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>
32 #include <linux/ratelimit.h>
34 #define CREATE_TRACE_POINTS
35 #include <trace/events/block.h>
38 #include "blk-cgroup.h"
40 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap
);
41 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap
);
42 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete
);
44 DEFINE_IDA(blk_queue_ida
);
47 * For the allocated request tables
49 static struct kmem_cache
*request_cachep
;
52 * For queue allocation
54 struct kmem_cache
*blk_requestq_cachep
;
57 * Controlling structure to kblockd
59 static struct workqueue_struct
*kblockd_workqueue
;
61 static void drive_stat_acct(struct request
*rq
, int new_io
)
63 struct hd_struct
*part
;
64 int rw
= rq_data_dir(rq
);
67 if (!blk_do_io_stat(rq
))
70 cpu
= part_stat_lock();
74 part_stat_inc(cpu
, part
, merges
[rw
]);
76 part
= disk_map_sector_rcu(rq
->rq_disk
, blk_rq_pos(rq
));
77 if (!hd_struct_try_get(part
)) {
79 * The partition is already being removed,
80 * the request will be accounted on the disk only
82 * We take a reference on disk->part0 although that
83 * partition will never be deleted, so we can treat
84 * it as any other partition.
86 part
= &rq
->rq_disk
->part0
;
89 part_round_stats(cpu
, part
);
90 part_inc_in_flight(part
, rw
);
97 void blk_queue_congestion_threshold(struct request_queue
*q
)
101 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) + 1;
102 if (nr
> q
->nr_requests
)
104 q
->nr_congestion_on
= nr
;
106 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) - (q
->nr_requests
/ 16) - 1;
109 q
->nr_congestion_off
= nr
;
113 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
116 * Locates the passed device's request queue and returns the address of its
119 * Will return NULL if the request queue cannot be located.
121 struct backing_dev_info
*blk_get_backing_dev_info(struct block_device
*bdev
)
123 struct backing_dev_info
*ret
= NULL
;
124 struct request_queue
*q
= bdev_get_queue(bdev
);
127 ret
= &q
->backing_dev_info
;
130 EXPORT_SYMBOL(blk_get_backing_dev_info
);
132 void blk_rq_init(struct request_queue
*q
, struct request
*rq
)
134 memset(rq
, 0, sizeof(*rq
));
136 INIT_LIST_HEAD(&rq
->queuelist
);
137 INIT_LIST_HEAD(&rq
->timeout_list
);
140 rq
->__sector
= (sector_t
) -1;
141 INIT_HLIST_NODE(&rq
->hash
);
142 RB_CLEAR_NODE(&rq
->rb_node
);
144 rq
->cmd_len
= BLK_MAX_CDB
;
147 rq
->start_time
= jiffies
;
148 set_start_time_ns(rq
);
151 EXPORT_SYMBOL(blk_rq_init
);
153 static void req_bio_endio(struct request
*rq
, struct bio
*bio
,
154 unsigned int nbytes
, int error
)
157 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
158 else if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
161 if (unlikely(nbytes
> bio
->bi_size
)) {
162 printk(KERN_ERR
"%s: want %u bytes done, %u left\n",
163 __func__
, nbytes
, bio
->bi_size
);
164 nbytes
= bio
->bi_size
;
167 if (unlikely(rq
->cmd_flags
& REQ_QUIET
))
168 set_bit(BIO_QUIET
, &bio
->bi_flags
);
170 bio
->bi_size
-= nbytes
;
171 bio
->bi_sector
+= (nbytes
>> 9);
173 if (bio_integrity(bio
))
174 bio_integrity_advance(bio
, nbytes
);
176 /* don't actually finish bio if it's part of flush sequence */
177 if (bio
->bi_size
== 0 && !(rq
->cmd_flags
& REQ_FLUSH_SEQ
))
178 bio_endio(bio
, error
);
181 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
185 printk(KERN_INFO
"%s: dev %s: type=%x, flags=%x\n", msg
,
186 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->cmd_type
,
189 printk(KERN_INFO
" sector %llu, nr/cnr %u/%u\n",
190 (unsigned long long)blk_rq_pos(rq
),
191 blk_rq_sectors(rq
), blk_rq_cur_sectors(rq
));
192 printk(KERN_INFO
" bio %p, biotail %p, buffer %p, len %u\n",
193 rq
->bio
, rq
->biotail
, rq
->buffer
, blk_rq_bytes(rq
));
195 if (rq
->cmd_type
== REQ_TYPE_BLOCK_PC
) {
196 printk(KERN_INFO
" cdb: ");
197 for (bit
= 0; bit
< BLK_MAX_CDB
; bit
++)
198 printk("%02x ", rq
->cmd
[bit
]);
202 EXPORT_SYMBOL(blk_dump_rq_flags
);
204 static void blk_delay_work(struct work_struct
*work
)
206 struct request_queue
*q
;
208 q
= container_of(work
, struct request_queue
, delay_work
.work
);
209 spin_lock_irq(q
->queue_lock
);
211 spin_unlock_irq(q
->queue_lock
);
215 * blk_delay_queue - restart queueing after defined interval
216 * @q: The &struct request_queue in question
217 * @msecs: Delay in msecs
220 * Sometimes queueing needs to be postponed for a little while, to allow
221 * resources to come back. This function will make sure that queueing is
222 * restarted around the specified time.
224 void blk_delay_queue(struct request_queue
*q
, unsigned long msecs
)
226 queue_delayed_work(kblockd_workqueue
, &q
->delay_work
,
227 msecs_to_jiffies(msecs
));
229 EXPORT_SYMBOL(blk_delay_queue
);
232 * blk_start_queue - restart a previously stopped queue
233 * @q: The &struct request_queue in question
236 * blk_start_queue() will clear the stop flag on the queue, and call
237 * the request_fn for the queue if it was in a stopped state when
238 * entered. Also see blk_stop_queue(). Queue lock must be held.
240 void blk_start_queue(struct request_queue
*q
)
242 WARN_ON(!irqs_disabled());
244 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
247 EXPORT_SYMBOL(blk_start_queue
);
250 * blk_stop_queue - stop a queue
251 * @q: The &struct request_queue in question
254 * The Linux block layer assumes that a block driver will consume all
255 * entries on the request queue when the request_fn strategy is called.
256 * Often this will not happen, because of hardware limitations (queue
257 * depth settings). If a device driver gets a 'queue full' response,
258 * or if it simply chooses not to queue more I/O at one point, it can
259 * call this function to prevent the request_fn from being called until
260 * the driver has signalled it's ready to go again. This happens by calling
261 * blk_start_queue() to restart queue operations. Queue lock must be held.
263 void blk_stop_queue(struct request_queue
*q
)
265 __cancel_delayed_work(&q
->delay_work
);
266 queue_flag_set(QUEUE_FLAG_STOPPED
, q
);
268 EXPORT_SYMBOL(blk_stop_queue
);
271 * blk_sync_queue - cancel any pending callbacks on a queue
275 * The block layer may perform asynchronous callback activity
276 * on a queue, such as calling the unplug function after a timeout.
277 * A block device may call blk_sync_queue to ensure that any
278 * such activity is cancelled, thus allowing it to release resources
279 * that the callbacks might use. The caller must already have made sure
280 * that its ->make_request_fn will not re-add plugging prior to calling
283 * This function does not cancel any asynchronous activity arising
284 * out of elevator or throttling code. That would require elevaotor_exit()
285 * and blkcg_exit_queue() to be called with queue lock initialized.
288 void blk_sync_queue(struct request_queue
*q
)
290 del_timer_sync(&q
->timeout
);
291 cancel_delayed_work_sync(&q
->delay_work
);
293 EXPORT_SYMBOL(blk_sync_queue
);
296 * __blk_run_queue - run a single device queue
297 * @q: The queue to run
300 * See @blk_run_queue. This variant must be called with the queue lock
301 * held and interrupts disabled.
303 void __blk_run_queue(struct request_queue
*q
)
305 if (unlikely(blk_queue_stopped(q
)))
310 EXPORT_SYMBOL(__blk_run_queue
);
313 * blk_run_queue_async - run a single device queue in workqueue context
314 * @q: The queue to run
317 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
320 void blk_run_queue_async(struct request_queue
*q
)
322 if (likely(!blk_queue_stopped(q
))) {
323 __cancel_delayed_work(&q
->delay_work
);
324 queue_delayed_work(kblockd_workqueue
, &q
->delay_work
, 0);
327 EXPORT_SYMBOL(blk_run_queue_async
);
330 * blk_run_queue - run a single device queue
331 * @q: The queue to run
334 * Invoke request handling on this queue, if it has pending work to do.
335 * May be used to restart queueing when a request has completed.
337 void blk_run_queue(struct request_queue
*q
)
341 spin_lock_irqsave(q
->queue_lock
, flags
);
343 spin_unlock_irqrestore(q
->queue_lock
, flags
);
345 EXPORT_SYMBOL(blk_run_queue
);
347 void blk_put_queue(struct request_queue
*q
)
349 kobject_put(&q
->kobj
);
351 EXPORT_SYMBOL(blk_put_queue
);
354 * blk_drain_queue - drain requests from request_queue
356 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
358 * Drain requests from @q. If @drain_all is set, all requests are drained.
359 * If not, only ELVPRIV requests are drained. The caller is responsible
360 * for ensuring that no new requests which need to be drained are queued.
362 void blk_drain_queue(struct request_queue
*q
, bool drain_all
)
369 spin_lock_irq(q
->queue_lock
);
372 * The caller might be trying to drain @q before its
373 * elevator is initialized.
376 elv_drain_elevator(q
);
378 blkcg_drain_queue(q
);
381 * This function might be called on a queue which failed
382 * driver init after queue creation or is not yet fully
383 * active yet. Some drivers (e.g. fd and loop) get unhappy
384 * in such cases. Kick queue iff dispatch queue has
385 * something on it and @q has request_fn set.
387 if (!list_empty(&q
->queue_head
) && q
->request_fn
)
390 drain
|= q
->nr_rqs_elvpriv
;
393 * Unfortunately, requests are queued at and tracked from
394 * multiple places and there's no single counter which can
395 * be drained. Check all the queues and counters.
398 drain
|= !list_empty(&q
->queue_head
);
399 for (i
= 0; i
< 2; i
++) {
400 drain
|= q
->nr_rqs
[i
];
401 drain
|= q
->in_flight
[i
];
402 drain
|= !list_empty(&q
->flush_queue
[i
]);
406 spin_unlock_irq(q
->queue_lock
);
414 * With queue marked dead, any woken up waiter will fail the
415 * allocation path, so the wakeup chaining is lost and we're
416 * left with hung waiters. We need to wake up those waiters.
419 spin_lock_irq(q
->queue_lock
);
420 for (i
= 0; i
< ARRAY_SIZE(q
->rq
.wait
); i
++)
421 wake_up_all(&q
->rq
.wait
[i
]);
422 spin_unlock_irq(q
->queue_lock
);
427 * blk_queue_bypass_start - enter queue bypass mode
428 * @q: queue of interest
430 * In bypass mode, only the dispatch FIFO queue of @q is used. This
431 * function makes @q enter bypass mode and drains all requests which were
432 * throttled or issued before. On return, it's guaranteed that no request
433 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
434 * inside queue or RCU read lock.
436 void blk_queue_bypass_start(struct request_queue
*q
)
440 spin_lock_irq(q
->queue_lock
);
441 drain
= !q
->bypass_depth
++;
442 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
443 spin_unlock_irq(q
->queue_lock
);
446 blk_drain_queue(q
, false);
447 /* ensure blk_queue_bypass() is %true inside RCU read lock */
451 EXPORT_SYMBOL_GPL(blk_queue_bypass_start
);
454 * blk_queue_bypass_end - leave queue bypass mode
455 * @q: queue of interest
457 * Leave bypass mode and restore the normal queueing behavior.
459 void blk_queue_bypass_end(struct request_queue
*q
)
461 spin_lock_irq(q
->queue_lock
);
462 if (!--q
->bypass_depth
)
463 queue_flag_clear(QUEUE_FLAG_BYPASS
, q
);
464 WARN_ON_ONCE(q
->bypass_depth
< 0);
465 spin_unlock_irq(q
->queue_lock
);
467 EXPORT_SYMBOL_GPL(blk_queue_bypass_end
);
470 * blk_cleanup_queue - shutdown a request queue
471 * @q: request queue to shutdown
473 * Mark @q DEAD, drain all pending requests, destroy and put it. All
474 * future requests will be failed immediately with -ENODEV.
476 void blk_cleanup_queue(struct request_queue
*q
)
478 spinlock_t
*lock
= q
->queue_lock
;
480 /* mark @q DEAD, no new request or merges will be allowed afterwards */
481 mutex_lock(&q
->sysfs_lock
);
482 queue_flag_set_unlocked(QUEUE_FLAG_DEAD
, q
);
486 * Dead queue is permanently in bypass mode till released. Note
487 * that, unlike blk_queue_bypass_start(), we aren't performing
488 * synchronize_rcu() after entering bypass mode to avoid the delay
489 * as some drivers create and destroy a lot of queues while
490 * probing. This is still safe because blk_release_queue() will be
491 * called only after the queue refcnt drops to zero and nothing,
492 * RCU or not, would be traversing the queue by then.
495 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
497 queue_flag_set(QUEUE_FLAG_NOMERGES
, q
);
498 queue_flag_set(QUEUE_FLAG_NOXMERGES
, q
);
499 queue_flag_set(QUEUE_FLAG_DEAD
, q
);
500 spin_unlock_irq(lock
);
501 mutex_unlock(&q
->sysfs_lock
);
503 /* drain all requests queued before DEAD marking */
504 blk_drain_queue(q
, true);
506 /* @q won't process any more request, flush async actions */
507 del_timer_sync(&q
->backing_dev_info
.laptop_mode_wb_timer
);
511 if (q
->queue_lock
!= &q
->__queue_lock
)
512 q
->queue_lock
= &q
->__queue_lock
;
513 spin_unlock_irq(lock
);
515 /* @q is and will stay empty, shutdown and put */
518 EXPORT_SYMBOL(blk_cleanup_queue
);
520 int blk_init_rl(struct request_list
*rl
, struct request_queue
*q
,
523 if (unlikely(rl
->rq_pool
))
527 rl
->count
[BLK_RW_SYNC
] = rl
->count
[BLK_RW_ASYNC
] = 0;
528 rl
->starved
[BLK_RW_SYNC
] = rl
->starved
[BLK_RW_ASYNC
] = 0;
529 init_waitqueue_head(&rl
->wait
[BLK_RW_SYNC
]);
530 init_waitqueue_head(&rl
->wait
[BLK_RW_ASYNC
]);
532 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, mempool_alloc_slab
,
533 mempool_free_slab
, request_cachep
,
541 void blk_exit_rl(struct request_list
*rl
)
544 mempool_destroy(rl
->rq_pool
);
547 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
549 return blk_alloc_queue_node(gfp_mask
, -1);
551 EXPORT_SYMBOL(blk_alloc_queue
);
553 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
555 struct request_queue
*q
;
558 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
559 gfp_mask
| __GFP_ZERO
, node_id
);
563 q
->id
= ida_simple_get(&blk_queue_ida
, 0, 0, gfp_mask
);
567 q
->backing_dev_info
.ra_pages
=
568 (VM_MAX_READAHEAD
* 1024) / PAGE_CACHE_SIZE
;
569 q
->backing_dev_info
.state
= 0;
570 q
->backing_dev_info
.capabilities
= BDI_CAP_MAP_COPY
;
571 q
->backing_dev_info
.name
= "block";
574 err
= bdi_init(&q
->backing_dev_info
);
578 setup_timer(&q
->backing_dev_info
.laptop_mode_wb_timer
,
579 laptop_mode_timer_fn
, (unsigned long) q
);
580 setup_timer(&q
->timeout
, blk_rq_timed_out_timer
, (unsigned long) q
);
581 INIT_LIST_HEAD(&q
->queue_head
);
582 INIT_LIST_HEAD(&q
->timeout_list
);
583 INIT_LIST_HEAD(&q
->icq_list
);
584 #ifdef CONFIG_BLK_CGROUP
585 INIT_LIST_HEAD(&q
->blkg_list
);
587 INIT_LIST_HEAD(&q
->flush_queue
[0]);
588 INIT_LIST_HEAD(&q
->flush_queue
[1]);
589 INIT_LIST_HEAD(&q
->flush_data_in_flight
);
590 INIT_DELAYED_WORK(&q
->delay_work
, blk_delay_work
);
592 kobject_init(&q
->kobj
, &blk_queue_ktype
);
594 mutex_init(&q
->sysfs_lock
);
595 spin_lock_init(&q
->__queue_lock
);
598 * By default initialize queue_lock to internal lock and driver can
599 * override it later if need be.
601 q
->queue_lock
= &q
->__queue_lock
;
604 * A queue starts its life with bypass turned on to avoid
605 * unnecessary bypass on/off overhead and nasty surprises during
606 * init. The initial bypass will be finished at the end of
607 * blk_init_allocated_queue().
610 __set_bit(QUEUE_FLAG_BYPASS
, &q
->queue_flags
);
612 if (blkcg_init_queue(q
))
618 ida_simple_remove(&blk_queue_ida
, q
->id
);
620 kmem_cache_free(blk_requestq_cachep
, q
);
623 EXPORT_SYMBOL(blk_alloc_queue_node
);
626 * blk_init_queue - prepare a request queue for use with a block device
627 * @rfn: The function to be called to process requests that have been
628 * placed on the queue.
629 * @lock: Request queue spin lock
632 * If a block device wishes to use the standard request handling procedures,
633 * which sorts requests and coalesces adjacent requests, then it must
634 * call blk_init_queue(). The function @rfn will be called when there
635 * are requests on the queue that need to be processed. If the device
636 * supports plugging, then @rfn may not be called immediately when requests
637 * are available on the queue, but may be called at some time later instead.
638 * Plugged queues are generally unplugged when a buffer belonging to one
639 * of the requests on the queue is needed, or due to memory pressure.
641 * @rfn is not required, or even expected, to remove all requests off the
642 * queue, but only as many as it can handle at a time. If it does leave
643 * requests on the queue, it is responsible for arranging that the requests
644 * get dealt with eventually.
646 * The queue spin lock must be held while manipulating the requests on the
647 * request queue; this lock will be taken also from interrupt context, so irq
648 * disabling is needed for it.
650 * Function returns a pointer to the initialized request queue, or %NULL if
654 * blk_init_queue() must be paired with a blk_cleanup_queue() call
655 * when the block device is deactivated (such as at module unload).
658 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
660 return blk_init_queue_node(rfn
, lock
, -1);
662 EXPORT_SYMBOL(blk_init_queue
);
664 struct request_queue
*
665 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
667 struct request_queue
*uninit_q
, *q
;
669 uninit_q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
673 q
= blk_init_allocated_queue(uninit_q
, rfn
, lock
);
675 blk_cleanup_queue(uninit_q
);
679 EXPORT_SYMBOL(blk_init_queue_node
);
681 struct request_queue
*
682 blk_init_allocated_queue(struct request_queue
*q
, request_fn_proc
*rfn
,
688 if (blk_init_rl(&q
->rq
, q
, GFP_KERNEL
))
692 q
->prep_rq_fn
= NULL
;
693 q
->unprep_rq_fn
= NULL
;
694 q
->queue_flags
= QUEUE_FLAG_DEFAULT
;
696 /* Override internal queue lock with supplied lock pointer */
698 q
->queue_lock
= lock
;
701 * This also sets hw/phys segments, boundary and size
703 blk_queue_make_request(q
, blk_queue_bio
);
705 q
->sg_reserved_size
= INT_MAX
;
708 if (elevator_init(q
, NULL
))
711 blk_queue_congestion_threshold(q
);
713 /* all done, end the initial bypass */
714 blk_queue_bypass_end(q
);
717 EXPORT_SYMBOL(blk_init_allocated_queue
);
719 bool blk_get_queue(struct request_queue
*q
)
721 if (likely(!blk_queue_dead(q
))) {
728 EXPORT_SYMBOL(blk_get_queue
);
730 static inline void blk_free_request(struct request_list
*rl
, struct request
*rq
)
732 if (rq
->cmd_flags
& REQ_ELVPRIV
) {
733 elv_put_request(rl
->q
, rq
);
735 put_io_context(rq
->elv
.icq
->ioc
);
738 mempool_free(rq
, rl
->rq_pool
);
742 * ioc_batching returns true if the ioc is a valid batching request and
743 * should be given priority access to a request.
745 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
751 * Make sure the process is able to allocate at least 1 request
752 * even if the batch times out, otherwise we could theoretically
755 return ioc
->nr_batch_requests
== q
->nr_batching
||
756 (ioc
->nr_batch_requests
> 0
757 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
761 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
762 * will cause the process to be a "batcher" on all queues in the system. This
763 * is the behaviour we want though - once it gets a wakeup it should be given
766 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
768 if (!ioc
|| ioc_batching(q
, ioc
))
771 ioc
->nr_batch_requests
= q
->nr_batching
;
772 ioc
->last_waited
= jiffies
;
775 static void __freed_request(struct request_list
*rl
, int sync
)
777 struct request_queue
*q
= rl
->q
;
779 if (rl
->count
[sync
] < queue_congestion_off_threshold(q
))
780 blk_clear_queue_congested(q
, sync
);
782 if (rl
->count
[sync
] + 1 <= q
->nr_requests
) {
783 if (waitqueue_active(&rl
->wait
[sync
]))
784 wake_up(&rl
->wait
[sync
]);
786 blk_clear_rl_full(rl
, sync
);
791 * A request has just been released. Account for it, update the full and
792 * congestion status, wake up any waiters. Called under q->queue_lock.
794 static void freed_request(struct request_list
*rl
, unsigned int flags
)
796 struct request_queue
*q
= rl
->q
;
797 int sync
= rw_is_sync(flags
);
801 if (flags
& REQ_ELVPRIV
)
804 __freed_request(rl
, sync
);
806 if (unlikely(rl
->starved
[sync
^ 1]))
807 __freed_request(rl
, sync
^ 1);
811 * Determine if elevator data should be initialized when allocating the
812 * request associated with @bio.
814 static bool blk_rq_should_init_elevator(struct bio
*bio
)
820 * Flush requests do not use the elevator so skip initialization.
821 * This allows a request to share the flush and elevator data.
823 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
))
830 * rq_ioc - determine io_context for request allocation
831 * @bio: request being allocated is for this bio (can be %NULL)
833 * Determine io_context to use for request allocation for @bio. May return
834 * %NULL if %current->io_context doesn't exist.
836 static struct io_context
*rq_ioc(struct bio
*bio
)
838 #ifdef CONFIG_BLK_CGROUP
839 if (bio
&& bio
->bi_ioc
)
842 return current
->io_context
;
846 * __get_request - get a free request
847 * @rl: request list to allocate from
848 * @rw_flags: RW and SYNC flags
849 * @bio: bio to allocate request for (can be %NULL)
850 * @gfp_mask: allocation mask
852 * Get a free request from @q. This function may fail under memory
853 * pressure or if @q is dead.
855 * Must be callled with @q->queue_lock held and,
856 * Returns %NULL on failure, with @q->queue_lock held.
857 * Returns !%NULL on success, with @q->queue_lock *not held*.
859 static struct request
*__get_request(struct request_list
*rl
, int rw_flags
,
860 struct bio
*bio
, gfp_t gfp_mask
)
862 struct request_queue
*q
= rl
->q
;
864 struct elevator_type
*et
= q
->elevator
->type
;
865 struct io_context
*ioc
= rq_ioc(bio
);
866 struct io_cq
*icq
= NULL
;
867 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
870 if (unlikely(blk_queue_dead(q
)))
873 may_queue
= elv_may_queue(q
, rw_flags
);
874 if (may_queue
== ELV_MQUEUE_NO
)
877 if (rl
->count
[is_sync
]+1 >= queue_congestion_on_threshold(q
)) {
878 if (rl
->count
[is_sync
]+1 >= q
->nr_requests
) {
880 * The queue will fill after this allocation, so set
881 * it as full, and mark this process as "batching".
882 * This process will be allowed to complete a batch of
883 * requests, others will be blocked.
885 if (!blk_rl_full(rl
, is_sync
)) {
886 ioc_set_batching(q
, ioc
);
887 blk_set_rl_full(rl
, is_sync
);
889 if (may_queue
!= ELV_MQUEUE_MUST
890 && !ioc_batching(q
, ioc
)) {
892 * The queue is full and the allocating
893 * process is not a "batcher", and not
894 * exempted by the IO scheduler
900 blk_set_queue_congested(q
, is_sync
);
904 * Only allow batching queuers to allocate up to 50% over the defined
905 * limit of requests, otherwise we could have thousands of requests
906 * allocated with any setting of ->nr_requests
908 if (rl
->count
[is_sync
] >= (3 * q
->nr_requests
/ 2))
911 q
->nr_rqs
[is_sync
]++;
912 rl
->count
[is_sync
]++;
913 rl
->starved
[is_sync
] = 0;
916 * Decide whether the new request will be managed by elevator. If
917 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
918 * prevent the current elevator from being destroyed until the new
919 * request is freed. This guarantees icq's won't be destroyed and
920 * makes creating new ones safe.
922 * Also, lookup icq while holding queue_lock. If it doesn't exist,
923 * it will be created after releasing queue_lock.
925 if (blk_rq_should_init_elevator(bio
) && !blk_queue_bypass(q
)) {
926 rw_flags
|= REQ_ELVPRIV
;
928 if (et
->icq_cache
&& ioc
)
929 icq
= ioc_lookup_icq(ioc
, q
);
932 if (blk_queue_io_stat(q
))
933 rw_flags
|= REQ_IO_STAT
;
934 spin_unlock_irq(q
->queue_lock
);
936 /* allocate and init request */
937 rq
= mempool_alloc(rl
->rq_pool
, gfp_mask
);
942 rq
->cmd_flags
= rw_flags
| REQ_ALLOCED
;
945 if (rw_flags
& REQ_ELVPRIV
) {
946 if (unlikely(et
->icq_cache
&& !icq
)) {
948 icq
= ioc_create_icq(ioc
, q
, gfp_mask
);
954 if (unlikely(elv_set_request(q
, rq
, bio
, gfp_mask
)))
957 /* @rq->elv.icq holds io_context until @rq is freed */
959 get_io_context(icq
->ioc
);
963 * ioc may be NULL here, and ioc_batching will be false. That's
964 * OK, if the queue is under the request limit then requests need
965 * not count toward the nr_batch_requests limit. There will always
966 * be some limit enforced by BLK_BATCH_TIME.
968 if (ioc_batching(q
, ioc
))
969 ioc
->nr_batch_requests
--;
971 trace_block_getrq(q
, bio
, rw_flags
& 1);
976 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
977 * and may fail indefinitely under memory pressure and thus
978 * shouldn't stall IO. Treat this request as !elvpriv. This will
979 * disturb iosched and blkcg but weird is bettern than dead.
981 printk_ratelimited(KERN_WARNING
"%s: request aux data allocation failed, iosched may be disturbed\n",
982 dev_name(q
->backing_dev_info
.dev
));
984 rq
->cmd_flags
&= ~REQ_ELVPRIV
;
987 spin_lock_irq(q
->queue_lock
);
989 spin_unlock_irq(q
->queue_lock
);
994 * Allocation failed presumably due to memory. Undo anything we
995 * might have messed up.
997 * Allocating task should really be put onto the front of the wait
998 * queue, but this is pretty rare.
1000 spin_lock_irq(q
->queue_lock
);
1001 freed_request(rl
, rw_flags
);
1004 * in the very unlikely event that allocation failed and no
1005 * requests for this direction was pending, mark us starved so that
1006 * freeing of a request in the other direction will notice
1007 * us. another possible fix would be to split the rq mempool into
1011 if (unlikely(rl
->count
[is_sync
] == 0))
1012 rl
->starved
[is_sync
] = 1;
1017 * get_request - get a free request
1018 * @q: request_queue to allocate request from
1019 * @rw_flags: RW and SYNC flags
1020 * @bio: bio to allocate request for (can be %NULL)
1021 * @gfp_mask: allocation mask
1023 * Get a free request from @q. If %__GFP_WAIT is set in @gfp_mask, this
1024 * function keeps retrying under memory pressure and fails iff @q is dead.
1026 * Must be callled with @q->queue_lock held and,
1027 * Returns %NULL on failure, with @q->queue_lock held.
1028 * Returns !%NULL on success, with @q->queue_lock *not held*.
1030 static struct request
*get_request(struct request_queue
*q
, int rw_flags
,
1031 struct bio
*bio
, gfp_t gfp_mask
)
1033 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
1035 struct request_list
*rl
= &q
->rq
;
1038 rq
= __get_request(&q
->rq
, rw_flags
, bio
, gfp_mask
);
1042 if (!(gfp_mask
& __GFP_WAIT
) || unlikely(blk_queue_dead(q
)))
1045 /* wait on @rl and retry */
1046 prepare_to_wait_exclusive(&rl
->wait
[is_sync
], &wait
,
1047 TASK_UNINTERRUPTIBLE
);
1049 trace_block_sleeprq(q
, bio
, rw_flags
& 1);
1051 spin_unlock_irq(q
->queue_lock
);
1055 * After sleeping, we become a "batching" process and will be able
1056 * to allocate at least one request, and up to a big batch of them
1057 * for a small period time. See ioc_batching, ioc_set_batching
1059 ioc_set_batching(q
, current
->io_context
);
1061 spin_lock_irq(q
->queue_lock
);
1062 finish_wait(&rl
->wait
[is_sync
], &wait
);
1067 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
1071 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
1073 /* create ioc upfront */
1074 create_io_context(gfp_mask
, q
->node
);
1076 spin_lock_irq(q
->queue_lock
);
1077 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
1079 spin_unlock_irq(q
->queue_lock
);
1080 /* q->queue_lock is unlocked at this point */
1084 EXPORT_SYMBOL(blk_get_request
);
1087 * blk_make_request - given a bio, allocate a corresponding struct request.
1088 * @q: target request queue
1089 * @bio: The bio describing the memory mappings that will be submitted for IO.
1090 * It may be a chained-bio properly constructed by block/bio layer.
1091 * @gfp_mask: gfp flags to be used for memory allocation
1093 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1094 * type commands. Where the struct request needs to be farther initialized by
1095 * the caller. It is passed a &struct bio, which describes the memory info of
1098 * The caller of blk_make_request must make sure that bi_io_vec
1099 * are set to describe the memory buffers. That bio_data_dir() will return
1100 * the needed direction of the request. (And all bio's in the passed bio-chain
1101 * are properly set accordingly)
1103 * If called under none-sleepable conditions, mapped bio buffers must not
1104 * need bouncing, by calling the appropriate masked or flagged allocator,
1105 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1108 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1109 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1110 * anything but the first bio in the chain. Otherwise you risk waiting for IO
1111 * completion of a bio that hasn't been submitted yet, thus resulting in a
1112 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1113 * of bio_alloc(), as that avoids the mempool deadlock.
1114 * If possible a big IO should be split into smaller parts when allocation
1115 * fails. Partial allocation should not be an error, or you risk a live-lock.
1117 struct request
*blk_make_request(struct request_queue
*q
, struct bio
*bio
,
1120 struct request
*rq
= blk_get_request(q
, bio_data_dir(bio
), gfp_mask
);
1123 return ERR_PTR(-ENOMEM
);
1126 struct bio
*bounce_bio
= bio
;
1129 blk_queue_bounce(q
, &bounce_bio
);
1130 ret
= blk_rq_append_bio(q
, rq
, bounce_bio
);
1131 if (unlikely(ret
)) {
1132 blk_put_request(rq
);
1133 return ERR_PTR(ret
);
1139 EXPORT_SYMBOL(blk_make_request
);
1142 * blk_requeue_request - put a request back on queue
1143 * @q: request queue where request should be inserted
1144 * @rq: request to be inserted
1147 * Drivers often keep queueing requests until the hardware cannot accept
1148 * more, when that condition happens we need to put the request back
1149 * on the queue. Must be called with queue lock held.
1151 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
1153 blk_delete_timer(rq
);
1154 blk_clear_rq_complete(rq
);
1155 trace_block_rq_requeue(q
, rq
);
1157 if (blk_rq_tagged(rq
))
1158 blk_queue_end_tag(q
, rq
);
1160 BUG_ON(blk_queued_rq(rq
));
1162 elv_requeue_request(q
, rq
);
1164 EXPORT_SYMBOL(blk_requeue_request
);
1166 static void add_acct_request(struct request_queue
*q
, struct request
*rq
,
1169 drive_stat_acct(rq
, 1);
1170 __elv_add_request(q
, rq
, where
);
1173 static void part_round_stats_single(int cpu
, struct hd_struct
*part
,
1176 if (now
== part
->stamp
)
1179 if (part_in_flight(part
)) {
1180 __part_stat_add(cpu
, part
, time_in_queue
,
1181 part_in_flight(part
) * (now
- part
->stamp
));
1182 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
1188 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1189 * @cpu: cpu number for stats access
1190 * @part: target partition
1192 * The average IO queue length and utilisation statistics are maintained
1193 * by observing the current state of the queue length and the amount of
1194 * time it has been in this state for.
1196 * Normally, that accounting is done on IO completion, but that can result
1197 * in more than a second's worth of IO being accounted for within any one
1198 * second, leading to >100% utilisation. To deal with that, we call this
1199 * function to do a round-off before returning the results when reading
1200 * /proc/diskstats. This accounts immediately for all queue usage up to
1201 * the current jiffies and restarts the counters again.
1203 void part_round_stats(int cpu
, struct hd_struct
*part
)
1205 unsigned long now
= jiffies
;
1208 part_round_stats_single(cpu
, &part_to_disk(part
)->part0
, now
);
1209 part_round_stats_single(cpu
, part
, now
);
1211 EXPORT_SYMBOL_GPL(part_round_stats
);
1214 * queue lock must be held
1216 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1220 if (unlikely(--req
->ref_count
))
1223 elv_completed_request(q
, req
);
1225 /* this is a bio leak */
1226 WARN_ON(req
->bio
!= NULL
);
1229 * Request may not have originated from ll_rw_blk. if not,
1230 * it didn't come out of our reserved rq pools
1232 if (req
->cmd_flags
& REQ_ALLOCED
) {
1233 unsigned int flags
= req
->cmd_flags
;
1235 BUG_ON(!list_empty(&req
->queuelist
));
1236 BUG_ON(!hlist_unhashed(&req
->hash
));
1238 blk_free_request(&q
->rq
, req
);
1239 freed_request(&q
->rq
, flags
);
1242 EXPORT_SYMBOL_GPL(__blk_put_request
);
1244 void blk_put_request(struct request
*req
)
1246 unsigned long flags
;
1247 struct request_queue
*q
= req
->q
;
1249 spin_lock_irqsave(q
->queue_lock
, flags
);
1250 __blk_put_request(q
, req
);
1251 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1253 EXPORT_SYMBOL(blk_put_request
);
1256 * blk_add_request_payload - add a payload to a request
1257 * @rq: request to update
1258 * @page: page backing the payload
1259 * @len: length of the payload.
1261 * This allows to later add a payload to an already submitted request by
1262 * a block driver. The driver needs to take care of freeing the payload
1265 * Note that this is a quite horrible hack and nothing but handling of
1266 * discard requests should ever use it.
1268 void blk_add_request_payload(struct request
*rq
, struct page
*page
,
1271 struct bio
*bio
= rq
->bio
;
1273 bio
->bi_io_vec
->bv_page
= page
;
1274 bio
->bi_io_vec
->bv_offset
= 0;
1275 bio
->bi_io_vec
->bv_len
= len
;
1279 bio
->bi_phys_segments
= 1;
1281 rq
->__data_len
= rq
->resid_len
= len
;
1282 rq
->nr_phys_segments
= 1;
1283 rq
->buffer
= bio_data(bio
);
1285 EXPORT_SYMBOL_GPL(blk_add_request_payload
);
1287 static bool bio_attempt_back_merge(struct request_queue
*q
, struct request
*req
,
1290 const int ff
= bio
->bi_rw
& REQ_FAILFAST_MASK
;
1292 if (!ll_back_merge_fn(q
, req
, bio
))
1295 trace_block_bio_backmerge(q
, bio
);
1297 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1298 blk_rq_set_mixed_merge(req
);
1300 req
->biotail
->bi_next
= bio
;
1302 req
->__data_len
+= bio
->bi_size
;
1303 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1305 drive_stat_acct(req
, 0);
1309 static bool bio_attempt_front_merge(struct request_queue
*q
,
1310 struct request
*req
, struct bio
*bio
)
1312 const int ff
= bio
->bi_rw
& REQ_FAILFAST_MASK
;
1314 if (!ll_front_merge_fn(q
, req
, bio
))
1317 trace_block_bio_frontmerge(q
, bio
);
1319 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1320 blk_rq_set_mixed_merge(req
);
1322 bio
->bi_next
= req
->bio
;
1326 * may not be valid. if the low level driver said
1327 * it didn't need a bounce buffer then it better
1328 * not touch req->buffer either...
1330 req
->buffer
= bio_data(bio
);
1331 req
->__sector
= bio
->bi_sector
;
1332 req
->__data_len
+= bio
->bi_size
;
1333 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1335 drive_stat_acct(req
, 0);
1340 * attempt_plug_merge - try to merge with %current's plugged list
1341 * @q: request_queue new bio is being queued at
1342 * @bio: new bio being queued
1343 * @request_count: out parameter for number of traversed plugged requests
1345 * Determine whether @bio being queued on @q can be merged with a request
1346 * on %current's plugged list. Returns %true if merge was successful,
1349 * Plugging coalesces IOs from the same issuer for the same purpose without
1350 * going through @q->queue_lock. As such it's more of an issuing mechanism
1351 * than scheduling, and the request, while may have elvpriv data, is not
1352 * added on the elevator at this point. In addition, we don't have
1353 * reliable access to the elevator outside queue lock. Only check basic
1354 * merging parameters without querying the elevator.
1356 static bool attempt_plug_merge(struct request_queue
*q
, struct bio
*bio
,
1357 unsigned int *request_count
)
1359 struct blk_plug
*plug
;
1363 plug
= current
->plug
;
1368 list_for_each_entry_reverse(rq
, &plug
->list
, queuelist
) {
1374 if (rq
->q
!= q
|| !blk_rq_merge_ok(rq
, bio
))
1377 el_ret
= blk_try_merge(rq
, bio
);
1378 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1379 ret
= bio_attempt_back_merge(q
, rq
, bio
);
1382 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1383 ret
= bio_attempt_front_merge(q
, rq
, bio
);
1392 void init_request_from_bio(struct request
*req
, struct bio
*bio
)
1394 req
->cmd_type
= REQ_TYPE_FS
;
1396 req
->cmd_flags
|= bio
->bi_rw
& REQ_COMMON_MASK
;
1397 if (bio
->bi_rw
& REQ_RAHEAD
)
1398 req
->cmd_flags
|= REQ_FAILFAST_MASK
;
1401 req
->__sector
= bio
->bi_sector
;
1402 req
->ioprio
= bio_prio(bio
);
1403 blk_rq_bio_prep(req
->q
, req
, bio
);
1406 void blk_queue_bio(struct request_queue
*q
, struct bio
*bio
)
1408 const bool sync
= !!(bio
->bi_rw
& REQ_SYNC
);
1409 struct blk_plug
*plug
;
1410 int el_ret
, rw_flags
, where
= ELEVATOR_INSERT_SORT
;
1411 struct request
*req
;
1412 unsigned int request_count
= 0;
1415 * low level driver can indicate that it wants pages above a
1416 * certain limit bounced to low memory (ie for highmem, or even
1417 * ISA dma in theory)
1419 blk_queue_bounce(q
, &bio
);
1421 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) {
1422 spin_lock_irq(q
->queue_lock
);
1423 where
= ELEVATOR_INSERT_FLUSH
;
1428 * Check if we can merge with the plugged list before grabbing
1431 if (attempt_plug_merge(q
, bio
, &request_count
))
1434 spin_lock_irq(q
->queue_lock
);
1436 el_ret
= elv_merge(q
, &req
, bio
);
1437 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1438 if (bio_attempt_back_merge(q
, req
, bio
)) {
1439 elv_bio_merged(q
, req
, bio
);
1440 if (!attempt_back_merge(q
, req
))
1441 elv_merged_request(q
, req
, el_ret
);
1444 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1445 if (bio_attempt_front_merge(q
, req
, bio
)) {
1446 elv_bio_merged(q
, req
, bio
);
1447 if (!attempt_front_merge(q
, req
))
1448 elv_merged_request(q
, req
, el_ret
);
1455 * This sync check and mask will be re-done in init_request_from_bio(),
1456 * but we need to set it earlier to expose the sync flag to the
1457 * rq allocator and io schedulers.
1459 rw_flags
= bio_data_dir(bio
);
1461 rw_flags
|= REQ_SYNC
;
1464 * Grab a free request. This is might sleep but can not fail.
1465 * Returns with the queue unlocked.
1467 req
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
1468 if (unlikely(!req
)) {
1469 bio_endio(bio
, -ENODEV
); /* @q is dead */
1474 * After dropping the lock and possibly sleeping here, our request
1475 * may now be mergeable after it had proven unmergeable (above).
1476 * We don't worry about that case for efficiency. It won't happen
1477 * often, and the elevators are able to handle it.
1479 init_request_from_bio(req
, bio
);
1481 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
))
1482 req
->cpu
= raw_smp_processor_id();
1484 plug
= current
->plug
;
1487 * If this is the first request added after a plug, fire
1488 * of a plug trace. If others have been added before, check
1489 * if we have multiple devices in this plug. If so, make a
1490 * note to sort the list before dispatch.
1492 if (list_empty(&plug
->list
))
1493 trace_block_plug(q
);
1495 if (!plug
->should_sort
) {
1496 struct request
*__rq
;
1498 __rq
= list_entry_rq(plug
->list
.prev
);
1500 plug
->should_sort
= 1;
1502 if (request_count
>= BLK_MAX_REQUEST_COUNT
) {
1503 blk_flush_plug_list(plug
, false);
1504 trace_block_plug(q
);
1507 list_add_tail(&req
->queuelist
, &plug
->list
);
1508 drive_stat_acct(req
, 1);
1510 spin_lock_irq(q
->queue_lock
);
1511 add_acct_request(q
, req
, where
);
1514 spin_unlock_irq(q
->queue_lock
);
1517 EXPORT_SYMBOL_GPL(blk_queue_bio
); /* for device mapper only */
1520 * If bio->bi_dev is a partition, remap the location
1522 static inline void blk_partition_remap(struct bio
*bio
)
1524 struct block_device
*bdev
= bio
->bi_bdev
;
1526 if (bio_sectors(bio
) && bdev
!= bdev
->bd_contains
) {
1527 struct hd_struct
*p
= bdev
->bd_part
;
1529 bio
->bi_sector
+= p
->start_sect
;
1530 bio
->bi_bdev
= bdev
->bd_contains
;
1532 trace_block_bio_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
1534 bio
->bi_sector
- p
->start_sect
);
1538 static void handle_bad_sector(struct bio
*bio
)
1540 char b
[BDEVNAME_SIZE
];
1542 printk(KERN_INFO
"attempt to access beyond end of device\n");
1543 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
1544 bdevname(bio
->bi_bdev
, b
),
1546 (unsigned long long)bio
->bi_sector
+ bio_sectors(bio
),
1547 (long long)(i_size_read(bio
->bi_bdev
->bd_inode
) >> 9));
1549 set_bit(BIO_EOF
, &bio
->bi_flags
);
1552 #ifdef CONFIG_FAIL_MAKE_REQUEST
1554 static DECLARE_FAULT_ATTR(fail_make_request
);
1556 static int __init
setup_fail_make_request(char *str
)
1558 return setup_fault_attr(&fail_make_request
, str
);
1560 __setup("fail_make_request=", setup_fail_make_request
);
1562 static bool should_fail_request(struct hd_struct
*part
, unsigned int bytes
)
1564 return part
->make_it_fail
&& should_fail(&fail_make_request
, bytes
);
1567 static int __init
fail_make_request_debugfs(void)
1569 struct dentry
*dir
= fault_create_debugfs_attr("fail_make_request",
1570 NULL
, &fail_make_request
);
1572 return IS_ERR(dir
) ? PTR_ERR(dir
) : 0;
1575 late_initcall(fail_make_request_debugfs
);
1577 #else /* CONFIG_FAIL_MAKE_REQUEST */
1579 static inline bool should_fail_request(struct hd_struct
*part
,
1585 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1588 * Check whether this bio extends beyond the end of the device.
1590 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
1597 /* Test device or partition size, when known. */
1598 maxsector
= i_size_read(bio
->bi_bdev
->bd_inode
) >> 9;
1600 sector_t sector
= bio
->bi_sector
;
1602 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
1604 * This may well happen - the kernel calls bread()
1605 * without checking the size of the device, e.g., when
1606 * mounting a device.
1608 handle_bad_sector(bio
);
1616 static noinline_for_stack
bool
1617 generic_make_request_checks(struct bio
*bio
)
1619 struct request_queue
*q
;
1620 int nr_sectors
= bio_sectors(bio
);
1622 char b
[BDEVNAME_SIZE
];
1623 struct hd_struct
*part
;
1627 if (bio_check_eod(bio
, nr_sectors
))
1630 q
= bdev_get_queue(bio
->bi_bdev
);
1633 "generic_make_request: Trying to access "
1634 "nonexistent block-device %s (%Lu)\n",
1635 bdevname(bio
->bi_bdev
, b
),
1636 (long long) bio
->bi_sector
);
1640 if (unlikely(!(bio
->bi_rw
& REQ_DISCARD
) &&
1641 nr_sectors
> queue_max_hw_sectors(q
))) {
1642 printk(KERN_ERR
"bio too big device %s (%u > %u)\n",
1643 bdevname(bio
->bi_bdev
, b
),
1645 queue_max_hw_sectors(q
));
1649 part
= bio
->bi_bdev
->bd_part
;
1650 if (should_fail_request(part
, bio
->bi_size
) ||
1651 should_fail_request(&part_to_disk(part
)->part0
,
1656 * If this device has partitions, remap block n
1657 * of partition p to block n+start(p) of the disk.
1659 blk_partition_remap(bio
);
1661 if (bio_integrity_enabled(bio
) && bio_integrity_prep(bio
))
1664 if (bio_check_eod(bio
, nr_sectors
))
1668 * Filter flush bio's early so that make_request based
1669 * drivers without flush support don't have to worry
1672 if ((bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) && !q
->flush_flags
) {
1673 bio
->bi_rw
&= ~(REQ_FLUSH
| REQ_FUA
);
1680 if ((bio
->bi_rw
& REQ_DISCARD
) &&
1681 (!blk_queue_discard(q
) ||
1682 ((bio
->bi_rw
& REQ_SECURE
) &&
1683 !blk_queue_secdiscard(q
)))) {
1689 * Various block parts want %current->io_context and lazy ioc
1690 * allocation ends up trading a lot of pain for a small amount of
1691 * memory. Just allocate it upfront. This may fail and block
1692 * layer knows how to live with it.
1694 create_io_context(GFP_ATOMIC
, q
->node
);
1696 if (blk_throtl_bio(q
, bio
))
1697 return false; /* throttled, will be resubmitted later */
1699 trace_block_bio_queue(q
, bio
);
1703 bio_endio(bio
, err
);
1708 * generic_make_request - hand a buffer to its device driver for I/O
1709 * @bio: The bio describing the location in memory and on the device.
1711 * generic_make_request() is used to make I/O requests of block
1712 * devices. It is passed a &struct bio, which describes the I/O that needs
1715 * generic_make_request() does not return any status. The
1716 * success/failure status of the request, along with notification of
1717 * completion, is delivered asynchronously through the bio->bi_end_io
1718 * function described (one day) else where.
1720 * The caller of generic_make_request must make sure that bi_io_vec
1721 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1722 * set to describe the device address, and the
1723 * bi_end_io and optionally bi_private are set to describe how
1724 * completion notification should be signaled.
1726 * generic_make_request and the drivers it calls may use bi_next if this
1727 * bio happens to be merged with someone else, and may resubmit the bio to
1728 * a lower device by calling into generic_make_request recursively, which
1729 * means the bio should NOT be touched after the call to ->make_request_fn.
1731 void generic_make_request(struct bio
*bio
)
1733 struct bio_list bio_list_on_stack
;
1735 if (!generic_make_request_checks(bio
))
1739 * We only want one ->make_request_fn to be active at a time, else
1740 * stack usage with stacked devices could be a problem. So use
1741 * current->bio_list to keep a list of requests submited by a
1742 * make_request_fn function. current->bio_list is also used as a
1743 * flag to say if generic_make_request is currently active in this
1744 * task or not. If it is NULL, then no make_request is active. If
1745 * it is non-NULL, then a make_request is active, and new requests
1746 * should be added at the tail
1748 if (current
->bio_list
) {
1749 bio_list_add(current
->bio_list
, bio
);
1753 /* following loop may be a bit non-obvious, and so deserves some
1755 * Before entering the loop, bio->bi_next is NULL (as all callers
1756 * ensure that) so we have a list with a single bio.
1757 * We pretend that we have just taken it off a longer list, so
1758 * we assign bio_list to a pointer to the bio_list_on_stack,
1759 * thus initialising the bio_list of new bios to be
1760 * added. ->make_request() may indeed add some more bios
1761 * through a recursive call to generic_make_request. If it
1762 * did, we find a non-NULL value in bio_list and re-enter the loop
1763 * from the top. In this case we really did just take the bio
1764 * of the top of the list (no pretending) and so remove it from
1765 * bio_list, and call into ->make_request() again.
1767 BUG_ON(bio
->bi_next
);
1768 bio_list_init(&bio_list_on_stack
);
1769 current
->bio_list
= &bio_list_on_stack
;
1771 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
1773 q
->make_request_fn(q
, bio
);
1775 bio
= bio_list_pop(current
->bio_list
);
1777 current
->bio_list
= NULL
; /* deactivate */
1779 EXPORT_SYMBOL(generic_make_request
);
1782 * submit_bio - submit a bio to the block device layer for I/O
1783 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1784 * @bio: The &struct bio which describes the I/O
1786 * submit_bio() is very similar in purpose to generic_make_request(), and
1787 * uses that function to do most of the work. Both are fairly rough
1788 * interfaces; @bio must be presetup and ready for I/O.
1791 void submit_bio(int rw
, struct bio
*bio
)
1793 int count
= bio_sectors(bio
);
1798 * If it's a regular read/write or a barrier with data attached,
1799 * go through the normal accounting stuff before submission.
1801 if (bio_has_data(bio
) && !(rw
& REQ_DISCARD
)) {
1803 count_vm_events(PGPGOUT
, count
);
1805 task_io_account_read(bio
->bi_size
);
1806 count_vm_events(PGPGIN
, count
);
1809 if (unlikely(block_dump
)) {
1810 char b
[BDEVNAME_SIZE
];
1811 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s (%u sectors)\n",
1812 current
->comm
, task_pid_nr(current
),
1813 (rw
& WRITE
) ? "WRITE" : "READ",
1814 (unsigned long long)bio
->bi_sector
,
1815 bdevname(bio
->bi_bdev
, b
),
1820 generic_make_request(bio
);
1822 EXPORT_SYMBOL(submit_bio
);
1825 * blk_rq_check_limits - Helper function to check a request for the queue limit
1827 * @rq: the request being checked
1830 * @rq may have been made based on weaker limitations of upper-level queues
1831 * in request stacking drivers, and it may violate the limitation of @q.
1832 * Since the block layer and the underlying device driver trust @rq
1833 * after it is inserted to @q, it should be checked against @q before
1834 * the insertion using this generic function.
1836 * This function should also be useful for request stacking drivers
1837 * in some cases below, so export this function.
1838 * Request stacking drivers like request-based dm may change the queue
1839 * limits while requests are in the queue (e.g. dm's table swapping).
1840 * Such request stacking drivers should check those requests agaist
1841 * the new queue limits again when they dispatch those requests,
1842 * although such checkings are also done against the old queue limits
1843 * when submitting requests.
1845 int blk_rq_check_limits(struct request_queue
*q
, struct request
*rq
)
1847 if (rq
->cmd_flags
& REQ_DISCARD
)
1850 if (blk_rq_sectors(rq
) > queue_max_sectors(q
) ||
1851 blk_rq_bytes(rq
) > queue_max_hw_sectors(q
) << 9) {
1852 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
1857 * queue's settings related to segment counting like q->bounce_pfn
1858 * may differ from that of other stacking queues.
1859 * Recalculate it to check the request correctly on this queue's
1862 blk_recalc_rq_segments(rq
);
1863 if (rq
->nr_phys_segments
> queue_max_segments(q
)) {
1864 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
1870 EXPORT_SYMBOL_GPL(blk_rq_check_limits
);
1873 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1874 * @q: the queue to submit the request
1875 * @rq: the request being queued
1877 int blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
1879 unsigned long flags
;
1880 int where
= ELEVATOR_INSERT_BACK
;
1882 if (blk_rq_check_limits(q
, rq
))
1886 should_fail_request(&rq
->rq_disk
->part0
, blk_rq_bytes(rq
)))
1889 spin_lock_irqsave(q
->queue_lock
, flags
);
1890 if (unlikely(blk_queue_dead(q
))) {
1891 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1896 * Submitting request must be dequeued before calling this function
1897 * because it will be linked to another request_queue
1899 BUG_ON(blk_queued_rq(rq
));
1901 if (rq
->cmd_flags
& (REQ_FLUSH
|REQ_FUA
))
1902 where
= ELEVATOR_INSERT_FLUSH
;
1904 add_acct_request(q
, rq
, where
);
1905 if (where
== ELEVATOR_INSERT_FLUSH
)
1907 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1911 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
1914 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1915 * @rq: request to examine
1918 * A request could be merge of IOs which require different failure
1919 * handling. This function determines the number of bytes which
1920 * can be failed from the beginning of the request without
1921 * crossing into area which need to be retried further.
1924 * The number of bytes to fail.
1927 * queue_lock must be held.
1929 unsigned int blk_rq_err_bytes(const struct request
*rq
)
1931 unsigned int ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
1932 unsigned int bytes
= 0;
1935 if (!(rq
->cmd_flags
& REQ_MIXED_MERGE
))
1936 return blk_rq_bytes(rq
);
1939 * Currently the only 'mixing' which can happen is between
1940 * different fastfail types. We can safely fail portions
1941 * which have all the failfast bits that the first one has -
1942 * the ones which are at least as eager to fail as the first
1945 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
1946 if ((bio
->bi_rw
& ff
) != ff
)
1948 bytes
+= bio
->bi_size
;
1951 /* this could lead to infinite loop */
1952 BUG_ON(blk_rq_bytes(rq
) && !bytes
);
1955 EXPORT_SYMBOL_GPL(blk_rq_err_bytes
);
1957 static void blk_account_io_completion(struct request
*req
, unsigned int bytes
)
1959 if (blk_do_io_stat(req
)) {
1960 const int rw
= rq_data_dir(req
);
1961 struct hd_struct
*part
;
1964 cpu
= part_stat_lock();
1966 part_stat_add(cpu
, part
, sectors
[rw
], bytes
>> 9);
1971 static void blk_account_io_done(struct request
*req
)
1974 * Account IO completion. flush_rq isn't accounted as a
1975 * normal IO on queueing nor completion. Accounting the
1976 * containing request is enough.
1978 if (blk_do_io_stat(req
) && !(req
->cmd_flags
& REQ_FLUSH_SEQ
)) {
1979 unsigned long duration
= jiffies
- req
->start_time
;
1980 const int rw
= rq_data_dir(req
);
1981 struct hd_struct
*part
;
1984 cpu
= part_stat_lock();
1987 part_stat_inc(cpu
, part
, ios
[rw
]);
1988 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
1989 part_round_stats(cpu
, part
);
1990 part_dec_in_flight(part
, rw
);
1992 hd_struct_put(part
);
1998 * blk_peek_request - peek at the top of a request queue
1999 * @q: request queue to peek at
2002 * Return the request at the top of @q. The returned request
2003 * should be started using blk_start_request() before LLD starts
2007 * Pointer to the request at the top of @q if available. Null
2011 * queue_lock must be held.
2013 struct request
*blk_peek_request(struct request_queue
*q
)
2018 while ((rq
= __elv_next_request(q
)) != NULL
) {
2019 if (!(rq
->cmd_flags
& REQ_STARTED
)) {
2021 * This is the first time the device driver
2022 * sees this request (possibly after
2023 * requeueing). Notify IO scheduler.
2025 if (rq
->cmd_flags
& REQ_SORTED
)
2026 elv_activate_rq(q
, rq
);
2029 * just mark as started even if we don't start
2030 * it, a request that has been delayed should
2031 * not be passed by new incoming requests
2033 rq
->cmd_flags
|= REQ_STARTED
;
2034 trace_block_rq_issue(q
, rq
);
2037 if (!q
->boundary_rq
|| q
->boundary_rq
== rq
) {
2038 q
->end_sector
= rq_end_sector(rq
);
2039 q
->boundary_rq
= NULL
;
2042 if (rq
->cmd_flags
& REQ_DONTPREP
)
2045 if (q
->dma_drain_size
&& blk_rq_bytes(rq
)) {
2047 * make sure space for the drain appears we
2048 * know we can do this because max_hw_segments
2049 * has been adjusted to be one fewer than the
2052 rq
->nr_phys_segments
++;
2058 ret
= q
->prep_rq_fn(q
, rq
);
2059 if (ret
== BLKPREP_OK
) {
2061 } else if (ret
== BLKPREP_DEFER
) {
2063 * the request may have been (partially) prepped.
2064 * we need to keep this request in the front to
2065 * avoid resource deadlock. REQ_STARTED will
2066 * prevent other fs requests from passing this one.
2068 if (q
->dma_drain_size
&& blk_rq_bytes(rq
) &&
2069 !(rq
->cmd_flags
& REQ_DONTPREP
)) {
2071 * remove the space for the drain we added
2072 * so that we don't add it again
2074 --rq
->nr_phys_segments
;
2079 } else if (ret
== BLKPREP_KILL
) {
2080 rq
->cmd_flags
|= REQ_QUIET
;
2082 * Mark this request as started so we don't trigger
2083 * any debug logic in the end I/O path.
2085 blk_start_request(rq
);
2086 __blk_end_request_all(rq
, -EIO
);
2088 printk(KERN_ERR
"%s: bad return=%d\n", __func__
, ret
);
2095 EXPORT_SYMBOL(blk_peek_request
);
2097 void blk_dequeue_request(struct request
*rq
)
2099 struct request_queue
*q
= rq
->q
;
2101 BUG_ON(list_empty(&rq
->queuelist
));
2102 BUG_ON(ELV_ON_HASH(rq
));
2104 list_del_init(&rq
->queuelist
);
2107 * the time frame between a request being removed from the lists
2108 * and to it is freed is accounted as io that is in progress at
2111 if (blk_account_rq(rq
)) {
2112 q
->in_flight
[rq_is_sync(rq
)]++;
2113 set_io_start_time_ns(rq
);
2118 * blk_start_request - start request processing on the driver
2119 * @req: request to dequeue
2122 * Dequeue @req and start timeout timer on it. This hands off the
2123 * request to the driver.
2125 * Block internal functions which don't want to start timer should
2126 * call blk_dequeue_request().
2129 * queue_lock must be held.
2131 void blk_start_request(struct request
*req
)
2133 blk_dequeue_request(req
);
2136 * We are now handing the request to the hardware, initialize
2137 * resid_len to full count and add the timeout handler.
2139 req
->resid_len
= blk_rq_bytes(req
);
2140 if (unlikely(blk_bidi_rq(req
)))
2141 req
->next_rq
->resid_len
= blk_rq_bytes(req
->next_rq
);
2145 EXPORT_SYMBOL(blk_start_request
);
2148 * blk_fetch_request - fetch a request from a request queue
2149 * @q: request queue to fetch a request from
2152 * Return the request at the top of @q. The request is started on
2153 * return and LLD can start processing it immediately.
2156 * Pointer to the request at the top of @q if available. Null
2160 * queue_lock must be held.
2162 struct request
*blk_fetch_request(struct request_queue
*q
)
2166 rq
= blk_peek_request(q
);
2168 blk_start_request(rq
);
2171 EXPORT_SYMBOL(blk_fetch_request
);
2174 * blk_update_request - Special helper function for request stacking drivers
2175 * @req: the request being processed
2176 * @error: %0 for success, < %0 for error
2177 * @nr_bytes: number of bytes to complete @req
2180 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2181 * the request structure even if @req doesn't have leftover.
2182 * If @req has leftover, sets it up for the next range of segments.
2184 * This special helper function is only for request stacking drivers
2185 * (e.g. request-based dm) so that they can handle partial completion.
2186 * Actual device drivers should use blk_end_request instead.
2188 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2189 * %false return from this function.
2192 * %false - this request doesn't have any more data
2193 * %true - this request has more data
2195 bool blk_update_request(struct request
*req
, int error
, unsigned int nr_bytes
)
2197 int total_bytes
, bio_nbytes
, next_idx
= 0;
2203 trace_block_rq_complete(req
->q
, req
);
2206 * For fs requests, rq is just carrier of independent bio's
2207 * and each partial completion should be handled separately.
2208 * Reset per-request error on each partial completion.
2210 * TODO: tj: This is too subtle. It would be better to let
2211 * low level drivers do what they see fit.
2213 if (req
->cmd_type
== REQ_TYPE_FS
)
2216 if (error
&& req
->cmd_type
== REQ_TYPE_FS
&&
2217 !(req
->cmd_flags
& REQ_QUIET
)) {
2222 error_type
= "recoverable transport";
2225 error_type
= "critical target";
2228 error_type
= "critical nexus";
2235 printk(KERN_ERR
"end_request: %s error, dev %s, sector %llu\n",
2236 error_type
, req
->rq_disk
? req
->rq_disk
->disk_name
: "?",
2237 (unsigned long long)blk_rq_pos(req
));
2240 blk_account_io_completion(req
, nr_bytes
);
2242 total_bytes
= bio_nbytes
= 0;
2243 while ((bio
= req
->bio
) != NULL
) {
2246 if (nr_bytes
>= bio
->bi_size
) {
2247 req
->bio
= bio
->bi_next
;
2248 nbytes
= bio
->bi_size
;
2249 req_bio_endio(req
, bio
, nbytes
, error
);
2253 int idx
= bio
->bi_idx
+ next_idx
;
2255 if (unlikely(idx
>= bio
->bi_vcnt
)) {
2256 blk_dump_rq_flags(req
, "__end_that");
2257 printk(KERN_ERR
"%s: bio idx %d >= vcnt %d\n",
2258 __func__
, idx
, bio
->bi_vcnt
);
2262 nbytes
= bio_iovec_idx(bio
, idx
)->bv_len
;
2263 BIO_BUG_ON(nbytes
> bio
->bi_size
);
2266 * not a complete bvec done
2268 if (unlikely(nbytes
> nr_bytes
)) {
2269 bio_nbytes
+= nr_bytes
;
2270 total_bytes
+= nr_bytes
;
2275 * advance to the next vector
2278 bio_nbytes
+= nbytes
;
2281 total_bytes
+= nbytes
;
2287 * end more in this run, or just return 'not-done'
2289 if (unlikely(nr_bytes
<= 0))
2299 * Reset counters so that the request stacking driver
2300 * can find how many bytes remain in the request
2303 req
->__data_len
= 0;
2308 * if the request wasn't completed, update state
2311 req_bio_endio(req
, bio
, bio_nbytes
, error
);
2312 bio
->bi_idx
+= next_idx
;
2313 bio_iovec(bio
)->bv_offset
+= nr_bytes
;
2314 bio_iovec(bio
)->bv_len
-= nr_bytes
;
2317 req
->__data_len
-= total_bytes
;
2318 req
->buffer
= bio_data(req
->bio
);
2320 /* update sector only for requests with clear definition of sector */
2321 if (req
->cmd_type
== REQ_TYPE_FS
|| (req
->cmd_flags
& REQ_DISCARD
))
2322 req
->__sector
+= total_bytes
>> 9;
2324 /* mixed attributes always follow the first bio */
2325 if (req
->cmd_flags
& REQ_MIXED_MERGE
) {
2326 req
->cmd_flags
&= ~REQ_FAILFAST_MASK
;
2327 req
->cmd_flags
|= req
->bio
->bi_rw
& REQ_FAILFAST_MASK
;
2331 * If total number of sectors is less than the first segment
2332 * size, something has gone terribly wrong.
2334 if (blk_rq_bytes(req
) < blk_rq_cur_bytes(req
)) {
2335 blk_dump_rq_flags(req
, "request botched");
2336 req
->__data_len
= blk_rq_cur_bytes(req
);
2339 /* recalculate the number of segments */
2340 blk_recalc_rq_segments(req
);
2344 EXPORT_SYMBOL_GPL(blk_update_request
);
2346 static bool blk_update_bidi_request(struct request
*rq
, int error
,
2347 unsigned int nr_bytes
,
2348 unsigned int bidi_bytes
)
2350 if (blk_update_request(rq
, error
, nr_bytes
))
2353 /* Bidi request must be completed as a whole */
2354 if (unlikely(blk_bidi_rq(rq
)) &&
2355 blk_update_request(rq
->next_rq
, error
, bidi_bytes
))
2358 if (blk_queue_add_random(rq
->q
))
2359 add_disk_randomness(rq
->rq_disk
);
2365 * blk_unprep_request - unprepare a request
2368 * This function makes a request ready for complete resubmission (or
2369 * completion). It happens only after all error handling is complete,
2370 * so represents the appropriate moment to deallocate any resources
2371 * that were allocated to the request in the prep_rq_fn. The queue
2372 * lock is held when calling this.
2374 void blk_unprep_request(struct request
*req
)
2376 struct request_queue
*q
= req
->q
;
2378 req
->cmd_flags
&= ~REQ_DONTPREP
;
2379 if (q
->unprep_rq_fn
)
2380 q
->unprep_rq_fn(q
, req
);
2382 EXPORT_SYMBOL_GPL(blk_unprep_request
);
2385 * queue lock must be held
2387 static void blk_finish_request(struct request
*req
, int error
)
2389 if (blk_rq_tagged(req
))
2390 blk_queue_end_tag(req
->q
, req
);
2392 BUG_ON(blk_queued_rq(req
));
2394 if (unlikely(laptop_mode
) && req
->cmd_type
== REQ_TYPE_FS
)
2395 laptop_io_completion(&req
->q
->backing_dev_info
);
2397 blk_delete_timer(req
);
2399 if (req
->cmd_flags
& REQ_DONTPREP
)
2400 blk_unprep_request(req
);
2403 blk_account_io_done(req
);
2406 req
->end_io(req
, error
);
2408 if (blk_bidi_rq(req
))
2409 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
2411 __blk_put_request(req
->q
, req
);
2416 * blk_end_bidi_request - Complete a bidi request
2417 * @rq: the request to complete
2418 * @error: %0 for success, < %0 for error
2419 * @nr_bytes: number of bytes to complete @rq
2420 * @bidi_bytes: number of bytes to complete @rq->next_rq
2423 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2424 * Drivers that supports bidi can safely call this member for any
2425 * type of request, bidi or uni. In the later case @bidi_bytes is
2429 * %false - we are done with this request
2430 * %true - still buffers pending for this request
2432 static bool blk_end_bidi_request(struct request
*rq
, int error
,
2433 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2435 struct request_queue
*q
= rq
->q
;
2436 unsigned long flags
;
2438 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2441 spin_lock_irqsave(q
->queue_lock
, flags
);
2442 blk_finish_request(rq
, error
);
2443 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2449 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2450 * @rq: the request to complete
2451 * @error: %0 for success, < %0 for error
2452 * @nr_bytes: number of bytes to complete @rq
2453 * @bidi_bytes: number of bytes to complete @rq->next_rq
2456 * Identical to blk_end_bidi_request() except that queue lock is
2457 * assumed to be locked on entry and remains so on return.
2460 * %false - we are done with this request
2461 * %true - still buffers pending for this request
2463 bool __blk_end_bidi_request(struct request
*rq
, int error
,
2464 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2466 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2469 blk_finish_request(rq
, error
);
2475 * blk_end_request - Helper function for drivers to complete the request.
2476 * @rq: the request being processed
2477 * @error: %0 for success, < %0 for error
2478 * @nr_bytes: number of bytes to complete
2481 * Ends I/O on a number of bytes attached to @rq.
2482 * If @rq has leftover, sets it up for the next range of segments.
2485 * %false - we are done with this request
2486 * %true - still buffers pending for this request
2488 bool blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2490 return blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2492 EXPORT_SYMBOL(blk_end_request
);
2495 * blk_end_request_all - Helper function for drives to finish the request.
2496 * @rq: the request to finish
2497 * @error: %0 for success, < %0 for error
2500 * Completely finish @rq.
2502 void blk_end_request_all(struct request
*rq
, int error
)
2505 unsigned int bidi_bytes
= 0;
2507 if (unlikely(blk_bidi_rq(rq
)))
2508 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2510 pending
= blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2513 EXPORT_SYMBOL(blk_end_request_all
);
2516 * blk_end_request_cur - Helper function to finish the current request chunk.
2517 * @rq: the request to finish the current chunk for
2518 * @error: %0 for success, < %0 for error
2521 * Complete the current consecutively mapped chunk from @rq.
2524 * %false - we are done with this request
2525 * %true - still buffers pending for this request
2527 bool blk_end_request_cur(struct request
*rq
, int error
)
2529 return blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2531 EXPORT_SYMBOL(blk_end_request_cur
);
2534 * blk_end_request_err - Finish a request till the next failure boundary.
2535 * @rq: the request to finish till the next failure boundary for
2536 * @error: must be negative errno
2539 * Complete @rq till the next failure boundary.
2542 * %false - we are done with this request
2543 * %true - still buffers pending for this request
2545 bool blk_end_request_err(struct request
*rq
, int error
)
2547 WARN_ON(error
>= 0);
2548 return blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2550 EXPORT_SYMBOL_GPL(blk_end_request_err
);
2553 * __blk_end_request - Helper function for drivers to complete the request.
2554 * @rq: the request being processed
2555 * @error: %0 for success, < %0 for error
2556 * @nr_bytes: number of bytes to complete
2559 * Must be called with queue lock held unlike blk_end_request().
2562 * %false - we are done with this request
2563 * %true - still buffers pending for this request
2565 bool __blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2567 return __blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2569 EXPORT_SYMBOL(__blk_end_request
);
2572 * __blk_end_request_all - Helper function for drives to finish the request.
2573 * @rq: the request to finish
2574 * @error: %0 for success, < %0 for error
2577 * Completely finish @rq. Must be called with queue lock held.
2579 void __blk_end_request_all(struct request
*rq
, int error
)
2582 unsigned int bidi_bytes
= 0;
2584 if (unlikely(blk_bidi_rq(rq
)))
2585 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2587 pending
= __blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2590 EXPORT_SYMBOL(__blk_end_request_all
);
2593 * __blk_end_request_cur - Helper function to finish the current request chunk.
2594 * @rq: the request to finish the current chunk for
2595 * @error: %0 for success, < %0 for error
2598 * Complete the current consecutively mapped chunk from @rq. Must
2599 * be called with queue lock held.
2602 * %false - we are done with this request
2603 * %true - still buffers pending for this request
2605 bool __blk_end_request_cur(struct request
*rq
, int error
)
2607 return __blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2609 EXPORT_SYMBOL(__blk_end_request_cur
);
2612 * __blk_end_request_err - Finish a request till the next failure boundary.
2613 * @rq: the request to finish till the next failure boundary for
2614 * @error: must be negative errno
2617 * Complete @rq till the next failure boundary. Must be called
2618 * with queue lock held.
2621 * %false - we are done with this request
2622 * %true - still buffers pending for this request
2624 bool __blk_end_request_err(struct request
*rq
, int error
)
2626 WARN_ON(error
>= 0);
2627 return __blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2629 EXPORT_SYMBOL_GPL(__blk_end_request_err
);
2631 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
2634 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2635 rq
->cmd_flags
|= bio
->bi_rw
& REQ_WRITE
;
2637 if (bio_has_data(bio
)) {
2638 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2639 rq
->buffer
= bio_data(bio
);
2641 rq
->__data_len
= bio
->bi_size
;
2642 rq
->bio
= rq
->biotail
= bio
;
2645 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2648 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2650 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2651 * @rq: the request to be flushed
2654 * Flush all pages in @rq.
2656 void rq_flush_dcache_pages(struct request
*rq
)
2658 struct req_iterator iter
;
2659 struct bio_vec
*bvec
;
2661 rq_for_each_segment(bvec
, rq
, iter
)
2662 flush_dcache_page(bvec
->bv_page
);
2664 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages
);
2668 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2669 * @q : the queue of the device being checked
2672 * Check if underlying low-level drivers of a device are busy.
2673 * If the drivers want to export their busy state, they must set own
2674 * exporting function using blk_queue_lld_busy() first.
2676 * Basically, this function is used only by request stacking drivers
2677 * to stop dispatching requests to underlying devices when underlying
2678 * devices are busy. This behavior helps more I/O merging on the queue
2679 * of the request stacking driver and prevents I/O throughput regression
2680 * on burst I/O load.
2683 * 0 - Not busy (The request stacking driver should dispatch request)
2684 * 1 - Busy (The request stacking driver should stop dispatching request)
2686 int blk_lld_busy(struct request_queue
*q
)
2689 return q
->lld_busy_fn(q
);
2693 EXPORT_SYMBOL_GPL(blk_lld_busy
);
2696 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2697 * @rq: the clone request to be cleaned up
2700 * Free all bios in @rq for a cloned request.
2702 void blk_rq_unprep_clone(struct request
*rq
)
2706 while ((bio
= rq
->bio
) != NULL
) {
2707 rq
->bio
= bio
->bi_next
;
2712 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone
);
2715 * Copy attributes of the original request to the clone request.
2716 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2718 static void __blk_rq_prep_clone(struct request
*dst
, struct request
*src
)
2720 dst
->cpu
= src
->cpu
;
2721 dst
->cmd_flags
= (src
->cmd_flags
& REQ_CLONE_MASK
) | REQ_NOMERGE
;
2722 dst
->cmd_type
= src
->cmd_type
;
2723 dst
->__sector
= blk_rq_pos(src
);
2724 dst
->__data_len
= blk_rq_bytes(src
);
2725 dst
->nr_phys_segments
= src
->nr_phys_segments
;
2726 dst
->ioprio
= src
->ioprio
;
2727 dst
->extra_len
= src
->extra_len
;
2731 * blk_rq_prep_clone - Helper function to setup clone request
2732 * @rq: the request to be setup
2733 * @rq_src: original request to be cloned
2734 * @bs: bio_set that bios for clone are allocated from
2735 * @gfp_mask: memory allocation mask for bio
2736 * @bio_ctr: setup function to be called for each clone bio.
2737 * Returns %0 for success, non %0 for failure.
2738 * @data: private data to be passed to @bio_ctr
2741 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2742 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2743 * are not copied, and copying such parts is the caller's responsibility.
2744 * Also, pages which the original bios are pointing to are not copied
2745 * and the cloned bios just point same pages.
2746 * So cloned bios must be completed before original bios, which means
2747 * the caller must complete @rq before @rq_src.
2749 int blk_rq_prep_clone(struct request
*rq
, struct request
*rq_src
,
2750 struct bio_set
*bs
, gfp_t gfp_mask
,
2751 int (*bio_ctr
)(struct bio
*, struct bio
*, void *),
2754 struct bio
*bio
, *bio_src
;
2759 blk_rq_init(NULL
, rq
);
2761 __rq_for_each_bio(bio_src
, rq_src
) {
2762 bio
= bio_alloc_bioset(gfp_mask
, bio_src
->bi_max_vecs
, bs
);
2766 __bio_clone(bio
, bio_src
);
2768 if (bio_integrity(bio_src
) &&
2769 bio_integrity_clone(bio
, bio_src
, gfp_mask
, bs
))
2772 if (bio_ctr
&& bio_ctr(bio
, bio_src
, data
))
2776 rq
->biotail
->bi_next
= bio
;
2779 rq
->bio
= rq
->biotail
= bio
;
2782 __blk_rq_prep_clone(rq
, rq_src
);
2789 blk_rq_unprep_clone(rq
);
2793 EXPORT_SYMBOL_GPL(blk_rq_prep_clone
);
2795 int kblockd_schedule_work(struct request_queue
*q
, struct work_struct
*work
)
2797 return queue_work(kblockd_workqueue
, work
);
2799 EXPORT_SYMBOL(kblockd_schedule_work
);
2801 int kblockd_schedule_delayed_work(struct request_queue
*q
,
2802 struct delayed_work
*dwork
, unsigned long delay
)
2804 return queue_delayed_work(kblockd_workqueue
, dwork
, delay
);
2806 EXPORT_SYMBOL(kblockd_schedule_delayed_work
);
2808 #define PLUG_MAGIC 0x91827364
2811 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2812 * @plug: The &struct blk_plug that needs to be initialized
2815 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2816 * pending I/O should the task end up blocking between blk_start_plug() and
2817 * blk_finish_plug(). This is important from a performance perspective, but
2818 * also ensures that we don't deadlock. For instance, if the task is blocking
2819 * for a memory allocation, memory reclaim could end up wanting to free a
2820 * page belonging to that request that is currently residing in our private
2821 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2822 * this kind of deadlock.
2824 void blk_start_plug(struct blk_plug
*plug
)
2826 struct task_struct
*tsk
= current
;
2828 plug
->magic
= PLUG_MAGIC
;
2829 INIT_LIST_HEAD(&plug
->list
);
2830 INIT_LIST_HEAD(&plug
->cb_list
);
2831 plug
->should_sort
= 0;
2834 * If this is a nested plug, don't actually assign it. It will be
2835 * flushed on its own.
2839 * Store ordering should not be needed here, since a potential
2840 * preempt will imply a full memory barrier
2845 EXPORT_SYMBOL(blk_start_plug
);
2847 static int plug_rq_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
2849 struct request
*rqa
= container_of(a
, struct request
, queuelist
);
2850 struct request
*rqb
= container_of(b
, struct request
, queuelist
);
2852 return !(rqa
->q
<= rqb
->q
);
2856 * If 'from_schedule' is true, then postpone the dispatch of requests
2857 * until a safe kblockd context. We due this to avoid accidental big
2858 * additional stack usage in driver dispatch, in places where the originally
2859 * plugger did not intend it.
2861 static void queue_unplugged(struct request_queue
*q
, unsigned int depth
,
2863 __releases(q
->queue_lock
)
2865 trace_block_unplug(q
, depth
, !from_schedule
);
2868 * Don't mess with dead queue.
2870 if (unlikely(blk_queue_dead(q
))) {
2871 spin_unlock(q
->queue_lock
);
2876 * If we are punting this to kblockd, then we can safely drop
2877 * the queue_lock before waking kblockd (which needs to take
2880 if (from_schedule
) {
2881 spin_unlock(q
->queue_lock
);
2882 blk_run_queue_async(q
);
2885 spin_unlock(q
->queue_lock
);
2890 static void flush_plug_callbacks(struct blk_plug
*plug
)
2892 LIST_HEAD(callbacks
);
2894 if (list_empty(&plug
->cb_list
))
2897 list_splice_init(&plug
->cb_list
, &callbacks
);
2899 while (!list_empty(&callbacks
)) {
2900 struct blk_plug_cb
*cb
= list_first_entry(&callbacks
,
2903 list_del(&cb
->list
);
2908 void blk_flush_plug_list(struct blk_plug
*plug
, bool from_schedule
)
2910 struct request_queue
*q
;
2911 unsigned long flags
;
2916 BUG_ON(plug
->magic
!= PLUG_MAGIC
);
2918 flush_plug_callbacks(plug
);
2919 if (list_empty(&plug
->list
))
2922 list_splice_init(&plug
->list
, &list
);
2924 if (plug
->should_sort
) {
2925 list_sort(NULL
, &list
, plug_rq_cmp
);
2926 plug
->should_sort
= 0;
2933 * Save and disable interrupts here, to avoid doing it for every
2934 * queue lock we have to take.
2936 local_irq_save(flags
);
2937 while (!list_empty(&list
)) {
2938 rq
= list_entry_rq(list
.next
);
2939 list_del_init(&rq
->queuelist
);
2943 * This drops the queue lock
2946 queue_unplugged(q
, depth
, from_schedule
);
2949 spin_lock(q
->queue_lock
);
2953 * Short-circuit if @q is dead
2955 if (unlikely(blk_queue_dead(q
))) {
2956 __blk_end_request_all(rq
, -ENODEV
);
2961 * rq is already accounted, so use raw insert
2963 if (rq
->cmd_flags
& (REQ_FLUSH
| REQ_FUA
))
2964 __elv_add_request(q
, rq
, ELEVATOR_INSERT_FLUSH
);
2966 __elv_add_request(q
, rq
, ELEVATOR_INSERT_SORT_MERGE
);
2972 * This drops the queue lock
2975 queue_unplugged(q
, depth
, from_schedule
);
2977 local_irq_restore(flags
);
2980 void blk_finish_plug(struct blk_plug
*plug
)
2982 blk_flush_plug_list(plug
, false);
2984 if (plug
== current
->plug
)
2985 current
->plug
= NULL
;
2987 EXPORT_SYMBOL(blk_finish_plug
);
2989 int __init
blk_dev_init(void)
2991 BUILD_BUG_ON(__REQ_NR_BITS
> 8 *
2992 sizeof(((struct request
*)0)->cmd_flags
));
2994 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
2995 kblockd_workqueue
= alloc_workqueue("kblockd",
2996 WQ_MEM_RECLAIM
| WQ_HIGHPRI
, 0);
2997 if (!kblockd_workqueue
)
2998 panic("Failed to create kblockd\n");
3000 request_cachep
= kmem_cache_create("blkdev_requests",
3001 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
3003 blk_requestq_cachep
= kmem_cache_create("blkdev_queue",
3004 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);