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>
32 #define CREATE_TRACE_POINTS
33 #include <trace/events/block.h>
37 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap
);
38 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap
);
39 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete
);
41 static int __make_request(struct request_queue
*q
, struct bio
*bio
);
44 * For the allocated request tables
46 static struct kmem_cache
*request_cachep
;
49 * For queue allocation
51 struct kmem_cache
*blk_requestq_cachep
;
54 * Controlling structure to kblockd
56 static struct workqueue_struct
*kblockd_workqueue
;
58 static void drive_stat_acct(struct request
*rq
, int new_io
)
60 struct hd_struct
*part
;
61 int rw
= rq_data_dir(rq
);
64 if (!blk_do_io_stat(rq
))
67 cpu
= part_stat_lock();
71 part_stat_inc(cpu
, part
, merges
[rw
]);
73 part
= disk_map_sector_rcu(rq
->rq_disk
, blk_rq_pos(rq
));
74 if (!hd_struct_try_get(part
)) {
76 * The partition is already being removed,
77 * the request will be accounted on the disk only
79 * We take a reference on disk->part0 although that
80 * partition will never be deleted, so we can treat
81 * it as any other partition.
83 part
= &rq
->rq_disk
->part0
;
86 part_round_stats(cpu
, part
);
87 part_inc_in_flight(part
, rw
);
94 void blk_queue_congestion_threshold(struct request_queue
*q
)
98 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) + 1;
99 if (nr
> q
->nr_requests
)
101 q
->nr_congestion_on
= nr
;
103 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) - (q
->nr_requests
/ 16) - 1;
106 q
->nr_congestion_off
= nr
;
110 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
113 * Locates the passed device's request queue and returns the address of its
116 * Will return NULL if the request queue cannot be located.
118 struct backing_dev_info
*blk_get_backing_dev_info(struct block_device
*bdev
)
120 struct backing_dev_info
*ret
= NULL
;
121 struct request_queue
*q
= bdev_get_queue(bdev
);
124 ret
= &q
->backing_dev_info
;
127 EXPORT_SYMBOL(blk_get_backing_dev_info
);
129 void blk_rq_init(struct request_queue
*q
, struct request
*rq
)
131 memset(rq
, 0, sizeof(*rq
));
133 INIT_LIST_HEAD(&rq
->queuelist
);
134 INIT_LIST_HEAD(&rq
->timeout_list
);
137 rq
->__sector
= (sector_t
) -1;
138 INIT_HLIST_NODE(&rq
->hash
);
139 RB_CLEAR_NODE(&rq
->rb_node
);
141 rq
->cmd_len
= BLK_MAX_CDB
;
144 rq
->start_time
= jiffies
;
145 set_start_time_ns(rq
);
148 EXPORT_SYMBOL(blk_rq_init
);
150 static void req_bio_endio(struct request
*rq
, struct bio
*bio
,
151 unsigned int nbytes
, int error
)
154 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
155 else if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
158 if (unlikely(nbytes
> bio
->bi_size
)) {
159 printk(KERN_ERR
"%s: want %u bytes done, %u left\n",
160 __func__
, nbytes
, bio
->bi_size
);
161 nbytes
= bio
->bi_size
;
164 if (unlikely(rq
->cmd_flags
& REQ_QUIET
))
165 set_bit(BIO_QUIET
, &bio
->bi_flags
);
167 bio
->bi_size
-= nbytes
;
168 bio
->bi_sector
+= (nbytes
>> 9);
170 if (bio_integrity(bio
))
171 bio_integrity_advance(bio
, nbytes
);
173 /* don't actually finish bio if it's part of flush sequence */
174 if (bio
->bi_size
== 0 && !(rq
->cmd_flags
& REQ_FLUSH_SEQ
))
175 bio_endio(bio
, error
);
178 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
182 printk(KERN_INFO
"%s: dev %s: type=%x, flags=%x\n", msg
,
183 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->cmd_type
,
186 printk(KERN_INFO
" sector %llu, nr/cnr %u/%u\n",
187 (unsigned long long)blk_rq_pos(rq
),
188 blk_rq_sectors(rq
), blk_rq_cur_sectors(rq
));
189 printk(KERN_INFO
" bio %p, biotail %p, buffer %p, len %u\n",
190 rq
->bio
, rq
->biotail
, rq
->buffer
, blk_rq_bytes(rq
));
192 if (rq
->cmd_type
== REQ_TYPE_BLOCK_PC
) {
193 printk(KERN_INFO
" cdb: ");
194 for (bit
= 0; bit
< BLK_MAX_CDB
; bit
++)
195 printk("%02x ", rq
->cmd
[bit
]);
199 EXPORT_SYMBOL(blk_dump_rq_flags
);
201 static void blk_delay_work(struct work_struct
*work
)
203 struct request_queue
*q
;
205 q
= container_of(work
, struct request_queue
, delay_work
.work
);
206 spin_lock_irq(q
->queue_lock
);
208 spin_unlock_irq(q
->queue_lock
);
212 * blk_delay_queue - restart queueing after defined interval
213 * @q: The &struct request_queue in question
214 * @msecs: Delay in msecs
217 * Sometimes queueing needs to be postponed for a little while, to allow
218 * resources to come back. This function will make sure that queueing is
219 * restarted around the specified time.
221 void blk_delay_queue(struct request_queue
*q
, unsigned long msecs
)
223 queue_delayed_work(kblockd_workqueue
, &q
->delay_work
,
224 msecs_to_jiffies(msecs
));
226 EXPORT_SYMBOL(blk_delay_queue
);
229 * blk_start_queue - restart a previously stopped queue
230 * @q: The &struct request_queue in question
233 * blk_start_queue() will clear the stop flag on the queue, and call
234 * the request_fn for the queue if it was in a stopped state when
235 * entered. Also see blk_stop_queue(). Queue lock must be held.
237 void blk_start_queue(struct request_queue
*q
)
239 WARN_ON(!irqs_disabled());
241 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
244 EXPORT_SYMBOL(blk_start_queue
);
247 * blk_stop_queue - stop a queue
248 * @q: The &struct request_queue in question
251 * The Linux block layer assumes that a block driver will consume all
252 * entries on the request queue when the request_fn strategy is called.
253 * Often this will not happen, because of hardware limitations (queue
254 * depth settings). If a device driver gets a 'queue full' response,
255 * or if it simply chooses not to queue more I/O at one point, it can
256 * call this function to prevent the request_fn from being called until
257 * the driver has signalled it's ready to go again. This happens by calling
258 * blk_start_queue() to restart queue operations. Queue lock must be held.
260 void blk_stop_queue(struct request_queue
*q
)
262 __cancel_delayed_work(&q
->delay_work
);
263 queue_flag_set(QUEUE_FLAG_STOPPED
, q
);
265 EXPORT_SYMBOL(blk_stop_queue
);
268 * blk_sync_queue - cancel any pending callbacks on a queue
272 * The block layer may perform asynchronous callback activity
273 * on a queue, such as calling the unplug function after a timeout.
274 * A block device may call blk_sync_queue to ensure that any
275 * such activity is cancelled, thus allowing it to release resources
276 * that the callbacks might use. The caller must already have made sure
277 * that its ->make_request_fn will not re-add plugging prior to calling
280 * This function does not cancel any asynchronous activity arising
281 * out of elevator or throttling code. That would require elevaotor_exit()
282 * and blk_throtl_exit() to be called with queue lock initialized.
285 void blk_sync_queue(struct request_queue
*q
)
287 del_timer_sync(&q
->timeout
);
288 cancel_delayed_work_sync(&q
->delay_work
);
290 EXPORT_SYMBOL(blk_sync_queue
);
293 * __blk_run_queue - run a single device queue
294 * @q: The queue to run
297 * See @blk_run_queue. This variant must be called with the queue lock
298 * held and interrupts disabled.
300 void __blk_run_queue(struct request_queue
*q
)
302 if (unlikely(blk_queue_stopped(q
)))
307 EXPORT_SYMBOL(__blk_run_queue
);
310 * blk_run_queue_async - run a single device queue in workqueue context
311 * @q: The queue to run
314 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
317 void blk_run_queue_async(struct request_queue
*q
)
319 if (likely(!blk_queue_stopped(q
))) {
320 __cancel_delayed_work(&q
->delay_work
);
321 queue_delayed_work(kblockd_workqueue
, &q
->delay_work
, 0);
324 EXPORT_SYMBOL(blk_run_queue_async
);
327 * blk_run_queue - run a single device queue
328 * @q: The queue to run
331 * Invoke request handling on this queue, if it has pending work to do.
332 * May be used to restart queueing when a request has completed.
334 void blk_run_queue(struct request_queue
*q
)
338 spin_lock_irqsave(q
->queue_lock
, flags
);
340 spin_unlock_irqrestore(q
->queue_lock
, flags
);
342 EXPORT_SYMBOL(blk_run_queue
);
344 void blk_put_queue(struct request_queue
*q
)
346 kobject_put(&q
->kobj
);
350 * Note: If a driver supplied the queue lock, it should not zap that lock
351 * unexpectedly as some queue cleanup components like elevator_exit() and
352 * blk_throtl_exit() need queue lock.
354 void blk_cleanup_queue(struct request_queue
*q
)
357 * We know we have process context here, so we can be a little
358 * cautious and ensure that pending block actions on this device
359 * are done before moving on. Going into this function, we should
360 * not have processes doing IO to this device.
364 del_timer_sync(&q
->backing_dev_info
.laptop_mode_wb_timer
);
365 mutex_lock(&q
->sysfs_lock
);
366 queue_flag_set_unlocked(QUEUE_FLAG_DEAD
, q
);
367 mutex_unlock(&q
->sysfs_lock
);
370 elevator_exit(q
->elevator
);
376 EXPORT_SYMBOL(blk_cleanup_queue
);
378 static int blk_init_free_list(struct request_queue
*q
)
380 struct request_list
*rl
= &q
->rq
;
382 if (unlikely(rl
->rq_pool
))
385 rl
->count
[BLK_RW_SYNC
] = rl
->count
[BLK_RW_ASYNC
] = 0;
386 rl
->starved
[BLK_RW_SYNC
] = rl
->starved
[BLK_RW_ASYNC
] = 0;
388 init_waitqueue_head(&rl
->wait
[BLK_RW_SYNC
]);
389 init_waitqueue_head(&rl
->wait
[BLK_RW_ASYNC
]);
391 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, mempool_alloc_slab
,
392 mempool_free_slab
, request_cachep
, q
->node
);
400 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
402 return blk_alloc_queue_node(gfp_mask
, -1);
404 EXPORT_SYMBOL(blk_alloc_queue
);
406 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
408 struct request_queue
*q
;
411 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
412 gfp_mask
| __GFP_ZERO
, node_id
);
416 q
->backing_dev_info
.ra_pages
=
417 (VM_MAX_READAHEAD
* 1024) / PAGE_CACHE_SIZE
;
418 q
->backing_dev_info
.state
= 0;
419 q
->backing_dev_info
.capabilities
= BDI_CAP_MAP_COPY
;
420 q
->backing_dev_info
.name
= "block";
422 err
= bdi_init(&q
->backing_dev_info
);
424 kmem_cache_free(blk_requestq_cachep
, q
);
428 if (blk_throtl_init(q
)) {
429 kmem_cache_free(blk_requestq_cachep
, q
);
433 setup_timer(&q
->backing_dev_info
.laptop_mode_wb_timer
,
434 laptop_mode_timer_fn
, (unsigned long) q
);
435 setup_timer(&q
->timeout
, blk_rq_timed_out_timer
, (unsigned long) q
);
436 INIT_LIST_HEAD(&q
->timeout_list
);
437 INIT_LIST_HEAD(&q
->flush_queue
[0]);
438 INIT_LIST_HEAD(&q
->flush_queue
[1]);
439 INIT_LIST_HEAD(&q
->flush_data_in_flight
);
440 INIT_DELAYED_WORK(&q
->delay_work
, blk_delay_work
);
442 kobject_init(&q
->kobj
, &blk_queue_ktype
);
444 mutex_init(&q
->sysfs_lock
);
445 spin_lock_init(&q
->__queue_lock
);
448 * By default initialize queue_lock to internal lock and driver can
449 * override it later if need be.
451 q
->queue_lock
= &q
->__queue_lock
;
455 EXPORT_SYMBOL(blk_alloc_queue_node
);
458 * blk_init_queue - prepare a request queue for use with a block device
459 * @rfn: The function to be called to process requests that have been
460 * placed on the queue.
461 * @lock: Request queue spin lock
464 * If a block device wishes to use the standard request handling procedures,
465 * which sorts requests and coalesces adjacent requests, then it must
466 * call blk_init_queue(). The function @rfn will be called when there
467 * are requests on the queue that need to be processed. If the device
468 * supports plugging, then @rfn may not be called immediately when requests
469 * are available on the queue, but may be called at some time later instead.
470 * Plugged queues are generally unplugged when a buffer belonging to one
471 * of the requests on the queue is needed, or due to memory pressure.
473 * @rfn is not required, or even expected, to remove all requests off the
474 * queue, but only as many as it can handle at a time. If it does leave
475 * requests on the queue, it is responsible for arranging that the requests
476 * get dealt with eventually.
478 * The queue spin lock must be held while manipulating the requests on the
479 * request queue; this lock will be taken also from interrupt context, so irq
480 * disabling is needed for it.
482 * Function returns a pointer to the initialized request queue, or %NULL if
486 * blk_init_queue() must be paired with a blk_cleanup_queue() call
487 * when the block device is deactivated (such as at module unload).
490 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
492 return blk_init_queue_node(rfn
, lock
, -1);
494 EXPORT_SYMBOL(blk_init_queue
);
496 struct request_queue
*
497 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
499 struct request_queue
*uninit_q
, *q
;
501 uninit_q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
505 q
= blk_init_allocated_queue_node(uninit_q
, rfn
, lock
, node_id
);
507 blk_cleanup_queue(uninit_q
);
511 EXPORT_SYMBOL(blk_init_queue_node
);
513 struct request_queue
*
514 blk_init_allocated_queue(struct request_queue
*q
, request_fn_proc
*rfn
,
517 return blk_init_allocated_queue_node(q
, rfn
, lock
, -1);
519 EXPORT_SYMBOL(blk_init_allocated_queue
);
521 struct request_queue
*
522 blk_init_allocated_queue_node(struct request_queue
*q
, request_fn_proc
*rfn
,
523 spinlock_t
*lock
, int node_id
)
529 if (blk_init_free_list(q
))
533 q
->prep_rq_fn
= NULL
;
534 q
->unprep_rq_fn
= NULL
;
535 q
->queue_flags
= QUEUE_FLAG_DEFAULT
;
537 /* Override internal queue lock with supplied lock pointer */
539 q
->queue_lock
= lock
;
542 * This also sets hw/phys segments, boundary and size
544 blk_queue_make_request(q
, __make_request
);
546 q
->sg_reserved_size
= INT_MAX
;
551 if (!elevator_init(q
, NULL
)) {
552 blk_queue_congestion_threshold(q
);
558 EXPORT_SYMBOL(blk_init_allocated_queue_node
);
560 int blk_get_queue(struct request_queue
*q
)
562 if (likely(!test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
))) {
563 kobject_get(&q
->kobj
);
570 static inline void blk_free_request(struct request_queue
*q
, struct request
*rq
)
572 if (rq
->cmd_flags
& REQ_ELVPRIV
)
573 elv_put_request(q
, rq
);
574 mempool_free(rq
, q
->rq
.rq_pool
);
577 static struct request
*
578 blk_alloc_request(struct request_queue
*q
, int flags
, int priv
, gfp_t gfp_mask
)
580 struct request
*rq
= mempool_alloc(q
->rq
.rq_pool
, gfp_mask
);
587 rq
->cmd_flags
= flags
| REQ_ALLOCED
;
590 if (unlikely(elv_set_request(q
, rq
, gfp_mask
))) {
591 mempool_free(rq
, q
->rq
.rq_pool
);
594 rq
->cmd_flags
|= REQ_ELVPRIV
;
601 * ioc_batching returns true if the ioc is a valid batching request and
602 * should be given priority access to a request.
604 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
610 * Make sure the process is able to allocate at least 1 request
611 * even if the batch times out, otherwise we could theoretically
614 return ioc
->nr_batch_requests
== q
->nr_batching
||
615 (ioc
->nr_batch_requests
> 0
616 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
620 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
621 * will cause the process to be a "batcher" on all queues in the system. This
622 * is the behaviour we want though - once it gets a wakeup it should be given
625 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
627 if (!ioc
|| ioc_batching(q
, ioc
))
630 ioc
->nr_batch_requests
= q
->nr_batching
;
631 ioc
->last_waited
= jiffies
;
634 static void __freed_request(struct request_queue
*q
, int sync
)
636 struct request_list
*rl
= &q
->rq
;
638 if (rl
->count
[sync
] < queue_congestion_off_threshold(q
))
639 blk_clear_queue_congested(q
, sync
);
641 if (rl
->count
[sync
] + 1 <= q
->nr_requests
) {
642 if (waitqueue_active(&rl
->wait
[sync
]))
643 wake_up(&rl
->wait
[sync
]);
645 blk_clear_queue_full(q
, sync
);
650 * A request has just been released. Account for it, update the full and
651 * congestion status, wake up any waiters. Called under q->queue_lock.
653 static void freed_request(struct request_queue
*q
, int sync
, int priv
)
655 struct request_list
*rl
= &q
->rq
;
661 __freed_request(q
, sync
);
663 if (unlikely(rl
->starved
[sync
^ 1]))
664 __freed_request(q
, sync
^ 1);
668 * Determine if elevator data should be initialized when allocating the
669 * request associated with @bio.
671 static bool blk_rq_should_init_elevator(struct bio
*bio
)
677 * Flush requests do not use the elevator so skip initialization.
678 * This allows a request to share the flush and elevator data.
680 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
))
687 * Get a free request, queue_lock must be held.
688 * Returns NULL on failure, with queue_lock held.
689 * Returns !NULL on success, with queue_lock *not held*.
691 static struct request
*get_request(struct request_queue
*q
, int rw_flags
,
692 struct bio
*bio
, gfp_t gfp_mask
)
694 struct request
*rq
= NULL
;
695 struct request_list
*rl
= &q
->rq
;
696 struct io_context
*ioc
= NULL
;
697 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
698 int may_queue
, priv
= 0;
700 may_queue
= elv_may_queue(q
, rw_flags
);
701 if (may_queue
== ELV_MQUEUE_NO
)
704 if (rl
->count
[is_sync
]+1 >= queue_congestion_on_threshold(q
)) {
705 if (rl
->count
[is_sync
]+1 >= q
->nr_requests
) {
706 ioc
= current_io_context(GFP_ATOMIC
, q
->node
);
708 * The queue will fill after this allocation, so set
709 * it as full, and mark this process as "batching".
710 * This process will be allowed to complete a batch of
711 * requests, others will be blocked.
713 if (!blk_queue_full(q
, is_sync
)) {
714 ioc_set_batching(q
, ioc
);
715 blk_set_queue_full(q
, is_sync
);
717 if (may_queue
!= ELV_MQUEUE_MUST
718 && !ioc_batching(q
, ioc
)) {
720 * The queue is full and the allocating
721 * process is not a "batcher", and not
722 * exempted by the IO scheduler
728 blk_set_queue_congested(q
, is_sync
);
732 * Only allow batching queuers to allocate up to 50% over the defined
733 * limit of requests, otherwise we could have thousands of requests
734 * allocated with any setting of ->nr_requests
736 if (rl
->count
[is_sync
] >= (3 * q
->nr_requests
/ 2))
739 rl
->count
[is_sync
]++;
740 rl
->starved
[is_sync
] = 0;
742 if (blk_rq_should_init_elevator(bio
)) {
743 priv
= !test_bit(QUEUE_FLAG_ELVSWITCH
, &q
->queue_flags
);
748 if (blk_queue_io_stat(q
))
749 rw_flags
|= REQ_IO_STAT
;
750 spin_unlock_irq(q
->queue_lock
);
752 rq
= blk_alloc_request(q
, rw_flags
, priv
, gfp_mask
);
755 * Allocation failed presumably due to memory. Undo anything
756 * we might have messed up.
758 * Allocating task should really be put onto the front of the
759 * wait queue, but this is pretty rare.
761 spin_lock_irq(q
->queue_lock
);
762 freed_request(q
, is_sync
, priv
);
765 * in the very unlikely event that allocation failed and no
766 * requests for this direction was pending, mark us starved
767 * so that freeing of a request in the other direction will
768 * notice us. another possible fix would be to split the
769 * rq mempool into READ and WRITE
772 if (unlikely(rl
->count
[is_sync
] == 0))
773 rl
->starved
[is_sync
] = 1;
779 * ioc may be NULL here, and ioc_batching will be false. That's
780 * OK, if the queue is under the request limit then requests need
781 * not count toward the nr_batch_requests limit. There will always
782 * be some limit enforced by BLK_BATCH_TIME.
784 if (ioc_batching(q
, ioc
))
785 ioc
->nr_batch_requests
--;
787 trace_block_getrq(q
, bio
, rw_flags
& 1);
793 * No available requests for this queue, wait for some requests to become
796 * Called with q->queue_lock held, and returns with it unlocked.
798 static struct request
*get_request_wait(struct request_queue
*q
, int rw_flags
,
801 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
804 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
807 struct io_context
*ioc
;
808 struct request_list
*rl
= &q
->rq
;
810 prepare_to_wait_exclusive(&rl
->wait
[is_sync
], &wait
,
811 TASK_UNINTERRUPTIBLE
);
813 trace_block_sleeprq(q
, bio
, rw_flags
& 1);
815 spin_unlock_irq(q
->queue_lock
);
819 * After sleeping, we become a "batching" process and
820 * will be able to allocate at least one request, and
821 * up to a big batch of them for a small period time.
822 * See ioc_batching, ioc_set_batching
824 ioc
= current_io_context(GFP_NOIO
, q
->node
);
825 ioc_set_batching(q
, ioc
);
827 spin_lock_irq(q
->queue_lock
);
828 finish_wait(&rl
->wait
[is_sync
], &wait
);
830 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
836 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
840 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
842 spin_lock_irq(q
->queue_lock
);
843 if (gfp_mask
& __GFP_WAIT
) {
844 rq
= get_request_wait(q
, rw
, NULL
);
846 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
848 spin_unlock_irq(q
->queue_lock
);
850 /* q->queue_lock is unlocked at this point */
854 EXPORT_SYMBOL(blk_get_request
);
857 * blk_make_request - given a bio, allocate a corresponding struct request.
858 * @q: target request queue
859 * @bio: The bio describing the memory mappings that will be submitted for IO.
860 * It may be a chained-bio properly constructed by block/bio layer.
861 * @gfp_mask: gfp flags to be used for memory allocation
863 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
864 * type commands. Where the struct request needs to be farther initialized by
865 * the caller. It is passed a &struct bio, which describes the memory info of
868 * The caller of blk_make_request must make sure that bi_io_vec
869 * are set to describe the memory buffers. That bio_data_dir() will return
870 * the needed direction of the request. (And all bio's in the passed bio-chain
871 * are properly set accordingly)
873 * If called under none-sleepable conditions, mapped bio buffers must not
874 * need bouncing, by calling the appropriate masked or flagged allocator,
875 * suitable for the target device. Otherwise the call to blk_queue_bounce will
878 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
879 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
880 * anything but the first bio in the chain. Otherwise you risk waiting for IO
881 * completion of a bio that hasn't been submitted yet, thus resulting in a
882 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
883 * of bio_alloc(), as that avoids the mempool deadlock.
884 * If possible a big IO should be split into smaller parts when allocation
885 * fails. Partial allocation should not be an error, or you risk a live-lock.
887 struct request
*blk_make_request(struct request_queue
*q
, struct bio
*bio
,
890 struct request
*rq
= blk_get_request(q
, bio_data_dir(bio
), gfp_mask
);
893 return ERR_PTR(-ENOMEM
);
896 struct bio
*bounce_bio
= bio
;
899 blk_queue_bounce(q
, &bounce_bio
);
900 ret
= blk_rq_append_bio(q
, rq
, bounce_bio
);
909 EXPORT_SYMBOL(blk_make_request
);
912 * blk_requeue_request - put a request back on queue
913 * @q: request queue where request should be inserted
914 * @rq: request to be inserted
917 * Drivers often keep queueing requests until the hardware cannot accept
918 * more, when that condition happens we need to put the request back
919 * on the queue. Must be called with queue lock held.
921 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
923 blk_delete_timer(rq
);
924 blk_clear_rq_complete(rq
);
925 trace_block_rq_requeue(q
, rq
);
927 if (blk_rq_tagged(rq
))
928 blk_queue_end_tag(q
, rq
);
930 BUG_ON(blk_queued_rq(rq
));
932 elv_requeue_request(q
, rq
);
934 EXPORT_SYMBOL(blk_requeue_request
);
936 static void add_acct_request(struct request_queue
*q
, struct request
*rq
,
939 drive_stat_acct(rq
, 1);
940 __elv_add_request(q
, rq
, where
);
944 * blk_insert_request - insert a special request into a request queue
945 * @q: request queue where request should be inserted
946 * @rq: request to be inserted
947 * @at_head: insert request at head or tail of queue
948 * @data: private data
951 * Many block devices need to execute commands asynchronously, so they don't
952 * block the whole kernel from preemption during request execution. This is
953 * accomplished normally by inserting aritficial requests tagged as
954 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
955 * be scheduled for actual execution by the request queue.
957 * We have the option of inserting the head or the tail of the queue.
958 * Typically we use the tail for new ioctls and so forth. We use the head
959 * of the queue for things like a QUEUE_FULL message from a device, or a
960 * host that is unable to accept a particular command.
962 void blk_insert_request(struct request_queue
*q
, struct request
*rq
,
963 int at_head
, void *data
)
965 int where
= at_head
? ELEVATOR_INSERT_FRONT
: ELEVATOR_INSERT_BACK
;
969 * tell I/O scheduler that this isn't a regular read/write (ie it
970 * must not attempt merges on this) and that it acts as a soft
973 rq
->cmd_type
= REQ_TYPE_SPECIAL
;
977 spin_lock_irqsave(q
->queue_lock
, flags
);
980 * If command is tagged, release the tag
982 if (blk_rq_tagged(rq
))
983 blk_queue_end_tag(q
, rq
);
985 add_acct_request(q
, rq
, where
);
987 spin_unlock_irqrestore(q
->queue_lock
, flags
);
989 EXPORT_SYMBOL(blk_insert_request
);
991 static void part_round_stats_single(int cpu
, struct hd_struct
*part
,
994 if (now
== part
->stamp
)
997 if (part_in_flight(part
)) {
998 __part_stat_add(cpu
, part
, time_in_queue
,
999 part_in_flight(part
) * (now
- part
->stamp
));
1000 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
1006 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1007 * @cpu: cpu number for stats access
1008 * @part: target partition
1010 * The average IO queue length and utilisation statistics are maintained
1011 * by observing the current state of the queue length and the amount of
1012 * time it has been in this state for.
1014 * Normally, that accounting is done on IO completion, but that can result
1015 * in more than a second's worth of IO being accounted for within any one
1016 * second, leading to >100% utilisation. To deal with that, we call this
1017 * function to do a round-off before returning the results when reading
1018 * /proc/diskstats. This accounts immediately for all queue usage up to
1019 * the current jiffies and restarts the counters again.
1021 void part_round_stats(int cpu
, struct hd_struct
*part
)
1023 unsigned long now
= jiffies
;
1026 part_round_stats_single(cpu
, &part_to_disk(part
)->part0
, now
);
1027 part_round_stats_single(cpu
, part
, now
);
1029 EXPORT_SYMBOL_GPL(part_round_stats
);
1032 * queue lock must be held
1034 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1038 if (unlikely(--req
->ref_count
))
1041 elv_completed_request(q
, req
);
1043 /* this is a bio leak */
1044 WARN_ON(req
->bio
!= NULL
);
1047 * Request may not have originated from ll_rw_blk. if not,
1048 * it didn't come out of our reserved rq pools
1050 if (req
->cmd_flags
& REQ_ALLOCED
) {
1051 int is_sync
= rq_is_sync(req
) != 0;
1052 int priv
= req
->cmd_flags
& REQ_ELVPRIV
;
1054 BUG_ON(!list_empty(&req
->queuelist
));
1055 BUG_ON(!hlist_unhashed(&req
->hash
));
1057 blk_free_request(q
, req
);
1058 freed_request(q
, is_sync
, priv
);
1061 EXPORT_SYMBOL_GPL(__blk_put_request
);
1063 void blk_put_request(struct request
*req
)
1065 unsigned long flags
;
1066 struct request_queue
*q
= req
->q
;
1068 spin_lock_irqsave(q
->queue_lock
, flags
);
1069 __blk_put_request(q
, req
);
1070 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1072 EXPORT_SYMBOL(blk_put_request
);
1075 * blk_add_request_payload - add a payload to a request
1076 * @rq: request to update
1077 * @page: page backing the payload
1078 * @len: length of the payload.
1080 * This allows to later add a payload to an already submitted request by
1081 * a block driver. The driver needs to take care of freeing the payload
1084 * Note that this is a quite horrible hack and nothing but handling of
1085 * discard requests should ever use it.
1087 void blk_add_request_payload(struct request
*rq
, struct page
*page
,
1090 struct bio
*bio
= rq
->bio
;
1092 bio
->bi_io_vec
->bv_page
= page
;
1093 bio
->bi_io_vec
->bv_offset
= 0;
1094 bio
->bi_io_vec
->bv_len
= len
;
1098 bio
->bi_phys_segments
= 1;
1100 rq
->__data_len
= rq
->resid_len
= len
;
1101 rq
->nr_phys_segments
= 1;
1102 rq
->buffer
= bio_data(bio
);
1104 EXPORT_SYMBOL_GPL(blk_add_request_payload
);
1106 static bool bio_attempt_back_merge(struct request_queue
*q
, struct request
*req
,
1109 const int ff
= bio
->bi_rw
& REQ_FAILFAST_MASK
;
1111 if (!ll_back_merge_fn(q
, req
, bio
))
1114 trace_block_bio_backmerge(q
, bio
);
1116 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1117 blk_rq_set_mixed_merge(req
);
1119 req
->biotail
->bi_next
= bio
;
1121 req
->__data_len
+= bio
->bi_size
;
1122 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1124 drive_stat_acct(req
, 0);
1125 elv_bio_merged(q
, req
, bio
);
1129 static bool bio_attempt_front_merge(struct request_queue
*q
,
1130 struct request
*req
, struct bio
*bio
)
1132 const int ff
= bio
->bi_rw
& REQ_FAILFAST_MASK
;
1134 if (!ll_front_merge_fn(q
, req
, bio
))
1137 trace_block_bio_frontmerge(q
, bio
);
1139 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1140 blk_rq_set_mixed_merge(req
);
1142 bio
->bi_next
= req
->bio
;
1146 * may not be valid. if the low level driver said
1147 * it didn't need a bounce buffer then it better
1148 * not touch req->buffer either...
1150 req
->buffer
= bio_data(bio
);
1151 req
->__sector
= bio
->bi_sector
;
1152 req
->__data_len
+= bio
->bi_size
;
1153 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1155 drive_stat_acct(req
, 0);
1156 elv_bio_merged(q
, req
, bio
);
1161 * Attempts to merge with the plugged list in the current process. Returns
1162 * true if merge was successful, otherwise false.
1164 static bool attempt_plug_merge(struct task_struct
*tsk
, struct request_queue
*q
,
1167 struct blk_plug
*plug
;
1175 list_for_each_entry_reverse(rq
, &plug
->list
, queuelist
) {
1181 el_ret
= elv_try_merge(rq
, bio
);
1182 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1183 ret
= bio_attempt_back_merge(q
, rq
, bio
);
1186 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1187 ret
= bio_attempt_front_merge(q
, rq
, bio
);
1196 void init_request_from_bio(struct request
*req
, struct bio
*bio
)
1198 req
->cpu
= bio
->bi_comp_cpu
;
1199 req
->cmd_type
= REQ_TYPE_FS
;
1201 req
->cmd_flags
|= bio
->bi_rw
& REQ_COMMON_MASK
;
1202 if (bio
->bi_rw
& REQ_RAHEAD
)
1203 req
->cmd_flags
|= REQ_FAILFAST_MASK
;
1206 req
->__sector
= bio
->bi_sector
;
1207 req
->ioprio
= bio_prio(bio
);
1208 blk_rq_bio_prep(req
->q
, req
, bio
);
1211 static int __make_request(struct request_queue
*q
, struct bio
*bio
)
1213 const bool sync
= !!(bio
->bi_rw
& REQ_SYNC
);
1214 struct blk_plug
*plug
;
1215 int el_ret
, rw_flags
, where
= ELEVATOR_INSERT_SORT
;
1216 struct request
*req
;
1219 * low level driver can indicate that it wants pages above a
1220 * certain limit bounced to low memory (ie for highmem, or even
1221 * ISA dma in theory)
1223 blk_queue_bounce(q
, &bio
);
1225 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) {
1226 spin_lock_irq(q
->queue_lock
);
1227 where
= ELEVATOR_INSERT_FLUSH
;
1232 * Check if we can merge with the plugged list before grabbing
1235 if (attempt_plug_merge(current
, q
, bio
))
1238 spin_lock_irq(q
->queue_lock
);
1240 el_ret
= elv_merge(q
, &req
, bio
);
1241 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1242 if (bio_attempt_back_merge(q
, req
, bio
)) {
1243 if (!attempt_back_merge(q
, req
))
1244 elv_merged_request(q
, req
, el_ret
);
1247 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1248 if (bio_attempt_front_merge(q
, req
, bio
)) {
1249 if (!attempt_front_merge(q
, req
))
1250 elv_merged_request(q
, req
, el_ret
);
1257 * This sync check and mask will be re-done in init_request_from_bio(),
1258 * but we need to set it earlier to expose the sync flag to the
1259 * rq allocator and io schedulers.
1261 rw_flags
= bio_data_dir(bio
);
1263 rw_flags
|= REQ_SYNC
;
1266 * Grab a free request. This is might sleep but can not fail.
1267 * Returns with the queue unlocked.
1269 req
= get_request_wait(q
, rw_flags
, bio
);
1272 * After dropping the lock and possibly sleeping here, our request
1273 * may now be mergeable after it had proven unmergeable (above).
1274 * We don't worry about that case for efficiency. It won't happen
1275 * often, and the elevators are able to handle it.
1277 init_request_from_bio(req
, bio
);
1279 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
) ||
1280 bio_flagged(bio
, BIO_CPU_AFFINE
)) {
1281 req
->cpu
= blk_cpu_to_group(get_cpu());
1285 plug
= current
->plug
;
1288 * If this is the first request added after a plug, fire
1289 * of a plug trace. If others have been added before, check
1290 * if we have multiple devices in this plug. If so, make a
1291 * note to sort the list before dispatch.
1293 if (list_empty(&plug
->list
))
1294 trace_block_plug(q
);
1295 else if (!plug
->should_sort
) {
1296 struct request
*__rq
;
1298 __rq
= list_entry_rq(plug
->list
.prev
);
1300 plug
->should_sort
= 1;
1302 list_add_tail(&req
->queuelist
, &plug
->list
);
1303 drive_stat_acct(req
, 1);
1305 spin_lock_irq(q
->queue_lock
);
1306 add_acct_request(q
, req
, where
);
1309 spin_unlock_irq(q
->queue_lock
);
1316 * If bio->bi_dev is a partition, remap the location
1318 static inline void blk_partition_remap(struct bio
*bio
)
1320 struct block_device
*bdev
= bio
->bi_bdev
;
1322 if (bio_sectors(bio
) && bdev
!= bdev
->bd_contains
) {
1323 struct hd_struct
*p
= bdev
->bd_part
;
1325 bio
->bi_sector
+= p
->start_sect
;
1326 bio
->bi_bdev
= bdev
->bd_contains
;
1328 trace_block_bio_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
1330 bio
->bi_sector
- p
->start_sect
);
1334 static void handle_bad_sector(struct bio
*bio
)
1336 char b
[BDEVNAME_SIZE
];
1338 printk(KERN_INFO
"attempt to access beyond end of device\n");
1339 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
1340 bdevname(bio
->bi_bdev
, b
),
1342 (unsigned long long)bio
->bi_sector
+ bio_sectors(bio
),
1343 (long long)(i_size_read(bio
->bi_bdev
->bd_inode
) >> 9));
1345 set_bit(BIO_EOF
, &bio
->bi_flags
);
1348 #ifdef CONFIG_FAIL_MAKE_REQUEST
1350 static DECLARE_FAULT_ATTR(fail_make_request
);
1352 static int __init
setup_fail_make_request(char *str
)
1354 return setup_fault_attr(&fail_make_request
, str
);
1356 __setup("fail_make_request=", setup_fail_make_request
);
1358 static int should_fail_request(struct bio
*bio
)
1360 struct hd_struct
*part
= bio
->bi_bdev
->bd_part
;
1362 if (part_to_disk(part
)->part0
.make_it_fail
|| part
->make_it_fail
)
1363 return should_fail(&fail_make_request
, bio
->bi_size
);
1368 static int __init
fail_make_request_debugfs(void)
1370 return init_fault_attr_dentries(&fail_make_request
,
1371 "fail_make_request");
1374 late_initcall(fail_make_request_debugfs
);
1376 #else /* CONFIG_FAIL_MAKE_REQUEST */
1378 static inline int should_fail_request(struct bio
*bio
)
1383 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1386 * Check whether this bio extends beyond the end of the device.
1388 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
1395 /* Test device or partition size, when known. */
1396 maxsector
= i_size_read(bio
->bi_bdev
->bd_inode
) >> 9;
1398 sector_t sector
= bio
->bi_sector
;
1400 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
1402 * This may well happen - the kernel calls bread()
1403 * without checking the size of the device, e.g., when
1404 * mounting a device.
1406 handle_bad_sector(bio
);
1415 * generic_make_request - hand a buffer to its device driver for I/O
1416 * @bio: The bio describing the location in memory and on the device.
1418 * generic_make_request() is used to make I/O requests of block
1419 * devices. It is passed a &struct bio, which describes the I/O that needs
1422 * generic_make_request() does not return any status. The
1423 * success/failure status of the request, along with notification of
1424 * completion, is delivered asynchronously through the bio->bi_end_io
1425 * function described (one day) else where.
1427 * The caller of generic_make_request must make sure that bi_io_vec
1428 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1429 * set to describe the device address, and the
1430 * bi_end_io and optionally bi_private are set to describe how
1431 * completion notification should be signaled.
1433 * generic_make_request and the drivers it calls may use bi_next if this
1434 * bio happens to be merged with someone else, and may change bi_dev and
1435 * bi_sector for remaps as it sees fit. So the values of these fields
1436 * should NOT be depended on after the call to generic_make_request.
1438 static inline void __generic_make_request(struct bio
*bio
)
1440 struct request_queue
*q
;
1441 sector_t old_sector
;
1442 int ret
, nr_sectors
= bio_sectors(bio
);
1448 if (bio_check_eod(bio
, nr_sectors
))
1452 * Resolve the mapping until finished. (drivers are
1453 * still free to implement/resolve their own stacking
1454 * by explicitly returning 0)
1456 * NOTE: we don't repeat the blk_size check for each new device.
1457 * Stacking drivers are expected to know what they are doing.
1462 char b
[BDEVNAME_SIZE
];
1464 q
= bdev_get_queue(bio
->bi_bdev
);
1467 "generic_make_request: Trying to access "
1468 "nonexistent block-device %s (%Lu)\n",
1469 bdevname(bio
->bi_bdev
, b
),
1470 (long long) bio
->bi_sector
);
1474 if (unlikely(!(bio
->bi_rw
& REQ_DISCARD
) &&
1475 nr_sectors
> queue_max_hw_sectors(q
))) {
1476 printk(KERN_ERR
"bio too big device %s (%u > %u)\n",
1477 bdevname(bio
->bi_bdev
, b
),
1479 queue_max_hw_sectors(q
));
1483 if (unlikely(test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)))
1486 if (should_fail_request(bio
))
1490 * If this device has partitions, remap block n
1491 * of partition p to block n+start(p) of the disk.
1493 blk_partition_remap(bio
);
1495 if (bio_integrity_enabled(bio
) && bio_integrity_prep(bio
))
1498 if (old_sector
!= -1)
1499 trace_block_bio_remap(q
, bio
, old_dev
, old_sector
);
1501 old_sector
= bio
->bi_sector
;
1502 old_dev
= bio
->bi_bdev
->bd_dev
;
1504 if (bio_check_eod(bio
, nr_sectors
))
1508 * Filter flush bio's early so that make_request based
1509 * drivers without flush support don't have to worry
1512 if ((bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) && !q
->flush_flags
) {
1513 bio
->bi_rw
&= ~(REQ_FLUSH
| REQ_FUA
);
1520 if ((bio
->bi_rw
& REQ_DISCARD
) &&
1521 (!blk_queue_discard(q
) ||
1522 ((bio
->bi_rw
& REQ_SECURE
) &&
1523 !blk_queue_secdiscard(q
)))) {
1528 if (blk_throtl_bio(q
, &bio
))
1532 * If bio = NULL, bio has been throttled and will be submitted
1538 trace_block_bio_queue(q
, bio
);
1540 ret
= q
->make_request_fn(q
, bio
);
1546 bio_endio(bio
, err
);
1550 * We only want one ->make_request_fn to be active at a time,
1551 * else stack usage with stacked devices could be a problem.
1552 * So use current->bio_list to keep a list of requests
1553 * submited by a make_request_fn function.
1554 * current->bio_list is also used as a flag to say if
1555 * generic_make_request is currently active in this task or not.
1556 * If it is NULL, then no make_request is active. If it is non-NULL,
1557 * then a make_request is active, and new requests should be added
1560 void generic_make_request(struct bio
*bio
)
1562 struct bio_list bio_list_on_stack
;
1564 if (current
->bio_list
) {
1565 /* make_request is active */
1566 bio_list_add(current
->bio_list
, bio
);
1569 /* following loop may be a bit non-obvious, and so deserves some
1571 * Before entering the loop, bio->bi_next is NULL (as all callers
1572 * ensure that) so we have a list with a single bio.
1573 * We pretend that we have just taken it off a longer list, so
1574 * we assign bio_list to a pointer to the bio_list_on_stack,
1575 * thus initialising the bio_list of new bios to be
1576 * added. __generic_make_request may indeed add some more bios
1577 * through a recursive call to generic_make_request. If it
1578 * did, we find a non-NULL value in bio_list and re-enter the loop
1579 * from the top. In this case we really did just take the bio
1580 * of the top of the list (no pretending) and so remove it from
1581 * bio_list, and call into __generic_make_request again.
1583 * The loop was structured like this to make only one call to
1584 * __generic_make_request (which is important as it is large and
1585 * inlined) and to keep the structure simple.
1587 BUG_ON(bio
->bi_next
);
1588 bio_list_init(&bio_list_on_stack
);
1589 current
->bio_list
= &bio_list_on_stack
;
1591 __generic_make_request(bio
);
1592 bio
= bio_list_pop(current
->bio_list
);
1594 current
->bio_list
= NULL
; /* deactivate */
1596 EXPORT_SYMBOL(generic_make_request
);
1599 * submit_bio - submit a bio to the block device layer for I/O
1600 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1601 * @bio: The &struct bio which describes the I/O
1603 * submit_bio() is very similar in purpose to generic_make_request(), and
1604 * uses that function to do most of the work. Both are fairly rough
1605 * interfaces; @bio must be presetup and ready for I/O.
1608 void submit_bio(int rw
, struct bio
*bio
)
1610 int count
= bio_sectors(bio
);
1615 * If it's a regular read/write or a barrier with data attached,
1616 * go through the normal accounting stuff before submission.
1618 if (bio_has_data(bio
) && !(rw
& REQ_DISCARD
)) {
1620 count_vm_events(PGPGOUT
, count
);
1622 task_io_account_read(bio
->bi_size
);
1623 count_vm_events(PGPGIN
, count
);
1626 if (unlikely(block_dump
)) {
1627 char b
[BDEVNAME_SIZE
];
1628 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s (%u sectors)\n",
1629 current
->comm
, task_pid_nr(current
),
1630 (rw
& WRITE
) ? "WRITE" : "READ",
1631 (unsigned long long)bio
->bi_sector
,
1632 bdevname(bio
->bi_bdev
, b
),
1637 generic_make_request(bio
);
1639 EXPORT_SYMBOL(submit_bio
);
1642 * blk_rq_check_limits - Helper function to check a request for the queue limit
1644 * @rq: the request being checked
1647 * @rq may have been made based on weaker limitations of upper-level queues
1648 * in request stacking drivers, and it may violate the limitation of @q.
1649 * Since the block layer and the underlying device driver trust @rq
1650 * after it is inserted to @q, it should be checked against @q before
1651 * the insertion using this generic function.
1653 * This function should also be useful for request stacking drivers
1654 * in some cases below, so export this function.
1655 * Request stacking drivers like request-based dm may change the queue
1656 * limits while requests are in the queue (e.g. dm's table swapping).
1657 * Such request stacking drivers should check those requests agaist
1658 * the new queue limits again when they dispatch those requests,
1659 * although such checkings are also done against the old queue limits
1660 * when submitting requests.
1662 int blk_rq_check_limits(struct request_queue
*q
, struct request
*rq
)
1664 if (rq
->cmd_flags
& REQ_DISCARD
)
1667 if (blk_rq_sectors(rq
) > queue_max_sectors(q
) ||
1668 blk_rq_bytes(rq
) > queue_max_hw_sectors(q
) << 9) {
1669 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
1674 * queue's settings related to segment counting like q->bounce_pfn
1675 * may differ from that of other stacking queues.
1676 * Recalculate it to check the request correctly on this queue's
1679 blk_recalc_rq_segments(rq
);
1680 if (rq
->nr_phys_segments
> queue_max_segments(q
)) {
1681 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
1687 EXPORT_SYMBOL_GPL(blk_rq_check_limits
);
1690 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1691 * @q: the queue to submit the request
1692 * @rq: the request being queued
1694 int blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
1696 unsigned long flags
;
1698 if (blk_rq_check_limits(q
, rq
))
1701 #ifdef CONFIG_FAIL_MAKE_REQUEST
1702 if (rq
->rq_disk
&& rq
->rq_disk
->part0
.make_it_fail
&&
1703 should_fail(&fail_make_request
, blk_rq_bytes(rq
)))
1707 spin_lock_irqsave(q
->queue_lock
, flags
);
1710 * Submitting request must be dequeued before calling this function
1711 * because it will be linked to another request_queue
1713 BUG_ON(blk_queued_rq(rq
));
1715 add_acct_request(q
, rq
, ELEVATOR_INSERT_BACK
);
1716 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1720 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
1723 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1724 * @rq: request to examine
1727 * A request could be merge of IOs which require different failure
1728 * handling. This function determines the number of bytes which
1729 * can be failed from the beginning of the request without
1730 * crossing into area which need to be retried further.
1733 * The number of bytes to fail.
1736 * queue_lock must be held.
1738 unsigned int blk_rq_err_bytes(const struct request
*rq
)
1740 unsigned int ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
1741 unsigned int bytes
= 0;
1744 if (!(rq
->cmd_flags
& REQ_MIXED_MERGE
))
1745 return blk_rq_bytes(rq
);
1748 * Currently the only 'mixing' which can happen is between
1749 * different fastfail types. We can safely fail portions
1750 * which have all the failfast bits that the first one has -
1751 * the ones which are at least as eager to fail as the first
1754 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
1755 if ((bio
->bi_rw
& ff
) != ff
)
1757 bytes
+= bio
->bi_size
;
1760 /* this could lead to infinite loop */
1761 BUG_ON(blk_rq_bytes(rq
) && !bytes
);
1764 EXPORT_SYMBOL_GPL(blk_rq_err_bytes
);
1766 static void blk_account_io_completion(struct request
*req
, unsigned int bytes
)
1768 if (blk_do_io_stat(req
)) {
1769 const int rw
= rq_data_dir(req
);
1770 struct hd_struct
*part
;
1773 cpu
= part_stat_lock();
1775 part_stat_add(cpu
, part
, sectors
[rw
], bytes
>> 9);
1780 static void blk_account_io_done(struct request
*req
)
1783 * Account IO completion. flush_rq isn't accounted as a
1784 * normal IO on queueing nor completion. Accounting the
1785 * containing request is enough.
1787 if (blk_do_io_stat(req
) && !(req
->cmd_flags
& REQ_FLUSH_SEQ
)) {
1788 unsigned long duration
= jiffies
- req
->start_time
;
1789 const int rw
= rq_data_dir(req
);
1790 struct hd_struct
*part
;
1793 cpu
= part_stat_lock();
1796 part_stat_inc(cpu
, part
, ios
[rw
]);
1797 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
1798 part_round_stats(cpu
, part
);
1799 part_dec_in_flight(part
, rw
);
1801 hd_struct_put(part
);
1807 * blk_peek_request - peek at the top of a request queue
1808 * @q: request queue to peek at
1811 * Return the request at the top of @q. The returned request
1812 * should be started using blk_start_request() before LLD starts
1816 * Pointer to the request at the top of @q if available. Null
1820 * queue_lock must be held.
1822 struct request
*blk_peek_request(struct request_queue
*q
)
1827 while ((rq
= __elv_next_request(q
)) != NULL
) {
1828 if (!(rq
->cmd_flags
& REQ_STARTED
)) {
1830 * This is the first time the device driver
1831 * sees this request (possibly after
1832 * requeueing). Notify IO scheduler.
1834 if (rq
->cmd_flags
& REQ_SORTED
)
1835 elv_activate_rq(q
, rq
);
1838 * just mark as started even if we don't start
1839 * it, a request that has been delayed should
1840 * not be passed by new incoming requests
1842 rq
->cmd_flags
|= REQ_STARTED
;
1843 trace_block_rq_issue(q
, rq
);
1846 if (!q
->boundary_rq
|| q
->boundary_rq
== rq
) {
1847 q
->end_sector
= rq_end_sector(rq
);
1848 q
->boundary_rq
= NULL
;
1851 if (rq
->cmd_flags
& REQ_DONTPREP
)
1854 if (q
->dma_drain_size
&& blk_rq_bytes(rq
)) {
1856 * make sure space for the drain appears we
1857 * know we can do this because max_hw_segments
1858 * has been adjusted to be one fewer than the
1861 rq
->nr_phys_segments
++;
1867 ret
= q
->prep_rq_fn(q
, rq
);
1868 if (ret
== BLKPREP_OK
) {
1870 } else if (ret
== BLKPREP_DEFER
) {
1872 * the request may have been (partially) prepped.
1873 * we need to keep this request in the front to
1874 * avoid resource deadlock. REQ_STARTED will
1875 * prevent other fs requests from passing this one.
1877 if (q
->dma_drain_size
&& blk_rq_bytes(rq
) &&
1878 !(rq
->cmd_flags
& REQ_DONTPREP
)) {
1880 * remove the space for the drain we added
1881 * so that we don't add it again
1883 --rq
->nr_phys_segments
;
1888 } else if (ret
== BLKPREP_KILL
) {
1889 rq
->cmd_flags
|= REQ_QUIET
;
1891 * Mark this request as started so we don't trigger
1892 * any debug logic in the end I/O path.
1894 blk_start_request(rq
);
1895 __blk_end_request_all(rq
, -EIO
);
1897 printk(KERN_ERR
"%s: bad return=%d\n", __func__
, ret
);
1904 EXPORT_SYMBOL(blk_peek_request
);
1906 void blk_dequeue_request(struct request
*rq
)
1908 struct request_queue
*q
= rq
->q
;
1910 BUG_ON(list_empty(&rq
->queuelist
));
1911 BUG_ON(ELV_ON_HASH(rq
));
1913 list_del_init(&rq
->queuelist
);
1916 * the time frame between a request being removed from the lists
1917 * and to it is freed is accounted as io that is in progress at
1920 if (blk_account_rq(rq
)) {
1921 q
->in_flight
[rq_is_sync(rq
)]++;
1922 set_io_start_time_ns(rq
);
1927 * blk_start_request - start request processing on the driver
1928 * @req: request to dequeue
1931 * Dequeue @req and start timeout timer on it. This hands off the
1932 * request to the driver.
1934 * Block internal functions which don't want to start timer should
1935 * call blk_dequeue_request().
1938 * queue_lock must be held.
1940 void blk_start_request(struct request
*req
)
1942 blk_dequeue_request(req
);
1945 * We are now handing the request to the hardware, initialize
1946 * resid_len to full count and add the timeout handler.
1948 req
->resid_len
= blk_rq_bytes(req
);
1949 if (unlikely(blk_bidi_rq(req
)))
1950 req
->next_rq
->resid_len
= blk_rq_bytes(req
->next_rq
);
1954 EXPORT_SYMBOL(blk_start_request
);
1957 * blk_fetch_request - fetch a request from a request queue
1958 * @q: request queue to fetch a request from
1961 * Return the request at the top of @q. The request is started on
1962 * return and LLD can start processing it immediately.
1965 * Pointer to the request at the top of @q if available. Null
1969 * queue_lock must be held.
1971 struct request
*blk_fetch_request(struct request_queue
*q
)
1975 rq
= blk_peek_request(q
);
1977 blk_start_request(rq
);
1980 EXPORT_SYMBOL(blk_fetch_request
);
1983 * blk_update_request - Special helper function for request stacking drivers
1984 * @req: the request being processed
1985 * @error: %0 for success, < %0 for error
1986 * @nr_bytes: number of bytes to complete @req
1989 * Ends I/O on a number of bytes attached to @req, but doesn't complete
1990 * the request structure even if @req doesn't have leftover.
1991 * If @req has leftover, sets it up for the next range of segments.
1993 * This special helper function is only for request stacking drivers
1994 * (e.g. request-based dm) so that they can handle partial completion.
1995 * Actual device drivers should use blk_end_request instead.
1997 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
1998 * %false return from this function.
2001 * %false - this request doesn't have any more data
2002 * %true - this request has more data
2004 bool blk_update_request(struct request
*req
, int error
, unsigned int nr_bytes
)
2006 int total_bytes
, bio_nbytes
, next_idx
= 0;
2012 trace_block_rq_complete(req
->q
, req
);
2015 * For fs requests, rq is just carrier of independent bio's
2016 * and each partial completion should be handled separately.
2017 * Reset per-request error on each partial completion.
2019 * TODO: tj: This is too subtle. It would be better to let
2020 * low level drivers do what they see fit.
2022 if (req
->cmd_type
== REQ_TYPE_FS
)
2025 if (error
&& req
->cmd_type
== REQ_TYPE_FS
&&
2026 !(req
->cmd_flags
& REQ_QUIET
)) {
2031 error_type
= "recoverable transport";
2034 error_type
= "critical target";
2037 error_type
= "critical nexus";
2044 printk(KERN_ERR
"end_request: %s error, dev %s, sector %llu\n",
2045 error_type
, req
->rq_disk
? req
->rq_disk
->disk_name
: "?",
2046 (unsigned long long)blk_rq_pos(req
));
2049 blk_account_io_completion(req
, nr_bytes
);
2051 total_bytes
= bio_nbytes
= 0;
2052 while ((bio
= req
->bio
) != NULL
) {
2055 if (nr_bytes
>= bio
->bi_size
) {
2056 req
->bio
= bio
->bi_next
;
2057 nbytes
= bio
->bi_size
;
2058 req_bio_endio(req
, bio
, nbytes
, error
);
2062 int idx
= bio
->bi_idx
+ next_idx
;
2064 if (unlikely(idx
>= bio
->bi_vcnt
)) {
2065 blk_dump_rq_flags(req
, "__end_that");
2066 printk(KERN_ERR
"%s: bio idx %d >= vcnt %d\n",
2067 __func__
, idx
, bio
->bi_vcnt
);
2071 nbytes
= bio_iovec_idx(bio
, idx
)->bv_len
;
2072 BIO_BUG_ON(nbytes
> bio
->bi_size
);
2075 * not a complete bvec done
2077 if (unlikely(nbytes
> nr_bytes
)) {
2078 bio_nbytes
+= nr_bytes
;
2079 total_bytes
+= nr_bytes
;
2084 * advance to the next vector
2087 bio_nbytes
+= nbytes
;
2090 total_bytes
+= nbytes
;
2096 * end more in this run, or just return 'not-done'
2098 if (unlikely(nr_bytes
<= 0))
2108 * Reset counters so that the request stacking driver
2109 * can find how many bytes remain in the request
2112 req
->__data_len
= 0;
2117 * if the request wasn't completed, update state
2120 req_bio_endio(req
, bio
, bio_nbytes
, error
);
2121 bio
->bi_idx
+= next_idx
;
2122 bio_iovec(bio
)->bv_offset
+= nr_bytes
;
2123 bio_iovec(bio
)->bv_len
-= nr_bytes
;
2126 req
->__data_len
-= total_bytes
;
2127 req
->buffer
= bio_data(req
->bio
);
2129 /* update sector only for requests with clear definition of sector */
2130 if (req
->cmd_type
== REQ_TYPE_FS
|| (req
->cmd_flags
& REQ_DISCARD
))
2131 req
->__sector
+= total_bytes
>> 9;
2133 /* mixed attributes always follow the first bio */
2134 if (req
->cmd_flags
& REQ_MIXED_MERGE
) {
2135 req
->cmd_flags
&= ~REQ_FAILFAST_MASK
;
2136 req
->cmd_flags
|= req
->bio
->bi_rw
& REQ_FAILFAST_MASK
;
2140 * If total number of sectors is less than the first segment
2141 * size, something has gone terribly wrong.
2143 if (blk_rq_bytes(req
) < blk_rq_cur_bytes(req
)) {
2144 blk_dump_rq_flags(req
, "request botched");
2145 req
->__data_len
= blk_rq_cur_bytes(req
);
2148 /* recalculate the number of segments */
2149 blk_recalc_rq_segments(req
);
2153 EXPORT_SYMBOL_GPL(blk_update_request
);
2155 static bool blk_update_bidi_request(struct request
*rq
, int error
,
2156 unsigned int nr_bytes
,
2157 unsigned int bidi_bytes
)
2159 if (blk_update_request(rq
, error
, nr_bytes
))
2162 /* Bidi request must be completed as a whole */
2163 if (unlikely(blk_bidi_rq(rq
)) &&
2164 blk_update_request(rq
->next_rq
, error
, bidi_bytes
))
2167 if (blk_queue_add_random(rq
->q
))
2168 add_disk_randomness(rq
->rq_disk
);
2174 * blk_unprep_request - unprepare a request
2177 * This function makes a request ready for complete resubmission (or
2178 * completion). It happens only after all error handling is complete,
2179 * so represents the appropriate moment to deallocate any resources
2180 * that were allocated to the request in the prep_rq_fn. The queue
2181 * lock is held when calling this.
2183 void blk_unprep_request(struct request
*req
)
2185 struct request_queue
*q
= req
->q
;
2187 req
->cmd_flags
&= ~REQ_DONTPREP
;
2188 if (q
->unprep_rq_fn
)
2189 q
->unprep_rq_fn(q
, req
);
2191 EXPORT_SYMBOL_GPL(blk_unprep_request
);
2194 * queue lock must be held
2196 static void blk_finish_request(struct request
*req
, int error
)
2198 if (blk_rq_tagged(req
))
2199 blk_queue_end_tag(req
->q
, req
);
2201 BUG_ON(blk_queued_rq(req
));
2203 if (unlikely(laptop_mode
) && req
->cmd_type
== REQ_TYPE_FS
)
2204 laptop_io_completion(&req
->q
->backing_dev_info
);
2206 blk_delete_timer(req
);
2208 if (req
->cmd_flags
& REQ_DONTPREP
)
2209 blk_unprep_request(req
);
2212 blk_account_io_done(req
);
2215 req
->end_io(req
, error
);
2217 if (blk_bidi_rq(req
))
2218 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
2220 __blk_put_request(req
->q
, req
);
2225 * blk_end_bidi_request - Complete a bidi request
2226 * @rq: the request to complete
2227 * @error: %0 for success, < %0 for error
2228 * @nr_bytes: number of bytes to complete @rq
2229 * @bidi_bytes: number of bytes to complete @rq->next_rq
2232 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2233 * Drivers that supports bidi can safely call this member for any
2234 * type of request, bidi or uni. In the later case @bidi_bytes is
2238 * %false - we are done with this request
2239 * %true - still buffers pending for this request
2241 static bool blk_end_bidi_request(struct request
*rq
, int error
,
2242 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2244 struct request_queue
*q
= rq
->q
;
2245 unsigned long flags
;
2247 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2250 spin_lock_irqsave(q
->queue_lock
, flags
);
2251 blk_finish_request(rq
, error
);
2252 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2258 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2259 * @rq: the request to complete
2260 * @error: %0 for success, < %0 for error
2261 * @nr_bytes: number of bytes to complete @rq
2262 * @bidi_bytes: number of bytes to complete @rq->next_rq
2265 * Identical to blk_end_bidi_request() except that queue lock is
2266 * assumed to be locked on entry and remains so on return.
2269 * %false - we are done with this request
2270 * %true - still buffers pending for this request
2272 static bool __blk_end_bidi_request(struct request
*rq
, int error
,
2273 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2275 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2278 blk_finish_request(rq
, error
);
2284 * blk_end_request - Helper function for drivers to complete the request.
2285 * @rq: the request being processed
2286 * @error: %0 for success, < %0 for error
2287 * @nr_bytes: number of bytes to complete
2290 * Ends I/O on a number of bytes attached to @rq.
2291 * If @rq has leftover, sets it up for the next range of segments.
2294 * %false - we are done with this request
2295 * %true - still buffers pending for this request
2297 bool blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2299 return blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2301 EXPORT_SYMBOL(blk_end_request
);
2304 * blk_end_request_all - Helper function for drives to finish the request.
2305 * @rq: the request to finish
2306 * @error: %0 for success, < %0 for error
2309 * Completely finish @rq.
2311 void blk_end_request_all(struct request
*rq
, int error
)
2314 unsigned int bidi_bytes
= 0;
2316 if (unlikely(blk_bidi_rq(rq
)))
2317 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2319 pending
= blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2322 EXPORT_SYMBOL(blk_end_request_all
);
2325 * blk_end_request_cur - Helper function to finish the current request chunk.
2326 * @rq: the request to finish the current chunk for
2327 * @error: %0 for success, < %0 for error
2330 * Complete the current consecutively mapped chunk from @rq.
2333 * %false - we are done with this request
2334 * %true - still buffers pending for this request
2336 bool blk_end_request_cur(struct request
*rq
, int error
)
2338 return blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2340 EXPORT_SYMBOL(blk_end_request_cur
);
2343 * blk_end_request_err - Finish a request till the next failure boundary.
2344 * @rq: the request to finish till the next failure boundary for
2345 * @error: must be negative errno
2348 * Complete @rq till the next failure boundary.
2351 * %false - we are done with this request
2352 * %true - still buffers pending for this request
2354 bool blk_end_request_err(struct request
*rq
, int error
)
2356 WARN_ON(error
>= 0);
2357 return blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2359 EXPORT_SYMBOL_GPL(blk_end_request_err
);
2362 * __blk_end_request - Helper function for drivers to complete the request.
2363 * @rq: the request being processed
2364 * @error: %0 for success, < %0 for error
2365 * @nr_bytes: number of bytes to complete
2368 * Must be called with queue lock held unlike blk_end_request().
2371 * %false - we are done with this request
2372 * %true - still buffers pending for this request
2374 bool __blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2376 return __blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2378 EXPORT_SYMBOL(__blk_end_request
);
2381 * __blk_end_request_all - Helper function for drives to finish the request.
2382 * @rq: the request to finish
2383 * @error: %0 for success, < %0 for error
2386 * Completely finish @rq. Must be called with queue lock held.
2388 void __blk_end_request_all(struct request
*rq
, int error
)
2391 unsigned int bidi_bytes
= 0;
2393 if (unlikely(blk_bidi_rq(rq
)))
2394 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2396 pending
= __blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2399 EXPORT_SYMBOL(__blk_end_request_all
);
2402 * __blk_end_request_cur - Helper function to finish the current request chunk.
2403 * @rq: the request to finish the current chunk for
2404 * @error: %0 for success, < %0 for error
2407 * Complete the current consecutively mapped chunk from @rq. Must
2408 * be called with queue lock held.
2411 * %false - we are done with this request
2412 * %true - still buffers pending for this request
2414 bool __blk_end_request_cur(struct request
*rq
, int error
)
2416 return __blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2418 EXPORT_SYMBOL(__blk_end_request_cur
);
2421 * __blk_end_request_err - Finish a request till the next failure boundary.
2422 * @rq: the request to finish till the next failure boundary for
2423 * @error: must be negative errno
2426 * Complete @rq till the next failure boundary. Must be called
2427 * with queue lock held.
2430 * %false - we are done with this request
2431 * %true - still buffers pending for this request
2433 bool __blk_end_request_err(struct request
*rq
, int error
)
2435 WARN_ON(error
>= 0);
2436 return __blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2438 EXPORT_SYMBOL_GPL(__blk_end_request_err
);
2440 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
2443 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2444 rq
->cmd_flags
|= bio
->bi_rw
& REQ_WRITE
;
2446 if (bio_has_data(bio
)) {
2447 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2448 rq
->buffer
= bio_data(bio
);
2450 rq
->__data_len
= bio
->bi_size
;
2451 rq
->bio
= rq
->biotail
= bio
;
2454 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2457 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2459 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2460 * @rq: the request to be flushed
2463 * Flush all pages in @rq.
2465 void rq_flush_dcache_pages(struct request
*rq
)
2467 struct req_iterator iter
;
2468 struct bio_vec
*bvec
;
2470 rq_for_each_segment(bvec
, rq
, iter
)
2471 flush_dcache_page(bvec
->bv_page
);
2473 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages
);
2477 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2478 * @q : the queue of the device being checked
2481 * Check if underlying low-level drivers of a device are busy.
2482 * If the drivers want to export their busy state, they must set own
2483 * exporting function using blk_queue_lld_busy() first.
2485 * Basically, this function is used only by request stacking drivers
2486 * to stop dispatching requests to underlying devices when underlying
2487 * devices are busy. This behavior helps more I/O merging on the queue
2488 * of the request stacking driver and prevents I/O throughput regression
2489 * on burst I/O load.
2492 * 0 - Not busy (The request stacking driver should dispatch request)
2493 * 1 - Busy (The request stacking driver should stop dispatching request)
2495 int blk_lld_busy(struct request_queue
*q
)
2498 return q
->lld_busy_fn(q
);
2502 EXPORT_SYMBOL_GPL(blk_lld_busy
);
2505 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2506 * @rq: the clone request to be cleaned up
2509 * Free all bios in @rq for a cloned request.
2511 void blk_rq_unprep_clone(struct request
*rq
)
2515 while ((bio
= rq
->bio
) != NULL
) {
2516 rq
->bio
= bio
->bi_next
;
2521 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone
);
2524 * Copy attributes of the original request to the clone request.
2525 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2527 static void __blk_rq_prep_clone(struct request
*dst
, struct request
*src
)
2529 dst
->cpu
= src
->cpu
;
2530 dst
->cmd_flags
= (src
->cmd_flags
& REQ_CLONE_MASK
) | REQ_NOMERGE
;
2531 dst
->cmd_type
= src
->cmd_type
;
2532 dst
->__sector
= blk_rq_pos(src
);
2533 dst
->__data_len
= blk_rq_bytes(src
);
2534 dst
->nr_phys_segments
= src
->nr_phys_segments
;
2535 dst
->ioprio
= src
->ioprio
;
2536 dst
->extra_len
= src
->extra_len
;
2540 * blk_rq_prep_clone - Helper function to setup clone request
2541 * @rq: the request to be setup
2542 * @rq_src: original request to be cloned
2543 * @bs: bio_set that bios for clone are allocated from
2544 * @gfp_mask: memory allocation mask for bio
2545 * @bio_ctr: setup function to be called for each clone bio.
2546 * Returns %0 for success, non %0 for failure.
2547 * @data: private data to be passed to @bio_ctr
2550 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2551 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2552 * are not copied, and copying such parts is the caller's responsibility.
2553 * Also, pages which the original bios are pointing to are not copied
2554 * and the cloned bios just point same pages.
2555 * So cloned bios must be completed before original bios, which means
2556 * the caller must complete @rq before @rq_src.
2558 int blk_rq_prep_clone(struct request
*rq
, struct request
*rq_src
,
2559 struct bio_set
*bs
, gfp_t gfp_mask
,
2560 int (*bio_ctr
)(struct bio
*, struct bio
*, void *),
2563 struct bio
*bio
, *bio_src
;
2568 blk_rq_init(NULL
, rq
);
2570 __rq_for_each_bio(bio_src
, rq_src
) {
2571 bio
= bio_alloc_bioset(gfp_mask
, bio_src
->bi_max_vecs
, bs
);
2575 __bio_clone(bio
, bio_src
);
2577 if (bio_integrity(bio_src
) &&
2578 bio_integrity_clone(bio
, bio_src
, gfp_mask
, bs
))
2581 if (bio_ctr
&& bio_ctr(bio
, bio_src
, data
))
2585 rq
->biotail
->bi_next
= bio
;
2588 rq
->bio
= rq
->biotail
= bio
;
2591 __blk_rq_prep_clone(rq
, rq_src
);
2598 blk_rq_unprep_clone(rq
);
2602 EXPORT_SYMBOL_GPL(blk_rq_prep_clone
);
2604 int kblockd_schedule_work(struct request_queue
*q
, struct work_struct
*work
)
2606 return queue_work(kblockd_workqueue
, work
);
2608 EXPORT_SYMBOL(kblockd_schedule_work
);
2610 int kblockd_schedule_delayed_work(struct request_queue
*q
,
2611 struct delayed_work
*dwork
, unsigned long delay
)
2613 return queue_delayed_work(kblockd_workqueue
, dwork
, delay
);
2615 EXPORT_SYMBOL(kblockd_schedule_delayed_work
);
2617 #define PLUG_MAGIC 0x91827364
2619 void blk_start_plug(struct blk_plug
*plug
)
2621 struct task_struct
*tsk
= current
;
2623 plug
->magic
= PLUG_MAGIC
;
2624 INIT_LIST_HEAD(&plug
->list
);
2625 INIT_LIST_HEAD(&plug
->cb_list
);
2626 plug
->should_sort
= 0;
2629 * If this is a nested plug, don't actually assign it. It will be
2630 * flushed on its own.
2634 * Store ordering should not be needed here, since a potential
2635 * preempt will imply a full memory barrier
2640 EXPORT_SYMBOL(blk_start_plug
);
2642 static int plug_rq_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
2644 struct request
*rqa
= container_of(a
, struct request
, queuelist
);
2645 struct request
*rqb
= container_of(b
, struct request
, queuelist
);
2647 return !(rqa
->q
<= rqb
->q
);
2651 * If 'from_schedule' is true, then postpone the dispatch of requests
2652 * until a safe kblockd context. We due this to avoid accidental big
2653 * additional stack usage in driver dispatch, in places where the originally
2654 * plugger did not intend it.
2656 static void queue_unplugged(struct request_queue
*q
, unsigned int depth
,
2658 __releases(q
->queue_lock
)
2660 trace_block_unplug(q
, depth
, !from_schedule
);
2663 * If we are punting this to kblockd, then we can safely drop
2664 * the queue_lock before waking kblockd (which needs to take
2667 if (from_schedule
) {
2668 spin_unlock(q
->queue_lock
);
2669 blk_run_queue_async(q
);
2672 spin_unlock(q
->queue_lock
);
2677 static void flush_plug_callbacks(struct blk_plug
*plug
)
2679 LIST_HEAD(callbacks
);
2681 if (list_empty(&plug
->cb_list
))
2684 list_splice_init(&plug
->cb_list
, &callbacks
);
2686 while (!list_empty(&callbacks
)) {
2687 struct blk_plug_cb
*cb
= list_first_entry(&callbacks
,
2690 list_del(&cb
->list
);
2695 void blk_flush_plug_list(struct blk_plug
*plug
, bool from_schedule
)
2697 struct request_queue
*q
;
2698 unsigned long flags
;
2703 BUG_ON(plug
->magic
!= PLUG_MAGIC
);
2705 flush_plug_callbacks(plug
);
2706 if (list_empty(&plug
->list
))
2709 list_splice_init(&plug
->list
, &list
);
2711 if (plug
->should_sort
) {
2712 list_sort(NULL
, &list
, plug_rq_cmp
);
2713 plug
->should_sort
= 0;
2720 * Save and disable interrupts here, to avoid doing it for every
2721 * queue lock we have to take.
2723 local_irq_save(flags
);
2724 while (!list_empty(&list
)) {
2725 rq
= list_entry_rq(list
.next
);
2726 list_del_init(&rq
->queuelist
);
2730 * This drops the queue lock
2733 queue_unplugged(q
, depth
, from_schedule
);
2736 spin_lock(q
->queue_lock
);
2739 * rq is already accounted, so use raw insert
2741 if (rq
->cmd_flags
& (REQ_FLUSH
| REQ_FUA
))
2742 __elv_add_request(q
, rq
, ELEVATOR_INSERT_FLUSH
);
2744 __elv_add_request(q
, rq
, ELEVATOR_INSERT_SORT_MERGE
);
2750 * This drops the queue lock
2753 queue_unplugged(q
, depth
, from_schedule
);
2755 local_irq_restore(flags
);
2758 void blk_finish_plug(struct blk_plug
*plug
)
2760 blk_flush_plug_list(plug
, false);
2762 if (plug
== current
->plug
)
2763 current
->plug
= NULL
;
2765 EXPORT_SYMBOL(blk_finish_plug
);
2767 int __init
blk_dev_init(void)
2769 BUILD_BUG_ON(__REQ_NR_BITS
> 8 *
2770 sizeof(((struct request
*)0)->cmd_flags
));
2772 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
2773 kblockd_workqueue
= alloc_workqueue("kblockd",
2774 WQ_MEM_RECLAIM
| WQ_HIGHPRI
, 0);
2775 if (!kblockd_workqueue
)
2776 panic("Failed to create kblockd\n");
2778 request_cachep
= kmem_cache_create("blkdev_requests",
2779 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
2781 blk_requestq_cachep
= kmem_cache_create("blkdev_queue",
2782 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);