938 WQ_MEM_RECLAIM
| WQ_HIGHPRI
, 0);
2939 if (!kblockd_workqueue
)
2940 panic("Failed to create kblockd\n");
2942 request_cachep
= kmem_cache_create("blkdev_requests",
2943 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
2945 blk_requestq_cachep
= kmem_cache_create("blkdev_queue",
2946 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);