2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
6 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
8 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
12 * This handles all read/write requests to block devices
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/backing-dev.h>
17 #include <linux/bio.h>
18 #include <linux/blkdev.h>
19 #include <linux/highmem.h>
21 #include <linux/kernel_stat.h>
22 #include <linux/string.h>
23 #include <linux/init.h>
24 #include <linux/completion.h>
25 #include <linux/slab.h>
26 #include <linux/swap.h>
27 #include <linux/writeback.h>
28 #include <linux/task_io_accounting_ops.h>
29 #include <linux/fault-inject.h>
30 #include <linux/list_sort.h>
32 #define CREATE_TRACE_POINTS
33 #include <trace/events/block.h>
37 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
38 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
39 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
41 static int __make_request(struct request_queue *q, struct bio *bio);
44 * For the allocated request tables
46 static struct kmem_cache *request_cachep;
49 * For queue allocation
51 struct kmem_cache *blk_requestq_cachep;
54 * Controlling structure to kblockd
56 static struct workqueue_struct *kblockd_workqueue;
58 static void drive_stat_acct(struct request *rq, int new_io)
60 struct hd_struct *part;
61 int rw = rq_data_dir(rq);
64 if (!blk_do_io_stat(rq))
67 cpu = part_stat_lock();
71 part_stat_inc(cpu, part, merges[rw]);
73 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
74 if (!hd_struct_try_get(part)) {
76 * The partition is already being removed,
77 * the request will be accounted on the disk only
79 * We take a reference on disk->part0 although that
80 * partition will never be deleted, so we can treat
81 * it as any other partition.
83 part = &rq->rq_disk->part0;
86 part_round_stats(cpu, part);
87 part_inc_in_flight(part, rw);
94 void blk_queue_congestion_threshold(struct request_queue *q)
98 nr = q->nr_requests - (q->nr_requests / 8) + 1;
99 if (nr > q->nr_requests)
101 q->nr_congestion_on = nr;
103 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
106 q->nr_congestion_off = nr;
110 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
113 * Locates the passed device's request queue and returns the address of its
116 * Will return NULL if the request queue cannot be located.
118 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
120 struct backing_dev_info *ret = NULL;
121 struct request_queue *q = bdev_get_queue(bdev);
124 ret = &q->backing_dev_info;
127 EXPORT_SYMBOL(blk_get_backing_dev_info);
129 void blk_rq_init(struct request_queue *q, struct request *rq)
131 memset(rq, 0, sizeof(*rq));
133 INIT_LIST_HEAD(&rq->queuelist);
134 INIT_LIST_HEAD(&rq->timeout_list);
137 rq->__sector = (sector_t) -1;
138 INIT_HLIST_NODE(&rq->hash);
139 RB_CLEAR_NODE(&rq->rb_node);
141 rq->cmd_len = BLK_MAX_CDB;
144 rq->start_time = jiffies;
145 set_start_time_ns(rq);
148 EXPORT_SYMBOL(blk_rq_init);
150 static void req_bio_endio(struct request *rq, struct bio *bio,
151 unsigned int nbytes, int error)
154 clear_bit(BIO_UPTODATE, &bio->bi_flags);
155 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
158 if (unlikely(nbytes > bio->bi_size)) {
159 printk(KERN_ERR "%s: want %u bytes done, %u left\n",
160 __func__, nbytes, bio->bi_size);
161 nbytes = bio->bi_size;
164 if (unlikely(rq->cmd_flags & REQ_QUIET))
165 set_bit(BIO_QUIET, &bio->bi_flags);
167 bio->bi_size -= nbytes;
168 bio->bi_sector += (nbytes >> 9);
170 if (bio_integrity(bio))
171 bio_integrity_advance(bio, nbytes);
173 /* don't actually finish bio if it's part of flush sequence */
174 if (bio->bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
175 bio_endio(bio, error);
178 void blk_dump_rq_flags(struct request *rq, char *msg)
182 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
183 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
186 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
187 (unsigned long long)blk_rq_pos(rq),
188 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
189 printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n",
190 rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq));
192 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
193 printk(KERN_INFO " cdb: ");
194 for (bit = 0; bit < BLK_MAX_CDB; bit++)
195 printk("%02x ", rq->cmd[bit]);
199 EXPORT_SYMBOL(blk_dump_rq_flags);
201 static void blk_delay_work(struct work_struct *work)
203 struct request_queue *q;
205 q = container_of(work, struct request_queue, delay_work.work);
206 spin_lock_irq(q->queue_lock);
208 spin_unlock_irq(q->queue_lock);
212 * blk_delay_queue - restart queueing after defined interval
213 * @q: The &struct request_queue in question
214 * @msecs: Delay in msecs
217 * Sometimes queueing needs to be postponed for a little while, to allow
218 * resources to come back. This function will make sure that queueing is
219 * restarted around the specified time.
221 void blk_delay_queue(struct request_queue *q, unsigned long msecs)
223 schedule_delayed_work(&q->delay_work, msecs_to_jiffies(msecs));
225 EXPORT_SYMBOL(blk_delay_queue);
228 * "plug" the device if there are no outstanding requests: this will
229 * force the transfer to start only after we have put all the requests
232 * This is called with interrupts off and no requests on the queue and
233 * with the queue lock held.
235 void blk_plug_device(struct request_queue *q)
237 WARN_ON(!irqs_disabled());
240 * don't plug a stopped queue, it must be paired with blk_start_queue()
241 * which will restart the queueing
243 if (blk_queue_stopped(q))
246 if (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED, q)) {
247 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
251 EXPORT_SYMBOL(blk_plug_device);
254 * blk_plug_device_unlocked - plug a device without queue lock held
255 * @q: The &struct request_queue to plug
258 * Like @blk_plug_device(), but grabs the queue lock and disables
261 void blk_plug_device_unlocked(struct request_queue *q)
265 spin_lock_irqsave(q->queue_lock, flags);
267 spin_unlock_irqrestore(q->queue_lock, flags);
269 EXPORT_SYMBOL(blk_plug_device_unlocked);
272 * remove the queue from the plugged list, if present. called with
273 * queue lock held and interrupts disabled.
275 int blk_remove_plug(struct request_queue *q)
277 WARN_ON(!irqs_disabled());
279 if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED, q))
282 del_timer(&q->unplug_timer);
285 EXPORT_SYMBOL(blk_remove_plug);
288 * remove the plug and let it rip..
290 void __generic_unplug_device(struct request_queue *q)
292 if (unlikely(blk_queue_stopped(q)))
294 if (!blk_remove_plug(q) && !blk_queue_nonrot(q))
301 * generic_unplug_device - fire a request queue
302 * @q: The &struct request_queue in question
305 * Linux uses plugging to build bigger requests queues before letting
306 * the device have at them. If a queue is plugged, the I/O scheduler
307 * is still adding and merging requests on the queue. Once the queue
308 * gets unplugged, the request_fn defined for the queue is invoked and
311 void generic_unplug_device(struct request_queue *q)
313 if (blk_queue_plugged(q)) {
314 spin_lock_irq(q->queue_lock);
315 __generic_unplug_device(q);
316 spin_unlock_irq(q->queue_lock);
319 EXPORT_SYMBOL(generic_unplug_device);
321 static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
324 struct request_queue *q = bdi->unplug_io_data;
329 void blk_unplug_work(struct work_struct *work)
331 struct request_queue *q =
332 container_of(work, struct request_queue, unplug_work);
334 trace_block_unplug_io(q);
338 void blk_unplug_timeout(unsigned long data)
340 struct request_queue *q = (struct request_queue *)data;
342 trace_block_unplug_timer(q);
343 kblockd_schedule_work(q, &q->unplug_work);
346 void blk_unplug(struct request_queue *q)
349 * devices don't necessarily have an ->unplug_fn defined
352 trace_block_unplug_io(q);
356 EXPORT_SYMBOL(blk_unplug);
359 * blk_start_queue - restart a previously stopped queue
360 * @q: The &struct request_queue in question
363 * blk_start_queue() will clear the stop flag on the queue, and call
364 * the request_fn for the queue if it was in a stopped state when
365 * entered. Also see blk_stop_queue(). Queue lock must be held.
367 void blk_start_queue(struct request_queue *q)
369 WARN_ON(!irqs_disabled());
371 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
374 EXPORT_SYMBOL(blk_start_queue);
377 * blk_stop_queue - stop a queue
378 * @q: The &struct request_queue in question
381 * The Linux block layer assumes that a block driver will consume all
382 * entries on the request queue when the request_fn strategy is called.
383 * Often this will not happen, because of hardware limitations (queue
384 * depth settings). If a device driver gets a 'queue full' response,
385 * or if it simply chooses not to queue more I/O at one point, it can
386 * call this function to prevent the request_fn from being called until
387 * the driver has signalled it's ready to go again. This happens by calling
388 * blk_start_queue() to restart queue operations. Queue lock must be held.
390 void blk_stop_queue(struct request_queue *q)
393 cancel_delayed_work(&q->delay_work);
394 queue_flag_set(QUEUE_FLAG_STOPPED, q);
396 EXPORT_SYMBOL(blk_stop_queue);
399 * blk_sync_queue - cancel any pending callbacks on a queue
403 * The block layer may perform asynchronous callback activity
404 * on a queue, such as calling the unplug function after a timeout.
405 * A block device may call blk_sync_queue to ensure that any
406 * such activity is cancelled, thus allowing it to release resources
407 * that the callbacks might use. The caller must already have made sure
408 * that its ->make_request_fn will not re-add plugging prior to calling
412 void blk_sync_queue(struct request_queue *q)
414 del_timer_sync(&q->unplug_timer);
415 del_timer_sync(&q->timeout);
416 cancel_work_sync(&q->unplug_work);
417 throtl_shutdown_timer_wq(q);
418 cancel_delayed_work_sync(&q->delay_work);
420 EXPORT_SYMBOL(blk_sync_queue);
423 * __blk_run_queue - run a single device queue
424 * @q: The queue to run
427 * See @blk_run_queue. This variant must be called with the queue lock
428 * held and interrupts disabled.
431 void __blk_run_queue(struct request_queue *q)
435 if (unlikely(blk_queue_stopped(q)))
438 if (elv_queue_empty(q))
442 * Only recurse once to avoid overrunning the stack, let the unplug
443 * handling reinvoke the handler shortly if we already got there.
445 if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) {
447 queue_flag_clear(QUEUE_FLAG_REENTER, q);
449 queue_flag_set(QUEUE_FLAG_PLUGGED, q);
450 kblockd_schedule_work(q, &q->unplug_work);
453 EXPORT_SYMBOL(__blk_run_queue);
456 * blk_run_queue - run a single device queue
457 * @q: The queue to run
460 * Invoke request handling on this queue, if it has pending work to do.
461 * May be used to restart queueing when a request has completed.
463 void blk_run_queue(struct request_queue *q)
467 spin_lock_irqsave(q->queue_lock, flags);
469 spin_unlock_irqrestore(q->queue_lock, flags);
471 EXPORT_SYMBOL(blk_run_queue);
473 void blk_put_queue(struct request_queue *q)
475 kobject_put(&q->kobj);
478 void blk_cleanup_queue(struct request_queue *q)
481 * We know we have process context here, so we can be a little
482 * cautious and ensure that pending block actions on this device
483 * are done before moving on. Going into this function, we should
484 * not have processes doing IO to this device.
488 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
489 mutex_lock(&q->sysfs_lock);
490 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
491 mutex_unlock(&q->sysfs_lock);
494 elevator_exit(q->elevator);
498 EXPORT_SYMBOL(blk_cleanup_queue);
500 static int blk_init_free_list(struct request_queue *q)
502 struct request_list *rl = &q->rq;
504 if (unlikely(rl->rq_pool))
507 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
508 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
510 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
511 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
513 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
514 mempool_free_slab, request_cachep, q->node);
522 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
524 return blk_alloc_queue_node(gfp_mask, -1);
526 EXPORT_SYMBOL(blk_alloc_queue);
528 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
530 struct request_queue *q;
533 q = kmem_cache_alloc_node(blk_requestq_cachep,
534 gfp_mask | __GFP_ZERO, node_id);
538 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
539 q->backing_dev_info.unplug_io_data = q;
540 q->backing_dev_info.ra_pages =
541 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
542 q->backing_dev_info.state = 0;
543 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
544 q->backing_dev_info.name = "block";
546 err = bdi_init(&q->backing_dev_info);
548 kmem_cache_free(blk_requestq_cachep, q);
552 if (blk_throtl_init(q)) {
553 kmem_cache_free(blk_requestq_cachep, q);
557 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
558 laptop_mode_timer_fn, (unsigned long) q);
559 init_timer(&q->unplug_timer);
560 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
561 INIT_LIST_HEAD(&q->timeout_list);
562 INIT_LIST_HEAD(&q->flush_queue[0]);
563 INIT_LIST_HEAD(&q->flush_queue[1]);
564 INIT_LIST_HEAD(&q->flush_data_in_flight);
565 INIT_WORK(&q->unplug_work, blk_unplug_work);
566 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
568 kobject_init(&q->kobj, &blk_queue_ktype);
570 mutex_init(&q->sysfs_lock);
571 spin_lock_init(&q->__queue_lock);
575 EXPORT_SYMBOL(blk_alloc_queue_node);
578 * blk_init_queue - prepare a request queue for use with a block device
579 * @rfn: The function to be called to process requests that have been
580 * placed on the queue.
581 * @lock: Request queue spin lock
584 * If a block device wishes to use the standard request handling procedures,
585 * which sorts requests and coalesces adjacent requests, then it must
586 * call blk_init_queue(). The function @rfn will be called when there
587 * are requests on the queue that need to be processed. If the device
588 * supports plugging, then @rfn may not be called immediately when requests
589 * are available on the queue, but may be called at some time later instead.
590 * Plugged queues are generally unplugged when a buffer belonging to one
591 * of the requests on the queue is needed, or due to memory pressure.
593 * @rfn is not required, or even expected, to remove all requests off the
594 * queue, but only as many as it can handle at a time. If it does leave
595 * requests on the queue, it is responsible for arranging that the requests
596 * get dealt with eventually.
598 * The queue spin lock must be held while manipulating the requests on the
599 * request queue; this lock will be taken also from interrupt context, so irq
600 * disabling is needed for it.
602 * Function returns a pointer to the initialized request queue, or %NULL if
606 * blk_init_queue() must be paired with a blk_cleanup_queue() call
607 * when the block device is deactivated (such as at module unload).
610 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
612 return blk_init_queue_node(rfn, lock, -1);
614 EXPORT_SYMBOL(blk_init_queue);
616 struct request_queue *
617 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
619 struct request_queue *uninit_q, *q;
621 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
625 q = blk_init_allocated_queue_node(uninit_q, rfn, lock, node_id);
627 blk_cleanup_queue(uninit_q);
631 EXPORT_SYMBOL(blk_init_queue_node);
633 struct request_queue *
634 blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
637 return blk_init_allocated_queue_node(q, rfn, lock, -1);
639 EXPORT_SYMBOL(blk_init_allocated_queue);
641 struct request_queue *
642 blk_init_allocated_queue_node(struct request_queue *q, request_fn_proc *rfn,
643 spinlock_t *lock, int node_id)
649 if (blk_init_free_list(q))
653 q->prep_rq_fn = NULL;
654 q->unprep_rq_fn = NULL;
655 q->unplug_fn = generic_unplug_device;
656 q->queue_flags = QUEUE_FLAG_DEFAULT;
657 q->queue_lock = lock;
660 * This also sets hw/phys segments, boundary and size
662 blk_queue_make_request(q, __make_request);
664 q->sg_reserved_size = INT_MAX;
669 if (!elevator_init(q, NULL)) {
670 blk_queue_congestion_threshold(q);
676 EXPORT_SYMBOL(blk_init_allocated_queue_node);
678 int blk_get_queue(struct request_queue *q)
680 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
681 kobject_get(&q->kobj);
688 static inline void blk_free_request(struct request_queue *q, struct request *rq)
690 BUG_ON(rq->cmd_flags & REQ_ON_PLUG);
692 if (rq->cmd_flags & REQ_ELVPRIV)
693 elv_put_request(q, rq);
694 mempool_free(rq, q->rq.rq_pool);
697 static struct request *
698 blk_alloc_request(struct request_queue *q, int flags, int priv, gfp_t gfp_mask)
700 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
707 rq->cmd_flags = flags | REQ_ALLOCED;
710 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
711 mempool_free(rq, q->rq.rq_pool);
714 rq->cmd_flags |= REQ_ELVPRIV;
721 * ioc_batching returns true if the ioc is a valid batching request and
722 * should be given priority access to a request.
724 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
730 * Make sure the process is able to allocate at least 1 request
731 * even if the batch times out, otherwise we could theoretically
734 return ioc->nr_batch_requests == q->nr_batching ||
735 (ioc->nr_batch_requests > 0
736 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
740 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
741 * will cause the process to be a "batcher" on all queues in the system. This
742 * is the behaviour we want though - once it gets a wakeup it should be given
745 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
747 if (!ioc || ioc_batching(q, ioc))
750 ioc->nr_batch_requests = q->nr_batching;
751 ioc->last_waited = jiffies;
754 static void __freed_request(struct request_queue *q, int sync)
756 struct request_list *rl = &q->rq;
758 if (rl->count[sync] < queue_congestion_off_threshold(q))
759 blk_clear_queue_congested(q, sync);
761 if (rl->count[sync] + 1 <= q->nr_requests) {
762 if (waitqueue_active(&rl->wait[sync]))
763 wake_up(&rl->wait[sync]);
765 blk_clear_queue_full(q, sync);
770 * A request has just been released. Account for it, update the full and
771 * congestion status, wake up any waiters. Called under q->queue_lock.
773 static void freed_request(struct request_queue *q, int sync, int priv)
775 struct request_list *rl = &q->rq;
781 __freed_request(q, sync);
783 if (unlikely(rl->starved[sync ^ 1]))
784 __freed_request(q, sync ^ 1);
788 * Determine if elevator data should be initialized when allocating the
789 * request associated with @bio.
791 static bool blk_rq_should_init_elevator(struct bio *bio)
797 * Flush requests do not use the elevator so skip initialization.
798 * This allows a request to share the flush and elevator data.
800 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
807 * Get a free request, queue_lock must be held.
808 * Returns NULL on failure, with queue_lock held.
809 * Returns !NULL on success, with queue_lock *not held*.
811 static struct request *get_request(struct request_queue *q, int rw_flags,
812 struct bio *bio, gfp_t gfp_mask)
814 struct request *rq = NULL;
815 struct request_list *rl = &q->rq;
816 struct io_context *ioc = NULL;
817 const bool is_sync = rw_is_sync(rw_flags) != 0;
818 int may_queue, priv = 0;
820 may_queue = elv_may_queue(q, rw_flags);
821 if (may_queue == ELV_MQUEUE_NO)
824 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
825 if (rl->count[is_sync]+1 >= q->nr_requests) {
826 ioc = current_io_context(GFP_ATOMIC, q->node);
828 * The queue will fill after this allocation, so set
829 * it as full, and mark this process as "batching".
830 * This process will be allowed to complete a batch of
831 * requests, others will be blocked.
833 if (!blk_queue_full(q, is_sync)) {
834 ioc_set_batching(q, ioc);
835 blk_set_queue_full(q, is_sync);
837 if (may_queue != ELV_MQUEUE_MUST
838 && !ioc_batching(q, ioc)) {
840 * The queue is full and the allocating
841 * process is not a "batcher", and not
842 * exempted by the IO scheduler
848 blk_set_queue_congested(q, is_sync);
852 * Only allow batching queuers to allocate up to 50% over the defined
853 * limit of requests, otherwise we could have thousands of requests
854 * allocated with any setting of ->nr_requests
856 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
859 rl->count[is_sync]++;
860 rl->starved[is_sync] = 0;
862 if (blk_rq_should_init_elevator(bio)) {
863 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
868 if (blk_queue_io_stat(q))
869 rw_flags |= REQ_IO_STAT;
870 spin_unlock_irq(q->queue_lock);
872 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
875 * Allocation failed presumably due to memory. Undo anything
876 * we might have messed up.
878 * Allocating task should really be put onto the front of the
879 * wait queue, but this is pretty rare.
881 spin_lock_irq(q->queue_lock);
882 freed_request(q, is_sync, priv);
885 * in the very unlikely event that allocation failed and no
886 * requests for this direction was pending, mark us starved
887 * so that freeing of a request in the other direction will
888 * notice us. another possible fix would be to split the
889 * rq mempool into READ and WRITE
892 if (unlikely(rl->count[is_sync] == 0))
893 rl->starved[is_sync] = 1;
899 * ioc may be NULL here, and ioc_batching will be false. That's
900 * OK, if the queue is under the request limit then requests need
901 * not count toward the nr_batch_requests limit. There will always
902 * be some limit enforced by BLK_BATCH_TIME.
904 if (ioc_batching(q, ioc))
905 ioc->nr_batch_requests--;
907 trace_block_getrq(q, bio, rw_flags & 1);
913 * No available requests for this queue, unplug the device and wait for some
914 * requests to become available.
916 * Called with q->queue_lock held, and returns with it unlocked.
918 static struct request *get_request_wait(struct request_queue *q, int rw_flags,
921 const bool is_sync = rw_is_sync(rw_flags) != 0;
924 rq = get_request(q, rw_flags, bio, GFP_NOIO);
927 struct io_context *ioc;
928 struct request_list *rl = &q->rq;
930 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
931 TASK_UNINTERRUPTIBLE);
933 trace_block_sleeprq(q, bio, rw_flags & 1);
935 __generic_unplug_device(q);
936 spin_unlock_irq(q->queue_lock);
940 * After sleeping, we become a "batching" process and
941 * will be able to allocate at least one request, and
942 * up to a big batch of them for a small period time.
943 * See ioc_batching, ioc_set_batching
945 ioc = current_io_context(GFP_NOIO, q->node);
946 ioc_set_batching(q, ioc);
948 spin_lock_irq(q->queue_lock);
949 finish_wait(&rl->wait[is_sync], &wait);
951 rq = get_request(q, rw_flags, bio, GFP_NOIO);
957 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
961 BUG_ON(rw != READ && rw != WRITE);
963 spin_lock_irq(q->queue_lock);
964 if (gfp_mask & __GFP_WAIT) {
965 rq = get_request_wait(q, rw, NULL);
967 rq = get_request(q, rw, NULL, gfp_mask);
969 spin_unlock_irq(q->queue_lock);
971 /* q->queue_lock is unlocked at this point */
975 EXPORT_SYMBOL(blk_get_request);
978 * blk_make_request - given a bio, allocate a corresponding struct request.
979 * @q: target request queue
980 * @bio: The bio describing the memory mappings that will be submitted for IO.
981 * It may be a chained-bio properly constructed by block/bio layer.
982 * @gfp_mask: gfp flags to be used for memory allocation
984 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
985 * type commands. Where the struct request needs to be farther initialized by
986 * the caller. It is passed a &struct bio, which describes the memory info of
989 * The caller of blk_make_request must make sure that bi_io_vec
990 * are set to describe the memory buffers. That bio_data_dir() will return
991 * the needed direction of the request. (And all bio's in the passed bio-chain
992 * are properly set accordingly)
994 * If called under none-sleepable conditions, mapped bio buffers must not
995 * need bouncing, by calling the appropriate masked or flagged allocator,
996 * suitable for the target device. Otherwise the call to blk_queue_bounce will
999 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1000 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1001 * anything but the first bio in the chain. Otherwise you risk waiting for IO
1002 * completion of a bio that hasn't been submitted yet, thus resulting in a
1003 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1004 * of bio_alloc(), as that avoids the mempool deadlock.
1005 * If possible a big IO should be split into smaller parts when allocation
1006 * fails. Partial allocation should not be an error, or you risk a live-lock.
1008 struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1011 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1014 return ERR_PTR(-ENOMEM);
1017 struct bio *bounce_bio = bio;
1020 blk_queue_bounce(q, &bounce_bio);
1021 ret = blk_rq_append_bio(q, rq, bounce_bio);
1022 if (unlikely(ret)) {
1023 blk_put_request(rq);
1024 return ERR_PTR(ret);
1030 EXPORT_SYMBOL(blk_make_request);
1033 * blk_requeue_request - put a request back on queue
1034 * @q: request queue where request should be inserted
1035 * @rq: request to be inserted
1038 * Drivers often keep queueing requests until the hardware cannot accept
1039 * more, when that condition happens we need to put the request back
1040 * on the queue. Must be called with queue lock held.
1042 void blk_requeue_request(struct request_queue *q, struct request *rq)
1044 blk_delete_timer(rq);
1045 blk_clear_rq_complete(rq);
1046 trace_block_rq_requeue(q, rq);
1048 if (blk_rq_tagged(rq))
1049 blk_queue_end_tag(q, rq);
1051 BUG_ON(blk_queued_rq(rq));
1053 elv_requeue_request(q, rq);
1055 EXPORT_SYMBOL(blk_requeue_request);
1057 static void add_acct_request(struct request_queue *q, struct request *rq,
1060 drive_stat_acct(rq, 1);
1061 __elv_add_request(q, rq, where, 0);
1065 * blk_insert_request - insert a special request into a request queue
1066 * @q: request queue where request should be inserted
1067 * @rq: request to be inserted
1068 * @at_head: insert request at head or tail of queue
1069 * @data: private data
1072 * Many block devices need to execute commands asynchronously, so they don't
1073 * block the whole kernel from preemption during request execution. This is
1074 * accomplished normally by inserting aritficial requests tagged as
1075 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
1076 * be scheduled for actual execution by the request queue.
1078 * We have the option of inserting the head or the tail of the queue.
1079 * Typically we use the tail for new ioctls and so forth. We use the head
1080 * of the queue for things like a QUEUE_FULL message from a device, or a
1081 * host that is unable to accept a particular command.
1083 void blk_insert_request(struct request_queue *q, struct request *rq,
1084 int at_head, void *data)
1086 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
1087 unsigned long flags;
1090 * tell I/O scheduler that this isn't a regular read/write (ie it
1091 * must not attempt merges on this) and that it acts as a soft
1094 rq->cmd_type = REQ_TYPE_SPECIAL;
1098 spin_lock_irqsave(q->queue_lock, flags);
1101 * If command is tagged, release the tag
1103 if (blk_rq_tagged(rq))
1104 blk_queue_end_tag(q, rq);
1106 add_acct_request(q, rq, where);
1108 spin_unlock_irqrestore(q->queue_lock, flags);
1110 EXPORT_SYMBOL(blk_insert_request);
1112 static void part_round_stats_single(int cpu, struct hd_struct *part,
1115 if (now == part->stamp)
1118 if (part_in_flight(part)) {
1119 __part_stat_add(cpu, part, time_in_queue,
1120 part_in_flight(part) * (now - part->stamp));
1121 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1127 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1128 * @cpu: cpu number for stats access
1129 * @part: target partition
1131 * The average IO queue length and utilisation statistics are maintained
1132 * by observing the current state of the queue length and the amount of
1133 * time it has been in this state for.
1135 * Normally, that accounting is done on IO completion, but that can result
1136 * in more than a second's worth of IO being accounted for within any one
1137 * second, leading to >100% utilisation. To deal with that, we call this
1138 * function to do a round-off before returning the results when reading
1139 * /proc/diskstats. This accounts immediately for all queue usage up to
1140 * the current jiffies and restarts the counters again.
1142 void part_round_stats(int cpu, struct hd_struct *part)
1144 unsigned long now = jiffies;
1147 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1148 part_round_stats_single(cpu, part, now);
1150 EXPORT_SYMBOL_GPL(part_round_stats);
1153 * queue lock must be held
1155 void __blk_put_request(struct request_queue *q, struct request *req)
1159 if (unlikely(--req->ref_count))
1162 elv_completed_request(q, req);
1164 /* this is a bio leak */
1165 WARN_ON(req->bio != NULL);
1168 * Request may not have originated from ll_rw_blk. if not,
1169 * it didn't come out of our reserved rq pools
1171 if (req->cmd_flags & REQ_ALLOCED) {
1172 int is_sync = rq_is_sync(req) != 0;
1173 int priv = req->cmd_flags & REQ_ELVPRIV;
1175 BUG_ON(!list_empty(&req->queuelist));
1176 BUG_ON(!hlist_unhashed(&req->hash));
1178 blk_free_request(q, req);
1179 freed_request(q, is_sync, priv);
1182 EXPORT_SYMBOL_GPL(__blk_put_request);
1184 void blk_put_request(struct request *req)
1186 unsigned long flags;
1187 struct request_queue *q = req->q;
1189 spin_lock_irqsave(q->queue_lock, flags);
1190 __blk_put_request(q, req);
1191 spin_unlock_irqrestore(q->queue_lock, flags);
1193 EXPORT_SYMBOL(blk_put_request);
1196 * blk_add_request_payload - add a payload to a request
1197 * @rq: request to update
1198 * @page: page backing the payload
1199 * @len: length of the payload.
1201 * This allows to later add a payload to an already submitted request by
1202 * a block driver. The driver needs to take care of freeing the payload
1205 * Note that this is a quite horrible hack and nothing but handling of
1206 * discard requests should ever use it.
1208 void blk_add_request_payload(struct request *rq, struct page *page,
1211 struct bio *bio = rq->bio;
1213 bio->bi_io_vec->bv_page = page;
1214 bio->bi_io_vec->bv_offset = 0;
1215 bio->bi_io_vec->bv_len = len;
1219 bio->bi_phys_segments = 1;
1221 rq->__data_len = rq->resid_len = len;
1222 rq->nr_phys_segments = 1;
1223 rq->buffer = bio_data(bio);
1225 EXPORT_SYMBOL_GPL(blk_add_request_payload);
1227 static bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1230 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1233 * Debug stuff, kill later
1235 if (!rq_mergeable(req)) {
1236 blk_dump_rq_flags(req, "back");
1240 if (!ll_back_merge_fn(q, req, bio))
1243 trace_block_bio_backmerge(q, bio);
1245 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1246 blk_rq_set_mixed_merge(req);
1248 req->biotail->bi_next = bio;
1250 req->__data_len += bio->bi_size;
1251 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1253 drive_stat_acct(req, 0);
1257 static bool bio_attempt_front_merge(struct request_queue *q,
1258 struct request *req, struct bio *bio)
1260 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1264 * Debug stuff, kill later
1266 if (!rq_mergeable(req)) {
1267 blk_dump_rq_flags(req, "front");
1271 if (!ll_front_merge_fn(q, req, bio))
1274 trace_block_bio_frontmerge(q, bio);
1276 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1277 blk_rq_set_mixed_merge(req);
1279 sector = bio->bi_sector;
1281 bio->bi_next = req->bio;
1285 * may not be valid. if the low level driver said
1286 * it didn't need a bounce buffer then it better
1287 * not touch req->buffer either...
1289 req->buffer = bio_data(bio);
1290 req->__sector = bio->bi_sector;
1291 req->__data_len += bio->bi_size;
1292 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1294 drive_stat_acct(req, 0);
1299 * Attempts to merge with the plugged list in the current process. Returns
1300 * true if merge was succesful, otherwise false.
1302 static bool attempt_plug_merge(struct task_struct *tsk, struct request_queue *q,
1305 struct blk_plug *plug;
1313 list_for_each_entry_reverse(rq, &plug->list, queuelist) {
1319 el_ret = elv_try_merge(rq, bio);
1320 if (el_ret == ELEVATOR_BACK_MERGE) {
1321 ret = bio_attempt_back_merge(q, rq, bio);
1324 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1325 ret = bio_attempt_front_merge(q, rq, bio);
1334 void init_request_from_bio(struct request *req, struct bio *bio)
1336 req->cpu = bio->bi_comp_cpu;
1337 req->cmd_type = REQ_TYPE_FS;
1339 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1340 if (bio->bi_rw & REQ_RAHEAD)
1341 req->cmd_flags |= REQ_FAILFAST_MASK;
1344 req->__sector = bio->bi_sector;
1345 req->ioprio = bio_prio(bio);
1346 blk_rq_bio_prep(req->q, req, bio);
1349 static int __make_request(struct request_queue *q, struct bio *bio)
1351 const bool sync = !!(bio->bi_rw & REQ_SYNC);
1352 struct blk_plug *plug;
1353 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1354 struct request *req;
1357 * low level driver can indicate that it wants pages above a
1358 * certain limit bounced to low memory (ie for highmem, or even
1359 * ISA dma in theory)
1361 blk_queue_bounce(q, &bio);
1363 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
1364 spin_lock_irq(q->queue_lock);
1365 where = ELEVATOR_INSERT_FLUSH;
1370 * Check if we can merge with the plugged list before grabbing
1373 if (attempt_plug_merge(current, q, bio))
1376 spin_lock_irq(q->queue_lock);
1378 el_ret = elv_merge(q, &req, bio);
1379 if (el_ret == ELEVATOR_BACK_MERGE) {
1380 BUG_ON(req->cmd_flags & REQ_ON_PLUG);
1381 if (bio_attempt_back_merge(q, req, bio)) {
1382 if (!attempt_back_merge(q, req))
1383 elv_merged_request(q, req, el_ret);
1386 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1387 BUG_ON(req->cmd_flags & REQ_ON_PLUG);
1388 if (bio_attempt_front_merge(q, req, bio)) {
1389 if (!attempt_front_merge(q, req))
1390 elv_merged_request(q, req, el_ret);
1397 * This sync check and mask will be re-done in init_request_from_bio(),
1398 * but we need to set it earlier to expose the sync flag to the
1399 * rq allocator and io schedulers.
1401 rw_flags = bio_data_dir(bio);
1403 rw_flags |= REQ_SYNC;
1406 * Grab a free request. This is might sleep but can not fail.
1407 * Returns with the queue unlocked.
1409 req = get_request_wait(q, rw_flags, bio);
1412 * After dropping the lock and possibly sleeping here, our request
1413 * may now be mergeable after it had proven unmergeable (above).
1414 * We don't worry about that case for efficiency. It won't happen
1415 * often, and the elevators are able to handle it.
1417 init_request_from_bio(req, bio);
1419 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) ||
1420 bio_flagged(bio, BIO_CPU_AFFINE)) {
1421 req->cpu = blk_cpu_to_group(get_cpu());
1425 plug = current->plug;
1426 if (plug && !sync) {
1427 if (!plug->should_sort && !list_empty(&plug->list)) {
1428 struct request *__rq;
1430 __rq = list_entry_rq(plug->list.prev);
1432 plug->should_sort = 1;
1435 * Debug flag, kill later
1437 req->cmd_flags |= REQ_ON_PLUG;
1438 list_add_tail(&req->queuelist, &plug->list);
1439 drive_stat_acct(req, 1);
1441 spin_lock_irq(q->queue_lock);
1442 add_acct_request(q, req, where);
1445 spin_unlock_irq(q->queue_lock);
1452 * If bio->bi_dev is a partition, remap the location
1454 static inline void blk_partition_remap(struct bio *bio)
1456 struct block_device *bdev = bio->bi_bdev;
1458 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1459 struct hd_struct *p = bdev->bd_part;
1461 bio->bi_sector += p->start_sect;
1462 bio->bi_bdev = bdev->bd_contains;
1464 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1466 bio->bi_sector - p->start_sect);
1470 static void handle_bad_sector(struct bio *bio)
1472 char b[BDEVNAME_SIZE];
1474 printk(KERN_INFO "attempt to access beyond end of device\n");
1475 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1476 bdevname(bio->bi_bdev, b),
1478 (unsigned long long)bio->bi_sector + bio_sectors(bio),
1479 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1481 set_bit(BIO_EOF, &bio->bi_flags);
1484 #ifdef CONFIG_FAIL_MAKE_REQUEST
1486 static DECLARE_FAULT_ATTR(fail_make_request);
1488 static int __init setup_fail_make_request(char *str)
1490 return setup_fault_attr(&fail_make_request, str);
1492 __setup("fail_make_request=", setup_fail_make_request);
1494 static int should_fail_request(struct bio *bio)
1496 struct hd_struct *part = bio->bi_bdev->bd_part;
1498 if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail)
1499 return should_fail(&fail_make_request, bio->bi_size);
1504 static int __init fail_make_request_debugfs(void)
1506 return init_fault_attr_dentries(&fail_make_request,
1507 "fail_make_request");
1510 late_initcall(fail_make_request_debugfs);
1512 #else /* CONFIG_FAIL_MAKE_REQUEST */
1514 static inline int should_fail_request(struct bio *bio)
1519 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1522 * Check whether this bio extends beyond the end of the device.
1524 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1531 /* Test device or partition size, when known. */
1532 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1534 sector_t sector = bio->bi_sector;
1536 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1538 * This may well happen - the kernel calls bread()
1539 * without checking the size of the device, e.g., when
1540 * mounting a device.
1542 handle_bad_sector(bio);
1551 * generic_make_request - hand a buffer to its device driver for I/O
1552 * @bio: The bio describing the location in memory and on the device.
1554 * generic_make_request() is used to make I/O requests of block
1555 * devices. It is passed a &struct bio, which describes the I/O that needs
1558 * generic_make_request() does not return any status. The
1559 * success/failure status of the request, along with notification of
1560 * completion, is delivered asynchronously through the bio->bi_end_io
1561 * function described (one day) else where.
1563 * The caller of generic_make_request must make sure that bi_io_vec
1564 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1565 * set to describe the device address, and the
1566 * bi_end_io and optionally bi_private are set to describe how
1567 * completion notification should be signaled.
1569 * generic_make_request and the drivers it calls may use bi_next if this
1570 * bio happens to be merged with someone else, and may change bi_dev and
1571 * bi_sector for remaps as it sees fit. So the values of these fields
1572 * should NOT be depended on after the call to generic_make_request.
1574 static inline void __generic_make_request(struct bio *bio)
1576 struct request_queue *q;
1577 sector_t old_sector;
1578 int ret, nr_sectors = bio_sectors(bio);
1584 if (bio_check_eod(bio, nr_sectors))
1588 * Resolve the mapping until finished. (drivers are
1589 * still free to implement/resolve their own stacking
1590 * by explicitly returning 0)
1592 * NOTE: we don't repeat the blk_size check for each new device.
1593 * Stacking drivers are expected to know what they are doing.
1598 char b[BDEVNAME_SIZE];
1600 q = bdev_get_queue(bio->bi_bdev);
1603 "generic_make_request: Trying to access "
1604 "nonexistent block-device %s (%Lu)\n",
1605 bdevname(bio->bi_bdev, b),
1606 (long long) bio->bi_sector);
1610 if (unlikely(!(bio->bi_rw & REQ_DISCARD) &&
1611 nr_sectors > queue_max_hw_sectors(q))) {
1612 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1613 bdevname(bio->bi_bdev, b),
1615 queue_max_hw_sectors(q));
1619 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
1622 if (should_fail_request(bio))
1626 * If this device has partitions, remap block n
1627 * of partition p to block n+start(p) of the disk.
1629 blk_partition_remap(bio);
1631 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1634 if (old_sector != -1)
1635 trace_block_bio_remap(q, bio, old_dev, old_sector);
1637 old_sector = bio->bi_sector;
1638 old_dev = bio->bi_bdev->bd_dev;
1640 if (bio_check_eod(bio, nr_sectors))
1644 * Filter flush bio's early so that make_request based
1645 * drivers without flush support don't have to worry
1648 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1649 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1656 if ((bio->bi_rw & REQ_DISCARD) &&
1657 (!blk_queue_discard(q) ||
1658 ((bio->bi_rw & REQ_SECURE) &&
1659 !blk_queue_secdiscard(q)))) {
1664 blk_throtl_bio(q, &bio);
1667 * If bio = NULL, bio has been throttled and will be submitted
1673 trace_block_bio_queue(q, bio);
1675 ret = q->make_request_fn(q, bio);
1681 bio_endio(bio, err);
1685 * We only want one ->make_request_fn to be active at a time,
1686 * else stack usage with stacked devices could be a problem.
1687 * So use current->bio_list to keep a list of requests
1688 * submited by a make_request_fn function.
1689 * current->bio_list is also used as a flag to say if
1690 * generic_make_request is currently active in this task or not.
1691 * If it is NULL, then no make_request is active. If it is non-NULL,
1692 * then a make_request is active, and new requests should be added
1695 void generic_make_request(struct bio *bio)
1697 struct bio_list bio_list_on_stack;
1699 if (current->bio_list) {
1700 /* make_request is active */
1701 bio_list_add(current->bio_list, bio);
1704 /* following loop may be a bit non-obvious, and so deserves some
1706 * Before entering the loop, bio->bi_next is NULL (as all callers
1707 * ensure that) so we have a list with a single bio.
1708 * We pretend that we have just taken it off a longer list, so
1709 * we assign bio_list to a pointer to the bio_list_on_stack,
1710 * thus initialising the bio_list of new bios to be
1711 * added. __generic_make_request may indeed add some more bios
1712 * through a recursive call to generic_make_request. If it
1713 * did, we find a non-NULL value in bio_list and re-enter the loop
1714 * from the top. In this case we really did just take the bio
1715 * of the top of the list (no pretending) and so remove it from
1716 * bio_list, and call into __generic_make_request again.
1718 * The loop was structured like this to make only one call to
1719 * __generic_make_request (which is important as it is large and
1720 * inlined) and to keep the structure simple.
1722 BUG_ON(bio->bi_next);
1723 bio_list_init(&bio_list_on_stack);
1724 current->bio_list = &bio_list_on_stack;
1726 __generic_make_request(bio);
1727 bio = bio_list_pop(current->bio_list);
1729 current->bio_list = NULL; /* deactivate */
1731 EXPORT_SYMBOL(generic_make_request);
1734 * submit_bio - submit a bio to the block device layer for I/O
1735 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1736 * @bio: The &struct bio which describes the I/O
1738 * submit_bio() is very similar in purpose to generic_make_request(), and
1739 * uses that function to do most of the work. Both are fairly rough
1740 * interfaces; @bio must be presetup and ready for I/O.
1743 void submit_bio(int rw, struct bio *bio)
1745 int count = bio_sectors(bio);
1750 * If it's a regular read/write or a barrier with data attached,
1751 * go through the normal accounting stuff before submission.
1753 if (bio_has_data(bio) && !(rw & REQ_DISCARD)) {
1755 count_vm_events(PGPGOUT, count);
1757 task_io_account_read(bio->bi_size);
1758 count_vm_events(PGPGIN, count);
1761 if (unlikely(block_dump)) {
1762 char b[BDEVNAME_SIZE];
1763 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1764 current->comm, task_pid_nr(current),
1765 (rw & WRITE) ? "WRITE" : "READ",
1766 (unsigned long long)bio->bi_sector,
1767 bdevname(bio->bi_bdev, b),
1772 generic_make_request(bio);
1774 EXPORT_SYMBOL(submit_bio);
1777 * blk_rq_check_limits - Helper function to check a request for the queue limit
1779 * @rq: the request being checked
1782 * @rq may have been made based on weaker limitations of upper-level queues
1783 * in request stacking drivers, and it may violate the limitation of @q.
1784 * Since the block layer and the underlying device driver trust @rq
1785 * after it is inserted to @q, it should be checked against @q before
1786 * the insertion using this generic function.
1788 * This function should also be useful for request stacking drivers
1789 * in some cases below, so export this function.
1790 * Request stacking drivers like request-based dm may change the queue
1791 * limits while requests are in the queue (e.g. dm's table swapping).
1792 * Such request stacking drivers should check those requests agaist
1793 * the new queue limits again when they dispatch those requests,
1794 * although such checkings are also done against the old queue limits
1795 * when submitting requests.
1797 int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1799 if (rq->cmd_flags & REQ_DISCARD)
1802 if (blk_rq_sectors(rq) > queue_max_sectors(q) ||
1803 blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) {
1804 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1809 * queue's settings related to segment counting like q->bounce_pfn
1810 * may differ from that of other stacking queues.
1811 * Recalculate it to check the request correctly on this queue's
1814 blk_recalc_rq_segments(rq);
1815 if (rq->nr_phys_segments > queue_max_segments(q)) {
1816 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1822 EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1825 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1826 * @q: the queue to submit the request
1827 * @rq: the request being queued
1829 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1831 unsigned long flags;
1833 if (blk_rq_check_limits(q, rq))
1836 #ifdef CONFIG_FAIL_MAKE_REQUEST
1837 if (rq->rq_disk && rq->rq_disk->part0.make_it_fail &&
1838 should_fail(&fail_make_request, blk_rq_bytes(rq)))
1842 spin_lock_irqsave(q->queue_lock, flags);
1845 * Submitting request must be dequeued before calling this function
1846 * because it will be linked to another request_queue
1848 BUG_ON(blk_queued_rq(rq));
1850 add_acct_request(q, rq, ELEVATOR_INSERT_BACK);
1851 spin_unlock_irqrestore(q->queue_lock, flags);
1855 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1858 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1859 * @rq: request to examine
1862 * A request could be merge of IOs which require different failure
1863 * handling. This function determines the number of bytes which
1864 * can be failed from the beginning of the request without
1865 * crossing into area which need to be retried further.
1868 * The number of bytes to fail.
1871 * queue_lock must be held.
1873 unsigned int blk_rq_err_bytes(const struct request *rq)
1875 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1876 unsigned int bytes = 0;
1879 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1880 return blk_rq_bytes(rq);
1883 * Currently the only 'mixing' which can happen is between
1884 * different fastfail types. We can safely fail portions
1885 * which have all the failfast bits that the first one has -
1886 * the ones which are at least as eager to fail as the first
1889 for (bio = rq->bio; bio; bio = bio->bi_next) {
1890 if ((bio->bi_rw & ff) != ff)
1892 bytes += bio->bi_size;
1895 /* this could lead to infinite loop */
1896 BUG_ON(blk_rq_bytes(rq) && !bytes);
1899 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1901 static void blk_account_io_completion(struct request *req, unsigned int bytes)
1903 if (blk_do_io_stat(req)) {
1904 const int rw = rq_data_dir(req);
1905 struct hd_struct *part;
1908 cpu = part_stat_lock();
1910 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1915 static void blk_account_io_done(struct request *req)
1918 * Account IO completion. flush_rq isn't accounted as a
1919 * normal IO on queueing nor completion. Accounting the
1920 * containing request is enough.
1922 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
1923 unsigned long duration = jiffies - req->start_time;
1924 const int rw = rq_data_dir(req);
1925 struct hd_struct *part;
1928 cpu = part_stat_lock();
1931 part_stat_inc(cpu, part, ios[rw]);
1932 part_stat_add(cpu, part, ticks[rw], duration);
1933 part_round_stats(cpu, part);
1934 part_dec_in_flight(part, rw);
1936 hd_struct_put(part);
1942 * blk_peek_request - peek at the top of a request queue
1943 * @q: request queue to peek at
1946 * Return the request at the top of @q. The returned request
1947 * should be started using blk_start_request() before LLD starts
1951 * Pointer to the request at the top of @q if available. Null
1955 * queue_lock must be held.
1957 struct request *blk_peek_request(struct request_queue *q)
1962 while ((rq = __elv_next_request(q)) != NULL) {
1963 if (!(rq->cmd_flags & REQ_STARTED)) {
1965 * This is the first time the device driver
1966 * sees this request (possibly after
1967 * requeueing). Notify IO scheduler.
1969 if (rq->cmd_flags & REQ_SORTED)
1970 elv_activate_rq(q, rq);
1973 * just mark as started even if we don't start
1974 * it, a request that has been delayed should
1975 * not be passed by new incoming requests
1977 rq->cmd_flags |= REQ_STARTED;
1978 trace_block_rq_issue(q, rq);
1981 if (!q->boundary_rq || q->boundary_rq == rq) {
1982 q->end_sector = rq_end_sector(rq);
1983 q->boundary_rq = NULL;
1986 if (rq->cmd_flags & REQ_DONTPREP)
1989 if (q->dma_drain_size && blk_rq_bytes(rq)) {
1991 * make sure space for the drain appears we
1992 * know we can do this because max_hw_segments
1993 * has been adjusted to be one fewer than the
1996 rq->nr_phys_segments++;
2002 ret = q->prep_rq_fn(q, rq);
2003 if (ret == BLKPREP_OK) {
2005 } else if (ret == BLKPREP_DEFER) {
2007 * the request may have been (partially) prepped.
2008 * we need to keep this request in the front to
2009 * avoid resource deadlock. REQ_STARTED will
2010 * prevent other fs requests from passing this one.
2012 if (q->dma_drain_size && blk_rq_bytes(rq) &&
2013 !(rq->cmd_flags & REQ_DONTPREP)) {
2015 * remove the space for the drain we added
2016 * so that we don't add it again
2018 --rq->nr_phys_segments;
2023 } else if (ret == BLKPREP_KILL) {
2024 rq->cmd_flags |= REQ_QUIET;
2026 * Mark this request as started so we don't trigger
2027 * any debug logic in the end I/O path.
2029 blk_start_request(rq);
2030 __blk_end_request_all(rq, -EIO);
2032 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2039 EXPORT_SYMBOL(blk_peek_request);
2041 void blk_dequeue_request(struct request *rq)
2043 struct request_queue *q = rq->q;
2045 BUG_ON(list_empty(&rq->queuelist));
2046 BUG_ON(ELV_ON_HASH(rq));
2048 list_del_init(&rq->queuelist);
2051 * the time frame between a request being removed from the lists
2052 * and to it is freed is accounted as io that is in progress at
2055 if (blk_account_rq(rq)) {
2056 q->in_flight[rq_is_sync(rq)]++;
2057 set_io_start_time_ns(rq);
2062 * blk_start_request - start request processing on the driver
2063 * @req: request to dequeue
2066 * Dequeue @req and start timeout timer on it. This hands off the
2067 * request to the driver.
2069 * Block internal functions which don't want to start timer should
2070 * call blk_dequeue_request().
2073 * queue_lock must be held.
2075 void blk_start_request(struct request *req)
2077 blk_dequeue_request(req);
2080 * We are now handing the request to the hardware, initialize
2081 * resid_len to full count and add the timeout handler.
2083 req->resid_len = blk_rq_bytes(req);
2084 if (unlikely(blk_bidi_rq(req)))
2085 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2089 EXPORT_SYMBOL(blk_start_request);
2092 * blk_fetch_request - fetch a request from a request queue
2093 * @q: request queue to fetch a request from
2096 * Return the request at the top of @q. The request is started on
2097 * return and LLD can start processing it immediately.
2100 * Pointer to the request at the top of @q if available. Null
2104 * queue_lock must be held.
2106 struct request *blk_fetch_request(struct request_queue *q)
2110 rq = blk_peek_request(q);
2112 blk_start_request(rq);
2115 EXPORT_SYMBOL(blk_fetch_request);
2118 * blk_update_request - Special helper function for request stacking drivers
2119 * @req: the request being processed
2120 * @error: %0 for success, < %0 for error
2121 * @nr_bytes: number of bytes to complete @req
2124 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2125 * the request structure even if @req doesn't have leftover.
2126 * If @req has leftover, sets it up for the next range of segments.
2128 * This special helper function is only for request stacking drivers
2129 * (e.g. request-based dm) so that they can handle partial completion.
2130 * Actual device drivers should use blk_end_request instead.
2132 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2133 * %false return from this function.
2136 * %false - this request doesn't have any more data
2137 * %true - this request has more data
2139 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2141 int total_bytes, bio_nbytes, next_idx = 0;
2147 trace_block_rq_complete(req->q, req);
2150 * For fs requests, rq is just carrier of independent bio's
2151 * and each partial completion should be handled separately.
2152 * Reset per-request error on each partial completion.
2154 * TODO: tj: This is too subtle. It would be better to let
2155 * low level drivers do what they see fit.
2157 if (req->cmd_type == REQ_TYPE_FS)
2160 if (error && req->cmd_type == REQ_TYPE_FS &&
2161 !(req->cmd_flags & REQ_QUIET)) {
2162 printk(KERN_ERR "end_request: I/O error, dev %s, sector %llu\n",
2163 req->rq_disk ? req->rq_disk->disk_name : "?",
2164 (unsigned long long)blk_rq_pos(req));
2167 blk_account_io_completion(req, nr_bytes);
2169 total_bytes = bio_nbytes = 0;
2170 while ((bio = req->bio) != NULL) {
2173 if (nr_bytes >= bio->bi_size) {
2174 req->bio = bio->bi_next;
2175 nbytes = bio->bi_size;
2176 req_bio_endio(req, bio, nbytes, error);
2180 int idx = bio->bi_idx + next_idx;
2182 if (unlikely(idx >= bio->bi_vcnt)) {
2183 blk_dump_rq_flags(req, "__end_that");
2184 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
2185 __func__, idx, bio->bi_vcnt);
2189 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2190 BIO_BUG_ON(nbytes > bio->bi_size);
2193 * not a complete bvec done
2195 if (unlikely(nbytes > nr_bytes)) {
2196 bio_nbytes += nr_bytes;
2197 total_bytes += nr_bytes;
2202 * advance to the next vector
2205 bio_nbytes += nbytes;
2208 total_bytes += nbytes;
2214 * end more in this run, or just return 'not-done'
2216 if (unlikely(nr_bytes <= 0))
2226 * Reset counters so that the request stacking driver
2227 * can find how many bytes remain in the request
2230 req->__data_len = 0;
2235 * if the request wasn't completed, update state
2238 req_bio_endio(req, bio, bio_nbytes, error);
2239 bio->bi_idx += next_idx;
2240 bio_iovec(bio)->bv_offset += nr_bytes;
2241 bio_iovec(bio)->bv_len -= nr_bytes;
2244 req->__data_len -= total_bytes;
2245 req->buffer = bio_data(req->bio);
2247 /* update sector only for requests with clear definition of sector */
2248 if (req->cmd_type == REQ_TYPE_FS || (req->cmd_flags & REQ_DISCARD))
2249 req->__sector += total_bytes >> 9;
2251 /* mixed attributes always follow the first bio */
2252 if (req->cmd_flags & REQ_MIXED_MERGE) {
2253 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2254 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2258 * If total number of sectors is less than the first segment
2259 * size, something has gone terribly wrong.
2261 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2262 printk(KERN_ERR "blk: request botched\n");
2263 req->__data_len = blk_rq_cur_bytes(req);
2266 /* recalculate the number of segments */
2267 blk_recalc_rq_segments(req);
2271 EXPORT_SYMBOL_GPL(blk_update_request);
2273 static bool blk_update_bidi_request(struct request *rq, int error,
2274 unsigned int nr_bytes,
2275 unsigned int bidi_bytes)
2277 if (blk_update_request(rq, error, nr_bytes))
2280 /* Bidi request must be completed as a whole */
2281 if (unlikely(blk_bidi_rq(rq)) &&
2282 blk_update_request(rq->next_rq, error, bidi_bytes))
2285 if (blk_queue_add_random(rq->q))
2286 add_disk_randomness(rq->rq_disk);
2292 * blk_unprep_request - unprepare a request
2295 * This function makes a request ready for complete resubmission (or
2296 * completion). It happens only after all error handling is complete,
2297 * so represents the appropriate moment to deallocate any resources
2298 * that were allocated to the request in the prep_rq_fn. The queue
2299 * lock is held when calling this.
2301 void blk_unprep_request(struct request *req)
2303 struct request_queue *q = req->q;
2305 req->cmd_flags &= ~REQ_DONTPREP;
2306 if (q->unprep_rq_fn)
2307 q->unprep_rq_fn(q, req);
2309 EXPORT_SYMBOL_GPL(blk_unprep_request);
2312 * queue lock must be held
2314 static void blk_finish_request(struct request *req, int error)
2316 if (blk_rq_tagged(req))
2317 blk_queue_end_tag(req->q, req);
2319 BUG_ON(blk_queued_rq(req));
2321 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2322 laptop_io_completion(&req->q->backing_dev_info);
2324 blk_delete_timer(req);
2326 if (req->cmd_flags & REQ_DONTPREP)
2327 blk_unprep_request(req);
2330 blk_account_io_done(req);
2333 req->end_io(req, error);
2335 if (blk_bidi_rq(req))
2336 __blk_put_request(req->next_rq->q, req->next_rq);
2338 __blk_put_request(req->q, req);
2343 * blk_end_bidi_request - Complete a bidi request
2344 * @rq: the request to complete
2345 * @error: %0 for success, < %0 for error
2346 * @nr_bytes: number of bytes to complete @rq
2347 * @bidi_bytes: number of bytes to complete @rq->next_rq
2350 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2351 * Drivers that supports bidi can safely call this member for any
2352 * type of request, bidi or uni. In the later case @bidi_bytes is
2356 * %false - we are done with this request
2357 * %true - still buffers pending for this request
2359 static bool blk_end_bidi_request(struct request *rq, int error,
2360 unsigned int nr_bytes, unsigned int bidi_bytes)
2362 struct request_queue *q = rq->q;
2363 unsigned long flags;
2365 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2368 spin_lock_irqsave(q->queue_lock, flags);
2369 blk_finish_request(rq, error);
2370 spin_unlock_irqrestore(q->queue_lock, flags);
2376 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2377 * @rq: the request to complete
2378 * @error: %0 for success, < %0 for error
2379 * @nr_bytes: number of bytes to complete @rq
2380 * @bidi_bytes: number of bytes to complete @rq->next_rq
2383 * Identical to blk_end_bidi_request() except that queue lock is
2384 * assumed to be locked on entry and remains so on return.
2387 * %false - we are done with this request
2388 * %true - still buffers pending for this request
2390 static bool __blk_end_bidi_request(struct request *rq, int error,
2391 unsigned int nr_bytes, unsigned int bidi_bytes)
2393 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2396 blk_finish_request(rq, error);
2402 * blk_end_request - Helper function for drivers to complete the request.
2403 * @rq: the request being processed
2404 * @error: %0 for success, < %0 for error
2405 * @nr_bytes: number of bytes to complete
2408 * Ends I/O on a number of bytes attached to @rq.
2409 * If @rq has leftover, sets it up for the next range of segments.
2412 * %false - we are done with this request
2413 * %true - still buffers pending for this request
2415 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2417 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2419 EXPORT_SYMBOL(blk_end_request);
2422 * blk_end_request_all - Helper function for drives to finish the request.
2423 * @rq: the request to finish
2424 * @error: %0 for success, < %0 for error
2427 * Completely finish @rq.
2429 void blk_end_request_all(struct request *rq, int error)
2432 unsigned int bidi_bytes = 0;
2434 if (unlikely(blk_bidi_rq(rq)))
2435 bidi_bytes = blk_rq_bytes(rq->next_rq);
2437 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2440 EXPORT_SYMBOL(blk_end_request_all);
2443 * blk_end_request_cur - Helper function to finish the current request chunk.
2444 * @rq: the request to finish the current chunk for
2445 * @error: %0 for success, < %0 for error
2448 * Complete the current consecutively mapped chunk from @rq.
2451 * %false - we are done with this request
2452 * %true - still buffers pending for this request
2454 bool blk_end_request_cur(struct request *rq, int error)
2456 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2458 EXPORT_SYMBOL(blk_end_request_cur);
2461 * blk_end_request_err - Finish a request till the next failure boundary.
2462 * @rq: the request to finish till the next failure boundary for
2463 * @error: must be negative errno
2466 * Complete @rq till the next failure boundary.
2469 * %false - we are done with this request
2470 * %true - still buffers pending for this request
2472 bool blk_end_request_err(struct request *rq, int error)
2474 WARN_ON(error >= 0);
2475 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2477 EXPORT_SYMBOL_GPL(blk_end_request_err);
2480 * __blk_end_request - Helper function for drivers to complete the request.
2481 * @rq: the request being processed
2482 * @error: %0 for success, < %0 for error
2483 * @nr_bytes: number of bytes to complete
2486 * Must be called with queue lock held unlike blk_end_request().
2489 * %false - we are done with this request
2490 * %true - still buffers pending for this request
2492 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2494 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2496 EXPORT_SYMBOL(__blk_end_request);
2499 * __blk_end_request_all - Helper function for drives to finish the request.
2500 * @rq: the request to finish
2501 * @error: %0 for success, < %0 for error
2504 * Completely finish @rq. Must be called with queue lock held.
2506 void __blk_end_request_all(struct request *rq, int error)
2509 unsigned int bidi_bytes = 0;
2511 if (unlikely(blk_bidi_rq(rq)))
2512 bidi_bytes = blk_rq_bytes(rq->next_rq);
2514 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2517 EXPORT_SYMBOL(__blk_end_request_all);
2520 * __blk_end_request_cur - Helper function to finish the current request chunk.
2521 * @rq: the request to finish the current chunk for
2522 * @error: %0 for success, < %0 for error
2525 * Complete the current consecutively mapped chunk from @rq. Must
2526 * be called with queue lock held.
2529 * %false - we are done with this request
2530 * %true - still buffers pending for this request
2532 bool __blk_end_request_cur(struct request *rq, int error)
2534 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2536 EXPORT_SYMBOL(__blk_end_request_cur);
2539 * __blk_end_request_err - Finish a request till the next failure boundary.
2540 * @rq: the request to finish till the next failure boundary for
2541 * @error: must be negative errno
2544 * Complete @rq till the next failure boundary. Must be called
2545 * with queue lock held.
2548 * %false - we are done with this request
2549 * %true - still buffers pending for this request
2551 bool __blk_end_request_err(struct request *rq, int error)
2553 WARN_ON(error >= 0);
2554 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2556 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2558 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2561 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2562 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
2564 if (bio_has_data(bio)) {
2565 rq->nr_phys_segments = bio_phys_segments(q, bio);
2566 rq->buffer = bio_data(bio);
2568 rq->__data_len = bio->bi_size;
2569 rq->bio = rq->biotail = bio;
2572 rq->rq_disk = bio->bi_bdev->bd_disk;
2575 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2577 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2578 * @rq: the request to be flushed
2581 * Flush all pages in @rq.
2583 void rq_flush_dcache_pages(struct request *rq)
2585 struct req_iterator iter;
2586 struct bio_vec *bvec;
2588 rq_for_each_segment(bvec, rq, iter)
2589 flush_dcache_page(bvec->bv_page);
2591 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2595 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2596 * @q : the queue of the device being checked
2599 * Check if underlying low-level drivers of a device are busy.
2600 * If the drivers want to export their busy state, they must set own
2601 * exporting function using blk_queue_lld_busy() first.
2603 * Basically, this function is used only by request stacking drivers
2604 * to stop dispatching requests to underlying devices when underlying
2605 * devices are busy. This behavior helps more I/O merging on the queue
2606 * of the request stacking driver and prevents I/O throughput regression
2607 * on burst I/O load.
2610 * 0 - Not busy (The request stacking driver should dispatch request)
2611 * 1 - Busy (The request stacking driver should stop dispatching request)
2613 int blk_lld_busy(struct request_queue *q)
2616 return q->lld_busy_fn(q);
2620 EXPORT_SYMBOL_GPL(blk_lld_busy);
2623 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2624 * @rq: the clone request to be cleaned up
2627 * Free all bios in @rq for a cloned request.
2629 void blk_rq_unprep_clone(struct request *rq)
2633 while ((bio = rq->bio) != NULL) {
2634 rq->bio = bio->bi_next;
2639 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2642 * Copy attributes of the original request to the clone request.
2643 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2645 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2647 dst->cpu = src->cpu;
2648 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
2649 dst->cmd_type = src->cmd_type;
2650 dst->__sector = blk_rq_pos(src);
2651 dst->__data_len = blk_rq_bytes(src);
2652 dst->nr_phys_segments = src->nr_phys_segments;
2653 dst->ioprio = src->ioprio;
2654 dst->extra_len = src->extra_len;
2658 * blk_rq_prep_clone - Helper function to setup clone request
2659 * @rq: the request to be setup
2660 * @rq_src: original request to be cloned
2661 * @bs: bio_set that bios for clone are allocated from
2662 * @gfp_mask: memory allocation mask for bio
2663 * @bio_ctr: setup function to be called for each clone bio.
2664 * Returns %0 for success, non %0 for failure.
2665 * @data: private data to be passed to @bio_ctr
2668 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2669 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2670 * are not copied, and copying such parts is the caller's responsibility.
2671 * Also, pages which the original bios are pointing to are not copied
2672 * and the cloned bios just point same pages.
2673 * So cloned bios must be completed before original bios, which means
2674 * the caller must complete @rq before @rq_src.
2676 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2677 struct bio_set *bs, gfp_t gfp_mask,
2678 int (*bio_ctr)(struct bio *, struct bio *, void *),
2681 struct bio *bio, *bio_src;
2686 blk_rq_init(NULL, rq);
2688 __rq_for_each_bio(bio_src, rq_src) {
2689 bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs);
2693 __bio_clone(bio, bio_src);
2695 if (bio_integrity(bio_src) &&
2696 bio_integrity_clone(bio, bio_src, gfp_mask, bs))
2699 if (bio_ctr && bio_ctr(bio, bio_src, data))
2703 rq->biotail->bi_next = bio;
2706 rq->bio = rq->biotail = bio;
2709 __blk_rq_prep_clone(rq, rq_src);
2716 blk_rq_unprep_clone(rq);
2720 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2722 int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
2724 return queue_work(kblockd_workqueue, work);
2726 EXPORT_SYMBOL(kblockd_schedule_work);
2728 int kblockd_schedule_delayed_work(struct request_queue *q,
2729 struct delayed_work *dwork, unsigned long delay)
2731 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2733 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2735 #define PLUG_MAGIC 0x91827364
2737 void blk_start_plug(struct blk_plug *plug)
2739 struct task_struct *tsk = current;
2741 plug->magic = PLUG_MAGIC;
2742 INIT_LIST_HEAD(&plug->list);
2743 plug->should_sort = 0;
2746 * If this is a nested plug, don't actually assign it. It will be
2747 * flushed on its own.
2751 * Store ordering should not be needed here, since a potential
2752 * preempt will imply a full memory barrier
2757 EXPORT_SYMBOL(blk_start_plug);
2759 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2761 struct request *rqa = container_of(a, struct request, queuelist);
2762 struct request *rqb = container_of(b, struct request, queuelist);
2764 return !(rqa->q == rqb->q);
2767 static void flush_plug_list(struct blk_plug *plug)
2769 struct request_queue *q;
2770 unsigned long flags;
2773 BUG_ON(plug->magic != PLUG_MAGIC);
2775 if (list_empty(&plug->list))
2778 if (plug->should_sort)
2779 list_sort(NULL, &plug->list, plug_rq_cmp);
2782 local_irq_save(flags);
2783 while (!list_empty(&plug->list)) {
2784 rq = list_entry_rq(plug->list.next);
2785 list_del_init(&rq->queuelist);
2786 BUG_ON(!(rq->cmd_flags & REQ_ON_PLUG));
2791 spin_unlock(q->queue_lock);
2794 spin_lock(q->queue_lock);
2796 rq->cmd_flags &= ~REQ_ON_PLUG;
2799 * rq is already accounted, so use raw insert
2801 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT, 0);
2806 spin_unlock(q->queue_lock);
2809 BUG_ON(!list_empty(&plug->list));
2810 local_irq_restore(flags);
2813 static void __blk_finish_plug(struct task_struct *tsk, struct blk_plug *plug)
2815 flush_plug_list(plug);
2817 if (plug == tsk->plug)
2821 void blk_finish_plug(struct blk_plug *plug)
2824 __blk_finish_plug(current, plug);
2826 EXPORT_SYMBOL(blk_finish_plug);
2828 void __blk_flush_plug(struct task_struct *tsk, struct blk_plug *plug)
2830 __blk_finish_plug(tsk, plug);
2833 EXPORT_SYMBOL(__blk_flush_plug);
2835 int __init blk_dev_init(void)
2837 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
2838 sizeof(((struct request *)0)->cmd_flags));
2840 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
2841 kblockd_workqueue = alloc_workqueue("kblockd",
2842 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
2843 if (!kblockd_workqueue)
2844 panic("Failed to create kblockd\n");
2846 request_cachep = kmem_cache_create("blkdev_requests",
2847 sizeof(struct request), 0, SLAB_PANIC, NULL);
2849 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
2850 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);