Merge branch 'next' of git://git.monstr.eu/linux-2.6-microblaze
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / block / blk-core.c
blobe6c05a97ee2ba94538222d76273d0d2fbae644cc
1 /*
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>
7 * - July2000
8 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
9 */
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>
20 #include <linux/mm.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>
36 #include "blk.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);
63 int cpu;
65 if (!blk_do_io_stat(rq))
66 return;
68 cpu = part_stat_lock();
70 if (!new_io) {
71 part = rq->part;
72 part_stat_inc(cpu, part, merges[rw]);
73 } else {
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;
85 hd_struct_get(part);
87 part_round_stats(cpu, part);
88 part_inc_in_flight(part, rw);
89 rq->part = part;
92 part_stat_unlock();
95 void blk_queue_congestion_threshold(struct request_queue *q)
97 int nr;
99 nr = q->nr_requests - (q->nr_requests / 8) + 1;
100 if (nr > q->nr_requests)
101 nr = q->nr_requests;
102 q->nr_congestion_on = nr;
104 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
105 if (nr < 1)
106 nr = 1;
107 q->nr_congestion_off = nr;
111 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
112 * @bdev: device
114 * Locates the passed device's request queue and returns the address of its
115 * backing_dev_info
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);
124 if (q)
125 ret = &q->backing_dev_info;
126 return ret;
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);
136 rq->cpu = -1;
137 rq->q = q;
138 rq->__sector = (sector_t) -1;
139 INIT_HLIST_NODE(&rq->hash);
140 RB_CLEAR_NODE(&rq->rb_node);
141 rq->cmd = rq->__cmd;
142 rq->cmd_len = BLK_MAX_CDB;
143 rq->tag = -1;
144 rq->ref_count = 1;
145 rq->start_time = jiffies;
146 set_start_time_ns(rq);
147 rq->part = NULL;
149 EXPORT_SYMBOL(blk_rq_init);
151 static void req_bio_endio(struct request *rq, struct bio *bio,
152 unsigned int nbytes, int error)
154 if (error)
155 clear_bit(BIO_UPTODATE, &bio->bi_flags);
156 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
157 error = -EIO;
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)
181 int bit;
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,
185 rq->cmd_flags);
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]);
197 printk("\n");
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);
208 __blk_run_queue(q);
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
217 * Description:
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
233 * Description:
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);
243 __blk_run_queue(q);
245 EXPORT_SYMBOL(blk_start_queue);
248 * blk_stop_queue - stop a queue
249 * @q: The &struct request_queue in question
251 * Description:
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
270 * @q: the queue
272 * Description:
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
279 * this function.
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
297 * Description:
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)))
304 return;
306 q->request_fn(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
314 * Description:
315 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
316 * of us.
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
331 * Description:
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)
337 unsigned long flags;
339 spin_lock_irqsave(q->queue_lock, flags);
340 __blk_run_queue(q);
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
353 * @q: queue to drain
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)
362 while (true) {
363 bool drain = false;
364 int i;
366 spin_lock_irq(q->queue_lock);
368 elv_drain_elevator(q);
369 if (drain_all)
370 blk_throtl_drain(q);
373 * This function might be called on a queue which failed
374 * driver init after queue creation. Some drivers
375 * (e.g. fd) get unhappy in such cases. Kick queue iff
376 * dispatch queue has something on it.
378 if (!list_empty(&q->queue_head))
379 __blk_run_queue(q);
381 drain |= q->rq.elvpriv;
384 * Unfortunately, requests are queued at and tracked from
385 * multiple places and there's no single counter which can
386 * be drained. Check all the queues and counters.
388 if (drain_all) {
389 drain |= !list_empty(&q->queue_head);
390 for (i = 0; i < 2; i++) {
391 drain |= q->rq.count[i];
392 drain |= q->in_flight[i];
393 drain |= !list_empty(&q->flush_queue[i]);
397 spin_unlock_irq(q->queue_lock);
399 if (!drain)
400 break;
401 msleep(10);
406 * blk_cleanup_queue - shutdown a request queue
407 * @q: request queue to shutdown
409 * Mark @q DEAD, drain all pending requests, destroy and put it. All
410 * future requests will be failed immediately with -ENODEV.
412 void blk_cleanup_queue(struct request_queue *q)
414 spinlock_t *lock = q->queue_lock;
416 /* mark @q DEAD, no new request or merges will be allowed afterwards */
417 mutex_lock(&q->sysfs_lock);
418 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
420 spin_lock_irq(lock);
421 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
422 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
423 queue_flag_set(QUEUE_FLAG_DEAD, q);
425 if (q->queue_lock != &q->__queue_lock)
426 q->queue_lock = &q->__queue_lock;
428 spin_unlock_irq(lock);
429 mutex_unlock(&q->sysfs_lock);
432 * Drain all requests queued before DEAD marking. The caller might
433 * be trying to tear down @q before its elevator is initialized, in
434 * which case we don't want to call into draining.
436 if (q->elevator)
437 blk_drain_queue(q, true);
439 /* @q won't process any more request, flush async actions */
440 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
441 blk_sync_queue(q);
443 /* @q is and will stay empty, shutdown and put */
444 blk_put_queue(q);
446 EXPORT_SYMBOL(blk_cleanup_queue);
448 static int blk_init_free_list(struct request_queue *q)
450 struct request_list *rl = &q->rq;
452 if (unlikely(rl->rq_pool))
453 return 0;
455 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
456 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
457 rl->elvpriv = 0;
458 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
459 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
461 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
462 mempool_free_slab, request_cachep, q->node);
464 if (!rl->rq_pool)
465 return -ENOMEM;
467 return 0;
470 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
472 return blk_alloc_queue_node(gfp_mask, -1);
474 EXPORT_SYMBOL(blk_alloc_queue);
476 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
478 struct request_queue *q;
479 int err;
481 q = kmem_cache_alloc_node(blk_requestq_cachep,
482 gfp_mask | __GFP_ZERO, node_id);
483 if (!q)
484 return NULL;
486 q->id = ida_simple_get(&blk_queue_ida, 0, 0, GFP_KERNEL);
487 if (q->id < 0)
488 goto fail_q;
490 q->backing_dev_info.ra_pages =
491 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
492 q->backing_dev_info.state = 0;
493 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
494 q->backing_dev_info.name = "block";
495 q->node = node_id;
497 err = bdi_init(&q->backing_dev_info);
498 if (err)
499 goto fail_id;
501 if (blk_throtl_init(q))
502 goto fail_id;
504 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
505 laptop_mode_timer_fn, (unsigned long) q);
506 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
507 INIT_LIST_HEAD(&q->timeout_list);
508 INIT_LIST_HEAD(&q->icq_list);
509 INIT_LIST_HEAD(&q->flush_queue[0]);
510 INIT_LIST_HEAD(&q->flush_queue[1]);
511 INIT_LIST_HEAD(&q->flush_data_in_flight);
512 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
514 kobject_init(&q->kobj, &blk_queue_ktype);
516 mutex_init(&q->sysfs_lock);
517 spin_lock_init(&q->__queue_lock);
520 * By default initialize queue_lock to internal lock and driver can
521 * override it later if need be.
523 q->queue_lock = &q->__queue_lock;
525 return q;
527 fail_id:
528 ida_simple_remove(&blk_queue_ida, q->id);
529 fail_q:
530 kmem_cache_free(blk_requestq_cachep, q);
531 return NULL;
533 EXPORT_SYMBOL(blk_alloc_queue_node);
536 * blk_init_queue - prepare a request queue for use with a block device
537 * @rfn: The function to be called to process requests that have been
538 * placed on the queue.
539 * @lock: Request queue spin lock
541 * Description:
542 * If a block device wishes to use the standard request handling procedures,
543 * which sorts requests and coalesces adjacent requests, then it must
544 * call blk_init_queue(). The function @rfn will be called when there
545 * are requests on the queue that need to be processed. If the device
546 * supports plugging, then @rfn may not be called immediately when requests
547 * are available on the queue, but may be called at some time later instead.
548 * Plugged queues are generally unplugged when a buffer belonging to one
549 * of the requests on the queue is needed, or due to memory pressure.
551 * @rfn is not required, or even expected, to remove all requests off the
552 * queue, but only as many as it can handle at a time. If it does leave
553 * requests on the queue, it is responsible for arranging that the requests
554 * get dealt with eventually.
556 * The queue spin lock must be held while manipulating the requests on the
557 * request queue; this lock will be taken also from interrupt context, so irq
558 * disabling is needed for it.
560 * Function returns a pointer to the initialized request queue, or %NULL if
561 * it didn't succeed.
563 * Note:
564 * blk_init_queue() must be paired with a blk_cleanup_queue() call
565 * when the block device is deactivated (such as at module unload).
568 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
570 return blk_init_queue_node(rfn, lock, -1);
572 EXPORT_SYMBOL(blk_init_queue);
574 struct request_queue *
575 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
577 struct request_queue *uninit_q, *q;
579 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
580 if (!uninit_q)
581 return NULL;
583 q = blk_init_allocated_queue(uninit_q, rfn, lock);
584 if (!q)
585 blk_cleanup_queue(uninit_q);
587 return q;
589 EXPORT_SYMBOL(blk_init_queue_node);
591 struct request_queue *
592 blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
593 spinlock_t *lock)
595 if (!q)
596 return NULL;
598 if (blk_init_free_list(q))
599 return NULL;
601 q->request_fn = rfn;
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 */
607 if (lock)
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;
618 * all done
620 if (!elevator_init(q, NULL)) {
621 blk_queue_congestion_threshold(q);
622 return q;
625 return NULL;
627 EXPORT_SYMBOL(blk_init_allocated_queue);
629 bool blk_get_queue(struct request_queue *q)
631 if (likely(!blk_queue_dead(q))) {
632 __blk_get_queue(q);
633 return true;
636 return false;
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 if (rq->elv.icq)
645 put_io_context(rq->elv.icq->ioc, q);
648 mempool_free(rq, q->rq.rq_pool);
651 static struct request *
652 blk_alloc_request(struct request_queue *q, struct io_cq *icq,
653 unsigned int flags, gfp_t gfp_mask)
655 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
657 if (!rq)
658 return NULL;
660 blk_rq_init(q, rq);
662 rq->cmd_flags = flags | REQ_ALLOCED;
664 if (flags & REQ_ELVPRIV) {
665 rq->elv.icq = icq;
666 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
667 mempool_free(rq, q->rq.rq_pool);
668 return NULL;
670 /* @rq->elv.icq holds on to io_context until @rq is freed */
671 if (icq)
672 get_io_context(icq->ioc);
675 return rq;
679 * ioc_batching returns true if the ioc is a valid batching request and
680 * should be given priority access to a request.
682 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
684 if (!ioc)
685 return 0;
688 * Make sure the process is able to allocate at least 1 request
689 * even if the batch times out, otherwise we could theoretically
690 * lose wakeups.
692 return ioc->nr_batch_requests == q->nr_batching ||
693 (ioc->nr_batch_requests > 0
694 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
698 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
699 * will cause the process to be a "batcher" on all queues in the system. This
700 * is the behaviour we want though - once it gets a wakeup it should be given
701 * a nice run.
703 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
705 if (!ioc || ioc_batching(q, ioc))
706 return;
708 ioc->nr_batch_requests = q->nr_batching;
709 ioc->last_waited = jiffies;
712 static void __freed_request(struct request_queue *q, int sync)
714 struct request_list *rl = &q->rq;
716 if (rl->count[sync] < queue_congestion_off_threshold(q))
717 blk_clear_queue_congested(q, sync);
719 if (rl->count[sync] + 1 <= q->nr_requests) {
720 if (waitqueue_active(&rl->wait[sync]))
721 wake_up(&rl->wait[sync]);
723 blk_clear_queue_full(q, sync);
728 * A request has just been released. Account for it, update the full and
729 * congestion status, wake up any waiters. Called under q->queue_lock.
731 static void freed_request(struct request_queue *q, unsigned int flags)
733 struct request_list *rl = &q->rq;
734 int sync = rw_is_sync(flags);
736 rl->count[sync]--;
737 if (flags & REQ_ELVPRIV)
738 rl->elvpriv--;
740 __freed_request(q, sync);
742 if (unlikely(rl->starved[sync ^ 1]))
743 __freed_request(q, sync ^ 1);
747 * Determine if elevator data should be initialized when allocating the
748 * request associated with @bio.
750 static bool blk_rq_should_init_elevator(struct bio *bio)
752 if (!bio)
753 return true;
756 * Flush requests do not use the elevator so skip initialization.
757 * This allows a request to share the flush and elevator data.
759 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
760 return false;
762 return true;
766 * get_request - get a free request
767 * @q: request_queue to allocate request from
768 * @rw_flags: RW and SYNC flags
769 * @bio: bio to allocate request for (can be %NULL)
770 * @gfp_mask: allocation mask
772 * Get a free request from @q. This function may fail under memory
773 * pressure or if @q is dead.
775 * Must be callled with @q->queue_lock held and,
776 * Returns %NULL on failure, with @q->queue_lock held.
777 * Returns !%NULL on success, with @q->queue_lock *not held*.
779 static struct request *get_request(struct request_queue *q, int rw_flags,
780 struct bio *bio, gfp_t gfp_mask)
782 struct request *rq = NULL;
783 struct request_list *rl = &q->rq;
784 struct elevator_type *et;
785 struct io_context *ioc;
786 struct io_cq *icq = NULL;
787 const bool is_sync = rw_is_sync(rw_flags) != 0;
788 bool retried = false;
789 int may_queue;
790 retry:
791 et = q->elevator->type;
792 ioc = current->io_context;
794 if (unlikely(blk_queue_dead(q)))
795 return NULL;
797 may_queue = elv_may_queue(q, rw_flags);
798 if (may_queue == ELV_MQUEUE_NO)
799 goto rq_starved;
801 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
802 if (rl->count[is_sync]+1 >= q->nr_requests) {
804 * We want ioc to record batching state. If it's
805 * not already there, creating a new one requires
806 * dropping queue_lock, which in turn requires
807 * retesting conditions to avoid queue hang.
809 if (!ioc && !retried) {
810 spin_unlock_irq(q->queue_lock);
811 create_io_context(current, gfp_mask, q->node);
812 spin_lock_irq(q->queue_lock);
813 retried = true;
814 goto retry;
818 * The queue will fill after this allocation, so set
819 * it as full, and mark this process as "batching".
820 * This process will be allowed to complete a batch of
821 * requests, others will be blocked.
823 if (!blk_queue_full(q, is_sync)) {
824 ioc_set_batching(q, ioc);
825 blk_set_queue_full(q, is_sync);
826 } else {
827 if (may_queue != ELV_MQUEUE_MUST
828 && !ioc_batching(q, ioc)) {
830 * The queue is full and the allocating
831 * process is not a "batcher", and not
832 * exempted by the IO scheduler
834 goto out;
838 blk_set_queue_congested(q, is_sync);
842 * Only allow batching queuers to allocate up to 50% over the defined
843 * limit of requests, otherwise we could have thousands of requests
844 * allocated with any setting of ->nr_requests
846 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
847 goto out;
849 rl->count[is_sync]++;
850 rl->starved[is_sync] = 0;
853 * Decide whether the new request will be managed by elevator. If
854 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
855 * prevent the current elevator from being destroyed until the new
856 * request is freed. This guarantees icq's won't be destroyed and
857 * makes creating new ones safe.
859 * Also, lookup icq while holding queue_lock. If it doesn't exist,
860 * it will be created after releasing queue_lock.
862 if (blk_rq_should_init_elevator(bio) &&
863 !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags)) {
864 rw_flags |= REQ_ELVPRIV;
865 rl->elvpriv++;
866 if (et->icq_cache && ioc)
867 icq = ioc_lookup_icq(ioc, q);
870 if (blk_queue_io_stat(q))
871 rw_flags |= REQ_IO_STAT;
872 spin_unlock_irq(q->queue_lock);
874 /* create icq if missing */
875 if (unlikely(et->icq_cache && !icq))
876 icq = ioc_create_icq(q, gfp_mask);
878 /* rqs are guaranteed to have icq on elv_set_request() if requested */
879 if (likely(!et->icq_cache || icq))
880 rq = blk_alloc_request(q, icq, rw_flags, gfp_mask);
882 if (unlikely(!rq)) {
884 * Allocation failed presumably due to memory. Undo anything
885 * we might have messed up.
887 * Allocating task should really be put onto the front of the
888 * wait queue, but this is pretty rare.
890 spin_lock_irq(q->queue_lock);
891 freed_request(q, rw_flags);
894 * in the very unlikely event that allocation failed and no
895 * requests for this direction was pending, mark us starved
896 * so that freeing of a request in the other direction will
897 * notice us. another possible fix would be to split the
898 * rq mempool into READ and WRITE
900 rq_starved:
901 if (unlikely(rl->count[is_sync] == 0))
902 rl->starved[is_sync] = 1;
904 goto out;
908 * ioc may be NULL here, and ioc_batching will be false. That's
909 * OK, if the queue is under the request limit then requests need
910 * not count toward the nr_batch_requests limit. There will always
911 * be some limit enforced by BLK_BATCH_TIME.
913 if (ioc_batching(q, ioc))
914 ioc->nr_batch_requests--;
916 trace_block_getrq(q, bio, rw_flags & 1);
917 out:
918 return rq;
922 * get_request_wait - get a free request with retry
923 * @q: request_queue to allocate request from
924 * @rw_flags: RW and SYNC flags
925 * @bio: bio to allocate request for (can be %NULL)
927 * Get a free request from @q. This function keeps retrying under memory
928 * pressure and fails iff @q is dead.
930 * Must be callled with @q->queue_lock held and,
931 * Returns %NULL on failure, with @q->queue_lock held.
932 * Returns !%NULL on success, with @q->queue_lock *not held*.
934 static struct request *get_request_wait(struct request_queue *q, int rw_flags,
935 struct bio *bio)
937 const bool is_sync = rw_is_sync(rw_flags) != 0;
938 struct request *rq;
940 rq = get_request(q, rw_flags, bio, GFP_NOIO);
941 while (!rq) {
942 DEFINE_WAIT(wait);
943 struct request_list *rl = &q->rq;
945 if (unlikely(blk_queue_dead(q)))
946 return NULL;
948 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
949 TASK_UNINTERRUPTIBLE);
951 trace_block_sleeprq(q, bio, rw_flags & 1);
953 spin_unlock_irq(q->queue_lock);
954 io_schedule();
957 * After sleeping, we become a "batching" process and
958 * will be able to allocate at least one request, and
959 * up to a big batch of them for a small period time.
960 * See ioc_batching, ioc_set_batching
962 create_io_context(current, GFP_NOIO, q->node);
963 ioc_set_batching(q, current->io_context);
965 spin_lock_irq(q->queue_lock);
966 finish_wait(&rl->wait[is_sync], &wait);
968 rq = get_request(q, rw_flags, bio, GFP_NOIO);
971 return rq;
974 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
976 struct request *rq;
978 BUG_ON(rw != READ && rw != WRITE);
980 spin_lock_irq(q->queue_lock);
981 if (gfp_mask & __GFP_WAIT)
982 rq = get_request_wait(q, rw, NULL);
983 else
984 rq = get_request(q, rw, NULL, gfp_mask);
985 if (!rq)
986 spin_unlock_irq(q->queue_lock);
987 /* q->queue_lock is unlocked at this point */
989 return rq;
991 EXPORT_SYMBOL(blk_get_request);
994 * blk_make_request - given a bio, allocate a corresponding struct request.
995 * @q: target request queue
996 * @bio: The bio describing the memory mappings that will be submitted for IO.
997 * It may be a chained-bio properly constructed by block/bio layer.
998 * @gfp_mask: gfp flags to be used for memory allocation
1000 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1001 * type commands. Where the struct request needs to be farther initialized by
1002 * the caller. It is passed a &struct bio, which describes the memory info of
1003 * the I/O transfer.
1005 * The caller of blk_make_request must make sure that bi_io_vec
1006 * are set to describe the memory buffers. That bio_data_dir() will return
1007 * the needed direction of the request. (And all bio's in the passed bio-chain
1008 * are properly set accordingly)
1010 * If called under none-sleepable conditions, mapped bio buffers must not
1011 * need bouncing, by calling the appropriate masked or flagged allocator,
1012 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1013 * BUG.
1015 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1016 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1017 * anything but the first bio in the chain. Otherwise you risk waiting for IO
1018 * completion of a bio that hasn't been submitted yet, thus resulting in a
1019 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1020 * of bio_alloc(), as that avoids the mempool deadlock.
1021 * If possible a big IO should be split into smaller parts when allocation
1022 * fails. Partial allocation should not be an error, or you risk a live-lock.
1024 struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1025 gfp_t gfp_mask)
1027 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1029 if (unlikely(!rq))
1030 return ERR_PTR(-ENOMEM);
1032 for_each_bio(bio) {
1033 struct bio *bounce_bio = bio;
1034 int ret;
1036 blk_queue_bounce(q, &bounce_bio);
1037 ret = blk_rq_append_bio(q, rq, bounce_bio);
1038 if (unlikely(ret)) {
1039 blk_put_request(rq);
1040 return ERR_PTR(ret);
1044 return rq;
1046 EXPORT_SYMBOL(blk_make_request);
1049 * blk_requeue_request - put a request back on queue
1050 * @q: request queue where request should be inserted
1051 * @rq: request to be inserted
1053 * Description:
1054 * Drivers often keep queueing requests until the hardware cannot accept
1055 * more, when that condition happens we need to put the request back
1056 * on the queue. Must be called with queue lock held.
1058 void blk_requeue_request(struct request_queue *q, struct request *rq)
1060 blk_delete_timer(rq);
1061 blk_clear_rq_complete(rq);
1062 trace_block_rq_requeue(q, rq);
1064 if (blk_rq_tagged(rq))
1065 blk_queue_end_tag(q, rq);
1067 BUG_ON(blk_queued_rq(rq));
1069 elv_requeue_request(q, rq);
1071 EXPORT_SYMBOL(blk_requeue_request);
1073 static void add_acct_request(struct request_queue *q, struct request *rq,
1074 int where)
1076 drive_stat_acct(rq, 1);
1077 __elv_add_request(q, rq, where);
1080 static void part_round_stats_single(int cpu, struct hd_struct *part,
1081 unsigned long now)
1083 if (now == part->stamp)
1084 return;
1086 if (part_in_flight(part)) {
1087 __part_stat_add(cpu, part, time_in_queue,
1088 part_in_flight(part) * (now - part->stamp));
1089 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1091 part->stamp = now;
1095 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1096 * @cpu: cpu number for stats access
1097 * @part: target partition
1099 * The average IO queue length and utilisation statistics are maintained
1100 * by observing the current state of the queue length and the amount of
1101 * time it has been in this state for.
1103 * Normally, that accounting is done on IO completion, but that can result
1104 * in more than a second's worth of IO being accounted for within any one
1105 * second, leading to >100% utilisation. To deal with that, we call this
1106 * function to do a round-off before returning the results when reading
1107 * /proc/diskstats. This accounts immediately for all queue usage up to
1108 * the current jiffies and restarts the counters again.
1110 void part_round_stats(int cpu, struct hd_struct *part)
1112 unsigned long now = jiffies;
1114 if (part->partno)
1115 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1116 part_round_stats_single(cpu, part, now);
1118 EXPORT_SYMBOL_GPL(part_round_stats);
1121 * queue lock must be held
1123 void __blk_put_request(struct request_queue *q, struct request *req)
1125 if (unlikely(!q))
1126 return;
1127 if (unlikely(--req->ref_count))
1128 return;
1130 elv_completed_request(q, req);
1132 /* this is a bio leak */
1133 WARN_ON(req->bio != NULL);
1136 * Request may not have originated from ll_rw_blk. if not,
1137 * it didn't come out of our reserved rq pools
1139 if (req->cmd_flags & REQ_ALLOCED) {
1140 unsigned int flags = req->cmd_flags;
1142 BUG_ON(!list_empty(&req->queuelist));
1143 BUG_ON(!hlist_unhashed(&req->hash));
1145 blk_free_request(q, req);
1146 freed_request(q, flags);
1149 EXPORT_SYMBOL_GPL(__blk_put_request);
1151 void blk_put_request(struct request *req)
1153 unsigned long flags;
1154 struct request_queue *q = req->q;
1156 spin_lock_irqsave(q->queue_lock, flags);
1157 __blk_put_request(q, req);
1158 spin_unlock_irqrestore(q->queue_lock, flags);
1160 EXPORT_SYMBOL(blk_put_request);
1163 * blk_add_request_payload - add a payload to a request
1164 * @rq: request to update
1165 * @page: page backing the payload
1166 * @len: length of the payload.
1168 * This allows to later add a payload to an already submitted request by
1169 * a block driver. The driver needs to take care of freeing the payload
1170 * itself.
1172 * Note that this is a quite horrible hack and nothing but handling of
1173 * discard requests should ever use it.
1175 void blk_add_request_payload(struct request *rq, struct page *page,
1176 unsigned int len)
1178 struct bio *bio = rq->bio;
1180 bio->bi_io_vec->bv_page = page;
1181 bio->bi_io_vec->bv_offset = 0;
1182 bio->bi_io_vec->bv_len = len;
1184 bio->bi_size = len;
1185 bio->bi_vcnt = 1;
1186 bio->bi_phys_segments = 1;
1188 rq->__data_len = rq->resid_len = len;
1189 rq->nr_phys_segments = 1;
1190 rq->buffer = bio_data(bio);
1192 EXPORT_SYMBOL_GPL(blk_add_request_payload);
1194 static bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1195 struct bio *bio)
1197 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1199 if (!ll_back_merge_fn(q, req, bio))
1200 return false;
1202 trace_block_bio_backmerge(q, bio);
1204 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1205 blk_rq_set_mixed_merge(req);
1207 req->biotail->bi_next = bio;
1208 req->biotail = bio;
1209 req->__data_len += bio->bi_size;
1210 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1212 drive_stat_acct(req, 0);
1213 elv_bio_merged(q, req, bio);
1214 return true;
1217 static bool bio_attempt_front_merge(struct request_queue *q,
1218 struct request *req, struct bio *bio)
1220 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1222 if (!ll_front_merge_fn(q, req, bio))
1223 return false;
1225 trace_block_bio_frontmerge(q, bio);
1227 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1228 blk_rq_set_mixed_merge(req);
1230 bio->bi_next = req->bio;
1231 req->bio = bio;
1234 * may not be valid. if the low level driver said
1235 * it didn't need a bounce buffer then it better
1236 * not touch req->buffer either...
1238 req->buffer = bio_data(bio);
1239 req->__sector = bio->bi_sector;
1240 req->__data_len += bio->bi_size;
1241 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1243 drive_stat_acct(req, 0);
1244 elv_bio_merged(q, req, bio);
1245 return true;
1249 * attempt_plug_merge - try to merge with %current's plugged list
1250 * @q: request_queue new bio is being queued at
1251 * @bio: new bio being queued
1252 * @request_count: out parameter for number of traversed plugged requests
1254 * Determine whether @bio being queued on @q can be merged with a request
1255 * on %current's plugged list. Returns %true if merge was successful,
1256 * otherwise %false.
1258 * This function is called without @q->queue_lock; however, elevator is
1259 * accessed iff there already are requests on the plugged list which in
1260 * turn guarantees validity of the elevator.
1262 * Note that, on successful merge, elevator operation
1263 * elevator_bio_merged_fn() will be called without queue lock. Elevator
1264 * must be ready for this.
1266 static bool attempt_plug_merge(struct request_queue *q, struct bio *bio,
1267 unsigned int *request_count)
1269 struct blk_plug *plug;
1270 struct request *rq;
1271 bool ret = false;
1273 plug = current->plug;
1274 if (!plug)
1275 goto out;
1276 *request_count = 0;
1278 list_for_each_entry_reverse(rq, &plug->list, queuelist) {
1279 int el_ret;
1281 (*request_count)++;
1283 if (rq->q != q)
1284 continue;
1286 el_ret = elv_try_merge(rq, bio);
1287 if (el_ret == ELEVATOR_BACK_MERGE) {
1288 ret = bio_attempt_back_merge(q, rq, bio);
1289 if (ret)
1290 break;
1291 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1292 ret = bio_attempt_front_merge(q, rq, bio);
1293 if (ret)
1294 break;
1297 out:
1298 return ret;
1301 void init_request_from_bio(struct request *req, struct bio *bio)
1303 req->cmd_type = REQ_TYPE_FS;
1305 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1306 if (bio->bi_rw & REQ_RAHEAD)
1307 req->cmd_flags |= REQ_FAILFAST_MASK;
1309 req->errors = 0;
1310 req->__sector = bio->bi_sector;
1311 req->ioprio = bio_prio(bio);
1312 blk_rq_bio_prep(req->q, req, bio);
1315 void blk_queue_bio(struct request_queue *q, struct bio *bio)
1317 const bool sync = !!(bio->bi_rw & REQ_SYNC);
1318 struct blk_plug *plug;
1319 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1320 struct request *req;
1321 unsigned int request_count = 0;
1324 * low level driver can indicate that it wants pages above a
1325 * certain limit bounced to low memory (ie for highmem, or even
1326 * ISA dma in theory)
1328 blk_queue_bounce(q, &bio);
1330 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
1331 spin_lock_irq(q->queue_lock);
1332 where = ELEVATOR_INSERT_FLUSH;
1333 goto get_rq;
1337 * Check if we can merge with the plugged list before grabbing
1338 * any locks.
1340 if (attempt_plug_merge(q, bio, &request_count))
1341 return;
1343 spin_lock_irq(q->queue_lock);
1345 el_ret = elv_merge(q, &req, bio);
1346 if (el_ret == ELEVATOR_BACK_MERGE) {
1347 if (bio_attempt_back_merge(q, req, bio)) {
1348 if (!attempt_back_merge(q, req))
1349 elv_merged_request(q, req, el_ret);
1350 goto out_unlock;
1352 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1353 if (bio_attempt_front_merge(q, req, bio)) {
1354 if (!attempt_front_merge(q, req))
1355 elv_merged_request(q, req, el_ret);
1356 goto out_unlock;
1360 get_rq:
1362 * This sync check and mask will be re-done in init_request_from_bio(),
1363 * but we need to set it earlier to expose the sync flag to the
1364 * rq allocator and io schedulers.
1366 rw_flags = bio_data_dir(bio);
1367 if (sync)
1368 rw_flags |= REQ_SYNC;
1371 * Grab a free request. This is might sleep but can not fail.
1372 * Returns with the queue unlocked.
1374 req = get_request_wait(q, rw_flags, bio);
1375 if (unlikely(!req)) {
1376 bio_endio(bio, -ENODEV); /* @q is dead */
1377 goto out_unlock;
1381 * After dropping the lock and possibly sleeping here, our request
1382 * may now be mergeable after it had proven unmergeable (above).
1383 * We don't worry about that case for efficiency. It won't happen
1384 * often, and the elevators are able to handle it.
1386 init_request_from_bio(req, bio);
1388 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
1389 req->cpu = raw_smp_processor_id();
1391 plug = current->plug;
1392 if (plug) {
1394 * If this is the first request added after a plug, fire
1395 * of a plug trace. If others have been added before, check
1396 * if we have multiple devices in this plug. If so, make a
1397 * note to sort the list before dispatch.
1399 if (list_empty(&plug->list))
1400 trace_block_plug(q);
1401 else {
1402 if (!plug->should_sort) {
1403 struct request *__rq;
1405 __rq = list_entry_rq(plug->list.prev);
1406 if (__rq->q != q)
1407 plug->should_sort = 1;
1409 if (request_count >= BLK_MAX_REQUEST_COUNT) {
1410 blk_flush_plug_list(plug, false);
1411 trace_block_plug(q);
1414 list_add_tail(&req->queuelist, &plug->list);
1415 drive_stat_acct(req, 1);
1416 } else {
1417 spin_lock_irq(q->queue_lock);
1418 add_acct_request(q, req, where);
1419 __blk_run_queue(q);
1420 out_unlock:
1421 spin_unlock_irq(q->queue_lock);
1424 EXPORT_SYMBOL_GPL(blk_queue_bio); /* for device mapper only */
1427 * If bio->bi_dev is a partition, remap the location
1429 static inline void blk_partition_remap(struct bio *bio)
1431 struct block_device *bdev = bio->bi_bdev;
1433 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1434 struct hd_struct *p = bdev->bd_part;
1436 bio->bi_sector += p->start_sect;
1437 bio->bi_bdev = bdev->bd_contains;
1439 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1440 bdev->bd_dev,
1441 bio->bi_sector - p->start_sect);
1445 static void handle_bad_sector(struct bio *bio)
1447 char b[BDEVNAME_SIZE];
1449 printk(KERN_INFO "attempt to access beyond end of device\n");
1450 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1451 bdevname(bio->bi_bdev, b),
1452 bio->bi_rw,
1453 (unsigned long long)bio->bi_sector + bio_sectors(bio),
1454 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1456 set_bit(BIO_EOF, &bio->bi_flags);
1459 #ifdef CONFIG_FAIL_MAKE_REQUEST
1461 static DECLARE_FAULT_ATTR(fail_make_request);
1463 static int __init setup_fail_make_request(char *str)
1465 return setup_fault_attr(&fail_make_request, str);
1467 __setup("fail_make_request=", setup_fail_make_request);
1469 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1471 return part->make_it_fail && should_fail(&fail_make_request, bytes);
1474 static int __init fail_make_request_debugfs(void)
1476 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1477 NULL, &fail_make_request);
1479 return IS_ERR(dir) ? PTR_ERR(dir) : 0;
1482 late_initcall(fail_make_request_debugfs);
1484 #else /* CONFIG_FAIL_MAKE_REQUEST */
1486 static inline bool should_fail_request(struct hd_struct *part,
1487 unsigned int bytes)
1489 return false;
1492 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1495 * Check whether this bio extends beyond the end of the device.
1497 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1499 sector_t maxsector;
1501 if (!nr_sectors)
1502 return 0;
1504 /* Test device or partition size, when known. */
1505 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1506 if (maxsector) {
1507 sector_t sector = bio->bi_sector;
1509 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1511 * This may well happen - the kernel calls bread()
1512 * without checking the size of the device, e.g., when
1513 * mounting a device.
1515 handle_bad_sector(bio);
1516 return 1;
1520 return 0;
1523 static noinline_for_stack bool
1524 generic_make_request_checks(struct bio *bio)
1526 struct request_queue *q;
1527 int nr_sectors = bio_sectors(bio);
1528 int err = -EIO;
1529 char b[BDEVNAME_SIZE];
1530 struct hd_struct *part;
1532 might_sleep();
1534 if (bio_check_eod(bio, nr_sectors))
1535 goto end_io;
1537 q = bdev_get_queue(bio->bi_bdev);
1538 if (unlikely(!q)) {
1539 printk(KERN_ERR
1540 "generic_make_request: Trying to access "
1541 "nonexistent block-device %s (%Lu)\n",
1542 bdevname(bio->bi_bdev, b),
1543 (long long) bio->bi_sector);
1544 goto end_io;
1547 if (unlikely(!(bio->bi_rw & REQ_DISCARD) &&
1548 nr_sectors > queue_max_hw_sectors(q))) {
1549 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1550 bdevname(bio->bi_bdev, b),
1551 bio_sectors(bio),
1552 queue_max_hw_sectors(q));
1553 goto end_io;
1556 part = bio->bi_bdev->bd_part;
1557 if (should_fail_request(part, bio->bi_size) ||
1558 should_fail_request(&part_to_disk(part)->part0,
1559 bio->bi_size))
1560 goto end_io;
1563 * If this device has partitions, remap block n
1564 * of partition p to block n+start(p) of the disk.
1566 blk_partition_remap(bio);
1568 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1569 goto end_io;
1571 if (bio_check_eod(bio, nr_sectors))
1572 goto end_io;
1575 * Filter flush bio's early so that make_request based
1576 * drivers without flush support don't have to worry
1577 * about them.
1579 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1580 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1581 if (!nr_sectors) {
1582 err = 0;
1583 goto end_io;
1587 if ((bio->bi_rw & REQ_DISCARD) &&
1588 (!blk_queue_discard(q) ||
1589 ((bio->bi_rw & REQ_SECURE) &&
1590 !blk_queue_secdiscard(q)))) {
1591 err = -EOPNOTSUPP;
1592 goto end_io;
1595 if (blk_throtl_bio(q, bio))
1596 return false; /* throttled, will be resubmitted later */
1598 trace_block_bio_queue(q, bio);
1599 return true;
1601 end_io:
1602 bio_endio(bio, err);
1603 return false;
1607 * generic_make_request - hand a buffer to its device driver for I/O
1608 * @bio: The bio describing the location in memory and on the device.
1610 * generic_make_request() is used to make I/O requests of block
1611 * devices. It is passed a &struct bio, which describes the I/O that needs
1612 * to be done.
1614 * generic_make_request() does not return any status. The
1615 * success/failure status of the request, along with notification of
1616 * completion, is delivered asynchronously through the bio->bi_end_io
1617 * function described (one day) else where.
1619 * The caller of generic_make_request must make sure that bi_io_vec
1620 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1621 * set to describe the device address, and the
1622 * bi_end_io and optionally bi_private are set to describe how
1623 * completion notification should be signaled.
1625 * generic_make_request and the drivers it calls may use bi_next if this
1626 * bio happens to be merged with someone else, and may resubmit the bio to
1627 * a lower device by calling into generic_make_request recursively, which
1628 * means the bio should NOT be touched after the call to ->make_request_fn.
1630 void generic_make_request(struct bio *bio)
1632 struct bio_list bio_list_on_stack;
1634 if (!generic_make_request_checks(bio))
1635 return;
1638 * We only want one ->make_request_fn to be active at a time, else
1639 * stack usage with stacked devices could be a problem. So use
1640 * current->bio_list to keep a list of requests submited by a
1641 * make_request_fn function. current->bio_list is also used as a
1642 * flag to say if generic_make_request is currently active in this
1643 * task or not. If it is NULL, then no make_request is active. If
1644 * it is non-NULL, then a make_request is active, and new requests
1645 * should be added at the tail
1647 if (current->bio_list) {
1648 bio_list_add(current->bio_list, bio);
1649 return;
1652 /* following loop may be a bit non-obvious, and so deserves some
1653 * explanation.
1654 * Before entering the loop, bio->bi_next is NULL (as all callers
1655 * ensure that) so we have a list with a single bio.
1656 * We pretend that we have just taken it off a longer list, so
1657 * we assign bio_list to a pointer to the bio_list_on_stack,
1658 * thus initialising the bio_list of new bios to be
1659 * added. ->make_request() may indeed add some more bios
1660 * through a recursive call to generic_make_request. If it
1661 * did, we find a non-NULL value in bio_list and re-enter the loop
1662 * from the top. In this case we really did just take the bio
1663 * of the top of the list (no pretending) and so remove it from
1664 * bio_list, and call into ->make_request() again.
1666 BUG_ON(bio->bi_next);
1667 bio_list_init(&bio_list_on_stack);
1668 current->bio_list = &bio_list_on_stack;
1669 do {
1670 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1672 q->make_request_fn(q, bio);
1674 bio = bio_list_pop(current->bio_list);
1675 } while (bio);
1676 current->bio_list = NULL; /* deactivate */
1678 EXPORT_SYMBOL(generic_make_request);
1681 * submit_bio - submit a bio to the block device layer for I/O
1682 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1683 * @bio: The &struct bio which describes the I/O
1685 * submit_bio() is very similar in purpose to generic_make_request(), and
1686 * uses that function to do most of the work. Both are fairly rough
1687 * interfaces; @bio must be presetup and ready for I/O.
1690 void submit_bio(int rw, struct bio *bio)
1692 int count = bio_sectors(bio);
1694 bio->bi_rw |= rw;
1697 * If it's a regular read/write or a barrier with data attached,
1698 * go through the normal accounting stuff before submission.
1700 if (bio_has_data(bio) && !(rw & REQ_DISCARD)) {
1701 if (rw & WRITE) {
1702 count_vm_events(PGPGOUT, count);
1703 } else {
1704 task_io_account_read(bio->bi_size);
1705 count_vm_events(PGPGIN, count);
1708 if (unlikely(block_dump)) {
1709 char b[BDEVNAME_SIZE];
1710 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1711 current->comm, task_pid_nr(current),
1712 (rw & WRITE) ? "WRITE" : "READ",
1713 (unsigned long long)bio->bi_sector,
1714 bdevname(bio->bi_bdev, b),
1715 count);
1719 generic_make_request(bio);
1721 EXPORT_SYMBOL(submit_bio);
1724 * blk_rq_check_limits - Helper function to check a request for the queue limit
1725 * @q: the queue
1726 * @rq: the request being checked
1728 * Description:
1729 * @rq may have been made based on weaker limitations of upper-level queues
1730 * in request stacking drivers, and it may violate the limitation of @q.
1731 * Since the block layer and the underlying device driver trust @rq
1732 * after it is inserted to @q, it should be checked against @q before
1733 * the insertion using this generic function.
1735 * This function should also be useful for request stacking drivers
1736 * in some cases below, so export this function.
1737 * Request stacking drivers like request-based dm may change the queue
1738 * limits while requests are in the queue (e.g. dm's table swapping).
1739 * Such request stacking drivers should check those requests agaist
1740 * the new queue limits again when they dispatch those requests,
1741 * although such checkings are also done against the old queue limits
1742 * when submitting requests.
1744 int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1746 if (rq->cmd_flags & REQ_DISCARD)
1747 return 0;
1749 if (blk_rq_sectors(rq) > queue_max_sectors(q) ||
1750 blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) {
1751 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1752 return -EIO;
1756 * queue's settings related to segment counting like q->bounce_pfn
1757 * may differ from that of other stacking queues.
1758 * Recalculate it to check the request correctly on this queue's
1759 * limitation.
1761 blk_recalc_rq_segments(rq);
1762 if (rq->nr_phys_segments > queue_max_segments(q)) {
1763 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1764 return -EIO;
1767 return 0;
1769 EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1772 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1773 * @q: the queue to submit the request
1774 * @rq: the request being queued
1776 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1778 unsigned long flags;
1779 int where = ELEVATOR_INSERT_BACK;
1781 if (blk_rq_check_limits(q, rq))
1782 return -EIO;
1784 if (rq->rq_disk &&
1785 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
1786 return -EIO;
1788 spin_lock_irqsave(q->queue_lock, flags);
1789 if (unlikely(blk_queue_dead(q))) {
1790 spin_unlock_irqrestore(q->queue_lock, flags);
1791 return -ENODEV;
1795 * Submitting request must be dequeued before calling this function
1796 * because it will be linked to another request_queue
1798 BUG_ON(blk_queued_rq(rq));
1800 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
1801 where = ELEVATOR_INSERT_FLUSH;
1803 add_acct_request(q, rq, where);
1804 if (where == ELEVATOR_INSERT_FLUSH)
1805 __blk_run_queue(q);
1806 spin_unlock_irqrestore(q->queue_lock, flags);
1808 return 0;
1810 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1813 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1814 * @rq: request to examine
1816 * Description:
1817 * A request could be merge of IOs which require different failure
1818 * handling. This function determines the number of bytes which
1819 * can be failed from the beginning of the request without
1820 * crossing into area which need to be retried further.
1822 * Return:
1823 * The number of bytes to fail.
1825 * Context:
1826 * queue_lock must be held.
1828 unsigned int blk_rq_err_bytes(const struct request *rq)
1830 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1831 unsigned int bytes = 0;
1832 struct bio *bio;
1834 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1835 return blk_rq_bytes(rq);
1838 * Currently the only 'mixing' which can happen is between
1839 * different fastfail types. We can safely fail portions
1840 * which have all the failfast bits that the first one has -
1841 * the ones which are at least as eager to fail as the first
1842 * one.
1844 for (bio = rq->bio; bio; bio = bio->bi_next) {
1845 if ((bio->bi_rw & ff) != ff)
1846 break;
1847 bytes += bio->bi_size;
1850 /* this could lead to infinite loop */
1851 BUG_ON(blk_rq_bytes(rq) && !bytes);
1852 return bytes;
1854 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1856 static void blk_account_io_completion(struct request *req, unsigned int bytes)
1858 if (blk_do_io_stat(req)) {
1859 const int rw = rq_data_dir(req);
1860 struct hd_struct *part;
1861 int cpu;
1863 cpu = part_stat_lock();
1864 part = req->part;
1865 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1866 part_stat_unlock();
1870 static void blk_account_io_done(struct request *req)
1873 * Account IO completion. flush_rq isn't accounted as a
1874 * normal IO on queueing nor completion. Accounting the
1875 * containing request is enough.
1877 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
1878 unsigned long duration = jiffies - req->start_time;
1879 const int rw = rq_data_dir(req);
1880 struct hd_struct *part;
1881 int cpu;
1883 cpu = part_stat_lock();
1884 part = req->part;
1886 part_stat_inc(cpu, part, ios[rw]);
1887 part_stat_add(cpu, part, ticks[rw], duration);
1888 part_round_stats(cpu, part);
1889 part_dec_in_flight(part, rw);
1891 hd_struct_put(part);
1892 part_stat_unlock();
1897 * blk_peek_request - peek at the top of a request queue
1898 * @q: request queue to peek at
1900 * Description:
1901 * Return the request at the top of @q. The returned request
1902 * should be started using blk_start_request() before LLD starts
1903 * processing it.
1905 * Return:
1906 * Pointer to the request at the top of @q if available. Null
1907 * otherwise.
1909 * Context:
1910 * queue_lock must be held.
1912 struct request *blk_peek_request(struct request_queue *q)
1914 struct request *rq;
1915 int ret;
1917 while ((rq = __elv_next_request(q)) != NULL) {
1918 if (!(rq->cmd_flags & REQ_STARTED)) {
1920 * This is the first time the device driver
1921 * sees this request (possibly after
1922 * requeueing). Notify IO scheduler.
1924 if (rq->cmd_flags & REQ_SORTED)
1925 elv_activate_rq(q, rq);
1928 * just mark as started even if we don't start
1929 * it, a request that has been delayed should
1930 * not be passed by new incoming requests
1932 rq->cmd_flags |= REQ_STARTED;
1933 trace_block_rq_issue(q, rq);
1936 if (!q->boundary_rq || q->boundary_rq == rq) {
1937 q->end_sector = rq_end_sector(rq);
1938 q->boundary_rq = NULL;
1941 if (rq->cmd_flags & REQ_DONTPREP)
1942 break;
1944 if (q->dma_drain_size && blk_rq_bytes(rq)) {
1946 * make sure space for the drain appears we
1947 * know we can do this because max_hw_segments
1948 * has been adjusted to be one fewer than the
1949 * device can handle
1951 rq->nr_phys_segments++;
1954 if (!q->prep_rq_fn)
1955 break;
1957 ret = q->prep_rq_fn(q, rq);
1958 if (ret == BLKPREP_OK) {
1959 break;
1960 } else if (ret == BLKPREP_DEFER) {
1962 * the request may have been (partially) prepped.
1963 * we need to keep this request in the front to
1964 * avoid resource deadlock. REQ_STARTED will
1965 * prevent other fs requests from passing this one.
1967 if (q->dma_drain_size && blk_rq_bytes(rq) &&
1968 !(rq->cmd_flags & REQ_DONTPREP)) {
1970 * remove the space for the drain we added
1971 * so that we don't add it again
1973 --rq->nr_phys_segments;
1976 rq = NULL;
1977 break;
1978 } else if (ret == BLKPREP_KILL) {
1979 rq->cmd_flags |= REQ_QUIET;
1981 * Mark this request as started so we don't trigger
1982 * any debug logic in the end I/O path.
1984 blk_start_request(rq);
1985 __blk_end_request_all(rq, -EIO);
1986 } else {
1987 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
1988 break;
1992 return rq;
1994 EXPORT_SYMBOL(blk_peek_request);
1996 void blk_dequeue_request(struct request *rq)
1998 struct request_queue *q = rq->q;
2000 BUG_ON(list_empty(&rq->queuelist));
2001 BUG_ON(ELV_ON_HASH(rq));
2003 list_del_init(&rq->queuelist);
2006 * the time frame between a request being removed from the lists
2007 * and to it is freed is accounted as io that is in progress at
2008 * the driver side.
2010 if (blk_account_rq(rq)) {
2011 q->in_flight[rq_is_sync(rq)]++;
2012 set_io_start_time_ns(rq);
2017 * blk_start_request - start request processing on the driver
2018 * @req: request to dequeue
2020 * Description:
2021 * Dequeue @req and start timeout timer on it. This hands off the
2022 * request to the driver.
2024 * Block internal functions which don't want to start timer should
2025 * call blk_dequeue_request().
2027 * Context:
2028 * queue_lock must be held.
2030 void blk_start_request(struct request *req)
2032 blk_dequeue_request(req);
2035 * We are now handing the request to the hardware, initialize
2036 * resid_len to full count and add the timeout handler.
2038 req->resid_len = blk_rq_bytes(req);
2039 if (unlikely(blk_bidi_rq(req)))
2040 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2042 blk_add_timer(req);
2044 EXPORT_SYMBOL(blk_start_request);
2047 * blk_fetch_request - fetch a request from a request queue
2048 * @q: request queue to fetch a request from
2050 * Description:
2051 * Return the request at the top of @q. The request is started on
2052 * return and LLD can start processing it immediately.
2054 * Return:
2055 * Pointer to the request at the top of @q if available. Null
2056 * otherwise.
2058 * Context:
2059 * queue_lock must be held.
2061 struct request *blk_fetch_request(struct request_queue *q)
2063 struct request *rq;
2065 rq = blk_peek_request(q);
2066 if (rq)
2067 blk_start_request(rq);
2068 return rq;
2070 EXPORT_SYMBOL(blk_fetch_request);
2073 * blk_update_request - Special helper function for request stacking drivers
2074 * @req: the request being processed
2075 * @error: %0 for success, < %0 for error
2076 * @nr_bytes: number of bytes to complete @req
2078 * Description:
2079 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2080 * the request structure even if @req doesn't have leftover.
2081 * If @req has leftover, sets it up for the next range of segments.
2083 * This special helper function is only for request stacking drivers
2084 * (e.g. request-based dm) so that they can handle partial completion.
2085 * Actual device drivers should use blk_end_request instead.
2087 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2088 * %false return from this function.
2090 * Return:
2091 * %false - this request doesn't have any more data
2092 * %true - this request has more data
2094 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2096 int total_bytes, bio_nbytes, next_idx = 0;
2097 struct bio *bio;
2099 if (!req->bio)
2100 return false;
2102 trace_block_rq_complete(req->q, req);
2105 * For fs requests, rq is just carrier of independent bio's
2106 * and each partial completion should be handled separately.
2107 * Reset per-request error on each partial completion.
2109 * TODO: tj: This is too subtle. It would be better to let
2110 * low level drivers do what they see fit.
2112 if (req->cmd_type == REQ_TYPE_FS)
2113 req->errors = 0;
2115 if (error && req->cmd_type == REQ_TYPE_FS &&
2116 !(req->cmd_flags & REQ_QUIET)) {
2117 char *error_type;
2119 switch (error) {
2120 case -ENOLINK:
2121 error_type = "recoverable transport";
2122 break;
2123 case -EREMOTEIO:
2124 error_type = "critical target";
2125 break;
2126 case -EBADE:
2127 error_type = "critical nexus";
2128 break;
2129 case -EIO:
2130 default:
2131 error_type = "I/O";
2132 break;
2134 printk(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2135 error_type, req->rq_disk ? req->rq_disk->disk_name : "?",
2136 (unsigned long long)blk_rq_pos(req));
2139 blk_account_io_completion(req, nr_bytes);
2141 total_bytes = bio_nbytes = 0;
2142 while ((bio = req->bio) != NULL) {
2143 int nbytes;
2145 if (nr_bytes >= bio->bi_size) {
2146 req->bio = bio->bi_next;
2147 nbytes = bio->bi_size;
2148 req_bio_endio(req, bio, nbytes, error);
2149 next_idx = 0;
2150 bio_nbytes = 0;
2151 } else {
2152 int idx = bio->bi_idx + next_idx;
2154 if (unlikely(idx >= bio->bi_vcnt)) {
2155 blk_dump_rq_flags(req, "__end_that");
2156 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
2157 __func__, idx, bio->bi_vcnt);
2158 break;
2161 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2162 BIO_BUG_ON(nbytes > bio->bi_size);
2165 * not a complete bvec done
2167 if (unlikely(nbytes > nr_bytes)) {
2168 bio_nbytes += nr_bytes;
2169 total_bytes += nr_bytes;
2170 break;
2174 * advance to the next vector
2176 next_idx++;
2177 bio_nbytes += nbytes;
2180 total_bytes += nbytes;
2181 nr_bytes -= nbytes;
2183 bio = req->bio;
2184 if (bio) {
2186 * end more in this run, or just return 'not-done'
2188 if (unlikely(nr_bytes <= 0))
2189 break;
2194 * completely done
2196 if (!req->bio) {
2198 * Reset counters so that the request stacking driver
2199 * can find how many bytes remain in the request
2200 * later.
2202 req->__data_len = 0;
2203 return false;
2207 * if the request wasn't completed, update state
2209 if (bio_nbytes) {
2210 req_bio_endio(req, bio, bio_nbytes, error);
2211 bio->bi_idx += next_idx;
2212 bio_iovec(bio)->bv_offset += nr_bytes;
2213 bio_iovec(bio)->bv_len -= nr_bytes;
2216 req->__data_len -= total_bytes;
2217 req->buffer = bio_data(req->bio);
2219 /* update sector only for requests with clear definition of sector */
2220 if (req->cmd_type == REQ_TYPE_FS || (req->cmd_flags & REQ_DISCARD))
2221 req->__sector += total_bytes >> 9;
2223 /* mixed attributes always follow the first bio */
2224 if (req->cmd_flags & REQ_MIXED_MERGE) {
2225 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2226 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2230 * If total number of sectors is less than the first segment
2231 * size, something has gone terribly wrong.
2233 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2234 blk_dump_rq_flags(req, "request botched");
2235 req->__data_len = blk_rq_cur_bytes(req);
2238 /* recalculate the number of segments */
2239 blk_recalc_rq_segments(req);
2241 return true;
2243 EXPORT_SYMBOL_GPL(blk_update_request);
2245 static bool blk_update_bidi_request(struct request *rq, int error,
2246 unsigned int nr_bytes,
2247 unsigned int bidi_bytes)
2249 if (blk_update_request(rq, error, nr_bytes))
2250 return true;
2252 /* Bidi request must be completed as a whole */
2253 if (unlikely(blk_bidi_rq(rq)) &&
2254 blk_update_request(rq->next_rq, error, bidi_bytes))
2255 return true;
2257 if (blk_queue_add_random(rq->q))
2258 add_disk_randomness(rq->rq_disk);
2260 return false;
2264 * blk_unprep_request - unprepare a request
2265 * @req: the request
2267 * This function makes a request ready for complete resubmission (or
2268 * completion). It happens only after all error handling is complete,
2269 * so represents the appropriate moment to deallocate any resources
2270 * that were allocated to the request in the prep_rq_fn. The queue
2271 * lock is held when calling this.
2273 void blk_unprep_request(struct request *req)
2275 struct request_queue *q = req->q;
2277 req->cmd_flags &= ~REQ_DONTPREP;
2278 if (q->unprep_rq_fn)
2279 q->unprep_rq_fn(q, req);
2281 EXPORT_SYMBOL_GPL(blk_unprep_request);
2284 * queue lock must be held
2286 static void blk_finish_request(struct request *req, int error)
2288 if (blk_rq_tagged(req))
2289 blk_queue_end_tag(req->q, req);
2291 BUG_ON(blk_queued_rq(req));
2293 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2294 laptop_io_completion(&req->q->backing_dev_info);
2296 blk_delete_timer(req);
2298 if (req->cmd_flags & REQ_DONTPREP)
2299 blk_unprep_request(req);
2302 blk_account_io_done(req);
2304 if (req->end_io)
2305 req->end_io(req, error);
2306 else {
2307 if (blk_bidi_rq(req))
2308 __blk_put_request(req->next_rq->q, req->next_rq);
2310 __blk_put_request(req->q, req);
2315 * blk_end_bidi_request - Complete a bidi request
2316 * @rq: the request to complete
2317 * @error: %0 for success, < %0 for error
2318 * @nr_bytes: number of bytes to complete @rq
2319 * @bidi_bytes: number of bytes to complete @rq->next_rq
2321 * Description:
2322 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2323 * Drivers that supports bidi can safely call this member for any
2324 * type of request, bidi or uni. In the later case @bidi_bytes is
2325 * just ignored.
2327 * Return:
2328 * %false - we are done with this request
2329 * %true - still buffers pending for this request
2331 static bool blk_end_bidi_request(struct request *rq, int error,
2332 unsigned int nr_bytes, unsigned int bidi_bytes)
2334 struct request_queue *q = rq->q;
2335 unsigned long flags;
2337 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2338 return true;
2340 spin_lock_irqsave(q->queue_lock, flags);
2341 blk_finish_request(rq, error);
2342 spin_unlock_irqrestore(q->queue_lock, flags);
2344 return false;
2348 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2349 * @rq: the request to complete
2350 * @error: %0 for success, < %0 for error
2351 * @nr_bytes: number of bytes to complete @rq
2352 * @bidi_bytes: number of bytes to complete @rq->next_rq
2354 * Description:
2355 * Identical to blk_end_bidi_request() except that queue lock is
2356 * assumed to be locked on entry and remains so on return.
2358 * Return:
2359 * %false - we are done with this request
2360 * %true - still buffers pending for this request
2362 bool __blk_end_bidi_request(struct request *rq, int error,
2363 unsigned int nr_bytes, unsigned int bidi_bytes)
2365 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2366 return true;
2368 blk_finish_request(rq, error);
2370 return false;
2374 * blk_end_request - Helper function for drivers to complete the request.
2375 * @rq: the request being processed
2376 * @error: %0 for success, < %0 for error
2377 * @nr_bytes: number of bytes to complete
2379 * Description:
2380 * Ends I/O on a number of bytes attached to @rq.
2381 * If @rq has leftover, sets it up for the next range of segments.
2383 * Return:
2384 * %false - we are done with this request
2385 * %true - still buffers pending for this request
2387 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2389 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2391 EXPORT_SYMBOL(blk_end_request);
2394 * blk_end_request_all - Helper function for drives to finish the request.
2395 * @rq: the request to finish
2396 * @error: %0 for success, < %0 for error
2398 * Description:
2399 * Completely finish @rq.
2401 void blk_end_request_all(struct request *rq, int error)
2403 bool pending;
2404 unsigned int bidi_bytes = 0;
2406 if (unlikely(blk_bidi_rq(rq)))
2407 bidi_bytes = blk_rq_bytes(rq->next_rq);
2409 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2410 BUG_ON(pending);
2412 EXPORT_SYMBOL(blk_end_request_all);
2415 * blk_end_request_cur - Helper function to finish the current request chunk.
2416 * @rq: the request to finish the current chunk for
2417 * @error: %0 for success, < %0 for error
2419 * Description:
2420 * Complete the current consecutively mapped chunk from @rq.
2422 * Return:
2423 * %false - we are done with this request
2424 * %true - still buffers pending for this request
2426 bool blk_end_request_cur(struct request *rq, int error)
2428 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2430 EXPORT_SYMBOL(blk_end_request_cur);
2433 * blk_end_request_err - Finish a request till the next failure boundary.
2434 * @rq: the request to finish till the next failure boundary for
2435 * @error: must be negative errno
2437 * Description:
2438 * Complete @rq till the next failure boundary.
2440 * Return:
2441 * %false - we are done with this request
2442 * %true - still buffers pending for this request
2444 bool blk_end_request_err(struct request *rq, int error)
2446 WARN_ON(error >= 0);
2447 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2449 EXPORT_SYMBOL_GPL(blk_end_request_err);
2452 * __blk_end_request - Helper function for drivers to complete the request.
2453 * @rq: the request being processed
2454 * @error: %0 for success, < %0 for error
2455 * @nr_bytes: number of bytes to complete
2457 * Description:
2458 * Must be called with queue lock held unlike blk_end_request().
2460 * Return:
2461 * %false - we are done with this request
2462 * %true - still buffers pending for this request
2464 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2466 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2468 EXPORT_SYMBOL(__blk_end_request);
2471 * __blk_end_request_all - Helper function for drives to finish the request.
2472 * @rq: the request to finish
2473 * @error: %0 for success, < %0 for error
2475 * Description:
2476 * Completely finish @rq. Must be called with queue lock held.
2478 void __blk_end_request_all(struct request *rq, int error)
2480 bool pending;
2481 unsigned int bidi_bytes = 0;
2483 if (unlikely(blk_bidi_rq(rq)))
2484 bidi_bytes = blk_rq_bytes(rq->next_rq);
2486 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2487 BUG_ON(pending);
2489 EXPORT_SYMBOL(__blk_end_request_all);
2492 * __blk_end_request_cur - Helper function to finish the current request chunk.
2493 * @rq: the request to finish the current chunk for
2494 * @error: %0 for success, < %0 for error
2496 * Description:
2497 * Complete the current consecutively mapped chunk from @rq. Must
2498 * be called with queue lock held.
2500 * Return:
2501 * %false - we are done with this request
2502 * %true - still buffers pending for this request
2504 bool __blk_end_request_cur(struct request *rq, int error)
2506 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2508 EXPORT_SYMBOL(__blk_end_request_cur);
2511 * __blk_end_request_err - Finish a request till the next failure boundary.
2512 * @rq: the request to finish till the next failure boundary for
2513 * @error: must be negative errno
2515 * Description:
2516 * Complete @rq till the next failure boundary. Must be called
2517 * with queue lock held.
2519 * Return:
2520 * %false - we are done with this request
2521 * %true - still buffers pending for this request
2523 bool __blk_end_request_err(struct request *rq, int error)
2525 WARN_ON(error >= 0);
2526 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2528 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2530 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2531 struct bio *bio)
2533 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2534 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
2536 if (bio_has_data(bio)) {
2537 rq->nr_phys_segments = bio_phys_segments(q, bio);
2538 rq->buffer = bio_data(bio);
2540 rq->__data_len = bio->bi_size;
2541 rq->bio = rq->biotail = bio;
2543 if (bio->bi_bdev)
2544 rq->rq_disk = bio->bi_bdev->bd_disk;
2547 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2549 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2550 * @rq: the request to be flushed
2552 * Description:
2553 * Flush all pages in @rq.
2555 void rq_flush_dcache_pages(struct request *rq)
2557 struct req_iterator iter;
2558 struct bio_vec *bvec;
2560 rq_for_each_segment(bvec, rq, iter)
2561 flush_dcache_page(bvec->bv_page);
2563 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2564 #endif
2567 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2568 * @q : the queue of the device being checked
2570 * Description:
2571 * Check if underlying low-level drivers of a device are busy.
2572 * If the drivers want to export their busy state, they must set own
2573 * exporting function using blk_queue_lld_busy() first.
2575 * Basically, this function is used only by request stacking drivers
2576 * to stop dispatching requests to underlying devices when underlying
2577 * devices are busy. This behavior helps more I/O merging on the queue
2578 * of the request stacking driver and prevents I/O throughput regression
2579 * on burst I/O load.
2581 * Return:
2582 * 0 - Not busy (The request stacking driver should dispatch request)
2583 * 1 - Busy (The request stacking driver should stop dispatching request)
2585 int blk_lld_busy(struct request_queue *q)
2587 if (q->lld_busy_fn)
2588 return q->lld_busy_fn(q);
2590 return 0;
2592 EXPORT_SYMBOL_GPL(blk_lld_busy);
2595 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2596 * @rq: the clone request to be cleaned up
2598 * Description:
2599 * Free all bios in @rq for a cloned request.
2601 void blk_rq_unprep_clone(struct request *rq)
2603 struct bio *bio;
2605 while ((bio = rq->bio) != NULL) {
2606 rq->bio = bio->bi_next;
2608 bio_put(bio);
2611 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2614 * Copy attributes of the original request to the clone request.
2615 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2617 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2619 dst->cpu = src->cpu;
2620 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
2621 dst->cmd_type = src->cmd_type;
2622 dst->__sector = blk_rq_pos(src);
2623 dst->__data_len = blk_rq_bytes(src);
2624 dst->nr_phys_segments = src->nr_phys_segments;
2625 dst->ioprio = src->ioprio;
2626 dst->extra_len = src->extra_len;
2630 * blk_rq_prep_clone - Helper function to setup clone request
2631 * @rq: the request to be setup
2632 * @rq_src: original request to be cloned
2633 * @bs: bio_set that bios for clone are allocated from
2634 * @gfp_mask: memory allocation mask for bio
2635 * @bio_ctr: setup function to be called for each clone bio.
2636 * Returns %0 for success, non %0 for failure.
2637 * @data: private data to be passed to @bio_ctr
2639 * Description:
2640 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2641 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2642 * are not copied, and copying such parts is the caller's responsibility.
2643 * Also, pages which the original bios are pointing to are not copied
2644 * and the cloned bios just point same pages.
2645 * So cloned bios must be completed before original bios, which means
2646 * the caller must complete @rq before @rq_src.
2648 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2649 struct bio_set *bs, gfp_t gfp_mask,
2650 int (*bio_ctr)(struct bio *, struct bio *, void *),
2651 void *data)
2653 struct bio *bio, *bio_src;
2655 if (!bs)
2656 bs = fs_bio_set;
2658 blk_rq_init(NULL, rq);
2660 __rq_for_each_bio(bio_src, rq_src) {
2661 bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs);
2662 if (!bio)
2663 goto free_and_out;
2665 __bio_clone(bio, bio_src);
2667 if (bio_integrity(bio_src) &&
2668 bio_integrity_clone(bio, bio_src, gfp_mask, bs))
2669 goto free_and_out;
2671 if (bio_ctr && bio_ctr(bio, bio_src, data))
2672 goto free_and_out;
2674 if (rq->bio) {
2675 rq->biotail->bi_next = bio;
2676 rq->biotail = bio;
2677 } else
2678 rq->bio = rq->biotail = bio;
2681 __blk_rq_prep_clone(rq, rq_src);
2683 return 0;
2685 free_and_out:
2686 if (bio)
2687 bio_free(bio, bs);
2688 blk_rq_unprep_clone(rq);
2690 return -ENOMEM;
2692 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2694 int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
2696 return queue_work(kblockd_workqueue, work);
2698 EXPORT_SYMBOL(kblockd_schedule_work);
2700 int kblockd_schedule_delayed_work(struct request_queue *q,
2701 struct delayed_work *dwork, unsigned long delay)
2703 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2705 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2707 #define PLUG_MAGIC 0x91827364
2710 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2711 * @plug: The &struct blk_plug that needs to be initialized
2713 * Description:
2714 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2715 * pending I/O should the task end up blocking between blk_start_plug() and
2716 * blk_finish_plug(). This is important from a performance perspective, but
2717 * also ensures that we don't deadlock. For instance, if the task is blocking
2718 * for a memory allocation, memory reclaim could end up wanting to free a
2719 * page belonging to that request that is currently residing in our private
2720 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2721 * this kind of deadlock.
2723 void blk_start_plug(struct blk_plug *plug)
2725 struct task_struct *tsk = current;
2727 plug->magic = PLUG_MAGIC;
2728 INIT_LIST_HEAD(&plug->list);
2729 INIT_LIST_HEAD(&plug->cb_list);
2730 plug->should_sort = 0;
2733 * If this is a nested plug, don't actually assign it. It will be
2734 * flushed on its own.
2736 if (!tsk->plug) {
2738 * Store ordering should not be needed here, since a potential
2739 * preempt will imply a full memory barrier
2741 tsk->plug = plug;
2744 EXPORT_SYMBOL(blk_start_plug);
2746 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2748 struct request *rqa = container_of(a, struct request, queuelist);
2749 struct request *rqb = container_of(b, struct request, queuelist);
2751 return !(rqa->q <= rqb->q);
2755 * If 'from_schedule' is true, then postpone the dispatch of requests
2756 * until a safe kblockd context. We due this to avoid accidental big
2757 * additional stack usage in driver dispatch, in places where the originally
2758 * plugger did not intend it.
2760 static void queue_unplugged(struct request_queue *q, unsigned int depth,
2761 bool from_schedule)
2762 __releases(q->queue_lock)
2764 trace_block_unplug(q, depth, !from_schedule);
2767 * Don't mess with dead queue.
2769 if (unlikely(blk_queue_dead(q))) {
2770 spin_unlock(q->queue_lock);
2771 return;
2775 * If we are punting this to kblockd, then we can safely drop
2776 * the queue_lock before waking kblockd (which needs to take
2777 * this lock).
2779 if (from_schedule) {
2780 spin_unlock(q->queue_lock);
2781 blk_run_queue_async(q);
2782 } else {
2783 __blk_run_queue(q);
2784 spin_unlock(q->queue_lock);
2789 static void flush_plug_callbacks(struct blk_plug *plug)
2791 LIST_HEAD(callbacks);
2793 if (list_empty(&plug->cb_list))
2794 return;
2796 list_splice_init(&plug->cb_list, &callbacks);
2798 while (!list_empty(&callbacks)) {
2799 struct blk_plug_cb *cb = list_first_entry(&callbacks,
2800 struct blk_plug_cb,
2801 list);
2802 list_del(&cb->list);
2803 cb->callback(cb);
2807 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
2809 struct request_queue *q;
2810 unsigned long flags;
2811 struct request *rq;
2812 LIST_HEAD(list);
2813 unsigned int depth;
2815 BUG_ON(plug->magic != PLUG_MAGIC);
2817 flush_plug_callbacks(plug);
2818 if (list_empty(&plug->list))
2819 return;
2821 list_splice_init(&plug->list, &list);
2823 if (plug->should_sort) {
2824 list_sort(NULL, &list, plug_rq_cmp);
2825 plug->should_sort = 0;
2828 q = NULL;
2829 depth = 0;
2832 * Save and disable interrupts here, to avoid doing it for every
2833 * queue lock we have to take.
2835 local_irq_save(flags);
2836 while (!list_empty(&list)) {
2837 rq = list_entry_rq(list.next);
2838 list_del_init(&rq->queuelist);
2839 BUG_ON(!rq->q);
2840 if (rq->q != q) {
2842 * This drops the queue lock
2844 if (q)
2845 queue_unplugged(q, depth, from_schedule);
2846 q = rq->q;
2847 depth = 0;
2848 spin_lock(q->queue_lock);
2852 * Short-circuit if @q is dead
2854 if (unlikely(blk_queue_dead(q))) {
2855 __blk_end_request_all(rq, -ENODEV);
2856 continue;
2860 * rq is already accounted, so use raw insert
2862 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
2863 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
2864 else
2865 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
2867 depth++;
2871 * This drops the queue lock
2873 if (q)
2874 queue_unplugged(q, depth, from_schedule);
2876 local_irq_restore(flags);
2879 void blk_finish_plug(struct blk_plug *plug)
2881 blk_flush_plug_list(plug, false);
2883 if (plug == current->plug)
2884 current->plug = NULL;
2886 EXPORT_SYMBOL(blk_finish_plug);
2888 int __init blk_dev_init(void)
2890 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
2891 sizeof(((struct request *)0)->cmd_flags));
2893 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
2894 kblockd_workqueue = alloc_workqueue("kblockd",
2895 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
2896 if (!kblockd_workqueue)
2897 panic("Failed to create kblockd\n");
2899 request_cachep = kmem_cache_create("blkdev_requests",
2900 sizeof(struct request), 0, SLAB_PANIC, NULL);
2902 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
2903 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
2905 return 0;