2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
6 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
8 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
12 * This handles all read/write requests to block devices
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/backing-dev.h>
17 #include <linux/bio.h>
18 #include <linux/blkdev.h>
19 #include <linux/highmem.h>
21 #include <linux/kernel_stat.h>
22 #include <linux/string.h>
23 #include <linux/init.h>
24 #include <linux/completion.h>
25 #include <linux/slab.h>
26 #include <linux/swap.h>
27 #include <linux/writeback.h>
28 #include <linux/task_io_accounting_ops.h>
29 #include <linux/fault-inject.h>
30 #include <linux/list_sort.h>
32 #define CREATE_TRACE_POINTS
33 #include <trace/events/block.h>
37 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap
);
38 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap
);
39 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete
);
41 static int __make_request(struct request_queue
*q
, struct bio
*bio
);
44 * For the allocated request tables
46 static struct kmem_cache
*request_cachep
;
49 * For queue allocation
51 struct kmem_cache
*blk_requestq_cachep
;
54 * Controlling structure to kblockd
56 static struct workqueue_struct
*kblockd_workqueue
;
58 static void drive_stat_acct(struct request
*rq
, int new_io
)
60 struct hd_struct
*part
;
61 int rw
= rq_data_dir(rq
);
64 if (!blk_do_io_stat(rq
))
67 cpu
= part_stat_lock();
71 part_stat_inc(cpu
, part
, merges
[rw
]);
73 part
= disk_map_sector_rcu(rq
->rq_disk
, blk_rq_pos(rq
));
74 if (!hd_struct_try_get(part
)) {
76 * The partition is already being removed,
77 * the request will be accounted on the disk only
79 * We take a reference on disk->part0 although that
80 * partition will never be deleted, so we can treat
81 * it as any other partition.
83 part
= &rq
->rq_disk
->part0
;
86 part_round_stats(cpu
, part
);
87 part_inc_in_flight(part
, rw
);
94 void blk_queue_congestion_threshold(struct request_queue
*q
)
98 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) + 1;
99 if (nr
> q
->nr_requests
)
101 q
->nr_congestion_on
= nr
;
103 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) - (q
->nr_requests
/ 16) - 1;
106 q
->nr_congestion_off
= nr
;
110 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
113 * Locates the passed device's request queue and returns the address of its
116 * Will return NULL if the request queue cannot be located.
118 struct backing_dev_info
*blk_get_backing_dev_info(struct block_device
*bdev
)
120 struct backing_dev_info
*ret
= NULL
;
121 struct request_queue
*q
= bdev_get_queue(bdev
);
124 ret
= &q
->backing_dev_info
;
127 EXPORT_SYMBOL(blk_get_backing_dev_info
);
129 void blk_rq_init(struct request_queue
*q
, struct request
*rq
)
131 memset(rq
, 0, sizeof(*rq
));
133 INIT_LIST_HEAD(&rq
->queuelist
);
134 INIT_LIST_HEAD(&rq
->timeout_list
);
137 rq
->__sector
= (sector_t
) -1;
138 INIT_HLIST_NODE(&rq
->hash
);
139 RB_CLEAR_NODE(&rq
->rb_node
);
141 rq
->cmd_len
= BLK_MAX_CDB
;
144 rq
->start_time
= jiffies
;
145 set_start_time_ns(rq
);
148 EXPORT_SYMBOL(blk_rq_init
);
150 static void req_bio_endio(struct request
*rq
, struct bio
*bio
,
151 unsigned int nbytes
, int error
)
154 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
155 else if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
158 if (unlikely(nbytes
> bio
->bi_size
)) {
159 printk(KERN_ERR
"%s: want %u bytes done, %u left\n",
160 __func__
, nbytes
, bio
->bi_size
);
161 nbytes
= bio
->bi_size
;
164 if (unlikely(rq
->cmd_flags
& REQ_QUIET
))
165 set_bit(BIO_QUIET
, &bio
->bi_flags
);
167 bio
->bi_size
-= nbytes
;
168 bio
->bi_sector
+= (nbytes
>> 9);
170 if (bio_integrity(bio
))
171 bio_integrity_advance(bio
, nbytes
);
173 /* don't actually finish bio if it's part of flush sequence */
174 if (bio
->bi_size
== 0 && !(rq
->cmd_flags
& REQ_FLUSH_SEQ
))
175 bio_endio(bio
, error
);
178 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
182 printk(KERN_INFO
"%s: dev %s: type=%x, flags=%x\n", msg
,
183 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->cmd_type
,
186 printk(KERN_INFO
" sector %llu, nr/cnr %u/%u\n",
187 (unsigned long long)blk_rq_pos(rq
),
188 blk_rq_sectors(rq
), blk_rq_cur_sectors(rq
));
189 printk(KERN_INFO
" bio %p, biotail %p, buffer %p, len %u\n",
190 rq
->bio
, rq
->biotail
, rq
->buffer
, blk_rq_bytes(rq
));
192 if (rq
->cmd_type
== REQ_TYPE_BLOCK_PC
) {
193 printk(KERN_INFO
" cdb: ");
194 for (bit
= 0; bit
< BLK_MAX_CDB
; bit
++)
195 printk("%02x ", rq
->cmd
[bit
]);
199 EXPORT_SYMBOL(blk_dump_rq_flags
);
202 * Make sure that plugs that were pending when this function was entered,
203 * are now complete and requests pushed to the queue.
205 static inline void queue_sync_plugs(struct request_queue
*q
)
208 * If the current process is plugged and has barriers submitted,
209 * we will livelock if we don't unplug first.
211 blk_flush_plug(current
);
214 static void blk_delay_work(struct work_struct
*work
)
216 struct request_queue
*q
;
218 q
= container_of(work
, struct request_queue
, delay_work
.work
);
219 spin_lock_irq(q
->queue_lock
);
221 spin_unlock_irq(q
->queue_lock
);
225 * blk_delay_queue - restart queueing after defined interval
226 * @q: The &struct request_queue in question
227 * @msecs: Delay in msecs
230 * Sometimes queueing needs to be postponed for a little while, to allow
231 * resources to come back. This function will make sure that queueing is
232 * restarted around the specified time.
234 void blk_delay_queue(struct request_queue
*q
, unsigned long msecs
)
236 schedule_delayed_work(&q
->delay_work
, msecs_to_jiffies(msecs
));
238 EXPORT_SYMBOL(blk_delay_queue
);
241 * blk_start_queue - restart a previously stopped queue
242 * @q: The &struct request_queue in question
245 * blk_start_queue() will clear the stop flag on the queue, and call
246 * the request_fn for the queue if it was in a stopped state when
247 * entered. Also see blk_stop_queue(). Queue lock must be held.
249 void blk_start_queue(struct request_queue
*q
)
251 WARN_ON(!irqs_disabled());
253 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
256 EXPORT_SYMBOL(blk_start_queue
);
259 * blk_stop_queue - stop a queue
260 * @q: The &struct request_queue in question
263 * The Linux block layer assumes that a block driver will consume all
264 * entries on the request queue when the request_fn strategy is called.
265 * Often this will not happen, because of hardware limitations (queue
266 * depth settings). If a device driver gets a 'queue full' response,
267 * or if it simply chooses not to queue more I/O at one point, it can
268 * call this function to prevent the request_fn from being called until
269 * the driver has signalled it's ready to go again. This happens by calling
270 * blk_start_queue() to restart queue operations. Queue lock must be held.
272 void blk_stop_queue(struct request_queue
*q
)
274 cancel_delayed_work(&q
->delay_work
);
275 queue_flag_set(QUEUE_FLAG_STOPPED
, q
);
277 EXPORT_SYMBOL(blk_stop_queue
);
280 * blk_sync_queue - cancel any pending callbacks on a queue
284 * The block layer may perform asynchronous callback activity
285 * on a queue, such as calling the unplug function after a timeout.
286 * A block device may call blk_sync_queue to ensure that any
287 * such activity is cancelled, thus allowing it to release resources
288 * that the callbacks might use. The caller must already have made sure
289 * that its ->make_request_fn will not re-add plugging prior to calling
293 void blk_sync_queue(struct request_queue
*q
)
295 del_timer_sync(&q
->timeout
);
296 throtl_shutdown_timer_wq(q
);
297 cancel_delayed_work_sync(&q
->delay_work
);
300 EXPORT_SYMBOL(blk_sync_queue
);
303 * __blk_run_queue - run a single device queue
304 * @q: The queue to run
307 * See @blk_run_queue. This variant must be called with the queue lock
308 * held and interrupts disabled.
311 void __blk_run_queue(struct request_queue
*q
)
313 if (unlikely(blk_queue_stopped(q
)))
317 * Only recurse once to avoid overrunning the stack, let the unplug
318 * handling reinvoke the handler shortly if we already got there.
320 if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER
, q
)) {
322 queue_flag_clear(QUEUE_FLAG_REENTER
, q
);
324 queue_delayed_work(kblockd_workqueue
, &q
->delay_work
, 0);
326 EXPORT_SYMBOL(__blk_run_queue
);
329 * blk_run_queue - run a single device queue
330 * @q: The queue to run
333 * Invoke request handling on this queue, if it has pending work to do.
334 * May be used to restart queueing when a request has completed.
336 void blk_run_queue(struct request_queue
*q
)
340 spin_lock_irqsave(q
->queue_lock
, flags
);
342 spin_unlock_irqrestore(q
->queue_lock
, flags
);
344 EXPORT_SYMBOL(blk_run_queue
);
346 void blk_put_queue(struct request_queue
*q
)
348 kobject_put(&q
->kobj
);
351 void blk_cleanup_queue(struct request_queue
*q
)
354 * We know we have process context here, so we can be a little
355 * cautious and ensure that pending block actions on this device
356 * are done before moving on. Going into this function, we should
357 * not have processes doing IO to this device.
361 del_timer_sync(&q
->backing_dev_info
.laptop_mode_wb_timer
);
362 mutex_lock(&q
->sysfs_lock
);
363 queue_flag_set_unlocked(QUEUE_FLAG_DEAD
, q
);
364 mutex_unlock(&q
->sysfs_lock
);
367 elevator_exit(q
->elevator
);
371 EXPORT_SYMBOL(blk_cleanup_queue
);
373 static int blk_init_free_list(struct request_queue
*q
)
375 struct request_list
*rl
= &q
->rq
;
377 if (unlikely(rl
->rq_pool
))
380 rl
->count
[BLK_RW_SYNC
] = rl
->count
[BLK_RW_ASYNC
] = 0;
381 rl
->starved
[BLK_RW_SYNC
] = rl
->starved
[BLK_RW_ASYNC
] = 0;
383 init_waitqueue_head(&rl
->wait
[BLK_RW_SYNC
]);
384 init_waitqueue_head(&rl
->wait
[BLK_RW_ASYNC
]);
386 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, mempool_alloc_slab
,
387 mempool_free_slab
, request_cachep
, q
->node
);
395 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
397 return blk_alloc_queue_node(gfp_mask
, -1);
399 EXPORT_SYMBOL(blk_alloc_queue
);
401 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
403 struct request_queue
*q
;
406 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
407 gfp_mask
| __GFP_ZERO
, node_id
);
411 q
->backing_dev_info
.ra_pages
=
412 (VM_MAX_READAHEAD
* 1024) / PAGE_CACHE_SIZE
;
413 q
->backing_dev_info
.state
= 0;
414 q
->backing_dev_info
.capabilities
= BDI_CAP_MAP_COPY
;
415 q
->backing_dev_info
.name
= "block";
417 err
= bdi_init(&q
->backing_dev_info
);
419 kmem_cache_free(blk_requestq_cachep
, q
);
423 if (blk_throtl_init(q
)) {
424 kmem_cache_free(blk_requestq_cachep
, q
);
428 setup_timer(&q
->backing_dev_info
.laptop_mode_wb_timer
,
429 laptop_mode_timer_fn
, (unsigned long) q
);
430 setup_timer(&q
->timeout
, blk_rq_timed_out_timer
, (unsigned long) q
);
431 INIT_LIST_HEAD(&q
->timeout_list
);
432 INIT_LIST_HEAD(&q
->flush_queue
[0]);
433 INIT_LIST_HEAD(&q
->flush_queue
[1]);
434 INIT_LIST_HEAD(&q
->flush_data_in_flight
);
435 INIT_DELAYED_WORK(&q
->delay_work
, blk_delay_work
);
437 kobject_init(&q
->kobj
, &blk_queue_ktype
);
439 mutex_init(&q
->sysfs_lock
);
440 spin_lock_init(&q
->__queue_lock
);
444 EXPORT_SYMBOL(blk_alloc_queue_node
);
447 * blk_init_queue - prepare a request queue for use with a block device
448 * @rfn: The function to be called to process requests that have been
449 * placed on the queue.
450 * @lock: Request queue spin lock
453 * If a block device wishes to use the standard request handling procedures,
454 * which sorts requests and coalesces adjacent requests, then it must
455 * call blk_init_queue(). The function @rfn will be called when there
456 * are requests on the queue that need to be processed. If the device
457 * supports plugging, then @rfn may not be called immediately when requests
458 * are available on the queue, but may be called at some time later instead.
459 * Plugged queues are generally unplugged when a buffer belonging to one
460 * of the requests on the queue is needed, or due to memory pressure.
462 * @rfn is not required, or even expected, to remove all requests off the
463 * queue, but only as many as it can handle at a time. If it does leave
464 * requests on the queue, it is responsible for arranging that the requests
465 * get dealt with eventually.
467 * The queue spin lock must be held while manipulating the requests on the
468 * request queue; this lock will be taken also from interrupt context, so irq
469 * disabling is needed for it.
471 * Function returns a pointer to the initialized request queue, or %NULL if
475 * blk_init_queue() must be paired with a blk_cleanup_queue() call
476 * when the block device is deactivated (such as at module unload).
479 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
481 return blk_init_queue_node(rfn
, lock
, -1);
483 EXPORT_SYMBOL(blk_init_queue
);
485 struct request_queue
*
486 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
488 struct request_queue
*uninit_q
, *q
;
490 uninit_q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
494 q
= blk_init_allocated_queue_node(uninit_q
, rfn
, lock
, node_id
);
496 blk_cleanup_queue(uninit_q
);
500 EXPORT_SYMBOL(blk_init_queue_node
);
502 struct request_queue
*
503 blk_init_allocated_queue(struct request_queue
*q
, request_fn_proc
*rfn
,
506 return blk_init_allocated_queue_node(q
, rfn
, lock
, -1);
508 EXPORT_SYMBOL(blk_init_allocated_queue
);
510 struct request_queue
*
511 blk_init_allocated_queue_node(struct request_queue
*q
, request_fn_proc
*rfn
,
512 spinlock_t
*lock
, int node_id
)
518 if (blk_init_free_list(q
))
522 q
->prep_rq_fn
= NULL
;
523 q
->unprep_rq_fn
= NULL
;
524 q
->queue_flags
= QUEUE_FLAG_DEFAULT
;
525 q
->queue_lock
= lock
;
528 * This also sets hw/phys segments, boundary and size
530 blk_queue_make_request(q
, __make_request
);
532 q
->sg_reserved_size
= INT_MAX
;
537 if (!elevator_init(q
, NULL
)) {
538 blk_queue_congestion_threshold(q
);
544 EXPORT_SYMBOL(blk_init_allocated_queue_node
);
546 int blk_get_queue(struct request_queue
*q
)
548 if (likely(!test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
))) {
549 kobject_get(&q
->kobj
);
556 static inline void blk_free_request(struct request_queue
*q
, struct request
*rq
)
558 BUG_ON(rq
->cmd_flags
& REQ_ON_PLUG
);
560 if (rq
->cmd_flags
& REQ_ELVPRIV
)
561 elv_put_request(q
, rq
);
562 mempool_free(rq
, q
->rq
.rq_pool
);
565 static struct request
*
566 blk_alloc_request(struct request_queue
*q
, int flags
, int priv
, gfp_t gfp_mask
)
568 struct request
*rq
= mempool_alloc(q
->rq
.rq_pool
, gfp_mask
);
575 rq
->cmd_flags
= flags
| REQ_ALLOCED
;
578 if (unlikely(elv_set_request(q
, rq
, gfp_mask
))) {
579 mempool_free(rq
, q
->rq
.rq_pool
);
582 rq
->cmd_flags
|= REQ_ELVPRIV
;
589 * ioc_batching returns true if the ioc is a valid batching request and
590 * should be given priority access to a request.
592 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
598 * Make sure the process is able to allocate at least 1 request
599 * even if the batch times out, otherwise we could theoretically
602 return ioc
->nr_batch_requests
== q
->nr_batching
||
603 (ioc
->nr_batch_requests
> 0
604 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
608 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
609 * will cause the process to be a "batcher" on all queues in the system. This
610 * is the behaviour we want though - once it gets a wakeup it should be given
613 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
615 if (!ioc
|| ioc_batching(q
, ioc
))
618 ioc
->nr_batch_requests
= q
->nr_batching
;
619 ioc
->last_waited
= jiffies
;
622 static void __freed_request(struct request_queue
*q
, int sync
)
624 struct request_list
*rl
= &q
->rq
;
626 if (rl
->count
[sync
] < queue_congestion_off_threshold(q
))
627 blk_clear_queue_congested(q
, sync
);
629 if (rl
->count
[sync
] + 1 <= q
->nr_requests
) {
630 if (waitqueue_active(&rl
->wait
[sync
]))
631 wake_up(&rl
->wait
[sync
]);
633 blk_clear_queue_full(q
, sync
);
638 * A request has just been released. Account for it, update the full and
639 * congestion status, wake up any waiters. Called under q->queue_lock.
641 static void freed_request(struct request_queue
*q
, int sync
, int priv
)
643 struct request_list
*rl
= &q
->rq
;
649 __freed_request(q
, sync
);
651 if (unlikely(rl
->starved
[sync
^ 1]))
652 __freed_request(q
, sync
^ 1);
656 * Determine if elevator data should be initialized when allocating the
657 * request associated with @bio.
659 static bool blk_rq_should_init_elevator(struct bio
*bio
)
665 * Flush requests do not use the elevator so skip initialization.
666 * This allows a request to share the flush and elevator data.
668 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
))
675 * Get a free request, queue_lock must be held.
676 * Returns NULL on failure, with queue_lock held.
677 * Returns !NULL on success, with queue_lock *not held*.
679 static struct request
*get_request(struct request_queue
*q
, int rw_flags
,
680 struct bio
*bio
, gfp_t gfp_mask
)
682 struct request
*rq
= NULL
;
683 struct request_list
*rl
= &q
->rq
;
684 struct io_context
*ioc
= NULL
;
685 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
686 int may_queue
, priv
= 0;
688 may_queue
= elv_may_queue(q
, rw_flags
);
689 if (may_queue
== ELV_MQUEUE_NO
)
692 if (rl
->count
[is_sync
]+1 >= queue_congestion_on_threshold(q
)) {
693 if (rl
->count
[is_sync
]+1 >= q
->nr_requests
) {
694 ioc
= current_io_context(GFP_ATOMIC
, q
->node
);
696 * The queue will fill after this allocation, so set
697 * it as full, and mark this process as "batching".
698 * This process will be allowed to complete a batch of
699 * requests, others will be blocked.
701 if (!blk_queue_full(q
, is_sync
)) {
702 ioc_set_batching(q
, ioc
);
703 blk_set_queue_full(q
, is_sync
);
705 if (may_queue
!= ELV_MQUEUE_MUST
706 && !ioc_batching(q
, ioc
)) {
708 * The queue is full and the allocating
709 * process is not a "batcher", and not
710 * exempted by the IO scheduler
716 blk_set_queue_congested(q
, is_sync
);
720 * Only allow batching queuers to allocate up to 50% over the defined
721 * limit of requests, otherwise we could have thousands of requests
722 * allocated with any setting of ->nr_requests
724 if (rl
->count
[is_sync
] >= (3 * q
->nr_requests
/ 2))
727 rl
->count
[is_sync
]++;
728 rl
->starved
[is_sync
] = 0;
730 if (blk_rq_should_init_elevator(bio
)) {
731 priv
= !test_bit(QUEUE_FLAG_ELVSWITCH
, &q
->queue_flags
);
736 if (blk_queue_io_stat(q
))
737 rw_flags
|= REQ_IO_STAT
;
738 spin_unlock_irq(q
->queue_lock
);
740 rq
= blk_alloc_request(q
, rw_flags
, priv
, gfp_mask
);
743 * Allocation failed presumably due to memory. Undo anything
744 * we might have messed up.
746 * Allocating task should really be put onto the front of the
747 * wait queue, but this is pretty rare.
749 spin_lock_irq(q
->queue_lock
);
750 freed_request(q
, is_sync
, priv
);
753 * in the very unlikely event that allocation failed and no
754 * requests for this direction was pending, mark us starved
755 * so that freeing of a request in the other direction will
756 * notice us. another possible fix would be to split the
757 * rq mempool into READ and WRITE
760 if (unlikely(rl
->count
[is_sync
] == 0))
761 rl
->starved
[is_sync
] = 1;
767 * ioc may be NULL here, and ioc_batching will be false. That's
768 * OK, if the queue is under the request limit then requests need
769 * not count toward the nr_batch_requests limit. There will always
770 * be some limit enforced by BLK_BATCH_TIME.
772 if (ioc_batching(q
, ioc
))
773 ioc
->nr_batch_requests
--;
775 trace_block_getrq(q
, bio
, rw_flags
& 1);
781 * No available requests for this queue, wait for some requests to become
784 * Called with q->queue_lock held, and returns with it unlocked.
786 static struct request
*get_request_wait(struct request_queue
*q
, int rw_flags
,
789 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
792 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
795 struct io_context
*ioc
;
796 struct request_list
*rl
= &q
->rq
;
798 prepare_to_wait_exclusive(&rl
->wait
[is_sync
], &wait
,
799 TASK_UNINTERRUPTIBLE
);
801 trace_block_sleeprq(q
, bio
, rw_flags
& 1);
803 spin_unlock_irq(q
->queue_lock
);
807 * After sleeping, we become a "batching" process and
808 * will be able to allocate at least one request, and
809 * up to a big batch of them for a small period time.
810 * See ioc_batching, ioc_set_batching
812 ioc
= current_io_context(GFP_NOIO
, q
->node
);
813 ioc_set_batching(q
, ioc
);
815 spin_lock_irq(q
->queue_lock
);
816 finish_wait(&rl
->wait
[is_sync
], &wait
);
818 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
824 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
828 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
830 spin_lock_irq(q
->queue_lock
);
831 if (gfp_mask
& __GFP_WAIT
) {
832 rq
= get_request_wait(q
, rw
, NULL
);
834 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
836 spin_unlock_irq(q
->queue_lock
);
838 /* q->queue_lock is unlocked at this point */
842 EXPORT_SYMBOL(blk_get_request
);
845 * blk_make_request - given a bio, allocate a corresponding struct request.
846 * @q: target request queue
847 * @bio: The bio describing the memory mappings that will be submitted for IO.
848 * It may be a chained-bio properly constructed by block/bio layer.
849 * @gfp_mask: gfp flags to be used for memory allocation
851 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
852 * type commands. Where the struct request needs to be farther initialized by
853 * the caller. It is passed a &struct bio, which describes the memory info of
856 * The caller of blk_make_request must make sure that bi_io_vec
857 * are set to describe the memory buffers. That bio_data_dir() will return
858 * the needed direction of the request. (And all bio's in the passed bio-chain
859 * are properly set accordingly)
861 * If called under none-sleepable conditions, mapped bio buffers must not
862 * need bouncing, by calling the appropriate masked or flagged allocator,
863 * suitable for the target device. Otherwise the call to blk_queue_bounce will
866 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
867 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
868 * anything but the first bio in the chain. Otherwise you risk waiting for IO
869 * completion of a bio that hasn't been submitted yet, thus resulting in a
870 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
871 * of bio_alloc(), as that avoids the mempool deadlock.
872 * If possible a big IO should be split into smaller parts when allocation
873 * fails. Partial allocation should not be an error, or you risk a live-lock.
875 struct request
*blk_make_request(struct request_queue
*q
, struct bio
*bio
,
878 struct request
*rq
= blk_get_request(q
, bio_data_dir(bio
), gfp_mask
);
881 return ERR_PTR(-ENOMEM
);
884 struct bio
*bounce_bio
= bio
;
887 blk_queue_bounce(q
, &bounce_bio
);
888 ret
= blk_rq_append_bio(q
, rq
, bounce_bio
);
897 EXPORT_SYMBOL(blk_make_request
);
900 * blk_requeue_request - put a request back on queue
901 * @q: request queue where request should be inserted
902 * @rq: request to be inserted
905 * Drivers often keep queueing requests until the hardware cannot accept
906 * more, when that condition happens we need to put the request back
907 * on the queue. Must be called with queue lock held.
909 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
911 blk_delete_timer(rq
);
912 blk_clear_rq_complete(rq
);
913 trace_block_rq_requeue(q
, rq
);
915 if (blk_rq_tagged(rq
))
916 blk_queue_end_tag(q
, rq
);
918 BUG_ON(blk_queued_rq(rq
));
920 elv_requeue_request(q
, rq
);
922 EXPORT_SYMBOL(blk_requeue_request
);
924 static void add_acct_request(struct request_queue
*q
, struct request
*rq
,
927 drive_stat_acct(rq
, 1);
928 __elv_add_request(q
, rq
, where
);
932 * blk_insert_request - insert a special request into a request queue
933 * @q: request queue where request should be inserted
934 * @rq: request to be inserted
935 * @at_head: insert request at head or tail of queue
936 * @data: private data
939 * Many block devices need to execute commands asynchronously, so they don't
940 * block the whole kernel from preemption during request execution. This is
941 * accomplished normally by inserting aritficial requests tagged as
942 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
943 * be scheduled for actual execution by the request queue.
945 * We have the option of inserting the head or the tail of the queue.
946 * Typically we use the tail for new ioctls and so forth. We use the head
947 * of the queue for things like a QUEUE_FULL message from a device, or a
948 * host that is unable to accept a particular command.
950 void blk_insert_request(struct request_queue
*q
, struct request
*rq
,
951 int at_head
, void *data
)
953 int where
= at_head
? ELEVATOR_INSERT_FRONT
: ELEVATOR_INSERT_BACK
;
957 * tell I/O scheduler that this isn't a regular read/write (ie it
958 * must not attempt merges on this) and that it acts as a soft
961 rq
->cmd_type
= REQ_TYPE_SPECIAL
;
965 spin_lock_irqsave(q
->queue_lock
, flags
);
968 * If command is tagged, release the tag
970 if (blk_rq_tagged(rq
))
971 blk_queue_end_tag(q
, rq
);
973 add_acct_request(q
, rq
, where
);
975 spin_unlock_irqrestore(q
->queue_lock
, flags
);
977 EXPORT_SYMBOL(blk_insert_request
);
979 static void part_round_stats_single(int cpu
, struct hd_struct
*part
,
982 if (now
== part
->stamp
)
985 if (part_in_flight(part
)) {
986 __part_stat_add(cpu
, part
, time_in_queue
,
987 part_in_flight(part
) * (now
- part
->stamp
));
988 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
994 * part_round_stats() - Round off the performance stats on a struct disk_stats.
995 * @cpu: cpu number for stats access
996 * @part: target partition
998 * The average IO queue length and utilisation statistics are maintained
999 * by observing the current state of the queue length and the amount of
1000 * time it has been in this state for.
1002 * Normally, that accounting is done on IO completion, but that can result
1003 * in more than a second's worth of IO being accounted for within any one
1004 * second, leading to >100% utilisation. To deal with that, we call this
1005 * function to do a round-off before returning the results when reading
1006 * /proc/diskstats. This accounts immediately for all queue usage up to
1007 * the current jiffies and restarts the counters again.
1009 void part_round_stats(int cpu
, struct hd_struct
*part
)
1011 unsigned long now
= jiffies
;
1014 part_round_stats_single(cpu
, &part_to_disk(part
)->part0
, now
);
1015 part_round_stats_single(cpu
, part
, now
);
1017 EXPORT_SYMBOL_GPL(part_round_stats
);
1020 * queue lock must be held
1022 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1026 if (unlikely(--req
->ref_count
))
1029 elv_completed_request(q
, req
);
1031 /* this is a bio leak */
1032 WARN_ON(req
->bio
!= NULL
);
1035 * Request may not have originated from ll_rw_blk. if not,
1036 * it didn't come out of our reserved rq pools
1038 if (req
->cmd_flags
& REQ_ALLOCED
) {
1039 int is_sync
= rq_is_sync(req
) != 0;
1040 int priv
= req
->cmd_flags
& REQ_ELVPRIV
;
1042 BUG_ON(!list_empty(&req
->queuelist
));
1043 BUG_ON(!hlist_unhashed(&req
->hash
));
1045 blk_free_request(q
, req
);
1046 freed_request(q
, is_sync
, priv
);
1049 EXPORT_SYMBOL_GPL(__blk_put_request
);
1051 void blk_put_request(struct request
*req
)
1053 unsigned long flags
;
1054 struct request_queue
*q
= req
->q
;
1056 spin_lock_irqsave(q
->queue_lock
, flags
);
1057 __blk_put_request(q
, req
);
1058 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1060 EXPORT_SYMBOL(blk_put_request
);
1063 * blk_add_request_payload - add a payload to a request
1064 * @rq: request to update
1065 * @page: page backing the payload
1066 * @len: length of the payload.
1068 * This allows to later add a payload to an already submitted request by
1069 * a block driver. The driver needs to take care of freeing the payload
1072 * Note that this is a quite horrible hack and nothing but handling of
1073 * discard requests should ever use it.
1075 void blk_add_request_payload(struct request
*rq
, struct page
*page
,
1078 struct bio
*bio
= rq
->bio
;
1080 bio
->bi_io_vec
->bv_page
= page
;
1081 bio
->bi_io_vec
->bv_offset
= 0;
1082 bio
->bi_io_vec
->bv_len
= len
;
1086 bio
->bi_phys_segments
= 1;
1088 rq
->__data_len
= rq
->resid_len
= len
;
1089 rq
->nr_phys_segments
= 1;
1090 rq
->buffer
= bio_data(bio
);
1092 EXPORT_SYMBOL_GPL(blk_add_request_payload
);
1094 static bool bio_attempt_back_merge(struct request_queue
*q
, struct request
*req
,
1097 const int ff
= bio
->bi_rw
& REQ_FAILFAST_MASK
;
1100 * Debug stuff, kill later
1102 if (!rq_mergeable(req
)) {
1103 blk_dump_rq_flags(req
, "back");
1107 if (!ll_back_merge_fn(q
, req
, bio
))
1110 trace_block_bio_backmerge(q
, bio
);
1112 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1113 blk_rq_set_mixed_merge(req
);
1115 req
->biotail
->bi_next
= bio
;
1117 req
->__data_len
+= bio
->bi_size
;
1118 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1120 drive_stat_acct(req
, 0);
1124 static bool bio_attempt_front_merge(struct request_queue
*q
,
1125 struct request
*req
, struct bio
*bio
)
1127 const int ff
= bio
->bi_rw
& REQ_FAILFAST_MASK
;
1131 * Debug stuff, kill later
1133 if (!rq_mergeable(req
)) {
1134 blk_dump_rq_flags(req
, "front");
1138 if (!ll_front_merge_fn(q
, req
, bio
))
1141 trace_block_bio_frontmerge(q
, bio
);
1143 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1144 blk_rq_set_mixed_merge(req
);
1146 sector
= bio
->bi_sector
;
1148 bio
->bi_next
= req
->bio
;
1152 * may not be valid. if the low level driver said
1153 * it didn't need a bounce buffer then it better
1154 * not touch req->buffer either...
1156 req
->buffer
= bio_data(bio
);
1157 req
->__sector
= bio
->bi_sector
;
1158 req
->__data_len
+= bio
->bi_size
;
1159 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1161 drive_stat_acct(req
, 0);
1166 * Attempts to merge with the plugged list in the current process. Returns
1167 * true if merge was succesful, otherwise false.
1169 static bool attempt_plug_merge(struct task_struct
*tsk
, struct request_queue
*q
,
1172 struct blk_plug
*plug
;
1180 list_for_each_entry_reverse(rq
, &plug
->list
, queuelist
) {
1186 el_ret
= elv_try_merge(rq
, bio
);
1187 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1188 ret
= bio_attempt_back_merge(q
, rq
, bio
);
1191 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1192 ret
= bio_attempt_front_merge(q
, rq
, bio
);
1201 void init_request_from_bio(struct request
*req
, struct bio
*bio
)
1203 req
->cpu
= bio
->bi_comp_cpu
;
1204 req
->cmd_type
= REQ_TYPE_FS
;
1206 req
->cmd_flags
|= bio
->bi_rw
& REQ_COMMON_MASK
;
1207 if (bio
->bi_rw
& REQ_RAHEAD
)
1208 req
->cmd_flags
|= REQ_FAILFAST_MASK
;
1211 req
->__sector
= bio
->bi_sector
;
1212 req
->ioprio
= bio_prio(bio
);
1213 blk_rq_bio_prep(req
->q
, req
, bio
);
1216 static int __make_request(struct request_queue
*q
, struct bio
*bio
)
1218 const bool sync
= !!(bio
->bi_rw
& REQ_SYNC
);
1219 struct blk_plug
*plug
;
1220 int el_ret
, rw_flags
, where
= ELEVATOR_INSERT_SORT
;
1221 struct request
*req
;
1224 * low level driver can indicate that it wants pages above a
1225 * certain limit bounced to low memory (ie for highmem, or even
1226 * ISA dma in theory)
1228 blk_queue_bounce(q
, &bio
);
1230 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) {
1231 spin_lock_irq(q
->queue_lock
);
1232 where
= ELEVATOR_INSERT_FLUSH
;
1237 * Check if we can merge with the plugged list before grabbing
1240 if (attempt_plug_merge(current
, q
, bio
))
1243 spin_lock_irq(q
->queue_lock
);
1245 el_ret
= elv_merge(q
, &req
, bio
);
1246 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1247 BUG_ON(req
->cmd_flags
& REQ_ON_PLUG
);
1248 if (bio_attempt_back_merge(q
, req
, bio
)) {
1249 if (!attempt_back_merge(q
, req
))
1250 elv_merged_request(q
, req
, el_ret
);
1253 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1254 BUG_ON(req
->cmd_flags
& REQ_ON_PLUG
);
1255 if (bio_attempt_front_merge(q
, req
, bio
)) {
1256 if (!attempt_front_merge(q
, req
))
1257 elv_merged_request(q
, req
, el_ret
);
1264 * This sync check and mask will be re-done in init_request_from_bio(),
1265 * but we need to set it earlier to expose the sync flag to the
1266 * rq allocator and io schedulers.
1268 rw_flags
= bio_data_dir(bio
);
1270 rw_flags
|= REQ_SYNC
;
1273 * Grab a free request. This is might sleep but can not fail.
1274 * Returns with the queue unlocked.
1276 req
= get_request_wait(q
, rw_flags
, bio
);
1279 * After dropping the lock and possibly sleeping here, our request
1280 * may now be mergeable after it had proven unmergeable (above).
1281 * We don't worry about that case for efficiency. It won't happen
1282 * often, and the elevators are able to handle it.
1284 init_request_from_bio(req
, bio
);
1286 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
) ||
1287 bio_flagged(bio
, BIO_CPU_AFFINE
)) {
1288 req
->cpu
= blk_cpu_to_group(get_cpu());
1292 plug
= current
->plug
;
1293 if (plug
&& !sync
) {
1294 if (!plug
->should_sort
&& !list_empty(&plug
->list
)) {
1295 struct request
*__rq
;
1297 __rq
= list_entry_rq(plug
->list
.prev
);
1299 plug
->should_sort
= 1;
1302 * Debug flag, kill later
1304 req
->cmd_flags
|= REQ_ON_PLUG
;
1305 list_add_tail(&req
->queuelist
, &plug
->list
);
1306 drive_stat_acct(req
, 1);
1308 spin_lock_irq(q
->queue_lock
);
1309 add_acct_request(q
, req
, where
);
1312 spin_unlock_irq(q
->queue_lock
);
1319 * If bio->bi_dev is a partition, remap the location
1321 static inline void blk_partition_remap(struct bio
*bio
)
1323 struct block_device
*bdev
= bio
->bi_bdev
;
1325 if (bio_sectors(bio
) && bdev
!= bdev
->bd_contains
) {
1326 struct hd_struct
*p
= bdev
->bd_part
;
1328 bio
->bi_sector
+= p
->start_sect
;
1329 bio
->bi_bdev
= bdev
->bd_contains
;
1331 trace_block_bio_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
1333 bio
->bi_sector
- p
->start_sect
);
1337 static void handle_bad_sector(struct bio
*bio
)
1339 char b
[BDEVNAME_SIZE
];
1341 printk(KERN_INFO
"attempt to access beyond end of device\n");
1342 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
1343 bdevname(bio
->bi_bdev
, b
),
1345 (unsigned long long)bio
->bi_sector
+ bio_sectors(bio
),
1346 (long long)(i_size_read(bio
->bi_bdev
->bd_inode
) >> 9));
1348 set_bit(BIO_EOF
, &bio
->bi_flags
);
1351 #ifdef CONFIG_FAIL_MAKE_REQUEST
1353 static DECLARE_FAULT_ATTR(fail_make_request
);
1355 static int __init
setup_fail_make_request(char *str
)
1357 return setup_fault_attr(&fail_make_request
, str
);
1359 __setup("fail_make_request=", setup_fail_make_request
);
1361 static int should_fail_request(struct bio
*bio
)
1363 struct hd_struct
*part
= bio
->bi_bdev
->bd_part
;
1365 if (part_to_disk(part
)->part0
.make_it_fail
|| part
->make_it_fail
)
1366 return should_fail(&fail_make_request
, bio
->bi_size
);
1371 static int __init
fail_make_request_debugfs(void)
1373 return init_fault_attr_dentries(&fail_make_request
,
1374 "fail_make_request");
1377 late_initcall(fail_make_request_debugfs
);
1379 #else /* CONFIG_FAIL_MAKE_REQUEST */
1381 static inline int should_fail_request(struct bio
*bio
)
1386 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1389 * Check whether this bio extends beyond the end of the device.
1391 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
1398 /* Test device or partition size, when known. */
1399 maxsector
= i_size_read(bio
->bi_bdev
->bd_inode
) >> 9;
1401 sector_t sector
= bio
->bi_sector
;
1403 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
1405 * This may well happen - the kernel calls bread()
1406 * without checking the size of the device, e.g., when
1407 * mounting a device.
1409 handle_bad_sector(bio
);
1418 * generic_make_request - hand a buffer to its device driver for I/O
1419 * @bio: The bio describing the location in memory and on the device.
1421 * generic_make_request() is used to make I/O requests of block
1422 * devices. It is passed a &struct bio, which describes the I/O that needs
1425 * generic_make_request() does not return any status. The
1426 * success/failure status of the request, along with notification of
1427 * completion, is delivered asynchronously through the bio->bi_end_io
1428 * function described (one day) else where.
1430 * The caller of generic_make_request must make sure that bi_io_vec
1431 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1432 * set to describe the device address, and the
1433 * bi_end_io and optionally bi_private are set to describe how
1434 * completion notification should be signaled.
1436 * generic_make_request and the drivers it calls may use bi_next if this
1437 * bio happens to be merged with someone else, and may change bi_dev and
1438 * bi_sector for remaps as it sees fit. So the values of these fields
1439 * should NOT be depended on after the call to generic_make_request.
1441 static inline void __generic_make_request(struct bio
*bio
)
1443 struct request_queue
*q
;
1444 sector_t old_sector
;
1445 int ret
, nr_sectors
= bio_sectors(bio
);
1451 if (bio_check_eod(bio
, nr_sectors
))
1455 * Resolve the mapping until finished. (drivers are
1456 * still free to implement/resolve their own stacking
1457 * by explicitly returning 0)
1459 * NOTE: we don't repeat the blk_size check for each new device.
1460 * Stacking drivers are expected to know what they are doing.
1465 char b
[BDEVNAME_SIZE
];
1467 q
= bdev_get_queue(bio
->bi_bdev
);
1470 "generic_make_request: Trying to access "
1471 "nonexistent block-device %s (%Lu)\n",
1472 bdevname(bio
->bi_bdev
, b
),
1473 (long long) bio
->bi_sector
);
1477 if (unlikely(!(bio
->bi_rw
& REQ_DISCARD
) &&
1478 nr_sectors
> queue_max_hw_sectors(q
))) {
1479 printk(KERN_ERR
"bio too big device %s (%u > %u)\n",
1480 bdevname(bio
->bi_bdev
, b
),
1482 queue_max_hw_sectors(q
));
1486 if (unlikely(test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)))
1489 if (should_fail_request(bio
))
1493 * If this device has partitions, remap block n
1494 * of partition p to block n+start(p) of the disk.
1496 blk_partition_remap(bio
);
1498 if (bio_integrity_enabled(bio
) && bio_integrity_prep(bio
))
1501 if (old_sector
!= -1)
1502 trace_block_bio_remap(q
, bio
, old_dev
, old_sector
);
1504 old_sector
= bio
->bi_sector
;
1505 old_dev
= bio
->bi_bdev
->bd_dev
;
1507 if (bio_check_eod(bio
, nr_sectors
))
1511 * Filter flush bio's early so that make_request based
1512 * drivers without flush support don't have to worry
1515 if ((bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) && !q
->flush_flags
) {
1516 bio
->bi_rw
&= ~(REQ_FLUSH
| REQ_FUA
);
1523 if ((bio
->bi_rw
& REQ_DISCARD
) &&
1524 (!blk_queue_discard(q
) ||
1525 ((bio
->bi_rw
& REQ_SECURE
) &&
1526 !blk_queue_secdiscard(q
)))) {
1531 blk_throtl_bio(q
, &bio
);
1534 * If bio = NULL, bio has been throttled and will be submitted
1540 trace_block_bio_queue(q
, bio
);
1542 ret
= q
->make_request_fn(q
, bio
);
1548 bio_endio(bio
, err
);
1552 * We only want one ->make_request_fn to be active at a time,
1553 * else stack usage with stacked devices could be a problem.
1554 * So use current->bio_list to keep a list of requests
1555 * submited by a make_request_fn function.
1556 * current->bio_list is also used as a flag to say if
1557 * generic_make_request is currently active in this task or not.
1558 * If it is NULL, then no make_request is active. If it is non-NULL,
1559 * then a make_request is active, and new requests should be added
1562 void generic_make_request(struct bio
*bio
)
1564 struct bio_list bio_list_on_stack
;
1566 if (current
->bio_list
) {
1567 /* make_request is active */
1568 bio_list_add(current
->bio_list
, bio
);
1571 /* following loop may be a bit non-obvious, and so deserves some
1573 * Before entering the loop, bio->bi_next is NULL (as all callers
1574 * ensure that) so we have a list with a single bio.
1575 * We pretend that we have just taken it off a longer list, so
1576 * we assign bio_list to a pointer to the bio_list_on_stack,
1577 * thus initialising the bio_list of new bios to be
1578 * added. __generic_make_request may indeed add some more bios
1579 * through a recursive call to generic_make_request. If it
1580 * did, we find a non-NULL value in bio_list and re-enter the loop
1581 * from the top. In this case we really did just take the bio
1582 * of the top of the list (no pretending) and so remove it from
1583 * bio_list, and call into __generic_make_request again.
1585 * The loop was structured like this to make only one call to
1586 * __generic_make_request (which is important as it is large and
1587 * inlined) and to keep the structure simple.
1589 BUG_ON(bio
->bi_next
);
1590 bio_list_init(&bio_list_on_stack
);
1591 current
->bio_list
= &bio_list_on_stack
;
1593 __generic_make_request(bio
);
1594 bio
= bio_list_pop(current
->bio_list
);
1596 current
->bio_list
= NULL
; /* deactivate */
1598 EXPORT_SYMBOL(generic_make_request
);
1601 * submit_bio - submit a bio to the block device layer for I/O
1602 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1603 * @bio: The &struct bio which describes the I/O
1605 * submit_bio() is very similar in purpose to generic_make_request(), and
1606 * uses that function to do most of the work. Both are fairly rough
1607 * interfaces; @bio must be presetup and ready for I/O.
1610 void submit_bio(int rw
, struct bio
*bio
)
1612 int count
= bio_sectors(bio
);
1617 * If it's a regular read/write or a barrier with data attached,
1618 * go through the normal accounting stuff before submission.
1620 if (bio_has_data(bio
) && !(rw
& REQ_DISCARD
)) {
1622 count_vm_events(PGPGOUT
, count
);
1624 task_io_account_read(bio
->bi_size
);
1625 count_vm_events(PGPGIN
, count
);
1628 if (unlikely(block_dump
)) {
1629 char b
[BDEVNAME_SIZE
];
1630 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s (%u sectors)\n",
1631 current
->comm
, task_pid_nr(current
),
1632 (rw
& WRITE
) ? "WRITE" : "READ",
1633 (unsigned long long)bio
->bi_sector
,
1634 bdevname(bio
->bi_bdev
, b
),
1639 generic_make_request(bio
);
1641 EXPORT_SYMBOL(submit_bio
);
1644 * blk_rq_check_limits - Helper function to check a request for the queue limit
1646 * @rq: the request being checked
1649 * @rq may have been made based on weaker limitations of upper-level queues
1650 * in request stacking drivers, and it may violate the limitation of @q.
1651 * Since the block layer and the underlying device driver trust @rq
1652 * after it is inserted to @q, it should be checked against @q before
1653 * the insertion using this generic function.
1655 * This function should also be useful for request stacking drivers
1656 * in some cases below, so export this function.
1657 * Request stacking drivers like request-based dm may change the queue
1658 * limits while requests are in the queue (e.g. dm's table swapping).
1659 * Such request stacking drivers should check those requests agaist
1660 * the new queue limits again when they dispatch those requests,
1661 * although such checkings are also done against the old queue limits
1662 * when submitting requests.
1664 int blk_rq_check_limits(struct request_queue
*q
, struct request
*rq
)
1666 if (rq
->cmd_flags
& REQ_DISCARD
)
1669 if (blk_rq_sectors(rq
) > queue_max_sectors(q
) ||
1670 blk_rq_bytes(rq
) > queue_max_hw_sectors(q
) << 9) {
1671 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
1676 * queue's settings related to segment counting like q->bounce_pfn
1677 * may differ from that of other stacking queues.
1678 * Recalculate it to check the request correctly on this queue's
1681 blk_recalc_rq_segments(rq
);
1682 if (rq
->nr_phys_segments
> queue_max_segments(q
)) {
1683 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
1689 EXPORT_SYMBOL_GPL(blk_rq_check_limits
);
1692 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1693 * @q: the queue to submit the request
1694 * @rq: the request being queued
1696 int blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
1698 unsigned long flags
;
1700 if (blk_rq_check_limits(q
, rq
))
1703 #ifdef CONFIG_FAIL_MAKE_REQUEST
1704 if (rq
->rq_disk
&& rq
->rq_disk
->part0
.make_it_fail
&&
1705 should_fail(&fail_make_request
, blk_rq_bytes(rq
)))
1709 spin_lock_irqsave(q
->queue_lock
, flags
);
1712 * Submitting request must be dequeued before calling this function
1713 * because it will be linked to another request_queue
1715 BUG_ON(blk_queued_rq(rq
));
1717 add_acct_request(q
, rq
, ELEVATOR_INSERT_BACK
);
1718 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1722 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
1725 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1726 * @rq: request to examine
1729 * A request could be merge of IOs which require different failure
1730 * handling. This function determines the number of bytes which
1731 * can be failed from the beginning of the request without
1732 * crossing into area which need to be retried further.
1735 * The number of bytes to fail.
1738 * queue_lock must be held.
1740 unsigned int blk_rq_err_bytes(const struct request
*rq
)
1742 unsigned int ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
1743 unsigned int bytes
= 0;
1746 if (!(rq
->cmd_flags
& REQ_MIXED_MERGE
))
1747 return blk_rq_bytes(rq
);
1750 * Currently the only 'mixing' which can happen is between
1751 * different fastfail types. We can safely fail portions
1752 * which have all the failfast bits that the first one has -
1753 * the ones which are at least as eager to fail as the first
1756 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
1757 if ((bio
->bi_rw
& ff
) != ff
)
1759 bytes
+= bio
->bi_size
;
1762 /* this could lead to infinite loop */
1763 BUG_ON(blk_rq_bytes(rq
) && !bytes
);
1766 EXPORT_SYMBOL_GPL(blk_rq_err_bytes
);
1768 static void blk_account_io_completion(struct request
*req
, unsigned int bytes
)
1770 if (blk_do_io_stat(req
)) {
1771 const int rw
= rq_data_dir(req
);
1772 struct hd_struct
*part
;
1775 cpu
= part_stat_lock();
1777 part_stat_add(cpu
, part
, sectors
[rw
], bytes
>> 9);
1782 static void blk_account_io_done(struct request
*req
)
1785 * Account IO completion. flush_rq isn't accounted as a
1786 * normal IO on queueing nor completion. Accounting the
1787 * containing request is enough.
1789 if (blk_do_io_stat(req
) && !(req
->cmd_flags
& REQ_FLUSH_SEQ
)) {
1790 unsigned long duration
= jiffies
- req
->start_time
;
1791 const int rw
= rq_data_dir(req
);
1792 struct hd_struct
*part
;
1795 cpu
= part_stat_lock();
1798 part_stat_inc(cpu
, part
, ios
[rw
]);
1799 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
1800 part_round_stats(cpu
, part
);
1801 part_dec_in_flight(part
, rw
);
1803 hd_struct_put(part
);
1809 * blk_peek_request - peek at the top of a request queue
1810 * @q: request queue to peek at
1813 * Return the request at the top of @q. The returned request
1814 * should be started using blk_start_request() before LLD starts
1818 * Pointer to the request at the top of @q if available. Null
1822 * queue_lock must be held.
1824 struct request
*blk_peek_request(struct request_queue
*q
)
1829 while ((rq
= __elv_next_request(q
)) != NULL
) {
1830 if (!(rq
->cmd_flags
& REQ_STARTED
)) {
1832 * This is the first time the device driver
1833 * sees this request (possibly after
1834 * requeueing). Notify IO scheduler.
1836 if (rq
->cmd_flags
& REQ_SORTED
)
1837 elv_activate_rq(q
, rq
);
1840 * just mark as started even if we don't start
1841 * it, a request that has been delayed should
1842 * not be passed by new incoming requests
1844 rq
->cmd_flags
|= REQ_STARTED
;
1845 trace_block_rq_issue(q
, rq
);
1848 if (!q
->boundary_rq
|| q
->boundary_rq
== rq
) {
1849 q
->end_sector
= rq_end_sector(rq
);
1850 q
->boundary_rq
= NULL
;
1853 if (rq
->cmd_flags
& REQ_DONTPREP
)
1856 if (q
->dma_drain_size
&& blk_rq_bytes(rq
)) {
1858 * make sure space for the drain appears we
1859 * know we can do this because max_hw_segments
1860 * has been adjusted to be one fewer than the
1863 rq
->nr_phys_segments
++;
1869 ret
= q
->prep_rq_fn(q
, rq
);
1870 if (ret
== BLKPREP_OK
) {
1872 } else if (ret
== BLKPREP_DEFER
) {
1874 * the request may have been (partially) prepped.
1875 * we need to keep this request in the front to
1876 * avoid resource deadlock. REQ_STARTED will
1877 * prevent other fs requests from passing this one.
1879 if (q
->dma_drain_size
&& blk_rq_bytes(rq
) &&
1880 !(rq
->cmd_flags
& REQ_DONTPREP
)) {
1882 * remove the space for the drain we added
1883 * so that we don't add it again
1885 --rq
->nr_phys_segments
;
1890 } else if (ret
== BLKPREP_KILL
) {
1891 rq
->cmd_flags
|= REQ_QUIET
;
1893 * Mark this request as started so we don't trigger
1894 * any debug logic in the end I/O path.
1896 blk_start_request(rq
);
1897 __blk_end_request_all(rq
, -EIO
);
1899 printk(KERN_ERR
"%s: bad return=%d\n", __func__
, ret
);
1906 EXPORT_SYMBOL(blk_peek_request
);
1908 void blk_dequeue_request(struct request
*rq
)
1910 struct request_queue
*q
= rq
->q
;
1912 BUG_ON(list_empty(&rq
->queuelist
));
1913 BUG_ON(ELV_ON_HASH(rq
));
1915 list_del_init(&rq
->queuelist
);
1918 * the time frame between a request being removed from the lists
1919 * and to it is freed is accounted as io that is in progress at
1922 if (blk_account_rq(rq
)) {
1923 q
->in_flight
[rq_is_sync(rq
)]++;
1924 set_io_start_time_ns(rq
);
1929 * blk_start_request - start request processing on the driver
1930 * @req: request to dequeue
1933 * Dequeue @req and start timeout timer on it. This hands off the
1934 * request to the driver.
1936 * Block internal functions which don't want to start timer should
1937 * call blk_dequeue_request().
1940 * queue_lock must be held.
1942 void blk_start_request(struct request
*req
)
1944 blk_dequeue_request(req
);
1947 * We are now handing the request to the hardware, initialize
1948 * resid_len to full count and add the timeout handler.
1950 req
->resid_len
= blk_rq_bytes(req
);
1951 if (unlikely(blk_bidi_rq(req
)))
1952 req
->next_rq
->resid_len
= blk_rq_bytes(req
->next_rq
);
1956 EXPORT_SYMBOL(blk_start_request
);
1959 * blk_fetch_request - fetch a request from a request queue
1960 * @q: request queue to fetch a request from
1963 * Return the request at the top of @q. The request is started on
1964 * return and LLD can start processing it immediately.
1967 * Pointer to the request at the top of @q if available. Null
1971 * queue_lock must be held.
1973 struct request
*blk_fetch_request(struct request_queue
*q
)
1977 rq
= blk_peek_request(q
);
1979 blk_start_request(rq
);
1982 EXPORT_SYMBOL(blk_fetch_request
);
1985 * blk_update_request - Special helper function for request stacking drivers
1986 * @req: the request being processed
1987 * @error: %0 for success, < %0 for error
1988 * @nr_bytes: number of bytes to complete @req
1991 * Ends I/O on a number of bytes attached to @req, but doesn't complete
1992 * the request structure even if @req doesn't have leftover.
1993 * If @req has leftover, sets it up for the next range of segments.
1995 * This special helper function is only for request stacking drivers
1996 * (e.g. request-based dm) so that they can handle partial completion.
1997 * Actual device drivers should use blk_end_request instead.
1999 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2000 * %false return from this function.
2003 * %false - this request doesn't have any more data
2004 * %true - this request has more data
2006 bool blk_update_request(struct request
*req
, int error
, unsigned int nr_bytes
)
2008 int total_bytes
, bio_nbytes
, next_idx
= 0;
2014 trace_block_rq_complete(req
->q
, req
);
2017 * For fs requests, rq is just carrier of independent bio's
2018 * and each partial completion should be handled separately.
2019 * Reset per-request error on each partial completion.
2021 * TODO: tj: This is too subtle. It would be better to let
2022 * low level drivers do what they see fit.
2024 if (req
->cmd_type
== REQ_TYPE_FS
)
2027 if (error
&& req
->cmd_type
== REQ_TYPE_FS
&&
2028 !(req
->cmd_flags
& REQ_QUIET
)) {
2029 printk(KERN_ERR
"end_request: I/O error, dev %s, sector %llu\n",
2030 req
->rq_disk
? req
->rq_disk
->disk_name
: "?",
2031 (unsigned long long)blk_rq_pos(req
));
2034 blk_account_io_completion(req
, nr_bytes
);
2036 total_bytes
= bio_nbytes
= 0;
2037 while ((bio
= req
->bio
) != NULL
) {
2040 if (nr_bytes
>= bio
->bi_size
) {
2041 req
->bio
= bio
->bi_next
;
2042 nbytes
= bio
->bi_size
;
2043 req_bio_endio(req
, bio
, nbytes
, error
);
2047 int idx
= bio
->bi_idx
+ next_idx
;
2049 if (unlikely(idx
>= bio
->bi_vcnt
)) {
2050 blk_dump_rq_flags(req
, "__end_that");
2051 printk(KERN_ERR
"%s: bio idx %d >= vcnt %d\n",
2052 __func__
, idx
, bio
->bi_vcnt
);
2056 nbytes
= bio_iovec_idx(bio
, idx
)->bv_len
;
2057 BIO_BUG_ON(nbytes
> bio
->bi_size
);
2060 * not a complete bvec done
2062 if (unlikely(nbytes
> nr_bytes
)) {
2063 bio_nbytes
+= nr_bytes
;
2064 total_bytes
+= nr_bytes
;
2069 * advance to the next vector
2072 bio_nbytes
+= nbytes
;
2075 total_bytes
+= nbytes
;
2081 * end more in this run, or just return 'not-done'
2083 if (unlikely(nr_bytes
<= 0))
2093 * Reset counters so that the request stacking driver
2094 * can find how many bytes remain in the request
2097 req
->__data_len
= 0;
2102 * if the request wasn't completed, update state
2105 req_bio_endio(req
, bio
, bio_nbytes
, error
);
2106 bio
->bi_idx
+= next_idx
;
2107 bio_iovec(bio
)->bv_offset
+= nr_bytes
;
2108 bio_iovec(bio
)->bv_len
-= nr_bytes
;
2111 req
->__data_len
-= total_bytes
;
2112 req
->buffer
= bio_data(req
->bio
);
2114 /* update sector only for requests with clear definition of sector */
2115 if (req
->cmd_type
== REQ_TYPE_FS
|| (req
->cmd_flags
& REQ_DISCARD
))
2116 req
->__sector
+= total_bytes
>> 9;
2118 /* mixed attributes always follow the first bio */
2119 if (req
->cmd_flags
& REQ_MIXED_MERGE
) {
2120 req
->cmd_flags
&= ~REQ_FAILFAST_MASK
;
2121 req
->cmd_flags
|= req
->bio
->bi_rw
& REQ_FAILFAST_MASK
;
2125 * If total number of sectors is less than the first segment
2126 * size, something has gone terribly wrong.
2128 if (blk_rq_bytes(req
) < blk_rq_cur_bytes(req
)) {
2129 printk(KERN_ERR
"blk: request botched\n");
2130 req
->__data_len
= blk_rq_cur_bytes(req
);
2133 /* recalculate the number of segments */
2134 blk_recalc_rq_segments(req
);
2138 EXPORT_SYMBOL_GPL(blk_update_request
);
2140 static bool blk_update_bidi_request(struct request
*rq
, int error
,
2141 unsigned int nr_bytes
,
2142 unsigned int bidi_bytes
)
2144 if (blk_update_request(rq
, error
, nr_bytes
))
2147 /* Bidi request must be completed as a whole */
2148 if (unlikely(blk_bidi_rq(rq
)) &&
2149 blk_update_request(rq
->next_rq
, error
, bidi_bytes
))
2152 if (blk_queue_add_random(rq
->q
))
2153 add_disk_randomness(rq
->rq_disk
);
2159 * blk_unprep_request - unprepare a request
2162 * This function makes a request ready for complete resubmission (or
2163 * completion). It happens only after all error handling is complete,
2164 * so represents the appropriate moment to deallocate any resources
2165 * that were allocated to the request in the prep_rq_fn. The queue
2166 * lock is held when calling this.
2168 void blk_unprep_request(struct request
*req
)
2170 struct request_queue
*q
= req
->q
;
2172 req
->cmd_flags
&= ~REQ_DONTPREP
;
2173 if (q
->unprep_rq_fn
)
2174 q
->unprep_rq_fn(q
, req
);
2176 EXPORT_SYMBOL_GPL(blk_unprep_request
);
2179 * queue lock must be held
2181 static void blk_finish_request(struct request
*req
, int error
)
2183 if (blk_rq_tagged(req
))
2184 blk_queue_end_tag(req
->q
, req
);
2186 BUG_ON(blk_queued_rq(req
));
2188 if (unlikely(laptop_mode
) && req
->cmd_type
== REQ_TYPE_FS
)
2189 laptop_io_completion(&req
->q
->backing_dev_info
);
2191 blk_delete_timer(req
);
2193 if (req
->cmd_flags
& REQ_DONTPREP
)
2194 blk_unprep_request(req
);
2197 blk_account_io_done(req
);
2200 req
->end_io(req
, error
);
2202 if (blk_bidi_rq(req
))
2203 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
2205 __blk_put_request(req
->q
, req
);
2210 * blk_end_bidi_request - Complete a bidi request
2211 * @rq: the request to complete
2212 * @error: %0 for success, < %0 for error
2213 * @nr_bytes: number of bytes to complete @rq
2214 * @bidi_bytes: number of bytes to complete @rq->next_rq
2217 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2218 * Drivers that supports bidi can safely call this member for any
2219 * type of request, bidi or uni. In the later case @bidi_bytes is
2223 * %false - we are done with this request
2224 * %true - still buffers pending for this request
2226 static bool blk_end_bidi_request(struct request
*rq
, int error
,
2227 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2229 struct request_queue
*q
= rq
->q
;
2230 unsigned long flags
;
2232 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2235 spin_lock_irqsave(q
->queue_lock
, flags
);
2236 blk_finish_request(rq
, error
);
2237 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2243 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2244 * @rq: the request to complete
2245 * @error: %0 for success, < %0 for error
2246 * @nr_bytes: number of bytes to complete @rq
2247 * @bidi_bytes: number of bytes to complete @rq->next_rq
2250 * Identical to blk_end_bidi_request() except that queue lock is
2251 * assumed to be locked on entry and remains so on return.
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 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2263 blk_finish_request(rq
, error
);
2269 * blk_end_request - Helper function for drivers to complete the request.
2270 * @rq: the request being processed
2271 * @error: %0 for success, < %0 for error
2272 * @nr_bytes: number of bytes to complete
2275 * Ends I/O on a number of bytes attached to @rq.
2276 * If @rq has leftover, sets it up for the next range of segments.
2279 * %false - we are done with this request
2280 * %true - still buffers pending for this request
2282 bool blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2284 return blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2286 EXPORT_SYMBOL(blk_end_request
);
2289 * blk_end_request_all - Helper function for drives to finish the request.
2290 * @rq: the request to finish
2291 * @error: %0 for success, < %0 for error
2294 * Completely finish @rq.
2296 void blk_end_request_all(struct request
*rq
, int error
)
2299 unsigned int bidi_bytes
= 0;
2301 if (unlikely(blk_bidi_rq(rq
)))
2302 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2304 pending
= blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2307 EXPORT_SYMBOL(blk_end_request_all
);
2310 * blk_end_request_cur - Helper function to finish the current request chunk.
2311 * @rq: the request to finish the current chunk for
2312 * @error: %0 for success, < %0 for error
2315 * Complete the current consecutively mapped chunk from @rq.
2318 * %false - we are done with this request
2319 * %true - still buffers pending for this request
2321 bool blk_end_request_cur(struct request
*rq
, int error
)
2323 return blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2325 EXPORT_SYMBOL(blk_end_request_cur
);
2328 * blk_end_request_err - Finish a request till the next failure boundary.
2329 * @rq: the request to finish till the next failure boundary for
2330 * @error: must be negative errno
2333 * Complete @rq till the next failure boundary.
2336 * %false - we are done with this request
2337 * %true - still buffers pending for this request
2339 bool blk_end_request_err(struct request
*rq
, int error
)
2341 WARN_ON(error
>= 0);
2342 return blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2344 EXPORT_SYMBOL_GPL(blk_end_request_err
);
2347 * __blk_end_request - Helper function for drivers to complete the request.
2348 * @rq: the request being processed
2349 * @error: %0 for success, < %0 for error
2350 * @nr_bytes: number of bytes to complete
2353 * Must be called with queue lock held unlike blk_end_request().
2356 * %false - we are done with this request
2357 * %true - still buffers pending for this request
2359 bool __blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2361 return __blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2363 EXPORT_SYMBOL(__blk_end_request
);
2366 * __blk_end_request_all - Helper function for drives to finish the request.
2367 * @rq: the request to finish
2368 * @error: %0 for success, < %0 for error
2371 * Completely finish @rq. Must be called with queue lock held.
2373 void __blk_end_request_all(struct request
*rq
, int error
)
2376 unsigned int bidi_bytes
= 0;
2378 if (unlikely(blk_bidi_rq(rq
)))
2379 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2381 pending
= __blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2384 EXPORT_SYMBOL(__blk_end_request_all
);
2387 * __blk_end_request_cur - Helper function to finish the current request chunk.
2388 * @rq: the request to finish the current chunk for
2389 * @error: %0 for success, < %0 for error
2392 * Complete the current consecutively mapped chunk from @rq. Must
2393 * be called with queue lock held.
2396 * %false - we are done with this request
2397 * %true - still buffers pending for this request
2399 bool __blk_end_request_cur(struct request
*rq
, int error
)
2401 return __blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2403 EXPORT_SYMBOL(__blk_end_request_cur
);
2406 * __blk_end_request_err - Finish a request till the next failure boundary.
2407 * @rq: the request to finish till the next failure boundary for
2408 * @error: must be negative errno
2411 * Complete @rq till the next failure boundary. Must be called
2412 * with queue lock held.
2415 * %false - we are done with this request
2416 * %true - still buffers pending for this request
2418 bool __blk_end_request_err(struct request
*rq
, int error
)
2420 WARN_ON(error
>= 0);
2421 return __blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2423 EXPORT_SYMBOL_GPL(__blk_end_request_err
);
2425 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
2428 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2429 rq
->cmd_flags
|= bio
->bi_rw
& REQ_WRITE
;
2431 if (bio_has_data(bio
)) {
2432 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2433 rq
->buffer
= bio_data(bio
);
2435 rq
->__data_len
= bio
->bi_size
;
2436 rq
->bio
= rq
->biotail
= bio
;
2439 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2442 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2444 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2445 * @rq: the request to be flushed
2448 * Flush all pages in @rq.
2450 void rq_flush_dcache_pages(struct request
*rq
)
2452 struct req_iterator iter
;
2453 struct bio_vec
*bvec
;
2455 rq_for_each_segment(bvec
, rq
, iter
)
2456 flush_dcache_page(bvec
->bv_page
);
2458 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages
);
2462 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2463 * @q : the queue of the device being checked
2466 * Check if underlying low-level drivers of a device are busy.
2467 * If the drivers want to export their busy state, they must set own
2468 * exporting function using blk_queue_lld_busy() first.
2470 * Basically, this function is used only by request stacking drivers
2471 * to stop dispatching requests to underlying devices when underlying
2472 * devices are busy. This behavior helps more I/O merging on the queue
2473 * of the request stacking driver and prevents I/O throughput regression
2474 * on burst I/O load.
2477 * 0 - Not busy (The request stacking driver should dispatch request)
2478 * 1 - Busy (The request stacking driver should stop dispatching request)
2480 int blk_lld_busy(struct request_queue
*q
)
2483 return q
->lld_busy_fn(q
);
2487 EXPORT_SYMBOL_GPL(blk_lld_busy
);
2490 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2491 * @rq: the clone request to be cleaned up
2494 * Free all bios in @rq for a cloned request.
2496 void blk_rq_unprep_clone(struct request
*rq
)
2500 while ((bio
= rq
->bio
) != NULL
) {
2501 rq
->bio
= bio
->bi_next
;
2506 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone
);
2509 * Copy attributes of the original request to the clone request.
2510 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2512 static void __blk_rq_prep_clone(struct request
*dst
, struct request
*src
)
2514 dst
->cpu
= src
->cpu
;
2515 dst
->cmd_flags
= (src
->cmd_flags
& REQ_CLONE_MASK
) | REQ_NOMERGE
;
2516 dst
->cmd_type
= src
->cmd_type
;
2517 dst
->__sector
= blk_rq_pos(src
);
2518 dst
->__data_len
= blk_rq_bytes(src
);
2519 dst
->nr_phys_segments
= src
->nr_phys_segments
;
2520 dst
->ioprio
= src
->ioprio
;
2521 dst
->extra_len
= src
->extra_len
;
2525 * blk_rq_prep_clone - Helper function to setup clone request
2526 * @rq: the request to be setup
2527 * @rq_src: original request to be cloned
2528 * @bs: bio_set that bios for clone are allocated from
2529 * @gfp_mask: memory allocation mask for bio
2530 * @bio_ctr: setup function to be called for each clone bio.
2531 * Returns %0 for success, non %0 for failure.
2532 * @data: private data to be passed to @bio_ctr
2535 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2536 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2537 * are not copied, and copying such parts is the caller's responsibility.
2538 * Also, pages which the original bios are pointing to are not copied
2539 * and the cloned bios just point same pages.
2540 * So cloned bios must be completed before original bios, which means
2541 * the caller must complete @rq before @rq_src.
2543 int blk_rq_prep_clone(struct request
*rq
, struct request
*rq_src
,
2544 struct bio_set
*bs
, gfp_t gfp_mask
,
2545 int (*bio_ctr
)(struct bio
*, struct bio
*, void *),
2548 struct bio
*bio
, *bio_src
;
2553 blk_rq_init(NULL
, rq
);
2555 __rq_for_each_bio(bio_src
, rq_src
) {
2556 bio
= bio_alloc_bioset(gfp_mask
, bio_src
->bi_max_vecs
, bs
);
2560 __bio_clone(bio
, bio_src
);
2562 if (bio_integrity(bio_src
) &&
2563 bio_integrity_clone(bio
, bio_src
, gfp_mask
, bs
))
2566 if (bio_ctr
&& bio_ctr(bio
, bio_src
, data
))
2570 rq
->biotail
->bi_next
= bio
;
2573 rq
->bio
= rq
->biotail
= bio
;
2576 __blk_rq_prep_clone(rq
, rq_src
);
2583 blk_rq_unprep_clone(rq
);
2587 EXPORT_SYMBOL_GPL(blk_rq_prep_clone
);
2589 int kblockd_schedule_work(struct request_queue
*q
, struct work_struct
*work
)
2591 return queue_work(kblockd_workqueue
, work
);
2593 EXPORT_SYMBOL(kblockd_schedule_work
);
2595 int kblockd_schedule_delayed_work(struct request_queue
*q
,
2596 struct delayed_work
*dwork
, unsigned long delay
)
2598 return queue_delayed_work(kblockd_workqueue
, dwork
, delay
);
2600 EXPORT_SYMBOL(kblockd_schedule_delayed_work
);
2602 #define PLUG_MAGIC 0x91827364
2604 void blk_start_plug(struct blk_plug
*plug
)
2606 struct task_struct
*tsk
= current
;
2608 plug
->magic
= PLUG_MAGIC
;
2609 INIT_LIST_HEAD(&plug
->list
);
2610 plug
->should_sort
= 0;
2613 * If this is a nested plug, don't actually assign it. It will be
2614 * flushed on its own.
2618 * Store ordering should not be needed here, since a potential
2619 * preempt will imply a full memory barrier
2624 EXPORT_SYMBOL(blk_start_plug
);
2626 static int plug_rq_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
2628 struct request
*rqa
= container_of(a
, struct request
, queuelist
);
2629 struct request
*rqb
= container_of(b
, struct request
, queuelist
);
2631 return !(rqa
->q
== rqb
->q
);
2634 static void flush_plug_list(struct blk_plug
*plug
)
2636 struct request_queue
*q
;
2637 unsigned long flags
;
2640 BUG_ON(plug
->magic
!= PLUG_MAGIC
);
2642 if (list_empty(&plug
->list
))
2645 if (plug
->should_sort
)
2646 list_sort(NULL
, &plug
->list
, plug_rq_cmp
);
2649 local_irq_save(flags
);
2650 while (!list_empty(&plug
->list
)) {
2651 rq
= list_entry_rq(plug
->list
.next
);
2652 list_del_init(&rq
->queuelist
);
2653 BUG_ON(!(rq
->cmd_flags
& REQ_ON_PLUG
));
2658 spin_unlock(q
->queue_lock
);
2661 spin_lock(q
->queue_lock
);
2663 rq
->cmd_flags
&= ~REQ_ON_PLUG
;
2666 * rq is already accounted, so use raw insert
2668 __elv_add_request(q
, rq
, ELEVATOR_INSERT_SORT
);
2673 spin_unlock(q
->queue_lock
);
2676 BUG_ON(!list_empty(&plug
->list
));
2677 local_irq_restore(flags
);
2680 static void __blk_finish_plug(struct task_struct
*tsk
, struct blk_plug
*plug
)
2682 flush_plug_list(plug
);
2684 if (plug
== tsk
->plug
)
2688 void blk_finish_plug(struct blk_plug
*plug
)
2691 __blk_finish_plug(current
, plug
);
2693 EXPORT_SYMBOL(blk_finish_plug
);
2695 void __blk_flush_plug(struct task_struct
*tsk
, struct blk_plug
*plug
)
2697 __blk_finish_plug(tsk
, plug
);
2700 EXPORT_SYMBOL(__blk_flush_plug
);
2702 int __init
blk_dev_init(void)
2704 BUILD_BUG_ON(__REQ_NR_BITS
> 8 *
2705 sizeof(((struct request
*)0)->cmd_flags
));
2707 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
2708 kblockd_workqueue
= alloc_workqueue("kblockd",
2709 WQ_MEM_RECLAIM
| WQ_HIGHPRI
, 0);
2710 if (!kblockd_workqueue
)
2711 panic("Failed to create kblockd\n");
2713 request_cachep
= kmem_cache_create("blkdev_requests",
2714 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
2716 blk_requestq_cachep
= kmem_cache_create("blkdev_queue",
2717 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);