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);
885 EXPORT_SYMBOL(blk_init_allocated_queue);
887 bool blk_get_queue(struct request_queue *q)
889 if (likely(!blk_queue_dying(q))) {
896 EXPORT_SYMBOL(blk_get_queue);
898 static inline void blk_free_request(struct request_list *rl, struct request *rq)
900 if (rq->cmd_flags & REQ_ELVPRIV) {
901 elv_put_request(rl->q, rq);
903 put_io_context(rq->elv.icq->ioc);
906 mempool_free(rq, rl->rq_pool);
910 * ioc_batching returns true if the ioc is a valid batching request and
911 * should be given priority access to a request.
913 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
919 * Make sure the process is able to allocate at least 1 request
920 * even if the batch times out, otherwise we could theoretically
923 return ioc->nr_batch_requests == q->nr_batching ||
924 (ioc->nr_batch_requests > 0
925 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
929 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
930 * will cause the process to be a "batcher" on all queues in the system. This
931 * is the behaviour we want though - once it gets a wakeup it should be given
934 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
936 if (!ioc || ioc_batching(q, ioc))
939 ioc->nr_batch_requests = q->nr_batching;
940 ioc->last_waited = jiffies;
943 static void __freed_request(struct request_list *rl, int sync)
945 struct request_queue *q = rl->q;
947 if (rl->count[sync] < queue_congestion_off_threshold(q))
948 blk_clear_congested(rl, sync);
950 if (rl->count[sync] + 1 <= q->nr_requests) {
951 if (waitqueue_active(&rl->wait[sync]))
952 wake_up(&rl->wait[sync]);
954 blk_clear_rl_full(rl, sync);
959 * A request has just been released. Account for it, update the full and
960 * congestion status, wake up any waiters. Called under q->queue_lock.
962 static void freed_request(struct request_list *rl, unsigned int flags)
964 struct request_queue *q = rl->q;
965 int sync = rw_is_sync(flags);
969 if (flags & REQ_ELVPRIV)
972 __freed_request(rl, sync);
974 if (unlikely(rl->starved[sync ^ 1]))
975 __freed_request(rl, sync ^ 1);
978 int blk_update_nr_requests(struct request_queue *q, unsigned int nr)
980 struct request_list *rl;
981 int on_thresh, off_thresh;
983 spin_lock_irq(q->queue_lock);
985 blk_queue_congestion_threshold(q);
986 on_thresh = queue_congestion_on_threshold(q);
987 off_thresh = queue_congestion_off_threshold(q);
989 blk_queue_for_each_rl(rl, q) {
990 if (rl->count[BLK_RW_SYNC] >= on_thresh)
991 blk_set_congested(rl, BLK_RW_SYNC);
992 else if (rl->count[BLK_RW_SYNC] < off_thresh)
993 blk_clear_congested(rl, BLK_RW_SYNC);
995 if (rl->count[BLK_RW_ASYNC] >= on_thresh)
996 blk_set_congested(rl, BLK_RW_ASYNC);
997 else if (rl->count[BLK_RW_ASYNC] < off_thresh)
998 blk_clear_congested(rl, BLK_RW_ASYNC);
1000 if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
1001 blk_set_rl_full(rl, BLK_RW_SYNC);
1003 blk_clear_rl_full(rl, BLK_RW_SYNC);
1004 wake_up(&rl->wait[BLK_RW_SYNC]);
1007 if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
1008 blk_set_rl_full(rl, BLK_RW_ASYNC);
1010 blk_clear_rl_full(rl, BLK_RW_ASYNC);
1011 wake_up(&rl->wait[BLK_RW_ASYNC]);
1015 spin_unlock_irq(q->queue_lock);
1020 * Determine if elevator data should be initialized when allocating the
1021 * request associated with @bio.
1023 static bool blk_rq_should_init_elevator(struct bio *bio)
1029 * Flush requests do not use the elevator so skip initialization.
1030 * This allows a request to share the flush and elevator data.
1032 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
1039 * rq_ioc - determine io_context for request allocation
1040 * @bio: request being allocated is for this bio (can be %NULL)
1042 * Determine io_context to use for request allocation for @bio. May return
1043 * %NULL if %current->io_context doesn't exist.
1045 static struct io_context *rq_ioc(struct bio *bio)
1047 #ifdef CONFIG_BLK_CGROUP
1048 if (bio && bio->bi_ioc)
1051 return current->io_context;
1055 * __get_request - get a free request
1056 * @rl: request list to allocate from
1057 * @rw_flags: RW and SYNC flags
1058 * @bio: bio to allocate request for (can be %NULL)
1059 * @gfp_mask: allocation mask
1061 * Get a free request from @q. This function may fail under memory
1062 * pressure or if @q is dead.
1064 * Must be called with @q->queue_lock held and,
1065 * Returns ERR_PTR on failure, with @q->queue_lock held.
1066 * Returns request pointer on success, with @q->queue_lock *not held*.
1068 static struct request *__get_request(struct request_list *rl, int rw_flags,
1069 struct bio *bio, gfp_t gfp_mask)
1071 struct request_queue *q = rl->q;
1073 struct elevator_type *et = q->elevator->type;
1074 struct io_context *ioc = rq_ioc(bio);
1075 struct io_cq *icq = NULL;
1076 const bool is_sync = rw_is_sync(rw_flags) != 0;
1079 if (unlikely(blk_queue_dying(q)))
1080 return ERR_PTR(-ENODEV);
1082 may_queue = elv_may_queue(q, rw_flags);
1083 if (may_queue == ELV_MQUEUE_NO)
1086 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
1087 if (rl->count[is_sync]+1 >= q->nr_requests) {
1089 * The queue will fill after this allocation, so set
1090 * it as full, and mark this process as "batching".
1091 * This process will be allowed to complete a batch of
1092 * requests, others will be blocked.
1094 if (!blk_rl_full(rl, is_sync)) {
1095 ioc_set_batching(q, ioc);
1096 blk_set_rl_full(rl, is_sync);
1098 if (may_queue != ELV_MQUEUE_MUST
1099 && !ioc_batching(q, ioc)) {
1101 * The queue is full and the allocating
1102 * process is not a "batcher", and not
1103 * exempted by the IO scheduler
1105 return ERR_PTR(-ENOMEM);
1109 blk_set_congested(rl, is_sync);
1113 * Only allow batching queuers to allocate up to 50% over the defined
1114 * limit of requests, otherwise we could have thousands of requests
1115 * allocated with any setting of ->nr_requests
1117 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
1118 return ERR_PTR(-ENOMEM);
1120 q->nr_rqs[is_sync]++;
1121 rl->count[is_sync]++;
1122 rl->starved[is_sync] = 0;
1125 * Decide whether the new request will be managed by elevator. If
1126 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
1127 * prevent the current elevator from being destroyed until the new
1128 * request is freed. This guarantees icq's won't be destroyed and
1129 * makes creating new ones safe.
1131 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1132 * it will be created after releasing queue_lock.
1134 if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
1135 rw_flags |= REQ_ELVPRIV;
1136 q->nr_rqs_elvpriv++;
1137 if (et->icq_cache && ioc)
1138 icq = ioc_lookup_icq(ioc, q);
1141 if (blk_queue_io_stat(q))
1142 rw_flags |= REQ_IO_STAT;
1143 spin_unlock_irq(q->queue_lock);
1145 /* allocate and init request */
1146 rq = mempool_alloc(rl->rq_pool, gfp_mask);
1151 blk_rq_set_rl(rq, rl);
1152 rq->cmd_flags = rw_flags | REQ_ALLOCED;
1155 if (rw_flags & REQ_ELVPRIV) {
1156 if (unlikely(et->icq_cache && !icq)) {
1158 icq = ioc_create_icq(ioc, q, gfp_mask);
1164 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1167 /* @rq->elv.icq holds io_context until @rq is freed */
1169 get_io_context(icq->ioc);
1173 * ioc may be NULL here, and ioc_batching will be false. That's
1174 * OK, if the queue is under the request limit then requests need
1175 * not count toward the nr_batch_requests limit. There will always
1176 * be some limit enforced by BLK_BATCH_TIME.
1178 if (ioc_batching(q, ioc))
1179 ioc->nr_batch_requests--;
1181 trace_block_getrq(q, bio, rw_flags & 1);
1186 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1187 * and may fail indefinitely under memory pressure and thus
1188 * shouldn't stall IO. Treat this request as !elvpriv. This will
1189 * disturb iosched and blkcg but weird is bettern than dead.
1191 printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1192 __func__, dev_name(q->backing_dev_info->dev));
1194 rq->cmd_flags &= ~REQ_ELVPRIV;
1197 spin_lock_irq(q->queue_lock);
1198 q->nr_rqs_elvpriv--;
1199 spin_unlock_irq(q->queue_lock);
1204 * Allocation failed presumably due to memory. Undo anything we
1205 * might have messed up.
1207 * Allocating task should really be put onto the front of the wait
1208 * queue, but this is pretty rare.
1210 spin_lock_irq(q->queue_lock);
1211 freed_request(rl, rw_flags);
1214 * in the very unlikely event that allocation failed and no
1215 * requests for this direction was pending, mark us starved so that
1216 * freeing of a request in the other direction will notice
1217 * us. another possible fix would be to split the rq mempool into
1221 if (unlikely(rl->count[is_sync] == 0))
1222 rl->starved[is_sync] = 1;
1223 return ERR_PTR(-ENOMEM);
1227 * get_request - get a free request
1228 * @q: request_queue to allocate request from
1229 * @rw_flags: RW and SYNC flags
1230 * @bio: bio to allocate request for (can be %NULL)
1231 * @gfp_mask: allocation mask
1233 * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask,
1234 * this function keeps retrying under memory pressure and fails iff @q is dead.
1236 * Must be called with @q->queue_lock held and,
1237 * Returns ERR_PTR on failure, with @q->queue_lock held.
1238 * Returns request pointer on success, with @q->queue_lock *not held*.
1240 static struct request *get_request(struct request_queue *q, int rw_flags,
1241 struct bio *bio, gfp_t gfp_mask)
1243 const bool is_sync = rw_is_sync(rw_flags) != 0;
1245 struct request_list *rl;
1248 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
1250 rq = __get_request(rl, rw_flags, bio, gfp_mask);
1254 if (!gfpflags_allow_blocking(gfp_mask) || unlikely(blk_queue_dying(q))) {
1259 /* wait on @rl and retry */
1260 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1261 TASK_UNINTERRUPTIBLE);
1263 trace_block_sleeprq(q, bio, rw_flags & 1);
1265 spin_unlock_irq(q->queue_lock);
1269 * After sleeping, we become a "batching" process and will be able
1270 * to allocate at least one request, and up to a big batch of them
1271 * for a small period time. See ioc_batching, ioc_set_batching
1273 ioc_set_batching(q, current->io_context);
1275 spin_lock_irq(q->queue_lock);
1276 finish_wait(&rl->wait[is_sync], &wait);
1281 static struct request *blk_old_get_request(struct request_queue *q, int rw,
1286 /* create ioc upfront */
1287 create_io_context(gfp_mask, q->node);
1289 spin_lock_irq(q->queue_lock);
1290 rq = get_request(q, rw, NULL, gfp_mask);
1292 spin_unlock_irq(q->queue_lock);
1293 /* q->queue_lock is unlocked at this point */
1298 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1301 return blk_mq_alloc_request(q, rw, gfp_mask, false);
1303 return blk_old_get_request(q, rw, gfp_mask);
1305 EXPORT_SYMBOL(blk_get_request);
1308 * blk_make_request - given a bio, allocate a corresponding struct request.
1309 * @q: target request queue
1310 * @bio: The bio describing the memory mappings that will be submitted for IO.
1311 * It may be a chained-bio properly constructed by block/bio layer.
1312 * @gfp_mask: gfp flags to be used for memory allocation
1314 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1315 * type commands. Where the struct request needs to be farther initialized by
1316 * the caller. It is passed a &struct bio, which describes the memory info of
1319 * The caller of blk_make_request must make sure that bi_io_vec
1320 * are set to describe the memory buffers. That bio_data_dir() will return
1321 * the needed direction of the request. (And all bio's in the passed bio-chain
1322 * are properly set accordingly)
1324 * If called under none-sleepable conditions, mapped bio buffers must not
1325 * need bouncing, by calling the appropriate masked or flagged allocator,
1326 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1329 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1330 * given to how you allocate bios. In particular, you cannot use
1331 * __GFP_DIRECT_RECLAIM for anything but the first bio in the chain. Otherwise
1332 * you risk waiting for IO completion of a bio that hasn't been submitted yet,
1333 * thus resulting in a deadlock. Alternatively bios should be allocated using
1334 * bio_kmalloc() instead of bio_alloc(), as that avoids the mempool deadlock.
1335 * If possible a big IO should be split into smaller parts when allocation
1336 * fails. Partial allocation should not be an error, or you risk a live-lock.
1338 struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1341 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1346 blk_rq_set_block_pc(rq);
1349 struct bio *bounce_bio = bio;
1352 blk_queue_bounce(q, &bounce_bio);
1353 ret = blk_rq_append_bio(q, rq, bounce_bio);
1354 if (unlikely(ret)) {
1355 blk_put_request(rq);
1356 return ERR_PTR(ret);
1362 EXPORT_SYMBOL(blk_make_request);
1365 * blk_rq_set_block_pc - initialize a request to type BLOCK_PC
1366 * @rq: request to be initialized
1369 void blk_rq_set_block_pc(struct request *rq)
1371 rq->cmd_type = REQ_TYPE_BLOCK_PC;
1373 rq->__sector = (sector_t) -1;
1374 rq->bio = rq->biotail = NULL;
1375 memset(rq->__cmd, 0, sizeof(rq->__cmd));
1377 EXPORT_SYMBOL(blk_rq_set_block_pc);
1380 * blk_requeue_request - put a request back on queue
1381 * @q: request queue where request should be inserted
1382 * @rq: request to be inserted
1385 * Drivers often keep queueing requests until the hardware cannot accept
1386 * more, when that condition happens we need to put the request back
1387 * on the queue. Must be called with queue lock held.
1389 void blk_requeue_request(struct request_queue *q, struct request *rq)
1391 blk_delete_timer(rq);
1392 blk_clear_rq_complete(rq);
1393 trace_block_rq_requeue(q, rq);
1395 if (rq->cmd_flags & REQ_QUEUED)
1396 blk_queue_end_tag(q, rq);
1398 BUG_ON(blk_queued_rq(rq));
1400 elv_requeue_request(q, rq);
1402 EXPORT_SYMBOL(blk_requeue_request);
1404 static void add_acct_request(struct request_queue *q, struct request *rq,
1407 blk_account_io_start(rq, true);
1408 __elv_add_request(q, rq, where);
1411 static void part_round_stats_single(int cpu, struct hd_struct *part,
1416 if (now == part->stamp)
1419 inflight = part_in_flight(part);
1421 __part_stat_add(cpu, part, time_in_queue,
1422 inflight * (now - part->stamp));
1423 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1429 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1430 * @cpu: cpu number for stats access
1431 * @part: target partition
1433 * The average IO queue length and utilisation statistics are maintained
1434 * by observing the current state of the queue length and the amount of
1435 * time it has been in this state for.
1437 * Normally, that accounting is done on IO completion, but that can result
1438 * in more than a second's worth of IO being accounted for within any one
1439 * second, leading to >100% utilisation. To deal with that, we call this
1440 * function to do a round-off before returning the results when reading
1441 * /proc/diskstats. This accounts immediately for all queue usage up to
1442 * the current jiffies and restarts the counters again.
1444 void part_round_stats(int cpu, struct hd_struct *part)
1446 unsigned long now = jiffies;
1449 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1450 part_round_stats_single(cpu, part, now);
1452 EXPORT_SYMBOL_GPL(part_round_stats);
1455 static void blk_pm_put_request(struct request *rq)
1457 if (rq->q->dev && !(rq->cmd_flags & REQ_PM) && rq->q->nr_pending) {
1458 if (!--rq->q->nr_pending)
1459 pm_runtime_mark_last_busy(rq->q->dev);
1463 static inline void blk_pm_put_request(struct request *rq) {}
1467 * queue lock must be held
1469 void __blk_put_request(struct request_queue *q, struct request *req)
1475 blk_mq_free_request(req);
1479 blk_pm_put_request(req);
1481 elv_completed_request(q, req);
1483 /* this is a bio leak */
1484 WARN_ON(req->bio != NULL);
1486 /* this is a bio leak if the bio is not tagged with BIO_DONTFREE */
1487 WARN_ON(req->bio && !bio_flagged(req->bio, BIO_DONTFREE));
1490 * Request may not have originated from ll_rw_blk. if not,
1491 * it didn't come out of our reserved rq pools
1493 if (req->cmd_flags & REQ_ALLOCED) {
1494 unsigned int flags = req->cmd_flags;
1495 struct request_list *rl = blk_rq_rl(req);
1497 BUG_ON(!list_empty(&req->queuelist));
1498 BUG_ON(ELV_ON_HASH(req));
1500 blk_free_request(rl, req);
1501 freed_request(rl, flags);
1505 EXPORT_SYMBOL_GPL(__blk_put_request);
1507 void blk_put_request(struct request *req)
1509 struct request_queue *q = req->q;
1512 blk_mq_free_request(req);
1514 unsigned long flags;
1516 spin_lock_irqsave(q->queue_lock, flags);
1517 __blk_put_request(q, req);
1518 spin_unlock_irqrestore(q->queue_lock, flags);
1521 EXPORT_SYMBOL(blk_put_request);
1524 * blk_add_request_payload - add a payload to a request
1525 * @rq: request to update
1526 * @page: page backing the payload
1527 * @len: length of the payload.
1529 * This allows to later add a payload to an already submitted request by
1530 * a block driver. The driver needs to take care of freeing the payload
1533 * Note that this is a quite horrible hack and nothing but handling of
1534 * discard requests should ever use it.
1536 void blk_add_request_payload(struct request *rq, struct page *page,
1539 struct bio *bio = rq->bio;
1541 bio->bi_io_vec->bv_page = page;
1542 bio->bi_io_vec->bv_offset = 0;
1543 bio->bi_io_vec->bv_len = len;
1545 bio->bi_iter.bi_size = len;
1547 bio->bi_phys_segments = 1;
1549 rq->__data_len = rq->resid_len = len;
1550 rq->nr_phys_segments = 1;
1552 EXPORT_SYMBOL_GPL(blk_add_request_payload);
1554 bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1557 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1559 if (!ll_back_merge_fn(q, req, bio))
1562 trace_block_bio_backmerge(q, req, bio);
1564 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1565 blk_rq_set_mixed_merge(req);
1567 req->biotail->bi_next = bio;
1569 req->__data_len += bio->bi_iter.bi_size;
1570 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1572 blk_account_io_start(req, false);
1576 bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1579 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1581 if (!ll_front_merge_fn(q, req, bio))
1584 trace_block_bio_frontmerge(q, req, bio);
1586 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1587 blk_rq_set_mixed_merge(req);
1589 bio->bi_next = req->bio;
1592 req->__sector = bio->bi_iter.bi_sector;
1593 req->__data_len += bio->bi_iter.bi_size;
1594 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1596 blk_account_io_start(req, false);
1601 * blk_attempt_plug_merge - try to merge with %current's plugged list
1602 * @q: request_queue new bio is being queued at
1603 * @bio: new bio being queued
1604 * @request_count: out parameter for number of traversed plugged requests
1605 * @same_queue_rq: pointer to &struct request that gets filled in when
1606 * another request associated with @q is found on the plug list
1607 * (optional, may be %NULL)
1609 * Determine whether @bio being queued on @q can be merged with a request
1610 * on %current's plugged list. Returns %true if merge was successful,
1613 * Plugging coalesces IOs from the same issuer for the same purpose without
1614 * going through @q->queue_lock. As such it's more of an issuing mechanism
1615 * than scheduling, and the request, while may have elvpriv data, is not
1616 * added on the elevator at this point. In addition, we don't have
1617 * reliable access to the elevator outside queue lock. Only check basic
1618 * merging parameters without querying the elevator.
1620 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1622 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1623 unsigned int *request_count,
1624 struct request **same_queue_rq)
1626 struct blk_plug *plug;
1629 struct list_head *plug_list;
1631 plug = current->plug;
1637 plug_list = &plug->mq_list;
1639 plug_list = &plug->list;
1641 list_for_each_entry_reverse(rq, plug_list, queuelist) {
1647 * Only blk-mq multiple hardware queues case checks the
1648 * rq in the same queue, there should be only one such
1652 *same_queue_rq = rq;
1655 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
1658 el_ret = blk_try_merge(rq, bio);
1659 if (el_ret == ELEVATOR_BACK_MERGE) {
1660 ret = bio_attempt_back_merge(q, rq, bio);
1663 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1664 ret = bio_attempt_front_merge(q, rq, bio);
1673 unsigned int blk_plug_queued_count(struct request_queue *q)
1675 struct blk_plug *plug;
1677 struct list_head *plug_list;
1678 unsigned int ret = 0;
1680 plug = current->plug;
1685 plug_list = &plug->mq_list;
1687 plug_list = &plug->list;
1689 list_for_each_entry(rq, plug_list, queuelist) {
1697 void init_request_from_bio(struct request *req, struct bio *bio)
1699 req->cmd_type = REQ_TYPE_FS;
1701 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1702 if (bio->bi_rw & REQ_RAHEAD)
1703 req->cmd_flags |= REQ_FAILFAST_MASK;
1706 req->__sector = bio->bi_iter.bi_sector;
1707 req->ioprio = bio_prio(bio);
1708 blk_rq_bio_prep(req->q, req, bio);
1710 EXPORT_SYMBOL(init_request_from_bio);
1712 static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio)
1714 const bool sync = !!(bio->bi_rw & REQ_SYNC);
1715 struct blk_plug *plug;
1716 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1717 struct request *req;
1718 unsigned int request_count = 0;
1721 * low level driver can indicate that it wants pages above a
1722 * certain limit bounced to low memory (ie for highmem, or even
1723 * ISA dma in theory)
1725 blk_queue_bounce(q, &bio);
1727 blk_queue_split(q, &bio, q->bio_split);
1729 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
1730 bio->bi_error = -EIO;
1732 return BLK_QC_T_NONE;
1735 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA | REQ_POST_FLUSH_BARRIER |
1737 spin_lock_irq(q->queue_lock);
1738 where = ELEVATOR_INSERT_FLUSH;
1743 * Check if we can merge with the plugged list before grabbing
1746 if (!blk_queue_nomerges(q)) {
1747 if (blk_attempt_plug_merge(q, bio, &request_count, NULL))
1748 return BLK_QC_T_NONE;
1750 request_count = blk_plug_queued_count(q);
1752 spin_lock_irq(q->queue_lock);
1754 el_ret = elv_merge(q, &req, bio);
1755 if (el_ret == ELEVATOR_BACK_MERGE) {
1756 if (bio_attempt_back_merge(q, req, bio)) {
1757 elv_bio_merged(q, req, bio);
1758 if (!attempt_back_merge(q, req))
1759 elv_merged_request(q, req, el_ret);
1762 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1763 if (bio_attempt_front_merge(q, req, bio)) {
1764 elv_bio_merged(q, req, bio);
1765 if (!attempt_front_merge(q, req))
1766 elv_merged_request(q, req, el_ret);
1773 * This sync check and mask will be re-done in init_request_from_bio(),
1774 * but we need to set it earlier to expose the sync flag to the
1775 * rq allocator and io schedulers.
1777 rw_flags = bio_data_dir(bio);
1779 rw_flags |= REQ_SYNC;
1782 * Grab a free request. This is might sleep but can not fail.
1783 * Returns with the queue unlocked.
1785 req = get_request(q, rw_flags, bio, GFP_NOIO);
1787 bio->bi_error = PTR_ERR(req);
1793 * After dropping the lock and possibly sleeping here, our request
1794 * may now be mergeable after it had proven unmergeable (above).
1795 * We don't worry about that case for efficiency. It won't happen
1796 * often, and the elevators are able to handle it.
1798 init_request_from_bio(req, bio);
1800 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
1801 req->cpu = raw_smp_processor_id();
1803 plug = current->plug;
1806 * If this is the first request added after a plug, fire
1810 trace_block_plug(q);
1812 if (request_count >= BLK_MAX_REQUEST_COUNT) {
1813 blk_flush_plug_list(plug, false);
1814 trace_block_plug(q);
1817 list_add_tail(&req->queuelist, &plug->list);
1818 blk_account_io_start(req, true);
1820 spin_lock_irq(q->queue_lock);
1821 add_acct_request(q, req, where);
1824 spin_unlock_irq(q->queue_lock);
1827 return BLK_QC_T_NONE;
1831 * If bio->bi_dev is a partition, remap the location
1833 static inline void blk_partition_remap(struct bio *bio)
1835 struct block_device *bdev = bio->bi_bdev;
1837 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1838 struct hd_struct *p = bdev->bd_part;
1840 bio->bi_iter.bi_sector += p->start_sect;
1841 bio->bi_bdev = bdev->bd_contains;
1843 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1845 bio->bi_iter.bi_sector - p->start_sect);
1849 static void handle_bad_sector(struct bio *bio)
1851 char b[BDEVNAME_SIZE];
1853 printk(KERN_INFO "attempt to access beyond end of device\n");
1854 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1855 bdevname(bio->bi_bdev, b),
1857 (unsigned long long)bio_end_sector(bio),
1858 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1861 #ifdef CONFIG_FAIL_MAKE_REQUEST
1863 static DECLARE_FAULT_ATTR(fail_make_request);
1865 static int __init setup_fail_make_request(char *str)
1867 return setup_fault_attr(&fail_make_request, str);
1869 __setup("fail_make_request=", setup_fail_make_request);
1871 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1873 return part->make_it_fail && should_fail(&fail_make_request, bytes);
1876 static int __init fail_make_request_debugfs(void)
1878 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1879 NULL, &fail_make_request);
1881 return PTR_ERR_OR_ZERO(dir);
1884 late_initcall(fail_make_request_debugfs);
1886 #else /* CONFIG_FAIL_MAKE_REQUEST */
1888 static inline bool should_fail_request(struct hd_struct *part,
1894 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1897 * Check whether this bio extends beyond the end of the device.
1899 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1906 /* Test device or partition size, when known. */
1907 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1909 sector_t sector = bio->bi_iter.bi_sector;
1911 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1913 * This may well happen - the kernel calls bread()
1914 * without checking the size of the device, e.g., when
1915 * mounting a device.
1917 handle_bad_sector(bio);
1925 static noinline_for_stack bool
1926 generic_make_request_checks(struct bio *bio)
1928 struct request_queue *q;
1929 int nr_sectors = bio_sectors(bio);
1931 char b[BDEVNAME_SIZE];
1932 struct hd_struct *part;
1936 if (bio_check_eod(bio, nr_sectors))
1939 q = bdev_get_queue(bio->bi_bdev);
1942 "generic_make_request: Trying to access "
1943 "nonexistent block-device %s (%Lu)\n",
1944 bdevname(bio->bi_bdev, b),
1945 (long long) bio->bi_iter.bi_sector);
1949 part = bio->bi_bdev->bd_part;
1950 if (should_fail_request(part, bio->bi_iter.bi_size) ||
1951 should_fail_request(&part_to_disk(part)->part0,
1952 bio->bi_iter.bi_size))
1956 * If this device has partitions, remap block n
1957 * of partition p to block n+start(p) of the disk.
1959 blk_partition_remap(bio);
1961 if (bio_check_eod(bio, nr_sectors))
1965 * Filter flush bio's early so that make_request based
1966 * drivers without flush support don't have to worry
1969 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1970 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1977 if ((bio->bi_rw & REQ_DISCARD) &&
1978 (!blk_queue_discard(q) ||
1979 ((bio->bi_rw & REQ_SECURE) && !blk_queue_secdiscard(q)))) {
1984 if (bio->bi_rw & REQ_WRITE_SAME && !bdev_write_same(bio->bi_bdev)) {
1990 * Various block parts want %current->io_context and lazy ioc
1991 * allocation ends up trading a lot of pain for a small amount of
1992 * memory. Just allocate it upfront. This may fail and block
1993 * layer knows how to live with it.
1995 create_io_context(GFP_ATOMIC, q->node);
1997 if (!blkcg_bio_issue_check(q, bio))
2000 trace_block_bio_queue(q, bio);
2004 bio->bi_error = err;
2010 * generic_make_request - hand a buffer to its device driver for I/O
2011 * @bio: The bio describing the location in memory and on the device.
2013 * generic_make_request() is used to make I/O requests of block
2014 * devices. It is passed a &struct bio, which describes the I/O that needs
2017 * generic_make_request() does not return any status. The
2018 * success/failure status of the request, along with notification of
2019 * completion, is delivered asynchronously through the bio->bi_end_io
2020 * function described (one day) else where.
2022 * The caller of generic_make_request must make sure that bi_io_vec
2023 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2024 * set to describe the device address, and the
2025 * bi_end_io and optionally bi_private are set to describe how
2026 * completion notification should be signaled.
2028 * generic_make_request and the drivers it calls may use bi_next if this
2029 * bio happens to be merged with someone else, and may resubmit the bio to
2030 * a lower device by calling into generic_make_request recursively, which
2031 * means the bio should NOT be touched after the call to ->make_request_fn.
2033 blk_qc_t generic_make_request(struct bio *bio)
2036 * bio_list_on_stack[0] contains bios submitted by the current
2038 * bio_list_on_stack[1] contains bios that were submitted before
2039 * the current make_request_fn, but that haven't been processed
2042 struct bio_list bio_list_on_stack[2];
2043 blk_qc_t ret = BLK_QC_T_NONE;
2045 if (!generic_make_request_checks(bio))
2049 * We only want one ->make_request_fn to be active at a time, else
2050 * stack usage with stacked devices could be a problem. So use
2051 * current->bio_list to keep a list of requests submited by a
2052 * make_request_fn function. current->bio_list is also used as a
2053 * flag to say if generic_make_request is currently active in this
2054 * task or not. If it is NULL, then no make_request is active. If
2055 * it is non-NULL, then a make_request is active, and new requests
2056 * should be added at the tail
2058 if (current->bio_list) {
2059 bio_list_add(¤t->bio_list[0], bio);
2063 /* following loop may be a bit non-obvious, and so deserves some
2065 * Before entering the loop, bio->bi_next is NULL (as all callers
2066 * ensure that) so we have a list with a single bio.
2067 * We pretend that we have just taken it off a longer list, so
2068 * we assign bio_list to a pointer to the bio_list_on_stack,
2069 * thus initialising the bio_list of new bios to be
2070 * added. ->make_request() may indeed add some more bios
2071 * through a recursive call to generic_make_request. If it
2072 * did, we find a non-NULL value in bio_list and re-enter the loop
2073 * from the top. In this case we really did just take the bio
2074 * of the top of the list (no pretending) and so remove it from
2075 * bio_list, and call into ->make_request() again.
2077 BUG_ON(bio->bi_next);
2078 bio_list_init(&bio_list_on_stack[0]);
2079 current->bio_list = bio_list_on_stack;
2081 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
2083 if (likely(blk_queue_enter(q, __GFP_DIRECT_RECLAIM) == 0)) {
2084 struct bio_list lower, same;
2086 /* Create a fresh bio_list for all subordinate requests */
2087 bio_list_on_stack[1] = bio_list_on_stack[0];
2088 bio_list_init(&bio_list_on_stack[0]);
2090 ret = q->make_request_fn(q, bio);
2093 /* sort new bios into those for a lower level
2094 * and those for the same level
2096 bio_list_init(&lower);
2097 bio_list_init(&same);
2098 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
2099 if (q == bdev_get_queue(bio->bi_bdev))
2100 bio_list_add(&same, bio);
2102 bio_list_add(&lower, bio);
2103 /* now assemble so we handle the lowest level first */
2104 bio_list_merge(&bio_list_on_stack[0], &lower);
2105 bio_list_merge(&bio_list_on_stack[0], &same);
2106 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
2110 bio = bio_list_pop(&bio_list_on_stack[0]);
2112 current->bio_list = NULL; /* deactivate */
2117 EXPORT_SYMBOL(generic_make_request);
2119 #ifdef CONFIG_BLK_DEV_IO_TRACE
2120 static inline struct task_struct *get_dirty_task(struct bio *bio)
2123 * Not all the pages in the bio are dirtied by the
2124 * same task but most likely it will be, since the
2125 * sectors accessed on the device must be adjacent.
2127 if (bio->bi_io_vec && bio->bi_io_vec->bv_page &&
2128 bio->bi_io_vec->bv_page->tsk_dirty)
2129 return bio->bi_io_vec->bv_page->tsk_dirty;
2134 static inline struct task_struct *get_dirty_task(struct bio *bio)
2140 #ifdef CONFIG_BLOCK_PERF_FRAMEWORK
2141 #define BLK_PERF_SIZE (1024 * 15)
2142 #define BLK_PERF_HIST_SIZE (sizeof(u32) * BLK_PERF_SIZE)
2144 struct blk_perf_stats {
2148 int buffers_alloced;
2149 ktime_t max_read_time;
2150 ktime_t max_write_time;
2151 ktime_t max_flush_time;
2152 ktime_t min_write_time;
2153 ktime_t min_read_time;
2154 ktime_t min_flush_time;
2155 ktime_t total_write_time;
2156 ktime_t total_read_time;
2157 u64 total_read_size;
2158 u64 total_write_size;
2163 static struct blk_perf_stats blk_perf;
2164 static struct dentry *blk_perf_debug_dir;
2166 static int alloc_histogram_buffers(void)
2170 if (!blk_perf.read_hist)
2171 blk_perf.read_hist = kzalloc(BLK_PERF_HIST_SIZE, GFP_KERNEL);
2173 if (!blk_perf.write_hist)
2174 blk_perf.write_hist = kzalloc(BLK_PERF_HIST_SIZE, GFP_KERNEL);
2176 if (!blk_perf.flush_hist)
2177 blk_perf.flush_hist = kzalloc(BLK_PERF_HIST_SIZE, GFP_KERNEL);
2179 if (!blk_perf.read_hist || !blk_perf.write_hist || !blk_perf.flush_hist)
2183 blk_perf.buffers_alloced = 1;
2187 static void clear_histogram_buffers(void)
2189 if (!blk_perf.buffers_alloced)
2191 memset(blk_perf.read_hist, 0, BLK_PERF_HIST_SIZE);
2192 memset(blk_perf.write_hist, 0, BLK_PERF_HIST_SIZE);
2193 memset(blk_perf.flush_hist, 0, BLK_PERF_HIST_SIZE);
2196 static int enable_perf(void *data, u64 val)
2200 if (!blk_perf.buffers_alloced)
2201 ret = alloc_histogram_buffers();
2206 spin_lock(&blk_perf.lock);
2207 blk_perf.is_enabled = val;
2208 spin_unlock(&blk_perf.lock);
2212 static int is_perf_enabled(void *data, u64 *val)
2214 spin_lock(&blk_perf.lock);
2215 *val = blk_perf.is_enabled;
2216 spin_unlock(&blk_perf.lock);
2220 DEFINE_SIMPLE_ATTRIBUTE(enable_perf_fops, is_perf_enabled, enable_perf,
2223 static char *blk_debug_buffer;
2224 static u32 blk_debug_data_size;
2225 static DEFINE_MUTEX(blk_perf_debug_buffer_mutex);
2227 static ssize_t blk_perf_read(struct file *file, char __user *buf,
2228 size_t count, loff_t *file_pos)
2232 mutex_lock(&blk_perf_debug_buffer_mutex);
2233 ret = simple_read_from_buffer(buf, count, file_pos, blk_debug_buffer,
2234 blk_debug_data_size);
2235 mutex_unlock(&blk_perf_debug_buffer_mutex);
2240 static int blk_debug_buffer_alloc(u32 buffer_size)
2244 mutex_lock(&blk_perf_debug_buffer_mutex);
2245 if (blk_debug_buffer != NULL) {
2246 pr_err("blk_debug_buffer is in use\n");
2250 blk_debug_buffer = kzalloc(buffer_size, GFP_KERNEL);
2251 if (!blk_debug_buffer)
2254 mutex_unlock(&blk_perf_debug_buffer_mutex);
2258 static int blk_perf_close(struct inode *inode, struct file *file)
2260 mutex_lock(&blk_perf_debug_buffer_mutex);
2261 blk_debug_data_size = 0;
2262 kfree(blk_debug_buffer);
2263 blk_debug_buffer = NULL;
2264 mutex_unlock(&blk_perf_debug_buffer_mutex);
2268 static u32 fill_basic_perf_info(char *buffer, u32 buffer_size)
2272 size += scnprintf(buffer + size, buffer_size - size, "\n");
2274 spin_lock(&blk_perf.lock);
2275 size += scnprintf(buffer + size, buffer_size - size,
2276 "max_read_time_ms: %llu\n",
2277 ktime_to_ms(blk_perf.max_read_time));
2279 size += scnprintf(buffer + size, buffer_size - size,
2280 "min_read_time_ms: %llu\n",
2281 ktime_to_ms(blk_perf.min_read_time));
2283 size += scnprintf(buffer + size, buffer_size - size,
2284 "total_read_time_ms: %llu\n",
2285 ktime_to_ms(blk_perf.total_read_time));
2287 size += scnprintf(buffer + size, buffer_size - size,
2288 "total_read_size: %llu\n\n",
2289 blk_perf.total_read_size);
2291 size += scnprintf(buffer + size, buffer_size - size,
2292 "max_write_time_ms: %llu\n",
2293 ktime_to_ms(blk_perf.max_write_time));
2295 size += scnprintf(buffer + size, buffer_size - size,
2296 "min_write_time_ms: %llu\n",
2297 ktime_to_ms(blk_perf.min_write_time));
2299 size += scnprintf(buffer + size, buffer_size - size,
2300 "total_write_time_ms: %llu\n",
2301 ktime_to_ms(blk_perf.total_write_time));
2303 size += scnprintf(buffer + size, buffer_size - size,
2304 "total_write_size: %llu\n\n",
2305 blk_perf.total_write_size);
2307 size += scnprintf(buffer + size, buffer_size - size,
2308 "max_flush_time_ms: %llu\n",
2309 ktime_to_ms(blk_perf.max_flush_time));
2311 size += scnprintf(buffer + size, buffer_size - size,
2312 "min_flush_time_ms: %llu\n\n",
2313 ktime_to_ms(blk_perf.min_flush_time));
2315 spin_unlock(&blk_perf.lock);
2320 static int basic_perf_open(struct inode *inode, struct file *file)
2325 buffer_size = BLK_PERF_HIST_SIZE;
2326 ret = blk_debug_buffer_alloc(buffer_size);
2330 mutex_lock(&blk_perf_debug_buffer_mutex);
2331 blk_debug_data_size = fill_basic_perf_info(blk_debug_buffer,
2333 mutex_unlock(&blk_perf_debug_buffer_mutex);
2338 static const struct file_operations basic_perf_ops = {
2339 .read = blk_perf_read,
2340 .release = blk_perf_close,
2341 .open = basic_perf_open,
2344 static int hist_open_helper(void *hist_buf)
2348 if (!blk_perf.buffers_alloced)
2351 ret = blk_debug_buffer_alloc(BLK_PERF_HIST_SIZE);
2355 spin_lock(&blk_perf.lock);
2356 memcpy(blk_debug_buffer, hist_buf, BLK_PERF_HIST_SIZE);
2357 spin_unlock(&blk_perf.lock);
2359 mutex_lock(&blk_perf_debug_buffer_mutex);
2360 blk_debug_data_size = BLK_PERF_HIST_SIZE;
2361 mutex_unlock(&blk_perf_debug_buffer_mutex);
2365 static int write_hist_open(struct inode *inode, struct file *file)
2367 return hist_open_helper(blk_perf.write_hist);
2370 static const struct file_operations write_hist_ops = {
2371 .read = blk_perf_read,
2372 .release = blk_perf_close,
2373 .open = write_hist_open,
2377 static int read_hist_open(struct inode *inode, struct file *file)
2379 return hist_open_helper(blk_perf.read_hist);
2382 static const struct file_operations read_hist_ops = {
2383 .read = blk_perf_read,
2384 .release = blk_perf_close,
2385 .open = read_hist_open,
2388 static int flush_hist_open(struct inode *inode, struct file *file)
2390 return hist_open_helper(blk_perf.flush_hist);
2393 static const struct file_operations flush_hist_ops = {
2394 .read = blk_perf_read,
2395 .release = blk_perf_close,
2396 .open = flush_hist_open,
2399 static void clear_perf_stats_helper(void)
2401 spin_lock(&blk_perf.lock);
2402 blk_perf.max_write_time = ktime_set(0, 0);
2403 blk_perf.max_read_time = ktime_set(0, 0);
2404 blk_perf.max_flush_time = ktime_set(0, 0);
2405 blk_perf.min_write_time = ktime_set(KTIME_MAX, 0);
2406 blk_perf.min_read_time = ktime_set(KTIME_MAX, 0);
2407 blk_perf.min_flush_time = ktime_set(KTIME_MAX, 0);
2408 blk_perf.total_write_time = ktime_set(0, 0);
2409 blk_perf.total_read_time = ktime_set(0, 0);
2410 blk_perf.total_read_size = 0;
2411 blk_perf.total_write_size = 0;
2412 blk_perf.is_enabled = 0;
2413 clear_histogram_buffers();
2414 spin_unlock(&blk_perf.lock);
2417 static int clear_perf_stats(void *data, u64 val)
2419 clear_perf_stats_helper();
2423 DEFINE_SIMPLE_ATTRIBUTE(clear_perf_stats_fops, NULL, clear_perf_stats,
2426 static void blk_debugfs_init(void)
2428 struct dentry *f_ent;
2430 blk_perf_debug_dir = debugfs_create_dir("block_perf", NULL);
2431 if (IS_ERR(blk_perf_debug_dir)) {
2432 pr_err("Failed to create block_perf debug_fs directory\n");
2436 f_ent = debugfs_create_file("basic_perf", 0400, blk_perf_debug_dir,
2437 NULL, &basic_perf_ops);
2438 if (IS_ERR(f_ent)) {
2439 pr_err("Failed to create debug_fs basic_perf file\n");
2443 f_ent = debugfs_create_file("write_hist", 0400, blk_perf_debug_dir,
2444 NULL, &write_hist_ops);
2445 if (IS_ERR(f_ent)) {
2446 pr_err("Failed to create debug_fs write_hist file\n");
2450 f_ent = debugfs_create_file("read_hist", 0400, blk_perf_debug_dir,
2451 NULL, &read_hist_ops);
2452 if (IS_ERR(f_ent)) {
2453 pr_err("Failed to create debug_fs read_hist file\n");
2457 f_ent = debugfs_create_file("flush_hist", 0400, blk_perf_debug_dir,
2458 NULL, &flush_hist_ops);
2459 if (IS_ERR(f_ent)) {
2460 pr_err("Failed to create debug_fs flush_hist file\n");
2464 f_ent = debugfs_create_file("enable_perf", 0600, blk_perf_debug_dir,
2465 NULL, &enable_perf_fops);
2466 if (IS_ERR(f_ent)) {
2467 pr_err("Failed to create debug_fs enable_perf file\n");
2471 f_ent = debugfs_create_file("clear_perf_stats", 0200,
2472 blk_perf_debug_dir, NULL,
2473 &clear_perf_stats_fops);
2474 if (IS_ERR(f_ent)) {
2475 pr_err("Failed to create debug_fs clear_perf_stats file\n");
2480 static void blk_init_perf(void)
2483 spin_lock_init(&blk_perf.lock);
2485 clear_perf_stats_helper();
2489 static void set_submit_info(struct bio *bio, unsigned int count)
2491 ktime_t submit_time;
2493 if (unlikely(blk_perf.is_enabled)) {
2494 submit_time = ktime_get();
2495 bio->submit_time.tv64 = submit_time.tv64;
2496 bio->blk_sector_count = count;
2500 bio->submit_time.tv64 = 0;
2501 bio->blk_sector_count = 0;
2504 void blk_update_perf_read_write_stats(ktime_t bio_process_time, int is_write,
2507 u32 bio_process_time_ms;
2509 bio_process_time_ms = ktime_to_ms(bio_process_time);
2510 if (bio_process_time_ms >= BLK_PERF_SIZE)
2511 bio_process_time_ms = BLK_PERF_SIZE - 1;
2514 if (ktime_after(bio_process_time, blk_perf.max_write_time))
2515 blk_perf.max_write_time = bio_process_time;
2517 if (ktime_before(bio_process_time, blk_perf.min_write_time))
2518 blk_perf.min_write_time = bio_process_time;
2519 blk_perf.total_write_time =
2520 ktime_add(blk_perf.total_write_time, bio_process_time);
2521 blk_perf.total_write_size += count;
2522 blk_perf.write_hist[bio_process_time_ms] += count;
2525 if (ktime_after(bio_process_time, blk_perf.max_read_time))
2526 blk_perf.max_read_time = bio_process_time;
2528 if (ktime_before(bio_process_time, blk_perf.min_read_time))
2529 blk_perf.min_read_time = bio_process_time;
2530 blk_perf.total_read_time =
2531 ktime_add(blk_perf.total_read_time, bio_process_time);
2532 blk_perf.total_read_size += count;
2533 blk_perf.read_hist[bio_process_time_ms] += count;
2536 void blk_update_perf_stats(struct bio *bio)
2538 ktime_t bio_process_time;
2539 u32 bio_process_time_ms;
2542 spin_lock(&blk_perf.lock);
2543 if (likely(!blk_perf.is_enabled))
2545 if (!bio->submit_time.tv64)
2547 bio_process_time = ktime_sub(ktime_get(), bio->submit_time);
2549 count = bio->blk_sector_count;
2554 if (bio->bi_rw & WRITE ||
2555 unlikely(bio->bi_rw & REQ_WRITE_SAME))
2558 blk_update_perf_read_write_stats(bio_process_time, is_write,
2562 bio_process_time_ms = ktime_to_ms(bio_process_time);
2563 if (bio_process_time_ms >= BLK_PERF_SIZE)
2564 bio_process_time_ms = BLK_PERF_SIZE - 1;
2566 if (ktime_after(bio_process_time, blk_perf.max_flush_time))
2567 blk_perf.max_flush_time = bio_process_time;
2569 if (ktime_before(bio_process_time, blk_perf.min_flush_time))
2570 blk_perf.min_flush_time = bio_process_time;
2572 blk_perf.flush_hist[bio_process_time_ms] += 1;
2575 spin_unlock(&blk_perf.lock);
2579 static inline void set_submit_info(struct bio *bio, unsigned int count)
2585 static inline void blk_init_perf(void)
2588 #endif /* #ifdef CONFIG_BLOCK_PERF_FRAMEWORK */
2591 * submit_bio - submit a bio to the block device layer for I/O
2592 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
2593 * @bio: The &struct bio which describes the I/O
2595 * submit_bio() is very similar in purpose to generic_make_request(), and
2596 * uses that function to do most of the work. Both are fairly rough
2597 * interfaces; @bio must be presetup and ready for I/O.
2600 blk_qc_t submit_bio(int rw, struct bio *bio)
2602 unsigned int count = 0;
2606 * If it's a regular read/write or a barrier with data attached,
2607 * go through the normal accounting stuff before submission.
2609 if (bio_has_data(bio)) {
2610 if (unlikely(rw & REQ_WRITE_SAME))
2611 count = bdev_logical_block_size(bio->bi_bdev) >> 9;
2613 count = bio_sectors(bio);
2616 count_vm_events(PGPGOUT, count);
2618 task_io_account_read(bio->bi_iter.bi_size);
2619 count_vm_events(PGPGIN, count);
2622 if (unlikely(block_dump)) {
2623 char b[BDEVNAME_SIZE];
2624 struct task_struct *tsk;
2626 tsk = get_dirty_task(bio);
2627 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
2628 tsk->comm, task_pid_nr(tsk),
2629 (rw & WRITE) ? "WRITE" : "READ",
2630 (unsigned long long)bio->bi_iter.bi_sector,
2631 bdevname(bio->bi_bdev, b),
2636 set_submit_info(bio, count);
2637 return generic_make_request(bio);
2639 EXPORT_SYMBOL(submit_bio);
2642 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2643 * for new the queue limits
2645 * @rq: the request being checked
2648 * @rq may have been made based on weaker limitations of upper-level queues
2649 * in request stacking drivers, and it may violate the limitation of @q.
2650 * Since the block layer and the underlying device driver trust @rq
2651 * after it is inserted to @q, it should be checked against @q before
2652 * the insertion using this generic function.
2654 * Request stacking drivers like request-based dm may change the queue
2655 * limits when retrying requests on other queues. Those requests need
2656 * to be checked against the new queue limits again during dispatch.
2658 static int blk_cloned_rq_check_limits(struct request_queue *q,
2661 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, rq->cmd_flags)) {
2662 printk(KERN_ERR "%s: over max size limit.\n", __func__);
2667 * queue's settings related to segment counting like q->bounce_pfn
2668 * may differ from that of other stacking queues.
2669 * Recalculate it to check the request correctly on this queue's
2672 blk_recalc_rq_segments(rq);
2673 if (rq->nr_phys_segments > queue_max_segments(q)) {
2674 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2682 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2683 * @q: the queue to submit the request
2684 * @rq: the request being queued
2686 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
2688 unsigned long flags;
2689 int where = ELEVATOR_INSERT_BACK;
2691 if (blk_cloned_rq_check_limits(q, rq))
2695 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
2699 if (blk_queue_io_stat(q))
2700 blk_account_io_start(rq, true);
2701 blk_mq_insert_request(rq, false, true, false);
2705 spin_lock_irqsave(q->queue_lock, flags);
2706 if (unlikely(blk_queue_dying(q))) {
2707 spin_unlock_irqrestore(q->queue_lock, flags);
2712 * Submitting request must be dequeued before calling this function
2713 * because it will be linked to another request_queue
2715 BUG_ON(blk_queued_rq(rq));
2717 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
2718 where = ELEVATOR_INSERT_FLUSH;
2720 add_acct_request(q, rq, where);
2721 if (where == ELEVATOR_INSERT_FLUSH)
2723 spin_unlock_irqrestore(q->queue_lock, flags);
2727 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2730 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2731 * @rq: request to examine
2734 * A request could be merge of IOs which require different failure
2735 * handling. This function determines the number of bytes which
2736 * can be failed from the beginning of the request without
2737 * crossing into area which need to be retried further.
2740 * The number of bytes to fail.
2743 * queue_lock must be held.
2745 unsigned int blk_rq_err_bytes(const struct request *rq)
2747 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2748 unsigned int bytes = 0;
2751 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
2752 return blk_rq_bytes(rq);
2755 * Currently the only 'mixing' which can happen is between
2756 * different fastfail types. We can safely fail portions
2757 * which have all the failfast bits that the first one has -
2758 * the ones which are at least as eager to fail as the first
2761 for (bio = rq->bio; bio; bio = bio->bi_next) {
2762 if ((bio->bi_rw & ff) != ff)
2764 bytes += bio->bi_iter.bi_size;
2767 /* this could lead to infinite loop */
2768 BUG_ON(blk_rq_bytes(rq) && !bytes);
2771 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2773 void blk_account_io_completion(struct request *req, unsigned int bytes)
2775 if (blk_do_io_stat(req)) {
2776 const int rw = rq_data_dir(req);
2777 struct hd_struct *part;
2780 cpu = part_stat_lock();
2782 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2787 void blk_account_io_done(struct request *req)
2790 * Account IO completion. flush_rq isn't accounted as a
2791 * normal IO on queueing nor completion. Accounting the
2792 * containing request is enough.
2794 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
2795 unsigned long duration = jiffies - req->start_time;
2796 const int rw = rq_data_dir(req);
2797 struct hd_struct *part;
2800 cpu = part_stat_lock();
2803 part_stat_inc(cpu, part, ios[rw]);
2804 part_stat_add(cpu, part, ticks[rw], duration);
2805 part_round_stats(cpu, part);
2806 part_dec_in_flight(part, rw);
2808 hd_struct_put(part);
2815 * Don't process normal requests when queue is suspended
2816 * or in the process of suspending/resuming
2818 static struct request *blk_pm_peek_request(struct request_queue *q,
2821 if (q->dev && (q->rpm_status == RPM_SUSPENDED ||
2822 (q->rpm_status != RPM_ACTIVE && !(rq->cmd_flags & REQ_PM))))
2828 static inline struct request *blk_pm_peek_request(struct request_queue *q,
2835 void blk_account_io_start(struct request *rq, bool new_io)
2837 struct hd_struct *part;
2838 int rw = rq_data_dir(rq);
2841 if (!blk_do_io_stat(rq))
2844 cpu = part_stat_lock();
2848 part_stat_inc(cpu, part, merges[rw]);
2850 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2851 if (!hd_struct_try_get(part)) {
2853 * The partition is already being removed,
2854 * the request will be accounted on the disk only
2856 * We take a reference on disk->part0 although that
2857 * partition will never be deleted, so we can treat
2858 * it as any other partition.
2860 part = &rq->rq_disk->part0;
2861 hd_struct_get(part);
2863 part_round_stats(cpu, part);
2864 part_inc_in_flight(part, rw);
2872 * blk_peek_request - peek at the top of a request queue
2873 * @q: request queue to peek at
2876 * Return the request at the top of @q. The returned request
2877 * should be started using blk_start_request() before LLD starts
2881 * Pointer to the request at the top of @q if available. Null
2885 * queue_lock must be held.
2887 struct request *blk_peek_request(struct request_queue *q)
2892 while ((rq = __elv_next_request(q)) != NULL) {
2894 rq = blk_pm_peek_request(q, rq);
2898 if (!(rq->cmd_flags & REQ_STARTED)) {
2900 * This is the first time the device driver
2901 * sees this request (possibly after
2902 * requeueing). Notify IO scheduler.
2904 if (rq->cmd_flags & REQ_SORTED)
2905 elv_activate_rq(q, rq);
2908 * just mark as started even if we don't start
2909 * it, a request that has been delayed should
2910 * not be passed by new incoming requests
2912 rq->cmd_flags |= REQ_STARTED;
2913 trace_block_rq_issue(q, rq);
2916 if (!q->boundary_rq || q->boundary_rq == rq) {
2917 q->end_sector = rq_end_sector(rq);
2918 q->boundary_rq = NULL;
2921 if (rq->cmd_flags & REQ_DONTPREP)
2924 if (q->dma_drain_size && blk_rq_bytes(rq)) {
2926 * make sure space for the drain appears we
2927 * know we can do this because max_hw_segments
2928 * has been adjusted to be one fewer than the
2931 rq->nr_phys_segments++;
2937 ret = q->prep_rq_fn(q, rq);
2938 if (ret == BLKPREP_OK) {
2940 } else if (ret == BLKPREP_DEFER) {
2942 * the request may have been (partially) prepped.
2943 * we need to keep this request in the front to
2944 * avoid resource deadlock. REQ_STARTED will
2945 * prevent other fs requests from passing this one.
2947 if (q->dma_drain_size && blk_rq_bytes(rq) &&
2948 !(rq->cmd_flags & REQ_DONTPREP)) {
2950 * remove the space for the drain we added
2951 * so that we don't add it again
2953 --rq->nr_phys_segments;
2958 } else if (ret == BLKPREP_KILL) {
2959 rq->cmd_flags |= REQ_QUIET;
2961 * Mark this request as started so we don't trigger
2962 * any debug logic in the end I/O path.
2964 blk_start_request(rq);
2965 __blk_end_request_all(rq, -EIO);
2967 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2974 EXPORT_SYMBOL(blk_peek_request);
2976 void blk_dequeue_request(struct request *rq)
2978 struct request_queue *q = rq->q;
2980 BUG_ON(list_empty(&rq->queuelist));
2981 BUG_ON(ELV_ON_HASH(rq));
2983 list_del_init(&rq->queuelist);
2986 * the time frame between a request being removed from the lists
2987 * and to it is freed is accounted as io that is in progress at
2990 if (blk_account_rq(rq)) {
2991 q->in_flight[rq_is_sync(rq)]++;
2992 set_io_start_time_ns(rq);
2997 * blk_start_request - start request processing on the driver
2998 * @req: request to dequeue
3001 * Dequeue @req and start timeout timer on it. This hands off the
3002 * request to the driver.
3004 * Block internal functions which don't want to start timer should
3005 * call blk_dequeue_request().
3008 * queue_lock must be held.
3010 void blk_start_request(struct request *req)
3012 blk_dequeue_request(req);
3015 * We are now handing the request to the hardware, initialize
3016 * resid_len to full count and add the timeout handler.
3018 req->resid_len = blk_rq_bytes(req);
3019 if (unlikely(blk_bidi_rq(req)))
3020 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
3022 BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags));
3025 EXPORT_SYMBOL(blk_start_request);
3028 * blk_fetch_request - fetch a request from a request queue
3029 * @q: request queue to fetch a request from
3032 * Return the request at the top of @q. The request is started on
3033 * return and LLD can start processing it immediately.
3036 * Pointer to the request at the top of @q if available. Null
3040 * queue_lock must be held.
3042 struct request *blk_fetch_request(struct request_queue *q)
3046 rq = blk_peek_request(q);
3048 blk_start_request(rq);
3051 EXPORT_SYMBOL(blk_fetch_request);
3054 * blk_update_request - Special helper function for request stacking drivers
3055 * @req: the request being processed
3056 * @error: %0 for success, < %0 for error
3057 * @nr_bytes: number of bytes to complete @req
3060 * Ends I/O on a number of bytes attached to @req, but doesn't complete
3061 * the request structure even if @req doesn't have leftover.
3062 * If @req has leftover, sets it up for the next range of segments.
3064 * This special helper function is only for request stacking drivers
3065 * (e.g. request-based dm) so that they can handle partial completion.
3066 * Actual device drivers should use blk_end_request instead.
3068 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
3069 * %false return from this function.
3072 * %false - this request doesn't have any more data
3073 * %true - this request has more data
3075 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
3079 trace_block_rq_complete(req->q, req, nr_bytes);
3085 * For fs requests, rq is just carrier of independent bio's
3086 * and each partial completion should be handled separately.
3087 * Reset per-request error on each partial completion.
3089 * TODO: tj: This is too subtle. It would be better to let
3090 * low level drivers do what they see fit.
3092 if (req->cmd_type == REQ_TYPE_FS)
3095 if (error && req->cmd_type == REQ_TYPE_FS &&
3096 !(req->cmd_flags & REQ_QUIET)) {
3101 error_type = "recoverable transport";
3104 error_type = "critical target";
3107 error_type = "critical nexus";
3110 error_type = "timeout";
3113 error_type = "critical space allocation";
3116 error_type = "critical medium";
3123 printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n",
3124 __func__, error_type, req->rq_disk ?
3125 req->rq_disk->disk_name : "?",
3126 (unsigned long long)blk_rq_pos(req));
3130 blk_account_io_completion(req, nr_bytes);
3135 * Check for this if flagged, Req based dm needs to perform
3136 * post processing, hence dont end bios or request.DM
3139 if (bio_flagged(req->bio, BIO_DONTFREE))
3143 struct bio *bio = req->bio;
3144 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
3146 if (bio_bytes == bio->bi_iter.bi_size)
3147 req->bio = bio->bi_next;
3149 req_bio_endio(req, bio, bio_bytes, error);
3151 total_bytes += bio_bytes;
3152 nr_bytes -= bio_bytes;
3163 * Reset counters so that the request stacking driver
3164 * can find how many bytes remain in the request
3167 req->__data_len = 0;
3171 req->__data_len -= total_bytes;
3173 /* update sector only for requests with clear definition of sector */
3174 if (req->cmd_type == REQ_TYPE_FS)
3175 req->__sector += total_bytes >> 9;
3177 /* mixed attributes always follow the first bio */
3178 if (req->cmd_flags & REQ_MIXED_MERGE) {
3179 req->cmd_flags &= ~REQ_FAILFAST_MASK;
3180 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
3184 * If total number of sectors is less than the first segment
3185 * size, something has gone terribly wrong.
3187 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
3188 blk_dump_rq_flags(req, "request botched");
3189 req->__data_len = blk_rq_cur_bytes(req);
3192 /* recalculate the number of segments */
3193 blk_recalc_rq_segments(req);
3197 EXPORT_SYMBOL_GPL(blk_update_request);
3199 static bool blk_update_bidi_request(struct request *rq, int error,
3200 unsigned int nr_bytes,
3201 unsigned int bidi_bytes)
3203 if (blk_update_request(rq, error, nr_bytes))
3206 /* Bidi request must be completed as a whole */
3207 if (unlikely(blk_bidi_rq(rq)) &&
3208 blk_update_request(rq->next_rq, error, bidi_bytes))
3211 if (blk_queue_add_random(rq->q))
3212 add_disk_randomness(rq->rq_disk);
3218 * blk_unprep_request - unprepare a request
3221 * This function makes a request ready for complete resubmission (or
3222 * completion). It happens only after all error handling is complete,
3223 * so represents the appropriate moment to deallocate any resources
3224 * that were allocated to the request in the prep_rq_fn. The queue
3225 * lock is held when calling this.
3227 void blk_unprep_request(struct request *req)
3229 struct request_queue *q = req->q;
3231 req->cmd_flags &= ~REQ_DONTPREP;
3232 if (q->unprep_rq_fn)
3233 q->unprep_rq_fn(q, req);
3235 EXPORT_SYMBOL_GPL(blk_unprep_request);
3238 * queue lock must be held
3240 void blk_finish_request(struct request *req, int error)
3242 if (req->cmd_flags & REQ_QUEUED)
3243 blk_queue_end_tag(req->q, req);
3245 BUG_ON(blk_queued_rq(req));
3247 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
3248 laptop_io_completion(req->q->backing_dev_info);
3250 blk_delete_timer(req);
3252 if (req->cmd_flags & REQ_DONTPREP)
3253 blk_unprep_request(req);
3255 blk_account_io_done(req);
3258 req->end_io(req, error);
3260 if (blk_bidi_rq(req))
3261 __blk_put_request(req->next_rq->q, req->next_rq);
3263 __blk_put_request(req->q, req);
3266 EXPORT_SYMBOL(blk_finish_request);
3269 * blk_end_bidi_request - Complete a bidi request
3270 * @rq: the request to complete
3271 * @error: %0 for success, < %0 for error
3272 * @nr_bytes: number of bytes to complete @rq
3273 * @bidi_bytes: number of bytes to complete @rq->next_rq
3276 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
3277 * Drivers that supports bidi can safely call this member for any
3278 * type of request, bidi or uni. In the later case @bidi_bytes is
3282 * %false - we are done with this request
3283 * %true - still buffers pending for this request
3285 static bool blk_end_bidi_request(struct request *rq, int error,
3286 unsigned int nr_bytes, unsigned int bidi_bytes)
3288 struct request_queue *q = rq->q;
3289 unsigned long flags;
3291 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
3294 spin_lock_irqsave(q->queue_lock, flags);
3295 blk_finish_request(rq, error);
3296 spin_unlock_irqrestore(q->queue_lock, flags);
3302 * __blk_end_bidi_request - Complete a bidi request with queue lock held
3303 * @rq: the request to complete
3304 * @error: %0 for success, < %0 for error
3305 * @nr_bytes: number of bytes to complete @rq
3306 * @bidi_bytes: number of bytes to complete @rq->next_rq
3309 * Identical to blk_end_bidi_request() except that queue lock is
3310 * assumed to be locked on entry and remains so on return.
3313 * %false - we are done with this request
3314 * %true - still buffers pending for this request
3316 bool __blk_end_bidi_request(struct request *rq, int error,
3317 unsigned int nr_bytes, unsigned int bidi_bytes)
3319 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
3322 blk_finish_request(rq, error);
3328 * blk_end_request - Helper function for drivers to complete the request.
3329 * @rq: the request being processed
3330 * @error: %0 for success, < %0 for error
3331 * @nr_bytes: number of bytes to complete
3334 * Ends I/O on a number of bytes attached to @rq.
3335 * If @rq has leftover, sets it up for the next range of segments.
3338 * %false - we are done with this request
3339 * %true - still buffers pending for this request
3341 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
3343 return blk_end_bidi_request(rq, error, nr_bytes, 0);
3345 EXPORT_SYMBOL(blk_end_request);
3348 * blk_end_request_all - Helper function for drives to finish the request.
3349 * @rq: the request to finish
3350 * @error: %0 for success, < %0 for error
3353 * Completely finish @rq.
3355 void blk_end_request_all(struct request *rq, int error)
3358 unsigned int bidi_bytes = 0;
3360 if (unlikely(blk_bidi_rq(rq)))
3361 bidi_bytes = blk_rq_bytes(rq->next_rq);
3363 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
3366 EXPORT_SYMBOL(blk_end_request_all);
3369 * blk_end_request_cur - Helper function to finish the current request chunk.
3370 * @rq: the request to finish the current chunk for
3371 * @error: %0 for success, < %0 for error
3374 * Complete the current consecutively mapped chunk from @rq.
3377 * %false - we are done with this request
3378 * %true - still buffers pending for this request
3380 bool blk_end_request_cur(struct request *rq, int error)
3382 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
3384 EXPORT_SYMBOL(blk_end_request_cur);
3387 * blk_end_request_err - Finish a request till the next failure boundary.
3388 * @rq: the request to finish till the next failure boundary for
3389 * @error: must be negative errno
3392 * Complete @rq till the next failure boundary.
3395 * %false - we are done with this request
3396 * %true - still buffers pending for this request
3398 bool blk_end_request_err(struct request *rq, int error)
3400 WARN_ON(error >= 0);
3401 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
3403 EXPORT_SYMBOL_GPL(blk_end_request_err);
3406 * __blk_end_request - Helper function for drivers to complete the request.
3407 * @rq: the request being processed
3408 * @error: %0 for success, < %0 for error
3409 * @nr_bytes: number of bytes to complete
3412 * Must be called with queue lock held unlike blk_end_request().
3415 * %false - we are done with this request
3416 * %true - still buffers pending for this request
3418 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
3420 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
3422 EXPORT_SYMBOL(__blk_end_request);
3425 * __blk_end_request_all - Helper function for drives to finish the request.
3426 * @rq: the request to finish
3427 * @error: %0 for success, < %0 for error
3430 * Completely finish @rq. Must be called with queue lock held.
3432 void __blk_end_request_all(struct request *rq, int error)
3435 unsigned int bidi_bytes = 0;
3437 if (unlikely(blk_bidi_rq(rq)))
3438 bidi_bytes = blk_rq_bytes(rq->next_rq);
3440 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
3443 EXPORT_SYMBOL(__blk_end_request_all);
3446 * __blk_end_request_cur - Helper function to finish the current request chunk.
3447 * @rq: the request to finish the current chunk for
3448 * @error: %0 for success, < %0 for error
3451 * Complete the current consecutively mapped chunk from @rq. Must
3452 * be called with queue lock held.
3455 * %false - we are done with this request
3456 * %true - still buffers pending for this request
3458 bool __blk_end_request_cur(struct request *rq, int error)
3460 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
3462 EXPORT_SYMBOL(__blk_end_request_cur);
3465 * __blk_end_request_err - Finish a request till the next failure boundary.
3466 * @rq: the request to finish till the next failure boundary for
3467 * @error: must be negative errno
3470 * Complete @rq till the next failure boundary. Must be called
3471 * with queue lock held.
3474 * %false - we are done with this request
3475 * %true - still buffers pending for this request
3477 bool __blk_end_request_err(struct request *rq, int error)
3479 WARN_ON(error >= 0);
3480 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
3482 EXPORT_SYMBOL_GPL(__blk_end_request_err);
3484 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
3487 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
3488 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
3490 if (bio_has_data(bio))
3491 rq->nr_phys_segments = bio_phys_segments(q, bio);
3493 rq->__data_len = bio->bi_iter.bi_size;
3494 rq->bio = rq->biotail = bio;
3497 rq->rq_disk = bio->bi_bdev->bd_disk;
3500 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
3502 * rq_flush_dcache_pages - Helper function to flush all pages in a request
3503 * @rq: the request to be flushed
3506 * Flush all pages in @rq.
3508 void rq_flush_dcache_pages(struct request *rq)
3510 struct req_iterator iter;
3511 struct bio_vec bvec;
3513 rq_for_each_segment(bvec, rq, iter)
3514 flush_dcache_page(bvec.bv_page);
3516 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
3520 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
3521 * @q : the queue of the device being checked
3524 * Check if underlying low-level drivers of a device are busy.
3525 * If the drivers want to export their busy state, they must set own
3526 * exporting function using blk_queue_lld_busy() first.
3528 * Basically, this function is used only by request stacking drivers
3529 * to stop dispatching requests to underlying devices when underlying
3530 * devices are busy. This behavior helps more I/O merging on the queue
3531 * of the request stacking driver and prevents I/O throughput regression
3532 * on burst I/O load.
3535 * 0 - Not busy (The request stacking driver should dispatch request)
3536 * 1 - Busy (The request stacking driver should stop dispatching request)
3538 int blk_lld_busy(struct request_queue *q)
3541 return q->lld_busy_fn(q);
3545 EXPORT_SYMBOL_GPL(blk_lld_busy);
3548 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
3549 * @rq: the clone request to be cleaned up
3552 * Free all bios in @rq for a cloned request.
3554 void blk_rq_unprep_clone(struct request *rq)
3558 while ((bio = rq->bio) != NULL) {
3559 rq->bio = bio->bi_next;
3564 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
3567 * Copy attributes of the original request to the clone request.
3568 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3570 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
3572 dst->cpu = src->cpu;
3573 dst->cmd_flags |= (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
3574 dst->cmd_type = src->cmd_type;
3575 dst->__sector = blk_rq_pos(src);
3576 dst->__data_len = blk_rq_bytes(src);
3577 dst->nr_phys_segments = src->nr_phys_segments;
3578 dst->ioprio = src->ioprio;
3579 dst->extra_len = src->extra_len;
3583 * blk_rq_prep_clone - Helper function to setup clone request
3584 * @rq: the request to be setup
3585 * @rq_src: original request to be cloned
3586 * @bs: bio_set that bios for clone are allocated from
3587 * @gfp_mask: memory allocation mask for bio
3588 * @bio_ctr: setup function to be called for each clone bio.
3589 * Returns %0 for success, non %0 for failure.
3590 * @data: private data to be passed to @bio_ctr
3593 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3594 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3595 * are not copied, and copying such parts is the caller's responsibility.
3596 * Also, pages which the original bios are pointing to are not copied
3597 * and the cloned bios just point same pages.
3598 * So cloned bios must be completed before original bios, which means
3599 * the caller must complete @rq before @rq_src.
3601 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
3602 struct bio_set *bs, gfp_t gfp_mask,
3603 int (*bio_ctr)(struct bio *, struct bio *, void *),
3606 struct bio *bio, *bio_src;
3611 __rq_for_each_bio(bio_src, rq_src) {
3612 bio = bio_clone_fast(bio_src, gfp_mask, bs);
3616 if (bio_ctr && bio_ctr(bio, bio_src, data))
3620 rq->biotail->bi_next = bio;
3623 rq->bio = rq->biotail = bio;
3626 __blk_rq_prep_clone(rq, rq_src);
3633 blk_rq_unprep_clone(rq);
3637 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
3639 int kblockd_schedule_work(struct work_struct *work)
3641 return queue_work(kblockd_workqueue, work);
3643 EXPORT_SYMBOL(kblockd_schedule_work);
3645 int kblockd_schedule_delayed_work(struct delayed_work *dwork,
3646 unsigned long delay)
3648 return queue_delayed_work(kblockd_workqueue, dwork, delay);
3650 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
3652 int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork,
3653 unsigned long delay)
3655 return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3657 EXPORT_SYMBOL(kblockd_schedule_delayed_work_on);
3660 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3661 * @plug: The &struct blk_plug that needs to be initialized
3664 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3665 * pending I/O should the task end up blocking between blk_start_plug() and
3666 * blk_finish_plug(). This is important from a performance perspective, but
3667 * also ensures that we don't deadlock. For instance, if the task is blocking
3668 * for a memory allocation, memory reclaim could end up wanting to free a
3669 * page belonging to that request that is currently residing in our private
3670 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3671 * this kind of deadlock.
3673 void blk_start_plug(struct blk_plug *plug)
3675 struct task_struct *tsk = current;
3678 * If this is a nested plug, don't actually assign it.
3683 INIT_LIST_HEAD(&plug->list);
3684 INIT_LIST_HEAD(&plug->mq_list);
3685 INIT_LIST_HEAD(&plug->cb_list);
3687 * Store ordering should not be needed here, since a potential
3688 * preempt will imply a full memory barrier
3692 EXPORT_SYMBOL(blk_start_plug);
3694 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
3696 struct request *rqa = container_of(a, struct request, queuelist);
3697 struct request *rqb = container_of(b, struct request, queuelist);
3699 return !(rqa->q < rqb->q ||
3700 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
3704 * If 'from_schedule' is true, then postpone the dispatch of requests
3705 * until a safe kblockd context. We due this to avoid accidental big
3706 * additional stack usage in driver dispatch, in places where the originally
3707 * plugger did not intend it.
3709 static void queue_unplugged(struct request_queue *q, unsigned int depth,
3711 __releases(q->queue_lock)
3713 trace_block_unplug(q, depth, !from_schedule);
3716 blk_run_queue_async(q);
3719 spin_unlock(q->queue_lock);
3722 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
3724 LIST_HEAD(callbacks);
3726 while (!list_empty(&plug->cb_list)) {
3727 list_splice_init(&plug->cb_list, &callbacks);
3729 while (!list_empty(&callbacks)) {
3730 struct blk_plug_cb *cb = list_first_entry(&callbacks,
3733 list_del(&cb->list);
3734 cb->callback(cb, from_schedule);
3739 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3742 struct blk_plug *plug = current->plug;
3743 struct blk_plug_cb *cb;
3748 list_for_each_entry(cb, &plug->cb_list, list)
3749 if (cb->callback == unplug && cb->data == data)
3752 /* Not currently on the callback list */
3753 BUG_ON(size < sizeof(*cb));
3754 cb = kzalloc(size, GFP_ATOMIC);
3757 cb->callback = unplug;
3758 list_add(&cb->list, &plug->cb_list);
3762 EXPORT_SYMBOL(blk_check_plugged);
3764 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
3766 struct request_queue *q;
3767 unsigned long flags;
3772 flush_plug_callbacks(plug, from_schedule);
3774 if (!list_empty(&plug->mq_list))
3775 blk_mq_flush_plug_list(plug, from_schedule);
3777 if (list_empty(&plug->list))
3780 list_splice_init(&plug->list, &list);
3782 list_sort(NULL, &list, plug_rq_cmp);
3788 * Save and disable interrupts here, to avoid doing it for every
3789 * queue lock we have to take.
3791 local_irq_save(flags);
3792 while (!list_empty(&list)) {
3793 rq = list_entry_rq(list.next);
3794 list_del_init(&rq->queuelist);
3798 * This drops the queue lock
3801 queue_unplugged(q, depth, from_schedule);
3804 spin_lock(q->queue_lock);
3808 * Short-circuit if @q is dead
3810 if (unlikely(blk_queue_dying(q))) {
3811 __blk_end_request_all(rq, -ENODEV);
3816 * rq is already accounted, so use raw insert
3818 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
3819 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3821 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
3827 * This drops the queue lock
3830 queue_unplugged(q, depth, from_schedule);
3832 local_irq_restore(flags);
3835 void blk_finish_plug(struct blk_plug *plug)
3837 if (plug != current->plug)
3839 blk_flush_plug_list(plug, false);
3841 current->plug = NULL;
3843 EXPORT_SYMBOL(blk_finish_plug);
3845 bool blk_poll(struct request_queue *q, blk_qc_t cookie)
3847 struct blk_plug *plug;
3850 if (!q->mq_ops || !q->mq_ops->poll || !blk_qc_t_valid(cookie) ||
3851 !test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
3854 plug = current->plug;
3856 blk_flush_plug_list(plug, false);
3858 state = current->state;
3859 while (!need_resched()) {
3860 unsigned int queue_num = blk_qc_t_to_queue_num(cookie);
3861 struct blk_mq_hw_ctx *hctx = q->queue_hw_ctx[queue_num];
3864 hctx->poll_invoked++;
3866 ret = q->mq_ops->poll(hctx, blk_qc_t_to_tag(cookie));
3868 hctx->poll_success++;
3869 set_current_state(TASK_RUNNING);
3873 if (signal_pending_state(state, current))
3874 set_current_state(TASK_RUNNING);
3876 if (current->state == TASK_RUNNING)
3888 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3889 * @q: the queue of the device
3890 * @dev: the device the queue belongs to
3893 * Initialize runtime-PM-related fields for @q and start auto suspend for
3894 * @dev. Drivers that want to take advantage of request-based runtime PM
3895 * should call this function after @dev has been initialized, and its
3896 * request queue @q has been allocated, and runtime PM for it can not happen
3897 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3898 * cases, driver should call this function before any I/O has taken place.
3900 * This function takes care of setting up using auto suspend for the device,
3901 * the autosuspend delay is set to -1 to make runtime suspend impossible
3902 * until an updated value is either set by user or by driver. Drivers do
3903 * not need to touch other autosuspend settings.
3905 * The block layer runtime PM is request based, so only works for drivers
3906 * that use request as their IO unit instead of those directly use bio's.
3908 void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3911 q->rpm_status = RPM_ACTIVE;
3912 pm_runtime_set_autosuspend_delay(q->dev, -1);
3913 pm_runtime_use_autosuspend(q->dev);
3915 EXPORT_SYMBOL(blk_pm_runtime_init);
3918 * blk_pre_runtime_suspend - Pre runtime suspend check
3919 * @q: the queue of the device
3922 * This function will check if runtime suspend is allowed for the device
3923 * by examining if there are any requests pending in the queue. If there
3924 * are requests pending, the device can not be runtime suspended; otherwise,
3925 * the queue's status will be updated to SUSPENDING and the driver can
3926 * proceed to suspend the device.
3928 * For the not allowed case, we mark last busy for the device so that
3929 * runtime PM core will try to autosuspend it some time later.
3931 * This function should be called near the start of the device's
3932 * runtime_suspend callback.
3935 * 0 - OK to runtime suspend the device
3936 * -EBUSY - Device should not be runtime suspended
3938 int blk_pre_runtime_suspend(struct request_queue *q)
3945 spin_lock_irq(q->queue_lock);
3946 if (q->nr_pending) {
3948 pm_runtime_mark_last_busy(q->dev);
3950 q->rpm_status = RPM_SUSPENDING;
3952 spin_unlock_irq(q->queue_lock);
3955 EXPORT_SYMBOL(blk_pre_runtime_suspend);
3958 * blk_post_runtime_suspend - Post runtime suspend processing
3959 * @q: the queue of the device
3960 * @err: return value of the device's runtime_suspend function
3963 * Update the queue's runtime status according to the return value of the
3964 * device's runtime suspend function and mark last busy for the device so
3965 * that PM core will try to auto suspend the device at a later time.
3967 * This function should be called near the end of the device's
3968 * runtime_suspend callback.
3970 void blk_post_runtime_suspend(struct request_queue *q, int err)
3975 spin_lock_irq(q->queue_lock);
3977 q->rpm_status = RPM_SUSPENDED;
3979 q->rpm_status = RPM_ACTIVE;
3980 pm_runtime_mark_last_busy(q->dev);
3982 spin_unlock_irq(q->queue_lock);
3984 EXPORT_SYMBOL(blk_post_runtime_suspend);
3987 * blk_pre_runtime_resume - Pre runtime resume processing
3988 * @q: the queue of the device
3991 * Update the queue's runtime status to RESUMING in preparation for the
3992 * runtime resume of the device.
3994 * This function should be called near the start of the device's
3995 * runtime_resume callback.
3997 void blk_pre_runtime_resume(struct request_queue *q)
4002 spin_lock_irq(q->queue_lock);
4003 q->rpm_status = RPM_RESUMING;
4004 spin_unlock_irq(q->queue_lock);
4006 EXPORT_SYMBOL(blk_pre_runtime_resume);
4009 * blk_post_runtime_resume - Post runtime resume processing
4010 * @q: the queue of the device
4011 * @err: return value of the device's runtime_resume function
4014 * Update the queue's runtime status according to the return value of the
4015 * device's runtime_resume function. If it is successfully resumed, process
4016 * the requests that are queued into the device's queue when it is resuming
4017 * and then mark last busy and initiate autosuspend for it.
4019 * This function should be called near the end of the device's
4020 * runtime_resume callback.
4022 void blk_post_runtime_resume(struct request_queue *q, int err)
4027 spin_lock_irq(q->queue_lock);
4029 q->rpm_status = RPM_ACTIVE;
4031 pm_runtime_mark_last_busy(q->dev);
4032 pm_request_autosuspend(q->dev);
4034 q->rpm_status = RPM_SUSPENDED;
4036 spin_unlock_irq(q->queue_lock);
4038 EXPORT_SYMBOL(blk_post_runtime_resume);
4041 int __init blk_dev_init(void)
4043 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
4044 FIELD_SIZEOF(struct request, cmd_flags));
4046 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
4047 kblockd_workqueue = alloc_workqueue("kblockd",
4048 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
4049 if (!kblockd_workqueue)
4050 panic("Failed to create kblockd\n");
4052 request_cachep = kmem_cache_create("blkdev_requests",
4053 sizeof(struct request), 0, SLAB_PANIC, NULL);
4055 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
4056 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
4062 * Blk IO latency support. We want this to be as cheap as possible, so doing
4063 * this lockless (and avoiding atomics), a few off by a few errors in this
4064 * code is not harmful, and we don't want to do anything that is
4066 * TODO : If necessary, we can make the histograms per-cpu and aggregate
4067 * them when printing them out.
4070 blk_latency_hist_show(char* name, struct io_latency_state *s, char *buf,
4074 int bytes_written = 0;
4075 u_int64_t num_elem, elem;
4079 num_elem = s->latency_elems;
4081 average = div64_u64(s->latency_sum, s->latency_elems);
4082 bytes_written += scnprintf(buf + bytes_written,
4083 buf_size - bytes_written,
4084 "IO svc_time %s Latency Histogram (n = %llu,"
4085 " average = %llu):\n", name, num_elem, average);
4087 i < ARRAY_SIZE(latency_x_axis_us);
4089 elem = s->latency_y_axis[i];
4090 pct = div64_u64(elem * 100, num_elem);
4091 bytes_written += scnprintf(buf + bytes_written,
4092 PAGE_SIZE - bytes_written,
4093 "\t< %6lluus%15llu%15d%%\n",
4094 latency_x_axis_us[i],
4097 /* Last element in y-axis table is overflow */
4098 elem = s->latency_y_axis[i];
4099 pct = div64_u64(elem * 100, num_elem);
4100 bytes_written += scnprintf(buf + bytes_written,
4101 PAGE_SIZE - bytes_written,
4102 "\t>=%6lluus%15llu%15d%%\n",
4103 latency_x_axis_us[i - 1], elem, pct);
4106 return bytes_written;
4108 EXPORT_SYMBOL(blk_latency_hist_show);