Loading...
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#include <linux/list_sort.h>
31
32#define CREATE_TRACE_POINTS
33#include <trace/events/block.h>
34
35#include "blk.h"
36
37EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
38EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
39EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
40
41static int __make_request(struct request_queue *q, struct bio *bio);
42
43/*
44 * For the allocated request tables
45 */
46static struct kmem_cache *request_cachep;
47
48/*
49 * For queue allocation
50 */
51struct kmem_cache *blk_requestq_cachep;
52
53/*
54 * Controlling structure to kblockd
55 */
56static struct workqueue_struct *kblockd_workqueue;
57
58static void drive_stat_acct(struct request *rq, int new_io)
59{
60 struct hd_struct *part;
61 int rw = rq_data_dir(rq);
62 int cpu;
63
64 if (!blk_do_io_stat(rq))
65 return;
66
67 cpu = part_stat_lock();
68
69 if (!new_io) {
70 part = rq->part;
71 part_stat_inc(cpu, part, merges[rw]);
72 } else {
73 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
74 if (!hd_struct_try_get(part)) {
75 /*
76 * The partition is already being removed,
77 * the request will be accounted on the disk only
78 *
79 * We take a reference on disk->part0 although that
80 * partition will never be deleted, so we can treat
81 * it as any other partition.
82 */
83 part = &rq->rq_disk->part0;
84 hd_struct_get(part);
85 }
86 part_round_stats(cpu, part);
87 part_inc_in_flight(part, rw);
88 rq->part = part;
89 }
90
91 part_stat_unlock();
92}
93
94void blk_queue_congestion_threshold(struct request_queue *q)
95{
96 int nr;
97
98 nr = q->nr_requests - (q->nr_requests / 8) + 1;
99 if (nr > q->nr_requests)
100 nr = q->nr_requests;
101 q->nr_congestion_on = nr;
102
103 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
104 if (nr < 1)
105 nr = 1;
106 q->nr_congestion_off = nr;
107}
108
109/**
110 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
111 * @bdev: device
112 *
113 * Locates the passed device's request queue and returns the address of its
114 * backing_dev_info
115 *
116 * Will return NULL if the request queue cannot be located.
117 */
118struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
119{
120 struct backing_dev_info *ret = NULL;
121 struct request_queue *q = bdev_get_queue(bdev);
122
123 if (q)
124 ret = &q->backing_dev_info;
125 return ret;
126}
127EXPORT_SYMBOL(blk_get_backing_dev_info);
128
129void blk_rq_init(struct request_queue *q, struct request *rq)
130{
131 memset(rq, 0, sizeof(*rq));
132
133 INIT_LIST_HEAD(&rq->queuelist);
134 INIT_LIST_HEAD(&rq->timeout_list);
135 rq->cpu = -1;
136 rq->q = q;
137 rq->__sector = (sector_t) -1;
138 INIT_HLIST_NODE(&rq->hash);
139 RB_CLEAR_NODE(&rq->rb_node);
140 rq->cmd = rq->__cmd;
141 rq->cmd_len = BLK_MAX_CDB;
142 rq->tag = -1;
143 rq->ref_count = 1;
144 rq->start_time = jiffies;
145 set_start_time_ns(rq);
146 rq->part = NULL;
147}
148EXPORT_SYMBOL(blk_rq_init);
149
150static void req_bio_endio(struct request *rq, struct bio *bio,
151 unsigned int nbytes, int error)
152{
153 if (error)
154 clear_bit(BIO_UPTODATE, &bio->bi_flags);
155 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
156 error = -EIO;
157
158 if (unlikely(nbytes > bio->bi_size)) {
159 printk(KERN_ERR "%s: want %u bytes done, %u left\n",
160 __func__, nbytes, bio->bi_size);
161 nbytes = bio->bi_size;
162 }
163
164 if (unlikely(rq->cmd_flags & REQ_QUIET))
165 set_bit(BIO_QUIET, &bio->bi_flags);
166
167 bio->bi_size -= nbytes;
168 bio->bi_sector += (nbytes >> 9);
169
170 if (bio_integrity(bio))
171 bio_integrity_advance(bio, nbytes);
172
173 /* don't actually finish bio if it's part of flush sequence */
174 if (bio->bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
175 bio_endio(bio, error);
176}
177
178void blk_dump_rq_flags(struct request *rq, char *msg)
179{
180 int bit;
181
182 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
183 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
184 rq->cmd_flags);
185
186 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
187 (unsigned long long)blk_rq_pos(rq),
188 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
189 printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n",
190 rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq));
191
192 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
193 printk(KERN_INFO " cdb: ");
194 for (bit = 0; bit < BLK_MAX_CDB; bit++)
195 printk("%02x ", rq->cmd[bit]);
196 printk("\n");
197 }
198}
199EXPORT_SYMBOL(blk_dump_rq_flags);
200
201static void blk_delay_work(struct work_struct *work)
202{
203 struct request_queue *q;
204
205 q = container_of(work, struct request_queue, delay_work.work);
206 spin_lock_irq(q->queue_lock);
207 __blk_run_queue(q);
208 spin_unlock_irq(q->queue_lock);
209}
210
211/**
212 * blk_delay_queue - restart queueing after defined interval
213 * @q: The &struct request_queue in question
214 * @msecs: Delay in msecs
215 *
216 * Description:
217 * Sometimes queueing needs to be postponed for a little while, to allow
218 * resources to come back. This function will make sure that queueing is
219 * restarted around the specified time.
220 */
221void blk_delay_queue(struct request_queue *q, unsigned long msecs)
222{
223 queue_delayed_work(kblockd_workqueue, &q->delay_work,
224 msecs_to_jiffies(msecs));
225}
226EXPORT_SYMBOL(blk_delay_queue);
227
228/**
229 * blk_start_queue - restart a previously stopped queue
230 * @q: The &struct request_queue in question
231 *
232 * Description:
233 * blk_start_queue() will clear the stop flag on the queue, and call
234 * the request_fn for the queue if it was in a stopped state when
235 * entered. Also see blk_stop_queue(). Queue lock must be held.
236 **/
237void blk_start_queue(struct request_queue *q)
238{
239 WARN_ON(!irqs_disabled());
240
241 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
242 __blk_run_queue(q);
243}
244EXPORT_SYMBOL(blk_start_queue);
245
246/**
247 * blk_stop_queue - stop a queue
248 * @q: The &struct request_queue in question
249 *
250 * Description:
251 * The Linux block layer assumes that a block driver will consume all
252 * entries on the request queue when the request_fn strategy is called.
253 * Often this will not happen, because of hardware limitations (queue
254 * depth settings). If a device driver gets a 'queue full' response,
255 * or if it simply chooses not to queue more I/O at one point, it can
256 * call this function to prevent the request_fn from being called until
257 * the driver has signalled it's ready to go again. This happens by calling
258 * blk_start_queue() to restart queue operations. Queue lock must be held.
259 **/
260void blk_stop_queue(struct request_queue *q)
261{
262 __cancel_delayed_work(&q->delay_work);
263 queue_flag_set(QUEUE_FLAG_STOPPED, q);
264}
265EXPORT_SYMBOL(blk_stop_queue);
266
267/**
268 * blk_sync_queue - cancel any pending callbacks on a queue
269 * @q: the queue
270 *
271 * Description:
272 * The block layer may perform asynchronous callback activity
273 * on a queue, such as calling the unplug function after a timeout.
274 * A block device may call blk_sync_queue to ensure that any
275 * such activity is cancelled, thus allowing it to release resources
276 * that the callbacks might use. The caller must already have made sure
277 * that its ->make_request_fn will not re-add plugging prior to calling
278 * this function.
279 *
280 * This function does not cancel any asynchronous activity arising
281 * out of elevator or throttling code. That would require elevaotor_exit()
282 * and blk_throtl_exit() to be called with queue lock initialized.
283 *
284 */
285void blk_sync_queue(struct request_queue *q)
286{
287 del_timer_sync(&q->timeout);
288 cancel_delayed_work_sync(&q->delay_work);
289}
290EXPORT_SYMBOL(blk_sync_queue);
291
292/**
293 * __blk_run_queue - run a single device queue
294 * @q: The queue to run
295 *
296 * Description:
297 * See @blk_run_queue. This variant must be called with the queue lock
298 * held and interrupts disabled.
299 */
300void __blk_run_queue(struct request_queue *q)
301{
302 if (unlikely(blk_queue_stopped(q)))
303 return;
304
305 q->request_fn(q);
306}
307EXPORT_SYMBOL(__blk_run_queue);
308
309/**
310 * blk_run_queue_async - run a single device queue in workqueue context
311 * @q: The queue to run
312 *
313 * Description:
314 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
315 * of us.
316 */
317void blk_run_queue_async(struct request_queue *q)
318{
319 if (likely(!blk_queue_stopped(q))) {
320 __cancel_delayed_work(&q->delay_work);
321 queue_delayed_work(kblockd_workqueue, &q->delay_work, 0);
322 }
323}
324EXPORT_SYMBOL(blk_run_queue_async);
325
326/**
327 * blk_run_queue - run a single device queue
328 * @q: The queue to run
329 *
330 * Description:
331 * Invoke request handling on this queue, if it has pending work to do.
332 * May be used to restart queueing when a request has completed.
333 */
334void blk_run_queue(struct request_queue *q)
335{
336 unsigned long flags;
337
338 spin_lock_irqsave(q->queue_lock, flags);
339 __blk_run_queue(q);
340 spin_unlock_irqrestore(q->queue_lock, flags);
341}
342EXPORT_SYMBOL(blk_run_queue);
343
344void blk_put_queue(struct request_queue *q)
345{
346 kobject_put(&q->kobj);
347}
348EXPORT_SYMBOL(blk_put_queue);
349
350/*
351 * Note: If a driver supplied the queue lock, it is disconnected
352 * by this function. The actual state of the lock doesn't matter
353 * here as the request_queue isn't accessible after this point
354 * (QUEUE_FLAG_DEAD is set) and no other requests will be queued.
355 */
356void blk_cleanup_queue(struct request_queue *q)
357{
358 /*
359 * We know we have process context here, so we can be a little
360 * cautious and ensure that pending block actions on this device
361 * are done before moving on. Going into this function, we should
362 * not have processes doing IO to this device.
363 */
364 blk_sync_queue(q);
365
366 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
367 mutex_lock(&q->sysfs_lock);
368 queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q);
369 mutex_unlock(&q->sysfs_lock);
370
371 if (q->queue_lock != &q->__queue_lock)
372 q->queue_lock = &q->__queue_lock;
373
374 blk_put_queue(q);
375}
376EXPORT_SYMBOL(blk_cleanup_queue);
377
378static int blk_init_free_list(struct request_queue *q)
379{
380 struct request_list *rl = &q->rq;
381
382 if (unlikely(rl->rq_pool))
383 return 0;
384
385 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
386 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
387 rl->elvpriv = 0;
388 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
389 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
390
391 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
392 mempool_free_slab, request_cachep, q->node);
393
394 if (!rl->rq_pool)
395 return -ENOMEM;
396
397 return 0;
398}
399
400struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
401{
402 return blk_alloc_queue_node(gfp_mask, -1);
403}
404EXPORT_SYMBOL(blk_alloc_queue);
405
406struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
407{
408 struct request_queue *q;
409 int err;
410
411 q = kmem_cache_alloc_node(blk_requestq_cachep,
412 gfp_mask | __GFP_ZERO, node_id);
413 if (!q)
414 return NULL;
415
416 q->backing_dev_info.ra_pages =
417 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
418 q->backing_dev_info.state = 0;
419 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
420 q->backing_dev_info.name = "block";
421
422 err = bdi_init(&q->backing_dev_info);
423 if (err) {
424 kmem_cache_free(blk_requestq_cachep, q);
425 return NULL;
426 }
427
428 if (blk_throtl_init(q)) {
429 kmem_cache_free(blk_requestq_cachep, q);
430 return NULL;
431 }
432
433 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
434 laptop_mode_timer_fn, (unsigned long) q);
435 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
436 INIT_LIST_HEAD(&q->timeout_list);
437 INIT_LIST_HEAD(&q->flush_queue[0]);
438 INIT_LIST_HEAD(&q->flush_queue[1]);
439 INIT_LIST_HEAD(&q->flush_data_in_flight);
440 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
441
442 kobject_init(&q->kobj, &blk_queue_ktype);
443
444 mutex_init(&q->sysfs_lock);
445 spin_lock_init(&q->__queue_lock);
446
447 /*
448 * By default initialize queue_lock to internal lock and driver can
449 * override it later if need be.
450 */
451 q->queue_lock = &q->__queue_lock;
452
453 return q;
454}
455EXPORT_SYMBOL(blk_alloc_queue_node);
456
457/**
458 * blk_init_queue - prepare a request queue for use with a block device
459 * @rfn: The function to be called to process requests that have been
460 * placed on the queue.
461 * @lock: Request queue spin lock
462 *
463 * Description:
464 * If a block device wishes to use the standard request handling procedures,
465 * which sorts requests and coalesces adjacent requests, then it must
466 * call blk_init_queue(). The function @rfn will be called when there
467 * are requests on the queue that need to be processed. If the device
468 * supports plugging, then @rfn may not be called immediately when requests
469 * are available on the queue, but may be called at some time later instead.
470 * Plugged queues are generally unplugged when a buffer belonging to one
471 * of the requests on the queue is needed, or due to memory pressure.
472 *
473 * @rfn is not required, or even expected, to remove all requests off the
474 * queue, but only as many as it can handle at a time. If it does leave
475 * requests on the queue, it is responsible for arranging that the requests
476 * get dealt with eventually.
477 *
478 * The queue spin lock must be held while manipulating the requests on the
479 * request queue; this lock will be taken also from interrupt context, so irq
480 * disabling is needed for it.
481 *
482 * Function returns a pointer to the initialized request queue, or %NULL if
483 * it didn't succeed.
484 *
485 * Note:
486 * blk_init_queue() must be paired with a blk_cleanup_queue() call
487 * when the block device is deactivated (such as at module unload).
488 **/
489
490struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
491{
492 return blk_init_queue_node(rfn, lock, -1);
493}
494EXPORT_SYMBOL(blk_init_queue);
495
496struct request_queue *
497blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
498{
499 struct request_queue *uninit_q, *q;
500
501 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
502 if (!uninit_q)
503 return NULL;
504
505 q = blk_init_allocated_queue_node(uninit_q, rfn, lock, node_id);
506 if (!q)
507 blk_cleanup_queue(uninit_q);
508
509 return q;
510}
511EXPORT_SYMBOL(blk_init_queue_node);
512
513struct request_queue *
514blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
515 spinlock_t *lock)
516{
517 return blk_init_allocated_queue_node(q, rfn, lock, -1);
518}
519EXPORT_SYMBOL(blk_init_allocated_queue);
520
521struct request_queue *
522blk_init_allocated_queue_node(struct request_queue *q, request_fn_proc *rfn,
523 spinlock_t *lock, int node_id)
524{
525 if (!q)
526 return NULL;
527
528 q->node = node_id;
529 if (blk_init_free_list(q))
530 return NULL;
531
532 q->request_fn = rfn;
533 q->prep_rq_fn = NULL;
534 q->unprep_rq_fn = NULL;
535 q->queue_flags = QUEUE_FLAG_DEFAULT;
536
537 /* Override internal queue lock with supplied lock pointer */
538 if (lock)
539 q->queue_lock = lock;
540
541 /*
542 * This also sets hw/phys segments, boundary and size
543 */
544 blk_queue_make_request(q, __make_request);
545
546 q->sg_reserved_size = INT_MAX;
547
548 /*
549 * all done
550 */
551 if (!elevator_init(q, NULL)) {
552 blk_queue_congestion_threshold(q);
553 return q;
554 }
555
556 return NULL;
557}
558EXPORT_SYMBOL(blk_init_allocated_queue_node);
559
560int blk_get_queue(struct request_queue *q)
561{
562 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
563 kobject_get(&q->kobj);
564 return 0;
565 }
566
567 return 1;
568}
569EXPORT_SYMBOL(blk_get_queue);
570
571static inline void blk_free_request(struct request_queue *q, struct request *rq)
572{
573 if (rq->cmd_flags & REQ_ELVPRIV)
574 elv_put_request(q, rq);
575 mempool_free(rq, q->rq.rq_pool);
576}
577
578static struct request *
579blk_alloc_request(struct request_queue *q, int flags, int priv, gfp_t gfp_mask)
580{
581 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
582
583 if (!rq)
584 return NULL;
585
586 blk_rq_init(q, rq);
587
588 rq->cmd_flags = flags | REQ_ALLOCED;
589
590 if (priv) {
591 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
592 mempool_free(rq, q->rq.rq_pool);
593 return NULL;
594 }
595 rq->cmd_flags |= REQ_ELVPRIV;
596 }
597
598 return rq;
599}
600
601/*
602 * ioc_batching returns true if the ioc is a valid batching request and
603 * should be given priority access to a request.
604 */
605static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
606{
607 if (!ioc)
608 return 0;
609
610 /*
611 * Make sure the process is able to allocate at least 1 request
612 * even if the batch times out, otherwise we could theoretically
613 * lose wakeups.
614 */
615 return ioc->nr_batch_requests == q->nr_batching ||
616 (ioc->nr_batch_requests > 0
617 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
618}
619
620/*
621 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
622 * will cause the process to be a "batcher" on all queues in the system. This
623 * is the behaviour we want though - once it gets a wakeup it should be given
624 * a nice run.
625 */
626static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
627{
628 if (!ioc || ioc_batching(q, ioc))
629 return;
630
631 ioc->nr_batch_requests = q->nr_batching;
632 ioc->last_waited = jiffies;
633}
634
635static void __freed_request(struct request_queue *q, int sync)
636{
637 struct request_list *rl = &q->rq;
638
639 if (rl->count[sync] < queue_congestion_off_threshold(q))
640 blk_clear_queue_congested(q, sync);
641
642 if (rl->count[sync] + 1 <= q->nr_requests) {
643 if (waitqueue_active(&rl->wait[sync]))
644 wake_up(&rl->wait[sync]);
645
646 blk_clear_queue_full(q, sync);
647 }
648}
649
650/*
651 * A request has just been released. Account for it, update the full and
652 * congestion status, wake up any waiters. Called under q->queue_lock.
653 */
654static void freed_request(struct request_queue *q, int sync, int priv)
655{
656 struct request_list *rl = &q->rq;
657
658 rl->count[sync]--;
659 if (priv)
660 rl->elvpriv--;
661
662 __freed_request(q, sync);
663
664 if (unlikely(rl->starved[sync ^ 1]))
665 __freed_request(q, sync ^ 1);
666}
667
668/*
669 * Determine if elevator data should be initialized when allocating the
670 * request associated with @bio.
671 */
672static bool blk_rq_should_init_elevator(struct bio *bio)
673{
674 if (!bio)
675 return true;
676
677 /*
678 * Flush requests do not use the elevator so skip initialization.
679 * This allows a request to share the flush and elevator data.
680 */
681 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
682 return false;
683
684 return true;
685}
686
687/*
688 * Get a free request, queue_lock must be held.
689 * Returns NULL on failure, with queue_lock held.
690 * Returns !NULL on success, with queue_lock *not held*.
691 */
692static struct request *get_request(struct request_queue *q, int rw_flags,
693 struct bio *bio, gfp_t gfp_mask)
694{
695 struct request *rq = NULL;
696 struct request_list *rl = &q->rq;
697 struct io_context *ioc = NULL;
698 const bool is_sync = rw_is_sync(rw_flags) != 0;
699 int may_queue, priv = 0;
700
701 may_queue = elv_may_queue(q, rw_flags);
702 if (may_queue == ELV_MQUEUE_NO)
703 goto rq_starved;
704
705 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
706 if (rl->count[is_sync]+1 >= q->nr_requests) {
707 ioc = current_io_context(GFP_ATOMIC, q->node);
708 /*
709 * The queue will fill after this allocation, so set
710 * it as full, and mark this process as "batching".
711 * This process will be allowed to complete a batch of
712 * requests, others will be blocked.
713 */
714 if (!blk_queue_full(q, is_sync)) {
715 ioc_set_batching(q, ioc);
716 blk_set_queue_full(q, is_sync);
717 } else {
718 if (may_queue != ELV_MQUEUE_MUST
719 && !ioc_batching(q, ioc)) {
720 /*
721 * The queue is full and the allocating
722 * process is not a "batcher", and not
723 * exempted by the IO scheduler
724 */
725 goto out;
726 }
727 }
728 }
729 blk_set_queue_congested(q, is_sync);
730 }
731
732 /*
733 * Only allow batching queuers to allocate up to 50% over the defined
734 * limit of requests, otherwise we could have thousands of requests
735 * allocated with any setting of ->nr_requests
736 */
737 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
738 goto out;
739
740 rl->count[is_sync]++;
741 rl->starved[is_sync] = 0;
742
743 if (blk_rq_should_init_elevator(bio)) {
744 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
745 if (priv)
746 rl->elvpriv++;
747 }
748
749 if (blk_queue_io_stat(q))
750 rw_flags |= REQ_IO_STAT;
751 spin_unlock_irq(q->queue_lock);
752
753 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
754 if (unlikely(!rq)) {
755 /*
756 * Allocation failed presumably due to memory. Undo anything
757 * we might have messed up.
758 *
759 * Allocating task should really be put onto the front of the
760 * wait queue, but this is pretty rare.
761 */
762 spin_lock_irq(q->queue_lock);
763 freed_request(q, is_sync, priv);
764
765 /*
766 * in the very unlikely event that allocation failed and no
767 * requests for this direction was pending, mark us starved
768 * so that freeing of a request in the other direction will
769 * notice us. another possible fix would be to split the
770 * rq mempool into READ and WRITE
771 */
772rq_starved:
773 if (unlikely(rl->count[is_sync] == 0))
774 rl->starved[is_sync] = 1;
775
776 goto out;
777 }
778
779 /*
780 * ioc may be NULL here, and ioc_batching will be false. That's
781 * OK, if the queue is under the request limit then requests need
782 * not count toward the nr_batch_requests limit. There will always
783 * be some limit enforced by BLK_BATCH_TIME.
784 */
785 if (ioc_batching(q, ioc))
786 ioc->nr_batch_requests--;
787
788 trace_block_getrq(q, bio, rw_flags & 1);
789out:
790 return rq;
791}
792
793/*
794 * No available requests for this queue, wait for some requests to become
795 * available.
796 *
797 * Called with q->queue_lock held, and returns with it unlocked.
798 */
799static struct request *get_request_wait(struct request_queue *q, int rw_flags,
800 struct bio *bio)
801{
802 const bool is_sync = rw_is_sync(rw_flags) != 0;
803 struct request *rq;
804
805 rq = get_request(q, rw_flags, bio, GFP_NOIO);
806 while (!rq) {
807 DEFINE_WAIT(wait);
808 struct io_context *ioc;
809 struct request_list *rl = &q->rq;
810
811 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
812 TASK_UNINTERRUPTIBLE);
813
814 trace_block_sleeprq(q, bio, rw_flags & 1);
815
816 spin_unlock_irq(q->queue_lock);
817 io_schedule();
818
819 /*
820 * After sleeping, we become a "batching" process and
821 * will be able to allocate at least one request, and
822 * up to a big batch of them for a small period time.
823 * See ioc_batching, ioc_set_batching
824 */
825 ioc = current_io_context(GFP_NOIO, q->node);
826 ioc_set_batching(q, ioc);
827
828 spin_lock_irq(q->queue_lock);
829 finish_wait(&rl->wait[is_sync], &wait);
830
831 rq = get_request(q, rw_flags, bio, GFP_NOIO);
832 };
833
834 return rq;
835}
836
837struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
838{
839 struct request *rq;
840
841 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
842 return NULL;
843
844 BUG_ON(rw != READ && rw != WRITE);
845
846 spin_lock_irq(q->queue_lock);
847 if (gfp_mask & __GFP_WAIT) {
848 rq = get_request_wait(q, rw, NULL);
849 } else {
850 rq = get_request(q, rw, NULL, gfp_mask);
851 if (!rq)
852 spin_unlock_irq(q->queue_lock);
853 }
854 /* q->queue_lock is unlocked at this point */
855
856 return rq;
857}
858EXPORT_SYMBOL(blk_get_request);
859
860/**
861 * blk_make_request - given a bio, allocate a corresponding struct request.
862 * @q: target request queue
863 * @bio: The bio describing the memory mappings that will be submitted for IO.
864 * It may be a chained-bio properly constructed by block/bio layer.
865 * @gfp_mask: gfp flags to be used for memory allocation
866 *
867 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
868 * type commands. Where the struct request needs to be farther initialized by
869 * the caller. It is passed a &struct bio, which describes the memory info of
870 * the I/O transfer.
871 *
872 * The caller of blk_make_request must make sure that bi_io_vec
873 * are set to describe the memory buffers. That bio_data_dir() will return
874 * the needed direction of the request. (And all bio's in the passed bio-chain
875 * are properly set accordingly)
876 *
877 * If called under none-sleepable conditions, mapped bio buffers must not
878 * need bouncing, by calling the appropriate masked or flagged allocator,
879 * suitable for the target device. Otherwise the call to blk_queue_bounce will
880 * BUG.
881 *
882 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
883 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
884 * anything but the first bio in the chain. Otherwise you risk waiting for IO
885 * completion of a bio that hasn't been submitted yet, thus resulting in a
886 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
887 * of bio_alloc(), as that avoids the mempool deadlock.
888 * If possible a big IO should be split into smaller parts when allocation
889 * fails. Partial allocation should not be an error, or you risk a live-lock.
890 */
891struct request *blk_make_request(struct request_queue *q, struct bio *bio,
892 gfp_t gfp_mask)
893{
894 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
895
896 if (unlikely(!rq))
897 return ERR_PTR(-ENOMEM);
898
899 for_each_bio(bio) {
900 struct bio *bounce_bio = bio;
901 int ret;
902
903 blk_queue_bounce(q, &bounce_bio);
904 ret = blk_rq_append_bio(q, rq, bounce_bio);
905 if (unlikely(ret)) {
906 blk_put_request(rq);
907 return ERR_PTR(ret);
908 }
909 }
910
911 return rq;
912}
913EXPORT_SYMBOL(blk_make_request);
914
915/**
916 * blk_requeue_request - put a request back on queue
917 * @q: request queue where request should be inserted
918 * @rq: request to be inserted
919 *
920 * Description:
921 * Drivers often keep queueing requests until the hardware cannot accept
922 * more, when that condition happens we need to put the request back
923 * on the queue. Must be called with queue lock held.
924 */
925void blk_requeue_request(struct request_queue *q, struct request *rq)
926{
927 blk_delete_timer(rq);
928 blk_clear_rq_complete(rq);
929 trace_block_rq_requeue(q, rq);
930
931 if (blk_rq_tagged(rq))
932 blk_queue_end_tag(q, rq);
933
934 BUG_ON(blk_queued_rq(rq));
935
936 elv_requeue_request(q, rq);
937}
938EXPORT_SYMBOL(blk_requeue_request);
939
940static void add_acct_request(struct request_queue *q, struct request *rq,
941 int where)
942{
943 drive_stat_acct(rq, 1);
944 __elv_add_request(q, rq, where);
945}
946
947/**
948 * blk_insert_request - insert a special request into a request queue
949 * @q: request queue where request should be inserted
950 * @rq: request to be inserted
951 * @at_head: insert request at head or tail of queue
952 * @data: private data
953 *
954 * Description:
955 * Many block devices need to execute commands asynchronously, so they don't
956 * block the whole kernel from preemption during request execution. This is
957 * accomplished normally by inserting aritficial requests tagged as
958 * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them
959 * be scheduled for actual execution by the request queue.
960 *
961 * We have the option of inserting the head or the tail of the queue.
962 * Typically we use the tail for new ioctls and so forth. We use the head
963 * of the queue for things like a QUEUE_FULL message from a device, or a
964 * host that is unable to accept a particular command.
965 */
966void blk_insert_request(struct request_queue *q, struct request *rq,
967 int at_head, void *data)
968{
969 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
970 unsigned long flags;
971
972 /*
973 * tell I/O scheduler that this isn't a regular read/write (ie it
974 * must not attempt merges on this) and that it acts as a soft
975 * barrier
976 */
977 rq->cmd_type = REQ_TYPE_SPECIAL;
978
979 rq->special = data;
980
981 spin_lock_irqsave(q->queue_lock, flags);
982
983 /*
984 * If command is tagged, release the tag
985 */
986 if (blk_rq_tagged(rq))
987 blk_queue_end_tag(q, rq);
988
989 add_acct_request(q, rq, where);
990 __blk_run_queue(q);
991 spin_unlock_irqrestore(q->queue_lock, flags);
992}
993EXPORT_SYMBOL(blk_insert_request);
994
995static void part_round_stats_single(int cpu, struct hd_struct *part,
996 unsigned long now)
997{
998 if (now == part->stamp)
999 return;
1000
1001 if (part_in_flight(part)) {
1002 __part_stat_add(cpu, part, time_in_queue,
1003 part_in_flight(part) * (now - part->stamp));
1004 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1005 }
1006 part->stamp = now;
1007}
1008
1009/**
1010 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1011 * @cpu: cpu number for stats access
1012 * @part: target partition
1013 *
1014 * The average IO queue length and utilisation statistics are maintained
1015 * by observing the current state of the queue length and the amount of
1016 * time it has been in this state for.
1017 *
1018 * Normally, that accounting is done on IO completion, but that can result
1019 * in more than a second's worth of IO being accounted for within any one
1020 * second, leading to >100% utilisation. To deal with that, we call this
1021 * function to do a round-off before returning the results when reading
1022 * /proc/diskstats. This accounts immediately for all queue usage up to
1023 * the current jiffies and restarts the counters again.
1024 */
1025void part_round_stats(int cpu, struct hd_struct *part)
1026{
1027 unsigned long now = jiffies;
1028
1029 if (part->partno)
1030 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1031 part_round_stats_single(cpu, part, now);
1032}
1033EXPORT_SYMBOL_GPL(part_round_stats);
1034
1035/*
1036 * queue lock must be held
1037 */
1038void __blk_put_request(struct request_queue *q, struct request *req)
1039{
1040 if (unlikely(!q))
1041 return;
1042 if (unlikely(--req->ref_count))
1043 return;
1044
1045 elv_completed_request(q, req);
1046
1047 /* this is a bio leak */
1048 WARN_ON(req->bio != NULL);
1049
1050 /*
1051 * Request may not have originated from ll_rw_blk. if not,
1052 * it didn't come out of our reserved rq pools
1053 */
1054 if (req->cmd_flags & REQ_ALLOCED) {
1055 int is_sync = rq_is_sync(req) != 0;
1056 int priv = req->cmd_flags & REQ_ELVPRIV;
1057
1058 BUG_ON(!list_empty(&req->queuelist));
1059 BUG_ON(!hlist_unhashed(&req->hash));
1060
1061 blk_free_request(q, req);
1062 freed_request(q, is_sync, priv);
1063 }
1064}
1065EXPORT_SYMBOL_GPL(__blk_put_request);
1066
1067void blk_put_request(struct request *req)
1068{
1069 unsigned long flags;
1070 struct request_queue *q = req->q;
1071
1072 spin_lock_irqsave(q->queue_lock, flags);
1073 __blk_put_request(q, req);
1074 spin_unlock_irqrestore(q->queue_lock, flags);
1075}
1076EXPORT_SYMBOL(blk_put_request);
1077
1078/**
1079 * blk_add_request_payload - add a payload to a request
1080 * @rq: request to update
1081 * @page: page backing the payload
1082 * @len: length of the payload.
1083 *
1084 * This allows to later add a payload to an already submitted request by
1085 * a block driver. The driver needs to take care of freeing the payload
1086 * itself.
1087 *
1088 * Note that this is a quite horrible hack and nothing but handling of
1089 * discard requests should ever use it.
1090 */
1091void blk_add_request_payload(struct request *rq, struct page *page,
1092 unsigned int len)
1093{
1094 struct bio *bio = rq->bio;
1095
1096 bio->bi_io_vec->bv_page = page;
1097 bio->bi_io_vec->bv_offset = 0;
1098 bio->bi_io_vec->bv_len = len;
1099
1100 bio->bi_size = len;
1101 bio->bi_vcnt = 1;
1102 bio->bi_phys_segments = 1;
1103
1104 rq->__data_len = rq->resid_len = len;
1105 rq->nr_phys_segments = 1;
1106 rq->buffer = bio_data(bio);
1107}
1108EXPORT_SYMBOL_GPL(blk_add_request_payload);
1109
1110static bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1111 struct bio *bio)
1112{
1113 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1114
1115 if (!ll_back_merge_fn(q, req, bio))
1116 return false;
1117
1118 trace_block_bio_backmerge(q, bio);
1119
1120 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1121 blk_rq_set_mixed_merge(req);
1122
1123 req->biotail->bi_next = bio;
1124 req->biotail = bio;
1125 req->__data_len += bio->bi_size;
1126 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1127
1128 drive_stat_acct(req, 0);
1129 elv_bio_merged(q, req, bio);
1130 return true;
1131}
1132
1133static bool bio_attempt_front_merge(struct request_queue *q,
1134 struct request *req, struct bio *bio)
1135{
1136 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1137
1138 if (!ll_front_merge_fn(q, req, bio))
1139 return false;
1140
1141 trace_block_bio_frontmerge(q, bio);
1142
1143 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1144 blk_rq_set_mixed_merge(req);
1145
1146 bio->bi_next = req->bio;
1147 req->bio = bio;
1148
1149 /*
1150 * may not be valid. if the low level driver said
1151 * it didn't need a bounce buffer then it better
1152 * not touch req->buffer either...
1153 */
1154 req->buffer = bio_data(bio);
1155 req->__sector = bio->bi_sector;
1156 req->__data_len += bio->bi_size;
1157 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1158
1159 drive_stat_acct(req, 0);
1160 elv_bio_merged(q, req, bio);
1161 return true;
1162}
1163
1164/*
1165 * Attempts to merge with the plugged list in the current process. Returns
1166 * true if merge was successful, otherwise false.
1167 */
1168static bool attempt_plug_merge(struct task_struct *tsk, struct request_queue *q,
1169 struct bio *bio, unsigned int *request_count)
1170{
1171 struct blk_plug *plug;
1172 struct request *rq;
1173 bool ret = false;
1174
1175 plug = tsk->plug;
1176 if (!plug)
1177 goto out;
1178 *request_count = 0;
1179
1180 list_for_each_entry_reverse(rq, &plug->list, queuelist) {
1181 int el_ret;
1182
1183 (*request_count)++;
1184
1185 if (rq->q != q)
1186 continue;
1187
1188 el_ret = elv_try_merge(rq, bio);
1189 if (el_ret == ELEVATOR_BACK_MERGE) {
1190 ret = bio_attempt_back_merge(q, rq, bio);
1191 if (ret)
1192 break;
1193 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1194 ret = bio_attempt_front_merge(q, rq, bio);
1195 if (ret)
1196 break;
1197 }
1198 }
1199out:
1200 return ret;
1201}
1202
1203void init_request_from_bio(struct request *req, struct bio *bio)
1204{
1205 req->cpu = bio->bi_comp_cpu;
1206 req->cmd_type = REQ_TYPE_FS;
1207
1208 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1209 if (bio->bi_rw & REQ_RAHEAD)
1210 req->cmd_flags |= REQ_FAILFAST_MASK;
1211
1212 req->errors = 0;
1213 req->__sector = bio->bi_sector;
1214 req->ioprio = bio_prio(bio);
1215 blk_rq_bio_prep(req->q, req, bio);
1216}
1217
1218static int __make_request(struct request_queue *q, struct bio *bio)
1219{
1220 const bool sync = !!(bio->bi_rw & REQ_SYNC);
1221 struct blk_plug *plug;
1222 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1223 struct request *req;
1224 unsigned int request_count = 0;
1225
1226 /*
1227 * low level driver can indicate that it wants pages above a
1228 * certain limit bounced to low memory (ie for highmem, or even
1229 * ISA dma in theory)
1230 */
1231 blk_queue_bounce(q, &bio);
1232
1233 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
1234 spin_lock_irq(q->queue_lock);
1235 where = ELEVATOR_INSERT_FLUSH;
1236 goto get_rq;
1237 }
1238
1239 /*
1240 * Check if we can merge with the plugged list before grabbing
1241 * any locks.
1242 */
1243 if (attempt_plug_merge(current, q, bio, &request_count))
1244 goto out;
1245
1246 spin_lock_irq(q->queue_lock);
1247
1248 el_ret = elv_merge(q, &req, bio);
1249 if (el_ret == ELEVATOR_BACK_MERGE) {
1250 if (bio_attempt_back_merge(q, req, bio)) {
1251 if (!attempt_back_merge(q, req))
1252 elv_merged_request(q, req, el_ret);
1253 goto out_unlock;
1254 }
1255 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1256 if (bio_attempt_front_merge(q, req, bio)) {
1257 if (!attempt_front_merge(q, req))
1258 elv_merged_request(q, req, el_ret);
1259 goto out_unlock;
1260 }
1261 }
1262
1263get_rq:
1264 /*
1265 * This sync check and mask will be re-done in init_request_from_bio(),
1266 * but we need to set it earlier to expose the sync flag to the
1267 * rq allocator and io schedulers.
1268 */
1269 rw_flags = bio_data_dir(bio);
1270 if (sync)
1271 rw_flags |= REQ_SYNC;
1272
1273 /*
1274 * Grab a free request. This is might sleep but can not fail.
1275 * Returns with the queue unlocked.
1276 */
1277 req = get_request_wait(q, rw_flags, bio);
1278
1279 /*
1280 * After dropping the lock and possibly sleeping here, our request
1281 * may now be mergeable after it had proven unmergeable (above).
1282 * We don't worry about that case for efficiency. It won't happen
1283 * often, and the elevators are able to handle it.
1284 */
1285 init_request_from_bio(req, bio);
1286
1287 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) ||
1288 bio_flagged(bio, BIO_CPU_AFFINE))
1289 req->cpu = raw_smp_processor_id();
1290
1291 plug = current->plug;
1292 if (plug) {
1293 /*
1294 * If this is the first request added after a plug, fire
1295 * of a plug trace. If others have been added before, check
1296 * if we have multiple devices in this plug. If so, make a
1297 * note to sort the list before dispatch.
1298 */
1299 if (list_empty(&plug->list))
1300 trace_block_plug(q);
1301 else if (!plug->should_sort) {
1302 struct request *__rq;
1303
1304 __rq = list_entry_rq(plug->list.prev);
1305 if (__rq->q != q)
1306 plug->should_sort = 1;
1307 }
1308 if (request_count >= BLK_MAX_REQUEST_COUNT)
1309 blk_flush_plug_list(plug, false);
1310 list_add_tail(&req->queuelist, &plug->list);
1311 drive_stat_acct(req, 1);
1312 } else {
1313 spin_lock_irq(q->queue_lock);
1314 add_acct_request(q, req, where);
1315 __blk_run_queue(q);
1316out_unlock:
1317 spin_unlock_irq(q->queue_lock);
1318 }
1319out:
1320 return 0;
1321}
1322
1323/*
1324 * If bio->bi_dev is a partition, remap the location
1325 */
1326static inline void blk_partition_remap(struct bio *bio)
1327{
1328 struct block_device *bdev = bio->bi_bdev;
1329
1330 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1331 struct hd_struct *p = bdev->bd_part;
1332
1333 bio->bi_sector += p->start_sect;
1334 bio->bi_bdev = bdev->bd_contains;
1335
1336 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1337 bdev->bd_dev,
1338 bio->bi_sector - p->start_sect);
1339 }
1340}
1341
1342static void handle_bad_sector(struct bio *bio)
1343{
1344 char b[BDEVNAME_SIZE];
1345
1346 printk(KERN_INFO "attempt to access beyond end of device\n");
1347 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1348 bdevname(bio->bi_bdev, b),
1349 bio->bi_rw,
1350 (unsigned long long)bio->bi_sector + bio_sectors(bio),
1351 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1352
1353 set_bit(BIO_EOF, &bio->bi_flags);
1354}
1355
1356#ifdef CONFIG_FAIL_MAKE_REQUEST
1357
1358static DECLARE_FAULT_ATTR(fail_make_request);
1359
1360static int __init setup_fail_make_request(char *str)
1361{
1362 return setup_fault_attr(&fail_make_request, str);
1363}
1364__setup("fail_make_request=", setup_fail_make_request);
1365
1366static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1367{
1368 return part->make_it_fail && should_fail(&fail_make_request, bytes);
1369}
1370
1371static int __init fail_make_request_debugfs(void)
1372{
1373 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1374 NULL, &fail_make_request);
1375
1376 return IS_ERR(dir) ? PTR_ERR(dir) : 0;
1377}
1378
1379late_initcall(fail_make_request_debugfs);
1380
1381#else /* CONFIG_FAIL_MAKE_REQUEST */
1382
1383static inline bool should_fail_request(struct hd_struct *part,
1384 unsigned int bytes)
1385{
1386 return false;
1387}
1388
1389#endif /* CONFIG_FAIL_MAKE_REQUEST */
1390
1391/*
1392 * Check whether this bio extends beyond the end of the device.
1393 */
1394static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1395{
1396 sector_t maxsector;
1397
1398 if (!nr_sectors)
1399 return 0;
1400
1401 /* Test device or partition size, when known. */
1402 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1403 if (maxsector) {
1404 sector_t sector = bio->bi_sector;
1405
1406 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1407 /*
1408 * This may well happen - the kernel calls bread()
1409 * without checking the size of the device, e.g., when
1410 * mounting a device.
1411 */
1412 handle_bad_sector(bio);
1413 return 1;
1414 }
1415 }
1416
1417 return 0;
1418}
1419
1420/**
1421 * generic_make_request - hand a buffer to its device driver for I/O
1422 * @bio: The bio describing the location in memory and on the device.
1423 *
1424 * generic_make_request() is used to make I/O requests of block
1425 * devices. It is passed a &struct bio, which describes the I/O that needs
1426 * to be done.
1427 *
1428 * generic_make_request() does not return any status. The
1429 * success/failure status of the request, along with notification of
1430 * completion, is delivered asynchronously through the bio->bi_end_io
1431 * function described (one day) else where.
1432 *
1433 * The caller of generic_make_request must make sure that bi_io_vec
1434 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1435 * set to describe the device address, and the
1436 * bi_end_io and optionally bi_private are set to describe how
1437 * completion notification should be signaled.
1438 *
1439 * generic_make_request and the drivers it calls may use bi_next if this
1440 * bio happens to be merged with someone else, and may change bi_dev and
1441 * bi_sector for remaps as it sees fit. So the values of these fields
1442 * should NOT be depended on after the call to generic_make_request.
1443 */
1444static inline void __generic_make_request(struct bio *bio)
1445{
1446 struct request_queue *q;
1447 sector_t old_sector;
1448 int ret, nr_sectors = bio_sectors(bio);
1449 dev_t old_dev;
1450 int err = -EIO;
1451
1452 might_sleep();
1453
1454 if (bio_check_eod(bio, nr_sectors))
1455 goto end_io;
1456
1457 /*
1458 * Resolve the mapping until finished. (drivers are
1459 * still free to implement/resolve their own stacking
1460 * by explicitly returning 0)
1461 *
1462 * NOTE: we don't repeat the blk_size check for each new device.
1463 * Stacking drivers are expected to know what they are doing.
1464 */
1465 old_sector = -1;
1466 old_dev = 0;
1467 do {
1468 char b[BDEVNAME_SIZE];
1469 struct hd_struct *part;
1470
1471 q = bdev_get_queue(bio->bi_bdev);
1472 if (unlikely(!q)) {
1473 printk(KERN_ERR
1474 "generic_make_request: Trying to access "
1475 "nonexistent block-device %s (%Lu)\n",
1476 bdevname(bio->bi_bdev, b),
1477 (long long) bio->bi_sector);
1478 goto end_io;
1479 }
1480
1481 if (unlikely(!(bio->bi_rw & REQ_DISCARD) &&
1482 nr_sectors > queue_max_hw_sectors(q))) {
1483 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1484 bdevname(bio->bi_bdev, b),
1485 bio_sectors(bio),
1486 queue_max_hw_sectors(q));
1487 goto end_io;
1488 }
1489
1490 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
1491 goto end_io;
1492
1493 part = bio->bi_bdev->bd_part;
1494 if (should_fail_request(part, bio->bi_size) ||
1495 should_fail_request(&part_to_disk(part)->part0,
1496 bio->bi_size))
1497 goto end_io;
1498
1499 /*
1500 * If this device has partitions, remap block n
1501 * of partition p to block n+start(p) of the disk.
1502 */
1503 blk_partition_remap(bio);
1504
1505 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1506 goto end_io;
1507
1508 if (old_sector != -1)
1509 trace_block_bio_remap(q, bio, old_dev, old_sector);
1510
1511 old_sector = bio->bi_sector;
1512 old_dev = bio->bi_bdev->bd_dev;
1513
1514 if (bio_check_eod(bio, nr_sectors))
1515 goto end_io;
1516
1517 /*
1518 * Filter flush bio's early so that make_request based
1519 * drivers without flush support don't have to worry
1520 * about them.
1521 */
1522 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1523 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1524 if (!nr_sectors) {
1525 err = 0;
1526 goto end_io;
1527 }
1528 }
1529
1530 if ((bio->bi_rw & REQ_DISCARD) &&
1531 (!blk_queue_discard(q) ||
1532 ((bio->bi_rw & REQ_SECURE) &&
1533 !blk_queue_secdiscard(q)))) {
1534 err = -EOPNOTSUPP;
1535 goto end_io;
1536 }
1537
1538 if (blk_throtl_bio(q, &bio))
1539 goto end_io;
1540
1541 /*
1542 * If bio = NULL, bio has been throttled and will be submitted
1543 * later.
1544 */
1545 if (!bio)
1546 break;
1547
1548 trace_block_bio_queue(q, bio);
1549
1550 ret = q->make_request_fn(q, bio);
1551 } while (ret);
1552
1553 return;
1554
1555end_io:
1556 bio_endio(bio, err);
1557}
1558
1559/*
1560 * We only want one ->make_request_fn to be active at a time,
1561 * else stack usage with stacked devices could be a problem.
1562 * So use current->bio_list to keep a list of requests
1563 * submited by a make_request_fn function.
1564 * current->bio_list is also used as a flag to say if
1565 * generic_make_request is currently active in this task or not.
1566 * If it is NULL, then no make_request is active. If it is non-NULL,
1567 * then a make_request is active, and new requests should be added
1568 * at the tail
1569 */
1570void generic_make_request(struct bio *bio)
1571{
1572 struct bio_list bio_list_on_stack;
1573
1574 if (current->bio_list) {
1575 /* make_request is active */
1576 bio_list_add(current->bio_list, bio);
1577 return;
1578 }
1579 /* following loop may be a bit non-obvious, and so deserves some
1580 * explanation.
1581 * Before entering the loop, bio->bi_next is NULL (as all callers
1582 * ensure that) so we have a list with a single bio.
1583 * We pretend that we have just taken it off a longer list, so
1584 * we assign bio_list to a pointer to the bio_list_on_stack,
1585 * thus initialising the bio_list of new bios to be
1586 * added. __generic_make_request may indeed add some more bios
1587 * through a recursive call to generic_make_request. If it
1588 * did, we find a non-NULL value in bio_list and re-enter the loop
1589 * from the top. In this case we really did just take the bio
1590 * of the top of the list (no pretending) and so remove it from
1591 * bio_list, and call into __generic_make_request again.
1592 *
1593 * The loop was structured like this to make only one call to
1594 * __generic_make_request (which is important as it is large and
1595 * inlined) and to keep the structure simple.
1596 */
1597 BUG_ON(bio->bi_next);
1598 bio_list_init(&bio_list_on_stack);
1599 current->bio_list = &bio_list_on_stack;
1600 do {
1601 __generic_make_request(bio);
1602 bio = bio_list_pop(current->bio_list);
1603 } while (bio);
1604 current->bio_list = NULL; /* deactivate */
1605}
1606EXPORT_SYMBOL(generic_make_request);
1607
1608/**
1609 * submit_bio - submit a bio to the block device layer for I/O
1610 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1611 * @bio: The &struct bio which describes the I/O
1612 *
1613 * submit_bio() is very similar in purpose to generic_make_request(), and
1614 * uses that function to do most of the work. Both are fairly rough
1615 * interfaces; @bio must be presetup and ready for I/O.
1616 *
1617 */
1618void submit_bio(int rw, struct bio *bio)
1619{
1620 int count = bio_sectors(bio);
1621
1622 bio->bi_rw |= rw;
1623
1624 /*
1625 * If it's a regular read/write or a barrier with data attached,
1626 * go through the normal accounting stuff before submission.
1627 */
1628 if (bio_has_data(bio) && !(rw & REQ_DISCARD)) {
1629 if (rw & WRITE) {
1630 count_vm_events(PGPGOUT, count);
1631 } else {
1632 task_io_account_read(bio->bi_size);
1633 count_vm_events(PGPGIN, count);
1634 }
1635
1636 if (unlikely(block_dump)) {
1637 char b[BDEVNAME_SIZE];
1638 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1639 current->comm, task_pid_nr(current),
1640 (rw & WRITE) ? "WRITE" : "READ",
1641 (unsigned long long)bio->bi_sector,
1642 bdevname(bio->bi_bdev, b),
1643 count);
1644 }
1645 }
1646
1647 generic_make_request(bio);
1648}
1649EXPORT_SYMBOL(submit_bio);
1650
1651/**
1652 * blk_rq_check_limits - Helper function to check a request for the queue limit
1653 * @q: the queue
1654 * @rq: the request being checked
1655 *
1656 * Description:
1657 * @rq may have been made based on weaker limitations of upper-level queues
1658 * in request stacking drivers, and it may violate the limitation of @q.
1659 * Since the block layer and the underlying device driver trust @rq
1660 * after it is inserted to @q, it should be checked against @q before
1661 * the insertion using this generic function.
1662 *
1663 * This function should also be useful for request stacking drivers
1664 * in some cases below, so export this function.
1665 * Request stacking drivers like request-based dm may change the queue
1666 * limits while requests are in the queue (e.g. dm's table swapping).
1667 * Such request stacking drivers should check those requests agaist
1668 * the new queue limits again when they dispatch those requests,
1669 * although such checkings are also done against the old queue limits
1670 * when submitting requests.
1671 */
1672int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1673{
1674 if (rq->cmd_flags & REQ_DISCARD)
1675 return 0;
1676
1677 if (blk_rq_sectors(rq) > queue_max_sectors(q) ||
1678 blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) {
1679 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1680 return -EIO;
1681 }
1682
1683 /*
1684 * queue's settings related to segment counting like q->bounce_pfn
1685 * may differ from that of other stacking queues.
1686 * Recalculate it to check the request correctly on this queue's
1687 * limitation.
1688 */
1689 blk_recalc_rq_segments(rq);
1690 if (rq->nr_phys_segments > queue_max_segments(q)) {
1691 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1692 return -EIO;
1693 }
1694
1695 return 0;
1696}
1697EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1698
1699/**
1700 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1701 * @q: the queue to submit the request
1702 * @rq: the request being queued
1703 */
1704int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1705{
1706 unsigned long flags;
1707 int where = ELEVATOR_INSERT_BACK;
1708
1709 if (blk_rq_check_limits(q, rq))
1710 return -EIO;
1711
1712 if (rq->rq_disk &&
1713 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
1714 return -EIO;
1715
1716 spin_lock_irqsave(q->queue_lock, flags);
1717
1718 /*
1719 * Submitting request must be dequeued before calling this function
1720 * because it will be linked to another request_queue
1721 */
1722 BUG_ON(blk_queued_rq(rq));
1723
1724 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
1725 where = ELEVATOR_INSERT_FLUSH;
1726
1727 add_acct_request(q, rq, where);
1728 spin_unlock_irqrestore(q->queue_lock, flags);
1729
1730 return 0;
1731}
1732EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1733
1734/**
1735 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1736 * @rq: request to examine
1737 *
1738 * Description:
1739 * A request could be merge of IOs which require different failure
1740 * handling. This function determines the number of bytes which
1741 * can be failed from the beginning of the request without
1742 * crossing into area which need to be retried further.
1743 *
1744 * Return:
1745 * The number of bytes to fail.
1746 *
1747 * Context:
1748 * queue_lock must be held.
1749 */
1750unsigned int blk_rq_err_bytes(const struct request *rq)
1751{
1752 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1753 unsigned int bytes = 0;
1754 struct bio *bio;
1755
1756 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1757 return blk_rq_bytes(rq);
1758
1759 /*
1760 * Currently the only 'mixing' which can happen is between
1761 * different fastfail types. We can safely fail portions
1762 * which have all the failfast bits that the first one has -
1763 * the ones which are at least as eager to fail as the first
1764 * one.
1765 */
1766 for (bio = rq->bio; bio; bio = bio->bi_next) {
1767 if ((bio->bi_rw & ff) != ff)
1768 break;
1769 bytes += bio->bi_size;
1770 }
1771
1772 /* this could lead to infinite loop */
1773 BUG_ON(blk_rq_bytes(rq) && !bytes);
1774 return bytes;
1775}
1776EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1777
1778static void blk_account_io_completion(struct request *req, unsigned int bytes)
1779{
1780 if (blk_do_io_stat(req)) {
1781 const int rw = rq_data_dir(req);
1782 struct hd_struct *part;
1783 int cpu;
1784
1785 cpu = part_stat_lock();
1786 part = req->part;
1787 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
1788 part_stat_unlock();
1789 }
1790}
1791
1792static void blk_account_io_done(struct request *req)
1793{
1794 /*
1795 * Account IO completion. flush_rq isn't accounted as a
1796 * normal IO on queueing nor completion. Accounting the
1797 * containing request is enough.
1798 */
1799 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
1800 unsigned long duration = jiffies - req->start_time;
1801 const int rw = rq_data_dir(req);
1802 struct hd_struct *part;
1803 int cpu;
1804
1805 cpu = part_stat_lock();
1806 part = req->part;
1807
1808 part_stat_inc(cpu, part, ios[rw]);
1809 part_stat_add(cpu, part, ticks[rw], duration);
1810 part_round_stats(cpu, part);
1811 part_dec_in_flight(part, rw);
1812
1813 hd_struct_put(part);
1814 part_stat_unlock();
1815 }
1816}
1817
1818/**
1819 * blk_peek_request - peek at the top of a request queue
1820 * @q: request queue to peek at
1821 *
1822 * Description:
1823 * Return the request at the top of @q. The returned request
1824 * should be started using blk_start_request() before LLD starts
1825 * processing it.
1826 *
1827 * Return:
1828 * Pointer to the request at the top of @q if available. Null
1829 * otherwise.
1830 *
1831 * Context:
1832 * queue_lock must be held.
1833 */
1834struct request *blk_peek_request(struct request_queue *q)
1835{
1836 struct request *rq;
1837 int ret;
1838
1839 while ((rq = __elv_next_request(q)) != NULL) {
1840 if (!(rq->cmd_flags & REQ_STARTED)) {
1841 /*
1842 * This is the first time the device driver
1843 * sees this request (possibly after
1844 * requeueing). Notify IO scheduler.
1845 */
1846 if (rq->cmd_flags & REQ_SORTED)
1847 elv_activate_rq(q, rq);
1848
1849 /*
1850 * just mark as started even if we don't start
1851 * it, a request that has been delayed should
1852 * not be passed by new incoming requests
1853 */
1854 rq->cmd_flags |= REQ_STARTED;
1855 trace_block_rq_issue(q, rq);
1856 }
1857
1858 if (!q->boundary_rq || q->boundary_rq == rq) {
1859 q->end_sector = rq_end_sector(rq);
1860 q->boundary_rq = NULL;
1861 }
1862
1863 if (rq->cmd_flags & REQ_DONTPREP)
1864 break;
1865
1866 if (q->dma_drain_size && blk_rq_bytes(rq)) {
1867 /*
1868 * make sure space for the drain appears we
1869 * know we can do this because max_hw_segments
1870 * has been adjusted to be one fewer than the
1871 * device can handle
1872 */
1873 rq->nr_phys_segments++;
1874 }
1875
1876 if (!q->prep_rq_fn)
1877 break;
1878
1879 ret = q->prep_rq_fn(q, rq);
1880 if (ret == BLKPREP_OK) {
1881 break;
1882 } else if (ret == BLKPREP_DEFER) {
1883 /*
1884 * the request may have been (partially) prepped.
1885 * we need to keep this request in the front to
1886 * avoid resource deadlock. REQ_STARTED will
1887 * prevent other fs requests from passing this one.
1888 */
1889 if (q->dma_drain_size && blk_rq_bytes(rq) &&
1890 !(rq->cmd_flags & REQ_DONTPREP)) {
1891 /*
1892 * remove the space for the drain we added
1893 * so that we don't add it again
1894 */
1895 --rq->nr_phys_segments;
1896 }
1897
1898 rq = NULL;
1899 break;
1900 } else if (ret == BLKPREP_KILL) {
1901 rq->cmd_flags |= REQ_QUIET;
1902 /*
1903 * Mark this request as started so we don't trigger
1904 * any debug logic in the end I/O path.
1905 */
1906 blk_start_request(rq);
1907 __blk_end_request_all(rq, -EIO);
1908 } else {
1909 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
1910 break;
1911 }
1912 }
1913
1914 return rq;
1915}
1916EXPORT_SYMBOL(blk_peek_request);
1917
1918void blk_dequeue_request(struct request *rq)
1919{
1920 struct request_queue *q = rq->q;
1921
1922 BUG_ON(list_empty(&rq->queuelist));
1923 BUG_ON(ELV_ON_HASH(rq));
1924
1925 list_del_init(&rq->queuelist);
1926
1927 /*
1928 * the time frame between a request being removed from the lists
1929 * and to it is freed is accounted as io that is in progress at
1930 * the driver side.
1931 */
1932 if (blk_account_rq(rq)) {
1933 q->in_flight[rq_is_sync(rq)]++;
1934 set_io_start_time_ns(rq);
1935 }
1936}
1937
1938/**
1939 * blk_start_request - start request processing on the driver
1940 * @req: request to dequeue
1941 *
1942 * Description:
1943 * Dequeue @req and start timeout timer on it. This hands off the
1944 * request to the driver.
1945 *
1946 * Block internal functions which don't want to start timer should
1947 * call blk_dequeue_request().
1948 *
1949 * Context:
1950 * queue_lock must be held.
1951 */
1952void blk_start_request(struct request *req)
1953{
1954 blk_dequeue_request(req);
1955
1956 /*
1957 * We are now handing the request to the hardware, initialize
1958 * resid_len to full count and add the timeout handler.
1959 */
1960 req->resid_len = blk_rq_bytes(req);
1961 if (unlikely(blk_bidi_rq(req)))
1962 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
1963
1964 blk_add_timer(req);
1965}
1966EXPORT_SYMBOL(blk_start_request);
1967
1968/**
1969 * blk_fetch_request - fetch a request from a request queue
1970 * @q: request queue to fetch a request from
1971 *
1972 * Description:
1973 * Return the request at the top of @q. The request is started on
1974 * return and LLD can start processing it immediately.
1975 *
1976 * Return:
1977 * Pointer to the request at the top of @q if available. Null
1978 * otherwise.
1979 *
1980 * Context:
1981 * queue_lock must be held.
1982 */
1983struct request *blk_fetch_request(struct request_queue *q)
1984{
1985 struct request *rq;
1986
1987 rq = blk_peek_request(q);
1988 if (rq)
1989 blk_start_request(rq);
1990 return rq;
1991}
1992EXPORT_SYMBOL(blk_fetch_request);
1993
1994/**
1995 * blk_update_request - Special helper function for request stacking drivers
1996 * @req: the request being processed
1997 * @error: %0 for success, < %0 for error
1998 * @nr_bytes: number of bytes to complete @req
1999 *
2000 * Description:
2001 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2002 * the request structure even if @req doesn't have leftover.
2003 * If @req has leftover, sets it up for the next range of segments.
2004 *
2005 * This special helper function is only for request stacking drivers
2006 * (e.g. request-based dm) so that they can handle partial completion.
2007 * Actual device drivers should use blk_end_request instead.
2008 *
2009 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2010 * %false return from this function.
2011 *
2012 * Return:
2013 * %false - this request doesn't have any more data
2014 * %true - this request has more data
2015 **/
2016bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2017{
2018 int total_bytes, bio_nbytes, next_idx = 0;
2019 struct bio *bio;
2020
2021 if (!req->bio)
2022 return false;
2023
2024 trace_block_rq_complete(req->q, req);
2025
2026 /*
2027 * For fs requests, rq is just carrier of independent bio's
2028 * and each partial completion should be handled separately.
2029 * Reset per-request error on each partial completion.
2030 *
2031 * TODO: tj: This is too subtle. It would be better to let
2032 * low level drivers do what they see fit.
2033 */
2034 if (req->cmd_type == REQ_TYPE_FS)
2035 req->errors = 0;
2036
2037 if (error && req->cmd_type == REQ_TYPE_FS &&
2038 !(req->cmd_flags & REQ_QUIET)) {
2039 char *error_type;
2040
2041 switch (error) {
2042 case -ENOLINK:
2043 error_type = "recoverable transport";
2044 break;
2045 case -EREMOTEIO:
2046 error_type = "critical target";
2047 break;
2048 case -EBADE:
2049 error_type = "critical nexus";
2050 break;
2051 case -EIO:
2052 default:
2053 error_type = "I/O";
2054 break;
2055 }
2056 printk(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2057 error_type, req->rq_disk ? req->rq_disk->disk_name : "?",
2058 (unsigned long long)blk_rq_pos(req));
2059 }
2060
2061 blk_account_io_completion(req, nr_bytes);
2062
2063 total_bytes = bio_nbytes = 0;
2064 while ((bio = req->bio) != NULL) {
2065 int nbytes;
2066
2067 if (nr_bytes >= bio->bi_size) {
2068 req->bio = bio->bi_next;
2069 nbytes = bio->bi_size;
2070 req_bio_endio(req, bio, nbytes, error);
2071 next_idx = 0;
2072 bio_nbytes = 0;
2073 } else {
2074 int idx = bio->bi_idx + next_idx;
2075
2076 if (unlikely(idx >= bio->bi_vcnt)) {
2077 blk_dump_rq_flags(req, "__end_that");
2078 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
2079 __func__, idx, bio->bi_vcnt);
2080 break;
2081 }
2082
2083 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2084 BIO_BUG_ON(nbytes > bio->bi_size);
2085
2086 /*
2087 * not a complete bvec done
2088 */
2089 if (unlikely(nbytes > nr_bytes)) {
2090 bio_nbytes += nr_bytes;
2091 total_bytes += nr_bytes;
2092 break;
2093 }
2094
2095 /*
2096 * advance to the next vector
2097 */
2098 next_idx++;
2099 bio_nbytes += nbytes;
2100 }
2101
2102 total_bytes += nbytes;
2103 nr_bytes -= nbytes;
2104
2105 bio = req->bio;
2106 if (bio) {
2107 /*
2108 * end more in this run, or just return 'not-done'
2109 */
2110 if (unlikely(nr_bytes <= 0))
2111 break;
2112 }
2113 }
2114
2115 /*
2116 * completely done
2117 */
2118 if (!req->bio) {
2119 /*
2120 * Reset counters so that the request stacking driver
2121 * can find how many bytes remain in the request
2122 * later.
2123 */
2124 req->__data_len = 0;
2125 return false;
2126 }
2127
2128 /*
2129 * if the request wasn't completed, update state
2130 */
2131 if (bio_nbytes) {
2132 req_bio_endio(req, bio, bio_nbytes, error);
2133 bio->bi_idx += next_idx;
2134 bio_iovec(bio)->bv_offset += nr_bytes;
2135 bio_iovec(bio)->bv_len -= nr_bytes;
2136 }
2137
2138 req->__data_len -= total_bytes;
2139 req->buffer = bio_data(req->bio);
2140
2141 /* update sector only for requests with clear definition of sector */
2142 if (req->cmd_type == REQ_TYPE_FS || (req->cmd_flags & REQ_DISCARD))
2143 req->__sector += total_bytes >> 9;
2144
2145 /* mixed attributes always follow the first bio */
2146 if (req->cmd_flags & REQ_MIXED_MERGE) {
2147 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2148 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2149 }
2150
2151 /*
2152 * If total number of sectors is less than the first segment
2153 * size, something has gone terribly wrong.
2154 */
2155 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2156 blk_dump_rq_flags(req, "request botched");
2157 req->__data_len = blk_rq_cur_bytes(req);
2158 }
2159
2160 /* recalculate the number of segments */
2161 blk_recalc_rq_segments(req);
2162
2163 return true;
2164}
2165EXPORT_SYMBOL_GPL(blk_update_request);
2166
2167static bool blk_update_bidi_request(struct request *rq, int error,
2168 unsigned int nr_bytes,
2169 unsigned int bidi_bytes)
2170{
2171 if (blk_update_request(rq, error, nr_bytes))
2172 return true;
2173
2174 /* Bidi request must be completed as a whole */
2175 if (unlikely(blk_bidi_rq(rq)) &&
2176 blk_update_request(rq->next_rq, error, bidi_bytes))
2177 return true;
2178
2179 if (blk_queue_add_random(rq->q))
2180 add_disk_randomness(rq->rq_disk);
2181
2182 return false;
2183}
2184
2185/**
2186 * blk_unprep_request - unprepare a request
2187 * @req: the request
2188 *
2189 * This function makes a request ready for complete resubmission (or
2190 * completion). It happens only after all error handling is complete,
2191 * so represents the appropriate moment to deallocate any resources
2192 * that were allocated to the request in the prep_rq_fn. The queue
2193 * lock is held when calling this.
2194 */
2195void blk_unprep_request(struct request *req)
2196{
2197 struct request_queue *q = req->q;
2198
2199 req->cmd_flags &= ~REQ_DONTPREP;
2200 if (q->unprep_rq_fn)
2201 q->unprep_rq_fn(q, req);
2202}
2203EXPORT_SYMBOL_GPL(blk_unprep_request);
2204
2205/*
2206 * queue lock must be held
2207 */
2208static void blk_finish_request(struct request *req, int error)
2209{
2210 if (blk_rq_tagged(req))
2211 blk_queue_end_tag(req->q, req);
2212
2213 BUG_ON(blk_queued_rq(req));
2214
2215 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2216 laptop_io_completion(&req->q->backing_dev_info);
2217
2218 blk_delete_timer(req);
2219
2220 if (req->cmd_flags & REQ_DONTPREP)
2221 blk_unprep_request(req);
2222
2223
2224 blk_account_io_done(req);
2225
2226 if (req->end_io)
2227 req->end_io(req, error);
2228 else {
2229 if (blk_bidi_rq(req))
2230 __blk_put_request(req->next_rq->q, req->next_rq);
2231
2232 __blk_put_request(req->q, req);
2233 }
2234}
2235
2236/**
2237 * blk_end_bidi_request - Complete a bidi request
2238 * @rq: the request to complete
2239 * @error: %0 for success, < %0 for error
2240 * @nr_bytes: number of bytes to complete @rq
2241 * @bidi_bytes: number of bytes to complete @rq->next_rq
2242 *
2243 * Description:
2244 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2245 * Drivers that supports bidi can safely call this member for any
2246 * type of request, bidi or uni. In the later case @bidi_bytes is
2247 * just ignored.
2248 *
2249 * Return:
2250 * %false - we are done with this request
2251 * %true - still buffers pending for this request
2252 **/
2253static bool blk_end_bidi_request(struct request *rq, int error,
2254 unsigned int nr_bytes, unsigned int bidi_bytes)
2255{
2256 struct request_queue *q = rq->q;
2257 unsigned long flags;
2258
2259 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2260 return true;
2261
2262 spin_lock_irqsave(q->queue_lock, flags);
2263 blk_finish_request(rq, error);
2264 spin_unlock_irqrestore(q->queue_lock, flags);
2265
2266 return false;
2267}
2268
2269/**
2270 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2271 * @rq: the request to complete
2272 * @error: %0 for success, < %0 for error
2273 * @nr_bytes: number of bytes to complete @rq
2274 * @bidi_bytes: number of bytes to complete @rq->next_rq
2275 *
2276 * Description:
2277 * Identical to blk_end_bidi_request() except that queue lock is
2278 * assumed to be locked on entry and remains so on return.
2279 *
2280 * Return:
2281 * %false - we are done with this request
2282 * %true - still buffers pending for this request
2283 **/
2284bool __blk_end_bidi_request(struct request *rq, int error,
2285 unsigned int nr_bytes, unsigned int bidi_bytes)
2286{
2287 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2288 return true;
2289
2290 blk_finish_request(rq, error);
2291
2292 return false;
2293}
2294
2295/**
2296 * blk_end_request - Helper function for drivers to complete the request.
2297 * @rq: the request being processed
2298 * @error: %0 for success, < %0 for error
2299 * @nr_bytes: number of bytes to complete
2300 *
2301 * Description:
2302 * Ends I/O on a number of bytes attached to @rq.
2303 * If @rq has leftover, sets it up for the next range of segments.
2304 *
2305 * Return:
2306 * %false - we are done with this request
2307 * %true - still buffers pending for this request
2308 **/
2309bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2310{
2311 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2312}
2313EXPORT_SYMBOL(blk_end_request);
2314
2315/**
2316 * blk_end_request_all - Helper function for drives to finish the request.
2317 * @rq: the request to finish
2318 * @error: %0 for success, < %0 for error
2319 *
2320 * Description:
2321 * Completely finish @rq.
2322 */
2323void blk_end_request_all(struct request *rq, int error)
2324{
2325 bool pending;
2326 unsigned int bidi_bytes = 0;
2327
2328 if (unlikely(blk_bidi_rq(rq)))
2329 bidi_bytes = blk_rq_bytes(rq->next_rq);
2330
2331 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2332 BUG_ON(pending);
2333}
2334EXPORT_SYMBOL(blk_end_request_all);
2335
2336/**
2337 * blk_end_request_cur - Helper function to finish the current request chunk.
2338 * @rq: the request to finish the current chunk for
2339 * @error: %0 for success, < %0 for error
2340 *
2341 * Description:
2342 * Complete the current consecutively mapped chunk from @rq.
2343 *
2344 * Return:
2345 * %false - we are done with this request
2346 * %true - still buffers pending for this request
2347 */
2348bool blk_end_request_cur(struct request *rq, int error)
2349{
2350 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2351}
2352EXPORT_SYMBOL(blk_end_request_cur);
2353
2354/**
2355 * blk_end_request_err - Finish a request till the next failure boundary.
2356 * @rq: the request to finish till the next failure boundary for
2357 * @error: must be negative errno
2358 *
2359 * Description:
2360 * Complete @rq till the next failure boundary.
2361 *
2362 * Return:
2363 * %false - we are done with this request
2364 * %true - still buffers pending for this request
2365 */
2366bool blk_end_request_err(struct request *rq, int error)
2367{
2368 WARN_ON(error >= 0);
2369 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2370}
2371EXPORT_SYMBOL_GPL(blk_end_request_err);
2372
2373/**
2374 * __blk_end_request - Helper function for drivers to complete the request.
2375 * @rq: the request being processed
2376 * @error: %0 for success, < %0 for error
2377 * @nr_bytes: number of bytes to complete
2378 *
2379 * Description:
2380 * Must be called with queue lock held unlike blk_end_request().
2381 *
2382 * Return:
2383 * %false - we are done with this request
2384 * %true - still buffers pending for this request
2385 **/
2386bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2387{
2388 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2389}
2390EXPORT_SYMBOL(__blk_end_request);
2391
2392/**
2393 * __blk_end_request_all - Helper function for drives to finish the request.
2394 * @rq: the request to finish
2395 * @error: %0 for success, < %0 for error
2396 *
2397 * Description:
2398 * Completely finish @rq. Must be called with queue lock held.
2399 */
2400void __blk_end_request_all(struct request *rq, int error)
2401{
2402 bool pending;
2403 unsigned int bidi_bytes = 0;
2404
2405 if (unlikely(blk_bidi_rq(rq)))
2406 bidi_bytes = blk_rq_bytes(rq->next_rq);
2407
2408 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2409 BUG_ON(pending);
2410}
2411EXPORT_SYMBOL(__blk_end_request_all);
2412
2413/**
2414 * __blk_end_request_cur - Helper function to finish the current request chunk.
2415 * @rq: the request to finish the current chunk for
2416 * @error: %0 for success, < %0 for error
2417 *
2418 * Description:
2419 * Complete the current consecutively mapped chunk from @rq. Must
2420 * be called with queue lock held.
2421 *
2422 * Return:
2423 * %false - we are done with this request
2424 * %true - still buffers pending for this request
2425 */
2426bool __blk_end_request_cur(struct request *rq, int error)
2427{
2428 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2429}
2430EXPORT_SYMBOL(__blk_end_request_cur);
2431
2432/**
2433 * __blk_end_request_err - Finish a request till the next failure boundary.
2434 * @rq: the request to finish till the next failure boundary for
2435 * @error: must be negative errno
2436 *
2437 * Description:
2438 * Complete @rq till the next failure boundary. Must be called
2439 * with queue lock held.
2440 *
2441 * Return:
2442 * %false - we are done with this request
2443 * %true - still buffers pending for this request
2444 */
2445bool __blk_end_request_err(struct request *rq, int error)
2446{
2447 WARN_ON(error >= 0);
2448 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2449}
2450EXPORT_SYMBOL_GPL(__blk_end_request_err);
2451
2452void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2453 struct bio *bio)
2454{
2455 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2456 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
2457
2458 if (bio_has_data(bio)) {
2459 rq->nr_phys_segments = bio_phys_segments(q, bio);
2460 rq->buffer = bio_data(bio);
2461 }
2462 rq->__data_len = bio->bi_size;
2463 rq->bio = rq->biotail = bio;
2464
2465 if (bio->bi_bdev)
2466 rq->rq_disk = bio->bi_bdev->bd_disk;
2467}
2468
2469#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2470/**
2471 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2472 * @rq: the request to be flushed
2473 *
2474 * Description:
2475 * Flush all pages in @rq.
2476 */
2477void rq_flush_dcache_pages(struct request *rq)
2478{
2479 struct req_iterator iter;
2480 struct bio_vec *bvec;
2481
2482 rq_for_each_segment(bvec, rq, iter)
2483 flush_dcache_page(bvec->bv_page);
2484}
2485EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2486#endif
2487
2488/**
2489 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2490 * @q : the queue of the device being checked
2491 *
2492 * Description:
2493 * Check if underlying low-level drivers of a device are busy.
2494 * If the drivers want to export their busy state, they must set own
2495 * exporting function using blk_queue_lld_busy() first.
2496 *
2497 * Basically, this function is used only by request stacking drivers
2498 * to stop dispatching requests to underlying devices when underlying
2499 * devices are busy. This behavior helps more I/O merging on the queue
2500 * of the request stacking driver and prevents I/O throughput regression
2501 * on burst I/O load.
2502 *
2503 * Return:
2504 * 0 - Not busy (The request stacking driver should dispatch request)
2505 * 1 - Busy (The request stacking driver should stop dispatching request)
2506 */
2507int blk_lld_busy(struct request_queue *q)
2508{
2509 if (q->lld_busy_fn)
2510 return q->lld_busy_fn(q);
2511
2512 return 0;
2513}
2514EXPORT_SYMBOL_GPL(blk_lld_busy);
2515
2516/**
2517 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2518 * @rq: the clone request to be cleaned up
2519 *
2520 * Description:
2521 * Free all bios in @rq for a cloned request.
2522 */
2523void blk_rq_unprep_clone(struct request *rq)
2524{
2525 struct bio *bio;
2526
2527 while ((bio = rq->bio) != NULL) {
2528 rq->bio = bio->bi_next;
2529
2530 bio_put(bio);
2531 }
2532}
2533EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2534
2535/*
2536 * Copy attributes of the original request to the clone request.
2537 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2538 */
2539static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2540{
2541 dst->cpu = src->cpu;
2542 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
2543 dst->cmd_type = src->cmd_type;
2544 dst->__sector = blk_rq_pos(src);
2545 dst->__data_len = blk_rq_bytes(src);
2546 dst->nr_phys_segments = src->nr_phys_segments;
2547 dst->ioprio = src->ioprio;
2548 dst->extra_len = src->extra_len;
2549}
2550
2551/**
2552 * blk_rq_prep_clone - Helper function to setup clone request
2553 * @rq: the request to be setup
2554 * @rq_src: original request to be cloned
2555 * @bs: bio_set that bios for clone are allocated from
2556 * @gfp_mask: memory allocation mask for bio
2557 * @bio_ctr: setup function to be called for each clone bio.
2558 * Returns %0 for success, non %0 for failure.
2559 * @data: private data to be passed to @bio_ctr
2560 *
2561 * Description:
2562 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2563 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2564 * are not copied, and copying such parts is the caller's responsibility.
2565 * Also, pages which the original bios are pointing to are not copied
2566 * and the cloned bios just point same pages.
2567 * So cloned bios must be completed before original bios, which means
2568 * the caller must complete @rq before @rq_src.
2569 */
2570int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2571 struct bio_set *bs, gfp_t gfp_mask,
2572 int (*bio_ctr)(struct bio *, struct bio *, void *),
2573 void *data)
2574{
2575 struct bio *bio, *bio_src;
2576
2577 if (!bs)
2578 bs = fs_bio_set;
2579
2580 blk_rq_init(NULL, rq);
2581
2582 __rq_for_each_bio(bio_src, rq_src) {
2583 bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs);
2584 if (!bio)
2585 goto free_and_out;
2586
2587 __bio_clone(bio, bio_src);
2588
2589 if (bio_integrity(bio_src) &&
2590 bio_integrity_clone(bio, bio_src, gfp_mask, bs))
2591 goto free_and_out;
2592
2593 if (bio_ctr && bio_ctr(bio, bio_src, data))
2594 goto free_and_out;
2595
2596 if (rq->bio) {
2597 rq->biotail->bi_next = bio;
2598 rq->biotail = bio;
2599 } else
2600 rq->bio = rq->biotail = bio;
2601 }
2602
2603 __blk_rq_prep_clone(rq, rq_src);
2604
2605 return 0;
2606
2607free_and_out:
2608 if (bio)
2609 bio_free(bio, bs);
2610 blk_rq_unprep_clone(rq);
2611
2612 return -ENOMEM;
2613}
2614EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2615
2616int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
2617{
2618 return queue_work(kblockd_workqueue, work);
2619}
2620EXPORT_SYMBOL(kblockd_schedule_work);
2621
2622int kblockd_schedule_delayed_work(struct request_queue *q,
2623 struct delayed_work *dwork, unsigned long delay)
2624{
2625 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2626}
2627EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2628
2629#define PLUG_MAGIC 0x91827364
2630
2631void blk_start_plug(struct blk_plug *plug)
2632{
2633 struct task_struct *tsk = current;
2634
2635 plug->magic = PLUG_MAGIC;
2636 INIT_LIST_HEAD(&plug->list);
2637 INIT_LIST_HEAD(&plug->cb_list);
2638 plug->should_sort = 0;
2639
2640 /*
2641 * If this is a nested plug, don't actually assign it. It will be
2642 * flushed on its own.
2643 */
2644 if (!tsk->plug) {
2645 /*
2646 * Store ordering should not be needed here, since a potential
2647 * preempt will imply a full memory barrier
2648 */
2649 tsk->plug = plug;
2650 }
2651}
2652EXPORT_SYMBOL(blk_start_plug);
2653
2654static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2655{
2656 struct request *rqa = container_of(a, struct request, queuelist);
2657 struct request *rqb = container_of(b, struct request, queuelist);
2658
2659 return !(rqa->q <= rqb->q);
2660}
2661
2662/*
2663 * If 'from_schedule' is true, then postpone the dispatch of requests
2664 * until a safe kblockd context. We due this to avoid accidental big
2665 * additional stack usage in driver dispatch, in places where the originally
2666 * plugger did not intend it.
2667 */
2668static void queue_unplugged(struct request_queue *q, unsigned int depth,
2669 bool from_schedule)
2670 __releases(q->queue_lock)
2671{
2672 trace_block_unplug(q, depth, !from_schedule);
2673
2674 /*
2675 * If we are punting this to kblockd, then we can safely drop
2676 * the queue_lock before waking kblockd (which needs to take
2677 * this lock).
2678 */
2679 if (from_schedule) {
2680 spin_unlock(q->queue_lock);
2681 blk_run_queue_async(q);
2682 } else {
2683 __blk_run_queue(q);
2684 spin_unlock(q->queue_lock);
2685 }
2686
2687}
2688
2689static void flush_plug_callbacks(struct blk_plug *plug)
2690{
2691 LIST_HEAD(callbacks);
2692
2693 if (list_empty(&plug->cb_list))
2694 return;
2695
2696 list_splice_init(&plug->cb_list, &callbacks);
2697
2698 while (!list_empty(&callbacks)) {
2699 struct blk_plug_cb *cb = list_first_entry(&callbacks,
2700 struct blk_plug_cb,
2701 list);
2702 list_del(&cb->list);
2703 cb->callback(cb);
2704 }
2705}
2706
2707void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
2708{
2709 struct request_queue *q;
2710 unsigned long flags;
2711 struct request *rq;
2712 LIST_HEAD(list);
2713 unsigned int depth;
2714
2715 BUG_ON(plug->magic != PLUG_MAGIC);
2716
2717 flush_plug_callbacks(plug);
2718 if (list_empty(&plug->list))
2719 return;
2720
2721 list_splice_init(&plug->list, &list);
2722
2723 if (plug->should_sort) {
2724 list_sort(NULL, &list, plug_rq_cmp);
2725 plug->should_sort = 0;
2726 }
2727
2728 q = NULL;
2729 depth = 0;
2730
2731 /*
2732 * Save and disable interrupts here, to avoid doing it for every
2733 * queue lock we have to take.
2734 */
2735 local_irq_save(flags);
2736 while (!list_empty(&list)) {
2737 rq = list_entry_rq(list.next);
2738 list_del_init(&rq->queuelist);
2739 BUG_ON(!rq->q);
2740 if (rq->q != q) {
2741 /*
2742 * This drops the queue lock
2743 */
2744 if (q)
2745 queue_unplugged(q, depth, from_schedule);
2746 q = rq->q;
2747 depth = 0;
2748 spin_lock(q->queue_lock);
2749 }
2750 /*
2751 * rq is already accounted, so use raw insert
2752 */
2753 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
2754 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
2755 else
2756 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
2757
2758 depth++;
2759 }
2760
2761 /*
2762 * This drops the queue lock
2763 */
2764 if (q)
2765 queue_unplugged(q, depth, from_schedule);
2766
2767 local_irq_restore(flags);
2768}
2769
2770void blk_finish_plug(struct blk_plug *plug)
2771{
2772 blk_flush_plug_list(plug, false);
2773
2774 if (plug == current->plug)
2775 current->plug = NULL;
2776}
2777EXPORT_SYMBOL(blk_finish_plug);
2778
2779int __init blk_dev_init(void)
2780{
2781 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
2782 sizeof(((struct request *)0)->cmd_flags));
2783
2784 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
2785 kblockd_workqueue = alloc_workqueue("kblockd",
2786 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
2787 if (!kblockd_workqueue)
2788 panic("Failed to create kblockd\n");
2789
2790 request_cachep = kmem_cache_create("blkdev_requests",
2791 sizeof(struct request), 0, SLAB_PANIC, NULL);
2792
2793 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
2794 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
2795
2796 return 0;
2797}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) 1991, 1992 Linus Torvalds
4 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
5 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
6 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
7 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
8 * - July2000
9 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
10 */
11
12/*
13 * This handles all read/write requests to block devices
14 */
15#include <linux/kernel.h>
16#include <linux/module.h>
17#include <linux/bio.h>
18#include <linux/blkdev.h>
19#include <linux/blk-pm.h>
20#include <linux/blk-integrity.h>
21#include <linux/highmem.h>
22#include <linux/mm.h>
23#include <linux/pagemap.h>
24#include <linux/kernel_stat.h>
25#include <linux/string.h>
26#include <linux/init.h>
27#include <linux/completion.h>
28#include <linux/slab.h>
29#include <linux/swap.h>
30#include <linux/writeback.h>
31#include <linux/task_io_accounting_ops.h>
32#include <linux/fault-inject.h>
33#include <linux/list_sort.h>
34#include <linux/delay.h>
35#include <linux/ratelimit.h>
36#include <linux/pm_runtime.h>
37#include <linux/t10-pi.h>
38#include <linux/debugfs.h>
39#include <linux/bpf.h>
40#include <linux/part_stat.h>
41#include <linux/sched/sysctl.h>
42#include <linux/blk-crypto.h>
43
44#define CREATE_TRACE_POINTS
45#include <trace/events/block.h>
46
47#include "blk.h"
48#include "blk-mq-sched.h"
49#include "blk-pm.h"
50#include "blk-cgroup.h"
51#include "blk-throttle.h"
52#include "blk-ioprio.h"
53
54struct dentry *blk_debugfs_root;
55
56EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
57EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
58EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
59EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
60EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
61EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_insert);
62
63static DEFINE_IDA(blk_queue_ida);
64
65/*
66 * For queue allocation
67 */
68static struct kmem_cache *blk_requestq_cachep;
69
70/*
71 * Controlling structure to kblockd
72 */
73static struct workqueue_struct *kblockd_workqueue;
74
75/**
76 * blk_queue_flag_set - atomically set a queue flag
77 * @flag: flag to be set
78 * @q: request queue
79 */
80void blk_queue_flag_set(unsigned int flag, struct request_queue *q)
81{
82 set_bit(flag, &q->queue_flags);
83}
84EXPORT_SYMBOL(blk_queue_flag_set);
85
86/**
87 * blk_queue_flag_clear - atomically clear a queue flag
88 * @flag: flag to be cleared
89 * @q: request queue
90 */
91void blk_queue_flag_clear(unsigned int flag, struct request_queue *q)
92{
93 clear_bit(flag, &q->queue_flags);
94}
95EXPORT_SYMBOL(blk_queue_flag_clear);
96
97#define REQ_OP_NAME(name) [REQ_OP_##name] = #name
98static const char *const blk_op_name[] = {
99 REQ_OP_NAME(READ),
100 REQ_OP_NAME(WRITE),
101 REQ_OP_NAME(FLUSH),
102 REQ_OP_NAME(DISCARD),
103 REQ_OP_NAME(SECURE_ERASE),
104 REQ_OP_NAME(ZONE_RESET),
105 REQ_OP_NAME(ZONE_RESET_ALL),
106 REQ_OP_NAME(ZONE_OPEN),
107 REQ_OP_NAME(ZONE_CLOSE),
108 REQ_OP_NAME(ZONE_FINISH),
109 REQ_OP_NAME(ZONE_APPEND),
110 REQ_OP_NAME(WRITE_ZEROES),
111 REQ_OP_NAME(DRV_IN),
112 REQ_OP_NAME(DRV_OUT),
113};
114#undef REQ_OP_NAME
115
116/**
117 * blk_op_str - Return string XXX in the REQ_OP_XXX.
118 * @op: REQ_OP_XXX.
119 *
120 * Description: Centralize block layer function to convert REQ_OP_XXX into
121 * string format. Useful in the debugging and tracing bio or request. For
122 * invalid REQ_OP_XXX it returns string "UNKNOWN".
123 */
124inline const char *blk_op_str(enum req_op op)
125{
126 const char *op_str = "UNKNOWN";
127
128 if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op])
129 op_str = blk_op_name[op];
130
131 return op_str;
132}
133EXPORT_SYMBOL_GPL(blk_op_str);
134
135static const struct {
136 int errno;
137 const char *name;
138} blk_errors[] = {
139 [BLK_STS_OK] = { 0, "" },
140 [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" },
141 [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" },
142 [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" },
143 [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" },
144 [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" },
145 [BLK_STS_RESV_CONFLICT] = { -EBADE, "reservation conflict" },
146 [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" },
147 [BLK_STS_PROTECTION] = { -EILSEQ, "protection" },
148 [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" },
149 [BLK_STS_DEV_RESOURCE] = { -EBUSY, "device resource" },
150 [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" },
151 [BLK_STS_OFFLINE] = { -ENODEV, "device offline" },
152
153 /* device mapper special case, should not leak out: */
154 [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" },
155
156 /* zone device specific errors */
157 [BLK_STS_ZONE_OPEN_RESOURCE] = { -ETOOMANYREFS, "open zones exceeded" },
158 [BLK_STS_ZONE_ACTIVE_RESOURCE] = { -EOVERFLOW, "active zones exceeded" },
159
160 /* Command duration limit device-side timeout */
161 [BLK_STS_DURATION_LIMIT] = { -ETIME, "duration limit exceeded" },
162
163 [BLK_STS_INVAL] = { -EINVAL, "invalid" },
164
165 /* everything else not covered above: */
166 [BLK_STS_IOERR] = { -EIO, "I/O" },
167};
168
169blk_status_t errno_to_blk_status(int errno)
170{
171 int i;
172
173 for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
174 if (blk_errors[i].errno == errno)
175 return (__force blk_status_t)i;
176 }
177
178 return BLK_STS_IOERR;
179}
180EXPORT_SYMBOL_GPL(errno_to_blk_status);
181
182int blk_status_to_errno(blk_status_t status)
183{
184 int idx = (__force int)status;
185
186 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
187 return -EIO;
188 return blk_errors[idx].errno;
189}
190EXPORT_SYMBOL_GPL(blk_status_to_errno);
191
192const char *blk_status_to_str(blk_status_t status)
193{
194 int idx = (__force int)status;
195
196 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
197 return "<null>";
198 return blk_errors[idx].name;
199}
200EXPORT_SYMBOL_GPL(blk_status_to_str);
201
202/**
203 * blk_sync_queue - cancel any pending callbacks on a queue
204 * @q: the queue
205 *
206 * Description:
207 * The block layer may perform asynchronous callback activity
208 * on a queue, such as calling the unplug function after a timeout.
209 * A block device may call blk_sync_queue to ensure that any
210 * such activity is cancelled, thus allowing it to release resources
211 * that the callbacks might use. The caller must already have made sure
212 * that its ->submit_bio will not re-add plugging prior to calling
213 * this function.
214 *
215 * This function does not cancel any asynchronous activity arising
216 * out of elevator or throttling code. That would require elevator_exit()
217 * and blkcg_exit_queue() to be called with queue lock initialized.
218 *
219 */
220void blk_sync_queue(struct request_queue *q)
221{
222 del_timer_sync(&q->timeout);
223 cancel_work_sync(&q->timeout_work);
224}
225EXPORT_SYMBOL(blk_sync_queue);
226
227/**
228 * blk_set_pm_only - increment pm_only counter
229 * @q: request queue pointer
230 */
231void blk_set_pm_only(struct request_queue *q)
232{
233 atomic_inc(&q->pm_only);
234}
235EXPORT_SYMBOL_GPL(blk_set_pm_only);
236
237void blk_clear_pm_only(struct request_queue *q)
238{
239 int pm_only;
240
241 pm_only = atomic_dec_return(&q->pm_only);
242 WARN_ON_ONCE(pm_only < 0);
243 if (pm_only == 0)
244 wake_up_all(&q->mq_freeze_wq);
245}
246EXPORT_SYMBOL_GPL(blk_clear_pm_only);
247
248static void blk_free_queue_rcu(struct rcu_head *rcu_head)
249{
250 struct request_queue *q = container_of(rcu_head,
251 struct request_queue, rcu_head);
252
253 percpu_ref_exit(&q->q_usage_counter);
254 kmem_cache_free(blk_requestq_cachep, q);
255}
256
257static void blk_free_queue(struct request_queue *q)
258{
259 blk_free_queue_stats(q->stats);
260 if (queue_is_mq(q))
261 blk_mq_release(q);
262
263 ida_free(&blk_queue_ida, q->id);
264 lockdep_unregister_key(&q->io_lock_cls_key);
265 lockdep_unregister_key(&q->q_lock_cls_key);
266 call_rcu(&q->rcu_head, blk_free_queue_rcu);
267}
268
269/**
270 * blk_put_queue - decrement the request_queue refcount
271 * @q: the request_queue structure to decrement the refcount for
272 *
273 * Decrements the refcount of the request_queue and free it when the refcount
274 * reaches 0.
275 */
276void blk_put_queue(struct request_queue *q)
277{
278 if (refcount_dec_and_test(&q->refs))
279 blk_free_queue(q);
280}
281EXPORT_SYMBOL(blk_put_queue);
282
283bool blk_queue_start_drain(struct request_queue *q)
284{
285 /*
286 * When queue DYING flag is set, we need to block new req
287 * entering queue, so we call blk_freeze_queue_start() to
288 * prevent I/O from crossing blk_queue_enter().
289 */
290 bool freeze = __blk_freeze_queue_start(q, current);
291 if (queue_is_mq(q))
292 blk_mq_wake_waiters(q);
293 /* Make blk_queue_enter() reexamine the DYING flag. */
294 wake_up_all(&q->mq_freeze_wq);
295
296 return freeze;
297}
298
299/**
300 * blk_queue_enter() - try to increase q->q_usage_counter
301 * @q: request queue pointer
302 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PM
303 */
304int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
305{
306 const bool pm = flags & BLK_MQ_REQ_PM;
307
308 while (!blk_try_enter_queue(q, pm)) {
309 if (flags & BLK_MQ_REQ_NOWAIT)
310 return -EAGAIN;
311
312 /*
313 * read pair of barrier in blk_freeze_queue_start(), we need to
314 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
315 * reading .mq_freeze_depth or queue dying flag, otherwise the
316 * following wait may never return if the two reads are
317 * reordered.
318 */
319 smp_rmb();
320 wait_event(q->mq_freeze_wq,
321 (!q->mq_freeze_depth &&
322 blk_pm_resume_queue(pm, q)) ||
323 blk_queue_dying(q));
324 if (blk_queue_dying(q))
325 return -ENODEV;
326 }
327
328 rwsem_acquire_read(&q->q_lockdep_map, 0, 0, _RET_IP_);
329 rwsem_release(&q->q_lockdep_map, _RET_IP_);
330 return 0;
331}
332
333int __bio_queue_enter(struct request_queue *q, struct bio *bio)
334{
335 while (!blk_try_enter_queue(q, false)) {
336 struct gendisk *disk = bio->bi_bdev->bd_disk;
337
338 if (bio->bi_opf & REQ_NOWAIT) {
339 if (test_bit(GD_DEAD, &disk->state))
340 goto dead;
341 bio_wouldblock_error(bio);
342 return -EAGAIN;
343 }
344
345 /*
346 * read pair of barrier in blk_freeze_queue_start(), we need to
347 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
348 * reading .mq_freeze_depth or queue dying flag, otherwise the
349 * following wait may never return if the two reads are
350 * reordered.
351 */
352 smp_rmb();
353 wait_event(q->mq_freeze_wq,
354 (!q->mq_freeze_depth &&
355 blk_pm_resume_queue(false, q)) ||
356 test_bit(GD_DEAD, &disk->state));
357 if (test_bit(GD_DEAD, &disk->state))
358 goto dead;
359 }
360
361 rwsem_acquire_read(&q->io_lockdep_map, 0, 0, _RET_IP_);
362 rwsem_release(&q->io_lockdep_map, _RET_IP_);
363 return 0;
364dead:
365 bio_io_error(bio);
366 return -ENODEV;
367}
368
369void blk_queue_exit(struct request_queue *q)
370{
371 percpu_ref_put(&q->q_usage_counter);
372}
373
374static void blk_queue_usage_counter_release(struct percpu_ref *ref)
375{
376 struct request_queue *q =
377 container_of(ref, struct request_queue, q_usage_counter);
378
379 wake_up_all(&q->mq_freeze_wq);
380}
381
382static void blk_rq_timed_out_timer(struct timer_list *t)
383{
384 struct request_queue *q = from_timer(q, t, timeout);
385
386 kblockd_schedule_work(&q->timeout_work);
387}
388
389static void blk_timeout_work(struct work_struct *work)
390{
391}
392
393struct request_queue *blk_alloc_queue(struct queue_limits *lim, int node_id)
394{
395 struct request_queue *q;
396 int error;
397
398 q = kmem_cache_alloc_node(blk_requestq_cachep, GFP_KERNEL | __GFP_ZERO,
399 node_id);
400 if (!q)
401 return ERR_PTR(-ENOMEM);
402
403 q->last_merge = NULL;
404
405 q->id = ida_alloc(&blk_queue_ida, GFP_KERNEL);
406 if (q->id < 0) {
407 error = q->id;
408 goto fail_q;
409 }
410
411 q->stats = blk_alloc_queue_stats();
412 if (!q->stats) {
413 error = -ENOMEM;
414 goto fail_id;
415 }
416
417 error = blk_set_default_limits(lim);
418 if (error)
419 goto fail_stats;
420 q->limits = *lim;
421
422 q->node = node_id;
423
424 atomic_set(&q->nr_active_requests_shared_tags, 0);
425
426 timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
427 INIT_WORK(&q->timeout_work, blk_timeout_work);
428 INIT_LIST_HEAD(&q->icq_list);
429
430 refcount_set(&q->refs, 1);
431 mutex_init(&q->debugfs_mutex);
432 mutex_init(&q->sysfs_lock);
433 mutex_init(&q->sysfs_dir_lock);
434 mutex_init(&q->limits_lock);
435 mutex_init(&q->rq_qos_mutex);
436 spin_lock_init(&q->queue_lock);
437
438 init_waitqueue_head(&q->mq_freeze_wq);
439 mutex_init(&q->mq_freeze_lock);
440
441 blkg_init_queue(q);
442
443 /*
444 * Init percpu_ref in atomic mode so that it's faster to shutdown.
445 * See blk_register_queue() for details.
446 */
447 error = percpu_ref_init(&q->q_usage_counter,
448 blk_queue_usage_counter_release,
449 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL);
450 if (error)
451 goto fail_stats;
452 lockdep_register_key(&q->io_lock_cls_key);
453 lockdep_register_key(&q->q_lock_cls_key);
454 lockdep_init_map(&q->io_lockdep_map, "&q->q_usage_counter(io)",
455 &q->io_lock_cls_key, 0);
456 lockdep_init_map(&q->q_lockdep_map, "&q->q_usage_counter(queue)",
457 &q->q_lock_cls_key, 0);
458
459 q->nr_requests = BLKDEV_DEFAULT_RQ;
460
461 return q;
462
463fail_stats:
464 blk_free_queue_stats(q->stats);
465fail_id:
466 ida_free(&blk_queue_ida, q->id);
467fail_q:
468 kmem_cache_free(blk_requestq_cachep, q);
469 return ERR_PTR(error);
470}
471
472/**
473 * blk_get_queue - increment the request_queue refcount
474 * @q: the request_queue structure to increment the refcount for
475 *
476 * Increment the refcount of the request_queue kobject.
477 *
478 * Context: Any context.
479 */
480bool blk_get_queue(struct request_queue *q)
481{
482 if (unlikely(blk_queue_dying(q)))
483 return false;
484 refcount_inc(&q->refs);
485 return true;
486}
487EXPORT_SYMBOL(blk_get_queue);
488
489#ifdef CONFIG_FAIL_MAKE_REQUEST
490
491static DECLARE_FAULT_ATTR(fail_make_request);
492
493static int __init setup_fail_make_request(char *str)
494{
495 return setup_fault_attr(&fail_make_request, str);
496}
497__setup("fail_make_request=", setup_fail_make_request);
498
499bool should_fail_request(struct block_device *part, unsigned int bytes)
500{
501 return bdev_test_flag(part, BD_MAKE_IT_FAIL) &&
502 should_fail(&fail_make_request, bytes);
503}
504
505static int __init fail_make_request_debugfs(void)
506{
507 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
508 NULL, &fail_make_request);
509
510 return PTR_ERR_OR_ZERO(dir);
511}
512
513late_initcall(fail_make_request_debugfs);
514#endif /* CONFIG_FAIL_MAKE_REQUEST */
515
516static inline void bio_check_ro(struct bio *bio)
517{
518 if (op_is_write(bio_op(bio)) && bdev_read_only(bio->bi_bdev)) {
519 if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
520 return;
521
522 if (bdev_test_flag(bio->bi_bdev, BD_RO_WARNED))
523 return;
524
525 bdev_set_flag(bio->bi_bdev, BD_RO_WARNED);
526
527 /*
528 * Use ioctl to set underlying disk of raid/dm to read-only
529 * will trigger this.
530 */
531 pr_warn("Trying to write to read-only block-device %pg\n",
532 bio->bi_bdev);
533 }
534}
535
536static noinline int should_fail_bio(struct bio *bio)
537{
538 if (should_fail_request(bdev_whole(bio->bi_bdev), bio->bi_iter.bi_size))
539 return -EIO;
540 return 0;
541}
542ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
543
544/*
545 * Check whether this bio extends beyond the end of the device or partition.
546 * This may well happen - the kernel calls bread() without checking the size of
547 * the device, e.g., when mounting a file system.
548 */
549static inline int bio_check_eod(struct bio *bio)
550{
551 sector_t maxsector = bdev_nr_sectors(bio->bi_bdev);
552 unsigned int nr_sectors = bio_sectors(bio);
553
554 if (nr_sectors &&
555 (nr_sectors > maxsector ||
556 bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
557 pr_info_ratelimited("%s: attempt to access beyond end of device\n"
558 "%pg: rw=%d, sector=%llu, nr_sectors = %u limit=%llu\n",
559 current->comm, bio->bi_bdev, bio->bi_opf,
560 bio->bi_iter.bi_sector, nr_sectors, maxsector);
561 return -EIO;
562 }
563 return 0;
564}
565
566/*
567 * Remap block n of partition p to block n+start(p) of the disk.
568 */
569static int blk_partition_remap(struct bio *bio)
570{
571 struct block_device *p = bio->bi_bdev;
572
573 if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
574 return -EIO;
575 if (bio_sectors(bio)) {
576 bio->bi_iter.bi_sector += p->bd_start_sect;
577 trace_block_bio_remap(bio, p->bd_dev,
578 bio->bi_iter.bi_sector -
579 p->bd_start_sect);
580 }
581 bio_set_flag(bio, BIO_REMAPPED);
582 return 0;
583}
584
585/*
586 * Check write append to a zoned block device.
587 */
588static inline blk_status_t blk_check_zone_append(struct request_queue *q,
589 struct bio *bio)
590{
591 int nr_sectors = bio_sectors(bio);
592
593 /* Only applicable to zoned block devices */
594 if (!bdev_is_zoned(bio->bi_bdev))
595 return BLK_STS_NOTSUPP;
596
597 /* The bio sector must point to the start of a sequential zone */
598 if (!bdev_is_zone_start(bio->bi_bdev, bio->bi_iter.bi_sector))
599 return BLK_STS_IOERR;
600
601 /*
602 * Not allowed to cross zone boundaries. Otherwise, the BIO will be
603 * split and could result in non-contiguous sectors being written in
604 * different zones.
605 */
606 if (nr_sectors > q->limits.chunk_sectors)
607 return BLK_STS_IOERR;
608
609 /* Make sure the BIO is small enough and will not get split */
610 if (nr_sectors > q->limits.max_zone_append_sectors)
611 return BLK_STS_IOERR;
612
613 bio->bi_opf |= REQ_NOMERGE;
614
615 return BLK_STS_OK;
616}
617
618static void __submit_bio(struct bio *bio)
619{
620 /* If plug is not used, add new plug here to cache nsecs time. */
621 struct blk_plug plug;
622
623 if (unlikely(!blk_crypto_bio_prep(&bio)))
624 return;
625
626 blk_start_plug(&plug);
627
628 if (!bdev_test_flag(bio->bi_bdev, BD_HAS_SUBMIT_BIO)) {
629 blk_mq_submit_bio(bio);
630 } else if (likely(bio_queue_enter(bio) == 0)) {
631 struct gendisk *disk = bio->bi_bdev->bd_disk;
632
633 if ((bio->bi_opf & REQ_POLLED) &&
634 !(disk->queue->limits.features & BLK_FEAT_POLL)) {
635 bio->bi_status = BLK_STS_NOTSUPP;
636 bio_endio(bio);
637 } else {
638 disk->fops->submit_bio(bio);
639 }
640 blk_queue_exit(disk->queue);
641 }
642
643 blk_finish_plug(&plug);
644}
645
646/*
647 * The loop in this function may be a bit non-obvious, and so deserves some
648 * explanation:
649 *
650 * - Before entering the loop, bio->bi_next is NULL (as all callers ensure
651 * that), so we have a list with a single bio.
652 * - We pretend that we have just taken it off a longer list, so we assign
653 * bio_list to a pointer to the bio_list_on_stack, thus initialising the
654 * bio_list of new bios to be added. ->submit_bio() may indeed add some more
655 * bios through a recursive call to submit_bio_noacct. If it did, we find a
656 * non-NULL value in bio_list and re-enter the loop from the top.
657 * - In this case we really did just take the bio of the top of the list (no
658 * pretending) and so remove it from bio_list, and call into ->submit_bio()
659 * again.
660 *
661 * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio.
662 * bio_list_on_stack[1] contains bios that were submitted before the current
663 * ->submit_bio, but that haven't been processed yet.
664 */
665static void __submit_bio_noacct(struct bio *bio)
666{
667 struct bio_list bio_list_on_stack[2];
668
669 BUG_ON(bio->bi_next);
670
671 bio_list_init(&bio_list_on_stack[0]);
672 current->bio_list = bio_list_on_stack;
673
674 do {
675 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
676 struct bio_list lower, same;
677
678 /*
679 * Create a fresh bio_list for all subordinate requests.
680 */
681 bio_list_on_stack[1] = bio_list_on_stack[0];
682 bio_list_init(&bio_list_on_stack[0]);
683
684 __submit_bio(bio);
685
686 /*
687 * Sort new bios into those for a lower level and those for the
688 * same level.
689 */
690 bio_list_init(&lower);
691 bio_list_init(&same);
692 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
693 if (q == bdev_get_queue(bio->bi_bdev))
694 bio_list_add(&same, bio);
695 else
696 bio_list_add(&lower, bio);
697
698 /*
699 * Now assemble so we handle the lowest level first.
700 */
701 bio_list_merge(&bio_list_on_stack[0], &lower);
702 bio_list_merge(&bio_list_on_stack[0], &same);
703 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
704 } while ((bio = bio_list_pop(&bio_list_on_stack[0])));
705
706 current->bio_list = NULL;
707}
708
709static void __submit_bio_noacct_mq(struct bio *bio)
710{
711 struct bio_list bio_list[2] = { };
712
713 current->bio_list = bio_list;
714
715 do {
716 __submit_bio(bio);
717 } while ((bio = bio_list_pop(&bio_list[0])));
718
719 current->bio_list = NULL;
720}
721
722void submit_bio_noacct_nocheck(struct bio *bio)
723{
724 blk_cgroup_bio_start(bio);
725 blkcg_bio_issue_init(bio);
726
727 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
728 trace_block_bio_queue(bio);
729 /*
730 * Now that enqueuing has been traced, we need to trace
731 * completion as well.
732 */
733 bio_set_flag(bio, BIO_TRACE_COMPLETION);
734 }
735
736 /*
737 * We only want one ->submit_bio to be active at a time, else stack
738 * usage with stacked devices could be a problem. Use current->bio_list
739 * to collect a list of requests submited by a ->submit_bio method while
740 * it is active, and then process them after it returned.
741 */
742 if (current->bio_list)
743 bio_list_add(¤t->bio_list[0], bio);
744 else if (!bdev_test_flag(bio->bi_bdev, BD_HAS_SUBMIT_BIO))
745 __submit_bio_noacct_mq(bio);
746 else
747 __submit_bio_noacct(bio);
748}
749
750static blk_status_t blk_validate_atomic_write_op_size(struct request_queue *q,
751 struct bio *bio)
752{
753 if (bio->bi_iter.bi_size > queue_atomic_write_unit_max_bytes(q))
754 return BLK_STS_INVAL;
755
756 if (bio->bi_iter.bi_size % queue_atomic_write_unit_min_bytes(q))
757 return BLK_STS_INVAL;
758
759 return BLK_STS_OK;
760}
761
762/**
763 * submit_bio_noacct - re-submit a bio to the block device layer for I/O
764 * @bio: The bio describing the location in memory and on the device.
765 *
766 * This is a version of submit_bio() that shall only be used for I/O that is
767 * resubmitted to lower level drivers by stacking block drivers. All file
768 * systems and other upper level users of the block layer should use
769 * submit_bio() instead.
770 */
771void submit_bio_noacct(struct bio *bio)
772{
773 struct block_device *bdev = bio->bi_bdev;
774 struct request_queue *q = bdev_get_queue(bdev);
775 blk_status_t status = BLK_STS_IOERR;
776
777 might_sleep();
778
779 /*
780 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
781 * if queue does not support NOWAIT.
782 */
783 if ((bio->bi_opf & REQ_NOWAIT) && !bdev_nowait(bdev))
784 goto not_supported;
785
786 if (should_fail_bio(bio))
787 goto end_io;
788 bio_check_ro(bio);
789 if (!bio_flagged(bio, BIO_REMAPPED)) {
790 if (unlikely(bio_check_eod(bio)))
791 goto end_io;
792 if (bdev_is_partition(bdev) &&
793 unlikely(blk_partition_remap(bio)))
794 goto end_io;
795 }
796
797 /*
798 * Filter flush bio's early so that bio based drivers without flush
799 * support don't have to worry about them.
800 */
801 if (op_is_flush(bio->bi_opf)) {
802 if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_WRITE &&
803 bio_op(bio) != REQ_OP_ZONE_APPEND))
804 goto end_io;
805 if (!bdev_write_cache(bdev)) {
806 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
807 if (!bio_sectors(bio)) {
808 status = BLK_STS_OK;
809 goto end_io;
810 }
811 }
812 }
813
814 switch (bio_op(bio)) {
815 case REQ_OP_READ:
816 break;
817 case REQ_OP_WRITE:
818 if (bio->bi_opf & REQ_ATOMIC) {
819 status = blk_validate_atomic_write_op_size(q, bio);
820 if (status != BLK_STS_OK)
821 goto end_io;
822 }
823 break;
824 case REQ_OP_FLUSH:
825 /*
826 * REQ_OP_FLUSH can't be submitted through bios, it is only
827 * synthetized in struct request by the flush state machine.
828 */
829 goto not_supported;
830 case REQ_OP_DISCARD:
831 if (!bdev_max_discard_sectors(bdev))
832 goto not_supported;
833 break;
834 case REQ_OP_SECURE_ERASE:
835 if (!bdev_max_secure_erase_sectors(bdev))
836 goto not_supported;
837 break;
838 case REQ_OP_ZONE_APPEND:
839 status = blk_check_zone_append(q, bio);
840 if (status != BLK_STS_OK)
841 goto end_io;
842 break;
843 case REQ_OP_WRITE_ZEROES:
844 if (!q->limits.max_write_zeroes_sectors)
845 goto not_supported;
846 break;
847 case REQ_OP_ZONE_RESET:
848 case REQ_OP_ZONE_OPEN:
849 case REQ_OP_ZONE_CLOSE:
850 case REQ_OP_ZONE_FINISH:
851 case REQ_OP_ZONE_RESET_ALL:
852 if (!bdev_is_zoned(bio->bi_bdev))
853 goto not_supported;
854 break;
855 case REQ_OP_DRV_IN:
856 case REQ_OP_DRV_OUT:
857 /*
858 * Driver private operations are only used with passthrough
859 * requests.
860 */
861 fallthrough;
862 default:
863 goto not_supported;
864 }
865
866 if (blk_throtl_bio(bio))
867 return;
868 submit_bio_noacct_nocheck(bio);
869 return;
870
871not_supported:
872 status = BLK_STS_NOTSUPP;
873end_io:
874 bio->bi_status = status;
875 bio_endio(bio);
876}
877EXPORT_SYMBOL(submit_bio_noacct);
878
879static void bio_set_ioprio(struct bio *bio)
880{
881 /* Nobody set ioprio so far? Initialize it based on task's nice value */
882 if (IOPRIO_PRIO_CLASS(bio->bi_ioprio) == IOPRIO_CLASS_NONE)
883 bio->bi_ioprio = get_current_ioprio();
884 blkcg_set_ioprio(bio);
885}
886
887/**
888 * submit_bio - submit a bio to the block device layer for I/O
889 * @bio: The &struct bio which describes the I/O
890 *
891 * submit_bio() is used to submit I/O requests to block devices. It is passed a
892 * fully set up &struct bio that describes the I/O that needs to be done. The
893 * bio will be send to the device described by the bi_bdev field.
894 *
895 * The success/failure status of the request, along with notification of
896 * completion, is delivered asynchronously through the ->bi_end_io() callback
897 * in @bio. The bio must NOT be touched by the caller until ->bi_end_io() has
898 * been called.
899 */
900void submit_bio(struct bio *bio)
901{
902 if (bio_op(bio) == REQ_OP_READ) {
903 task_io_account_read(bio->bi_iter.bi_size);
904 count_vm_events(PGPGIN, bio_sectors(bio));
905 } else if (bio_op(bio) == REQ_OP_WRITE) {
906 count_vm_events(PGPGOUT, bio_sectors(bio));
907 }
908
909 bio_set_ioprio(bio);
910 submit_bio_noacct(bio);
911}
912EXPORT_SYMBOL(submit_bio);
913
914/**
915 * bio_poll - poll for BIO completions
916 * @bio: bio to poll for
917 * @iob: batches of IO
918 * @flags: BLK_POLL_* flags that control the behavior
919 *
920 * Poll for completions on queue associated with the bio. Returns number of
921 * completed entries found.
922 *
923 * Note: the caller must either be the context that submitted @bio, or
924 * be in a RCU critical section to prevent freeing of @bio.
925 */
926int bio_poll(struct bio *bio, struct io_comp_batch *iob, unsigned int flags)
927{
928 blk_qc_t cookie = READ_ONCE(bio->bi_cookie);
929 struct block_device *bdev;
930 struct request_queue *q;
931 int ret = 0;
932
933 bdev = READ_ONCE(bio->bi_bdev);
934 if (!bdev)
935 return 0;
936
937 q = bdev_get_queue(bdev);
938 if (cookie == BLK_QC_T_NONE)
939 return 0;
940
941 blk_flush_plug(current->plug, false);
942
943 /*
944 * We need to be able to enter a frozen queue, similar to how
945 * timeouts also need to do that. If that is blocked, then we can
946 * have pending IO when a queue freeze is started, and then the
947 * wait for the freeze to finish will wait for polled requests to
948 * timeout as the poller is preventer from entering the queue and
949 * completing them. As long as we prevent new IO from being queued,
950 * that should be all that matters.
951 */
952 if (!percpu_ref_tryget(&q->q_usage_counter))
953 return 0;
954 if (queue_is_mq(q)) {
955 ret = blk_mq_poll(q, cookie, iob, flags);
956 } else {
957 struct gendisk *disk = q->disk;
958
959 if ((q->limits.features & BLK_FEAT_POLL) && disk &&
960 disk->fops->poll_bio)
961 ret = disk->fops->poll_bio(bio, iob, flags);
962 }
963 blk_queue_exit(q);
964 return ret;
965}
966EXPORT_SYMBOL_GPL(bio_poll);
967
968/*
969 * Helper to implement file_operations.iopoll. Requires the bio to be stored
970 * in iocb->private, and cleared before freeing the bio.
971 */
972int iocb_bio_iopoll(struct kiocb *kiocb, struct io_comp_batch *iob,
973 unsigned int flags)
974{
975 struct bio *bio;
976 int ret = 0;
977
978 /*
979 * Note: the bio cache only uses SLAB_TYPESAFE_BY_RCU, so bio can
980 * point to a freshly allocated bio at this point. If that happens
981 * we have a few cases to consider:
982 *
983 * 1) the bio is beeing initialized and bi_bdev is NULL. We can just
984 * simply nothing in this case
985 * 2) the bio points to a not poll enabled device. bio_poll will catch
986 * this and return 0
987 * 3) the bio points to a poll capable device, including but not
988 * limited to the one that the original bio pointed to. In this
989 * case we will call into the actual poll method and poll for I/O,
990 * even if we don't need to, but it won't cause harm either.
991 *
992 * For cases 2) and 3) above the RCU grace period ensures that bi_bdev
993 * is still allocated. Because partitions hold a reference to the whole
994 * device bdev and thus disk, the disk is also still valid. Grabbing
995 * a reference to the queue in bio_poll() ensures the hctxs and requests
996 * are still valid as well.
997 */
998 rcu_read_lock();
999 bio = READ_ONCE(kiocb->private);
1000 if (bio)
1001 ret = bio_poll(bio, iob, flags);
1002 rcu_read_unlock();
1003
1004 return ret;
1005}
1006EXPORT_SYMBOL_GPL(iocb_bio_iopoll);
1007
1008void update_io_ticks(struct block_device *part, unsigned long now, bool end)
1009{
1010 unsigned long stamp;
1011again:
1012 stamp = READ_ONCE(part->bd_stamp);
1013 if (unlikely(time_after(now, stamp)) &&
1014 likely(try_cmpxchg(&part->bd_stamp, &stamp, now)) &&
1015 (end || part_in_flight(part)))
1016 __part_stat_add(part, io_ticks, now - stamp);
1017
1018 if (bdev_is_partition(part)) {
1019 part = bdev_whole(part);
1020 goto again;
1021 }
1022}
1023
1024unsigned long bdev_start_io_acct(struct block_device *bdev, enum req_op op,
1025 unsigned long start_time)
1026{
1027 part_stat_lock();
1028 update_io_ticks(bdev, start_time, false);
1029 part_stat_local_inc(bdev, in_flight[op_is_write(op)]);
1030 part_stat_unlock();
1031
1032 return start_time;
1033}
1034EXPORT_SYMBOL(bdev_start_io_acct);
1035
1036/**
1037 * bio_start_io_acct - start I/O accounting for bio based drivers
1038 * @bio: bio to start account for
1039 *
1040 * Returns the start time that should be passed back to bio_end_io_acct().
1041 */
1042unsigned long bio_start_io_acct(struct bio *bio)
1043{
1044 return bdev_start_io_acct(bio->bi_bdev, bio_op(bio), jiffies);
1045}
1046EXPORT_SYMBOL_GPL(bio_start_io_acct);
1047
1048void bdev_end_io_acct(struct block_device *bdev, enum req_op op,
1049 unsigned int sectors, unsigned long start_time)
1050{
1051 const int sgrp = op_stat_group(op);
1052 unsigned long now = READ_ONCE(jiffies);
1053 unsigned long duration = now - start_time;
1054
1055 part_stat_lock();
1056 update_io_ticks(bdev, now, true);
1057 part_stat_inc(bdev, ios[sgrp]);
1058 part_stat_add(bdev, sectors[sgrp], sectors);
1059 part_stat_add(bdev, nsecs[sgrp], jiffies_to_nsecs(duration));
1060 part_stat_local_dec(bdev, in_flight[op_is_write(op)]);
1061 part_stat_unlock();
1062}
1063EXPORT_SYMBOL(bdev_end_io_acct);
1064
1065void bio_end_io_acct_remapped(struct bio *bio, unsigned long start_time,
1066 struct block_device *orig_bdev)
1067{
1068 bdev_end_io_acct(orig_bdev, bio_op(bio), bio_sectors(bio), start_time);
1069}
1070EXPORT_SYMBOL_GPL(bio_end_io_acct_remapped);
1071
1072/**
1073 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1074 * @q : the queue of the device being checked
1075 *
1076 * Description:
1077 * Check if underlying low-level drivers of a device are busy.
1078 * If the drivers want to export their busy state, they must set own
1079 * exporting function using blk_queue_lld_busy() first.
1080 *
1081 * Basically, this function is used only by request stacking drivers
1082 * to stop dispatching requests to underlying devices when underlying
1083 * devices are busy. This behavior helps more I/O merging on the queue
1084 * of the request stacking driver and prevents I/O throughput regression
1085 * on burst I/O load.
1086 *
1087 * Return:
1088 * 0 - Not busy (The request stacking driver should dispatch request)
1089 * 1 - Busy (The request stacking driver should stop dispatching request)
1090 */
1091int blk_lld_busy(struct request_queue *q)
1092{
1093 if (queue_is_mq(q) && q->mq_ops->busy)
1094 return q->mq_ops->busy(q);
1095
1096 return 0;
1097}
1098EXPORT_SYMBOL_GPL(blk_lld_busy);
1099
1100int kblockd_schedule_work(struct work_struct *work)
1101{
1102 return queue_work(kblockd_workqueue, work);
1103}
1104EXPORT_SYMBOL(kblockd_schedule_work);
1105
1106int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
1107 unsigned long delay)
1108{
1109 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
1110}
1111EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
1112
1113void blk_start_plug_nr_ios(struct blk_plug *plug, unsigned short nr_ios)
1114{
1115 struct task_struct *tsk = current;
1116
1117 /*
1118 * If this is a nested plug, don't actually assign it.
1119 */
1120 if (tsk->plug)
1121 return;
1122
1123 plug->cur_ktime = 0;
1124 rq_list_init(&plug->mq_list);
1125 rq_list_init(&plug->cached_rqs);
1126 plug->nr_ios = min_t(unsigned short, nr_ios, BLK_MAX_REQUEST_COUNT);
1127 plug->rq_count = 0;
1128 plug->multiple_queues = false;
1129 plug->has_elevator = false;
1130 INIT_LIST_HEAD(&plug->cb_list);
1131
1132 /*
1133 * Store ordering should not be needed here, since a potential
1134 * preempt will imply a full memory barrier
1135 */
1136 tsk->plug = plug;
1137}
1138
1139/**
1140 * blk_start_plug - initialize blk_plug and track it inside the task_struct
1141 * @plug: The &struct blk_plug that needs to be initialized
1142 *
1143 * Description:
1144 * blk_start_plug() indicates to the block layer an intent by the caller
1145 * to submit multiple I/O requests in a batch. The block layer may use
1146 * this hint to defer submitting I/Os from the caller until blk_finish_plug()
1147 * is called. However, the block layer may choose to submit requests
1148 * before a call to blk_finish_plug() if the number of queued I/Os
1149 * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1150 * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if
1151 * the task schedules (see below).
1152 *
1153 * Tracking blk_plug inside the task_struct will help with auto-flushing the
1154 * pending I/O should the task end up blocking between blk_start_plug() and
1155 * blk_finish_plug(). This is important from a performance perspective, but
1156 * also ensures that we don't deadlock. For instance, if the task is blocking
1157 * for a memory allocation, memory reclaim could end up wanting to free a
1158 * page belonging to that request that is currently residing in our private
1159 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
1160 * this kind of deadlock.
1161 */
1162void blk_start_plug(struct blk_plug *plug)
1163{
1164 blk_start_plug_nr_ios(plug, 1);
1165}
1166EXPORT_SYMBOL(blk_start_plug);
1167
1168static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
1169{
1170 LIST_HEAD(callbacks);
1171
1172 while (!list_empty(&plug->cb_list)) {
1173 list_splice_init(&plug->cb_list, &callbacks);
1174
1175 while (!list_empty(&callbacks)) {
1176 struct blk_plug_cb *cb = list_first_entry(&callbacks,
1177 struct blk_plug_cb,
1178 list);
1179 list_del(&cb->list);
1180 cb->callback(cb, from_schedule);
1181 }
1182 }
1183}
1184
1185struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
1186 int size)
1187{
1188 struct blk_plug *plug = current->plug;
1189 struct blk_plug_cb *cb;
1190
1191 if (!plug)
1192 return NULL;
1193
1194 list_for_each_entry(cb, &plug->cb_list, list)
1195 if (cb->callback == unplug && cb->data == data)
1196 return cb;
1197
1198 /* Not currently on the callback list */
1199 BUG_ON(size < sizeof(*cb));
1200 cb = kzalloc(size, GFP_ATOMIC);
1201 if (cb) {
1202 cb->data = data;
1203 cb->callback = unplug;
1204 list_add(&cb->list, &plug->cb_list);
1205 }
1206 return cb;
1207}
1208EXPORT_SYMBOL(blk_check_plugged);
1209
1210void __blk_flush_plug(struct blk_plug *plug, bool from_schedule)
1211{
1212 if (!list_empty(&plug->cb_list))
1213 flush_plug_callbacks(plug, from_schedule);
1214 blk_mq_flush_plug_list(plug, from_schedule);
1215 /*
1216 * Unconditionally flush out cached requests, even if the unplug
1217 * event came from schedule. Since we know hold references to the
1218 * queue for cached requests, we don't want a blocked task holding
1219 * up a queue freeze/quiesce event.
1220 */
1221 if (unlikely(!rq_list_empty(&plug->cached_rqs)))
1222 blk_mq_free_plug_rqs(plug);
1223
1224 plug->cur_ktime = 0;
1225 current->flags &= ~PF_BLOCK_TS;
1226}
1227
1228/**
1229 * blk_finish_plug - mark the end of a batch of submitted I/O
1230 * @plug: The &struct blk_plug passed to blk_start_plug()
1231 *
1232 * Description:
1233 * Indicate that a batch of I/O submissions is complete. This function
1234 * must be paired with an initial call to blk_start_plug(). The intent
1235 * is to allow the block layer to optimize I/O submission. See the
1236 * documentation for blk_start_plug() for more information.
1237 */
1238void blk_finish_plug(struct blk_plug *plug)
1239{
1240 if (plug == current->plug) {
1241 __blk_flush_plug(plug, false);
1242 current->plug = NULL;
1243 }
1244}
1245EXPORT_SYMBOL(blk_finish_plug);
1246
1247void blk_io_schedule(void)
1248{
1249 /* Prevent hang_check timer from firing at us during very long I/O */
1250 unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2;
1251
1252 if (timeout)
1253 io_schedule_timeout(timeout);
1254 else
1255 io_schedule();
1256}
1257EXPORT_SYMBOL_GPL(blk_io_schedule);
1258
1259int __init blk_dev_init(void)
1260{
1261 BUILD_BUG_ON((__force u32)REQ_OP_LAST >= (1 << REQ_OP_BITS));
1262 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1263 sizeof_field(struct request, cmd_flags));
1264 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1265 sizeof_field(struct bio, bi_opf));
1266
1267 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
1268 kblockd_workqueue = alloc_workqueue("kblockd",
1269 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1270 if (!kblockd_workqueue)
1271 panic("Failed to create kblockd\n");
1272
1273 blk_requestq_cachep = KMEM_CACHE(request_queue, SLAB_PANIC);
1274
1275 blk_debugfs_root = debugfs_create_dir("block", NULL);
1276
1277 return 0;
1278}