xhci: STFU: Be quieter during URB submission and completion.
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / block / blk-core.c
blob3fe00a14822a203cacc0529ad0610cf5ce7dae06
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>
32 #define CREATE_TRACE_POINTS
33 #include <trace/events/block.h>
35 #include "blk.h"
37 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
38 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
39 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
41 static int __make_request(struct request_queue *q, struct bio *bio);
44 * For the allocated request tables
46 static struct kmem_cache *request_cachep;
49 * For queue allocation
51 struct kmem_cache *blk_requestq_cachep;
54 * Controlling structure to kblockd
56 static struct workqueue_struct *kblockd_workqueue;
58 static void drive_stat_acct(struct request *rq, int new_io)
60 struct hd_struct *part;
61 int rw = rq_data_dir(rq);
62 int cpu;
64 if (!blk_do_io_stat(rq))
65 return;
67 cpu = part_stat_lock();
69 if (!new_io) {
70 part = rq->part;
71 part_stat_inc(cpu, part, merges[rw]);
72 } else {
73 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
74 if (!hd_struct_try_get(part)) {
76 * The partition is already being removed,
77 * the request will be accounted on the disk only
79 * We take a reference on disk->part0 although that
80 * partition will never be deleted, so we can treat
81 * it as any other partition.
83 part = &rq->rq_disk->part0;
84 hd_struct_get(part);
86 part_round_stats(cpu, part);
87 part_inc_in_flight(part, rw);
88 rq->part = part;
91 part_stat_unlock();
94 void blk_queue_congestion_threshold(struct request_queue *q)
96 int nr;
98 nr = q->nr_requests - (q->nr_requests / 8) + 1;
99 if (nr > q->nr_requests)
100 nr = q->nr_requests;
101 q->nr_congestion_on = nr;
103 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
104 if (nr < 1)
105 nr = 1;
106 q->nr_congestion_off = nr;
110 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
111 * @bdev: device
113 * Locates the passed device's request queue and returns the address of its
114 * backing_dev_info
116 * Will return NULL if the request queue cannot be located.
118 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
120 struct backing_dev_info *ret = NULL;
121 struct request_queue *q = bdev_get_queue(bdev);
123 if (q)
124 ret = &q->backing_dev_info;
125 return ret;
127 EXPORT_SYMBOL(blk_get_backing_dev_info);
129 void blk_rq_init(struct request_queue *q, struct request *rq)
131 memset(rq, 0, sizeof(*rq));
133 INIT_LIST_HEAD(&rq->queuelist);
134 INIT_LIST_HEAD(&rq->timeout_list);
135 rq->cpu = -1;
136 rq->q = q;
137 rq->__sector = (sector_t) -1;
138 INIT_HLIST_NODE(&rq->hash);
139 RB_CLEAR_NODE(&rq->rb_node);
140 rq->cmd = rq->__cmd;
141 rq->cmd_len = BLK_MAX_CDB;
142 rq->tag = -1;
143 rq->ref_count = 1;
144 rq->start_time = jiffies;
145 set_start_time_ns(rq);
146 rq->part = NULL;
148 EXPORT_SYMBOL(blk_rq_init);
150 static void req_bio_endio(struct request *rq, struct bio *bio,
151 unsigned int nbytes, int error)
153 if (error)
154 clear_bit(BIO_UPTODATE, &bio->bi_flags);
155 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
156 error = -EIO;
158 if (unlikely(nbytes > bio->bi_size)) {
159 printk(KERN_ERR "%s: want %u bytes done, %u left\n",
160 __func__, nbytes, bio->bi_size);
161 nbytes = bio->bi_size;
164 if (unlikely(rq->cmd_flags & REQ_QUIET))
165 set_bit(BIO_QUIET, &bio->bi_flags);
167 bio->bi_size -= nbytes;
168 bio->bi_sector += (nbytes >> 9);
170 if (bio_integrity(bio))
171 bio_integrity_advance(bio, nbytes);
173 /* don't actually finish bio if it's part of flush sequence */
174 if (bio->bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
175 bio_endio(bio, error);
178 void blk_dump_rq_flags(struct request *rq, char *msg)
180 int bit;
182 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
183 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
184 rq->cmd_flags);
186 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
187 (unsigned long long)blk_rq_pos(rq),
188 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
189 printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n",
190 rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq));
192 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
193 printk(KERN_INFO " cdb: ");
194 for (bit = 0; bit < BLK_MAX_CDB; bit++)
195 printk("%02x ", rq->cmd[bit]);
196 printk("\n");
199 EXPORT_SYMBOL(blk_dump_rq_flags);
201 static void blk_delay_work(struct work_struct *work)
203 struct request_queue *q;
205 q = container_of(work, struct request_queue, delay_work.work);
206 spin_lock_irq(q->queue_lock);
207 __blk_run_queue(q);
208 spin_unlock_irq(q->queue_lock);
212 * blk_delay_queue - restart queueing after defined interval
213 * @q: The &struct request_queue in question
214 * @msecs: Delay in msecs
216 * Description:
217 * Sometimes queueing needs to be postponed for a little while, to allow
218 * resources to come back. This function will make sure that queueing is
219 * restarted around the specified time.
221 void blk_delay_queue(struct request_queue *q, unsigned long msecs)
223 queue_delayed_work(kblockd_workqueue, &q->delay_work,
224 msecs_to_jiffies(msecs));
226 EXPORT_SYMBOL(blk_delay_queue);
229 * blk_start_queue - restart a previously stopped queue
230 * @q: The &struct request_queue in question
232 * Description:
233 * blk_start_queue() will clear the stop flag on the queue, and call
234 * the request_fn for the queue if it was in a stopped state when
235 * entered. Also see blk_stop_queue(). Queue lock must be held.
237 void blk_start_queue(struct request_queue *q)
239 WARN_ON(!irqs_disabled());
241 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
242 __blk_run_queue(q);
244 EXPORT_SYMBOL(blk_start_queue);
247 * blk_stop_queue - stop a queue
248 * @q: The &struct request_queue in question
250 * Description:
251 * The Linux block layer assumes that a block driver will consume all
252 * entries on the request queue when the request_fn strategy is called.
253 * Often this will not happen, because of hardware limitations (queue
254 * depth settings). If a device driver gets a 'queue full' response,
255 * or if it simply chooses not to queue more I/O at one point, it can
256 * call this function to prevent the request_fn from being called until
257 * the driver has signalled it's ready to go again. This happens by calling
258 * blk_start_queue() to restart queue operations. Queue lock must be held.
260 void blk_stop_queue(struct request_queue *q)
262 __cancel_delayed_work(&q->delay_work);
263 queue_flag_set(QUEUE_FLAG_STOPPED, q);
265 EXPORT_SYMBOL(blk_stop_queue);
268 * blk_sync_queue - cancel any pending callbacks on a queue
269 * @q: the queue
271 * Description:
272 * The block layer may perform asynchronous callback activity
273 * on a queue, such as calling the unplug function after a timeout.
274 * A block device may call blk_sync_queue to ensure that any
275 * such activity is cancelled, thus allowing it to release resources
276 * that the callbacks might use. The caller must already have made sure
277 * that its ->make_request_fn will not re-add plugging prior to calling
278 * this function.
280 * This function does not cancel any asynchronous activity arising
281 * out of elevator or throttling code. That would require elevaotor_exit()
282 * and blk_throtl_exit() to be called with queue lock initialized.
285 void blk_sync_queue(struct request_queue *q)
287 del_timer_sync(&q->timeout);
288 cancel_delayed_work_sync(&q->delay_work);
290 EXPORT_SYMBOL(blk_sync_queue);
293 * __blk_run_queue - run a single device queue
294 * @q: The queue to run
296 * Description:
297 * See @blk_run_queue. This variant must be called with the queue lock
298 * held and interrupts disabled.
300 void __blk_run_queue(struct request_queue *q)
302 if (unlikely(blk_queue_stopped(q)))
303 return;
305 q->request_fn(q);
307 EXPORT_SYMBOL(__blk_run_queue);
310 * blk_run_queue_async - run a single device queue in workqueue context
311 * @q: The queue to run
313 * Description:
314 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
315 * of us.
317 void blk_run_queue_async(struct request_queue *q)
319 if (likely(!blk_queue_stopped(q))) {
320 __cancel_delayed_work(&q->delay_work);
321 queue_delayed_work(kblockd_workqueue, &q->delay_work, 0);
324 EXPORT_SYMBOL(blk_run_queue_async);
327 * blk_run_queue - run a single device queue
328 * @q: The queue to run
330 * Description:
331 * Invoke request handling on this queue, if it has pending work to do.
332 * May be used to restart queueing when a request has completed.
334 void blk_run_queue(struct request_queue *q)
336 unsigned long flags;
338 spin_lock_irqsave(q->queue_lock, flags);
339 __blk_run_queue(q);
340 spin_unlock_irqrestore(q->queue_lock, flags);
342 EXPORT_SYMBOL(blk_run_queue);
344 void blk_put_queue(struct request_queue *q)
346 kobject_put(&q->kobj);
350 * Note: If a driver supplied the queue lock, it should not zap that lock
351 * unexpectedly as some queue cleanup components like elevator_exit() and
352 * blk_throtl_exit() need queue lock.
354 void blk_cleanup_queue(struct request_queue *q)
357 * We know we have process context here, so we can be a little
358 * cautious and ensure that pending block actions on this device
359 * are done before moving on. Going into this function, we should
360 * not have processes doing IO to this device.
362 blk_sync_queue(q);
364 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
365 mutex_lock(&q->sysfs_lock);
366 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
367 mutex_unlock(&q->sysfs_lock);
369 if (q->elevator)
370 elevator_exit(q->elevator);
372 blk_throtl_exit(q);
374 blk_put_queue(q);
376 EXPORT_SYMBOL(blk_cleanup_queue);
378 static int blk_init_free_list(struct request_queue *q)
380 struct request_list *rl = &q->rq;
382 if (unlikely(rl->rq_pool))
383 return 0;
385 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
386 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
387 rl->elvpriv = 0;
388 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
389 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
391 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
392 mempool_free_slab, request_cachep, q->node);
394 if (!rl->rq_pool)
395 return -ENOMEM;
397 return 0;
400 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
402 return blk_alloc_queue_node(gfp_mask, -1);
404 EXPORT_SYMBOL(blk_alloc_queue);
406 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
408 struct request_queue *q;
409 int err;
411 q = kmem_cache_alloc_node(blk_requestq_cachep,
412 gfp_mask | __GFP_ZERO, node_id);
413 if (!q)
414 return NULL;
416 q->backing_dev_info.ra_pages =
417 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
418 q->backing_dev_info.state = 0;
419 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
420 q->backing_dev_info.name = "block";
422 err = bdi_init(&q->backing_dev_info);
423 if (err) {
424 kmem_cache_free(blk_requestq_cachep, q);
425 return NULL;
428 if (blk_throtl_init(q)) {
429 kmem_cache_free(blk_requestq_cachep, q);
430 return NULL;
433 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
434 laptop_mode_timer_fn, (unsigned long) q);
435 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
436 INIT_LIST_HEAD(&q->timeout_list);
437 INIT_LIST_HEAD(&q->flush_queue[0]);
438 INIT_LIST_HEAD(&q->flush_queue[1]);
439 INIT_LIST_HEAD(&q->flush_data_in_flight);
440 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
442 kobject_init(&q->kobj, &blk_queue_ktype);
444 mutex_init(&q->sysfs_lock);
445 spin_lock_init(&q->__queue_lock);
448 * By default initialize queue_lock to internal lock and driver can
449 * override it later if need be.
451 q->queue_lock = &q->__queue_lock;
453 return q;
455 EXPORT_SYMBOL(blk_alloc_queue_node);
458 * blk_init_queue - prepare a request queue for use with a block device
459 * @rfn: The function to be called to process requests that have been
460 * placed on the queue.
461 * @lock: Request queue spin lock
463 * Description:
464 * If a block device wishes to use the standard request handling procedures,
465 * which sorts requests and coalesces adjacent requests, then it must
466 * call blk_init_queue(). The function @rfn will be called when there
467 * are requests on the queue that need to be processed. If the device
468 * supports plugging, then @rfn may not be called immediately when requests
469 * are available on the queue, but may be called at some time later instead.
470 * Plugged queues are generally unplugged when a buffer belonging to one
471 * of the requests on the queue is needed, or due to memory pressure.
473 * @rfn is not required, or even expected, to remove all requests off the
474 * queue, but only as many as it can handle at a time. If it does leave
475 * requests on the queue, it is responsible for arranging that the requests
476 * get dealt with eventually.
478 * The queue spin lock must be held while manipulating the requests on the
479 * request queue; this lock will be taken also from interrupt context, so irq
480 * disabling is needed for it.
482 * Function returns a pointer to the initialized request queue, or %NULL if
483 * it didn't succeed.
485 * Note:
486 * blk_init_queue() must be paired with a blk_cleanup_queue() call
487 * when the block device is deactivated (such as at module unload).
490 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
492 return blk_init_queue_node(rfn, lock, -1);
494 EXPORT_SYMBOL(blk_init_queue);
496 struct request_queue *
497 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
499 struct request_queue *uninit_q, *q;
501 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
502 if (!uninit_q)
503 return NULL;
505 q = blk_init_allocated_queue_node(uninit_q, rfn, lock, node_id);
506 if (!q)
507 blk_cleanup_queue(uninit_q);
509 return q;
511 EXPORT_SYMBOL(blk_init_queue_node);
513 struct request_queue *
514 blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
515 spinlock_t *lock)
517 return blk_init_allocated_queue_node(q, rfn, lock, -1);
519 EXPORT_SYMBOL(blk_init_allocated_queue);
521 struct request_queue *
522 blk_init_allocated_queue_node(struct request_queue *q, request_fn_proc *rfn,
523 spinlock_t *lock, int node_id)
525 if (!q)
526 return NULL;
528 q->node = node_id;
529 if (blk_init_free_list(q))
530 return NULL;
532 q->request_fn = rfn;
533 q->prep_rq_fn = NULL;
534 q->unprep_rq_fn = NULL;
535 q->queue_flags = QUEUE_FLAG_DEFAULT;
537 /* Override internal queue lock with supplied lock pointer */
538 if (lock)
539 q->queue_lock = lock;
542 * This also sets hw/phys segments, boundary and size
544 blk_queue_make_request(q, __make_request);
546 q->sg_reserved_size = INT_MAX;
549 * all done
551 if (!elevator_init(q, NULL)) {
552 blk_queue_congestion_threshold(q);
553 return q;
556 return NULL;
558 EXPORT_SYMBOL(blk_init_allocated_queue_node);
560 int blk_get_queue(struct request_queue *q)
562 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
563 kobject_get(&q->kobj);
564 return 0;
567 return 1;
570 static inline void blk_free_request(struct request_queue *q, struct request *rq)
572 BUG_ON(rq->cmd_flags & REQ_ON_PLUG);
574 if (rq->cmd_flags & REQ_ELVPRIV)
575 elv_put_request(q, rq);
576 mempool_free(rq, q->rq.rq_pool);
579 static struct request *
580 blk_alloc_request(struct request_queue *q, int flags, int priv, gfp_t gfp_mask)
582 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
584 if (!rq)
585 return NULL;
587 blk_rq_init(q, rq);
589 rq->cmd_flags = flags | REQ_ALLOCED;
591 if (priv) {
592 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
593 mempool_free(rq, q->rq.rq_pool);
594 return NULL;
596 rq->cmd_flags |= REQ_ELVPRIV;
599 return rq;
603 * ioc_batching returns true if the ioc is a valid batching request and
604 * should be given priority access to a request.
606 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
608 if (!ioc)
609 return 0;
612 * Make sure the process is able to allocate at least 1 request
613 * even if the batch times out, otherwise we could theoretically
614 * lose wakeups.
616 return ioc->nr_batch_requests == q->nr_batching ||
617 (ioc->nr_batch_requests > 0
618 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
622 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
623 * will cause the process to be a "batcher" on all queues in the system. This
624 * is the behaviour we want though - once it gets a wakeup it should be given
625 * a nice run.
627 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
629 if (!ioc || ioc_batching(q, ioc))
630 return;
632 ioc->nr_batch_requests = q->nr_batching;
633 ioc->last_waited = jiffies;
636 static void __freed_request(struct request_queue *q, int sync)
638 struct request_list *rl = &q->rq;
640 if (rl->count[sync] < queue_congestion_off_threshold(q))
641 blk_clear_queue_congested(q, sync);
643 if (rl->count[sync] + 1 <= q->nr_requests) {
644 if (waitqueue_active(&rl->wait[sync]))
645 wake_up(&rl->wait[sync]);
647 blk_clear_queue_full(q, sync);
652 * A request has just been released. Account for it, update the full and
653 * congestion status, wake up any waiters. Called under q->queue_lock.
655 static void freed_request(struct request_queue *q, int sync, int priv)
657 struct request_list *rl = &q->rq;
659 rl->count[sync]--;
660 if (priv)
661 rl->elvpriv--;
663 __freed_request(q, sync);
665 if (unlikely(rl->starved[sync ^ 1]))
666 __freed_request(q, sync ^ 1);
670 * Determine if elevator data should be initialized when allocating the
671 * request associated with @bio.
673 static bool blk_rq_should_init_elevator(struct bio *bio)
675 if (!bio)
676 return true;
679 * Flush requests do not use the elevator so skip initialization.
680 * This allows a request to share the flush and elevator data.
682 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
683 return false;
685 return true;
689 * Get a free request, queue_lock must be held.
690 * Returns NULL on failure, with queue_lock held.
691 * Returns !NULL on success, with queue_lock *not held*.
693 static struct request *get_request(struct request_queue *q, int rw_flags,
694 struct bio *bio, gfp_t gfp_mask)
696 struct request *rq = NULL;
697 struct request_list *rl = &q->rq;
698 struct io_context *ioc = NULL;
699 const bool is_sync = rw_is_sync(rw_flags) != 0;
700 int may_queue, priv = 0;
702 may_queue = elv_may_queue(q, rw_flags);
703 if (may_queue == ELV_MQUEUE_NO)
704 goto rq_starved;
706 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
707 if (rl->count[is_sync]+1 >= q->nr_requests) {
708 ioc = current_io_context(GFP_ATOMIC, q->node);
710 * The queue will fill after this allocation, so set
711 * it as full, and mark this process as "batching".
712 * This process will be allowed to complete a batch of
713 * requests, others will be blocked.
715 if (!blk_queue_full(q, is_sync)) {
716 ioc_set_batching(q, ioc);
717 blk_set_queue_full(q, is_sync);
718 } else {
719 if (may_queue != ELV_MQUEUE_MUST
720 && !ioc_batching(q, ioc)) {
722 * The queue is full and the allocating
723 * process is not a "batcher", and not
724 * exempted by the IO scheduler
726 goto out;
730 blk_set_queue_congested(q, is_sync);
734 * Only allow batching queuers to allocate up to 50% over the defined
735 * limit of requests, otherwise we could have thousands of requests
736 * allocated with any setting of ->nr_requests
738 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
739 goto out;
741 rl->count[is_sync]++;
742 rl->starved[is_sync] = 0;
744 if (blk_rq_should_init_elevator(bio)) {
745 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
746 if (priv)
747 rl->elvpriv++;
750 if (blk_queue_io_stat(q))
751 rw_flags |= REQ_IO_STAT;
752 spin_unlock_irq(q->queue_lock);
754 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
755 if (unlikely(!rq)) {
757 * Allocation failed presumably due to memory. Undo anything
758 * we might have messed up.
760 * Allocating task should really be put onto the front of the
761 * wait queue, but this is pretty rare.
763 spin_lock_irq(q->queue_lock);
764 freed_request(q, is_sync, priv);
767 * in the very unlikely event that allocation failed and no
768 * requests for this direction was pending, mark us starved
769 * so that freeing of a request in the other direction will
770 * notice us. another possible fix would be to split the
771 * rq mempool into READ and WRITE
773 rq_starved:
774 if (unlikely(rl->count[is_sync] == 0))
775 rl->starved[is_sync] = 1;
777 goto out;
781 * ioc may be NULL here, and ioc_batching will be false. That's
782 * OK, if the queue is under the request limit then requests need
783 * not count toward the nr_batch_requests limit. There will always
784 * be some limit enforced by BLK_BATCH_TIME.
786 if (ioc_batching(q, ioc))
787 ioc->nr_batch_requests--;
789 trace_block_getrq(q, bio, rw_flags & 1);
790 out:
791 return rq;
795 * No available requests for this queue, wait for some requests to become
796 * available.
798 * Called with q->queue_lock held, and returns with it unlocked.
800 static struct request *get_request_wait(struct request_queue *q, int rw_flags,
801 struct bio *bio)
803 const bool is_sync = rw_is_sync(rw_flags) != 0;
804 struct request *rq;
806 rq = get_request(q, rw_flags, bio, GFP_NOIO);
807 while (!rq) {
808 DEFINE_WAIT(wait);
809 struct io_context *ioc;
810 struct request_list *rl = &q->rq;
812 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
813 TASK_UNINTERRUPTIBLE);
815 trace_block_sleeprq(q, bio, rw_flags & 1);
817 spin_unlock_irq(q->queue_lock);
818 io_schedule();
821 * After sleeping, we become a "batching" process and
822 * will be able to allocate at least one request, and
823 * up to a big batch of them for a small period time.
824 * See ioc_batching, ioc_set_batching
826 ioc = current_io_context(GFP_NOIO, q->node);
827 ioc_set_batching(q, ioc);
829 spin_lock_irq(q->queue_lock);
830 finish_wait(&rl->wait[is_sync], &wait);
832 rq = get_request(q, rw_flags, bio, GFP_NOIO);
835 return rq;
838 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
840 struct request *rq;
842 BUG_ON(rw != READ && rw != WRITE);
844 spin_lock_irq(q->queue_lock);
845 if (gfp_mask & __GFP_WAIT) {
846 rq = get_request_wait(q, rw, NULL);
847 } else {
848 rq = get_request(q, rw, NULL, gfp_mask);
849 if (!rq)
850 spin_unlock_irq(q->queue_lock);
852 /* q->queue_lock is unlocked at this point */
854 return rq;
856 EXPORT_SYMBOL(blk_get_request);
859 * blk_make_request - given a bio, allocate a corresponding struct request.
860 * @q: target request queue
861 * @bio: The bio describing the memory mappings that will be submitted for IO.
862 * It may be a chained-bio properly constructed by block/bio layer.
863 * @gfp_mask: gfp flags to be used for memory allocation
865 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
866 * type commands. Where the struct request needs to be farther initialized by
867 * the caller. It is passed a &struct bio, which describes the memory info of
868 * the I/O transfer.
870 * The caller of blk_make_request must make sure that bi_io_vec
871 * are set to describe the memory buffers. That bio_data_dir() will return
872 * the needed direction of the request. (And all bio's in the passed bio-chain
873 * are properly set accordingly)
875 * If called under none-sleepable conditions, mapped bio buffers must not
876 * need bouncing, by calling the appropriate masked or flagged allocator,
877 * suitable for the target device. Otherwise the call to blk_queue_bounce will
878 * BUG.
880 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
881 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
882 * anything but the first bio in the chain. Otherwise you risk waiting for IO
883 * completion of a bio that hasn't been submitted yet, thus resulting in a
884 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
885 * of bio_alloc(), as that avoids the mempool deadlock.
886 * If possible a big IO should be split into smaller parts when allocation
887 * fails. Partial allocation should not be an error, or you risk a live-lock.
889 struct request *blk_make_request(struct request_queue *q, struct bio *bio,
890 gfp_t gfp_mask)
892 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
894 if (unlikely(!rq))
895 return ERR_PTR(-ENOMEM);
897 for_each_bio(bio) {
898 struct bio *bounce_bio = bio;
899 int ret;
901 blk_queue_bounce(q, &bounce_bio);
902 ret = blk_rq_append_bio(q, rq, bounce_bio);
903 if (unlikely(ret)) {
904 blk_put_request(rq);
905 return ERR_PTR(ret);
909 return rq;
911 EXPORT_SYMBOL(blk_make_request);
914 * blk_requeue_request - put a request back on queue
915 * @q: request queue where request should be inserted
916 * @rq: request to be inserted
918 * Description:
919 * Drivers often keep queueing requests until the hardware cannot accept
920 * more, when that condition happens we need to put the request back
921 * on the queue. Must be called with queue lock held.
923 void blk_requeue_request(struct request_queue *q, struct request *rq)
925 blk_delete_timer(rq);
926 blk_clear_rq_complete(rq);
927 trace_block_rq_requeue(q, rq);
929 if (blk_rq_tagged(rq))
930 blk_queue_end_tag(q, rq);
932 BUG_ON(blk_queued_rq(rq));
934 elv_requeue_request(q, rq);
936 EXPORT_SYMBOL(blk_requeue_request);
938 static void add_acct_request(struct request_queue *q, struct request *rq,
939 int where)
941 drive_stat_acct(rq, 1);
942 __elv_add_request(q, rq, where);
946 * blk_insert_request - insert a special request into a request queue
947 * @q: request queue where request should be inserted
948 * @rq: request to be inserted
949 * @at_head: insert request at head or tail of queue
950 * @data: private data
952 * Description:
953 * Many block devices need to execute commands asynchronously, so they don't
954 * block the whole kernel from preemption during request execution. This is
955 * accomplished normally by inserting aritficial requests tagged as
956 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
957 * be scheduled for actual execution by the request queue.
959 * We have the option of inserting the head or the tail of the queue.
960 * Typically we use the tail for new ioctls and so forth. We use the head
961 * of the queue for things like a QUEUE_FULL message from a device, or a
962 * host that is unable to accept a particular command.
964 void blk_insert_request(struct request_queue *q, struct request *rq,
965 int at_head, void *data)
967 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
968 unsigned long flags;
971 * tell I/O scheduler that this isn't a regular read/write (ie it
972 * must not attempt merges on this) and that it acts as a soft
973 * barrier
975 rq->cmd_type = REQ_TYPE_SPECIAL;
977 rq->special = data;
979 spin_lock_irqsave(q->queue_lock, flags);
982 * If command is tagged, release the tag
984 if (blk_rq_tagged(rq))
985 blk_queue_end_tag(q, rq);
987 add_acct_request(q, rq, where);
988 __blk_run_queue(q);
989 spin_unlock_irqrestore(q->queue_lock, flags);
991 EXPORT_SYMBOL(blk_insert_request);
993 static void part_round_stats_single(int cpu, struct hd_struct *part,
994 unsigned long now)
996 if (now == part->stamp)
997 return;
999 if (part_in_flight(part)) {
1000 __part_stat_add(cpu, part, time_in_queue,
1001 part_in_flight(part) * (now - part->stamp));
1002 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1004 part->stamp = now;
1008 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1009 * @cpu: cpu number for stats access
1010 * @part: target partition
1012 * The average IO queue length and utilisation statistics are maintained
1013 * by observing the current state of the queue length and the amount of
1014 * time it has been in this state for.
1016 * Normally, that accounting is done on IO completion, but that can result
1017 * in more than a second's worth of IO being accounted for within any one
1018 * second, leading to >100% utilisation. To deal with that, we call this
1019 * function to do a round-off before returning the results when reading
1020 * /proc/diskstats. This accounts immediately for all queue usage up to
1021 * the current jiffies and restarts the counters again.
1023 void part_round_stats(int cpu, struct hd_struct *part)
1025 unsigned long now = jiffies;
1027 if (part->partno)
1028 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1029 part_round_stats_single(cpu, part, now);
1031 EXPORT_SYMBOL_GPL(part_round_stats);
1034 * queue lock must be held
1036 void __blk_put_request(struct request_queue *q, struct request *req)
1038 if (unlikely(!q))
1039 return;
1040 if (unlikely(--req->ref_count))
1041 return;
1043 elv_completed_request(q, req);
1045 /* this is a bio leak */
1046 WARN_ON(req->bio != NULL);
1049 * Request may not have originated from ll_rw_blk. if not,
1050 * it didn't come out of our reserved rq pools
1052 if (req->cmd_flags & REQ_ALLOCED) {
1053 int is_sync = rq_is_sync(req) != 0;
1054 int priv = req->cmd_flags & REQ_ELVPRIV;
1056 BUG_ON(!list_empty(&req->queuelist));
1057 BUG_ON(!hlist_unhashed(&req->hash));
1059 blk_free_request(q, req);
1060 freed_request(q, is_sync, priv);
1063 EXPORT_SYMBOL_GPL(__blk_put_request);
1065 void blk_put_request(struct request *req)
1067 unsigned long flags;
1068 struct request_queue *q = req->q;
1070 spin_lock_irqsave(q->queue_lock, flags);
1071 __blk_put_request(q, req);
1072 spin_unlock_irqrestore(q->queue_lock, flags);
1074 EXPORT_SYMBOL(blk_put_request);
1077 * blk_add_request_payload - add a payload to a request
1078 * @rq: request to update
1079 * @page: page backing the payload
1080 * @len: length of the payload.
1082 * This allows to later add a payload to an already submitted request by
1083 * a block driver. The driver needs to take care of freeing the payload
1084 * itself.
1086 * Note that this is a quite horrible hack and nothing but handling of
1087 * discard requests should ever use it.
1089 void blk_add_request_payload(struct request *rq, struct page *page,
1090 unsigned int len)
1092 struct bio *bio = rq->bio;
1094 bio->bi_io_vec->bv_page = page;
1095 bio->bi_io_vec->bv_offset = 0;
1096 bio->bi_io_vec->bv_len = len;
1098 bio->bi_size = len;
1099 bio->bi_vcnt = 1;
1100 bio->bi_phys_segments = 1;
1102 rq->__data_len = rq->resid_len = len;
1103 rq->nr_phys_segments = 1;
1104 rq->buffer = bio_data(bio);
1106 EXPORT_SYMBOL_GPL(blk_add_request_payload);
1108 static bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1109 struct bio *bio)
1111 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1114 * Debug stuff, kill later
1116 if (!rq_mergeable(req)) {
1117 blk_dump_rq_flags(req, "back");
1118 return false;
1121 if (!ll_back_merge_fn(q, req, bio))
1122 return false;
1124 trace_block_bio_backmerge(q, bio);
1126 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1127 blk_rq_set_mixed_merge(req);
1129 req->biotail->bi_next = bio;
1130 req->biotail = bio;
1131 req->__data_len += bio->bi_size;
1132 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1134 drive_stat_acct(req, 0);
1135 return true;
1138 static bool bio_attempt_front_merge(struct request_queue *q,
1139 struct request *req, struct bio *bio)
1141 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1142 sector_t sector;
1145 * Debug stuff, kill later
1147 if (!rq_mergeable(req)) {
1148 blk_dump_rq_flags(req, "front");
1149 return false;
1152 if (!ll_front_merge_fn(q, req, bio))
1153 return false;
1155 trace_block_bio_frontmerge(q, bio);
1157 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1158 blk_rq_set_mixed_merge(req);
1160 sector = bio->bi_sector;
1162 bio->bi_next = req->bio;
1163 req->bio = bio;
1166 * may not be valid. if the low level driver said
1167 * it didn't need a bounce buffer then it better
1168 * not touch req->buffer either...
1170 req->buffer = bio_data(bio);
1171 req->__sector = bio->bi_sector;
1172 req->__data_len += bio->bi_size;
1173 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1175 drive_stat_acct(req, 0);
1176 return true;
1180 * Attempts to merge with the plugged list in the current process. Returns
1181 * true if merge was successful, otherwise false.
1183 static bool attempt_plug_merge(struct task_struct *tsk, struct request_queue *q,
1184 struct bio *bio)
1186 struct blk_plug *plug;
1187 struct request *rq;
1188 bool ret = false;
1190 plug = tsk->plug;
1191 if (!plug)
1192 goto out;
1194 list_for_each_entry_reverse(rq, &plug->list, queuelist) {
1195 int el_ret;
1197 if (rq->q != q)
1198 continue;
1200 el_ret = elv_try_merge(rq, bio);
1201 if (el_ret == ELEVATOR_BACK_MERGE) {
1202 ret = bio_attempt_back_merge(q, rq, bio);
1203 if (ret)
1204 break;
1205 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1206 ret = bio_attempt_front_merge(q, rq, bio);
1207 if (ret)
1208 break;
1211 out:
1212 return ret;
1215 void init_request_from_bio(struct request *req, struct bio *bio)
1217 req->cpu = bio->bi_comp_cpu;
1218 req->cmd_type = REQ_TYPE_FS;
1220 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1221 if (bio->bi_rw & REQ_RAHEAD)
1222 req->cmd_flags |= REQ_FAILFAST_MASK;
1224 req->errors = 0;
1225 req->__sector = bio->bi_sector;
1226 req->ioprio = bio_prio(bio);
1227 blk_rq_bio_prep(req->q, req, bio);
1230 static int __make_request(struct request_queue *q, struct bio *bio)
1232 const bool sync = !!(bio->bi_rw & REQ_SYNC);
1233 struct blk_plug *plug;
1234 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1235 struct request *req;
1238 * low level driver can indicate that it wants pages above a
1239 * certain limit bounced to low memory (ie for highmem, or even
1240 * ISA dma in theory)
1242 blk_queue_bounce(q, &bio);
1244 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
1245 spin_lock_irq(q->queue_lock);
1246 where = ELEVATOR_INSERT_FLUSH;
1247 goto get_rq;
1251 * Check if we can merge with the plugged list before grabbing
1252 * any locks.
1254 if (attempt_plug_merge(current, q, bio))
1255 goto out;
1257 spin_lock_irq(q->queue_lock);
1259 el_ret = elv_merge(q, &req, bio);
1260 if (el_ret == ELEVATOR_BACK_MERGE) {
1261 BUG_ON(req->cmd_flags & REQ_ON_PLUG);
1262 if (bio_attempt_back_merge(q, req, bio)) {
1263 if (!attempt_back_merge(q, req))
1264 elv_merged_request(q, req, el_ret);
1265 goto out_unlock;
1267 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1268 BUG_ON(req->cmd_flags & REQ_ON_PLUG);
1269 if (bio_attempt_front_merge(q, req, bio)) {
1270 if (!attempt_front_merge(q, req))
1271 elv_merged_request(q, req, el_ret);
1272 goto out_unlock;
1276 get_rq:
1278 * This sync check and mask will be re-done in init_request_from_bio(),
1279 * but we need to set it earlier to expose the sync flag to the
1280 * rq allocator and io schedulers.
1282 rw_flags = bio_data_dir(bio);
1283 if (sync)
1284 rw_flags |= REQ_SYNC;
1287 * Grab a free request. This is might sleep but can not fail.
1288 * Returns with the queue unlocked.
1290 req = get_request_wait(q, rw_flags, bio);
1293 * After dropping the lock and possibly sleeping here, our request
1294 * may now be mergeable after it had proven unmergeable (above).
1295 * We don't worry about that case for efficiency. It won't happen
1296 * often, and the elevators are able to handle it.
1298 init_request_from_bio(req, bio);
1300 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) ||
1301 bio_flagged(bio, BIO_CPU_AFFINE)) {
1302 req->cpu = blk_cpu_to_group(get_cpu());
1303 put_cpu();
1306 plug = current->plug;
1307 if (plug) {
1309 * If this is the first request added after a plug, fire
1310 * of a plug trace. If others have been added before, check
1311 * if we have multiple devices in this plug. If so, make a
1312 * note to sort the list before dispatch.
1314 if (list_empty(&plug->list))
1315 trace_block_plug(q);
1316 else if (!plug->should_sort) {
1317 struct request *__rq;
1319 __rq = list_entry_rq(plug->list.prev);
1320 if (__rq->q != q)
1321 plug->should_sort = 1;
1324 * Debug flag, kill later
1326 req->cmd_flags |= REQ_ON_PLUG;
1327 list_add_tail(&req->queuelist, &plug->list);
1328 drive_stat_acct(req, 1);
1329 } else {
1330 spin_lock_irq(q->queue_lock);
1331 add_acct_request(q, req, where);
1332 __blk_run_queue(q);
1333 out_unlock:
1334 spin_unlock_irq(q->queue_lock);
1336 out:
1337 return 0;
1341 * If bio->bi_dev is a partition, remap the location
1343 static inline void blk_partition_remap(struct bio *bio)
1345 struct block_device *bdev = bio->bi_bdev;
1347 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1348 struct hd_struct *p = bdev->bd_part;
1350 bio->bi_sector += p->start_sect;
1351 bio->bi_bdev = bdev->bd_contains;
1353 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1354 bdev->bd_dev,
1355 bio->bi_sector - p->start_sect);
1359 static void handle_bad_sector(struct bio *bio)
1361 char b[BDEVNAME_SIZE];
1363 printk(KERN_INFO "attempt to access beyond end of device\n");
1364 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1365 bdevname(bio->bi_bdev, b),
1366 bio->bi_rw,
1367 (unsigned long long)bio->bi_sector + bio_sectors(bio),
1368 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1370 set_bit(BIO_EOF, &bio->bi_flags);
1373 #ifdef CONFIG_FAIL_MAKE_REQUEST
1375 static DECLARE_FAULT_ATTR(fail_make_request);
1377 static int __init setup_fail_make_request(char *str)
1379 return setup_fault_attr(&fail_make_request, str);
1381 __setup("fail_make_request=", setup_fail_make_request);
1383 static int should_fail_request(struct bio *bio)
1385 struct hd_struct *part = bio->bi_bdev->bd_part;
1387 if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail)
1388 return should_fail(&fail_make_request, bio->bi_size);
1390 return 0;
1393 static int __init fail_make_request_debugfs(void)
1395 return init_fault_attr_dentries(&fail_make_request,
1396 "fail_make_request");
1399 late_initcall(fail_make_request_debugfs);
1401 #else /* CONFIG_FAIL_MAKE_REQUEST */
1403 static inline int should_fail_request(struct bio *bio)
1405 return 0;
1408 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1411 * Check whether this bio extends beyond the end of the device.
1413 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1415 sector_t maxsector;
1417 if (!nr_sectors)
1418 return 0;
1420 /* Test device or partition size, when known. */
1421 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1422 if (maxsector) {
1423 sector_t sector = bio->bi_sector;
1425 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1427 * This may well happen - the kernel calls bread()
1428 * without checking the size of the device, e.g., when
1429 * mounting a device.
1431 handle_bad_sector(bio);
1432 return 1;
1436 return 0;
1440 * generic_make_request - hand a buffer to its device driver for I/O
1441 * @bio: The bio describing the location in memory and on the device.
1443 * generic_make_request() is used to make I/O requests of block
1444 * devices. It is passed a &struct bio, which describes the I/O that needs
1445 * to be done.
1447 * generic_make_request() does not return any status. The
1448 * success/failure status of the request, along with notification of
1449 * completion, is delivered asynchronously through the bio->bi_end_io
1450 * function described (one day) else where.
1452 * The caller of generic_make_request must make sure that bi_io_vec
1453 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1454 * set to describe the device address, and the
1455 * bi_end_io and optionally bi_private are set to describe how
1456 * completion notification should be signaled.
1458 * generic_make_request and the drivers it calls may use bi_next if this
1459 * bio happens to be merged with someone else, and may change bi_dev and
1460 * bi_sector for remaps as it sees fit. So the values of these fields
1461 * should NOT be depended on after the call to generic_make_request.
1463 static inline void __generic_make_request(struct bio *bio)
1465 struct request_queue *q;
1466 sector_t old_sector;
1467 int ret, nr_sectors = bio_sectors(bio);
1468 dev_t old_dev;
1469 int err = -EIO;
1471 might_sleep();
1473 if (bio_check_eod(bio, nr_sectors))
1474 goto end_io;
1477 * Resolve the mapping until finished. (drivers are
1478 * still free to implement/resolve their own stacking
1479 * by explicitly returning 0)
1481 * NOTE: we don't repeat the blk_size check for each new device.
1482 * Stacking drivers are expected to know what they are doing.
1484 old_sector = -1;
1485 old_dev = 0;
1486 do {
1487 char b[BDEVNAME_SIZE];
1489 q = bdev_get_queue(bio->bi_bdev);
1490 if (unlikely(!q)) {
1491 printk(KERN_ERR
1492 "generic_make_request: Trying to access "
1493 "nonexistent block-device %s (%Lu)\n",
1494 bdevname(bio->bi_bdev, b),
1495 (long long) bio->bi_sector);
1496 goto end_io;
1499 if (unlikely(!(bio->bi_rw & REQ_DISCARD) &&
1500 nr_sectors > queue_max_hw_sectors(q))) {
1501 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1502 bdevname(bio->bi_bdev, b),
1503 bio_sectors(bio),
1504 queue_max_hw_sectors(q));
1505 goto end_io;
1508 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
1509 goto end_io;
1511 if (should_fail_request(bio))
1512 goto end_io;
1515 * If this device has partitions, remap block n
1516 * of partition p to block n+start(p) of the disk.
1518 blk_partition_remap(bio);
1520 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1521 goto end_io;
1523 if (old_sector != -1)
1524 trace_block_bio_remap(q, bio, old_dev, old_sector);
1526 old_sector = bio->bi_sector;
1527 old_dev = bio->bi_bdev->bd_dev;
1529 if (bio_check_eod(bio, nr_sectors))
1530 goto end_io;
1533 * Filter flush bio's early so that make_request based
1534 * drivers without flush support don't have to worry
1535 * about them.
1537 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1538 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1539 if (!nr_sectors) {
1540 err = 0;
1541 goto end_io;
1545 if ((bio->bi_rw & REQ_DISCARD) &&
1546 (!blk_queue_discard(q) ||
1547 ((bio->bi_rw & REQ_SECURE) &&
1548 !blk_queue_secdiscard(q)))) {
1549 err = -EOPNOTSUPP;
1550 goto end_io;
1553 blk_throtl_bio(q, &bio);
1556 * If bio = NULL, bio has been throttled and will be submitted
1557 * later.
1559 if (!bio)
1560 break;
1562 trace_block_bio_queue(q, bio);
1564 ret = q->make_request_fn(q, bio);
1565 } while (ret);
1567 return;
1569 end_io:
1570 bio_endio(bio, err);
1574 * We only want one ->make_request_fn to be active at a time,
1575 * else stack usage with stacked devices could be a problem.
1576 * So use current->bio_list to keep a list of requests
1577 * submited by a make_request_fn function.
1578 * current->bio_list is also used as a flag to say if
1579 * generic_make_request is currently active in this task or not.
1580 * If it is NULL, then no make_request is active. If it is non-NULL,
1581 * then a make_request is active, and new requests should be added
1582 * at the tail
1584 void generic_make_request(struct bio *bio)
1586 struct bio_list bio_list_on_stack;
1588 if (current->bio_list) {
1589 /* make_request is active */
1590 bio_list_add(current->bio_list, bio);
1591 return;
1593 /* following loop may be a bit non-obvious, and so deserves some
1594 * explanation.
1595 * Before entering the loop, bio->bi_next is NULL (as all callers
1596 * ensure that) so we have a list with a single bio.
1597 * We pretend that we have just taken it off a longer list, so
1598 * we assign bio_list to a pointer to the bio_list_on_stack,
1599 * thus initialising the bio_list of new bios to be
1600 * added. __generic_make_request may indeed add some more bios
1601 * through a recursive call to generic_make_request. If it
1602 * did, we find a non-NULL value in bio_list and re-enter the loop
1603 * from the top. In this case we really did just take the bio
1604 * of the top of the list (no pretending) and so remove it from
1605 * bio_list, and call into __generic_make_request again.
1607 * The loop was structured like this to make only one call to
1608 * __generic_make_request (which is important as it is large and
1609 * inlined) and to keep the structure simple.
1611 BUG_ON(bio->bi_next);
1612 bio_list_init(&bio_list_on_stack);
1613 current->bio_list = &bio_list_on_stack;
1614 do {
1615 __generic_make_request(bio);
1616 bio = bio_list_pop(current->bio_list);
1617 } while (bio);
1618 current->bio_list = NULL; /* deactivate */
1620 EXPORT_SYMBOL(generic_make_request);
1623 * submit_bio - submit a bio to the block device layer for I/O
1624 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1625 * @bio: The &struct bio which describes the I/O
1627 * submit_bio() is very similar in purpose to generic_make_request(), and
1628 * uses that function to do most of the work. Both are fairly rough
1629 * interfaces; @bio must be presetup and ready for I/O.
1632 void submit_bio(int rw, struct bio *bio)
1634 int count = bio_sectors(bio);
1636 bio->bi_rw |= rw;
1639 * If it's a regular read/write or a barrier with data attached,
1640 * go through the normal accounting stuff before submission.
1642 if (bio_has_data(bio) && !(rw & REQ_DISCARD)) {
1643 if (rw & WRITE) {
1644 count_vm_events(PGPGOUT, count);
1645 } else {
1646 task_io_account_read(bio->bi_size);
1647 count_vm_events(PGPGIN, count);
1650 if (unlikely(block_dump)) {
1651 char b[BDEVNAME_SIZE];
1652 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1653 current->comm, task_pid_nr(current),
1654 (rw & WRITE) ? "WRITE" : "READ",
1655 (unsigned long long)bio->bi_sector,
1656 bdevname(bio->bi_bdev, b),
1657 count);
1661 generic_make_request(bio);
1663 EXPORT_SYMBOL(submit_bio);
1666 * blk_rq_check_limits - Helper function to check a request for the queue limit
1667 * @q: the queue
1668 * @rq: the request being checked
1670 * Description:
1671 * @rq may have been made based on weaker limitations of upper-level queues
1672 * in request stacking drivers, and it may violate the limitation of @q.
1673 * Since the block layer and the underlying device driver trust @rq
1674 * after it is inserted to @q, it should be checked against @q before
1675 * the insertion using this generic function.
1677 * This function should also be useful for request stacking drivers
1678 * in some cases below, so export this function.
1679 * Request stacking drivers like request-based dm may change the queue
1680 * limits while requests are in the queue (e.g. dm's table swapping).
1681 * Such request stacking drivers should check those requests agaist
1682 * the new queue limits again when they dispatch those requests,
1683 * although such checkings are also done against the old queue limits
1684 * when submitting requests.
1686 int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1688 if (rq->cmd_flags & REQ_DISCARD)
1689 return 0;
1691 if (blk_rq_sectors(rq) > queue_max_sectors(q) ||
1692 blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) {
1693 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1694 return -EIO;
1698 * queue's settings related to segment counting like q->bounce_pfn
1699 * may differ from that of other stacking queues.
1700 * Recalculate it to check the request correctly on this queue's
1701 * limitation.
1703 blk_recalc_rq_segments(rq);
1704 if (rq->nr_phys_segments > queue_max_segments(q)) {
1705 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1706 return -EIO;
1709 return 0;
1711 EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1714 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1715 * @q: the queue to submit the request
1716 * @rq: the request being queued
1718 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1720 unsigned long flags;
1722 if (blk_rq_check_limits(q, rq))
1723 return -EIO;
1725 #ifdef CONFIG_FAIL_MAKE_REQUEST
1726 if (rq->rq_disk && rq->rq_disk->part0.make_it_fail &&
1727 should_fail(&fail_make_request, blk_rq_bytes(rq)))
1728 return -EIO;
1729 #endif
1731 spin_lock_irqsave(q->queue_lock, flags);
1734 * Submitting request must be dequeued before calling this function
1735 * because it will be linked to another request_queue
1737 BUG_ON(blk_queued_rq(rq));
1739 add_acct_request(q, rq, ELEVATOR_INSERT_BACK);
1740 spin_unlock_irqrestore(q->queue_lock, flags);
1742 return 0;
1744 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1747 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1748 * @rq: request to examine
1750 * Description:
1751 * A request could be merge of IOs which require different failure
1752 * handling. This function determines the number of bytes which
1753 * can be failed from the beginning of the request without
1754 * crossing into area which need to be retried further.
1756 * Return:
1757 * The number of bytes to fail.
1759 * Context:
1760 * queue_lock must be held.
1762 unsigned int blk_rq_err_bytes(const struct request *rq)
1764 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1765 unsigned int bytes = 0;
1766 struct bio *bio;
1768 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1769 return blk_rq_bytes(rq);
1772 * Currently the only 'mixing' which can happen is between
1773 * different fastfail types. We can safely fail portions
1774 * which have all the failfast bits that the first one has -
1775 * the ones which are at least as eager to fail as the first
1776 * one.
1778 for (bio = rq->bio; bio; bio = bio->bi_next) {
1779 if ((bio->bi_rw & ff) != ff)
1780 break;
1781 bytes += bio->bi_size;
1784 /* this could lead to infinite loop */
1785 BUG_ON(blk_rq_bytes(rq) && !bytes);
1786 return bytes;
1788 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1790 static void blk_account_io_completion(struct request *req, unsigned int bytes)
1792 if (blk_do_io_stat(req)) {
1793 const int rw = rq_data_dir(req);
1794 struct hd_struct *part;
1795 int cpu;
1797 cpu = part_stat_lock();
1798 part = req->part;
1799 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1800 part_stat_unlock();
1804 static void blk_account_io_done(struct request *req)
1807 * Account IO completion. flush_rq isn't accounted as a
1808 * normal IO on queueing nor completion. Accounting the
1809 * containing request is enough.
1811 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
1812 unsigned long duration = jiffies - req->start_time;
1813 const int rw = rq_data_dir(req);
1814 struct hd_struct *part;
1815 int cpu;
1817 cpu = part_stat_lock();
1818 part = req->part;
1820 part_stat_inc(cpu, part, ios[rw]);
1821 part_stat_add(cpu, part, ticks[rw], duration);
1822 part_round_stats(cpu, part);
1823 part_dec_in_flight(part, rw);
1825 hd_struct_put(part);
1826 part_stat_unlock();
1831 * blk_peek_request - peek at the top of a request queue
1832 * @q: request queue to peek at
1834 * Description:
1835 * Return the request at the top of @q. The returned request
1836 * should be started using blk_start_request() before LLD starts
1837 * processing it.
1839 * Return:
1840 * Pointer to the request at the top of @q if available. Null
1841 * otherwise.
1843 * Context:
1844 * queue_lock must be held.
1846 struct request *blk_peek_request(struct request_queue *q)
1848 struct request *rq;
1849 int ret;
1851 while ((rq = __elv_next_request(q)) != NULL) {
1852 if (!(rq->cmd_flags & REQ_STARTED)) {
1854 * This is the first time the device driver
1855 * sees this request (possibly after
1856 * requeueing). Notify IO scheduler.
1858 if (rq->cmd_flags & REQ_SORTED)
1859 elv_activate_rq(q, rq);
1862 * just mark as started even if we don't start
1863 * it, a request that has been delayed should
1864 * not be passed by new incoming requests
1866 rq->cmd_flags |= REQ_STARTED;
1867 trace_block_rq_issue(q, rq);
1870 if (!q->boundary_rq || q->boundary_rq == rq) {
1871 q->end_sector = rq_end_sector(rq);
1872 q->boundary_rq = NULL;
1875 if (rq->cmd_flags & REQ_DONTPREP)
1876 break;
1878 if (q->dma_drain_size && blk_rq_bytes(rq)) {
1880 * make sure space for the drain appears we
1881 * know we can do this because max_hw_segments
1882 * has been adjusted to be one fewer than the
1883 * device can handle
1885 rq->nr_phys_segments++;
1888 if (!q->prep_rq_fn)
1889 break;
1891 ret = q->prep_rq_fn(q, rq);
1892 if (ret == BLKPREP_OK) {
1893 break;
1894 } else if (ret == BLKPREP_DEFER) {
1896 * the request may have been (partially) prepped.
1897 * we need to keep this request in the front to
1898 * avoid resource deadlock. REQ_STARTED will
1899 * prevent other fs requests from passing this one.
1901 if (q->dma_drain_size && blk_rq_bytes(rq) &&
1902 !(rq->cmd_flags & REQ_DONTPREP)) {
1904 * remove the space for the drain we added
1905 * so that we don't add it again
1907 --rq->nr_phys_segments;
1910 rq = NULL;
1911 break;
1912 } else if (ret == BLKPREP_KILL) {
1913 rq->cmd_flags |= REQ_QUIET;
1915 * Mark this request as started so we don't trigger
1916 * any debug logic in the end I/O path.
1918 blk_start_request(rq);
1919 __blk_end_request_all(rq, -EIO);
1920 } else {
1921 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
1922 break;
1926 return rq;
1928 EXPORT_SYMBOL(blk_peek_request);
1930 void blk_dequeue_request(struct request *rq)
1932 struct request_queue *q = rq->q;
1934 BUG_ON(list_empty(&rq->queuelist));
1935 BUG_ON(ELV_ON_HASH(rq));
1937 list_del_init(&rq->queuelist);
1940 * the time frame between a request being removed from the lists
1941 * and to it is freed is accounted as io that is in progress at
1942 * the driver side.
1944 if (blk_account_rq(rq)) {
1945 q->in_flight[rq_is_sync(rq)]++;
1946 set_io_start_time_ns(rq);
1951 * blk_start_request - start request processing on the driver
1952 * @req: request to dequeue
1954 * Description:
1955 * Dequeue @req and start timeout timer on it. This hands off the
1956 * request to the driver.
1958 * Block internal functions which don't want to start timer should
1959 * call blk_dequeue_request().
1961 * Context:
1962 * queue_lock must be held.
1964 void blk_start_request(struct request *req)
1966 blk_dequeue_request(req);
1969 * We are now handing the request to the hardware, initialize
1970 * resid_len to full count and add the timeout handler.
1972 req->resid_len = blk_rq_bytes(req);
1973 if (unlikely(blk_bidi_rq(req)))
1974 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
1976 blk_add_timer(req);
1978 EXPORT_SYMBOL(blk_start_request);
1981 * blk_fetch_request - fetch a request from a request queue
1982 * @q: request queue to fetch a request from
1984 * Description:
1985 * Return the request at the top of @q. The request is started on
1986 * return and LLD can start processing it immediately.
1988 * Return:
1989 * Pointer to the request at the top of @q if available. Null
1990 * otherwise.
1992 * Context:
1993 * queue_lock must be held.
1995 struct request *blk_fetch_request(struct request_queue *q)
1997 struct request *rq;
1999 rq = blk_peek_request(q);
2000 if (rq)
2001 blk_start_request(rq);
2002 return rq;
2004 EXPORT_SYMBOL(blk_fetch_request);
2007 * blk_update_request - Special helper function for request stacking drivers
2008 * @req: the request being processed
2009 * @error: %0 for success, < %0 for error
2010 * @nr_bytes: number of bytes to complete @req
2012 * Description:
2013 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2014 * the request structure even if @req doesn't have leftover.
2015 * If @req has leftover, sets it up for the next range of segments.
2017 * This special helper function is only for request stacking drivers
2018 * (e.g. request-based dm) so that they can handle partial completion.
2019 * Actual device drivers should use blk_end_request instead.
2021 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2022 * %false return from this function.
2024 * Return:
2025 * %false - this request doesn't have any more data
2026 * %true - this request has more data
2028 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2030 int total_bytes, bio_nbytes, next_idx = 0;
2031 struct bio *bio;
2033 if (!req->bio)
2034 return false;
2036 trace_block_rq_complete(req->q, req);
2039 * For fs requests, rq is just carrier of independent bio's
2040 * and each partial completion should be handled separately.
2041 * Reset per-request error on each partial completion.
2043 * TODO: tj: This is too subtle. It would be better to let
2044 * low level drivers do what they see fit.
2046 if (req->cmd_type == REQ_TYPE_FS)
2047 req->errors = 0;
2049 if (error && req->cmd_type == REQ_TYPE_FS &&
2050 !(req->cmd_flags & REQ_QUIET)) {
2051 char *error_type;
2053 switch (error) {
2054 case -ENOLINK:
2055 error_type = "recoverable transport";
2056 break;
2057 case -EREMOTEIO:
2058 error_type = "critical target";
2059 break;
2060 case -EBADE:
2061 error_type = "critical nexus";
2062 break;
2063 case -EIO:
2064 default:
2065 error_type = "I/O";
2066 break;
2068 printk(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2069 error_type, req->rq_disk ? req->rq_disk->disk_name : "?",
2070 (unsigned long long)blk_rq_pos(req));
2073 blk_account_io_completion(req, nr_bytes);
2075 total_bytes = bio_nbytes = 0;
2076 while ((bio = req->bio) != NULL) {
2077 int nbytes;
2079 if (nr_bytes >= bio->bi_size) {
2080 req->bio = bio->bi_next;
2081 nbytes = bio->bi_size;
2082 req_bio_endio(req, bio, nbytes, error);
2083 next_idx = 0;
2084 bio_nbytes = 0;
2085 } else {
2086 int idx = bio->bi_idx + next_idx;
2088 if (unlikely(idx >= bio->bi_vcnt)) {
2089 blk_dump_rq_flags(req, "__end_that");
2090 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
2091 __func__, idx, bio->bi_vcnt);
2092 break;
2095 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2096 BIO_BUG_ON(nbytes > bio->bi_size);
2099 * not a complete bvec done
2101 if (unlikely(nbytes > nr_bytes)) {
2102 bio_nbytes += nr_bytes;
2103 total_bytes += nr_bytes;
2104 break;
2108 * advance to the next vector
2110 next_idx++;
2111 bio_nbytes += nbytes;
2114 total_bytes += nbytes;
2115 nr_bytes -= nbytes;
2117 bio = req->bio;
2118 if (bio) {
2120 * end more in this run, or just return 'not-done'
2122 if (unlikely(nr_bytes <= 0))
2123 break;
2128 * completely done
2130 if (!req->bio) {
2132 * Reset counters so that the request stacking driver
2133 * can find how many bytes remain in the request
2134 * later.
2136 req->__data_len = 0;
2137 return false;
2141 * if the request wasn't completed, update state
2143 if (bio_nbytes) {
2144 req_bio_endio(req, bio, bio_nbytes, error);
2145 bio->bi_idx += next_idx;
2146 bio_iovec(bio)->bv_offset += nr_bytes;
2147 bio_iovec(bio)->bv_len -= nr_bytes;
2150 req->__data_len -= total_bytes;
2151 req->buffer = bio_data(req->bio);
2153 /* update sector only for requests with clear definition of sector */
2154 if (req->cmd_type == REQ_TYPE_FS || (req->cmd_flags & REQ_DISCARD))
2155 req->__sector += total_bytes >> 9;
2157 /* mixed attributes always follow the first bio */
2158 if (req->cmd_flags & REQ_MIXED_MERGE) {
2159 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2160 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2164 * If total number of sectors is less than the first segment
2165 * size, something has gone terribly wrong.
2167 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2168 blk_dump_rq_flags(req, "request botched");
2169 req->__data_len = blk_rq_cur_bytes(req);
2172 /* recalculate the number of segments */
2173 blk_recalc_rq_segments(req);
2175 return true;
2177 EXPORT_SYMBOL_GPL(blk_update_request);
2179 static bool blk_update_bidi_request(struct request *rq, int error,
2180 unsigned int nr_bytes,
2181 unsigned int bidi_bytes)
2183 if (blk_update_request(rq, error, nr_bytes))
2184 return true;
2186 /* Bidi request must be completed as a whole */
2187 if (unlikely(blk_bidi_rq(rq)) &&
2188 blk_update_request(rq->next_rq, error, bidi_bytes))
2189 return true;
2191 if (blk_queue_add_random(rq->q))
2192 add_disk_randomness(rq->rq_disk);
2194 return false;
2198 * blk_unprep_request - unprepare a request
2199 * @req: the request
2201 * This function makes a request ready for complete resubmission (or
2202 * completion). It happens only after all error handling is complete,
2203 * so represents the appropriate moment to deallocate any resources
2204 * that were allocated to the request in the prep_rq_fn. The queue
2205 * lock is held when calling this.
2207 void blk_unprep_request(struct request *req)
2209 struct request_queue *q = req->q;
2211 req->cmd_flags &= ~REQ_DONTPREP;
2212 if (q->unprep_rq_fn)
2213 q->unprep_rq_fn(q, req);
2215 EXPORT_SYMBOL_GPL(blk_unprep_request);
2218 * queue lock must be held
2220 static void blk_finish_request(struct request *req, int error)
2222 if (blk_rq_tagged(req))
2223 blk_queue_end_tag(req->q, req);
2225 BUG_ON(blk_queued_rq(req));
2227 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2228 laptop_io_completion(&req->q->backing_dev_info);
2230 blk_delete_timer(req);
2232 if (req->cmd_flags & REQ_DONTPREP)
2233 blk_unprep_request(req);
2236 blk_account_io_done(req);
2238 if (req->end_io)
2239 req->end_io(req, error);
2240 else {
2241 if (blk_bidi_rq(req))
2242 __blk_put_request(req->next_rq->q, req->next_rq);
2244 __blk_put_request(req->q, req);
2249 * blk_end_bidi_request - Complete a bidi request
2250 * @rq: the request to complete
2251 * @error: %0 for success, < %0 for error
2252 * @nr_bytes: number of bytes to complete @rq
2253 * @bidi_bytes: number of bytes to complete @rq->next_rq
2255 * Description:
2256 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2257 * Drivers that supports bidi can safely call this member for any
2258 * type of request, bidi or uni. In the later case @bidi_bytes is
2259 * just ignored.
2261 * Return:
2262 * %false - we are done with this request
2263 * %true - still buffers pending for this request
2265 static bool blk_end_bidi_request(struct request *rq, int error,
2266 unsigned int nr_bytes, unsigned int bidi_bytes)
2268 struct request_queue *q = rq->q;
2269 unsigned long flags;
2271 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2272 return true;
2274 spin_lock_irqsave(q->queue_lock, flags);
2275 blk_finish_request(rq, error);
2276 spin_unlock_irqrestore(q->queue_lock, flags);
2278 return false;
2282 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2283 * @rq: the request to complete
2284 * @error: %0 for success, < %0 for error
2285 * @nr_bytes: number of bytes to complete @rq
2286 * @bidi_bytes: number of bytes to complete @rq->next_rq
2288 * Description:
2289 * Identical to blk_end_bidi_request() except that queue lock is
2290 * assumed to be locked on entry and remains so on return.
2292 * Return:
2293 * %false - we are done with this request
2294 * %true - still buffers pending for this request
2296 static bool __blk_end_bidi_request(struct request *rq, int error,
2297 unsigned int nr_bytes, unsigned int bidi_bytes)
2299 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2300 return true;
2302 blk_finish_request(rq, error);
2304 return false;
2308 * blk_end_request - Helper function for drivers to complete the request.
2309 * @rq: the request being processed
2310 * @error: %0 for success, < %0 for error
2311 * @nr_bytes: number of bytes to complete
2313 * Description:
2314 * Ends I/O on a number of bytes attached to @rq.
2315 * If @rq has leftover, sets it up for the next range of segments.
2317 * Return:
2318 * %false - we are done with this request
2319 * %true - still buffers pending for this request
2321 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2323 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2325 EXPORT_SYMBOL(blk_end_request);
2328 * blk_end_request_all - Helper function for drives to finish the request.
2329 * @rq: the request to finish
2330 * @error: %0 for success, < %0 for error
2332 * Description:
2333 * Completely finish @rq.
2335 void blk_end_request_all(struct request *rq, int error)
2337 bool pending;
2338 unsigned int bidi_bytes = 0;
2340 if (unlikely(blk_bidi_rq(rq)))
2341 bidi_bytes = blk_rq_bytes(rq->next_rq);
2343 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2344 BUG_ON(pending);
2346 EXPORT_SYMBOL(blk_end_request_all);
2349 * blk_end_request_cur - Helper function to finish the current request chunk.
2350 * @rq: the request to finish the current chunk for
2351 * @error: %0 for success, < %0 for error
2353 * Description:
2354 * Complete the current consecutively mapped chunk from @rq.
2356 * Return:
2357 * %false - we are done with this request
2358 * %true - still buffers pending for this request
2360 bool blk_end_request_cur(struct request *rq, int error)
2362 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2364 EXPORT_SYMBOL(blk_end_request_cur);
2367 * blk_end_request_err - Finish a request till the next failure boundary.
2368 * @rq: the request to finish till the next failure boundary for
2369 * @error: must be negative errno
2371 * Description:
2372 * Complete @rq till the next failure boundary.
2374 * Return:
2375 * %false - we are done with this request
2376 * %true - still buffers pending for this request
2378 bool blk_end_request_err(struct request *rq, int error)
2380 WARN_ON(error >= 0);
2381 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2383 EXPORT_SYMBOL_GPL(blk_end_request_err);
2386 * __blk_end_request - Helper function for drivers to complete the request.
2387 * @rq: the request being processed
2388 * @error: %0 for success, < %0 for error
2389 * @nr_bytes: number of bytes to complete
2391 * Description:
2392 * Must be called with queue lock held unlike blk_end_request().
2394 * Return:
2395 * %false - we are done with this request
2396 * %true - still buffers pending for this request
2398 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2400 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2402 EXPORT_SYMBOL(__blk_end_request);
2405 * __blk_end_request_all - Helper function for drives to finish the request.
2406 * @rq: the request to finish
2407 * @error: %0 for success, < %0 for error
2409 * Description:
2410 * Completely finish @rq. Must be called with queue lock held.
2412 void __blk_end_request_all(struct request *rq, int error)
2414 bool pending;
2415 unsigned int bidi_bytes = 0;
2417 if (unlikely(blk_bidi_rq(rq)))
2418 bidi_bytes = blk_rq_bytes(rq->next_rq);
2420 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2421 BUG_ON(pending);
2423 EXPORT_SYMBOL(__blk_end_request_all);
2426 * __blk_end_request_cur - Helper function to finish the current request chunk.
2427 * @rq: the request to finish the current chunk for
2428 * @error: %0 for success, < %0 for error
2430 * Description:
2431 * Complete the current consecutively mapped chunk from @rq. Must
2432 * be called with queue lock held.
2434 * Return:
2435 * %false - we are done with this request
2436 * %true - still buffers pending for this request
2438 bool __blk_end_request_cur(struct request *rq, int error)
2440 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2442 EXPORT_SYMBOL(__blk_end_request_cur);
2445 * __blk_end_request_err - Finish a request till the next failure boundary.
2446 * @rq: the request to finish till the next failure boundary for
2447 * @error: must be negative errno
2449 * Description:
2450 * Complete @rq till the next failure boundary. Must be called
2451 * with queue lock held.
2453 * Return:
2454 * %false - we are done with this request
2455 * %true - still buffers pending for this request
2457 bool __blk_end_request_err(struct request *rq, int error)
2459 WARN_ON(error >= 0);
2460 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2462 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2464 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2465 struct bio *bio)
2467 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2468 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
2470 if (bio_has_data(bio)) {
2471 rq->nr_phys_segments = bio_phys_segments(q, bio);
2472 rq->buffer = bio_data(bio);
2474 rq->__data_len = bio->bi_size;
2475 rq->bio = rq->biotail = bio;
2477 if (bio->bi_bdev)
2478 rq->rq_disk = bio->bi_bdev->bd_disk;
2481 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2483 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2484 * @rq: the request to be flushed
2486 * Description:
2487 * Flush all pages in @rq.
2489 void rq_flush_dcache_pages(struct request *rq)
2491 struct req_iterator iter;
2492 struct bio_vec *bvec;
2494 rq_for_each_segment(bvec, rq, iter)
2495 flush_dcache_page(bvec->bv_page);
2497 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2498 #endif
2501 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2502 * @q : the queue of the device being checked
2504 * Description:
2505 * Check if underlying low-level drivers of a device are busy.
2506 * If the drivers want to export their busy state, they must set own
2507 * exporting function using blk_queue_lld_busy() first.
2509 * Basically, this function is used only by request stacking drivers
2510 * to stop dispatching requests to underlying devices when underlying
2511 * devices are busy. This behavior helps more I/O merging on the queue
2512 * of the request stacking driver and prevents I/O throughput regression
2513 * on burst I/O load.
2515 * Return:
2516 * 0 - Not busy (The request stacking driver should dispatch request)
2517 * 1 - Busy (The request stacking driver should stop dispatching request)
2519 int blk_lld_busy(struct request_queue *q)
2521 if (q->lld_busy_fn)
2522 return q->lld_busy_fn(q);
2524 return 0;
2526 EXPORT_SYMBOL_GPL(blk_lld_busy);
2529 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2530 * @rq: the clone request to be cleaned up
2532 * Description:
2533 * Free all bios in @rq for a cloned request.
2535 void blk_rq_unprep_clone(struct request *rq)
2537 struct bio *bio;
2539 while ((bio = rq->bio) != NULL) {
2540 rq->bio = bio->bi_next;
2542 bio_put(bio);
2545 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2548 * Copy attributes of the original request to the clone request.
2549 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2551 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2553 dst->cpu = src->cpu;
2554 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
2555 dst->cmd_type = src->cmd_type;
2556 dst->__sector = blk_rq_pos(src);
2557 dst->__data_len = blk_rq_bytes(src);
2558 dst->nr_phys_segments = src->nr_phys_segments;
2559 dst->ioprio = src->ioprio;
2560 dst->extra_len = src->extra_len;
2564 * blk_rq_prep_clone - Helper function to setup clone request
2565 * @rq: the request to be setup
2566 * @rq_src: original request to be cloned
2567 * @bs: bio_set that bios for clone are allocated from
2568 * @gfp_mask: memory allocation mask for bio
2569 * @bio_ctr: setup function to be called for each clone bio.
2570 * Returns %0 for success, non %0 for failure.
2571 * @data: private data to be passed to @bio_ctr
2573 * Description:
2574 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2575 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2576 * are not copied, and copying such parts is the caller's responsibility.
2577 * Also, pages which the original bios are pointing to are not copied
2578 * and the cloned bios just point same pages.
2579 * So cloned bios must be completed before original bios, which means
2580 * the caller must complete @rq before @rq_src.
2582 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2583 struct bio_set *bs, gfp_t gfp_mask,
2584 int (*bio_ctr)(struct bio *, struct bio *, void *),
2585 void *data)
2587 struct bio *bio, *bio_src;
2589 if (!bs)
2590 bs = fs_bio_set;
2592 blk_rq_init(NULL, rq);
2594 __rq_for_each_bio(bio_src, rq_src) {
2595 bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs);
2596 if (!bio)
2597 goto free_and_out;
2599 __bio_clone(bio, bio_src);
2601 if (bio_integrity(bio_src) &&
2602 bio_integrity_clone(bio, bio_src, gfp_mask, bs))
2603 goto free_and_out;
2605 if (bio_ctr && bio_ctr(bio, bio_src, data))
2606 goto free_and_out;
2608 if (rq->bio) {
2609 rq->biotail->bi_next = bio;
2610 rq->biotail = bio;
2611 } else
2612 rq->bio = rq->biotail = bio;
2615 __blk_rq_prep_clone(rq, rq_src);
2617 return 0;
2619 free_and_out:
2620 if (bio)
2621 bio_free(bio, bs);
2622 blk_rq_unprep_clone(rq);
2624 return -ENOMEM;
2626 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2628 int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
2630 return queue_work(kblockd_workqueue, work);
2632 EXPORT_SYMBOL(kblockd_schedule_work);
2634 int kblockd_schedule_delayed_work(struct request_queue *q,
2635 struct delayed_work *dwork, unsigned long delay)
2637 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2639 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2641 #define PLUG_MAGIC 0x91827364
2643 void blk_start_plug(struct blk_plug *plug)
2645 struct task_struct *tsk = current;
2647 plug->magic = PLUG_MAGIC;
2648 INIT_LIST_HEAD(&plug->list);
2649 INIT_LIST_HEAD(&plug->cb_list);
2650 plug->should_sort = 0;
2653 * If this is a nested plug, don't actually assign it. It will be
2654 * flushed on its own.
2656 if (!tsk->plug) {
2658 * Store ordering should not be needed here, since a potential
2659 * preempt will imply a full memory barrier
2661 tsk->plug = plug;
2664 EXPORT_SYMBOL(blk_start_plug);
2666 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2668 struct request *rqa = container_of(a, struct request, queuelist);
2669 struct request *rqb = container_of(b, struct request, queuelist);
2671 return !(rqa->q <= rqb->q);
2675 * If 'from_schedule' is true, then postpone the dispatch of requests
2676 * until a safe kblockd context. We due this to avoid accidental big
2677 * additional stack usage in driver dispatch, in places where the originally
2678 * plugger did not intend it.
2680 static void queue_unplugged(struct request_queue *q, unsigned int depth,
2681 bool from_schedule)
2682 __releases(q->queue_lock)
2684 trace_block_unplug(q, depth, !from_schedule);
2687 * If we are punting this to kblockd, then we can safely drop
2688 * the queue_lock before waking kblockd (which needs to take
2689 * this lock).
2691 if (from_schedule) {
2692 spin_unlock(q->queue_lock);
2693 blk_run_queue_async(q);
2694 } else {
2695 __blk_run_queue(q);
2696 spin_unlock(q->queue_lock);
2701 static void flush_plug_callbacks(struct blk_plug *plug)
2703 LIST_HEAD(callbacks);
2705 if (list_empty(&plug->cb_list))
2706 return;
2708 list_splice_init(&plug->cb_list, &callbacks);
2710 while (!list_empty(&callbacks)) {
2711 struct blk_plug_cb *cb = list_first_entry(&callbacks,
2712 struct blk_plug_cb,
2713 list);
2714 list_del(&cb->list);
2715 cb->callback(cb);
2719 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
2721 struct request_queue *q;
2722 unsigned long flags;
2723 struct request *rq;
2724 LIST_HEAD(list);
2725 unsigned int depth;
2727 BUG_ON(plug->magic != PLUG_MAGIC);
2729 flush_plug_callbacks(plug);
2730 if (list_empty(&plug->list))
2731 return;
2733 list_splice_init(&plug->list, &list);
2735 if (plug->should_sort) {
2736 list_sort(NULL, &list, plug_rq_cmp);
2737 plug->should_sort = 0;
2740 q = NULL;
2741 depth = 0;
2744 * Save and disable interrupts here, to avoid doing it for every
2745 * queue lock we have to take.
2747 local_irq_save(flags);
2748 while (!list_empty(&list)) {
2749 rq = list_entry_rq(list.next);
2750 list_del_init(&rq->queuelist);
2751 BUG_ON(!(rq->cmd_flags & REQ_ON_PLUG));
2752 BUG_ON(!rq->q);
2753 if (rq->q != q) {
2755 * This drops the queue lock
2757 if (q)
2758 queue_unplugged(q, depth, from_schedule);
2759 q = rq->q;
2760 depth = 0;
2761 spin_lock(q->queue_lock);
2763 rq->cmd_flags &= ~REQ_ON_PLUG;
2766 * rq is already accounted, so use raw insert
2768 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
2769 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
2770 else
2771 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
2773 depth++;
2777 * This drops the queue lock
2779 if (q)
2780 queue_unplugged(q, depth, from_schedule);
2782 local_irq_restore(flags);
2785 void blk_finish_plug(struct blk_plug *plug)
2787 blk_flush_plug_list(plug, false);
2789 if (plug == current->plug)
2790 current->plug = NULL;
2792 EXPORT_SYMBOL(blk_finish_plug);
2794 int __init blk_dev_init(void)
2796 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
2797 sizeof(((struct request *)0)->cmd_flags));
2799 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
2800 kblockd_workqueue = alloc_workqueue("kblockd",
2801 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
2802 if (!kblockd_workqueue)
2803 panic("Failed to create kblockd\n");
2805 request_cachep = kmem_cache_create("blkdev_requests",
2806 sizeof(struct request), 0, SLAB_PANIC, NULL);
2808 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
2809 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
2811 return 0;