2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
6 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
8 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
12 * This handles all read/write requests to block devices
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/backing-dev.h>
17 #include <linux/bio.h>
18 #include <linux/blkdev.h>
19 #include <linux/highmem.h>
21 #include <linux/kernel_stat.h>
22 #include <linux/string.h>
23 #include <linux/init.h>
24 #include <linux/completion.h>
25 #include <linux/slab.h>
26 #include <linux/swap.h>
27 #include <linux/writeback.h>
28 #include <linux/task_io_accounting_ops.h>
29 #include <linux/fault-inject.h>
30 #include <linux/list_sort.h>
31 #include <linux/delay.h>
33 #define CREATE_TRACE_POINTS
34 #include <trace/events/block.h>
38 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap
);
39 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap
);
40 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete
);
42 DEFINE_IDA(blk_queue_ida
);
45 * For the allocated request tables
47 static struct kmem_cache
*request_cachep
;
50 * For queue allocation
52 struct kmem_cache
*blk_requestq_cachep
;
55 * Controlling structure to kblockd
57 static struct workqueue_struct
*kblockd_workqueue
;
59 static void drive_stat_acct(struct request
*rq
, int new_io
)
61 struct hd_struct
*part
;
62 int rw
= rq_data_dir(rq
);
65 if (!blk_do_io_stat(rq
))
68 cpu
= part_stat_lock();
72 part_stat_inc(cpu
, part
, merges
[rw
]);
74 part
= disk_map_sector_rcu(rq
->rq_disk
, blk_rq_pos(rq
));
75 if (!hd_struct_try_get(part
)) {
77 * The partition is already being removed,
78 * the request will be accounted on the disk only
80 * We take a reference on disk->part0 although that
81 * partition will never be deleted, so we can treat
82 * it as any other partition.
84 part
= &rq
->rq_disk
->part0
;
87 part_round_stats(cpu
, part
);
88 part_inc_in_flight(part
, rw
);
95 void blk_queue_congestion_threshold(struct request_queue
*q
)
99 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) + 1;
100 if (nr
> q
->nr_requests
)
102 q
->nr_congestion_on
= nr
;
104 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) - (q
->nr_requests
/ 16) - 1;
107 q
->nr_congestion_off
= nr
;
111 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
114 * Locates the passed device's request queue and returns the address of its
117 * Will return NULL if the request queue cannot be located.
119 struct backing_dev_info
*blk_get_backing_dev_info(struct block_device
*bdev
)
121 struct backing_dev_info
*ret
= NULL
;
122 struct request_queue
*q
= bdev_get_queue(bdev
);
125 ret
= &q
->backing_dev_info
;
128 EXPORT_SYMBOL(blk_get_backing_dev_info
);
130 void blk_rq_init(struct request_queue
*q
, struct request
*rq
)
132 memset(rq
, 0, sizeof(*rq
));
134 INIT_LIST_HEAD(&rq
->queuelist
);
135 INIT_LIST_HEAD(&rq
->timeout_list
);
138 rq
->__sector
= (sector_t
) -1;
139 INIT_HLIST_NODE(&rq
->hash
);
140 RB_CLEAR_NODE(&rq
->rb_node
);
142 rq
->cmd_len
= BLK_MAX_CDB
;
145 rq
->start_time
= jiffies
;
146 set_start_time_ns(rq
);
149 EXPORT_SYMBOL(blk_rq_init
);
151 static void req_bio_endio(struct request
*rq
, struct bio
*bio
,
152 unsigned int nbytes
, int error
)
155 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
156 else if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
159 if (unlikely(nbytes
> bio
->bi_size
)) {
160 printk(KERN_ERR
"%s: want %u bytes done, %u left\n",
161 __func__
, nbytes
, bio
->bi_size
);
162 nbytes
= bio
->bi_size
;
165 if (unlikely(rq
->cmd_flags
& REQ_QUIET
))
166 set_bit(BIO_QUIET
, &bio
->bi_flags
);
168 bio
->bi_size
-= nbytes
;
169 bio
->bi_sector
+= (nbytes
>> 9);
171 if (bio_integrity(bio
))
172 bio_integrity_advance(bio
, nbytes
);
174 /* don't actually finish bio if it's part of flush sequence */
175 if (bio
->bi_size
== 0 && !(rq
->cmd_flags
& REQ_FLUSH_SEQ
))
176 bio_endio(bio
, error
);
179 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
183 printk(KERN_INFO
"%s: dev %s: type=%x, flags=%x\n", msg
,
184 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->cmd_type
,
187 printk(KERN_INFO
" sector %llu, nr/cnr %u/%u\n",
188 (unsigned long long)blk_rq_pos(rq
),
189 blk_rq_sectors(rq
), blk_rq_cur_sectors(rq
));
190 printk(KERN_INFO
" bio %p, biotail %p, buffer %p, len %u\n",
191 rq
->bio
, rq
->biotail
, rq
->buffer
, blk_rq_bytes(rq
));
193 if (rq
->cmd_type
== REQ_TYPE_BLOCK_PC
) {
194 printk(KERN_INFO
" cdb: ");
195 for (bit
= 0; bit
< BLK_MAX_CDB
; bit
++)
196 printk("%02x ", rq
->cmd
[bit
]);
200 EXPORT_SYMBOL(blk_dump_rq_flags
);
202 static void blk_delay_work(struct work_struct
*work
)
204 struct request_queue
*q
;
206 q
= container_of(work
, struct request_queue
, delay_work
.work
);
207 spin_lock_irq(q
->queue_lock
);
209 spin_unlock_irq(q
->queue_lock
);
213 * blk_delay_queue - restart queueing after defined interval
214 * @q: The &struct request_queue in question
215 * @msecs: Delay in msecs
218 * Sometimes queueing needs to be postponed for a little while, to allow
219 * resources to come back. This function will make sure that queueing is
220 * restarted around the specified time.
222 void blk_delay_queue(struct request_queue
*q
, unsigned long msecs
)
224 queue_delayed_work(kblockd_workqueue
, &q
->delay_work
,
225 msecs_to_jiffies(msecs
));
227 EXPORT_SYMBOL(blk_delay_queue
);
230 * blk_start_queue - restart a previously stopped queue
231 * @q: The &struct request_queue in question
234 * blk_start_queue() will clear the stop flag on the queue, and call
235 * the request_fn for the queue if it was in a stopped state when
236 * entered. Also see blk_stop_queue(). Queue lock must be held.
238 void blk_start_queue(struct request_queue
*q
)
240 WARN_ON(!irqs_disabled());
242 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
245 EXPORT_SYMBOL(blk_start_queue
);
248 * blk_stop_queue - stop a queue
249 * @q: The &struct request_queue in question
252 * The Linux block layer assumes that a block driver will consume all
253 * entries on the request queue when the request_fn strategy is called.
254 * Often this will not happen, because of hardware limitations (queue
255 * depth settings). If a device driver gets a 'queue full' response,
256 * or if it simply chooses not to queue more I/O at one point, it can
257 * call this function to prevent the request_fn from being called until
258 * the driver has signalled it's ready to go again. This happens by calling
259 * blk_start_queue() to restart queue operations. Queue lock must be held.
261 void blk_stop_queue(struct request_queue
*q
)
263 __cancel_delayed_work(&q
->delay_work
);
264 queue_flag_set(QUEUE_FLAG_STOPPED
, q
);
266 EXPORT_SYMBOL(blk_stop_queue
);
269 * blk_sync_queue - cancel any pending callbacks on a queue
273 * The block layer may perform asynchronous callback activity
274 * on a queue, such as calling the unplug function after a timeout.
275 * A block device may call blk_sync_queue to ensure that any
276 * such activity is cancelled, thus allowing it to release resources
277 * that the callbacks might use. The caller must already have made sure
278 * that its ->make_request_fn will not re-add plugging prior to calling
281 * This function does not cancel any asynchronous activity arising
282 * out of elevator or throttling code. That would require elevaotor_exit()
283 * and blk_throtl_exit() to be called with queue lock initialized.
286 void blk_sync_queue(struct request_queue
*q
)
288 del_timer_sync(&q
->timeout
);
289 cancel_delayed_work_sync(&q
->delay_work
);
291 EXPORT_SYMBOL(blk_sync_queue
);
294 * __blk_run_queue - run a single device queue
295 * @q: The queue to run
298 * See @blk_run_queue. This variant must be called with the queue lock
299 * held and interrupts disabled.
301 void __blk_run_queue(struct request_queue
*q
)
303 if (unlikely(blk_queue_stopped(q
)))
308 EXPORT_SYMBOL(__blk_run_queue
);
311 * blk_run_queue_async - run a single device queue in workqueue context
312 * @q: The queue to run
315 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
318 void blk_run_queue_async(struct request_queue
*q
)
320 if (likely(!blk_queue_stopped(q
))) {
321 __cancel_delayed_work(&q
->delay_work
);
322 queue_delayed_work(kblockd_workqueue
, &q
->delay_work
, 0);
325 EXPORT_SYMBOL(blk_run_queue_async
);
328 * blk_run_queue - run a single device queue
329 * @q: The queue to run
332 * Invoke request handling on this queue, if it has pending work to do.
333 * May be used to restart queueing when a request has completed.
335 void blk_run_queue(struct request_queue
*q
)
339 spin_lock_irqsave(q
->queue_lock
, flags
);
341 spin_unlock_irqrestore(q
->queue_lock
, flags
);
343 EXPORT_SYMBOL(blk_run_queue
);
345 void blk_put_queue(struct request_queue
*q
)
347 kobject_put(&q
->kobj
);
349 EXPORT_SYMBOL(blk_put_queue
);
352 * blk_drain_queue - drain requests from request_queue
354 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
356 * Drain requests from @q. If @drain_all is set, all requests are drained.
357 * If not, only ELVPRIV requests are drained. The caller is responsible
358 * for ensuring that no new requests which need to be drained are queued.
360 void blk_drain_queue(struct request_queue
*q
, bool drain_all
)
366 spin_lock_irq(q
->queue_lock
);
368 elv_drain_elevator(q
);
374 drain
|= q
->rq
.elvpriv
;
377 * Unfortunately, requests are queued at and tracked from
378 * multiple places and there's no single counter which can
379 * be drained. Check all the queues and counters.
382 drain
|= !list_empty(&q
->queue_head
);
383 for (i
= 0; i
< 2; i
++) {
384 drain
|= q
->rq
.count
[i
];
385 drain
|= q
->in_flight
[i
];
386 drain
|= !list_empty(&q
->flush_queue
[i
]);
390 spin_unlock_irq(q
->queue_lock
);
399 * blk_cleanup_queue - shutdown a request queue
400 * @q: request queue to shutdown
402 * Mark @q DEAD, drain all pending requests, destroy and put it. All
403 * future requests will be failed immediately with -ENODEV.
405 void blk_cleanup_queue(struct request_queue
*q
)
407 spinlock_t
*lock
= q
->queue_lock
;
409 /* mark @q DEAD, no new request or merges will be allowed afterwards */
410 mutex_lock(&q
->sysfs_lock
);
411 queue_flag_set_unlocked(QUEUE_FLAG_DEAD
, q
);
414 queue_flag_set(QUEUE_FLAG_NOMERGES
, q
);
415 queue_flag_set(QUEUE_FLAG_NOXMERGES
, q
);
416 queue_flag_set(QUEUE_FLAG_DEAD
, q
);
418 if (q
->queue_lock
!= &q
->__queue_lock
)
419 q
->queue_lock
= &q
->__queue_lock
;
421 spin_unlock_irq(lock
);
422 mutex_unlock(&q
->sysfs_lock
);
425 * Drain all requests queued before DEAD marking. The caller might
426 * be trying to tear down @q before its elevator is initialized, in
427 * which case we don't want to call into draining.
430 blk_drain_queue(q
, true);
432 /* @q won't process any more request, flush async actions */
433 del_timer_sync(&q
->backing_dev_info
.laptop_mode_wb_timer
);
436 /* @q is and will stay empty, shutdown and put */
439 EXPORT_SYMBOL(blk_cleanup_queue
);
441 static int blk_init_free_list(struct request_queue
*q
)
443 struct request_list
*rl
= &q
->rq
;
445 if (unlikely(rl
->rq_pool
))
448 rl
->count
[BLK_RW_SYNC
] = rl
->count
[BLK_RW_ASYNC
] = 0;
449 rl
->starved
[BLK_RW_SYNC
] = rl
->starved
[BLK_RW_ASYNC
] = 0;
451 init_waitqueue_head(&rl
->wait
[BLK_RW_SYNC
]);
452 init_waitqueue_head(&rl
->wait
[BLK_RW_ASYNC
]);
454 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, mempool_alloc_slab
,
455 mempool_free_slab
, request_cachep
, q
->node
);
463 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
465 return blk_alloc_queue_node(gfp_mask
, -1);
467 EXPORT_SYMBOL(blk_alloc_queue
);
469 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
471 struct request_queue
*q
;
474 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
475 gfp_mask
| __GFP_ZERO
, node_id
);
479 q
->id
= ida_simple_get(&blk_queue_ida
, 0, 0, GFP_KERNEL
);
483 q
->backing_dev_info
.ra_pages
=
484 (VM_MAX_READAHEAD
* 1024) / PAGE_CACHE_SIZE
;
485 q
->backing_dev_info
.state
= 0;
486 q
->backing_dev_info
.capabilities
= BDI_CAP_MAP_COPY
;
487 q
->backing_dev_info
.name
= "block";
489 err
= bdi_init(&q
->backing_dev_info
);
493 if (blk_throtl_init(q
))
496 setup_timer(&q
->backing_dev_info
.laptop_mode_wb_timer
,
497 laptop_mode_timer_fn
, (unsigned long) q
);
498 setup_timer(&q
->timeout
, blk_rq_timed_out_timer
, (unsigned long) q
);
499 INIT_LIST_HEAD(&q
->timeout_list
);
500 INIT_LIST_HEAD(&q
->flush_queue
[0]);
501 INIT_LIST_HEAD(&q
->flush_queue
[1]);
502 INIT_LIST_HEAD(&q
->flush_data_in_flight
);
503 INIT_DELAYED_WORK(&q
->delay_work
, blk_delay_work
);
505 kobject_init(&q
->kobj
, &blk_queue_ktype
);
507 mutex_init(&q
->sysfs_lock
);
508 spin_lock_init(&q
->__queue_lock
);
511 * By default initialize queue_lock to internal lock and driver can
512 * override it later if need be.
514 q
->queue_lock
= &q
->__queue_lock
;
519 ida_simple_remove(&blk_queue_ida
, q
->id
);
521 kmem_cache_free(blk_requestq_cachep
, q
);
524 EXPORT_SYMBOL(blk_alloc_queue_node
);
527 * blk_init_queue - prepare a request queue for use with a block device
528 * @rfn: The function to be called to process requests that have been
529 * placed on the queue.
530 * @lock: Request queue spin lock
533 * If a block device wishes to use the standard request handling procedures,
534 * which sorts requests and coalesces adjacent requests, then it must
535 * call blk_init_queue(). The function @rfn will be called when there
536 * are requests on the queue that need to be processed. If the device
537 * supports plugging, then @rfn may not be called immediately when requests
538 * are available on the queue, but may be called at some time later instead.
539 * Plugged queues are generally unplugged when a buffer belonging to one
540 * of the requests on the queue is needed, or due to memory pressure.
542 * @rfn is not required, or even expected, to remove all requests off the
543 * queue, but only as many as it can handle at a time. If it does leave
544 * requests on the queue, it is responsible for arranging that the requests
545 * get dealt with eventually.
547 * The queue spin lock must be held while manipulating the requests on the
548 * request queue; this lock will be taken also from interrupt context, so irq
549 * disabling is needed for it.
551 * Function returns a pointer to the initialized request queue, or %NULL if
555 * blk_init_queue() must be paired with a blk_cleanup_queue() call
556 * when the block device is deactivated (such as at module unload).
559 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
561 return blk_init_queue_node(rfn
, lock
, -1);
563 EXPORT_SYMBOL(blk_init_queue
);
565 struct request_queue
*
566 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
568 struct request_queue
*uninit_q
, *q
;
570 uninit_q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
574 q
= blk_init_allocated_queue_node(uninit_q
, rfn
, lock
, node_id
);
576 blk_cleanup_queue(uninit_q
);
580 EXPORT_SYMBOL(blk_init_queue_node
);
582 struct request_queue
*
583 blk_init_allocated_queue(struct request_queue
*q
, request_fn_proc
*rfn
,
586 return blk_init_allocated_queue_node(q
, rfn
, lock
, -1);
588 EXPORT_SYMBOL(blk_init_allocated_queue
);
590 struct request_queue
*
591 blk_init_allocated_queue_node(struct request_queue
*q
, request_fn_proc
*rfn
,
592 spinlock_t
*lock
, int node_id
)
598 if (blk_init_free_list(q
))
602 q
->prep_rq_fn
= NULL
;
603 q
->unprep_rq_fn
= NULL
;
604 q
->queue_flags
= QUEUE_FLAG_DEFAULT
;
606 /* Override internal queue lock with supplied lock pointer */
608 q
->queue_lock
= lock
;
611 * This also sets hw/phys segments, boundary and size
613 blk_queue_make_request(q
, blk_queue_bio
);
615 q
->sg_reserved_size
= INT_MAX
;
620 if (!elevator_init(q
, NULL
)) {
621 blk_queue_congestion_threshold(q
);
627 EXPORT_SYMBOL(blk_init_allocated_queue_node
);
629 bool blk_get_queue(struct request_queue
*q
)
631 if (likely(!blk_queue_dead(q
))) {
638 EXPORT_SYMBOL(blk_get_queue
);
640 static inline void blk_free_request(struct request_queue
*q
, struct request
*rq
)
642 if (rq
->cmd_flags
& REQ_ELVPRIV
)
643 elv_put_request(q
, rq
);
644 mempool_free(rq
, q
->rq
.rq_pool
);
647 static struct request
*
648 blk_alloc_request(struct request_queue
*q
, unsigned int flags
, gfp_t gfp_mask
)
650 struct request
*rq
= mempool_alloc(q
->rq
.rq_pool
, gfp_mask
);
657 rq
->cmd_flags
= flags
| REQ_ALLOCED
;
659 if ((flags
& REQ_ELVPRIV
) &&
660 unlikely(elv_set_request(q
, rq
, gfp_mask
))) {
661 mempool_free(rq
, q
->rq
.rq_pool
);
669 * ioc_batching returns true if the ioc is a valid batching request and
670 * should be given priority access to a request.
672 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
678 * Make sure the process is able to allocate at least 1 request
679 * even if the batch times out, otherwise we could theoretically
682 return ioc
->nr_batch_requests
== q
->nr_batching
||
683 (ioc
->nr_batch_requests
> 0
684 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
688 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
689 * will cause the process to be a "batcher" on all queues in the system. This
690 * is the behaviour we want though - once it gets a wakeup it should be given
693 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
695 if (!ioc
|| ioc_batching(q
, ioc
))
698 ioc
->nr_batch_requests
= q
->nr_batching
;
699 ioc
->last_waited
= jiffies
;
702 static void __freed_request(struct request_queue
*q
, int sync
)
704 struct request_list
*rl
= &q
->rq
;
706 if (rl
->count
[sync
] < queue_congestion_off_threshold(q
))
707 blk_clear_queue_congested(q
, sync
);
709 if (rl
->count
[sync
] + 1 <= q
->nr_requests
) {
710 if (waitqueue_active(&rl
->wait
[sync
]))
711 wake_up(&rl
->wait
[sync
]);
713 blk_clear_queue_full(q
, sync
);
718 * A request has just been released. Account for it, update the full and
719 * congestion status, wake up any waiters. Called under q->queue_lock.
721 static void freed_request(struct request_queue
*q
, unsigned int flags
)
723 struct request_list
*rl
= &q
->rq
;
724 int sync
= rw_is_sync(flags
);
727 if (flags
& REQ_ELVPRIV
)
730 __freed_request(q
, sync
);
732 if (unlikely(rl
->starved
[sync
^ 1]))
733 __freed_request(q
, sync
^ 1);
737 * Determine if elevator data should be initialized when allocating the
738 * request associated with @bio.
740 static bool blk_rq_should_init_elevator(struct bio
*bio
)
746 * Flush requests do not use the elevator so skip initialization.
747 * This allows a request to share the flush and elevator data.
749 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
))
756 * get_request - get a free request
757 * @q: request_queue to allocate request from
758 * @rw_flags: RW and SYNC flags
759 * @bio: bio to allocate request for (can be %NULL)
760 * @gfp_mask: allocation mask
762 * Get a free request from @q. This function may fail under memory
763 * pressure or if @q is dead.
765 * Must be callled with @q->queue_lock held and,
766 * Returns %NULL on failure, with @q->queue_lock held.
767 * Returns !%NULL on success, with @q->queue_lock *not held*.
769 static struct request
*get_request(struct request_queue
*q
, int rw_flags
,
770 struct bio
*bio
, gfp_t gfp_mask
)
772 struct request
*rq
= NULL
;
773 struct request_list
*rl
= &q
->rq
;
774 struct io_context
*ioc
;
775 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
776 bool retried
= false;
779 ioc
= current
->io_context
;
781 if (unlikely(blk_queue_dead(q
)))
784 may_queue
= elv_may_queue(q
, rw_flags
);
785 if (may_queue
== ELV_MQUEUE_NO
)
788 if (rl
->count
[is_sync
]+1 >= queue_congestion_on_threshold(q
)) {
789 if (rl
->count
[is_sync
]+1 >= q
->nr_requests
) {
791 * We want ioc to record batching state. If it's
792 * not already there, creating a new one requires
793 * dropping queue_lock, which in turn requires
794 * retesting conditions to avoid queue hang.
796 if (!ioc
&& !retried
) {
797 spin_unlock_irq(q
->queue_lock
);
798 create_io_context(current
, gfp_mask
, q
->node
);
799 spin_lock_irq(q
->queue_lock
);
805 * The queue will fill after this allocation, so set
806 * it as full, and mark this process as "batching".
807 * This process will be allowed to complete a batch of
808 * requests, others will be blocked.
810 if (!blk_queue_full(q
, is_sync
)) {
811 ioc_set_batching(q
, ioc
);
812 blk_set_queue_full(q
, is_sync
);
814 if (may_queue
!= ELV_MQUEUE_MUST
815 && !ioc_batching(q
, ioc
)) {
817 * The queue is full and the allocating
818 * process is not a "batcher", and not
819 * exempted by the IO scheduler
825 blk_set_queue_congested(q
, is_sync
);
829 * Only allow batching queuers to allocate up to 50% over the defined
830 * limit of requests, otherwise we could have thousands of requests
831 * allocated with any setting of ->nr_requests
833 if (rl
->count
[is_sync
] >= (3 * q
->nr_requests
/ 2))
836 rl
->count
[is_sync
]++;
837 rl
->starved
[is_sync
] = 0;
839 if (blk_rq_should_init_elevator(bio
) &&
840 !test_bit(QUEUE_FLAG_ELVSWITCH
, &q
->queue_flags
)) {
841 rw_flags
|= REQ_ELVPRIV
;
845 if (blk_queue_io_stat(q
))
846 rw_flags
|= REQ_IO_STAT
;
847 spin_unlock_irq(q
->queue_lock
);
849 rq
= blk_alloc_request(q
, rw_flags
, gfp_mask
);
852 * Allocation failed presumably due to memory. Undo anything
853 * we might have messed up.
855 * Allocating task should really be put onto the front of the
856 * wait queue, but this is pretty rare.
858 spin_lock_irq(q
->queue_lock
);
859 freed_request(q
, rw_flags
);
862 * in the very unlikely event that allocation failed and no
863 * requests for this direction was pending, mark us starved
864 * so that freeing of a request in the other direction will
865 * notice us. another possible fix would be to split the
866 * rq mempool into READ and WRITE
869 if (unlikely(rl
->count
[is_sync
] == 0))
870 rl
->starved
[is_sync
] = 1;
876 * ioc may be NULL here, and ioc_batching will be false. That's
877 * OK, if the queue is under the request limit then requests need
878 * not count toward the nr_batch_requests limit. There will always
879 * be some limit enforced by BLK_BATCH_TIME.
881 if (ioc_batching(q
, ioc
))
882 ioc
->nr_batch_requests
--;
884 trace_block_getrq(q
, bio
, rw_flags
& 1);
890 * get_request_wait - get a free request with retry
891 * @q: request_queue to allocate request from
892 * @rw_flags: RW and SYNC flags
893 * @bio: bio to allocate request for (can be %NULL)
895 * Get a free request from @q. This function keeps retrying under memory
896 * pressure and fails iff @q is dead.
898 * Must be callled with @q->queue_lock held and,
899 * Returns %NULL on failure, with @q->queue_lock held.
900 * Returns !%NULL on success, with @q->queue_lock *not held*.
902 static struct request
*get_request_wait(struct request_queue
*q
, int rw_flags
,
905 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
908 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
911 struct request_list
*rl
= &q
->rq
;
913 if (unlikely(blk_queue_dead(q
)))
916 prepare_to_wait_exclusive(&rl
->wait
[is_sync
], &wait
,
917 TASK_UNINTERRUPTIBLE
);
919 trace_block_sleeprq(q
, bio
, rw_flags
& 1);
921 spin_unlock_irq(q
->queue_lock
);
925 * After sleeping, we become a "batching" process and
926 * will be able to allocate at least one request, and
927 * up to a big batch of them for a small period time.
928 * See ioc_batching, ioc_set_batching
930 create_io_context(current
, GFP_NOIO
, q
->node
);
931 ioc_set_batching(q
, current
->io_context
);
933 spin_lock_irq(q
->queue_lock
);
934 finish_wait(&rl
->wait
[is_sync
], &wait
);
936 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
942 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
946 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
948 spin_lock_irq(q
->queue_lock
);
949 if (gfp_mask
& __GFP_WAIT
)
950 rq
= get_request_wait(q
, rw
, NULL
);
952 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
954 spin_unlock_irq(q
->queue_lock
);
955 /* q->queue_lock is unlocked at this point */
959 EXPORT_SYMBOL(blk_get_request
);
962 * blk_make_request - given a bio, allocate a corresponding struct request.
963 * @q: target request queue
964 * @bio: The bio describing the memory mappings that will be submitted for IO.
965 * It may be a chained-bio properly constructed by block/bio layer.
966 * @gfp_mask: gfp flags to be used for memory allocation
968 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
969 * type commands. Where the struct request needs to be farther initialized by
970 * the caller. It is passed a &struct bio, which describes the memory info of
973 * The caller of blk_make_request must make sure that bi_io_vec
974 * are set to describe the memory buffers. That bio_data_dir() will return
975 * the needed direction of the request. (And all bio's in the passed bio-chain
976 * are properly set accordingly)
978 * If called under none-sleepable conditions, mapped bio buffers must not
979 * need bouncing, by calling the appropriate masked or flagged allocator,
980 * suitable for the target device. Otherwise the call to blk_queue_bounce will
983 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
984 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
985 * anything but the first bio in the chain. Otherwise you risk waiting for IO
986 * completion of a bio that hasn't been submitted yet, thus resulting in a
987 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
988 * of bio_alloc(), as that avoids the mempool deadlock.
989 * If possible a big IO should be split into smaller parts when allocation
990 * fails. Partial allocation should not be an error, or you risk a live-lock.
992 struct request
*blk_make_request(struct request_queue
*q
, struct bio
*bio
,
995 struct request
*rq
= blk_get_request(q
, bio_data_dir(bio
), gfp_mask
);
998 return ERR_PTR(-ENOMEM
);
1001 struct bio
*bounce_bio
= bio
;
1004 blk_queue_bounce(q
, &bounce_bio
);
1005 ret
= blk_rq_append_bio(q
, rq
, bounce_bio
);
1006 if (unlikely(ret
)) {
1007 blk_put_request(rq
);
1008 return ERR_PTR(ret
);
1014 EXPORT_SYMBOL(blk_make_request
);
1017 * blk_requeue_request - put a request back on queue
1018 * @q: request queue where request should be inserted
1019 * @rq: request to be inserted
1022 * Drivers often keep queueing requests until the hardware cannot accept
1023 * more, when that condition happens we need to put the request back
1024 * on the queue. Must be called with queue lock held.
1026 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
1028 blk_delete_timer(rq
);
1029 blk_clear_rq_complete(rq
);
1030 trace_block_rq_requeue(q
, rq
);
1032 if (blk_rq_tagged(rq
))
1033 blk_queue_end_tag(q
, rq
);
1035 BUG_ON(blk_queued_rq(rq
));
1037 elv_requeue_request(q
, rq
);
1039 EXPORT_SYMBOL(blk_requeue_request
);
1041 static void add_acct_request(struct request_queue
*q
, struct request
*rq
,
1044 drive_stat_acct(rq
, 1);
1045 __elv_add_request(q
, rq
, where
);
1048 static void part_round_stats_single(int cpu
, struct hd_struct
*part
,
1051 if (now
== part
->stamp
)
1054 if (part_in_flight(part
)) {
1055 __part_stat_add(cpu
, part
, time_in_queue
,
1056 part_in_flight(part
) * (now
- part
->stamp
));
1057 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
1063 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1064 * @cpu: cpu number for stats access
1065 * @part: target partition
1067 * The average IO queue length and utilisation statistics are maintained
1068 * by observing the current state of the queue length and the amount of
1069 * time it has been in this state for.
1071 * Normally, that accounting is done on IO completion, but that can result
1072 * in more than a second's worth of IO being accounted for within any one
1073 * second, leading to >100% utilisation. To deal with that, we call this
1074 * function to do a round-off before returning the results when reading
1075 * /proc/diskstats. This accounts immediately for all queue usage up to
1076 * the current jiffies and restarts the counters again.
1078 void part_round_stats(int cpu
, struct hd_struct
*part
)
1080 unsigned long now
= jiffies
;
1083 part_round_stats_single(cpu
, &part_to_disk(part
)->part0
, now
);
1084 part_round_stats_single(cpu
, part
, now
);
1086 EXPORT_SYMBOL_GPL(part_round_stats
);
1089 * queue lock must be held
1091 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1095 if (unlikely(--req
->ref_count
))
1098 elv_completed_request(q
, req
);
1100 /* this is a bio leak */
1101 WARN_ON(req
->bio
!= NULL
);
1104 * Request may not have originated from ll_rw_blk. if not,
1105 * it didn't come out of our reserved rq pools
1107 if (req
->cmd_flags
& REQ_ALLOCED
) {
1108 unsigned int flags
= req
->cmd_flags
;
1110 BUG_ON(!list_empty(&req
->queuelist
));
1111 BUG_ON(!hlist_unhashed(&req
->hash
));
1113 blk_free_request(q
, req
);
1114 freed_request(q
, flags
);
1117 EXPORT_SYMBOL_GPL(__blk_put_request
);
1119 void blk_put_request(struct request
*req
)
1121 unsigned long flags
;
1122 struct request_queue
*q
= req
->q
;
1124 spin_lock_irqsave(q
->queue_lock
, flags
);
1125 __blk_put_request(q
, req
);
1126 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1128 EXPORT_SYMBOL(blk_put_request
);
1131 * blk_add_request_payload - add a payload to a request
1132 * @rq: request to update
1133 * @page: page backing the payload
1134 * @len: length of the payload.
1136 * This allows to later add a payload to an already submitted request by
1137 * a block driver. The driver needs to take care of freeing the payload
1140 * Note that this is a quite horrible hack and nothing but handling of
1141 * discard requests should ever use it.
1143 void blk_add_request_payload(struct request
*rq
, struct page
*page
,
1146 struct bio
*bio
= rq
->bio
;
1148 bio
->bi_io_vec
->bv_page
= page
;
1149 bio
->bi_io_vec
->bv_offset
= 0;
1150 bio
->bi_io_vec
->bv_len
= len
;
1154 bio
->bi_phys_segments
= 1;
1156 rq
->__data_len
= rq
->resid_len
= len
;
1157 rq
->nr_phys_segments
= 1;
1158 rq
->buffer
= bio_data(bio
);
1160 EXPORT_SYMBOL_GPL(blk_add_request_payload
);
1162 static bool bio_attempt_back_merge(struct request_queue
*q
, struct request
*req
,
1165 const int ff
= bio
->bi_rw
& REQ_FAILFAST_MASK
;
1167 if (!ll_back_merge_fn(q
, req
, bio
))
1170 trace_block_bio_backmerge(q
, bio
);
1172 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1173 blk_rq_set_mixed_merge(req
);
1175 req
->biotail
->bi_next
= bio
;
1177 req
->__data_len
+= bio
->bi_size
;
1178 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1180 drive_stat_acct(req
, 0);
1181 elv_bio_merged(q
, req
, bio
);
1185 static bool bio_attempt_front_merge(struct request_queue
*q
,
1186 struct request
*req
, struct bio
*bio
)
1188 const int ff
= bio
->bi_rw
& REQ_FAILFAST_MASK
;
1190 if (!ll_front_merge_fn(q
, req
, bio
))
1193 trace_block_bio_frontmerge(q
, bio
);
1195 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1196 blk_rq_set_mixed_merge(req
);
1198 bio
->bi_next
= req
->bio
;
1202 * may not be valid. if the low level driver said
1203 * it didn't need a bounce buffer then it better
1204 * not touch req->buffer either...
1206 req
->buffer
= bio_data(bio
);
1207 req
->__sector
= bio
->bi_sector
;
1208 req
->__data_len
+= bio
->bi_size
;
1209 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1211 drive_stat_acct(req
, 0);
1212 elv_bio_merged(q
, req
, bio
);
1217 * attempt_plug_merge - try to merge with %current's plugged list
1218 * @q: request_queue new bio is being queued at
1219 * @bio: new bio being queued
1220 * @request_count: out parameter for number of traversed plugged requests
1222 * Determine whether @bio being queued on @q can be merged with a request
1223 * on %current's plugged list. Returns %true if merge was successful,
1226 * This function is called without @q->queue_lock; however, elevator is
1227 * accessed iff there already are requests on the plugged list which in
1228 * turn guarantees validity of the elevator.
1230 * Note that, on successful merge, elevator operation
1231 * elevator_bio_merged_fn() will be called without queue lock. Elevator
1232 * must be ready for this.
1234 static bool attempt_plug_merge(struct request_queue
*q
, struct bio
*bio
,
1235 unsigned int *request_count
)
1237 struct blk_plug
*plug
;
1241 plug
= current
->plug
;
1246 list_for_each_entry_reverse(rq
, &plug
->list
, queuelist
) {
1254 el_ret
= elv_try_merge(rq
, bio
);
1255 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1256 ret
= bio_attempt_back_merge(q
, rq
, bio
);
1259 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1260 ret
= bio_attempt_front_merge(q
, rq
, bio
);
1269 void init_request_from_bio(struct request
*req
, struct bio
*bio
)
1271 req
->cmd_type
= REQ_TYPE_FS
;
1273 req
->cmd_flags
|= bio
->bi_rw
& REQ_COMMON_MASK
;
1274 if (bio
->bi_rw
& REQ_RAHEAD
)
1275 req
->cmd_flags
|= REQ_FAILFAST_MASK
;
1278 req
->__sector
= bio
->bi_sector
;
1279 req
->ioprio
= bio_prio(bio
);
1280 blk_rq_bio_prep(req
->q
, req
, bio
);
1283 void blk_queue_bio(struct request_queue
*q
, struct bio
*bio
)
1285 const bool sync
= !!(bio
->bi_rw
& REQ_SYNC
);
1286 struct blk_plug
*plug
;
1287 int el_ret
, rw_flags
, where
= ELEVATOR_INSERT_SORT
;
1288 struct request
*req
;
1289 unsigned int request_count
= 0;
1292 * low level driver can indicate that it wants pages above a
1293 * certain limit bounced to low memory (ie for highmem, or even
1294 * ISA dma in theory)
1296 blk_queue_bounce(q
, &bio
);
1298 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) {
1299 spin_lock_irq(q
->queue_lock
);
1300 where
= ELEVATOR_INSERT_FLUSH
;
1305 * Check if we can merge with the plugged list before grabbing
1308 if (attempt_plug_merge(q
, bio
, &request_count
))
1311 spin_lock_irq(q
->queue_lock
);
1313 el_ret
= elv_merge(q
, &req
, bio
);
1314 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1315 if (bio_attempt_back_merge(q
, req
, bio
)) {
1316 if (!attempt_back_merge(q
, req
))
1317 elv_merged_request(q
, req
, el_ret
);
1320 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1321 if (bio_attempt_front_merge(q
, req
, bio
)) {
1322 if (!attempt_front_merge(q
, req
))
1323 elv_merged_request(q
, req
, el_ret
);
1330 * This sync check and mask will be re-done in init_request_from_bio(),
1331 * but we need to set it earlier to expose the sync flag to the
1332 * rq allocator and io schedulers.
1334 rw_flags
= bio_data_dir(bio
);
1336 rw_flags
|= REQ_SYNC
;
1339 * Grab a free request. This is might sleep but can not fail.
1340 * Returns with the queue unlocked.
1342 req
= get_request_wait(q
, rw_flags
, bio
);
1343 if (unlikely(!req
)) {
1344 bio_endio(bio
, -ENODEV
); /* @q is dead */
1349 * After dropping the lock and possibly sleeping here, our request
1350 * may now be mergeable after it had proven unmergeable (above).
1351 * We don't worry about that case for efficiency. It won't happen
1352 * often, and the elevators are able to handle it.
1354 init_request_from_bio(req
, bio
);
1356 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
))
1357 req
->cpu
= raw_smp_processor_id();
1359 plug
= current
->plug
;
1362 * If this is the first request added after a plug, fire
1363 * of a plug trace. If others have been added before, check
1364 * if we have multiple devices in this plug. If so, make a
1365 * note to sort the list before dispatch.
1367 if (list_empty(&plug
->list
))
1368 trace_block_plug(q
);
1370 if (!plug
->should_sort
) {
1371 struct request
*__rq
;
1373 __rq
= list_entry_rq(plug
->list
.prev
);
1375 plug
->should_sort
= 1;
1377 if (request_count
>= BLK_MAX_REQUEST_COUNT
) {
1378 blk_flush_plug_list(plug
, false);
1379 trace_block_plug(q
);
1382 list_add_tail(&req
->queuelist
, &plug
->list
);
1383 drive_stat_acct(req
, 1);
1385 spin_lock_irq(q
->queue_lock
);
1386 add_acct_request(q
, req
, where
);
1389 spin_unlock_irq(q
->queue_lock
);
1392 EXPORT_SYMBOL_GPL(blk_queue_bio
); /* for device mapper only */
1395 * If bio->bi_dev is a partition, remap the location
1397 static inline void blk_partition_remap(struct bio
*bio
)
1399 struct block_device
*bdev
= bio
->bi_bdev
;
1401 if (bio_sectors(bio
) && bdev
!= bdev
->bd_contains
) {
1402 struct hd_struct
*p
= bdev
->bd_part
;
1404 bio
->bi_sector
+= p
->start_sect
;
1405 bio
->bi_bdev
= bdev
->bd_contains
;
1407 trace_block_bio_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
1409 bio
->bi_sector
- p
->start_sect
);
1413 static void handle_bad_sector(struct bio
*bio
)
1415 char b
[BDEVNAME_SIZE
];
1417 printk(KERN_INFO
"attempt to access beyond end of device\n");
1418 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
1419 bdevname(bio
->bi_bdev
, b
),
1421 (unsigned long long)bio
->bi_sector
+ bio_sectors(bio
),
1422 (long long)(i_size_read(bio
->bi_bdev
->bd_inode
) >> 9));
1424 set_bit(BIO_EOF
, &bio
->bi_flags
);
1427 #ifdef CONFIG_FAIL_MAKE_REQUEST
1429 static DECLARE_FAULT_ATTR(fail_make_request
);
1431 static int __init
setup_fail_make_request(char *str
)
1433 return setup_fault_attr(&fail_make_request
, str
);
1435 __setup("fail_make_request=", setup_fail_make_request
);
1437 static bool should_fail_request(struct hd_struct
*part
, unsigned int bytes
)
1439 return part
->make_it_fail
&& should_fail(&fail_make_request
, bytes
);
1442 static int __init
fail_make_request_debugfs(void)
1444 struct dentry
*dir
= fault_create_debugfs_attr("fail_make_request",
1445 NULL
, &fail_make_request
);
1447 return IS_ERR(dir
) ? PTR_ERR(dir
) : 0;
1450 late_initcall(fail_make_request_debugfs
);
1452 #else /* CONFIG_FAIL_MAKE_REQUEST */
1454 static inline bool should_fail_request(struct hd_struct
*part
,
1460 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1463 * Check whether this bio extends beyond the end of the device.
1465 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
1472 /* Test device or partition size, when known. */
1473 maxsector
= i_size_read(bio
->bi_bdev
->bd_inode
) >> 9;
1475 sector_t sector
= bio
->bi_sector
;
1477 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
1479 * This may well happen - the kernel calls bread()
1480 * without checking the size of the device, e.g., when
1481 * mounting a device.
1483 handle_bad_sector(bio
);
1491 static noinline_for_stack
bool
1492 generic_make_request_checks(struct bio
*bio
)
1494 struct request_queue
*q
;
1495 int nr_sectors
= bio_sectors(bio
);
1497 char b
[BDEVNAME_SIZE
];
1498 struct hd_struct
*part
;
1502 if (bio_check_eod(bio
, nr_sectors
))
1505 q
= bdev_get_queue(bio
->bi_bdev
);
1508 "generic_make_request: Trying to access "
1509 "nonexistent block-device %s (%Lu)\n",
1510 bdevname(bio
->bi_bdev
, b
),
1511 (long long) bio
->bi_sector
);
1515 if (unlikely(!(bio
->bi_rw
& REQ_DISCARD
) &&
1516 nr_sectors
> queue_max_hw_sectors(q
))) {
1517 printk(KERN_ERR
"bio too big device %s (%u > %u)\n",
1518 bdevname(bio
->bi_bdev
, b
),
1520 queue_max_hw_sectors(q
));
1524 part
= bio
->bi_bdev
->bd_part
;
1525 if (should_fail_request(part
, bio
->bi_size
) ||
1526 should_fail_request(&part_to_disk(part
)->part0
,
1531 * If this device has partitions, remap block n
1532 * of partition p to block n+start(p) of the disk.
1534 blk_partition_remap(bio
);
1536 if (bio_integrity_enabled(bio
) && bio_integrity_prep(bio
))
1539 if (bio_check_eod(bio
, nr_sectors
))
1543 * Filter flush bio's early so that make_request based
1544 * drivers without flush support don't have to worry
1547 if ((bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) && !q
->flush_flags
) {
1548 bio
->bi_rw
&= ~(REQ_FLUSH
| REQ_FUA
);
1555 if ((bio
->bi_rw
& REQ_DISCARD
) &&
1556 (!blk_queue_discard(q
) ||
1557 ((bio
->bi_rw
& REQ_SECURE
) &&
1558 !blk_queue_secdiscard(q
)))) {
1563 if (blk_throtl_bio(q
, bio
))
1564 return false; /* throttled, will be resubmitted later */
1566 trace_block_bio_queue(q
, bio
);
1570 bio_endio(bio
, err
);
1575 * generic_make_request - hand a buffer to its device driver for I/O
1576 * @bio: The bio describing the location in memory and on the device.
1578 * generic_make_request() is used to make I/O requests of block
1579 * devices. It is passed a &struct bio, which describes the I/O that needs
1582 * generic_make_request() does not return any status. The
1583 * success/failure status of the request, along with notification of
1584 * completion, is delivered asynchronously through the bio->bi_end_io
1585 * function described (one day) else where.
1587 * The caller of generic_make_request must make sure that bi_io_vec
1588 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1589 * set to describe the device address, and the
1590 * bi_end_io and optionally bi_private are set to describe how
1591 * completion notification should be signaled.
1593 * generic_make_request and the drivers it calls may use bi_next if this
1594 * bio happens to be merged with someone else, and may resubmit the bio to
1595 * a lower device by calling into generic_make_request recursively, which
1596 * means the bio should NOT be touched after the call to ->make_request_fn.
1598 void generic_make_request(struct bio
*bio
)
1600 struct bio_list bio_list_on_stack
;
1602 if (!generic_make_request_checks(bio
))
1606 * We only want one ->make_request_fn to be active at a time, else
1607 * stack usage with stacked devices could be a problem. So use
1608 * current->bio_list to keep a list of requests submited by a
1609 * make_request_fn function. current->bio_list is also used as a
1610 * flag to say if generic_make_request is currently active in this
1611 * task or not. If it is NULL, then no make_request is active. If
1612 * it is non-NULL, then a make_request is active, and new requests
1613 * should be added at the tail
1615 if (current
->bio_list
) {
1616 bio_list_add(current
->bio_list
, bio
);
1620 /* following loop may be a bit non-obvious, and so deserves some
1622 * Before entering the loop, bio->bi_next is NULL (as all callers
1623 * ensure that) so we have a list with a single bio.
1624 * We pretend that we have just taken it off a longer list, so
1625 * we assign bio_list to a pointer to the bio_list_on_stack,
1626 * thus initialising the bio_list of new bios to be
1627 * added. ->make_request() may indeed add some more bios
1628 * through a recursive call to generic_make_request. If it
1629 * did, we find a non-NULL value in bio_list and re-enter the loop
1630 * from the top. In this case we really did just take the bio
1631 * of the top of the list (no pretending) and so remove it from
1632 * bio_list, and call into ->make_request() again.
1634 BUG_ON(bio
->bi_next
);
1635 bio_list_init(&bio_list_on_stack
);
1636 current
->bio_list
= &bio_list_on_stack
;
1638 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
1640 q
->make_request_fn(q
, bio
);
1642 bio
= bio_list_pop(current
->bio_list
);
1644 current
->bio_list
= NULL
; /* deactivate */
1646 EXPORT_SYMBOL(generic_make_request
);
1649 * submit_bio - submit a bio to the block device layer for I/O
1650 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1651 * @bio: The &struct bio which describes the I/O
1653 * submit_bio() is very similar in purpose to generic_make_request(), and
1654 * uses that function to do most of the work. Both are fairly rough
1655 * interfaces; @bio must be presetup and ready for I/O.
1658 void submit_bio(int rw
, struct bio
*bio
)
1660 int count
= bio_sectors(bio
);
1665 * If it's a regular read/write or a barrier with data attached,
1666 * go through the normal accounting stuff before submission.
1668 if (bio_has_data(bio
) && !(rw
& REQ_DISCARD
)) {
1670 count_vm_events(PGPGOUT
, count
);
1672 task_io_account_read(bio
->bi_size
);
1673 count_vm_events(PGPGIN
, count
);
1676 if (unlikely(block_dump
)) {
1677 char b
[BDEVNAME_SIZE
];
1678 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s (%u sectors)\n",
1679 current
->comm
, task_pid_nr(current
),
1680 (rw
& WRITE
) ? "WRITE" : "READ",
1681 (unsigned long long)bio
->bi_sector
,
1682 bdevname(bio
->bi_bdev
, b
),
1687 generic_make_request(bio
);
1689 EXPORT_SYMBOL(submit_bio
);
1692 * blk_rq_check_limits - Helper function to check a request for the queue limit
1694 * @rq: the request being checked
1697 * @rq may have been made based on weaker limitations of upper-level queues
1698 * in request stacking drivers, and it may violate the limitation of @q.
1699 * Since the block layer and the underlying device driver trust @rq
1700 * after it is inserted to @q, it should be checked against @q before
1701 * the insertion using this generic function.
1703 * This function should also be useful for request stacking drivers
1704 * in some cases below, so export this function.
1705 * Request stacking drivers like request-based dm may change the queue
1706 * limits while requests are in the queue (e.g. dm's table swapping).
1707 * Such request stacking drivers should check those requests agaist
1708 * the new queue limits again when they dispatch those requests,
1709 * although such checkings are also done against the old queue limits
1710 * when submitting requests.
1712 int blk_rq_check_limits(struct request_queue
*q
, struct request
*rq
)
1714 if (rq
->cmd_flags
& REQ_DISCARD
)
1717 if (blk_rq_sectors(rq
) > queue_max_sectors(q
) ||
1718 blk_rq_bytes(rq
) > queue_max_hw_sectors(q
) << 9) {
1719 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
1724 * queue's settings related to segment counting like q->bounce_pfn
1725 * may differ from that of other stacking queues.
1726 * Recalculate it to check the request correctly on this queue's
1729 blk_recalc_rq_segments(rq
);
1730 if (rq
->nr_phys_segments
> queue_max_segments(q
)) {
1731 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
1737 EXPORT_SYMBOL_GPL(blk_rq_check_limits
);
1740 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1741 * @q: the queue to submit the request
1742 * @rq: the request being queued
1744 int blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
1746 unsigned long flags
;
1747 int where
= ELEVATOR_INSERT_BACK
;
1749 if (blk_rq_check_limits(q
, rq
))
1753 should_fail_request(&rq
->rq_disk
->part0
, blk_rq_bytes(rq
)))
1756 spin_lock_irqsave(q
->queue_lock
, flags
);
1757 if (unlikely(blk_queue_dead(q
))) {
1758 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1763 * Submitting request must be dequeued before calling this function
1764 * because it will be linked to another request_queue
1766 BUG_ON(blk_queued_rq(rq
));
1768 if (rq
->cmd_flags
& (REQ_FLUSH
|REQ_FUA
))
1769 where
= ELEVATOR_INSERT_FLUSH
;
1771 add_acct_request(q
, rq
, where
);
1772 if (where
== ELEVATOR_INSERT_FLUSH
)
1774 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1778 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
1781 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1782 * @rq: request to examine
1785 * A request could be merge of IOs which require different failure
1786 * handling. This function determines the number of bytes which
1787 * can be failed from the beginning of the request without
1788 * crossing into area which need to be retried further.
1791 * The number of bytes to fail.
1794 * queue_lock must be held.
1796 unsigned int blk_rq_err_bytes(const struct request
*rq
)
1798 unsigned int ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
1799 unsigned int bytes
= 0;
1802 if (!(rq
->cmd_flags
& REQ_MIXED_MERGE
))
1803 return blk_rq_bytes(rq
);
1806 * Currently the only 'mixing' which can happen is between
1807 * different fastfail types. We can safely fail portions
1808 * which have all the failfast bits that the first one has -
1809 * the ones which are at least as eager to fail as the first
1812 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
1813 if ((bio
->bi_rw
& ff
) != ff
)
1815 bytes
+= bio
->bi_size
;
1818 /* this could lead to infinite loop */
1819 BUG_ON(blk_rq_bytes(rq
) && !bytes
);
1822 EXPORT_SYMBOL_GPL(blk_rq_err_bytes
);
1824 static void blk_account_io_completion(struct request
*req
, unsigned int bytes
)
1826 if (blk_do_io_stat(req
)) {
1827 const int rw
= rq_data_dir(req
);
1828 struct hd_struct
*part
;
1831 cpu
= part_stat_lock();
1833 part_stat_add(cpu
, part
, sectors
[rw
], bytes
>> 9);
1838 static void blk_account_io_done(struct request
*req
)
1841 * Account IO completion. flush_rq isn't accounted as a
1842 * normal IO on queueing nor completion. Accounting the
1843 * containing request is enough.
1845 if (blk_do_io_stat(req
) && !(req
->cmd_flags
& REQ_FLUSH_SEQ
)) {
1846 unsigned long duration
= jiffies
- req
->start_time
;
1847 const int rw
= rq_data_dir(req
);
1848 struct hd_struct
*part
;
1851 cpu
= part_stat_lock();
1854 part_stat_inc(cpu
, part
, ios
[rw
]);
1855 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
1856 part_round_stats(cpu
, part
);
1857 part_dec_in_flight(part
, rw
);
1859 hd_struct_put(part
);
1865 * blk_peek_request - peek at the top of a request queue
1866 * @q: request queue to peek at
1869 * Return the request at the top of @q. The returned request
1870 * should be started using blk_start_request() before LLD starts
1874 * Pointer to the request at the top of @q if available. Null
1878 * queue_lock must be held.
1880 struct request
*blk_peek_request(struct request_queue
*q
)
1885 while ((rq
= __elv_next_request(q
)) != NULL
) {
1886 if (!(rq
->cmd_flags
& REQ_STARTED
)) {
1888 * This is the first time the device driver
1889 * sees this request (possibly after
1890 * requeueing). Notify IO scheduler.
1892 if (rq
->cmd_flags
& REQ_SORTED
)
1893 elv_activate_rq(q
, rq
);
1896 * just mark as started even if we don't start
1897 * it, a request that has been delayed should
1898 * not be passed by new incoming requests
1900 rq
->cmd_flags
|= REQ_STARTED
;
1901 trace_block_rq_issue(q
, rq
);
1904 if (!q
->boundary_rq
|| q
->boundary_rq
== rq
) {
1905 q
->end_sector
= rq_end_sector(rq
);
1906 q
->boundary_rq
= NULL
;
1909 if (rq
->cmd_flags
& REQ_DONTPREP
)
1912 if (q
->dma_drain_size
&& blk_rq_bytes(rq
)) {
1914 * make sure space for the drain appears we
1915 * know we can do this because max_hw_segments
1916 * has been adjusted to be one fewer than the
1919 rq
->nr_phys_segments
++;
1925 ret
= q
->prep_rq_fn(q
, rq
);
1926 if (ret
== BLKPREP_OK
) {
1928 } else if (ret
== BLKPREP_DEFER
) {
1930 * the request may have been (partially) prepped.
1931 * we need to keep this request in the front to
1932 * avoid resource deadlock. REQ_STARTED will
1933 * prevent other fs requests from passing this one.
1935 if (q
->dma_drain_size
&& blk_rq_bytes(rq
) &&
1936 !(rq
->cmd_flags
& REQ_DONTPREP
)) {
1938 * remove the space for the drain we added
1939 * so that we don't add it again
1941 --rq
->nr_phys_segments
;
1946 } else if (ret
== BLKPREP_KILL
) {
1947 rq
->cmd_flags
|= REQ_QUIET
;
1949 * Mark this request as started so we don't trigger
1950 * any debug logic in the end I/O path.
1952 blk_start_request(rq
);
1953 __blk_end_request_all(rq
, -EIO
);
1955 printk(KERN_ERR
"%s: bad return=%d\n", __func__
, ret
);
1962 EXPORT_SYMBOL(blk_peek_request
);
1964 void blk_dequeue_request(struct request
*rq
)
1966 struct request_queue
*q
= rq
->q
;
1968 BUG_ON(list_empty(&rq
->queuelist
));
1969 BUG_ON(ELV_ON_HASH(rq
));
1971 list_del_init(&rq
->queuelist
);
1974 * the time frame between a request being removed from the lists
1975 * and to it is freed is accounted as io that is in progress at
1978 if (blk_account_rq(rq
)) {
1979 q
->in_flight
[rq_is_sync(rq
)]++;
1980 set_io_start_time_ns(rq
);
1985 * blk_start_request - start request processing on the driver
1986 * @req: request to dequeue
1989 * Dequeue @req and start timeout timer on it. This hands off the
1990 * request to the driver.
1992 * Block internal functions which don't want to start timer should
1993 * call blk_dequeue_request().
1996 * queue_lock must be held.
1998 void blk_start_request(struct request
*req
)
2000 blk_dequeue_request(req
);
2003 * We are now handing the request to the hardware, initialize
2004 * resid_len to full count and add the timeout handler.
2006 req
->resid_len
= blk_rq_bytes(req
);
2007 if (unlikely(blk_bidi_rq(req
)))
2008 req
->next_rq
->resid_len
= blk_rq_bytes(req
->next_rq
);
2012 EXPORT_SYMBOL(blk_start_request
);
2015 * blk_fetch_request - fetch a request from a request queue
2016 * @q: request queue to fetch a request from
2019 * Return the request at the top of @q. The request is started on
2020 * return and LLD can start processing it immediately.
2023 * Pointer to the request at the top of @q if available. Null
2027 * queue_lock must be held.
2029 struct request
*blk_fetch_request(struct request_queue
*q
)
2033 rq
= blk_peek_request(q
);
2035 blk_start_request(rq
);
2038 EXPORT_SYMBOL(blk_fetch_request
);
2041 * blk_update_request - Special helper function for request stacking drivers
2042 * @req: the request being processed
2043 * @error: %0 for success, < %0 for error
2044 * @nr_bytes: number of bytes to complete @req
2047 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2048 * the request structure even if @req doesn't have leftover.
2049 * If @req has leftover, sets it up for the next range of segments.
2051 * This special helper function is only for request stacking drivers
2052 * (e.g. request-based dm) so that they can handle partial completion.
2053 * Actual device drivers should use blk_end_request instead.
2055 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2056 * %false return from this function.
2059 * %false - this request doesn't have any more data
2060 * %true - this request has more data
2062 bool blk_update_request(struct request
*req
, int error
, unsigned int nr_bytes
)
2064 int total_bytes
, bio_nbytes
, next_idx
= 0;
2070 trace_block_rq_complete(req
->q
, req
);
2073 * For fs requests, rq is just carrier of independent bio's
2074 * and each partial completion should be handled separately.
2075 * Reset per-request error on each partial completion.
2077 * TODO: tj: This is too subtle. It would be better to let
2078 * low level drivers do what they see fit.
2080 if (req
->cmd_type
== REQ_TYPE_FS
)
2083 if (error
&& req
->cmd_type
== REQ_TYPE_FS
&&
2084 !(req
->cmd_flags
& REQ_QUIET
)) {
2089 error_type
= "recoverable transport";
2092 error_type
= "critical target";
2095 error_type
= "critical nexus";
2102 printk(KERN_ERR
"end_request: %s error, dev %s, sector %llu\n",
2103 error_type
, req
->rq_disk
? req
->rq_disk
->disk_name
: "?",
2104 (unsigned long long)blk_rq_pos(req
));
2107 blk_account_io_completion(req
, nr_bytes
);
2109 total_bytes
= bio_nbytes
= 0;
2110 while ((bio
= req
->bio
) != NULL
) {
2113 if (nr_bytes
>= bio
->bi_size
) {
2114 req
->bio
= bio
->bi_next
;
2115 nbytes
= bio
->bi_size
;
2116 req_bio_endio(req
, bio
, nbytes
, error
);
2120 int idx
= bio
->bi_idx
+ next_idx
;
2122 if (unlikely(idx
>= bio
->bi_vcnt
)) {
2123 blk_dump_rq_flags(req
, "__end_that");
2124 printk(KERN_ERR
"%s: bio idx %d >= vcnt %d\n",
2125 __func__
, idx
, bio
->bi_vcnt
);
2129 nbytes
= bio_iovec_idx(bio
, idx
)->bv_len
;
2130 BIO_BUG_ON(nbytes
> bio
->bi_size
);
2133 * not a complete bvec done
2135 if (unlikely(nbytes
> nr_bytes
)) {
2136 bio_nbytes
+= nr_bytes
;
2137 total_bytes
+= nr_bytes
;
2142 * advance to the next vector
2145 bio_nbytes
+= nbytes
;
2148 total_bytes
+= nbytes
;
2154 * end more in this run, or just return 'not-done'
2156 if (unlikely(nr_bytes
<= 0))
2166 * Reset counters so that the request stacking driver
2167 * can find how many bytes remain in the request
2170 req
->__data_len
= 0;
2175 * if the request wasn't completed, update state
2178 req_bio_endio(req
, bio
, bio_nbytes
, error
);
2179 bio
->bi_idx
+= next_idx
;
2180 bio_iovec(bio
)->bv_offset
+= nr_bytes
;
2181 bio_iovec(bio
)->bv_len
-= nr_bytes
;
2184 req
->__data_len
-= total_bytes
;
2185 req
->buffer
= bio_data(req
->bio
);
2187 /* update sector only for requests with clear definition of sector */
2188 if (req
->cmd_type
== REQ_TYPE_FS
|| (req
->cmd_flags
& REQ_DISCARD
))
2189 req
->__sector
+= total_bytes
>> 9;
2191 /* mixed attributes always follow the first bio */
2192 if (req
->cmd_flags
& REQ_MIXED_MERGE
) {
2193 req
->cmd_flags
&= ~REQ_FAILFAST_MASK
;
2194 req
->cmd_flags
|= req
->bio
->bi_rw
& REQ_FAILFAST_MASK
;
2198 * If total number of sectors is less than the first segment
2199 * size, something has gone terribly wrong.
2201 if (blk_rq_bytes(req
) < blk_rq_cur_bytes(req
)) {
2202 blk_dump_rq_flags(req
, "request botched");
2203 req
->__data_len
= blk_rq_cur_bytes(req
);
2206 /* recalculate the number of segments */
2207 blk_recalc_rq_segments(req
);
2211 EXPORT_SYMBOL_GPL(blk_update_request
);
2213 static bool blk_update_bidi_request(struct request
*rq
, int error
,
2214 unsigned int nr_bytes
,
2215 unsigned int bidi_bytes
)
2217 if (blk_update_request(rq
, error
, nr_bytes
))
2220 /* Bidi request must be completed as a whole */
2221 if (unlikely(blk_bidi_rq(rq
)) &&
2222 blk_update_request(rq
->next_rq
, error
, bidi_bytes
))
2225 if (blk_queue_add_random(rq
->q
))
2226 add_disk_randomness(rq
->rq_disk
);
2232 * blk_unprep_request - unprepare a request
2235 * This function makes a request ready for complete resubmission (or
2236 * completion). It happens only after all error handling is complete,
2237 * so represents the appropriate moment to deallocate any resources
2238 * that were allocated to the request in the prep_rq_fn. The queue
2239 * lock is held when calling this.
2241 void blk_unprep_request(struct request
*req
)
2243 struct request_queue
*q
= req
->q
;
2245 req
->cmd_flags
&= ~REQ_DONTPREP
;
2246 if (q
->unprep_rq_fn
)
2247 q
->unprep_rq_fn(q
, req
);
2249 EXPORT_SYMBOL_GPL(blk_unprep_request
);
2252 * queue lock must be held
2254 static void blk_finish_request(struct request
*req
, int error
)
2256 if (blk_rq_tagged(req
))
2257 blk_queue_end_tag(req
->q
, req
);
2259 BUG_ON(blk_queued_rq(req
));
2261 if (unlikely(laptop_mode
) && req
->cmd_type
== REQ_TYPE_FS
)
2262 laptop_io_completion(&req
->q
->backing_dev_info
);
2264 blk_delete_timer(req
);
2266 if (req
->cmd_flags
& REQ_DONTPREP
)
2267 blk_unprep_request(req
);
2270 blk_account_io_done(req
);
2273 req
->end_io(req
, error
);
2275 if (blk_bidi_rq(req
))
2276 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
2278 __blk_put_request(req
->q
, req
);
2283 * blk_end_bidi_request - Complete a bidi request
2284 * @rq: the request to complete
2285 * @error: %0 for success, < %0 for error
2286 * @nr_bytes: number of bytes to complete @rq
2287 * @bidi_bytes: number of bytes to complete @rq->next_rq
2290 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2291 * Drivers that supports bidi can safely call this member for any
2292 * type of request, bidi or uni. In the later case @bidi_bytes is
2296 * %false - we are done with this request
2297 * %true - still buffers pending for this request
2299 static bool blk_end_bidi_request(struct request
*rq
, int error
,
2300 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2302 struct request_queue
*q
= rq
->q
;
2303 unsigned long flags
;
2305 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2308 spin_lock_irqsave(q
->queue_lock
, flags
);
2309 blk_finish_request(rq
, error
);
2310 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2316 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2317 * @rq: the request to complete
2318 * @error: %0 for success, < %0 for error
2319 * @nr_bytes: number of bytes to complete @rq
2320 * @bidi_bytes: number of bytes to complete @rq->next_rq
2323 * Identical to blk_end_bidi_request() except that queue lock is
2324 * assumed to be locked on entry and remains so on return.
2327 * %false - we are done with this request
2328 * %true - still buffers pending for this request
2330 bool __blk_end_bidi_request(struct request
*rq
, int error
,
2331 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2333 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2336 blk_finish_request(rq
, error
);
2342 * blk_end_request - Helper function for drivers to complete the request.
2343 * @rq: the request being processed
2344 * @error: %0 for success, < %0 for error
2345 * @nr_bytes: number of bytes to complete
2348 * Ends I/O on a number of bytes attached to @rq.
2349 * If @rq has leftover, sets it up for the next range of segments.
2352 * %false - we are done with this request
2353 * %true - still buffers pending for this request
2355 bool blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2357 return blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2359 EXPORT_SYMBOL(blk_end_request
);
2362 * blk_end_request_all - Helper function for drives to finish the request.
2363 * @rq: the request to finish
2364 * @error: %0 for success, < %0 for error
2367 * Completely finish @rq.
2369 void blk_end_request_all(struct request
*rq
, int error
)
2372 unsigned int bidi_bytes
= 0;
2374 if (unlikely(blk_bidi_rq(rq
)))
2375 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2377 pending
= blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2380 EXPORT_SYMBOL(blk_end_request_all
);
2383 * blk_end_request_cur - Helper function to finish the current request chunk.
2384 * @rq: the request to finish the current chunk for
2385 * @error: %0 for success, < %0 for error
2388 * Complete the current consecutively mapped chunk from @rq.
2391 * %false - we are done with this request
2392 * %true - still buffers pending for this request
2394 bool blk_end_request_cur(struct request
*rq
, int error
)
2396 return blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2398 EXPORT_SYMBOL(blk_end_request_cur
);
2401 * blk_end_request_err - Finish a request till the next failure boundary.
2402 * @rq: the request to finish till the next failure boundary for
2403 * @error: must be negative errno
2406 * Complete @rq till the next failure boundary.
2409 * %false - we are done with this request
2410 * %true - still buffers pending for this request
2412 bool blk_end_request_err(struct request
*rq
, int error
)
2414 WARN_ON(error
>= 0);
2415 return blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2417 EXPORT_SYMBOL_GPL(blk_end_request_err
);
2420 * __blk_end_request - Helper function for drivers to complete the request.
2421 * @rq: the request being processed
2422 * @error: %0 for success, < %0 for error
2423 * @nr_bytes: number of bytes to complete
2426 * Must be called with queue lock held unlike blk_end_request().
2429 * %false - we are done with this request
2430 * %true - still buffers pending for this request
2432 bool __blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2434 return __blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2436 EXPORT_SYMBOL(__blk_end_request
);
2439 * __blk_end_request_all - Helper function for drives to finish the request.
2440 * @rq: the request to finish
2441 * @error: %0 for success, < %0 for error
2444 * Completely finish @rq. Must be called with queue lock held.
2446 void __blk_end_request_all(struct request
*rq
, int error
)
2449 unsigned int bidi_bytes
= 0;
2451 if (unlikely(blk_bidi_rq(rq
)))
2452 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2454 pending
= __blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2457 EXPORT_SYMBOL(__blk_end_request_all
);
2460 * __blk_end_request_cur - Helper function to finish the current request chunk.
2461 * @rq: the request to finish the current chunk for
2462 * @error: %0 for success, < %0 for error
2465 * Complete the current consecutively mapped chunk from @rq. Must
2466 * be called with queue lock held.
2469 * %false - we are done with this request
2470 * %true - still buffers pending for this request
2472 bool __blk_end_request_cur(struct request
*rq
, int error
)
2474 return __blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2476 EXPORT_SYMBOL(__blk_end_request_cur
);
2479 * __blk_end_request_err - Finish a request till the next failure boundary.
2480 * @rq: the request to finish till the next failure boundary for
2481 * @error: must be negative errno
2484 * Complete @rq till the next failure boundary. Must be called
2485 * with queue lock held.
2488 * %false - we are done with this request
2489 * %true - still buffers pending for this request
2491 bool __blk_end_request_err(struct request
*rq
, int error
)
2493 WARN_ON(error
>= 0);
2494 return __blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2496 EXPORT_SYMBOL_GPL(__blk_end_request_err
);
2498 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
2501 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2502 rq
->cmd_flags
|= bio
->bi_rw
& REQ_WRITE
;
2504 if (bio_has_data(bio
)) {
2505 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2506 rq
->buffer
= bio_data(bio
);
2508 rq
->__data_len
= bio
->bi_size
;
2509 rq
->bio
= rq
->biotail
= bio
;
2512 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2515 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2517 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2518 * @rq: the request to be flushed
2521 * Flush all pages in @rq.
2523 void rq_flush_dcache_pages(struct request
*rq
)
2525 struct req_iterator iter
;
2526 struct bio_vec
*bvec
;
2528 rq_for_each_segment(bvec
, rq
, iter
)
2529 flush_dcache_page(bvec
->bv_page
);
2531 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages
);
2535 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2536 * @q : the queue of the device being checked
2539 * Check if underlying low-level drivers of a device are busy.
2540 * If the drivers want to export their busy state, they must set own
2541 * exporting function using blk_queue_lld_busy() first.
2543 * Basically, this function is used only by request stacking drivers
2544 * to stop dispatching requests to underlying devices when underlying
2545 * devices are busy. This behavior helps more I/O merging on the queue
2546 * of the request stacking driver and prevents I/O throughput regression
2547 * on burst I/O load.
2550 * 0 - Not busy (The request stacking driver should dispatch request)
2551 * 1 - Busy (The request stacking driver should stop dispatching request)
2553 int blk_lld_busy(struct request_queue
*q
)
2556 return q
->lld_busy_fn(q
);
2560 EXPORT_SYMBOL_GPL(blk_lld_busy
);
2563 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2564 * @rq: the clone request to be cleaned up
2567 * Free all bios in @rq for a cloned request.
2569 void blk_rq_unprep_clone(struct request
*rq
)
2573 while ((bio
= rq
->bio
) != NULL
) {
2574 rq
->bio
= bio
->bi_next
;
2579 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone
);
2582 * Copy attributes of the original request to the clone request.
2583 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2585 static void __blk_rq_prep_clone(struct request
*dst
, struct request
*src
)
2587 dst
->cpu
= src
->cpu
;
2588 dst
->cmd_flags
= (src
->cmd_flags
& REQ_CLONE_MASK
) | REQ_NOMERGE
;
2589 dst
->cmd_type
= src
->cmd_type
;
2590 dst
->__sector
= blk_rq_pos(src
);
2591 dst
->__data_len
= blk_rq_bytes(src
);
2592 dst
->nr_phys_segments
= src
->nr_phys_segments
;
2593 dst
->ioprio
= src
->ioprio
;
2594 dst
->extra_len
= src
->extra_len
;
2598 * blk_rq_prep_clone - Helper function to setup clone request
2599 * @rq: the request to be setup
2600 * @rq_src: original request to be cloned
2601 * @bs: bio_set that bios for clone are allocated from
2602 * @gfp_mask: memory allocation mask for bio
2603 * @bio_ctr: setup function to be called for each clone bio.
2604 * Returns %0 for success, non %0 for failure.
2605 * @data: private data to be passed to @bio_ctr
2608 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2609 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2610 * are not copied, and copying such parts is the caller's responsibility.
2611 * Also, pages which the original bios are pointing to are not copied
2612 * and the cloned bios just point same pages.
2613 * So cloned bios must be completed before original bios, which means
2614 * the caller must complete @rq before @rq_src.
2616 int blk_rq_prep_clone(struct request
*rq
, struct request
*rq_src
,
2617 struct bio_set
*bs
, gfp_t gfp_mask
,
2618 int (*bio_ctr
)(struct bio
*, struct bio
*, void *),
2621 struct bio
*bio
, *bio_src
;
2626 blk_rq_init(NULL
, rq
);
2628 __rq_for_each_bio(bio_src
, rq_src
) {
2629 bio
= bio_alloc_bioset(gfp_mask
, bio_src
->bi_max_vecs
, bs
);
2633 __bio_clone(bio
, bio_src
);
2635 if (bio_integrity(bio_src
) &&
2636 bio_integrity_clone(bio
, bio_src
, gfp_mask
, bs
))
2639 if (bio_ctr
&& bio_ctr(bio
, bio_src
, data
))
2643 rq
->biotail
->bi_next
= bio
;
2646 rq
->bio
= rq
->biotail
= bio
;
2649 __blk_rq_prep_clone(rq
, rq_src
);
2656 blk_rq_unprep_clone(rq
);
2660 EXPORT_SYMBOL_GPL(blk_rq_prep_clone
);
2662 int kblockd_schedule_work(struct request_queue
*q
, struct work_struct
*work
)
2664 return queue_work(kblockd_workqueue
, work
);
2666 EXPORT_SYMBOL(kblockd_schedule_work
);
2668 int kblockd_schedule_delayed_work(struct request_queue
*q
,
2669 struct delayed_work
*dwork
, unsigned long delay
)
2671 return queue_delayed_work(kblockd_workqueue
, dwork
, delay
);
2673 EXPORT_SYMBOL(kblockd_schedule_delayed_work
);
2675 #define PLUG_MAGIC 0x91827364
2678 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2679 * @plug: The &struct blk_plug that needs to be initialized
2682 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2683 * pending I/O should the task end up blocking between blk_start_plug() and
2684 * blk_finish_plug(). This is important from a performance perspective, but
2685 * also ensures that we don't deadlock. For instance, if the task is blocking
2686 * for a memory allocation, memory reclaim could end up wanting to free a
2687 * page belonging to that request that is currently residing in our private
2688 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2689 * this kind of deadlock.
2691 void blk_start_plug(struct blk_plug
*plug
)
2693 struct task_struct
*tsk
= current
;
2695 plug
->magic
= PLUG_MAGIC
;
2696 INIT_LIST_HEAD(&plug
->list
);
2697 INIT_LIST_HEAD(&plug
->cb_list
);
2698 plug
->should_sort
= 0;
2701 * If this is a nested plug, don't actually assign it. It will be
2702 * flushed on its own.
2706 * Store ordering should not be needed here, since a potential
2707 * preempt will imply a full memory barrier
2712 EXPORT_SYMBOL(blk_start_plug
);
2714 static int plug_rq_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
2716 struct request
*rqa
= container_of(a
, struct request
, queuelist
);
2717 struct request
*rqb
= container_of(b
, struct request
, queuelist
);
2719 return !(rqa
->q
<= rqb
->q
);
2723 * If 'from_schedule' is true, then postpone the dispatch of requests
2724 * until a safe kblockd context. We due this to avoid accidental big
2725 * additional stack usage in driver dispatch, in places where the originally
2726 * plugger did not intend it.
2728 static void queue_unplugged(struct request_queue
*q
, unsigned int depth
,
2730 __releases(q
->queue_lock
)
2732 trace_block_unplug(q
, depth
, !from_schedule
);
2735 * Don't mess with dead queue.
2737 if (unlikely(blk_queue_dead(q
))) {
2738 spin_unlock(q
->queue_lock
);
2743 * If we are punting this to kblockd, then we can safely drop
2744 * the queue_lock before waking kblockd (which needs to take
2747 if (from_schedule
) {
2748 spin_unlock(q
->queue_lock
);
2749 blk_run_queue_async(q
);
2752 spin_unlock(q
->queue_lock
);
2757 static void flush_plug_callbacks(struct blk_plug
*plug
)
2759 LIST_HEAD(callbacks
);
2761 if (list_empty(&plug
->cb_list
))
2764 list_splice_init(&plug
->cb_list
, &callbacks
);
2766 while (!list_empty(&callbacks
)) {
2767 struct blk_plug_cb
*cb
= list_first_entry(&callbacks
,
2770 list_del(&cb
->list
);
2775 void blk_flush_plug_list(struct blk_plug
*plug
, bool from_schedule
)
2777 struct request_queue
*q
;
2778 unsigned long flags
;
2783 BUG_ON(plug
->magic
!= PLUG_MAGIC
);
2785 flush_plug_callbacks(plug
);
2786 if (list_empty(&plug
->list
))
2789 list_splice_init(&plug
->list
, &list
);
2791 if (plug
->should_sort
) {
2792 list_sort(NULL
, &list
, plug_rq_cmp
);
2793 plug
->should_sort
= 0;
2800 * Save and disable interrupts here, to avoid doing it for every
2801 * queue lock we have to take.
2803 local_irq_save(flags
);
2804 while (!list_empty(&list
)) {
2805 rq
= list_entry_rq(list
.next
);
2806 list_del_init(&rq
->queuelist
);
2810 * This drops the queue lock
2813 queue_unplugged(q
, depth
, from_schedule
);
2816 spin_lock(q
->queue_lock
);
2820 * Short-circuit if @q is dead
2822 if (unlikely(blk_queue_dead(q
))) {
2823 __blk_end_request_all(rq
, -ENODEV
);
2828 * rq is already accounted, so use raw insert
2830 if (rq
->cmd_flags
& (REQ_FLUSH
| REQ_FUA
))
2831 __elv_add_request(q
, rq
, ELEVATOR_INSERT_FLUSH
);
2833 __elv_add_request(q
, rq
, ELEVATOR_INSERT_SORT_MERGE
);
2839 * This drops the queue lock
2842 queue_unplugged(q
, depth
, from_schedule
);
2844 local_irq_restore(flags
);
2847 void blk_finish_plug(struct blk_plug
*plug
)
2849 blk_flush_plug_list(plug
, false);
2851 if (plug
== current
->plug
)
2852 current
->plug
= NULL
;
2854 EXPORT_SYMBOL(blk_finish_plug
);
2856 int __init
blk_dev_init(void)
2858 BUILD_BUG_ON(__REQ_NR_BITS
> 8 *
2859 sizeof(((struct request
*)0)->cmd_flags
));
2861 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
2862 kblockd_workqueue
= alloc_workqueue("kblockd",
2863 WQ_MEM_RECLAIM
| WQ_HIGHPRI
, 0);
2864 if (!kblockd_workqueue
)
2865 panic("Failed to create kblockd\n");
2867 request_cachep
= kmem_cache_create("blkdev_requests",
2868 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
2870 blk_requestq_cachep
= kmem_cache_create("blkdev_queue",
2871 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);