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#include <linux/debugfs.h>
37#include <linux/bpf.h>
38
39#define CREATE_TRACE_POINTS
40#include <trace/events/block.h>
41
42#include "blk.h"
43#include "blk-mq.h"
44#include "blk-mq-sched.h"
45#include "blk-pm.h"
46#include "blk-rq-qos.h"
47
48#ifdef CONFIG_DEBUG_FS
49struct dentry *blk_debugfs_root;
50#endif
51
52EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
53EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
54EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
55EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
56EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
57
58DEFINE_IDA(blk_queue_ida);
59
60/*
61 * For queue allocation
62 */
63struct kmem_cache *blk_requestq_cachep;
64
65/*
66 * Controlling structure to kblockd
67 */
68static struct workqueue_struct *kblockd_workqueue;
69
70/**
71 * blk_queue_flag_set - atomically set a queue flag
72 * @flag: flag to be set
73 * @q: request queue
74 */
75void blk_queue_flag_set(unsigned int flag, struct request_queue *q)
76{
77 set_bit(flag, &q->queue_flags);
78}
79EXPORT_SYMBOL(blk_queue_flag_set);
80
81/**
82 * blk_queue_flag_clear - atomically clear a queue flag
83 * @flag: flag to be cleared
84 * @q: request queue
85 */
86void blk_queue_flag_clear(unsigned int flag, struct request_queue *q)
87{
88 clear_bit(flag, &q->queue_flags);
89}
90EXPORT_SYMBOL(blk_queue_flag_clear);
91
92/**
93 * blk_queue_flag_test_and_set - atomically test and set a queue flag
94 * @flag: flag to be set
95 * @q: request queue
96 *
97 * Returns the previous value of @flag - 0 if the flag was not set and 1 if
98 * the flag was already set.
99 */
100bool blk_queue_flag_test_and_set(unsigned int flag, struct request_queue *q)
101{
102 return test_and_set_bit(flag, &q->queue_flags);
103}
104EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set);
105
106void blk_rq_init(struct request_queue *q, struct request *rq)
107{
108 memset(rq, 0, sizeof(*rq));
109
110 INIT_LIST_HEAD(&rq->queuelist);
111 rq->q = q;
112 rq->__sector = (sector_t) -1;
113 INIT_HLIST_NODE(&rq->hash);
114 RB_CLEAR_NODE(&rq->rb_node);
115 rq->tag = -1;
116 rq->internal_tag = -1;
117 rq->start_time_ns = ktime_get_ns();
118 rq->part = NULL;
119}
120EXPORT_SYMBOL(blk_rq_init);
121
122static const struct {
123 int errno;
124 const char *name;
125} blk_errors[] = {
126 [BLK_STS_OK] = { 0, "" },
127 [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" },
128 [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" },
129 [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" },
130 [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" },
131 [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" },
132 [BLK_STS_NEXUS] = { -EBADE, "critical nexus" },
133 [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" },
134 [BLK_STS_PROTECTION] = { -EILSEQ, "protection" },
135 [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" },
136 [BLK_STS_DEV_RESOURCE] = { -EBUSY, "device resource" },
137 [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" },
138
139 /* device mapper special case, should not leak out: */
140 [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" },
141
142 /* everything else not covered above: */
143 [BLK_STS_IOERR] = { -EIO, "I/O" },
144};
145
146blk_status_t errno_to_blk_status(int errno)
147{
148 int i;
149
150 for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
151 if (blk_errors[i].errno == errno)
152 return (__force blk_status_t)i;
153 }
154
155 return BLK_STS_IOERR;
156}
157EXPORT_SYMBOL_GPL(errno_to_blk_status);
158
159int blk_status_to_errno(blk_status_t status)
160{
161 int idx = (__force int)status;
162
163 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
164 return -EIO;
165 return blk_errors[idx].errno;
166}
167EXPORT_SYMBOL_GPL(blk_status_to_errno);
168
169static void print_req_error(struct request *req, blk_status_t status)
170{
171 int idx = (__force int)status;
172
173 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
174 return;
175
176 printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu flags %x\n",
177 __func__, blk_errors[idx].name,
178 req->rq_disk ? req->rq_disk->disk_name : "?",
179 (unsigned long long)blk_rq_pos(req),
180 req->cmd_flags);
181}
182
183static void req_bio_endio(struct request *rq, struct bio *bio,
184 unsigned int nbytes, blk_status_t error)
185{
186 if (error)
187 bio->bi_status = error;
188
189 if (unlikely(rq->rq_flags & RQF_QUIET))
190 bio_set_flag(bio, BIO_QUIET);
191
192 bio_advance(bio, nbytes);
193
194 /* don't actually finish bio if it's part of flush sequence */
195 if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ))
196 bio_endio(bio);
197}
198
199void blk_dump_rq_flags(struct request *rq, char *msg)
200{
201 printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg,
202 rq->rq_disk ? rq->rq_disk->disk_name : "?",
203 (unsigned long long) rq->cmd_flags);
204
205 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
206 (unsigned long long)blk_rq_pos(rq),
207 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
208 printk(KERN_INFO " bio %p, biotail %p, len %u\n",
209 rq->bio, rq->biotail, blk_rq_bytes(rq));
210}
211EXPORT_SYMBOL(blk_dump_rq_flags);
212
213/**
214 * blk_sync_queue - cancel any pending callbacks on a queue
215 * @q: the queue
216 *
217 * Description:
218 * The block layer may perform asynchronous callback activity
219 * on a queue, such as calling the unplug function after a timeout.
220 * A block device may call blk_sync_queue to ensure that any
221 * such activity is cancelled, thus allowing it to release resources
222 * that the callbacks might use. The caller must already have made sure
223 * that its ->make_request_fn will not re-add plugging prior to calling
224 * this function.
225 *
226 * This function does not cancel any asynchronous activity arising
227 * out of elevator or throttling code. That would require elevator_exit()
228 * and blkcg_exit_queue() to be called with queue lock initialized.
229 *
230 */
231void blk_sync_queue(struct request_queue *q)
232{
233 del_timer_sync(&q->timeout);
234 cancel_work_sync(&q->timeout_work);
235
236 if (queue_is_mq(q)) {
237 struct blk_mq_hw_ctx *hctx;
238 int i;
239
240 cancel_delayed_work_sync(&q->requeue_work);
241 queue_for_each_hw_ctx(q, hctx, i)
242 cancel_delayed_work_sync(&hctx->run_work);
243 }
244}
245EXPORT_SYMBOL(blk_sync_queue);
246
247/**
248 * blk_set_pm_only - increment pm_only counter
249 * @q: request queue pointer
250 */
251void blk_set_pm_only(struct request_queue *q)
252{
253 atomic_inc(&q->pm_only);
254}
255EXPORT_SYMBOL_GPL(blk_set_pm_only);
256
257void blk_clear_pm_only(struct request_queue *q)
258{
259 int pm_only;
260
261 pm_only = atomic_dec_return(&q->pm_only);
262 WARN_ON_ONCE(pm_only < 0);
263 if (pm_only == 0)
264 wake_up_all(&q->mq_freeze_wq);
265}
266EXPORT_SYMBOL_GPL(blk_clear_pm_only);
267
268void blk_put_queue(struct request_queue *q)
269{
270 kobject_put(&q->kobj);
271}
272EXPORT_SYMBOL(blk_put_queue);
273
274void blk_set_queue_dying(struct request_queue *q)
275{
276 blk_queue_flag_set(QUEUE_FLAG_DYING, q);
277
278 /*
279 * When queue DYING flag is set, we need to block new req
280 * entering queue, so we call blk_freeze_queue_start() to
281 * prevent I/O from crossing blk_queue_enter().
282 */
283 blk_freeze_queue_start(q);
284
285 if (queue_is_mq(q))
286 blk_mq_wake_waiters(q);
287
288 /* Make blk_queue_enter() reexamine the DYING flag. */
289 wake_up_all(&q->mq_freeze_wq);
290}
291EXPORT_SYMBOL_GPL(blk_set_queue_dying);
292
293/* Unconfigure the I/O scheduler and dissociate from the cgroup controller. */
294void blk_exit_queue(struct request_queue *q)
295{
296 /*
297 * Since the I/O scheduler exit code may access cgroup information,
298 * perform I/O scheduler exit before disassociating from the block
299 * cgroup controller.
300 */
301 if (q->elevator) {
302 ioc_clear_queue(q);
303 elevator_exit(q, q->elevator);
304 q->elevator = NULL;
305 }
306
307 /*
308 * Remove all references to @q from the block cgroup controller before
309 * restoring @q->queue_lock to avoid that restoring this pointer causes
310 * e.g. blkcg_print_blkgs() to crash.
311 */
312 blkcg_exit_queue(q);
313
314 /*
315 * Since the cgroup code may dereference the @q->backing_dev_info
316 * pointer, only decrease its reference count after having removed the
317 * association with the block cgroup controller.
318 */
319 bdi_put(q->backing_dev_info);
320}
321
322/**
323 * blk_cleanup_queue - shutdown a request queue
324 * @q: request queue to shutdown
325 *
326 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
327 * put it. All future requests will be failed immediately with -ENODEV.
328 */
329void blk_cleanup_queue(struct request_queue *q)
330{
331 /* mark @q DYING, no new request or merges will be allowed afterwards */
332 mutex_lock(&q->sysfs_lock);
333 blk_set_queue_dying(q);
334
335 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q);
336 blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
337 blk_queue_flag_set(QUEUE_FLAG_DYING, q);
338 mutex_unlock(&q->sysfs_lock);
339
340 /*
341 * Drain all requests queued before DYING marking. Set DEAD flag to
342 * prevent that q->request_fn() gets invoked after draining finished.
343 */
344 blk_freeze_queue(q);
345
346 rq_qos_exit(q);
347
348 blk_queue_flag_set(QUEUE_FLAG_DEAD, q);
349
350 /*
351 * make sure all in-progress dispatch are completed because
352 * blk_freeze_queue() can only complete all requests, and
353 * dispatch may still be in-progress since we dispatch requests
354 * from more than one contexts.
355 *
356 * We rely on driver to deal with the race in case that queue
357 * initialization isn't done.
358 */
359 if (queue_is_mq(q) && blk_queue_init_done(q))
360 blk_mq_quiesce_queue(q);
361
362 /* for synchronous bio-based driver finish in-flight integrity i/o */
363 blk_flush_integrity();
364
365 /* @q won't process any more request, flush async actions */
366 del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer);
367 blk_sync_queue(q);
368
369 /*
370 * I/O scheduler exit is only safe after the sysfs scheduler attribute
371 * has been removed.
372 */
373 WARN_ON_ONCE(q->kobj.state_in_sysfs);
374
375 blk_exit_queue(q);
376
377 if (queue_is_mq(q))
378 blk_mq_free_queue(q);
379
380 percpu_ref_exit(&q->q_usage_counter);
381
382 /* @q is and will stay empty, shutdown and put */
383 blk_put_queue(q);
384}
385EXPORT_SYMBOL(blk_cleanup_queue);
386
387struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
388{
389 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
390}
391EXPORT_SYMBOL(blk_alloc_queue);
392
393/**
394 * blk_queue_enter() - try to increase q->q_usage_counter
395 * @q: request queue pointer
396 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PREEMPT
397 */
398int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
399{
400 const bool pm = flags & BLK_MQ_REQ_PREEMPT;
401
402 while (true) {
403 bool success = false;
404
405 rcu_read_lock();
406 if (percpu_ref_tryget_live(&q->q_usage_counter)) {
407 /*
408 * The code that increments the pm_only counter is
409 * responsible for ensuring that that counter is
410 * globally visible before the queue is unfrozen.
411 */
412 if (pm || !blk_queue_pm_only(q)) {
413 success = true;
414 } else {
415 percpu_ref_put(&q->q_usage_counter);
416 }
417 }
418 rcu_read_unlock();
419
420 if (success)
421 return 0;
422
423 if (flags & BLK_MQ_REQ_NOWAIT)
424 return -EBUSY;
425
426 /*
427 * read pair of barrier in blk_freeze_queue_start(),
428 * we need to order reading __PERCPU_REF_DEAD flag of
429 * .q_usage_counter and reading .mq_freeze_depth or
430 * queue dying flag, otherwise the following wait may
431 * never return if the two reads are reordered.
432 */
433 smp_rmb();
434
435 wait_event(q->mq_freeze_wq,
436 (atomic_read(&q->mq_freeze_depth) == 0 &&
437 (pm || (blk_pm_request_resume(q),
438 !blk_queue_pm_only(q)))) ||
439 blk_queue_dying(q));
440 if (blk_queue_dying(q))
441 return -ENODEV;
442 }
443}
444
445void blk_queue_exit(struct request_queue *q)
446{
447 percpu_ref_put(&q->q_usage_counter);
448}
449
450static void blk_queue_usage_counter_release(struct percpu_ref *ref)
451{
452 struct request_queue *q =
453 container_of(ref, struct request_queue, q_usage_counter);
454
455 wake_up_all(&q->mq_freeze_wq);
456}
457
458static void blk_rq_timed_out_timer(struct timer_list *t)
459{
460 struct request_queue *q = from_timer(q, t, timeout);
461
462 kblockd_schedule_work(&q->timeout_work);
463}
464
465static void blk_timeout_work(struct work_struct *work)
466{
467}
468
469/**
470 * blk_alloc_queue_node - allocate a request queue
471 * @gfp_mask: memory allocation flags
472 * @node_id: NUMA node to allocate memory from
473 */
474struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
475{
476 struct request_queue *q;
477 int ret;
478
479 q = kmem_cache_alloc_node(blk_requestq_cachep,
480 gfp_mask | __GFP_ZERO, node_id);
481 if (!q)
482 return NULL;
483
484 INIT_LIST_HEAD(&q->queue_head);
485 q->last_merge = NULL;
486
487 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
488 if (q->id < 0)
489 goto fail_q;
490
491 ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
492 if (ret)
493 goto fail_id;
494
495 q->backing_dev_info = bdi_alloc_node(gfp_mask, node_id);
496 if (!q->backing_dev_info)
497 goto fail_split;
498
499 q->stats = blk_alloc_queue_stats();
500 if (!q->stats)
501 goto fail_stats;
502
503 q->backing_dev_info->ra_pages = VM_READAHEAD_PAGES;
504 q->backing_dev_info->capabilities = BDI_CAP_CGROUP_WRITEBACK;
505 q->backing_dev_info->name = "block";
506 q->node = node_id;
507
508 timer_setup(&q->backing_dev_info->laptop_mode_wb_timer,
509 laptop_mode_timer_fn, 0);
510 timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
511 INIT_WORK(&q->timeout_work, blk_timeout_work);
512 INIT_LIST_HEAD(&q->icq_list);
513#ifdef CONFIG_BLK_CGROUP
514 INIT_LIST_HEAD(&q->blkg_list);
515#endif
516
517 kobject_init(&q->kobj, &blk_queue_ktype);
518
519#ifdef CONFIG_BLK_DEV_IO_TRACE
520 mutex_init(&q->blk_trace_mutex);
521#endif
522 mutex_init(&q->sysfs_lock);
523 spin_lock_init(&q->queue_lock);
524
525 init_waitqueue_head(&q->mq_freeze_wq);
526
527 /*
528 * Init percpu_ref in atomic mode so that it's faster to shutdown.
529 * See blk_register_queue() for details.
530 */
531 if (percpu_ref_init(&q->q_usage_counter,
532 blk_queue_usage_counter_release,
533 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
534 goto fail_bdi;
535
536 if (blkcg_init_queue(q))
537 goto fail_ref;
538
539 return q;
540
541fail_ref:
542 percpu_ref_exit(&q->q_usage_counter);
543fail_bdi:
544 blk_free_queue_stats(q->stats);
545fail_stats:
546 bdi_put(q->backing_dev_info);
547fail_split:
548 bioset_exit(&q->bio_split);
549fail_id:
550 ida_simple_remove(&blk_queue_ida, q->id);
551fail_q:
552 kmem_cache_free(blk_requestq_cachep, q);
553 return NULL;
554}
555EXPORT_SYMBOL(blk_alloc_queue_node);
556
557bool blk_get_queue(struct request_queue *q)
558{
559 if (likely(!blk_queue_dying(q))) {
560 __blk_get_queue(q);
561 return true;
562 }
563
564 return false;
565}
566EXPORT_SYMBOL(blk_get_queue);
567
568/**
569 * blk_get_request - allocate a request
570 * @q: request queue to allocate a request for
571 * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC.
572 * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
573 */
574struct request *blk_get_request(struct request_queue *q, unsigned int op,
575 blk_mq_req_flags_t flags)
576{
577 struct request *req;
578
579 WARN_ON_ONCE(op & REQ_NOWAIT);
580 WARN_ON_ONCE(flags & ~(BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_PREEMPT));
581
582 req = blk_mq_alloc_request(q, op, flags);
583 if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn)
584 q->mq_ops->initialize_rq_fn(req);
585
586 return req;
587}
588EXPORT_SYMBOL(blk_get_request);
589
590void blk_put_request(struct request *req)
591{
592 blk_mq_free_request(req);
593}
594EXPORT_SYMBOL(blk_put_request);
595
596bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
597 struct bio *bio)
598{
599 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
600
601 if (!ll_back_merge_fn(q, req, bio))
602 return false;
603
604 trace_block_bio_backmerge(q, req, bio);
605
606 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
607 blk_rq_set_mixed_merge(req);
608
609 req->biotail->bi_next = bio;
610 req->biotail = bio;
611 req->__data_len += bio->bi_iter.bi_size;
612
613 blk_account_io_start(req, false);
614 return true;
615}
616
617bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
618 struct bio *bio)
619{
620 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
621
622 if (!ll_front_merge_fn(q, req, bio))
623 return false;
624
625 trace_block_bio_frontmerge(q, req, bio);
626
627 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
628 blk_rq_set_mixed_merge(req);
629
630 bio->bi_next = req->bio;
631 req->bio = bio;
632
633 req->__sector = bio->bi_iter.bi_sector;
634 req->__data_len += bio->bi_iter.bi_size;
635
636 blk_account_io_start(req, false);
637 return true;
638}
639
640bool bio_attempt_discard_merge(struct request_queue *q, struct request *req,
641 struct bio *bio)
642{
643 unsigned short segments = blk_rq_nr_discard_segments(req);
644
645 if (segments >= queue_max_discard_segments(q))
646 goto no_merge;
647 if (blk_rq_sectors(req) + bio_sectors(bio) >
648 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
649 goto no_merge;
650
651 req->biotail->bi_next = bio;
652 req->biotail = bio;
653 req->__data_len += bio->bi_iter.bi_size;
654 req->nr_phys_segments = segments + 1;
655
656 blk_account_io_start(req, false);
657 return true;
658no_merge:
659 req_set_nomerge(q, req);
660 return false;
661}
662
663/**
664 * blk_attempt_plug_merge - try to merge with %current's plugged list
665 * @q: request_queue new bio is being queued at
666 * @bio: new bio being queued
667 * @same_queue_rq: pointer to &struct request that gets filled in when
668 * another request associated with @q is found on the plug list
669 * (optional, may be %NULL)
670 *
671 * Determine whether @bio being queued on @q can be merged with a request
672 * on %current's plugged list. Returns %true if merge was successful,
673 * otherwise %false.
674 *
675 * Plugging coalesces IOs from the same issuer for the same purpose without
676 * going through @q->queue_lock. As such it's more of an issuing mechanism
677 * than scheduling, and the request, while may have elvpriv data, is not
678 * added on the elevator at this point. In addition, we don't have
679 * reliable access to the elevator outside queue lock. Only check basic
680 * merging parameters without querying the elevator.
681 *
682 * Caller must ensure !blk_queue_nomerges(q) beforehand.
683 */
684bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
685 struct request **same_queue_rq)
686{
687 struct blk_plug *plug;
688 struct request *rq;
689 struct list_head *plug_list;
690
691 plug = current->plug;
692 if (!plug)
693 return false;
694
695 plug_list = &plug->mq_list;
696
697 list_for_each_entry_reverse(rq, plug_list, queuelist) {
698 bool merged = false;
699
700 if (rq->q == q && same_queue_rq) {
701 /*
702 * Only blk-mq multiple hardware queues case checks the
703 * rq in the same queue, there should be only one such
704 * rq in a queue
705 **/
706 *same_queue_rq = rq;
707 }
708
709 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
710 continue;
711
712 switch (blk_try_merge(rq, bio)) {
713 case ELEVATOR_BACK_MERGE:
714 merged = bio_attempt_back_merge(q, rq, bio);
715 break;
716 case ELEVATOR_FRONT_MERGE:
717 merged = bio_attempt_front_merge(q, rq, bio);
718 break;
719 case ELEVATOR_DISCARD_MERGE:
720 merged = bio_attempt_discard_merge(q, rq, bio);
721 break;
722 default:
723 break;
724 }
725
726 if (merged)
727 return true;
728 }
729
730 return false;
731}
732
733void blk_init_request_from_bio(struct request *req, struct bio *bio)
734{
735 if (bio->bi_opf & REQ_RAHEAD)
736 req->cmd_flags |= REQ_FAILFAST_MASK;
737
738 req->__sector = bio->bi_iter.bi_sector;
739 req->ioprio = bio_prio(bio);
740 req->write_hint = bio->bi_write_hint;
741 blk_rq_bio_prep(req->q, req, bio);
742}
743EXPORT_SYMBOL_GPL(blk_init_request_from_bio);
744
745static void handle_bad_sector(struct bio *bio, sector_t maxsector)
746{
747 char b[BDEVNAME_SIZE];
748
749 printk(KERN_INFO "attempt to access beyond end of device\n");
750 printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n",
751 bio_devname(bio, b), bio->bi_opf,
752 (unsigned long long)bio_end_sector(bio),
753 (long long)maxsector);
754}
755
756#ifdef CONFIG_FAIL_MAKE_REQUEST
757
758static DECLARE_FAULT_ATTR(fail_make_request);
759
760static int __init setup_fail_make_request(char *str)
761{
762 return setup_fault_attr(&fail_make_request, str);
763}
764__setup("fail_make_request=", setup_fail_make_request);
765
766static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
767{
768 return part->make_it_fail && should_fail(&fail_make_request, bytes);
769}
770
771static int __init fail_make_request_debugfs(void)
772{
773 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
774 NULL, &fail_make_request);
775
776 return PTR_ERR_OR_ZERO(dir);
777}
778
779late_initcall(fail_make_request_debugfs);
780
781#else /* CONFIG_FAIL_MAKE_REQUEST */
782
783static inline bool should_fail_request(struct hd_struct *part,
784 unsigned int bytes)
785{
786 return false;
787}
788
789#endif /* CONFIG_FAIL_MAKE_REQUEST */
790
791static inline bool bio_check_ro(struct bio *bio, struct hd_struct *part)
792{
793 const int op = bio_op(bio);
794
795 if (part->policy && op_is_write(op)) {
796 char b[BDEVNAME_SIZE];
797
798 if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
799 return false;
800
801 WARN_ONCE(1,
802 "generic_make_request: Trying to write "
803 "to read-only block-device %s (partno %d)\n",
804 bio_devname(bio, b), part->partno);
805 /* Older lvm-tools actually trigger this */
806 return false;
807 }
808
809 return false;
810}
811
812static noinline int should_fail_bio(struct bio *bio)
813{
814 if (should_fail_request(&bio->bi_disk->part0, bio->bi_iter.bi_size))
815 return -EIO;
816 return 0;
817}
818ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
819
820/*
821 * Check whether this bio extends beyond the end of the device or partition.
822 * This may well happen - the kernel calls bread() without checking the size of
823 * the device, e.g., when mounting a file system.
824 */
825static inline int bio_check_eod(struct bio *bio, sector_t maxsector)
826{
827 unsigned int nr_sectors = bio_sectors(bio);
828
829 if (nr_sectors && maxsector &&
830 (nr_sectors > maxsector ||
831 bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
832 handle_bad_sector(bio, maxsector);
833 return -EIO;
834 }
835 return 0;
836}
837
838/*
839 * Remap block n of partition p to block n+start(p) of the disk.
840 */
841static inline int blk_partition_remap(struct bio *bio)
842{
843 struct hd_struct *p;
844 int ret = -EIO;
845
846 rcu_read_lock();
847 p = __disk_get_part(bio->bi_disk, bio->bi_partno);
848 if (unlikely(!p))
849 goto out;
850 if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
851 goto out;
852 if (unlikely(bio_check_ro(bio, p)))
853 goto out;
854
855 /*
856 * Zone reset does not include bi_size so bio_sectors() is always 0.
857 * Include a test for the reset op code and perform the remap if needed.
858 */
859 if (bio_sectors(bio) || bio_op(bio) == REQ_OP_ZONE_RESET) {
860 if (bio_check_eod(bio, part_nr_sects_read(p)))
861 goto out;
862 bio->bi_iter.bi_sector += p->start_sect;
863 trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p),
864 bio->bi_iter.bi_sector - p->start_sect);
865 }
866 bio->bi_partno = 0;
867 ret = 0;
868out:
869 rcu_read_unlock();
870 return ret;
871}
872
873static noinline_for_stack bool
874generic_make_request_checks(struct bio *bio)
875{
876 struct request_queue *q;
877 int nr_sectors = bio_sectors(bio);
878 blk_status_t status = BLK_STS_IOERR;
879 char b[BDEVNAME_SIZE];
880
881 might_sleep();
882
883 q = bio->bi_disk->queue;
884 if (unlikely(!q)) {
885 printk(KERN_ERR
886 "generic_make_request: Trying to access "
887 "nonexistent block-device %s (%Lu)\n",
888 bio_devname(bio, b), (long long)bio->bi_iter.bi_sector);
889 goto end_io;
890 }
891
892 /*
893 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
894 * if queue is not a request based queue.
895 */
896 if ((bio->bi_opf & REQ_NOWAIT) && !queue_is_mq(q))
897 goto not_supported;
898
899 if (should_fail_bio(bio))
900 goto end_io;
901
902 if (bio->bi_partno) {
903 if (unlikely(blk_partition_remap(bio)))
904 goto end_io;
905 } else {
906 if (unlikely(bio_check_ro(bio, &bio->bi_disk->part0)))
907 goto end_io;
908 if (unlikely(bio_check_eod(bio, get_capacity(bio->bi_disk))))
909 goto end_io;
910 }
911
912 /*
913 * Filter flush bio's early so that make_request based
914 * drivers without flush support don't have to worry
915 * about them.
916 */
917 if (op_is_flush(bio->bi_opf) &&
918 !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
919 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
920 if (!nr_sectors) {
921 status = BLK_STS_OK;
922 goto end_io;
923 }
924 }
925
926 if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
927 bio->bi_opf &= ~REQ_HIPRI;
928
929 switch (bio_op(bio)) {
930 case REQ_OP_DISCARD:
931 if (!blk_queue_discard(q))
932 goto not_supported;
933 break;
934 case REQ_OP_SECURE_ERASE:
935 if (!blk_queue_secure_erase(q))
936 goto not_supported;
937 break;
938 case REQ_OP_WRITE_SAME:
939 if (!q->limits.max_write_same_sectors)
940 goto not_supported;
941 break;
942 case REQ_OP_ZONE_RESET:
943 if (!blk_queue_is_zoned(q))
944 goto not_supported;
945 break;
946 case REQ_OP_WRITE_ZEROES:
947 if (!q->limits.max_write_zeroes_sectors)
948 goto not_supported;
949 break;
950 default:
951 break;
952 }
953
954 /*
955 * Various block parts want %current->io_context and lazy ioc
956 * allocation ends up trading a lot of pain for a small amount of
957 * memory. Just allocate it upfront. This may fail and block
958 * layer knows how to live with it.
959 */
960 create_io_context(GFP_ATOMIC, q->node);
961
962 if (!blkcg_bio_issue_check(q, bio))
963 return false;
964
965 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
966 trace_block_bio_queue(q, bio);
967 /* Now that enqueuing has been traced, we need to trace
968 * completion as well.
969 */
970 bio_set_flag(bio, BIO_TRACE_COMPLETION);
971 }
972 return true;
973
974not_supported:
975 status = BLK_STS_NOTSUPP;
976end_io:
977 bio->bi_status = status;
978 bio_endio(bio);
979 return false;
980}
981
982/**
983 * generic_make_request - hand a buffer to its device driver for I/O
984 * @bio: The bio describing the location in memory and on the device.
985 *
986 * generic_make_request() is used to make I/O requests of block
987 * devices. It is passed a &struct bio, which describes the I/O that needs
988 * to be done.
989 *
990 * generic_make_request() does not return any status. The
991 * success/failure status of the request, along with notification of
992 * completion, is delivered asynchronously through the bio->bi_end_io
993 * function described (one day) else where.
994 *
995 * The caller of generic_make_request must make sure that bi_io_vec
996 * are set to describe the memory buffer, and that bi_dev and bi_sector are
997 * set to describe the device address, and the
998 * bi_end_io and optionally bi_private are set to describe how
999 * completion notification should be signaled.
1000 *
1001 * generic_make_request and the drivers it calls may use bi_next if this
1002 * bio happens to be merged with someone else, and may resubmit the bio to
1003 * a lower device by calling into generic_make_request recursively, which
1004 * means the bio should NOT be touched after the call to ->make_request_fn.
1005 */
1006blk_qc_t generic_make_request(struct bio *bio)
1007{
1008 /*
1009 * bio_list_on_stack[0] contains bios submitted by the current
1010 * make_request_fn.
1011 * bio_list_on_stack[1] contains bios that were submitted before
1012 * the current make_request_fn, but that haven't been processed
1013 * yet.
1014 */
1015 struct bio_list bio_list_on_stack[2];
1016 blk_mq_req_flags_t flags = 0;
1017 struct request_queue *q = bio->bi_disk->queue;
1018 blk_qc_t ret = BLK_QC_T_NONE;
1019
1020 if (bio->bi_opf & REQ_NOWAIT)
1021 flags = BLK_MQ_REQ_NOWAIT;
1022 if (bio_flagged(bio, BIO_QUEUE_ENTERED))
1023 blk_queue_enter_live(q);
1024 else if (blk_queue_enter(q, flags) < 0) {
1025 if (!blk_queue_dying(q) && (bio->bi_opf & REQ_NOWAIT))
1026 bio_wouldblock_error(bio);
1027 else
1028 bio_io_error(bio);
1029 return ret;
1030 }
1031
1032 if (!generic_make_request_checks(bio))
1033 goto out;
1034
1035 /*
1036 * We only want one ->make_request_fn to be active at a time, else
1037 * stack usage with stacked devices could be a problem. So use
1038 * current->bio_list to keep a list of requests submited by a
1039 * make_request_fn function. current->bio_list is also used as a
1040 * flag to say if generic_make_request is currently active in this
1041 * task or not. If it is NULL, then no make_request is active. If
1042 * it is non-NULL, then a make_request is active, and new requests
1043 * should be added at the tail
1044 */
1045 if (current->bio_list) {
1046 bio_list_add(&current->bio_list[0], bio);
1047 goto out;
1048 }
1049
1050 /* following loop may be a bit non-obvious, and so deserves some
1051 * explanation.
1052 * Before entering the loop, bio->bi_next is NULL (as all callers
1053 * ensure that) so we have a list with a single bio.
1054 * We pretend that we have just taken it off a longer list, so
1055 * we assign bio_list to a pointer to the bio_list_on_stack,
1056 * thus initialising the bio_list of new bios to be
1057 * added. ->make_request() may indeed add some more bios
1058 * through a recursive call to generic_make_request. If it
1059 * did, we find a non-NULL value in bio_list and re-enter the loop
1060 * from the top. In this case we really did just take the bio
1061 * of the top of the list (no pretending) and so remove it from
1062 * bio_list, and call into ->make_request() again.
1063 */
1064 BUG_ON(bio->bi_next);
1065 bio_list_init(&bio_list_on_stack[0]);
1066 current->bio_list = bio_list_on_stack;
1067 do {
1068 bool enter_succeeded = true;
1069
1070 if (unlikely(q != bio->bi_disk->queue)) {
1071 if (q)
1072 blk_queue_exit(q);
1073 q = bio->bi_disk->queue;
1074 flags = 0;
1075 if (bio->bi_opf & REQ_NOWAIT)
1076 flags = BLK_MQ_REQ_NOWAIT;
1077 if (blk_queue_enter(q, flags) < 0) {
1078 enter_succeeded = false;
1079 q = NULL;
1080 }
1081 }
1082
1083 if (enter_succeeded) {
1084 struct bio_list lower, same;
1085
1086 /* Create a fresh bio_list for all subordinate requests */
1087 bio_list_on_stack[1] = bio_list_on_stack[0];
1088 bio_list_init(&bio_list_on_stack[0]);
1089 ret = q->make_request_fn(q, bio);
1090
1091 /* sort new bios into those for a lower level
1092 * and those for the same level
1093 */
1094 bio_list_init(&lower);
1095 bio_list_init(&same);
1096 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
1097 if (q == bio->bi_disk->queue)
1098 bio_list_add(&same, bio);
1099 else
1100 bio_list_add(&lower, bio);
1101 /* now assemble so we handle the lowest level first */
1102 bio_list_merge(&bio_list_on_stack[0], &lower);
1103 bio_list_merge(&bio_list_on_stack[0], &same);
1104 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
1105 } else {
1106 if (unlikely(!blk_queue_dying(q) &&
1107 (bio->bi_opf & REQ_NOWAIT)))
1108 bio_wouldblock_error(bio);
1109 else
1110 bio_io_error(bio);
1111 }
1112 bio = bio_list_pop(&bio_list_on_stack[0]);
1113 } while (bio);
1114 current->bio_list = NULL; /* deactivate */
1115
1116out:
1117 if (q)
1118 blk_queue_exit(q);
1119 return ret;
1120}
1121EXPORT_SYMBOL(generic_make_request);
1122
1123/**
1124 * direct_make_request - hand a buffer directly to its device driver for I/O
1125 * @bio: The bio describing the location in memory and on the device.
1126 *
1127 * This function behaves like generic_make_request(), but does not protect
1128 * against recursion. Must only be used if the called driver is known
1129 * to not call generic_make_request (or direct_make_request) again from
1130 * its make_request function. (Calling direct_make_request again from
1131 * a workqueue is perfectly fine as that doesn't recurse).
1132 */
1133blk_qc_t direct_make_request(struct bio *bio)
1134{
1135 struct request_queue *q = bio->bi_disk->queue;
1136 bool nowait = bio->bi_opf & REQ_NOWAIT;
1137 blk_qc_t ret;
1138
1139 if (!generic_make_request_checks(bio))
1140 return BLK_QC_T_NONE;
1141
1142 if (unlikely(blk_queue_enter(q, nowait ? BLK_MQ_REQ_NOWAIT : 0))) {
1143 if (nowait && !blk_queue_dying(q))
1144 bio->bi_status = BLK_STS_AGAIN;
1145 else
1146 bio->bi_status = BLK_STS_IOERR;
1147 bio_endio(bio);
1148 return BLK_QC_T_NONE;
1149 }
1150
1151 ret = q->make_request_fn(q, bio);
1152 blk_queue_exit(q);
1153 return ret;
1154}
1155EXPORT_SYMBOL_GPL(direct_make_request);
1156
1157/**
1158 * submit_bio - submit a bio to the block device layer for I/O
1159 * @bio: The &struct bio which describes the I/O
1160 *
1161 * submit_bio() is very similar in purpose to generic_make_request(), and
1162 * uses that function to do most of the work. Both are fairly rough
1163 * interfaces; @bio must be presetup and ready for I/O.
1164 *
1165 */
1166blk_qc_t submit_bio(struct bio *bio)
1167{
1168 /*
1169 * If it's a regular read/write or a barrier with data attached,
1170 * go through the normal accounting stuff before submission.
1171 */
1172 if (bio_has_data(bio)) {
1173 unsigned int count;
1174
1175 if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
1176 count = queue_logical_block_size(bio->bi_disk->queue) >> 9;
1177 else
1178 count = bio_sectors(bio);
1179
1180 if (op_is_write(bio_op(bio))) {
1181 count_vm_events(PGPGOUT, count);
1182 } else {
1183 task_io_account_read(bio->bi_iter.bi_size);
1184 count_vm_events(PGPGIN, count);
1185 }
1186
1187 if (unlikely(block_dump)) {
1188 char b[BDEVNAME_SIZE];
1189 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1190 current->comm, task_pid_nr(current),
1191 op_is_write(bio_op(bio)) ? "WRITE" : "READ",
1192 (unsigned long long)bio->bi_iter.bi_sector,
1193 bio_devname(bio, b), count);
1194 }
1195 }
1196
1197 return generic_make_request(bio);
1198}
1199EXPORT_SYMBOL(submit_bio);
1200
1201/**
1202 * blk_cloned_rq_check_limits - Helper function to check a cloned request
1203 * for new the queue limits
1204 * @q: the queue
1205 * @rq: the request being checked
1206 *
1207 * Description:
1208 * @rq may have been made based on weaker limitations of upper-level queues
1209 * in request stacking drivers, and it may violate the limitation of @q.
1210 * Since the block layer and the underlying device driver trust @rq
1211 * after it is inserted to @q, it should be checked against @q before
1212 * the insertion using this generic function.
1213 *
1214 * Request stacking drivers like request-based dm may change the queue
1215 * limits when retrying requests on other queues. Those requests need
1216 * to be checked against the new queue limits again during dispatch.
1217 */
1218static int blk_cloned_rq_check_limits(struct request_queue *q,
1219 struct request *rq)
1220{
1221 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) {
1222 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1223 return -EIO;
1224 }
1225
1226 /*
1227 * queue's settings related to segment counting like q->bounce_pfn
1228 * may differ from that of other stacking queues.
1229 * Recalculate it to check the request correctly on this queue's
1230 * limitation.
1231 */
1232 blk_recalc_rq_segments(rq);
1233 if (rq->nr_phys_segments > queue_max_segments(q)) {
1234 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1235 return -EIO;
1236 }
1237
1238 return 0;
1239}
1240
1241/**
1242 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1243 * @q: the queue to submit the request
1244 * @rq: the request being queued
1245 */
1246blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1247{
1248 blk_qc_t unused;
1249
1250 if (blk_cloned_rq_check_limits(q, rq))
1251 return BLK_STS_IOERR;
1252
1253 if (rq->rq_disk &&
1254 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
1255 return BLK_STS_IOERR;
1256
1257 if (blk_queue_io_stat(q))
1258 blk_account_io_start(rq, true);
1259
1260 /*
1261 * Since we have a scheduler attached on the top device,
1262 * bypass a potential scheduler on the bottom device for
1263 * insert.
1264 */
1265 return blk_mq_try_issue_directly(rq->mq_hctx, rq, &unused, true, true);
1266}
1267EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1268
1269/**
1270 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1271 * @rq: request to examine
1272 *
1273 * Description:
1274 * A request could be merge of IOs which require different failure
1275 * handling. This function determines the number of bytes which
1276 * can be failed from the beginning of the request without
1277 * crossing into area which need to be retried further.
1278 *
1279 * Return:
1280 * The number of bytes to fail.
1281 */
1282unsigned int blk_rq_err_bytes(const struct request *rq)
1283{
1284 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1285 unsigned int bytes = 0;
1286 struct bio *bio;
1287
1288 if (!(rq->rq_flags & RQF_MIXED_MERGE))
1289 return blk_rq_bytes(rq);
1290
1291 /*
1292 * Currently the only 'mixing' which can happen is between
1293 * different fastfail types. We can safely fail portions
1294 * which have all the failfast bits that the first one has -
1295 * the ones which are at least as eager to fail as the first
1296 * one.
1297 */
1298 for (bio = rq->bio; bio; bio = bio->bi_next) {
1299 if ((bio->bi_opf & ff) != ff)
1300 break;
1301 bytes += bio->bi_iter.bi_size;
1302 }
1303
1304 /* this could lead to infinite loop */
1305 BUG_ON(blk_rq_bytes(rq) && !bytes);
1306 return bytes;
1307}
1308EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1309
1310void blk_account_io_completion(struct request *req, unsigned int bytes)
1311{
1312 if (blk_do_io_stat(req)) {
1313 const int sgrp = op_stat_group(req_op(req));
1314 struct hd_struct *part;
1315
1316 part_stat_lock();
1317 part = req->part;
1318 part_stat_add(part, sectors[sgrp], bytes >> 9);
1319 part_stat_unlock();
1320 }
1321}
1322
1323void blk_account_io_done(struct request *req, u64 now)
1324{
1325 /*
1326 * Account IO completion. flush_rq isn't accounted as a
1327 * normal IO on queueing nor completion. Accounting the
1328 * containing request is enough.
1329 */
1330 if (blk_do_io_stat(req) && !(req->rq_flags & RQF_FLUSH_SEQ)) {
1331 const int sgrp = op_stat_group(req_op(req));
1332 struct hd_struct *part;
1333
1334 part_stat_lock();
1335 part = req->part;
1336
1337 update_io_ticks(part, jiffies);
1338 part_stat_inc(part, ios[sgrp]);
1339 part_stat_add(part, nsecs[sgrp], now - req->start_time_ns);
1340 part_stat_add(part, time_in_queue, nsecs_to_jiffies64(now - req->start_time_ns));
1341 part_dec_in_flight(req->q, part, rq_data_dir(req));
1342
1343 hd_struct_put(part);
1344 part_stat_unlock();
1345 }
1346}
1347
1348void blk_account_io_start(struct request *rq, bool new_io)
1349{
1350 struct hd_struct *part;
1351 int rw = rq_data_dir(rq);
1352
1353 if (!blk_do_io_stat(rq))
1354 return;
1355
1356 part_stat_lock();
1357
1358 if (!new_io) {
1359 part = rq->part;
1360 part_stat_inc(part, merges[rw]);
1361 } else {
1362 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
1363 if (!hd_struct_try_get(part)) {
1364 /*
1365 * The partition is already being removed,
1366 * the request will be accounted on the disk only
1367 *
1368 * We take a reference on disk->part0 although that
1369 * partition will never be deleted, so we can treat
1370 * it as any other partition.
1371 */
1372 part = &rq->rq_disk->part0;
1373 hd_struct_get(part);
1374 }
1375 part_inc_in_flight(rq->q, part, rw);
1376 rq->part = part;
1377 }
1378
1379 update_io_ticks(part, jiffies);
1380
1381 part_stat_unlock();
1382}
1383
1384/*
1385 * Steal bios from a request and add them to a bio list.
1386 * The request must not have been partially completed before.
1387 */
1388void blk_steal_bios(struct bio_list *list, struct request *rq)
1389{
1390 if (rq->bio) {
1391 if (list->tail)
1392 list->tail->bi_next = rq->bio;
1393 else
1394 list->head = rq->bio;
1395 list->tail = rq->biotail;
1396
1397 rq->bio = NULL;
1398 rq->biotail = NULL;
1399 }
1400
1401 rq->__data_len = 0;
1402}
1403EXPORT_SYMBOL_GPL(blk_steal_bios);
1404
1405/**
1406 * blk_update_request - Special helper function for request stacking drivers
1407 * @req: the request being processed
1408 * @error: block status code
1409 * @nr_bytes: number of bytes to complete @req
1410 *
1411 * Description:
1412 * Ends I/O on a number of bytes attached to @req, but doesn't complete
1413 * the request structure even if @req doesn't have leftover.
1414 * If @req has leftover, sets it up for the next range of segments.
1415 *
1416 * This special helper function is only for request stacking drivers
1417 * (e.g. request-based dm) so that they can handle partial completion.
1418 * Actual device drivers should use blk_end_request instead.
1419 *
1420 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
1421 * %false return from this function.
1422 *
1423 * Note:
1424 * The RQF_SPECIAL_PAYLOAD flag is ignored on purpose in both
1425 * blk_rq_bytes() and in blk_update_request().
1426 *
1427 * Return:
1428 * %false - this request doesn't have any more data
1429 * %true - this request has more data
1430 **/
1431bool blk_update_request(struct request *req, blk_status_t error,
1432 unsigned int nr_bytes)
1433{
1434 int total_bytes;
1435
1436 trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes);
1437
1438 if (!req->bio)
1439 return false;
1440
1441 if (unlikely(error && !blk_rq_is_passthrough(req) &&
1442 !(req->rq_flags & RQF_QUIET)))
1443 print_req_error(req, error);
1444
1445 blk_account_io_completion(req, nr_bytes);
1446
1447 total_bytes = 0;
1448 while (req->bio) {
1449 struct bio *bio = req->bio;
1450 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
1451
1452 if (bio_bytes == bio->bi_iter.bi_size)
1453 req->bio = bio->bi_next;
1454
1455 /* Completion has already been traced */
1456 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1457 req_bio_endio(req, bio, bio_bytes, error);
1458
1459 total_bytes += bio_bytes;
1460 nr_bytes -= bio_bytes;
1461
1462 if (!nr_bytes)
1463 break;
1464 }
1465
1466 /*
1467 * completely done
1468 */
1469 if (!req->bio) {
1470 /*
1471 * Reset counters so that the request stacking driver
1472 * can find how many bytes remain in the request
1473 * later.
1474 */
1475 req->__data_len = 0;
1476 return false;
1477 }
1478
1479 req->__data_len -= total_bytes;
1480
1481 /* update sector only for requests with clear definition of sector */
1482 if (!blk_rq_is_passthrough(req))
1483 req->__sector += total_bytes >> 9;
1484
1485 /* mixed attributes always follow the first bio */
1486 if (req->rq_flags & RQF_MIXED_MERGE) {
1487 req->cmd_flags &= ~REQ_FAILFAST_MASK;
1488 req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
1489 }
1490
1491 if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) {
1492 /*
1493 * If total number of sectors is less than the first segment
1494 * size, something has gone terribly wrong.
1495 */
1496 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
1497 blk_dump_rq_flags(req, "request botched");
1498 req->__data_len = blk_rq_cur_bytes(req);
1499 }
1500
1501 /* recalculate the number of segments */
1502 blk_recalc_rq_segments(req);
1503 }
1504
1505 return true;
1506}
1507EXPORT_SYMBOL_GPL(blk_update_request);
1508
1509void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
1510 struct bio *bio)
1511{
1512 if (bio_has_data(bio))
1513 rq->nr_phys_segments = bio_phys_segments(q, bio);
1514 else if (bio_op(bio) == REQ_OP_DISCARD)
1515 rq->nr_phys_segments = 1;
1516
1517 rq->__data_len = bio->bi_iter.bi_size;
1518 rq->bio = rq->biotail = bio;
1519
1520 if (bio->bi_disk)
1521 rq->rq_disk = bio->bi_disk;
1522}
1523
1524#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1525/**
1526 * rq_flush_dcache_pages - Helper function to flush all pages in a request
1527 * @rq: the request to be flushed
1528 *
1529 * Description:
1530 * Flush all pages in @rq.
1531 */
1532void rq_flush_dcache_pages(struct request *rq)
1533{
1534 struct req_iterator iter;
1535 struct bio_vec bvec;
1536
1537 rq_for_each_segment(bvec, rq, iter)
1538 flush_dcache_page(bvec.bv_page);
1539}
1540EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
1541#endif
1542
1543/**
1544 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1545 * @q : the queue of the device being checked
1546 *
1547 * Description:
1548 * Check if underlying low-level drivers of a device are busy.
1549 * If the drivers want to export their busy state, they must set own
1550 * exporting function using blk_queue_lld_busy() first.
1551 *
1552 * Basically, this function is used only by request stacking drivers
1553 * to stop dispatching requests to underlying devices when underlying
1554 * devices are busy. This behavior helps more I/O merging on the queue
1555 * of the request stacking driver and prevents I/O throughput regression
1556 * on burst I/O load.
1557 *
1558 * Return:
1559 * 0 - Not busy (The request stacking driver should dispatch request)
1560 * 1 - Busy (The request stacking driver should stop dispatching request)
1561 */
1562int blk_lld_busy(struct request_queue *q)
1563{
1564 if (queue_is_mq(q) && q->mq_ops->busy)
1565 return q->mq_ops->busy(q);
1566
1567 return 0;
1568}
1569EXPORT_SYMBOL_GPL(blk_lld_busy);
1570
1571/**
1572 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
1573 * @rq: the clone request to be cleaned up
1574 *
1575 * Description:
1576 * Free all bios in @rq for a cloned request.
1577 */
1578void blk_rq_unprep_clone(struct request *rq)
1579{
1580 struct bio *bio;
1581
1582 while ((bio = rq->bio) != NULL) {
1583 rq->bio = bio->bi_next;
1584
1585 bio_put(bio);
1586 }
1587}
1588EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
1589
1590/*
1591 * Copy attributes of the original request to the clone request.
1592 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
1593 */
1594static void __blk_rq_prep_clone(struct request *dst, struct request *src)
1595{
1596 dst->__sector = blk_rq_pos(src);
1597 dst->__data_len = blk_rq_bytes(src);
1598 if (src->rq_flags & RQF_SPECIAL_PAYLOAD) {
1599 dst->rq_flags |= RQF_SPECIAL_PAYLOAD;
1600 dst->special_vec = src->special_vec;
1601 }
1602 dst->nr_phys_segments = src->nr_phys_segments;
1603 dst->ioprio = src->ioprio;
1604 dst->extra_len = src->extra_len;
1605}
1606
1607/**
1608 * blk_rq_prep_clone - Helper function to setup clone request
1609 * @rq: the request to be setup
1610 * @rq_src: original request to be cloned
1611 * @bs: bio_set that bios for clone are allocated from
1612 * @gfp_mask: memory allocation mask for bio
1613 * @bio_ctr: setup function to be called for each clone bio.
1614 * Returns %0 for success, non %0 for failure.
1615 * @data: private data to be passed to @bio_ctr
1616 *
1617 * Description:
1618 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
1619 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
1620 * are not copied, and copying such parts is the caller's responsibility.
1621 * Also, pages which the original bios are pointing to are not copied
1622 * and the cloned bios just point same pages.
1623 * So cloned bios must be completed before original bios, which means
1624 * the caller must complete @rq before @rq_src.
1625 */
1626int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
1627 struct bio_set *bs, gfp_t gfp_mask,
1628 int (*bio_ctr)(struct bio *, struct bio *, void *),
1629 void *data)
1630{
1631 struct bio *bio, *bio_src;
1632
1633 if (!bs)
1634 bs = &fs_bio_set;
1635
1636 __rq_for_each_bio(bio_src, rq_src) {
1637 bio = bio_clone_fast(bio_src, gfp_mask, bs);
1638 if (!bio)
1639 goto free_and_out;
1640
1641 if (bio_ctr && bio_ctr(bio, bio_src, data))
1642 goto free_and_out;
1643
1644 if (rq->bio) {
1645 rq->biotail->bi_next = bio;
1646 rq->biotail = bio;
1647 } else
1648 rq->bio = rq->biotail = bio;
1649 }
1650
1651 __blk_rq_prep_clone(rq, rq_src);
1652
1653 return 0;
1654
1655free_and_out:
1656 if (bio)
1657 bio_put(bio);
1658 blk_rq_unprep_clone(rq);
1659
1660 return -ENOMEM;
1661}
1662EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
1663
1664int kblockd_schedule_work(struct work_struct *work)
1665{
1666 return queue_work(kblockd_workqueue, work);
1667}
1668EXPORT_SYMBOL(kblockd_schedule_work);
1669
1670int kblockd_schedule_work_on(int cpu, struct work_struct *work)
1671{
1672 return queue_work_on(cpu, kblockd_workqueue, work);
1673}
1674EXPORT_SYMBOL(kblockd_schedule_work_on);
1675
1676int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
1677 unsigned long delay)
1678{
1679 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
1680}
1681EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
1682
1683/**
1684 * blk_start_plug - initialize blk_plug and track it inside the task_struct
1685 * @plug: The &struct blk_plug that needs to be initialized
1686 *
1687 * Description:
1688 * blk_start_plug() indicates to the block layer an intent by the caller
1689 * to submit multiple I/O requests in a batch. The block layer may use
1690 * this hint to defer submitting I/Os from the caller until blk_finish_plug()
1691 * is called. However, the block layer may choose to submit requests
1692 * before a call to blk_finish_plug() if the number of queued I/Os
1693 * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1694 * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if
1695 * the task schedules (see below).
1696 *
1697 * Tracking blk_plug inside the task_struct will help with auto-flushing the
1698 * pending I/O should the task end up blocking between blk_start_plug() and
1699 * blk_finish_plug(). This is important from a performance perspective, but
1700 * also ensures that we don't deadlock. For instance, if the task is blocking
1701 * for a memory allocation, memory reclaim could end up wanting to free a
1702 * page belonging to that request that is currently residing in our private
1703 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
1704 * this kind of deadlock.
1705 */
1706void blk_start_plug(struct blk_plug *plug)
1707{
1708 struct task_struct *tsk = current;
1709
1710 /*
1711 * If this is a nested plug, don't actually assign it.
1712 */
1713 if (tsk->plug)
1714 return;
1715
1716 INIT_LIST_HEAD(&plug->mq_list);
1717 INIT_LIST_HEAD(&plug->cb_list);
1718 plug->rq_count = 0;
1719 plug->multiple_queues = false;
1720
1721 /*
1722 * Store ordering should not be needed here, since a potential
1723 * preempt will imply a full memory barrier
1724 */
1725 tsk->plug = plug;
1726}
1727EXPORT_SYMBOL(blk_start_plug);
1728
1729static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
1730{
1731 LIST_HEAD(callbacks);
1732
1733 while (!list_empty(&plug->cb_list)) {
1734 list_splice_init(&plug->cb_list, &callbacks);
1735
1736 while (!list_empty(&callbacks)) {
1737 struct blk_plug_cb *cb = list_first_entry(&callbacks,
1738 struct blk_plug_cb,
1739 list);
1740 list_del(&cb->list);
1741 cb->callback(cb, from_schedule);
1742 }
1743 }
1744}
1745
1746struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
1747 int size)
1748{
1749 struct blk_plug *plug = current->plug;
1750 struct blk_plug_cb *cb;
1751
1752 if (!plug)
1753 return NULL;
1754
1755 list_for_each_entry(cb, &plug->cb_list, list)
1756 if (cb->callback == unplug && cb->data == data)
1757 return cb;
1758
1759 /* Not currently on the callback list */
1760 BUG_ON(size < sizeof(*cb));
1761 cb = kzalloc(size, GFP_ATOMIC);
1762 if (cb) {
1763 cb->data = data;
1764 cb->callback = unplug;
1765 list_add(&cb->list, &plug->cb_list);
1766 }
1767 return cb;
1768}
1769EXPORT_SYMBOL(blk_check_plugged);
1770
1771void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
1772{
1773 flush_plug_callbacks(plug, from_schedule);
1774
1775 if (!list_empty(&plug->mq_list))
1776 blk_mq_flush_plug_list(plug, from_schedule);
1777}
1778
1779/**
1780 * blk_finish_plug - mark the end of a batch of submitted I/O
1781 * @plug: The &struct blk_plug passed to blk_start_plug()
1782 *
1783 * Description:
1784 * Indicate that a batch of I/O submissions is complete. This function
1785 * must be paired with an initial call to blk_start_plug(). The intent
1786 * is to allow the block layer to optimize I/O submission. See the
1787 * documentation for blk_start_plug() for more information.
1788 */
1789void blk_finish_plug(struct blk_plug *plug)
1790{
1791 if (plug != current->plug)
1792 return;
1793 blk_flush_plug_list(plug, false);
1794
1795 current->plug = NULL;
1796}
1797EXPORT_SYMBOL(blk_finish_plug);
1798
1799int __init blk_dev_init(void)
1800{
1801 BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
1802 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1803 FIELD_SIZEOF(struct request, cmd_flags));
1804 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1805 FIELD_SIZEOF(struct bio, bi_opf));
1806
1807 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
1808 kblockd_workqueue = alloc_workqueue("kblockd",
1809 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1810 if (!kblockd_workqueue)
1811 panic("Failed to create kblockd\n");
1812
1813 blk_requestq_cachep = kmem_cache_create("request_queue",
1814 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
1815
1816#ifdef CONFIG_DEBUG_FS
1817 blk_debugfs_root = debugfs_create_dir("block", NULL);
1818#endif
1819
1820 return 0;
1821}
1822