Loading...
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
53struct dentry *blk_debugfs_root;
54
55EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
56EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
57EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
58EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
59EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
60EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_insert);
61
62static DEFINE_IDA(blk_queue_ida);
63
64/*
65 * For queue allocation
66 */
67static struct kmem_cache *blk_requestq_cachep;
68
69/*
70 * Controlling structure to kblockd
71 */
72static struct workqueue_struct *kblockd_workqueue;
73
74/**
75 * blk_queue_flag_set - atomically set a queue flag
76 * @flag: flag to be set
77 * @q: request queue
78 */
79void blk_queue_flag_set(unsigned int flag, struct request_queue *q)
80{
81 set_bit(flag, &q->queue_flags);
82}
83EXPORT_SYMBOL(blk_queue_flag_set);
84
85/**
86 * blk_queue_flag_clear - atomically clear a queue flag
87 * @flag: flag to be cleared
88 * @q: request queue
89 */
90void blk_queue_flag_clear(unsigned int flag, struct request_queue *q)
91{
92 clear_bit(flag, &q->queue_flags);
93}
94EXPORT_SYMBOL(blk_queue_flag_clear);
95
96/**
97 * blk_queue_flag_test_and_set - atomically test and set a queue flag
98 * @flag: flag to be set
99 * @q: request queue
100 *
101 * Returns the previous value of @flag - 0 if the flag was not set and 1 if
102 * the flag was already set.
103 */
104bool blk_queue_flag_test_and_set(unsigned int flag, struct request_queue *q)
105{
106 return test_and_set_bit(flag, &q->queue_flags);
107}
108EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set);
109
110#define REQ_OP_NAME(name) [REQ_OP_##name] = #name
111static const char *const blk_op_name[] = {
112 REQ_OP_NAME(READ),
113 REQ_OP_NAME(WRITE),
114 REQ_OP_NAME(FLUSH),
115 REQ_OP_NAME(DISCARD),
116 REQ_OP_NAME(SECURE_ERASE),
117 REQ_OP_NAME(ZONE_RESET),
118 REQ_OP_NAME(ZONE_RESET_ALL),
119 REQ_OP_NAME(ZONE_OPEN),
120 REQ_OP_NAME(ZONE_CLOSE),
121 REQ_OP_NAME(ZONE_FINISH),
122 REQ_OP_NAME(ZONE_APPEND),
123 REQ_OP_NAME(WRITE_ZEROES),
124 REQ_OP_NAME(DRV_IN),
125 REQ_OP_NAME(DRV_OUT),
126};
127#undef REQ_OP_NAME
128
129/**
130 * blk_op_str - Return string XXX in the REQ_OP_XXX.
131 * @op: REQ_OP_XXX.
132 *
133 * Description: Centralize block layer function to convert REQ_OP_XXX into
134 * string format. Useful in the debugging and tracing bio or request. For
135 * invalid REQ_OP_XXX it returns string "UNKNOWN".
136 */
137inline const char *blk_op_str(enum req_op op)
138{
139 const char *op_str = "UNKNOWN";
140
141 if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op])
142 op_str = blk_op_name[op];
143
144 return op_str;
145}
146EXPORT_SYMBOL_GPL(blk_op_str);
147
148static const struct {
149 int errno;
150 const char *name;
151} blk_errors[] = {
152 [BLK_STS_OK] = { 0, "" },
153 [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" },
154 [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" },
155 [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" },
156 [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" },
157 [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" },
158 [BLK_STS_NEXUS] = { -EBADE, "critical nexus" },
159 [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" },
160 [BLK_STS_PROTECTION] = { -EILSEQ, "protection" },
161 [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" },
162 [BLK_STS_DEV_RESOURCE] = { -EBUSY, "device resource" },
163 [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" },
164 [BLK_STS_OFFLINE] = { -ENODEV, "device offline" },
165
166 /* device mapper special case, should not leak out: */
167 [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" },
168
169 /* zone device specific errors */
170 [BLK_STS_ZONE_OPEN_RESOURCE] = { -ETOOMANYREFS, "open zones exceeded" },
171 [BLK_STS_ZONE_ACTIVE_RESOURCE] = { -EOVERFLOW, "active zones exceeded" },
172
173 /* everything else not covered above: */
174 [BLK_STS_IOERR] = { -EIO, "I/O" },
175};
176
177blk_status_t errno_to_blk_status(int errno)
178{
179 int i;
180
181 for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
182 if (blk_errors[i].errno == errno)
183 return (__force blk_status_t)i;
184 }
185
186 return BLK_STS_IOERR;
187}
188EXPORT_SYMBOL_GPL(errno_to_blk_status);
189
190int blk_status_to_errno(blk_status_t status)
191{
192 int idx = (__force int)status;
193
194 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
195 return -EIO;
196 return blk_errors[idx].errno;
197}
198EXPORT_SYMBOL_GPL(blk_status_to_errno);
199
200const char *blk_status_to_str(blk_status_t status)
201{
202 int idx = (__force int)status;
203
204 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
205 return "<null>";
206 return blk_errors[idx].name;
207}
208
209/**
210 * blk_sync_queue - cancel any pending callbacks on a queue
211 * @q: the queue
212 *
213 * Description:
214 * The block layer may perform asynchronous callback activity
215 * on a queue, such as calling the unplug function after a timeout.
216 * A block device may call blk_sync_queue to ensure that any
217 * such activity is cancelled, thus allowing it to release resources
218 * that the callbacks might use. The caller must already have made sure
219 * that its ->submit_bio will not re-add plugging prior to calling
220 * this function.
221 *
222 * This function does not cancel any asynchronous activity arising
223 * out of elevator or throttling code. That would require elevator_exit()
224 * and blkcg_exit_queue() to be called with queue lock initialized.
225 *
226 */
227void blk_sync_queue(struct request_queue *q)
228{
229 del_timer_sync(&q->timeout);
230 cancel_work_sync(&q->timeout_work);
231}
232EXPORT_SYMBOL(blk_sync_queue);
233
234/**
235 * blk_set_pm_only - increment pm_only counter
236 * @q: request queue pointer
237 */
238void blk_set_pm_only(struct request_queue *q)
239{
240 atomic_inc(&q->pm_only);
241}
242EXPORT_SYMBOL_GPL(blk_set_pm_only);
243
244void blk_clear_pm_only(struct request_queue *q)
245{
246 int pm_only;
247
248 pm_only = atomic_dec_return(&q->pm_only);
249 WARN_ON_ONCE(pm_only < 0);
250 if (pm_only == 0)
251 wake_up_all(&q->mq_freeze_wq);
252}
253EXPORT_SYMBOL_GPL(blk_clear_pm_only);
254
255static void blk_free_queue_rcu(struct rcu_head *rcu_head)
256{
257 struct request_queue *q = container_of(rcu_head,
258 struct request_queue, rcu_head);
259
260 percpu_ref_exit(&q->q_usage_counter);
261 kmem_cache_free(blk_requestq_cachep, q);
262}
263
264static void blk_free_queue(struct request_queue *q)
265{
266 if (q->poll_stat)
267 blk_stat_remove_callback(q, q->poll_cb);
268 blk_stat_free_callback(q->poll_cb);
269
270 blk_free_queue_stats(q->stats);
271 kfree(q->poll_stat);
272
273 if (queue_is_mq(q))
274 blk_mq_release(q);
275
276 ida_free(&blk_queue_ida, q->id);
277 call_rcu(&q->rcu_head, blk_free_queue_rcu);
278}
279
280/**
281 * blk_put_queue - decrement the request_queue refcount
282 * @q: the request_queue structure to decrement the refcount for
283 *
284 * Decrements the refcount of the request_queue and free it when the refcount
285 * reaches 0.
286 */
287void blk_put_queue(struct request_queue *q)
288{
289 if (refcount_dec_and_test(&q->refs))
290 blk_free_queue(q);
291}
292EXPORT_SYMBOL(blk_put_queue);
293
294void blk_queue_start_drain(struct request_queue *q)
295{
296 /*
297 * When queue DYING flag is set, we need to block new req
298 * entering queue, so we call blk_freeze_queue_start() to
299 * prevent I/O from crossing blk_queue_enter().
300 */
301 blk_freeze_queue_start(q);
302 if (queue_is_mq(q))
303 blk_mq_wake_waiters(q);
304 /* Make blk_queue_enter() reexamine the DYING flag. */
305 wake_up_all(&q->mq_freeze_wq);
306}
307
308/**
309 * blk_queue_enter() - try to increase q->q_usage_counter
310 * @q: request queue pointer
311 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PM
312 */
313int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
314{
315 const bool pm = flags & BLK_MQ_REQ_PM;
316
317 while (!blk_try_enter_queue(q, pm)) {
318 if (flags & BLK_MQ_REQ_NOWAIT)
319 return -EAGAIN;
320
321 /*
322 * read pair of barrier in blk_freeze_queue_start(), we need to
323 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
324 * reading .mq_freeze_depth or queue dying flag, otherwise the
325 * following wait may never return if the two reads are
326 * reordered.
327 */
328 smp_rmb();
329 wait_event(q->mq_freeze_wq,
330 (!q->mq_freeze_depth &&
331 blk_pm_resume_queue(pm, q)) ||
332 blk_queue_dying(q));
333 if (blk_queue_dying(q))
334 return -ENODEV;
335 }
336
337 return 0;
338}
339
340int __bio_queue_enter(struct request_queue *q, struct bio *bio)
341{
342 while (!blk_try_enter_queue(q, false)) {
343 struct gendisk *disk = bio->bi_bdev->bd_disk;
344
345 if (bio->bi_opf & REQ_NOWAIT) {
346 if (test_bit(GD_DEAD, &disk->state))
347 goto dead;
348 bio_wouldblock_error(bio);
349 return -EAGAIN;
350 }
351
352 /*
353 * read pair of barrier in blk_freeze_queue_start(), we need to
354 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
355 * reading .mq_freeze_depth or queue dying flag, otherwise the
356 * following wait may never return if the two reads are
357 * reordered.
358 */
359 smp_rmb();
360 wait_event(q->mq_freeze_wq,
361 (!q->mq_freeze_depth &&
362 blk_pm_resume_queue(false, q)) ||
363 test_bit(GD_DEAD, &disk->state));
364 if (test_bit(GD_DEAD, &disk->state))
365 goto dead;
366 }
367
368 return 0;
369dead:
370 bio_io_error(bio);
371 return -ENODEV;
372}
373
374void blk_queue_exit(struct request_queue *q)
375{
376 percpu_ref_put(&q->q_usage_counter);
377}
378
379static void blk_queue_usage_counter_release(struct percpu_ref *ref)
380{
381 struct request_queue *q =
382 container_of(ref, struct request_queue, q_usage_counter);
383
384 wake_up_all(&q->mq_freeze_wq);
385}
386
387static void blk_rq_timed_out_timer(struct timer_list *t)
388{
389 struct request_queue *q = from_timer(q, t, timeout);
390
391 kblockd_schedule_work(&q->timeout_work);
392}
393
394static void blk_timeout_work(struct work_struct *work)
395{
396}
397
398struct request_queue *blk_alloc_queue(int node_id)
399{
400 struct request_queue *q;
401
402 q = kmem_cache_alloc_node(blk_requestq_cachep, GFP_KERNEL | __GFP_ZERO,
403 node_id);
404 if (!q)
405 return NULL;
406
407 q->last_merge = NULL;
408
409 q->id = ida_alloc(&blk_queue_ida, GFP_KERNEL);
410 if (q->id < 0)
411 goto fail_q;
412
413 q->stats = blk_alloc_queue_stats();
414 if (!q->stats)
415 goto fail_id;
416
417 q->node = node_id;
418
419 atomic_set(&q->nr_active_requests_shared_tags, 0);
420
421 timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
422 INIT_WORK(&q->timeout_work, blk_timeout_work);
423 INIT_LIST_HEAD(&q->icq_list);
424
425 refcount_set(&q->refs, 1);
426 mutex_init(&q->debugfs_mutex);
427 mutex_init(&q->sysfs_lock);
428 mutex_init(&q->sysfs_dir_lock);
429 spin_lock_init(&q->queue_lock);
430
431 init_waitqueue_head(&q->mq_freeze_wq);
432 mutex_init(&q->mq_freeze_lock);
433
434 /*
435 * Init percpu_ref in atomic mode so that it's faster to shutdown.
436 * See blk_register_queue() for details.
437 */
438 if (percpu_ref_init(&q->q_usage_counter,
439 blk_queue_usage_counter_release,
440 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
441 goto fail_stats;
442
443 blk_set_default_limits(&q->limits);
444 q->nr_requests = BLKDEV_DEFAULT_RQ;
445
446 return q;
447
448fail_stats:
449 blk_free_queue_stats(q->stats);
450fail_id:
451 ida_free(&blk_queue_ida, q->id);
452fail_q:
453 kmem_cache_free(blk_requestq_cachep, q);
454 return NULL;
455}
456
457/**
458 * blk_get_queue - increment the request_queue refcount
459 * @q: the request_queue structure to increment the refcount for
460 *
461 * Increment the refcount of the request_queue kobject.
462 *
463 * Context: Any context.
464 */
465bool blk_get_queue(struct request_queue *q)
466{
467 if (unlikely(blk_queue_dying(q)))
468 return false;
469 refcount_inc(&q->refs);
470 return true;
471}
472EXPORT_SYMBOL(blk_get_queue);
473
474#ifdef CONFIG_FAIL_MAKE_REQUEST
475
476static DECLARE_FAULT_ATTR(fail_make_request);
477
478static int __init setup_fail_make_request(char *str)
479{
480 return setup_fault_attr(&fail_make_request, str);
481}
482__setup("fail_make_request=", setup_fail_make_request);
483
484bool should_fail_request(struct block_device *part, unsigned int bytes)
485{
486 return part->bd_make_it_fail && should_fail(&fail_make_request, bytes);
487}
488
489static int __init fail_make_request_debugfs(void)
490{
491 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
492 NULL, &fail_make_request);
493
494 return PTR_ERR_OR_ZERO(dir);
495}
496
497late_initcall(fail_make_request_debugfs);
498#endif /* CONFIG_FAIL_MAKE_REQUEST */
499
500static inline void bio_check_ro(struct bio *bio)
501{
502 if (op_is_write(bio_op(bio)) && bdev_read_only(bio->bi_bdev)) {
503 if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
504 return;
505 pr_warn("Trying to write to read-only block-device %pg\n",
506 bio->bi_bdev);
507 /* Older lvm-tools actually trigger this */
508 }
509}
510
511static noinline int should_fail_bio(struct bio *bio)
512{
513 if (should_fail_request(bdev_whole(bio->bi_bdev), bio->bi_iter.bi_size))
514 return -EIO;
515 return 0;
516}
517ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
518
519/*
520 * Check whether this bio extends beyond the end of the device or partition.
521 * This may well happen - the kernel calls bread() without checking the size of
522 * the device, e.g., when mounting a file system.
523 */
524static inline int bio_check_eod(struct bio *bio)
525{
526 sector_t maxsector = bdev_nr_sectors(bio->bi_bdev);
527 unsigned int nr_sectors = bio_sectors(bio);
528
529 if (nr_sectors && maxsector &&
530 (nr_sectors > maxsector ||
531 bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
532 pr_info_ratelimited("%s: attempt to access beyond end of device\n"
533 "%pg: rw=%d, sector=%llu, nr_sectors = %u limit=%llu\n",
534 current->comm, bio->bi_bdev, bio->bi_opf,
535 bio->bi_iter.bi_sector, nr_sectors, maxsector);
536 return -EIO;
537 }
538 return 0;
539}
540
541/*
542 * Remap block n of partition p to block n+start(p) of the disk.
543 */
544static int blk_partition_remap(struct bio *bio)
545{
546 struct block_device *p = bio->bi_bdev;
547
548 if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
549 return -EIO;
550 if (bio_sectors(bio)) {
551 bio->bi_iter.bi_sector += p->bd_start_sect;
552 trace_block_bio_remap(bio, p->bd_dev,
553 bio->bi_iter.bi_sector -
554 p->bd_start_sect);
555 }
556 bio_set_flag(bio, BIO_REMAPPED);
557 return 0;
558}
559
560/*
561 * Check write append to a zoned block device.
562 */
563static inline blk_status_t blk_check_zone_append(struct request_queue *q,
564 struct bio *bio)
565{
566 int nr_sectors = bio_sectors(bio);
567
568 /* Only applicable to zoned block devices */
569 if (!bdev_is_zoned(bio->bi_bdev))
570 return BLK_STS_NOTSUPP;
571
572 /* The bio sector must point to the start of a sequential zone */
573 if (bio->bi_iter.bi_sector & (bdev_zone_sectors(bio->bi_bdev) - 1) ||
574 !bio_zone_is_seq(bio))
575 return BLK_STS_IOERR;
576
577 /*
578 * Not allowed to cross zone boundaries. Otherwise, the BIO will be
579 * split and could result in non-contiguous sectors being written in
580 * different zones.
581 */
582 if (nr_sectors > q->limits.chunk_sectors)
583 return BLK_STS_IOERR;
584
585 /* Make sure the BIO is small enough and will not get split */
586 if (nr_sectors > q->limits.max_zone_append_sectors)
587 return BLK_STS_IOERR;
588
589 bio->bi_opf |= REQ_NOMERGE;
590
591 return BLK_STS_OK;
592}
593
594static void __submit_bio(struct bio *bio)
595{
596 struct gendisk *disk = bio->bi_bdev->bd_disk;
597
598 if (unlikely(!blk_crypto_bio_prep(&bio)))
599 return;
600
601 if (!disk->fops->submit_bio) {
602 blk_mq_submit_bio(bio);
603 } else if (likely(bio_queue_enter(bio) == 0)) {
604 disk->fops->submit_bio(bio);
605 blk_queue_exit(disk->queue);
606 }
607}
608
609/*
610 * The loop in this function may be a bit non-obvious, and so deserves some
611 * explanation:
612 *
613 * - Before entering the loop, bio->bi_next is NULL (as all callers ensure
614 * that), so we have a list with a single bio.
615 * - We pretend that we have just taken it off a longer list, so we assign
616 * bio_list to a pointer to the bio_list_on_stack, thus initialising the
617 * bio_list of new bios to be added. ->submit_bio() may indeed add some more
618 * bios through a recursive call to submit_bio_noacct. If it did, we find a
619 * non-NULL value in bio_list and re-enter the loop from the top.
620 * - In this case we really did just take the bio of the top of the list (no
621 * pretending) and so remove it from bio_list, and call into ->submit_bio()
622 * again.
623 *
624 * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio.
625 * bio_list_on_stack[1] contains bios that were submitted before the current
626 * ->submit_bio, but that haven't been processed yet.
627 */
628static void __submit_bio_noacct(struct bio *bio)
629{
630 struct bio_list bio_list_on_stack[2];
631
632 BUG_ON(bio->bi_next);
633
634 bio_list_init(&bio_list_on_stack[0]);
635 current->bio_list = bio_list_on_stack;
636
637 do {
638 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
639 struct bio_list lower, same;
640
641 /*
642 * Create a fresh bio_list for all subordinate requests.
643 */
644 bio_list_on_stack[1] = bio_list_on_stack[0];
645 bio_list_init(&bio_list_on_stack[0]);
646
647 __submit_bio(bio);
648
649 /*
650 * Sort new bios into those for a lower level and those for the
651 * same level.
652 */
653 bio_list_init(&lower);
654 bio_list_init(&same);
655 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
656 if (q == bdev_get_queue(bio->bi_bdev))
657 bio_list_add(&same, bio);
658 else
659 bio_list_add(&lower, bio);
660
661 /*
662 * Now assemble so we handle the lowest level first.
663 */
664 bio_list_merge(&bio_list_on_stack[0], &lower);
665 bio_list_merge(&bio_list_on_stack[0], &same);
666 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
667 } while ((bio = bio_list_pop(&bio_list_on_stack[0])));
668
669 current->bio_list = NULL;
670}
671
672static void __submit_bio_noacct_mq(struct bio *bio)
673{
674 struct bio_list bio_list[2] = { };
675
676 current->bio_list = bio_list;
677
678 do {
679 __submit_bio(bio);
680 } while ((bio = bio_list_pop(&bio_list[0])));
681
682 current->bio_list = NULL;
683}
684
685void submit_bio_noacct_nocheck(struct bio *bio)
686{
687 /*
688 * We only want one ->submit_bio to be active at a time, else stack
689 * usage with stacked devices could be a problem. Use current->bio_list
690 * to collect a list of requests submited by a ->submit_bio method while
691 * it is active, and then process them after it returned.
692 */
693 if (current->bio_list)
694 bio_list_add(¤t->bio_list[0], bio);
695 else if (!bio->bi_bdev->bd_disk->fops->submit_bio)
696 __submit_bio_noacct_mq(bio);
697 else
698 __submit_bio_noacct(bio);
699}
700
701/**
702 * submit_bio_noacct - re-submit a bio to the block device layer for I/O
703 * @bio: The bio describing the location in memory and on the device.
704 *
705 * This is a version of submit_bio() that shall only be used for I/O that is
706 * resubmitted to lower level drivers by stacking block drivers. All file
707 * systems and other upper level users of the block layer should use
708 * submit_bio() instead.
709 */
710void submit_bio_noacct(struct bio *bio)
711{
712 struct block_device *bdev = bio->bi_bdev;
713 struct request_queue *q = bdev_get_queue(bdev);
714 blk_status_t status = BLK_STS_IOERR;
715 struct blk_plug *plug;
716
717 might_sleep();
718
719 plug = blk_mq_plug(bio);
720 if (plug && plug->nowait)
721 bio->bi_opf |= REQ_NOWAIT;
722
723 /*
724 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
725 * if queue does not support NOWAIT.
726 */
727 if ((bio->bi_opf & REQ_NOWAIT) && !bdev_nowait(bdev))
728 goto not_supported;
729
730 if (should_fail_bio(bio))
731 goto end_io;
732 bio_check_ro(bio);
733 if (!bio_flagged(bio, BIO_REMAPPED)) {
734 if (unlikely(bio_check_eod(bio)))
735 goto end_io;
736 if (bdev->bd_partno && unlikely(blk_partition_remap(bio)))
737 goto end_io;
738 }
739
740 /*
741 * Filter flush bio's early so that bio based drivers without flush
742 * support don't have to worry about them.
743 */
744 if (op_is_flush(bio->bi_opf) &&
745 !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
746 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
747 if (!bio_sectors(bio)) {
748 status = BLK_STS_OK;
749 goto end_io;
750 }
751 }
752
753 if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
754 bio_clear_polled(bio);
755
756 switch (bio_op(bio)) {
757 case REQ_OP_DISCARD:
758 if (!bdev_max_discard_sectors(bdev))
759 goto not_supported;
760 break;
761 case REQ_OP_SECURE_ERASE:
762 if (!bdev_max_secure_erase_sectors(bdev))
763 goto not_supported;
764 break;
765 case REQ_OP_ZONE_APPEND:
766 status = blk_check_zone_append(q, bio);
767 if (status != BLK_STS_OK)
768 goto end_io;
769 break;
770 case REQ_OP_ZONE_RESET:
771 case REQ_OP_ZONE_OPEN:
772 case REQ_OP_ZONE_CLOSE:
773 case REQ_OP_ZONE_FINISH:
774 if (!bdev_is_zoned(bio->bi_bdev))
775 goto not_supported;
776 break;
777 case REQ_OP_ZONE_RESET_ALL:
778 if (!bdev_is_zoned(bio->bi_bdev) || !blk_queue_zone_resetall(q))
779 goto not_supported;
780 break;
781 case REQ_OP_WRITE_ZEROES:
782 if (!q->limits.max_write_zeroes_sectors)
783 goto not_supported;
784 break;
785 default:
786 break;
787 }
788
789 if (blk_throtl_bio(bio))
790 return;
791
792 blk_cgroup_bio_start(bio);
793 blkcg_bio_issue_init(bio);
794
795 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
796 trace_block_bio_queue(bio);
797 /* Now that enqueuing has been traced, we need to trace
798 * completion as well.
799 */
800 bio_set_flag(bio, BIO_TRACE_COMPLETION);
801 }
802 submit_bio_noacct_nocheck(bio);
803 return;
804
805not_supported:
806 status = BLK_STS_NOTSUPP;
807end_io:
808 bio->bi_status = status;
809 bio_endio(bio);
810}
811EXPORT_SYMBOL(submit_bio_noacct);
812
813/**
814 * submit_bio - submit a bio to the block device layer for I/O
815 * @bio: The &struct bio which describes the I/O
816 *
817 * submit_bio() is used to submit I/O requests to block devices. It is passed a
818 * fully set up &struct bio that describes the I/O that needs to be done. The
819 * bio will be send to the device described by the bi_bdev field.
820 *
821 * The success/failure status of the request, along with notification of
822 * completion, is delivered asynchronously through the ->bi_end_io() callback
823 * in @bio. The bio must NOT be touched by the caller until ->bi_end_io() has
824 * been called.
825 */
826void submit_bio(struct bio *bio)
827{
828 if (blkcg_punt_bio_submit(bio))
829 return;
830
831 if (bio_op(bio) == REQ_OP_READ) {
832 task_io_account_read(bio->bi_iter.bi_size);
833 count_vm_events(PGPGIN, bio_sectors(bio));
834 } else if (bio_op(bio) == REQ_OP_WRITE) {
835 count_vm_events(PGPGOUT, bio_sectors(bio));
836 }
837
838 submit_bio_noacct(bio);
839}
840EXPORT_SYMBOL(submit_bio);
841
842/**
843 * bio_poll - poll for BIO completions
844 * @bio: bio to poll for
845 * @iob: batches of IO
846 * @flags: BLK_POLL_* flags that control the behavior
847 *
848 * Poll for completions on queue associated with the bio. Returns number of
849 * completed entries found.
850 *
851 * Note: the caller must either be the context that submitted @bio, or
852 * be in a RCU critical section to prevent freeing of @bio.
853 */
854int bio_poll(struct bio *bio, struct io_comp_batch *iob, unsigned int flags)
855{
856 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
857 blk_qc_t cookie = READ_ONCE(bio->bi_cookie);
858 int ret = 0;
859
860 if (cookie == BLK_QC_T_NONE ||
861 !test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
862 return 0;
863
864 /*
865 * As the requests that require a zone lock are not plugged in the
866 * first place, directly accessing the plug instead of using
867 * blk_mq_plug() should not have any consequences during flushing for
868 * zoned devices.
869 */
870 blk_flush_plug(current->plug, false);
871
872 if (bio_queue_enter(bio))
873 return 0;
874 if (queue_is_mq(q)) {
875 ret = blk_mq_poll(q, cookie, iob, flags);
876 } else {
877 struct gendisk *disk = q->disk;
878
879 if (disk && disk->fops->poll_bio)
880 ret = disk->fops->poll_bio(bio, iob, flags);
881 }
882 blk_queue_exit(q);
883 return ret;
884}
885EXPORT_SYMBOL_GPL(bio_poll);
886
887/*
888 * Helper to implement file_operations.iopoll. Requires the bio to be stored
889 * in iocb->private, and cleared before freeing the bio.
890 */
891int iocb_bio_iopoll(struct kiocb *kiocb, struct io_comp_batch *iob,
892 unsigned int flags)
893{
894 struct bio *bio;
895 int ret = 0;
896
897 /*
898 * Note: the bio cache only uses SLAB_TYPESAFE_BY_RCU, so bio can
899 * point to a freshly allocated bio at this point. If that happens
900 * we have a few cases to consider:
901 *
902 * 1) the bio is beeing initialized and bi_bdev is NULL. We can just
903 * simply nothing in this case
904 * 2) the bio points to a not poll enabled device. bio_poll will catch
905 * this and return 0
906 * 3) the bio points to a poll capable device, including but not
907 * limited to the one that the original bio pointed to. In this
908 * case we will call into the actual poll method and poll for I/O,
909 * even if we don't need to, but it won't cause harm either.
910 *
911 * For cases 2) and 3) above the RCU grace period ensures that bi_bdev
912 * is still allocated. Because partitions hold a reference to the whole
913 * device bdev and thus disk, the disk is also still valid. Grabbing
914 * a reference to the queue in bio_poll() ensures the hctxs and requests
915 * are still valid as well.
916 */
917 rcu_read_lock();
918 bio = READ_ONCE(kiocb->private);
919 if (bio && bio->bi_bdev)
920 ret = bio_poll(bio, iob, flags);
921 rcu_read_unlock();
922
923 return ret;
924}
925EXPORT_SYMBOL_GPL(iocb_bio_iopoll);
926
927void update_io_ticks(struct block_device *part, unsigned long now, bool end)
928{
929 unsigned long stamp;
930again:
931 stamp = READ_ONCE(part->bd_stamp);
932 if (unlikely(time_after(now, stamp))) {
933 if (likely(try_cmpxchg(&part->bd_stamp, &stamp, now)))
934 __part_stat_add(part, io_ticks, end ? now - stamp : 1);
935 }
936 if (part->bd_partno) {
937 part = bdev_whole(part);
938 goto again;
939 }
940}
941
942unsigned long bdev_start_io_acct(struct block_device *bdev,
943 unsigned int sectors, enum req_op op,
944 unsigned long start_time)
945{
946 const int sgrp = op_stat_group(op);
947
948 part_stat_lock();
949 update_io_ticks(bdev, start_time, false);
950 part_stat_inc(bdev, ios[sgrp]);
951 part_stat_add(bdev, sectors[sgrp], sectors);
952 part_stat_local_inc(bdev, in_flight[op_is_write(op)]);
953 part_stat_unlock();
954
955 return start_time;
956}
957EXPORT_SYMBOL(bdev_start_io_acct);
958
959/**
960 * bio_start_io_acct - start I/O accounting for bio based drivers
961 * @bio: bio to start account for
962 *
963 * Returns the start time that should be passed back to bio_end_io_acct().
964 */
965unsigned long bio_start_io_acct(struct bio *bio)
966{
967 return bdev_start_io_acct(bio->bi_bdev, bio_sectors(bio),
968 bio_op(bio), jiffies);
969}
970EXPORT_SYMBOL_GPL(bio_start_io_acct);
971
972void bdev_end_io_acct(struct block_device *bdev, enum req_op op,
973 unsigned long start_time)
974{
975 const int sgrp = op_stat_group(op);
976 unsigned long now = READ_ONCE(jiffies);
977 unsigned long duration = now - start_time;
978
979 part_stat_lock();
980 update_io_ticks(bdev, now, true);
981 part_stat_add(bdev, nsecs[sgrp], jiffies_to_nsecs(duration));
982 part_stat_local_dec(bdev, in_flight[op_is_write(op)]);
983 part_stat_unlock();
984}
985EXPORT_SYMBOL(bdev_end_io_acct);
986
987void bio_end_io_acct_remapped(struct bio *bio, unsigned long start_time,
988 struct block_device *orig_bdev)
989{
990 bdev_end_io_acct(orig_bdev, bio_op(bio), start_time);
991}
992EXPORT_SYMBOL_GPL(bio_end_io_acct_remapped);
993
994/**
995 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
996 * @q : the queue of the device being checked
997 *
998 * Description:
999 * Check if underlying low-level drivers of a device are busy.
1000 * If the drivers want to export their busy state, they must set own
1001 * exporting function using blk_queue_lld_busy() first.
1002 *
1003 * Basically, this function is used only by request stacking drivers
1004 * to stop dispatching requests to underlying devices when underlying
1005 * devices are busy. This behavior helps more I/O merging on the queue
1006 * of the request stacking driver and prevents I/O throughput regression
1007 * on burst I/O load.
1008 *
1009 * Return:
1010 * 0 - Not busy (The request stacking driver should dispatch request)
1011 * 1 - Busy (The request stacking driver should stop dispatching request)
1012 */
1013int blk_lld_busy(struct request_queue *q)
1014{
1015 if (queue_is_mq(q) && q->mq_ops->busy)
1016 return q->mq_ops->busy(q);
1017
1018 return 0;
1019}
1020EXPORT_SYMBOL_GPL(blk_lld_busy);
1021
1022int kblockd_schedule_work(struct work_struct *work)
1023{
1024 return queue_work(kblockd_workqueue, work);
1025}
1026EXPORT_SYMBOL(kblockd_schedule_work);
1027
1028int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
1029 unsigned long delay)
1030{
1031 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
1032}
1033EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
1034
1035void blk_start_plug_nr_ios(struct blk_plug *plug, unsigned short nr_ios)
1036{
1037 struct task_struct *tsk = current;
1038
1039 /*
1040 * If this is a nested plug, don't actually assign it.
1041 */
1042 if (tsk->plug)
1043 return;
1044
1045 plug->mq_list = NULL;
1046 plug->cached_rq = NULL;
1047 plug->nr_ios = min_t(unsigned short, nr_ios, BLK_MAX_REQUEST_COUNT);
1048 plug->rq_count = 0;
1049 plug->multiple_queues = false;
1050 plug->has_elevator = false;
1051 plug->nowait = false;
1052 INIT_LIST_HEAD(&plug->cb_list);
1053
1054 /*
1055 * Store ordering should not be needed here, since a potential
1056 * preempt will imply a full memory barrier
1057 */
1058 tsk->plug = plug;
1059}
1060
1061/**
1062 * blk_start_plug - initialize blk_plug and track it inside the task_struct
1063 * @plug: The &struct blk_plug that needs to be initialized
1064 *
1065 * Description:
1066 * blk_start_plug() indicates to the block layer an intent by the caller
1067 * to submit multiple I/O requests in a batch. The block layer may use
1068 * this hint to defer submitting I/Os from the caller until blk_finish_plug()
1069 * is called. However, the block layer may choose to submit requests
1070 * before a call to blk_finish_plug() if the number of queued I/Os
1071 * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1072 * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if
1073 * the task schedules (see below).
1074 *
1075 * Tracking blk_plug inside the task_struct will help with auto-flushing the
1076 * pending I/O should the task end up blocking between blk_start_plug() and
1077 * blk_finish_plug(). This is important from a performance perspective, but
1078 * also ensures that we don't deadlock. For instance, if the task is blocking
1079 * for a memory allocation, memory reclaim could end up wanting to free a
1080 * page belonging to that request that is currently residing in our private
1081 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
1082 * this kind of deadlock.
1083 */
1084void blk_start_plug(struct blk_plug *plug)
1085{
1086 blk_start_plug_nr_ios(plug, 1);
1087}
1088EXPORT_SYMBOL(blk_start_plug);
1089
1090static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
1091{
1092 LIST_HEAD(callbacks);
1093
1094 while (!list_empty(&plug->cb_list)) {
1095 list_splice_init(&plug->cb_list, &callbacks);
1096
1097 while (!list_empty(&callbacks)) {
1098 struct blk_plug_cb *cb = list_first_entry(&callbacks,
1099 struct blk_plug_cb,
1100 list);
1101 list_del(&cb->list);
1102 cb->callback(cb, from_schedule);
1103 }
1104 }
1105}
1106
1107struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
1108 int size)
1109{
1110 struct blk_plug *plug = current->plug;
1111 struct blk_plug_cb *cb;
1112
1113 if (!plug)
1114 return NULL;
1115
1116 list_for_each_entry(cb, &plug->cb_list, list)
1117 if (cb->callback == unplug && cb->data == data)
1118 return cb;
1119
1120 /* Not currently on the callback list */
1121 BUG_ON(size < sizeof(*cb));
1122 cb = kzalloc(size, GFP_ATOMIC);
1123 if (cb) {
1124 cb->data = data;
1125 cb->callback = unplug;
1126 list_add(&cb->list, &plug->cb_list);
1127 }
1128 return cb;
1129}
1130EXPORT_SYMBOL(blk_check_plugged);
1131
1132void __blk_flush_plug(struct blk_plug *plug, bool from_schedule)
1133{
1134 if (!list_empty(&plug->cb_list))
1135 flush_plug_callbacks(plug, from_schedule);
1136 if (!rq_list_empty(plug->mq_list))
1137 blk_mq_flush_plug_list(plug, from_schedule);
1138 /*
1139 * Unconditionally flush out cached requests, even if the unplug
1140 * event came from schedule. Since we know hold references to the
1141 * queue for cached requests, we don't want a blocked task holding
1142 * up a queue freeze/quiesce event.
1143 */
1144 if (unlikely(!rq_list_empty(plug->cached_rq)))
1145 blk_mq_free_plug_rqs(plug);
1146}
1147
1148/**
1149 * blk_finish_plug - mark the end of a batch of submitted I/O
1150 * @plug: The &struct blk_plug passed to blk_start_plug()
1151 *
1152 * Description:
1153 * Indicate that a batch of I/O submissions is complete. This function
1154 * must be paired with an initial call to blk_start_plug(). The intent
1155 * is to allow the block layer to optimize I/O submission. See the
1156 * documentation for blk_start_plug() for more information.
1157 */
1158void blk_finish_plug(struct blk_plug *plug)
1159{
1160 if (plug == current->plug) {
1161 __blk_flush_plug(plug, false);
1162 current->plug = NULL;
1163 }
1164}
1165EXPORT_SYMBOL(blk_finish_plug);
1166
1167void blk_io_schedule(void)
1168{
1169 /* Prevent hang_check timer from firing at us during very long I/O */
1170 unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2;
1171
1172 if (timeout)
1173 io_schedule_timeout(timeout);
1174 else
1175 io_schedule();
1176}
1177EXPORT_SYMBOL_GPL(blk_io_schedule);
1178
1179int __init blk_dev_init(void)
1180{
1181 BUILD_BUG_ON((__force u32)REQ_OP_LAST >= (1 << REQ_OP_BITS));
1182 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1183 sizeof_field(struct request, cmd_flags));
1184 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1185 sizeof_field(struct bio, bi_opf));
1186
1187 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
1188 kblockd_workqueue = alloc_workqueue("kblockd",
1189 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1190 if (!kblockd_workqueue)
1191 panic("Failed to create kblockd\n");
1192
1193 blk_requestq_cachep = kmem_cache_create("request_queue",
1194 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
1195
1196 blk_debugfs_root = debugfs_create_dir("block", NULL);
1197
1198 return 0;
1199}
1/*
2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
6 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
7 * - July2000
8 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
9 */
10
11/*
12 * This handles all read/write requests to block devices
13 */
14#include <linux/kernel.h>
15#include <linux/module.h>
16#include <linux/backing-dev.h>
17#include <linux/bio.h>
18#include <linux/blkdev.h>
19#include <linux/blk-mq.h>
20#include <linux/highmem.h>
21#include <linux/mm.h>
22#include <linux/kernel_stat.h>
23#include <linux/string.h>
24#include <linux/init.h>
25#include <linux/completion.h>
26#include <linux/slab.h>
27#include <linux/swap.h>
28#include <linux/writeback.h>
29#include <linux/task_io_accounting_ops.h>
30#include <linux/fault-inject.h>
31#include <linux/list_sort.h>
32#include <linux/delay.h>
33#include <linux/ratelimit.h>
34#include <linux/pm_runtime.h>
35#include <linux/blk-cgroup.h>
36
37#define CREATE_TRACE_POINTS
38#include <trace/events/block.h>
39
40#include "blk.h"
41#include "blk-mq.h"
42#include "blk-wbt.h"
43
44EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
45EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
46EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
47EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
48EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
49
50DEFINE_IDA(blk_queue_ida);
51
52/*
53 * For the allocated request tables
54 */
55struct kmem_cache *request_cachep;
56
57/*
58 * For queue allocation
59 */
60struct kmem_cache *blk_requestq_cachep;
61
62/*
63 * Controlling structure to kblockd
64 */
65static struct workqueue_struct *kblockd_workqueue;
66
67static void blk_clear_congested(struct request_list *rl, int sync)
68{
69#ifdef CONFIG_CGROUP_WRITEBACK
70 clear_wb_congested(rl->blkg->wb_congested, sync);
71#else
72 /*
73 * If !CGROUP_WRITEBACK, all blkg's map to bdi->wb and we shouldn't
74 * flip its congestion state for events on other blkcgs.
75 */
76 if (rl == &rl->q->root_rl)
77 clear_wb_congested(rl->q->backing_dev_info.wb.congested, sync);
78#endif
79}
80
81static void blk_set_congested(struct request_list *rl, int sync)
82{
83#ifdef CONFIG_CGROUP_WRITEBACK
84 set_wb_congested(rl->blkg->wb_congested, sync);
85#else
86 /* see blk_clear_congested() */
87 if (rl == &rl->q->root_rl)
88 set_wb_congested(rl->q->backing_dev_info.wb.congested, sync);
89#endif
90}
91
92void blk_queue_congestion_threshold(struct request_queue *q)
93{
94 int nr;
95
96 nr = q->nr_requests - (q->nr_requests / 8) + 1;
97 if (nr > q->nr_requests)
98 nr = q->nr_requests;
99 q->nr_congestion_on = nr;
100
101 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
102 if (nr < 1)
103 nr = 1;
104 q->nr_congestion_off = nr;
105}
106
107/**
108 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
109 * @bdev: device
110 *
111 * Locates the passed device's request queue and returns the address of its
112 * backing_dev_info. This function can only be called if @bdev is opened
113 * and the return value is never NULL.
114 */
115struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
116{
117 struct request_queue *q = bdev_get_queue(bdev);
118
119 return &q->backing_dev_info;
120}
121EXPORT_SYMBOL(blk_get_backing_dev_info);
122
123void blk_rq_init(struct request_queue *q, struct request *rq)
124{
125 memset(rq, 0, sizeof(*rq));
126
127 INIT_LIST_HEAD(&rq->queuelist);
128 INIT_LIST_HEAD(&rq->timeout_list);
129 rq->cpu = -1;
130 rq->q = q;
131 rq->__sector = (sector_t) -1;
132 INIT_HLIST_NODE(&rq->hash);
133 RB_CLEAR_NODE(&rq->rb_node);
134 rq->cmd = rq->__cmd;
135 rq->cmd_len = BLK_MAX_CDB;
136 rq->tag = -1;
137 rq->start_time = jiffies;
138 set_start_time_ns(rq);
139 rq->part = NULL;
140}
141EXPORT_SYMBOL(blk_rq_init);
142
143static void req_bio_endio(struct request *rq, struct bio *bio,
144 unsigned int nbytes, int error)
145{
146 if (error)
147 bio->bi_error = error;
148
149 if (unlikely(rq->rq_flags & RQF_QUIET))
150 bio_set_flag(bio, BIO_QUIET);
151
152 bio_advance(bio, nbytes);
153
154 /* don't actually finish bio if it's part of flush sequence */
155 if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ))
156 bio_endio(bio);
157}
158
159void blk_dump_rq_flags(struct request *rq, char *msg)
160{
161 int bit;
162
163 printk(KERN_INFO "%s: dev %s: type=%x, flags=%llx\n", msg,
164 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
165 (unsigned long long) rq->cmd_flags);
166
167 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
168 (unsigned long long)blk_rq_pos(rq),
169 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
170 printk(KERN_INFO " bio %p, biotail %p, len %u\n",
171 rq->bio, rq->biotail, blk_rq_bytes(rq));
172
173 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
174 printk(KERN_INFO " cdb: ");
175 for (bit = 0; bit < BLK_MAX_CDB; bit++)
176 printk("%02x ", rq->cmd[bit]);
177 printk("\n");
178 }
179}
180EXPORT_SYMBOL(blk_dump_rq_flags);
181
182static void blk_delay_work(struct work_struct *work)
183{
184 struct request_queue *q;
185
186 q = container_of(work, struct request_queue, delay_work.work);
187 spin_lock_irq(q->queue_lock);
188 __blk_run_queue(q);
189 spin_unlock_irq(q->queue_lock);
190}
191
192/**
193 * blk_delay_queue - restart queueing after defined interval
194 * @q: The &struct request_queue in question
195 * @msecs: Delay in msecs
196 *
197 * Description:
198 * Sometimes queueing needs to be postponed for a little while, to allow
199 * resources to come back. This function will make sure that queueing is
200 * restarted around the specified time. Queue lock must be held.
201 */
202void blk_delay_queue(struct request_queue *q, unsigned long msecs)
203{
204 if (likely(!blk_queue_dead(q)))
205 queue_delayed_work(kblockd_workqueue, &q->delay_work,
206 msecs_to_jiffies(msecs));
207}
208EXPORT_SYMBOL(blk_delay_queue);
209
210/**
211 * blk_start_queue_async - asynchronously restart a previously stopped queue
212 * @q: The &struct request_queue in question
213 *
214 * Description:
215 * blk_start_queue_async() will clear the stop flag on the queue, and
216 * ensure that the request_fn for the queue is run from an async
217 * context.
218 **/
219void blk_start_queue_async(struct request_queue *q)
220{
221 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
222 blk_run_queue_async(q);
223}
224EXPORT_SYMBOL(blk_start_queue_async);
225
226/**
227 * blk_start_queue - restart a previously stopped queue
228 * @q: The &struct request_queue in question
229 *
230 * Description:
231 * blk_start_queue() will clear the stop flag on the queue, and call
232 * the request_fn for the queue if it was in a stopped state when
233 * entered. Also see blk_stop_queue(). Queue lock must be held.
234 **/
235void blk_start_queue(struct request_queue *q)
236{
237 WARN_ON(!irqs_disabled());
238
239 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
240 __blk_run_queue(q);
241}
242EXPORT_SYMBOL(blk_start_queue);
243
244/**
245 * blk_stop_queue - stop a queue
246 * @q: The &struct request_queue in question
247 *
248 * Description:
249 * The Linux block layer assumes that a block driver will consume all
250 * entries on the request queue when the request_fn strategy is called.
251 * Often this will not happen, because of hardware limitations (queue
252 * depth settings). If a device driver gets a 'queue full' response,
253 * or if it simply chooses not to queue more I/O at one point, it can
254 * call this function to prevent the request_fn from being called until
255 * the driver has signalled it's ready to go again. This happens by calling
256 * blk_start_queue() to restart queue operations. Queue lock must be held.
257 **/
258void blk_stop_queue(struct request_queue *q)
259{
260 cancel_delayed_work(&q->delay_work);
261 queue_flag_set(QUEUE_FLAG_STOPPED, q);
262}
263EXPORT_SYMBOL(blk_stop_queue);
264
265/**
266 * blk_sync_queue - cancel any pending callbacks on a queue
267 * @q: the queue
268 *
269 * Description:
270 * The block layer may perform asynchronous callback activity
271 * on a queue, such as calling the unplug function after a timeout.
272 * A block device may call blk_sync_queue to ensure that any
273 * such activity is cancelled, thus allowing it to release resources
274 * that the callbacks might use. The caller must already have made sure
275 * that its ->make_request_fn will not re-add plugging prior to calling
276 * this function.
277 *
278 * This function does not cancel any asynchronous activity arising
279 * out of elevator or throttling code. That would require elevator_exit()
280 * and blkcg_exit_queue() to be called with queue lock initialized.
281 *
282 */
283void blk_sync_queue(struct request_queue *q)
284{
285 del_timer_sync(&q->timeout);
286
287 if (q->mq_ops) {
288 struct blk_mq_hw_ctx *hctx;
289 int i;
290
291 queue_for_each_hw_ctx(q, hctx, i) {
292 cancel_work_sync(&hctx->run_work);
293 cancel_delayed_work_sync(&hctx->delay_work);
294 }
295 } else {
296 cancel_delayed_work_sync(&q->delay_work);
297 }
298}
299EXPORT_SYMBOL(blk_sync_queue);
300
301/**
302 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
303 * @q: The queue to run
304 *
305 * Description:
306 * Invoke request handling on a queue if there are any pending requests.
307 * May be used to restart request handling after a request has completed.
308 * This variant runs the queue whether or not the queue has been
309 * stopped. Must be called with the queue lock held and interrupts
310 * disabled. See also @blk_run_queue.
311 */
312inline void __blk_run_queue_uncond(struct request_queue *q)
313{
314 if (unlikely(blk_queue_dead(q)))
315 return;
316
317 /*
318 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
319 * the queue lock internally. As a result multiple threads may be
320 * running such a request function concurrently. Keep track of the
321 * number of active request_fn invocations such that blk_drain_queue()
322 * can wait until all these request_fn calls have finished.
323 */
324 q->request_fn_active++;
325 q->request_fn(q);
326 q->request_fn_active--;
327}
328EXPORT_SYMBOL_GPL(__blk_run_queue_uncond);
329
330/**
331 * __blk_run_queue - run a single device queue
332 * @q: The queue to run
333 *
334 * Description:
335 * See @blk_run_queue. This variant must be called with the queue lock
336 * held and interrupts disabled.
337 */
338void __blk_run_queue(struct request_queue *q)
339{
340 if (unlikely(blk_queue_stopped(q)))
341 return;
342
343 __blk_run_queue_uncond(q);
344}
345EXPORT_SYMBOL(__blk_run_queue);
346
347/**
348 * blk_run_queue_async - run a single device queue in workqueue context
349 * @q: The queue to run
350 *
351 * Description:
352 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
353 * of us. The caller must hold the queue lock.
354 */
355void blk_run_queue_async(struct request_queue *q)
356{
357 if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
358 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
359}
360EXPORT_SYMBOL(blk_run_queue_async);
361
362/**
363 * blk_run_queue - run a single device queue
364 * @q: The queue to run
365 *
366 * Description:
367 * Invoke request handling on this queue, if it has pending work to do.
368 * May be used to restart queueing when a request has completed.
369 */
370void blk_run_queue(struct request_queue *q)
371{
372 unsigned long flags;
373
374 spin_lock_irqsave(q->queue_lock, flags);
375 __blk_run_queue(q);
376 spin_unlock_irqrestore(q->queue_lock, flags);
377}
378EXPORT_SYMBOL(blk_run_queue);
379
380void blk_put_queue(struct request_queue *q)
381{
382 kobject_put(&q->kobj);
383}
384EXPORT_SYMBOL(blk_put_queue);
385
386/**
387 * __blk_drain_queue - drain requests from request_queue
388 * @q: queue to drain
389 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
390 *
391 * Drain requests from @q. If @drain_all is set, all requests are drained.
392 * If not, only ELVPRIV requests are drained. The caller is responsible
393 * for ensuring that no new requests which need to be drained are queued.
394 */
395static void __blk_drain_queue(struct request_queue *q, bool drain_all)
396 __releases(q->queue_lock)
397 __acquires(q->queue_lock)
398{
399 int i;
400
401 lockdep_assert_held(q->queue_lock);
402
403 while (true) {
404 bool drain = false;
405
406 /*
407 * The caller might be trying to drain @q before its
408 * elevator is initialized.
409 */
410 if (q->elevator)
411 elv_drain_elevator(q);
412
413 blkcg_drain_queue(q);
414
415 /*
416 * This function might be called on a queue which failed
417 * driver init after queue creation or is not yet fully
418 * active yet. Some drivers (e.g. fd and loop) get unhappy
419 * in such cases. Kick queue iff dispatch queue has
420 * something on it and @q has request_fn set.
421 */
422 if (!list_empty(&q->queue_head) && q->request_fn)
423 __blk_run_queue(q);
424
425 drain |= q->nr_rqs_elvpriv;
426 drain |= q->request_fn_active;
427
428 /*
429 * Unfortunately, requests are queued at and tracked from
430 * multiple places and there's no single counter which can
431 * be drained. Check all the queues and counters.
432 */
433 if (drain_all) {
434 struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
435 drain |= !list_empty(&q->queue_head);
436 for (i = 0; i < 2; i++) {
437 drain |= q->nr_rqs[i];
438 drain |= q->in_flight[i];
439 if (fq)
440 drain |= !list_empty(&fq->flush_queue[i]);
441 }
442 }
443
444 if (!drain)
445 break;
446
447 spin_unlock_irq(q->queue_lock);
448
449 msleep(10);
450
451 spin_lock_irq(q->queue_lock);
452 }
453
454 /*
455 * With queue marked dead, any woken up waiter will fail the
456 * allocation path, so the wakeup chaining is lost and we're
457 * left with hung waiters. We need to wake up those waiters.
458 */
459 if (q->request_fn) {
460 struct request_list *rl;
461
462 blk_queue_for_each_rl(rl, q)
463 for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
464 wake_up_all(&rl->wait[i]);
465 }
466}
467
468/**
469 * blk_queue_bypass_start - enter queue bypass mode
470 * @q: queue of interest
471 *
472 * In bypass mode, only the dispatch FIFO queue of @q is used. This
473 * function makes @q enter bypass mode and drains all requests which were
474 * throttled or issued before. On return, it's guaranteed that no request
475 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
476 * inside queue or RCU read lock.
477 */
478void blk_queue_bypass_start(struct request_queue *q)
479{
480 spin_lock_irq(q->queue_lock);
481 q->bypass_depth++;
482 queue_flag_set(QUEUE_FLAG_BYPASS, q);
483 spin_unlock_irq(q->queue_lock);
484
485 /*
486 * Queues start drained. Skip actual draining till init is
487 * complete. This avoids lenghty delays during queue init which
488 * can happen many times during boot.
489 */
490 if (blk_queue_init_done(q)) {
491 spin_lock_irq(q->queue_lock);
492 __blk_drain_queue(q, false);
493 spin_unlock_irq(q->queue_lock);
494
495 /* ensure blk_queue_bypass() is %true inside RCU read lock */
496 synchronize_rcu();
497 }
498}
499EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
500
501/**
502 * blk_queue_bypass_end - leave queue bypass mode
503 * @q: queue of interest
504 *
505 * Leave bypass mode and restore the normal queueing behavior.
506 */
507void blk_queue_bypass_end(struct request_queue *q)
508{
509 spin_lock_irq(q->queue_lock);
510 if (!--q->bypass_depth)
511 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
512 WARN_ON_ONCE(q->bypass_depth < 0);
513 spin_unlock_irq(q->queue_lock);
514}
515EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
516
517void blk_set_queue_dying(struct request_queue *q)
518{
519 spin_lock_irq(q->queue_lock);
520 queue_flag_set(QUEUE_FLAG_DYING, q);
521 spin_unlock_irq(q->queue_lock);
522
523 if (q->mq_ops)
524 blk_mq_wake_waiters(q);
525 else {
526 struct request_list *rl;
527
528 blk_queue_for_each_rl(rl, q) {
529 if (rl->rq_pool) {
530 wake_up(&rl->wait[BLK_RW_SYNC]);
531 wake_up(&rl->wait[BLK_RW_ASYNC]);
532 }
533 }
534 }
535}
536EXPORT_SYMBOL_GPL(blk_set_queue_dying);
537
538/**
539 * blk_cleanup_queue - shutdown a request queue
540 * @q: request queue to shutdown
541 *
542 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
543 * put it. All future requests will be failed immediately with -ENODEV.
544 */
545void blk_cleanup_queue(struct request_queue *q)
546{
547 spinlock_t *lock = q->queue_lock;
548
549 /* mark @q DYING, no new request or merges will be allowed afterwards */
550 mutex_lock(&q->sysfs_lock);
551 blk_set_queue_dying(q);
552 spin_lock_irq(lock);
553
554 /*
555 * A dying queue is permanently in bypass mode till released. Note
556 * that, unlike blk_queue_bypass_start(), we aren't performing
557 * synchronize_rcu() after entering bypass mode to avoid the delay
558 * as some drivers create and destroy a lot of queues while
559 * probing. This is still safe because blk_release_queue() will be
560 * called only after the queue refcnt drops to zero and nothing,
561 * RCU or not, would be traversing the queue by then.
562 */
563 q->bypass_depth++;
564 queue_flag_set(QUEUE_FLAG_BYPASS, q);
565
566 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
567 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
568 queue_flag_set(QUEUE_FLAG_DYING, q);
569 spin_unlock_irq(lock);
570 mutex_unlock(&q->sysfs_lock);
571
572 /*
573 * Drain all requests queued before DYING marking. Set DEAD flag to
574 * prevent that q->request_fn() gets invoked after draining finished.
575 */
576 blk_freeze_queue(q);
577 spin_lock_irq(lock);
578 if (!q->mq_ops)
579 __blk_drain_queue(q, true);
580 queue_flag_set(QUEUE_FLAG_DEAD, q);
581 spin_unlock_irq(lock);
582
583 /* for synchronous bio-based driver finish in-flight integrity i/o */
584 blk_flush_integrity();
585
586 /* @q won't process any more request, flush async actions */
587 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
588 blk_sync_queue(q);
589
590 if (q->mq_ops)
591 blk_mq_free_queue(q);
592 percpu_ref_exit(&q->q_usage_counter);
593
594 spin_lock_irq(lock);
595 if (q->queue_lock != &q->__queue_lock)
596 q->queue_lock = &q->__queue_lock;
597 spin_unlock_irq(lock);
598
599 bdi_unregister(&q->backing_dev_info);
600
601 /* @q is and will stay empty, shutdown and put */
602 blk_put_queue(q);
603}
604EXPORT_SYMBOL(blk_cleanup_queue);
605
606/* Allocate memory local to the request queue */
607static void *alloc_request_struct(gfp_t gfp_mask, void *data)
608{
609 int nid = (int)(long)data;
610 return kmem_cache_alloc_node(request_cachep, gfp_mask, nid);
611}
612
613static void free_request_struct(void *element, void *unused)
614{
615 kmem_cache_free(request_cachep, element);
616}
617
618int blk_init_rl(struct request_list *rl, struct request_queue *q,
619 gfp_t gfp_mask)
620{
621 if (unlikely(rl->rq_pool))
622 return 0;
623
624 rl->q = q;
625 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
626 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
627 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
628 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
629
630 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, alloc_request_struct,
631 free_request_struct,
632 (void *)(long)q->node, gfp_mask,
633 q->node);
634 if (!rl->rq_pool)
635 return -ENOMEM;
636
637 return 0;
638}
639
640void blk_exit_rl(struct request_list *rl)
641{
642 if (rl->rq_pool)
643 mempool_destroy(rl->rq_pool);
644}
645
646struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
647{
648 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
649}
650EXPORT_SYMBOL(blk_alloc_queue);
651
652int blk_queue_enter(struct request_queue *q, bool nowait)
653{
654 while (true) {
655 int ret;
656
657 if (percpu_ref_tryget_live(&q->q_usage_counter))
658 return 0;
659
660 if (nowait)
661 return -EBUSY;
662
663 ret = wait_event_interruptible(q->mq_freeze_wq,
664 !atomic_read(&q->mq_freeze_depth) ||
665 blk_queue_dying(q));
666 if (blk_queue_dying(q))
667 return -ENODEV;
668 if (ret)
669 return ret;
670 }
671}
672
673void blk_queue_exit(struct request_queue *q)
674{
675 percpu_ref_put(&q->q_usage_counter);
676}
677
678static void blk_queue_usage_counter_release(struct percpu_ref *ref)
679{
680 struct request_queue *q =
681 container_of(ref, struct request_queue, q_usage_counter);
682
683 wake_up_all(&q->mq_freeze_wq);
684}
685
686static void blk_rq_timed_out_timer(unsigned long data)
687{
688 struct request_queue *q = (struct request_queue *)data;
689
690 kblockd_schedule_work(&q->timeout_work);
691}
692
693struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
694{
695 struct request_queue *q;
696 int err;
697
698 q = kmem_cache_alloc_node(blk_requestq_cachep,
699 gfp_mask | __GFP_ZERO, node_id);
700 if (!q)
701 return NULL;
702
703 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
704 if (q->id < 0)
705 goto fail_q;
706
707 q->bio_split = bioset_create(BIO_POOL_SIZE, 0);
708 if (!q->bio_split)
709 goto fail_id;
710
711 q->backing_dev_info.ra_pages =
712 (VM_MAX_READAHEAD * 1024) / PAGE_SIZE;
713 q->backing_dev_info.capabilities = BDI_CAP_CGROUP_WRITEBACK;
714 q->backing_dev_info.name = "block";
715 q->node = node_id;
716
717 err = bdi_init(&q->backing_dev_info);
718 if (err)
719 goto fail_split;
720
721 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
722 laptop_mode_timer_fn, (unsigned long) q);
723 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
724 INIT_LIST_HEAD(&q->queue_head);
725 INIT_LIST_HEAD(&q->timeout_list);
726 INIT_LIST_HEAD(&q->icq_list);
727#ifdef CONFIG_BLK_CGROUP
728 INIT_LIST_HEAD(&q->blkg_list);
729#endif
730 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
731
732 kobject_init(&q->kobj, &blk_queue_ktype);
733
734 mutex_init(&q->sysfs_lock);
735 spin_lock_init(&q->__queue_lock);
736
737 /*
738 * By default initialize queue_lock to internal lock and driver can
739 * override it later if need be.
740 */
741 q->queue_lock = &q->__queue_lock;
742
743 /*
744 * A queue starts its life with bypass turned on to avoid
745 * unnecessary bypass on/off overhead and nasty surprises during
746 * init. The initial bypass will be finished when the queue is
747 * registered by blk_register_queue().
748 */
749 q->bypass_depth = 1;
750 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
751
752 init_waitqueue_head(&q->mq_freeze_wq);
753
754 /*
755 * Init percpu_ref in atomic mode so that it's faster to shutdown.
756 * See blk_register_queue() for details.
757 */
758 if (percpu_ref_init(&q->q_usage_counter,
759 blk_queue_usage_counter_release,
760 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
761 goto fail_bdi;
762
763 if (blkcg_init_queue(q))
764 goto fail_ref;
765
766 return q;
767
768fail_ref:
769 percpu_ref_exit(&q->q_usage_counter);
770fail_bdi:
771 bdi_destroy(&q->backing_dev_info);
772fail_split:
773 bioset_free(q->bio_split);
774fail_id:
775 ida_simple_remove(&blk_queue_ida, q->id);
776fail_q:
777 kmem_cache_free(blk_requestq_cachep, q);
778 return NULL;
779}
780EXPORT_SYMBOL(blk_alloc_queue_node);
781
782/**
783 * blk_init_queue - prepare a request queue for use with a block device
784 * @rfn: The function to be called to process requests that have been
785 * placed on the queue.
786 * @lock: Request queue spin lock
787 *
788 * Description:
789 * If a block device wishes to use the standard request handling procedures,
790 * which sorts requests and coalesces adjacent requests, then it must
791 * call blk_init_queue(). The function @rfn will be called when there
792 * are requests on the queue that need to be processed. If the device
793 * supports plugging, then @rfn may not be called immediately when requests
794 * are available on the queue, but may be called at some time later instead.
795 * Plugged queues are generally unplugged when a buffer belonging to one
796 * of the requests on the queue is needed, or due to memory pressure.
797 *
798 * @rfn is not required, or even expected, to remove all requests off the
799 * queue, but only as many as it can handle at a time. If it does leave
800 * requests on the queue, it is responsible for arranging that the requests
801 * get dealt with eventually.
802 *
803 * The queue spin lock must be held while manipulating the requests on the
804 * request queue; this lock will be taken also from interrupt context, so irq
805 * disabling is needed for it.
806 *
807 * Function returns a pointer to the initialized request queue, or %NULL if
808 * it didn't succeed.
809 *
810 * Note:
811 * blk_init_queue() must be paired with a blk_cleanup_queue() call
812 * when the block device is deactivated (such as at module unload).
813 **/
814
815struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
816{
817 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
818}
819EXPORT_SYMBOL(blk_init_queue);
820
821struct request_queue *
822blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
823{
824 struct request_queue *uninit_q, *q;
825
826 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
827 if (!uninit_q)
828 return NULL;
829
830 q = blk_init_allocated_queue(uninit_q, rfn, lock);
831 if (!q)
832 blk_cleanup_queue(uninit_q);
833
834 return q;
835}
836EXPORT_SYMBOL(blk_init_queue_node);
837
838static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio);
839
840struct request_queue *
841blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
842 spinlock_t *lock)
843{
844 if (!q)
845 return NULL;
846
847 q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, 0);
848 if (!q->fq)
849 return NULL;
850
851 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
852 goto fail;
853
854 INIT_WORK(&q->timeout_work, blk_timeout_work);
855 q->request_fn = rfn;
856 q->prep_rq_fn = NULL;
857 q->unprep_rq_fn = NULL;
858 q->queue_flags |= QUEUE_FLAG_DEFAULT;
859
860 /* Override internal queue lock with supplied lock pointer */
861 if (lock)
862 q->queue_lock = lock;
863
864 /*
865 * This also sets hw/phys segments, boundary and size
866 */
867 blk_queue_make_request(q, blk_queue_bio);
868
869 q->sg_reserved_size = INT_MAX;
870
871 /* Protect q->elevator from elevator_change */
872 mutex_lock(&q->sysfs_lock);
873
874 /* init elevator */
875 if (elevator_init(q, NULL)) {
876 mutex_unlock(&q->sysfs_lock);
877 goto fail;
878 }
879
880 mutex_unlock(&q->sysfs_lock);
881
882 return q;
883
884fail:
885 blk_free_flush_queue(q->fq);
886 wbt_exit(q);
887 return NULL;
888}
889EXPORT_SYMBOL(blk_init_allocated_queue);
890
891bool blk_get_queue(struct request_queue *q)
892{
893 if (likely(!blk_queue_dying(q))) {
894 __blk_get_queue(q);
895 return true;
896 }
897
898 return false;
899}
900EXPORT_SYMBOL(blk_get_queue);
901
902static inline void blk_free_request(struct request_list *rl, struct request *rq)
903{
904 if (rq->rq_flags & RQF_ELVPRIV) {
905 elv_put_request(rl->q, rq);
906 if (rq->elv.icq)
907 put_io_context(rq->elv.icq->ioc);
908 }
909
910 mempool_free(rq, rl->rq_pool);
911}
912
913/*
914 * ioc_batching returns true if the ioc is a valid batching request and
915 * should be given priority access to a request.
916 */
917static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
918{
919 if (!ioc)
920 return 0;
921
922 /*
923 * Make sure the process is able to allocate at least 1 request
924 * even if the batch times out, otherwise we could theoretically
925 * lose wakeups.
926 */
927 return ioc->nr_batch_requests == q->nr_batching ||
928 (ioc->nr_batch_requests > 0
929 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
930}
931
932/*
933 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
934 * will cause the process to be a "batcher" on all queues in the system. This
935 * is the behaviour we want though - once it gets a wakeup it should be given
936 * a nice run.
937 */
938static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
939{
940 if (!ioc || ioc_batching(q, ioc))
941 return;
942
943 ioc->nr_batch_requests = q->nr_batching;
944 ioc->last_waited = jiffies;
945}
946
947static void __freed_request(struct request_list *rl, int sync)
948{
949 struct request_queue *q = rl->q;
950
951 if (rl->count[sync] < queue_congestion_off_threshold(q))
952 blk_clear_congested(rl, sync);
953
954 if (rl->count[sync] + 1 <= q->nr_requests) {
955 if (waitqueue_active(&rl->wait[sync]))
956 wake_up(&rl->wait[sync]);
957
958 blk_clear_rl_full(rl, sync);
959 }
960}
961
962/*
963 * A request has just been released. Account for it, update the full and
964 * congestion status, wake up any waiters. Called under q->queue_lock.
965 */
966static void freed_request(struct request_list *rl, bool sync,
967 req_flags_t rq_flags)
968{
969 struct request_queue *q = rl->q;
970
971 q->nr_rqs[sync]--;
972 rl->count[sync]--;
973 if (rq_flags & RQF_ELVPRIV)
974 q->nr_rqs_elvpriv--;
975
976 __freed_request(rl, sync);
977
978 if (unlikely(rl->starved[sync ^ 1]))
979 __freed_request(rl, sync ^ 1);
980}
981
982int blk_update_nr_requests(struct request_queue *q, unsigned int nr)
983{
984 struct request_list *rl;
985 int on_thresh, off_thresh;
986
987 spin_lock_irq(q->queue_lock);
988 q->nr_requests = nr;
989 blk_queue_congestion_threshold(q);
990 on_thresh = queue_congestion_on_threshold(q);
991 off_thresh = queue_congestion_off_threshold(q);
992
993 blk_queue_for_each_rl(rl, q) {
994 if (rl->count[BLK_RW_SYNC] >= on_thresh)
995 blk_set_congested(rl, BLK_RW_SYNC);
996 else if (rl->count[BLK_RW_SYNC] < off_thresh)
997 blk_clear_congested(rl, BLK_RW_SYNC);
998
999 if (rl->count[BLK_RW_ASYNC] >= on_thresh)
1000 blk_set_congested(rl, BLK_RW_ASYNC);
1001 else if (rl->count[BLK_RW_ASYNC] < off_thresh)
1002 blk_clear_congested(rl, BLK_RW_ASYNC);
1003
1004 if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
1005 blk_set_rl_full(rl, BLK_RW_SYNC);
1006 } else {
1007 blk_clear_rl_full(rl, BLK_RW_SYNC);
1008 wake_up(&rl->wait[BLK_RW_SYNC]);
1009 }
1010
1011 if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
1012 blk_set_rl_full(rl, BLK_RW_ASYNC);
1013 } else {
1014 blk_clear_rl_full(rl, BLK_RW_ASYNC);
1015 wake_up(&rl->wait[BLK_RW_ASYNC]);
1016 }
1017 }
1018
1019 spin_unlock_irq(q->queue_lock);
1020 return 0;
1021}
1022
1023/*
1024 * Determine if elevator data should be initialized when allocating the
1025 * request associated with @bio.
1026 */
1027static bool blk_rq_should_init_elevator(struct bio *bio)
1028{
1029 if (!bio)
1030 return true;
1031
1032 /*
1033 * Flush requests do not use the elevator so skip initialization.
1034 * This allows a request to share the flush and elevator data.
1035 */
1036 if (bio->bi_opf & (REQ_PREFLUSH | REQ_FUA))
1037 return false;
1038
1039 return true;
1040}
1041
1042/**
1043 * rq_ioc - determine io_context for request allocation
1044 * @bio: request being allocated is for this bio (can be %NULL)
1045 *
1046 * Determine io_context to use for request allocation for @bio. May return
1047 * %NULL if %current->io_context doesn't exist.
1048 */
1049static struct io_context *rq_ioc(struct bio *bio)
1050{
1051#ifdef CONFIG_BLK_CGROUP
1052 if (bio && bio->bi_ioc)
1053 return bio->bi_ioc;
1054#endif
1055 return current->io_context;
1056}
1057
1058/**
1059 * __get_request - get a free request
1060 * @rl: request list to allocate from
1061 * @op: operation and flags
1062 * @bio: bio to allocate request for (can be %NULL)
1063 * @gfp_mask: allocation mask
1064 *
1065 * Get a free request from @q. This function may fail under memory
1066 * pressure or if @q is dead.
1067 *
1068 * Must be called with @q->queue_lock held and,
1069 * Returns ERR_PTR on failure, with @q->queue_lock held.
1070 * Returns request pointer on success, with @q->queue_lock *not held*.
1071 */
1072static struct request *__get_request(struct request_list *rl, unsigned int op,
1073 struct bio *bio, gfp_t gfp_mask)
1074{
1075 struct request_queue *q = rl->q;
1076 struct request *rq;
1077 struct elevator_type *et = q->elevator->type;
1078 struct io_context *ioc = rq_ioc(bio);
1079 struct io_cq *icq = NULL;
1080 const bool is_sync = op_is_sync(op);
1081 int may_queue;
1082 req_flags_t rq_flags = RQF_ALLOCED;
1083
1084 if (unlikely(blk_queue_dying(q)))
1085 return ERR_PTR(-ENODEV);
1086
1087 may_queue = elv_may_queue(q, op);
1088 if (may_queue == ELV_MQUEUE_NO)
1089 goto rq_starved;
1090
1091 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
1092 if (rl->count[is_sync]+1 >= q->nr_requests) {
1093 /*
1094 * The queue will fill after this allocation, so set
1095 * it as full, and mark this process as "batching".
1096 * This process will be allowed to complete a batch of
1097 * requests, others will be blocked.
1098 */
1099 if (!blk_rl_full(rl, is_sync)) {
1100 ioc_set_batching(q, ioc);
1101 blk_set_rl_full(rl, is_sync);
1102 } else {
1103 if (may_queue != ELV_MQUEUE_MUST
1104 && !ioc_batching(q, ioc)) {
1105 /*
1106 * The queue is full and the allocating
1107 * process is not a "batcher", and not
1108 * exempted by the IO scheduler
1109 */
1110 return ERR_PTR(-ENOMEM);
1111 }
1112 }
1113 }
1114 blk_set_congested(rl, is_sync);
1115 }
1116
1117 /*
1118 * Only allow batching queuers to allocate up to 50% over the defined
1119 * limit of requests, otherwise we could have thousands of requests
1120 * allocated with any setting of ->nr_requests
1121 */
1122 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
1123 return ERR_PTR(-ENOMEM);
1124
1125 q->nr_rqs[is_sync]++;
1126 rl->count[is_sync]++;
1127 rl->starved[is_sync] = 0;
1128
1129 /*
1130 * Decide whether the new request will be managed by elevator. If
1131 * so, mark @rq_flags and increment elvpriv. Non-zero elvpriv will
1132 * prevent the current elevator from being destroyed until the new
1133 * request is freed. This guarantees icq's won't be destroyed and
1134 * makes creating new ones safe.
1135 *
1136 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1137 * it will be created after releasing queue_lock.
1138 */
1139 if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
1140 rq_flags |= RQF_ELVPRIV;
1141 q->nr_rqs_elvpriv++;
1142 if (et->icq_cache && ioc)
1143 icq = ioc_lookup_icq(ioc, q);
1144 }
1145
1146 if (blk_queue_io_stat(q))
1147 rq_flags |= RQF_IO_STAT;
1148 spin_unlock_irq(q->queue_lock);
1149
1150 /* allocate and init request */
1151 rq = mempool_alloc(rl->rq_pool, gfp_mask);
1152 if (!rq)
1153 goto fail_alloc;
1154
1155 blk_rq_init(q, rq);
1156 blk_rq_set_rl(rq, rl);
1157 blk_rq_set_prio(rq, ioc);
1158 rq->cmd_flags = op;
1159 rq->rq_flags = rq_flags;
1160
1161 /* init elvpriv */
1162 if (rq_flags & RQF_ELVPRIV) {
1163 if (unlikely(et->icq_cache && !icq)) {
1164 if (ioc)
1165 icq = ioc_create_icq(ioc, q, gfp_mask);
1166 if (!icq)
1167 goto fail_elvpriv;
1168 }
1169
1170 rq->elv.icq = icq;
1171 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1172 goto fail_elvpriv;
1173
1174 /* @rq->elv.icq holds io_context until @rq is freed */
1175 if (icq)
1176 get_io_context(icq->ioc);
1177 }
1178out:
1179 /*
1180 * ioc may be NULL here, and ioc_batching will be false. That's
1181 * OK, if the queue is under the request limit then requests need
1182 * not count toward the nr_batch_requests limit. There will always
1183 * be some limit enforced by BLK_BATCH_TIME.
1184 */
1185 if (ioc_batching(q, ioc))
1186 ioc->nr_batch_requests--;
1187
1188 trace_block_getrq(q, bio, op);
1189 return rq;
1190
1191fail_elvpriv:
1192 /*
1193 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1194 * and may fail indefinitely under memory pressure and thus
1195 * shouldn't stall IO. Treat this request as !elvpriv. This will
1196 * disturb iosched and blkcg but weird is bettern than dead.
1197 */
1198 printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1199 __func__, dev_name(q->backing_dev_info.dev));
1200
1201 rq->rq_flags &= ~RQF_ELVPRIV;
1202 rq->elv.icq = NULL;
1203
1204 spin_lock_irq(q->queue_lock);
1205 q->nr_rqs_elvpriv--;
1206 spin_unlock_irq(q->queue_lock);
1207 goto out;
1208
1209fail_alloc:
1210 /*
1211 * Allocation failed presumably due to memory. Undo anything we
1212 * might have messed up.
1213 *
1214 * Allocating task should really be put onto the front of the wait
1215 * queue, but this is pretty rare.
1216 */
1217 spin_lock_irq(q->queue_lock);
1218 freed_request(rl, is_sync, rq_flags);
1219
1220 /*
1221 * in the very unlikely event that allocation failed and no
1222 * requests for this direction was pending, mark us starved so that
1223 * freeing of a request in the other direction will notice
1224 * us. another possible fix would be to split the rq mempool into
1225 * READ and WRITE
1226 */
1227rq_starved:
1228 if (unlikely(rl->count[is_sync] == 0))
1229 rl->starved[is_sync] = 1;
1230 return ERR_PTR(-ENOMEM);
1231}
1232
1233/**
1234 * get_request - get a free request
1235 * @q: request_queue to allocate request from
1236 * @op: operation and flags
1237 * @bio: bio to allocate request for (can be %NULL)
1238 * @gfp_mask: allocation mask
1239 *
1240 * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask,
1241 * this function keeps retrying under memory pressure and fails iff @q is dead.
1242 *
1243 * Must be called with @q->queue_lock held and,
1244 * Returns ERR_PTR on failure, with @q->queue_lock held.
1245 * Returns request pointer on success, with @q->queue_lock *not held*.
1246 */
1247static struct request *get_request(struct request_queue *q, unsigned int op,
1248 struct bio *bio, gfp_t gfp_mask)
1249{
1250 const bool is_sync = op_is_sync(op);
1251 DEFINE_WAIT(wait);
1252 struct request_list *rl;
1253 struct request *rq;
1254
1255 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
1256retry:
1257 rq = __get_request(rl, op, bio, gfp_mask);
1258 if (!IS_ERR(rq))
1259 return rq;
1260
1261 if (!gfpflags_allow_blocking(gfp_mask) || unlikely(blk_queue_dying(q))) {
1262 blk_put_rl(rl);
1263 return rq;
1264 }
1265
1266 /* wait on @rl and retry */
1267 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1268 TASK_UNINTERRUPTIBLE);
1269
1270 trace_block_sleeprq(q, bio, op);
1271
1272 spin_unlock_irq(q->queue_lock);
1273 io_schedule();
1274
1275 /*
1276 * After sleeping, we become a "batching" process and will be able
1277 * to allocate at least one request, and up to a big batch of them
1278 * for a small period time. See ioc_batching, ioc_set_batching
1279 */
1280 ioc_set_batching(q, current->io_context);
1281
1282 spin_lock_irq(q->queue_lock);
1283 finish_wait(&rl->wait[is_sync], &wait);
1284
1285 goto retry;
1286}
1287
1288static struct request *blk_old_get_request(struct request_queue *q, int rw,
1289 gfp_t gfp_mask)
1290{
1291 struct request *rq;
1292
1293 BUG_ON(rw != READ && rw != WRITE);
1294
1295 /* create ioc upfront */
1296 create_io_context(gfp_mask, q->node);
1297
1298 spin_lock_irq(q->queue_lock);
1299 rq = get_request(q, rw, NULL, gfp_mask);
1300 if (IS_ERR(rq)) {
1301 spin_unlock_irq(q->queue_lock);
1302 return rq;
1303 }
1304
1305 /* q->queue_lock is unlocked at this point */
1306 rq->__data_len = 0;
1307 rq->__sector = (sector_t) -1;
1308 rq->bio = rq->biotail = NULL;
1309 return rq;
1310}
1311
1312struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1313{
1314 if (q->mq_ops)
1315 return blk_mq_alloc_request(q, rw,
1316 (gfp_mask & __GFP_DIRECT_RECLAIM) ?
1317 0 : BLK_MQ_REQ_NOWAIT);
1318 else
1319 return blk_old_get_request(q, rw, gfp_mask);
1320}
1321EXPORT_SYMBOL(blk_get_request);
1322
1323/**
1324 * blk_rq_set_block_pc - initialize a request to type BLOCK_PC
1325 * @rq: request to be initialized
1326 *
1327 */
1328void blk_rq_set_block_pc(struct request *rq)
1329{
1330 rq->cmd_type = REQ_TYPE_BLOCK_PC;
1331 memset(rq->__cmd, 0, sizeof(rq->__cmd));
1332}
1333EXPORT_SYMBOL(blk_rq_set_block_pc);
1334
1335/**
1336 * blk_requeue_request - put a request back on queue
1337 * @q: request queue where request should be inserted
1338 * @rq: request to be inserted
1339 *
1340 * Description:
1341 * Drivers often keep queueing requests until the hardware cannot accept
1342 * more, when that condition happens we need to put the request back
1343 * on the queue. Must be called with queue lock held.
1344 */
1345void blk_requeue_request(struct request_queue *q, struct request *rq)
1346{
1347 blk_delete_timer(rq);
1348 blk_clear_rq_complete(rq);
1349 trace_block_rq_requeue(q, rq);
1350 wbt_requeue(q->rq_wb, &rq->issue_stat);
1351
1352 if (rq->rq_flags & RQF_QUEUED)
1353 blk_queue_end_tag(q, rq);
1354
1355 BUG_ON(blk_queued_rq(rq));
1356
1357 elv_requeue_request(q, rq);
1358}
1359EXPORT_SYMBOL(blk_requeue_request);
1360
1361static void add_acct_request(struct request_queue *q, struct request *rq,
1362 int where)
1363{
1364 blk_account_io_start(rq, true);
1365 __elv_add_request(q, rq, where);
1366}
1367
1368static void part_round_stats_single(int cpu, struct hd_struct *part,
1369 unsigned long now)
1370{
1371 int inflight;
1372
1373 if (now == part->stamp)
1374 return;
1375
1376 inflight = part_in_flight(part);
1377 if (inflight) {
1378 __part_stat_add(cpu, part, time_in_queue,
1379 inflight * (now - part->stamp));
1380 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1381 }
1382 part->stamp = now;
1383}
1384
1385/**
1386 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1387 * @cpu: cpu number for stats access
1388 * @part: target partition
1389 *
1390 * The average IO queue length and utilisation statistics are maintained
1391 * by observing the current state of the queue length and the amount of
1392 * time it has been in this state for.
1393 *
1394 * Normally, that accounting is done on IO completion, but that can result
1395 * in more than a second's worth of IO being accounted for within any one
1396 * second, leading to >100% utilisation. To deal with that, we call this
1397 * function to do a round-off before returning the results when reading
1398 * /proc/diskstats. This accounts immediately for all queue usage up to
1399 * the current jiffies and restarts the counters again.
1400 */
1401void part_round_stats(int cpu, struct hd_struct *part)
1402{
1403 unsigned long now = jiffies;
1404
1405 if (part->partno)
1406 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1407 part_round_stats_single(cpu, part, now);
1408}
1409EXPORT_SYMBOL_GPL(part_round_stats);
1410
1411#ifdef CONFIG_PM
1412static void blk_pm_put_request(struct request *rq)
1413{
1414 if (rq->q->dev && !(rq->rq_flags & RQF_PM) && !--rq->q->nr_pending)
1415 pm_runtime_mark_last_busy(rq->q->dev);
1416}
1417#else
1418static inline void blk_pm_put_request(struct request *rq) {}
1419#endif
1420
1421/*
1422 * queue lock must be held
1423 */
1424void __blk_put_request(struct request_queue *q, struct request *req)
1425{
1426 req_flags_t rq_flags = req->rq_flags;
1427
1428 if (unlikely(!q))
1429 return;
1430
1431 if (q->mq_ops) {
1432 blk_mq_free_request(req);
1433 return;
1434 }
1435
1436 blk_pm_put_request(req);
1437
1438 elv_completed_request(q, req);
1439
1440 /* this is a bio leak */
1441 WARN_ON(req->bio != NULL);
1442
1443 wbt_done(q->rq_wb, &req->issue_stat);
1444
1445 /*
1446 * Request may not have originated from ll_rw_blk. if not,
1447 * it didn't come out of our reserved rq pools
1448 */
1449 if (rq_flags & RQF_ALLOCED) {
1450 struct request_list *rl = blk_rq_rl(req);
1451 bool sync = op_is_sync(req->cmd_flags);
1452
1453 BUG_ON(!list_empty(&req->queuelist));
1454 BUG_ON(ELV_ON_HASH(req));
1455
1456 blk_free_request(rl, req);
1457 freed_request(rl, sync, rq_flags);
1458 blk_put_rl(rl);
1459 }
1460}
1461EXPORT_SYMBOL_GPL(__blk_put_request);
1462
1463void blk_put_request(struct request *req)
1464{
1465 struct request_queue *q = req->q;
1466
1467 if (q->mq_ops)
1468 blk_mq_free_request(req);
1469 else {
1470 unsigned long flags;
1471
1472 spin_lock_irqsave(q->queue_lock, flags);
1473 __blk_put_request(q, req);
1474 spin_unlock_irqrestore(q->queue_lock, flags);
1475 }
1476}
1477EXPORT_SYMBOL(blk_put_request);
1478
1479bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1480 struct bio *bio)
1481{
1482 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
1483
1484 if (!ll_back_merge_fn(q, req, bio))
1485 return false;
1486
1487 trace_block_bio_backmerge(q, req, bio);
1488
1489 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1490 blk_rq_set_mixed_merge(req);
1491
1492 req->biotail->bi_next = bio;
1493 req->biotail = bio;
1494 req->__data_len += bio->bi_iter.bi_size;
1495 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1496
1497 blk_account_io_start(req, false);
1498 return true;
1499}
1500
1501bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1502 struct bio *bio)
1503{
1504 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
1505
1506 if (!ll_front_merge_fn(q, req, bio))
1507 return false;
1508
1509 trace_block_bio_frontmerge(q, req, bio);
1510
1511 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1512 blk_rq_set_mixed_merge(req);
1513
1514 bio->bi_next = req->bio;
1515 req->bio = bio;
1516
1517 req->__sector = bio->bi_iter.bi_sector;
1518 req->__data_len += bio->bi_iter.bi_size;
1519 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1520
1521 blk_account_io_start(req, false);
1522 return true;
1523}
1524
1525/**
1526 * blk_attempt_plug_merge - try to merge with %current's plugged list
1527 * @q: request_queue new bio is being queued at
1528 * @bio: new bio being queued
1529 * @request_count: out parameter for number of traversed plugged requests
1530 * @same_queue_rq: pointer to &struct request that gets filled in when
1531 * another request associated with @q is found on the plug list
1532 * (optional, may be %NULL)
1533 *
1534 * Determine whether @bio being queued on @q can be merged with a request
1535 * on %current's plugged list. Returns %true if merge was successful,
1536 * otherwise %false.
1537 *
1538 * Plugging coalesces IOs from the same issuer for the same purpose without
1539 * going through @q->queue_lock. As such it's more of an issuing mechanism
1540 * than scheduling, and the request, while may have elvpriv data, is not
1541 * added on the elevator at this point. In addition, we don't have
1542 * reliable access to the elevator outside queue lock. Only check basic
1543 * merging parameters without querying the elevator.
1544 *
1545 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1546 */
1547bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1548 unsigned int *request_count,
1549 struct request **same_queue_rq)
1550{
1551 struct blk_plug *plug;
1552 struct request *rq;
1553 bool ret = false;
1554 struct list_head *plug_list;
1555
1556 plug = current->plug;
1557 if (!plug)
1558 goto out;
1559 *request_count = 0;
1560
1561 if (q->mq_ops)
1562 plug_list = &plug->mq_list;
1563 else
1564 plug_list = &plug->list;
1565
1566 list_for_each_entry_reverse(rq, plug_list, queuelist) {
1567 int el_ret;
1568
1569 if (rq->q == q) {
1570 (*request_count)++;
1571 /*
1572 * Only blk-mq multiple hardware queues case checks the
1573 * rq in the same queue, there should be only one such
1574 * rq in a queue
1575 **/
1576 if (same_queue_rq)
1577 *same_queue_rq = rq;
1578 }
1579
1580 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
1581 continue;
1582
1583 el_ret = blk_try_merge(rq, bio);
1584 if (el_ret == ELEVATOR_BACK_MERGE) {
1585 ret = bio_attempt_back_merge(q, rq, bio);
1586 if (ret)
1587 break;
1588 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1589 ret = bio_attempt_front_merge(q, rq, bio);
1590 if (ret)
1591 break;
1592 }
1593 }
1594out:
1595 return ret;
1596}
1597
1598unsigned int blk_plug_queued_count(struct request_queue *q)
1599{
1600 struct blk_plug *plug;
1601 struct request *rq;
1602 struct list_head *plug_list;
1603 unsigned int ret = 0;
1604
1605 plug = current->plug;
1606 if (!plug)
1607 goto out;
1608
1609 if (q->mq_ops)
1610 plug_list = &plug->mq_list;
1611 else
1612 plug_list = &plug->list;
1613
1614 list_for_each_entry(rq, plug_list, queuelist) {
1615 if (rq->q == q)
1616 ret++;
1617 }
1618out:
1619 return ret;
1620}
1621
1622void init_request_from_bio(struct request *req, struct bio *bio)
1623{
1624 req->cmd_type = REQ_TYPE_FS;
1625 if (bio->bi_opf & REQ_RAHEAD)
1626 req->cmd_flags |= REQ_FAILFAST_MASK;
1627
1628 req->errors = 0;
1629 req->__sector = bio->bi_iter.bi_sector;
1630 if (ioprio_valid(bio_prio(bio)))
1631 req->ioprio = bio_prio(bio);
1632 blk_rq_bio_prep(req->q, req, bio);
1633}
1634
1635static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio)
1636{
1637 struct blk_plug *plug;
1638 int el_ret, where = ELEVATOR_INSERT_SORT;
1639 struct request *req;
1640 unsigned int request_count = 0;
1641 unsigned int wb_acct;
1642
1643 /*
1644 * low level driver can indicate that it wants pages above a
1645 * certain limit bounced to low memory (ie for highmem, or even
1646 * ISA dma in theory)
1647 */
1648 blk_queue_bounce(q, &bio);
1649
1650 blk_queue_split(q, &bio, q->bio_split);
1651
1652 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
1653 bio->bi_error = -EIO;
1654 bio_endio(bio);
1655 return BLK_QC_T_NONE;
1656 }
1657
1658 if (bio->bi_opf & (REQ_PREFLUSH | REQ_FUA)) {
1659 spin_lock_irq(q->queue_lock);
1660 where = ELEVATOR_INSERT_FLUSH;
1661 goto get_rq;
1662 }
1663
1664 /*
1665 * Check if we can merge with the plugged list before grabbing
1666 * any locks.
1667 */
1668 if (!blk_queue_nomerges(q)) {
1669 if (blk_attempt_plug_merge(q, bio, &request_count, NULL))
1670 return BLK_QC_T_NONE;
1671 } else
1672 request_count = blk_plug_queued_count(q);
1673
1674 spin_lock_irq(q->queue_lock);
1675
1676 el_ret = elv_merge(q, &req, bio);
1677 if (el_ret == ELEVATOR_BACK_MERGE) {
1678 if (bio_attempt_back_merge(q, req, bio)) {
1679 elv_bio_merged(q, req, bio);
1680 if (!attempt_back_merge(q, req))
1681 elv_merged_request(q, req, el_ret);
1682 goto out_unlock;
1683 }
1684 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1685 if (bio_attempt_front_merge(q, req, bio)) {
1686 elv_bio_merged(q, req, bio);
1687 if (!attempt_front_merge(q, req))
1688 elv_merged_request(q, req, el_ret);
1689 goto out_unlock;
1690 }
1691 }
1692
1693get_rq:
1694 wb_acct = wbt_wait(q->rq_wb, bio, q->queue_lock);
1695
1696 /*
1697 * Grab a free request. This is might sleep but can not fail.
1698 * Returns with the queue unlocked.
1699 */
1700 req = get_request(q, bio->bi_opf, bio, GFP_NOIO);
1701 if (IS_ERR(req)) {
1702 __wbt_done(q->rq_wb, wb_acct);
1703 bio->bi_error = PTR_ERR(req);
1704 bio_endio(bio);
1705 goto out_unlock;
1706 }
1707
1708 wbt_track(&req->issue_stat, wb_acct);
1709
1710 /*
1711 * After dropping the lock and possibly sleeping here, our request
1712 * may now be mergeable after it had proven unmergeable (above).
1713 * We don't worry about that case for efficiency. It won't happen
1714 * often, and the elevators are able to handle it.
1715 */
1716 init_request_from_bio(req, bio);
1717
1718 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
1719 req->cpu = raw_smp_processor_id();
1720
1721 plug = current->plug;
1722 if (plug) {
1723 /*
1724 * If this is the first request added after a plug, fire
1725 * of a plug trace.
1726 *
1727 * @request_count may become stale because of schedule
1728 * out, so check plug list again.
1729 */
1730 if (!request_count || list_empty(&plug->list))
1731 trace_block_plug(q);
1732 else {
1733 struct request *last = list_entry_rq(plug->list.prev);
1734 if (request_count >= BLK_MAX_REQUEST_COUNT ||
1735 blk_rq_bytes(last) >= BLK_PLUG_FLUSH_SIZE) {
1736 blk_flush_plug_list(plug, false);
1737 trace_block_plug(q);
1738 }
1739 }
1740 list_add_tail(&req->queuelist, &plug->list);
1741 blk_account_io_start(req, true);
1742 } else {
1743 spin_lock_irq(q->queue_lock);
1744 add_acct_request(q, req, where);
1745 __blk_run_queue(q);
1746out_unlock:
1747 spin_unlock_irq(q->queue_lock);
1748 }
1749
1750 return BLK_QC_T_NONE;
1751}
1752
1753/*
1754 * If bio->bi_dev is a partition, remap the location
1755 */
1756static inline void blk_partition_remap(struct bio *bio)
1757{
1758 struct block_device *bdev = bio->bi_bdev;
1759
1760 /*
1761 * Zone reset does not include bi_size so bio_sectors() is always 0.
1762 * Include a test for the reset op code and perform the remap if needed.
1763 */
1764 if (bdev != bdev->bd_contains &&
1765 (bio_sectors(bio) || bio_op(bio) == REQ_OP_ZONE_RESET)) {
1766 struct hd_struct *p = bdev->bd_part;
1767
1768 bio->bi_iter.bi_sector += p->start_sect;
1769 bio->bi_bdev = bdev->bd_contains;
1770
1771 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1772 bdev->bd_dev,
1773 bio->bi_iter.bi_sector - p->start_sect);
1774 }
1775}
1776
1777static void handle_bad_sector(struct bio *bio)
1778{
1779 char b[BDEVNAME_SIZE];
1780
1781 printk(KERN_INFO "attempt to access beyond end of device\n");
1782 printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n",
1783 bdevname(bio->bi_bdev, b),
1784 bio->bi_opf,
1785 (unsigned long long)bio_end_sector(bio),
1786 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1787}
1788
1789#ifdef CONFIG_FAIL_MAKE_REQUEST
1790
1791static DECLARE_FAULT_ATTR(fail_make_request);
1792
1793static int __init setup_fail_make_request(char *str)
1794{
1795 return setup_fault_attr(&fail_make_request, str);
1796}
1797__setup("fail_make_request=", setup_fail_make_request);
1798
1799static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1800{
1801 return part->make_it_fail && should_fail(&fail_make_request, bytes);
1802}
1803
1804static int __init fail_make_request_debugfs(void)
1805{
1806 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1807 NULL, &fail_make_request);
1808
1809 return PTR_ERR_OR_ZERO(dir);
1810}
1811
1812late_initcall(fail_make_request_debugfs);
1813
1814#else /* CONFIG_FAIL_MAKE_REQUEST */
1815
1816static inline bool should_fail_request(struct hd_struct *part,
1817 unsigned int bytes)
1818{
1819 return false;
1820}
1821
1822#endif /* CONFIG_FAIL_MAKE_REQUEST */
1823
1824/*
1825 * Check whether this bio extends beyond the end of the device.
1826 */
1827static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1828{
1829 sector_t maxsector;
1830
1831 if (!nr_sectors)
1832 return 0;
1833
1834 /* Test device or partition size, when known. */
1835 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1836 if (maxsector) {
1837 sector_t sector = bio->bi_iter.bi_sector;
1838
1839 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1840 /*
1841 * This may well happen - the kernel calls bread()
1842 * without checking the size of the device, e.g., when
1843 * mounting a device.
1844 */
1845 handle_bad_sector(bio);
1846 return 1;
1847 }
1848 }
1849
1850 return 0;
1851}
1852
1853static noinline_for_stack bool
1854generic_make_request_checks(struct bio *bio)
1855{
1856 struct request_queue *q;
1857 int nr_sectors = bio_sectors(bio);
1858 int err = -EIO;
1859 char b[BDEVNAME_SIZE];
1860 struct hd_struct *part;
1861
1862 might_sleep();
1863
1864 if (bio_check_eod(bio, nr_sectors))
1865 goto end_io;
1866
1867 q = bdev_get_queue(bio->bi_bdev);
1868 if (unlikely(!q)) {
1869 printk(KERN_ERR
1870 "generic_make_request: Trying to access "
1871 "nonexistent block-device %s (%Lu)\n",
1872 bdevname(bio->bi_bdev, b),
1873 (long long) bio->bi_iter.bi_sector);
1874 goto end_io;
1875 }
1876
1877 part = bio->bi_bdev->bd_part;
1878 if (should_fail_request(part, bio->bi_iter.bi_size) ||
1879 should_fail_request(&part_to_disk(part)->part0,
1880 bio->bi_iter.bi_size))
1881 goto end_io;
1882
1883 /*
1884 * If this device has partitions, remap block n
1885 * of partition p to block n+start(p) of the disk.
1886 */
1887 blk_partition_remap(bio);
1888
1889 if (bio_check_eod(bio, nr_sectors))
1890 goto end_io;
1891
1892 /*
1893 * Filter flush bio's early so that make_request based
1894 * drivers without flush support don't have to worry
1895 * about them.
1896 */
1897 if ((bio->bi_opf & (REQ_PREFLUSH | REQ_FUA)) &&
1898 !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
1899 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
1900 if (!nr_sectors) {
1901 err = 0;
1902 goto end_io;
1903 }
1904 }
1905
1906 switch (bio_op(bio)) {
1907 case REQ_OP_DISCARD:
1908 if (!blk_queue_discard(q))
1909 goto not_supported;
1910 break;
1911 case REQ_OP_SECURE_ERASE:
1912 if (!blk_queue_secure_erase(q))
1913 goto not_supported;
1914 break;
1915 case REQ_OP_WRITE_SAME:
1916 if (!bdev_write_same(bio->bi_bdev))
1917 goto not_supported;
1918 break;
1919 case REQ_OP_ZONE_REPORT:
1920 case REQ_OP_ZONE_RESET:
1921 if (!bdev_is_zoned(bio->bi_bdev))
1922 goto not_supported;
1923 break;
1924 case REQ_OP_WRITE_ZEROES:
1925 if (!bdev_write_zeroes_sectors(bio->bi_bdev))
1926 goto not_supported;
1927 break;
1928 default:
1929 break;
1930 }
1931
1932 /*
1933 * Various block parts want %current->io_context and lazy ioc
1934 * allocation ends up trading a lot of pain for a small amount of
1935 * memory. Just allocate it upfront. This may fail and block
1936 * layer knows how to live with it.
1937 */
1938 create_io_context(GFP_ATOMIC, q->node);
1939
1940 if (!blkcg_bio_issue_check(q, bio))
1941 return false;
1942
1943 trace_block_bio_queue(q, bio);
1944 return true;
1945
1946not_supported:
1947 err = -EOPNOTSUPP;
1948end_io:
1949 bio->bi_error = err;
1950 bio_endio(bio);
1951 return false;
1952}
1953
1954/**
1955 * generic_make_request - hand a buffer to its device driver for I/O
1956 * @bio: The bio describing the location in memory and on the device.
1957 *
1958 * generic_make_request() is used to make I/O requests of block
1959 * devices. It is passed a &struct bio, which describes the I/O that needs
1960 * to be done.
1961 *
1962 * generic_make_request() does not return any status. The
1963 * success/failure status of the request, along with notification of
1964 * completion, is delivered asynchronously through the bio->bi_end_io
1965 * function described (one day) else where.
1966 *
1967 * The caller of generic_make_request must make sure that bi_io_vec
1968 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1969 * set to describe the device address, and the
1970 * bi_end_io and optionally bi_private are set to describe how
1971 * completion notification should be signaled.
1972 *
1973 * generic_make_request and the drivers it calls may use bi_next if this
1974 * bio happens to be merged with someone else, and may resubmit the bio to
1975 * a lower device by calling into generic_make_request recursively, which
1976 * means the bio should NOT be touched after the call to ->make_request_fn.
1977 */
1978blk_qc_t generic_make_request(struct bio *bio)
1979{
1980 /*
1981 * bio_list_on_stack[0] contains bios submitted by the current
1982 * make_request_fn.
1983 * bio_list_on_stack[1] contains bios that were submitted before
1984 * the current make_request_fn, but that haven't been processed
1985 * yet.
1986 */
1987 struct bio_list bio_list_on_stack[2];
1988 blk_qc_t ret = BLK_QC_T_NONE;
1989
1990 if (!generic_make_request_checks(bio))
1991 goto out;
1992
1993 /*
1994 * We only want one ->make_request_fn to be active at a time, else
1995 * stack usage with stacked devices could be a problem. So use
1996 * current->bio_list to keep a list of requests submited by a
1997 * make_request_fn function. current->bio_list is also used as a
1998 * flag to say if generic_make_request is currently active in this
1999 * task or not. If it is NULL, then no make_request is active. If
2000 * it is non-NULL, then a make_request is active, and new requests
2001 * should be added at the tail
2002 */
2003 if (current->bio_list) {
2004 bio_list_add(¤t->bio_list[0], bio);
2005 goto out;
2006 }
2007
2008 /* following loop may be a bit non-obvious, and so deserves some
2009 * explanation.
2010 * Before entering the loop, bio->bi_next is NULL (as all callers
2011 * ensure that) so we have a list with a single bio.
2012 * We pretend that we have just taken it off a longer list, so
2013 * we assign bio_list to a pointer to the bio_list_on_stack,
2014 * thus initialising the bio_list of new bios to be
2015 * added. ->make_request() may indeed add some more bios
2016 * through a recursive call to generic_make_request. If it
2017 * did, we find a non-NULL value in bio_list and re-enter the loop
2018 * from the top. In this case we really did just take the bio
2019 * of the top of the list (no pretending) and so remove it from
2020 * bio_list, and call into ->make_request() again.
2021 */
2022 BUG_ON(bio->bi_next);
2023 bio_list_init(&bio_list_on_stack[0]);
2024 current->bio_list = bio_list_on_stack;
2025 do {
2026 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
2027
2028 if (likely(blk_queue_enter(q, false) == 0)) {
2029 struct bio_list lower, same;
2030
2031 /* Create a fresh bio_list for all subordinate requests */
2032 bio_list_on_stack[1] = bio_list_on_stack[0];
2033 bio_list_init(&bio_list_on_stack[0]);
2034 ret = q->make_request_fn(q, bio);
2035
2036 blk_queue_exit(q);
2037
2038 /* sort new bios into those for a lower level
2039 * and those for the same level
2040 */
2041 bio_list_init(&lower);
2042 bio_list_init(&same);
2043 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
2044 if (q == bdev_get_queue(bio->bi_bdev))
2045 bio_list_add(&same, bio);
2046 else
2047 bio_list_add(&lower, bio);
2048 /* now assemble so we handle the lowest level first */
2049 bio_list_merge(&bio_list_on_stack[0], &lower);
2050 bio_list_merge(&bio_list_on_stack[0], &same);
2051 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
2052 } else {
2053 bio_io_error(bio);
2054 }
2055 bio = bio_list_pop(&bio_list_on_stack[0]);
2056 } while (bio);
2057 current->bio_list = NULL; /* deactivate */
2058
2059out:
2060 return ret;
2061}
2062EXPORT_SYMBOL(generic_make_request);
2063
2064/**
2065 * submit_bio - submit a bio to the block device layer for I/O
2066 * @bio: The &struct bio which describes the I/O
2067 *
2068 * submit_bio() is very similar in purpose to generic_make_request(), and
2069 * uses that function to do most of the work. Both are fairly rough
2070 * interfaces; @bio must be presetup and ready for I/O.
2071 *
2072 */
2073blk_qc_t submit_bio(struct bio *bio)
2074{
2075 /*
2076 * If it's a regular read/write or a barrier with data attached,
2077 * go through the normal accounting stuff before submission.
2078 */
2079 if (bio_has_data(bio)) {
2080 unsigned int count;
2081
2082 if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
2083 count = bdev_logical_block_size(bio->bi_bdev) >> 9;
2084 else
2085 count = bio_sectors(bio);
2086
2087 if (op_is_write(bio_op(bio))) {
2088 count_vm_events(PGPGOUT, count);
2089 } else {
2090 task_io_account_read(bio->bi_iter.bi_size);
2091 count_vm_events(PGPGIN, count);
2092 }
2093
2094 if (unlikely(block_dump)) {
2095 char b[BDEVNAME_SIZE];
2096 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
2097 current->comm, task_pid_nr(current),
2098 op_is_write(bio_op(bio)) ? "WRITE" : "READ",
2099 (unsigned long long)bio->bi_iter.bi_sector,
2100 bdevname(bio->bi_bdev, b),
2101 count);
2102 }
2103 }
2104
2105 return generic_make_request(bio);
2106}
2107EXPORT_SYMBOL(submit_bio);
2108
2109/**
2110 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2111 * for new the queue limits
2112 * @q: the queue
2113 * @rq: the request being checked
2114 *
2115 * Description:
2116 * @rq may have been made based on weaker limitations of upper-level queues
2117 * in request stacking drivers, and it may violate the limitation of @q.
2118 * Since the block layer and the underlying device driver trust @rq
2119 * after it is inserted to @q, it should be checked against @q before
2120 * the insertion using this generic function.
2121 *
2122 * Request stacking drivers like request-based dm may change the queue
2123 * limits when retrying requests on other queues. Those requests need
2124 * to be checked against the new queue limits again during dispatch.
2125 */
2126static int blk_cloned_rq_check_limits(struct request_queue *q,
2127 struct request *rq)
2128{
2129 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) {
2130 printk(KERN_ERR "%s: over max size limit.\n", __func__);
2131 return -EIO;
2132 }
2133
2134 /*
2135 * queue's settings related to segment counting like q->bounce_pfn
2136 * may differ from that of other stacking queues.
2137 * Recalculate it to check the request correctly on this queue's
2138 * limitation.
2139 */
2140 blk_recalc_rq_segments(rq);
2141 if (rq->nr_phys_segments > queue_max_segments(q)) {
2142 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2143 return -EIO;
2144 }
2145
2146 return 0;
2147}
2148
2149/**
2150 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2151 * @q: the queue to submit the request
2152 * @rq: the request being queued
2153 */
2154int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
2155{
2156 unsigned long flags;
2157 int where = ELEVATOR_INSERT_BACK;
2158
2159 if (blk_cloned_rq_check_limits(q, rq))
2160 return -EIO;
2161
2162 if (rq->rq_disk &&
2163 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
2164 return -EIO;
2165
2166 if (q->mq_ops) {
2167 if (blk_queue_io_stat(q))
2168 blk_account_io_start(rq, true);
2169 blk_mq_insert_request(rq, false, true, false);
2170 return 0;
2171 }
2172
2173 spin_lock_irqsave(q->queue_lock, flags);
2174 if (unlikely(blk_queue_dying(q))) {
2175 spin_unlock_irqrestore(q->queue_lock, flags);
2176 return -ENODEV;
2177 }
2178
2179 /*
2180 * Submitting request must be dequeued before calling this function
2181 * because it will be linked to another request_queue
2182 */
2183 BUG_ON(blk_queued_rq(rq));
2184
2185 if (rq->cmd_flags & (REQ_PREFLUSH | REQ_FUA))
2186 where = ELEVATOR_INSERT_FLUSH;
2187
2188 add_acct_request(q, rq, where);
2189 if (where == ELEVATOR_INSERT_FLUSH)
2190 __blk_run_queue(q);
2191 spin_unlock_irqrestore(q->queue_lock, flags);
2192
2193 return 0;
2194}
2195EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2196
2197/**
2198 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2199 * @rq: request to examine
2200 *
2201 * Description:
2202 * A request could be merge of IOs which require different failure
2203 * handling. This function determines the number of bytes which
2204 * can be failed from the beginning of the request without
2205 * crossing into area which need to be retried further.
2206 *
2207 * Return:
2208 * The number of bytes to fail.
2209 *
2210 * Context:
2211 * queue_lock must be held.
2212 */
2213unsigned int blk_rq_err_bytes(const struct request *rq)
2214{
2215 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2216 unsigned int bytes = 0;
2217 struct bio *bio;
2218
2219 if (!(rq->rq_flags & RQF_MIXED_MERGE))
2220 return blk_rq_bytes(rq);
2221
2222 /*
2223 * Currently the only 'mixing' which can happen is between
2224 * different fastfail types. We can safely fail portions
2225 * which have all the failfast bits that the first one has -
2226 * the ones which are at least as eager to fail as the first
2227 * one.
2228 */
2229 for (bio = rq->bio; bio; bio = bio->bi_next) {
2230 if ((bio->bi_opf & ff) != ff)
2231 break;
2232 bytes += bio->bi_iter.bi_size;
2233 }
2234
2235 /* this could lead to infinite loop */
2236 BUG_ON(blk_rq_bytes(rq) && !bytes);
2237 return bytes;
2238}
2239EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2240
2241void blk_account_io_completion(struct request *req, unsigned int bytes)
2242{
2243 if (blk_do_io_stat(req)) {
2244 const int rw = rq_data_dir(req);
2245 struct hd_struct *part;
2246 int cpu;
2247
2248 cpu = part_stat_lock();
2249 part = req->part;
2250 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2251 part_stat_unlock();
2252 }
2253}
2254
2255void blk_account_io_done(struct request *req)
2256{
2257 /*
2258 * Account IO completion. flush_rq isn't accounted as a
2259 * normal IO on queueing nor completion. Accounting the
2260 * containing request is enough.
2261 */
2262 if (blk_do_io_stat(req) && !(req->rq_flags & RQF_FLUSH_SEQ)) {
2263 unsigned long duration = jiffies - req->start_time;
2264 const int rw = rq_data_dir(req);
2265 struct hd_struct *part;
2266 int cpu;
2267
2268 cpu = part_stat_lock();
2269 part = req->part;
2270
2271 part_stat_inc(cpu, part, ios[rw]);
2272 part_stat_add(cpu, part, ticks[rw], duration);
2273 part_round_stats(cpu, part);
2274 part_dec_in_flight(part, rw);
2275
2276 hd_struct_put(part);
2277 part_stat_unlock();
2278 }
2279}
2280
2281#ifdef CONFIG_PM
2282/*
2283 * Don't process normal requests when queue is suspended
2284 * or in the process of suspending/resuming
2285 */
2286static struct request *blk_pm_peek_request(struct request_queue *q,
2287 struct request *rq)
2288{
2289 if (q->dev && (q->rpm_status == RPM_SUSPENDED ||
2290 (q->rpm_status != RPM_ACTIVE && !(rq->rq_flags & RQF_PM))))
2291 return NULL;
2292 else
2293 return rq;
2294}
2295#else
2296static inline struct request *blk_pm_peek_request(struct request_queue *q,
2297 struct request *rq)
2298{
2299 return rq;
2300}
2301#endif
2302
2303void blk_account_io_start(struct request *rq, bool new_io)
2304{
2305 struct hd_struct *part;
2306 int rw = rq_data_dir(rq);
2307 int cpu;
2308
2309 if (!blk_do_io_stat(rq))
2310 return;
2311
2312 cpu = part_stat_lock();
2313
2314 if (!new_io) {
2315 part = rq->part;
2316 part_stat_inc(cpu, part, merges[rw]);
2317 } else {
2318 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2319 if (!hd_struct_try_get(part)) {
2320 /*
2321 * The partition is already being removed,
2322 * the request will be accounted on the disk only
2323 *
2324 * We take a reference on disk->part0 although that
2325 * partition will never be deleted, so we can treat
2326 * it as any other partition.
2327 */
2328 part = &rq->rq_disk->part0;
2329 hd_struct_get(part);
2330 }
2331 part_round_stats(cpu, part);
2332 part_inc_in_flight(part, rw);
2333 rq->part = part;
2334 }
2335
2336 part_stat_unlock();
2337}
2338
2339/**
2340 * blk_peek_request - peek at the top of a request queue
2341 * @q: request queue to peek at
2342 *
2343 * Description:
2344 * Return the request at the top of @q. The returned request
2345 * should be started using blk_start_request() before LLD starts
2346 * processing it.
2347 *
2348 * Return:
2349 * Pointer to the request at the top of @q if available. Null
2350 * otherwise.
2351 *
2352 * Context:
2353 * queue_lock must be held.
2354 */
2355struct request *blk_peek_request(struct request_queue *q)
2356{
2357 struct request *rq;
2358 int ret;
2359
2360 while ((rq = __elv_next_request(q)) != NULL) {
2361
2362 rq = blk_pm_peek_request(q, rq);
2363 if (!rq)
2364 break;
2365
2366 if (!(rq->rq_flags & RQF_STARTED)) {
2367 /*
2368 * This is the first time the device driver
2369 * sees this request (possibly after
2370 * requeueing). Notify IO scheduler.
2371 */
2372 if (rq->rq_flags & RQF_SORTED)
2373 elv_activate_rq(q, rq);
2374
2375 /*
2376 * just mark as started even if we don't start
2377 * it, a request that has been delayed should
2378 * not be passed by new incoming requests
2379 */
2380 rq->rq_flags |= RQF_STARTED;
2381 trace_block_rq_issue(q, rq);
2382 }
2383
2384 if (!q->boundary_rq || q->boundary_rq == rq) {
2385 q->end_sector = rq_end_sector(rq);
2386 q->boundary_rq = NULL;
2387 }
2388
2389 if (rq->rq_flags & RQF_DONTPREP)
2390 break;
2391
2392 if (q->dma_drain_size && blk_rq_bytes(rq)) {
2393 /*
2394 * make sure space for the drain appears we
2395 * know we can do this because max_hw_segments
2396 * has been adjusted to be one fewer than the
2397 * device can handle
2398 */
2399 rq->nr_phys_segments++;
2400 }
2401
2402 if (!q->prep_rq_fn)
2403 break;
2404
2405 ret = q->prep_rq_fn(q, rq);
2406 if (ret == BLKPREP_OK) {
2407 break;
2408 } else if (ret == BLKPREP_DEFER) {
2409 /*
2410 * the request may have been (partially) prepped.
2411 * we need to keep this request in the front to
2412 * avoid resource deadlock. RQF_STARTED will
2413 * prevent other fs requests from passing this one.
2414 */
2415 if (q->dma_drain_size && blk_rq_bytes(rq) &&
2416 !(rq->rq_flags & RQF_DONTPREP)) {
2417 /*
2418 * remove the space for the drain we added
2419 * so that we don't add it again
2420 */
2421 --rq->nr_phys_segments;
2422 }
2423
2424 rq = NULL;
2425 break;
2426 } else if (ret == BLKPREP_KILL || ret == BLKPREP_INVALID) {
2427 int err = (ret == BLKPREP_INVALID) ? -EREMOTEIO : -EIO;
2428
2429 rq->rq_flags |= RQF_QUIET;
2430 /*
2431 * Mark this request as started so we don't trigger
2432 * any debug logic in the end I/O path.
2433 */
2434 blk_start_request(rq);
2435 __blk_end_request_all(rq, err);
2436 } else {
2437 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2438 break;
2439 }
2440 }
2441
2442 return rq;
2443}
2444EXPORT_SYMBOL(blk_peek_request);
2445
2446void blk_dequeue_request(struct request *rq)
2447{
2448 struct request_queue *q = rq->q;
2449
2450 BUG_ON(list_empty(&rq->queuelist));
2451 BUG_ON(ELV_ON_HASH(rq));
2452
2453 list_del_init(&rq->queuelist);
2454
2455 /*
2456 * the time frame between a request being removed from the lists
2457 * and to it is freed is accounted as io that is in progress at
2458 * the driver side.
2459 */
2460 if (blk_account_rq(rq)) {
2461 q->in_flight[rq_is_sync(rq)]++;
2462 set_io_start_time_ns(rq);
2463 }
2464}
2465
2466/**
2467 * blk_start_request - start request processing on the driver
2468 * @req: request to dequeue
2469 *
2470 * Description:
2471 * Dequeue @req and start timeout timer on it. This hands off the
2472 * request to the driver.
2473 *
2474 * Block internal functions which don't want to start timer should
2475 * call blk_dequeue_request().
2476 *
2477 * Context:
2478 * queue_lock must be held.
2479 */
2480void blk_start_request(struct request *req)
2481{
2482 blk_dequeue_request(req);
2483
2484 if (test_bit(QUEUE_FLAG_STATS, &req->q->queue_flags)) {
2485 blk_stat_set_issue_time(&req->issue_stat);
2486 req->rq_flags |= RQF_STATS;
2487 wbt_issue(req->q->rq_wb, &req->issue_stat);
2488 }
2489
2490 /*
2491 * We are now handing the request to the hardware, initialize
2492 * resid_len to full count and add the timeout handler.
2493 */
2494 req->resid_len = blk_rq_bytes(req);
2495 if (unlikely(blk_bidi_rq(req)))
2496 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2497
2498 BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags));
2499 blk_add_timer(req);
2500}
2501EXPORT_SYMBOL(blk_start_request);
2502
2503/**
2504 * blk_fetch_request - fetch a request from a request queue
2505 * @q: request queue to fetch a request from
2506 *
2507 * Description:
2508 * Return the request at the top of @q. The request is started on
2509 * return and LLD can start processing it immediately.
2510 *
2511 * Return:
2512 * Pointer to the request at the top of @q if available. Null
2513 * otherwise.
2514 *
2515 * Context:
2516 * queue_lock must be held.
2517 */
2518struct request *blk_fetch_request(struct request_queue *q)
2519{
2520 struct request *rq;
2521
2522 rq = blk_peek_request(q);
2523 if (rq)
2524 blk_start_request(rq);
2525 return rq;
2526}
2527EXPORT_SYMBOL(blk_fetch_request);
2528
2529/**
2530 * blk_update_request - Special helper function for request stacking drivers
2531 * @req: the request being processed
2532 * @error: %0 for success, < %0 for error
2533 * @nr_bytes: number of bytes to complete @req
2534 *
2535 * Description:
2536 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2537 * the request structure even if @req doesn't have leftover.
2538 * If @req has leftover, sets it up for the next range of segments.
2539 *
2540 * This special helper function is only for request stacking drivers
2541 * (e.g. request-based dm) so that they can handle partial completion.
2542 * Actual device drivers should use blk_end_request instead.
2543 *
2544 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2545 * %false return from this function.
2546 *
2547 * Return:
2548 * %false - this request doesn't have any more data
2549 * %true - this request has more data
2550 **/
2551bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2552{
2553 int total_bytes;
2554
2555 trace_block_rq_complete(req->q, req, nr_bytes);
2556
2557 if (!req->bio)
2558 return false;
2559
2560 /*
2561 * For fs requests, rq is just carrier of independent bio's
2562 * and each partial completion should be handled separately.
2563 * Reset per-request error on each partial completion.
2564 *
2565 * TODO: tj: This is too subtle. It would be better to let
2566 * low level drivers do what they see fit.
2567 */
2568 if (req->cmd_type == REQ_TYPE_FS)
2569 req->errors = 0;
2570
2571 if (error && req->cmd_type == REQ_TYPE_FS &&
2572 !(req->rq_flags & RQF_QUIET)) {
2573 char *error_type;
2574
2575 switch (error) {
2576 case -ENOLINK:
2577 error_type = "recoverable transport";
2578 break;
2579 case -EREMOTEIO:
2580 error_type = "critical target";
2581 break;
2582 case -EBADE:
2583 error_type = "critical nexus";
2584 break;
2585 case -ETIMEDOUT:
2586 error_type = "timeout";
2587 break;
2588 case -ENOSPC:
2589 error_type = "critical space allocation";
2590 break;
2591 case -ENODATA:
2592 error_type = "critical medium";
2593 break;
2594 case -EIO:
2595 default:
2596 error_type = "I/O";
2597 break;
2598 }
2599 printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n",
2600 __func__, error_type, req->rq_disk ?
2601 req->rq_disk->disk_name : "?",
2602 (unsigned long long)blk_rq_pos(req));
2603
2604 }
2605
2606 blk_account_io_completion(req, nr_bytes);
2607
2608 total_bytes = 0;
2609 while (req->bio) {
2610 struct bio *bio = req->bio;
2611 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
2612
2613 if (bio_bytes == bio->bi_iter.bi_size)
2614 req->bio = bio->bi_next;
2615
2616 req_bio_endio(req, bio, bio_bytes, error);
2617
2618 total_bytes += bio_bytes;
2619 nr_bytes -= bio_bytes;
2620
2621 if (!nr_bytes)
2622 break;
2623 }
2624
2625 /*
2626 * completely done
2627 */
2628 if (!req->bio) {
2629 /*
2630 * Reset counters so that the request stacking driver
2631 * can find how many bytes remain in the request
2632 * later.
2633 */
2634 req->__data_len = 0;
2635 return false;
2636 }
2637
2638 WARN_ON_ONCE(req->rq_flags & RQF_SPECIAL_PAYLOAD);
2639
2640 req->__data_len -= total_bytes;
2641
2642 /* update sector only for requests with clear definition of sector */
2643 if (req->cmd_type == REQ_TYPE_FS)
2644 req->__sector += total_bytes >> 9;
2645
2646 /* mixed attributes always follow the first bio */
2647 if (req->rq_flags & RQF_MIXED_MERGE) {
2648 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2649 req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
2650 }
2651
2652 /*
2653 * If total number of sectors is less than the first segment
2654 * size, something has gone terribly wrong.
2655 */
2656 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2657 blk_dump_rq_flags(req, "request botched");
2658 req->__data_len = blk_rq_cur_bytes(req);
2659 }
2660
2661 /* recalculate the number of segments */
2662 blk_recalc_rq_segments(req);
2663
2664 return true;
2665}
2666EXPORT_SYMBOL_GPL(blk_update_request);
2667
2668static bool blk_update_bidi_request(struct request *rq, int error,
2669 unsigned int nr_bytes,
2670 unsigned int bidi_bytes)
2671{
2672 if (blk_update_request(rq, error, nr_bytes))
2673 return true;
2674
2675 /* Bidi request must be completed as a whole */
2676 if (unlikely(blk_bidi_rq(rq)) &&
2677 blk_update_request(rq->next_rq, error, bidi_bytes))
2678 return true;
2679
2680 if (blk_queue_add_random(rq->q))
2681 add_disk_randomness(rq->rq_disk);
2682
2683 return false;
2684}
2685
2686/**
2687 * blk_unprep_request - unprepare a request
2688 * @req: the request
2689 *
2690 * This function makes a request ready for complete resubmission (or
2691 * completion). It happens only after all error handling is complete,
2692 * so represents the appropriate moment to deallocate any resources
2693 * that were allocated to the request in the prep_rq_fn. The queue
2694 * lock is held when calling this.
2695 */
2696void blk_unprep_request(struct request *req)
2697{
2698 struct request_queue *q = req->q;
2699
2700 req->rq_flags &= ~RQF_DONTPREP;
2701 if (q->unprep_rq_fn)
2702 q->unprep_rq_fn(q, req);
2703}
2704EXPORT_SYMBOL_GPL(blk_unprep_request);
2705
2706/*
2707 * queue lock must be held
2708 */
2709void blk_finish_request(struct request *req, int error)
2710{
2711 struct request_queue *q = req->q;
2712
2713 if (req->rq_flags & RQF_STATS)
2714 blk_stat_add(&q->rq_stats[rq_data_dir(req)], req);
2715
2716 if (req->rq_flags & RQF_QUEUED)
2717 blk_queue_end_tag(q, req);
2718
2719 BUG_ON(blk_queued_rq(req));
2720
2721 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2722 laptop_io_completion(&req->q->backing_dev_info);
2723
2724 blk_delete_timer(req);
2725
2726 if (req->rq_flags & RQF_DONTPREP)
2727 blk_unprep_request(req);
2728
2729 blk_account_io_done(req);
2730
2731 if (req->end_io) {
2732 wbt_done(req->q->rq_wb, &req->issue_stat);
2733 req->end_io(req, error);
2734 } else {
2735 if (blk_bidi_rq(req))
2736 __blk_put_request(req->next_rq->q, req->next_rq);
2737
2738 __blk_put_request(q, req);
2739 }
2740}
2741EXPORT_SYMBOL(blk_finish_request);
2742
2743/**
2744 * blk_end_bidi_request - Complete a bidi request
2745 * @rq: the request to complete
2746 * @error: %0 for success, < %0 for error
2747 * @nr_bytes: number of bytes to complete @rq
2748 * @bidi_bytes: number of bytes to complete @rq->next_rq
2749 *
2750 * Description:
2751 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2752 * Drivers that supports bidi can safely call this member for any
2753 * type of request, bidi or uni. In the later case @bidi_bytes is
2754 * just ignored.
2755 *
2756 * Return:
2757 * %false - we are done with this request
2758 * %true - still buffers pending for this request
2759 **/
2760static bool blk_end_bidi_request(struct request *rq, int error,
2761 unsigned int nr_bytes, unsigned int bidi_bytes)
2762{
2763 struct request_queue *q = rq->q;
2764 unsigned long flags;
2765
2766 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2767 return true;
2768
2769 spin_lock_irqsave(q->queue_lock, flags);
2770 blk_finish_request(rq, error);
2771 spin_unlock_irqrestore(q->queue_lock, flags);
2772
2773 return false;
2774}
2775
2776/**
2777 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2778 * @rq: the request to complete
2779 * @error: %0 for success, < %0 for error
2780 * @nr_bytes: number of bytes to complete @rq
2781 * @bidi_bytes: number of bytes to complete @rq->next_rq
2782 *
2783 * Description:
2784 * Identical to blk_end_bidi_request() except that queue lock is
2785 * assumed to be locked on entry and remains so on return.
2786 *
2787 * Return:
2788 * %false - we are done with this request
2789 * %true - still buffers pending for this request
2790 **/
2791bool __blk_end_bidi_request(struct request *rq, int error,
2792 unsigned int nr_bytes, unsigned int bidi_bytes)
2793{
2794 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2795 return true;
2796
2797 blk_finish_request(rq, error);
2798
2799 return false;
2800}
2801
2802/**
2803 * blk_end_request - Helper function for drivers to complete the request.
2804 * @rq: the request being processed
2805 * @error: %0 for success, < %0 for error
2806 * @nr_bytes: number of bytes to complete
2807 *
2808 * Description:
2809 * Ends I/O on a number of bytes attached to @rq.
2810 * If @rq has leftover, sets it up for the next range of segments.
2811 *
2812 * Return:
2813 * %false - we are done with this request
2814 * %true - still buffers pending for this request
2815 **/
2816bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2817{
2818 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2819}
2820EXPORT_SYMBOL(blk_end_request);
2821
2822/**
2823 * blk_end_request_all - Helper function for drives to finish the request.
2824 * @rq: the request to finish
2825 * @error: %0 for success, < %0 for error
2826 *
2827 * Description:
2828 * Completely finish @rq.
2829 */
2830void blk_end_request_all(struct request *rq, int error)
2831{
2832 bool pending;
2833 unsigned int bidi_bytes = 0;
2834
2835 if (unlikely(blk_bidi_rq(rq)))
2836 bidi_bytes = blk_rq_bytes(rq->next_rq);
2837
2838 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2839 BUG_ON(pending);
2840}
2841EXPORT_SYMBOL(blk_end_request_all);
2842
2843/**
2844 * blk_end_request_cur - Helper function to finish the current request chunk.
2845 * @rq: the request to finish the current chunk for
2846 * @error: %0 for success, < %0 for error
2847 *
2848 * Description:
2849 * Complete the current consecutively mapped chunk from @rq.
2850 *
2851 * Return:
2852 * %false - we are done with this request
2853 * %true - still buffers pending for this request
2854 */
2855bool blk_end_request_cur(struct request *rq, int error)
2856{
2857 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2858}
2859EXPORT_SYMBOL(blk_end_request_cur);
2860
2861/**
2862 * blk_end_request_err - Finish a request till the next failure boundary.
2863 * @rq: the request to finish till the next failure boundary for
2864 * @error: must be negative errno
2865 *
2866 * Description:
2867 * Complete @rq till the next failure boundary.
2868 *
2869 * Return:
2870 * %false - we are done with this request
2871 * %true - still buffers pending for this request
2872 */
2873bool blk_end_request_err(struct request *rq, int error)
2874{
2875 WARN_ON(error >= 0);
2876 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2877}
2878EXPORT_SYMBOL_GPL(blk_end_request_err);
2879
2880/**
2881 * __blk_end_request - Helper function for drivers to complete the request.
2882 * @rq: the request being processed
2883 * @error: %0 for success, < %0 for error
2884 * @nr_bytes: number of bytes to complete
2885 *
2886 * Description:
2887 * Must be called with queue lock held unlike blk_end_request().
2888 *
2889 * Return:
2890 * %false - we are done with this request
2891 * %true - still buffers pending for this request
2892 **/
2893bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2894{
2895 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2896}
2897EXPORT_SYMBOL(__blk_end_request);
2898
2899/**
2900 * __blk_end_request_all - Helper function for drives to finish the request.
2901 * @rq: the request to finish
2902 * @error: %0 for success, < %0 for error
2903 *
2904 * Description:
2905 * Completely finish @rq. Must be called with queue lock held.
2906 */
2907void __blk_end_request_all(struct request *rq, int error)
2908{
2909 bool pending;
2910 unsigned int bidi_bytes = 0;
2911
2912 if (unlikely(blk_bidi_rq(rq)))
2913 bidi_bytes = blk_rq_bytes(rq->next_rq);
2914
2915 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2916 BUG_ON(pending);
2917}
2918EXPORT_SYMBOL(__blk_end_request_all);
2919
2920/**
2921 * __blk_end_request_cur - Helper function to finish the current request chunk.
2922 * @rq: the request to finish the current chunk for
2923 * @error: %0 for success, < %0 for error
2924 *
2925 * Description:
2926 * Complete the current consecutively mapped chunk from @rq. Must
2927 * be called with queue lock held.
2928 *
2929 * Return:
2930 * %false - we are done with this request
2931 * %true - still buffers pending for this request
2932 */
2933bool __blk_end_request_cur(struct request *rq, int error)
2934{
2935 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2936}
2937EXPORT_SYMBOL(__blk_end_request_cur);
2938
2939/**
2940 * __blk_end_request_err - Finish a request till the next failure boundary.
2941 * @rq: the request to finish till the next failure boundary for
2942 * @error: must be negative errno
2943 *
2944 * Description:
2945 * Complete @rq till the next failure boundary. Must be called
2946 * with queue lock held.
2947 *
2948 * Return:
2949 * %false - we are done with this request
2950 * %true - still buffers pending for this request
2951 */
2952bool __blk_end_request_err(struct request *rq, int error)
2953{
2954 WARN_ON(error >= 0);
2955 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2956}
2957EXPORT_SYMBOL_GPL(__blk_end_request_err);
2958
2959void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2960 struct bio *bio)
2961{
2962 if (bio_has_data(bio))
2963 rq->nr_phys_segments = bio_phys_segments(q, bio);
2964
2965 rq->__data_len = bio->bi_iter.bi_size;
2966 rq->bio = rq->biotail = bio;
2967
2968 if (bio->bi_bdev)
2969 rq->rq_disk = bio->bi_bdev->bd_disk;
2970}
2971
2972#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2973/**
2974 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2975 * @rq: the request to be flushed
2976 *
2977 * Description:
2978 * Flush all pages in @rq.
2979 */
2980void rq_flush_dcache_pages(struct request *rq)
2981{
2982 struct req_iterator iter;
2983 struct bio_vec bvec;
2984
2985 rq_for_each_segment(bvec, rq, iter)
2986 flush_dcache_page(bvec.bv_page);
2987}
2988EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2989#endif
2990
2991/**
2992 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2993 * @q : the queue of the device being checked
2994 *
2995 * Description:
2996 * Check if underlying low-level drivers of a device are busy.
2997 * If the drivers want to export their busy state, they must set own
2998 * exporting function using blk_queue_lld_busy() first.
2999 *
3000 * Basically, this function is used only by request stacking drivers
3001 * to stop dispatching requests to underlying devices when underlying
3002 * devices are busy. This behavior helps more I/O merging on the queue
3003 * of the request stacking driver and prevents I/O throughput regression
3004 * on burst I/O load.
3005 *
3006 * Return:
3007 * 0 - Not busy (The request stacking driver should dispatch request)
3008 * 1 - Busy (The request stacking driver should stop dispatching request)
3009 */
3010int blk_lld_busy(struct request_queue *q)
3011{
3012 if (q->lld_busy_fn)
3013 return q->lld_busy_fn(q);
3014
3015 return 0;
3016}
3017EXPORT_SYMBOL_GPL(blk_lld_busy);
3018
3019/**
3020 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
3021 * @rq: the clone request to be cleaned up
3022 *
3023 * Description:
3024 * Free all bios in @rq for a cloned request.
3025 */
3026void blk_rq_unprep_clone(struct request *rq)
3027{
3028 struct bio *bio;
3029
3030 while ((bio = rq->bio) != NULL) {
3031 rq->bio = bio->bi_next;
3032
3033 bio_put(bio);
3034 }
3035}
3036EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
3037
3038/*
3039 * Copy attributes of the original request to the clone request.
3040 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3041 */
3042static void __blk_rq_prep_clone(struct request *dst, struct request *src)
3043{
3044 dst->cpu = src->cpu;
3045 dst->cmd_flags = src->cmd_flags | REQ_NOMERGE;
3046 dst->cmd_type = src->cmd_type;
3047 dst->__sector = blk_rq_pos(src);
3048 dst->__data_len = blk_rq_bytes(src);
3049 dst->nr_phys_segments = src->nr_phys_segments;
3050 dst->ioprio = src->ioprio;
3051 dst->extra_len = src->extra_len;
3052}
3053
3054/**
3055 * blk_rq_prep_clone - Helper function to setup clone request
3056 * @rq: the request to be setup
3057 * @rq_src: original request to be cloned
3058 * @bs: bio_set that bios for clone are allocated from
3059 * @gfp_mask: memory allocation mask for bio
3060 * @bio_ctr: setup function to be called for each clone bio.
3061 * Returns %0 for success, non %0 for failure.
3062 * @data: private data to be passed to @bio_ctr
3063 *
3064 * Description:
3065 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3066 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3067 * are not copied, and copying such parts is the caller's responsibility.
3068 * Also, pages which the original bios are pointing to are not copied
3069 * and the cloned bios just point same pages.
3070 * So cloned bios must be completed before original bios, which means
3071 * the caller must complete @rq before @rq_src.
3072 */
3073int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
3074 struct bio_set *bs, gfp_t gfp_mask,
3075 int (*bio_ctr)(struct bio *, struct bio *, void *),
3076 void *data)
3077{
3078 struct bio *bio, *bio_src;
3079
3080 if (!bs)
3081 bs = fs_bio_set;
3082
3083 __rq_for_each_bio(bio_src, rq_src) {
3084 bio = bio_clone_fast(bio_src, gfp_mask, bs);
3085 if (!bio)
3086 goto free_and_out;
3087
3088 if (bio_ctr && bio_ctr(bio, bio_src, data))
3089 goto free_and_out;
3090
3091 if (rq->bio) {
3092 rq->biotail->bi_next = bio;
3093 rq->biotail = bio;
3094 } else
3095 rq->bio = rq->biotail = bio;
3096 }
3097
3098 __blk_rq_prep_clone(rq, rq_src);
3099
3100 return 0;
3101
3102free_and_out:
3103 if (bio)
3104 bio_put(bio);
3105 blk_rq_unprep_clone(rq);
3106
3107 return -ENOMEM;
3108}
3109EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
3110
3111int kblockd_schedule_work(struct work_struct *work)
3112{
3113 return queue_work(kblockd_workqueue, work);
3114}
3115EXPORT_SYMBOL(kblockd_schedule_work);
3116
3117int kblockd_schedule_work_on(int cpu, struct work_struct *work)
3118{
3119 return queue_work_on(cpu, kblockd_workqueue, work);
3120}
3121EXPORT_SYMBOL(kblockd_schedule_work_on);
3122
3123int kblockd_schedule_delayed_work(struct delayed_work *dwork,
3124 unsigned long delay)
3125{
3126 return queue_delayed_work(kblockd_workqueue, dwork, delay);
3127}
3128EXPORT_SYMBOL(kblockd_schedule_delayed_work);
3129
3130int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork,
3131 unsigned long delay)
3132{
3133 return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3134}
3135EXPORT_SYMBOL(kblockd_schedule_delayed_work_on);
3136
3137/**
3138 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3139 * @plug: The &struct blk_plug that needs to be initialized
3140 *
3141 * Description:
3142 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3143 * pending I/O should the task end up blocking between blk_start_plug() and
3144 * blk_finish_plug(). This is important from a performance perspective, but
3145 * also ensures that we don't deadlock. For instance, if the task is blocking
3146 * for a memory allocation, memory reclaim could end up wanting to free a
3147 * page belonging to that request that is currently residing in our private
3148 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3149 * this kind of deadlock.
3150 */
3151void blk_start_plug(struct blk_plug *plug)
3152{
3153 struct task_struct *tsk = current;
3154
3155 /*
3156 * If this is a nested plug, don't actually assign it.
3157 */
3158 if (tsk->plug)
3159 return;
3160
3161 INIT_LIST_HEAD(&plug->list);
3162 INIT_LIST_HEAD(&plug->mq_list);
3163 INIT_LIST_HEAD(&plug->cb_list);
3164 /*
3165 * Store ordering should not be needed here, since a potential
3166 * preempt will imply a full memory barrier
3167 */
3168 tsk->plug = plug;
3169}
3170EXPORT_SYMBOL(blk_start_plug);
3171
3172static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
3173{
3174 struct request *rqa = container_of(a, struct request, queuelist);
3175 struct request *rqb = container_of(b, struct request, queuelist);
3176
3177 return !(rqa->q < rqb->q ||
3178 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
3179}
3180
3181/*
3182 * If 'from_schedule' is true, then postpone the dispatch of requests
3183 * until a safe kblockd context. We due this to avoid accidental big
3184 * additional stack usage in driver dispatch, in places where the originally
3185 * plugger did not intend it.
3186 */
3187static void queue_unplugged(struct request_queue *q, unsigned int depth,
3188 bool from_schedule)
3189 __releases(q->queue_lock)
3190{
3191 trace_block_unplug(q, depth, !from_schedule);
3192
3193 if (from_schedule)
3194 blk_run_queue_async(q);
3195 else
3196 __blk_run_queue(q);
3197 spin_unlock(q->queue_lock);
3198}
3199
3200static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
3201{
3202 LIST_HEAD(callbacks);
3203
3204 while (!list_empty(&plug->cb_list)) {
3205 list_splice_init(&plug->cb_list, &callbacks);
3206
3207 while (!list_empty(&callbacks)) {
3208 struct blk_plug_cb *cb = list_first_entry(&callbacks,
3209 struct blk_plug_cb,
3210 list);
3211 list_del(&cb->list);
3212 cb->callback(cb, from_schedule);
3213 }
3214 }
3215}
3216
3217struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3218 int size)
3219{
3220 struct blk_plug *plug = current->plug;
3221 struct blk_plug_cb *cb;
3222
3223 if (!plug)
3224 return NULL;
3225
3226 list_for_each_entry(cb, &plug->cb_list, list)
3227 if (cb->callback == unplug && cb->data == data)
3228 return cb;
3229
3230 /* Not currently on the callback list */
3231 BUG_ON(size < sizeof(*cb));
3232 cb = kzalloc(size, GFP_ATOMIC);
3233 if (cb) {
3234 cb->data = data;
3235 cb->callback = unplug;
3236 list_add(&cb->list, &plug->cb_list);
3237 }
3238 return cb;
3239}
3240EXPORT_SYMBOL(blk_check_plugged);
3241
3242void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
3243{
3244 struct request_queue *q;
3245 unsigned long flags;
3246 struct request *rq;
3247 LIST_HEAD(list);
3248 unsigned int depth;
3249
3250 flush_plug_callbacks(plug, from_schedule);
3251
3252 if (!list_empty(&plug->mq_list))
3253 blk_mq_flush_plug_list(plug, from_schedule);
3254
3255 if (list_empty(&plug->list))
3256 return;
3257
3258 list_splice_init(&plug->list, &list);
3259
3260 list_sort(NULL, &list, plug_rq_cmp);
3261
3262 q = NULL;
3263 depth = 0;
3264
3265 /*
3266 * Save and disable interrupts here, to avoid doing it for every
3267 * queue lock we have to take.
3268 */
3269 local_irq_save(flags);
3270 while (!list_empty(&list)) {
3271 rq = list_entry_rq(list.next);
3272 list_del_init(&rq->queuelist);
3273 BUG_ON(!rq->q);
3274 if (rq->q != q) {
3275 /*
3276 * This drops the queue lock
3277 */
3278 if (q)
3279 queue_unplugged(q, depth, from_schedule);
3280 q = rq->q;
3281 depth = 0;
3282 spin_lock(q->queue_lock);
3283 }
3284
3285 /*
3286 * Short-circuit if @q is dead
3287 */
3288 if (unlikely(blk_queue_dying(q))) {
3289 __blk_end_request_all(rq, -ENODEV);
3290 continue;
3291 }
3292
3293 /*
3294 * rq is already accounted, so use raw insert
3295 */
3296 if (rq->cmd_flags & (REQ_PREFLUSH | REQ_FUA))
3297 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3298 else
3299 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
3300
3301 depth++;
3302 }
3303
3304 /*
3305 * This drops the queue lock
3306 */
3307 if (q)
3308 queue_unplugged(q, depth, from_schedule);
3309
3310 local_irq_restore(flags);
3311}
3312
3313void blk_finish_plug(struct blk_plug *plug)
3314{
3315 if (plug != current->plug)
3316 return;
3317 blk_flush_plug_list(plug, false);
3318
3319 current->plug = NULL;
3320}
3321EXPORT_SYMBOL(blk_finish_plug);
3322
3323#ifdef CONFIG_PM
3324/**
3325 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3326 * @q: the queue of the device
3327 * @dev: the device the queue belongs to
3328 *
3329 * Description:
3330 * Initialize runtime-PM-related fields for @q and start auto suspend for
3331 * @dev. Drivers that want to take advantage of request-based runtime PM
3332 * should call this function after @dev has been initialized, and its
3333 * request queue @q has been allocated, and runtime PM for it can not happen
3334 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3335 * cases, driver should call this function before any I/O has taken place.
3336 *
3337 * This function takes care of setting up using auto suspend for the device,
3338 * the autosuspend delay is set to -1 to make runtime suspend impossible
3339 * until an updated value is either set by user or by driver. Drivers do
3340 * not need to touch other autosuspend settings.
3341 *
3342 * The block layer runtime PM is request based, so only works for drivers
3343 * that use request as their IO unit instead of those directly use bio's.
3344 */
3345void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3346{
3347 q->dev = dev;
3348 q->rpm_status = RPM_ACTIVE;
3349 pm_runtime_set_autosuspend_delay(q->dev, -1);
3350 pm_runtime_use_autosuspend(q->dev);
3351}
3352EXPORT_SYMBOL(blk_pm_runtime_init);
3353
3354/**
3355 * blk_pre_runtime_suspend - Pre runtime suspend check
3356 * @q: the queue of the device
3357 *
3358 * Description:
3359 * This function will check if runtime suspend is allowed for the device
3360 * by examining if there are any requests pending in the queue. If there
3361 * are requests pending, the device can not be runtime suspended; otherwise,
3362 * the queue's status will be updated to SUSPENDING and the driver can
3363 * proceed to suspend the device.
3364 *
3365 * For the not allowed case, we mark last busy for the device so that
3366 * runtime PM core will try to autosuspend it some time later.
3367 *
3368 * This function should be called near the start of the device's
3369 * runtime_suspend callback.
3370 *
3371 * Return:
3372 * 0 - OK to runtime suspend the device
3373 * -EBUSY - Device should not be runtime suspended
3374 */
3375int blk_pre_runtime_suspend(struct request_queue *q)
3376{
3377 int ret = 0;
3378
3379 if (!q->dev)
3380 return ret;
3381
3382 spin_lock_irq(q->queue_lock);
3383 if (q->nr_pending) {
3384 ret = -EBUSY;
3385 pm_runtime_mark_last_busy(q->dev);
3386 } else {
3387 q->rpm_status = RPM_SUSPENDING;
3388 }
3389 spin_unlock_irq(q->queue_lock);
3390 return ret;
3391}
3392EXPORT_SYMBOL(blk_pre_runtime_suspend);
3393
3394/**
3395 * blk_post_runtime_suspend - Post runtime suspend processing
3396 * @q: the queue of the device
3397 * @err: return value of the device's runtime_suspend function
3398 *
3399 * Description:
3400 * Update the queue's runtime status according to the return value of the
3401 * device's runtime suspend function and mark last busy for the device so
3402 * that PM core will try to auto suspend the device at a later time.
3403 *
3404 * This function should be called near the end of the device's
3405 * runtime_suspend callback.
3406 */
3407void blk_post_runtime_suspend(struct request_queue *q, int err)
3408{
3409 if (!q->dev)
3410 return;
3411
3412 spin_lock_irq(q->queue_lock);
3413 if (!err) {
3414 q->rpm_status = RPM_SUSPENDED;
3415 } else {
3416 q->rpm_status = RPM_ACTIVE;
3417 pm_runtime_mark_last_busy(q->dev);
3418 }
3419 spin_unlock_irq(q->queue_lock);
3420}
3421EXPORT_SYMBOL(blk_post_runtime_suspend);
3422
3423/**
3424 * blk_pre_runtime_resume - Pre runtime resume processing
3425 * @q: the queue of the device
3426 *
3427 * Description:
3428 * Update the queue's runtime status to RESUMING in preparation for the
3429 * runtime resume of the device.
3430 *
3431 * This function should be called near the start of the device's
3432 * runtime_resume callback.
3433 */
3434void blk_pre_runtime_resume(struct request_queue *q)
3435{
3436 if (!q->dev)
3437 return;
3438
3439 spin_lock_irq(q->queue_lock);
3440 q->rpm_status = RPM_RESUMING;
3441 spin_unlock_irq(q->queue_lock);
3442}
3443EXPORT_SYMBOL(blk_pre_runtime_resume);
3444
3445/**
3446 * blk_post_runtime_resume - Post runtime resume processing
3447 * @q: the queue of the device
3448 * @err: return value of the device's runtime_resume function
3449 *
3450 * Description:
3451 * Update the queue's runtime status according to the return value of the
3452 * device's runtime_resume function. If it is successfully resumed, process
3453 * the requests that are queued into the device's queue when it is resuming
3454 * and then mark last busy and initiate autosuspend for it.
3455 *
3456 * This function should be called near the end of the device's
3457 * runtime_resume callback.
3458 */
3459void blk_post_runtime_resume(struct request_queue *q, int err)
3460{
3461 if (!q->dev)
3462 return;
3463
3464 spin_lock_irq(q->queue_lock);
3465 if (!err) {
3466 q->rpm_status = RPM_ACTIVE;
3467 __blk_run_queue(q);
3468 pm_runtime_mark_last_busy(q->dev);
3469 pm_request_autosuspend(q->dev);
3470 } else {
3471 q->rpm_status = RPM_SUSPENDED;
3472 }
3473 spin_unlock_irq(q->queue_lock);
3474}
3475EXPORT_SYMBOL(blk_post_runtime_resume);
3476
3477/**
3478 * blk_set_runtime_active - Force runtime status of the queue to be active
3479 * @q: the queue of the device
3480 *
3481 * If the device is left runtime suspended during system suspend the resume
3482 * hook typically resumes the device and corrects runtime status
3483 * accordingly. However, that does not affect the queue runtime PM status
3484 * which is still "suspended". This prevents processing requests from the
3485 * queue.
3486 *
3487 * This function can be used in driver's resume hook to correct queue
3488 * runtime PM status and re-enable peeking requests from the queue. It
3489 * should be called before first request is added to the queue.
3490 */
3491void blk_set_runtime_active(struct request_queue *q)
3492{
3493 spin_lock_irq(q->queue_lock);
3494 q->rpm_status = RPM_ACTIVE;
3495 pm_runtime_mark_last_busy(q->dev);
3496 pm_request_autosuspend(q->dev);
3497 spin_unlock_irq(q->queue_lock);
3498}
3499EXPORT_SYMBOL(blk_set_runtime_active);
3500#endif
3501
3502int __init blk_dev_init(void)
3503{
3504 BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
3505 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
3506 FIELD_SIZEOF(struct request, cmd_flags));
3507 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
3508 FIELD_SIZEOF(struct bio, bi_opf));
3509
3510 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3511 kblockd_workqueue = alloc_workqueue("kblockd",
3512 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
3513 if (!kblockd_workqueue)
3514 panic("Failed to create kblockd\n");
3515
3516 request_cachep = kmem_cache_create("blkdev_requests",
3517 sizeof(struct request), 0, SLAB_PANIC, NULL);
3518
3519 blk_requestq_cachep = kmem_cache_create("request_queue",
3520 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
3521
3522 return 0;
3523}