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
15 #ifdef CONFIG_BLOCK_PERF_FRAMEWORK
16 #define DRIVER_NAME "Block"
17 #define pr_fmt(fmt) DRIVER_NAME ": %s: " fmt, __func__
20 #include <linux/kernel.h>
21 #include <linux/module.h>
22 #include <linux/backing-dev.h>
23 #include <linux/bio.h>
24 #include <linux/blkdev.h>
25 #include <linux/blk-mq.h>
26 #include <linux/highmem.h>
28 #include <linux/kernel_stat.h>
29 #include <linux/string.h>
30 #include <linux/init.h>
31 #include <linux/completion.h>
32 #include <linux/slab.h>
33 #include <linux/swap.h>
34 #include <linux/writeback.h>
35 #include <linux/task_io_accounting_ops.h>
36 #include <linux/fault-inject.h>
37 #include <linux/list_sort.h>
38 #include <linux/delay.h>
39 #include <linux/ratelimit.h>
40 #include <linux/pm_runtime.h>
41 #include <linux/blk-cgroup.h>
43 #ifdef CONFIG_BLOCK_PERF_FRAMEWORK
44 #include <linux/ktime.h>
45 #include <linux/spinlock.h>
46 #include <linux/debugfs.h>
49 #define CREATE_TRACE_POINTS
50 #include <trace/events/block.h>
55 #include <linux/math64.h>
57 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
58 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
59 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
60 EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
61 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
63 DEFINE_IDA(blk_queue_ida);
66 * For the allocated request tables
68 struct kmem_cache *request_cachep = NULL;
71 * For queue allocation
73 struct kmem_cache *blk_requestq_cachep;
76 * Controlling structure to kblockd
78 static struct workqueue_struct *kblockd_workqueue;
80 static void blk_clear_congested(struct request_list *rl, int sync)
82 #ifdef CONFIG_CGROUP_WRITEBACK
83 clear_wb_congested(rl->blkg->wb_congested, sync);
86 * If !CGROUP_WRITEBACK, all blkg's map to bdi->wb and we shouldn't
87 * flip its congestion state for events on other blkcgs.
89 if (rl == &rl->q->root_rl)
90 clear_wb_congested(rl->q->backing_dev_info->wb.congested, sync);
94 static void blk_set_congested(struct request_list *rl, int sync)
96 #ifdef CONFIG_CGROUP_WRITEBACK
97 set_wb_congested(rl->blkg->wb_congested, sync);
99 /* see blk_clear_congested() */
100 if (rl == &rl->q->root_rl)
101 set_wb_congested(rl->q->backing_dev_info->wb.congested, sync);
105 void blk_queue_congestion_threshold(struct request_queue *q)
109 nr = q->nr_requests - (q->nr_requests / 8) + 1;
110 if (nr > q->nr_requests)
112 q->nr_congestion_on = nr;
114 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
117 q->nr_congestion_off = nr;
121 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
124 * Locates the passed device's request queue and returns the address of its
125 * backing_dev_info. The return value is never NULL however we may return
126 * &noop_backing_dev_info if the bdev is not currently open.
128 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
132 EXPORT_SYMBOL(blk_get_backing_dev_info);
134 void blk_rq_init(struct request_queue *q, struct request *rq)
136 memset(rq, 0, sizeof(*rq));
138 INIT_LIST_HEAD(&rq->queuelist);
139 INIT_LIST_HEAD(&rq->timeout_list);
142 rq->__sector = (sector_t) -1;
143 INIT_HLIST_NODE(&rq->hash);
144 RB_CLEAR_NODE(&rq->rb_node);
146 rq->cmd_len = BLK_MAX_CDB;
148 rq->start_time = jiffies;
149 set_start_time_ns(rq);
152 EXPORT_SYMBOL(blk_rq_init);
154 static void req_bio_endio(struct request *rq, struct bio *bio,
155 unsigned int nbytes, int error)
158 bio->bi_error = error;
160 if (unlikely(rq->cmd_flags & REQ_QUIET))
161 bio_set_flag(bio, BIO_QUIET);
163 bio_advance(bio, nbytes);
165 /* don't actually finish bio if it's part of flush sequence */
166 if (bio->bi_iter.bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
170 void blk_dump_rq_flags(struct request *rq, char *msg)
174 printk(KERN_INFO "%s: dev %s: type=%x, flags=%llx\n", msg,
175 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
176 (unsigned long long) rq->cmd_flags);
178 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
179 (unsigned long long)blk_rq_pos(rq),
180 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
181 printk(KERN_INFO " bio %p, biotail %p, len %u\n",
182 rq->bio, rq->biotail, blk_rq_bytes(rq));
184 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
185 printk(KERN_INFO " cdb: ");
186 for (bit = 0; bit < BLK_MAX_CDB; bit++)
187 printk("%02x ", rq->cmd[bit]);
191 EXPORT_SYMBOL(blk_dump_rq_flags);
193 static void blk_delay_work(struct work_struct *work)
195 struct request_queue *q;
197 q = container_of(work, struct request_queue, delay_work.work);
198 spin_lock_irq(q->queue_lock);
200 spin_unlock_irq(q->queue_lock);
204 * blk_delay_queue - restart queueing after defined interval
205 * @q: The &struct request_queue in question
206 * @msecs: Delay in msecs
209 * Sometimes queueing needs to be postponed for a little while, to allow
210 * resources to come back. This function will make sure that queueing is
211 * restarted around the specified time. Queue lock must be held.
213 void blk_delay_queue(struct request_queue *q, unsigned long msecs)
215 if (likely(!blk_queue_dead(q)))
216 queue_delayed_work(kblockd_workqueue, &q->delay_work,
217 msecs_to_jiffies(msecs));
219 EXPORT_SYMBOL(blk_delay_queue);
222 * blk_start_queue_async - asynchronously restart a previously stopped queue
223 * @q: The &struct request_queue in question
226 * blk_start_queue_async() will clear the stop flag on the queue, and
227 * ensure that the request_fn for the queue is run from an async
230 void blk_start_queue_async(struct request_queue *q)
232 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
233 blk_run_queue_async(q);
235 EXPORT_SYMBOL(blk_start_queue_async);
238 * blk_start_queue - restart a previously stopped queue
239 * @q: The &struct request_queue in question
242 * blk_start_queue() will clear the stop flag on the queue, and call
243 * the request_fn for the queue if it was in a stopped state when
244 * entered. Also see blk_stop_queue(). Queue lock must be held.
246 void blk_start_queue(struct request_queue *q)
248 WARN_ON(!in_interrupt() && !irqs_disabled());
250 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
253 EXPORT_SYMBOL(blk_start_queue);
256 * blk_stop_queue - stop a queue
257 * @q: The &struct request_queue in question
260 * The Linux block layer assumes that a block driver will consume all
261 * entries on the request queue when the request_fn strategy is called.
262 * Often this will not happen, because of hardware limitations (queue
263 * depth settings). If a device driver gets a 'queue full' response,
264 * or if it simply chooses not to queue more I/O at one point, it can
265 * call this function to prevent the request_fn from being called until
266 * the driver has signalled it's ready to go again. This happens by calling
267 * blk_start_queue() to restart queue operations. Queue lock must be held.
269 void blk_stop_queue(struct request_queue *q)
271 cancel_delayed_work(&q->delay_work);
272 queue_flag_set(QUEUE_FLAG_STOPPED, q);
274 EXPORT_SYMBOL(blk_stop_queue);
277 * blk_sync_queue - cancel any pending callbacks on a queue
281 * The block layer may perform asynchronous callback activity
282 * on a queue, such as calling the unplug function after a timeout.
283 * A block device may call blk_sync_queue to ensure that any
284 * such activity is cancelled, thus allowing it to release resources
285 * that the callbacks might use. The caller must already have made sure
286 * that its ->make_request_fn will not re-add plugging prior to calling
289 * This function does not cancel any asynchronous activity arising
290 * out of elevator or throttling code. That would require elevator_exit()
291 * and blkcg_exit_queue() to be called with queue lock initialized.
294 void blk_sync_queue(struct request_queue *q)
296 del_timer_sync(&q->timeout);
299 struct blk_mq_hw_ctx *hctx;
302 queue_for_each_hw_ctx(q, hctx, i) {
303 cancel_delayed_work_sync(&hctx->run_work);
304 cancel_delayed_work_sync(&hctx->delay_work);
307 cancel_delayed_work_sync(&q->delay_work);
310 EXPORT_SYMBOL(blk_sync_queue);
313 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
314 * @q: The queue to run
317 * Invoke request handling on a queue if there are any pending requests.
318 * May be used to restart request handling after a request has completed.
319 * This variant runs the queue whether or not the queue has been
320 * stopped. Must be called with the queue lock held and interrupts
321 * disabled. See also @blk_run_queue.
323 inline void __blk_run_queue_uncond(struct request_queue *q)
325 if (unlikely(blk_queue_dead(q)))
329 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
330 * the queue lock internally. As a result multiple threads may be
331 * running such a request function concurrently. Keep track of the
332 * number of active request_fn invocations such that blk_drain_queue()
333 * can wait until all these request_fn calls have finished.
335 q->request_fn_active++;
337 q->request_fn_active--;
339 EXPORT_SYMBOL_GPL(__blk_run_queue_uncond);
342 * __blk_run_queue - run a single device queue
343 * @q: The queue to run
346 * See @blk_run_queue. This variant must be called with the queue lock
347 * held and interrupts disabled.
349 void __blk_run_queue(struct request_queue *q)
351 if (unlikely(blk_queue_stopped(q)))
354 __blk_run_queue_uncond(q);
356 EXPORT_SYMBOL(__blk_run_queue);
359 * blk_run_queue_async - run a single device queue in workqueue context
360 * @q: The queue to run
363 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
364 * of us. The caller must hold the queue lock.
366 void blk_run_queue_async(struct request_queue *q)
368 if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
369 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
371 EXPORT_SYMBOL(blk_run_queue_async);
374 * blk_run_queue - run a single device queue
375 * @q: The queue to run
378 * Invoke request handling on this queue, if it has pending work to do.
379 * May be used to restart queueing when a request has completed.
381 void blk_run_queue(struct request_queue *q)
385 spin_lock_irqsave(q->queue_lock, flags);
387 spin_unlock_irqrestore(q->queue_lock, flags);
389 EXPORT_SYMBOL(blk_run_queue);
391 void blk_put_queue(struct request_queue *q)
393 kobject_put(&q->kobj);
395 EXPORT_SYMBOL(blk_put_queue);
398 * __blk_drain_queue - drain requests from request_queue
400 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
402 * Drain requests from @q. If @drain_all is set, all requests are drained.
403 * If not, only ELVPRIV requests are drained. The caller is responsible
404 * for ensuring that no new requests which need to be drained are queued.
406 static void __blk_drain_queue(struct request_queue *q, bool drain_all)
407 __releases(q->queue_lock)
408 __acquires(q->queue_lock)
412 lockdep_assert_held(q->queue_lock);
418 * The caller might be trying to drain @q before its
419 * elevator is initialized.
422 elv_drain_elevator(q);
424 blkcg_drain_queue(q);
427 * This function might be called on a queue which failed
428 * driver init after queue creation or is not yet fully
429 * active yet. Some drivers (e.g. fd and loop) get unhappy
430 * in such cases. Kick queue iff dispatch queue has
431 * something on it and @q has request_fn set.
433 if (!list_empty(&q->queue_head) && q->request_fn)
436 drain |= q->nr_rqs_elvpriv;
437 drain |= q->request_fn_active;
440 * Unfortunately, requests are queued at and tracked from
441 * multiple places and there's no single counter which can
442 * be drained. Check all the queues and counters.
445 struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
446 drain |= !list_empty(&q->queue_head);
447 for (i = 0; i < 2; i++) {
448 drain |= q->nr_rqs[i];
449 drain |= q->in_flight[i];
451 drain |= !list_empty(&fq->flush_queue[i]);
458 spin_unlock_irq(q->queue_lock);
462 spin_lock_irq(q->queue_lock);
466 * With queue marked dead, any woken up waiter will fail the
467 * allocation path, so the wakeup chaining is lost and we're
468 * left with hung waiters. We need to wake up those waiters.
471 struct request_list *rl;
473 blk_queue_for_each_rl(rl, q)
474 for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
475 wake_up_all(&rl->wait[i]);
480 * blk_queue_bypass_start - enter queue bypass mode
481 * @q: queue of interest
483 * In bypass mode, only the dispatch FIFO queue of @q is used. This
484 * function makes @q enter bypass mode and drains all requests which were
485 * throttled or issued before. On return, it's guaranteed that no request
486 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
487 * inside queue or RCU read lock.
489 void blk_queue_bypass_start(struct request_queue *q)
491 spin_lock_irq(q->queue_lock);
493 queue_flag_set(QUEUE_FLAG_BYPASS, q);
494 spin_unlock_irq(q->queue_lock);
497 * Queues start drained. Skip actual draining till init is
498 * complete. This avoids lenghty delays during queue init which
499 * can happen many times during boot.
501 if (blk_queue_init_done(q)) {
502 spin_lock_irq(q->queue_lock);
503 __blk_drain_queue(q, false);
504 spin_unlock_irq(q->queue_lock);
506 /* ensure blk_queue_bypass() is %true inside RCU read lock */
510 EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
513 * blk_queue_bypass_end - leave queue bypass mode
514 * @q: queue of interest
516 * Leave bypass mode and restore the normal queueing behavior.
518 void blk_queue_bypass_end(struct request_queue *q)
520 spin_lock_irq(q->queue_lock);
521 if (!--q->bypass_depth)
522 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
523 WARN_ON_ONCE(q->bypass_depth < 0);
524 spin_unlock_irq(q->queue_lock);
526 EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
528 void blk_set_queue_dying(struct request_queue *q)
530 spin_lock_irq(q->queue_lock);
531 queue_flag_set(QUEUE_FLAG_DYING, q);
532 spin_unlock_irq(q->queue_lock);
535 blk_mq_wake_waiters(q);
537 struct request_list *rl;
539 blk_queue_for_each_rl(rl, q) {
541 wake_up_all(&rl->wait[BLK_RW_SYNC]);
542 wake_up_all(&rl->wait[BLK_RW_ASYNC]);
547 EXPORT_SYMBOL_GPL(blk_set_queue_dying);
550 * blk_cleanup_queue - shutdown a request queue
551 * @q: request queue to shutdown
553 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
554 * put it. All future requests will be failed immediately with -ENODEV.
556 void blk_cleanup_queue(struct request_queue *q)
558 spinlock_t *lock = q->queue_lock;
560 /* mark @q DYING, no new request or merges will be allowed afterwards */
561 mutex_lock(&q->sysfs_lock);
562 blk_set_queue_dying(q);
566 * A dying queue is permanently in bypass mode till released. Note
567 * that, unlike blk_queue_bypass_start(), we aren't performing
568 * synchronize_rcu() after entering bypass mode to avoid the delay
569 * as some drivers create and destroy a lot of queues while
570 * probing. This is still safe because blk_release_queue() will be
571 * called only after the queue refcnt drops to zero and nothing,
572 * RCU or not, would be traversing the queue by then.
575 queue_flag_set(QUEUE_FLAG_BYPASS, q);
577 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
578 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
579 queue_flag_set(QUEUE_FLAG_DYING, q);
580 spin_unlock_irq(lock);
581 mutex_unlock(&q->sysfs_lock);
584 * Drain all requests queued before DYING marking. Set DEAD flag to
585 * prevent that q->request_fn() gets invoked after draining finished.
590 __blk_drain_queue(q, true);
591 queue_flag_set(QUEUE_FLAG_DEAD, q);
592 spin_unlock_irq(lock);
594 /* for synchronous bio-based driver finish in-flight integrity i/o */
595 blk_flush_integrity();
597 /* @q won't process any more request, flush async actions */
598 del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer);
602 blk_mq_free_queue(q);
603 percpu_ref_exit(&q->q_usage_counter);
606 if (q->queue_lock != &q->__queue_lock)
607 q->queue_lock = &q->__queue_lock;
608 spin_unlock_irq(lock);
610 /* @q is and will stay empty, shutdown and put */
613 EXPORT_SYMBOL(blk_cleanup_queue);
615 /* Allocate memory local to the request queue */
616 static void *alloc_request_struct(gfp_t gfp_mask, void *data)
618 int nid = (int)(long)data;
619 return kmem_cache_alloc_node(request_cachep, gfp_mask, nid);
622 static void free_request_struct(void *element, void *unused)
624 kmem_cache_free(request_cachep, element);
627 int blk_init_rl(struct request_list *rl, struct request_queue *q,
630 if (unlikely(rl->rq_pool))
634 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
635 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
636 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
637 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
639 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, alloc_request_struct,
641 (void *)(long)q->node, gfp_mask,
649 void blk_exit_rl(struct request_list *rl)
652 mempool_destroy(rl->rq_pool);
655 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
657 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
659 EXPORT_SYMBOL(blk_alloc_queue);
661 int blk_queue_enter(struct request_queue *q, gfp_t gfp)
664 if (percpu_ref_tryget_live(&q->q_usage_counter))
667 if (!gfpflags_allow_blocking(gfp))
670 wait_event(q->mq_freeze_wq,
671 !atomic_read(&q->mq_freeze_depth) ||
673 if (blk_queue_dying(q))
678 void blk_queue_exit(struct request_queue *q)
680 percpu_ref_put(&q->q_usage_counter);
683 static void blk_queue_usage_counter_release(struct percpu_ref *ref)
685 struct request_queue *q =
686 container_of(ref, struct request_queue, q_usage_counter);
688 wake_up_all(&q->mq_freeze_wq);
691 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
693 struct request_queue *q;
695 q = kmem_cache_alloc_node(blk_requestq_cachep,
696 gfp_mask | __GFP_ZERO, node_id);
700 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
704 q->bio_split = bioset_create(BIO_POOL_SIZE, 0);
708 q->backing_dev_info = bdi_alloc_node(gfp_mask, node_id);
709 if (!q->backing_dev_info)
712 q->backing_dev_info->ra_pages =
713 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
714 q->backing_dev_info->capabilities = BDI_CAP_CGROUP_WRITEBACK;
715 q->backing_dev_info->name = "block";
718 setup_timer(&q->backing_dev_info->laptop_mode_wb_timer,
719 laptop_mode_timer_fn, (unsigned long) q);
720 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
721 INIT_LIST_HEAD(&q->queue_head);
722 INIT_LIST_HEAD(&q->timeout_list);
723 INIT_LIST_HEAD(&q->icq_list);
724 #ifdef CONFIG_BLK_CGROUP
725 INIT_LIST_HEAD(&q->blkg_list);
727 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
729 kobject_init(&q->kobj, &blk_queue_ktype);
731 mutex_init(&q->sysfs_lock);
732 spin_lock_init(&q->__queue_lock);
735 * By default initialize queue_lock to internal lock and driver can
736 * override it later if need be.
738 q->queue_lock = &q->__queue_lock;
741 * A queue starts its life with bypass turned on to avoid
742 * unnecessary bypass on/off overhead and nasty surprises during
743 * init. The initial bypass will be finished when the queue is
744 * registered by blk_register_queue().
747 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
749 init_waitqueue_head(&q->mq_freeze_wq);
752 * Init percpu_ref in atomic mode so that it's faster to shutdown.
753 * See blk_register_queue() for details.
755 if (percpu_ref_init(&q->q_usage_counter,
756 blk_queue_usage_counter_release,
757 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
760 if (blkcg_init_queue(q))
766 percpu_ref_exit(&q->q_usage_counter);
768 bdi_put(q->backing_dev_info);
770 bioset_free(q->bio_split);
772 ida_simple_remove(&blk_queue_ida, q->id);
774 kmem_cache_free(blk_requestq_cachep, q);
777 EXPORT_SYMBOL(blk_alloc_queue_node);
780 * blk_init_queue - prepare a request queue for use with a block device
781 * @rfn: The function to be called to process requests that have been
782 * placed on the queue.
783 * @lock: Request queue spin lock
786 * If a block device wishes to use the standard request handling procedures,
787 * which sorts requests and coalesces adjacent requests, then it must
788 * call blk_init_queue(). The function @rfn will be called when there
789 * are requests on the queue that need to be processed. If the device
790 * supports plugging, then @rfn may not be called immediately when requests
791 * are available on the queue, but may be called at some time later instead.
792 * Plugged queues are generally unplugged when a buffer belonging to one
793 * of the requests on the queue is needed, or due to memory pressure.
795 * @rfn is not required, or even expected, to remove all requests off the
796 * queue, but only as many as it can handle at a time. If it does leave
797 * requests on the queue, it is responsible for arranging that the requests
798 * get dealt with eventually.
800 * The queue spin lock must be held while manipulating the requests on the
801 * request queue; this lock will be taken also from interrupt context, so irq
802 * disabling is needed for it.
804 * Function returns a pointer to the initialized request queue, or %NULL if
808 * blk_init_queue() must be paired with a blk_cleanup_queue() call
809 * when the block device is deactivated (such as at module unload).
812 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
814 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
816 EXPORT_SYMBOL(blk_init_queue);
818 struct request_queue *
819 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
821 struct request_queue *uninit_q, *q;
823 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
827 q = blk_init_allocated_queue(uninit_q, rfn, lock);
829 blk_cleanup_queue(uninit_q);
833 EXPORT_SYMBOL(blk_init_queue_node);
835 static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio);
837 struct request_queue *
838 blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
844 q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, 0);
848 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
852 q->prep_rq_fn = NULL;
853 q->unprep_rq_fn = NULL;
854 q->queue_flags |= QUEUE_FLAG_DEFAULT;
856 /* Override internal queue lock with supplied lock pointer */
858 q->queue_lock = lock;
861 * This also sets hw/phys segments, boundary and size
863 blk_queue_make_request(q, blk_queue_bio);
865 q->sg_reserved_size = INT_MAX;
867 /* Protect q->elevator from elevator_change */
868 mutex_lock(&q->sysfs_lock);
871 if (elevator_init(q, NULL)) {
872 mutex_unlock(&q->sysfs_lock);
876 mutex_unlock(&q->sysfs_lock);
881 blk_free_flush_queue(q->fq);
884 EXPORT_SYMBOL(blk_init_allocated_queue);
886 bool blk_get_queue(struct request_queue *q)
888 if (likely(!blk_queue_dying(q))) {
895 EXPORT_SYMBOL(blk_get_queue);
897 static inline void blk_free_request(struct request_list *rl, struct request *rq)
899 if (rq->cmd_flags & REQ_ELVPRIV) {
900 elv_put_request(rl->q, rq);
902 put_io_context(rq->elv.icq->ioc);
905 mempool_free(rq, rl->rq_pool);
909 * ioc_batching returns true if the ioc is a valid batching request and
910 * should be given priority access to a request.
912 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
918 * Make sure the process is able to allocate at least 1 request
919 * even if the batch times out, otherwise we could theoretically
922 return ioc->nr_batch_requests == q->nr_batching ||
923 (ioc->nr_batch_requests > 0
924 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
928 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
929 * will cause the process to be a "batcher" on all queues in the system. This
930 * is the behaviour we want though - once it gets a wakeup it should be given
933 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
935 if (!ioc || ioc_batching(q, ioc))
938 ioc->nr_batch_requests = q->nr_batching;
939 ioc->last_waited = jiffies;
942 static void __freed_request(struct request_list *rl, int sync)
944 struct request_queue *q = rl->q;
946 if (rl->count[sync] < queue_congestion_off_threshold(q))
947 blk_clear_congested(rl, sync);
949 if (rl->count[sync] + 1 <= q->nr_requests) {
950 if (waitqueue_active(&rl->wait[sync]))
951 wake_up(&rl->wait[sync]);
953 blk_clear_rl_full(rl, sync);
958 * A request has just been released. Account for it, update the full and
959 * congestion status, wake up any waiters. Called under q->queue_lock.
961 static void freed_request(struct request_list *rl, unsigned int flags)
963 struct request_queue *q = rl->q;
964 int sync = rw_is_sync(flags);
968 if (flags & REQ_ELVPRIV)
971 __freed_request(rl, sync);
973 if (unlikely(rl->starved[sync ^ 1]))
974 __freed_request(rl, sync ^ 1);
977 int blk_update_nr_requests(struct request_queue *q, unsigned int nr)
979 struct request_list *rl;
980 int on_thresh, off_thresh;
982 spin_lock_irq(q->queue_lock);
984 blk_queue_congestion_threshold(q);
985 on_thresh = queue_congestion_on_threshold(q);
986 off_thresh = queue_congestion_off_threshold(q);
988 blk_queue_for_each_rl(rl, q) {
989 if (rl->count[BLK_RW_SYNC] >= on_thresh)
990 blk_set_congested(rl, BLK_RW_SYNC);
991 else if (rl->count[BLK_RW_SYNC] < off_thresh)
992 blk_clear_congested(rl, BLK_RW_SYNC);
994 if (rl->count[BLK_RW_ASYNC] >= on_thresh)
995 blk_set_congested(rl, BLK_RW_ASYNC);
996 else if (rl->count[BLK_RW_ASYNC] < off_thresh)
997 blk_clear_congested(rl, BLK_RW_ASYNC);
999 if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
1000 blk_set_rl_full(rl, BLK_RW_SYNC);
1002 blk_clear_rl_full(rl, BLK_RW_SYNC);
1003 wake_up(&rl->wait[BLK_RW_SYNC]);
1006 if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
1007 blk_set_rl_full(rl, BLK_RW_ASYNC);
1009 blk_clear_rl_full(rl, BLK_RW_ASYNC);
1010 wake_up(&rl->wait[BLK_RW_ASYNC]);
1014 spin_unlock_irq(q->queue_lock);
1019 * Determine if elevator data should be initialized when allocating the
1020 * request associated with @bio.
1022 static bool blk_rq_should_init_elevator(struct bio *bio)
1028 * Flush requests do not use the elevator so skip initialization.
1029 * This allows a request to share the flush and elevator data.
1031 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
1038 * rq_ioc - determine io_context for request allocation
1039 * @bio: request being allocated is for this bio (can be %NULL)
1041 * Determine io_context to use for request allocation for @bio. May return
1042 * %NULL if %current->io_context doesn't exist.
1044 static struct io_context *rq_ioc(struct bio *bio)
1046 #ifdef CONFIG_BLK_CGROUP
1047 if (bio && bio->bi_ioc)
1050 return current->io_context;
1054 * __get_request - get a free request
1055 * @rl: request list to allocate from
1056 * @rw_flags: RW and SYNC flags
1057 * @bio: bio to allocate request for (can be %NULL)
1058 * @gfp_mask: allocation mask
1060 * Get a free request from @q. This function may fail under memory
1061 * pressure or if @q is dead.
1063 * Must be called with @q->queue_lock held and,
1064 * Returns ERR_PTR on failure, with @q->queue_lock held.
1065 * Returns request pointer on success, with @q->queue_lock *not held*.
1067 static struct request *__get_request(struct request_list *rl, int rw_flags,
1068 struct bio *bio, gfp_t gfp_mask)
1070 struct request_queue *q = rl->q;
1072 struct elevator_type *et = q->elevator->type;
1073 struct io_context *ioc = rq_ioc(bio);
1074 struct io_cq *icq = NULL;
1075 const bool is_sync = rw_is_sync(rw_flags) != 0;
1078 if (unlikely(blk_queue_dying(q)))
1079 return ERR_PTR(-ENODEV);
1081 may_queue = elv_may_queue(q, rw_flags);
1082 if (may_queue == ELV_MQUEUE_NO)
1085 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
1086 if (rl->count[is_sync]+1 >= q->nr_requests) {
1088 * The queue will fill after this allocation, so set
1089 * it as full, and mark this process as "batching".
1090 * This process will be allowed to complete a batch of
1091 * requests, others will be blocked.
1093 if (!blk_rl_full(rl, is_sync)) {
1094 ioc_set_batching(q, ioc);
1095 blk_set_rl_full(rl, is_sync);
1097 if (may_queue != ELV_MQUEUE_MUST
1098 && !ioc_batching(q, ioc)) {
1100 * The queue is full and the allocating
1101 * process is not a "batcher", and not
1102 * exempted by the IO scheduler
1104 return ERR_PTR(-ENOMEM);
1108 blk_set_congested(rl, is_sync);
1112 * Only allow batching queuers to allocate up to 50% over the defined
1113 * limit of requests, otherwise we could have thousands of requests
1114 * allocated with any setting of ->nr_requests
1116 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
1117 return ERR_PTR(-ENOMEM);
1119 q->nr_rqs[is_sync]++;
1120 rl->count[is_sync]++;
1121 rl->starved[is_sync] = 0;
1124 * Decide whether the new request will be managed by elevator. If
1125 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
1126 * prevent the current elevator from being destroyed until the new
1127 * request is freed. This guarantees icq's won't be destroyed and
1128 * makes creating new ones safe.
1130 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1131 * it will be created after releasing queue_lock.
1133 if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
1134 rw_flags |= REQ_ELVPRIV;
1135 q->nr_rqs_elvpriv++;
1136 if (et->icq_cache && ioc)
1137 icq = ioc_lookup_icq(ioc, q);
1140 if (blk_queue_io_stat(q))
1141 rw_flags |= REQ_IO_STAT;
1142 spin_unlock_irq(q->queue_lock);
1144 /* allocate and init request */
1145 rq = mempool_alloc(rl->rq_pool, gfp_mask);
1150 blk_rq_set_rl(rq, rl);
1151 rq->cmd_flags = rw_flags | REQ_ALLOCED;
1154 if (rw_flags & REQ_ELVPRIV) {
1155 if (unlikely(et->icq_cache && !icq)) {
1157 icq = ioc_create_icq(ioc, q, gfp_mask);
1163 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1166 /* @rq->elv.icq holds io_context until @rq is freed */
1168 get_io_context(icq->ioc);
1172 * ioc may be NULL here, and ioc_batching will be false. That's
1173 * OK, if the queue is under the request limit then requests need
1174 * not count toward the nr_batch_requests limit. There will always
1175 * be some limit enforced by BLK_BATCH_TIME.
1177 if (ioc_batching(q, ioc))
1178 ioc->nr_batch_requests--;
1180 trace_block_getrq(q, bio, rw_flags & 1);
1185 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1186 * and may fail indefinitely under memory pressure and thus
1187 * shouldn't stall IO. Treat this request as !elvpriv. This will
1188 * disturb iosched and blkcg but weird is bettern than dead.
1190 printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1191 __func__, dev_name(q->backing_dev_info->dev));
1193 rq->cmd_flags &= ~REQ_ELVPRIV;
1196 spin_lock_irq(q->queue_lock);
1197 q->nr_rqs_elvpriv--;
1198 spin_unlock_irq(q->queue_lock);
1203 * Allocation failed presumably due to memory. Undo anything we
1204 * might have messed up.
1206 * Allocating task should really be put onto the front of the wait
1207 * queue, but this is pretty rare.
1209 spin_lock_irq(q->queue_lock);
1210 freed_request(rl, rw_flags);
1213 * in the very unlikely event that allocation failed and no
1214 * requests for this direction was pending, mark us starved so that
1215 * freeing of a request in the other direction will notice
1216 * us. another possible fix would be to split the rq mempool into
1220 if (unlikely(rl->count[is_sync] == 0))
1221 rl->starved[is_sync] = 1;
1222 return ERR_PTR(-ENOMEM);
1226 * get_request - get a free request
1227 * @q: request_queue to allocate request from
1228 * @rw_flags: RW and SYNC flags
1229 * @bio: bio to allocate request for (can be %NULL)
1230 * @gfp_mask: allocation mask
1232 * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask,
1233 * this function keeps retrying under memory pressure and fails iff @q is dead.
1235 * Must be called with @q->queue_lock held and,
1236 * Returns ERR_PTR on failure, with @q->queue_lock held.
1237 * Returns request pointer on success, with @q->queue_lock *not held*.
1239 static struct request *get_request(struct request_queue *q, int rw_flags,
1240 struct bio *bio, gfp_t gfp_mask)
1242 const bool is_sync = rw_is_sync(rw_flags) != 0;
1244 struct request_list *rl;
1247 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
1249 rq = __get_request(rl, rw_flags, bio, gfp_mask);
1253 if (!gfpflags_allow_blocking(gfp_mask) || unlikely(blk_queue_dying(q))) {
1258 /* wait on @rl and retry */
1259 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1260 TASK_UNINTERRUPTIBLE);
1262 trace_block_sleeprq(q, bio, rw_flags & 1);
1264 spin_unlock_irq(q->queue_lock);
1268 * After sleeping, we become a "batching" process and will be able
1269 * to allocate at least one request, and up to a big batch of them
1270 * for a small period time. See ioc_batching, ioc_set_batching
1272 ioc_set_batching(q, current->io_context);
1274 spin_lock_irq(q->queue_lock);
1275 finish_wait(&rl->wait[is_sync], &wait);
1280 static struct request *blk_old_get_request(struct request_queue *q, int rw,
1285 /* create ioc upfront */
1286 create_io_context(gfp_mask, q->node);
1288 spin_lock_irq(q->queue_lock);
1289 rq = get_request(q, rw, NULL, gfp_mask);
1291 spin_unlock_irq(q->queue_lock);
1292 /* q->queue_lock is unlocked at this point */
1297 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1300 return blk_mq_alloc_request(q, rw, gfp_mask, false);
1302 return blk_old_get_request(q, rw, gfp_mask);
1304 EXPORT_SYMBOL(blk_get_request);
1307 * blk_make_request - given a bio, allocate a corresponding struct request.
1308 * @q: target request queue
1309 * @bio: The bio describing the memory mappings that will be submitted for IO.
1310 * It may be a chained-bio properly constructed by block/bio layer.
1311 * @gfp_mask: gfp flags to be used for memory allocation
1313 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1314 * type commands. Where the struct request needs to be farther initialized by
1315 * the caller. It is passed a &struct bio, which describes the memory info of
1318 * The caller of blk_make_request must make sure that bi_io_vec
1319 * are set to describe the memory buffers. That bio_data_dir() will return
1320 * the needed direction of the request. (And all bio's in the passed bio-chain
1321 * are properly set accordingly)
1323 * If called under none-sleepable conditions, mapped bio buffers must not
1324 * need bouncing, by calling the appropriate masked or flagged allocator,
1325 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1328 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1329 * given to how you allocate bios. In particular, you cannot use
1330 * __GFP_DIRECT_RECLAIM for anything but the first bio in the chain. Otherwise
1331 * you risk waiting for IO completion of a bio that hasn't been submitted yet,
1332 * thus resulting in a deadlock. Alternatively bios should be allocated using
1333 * bio_kmalloc() instead of bio_alloc(), as that avoids the mempool deadlock.
1334 * If possible a big IO should be split into smaller parts when allocation
1335 * fails. Partial allocation should not be an error, or you risk a live-lock.
1337 struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1340 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1345 blk_rq_set_block_pc(rq);
1348 struct bio *bounce_bio = bio;
1351 blk_queue_bounce(q, &bounce_bio);
1352 ret = blk_rq_append_bio(q, rq, bounce_bio);
1353 if (unlikely(ret)) {
1354 blk_put_request(rq);
1355 return ERR_PTR(ret);
1361 EXPORT_SYMBOL(blk_make_request);
1364 * blk_rq_set_block_pc - initialize a request to type BLOCK_PC
1365 * @rq: request to be initialized
1368 void blk_rq_set_block_pc(struct request *rq)
1370 rq->cmd_type = REQ_TYPE_BLOCK_PC;
1372 rq->__sector = (sector_t) -1;
1373 rq->bio = rq->biotail = NULL;
1374 memset(rq->__cmd, 0, sizeof(rq->__cmd));
1376 EXPORT_SYMBOL(blk_rq_set_block_pc);
1379 * blk_requeue_request - put a request back on queue
1380 * @q: request queue where request should be inserted
1381 * @rq: request to be inserted
1384 * Drivers often keep queueing requests until the hardware cannot accept
1385 * more, when that condition happens we need to put the request back
1386 * on the queue. Must be called with queue lock held.
1388 void blk_requeue_request(struct request_queue *q, struct request *rq)
1390 blk_delete_timer(rq);
1391 blk_clear_rq_complete(rq);
1392 trace_block_rq_requeue(q, rq);
1394 if (rq->cmd_flags & REQ_QUEUED)
1395 blk_queue_end_tag(q, rq);
1397 BUG_ON(blk_queued_rq(rq));
1399 elv_requeue_request(q, rq);
1401 EXPORT_SYMBOL(blk_requeue_request);
1403 static void add_acct_request(struct request_queue *q, struct request *rq,
1406 blk_account_io_start(rq, true);
1407 __elv_add_request(q, rq, where);
1410 static void part_round_stats_single(int cpu, struct hd_struct *part,
1415 if (now == part->stamp)
1418 inflight = part_in_flight(part);
1420 __part_stat_add(cpu, part, time_in_queue,
1421 inflight * (now - part->stamp));
1422 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1428 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1429 * @cpu: cpu number for stats access
1430 * @part: target partition
1432 * The average IO queue length and utilisation statistics are maintained
1433 * by observing the current state of the queue length and the amount of
1434 * time it has been in this state for.
1436 * Normally, that accounting is done on IO completion, but that can result
1437 * in more than a second's worth of IO being accounted for within any one
1438 * second, leading to >100% utilisation. To deal with that, we call this
1439 * function to do a round-off before returning the results when reading
1440 * /proc/diskstats. This accounts immediately for all queue usage up to
1441 * the current jiffies and restarts the counters again.
1443 void part_round_stats(int cpu, struct hd_struct *part)
1445 unsigned long now = jiffies;
1448 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1449 part_round_stats_single(cpu, part, now);
1451 EXPORT_SYMBOL_GPL(part_round_stats);
1454 static void blk_pm_put_request(struct request *rq)
1456 if (rq->q->dev && !(rq->cmd_flags & REQ_PM) && rq->q->nr_pending) {
1457 if (!--rq->q->nr_pending)
1458 pm_runtime_mark_last_busy(rq->q->dev);
1462 static inline void blk_pm_put_request(struct request *rq) {}
1466 * queue lock must be held
1468 void __blk_put_request(struct request_queue *q, struct request *req)
1474 blk_mq_free_request(req);
1478 blk_pm_put_request(req);
1480 elv_completed_request(q, req);
1482 /* this is a bio leak */
1483 WARN_ON(req->bio != NULL);
1485 /* this is a bio leak if the bio is not tagged with BIO_DONTFREE */
1486 WARN_ON(req->bio && !bio_flagged(req->bio, BIO_DONTFREE));
1489 * Request may not have originated from ll_rw_blk. if not,
1490 * it didn't come out of our reserved rq pools
1492 if (req->cmd_flags & REQ_ALLOCED) {
1493 unsigned int flags = req->cmd_flags;
1494 struct request_list *rl = blk_rq_rl(req);
1496 BUG_ON(!list_empty(&req->queuelist));
1497 BUG_ON(ELV_ON_HASH(req));
1499 blk_free_request(rl, req);
1500 freed_request(rl, flags);
1504 EXPORT_SYMBOL_GPL(__blk_put_request);
1506 void blk_put_request(struct request *req)
1508 struct request_queue *q = req->q;
1511 blk_mq_free_request(req);
1513 unsigned long flags;
1515 spin_lock_irqsave(q->queue_lock, flags);
1516 __blk_put_request(q, req);
1517 spin_unlock_irqrestore(q->queue_lock, flags);
1520 EXPORT_SYMBOL(blk_put_request);
1523 * blk_add_request_payload - add a payload to a request
1524 * @rq: request to update
1525 * @page: page backing the payload
1526 * @len: length of the payload.
1528 * This allows to later add a payload to an already submitted request by
1529 * a block driver. The driver needs to take care of freeing the payload
1532 * Note that this is a quite horrible hack and nothing but handling of
1533 * discard requests should ever use it.
1535 void blk_add_request_payload(struct request *rq, struct page *page,
1538 struct bio *bio = rq->bio;
1540 bio->bi_io_vec->bv_page = page;
1541 bio->bi_io_vec->bv_offset = 0;
1542 bio->bi_io_vec->bv_len = len;
1544 bio->bi_iter.bi_size = len;
1546 bio->bi_phys_segments = 1;
1548 rq->__data_len = rq->resid_len = len;
1549 rq->nr_phys_segments = 1;
1551 EXPORT_SYMBOL_GPL(blk_add_request_payload);
1553 bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1556 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1558 if (!ll_back_merge_fn(q, req, bio))
1561 trace_block_bio_backmerge(q, req, bio);
1563 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1564 blk_rq_set_mixed_merge(req);
1566 req->biotail->bi_next = bio;
1568 req->__data_len += bio->bi_iter.bi_size;
1569 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1571 blk_account_io_start(req, false);
1575 bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1578 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1580 if (!ll_front_merge_fn(q, req, bio))
1583 trace_block_bio_frontmerge(q, req, bio);
1585 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1586 blk_rq_set_mixed_merge(req);
1588 bio->bi_next = req->bio;
1591 req->__sector = bio->bi_iter.bi_sector;
1592 req->__data_len += bio->bi_iter.bi_size;
1593 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1595 blk_account_io_start(req, false);
1600 * blk_attempt_plug_merge - try to merge with %current's plugged list
1601 * @q: request_queue new bio is being queued at
1602 * @bio: new bio being queued
1603 * @request_count: out parameter for number of traversed plugged requests
1604 * @same_queue_rq: pointer to &struct request that gets filled in when
1605 * another request associated with @q is found on the plug list
1606 * (optional, may be %NULL)
1608 * Determine whether @bio being queued on @q can be merged with a request
1609 * on %current's plugged list. Returns %true if merge was successful,
1612 * Plugging coalesces IOs from the same issuer for the same purpose without
1613 * going through @q->queue_lock. As such it's more of an issuing mechanism
1614 * than scheduling, and the request, while may have elvpriv data, is not
1615 * added on the elevator at this point. In addition, we don't have
1616 * reliable access to the elevator outside queue lock. Only check basic
1617 * merging parameters without querying the elevator.
1619 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1621 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1622 unsigned int *request_count,
1623 struct request **same_queue_rq)
1625 struct blk_plug *plug;
1628 struct list_head *plug_list;
1630 plug = current->plug;
1636 plug_list = &plug->mq_list;
1638 plug_list = &plug->list;
1640 list_for_each_entry_reverse(rq, plug_list, queuelist) {
1646 * Only blk-mq multiple hardware queues case checks the
1647 * rq in the same queue, there should be only one such
1651 *same_queue_rq = rq;
1654 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
1657 el_ret = blk_try_merge(rq, bio);
1658 if (el_ret == ELEVATOR_BACK_MERGE) {
1659 ret = bio_attempt_back_merge(q, rq, bio);
1662 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1663 ret = bio_attempt_front_merge(q, rq, bio);
1672 unsigned int blk_plug_queued_count(struct request_queue *q)
1674 struct blk_plug *plug;
1676 struct list_head *plug_list;
1677 unsigned int ret = 0;
1679 plug = current->plug;
1684 plug_list = &plug->mq_list;
1686 plug_list = &plug->list;
1688 list_for_each_entry(rq, plug_list, queuelist) {
1696 void init_request_from_bio(struct request *req, struct bio *bio)
1698 req->cmd_type = REQ_TYPE_FS;
1700 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1701 if (bio->bi_rw & REQ_RAHEAD)
1702 req->cmd_flags |= REQ_FAILFAST_MASK;
1705 req->__sector = bio->bi_iter.bi_sector;
1706 req->ioprio = bio_prio(bio);
1707 blk_rq_bio_prep(req->q, req, bio);
1709 EXPORT_SYMBOL(init_request_from_bio);
1711 static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio)
1713 const bool sync = !!(bio->bi_rw & REQ_SYNC);
1714 struct blk_plug *plug;
1715 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1716 struct request *req;
1717 unsigned int request_count = 0;
1720 * low level driver can indicate that it wants pages above a
1721 * certain limit bounced to low memory (ie for highmem, or even
1722 * ISA dma in theory)
1724 blk_queue_bounce(q, &bio);
1726 blk_queue_split(q, &bio, q->bio_split);
1728 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
1729 bio->bi_error = -EIO;
1731 return BLK_QC_T_NONE;
1734 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA | REQ_POST_FLUSH_BARRIER |
1736 spin_lock_irq(q->queue_lock);
1737 where = ELEVATOR_INSERT_FLUSH;
1742 * Check if we can merge with the plugged list before grabbing
1745 if (!blk_queue_nomerges(q)) {
1746 if (blk_attempt_plug_merge(q, bio, &request_count, NULL))
1747 return BLK_QC_T_NONE;
1749 request_count = blk_plug_queued_count(q);
1751 spin_lock_irq(q->queue_lock);
1753 el_ret = elv_merge(q, &req, bio);
1754 if (el_ret == ELEVATOR_BACK_MERGE) {
1755 if (bio_attempt_back_merge(q, req, bio)) {
1756 elv_bio_merged(q, req, bio);
1757 if (!attempt_back_merge(q, req))
1758 elv_merged_request(q, req, el_ret);
1761 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1762 if (bio_attempt_front_merge(q, req, bio)) {
1763 elv_bio_merged(q, req, bio);
1764 if (!attempt_front_merge(q, req))
1765 elv_merged_request(q, req, el_ret);
1772 * This sync check and mask will be re-done in init_request_from_bio(),
1773 * but we need to set it earlier to expose the sync flag to the
1774 * rq allocator and io schedulers.
1776 rw_flags = bio_data_dir(bio);
1778 rw_flags |= REQ_SYNC;
1781 * Grab a free request. This is might sleep but can not fail.
1782 * Returns with the queue unlocked.
1784 req = get_request(q, rw_flags, bio, GFP_NOIO);
1786 bio->bi_error = PTR_ERR(req);
1792 * After dropping the lock and possibly sleeping here, our request
1793 * may now be mergeable after it had proven unmergeable (above).
1794 * We don't worry about that case for efficiency. It won't happen
1795 * often, and the elevators are able to handle it.
1797 init_request_from_bio(req, bio);
1799 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
1800 req->cpu = raw_smp_processor_id();
1802 plug = current->plug;
1805 * If this is the first request added after a plug, fire
1809 trace_block_plug(q);
1811 if (request_count >= BLK_MAX_REQUEST_COUNT) {
1812 blk_flush_plug_list(plug, false);
1813 trace_block_plug(q);
1816 list_add_tail(&req->queuelist, &plug->list);
1817 blk_account_io_start(req, true);
1819 spin_lock_irq(q->queue_lock);
1820 add_acct_request(q, req, where);
1823 spin_unlock_irq(q->queue_lock);
1826 return BLK_QC_T_NONE;
1830 * If bio->bi_dev is a partition, remap the location
1832 static inline void blk_partition_remap(struct bio *bio)
1834 struct block_device *bdev = bio->bi_bdev;
1836 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1837 struct hd_struct *p = bdev->bd_part;
1839 bio->bi_iter.bi_sector += p->start_sect;
1840 bio->bi_bdev = bdev->bd_contains;
1842 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1844 bio->bi_iter.bi_sector - p->start_sect);
1848 static void handle_bad_sector(struct bio *bio)
1850 char b[BDEVNAME_SIZE];
1852 printk(KERN_INFO "attempt to access beyond end of device\n");
1853 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1854 bdevname(bio->bi_bdev, b),
1856 (unsigned long long)bio_end_sector(bio),
1857 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1860 #ifdef CONFIG_FAIL_MAKE_REQUEST
1862 static DECLARE_FAULT_ATTR(fail_make_request);
1864 static int __init setup_fail_make_request(char *str)
1866 return setup_fault_attr(&fail_make_request, str);
1868 __setup("fail_make_request=", setup_fail_make_request);
1870 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1872 return part->make_it_fail && should_fail(&fail_make_request, bytes);
1875 static int __init fail_make_request_debugfs(void)
1877 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1878 NULL, &fail_make_request);
1880 return PTR_ERR_OR_ZERO(dir);
1883 late_initcall(fail_make_request_debugfs);
1885 #else /* CONFIG_FAIL_MAKE_REQUEST */
1887 static inline bool should_fail_request(struct hd_struct *part,
1893 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1896 * Check whether this bio extends beyond the end of the device.
1898 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1905 /* Test device or partition size, when known. */
1906 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1908 sector_t sector = bio->bi_iter.bi_sector;
1910 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1912 * This may well happen - the kernel calls bread()
1913 * without checking the size of the device, e.g., when
1914 * mounting a device.
1916 handle_bad_sector(bio);
1924 static noinline_for_stack bool
1925 generic_make_request_checks(struct bio *bio)
1927 struct request_queue *q;
1928 int nr_sectors = bio_sectors(bio);
1930 char b[BDEVNAME_SIZE];
1931 struct hd_struct *part;
1935 if (bio_check_eod(bio, nr_sectors))
1938 q = bdev_get_queue(bio->bi_bdev);
1941 "generic_make_request: Trying to access "
1942 "nonexistent block-device %s (%Lu)\n",
1943 bdevname(bio->bi_bdev, b),
1944 (long long) bio->bi_iter.bi_sector);
1948 part = bio->bi_bdev->bd_part;
1949 if (should_fail_request(part, bio->bi_iter.bi_size) ||
1950 should_fail_request(&part_to_disk(part)->part0,
1951 bio->bi_iter.bi_size))
1955 * If this device has partitions, remap block n
1956 * of partition p to block n+start(p) of the disk.
1958 blk_partition_remap(bio);
1960 if (bio_check_eod(bio, nr_sectors))
1964 * Filter flush bio's early so that make_request based
1965 * drivers without flush support don't have to worry
1968 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1969 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1976 if ((bio->bi_rw & REQ_DISCARD) &&
1977 (!blk_queue_discard(q) ||
1978 ((bio->bi_rw & REQ_SECURE) && !blk_queue_secdiscard(q)))) {
1983 if (bio->bi_rw & REQ_WRITE_SAME && !bdev_write_same(bio->bi_bdev)) {
1989 * Various block parts want %current->io_context and lazy ioc
1990 * allocation ends up trading a lot of pain for a small amount of
1991 * memory. Just allocate it upfront. This may fail and block
1992 * layer knows how to live with it.
1994 create_io_context(GFP_ATOMIC, q->node);
1996 if (!blkcg_bio_issue_check(q, bio))
1999 trace_block_bio_queue(q, bio);
2003 bio->bi_error = err;
2009 * generic_make_request - hand a buffer to its device driver for I/O
2010 * @bio: The bio describing the location in memory and on the device.
2012 * generic_make_request() is used to make I/O requests of block
2013 * devices. It is passed a &struct bio, which describes the I/O that needs
2016 * generic_make_request() does not return any status. The
2017 * success/failure status of the request, along with notification of
2018 * completion, is delivered asynchronously through the bio->bi_end_io
2019 * function described (one day) else where.
2021 * The caller of generic_make_request must make sure that bi_io_vec
2022 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2023 * set to describe the device address, and the
2024 * bi_end_io and optionally bi_private are set to describe how
2025 * completion notification should be signaled.
2027 * generic_make_request and the drivers it calls may use bi_next if this
2028 * bio happens to be merged with someone else, and may resubmit the bio to
2029 * a lower device by calling into generic_make_request recursively, which
2030 * means the bio should NOT be touched after the call to ->make_request_fn.
2032 blk_qc_t generic_make_request(struct bio *bio)
2035 * bio_list_on_stack[0] contains bios submitted by the current
2037 * bio_list_on_stack[1] contains bios that were submitted before
2038 * the current make_request_fn, but that haven't been processed
2041 struct bio_list bio_list_on_stack[2];
2042 blk_qc_t ret = BLK_QC_T_NONE;
2044 if (!generic_make_request_checks(bio))
2048 * We only want one ->make_request_fn to be active at a time, else
2049 * stack usage with stacked devices could be a problem. So use
2050 * current->bio_list to keep a list of requests submited by a
2051 * make_request_fn function. current->bio_list is also used as a
2052 * flag to say if generic_make_request is currently active in this
2053 * task or not. If it is NULL, then no make_request is active. If
2054 * it is non-NULL, then a make_request is active, and new requests
2055 * should be added at the tail
2057 if (current->bio_list) {
2058 bio_list_add(¤t->bio_list[0], bio);
2062 /* following loop may be a bit non-obvious, and so deserves some
2064 * Before entering the loop, bio->bi_next is NULL (as all callers
2065 * ensure that) so we have a list with a single bio.
2066 * We pretend that we have just taken it off a longer list, so
2067 * we assign bio_list to a pointer to the bio_list_on_stack,
2068 * thus initialising the bio_list of new bios to be
2069 * added. ->make_request() may indeed add some more bios
2070 * through a recursive call to generic_make_request. If it
2071 * did, we find a non-NULL value in bio_list and re-enter the loop
2072 * from the top. In this case we really did just take the bio
2073 * of the top of the list (no pretending) and so remove it from
2074 * bio_list, and call into ->make_request() again.
2076 BUG_ON(bio->bi_next);
2077 bio_list_init(&bio_list_on_stack[0]);
2078 current->bio_list = bio_list_on_stack;
2080 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
2082 if (likely(blk_queue_enter(q, __GFP_DIRECT_RECLAIM) == 0)) {
2083 struct bio_list lower, same;
2085 /* Create a fresh bio_list for all subordinate requests */
2086 bio_list_on_stack[1] = bio_list_on_stack[0];
2087 bio_list_init(&bio_list_on_stack[0]);
2089 ret = q->make_request_fn(q, bio);
2092 /* sort new bios into those for a lower level
2093 * and those for the same level
2095 bio_list_init(&lower);
2096 bio_list_init(&same);
2097 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
2098 if (q == bdev_get_queue(bio->bi_bdev))
2099 bio_list_add(&same, bio);
2101 bio_list_add(&lower, bio);
2102 /* now assemble so we handle the lowest level first */
2103 bio_list_merge(&bio_list_on_stack[0], &lower);
2104 bio_list_merge(&bio_list_on_stack[0], &same);
2105 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
2109 bio = bio_list_pop(&bio_list_on_stack[0]);
2111 current->bio_list = NULL; /* deactivate */
2116 EXPORT_SYMBOL(generic_make_request);
2118 #ifdef CONFIG_BLK_DEV_IO_TRACE
2119 static inline struct task_struct *get_dirty_task(struct bio *bio)
2122 * Not all the pages in the bio are dirtied by the
2123 * same task but most likely it will be, since the
2124 * sectors accessed on the device must be adjacent.
2126 if (bio->bi_io_vec && bio->bi_io_vec->bv_page &&
2127 bio->bi_io_vec->bv_page->tsk_dirty)
2128 return bio->bi_io_vec->bv_page->tsk_dirty;
2133 static inline struct task_struct *get_dirty_task(struct bio *bio)
2139 #ifdef CONFIG_BLOCK_PERF_FRAMEWORK
2140 #define BLK_PERF_SIZE (1024 * 15)
2141 #define BLK_PERF_HIST_SIZE (sizeof(u32) * BLK_PERF_SIZE)
2143 struct blk_perf_stats {
2147 int buffers_alloced;
2148 ktime_t max_read_time;
2149 ktime_t max_write_time;
2150 ktime_t max_flush_time;
2151 ktime_t min_write_time;
2152 ktime_t min_read_time;
2153 ktime_t min_flush_time;
2154 ktime_t total_write_time;
2155 ktime_t total_read_time;
2156 u64 total_read_size;
2157 u64 total_write_size;
2162 static struct blk_perf_stats blk_perf;
2163 static struct dentry *blk_perf_debug_dir;
2165 static int alloc_histogram_buffers(void)
2169 if (!blk_perf.read_hist)
2170 blk_perf.read_hist = kzalloc(BLK_PERF_HIST_SIZE, GFP_KERNEL);
2172 if (!blk_perf.write_hist)
2173 blk_perf.write_hist = kzalloc(BLK_PERF_HIST_SIZE, GFP_KERNEL);
2175 if (!blk_perf.flush_hist)
2176 blk_perf.flush_hist = kzalloc(BLK_PERF_HIST_SIZE, GFP_KERNEL);
2178 if (!blk_perf.read_hist || !blk_perf.write_hist || !blk_perf.flush_hist)
2182 blk_perf.buffers_alloced = 1;
2186 static void clear_histogram_buffers(void)
2188 if (!blk_perf.buffers_alloced)
2190 memset(blk_perf.read_hist, 0, BLK_PERF_HIST_SIZE);
2191 memset(blk_perf.write_hist, 0, BLK_PERF_HIST_SIZE);
2192 memset(blk_perf.flush_hist, 0, BLK_PERF_HIST_SIZE);
2195 static int enable_perf(void *data, u64 val)
2199 if (!blk_perf.buffers_alloced)
2200 ret = alloc_histogram_buffers();
2205 spin_lock(&blk_perf.lock);
2206 blk_perf.is_enabled = val;
2207 spin_unlock(&blk_perf.lock);
2211 static int is_perf_enabled(void *data, u64 *val)
2213 spin_lock(&blk_perf.lock);
2214 *val = blk_perf.is_enabled;
2215 spin_unlock(&blk_perf.lock);
2219 DEFINE_SIMPLE_ATTRIBUTE(enable_perf_fops, is_perf_enabled, enable_perf,
2222 static char *blk_debug_buffer;
2223 static u32 blk_debug_data_size;
2224 static DEFINE_MUTEX(blk_perf_debug_buffer_mutex);
2226 static ssize_t blk_perf_read(struct file *file, char __user *buf,
2227 size_t count, loff_t *file_pos)
2231 mutex_lock(&blk_perf_debug_buffer_mutex);
2232 ret = simple_read_from_buffer(buf, count, file_pos, blk_debug_buffer,
2233 blk_debug_data_size);
2234 mutex_unlock(&blk_perf_debug_buffer_mutex);
2239 static int blk_debug_buffer_alloc(u32 buffer_size)
2243 mutex_lock(&blk_perf_debug_buffer_mutex);
2244 if (blk_debug_buffer != NULL) {
2245 pr_err("blk_debug_buffer is in use\n");
2249 blk_debug_buffer = kzalloc(buffer_size, GFP_KERNEL);
2250 if (!blk_debug_buffer)
2253 mutex_unlock(&blk_perf_debug_buffer_mutex);
2257 static int blk_perf_close(struct inode *inode, struct file *file)
2259 mutex_lock(&blk_perf_debug_buffer_mutex);
2260 blk_debug_data_size = 0;
2261 kfree(blk_debug_buffer);
2262 blk_debug_buffer = NULL;
2263 mutex_unlock(&blk_perf_debug_buffer_mutex);
2267 static u32 fill_basic_perf_info(char *buffer, u32 buffer_size)
2271 size += scnprintf(buffer + size, buffer_size - size, "\n");
2273 spin_lock(&blk_perf.lock);
2274 size += scnprintf(buffer + size, buffer_size - size,
2275 "max_read_time_ms: %llu\n",
2276 ktime_to_ms(blk_perf.max_read_time));
2278 size += scnprintf(buffer + size, buffer_size - size,
2279 "min_read_time_ms: %llu\n",
2280 ktime_to_ms(blk_perf.min_read_time));
2282 size += scnprintf(buffer + size, buffer_size - size,
2283 "total_read_time_ms: %llu\n",
2284 ktime_to_ms(blk_perf.total_read_time));
2286 size += scnprintf(buffer + size, buffer_size - size,
2287 "total_read_size: %llu\n\n",
2288 blk_perf.total_read_size);
2290 size += scnprintf(buffer + size, buffer_size - size,
2291 "max_write_time_ms: %llu\n",
2292 ktime_to_ms(blk_perf.max_write_time));
2294 size += scnprintf(buffer + size, buffer_size - size,
2295 "min_write_time_ms: %llu\n",
2296 ktime_to_ms(blk_perf.min_write_time));
2298 size += scnprintf(buffer + size, buffer_size - size,
2299 "total_write_time_ms: %llu\n",
2300 ktime_to_ms(blk_perf.total_write_time));
2302 size += scnprintf(buffer + size, buffer_size - size,
2303 "total_write_size: %llu\n\n",
2304 blk_perf.total_write_size);
2306 size += scnprintf(buffer + size, buffer_size - size,
2307 "max_flush_time_ms: %llu\n",
2308 ktime_to_ms(blk_perf.max_flush_time));
2310 size += scnprintf(buffer + size, buffer_size - size,
2311 "min_flush_time_ms: %llu\n\n",
2312 ktime_to_ms(blk_perf.min_flush_time));
2314 spin_unlock(&blk_perf.lock);
2319 static int basic_perf_open(struct inode *inode, struct file *file)
2324 buffer_size = BLK_PERF_HIST_SIZE;
2325 ret = blk_debug_buffer_alloc(buffer_size);
2329 mutex_lock(&blk_perf_debug_buffer_mutex);
2330 blk_debug_data_size = fill_basic_perf_info(blk_debug_buffer,
2332 mutex_unlock(&blk_perf_debug_buffer_mutex);
2337 static const struct file_operations basic_perf_ops = {
2338 .read = blk_perf_read,
2339 .release = blk_perf_close,
2340 .open = basic_perf_open,
2343 static int hist_open_helper(void *hist_buf)
2347 if (!blk_perf.buffers_alloced)
2350 ret = blk_debug_buffer_alloc(BLK_PERF_HIST_SIZE);
2354 spin_lock(&blk_perf.lock);
2355 memcpy(blk_debug_buffer, hist_buf, BLK_PERF_HIST_SIZE);
2356 spin_unlock(&blk_perf.lock);
2358 mutex_lock(&blk_perf_debug_buffer_mutex);
2359 blk_debug_data_size = BLK_PERF_HIST_SIZE;
2360 mutex_unlock(&blk_perf_debug_buffer_mutex);
2364 static int write_hist_open(struct inode *inode, struct file *file)
2366 return hist_open_helper(blk_perf.write_hist);
2369 static const struct file_operations write_hist_ops = {
2370 .read = blk_perf_read,
2371 .release = blk_perf_close,
2372 .open = write_hist_open,
2376 static int read_hist_open(struct inode *inode, struct file *file)
2378 return hist_open_helper(blk_perf.read_hist);
2381 static const struct file_operations read_hist_ops = {
2382 .read = blk_perf_read,
2383 .release = blk_perf_close,
2384 .open = read_hist_open,
2387 static int flush_hist_open(struct inode *inode, struct file *file)
2389 return hist_open_helper(blk_perf.flush_hist);
2392 static const struct file_operations flush_hist_ops = {
2393 .read = blk_perf_read,
2394 .release = blk_perf_close,
2395 .open = flush_hist_open,
2398 static void clear_perf_stats_helper(void)
2400 spin_lock(&blk_perf.lock);
2401 blk_perf.max_write_time = ktime_set(0, 0);
2402 blk_perf.max_read_time = ktime_set(0, 0);
2403 blk_perf.max_flush_time = ktime_set(0, 0);
2404 blk_perf.min_write_time = ktime_set(KTIME_MAX, 0);
2405 blk_perf.min_read_time = ktime_set(KTIME_MAX, 0);
2406 blk_perf.min_flush_time = ktime_set(KTIME_MAX, 0);
2407 blk_perf.total_write_time = ktime_set(0, 0);
2408 blk_perf.total_read_time = ktime_set(0, 0);
2409 blk_perf.total_read_size = 0;
2410 blk_perf.total_write_size = 0;
2411 blk_perf.is_enabled = 0;
2412 clear_histogram_buffers();
2413 spin_unlock(&blk_perf.lock);
2416 static int clear_perf_stats(void *data, u64 val)
2418 clear_perf_stats_helper();
2422 DEFINE_SIMPLE_ATTRIBUTE(clear_perf_stats_fops, NULL, clear_perf_stats,
2425 static void blk_debugfs_init(void)
2427 struct dentry *f_ent;
2429 blk_perf_debug_dir = debugfs_create_dir("block_perf", NULL);
2430 if (IS_ERR(blk_perf_debug_dir)) {
2431 pr_err("Failed to create block_perf debug_fs directory\n");
2435 f_ent = debugfs_create_file("basic_perf", 0400, blk_perf_debug_dir,
2436 NULL, &basic_perf_ops);
2437 if (IS_ERR(f_ent)) {
2438 pr_err("Failed to create debug_fs basic_perf file\n");
2442 f_ent = debugfs_create_file("write_hist", 0400, blk_perf_debug_dir,
2443 NULL, &write_hist_ops);
2444 if (IS_ERR(f_ent)) {
2445 pr_err("Failed to create debug_fs write_hist file\n");
2449 f_ent = debugfs_create_file("read_hist", 0400, blk_perf_debug_dir,
2450 NULL, &read_hist_ops);
2451 if (IS_ERR(f_ent)) {
2452 pr_err("Failed to create debug_fs read_hist file\n");
2456 f_ent = debugfs_create_file("flush_hist", 0400, blk_perf_debug_dir,
2457 NULL, &flush_hist_ops);
2458 if (IS_ERR(f_ent)) {
2459 pr_err("Failed to create debug_fs flush_hist file\n");
2463 f_ent = debugfs_create_file("enable_perf", 0600, blk_perf_debug_dir,
2464 NULL, &enable_perf_fops);
2465 if (IS_ERR(f_ent)) {
2466 pr_err("Failed to create debug_fs enable_perf file\n");
2470 f_ent = debugfs_create_file("clear_perf_stats", 0200,
2471 blk_perf_debug_dir, NULL,
2472 &clear_perf_stats_fops);
2473 if (IS_ERR(f_ent)) {
2474 pr_err("Failed to create debug_fs clear_perf_stats file\n");
2479 static void blk_init_perf(void)
2482 spin_lock_init(&blk_perf.lock);
2484 clear_perf_stats_helper();
2488 static void set_submit_info(struct bio *bio, unsigned int count)
2490 ktime_t submit_time;
2492 if (unlikely(blk_perf.is_enabled)) {
2493 submit_time = ktime_get();
2494 bio->submit_time.tv64 = submit_time.tv64;
2495 bio->blk_sector_count = count;
2499 bio->submit_time.tv64 = 0;
2500 bio->blk_sector_count = 0;
2503 void blk_update_perf_read_write_stats(ktime_t bio_process_time, int is_write,
2506 u32 bio_process_time_ms;
2508 bio_process_time_ms = ktime_to_ms(bio_process_time);
2509 if (bio_process_time_ms >= BLK_PERF_SIZE)
2510 bio_process_time_ms = BLK_PERF_SIZE - 1;
2513 if (ktime_after(bio_process_time, blk_perf.max_write_time))
2514 blk_perf.max_write_time = bio_process_time;
2516 if (ktime_before(bio_process_time, blk_perf.min_write_time))
2517 blk_perf.min_write_time = bio_process_time;
2518 blk_perf.total_write_time =
2519 ktime_add(blk_perf.total_write_time, bio_process_time);
2520 blk_perf.total_write_size += count;
2521 blk_perf.write_hist[bio_process_time_ms] += count;
2524 if (ktime_after(bio_process_time, blk_perf.max_read_time))
2525 blk_perf.max_read_time = bio_process_time;
2527 if (ktime_before(bio_process_time, blk_perf.min_read_time))
2528 blk_perf.min_read_time = bio_process_time;
2529 blk_perf.total_read_time =
2530 ktime_add(blk_perf.total_read_time, bio_process_time);
2531 blk_perf.total_read_size += count;
2532 blk_perf.read_hist[bio_process_time_ms] += count;
2535 void blk_update_perf_stats(struct bio *bio)
2537 ktime_t bio_process_time;
2538 u32 bio_process_time_ms;
2541 spin_lock(&blk_perf.lock);
2542 if (likely(!blk_perf.is_enabled))
2544 if (!bio->submit_time.tv64)
2546 bio_process_time = ktime_sub(ktime_get(), bio->submit_time);
2548 count = bio->blk_sector_count;
2553 if (bio->bi_rw & WRITE ||
2554 unlikely(bio->bi_rw & REQ_WRITE_SAME))
2557 blk_update_perf_read_write_stats(bio_process_time, is_write,
2561 bio_process_time_ms = ktime_to_ms(bio_process_time);
2562 if (bio_process_time_ms >= BLK_PERF_SIZE)
2563 bio_process_time_ms = BLK_PERF_SIZE - 1;
2565 if (ktime_after(bio_process_time, blk_perf.max_flush_time))
2566 blk_perf.max_flush_time = bio_process_time;
2568 if (ktime_before(bio_process_time, blk_perf.min_flush_time))
2569 blk_perf.min_flush_time = bio_process_time;
2571 blk_perf.flush_hist[bio_process_time_ms] += 1;
2574 spin_unlock(&blk_perf.lock);
2578 static inline void set_submit_info(struct bio *bio, unsigned int count)
2584 static inline void blk_init_perf(void)
2587 #endif /* #ifdef CONFIG_BLOCK_PERF_FRAMEWORK */
2590 * submit_bio - submit a bio to the block device layer for I/O
2591 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
2592 * @bio: The &struct bio which describes the I/O
2594 * submit_bio() is very similar in purpose to generic_make_request(), and
2595 * uses that function to do most of the work. Both are fairly rough
2596 * interfaces; @bio must be presetup and ready for I/O.
2599 blk_qc_t submit_bio(int rw, struct bio *bio)
2601 unsigned int count = 0;
2605 * If it's a regular read/write or a barrier with data attached,
2606 * go through the normal accounting stuff before submission.
2608 if (bio_has_data(bio)) {
2609 if (unlikely(rw & REQ_WRITE_SAME))
2610 count = bdev_logical_block_size(bio->bi_bdev) >> 9;
2612 count = bio_sectors(bio);
2615 count_vm_events(PGPGOUT, count);
2617 task_io_account_read(bio->bi_iter.bi_size);
2618 count_vm_events(PGPGIN, count);
2621 if (unlikely(block_dump)) {
2622 char b[BDEVNAME_SIZE];
2623 struct task_struct *tsk;
2625 tsk = get_dirty_task(bio);
2626 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
2627 tsk->comm, task_pid_nr(tsk),
2628 (rw & WRITE) ? "WRITE" : "READ",
2629 (unsigned long long)bio->bi_iter.bi_sector,
2630 bdevname(bio->bi_bdev, b),
2635 set_submit_info(bio, count);
2636 return generic_make_request(bio);
2638 EXPORT_SYMBOL(submit_bio);
2641 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2642 * for new the queue limits
2644 * @rq: the request being checked
2647 * @rq may have been made based on weaker limitations of upper-level queues
2648 * in request stacking drivers, and it may violate the limitation of @q.
2649 * Since the block layer and the underlying device driver trust @rq
2650 * after it is inserted to @q, it should be checked against @q before
2651 * the insertion using this generic function.
2653 * Request stacking drivers like request-based dm may change the queue
2654 * limits when retrying requests on other queues. Those requests need
2655 * to be checked against the new queue limits again during dispatch.
2657 static int blk_cloned_rq_check_limits(struct request_queue *q,
2660 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, rq->cmd_flags)) {
2661 printk(KERN_ERR "%s: over max size limit.\n", __func__);
2666 * queue's settings related to segment counting like q->bounce_pfn
2667 * may differ from that of other stacking queues.
2668 * Recalculate it to check the request correctly on this queue's
2671 blk_recalc_rq_segments(rq);
2672 if (rq->nr_phys_segments > queue_max_segments(q)) {
2673 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2681 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2682 * @q: the queue to submit the request
2683 * @rq: the request being queued
2685 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
2687 unsigned long flags;
2688 int where = ELEVATOR_INSERT_BACK;
2690 if (blk_cloned_rq_check_limits(q, rq))
2694 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
2698 if (blk_queue_io_stat(q))
2699 blk_account_io_start(rq, true);
2700 blk_mq_insert_request(rq, false, true, false);
2704 spin_lock_irqsave(q->queue_lock, flags);
2705 if (unlikely(blk_queue_dying(q))) {
2706 spin_unlock_irqrestore(q->queue_lock, flags);
2711 * Submitting request must be dequeued before calling this function
2712 * because it will be linked to another request_queue
2714 BUG_ON(blk_queued_rq(rq));
2716 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
2717 where = ELEVATOR_INSERT_FLUSH;
2719 add_acct_request(q, rq, where);
2720 if (where == ELEVATOR_INSERT_FLUSH)
2722 spin_unlock_irqrestore(q->queue_lock, flags);
2726 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2729 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2730 * @rq: request to examine
2733 * A request could be merge of IOs which require different failure
2734 * handling. This function determines the number of bytes which
2735 * can be failed from the beginning of the request without
2736 * crossing into area which need to be retried further.
2739 * The number of bytes to fail.
2742 * queue_lock must be held.
2744 unsigned int blk_rq_err_bytes(const struct request *rq)
2746 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2747 unsigned int bytes = 0;
2750 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
2751 return blk_rq_bytes(rq);
2754 * Currently the only 'mixing' which can happen is between
2755 * different fastfail types. We can safely fail portions
2756 * which have all the failfast bits that the first one has -
2757 * the ones which are at least as eager to fail as the first
2760 for (bio = rq->bio; bio; bio = bio->bi_next) {
2761 if ((bio->bi_rw & ff) != ff)
2763 bytes += bio->bi_iter.bi_size;
2766 /* this could lead to infinite loop */
2767 BUG_ON(blk_rq_bytes(rq) && !bytes);
2770 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2772 void blk_account_io_completion(struct request *req, unsigned int bytes)
2774 if (blk_do_io_stat(req)) {
2775 const int rw = rq_data_dir(req);
2776 struct hd_struct *part;
2779 cpu = part_stat_lock();
2781 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2786 void blk_account_io_done(struct request *req)
2789 * Account IO completion. flush_rq isn't accounted as a
2790 * normal IO on queueing nor completion. Accounting the
2791 * containing request is enough.
2793 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
2794 unsigned long duration = jiffies - req->start_time;
2795 const int rw = rq_data_dir(req);
2796 struct hd_struct *part;
2799 cpu = part_stat_lock();
2802 part_stat_inc(cpu, part, ios[rw]);
2803 part_stat_add(cpu, part, ticks[rw], duration);
2804 part_round_stats(cpu, part);
2805 part_dec_in_flight(part, rw);
2807 hd_struct_put(part);
2814 * Don't process normal requests when queue is suspended
2815 * or in the process of suspending/resuming
2817 static struct request *blk_pm_peek_request(struct request_queue *q,
2820 if (q->dev && (q->rpm_status == RPM_SUSPENDED ||
2821 (q->rpm_status != RPM_ACTIVE && !(rq->cmd_flags & REQ_PM))))
2827 static inline struct request *blk_pm_peek_request(struct request_queue *q,
2834 void blk_account_io_start(struct request *rq, bool new_io)
2836 struct hd_struct *part;
2837 int rw = rq_data_dir(rq);
2840 if (!blk_do_io_stat(rq))
2843 cpu = part_stat_lock();
2847 part_stat_inc(cpu, part, merges[rw]);
2849 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2850 if (!hd_struct_try_get(part)) {
2852 * The partition is already being removed,
2853 * the request will be accounted on the disk only
2855 * We take a reference on disk->part0 although that
2856 * partition will never be deleted, so we can treat
2857 * it as any other partition.
2859 part = &rq->rq_disk->part0;
2860 hd_struct_get(part);
2862 part_round_stats(cpu, part);
2863 part_inc_in_flight(part, rw);
2871 * blk_peek_request - peek at the top of a request queue
2872 * @q: request queue to peek at
2875 * Return the request at the top of @q. The returned request
2876 * should be started using blk_start_request() before LLD starts
2880 * Pointer to the request at the top of @q if available. Null
2884 * queue_lock must be held.
2886 struct request *blk_peek_request(struct request_queue *q)
2891 while ((rq = __elv_next_request(q)) != NULL) {
2893 rq = blk_pm_peek_request(q, rq);
2897 if (!(rq->cmd_flags & REQ_STARTED)) {
2899 * This is the first time the device driver
2900 * sees this request (possibly after
2901 * requeueing). Notify IO scheduler.
2903 if (rq->cmd_flags & REQ_SORTED)
2904 elv_activate_rq(q, rq);
2907 * just mark as started even if we don't start
2908 * it, a request that has been delayed should
2909 * not be passed by new incoming requests
2911 rq->cmd_flags |= REQ_STARTED;
2912 trace_block_rq_issue(q, rq);
2915 if (!q->boundary_rq || q->boundary_rq == rq) {
2916 q->end_sector = rq_end_sector(rq);
2917 q->boundary_rq = NULL;
2920 if (rq->cmd_flags & REQ_DONTPREP)
2923 if (q->dma_drain_size && blk_rq_bytes(rq)) {
2925 * make sure space for the drain appears we
2926 * know we can do this because max_hw_segments
2927 * has been adjusted to be one fewer than the
2930 rq->nr_phys_segments++;
2936 ret = q->prep_rq_fn(q, rq);
2937 if (ret == BLKPREP_OK) {
2939 } else if (ret == BLKPREP_DEFER) {
2941 * the request may have been (partially) prepped.
2942 * we need to keep this request in the front to
2943 * avoid resource deadlock. REQ_STARTED will
2944 * prevent other fs requests from passing this one.
2946 if (q->dma_drain_size && blk_rq_bytes(rq) &&
2947 !(rq->cmd_flags & REQ_DONTPREP)) {
2949 * remove the space for the drain we added
2950 * so that we don't add it again
2952 --rq->nr_phys_segments;
2957 } else if (ret == BLKPREP_KILL) {
2958 rq->cmd_flags |= REQ_QUIET;
2960 * Mark this request as started so we don't trigger
2961 * any debug logic in the end I/O path.
2963 blk_start_request(rq);
2964 __blk_end_request_all(rq, -EIO);
2966 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2973 EXPORT_SYMBOL(blk_peek_request);
2975 void blk_dequeue_request(struct request *rq)
2977 struct request_queue *q = rq->q;
2979 BUG_ON(list_empty(&rq->queuelist));
2980 BUG_ON(ELV_ON_HASH(rq));
2982 list_del_init(&rq->queuelist);
2985 * the time frame between a request being removed from the lists
2986 * and to it is freed is accounted as io that is in progress at
2989 if (blk_account_rq(rq)) {
2990 q->in_flight[rq_is_sync(rq)]++;
2991 set_io_start_time_ns(rq);
2996 * blk_start_request - start request processing on the driver
2997 * @req: request to dequeue
3000 * Dequeue @req and start timeout timer on it. This hands off the
3001 * request to the driver.
3003 * Block internal functions which don't want to start timer should
3004 * call blk_dequeue_request().
3007 * queue_lock must be held.
3009 void blk_start_request(struct request *req)
3011 blk_dequeue_request(req);
3014 * We are now handing the request to the hardware, initialize
3015 * resid_len to full count and add the timeout handler.
3017 req->resid_len = blk_rq_bytes(req);
3018 if (unlikely(blk_bidi_rq(req)))
3019 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
3021 BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags));
3024 EXPORT_SYMBOL(blk_start_request);
3027 * blk_fetch_request - fetch a request from a request queue
3028 * @q: request queue to fetch a request from
3031 * Return the request at the top of @q. The request is started on
3032 * return and LLD can start processing it immediately.
3035 * Pointer to the request at the top of @q if available. Null
3039 * queue_lock must be held.
3041 struct request *blk_fetch_request(struct request_queue *q)
3045 rq = blk_peek_request(q);
3047 blk_start_request(rq);
3050 EXPORT_SYMBOL(blk_fetch_request);
3053 * blk_update_request - Special helper function for request stacking drivers
3054 * @req: the request being processed
3055 * @error: %0 for success, < %0 for error
3056 * @nr_bytes: number of bytes to complete @req
3059 * Ends I/O on a number of bytes attached to @req, but doesn't complete
3060 * the request structure even if @req doesn't have leftover.
3061 * If @req has leftover, sets it up for the next range of segments.
3063 * This special helper function is only for request stacking drivers
3064 * (e.g. request-based dm) so that they can handle partial completion.
3065 * Actual device drivers should use blk_end_request instead.
3067 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
3068 * %false return from this function.
3071 * %false - this request doesn't have any more data
3072 * %true - this request has more data
3074 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
3078 trace_block_rq_complete(req->q, req, nr_bytes);
3084 * For fs requests, rq is just carrier of independent bio's
3085 * and each partial completion should be handled separately.
3086 * Reset per-request error on each partial completion.
3088 * TODO: tj: This is too subtle. It would be better to let
3089 * low level drivers do what they see fit.
3091 if (req->cmd_type == REQ_TYPE_FS)
3094 if (error && req->cmd_type == REQ_TYPE_FS &&
3095 !(req->cmd_flags & REQ_QUIET)) {
3100 error_type = "recoverable transport";
3103 error_type = "critical target";
3106 error_type = "critical nexus";
3109 error_type = "timeout";
3112 error_type = "critical space allocation";
3115 error_type = "critical medium";
3122 printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n",
3123 __func__, error_type, req->rq_disk ?
3124 req->rq_disk->disk_name : "?",
3125 (unsigned long long)blk_rq_pos(req));
3129 blk_account_io_completion(req, nr_bytes);
3134 * Check for this if flagged, Req based dm needs to perform
3135 * post processing, hence dont end bios or request.DM
3138 if (bio_flagged(req->bio, BIO_DONTFREE))
3142 struct bio *bio = req->bio;
3143 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
3145 if (bio_bytes == bio->bi_iter.bi_size)
3146 req->bio = bio->bi_next;
3148 req_bio_endio(req, bio, bio_bytes, error);
3150 total_bytes += bio_bytes;
3151 nr_bytes -= bio_bytes;
3162 * Reset counters so that the request stacking driver
3163 * can find how many bytes remain in the request
3166 req->__data_len = 0;
3170 req->__data_len -= total_bytes;
3172 /* update sector only for requests with clear definition of sector */
3173 if (req->cmd_type == REQ_TYPE_FS)
3174 req->__sector += total_bytes >> 9;
3176 /* mixed attributes always follow the first bio */
3177 if (req->cmd_flags & REQ_MIXED_MERGE) {
3178 req->cmd_flags &= ~REQ_FAILFAST_MASK;
3179 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
3183 * If total number of sectors is less than the first segment
3184 * size, something has gone terribly wrong.
3186 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
3187 blk_dump_rq_flags(req, "request botched");
3188 req->__data_len = blk_rq_cur_bytes(req);
3191 /* recalculate the number of segments */
3192 blk_recalc_rq_segments(req);
3196 EXPORT_SYMBOL_GPL(blk_update_request);
3198 static bool blk_update_bidi_request(struct request *rq, int error,
3199 unsigned int nr_bytes,
3200 unsigned int bidi_bytes)
3202 if (blk_update_request(rq, error, nr_bytes))
3205 /* Bidi request must be completed as a whole */
3206 if (unlikely(blk_bidi_rq(rq)) &&
3207 blk_update_request(rq->next_rq, error, bidi_bytes))
3210 if (blk_queue_add_random(rq->q))
3211 add_disk_randomness(rq->rq_disk);
3217 * blk_unprep_request - unprepare a request
3220 * This function makes a request ready for complete resubmission (or
3221 * completion). It happens only after all error handling is complete,
3222 * so represents the appropriate moment to deallocate any resources
3223 * that were allocated to the request in the prep_rq_fn. The queue
3224 * lock is held when calling this.
3226 void blk_unprep_request(struct request *req)
3228 struct request_queue *q = req->q;
3230 req->cmd_flags &= ~REQ_DONTPREP;
3231 if (q->unprep_rq_fn)
3232 q->unprep_rq_fn(q, req);
3234 EXPORT_SYMBOL_GPL(blk_unprep_request);
3237 * queue lock must be held
3239 void blk_finish_request(struct request *req, int error)
3241 if (req->cmd_flags & REQ_QUEUED)
3242 blk_queue_end_tag(req->q, req);
3244 BUG_ON(blk_queued_rq(req));
3246 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
3247 laptop_io_completion(req->q->backing_dev_info);
3249 blk_delete_timer(req);
3251 if (req->cmd_flags & REQ_DONTPREP)
3252 blk_unprep_request(req);
3254 blk_account_io_done(req);
3257 req->end_io(req, error);
3259 if (blk_bidi_rq(req))
3260 __blk_put_request(req->next_rq->q, req->next_rq);
3262 __blk_put_request(req->q, req);
3265 EXPORT_SYMBOL(blk_finish_request);
3268 * blk_end_bidi_request - Complete a bidi request
3269 * @rq: the request to complete
3270 * @error: %0 for success, < %0 for error
3271 * @nr_bytes: number of bytes to complete @rq
3272 * @bidi_bytes: number of bytes to complete @rq->next_rq
3275 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
3276 * Drivers that supports bidi can safely call this member for any
3277 * type of request, bidi or uni. In the later case @bidi_bytes is
3281 * %false - we are done with this request
3282 * %true - still buffers pending for this request
3284 static bool blk_end_bidi_request(struct request *rq, int error,
3285 unsigned int nr_bytes, unsigned int bidi_bytes)
3287 struct request_queue *q = rq->q;
3288 unsigned long flags;
3290 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
3293 spin_lock_irqsave(q->queue_lock, flags);
3294 blk_finish_request(rq, error);
3295 spin_unlock_irqrestore(q->queue_lock, flags);
3301 * __blk_end_bidi_request - Complete a bidi request with queue lock held
3302 * @rq: the request to complete
3303 * @error: %0 for success, < %0 for error
3304 * @nr_bytes: number of bytes to complete @rq
3305 * @bidi_bytes: number of bytes to complete @rq->next_rq
3308 * Identical to blk_end_bidi_request() except that queue lock is
3309 * assumed to be locked on entry and remains so on return.
3312 * %false - we are done with this request
3313 * %true - still buffers pending for this request
3315 bool __blk_end_bidi_request(struct request *rq, int error,
3316 unsigned int nr_bytes, unsigned int bidi_bytes)
3318 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
3321 blk_finish_request(rq, error);
3327 * blk_end_request - Helper function for drivers to complete the request.
3328 * @rq: the request being processed
3329 * @error: %0 for success, < %0 for error
3330 * @nr_bytes: number of bytes to complete
3333 * Ends I/O on a number of bytes attached to @rq.
3334 * If @rq has leftover, sets it up for the next range of segments.
3337 * %false - we are done with this request
3338 * %true - still buffers pending for this request
3340 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
3342 return blk_end_bidi_request(rq, error, nr_bytes, 0);
3344 EXPORT_SYMBOL(blk_end_request);
3347 * blk_end_request_all - Helper function for drives to finish the request.
3348 * @rq: the request to finish
3349 * @error: %0 for success, < %0 for error
3352 * Completely finish @rq.
3354 void blk_end_request_all(struct request *rq, int error)
3357 unsigned int bidi_bytes = 0;
3359 if (unlikely(blk_bidi_rq(rq)))
3360 bidi_bytes = blk_rq_bytes(rq->next_rq);
3362 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
3365 EXPORT_SYMBOL(blk_end_request_all);
3368 * blk_end_request_cur - Helper function to finish the current request chunk.
3369 * @rq: the request to finish the current chunk for
3370 * @error: %0 for success, < %0 for error
3373 * Complete the current consecutively mapped chunk from @rq.
3376 * %false - we are done with this request
3377 * %true - still buffers pending for this request
3379 bool blk_end_request_cur(struct request *rq, int error)
3381 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
3383 EXPORT_SYMBOL(blk_end_request_cur);
3386 * blk_end_request_err - Finish a request till the next failure boundary.
3387 * @rq: the request to finish till the next failure boundary for
3388 * @error: must be negative errno
3391 * Complete @rq till the next failure boundary.
3394 * %false - we are done with this request
3395 * %true - still buffers pending for this request
3397 bool blk_end_request_err(struct request *rq, int error)
3399 WARN_ON(error >= 0);
3400 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
3402 EXPORT_SYMBOL_GPL(blk_end_request_err);
3405 * __blk_end_request - Helper function for drivers to complete the request.
3406 * @rq: the request being processed
3407 * @error: %0 for success, < %0 for error
3408 * @nr_bytes: number of bytes to complete
3411 * Must be called with queue lock held unlike blk_end_request().
3414 * %false - we are done with this request
3415 * %true - still buffers pending for this request
3417 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
3419 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
3421 EXPORT_SYMBOL(__blk_end_request);
3424 * __blk_end_request_all - Helper function for drives to finish the request.
3425 * @rq: the request to finish
3426 * @error: %0 for success, < %0 for error
3429 * Completely finish @rq. Must be called with queue lock held.
3431 void __blk_end_request_all(struct request *rq, int error)
3434 unsigned int bidi_bytes = 0;
3436 if (unlikely(blk_bidi_rq(rq)))
3437 bidi_bytes = blk_rq_bytes(rq->next_rq);
3439 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
3442 EXPORT_SYMBOL(__blk_end_request_all);
3445 * __blk_end_request_cur - Helper function to finish the current request chunk.
3446 * @rq: the request to finish the current chunk for
3447 * @error: %0 for success, < %0 for error
3450 * Complete the current consecutively mapped chunk from @rq. Must
3451 * be called with queue lock held.
3454 * %false - we are done with this request
3455 * %true - still buffers pending for this request
3457 bool __blk_end_request_cur(struct request *rq, int error)
3459 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
3461 EXPORT_SYMBOL(__blk_end_request_cur);
3464 * __blk_end_request_err - Finish a request till the next failure boundary.
3465 * @rq: the request to finish till the next failure boundary for
3466 * @error: must be negative errno
3469 * Complete @rq till the next failure boundary. Must be called
3470 * with queue lock held.
3473 * %false - we are done with this request
3474 * %true - still buffers pending for this request
3476 bool __blk_end_request_err(struct request *rq, int error)
3478 WARN_ON(error >= 0);
3479 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
3481 EXPORT_SYMBOL_GPL(__blk_end_request_err);
3483 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
3486 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
3487 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
3489 if (bio_has_data(bio))
3490 rq->nr_phys_segments = bio_phys_segments(q, bio);
3492 rq->__data_len = bio->bi_iter.bi_size;
3493 rq->bio = rq->biotail = bio;
3496 rq->rq_disk = bio->bi_bdev->bd_disk;
3499 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
3501 * rq_flush_dcache_pages - Helper function to flush all pages in a request
3502 * @rq: the request to be flushed
3505 * Flush all pages in @rq.
3507 void rq_flush_dcache_pages(struct request *rq)
3509 struct req_iterator iter;
3510 struct bio_vec bvec;
3512 rq_for_each_segment(bvec, rq, iter)
3513 flush_dcache_page(bvec.bv_page);
3515 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
3519 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
3520 * @q : the queue of the device being checked
3523 * Check if underlying low-level drivers of a device are busy.
3524 * If the drivers want to export their busy state, they must set own
3525 * exporting function using blk_queue_lld_busy() first.
3527 * Basically, this function is used only by request stacking drivers
3528 * to stop dispatching requests to underlying devices when underlying
3529 * devices are busy. This behavior helps more I/O merging on the queue
3530 * of the request stacking driver and prevents I/O throughput regression
3531 * on burst I/O load.
3534 * 0 - Not busy (The request stacking driver should dispatch request)
3535 * 1 - Busy (The request stacking driver should stop dispatching request)
3537 int blk_lld_busy(struct request_queue *q)
3540 return q->lld_busy_fn(q);
3544 EXPORT_SYMBOL_GPL(blk_lld_busy);
3547 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
3548 * @rq: the clone request to be cleaned up
3551 * Free all bios in @rq for a cloned request.
3553 void blk_rq_unprep_clone(struct request *rq)
3557 while ((bio = rq->bio) != NULL) {
3558 rq->bio = bio->bi_next;
3563 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
3566 * Copy attributes of the original request to the clone request.
3567 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3569 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
3571 dst->cpu = src->cpu;
3572 dst->cmd_flags |= (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
3573 dst->cmd_type = src->cmd_type;
3574 dst->__sector = blk_rq_pos(src);
3575 dst->__data_len = blk_rq_bytes(src);
3576 dst->nr_phys_segments = src->nr_phys_segments;
3577 dst->ioprio = src->ioprio;
3578 dst->extra_len = src->extra_len;
3582 * blk_rq_prep_clone - Helper function to setup clone request
3583 * @rq: the request to be setup
3584 * @rq_src: original request to be cloned
3585 * @bs: bio_set that bios for clone are allocated from
3586 * @gfp_mask: memory allocation mask for bio
3587 * @bio_ctr: setup function to be called for each clone bio.
3588 * Returns %0 for success, non %0 for failure.
3589 * @data: private data to be passed to @bio_ctr
3592 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3593 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3594 * are not copied, and copying such parts is the caller's responsibility.
3595 * Also, pages which the original bios are pointing to are not copied
3596 * and the cloned bios just point same pages.
3597 * So cloned bios must be completed before original bios, which means
3598 * the caller must complete @rq before @rq_src.
3600 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
3601 struct bio_set *bs, gfp_t gfp_mask,
3602 int (*bio_ctr)(struct bio *, struct bio *, void *),
3605 struct bio *bio, *bio_src;
3610 __rq_for_each_bio(bio_src, rq_src) {
3611 bio = bio_clone_fast(bio_src, gfp_mask, bs);
3615 if (bio_ctr && bio_ctr(bio, bio_src, data))
3619 rq->biotail->bi_next = bio;
3622 rq->bio = rq->biotail = bio;
3625 __blk_rq_prep_clone(rq, rq_src);
3632 blk_rq_unprep_clone(rq);
3636 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
3638 int kblockd_schedule_work(struct work_struct *work)
3640 return queue_work(kblockd_workqueue, work);
3642 EXPORT_SYMBOL(kblockd_schedule_work);
3644 int kblockd_schedule_delayed_work(struct delayed_work *dwork,
3645 unsigned long delay)
3647 return queue_delayed_work(kblockd_workqueue, dwork, delay);
3649 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
3651 int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork,
3652 unsigned long delay)
3654 return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3656 EXPORT_SYMBOL(kblockd_schedule_delayed_work_on);
3659 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3660 * @plug: The &struct blk_plug that needs to be initialized
3663 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3664 * pending I/O should the task end up blocking between blk_start_plug() and
3665 * blk_finish_plug(). This is important from a performance perspective, but
3666 * also ensures that we don't deadlock. For instance, if the task is blocking
3667 * for a memory allocation, memory reclaim could end up wanting to free a
3668 * page belonging to that request that is currently residing in our private
3669 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3670 * this kind of deadlock.
3672 void blk_start_plug(struct blk_plug *plug)
3674 struct task_struct *tsk = current;
3677 * If this is a nested plug, don't actually assign it.
3682 INIT_LIST_HEAD(&plug->list);
3683 INIT_LIST_HEAD(&plug->mq_list);
3684 INIT_LIST_HEAD(&plug->cb_list);
3686 * Store ordering should not be needed here, since a potential
3687 * preempt will imply a full memory barrier
3691 EXPORT_SYMBOL(blk_start_plug);
3693 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
3695 struct request *rqa = container_of(a, struct request, queuelist);
3696 struct request *rqb = container_of(b, struct request, queuelist);
3698 return !(rqa->q < rqb->q ||
3699 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
3703 * If 'from_schedule' is true, then postpone the dispatch of requests
3704 * until a safe kblockd context. We due this to avoid accidental big
3705 * additional stack usage in driver dispatch, in places where the originally
3706 * plugger did not intend it.
3708 static void queue_unplugged(struct request_queue *q, unsigned int depth,
3710 __releases(q->queue_lock)
3712 trace_block_unplug(q, depth, !from_schedule);
3715 blk_run_queue_async(q);
3718 spin_unlock(q->queue_lock);
3721 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
3723 LIST_HEAD(callbacks);
3725 while (!list_empty(&plug->cb_list)) {
3726 list_splice_init(&plug->cb_list, &callbacks);
3728 while (!list_empty(&callbacks)) {
3729 struct blk_plug_cb *cb = list_first_entry(&callbacks,
3732 list_del(&cb->list);
3733 cb->callback(cb, from_schedule);
3738 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3741 struct blk_plug *plug = current->plug;
3742 struct blk_plug_cb *cb;
3747 list_for_each_entry(cb, &plug->cb_list, list)
3748 if (cb->callback == unplug && cb->data == data)
3751 /* Not currently on the callback list */
3752 BUG_ON(size < sizeof(*cb));
3753 cb = kzalloc(size, GFP_ATOMIC);
3756 cb->callback = unplug;
3757 list_add(&cb->list, &plug->cb_list);
3761 EXPORT_SYMBOL(blk_check_plugged);
3763 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
3765 struct request_queue *q;
3766 unsigned long flags;
3771 flush_plug_callbacks(plug, from_schedule);
3773 if (!list_empty(&plug->mq_list))
3774 blk_mq_flush_plug_list(plug, from_schedule);
3776 if (list_empty(&plug->list))
3779 list_splice_init(&plug->list, &list);
3781 list_sort(NULL, &list, plug_rq_cmp);
3787 * Save and disable interrupts here, to avoid doing it for every
3788 * queue lock we have to take.
3790 local_irq_save(flags);
3791 while (!list_empty(&list)) {
3792 rq = list_entry_rq(list.next);
3793 list_del_init(&rq->queuelist);
3797 * This drops the queue lock
3800 queue_unplugged(q, depth, from_schedule);
3803 spin_lock(q->queue_lock);
3807 * Short-circuit if @q is dead
3809 if (unlikely(blk_queue_dying(q))) {
3810 __blk_end_request_all(rq, -ENODEV);
3815 * rq is already accounted, so use raw insert
3817 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
3818 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3820 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
3826 * This drops the queue lock
3829 queue_unplugged(q, depth, from_schedule);
3831 local_irq_restore(flags);
3834 void blk_finish_plug(struct blk_plug *plug)
3836 if (plug != current->plug)
3838 blk_flush_plug_list(plug, false);
3840 current->plug = NULL;
3842 EXPORT_SYMBOL(blk_finish_plug);
3844 bool blk_poll(struct request_queue *q, blk_qc_t cookie)
3846 struct blk_plug *plug;
3849 if (!q->mq_ops || !q->mq_ops->poll || !blk_qc_t_valid(cookie) ||
3850 !test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
3853 plug = current->plug;
3855 blk_flush_plug_list(plug, false);
3857 state = current->state;
3858 while (!need_resched()) {
3859 unsigned int queue_num = blk_qc_t_to_queue_num(cookie);
3860 struct blk_mq_hw_ctx *hctx = q->queue_hw_ctx[queue_num];
3863 hctx->poll_invoked++;
3865 ret = q->mq_ops->poll(hctx, blk_qc_t_to_tag(cookie));
3867 hctx->poll_success++;
3868 set_current_state(TASK_RUNNING);
3872 if (signal_pending_state(state, current))
3873 set_current_state(TASK_RUNNING);
3875 if (current->state == TASK_RUNNING)
3887 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3888 * @q: the queue of the device
3889 * @dev: the device the queue belongs to
3892 * Initialize runtime-PM-related fields for @q and start auto suspend for
3893 * @dev. Drivers that want to take advantage of request-based runtime PM
3894 * should call this function after @dev has been initialized, and its
3895 * request queue @q has been allocated, and runtime PM for it can not happen
3896 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3897 * cases, driver should call this function before any I/O has taken place.
3899 * This function takes care of setting up using auto suspend for the device,
3900 * the autosuspend delay is set to -1 to make runtime suspend impossible
3901 * until an updated value is either set by user or by driver. Drivers do
3902 * not need to touch other autosuspend settings.
3904 * The block layer runtime PM is request based, so only works for drivers
3905 * that use request as their IO unit instead of those directly use bio's.
3907 void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3910 q->rpm_status = RPM_ACTIVE;
3911 pm_runtime_set_autosuspend_delay(q->dev, -1);
3912 pm_runtime_use_autosuspend(q->dev);
3914 EXPORT_SYMBOL(blk_pm_runtime_init);
3917 * blk_pre_runtime_suspend - Pre runtime suspend check
3918 * @q: the queue of the device
3921 * This function will check if runtime suspend is allowed for the device
3922 * by examining if there are any requests pending in the queue. If there
3923 * are requests pending, the device can not be runtime suspended; otherwise,
3924 * the queue's status will be updated to SUSPENDING and the driver can
3925 * proceed to suspend the device.
3927 * For the not allowed case, we mark last busy for the device so that
3928 * runtime PM core will try to autosuspend it some time later.
3930 * This function should be called near the start of the device's
3931 * runtime_suspend callback.
3934 * 0 - OK to runtime suspend the device
3935 * -EBUSY - Device should not be runtime suspended
3937 int blk_pre_runtime_suspend(struct request_queue *q)
3944 spin_lock_irq(q->queue_lock);
3945 if (q->nr_pending) {
3947 pm_runtime_mark_last_busy(q->dev);
3949 q->rpm_status = RPM_SUSPENDING;
3951 spin_unlock_irq(q->queue_lock);
3954 EXPORT_SYMBOL(blk_pre_runtime_suspend);
3957 * blk_post_runtime_suspend - Post runtime suspend processing
3958 * @q: the queue of the device
3959 * @err: return value of the device's runtime_suspend function
3962 * Update the queue's runtime status according to the return value of the
3963 * device's runtime suspend function and mark last busy for the device so
3964 * that PM core will try to auto suspend the device at a later time.
3966 * This function should be called near the end of the device's
3967 * runtime_suspend callback.
3969 void blk_post_runtime_suspend(struct request_queue *q, int err)
3974 spin_lock_irq(q->queue_lock);
3976 q->rpm_status = RPM_SUSPENDED;
3978 q->rpm_status = RPM_ACTIVE;
3979 pm_runtime_mark_last_busy(q->dev);
3981 spin_unlock_irq(q->queue_lock);
3983 EXPORT_SYMBOL(blk_post_runtime_suspend);
3986 * blk_pre_runtime_resume - Pre runtime resume processing
3987 * @q: the queue of the device
3990 * Update the queue's runtime status to RESUMING in preparation for the
3991 * runtime resume of the device.
3993 * This function should be called near the start of the device's
3994 * runtime_resume callback.
3996 void blk_pre_runtime_resume(struct request_queue *q)
4001 spin_lock_irq(q->queue_lock);
4002 q->rpm_status = RPM_RESUMING;
4003 spin_unlock_irq(q->queue_lock);
4005 EXPORT_SYMBOL(blk_pre_runtime_resume);
4008 * blk_post_runtime_resume - Post runtime resume processing
4009 * @q: the queue of the device
4010 * @err: return value of the device's runtime_resume function
4013 * Update the queue's runtime status according to the return value of the
4014 * device's runtime_resume function. If it is successfully resumed, process
4015 * the requests that are queued into the device's queue when it is resuming
4016 * and then mark last busy and initiate autosuspend for it.
4018 * This function should be called near the end of the device's
4019 * runtime_resume callback.
4021 void blk_post_runtime_resume(struct request_queue *q, int err)
4026 spin_lock_irq(q->queue_lock);
4028 q->rpm_status = RPM_ACTIVE;
4030 pm_runtime_mark_last_busy(q->dev);
4031 pm_request_autosuspend(q->dev);
4033 q->rpm_status = RPM_SUSPENDED;
4035 spin_unlock_irq(q->queue_lock);
4037 EXPORT_SYMBOL(blk_post_runtime_resume);
4040 int __init blk_dev_init(void)
4042 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
4043 FIELD_SIZEOF(struct request, cmd_flags));
4045 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
4046 kblockd_workqueue = alloc_workqueue("kblockd",
4047 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
4048 if (!kblockd_workqueue)
4049 panic("Failed to create kblockd\n");
4051 request_cachep = kmem_cache_create("blkdev_requests",
4052 sizeof(struct request), 0, SLAB_PANIC, NULL);
4054 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
4055 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
4061 * Blk IO latency support. We want this to be as cheap as possible, so doing
4062 * this lockless (and avoiding atomics), a few off by a few errors in this
4063 * code is not harmful, and we don't want to do anything that is
4065 * TODO : If necessary, we can make the histograms per-cpu and aggregate
4066 * them when printing them out.
4069 blk_latency_hist_show(char* name, struct io_latency_state *s, char *buf,
4073 int bytes_written = 0;
4074 u_int64_t num_elem, elem;
4078 num_elem = s->latency_elems;
4080 average = div64_u64(s->latency_sum, s->latency_elems);
4081 bytes_written += scnprintf(buf + bytes_written,
4082 buf_size - bytes_written,
4083 "IO svc_time %s Latency Histogram (n = %llu,"
4084 " average = %llu):\n", name, num_elem, average);
4086 i < ARRAY_SIZE(latency_x_axis_us);
4088 elem = s->latency_y_axis[i];
4089 pct = div64_u64(elem * 100, num_elem);
4090 bytes_written += scnprintf(buf + bytes_written,
4091 PAGE_SIZE - bytes_written,
4092 "\t< %6lluus%15llu%15d%%\n",
4093 latency_x_axis_us[i],
4096 /* Last element in y-axis table is overflow */
4097 elem = s->latency_y_axis[i];
4098 pct = div64_u64(elem * 100, num_elem);
4099 bytes_written += scnprintf(buf + bytes_written,
4100 PAGE_SIZE - bytes_written,
4101 "\t>=%6lluus%15llu%15d%%\n",
4102 latency_x_axis_us[i - 1], elem, pct);
4105 return bytes_written;
4107 EXPORT_SYMBOL(blk_latency_hist_show);