2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
6 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
8 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
12 * This handles all read/write requests to block devices
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/backing-dev.h>
17 #include <linux/bio.h>
18 #include <linux/blkdev.h>
19 #include <linux/highmem.h>
21 #include <linux/kernel_stat.h>
22 #include <linux/string.h>
23 #include <linux/init.h>
24 #include <linux/completion.h>
25 #include <linux/slab.h>
26 #include <linux/swap.h>
27 #include <linux/writeback.h>
28 #include <linux/task_io_accounting_ops.h>
29 #include <linux/fault-inject.h>
30 #include <linux/list_sort.h>
32 #define CREATE_TRACE_POINTS
33 #include <trace/events/block.h>
37 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap
);
38 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap
);
39 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete
);
41 static int __make_request(struct request_queue
*q
, struct bio
*bio
);
44 * For the allocated request tables
46 static struct kmem_cache
*request_cachep
;
49 * For queue allocation
51 struct kmem_cache
*blk_requestq_cachep
;
54 * Controlling structure to kblockd
56 static struct workqueue_struct
*kblockd_workqueue
;
58 static void drive_stat_acct(struct request
*rq
, int new_io
)
60 struct hd_struct
*part
;
61 int rw
= rq_data_dir(rq
);
64 if (!blk_do_io_stat(rq
))
67 cpu
= part_stat_lock();
71 part_stat_inc(cpu
, part
, merges
[rw
]);
73 part
= disk_map_sector_rcu(rq
->rq_disk
, blk_rq_pos(rq
));
74 if (!hd_struct_try_get(part
)) {
76 * The partition is already being removed,
77 * the request will be accounted on the disk only
79 * We take a reference on disk->part0 although that
80 * partition will never be deleted, so we can treat
81 * it as any other partition.
83 part
= &rq
->rq_disk
->part0
;
86 part_round_stats(cpu
, part
);
87 part_inc_in_flight(part
, rw
);
94 void blk_queue_congestion_threshold(struct request_queue
*q
)
98 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) + 1;
99 if (nr
> q
->nr_requests
)
101 q
->nr_congestion_on
= nr
;
103 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) - (q
->nr_requests
/ 16) - 1;
106 q
->nr_congestion_off
= nr
;
110 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
113 * Locates the passed device's request queue and returns the address of its
116 * Will return NULL if the request queue cannot be located.
118 struct backing_dev_info
*blk_get_backing_dev_info(struct block_device
*bdev
)
120 struct backing_dev_info
*ret
= NULL
;
121 struct request_queue
*q
= bdev_get_queue(bdev
);
124 ret
= &q
->backing_dev_info
;
127 EXPORT_SYMBOL(blk_get_backing_dev_info
);
129 void blk_rq_init(struct request_queue
*q
, struct request
*rq
)
131 memset(rq
, 0, sizeof(*rq
));
133 INIT_LIST_HEAD(&rq
->queuelist
);
134 INIT_LIST_HEAD(&rq
->timeout_list
);
137 rq
->__sector
= (sector_t
) -1;
138 INIT_HLIST_NODE(&rq
->hash
);
139 RB_CLEAR_NODE(&rq
->rb_node
);
141 rq
->cmd_len
= BLK_MAX_CDB
;
144 rq
->start_time
= jiffies
;
145 set_start_time_ns(rq
);
148 EXPORT_SYMBOL(blk_rq_init
);
150 static void req_bio_endio(struct request
*rq
, struct bio
*bio
,
151 unsigned int nbytes
, int error
)
154 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
155 else if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
158 if (unlikely(nbytes
> bio
->bi_size
)) {
159 printk(KERN_ERR
"%s: want %u bytes done, %u left\n",
160 __func__
, nbytes
, bio
->bi_size
);
161 nbytes
= bio
->bi_size
;
164 if (unlikely(rq
->cmd_flags
& REQ_QUIET
))
165 set_bit(BIO_QUIET
, &bio
->bi_flags
);
167 bio
->bi_size
-= nbytes
;
168 bio
->bi_sector
+= (nbytes
>> 9);
170 if (bio_integrity(bio
))
171 bio_integrity_advance(bio
, nbytes
);
173 /* don't actually finish bio if it's part of flush sequence */
174 if (bio
->bi_size
== 0 && !(rq
->cmd_flags
& REQ_FLUSH_SEQ
))
175 bio_endio(bio
, error
);
178 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
182 printk(KERN_INFO
"%s: dev %s: type=%x, flags=%x\n", msg
,
183 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->cmd_type
,
186 printk(KERN_INFO
" sector %llu, nr/cnr %u/%u\n",
187 (unsigned long long)blk_rq_pos(rq
),
188 blk_rq_sectors(rq
), blk_rq_cur_sectors(rq
));
189 printk(KERN_INFO
" bio %p, biotail %p, buffer %p, len %u\n",
190 rq
->bio
, rq
->biotail
, rq
->buffer
, blk_rq_bytes(rq
));
192 if (rq
->cmd_type
== REQ_TYPE_BLOCK_PC
) {
193 printk(KERN_INFO
" cdb: ");
194 for (bit
= 0; bit
< BLK_MAX_CDB
; bit
++)
195 printk("%02x ", rq
->cmd
[bit
]);
199 EXPORT_SYMBOL(blk_dump_rq_flags
);
201 static void blk_delay_work(struct work_struct
*work
)
203 struct request_queue
*q
;
205 q
= container_of(work
, struct request_queue
, delay_work
.work
);
206 spin_lock_irq(q
->queue_lock
);
207 __blk_run_queue(q
, false);
208 spin_unlock_irq(q
->queue_lock
);
212 * blk_delay_queue - restart queueing after defined interval
213 * @q: The &struct request_queue in question
214 * @msecs: Delay in msecs
217 * Sometimes queueing needs to be postponed for a little while, to allow
218 * resources to come back. This function will make sure that queueing is
219 * restarted around the specified time.
221 void blk_delay_queue(struct request_queue
*q
, unsigned long msecs
)
223 queue_delayed_work(kblockd_workqueue
, &q
->delay_work
,
224 msecs_to_jiffies(msecs
));
226 EXPORT_SYMBOL(blk_delay_queue
);
229 * blk_start_queue - restart a previously stopped queue
230 * @q: The &struct request_queue in question
233 * blk_start_queue() will clear the stop flag on the queue, and call
234 * the request_fn for the queue if it was in a stopped state when
235 * entered. Also see blk_stop_queue(). Queue lock must be held.
237 void blk_start_queue(struct request_queue
*q
)
239 WARN_ON(!irqs_disabled());
241 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
242 __blk_run_queue(q
, false);
244 EXPORT_SYMBOL(blk_start_queue
);
247 * blk_stop_queue - stop a queue
248 * @q: The &struct request_queue in question
251 * The Linux block layer assumes that a block driver will consume all
252 * entries on the request queue when the request_fn strategy is called.
253 * Often this will not happen, because of hardware limitations (queue
254 * depth settings). If a device driver gets a 'queue full' response,
255 * or if it simply chooses not to queue more I/O at one point, it can
256 * call this function to prevent the request_fn from being called until
257 * the driver has signalled it's ready to go again. This happens by calling
258 * blk_start_queue() to restart queue operations. Queue lock must be held.
260 void blk_stop_queue(struct request_queue
*q
)
262 __cancel_delayed_work(&q
->delay_work
);
263 queue_flag_set(QUEUE_FLAG_STOPPED
, q
);
265 EXPORT_SYMBOL(blk_stop_queue
);
268 * blk_sync_queue - cancel any pending callbacks on a queue
272 * The block layer may perform asynchronous callback activity
273 * on a queue, such as calling the unplug function after a timeout.
274 * A block device may call blk_sync_queue to ensure that any
275 * such activity is cancelled, thus allowing it to release resources
276 * that the callbacks might use. The caller must already have made sure
277 * that its ->make_request_fn will not re-add plugging prior to calling
280 * This function does not cancel any asynchronous activity arising
281 * out of elevator or throttling code. That would require elevaotor_exit()
282 * and blk_throtl_exit() to be called with queue lock initialized.
285 void blk_sync_queue(struct request_queue
*q
)
287 del_timer_sync(&q
->timeout
);
288 cancel_delayed_work_sync(&q
->delay_work
);
290 EXPORT_SYMBOL(blk_sync_queue
);
293 * __blk_run_queue - run a single device queue
294 * @q: The queue to run
295 * @force_kblockd: Don't run @q->request_fn directly. Use kblockd.
298 * See @blk_run_queue. This variant must be called with the queue lock
299 * held and interrupts disabled. If force_kblockd is true, then it is
300 * safe to call this without holding the queue lock.
303 void __blk_run_queue(struct request_queue
*q
, bool force_kblockd
)
305 if (unlikely(blk_queue_stopped(q
)))
309 * Only recurse once to avoid overrunning the stack, let the unplug
310 * handling reinvoke the handler shortly if we already got there.
312 if (!force_kblockd
&& !queue_flag_test_and_set(QUEUE_FLAG_REENTER
, q
)) {
314 queue_flag_clear(QUEUE_FLAG_REENTER
, q
);
316 queue_delayed_work(kblockd_workqueue
, &q
->delay_work
, 0);
318 EXPORT_SYMBOL(__blk_run_queue
);
321 * blk_run_queue - run a single device queue
322 * @q: The queue to run
325 * Invoke request handling on this queue, if it has pending work to do.
326 * May be used to restart queueing when a request has completed.
328 void blk_run_queue(struct request_queue
*q
)
332 spin_lock_irqsave(q
->queue_lock
, flags
);
333 __blk_run_queue(q
, false);
334 spin_unlock_irqrestore(q
->queue_lock
, flags
);
336 EXPORT_SYMBOL(blk_run_queue
);
338 void blk_put_queue(struct request_queue
*q
)
340 kobject_put(&q
->kobj
);
344 * Note: If a driver supplied the queue lock, it should not zap that lock
345 * unexpectedly as some queue cleanup components like elevator_exit() and
346 * blk_throtl_exit() need queue lock.
348 void blk_cleanup_queue(struct request_queue
*q
)
351 * We know we have process context here, so we can be a little
352 * cautious and ensure that pending block actions on this device
353 * are done before moving on. Going into this function, we should
354 * not have processes doing IO to this device.
358 del_timer_sync(&q
->backing_dev_info
.laptop_mode_wb_timer
);
359 mutex_lock(&q
->sysfs_lock
);
360 queue_flag_set_unlocked(QUEUE_FLAG_DEAD
, q
);
361 mutex_unlock(&q
->sysfs_lock
);
364 elevator_exit(q
->elevator
);
370 EXPORT_SYMBOL(blk_cleanup_queue
);
372 static int blk_init_free_list(struct request_queue
*q
)
374 struct request_list
*rl
= &q
->rq
;
376 if (unlikely(rl
->rq_pool
))
379 rl
->count
[BLK_RW_SYNC
] = rl
->count
[BLK_RW_ASYNC
] = 0;
380 rl
->starved
[BLK_RW_SYNC
] = rl
->starved
[BLK_RW_ASYNC
] = 0;
382 init_waitqueue_head(&rl
->wait
[BLK_RW_SYNC
]);
383 init_waitqueue_head(&rl
->wait
[BLK_RW_ASYNC
]);
385 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, mempool_alloc_slab
,
386 mempool_free_slab
, request_cachep
, q
->node
);
394 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
396 return blk_alloc_queue_node(gfp_mask
, -1);
398 EXPORT_SYMBOL(blk_alloc_queue
);
400 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
402 struct request_queue
*q
;
405 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
406 gfp_mask
| __GFP_ZERO
, node_id
);
410 q
->backing_dev_info
.ra_pages
=
411 (VM_MAX_READAHEAD
* 1024) / PAGE_CACHE_SIZE
;
412 q
->backing_dev_info
.state
= 0;
413 q
->backing_dev_info
.capabilities
= BDI_CAP_MAP_COPY
;
414 q
->backing_dev_info
.name
= "block";
416 err
= bdi_init(&q
->backing_dev_info
);
418 kmem_cache_free(blk_requestq_cachep
, q
);
422 if (blk_throtl_init(q
)) {
423 kmem_cache_free(blk_requestq_cachep
, q
);
427 setup_timer(&q
->backing_dev_info
.laptop_mode_wb_timer
,
428 laptop_mode_timer_fn
, (unsigned long) q
);
429 setup_timer(&q
->timeout
, blk_rq_timed_out_timer
, (unsigned long) q
);
430 INIT_LIST_HEAD(&q
->timeout_list
);
431 INIT_LIST_HEAD(&q
->flush_queue
[0]);
432 INIT_LIST_HEAD(&q
->flush_queue
[1]);
433 INIT_LIST_HEAD(&q
->flush_data_in_flight
);
434 INIT_DELAYED_WORK(&q
->delay_work
, blk_delay_work
);
436 kobject_init(&q
->kobj
, &blk_queue_ktype
);
438 mutex_init(&q
->sysfs_lock
);
439 spin_lock_init(&q
->__queue_lock
);
442 * By default initialize queue_lock to internal lock and driver can
443 * override it later if need be.
445 q
->queue_lock
= &q
->__queue_lock
;
449 EXPORT_SYMBOL(blk_alloc_queue_node
);
452 * blk_init_queue - prepare a request queue for use with a block device
453 * @rfn: The function to be called to process requests that have been
454 * placed on the queue.
455 * @lock: Request queue spin lock
458 * If a block device wishes to use the standard request handling procedures,
459 * which sorts requests and coalesces adjacent requests, then it must
460 * call blk_init_queue(). The function @rfn will be called when there
461 * are requests on the queue that need to be processed. If the device
462 * supports plugging, then @rfn may not be called immediately when requests
463 * are available on the queue, but may be called at some time later instead.
464 * Plugged queues are generally unplugged when a buffer belonging to one
465 * of the requests on the queue is needed, or due to memory pressure.
467 * @rfn is not required, or even expected, to remove all requests off the
468 * queue, but only as many as it can handle at a time. If it does leave
469 * requests on the queue, it is responsible for arranging that the requests
470 * get dealt with eventually.
472 * The queue spin lock must be held while manipulating the requests on the
473 * request queue; this lock will be taken also from interrupt context, so irq
474 * disabling is needed for it.
476 * Function returns a pointer to the initialized request queue, or %NULL if
480 * blk_init_queue() must be paired with a blk_cleanup_queue() call
481 * when the block device is deactivated (such as at module unload).
484 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
486 return blk_init_queue_node(rfn
, lock
, -1);
488 EXPORT_SYMBOL(blk_init_queue
);
490 struct request_queue
*
491 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
493 struct request_queue
*uninit_q
, *q
;
495 uninit_q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
499 q
= blk_init_allocated_queue_node(uninit_q
, rfn
, lock
, node_id
);
501 blk_cleanup_queue(uninit_q
);
505 EXPORT_SYMBOL(blk_init_queue_node
);
507 struct request_queue
*
508 blk_init_allocated_queue(struct request_queue
*q
, request_fn_proc
*rfn
,
511 return blk_init_allocated_queue_node(q
, rfn
, lock
, -1);
513 EXPORT_SYMBOL(blk_init_allocated_queue
);
515 struct request_queue
*
516 blk_init_allocated_queue_node(struct request_queue
*q
, request_fn_proc
*rfn
,
517 spinlock_t
*lock
, int node_id
)
523 if (blk_init_free_list(q
))
527 q
->prep_rq_fn
= NULL
;
528 q
->unprep_rq_fn
= NULL
;
529 q
->queue_flags
= QUEUE_FLAG_DEFAULT
;
531 /* Override internal queue lock with supplied lock pointer */
533 q
->queue_lock
= lock
;
536 * This also sets hw/phys segments, boundary and size
538 blk_queue_make_request(q
, __make_request
);
540 q
->sg_reserved_size
= INT_MAX
;
545 if (!elevator_init(q
, NULL
)) {
546 blk_queue_congestion_threshold(q
);
552 EXPORT_SYMBOL(blk_init_allocated_queue_node
);
554 int blk_get_queue(struct request_queue
*q
)
556 if (likely(!test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
))) {
557 kobject_get(&q
->kobj
);
564 static inline void blk_free_request(struct request_queue
*q
, struct request
*rq
)
566 BUG_ON(rq
->cmd_flags
& REQ_ON_PLUG
);
568 if (rq
->cmd_flags
& REQ_ELVPRIV
)
569 elv_put_request(q
, rq
);
570 mempool_free(rq
, q
->rq
.rq_pool
);
573 static struct request
*
574 blk_alloc_request(struct request_queue
*q
, int flags
, int priv
, gfp_t gfp_mask
)
576 struct request
*rq
= mempool_alloc(q
->rq
.rq_pool
, gfp_mask
);
583 rq
->cmd_flags
= flags
| REQ_ALLOCED
;
586 if (unlikely(elv_set_request(q
, rq
, gfp_mask
))) {
587 mempool_free(rq
, q
->rq
.rq_pool
);
590 rq
->cmd_flags
|= REQ_ELVPRIV
;
597 * ioc_batching returns true if the ioc is a valid batching request and
598 * should be given priority access to a request.
600 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
606 * Make sure the process is able to allocate at least 1 request
607 * even if the batch times out, otherwise we could theoretically
610 return ioc
->nr_batch_requests
== q
->nr_batching
||
611 (ioc
->nr_batch_requests
> 0
612 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
616 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
617 * will cause the process to be a "batcher" on all queues in the system. This
618 * is the behaviour we want though - once it gets a wakeup it should be given
621 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
623 if (!ioc
|| ioc_batching(q
, ioc
))
626 ioc
->nr_batch_requests
= q
->nr_batching
;
627 ioc
->last_waited
= jiffies
;
630 static void __freed_request(struct request_queue
*q
, int sync
)
632 struct request_list
*rl
= &q
->rq
;
634 if (rl
->count
[sync
] < queue_congestion_off_threshold(q
))
635 blk_clear_queue_congested(q
, sync
);
637 if (rl
->count
[sync
] + 1 <= q
->nr_requests
) {
638 if (waitqueue_active(&rl
->wait
[sync
]))
639 wake_up(&rl
->wait
[sync
]);
641 blk_clear_queue_full(q
, sync
);
646 * A request has just been released. Account for it, update the full and
647 * congestion status, wake up any waiters. Called under q->queue_lock.
649 static void freed_request(struct request_queue
*q
, int sync
, int priv
)
651 struct request_list
*rl
= &q
->rq
;
657 __freed_request(q
, sync
);
659 if (unlikely(rl
->starved
[sync
^ 1]))
660 __freed_request(q
, sync
^ 1);
664 * Determine if elevator data should be initialized when allocating the
665 * request associated with @bio.
667 static bool blk_rq_should_init_elevator(struct bio
*bio
)
673 * Flush requests do not use the elevator so skip initialization.
674 * This allows a request to share the flush and elevator data.
676 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
))
683 * Get a free request, queue_lock must be held.
684 * Returns NULL on failure, with queue_lock held.
685 * Returns !NULL on success, with queue_lock *not held*.
687 static struct request
*get_request(struct request_queue
*q
, int rw_flags
,
688 struct bio
*bio
, gfp_t gfp_mask
)
690 struct request
*rq
= NULL
;
691 struct request_list
*rl
= &q
->rq
;
692 struct io_context
*ioc
= NULL
;
693 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
694 int may_queue
, priv
= 0;
696 may_queue
= elv_may_queue(q
, rw_flags
);
697 if (may_queue
== ELV_MQUEUE_NO
)
700 if (rl
->count
[is_sync
]+1 >= queue_congestion_on_threshold(q
)) {
701 if (rl
->count
[is_sync
]+1 >= q
->nr_requests
) {
702 ioc
= current_io_context(GFP_ATOMIC
, q
->node
);
704 * The queue will fill after this allocation, so set
705 * it as full, and mark this process as "batching".
706 * This process will be allowed to complete a batch of
707 * requests, others will be blocked.
709 if (!blk_queue_full(q
, is_sync
)) {
710 ioc_set_batching(q
, ioc
);
711 blk_set_queue_full(q
, is_sync
);
713 if (may_queue
!= ELV_MQUEUE_MUST
714 && !ioc_batching(q
, ioc
)) {
716 * The queue is full and the allocating
717 * process is not a "batcher", and not
718 * exempted by the IO scheduler
724 blk_set_queue_congested(q
, is_sync
);
728 * Only allow batching queuers to allocate up to 50% over the defined
729 * limit of requests, otherwise we could have thousands of requests
730 * allocated with any setting of ->nr_requests
732 if (rl
->count
[is_sync
] >= (3 * q
->nr_requests
/ 2))
735 rl
->count
[is_sync
]++;
736 rl
->starved
[is_sync
] = 0;
738 if (blk_rq_should_init_elevator(bio
)) {
739 priv
= !test_bit(QUEUE_FLAG_ELVSWITCH
, &q
->queue_flags
);
744 if (blk_queue_io_stat(q
))
745 rw_flags
|= REQ_IO_STAT
;
746 spin_unlock_irq(q
->queue_lock
);
748 rq
= blk_alloc_request(q
, rw_flags
, priv
, gfp_mask
);
751 * Allocation failed presumably due to memory. Undo anything
752 * we might have messed up.
754 * Allocating task should really be put onto the front of the
755 * wait queue, but this is pretty rare.
757 spin_lock_irq(q
->queue_lock
);
758 freed_request(q
, is_sync
, priv
);
761 * in the very unlikely event that allocation failed and no
762 * requests for this direction was pending, mark us starved
763 * so that freeing of a request in the other direction will
764 * notice us. another possible fix would be to split the
765 * rq mempool into READ and WRITE
768 if (unlikely(rl
->count
[is_sync
] == 0))
769 rl
->starved
[is_sync
] = 1;
775 * ioc may be NULL here, and ioc_batching will be false. That's
776 * OK, if the queue is under the request limit then requests need
777 * not count toward the nr_batch_requests limit. There will always
778 * be some limit enforced by BLK_BATCH_TIME.
780 if (ioc_batching(q
, ioc
))
781 ioc
->nr_batch_requests
--;
783 trace_block_getrq(q
, bio
, rw_flags
& 1);
789 * No available requests for this queue, wait for some requests to become
792 * Called with q->queue_lock held, and returns with it unlocked.
794 static struct request
*get_request_wait(struct request_queue
*q
, int rw_flags
,
797 const bool is_sync
= rw_is_sync(rw_flags
) != 0;
800 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
803 struct io_context
*ioc
;
804 struct request_list
*rl
= &q
->rq
;
806 prepare_to_wait_exclusive(&rl
->wait
[is_sync
], &wait
,
807 TASK_UNINTERRUPTIBLE
);
809 trace_block_sleeprq(q
, bio
, rw_flags
& 1);
811 spin_unlock_irq(q
->queue_lock
);
815 * After sleeping, we become a "batching" process and
816 * will be able to allocate at least one request, and
817 * up to a big batch of them for a small period time.
818 * See ioc_batching, ioc_set_batching
820 ioc
= current_io_context(GFP_NOIO
, q
->node
);
821 ioc_set_batching(q
, ioc
);
823 spin_lock_irq(q
->queue_lock
);
824 finish_wait(&rl
->wait
[is_sync
], &wait
);
826 rq
= get_request(q
, rw_flags
, bio
, GFP_NOIO
);
832 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
836 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
838 spin_lock_irq(q
->queue_lock
);
839 if (gfp_mask
& __GFP_WAIT
) {
840 rq
= get_request_wait(q
, rw
, NULL
);
842 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
844 spin_unlock_irq(q
->queue_lock
);
846 /* q->queue_lock is unlocked at this point */
850 EXPORT_SYMBOL(blk_get_request
);
853 * blk_make_request - given a bio, allocate a corresponding struct request.
854 * @q: target request queue
855 * @bio: The bio describing the memory mappings that will be submitted for IO.
856 * It may be a chained-bio properly constructed by block/bio layer.
857 * @gfp_mask: gfp flags to be used for memory allocation
859 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
860 * type commands. Where the struct request needs to be farther initialized by
861 * the caller. It is passed a &struct bio, which describes the memory info of
864 * The caller of blk_make_request must make sure that bi_io_vec
865 * are set to describe the memory buffers. That bio_data_dir() will return
866 * the needed direction of the request. (And all bio's in the passed bio-chain
867 * are properly set accordingly)
869 * If called under none-sleepable conditions, mapped bio buffers must not
870 * need bouncing, by calling the appropriate masked or flagged allocator,
871 * suitable for the target device. Otherwise the call to blk_queue_bounce will
874 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
875 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
876 * anything but the first bio in the chain. Otherwise you risk waiting for IO
877 * completion of a bio that hasn't been submitted yet, thus resulting in a
878 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
879 * of bio_alloc(), as that avoids the mempool deadlock.
880 * If possible a big IO should be split into smaller parts when allocation
881 * fails. Partial allocation should not be an error, or you risk a live-lock.
883 struct request
*blk_make_request(struct request_queue
*q
, struct bio
*bio
,
886 struct request
*rq
= blk_get_request(q
, bio_data_dir(bio
), gfp_mask
);
889 return ERR_PTR(-ENOMEM
);
892 struct bio
*bounce_bio
= bio
;
895 blk_queue_bounce(q
, &bounce_bio
);
896 ret
= blk_rq_append_bio(q
, rq
, bounce_bio
);
905 EXPORT_SYMBOL(blk_make_request
);
908 * blk_requeue_request - put a request back on queue
909 * @q: request queue where request should be inserted
910 * @rq: request to be inserted
913 * Drivers often keep queueing requests until the hardware cannot accept
914 * more, when that condition happens we need to put the request back
915 * on the queue. Must be called with queue lock held.
917 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
919 blk_delete_timer(rq
);
920 blk_clear_rq_complete(rq
);
921 trace_block_rq_requeue(q
, rq
);
923 if (blk_rq_tagged(rq
))
924 blk_queue_end_tag(q
, rq
);
926 BUG_ON(blk_queued_rq(rq
));
928 elv_requeue_request(q
, rq
);
930 EXPORT_SYMBOL(blk_requeue_request
);
932 static void add_acct_request(struct request_queue
*q
, struct request
*rq
,
935 drive_stat_acct(rq
, 1);
936 __elv_add_request(q
, rq
, where
);
940 * blk_insert_request - insert a special request into a request queue
941 * @q: request queue where request should be inserted
942 * @rq: request to be inserted
943 * @at_head: insert request at head or tail of queue
944 * @data: private data
947 * Many block devices need to execute commands asynchronously, so they don't
948 * block the whole kernel from preemption during request execution. This is
949 * accomplished normally by inserting aritficial requests tagged as
950 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
951 * be scheduled for actual execution by the request queue.
953 * We have the option of inserting the head or the tail of the queue.
954 * Typically we use the tail for new ioctls and so forth. We use the head
955 * of the queue for things like a QUEUE_FULL message from a device, or a
956 * host that is unable to accept a particular command.
958 void blk_insert_request(struct request_queue
*q
, struct request
*rq
,
959 int at_head
, void *data
)
961 int where
= at_head
? ELEVATOR_INSERT_FRONT
: ELEVATOR_INSERT_BACK
;
965 * tell I/O scheduler that this isn't a regular read/write (ie it
966 * must not attempt merges on this) and that it acts as a soft
969 rq
->cmd_type
= REQ_TYPE_SPECIAL
;
973 spin_lock_irqsave(q
->queue_lock
, flags
);
976 * If command is tagged, release the tag
978 if (blk_rq_tagged(rq
))
979 blk_queue_end_tag(q
, rq
);
981 add_acct_request(q
, rq
, where
);
982 __blk_run_queue(q
, false);
983 spin_unlock_irqrestore(q
->queue_lock
, flags
);
985 EXPORT_SYMBOL(blk_insert_request
);
987 static void part_round_stats_single(int cpu
, struct hd_struct
*part
,
990 if (now
== part
->stamp
)
993 if (part_in_flight(part
)) {
994 __part_stat_add(cpu
, part
, time_in_queue
,
995 part_in_flight(part
) * (now
- part
->stamp
));
996 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
1002 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1003 * @cpu: cpu number for stats access
1004 * @part: target partition
1006 * The average IO queue length and utilisation statistics are maintained
1007 * by observing the current state of the queue length and the amount of
1008 * time it has been in this state for.
1010 * Normally, that accounting is done on IO completion, but that can result
1011 * in more than a second's worth of IO being accounted for within any one
1012 * second, leading to >100% utilisation. To deal with that, we call this
1013 * function to do a round-off before returning the results when reading
1014 * /proc/diskstats. This accounts immediately for all queue usage up to
1015 * the current jiffies and restarts the counters again.
1017 void part_round_stats(int cpu
, struct hd_struct
*part
)
1019 unsigned long now
= jiffies
;
1022 part_round_stats_single(cpu
, &part_to_disk(part
)->part0
, now
);
1023 part_round_stats_single(cpu
, part
, now
);
1025 EXPORT_SYMBOL_GPL(part_round_stats
);
1028 * queue lock must be held
1030 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1034 if (unlikely(--req
->ref_count
))
1037 elv_completed_request(q
, req
);
1039 /* this is a bio leak */
1040 WARN_ON(req
->bio
!= NULL
);
1043 * Request may not have originated from ll_rw_blk. if not,
1044 * it didn't come out of our reserved rq pools
1046 if (req
->cmd_flags
& REQ_ALLOCED
) {
1047 int is_sync
= rq_is_sync(req
) != 0;
1048 int priv
= req
->cmd_flags
& REQ_ELVPRIV
;
1050 BUG_ON(!list_empty(&req
->queuelist
));
1051 BUG_ON(!hlist_unhashed(&req
->hash
));
1053 blk_free_request(q
, req
);
1054 freed_request(q
, is_sync
, priv
);
1057 EXPORT_SYMBOL_GPL(__blk_put_request
);
1059 void blk_put_request(struct request
*req
)
1061 unsigned long flags
;
1062 struct request_queue
*q
= req
->q
;
1064 spin_lock_irqsave(q
->queue_lock
, flags
);
1065 __blk_put_request(q
, req
);
1066 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1068 EXPORT_SYMBOL(blk_put_request
);
1071 * blk_add_request_payload - add a payload to a request
1072 * @rq: request to update
1073 * @page: page backing the payload
1074 * @len: length of the payload.
1076 * This allows to later add a payload to an already submitted request by
1077 * a block driver. The driver needs to take care of freeing the payload
1080 * Note that this is a quite horrible hack and nothing but handling of
1081 * discard requests should ever use it.
1083 void blk_add_request_payload(struct request
*rq
, struct page
*page
,
1086 struct bio
*bio
= rq
->bio
;
1088 bio
->bi_io_vec
->bv_page
= page
;
1089 bio
->bi_io_vec
->bv_offset
= 0;
1090 bio
->bi_io_vec
->bv_len
= len
;
1094 bio
->bi_phys_segments
= 1;
1096 rq
->__data_len
= rq
->resid_len
= len
;
1097 rq
->nr_phys_segments
= 1;
1098 rq
->buffer
= bio_data(bio
);
1100 EXPORT_SYMBOL_GPL(blk_add_request_payload
);
1102 static bool bio_attempt_back_merge(struct request_queue
*q
, struct request
*req
,
1105 const int ff
= bio
->bi_rw
& REQ_FAILFAST_MASK
;
1108 * Debug stuff, kill later
1110 if (!rq_mergeable(req
)) {
1111 blk_dump_rq_flags(req
, "back");
1115 if (!ll_back_merge_fn(q
, req
, bio
))
1118 trace_block_bio_backmerge(q
, bio
);
1120 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1121 blk_rq_set_mixed_merge(req
);
1123 req
->biotail
->bi_next
= bio
;
1125 req
->__data_len
+= bio
->bi_size
;
1126 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1128 drive_stat_acct(req
, 0);
1132 static bool bio_attempt_front_merge(struct request_queue
*q
,
1133 struct request
*req
, struct bio
*bio
)
1135 const int ff
= bio
->bi_rw
& REQ_FAILFAST_MASK
;
1139 * Debug stuff, kill later
1141 if (!rq_mergeable(req
)) {
1142 blk_dump_rq_flags(req
, "front");
1146 if (!ll_front_merge_fn(q
, req
, bio
))
1149 trace_block_bio_frontmerge(q
, bio
);
1151 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1152 blk_rq_set_mixed_merge(req
);
1154 sector
= bio
->bi_sector
;
1156 bio
->bi_next
= req
->bio
;
1160 * may not be valid. if the low level driver said
1161 * it didn't need a bounce buffer then it better
1162 * not touch req->buffer either...
1164 req
->buffer
= bio_data(bio
);
1165 req
->__sector
= bio
->bi_sector
;
1166 req
->__data_len
+= bio
->bi_size
;
1167 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1169 drive_stat_acct(req
, 0);
1174 * Attempts to merge with the plugged list in the current process. Returns
1175 * true if merge was successful, otherwise false.
1177 static bool attempt_plug_merge(struct task_struct
*tsk
, struct request_queue
*q
,
1180 struct blk_plug
*plug
;
1188 list_for_each_entry_reverse(rq
, &plug
->list
, queuelist
) {
1194 el_ret
= elv_try_merge(rq
, bio
);
1195 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1196 ret
= bio_attempt_back_merge(q
, rq
, bio
);
1199 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1200 ret
= bio_attempt_front_merge(q
, rq
, bio
);
1209 void init_request_from_bio(struct request
*req
, struct bio
*bio
)
1211 req
->cpu
= bio
->bi_comp_cpu
;
1212 req
->cmd_type
= REQ_TYPE_FS
;
1214 req
->cmd_flags
|= bio
->bi_rw
& REQ_COMMON_MASK
;
1215 if (bio
->bi_rw
& REQ_RAHEAD
)
1216 req
->cmd_flags
|= REQ_FAILFAST_MASK
;
1219 req
->__sector
= bio
->bi_sector
;
1220 req
->ioprio
= bio_prio(bio
);
1221 blk_rq_bio_prep(req
->q
, req
, bio
);
1224 static int __make_request(struct request_queue
*q
, struct bio
*bio
)
1226 const bool sync
= !!(bio
->bi_rw
& REQ_SYNC
);
1227 struct blk_plug
*plug
;
1228 int el_ret
, rw_flags
, where
= ELEVATOR_INSERT_SORT
;
1229 struct request
*req
;
1232 * low level driver can indicate that it wants pages above a
1233 * certain limit bounced to low memory (ie for highmem, or even
1234 * ISA dma in theory)
1236 blk_queue_bounce(q
, &bio
);
1238 if (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) {
1239 spin_lock_irq(q
->queue_lock
);
1240 where
= ELEVATOR_INSERT_FLUSH
;
1245 * Check if we can merge with the plugged list before grabbing
1248 if (attempt_plug_merge(current
, q
, bio
))
1251 spin_lock_irq(q
->queue_lock
);
1253 el_ret
= elv_merge(q
, &req
, bio
);
1254 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1255 BUG_ON(req
->cmd_flags
& REQ_ON_PLUG
);
1256 if (bio_attempt_back_merge(q
, req
, bio
)) {
1257 if (!attempt_back_merge(q
, req
))
1258 elv_merged_request(q
, req
, el_ret
);
1261 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1262 BUG_ON(req
->cmd_flags
& REQ_ON_PLUG
);
1263 if (bio_attempt_front_merge(q
, req
, bio
)) {
1264 if (!attempt_front_merge(q
, req
))
1265 elv_merged_request(q
, req
, el_ret
);
1272 * This sync check and mask will be re-done in init_request_from_bio(),
1273 * but we need to set it earlier to expose the sync flag to the
1274 * rq allocator and io schedulers.
1276 rw_flags
= bio_data_dir(bio
);
1278 rw_flags
|= REQ_SYNC
;
1281 * Grab a free request. This is might sleep but can not fail.
1282 * Returns with the queue unlocked.
1284 req
= get_request_wait(q
, rw_flags
, bio
);
1287 * After dropping the lock and possibly sleeping here, our request
1288 * may now be mergeable after it had proven unmergeable (above).
1289 * We don't worry about that case for efficiency. It won't happen
1290 * often, and the elevators are able to handle it.
1292 init_request_from_bio(req
, bio
);
1294 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
) ||
1295 bio_flagged(bio
, BIO_CPU_AFFINE
)) {
1296 req
->cpu
= blk_cpu_to_group(get_cpu());
1300 plug
= current
->plug
;
1303 * If this is the first request added after a plug, fire
1304 * of a plug trace. If others have been added before, check
1305 * if we have multiple devices in this plug. If so, make a
1306 * note to sort the list before dispatch.
1308 if (list_empty(&plug
->list
))
1309 trace_block_plug(q
);
1310 else if (!plug
->should_sort
) {
1311 struct request
*__rq
;
1313 __rq
= list_entry_rq(plug
->list
.prev
);
1315 plug
->should_sort
= 1;
1318 * Debug flag, kill later
1320 req
->cmd_flags
|= REQ_ON_PLUG
;
1321 list_add_tail(&req
->queuelist
, &plug
->list
);
1322 drive_stat_acct(req
, 1);
1324 spin_lock_irq(q
->queue_lock
);
1325 add_acct_request(q
, req
, where
);
1326 __blk_run_queue(q
, false);
1328 spin_unlock_irq(q
->queue_lock
);
1335 * If bio->bi_dev is a partition, remap the location
1337 static inline void blk_partition_remap(struct bio
*bio
)
1339 struct block_device
*bdev
= bio
->bi_bdev
;
1341 if (bio_sectors(bio
) && bdev
!= bdev
->bd_contains
) {
1342 struct hd_struct
*p
= bdev
->bd_part
;
1344 bio
->bi_sector
+= p
->start_sect
;
1345 bio
->bi_bdev
= bdev
->bd_contains
;
1347 trace_block_bio_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
1349 bio
->bi_sector
- p
->start_sect
);
1353 static void handle_bad_sector(struct bio
*bio
)
1355 char b
[BDEVNAME_SIZE
];
1357 printk(KERN_INFO
"attempt to access beyond end of device\n");
1358 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
1359 bdevname(bio
->bi_bdev
, b
),
1361 (unsigned long long)bio
->bi_sector
+ bio_sectors(bio
),
1362 (long long)(i_size_read(bio
->bi_bdev
->bd_inode
) >> 9));
1364 set_bit(BIO_EOF
, &bio
->bi_flags
);
1367 #ifdef CONFIG_FAIL_MAKE_REQUEST
1369 static DECLARE_FAULT_ATTR(fail_make_request
);
1371 static int __init
setup_fail_make_request(char *str
)
1373 return setup_fault_attr(&fail_make_request
, str
);
1375 __setup("fail_make_request=", setup_fail_make_request
);
1377 static int should_fail_request(struct bio
*bio
)
1379 struct hd_struct
*part
= bio
->bi_bdev
->bd_part
;
1381 if (part_to_disk(part
)->part0
.make_it_fail
|| part
->make_it_fail
)
1382 return should_fail(&fail_make_request
, bio
->bi_size
);
1387 static int __init
fail_make_request_debugfs(void)
1389 return init_fault_attr_dentries(&fail_make_request
,
1390 "fail_make_request");
1393 late_initcall(fail_make_request_debugfs
);
1395 #else /* CONFIG_FAIL_MAKE_REQUEST */
1397 static inline int should_fail_request(struct bio
*bio
)
1402 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1405 * Check whether this bio extends beyond the end of the device.
1407 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
1414 /* Test device or partition size, when known. */
1415 maxsector
= i_size_read(bio
->bi_bdev
->bd_inode
) >> 9;
1417 sector_t sector
= bio
->bi_sector
;
1419 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
1421 * This may well happen - the kernel calls bread()
1422 * without checking the size of the device, e.g., when
1423 * mounting a device.
1425 handle_bad_sector(bio
);
1434 * generic_make_request - hand a buffer to its device driver for I/O
1435 * @bio: The bio describing the location in memory and on the device.
1437 * generic_make_request() is used to make I/O requests of block
1438 * devices. It is passed a &struct bio, which describes the I/O that needs
1441 * generic_make_request() does not return any status. The
1442 * success/failure status of the request, along with notification of
1443 * completion, is delivered asynchronously through the bio->bi_end_io
1444 * function described (one day) else where.
1446 * The caller of generic_make_request must make sure that bi_io_vec
1447 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1448 * set to describe the device address, and the
1449 * bi_end_io and optionally bi_private are set to describe how
1450 * completion notification should be signaled.
1452 * generic_make_request and the drivers it calls may use bi_next if this
1453 * bio happens to be merged with someone else, and may change bi_dev and
1454 * bi_sector for remaps as it sees fit. So the values of these fields
1455 * should NOT be depended on after the call to generic_make_request.
1457 static inline void __generic_make_request(struct bio
*bio
)
1459 struct request_queue
*q
;
1460 sector_t old_sector
;
1461 int ret
, nr_sectors
= bio_sectors(bio
);
1467 if (bio_check_eod(bio
, nr_sectors
))
1471 * Resolve the mapping until finished. (drivers are
1472 * still free to implement/resolve their own stacking
1473 * by explicitly returning 0)
1475 * NOTE: we don't repeat the blk_size check for each new device.
1476 * Stacking drivers are expected to know what they are doing.
1481 char b
[BDEVNAME_SIZE
];
1483 q
= bdev_get_queue(bio
->bi_bdev
);
1486 "generic_make_request: Trying to access "
1487 "nonexistent block-device %s (%Lu)\n",
1488 bdevname(bio
->bi_bdev
, b
),
1489 (long long) bio
->bi_sector
);
1493 if (unlikely(!(bio
->bi_rw
& REQ_DISCARD
) &&
1494 nr_sectors
> queue_max_hw_sectors(q
))) {
1495 printk(KERN_ERR
"bio too big device %s (%u > %u)\n",
1496 bdevname(bio
->bi_bdev
, b
),
1498 queue_max_hw_sectors(q
));
1502 if (unlikely(test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)))
1505 if (should_fail_request(bio
))
1509 * If this device has partitions, remap block n
1510 * of partition p to block n+start(p) of the disk.
1512 blk_partition_remap(bio
);
1514 if (bio_integrity_enabled(bio
) && bio_integrity_prep(bio
))
1517 if (old_sector
!= -1)
1518 trace_block_bio_remap(q
, bio
, old_dev
, old_sector
);
1520 old_sector
= bio
->bi_sector
;
1521 old_dev
= bio
->bi_bdev
->bd_dev
;
1523 if (bio_check_eod(bio
, nr_sectors
))
1527 * Filter flush bio's early so that make_request based
1528 * drivers without flush support don't have to worry
1531 if ((bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
)) && !q
->flush_flags
) {
1532 bio
->bi_rw
&= ~(REQ_FLUSH
| REQ_FUA
);
1539 if ((bio
->bi_rw
& REQ_DISCARD
) &&
1540 (!blk_queue_discard(q
) ||
1541 ((bio
->bi_rw
& REQ_SECURE
) &&
1542 !blk_queue_secdiscard(q
)))) {
1547 blk_throtl_bio(q
, &bio
);
1550 * If bio = NULL, bio has been throttled and will be submitted
1556 trace_block_bio_queue(q
, bio
);
1558 ret
= q
->make_request_fn(q
, bio
);
1564 bio_endio(bio
, err
);
1568 * We only want one ->make_request_fn to be active at a time,
1569 * else stack usage with stacked devices could be a problem.
1570 * So use current->bio_list to keep a list of requests
1571 * submited by a make_request_fn function.
1572 * current->bio_list is also used as a flag to say if
1573 * generic_make_request is currently active in this task or not.
1574 * If it is NULL, then no make_request is active. If it is non-NULL,
1575 * then a make_request is active, and new requests should be added
1578 void generic_make_request(struct bio
*bio
)
1580 struct bio_list bio_list_on_stack
;
1582 if (current
->bio_list
) {
1583 /* make_request is active */
1584 bio_list_add(current
->bio_list
, bio
);
1587 /* following loop may be a bit non-obvious, and so deserves some
1589 * Before entering the loop, bio->bi_next is NULL (as all callers
1590 * ensure that) so we have a list with a single bio.
1591 * We pretend that we have just taken it off a longer list, so
1592 * we assign bio_list to a pointer to the bio_list_on_stack,
1593 * thus initialising the bio_list of new bios to be
1594 * added. __generic_make_request may indeed add some more bios
1595 * through a recursive call to generic_make_request. If it
1596 * did, we find a non-NULL value in bio_list and re-enter the loop
1597 * from the top. In this case we really did just take the bio
1598 * of the top of the list (no pretending) and so remove it from
1599 * bio_list, and call into __generic_make_request again.
1601 * The loop was structured like this to make only one call to
1602 * __generic_make_request (which is important as it is large and
1603 * inlined) and to keep the structure simple.
1605 BUG_ON(bio
->bi_next
);
1606 bio_list_init(&bio_list_on_stack
);
1607 current
->bio_list
= &bio_list_on_stack
;
1609 __generic_make_request(bio
);
1610 bio
= bio_list_pop(current
->bio_list
);
1612 current
->bio_list
= NULL
; /* deactivate */
1614 EXPORT_SYMBOL(generic_make_request
);
1617 * submit_bio - submit a bio to the block device layer for I/O
1618 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1619 * @bio: The &struct bio which describes the I/O
1621 * submit_bio() is very similar in purpose to generic_make_request(), and
1622 * uses that function to do most of the work. Both are fairly rough
1623 * interfaces; @bio must be presetup and ready for I/O.
1626 void submit_bio(int rw
, struct bio
*bio
)
1628 int count
= bio_sectors(bio
);
1633 * If it's a regular read/write or a barrier with data attached,
1634 * go through the normal accounting stuff before submission.
1636 if (bio_has_data(bio
) && !(rw
& REQ_DISCARD
)) {
1638 count_vm_events(PGPGOUT
, count
);
1640 task_io_account_read(bio
->bi_size
);
1641 count_vm_events(PGPGIN
, count
);
1644 if (unlikely(block_dump
)) {
1645 char b
[BDEVNAME_SIZE
];
1646 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s (%u sectors)\n",
1647 current
->comm
, task_pid_nr(current
),
1648 (rw
& WRITE
) ? "WRITE" : "READ",
1649 (unsigned long long)bio
->bi_sector
,
1650 bdevname(bio
->bi_bdev
, b
),
1655 generic_make_request(bio
);
1657 EXPORT_SYMBOL(submit_bio
);
1660 * blk_rq_check_limits - Helper function to check a request for the queue limit
1662 * @rq: the request being checked
1665 * @rq may have been made based on weaker limitations of upper-level queues
1666 * in request stacking drivers, and it may violate the limitation of @q.
1667 * Since the block layer and the underlying device driver trust @rq
1668 * after it is inserted to @q, it should be checked against @q before
1669 * the insertion using this generic function.
1671 * This function should also be useful for request stacking drivers
1672 * in some cases below, so export this function.
1673 * Request stacking drivers like request-based dm may change the queue
1674 * limits while requests are in the queue (e.g. dm's table swapping).
1675 * Such request stacking drivers should check those requests agaist
1676 * the new queue limits again when they dispatch those requests,
1677 * although such checkings are also done against the old queue limits
1678 * when submitting requests.
1680 int blk_rq_check_limits(struct request_queue
*q
, struct request
*rq
)
1682 if (rq
->cmd_flags
& REQ_DISCARD
)
1685 if (blk_rq_sectors(rq
) > queue_max_sectors(q
) ||
1686 blk_rq_bytes(rq
) > queue_max_hw_sectors(q
) << 9) {
1687 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
1692 * queue's settings related to segment counting like q->bounce_pfn
1693 * may differ from that of other stacking queues.
1694 * Recalculate it to check the request correctly on this queue's
1697 blk_recalc_rq_segments(rq
);
1698 if (rq
->nr_phys_segments
> queue_max_segments(q
)) {
1699 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
1705 EXPORT_SYMBOL_GPL(blk_rq_check_limits
);
1708 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1709 * @q: the queue to submit the request
1710 * @rq: the request being queued
1712 int blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
1714 unsigned long flags
;
1716 if (blk_rq_check_limits(q
, rq
))
1719 #ifdef CONFIG_FAIL_MAKE_REQUEST
1720 if (rq
->rq_disk
&& rq
->rq_disk
->part0
.make_it_fail
&&
1721 should_fail(&fail_make_request
, blk_rq_bytes(rq
)))
1725 spin_lock_irqsave(q
->queue_lock
, flags
);
1728 * Submitting request must be dequeued before calling this function
1729 * because it will be linked to another request_queue
1731 BUG_ON(blk_queued_rq(rq
));
1733 add_acct_request(q
, rq
, ELEVATOR_INSERT_BACK
);
1734 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1738 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
1741 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1742 * @rq: request to examine
1745 * A request could be merge of IOs which require different failure
1746 * handling. This function determines the number of bytes which
1747 * can be failed from the beginning of the request without
1748 * crossing into area which need to be retried further.
1751 * The number of bytes to fail.
1754 * queue_lock must be held.
1756 unsigned int blk_rq_err_bytes(const struct request
*rq
)
1758 unsigned int ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
1759 unsigned int bytes
= 0;
1762 if (!(rq
->cmd_flags
& REQ_MIXED_MERGE
))
1763 return blk_rq_bytes(rq
);
1766 * Currently the only 'mixing' which can happen is between
1767 * different fastfail types. We can safely fail portions
1768 * which have all the failfast bits that the first one has -
1769 * the ones which are at least as eager to fail as the first
1772 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
1773 if ((bio
->bi_rw
& ff
) != ff
)
1775 bytes
+= bio
->bi_size
;
1778 /* this could lead to infinite loop */
1779 BUG_ON(blk_rq_bytes(rq
) && !bytes
);
1782 EXPORT_SYMBOL_GPL(blk_rq_err_bytes
);
1784 static void blk_account_io_completion(struct request
*req
, unsigned int bytes
)
1786 if (blk_do_io_stat(req
)) {
1787 const int rw
= rq_data_dir(req
);
1788 struct hd_struct
*part
;
1791 cpu
= part_stat_lock();
1793 part_stat_add(cpu
, part
, sectors
[rw
], bytes
>> 9);
1798 static void blk_account_io_done(struct request
*req
)
1801 * Account IO completion. flush_rq isn't accounted as a
1802 * normal IO on queueing nor completion. Accounting the
1803 * containing request is enough.
1805 if (blk_do_io_stat(req
) && !(req
->cmd_flags
& REQ_FLUSH_SEQ
)) {
1806 unsigned long duration
= jiffies
- req
->start_time
;
1807 const int rw
= rq_data_dir(req
);
1808 struct hd_struct
*part
;
1811 cpu
= part_stat_lock();
1814 part_stat_inc(cpu
, part
, ios
[rw
]);
1815 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
1816 part_round_stats(cpu
, part
);
1817 part_dec_in_flight(part
, rw
);
1819 hd_struct_put(part
);
1825 * blk_peek_request - peek at the top of a request queue
1826 * @q: request queue to peek at
1829 * Return the request at the top of @q. The returned request
1830 * should be started using blk_start_request() before LLD starts
1834 * Pointer to the request at the top of @q if available. Null
1838 * queue_lock must be held.
1840 struct request
*blk_peek_request(struct request_queue
*q
)
1845 while ((rq
= __elv_next_request(q
)) != NULL
) {
1846 if (!(rq
->cmd_flags
& REQ_STARTED
)) {
1848 * This is the first time the device driver
1849 * sees this request (possibly after
1850 * requeueing). Notify IO scheduler.
1852 if (rq
->cmd_flags
& REQ_SORTED
)
1853 elv_activate_rq(q
, rq
);
1856 * just mark as started even if we don't start
1857 * it, a request that has been delayed should
1858 * not be passed by new incoming requests
1860 rq
->cmd_flags
|= REQ_STARTED
;
1861 trace_block_rq_issue(q
, rq
);
1864 if (!q
->boundary_rq
|| q
->boundary_rq
== rq
) {
1865 q
->end_sector
= rq_end_sector(rq
);
1866 q
->boundary_rq
= NULL
;
1869 if (rq
->cmd_flags
& REQ_DONTPREP
)
1872 if (q
->dma_drain_size
&& blk_rq_bytes(rq
)) {
1874 * make sure space for the drain appears we
1875 * know we can do this because max_hw_segments
1876 * has been adjusted to be one fewer than the
1879 rq
->nr_phys_segments
++;
1885 ret
= q
->prep_rq_fn(q
, rq
);
1886 if (ret
== BLKPREP_OK
) {
1888 } else if (ret
== BLKPREP_DEFER
) {
1890 * the request may have been (partially) prepped.
1891 * we need to keep this request in the front to
1892 * avoid resource deadlock. REQ_STARTED will
1893 * prevent other fs requests from passing this one.
1895 if (q
->dma_drain_size
&& blk_rq_bytes(rq
) &&
1896 !(rq
->cmd_flags
& REQ_DONTPREP
)) {
1898 * remove the space for the drain we added
1899 * so that we don't add it again
1901 --rq
->nr_phys_segments
;
1906 } else if (ret
== BLKPREP_KILL
) {
1907 rq
->cmd_flags
|= REQ_QUIET
;
1909 * Mark this request as started so we don't trigger
1910 * any debug logic in the end I/O path.
1912 blk_start_request(rq
);
1913 __blk_end_request_all(rq
, -EIO
);
1915 printk(KERN_ERR
"%s: bad return=%d\n", __func__
, ret
);
1922 EXPORT_SYMBOL(blk_peek_request
);
1924 void blk_dequeue_request(struct request
*rq
)
1926 struct request_queue
*q
= rq
->q
;
1928 BUG_ON(list_empty(&rq
->queuelist
));
1929 BUG_ON(ELV_ON_HASH(rq
));
1931 list_del_init(&rq
->queuelist
);
1934 * the time frame between a request being removed from the lists
1935 * and to it is freed is accounted as io that is in progress at
1938 if (blk_account_rq(rq
)) {
1939 q
->in_flight
[rq_is_sync(rq
)]++;
1940 set_io_start_time_ns(rq
);
1945 * blk_start_request - start request processing on the driver
1946 * @req: request to dequeue
1949 * Dequeue @req and start timeout timer on it. This hands off the
1950 * request to the driver.
1952 * Block internal functions which don't want to start timer should
1953 * call blk_dequeue_request().
1956 * queue_lock must be held.
1958 void blk_start_request(struct request
*req
)
1960 blk_dequeue_request(req
);
1963 * We are now handing the request to the hardware, initialize
1964 * resid_len to full count and add the timeout handler.
1966 req
->resid_len
= blk_rq_bytes(req
);
1967 if (unlikely(blk_bidi_rq(req
)))
1968 req
->next_rq
->resid_len
= blk_rq_bytes(req
->next_rq
);
1972 EXPORT_SYMBOL(blk_start_request
);
1975 * blk_fetch_request - fetch a request from a request queue
1976 * @q: request queue to fetch a request from
1979 * Return the request at the top of @q. The request is started on
1980 * return and LLD can start processing it immediately.
1983 * Pointer to the request at the top of @q if available. Null
1987 * queue_lock must be held.
1989 struct request
*blk_fetch_request(struct request_queue
*q
)
1993 rq
= blk_peek_request(q
);
1995 blk_start_request(rq
);
1998 EXPORT_SYMBOL(blk_fetch_request
);
2001 * blk_update_request - Special helper function for request stacking drivers
2002 * @req: the request being processed
2003 * @error: %0 for success, < %0 for error
2004 * @nr_bytes: number of bytes to complete @req
2007 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2008 * the request structure even if @req doesn't have leftover.
2009 * If @req has leftover, sets it up for the next range of segments.
2011 * This special helper function is only for request stacking drivers
2012 * (e.g. request-based dm) so that they can handle partial completion.
2013 * Actual device drivers should use blk_end_request instead.
2015 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2016 * %false return from this function.
2019 * %false - this request doesn't have any more data
2020 * %true - this request has more data
2022 bool blk_update_request(struct request
*req
, int error
, unsigned int nr_bytes
)
2024 int total_bytes
, bio_nbytes
, next_idx
= 0;
2030 trace_block_rq_complete(req
->q
, req
);
2033 * For fs requests, rq is just carrier of independent bio's
2034 * and each partial completion should be handled separately.
2035 * Reset per-request error on each partial completion.
2037 * TODO: tj: This is too subtle. It would be better to let
2038 * low level drivers do what they see fit.
2040 if (req
->cmd_type
== REQ_TYPE_FS
)
2043 if (error
&& req
->cmd_type
== REQ_TYPE_FS
&&
2044 !(req
->cmd_flags
& REQ_QUIET
)) {
2049 error_type
= "recoverable transport";
2052 error_type
= "critical target";
2055 error_type
= "critical nexus";
2062 printk(KERN_ERR
"end_request: %s error, dev %s, sector %llu\n",
2063 error_type
, req
->rq_disk
? req
->rq_disk
->disk_name
: "?",
2064 (unsigned long long)blk_rq_pos(req
));
2067 blk_account_io_completion(req
, nr_bytes
);
2069 total_bytes
= bio_nbytes
= 0;
2070 while ((bio
= req
->bio
) != NULL
) {
2073 if (nr_bytes
>= bio
->bi_size
) {
2074 req
->bio
= bio
->bi_next
;
2075 nbytes
= bio
->bi_size
;
2076 req_bio_endio(req
, bio
, nbytes
, error
);
2080 int idx
= bio
->bi_idx
+ next_idx
;
2082 if (unlikely(idx
>= bio
->bi_vcnt
)) {
2083 blk_dump_rq_flags(req
, "__end_that");
2084 printk(KERN_ERR
"%s: bio idx %d >= vcnt %d\n",
2085 __func__
, idx
, bio
->bi_vcnt
);
2089 nbytes
= bio_iovec_idx(bio
, idx
)->bv_len
;
2090 BIO_BUG_ON(nbytes
> bio
->bi_size
);
2093 * not a complete bvec done
2095 if (unlikely(nbytes
> nr_bytes
)) {
2096 bio_nbytes
+= nr_bytes
;
2097 total_bytes
+= nr_bytes
;
2102 * advance to the next vector
2105 bio_nbytes
+= nbytes
;
2108 total_bytes
+= nbytes
;
2114 * end more in this run, or just return 'not-done'
2116 if (unlikely(nr_bytes
<= 0))
2126 * Reset counters so that the request stacking driver
2127 * can find how many bytes remain in the request
2130 req
->__data_len
= 0;
2135 * if the request wasn't completed, update state
2138 req_bio_endio(req
, bio
, bio_nbytes
, error
);
2139 bio
->bi_idx
+= next_idx
;
2140 bio_iovec(bio
)->bv_offset
+= nr_bytes
;
2141 bio_iovec(bio
)->bv_len
-= nr_bytes
;
2144 req
->__data_len
-= total_bytes
;
2145 req
->buffer
= bio_data(req
->bio
);
2147 /* update sector only for requests with clear definition of sector */
2148 if (req
->cmd_type
== REQ_TYPE_FS
|| (req
->cmd_flags
& REQ_DISCARD
))
2149 req
->__sector
+= total_bytes
>> 9;
2151 /* mixed attributes always follow the first bio */
2152 if (req
->cmd_flags
& REQ_MIXED_MERGE
) {
2153 req
->cmd_flags
&= ~REQ_FAILFAST_MASK
;
2154 req
->cmd_flags
|= req
->bio
->bi_rw
& REQ_FAILFAST_MASK
;
2158 * If total number of sectors is less than the first segment
2159 * size, something has gone terribly wrong.
2161 if (blk_rq_bytes(req
) < blk_rq_cur_bytes(req
)) {
2162 blk_dump_rq_flags(req
, "request botched");
2163 req
->__data_len
= blk_rq_cur_bytes(req
);
2166 /* recalculate the number of segments */
2167 blk_recalc_rq_segments(req
);
2171 EXPORT_SYMBOL_GPL(blk_update_request
);
2173 static bool blk_update_bidi_request(struct request
*rq
, int error
,
2174 unsigned int nr_bytes
,
2175 unsigned int bidi_bytes
)
2177 if (blk_update_request(rq
, error
, nr_bytes
))
2180 /* Bidi request must be completed as a whole */
2181 if (unlikely(blk_bidi_rq(rq
)) &&
2182 blk_update_request(rq
->next_rq
, error
, bidi_bytes
))
2185 if (blk_queue_add_random(rq
->q
))
2186 add_disk_randomness(rq
->rq_disk
);
2192 * blk_unprep_request - unprepare a request
2195 * This function makes a request ready for complete resubmission (or
2196 * completion). It happens only after all error handling is complete,
2197 * so represents the appropriate moment to deallocate any resources
2198 * that were allocated to the request in the prep_rq_fn. The queue
2199 * lock is held when calling this.
2201 void blk_unprep_request(struct request
*req
)
2203 struct request_queue
*q
= req
->q
;
2205 req
->cmd_flags
&= ~REQ_DONTPREP
;
2206 if (q
->unprep_rq_fn
)
2207 q
->unprep_rq_fn(q
, req
);
2209 EXPORT_SYMBOL_GPL(blk_unprep_request
);
2212 * queue lock must be held
2214 static void blk_finish_request(struct request
*req
, int error
)
2216 if (blk_rq_tagged(req
))
2217 blk_queue_end_tag(req
->q
, req
);
2219 BUG_ON(blk_queued_rq(req
));
2221 if (unlikely(laptop_mode
) && req
->cmd_type
== REQ_TYPE_FS
)
2222 laptop_io_completion(&req
->q
->backing_dev_info
);
2224 blk_delete_timer(req
);
2226 if (req
->cmd_flags
& REQ_DONTPREP
)
2227 blk_unprep_request(req
);
2230 blk_account_io_done(req
);
2233 req
->end_io(req
, error
);
2235 if (blk_bidi_rq(req
))
2236 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
2238 __blk_put_request(req
->q
, req
);
2243 * blk_end_bidi_request - Complete a bidi request
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 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2251 * Drivers that supports bidi can safely call this member for any
2252 * type of request, bidi or uni. In the later case @bidi_bytes is
2256 * %false - we are done with this request
2257 * %true - still buffers pending for this request
2259 static bool blk_end_bidi_request(struct request
*rq
, int error
,
2260 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2262 struct request_queue
*q
= rq
->q
;
2263 unsigned long flags
;
2265 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2268 spin_lock_irqsave(q
->queue_lock
, flags
);
2269 blk_finish_request(rq
, error
);
2270 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2276 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2277 * @rq: the request to complete
2278 * @error: %0 for success, < %0 for error
2279 * @nr_bytes: number of bytes to complete @rq
2280 * @bidi_bytes: number of bytes to complete @rq->next_rq
2283 * Identical to blk_end_bidi_request() except that queue lock is
2284 * assumed to be locked on entry and remains so on return.
2287 * %false - we are done with this request
2288 * %true - still buffers pending for this request
2290 static bool __blk_end_bidi_request(struct request
*rq
, int error
,
2291 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2293 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2296 blk_finish_request(rq
, error
);
2302 * blk_end_request - Helper function for drivers to complete the request.
2303 * @rq: the request being processed
2304 * @error: %0 for success, < %0 for error
2305 * @nr_bytes: number of bytes to complete
2308 * Ends I/O on a number of bytes attached to @rq.
2309 * If @rq has leftover, sets it up for the next range of segments.
2312 * %false - we are done with this request
2313 * %true - still buffers pending for this request
2315 bool blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2317 return blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2319 EXPORT_SYMBOL(blk_end_request
);
2322 * blk_end_request_all - Helper function for drives to finish the request.
2323 * @rq: the request to finish
2324 * @error: %0 for success, < %0 for error
2327 * Completely finish @rq.
2329 void blk_end_request_all(struct request
*rq
, int error
)
2332 unsigned int bidi_bytes
= 0;
2334 if (unlikely(blk_bidi_rq(rq
)))
2335 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2337 pending
= blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2340 EXPORT_SYMBOL(blk_end_request_all
);
2343 * blk_end_request_cur - Helper function to finish the current request chunk.
2344 * @rq: the request to finish the current chunk for
2345 * @error: %0 for success, < %0 for error
2348 * Complete the current consecutively mapped chunk from @rq.
2351 * %false - we are done with this request
2352 * %true - still buffers pending for this request
2354 bool blk_end_request_cur(struct request
*rq
, int error
)
2356 return blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2358 EXPORT_SYMBOL(blk_end_request_cur
);
2361 * blk_end_request_err - Finish a request till the next failure boundary.
2362 * @rq: the request to finish till the next failure boundary for
2363 * @error: must be negative errno
2366 * Complete @rq till the next failure boundary.
2369 * %false - we are done with this request
2370 * %true - still buffers pending for this request
2372 bool blk_end_request_err(struct request
*rq
, int error
)
2374 WARN_ON(error
>= 0);
2375 return blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2377 EXPORT_SYMBOL_GPL(blk_end_request_err
);
2380 * __blk_end_request - Helper function for drivers to complete the request.
2381 * @rq: the request being processed
2382 * @error: %0 for success, < %0 for error
2383 * @nr_bytes: number of bytes to complete
2386 * Must be called with queue lock held unlike blk_end_request().
2389 * %false - we are done with this request
2390 * %true - still buffers pending for this request
2392 bool __blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2394 return __blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2396 EXPORT_SYMBOL(__blk_end_request
);
2399 * __blk_end_request_all - Helper function for drives to finish the request.
2400 * @rq: the request to finish
2401 * @error: %0 for success, < %0 for error
2404 * Completely finish @rq. Must be called with queue lock held.
2406 void __blk_end_request_all(struct request
*rq
, int error
)
2409 unsigned int bidi_bytes
= 0;
2411 if (unlikely(blk_bidi_rq(rq
)))
2412 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2414 pending
= __blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2417 EXPORT_SYMBOL(__blk_end_request_all
);
2420 * __blk_end_request_cur - Helper function to finish the current request chunk.
2421 * @rq: the request to finish the current chunk for
2422 * @error: %0 for success, < %0 for error
2425 * Complete the current consecutively mapped chunk from @rq. Must
2426 * be called with queue lock held.
2429 * %false - we are done with this request
2430 * %true - still buffers pending for this request
2432 bool __blk_end_request_cur(struct request
*rq
, int error
)
2434 return __blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2436 EXPORT_SYMBOL(__blk_end_request_cur
);
2439 * __blk_end_request_err - Finish a request till the next failure boundary.
2440 * @rq: the request to finish till the next failure boundary for
2441 * @error: must be negative errno
2444 * Complete @rq till the next failure boundary. Must be called
2445 * with queue lock held.
2448 * %false - we are done with this request
2449 * %true - still buffers pending for this request
2451 bool __blk_end_request_err(struct request
*rq
, int error
)
2453 WARN_ON(error
>= 0);
2454 return __blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2456 EXPORT_SYMBOL_GPL(__blk_end_request_err
);
2458 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
2461 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2462 rq
->cmd_flags
|= bio
->bi_rw
& REQ_WRITE
;
2464 if (bio_has_data(bio
)) {
2465 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2466 rq
->buffer
= bio_data(bio
);
2468 rq
->__data_len
= bio
->bi_size
;
2469 rq
->bio
= rq
->biotail
= bio
;
2472 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2475 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2477 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2478 * @rq: the request to be flushed
2481 * Flush all pages in @rq.
2483 void rq_flush_dcache_pages(struct request
*rq
)
2485 struct req_iterator iter
;
2486 struct bio_vec
*bvec
;
2488 rq_for_each_segment(bvec
, rq
, iter
)
2489 flush_dcache_page(bvec
->bv_page
);
2491 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages
);
2495 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2496 * @q : the queue of the device being checked
2499 * Check if underlying low-level drivers of a device are busy.
2500 * If the drivers want to export their busy state, they must set own
2501 * exporting function using blk_queue_lld_busy() first.
2503 * Basically, this function is used only by request stacking drivers
2504 * to stop dispatching requests to underlying devices when underlying
2505 * devices are busy. This behavior helps more I/O merging on the queue
2506 * of the request stacking driver and prevents I/O throughput regression
2507 * on burst I/O load.
2510 * 0 - Not busy (The request stacking driver should dispatch request)
2511 * 1 - Busy (The request stacking driver should stop dispatching request)
2513 int blk_lld_busy(struct request_queue
*q
)
2516 return q
->lld_busy_fn(q
);
2520 EXPORT_SYMBOL_GPL(blk_lld_busy
);
2523 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2524 * @rq: the clone request to be cleaned up
2527 * Free all bios in @rq for a cloned request.
2529 void blk_rq_unprep_clone(struct request
*rq
)
2533 while ((bio
= rq
->bio
) != NULL
) {
2534 rq
->bio
= bio
->bi_next
;
2539 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone
);
2542 * Copy attributes of the original request to the clone request.
2543 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2545 static void __blk_rq_prep_clone(struct request
*dst
, struct request
*src
)
2547 dst
->cpu
= src
->cpu
;
2548 dst
->cmd_flags
= (src
->cmd_flags
& REQ_CLONE_MASK
) | REQ_NOMERGE
;
2549 dst
->cmd_type
= src
->cmd_type
;
2550 dst
->__sector
= blk_rq_pos(src
);
2551 dst
->__data_len
= blk_rq_bytes(src
);
2552 dst
->nr_phys_segments
= src
->nr_phys_segments
;
2553 dst
->ioprio
= src
->ioprio
;
2554 dst
->extra_len
= src
->extra_len
;
2558 * blk_rq_prep_clone - Helper function to setup clone request
2559 * @rq: the request to be setup
2560 * @rq_src: original request to be cloned
2561 * @bs: bio_set that bios for clone are allocated from
2562 * @gfp_mask: memory allocation mask for bio
2563 * @bio_ctr: setup function to be called for each clone bio.
2564 * Returns %0 for success, non %0 for failure.
2565 * @data: private data to be passed to @bio_ctr
2568 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2569 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2570 * are not copied, and copying such parts is the caller's responsibility.
2571 * Also, pages which the original bios are pointing to are not copied
2572 * and the cloned bios just point same pages.
2573 * So cloned bios must be completed before original bios, which means
2574 * the caller must complete @rq before @rq_src.
2576 int blk_rq_prep_clone(struct request
*rq
, struct request
*rq_src
,
2577 struct bio_set
*bs
, gfp_t gfp_mask
,
2578 int (*bio_ctr
)(struct bio
*, struct bio
*, void *),
2581 struct bio
*bio
, *bio_src
;
2586 blk_rq_init(NULL
, rq
);
2588 __rq_for_each_bio(bio_src
, rq_src
) {
2589 bio
= bio_alloc_bioset(gfp_mask
, bio_src
->bi_max_vecs
, bs
);
2593 __bio_clone(bio
, bio_src
);
2595 if (bio_integrity(bio_src
) &&
2596 bio_integrity_clone(bio
, bio_src
, gfp_mask
, bs
))
2599 if (bio_ctr
&& bio_ctr(bio
, bio_src
, data
))
2603 rq
->biotail
->bi_next
= bio
;
2606 rq
->bio
= rq
->biotail
= bio
;
2609 __blk_rq_prep_clone(rq
, rq_src
);
2616 blk_rq_unprep_clone(rq
);
2620 EXPORT_SYMBOL_GPL(blk_rq_prep_clone
);
2622 int kblockd_schedule_work(struct request_queue
*q
, struct work_struct
*work
)
2624 return queue_work(kblockd_workqueue
, work
);
2626 EXPORT_SYMBOL(kblockd_schedule_work
);
2628 int kblockd_schedule_delayed_work(struct request_queue
*q
,
2629 struct delayed_work
*dwork
, unsigned long delay
)
2631 return queue_delayed_work(kblockd_workqueue
, dwork
, delay
);
2633 EXPORT_SYMBOL(kblockd_schedule_delayed_work
);
2635 #define PLUG_MAGIC 0x91827364
2637 void blk_start_plug(struct blk_plug
*plug
)
2639 struct task_struct
*tsk
= current
;
2641 plug
->magic
= PLUG_MAGIC
;
2642 INIT_LIST_HEAD(&plug
->list
);
2643 INIT_LIST_HEAD(&plug
->cb_list
);
2644 plug
->should_sort
= 0;
2647 * If this is a nested plug, don't actually assign it. It will be
2648 * flushed on its own.
2652 * Store ordering should not be needed here, since a potential
2653 * preempt will imply a full memory barrier
2658 EXPORT_SYMBOL(blk_start_plug
);
2660 static int plug_rq_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
2662 struct request
*rqa
= container_of(a
, struct request
, queuelist
);
2663 struct request
*rqb
= container_of(b
, struct request
, queuelist
);
2665 return !(rqa
->q
<= rqb
->q
);
2669 * If 'from_schedule' is true, then postpone the dispatch of requests
2670 * until a safe kblockd context. We due this to avoid accidental big
2671 * additional stack usage in driver dispatch, in places where the originally
2672 * plugger did not intend it.
2674 static void queue_unplugged(struct request_queue
*q
, unsigned int depth
,
2676 __releases(q
->queue_lock
)
2678 trace_block_unplug(q
, depth
, !from_schedule
);
2681 * If we are punting this to kblockd, then we can safely drop
2682 * the queue_lock before waking kblockd (which needs to take
2685 if (from_schedule
) {
2686 spin_unlock(q
->queue_lock
);
2687 __blk_run_queue(q
, true);
2689 __blk_run_queue(q
, false);
2690 spin_unlock(q
->queue_lock
);
2695 static void flush_plug_callbacks(struct blk_plug
*plug
)
2697 LIST_HEAD(callbacks
);
2699 if (list_empty(&plug
->cb_list
))
2702 list_splice_init(&plug
->cb_list
, &callbacks
);
2704 while (!list_empty(&callbacks
)) {
2705 struct blk_plug_cb
*cb
= list_first_entry(&callbacks
,
2708 list_del(&cb
->list
);
2713 void blk_flush_plug_list(struct blk_plug
*plug
, bool from_schedule
)
2715 struct request_queue
*q
;
2716 unsigned long flags
;
2721 BUG_ON(plug
->magic
!= PLUG_MAGIC
);
2723 flush_plug_callbacks(plug
);
2724 if (list_empty(&plug
->list
))
2727 list_splice_init(&plug
->list
, &list
);
2729 if (plug
->should_sort
) {
2730 list_sort(NULL
, &list
, plug_rq_cmp
);
2731 plug
->should_sort
= 0;
2738 * Save and disable interrupts here, to avoid doing it for every
2739 * queue lock we have to take.
2741 local_irq_save(flags
);
2742 while (!list_empty(&list
)) {
2743 rq
= list_entry_rq(list
.next
);
2744 list_del_init(&rq
->queuelist
);
2745 BUG_ON(!(rq
->cmd_flags
& REQ_ON_PLUG
));
2749 * This drops the queue lock
2752 queue_unplugged(q
, depth
, from_schedule
);
2755 spin_lock(q
->queue_lock
);
2757 rq
->cmd_flags
&= ~REQ_ON_PLUG
;
2760 * rq is already accounted, so use raw insert
2762 if (rq
->cmd_flags
& (REQ_FLUSH
| REQ_FUA
))
2763 __elv_add_request(q
, rq
, ELEVATOR_INSERT_FLUSH
);
2765 __elv_add_request(q
, rq
, ELEVATOR_INSERT_SORT_MERGE
);
2771 * This drops the queue lock
2774 queue_unplugged(q
, depth
, from_schedule
);
2776 local_irq_restore(flags
);
2778 EXPORT_SYMBOL(blk_flush_plug_list
);
2780 void blk_finish_plug(struct blk_plug
*plug
)
2782 blk_flush_plug_list(plug
, false);
2784 if (plug
== current
->plug
)
2785 current
->plug
= NULL
;
2787 EXPORT_SYMBOL(blk_finish_plug
);
2789 int __init
blk_dev_init(void)
2791 BUILD_BUG_ON(__REQ_NR_BITS
> 8 *
2792 sizeof(((struct request
*)0)->cmd_flags
));
2794 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
2795 kblockd_workqueue
= alloc_workqueue("kblockd",
2796 WQ_MEM_RECLAIM
| WQ_HIGHPRI
, 0);
2797 if (!kblockd_workqueue
)
2798 panic("Failed to create kblockd\n");
2800 request_cachep
= kmem_cache_create("blkdev_requests",
2801 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
2803 blk_requestq_cachep
= kmem_cache_create("blkdev_queue",
2804 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);