Merge branch 'firewire-kernel-streaming' of git://git.alsa-project.org/alsa-kprivate
[firewire-audio.git] / block / blk-core.c
blob518dd423a5fef4c4dcf1d2cdc7af78e6f8c3a778
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>
31 #define CREATE_TRACE_POINTS
32 #include <trace/events/block.h>
34 #include "blk.h"
36 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
37 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
38 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
40 static int __make_request(struct request_queue *q, struct bio *bio);
43 * For the allocated request tables
45 static struct kmem_cache *request_cachep;
48 * For queue allocation
50 struct kmem_cache *blk_requestq_cachep;
53 * Controlling structure to kblockd
55 static struct workqueue_struct *kblockd_workqueue;
57 static void drive_stat_acct(struct request *rq, int new_io)
59 struct hd_struct *part;
60 int rw = rq_data_dir(rq);
61 int cpu;
63 if (!blk_do_io_stat(rq))
64 return;
66 cpu = part_stat_lock();
68 if (!new_io) {
69 part = rq->part;
70 part_stat_inc(cpu, part, merges[rw]);
71 } else {
72 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
73 if (!hd_struct_try_get(part)) {
75 * The partition is already being removed,
76 * the request will be accounted on the disk only
78 * We take a reference on disk->part0 although that
79 * partition will never be deleted, so we can treat
80 * it as any other partition.
82 part = &rq->rq_disk->part0;
83 hd_struct_get(part);
85 part_round_stats(cpu, part);
86 part_inc_in_flight(part, rw);
87 rq->part = part;
90 part_stat_unlock();
93 void blk_queue_congestion_threshold(struct request_queue *q)
95 int nr;
97 nr = q->nr_requests - (q->nr_requests / 8) + 1;
98 if (nr > q->nr_requests)
99 nr = q->nr_requests;
100 q->nr_congestion_on = nr;
102 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
103 if (nr < 1)
104 nr = 1;
105 q->nr_congestion_off = nr;
109 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
110 * @bdev: device
112 * Locates the passed device's request queue and returns the address of its
113 * backing_dev_info
115 * Will return NULL if the request queue cannot be located.
117 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
119 struct backing_dev_info *ret = NULL;
120 struct request_queue *q = bdev_get_queue(bdev);
122 if (q)
123 ret = &q->backing_dev_info;
124 return ret;
126 EXPORT_SYMBOL(blk_get_backing_dev_info);
128 void blk_rq_init(struct request_queue *q, struct request *rq)
130 memset(rq, 0, sizeof(*rq));
132 INIT_LIST_HEAD(&rq->queuelist);
133 INIT_LIST_HEAD(&rq->timeout_list);
134 rq->cpu = -1;
135 rq->q = q;
136 rq->__sector = (sector_t) -1;
137 INIT_HLIST_NODE(&rq->hash);
138 RB_CLEAR_NODE(&rq->rb_node);
139 rq->cmd = rq->__cmd;
140 rq->cmd_len = BLK_MAX_CDB;
141 rq->tag = -1;
142 rq->ref_count = 1;
143 rq->start_time = jiffies;
144 set_start_time_ns(rq);
145 rq->part = NULL;
147 EXPORT_SYMBOL(blk_rq_init);
149 static void req_bio_endio(struct request *rq, struct bio *bio,
150 unsigned int nbytes, int error)
152 struct request_queue *q = rq->q;
154 if (&q->flush_rq != rq) {
155 if (error)
156 clear_bit(BIO_UPTODATE, &bio->bi_flags);
157 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
158 error = -EIO;
160 if (unlikely(nbytes > bio->bi_size)) {
161 printk(KERN_ERR "%s: want %u bytes done, %u left\n",
162 __func__, nbytes, bio->bi_size);
163 nbytes = bio->bi_size;
166 if (unlikely(rq->cmd_flags & REQ_QUIET))
167 set_bit(BIO_QUIET, &bio->bi_flags);
169 bio->bi_size -= nbytes;
170 bio->bi_sector += (nbytes >> 9);
172 if (bio_integrity(bio))
173 bio_integrity_advance(bio, nbytes);
175 if (bio->bi_size == 0)
176 bio_endio(bio, error);
177 } else {
179 * Okay, this is the sequenced flush request in
180 * progress, just record the error;
182 if (error && !q->flush_err)
183 q->flush_err = error;
187 void blk_dump_rq_flags(struct request *rq, char *msg)
189 int bit;
191 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
192 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
193 rq->cmd_flags);
195 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
196 (unsigned long long)blk_rq_pos(rq),
197 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
198 printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n",
199 rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq));
201 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
202 printk(KERN_INFO " cdb: ");
203 for (bit = 0; bit < BLK_MAX_CDB; bit++)
204 printk("%02x ", rq->cmd[bit]);
205 printk("\n");
208 EXPORT_SYMBOL(blk_dump_rq_flags);
211 * "plug" the device if there are no outstanding requests: this will
212 * force the transfer to start only after we have put all the requests
213 * on the list.
215 * This is called with interrupts off and no requests on the queue and
216 * with the queue lock held.
218 void blk_plug_device(struct request_queue *q)
220 WARN_ON(!irqs_disabled());
223 * don't plug a stopped queue, it must be paired with blk_start_queue()
224 * which will restart the queueing
226 if (blk_queue_stopped(q))
227 return;
229 if (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED, q)) {
230 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
231 trace_block_plug(q);
234 EXPORT_SYMBOL(blk_plug_device);
237 * blk_plug_device_unlocked - plug a device without queue lock held
238 * @q: The &struct request_queue to plug
240 * Description:
241 * Like @blk_plug_device(), but grabs the queue lock and disables
242 * interrupts.
244 void blk_plug_device_unlocked(struct request_queue *q)
246 unsigned long flags;
248 spin_lock_irqsave(q->queue_lock, flags);
249 blk_plug_device(q);
250 spin_unlock_irqrestore(q->queue_lock, flags);
252 EXPORT_SYMBOL(blk_plug_device_unlocked);
255 * remove the queue from the plugged list, if present. called with
256 * queue lock held and interrupts disabled.
258 int blk_remove_plug(struct request_queue *q)
260 WARN_ON(!irqs_disabled());
262 if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED, q))
263 return 0;
265 del_timer(&q->unplug_timer);
266 return 1;
268 EXPORT_SYMBOL(blk_remove_plug);
271 * remove the plug and let it rip..
273 void __generic_unplug_device(struct request_queue *q)
275 if (unlikely(blk_queue_stopped(q)))
276 return;
277 if (!blk_remove_plug(q) && !blk_queue_nonrot(q))
278 return;
280 q->request_fn(q);
284 * generic_unplug_device - fire a request queue
285 * @q: The &struct request_queue in question
287 * Description:
288 * Linux uses plugging to build bigger requests queues before letting
289 * the device have at them. If a queue is plugged, the I/O scheduler
290 * is still adding and merging requests on the queue. Once the queue
291 * gets unplugged, the request_fn defined for the queue is invoked and
292 * transfers started.
294 void generic_unplug_device(struct request_queue *q)
296 if (blk_queue_plugged(q)) {
297 spin_lock_irq(q->queue_lock);
298 __generic_unplug_device(q);
299 spin_unlock_irq(q->queue_lock);
302 EXPORT_SYMBOL(generic_unplug_device);
304 static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
305 struct page *page)
307 struct request_queue *q = bdi->unplug_io_data;
309 blk_unplug(q);
312 void blk_unplug_work(struct work_struct *work)
314 struct request_queue *q =
315 container_of(work, struct request_queue, unplug_work);
317 trace_block_unplug_io(q);
318 q->unplug_fn(q);
321 void blk_unplug_timeout(unsigned long data)
323 struct request_queue *q = (struct request_queue *)data;
325 trace_block_unplug_timer(q);
326 kblockd_schedule_work(q, &q->unplug_work);
329 void blk_unplug(struct request_queue *q)
332 * devices don't necessarily have an ->unplug_fn defined
334 if (q->unplug_fn) {
335 trace_block_unplug_io(q);
336 q->unplug_fn(q);
339 EXPORT_SYMBOL(blk_unplug);
342 * blk_start_queue - restart a previously stopped queue
343 * @q: The &struct request_queue in question
345 * Description:
346 * blk_start_queue() will clear the stop flag on the queue, and call
347 * the request_fn for the queue if it was in a stopped state when
348 * entered. Also see blk_stop_queue(). Queue lock must be held.
350 void blk_start_queue(struct request_queue *q)
352 WARN_ON(!irqs_disabled());
354 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
355 __blk_run_queue(q, false);
357 EXPORT_SYMBOL(blk_start_queue);
360 * blk_stop_queue - stop a queue
361 * @q: The &struct request_queue in question
363 * Description:
364 * The Linux block layer assumes that a block driver will consume all
365 * entries on the request queue when the request_fn strategy is called.
366 * Often this will not happen, because of hardware limitations (queue
367 * depth settings). If a device driver gets a 'queue full' response,
368 * or if it simply chooses not to queue more I/O at one point, it can
369 * call this function to prevent the request_fn from being called until
370 * the driver has signalled it's ready to go again. This happens by calling
371 * blk_start_queue() to restart queue operations. Queue lock must be held.
373 void blk_stop_queue(struct request_queue *q)
375 blk_remove_plug(q);
376 queue_flag_set(QUEUE_FLAG_STOPPED, q);
378 EXPORT_SYMBOL(blk_stop_queue);
381 * blk_sync_queue - cancel any pending callbacks on a queue
382 * @q: the queue
384 * Description:
385 * The block layer may perform asynchronous callback activity
386 * on a queue, such as calling the unplug function after a timeout.
387 * A block device may call blk_sync_queue to ensure that any
388 * such activity is cancelled, thus allowing it to release resources
389 * that the callbacks might use. The caller must already have made sure
390 * that its ->make_request_fn will not re-add plugging prior to calling
391 * this function.
394 void blk_sync_queue(struct request_queue *q)
396 del_timer_sync(&q->unplug_timer);
397 del_timer_sync(&q->timeout);
398 cancel_work_sync(&q->unplug_work);
399 throtl_shutdown_timer_wq(q);
401 EXPORT_SYMBOL(blk_sync_queue);
404 * __blk_run_queue - run a single device queue
405 * @q: The queue to run
406 * @force_kblockd: Don't run @q->request_fn directly. Use kblockd.
408 * Description:
409 * See @blk_run_queue. This variant must be called with the queue lock
410 * held and interrupts disabled.
413 void __blk_run_queue(struct request_queue *q, bool force_kblockd)
415 blk_remove_plug(q);
417 if (unlikely(blk_queue_stopped(q)))
418 return;
420 if (elv_queue_empty(q))
421 return;
424 * Only recurse once to avoid overrunning the stack, let the unplug
425 * handling reinvoke the handler shortly if we already got there.
427 if (!force_kblockd && !queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) {
428 q->request_fn(q);
429 queue_flag_clear(QUEUE_FLAG_REENTER, q);
430 } else {
431 queue_flag_set(QUEUE_FLAG_PLUGGED, q);
432 kblockd_schedule_work(q, &q->unplug_work);
435 EXPORT_SYMBOL(__blk_run_queue);
438 * blk_run_queue - run a single device queue
439 * @q: The queue to run
441 * Description:
442 * Invoke request handling on this queue, if it has pending work to do.
443 * May be used to restart queueing when a request has completed.
445 void blk_run_queue(struct request_queue *q)
447 unsigned long flags;
449 spin_lock_irqsave(q->queue_lock, flags);
450 __blk_run_queue(q, false);
451 spin_unlock_irqrestore(q->queue_lock, flags);
453 EXPORT_SYMBOL(blk_run_queue);
455 void blk_put_queue(struct request_queue *q)
457 kobject_put(&q->kobj);
460 void blk_cleanup_queue(struct request_queue *q)
463 * We know we have process context here, so we can be a little
464 * cautious and ensure that pending block actions on this device
465 * are done before moving on. Going into this function, we should
466 * not have processes doing IO to this device.
468 blk_sync_queue(q);
470 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
471 mutex_lock(&q->sysfs_lock);
472 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
473 mutex_unlock(&q->sysfs_lock);
475 if (q->elevator)
476 elevator_exit(q->elevator);
478 blk_put_queue(q);
480 EXPORT_SYMBOL(blk_cleanup_queue);
482 static int blk_init_free_list(struct request_queue *q)
484 struct request_list *rl = &q->rq;
486 if (unlikely(rl->rq_pool))
487 return 0;
489 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
490 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
491 rl->elvpriv = 0;
492 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
493 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
495 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
496 mempool_free_slab, request_cachep, q->node);
498 if (!rl->rq_pool)
499 return -ENOMEM;
501 return 0;
504 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
506 return blk_alloc_queue_node(gfp_mask, -1);
508 EXPORT_SYMBOL(blk_alloc_queue);
510 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
512 struct request_queue *q;
513 int err;
515 q = kmem_cache_alloc_node(blk_requestq_cachep,
516 gfp_mask | __GFP_ZERO, node_id);
517 if (!q)
518 return NULL;
520 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
521 q->backing_dev_info.unplug_io_data = q;
522 q->backing_dev_info.ra_pages =
523 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
524 q->backing_dev_info.state = 0;
525 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
526 q->backing_dev_info.name = "block";
528 err = bdi_init(&q->backing_dev_info);
529 if (err) {
530 kmem_cache_free(blk_requestq_cachep, q);
531 return NULL;
534 if (blk_throtl_init(q)) {
535 kmem_cache_free(blk_requestq_cachep, q);
536 return NULL;
539 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
540 laptop_mode_timer_fn, (unsigned long) q);
541 init_timer(&q->unplug_timer);
542 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
543 INIT_LIST_HEAD(&q->timeout_list);
544 INIT_LIST_HEAD(&q->pending_flushes);
545 INIT_WORK(&q->unplug_work, blk_unplug_work);
547 kobject_init(&q->kobj, &blk_queue_ktype);
549 mutex_init(&q->sysfs_lock);
550 spin_lock_init(&q->__queue_lock);
552 return q;
554 EXPORT_SYMBOL(blk_alloc_queue_node);
557 * blk_init_queue - prepare a request queue for use with a block device
558 * @rfn: The function to be called to process requests that have been
559 * placed on the queue.
560 * @lock: Request queue spin lock
562 * Description:
563 * If a block device wishes to use the standard request handling procedures,
564 * which sorts requests and coalesces adjacent requests, then it must
565 * call blk_init_queue(). The function @rfn will be called when there
566 * are requests on the queue that need to be processed. If the device
567 * supports plugging, then @rfn may not be called immediately when requests
568 * are available on the queue, but may be called at some time later instead.
569 * Plugged queues are generally unplugged when a buffer belonging to one
570 * of the requests on the queue is needed, or due to memory pressure.
572 * @rfn is not required, or even expected, to remove all requests off the
573 * queue, but only as many as it can handle at a time. If it does leave
574 * requests on the queue, it is responsible for arranging that the requests
575 * get dealt with eventually.
577 * The queue spin lock must be held while manipulating the requests on the
578 * request queue; this lock will be taken also from interrupt context, so irq
579 * disabling is needed for it.
581 * Function returns a pointer to the initialized request queue, or %NULL if
582 * it didn't succeed.
584 * Note:
585 * blk_init_queue() must be paired with a blk_cleanup_queue() call
586 * when the block device is deactivated (such as at module unload).
589 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
591 return blk_init_queue_node(rfn, lock, -1);
593 EXPORT_SYMBOL(blk_init_queue);
595 struct request_queue *
596 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
598 struct request_queue *uninit_q, *q;
600 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
601 if (!uninit_q)
602 return NULL;
604 q = blk_init_allocated_queue_node(uninit_q, rfn, lock, node_id);
605 if (!q)
606 blk_cleanup_queue(uninit_q);
608 return q;
610 EXPORT_SYMBOL(blk_init_queue_node);
612 struct request_queue *
613 blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
614 spinlock_t *lock)
616 return blk_init_allocated_queue_node(q, rfn, lock, -1);
618 EXPORT_SYMBOL(blk_init_allocated_queue);
620 struct request_queue *
621 blk_init_allocated_queue_node(struct request_queue *q, request_fn_proc *rfn,
622 spinlock_t *lock, int node_id)
624 if (!q)
625 return NULL;
627 q->node = node_id;
628 if (blk_init_free_list(q))
629 return NULL;
631 q->request_fn = rfn;
632 q->prep_rq_fn = NULL;
633 q->unprep_rq_fn = NULL;
634 q->unplug_fn = generic_unplug_device;
635 q->queue_flags = QUEUE_FLAG_DEFAULT;
636 q->queue_lock = lock;
639 * This also sets hw/phys segments, boundary and size
641 blk_queue_make_request(q, __make_request);
643 q->sg_reserved_size = INT_MAX;
646 * all done
648 if (!elevator_init(q, NULL)) {
649 blk_queue_congestion_threshold(q);
650 return q;
653 return NULL;
655 EXPORT_SYMBOL(blk_init_allocated_queue_node);
657 int blk_get_queue(struct request_queue *q)
659 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
660 kobject_get(&q->kobj);
661 return 0;
664 return 1;
667 static inline void blk_free_request(struct request_queue *q, struct request *rq)
669 if (rq->cmd_flags & REQ_ELVPRIV)
670 elv_put_request(q, rq);
671 mempool_free(rq, q->rq.rq_pool);
674 static struct request *
675 blk_alloc_request(struct request_queue *q, int flags, int priv, gfp_t gfp_mask)
677 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
679 if (!rq)
680 return NULL;
682 blk_rq_init(q, rq);
684 rq->cmd_flags = flags | REQ_ALLOCED;
686 if (priv) {
687 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
688 mempool_free(rq, q->rq.rq_pool);
689 return NULL;
691 rq->cmd_flags |= REQ_ELVPRIV;
694 return rq;
698 * ioc_batching returns true if the ioc is a valid batching request and
699 * should be given priority access to a request.
701 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
703 if (!ioc)
704 return 0;
707 * Make sure the process is able to allocate at least 1 request
708 * even if the batch times out, otherwise we could theoretically
709 * lose wakeups.
711 return ioc->nr_batch_requests == q->nr_batching ||
712 (ioc->nr_batch_requests > 0
713 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
717 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
718 * will cause the process to be a "batcher" on all queues in the system. This
719 * is the behaviour we want though - once it gets a wakeup it should be given
720 * a nice run.
722 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
724 if (!ioc || ioc_batching(q, ioc))
725 return;
727 ioc->nr_batch_requests = q->nr_batching;
728 ioc->last_waited = jiffies;
731 static void __freed_request(struct request_queue *q, int sync)
733 struct request_list *rl = &q->rq;
735 if (rl->count[sync] < queue_congestion_off_threshold(q))
736 blk_clear_queue_congested(q, sync);
738 if (rl->count[sync] + 1 <= q->nr_requests) {
739 if (waitqueue_active(&rl->wait[sync]))
740 wake_up(&rl->wait[sync]);
742 blk_clear_queue_full(q, sync);
747 * A request has just been released. Account for it, update the full and
748 * congestion status, wake up any waiters. Called under q->queue_lock.
750 static void freed_request(struct request_queue *q, int sync, int priv)
752 struct request_list *rl = &q->rq;
754 rl->count[sync]--;
755 if (priv)
756 rl->elvpriv--;
758 __freed_request(q, sync);
760 if (unlikely(rl->starved[sync ^ 1]))
761 __freed_request(q, sync ^ 1);
765 * Get a free request, queue_lock must be held.
766 * Returns NULL on failure, with queue_lock held.
767 * Returns !NULL on success, with queue_lock *not held*.
769 static struct request *get_request(struct request_queue *q, int rw_flags,
770 struct bio *bio, gfp_t gfp_mask)
772 struct request *rq = NULL;
773 struct request_list *rl = &q->rq;
774 struct io_context *ioc = NULL;
775 const bool is_sync = rw_is_sync(rw_flags) != 0;
776 int may_queue, priv;
778 may_queue = elv_may_queue(q, rw_flags);
779 if (may_queue == ELV_MQUEUE_NO)
780 goto rq_starved;
782 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
783 if (rl->count[is_sync]+1 >= q->nr_requests) {
784 ioc = current_io_context(GFP_ATOMIC, q->node);
786 * The queue will fill after this allocation, so set
787 * it as full, and mark this process as "batching".
788 * This process will be allowed to complete a batch of
789 * requests, others will be blocked.
791 if (!blk_queue_full(q, is_sync)) {
792 ioc_set_batching(q, ioc);
793 blk_set_queue_full(q, is_sync);
794 } else {
795 if (may_queue != ELV_MQUEUE_MUST
796 && !ioc_batching(q, ioc)) {
798 * The queue is full and the allocating
799 * process is not a "batcher", and not
800 * exempted by the IO scheduler
802 goto out;
806 blk_set_queue_congested(q, is_sync);
810 * Only allow batching queuers to allocate up to 50% over the defined
811 * limit of requests, otherwise we could have thousands of requests
812 * allocated with any setting of ->nr_requests
814 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
815 goto out;
817 rl->count[is_sync]++;
818 rl->starved[is_sync] = 0;
820 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
821 if (priv)
822 rl->elvpriv++;
824 if (blk_queue_io_stat(q))
825 rw_flags |= REQ_IO_STAT;
826 spin_unlock_irq(q->queue_lock);
828 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
829 if (unlikely(!rq)) {
831 * Allocation failed presumably due to memory. Undo anything
832 * we might have messed up.
834 * Allocating task should really be put onto the front of the
835 * wait queue, but this is pretty rare.
837 spin_lock_irq(q->queue_lock);
838 freed_request(q, is_sync, priv);
841 * in the very unlikely event that allocation failed and no
842 * requests for this direction was pending, mark us starved
843 * so that freeing of a request in the other direction will
844 * notice us. another possible fix would be to split the
845 * rq mempool into READ and WRITE
847 rq_starved:
848 if (unlikely(rl->count[is_sync] == 0))
849 rl->starved[is_sync] = 1;
851 goto out;
855 * ioc may be NULL here, and ioc_batching will be false. That's
856 * OK, if the queue is under the request limit then requests need
857 * not count toward the nr_batch_requests limit. There will always
858 * be some limit enforced by BLK_BATCH_TIME.
860 if (ioc_batching(q, ioc))
861 ioc->nr_batch_requests--;
863 trace_block_getrq(q, bio, rw_flags & 1);
864 out:
865 return rq;
869 * No available requests for this queue, unplug the device and wait for some
870 * requests to become available.
872 * Called with q->queue_lock held, and returns with it unlocked.
874 static struct request *get_request_wait(struct request_queue *q, int rw_flags,
875 struct bio *bio)
877 const bool is_sync = rw_is_sync(rw_flags) != 0;
878 struct request *rq;
880 rq = get_request(q, rw_flags, bio, GFP_NOIO);
881 while (!rq) {
882 DEFINE_WAIT(wait);
883 struct io_context *ioc;
884 struct request_list *rl = &q->rq;
886 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
887 TASK_UNINTERRUPTIBLE);
889 trace_block_sleeprq(q, bio, rw_flags & 1);
891 __generic_unplug_device(q);
892 spin_unlock_irq(q->queue_lock);
893 io_schedule();
896 * After sleeping, we become a "batching" process and
897 * will be able to allocate at least one request, and
898 * up to a big batch of them for a small period time.
899 * See ioc_batching, ioc_set_batching
901 ioc = current_io_context(GFP_NOIO, q->node);
902 ioc_set_batching(q, ioc);
904 spin_lock_irq(q->queue_lock);
905 finish_wait(&rl->wait[is_sync], &wait);
907 rq = get_request(q, rw_flags, bio, GFP_NOIO);
910 return rq;
913 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
915 struct request *rq;
917 BUG_ON(rw != READ && rw != WRITE);
919 spin_lock_irq(q->queue_lock);
920 if (gfp_mask & __GFP_WAIT) {
921 rq = get_request_wait(q, rw, NULL);
922 } else {
923 rq = get_request(q, rw, NULL, gfp_mask);
924 if (!rq)
925 spin_unlock_irq(q->queue_lock);
927 /* q->queue_lock is unlocked at this point */
929 return rq;
931 EXPORT_SYMBOL(blk_get_request);
934 * blk_make_request - given a bio, allocate a corresponding struct request.
935 * @q: target request queue
936 * @bio: The bio describing the memory mappings that will be submitted for IO.
937 * It may be a chained-bio properly constructed by block/bio layer.
938 * @gfp_mask: gfp flags to be used for memory allocation
940 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
941 * type commands. Where the struct request needs to be farther initialized by
942 * the caller. It is passed a &struct bio, which describes the memory info of
943 * the I/O transfer.
945 * The caller of blk_make_request must make sure that bi_io_vec
946 * are set to describe the memory buffers. That bio_data_dir() will return
947 * the needed direction of the request. (And all bio's in the passed bio-chain
948 * are properly set accordingly)
950 * If called under none-sleepable conditions, mapped bio buffers must not
951 * need bouncing, by calling the appropriate masked or flagged allocator,
952 * suitable for the target device. Otherwise the call to blk_queue_bounce will
953 * BUG.
955 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
956 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
957 * anything but the first bio in the chain. Otherwise you risk waiting for IO
958 * completion of a bio that hasn't been submitted yet, thus resulting in a
959 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
960 * of bio_alloc(), as that avoids the mempool deadlock.
961 * If possible a big IO should be split into smaller parts when allocation
962 * fails. Partial allocation should not be an error, or you risk a live-lock.
964 struct request *blk_make_request(struct request_queue *q, struct bio *bio,
965 gfp_t gfp_mask)
967 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
969 if (unlikely(!rq))
970 return ERR_PTR(-ENOMEM);
972 for_each_bio(bio) {
973 struct bio *bounce_bio = bio;
974 int ret;
976 blk_queue_bounce(q, &bounce_bio);
977 ret = blk_rq_append_bio(q, rq, bounce_bio);
978 if (unlikely(ret)) {
979 blk_put_request(rq);
980 return ERR_PTR(ret);
984 return rq;
986 EXPORT_SYMBOL(blk_make_request);
989 * blk_requeue_request - put a request back on queue
990 * @q: request queue where request should be inserted
991 * @rq: request to be inserted
993 * Description:
994 * Drivers often keep queueing requests until the hardware cannot accept
995 * more, when that condition happens we need to put the request back
996 * on the queue. Must be called with queue lock held.
998 void blk_requeue_request(struct request_queue *q, struct request *rq)
1000 blk_delete_timer(rq);
1001 blk_clear_rq_complete(rq);
1002 trace_block_rq_requeue(q, rq);
1004 if (blk_rq_tagged(rq))
1005 blk_queue_end_tag(q, rq);
1007 BUG_ON(blk_queued_rq(rq));
1009 elv_requeue_request(q, rq);
1011 EXPORT_SYMBOL(blk_requeue_request);
1014 * blk_insert_request - insert a special request into a request queue
1015 * @q: request queue where request should be inserted
1016 * @rq: request to be inserted
1017 * @at_head: insert request at head or tail of queue
1018 * @data: private data
1020 * Description:
1021 * Many block devices need to execute commands asynchronously, so they don't
1022 * block the whole kernel from preemption during request execution. This is
1023 * accomplished normally by inserting aritficial requests tagged as
1024 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
1025 * be scheduled for actual execution by the request queue.
1027 * We have the option of inserting the head or the tail of the queue.
1028 * Typically we use the tail for new ioctls and so forth. We use the head
1029 * of the queue for things like a QUEUE_FULL message from a device, or a
1030 * host that is unable to accept a particular command.
1032 void blk_insert_request(struct request_queue *q, struct request *rq,
1033 int at_head, void *data)
1035 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
1036 unsigned long flags;
1039 * tell I/O scheduler that this isn't a regular read/write (ie it
1040 * must not attempt merges on this) and that it acts as a soft
1041 * barrier
1043 rq->cmd_type = REQ_TYPE_SPECIAL;
1045 rq->special = data;
1047 spin_lock_irqsave(q->queue_lock, flags);
1050 * If command is tagged, release the tag
1052 if (blk_rq_tagged(rq))
1053 blk_queue_end_tag(q, rq);
1055 drive_stat_acct(rq, 1);
1056 __elv_add_request(q, rq, where, 0);
1057 __blk_run_queue(q, false);
1058 spin_unlock_irqrestore(q->queue_lock, flags);
1060 EXPORT_SYMBOL(blk_insert_request);
1062 static void part_round_stats_single(int cpu, struct hd_struct *part,
1063 unsigned long now)
1065 if (now == part->stamp)
1066 return;
1068 if (part_in_flight(part)) {
1069 __part_stat_add(cpu, part, time_in_queue,
1070 part_in_flight(part) * (now - part->stamp));
1071 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1073 part->stamp = now;
1077 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1078 * @cpu: cpu number for stats access
1079 * @part: target partition
1081 * The average IO queue length and utilisation statistics are maintained
1082 * by observing the current state of the queue length and the amount of
1083 * time it has been in this state for.
1085 * Normally, that accounting is done on IO completion, but that can result
1086 * in more than a second's worth of IO being accounted for within any one
1087 * second, leading to >100% utilisation. To deal with that, we call this
1088 * function to do a round-off before returning the results when reading
1089 * /proc/diskstats. This accounts immediately for all queue usage up to
1090 * the current jiffies and restarts the counters again.
1092 void part_round_stats(int cpu, struct hd_struct *part)
1094 unsigned long now = jiffies;
1096 if (part->partno)
1097 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1098 part_round_stats_single(cpu, part, now);
1100 EXPORT_SYMBOL_GPL(part_round_stats);
1103 * queue lock must be held
1105 void __blk_put_request(struct request_queue *q, struct request *req)
1107 if (unlikely(!q))
1108 return;
1109 if (unlikely(--req->ref_count))
1110 return;
1112 elv_completed_request(q, req);
1114 /* this is a bio leak */
1115 WARN_ON(req->bio != NULL);
1118 * Request may not have originated from ll_rw_blk. if not,
1119 * it didn't come out of our reserved rq pools
1121 if (req->cmd_flags & REQ_ALLOCED) {
1122 int is_sync = rq_is_sync(req) != 0;
1123 int priv = req->cmd_flags & REQ_ELVPRIV;
1125 BUG_ON(!list_empty(&req->queuelist));
1126 BUG_ON(!hlist_unhashed(&req->hash));
1128 blk_free_request(q, req);
1129 freed_request(q, is_sync, priv);
1132 EXPORT_SYMBOL_GPL(__blk_put_request);
1134 void blk_put_request(struct request *req)
1136 unsigned long flags;
1137 struct request_queue *q = req->q;
1139 spin_lock_irqsave(q->queue_lock, flags);
1140 __blk_put_request(q, req);
1141 spin_unlock_irqrestore(q->queue_lock, flags);
1143 EXPORT_SYMBOL(blk_put_request);
1146 * blk_add_request_payload - add a payload to a request
1147 * @rq: request to update
1148 * @page: page backing the payload
1149 * @len: length of the payload.
1151 * This allows to later add a payload to an already submitted request by
1152 * a block driver. The driver needs to take care of freeing the payload
1153 * itself.
1155 * Note that this is a quite horrible hack and nothing but handling of
1156 * discard requests should ever use it.
1158 void blk_add_request_payload(struct request *rq, struct page *page,
1159 unsigned int len)
1161 struct bio *bio = rq->bio;
1163 bio->bi_io_vec->bv_page = page;
1164 bio->bi_io_vec->bv_offset = 0;
1165 bio->bi_io_vec->bv_len = len;
1167 bio->bi_size = len;
1168 bio->bi_vcnt = 1;
1169 bio->bi_phys_segments = 1;
1171 rq->__data_len = rq->resid_len = len;
1172 rq->nr_phys_segments = 1;
1173 rq->buffer = bio_data(bio);
1175 EXPORT_SYMBOL_GPL(blk_add_request_payload);
1177 void init_request_from_bio(struct request *req, struct bio *bio)
1179 req->cpu = bio->bi_comp_cpu;
1180 req->cmd_type = REQ_TYPE_FS;
1182 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1183 if (bio->bi_rw & REQ_RAHEAD)
1184 req->cmd_flags |= REQ_FAILFAST_MASK;
1186 req->errors = 0;
1187 req->__sector = bio->bi_sector;
1188 req->ioprio = bio_prio(bio);
1189 blk_rq_bio_prep(req->q, req, bio);
1193 * Only disabling plugging for non-rotational devices if it does tagging
1194 * as well, otherwise we do need the proper merging
1196 static inline bool queue_should_plug(struct request_queue *q)
1198 return !(blk_queue_nonrot(q) && blk_queue_tagged(q));
1201 static int __make_request(struct request_queue *q, struct bio *bio)
1203 struct request *req;
1204 int el_ret;
1205 unsigned int bytes = bio->bi_size;
1206 const unsigned short prio = bio_prio(bio);
1207 const bool sync = !!(bio->bi_rw & REQ_SYNC);
1208 const bool unplug = !!(bio->bi_rw & REQ_UNPLUG);
1209 const unsigned long ff = bio->bi_rw & REQ_FAILFAST_MASK;
1210 int where = ELEVATOR_INSERT_SORT;
1211 int rw_flags;
1214 * low level driver can indicate that it wants pages above a
1215 * certain limit bounced to low memory (ie for highmem, or even
1216 * ISA dma in theory)
1218 blk_queue_bounce(q, &bio);
1220 spin_lock_irq(q->queue_lock);
1222 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
1223 where = ELEVATOR_INSERT_FRONT;
1224 goto get_rq;
1227 if (elv_queue_empty(q))
1228 goto get_rq;
1230 el_ret = elv_merge(q, &req, bio);
1231 switch (el_ret) {
1232 case ELEVATOR_BACK_MERGE:
1233 BUG_ON(!rq_mergeable(req));
1235 if (!ll_back_merge_fn(q, req, bio))
1236 break;
1238 trace_block_bio_backmerge(q, bio);
1240 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1241 blk_rq_set_mixed_merge(req);
1243 req->biotail->bi_next = bio;
1244 req->biotail = bio;
1245 req->__data_len += bytes;
1246 req->ioprio = ioprio_best(req->ioprio, prio);
1247 if (!blk_rq_cpu_valid(req))
1248 req->cpu = bio->bi_comp_cpu;
1249 drive_stat_acct(req, 0);
1250 elv_bio_merged(q, req, bio);
1251 if (!attempt_back_merge(q, req))
1252 elv_merged_request(q, req, el_ret);
1253 goto out;
1255 case ELEVATOR_FRONT_MERGE:
1256 BUG_ON(!rq_mergeable(req));
1258 if (!ll_front_merge_fn(q, req, bio))
1259 break;
1261 trace_block_bio_frontmerge(q, bio);
1263 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) {
1264 blk_rq_set_mixed_merge(req);
1265 req->cmd_flags &= ~REQ_FAILFAST_MASK;
1266 req->cmd_flags |= ff;
1269 bio->bi_next = req->bio;
1270 req->bio = bio;
1273 * may not be valid. if the low level driver said
1274 * it didn't need a bounce buffer then it better
1275 * not touch req->buffer either...
1277 req->buffer = bio_data(bio);
1278 req->__sector = bio->bi_sector;
1279 req->__data_len += bytes;
1280 req->ioprio = ioprio_best(req->ioprio, prio);
1281 if (!blk_rq_cpu_valid(req))
1282 req->cpu = bio->bi_comp_cpu;
1283 drive_stat_acct(req, 0);
1284 elv_bio_merged(q, req, bio);
1285 if (!attempt_front_merge(q, req))
1286 elv_merged_request(q, req, el_ret);
1287 goto out;
1289 /* ELV_NO_MERGE: elevator says don't/can't merge. */
1290 default:
1294 get_rq:
1296 * This sync check and mask will be re-done in init_request_from_bio(),
1297 * but we need to set it earlier to expose the sync flag to the
1298 * rq allocator and io schedulers.
1300 rw_flags = bio_data_dir(bio);
1301 if (sync)
1302 rw_flags |= REQ_SYNC;
1305 * Grab a free request. This is might sleep but can not fail.
1306 * Returns with the queue unlocked.
1308 req = get_request_wait(q, rw_flags, bio);
1311 * After dropping the lock and possibly sleeping here, our request
1312 * may now be mergeable after it had proven unmergeable (above).
1313 * We don't worry about that case for efficiency. It won't happen
1314 * often, and the elevators are able to handle it.
1316 init_request_from_bio(req, bio);
1318 spin_lock_irq(q->queue_lock);
1319 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) ||
1320 bio_flagged(bio, BIO_CPU_AFFINE))
1321 req->cpu = blk_cpu_to_group(smp_processor_id());
1322 if (queue_should_plug(q) && elv_queue_empty(q))
1323 blk_plug_device(q);
1325 /* insert the request into the elevator */
1326 drive_stat_acct(req, 1);
1327 __elv_add_request(q, req, where, 0);
1328 out:
1329 if (unplug || !queue_should_plug(q))
1330 __generic_unplug_device(q);
1331 spin_unlock_irq(q->queue_lock);
1332 return 0;
1336 * If bio->bi_dev is a partition, remap the location
1338 static inline void blk_partition_remap(struct bio *bio)
1340 struct block_device *bdev = bio->bi_bdev;
1342 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1343 struct hd_struct *p = bdev->bd_part;
1345 bio->bi_sector += p->start_sect;
1346 bio->bi_bdev = bdev->bd_contains;
1348 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1349 bdev->bd_dev,
1350 bio->bi_sector - p->start_sect);
1354 static void handle_bad_sector(struct bio *bio)
1356 char b[BDEVNAME_SIZE];
1358 printk(KERN_INFO "attempt to access beyond end of device\n");
1359 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1360 bdevname(bio->bi_bdev, b),
1361 bio->bi_rw,
1362 (unsigned long long)bio->bi_sector + bio_sectors(bio),
1363 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1365 set_bit(BIO_EOF, &bio->bi_flags);
1368 #ifdef CONFIG_FAIL_MAKE_REQUEST
1370 static DECLARE_FAULT_ATTR(fail_make_request);
1372 static int __init setup_fail_make_request(char *str)
1374 return setup_fault_attr(&fail_make_request, str);
1376 __setup("fail_make_request=", setup_fail_make_request);
1378 static int should_fail_request(struct bio *bio)
1380 struct hd_struct *part = bio->bi_bdev->bd_part;
1382 if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail)
1383 return should_fail(&fail_make_request, bio->bi_size);
1385 return 0;
1388 static int __init fail_make_request_debugfs(void)
1390 return init_fault_attr_dentries(&fail_make_request,
1391 "fail_make_request");
1394 late_initcall(fail_make_request_debugfs);
1396 #else /* CONFIG_FAIL_MAKE_REQUEST */
1398 static inline int should_fail_request(struct bio *bio)
1400 return 0;
1403 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1406 * Check whether this bio extends beyond the end of the device.
1408 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1410 sector_t maxsector;
1412 if (!nr_sectors)
1413 return 0;
1415 /* Test device or partition size, when known. */
1416 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1417 if (maxsector) {
1418 sector_t sector = bio->bi_sector;
1420 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1422 * This may well happen - the kernel calls bread()
1423 * without checking the size of the device, e.g., when
1424 * mounting a device.
1426 handle_bad_sector(bio);
1427 return 1;
1431 return 0;
1435 * generic_make_request - hand a buffer to its device driver for I/O
1436 * @bio: The bio describing the location in memory and on the device.
1438 * generic_make_request() is used to make I/O requests of block
1439 * devices. It is passed a &struct bio, which describes the I/O that needs
1440 * to be done.
1442 * generic_make_request() does not return any status. The
1443 * success/failure status of the request, along with notification of
1444 * completion, is delivered asynchronously through the bio->bi_end_io
1445 * function described (one day) else where.
1447 * The caller of generic_make_request must make sure that bi_io_vec
1448 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1449 * set to describe the device address, and the
1450 * bi_end_io and optionally bi_private are set to describe how
1451 * completion notification should be signaled.
1453 * generic_make_request and the drivers it calls may use bi_next if this
1454 * bio happens to be merged with someone else, and may change bi_dev and
1455 * bi_sector for remaps as it sees fit. So the values of these fields
1456 * should NOT be depended on after the call to generic_make_request.
1458 static inline void __generic_make_request(struct bio *bio)
1460 struct request_queue *q;
1461 sector_t old_sector;
1462 int ret, nr_sectors = bio_sectors(bio);
1463 dev_t old_dev;
1464 int err = -EIO;
1466 might_sleep();
1468 if (bio_check_eod(bio, nr_sectors))
1469 goto end_io;
1472 * Resolve the mapping until finished. (drivers are
1473 * still free to implement/resolve their own stacking
1474 * by explicitly returning 0)
1476 * NOTE: we don't repeat the blk_size check for each new device.
1477 * Stacking drivers are expected to know what they are doing.
1479 old_sector = -1;
1480 old_dev = 0;
1481 do {
1482 char b[BDEVNAME_SIZE];
1484 q = bdev_get_queue(bio->bi_bdev);
1485 if (unlikely(!q)) {
1486 printk(KERN_ERR
1487 "generic_make_request: Trying to access "
1488 "nonexistent block-device %s (%Lu)\n",
1489 bdevname(bio->bi_bdev, b),
1490 (long long) bio->bi_sector);
1491 goto end_io;
1494 if (unlikely(!(bio->bi_rw & REQ_DISCARD) &&
1495 nr_sectors > queue_max_hw_sectors(q))) {
1496 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1497 bdevname(bio->bi_bdev, b),
1498 bio_sectors(bio),
1499 queue_max_hw_sectors(q));
1500 goto end_io;
1503 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
1504 goto end_io;
1506 if (should_fail_request(bio))
1507 goto end_io;
1510 * If this device has partitions, remap block n
1511 * of partition p to block n+start(p) of the disk.
1513 blk_partition_remap(bio);
1515 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1516 goto end_io;
1518 if (old_sector != -1)
1519 trace_block_bio_remap(q, bio, old_dev, old_sector);
1521 old_sector = bio->bi_sector;
1522 old_dev = bio->bi_bdev->bd_dev;
1524 if (bio_check_eod(bio, nr_sectors))
1525 goto end_io;
1528 * Filter flush bio's early so that make_request based
1529 * drivers without flush support don't have to worry
1530 * about them.
1532 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1533 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1534 if (!nr_sectors) {
1535 err = 0;
1536 goto end_io;
1540 if ((bio->bi_rw & REQ_DISCARD) &&
1541 (!blk_queue_discard(q) ||
1542 ((bio->bi_rw & REQ_SECURE) &&
1543 !blk_queue_secdiscard(q)))) {
1544 err = -EOPNOTSUPP;
1545 goto end_io;
1548 blk_throtl_bio(q, &bio);
1551 * If bio = NULL, bio has been throttled and will be submitted
1552 * later.
1554 if (!bio)
1555 break;
1557 trace_block_bio_queue(q, bio);
1559 ret = q->make_request_fn(q, bio);
1560 } while (ret);
1562 return;
1564 end_io:
1565 bio_endio(bio, err);
1569 * We only want one ->make_request_fn to be active at a time,
1570 * else stack usage with stacked devices could be a problem.
1571 * So use current->bio_list to keep a list of requests
1572 * submited by a make_request_fn function.
1573 * current->bio_list is also used as a flag to say if
1574 * generic_make_request is currently active in this task or not.
1575 * If it is NULL, then no make_request is active. If it is non-NULL,
1576 * then a make_request is active, and new requests should be added
1577 * at the tail
1579 void generic_make_request(struct bio *bio)
1581 struct bio_list bio_list_on_stack;
1583 if (current->bio_list) {
1584 /* make_request is active */
1585 bio_list_add(current->bio_list, bio);
1586 return;
1588 /* following loop may be a bit non-obvious, and so deserves some
1589 * explanation.
1590 * Before entering the loop, bio->bi_next is NULL (as all callers
1591 * ensure that) so we have a list with a single bio.
1592 * We pretend that we have just taken it off a longer list, so
1593 * we assign bio_list to a pointer to the bio_list_on_stack,
1594 * thus initialising the bio_list of new bios to be
1595 * added. __generic_make_request may indeed add some more bios
1596 * through a recursive call to generic_make_request. If it
1597 * did, we find a non-NULL value in bio_list and re-enter the loop
1598 * from the top. In this case we really did just take the bio
1599 * of the top of the list (no pretending) and so remove it from
1600 * bio_list, and call into __generic_make_request again.
1602 * The loop was structured like this to make only one call to
1603 * __generic_make_request (which is important as it is large and
1604 * inlined) and to keep the structure simple.
1606 BUG_ON(bio->bi_next);
1607 bio_list_init(&bio_list_on_stack);
1608 current->bio_list = &bio_list_on_stack;
1609 do {
1610 __generic_make_request(bio);
1611 bio = bio_list_pop(current->bio_list);
1612 } while (bio);
1613 current->bio_list = NULL; /* deactivate */
1615 EXPORT_SYMBOL(generic_make_request);
1618 * submit_bio - submit a bio to the block device layer for I/O
1619 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1620 * @bio: The &struct bio which describes the I/O
1622 * submit_bio() is very similar in purpose to generic_make_request(), and
1623 * uses that function to do most of the work. Both are fairly rough
1624 * interfaces; @bio must be presetup and ready for I/O.
1627 void submit_bio(int rw, struct bio *bio)
1629 int count = bio_sectors(bio);
1631 bio->bi_rw |= rw;
1634 * If it's a regular read/write or a barrier with data attached,
1635 * go through the normal accounting stuff before submission.
1637 if (bio_has_data(bio) && !(rw & REQ_DISCARD)) {
1638 if (rw & WRITE) {
1639 count_vm_events(PGPGOUT, count);
1640 } else {
1641 task_io_account_read(bio->bi_size);
1642 count_vm_events(PGPGIN, count);
1645 if (unlikely(block_dump)) {
1646 char b[BDEVNAME_SIZE];
1647 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1648 current->comm, task_pid_nr(current),
1649 (rw & WRITE) ? "WRITE" : "READ",
1650 (unsigned long long)bio->bi_sector,
1651 bdevname(bio->bi_bdev, b),
1652 count);
1656 generic_make_request(bio);
1658 EXPORT_SYMBOL(submit_bio);
1661 * blk_rq_check_limits - Helper function to check a request for the queue limit
1662 * @q: the queue
1663 * @rq: the request being checked
1665 * Description:
1666 * @rq may have been made based on weaker limitations of upper-level queues
1667 * in request stacking drivers, and it may violate the limitation of @q.
1668 * Since the block layer and the underlying device driver trust @rq
1669 * after it is inserted to @q, it should be checked against @q before
1670 * the insertion using this generic function.
1672 * This function should also be useful for request stacking drivers
1673 * in some cases below, so export this function.
1674 * Request stacking drivers like request-based dm may change the queue
1675 * limits while requests are in the queue (e.g. dm's table swapping).
1676 * Such request stacking drivers should check those requests agaist
1677 * the new queue limits again when they dispatch those requests,
1678 * although such checkings are also done against the old queue limits
1679 * when submitting requests.
1681 int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1683 if (rq->cmd_flags & REQ_DISCARD)
1684 return 0;
1686 if (blk_rq_sectors(rq) > queue_max_sectors(q) ||
1687 blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) {
1688 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1689 return -EIO;
1693 * queue's settings related to segment counting like q->bounce_pfn
1694 * may differ from that of other stacking queues.
1695 * Recalculate it to check the request correctly on this queue's
1696 * limitation.
1698 blk_recalc_rq_segments(rq);
1699 if (rq->nr_phys_segments > queue_max_segments(q)) {
1700 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1701 return -EIO;
1704 return 0;
1706 EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1709 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1710 * @q: the queue to submit the request
1711 * @rq: the request being queued
1713 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1715 unsigned long flags;
1717 if (blk_rq_check_limits(q, rq))
1718 return -EIO;
1720 #ifdef CONFIG_FAIL_MAKE_REQUEST
1721 if (rq->rq_disk && rq->rq_disk->part0.make_it_fail &&
1722 should_fail(&fail_make_request, blk_rq_bytes(rq)))
1723 return -EIO;
1724 #endif
1726 spin_lock_irqsave(q->queue_lock, flags);
1729 * Submitting request must be dequeued before calling this function
1730 * because it will be linked to another request_queue
1732 BUG_ON(blk_queued_rq(rq));
1734 drive_stat_acct(rq, 1);
1735 __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0);
1737 spin_unlock_irqrestore(q->queue_lock, flags);
1739 return 0;
1741 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1744 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1745 * @rq: request to examine
1747 * Description:
1748 * A request could be merge of IOs which require different failure
1749 * handling. This function determines the number of bytes which
1750 * can be failed from the beginning of the request without
1751 * crossing into area which need to be retried further.
1753 * Return:
1754 * The number of bytes to fail.
1756 * Context:
1757 * queue_lock must be held.
1759 unsigned int blk_rq_err_bytes(const struct request *rq)
1761 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1762 unsigned int bytes = 0;
1763 struct bio *bio;
1765 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1766 return blk_rq_bytes(rq);
1769 * Currently the only 'mixing' which can happen is between
1770 * different fastfail types. We can safely fail portions
1771 * which have all the failfast bits that the first one has -
1772 * the ones which are at least as eager to fail as the first
1773 * one.
1775 for (bio = rq->bio; bio; bio = bio->bi_next) {
1776 if ((bio->bi_rw & ff) != ff)
1777 break;
1778 bytes += bio->bi_size;
1781 /* this could lead to infinite loop */
1782 BUG_ON(blk_rq_bytes(rq) && !bytes);
1783 return bytes;
1785 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1787 static void blk_account_io_completion(struct request *req, unsigned int bytes)
1789 if (blk_do_io_stat(req)) {
1790 const int rw = rq_data_dir(req);
1791 struct hd_struct *part;
1792 int cpu;
1794 cpu = part_stat_lock();
1795 part = req->part;
1796 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1797 part_stat_unlock();
1801 static void blk_account_io_done(struct request *req)
1804 * Account IO completion. flush_rq isn't accounted as a
1805 * normal IO on queueing nor completion. Accounting the
1806 * containing request is enough.
1808 if (blk_do_io_stat(req) && req != &req->q->flush_rq) {
1809 unsigned long duration = jiffies - req->start_time;
1810 const int rw = rq_data_dir(req);
1811 struct hd_struct *part;
1812 int cpu;
1814 cpu = part_stat_lock();
1815 part = req->part;
1817 part_stat_inc(cpu, part, ios[rw]);
1818 part_stat_add(cpu, part, ticks[rw], duration);
1819 part_round_stats(cpu, part);
1820 part_dec_in_flight(part, rw);
1822 hd_struct_put(part);
1823 part_stat_unlock();
1828 * blk_peek_request - peek at the top of a request queue
1829 * @q: request queue to peek at
1831 * Description:
1832 * Return the request at the top of @q. The returned request
1833 * should be started using blk_start_request() before LLD starts
1834 * processing it.
1836 * Return:
1837 * Pointer to the request at the top of @q if available. Null
1838 * otherwise.
1840 * Context:
1841 * queue_lock must be held.
1843 struct request *blk_peek_request(struct request_queue *q)
1845 struct request *rq;
1846 int ret;
1848 while ((rq = __elv_next_request(q)) != NULL) {
1849 if (!(rq->cmd_flags & REQ_STARTED)) {
1851 * This is the first time the device driver
1852 * sees this request (possibly after
1853 * requeueing). Notify IO scheduler.
1855 if (rq->cmd_flags & REQ_SORTED)
1856 elv_activate_rq(q, rq);
1859 * just mark as started even if we don't start
1860 * it, a request that has been delayed should
1861 * not be passed by new incoming requests
1863 rq->cmd_flags |= REQ_STARTED;
1864 trace_block_rq_issue(q, rq);
1867 if (!q->boundary_rq || q->boundary_rq == rq) {
1868 q->end_sector = rq_end_sector(rq);
1869 q->boundary_rq = NULL;
1872 if (rq->cmd_flags & REQ_DONTPREP)
1873 break;
1875 if (q->dma_drain_size && blk_rq_bytes(rq)) {
1877 * make sure space for the drain appears we
1878 * know we can do this because max_hw_segments
1879 * has been adjusted to be one fewer than the
1880 * device can handle
1882 rq->nr_phys_segments++;
1885 if (!q->prep_rq_fn)
1886 break;
1888 ret = q->prep_rq_fn(q, rq);
1889 if (ret == BLKPREP_OK) {
1890 break;
1891 } else if (ret == BLKPREP_DEFER) {
1893 * the request may have been (partially) prepped.
1894 * we need to keep this request in the front to
1895 * avoid resource deadlock. REQ_STARTED will
1896 * prevent other fs requests from passing this one.
1898 if (q->dma_drain_size && blk_rq_bytes(rq) &&
1899 !(rq->cmd_flags & REQ_DONTPREP)) {
1901 * remove the space for the drain we added
1902 * so that we don't add it again
1904 --rq->nr_phys_segments;
1907 rq = NULL;
1908 break;
1909 } else if (ret == BLKPREP_KILL) {
1910 rq->cmd_flags |= REQ_QUIET;
1912 * Mark this request as started so we don't trigger
1913 * any debug logic in the end I/O path.
1915 blk_start_request(rq);
1916 __blk_end_request_all(rq, -EIO);
1917 } else {
1918 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
1919 break;
1923 return rq;
1925 EXPORT_SYMBOL(blk_peek_request);
1927 void blk_dequeue_request(struct request *rq)
1929 struct request_queue *q = rq->q;
1931 BUG_ON(list_empty(&rq->queuelist));
1932 BUG_ON(ELV_ON_HASH(rq));
1934 list_del_init(&rq->queuelist);
1937 * the time frame between a request being removed from the lists
1938 * and to it is freed is accounted as io that is in progress at
1939 * the driver side.
1941 if (blk_account_rq(rq)) {
1942 q->in_flight[rq_is_sync(rq)]++;
1943 set_io_start_time_ns(rq);
1948 * blk_start_request - start request processing on the driver
1949 * @req: request to dequeue
1951 * Description:
1952 * Dequeue @req and start timeout timer on it. This hands off the
1953 * request to the driver.
1955 * Block internal functions which don't want to start timer should
1956 * call blk_dequeue_request().
1958 * Context:
1959 * queue_lock must be held.
1961 void blk_start_request(struct request *req)
1963 blk_dequeue_request(req);
1966 * We are now handing the request to the hardware, initialize
1967 * resid_len to full count and add the timeout handler.
1969 req->resid_len = blk_rq_bytes(req);
1970 if (unlikely(blk_bidi_rq(req)))
1971 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
1973 blk_add_timer(req);
1975 EXPORT_SYMBOL(blk_start_request);
1978 * blk_fetch_request - fetch a request from a request queue
1979 * @q: request queue to fetch a request from
1981 * Description:
1982 * Return the request at the top of @q. The request is started on
1983 * return and LLD can start processing it immediately.
1985 * Return:
1986 * Pointer to the request at the top of @q if available. Null
1987 * otherwise.
1989 * Context:
1990 * queue_lock must be held.
1992 struct request *blk_fetch_request(struct request_queue *q)
1994 struct request *rq;
1996 rq = blk_peek_request(q);
1997 if (rq)
1998 blk_start_request(rq);
1999 return rq;
2001 EXPORT_SYMBOL(blk_fetch_request);
2004 * blk_update_request - Special helper function for request stacking drivers
2005 * @req: the request being processed
2006 * @error: %0 for success, < %0 for error
2007 * @nr_bytes: number of bytes to complete @req
2009 * Description:
2010 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2011 * the request structure even if @req doesn't have leftover.
2012 * If @req has leftover, sets it up for the next range of segments.
2014 * This special helper function is only for request stacking drivers
2015 * (e.g. request-based dm) so that they can handle partial completion.
2016 * Actual device drivers should use blk_end_request instead.
2018 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2019 * %false return from this function.
2021 * Return:
2022 * %false - this request doesn't have any more data
2023 * %true - this request has more data
2025 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2027 int total_bytes, bio_nbytes, next_idx = 0;
2028 struct bio *bio;
2030 if (!req->bio)
2031 return false;
2033 trace_block_rq_complete(req->q, req);
2036 * For fs requests, rq is just carrier of independent bio's
2037 * and each partial completion should be handled separately.
2038 * Reset per-request error on each partial completion.
2040 * TODO: tj: This is too subtle. It would be better to let
2041 * low level drivers do what they see fit.
2043 if (req->cmd_type == REQ_TYPE_FS)
2044 req->errors = 0;
2046 if (error && req->cmd_type == REQ_TYPE_FS &&
2047 !(req->cmd_flags & REQ_QUIET)) {
2048 printk(KERN_ERR "end_request: I/O error, dev %s, sector %llu\n",
2049 req->rq_disk ? req->rq_disk->disk_name : "?",
2050 (unsigned long long)blk_rq_pos(req));
2053 blk_account_io_completion(req, nr_bytes);
2055 total_bytes = bio_nbytes = 0;
2056 while ((bio = req->bio) != NULL) {
2057 int nbytes;
2059 if (nr_bytes >= bio->bi_size) {
2060 req->bio = bio->bi_next;
2061 nbytes = bio->bi_size;
2062 req_bio_endio(req, bio, nbytes, error);
2063 next_idx = 0;
2064 bio_nbytes = 0;
2065 } else {
2066 int idx = bio->bi_idx + next_idx;
2068 if (unlikely(idx >= bio->bi_vcnt)) {
2069 blk_dump_rq_flags(req, "__end_that");
2070 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
2071 __func__, idx, bio->bi_vcnt);
2072 break;
2075 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2076 BIO_BUG_ON(nbytes > bio->bi_size);
2079 * not a complete bvec done
2081 if (unlikely(nbytes > nr_bytes)) {
2082 bio_nbytes += nr_bytes;
2083 total_bytes += nr_bytes;
2084 break;
2088 * advance to the next vector
2090 next_idx++;
2091 bio_nbytes += nbytes;
2094 total_bytes += nbytes;
2095 nr_bytes -= nbytes;
2097 bio = req->bio;
2098 if (bio) {
2100 * end more in this run, or just return 'not-done'
2102 if (unlikely(nr_bytes <= 0))
2103 break;
2108 * completely done
2110 if (!req->bio) {
2112 * Reset counters so that the request stacking driver
2113 * can find how many bytes remain in the request
2114 * later.
2116 req->__data_len = 0;
2117 return false;
2121 * if the request wasn't completed, update state
2123 if (bio_nbytes) {
2124 req_bio_endio(req, bio, bio_nbytes, error);
2125 bio->bi_idx += next_idx;
2126 bio_iovec(bio)->bv_offset += nr_bytes;
2127 bio_iovec(bio)->bv_len -= nr_bytes;
2130 req->__data_len -= total_bytes;
2131 req->buffer = bio_data(req->bio);
2133 /* update sector only for requests with clear definition of sector */
2134 if (req->cmd_type == REQ_TYPE_FS || (req->cmd_flags & REQ_DISCARD))
2135 req->__sector += total_bytes >> 9;
2137 /* mixed attributes always follow the first bio */
2138 if (req->cmd_flags & REQ_MIXED_MERGE) {
2139 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2140 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2144 * If total number of sectors is less than the first segment
2145 * size, something has gone terribly wrong.
2147 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2148 printk(KERN_ERR "blk: request botched\n");
2149 req->__data_len = blk_rq_cur_bytes(req);
2152 /* recalculate the number of segments */
2153 blk_recalc_rq_segments(req);
2155 return true;
2157 EXPORT_SYMBOL_GPL(blk_update_request);
2159 static bool blk_update_bidi_request(struct request *rq, int error,
2160 unsigned int nr_bytes,
2161 unsigned int bidi_bytes)
2163 if (blk_update_request(rq, error, nr_bytes))
2164 return true;
2166 /* Bidi request must be completed as a whole */
2167 if (unlikely(blk_bidi_rq(rq)) &&
2168 blk_update_request(rq->next_rq, error, bidi_bytes))
2169 return true;
2171 if (blk_queue_add_random(rq->q))
2172 add_disk_randomness(rq->rq_disk);
2174 return false;
2178 * blk_unprep_request - unprepare a request
2179 * @req: the request
2181 * This function makes a request ready for complete resubmission (or
2182 * completion). It happens only after all error handling is complete,
2183 * so represents the appropriate moment to deallocate any resources
2184 * that were allocated to the request in the prep_rq_fn. The queue
2185 * lock is held when calling this.
2187 void blk_unprep_request(struct request *req)
2189 struct request_queue *q = req->q;
2191 req->cmd_flags &= ~REQ_DONTPREP;
2192 if (q->unprep_rq_fn)
2193 q->unprep_rq_fn(q, req);
2195 EXPORT_SYMBOL_GPL(blk_unprep_request);
2198 * queue lock must be held
2200 static void blk_finish_request(struct request *req, int error)
2202 if (blk_rq_tagged(req))
2203 blk_queue_end_tag(req->q, req);
2205 BUG_ON(blk_queued_rq(req));
2207 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2208 laptop_io_completion(&req->q->backing_dev_info);
2210 blk_delete_timer(req);
2212 if (req->cmd_flags & REQ_DONTPREP)
2213 blk_unprep_request(req);
2216 blk_account_io_done(req);
2218 if (req->end_io)
2219 req->end_io(req, error);
2220 else {
2221 if (blk_bidi_rq(req))
2222 __blk_put_request(req->next_rq->q, req->next_rq);
2224 __blk_put_request(req->q, req);
2229 * blk_end_bidi_request - Complete a bidi request
2230 * @rq: the request to complete
2231 * @error: %0 for success, < %0 for error
2232 * @nr_bytes: number of bytes to complete @rq
2233 * @bidi_bytes: number of bytes to complete @rq->next_rq
2235 * Description:
2236 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2237 * Drivers that supports bidi can safely call this member for any
2238 * type of request, bidi or uni. In the later case @bidi_bytes is
2239 * just ignored.
2241 * Return:
2242 * %false - we are done with this request
2243 * %true - still buffers pending for this request
2245 static bool blk_end_bidi_request(struct request *rq, int error,
2246 unsigned int nr_bytes, unsigned int bidi_bytes)
2248 struct request_queue *q = rq->q;
2249 unsigned long flags;
2251 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2252 return true;
2254 spin_lock_irqsave(q->queue_lock, flags);
2255 blk_finish_request(rq, error);
2256 spin_unlock_irqrestore(q->queue_lock, flags);
2258 return false;
2262 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2263 * @rq: the request to complete
2264 * @error: %0 for success, < %0 for error
2265 * @nr_bytes: number of bytes to complete @rq
2266 * @bidi_bytes: number of bytes to complete @rq->next_rq
2268 * Description:
2269 * Identical to blk_end_bidi_request() except that queue lock is
2270 * assumed to be locked on entry and remains so on return.
2272 * Return:
2273 * %false - we are done with this request
2274 * %true - still buffers pending for this request
2276 static bool __blk_end_bidi_request(struct request *rq, int error,
2277 unsigned int nr_bytes, unsigned int bidi_bytes)
2279 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2280 return true;
2282 blk_finish_request(rq, error);
2284 return false;
2288 * blk_end_request - Helper function for drivers to complete the request.
2289 * @rq: the request being processed
2290 * @error: %0 for success, < %0 for error
2291 * @nr_bytes: number of bytes to complete
2293 * Description:
2294 * Ends I/O on a number of bytes attached to @rq.
2295 * If @rq has leftover, sets it up for the next range of segments.
2297 * Return:
2298 * %false - we are done with this request
2299 * %true - still buffers pending for this request
2301 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2303 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2305 EXPORT_SYMBOL(blk_end_request);
2308 * blk_end_request_all - Helper function for drives to finish the request.
2309 * @rq: the request to finish
2310 * @error: %0 for success, < %0 for error
2312 * Description:
2313 * Completely finish @rq.
2315 void blk_end_request_all(struct request *rq, int error)
2317 bool pending;
2318 unsigned int bidi_bytes = 0;
2320 if (unlikely(blk_bidi_rq(rq)))
2321 bidi_bytes = blk_rq_bytes(rq->next_rq);
2323 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2324 BUG_ON(pending);
2326 EXPORT_SYMBOL(blk_end_request_all);
2329 * blk_end_request_cur - Helper function to finish the current request chunk.
2330 * @rq: the request to finish the current chunk for
2331 * @error: %0 for success, < %0 for error
2333 * Description:
2334 * Complete the current consecutively mapped chunk from @rq.
2336 * Return:
2337 * %false - we are done with this request
2338 * %true - still buffers pending for this request
2340 bool blk_end_request_cur(struct request *rq, int error)
2342 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2344 EXPORT_SYMBOL(blk_end_request_cur);
2347 * blk_end_request_err - Finish a request till the next failure boundary.
2348 * @rq: the request to finish till the next failure boundary for
2349 * @error: must be negative errno
2351 * Description:
2352 * Complete @rq till the next failure boundary.
2354 * Return:
2355 * %false - we are done with this request
2356 * %true - still buffers pending for this request
2358 bool blk_end_request_err(struct request *rq, int error)
2360 WARN_ON(error >= 0);
2361 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2363 EXPORT_SYMBOL_GPL(blk_end_request_err);
2366 * __blk_end_request - Helper function for drivers to complete the request.
2367 * @rq: the request being processed
2368 * @error: %0 for success, < %0 for error
2369 * @nr_bytes: number of bytes to complete
2371 * Description:
2372 * Must be called with queue lock held unlike blk_end_request().
2374 * Return:
2375 * %false - we are done with this request
2376 * %true - still buffers pending for this request
2378 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2380 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2382 EXPORT_SYMBOL(__blk_end_request);
2385 * __blk_end_request_all - Helper function for drives to finish the request.
2386 * @rq: the request to finish
2387 * @error: %0 for success, < %0 for error
2389 * Description:
2390 * Completely finish @rq. Must be called with queue lock held.
2392 void __blk_end_request_all(struct request *rq, int error)
2394 bool pending;
2395 unsigned int bidi_bytes = 0;
2397 if (unlikely(blk_bidi_rq(rq)))
2398 bidi_bytes = blk_rq_bytes(rq->next_rq);
2400 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2401 BUG_ON(pending);
2403 EXPORT_SYMBOL(__blk_end_request_all);
2406 * __blk_end_request_cur - Helper function to finish the current request chunk.
2407 * @rq: the request to finish the current chunk for
2408 * @error: %0 for success, < %0 for error
2410 * Description:
2411 * Complete the current consecutively mapped chunk from @rq. Must
2412 * be called with queue lock held.
2414 * Return:
2415 * %false - we are done with this request
2416 * %true - still buffers pending for this request
2418 bool __blk_end_request_cur(struct request *rq, int error)
2420 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2422 EXPORT_SYMBOL(__blk_end_request_cur);
2425 * __blk_end_request_err - Finish a request till the next failure boundary.
2426 * @rq: the request to finish till the next failure boundary for
2427 * @error: must be negative errno
2429 * Description:
2430 * Complete @rq till the next failure boundary. Must be called
2431 * with queue lock held.
2433 * Return:
2434 * %false - we are done with this request
2435 * %true - still buffers pending for this request
2437 bool __blk_end_request_err(struct request *rq, int error)
2439 WARN_ON(error >= 0);
2440 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2442 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2444 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2445 struct bio *bio)
2447 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2448 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
2450 if (bio_has_data(bio)) {
2451 rq->nr_phys_segments = bio_phys_segments(q, bio);
2452 rq->buffer = bio_data(bio);
2454 rq->__data_len = bio->bi_size;
2455 rq->bio = rq->biotail = bio;
2457 if (bio->bi_bdev)
2458 rq->rq_disk = bio->bi_bdev->bd_disk;
2461 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2463 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2464 * @rq: the request to be flushed
2466 * Description:
2467 * Flush all pages in @rq.
2469 void rq_flush_dcache_pages(struct request *rq)
2471 struct req_iterator iter;
2472 struct bio_vec *bvec;
2474 rq_for_each_segment(bvec, rq, iter)
2475 flush_dcache_page(bvec->bv_page);
2477 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2478 #endif
2481 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2482 * @q : the queue of the device being checked
2484 * Description:
2485 * Check if underlying low-level drivers of a device are busy.
2486 * If the drivers want to export their busy state, they must set own
2487 * exporting function using blk_queue_lld_busy() first.
2489 * Basically, this function is used only by request stacking drivers
2490 * to stop dispatching requests to underlying devices when underlying
2491 * devices are busy. This behavior helps more I/O merging on the queue
2492 * of the request stacking driver and prevents I/O throughput regression
2493 * on burst I/O load.
2495 * Return:
2496 * 0 - Not busy (The request stacking driver should dispatch request)
2497 * 1 - Busy (The request stacking driver should stop dispatching request)
2499 int blk_lld_busy(struct request_queue *q)
2501 if (q->lld_busy_fn)
2502 return q->lld_busy_fn(q);
2504 return 0;
2506 EXPORT_SYMBOL_GPL(blk_lld_busy);
2509 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2510 * @rq: the clone request to be cleaned up
2512 * Description:
2513 * Free all bios in @rq for a cloned request.
2515 void blk_rq_unprep_clone(struct request *rq)
2517 struct bio *bio;
2519 while ((bio = rq->bio) != NULL) {
2520 rq->bio = bio->bi_next;
2522 bio_put(bio);
2525 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2528 * Copy attributes of the original request to the clone request.
2529 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2531 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2533 dst->cpu = src->cpu;
2534 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
2535 dst->cmd_type = src->cmd_type;
2536 dst->__sector = blk_rq_pos(src);
2537 dst->__data_len = blk_rq_bytes(src);
2538 dst->nr_phys_segments = src->nr_phys_segments;
2539 dst->ioprio = src->ioprio;
2540 dst->extra_len = src->extra_len;
2544 * blk_rq_prep_clone - Helper function to setup clone request
2545 * @rq: the request to be setup
2546 * @rq_src: original request to be cloned
2547 * @bs: bio_set that bios for clone are allocated from
2548 * @gfp_mask: memory allocation mask for bio
2549 * @bio_ctr: setup function to be called for each clone bio.
2550 * Returns %0 for success, non %0 for failure.
2551 * @data: private data to be passed to @bio_ctr
2553 * Description:
2554 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2555 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2556 * are not copied, and copying such parts is the caller's responsibility.
2557 * Also, pages which the original bios are pointing to are not copied
2558 * and the cloned bios just point same pages.
2559 * So cloned bios must be completed before original bios, which means
2560 * the caller must complete @rq before @rq_src.
2562 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2563 struct bio_set *bs, gfp_t gfp_mask,
2564 int (*bio_ctr)(struct bio *, struct bio *, void *),
2565 void *data)
2567 struct bio *bio, *bio_src;
2569 if (!bs)
2570 bs = fs_bio_set;
2572 blk_rq_init(NULL, rq);
2574 __rq_for_each_bio(bio_src, rq_src) {
2575 bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs);
2576 if (!bio)
2577 goto free_and_out;
2579 __bio_clone(bio, bio_src);
2581 if (bio_integrity(bio_src) &&
2582 bio_integrity_clone(bio, bio_src, gfp_mask, bs))
2583 goto free_and_out;
2585 if (bio_ctr && bio_ctr(bio, bio_src, data))
2586 goto free_and_out;
2588 if (rq->bio) {
2589 rq->biotail->bi_next = bio;
2590 rq->biotail = bio;
2591 } else
2592 rq->bio = rq->biotail = bio;
2595 __blk_rq_prep_clone(rq, rq_src);
2597 return 0;
2599 free_and_out:
2600 if (bio)
2601 bio_free(bio, bs);
2602 blk_rq_unprep_clone(rq);
2604 return -ENOMEM;
2606 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2608 int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
2610 return queue_work(kblockd_workqueue, work);
2612 EXPORT_SYMBOL(kblockd_schedule_work);
2614 int __init blk_dev_init(void)
2616 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
2617 sizeof(((struct request *)0)->cmd_flags));
2619 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
2620 kblockd_workqueue = alloc_workqueue("kblockd",
2621 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
2622 if (!kblockd_workqueue)
2623 panic("Failed to create kblockd\n");
2625 request_cachep = kmem_cache_create("blkdev_requests",
2626 sizeof(struct request), 0, SLAB_PANIC, NULL);
2628 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
2629 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
2631 return 0;