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>
31 #define CREATE_TRACE_POINTS
32 #include <trace/events/block.h>
36 EXPORT_TRACEPOINT_SYMBOL_GPL(block_remap
);
37 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap
);
38 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete
);
40 static int __make_request(struct request_queue
*q
, struct bio
*bio
);
43 * For the allocated request tables
45 static struct kmem_cache
*request_cachep
;
48 * For queue allocation
50 struct kmem_cache
*blk_requestq_cachep
;
53 * Controlling structure to kblockd
55 static struct workqueue_struct
*kblockd_workqueue
;
57 static void drive_stat_acct(struct request
*rq
, int new_io
)
59 struct hd_struct
*part
;
60 int rw
= rq_data_dir(rq
);
63 if (!blk_do_io_stat(rq
))
66 cpu
= part_stat_lock();
67 part
= disk_map_sector_rcu(rq
->rq_disk
, blk_rq_pos(rq
));
70 part_stat_inc(cpu
, part
, merges
[rw
]);
72 part_round_stats(cpu
, part
);
73 part_inc_in_flight(part
, rw
);
79 void blk_queue_congestion_threshold(struct request_queue
*q
)
83 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) + 1;
84 if (nr
> q
->nr_requests
)
86 q
->nr_congestion_on
= nr
;
88 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) - (q
->nr_requests
/ 16) - 1;
91 q
->nr_congestion_off
= nr
;
95 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
98 * Locates the passed device's request queue and returns the address of its
101 * Will return NULL if the request queue cannot be located.
103 struct backing_dev_info
*blk_get_backing_dev_info(struct block_device
*bdev
)
105 struct backing_dev_info
*ret
= NULL
;
106 struct request_queue
*q
= bdev_get_queue(bdev
);
109 ret
= &q
->backing_dev_info
;
112 EXPORT_SYMBOL(blk_get_backing_dev_info
);
114 void blk_rq_init(struct request_queue
*q
, struct request
*rq
)
116 memset(rq
, 0, sizeof(*rq
));
118 INIT_LIST_HEAD(&rq
->queuelist
);
119 INIT_LIST_HEAD(&rq
->timeout_list
);
122 rq
->__sector
= (sector_t
) -1;
123 INIT_HLIST_NODE(&rq
->hash
);
124 RB_CLEAR_NODE(&rq
->rb_node
);
126 rq
->cmd_len
= BLK_MAX_CDB
;
129 rq
->start_time
= jiffies
;
130 set_start_time_ns(rq
);
132 EXPORT_SYMBOL(blk_rq_init
);
134 static void req_bio_endio(struct request
*rq
, struct bio
*bio
,
135 unsigned int nbytes
, int error
)
137 struct request_queue
*q
= rq
->q
;
139 if (&q
->flush_rq
!= rq
) {
141 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
142 else if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
145 if (unlikely(nbytes
> bio
->bi_size
)) {
146 printk(KERN_ERR
"%s: want %u bytes done, %u left\n",
147 __func__
, nbytes
, bio
->bi_size
);
148 nbytes
= bio
->bi_size
;
151 if (unlikely(rq
->cmd_flags
& REQ_QUIET
))
152 set_bit(BIO_QUIET
, &bio
->bi_flags
);
154 bio
->bi_size
-= nbytes
;
155 bio
->bi_sector
+= (nbytes
>> 9);
157 if (bio_integrity(bio
))
158 bio_integrity_advance(bio
, nbytes
);
160 if (bio
->bi_size
== 0)
161 bio_endio(bio
, error
);
164 * Okay, this is the sequenced flush request in
165 * progress, just record the error;
167 if (error
&& !q
->flush_err
)
168 q
->flush_err
= error
;
172 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
176 printk(KERN_INFO
"%s: dev %s: type=%x, flags=%x\n", msg
,
177 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->cmd_type
,
180 printk(KERN_INFO
" sector %llu, nr/cnr %u/%u\n",
181 (unsigned long long)blk_rq_pos(rq
),
182 blk_rq_sectors(rq
), blk_rq_cur_sectors(rq
));
183 printk(KERN_INFO
" bio %p, biotail %p, buffer %p, len %u\n",
184 rq
->bio
, rq
->biotail
, rq
->buffer
, blk_rq_bytes(rq
));
186 if (rq
->cmd_type
== REQ_TYPE_BLOCK_PC
) {
187 printk(KERN_INFO
" cdb: ");
188 for (bit
= 0; bit
< BLK_MAX_CDB
; bit
++)
189 printk("%02x ", rq
->cmd
[bit
]);
193 EXPORT_SYMBOL(blk_dump_rq_flags
);
196 * "plug" the device if there are no outstanding requests: this will
197 * force the transfer to start only after we have put all the requests
200 * This is called with interrupts off and no requests on the queue and
201 * with the queue lock held.
203 void blk_plug_device(struct request_queue
*q
)
205 WARN_ON(!irqs_disabled());
208 * don't plug a stopped queue, it must be paired with blk_start_queue()
209 * which will restart the queueing
211 if (blk_queue_stopped(q
))
214 if (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED
, q
)) {
215 mod_timer(&q
->unplug_timer
, jiffies
+ q
->unplug_delay
);
219 EXPORT_SYMBOL(blk_plug_device
);
222 * blk_plug_device_unlocked - plug a device without queue lock held
223 * @q: The &struct request_queue to plug
226 * Like @blk_plug_device(), but grabs the queue lock and disables
229 void blk_plug_device_unlocked(struct request_queue
*q
)
233 spin_lock_irqsave(q
->queue_lock
, flags
);
235 spin_unlock_irqrestore(q
->queue_lock
, flags
);
237 EXPORT_SYMBOL(blk_plug_device_unlocked
);
240 * remove the queue from the plugged list, if present. called with
241 * queue lock held and interrupts disabled.
243 int blk_remove_plug(struct request_queue
*q
)
245 WARN_ON(!irqs_disabled());
247 if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED
, q
))
250 del_timer(&q
->unplug_timer
);
253 EXPORT_SYMBOL(blk_remove_plug
);
256 * remove the plug and let it rip..
258 void __generic_unplug_device(struct request_queue
*q
)
260 if (unlikely(blk_queue_stopped(q
)))
262 if (!blk_remove_plug(q
) && !blk_queue_nonrot(q
))
269 * generic_unplug_device - fire a request queue
270 * @q: The &struct request_queue in question
273 * Linux uses plugging to build bigger requests queues before letting
274 * the device have at them. If a queue is plugged, the I/O scheduler
275 * is still adding and merging requests on the queue. Once the queue
276 * gets unplugged, the request_fn defined for the queue is invoked and
279 void generic_unplug_device(struct request_queue
*q
)
281 if (blk_queue_plugged(q
)) {
282 spin_lock_irq(q
->queue_lock
);
283 __generic_unplug_device(q
);
284 spin_unlock_irq(q
->queue_lock
);
287 EXPORT_SYMBOL(generic_unplug_device
);
289 static void blk_backing_dev_unplug(struct backing_dev_info
*bdi
,
292 struct request_queue
*q
= bdi
->unplug_io_data
;
297 void blk_unplug_work(struct work_struct
*work
)
299 struct request_queue
*q
=
300 container_of(work
, struct request_queue
, unplug_work
);
302 trace_block_unplug_io(q
);
306 void blk_unplug_timeout(unsigned long data
)
308 struct request_queue
*q
= (struct request_queue
*)data
;
310 trace_block_unplug_timer(q
);
311 kblockd_schedule_work(q
, &q
->unplug_work
);
314 void blk_unplug(struct request_queue
*q
)
317 * devices don't necessarily have an ->unplug_fn defined
320 trace_block_unplug_io(q
);
324 EXPORT_SYMBOL(blk_unplug
);
327 * blk_start_queue - restart a previously stopped queue
328 * @q: The &struct request_queue in question
331 * blk_start_queue() will clear the stop flag on the queue, and call
332 * the request_fn for the queue if it was in a stopped state when
333 * entered. Also see blk_stop_queue(). Queue lock must be held.
335 void blk_start_queue(struct request_queue
*q
)
337 WARN_ON(!irqs_disabled());
339 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
342 EXPORT_SYMBOL(blk_start_queue
);
345 * blk_stop_queue - stop a queue
346 * @q: The &struct request_queue in question
349 * The Linux block layer assumes that a block driver will consume all
350 * entries on the request queue when the request_fn strategy is called.
351 * Often this will not happen, because of hardware limitations (queue
352 * depth settings). If a device driver gets a 'queue full' response,
353 * or if it simply chooses not to queue more I/O at one point, it can
354 * call this function to prevent the request_fn from being called until
355 * the driver has signalled it's ready to go again. This happens by calling
356 * blk_start_queue() to restart queue operations. Queue lock must be held.
358 void blk_stop_queue(struct request_queue
*q
)
361 queue_flag_set(QUEUE_FLAG_STOPPED
, q
);
363 EXPORT_SYMBOL(blk_stop_queue
);
366 * blk_sync_queue - cancel any pending callbacks on a queue
370 * The block layer may perform asynchronous callback activity
371 * on a queue, such as calling the unplug function after a timeout.
372 * A block device may call blk_sync_queue to ensure that any
373 * such activity is cancelled, thus allowing it to release resources
374 * that the callbacks might use. The caller must already have made sure
375 * that its ->make_request_fn will not re-add plugging prior to calling
379 void blk_sync_queue(struct request_queue
*q
)
381 del_timer_sync(&q
->unplug_timer
);
382 del_timer_sync(&q
->timeout
);
383 cancel_work_sync(&q
->unplug_work
);
385 EXPORT_SYMBOL(blk_sync_queue
);
388 * __blk_run_queue - run a single device queue
389 * @q: The queue to run
392 * See @blk_run_queue. This variant must be called with the queue lock
393 * held and interrupts disabled.
396 void __blk_run_queue(struct request_queue
*q
)
400 if (unlikely(blk_queue_stopped(q
)))
403 if (elv_queue_empty(q
))
407 * Only recurse once to avoid overrunning the stack, let the unplug
408 * handling reinvoke the handler shortly if we already got there.
410 if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER
, q
)) {
412 queue_flag_clear(QUEUE_FLAG_REENTER
, q
);
414 queue_flag_set(QUEUE_FLAG_PLUGGED
, q
);
415 kblockd_schedule_work(q
, &q
->unplug_work
);
418 EXPORT_SYMBOL(__blk_run_queue
);
421 * blk_run_queue - run a single device queue
422 * @q: The queue to run
425 * Invoke request handling on this queue, if it has pending work to do.
426 * May be used to restart queueing when a request has completed.
428 void blk_run_queue(struct request_queue
*q
)
432 spin_lock_irqsave(q
->queue_lock
, flags
);
434 spin_unlock_irqrestore(q
->queue_lock
, flags
);
436 EXPORT_SYMBOL(blk_run_queue
);
438 void blk_put_queue(struct request_queue
*q
)
440 kobject_put(&q
->kobj
);
443 void blk_cleanup_queue(struct request_queue
*q
)
446 * We know we have process context here, so we can be a little
447 * cautious and ensure that pending block actions on this device
448 * are done before moving on. Going into this function, we should
449 * not have processes doing IO to this device.
453 del_timer_sync(&q
->backing_dev_info
.laptop_mode_wb_timer
);
454 mutex_lock(&q
->sysfs_lock
);
455 queue_flag_set_unlocked(QUEUE_FLAG_DEAD
, q
);
456 mutex_unlock(&q
->sysfs_lock
);
459 elevator_exit(q
->elevator
);
463 EXPORT_SYMBOL(blk_cleanup_queue
);
465 static int blk_init_free_list(struct request_queue
*q
)
467 struct request_list
*rl
= &q
->rq
;
469 if (unlikely(rl
->rq_pool
))
472 rl
->count
[BLK_RW_SYNC
] = rl
->count
[BLK_RW_ASYNC
] = 0;
473 rl
->starved
[BLK_RW_SYNC
] = rl
->starved
[BLK_RW_ASYNC
] = 0;
475 init_waitqueue_head(&rl
->wait
[BLK_RW_SYNC
]);
476 init_waitqueue_head(&rl
->wait
[BLK_RW_ASYNC
]);
478 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, mempool_alloc_slab
,
479 mempool_free_slab
, request_cachep
, q
->node
);
487 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
489 return blk_alloc_queue_node(gfp_mask
, -1);
491 EXPORT_SYMBOL(blk_alloc_queue
);
493 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
495 struct request_queue
*q
;
498 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
499 gfp_mask
| __GFP_ZERO
, node_id
);
503 q
->backing_dev_info
.unplug_io_fn
= blk_backing_dev_unplug
;
504 q
->backing_dev_info
.unplug_io_data
= q
;
505 q
->backing_dev_info
.ra_pages
=
506 (VM_MAX_READAHEAD
* 1024) / PAGE_CACHE_SIZE
;
507 q
->backing_dev_info
.state
= 0;
508 q
->backing_dev_info
.capabilities
= BDI_CAP_MAP_COPY
;
509 q
->backing_dev_info
.name
= "block";
511 err
= bdi_init(&q
->backing_dev_info
);
513 kmem_cache_free(blk_requestq_cachep
, q
);
517 setup_timer(&q
->backing_dev_info
.laptop_mode_wb_timer
,
518 laptop_mode_timer_fn
, (unsigned long) q
);
519 init_timer(&q
->unplug_timer
);
520 setup_timer(&q
->timeout
, blk_rq_timed_out_timer
, (unsigned long) q
);
521 INIT_LIST_HEAD(&q
->timeout_list
);
522 INIT_LIST_HEAD(&q
->pending_flushes
);
523 INIT_WORK(&q
->unplug_work
, blk_unplug_work
);
525 kobject_init(&q
->kobj
, &blk_queue_ktype
);
527 mutex_init(&q
->sysfs_lock
);
528 spin_lock_init(&q
->__queue_lock
);
532 EXPORT_SYMBOL(blk_alloc_queue_node
);
535 * blk_init_queue - prepare a request queue for use with a block device
536 * @rfn: The function to be called to process requests that have been
537 * placed on the queue.
538 * @lock: Request queue spin lock
541 * If a block device wishes to use the standard request handling procedures,
542 * which sorts requests and coalesces adjacent requests, then it must
543 * call blk_init_queue(). The function @rfn will be called when there
544 * are requests on the queue that need to be processed. If the device
545 * supports plugging, then @rfn may not be called immediately when requests
546 * are available on the queue, but may be called at some time later instead.
547 * Plugged queues are generally unplugged when a buffer belonging to one
548 * of the requests on the queue is needed, or due to memory pressure.
550 * @rfn is not required, or even expected, to remove all requests off the
551 * queue, but only as many as it can handle at a time. If it does leave
552 * requests on the queue, it is responsible for arranging that the requests
553 * get dealt with eventually.
555 * The queue spin lock must be held while manipulating the requests on the
556 * request queue; this lock will be taken also from interrupt context, so irq
557 * disabling is needed for it.
559 * Function returns a pointer to the initialized request queue, or %NULL if
563 * blk_init_queue() must be paired with a blk_cleanup_queue() call
564 * when the block device is deactivated (such as at module unload).
567 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
569 return blk_init_queue_node(rfn
, lock
, -1);
571 EXPORT_SYMBOL(blk_init_queue
);
573 struct request_queue
*
574 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
576 struct request_queue
*uninit_q
, *q
;
578 uninit_q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
582 q
= blk_init_allocated_queue_node(uninit_q
, rfn
, lock
, node_id
);
584 blk_cleanup_queue(uninit_q
);
588 EXPORT_SYMBOL(blk_init_queue_node
);
590 struct request_queue
*
591 blk_init_allocated_queue(struct request_queue
*q
, request_fn_proc
*rfn
,
594 return blk_init_allocated_queue_node(q
, rfn
, lock
, -1);
596 EXPORT_SYMBOL(blk_init_allocated_queue
);
598 struct request_queue
*
599 blk_init_allocated_queue_node(struct request_queue
*q
, request_fn_proc
*rfn
,
600 spinlock_t
*lock
, int node_id
)
606 if (blk_init_free_list(q
))
610 q
->prep_rq_fn
= NULL
;
611 q
->unprep_rq_fn
= NULL
;
612 q
->unplug_fn
= generic_unplug_device
;
613 q
->queue_flags
= QUEUE_FLAG_DEFAULT
;
614 q
->queue_lock
= lock
;
617 * This also sets hw/phys segments, boundary and size
619 blk_queue_make_request(q
, __make_request
);
621 q
->sg_reserved_size
= INT_MAX
;
626 if (!elevator_init(q
, NULL
)) {
627 blk_queue_congestion_threshold(q
);
633 EXPORT_SYMBOL(blk_init_allocated_queue_node
);
635 int blk_get_queue(struct request_queue
*q
)
637 if (likely(!test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
))) {
638 kobject_get(&q
->kobj
);
645 static inline void blk_free_request(struct request_queue
*q
, struct request
*rq
)
647 if (rq
->cmd_flags
& REQ_ELVPRIV
)
648 elv_put_request(q
, rq
);
649 mempool_free(rq
, q
->rq
.rq_pool
);
652 static struct request
*
653 blk_alloc_request(struct request_queue
*q
, int flags
, int priv
, gfp_t gfp_mask
)
655 struct request
*rq
= mempool_alloc(q
->rq
.rq_pool
, gfp_mask
);
662 rq
->cmd_flags
= flags
| REQ_ALLOCED
;
665 if (unlikely(elv_set_request(q
, rq
, gfp_mask
))) {
666 mempool_free(rq
, q
->rq
.rq_pool
);
669 rq
->cmd_flags
|= REQ_ELVPRIV
;
676 * ioc_batching returns true if the ioc is a valid batching request and
677 * should be given priority access to a request.
679 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
685 * Make sure the process is able to allocate at least 1 request
686 * even if the batch times out, otherwise we could theoretically
689 return ioc
->nr_batch_requests
== q
->nr_batching
||
690 (ioc
->nr_batch_requests
> 0
691 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
695 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
696 * will cause the process to be a "batcher" on all queues in the system. This
697 * is the behaviour we want though - once it gets a wakeup it should be given
700 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
702 if (!ioc
|| ioc_batching(q
, ioc
))
705 ioc
->nr_batch_requests
= q
->nr_batching
;
706 ioc
->last_waited
= jiffies
;
709 static void __freed_request(struct request_queue
*q
, int sync
)
711 struct request_list
*rl
= &q
->rq
;
713 if (rl
->count
[sync
] < queue_congestion_off_threshold(q
))
714 blk_clear_queue_congested(q
, sync
);
716 if (rl
->count
[sync
] + 1 <= q
->nr_requests
) {
717 if (waitqueue_active(&rl
->wait
[sync
]))
718 wake_up(&rl
->wait
[sync
]);
720 blk_clear_queue_full(q
, sync
);
725 * A request has just been released. Account for it, update the full and
726 * congestion status, wake up any waiters. Called under q->queue_lock.
728 static void freed_request(struct request_queue
*q
, int sync
, int priv
)
730 struct request_list
*rl
= &q
->rq
;
736 __freed_request(q
, sync
);
738 if (unlikely(rl
->starved
[sync
^ 1]))
739 __freed_request(q
, sync
^ 1);
743 * Get a free request, queue_lock must be held.
744 * Returns NULL on failure, with queue_lock held.
745 * Returns !NULL on success, with queue_lock *not held*.
747 static struct request
*get_request(struct request_queue
*q
, int rw_flags
,
748 struct bio
*bio
, gfp_t gfp_mask
)
750 struct request
*rq
= NULL
;
751 struct request_list
*rl
= &q
->rq
;
752 struct io_context
*ioc
= NULL
;
753 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
756 may_queue
= elv_may_queue(q
, rw_flags
);
757 if (may_queue
== ELV_MQUEUE_NO
)
760 if (rl
->count
[is_sync
]+1 >= queue_congestion_on_threshold(q
)) {
761 if (rl
->count
[is_sync
]+1 >= q
->nr_requests
) {
762 ioc
= current_io_context(GFP_ATOMIC
, q
->node
);
764 * The queue will fill after this allocation, so set
765 * it as full, and mark this process as "batching".
766 * This process will be allowed to complete a batch of
767 * requests, others will be blocked.
769 if (!blk_queue_full(q
, is_sync
)) {
770 ioc_set_batching(q
, ioc
);
771 blk_set_queue_full(q
, is_sync
);
773 if (may_queue
!= ELV_MQUEUE_MUST
774 && !ioc_batching(q
, ioc
)) {
776 * The queue is full and the allocating
777 * process is not a "batcher", and not
778 * exempted by the IO scheduler
784 blk_set_queue_congested(q
, is_sync
);
788 * Only allow batching queuers to allocate up to 50% over the defined
789 * limit of requests, otherwise we could have thousands of requests
790 * allocated with any setting of ->nr_requests
792 if (rl
->count
[is_sync
] >= (3 * q
->nr_requests
/ 2))
795 rl
->count
[is_sync
]++;
796 rl
->starved
[is_sync
] = 0;
798 priv
= !test_bit(QUEUE_FLAG_ELVSWITCH
, &q
->queue_flags
);
802 if (blk_queue_io_stat(q
))
803 rw_flags
|= REQ_IO_STAT
;
804 spin_unlock_irq(q
->queue_lock
);
806 rq
= blk_alloc_request(q
, rw_flags
, priv
, gfp_mask
);
809 * Allocation failed presumably due to memory. Undo anything
810 * we might have messed up.
812 * Allocating task should really be put onto the front of the
813 * wait queue, but this is pretty rare.
815 spin_lock_irq(q
->queue_lock
);
816 freed_request(q
, is_sync
, priv
);
819 * in the very unlikely event that allocation failed and no
820 * requests for this direction was pending, mark us starved
821 * so that freeing of a request in the other direction will
822 * notice us. another possible fix would be to split the
823 * rq mempool into READ and WRITE
826 if (unlikely(rl
->count
[is_sync
] == 0))
827 rl
->starved
[is_sync
] = 1;
833 * ioc may be NULL here, and ioc_batching will be false. That's
834 * OK, if the queue is under the request limit then requests need
835 * not count toward the nr_batch_requests limit. There will always
836 * be some limit enforced by BLK_BATCH_TIME.
838 if (ioc_batching(q
, ioc
))
839 ioc
->nr_batch_requests
--;
841 trace_block_getrq(q
, bio
, rw_flags
& 1);
847 * No available requests for this queue, unplug the device and wait for some
848 * requests to become available.
850 * Called with q->queue_lock held, and returns with it unlocked.
852 static struct request
*get_request_wait(struct request_queue
*q
, int rw_flags
,
855 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
858 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
861 struct io_context
*ioc
;
862 struct request_list
*rl
= &q
->rq
;
864 prepare_to_wait_exclusive(&rl
->wait
[is_sync
], &wait
,
865 TASK_UNINTERRUPTIBLE
);
867 trace_block_sleeprq(q
, bio
, rw_flags
& 1);
869 __generic_unplug_device(q
);
870 spin_unlock_irq(q
->queue_lock
);
874 * After sleeping, we become a "batching" process and
875 * will be able to allocate at least one request, and
876 * up to a big batch of them for a small period time.
877 * See ioc_batching, ioc_set_batching
879 ioc
= current_io_context(GFP_NOIO
, q
->node
);
880 ioc_set_batching(q
, ioc
);
882 spin_lock_irq(q
->queue_lock
);
883 finish_wait(&rl
->wait
[is_sync
], &wait
);
885 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
891 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
895 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
897 spin_lock_irq(q
->queue_lock
);
898 if (gfp_mask
& __GFP_WAIT
) {
899 rq
= get_request_wait(q
, rw
, NULL
);
901 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
903 spin_unlock_irq(q
->queue_lock
);
905 /* q->queue_lock is unlocked at this point */
909 EXPORT_SYMBOL(blk_get_request
);
912 * blk_make_request - given a bio, allocate a corresponding struct request.
913 * @q: target request queue
914 * @bio: The bio describing the memory mappings that will be submitted for IO.
915 * It may be a chained-bio properly constructed by block/bio layer.
916 * @gfp_mask: gfp flags to be used for memory allocation
918 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
919 * type commands. Where the struct request needs to be farther initialized by
920 * the caller. It is passed a &struct bio, which describes the memory info of
923 * The caller of blk_make_request must make sure that bi_io_vec
924 * are set to describe the memory buffers. That bio_data_dir() will return
925 * the needed direction of the request. (And all bio's in the passed bio-chain
926 * are properly set accordingly)
928 * If called under none-sleepable conditions, mapped bio buffers must not
929 * need bouncing, by calling the appropriate masked or flagged allocator,
930 * suitable for the target device. Otherwise the call to blk_queue_bounce will
933 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
934 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
935 * anything but the first bio in the chain. Otherwise you risk waiting for IO
936 * completion of a bio that hasn't been submitted yet, thus resulting in a
937 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
938 * of bio_alloc(), as that avoids the mempool deadlock.
939 * If possible a big IO should be split into smaller parts when allocation
940 * fails. Partial allocation should not be an error, or you risk a live-lock.
942 struct request
*blk_make_request(struct request_queue
*q
, struct bio
*bio
,
945 struct request
*rq
= blk_get_request(q
, bio_data_dir(bio
), gfp_mask
);
948 return ERR_PTR(-ENOMEM
);
951 struct bio
*bounce_bio
= bio
;
954 blk_queue_bounce(q
, &bounce_bio
);
955 ret
= blk_rq_append_bio(q
, rq
, bounce_bio
);
964 EXPORT_SYMBOL(blk_make_request
);
967 * blk_requeue_request - put a request back on queue
968 * @q: request queue where request should be inserted
969 * @rq: request to be inserted
972 * Drivers often keep queueing requests until the hardware cannot accept
973 * more, when that condition happens we need to put the request back
974 * on the queue. Must be called with queue lock held.
976 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
978 blk_delete_timer(rq
);
979 blk_clear_rq_complete(rq
);
980 trace_block_rq_requeue(q
, rq
);
982 if (blk_rq_tagged(rq
))
983 blk_queue_end_tag(q
, rq
);
985 BUG_ON(blk_queued_rq(rq
));
987 elv_requeue_request(q
, rq
);
989 EXPORT_SYMBOL(blk_requeue_request
);
992 * blk_insert_request - insert a special request into a request queue
993 * @q: request queue where request should be inserted
994 * @rq: request to be inserted
995 * @at_head: insert request at head or tail of queue
996 * @data: private data
999 * Many block devices need to execute commands asynchronously, so they don't
1000 * block the whole kernel from preemption during request execution. This is
1001 * accomplished normally by inserting aritficial requests tagged as
1002 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
1003 * be scheduled for actual execution by the request queue.
1005 * We have the option of inserting the head or the tail of the queue.
1006 * Typically we use the tail for new ioctls and so forth. We use the head
1007 * of the queue for things like a QUEUE_FULL message from a device, or a
1008 * host that is unable to accept a particular command.
1010 void blk_insert_request(struct request_queue
*q
, struct request
*rq
,
1011 int at_head
, void *data
)
1013 int where
= at_head
? ELEVATOR_INSERT_FRONT
: ELEVATOR_INSERT_BACK
;
1014 unsigned long flags
;
1017 * tell I/O scheduler that this isn't a regular read/write (ie it
1018 * must not attempt merges on this) and that it acts as a soft
1021 rq
->cmd_type
= REQ_TYPE_SPECIAL
;
1025 spin_lock_irqsave(q
->queue_lock
, flags
);
1028 * If command is tagged, release the tag
1030 if (blk_rq_tagged(rq
))
1031 blk_queue_end_tag(q
, rq
);
1033 drive_stat_acct(rq
, 1);
1034 __elv_add_request(q
, rq
, where
, 0);
1036 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1038 EXPORT_SYMBOL(blk_insert_request
);
1040 static void part_round_stats_single(int cpu
, struct hd_struct
*part
,
1043 if (now
== part
->stamp
)
1046 if (part_in_flight(part
)) {
1047 __part_stat_add(cpu
, part
, time_in_queue
,
1048 part_in_flight(part
) * (now
- part
->stamp
));
1049 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
1055 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1056 * @cpu: cpu number for stats access
1057 * @part: target partition
1059 * The average IO queue length and utilisation statistics are maintained
1060 * by observing the current state of the queue length and the amount of
1061 * time it has been in this state for.
1063 * Normally, that accounting is done on IO completion, but that can result
1064 * in more than a second's worth of IO being accounted for within any one
1065 * second, leading to >100% utilisation. To deal with that, we call this
1066 * function to do a round-off before returning the results when reading
1067 * /proc/diskstats. This accounts immediately for all queue usage up to
1068 * the current jiffies and restarts the counters again.
1070 void part_round_stats(int cpu
, struct hd_struct
*part
)
1072 unsigned long now
= jiffies
;
1075 part_round_stats_single(cpu
, &part_to_disk(part
)->part0
, now
);
1076 part_round_stats_single(cpu
, part
, now
);
1078 EXPORT_SYMBOL_GPL(part_round_stats
);
1081 * queue lock must be held
1083 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1087 if (unlikely(--req
->ref_count
))
1090 elv_completed_request(q
, req
);
1092 /* this is a bio leak */
1093 WARN_ON(req
->bio
!= NULL
);
1096 * Request may not have originated from ll_rw_blk. if not,
1097 * it didn't come out of our reserved rq pools
1099 if (req
->cmd_flags
& REQ_ALLOCED
) {
1100 int is_sync
= rq_is_sync(req
) != 0;
1101 int priv
= req
->cmd_flags
& REQ_ELVPRIV
;
1103 BUG_ON(!list_empty(&req
->queuelist
));
1104 BUG_ON(!hlist_unhashed(&req
->hash
));
1106 blk_free_request(q
, req
);
1107 freed_request(q
, is_sync
, priv
);
1110 EXPORT_SYMBOL_GPL(__blk_put_request
);
1112 void blk_put_request(struct request
*req
)
1114 unsigned long flags
;
1115 struct request_queue
*q
= req
->q
;
1117 spin_lock_irqsave(q
->queue_lock
, flags
);
1118 __blk_put_request(q
, req
);
1119 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1121 EXPORT_SYMBOL(blk_put_request
);
1124 * blk_add_request_payload - add a payload to a request
1125 * @rq: request to update
1126 * @page: page backing the payload
1127 * @len: length of the payload.
1129 * This allows to later add a payload to an already submitted request by
1130 * a block driver. The driver needs to take care of freeing the payload
1133 * Note that this is a quite horrible hack and nothing but handling of
1134 * discard requests should ever use it.
1136 void blk_add_request_payload(struct request
*rq
, struct page
*page
,
1139 struct bio
*bio
= rq
->bio
;
1141 bio
->bi_io_vec
->bv_page
= page
;
1142 bio
->bi_io_vec
->bv_offset
= 0;
1143 bio
->bi_io_vec
->bv_len
= len
;
1147 bio
->bi_phys_segments
= 1;
1149 rq
->__data_len
= rq
->resid_len
= len
;
1150 rq
->nr_phys_segments
= 1;
1151 rq
->buffer
= bio_data(bio
);
1153 EXPORT_SYMBOL_GPL(blk_add_request_payload
);
1155 void init_request_from_bio(struct request
*req
, struct bio
*bio
)
1157 req
->cpu
= bio
->bi_comp_cpu
;
1158 req
->cmd_type
= REQ_TYPE_FS
;
1160 req
->cmd_flags
|= bio
->bi_rw
& REQ_COMMON_MASK
;
1161 if (bio
->bi_rw
& REQ_RAHEAD
)
1162 req
->cmd_flags
|= REQ_FAILFAST_MASK
;
1165 req
->__sector
= bio
->bi_sector
;
1166 req
->ioprio
= bio_prio(bio
);
1167 blk_rq_bio_prep(req
->q
, req
, bio
);
1171 * Only disabling plugging for non-rotational devices if it does tagging
1172 * as well, otherwise we do need the proper merging
1174 static inline bool queue_should_plug(struct request_queue
*q
)
1176 return !(blk_queue_nonrot(q
) && blk_queue_tagged(q
));
1179 static int __make_request(struct request_queue
*q
, struct bio
*bio
)
1181 struct request
*req
;
1183 unsigned int bytes
= bio
->bi_size
;
1184 const unsigned short prio
= bio_prio(bio
);
1185 const bool sync
= (bio
->bi_rw
& REQ_SYNC
);
1186 const bool unplug
= (bio
->bi_rw
& REQ_UNPLUG
);
1187 const unsigned int ff
= bio
->bi_rw
& REQ_FAILFAST_MASK
;
1188 int where
= ELEVATOR_INSERT_SORT
;
1191 /* REQ_HARDBARRIER is no more */
1192 if (WARN_ONCE(bio
->bi_rw
& REQ_HARDBARRIER
,
1193 "block: HARDBARRIER is deprecated, use FLUSH/FUA instead\n")) {
1194 bio_endio(bio
, -EOPNOTSUPP
);
1199 * low level driver can indicate that it wants pages above a
1200 * certain limit bounced to low memory (ie for highmem, or even
1201 * ISA dma in theory)
1203 blk_queue_bounce(q
, &bio
);
1205 spin_lock_irq(q
->queue_lock
);
1207 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) {
1208 where
= ELEVATOR_INSERT_FRONT
;
1212 if (elv_queue_empty(q
))
1215 el_ret
= elv_merge(q
, &req
, bio
);
1217 case ELEVATOR_BACK_MERGE
:
1218 BUG_ON(!rq_mergeable(req
));
1220 if (!ll_back_merge_fn(q
, req
, bio
))
1223 trace_block_bio_backmerge(q
, bio
);
1225 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1226 blk_rq_set_mixed_merge(req
);
1228 req
->biotail
->bi_next
= bio
;
1230 req
->__data_len
+= bytes
;
1231 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
1232 if (!blk_rq_cpu_valid(req
))
1233 req
->cpu
= bio
->bi_comp_cpu
;
1234 drive_stat_acct(req
, 0);
1235 elv_bio_merged(q
, req
, bio
);
1236 if (!attempt_back_merge(q
, req
))
1237 elv_merged_request(q
, req
, el_ret
);
1240 case ELEVATOR_FRONT_MERGE
:
1241 BUG_ON(!rq_mergeable(req
));
1243 if (!ll_front_merge_fn(q
, req
, bio
))
1246 trace_block_bio_frontmerge(q
, bio
);
1248 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
) {
1249 blk_rq_set_mixed_merge(req
);
1250 req
->cmd_flags
&= ~REQ_FAILFAST_MASK
;
1251 req
->cmd_flags
|= ff
;
1254 bio
->bi_next
= req
->bio
;
1258 * may not be valid. if the low level driver said
1259 * it didn't need a bounce buffer then it better
1260 * not touch req->buffer either...
1262 req
->buffer
= bio_data(bio
);
1263 req
->__sector
= bio
->bi_sector
;
1264 req
->__data_len
+= bytes
;
1265 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
1266 if (!blk_rq_cpu_valid(req
))
1267 req
->cpu
= bio
->bi_comp_cpu
;
1268 drive_stat_acct(req
, 0);
1269 elv_bio_merged(q
, req
, bio
);
1270 if (!attempt_front_merge(q
, req
))
1271 elv_merged_request(q
, req
, el_ret
);
1274 /* ELV_NO_MERGE: elevator says don't/can't merge. */
1281 * This sync check and mask will be re-done in init_request_from_bio(),
1282 * but we need to set it earlier to expose the sync flag to the
1283 * rq allocator and io schedulers.
1285 rw_flags
= bio_data_dir(bio
);
1287 rw_flags
|= REQ_SYNC
;
1290 * Grab a free request. This is might sleep but can not fail.
1291 * Returns with the queue unlocked.
1293 req
= get_request_wait(q
, rw_flags
, bio
);
1296 * After dropping the lock and possibly sleeping here, our request
1297 * may now be mergeable after it had proven unmergeable (above).
1298 * We don't worry about that case for efficiency. It won't happen
1299 * often, and the elevators are able to handle it.
1301 init_request_from_bio(req
, bio
);
1303 spin_lock_irq(q
->queue_lock
);
1304 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
) ||
1305 bio_flagged(bio
, BIO_CPU_AFFINE
))
1306 req
->cpu
= blk_cpu_to_group(smp_processor_id());
1307 if (queue_should_plug(q
) && elv_queue_empty(q
))
1310 /* insert the request into the elevator */
1311 drive_stat_acct(req
, 1);
1312 __elv_add_request(q
, req
, where
, 0);
1314 if (unplug
|| !queue_should_plug(q
))
1315 __generic_unplug_device(q
);
1316 spin_unlock_irq(q
->queue_lock
);
1321 * If bio->bi_dev is a partition, remap the location
1323 static inline void blk_partition_remap(struct bio
*bio
)
1325 struct block_device
*bdev
= bio
->bi_bdev
;
1327 if (bio_sectors(bio
) && bdev
!= bdev
->bd_contains
) {
1328 struct hd_struct
*p
= bdev
->bd_part
;
1330 bio
->bi_sector
+= p
->start_sect
;
1331 bio
->bi_bdev
= bdev
->bd_contains
;
1333 trace_block_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
1335 bio
->bi_sector
- p
->start_sect
);
1339 static void handle_bad_sector(struct bio
*bio
)
1341 char b
[BDEVNAME_SIZE
];
1343 printk(KERN_INFO
"attempt to access beyond end of device\n");
1344 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
1345 bdevname(bio
->bi_bdev
, b
),
1347 (unsigned long long)bio
->bi_sector
+ bio_sectors(bio
),
1348 (long long)(bio
->bi_bdev
->bd_inode
->i_size
>> 9));
1350 set_bit(BIO_EOF
, &bio
->bi_flags
);
1353 #ifdef CONFIG_FAIL_MAKE_REQUEST
1355 static DECLARE_FAULT_ATTR(fail_make_request
);
1357 static int __init
setup_fail_make_request(char *str
)
1359 return setup_fault_attr(&fail_make_request
, str
);
1361 __setup("fail_make_request=", setup_fail_make_request
);
1363 static int should_fail_request(struct bio
*bio
)
1365 struct hd_struct
*part
= bio
->bi_bdev
->bd_part
;
1367 if (part_to_disk(part
)->part0
.make_it_fail
|| part
->make_it_fail
)
1368 return should_fail(&fail_make_request
, bio
->bi_size
);
1373 static int __init
fail_make_request_debugfs(void)
1375 return init_fault_attr_dentries(&fail_make_request
,
1376 "fail_make_request");
1379 late_initcall(fail_make_request_debugfs
);
1381 #else /* CONFIG_FAIL_MAKE_REQUEST */
1383 static inline int should_fail_request(struct bio
*bio
)
1388 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1391 * Check whether this bio extends beyond the end of the device.
1393 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
1400 /* Test device or partition size, when known. */
1401 maxsector
= bio
->bi_bdev
->bd_inode
->i_size
>> 9;
1403 sector_t sector
= bio
->bi_sector
;
1405 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
1407 * This may well happen - the kernel calls bread()
1408 * without checking the size of the device, e.g., when
1409 * mounting a device.
1411 handle_bad_sector(bio
);
1420 * generic_make_request - hand a buffer to its device driver for I/O
1421 * @bio: The bio describing the location in memory and on the device.
1423 * generic_make_request() is used to make I/O requests of block
1424 * devices. It is passed a &struct bio, which describes the I/O that needs
1427 * generic_make_request() does not return any status. The
1428 * success/failure status of the request, along with notification of
1429 * completion, is delivered asynchronously through the bio->bi_end_io
1430 * function described (one day) else where.
1432 * The caller of generic_make_request must make sure that bi_io_vec
1433 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1434 * set to describe the device address, and the
1435 * bi_end_io and optionally bi_private are set to describe how
1436 * completion notification should be signaled.
1438 * generic_make_request and the drivers it calls may use bi_next if this
1439 * bio happens to be merged with someone else, and may change bi_dev and
1440 * bi_sector for remaps as it sees fit. So the values of these fields
1441 * should NOT be depended on after the call to generic_make_request.
1443 static inline void __generic_make_request(struct bio
*bio
)
1445 struct request_queue
*q
;
1446 sector_t old_sector
;
1447 int ret
, nr_sectors
= bio_sectors(bio
);
1453 if (bio_check_eod(bio
, nr_sectors
))
1457 * Resolve the mapping until finished. (drivers are
1458 * still free to implement/resolve their own stacking
1459 * by explicitly returning 0)
1461 * NOTE: we don't repeat the blk_size check for each new device.
1462 * Stacking drivers are expected to know what they are doing.
1467 char b
[BDEVNAME_SIZE
];
1469 q
= bdev_get_queue(bio
->bi_bdev
);
1472 "generic_make_request: Trying to access "
1473 "nonexistent block-device %s (%Lu)\n",
1474 bdevname(bio
->bi_bdev
, b
),
1475 (long long) bio
->bi_sector
);
1479 if (unlikely(!(bio
->bi_rw
& REQ_DISCARD
) &&
1480 nr_sectors
> queue_max_hw_sectors(q
))) {
1481 printk(KERN_ERR
"bio too big device %s (%u > %u)\n",
1482 bdevname(bio
->bi_bdev
, b
),
1484 queue_max_hw_sectors(q
));
1488 if (unlikely(test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)))
1491 if (should_fail_request(bio
))
1495 * If this device has partitions, remap block n
1496 * of partition p to block n+start(p) of the disk.
1498 blk_partition_remap(bio
);
1500 if (bio_integrity_enabled(bio
) && bio_integrity_prep(bio
))
1503 if (old_sector
!= -1)
1504 trace_block_remap(q
, bio
, old_dev
, old_sector
);
1506 old_sector
= bio
->bi_sector
;
1507 old_dev
= bio
->bi_bdev
->bd_dev
;
1509 if (bio_check_eod(bio
, nr_sectors
))
1513 * Filter flush bio's early so that make_request based
1514 * drivers without flush support don't have to worry
1517 if ((bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) && !q
->flush_flags
) {
1518 bio
->bi_rw
&= ~(REQ_FLUSH
| REQ_FUA
);
1525 if ((bio
->bi_rw
& REQ_DISCARD
) &&
1526 (!blk_queue_discard(q
) ||
1527 ((bio
->bi_rw
& REQ_SECURE
) &&
1528 !blk_queue_secdiscard(q
)))) {
1533 trace_block_bio_queue(q
, bio
);
1535 ret
= q
->make_request_fn(q
, bio
);
1541 bio_endio(bio
, err
);
1545 * We only want one ->make_request_fn to be active at a time,
1546 * else stack usage with stacked devices could be a problem.
1547 * So use current->bio_list to keep a list of requests
1548 * submited by a make_request_fn function.
1549 * current->bio_list is also used as a flag to say if
1550 * generic_make_request is currently active in this task or not.
1551 * If it is NULL, then no make_request is active. If it is non-NULL,
1552 * then a make_request is active, and new requests should be added
1555 void generic_make_request(struct bio
*bio
)
1557 struct bio_list bio_list_on_stack
;
1559 if (current
->bio_list
) {
1560 /* make_request is active */
1561 bio_list_add(current
->bio_list
, bio
);
1564 /* following loop may be a bit non-obvious, and so deserves some
1566 * Before entering the loop, bio->bi_next is NULL (as all callers
1567 * ensure that) so we have a list with a single bio.
1568 * We pretend that we have just taken it off a longer list, so
1569 * we assign bio_list to a pointer to the bio_list_on_stack,
1570 * thus initialising the bio_list of new bios to be
1571 * added. __generic_make_request may indeed add some more bios
1572 * through a recursive call to generic_make_request. If it
1573 * did, we find a non-NULL value in bio_list and re-enter the loop
1574 * from the top. In this case we really did just take the bio
1575 * of the top of the list (no pretending) and so remove it from
1576 * bio_list, and call into __generic_make_request again.
1578 * The loop was structured like this to make only one call to
1579 * __generic_make_request (which is important as it is large and
1580 * inlined) and to keep the structure simple.
1582 BUG_ON(bio
->bi_next
);
1583 bio_list_init(&bio_list_on_stack
);
1584 current
->bio_list
= &bio_list_on_stack
;
1586 __generic_make_request(bio
);
1587 bio
= bio_list_pop(current
->bio_list
);
1589 current
->bio_list
= NULL
; /* deactivate */
1591 EXPORT_SYMBOL(generic_make_request
);
1594 * submit_bio - submit a bio to the block device layer for I/O
1595 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1596 * @bio: The &struct bio which describes the I/O
1598 * submit_bio() is very similar in purpose to generic_make_request(), and
1599 * uses that function to do most of the work. Both are fairly rough
1600 * interfaces; @bio must be presetup and ready for I/O.
1603 void submit_bio(int rw
, struct bio
*bio
)
1605 int count
= bio_sectors(bio
);
1610 * If it's a regular read/write or a barrier with data attached,
1611 * go through the normal accounting stuff before submission.
1613 if (bio_has_data(bio
) && !(rw
& REQ_DISCARD
)) {
1615 count_vm_events(PGPGOUT
, count
);
1617 task_io_account_read(bio
->bi_size
);
1618 count_vm_events(PGPGIN
, count
);
1621 if (unlikely(block_dump
)) {
1622 char b
[BDEVNAME_SIZE
];
1623 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s\n",
1624 current
->comm
, task_pid_nr(current
),
1625 (rw
& WRITE
) ? "WRITE" : "READ",
1626 (unsigned long long)bio
->bi_sector
,
1627 bdevname(bio
->bi_bdev
, b
));
1631 generic_make_request(bio
);
1633 EXPORT_SYMBOL(submit_bio
);
1636 * blk_rq_check_limits - Helper function to check a request for the queue limit
1638 * @rq: the request being checked
1641 * @rq may have been made based on weaker limitations of upper-level queues
1642 * in request stacking drivers, and it may violate the limitation of @q.
1643 * Since the block layer and the underlying device driver trust @rq
1644 * after it is inserted to @q, it should be checked against @q before
1645 * the insertion using this generic function.
1647 * This function should also be useful for request stacking drivers
1648 * in some cases below, so export this fuction.
1649 * Request stacking drivers like request-based dm may change the queue
1650 * limits while requests are in the queue (e.g. dm's table swapping).
1651 * Such request stacking drivers should check those requests agaist
1652 * the new queue limits again when they dispatch those requests,
1653 * although such checkings are also done against the old queue limits
1654 * when submitting requests.
1656 int blk_rq_check_limits(struct request_queue
*q
, struct request
*rq
)
1658 if (rq
->cmd_flags
& REQ_DISCARD
)
1661 if (blk_rq_sectors(rq
) > queue_max_sectors(q
) ||
1662 blk_rq_bytes(rq
) > queue_max_hw_sectors(q
) << 9) {
1663 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
1668 * queue's settings related to segment counting like q->bounce_pfn
1669 * may differ from that of other stacking queues.
1670 * Recalculate it to check the request correctly on this queue's
1673 blk_recalc_rq_segments(rq
);
1674 if (rq
->nr_phys_segments
> queue_max_segments(q
)) {
1675 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
1681 EXPORT_SYMBOL_GPL(blk_rq_check_limits
);
1684 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1685 * @q: the queue to submit the request
1686 * @rq: the request being queued
1688 int blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
1690 unsigned long flags
;
1692 if (blk_rq_check_limits(q
, rq
))
1695 #ifdef CONFIG_FAIL_MAKE_REQUEST
1696 if (rq
->rq_disk
&& rq
->rq_disk
->part0
.make_it_fail
&&
1697 should_fail(&fail_make_request
, blk_rq_bytes(rq
)))
1701 spin_lock_irqsave(q
->queue_lock
, flags
);
1704 * Submitting request must be dequeued before calling this function
1705 * because it will be linked to another request_queue
1707 BUG_ON(blk_queued_rq(rq
));
1709 drive_stat_acct(rq
, 1);
1710 __elv_add_request(q
, rq
, ELEVATOR_INSERT_BACK
, 0);
1712 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1716 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
1719 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1720 * @rq: request to examine
1723 * A request could be merge of IOs which require different failure
1724 * handling. This function determines the number of bytes which
1725 * can be failed from the beginning of the request without
1726 * crossing into area which need to be retried further.
1729 * The number of bytes to fail.
1732 * queue_lock must be held.
1734 unsigned int blk_rq_err_bytes(const struct request
*rq
)
1736 unsigned int ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
1737 unsigned int bytes
= 0;
1740 if (!(rq
->cmd_flags
& REQ_MIXED_MERGE
))
1741 return blk_rq_bytes(rq
);
1744 * Currently the only 'mixing' which can happen is between
1745 * different fastfail types. We can safely fail portions
1746 * which have all the failfast bits that the first one has -
1747 * the ones which are at least as eager to fail as the first
1750 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
1751 if ((bio
->bi_rw
& ff
) != ff
)
1753 bytes
+= bio
->bi_size
;
1756 /* this could lead to infinite loop */
1757 BUG_ON(blk_rq_bytes(rq
) && !bytes
);
1760 EXPORT_SYMBOL_GPL(blk_rq_err_bytes
);
1762 static void blk_account_io_completion(struct request
*req
, unsigned int bytes
)
1764 if (blk_do_io_stat(req
)) {
1765 const int rw
= rq_data_dir(req
);
1766 struct hd_struct
*part
;
1769 cpu
= part_stat_lock();
1770 part
= disk_map_sector_rcu(req
->rq_disk
, blk_rq_pos(req
));
1771 part_stat_add(cpu
, part
, sectors
[rw
], bytes
>> 9);
1776 static void blk_account_io_done(struct request
*req
)
1779 * Account IO completion. flush_rq isn't accounted as a
1780 * normal IO on queueing nor completion. Accounting the
1781 * containing request is enough.
1783 if (blk_do_io_stat(req
) && req
!= &req
->q
->flush_rq
) {
1784 unsigned long duration
= jiffies
- req
->start_time
;
1785 const int rw
= rq_data_dir(req
);
1786 struct hd_struct
*part
;
1789 cpu
= part_stat_lock();
1790 part
= disk_map_sector_rcu(req
->rq_disk
, blk_rq_pos(req
));
1792 part_stat_inc(cpu
, part
, ios
[rw
]);
1793 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
1794 part_round_stats(cpu
, part
);
1795 part_dec_in_flight(part
, rw
);
1802 * blk_peek_request - peek at the top of a request queue
1803 * @q: request queue to peek at
1806 * Return the request at the top of @q. The returned request
1807 * should be started using blk_start_request() before LLD starts
1811 * Pointer to the request at the top of @q if available. Null
1815 * queue_lock must be held.
1817 struct request
*blk_peek_request(struct request_queue
*q
)
1822 while ((rq
= __elv_next_request(q
)) != NULL
) {
1823 if (!(rq
->cmd_flags
& REQ_STARTED
)) {
1825 * This is the first time the device driver
1826 * sees this request (possibly after
1827 * requeueing). Notify IO scheduler.
1829 if (rq
->cmd_flags
& REQ_SORTED
)
1830 elv_activate_rq(q
, rq
);
1833 * just mark as started even if we don't start
1834 * it, a request that has been delayed should
1835 * not be passed by new incoming requests
1837 rq
->cmd_flags
|= REQ_STARTED
;
1838 trace_block_rq_issue(q
, rq
);
1841 if (!q
->boundary_rq
|| q
->boundary_rq
== rq
) {
1842 q
->end_sector
= rq_end_sector(rq
);
1843 q
->boundary_rq
= NULL
;
1846 if (rq
->cmd_flags
& REQ_DONTPREP
)
1849 if (q
->dma_drain_size
&& blk_rq_bytes(rq
)) {
1851 * make sure space for the drain appears we
1852 * know we can do this because max_hw_segments
1853 * has been adjusted to be one fewer than the
1856 rq
->nr_phys_segments
++;
1862 ret
= q
->prep_rq_fn(q
, rq
);
1863 if (ret
== BLKPREP_OK
) {
1865 } else if (ret
== BLKPREP_DEFER
) {
1867 * the request may have been (partially) prepped.
1868 * we need to keep this request in the front to
1869 * avoid resource deadlock. REQ_STARTED will
1870 * prevent other fs requests from passing this one.
1872 if (q
->dma_drain_size
&& blk_rq_bytes(rq
) &&
1873 !(rq
->cmd_flags
& REQ_DONTPREP
)) {
1875 * remove the space for the drain we added
1876 * so that we don't add it again
1878 --rq
->nr_phys_segments
;
1883 } else if (ret
== BLKPREP_KILL
) {
1884 rq
->cmd_flags
|= REQ_QUIET
;
1886 * Mark this request as started so we don't trigger
1887 * any debug logic in the end I/O path.
1889 blk_start_request(rq
);
1890 __blk_end_request_all(rq
, -EIO
);
1892 printk(KERN_ERR
"%s: bad return=%d\n", __func__
, ret
);
1899 EXPORT_SYMBOL(blk_peek_request
);
1901 void blk_dequeue_request(struct request
*rq
)
1903 struct request_queue
*q
= rq
->q
;
1905 BUG_ON(list_empty(&rq
->queuelist
));
1906 BUG_ON(ELV_ON_HASH(rq
));
1908 list_del_init(&rq
->queuelist
);
1911 * the time frame between a request being removed from the lists
1912 * and to it is freed is accounted as io that is in progress at
1915 if (blk_account_rq(rq
)) {
1916 q
->in_flight
[rq_is_sync(rq
)]++;
1917 set_io_start_time_ns(rq
);
1922 * blk_start_request - start request processing on the driver
1923 * @req: request to dequeue
1926 * Dequeue @req and start timeout timer on it. This hands off the
1927 * request to the driver.
1929 * Block internal functions which don't want to start timer should
1930 * call blk_dequeue_request().
1933 * queue_lock must be held.
1935 void blk_start_request(struct request
*req
)
1937 blk_dequeue_request(req
);
1940 * We are now handing the request to the hardware, initialize
1941 * resid_len to full count and add the timeout handler.
1943 req
->resid_len
= blk_rq_bytes(req
);
1944 if (unlikely(blk_bidi_rq(req
)))
1945 req
->next_rq
->resid_len
= blk_rq_bytes(req
->next_rq
);
1949 EXPORT_SYMBOL(blk_start_request
);
1952 * blk_fetch_request - fetch a request from a request queue
1953 * @q: request queue to fetch a request from
1956 * Return the request at the top of @q. The request is started on
1957 * return and LLD can start processing it immediately.
1960 * Pointer to the request at the top of @q if available. Null
1964 * queue_lock must be held.
1966 struct request
*blk_fetch_request(struct request_queue
*q
)
1970 rq
= blk_peek_request(q
);
1972 blk_start_request(rq
);
1975 EXPORT_SYMBOL(blk_fetch_request
);
1978 * blk_update_request - Special helper function for request stacking drivers
1979 * @req: the request being processed
1980 * @error: %0 for success, < %0 for error
1981 * @nr_bytes: number of bytes to complete @req
1984 * Ends I/O on a number of bytes attached to @req, but doesn't complete
1985 * the request structure even if @req doesn't have leftover.
1986 * If @req has leftover, sets it up for the next range of segments.
1988 * This special helper function is only for request stacking drivers
1989 * (e.g. request-based dm) so that they can handle partial completion.
1990 * Actual device drivers should use blk_end_request instead.
1992 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
1993 * %false return from this function.
1996 * %false - this request doesn't have any more data
1997 * %true - this request has more data
1999 bool blk_update_request(struct request
*req
, int error
, unsigned int nr_bytes
)
2001 int total_bytes
, bio_nbytes
, next_idx
= 0;
2007 trace_block_rq_complete(req
->q
, req
);
2010 * For fs requests, rq is just carrier of independent bio's
2011 * and each partial completion should be handled separately.
2012 * Reset per-request error on each partial completion.
2014 * TODO: tj: This is too subtle. It would be better to let
2015 * low level drivers do what they see fit.
2017 if (req
->cmd_type
== REQ_TYPE_FS
)
2020 if (error
&& req
->cmd_type
== REQ_TYPE_FS
&&
2021 !(req
->cmd_flags
& REQ_QUIET
)) {
2022 printk(KERN_ERR
"end_request: I/O error, dev %s, sector %llu\n",
2023 req
->rq_disk
? req
->rq_disk
->disk_name
: "?",
2024 (unsigned long long)blk_rq_pos(req
));
2027 blk_account_io_completion(req
, nr_bytes
);
2029 total_bytes
= bio_nbytes
= 0;
2030 while ((bio
= req
->bio
) != NULL
) {
2033 if (nr_bytes
>= bio
->bi_size
) {
2034 req
->bio
= bio
->bi_next
;
2035 nbytes
= bio
->bi_size
;
2036 req_bio_endio(req
, bio
, nbytes
, error
);
2040 int idx
= bio
->bi_idx
+ next_idx
;
2042 if (unlikely(idx
>= bio
->bi_vcnt
)) {
2043 blk_dump_rq_flags(req
, "__end_that");
2044 printk(KERN_ERR
"%s: bio idx %d >= vcnt %d\n",
2045 __func__
, idx
, bio
->bi_vcnt
);
2049 nbytes
= bio_iovec_idx(bio
, idx
)->bv_len
;
2050 BIO_BUG_ON(nbytes
> bio
->bi_size
);
2053 * not a complete bvec done
2055 if (unlikely(nbytes
> nr_bytes
)) {
2056 bio_nbytes
+= nr_bytes
;
2057 total_bytes
+= nr_bytes
;
2062 * advance to the next vector
2065 bio_nbytes
+= nbytes
;
2068 total_bytes
+= nbytes
;
2074 * end more in this run, or just return 'not-done'
2076 if (unlikely(nr_bytes
<= 0))
2086 * Reset counters so that the request stacking driver
2087 * can find how many bytes remain in the request
2090 req
->__data_len
= 0;
2095 * if the request wasn't completed, update state
2098 req_bio_endio(req
, bio
, bio_nbytes
, error
);
2099 bio
->bi_idx
+= next_idx
;
2100 bio_iovec(bio
)->bv_offset
+= nr_bytes
;
2101 bio_iovec(bio
)->bv_len
-= nr_bytes
;
2104 req
->__data_len
-= total_bytes
;
2105 req
->buffer
= bio_data(req
->bio
);
2107 /* update sector only for requests with clear definition of sector */
2108 if (req
->cmd_type
== REQ_TYPE_FS
|| (req
->cmd_flags
& REQ_DISCARD
))
2109 req
->__sector
+= total_bytes
>> 9;
2111 /* mixed attributes always follow the first bio */
2112 if (req
->cmd_flags
& REQ_MIXED_MERGE
) {
2113 req
->cmd_flags
&= ~REQ_FAILFAST_MASK
;
2114 req
->cmd_flags
|= req
->bio
->bi_rw
& REQ_FAILFAST_MASK
;
2118 * If total number of sectors is less than the first segment
2119 * size, something has gone terribly wrong.
2121 if (blk_rq_bytes(req
) < blk_rq_cur_bytes(req
)) {
2122 printk(KERN_ERR
"blk: request botched\n");
2123 req
->__data_len
= blk_rq_cur_bytes(req
);
2126 /* recalculate the number of segments */
2127 blk_recalc_rq_segments(req
);
2131 EXPORT_SYMBOL_GPL(blk_update_request
);
2133 static bool blk_update_bidi_request(struct request
*rq
, int error
,
2134 unsigned int nr_bytes
,
2135 unsigned int bidi_bytes
)
2137 if (blk_update_request(rq
, error
, nr_bytes
))
2140 /* Bidi request must be completed as a whole */
2141 if (unlikely(blk_bidi_rq(rq
)) &&
2142 blk_update_request(rq
->next_rq
, error
, bidi_bytes
))
2145 if (blk_queue_add_random(rq
->q
))
2146 add_disk_randomness(rq
->rq_disk
);
2152 * blk_unprep_request - unprepare a request
2155 * This function makes a request ready for complete resubmission (or
2156 * completion). It happens only after all error handling is complete,
2157 * so represents the appropriate moment to deallocate any resources
2158 * that were allocated to the request in the prep_rq_fn. The queue
2159 * lock is held when calling this.
2161 void blk_unprep_request(struct request
*req
)
2163 struct request_queue
*q
= req
->q
;
2165 req
->cmd_flags
&= ~REQ_DONTPREP
;
2166 if (q
->unprep_rq_fn
)
2167 q
->unprep_rq_fn(q
, req
);
2169 EXPORT_SYMBOL_GPL(blk_unprep_request
);
2172 * queue lock must be held
2174 static void blk_finish_request(struct request
*req
, int error
)
2176 if (blk_rq_tagged(req
))
2177 blk_queue_end_tag(req
->q
, req
);
2179 BUG_ON(blk_queued_rq(req
));
2181 if (unlikely(laptop_mode
) && req
->cmd_type
== REQ_TYPE_FS
)
2182 laptop_io_completion(&req
->q
->backing_dev_info
);
2184 blk_delete_timer(req
);
2186 if (req
->cmd_flags
& REQ_DONTPREP
)
2187 blk_unprep_request(req
);
2190 blk_account_io_done(req
);
2193 req
->end_io(req
, error
);
2195 if (blk_bidi_rq(req
))
2196 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
2198 __blk_put_request(req
->q
, req
);
2203 * blk_end_bidi_request - Complete a bidi request
2204 * @rq: the request to complete
2205 * @error: %0 for success, < %0 for error
2206 * @nr_bytes: number of bytes to complete @rq
2207 * @bidi_bytes: number of bytes to complete @rq->next_rq
2210 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2211 * Drivers that supports bidi can safely call this member for any
2212 * type of request, bidi or uni. In the later case @bidi_bytes is
2216 * %false - we are done with this request
2217 * %true - still buffers pending for this request
2219 static bool blk_end_bidi_request(struct request
*rq
, int error
,
2220 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2222 struct request_queue
*q
= rq
->q
;
2223 unsigned long flags
;
2225 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2228 spin_lock_irqsave(q
->queue_lock
, flags
);
2229 blk_finish_request(rq
, error
);
2230 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2236 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2237 * @rq: the request to complete
2238 * @error: %0 for success, < %0 for error
2239 * @nr_bytes: number of bytes to complete @rq
2240 * @bidi_bytes: number of bytes to complete @rq->next_rq
2243 * Identical to blk_end_bidi_request() except that queue lock is
2244 * assumed to be locked on entry and remains so on return.
2247 * %false - we are done with this request
2248 * %true - still buffers pending for this request
2250 static bool __blk_end_bidi_request(struct request
*rq
, int error
,
2251 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2253 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2256 blk_finish_request(rq
, error
);
2262 * blk_end_request - Helper function for drivers to complete the request.
2263 * @rq: the request being processed
2264 * @error: %0 for success, < %0 for error
2265 * @nr_bytes: number of bytes to complete
2268 * Ends I/O on a number of bytes attached to @rq.
2269 * If @rq has leftover, sets it up for the next range of segments.
2272 * %false - we are done with this request
2273 * %true - still buffers pending for this request
2275 bool blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2277 return blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2279 EXPORT_SYMBOL(blk_end_request
);
2282 * blk_end_request_all - Helper function for drives to finish the request.
2283 * @rq: the request to finish
2284 * @error: %0 for success, < %0 for error
2287 * Completely finish @rq.
2289 void blk_end_request_all(struct request
*rq
, int error
)
2292 unsigned int bidi_bytes
= 0;
2294 if (unlikely(blk_bidi_rq(rq
)))
2295 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2297 pending
= blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2300 EXPORT_SYMBOL(blk_end_request_all
);
2303 * blk_end_request_cur - Helper function to finish the current request chunk.
2304 * @rq: the request to finish the current chunk for
2305 * @error: %0 for success, < %0 for error
2308 * Complete the current consecutively mapped chunk from @rq.
2311 * %false - we are done with this request
2312 * %true - still buffers pending for this request
2314 bool blk_end_request_cur(struct request
*rq
, int error
)
2316 return blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2318 EXPORT_SYMBOL(blk_end_request_cur
);
2321 * blk_end_request_err - Finish a request till the next failure boundary.
2322 * @rq: the request to finish till the next failure boundary for
2323 * @error: must be negative errno
2326 * Complete @rq till the next failure boundary.
2329 * %false - we are done with this request
2330 * %true - still buffers pending for this request
2332 bool blk_end_request_err(struct request
*rq
, int error
)
2334 WARN_ON(error
>= 0);
2335 return blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2337 EXPORT_SYMBOL_GPL(blk_end_request_err
);
2340 * __blk_end_request - Helper function for drivers to complete the request.
2341 * @rq: the request being processed
2342 * @error: %0 for success, < %0 for error
2343 * @nr_bytes: number of bytes to complete
2346 * Must be called with queue lock held unlike blk_end_request().
2349 * %false - we are done with this request
2350 * %true - still buffers pending for this request
2352 bool __blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2354 return __blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2356 EXPORT_SYMBOL(__blk_end_request
);
2359 * __blk_end_request_all - Helper function for drives to finish the request.
2360 * @rq: the request to finish
2361 * @error: %0 for success, < %0 for error
2364 * Completely finish @rq. Must be called with queue lock held.
2366 void __blk_end_request_all(struct request
*rq
, int error
)
2369 unsigned int bidi_bytes
= 0;
2371 if (unlikely(blk_bidi_rq(rq
)))
2372 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2374 pending
= __blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2377 EXPORT_SYMBOL(__blk_end_request_all
);
2380 * __blk_end_request_cur - Helper function to finish the current request chunk.
2381 * @rq: the request to finish the current chunk for
2382 * @error: %0 for success, < %0 for error
2385 * Complete the current consecutively mapped chunk from @rq. Must
2386 * be called with queue lock held.
2389 * %false - we are done with this request
2390 * %true - still buffers pending for this request
2392 bool __blk_end_request_cur(struct request
*rq
, int error
)
2394 return __blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2396 EXPORT_SYMBOL(__blk_end_request_cur
);
2399 * __blk_end_request_err - Finish a request till the next failure boundary.
2400 * @rq: the request to finish till the next failure boundary for
2401 * @error: must be negative errno
2404 * Complete @rq till the next failure boundary. Must be called
2405 * with queue lock held.
2408 * %false - we are done with this request
2409 * %true - still buffers pending for this request
2411 bool __blk_end_request_err(struct request
*rq
, int error
)
2413 WARN_ON(error
>= 0);
2414 return __blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2416 EXPORT_SYMBOL_GPL(__blk_end_request_err
);
2418 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
2421 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2422 rq
->cmd_flags
|= bio
->bi_rw
& REQ_WRITE
;
2424 if (bio_has_data(bio
)) {
2425 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2426 rq
->buffer
= bio_data(bio
);
2428 rq
->__data_len
= bio
->bi_size
;
2429 rq
->bio
= rq
->biotail
= bio
;
2432 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2435 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2437 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2438 * @rq: the request to be flushed
2441 * Flush all pages in @rq.
2443 void rq_flush_dcache_pages(struct request
*rq
)
2445 struct req_iterator iter
;
2446 struct bio_vec
*bvec
;
2448 rq_for_each_segment(bvec
, rq
, iter
)
2449 flush_dcache_page(bvec
->bv_page
);
2451 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages
);
2455 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2456 * @q : the queue of the device being checked
2459 * Check if underlying low-level drivers of a device are busy.
2460 * If the drivers want to export their busy state, they must set own
2461 * exporting function using blk_queue_lld_busy() first.
2463 * Basically, this function is used only by request stacking drivers
2464 * to stop dispatching requests to underlying devices when underlying
2465 * devices are busy. This behavior helps more I/O merging on the queue
2466 * of the request stacking driver and prevents I/O throughput regression
2467 * on burst I/O load.
2470 * 0 - Not busy (The request stacking driver should dispatch request)
2471 * 1 - Busy (The request stacking driver should stop dispatching request)
2473 int blk_lld_busy(struct request_queue
*q
)
2476 return q
->lld_busy_fn(q
);
2480 EXPORT_SYMBOL_GPL(blk_lld_busy
);
2483 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2484 * @rq: the clone request to be cleaned up
2487 * Free all bios in @rq for a cloned request.
2489 void blk_rq_unprep_clone(struct request
*rq
)
2493 while ((bio
= rq
->bio
) != NULL
) {
2494 rq
->bio
= bio
->bi_next
;
2499 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone
);
2502 * Copy attributes of the original request to the clone request.
2503 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2505 static void __blk_rq_prep_clone(struct request
*dst
, struct request
*src
)
2507 dst
->cpu
= src
->cpu
;
2508 dst
->cmd_flags
= (rq_data_dir(src
) | REQ_NOMERGE
);
2509 if (src
->cmd_flags
& REQ_DISCARD
)
2510 dst
->cmd_flags
|= REQ_DISCARD
;
2511 dst
->cmd_type
= src
->cmd_type
;
2512 dst
->__sector
= blk_rq_pos(src
);
2513 dst
->__data_len
= blk_rq_bytes(src
);
2514 dst
->nr_phys_segments
= src
->nr_phys_segments
;
2515 dst
->ioprio
= src
->ioprio
;
2516 dst
->extra_len
= src
->extra_len
;
2520 * blk_rq_prep_clone - Helper function to setup clone request
2521 * @rq: the request to be setup
2522 * @rq_src: original request to be cloned
2523 * @bs: bio_set that bios for clone are allocated from
2524 * @gfp_mask: memory allocation mask for bio
2525 * @bio_ctr: setup function to be called for each clone bio.
2526 * Returns %0 for success, non %0 for failure.
2527 * @data: private data to be passed to @bio_ctr
2530 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2531 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2532 * are not copied, and copying such parts is the caller's responsibility.
2533 * Also, pages which the original bios are pointing to are not copied
2534 * and the cloned bios just point same pages.
2535 * So cloned bios must be completed before original bios, which means
2536 * the caller must complete @rq before @rq_src.
2538 int blk_rq_prep_clone(struct request
*rq
, struct request
*rq_src
,
2539 struct bio_set
*bs
, gfp_t gfp_mask
,
2540 int (*bio_ctr
)(struct bio
*, struct bio
*, void *),
2543 struct bio
*bio
, *bio_src
;
2548 blk_rq_init(NULL
, rq
);
2550 __rq_for_each_bio(bio_src
, rq_src
) {
2551 bio
= bio_alloc_bioset(gfp_mask
, bio_src
->bi_max_vecs
, bs
);
2555 __bio_clone(bio
, bio_src
);
2557 if (bio_integrity(bio_src
) &&
2558 bio_integrity_clone(bio
, bio_src
, gfp_mask
, bs
))
2561 if (bio_ctr
&& bio_ctr(bio
, bio_src
, data
))
2565 rq
->biotail
->bi_next
= bio
;
2568 rq
->bio
= rq
->biotail
= bio
;
2571 __blk_rq_prep_clone(rq
, rq_src
);
2578 blk_rq_unprep_clone(rq
);
2582 EXPORT_SYMBOL_GPL(blk_rq_prep_clone
);
2584 int kblockd_schedule_work(struct request_queue
*q
, struct work_struct
*work
)
2586 return queue_work(kblockd_workqueue
, work
);
2588 EXPORT_SYMBOL(kblockd_schedule_work
);
2590 int __init
blk_dev_init(void)
2592 BUILD_BUG_ON(__REQ_NR_BITS
> 8 *
2593 sizeof(((struct request
*)0)->cmd_flags
));
2595 kblockd_workqueue
= create_workqueue("kblockd");
2596 if (!kblockd_workqueue
)
2597 panic("Failed to create kblockd\n");
2599 request_cachep
= kmem_cache_create("blkdev_requests",
2600 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
2602 blk_requestq_cachep
= kmem_cache_create("blkdev_queue",
2603 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);