xref: /openbmc/linux/block/blk-core.c (revision 9d56dd3b083a3bec56e9da35ce07baca81030b03)
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/highmem.h>
20 #include <linux/mm.h>
21 #include <linux/kernel_stat.h>
22 #include <linux/string.h>
23 #include <linux/init.h>
24 #include <linux/completion.h>
25 #include <linux/slab.h>
26 #include <linux/swap.h>
27 #include <linux/writeback.h>
28 #include <linux/task_io_accounting_ops.h>
29 #include <linux/fault-inject.h>
30 
31 #define CREATE_TRACE_POINTS
32 #include <trace/events/block.h>
33 
34 #include "blk.h"
35 
36 EXPORT_TRACEPOINT_SYMBOL_GPL(block_remap);
37 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
38 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
39 
40 static int __make_request(struct request_queue *q, struct bio *bio);
41 
42 /*
43  * For the allocated request tables
44  */
45 static struct kmem_cache *request_cachep;
46 
47 /*
48  * For queue allocation
49  */
50 struct kmem_cache *blk_requestq_cachep;
51 
52 /*
53  * Controlling structure to kblockd
54  */
55 static struct workqueue_struct *kblockd_workqueue;
56 
57 static void drive_stat_acct(struct request *rq, int new_io)
58 {
59 	struct hd_struct *part;
60 	int rw = rq_data_dir(rq);
61 	int cpu;
62 
63 	if (!blk_do_io_stat(rq))
64 		return;
65 
66 	cpu = part_stat_lock();
67 	part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
68 
69 	if (!new_io)
70 		part_stat_inc(cpu, part, merges[rw]);
71 	else {
72 		part_round_stats(cpu, part);
73 		part_inc_in_flight(part, rw);
74 	}
75 
76 	part_stat_unlock();
77 }
78 
79 void blk_queue_congestion_threshold(struct request_queue *q)
80 {
81 	int nr;
82 
83 	nr = q->nr_requests - (q->nr_requests / 8) + 1;
84 	if (nr > q->nr_requests)
85 		nr = q->nr_requests;
86 	q->nr_congestion_on = nr;
87 
88 	nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
89 	if (nr < 1)
90 		nr = 1;
91 	q->nr_congestion_off = nr;
92 }
93 
94 /**
95  * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
96  * @bdev:	device
97  *
98  * Locates the passed device's request queue and returns the address of its
99  * backing_dev_info
100  *
101  * Will return NULL if the request queue cannot be located.
102  */
103 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
104 {
105 	struct backing_dev_info *ret = NULL;
106 	struct request_queue *q = bdev_get_queue(bdev);
107 
108 	if (q)
109 		ret = &q->backing_dev_info;
110 	return ret;
111 }
112 EXPORT_SYMBOL(blk_get_backing_dev_info);
113 
114 void blk_rq_init(struct request_queue *q, struct request *rq)
115 {
116 	memset(rq, 0, sizeof(*rq));
117 
118 	INIT_LIST_HEAD(&rq->queuelist);
119 	INIT_LIST_HEAD(&rq->timeout_list);
120 	rq->cpu = -1;
121 	rq->q = q;
122 	rq->__sector = (sector_t) -1;
123 	INIT_HLIST_NODE(&rq->hash);
124 	RB_CLEAR_NODE(&rq->rb_node);
125 	rq->cmd = rq->__cmd;
126 	rq->cmd_len = BLK_MAX_CDB;
127 	rq->tag = -1;
128 	rq->ref_count = 1;
129 	rq->start_time = jiffies;
130 }
131 EXPORT_SYMBOL(blk_rq_init);
132 
133 static void req_bio_endio(struct request *rq, struct bio *bio,
134 			  unsigned int nbytes, int error)
135 {
136 	struct request_queue *q = rq->q;
137 
138 	if (&q->bar_rq != rq) {
139 		if (error)
140 			clear_bit(BIO_UPTODATE, &bio->bi_flags);
141 		else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
142 			error = -EIO;
143 
144 		if (unlikely(nbytes > bio->bi_size)) {
145 			printk(KERN_ERR "%s: want %u bytes done, %u left\n",
146 			       __func__, nbytes, bio->bi_size);
147 			nbytes = bio->bi_size;
148 		}
149 
150 		if (unlikely(rq->cmd_flags & REQ_QUIET))
151 			set_bit(BIO_QUIET, &bio->bi_flags);
152 
153 		bio->bi_size -= nbytes;
154 		bio->bi_sector += (nbytes >> 9);
155 
156 		if (bio_integrity(bio))
157 			bio_integrity_advance(bio, nbytes);
158 
159 		if (bio->bi_size == 0)
160 			bio_endio(bio, error);
161 	} else {
162 
163 		/*
164 		 * Okay, this is the barrier request in progress, just
165 		 * record the error;
166 		 */
167 		if (error && !q->orderr)
168 			q->orderr = error;
169 	}
170 }
171 
172 void blk_dump_rq_flags(struct request *rq, char *msg)
173 {
174 	int bit;
175 
176 	printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
177 		rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
178 		rq->cmd_flags);
179 
180 	printk(KERN_INFO "  sector %llu, nr/cnr %u/%u\n",
181 	       (unsigned long long)blk_rq_pos(rq),
182 	       blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
183 	printk(KERN_INFO "  bio %p, biotail %p, buffer %p, len %u\n",
184 	       rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq));
185 
186 	if (blk_pc_request(rq)) {
187 		printk(KERN_INFO "  cdb: ");
188 		for (bit = 0; bit < BLK_MAX_CDB; bit++)
189 			printk("%02x ", rq->cmd[bit]);
190 		printk("\n");
191 	}
192 }
193 EXPORT_SYMBOL(blk_dump_rq_flags);
194 
195 /*
196  * "plug" the device if there are no outstanding requests: this will
197  * force the transfer to start only after we have put all the requests
198  * on the list.
199  *
200  * This is called with interrupts off and no requests on the queue and
201  * with the queue lock held.
202  */
203 void blk_plug_device(struct request_queue *q)
204 {
205 	WARN_ON(!irqs_disabled());
206 
207 	/*
208 	 * don't plug a stopped queue, it must be paired with blk_start_queue()
209 	 * which will restart the queueing
210 	 */
211 	if (blk_queue_stopped(q))
212 		return;
213 
214 	if (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED, q)) {
215 		mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
216 		trace_block_plug(q);
217 	}
218 }
219 EXPORT_SYMBOL(blk_plug_device);
220 
221 /**
222  * blk_plug_device_unlocked - plug a device without queue lock held
223  * @q:    The &struct request_queue to plug
224  *
225  * Description:
226  *   Like @blk_plug_device(), but grabs the queue lock and disables
227  *   interrupts.
228  **/
229 void blk_plug_device_unlocked(struct request_queue *q)
230 {
231 	unsigned long flags;
232 
233 	spin_lock_irqsave(q->queue_lock, flags);
234 	blk_plug_device(q);
235 	spin_unlock_irqrestore(q->queue_lock, flags);
236 }
237 EXPORT_SYMBOL(blk_plug_device_unlocked);
238 
239 /*
240  * remove the queue from the plugged list, if present. called with
241  * queue lock held and interrupts disabled.
242  */
243 int blk_remove_plug(struct request_queue *q)
244 {
245 	WARN_ON(!irqs_disabled());
246 
247 	if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED, q))
248 		return 0;
249 
250 	del_timer(&q->unplug_timer);
251 	return 1;
252 }
253 EXPORT_SYMBOL(blk_remove_plug);
254 
255 /*
256  * remove the plug and let it rip..
257  */
258 void __generic_unplug_device(struct request_queue *q)
259 {
260 	if (unlikely(blk_queue_stopped(q)))
261 		return;
262 	if (!blk_remove_plug(q) && !blk_queue_nonrot(q))
263 		return;
264 
265 	q->request_fn(q);
266 }
267 
268 /**
269  * generic_unplug_device - fire a request queue
270  * @q:    The &struct request_queue in question
271  *
272  * Description:
273  *   Linux uses plugging to build bigger requests queues before letting
274  *   the device have at them. If a queue is plugged, the I/O scheduler
275  *   is still adding and merging requests on the queue. Once the queue
276  *   gets unplugged, the request_fn defined for the queue is invoked and
277  *   transfers started.
278  **/
279 void generic_unplug_device(struct request_queue *q)
280 {
281 	if (blk_queue_plugged(q)) {
282 		spin_lock_irq(q->queue_lock);
283 		__generic_unplug_device(q);
284 		spin_unlock_irq(q->queue_lock);
285 	}
286 }
287 EXPORT_SYMBOL(generic_unplug_device);
288 
289 static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
290 				   struct page *page)
291 {
292 	struct request_queue *q = bdi->unplug_io_data;
293 
294 	blk_unplug(q);
295 }
296 
297 void blk_unplug_work(struct work_struct *work)
298 {
299 	struct request_queue *q =
300 		container_of(work, struct request_queue, unplug_work);
301 
302 	trace_block_unplug_io(q);
303 	q->unplug_fn(q);
304 }
305 
306 void blk_unplug_timeout(unsigned long data)
307 {
308 	struct request_queue *q = (struct request_queue *)data;
309 
310 	trace_block_unplug_timer(q);
311 	kblockd_schedule_work(q, &q->unplug_work);
312 }
313 
314 void blk_unplug(struct request_queue *q)
315 {
316 	/*
317 	 * devices don't necessarily have an ->unplug_fn defined
318 	 */
319 	if (q->unplug_fn) {
320 		trace_block_unplug_io(q);
321 		q->unplug_fn(q);
322 	}
323 }
324 EXPORT_SYMBOL(blk_unplug);
325 
326 /**
327  * blk_start_queue - restart a previously stopped queue
328  * @q:    The &struct request_queue in question
329  *
330  * Description:
331  *   blk_start_queue() will clear the stop flag on the queue, and call
332  *   the request_fn for the queue if it was in a stopped state when
333  *   entered. Also see blk_stop_queue(). Queue lock must be held.
334  **/
335 void blk_start_queue(struct request_queue *q)
336 {
337 	WARN_ON(!irqs_disabled());
338 
339 	queue_flag_clear(QUEUE_FLAG_STOPPED, q);
340 	__blk_run_queue(q);
341 }
342 EXPORT_SYMBOL(blk_start_queue);
343 
344 /**
345  * blk_stop_queue - stop a queue
346  * @q:    The &struct request_queue in question
347  *
348  * Description:
349  *   The Linux block layer assumes that a block driver will consume all
350  *   entries on the request queue when the request_fn strategy is called.
351  *   Often this will not happen, because of hardware limitations (queue
352  *   depth settings). If a device driver gets a 'queue full' response,
353  *   or if it simply chooses not to queue more I/O at one point, it can
354  *   call this function to prevent the request_fn from being called until
355  *   the driver has signalled it's ready to go again. This happens by calling
356  *   blk_start_queue() to restart queue operations. Queue lock must be held.
357  **/
358 void blk_stop_queue(struct request_queue *q)
359 {
360 	blk_remove_plug(q);
361 	queue_flag_set(QUEUE_FLAG_STOPPED, q);
362 }
363 EXPORT_SYMBOL(blk_stop_queue);
364 
365 /**
366  * blk_sync_queue - cancel any pending callbacks on a queue
367  * @q: the queue
368  *
369  * Description:
370  *     The block layer may perform asynchronous callback activity
371  *     on a queue, such as calling the unplug function after a timeout.
372  *     A block device may call blk_sync_queue to ensure that any
373  *     such activity is cancelled, thus allowing it to release resources
374  *     that the callbacks might use. The caller must already have made sure
375  *     that its ->make_request_fn will not re-add plugging prior to calling
376  *     this function.
377  *
378  */
379 void blk_sync_queue(struct request_queue *q)
380 {
381 	del_timer_sync(&q->unplug_timer);
382 	del_timer_sync(&q->timeout);
383 	cancel_work_sync(&q->unplug_work);
384 }
385 EXPORT_SYMBOL(blk_sync_queue);
386 
387 /**
388  * __blk_run_queue - run a single device queue
389  * @q:	The queue to run
390  *
391  * Description:
392  *    See @blk_run_queue. This variant must be called with the queue lock
393  *    held and interrupts disabled.
394  *
395  */
396 void __blk_run_queue(struct request_queue *q)
397 {
398 	blk_remove_plug(q);
399 
400 	if (unlikely(blk_queue_stopped(q)))
401 		return;
402 
403 	if (elv_queue_empty(q))
404 		return;
405 
406 	/*
407 	 * Only recurse once to avoid overrunning the stack, let the unplug
408 	 * handling reinvoke the handler shortly if we already got there.
409 	 */
410 	if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) {
411 		q->request_fn(q);
412 		queue_flag_clear(QUEUE_FLAG_REENTER, q);
413 	} else {
414 		queue_flag_set(QUEUE_FLAG_PLUGGED, q);
415 		kblockd_schedule_work(q, &q->unplug_work);
416 	}
417 }
418 EXPORT_SYMBOL(__blk_run_queue);
419 
420 /**
421  * blk_run_queue - run a single device queue
422  * @q: The queue to run
423  *
424  * Description:
425  *    Invoke request handling on this queue, if it has pending work to do.
426  *    May be used to restart queueing when a request has completed.
427  */
428 void blk_run_queue(struct request_queue *q)
429 {
430 	unsigned long flags;
431 
432 	spin_lock_irqsave(q->queue_lock, flags);
433 	__blk_run_queue(q);
434 	spin_unlock_irqrestore(q->queue_lock, flags);
435 }
436 EXPORT_SYMBOL(blk_run_queue);
437 
438 void blk_put_queue(struct request_queue *q)
439 {
440 	kobject_put(&q->kobj);
441 }
442 
443 void blk_cleanup_queue(struct request_queue *q)
444 {
445 	/*
446 	 * We know we have process context here, so we can be a little
447 	 * cautious and ensure that pending block actions on this device
448 	 * are done before moving on. Going into this function, we should
449 	 * not have processes doing IO to this device.
450 	 */
451 	blk_sync_queue(q);
452 
453 	mutex_lock(&q->sysfs_lock);
454 	queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
455 	mutex_unlock(&q->sysfs_lock);
456 
457 	if (q->elevator)
458 		elevator_exit(q->elevator);
459 
460 	blk_put_queue(q);
461 }
462 EXPORT_SYMBOL(blk_cleanup_queue);
463 
464 static int blk_init_free_list(struct request_queue *q)
465 {
466 	struct request_list *rl = &q->rq;
467 
468 	rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
469 	rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
470 	rl->elvpriv = 0;
471 	init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
472 	init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
473 
474 	rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
475 				mempool_free_slab, request_cachep, q->node);
476 
477 	if (!rl->rq_pool)
478 		return -ENOMEM;
479 
480 	return 0;
481 }
482 
483 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
484 {
485 	return blk_alloc_queue_node(gfp_mask, -1);
486 }
487 EXPORT_SYMBOL(blk_alloc_queue);
488 
489 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
490 {
491 	struct request_queue *q;
492 	int err;
493 
494 	q = kmem_cache_alloc_node(blk_requestq_cachep,
495 				gfp_mask | __GFP_ZERO, node_id);
496 	if (!q)
497 		return NULL;
498 
499 	q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
500 	q->backing_dev_info.unplug_io_data = q;
501 	q->backing_dev_info.ra_pages =
502 			(VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
503 	q->backing_dev_info.state = 0;
504 	q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
505 	q->backing_dev_info.name = "block";
506 
507 	err = bdi_init(&q->backing_dev_info);
508 	if (err) {
509 		kmem_cache_free(blk_requestq_cachep, q);
510 		return NULL;
511 	}
512 
513 	init_timer(&q->unplug_timer);
514 	setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
515 	INIT_LIST_HEAD(&q->timeout_list);
516 	INIT_WORK(&q->unplug_work, blk_unplug_work);
517 
518 	kobject_init(&q->kobj, &blk_queue_ktype);
519 
520 	mutex_init(&q->sysfs_lock);
521 	spin_lock_init(&q->__queue_lock);
522 
523 	return q;
524 }
525 EXPORT_SYMBOL(blk_alloc_queue_node);
526 
527 /**
528  * blk_init_queue  - prepare a request queue for use with a block device
529  * @rfn:  The function to be called to process requests that have been
530  *        placed on the queue.
531  * @lock: Request queue spin lock
532  *
533  * Description:
534  *    If a block device wishes to use the standard request handling procedures,
535  *    which sorts requests and coalesces adjacent requests, then it must
536  *    call blk_init_queue().  The function @rfn will be called when there
537  *    are requests on the queue that need to be processed.  If the device
538  *    supports plugging, then @rfn may not be called immediately when requests
539  *    are available on the queue, but may be called at some time later instead.
540  *    Plugged queues are generally unplugged when a buffer belonging to one
541  *    of the requests on the queue is needed, or due to memory pressure.
542  *
543  *    @rfn is not required, or even expected, to remove all requests off the
544  *    queue, but only as many as it can handle at a time.  If it does leave
545  *    requests on the queue, it is responsible for arranging that the requests
546  *    get dealt with eventually.
547  *
548  *    The queue spin lock must be held while manipulating the requests on the
549  *    request queue; this lock will be taken also from interrupt context, so irq
550  *    disabling is needed for it.
551  *
552  *    Function returns a pointer to the initialized request queue, or %NULL if
553  *    it didn't succeed.
554  *
555  * Note:
556  *    blk_init_queue() must be paired with a blk_cleanup_queue() call
557  *    when the block device is deactivated (such as at module unload).
558  **/
559 
560 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
561 {
562 	return blk_init_queue_node(rfn, lock, -1);
563 }
564 EXPORT_SYMBOL(blk_init_queue);
565 
566 struct request_queue *
567 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
568 {
569 	struct request_queue *q = blk_alloc_queue_node(GFP_KERNEL, node_id);
570 
571 	if (!q)
572 		return NULL;
573 
574 	q->node = node_id;
575 	if (blk_init_free_list(q)) {
576 		kmem_cache_free(blk_requestq_cachep, q);
577 		return NULL;
578 	}
579 
580 	q->request_fn		= rfn;
581 	q->prep_rq_fn		= NULL;
582 	q->unplug_fn		= generic_unplug_device;
583 	q->queue_flags		= QUEUE_FLAG_DEFAULT;
584 	q->queue_lock		= lock;
585 
586 	/*
587 	 * This also sets hw/phys segments, boundary and size
588 	 */
589 	blk_queue_make_request(q, __make_request);
590 
591 	q->sg_reserved_size = INT_MAX;
592 
593 	/*
594 	 * all done
595 	 */
596 	if (!elevator_init(q, NULL)) {
597 		blk_queue_congestion_threshold(q);
598 		return q;
599 	}
600 
601 	blk_put_queue(q);
602 	return NULL;
603 }
604 EXPORT_SYMBOL(blk_init_queue_node);
605 
606 int blk_get_queue(struct request_queue *q)
607 {
608 	if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
609 		kobject_get(&q->kobj);
610 		return 0;
611 	}
612 
613 	return 1;
614 }
615 
616 static inline void blk_free_request(struct request_queue *q, struct request *rq)
617 {
618 	if (rq->cmd_flags & REQ_ELVPRIV)
619 		elv_put_request(q, rq);
620 	mempool_free(rq, q->rq.rq_pool);
621 }
622 
623 static struct request *
624 blk_alloc_request(struct request_queue *q, int flags, int priv, gfp_t gfp_mask)
625 {
626 	struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
627 
628 	if (!rq)
629 		return NULL;
630 
631 	blk_rq_init(q, rq);
632 
633 	rq->cmd_flags = flags | REQ_ALLOCED;
634 
635 	if (priv) {
636 		if (unlikely(elv_set_request(q, rq, gfp_mask))) {
637 			mempool_free(rq, q->rq.rq_pool);
638 			return NULL;
639 		}
640 		rq->cmd_flags |= REQ_ELVPRIV;
641 	}
642 
643 	return rq;
644 }
645 
646 /*
647  * ioc_batching returns true if the ioc is a valid batching request and
648  * should be given priority access to a request.
649  */
650 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
651 {
652 	if (!ioc)
653 		return 0;
654 
655 	/*
656 	 * Make sure the process is able to allocate at least 1 request
657 	 * even if the batch times out, otherwise we could theoretically
658 	 * lose wakeups.
659 	 */
660 	return ioc->nr_batch_requests == q->nr_batching ||
661 		(ioc->nr_batch_requests > 0
662 		&& time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
663 }
664 
665 /*
666  * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
667  * will cause the process to be a "batcher" on all queues in the system. This
668  * is the behaviour we want though - once it gets a wakeup it should be given
669  * a nice run.
670  */
671 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
672 {
673 	if (!ioc || ioc_batching(q, ioc))
674 		return;
675 
676 	ioc->nr_batch_requests = q->nr_batching;
677 	ioc->last_waited = jiffies;
678 }
679 
680 static void __freed_request(struct request_queue *q, int sync)
681 {
682 	struct request_list *rl = &q->rq;
683 
684 	if (rl->count[sync] < queue_congestion_off_threshold(q))
685 		blk_clear_queue_congested(q, sync);
686 
687 	if (rl->count[sync] + 1 <= q->nr_requests) {
688 		if (waitqueue_active(&rl->wait[sync]))
689 			wake_up(&rl->wait[sync]);
690 
691 		blk_clear_queue_full(q, sync);
692 	}
693 }
694 
695 /*
696  * A request has just been released.  Account for it, update the full and
697  * congestion status, wake up any waiters.   Called under q->queue_lock.
698  */
699 static void freed_request(struct request_queue *q, int sync, int priv)
700 {
701 	struct request_list *rl = &q->rq;
702 
703 	rl->count[sync]--;
704 	if (priv)
705 		rl->elvpriv--;
706 
707 	__freed_request(q, sync);
708 
709 	if (unlikely(rl->starved[sync ^ 1]))
710 		__freed_request(q, sync ^ 1);
711 }
712 
713 /*
714  * Get a free request, queue_lock must be held.
715  * Returns NULL on failure, with queue_lock held.
716  * Returns !NULL on success, with queue_lock *not held*.
717  */
718 static struct request *get_request(struct request_queue *q, int rw_flags,
719 				   struct bio *bio, gfp_t gfp_mask)
720 {
721 	struct request *rq = NULL;
722 	struct request_list *rl = &q->rq;
723 	struct io_context *ioc = NULL;
724 	const bool is_sync = rw_is_sync(rw_flags) != 0;
725 	int may_queue, priv;
726 
727 	may_queue = elv_may_queue(q, rw_flags);
728 	if (may_queue == ELV_MQUEUE_NO)
729 		goto rq_starved;
730 
731 	if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
732 		if (rl->count[is_sync]+1 >= q->nr_requests) {
733 			ioc = current_io_context(GFP_ATOMIC, q->node);
734 			/*
735 			 * The queue will fill after this allocation, so set
736 			 * it as full, and mark this process as "batching".
737 			 * This process will be allowed to complete a batch of
738 			 * requests, others will be blocked.
739 			 */
740 			if (!blk_queue_full(q, is_sync)) {
741 				ioc_set_batching(q, ioc);
742 				blk_set_queue_full(q, is_sync);
743 			} else {
744 				if (may_queue != ELV_MQUEUE_MUST
745 						&& !ioc_batching(q, ioc)) {
746 					/*
747 					 * The queue is full and the allocating
748 					 * process is not a "batcher", and not
749 					 * exempted by the IO scheduler
750 					 */
751 					goto out;
752 				}
753 			}
754 		}
755 		blk_set_queue_congested(q, is_sync);
756 	}
757 
758 	/*
759 	 * Only allow batching queuers to allocate up to 50% over the defined
760 	 * limit of requests, otherwise we could have thousands of requests
761 	 * allocated with any setting of ->nr_requests
762 	 */
763 	if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
764 		goto out;
765 
766 	rl->count[is_sync]++;
767 	rl->starved[is_sync] = 0;
768 
769 	priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
770 	if (priv)
771 		rl->elvpriv++;
772 
773 	if (blk_queue_io_stat(q))
774 		rw_flags |= REQ_IO_STAT;
775 	spin_unlock_irq(q->queue_lock);
776 
777 	rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
778 	if (unlikely(!rq)) {
779 		/*
780 		 * Allocation failed presumably due to memory. Undo anything
781 		 * we might have messed up.
782 		 *
783 		 * Allocating task should really be put onto the front of the
784 		 * wait queue, but this is pretty rare.
785 		 */
786 		spin_lock_irq(q->queue_lock);
787 		freed_request(q, is_sync, priv);
788 
789 		/*
790 		 * in the very unlikely event that allocation failed and no
791 		 * requests for this direction was pending, mark us starved
792 		 * so that freeing of a request in the other direction will
793 		 * notice us. another possible fix would be to split the
794 		 * rq mempool into READ and WRITE
795 		 */
796 rq_starved:
797 		if (unlikely(rl->count[is_sync] == 0))
798 			rl->starved[is_sync] = 1;
799 
800 		goto out;
801 	}
802 
803 	/*
804 	 * ioc may be NULL here, and ioc_batching will be false. That's
805 	 * OK, if the queue is under the request limit then requests need
806 	 * not count toward the nr_batch_requests limit. There will always
807 	 * be some limit enforced by BLK_BATCH_TIME.
808 	 */
809 	if (ioc_batching(q, ioc))
810 		ioc->nr_batch_requests--;
811 
812 	trace_block_getrq(q, bio, rw_flags & 1);
813 out:
814 	return rq;
815 }
816 
817 /*
818  * No available requests for this queue, unplug the device and wait for some
819  * requests to become available.
820  *
821  * Called with q->queue_lock held, and returns with it unlocked.
822  */
823 static struct request *get_request_wait(struct request_queue *q, int rw_flags,
824 					struct bio *bio)
825 {
826 	const bool is_sync = rw_is_sync(rw_flags) != 0;
827 	struct request *rq;
828 
829 	rq = get_request(q, rw_flags, bio, GFP_NOIO);
830 	while (!rq) {
831 		DEFINE_WAIT(wait);
832 		struct io_context *ioc;
833 		struct request_list *rl = &q->rq;
834 
835 		prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
836 				TASK_UNINTERRUPTIBLE);
837 
838 		trace_block_sleeprq(q, bio, rw_flags & 1);
839 
840 		__generic_unplug_device(q);
841 		spin_unlock_irq(q->queue_lock);
842 		io_schedule();
843 
844 		/*
845 		 * After sleeping, we become a "batching" process and
846 		 * will be able to allocate at least one request, and
847 		 * up to a big batch of them for a small period time.
848 		 * See ioc_batching, ioc_set_batching
849 		 */
850 		ioc = current_io_context(GFP_NOIO, q->node);
851 		ioc_set_batching(q, ioc);
852 
853 		spin_lock_irq(q->queue_lock);
854 		finish_wait(&rl->wait[is_sync], &wait);
855 
856 		rq = get_request(q, rw_flags, bio, GFP_NOIO);
857 	};
858 
859 	return rq;
860 }
861 
862 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
863 {
864 	struct request *rq;
865 
866 	BUG_ON(rw != READ && rw != WRITE);
867 
868 	spin_lock_irq(q->queue_lock);
869 	if (gfp_mask & __GFP_WAIT) {
870 		rq = get_request_wait(q, rw, NULL);
871 	} else {
872 		rq = get_request(q, rw, NULL, gfp_mask);
873 		if (!rq)
874 			spin_unlock_irq(q->queue_lock);
875 	}
876 	/* q->queue_lock is unlocked at this point */
877 
878 	return rq;
879 }
880 EXPORT_SYMBOL(blk_get_request);
881 
882 /**
883  * blk_make_request - given a bio, allocate a corresponding struct request.
884  * @q: target request queue
885  * @bio:  The bio describing the memory mappings that will be submitted for IO.
886  *        It may be a chained-bio properly constructed by block/bio layer.
887  * @gfp_mask: gfp flags to be used for memory allocation
888  *
889  * blk_make_request is the parallel of generic_make_request for BLOCK_PC
890  * type commands. Where the struct request needs to be farther initialized by
891  * the caller. It is passed a &struct bio, which describes the memory info of
892  * the I/O transfer.
893  *
894  * The caller of blk_make_request must make sure that bi_io_vec
895  * are set to describe the memory buffers. That bio_data_dir() will return
896  * the needed direction of the request. (And all bio's in the passed bio-chain
897  * are properly set accordingly)
898  *
899  * If called under none-sleepable conditions, mapped bio buffers must not
900  * need bouncing, by calling the appropriate masked or flagged allocator,
901  * suitable for the target device. Otherwise the call to blk_queue_bounce will
902  * BUG.
903  *
904  * WARNING: When allocating/cloning a bio-chain, careful consideration should be
905  * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
906  * anything but the first bio in the chain. Otherwise you risk waiting for IO
907  * completion of a bio that hasn't been submitted yet, thus resulting in a
908  * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
909  * of bio_alloc(), as that avoids the mempool deadlock.
910  * If possible a big IO should be split into smaller parts when allocation
911  * fails. Partial allocation should not be an error, or you risk a live-lock.
912  */
913 struct request *blk_make_request(struct request_queue *q, struct bio *bio,
914 				 gfp_t gfp_mask)
915 {
916 	struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
917 
918 	if (unlikely(!rq))
919 		return ERR_PTR(-ENOMEM);
920 
921 	for_each_bio(bio) {
922 		struct bio *bounce_bio = bio;
923 		int ret;
924 
925 		blk_queue_bounce(q, &bounce_bio);
926 		ret = blk_rq_append_bio(q, rq, bounce_bio);
927 		if (unlikely(ret)) {
928 			blk_put_request(rq);
929 			return ERR_PTR(ret);
930 		}
931 	}
932 
933 	return rq;
934 }
935 EXPORT_SYMBOL(blk_make_request);
936 
937 /**
938  * blk_requeue_request - put a request back on queue
939  * @q:		request queue where request should be inserted
940  * @rq:		request to be inserted
941  *
942  * Description:
943  *    Drivers often keep queueing requests until the hardware cannot accept
944  *    more, when that condition happens we need to put the request back
945  *    on the queue. Must be called with queue lock held.
946  */
947 void blk_requeue_request(struct request_queue *q, struct request *rq)
948 {
949 	blk_delete_timer(rq);
950 	blk_clear_rq_complete(rq);
951 	trace_block_rq_requeue(q, rq);
952 
953 	if (blk_rq_tagged(rq))
954 		blk_queue_end_tag(q, rq);
955 
956 	BUG_ON(blk_queued_rq(rq));
957 
958 	elv_requeue_request(q, rq);
959 }
960 EXPORT_SYMBOL(blk_requeue_request);
961 
962 /**
963  * blk_insert_request - insert a special request into a request queue
964  * @q:		request queue where request should be inserted
965  * @rq:		request to be inserted
966  * @at_head:	insert request at head or tail of queue
967  * @data:	private data
968  *
969  * Description:
970  *    Many block devices need to execute commands asynchronously, so they don't
971  *    block the whole kernel from preemption during request execution.  This is
972  *    accomplished normally by inserting aritficial requests tagged as
973  *    REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
974  *    be scheduled for actual execution by the request queue.
975  *
976  *    We have the option of inserting the head or the tail of the queue.
977  *    Typically we use the tail for new ioctls and so forth.  We use the head
978  *    of the queue for things like a QUEUE_FULL message from a device, or a
979  *    host that is unable to accept a particular command.
980  */
981 void blk_insert_request(struct request_queue *q, struct request *rq,
982 			int at_head, void *data)
983 {
984 	int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
985 	unsigned long flags;
986 
987 	/*
988 	 * tell I/O scheduler that this isn't a regular read/write (ie it
989 	 * must not attempt merges on this) and that it acts as a soft
990 	 * barrier
991 	 */
992 	rq->cmd_type = REQ_TYPE_SPECIAL;
993 
994 	rq->special = data;
995 
996 	spin_lock_irqsave(q->queue_lock, flags);
997 
998 	/*
999 	 * If command is tagged, release the tag
1000 	 */
1001 	if (blk_rq_tagged(rq))
1002 		blk_queue_end_tag(q, rq);
1003 
1004 	drive_stat_acct(rq, 1);
1005 	__elv_add_request(q, rq, where, 0);
1006 	__blk_run_queue(q);
1007 	spin_unlock_irqrestore(q->queue_lock, flags);
1008 }
1009 EXPORT_SYMBOL(blk_insert_request);
1010 
1011 /*
1012  * add-request adds a request to the linked list.
1013  * queue lock is held and interrupts disabled, as we muck with the
1014  * request queue list.
1015  */
1016 static inline void add_request(struct request_queue *q, struct request *req)
1017 {
1018 	drive_stat_acct(req, 1);
1019 
1020 	/*
1021 	 * elevator indicated where it wants this request to be
1022 	 * inserted at elevator_merge time
1023 	 */
1024 	__elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0);
1025 }
1026 
1027 static void part_round_stats_single(int cpu, struct hd_struct *part,
1028 				    unsigned long now)
1029 {
1030 	if (now == part->stamp)
1031 		return;
1032 
1033 	if (part_in_flight(part)) {
1034 		__part_stat_add(cpu, part, time_in_queue,
1035 				part_in_flight(part) * (now - part->stamp));
1036 		__part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1037 	}
1038 	part->stamp = now;
1039 }
1040 
1041 /**
1042  * part_round_stats() - Round off the performance stats on a struct disk_stats.
1043  * @cpu: cpu number for stats access
1044  * @part: target partition
1045  *
1046  * The average IO queue length and utilisation statistics are maintained
1047  * by observing the current state of the queue length and the amount of
1048  * time it has been in this state for.
1049  *
1050  * Normally, that accounting is done on IO completion, but that can result
1051  * in more than a second's worth of IO being accounted for within any one
1052  * second, leading to >100% utilisation.  To deal with that, we call this
1053  * function to do a round-off before returning the results when reading
1054  * /proc/diskstats.  This accounts immediately for all queue usage up to
1055  * the current jiffies and restarts the counters again.
1056  */
1057 void part_round_stats(int cpu, struct hd_struct *part)
1058 {
1059 	unsigned long now = jiffies;
1060 
1061 	if (part->partno)
1062 		part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1063 	part_round_stats_single(cpu, part, now);
1064 }
1065 EXPORT_SYMBOL_GPL(part_round_stats);
1066 
1067 /*
1068  * queue lock must be held
1069  */
1070 void __blk_put_request(struct request_queue *q, struct request *req)
1071 {
1072 	if (unlikely(!q))
1073 		return;
1074 	if (unlikely(--req->ref_count))
1075 		return;
1076 
1077 	elv_completed_request(q, req);
1078 
1079 	/* this is a bio leak */
1080 	WARN_ON(req->bio != NULL);
1081 
1082 	/*
1083 	 * Request may not have originated from ll_rw_blk. if not,
1084 	 * it didn't come out of our reserved rq pools
1085 	 */
1086 	if (req->cmd_flags & REQ_ALLOCED) {
1087 		int is_sync = rq_is_sync(req) != 0;
1088 		int priv = req->cmd_flags & REQ_ELVPRIV;
1089 
1090 		BUG_ON(!list_empty(&req->queuelist));
1091 		BUG_ON(!hlist_unhashed(&req->hash));
1092 
1093 		blk_free_request(q, req);
1094 		freed_request(q, is_sync, priv);
1095 	}
1096 }
1097 EXPORT_SYMBOL_GPL(__blk_put_request);
1098 
1099 void blk_put_request(struct request *req)
1100 {
1101 	unsigned long flags;
1102 	struct request_queue *q = req->q;
1103 
1104 	spin_lock_irqsave(q->queue_lock, flags);
1105 	__blk_put_request(q, req);
1106 	spin_unlock_irqrestore(q->queue_lock, flags);
1107 }
1108 EXPORT_SYMBOL(blk_put_request);
1109 
1110 void init_request_from_bio(struct request *req, struct bio *bio)
1111 {
1112 	req->cpu = bio->bi_comp_cpu;
1113 	req->cmd_type = REQ_TYPE_FS;
1114 
1115 	/*
1116 	 * Inherit FAILFAST from bio (for read-ahead, and explicit
1117 	 * FAILFAST).  FAILFAST flags are identical for req and bio.
1118 	 */
1119 	if (bio_rw_flagged(bio, BIO_RW_AHEAD))
1120 		req->cmd_flags |= REQ_FAILFAST_MASK;
1121 	else
1122 		req->cmd_flags |= bio->bi_rw & REQ_FAILFAST_MASK;
1123 
1124 	if (unlikely(bio_rw_flagged(bio, BIO_RW_DISCARD))) {
1125 		req->cmd_flags |= REQ_DISCARD;
1126 		if (bio_rw_flagged(bio, BIO_RW_BARRIER))
1127 			req->cmd_flags |= REQ_SOFTBARRIER;
1128 	} else if (unlikely(bio_rw_flagged(bio, BIO_RW_BARRIER)))
1129 		req->cmd_flags |= REQ_HARDBARRIER;
1130 
1131 	if (bio_rw_flagged(bio, BIO_RW_SYNCIO))
1132 		req->cmd_flags |= REQ_RW_SYNC;
1133 	if (bio_rw_flagged(bio, BIO_RW_META))
1134 		req->cmd_flags |= REQ_RW_META;
1135 	if (bio_rw_flagged(bio, BIO_RW_NOIDLE))
1136 		req->cmd_flags |= REQ_NOIDLE;
1137 
1138 	req->errors = 0;
1139 	req->__sector = bio->bi_sector;
1140 	req->ioprio = bio_prio(bio);
1141 	blk_rq_bio_prep(req->q, req, bio);
1142 }
1143 
1144 /*
1145  * Only disabling plugging for non-rotational devices if it does tagging
1146  * as well, otherwise we do need the proper merging
1147  */
1148 static inline bool queue_should_plug(struct request_queue *q)
1149 {
1150 	return !(blk_queue_nonrot(q) && blk_queue_queuing(q));
1151 }
1152 
1153 static int __make_request(struct request_queue *q, struct bio *bio)
1154 {
1155 	struct request *req;
1156 	int el_ret;
1157 	unsigned int bytes = bio->bi_size;
1158 	const unsigned short prio = bio_prio(bio);
1159 	const bool sync = bio_rw_flagged(bio, BIO_RW_SYNCIO);
1160 	const bool unplug = bio_rw_flagged(bio, BIO_RW_UNPLUG);
1161 	const unsigned int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1162 	int rw_flags;
1163 
1164 	if (bio_rw_flagged(bio, BIO_RW_BARRIER) &&
1165 	    (q->next_ordered == QUEUE_ORDERED_NONE)) {
1166 		bio_endio(bio, -EOPNOTSUPP);
1167 		return 0;
1168 	}
1169 	/*
1170 	 * low level driver can indicate that it wants pages above a
1171 	 * certain limit bounced to low memory (ie for highmem, or even
1172 	 * ISA dma in theory)
1173 	 */
1174 	blk_queue_bounce(q, &bio);
1175 
1176 	spin_lock_irq(q->queue_lock);
1177 
1178 	if (unlikely(bio_rw_flagged(bio, BIO_RW_BARRIER)) || elv_queue_empty(q))
1179 		goto get_rq;
1180 
1181 	el_ret = elv_merge(q, &req, bio);
1182 	switch (el_ret) {
1183 	case ELEVATOR_BACK_MERGE:
1184 		BUG_ON(!rq_mergeable(req));
1185 
1186 		if (!ll_back_merge_fn(q, req, bio))
1187 			break;
1188 
1189 		trace_block_bio_backmerge(q, bio);
1190 
1191 		if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1192 			blk_rq_set_mixed_merge(req);
1193 
1194 		req->biotail->bi_next = bio;
1195 		req->biotail = bio;
1196 		req->__data_len += bytes;
1197 		req->ioprio = ioprio_best(req->ioprio, prio);
1198 		if (!blk_rq_cpu_valid(req))
1199 			req->cpu = bio->bi_comp_cpu;
1200 		drive_stat_acct(req, 0);
1201 		if (!attempt_back_merge(q, req))
1202 			elv_merged_request(q, req, el_ret);
1203 		goto out;
1204 
1205 	case ELEVATOR_FRONT_MERGE:
1206 		BUG_ON(!rq_mergeable(req));
1207 
1208 		if (!ll_front_merge_fn(q, req, bio))
1209 			break;
1210 
1211 		trace_block_bio_frontmerge(q, bio);
1212 
1213 		if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) {
1214 			blk_rq_set_mixed_merge(req);
1215 			req->cmd_flags &= ~REQ_FAILFAST_MASK;
1216 			req->cmd_flags |= ff;
1217 		}
1218 
1219 		bio->bi_next = req->bio;
1220 		req->bio = bio;
1221 
1222 		/*
1223 		 * may not be valid. if the low level driver said
1224 		 * it didn't need a bounce buffer then it better
1225 		 * not touch req->buffer either...
1226 		 */
1227 		req->buffer = bio_data(bio);
1228 		req->__sector = bio->bi_sector;
1229 		req->__data_len += bytes;
1230 		req->ioprio = ioprio_best(req->ioprio, prio);
1231 		if (!blk_rq_cpu_valid(req))
1232 			req->cpu = bio->bi_comp_cpu;
1233 		drive_stat_acct(req, 0);
1234 		if (!attempt_front_merge(q, req))
1235 			elv_merged_request(q, req, el_ret);
1236 		goto out;
1237 
1238 	/* ELV_NO_MERGE: elevator says don't/can't merge. */
1239 	default:
1240 		;
1241 	}
1242 
1243 get_rq:
1244 	/*
1245 	 * This sync check and mask will be re-done in init_request_from_bio(),
1246 	 * but we need to set it earlier to expose the sync flag to the
1247 	 * rq allocator and io schedulers.
1248 	 */
1249 	rw_flags = bio_data_dir(bio);
1250 	if (sync)
1251 		rw_flags |= REQ_RW_SYNC;
1252 
1253 	/*
1254 	 * Grab a free request. This is might sleep but can not fail.
1255 	 * Returns with the queue unlocked.
1256 	 */
1257 	req = get_request_wait(q, rw_flags, bio);
1258 
1259 	/*
1260 	 * After dropping the lock and possibly sleeping here, our request
1261 	 * may now be mergeable after it had proven unmergeable (above).
1262 	 * We don't worry about that case for efficiency. It won't happen
1263 	 * often, and the elevators are able to handle it.
1264 	 */
1265 	init_request_from_bio(req, bio);
1266 
1267 	spin_lock_irq(q->queue_lock);
1268 	if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) ||
1269 	    bio_flagged(bio, BIO_CPU_AFFINE))
1270 		req->cpu = blk_cpu_to_group(smp_processor_id());
1271 	if (queue_should_plug(q) && elv_queue_empty(q))
1272 		blk_plug_device(q);
1273 	add_request(q, req);
1274 out:
1275 	if (unplug || !queue_should_plug(q))
1276 		__generic_unplug_device(q);
1277 	spin_unlock_irq(q->queue_lock);
1278 	return 0;
1279 }
1280 
1281 /*
1282  * If bio->bi_dev is a partition, remap the location
1283  */
1284 static inline void blk_partition_remap(struct bio *bio)
1285 {
1286 	struct block_device *bdev = bio->bi_bdev;
1287 
1288 	if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1289 		struct hd_struct *p = bdev->bd_part;
1290 
1291 		bio->bi_sector += p->start_sect;
1292 		bio->bi_bdev = bdev->bd_contains;
1293 
1294 		trace_block_remap(bdev_get_queue(bio->bi_bdev), bio,
1295 				    bdev->bd_dev,
1296 				    bio->bi_sector - p->start_sect);
1297 	}
1298 }
1299 
1300 static void handle_bad_sector(struct bio *bio)
1301 {
1302 	char b[BDEVNAME_SIZE];
1303 
1304 	printk(KERN_INFO "attempt to access beyond end of device\n");
1305 	printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1306 			bdevname(bio->bi_bdev, b),
1307 			bio->bi_rw,
1308 			(unsigned long long)bio->bi_sector + bio_sectors(bio),
1309 			(long long)(bio->bi_bdev->bd_inode->i_size >> 9));
1310 
1311 	set_bit(BIO_EOF, &bio->bi_flags);
1312 }
1313 
1314 #ifdef CONFIG_FAIL_MAKE_REQUEST
1315 
1316 static DECLARE_FAULT_ATTR(fail_make_request);
1317 
1318 static int __init setup_fail_make_request(char *str)
1319 {
1320 	return setup_fault_attr(&fail_make_request, str);
1321 }
1322 __setup("fail_make_request=", setup_fail_make_request);
1323 
1324 static int should_fail_request(struct bio *bio)
1325 {
1326 	struct hd_struct *part = bio->bi_bdev->bd_part;
1327 
1328 	if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail)
1329 		return should_fail(&fail_make_request, bio->bi_size);
1330 
1331 	return 0;
1332 }
1333 
1334 static int __init fail_make_request_debugfs(void)
1335 {
1336 	return init_fault_attr_dentries(&fail_make_request,
1337 					"fail_make_request");
1338 }
1339 
1340 late_initcall(fail_make_request_debugfs);
1341 
1342 #else /* CONFIG_FAIL_MAKE_REQUEST */
1343 
1344 static inline int should_fail_request(struct bio *bio)
1345 {
1346 	return 0;
1347 }
1348 
1349 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1350 
1351 /*
1352  * Check whether this bio extends beyond the end of the device.
1353  */
1354 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1355 {
1356 	sector_t maxsector;
1357 
1358 	if (!nr_sectors)
1359 		return 0;
1360 
1361 	/* Test device or partition size, when known. */
1362 	maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
1363 	if (maxsector) {
1364 		sector_t sector = bio->bi_sector;
1365 
1366 		if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1367 			/*
1368 			 * This may well happen - the kernel calls bread()
1369 			 * without checking the size of the device, e.g., when
1370 			 * mounting a device.
1371 			 */
1372 			handle_bad_sector(bio);
1373 			return 1;
1374 		}
1375 	}
1376 
1377 	return 0;
1378 }
1379 
1380 /**
1381  * generic_make_request - hand a buffer to its device driver for I/O
1382  * @bio:  The bio describing the location in memory and on the device.
1383  *
1384  * generic_make_request() is used to make I/O requests of block
1385  * devices. It is passed a &struct bio, which describes the I/O that needs
1386  * to be done.
1387  *
1388  * generic_make_request() does not return any status.  The
1389  * success/failure status of the request, along with notification of
1390  * completion, is delivered asynchronously through the bio->bi_end_io
1391  * function described (one day) else where.
1392  *
1393  * The caller of generic_make_request must make sure that bi_io_vec
1394  * are set to describe the memory buffer, and that bi_dev and bi_sector are
1395  * set to describe the device address, and the
1396  * bi_end_io and optionally bi_private are set to describe how
1397  * completion notification should be signaled.
1398  *
1399  * generic_make_request and the drivers it calls may use bi_next if this
1400  * bio happens to be merged with someone else, and may change bi_dev and
1401  * bi_sector for remaps as it sees fit.  So the values of these fields
1402  * should NOT be depended on after the call to generic_make_request.
1403  */
1404 static inline void __generic_make_request(struct bio *bio)
1405 {
1406 	struct request_queue *q;
1407 	sector_t old_sector;
1408 	int ret, nr_sectors = bio_sectors(bio);
1409 	dev_t old_dev;
1410 	int err = -EIO;
1411 
1412 	might_sleep();
1413 
1414 	if (bio_check_eod(bio, nr_sectors))
1415 		goto end_io;
1416 
1417 	/*
1418 	 * Resolve the mapping until finished. (drivers are
1419 	 * still free to implement/resolve their own stacking
1420 	 * by explicitly returning 0)
1421 	 *
1422 	 * NOTE: we don't repeat the blk_size check for each new device.
1423 	 * Stacking drivers are expected to know what they are doing.
1424 	 */
1425 	old_sector = -1;
1426 	old_dev = 0;
1427 	do {
1428 		char b[BDEVNAME_SIZE];
1429 
1430 		q = bdev_get_queue(bio->bi_bdev);
1431 		if (unlikely(!q)) {
1432 			printk(KERN_ERR
1433 			       "generic_make_request: Trying to access "
1434 				"nonexistent block-device %s (%Lu)\n",
1435 				bdevname(bio->bi_bdev, b),
1436 				(long long) bio->bi_sector);
1437 			goto end_io;
1438 		}
1439 
1440 		if (unlikely(!bio_rw_flagged(bio, BIO_RW_DISCARD) &&
1441 			     nr_sectors > queue_max_hw_sectors(q))) {
1442 			printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1443 			       bdevname(bio->bi_bdev, b),
1444 			       bio_sectors(bio),
1445 			       queue_max_hw_sectors(q));
1446 			goto end_io;
1447 		}
1448 
1449 		if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
1450 			goto end_io;
1451 
1452 		if (should_fail_request(bio))
1453 			goto end_io;
1454 
1455 		/*
1456 		 * If this device has partitions, remap block n
1457 		 * of partition p to block n+start(p) of the disk.
1458 		 */
1459 		blk_partition_remap(bio);
1460 
1461 		if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1462 			goto end_io;
1463 
1464 		if (old_sector != -1)
1465 			trace_block_remap(q, bio, old_dev, old_sector);
1466 
1467 		old_sector = bio->bi_sector;
1468 		old_dev = bio->bi_bdev->bd_dev;
1469 
1470 		if (bio_check_eod(bio, nr_sectors))
1471 			goto end_io;
1472 
1473 		if (bio_rw_flagged(bio, BIO_RW_DISCARD) &&
1474 		    !blk_queue_discard(q)) {
1475 			err = -EOPNOTSUPP;
1476 			goto end_io;
1477 		}
1478 
1479 		trace_block_bio_queue(q, bio);
1480 
1481 		ret = q->make_request_fn(q, bio);
1482 	} while (ret);
1483 
1484 	return;
1485 
1486 end_io:
1487 	bio_endio(bio, err);
1488 }
1489 
1490 /*
1491  * We only want one ->make_request_fn to be active at a time,
1492  * else stack usage with stacked devices could be a problem.
1493  * So use current->bio_{list,tail} to keep a list of requests
1494  * submited by a make_request_fn function.
1495  * current->bio_tail is also used as a flag to say if
1496  * generic_make_request is currently active in this task or not.
1497  * If it is NULL, then no make_request is active.  If it is non-NULL,
1498  * then a make_request is active, and new requests should be added
1499  * at the tail
1500  */
1501 void generic_make_request(struct bio *bio)
1502 {
1503 	if (current->bio_tail) {
1504 		/* make_request is active */
1505 		*(current->bio_tail) = bio;
1506 		bio->bi_next = NULL;
1507 		current->bio_tail = &bio->bi_next;
1508 		return;
1509 	}
1510 	/* following loop may be a bit non-obvious, and so deserves some
1511 	 * explanation.
1512 	 * Before entering the loop, bio->bi_next is NULL (as all callers
1513 	 * ensure that) so we have a list with a single bio.
1514 	 * We pretend that we have just taken it off a longer list, so
1515 	 * we assign bio_list to the next (which is NULL) and bio_tail
1516 	 * to &bio_list, thus initialising the bio_list of new bios to be
1517 	 * added.  __generic_make_request may indeed add some more bios
1518 	 * through a recursive call to generic_make_request.  If it
1519 	 * did, we find a non-NULL value in bio_list and re-enter the loop
1520 	 * from the top.  In this case we really did just take the bio
1521 	 * of the top of the list (no pretending) and so fixup bio_list and
1522 	 * bio_tail or bi_next, and call into __generic_make_request again.
1523 	 *
1524 	 * The loop was structured like this to make only one call to
1525 	 * __generic_make_request (which is important as it is large and
1526 	 * inlined) and to keep the structure simple.
1527 	 */
1528 	BUG_ON(bio->bi_next);
1529 	do {
1530 		current->bio_list = bio->bi_next;
1531 		if (bio->bi_next == NULL)
1532 			current->bio_tail = &current->bio_list;
1533 		else
1534 			bio->bi_next = NULL;
1535 		__generic_make_request(bio);
1536 		bio = current->bio_list;
1537 	} while (bio);
1538 	current->bio_tail = NULL; /* deactivate */
1539 }
1540 EXPORT_SYMBOL(generic_make_request);
1541 
1542 /**
1543  * submit_bio - submit a bio to the block device layer for I/O
1544  * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1545  * @bio: The &struct bio which describes the I/O
1546  *
1547  * submit_bio() is very similar in purpose to generic_make_request(), and
1548  * uses that function to do most of the work. Both are fairly rough
1549  * interfaces; @bio must be presetup and ready for I/O.
1550  *
1551  */
1552 void submit_bio(int rw, struct bio *bio)
1553 {
1554 	int count = bio_sectors(bio);
1555 
1556 	bio->bi_rw |= rw;
1557 
1558 	/*
1559 	 * If it's a regular read/write or a barrier with data attached,
1560 	 * go through the normal accounting stuff before submission.
1561 	 */
1562 	if (bio_has_data(bio)) {
1563 		if (rw & WRITE) {
1564 			count_vm_events(PGPGOUT, count);
1565 		} else {
1566 			task_io_account_read(bio->bi_size);
1567 			count_vm_events(PGPGIN, count);
1568 		}
1569 
1570 		if (unlikely(block_dump)) {
1571 			char b[BDEVNAME_SIZE];
1572 			printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n",
1573 			current->comm, task_pid_nr(current),
1574 				(rw & WRITE) ? "WRITE" : "READ",
1575 				(unsigned long long)bio->bi_sector,
1576 				bdevname(bio->bi_bdev, b));
1577 		}
1578 	}
1579 
1580 	generic_make_request(bio);
1581 }
1582 EXPORT_SYMBOL(submit_bio);
1583 
1584 /**
1585  * blk_rq_check_limits - Helper function to check a request for the queue limit
1586  * @q:  the queue
1587  * @rq: the request being checked
1588  *
1589  * Description:
1590  *    @rq may have been made based on weaker limitations of upper-level queues
1591  *    in request stacking drivers, and it may violate the limitation of @q.
1592  *    Since the block layer and the underlying device driver trust @rq
1593  *    after it is inserted to @q, it should be checked against @q before
1594  *    the insertion using this generic function.
1595  *
1596  *    This function should also be useful for request stacking drivers
1597  *    in some cases below, so export this fuction.
1598  *    Request stacking drivers like request-based dm may change the queue
1599  *    limits while requests are in the queue (e.g. dm's table swapping).
1600  *    Such request stacking drivers should check those requests agaist
1601  *    the new queue limits again when they dispatch those requests,
1602  *    although such checkings are also done against the old queue limits
1603  *    when submitting requests.
1604  */
1605 int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1606 {
1607 	if (blk_rq_sectors(rq) > queue_max_sectors(q) ||
1608 	    blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) {
1609 		printk(KERN_ERR "%s: over max size limit.\n", __func__);
1610 		return -EIO;
1611 	}
1612 
1613 	/*
1614 	 * queue's settings related to segment counting like q->bounce_pfn
1615 	 * may differ from that of other stacking queues.
1616 	 * Recalculate it to check the request correctly on this queue's
1617 	 * limitation.
1618 	 */
1619 	blk_recalc_rq_segments(rq);
1620 	if (rq->nr_phys_segments > queue_max_phys_segments(q) ||
1621 	    rq->nr_phys_segments > queue_max_hw_segments(q)) {
1622 		printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1623 		return -EIO;
1624 	}
1625 
1626 	return 0;
1627 }
1628 EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1629 
1630 /**
1631  * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1632  * @q:  the queue to submit the request
1633  * @rq: the request being queued
1634  */
1635 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1636 {
1637 	unsigned long flags;
1638 
1639 	if (blk_rq_check_limits(q, rq))
1640 		return -EIO;
1641 
1642 #ifdef CONFIG_FAIL_MAKE_REQUEST
1643 	if (rq->rq_disk && rq->rq_disk->part0.make_it_fail &&
1644 	    should_fail(&fail_make_request, blk_rq_bytes(rq)))
1645 		return -EIO;
1646 #endif
1647 
1648 	spin_lock_irqsave(q->queue_lock, flags);
1649 
1650 	/*
1651 	 * Submitting request must be dequeued before calling this function
1652 	 * because it will be linked to another request_queue
1653 	 */
1654 	BUG_ON(blk_queued_rq(rq));
1655 
1656 	drive_stat_acct(rq, 1);
1657 	__elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0);
1658 
1659 	spin_unlock_irqrestore(q->queue_lock, flags);
1660 
1661 	return 0;
1662 }
1663 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1664 
1665 /**
1666  * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1667  * @rq: request to examine
1668  *
1669  * Description:
1670  *     A request could be merge of IOs which require different failure
1671  *     handling.  This function determines the number of bytes which
1672  *     can be failed from the beginning of the request without
1673  *     crossing into area which need to be retried further.
1674  *
1675  * Return:
1676  *     The number of bytes to fail.
1677  *
1678  * Context:
1679  *     queue_lock must be held.
1680  */
1681 unsigned int blk_rq_err_bytes(const struct request *rq)
1682 {
1683 	unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1684 	unsigned int bytes = 0;
1685 	struct bio *bio;
1686 
1687 	if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1688 		return blk_rq_bytes(rq);
1689 
1690 	/*
1691 	 * Currently the only 'mixing' which can happen is between
1692 	 * different fastfail types.  We can safely fail portions
1693 	 * which have all the failfast bits that the first one has -
1694 	 * the ones which are at least as eager to fail as the first
1695 	 * one.
1696 	 */
1697 	for (bio = rq->bio; bio; bio = bio->bi_next) {
1698 		if ((bio->bi_rw & ff) != ff)
1699 			break;
1700 		bytes += bio->bi_size;
1701 	}
1702 
1703 	/* this could lead to infinite loop */
1704 	BUG_ON(blk_rq_bytes(rq) && !bytes);
1705 	return bytes;
1706 }
1707 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1708 
1709 static void blk_account_io_completion(struct request *req, unsigned int bytes)
1710 {
1711 	if (blk_do_io_stat(req)) {
1712 		const int rw = rq_data_dir(req);
1713 		struct hd_struct *part;
1714 		int cpu;
1715 
1716 		cpu = part_stat_lock();
1717 		part = disk_map_sector_rcu(req->rq_disk, blk_rq_pos(req));
1718 		part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1719 		part_stat_unlock();
1720 	}
1721 }
1722 
1723 static void blk_account_io_done(struct request *req)
1724 {
1725 	/*
1726 	 * Account IO completion.  bar_rq isn't accounted as a normal
1727 	 * IO on queueing nor completion.  Accounting the containing
1728 	 * request is enough.
1729 	 */
1730 	if (blk_do_io_stat(req) && req != &req->q->bar_rq) {
1731 		unsigned long duration = jiffies - req->start_time;
1732 		const int rw = rq_data_dir(req);
1733 		struct hd_struct *part;
1734 		int cpu;
1735 
1736 		cpu = part_stat_lock();
1737 		part = disk_map_sector_rcu(req->rq_disk, blk_rq_pos(req));
1738 
1739 		part_stat_inc(cpu, part, ios[rw]);
1740 		part_stat_add(cpu, part, ticks[rw], duration);
1741 		part_round_stats(cpu, part);
1742 		part_dec_in_flight(part, rw);
1743 
1744 		part_stat_unlock();
1745 	}
1746 }
1747 
1748 /**
1749  * blk_peek_request - peek at the top of a request queue
1750  * @q: request queue to peek at
1751  *
1752  * Description:
1753  *     Return the request at the top of @q.  The returned request
1754  *     should be started using blk_start_request() before LLD starts
1755  *     processing it.
1756  *
1757  * Return:
1758  *     Pointer to the request at the top of @q if available.  Null
1759  *     otherwise.
1760  *
1761  * Context:
1762  *     queue_lock must be held.
1763  */
1764 struct request *blk_peek_request(struct request_queue *q)
1765 {
1766 	struct request *rq;
1767 	int ret;
1768 
1769 	while ((rq = __elv_next_request(q)) != NULL) {
1770 		if (!(rq->cmd_flags & REQ_STARTED)) {
1771 			/*
1772 			 * This is the first time the device driver
1773 			 * sees this request (possibly after
1774 			 * requeueing).  Notify IO scheduler.
1775 			 */
1776 			if (blk_sorted_rq(rq))
1777 				elv_activate_rq(q, rq);
1778 
1779 			/*
1780 			 * just mark as started even if we don't start
1781 			 * it, a request that has been delayed should
1782 			 * not be passed by new incoming requests
1783 			 */
1784 			rq->cmd_flags |= REQ_STARTED;
1785 			trace_block_rq_issue(q, rq);
1786 		}
1787 
1788 		if (!q->boundary_rq || q->boundary_rq == rq) {
1789 			q->end_sector = rq_end_sector(rq);
1790 			q->boundary_rq = NULL;
1791 		}
1792 
1793 		if (rq->cmd_flags & REQ_DONTPREP)
1794 			break;
1795 
1796 		if (q->dma_drain_size && blk_rq_bytes(rq)) {
1797 			/*
1798 			 * make sure space for the drain appears we
1799 			 * know we can do this because max_hw_segments
1800 			 * has been adjusted to be one fewer than the
1801 			 * device can handle
1802 			 */
1803 			rq->nr_phys_segments++;
1804 		}
1805 
1806 		if (!q->prep_rq_fn)
1807 			break;
1808 
1809 		ret = q->prep_rq_fn(q, rq);
1810 		if (ret == BLKPREP_OK) {
1811 			break;
1812 		} else if (ret == BLKPREP_DEFER) {
1813 			/*
1814 			 * the request may have been (partially) prepped.
1815 			 * we need to keep this request in the front to
1816 			 * avoid resource deadlock.  REQ_STARTED will
1817 			 * prevent other fs requests from passing this one.
1818 			 */
1819 			if (q->dma_drain_size && blk_rq_bytes(rq) &&
1820 			    !(rq->cmd_flags & REQ_DONTPREP)) {
1821 				/*
1822 				 * remove the space for the drain we added
1823 				 * so that we don't add it again
1824 				 */
1825 				--rq->nr_phys_segments;
1826 			}
1827 
1828 			rq = NULL;
1829 			break;
1830 		} else if (ret == BLKPREP_KILL) {
1831 			rq->cmd_flags |= REQ_QUIET;
1832 			/*
1833 			 * Mark this request as started so we don't trigger
1834 			 * any debug logic in the end I/O path.
1835 			 */
1836 			blk_start_request(rq);
1837 			__blk_end_request_all(rq, -EIO);
1838 		} else {
1839 			printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
1840 			break;
1841 		}
1842 	}
1843 
1844 	return rq;
1845 }
1846 EXPORT_SYMBOL(blk_peek_request);
1847 
1848 void blk_dequeue_request(struct request *rq)
1849 {
1850 	struct request_queue *q = rq->q;
1851 
1852 	BUG_ON(list_empty(&rq->queuelist));
1853 	BUG_ON(ELV_ON_HASH(rq));
1854 
1855 	list_del_init(&rq->queuelist);
1856 
1857 	/*
1858 	 * the time frame between a request being removed from the lists
1859 	 * and to it is freed is accounted as io that is in progress at
1860 	 * the driver side.
1861 	 */
1862 	if (blk_account_rq(rq)) {
1863 		q->in_flight[rq_is_sync(rq)]++;
1864 		/*
1865 		 * Mark this device as supporting hardware queuing, if
1866 		 * we have more IOs in flight than 4.
1867 		 */
1868 		if (!blk_queue_queuing(q) && queue_in_flight(q) > 4)
1869 			set_bit(QUEUE_FLAG_CQ, &q->queue_flags);
1870 	}
1871 }
1872 
1873 /**
1874  * blk_start_request - start request processing on the driver
1875  * @req: request to dequeue
1876  *
1877  * Description:
1878  *     Dequeue @req and start timeout timer on it.  This hands off the
1879  *     request to the driver.
1880  *
1881  *     Block internal functions which don't want to start timer should
1882  *     call blk_dequeue_request().
1883  *
1884  * Context:
1885  *     queue_lock must be held.
1886  */
1887 void blk_start_request(struct request *req)
1888 {
1889 	blk_dequeue_request(req);
1890 
1891 	/*
1892 	 * We are now handing the request to the hardware, initialize
1893 	 * resid_len to full count and add the timeout handler.
1894 	 */
1895 	req->resid_len = blk_rq_bytes(req);
1896 	if (unlikely(blk_bidi_rq(req)))
1897 		req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
1898 
1899 	blk_add_timer(req);
1900 }
1901 EXPORT_SYMBOL(blk_start_request);
1902 
1903 /**
1904  * blk_fetch_request - fetch a request from a request queue
1905  * @q: request queue to fetch a request from
1906  *
1907  * Description:
1908  *     Return the request at the top of @q.  The request is started on
1909  *     return and LLD can start processing it immediately.
1910  *
1911  * Return:
1912  *     Pointer to the request at the top of @q if available.  Null
1913  *     otherwise.
1914  *
1915  * Context:
1916  *     queue_lock must be held.
1917  */
1918 struct request *blk_fetch_request(struct request_queue *q)
1919 {
1920 	struct request *rq;
1921 
1922 	rq = blk_peek_request(q);
1923 	if (rq)
1924 		blk_start_request(rq);
1925 	return rq;
1926 }
1927 EXPORT_SYMBOL(blk_fetch_request);
1928 
1929 /**
1930  * blk_update_request - Special helper function for request stacking drivers
1931  * @req:      the request being processed
1932  * @error:    %0 for success, < %0 for error
1933  * @nr_bytes: number of bytes to complete @req
1934  *
1935  * Description:
1936  *     Ends I/O on a number of bytes attached to @req, but doesn't complete
1937  *     the request structure even if @req doesn't have leftover.
1938  *     If @req has leftover, sets it up for the next range of segments.
1939  *
1940  *     This special helper function is only for request stacking drivers
1941  *     (e.g. request-based dm) so that they can handle partial completion.
1942  *     Actual device drivers should use blk_end_request instead.
1943  *
1944  *     Passing the result of blk_rq_bytes() as @nr_bytes guarantees
1945  *     %false return from this function.
1946  *
1947  * Return:
1948  *     %false - this request doesn't have any more data
1949  *     %true  - this request has more data
1950  **/
1951 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
1952 {
1953 	int total_bytes, bio_nbytes, next_idx = 0;
1954 	struct bio *bio;
1955 
1956 	if (!req->bio)
1957 		return false;
1958 
1959 	trace_block_rq_complete(req->q, req);
1960 
1961 	/*
1962 	 * For fs requests, rq is just carrier of independent bio's
1963 	 * and each partial completion should be handled separately.
1964 	 * Reset per-request error on each partial completion.
1965 	 *
1966 	 * TODO: tj: This is too subtle.  It would be better to let
1967 	 * low level drivers do what they see fit.
1968 	 */
1969 	if (blk_fs_request(req))
1970 		req->errors = 0;
1971 
1972 	if (error && (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))) {
1973 		printk(KERN_ERR "end_request: I/O error, dev %s, sector %llu\n",
1974 				req->rq_disk ? req->rq_disk->disk_name : "?",
1975 				(unsigned long long)blk_rq_pos(req));
1976 	}
1977 
1978 	blk_account_io_completion(req, nr_bytes);
1979 
1980 	total_bytes = bio_nbytes = 0;
1981 	while ((bio = req->bio) != NULL) {
1982 		int nbytes;
1983 
1984 		if (nr_bytes >= bio->bi_size) {
1985 			req->bio = bio->bi_next;
1986 			nbytes = bio->bi_size;
1987 			req_bio_endio(req, bio, nbytes, error);
1988 			next_idx = 0;
1989 			bio_nbytes = 0;
1990 		} else {
1991 			int idx = bio->bi_idx + next_idx;
1992 
1993 			if (unlikely(idx >= bio->bi_vcnt)) {
1994 				blk_dump_rq_flags(req, "__end_that");
1995 				printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
1996 				       __func__, idx, bio->bi_vcnt);
1997 				break;
1998 			}
1999 
2000 			nbytes = bio_iovec_idx(bio, idx)->bv_len;
2001 			BIO_BUG_ON(nbytes > bio->bi_size);
2002 
2003 			/*
2004 			 * not a complete bvec done
2005 			 */
2006 			if (unlikely(nbytes > nr_bytes)) {
2007 				bio_nbytes += nr_bytes;
2008 				total_bytes += nr_bytes;
2009 				break;
2010 			}
2011 
2012 			/*
2013 			 * advance to the next vector
2014 			 */
2015 			next_idx++;
2016 			bio_nbytes += nbytes;
2017 		}
2018 
2019 		total_bytes += nbytes;
2020 		nr_bytes -= nbytes;
2021 
2022 		bio = req->bio;
2023 		if (bio) {
2024 			/*
2025 			 * end more in this run, or just return 'not-done'
2026 			 */
2027 			if (unlikely(nr_bytes <= 0))
2028 				break;
2029 		}
2030 	}
2031 
2032 	/*
2033 	 * completely done
2034 	 */
2035 	if (!req->bio) {
2036 		/*
2037 		 * Reset counters so that the request stacking driver
2038 		 * can find how many bytes remain in the request
2039 		 * later.
2040 		 */
2041 		req->__data_len = 0;
2042 		return false;
2043 	}
2044 
2045 	/*
2046 	 * if the request wasn't completed, update state
2047 	 */
2048 	if (bio_nbytes) {
2049 		req_bio_endio(req, bio, bio_nbytes, error);
2050 		bio->bi_idx += next_idx;
2051 		bio_iovec(bio)->bv_offset += nr_bytes;
2052 		bio_iovec(bio)->bv_len -= nr_bytes;
2053 	}
2054 
2055 	req->__data_len -= total_bytes;
2056 	req->buffer = bio_data(req->bio);
2057 
2058 	/* update sector only for requests with clear definition of sector */
2059 	if (blk_fs_request(req) || blk_discard_rq(req))
2060 		req->__sector += total_bytes >> 9;
2061 
2062 	/* mixed attributes always follow the first bio */
2063 	if (req->cmd_flags & REQ_MIXED_MERGE) {
2064 		req->cmd_flags &= ~REQ_FAILFAST_MASK;
2065 		req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2066 	}
2067 
2068 	/*
2069 	 * If total number of sectors is less than the first segment
2070 	 * size, something has gone terribly wrong.
2071 	 */
2072 	if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2073 		printk(KERN_ERR "blk: request botched\n");
2074 		req->__data_len = blk_rq_cur_bytes(req);
2075 	}
2076 
2077 	/* recalculate the number of segments */
2078 	blk_recalc_rq_segments(req);
2079 
2080 	return true;
2081 }
2082 EXPORT_SYMBOL_GPL(blk_update_request);
2083 
2084 static bool blk_update_bidi_request(struct request *rq, int error,
2085 				    unsigned int nr_bytes,
2086 				    unsigned int bidi_bytes)
2087 {
2088 	if (blk_update_request(rq, error, nr_bytes))
2089 		return true;
2090 
2091 	/* Bidi request must be completed as a whole */
2092 	if (unlikely(blk_bidi_rq(rq)) &&
2093 	    blk_update_request(rq->next_rq, error, bidi_bytes))
2094 		return true;
2095 
2096 	add_disk_randomness(rq->rq_disk);
2097 
2098 	return false;
2099 }
2100 
2101 /*
2102  * queue lock must be held
2103  */
2104 static void blk_finish_request(struct request *req, int error)
2105 {
2106 	if (blk_rq_tagged(req))
2107 		blk_queue_end_tag(req->q, req);
2108 
2109 	BUG_ON(blk_queued_rq(req));
2110 
2111 	if (unlikely(laptop_mode) && blk_fs_request(req))
2112 		laptop_io_completion();
2113 
2114 	blk_delete_timer(req);
2115 
2116 	blk_account_io_done(req);
2117 
2118 	if (req->end_io)
2119 		req->end_io(req, error);
2120 	else {
2121 		if (blk_bidi_rq(req))
2122 			__blk_put_request(req->next_rq->q, req->next_rq);
2123 
2124 		__blk_put_request(req->q, req);
2125 	}
2126 }
2127 
2128 /**
2129  * blk_end_bidi_request - Complete a bidi request
2130  * @rq:         the request to complete
2131  * @error:      %0 for success, < %0 for error
2132  * @nr_bytes:   number of bytes to complete @rq
2133  * @bidi_bytes: number of bytes to complete @rq->next_rq
2134  *
2135  * Description:
2136  *     Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2137  *     Drivers that supports bidi can safely call this member for any
2138  *     type of request, bidi or uni.  In the later case @bidi_bytes is
2139  *     just ignored.
2140  *
2141  * Return:
2142  *     %false - we are done with this request
2143  *     %true  - still buffers pending for this request
2144  **/
2145 static bool blk_end_bidi_request(struct request *rq, int error,
2146 				 unsigned int nr_bytes, unsigned int bidi_bytes)
2147 {
2148 	struct request_queue *q = rq->q;
2149 	unsigned long flags;
2150 
2151 	if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2152 		return true;
2153 
2154 	spin_lock_irqsave(q->queue_lock, flags);
2155 	blk_finish_request(rq, error);
2156 	spin_unlock_irqrestore(q->queue_lock, flags);
2157 
2158 	return false;
2159 }
2160 
2161 /**
2162  * __blk_end_bidi_request - Complete a bidi request with queue lock held
2163  * @rq:         the request to complete
2164  * @error:      %0 for success, < %0 for error
2165  * @nr_bytes:   number of bytes to complete @rq
2166  * @bidi_bytes: number of bytes to complete @rq->next_rq
2167  *
2168  * Description:
2169  *     Identical to blk_end_bidi_request() except that queue lock is
2170  *     assumed to be locked on entry and remains so on return.
2171  *
2172  * Return:
2173  *     %false - we are done with this request
2174  *     %true  - still buffers pending for this request
2175  **/
2176 static bool __blk_end_bidi_request(struct request *rq, int error,
2177 				   unsigned int nr_bytes, unsigned int bidi_bytes)
2178 {
2179 	if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2180 		return true;
2181 
2182 	blk_finish_request(rq, error);
2183 
2184 	return false;
2185 }
2186 
2187 /**
2188  * blk_end_request - Helper function for drivers to complete the request.
2189  * @rq:       the request being processed
2190  * @error:    %0 for success, < %0 for error
2191  * @nr_bytes: number of bytes to complete
2192  *
2193  * Description:
2194  *     Ends I/O on a number of bytes attached to @rq.
2195  *     If @rq has leftover, sets it up for the next range of segments.
2196  *
2197  * Return:
2198  *     %false - we are done with this request
2199  *     %true  - still buffers pending for this request
2200  **/
2201 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2202 {
2203 	return blk_end_bidi_request(rq, error, nr_bytes, 0);
2204 }
2205 EXPORT_SYMBOL(blk_end_request);
2206 
2207 /**
2208  * blk_end_request_all - Helper function for drives to finish the request.
2209  * @rq: the request to finish
2210  * @error: %0 for success, < %0 for error
2211  *
2212  * Description:
2213  *     Completely finish @rq.
2214  */
2215 void blk_end_request_all(struct request *rq, int error)
2216 {
2217 	bool pending;
2218 	unsigned int bidi_bytes = 0;
2219 
2220 	if (unlikely(blk_bidi_rq(rq)))
2221 		bidi_bytes = blk_rq_bytes(rq->next_rq);
2222 
2223 	pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2224 	BUG_ON(pending);
2225 }
2226 EXPORT_SYMBOL(blk_end_request_all);
2227 
2228 /**
2229  * blk_end_request_cur - Helper function to finish the current request chunk.
2230  * @rq: the request to finish the current chunk for
2231  * @error: %0 for success, < %0 for error
2232  *
2233  * Description:
2234  *     Complete the current consecutively mapped chunk from @rq.
2235  *
2236  * Return:
2237  *     %false - we are done with this request
2238  *     %true  - still buffers pending for this request
2239  */
2240 bool blk_end_request_cur(struct request *rq, int error)
2241 {
2242 	return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2243 }
2244 EXPORT_SYMBOL(blk_end_request_cur);
2245 
2246 /**
2247  * blk_end_request_err - Finish a request till the next failure boundary.
2248  * @rq: the request to finish till the next failure boundary for
2249  * @error: must be negative errno
2250  *
2251  * Description:
2252  *     Complete @rq till the next failure boundary.
2253  *
2254  * Return:
2255  *     %false - we are done with this request
2256  *     %true  - still buffers pending for this request
2257  */
2258 bool blk_end_request_err(struct request *rq, int error)
2259 {
2260 	WARN_ON(error >= 0);
2261 	return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2262 }
2263 EXPORT_SYMBOL_GPL(blk_end_request_err);
2264 
2265 /**
2266  * __blk_end_request - Helper function for drivers to complete the request.
2267  * @rq:       the request being processed
2268  * @error:    %0 for success, < %0 for error
2269  * @nr_bytes: number of bytes to complete
2270  *
2271  * Description:
2272  *     Must be called with queue lock held unlike blk_end_request().
2273  *
2274  * Return:
2275  *     %false - we are done with this request
2276  *     %true  - still buffers pending for this request
2277  **/
2278 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2279 {
2280 	return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2281 }
2282 EXPORT_SYMBOL(__blk_end_request);
2283 
2284 /**
2285  * __blk_end_request_all - Helper function for drives to finish the request.
2286  * @rq: the request to finish
2287  * @error: %0 for success, < %0 for error
2288  *
2289  * Description:
2290  *     Completely finish @rq.  Must be called with queue lock held.
2291  */
2292 void __blk_end_request_all(struct request *rq, int error)
2293 {
2294 	bool pending;
2295 	unsigned int bidi_bytes = 0;
2296 
2297 	if (unlikely(blk_bidi_rq(rq)))
2298 		bidi_bytes = blk_rq_bytes(rq->next_rq);
2299 
2300 	pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2301 	BUG_ON(pending);
2302 }
2303 EXPORT_SYMBOL(__blk_end_request_all);
2304 
2305 /**
2306  * __blk_end_request_cur - Helper function to finish the current request chunk.
2307  * @rq: the request to finish the current chunk for
2308  * @error: %0 for success, < %0 for error
2309  *
2310  * Description:
2311  *     Complete the current consecutively mapped chunk from @rq.  Must
2312  *     be called with queue lock held.
2313  *
2314  * Return:
2315  *     %false - we are done with this request
2316  *     %true  - still buffers pending for this request
2317  */
2318 bool __blk_end_request_cur(struct request *rq, int error)
2319 {
2320 	return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2321 }
2322 EXPORT_SYMBOL(__blk_end_request_cur);
2323 
2324 /**
2325  * __blk_end_request_err - Finish a request till the next failure boundary.
2326  * @rq: the request to finish till the next failure boundary for
2327  * @error: must be negative errno
2328  *
2329  * Description:
2330  *     Complete @rq till the next failure boundary.  Must be called
2331  *     with queue lock held.
2332  *
2333  * Return:
2334  *     %false - we are done with this request
2335  *     %true  - still buffers pending for this request
2336  */
2337 bool __blk_end_request_err(struct request *rq, int error)
2338 {
2339 	WARN_ON(error >= 0);
2340 	return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2341 }
2342 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2343 
2344 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2345 		     struct bio *bio)
2346 {
2347 	/* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2348 	rq->cmd_flags |= bio->bi_rw & REQ_RW;
2349 
2350 	if (bio_has_data(bio)) {
2351 		rq->nr_phys_segments = bio_phys_segments(q, bio);
2352 		rq->buffer = bio_data(bio);
2353 	}
2354 	rq->__data_len = bio->bi_size;
2355 	rq->bio = rq->biotail = bio;
2356 
2357 	if (bio->bi_bdev)
2358 		rq->rq_disk = bio->bi_bdev->bd_disk;
2359 }
2360 
2361 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2362 /**
2363  * rq_flush_dcache_pages - Helper function to flush all pages in a request
2364  * @rq: the request to be flushed
2365  *
2366  * Description:
2367  *     Flush all pages in @rq.
2368  */
2369 void rq_flush_dcache_pages(struct request *rq)
2370 {
2371 	struct req_iterator iter;
2372 	struct bio_vec *bvec;
2373 
2374 	rq_for_each_segment(bvec, rq, iter)
2375 		flush_dcache_page(bvec->bv_page);
2376 }
2377 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2378 #endif
2379 
2380 /**
2381  * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2382  * @q : the queue of the device being checked
2383  *
2384  * Description:
2385  *    Check if underlying low-level drivers of a device are busy.
2386  *    If the drivers want to export their busy state, they must set own
2387  *    exporting function using blk_queue_lld_busy() first.
2388  *
2389  *    Basically, this function is used only by request stacking drivers
2390  *    to stop dispatching requests to underlying devices when underlying
2391  *    devices are busy.  This behavior helps more I/O merging on the queue
2392  *    of the request stacking driver and prevents I/O throughput regression
2393  *    on burst I/O load.
2394  *
2395  * Return:
2396  *    0 - Not busy (The request stacking driver should dispatch request)
2397  *    1 - Busy (The request stacking driver should stop dispatching request)
2398  */
2399 int blk_lld_busy(struct request_queue *q)
2400 {
2401 	if (q->lld_busy_fn)
2402 		return q->lld_busy_fn(q);
2403 
2404 	return 0;
2405 }
2406 EXPORT_SYMBOL_GPL(blk_lld_busy);
2407 
2408 /**
2409  * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2410  * @rq: the clone request to be cleaned up
2411  *
2412  * Description:
2413  *     Free all bios in @rq for a cloned request.
2414  */
2415 void blk_rq_unprep_clone(struct request *rq)
2416 {
2417 	struct bio *bio;
2418 
2419 	while ((bio = rq->bio) != NULL) {
2420 		rq->bio = bio->bi_next;
2421 
2422 		bio_put(bio);
2423 	}
2424 }
2425 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2426 
2427 /*
2428  * Copy attributes of the original request to the clone request.
2429  * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2430  */
2431 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2432 {
2433 	dst->cpu = src->cpu;
2434 	dst->cmd_flags = (rq_data_dir(src) | REQ_NOMERGE);
2435 	dst->cmd_type = src->cmd_type;
2436 	dst->__sector = blk_rq_pos(src);
2437 	dst->__data_len = blk_rq_bytes(src);
2438 	dst->nr_phys_segments = src->nr_phys_segments;
2439 	dst->ioprio = src->ioprio;
2440 	dst->extra_len = src->extra_len;
2441 }
2442 
2443 /**
2444  * blk_rq_prep_clone - Helper function to setup clone request
2445  * @rq: the request to be setup
2446  * @rq_src: original request to be cloned
2447  * @bs: bio_set that bios for clone are allocated from
2448  * @gfp_mask: memory allocation mask for bio
2449  * @bio_ctr: setup function to be called for each clone bio.
2450  *           Returns %0 for success, non %0 for failure.
2451  * @data: private data to be passed to @bio_ctr
2452  *
2453  * Description:
2454  *     Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2455  *     The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2456  *     are not copied, and copying such parts is the caller's responsibility.
2457  *     Also, pages which the original bios are pointing to are not copied
2458  *     and the cloned bios just point same pages.
2459  *     So cloned bios must be completed before original bios, which means
2460  *     the caller must complete @rq before @rq_src.
2461  */
2462 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2463 		      struct bio_set *bs, gfp_t gfp_mask,
2464 		      int (*bio_ctr)(struct bio *, struct bio *, void *),
2465 		      void *data)
2466 {
2467 	struct bio *bio, *bio_src;
2468 
2469 	if (!bs)
2470 		bs = fs_bio_set;
2471 
2472 	blk_rq_init(NULL, rq);
2473 
2474 	__rq_for_each_bio(bio_src, rq_src) {
2475 		bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs);
2476 		if (!bio)
2477 			goto free_and_out;
2478 
2479 		__bio_clone(bio, bio_src);
2480 
2481 		if (bio_integrity(bio_src) &&
2482 		    bio_integrity_clone(bio, bio_src, gfp_mask, bs))
2483 			goto free_and_out;
2484 
2485 		if (bio_ctr && bio_ctr(bio, bio_src, data))
2486 			goto free_and_out;
2487 
2488 		if (rq->bio) {
2489 			rq->biotail->bi_next = bio;
2490 			rq->biotail = bio;
2491 		} else
2492 			rq->bio = rq->biotail = bio;
2493 	}
2494 
2495 	__blk_rq_prep_clone(rq, rq_src);
2496 
2497 	return 0;
2498 
2499 free_and_out:
2500 	if (bio)
2501 		bio_free(bio, bs);
2502 	blk_rq_unprep_clone(rq);
2503 
2504 	return -ENOMEM;
2505 }
2506 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2507 
2508 int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
2509 {
2510 	return queue_work(kblockd_workqueue, work);
2511 }
2512 EXPORT_SYMBOL(kblockd_schedule_work);
2513 
2514 int __init blk_dev_init(void)
2515 {
2516 	BUILD_BUG_ON(__REQ_NR_BITS > 8 *
2517 			sizeof(((struct request *)0)->cmd_flags));
2518 
2519 	kblockd_workqueue = create_workqueue("kblockd");
2520 	if (!kblockd_workqueue)
2521 		panic("Failed to create kblockd\n");
2522 
2523 	request_cachep = kmem_cache_create("blkdev_requests",
2524 			sizeof(struct request), 0, SLAB_PANIC, NULL);
2525 
2526 	blk_requestq_cachep = kmem_cache_create("blkdev_queue",
2527 			sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
2528 
2529 	return 0;
2530 }
2531 
2532