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_bio_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();
70 part_stat_inc(cpu
, part
, merges
[rw
]);
72 part
= disk_map_sector_rcu(rq
->rq_disk
, blk_rq_pos(rq
));
73 if (!hd_struct_try_get(part
)) {
75 * The partition is already being removed,
76 * the request will be accounted on the disk only
78 * We take a reference on disk->part0 although that
79 * partition will never be deleted, so we can treat
80 * it as any other partition.
82 part
= &rq
->rq_disk
->part0
;
85 part_round_stats(cpu
, part
);
86 part_inc_in_flight(part
, rw
);
93 void blk_queue_congestion_threshold(struct request_queue
*q
)
97 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) + 1;
98 if (nr
> q
->nr_requests
)
100 q
->nr_congestion_on
= nr
;
102 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) - (q
->nr_requests
/ 16) - 1;
105 q
->nr_congestion_off
= nr
;
109 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
112 * Locates the passed device's request queue and returns the address of its
115 * Will return NULL if the request queue cannot be located.
117 struct backing_dev_info
*blk_get_backing_dev_info(struct block_device
*bdev
)
119 struct backing_dev_info
*ret
= NULL
;
120 struct request_queue
*q
= bdev_get_queue(bdev
);
123 ret
= &q
->backing_dev_info
;
126 EXPORT_SYMBOL(blk_get_backing_dev_info
);
128 void blk_rq_init(struct request_queue
*q
, struct request
*rq
)
130 memset(rq
, 0, sizeof(*rq
));
132 INIT_LIST_HEAD(&rq
->queuelist
);
133 INIT_LIST_HEAD(&rq
->timeout_list
);
136 rq
->__sector
= (sector_t
) -1;
137 INIT_HLIST_NODE(&rq
->hash
);
138 RB_CLEAR_NODE(&rq
->rb_node
);
140 rq
->cmd_len
= BLK_MAX_CDB
;
143 rq
->start_time
= jiffies
;
144 set_start_time_ns(rq
);
147 EXPORT_SYMBOL(blk_rq_init
);
149 static void req_bio_endio(struct request
*rq
, struct bio
*bio
,
150 unsigned int nbytes
, int error
)
153 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
154 else if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
157 if (unlikely(nbytes
> bio
->bi_size
)) {
158 printk(KERN_ERR
"%s: want %u bytes done, %u left\n",
159 __func__
, nbytes
, bio
->bi_size
);
160 nbytes
= bio
->bi_size
;
163 if (unlikely(rq
->cmd_flags
& REQ_QUIET
))
164 set_bit(BIO_QUIET
, &bio
->bi_flags
);
166 bio
->bi_size
-= nbytes
;
167 bio
->bi_sector
+= (nbytes
>> 9);
169 if (bio_integrity(bio
))
170 bio_integrity_advance(bio
, nbytes
);
172 /* don't actually finish bio if it's part of flush sequence */
173 if (bio
->bi_size
== 0 && !(rq
->cmd_flags
& REQ_FLUSH_SEQ
))
174 bio_endio(bio
, error
);
177 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
181 printk(KERN_INFO
"%s: dev %s: type=%x, flags=%x\n", msg
,
182 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->cmd_type
,
185 printk(KERN_INFO
" sector %llu, nr/cnr %u/%u\n",
186 (unsigned long long)blk_rq_pos(rq
),
187 blk_rq_sectors(rq
), blk_rq_cur_sectors(rq
));
188 printk(KERN_INFO
" bio %p, biotail %p, buffer %p, len %u\n",
189 rq
->bio
, rq
->biotail
, rq
->buffer
, blk_rq_bytes(rq
));
191 if (rq
->cmd_type
== REQ_TYPE_BLOCK_PC
) {
192 printk(KERN_INFO
" cdb: ");
193 for (bit
= 0; bit
< BLK_MAX_CDB
; bit
++)
194 printk("%02x ", rq
->cmd
[bit
]);
198 EXPORT_SYMBOL(blk_dump_rq_flags
);
201 * "plug" the device if there are no outstanding requests: this will
202 * force the transfer to start only after we have put all the requests
205 * This is called with interrupts off and no requests on the queue and
206 * with the queue lock held.
208 void blk_plug_device(struct request_queue
*q
)
210 WARN_ON(!irqs_disabled());
213 * don't plug a stopped queue, it must be paired with blk_start_queue()
214 * which will restart the queueing
216 if (blk_queue_stopped(q
))
219 if (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED
, q
)) {
220 mod_timer(&q
->unplug_timer
, jiffies
+ q
->unplug_delay
);
224 EXPORT_SYMBOL(blk_plug_device
);
227 * blk_plug_device_unlocked - plug a device without queue lock held
228 * @q: The &struct request_queue to plug
231 * Like @blk_plug_device(), but grabs the queue lock and disables
234 void blk_plug_device_unlocked(struct request_queue
*q
)
238 spin_lock_irqsave(q
->queue_lock
, flags
);
240 spin_unlock_irqrestore(q
->queue_lock
, flags
);
242 EXPORT_SYMBOL(blk_plug_device_unlocked
);
245 * remove the queue from the plugged list, if present. called with
246 * queue lock held and interrupts disabled.
248 int blk_remove_plug(struct request_queue
*q
)
250 WARN_ON(!irqs_disabled());
252 if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED
, q
))
255 del_timer(&q
->unplug_timer
);
258 EXPORT_SYMBOL(blk_remove_plug
);
261 * remove the plug and let it rip..
263 void __generic_unplug_device(struct request_queue
*q
)
265 if (unlikely(blk_queue_stopped(q
)))
267 if (!blk_remove_plug(q
) && !blk_queue_nonrot(q
))
274 * generic_unplug_device - fire a request queue
275 * @q: The &struct request_queue in question
278 * Linux uses plugging to build bigger requests queues before letting
279 * the device have at them. If a queue is plugged, the I/O scheduler
280 * is still adding and merging requests on the queue. Once the queue
281 * gets unplugged, the request_fn defined for the queue is invoked and
284 void generic_unplug_device(struct request_queue
*q
)
286 if (blk_queue_plugged(q
)) {
287 spin_lock_irq(q
->queue_lock
);
288 __generic_unplug_device(q
);
289 spin_unlock_irq(q
->queue_lock
);
292 EXPORT_SYMBOL(generic_unplug_device
);
294 static void blk_backing_dev_unplug(struct backing_dev_info
*bdi
,
297 struct request_queue
*q
= bdi
->unplug_io_data
;
302 void blk_unplug_work(struct work_struct
*work
)
304 struct request_queue
*q
=
305 container_of(work
, struct request_queue
, unplug_work
);
307 trace_block_unplug_io(q
);
311 void blk_unplug_timeout(unsigned long data
)
313 struct request_queue
*q
= (struct request_queue
*)data
;
315 trace_block_unplug_timer(q
);
316 kblockd_schedule_work(q
, &q
->unplug_work
);
319 void blk_unplug(struct request_queue
*q
)
322 * devices don't necessarily have an ->unplug_fn defined
325 trace_block_unplug_io(q
);
329 EXPORT_SYMBOL(blk_unplug
);
332 * blk_start_queue - restart a previously stopped queue
333 * @q: The &struct request_queue in question
336 * blk_start_queue() will clear the stop flag on the queue, and call
337 * the request_fn for the queue if it was in a stopped state when
338 * entered. Also see blk_stop_queue(). Queue lock must be held.
340 void blk_start_queue(struct request_queue
*q
)
342 WARN_ON(!irqs_disabled());
344 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
347 EXPORT_SYMBOL(blk_start_queue
);
350 * blk_stop_queue - stop a queue
351 * @q: The &struct request_queue in question
354 * The Linux block layer assumes that a block driver will consume all
355 * entries on the request queue when the request_fn strategy is called.
356 * Often this will not happen, because of hardware limitations (queue
357 * depth settings). If a device driver gets a 'queue full' response,
358 * or if it simply chooses not to queue more I/O at one point, it can
359 * call this function to prevent the request_fn from being called until
360 * the driver has signalled it's ready to go again. This happens by calling
361 * blk_start_queue() to restart queue operations. Queue lock must be held.
363 void blk_stop_queue(struct request_queue
*q
)
366 queue_flag_set(QUEUE_FLAG_STOPPED
, q
);
368 EXPORT_SYMBOL(blk_stop_queue
);
371 * blk_sync_queue - cancel any pending callbacks on a queue
375 * The block layer may perform asynchronous callback activity
376 * on a queue, such as calling the unplug function after a timeout.
377 * A block device may call blk_sync_queue to ensure that any
378 * such activity is cancelled, thus allowing it to release resources
379 * that the callbacks might use. The caller must already have made sure
380 * that its ->make_request_fn will not re-add plugging prior to calling
384 void blk_sync_queue(struct request_queue
*q
)
386 del_timer_sync(&q
->unplug_timer
);
387 del_timer_sync(&q
->timeout
);
388 cancel_work_sync(&q
->unplug_work
);
389 throtl_shutdown_timer_wq(q
);
391 EXPORT_SYMBOL(blk_sync_queue
);
394 * __blk_run_queue - run a single device queue
395 * @q: The queue to run
398 * See @blk_run_queue. This variant must be called with the queue lock
399 * held and interrupts disabled.
402 void __blk_run_queue(struct request_queue
*q
)
406 if (unlikely(blk_queue_stopped(q
)))
409 if (elv_queue_empty(q
))
413 * Only recurse once to avoid overrunning the stack, let the unplug
414 * handling reinvoke the handler shortly if we already got there.
416 if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER
, q
)) {
418 queue_flag_clear(QUEUE_FLAG_REENTER
, q
);
420 queue_flag_set(QUEUE_FLAG_PLUGGED
, q
);
421 kblockd_schedule_work(q
, &q
->unplug_work
);
424 EXPORT_SYMBOL(__blk_run_queue
);
427 * blk_run_queue - run a single device queue
428 * @q: The queue to run
431 * Invoke request handling on this queue, if it has pending work to do.
432 * May be used to restart queueing when a request has completed.
434 void blk_run_queue(struct request_queue
*q
)
438 spin_lock_irqsave(q
->queue_lock
, flags
);
440 spin_unlock_irqrestore(q
->queue_lock
, flags
);
442 EXPORT_SYMBOL(blk_run_queue
);
444 void blk_put_queue(struct request_queue
*q
)
446 kobject_put(&q
->kobj
);
449 void blk_cleanup_queue(struct request_queue
*q
)
452 * We know we have process context here, so we can be a little
453 * cautious and ensure that pending block actions on this device
454 * are done before moving on. Going into this function, we should
455 * not have processes doing IO to this device.
459 del_timer_sync(&q
->backing_dev_info
.laptop_mode_wb_timer
);
460 mutex_lock(&q
->sysfs_lock
);
461 queue_flag_set_unlocked(QUEUE_FLAG_DEAD
, q
);
462 mutex_unlock(&q
->sysfs_lock
);
465 elevator_exit(q
->elevator
);
469 EXPORT_SYMBOL(blk_cleanup_queue
);
471 static int blk_init_free_list(struct request_queue
*q
)
473 struct request_list
*rl
= &q
->rq
;
475 if (unlikely(rl
->rq_pool
))
478 rl
->count
[BLK_RW_SYNC
] = rl
->count
[BLK_RW_ASYNC
] = 0;
479 rl
->starved
[BLK_RW_SYNC
] = rl
->starved
[BLK_RW_ASYNC
] = 0;
481 init_waitqueue_head(&rl
->wait
[BLK_RW_SYNC
]);
482 init_waitqueue_head(&rl
->wait
[BLK_RW_ASYNC
]);
484 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, mempool_alloc_slab
,
485 mempool_free_slab
, request_cachep
, q
->node
);
493 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
495 return blk_alloc_queue_node(gfp_mask
, -1);
497 EXPORT_SYMBOL(blk_alloc_queue
);
499 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
501 struct request_queue
*q
;
504 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
505 gfp_mask
| __GFP_ZERO
, node_id
);
509 q
->backing_dev_info
.unplug_io_fn
= blk_backing_dev_unplug
;
510 q
->backing_dev_info
.unplug_io_data
= q
;
511 q
->backing_dev_info
.ra_pages
=
512 (VM_MAX_READAHEAD
* 1024) / PAGE_CACHE_SIZE
;
513 q
->backing_dev_info
.state
= 0;
514 q
->backing_dev_info
.capabilities
= BDI_CAP_MAP_COPY
;
515 q
->backing_dev_info
.name
= "block";
517 err
= bdi_init(&q
->backing_dev_info
);
519 kmem_cache_free(blk_requestq_cachep
, q
);
523 if (blk_throtl_init(q
)) {
524 kmem_cache_free(blk_requestq_cachep
, q
);
528 setup_timer(&q
->backing_dev_info
.laptop_mode_wb_timer
,
529 laptop_mode_timer_fn
, (unsigned long) q
);
530 init_timer(&q
->unplug_timer
);
531 setup_timer(&q
->timeout
, blk_rq_timed_out_timer
, (unsigned long) q
);
532 INIT_LIST_HEAD(&q
->timeout_list
);
533 INIT_LIST_HEAD(&q
->flush_queue
[0]);
534 INIT_LIST_HEAD(&q
->flush_queue
[1]);
535 INIT_LIST_HEAD(&q
->flush_data_in_flight
);
536 INIT_WORK(&q
->unplug_work
, blk_unplug_work
);
538 kobject_init(&q
->kobj
, &blk_queue_ktype
);
540 mutex_init(&q
->sysfs_lock
);
541 spin_lock_init(&q
->__queue_lock
);
545 EXPORT_SYMBOL(blk_alloc_queue_node
);
548 * blk_init_queue - prepare a request queue for use with a block device
549 * @rfn: The function to be called to process requests that have been
550 * placed on the queue.
551 * @lock: Request queue spin lock
554 * If a block device wishes to use the standard request handling procedures,
555 * which sorts requests and coalesces adjacent requests, then it must
556 * call blk_init_queue(). The function @rfn will be called when there
557 * are requests on the queue that need to be processed. If the device
558 * supports plugging, then @rfn may not be called immediately when requests
559 * are available on the queue, but may be called at some time later instead.
560 * Plugged queues are generally unplugged when a buffer belonging to one
561 * of the requests on the queue is needed, or due to memory pressure.
563 * @rfn is not required, or even expected, to remove all requests off the
564 * queue, but only as many as it can handle at a time. If it does leave
565 * requests on the queue, it is responsible for arranging that the requests
566 * get dealt with eventually.
568 * The queue spin lock must be held while manipulating the requests on the
569 * request queue; this lock will be taken also from interrupt context, so irq
570 * disabling is needed for it.
572 * Function returns a pointer to the initialized request queue, or %NULL if
576 * blk_init_queue() must be paired with a blk_cleanup_queue() call
577 * when the block device is deactivated (such as at module unload).
580 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
582 return blk_init_queue_node(rfn
, lock
, -1);
584 EXPORT_SYMBOL(blk_init_queue
);
586 struct request_queue
*
587 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
589 struct request_queue
*uninit_q
, *q
;
591 uninit_q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
595 q
= blk_init_allocated_queue_node(uninit_q
, rfn
, lock
, node_id
);
597 blk_cleanup_queue(uninit_q
);
601 EXPORT_SYMBOL(blk_init_queue_node
);
603 struct request_queue
*
604 blk_init_allocated_queue(struct request_queue
*q
, request_fn_proc
*rfn
,
607 return blk_init_allocated_queue_node(q
, rfn
, lock
, -1);
609 EXPORT_SYMBOL(blk_init_allocated_queue
);
611 struct request_queue
*
612 blk_init_allocated_queue_node(struct request_queue
*q
, request_fn_proc
*rfn
,
613 spinlock_t
*lock
, int node_id
)
619 if (blk_init_free_list(q
))
623 q
->prep_rq_fn
= NULL
;
624 q
->unprep_rq_fn
= NULL
;
625 q
->unplug_fn
= generic_unplug_device
;
626 q
->queue_flags
= QUEUE_FLAG_DEFAULT
;
627 q
->queue_lock
= lock
;
630 * This also sets hw/phys segments, boundary and size
632 blk_queue_make_request(q
, __make_request
);
634 q
->sg_reserved_size
= INT_MAX
;
639 if (!elevator_init(q
, NULL
)) {
640 blk_queue_congestion_threshold(q
);
646 EXPORT_SYMBOL(blk_init_allocated_queue_node
);
648 int blk_get_queue(struct request_queue
*q
)
650 if (likely(!test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
))) {
651 kobject_get(&q
->kobj
);
658 static inline void blk_free_request(struct request_queue
*q
, struct request
*rq
)
660 if (rq
->cmd_flags
& REQ_ELVPRIV
)
661 elv_put_request(q
, rq
);
662 mempool_free(rq
, q
->rq
.rq_pool
);
665 static struct request
*
666 blk_alloc_request(struct request_queue
*q
, int flags
, int priv
, gfp_t gfp_mask
)
668 struct request
*rq
= mempool_alloc(q
->rq
.rq_pool
, gfp_mask
);
675 rq
->cmd_flags
= flags
| REQ_ALLOCED
;
678 if (unlikely(elv_set_request(q
, rq
, gfp_mask
))) {
679 mempool_free(rq
, q
->rq
.rq_pool
);
682 rq
->cmd_flags
|= REQ_ELVPRIV
;
689 * ioc_batching returns true if the ioc is a valid batching request and
690 * should be given priority access to a request.
692 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
698 * Make sure the process is able to allocate at least 1 request
699 * even if the batch times out, otherwise we could theoretically
702 return ioc
->nr_batch_requests
== q
->nr_batching
||
703 (ioc
->nr_batch_requests
> 0
704 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
708 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
709 * will cause the process to be a "batcher" on all queues in the system. This
710 * is the behaviour we want though - once it gets a wakeup it should be given
713 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
715 if (!ioc
|| ioc_batching(q
, ioc
))
718 ioc
->nr_batch_requests
= q
->nr_batching
;
719 ioc
->last_waited
= jiffies
;
722 static void __freed_request(struct request_queue
*q
, int sync
)
724 struct request_list
*rl
= &q
->rq
;
726 if (rl
->count
[sync
] < queue_congestion_off_threshold(q
))
727 blk_clear_queue_congested(q
, sync
);
729 if (rl
->count
[sync
] + 1 <= q
->nr_requests
) {
730 if (waitqueue_active(&rl
->wait
[sync
]))
731 wake_up(&rl
->wait
[sync
]);
733 blk_clear_queue_full(q
, sync
);
738 * A request has just been released. Account for it, update the full and
739 * congestion status, wake up any waiters. Called under q->queue_lock.
741 static void freed_request(struct request_queue
*q
, int sync
, int priv
)
743 struct request_list
*rl
= &q
->rq
;
749 __freed_request(q
, sync
);
751 if (unlikely(rl
->starved
[sync
^ 1]))
752 __freed_request(q
, sync
^ 1);
756 * Determine if elevator data should be initialized when allocating the
757 * request associated with @bio.
759 static bool blk_rq_should_init_elevator(struct bio
*bio
)
765 * Flush requests do not use the elevator so skip initialization.
766 * This allows a request to share the flush and elevator data.
768 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
))
775 * Get a free request, queue_lock must be held.
776 * Returns NULL on failure, with queue_lock held.
777 * Returns !NULL on success, with queue_lock *not held*.
779 static struct request
*get_request(struct request_queue
*q
, int rw_flags
,
780 struct bio
*bio
, gfp_t gfp_mask
)
782 struct request
*rq
= NULL
;
783 struct request_list
*rl
= &q
->rq
;
784 struct io_context
*ioc
= NULL
;
785 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
786 int may_queue
, priv
= 0;
788 may_queue
= elv_may_queue(q
, rw_flags
);
789 if (may_queue
== ELV_MQUEUE_NO
)
792 if (rl
->count
[is_sync
]+1 >= queue_congestion_on_threshold(q
)) {
793 if (rl
->count
[is_sync
]+1 >= q
->nr_requests
) {
794 ioc
= current_io_context(GFP_ATOMIC
, q
->node
);
796 * The queue will fill after this allocation, so set
797 * it as full, and mark this process as "batching".
798 * This process will be allowed to complete a batch of
799 * requests, others will be blocked.
801 if (!blk_queue_full(q
, is_sync
)) {
802 ioc_set_batching(q
, ioc
);
803 blk_set_queue_full(q
, is_sync
);
805 if (may_queue
!= ELV_MQUEUE_MUST
806 && !ioc_batching(q
, ioc
)) {
808 * The queue is full and the allocating
809 * process is not a "batcher", and not
810 * exempted by the IO scheduler
816 blk_set_queue_congested(q
, is_sync
);
820 * Only allow batching queuers to allocate up to 50% over the defined
821 * limit of requests, otherwise we could have thousands of requests
822 * allocated with any setting of ->nr_requests
824 if (rl
->count
[is_sync
] >= (3 * q
->nr_requests
/ 2))
827 rl
->count
[is_sync
]++;
828 rl
->starved
[is_sync
] = 0;
830 if (blk_rq_should_init_elevator(bio
)) {
831 priv
= !test_bit(QUEUE_FLAG_ELVSWITCH
, &q
->queue_flags
);
836 if (blk_queue_io_stat(q
))
837 rw_flags
|= REQ_IO_STAT
;
838 spin_unlock_irq(q
->queue_lock
);
840 rq
= blk_alloc_request(q
, rw_flags
, priv
, gfp_mask
);
843 * Allocation failed presumably due to memory. Undo anything
844 * we might have messed up.
846 * Allocating task should really be put onto the front of the
847 * wait queue, but this is pretty rare.
849 spin_lock_irq(q
->queue_lock
);
850 freed_request(q
, is_sync
, priv
);
853 * in the very unlikely event that allocation failed and no
854 * requests for this direction was pending, mark us starved
855 * so that freeing of a request in the other direction will
856 * notice us. another possible fix would be to split the
857 * rq mempool into READ and WRITE
860 if (unlikely(rl
->count
[is_sync
] == 0))
861 rl
->starved
[is_sync
] = 1;
867 * ioc may be NULL here, and ioc_batching will be false. That's
868 * OK, if the queue is under the request limit then requests need
869 * not count toward the nr_batch_requests limit. There will always
870 * be some limit enforced by BLK_BATCH_TIME.
872 if (ioc_batching(q
, ioc
))
873 ioc
->nr_batch_requests
--;
875 trace_block_getrq(q
, bio
, rw_flags
& 1);
881 * No available requests for this queue, unplug the device and wait for some
882 * requests to become available.
884 * Called with q->queue_lock held, and returns with it unlocked.
886 static struct request
*get_request_wait(struct request_queue
*q
, int rw_flags
,
889 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
892 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
895 struct io_context
*ioc
;
896 struct request_list
*rl
= &q
->rq
;
898 prepare_to_wait_exclusive(&rl
->wait
[is_sync
], &wait
,
899 TASK_UNINTERRUPTIBLE
);
901 trace_block_sleeprq(q
, bio
, rw_flags
& 1);
903 __generic_unplug_device(q
);
904 spin_unlock_irq(q
->queue_lock
);
908 * After sleeping, we become a "batching" process and
909 * will be able to allocate at least one request, and
910 * up to a big batch of them for a small period time.
911 * See ioc_batching, ioc_set_batching
913 ioc
= current_io_context(GFP_NOIO
, q
->node
);
914 ioc_set_batching(q
, ioc
);
916 spin_lock_irq(q
->queue_lock
);
917 finish_wait(&rl
->wait
[is_sync
], &wait
);
919 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
925 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
929 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
931 spin_lock_irq(q
->queue_lock
);
932 if (gfp_mask
& __GFP_WAIT
) {
933 rq
= get_request_wait(q
, rw
, NULL
);
935 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
937 spin_unlock_irq(q
->queue_lock
);
939 /* q->queue_lock is unlocked at this point */
943 EXPORT_SYMBOL(blk_get_request
);
946 * blk_make_request - given a bio, allocate a corresponding struct request.
947 * @q: target request queue
948 * @bio: The bio describing the memory mappings that will be submitted for IO.
949 * It may be a chained-bio properly constructed by block/bio layer.
950 * @gfp_mask: gfp flags to be used for memory allocation
952 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
953 * type commands. Where the struct request needs to be farther initialized by
954 * the caller. It is passed a &struct bio, which describes the memory info of
957 * The caller of blk_make_request must make sure that bi_io_vec
958 * are set to describe the memory buffers. That bio_data_dir() will return
959 * the needed direction of the request. (And all bio's in the passed bio-chain
960 * are properly set accordingly)
962 * If called under none-sleepable conditions, mapped bio buffers must not
963 * need bouncing, by calling the appropriate masked or flagged allocator,
964 * suitable for the target device. Otherwise the call to blk_queue_bounce will
967 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
968 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
969 * anything but the first bio in the chain. Otherwise you risk waiting for IO
970 * completion of a bio that hasn't been submitted yet, thus resulting in a
971 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
972 * of bio_alloc(), as that avoids the mempool deadlock.
973 * If possible a big IO should be split into smaller parts when allocation
974 * fails. Partial allocation should not be an error, or you risk a live-lock.
976 struct request
*blk_make_request(struct request_queue
*q
, struct bio
*bio
,
979 struct request
*rq
= blk_get_request(q
, bio_data_dir(bio
), gfp_mask
);
982 return ERR_PTR(-ENOMEM
);
985 struct bio
*bounce_bio
= bio
;
988 blk_queue_bounce(q
, &bounce_bio
);
989 ret
= blk_rq_append_bio(q
, rq
, bounce_bio
);
998 EXPORT_SYMBOL(blk_make_request
);
1001 * blk_requeue_request - put a request back on queue
1002 * @q: request queue where request should be inserted
1003 * @rq: request to be inserted
1006 * Drivers often keep queueing requests until the hardware cannot accept
1007 * more, when that condition happens we need to put the request back
1008 * on the queue. Must be called with queue lock held.
1010 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
1012 blk_delete_timer(rq
);
1013 blk_clear_rq_complete(rq
);
1014 trace_block_rq_requeue(q
, rq
);
1016 if (blk_rq_tagged(rq
))
1017 blk_queue_end_tag(q
, rq
);
1019 BUG_ON(blk_queued_rq(rq
));
1021 elv_requeue_request(q
, rq
);
1023 EXPORT_SYMBOL(blk_requeue_request
);
1026 * blk_insert_request - insert a special request into a request queue
1027 * @q: request queue where request should be inserted
1028 * @rq: request to be inserted
1029 * @at_head: insert request at head or tail of queue
1030 * @data: private data
1033 * Many block devices need to execute commands asynchronously, so they don't
1034 * block the whole kernel from preemption during request execution. This is
1035 * accomplished normally by inserting aritficial requests tagged as
1036 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
1037 * be scheduled for actual execution by the request queue.
1039 * We have the option of inserting the head or the tail of the queue.
1040 * Typically we use the tail for new ioctls and so forth. We use the head
1041 * of the queue for things like a QUEUE_FULL message from a device, or a
1042 * host that is unable to accept a particular command.
1044 void blk_insert_request(struct request_queue
*q
, struct request
*rq
,
1045 int at_head
, void *data
)
1047 int where
= at_head
? ELEVATOR_INSERT_FRONT
: ELEVATOR_INSERT_BACK
;
1048 unsigned long flags
;
1051 * tell I/O scheduler that this isn't a regular read/write (ie it
1052 * must not attempt merges on this) and that it acts as a soft
1055 rq
->cmd_type
= REQ_TYPE_SPECIAL
;
1059 spin_lock_irqsave(q
->queue_lock
, flags
);
1062 * If command is tagged, release the tag
1064 if (blk_rq_tagged(rq
))
1065 blk_queue_end_tag(q
, rq
);
1067 drive_stat_acct(rq
, 1);
1068 __elv_add_request(q
, rq
, where
, 0);
1070 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1072 EXPORT_SYMBOL(blk_insert_request
);
1074 static void part_round_stats_single(int cpu
, struct hd_struct
*part
,
1077 if (now
== part
->stamp
)
1080 if (part_in_flight(part
)) {
1081 __part_stat_add(cpu
, part
, time_in_queue
,
1082 part_in_flight(part
) * (now
- part
->stamp
));
1083 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
1089 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1090 * @cpu: cpu number for stats access
1091 * @part: target partition
1093 * The average IO queue length and utilisation statistics are maintained
1094 * by observing the current state of the queue length and the amount of
1095 * time it has been in this state for.
1097 * Normally, that accounting is done on IO completion, but that can result
1098 * in more than a second's worth of IO being accounted for within any one
1099 * second, leading to >100% utilisation. To deal with that, we call this
1100 * function to do a round-off before returning the results when reading
1101 * /proc/diskstats. This accounts immediately for all queue usage up to
1102 * the current jiffies and restarts the counters again.
1104 void part_round_stats(int cpu
, struct hd_struct
*part
)
1106 unsigned long now
= jiffies
;
1109 part_round_stats_single(cpu
, &part_to_disk(part
)->part0
, now
);
1110 part_round_stats_single(cpu
, part
, now
);
1112 EXPORT_SYMBOL_GPL(part_round_stats
);
1115 * queue lock must be held
1117 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1121 if (unlikely(--req
->ref_count
))
1124 elv_completed_request(q
, req
);
1126 /* this is a bio leak */
1127 WARN_ON(req
->bio
!= NULL
);
1130 * Request may not have originated from ll_rw_blk. if not,
1131 * it didn't come out of our reserved rq pools
1133 if (req
->cmd_flags
& REQ_ALLOCED
) {
1134 int is_sync
= rq_is_sync(req
) != 0;
1135 int priv
= req
->cmd_flags
& REQ_ELVPRIV
;
1137 BUG_ON(!list_empty(&req
->queuelist
));
1138 BUG_ON(!hlist_unhashed(&req
->hash
));
1140 blk_free_request(q
, req
);
1141 freed_request(q
, is_sync
, priv
);
1144 EXPORT_SYMBOL_GPL(__blk_put_request
);
1146 void blk_put_request(struct request
*req
)
1148 unsigned long flags
;
1149 struct request_queue
*q
= req
->q
;
1151 spin_lock_irqsave(q
->queue_lock
, flags
);
1152 __blk_put_request(q
, req
);
1153 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1155 EXPORT_SYMBOL(blk_put_request
);
1158 * blk_add_request_payload - add a payload to a request
1159 * @rq: request to update
1160 * @page: page backing the payload
1161 * @len: length of the payload.
1163 * This allows to later add a payload to an already submitted request by
1164 * a block driver. The driver needs to take care of freeing the payload
1167 * Note that this is a quite horrible hack and nothing but handling of
1168 * discard requests should ever use it.
1170 void blk_add_request_payload(struct request
*rq
, struct page
*page
,
1173 struct bio
*bio
= rq
->bio
;
1175 bio
->bi_io_vec
->bv_page
= page
;
1176 bio
->bi_io_vec
->bv_offset
= 0;
1177 bio
->bi_io_vec
->bv_len
= len
;
1181 bio
->bi_phys_segments
= 1;
1183 rq
->__data_len
= rq
->resid_len
= len
;
1184 rq
->nr_phys_segments
= 1;
1185 rq
->buffer
= bio_data(bio
);
1187 EXPORT_SYMBOL_GPL(blk_add_request_payload
);
1189 void init_request_from_bio(struct request
*req
, struct bio
*bio
)
1191 req
->cpu
= bio
->bi_comp_cpu
;
1192 req
->cmd_type
= REQ_TYPE_FS
;
1194 req
->cmd_flags
|= bio
->bi_rw
& REQ_COMMON_MASK
;
1195 if (bio
->bi_rw
& REQ_RAHEAD
)
1196 req
->cmd_flags
|= REQ_FAILFAST_MASK
;
1199 req
->__sector
= bio
->bi_sector
;
1200 req
->ioprio
= bio_prio(bio
);
1201 blk_rq_bio_prep(req
->q
, req
, bio
);
1205 * Only disabling plugging for non-rotational devices if it does tagging
1206 * as well, otherwise we do need the proper merging
1208 static inline bool queue_should_plug(struct request_queue
*q
)
1210 return !(blk_queue_nonrot(q
) && blk_queue_tagged(q
));
1213 static int __make_request(struct request_queue
*q
, struct bio
*bio
)
1215 struct request
*req
;
1217 unsigned int bytes
= bio
->bi_size
;
1218 const unsigned short prio
= bio_prio(bio
);
1219 const bool sync
= !!(bio
->bi_rw
& REQ_SYNC
);
1220 const bool unplug
= !!(bio
->bi_rw
& REQ_UNPLUG
);
1221 const unsigned long ff
= bio
->bi_rw
& REQ_FAILFAST_MASK
;
1222 int where
= ELEVATOR_INSERT_SORT
;
1226 * low level driver can indicate that it wants pages above a
1227 * certain limit bounced to low memory (ie for highmem, or even
1228 * ISA dma in theory)
1230 blk_queue_bounce(q
, &bio
);
1232 spin_lock_irq(q
->queue_lock
);
1234 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) {
1235 where
= ELEVATOR_INSERT_FLUSH
;
1239 if (elv_queue_empty(q
))
1242 el_ret
= elv_merge(q
, &req
, bio
);
1244 case ELEVATOR_BACK_MERGE
:
1245 BUG_ON(!rq_mergeable(req
));
1247 if (!ll_back_merge_fn(q
, req
, bio
))
1250 trace_block_bio_backmerge(q
, bio
);
1252 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1253 blk_rq_set_mixed_merge(req
);
1255 req
->biotail
->bi_next
= bio
;
1257 req
->__data_len
+= bytes
;
1258 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
1259 if (!blk_rq_cpu_valid(req
))
1260 req
->cpu
= bio
->bi_comp_cpu
;
1261 drive_stat_acct(req
, 0);
1262 elv_bio_merged(q
, req
, bio
);
1263 if (!attempt_back_merge(q
, req
))
1264 elv_merged_request(q
, req
, el_ret
);
1267 case ELEVATOR_FRONT_MERGE
:
1268 BUG_ON(!rq_mergeable(req
));
1270 if (!ll_front_merge_fn(q
, req
, bio
))
1273 trace_block_bio_frontmerge(q
, bio
);
1275 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
) {
1276 blk_rq_set_mixed_merge(req
);
1277 req
->cmd_flags
&= ~REQ_FAILFAST_MASK
;
1278 req
->cmd_flags
|= ff
;
1281 bio
->bi_next
= req
->bio
;
1285 * may not be valid. if the low level driver said
1286 * it didn't need a bounce buffer then it better
1287 * not touch req->buffer either...
1289 req
->buffer
= bio_data(bio
);
1290 req
->__sector
= bio
->bi_sector
;
1291 req
->__data_len
+= bytes
;
1292 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
1293 if (!blk_rq_cpu_valid(req
))
1294 req
->cpu
= bio
->bi_comp_cpu
;
1295 drive_stat_acct(req
, 0);
1296 elv_bio_merged(q
, req
, bio
);
1297 if (!attempt_front_merge(q
, req
))
1298 elv_merged_request(q
, req
, el_ret
);
1301 /* ELV_NO_MERGE: elevator says don't/can't merge. */
1308 * This sync check and mask will be re-done in init_request_from_bio(),
1309 * but we need to set it earlier to expose the sync flag to the
1310 * rq allocator and io schedulers.
1312 rw_flags
= bio_data_dir(bio
);
1314 rw_flags
|= REQ_SYNC
;
1317 * Grab a free request. This is might sleep but can not fail.
1318 * Returns with the queue unlocked.
1320 req
= get_request_wait(q
, rw_flags
, bio
);
1323 * After dropping the lock and possibly sleeping here, our request
1324 * may now be mergeable after it had proven unmergeable (above).
1325 * We don't worry about that case for efficiency. It won't happen
1326 * often, and the elevators are able to handle it.
1328 init_request_from_bio(req
, bio
);
1330 spin_lock_irq(q
->queue_lock
);
1331 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
) ||
1332 bio_flagged(bio
, BIO_CPU_AFFINE
))
1333 req
->cpu
= blk_cpu_to_group(smp_processor_id());
1334 if (queue_should_plug(q
) && elv_queue_empty(q
))
1337 /* insert the request into the elevator */
1338 drive_stat_acct(req
, 1);
1339 __elv_add_request(q
, req
, where
, 0);
1341 if (unplug
|| !queue_should_plug(q
))
1342 __generic_unplug_device(q
);
1343 spin_unlock_irq(q
->queue_lock
);
1348 * If bio->bi_dev is a partition, remap the location
1350 static inline void blk_partition_remap(struct bio
*bio
)
1352 struct block_device
*bdev
= bio
->bi_bdev
;
1354 if (bio_sectors(bio
) && bdev
!= bdev
->bd_contains
) {
1355 struct hd_struct
*p
= bdev
->bd_part
;
1357 bio
->bi_sector
+= p
->start_sect
;
1358 bio
->bi_bdev
= bdev
->bd_contains
;
1360 trace_block_bio_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
1362 bio
->bi_sector
- p
->start_sect
);
1366 static void handle_bad_sector(struct bio
*bio
)
1368 char b
[BDEVNAME_SIZE
];
1370 printk(KERN_INFO
"attempt to access beyond end of device\n");
1371 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
1372 bdevname(bio
->bi_bdev
, b
),
1374 (unsigned long long)bio
->bi_sector
+ bio_sectors(bio
),
1375 (long long)(i_size_read(bio
->bi_bdev
->bd_inode
) >> 9));
1377 set_bit(BIO_EOF
, &bio
->bi_flags
);
1380 #ifdef CONFIG_FAIL_MAKE_REQUEST
1382 static DECLARE_FAULT_ATTR(fail_make_request
);
1384 static int __init
setup_fail_make_request(char *str
)
1386 return setup_fault_attr(&fail_make_request
, str
);
1388 __setup("fail_make_request=", setup_fail_make_request
);
1390 static int should_fail_request(struct bio
*bio
)
1392 struct hd_struct
*part
= bio
->bi_bdev
->bd_part
;
1394 if (part_to_disk(part
)->part0
.make_it_fail
|| part
->make_it_fail
)
1395 return should_fail(&fail_make_request
, bio
->bi_size
);
1400 static int __init
fail_make_request_debugfs(void)
1402 return init_fault_attr_dentries(&fail_make_request
,
1403 "fail_make_request");
1406 late_initcall(fail_make_request_debugfs
);
1408 #else /* CONFIG_FAIL_MAKE_REQUEST */
1410 static inline int should_fail_request(struct bio
*bio
)
1415 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1418 * Check whether this bio extends beyond the end of the device.
1420 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
1427 /* Test device or partition size, when known. */
1428 maxsector
= i_size_read(bio
->bi_bdev
->bd_inode
) >> 9;
1430 sector_t sector
= bio
->bi_sector
;
1432 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
1434 * This may well happen - the kernel calls bread()
1435 * without checking the size of the device, e.g., when
1436 * mounting a device.
1438 handle_bad_sector(bio
);
1447 * generic_make_request - hand a buffer to its device driver for I/O
1448 * @bio: The bio describing the location in memory and on the device.
1450 * generic_make_request() is used to make I/O requests of block
1451 * devices. It is passed a &struct bio, which describes the I/O that needs
1454 * generic_make_request() does not return any status. The
1455 * success/failure status of the request, along with notification of
1456 * completion, is delivered asynchronously through the bio->bi_end_io
1457 * function described (one day) else where.
1459 * The caller of generic_make_request must make sure that bi_io_vec
1460 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1461 * set to describe the device address, and the
1462 * bi_end_io and optionally bi_private are set to describe how
1463 * completion notification should be signaled.
1465 * generic_make_request and the drivers it calls may use bi_next if this
1466 * bio happens to be merged with someone else, and may change bi_dev and
1467 * bi_sector for remaps as it sees fit. So the values of these fields
1468 * should NOT be depended on after the call to generic_make_request.
1470 static inline void __generic_make_request(struct bio
*bio
)
1472 struct request_queue
*q
;
1473 sector_t old_sector
;
1474 int ret
, nr_sectors
= bio_sectors(bio
);
1480 if (bio_check_eod(bio
, nr_sectors
))
1484 * Resolve the mapping until finished. (drivers are
1485 * still free to implement/resolve their own stacking
1486 * by explicitly returning 0)
1488 * NOTE: we don't repeat the blk_size check for each new device.
1489 * Stacking drivers are expected to know what they are doing.
1494 char b
[BDEVNAME_SIZE
];
1496 q
= bdev_get_queue(bio
->bi_bdev
);
1499 "generic_make_request: Trying to access "
1500 "nonexistent block-device %s (%Lu)\n",
1501 bdevname(bio
->bi_bdev
, b
),
1502 (long long) bio
->bi_sector
);
1506 if (unlikely(!(bio
->bi_rw
& REQ_DISCARD
) &&
1507 nr_sectors
> queue_max_hw_sectors(q
))) {
1508 printk(KERN_ERR
"bio too big device %s (%u > %u)\n",
1509 bdevname(bio
->bi_bdev
, b
),
1511 queue_max_hw_sectors(q
));
1515 if (unlikely(test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)))
1518 if (should_fail_request(bio
))
1522 * If this device has partitions, remap block n
1523 * of partition p to block n+start(p) of the disk.
1525 blk_partition_remap(bio
);
1527 if (bio_integrity_enabled(bio
) && bio_integrity_prep(bio
))
1530 if (old_sector
!= -1)
1531 trace_block_bio_remap(q
, bio
, old_dev
, old_sector
);
1533 old_sector
= bio
->bi_sector
;
1534 old_dev
= bio
->bi_bdev
->bd_dev
;
1536 if (bio_check_eod(bio
, nr_sectors
))
1540 * Filter flush bio's early so that make_request based
1541 * drivers without flush support don't have to worry
1544 if ((bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) && !q
->flush_flags
) {
1545 bio
->bi_rw
&= ~(REQ_FLUSH
| REQ_FUA
);
1552 if ((bio
->bi_rw
& REQ_DISCARD
) &&
1553 (!blk_queue_discard(q
) ||
1554 ((bio
->bi_rw
& REQ_SECURE
) &&
1555 !blk_queue_secdiscard(q
)))) {
1560 blk_throtl_bio(q
, &bio
);
1563 * If bio = NULL, bio has been throttled and will be submitted
1569 trace_block_bio_queue(q
, bio
);
1571 ret
= q
->make_request_fn(q
, bio
);
1577 bio_endio(bio
, err
);
1581 * We only want one ->make_request_fn to be active at a time,
1582 * else stack usage with stacked devices could be a problem.
1583 * So use current->bio_list to keep a list of requests
1584 * submited by a make_request_fn function.
1585 * current->bio_list is also used as a flag to say if
1586 * generic_make_request is currently active in this task or not.
1587 * If it is NULL, then no make_request is active. If it is non-NULL,
1588 * then a make_request is active, and new requests should be added
1591 void generic_make_request(struct bio
*bio
)
1593 struct bio_list bio_list_on_stack
;
1595 if (current
->bio_list
) {
1596 /* make_request is active */
1597 bio_list_add(current
->bio_list
, bio
);
1600 /* following loop may be a bit non-obvious, and so deserves some
1602 * Before entering the loop, bio->bi_next is NULL (as all callers
1603 * ensure that) so we have a list with a single bio.
1604 * We pretend that we have just taken it off a longer list, so
1605 * we assign bio_list to a pointer to the bio_list_on_stack,
1606 * thus initialising the bio_list of new bios to be
1607 * added. __generic_make_request may indeed add some more bios
1608 * through a recursive call to generic_make_request. If it
1609 * did, we find a non-NULL value in bio_list and re-enter the loop
1610 * from the top. In this case we really did just take the bio
1611 * of the top of the list (no pretending) and so remove it from
1612 * bio_list, and call into __generic_make_request again.
1614 * The loop was structured like this to make only one call to
1615 * __generic_make_request (which is important as it is large and
1616 * inlined) and to keep the structure simple.
1618 BUG_ON(bio
->bi_next
);
1619 bio_list_init(&bio_list_on_stack
);
1620 current
->bio_list
= &bio_list_on_stack
;
1622 __generic_make_request(bio
);
1623 bio
= bio_list_pop(current
->bio_list
);
1625 current
->bio_list
= NULL
; /* deactivate */
1627 EXPORT_SYMBOL(generic_make_request
);
1630 * submit_bio - submit a bio to the block device layer for I/O
1631 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1632 * @bio: The &struct bio which describes the I/O
1634 * submit_bio() is very similar in purpose to generic_make_request(), and
1635 * uses that function to do most of the work. Both are fairly rough
1636 * interfaces; @bio must be presetup and ready for I/O.
1639 void submit_bio(int rw
, struct bio
*bio
)
1641 int count
= bio_sectors(bio
);
1646 * If it's a regular read/write or a barrier with data attached,
1647 * go through the normal accounting stuff before submission.
1649 if (bio_has_data(bio
) && !(rw
& REQ_DISCARD
)) {
1651 count_vm_events(PGPGOUT
, count
);
1653 task_io_account_read(bio
->bi_size
);
1654 count_vm_events(PGPGIN
, count
);
1657 if (unlikely(block_dump
)) {
1658 char b
[BDEVNAME_SIZE
];
1659 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s (%u sectors)\n",
1660 current
->comm
, task_pid_nr(current
),
1661 (rw
& WRITE
) ? "WRITE" : "READ",
1662 (unsigned long long)bio
->bi_sector
,
1663 bdevname(bio
->bi_bdev
, b
),
1668 generic_make_request(bio
);
1670 EXPORT_SYMBOL(submit_bio
);
1673 * blk_rq_check_limits - Helper function to check a request for the queue limit
1675 * @rq: the request being checked
1678 * @rq may have been made based on weaker limitations of upper-level queues
1679 * in request stacking drivers, and it may violate the limitation of @q.
1680 * Since the block layer and the underlying device driver trust @rq
1681 * after it is inserted to @q, it should be checked against @q before
1682 * the insertion using this generic function.
1684 * This function should also be useful for request stacking drivers
1685 * in some cases below, so export this function.
1686 * Request stacking drivers like request-based dm may change the queue
1687 * limits while requests are in the queue (e.g. dm's table swapping).
1688 * Such request stacking drivers should check those requests agaist
1689 * the new queue limits again when they dispatch those requests,
1690 * although such checkings are also done against the old queue limits
1691 * when submitting requests.
1693 int blk_rq_check_limits(struct request_queue
*q
, struct request
*rq
)
1695 if (rq
->cmd_flags
& REQ_DISCARD
)
1698 if (blk_rq_sectors(rq
) > queue_max_sectors(q
) ||
1699 blk_rq_bytes(rq
) > queue_max_hw_sectors(q
) << 9) {
1700 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
1705 * queue's settings related to segment counting like q->bounce_pfn
1706 * may differ from that of other stacking queues.
1707 * Recalculate it to check the request correctly on this queue's
1710 blk_recalc_rq_segments(rq
);
1711 if (rq
->nr_phys_segments
> queue_max_segments(q
)) {
1712 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
1718 EXPORT_SYMBOL_GPL(blk_rq_check_limits
);
1721 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1722 * @q: the queue to submit the request
1723 * @rq: the request being queued
1725 int blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
1727 unsigned long flags
;
1729 if (blk_rq_check_limits(q
, rq
))
1732 #ifdef CONFIG_FAIL_MAKE_REQUEST
1733 if (rq
->rq_disk
&& rq
->rq_disk
->part0
.make_it_fail
&&
1734 should_fail(&fail_make_request
, blk_rq_bytes(rq
)))
1738 spin_lock_irqsave(q
->queue_lock
, flags
);
1741 * Submitting request must be dequeued before calling this function
1742 * because it will be linked to another request_queue
1744 BUG_ON(blk_queued_rq(rq
));
1746 drive_stat_acct(rq
, 1);
1747 __elv_add_request(q
, rq
, ELEVATOR_INSERT_BACK
, 0);
1749 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1753 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
1756 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1757 * @rq: request to examine
1760 * A request could be merge of IOs which require different failure
1761 * handling. This function determines the number of bytes which
1762 * can be failed from the beginning of the request without
1763 * crossing into area which need to be retried further.
1766 * The number of bytes to fail.
1769 * queue_lock must be held.
1771 unsigned int blk_rq_err_bytes(const struct request
*rq
)
1773 unsigned int ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
1774 unsigned int bytes
= 0;
1777 if (!(rq
->cmd_flags
& REQ_MIXED_MERGE
))
1778 return blk_rq_bytes(rq
);
1781 * Currently the only 'mixing' which can happen is between
1782 * different fastfail types. We can safely fail portions
1783 * which have all the failfast bits that the first one has -
1784 * the ones which are at least as eager to fail as the first
1787 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
1788 if ((bio
->bi_rw
& ff
) != ff
)
1790 bytes
+= bio
->bi_size
;
1793 /* this could lead to infinite loop */
1794 BUG_ON(blk_rq_bytes(rq
) && !bytes
);
1797 EXPORT_SYMBOL_GPL(blk_rq_err_bytes
);
1799 static void blk_account_io_completion(struct request
*req
, unsigned int bytes
)
1801 if (blk_do_io_stat(req
)) {
1802 const int rw
= rq_data_dir(req
);
1803 struct hd_struct
*part
;
1806 cpu
= part_stat_lock();
1808 part_stat_add(cpu
, part
, sectors
[rw
], bytes
>> 9);
1813 static void blk_account_io_done(struct request
*req
)
1816 * Account IO completion. flush_rq isn't accounted as a
1817 * normal IO on queueing nor completion. Accounting the
1818 * containing request is enough.
1820 if (blk_do_io_stat(req
) && !(req
->cmd_flags
& REQ_FLUSH_SEQ
)) {
1821 unsigned long duration
= jiffies
- req
->start_time
;
1822 const int rw
= rq_data_dir(req
);
1823 struct hd_struct
*part
;
1826 cpu
= part_stat_lock();
1829 part_stat_inc(cpu
, part
, ios
[rw
]);
1830 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
1831 part_round_stats(cpu
, part
);
1832 part_dec_in_flight(part
, rw
);
1834 hd_struct_put(part
);
1840 * blk_peek_request - peek at the top of a request queue
1841 * @q: request queue to peek at
1844 * Return the request at the top of @q. The returned request
1845 * should be started using blk_start_request() before LLD starts
1849 * Pointer to the request at the top of @q if available. Null
1853 * queue_lock must be held.
1855 struct request
*blk_peek_request(struct request_queue
*q
)
1860 while ((rq
= __elv_next_request(q
)) != NULL
) {
1861 if (!(rq
->cmd_flags
& REQ_STARTED
)) {
1863 * This is the first time the device driver
1864 * sees this request (possibly after
1865 * requeueing). Notify IO scheduler.
1867 if (rq
->cmd_flags
& REQ_SORTED
)
1868 elv_activate_rq(q
, rq
);
1871 * just mark as started even if we don't start
1872 * it, a request that has been delayed should
1873 * not be passed by new incoming requests
1875 rq
->cmd_flags
|= REQ_STARTED
;
1876 trace_block_rq_issue(q
, rq
);
1879 if (!q
->boundary_rq
|| q
->boundary_rq
== rq
) {
1880 q
->end_sector
= rq_end_sector(rq
);
1881 q
->boundary_rq
= NULL
;
1884 if (rq
->cmd_flags
& REQ_DONTPREP
)
1887 if (q
->dma_drain_size
&& blk_rq_bytes(rq
)) {
1889 * make sure space for the drain appears we
1890 * know we can do this because max_hw_segments
1891 * has been adjusted to be one fewer than the
1894 rq
->nr_phys_segments
++;
1900 ret
= q
->prep_rq_fn(q
, rq
);
1901 if (ret
== BLKPREP_OK
) {
1903 } else if (ret
== BLKPREP_DEFER
) {
1905 * the request may have been (partially) prepped.
1906 * we need to keep this request in the front to
1907 * avoid resource deadlock. REQ_STARTED will
1908 * prevent other fs requests from passing this one.
1910 if (q
->dma_drain_size
&& blk_rq_bytes(rq
) &&
1911 !(rq
->cmd_flags
& REQ_DONTPREP
)) {
1913 * remove the space for the drain we added
1914 * so that we don't add it again
1916 --rq
->nr_phys_segments
;
1921 } else if (ret
== BLKPREP_KILL
) {
1922 rq
->cmd_flags
|= REQ_QUIET
;
1924 * Mark this request as started so we don't trigger
1925 * any debug logic in the end I/O path.
1927 blk_start_request(rq
);
1928 __blk_end_request_all(rq
, -EIO
);
1930 printk(KERN_ERR
"%s: bad return=%d\n", __func__
, ret
);
1937 EXPORT_SYMBOL(blk_peek_request
);
1939 void blk_dequeue_request(struct request
*rq
)
1941 struct request_queue
*q
= rq
->q
;
1943 BUG_ON(list_empty(&rq
->queuelist
));
1944 BUG_ON(ELV_ON_HASH(rq
));
1946 list_del_init(&rq
->queuelist
);
1949 * the time frame between a request being removed from the lists
1950 * and to it is freed is accounted as io that is in progress at
1953 if (blk_account_rq(rq
)) {
1954 q
->in_flight
[rq_is_sync(rq
)]++;
1955 set_io_start_time_ns(rq
);
1960 * blk_start_request - start request processing on the driver
1961 * @req: request to dequeue
1964 * Dequeue @req and start timeout timer on it. This hands off the
1965 * request to the driver.
1967 * Block internal functions which don't want to start timer should
1968 * call blk_dequeue_request().
1971 * queue_lock must be held.
1973 void blk_start_request(struct request
*req
)
1975 blk_dequeue_request(req
);
1978 * We are now handing the request to the hardware, initialize
1979 * resid_len to full count and add the timeout handler.
1981 req
->resid_len
= blk_rq_bytes(req
);
1982 if (unlikely(blk_bidi_rq(req
)))
1983 req
->next_rq
->resid_len
= blk_rq_bytes(req
->next_rq
);
1987 EXPORT_SYMBOL(blk_start_request
);
1990 * blk_fetch_request - fetch a request from a request queue
1991 * @q: request queue to fetch a request from
1994 * Return the request at the top of @q. The request is started on
1995 * return and LLD can start processing it immediately.
1998 * Pointer to the request at the top of @q if available. Null
2002 * queue_lock must be held.
2004 struct request
*blk_fetch_request(struct request_queue
*q
)
2008 rq
= blk_peek_request(q
);
2010 blk_start_request(rq
);
2013 EXPORT_SYMBOL(blk_fetch_request
);
2016 * blk_update_request - Special helper function for request stacking drivers
2017 * @req: the request being processed
2018 * @error: %0 for success, < %0 for error
2019 * @nr_bytes: number of bytes to complete @req
2022 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2023 * the request structure even if @req doesn't have leftover.
2024 * If @req has leftover, sets it up for the next range of segments.
2026 * This special helper function is only for request stacking drivers
2027 * (e.g. request-based dm) so that they can handle partial completion.
2028 * Actual device drivers should use blk_end_request instead.
2030 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2031 * %false return from this function.
2034 * %false - this request doesn't have any more data
2035 * %true - this request has more data
2037 bool blk_update_request(struct request
*req
, int error
, unsigned int nr_bytes
)
2039 int total_bytes
, bio_nbytes
, next_idx
= 0;
2045 trace_block_rq_complete(req
->q
, req
);
2048 * For fs requests, rq is just carrier of independent bio's
2049 * and each partial completion should be handled separately.
2050 * Reset per-request error on each partial completion.
2052 * TODO: tj: This is too subtle. It would be better to let
2053 * low level drivers do what they see fit.
2055 if (req
->cmd_type
== REQ_TYPE_FS
)
2058 if (error
&& req
->cmd_type
== REQ_TYPE_FS
&&
2059 !(req
->cmd_flags
& REQ_QUIET
)) {
2060 printk(KERN_ERR
"end_request: I/O error, dev %s, sector %llu\n",
2061 req
->rq_disk
? req
->rq_disk
->disk_name
: "?",
2062 (unsigned long long)blk_rq_pos(req
));
2065 blk_account_io_completion(req
, nr_bytes
);
2067 total_bytes
= bio_nbytes
= 0;
2068 while ((bio
= req
->bio
) != NULL
) {
2071 if (nr_bytes
>= bio
->bi_size
) {
2072 req
->bio
= bio
->bi_next
;
2073 nbytes
= bio
->bi_size
;
2074 req_bio_endio(req
, bio
, nbytes
, error
);
2078 int idx
= bio
->bi_idx
+ next_idx
;
2080 if (unlikely(idx
>= bio
->bi_vcnt
)) {
2081 blk_dump_rq_flags(req
, "__end_that");
2082 printk(KERN_ERR
"%s: bio idx %d >= vcnt %d\n",
2083 __func__
, idx
, bio
->bi_vcnt
);
2087 nbytes
= bio_iovec_idx(bio
, idx
)->bv_len
;
2088 BIO_BUG_ON(nbytes
> bio
->bi_size
);
2091 * not a complete bvec done
2093 if (unlikely(nbytes
> nr_bytes
)) {
2094 bio_nbytes
+= nr_bytes
;
2095 total_bytes
+= nr_bytes
;
2100 * advance to the next vector
2103 bio_nbytes
+= nbytes
;
2106 total_bytes
+= nbytes
;
2112 * end more in this run, or just return 'not-done'
2114 if (unlikely(nr_bytes
<= 0))
2124 * Reset counters so that the request stacking driver
2125 * can find how many bytes remain in the request
2128 req
->__data_len
= 0;
2133 * if the request wasn't completed, update state
2136 req_bio_endio(req
, bio
, bio_nbytes
, error
);
2137 bio
->bi_idx
+= next_idx
;
2138 bio_iovec(bio
)->bv_offset
+= nr_bytes
;
2139 bio_iovec(bio
)->bv_len
-= nr_bytes
;
2142 req
->__data_len
-= total_bytes
;
2143 req
->buffer
= bio_data(req
->bio
);
2145 /* update sector only for requests with clear definition of sector */
2146 if (req
->cmd_type
== REQ_TYPE_FS
|| (req
->cmd_flags
& REQ_DISCARD
))
2147 req
->__sector
+= total_bytes
>> 9;
2149 /* mixed attributes always follow the first bio */
2150 if (req
->cmd_flags
& REQ_MIXED_MERGE
) {
2151 req
->cmd_flags
&= ~REQ_FAILFAST_MASK
;
2152 req
->cmd_flags
|= req
->bio
->bi_rw
& REQ_FAILFAST_MASK
;
2156 * If total number of sectors is less than the first segment
2157 * size, something has gone terribly wrong.
2159 if (blk_rq_bytes(req
) < blk_rq_cur_bytes(req
)) {
2160 printk(KERN_ERR
"blk: request botched\n");
2161 req
->__data_len
= blk_rq_cur_bytes(req
);
2164 /* recalculate the number of segments */
2165 blk_recalc_rq_segments(req
);
2169 EXPORT_SYMBOL_GPL(blk_update_request
);
2171 static bool blk_update_bidi_request(struct request
*rq
, int error
,
2172 unsigned int nr_bytes
,
2173 unsigned int bidi_bytes
)
2175 if (blk_update_request(rq
, error
, nr_bytes
))
2178 /* Bidi request must be completed as a whole */
2179 if (unlikely(blk_bidi_rq(rq
)) &&
2180 blk_update_request(rq
->next_rq
, error
, bidi_bytes
))
2183 if (blk_queue_add_random(rq
->q
))
2184 add_disk_randomness(rq
->rq_disk
);
2190 * blk_unprep_request - unprepare a request
2193 * This function makes a request ready for complete resubmission (or
2194 * completion). It happens only after all error handling is complete,
2195 * so represents the appropriate moment to deallocate any resources
2196 * that were allocated to the request in the prep_rq_fn. The queue
2197 * lock is held when calling this.
2199 void blk_unprep_request(struct request
*req
)
2201 struct request_queue
*q
= req
->q
;
2203 req
->cmd_flags
&= ~REQ_DONTPREP
;
2204 if (q
->unprep_rq_fn
)
2205 q
->unprep_rq_fn(q
, req
);
2207 EXPORT_SYMBOL_GPL(blk_unprep_request
);
2210 * queue lock must be held
2212 static void blk_finish_request(struct request
*req
, int error
)
2214 if (blk_rq_tagged(req
))
2215 blk_queue_end_tag(req
->q
, req
);
2217 BUG_ON(blk_queued_rq(req
));
2219 if (unlikely(laptop_mode
) && req
->cmd_type
== REQ_TYPE_FS
)
2220 laptop_io_completion(&req
->q
->backing_dev_info
);
2222 blk_delete_timer(req
);
2224 if (req
->cmd_flags
& REQ_DONTPREP
)
2225 blk_unprep_request(req
);
2228 blk_account_io_done(req
);
2231 req
->end_io(req
, error
);
2233 if (blk_bidi_rq(req
))
2234 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
2236 __blk_put_request(req
->q
, req
);
2241 * blk_end_bidi_request - Complete a bidi request
2242 * @rq: the request to complete
2243 * @error: %0 for success, < %0 for error
2244 * @nr_bytes: number of bytes to complete @rq
2245 * @bidi_bytes: number of bytes to complete @rq->next_rq
2248 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2249 * Drivers that supports bidi can safely call this member for any
2250 * type of request, bidi or uni. In the later case @bidi_bytes is
2254 * %false - we are done with this request
2255 * %true - still buffers pending for this request
2257 static bool blk_end_bidi_request(struct request
*rq
, int error
,
2258 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2260 struct request_queue
*q
= rq
->q
;
2261 unsigned long flags
;
2263 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2266 spin_lock_irqsave(q
->queue_lock
, flags
);
2267 blk_finish_request(rq
, error
);
2268 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2274 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2275 * @rq: the request to complete
2276 * @error: %0 for success, < %0 for error
2277 * @nr_bytes: number of bytes to complete @rq
2278 * @bidi_bytes: number of bytes to complete @rq->next_rq
2281 * Identical to blk_end_bidi_request() except that queue lock is
2282 * assumed to be locked on entry and remains so on return.
2285 * %false - we are done with this request
2286 * %true - still buffers pending for this request
2288 static bool __blk_end_bidi_request(struct request
*rq
, int error
,
2289 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2291 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2294 blk_finish_request(rq
, error
);
2300 * blk_end_request - Helper function for drivers to complete the request.
2301 * @rq: the request being processed
2302 * @error: %0 for success, < %0 for error
2303 * @nr_bytes: number of bytes to complete
2306 * Ends I/O on a number of bytes attached to @rq.
2307 * If @rq has leftover, sets it up for the next range of segments.
2310 * %false - we are done with this request
2311 * %true - still buffers pending for this request
2313 bool blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2315 return blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2317 EXPORT_SYMBOL(blk_end_request
);
2320 * blk_end_request_all - Helper function for drives to finish the request.
2321 * @rq: the request to finish
2322 * @error: %0 for success, < %0 for error
2325 * Completely finish @rq.
2327 void blk_end_request_all(struct request
*rq
, int error
)
2330 unsigned int bidi_bytes
= 0;
2332 if (unlikely(blk_bidi_rq(rq
)))
2333 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2335 pending
= blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2338 EXPORT_SYMBOL(blk_end_request_all
);
2341 * blk_end_request_cur - Helper function to finish the current request chunk.
2342 * @rq: the request to finish the current chunk for
2343 * @error: %0 for success, < %0 for error
2346 * Complete the current consecutively mapped chunk from @rq.
2349 * %false - we are done with this request
2350 * %true - still buffers pending for this request
2352 bool blk_end_request_cur(struct request
*rq
, int error
)
2354 return blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2356 EXPORT_SYMBOL(blk_end_request_cur
);
2359 * blk_end_request_err - Finish a request till the next failure boundary.
2360 * @rq: the request to finish till the next failure boundary for
2361 * @error: must be negative errno
2364 * Complete @rq till the next failure boundary.
2367 * %false - we are done with this request
2368 * %true - still buffers pending for this request
2370 bool blk_end_request_err(struct request
*rq
, int error
)
2372 WARN_ON(error
>= 0);
2373 return blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2375 EXPORT_SYMBOL_GPL(blk_end_request_err
);
2378 * __blk_end_request - Helper function for drivers to complete the request.
2379 * @rq: the request being processed
2380 * @error: %0 for success, < %0 for error
2381 * @nr_bytes: number of bytes to complete
2384 * Must be called with queue lock held unlike blk_end_request().
2387 * %false - we are done with this request
2388 * %true - still buffers pending for this request
2390 bool __blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2392 return __blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2394 EXPORT_SYMBOL(__blk_end_request
);
2397 * __blk_end_request_all - Helper function for drives to finish the request.
2398 * @rq: the request to finish
2399 * @error: %0 for success, < %0 for error
2402 * Completely finish @rq. Must be called with queue lock held.
2404 void __blk_end_request_all(struct request
*rq
, int error
)
2407 unsigned int bidi_bytes
= 0;
2409 if (unlikely(blk_bidi_rq(rq
)))
2410 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2412 pending
= __blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2415 EXPORT_SYMBOL(__blk_end_request_all
);
2418 * __blk_end_request_cur - Helper function to finish the current request chunk.
2419 * @rq: the request to finish the current chunk for
2420 * @error: %0 for success, < %0 for error
2423 * Complete the current consecutively mapped chunk from @rq. Must
2424 * be called with queue lock held.
2427 * %false - we are done with this request
2428 * %true - still buffers pending for this request
2430 bool __blk_end_request_cur(struct request
*rq
, int error
)
2432 return __blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2434 EXPORT_SYMBOL(__blk_end_request_cur
);
2437 * __blk_end_request_err - Finish a request till the next failure boundary.
2438 * @rq: the request to finish till the next failure boundary for
2439 * @error: must be negative errno
2442 * Complete @rq till the next failure boundary. Must be called
2443 * with queue lock held.
2446 * %false - we are done with this request
2447 * %true - still buffers pending for this request
2449 bool __blk_end_request_err(struct request
*rq
, int error
)
2451 WARN_ON(error
>= 0);
2452 return __blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2454 EXPORT_SYMBOL_GPL(__blk_end_request_err
);
2456 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
2459 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2460 rq
->cmd_flags
|= bio
->bi_rw
& REQ_WRITE
;
2462 if (bio_has_data(bio
)) {
2463 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2464 rq
->buffer
= bio_data(bio
);
2466 rq
->__data_len
= bio
->bi_size
;
2467 rq
->bio
= rq
->biotail
= bio
;
2470 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2473 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2475 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2476 * @rq: the request to be flushed
2479 * Flush all pages in @rq.
2481 void rq_flush_dcache_pages(struct request
*rq
)
2483 struct req_iterator iter
;
2484 struct bio_vec
*bvec
;
2486 rq_for_each_segment(bvec
, rq
, iter
)
2487 flush_dcache_page(bvec
->bv_page
);
2489 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages
);
2493 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2494 * @q : the queue of the device being checked
2497 * Check if underlying low-level drivers of a device are busy.
2498 * If the drivers want to export their busy state, they must set own
2499 * exporting function using blk_queue_lld_busy() first.
2501 * Basically, this function is used only by request stacking drivers
2502 * to stop dispatching requests to underlying devices when underlying
2503 * devices are busy. This behavior helps more I/O merging on the queue
2504 * of the request stacking driver and prevents I/O throughput regression
2505 * on burst I/O load.
2508 * 0 - Not busy (The request stacking driver should dispatch request)
2509 * 1 - Busy (The request stacking driver should stop dispatching request)
2511 int blk_lld_busy(struct request_queue
*q
)
2514 return q
->lld_busy_fn(q
);
2518 EXPORT_SYMBOL_GPL(blk_lld_busy
);
2521 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2522 * @rq: the clone request to be cleaned up
2525 * Free all bios in @rq for a cloned request.
2527 void blk_rq_unprep_clone(struct request
*rq
)
2531 while ((bio
= rq
->bio
) != NULL
) {
2532 rq
->bio
= bio
->bi_next
;
2537 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone
);
2540 * Copy attributes of the original request to the clone request.
2541 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2543 static void __blk_rq_prep_clone(struct request
*dst
, struct request
*src
)
2545 dst
->cpu
= src
->cpu
;
2546 dst
->cmd_flags
= (src
->cmd_flags
& REQ_CLONE_MASK
) | REQ_NOMERGE
;
2547 dst
->cmd_type
= src
->cmd_type
;
2548 dst
->__sector
= blk_rq_pos(src
);
2549 dst
->__data_len
= blk_rq_bytes(src
);
2550 dst
->nr_phys_segments
= src
->nr_phys_segments
;
2551 dst
->ioprio
= src
->ioprio
;
2552 dst
->extra_len
= src
->extra_len
;
2556 * blk_rq_prep_clone - Helper function to setup clone request
2557 * @rq: the request to be setup
2558 * @rq_src: original request to be cloned
2559 * @bs: bio_set that bios for clone are allocated from
2560 * @gfp_mask: memory allocation mask for bio
2561 * @bio_ctr: setup function to be called for each clone bio.
2562 * Returns %0 for success, non %0 for failure.
2563 * @data: private data to be passed to @bio_ctr
2566 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2567 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2568 * are not copied, and copying such parts is the caller's responsibility.
2569 * Also, pages which the original bios are pointing to are not copied
2570 * and the cloned bios just point same pages.
2571 * So cloned bios must be completed before original bios, which means
2572 * the caller must complete @rq before @rq_src.
2574 int blk_rq_prep_clone(struct request
*rq
, struct request
*rq_src
,
2575 struct bio_set
*bs
, gfp_t gfp_mask
,
2576 int (*bio_ctr
)(struct bio
*, struct bio
*, void *),
2579 struct bio
*bio
, *bio_src
;
2584 blk_rq_init(NULL
, rq
);
2586 __rq_for_each_bio(bio_src
, rq_src
) {
2587 bio
= bio_alloc_bioset(gfp_mask
, bio_src
->bi_max_vecs
, bs
);
2591 __bio_clone(bio
, bio_src
);
2593 if (bio_integrity(bio_src
) &&
2594 bio_integrity_clone(bio
, bio_src
, gfp_mask
, bs
))
2597 if (bio_ctr
&& bio_ctr(bio
, bio_src
, data
))
2601 rq
->biotail
->bi_next
= bio
;
2604 rq
->bio
= rq
->biotail
= bio
;
2607 __blk_rq_prep_clone(rq
, rq_src
);
2614 blk_rq_unprep_clone(rq
);
2618 EXPORT_SYMBOL_GPL(blk_rq_prep_clone
);
2620 int kblockd_schedule_work(struct request_queue
*q
, struct work_struct
*work
)
2622 return queue_work(kblockd_workqueue
, work
);
2624 EXPORT_SYMBOL(kblockd_schedule_work
);
2626 int kblockd_schedule_delayed_work(struct request_queue
*q
,
2627 struct delayed_work
*dwork
, unsigned long delay
)
2629 return queue_delayed_work(kblockd_workqueue
, dwork
, delay
);
2631 EXPORT_SYMBOL(kblockd_schedule_delayed_work
);
2633 int __init
blk_dev_init(void)
2635 BUILD_BUG_ON(__REQ_NR_BITS
> 8 *
2636 sizeof(((struct request
*)0)->cmd_flags
));
2638 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
2639 kblockd_workqueue
= alloc_workqueue("kblockd",
2640 WQ_MEM_RECLAIM
| WQ_HIGHPRI
, 0);
2641 if (!kblockd_workqueue
)
2642 panic("Failed to create kblockd\n");
2644 request_cachep
= kmem_cache_create("blkdev_requests",
2645 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
2647 blk_requestq_cachep
= kmem_cache_create("blkdev_queue",
2648 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);