RTC: fix typo in drivers/rtc/rtc-at91sam9.c
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / block / blk-core.c
blob2f4002f79a24b3cf242c870282d96859dc475dc9
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);
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
407 * Description:
408 * See @blk_run_queue. This variant must be called with the queue lock
409 * held and interrupts disabled.
412 void __blk_run_queue(struct request_queue *q)
414 blk_remove_plug(q);
416 if (unlikely(blk_queue_stopped(q)))
417 return;
419 if (elv_queue_empty(q))
420 return;
423 * Only recurse once to avoid overrunning the stack, let the unplug
424 * handling reinvoke the handler shortly if we already got there.
426 if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) {
427 q->request_fn(q);
428 queue_flag_clear(QUEUE_FLAG_REENTER, q);
429 } else {
430 queue_flag_set(QUEUE_FLAG_PLUGGED, q);
431 kblockd_schedule_work(q, &q->unplug_work);
434 EXPORT_SYMBOL(__blk_run_queue);
437 * blk_run_queue - run a single device queue
438 * @q: The queue to run
440 * Description:
441 * Invoke request handling on this queue, if it has pending work to do.
442 * May be used to restart queueing when a request has completed.
444 void blk_run_queue(struct request_queue *q)
446 unsigned long flags;
448 spin_lock_irqsave(q->queue_lock, flags);
449 __blk_run_queue(q);
450 spin_unlock_irqrestore(q->queue_lock, flags);
452 EXPORT_SYMBOL(blk_run_queue);
454 void blk_put_queue(struct request_queue *q)
456 kobject_put(&q->kobj);
459 void blk_cleanup_queue(struct request_queue *q)
462 * We know we have process context here, so we can be a little
463 * cautious and ensure that pending block actions on this device
464 * are done before moving on. Going into this function, we should
465 * not have processes doing IO to this device.
467 blk_sync_queue(q);
469 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
470 mutex_lock(&q->sysfs_lock);
471 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
472 mutex_unlock(&q->sysfs_lock);
474 if (q->elevator)
475 elevator_exit(q->elevator);
477 blk_put_queue(q);
479 EXPORT_SYMBOL(blk_cleanup_queue);
481 static int blk_init_free_list(struct request_queue *q)
483 struct request_list *rl = &q->rq;
485 if (unlikely(rl->rq_pool))
486 return 0;
488 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
489 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
490 rl->elvpriv = 0;
491 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
492 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
494 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
495 mempool_free_slab, request_cachep, q->node);
497 if (!rl->rq_pool)
498 return -ENOMEM;
500 return 0;
503 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
505 return blk_alloc_queue_node(gfp_mask, -1);
507 EXPORT_SYMBOL(blk_alloc_queue);
509 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
511 struct request_queue *q;
512 int err;
514 q = kmem_cache_alloc_node(blk_requestq_cachep,
515 gfp_mask | __GFP_ZERO, node_id);
516 if (!q)
517 return NULL;
519 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
520 q->backing_dev_info.unplug_io_data = q;
521 q->backing_dev_info.ra_pages =
522 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
523 q->backing_dev_info.state = 0;
524 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
525 q->backing_dev_info.name = "block";
527 err = bdi_init(&q->backing_dev_info);
528 if (err) {
529 kmem_cache_free(blk_requestq_cachep, q);
530 return NULL;
533 if (blk_throtl_init(q)) {
534 kmem_cache_free(blk_requestq_cachep, q);
535 return NULL;
538 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
539 laptop_mode_timer_fn, (unsigned long) q);
540 init_timer(&q->unplug_timer);
541 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
542 INIT_LIST_HEAD(&q->timeout_list);
543 INIT_LIST_HEAD(&q->pending_flushes);
544 INIT_WORK(&q->unplug_work, blk_unplug_work);
546 kobject_init(&q->kobj, &blk_queue_ktype);
548 mutex_init(&q->sysfs_lock);
549 spin_lock_init(&q->__queue_lock);
551 return q;
553 EXPORT_SYMBOL(blk_alloc_queue_node);
556 * blk_init_queue - prepare a request queue for use with a block device
557 * @rfn: The function to be called to process requests that have been
558 * placed on the queue.
559 * @lock: Request queue spin lock
561 * Description:
562 * If a block device wishes to use the standard request handling procedures,
563 * which sorts requests and coalesces adjacent requests, then it must
564 * call blk_init_queue(). The function @rfn will be called when there
565 * are requests on the queue that need to be processed. If the device
566 * supports plugging, then @rfn may not be called immediately when requests
567 * are available on the queue, but may be called at some time later instead.
568 * Plugged queues are generally unplugged when a buffer belonging to one
569 * of the requests on the queue is needed, or due to memory pressure.
571 * @rfn is not required, or even expected, to remove all requests off the
572 * queue, but only as many as it can handle at a time. If it does leave
573 * requests on the queue, it is responsible for arranging that the requests
574 * get dealt with eventually.
576 * The queue spin lock must be held while manipulating the requests on the
577 * request queue; this lock will be taken also from interrupt context, so irq
578 * disabling is needed for it.
580 * Function returns a pointer to the initialized request queue, or %NULL if
581 * it didn't succeed.
583 * Note:
584 * blk_init_queue() must be paired with a blk_cleanup_queue() call
585 * when the block device is deactivated (such as at module unload).
588 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
590 return blk_init_queue_node(rfn, lock, -1);
592 EXPORT_SYMBOL(blk_init_queue);
594 struct request_queue *
595 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
597 struct request_queue *uninit_q, *q;
599 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
600 if (!uninit_q)
601 return NULL;
603 q = blk_init_allocated_queue_node(uninit_q, rfn, lock, node_id);
604 if (!q)
605 blk_cleanup_queue(uninit_q);
607 return q;
609 EXPORT_SYMBOL(blk_init_queue_node);
611 struct request_queue *
612 blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
613 spinlock_t *lock)
615 return blk_init_allocated_queue_node(q, rfn, lock, -1);
617 EXPORT_SYMBOL(blk_init_allocated_queue);
619 struct request_queue *
620 blk_init_allocated_queue_node(struct request_queue *q, request_fn_proc *rfn,
621 spinlock_t *lock, int node_id)
623 if (!q)
624 return NULL;
626 q->node = node_id;
627 if (blk_init_free_list(q))
628 return NULL;
630 q->request_fn = rfn;
631 q->prep_rq_fn = NULL;
632 q->unprep_rq_fn = NULL;
633 q->unplug_fn = generic_unplug_device;
634 q->queue_flags = QUEUE_FLAG_DEFAULT;
635 q->queue_lock = lock;
638 * This also sets hw/phys segments, boundary and size
640 blk_queue_make_request(q, __make_request);
642 q->sg_reserved_size = INT_MAX;
645 * all done
647 if (!elevator_init(q, NULL)) {
648 blk_queue_congestion_threshold(q);
649 return q;
652 return NULL;
654 EXPORT_SYMBOL(blk_init_allocated_queue_node);
656 int blk_get_queue(struct request_queue *q)
658 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
659 kobject_get(&q->kobj);
660 return 0;
663 return 1;
666 static inline void blk_free_request(struct request_queue *q, struct request *rq)
668 if (rq->cmd_flags & REQ_ELVPRIV)
669 elv_put_request(q, rq);
670 mempool_free(rq, q->rq.rq_pool);
673 static struct request *
674 blk_alloc_request(struct request_queue *q, int flags, int priv, gfp_t gfp_mask)
676 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
678 if (!rq)
679 return NULL;
681 blk_rq_init(q, rq);
683 rq->cmd_flags = flags | REQ_ALLOCED;
685 if (priv) {
686 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
687 mempool_free(rq, q->rq.rq_pool);
688 return NULL;
690 rq->cmd_flags |= REQ_ELVPRIV;
693 return rq;
697 * ioc_batching returns true if the ioc is a valid batching request and
698 * should be given priority access to a request.
700 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
702 if (!ioc)
703 return 0;
706 * Make sure the process is able to allocate at least 1 request
707 * even if the batch times out, otherwise we could theoretically
708 * lose wakeups.
710 return ioc->nr_batch_requests == q->nr_batching ||
711 (ioc->nr_batch_requests > 0
712 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
716 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
717 * will cause the process to be a "batcher" on all queues in the system. This
718 * is the behaviour we want though - once it gets a wakeup it should be given
719 * a nice run.
721 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
723 if (!ioc || ioc_batching(q, ioc))
724 return;
726 ioc->nr_batch_requests = q->nr_batching;
727 ioc->last_waited = jiffies;
730 static void __freed_request(struct request_queue *q, int sync)
732 struct request_list *rl = &q->rq;
734 if (rl->count[sync] < queue_congestion_off_threshold(q))
735 blk_clear_queue_congested(q, sync);
737 if (rl->count[sync] + 1 <= q->nr_requests) {
738 if (waitqueue_active(&rl->wait[sync]))
739 wake_up(&rl->wait[sync]);
741 blk_clear_queue_full(q, sync);
746 * A request has just been released. Account for it, update the full and
747 * congestion status, wake up any waiters. Called under q->queue_lock.
749 static void freed_request(struct request_queue *q, int sync, int priv)
751 struct request_list *rl = &q->rq;
753 rl->count[sync]--;
754 if (priv)
755 rl->elvpriv--;
757 __freed_request(q, sync);
759 if (unlikely(rl->starved[sync ^ 1]))
760 __freed_request(q, sync ^ 1);
764 * Get a free request, queue_lock must be held.
765 * Returns NULL on failure, with queue_lock held.
766 * Returns !NULL on success, with queue_lock *not held*.
768 static struct request *get_request(struct request_queue *q, int rw_flags,
769 struct bio *bio, gfp_t gfp_mask)
771 struct request *rq = NULL;
772 struct request_list *rl = &q->rq;
773 struct io_context *ioc = NULL;
774 const bool is_sync = rw_is_sync(rw_flags) != 0;
775 int may_queue, priv;
777 may_queue = elv_may_queue(q, rw_flags);
778 if (may_queue == ELV_MQUEUE_NO)
779 goto rq_starved;
781 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
782 if (rl->count[is_sync]+1 >= q->nr_requests) {
783 ioc = current_io_context(GFP_ATOMIC, q->node);
785 * The queue will fill after this allocation, so set
786 * it as full, and mark this process as "batching".
787 * This process will be allowed to complete a batch of
788 * requests, others will be blocked.
790 if (!blk_queue_full(q, is_sync)) {
791 ioc_set_batching(q, ioc);
792 blk_set_queue_full(q, is_sync);
793 } else {
794 if (may_queue != ELV_MQUEUE_MUST
795 && !ioc_batching(q, ioc)) {
797 * The queue is full and the allocating
798 * process is not a "batcher", and not
799 * exempted by the IO scheduler
801 goto out;
805 blk_set_queue_congested(q, is_sync);
809 * Only allow batching queuers to allocate up to 50% over the defined
810 * limit of requests, otherwise we could have thousands of requests
811 * allocated with any setting of ->nr_requests
813 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
814 goto out;
816 rl->count[is_sync]++;
817 rl->starved[is_sync] = 0;
819 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
820 if (priv)
821 rl->elvpriv++;
823 if (blk_queue_io_stat(q))
824 rw_flags |= REQ_IO_STAT;
825 spin_unlock_irq(q->queue_lock);
827 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
828 if (unlikely(!rq)) {
830 * Allocation failed presumably due to memory. Undo anything
831 * we might have messed up.
833 * Allocating task should really be put onto the front of the
834 * wait queue, but this is pretty rare.
836 spin_lock_irq(q->queue_lock);
837 freed_request(q, is_sync, priv);
840 * in the very unlikely event that allocation failed and no
841 * requests for this direction was pending, mark us starved
842 * so that freeing of a request in the other direction will
843 * notice us. another possible fix would be to split the
844 * rq mempool into READ and WRITE
846 rq_starved:
847 if (unlikely(rl->count[is_sync] == 0))
848 rl->starved[is_sync] = 1;
850 goto out;
854 * ioc may be NULL here, and ioc_batching will be false. That's
855 * OK, if the queue is under the request limit then requests need
856 * not count toward the nr_batch_requests limit. There will always
857 * be some limit enforced by BLK_BATCH_TIME.
859 if (ioc_batching(q, ioc))
860 ioc->nr_batch_requests--;
862 trace_block_getrq(q, bio, rw_flags & 1);
863 out:
864 return rq;
868 * No available requests for this queue, unplug the device and wait for some
869 * requests to become available.
871 * Called with q->queue_lock held, and returns with it unlocked.
873 static struct request *get_request_wait(struct request_queue *q, int rw_flags,
874 struct bio *bio)
876 const bool is_sync = rw_is_sync(rw_flags) != 0;
877 struct request *rq;
879 rq = get_request(q, rw_flags, bio, GFP_NOIO);
880 while (!rq) {
881 DEFINE_WAIT(wait);
882 struct io_context *ioc;
883 struct request_list *rl = &q->rq;
885 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
886 TASK_UNINTERRUPTIBLE);
888 trace_block_sleeprq(q, bio, rw_flags & 1);
890 __generic_unplug_device(q);
891 spin_unlock_irq(q->queue_lock);
892 io_schedule();
895 * After sleeping, we become a "batching" process and
896 * will be able to allocate at least one request, and
897 * up to a big batch of them for a small period time.
898 * See ioc_batching, ioc_set_batching
900 ioc = current_io_context(GFP_NOIO, q->node);
901 ioc_set_batching(q, ioc);
903 spin_lock_irq(q->queue_lock);
904 finish_wait(&rl->wait[is_sync], &wait);
906 rq = get_request(q, rw_flags, bio, GFP_NOIO);
909 return rq;
912 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
914 struct request *rq;
916 BUG_ON(rw != READ && rw != WRITE);
918 spin_lock_irq(q->queue_lock);
919 if (gfp_mask & __GFP_WAIT) {
920 rq = get_request_wait(q, rw, NULL);
921 } else {
922 rq = get_request(q, rw, NULL, gfp_mask);
923 if (!rq)
924 spin_unlock_irq(q->queue_lock);
926 /* q->queue_lock is unlocked at this point */
928 return rq;
930 EXPORT_SYMBOL(blk_get_request);
933 * blk_make_request - given a bio, allocate a corresponding struct request.
934 * @q: target request queue
935 * @bio: The bio describing the memory mappings that will be submitted for IO.
936 * It may be a chained-bio properly constructed by block/bio layer.
937 * @gfp_mask: gfp flags to be used for memory allocation
939 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
940 * type commands. Where the struct request needs to be farther initialized by
941 * the caller. It is passed a &struct bio, which describes the memory info of
942 * the I/O transfer.
944 * The caller of blk_make_request must make sure that bi_io_vec
945 * are set to describe the memory buffers. That bio_data_dir() will return
946 * the needed direction of the request. (And all bio's in the passed bio-chain
947 * are properly set accordingly)
949 * If called under none-sleepable conditions, mapped bio buffers must not
950 * need bouncing, by calling the appropriate masked or flagged allocator,
951 * suitable for the target device. Otherwise the call to blk_queue_bounce will
952 * BUG.
954 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
955 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
956 * anything but the first bio in the chain. Otherwise you risk waiting for IO
957 * completion of a bio that hasn't been submitted yet, thus resulting in a
958 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
959 * of bio_alloc(), as that avoids the mempool deadlock.
960 * If possible a big IO should be split into smaller parts when allocation
961 * fails. Partial allocation should not be an error, or you risk a live-lock.
963 struct request *blk_make_request(struct request_queue *q, struct bio *bio,
964 gfp_t gfp_mask)
966 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
968 if (unlikely(!rq))
969 return ERR_PTR(-ENOMEM);
971 for_each_bio(bio) {
972 struct bio *bounce_bio = bio;
973 int ret;
975 blk_queue_bounce(q, &bounce_bio);
976 ret = blk_rq_append_bio(q, rq, bounce_bio);
977 if (unlikely(ret)) {
978 blk_put_request(rq);
979 return ERR_PTR(ret);
983 return rq;
985 EXPORT_SYMBOL(blk_make_request);
988 * blk_requeue_request - put a request back on queue
989 * @q: request queue where request should be inserted
990 * @rq: request to be inserted
992 * Description:
993 * Drivers often keep queueing requests until the hardware cannot accept
994 * more, when that condition happens we need to put the request back
995 * on the queue. Must be called with queue lock held.
997 void blk_requeue_request(struct request_queue *q, struct request *rq)
999 blk_delete_timer(rq);
1000 blk_clear_rq_complete(rq);
1001 trace_block_rq_requeue(q, rq);
1003 if (blk_rq_tagged(rq))
1004 blk_queue_end_tag(q, rq);
1006 BUG_ON(blk_queued_rq(rq));
1008 elv_requeue_request(q, rq);
1010 EXPORT_SYMBOL(blk_requeue_request);
1013 * blk_insert_request - insert a special request into a request queue
1014 * @q: request queue where request should be inserted
1015 * @rq: request to be inserted
1016 * @at_head: insert request at head or tail of queue
1017 * @data: private data
1019 * Description:
1020 * Many block devices need to execute commands asynchronously, so they don't
1021 * block the whole kernel from preemption during request execution. This is
1022 * accomplished normally by inserting aritficial requests tagged as
1023 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
1024 * be scheduled for actual execution by the request queue.
1026 * We have the option of inserting the head or the tail of the queue.
1027 * Typically we use the tail for new ioctls and so forth. We use the head
1028 * of the queue for things like a QUEUE_FULL message from a device, or a
1029 * host that is unable to accept a particular command.
1031 void blk_insert_request(struct request_queue *q, struct request *rq,
1032 int at_head, void *data)
1034 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
1035 unsigned long flags;
1038 * tell I/O scheduler that this isn't a regular read/write (ie it
1039 * must not attempt merges on this) and that it acts as a soft
1040 * barrier
1042 rq->cmd_type = REQ_TYPE_SPECIAL;
1044 rq->special = data;
1046 spin_lock_irqsave(q->queue_lock, flags);
1049 * If command is tagged, release the tag
1051 if (blk_rq_tagged(rq))
1052 blk_queue_end_tag(q, rq);
1054 drive_stat_acct(rq, 1);
1055 __elv_add_request(q, rq, where, 0);
1056 __blk_run_queue(q);
1057 spin_unlock_irqrestore(q->queue_lock, flags);
1059 EXPORT_SYMBOL(blk_insert_request);
1061 static void part_round_stats_single(int cpu, struct hd_struct *part,
1062 unsigned long now)
1064 if (now == part->stamp)
1065 return;
1067 if (part_in_flight(part)) {
1068 __part_stat_add(cpu, part, time_in_queue,
1069 part_in_flight(part) * (now - part->stamp));
1070 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1072 part->stamp = now;
1076 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1077 * @cpu: cpu number for stats access
1078 * @part: target partition
1080 * The average IO queue length and utilisation statistics are maintained
1081 * by observing the current state of the queue length and the amount of
1082 * time it has been in this state for.
1084 * Normally, that accounting is done on IO completion, but that can result
1085 * in more than a second's worth of IO being accounted for within any one
1086 * second, leading to >100% utilisation. To deal with that, we call this
1087 * function to do a round-off before returning the results when reading
1088 * /proc/diskstats. This accounts immediately for all queue usage up to
1089 * the current jiffies and restarts the counters again.
1091 void part_round_stats(int cpu, struct hd_struct *part)
1093 unsigned long now = jiffies;
1095 if (part->partno)
1096 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1097 part_round_stats_single(cpu, part, now);
1099 EXPORT_SYMBOL_GPL(part_round_stats);
1102 * queue lock must be held
1104 void __blk_put_request(struct request_queue *q, struct request *req)
1106 if (unlikely(!q))
1107 return;
1108 if (unlikely(--req->ref_count))
1109 return;
1111 elv_completed_request(q, req);
1113 /* this is a bio leak */
1114 WARN_ON(req->bio != NULL);
1117 * Request may not have originated from ll_rw_blk. if not,
1118 * it didn't come out of our reserved rq pools
1120 if (req->cmd_flags & REQ_ALLOCED) {
1121 int is_sync = rq_is_sync(req) != 0;
1122 int priv = req->cmd_flags & REQ_ELVPRIV;
1124 BUG_ON(!list_empty(&req->queuelist));
1125 BUG_ON(!hlist_unhashed(&req->hash));
1127 blk_free_request(q, req);
1128 freed_request(q, is_sync, priv);
1131 EXPORT_SYMBOL_GPL(__blk_put_request);
1133 void blk_put_request(struct request *req)
1135 unsigned long flags;
1136 struct request_queue *q = req->q;
1138 spin_lock_irqsave(q->queue_lock, flags);
1139 __blk_put_request(q, req);
1140 spin_unlock_irqrestore(q->queue_lock, flags);
1142 EXPORT_SYMBOL(blk_put_request);
1145 * blk_add_request_payload - add a payload to a request
1146 * @rq: request to update
1147 * @page: page backing the payload
1148 * @len: length of the payload.
1150 * This allows to later add a payload to an already submitted request by
1151 * a block driver. The driver needs to take care of freeing the payload
1152 * itself.
1154 * Note that this is a quite horrible hack and nothing but handling of
1155 * discard requests should ever use it.
1157 void blk_add_request_payload(struct request *rq, struct page *page,
1158 unsigned int len)
1160 struct bio *bio = rq->bio;
1162 bio->bi_io_vec->bv_page = page;
1163 bio->bi_io_vec->bv_offset = 0;
1164 bio->bi_io_vec->bv_len = len;
1166 bio->bi_size = len;
1167 bio->bi_vcnt = 1;
1168 bio->bi_phys_segments = 1;
1170 rq->__data_len = rq->resid_len = len;
1171 rq->nr_phys_segments = 1;
1172 rq->buffer = bio_data(bio);
1174 EXPORT_SYMBOL_GPL(blk_add_request_payload);
1176 void init_request_from_bio(struct request *req, struct bio *bio)
1178 req->cpu = bio->bi_comp_cpu;
1179 req->cmd_type = REQ_TYPE_FS;
1181 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1182 if (bio->bi_rw & REQ_RAHEAD)
1183 req->cmd_flags |= REQ_FAILFAST_MASK;
1185 req->errors = 0;
1186 req->__sector = bio->bi_sector;
1187 req->ioprio = bio_prio(bio);
1188 blk_rq_bio_prep(req->q, req, bio);
1192 * Only disabling plugging for non-rotational devices if it does tagging
1193 * as well, otherwise we do need the proper merging
1195 static inline bool queue_should_plug(struct request_queue *q)
1197 return !(blk_queue_nonrot(q) && blk_queue_tagged(q));
1200 static int __make_request(struct request_queue *q, struct bio *bio)
1202 struct request *req;
1203 int el_ret;
1204 unsigned int bytes = bio->bi_size;
1205 const unsigned short prio = bio_prio(bio);
1206 const bool sync = !!(bio->bi_rw & REQ_SYNC);
1207 const bool unplug = !!(bio->bi_rw & REQ_UNPLUG);
1208 const unsigned long ff = bio->bi_rw & REQ_FAILFAST_MASK;
1209 int where = ELEVATOR_INSERT_SORT;
1210 int rw_flags;
1213 * low level driver can indicate that it wants pages above a
1214 * certain limit bounced to low memory (ie for highmem, or even
1215 * ISA dma in theory)
1217 blk_queue_bounce(q, &bio);
1219 spin_lock_irq(q->queue_lock);
1221 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
1222 where = ELEVATOR_INSERT_FRONT;
1223 goto get_rq;
1226 if (elv_queue_empty(q))
1227 goto get_rq;
1229 el_ret = elv_merge(q, &req, bio);
1230 switch (el_ret) {
1231 case ELEVATOR_BACK_MERGE:
1232 BUG_ON(!rq_mergeable(req));
1234 if (!ll_back_merge_fn(q, req, bio))
1235 break;
1237 trace_block_bio_backmerge(q, bio);
1239 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1240 blk_rq_set_mixed_merge(req);
1242 req->biotail->bi_next = bio;
1243 req->biotail = bio;
1244 req->__data_len += bytes;
1245 req->ioprio = ioprio_best(req->ioprio, prio);
1246 if (!blk_rq_cpu_valid(req))
1247 req->cpu = bio->bi_comp_cpu;
1248 drive_stat_acct(req, 0);
1249 elv_bio_merged(q, req, bio);
1250 if (!attempt_back_merge(q, req))
1251 elv_merged_request(q, req, el_ret);
1252 goto out;
1254 case ELEVATOR_FRONT_MERGE:
1255 BUG_ON(!rq_mergeable(req));
1257 if (!ll_front_merge_fn(q, req, bio))
1258 break;
1260 trace_block_bio_frontmerge(q, bio);
1262 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) {
1263 blk_rq_set_mixed_merge(req);
1264 req->cmd_flags &= ~REQ_FAILFAST_MASK;
1265 req->cmd_flags |= ff;
1268 bio->bi_next = req->bio;
1269 req->bio = bio;
1272 * may not be valid. if the low level driver said
1273 * it didn't need a bounce buffer then it better
1274 * not touch req->buffer either...
1276 req->buffer = bio_data(bio);
1277 req->__sector = bio->bi_sector;
1278 req->__data_len += bytes;
1279 req->ioprio = ioprio_best(req->ioprio, prio);
1280 if (!blk_rq_cpu_valid(req))
1281 req->cpu = bio->bi_comp_cpu;
1282 drive_stat_acct(req, 0);
1283 elv_bio_merged(q, req, bio);
1284 if (!attempt_front_merge(q, req))
1285 elv_merged_request(q, req, el_ret);
1286 goto out;
1288 /* ELV_NO_MERGE: elevator says don't/can't merge. */
1289 default:
1293 get_rq:
1295 * This sync check and mask will be re-done in init_request_from_bio(),
1296 * but we need to set it earlier to expose the sync flag to the
1297 * rq allocator and io schedulers.
1299 rw_flags = bio_data_dir(bio);
1300 if (sync)
1301 rw_flags |= REQ_SYNC;
1304 * Grab a free request. This is might sleep but can not fail.
1305 * Returns with the queue unlocked.
1307 req = get_request_wait(q, rw_flags, bio);
1310 * After dropping the lock and possibly sleeping here, our request
1311 * may now be mergeable after it had proven unmergeable (above).
1312 * We don't worry about that case for efficiency. It won't happen
1313 * often, and the elevators are able to handle it.
1315 init_request_from_bio(req, bio);
1317 spin_lock_irq(q->queue_lock);
1318 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) ||
1319 bio_flagged(bio, BIO_CPU_AFFINE))
1320 req->cpu = blk_cpu_to_group(smp_processor_id());
1321 if (queue_should_plug(q) && elv_queue_empty(q))
1322 blk_plug_device(q);
1324 /* insert the request into the elevator */
1325 drive_stat_acct(req, 1);
1326 __elv_add_request(q, req, where, 0);
1327 out:
1328 if (unplug || !queue_should_plug(q))
1329 __generic_unplug_device(q);
1330 spin_unlock_irq(q->queue_lock);
1331 return 0;
1335 * If bio->bi_dev is a partition, remap the location
1337 static inline void blk_partition_remap(struct bio *bio)
1339 struct block_device *bdev = bio->bi_bdev;
1341 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1342 struct hd_struct *p = bdev->bd_part;
1344 bio->bi_sector += p->start_sect;
1345 bio->bi_bdev = bdev->bd_contains;
1347 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1348 bdev->bd_dev,
1349 bio->bi_sector - p->start_sect);
1353 static void handle_bad_sector(struct bio *bio)
1355 char b[BDEVNAME_SIZE];
1357 printk(KERN_INFO "attempt to access beyond end of device\n");
1358 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1359 bdevname(bio->bi_bdev, b),
1360 bio->bi_rw,
1361 (unsigned long long)bio->bi_sector + bio_sectors(bio),
1362 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1364 set_bit(BIO_EOF, &bio->bi_flags);
1367 #ifdef CONFIG_FAIL_MAKE_REQUEST
1369 static DECLARE_FAULT_ATTR(fail_make_request);
1371 static int __init setup_fail_make_request(char *str)
1373 return setup_fault_attr(&fail_make_request, str);
1375 __setup("fail_make_request=", setup_fail_make_request);
1377 static int should_fail_request(struct bio *bio)
1379 struct hd_struct *part = bio->bi_bdev->bd_part;
1381 if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail)
1382 return should_fail(&fail_make_request, bio->bi_size);
1384 return 0;
1387 static int __init fail_make_request_debugfs(void)
1389 return init_fault_attr_dentries(&fail_make_request,
1390 "fail_make_request");
1393 late_initcall(fail_make_request_debugfs);
1395 #else /* CONFIG_FAIL_MAKE_REQUEST */
1397 static inline int should_fail_request(struct bio *bio)
1399 return 0;
1402 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1405 * Check whether this bio extends beyond the end of the device.
1407 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1409 sector_t maxsector;
1411 if (!nr_sectors)
1412 return 0;
1414 /* Test device or partition size, when known. */
1415 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1416 if (maxsector) {
1417 sector_t sector = bio->bi_sector;
1419 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1421 * This may well happen - the kernel calls bread()
1422 * without checking the size of the device, e.g., when
1423 * mounting a device.
1425 handle_bad_sector(bio);
1426 return 1;
1430 return 0;
1434 * generic_make_request - hand a buffer to its device driver for I/O
1435 * @bio: The bio describing the location in memory and on the device.
1437 * generic_make_request() is used to make I/O requests of block
1438 * devices. It is passed a &struct bio, which describes the I/O that needs
1439 * to be done.
1441 * generic_make_request() does not return any status. The
1442 * success/failure status of the request, along with notification of
1443 * completion, is delivered asynchronously through the bio->bi_end_io
1444 * function described (one day) else where.
1446 * The caller of generic_make_request must make sure that bi_io_vec
1447 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1448 * set to describe the device address, and the
1449 * bi_end_io and optionally bi_private are set to describe how
1450 * completion notification should be signaled.
1452 * generic_make_request and the drivers it calls may use bi_next if this
1453 * bio happens to be merged with someone else, and may change bi_dev and
1454 * bi_sector for remaps as it sees fit. So the values of these fields
1455 * should NOT be depended on after the call to generic_make_request.
1457 static inline void __generic_make_request(struct bio *bio)
1459 struct request_queue *q;
1460 sector_t old_sector;
1461 int ret, nr_sectors = bio_sectors(bio);
1462 dev_t old_dev;
1463 int err = -EIO;
1465 might_sleep();
1467 if (bio_check_eod(bio, nr_sectors))
1468 goto end_io;
1471 * Resolve the mapping until finished. (drivers are
1472 * still free to implement/resolve their own stacking
1473 * by explicitly returning 0)
1475 * NOTE: we don't repeat the blk_size check for each new device.
1476 * Stacking drivers are expected to know what they are doing.
1478 old_sector = -1;
1479 old_dev = 0;
1480 do {
1481 char b[BDEVNAME_SIZE];
1483 q = bdev_get_queue(bio->bi_bdev);
1484 if (unlikely(!q)) {
1485 printk(KERN_ERR
1486 "generic_make_request: Trying to access "
1487 "nonexistent block-device %s (%Lu)\n",
1488 bdevname(bio->bi_bdev, b),
1489 (long long) bio->bi_sector);
1490 goto end_io;
1493 if (unlikely(!(bio->bi_rw & REQ_DISCARD) &&
1494 nr_sectors > queue_max_hw_sectors(q))) {
1495 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1496 bdevname(bio->bi_bdev, b),
1497 bio_sectors(bio),
1498 queue_max_hw_sectors(q));
1499 goto end_io;
1502 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
1503 goto end_io;
1505 if (should_fail_request(bio))
1506 goto end_io;
1509 * If this device has partitions, remap block n
1510 * of partition p to block n+start(p) of the disk.
1512 blk_partition_remap(bio);
1514 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1515 goto end_io;
1517 if (old_sector != -1)
1518 trace_block_bio_remap(q, bio, old_dev, old_sector);
1520 old_sector = bio->bi_sector;
1521 old_dev = bio->bi_bdev->bd_dev;
1523 if (bio_check_eod(bio, nr_sectors))
1524 goto end_io;
1527 * Filter flush bio's early so that make_request based
1528 * drivers without flush support don't have to worry
1529 * about them.
1531 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1532 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1533 if (!nr_sectors) {
1534 err = 0;
1535 goto end_io;
1539 if ((bio->bi_rw & REQ_DISCARD) &&
1540 (!blk_queue_discard(q) ||
1541 ((bio->bi_rw & REQ_SECURE) &&
1542 !blk_queue_secdiscard(q)))) {
1543 err = -EOPNOTSUPP;
1544 goto end_io;
1547 blk_throtl_bio(q, &bio);
1550 * If bio = NULL, bio has been throttled and will be submitted
1551 * later.
1553 if (!bio)
1554 break;
1556 trace_block_bio_queue(q, bio);
1558 ret = q->make_request_fn(q, bio);
1559 } while (ret);
1561 return;
1563 end_io:
1564 bio_endio(bio, err);
1568 * We only want one ->make_request_fn to be active at a time,
1569 * else stack usage with stacked devices could be a problem.
1570 * So use current->bio_list to keep a list of requests
1571 * submited by a make_request_fn function.
1572 * current->bio_list is also used as a flag to say if
1573 * generic_make_request is currently active in this task or not.
1574 * If it is NULL, then no make_request is active. If it is non-NULL,
1575 * then a make_request is active, and new requests should be added
1576 * at the tail
1578 void generic_make_request(struct bio *bio)
1580 struct bio_list bio_list_on_stack;
1582 if (current->bio_list) {
1583 /* make_request is active */
1584 bio_list_add(current->bio_list, bio);
1585 return;
1587 /* following loop may be a bit non-obvious, and so deserves some
1588 * explanation.
1589 * Before entering the loop, bio->bi_next is NULL (as all callers
1590 * ensure that) so we have a list with a single bio.
1591 * We pretend that we have just taken it off a longer list, so
1592 * we assign bio_list to a pointer to the bio_list_on_stack,
1593 * thus initialising the bio_list of new bios to be
1594 * added. __generic_make_request may indeed add some more bios
1595 * through a recursive call to generic_make_request. If it
1596 * did, we find a non-NULL value in bio_list and re-enter the loop
1597 * from the top. In this case we really did just take the bio
1598 * of the top of the list (no pretending) and so remove it from
1599 * bio_list, and call into __generic_make_request again.
1601 * The loop was structured like this to make only one call to
1602 * __generic_make_request (which is important as it is large and
1603 * inlined) and to keep the structure simple.
1605 BUG_ON(bio->bi_next);
1606 bio_list_init(&bio_list_on_stack);
1607 current->bio_list = &bio_list_on_stack;
1608 do {
1609 __generic_make_request(bio);
1610 bio = bio_list_pop(current->bio_list);
1611 } while (bio);
1612 current->bio_list = NULL; /* deactivate */
1614 EXPORT_SYMBOL(generic_make_request);
1617 * submit_bio - submit a bio to the block device layer for I/O
1618 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1619 * @bio: The &struct bio which describes the I/O
1621 * submit_bio() is very similar in purpose to generic_make_request(), and
1622 * uses that function to do most of the work. Both are fairly rough
1623 * interfaces; @bio must be presetup and ready for I/O.
1626 void submit_bio(int rw, struct bio *bio)
1628 int count = bio_sectors(bio);
1630 bio->bi_rw |= rw;
1633 * If it's a regular read/write or a barrier with data attached,
1634 * go through the normal accounting stuff before submission.
1636 if (bio_has_data(bio) && !(rw & REQ_DISCARD)) {
1637 if (rw & WRITE) {
1638 count_vm_events(PGPGOUT, count);
1639 } else {
1640 task_io_account_read(bio->bi_size);
1641 count_vm_events(PGPGIN, count);
1644 if (unlikely(block_dump)) {
1645 char b[BDEVNAME_SIZE];
1646 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1647 current->comm, task_pid_nr(current),
1648 (rw & WRITE) ? "WRITE" : "READ",
1649 (unsigned long long)bio->bi_sector,
1650 bdevname(bio->bi_bdev, b),
1651 count);
1655 generic_make_request(bio);
1657 EXPORT_SYMBOL(submit_bio);
1660 * blk_rq_check_limits - Helper function to check a request for the queue limit
1661 * @q: the queue
1662 * @rq: the request being checked
1664 * Description:
1665 * @rq may have been made based on weaker limitations of upper-level queues
1666 * in request stacking drivers, and it may violate the limitation of @q.
1667 * Since the block layer and the underlying device driver trust @rq
1668 * after it is inserted to @q, it should be checked against @q before
1669 * the insertion using this generic function.
1671 * This function should also be useful for request stacking drivers
1672 * in some cases below, so export this function.
1673 * Request stacking drivers like request-based dm may change the queue
1674 * limits while requests are in the queue (e.g. dm's table swapping).
1675 * Such request stacking drivers should check those requests agaist
1676 * the new queue limits again when they dispatch those requests,
1677 * although such checkings are also done against the old queue limits
1678 * when submitting requests.
1680 int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1682 if (rq->cmd_flags & REQ_DISCARD)
1683 return 0;
1685 if (blk_rq_sectors(rq) > queue_max_sectors(q) ||
1686 blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) {
1687 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1688 return -EIO;
1692 * queue's settings related to segment counting like q->bounce_pfn
1693 * may differ from that of other stacking queues.
1694 * Recalculate it to check the request correctly on this queue's
1695 * limitation.
1697 blk_recalc_rq_segments(rq);
1698 if (rq->nr_phys_segments > queue_max_segments(q)) {
1699 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1700 return -EIO;
1703 return 0;
1705 EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1708 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1709 * @q: the queue to submit the request
1710 * @rq: the request being queued
1712 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1714 unsigned long flags;
1716 if (blk_rq_check_limits(q, rq))
1717 return -EIO;
1719 #ifdef CONFIG_FAIL_MAKE_REQUEST
1720 if (rq->rq_disk && rq->rq_disk->part0.make_it_fail &&
1721 should_fail(&fail_make_request, blk_rq_bytes(rq)))
1722 return -EIO;
1723 #endif
1725 spin_lock_irqsave(q->queue_lock, flags);
1728 * Submitting request must be dequeued before calling this function
1729 * because it will be linked to another request_queue
1731 BUG_ON(blk_queued_rq(rq));
1733 drive_stat_acct(rq, 1);
1734 __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0);
1736 spin_unlock_irqrestore(q->queue_lock, flags);
1738 return 0;
1740 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1743 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1744 * @rq: request to examine
1746 * Description:
1747 * A request could be merge of IOs which require different failure
1748 * handling. This function determines the number of bytes which
1749 * can be failed from the beginning of the request without
1750 * crossing into area which need to be retried further.
1752 * Return:
1753 * The number of bytes to fail.
1755 * Context:
1756 * queue_lock must be held.
1758 unsigned int blk_rq_err_bytes(const struct request *rq)
1760 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1761 unsigned int bytes = 0;
1762 struct bio *bio;
1764 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1765 return blk_rq_bytes(rq);
1768 * Currently the only 'mixing' which can happen is between
1769 * different fastfail types. We can safely fail portions
1770 * which have all the failfast bits that the first one has -
1771 * the ones which are at least as eager to fail as the first
1772 * one.
1774 for (bio = rq->bio; bio; bio = bio->bi_next) {
1775 if ((bio->bi_rw & ff) != ff)
1776 break;
1777 bytes += bio->bi_size;
1780 /* this could lead to infinite loop */
1781 BUG_ON(blk_rq_bytes(rq) && !bytes);
1782 return bytes;
1784 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1786 static void blk_account_io_completion(struct request *req, unsigned int bytes)
1788 if (blk_do_io_stat(req)) {
1789 const int rw = rq_data_dir(req);
1790 struct hd_struct *part;
1791 int cpu;
1793 cpu = part_stat_lock();
1794 part = req->part;
1795 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1796 part_stat_unlock();
1800 static void blk_account_io_done(struct request *req)
1803 * Account IO completion. flush_rq isn't accounted as a
1804 * normal IO on queueing nor completion. Accounting the
1805 * containing request is enough.
1807 if (blk_do_io_stat(req) && req != &req->q->flush_rq) {
1808 unsigned long duration = jiffies - req->start_time;
1809 const int rw = rq_data_dir(req);
1810 struct hd_struct *part;
1811 int cpu;
1813 cpu = part_stat_lock();
1814 part = req->part;
1816 part_stat_inc(cpu, part, ios[rw]);
1817 part_stat_add(cpu, part, ticks[rw], duration);
1818 part_round_stats(cpu, part);
1819 part_dec_in_flight(part, rw);
1821 hd_struct_put(part);
1822 part_stat_unlock();
1827 * blk_peek_request - peek at the top of a request queue
1828 * @q: request queue to peek at
1830 * Description:
1831 * Return the request at the top of @q. The returned request
1832 * should be started using blk_start_request() before LLD starts
1833 * processing it.
1835 * Return:
1836 * Pointer to the request at the top of @q if available. Null
1837 * otherwise.
1839 * Context:
1840 * queue_lock must be held.
1842 struct request *blk_peek_request(struct request_queue *q)
1844 struct request *rq;
1845 int ret;
1847 while ((rq = __elv_next_request(q)) != NULL) {
1848 if (!(rq->cmd_flags & REQ_STARTED)) {
1850 * This is the first time the device driver
1851 * sees this request (possibly after
1852 * requeueing). Notify IO scheduler.
1854 if (rq->cmd_flags & REQ_SORTED)
1855 elv_activate_rq(q, rq);
1858 * just mark as started even if we don't start
1859 * it, a request that has been delayed should
1860 * not be passed by new incoming requests
1862 rq->cmd_flags |= REQ_STARTED;
1863 trace_block_rq_issue(q, rq);
1866 if (!q->boundary_rq || q->boundary_rq == rq) {
1867 q->end_sector = rq_end_sector(rq);
1868 q->boundary_rq = NULL;
1871 if (rq->cmd_flags & REQ_DONTPREP)
1872 break;
1874 if (q->dma_drain_size && blk_rq_bytes(rq)) {
1876 * make sure space for the drain appears we
1877 * know we can do this because max_hw_segments
1878 * has been adjusted to be one fewer than the
1879 * device can handle
1881 rq->nr_phys_segments++;
1884 if (!q->prep_rq_fn)
1885 break;
1887 ret = q->prep_rq_fn(q, rq);
1888 if (ret == BLKPREP_OK) {
1889 break;
1890 } else if (ret == BLKPREP_DEFER) {
1892 * the request may have been (partially) prepped.
1893 * we need to keep this request in the front to
1894 * avoid resource deadlock. REQ_STARTED will
1895 * prevent other fs requests from passing this one.
1897 if (q->dma_drain_size && blk_rq_bytes(rq) &&
1898 !(rq->cmd_flags & REQ_DONTPREP)) {
1900 * remove the space for the drain we added
1901 * so that we don't add it again
1903 --rq->nr_phys_segments;
1906 rq = NULL;
1907 break;
1908 } else if (ret == BLKPREP_KILL) {
1909 rq->cmd_flags |= REQ_QUIET;
1911 * Mark this request as started so we don't trigger
1912 * any debug logic in the end I/O path.
1914 blk_start_request(rq);
1915 __blk_end_request_all(rq, -EIO);
1916 } else {
1917 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
1918 break;
1922 return rq;
1924 EXPORT_SYMBOL(blk_peek_request);
1926 void blk_dequeue_request(struct request *rq)
1928 struct request_queue *q = rq->q;
1930 BUG_ON(list_empty(&rq->queuelist));
1931 BUG_ON(ELV_ON_HASH(rq));
1933 list_del_init(&rq->queuelist);
1936 * the time frame between a request being removed from the lists
1937 * and to it is freed is accounted as io that is in progress at
1938 * the driver side.
1940 if (blk_account_rq(rq)) {
1941 q->in_flight[rq_is_sync(rq)]++;
1942 set_io_start_time_ns(rq);
1947 * blk_start_request - start request processing on the driver
1948 * @req: request to dequeue
1950 * Description:
1951 * Dequeue @req and start timeout timer on it. This hands off the
1952 * request to the driver.
1954 * Block internal functions which don't want to start timer should
1955 * call blk_dequeue_request().
1957 * Context:
1958 * queue_lock must be held.
1960 void blk_start_request(struct request *req)
1962 blk_dequeue_request(req);
1965 * We are now handing the request to the hardware, initialize
1966 * resid_len to full count and add the timeout handler.
1968 req->resid_len = blk_rq_bytes(req);
1969 if (unlikely(blk_bidi_rq(req)))
1970 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
1972 blk_add_timer(req);
1974 EXPORT_SYMBOL(blk_start_request);
1977 * blk_fetch_request - fetch a request from a request queue
1978 * @q: request queue to fetch a request from
1980 * Description:
1981 * Return the request at the top of @q. The request is started on
1982 * return and LLD can start processing it immediately.
1984 * Return:
1985 * Pointer to the request at the top of @q if available. Null
1986 * otherwise.
1988 * Context:
1989 * queue_lock must be held.
1991 struct request *blk_fetch_request(struct request_queue *q)
1993 struct request *rq;
1995 rq = blk_peek_request(q);
1996 if (rq)
1997 blk_start_request(rq);
1998 return rq;
2000 EXPORT_SYMBOL(blk_fetch_request);
2003 * blk_update_request - Special helper function for request stacking drivers
2004 * @req: the request being processed
2005 * @error: %0 for success, < %0 for error
2006 * @nr_bytes: number of bytes to complete @req
2008 * Description:
2009 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2010 * the request structure even if @req doesn't have leftover.
2011 * If @req has leftover, sets it up for the next range of segments.
2013 * This special helper function is only for request stacking drivers
2014 * (e.g. request-based dm) so that they can handle partial completion.
2015 * Actual device drivers should use blk_end_request instead.
2017 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2018 * %false return from this function.
2020 * Return:
2021 * %false - this request doesn't have any more data
2022 * %true - this request has more data
2024 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2026 int total_bytes, bio_nbytes, next_idx = 0;
2027 struct bio *bio;
2029 if (!req->bio)
2030 return false;
2032 trace_block_rq_complete(req->q, req);
2035 * For fs requests, rq is just carrier of independent bio's
2036 * and each partial completion should be handled separately.
2037 * Reset per-request error on each partial completion.
2039 * TODO: tj: This is too subtle. It would be better to let
2040 * low level drivers do what they see fit.
2042 if (req->cmd_type == REQ_TYPE_FS)
2043 req->errors = 0;
2045 if (error && req->cmd_type == REQ_TYPE_FS &&
2046 !(req->cmd_flags & REQ_QUIET)) {
2047 printk(KERN_ERR "end_request: I/O error, dev %s, sector %llu\n",
2048 req->rq_disk ? req->rq_disk->disk_name : "?",
2049 (unsigned long long)blk_rq_pos(req));
2052 blk_account_io_completion(req, nr_bytes);
2054 total_bytes = bio_nbytes = 0;
2055 while ((bio = req->bio) != NULL) {
2056 int nbytes;
2058 if (nr_bytes >= bio->bi_size) {
2059 req->bio = bio->bi_next;
2060 nbytes = bio->bi_size;
2061 req_bio_endio(req, bio, nbytes, error);
2062 next_idx = 0;
2063 bio_nbytes = 0;
2064 } else {
2065 int idx = bio->bi_idx + next_idx;
2067 if (unlikely(idx >= bio->bi_vcnt)) {
2068 blk_dump_rq_flags(req, "__end_that");
2069 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
2070 __func__, idx, bio->bi_vcnt);
2071 break;
2074 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2075 BIO_BUG_ON(nbytes > bio->bi_size);
2078 * not a complete bvec done
2080 if (unlikely(nbytes > nr_bytes)) {
2081 bio_nbytes += nr_bytes;
2082 total_bytes += nr_bytes;
2083 break;
2087 * advance to the next vector
2089 next_idx++;
2090 bio_nbytes += nbytes;
2093 total_bytes += nbytes;
2094 nr_bytes -= nbytes;
2096 bio = req->bio;
2097 if (bio) {
2099 * end more in this run, or just return 'not-done'
2101 if (unlikely(nr_bytes <= 0))
2102 break;
2107 * completely done
2109 if (!req->bio) {
2111 * Reset counters so that the request stacking driver
2112 * can find how many bytes remain in the request
2113 * later.
2115 req->__data_len = 0;
2116 return false;
2120 * if the request wasn't completed, update state
2122 if (bio_nbytes) {
2123 req_bio_endio(req, bio, bio_nbytes, error);
2124 bio->bi_idx += next_idx;
2125 bio_iovec(bio)->bv_offset += nr_bytes;
2126 bio_iovec(bio)->bv_len -= nr_bytes;
2129 req->__data_len -= total_bytes;
2130 req->buffer = bio_data(req->bio);
2132 /* update sector only for requests with clear definition of sector */
2133 if (req->cmd_type == REQ_TYPE_FS || (req->cmd_flags & REQ_DISCARD))
2134 req->__sector += total_bytes >> 9;
2136 /* mixed attributes always follow the first bio */
2137 if (req->cmd_flags & REQ_MIXED_MERGE) {
2138 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2139 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2143 * If total number of sectors is less than the first segment
2144 * size, something has gone terribly wrong.
2146 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2147 printk(KERN_ERR "blk: request botched\n");
2148 req->__data_len = blk_rq_cur_bytes(req);
2151 /* recalculate the number of segments */
2152 blk_recalc_rq_segments(req);
2154 return true;
2156 EXPORT_SYMBOL_GPL(blk_update_request);
2158 static bool blk_update_bidi_request(struct request *rq, int error,
2159 unsigned int nr_bytes,
2160 unsigned int bidi_bytes)
2162 if (blk_update_request(rq, error, nr_bytes))
2163 return true;
2165 /* Bidi request must be completed as a whole */
2166 if (unlikely(blk_bidi_rq(rq)) &&
2167 blk_update_request(rq->next_rq, error, bidi_bytes))
2168 return true;
2170 if (blk_queue_add_random(rq->q))
2171 add_disk_randomness(rq->rq_disk);
2173 return false;
2177 * blk_unprep_request - unprepare a request
2178 * @req: the request
2180 * This function makes a request ready for complete resubmission (or
2181 * completion). It happens only after all error handling is complete,
2182 * so represents the appropriate moment to deallocate any resources
2183 * that were allocated to the request in the prep_rq_fn. The queue
2184 * lock is held when calling this.
2186 void blk_unprep_request(struct request *req)
2188 struct request_queue *q = req->q;
2190 req->cmd_flags &= ~REQ_DONTPREP;
2191 if (q->unprep_rq_fn)
2192 q->unprep_rq_fn(q, req);
2194 EXPORT_SYMBOL_GPL(blk_unprep_request);
2197 * queue lock must be held
2199 static void blk_finish_request(struct request *req, int error)
2201 if (blk_rq_tagged(req))
2202 blk_queue_end_tag(req->q, req);
2204 BUG_ON(blk_queued_rq(req));
2206 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2207 laptop_io_completion(&req->q->backing_dev_info);
2209 blk_delete_timer(req);
2211 if (req->cmd_flags & REQ_DONTPREP)
2212 blk_unprep_request(req);
2215 blk_account_io_done(req);
2217 if (req->end_io)
2218 req->end_io(req, error);
2219 else {
2220 if (blk_bidi_rq(req))
2221 __blk_put_request(req->next_rq->q, req->next_rq);
2223 __blk_put_request(req->q, req);
2228 * blk_end_bidi_request - Complete a bidi request
2229 * @rq: the request to complete
2230 * @error: %0 for success, < %0 for error
2231 * @nr_bytes: number of bytes to complete @rq
2232 * @bidi_bytes: number of bytes to complete @rq->next_rq
2234 * Description:
2235 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2236 * Drivers that supports bidi can safely call this member for any
2237 * type of request, bidi or uni. In the later case @bidi_bytes is
2238 * just ignored.
2240 * Return:
2241 * %false - we are done with this request
2242 * %true - still buffers pending for this request
2244 static bool blk_end_bidi_request(struct request *rq, int error,
2245 unsigned int nr_bytes, unsigned int bidi_bytes)
2247 struct request_queue *q = rq->q;
2248 unsigned long flags;
2250 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2251 return true;
2253 spin_lock_irqsave(q->queue_lock, flags);
2254 blk_finish_request(rq, error);
2255 spin_unlock_irqrestore(q->queue_lock, flags);
2257 return false;
2261 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2262 * @rq: the request to complete
2263 * @error: %0 for success, < %0 for error
2264 * @nr_bytes: number of bytes to complete @rq
2265 * @bidi_bytes: number of bytes to complete @rq->next_rq
2267 * Description:
2268 * Identical to blk_end_bidi_request() except that queue lock is
2269 * assumed to be locked on entry and remains so on return.
2271 * Return:
2272 * %false - we are done with this request
2273 * %true - still buffers pending for this request
2275 static bool __blk_end_bidi_request(struct request *rq, int error,
2276 unsigned int nr_bytes, unsigned int bidi_bytes)
2278 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2279 return true;
2281 blk_finish_request(rq, error);
2283 return false;
2287 * blk_end_request - Helper function for drivers to complete the request.
2288 * @rq: the request being processed
2289 * @error: %0 for success, < %0 for error
2290 * @nr_bytes: number of bytes to complete
2292 * Description:
2293 * Ends I/O on a number of bytes attached to @rq.
2294 * If @rq has leftover, sets it up for the next range of segments.
2296 * Return:
2297 * %false - we are done with this request
2298 * %true - still buffers pending for this request
2300 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2302 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2304 EXPORT_SYMBOL(blk_end_request);
2307 * blk_end_request_all - Helper function for drives to finish the request.
2308 * @rq: the request to finish
2309 * @error: %0 for success, < %0 for error
2311 * Description:
2312 * Completely finish @rq.
2314 void blk_end_request_all(struct request *rq, int error)
2316 bool pending;
2317 unsigned int bidi_bytes = 0;
2319 if (unlikely(blk_bidi_rq(rq)))
2320 bidi_bytes = blk_rq_bytes(rq->next_rq);
2322 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2323 BUG_ON(pending);
2325 EXPORT_SYMBOL(blk_end_request_all);
2328 * blk_end_request_cur - Helper function to finish the current request chunk.
2329 * @rq: the request to finish the current chunk for
2330 * @error: %0 for success, < %0 for error
2332 * Description:
2333 * Complete the current consecutively mapped chunk from @rq.
2335 * Return:
2336 * %false - we are done with this request
2337 * %true - still buffers pending for this request
2339 bool blk_end_request_cur(struct request *rq, int error)
2341 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2343 EXPORT_SYMBOL(blk_end_request_cur);
2346 * blk_end_request_err - Finish a request till the next failure boundary.
2347 * @rq: the request to finish till the next failure boundary for
2348 * @error: must be negative errno
2350 * Description:
2351 * Complete @rq till the next failure boundary.
2353 * Return:
2354 * %false - we are done with this request
2355 * %true - still buffers pending for this request
2357 bool blk_end_request_err(struct request *rq, int error)
2359 WARN_ON(error >= 0);
2360 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2362 EXPORT_SYMBOL_GPL(blk_end_request_err);
2365 * __blk_end_request - Helper function for drivers to complete the request.
2366 * @rq: the request being processed
2367 * @error: %0 for success, < %0 for error
2368 * @nr_bytes: number of bytes to complete
2370 * Description:
2371 * Must be called with queue lock held unlike blk_end_request().
2373 * Return:
2374 * %false - we are done with this request
2375 * %true - still buffers pending for this request
2377 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2379 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2381 EXPORT_SYMBOL(__blk_end_request);
2384 * __blk_end_request_all - Helper function for drives to finish the request.
2385 * @rq: the request to finish
2386 * @error: %0 for success, < %0 for error
2388 * Description:
2389 * Completely finish @rq. Must be called with queue lock held.
2391 void __blk_end_request_all(struct request *rq, int error)
2393 bool pending;
2394 unsigned int bidi_bytes = 0;
2396 if (unlikely(blk_bidi_rq(rq)))
2397 bidi_bytes = blk_rq_bytes(rq->next_rq);
2399 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2400 BUG_ON(pending);
2402 EXPORT_SYMBOL(__blk_end_request_all);
2405 * __blk_end_request_cur - Helper function to finish the current request chunk.
2406 * @rq: the request to finish the current chunk for
2407 * @error: %0 for success, < %0 for error
2409 * Description:
2410 * Complete the current consecutively mapped chunk from @rq. Must
2411 * be called with queue lock held.
2413 * Return:
2414 * %false - we are done with this request
2415 * %true - still buffers pending for this request
2417 bool __blk_end_request_cur(struct request *rq, int error)
2419 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2421 EXPORT_SYMBOL(__blk_end_request_cur);
2424 * __blk_end_request_err - Finish a request till the next failure boundary.
2425 * @rq: the request to finish till the next failure boundary for
2426 * @error: must be negative errno
2428 * Description:
2429 * Complete @rq till the next failure boundary. Must be called
2430 * with queue lock held.
2432 * Return:
2433 * %false - we are done with this request
2434 * %true - still buffers pending for this request
2436 bool __blk_end_request_err(struct request *rq, int error)
2438 WARN_ON(error >= 0);
2439 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2441 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2443 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2444 struct bio *bio)
2446 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2447 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
2449 if (bio_has_data(bio)) {
2450 rq->nr_phys_segments = bio_phys_segments(q, bio);
2451 rq->buffer = bio_data(bio);
2453 rq->__data_len = bio->bi_size;
2454 rq->bio = rq->biotail = bio;
2456 if (bio->bi_bdev)
2457 rq->rq_disk = bio->bi_bdev->bd_disk;
2460 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2462 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2463 * @rq: the request to be flushed
2465 * Description:
2466 * Flush all pages in @rq.
2468 void rq_flush_dcache_pages(struct request *rq)
2470 struct req_iterator iter;
2471 struct bio_vec *bvec;
2473 rq_for_each_segment(bvec, rq, iter)
2474 flush_dcache_page(bvec->bv_page);
2476 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2477 #endif
2480 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2481 * @q : the queue of the device being checked
2483 * Description:
2484 * Check if underlying low-level drivers of a device are busy.
2485 * If the drivers want to export their busy state, they must set own
2486 * exporting function using blk_queue_lld_busy() first.
2488 * Basically, this function is used only by request stacking drivers
2489 * to stop dispatching requests to underlying devices when underlying
2490 * devices are busy. This behavior helps more I/O merging on the queue
2491 * of the request stacking driver and prevents I/O throughput regression
2492 * on burst I/O load.
2494 * Return:
2495 * 0 - Not busy (The request stacking driver should dispatch request)
2496 * 1 - Busy (The request stacking driver should stop dispatching request)
2498 int blk_lld_busy(struct request_queue *q)
2500 if (q->lld_busy_fn)
2501 return q->lld_busy_fn(q);
2503 return 0;
2505 EXPORT_SYMBOL_GPL(blk_lld_busy);
2508 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2509 * @rq: the clone request to be cleaned up
2511 * Description:
2512 * Free all bios in @rq for a cloned request.
2514 void blk_rq_unprep_clone(struct request *rq)
2516 struct bio *bio;
2518 while ((bio = rq->bio) != NULL) {
2519 rq->bio = bio->bi_next;
2521 bio_put(bio);
2524 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2527 * Copy attributes of the original request to the clone request.
2528 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2530 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2532 dst->cpu = src->cpu;
2533 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
2534 dst->cmd_type = src->cmd_type;
2535 dst->__sector = blk_rq_pos(src);
2536 dst->__data_len = blk_rq_bytes(src);
2537 dst->nr_phys_segments = src->nr_phys_segments;
2538 dst->ioprio = src->ioprio;
2539 dst->extra_len = src->extra_len;
2543 * blk_rq_prep_clone - Helper function to setup clone request
2544 * @rq: the request to be setup
2545 * @rq_src: original request to be cloned
2546 * @bs: bio_set that bios for clone are allocated from
2547 * @gfp_mask: memory allocation mask for bio
2548 * @bio_ctr: setup function to be called for each clone bio.
2549 * Returns %0 for success, non %0 for failure.
2550 * @data: private data to be passed to @bio_ctr
2552 * Description:
2553 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2554 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2555 * are not copied, and copying such parts is the caller's responsibility.
2556 * Also, pages which the original bios are pointing to are not copied
2557 * and the cloned bios just point same pages.
2558 * So cloned bios must be completed before original bios, which means
2559 * the caller must complete @rq before @rq_src.
2561 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2562 struct bio_set *bs, gfp_t gfp_mask,
2563 int (*bio_ctr)(struct bio *, struct bio *, void *),
2564 void *data)
2566 struct bio *bio, *bio_src;
2568 if (!bs)
2569 bs = fs_bio_set;
2571 blk_rq_init(NULL, rq);
2573 __rq_for_each_bio(bio_src, rq_src) {
2574 bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs);
2575 if (!bio)
2576 goto free_and_out;
2578 __bio_clone(bio, bio_src);
2580 if (bio_integrity(bio_src) &&
2581 bio_integrity_clone(bio, bio_src, gfp_mask, bs))
2582 goto free_and_out;
2584 if (bio_ctr && bio_ctr(bio, bio_src, data))
2585 goto free_and_out;
2587 if (rq->bio) {
2588 rq->biotail->bi_next = bio;
2589 rq->biotail = bio;
2590 } else
2591 rq->bio = rq->biotail = bio;
2594 __blk_rq_prep_clone(rq, rq_src);
2596 return 0;
2598 free_and_out:
2599 if (bio)
2600 bio_free(bio, bs);
2601 blk_rq_unprep_clone(rq);
2603 return -ENOMEM;
2605 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2607 int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
2609 return queue_work(kblockd_workqueue, work);
2611 EXPORT_SYMBOL(kblockd_schedule_work);
2613 int kblockd_schedule_delayed_work(struct request_queue *q,
2614 struct delayed_work *dwork, unsigned long delay)
2616 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2618 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2620 int __init blk_dev_init(void)
2622 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
2623 sizeof(((struct request *)0)->cmd_flags));
2625 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
2626 kblockd_workqueue = alloc_workqueue("kblockd",
2627 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
2628 if (!kblockd_workqueue)
2629 panic("Failed to create kblockd\n");
2631 request_cachep = kmem_cache_create("blkdev_requests",
2632 sizeof(struct request), 0, SLAB_PANIC, NULL);
2634 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
2635 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
2637 return 0;