path: root/block/blk-core.c (plain)
blob: c13af6192a355efbd75ca9073f236deec7d9f41c

1	/*
2	* Copyright (C) 1991, 1992 Linus Torvalds
3	* Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4	* Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5	* Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
6	* kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
7	* - July2000
8	* bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
9	*/
10
11	/*
12	* This handles all read/write requests to block devices
13	*/
14	#include <linux/kernel.h>
15	#include <linux/module.h>
16	#include <linux/backing-dev.h>
17	#include <linux/bio.h>
18	#include <linux/blkdev.h>
19	#include <linux/blk-mq.h>
20	#include <linux/highmem.h>
21	#include <linux/mm.h>
22	#include <linux/kernel_stat.h>
23	#include <linux/string.h>
24	#include <linux/init.h>
25	#include <linux/completion.h>
26	#include <linux/slab.h>
27	#include <linux/swap.h>
28	#include <linux/writeback.h>
29	#include <linux/task_io_accounting_ops.h>
30	#include <linux/fault-inject.h>
31	#include <linux/list_sort.h>
32	#include <linux/delay.h>
33	#include <linux/ratelimit.h>
34	#include <linux/pm_runtime.h>
35	#include <linux/blk-cgroup.h>
36
37	#define CREATE_TRACE_POINTS
38	#include <trace/events/block.h>
39
40	#include "blk.h"
41	#include "blk-mq.h"
42
43	#include <linux/math64.h>
44
45	EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
46	EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
47	EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
48	EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
49	EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
50
51	DEFINE_IDA(blk_queue_ida);
52
53	/*
54	* For the allocated request tables
55	*/
56	struct kmem_cache *request_cachep;
57
58	/*
59	* For queue allocation
60	*/
61	struct kmem_cache *blk_requestq_cachep;
62
63	/*
64	* Controlling structure to kblockd
65	*/
66	static struct workqueue_struct *kblockd_workqueue;
67
68	static void blk_clear_congested(struct request_list *rl, int sync)
69	{
70	#ifdef CONFIG_CGROUP_WRITEBACK
71	clear_wb_congested(rl->blkg->wb_congested, sync);
72	#else
73	/*
74	* If !CGROUP_WRITEBACK, all blkg's map to bdi->wb and we shouldn't
75	* flip its congestion state for events on other blkcgs.
76	*/
77	if (rl == &rl->q->root_rl)
78	clear_wb_congested(rl->q->backing_dev_info.wb.congested, sync);
79	#endif
80	}
81
82	static void blk_set_congested(struct request_list *rl, int sync)
83	{
84	#ifdef CONFIG_CGROUP_WRITEBACK
85	set_wb_congested(rl->blkg->wb_congested, sync);
86	#else
87	/* see blk_clear_congested() */
88	if (rl == &rl->q->root_rl)
89	set_wb_congested(rl->q->backing_dev_info.wb.congested, sync);
90	#endif
91	}
92
93	void blk_queue_congestion_threshold(struct request_queue *q)
94	{
95	int nr;
96
97	nr = q->nr_requests - (q->nr_requests / 8) + 1;
98	if (nr > q->nr_requests)
99	nr = q->nr_requests;
100	q->nr_congestion_on = nr;
101
102	nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
103	if (nr < 1)
104	nr = 1;
105	q->nr_congestion_off = nr;
106	}
107
108	/**
109	* blk_get_backing_dev_info - get the address of a queue's backing_dev_info
110	* @bdev: device
111	*
112	* Locates the passed device's request queue and returns the address of its
113	* backing_dev_info. This function can only be called if @bdev is opened
114	* and the return value is never NULL.
115	*/
116	struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
117	{
118	struct request_queue *q = bdev_get_queue(bdev);
119
120	return &q->backing_dev_info;
121	}
122	EXPORT_SYMBOL(blk_get_backing_dev_info);
123
124	void blk_rq_init(struct request_queue *q, struct request *rq)
125	{
126	memset(rq, 0, sizeof(*rq));
127
128	INIT_LIST_HEAD(&rq->queuelist);
129	INIT_LIST_HEAD(&rq->timeout_list);
130	rq->cpu = -1;
131	rq->q = q;
132	rq->__sector = (sector_t) -1;
133	INIT_HLIST_NODE(&rq->hash);
134	RB_CLEAR_NODE(&rq->rb_node);
135	rq->cmd = rq->__cmd;
136	rq->cmd_len = BLK_MAX_CDB;
137	rq->tag = -1;
138	rq->start_time = jiffies;
139	set_start_time_ns(rq);
140	rq->part = NULL;
141	}
142	EXPORT_SYMBOL(blk_rq_init);
143
144	static void req_bio_endio(struct request *rq, struct bio *bio,
145	unsigned int nbytes, int error)
146	{
147	if (error)
148	bio->bi_error = error;
149
150	if (unlikely(rq->cmd_flags & REQ_QUIET))
151	bio_set_flag(bio, BIO_QUIET);
152
153	bio_advance(bio, nbytes);
154
155	/* don't actually finish bio if it's part of flush sequence */
156	if (bio->bi_iter.bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
157	bio_endio(bio);
158	}
159
160	void blk_dump_rq_flags(struct request *rq, char *msg)
161	{
162	int bit;
163
164	printk(KERN_INFO "%s: dev %s: type=%x, flags=%llx\n", msg,
165	rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
166	(unsigned long long) rq->cmd_flags);
167
168	printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
169	(unsigned long long)blk_rq_pos(rq),
170	blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
171	printk(KERN_INFO " bio %p, biotail %p, len %u\n",
172	rq->bio, rq->biotail, blk_rq_bytes(rq));
173
174	if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
175	printk(KERN_INFO " cdb: ");
176	for (bit = 0; bit < BLK_MAX_CDB; bit++)
177	printk("%02x ", rq->cmd[bit]);
178	printk("\n");
179	}
180	}
181	EXPORT_SYMBOL(blk_dump_rq_flags);
182
183	static void blk_delay_work(struct work_struct *work)
184	{
185	struct request_queue *q;
186
187	q = container_of(work, struct request_queue, delay_work.work);
188	spin_lock_irq(q->queue_lock);
189	__blk_run_queue(q);
190	spin_unlock_irq(q->queue_lock);
191	}
192
193	/**
194	* blk_delay_queue - restart queueing after defined interval
195	* @q: The &struct request_queue in question
196	* @msecs: Delay in msecs
197	*
198	* Description:
199	* Sometimes queueing needs to be postponed for a little while, to allow
200	* resources to come back. This function will make sure that queueing is
201	* restarted around the specified time. Queue lock must be held.
202	*/
203	void blk_delay_queue(struct request_queue *q, unsigned long msecs)
204	{
205	if (likely(!blk_queue_dead(q)))
206	queue_delayed_work(kblockd_workqueue, &q->delay_work,
207	msecs_to_jiffies(msecs));
208	}
209	EXPORT_SYMBOL(blk_delay_queue);
210
211	/**
212	* blk_start_queue_async - asynchronously restart a previously stopped queue
213	* @q: The &struct request_queue in question
214	*
215	* Description:
216	* blk_start_queue_async() will clear the stop flag on the queue, and
217	* ensure that the request_fn for the queue is run from an async
218	* context.
219	**/
220	void blk_start_queue_async(struct request_queue *q)
221	{
222	queue_flag_clear(QUEUE_FLAG_STOPPED, q);
223	blk_run_queue_async(q);
224	}
225	EXPORT_SYMBOL(blk_start_queue_async);
226
227	/**
228	* blk_start_queue - restart a previously stopped queue
229	* @q: The &struct request_queue in question
230	*
231	* Description:
232	* blk_start_queue() will clear the stop flag on the queue, and call
233	* the request_fn for the queue if it was in a stopped state when
234	* entered. Also see blk_stop_queue(). Queue lock must be held.
235	**/
236	void blk_start_queue(struct request_queue *q)
237	{
238	WARN_ON(!in_interrupt() && !irqs_disabled());
239
240	queue_flag_clear(QUEUE_FLAG_STOPPED, q);
241	__blk_run_queue(q);
242	}
243	EXPORT_SYMBOL(blk_start_queue);
244
245	/**
246	* blk_stop_queue - stop a queue
247	* @q: The &struct request_queue in question
248	*
249	* Description:
250	* The Linux block layer assumes that a block driver will consume all
251	* entries on the request queue when the request_fn strategy is called.
252	* Often this will not happen, because of hardware limitations (queue
253	* depth settings). If a device driver gets a 'queue full' response,
254	* or if it simply chooses not to queue more I/O at one point, it can
255	* call this function to prevent the request_fn from being called until
256	* the driver has signalled it's ready to go again. This happens by calling
257	* blk_start_queue() to restart queue operations. Queue lock must be held.
258	**/
259	void blk_stop_queue(struct request_queue *q)
260	{
261	cancel_delayed_work(&q->delay_work);
262	queue_flag_set(QUEUE_FLAG_STOPPED, q);
263	}
264	EXPORT_SYMBOL(blk_stop_queue);
265
266	/**
267	* blk_sync_queue - cancel any pending callbacks on a queue
268	* @q: the queue
269	*
270	* Description:
271	* The block layer may perform asynchronous callback activity
272	* on a queue, such as calling the unplug function after a timeout.
273	* A block device may call blk_sync_queue to ensure that any
274	* such activity is cancelled, thus allowing it to release resources
275	* that the callbacks might use. The caller must already have made sure
276	* that its ->make_request_fn will not re-add plugging prior to calling
277	* this function.
278	*
279	* This function does not cancel any asynchronous activity arising
280	* out of elevator or throttling code. That would require elevator_exit()
281	* and blkcg_exit_queue() to be called with queue lock initialized.
282	*
283	*/
284	void blk_sync_queue(struct request_queue *q)
285	{
286	del_timer_sync(&q->timeout);
287	cancel_work_sync(&q->timeout_work);
288
289	if (q->mq_ops) {
290	struct blk_mq_hw_ctx *hctx;
291	int i;
292
293	queue_for_each_hw_ctx(q, hctx, i) {
294	cancel_work_sync(&hctx->run_work);
295	cancel_delayed_work_sync(&hctx->delay_work);
296	}
297	} else {
298	cancel_delayed_work_sync(&q->delay_work);
299	}
300	}
301	EXPORT_SYMBOL(blk_sync_queue);
302
303	/**
304	* __blk_run_queue_uncond - run a queue whether or not it has been stopped
305	* @q: The queue to run
306	*
307	* Description:
308	* Invoke request handling on a queue if there are any pending requests.
309	* May be used to restart request handling after a request has completed.
310	* This variant runs the queue whether or not the queue has been
311	* stopped. Must be called with the queue lock held and interrupts
312	* disabled. See also @blk_run_queue.
313	*/
314	inline void __blk_run_queue_uncond(struct request_queue *q)
315	{
316	if (unlikely(blk_queue_dead(q)))
317	return;
318
319	/*
320	* Some request_fn implementations, e.g. scsi_request_fn(), unlock
321	* the queue lock internally. As a result multiple threads may be
322	* running such a request function concurrently. Keep track of the
323	* number of active request_fn invocations such that blk_drain_queue()
324	* can wait until all these request_fn calls have finished.
325	*/
326	q->request_fn_active++;
327	q->request_fn(q);
328	q->request_fn_active--;
329	}
330	EXPORT_SYMBOL_GPL(__blk_run_queue_uncond);
331
332	/**
333	* __blk_run_queue - run a single device queue
334	* @q: The queue to run
335	*
336	* Description:
337	* See @blk_run_queue. This variant must be called with the queue lock
338	* held and interrupts disabled.
339	*/
340	void __blk_run_queue(struct request_queue *q)
341	{
342	if (unlikely(blk_queue_stopped(q)))
343	return;
344
345	__blk_run_queue_uncond(q);
346	}
347	EXPORT_SYMBOL(__blk_run_queue);
348
349	/**
350	* blk_run_queue_async - run a single device queue in workqueue context
351	* @q: The queue to run
352	*
353	* Description:
354	* Tells kblockd to perform the equivalent of @blk_run_queue on behalf
355	* of us. The caller must hold the queue lock.
356	*/
357	void blk_run_queue_async(struct request_queue *q)
358	{
359	if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
360	mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
361	}
362	EXPORT_SYMBOL(blk_run_queue_async);
363
364	/**
365	* blk_run_queue - run a single device queue
366	* @q: The queue to run
367	*
368	* Description:
369	* Invoke request handling on this queue, if it has pending work to do.
370	* May be used to restart queueing when a request has completed.
371	*/
372	void blk_run_queue(struct request_queue *q)
373	{
374	unsigned long flags;
375
376	spin_lock_irqsave(q->queue_lock, flags);
377	__blk_run_queue(q);
378	spin_unlock_irqrestore(q->queue_lock, flags);
379	}
380	EXPORT_SYMBOL(blk_run_queue);
381
382	void blk_put_queue(struct request_queue *q)
383	{
384	kobject_put(&q->kobj);
385	}
386	EXPORT_SYMBOL(blk_put_queue);
387
388	/**
389	* __blk_drain_queue - drain requests from request_queue
390	* @q: queue to drain
391	* @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
392	*
393	* Drain requests from @q. If @drain_all is set, all requests are drained.
394	* If not, only ELVPRIV requests are drained. The caller is responsible
395	* for ensuring that no new requests which need to be drained are queued.
396	*/
397	static void __blk_drain_queue(struct request_queue *q, bool drain_all)
398	__releases(q->queue_lock)
399	__acquires(q->queue_lock)
400	{
401	int i;
402
403	lockdep_assert_held(q->queue_lock);
404
405	while (true) {
406	bool drain = false;
407
408	/*
409	* The caller might be trying to drain @q before its
410	* elevator is initialized.
411	*/
412	if (q->elevator)
413	elv_drain_elevator(q);
414
415	blkcg_drain_queue(q);
416
417	/*
418	* This function might be called on a queue which failed
419	* driver init after queue creation or is not yet fully
420	* active yet. Some drivers (e.g. fd and loop) get unhappy
421	* in such cases. Kick queue iff dispatch queue has
422	* something on it and @q has request_fn set.
423	*/
424	if (!list_empty(&q->queue_head) && q->request_fn)
425	__blk_run_queue(q);
426
427	drain |= q->nr_rqs_elvpriv;
428	drain |= q->request_fn_active;
429
430	/*
431	* Unfortunately, requests are queued at and tracked from
432	* multiple places and there's no single counter which can
433	* be drained. Check all the queues and counters.
434	*/
435	if (drain_all) {
436	struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
437	drain |= !list_empty(&q->queue_head);
438	for (i = 0; i < 2; i++) {
439	drain |= q->nr_rqs[i];
440	drain |= q->in_flight[i];
441	if (fq)
442	drain |= !list_empty(&fq->flush_queue[i]);
443	}
444	}
445
446	if (!drain)
447	break;
448
449	spin_unlock_irq(q->queue_lock);
450
451	msleep(10);
452
453	spin_lock_irq(q->queue_lock);
454	}
455
456	/*
457	* With queue marked dead, any woken up waiter will fail the
458	* allocation path, so the wakeup chaining is lost and we're
459	* left with hung waiters. We need to wake up those waiters.
460	*/
461	if (q->request_fn) {
462	struct request_list *rl;
463
464	blk_queue_for_each_rl(rl, q)
465	for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
466	wake_up_all(&rl->wait[i]);
467	}
468	}
469
470	/**
471	* blk_queue_bypass_start - enter queue bypass mode
472	* @q: queue of interest
473	*
474	* In bypass mode, only the dispatch FIFO queue of @q is used. This
475	* function makes @q enter bypass mode and drains all requests which were
476	* throttled or issued before. On return, it's guaranteed that no request
477	* is being throttled or has ELVPRIV set and blk_queue_bypass() %true
478	* inside queue or RCU read lock.
479	*/
480	void blk_queue_bypass_start(struct request_queue *q)
481	{
482	spin_lock_irq(q->queue_lock);
483	q->bypass_depth++;
484	queue_flag_set(QUEUE_FLAG_BYPASS, q);
485	spin_unlock_irq(q->queue_lock);
486
487	/*
488	* Queues start drained. Skip actual draining till init is
489	* complete. This avoids lenghty delays during queue init which
490	* can happen many times during boot.
491	*/
492	if (blk_queue_init_done(q)) {
493	spin_lock_irq(q->queue_lock);
494	__blk_drain_queue(q, false);
495	spin_unlock_irq(q->queue_lock);
496
497	/* ensure blk_queue_bypass() is %true inside RCU read lock */
498	synchronize_rcu();
499	}
500	}
501	EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
502
503	/**
504	* blk_queue_bypass_end - leave queue bypass mode
505	* @q: queue of interest
506	*
507	* Leave bypass mode and restore the normal queueing behavior.
508	*/
509	void blk_queue_bypass_end(struct request_queue *q)
510	{
511	spin_lock_irq(q->queue_lock);
512	if (!--q->bypass_depth)
513	queue_flag_clear(QUEUE_FLAG_BYPASS, q);
514	WARN_ON_ONCE(q->bypass_depth < 0);
515	spin_unlock_irq(q->queue_lock);
516	}
517	EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
518
519	void blk_set_queue_dying(struct request_queue *q)
520	{
521	spin_lock_irq(q->queue_lock);
522	queue_flag_set(QUEUE_FLAG_DYING, q);
523	spin_unlock_irq(q->queue_lock);
524
525	if (q->mq_ops)
526	blk_mq_wake_waiters(q);
527	else {
528	struct request_list *rl;
529
530	blk_queue_for_each_rl(rl, q) {
531	if (rl->rq_pool) {
532	wake_up_all(&rl->wait[BLK_RW_SYNC]);
533	wake_up_all(&rl->wait[BLK_RW_ASYNC]);
534	}
535	}
536	}
537	}
538	EXPORT_SYMBOL_GPL(blk_set_queue_dying);
539
540	/**
541	* blk_cleanup_queue - shutdown a request queue
542	* @q: request queue to shutdown
543	*
544	* Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
545	* put it. All future requests will be failed immediately with -ENODEV.
546	*/
547	void blk_cleanup_queue(struct request_queue *q)
548	{
549	spinlock_t *lock = q->queue_lock;
550
551	/* mark @q DYING, no new request or merges will be allowed afterwards */
552	mutex_lock(&q->sysfs_lock);
553	blk_set_queue_dying(q);
554	spin_lock_irq(lock);
555
556	/*
557	* A dying queue is permanently in bypass mode till released. Note
558	* that, unlike blk_queue_bypass_start(), we aren't performing
559	* synchronize_rcu() after entering bypass mode to avoid the delay
560	* as some drivers create and destroy a lot of queues while
561	* probing. This is still safe because blk_release_queue() will be
562	* called only after the queue refcnt drops to zero and nothing,
563	* RCU or not, would be traversing the queue by then.
564	*/
565	q->bypass_depth++;
566	queue_flag_set(QUEUE_FLAG_BYPASS, q);
567
568	queue_flag_set(QUEUE_FLAG_NOMERGES, q);
569	queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
570	queue_flag_set(QUEUE_FLAG_DYING, q);
571	spin_unlock_irq(lock);
572	mutex_unlock(&q->sysfs_lock);
573
574	/*
575	* Drain all requests queued before DYING marking. Set DEAD flag to
576	* prevent that q->request_fn() gets invoked after draining finished.
577	*/
578	blk_freeze_queue(q);
579	spin_lock_irq(lock);
580	if (!q->mq_ops)
581	__blk_drain_queue(q, true);
582	queue_flag_set(QUEUE_FLAG_DEAD, q);
583	spin_unlock_irq(lock);
584
585	/* for synchronous bio-based driver finish in-flight integrity i/o */
586	blk_flush_integrity();
587
588	/* @q won't process any more request, flush async actions */
589	del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
590	blk_sync_queue(q);
591
592	if (q->mq_ops)
593	blk_mq_free_queue(q);
594	percpu_ref_exit(&q->q_usage_counter);
595
596	spin_lock_irq(lock);
597	if (q->queue_lock != &q->__queue_lock)
598	q->queue_lock = &q->__queue_lock;
599	spin_unlock_irq(lock);
600
601	bdi_unregister(&q->backing_dev_info);
602
603	/* @q is and will stay empty, shutdown and put */
604	blk_put_queue(q);
605	}
606	EXPORT_SYMBOL(blk_cleanup_queue);
607
608	/* Allocate memory local to the request queue */
609	static void *alloc_request_struct(gfp_t gfp_mask, void *data)
610	{
611	int nid = (int)(long)data;
612	return kmem_cache_alloc_node(request_cachep, gfp_mask, nid);
613	}
614
615	static void free_request_struct(void *element, void *unused)
616	{
617	kmem_cache_free(request_cachep, element);
618	}
619
620	int blk_init_rl(struct request_list *rl, struct request_queue *q,
621	gfp_t gfp_mask)
622	{
623	if (unlikely(rl->rq_pool))
624	return 0;
625
626	rl->q = q;
627	rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
628	rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
629	init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
630	init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
631
632	rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, alloc_request_struct,
633	free_request_struct,
634	(void *)(long)q->node, gfp_mask,
635	q->node);
636	if (!rl->rq_pool)
637	return -ENOMEM;
638
639	return 0;
640	}
641
642	void blk_exit_rl(struct request_list *rl)
643	{
644	if (rl->rq_pool)
645	mempool_destroy(rl->rq_pool);
646	}
647
648	struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
649	{
650	return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
651	}
652	EXPORT_SYMBOL(blk_alloc_queue);
653
654	int blk_queue_enter(struct request_queue *q, bool nowait)
655	{
656	while (true) {
657
658	if (percpu_ref_tryget_live(&q->q_usage_counter))
659	return 0;
660
661	if (nowait)
662	return -EBUSY;
663
664	wait_event(q->mq_freeze_wq,
665	!atomic_read(&q->mq_freeze_depth) ||
666	blk_queue_dying(q));
667	if (blk_queue_dying(q))
668	return -ENODEV;
669	}
670	}
671
672	void blk_queue_exit(struct request_queue *q)
673	{
674	percpu_ref_put(&q->q_usage_counter);
675	}
676
677	static void blk_queue_usage_counter_release(struct percpu_ref *ref)
678	{
679	struct request_queue *q =
680	container_of(ref, struct request_queue, q_usage_counter);
681
682	wake_up_all(&q->mq_freeze_wq);
683	}
684
685	static void blk_rq_timed_out_timer(unsigned long data)
686	{
687	struct request_queue *q = (struct request_queue *)data;
688
689	kblockd_schedule_work(&q->timeout_work);
690	}
691
692	struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
693	{
694	struct request_queue *q;
695	int err;
696
697	q = kmem_cache_alloc_node(blk_requestq_cachep,
698	gfp_mask | __GFP_ZERO, node_id);
699	if (!q)
700	return NULL;
701
702	q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
703	if (q->id < 0)
704	goto fail_q;
705
706	q->bio_split = bioset_create(BIO_POOL_SIZE, 0);
707	if (!q->bio_split)
708	goto fail_id;
709
710	q->backing_dev_info.ra_pages =
711	(VM_MAX_READAHEAD * 1024) / PAGE_SIZE;
712	q->backing_dev_info.capabilities = BDI_CAP_CGROUP_WRITEBACK;
713	q->backing_dev_info.name = "block";
714	q->node = node_id;
715
716	err = bdi_init(&q->backing_dev_info);
717	if (err)
718	goto fail_split;
719
720	setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
721	laptop_mode_timer_fn, (unsigned long) q);
722	setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
723	INIT_WORK(&q->timeout_work, NULL);
724	INIT_LIST_HEAD(&q->queue_head);
725	INIT_LIST_HEAD(&q->timeout_list);
726	INIT_LIST_HEAD(&q->icq_list);
727	#ifdef CONFIG_BLK_CGROUP
728	INIT_LIST_HEAD(&q->blkg_list);
729	#endif
730	INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
731
732	kobject_init(&q->kobj, &blk_queue_ktype);
733
734	mutex_init(&q->sysfs_lock);
735	spin_lock_init(&q->__queue_lock);
736
737	/*
738	* By default initialize queue_lock to internal lock and driver can
739	* override it later if need be.
740	*/
741	q->queue_lock = &q->__queue_lock;
742
743	/*
744	* A queue starts its life with bypass turned on to avoid
745	* unnecessary bypass on/off overhead and nasty surprises during
746	* init. The initial bypass will be finished when the queue is
747	* registered by blk_register_queue().
748	*/
749	q->bypass_depth = 1;
750	__set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
751
752	init_waitqueue_head(&q->mq_freeze_wq);
753
754	/*
755	* Init percpu_ref in atomic mode so that it's faster to shutdown.
756	* See blk_register_queue() for details.
757	*/
758	if (percpu_ref_init(&q->q_usage_counter,
759	blk_queue_usage_counter_release,
760	PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
761	goto fail_bdi;
762
763	if (blkcg_init_queue(q))
764	goto fail_ref;
765
766	return q;
767
768	fail_ref:
769	percpu_ref_exit(&q->q_usage_counter);
770	fail_bdi:
771	bdi_destroy(&q->backing_dev_info);
772	fail_split:
773	bioset_free(q->bio_split);
774	fail_id:
775	ida_simple_remove(&blk_queue_ida, q->id);
776	fail_q:
777	kmem_cache_free(blk_requestq_cachep, q);
778	return NULL;
779	}
780	EXPORT_SYMBOL(blk_alloc_queue_node);
781
782	/**
783	* blk_init_queue - prepare a request queue for use with a block device
784	* @rfn: The function to be called to process requests that have been
785	* placed on the queue.
786	* @lock: Request queue spin lock
787	*
788	* Description:
789	* If a block device wishes to use the standard request handling procedures,
790	* which sorts requests and coalesces adjacent requests, then it must
791	* call blk_init_queue(). The function @rfn will be called when there
792	* are requests on the queue that need to be processed. If the device
793	* supports plugging, then @rfn may not be called immediately when requests
794	* are available on the queue, but may be called at some time later instead.
795	* Plugged queues are generally unplugged when a buffer belonging to one
796	* of the requests on the queue is needed, or due to memory pressure.
797	*
798	* @rfn is not required, or even expected, to remove all requests off the
799	* queue, but only as many as it can handle at a time. If it does leave
800	* requests on the queue, it is responsible for arranging that the requests
801	* get dealt with eventually.
802	*
803	* The queue spin lock must be held while manipulating the requests on the
804	* request queue; this lock will be taken also from interrupt context, so irq
805	* disabling is needed for it.
806	*
807	* Function returns a pointer to the initialized request queue, or %NULL if
808	* it didn't succeed.
809	*
810	* Note:
811	* blk_init_queue() must be paired with a blk_cleanup_queue() call
812	* when the block device is deactivated (such as at module unload).
813	**/
814
815	struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
816	{
817	return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
818	}
819	EXPORT_SYMBOL(blk_init_queue);
820
821	struct request_queue *
822	blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
823	{
824	struct request_queue *uninit_q, *q;
825
826	uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
827	if (!uninit_q)
828	return NULL;
829
830	q = blk_init_allocated_queue(uninit_q, rfn, lock);
831	if (!q)
832	blk_cleanup_queue(uninit_q);
833
834	return q;
835	}
836	EXPORT_SYMBOL(blk_init_queue_node);
837
838	static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio);
839
840	struct request_queue *
841	blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
842	spinlock_t *lock)
843	{
844	if (!q)
845	return NULL;
846
847	q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, 0);
848	if (!q->fq)
849	return NULL;
850
851	if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
852	goto fail;
853
854	INIT_WORK(&q->timeout_work, blk_timeout_work);
855	q->request_fn = rfn;
856	q->prep_rq_fn = NULL;
857	q->unprep_rq_fn = NULL;
858	q->queue_flags |= QUEUE_FLAG_DEFAULT;
859
860	/* Override internal queue lock with supplied lock pointer */
861	if (lock)
862	q->queue_lock = lock;
863
864	/*
865	* This also sets hw/phys segments, boundary and size
866	*/
867	blk_queue_make_request(q, blk_queue_bio);
868
869	q->sg_reserved_size = INT_MAX;
870
871	/* Protect q->elevator from elevator_change */
872	mutex_lock(&q->sysfs_lock);
873
874	/* init elevator */
875	if (elevator_init(q, NULL)) {
876	mutex_unlock(&q->sysfs_lock);
877	goto fail;
878	}
879
880	mutex_unlock(&q->sysfs_lock);
881
882	return q;
883
884	fail:
885	blk_free_flush_queue(q->fq);
886	return NULL;
887	}
888	EXPORT_SYMBOL(blk_init_allocated_queue);
889
890	bool blk_get_queue(struct request_queue *q)
891	{
892	if (likely(!blk_queue_dying(q))) {
893	__blk_get_queue(q);
894	return true;
895	}
896
897	return false;
898	}
899	EXPORT_SYMBOL(blk_get_queue);
900
901	static inline void blk_free_request(struct request_list *rl, struct request *rq)
902	{
903	if (rq->cmd_flags & REQ_ELVPRIV) {
904	elv_put_request(rl->q, rq);
905	if (rq->elv.icq)
906	put_io_context(rq->elv.icq->ioc);
907	}
908
909	mempool_free(rq, rl->rq_pool);
910	}
911
912	/*
913	* ioc_batching returns true if the ioc is a valid batching request and
914	* should be given priority access to a request.
915	*/
916	static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
917	{
918	if (!ioc)
919	return 0;
920
921	/*
922	* Make sure the process is able to allocate at least 1 request
923	* even if the batch times out, otherwise we could theoretically
924	* lose wakeups.
925	*/
926	return ioc->nr_batch_requests == q->nr_batching ||
927	(ioc->nr_batch_requests > 0
928	&& time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
929	}
930
931	/*
932	* ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
933	* will cause the process to be a "batcher" on all queues in the system. This
934	* is the behaviour we want though - once it gets a wakeup it should be given
935	* a nice run.
936	*/
937	static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
938	{
939	if (!ioc || ioc_batching(q, ioc))
940	return;
941
942	ioc->nr_batch_requests = q->nr_batching;
943	ioc->last_waited = jiffies;
944	}
945
946	static void __freed_request(struct request_list *rl, int sync)
947	{
948	struct request_queue *q = rl->q;
949
950	if (rl->count[sync] < queue_congestion_off_threshold(q))
951	blk_clear_congested(rl, sync);
952
953	if (rl->count[sync] + 1 <= q->nr_requests) {
954	if (waitqueue_active(&rl->wait[sync]))
955	wake_up(&rl->wait[sync]);
956
957	blk_clear_rl_full(rl, sync);
958	}
959	}
960
961	/*
962	* A request has just been released. Account for it, update the full and
963	* congestion status, wake up any waiters. Called under q->queue_lock.
964	*/
965	static void freed_request(struct request_list *rl, int op, unsigned int flags)
966	{
967	struct request_queue *q = rl->q;
968	int sync = rw_is_sync(op, flags);
969
970	q->nr_rqs[sync]--;
971	rl->count[sync]--;
972	if (flags & REQ_ELVPRIV)
973	q->nr_rqs_elvpriv--;
974
975	__freed_request(rl, sync);
976
977	if (unlikely(rl->starved[sync ^ 1]))
978	__freed_request(rl, sync ^ 1);
979	}
980
981	int blk_update_nr_requests(struct request_queue *q, unsigned int nr)
982	{
983	struct request_list *rl;
984	int on_thresh, off_thresh;
985
986	spin_lock_irq(q->queue_lock);
987	q->nr_requests = nr;
988	blk_queue_congestion_threshold(q);
989	on_thresh = queue_congestion_on_threshold(q);
990	off_thresh = queue_congestion_off_threshold(q);
991
992	blk_queue_for_each_rl(rl, q) {
993	if (rl->count[BLK_RW_SYNC] >= on_thresh)
994	blk_set_congested(rl, BLK_RW_SYNC);
995	else if (rl->count[BLK_RW_SYNC] < off_thresh)
996	blk_clear_congested(rl, BLK_RW_SYNC);
997
998	if (rl->count[BLK_RW_ASYNC] >= on_thresh)
999	blk_set_congested(rl, BLK_RW_ASYNC);
1000	else if (rl->count[BLK_RW_ASYNC] < off_thresh)
1001	blk_clear_congested(rl, BLK_RW_ASYNC);
1002
1003	if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
1004	blk_set_rl_full(rl, BLK_RW_SYNC);
1005	} else {
1006	blk_clear_rl_full(rl, BLK_RW_SYNC);
1007	wake_up(&rl->wait[BLK_RW_SYNC]);
1008	}
1009
1010	if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
1011	blk_set_rl_full(rl, BLK_RW_ASYNC);
1012	} else {
1013	blk_clear_rl_full(rl, BLK_RW_ASYNC);
1014	wake_up(&rl->wait[BLK_RW_ASYNC]);
1015	}
1016	}
1017
1018	spin_unlock_irq(q->queue_lock);
1019	return 0;
1020	}
1021
1022	/*
1023	* Determine if elevator data should be initialized when allocating the
1024	* request associated with @bio.
1025	*/
1026	static bool blk_rq_should_init_elevator(struct bio *bio)
1027	{
1028	if (!bio)
1029	return true;
1030
1031	/*
1032	* Flush requests do not use the elevator so skip initialization.
1033	* This allows a request to share the flush and elevator data.
1034	*/
1035	if (bio->bi_opf & (REQ_PREFLUSH | REQ_FUA))
1036	return false;
1037
1038	return true;
1039	}
1040
1041	/**
1042	* rq_ioc - determine io_context for request allocation
1043	* @bio: request being allocated is for this bio (can be %NULL)
1044	*
1045	* Determine io_context to use for request allocation for @bio. May return
1046	* %NULL if %current->io_context doesn't exist.
1047	*/
1048	static struct io_context *rq_ioc(struct bio *bio)
1049	{
1050	#ifdef CONFIG_BLK_CGROUP
1051	if (bio && bio->bi_ioc)
1052	return bio->bi_ioc;
1053	#endif
1054	return current->io_context;
1055	}
1056
1057	/**
1058	* __get_request - get a free request
1059	* @rl: request list to allocate from
1060	* @op: REQ_OP_READ/REQ_OP_WRITE
1061	* @op_flags: rq_flag_bits
1062	* @bio: bio to allocate request for (can be %NULL)
1063	* @gfp_mask: allocation mask
1064	*
1065	* Get a free request from @q. This function may fail under memory
1066	* pressure or if @q is dead.
1067	*
1068	* Must be called with @q->queue_lock held and,
1069	* Returns ERR_PTR on failure, with @q->queue_lock held.
1070	* Returns request pointer on success, with @q->queue_lock *not held*.
1071	*/
1072	static struct request *__get_request(struct request_list *rl, int op,
1073	int op_flags, struct bio *bio,
1074	gfp_t gfp_mask)
1075	{
1076	struct request_queue *q = rl->q;
1077	struct request *rq;
1078	struct elevator_type *et = q->elevator->type;
1079	struct io_context *ioc = rq_ioc(bio);
1080	struct io_cq *icq = NULL;
1081	const bool is_sync = rw_is_sync(op, op_flags) != 0;
1082	int may_queue;
1083
1084	if (unlikely(blk_queue_dying(q)))
1085	return ERR_PTR(-ENODEV);
1086
1087	may_queue = elv_may_queue(q, op, op_flags);
1088	if (may_queue == ELV_MQUEUE_NO)
1089	goto rq_starved;
1090
1091	if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
1092	if (rl->count[is_sync]+1 >= q->nr_requests) {
1093	/*
1094	* The queue will fill after this allocation, so set
1095	* it as full, and mark this process as "batching".
1096	* This process will be allowed to complete a batch of
1097	* requests, others will be blocked.
1098	*/
1099	if (!blk_rl_full(rl, is_sync)) {
1100	ioc_set_batching(q, ioc);
1101	blk_set_rl_full(rl, is_sync);
1102	} else {
1103	if (may_queue != ELV_MQUEUE_MUST
1104	&& !ioc_batching(q, ioc)) {
1105	/*
1106	* The queue is full and the allocating
1107	* process is not a "batcher", and not
1108	* exempted by the IO scheduler
1109	*/
1110	return ERR_PTR(-ENOMEM);
1111	}
1112	}
1113	}
1114	blk_set_congested(rl, is_sync);
1115	}
1116
1117	/*
1118	* Only allow batching queuers to allocate up to 50% over the defined
1119	* limit of requests, otherwise we could have thousands of requests
1120	* allocated with any setting of ->nr_requests
1121	*/
1122	if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
1123	return ERR_PTR(-ENOMEM);
1124
1125	q->nr_rqs[is_sync]++;
1126	rl->count[is_sync]++;
1127	rl->starved[is_sync] = 0;
1128
1129	/*
1130	* Decide whether the new request will be managed by elevator. If
1131	* so, mark @op_flags and increment elvpriv. Non-zero elvpriv will
1132	* prevent the current elevator from being destroyed until the new
1133	* request is freed. This guarantees icq's won't be destroyed and
1134	* makes creating new ones safe.
1135	*
1136	* Also, lookup icq while holding queue_lock. If it doesn't exist,
1137	* it will be created after releasing queue_lock.
1138	*/
1139	if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
1140	op_flags |= REQ_ELVPRIV;
1141	q->nr_rqs_elvpriv++;
1142	if (et->icq_cache && ioc)
1143	icq = ioc_lookup_icq(ioc, q);
1144	}
1145
1146	if (blk_queue_io_stat(q))
1147	op_flags |= REQ_IO_STAT;
1148	spin_unlock_irq(q->queue_lock);
1149
1150	/* allocate and init request */
1151	rq = mempool_alloc(rl->rq_pool, gfp_mask);
1152	if (!rq)
1153	goto fail_alloc;
1154
1155	blk_rq_init(q, rq);
1156	blk_rq_set_rl(rq, rl);
1157	req_set_op_attrs(rq, op, op_flags | REQ_ALLOCED);
1158
1159	/* init elvpriv */
1160	if (op_flags & REQ_ELVPRIV) {
1161	if (unlikely(et->icq_cache && !icq)) {
1162	if (ioc)
1163	icq = ioc_create_icq(ioc, q, gfp_mask);
1164	if (!icq)
1165	goto fail_elvpriv;
1166	}
1167
1168	rq->elv.icq = icq;
1169	if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1170	goto fail_elvpriv;
1171
1172	/* @rq->elv.icq holds io_context until @rq is freed */
1173	if (icq)
1174	get_io_context(icq->ioc);
1175	}
1176	out:
1177	/*
1178	* ioc may be NULL here, and ioc_batching will be false. That's
1179	* OK, if the queue is under the request limit then requests need
1180	* not count toward the nr_batch_requests limit. There will always
1181	* be some limit enforced by BLK_BATCH_TIME.
1182	*/
1183	if (ioc_batching(q, ioc))
1184	ioc->nr_batch_requests--;
1185
1186	trace_block_getrq(q, bio, op);
1187	return rq;
1188
1189	fail_elvpriv:
1190	/*
1191	* elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1192	* and may fail indefinitely under memory pressure and thus
1193	* shouldn't stall IO. Treat this request as !elvpriv. This will
1194	* disturb iosched and blkcg but weird is bettern than dead.
1195	*/
1196	printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1197	__func__, dev_name(q->backing_dev_info.dev));
1198
1199	rq->cmd_flags &= ~REQ_ELVPRIV;
1200	rq->elv.icq = NULL;
1201
1202	spin_lock_irq(q->queue_lock);
1203	q->nr_rqs_elvpriv--;
1204	spin_unlock_irq(q->queue_lock);
1205	goto out;
1206
1207	fail_alloc:
1208	/*
1209	* Allocation failed presumably due to memory. Undo anything we
1210	* might have messed up.
1211	*
1212	* Allocating task should really be put onto the front of the wait
1213	* queue, but this is pretty rare.
1214	*/
1215	spin_lock_irq(q->queue_lock);
1216	freed_request(rl, op, op_flags);
1217
1218	/*
1219	* in the very unlikely event that allocation failed and no
1220	* requests for this direction was pending, mark us starved so that
1221	* freeing of a request in the other direction will notice
1222	* us. another possible fix would be to split the rq mempool into
1223	* READ and WRITE
1224	*/
1225	rq_starved:
1226	if (unlikely(rl->count[is_sync] == 0))
1227	rl->starved[is_sync] = 1;
1228	return ERR_PTR(-ENOMEM);
1229	}
1230
1231	/**
1232	* get_request - get a free request
1233	* @q: request_queue to allocate request from
1234	* @op: REQ_OP_READ/REQ_OP_WRITE
1235	* @op_flags: rq_flag_bits
1236	* @bio: bio to allocate request for (can be %NULL)
1237	* @gfp_mask: allocation mask
1238	*
1239	* Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask,
1240	* this function keeps retrying under memory pressure and fails iff @q is dead.
1241	*
1242	* Must be called with @q->queue_lock held and,
1243	* Returns ERR_PTR on failure, with @q->queue_lock held.
1244	* Returns request pointer on success, with @q->queue_lock *not held*.
1245	*/
1246	static struct request *get_request(struct request_queue *q, int op,
1247	int op_flags, struct bio *bio,
1248	gfp_t gfp_mask)
1249	{
1250	const bool is_sync = rw_is_sync(op, op_flags) != 0;
1251	DEFINE_WAIT(wait);
1252	struct request_list *rl;
1253	struct request *rq;
1254
1255	rl = blk_get_rl(q, bio); /* transferred to @rq on success */
1256	retry:
1257	rq = __get_request(rl, op, op_flags, bio, gfp_mask);
1258	if (!IS_ERR(rq))
1259	return rq;
1260
1261	if (!gfpflags_allow_blocking(gfp_mask) || unlikely(blk_queue_dying(q))) {
1262	blk_put_rl(rl);
1263	return rq;
1264	}
1265
1266	/* wait on @rl and retry */
1267	prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1268	TASK_UNINTERRUPTIBLE);
1269
1270	trace_block_sleeprq(q, bio, op);
1271
1272	spin_unlock_irq(q->queue_lock);
1273	io_schedule();
1274
1275	/*
1276	* After sleeping, we become a "batching" process and will be able
1277	* to allocate at least one request, and up to a big batch of them
1278	* for a small period time. See ioc_batching, ioc_set_batching
1279	*/
1280	ioc_set_batching(q, current->io_context);
1281
1282	spin_lock_irq(q->queue_lock);
1283	finish_wait(&rl->wait[is_sync], &wait);
1284
1285	goto retry;
1286	}
1287
1288	static struct request *blk_old_get_request(struct request_queue *q, int rw,
1289	gfp_t gfp_mask)
1290	{
1291	struct request *rq;
1292
1293	BUG_ON(rw != READ && rw != WRITE);
1294
1295	/* create ioc upfront */
1296	create_io_context(gfp_mask, q->node);
1297
1298	spin_lock_irq(q->queue_lock);
1299	rq = get_request(q, rw, 0, NULL, gfp_mask);
1300	if (IS_ERR(rq)) {
1301	spin_unlock_irq(q->queue_lock);
1302	return rq;
1303	}
1304
1305	/* q->queue_lock is unlocked at this point */
1306	rq->__data_len = 0;
1307	rq->__sector = (sector_t) -1;
1308	rq->bio = rq->biotail = NULL;
1309	return rq;
1310	}
1311
1312	struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1313	{
1314	if (q->mq_ops)
1315	return blk_mq_alloc_request(q, rw,
1316	(gfp_mask & __GFP_DIRECT_RECLAIM) ?
1317	0 : BLK_MQ_REQ_NOWAIT);
1318	else
1319	return blk_old_get_request(q, rw, gfp_mask);
1320	}
1321	EXPORT_SYMBOL(blk_get_request);
1322
1323	/**
1324	* blk_rq_set_block_pc - initialize a request to type BLOCK_PC
1325	* @rq: request to be initialized
1326	*
1327	*/
1328	void blk_rq_set_block_pc(struct request *rq)
1329	{
1330	rq->cmd_type = REQ_TYPE_BLOCK_PC;
1331	memset(rq->__cmd, 0, sizeof(rq->__cmd));
1332	}
1333	EXPORT_SYMBOL(blk_rq_set_block_pc);
1334
1335	/**
1336	* blk_requeue_request - put a request back on queue
1337	* @q: request queue where request should be inserted
1338	* @rq: request to be inserted
1339	*
1340	* Description:
1341	* Drivers often keep queueing requests until the hardware cannot accept
1342	* more, when that condition happens we need to put the request back
1343	* on the queue. Must be called with queue lock held.
1344	*/
1345	void blk_requeue_request(struct request_queue *q, struct request *rq)
1346	{
1347	blk_delete_timer(rq);
1348	blk_clear_rq_complete(rq);
1349	trace_block_rq_requeue(q, rq);
1350
1351	if (rq->cmd_flags & REQ_QUEUED)
1352	blk_queue_end_tag(q, rq);
1353
1354	BUG_ON(blk_queued_rq(rq));
1355
1356	elv_requeue_request(q, rq);
1357	}
1358	EXPORT_SYMBOL(blk_requeue_request);
1359
1360	static void add_acct_request(struct request_queue *q, struct request *rq,
1361	int where)
1362	{
1363	blk_account_io_start(rq, true);
1364	__elv_add_request(q, rq, where);
1365	}
1366
1367	static void part_round_stats_single(int cpu, struct hd_struct *part,
1368	unsigned long now)
1369	{
1370	int inflight;
1371
1372	if (now == part->stamp)
1373	return;
1374
1375	inflight = part_in_flight(part);
1376	if (inflight) {
1377	__part_stat_add(cpu, part, time_in_queue,
1378	inflight * (now - part->stamp));
1379	__part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1380	}
1381	part->stamp = now;
1382	}
1383
1384	/**
1385	* part_round_stats() - Round off the performance stats on a struct disk_stats.
1386	* @cpu: cpu number for stats access
1387	* @part: target partition
1388	*
1389	* The average IO queue length and utilisation statistics are maintained
1390	* by observing the current state of the queue length and the amount of
1391	* time it has been in this state for.
1392	*
1393	* Normally, that accounting is done on IO completion, but that can result
1394	* in more than a second's worth of IO being accounted for within any one
1395	* second, leading to >100% utilisation. To deal with that, we call this
1396	* function to do a round-off before returning the results when reading
1397	* /proc/diskstats. This accounts immediately for all queue usage up to
1398	* the current jiffies and restarts the counters again.
1399	*/
1400	void part_round_stats(int cpu, struct hd_struct *part)
1401	{
1402	unsigned long now = jiffies;
1403
1404	if (part->partno)
1405	part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1406	part_round_stats_single(cpu, part, now);
1407	}
1408	EXPORT_SYMBOL_GPL(part_round_stats);
1409
1410	#ifdef CONFIG_PM
1411	static void blk_pm_put_request(struct request *rq)
1412	{
1413	if (rq->q->dev && !(rq->cmd_flags & REQ_PM) && !--rq->q->nr_pending)
1414	pm_runtime_mark_last_busy(rq->q->dev);
1415	}
1416	#else
1417	static inline void blk_pm_put_request(struct request *rq) {}
1418	#endif
1419
1420	/*
1421	* queue lock must be held
1422	*/
1423	void __blk_put_request(struct request_queue *q, struct request *req)
1424	{
1425	if (unlikely(!q))
1426	return;
1427
1428	if (q->mq_ops) {
1429	blk_mq_free_request(req);
1430	return;
1431	}
1432
1433	blk_pm_put_request(req);
1434
1435	elv_completed_request(q, req);
1436
1437	/* this is a bio leak */
1438	WARN_ON(req->bio != NULL);
1439
1440	/*
1441	* Request may not have originated from ll_rw_blk. if not,
1442	* it didn't come out of our reserved rq pools
1443	*/
1444	if (req->cmd_flags & REQ_ALLOCED) {
1445	unsigned int flags = req->cmd_flags;
1446	int op = req_op(req);
1447	struct request_list *rl = blk_rq_rl(req);
1448
1449	BUG_ON(!list_empty(&req->queuelist));
1450	BUG_ON(ELV_ON_HASH(req));
1451
1452	blk_free_request(rl, req);
1453	freed_request(rl, op, flags);
1454	blk_put_rl(rl);
1455	}
1456	}
1457	EXPORT_SYMBOL_GPL(__blk_put_request);
1458
1459	void blk_put_request(struct request *req)
1460	{
1461	struct request_queue *q = req->q;
1462
1463	if (q->mq_ops)
1464	blk_mq_free_request(req);
1465	else {
1466	unsigned long flags;
1467
1468	spin_lock_irqsave(q->queue_lock, flags);
1469	__blk_put_request(q, req);
1470	spin_unlock_irqrestore(q->queue_lock, flags);
1471	}
1472	}
1473	EXPORT_SYMBOL(blk_put_request);
1474
1475	/**
1476	* blk_add_request_payload - add a payload to a request
1477	* @rq: request to update
1478	* @page: page backing the payload
1479	* @offset: offset in page
1480	* @len: length of the payload.
1481	*
1482	* This allows to later add a payload to an already submitted request by
1483	* a block driver. The driver needs to take care of freeing the payload
1484	* itself.
1485	*
1486	* Note that this is a quite horrible hack and nothing but handling of
1487	* discard requests should ever use it.
1488	*/
1489	void blk_add_request_payload(struct request *rq, struct page *page,
1490	int offset, unsigned int len)
1491	{
1492	struct bio *bio = rq->bio;
1493
1494	bio->bi_io_vec->bv_page = page;
1495	bio->bi_io_vec->bv_offset = offset;
1496	bio->bi_io_vec->bv_len = len;
1497
1498	bio->bi_iter.bi_size = len;
1499	bio->bi_vcnt = 1;
1500	bio->bi_phys_segments = 1;
1501
1502	rq->__data_len = rq->resid_len = len;
1503	rq->nr_phys_segments = 1;
1504	}
1505	EXPORT_SYMBOL_GPL(blk_add_request_payload);
1506
1507	bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1508	struct bio *bio)
1509	{
1510	const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
1511
1512	if (!ll_back_merge_fn(q, req, bio))
1513	return false;
1514
1515	trace_block_bio_backmerge(q, req, bio);
1516
1517	if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1518	blk_rq_set_mixed_merge(req);
1519
1520	req->biotail->bi_next = bio;
1521	req->biotail = bio;
1522	req->__data_len += bio->bi_iter.bi_size;
1523	req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1524
1525	blk_account_io_start(req, false);
1526	return true;
1527	}
1528
1529	bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1530	struct bio *bio)
1531	{
1532	const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
1533
1534	if (!ll_front_merge_fn(q, req, bio))
1535	return false;
1536
1537	trace_block_bio_frontmerge(q, req, bio);
1538
1539	if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1540	blk_rq_set_mixed_merge(req);
1541
1542	bio->bi_next = req->bio;
1543	req->bio = bio;
1544
1545	req->__sector = bio->bi_iter.bi_sector;
1546	req->__data_len += bio->bi_iter.bi_size;
1547	req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1548
1549	blk_account_io_start(req, false);
1550	return true;
1551	}
1552
1553	/**
1554	* blk_attempt_plug_merge - try to merge with %current's plugged list
1555	* @q: request_queue new bio is being queued at
1556	* @bio: new bio being queued
1557	* @request_count: out parameter for number of traversed plugged requests
1558	* @same_queue_rq: pointer to &struct request that gets filled in when
1559	* another request associated with @q is found on the plug list
1560	* (optional, may be %NULL)
1561	*
1562	* Determine whether @bio being queued on @q can be merged with a request
1563	* on %current's plugged list. Returns %true if merge was successful,
1564	* otherwise %false.
1565	*
1566	* Plugging coalesces IOs from the same issuer for the same purpose without
1567	* going through @q->queue_lock. As such it's more of an issuing mechanism
1568	* than scheduling, and the request, while may have elvpriv data, is not
1569	* added on the elevator at this point. In addition, we don't have
1570	* reliable access to the elevator outside queue lock. Only check basic
1571	* merging parameters without querying the elevator.
1572	*
1573	* Caller must ensure !blk_queue_nomerges(q) beforehand.
1574	*/
1575	bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1576	unsigned int *request_count,
1577	struct request **same_queue_rq)
1578	{
1579	struct blk_plug *plug;
1580	struct request *rq;
1581	bool ret = false;
1582	struct list_head *plug_list;
1583
1584	plug = current->plug;
1585	if (!plug)
1586	goto out;
1587	*request_count = 0;
1588
1589	if (q->mq_ops)
1590	plug_list = &plug->mq_list;
1591	else
1592	plug_list = &plug->list;
1593
1594	list_for_each_entry_reverse(rq, plug_list, queuelist) {
1595	int el_ret;
1596
1597	if (rq->q == q) {
1598	(*request_count)++;
1599	/*
1600	* Only blk-mq multiple hardware queues case checks the
1601	* rq in the same queue, there should be only one such
1602	* rq in a queue
1603	**/
1604	if (same_queue_rq)
1605	*same_queue_rq = rq;
1606	}
1607
1608	if (rq->q != q || !blk_rq_merge_ok(rq, bio))
1609	continue;
1610
1611	el_ret = blk_try_merge(rq, bio);
1612	if (el_ret == ELEVATOR_BACK_MERGE) {
1613	ret = bio_attempt_back_merge(q, rq, bio);
1614	if (ret)
1615	break;
1616	} else if (el_ret == ELEVATOR_FRONT_MERGE) {
1617	ret = bio_attempt_front_merge(q, rq, bio);
1618	if (ret)
1619	break;
1620	}
1621	}
1622	out:
1623	return ret;
1624	}
1625
1626	unsigned int blk_plug_queued_count(struct request_queue *q)
1627	{
1628	struct blk_plug *plug;
1629	struct request *rq;
1630	struct list_head *plug_list;
1631	unsigned int ret = 0;
1632
1633	plug = current->plug;
1634	if (!plug)
1635	goto out;
1636
1637	if (q->mq_ops)
1638	plug_list = &plug->mq_list;
1639	else
1640	plug_list = &plug->list;
1641
1642	list_for_each_entry(rq, plug_list, queuelist) {
1643	if (rq->q == q)
1644	ret++;
1645	}
1646	out:
1647	return ret;
1648	}
1649
1650	void init_request_from_bio(struct request *req, struct bio *bio)
1651	{
1652	req->cmd_type = REQ_TYPE_FS;
1653
1654	req->cmd_flags |= bio->bi_opf & REQ_COMMON_MASK;
1655	if (bio->bi_opf & REQ_RAHEAD)
1656	req->cmd_flags |= REQ_FAILFAST_MASK;
1657
1658	req->errors = 0;
1659	req->__sector = bio->bi_iter.bi_sector;
1660	req->ioprio = bio_prio(bio);
1661	blk_rq_bio_prep(req->q, req, bio);
1662	}
1663
1664	static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio)
1665	{
1666	const bool sync = !!(bio->bi_opf & REQ_SYNC);
1667	struct blk_plug *plug;
1668	int el_ret, rw_flags = 0, where = ELEVATOR_INSERT_SORT;
1669	struct request *req;
1670	unsigned int request_count = 0;
1671
1672	/*
1673	* low level driver can indicate that it wants pages above a
1674	* certain limit bounced to low memory (ie for highmem, or even
1675	* ISA dma in theory)
1676	*/
1677	blk_queue_bounce(q, &bio);
1678
1679	blk_queue_split(q, &bio, q->bio_split);
1680
1681	if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
1682	bio->bi_error = -EIO;
1683	bio_endio(bio);
1684	return BLK_QC_T_NONE;
1685	}
1686
1687	if (bio->bi_opf & (REQ_PREFLUSH | REQ_FUA)) {
1688	spin_lock_irq(q->queue_lock);
1689	where = ELEVATOR_INSERT_FLUSH;
1690	goto get_rq;
1691	}
1692
1693	/*
1694	* Check if we can merge with the plugged list before grabbing
1695	* any locks.
1696	*/
1697	if (!blk_queue_nomerges(q)) {
1698	if (blk_attempt_plug_merge(q, bio, &request_count, NULL))
1699	return BLK_QC_T_NONE;
1700	} else
1701	request_count = blk_plug_queued_count(q);
1702
1703	spin_lock_irq(q->queue_lock);
1704
1705	el_ret = elv_merge(q, &req, bio);
1706	if (el_ret == ELEVATOR_BACK_MERGE) {
1707	if (bio_attempt_back_merge(q, req, bio)) {
1708	elv_bio_merged(q, req, bio);
1709	if (!attempt_back_merge(q, req))
1710	elv_merged_request(q, req, el_ret);
1711	goto out_unlock;
1712	}
1713	} else if (el_ret == ELEVATOR_FRONT_MERGE) {
1714	if (bio_attempt_front_merge(q, req, bio)) {
1715	elv_bio_merged(q, req, bio);
1716	if (!attempt_front_merge(q, req))
1717	elv_merged_request(q, req, el_ret);
1718	goto out_unlock;
1719	}
1720	}
1721
1722	get_rq:
1723	/*
1724	* This sync check and mask will be re-done in init_request_from_bio(),
1725	* but we need to set it earlier to expose the sync flag to the
1726	* rq allocator and io schedulers.
1727	*/
1728	if (sync)
1729	rw_flags |= REQ_SYNC;
1730
1731	/*
1732	* Add in META/PRIO flags, if set, before we get to the IO scheduler
1733	*/
1734	rw_flags |= (bio->bi_opf & (REQ_META | REQ_PRIO));
1735
1736	/*
1737	* Grab a free request. This is might sleep but can not fail.
1738	* Returns with the queue unlocked.
1739	*/
1740	req = get_request(q, bio_data_dir(bio), rw_flags, bio, GFP_NOIO);
1741	if (IS_ERR(req)) {
1742	bio->bi_error = PTR_ERR(req);
1743	bio_endio(bio);
1744	goto out_unlock;
1745	}
1746
1747	/*
1748	* After dropping the lock and possibly sleeping here, our request
1749	* may now be mergeable after it had proven unmergeable (above).
1750	* We don't worry about that case for efficiency. It won't happen
1751	* often, and the elevators are able to handle it.
1752	*/
1753	init_request_from_bio(req, bio);
1754
1755	if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
1756	req->cpu = raw_smp_processor_id();
1757
1758	plug = current->plug;
1759	if (plug) {
1760	/*
1761	* If this is the first request added after a plug, fire
1762	* of a plug trace.
1763	*/
1764	if (!request_count)
1765	trace_block_plug(q);
1766	else {
1767	if (request_count >= BLK_MAX_REQUEST_COUNT) {
1768	blk_flush_plug_list(plug, false);
1769	trace_block_plug(q);
1770	}
1771	}
1772	list_add_tail(&req->queuelist, &plug->list);
1773	blk_account_io_start(req, true);
1774	} else {
1775	spin_lock_irq(q->queue_lock);
1776	add_acct_request(q, req, where);
1777	__blk_run_queue(q);
1778	out_unlock:
1779	spin_unlock_irq(q->queue_lock);
1780	}
1781
1782	return BLK_QC_T_NONE;
1783	}
1784
1785	/*
1786	* If bio->bi_dev is a partition, remap the location
1787	*/
1788	static inline void blk_partition_remap(struct bio *bio)
1789	{
1790	struct block_device *bdev = bio->bi_bdev;
1791
1792	if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1793	struct hd_struct *p = bdev->bd_part;
1794
1795	bio->bi_iter.bi_sector += p->start_sect;
1796	bio->bi_bdev = bdev->bd_contains;
1797
1798	trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1799	bdev->bd_dev,
1800	bio->bi_iter.bi_sector - p->start_sect);
1801	}
1802	}
1803
1804	static void handle_bad_sector(struct bio *bio)
1805	{
1806	char b[BDEVNAME_SIZE];
1807
1808	printk(KERN_INFO "attempt to access beyond end of device\n");
1809	printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n",
1810	bdevname(bio->bi_bdev, b),
1811	bio->bi_opf,
1812	(unsigned long long)bio_end_sector(bio),
1813	(long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1814	}
1815
1816	#ifdef CONFIG_FAIL_MAKE_REQUEST
1817
1818	static DECLARE_FAULT_ATTR(fail_make_request);
1819
1820	static int __init setup_fail_make_request(char *str)
1821	{
1822	return setup_fault_attr(&fail_make_request, str);
1823	}
1824	__setup("fail_make_request=", setup_fail_make_request);
1825
1826	static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1827	{
1828	return part->make_it_fail && should_fail(&fail_make_request, bytes);
1829	}
1830
1831	static int __init fail_make_request_debugfs(void)
1832	{
1833	struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1834	NULL, &fail_make_request);
1835
1836	return PTR_ERR_OR_ZERO(dir);
1837	}
1838
1839	late_initcall(fail_make_request_debugfs);
1840
1841	#else /* CONFIG_FAIL_MAKE_REQUEST */
1842
1843	static inline bool should_fail_request(struct hd_struct *part,
1844	unsigned int bytes)
1845	{
1846	return false;
1847	}
1848
1849	#endif /* CONFIG_FAIL_MAKE_REQUEST */
1850
1851	/*
1852	* Check whether this bio extends beyond the end of the device.
1853	*/
1854	static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1855	{
1856	sector_t maxsector;
1857
1858	if (!nr_sectors)
1859	return 0;
1860
1861	/* Test device or partition size, when known. */
1862	maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1863	if (maxsector) {
1864	sector_t sector = bio->bi_iter.bi_sector;
1865
1866	if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1867	/*
1868	* This may well happen - the kernel calls bread()
1869	* without checking the size of the device, e.g., when
1870	* mounting a device.
1871	*/
1872	handle_bad_sector(bio);
1873	return 1;
1874	}
1875	}
1876
1877	return 0;
1878	}
1879
1880	static noinline_for_stack bool
1881	generic_make_request_checks(struct bio *bio)
1882	{
1883	struct request_queue *q;
1884	int nr_sectors = bio_sectors(bio);
1885	int err = -EIO;
1886	char b[BDEVNAME_SIZE];
1887	struct hd_struct *part;
1888
1889	might_sleep();
1890
1891	if (bio_check_eod(bio, nr_sectors))
1892	goto end_io;
1893
1894	q = bdev_get_queue(bio->bi_bdev);
1895	if (unlikely(!q)) {
1896	printk(KERN_ERR
1897	"generic_make_request: Trying to access "
1898	"nonexistent block-device %s (%Lu)\n",
1899	bdevname(bio->bi_bdev, b),
1900	(long long) bio->bi_iter.bi_sector);
1901	goto end_io;
1902	}
1903
1904	part = bio->bi_bdev->bd_part;
1905	if (should_fail_request(part, bio->bi_iter.bi_size) ||
1906	should_fail_request(&part_to_disk(part)->part0,
1907	bio->bi_iter.bi_size))
1908	goto end_io;
1909
1910	/*
1911	* If this device has partitions, remap block n
1912	* of partition p to block n+start(p) of the disk.
1913	*/
1914	blk_partition_remap(bio);
1915
1916	if (bio_check_eod(bio, nr_sectors))
1917	goto end_io;
1918
1919	/*
1920	* Filter flush bio's early so that make_request based
1921	* drivers without flush support don't have to worry
1922	* about them.
1923	*/
1924	if ((bio->bi_opf & (REQ_PREFLUSH | REQ_FUA)) &&
1925	!test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
1926	bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
1927	if (!nr_sectors) {
1928	err = 0;
1929	goto end_io;
1930	}
1931	}
1932
1933	switch (bio_op(bio)) {
1934	case REQ_OP_DISCARD:
1935	if (!blk_queue_discard(q))
1936	goto not_supported;
1937	break;
1938	case REQ_OP_SECURE_ERASE:
1939	if (!blk_queue_secure_erase(q))
1940	goto not_supported;
1941	break;
1942	case REQ_OP_WRITE_SAME:
1943	if (!bdev_write_same(bio->bi_bdev))
1944	goto not_supported;
1945	break;
1946	default:
1947	break;
1948	}
1949
1950	/*
1951	* Various block parts want %current->io_context and lazy ioc
1952	* allocation ends up trading a lot of pain for a small amount of
1953	* memory. Just allocate it upfront. This may fail and block
1954	* layer knows how to live with it.
1955	*/
1956	create_io_context(GFP_ATOMIC, q->node);
1957
1958	if (!blkcg_bio_issue_check(q, bio))
1959	return false;
1960
1961	trace_block_bio_queue(q, bio);
1962	return true;
1963
1964	not_supported:
1965	err = -EOPNOTSUPP;
1966	end_io:
1967	bio->bi_error = err;
1968	bio_endio(bio);
1969	return false;
1970	}
1971
1972	/**
1973	* generic_make_request - hand a buffer to its device driver for I/O
1974	* @bio: The bio describing the location in memory and on the device.
1975	*
1976	* generic_make_request() is used to make I/O requests of block
1977	* devices. It is passed a &struct bio, which describes the I/O that needs
1978	* to be done.
1979	*
1980	* generic_make_request() does not return any status. The
1981	* success/failure status of the request, along with notification of
1982	* completion, is delivered asynchronously through the bio->bi_end_io
1983	* function described (one day) else where.
1984	*
1985	* The caller of generic_make_request must make sure that bi_io_vec
1986	* are set to describe the memory buffer, and that bi_dev and bi_sector are
1987	* set to describe the device address, and the
1988	* bi_end_io and optionally bi_private are set to describe how
1989	* completion notification should be signaled.
1990	*
1991	* generic_make_request and the drivers it calls may use bi_next if this
1992	* bio happens to be merged with someone else, and may resubmit the bio to
1993	* a lower device by calling into generic_make_request recursively, which
1994	* means the bio should NOT be touched after the call to ->make_request_fn.
1995	*/
1996	blk_qc_t generic_make_request(struct bio *bio)
1997	{
1998	/*
1999	* bio_list_on_stack[0] contains bios submitted by the current
2000	* make_request_fn.
2001	* bio_list_on_stack[1] contains bios that were submitted before
2002	* the current make_request_fn, but that haven't been processed
2003	* yet.
2004	*/
2005	struct bio_list bio_list_on_stack[2];
2006	blk_qc_t ret = BLK_QC_T_NONE;
2007
2008	if (!generic_make_request_checks(bio))
2009	goto out;
2010
2011	/*
2012	* We only want one ->make_request_fn to be active at a time, else
2013	* stack usage with stacked devices could be a problem. So use
2014	* current->bio_list to keep a list of requests submited by a
2015	* make_request_fn function. current->bio_list is also used as a
2016	* flag to say if generic_make_request is currently active in this
2017	* task or not. If it is NULL, then no make_request is active. If
2018	* it is non-NULL, then a make_request is active, and new requests
2019	* should be added at the tail
2020	*/
2021	if (current->bio_list) {
2022	bio_list_add(&current->bio_list[0], bio);
2023	goto out;
2024	}
2025
2026	/* following loop may be a bit non-obvious, and so deserves some
2027	* explanation.
2028	* Before entering the loop, bio->bi_next is NULL (as all callers
2029	* ensure that) so we have a list with a single bio.
2030	* We pretend that we have just taken it off a longer list, so
2031	* we assign bio_list to a pointer to the bio_list_on_stack,
2032	* thus initialising the bio_list of new bios to be
2033	* added. ->make_request() may indeed add some more bios
2034	* through a recursive call to generic_make_request. If it
2035	* did, we find a non-NULL value in bio_list and re-enter the loop
2036	* from the top. In this case we really did just take the bio
2037	* of the top of the list (no pretending) and so remove it from
2038	* bio_list, and call into ->make_request() again.
2039	*/
2040	BUG_ON(bio->bi_next);
2041	bio_list_init(&bio_list_on_stack[0]);
2042	current->bio_list = bio_list_on_stack;
2043	do {
2044	struct request_queue *q = bdev_get_queue(bio->bi_bdev);
2045
2046	if (likely(blk_queue_enter(q, false) == 0)) {
2047	struct bio_list lower, same;
2048
2049	/* Create a fresh bio_list for all subordinate requests */
2050	bio_list_on_stack[1] = bio_list_on_stack[0];
2051	bio_list_init(&bio_list_on_stack[0]);
2052	ret = q->make_request_fn(q, bio);
2053
2054	blk_queue_exit(q);
2055
2056	/* sort new bios into those for a lower level
2057	* and those for the same level
2058	*/
2059	bio_list_init(&lower);
2060	bio_list_init(&same);
2061	while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
2062	if (q == bdev_get_queue(bio->bi_bdev))
2063	bio_list_add(&same, bio);
2064	else
2065	bio_list_add(&lower, bio);
2066	/* now assemble so we handle the lowest level first */
2067	bio_list_merge(&bio_list_on_stack[0], &lower);
2068	bio_list_merge(&bio_list_on_stack[0], &same);
2069	bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
2070	} else {
2071	bio_io_error(bio);
2072	}
2073	bio = bio_list_pop(&bio_list_on_stack[0]);
2074	} while (bio);
2075	current->bio_list = NULL; /* deactivate */
2076
2077	out:
2078	return ret;
2079	}
2080	EXPORT_SYMBOL(generic_make_request);
2081
2082	/**
2083	* submit_bio - submit a bio to the block device layer for I/O
2084	* @bio: The &struct bio which describes the I/O
2085	*
2086	* submit_bio() is very similar in purpose to generic_make_request(), and
2087	* uses that function to do most of the work. Both are fairly rough
2088	* interfaces; @bio must be presetup and ready for I/O.
2089	*
2090	*/
2091	blk_qc_t submit_bio(struct bio *bio)
2092	{
2093	/*
2094	* If it's a regular read/write or a barrier with data attached,
2095	* go through the normal accounting stuff before submission.
2096	*/
2097	if (bio_has_data(bio)) {
2098	unsigned int count;
2099
2100	if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
2101	count = bdev_logical_block_size(bio->bi_bdev) >> 9;
2102	else
2103	count = bio_sectors(bio);
2104
2105	if (op_is_write(bio_op(bio))) {
2106	count_vm_events(PGPGOUT, count);
2107	} else {
2108	task_io_account_read(bio->bi_iter.bi_size);
2109	count_vm_events(PGPGIN, count);
2110	}
2111
2112	if (unlikely(block_dump)) {
2113	char b[BDEVNAME_SIZE];
2114	printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
2115	current->comm, task_pid_nr(current),
2116	op_is_write(bio_op(bio)) ? "WRITE" : "READ",
2117	(unsigned long long)bio->bi_iter.bi_sector,
2118	bdevname(bio->bi_bdev, b),
2119	count);
2120	}
2121	}
2122
2123	return generic_make_request(bio);
2124	}
2125	EXPORT_SYMBOL(submit_bio);
2126
2127	/**
2128	* blk_cloned_rq_check_limits - Helper function to check a cloned request
2129	* for new the queue limits
2130	* @q: the queue
2131	* @rq: the request being checked
2132	*
2133	* Description:
2134	* @rq may have been made based on weaker limitations of upper-level queues
2135	* in request stacking drivers, and it may violate the limitation of @q.
2136	* Since the block layer and the underlying device driver trust @rq
2137	* after it is inserted to @q, it should be checked against @q before
2138	* the insertion using this generic function.
2139	*
2140	* Request stacking drivers like request-based dm may change the queue
2141	* limits when retrying requests on other queues. Those requests need
2142	* to be checked against the new queue limits again during dispatch.
2143	*/
2144	static int blk_cloned_rq_check_limits(struct request_queue *q,
2145	struct request *rq)
2146	{
2147	if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) {
2148	printk(KERN_ERR "%s: over max size limit.\n", __func__);
2149	return -EIO;
2150	}
2151
2152	/*
2153	* queue's settings related to segment counting like q->bounce_pfn
2154	* may differ from that of other stacking queues.
2155	* Recalculate it to check the request correctly on this queue's
2156	* limitation.
2157	*/
2158	blk_recalc_rq_segments(rq);
2159	if (rq->nr_phys_segments > queue_max_segments(q)) {
2160	printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2161	return -EIO;
2162	}
2163
2164	return 0;
2165	}
2166
2167	/**
2168	* blk_insert_cloned_request - Helper for stacking drivers to submit a request
2169	* @q: the queue to submit the request
2170	* @rq: the request being queued
2171	*/
2172	int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
2173	{
2174	unsigned long flags;
2175	int where = ELEVATOR_INSERT_BACK;
2176
2177	if (blk_cloned_rq_check_limits(q, rq))
2178	return -EIO;
2179
2180	if (rq->rq_disk &&
2181	should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
2182	return -EIO;
2183
2184	if (q->mq_ops) {
2185	if (blk_queue_io_stat(q))
2186	blk_account_io_start(rq, true);
2187	blk_mq_insert_request(rq, false, true, false);
2188	return 0;
2189	}
2190
2191	spin_lock_irqsave(q->queue_lock, flags);
2192	if (unlikely(blk_queue_dying(q))) {
2193	spin_unlock_irqrestore(q->queue_lock, flags);
2194	return -ENODEV;
2195	}
2196
2197	/*
2198	* Submitting request must be dequeued before calling this function
2199	* because it will be linked to another request_queue
2200	*/
2201	BUG_ON(blk_queued_rq(rq));
2202
2203	if (rq->cmd_flags & (REQ_PREFLUSH | REQ_FUA))
2204	where = ELEVATOR_INSERT_FLUSH;
2205
2206	add_acct_request(q, rq, where);
2207	if (where == ELEVATOR_INSERT_FLUSH)
2208	__blk_run_queue(q);
2209	spin_unlock_irqrestore(q->queue_lock, flags);
2210
2211	return 0;
2212	}
2213	EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2214
2215	/**
2216	* blk_rq_err_bytes - determine number of bytes till the next failure boundary
2217	* @rq: request to examine
2218	*
2219	* Description:
2220	* A request could be merge of IOs which require different failure
2221	* handling. This function determines the number of bytes which
2222	* can be failed from the beginning of the request without
2223	* crossing into area which need to be retried further.
2224	*
2225	* Return:
2226	* The number of bytes to fail.
2227	*
2228	* Context:
2229	* queue_lock must be held.
2230	*/
2231	unsigned int blk_rq_err_bytes(const struct request *rq)
2232	{
2233	unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2234	unsigned int bytes = 0;
2235	struct bio *bio;
2236
2237	if (!(rq->cmd_flags & REQ_MIXED_MERGE))
2238	return blk_rq_bytes(rq);
2239
2240	/*
2241	* Currently the only 'mixing' which can happen is between
2242	* different fastfail types. We can safely fail portions
2243	* which have all the failfast bits that the first one has -
2244	* the ones which are at least as eager to fail as the first
2245	* one.
2246	*/
2247	for (bio = rq->bio; bio; bio = bio->bi_next) {
2248	if ((bio->bi_opf & ff) != ff)
2249	break;
2250	bytes += bio->bi_iter.bi_size;
2251	}
2252
2253	/* this could lead to infinite loop */
2254	BUG_ON(blk_rq_bytes(rq) && !bytes);
2255	return bytes;
2256	}
2257	EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2258
2259	void blk_account_io_completion(struct request *req, unsigned int bytes)
2260	{
2261	if (blk_do_io_stat(req)) {
2262	const int rw = rq_data_dir(req);
2263	struct hd_struct *part;
2264	int cpu;
2265
2266	cpu = part_stat_lock();
2267	part = req->part;
2268	part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2269	part_stat_unlock();
2270	}
2271	}
2272
2273	void blk_account_io_done(struct request *req)
2274	{
2275	/*
2276	* Account IO completion. flush_rq isn't accounted as a
2277	* normal IO on queueing nor completion. Accounting the
2278	* containing request is enough.
2279	*/
2280	if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
2281	unsigned long duration = jiffies - req->start_time;
2282	const int rw = rq_data_dir(req);
2283	struct hd_struct *part;
2284	int cpu;
2285
2286	cpu = part_stat_lock();
2287	part = req->part;
2288
2289	part_stat_inc(cpu, part, ios[rw]);
2290	part_stat_add(cpu, part, ticks[rw], duration);
2291	part_round_stats(cpu, part);
2292	part_dec_in_flight(part, rw);
2293
2294	hd_struct_put(part);
2295	part_stat_unlock();
2296	}
2297	}
2298
2299	#ifdef CONFIG_PM
2300	/*
2301	* Don't process normal requests when queue is suspended
2302	* or in the process of suspending/resuming
2303	*/
2304	static struct request *blk_pm_peek_request(struct request_queue *q,
2305	struct request *rq)
2306	{
2307	if (q->dev && (q->rpm_status == RPM_SUSPENDED ||
2308	(q->rpm_status != RPM_ACTIVE && !(rq->cmd_flags & REQ_PM))))
2309	return NULL;
2310	else
2311	return rq;
2312	}
2313	#else
2314	static inline struct request *blk_pm_peek_request(struct request_queue *q,
2315	struct request *rq)
2316	{
2317	return rq;
2318	}
2319	#endif
2320
2321	void blk_account_io_start(struct request *rq, bool new_io)
2322	{
2323	struct hd_struct *part;
2324	int rw = rq_data_dir(rq);
2325	int cpu;
2326
2327	if (!blk_do_io_stat(rq))
2328	return;
2329
2330	cpu = part_stat_lock();
2331
2332	if (!new_io) {
2333	part = rq->part;
2334	part_stat_inc(cpu, part, merges[rw]);
2335	} else {
2336	part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2337	if (!hd_struct_try_get(part)) {
2338	/*
2339	* The partition is already being removed,
2340	* the request will be accounted on the disk only
2341	*
2342	* We take a reference on disk->part0 although that
2343	* partition will never be deleted, so we can treat
2344	* it as any other partition.
2345	*/
2346	part = &rq->rq_disk->part0;
2347	hd_struct_get(part);
2348	}
2349	part_round_stats(cpu, part);
2350	part_inc_in_flight(part, rw);
2351	rq->part = part;
2352	}
2353
2354	part_stat_unlock();
2355	}
2356
2357	/**
2358	* blk_peek_request - peek at the top of a request queue
2359	* @q: request queue to peek at
2360	*
2361	* Description:
2362	* Return the request at the top of @q. The returned request
2363	* should be started using blk_start_request() before LLD starts
2364	* processing it.
2365	*
2366	* Return:
2367	* Pointer to the request at the top of @q if available. Null
2368	* otherwise.
2369	*
2370	* Context:
2371	* queue_lock must be held.
2372	*/
2373	struct request *blk_peek_request(struct request_queue *q)
2374	{
2375	struct request *rq;
2376	int ret;
2377
2378	while ((rq = __elv_next_request(q)) != NULL) {
2379
2380	rq = blk_pm_peek_request(q, rq);
2381	if (!rq)
2382	break;
2383
2384	if (!(rq->cmd_flags & REQ_STARTED)) {
2385	/*
2386	* This is the first time the device driver
2387	* sees this request (possibly after
2388	* requeueing). Notify IO scheduler.
2389	*/
2390	if (rq->cmd_flags & REQ_SORTED)
2391	elv_activate_rq(q, rq);
2392
2393	/*
2394	* just mark as started even if we don't start
2395	* it, a request that has been delayed should
2396	* not be passed by new incoming requests
2397	*/
2398	rq->cmd_flags |= REQ_STARTED;
2399	trace_block_rq_issue(q, rq);
2400	}
2401
2402	if (!q->boundary_rq || q->boundary_rq == rq) {
2403	q->end_sector = rq_end_sector(rq);
2404	q->boundary_rq = NULL;
2405	}
2406
2407	if (rq->cmd_flags & REQ_DONTPREP)
2408	break;
2409
2410	if (q->dma_drain_size && blk_rq_bytes(rq)) {
2411	/*
2412	* make sure space for the drain appears we
2413	* know we can do this because max_hw_segments
2414	* has been adjusted to be one fewer than the
2415	* device can handle
2416	*/
2417	rq->nr_phys_segments++;
2418	}
2419
2420	if (!q->prep_rq_fn)
2421	break;
2422
2423	ret = q->prep_rq_fn(q, rq);
2424	if (ret == BLKPREP_OK) {
2425	break;
2426	} else if (ret == BLKPREP_DEFER) {
2427	/*
2428	* the request may have been (partially) prepped.
2429	* we need to keep this request in the front to
2430	* avoid resource deadlock. REQ_STARTED will
2431	* prevent other fs requests from passing this one.
2432	*/
2433	if (q->dma_drain_size && blk_rq_bytes(rq) &&
2434	!(rq->cmd_flags & REQ_DONTPREP)) {
2435	/*
2436	* remove the space for the drain we added
2437	* so that we don't add it again
2438	*/
2439	--rq->nr_phys_segments;
2440	}
2441
2442	rq = NULL;
2443	break;
2444	} else if (ret == BLKPREP_KILL || ret == BLKPREP_INVALID) {
2445	int err = (ret == BLKPREP_INVALID) ? -EREMOTEIO : -EIO;
2446
2447	rq->cmd_flags |= REQ_QUIET;
2448	/*
2449	* Mark this request as started so we don't trigger
2450	* any debug logic in the end I/O path.
2451	*/
2452	blk_start_request(rq);
2453	__blk_end_request_all(rq, err);
2454	} else {
2455	printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2456	break;
2457	}
2458	}
2459
2460	return rq;
2461	}
2462	EXPORT_SYMBOL(blk_peek_request);
2463
2464	void blk_dequeue_request(struct request *rq)
2465	{
2466	struct request_queue *q = rq->q;
2467
2468	BUG_ON(list_empty(&rq->queuelist));
2469	BUG_ON(ELV_ON_HASH(rq));
2470
2471	list_del_init(&rq->queuelist);
2472
2473	/*
2474	* the time frame between a request being removed from the lists
2475	* and to it is freed is accounted as io that is in progress at
2476	* the driver side.
2477	*/
2478	if (blk_account_rq(rq)) {
2479	q->in_flight[rq_is_sync(rq)]++;
2480	set_io_start_time_ns(rq);
2481	}
2482	}
2483
2484	/**
2485	* blk_start_request - start request processing on the driver
2486	* @req: request to dequeue
2487	*
2488	* Description:
2489	* Dequeue @req and start timeout timer on it. This hands off the
2490	* request to the driver.
2491	*
2492	* Block internal functions which don't want to start timer should
2493	* call blk_dequeue_request().
2494	*
2495	* Context:
2496	* queue_lock must be held.
2497	*/
2498	void blk_start_request(struct request *req)
2499	{
2500	blk_dequeue_request(req);
2501
2502	/*
2503	* We are now handing the request to the hardware, initialize
2504	* resid_len to full count and add the timeout handler.
2505	*/
2506	req->resid_len = blk_rq_bytes(req);
2507	if (unlikely(blk_bidi_rq(req)))
2508	req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2509
2510	BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags));
2511	blk_add_timer(req);
2512	}
2513	EXPORT_SYMBOL(blk_start_request);
2514
2515	/**
2516	* blk_fetch_request - fetch a request from a request queue
2517	* @q: request queue to fetch a request from
2518	*
2519	* Description:
2520	* Return the request at the top of @q. The request is started on
2521	* return and LLD can start processing it immediately.
2522	*
2523	* Return:
2524	* Pointer to the request at the top of @q if available. Null
2525	* otherwise.
2526	*
2527	* Context:
2528	* queue_lock must be held.
2529	*/
2530	struct request *blk_fetch_request(struct request_queue *q)
2531	{
2532	struct request *rq;
2533
2534	rq = blk_peek_request(q);
2535	if (rq)
2536	blk_start_request(rq);
2537	return rq;
2538	}
2539	EXPORT_SYMBOL(blk_fetch_request);
2540
2541	/**
2542	* blk_update_request - Special helper function for request stacking drivers
2543	* @req: the request being processed
2544	* @error: %0 for success, < %0 for error
2545	* @nr_bytes: number of bytes to complete @req
2546	*
2547	* Description:
2548	* Ends I/O on a number of bytes attached to @req, but doesn't complete
2549	* the request structure even if @req doesn't have leftover.
2550	* If @req has leftover, sets it up for the next range of segments.
2551	*
2552	* This special helper function is only for request stacking drivers
2553	* (e.g. request-based dm) so that they can handle partial completion.
2554	* Actual device drivers should use blk_end_request instead.
2555	*
2556	* Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2557	* %false return from this function.
2558	*
2559	* Return:
2560	* %false - this request doesn't have any more data
2561	* %true - this request has more data
2562	**/
2563	bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2564	{
2565	int total_bytes;
2566
2567	trace_block_rq_complete(req->q, req, nr_bytes);
2568
2569	if (!req->bio)
2570	return false;
2571
2572	/*
2573	* For fs requests, rq is just carrier of independent bio's
2574	* and each partial completion should be handled separately.
2575	* Reset per-request error on each partial completion.
2576	*
2577	* TODO: tj: This is too subtle. It would be better to let
2578	* low level drivers do what they see fit.
2579	*/
2580	if (req->cmd_type == REQ_TYPE_FS)
2581	req->errors = 0;
2582
2583	if (error && req->cmd_type == REQ_TYPE_FS &&
2584	!(req->cmd_flags & REQ_QUIET)) {
2585	char *error_type;
2586
2587	switch (error) {
2588	case -ENOLINK:
2589	error_type = "recoverable transport";
2590	break;
2591	case -EREMOTEIO:
2592	error_type = "critical target";
2593	break;
2594	case -EBADE:
2595	error_type = "critical nexus";
2596	break;
2597	case -ETIMEDOUT:
2598	error_type = "timeout";
2599	break;
2600	case -ENOSPC:
2601	error_type = "critical space allocation";
2602	break;
2603	case -ENODATA:
2604	error_type = "critical medium";
2605	break;
2606	case -EIO:
2607	default:
2608	error_type = "I/O";
2609	break;
2610	}
2611	printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n",
2612	__func__, error_type, req->rq_disk ?
2613	req->rq_disk->disk_name : "?",
2614	(unsigned long long)blk_rq_pos(req));
2615
2616	}
2617
2618	blk_account_io_completion(req, nr_bytes);
2619
2620	total_bytes = 0;
2621	while (req->bio) {
2622	struct bio *bio = req->bio;
2623	unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
2624
2625	if (bio_bytes == bio->bi_iter.bi_size)
2626	req->bio = bio->bi_next;
2627
2628	req_bio_endio(req, bio, bio_bytes, error);
2629
2630	total_bytes += bio_bytes;
2631	nr_bytes -= bio_bytes;
2632
2633	if (!nr_bytes)
2634	break;
2635	}
2636
2637	/*
2638	* completely done
2639	*/
2640	if (!req->bio) {
2641	/*
2642	* Reset counters so that the request stacking driver
2643	* can find how many bytes remain in the request
2644	* later.
2645	*/
2646	req->__data_len = 0;
2647	return false;
2648	}
2649
2650	req->__data_len -= total_bytes;
2651
2652	/* update sector only for requests with clear definition of sector */
2653	if (req->cmd_type == REQ_TYPE_FS)
2654	req->__sector += total_bytes >> 9;
2655
2656	/* mixed attributes always follow the first bio */
2657	if (req->cmd_flags & REQ_MIXED_MERGE) {
2658	req->cmd_flags &= ~REQ_FAILFAST_MASK;
2659	req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
2660	}
2661
2662	/*
2663	* If total number of sectors is less than the first segment
2664	* size, something has gone terribly wrong.
2665	*/
2666	if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2667	blk_dump_rq_flags(req, "request botched");
2668	req->__data_len = blk_rq_cur_bytes(req);
2669	}
2670
2671	/* recalculate the number of segments */
2672	blk_recalc_rq_segments(req);
2673
2674	return true;
2675	}
2676	EXPORT_SYMBOL_GPL(blk_update_request);
2677
2678	static bool blk_update_bidi_request(struct request *rq, int error,
2679	unsigned int nr_bytes,
2680	unsigned int bidi_bytes)
2681	{
2682	if (blk_update_request(rq, error, nr_bytes))
2683	return true;
2684
2685	/* Bidi request must be completed as a whole */
2686	if (unlikely(blk_bidi_rq(rq)) &&
2687	blk_update_request(rq->next_rq, error, bidi_bytes))
2688	return true;
2689
2690	if (blk_queue_add_random(rq->q))
2691	add_disk_randomness(rq->rq_disk);
2692
2693	return false;
2694	}
2695
2696	/**
2697	* blk_unprep_request - unprepare a request
2698	* @req: the request
2699	*
2700	* This function makes a request ready for complete resubmission (or
2701	* completion). It happens only after all error handling is complete,
2702	* so represents the appropriate moment to deallocate any resources
2703	* that were allocated to the request in the prep_rq_fn. The queue
2704	* lock is held when calling this.
2705	*/
2706	void blk_unprep_request(struct request *req)
2707	{
2708	struct request_queue *q = req->q;
2709
2710	req->cmd_flags &= ~REQ_DONTPREP;
2711	if (q->unprep_rq_fn)
2712	q->unprep_rq_fn(q, req);
2713	}
2714	EXPORT_SYMBOL_GPL(blk_unprep_request);
2715
2716	/*
2717	* queue lock must be held
2718	*/
2719	void blk_finish_request(struct request *req, int error)
2720	{
2721	if (req->cmd_flags & REQ_QUEUED)
2722	blk_queue_end_tag(req->q, req);
2723
2724	BUG_ON(blk_queued_rq(req));
2725
2726	if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2727	laptop_io_completion(&req->q->backing_dev_info);
2728
2729	blk_delete_timer(req);
2730
2731	if (req->cmd_flags & REQ_DONTPREP)
2732	blk_unprep_request(req);
2733
2734	blk_account_io_done(req);
2735
2736	if (req->end_io)
2737	req->end_io(req, error);
2738	else {
2739	if (blk_bidi_rq(req))
2740	__blk_put_request(req->next_rq->q, req->next_rq);
2741
2742	__blk_put_request(req->q, req);
2743	}
2744	}
2745	EXPORT_SYMBOL(blk_finish_request);
2746
2747	/**
2748	* blk_end_bidi_request - Complete a bidi request
2749	* @rq: the request to complete
2750	* @error: %0 for success, < %0 for error
2751	* @nr_bytes: number of bytes to complete @rq
2752	* @bidi_bytes: number of bytes to complete @rq->next_rq
2753	*
2754	* Description:
2755	* Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2756	* Drivers that supports bidi can safely call this member for any
2757	* type of request, bidi or uni. In the later case @bidi_bytes is
2758	* just ignored.
2759	*
2760	* Return:
2761	* %false - we are done with this request
2762	* %true - still buffers pending for this request
2763	**/
2764	static bool blk_end_bidi_request(struct request *rq, int error,
2765	unsigned int nr_bytes, unsigned int bidi_bytes)
2766	{
2767	struct request_queue *q = rq->q;
2768	unsigned long flags;
2769
2770	if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2771	return true;
2772
2773	spin_lock_irqsave(q->queue_lock, flags);
2774	blk_finish_request(rq, error);
2775	spin_unlock_irqrestore(q->queue_lock, flags);
2776
2777	return false;
2778	}
2779
2780	/**
2781	* __blk_end_bidi_request - Complete a bidi request with queue lock held
2782	* @rq: the request to complete
2783	* @error: %0 for success, < %0 for error
2784	* @nr_bytes: number of bytes to complete @rq
2785	* @bidi_bytes: number of bytes to complete @rq->next_rq
2786	*
2787	* Description:
2788	* Identical to blk_end_bidi_request() except that queue lock is
2789	* assumed to be locked on entry and remains so on return.
2790	*
2791	* Return:
2792	* %false - we are done with this request
2793	* %true - still buffers pending for this request
2794	**/
2795	bool __blk_end_bidi_request(struct request *rq, int error,
2796	unsigned int nr_bytes, unsigned int bidi_bytes)
2797	{
2798	if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2799	return true;
2800
2801	blk_finish_request(rq, error);
2802
2803	return false;
2804	}
2805
2806	/**
2807	* blk_end_request - Helper function for drivers to complete the request.
2808	* @rq: the request being processed
2809	* @error: %0 for success, < %0 for error
2810	* @nr_bytes: number of bytes to complete
2811	*
2812	* Description:
2813	* Ends I/O on a number of bytes attached to @rq.
2814	* If @rq has leftover, sets it up for the next range of segments.
2815	*
2816	* Return:
2817	* %false - we are done with this request
2818	* %true - still buffers pending for this request
2819	**/
2820	bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2821	{
2822	return blk_end_bidi_request(rq, error, nr_bytes, 0);
2823	}
2824	EXPORT_SYMBOL(blk_end_request);
2825
2826	/**
2827	* blk_end_request_all - Helper function for drives to finish the request.
2828	* @rq: the request to finish
2829	* @error: %0 for success, < %0 for error
2830	*
2831	* Description:
2832	* Completely finish @rq.
2833	*/
2834	void blk_end_request_all(struct request *rq, int error)
2835	{
2836	bool pending;
2837	unsigned int bidi_bytes = 0;
2838
2839	if (unlikely(blk_bidi_rq(rq)))
2840	bidi_bytes = blk_rq_bytes(rq->next_rq);
2841
2842	pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2843	BUG_ON(pending);
2844	}
2845	EXPORT_SYMBOL(blk_end_request_all);
2846
2847	/**
2848	* blk_end_request_cur - Helper function to finish the current request chunk.
2849	* @rq: the request to finish the current chunk for
2850	* @error: %0 for success, < %0 for error
2851	*
2852	* Description:
2853	* Complete the current consecutively mapped chunk from @rq.
2854	*
2855	* Return:
2856	* %false - we are done with this request
2857	* %true - still buffers pending for this request
2858	*/
2859	bool blk_end_request_cur(struct request *rq, int error)
2860	{
2861	return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2862	}
2863	EXPORT_SYMBOL(blk_end_request_cur);
2864
2865	/**
2866	* blk_end_request_err - Finish a request till the next failure boundary.
2867	* @rq: the request to finish till the next failure boundary for
2868	* @error: must be negative errno
2869	*
2870	* Description:
2871	* Complete @rq till the next failure boundary.
2872	*
2873	* Return:
2874	* %false - we are done with this request
2875	* %true - still buffers pending for this request
2876	*/
2877	bool blk_end_request_err(struct request *rq, int error)
2878	{
2879	WARN_ON(error >= 0);
2880	return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2881	}
2882	EXPORT_SYMBOL_GPL(blk_end_request_err);
2883
2884	/**
2885	* __blk_end_request - Helper function for drivers to complete the request.
2886	* @rq: the request being processed
2887	* @error: %0 for success, < %0 for error
2888	* @nr_bytes: number of bytes to complete
2889	*
2890	* Description:
2891	* Must be called with queue lock held unlike blk_end_request().
2892	*
2893	* Return:
2894	* %false - we are done with this request
2895	* %true - still buffers pending for this request
2896	**/
2897	bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2898	{
2899	return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2900	}
2901	EXPORT_SYMBOL(__blk_end_request);
2902
2903	/**
2904	* __blk_end_request_all - Helper function for drives to finish the request.
2905	* @rq: the request to finish
2906	* @error: %0 for success, < %0 for error
2907	*
2908	* Description:
2909	* Completely finish @rq. Must be called with queue lock held.
2910	*/
2911	void __blk_end_request_all(struct request *rq, int error)
2912	{
2913	bool pending;
2914	unsigned int bidi_bytes = 0;
2915
2916	if (unlikely(blk_bidi_rq(rq)))
2917	bidi_bytes = blk_rq_bytes(rq->next_rq);
2918
2919	pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2920	BUG_ON(pending);
2921	}
2922	EXPORT_SYMBOL(__blk_end_request_all);
2923
2924	/**
2925	* __blk_end_request_cur - Helper function to finish the current request chunk.
2926	* @rq: the request to finish the current chunk for
2927	* @error: %0 for success, < %0 for error
2928	*
2929	* Description:
2930	* Complete the current consecutively mapped chunk from @rq. Must
2931	* be called with queue lock held.
2932	*
2933	* Return:
2934	* %false - we are done with this request
2935	* %true - still buffers pending for this request
2936	*/
2937	bool __blk_end_request_cur(struct request *rq, int error)
2938	{
2939	return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2940	}
2941	EXPORT_SYMBOL(__blk_end_request_cur);
2942
2943	/**
2944	* __blk_end_request_err - Finish a request till the next failure boundary.
2945	* @rq: the request to finish till the next failure boundary for
2946	* @error: must be negative errno
2947	*
2948	* Description:
2949	* Complete @rq till the next failure boundary. Must be called
2950	* with queue lock held.
2951	*
2952	* Return:
2953	* %false - we are done with this request
2954	* %true - still buffers pending for this request
2955	*/
2956	bool __blk_end_request_err(struct request *rq, int error)
2957	{
2958	WARN_ON(error >= 0);
2959	return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2960	}
2961	EXPORT_SYMBOL_GPL(__blk_end_request_err);
2962
2963	void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2964	struct bio *bio)
2965	{
2966	req_set_op(rq, bio_op(bio));
2967
2968	if (bio_has_data(bio))
2969	rq->nr_phys_segments = bio_phys_segments(q, bio);
2970
2971	rq->__data_len = bio->bi_iter.bi_size;
2972	rq->bio = rq->biotail = bio;
2973
2974	if (bio->bi_bdev)
2975	rq->rq_disk = bio->bi_bdev->bd_disk;
2976	}
2977
2978	#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2979	/**
2980	* rq_flush_dcache_pages - Helper function to flush all pages in a request
2981	* @rq: the request to be flushed
2982	*
2983	* Description:
2984	* Flush all pages in @rq.
2985	*/
2986	void rq_flush_dcache_pages(struct request *rq)
2987	{
2988	struct req_iterator iter;
2989	struct bio_vec bvec;
2990
2991	rq_for_each_segment(bvec, rq, iter)
2992	flush_dcache_page(bvec.bv_page);
2993	}
2994	EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2995	#endif
2996
2997	/**
2998	* blk_lld_busy - Check if underlying low-level drivers of a device are busy
2999	* @q : the queue of the device being checked
3000	*
3001	* Description:
3002	* Check if underlying low-level drivers of a device are busy.
3003	* If the drivers want to export their busy state, they must set own
3004	* exporting function using blk_queue_lld_busy() first.
3005	*
3006	* Basically, this function is used only by request stacking drivers
3007	* to stop dispatching requests to underlying devices when underlying
3008	* devices are busy. This behavior helps more I/O merging on the queue
3009	* of the request stacking driver and prevents I/O throughput regression
3010	* on burst I/O load.
3011	*
3012	* Return:
3013	* 0 - Not busy (The request stacking driver should dispatch request)
3014	* 1 - Busy (The request stacking driver should stop dispatching request)
3015	*/
3016	int blk_lld_busy(struct request_queue *q)
3017	{
3018	if (q->lld_busy_fn)
3019	return q->lld_busy_fn(q);
3020
3021	return 0;
3022	}
3023	EXPORT_SYMBOL_GPL(blk_lld_busy);
3024
3025	/**
3026	* blk_rq_unprep_clone - Helper function to free all bios in a cloned request
3027	* @rq: the clone request to be cleaned up
3028	*
3029	* Description:
3030	* Free all bios in @rq for a cloned request.
3031	*/
3032	void blk_rq_unprep_clone(struct request *rq)
3033	{
3034	struct bio *bio;
3035
3036	while ((bio = rq->bio) != NULL) {
3037	rq->bio = bio->bi_next;
3038
3039	bio_put(bio);
3040	}
3041	}
3042	EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
3043
3044	/*
3045	* Copy attributes of the original request to the clone request.
3046	* The actual data parts (e.g. ->cmd, ->sense) are not copied.
3047	*/
3048	static void __blk_rq_prep_clone(struct request *dst, struct request *src)
3049	{
3050	dst->cpu = src->cpu;
3051	req_set_op_attrs(dst, req_op(src),
3052	(src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE);
3053	dst->cmd_type = src->cmd_type;
3054	dst->__sector = blk_rq_pos(src);
3055	dst->__data_len = blk_rq_bytes(src);
3056	dst->nr_phys_segments = src->nr_phys_segments;
3057	dst->ioprio = src->ioprio;
3058	dst->extra_len = src->extra_len;
3059	}
3060
3061	/**
3062	* blk_rq_prep_clone - Helper function to setup clone request
3063	* @rq: the request to be setup
3064	* @rq_src: original request to be cloned
3065	* @bs: bio_set that bios for clone are allocated from
3066	* @gfp_mask: memory allocation mask for bio
3067	* @bio_ctr: setup function to be called for each clone bio.
3068	* Returns %0 for success, non %0 for failure.
3069	* @data: private data to be passed to @bio_ctr
3070	*
3071	* Description:
3072	* Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3073	* The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3074	* are not copied, and copying such parts is the caller's responsibility.
3075	* Also, pages which the original bios are pointing to are not copied
3076	* and the cloned bios just point same pages.
3077	* So cloned bios must be completed before original bios, which means
3078	* the caller must complete @rq before @rq_src.
3079	*/
3080	int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
3081	struct bio_set *bs, gfp_t gfp_mask,
3082	int (*bio_ctr)(struct bio *, struct bio *, void *),
3083	void *data)
3084	{
3085	struct bio *bio, *bio_src;
3086
3087	if (!bs)
3088	bs = fs_bio_set;
3089
3090	__rq_for_each_bio(bio_src, rq_src) {
3091	bio = bio_clone_fast(bio_src, gfp_mask, bs);
3092	if (!bio)
3093	goto free_and_out;
3094
3095	if (bio_ctr && bio_ctr(bio, bio_src, data))
3096	goto free_and_out;
3097
3098	if (rq->bio) {
3099	rq->biotail->bi_next = bio;
3100	rq->biotail = bio;
3101	} else
3102	rq->bio = rq->biotail = bio;
3103	}
3104
3105	__blk_rq_prep_clone(rq, rq_src);
3106
3107	return 0;
3108
3109	free_and_out:
3110	if (bio)
3111	bio_put(bio);
3112	blk_rq_unprep_clone(rq);
3113
3114	return -ENOMEM;
3115	}
3116	EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
3117
3118	int kblockd_schedule_work(struct work_struct *work)
3119	{
3120	return queue_work(kblockd_workqueue, work);
3121	}
3122	EXPORT_SYMBOL(kblockd_schedule_work);
3123
3124	int kblockd_schedule_work_on(int cpu, struct work_struct *work)
3125	{
3126	return queue_work_on(cpu, kblockd_workqueue, work);
3127	}
3128	EXPORT_SYMBOL(kblockd_schedule_work_on);
3129
3130	int kblockd_schedule_delayed_work(struct delayed_work *dwork,
3131	unsigned long delay)
3132	{
3133	return queue_delayed_work(kblockd_workqueue, dwork, delay);
3134	}
3135	EXPORT_SYMBOL(kblockd_schedule_delayed_work);
3136
3137	int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork,
3138	unsigned long delay)
3139	{
3140	return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3141	}
3142	EXPORT_SYMBOL(kblockd_schedule_delayed_work_on);
3143
3144	/**
3145	* blk_start_plug - initialize blk_plug and track it inside the task_struct
3146	* @plug: The &struct blk_plug that needs to be initialized
3147	*
3148	* Description:
3149	* Tracking blk_plug inside the task_struct will help with auto-flushing the
3150	* pending I/O should the task end up blocking between blk_start_plug() and
3151	* blk_finish_plug(). This is important from a performance perspective, but
3152	* also ensures that we don't deadlock. For instance, if the task is blocking
3153	* for a memory allocation, memory reclaim could end up wanting to free a
3154	* page belonging to that request that is currently residing in our private
3155	* plug. By flushing the pending I/O when the process goes to sleep, we avoid
3156	* this kind of deadlock.
3157	*/
3158	void blk_start_plug(struct blk_plug *plug)
3159	{
3160	struct task_struct *tsk = current;
3161
3162	/*
3163	* If this is a nested plug, don't actually assign it.
3164	*/
3165	if (tsk->plug)
3166	return;
3167
3168	INIT_LIST_HEAD(&plug->list);
3169	INIT_LIST_HEAD(&plug->mq_list);
3170	INIT_LIST_HEAD(&plug->cb_list);
3171	/*
3172	* Store ordering should not be needed here, since a potential
3173	* preempt will imply a full memory barrier
3174	*/
3175	tsk->plug = plug;
3176	}
3177	EXPORT_SYMBOL(blk_start_plug);
3178
3179	static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
3180	{
3181	struct request *rqa = container_of(a, struct request, queuelist);
3182	struct request *rqb = container_of(b, struct request, queuelist);
3183
3184	return !(rqa->q < rqb->q ||
3185	(rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
3186	}
3187
3188	/*
3189	* If 'from_schedule' is true, then postpone the dispatch of requests
3190	* until a safe kblockd context. We due this to avoid accidental big
3191	* additional stack usage in driver dispatch, in places where the originally
3192	* plugger did not intend it.
3193	*/
3194	static void queue_unplugged(struct request_queue *q, unsigned int depth,
3195	bool from_schedule)
3196	__releases(q->queue_lock)
3197	{
3198	trace_block_unplug(q, depth, !from_schedule);
3199
3200	if (from_schedule)
3201	blk_run_queue_async(q);
3202	else
3203	__blk_run_queue(q);
3204	spin_unlock(q->queue_lock);
3205	}
3206
3207	static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
3208	{
3209	LIST_HEAD(callbacks);
3210
3211	while (!list_empty(&plug->cb_list)) {
3212	list_splice_init(&plug->cb_list, &callbacks);
3213
3214	while (!list_empty(&callbacks)) {
3215	struct blk_plug_cb *cb = list_first_entry(&callbacks,
3216	struct blk_plug_cb,
3217	list);
3218	list_del(&cb->list);
3219	cb->callback(cb, from_schedule);
3220	}
3221	}
3222	}
3223
3224	struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3225	int size)
3226	{
3227	struct blk_plug *plug = current->plug;
3228	struct blk_plug_cb *cb;
3229
3230	if (!plug)
3231	return NULL;
3232
3233	list_for_each_entry(cb, &plug->cb_list, list)
3234	if (cb->callback == unplug && cb->data == data)
3235	return cb;
3236
3237	/* Not currently on the callback list */
3238	BUG_ON(size < sizeof(*cb));
3239	cb = kzalloc(size, GFP_ATOMIC);
3240	if (cb) {
3241	cb->data = data;
3242	cb->callback = unplug;
3243	list_add(&cb->list, &plug->cb_list);
3244	}
3245	return cb;
3246	}
3247	EXPORT_SYMBOL(blk_check_plugged);
3248
3249	void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
3250	{
3251	struct request_queue *q;
3252	unsigned long flags;
3253	struct request *rq;
3254	LIST_HEAD(list);
3255	unsigned int depth;
3256
3257	flush_plug_callbacks(plug, from_schedule);
3258
3259	if (!list_empty(&plug->mq_list))
3260	blk_mq_flush_plug_list(plug, from_schedule);
3261
3262	if (list_empty(&plug->list))
3263	return;
3264
3265	list_splice_init(&plug->list, &list);
3266
3267	list_sort(NULL, &list, plug_rq_cmp);
3268
3269	q = NULL;
3270	depth = 0;
3271
3272	/*
3273	* Save and disable interrupts here, to avoid doing it for every
3274	* queue lock we have to take.
3275	*/
3276	local_irq_save(flags);
3277	while (!list_empty(&list)) {
3278	rq = list_entry_rq(list.next);
3279	list_del_init(&rq->queuelist);
3280	BUG_ON(!rq->q);
3281	if (rq->q != q) {
3282	/*
3283	* This drops the queue lock
3284	*/
3285	if (q)
3286	queue_unplugged(q, depth, from_schedule);
3287	q = rq->q;
3288	depth = 0;
3289	spin_lock(q->queue_lock);
3290	}
3291
3292	/*
3293	* Short-circuit if @q is dead
3294	*/
3295	if (unlikely(blk_queue_dying(q))) {
3296	__blk_end_request_all(rq, -ENODEV);
3297	continue;
3298	}
3299
3300	/*
3301	* rq is already accounted, so use raw insert
3302	*/
3303	if (rq->cmd_flags & (REQ_PREFLUSH | REQ_FUA))
3304	__elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3305	else
3306	__elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
3307
3308	depth++;
3309	}
3310
3311	/*
3312	* This drops the queue lock
3313	*/
3314	if (q)
3315	queue_unplugged(q, depth, from_schedule);
3316
3317	local_irq_restore(flags);
3318	}
3319
3320	void blk_finish_plug(struct blk_plug *plug)
3321	{
3322	if (plug != current->plug)
3323	return;
3324	blk_flush_plug_list(plug, false);
3325
3326	current->plug = NULL;
3327	}
3328	EXPORT_SYMBOL(blk_finish_plug);
3329
3330	bool blk_poll(struct request_queue *q, blk_qc_t cookie)
3331	{
3332	struct blk_plug *plug;
3333	long state;
3334	unsigned int queue_num;
3335	struct blk_mq_hw_ctx *hctx;
3336
3337	if (!q->mq_ops || !q->mq_ops->poll || !blk_qc_t_valid(cookie) ||
3338	!test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
3339	return false;
3340
3341	queue_num = blk_qc_t_to_queue_num(cookie);
3342	hctx = q->queue_hw_ctx[queue_num];
3343	hctx->poll_considered++;
3344
3345	plug = current->plug;
3346	if (plug)
3347	blk_flush_plug_list(plug, false);
3348
3349	state = current->state;
3350	while (!need_resched()) {
3351	int ret;
3352
3353	hctx->poll_invoked++;
3354
3355	ret = q->mq_ops->poll(hctx, blk_qc_t_to_tag(cookie));
3356	if (ret > 0) {
3357	hctx->poll_success++;
3358	set_current_state(TASK_RUNNING);
3359	return true;
3360	}
3361
3362	if (signal_pending_state(state, current))
3363	set_current_state(TASK_RUNNING);
3364
3365	if (current->state == TASK_RUNNING)
3366	return true;
3367	if (ret < 0)
3368	break;
3369	cpu_relax();
3370	}
3371
3372	return false;
3373	}
3374	EXPORT_SYMBOL_GPL(blk_poll);
3375
3376	#ifdef CONFIG_PM
3377	/**
3378	* blk_pm_runtime_init - Block layer runtime PM initialization routine
3379	* @q: the queue of the device
3380	* @dev: the device the queue belongs to
3381	*
3382	* Description:
3383	* Initialize runtime-PM-related fields for @q and start auto suspend for
3384	* @dev. Drivers that want to take advantage of request-based runtime PM
3385	* should call this function after @dev has been initialized, and its
3386	* request queue @q has been allocated, and runtime PM for it can not happen
3387	* yet(either due to disabled/forbidden or its usage_count > 0). In most
3388	* cases, driver should call this function before any I/O has taken place.
3389	*
3390	* This function takes care of setting up using auto suspend for the device,
3391	* the autosuspend delay is set to -1 to make runtime suspend impossible
3392	* until an updated value is either set by user or by driver. Drivers do
3393	* not need to touch other autosuspend settings.
3394	*
3395	* The block layer runtime PM is request based, so only works for drivers
3396	* that use request as their IO unit instead of those directly use bio's.
3397	*/
3398	void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3399	{
3400	q->dev = dev;
3401	q->rpm_status = RPM_ACTIVE;
3402	pm_runtime_set_autosuspend_delay(q->dev, -1);
3403	pm_runtime_use_autosuspend(q->dev);
3404	}
3405	EXPORT_SYMBOL(blk_pm_runtime_init);
3406
3407	/**
3408	* blk_pre_runtime_suspend - Pre runtime suspend check
3409	* @q: the queue of the device
3410	*
3411	* Description:
3412	* This function will check if runtime suspend is allowed for the device
3413	* by examining if there are any requests pending in the queue. If there
3414	* are requests pending, the device can not be runtime suspended; otherwise,
3415	* the queue's status will be updated to SUSPENDING and the driver can
3416	* proceed to suspend the device.
3417	*
3418	* For the not allowed case, we mark last busy for the device so that
3419	* runtime PM core will try to autosuspend it some time later.
3420	*
3421	* This function should be called near the start of the device's
3422	* runtime_suspend callback.
3423	*
3424	* Return:
3425	* 0 - OK to runtime suspend the device
3426	* -EBUSY - Device should not be runtime suspended
3427	*/
3428	int blk_pre_runtime_suspend(struct request_queue *q)
3429	{
3430	int ret = 0;
3431
3432	if (!q->dev)
3433	return ret;
3434
3435	spin_lock_irq(q->queue_lock);
3436	if (q->nr_pending) {
3437	ret = -EBUSY;
3438	pm_runtime_mark_last_busy(q->dev);
3439	} else {
3440	q->rpm_status = RPM_SUSPENDING;
3441	}
3442	spin_unlock_irq(q->queue_lock);
3443	return ret;
3444	}
3445	EXPORT_SYMBOL(blk_pre_runtime_suspend);
3446
3447	/**
3448	* blk_post_runtime_suspend - Post runtime suspend processing
3449	* @q: the queue of the device
3450	* @err: return value of the device's runtime_suspend function
3451	*
3452	* Description:
3453	* Update the queue's runtime status according to the return value of the
3454	* device's runtime suspend function and mark last busy for the device so
3455	* that PM core will try to auto suspend the device at a later time.
3456	*
3457	* This function should be called near the end of the device's
3458	* runtime_suspend callback.
3459	*/
3460	void blk_post_runtime_suspend(struct request_queue *q, int err)
3461	{
3462	if (!q->dev)
3463	return;
3464
3465	spin_lock_irq(q->queue_lock);
3466	if (!err) {
3467	q->rpm_status = RPM_SUSPENDED;
3468	} else {
3469	q->rpm_status = RPM_ACTIVE;
3470	pm_runtime_mark_last_busy(q->dev);
3471	}
3472	spin_unlock_irq(q->queue_lock);
3473	}
3474	EXPORT_SYMBOL(blk_post_runtime_suspend);
3475
3476	/**
3477	* blk_pre_runtime_resume - Pre runtime resume processing
3478	* @q: the queue of the device
3479	*
3480	* Description:
3481	* Update the queue's runtime status to RESUMING in preparation for the
3482	* runtime resume of the device.
3483	*
3484	* This function should be called near the start of the device's
3485	* runtime_resume callback.
3486	*/
3487	void blk_pre_runtime_resume(struct request_queue *q)
3488	{
3489	if (!q->dev)
3490	return;
3491
3492	spin_lock_irq(q->queue_lock);
3493	q->rpm_status = RPM_RESUMING;
3494	spin_unlock_irq(q->queue_lock);
3495	}
3496	EXPORT_SYMBOL(blk_pre_runtime_resume);
3497
3498	/**
3499	* blk_post_runtime_resume - Post runtime resume processing
3500	* @q: the queue of the device
3501	* @err: return value of the device's runtime_resume function
3502	*
3503	* Description:
3504	* Update the queue's runtime status according to the return value of the
3505	* device's runtime_resume function. If it is successfully resumed, process
3506	* the requests that are queued into the device's queue when it is resuming
3507	* and then mark last busy and initiate autosuspend for it.
3508	*
3509	* This function should be called near the end of the device's
3510	* runtime_resume callback.
3511	*/
3512	void blk_post_runtime_resume(struct request_queue *q, int err)
3513	{
3514	if (!q->dev)
3515	return;
3516
3517	spin_lock_irq(q->queue_lock);
3518	if (!err) {
3519	q->rpm_status = RPM_ACTIVE;
3520	__blk_run_queue(q);
3521	pm_runtime_mark_last_busy(q->dev);
3522	pm_request_autosuspend(q->dev);
3523	} else {
3524	q->rpm_status = RPM_SUSPENDED;
3525	}
3526	spin_unlock_irq(q->queue_lock);
3527	}
3528	EXPORT_SYMBOL(blk_post_runtime_resume);
3529
3530	/**
3531	* blk_set_runtime_active - Force runtime status of the queue to be active
3532	* @q: the queue of the device
3533	*
3534	* If the device is left runtime suspended during system suspend the resume
3535	* hook typically resumes the device and corrects runtime status
3536	* accordingly. However, that does not affect the queue runtime PM status
3537	* which is still "suspended". This prevents processing requests from the
3538	* queue.
3539	*
3540	* This function can be used in driver's resume hook to correct queue
3541	* runtime PM status and re-enable peeking requests from the queue. It
3542	* should be called before first request is added to the queue.
3543	*/
3544	void blk_set_runtime_active(struct request_queue *q)
3545	{
3546	spin_lock_irq(q->queue_lock);
3547	q->rpm_status = RPM_ACTIVE;
3548	pm_runtime_mark_last_busy(q->dev);
3549	pm_request_autosuspend(q->dev);
3550	spin_unlock_irq(q->queue_lock);
3551	}
3552	EXPORT_SYMBOL(blk_set_runtime_active);
3553	#endif
3554
3555	int __init blk_dev_init(void)
3556	{
3557	BUILD_BUG_ON(__REQ_NR_BITS > 8 *
3558	FIELD_SIZEOF(struct request, cmd_flags));
3559
3560	/* used for unplugging and affects IO latency/throughput - HIGHPRI */
3561	kblockd_workqueue = alloc_workqueue("kblockd",
3562	WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
3563	if (!kblockd_workqueue)
3564	panic("Failed to create kblockd\n");
3565
3566	request_cachep = kmem_cache_create("blkdev_requests",
3567	sizeof(struct request), 0, SLAB_PANIC, NULL);
3568
3569	blk_requestq_cachep = kmem_cache_create("request_queue",
3570	sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
3571
3572	return 0;
3573	}
3574
3575	/*
3576	* Blk IO latency support. We want this to be as cheap as possible, so doing
3577	* this lockless (and avoiding atomics), a few off by a few errors in this
3578	* code is not harmful, and we don't want to do anything that is
3579	* perf-impactful.
3580	* TODO : If necessary, we can make the histograms per-cpu and aggregate
3581	* them when printing them out.
3582	*/
3583	ssize_t
3584	blk_latency_hist_show(char* name, struct io_latency_state *s, char *buf,
3585	int buf_size)
3586	{
3587	int i;
3588	int bytes_written = 0;
3589	u_int64_t num_elem, elem;
3590	int pct;
3591	u_int64_t average;
3592
3593	num_elem = s->latency_elems;
3594	if (num_elem > 0) {
3595	average = div64_u64(s->latency_sum, s->latency_elems);
3596	bytes_written += scnprintf(buf + bytes_written,
3597	buf_size - bytes_written,
3598	"IO svc_time %s Latency Histogram (n = %llu,"
3599	" average = %llu):\n", name, num_elem, average);
3600	for (i = 0;
3601	i < ARRAY_SIZE(latency_x_axis_us);
3602	i++) {
3603	elem = s->latency_y_axis[i];
3604	pct = div64_u64(elem * 100, num_elem);
3605	bytes_written += scnprintf(buf + bytes_written,
3606	PAGE_SIZE - bytes_written,
3607	"\t< %6lluus%15llu%15d%%\n",
3608	latency_x_axis_us[i],
3609	elem, pct);
3610	}
3611	/* Last element in y-axis table is overflow */
3612	elem = s->latency_y_axis[i];
3613	pct = div64_u64(elem * 100, num_elem);
3614	bytes_written += scnprintf(buf + bytes_written,
3615	PAGE_SIZE - bytes_written,
3616	"\t>=%6lluus%15llu%15d%%\n",
3617	latency_x_axis_us[i - 1], elem, pct);
3618	}
3619
3620	return bytes_written;
3621	}
3622	EXPORT_SYMBOL(blk_latency_hist_show);
3623