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>
37 #include "blk-cgroup.h"
39 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap
);
40 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap
);
41 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete
);
43 DEFINE_IDA(blk_queue_ida
);
46 * For the allocated request tables
48 static struct kmem_cache
*request_cachep
;
51 * For queue allocation
53 struct kmem_cache
*blk_requestq_cachep
;
56 * Controlling structure to kblockd
58 static struct workqueue_struct
*kblockd_workqueue
;
60 static void drive_stat_acct(struct request
*rq
, int new_io
)
62 struct hd_struct
*part
;
63 int rw
= rq_data_dir(rq
);
66 if (!blk_do_io_stat(rq
))
69 cpu
= part_stat_lock();
73 part_stat_inc(cpu
, part
, merges
[rw
]);
75 part
= disk_map_sector_rcu(rq
->rq_disk
, blk_rq_pos(rq
));
76 if (!hd_struct_try_get(part
)) {
78 * The partition is already being removed,
79 * the request will be accounted on the disk only
81 * We take a reference on disk->part0 although that
82 * partition will never be deleted, so we can treat
83 * it as any other partition.
85 part
= &rq
->rq_disk
->part0
;
88 part_round_stats(cpu
, part
);
89 part_inc_in_flight(part
, rw
);
96 void blk_queue_congestion_threshold(struct request_queue
*q
)
100 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) + 1;
101 if (nr
> q
->nr_requests
)
103 q
->nr_congestion_on
= nr
;
105 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) - (q
->nr_requests
/ 16) - 1;
108 q
->nr_congestion_off
= nr
;
112 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
115 * Locates the passed device's request queue and returns the address of its
118 * Will return NULL if the request queue cannot be located.
120 struct backing_dev_info
*blk_get_backing_dev_info(struct block_device
*bdev
)
122 struct backing_dev_info
*ret
= NULL
;
123 struct request_queue
*q
= bdev_get_queue(bdev
);
126 ret
= &q
->backing_dev_info
;
129 EXPORT_SYMBOL(blk_get_backing_dev_info
);
131 void blk_rq_init(struct request_queue
*q
, struct request
*rq
)
133 memset(rq
, 0, sizeof(*rq
));
135 INIT_LIST_HEAD(&rq
->queuelist
);
136 INIT_LIST_HEAD(&rq
->timeout_list
);
139 rq
->__sector
= (sector_t
) -1;
140 INIT_HLIST_NODE(&rq
->hash
);
141 RB_CLEAR_NODE(&rq
->rb_node
);
143 rq
->cmd_len
= BLK_MAX_CDB
;
146 rq
->start_time
= jiffies
;
147 set_start_time_ns(rq
);
150 EXPORT_SYMBOL(blk_rq_init
);
152 static void req_bio_endio(struct request
*rq
, struct bio
*bio
,
153 unsigned int nbytes
, int error
)
156 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
157 else if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
160 if (unlikely(nbytes
> bio
->bi_size
)) {
161 printk(KERN_ERR
"%s: want %u bytes done, %u left\n",
162 __func__
, nbytes
, bio
->bi_size
);
163 nbytes
= bio
->bi_size
;
166 if (unlikely(rq
->cmd_flags
& REQ_QUIET
))
167 set_bit(BIO_QUIET
, &bio
->bi_flags
);
169 bio
->bi_size
-= nbytes
;
170 bio
->bi_sector
+= (nbytes
>> 9);
172 if (bio_integrity(bio
))
173 bio_integrity_advance(bio
, nbytes
);
175 /* don't actually finish bio if it's part of flush sequence */
176 if (bio
->bi_size
== 0 && !(rq
->cmd_flags
& REQ_FLUSH_SEQ
))
177 bio_endio(bio
, error
);
180 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
184 printk(KERN_INFO
"%s: dev %s: type=%x, flags=%x\n", msg
,
185 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->cmd_type
,
188 printk(KERN_INFO
" sector %llu, nr/cnr %u/%u\n",
189 (unsigned long long)blk_rq_pos(rq
),
190 blk_rq_sectors(rq
), blk_rq_cur_sectors(rq
));
191 printk(KERN_INFO
" bio %p, biotail %p, buffer %p, len %u\n",
192 rq
->bio
, rq
->biotail
, rq
->buffer
, blk_rq_bytes(rq
));
194 if (rq
->cmd_type
== REQ_TYPE_BLOCK_PC
) {
195 printk(KERN_INFO
" cdb: ");
196 for (bit
= 0; bit
< BLK_MAX_CDB
; bit
++)
197 printk("%02x ", rq
->cmd
[bit
]);
201 EXPORT_SYMBOL(blk_dump_rq_flags
);
203 static void blk_delay_work(struct work_struct
*work
)
205 struct request_queue
*q
;
207 q
= container_of(work
, struct request_queue
, delay_work
.work
);
208 spin_lock_irq(q
->queue_lock
);
210 spin_unlock_irq(q
->queue_lock
);
214 * blk_delay_queue - restart queueing after defined interval
215 * @q: The &struct request_queue in question
216 * @msecs: Delay in msecs
219 * Sometimes queueing needs to be postponed for a little while, to allow
220 * resources to come back. This function will make sure that queueing is
221 * restarted around the specified time.
223 void blk_delay_queue(struct request_queue
*q
, unsigned long msecs
)
225 queue_delayed_work(kblockd_workqueue
, &q
->delay_work
,
226 msecs_to_jiffies(msecs
));
228 EXPORT_SYMBOL(blk_delay_queue
);
231 * blk_start_queue - restart a previously stopped queue
232 * @q: The &struct request_queue in question
235 * blk_start_queue() will clear the stop flag on the queue, and call
236 * the request_fn for the queue if it was in a stopped state when
237 * entered. Also see blk_stop_queue(). Queue lock must be held.
239 void blk_start_queue(struct request_queue
*q
)
241 WARN_ON(!irqs_disabled());
243 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
246 EXPORT_SYMBOL(blk_start_queue
);
249 * blk_stop_queue - stop a queue
250 * @q: The &struct request_queue in question
253 * The Linux block layer assumes that a block driver will consume all
254 * entries on the request queue when the request_fn strategy is called.
255 * Often this will not happen, because of hardware limitations (queue
256 * depth settings). If a device driver gets a 'queue full' response,
257 * or if it simply chooses not to queue more I/O at one point, it can
258 * call this function to prevent the request_fn from being called until
259 * the driver has signalled it's ready to go again. This happens by calling
260 * blk_start_queue() to restart queue operations. Queue lock must be held.
262 void blk_stop_queue(struct request_queue
*q
)
264 __cancel_delayed_work(&q
->delay_work
);
265 queue_flag_set(QUEUE_FLAG_STOPPED
, q
);
267 EXPORT_SYMBOL(blk_stop_queue
);
270 * blk_sync_queue - cancel any pending callbacks on a queue
274 * The block layer may perform asynchronous callback activity
275 * on a queue, such as calling the unplug function after a timeout.
276 * A block device may call blk_sync_queue to ensure that any
277 * such activity is cancelled, thus allowing it to release resources
278 * that the callbacks might use. The caller must already have made sure
279 * that its ->make_request_fn will not re-add plugging prior to calling
282 * This function does not cancel any asynchronous activity arising
283 * out of elevator or throttling code. That would require elevaotor_exit()
284 * and blkcg_exit_queue() to be called with queue lock initialized.
287 void blk_sync_queue(struct request_queue
*q
)
289 del_timer_sync(&q
->timeout
);
290 cancel_delayed_work_sync(&q
->delay_work
);
292 EXPORT_SYMBOL(blk_sync_queue
);
295 * __blk_run_queue - run a single device queue
296 * @q: The queue to run
299 * See @blk_run_queue. This variant must be called with the queue lock
300 * held and interrupts disabled.
302 void __blk_run_queue(struct request_queue
*q
)
304 if (unlikely(blk_queue_stopped(q
)))
309 EXPORT_SYMBOL(__blk_run_queue
);
312 * blk_run_queue_async - run a single device queue in workqueue context
313 * @q: The queue to run
316 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
319 void blk_run_queue_async(struct request_queue
*q
)
321 if (likely(!blk_queue_stopped(q
))) {
322 __cancel_delayed_work(&q
->delay_work
);
323 queue_delayed_work(kblockd_workqueue
, &q
->delay_work
, 0);
326 EXPORT_SYMBOL(blk_run_queue_async
);
329 * blk_run_queue - run a single device queue
330 * @q: The queue to run
333 * Invoke request handling on this queue, if it has pending work to do.
334 * May be used to restart queueing when a request has completed.
336 void blk_run_queue(struct request_queue
*q
)
340 spin_lock_irqsave(q
->queue_lock
, flags
);
342 spin_unlock_irqrestore(q
->queue_lock
, flags
);
344 EXPORT_SYMBOL(blk_run_queue
);
346 void blk_put_queue(struct request_queue
*q
)
348 kobject_put(&q
->kobj
);
350 EXPORT_SYMBOL(blk_put_queue
);
353 * blk_drain_queue - drain requests from request_queue
355 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
357 * Drain requests from @q. If @drain_all is set, all requests are drained.
358 * If not, only ELVPRIV requests are drained. The caller is responsible
359 * for ensuring that no new requests which need to be drained are queued.
361 void blk_drain_queue(struct request_queue
*q
, bool drain_all
)
367 spin_lock_irq(q
->queue_lock
);
370 * The caller might be trying to drain @q before its
371 * elevator is initialized.
374 elv_drain_elevator(q
);
376 blkcg_drain_queue(q
);
379 * This function might be called on a queue which failed
380 * driver init after queue creation or is not yet fully
381 * active yet. Some drivers (e.g. fd and loop) get unhappy
382 * in such cases. Kick queue iff dispatch queue has
383 * something on it and @q has request_fn set.
385 if (!list_empty(&q
->queue_head
) && q
->request_fn
)
388 drain
|= q
->rq
.elvpriv
;
391 * Unfortunately, requests are queued at and tracked from
392 * multiple places and there's no single counter which can
393 * be drained. Check all the queues and counters.
396 drain
|= !list_empty(&q
->queue_head
);
397 for (i
= 0; i
< 2; i
++) {
398 drain
|= q
->rq
.count
[i
];
399 drain
|= q
->in_flight
[i
];
400 drain
|= !list_empty(&q
->flush_queue
[i
]);
404 spin_unlock_irq(q
->queue_lock
);
413 * blk_queue_bypass_start - enter queue bypass mode
414 * @q: queue of interest
416 * In bypass mode, only the dispatch FIFO queue of @q is used. This
417 * function makes @q enter bypass mode and drains all requests which were
418 * throttled or issued before. On return, it's guaranteed that no request
419 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
420 * inside queue or RCU read lock.
422 void blk_queue_bypass_start(struct request_queue
*q
)
426 spin_lock_irq(q
->queue_lock
);
427 drain
= !q
->bypass_depth
++;
428 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
429 spin_unlock_irq(q
->queue_lock
);
432 blk_drain_queue(q
, false);
433 /* ensure blk_queue_bypass() is %true inside RCU read lock */
437 EXPORT_SYMBOL_GPL(blk_queue_bypass_start
);
440 * blk_queue_bypass_end - leave queue bypass mode
441 * @q: queue of interest
443 * Leave bypass mode and restore the normal queueing behavior.
445 void blk_queue_bypass_end(struct request_queue
*q
)
447 spin_lock_irq(q
->queue_lock
);
448 if (!--q
->bypass_depth
)
449 queue_flag_clear(QUEUE_FLAG_BYPASS
, q
);
450 WARN_ON_ONCE(q
->bypass_depth
< 0);
451 spin_unlock_irq(q
->queue_lock
);
453 EXPORT_SYMBOL_GPL(blk_queue_bypass_end
);
456 * blk_cleanup_queue - shutdown a request queue
457 * @q: request queue to shutdown
459 * Mark @q DEAD, drain all pending requests, destroy and put it. All
460 * future requests will be failed immediately with -ENODEV.
462 void blk_cleanup_queue(struct request_queue
*q
)
464 spinlock_t
*lock
= q
->queue_lock
;
466 /* mark @q DEAD, no new request or merges will be allowed afterwards */
467 mutex_lock(&q
->sysfs_lock
);
468 queue_flag_set_unlocked(QUEUE_FLAG_DEAD
, q
);
473 * Dead queue is permanently in bypass mode till released. Note
474 * that, unlike blk_queue_bypass_start(), we aren't performing
475 * synchronize_rcu() after entering bypass mode to avoid the delay
476 * as some drivers create and destroy a lot of queues while
477 * probing. This is still safe because blk_release_queue() will be
478 * called only after the queue refcnt drops to zero and nothing,
479 * RCU or not, would be traversing the queue by then.
482 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
484 queue_flag_set(QUEUE_FLAG_NOMERGES
, q
);
485 queue_flag_set(QUEUE_FLAG_NOXMERGES
, q
);
486 queue_flag_set(QUEUE_FLAG_DEAD
, q
);
488 if (q
->queue_lock
!= &q
->__queue_lock
)
489 q
->queue_lock
= &q
->__queue_lock
;
491 spin_unlock_irq(lock
);
492 mutex_unlock(&q
->sysfs_lock
);
494 /* drain all requests queued before DEAD marking */
495 blk_drain_queue(q
, true);
497 /* @q won't process any more request, flush async actions */
498 del_timer_sync(&q
->backing_dev_info
.laptop_mode_wb_timer
);
501 /* @q is and will stay empty, shutdown and put */
504 EXPORT_SYMBOL(blk_cleanup_queue
);
506 static int blk_init_free_list(struct request_queue
*q
)
508 struct request_list
*rl
= &q
->rq
;
510 if (unlikely(rl
->rq_pool
))
513 rl
->count
[BLK_RW_SYNC
] = rl
->count
[BLK_RW_ASYNC
] = 0;
514 rl
->starved
[BLK_RW_SYNC
] = rl
->starved
[BLK_RW_ASYNC
] = 0;
516 init_waitqueue_head(&rl
->wait
[BLK_RW_SYNC
]);
517 init_waitqueue_head(&rl
->wait
[BLK_RW_ASYNC
]);
519 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, mempool_alloc_slab
,
520 mempool_free_slab
, request_cachep
, q
->node
);
528 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
530 return blk_alloc_queue_node(gfp_mask
, -1);
532 EXPORT_SYMBOL(blk_alloc_queue
);
534 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
536 struct request_queue
*q
;
539 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
540 gfp_mask
| __GFP_ZERO
, node_id
);
544 q
->id
= ida_simple_get(&blk_queue_ida
, 0, 0, GFP_KERNEL
);
548 q
->backing_dev_info
.ra_pages
=
549 (VM_MAX_READAHEAD
* 1024) / PAGE_CACHE_SIZE
;
550 q
->backing_dev_info
.state
= 0;
551 q
->backing_dev_info
.capabilities
= BDI_CAP_MAP_COPY
;
552 q
->backing_dev_info
.name
= "block";
555 err
= bdi_init(&q
->backing_dev_info
);
559 setup_timer(&q
->backing_dev_info
.laptop_mode_wb_timer
,
560 laptop_mode_timer_fn
, (unsigned long) q
);
561 setup_timer(&q
->timeout
, blk_rq_timed_out_timer
, (unsigned long) q
);
562 INIT_LIST_HEAD(&q
->queue_head
);
563 INIT_LIST_HEAD(&q
->timeout_list
);
564 INIT_LIST_HEAD(&q
->icq_list
);
565 #ifdef CONFIG_BLK_CGROUP
566 INIT_LIST_HEAD(&q
->blkg_list
);
568 INIT_LIST_HEAD(&q
->flush_queue
[0]);
569 INIT_LIST_HEAD(&q
->flush_queue
[1]);
570 INIT_LIST_HEAD(&q
->flush_data_in_flight
);
571 INIT_DELAYED_WORK(&q
->delay_work
, blk_delay_work
);
573 kobject_init(&q
->kobj
, &blk_queue_ktype
);
575 mutex_init(&q
->sysfs_lock
);
576 spin_lock_init(&q
->__queue_lock
);
579 * By default initialize queue_lock to internal lock and driver can
580 * override it later if need be.
582 q
->queue_lock
= &q
->__queue_lock
;
585 * A queue starts its life with bypass turned on to avoid
586 * unnecessary bypass on/off overhead and nasty surprises during
587 * init. The initial bypass will be finished at the end of
588 * blk_init_allocated_queue().
591 __set_bit(QUEUE_FLAG_BYPASS
, &q
->queue_flags
);
593 if (blkcg_init_queue(q
))
599 ida_simple_remove(&blk_queue_ida
, q
->id
);
601 kmem_cache_free(blk_requestq_cachep
, q
);
604 EXPORT_SYMBOL(blk_alloc_queue_node
);
607 * blk_init_queue - prepare a request queue for use with a block device
608 * @rfn: The function to be called to process requests that have been
609 * placed on the queue.
610 * @lock: Request queue spin lock
613 * If a block device wishes to use the standard request handling procedures,
614 * which sorts requests and coalesces adjacent requests, then it must
615 * call blk_init_queue(). The function @rfn will be called when there
616 * are requests on the queue that need to be processed. If the device
617 * supports plugging, then @rfn may not be called immediately when requests
618 * are available on the queue, but may be called at some time later instead.
619 * Plugged queues are generally unplugged when a buffer belonging to one
620 * of the requests on the queue is needed, or due to memory pressure.
622 * @rfn is not required, or even expected, to remove all requests off the
623 * queue, but only as many as it can handle at a time. If it does leave
624 * requests on the queue, it is responsible for arranging that the requests
625 * get dealt with eventually.
627 * The queue spin lock must be held while manipulating the requests on the
628 * request queue; this lock will be taken also from interrupt context, so irq
629 * disabling is needed for it.
631 * Function returns a pointer to the initialized request queue, or %NULL if
635 * blk_init_queue() must be paired with a blk_cleanup_queue() call
636 * when the block device is deactivated (such as at module unload).
639 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
641 return blk_init_queue_node(rfn
, lock
, -1);
643 EXPORT_SYMBOL(blk_init_queue
);
645 struct request_queue
*
646 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
648 struct request_queue
*uninit_q
, *q
;
650 uninit_q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
654 q
= blk_init_allocated_queue(uninit_q
, rfn
, lock
);
656 blk_cleanup_queue(uninit_q
);
660 EXPORT_SYMBOL(blk_init_queue_node
);
662 struct request_queue
*
663 blk_init_allocated_queue(struct request_queue
*q
, request_fn_proc
*rfn
,
669 if (blk_init_free_list(q
))
673 q
->prep_rq_fn
= NULL
;
674 q
->unprep_rq_fn
= NULL
;
675 q
->queue_flags
= QUEUE_FLAG_DEFAULT
;
677 /* Override internal queue lock with supplied lock pointer */
679 q
->queue_lock
= lock
;
682 * This also sets hw/phys segments, boundary and size
684 blk_queue_make_request(q
, blk_queue_bio
);
686 q
->sg_reserved_size
= INT_MAX
;
689 if (elevator_init(q
, NULL
))
692 blk_queue_congestion_threshold(q
);
694 /* all done, end the initial bypass */
695 blk_queue_bypass_end(q
);
698 EXPORT_SYMBOL(blk_init_allocated_queue
);
700 bool blk_get_queue(struct request_queue
*q
)
702 if (likely(!blk_queue_dead(q
))) {
709 EXPORT_SYMBOL(blk_get_queue
);
711 static inline void blk_free_request(struct request_queue
*q
, struct request
*rq
)
713 if (rq
->cmd_flags
& REQ_ELVPRIV
) {
714 elv_put_request(q
, rq
);
716 put_io_context(rq
->elv
.icq
->ioc
);
719 mempool_free(rq
, q
->rq
.rq_pool
);
723 * ioc_batching returns true if the ioc is a valid batching request and
724 * should be given priority access to a request.
726 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
732 * Make sure the process is able to allocate at least 1 request
733 * even if the batch times out, otherwise we could theoretically
736 return ioc
->nr_batch_requests
== q
->nr_batching
||
737 (ioc
->nr_batch_requests
> 0
738 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
742 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
743 * will cause the process to be a "batcher" on all queues in the system. This
744 * is the behaviour we want though - once it gets a wakeup it should be given
747 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
749 if (!ioc
|| ioc_batching(q
, ioc
))
752 ioc
->nr_batch_requests
= q
->nr_batching
;
753 ioc
->last_waited
= jiffies
;
756 static void __freed_request(struct request_queue
*q
, int sync
)
758 struct request_list
*rl
= &q
->rq
;
760 if (rl
->count
[sync
] < queue_congestion_off_threshold(q
))
761 blk_clear_queue_congested(q
, sync
);
763 if (rl
->count
[sync
] + 1 <= q
->nr_requests
) {
764 if (waitqueue_active(&rl
->wait
[sync
]))
765 wake_up(&rl
->wait
[sync
]);
767 blk_clear_queue_full(q
, sync
);
772 * A request has just been released. Account for it, update the full and
773 * congestion status, wake up any waiters. Called under q->queue_lock.
775 static void freed_request(struct request_queue
*q
, unsigned int flags
)
777 struct request_list
*rl
= &q
->rq
;
778 int sync
= rw_is_sync(flags
);
781 if (flags
& REQ_ELVPRIV
)
784 __freed_request(q
, sync
);
786 if (unlikely(rl
->starved
[sync
^ 1]))
787 __freed_request(q
, sync
^ 1);
791 * Determine if elevator data should be initialized when allocating the
792 * request associated with @bio.
794 static bool blk_rq_should_init_elevator(struct bio
*bio
)
800 * Flush requests do not use the elevator so skip initialization.
801 * This allows a request to share the flush and elevator data.
803 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
))
810 * rq_ioc - determine io_context for request allocation
811 * @bio: request being allocated is for this bio (can be %NULL)
813 * Determine io_context to use for request allocation for @bio. May return
814 * %NULL if %current->io_context doesn't exist.
816 static struct io_context
*rq_ioc(struct bio
*bio
)
818 #ifdef CONFIG_BLK_CGROUP
819 if (bio
&& bio
->bi_ioc
)
822 return current
->io_context
;
826 * get_request - get a free request
827 * @q: request_queue to allocate request from
828 * @rw_flags: RW and SYNC flags
829 * @bio: bio to allocate request for (can be %NULL)
830 * @gfp_mask: allocation mask
832 * Get a free request from @q. This function may fail under memory
833 * pressure or if @q is dead.
835 * Must be callled with @q->queue_lock held and,
836 * Returns %NULL on failure, with @q->queue_lock held.
837 * Returns !%NULL on success, with @q->queue_lock *not held*.
839 static struct request
*get_request(struct request_queue
*q
, int rw_flags
,
840 struct bio
*bio
, gfp_t gfp_mask
)
843 struct request_list
*rl
= &q
->rq
;
844 struct elevator_type
*et
;
845 struct io_context
*ioc
;
846 struct io_cq
*icq
= NULL
;
847 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
848 bool retried
= false;
851 et
= q
->elevator
->type
;
854 if (unlikely(blk_queue_dead(q
)))
857 may_queue
= elv_may_queue(q
, rw_flags
);
858 if (may_queue
== ELV_MQUEUE_NO
)
861 if (rl
->count
[is_sync
]+1 >= queue_congestion_on_threshold(q
)) {
862 if (rl
->count
[is_sync
]+1 >= q
->nr_requests
) {
864 * We want ioc to record batching state. If it's
865 * not already there, creating a new one requires
866 * dropping queue_lock, which in turn requires
867 * retesting conditions to avoid queue hang.
869 if (!ioc
&& !retried
) {
870 spin_unlock_irq(q
->queue_lock
);
871 create_io_context(gfp_mask
, q
->node
);
872 spin_lock_irq(q
->queue_lock
);
878 * The queue will fill after this allocation, so set
879 * it as full, and mark this process as "batching".
880 * This process will be allowed to complete a batch of
881 * requests, others will be blocked.
883 if (!blk_queue_full(q
, is_sync
)) {
884 ioc_set_batching(q
, ioc
);
885 blk_set_queue_full(q
, is_sync
);
887 if (may_queue
!= ELV_MQUEUE_MUST
888 && !ioc_batching(q
, ioc
)) {
890 * The queue is full and the allocating
891 * process is not a "batcher", and not
892 * exempted by the IO scheduler
898 blk_set_queue_congested(q
, is_sync
);
902 * Only allow batching queuers to allocate up to 50% over the defined
903 * limit of requests, otherwise we could have thousands of requests
904 * allocated with any setting of ->nr_requests
906 if (rl
->count
[is_sync
] >= (3 * q
->nr_requests
/ 2))
909 rl
->count
[is_sync
]++;
910 rl
->starved
[is_sync
] = 0;
913 * Decide whether the new request will be managed by elevator. If
914 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
915 * prevent the current elevator from being destroyed until the new
916 * request is freed. This guarantees icq's won't be destroyed and
917 * makes creating new ones safe.
919 * Also, lookup icq while holding queue_lock. If it doesn't exist,
920 * it will be created after releasing queue_lock.
922 if (blk_rq_should_init_elevator(bio
) && !blk_queue_bypass(q
)) {
923 rw_flags
|= REQ_ELVPRIV
;
925 if (et
->icq_cache
&& ioc
)
926 icq
= ioc_lookup_icq(ioc
, q
);
929 if (blk_queue_io_stat(q
))
930 rw_flags
|= REQ_IO_STAT
;
931 spin_unlock_irq(q
->queue_lock
);
933 /* create icq if missing */
934 if ((rw_flags
& REQ_ELVPRIV
) && unlikely(et
->icq_cache
&& !icq
)) {
935 create_io_context(gfp_mask
, q
->node
);
939 icq
= ioc_create_icq(ioc
, q
, gfp_mask
);
944 /* allocate and init request */
945 rq
= mempool_alloc(q
->rq
.rq_pool
, gfp_mask
);
950 rq
->cmd_flags
= rw_flags
| REQ_ALLOCED
;
952 if (rw_flags
& REQ_ELVPRIV
) {
954 if (unlikely(elv_set_request(q
, rq
, bio
, gfp_mask
))) {
955 mempool_free(rq
, q
->rq
.rq_pool
);
958 /* @rq->elv.icq holds on to io_context until @rq is freed */
960 get_io_context(icq
->ioc
);
964 * ioc may be NULL here, and ioc_batching will be false. That's
965 * OK, if the queue is under the request limit then requests need
966 * not count toward the nr_batch_requests limit. There will always
967 * be some limit enforced by BLK_BATCH_TIME.
969 if (ioc_batching(q
, ioc
))
970 ioc
->nr_batch_requests
--;
972 trace_block_getrq(q
, bio
, rw_flags
& 1);
977 * Allocation failed presumably due to memory. Undo anything we
978 * might have messed up.
980 * Allocating task should really be put onto the front of the wait
981 * queue, but this is pretty rare.
983 spin_lock_irq(q
->queue_lock
);
984 freed_request(q
, rw_flags
);
987 * in the very unlikely event that allocation failed and no
988 * requests for this direction was pending, mark us starved so that
989 * freeing of a request in the other direction will notice
990 * us. another possible fix would be to split the rq mempool into
994 if (unlikely(rl
->count
[is_sync
] == 0))
995 rl
->starved
[is_sync
] = 1;
1000 * get_request_wait - get a free request with retry
1001 * @q: request_queue to allocate request from
1002 * @rw_flags: RW and SYNC flags
1003 * @bio: bio to allocate request for (can be %NULL)
1005 * Get a free request from @q. This function keeps retrying under memory
1006 * pressure and fails iff @q is dead.
1008 * Must be callled with @q->queue_lock held and,
1009 * Returns %NULL on failure, with @q->queue_lock held.
1010 * Returns !%NULL on success, with @q->queue_lock *not held*.
1012 static struct request
*get_request_wait(struct request_queue
*q
, int rw_flags
,
1015 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
1018 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
1021 struct request_list
*rl
= &q
->rq
;
1023 if (unlikely(blk_queue_dead(q
)))
1026 prepare_to_wait_exclusive(&rl
->wait
[is_sync
], &wait
,
1027 TASK_UNINTERRUPTIBLE
);
1029 trace_block_sleeprq(q
, bio
, rw_flags
& 1);
1031 spin_unlock_irq(q
->queue_lock
);
1035 * After sleeping, we become a "batching" process and
1036 * will be able to allocate at least one request, and
1037 * up to a big batch of them for a small period time.
1038 * See ioc_batching, ioc_set_batching
1040 create_io_context(GFP_NOIO
, q
->node
);
1041 ioc_set_batching(q
, current
->io_context
);
1043 spin_lock_irq(q
->queue_lock
);
1044 finish_wait(&rl
->wait
[is_sync
], &wait
);
1046 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
1052 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
1056 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
1058 spin_lock_irq(q
->queue_lock
);
1059 if (gfp_mask
& __GFP_WAIT
)
1060 rq
= get_request_wait(q
, rw
, NULL
);
1062 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
1064 spin_unlock_irq(q
->queue_lock
);
1065 /* q->queue_lock is unlocked at this point */
1069 EXPORT_SYMBOL(blk_get_request
);
1072 * blk_make_request - given a bio, allocate a corresponding struct request.
1073 * @q: target request queue
1074 * @bio: The bio describing the memory mappings that will be submitted for IO.
1075 * It may be a chained-bio properly constructed by block/bio layer.
1076 * @gfp_mask: gfp flags to be used for memory allocation
1078 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1079 * type commands. Where the struct request needs to be farther initialized by
1080 * the caller. It is passed a &struct bio, which describes the memory info of
1083 * The caller of blk_make_request must make sure that bi_io_vec
1084 * are set to describe the memory buffers. That bio_data_dir() will return
1085 * the needed direction of the request. (And all bio's in the passed bio-chain
1086 * are properly set accordingly)
1088 * If called under none-sleepable conditions, mapped bio buffers must not
1089 * need bouncing, by calling the appropriate masked or flagged allocator,
1090 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1093 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1094 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1095 * anything but the first bio in the chain. Otherwise you risk waiting for IO
1096 * completion of a bio that hasn't been submitted yet, thus resulting in a
1097 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1098 * of bio_alloc(), as that avoids the mempool deadlock.
1099 * If possible a big IO should be split into smaller parts when allocation
1100 * fails. Partial allocation should not be an error, or you risk a live-lock.
1102 struct request
*blk_make_request(struct request_queue
*q
, struct bio
*bio
,
1105 struct request
*rq
= blk_get_request(q
, bio_data_dir(bio
), gfp_mask
);
1108 return ERR_PTR(-ENOMEM
);
1111 struct bio
*bounce_bio
= bio
;
1114 blk_queue_bounce(q
, &bounce_bio
);
1115 ret
= blk_rq_append_bio(q
, rq
, bounce_bio
);
1116 if (unlikely(ret
)) {
1117 blk_put_request(rq
);
1118 return ERR_PTR(ret
);
1124 EXPORT_SYMBOL(blk_make_request
);
1127 * blk_requeue_request - put a request back on queue
1128 * @q: request queue where request should be inserted
1129 * @rq: request to be inserted
1132 * Drivers often keep queueing requests until the hardware cannot accept
1133 * more, when that condition happens we need to put the request back
1134 * on the queue. Must be called with queue lock held.
1136 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
1138 blk_delete_timer(rq
);
1139 blk_clear_rq_complete(rq
);
1140 trace_block_rq_requeue(q
, rq
);
1142 if (blk_rq_tagged(rq
))
1143 blk_queue_end_tag(q
, rq
);
1145 BUG_ON(blk_queued_rq(rq
));
1147 elv_requeue_request(q
, rq
);
1149 EXPORT_SYMBOL(blk_requeue_request
);
1151 static void add_acct_request(struct request_queue
*q
, struct request
*rq
,
1154 drive_stat_acct(rq
, 1);
1155 __elv_add_request(q
, rq
, where
);
1158 static void part_round_stats_single(int cpu
, struct hd_struct
*part
,
1161 if (now
== part
->stamp
)
1164 if (part_in_flight(part
)) {
1165 __part_stat_add(cpu
, part
, time_in_queue
,
1166 part_in_flight(part
) * (now
- part
->stamp
));
1167 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
1173 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1174 * @cpu: cpu number for stats access
1175 * @part: target partition
1177 * The average IO queue length and utilisation statistics are maintained
1178 * by observing the current state of the queue length and the amount of
1179 * time it has been in this state for.
1181 * Normally, that accounting is done on IO completion, but that can result
1182 * in more than a second's worth of IO being accounted for within any one
1183 * second, leading to >100% utilisation. To deal with that, we call this
1184 * function to do a round-off before returning the results when reading
1185 * /proc/diskstats. This accounts immediately for all queue usage up to
1186 * the current jiffies and restarts the counters again.
1188 void part_round_stats(int cpu
, struct hd_struct
*part
)
1190 unsigned long now
= jiffies
;
1193 part_round_stats_single(cpu
, &part_to_disk(part
)->part0
, now
);
1194 part_round_stats_single(cpu
, part
, now
);
1196 EXPORT_SYMBOL_GPL(part_round_stats
);
1199 * queue lock must be held
1201 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1205 if (unlikely(--req
->ref_count
))
1208 elv_completed_request(q
, req
);
1210 /* this is a bio leak */
1211 WARN_ON(req
->bio
!= NULL
);
1214 * Request may not have originated from ll_rw_blk. if not,
1215 * it didn't come out of our reserved rq pools
1217 if (req
->cmd_flags
& REQ_ALLOCED
) {
1218 unsigned int flags
= req
->cmd_flags
;
1220 BUG_ON(!list_empty(&req
->queuelist
));
1221 BUG_ON(!hlist_unhashed(&req
->hash
));
1223 blk_free_request(q
, req
);
1224 freed_request(q
, flags
);
1227 EXPORT_SYMBOL_GPL(__blk_put_request
);
1229 void blk_put_request(struct request
*req
)
1231 unsigned long flags
;
1232 struct request_queue
*q
= req
->q
;
1234 spin_lock_irqsave(q
->queue_lock
, flags
);
1235 __blk_put_request(q
, req
);
1236 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1238 EXPORT_SYMBOL(blk_put_request
);
1241 * blk_add_request_payload - add a payload to a request
1242 * @rq: request to update
1243 * @page: page backing the payload
1244 * @len: length of the payload.
1246 * This allows to later add a payload to an already submitted request by
1247 * a block driver. The driver needs to take care of freeing the payload
1250 * Note that this is a quite horrible hack and nothing but handling of
1251 * discard requests should ever use it.
1253 void blk_add_request_payload(struct request
*rq
, struct page
*page
,
1256 struct bio
*bio
= rq
->bio
;
1258 bio
->bi_io_vec
->bv_page
= page
;
1259 bio
->bi_io_vec
->bv_offset
= 0;
1260 bio
->bi_io_vec
->bv_len
= len
;
1264 bio
->bi_phys_segments
= 1;
1266 rq
->__data_len
= rq
->resid_len
= len
;
1267 rq
->nr_phys_segments
= 1;
1268 rq
->buffer
= bio_data(bio
);
1270 EXPORT_SYMBOL_GPL(blk_add_request_payload
);
1272 static bool bio_attempt_back_merge(struct request_queue
*q
, struct request
*req
,
1275 const int ff
= bio
->bi_rw
& REQ_FAILFAST_MASK
;
1277 if (!ll_back_merge_fn(q
, req
, bio
))
1280 trace_block_bio_backmerge(q
, bio
);
1282 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1283 blk_rq_set_mixed_merge(req
);
1285 req
->biotail
->bi_next
= bio
;
1287 req
->__data_len
+= bio
->bi_size
;
1288 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1290 drive_stat_acct(req
, 0);
1294 static bool bio_attempt_front_merge(struct request_queue
*q
,
1295 struct request
*req
, struct bio
*bio
)
1297 const int ff
= bio
->bi_rw
& REQ_FAILFAST_MASK
;
1299 if (!ll_front_merge_fn(q
, req
, bio
))
1302 trace_block_bio_frontmerge(q
, bio
);
1304 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1305 blk_rq_set_mixed_merge(req
);
1307 bio
->bi_next
= req
->bio
;
1311 * may not be valid. if the low level driver said
1312 * it didn't need a bounce buffer then it better
1313 * not touch req->buffer either...
1315 req
->buffer
= bio_data(bio
);
1316 req
->__sector
= bio
->bi_sector
;
1317 req
->__data_len
+= bio
->bi_size
;
1318 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1320 drive_stat_acct(req
, 0);
1325 * attempt_plug_merge - try to merge with %current's plugged list
1326 * @q: request_queue new bio is being queued at
1327 * @bio: new bio being queued
1328 * @request_count: out parameter for number of traversed plugged requests
1330 * Determine whether @bio being queued on @q can be merged with a request
1331 * on %current's plugged list. Returns %true if merge was successful,
1334 * Plugging coalesces IOs from the same issuer for the same purpose without
1335 * going through @q->queue_lock. As such it's more of an issuing mechanism
1336 * than scheduling, and the request, while may have elvpriv data, is not
1337 * added on the elevator at this point. In addition, we don't have
1338 * reliable access to the elevator outside queue lock. Only check basic
1339 * merging parameters without querying the elevator.
1341 static bool attempt_plug_merge(struct request_queue
*q
, struct bio
*bio
,
1342 unsigned int *request_count
)
1344 struct blk_plug
*plug
;
1348 plug
= current
->plug
;
1353 list_for_each_entry_reverse(rq
, &plug
->list
, queuelist
) {
1358 if (rq
->q
!= q
|| !blk_rq_merge_ok(rq
, bio
))
1361 el_ret
= blk_try_merge(rq
, bio
);
1362 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1363 ret
= bio_attempt_back_merge(q
, rq
, bio
);
1366 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1367 ret
= bio_attempt_front_merge(q
, rq
, bio
);
1376 void init_request_from_bio(struct request
*req
, struct bio
*bio
)
1378 req
->cmd_type
= REQ_TYPE_FS
;
1380 req
->cmd_flags
|= bio
->bi_rw
& REQ_COMMON_MASK
;
1381 if (bio
->bi_rw
& REQ_RAHEAD
)
1382 req
->cmd_flags
|= REQ_FAILFAST_MASK
;
1385 req
->__sector
= bio
->bi_sector
;
1386 req
->ioprio
= bio_prio(bio
);
1387 blk_rq_bio_prep(req
->q
, req
, bio
);
1390 void blk_queue_bio(struct request_queue
*q
, struct bio
*bio
)
1392 const bool sync
= !!(bio
->bi_rw
& REQ_SYNC
);
1393 struct blk_plug
*plug
;
1394 int el_ret
, rw_flags
, where
= ELEVATOR_INSERT_SORT
;
1395 struct request
*req
;
1396 unsigned int request_count
= 0;
1399 * low level driver can indicate that it wants pages above a
1400 * certain limit bounced to low memory (ie for highmem, or even
1401 * ISA dma in theory)
1403 blk_queue_bounce(q
, &bio
);
1405 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) {
1406 spin_lock_irq(q
->queue_lock
);
1407 where
= ELEVATOR_INSERT_FLUSH
;
1412 * Check if we can merge with the plugged list before grabbing
1415 if (attempt_plug_merge(q
, bio
, &request_count
))
1418 spin_lock_irq(q
->queue_lock
);
1420 el_ret
= elv_merge(q
, &req
, bio
);
1421 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1422 if (bio_attempt_back_merge(q
, req
, bio
)) {
1423 elv_bio_merged(q
, req
, bio
);
1424 if (!attempt_back_merge(q
, req
))
1425 elv_merged_request(q
, req
, el_ret
);
1428 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1429 if (bio_attempt_front_merge(q
, req
, bio
)) {
1430 elv_bio_merged(q
, req
, bio
);
1431 if (!attempt_front_merge(q
, req
))
1432 elv_merged_request(q
, req
, el_ret
);
1439 * This sync check and mask will be re-done in init_request_from_bio(),
1440 * but we need to set it earlier to expose the sync flag to the
1441 * rq allocator and io schedulers.
1443 rw_flags
= bio_data_dir(bio
);
1445 rw_flags
|= REQ_SYNC
;
1448 * Grab a free request. This is might sleep but can not fail.
1449 * Returns with the queue unlocked.
1451 req
= get_request_wait(q
, rw_flags
, bio
);
1452 if (unlikely(!req
)) {
1453 bio_endio(bio
, -ENODEV
); /* @q is dead */
1458 * After dropping the lock and possibly sleeping here, our request
1459 * may now be mergeable after it had proven unmergeable (above).
1460 * We don't worry about that case for efficiency. It won't happen
1461 * often, and the elevators are able to handle it.
1463 init_request_from_bio(req
, bio
);
1465 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
))
1466 req
->cpu
= raw_smp_processor_id();
1468 plug
= current
->plug
;
1471 * If this is the first request added after a plug, fire
1472 * of a plug trace. If others have been added before, check
1473 * if we have multiple devices in this plug. If so, make a
1474 * note to sort the list before dispatch.
1476 if (list_empty(&plug
->list
))
1477 trace_block_plug(q
);
1479 if (!plug
->should_sort
) {
1480 struct request
*__rq
;
1482 __rq
= list_entry_rq(plug
->list
.prev
);
1484 plug
->should_sort
= 1;
1486 if (request_count
>= BLK_MAX_REQUEST_COUNT
) {
1487 blk_flush_plug_list(plug
, false);
1488 trace_block_plug(q
);
1491 list_add_tail(&req
->queuelist
, &plug
->list
);
1492 drive_stat_acct(req
, 1);
1494 spin_lock_irq(q
->queue_lock
);
1495 add_acct_request(q
, req
, where
);
1498 spin_unlock_irq(q
->queue_lock
);
1501 EXPORT_SYMBOL_GPL(blk_queue_bio
); /* for device mapper only */
1504 * If bio->bi_dev is a partition, remap the location
1506 static inline void blk_partition_remap(struct bio
*bio
)
1508 struct block_device
*bdev
= bio
->bi_bdev
;
1510 if (bio_sectors(bio
) && bdev
!= bdev
->bd_contains
) {
1511 struct hd_struct
*p
= bdev
->bd_part
;
1513 bio
->bi_sector
+= p
->start_sect
;
1514 bio
->bi_bdev
= bdev
->bd_contains
;
1516 trace_block_bio_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
1518 bio
->bi_sector
- p
->start_sect
);
1522 static void handle_bad_sector(struct bio
*bio
)
1524 char b
[BDEVNAME_SIZE
];
1526 printk(KERN_INFO
"attempt to access beyond end of device\n");
1527 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
1528 bdevname(bio
->bi_bdev
, b
),
1530 (unsigned long long)bio
->bi_sector
+ bio_sectors(bio
),
1531 (long long)(i_size_read(bio
->bi_bdev
->bd_inode
) >> 9));
1533 set_bit(BIO_EOF
, &bio
->bi_flags
);
1536 #ifdef CONFIG_FAIL_MAKE_REQUEST
1538 static DECLARE_FAULT_ATTR(fail_make_request
);
1540 static int __init
setup_fail_make_request(char *str
)
1542 return setup_fault_attr(&fail_make_request
, str
);
1544 __setup("fail_make_request=", setup_fail_make_request
);
1546 static bool should_fail_request(struct hd_struct
*part
, unsigned int bytes
)
1548 return part
->make_it_fail
&& should_fail(&fail_make_request
, bytes
);
1551 static int __init
fail_make_request_debugfs(void)
1553 struct dentry
*dir
= fault_create_debugfs_attr("fail_make_request",
1554 NULL
, &fail_make_request
);
1556 return IS_ERR(dir
) ? PTR_ERR(dir
) : 0;
1559 late_initcall(fail_make_request_debugfs
);
1561 #else /* CONFIG_FAIL_MAKE_REQUEST */
1563 static inline bool should_fail_request(struct hd_struct
*part
,
1569 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1572 * Check whether this bio extends beyond the end of the device.
1574 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
1581 /* Test device or partition size, when known. */
1582 maxsector
= i_size_read(bio
->bi_bdev
->bd_inode
) >> 9;
1584 sector_t sector
= bio
->bi_sector
;
1586 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
1588 * This may well happen - the kernel calls bread()
1589 * without checking the size of the device, e.g., when
1590 * mounting a device.
1592 handle_bad_sector(bio
);
1600 static noinline_for_stack
bool
1601 generic_make_request_checks(struct bio
*bio
)
1603 struct request_queue
*q
;
1604 int nr_sectors
= bio_sectors(bio
);
1606 char b
[BDEVNAME_SIZE
];
1607 struct hd_struct
*part
;
1611 if (bio_check_eod(bio
, nr_sectors
))
1614 q
= bdev_get_queue(bio
->bi_bdev
);
1617 "generic_make_request: Trying to access "
1618 "nonexistent block-device %s (%Lu)\n",
1619 bdevname(bio
->bi_bdev
, b
),
1620 (long long) bio
->bi_sector
);
1624 if (unlikely(!(bio
->bi_rw
& REQ_DISCARD
) &&
1625 nr_sectors
> queue_max_hw_sectors(q
))) {
1626 printk(KERN_ERR
"bio too big device %s (%u > %u)\n",
1627 bdevname(bio
->bi_bdev
, b
),
1629 queue_max_hw_sectors(q
));
1633 part
= bio
->bi_bdev
->bd_part
;
1634 if (should_fail_request(part
, bio
->bi_size
) ||
1635 should_fail_request(&part_to_disk(part
)->part0
,
1640 * If this device has partitions, remap block n
1641 * of partition p to block n+start(p) of the disk.
1643 blk_partition_remap(bio
);
1645 if (bio_integrity_enabled(bio
) && bio_integrity_prep(bio
))
1648 if (bio_check_eod(bio
, nr_sectors
))
1652 * Filter flush bio's early so that make_request based
1653 * drivers without flush support don't have to worry
1656 if ((bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) && !q
->flush_flags
) {
1657 bio
->bi_rw
&= ~(REQ_FLUSH
| REQ_FUA
);
1664 if ((bio
->bi_rw
& REQ_DISCARD
) &&
1665 (!blk_queue_discard(q
) ||
1666 ((bio
->bi_rw
& REQ_SECURE
) &&
1667 !blk_queue_secdiscard(q
)))) {
1672 if (blk_throtl_bio(q
, bio
))
1673 return false; /* throttled, will be resubmitted later */
1675 trace_block_bio_queue(q
, bio
);
1679 bio_endio(bio
, err
);
1684 * generic_make_request - hand a buffer to its device driver for I/O
1685 * @bio: The bio describing the location in memory and on the device.
1687 * generic_make_request() is used to make I/O requests of block
1688 * devices. It is passed a &struct bio, which describes the I/O that needs
1691 * generic_make_request() does not return any status. The
1692 * success/failure status of the request, along with notification of
1693 * completion, is delivered asynchronously through the bio->bi_end_io
1694 * function described (one day) else where.
1696 * The caller of generic_make_request must make sure that bi_io_vec
1697 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1698 * set to describe the device address, and the
1699 * bi_end_io and optionally bi_private are set to describe how
1700 * completion notification should be signaled.
1702 * generic_make_request and the drivers it calls may use bi_next if this
1703 * bio happens to be merged with someone else, and may resubmit the bio to
1704 * a lower device by calling into generic_make_request recursively, which
1705 * means the bio should NOT be touched after the call to ->make_request_fn.
1707 void generic_make_request(struct bio
*bio
)
1709 struct bio_list bio_list_on_stack
;
1711 if (!generic_make_request_checks(bio
))
1715 * We only want one ->make_request_fn to be active at a time, else
1716 * stack usage with stacked devices could be a problem. So use
1717 * current->bio_list to keep a list of requests submited by a
1718 * make_request_fn function. current->bio_list is also used as a
1719 * flag to say if generic_make_request is currently active in this
1720 * task or not. If it is NULL, then no make_request is active. If
1721 * it is non-NULL, then a make_request is active, and new requests
1722 * should be added at the tail
1724 if (current
->bio_list
) {
1725 bio_list_add(current
->bio_list
, bio
);
1729 /* following loop may be a bit non-obvious, and so deserves some
1731 * Before entering the loop, bio->bi_next is NULL (as all callers
1732 * ensure that) so we have a list with a single bio.
1733 * We pretend that we have just taken it off a longer list, so
1734 * we assign bio_list to a pointer to the bio_list_on_stack,
1735 * thus initialising the bio_list of new bios to be
1736 * added. ->make_request() may indeed add some more bios
1737 * through a recursive call to generic_make_request. If it
1738 * did, we find a non-NULL value in bio_list and re-enter the loop
1739 * from the top. In this case we really did just take the bio
1740 * of the top of the list (no pretending) and so remove it from
1741 * bio_list, and call into ->make_request() again.
1743 BUG_ON(bio
->bi_next
);
1744 bio_list_init(&bio_list_on_stack
);
1745 current
->bio_list
= &bio_list_on_stack
;
1747 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
1749 q
->make_request_fn(q
, bio
);
1751 bio
= bio_list_pop(current
->bio_list
);
1753 current
->bio_list
= NULL
; /* deactivate */
1755 EXPORT_SYMBOL(generic_make_request
);
1758 * submit_bio - submit a bio to the block device layer for I/O
1759 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1760 * @bio: The &struct bio which describes the I/O
1762 * submit_bio() is very similar in purpose to generic_make_request(), and
1763 * uses that function to do most of the work. Both are fairly rough
1764 * interfaces; @bio must be presetup and ready for I/O.
1767 void submit_bio(int rw
, struct bio
*bio
)
1769 int count
= bio_sectors(bio
);
1774 * If it's a regular read/write or a barrier with data attached,
1775 * go through the normal accounting stuff before submission.
1777 if (bio_has_data(bio
) && !(rw
& REQ_DISCARD
)) {
1779 count_vm_events(PGPGOUT
, count
);
1781 task_io_account_read(bio
->bi_size
);
1782 count_vm_events(PGPGIN
, count
);
1785 if (unlikely(block_dump
)) {
1786 char b
[BDEVNAME_SIZE
];
1787 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s (%u sectors)\n",
1788 current
->comm
, task_pid_nr(current
),
1789 (rw
& WRITE
) ? "WRITE" : "READ",
1790 (unsigned long long)bio
->bi_sector
,
1791 bdevname(bio
->bi_bdev
, b
),
1796 generic_make_request(bio
);
1798 EXPORT_SYMBOL(submit_bio
);
1801 * blk_rq_check_limits - Helper function to check a request for the queue limit
1803 * @rq: the request being checked
1806 * @rq may have been made based on weaker limitations of upper-level queues
1807 * in request stacking drivers, and it may violate the limitation of @q.
1808 * Since the block layer and the underlying device driver trust @rq
1809 * after it is inserted to @q, it should be checked against @q before
1810 * the insertion using this generic function.
1812 * This function should also be useful for request stacking drivers
1813 * in some cases below, so export this function.
1814 * Request stacking drivers like request-based dm may change the queue
1815 * limits while requests are in the queue (e.g. dm's table swapping).
1816 * Such request stacking drivers should check those requests agaist
1817 * the new queue limits again when they dispatch those requests,
1818 * although such checkings are also done against the old queue limits
1819 * when submitting requests.
1821 int blk_rq_check_limits(struct request_queue
*q
, struct request
*rq
)
1823 if (rq
->cmd_flags
& REQ_DISCARD
)
1826 if (blk_rq_sectors(rq
) > queue_max_sectors(q
) ||
1827 blk_rq_bytes(rq
) > queue_max_hw_sectors(q
) << 9) {
1828 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
1833 * queue's settings related to segment counting like q->bounce_pfn
1834 * may differ from that of other stacking queues.
1835 * Recalculate it to check the request correctly on this queue's
1838 blk_recalc_rq_segments(rq
);
1839 if (rq
->nr_phys_segments
> queue_max_segments(q
)) {
1840 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
1846 EXPORT_SYMBOL_GPL(blk_rq_check_limits
);
1849 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1850 * @q: the queue to submit the request
1851 * @rq: the request being queued
1853 int blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
1855 unsigned long flags
;
1856 int where
= ELEVATOR_INSERT_BACK
;
1858 if (blk_rq_check_limits(q
, rq
))
1862 should_fail_request(&rq
->rq_disk
->part0
, blk_rq_bytes(rq
)))
1865 spin_lock_irqsave(q
->queue_lock
, flags
);
1866 if (unlikely(blk_queue_dead(q
))) {
1867 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1872 * Submitting request must be dequeued before calling this function
1873 * because it will be linked to another request_queue
1875 BUG_ON(blk_queued_rq(rq
));
1877 if (rq
->cmd_flags
& (REQ_FLUSH
|REQ_FUA
))
1878 where
= ELEVATOR_INSERT_FLUSH
;
1880 add_acct_request(q
, rq
, where
);
1881 if (where
== ELEVATOR_INSERT_FLUSH
)
1883 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1887 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
1890 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1891 * @rq: request to examine
1894 * A request could be merge of IOs which require different failure
1895 * handling. This function determines the number of bytes which
1896 * can be failed from the beginning of the request without
1897 * crossing into area which need to be retried further.
1900 * The number of bytes to fail.
1903 * queue_lock must be held.
1905 unsigned int blk_rq_err_bytes(const struct request
*rq
)
1907 unsigned int ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
1908 unsigned int bytes
= 0;
1911 if (!(rq
->cmd_flags
& REQ_MIXED_MERGE
))
1912 return blk_rq_bytes(rq
);
1915 * Currently the only 'mixing' which can happen is between
1916 * different fastfail types. We can safely fail portions
1917 * which have all the failfast bits that the first one has -
1918 * the ones which are at least as eager to fail as the first
1921 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
1922 if ((bio
->bi_rw
& ff
) != ff
)
1924 bytes
+= bio
->bi_size
;
1927 /* this could lead to infinite loop */
1928 BUG_ON(blk_rq_bytes(rq
) && !bytes
);
1931 EXPORT_SYMBOL_GPL(blk_rq_err_bytes
);
1933 static void blk_account_io_completion(struct request
*req
, unsigned int bytes
)
1935 if (blk_do_io_stat(req
)) {
1936 const int rw
= rq_data_dir(req
);
1937 struct hd_struct
*part
;
1940 cpu
= part_stat_lock();
1942 part_stat_add(cpu
, part
, sectors
[rw
], bytes
>> 9);
1947 static void blk_account_io_done(struct request
*req
)
1950 * Account IO completion. flush_rq isn't accounted as a
1951 * normal IO on queueing nor completion. Accounting the
1952 * containing request is enough.
1954 if (blk_do_io_stat(req
) && !(req
->cmd_flags
& REQ_FLUSH_SEQ
)) {
1955 unsigned long duration
= jiffies
- req
->start_time
;
1956 const int rw
= rq_data_dir(req
);
1957 struct hd_struct
*part
;
1960 cpu
= part_stat_lock();
1963 part_stat_inc(cpu
, part
, ios
[rw
]);
1964 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
1965 part_round_stats(cpu
, part
);
1966 part_dec_in_flight(part
, rw
);
1968 hd_struct_put(part
);
1974 * blk_peek_request - peek at the top of a request queue
1975 * @q: request queue to peek at
1978 * Return the request at the top of @q. The returned request
1979 * should be started using blk_start_request() before LLD starts
1983 * Pointer to the request at the top of @q if available. Null
1987 * queue_lock must be held.
1989 struct request
*blk_peek_request(struct request_queue
*q
)
1994 while ((rq
= __elv_next_request(q
)) != NULL
) {
1995 if (!(rq
->cmd_flags
& REQ_STARTED
)) {
1997 * This is the first time the device driver
1998 * sees this request (possibly after
1999 * requeueing). Notify IO scheduler.
2001 if (rq
->cmd_flags
& REQ_SORTED
)
2002 elv_activate_rq(q
, rq
);
2005 * just mark as started even if we don't start
2006 * it, a request that has been delayed should
2007 * not be passed by new incoming requests
2009 rq
->cmd_flags
|= REQ_STARTED
;
2010 trace_block_rq_issue(q
, rq
);
2013 if (!q
->boundary_rq
|| q
->boundary_rq
== rq
) {
2014 q
->end_sector
= rq_end_sector(rq
);
2015 q
->boundary_rq
= NULL
;
2018 if (rq
->cmd_flags
& REQ_DONTPREP
)
2021 if (q
->dma_drain_size
&& blk_rq_bytes(rq
)) {
2023 * make sure space for the drain appears we
2024 * know we can do this because max_hw_segments
2025 * has been adjusted to be one fewer than the
2028 rq
->nr_phys_segments
++;
2034 ret
= q
->prep_rq_fn(q
, rq
);
2035 if (ret
== BLKPREP_OK
) {
2037 } else if (ret
== BLKPREP_DEFER
) {
2039 * the request may have been (partially) prepped.
2040 * we need to keep this request in the front to
2041 * avoid resource deadlock. REQ_STARTED will
2042 * prevent other fs requests from passing this one.
2044 if (q
->dma_drain_size
&& blk_rq_bytes(rq
) &&
2045 !(rq
->cmd_flags
& REQ_DONTPREP
)) {
2047 * remove the space for the drain we added
2048 * so that we don't add it again
2050 --rq
->nr_phys_segments
;
2055 } else if (ret
== BLKPREP_KILL
) {
2056 rq
->cmd_flags
|= REQ_QUIET
;
2058 * Mark this request as started so we don't trigger
2059 * any debug logic in the end I/O path.
2061 blk_start_request(rq
);
2062 __blk_end_request_all(rq
, -EIO
);
2064 printk(KERN_ERR
"%s: bad return=%d\n", __func__
, ret
);
2071 EXPORT_SYMBOL(blk_peek_request
);
2073 void blk_dequeue_request(struct request
*rq
)
2075 struct request_queue
*q
= rq
->q
;
2077 BUG_ON(list_empty(&rq
->queuelist
));
2078 BUG_ON(ELV_ON_HASH(rq
));
2080 list_del_init(&rq
->queuelist
);
2083 * the time frame between a request being removed from the lists
2084 * and to it is freed is accounted as io that is in progress at
2087 if (blk_account_rq(rq
)) {
2088 q
->in_flight
[rq_is_sync(rq
)]++;
2089 set_io_start_time_ns(rq
);
2094 * blk_start_request - start request processing on the driver
2095 * @req: request to dequeue
2098 * Dequeue @req and start timeout timer on it. This hands off the
2099 * request to the driver.
2101 * Block internal functions which don't want to start timer should
2102 * call blk_dequeue_request().
2105 * queue_lock must be held.
2107 void blk_start_request(struct request
*req
)
2109 blk_dequeue_request(req
);
2112 * We are now handing the request to the hardware, initialize
2113 * resid_len to full count and add the timeout handler.
2115 req
->resid_len
= blk_rq_bytes(req
);
2116 if (unlikely(blk_bidi_rq(req
)))
2117 req
->next_rq
->resid_len
= blk_rq_bytes(req
->next_rq
);
2121 EXPORT_SYMBOL(blk_start_request
);
2124 * blk_fetch_request - fetch a request from a request queue
2125 * @q: request queue to fetch a request from
2128 * Return the request at the top of @q. The request is started on
2129 * return and LLD can start processing it immediately.
2132 * Pointer to the request at the top of @q if available. Null
2136 * queue_lock must be held.
2138 struct request
*blk_fetch_request(struct request_queue
*q
)
2142 rq
= blk_peek_request(q
);
2144 blk_start_request(rq
);
2147 EXPORT_SYMBOL(blk_fetch_request
);
2150 * blk_update_request - Special helper function for request stacking drivers
2151 * @req: the request being processed
2152 * @error: %0 for success, < %0 for error
2153 * @nr_bytes: number of bytes to complete @req
2156 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2157 * the request structure even if @req doesn't have leftover.
2158 * If @req has leftover, sets it up for the next range of segments.
2160 * This special helper function is only for request stacking drivers
2161 * (e.g. request-based dm) so that they can handle partial completion.
2162 * Actual device drivers should use blk_end_request instead.
2164 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2165 * %false return from this function.
2168 * %false - this request doesn't have any more data
2169 * %true - this request has more data
2171 bool blk_update_request(struct request
*req
, int error
, unsigned int nr_bytes
)
2173 int total_bytes
, bio_nbytes
, next_idx
= 0;
2179 trace_block_rq_complete(req
->q
, req
);
2182 * For fs requests, rq is just carrier of independent bio's
2183 * and each partial completion should be handled separately.
2184 * Reset per-request error on each partial completion.
2186 * TODO: tj: This is too subtle. It would be better to let
2187 * low level drivers do what they see fit.
2189 if (req
->cmd_type
== REQ_TYPE_FS
)
2192 if (error
&& req
->cmd_type
== REQ_TYPE_FS
&&
2193 !(req
->cmd_flags
& REQ_QUIET
)) {
2198 error_type
= "recoverable transport";
2201 error_type
= "critical target";
2204 error_type
= "critical nexus";
2211 printk(KERN_ERR
"end_request: %s error, dev %s, sector %llu\n",
2212 error_type
, req
->rq_disk
? req
->rq_disk
->disk_name
: "?",
2213 (unsigned long long)blk_rq_pos(req
));
2216 blk_account_io_completion(req
, nr_bytes
);
2218 total_bytes
= bio_nbytes
= 0;
2219 while ((bio
= req
->bio
) != NULL
) {
2222 if (nr_bytes
>= bio
->bi_size
) {
2223 req
->bio
= bio
->bi_next
;
2224 nbytes
= bio
->bi_size
;
2225 req_bio_endio(req
, bio
, nbytes
, error
);
2229 int idx
= bio
->bi_idx
+ next_idx
;
2231 if (unlikely(idx
>= bio
->bi_vcnt
)) {
2232 blk_dump_rq_flags(req
, "__end_that");
2233 printk(KERN_ERR
"%s: bio idx %d >= vcnt %d\n",
2234 __func__
, idx
, bio
->bi_vcnt
);
2238 nbytes
= bio_iovec_idx(bio
, idx
)->bv_len
;
2239 BIO_BUG_ON(nbytes
> bio
->bi_size
);
2242 * not a complete bvec done
2244 if (unlikely(nbytes
> nr_bytes
)) {
2245 bio_nbytes
+= nr_bytes
;
2246 total_bytes
+= nr_bytes
;
2251 * advance to the next vector
2254 bio_nbytes
+= nbytes
;
2257 total_bytes
+= nbytes
;
2263 * end more in this run, or just return 'not-done'
2265 if (unlikely(nr_bytes
<= 0))
2275 * Reset counters so that the request stacking driver
2276 * can find how many bytes remain in the request
2279 req
->__data_len
= 0;
2284 * if the request wasn't completed, update state
2287 req_bio_endio(req
, bio
, bio_nbytes
, error
);
2288 bio
->bi_idx
+= next_idx
;
2289 bio_iovec(bio
)->bv_offset
+= nr_bytes
;
2290 bio_iovec(bio
)->bv_len
-= nr_bytes
;
2293 req
->__data_len
-= total_bytes
;
2294 req
->buffer
= bio_data(req
->bio
);
2296 /* update sector only for requests with clear definition of sector */
2297 if (req
->cmd_type
== REQ_TYPE_FS
|| (req
->cmd_flags
& REQ_DISCARD
))
2298 req
->__sector
+= total_bytes
>> 9;
2300 /* mixed attributes always follow the first bio */
2301 if (req
->cmd_flags
& REQ_MIXED_MERGE
) {
2302 req
->cmd_flags
&= ~REQ_FAILFAST_MASK
;
2303 req
->cmd_flags
|= req
->bio
->bi_rw
& REQ_FAILFAST_MASK
;
2307 * If total number of sectors is less than the first segment
2308 * size, something has gone terribly wrong.
2310 if (blk_rq_bytes(req
) < blk_rq_cur_bytes(req
)) {
2311 blk_dump_rq_flags(req
, "request botched");
2312 req
->__data_len
= blk_rq_cur_bytes(req
);
2315 /* recalculate the number of segments */
2316 blk_recalc_rq_segments(req
);
2320 EXPORT_SYMBOL_GPL(blk_update_request
);
2322 static bool blk_update_bidi_request(struct request
*rq
, int error
,
2323 unsigned int nr_bytes
,
2324 unsigned int bidi_bytes
)
2326 if (blk_update_request(rq
, error
, nr_bytes
))
2329 /* Bidi request must be completed as a whole */
2330 if (unlikely(blk_bidi_rq(rq
)) &&
2331 blk_update_request(rq
->next_rq
, error
, bidi_bytes
))
2334 if (blk_queue_add_random(rq
->q
))
2335 add_disk_randomness(rq
->rq_disk
);
2341 * blk_unprep_request - unprepare a request
2344 * This function makes a request ready for complete resubmission (or
2345 * completion). It happens only after all error handling is complete,
2346 * so represents the appropriate moment to deallocate any resources
2347 * that were allocated to the request in the prep_rq_fn. The queue
2348 * lock is held when calling this.
2350 void blk_unprep_request(struct request
*req
)
2352 struct request_queue
*q
= req
->q
;
2354 req
->cmd_flags
&= ~REQ_DONTPREP
;
2355 if (q
->unprep_rq_fn
)
2356 q
->unprep_rq_fn(q
, req
);
2358 EXPORT_SYMBOL_GPL(blk_unprep_request
);
2361 * queue lock must be held
2363 static void blk_finish_request(struct request
*req
, int error
)
2365 if (blk_rq_tagged(req
))
2366 blk_queue_end_tag(req
->q
, req
);
2368 BUG_ON(blk_queued_rq(req
));
2370 if (unlikely(laptop_mode
) && req
->cmd_type
== REQ_TYPE_FS
)
2371 laptop_io_completion(&req
->q
->backing_dev_info
);
2373 blk_delete_timer(req
);
2375 if (req
->cmd_flags
& REQ_DONTPREP
)
2376 blk_unprep_request(req
);
2379 blk_account_io_done(req
);
2382 req
->end_io(req
, error
);
2384 if (blk_bidi_rq(req
))
2385 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
2387 __blk_put_request(req
->q
, req
);
2392 * blk_end_bidi_request - Complete a bidi request
2393 * @rq: the request to complete
2394 * @error: %0 for success, < %0 for error
2395 * @nr_bytes: number of bytes to complete @rq
2396 * @bidi_bytes: number of bytes to complete @rq->next_rq
2399 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2400 * Drivers that supports bidi can safely call this member for any
2401 * type of request, bidi or uni. In the later case @bidi_bytes is
2405 * %false - we are done with this request
2406 * %true - still buffers pending for this request
2408 static bool blk_end_bidi_request(struct request
*rq
, int error
,
2409 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2411 struct request_queue
*q
= rq
->q
;
2412 unsigned long flags
;
2414 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2417 spin_lock_irqsave(q
->queue_lock
, flags
);
2418 blk_finish_request(rq
, error
);
2419 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2425 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2426 * @rq: the request to complete
2427 * @error: %0 for success, < %0 for error
2428 * @nr_bytes: number of bytes to complete @rq
2429 * @bidi_bytes: number of bytes to complete @rq->next_rq
2432 * Identical to blk_end_bidi_request() except that queue lock is
2433 * assumed to be locked on entry and remains so on return.
2436 * %false - we are done with this request
2437 * %true - still buffers pending for this request
2439 bool __blk_end_bidi_request(struct request
*rq
, int error
,
2440 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2442 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2445 blk_finish_request(rq
, error
);
2451 * blk_end_request - Helper function for drivers to complete the request.
2452 * @rq: the request being processed
2453 * @error: %0 for success, < %0 for error
2454 * @nr_bytes: number of bytes to complete
2457 * Ends I/O on a number of bytes attached to @rq.
2458 * If @rq has leftover, sets it up for the next range of segments.
2461 * %false - we are done with this request
2462 * %true - still buffers pending for this request
2464 bool blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2466 return blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2468 EXPORT_SYMBOL(blk_end_request
);
2471 * blk_end_request_all - Helper function for drives to finish the request.
2472 * @rq: the request to finish
2473 * @error: %0 for success, < %0 for error
2476 * Completely finish @rq.
2478 void blk_end_request_all(struct request
*rq
, int error
)
2481 unsigned int bidi_bytes
= 0;
2483 if (unlikely(blk_bidi_rq(rq
)))
2484 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2486 pending
= blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2489 EXPORT_SYMBOL(blk_end_request_all
);
2492 * blk_end_request_cur - Helper function to finish the current request chunk.
2493 * @rq: the request to finish the current chunk for
2494 * @error: %0 for success, < %0 for error
2497 * Complete the current consecutively mapped chunk from @rq.
2500 * %false - we are done with this request
2501 * %true - still buffers pending for this request
2503 bool blk_end_request_cur(struct request
*rq
, int error
)
2505 return blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2507 EXPORT_SYMBOL(blk_end_request_cur
);
2510 * blk_end_request_err - Finish a request till the next failure boundary.
2511 * @rq: the request to finish till the next failure boundary for
2512 * @error: must be negative errno
2515 * Complete @rq till the next failure boundary.
2518 * %false - we are done with this request
2519 * %true - still buffers pending for this request
2521 bool blk_end_request_err(struct request
*rq
, int error
)
2523 WARN_ON(error
>= 0);
2524 return blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2526 EXPORT_SYMBOL_GPL(blk_end_request_err
);
2529 * __blk_end_request - Helper function for drivers to complete the request.
2530 * @rq: the request being processed
2531 * @error: %0 for success, < %0 for error
2532 * @nr_bytes: number of bytes to complete
2535 * Must be called with queue lock held unlike blk_end_request().
2538 * %false - we are done with this request
2539 * %true - still buffers pending for this request
2541 bool __blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2543 return __blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2545 EXPORT_SYMBOL(__blk_end_request
);
2548 * __blk_end_request_all - Helper function for drives to finish the request.
2549 * @rq: the request to finish
2550 * @error: %0 for success, < %0 for error
2553 * Completely finish @rq. Must be called with queue lock held.
2555 void __blk_end_request_all(struct request
*rq
, int error
)
2558 unsigned int bidi_bytes
= 0;
2560 if (unlikely(blk_bidi_rq(rq
)))
2561 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2563 pending
= __blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2566 EXPORT_SYMBOL(__blk_end_request_all
);
2569 * __blk_end_request_cur - Helper function to finish the current request chunk.
2570 * @rq: the request to finish the current chunk for
2571 * @error: %0 for success, < %0 for error
2574 * Complete the current consecutively mapped chunk from @rq. Must
2575 * be called with queue lock held.
2578 * %false - we are done with this request
2579 * %true - still buffers pending for this request
2581 bool __blk_end_request_cur(struct request
*rq
, int error
)
2583 return __blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2585 EXPORT_SYMBOL(__blk_end_request_cur
);
2588 * __blk_end_request_err - Finish a request till the next failure boundary.
2589 * @rq: the request to finish till the next failure boundary for
2590 * @error: must be negative errno
2593 * Complete @rq till the next failure boundary. Must be called
2594 * with queue lock held.
2597 * %false - we are done with this request
2598 * %true - still buffers pending for this request
2600 bool __blk_end_request_err(struct request
*rq
, int error
)
2602 WARN_ON(error
>= 0);
2603 return __blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2605 EXPORT_SYMBOL_GPL(__blk_end_request_err
);
2607 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
2610 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2611 rq
->cmd_flags
|= bio
->bi_rw
& REQ_WRITE
;
2613 if (bio_has_data(bio
)) {
2614 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2615 rq
->buffer
= bio_data(bio
);
2617 rq
->__data_len
= bio
->bi_size
;
2618 rq
->bio
= rq
->biotail
= bio
;
2621 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2624 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2626 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2627 * @rq: the request to be flushed
2630 * Flush all pages in @rq.
2632 void rq_flush_dcache_pages(struct request
*rq
)
2634 struct req_iterator iter
;
2635 struct bio_vec
*bvec
;
2637 rq_for_each_segment(bvec
, rq
, iter
)
2638 flush_dcache_page(bvec
->bv_page
);
2640 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages
);
2644 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2645 * @q : the queue of the device being checked
2648 * Check if underlying low-level drivers of a device are busy.
2649 * If the drivers want to export their busy state, they must set own
2650 * exporting function using blk_queue_lld_busy() first.
2652 * Basically, this function is used only by request stacking drivers
2653 * to stop dispatching requests to underlying devices when underlying
2654 * devices are busy. This behavior helps more I/O merging on the queue
2655 * of the request stacking driver and prevents I/O throughput regression
2656 * on burst I/O load.
2659 * 0 - Not busy (The request stacking driver should dispatch request)
2660 * 1 - Busy (The request stacking driver should stop dispatching request)
2662 int blk_lld_busy(struct request_queue
*q
)
2665 return q
->lld_busy_fn(q
);
2669 EXPORT_SYMBOL_GPL(blk_lld_busy
);
2672 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2673 * @rq: the clone request to be cleaned up
2676 * Free all bios in @rq for a cloned request.
2678 void blk_rq_unprep_clone(struct request
*rq
)
2682 while ((bio
= rq
->bio
) != NULL
) {
2683 rq
->bio
= bio
->bi_next
;
2688 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone
);
2691 * Copy attributes of the original request to the clone request.
2692 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2694 static void __blk_rq_prep_clone(struct request
*dst
, struct request
*src
)
2696 dst
->cpu
= src
->cpu
;
2697 dst
->cmd_flags
= (src
->cmd_flags
& REQ_CLONE_MASK
) | REQ_NOMERGE
;
2698 dst
->cmd_type
= src
->cmd_type
;
2699 dst
->__sector
= blk_rq_pos(src
);
2700 dst
->__data_len
= blk_rq_bytes(src
);
2701 dst
->nr_phys_segments
= src
->nr_phys_segments
;
2702 dst
->ioprio
= src
->ioprio
;
2703 dst
->extra_len
= src
->extra_len
;
2707 * blk_rq_prep_clone - Helper function to setup clone request
2708 * @rq: the request to be setup
2709 * @rq_src: original request to be cloned
2710 * @bs: bio_set that bios for clone are allocated from
2711 * @gfp_mask: memory allocation mask for bio
2712 * @bio_ctr: setup function to be called for each clone bio.
2713 * Returns %0 for success, non %0 for failure.
2714 * @data: private data to be passed to @bio_ctr
2717 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2718 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2719 * are not copied, and copying such parts is the caller's responsibility.
2720 * Also, pages which the original bios are pointing to are not copied
2721 * and the cloned bios just point same pages.
2722 * So cloned bios must be completed before original bios, which means
2723 * the caller must complete @rq before @rq_src.
2725 int blk_rq_prep_clone(struct request
*rq
, struct request
*rq_src
,
2726 struct bio_set
*bs
, gfp_t gfp_mask
,
2727 int (*bio_ctr
)(struct bio
*, struct bio
*, void *),
2730 struct bio
*bio
, *bio_src
;
2735 blk_rq_init(NULL
, rq
);
2737 __rq_for_each_bio(bio_src
, rq_src
) {
2738 bio
= bio_alloc_bioset(gfp_mask
, bio_src
->bi_max_vecs
, bs
);
2742 __bio_clone(bio
, bio_src
);
2744 if (bio_integrity(bio_src
) &&
2745 bio_integrity_clone(bio
, bio_src
, gfp_mask
, bs
))
2748 if (bio_ctr
&& bio_ctr(bio
, bio_src
, data
))
2752 rq
->biotail
->bi_next
= bio
;
2755 rq
->bio
= rq
->biotail
= bio
;
2758 __blk_rq_prep_clone(rq
, rq_src
);
2765 blk_rq_unprep_clone(rq
);
2769 EXPORT_SYMBOL_GPL(blk_rq_prep_clone
);
2771 int kblockd_schedule_work(struct request_queue
*q
, struct work_struct
*work
)
2773 return queue_work(kblockd_workqueue
, work
);
2775 EXPORT_SYMBOL(kblockd_schedule_work
);
2777 int kblockd_schedule_delayed_work(struct request_queue
*q
,
2778 struct delayed_work
*dwork
, unsigned long delay
)
2780 return queue_delayed_work(kblockd_workqueue
, dwork
, delay
);
2782 EXPORT_SYMBOL(kblockd_schedule_delayed_work
);
2784 #define PLUG_MAGIC 0x91827364
2787 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2788 * @plug: The &struct blk_plug that needs to be initialized
2791 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2792 * pending I/O should the task end up blocking between blk_start_plug() and
2793 * blk_finish_plug(). This is important from a performance perspective, but
2794 * also ensures that we don't deadlock. For instance, if the task is blocking
2795 * for a memory allocation, memory reclaim could end up wanting to free a
2796 * page belonging to that request that is currently residing in our private
2797 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2798 * this kind of deadlock.
2800 void blk_start_plug(struct blk_plug
*plug
)
2802 struct task_struct
*tsk
= current
;
2804 plug
->magic
= PLUG_MAGIC
;
2805 INIT_LIST_HEAD(&plug
->list
);
2806 INIT_LIST_HEAD(&plug
->cb_list
);
2807 plug
->should_sort
= 0;
2810 * If this is a nested plug, don't actually assign it. It will be
2811 * flushed on its own.
2815 * Store ordering should not be needed here, since a potential
2816 * preempt will imply a full memory barrier
2821 EXPORT_SYMBOL(blk_start_plug
);
2823 static int plug_rq_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
2825 struct request
*rqa
= container_of(a
, struct request
, queuelist
);
2826 struct request
*rqb
= container_of(b
, struct request
, queuelist
);
2828 return !(rqa
->q
<= rqb
->q
);
2832 * If 'from_schedule' is true, then postpone the dispatch of requests
2833 * until a safe kblockd context. We due this to avoid accidental big
2834 * additional stack usage in driver dispatch, in places where the originally
2835 * plugger did not intend it.
2837 static void queue_unplugged(struct request_queue
*q
, unsigned int depth
,
2839 __releases(q
->queue_lock
)
2841 trace_block_unplug(q
, depth
, !from_schedule
);
2844 * Don't mess with dead queue.
2846 if (unlikely(blk_queue_dead(q
))) {
2847 spin_unlock(q
->queue_lock
);
2852 * If we are punting this to kblockd, then we can safely drop
2853 * the queue_lock before waking kblockd (which needs to take
2856 if (from_schedule
) {
2857 spin_unlock(q
->queue_lock
);
2858 blk_run_queue_async(q
);
2861 spin_unlock(q
->queue_lock
);
2866 static void flush_plug_callbacks(struct blk_plug
*plug
)
2868 LIST_HEAD(callbacks
);
2870 if (list_empty(&plug
->cb_list
))
2873 list_splice_init(&plug
->cb_list
, &callbacks
);
2875 while (!list_empty(&callbacks
)) {
2876 struct blk_plug_cb
*cb
= list_first_entry(&callbacks
,
2879 list_del(&cb
->list
);
2884 void blk_flush_plug_list(struct blk_plug
*plug
, bool from_schedule
)
2886 struct request_queue
*q
;
2887 unsigned long flags
;
2892 BUG_ON(plug
->magic
!= PLUG_MAGIC
);
2894 flush_plug_callbacks(plug
);
2895 if (list_empty(&plug
->list
))
2898 list_splice_init(&plug
->list
, &list
);
2900 if (plug
->should_sort
) {
2901 list_sort(NULL
, &list
, plug_rq_cmp
);
2902 plug
->should_sort
= 0;
2909 * Save and disable interrupts here, to avoid doing it for every
2910 * queue lock we have to take.
2912 local_irq_save(flags
);
2913 while (!list_empty(&list
)) {
2914 rq
= list_entry_rq(list
.next
);
2915 list_del_init(&rq
->queuelist
);
2919 * This drops the queue lock
2922 queue_unplugged(q
, depth
, from_schedule
);
2925 spin_lock(q
->queue_lock
);
2929 * Short-circuit if @q is dead
2931 if (unlikely(blk_queue_dead(q
))) {
2932 __blk_end_request_all(rq
, -ENODEV
);
2937 * rq is already accounted, so use raw insert
2939 if (rq
->cmd_flags
& (REQ_FLUSH
| REQ_FUA
))
2940 __elv_add_request(q
, rq
, ELEVATOR_INSERT_FLUSH
);
2942 __elv_add_request(q
, rq
, ELEVATOR_INSERT_SORT_MERGE
);
2948 * This drops the queue lock
2951 queue_unplugged(q
, depth
, from_schedule
);
2953 local_irq_restore(flags
);
2956 void blk_finish_plug(struct blk_plug
*plug
)
2958 blk_flush_plug_list(plug
, false);
2960 if (plug
== current
->plug
)
2961 current
->plug
= NULL
;
2963 EXPORT_SYMBOL(blk_finish_plug
);
2965 int __init
blk_dev_init(void)
2967 BUILD_BUG_ON(__REQ_NR_BITS
> 8 *
2968 sizeof(((struct request
*)0)->cmd_flags
));
2970 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
2971 kblockd_workqueue
= alloc_workqueue("kblockd",
2972 WQ_MEM_RECLAIM
| WQ_HIGHPRI
, 0);
2973 if (!kblockd_workqueue
)
2974 panic("Failed to create kblockd\n");
2976 request_cachep
= kmem_cache_create("blkdev_requests",
2977 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
2979 blk_requestq_cachep
= kmem_cache_create("blkdev_queue",
2980 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);