Loading...
1/*
2 * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
4 *
5 * This file is released under the GPL.
6 */
7
8#include "dm-core.h"
9#include "dm-rq.h"
10#include "dm-uevent.h"
11
12#include <linux/init.h>
13#include <linux/module.h>
14#include <linux/mutex.h>
15#include <linux/sched/mm.h>
16#include <linux/sched/signal.h>
17#include <linux/blkpg.h>
18#include <linux/bio.h>
19#include <linux/mempool.h>
20#include <linux/dax.h>
21#include <linux/slab.h>
22#include <linux/idr.h>
23#include <linux/uio.h>
24#include <linux/hdreg.h>
25#include <linux/delay.h>
26#include <linux/wait.h>
27#include <linux/pr.h>
28#include <linux/refcount.h>
29#include <linux/part_stat.h>
30#include <linux/blk-crypto.h>
31
32#define DM_MSG_PREFIX "core"
33
34/*
35 * Cookies are numeric values sent with CHANGE and REMOVE
36 * uevents while resuming, removing or renaming the device.
37 */
38#define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
39#define DM_COOKIE_LENGTH 24
40
41static const char *_name = DM_NAME;
42
43static unsigned int major = 0;
44static unsigned int _major = 0;
45
46static DEFINE_IDR(_minor_idr);
47
48static DEFINE_SPINLOCK(_minor_lock);
49
50static void do_deferred_remove(struct work_struct *w);
51
52static DECLARE_WORK(deferred_remove_work, do_deferred_remove);
53
54static struct workqueue_struct *deferred_remove_workqueue;
55
56atomic_t dm_global_event_nr = ATOMIC_INIT(0);
57DECLARE_WAIT_QUEUE_HEAD(dm_global_eventq);
58
59void dm_issue_global_event(void)
60{
61 atomic_inc(&dm_global_event_nr);
62 wake_up(&dm_global_eventq);
63}
64
65/*
66 * One of these is allocated (on-stack) per original bio.
67 */
68struct clone_info {
69 struct dm_table *map;
70 struct bio *bio;
71 struct dm_io *io;
72 sector_t sector;
73 unsigned sector_count;
74};
75
76/*
77 * One of these is allocated per clone bio.
78 */
79#define DM_TIO_MAGIC 7282014
80struct dm_target_io {
81 unsigned magic;
82 struct dm_io *io;
83 struct dm_target *ti;
84 unsigned target_bio_nr;
85 unsigned *len_ptr;
86 bool inside_dm_io;
87 struct bio clone;
88};
89
90/*
91 * One of these is allocated per original bio.
92 * It contains the first clone used for that original.
93 */
94#define DM_IO_MAGIC 5191977
95struct dm_io {
96 unsigned magic;
97 struct mapped_device *md;
98 blk_status_t status;
99 atomic_t io_count;
100 struct bio *orig_bio;
101 unsigned long start_time;
102 spinlock_t endio_lock;
103 struct dm_stats_aux stats_aux;
104 /* last member of dm_target_io is 'struct bio' */
105 struct dm_target_io tio;
106};
107
108void *dm_per_bio_data(struct bio *bio, size_t data_size)
109{
110 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
111 if (!tio->inside_dm_io)
112 return (char *)bio - offsetof(struct dm_target_io, clone) - data_size;
113 return (char *)bio - offsetof(struct dm_target_io, clone) - offsetof(struct dm_io, tio) - data_size;
114}
115EXPORT_SYMBOL_GPL(dm_per_bio_data);
116
117struct bio *dm_bio_from_per_bio_data(void *data, size_t data_size)
118{
119 struct dm_io *io = (struct dm_io *)((char *)data + data_size);
120 if (io->magic == DM_IO_MAGIC)
121 return (struct bio *)((char *)io + offsetof(struct dm_io, tio) + offsetof(struct dm_target_io, clone));
122 BUG_ON(io->magic != DM_TIO_MAGIC);
123 return (struct bio *)((char *)io + offsetof(struct dm_target_io, clone));
124}
125EXPORT_SYMBOL_GPL(dm_bio_from_per_bio_data);
126
127unsigned dm_bio_get_target_bio_nr(const struct bio *bio)
128{
129 return container_of(bio, struct dm_target_io, clone)->target_bio_nr;
130}
131EXPORT_SYMBOL_GPL(dm_bio_get_target_bio_nr);
132
133#define MINOR_ALLOCED ((void *)-1)
134
135/*
136 * Bits for the md->flags field.
137 */
138#define DMF_BLOCK_IO_FOR_SUSPEND 0
139#define DMF_SUSPENDED 1
140#define DMF_FROZEN 2
141#define DMF_FREEING 3
142#define DMF_DELETING 4
143#define DMF_NOFLUSH_SUSPENDING 5
144#define DMF_DEFERRED_REMOVE 6
145#define DMF_SUSPENDED_INTERNALLY 7
146#define DMF_POST_SUSPENDING 8
147
148#define DM_NUMA_NODE NUMA_NO_NODE
149static int dm_numa_node = DM_NUMA_NODE;
150
151/*
152 * For mempools pre-allocation at the table loading time.
153 */
154struct dm_md_mempools {
155 struct bio_set bs;
156 struct bio_set io_bs;
157};
158
159struct table_device {
160 struct list_head list;
161 refcount_t count;
162 struct dm_dev dm_dev;
163};
164
165/*
166 * Bio-based DM's mempools' reserved IOs set by the user.
167 */
168#define RESERVED_BIO_BASED_IOS 16
169static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
170
171static int __dm_get_module_param_int(int *module_param, int min, int max)
172{
173 int param = READ_ONCE(*module_param);
174 int modified_param = 0;
175 bool modified = true;
176
177 if (param < min)
178 modified_param = min;
179 else if (param > max)
180 modified_param = max;
181 else
182 modified = false;
183
184 if (modified) {
185 (void)cmpxchg(module_param, param, modified_param);
186 param = modified_param;
187 }
188
189 return param;
190}
191
192unsigned __dm_get_module_param(unsigned *module_param,
193 unsigned def, unsigned max)
194{
195 unsigned param = READ_ONCE(*module_param);
196 unsigned modified_param = 0;
197
198 if (!param)
199 modified_param = def;
200 else if (param > max)
201 modified_param = max;
202
203 if (modified_param) {
204 (void)cmpxchg(module_param, param, modified_param);
205 param = modified_param;
206 }
207
208 return param;
209}
210
211unsigned dm_get_reserved_bio_based_ios(void)
212{
213 return __dm_get_module_param(&reserved_bio_based_ios,
214 RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS);
215}
216EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
217
218static unsigned dm_get_numa_node(void)
219{
220 return __dm_get_module_param_int(&dm_numa_node,
221 DM_NUMA_NODE, num_online_nodes() - 1);
222}
223
224static int __init local_init(void)
225{
226 int r;
227
228 r = dm_uevent_init();
229 if (r)
230 return r;
231
232 deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1);
233 if (!deferred_remove_workqueue) {
234 r = -ENOMEM;
235 goto out_uevent_exit;
236 }
237
238 _major = major;
239 r = register_blkdev(_major, _name);
240 if (r < 0)
241 goto out_free_workqueue;
242
243 if (!_major)
244 _major = r;
245
246 return 0;
247
248out_free_workqueue:
249 destroy_workqueue(deferred_remove_workqueue);
250out_uevent_exit:
251 dm_uevent_exit();
252
253 return r;
254}
255
256static void local_exit(void)
257{
258 flush_scheduled_work();
259 destroy_workqueue(deferred_remove_workqueue);
260
261 unregister_blkdev(_major, _name);
262 dm_uevent_exit();
263
264 _major = 0;
265
266 DMINFO("cleaned up");
267}
268
269static int (*_inits[])(void) __initdata = {
270 local_init,
271 dm_target_init,
272 dm_linear_init,
273 dm_stripe_init,
274 dm_io_init,
275 dm_kcopyd_init,
276 dm_interface_init,
277 dm_statistics_init,
278};
279
280static void (*_exits[])(void) = {
281 local_exit,
282 dm_target_exit,
283 dm_linear_exit,
284 dm_stripe_exit,
285 dm_io_exit,
286 dm_kcopyd_exit,
287 dm_interface_exit,
288 dm_statistics_exit,
289};
290
291static int __init dm_init(void)
292{
293 const int count = ARRAY_SIZE(_inits);
294
295 int r, i;
296
297 for (i = 0; i < count; i++) {
298 r = _inits[i]();
299 if (r)
300 goto bad;
301 }
302
303 return 0;
304
305 bad:
306 while (i--)
307 _exits[i]();
308
309 return r;
310}
311
312static void __exit dm_exit(void)
313{
314 int i = ARRAY_SIZE(_exits);
315
316 while (i--)
317 _exits[i]();
318
319 /*
320 * Should be empty by this point.
321 */
322 idr_destroy(&_minor_idr);
323}
324
325/*
326 * Block device functions
327 */
328int dm_deleting_md(struct mapped_device *md)
329{
330 return test_bit(DMF_DELETING, &md->flags);
331}
332
333static int dm_blk_open(struct block_device *bdev, fmode_t mode)
334{
335 struct mapped_device *md;
336
337 spin_lock(&_minor_lock);
338
339 md = bdev->bd_disk->private_data;
340 if (!md)
341 goto out;
342
343 if (test_bit(DMF_FREEING, &md->flags) ||
344 dm_deleting_md(md)) {
345 md = NULL;
346 goto out;
347 }
348
349 dm_get(md);
350 atomic_inc(&md->open_count);
351out:
352 spin_unlock(&_minor_lock);
353
354 return md ? 0 : -ENXIO;
355}
356
357static void dm_blk_close(struct gendisk *disk, fmode_t mode)
358{
359 struct mapped_device *md;
360
361 spin_lock(&_minor_lock);
362
363 md = disk->private_data;
364 if (WARN_ON(!md))
365 goto out;
366
367 if (atomic_dec_and_test(&md->open_count) &&
368 (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
369 queue_work(deferred_remove_workqueue, &deferred_remove_work);
370
371 dm_put(md);
372out:
373 spin_unlock(&_minor_lock);
374}
375
376int dm_open_count(struct mapped_device *md)
377{
378 return atomic_read(&md->open_count);
379}
380
381/*
382 * Guarantees nothing is using the device before it's deleted.
383 */
384int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
385{
386 int r = 0;
387
388 spin_lock(&_minor_lock);
389
390 if (dm_open_count(md)) {
391 r = -EBUSY;
392 if (mark_deferred)
393 set_bit(DMF_DEFERRED_REMOVE, &md->flags);
394 } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
395 r = -EEXIST;
396 else
397 set_bit(DMF_DELETING, &md->flags);
398
399 spin_unlock(&_minor_lock);
400
401 return r;
402}
403
404int dm_cancel_deferred_remove(struct mapped_device *md)
405{
406 int r = 0;
407
408 spin_lock(&_minor_lock);
409
410 if (test_bit(DMF_DELETING, &md->flags))
411 r = -EBUSY;
412 else
413 clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
414
415 spin_unlock(&_minor_lock);
416
417 return r;
418}
419
420static void do_deferred_remove(struct work_struct *w)
421{
422 dm_deferred_remove();
423}
424
425sector_t dm_get_size(struct mapped_device *md)
426{
427 return get_capacity(md->disk);
428}
429
430struct request_queue *dm_get_md_queue(struct mapped_device *md)
431{
432 return md->queue;
433}
434
435struct dm_stats *dm_get_stats(struct mapped_device *md)
436{
437 return &md->stats;
438}
439
440static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
441{
442 struct mapped_device *md = bdev->bd_disk->private_data;
443
444 return dm_get_geometry(md, geo);
445}
446
447#ifdef CONFIG_BLK_DEV_ZONED
448int dm_report_zones_cb(struct blk_zone *zone, unsigned int idx, void *data)
449{
450 struct dm_report_zones_args *args = data;
451 sector_t sector_diff = args->tgt->begin - args->start;
452
453 /*
454 * Ignore zones beyond the target range.
455 */
456 if (zone->start >= args->start + args->tgt->len)
457 return 0;
458
459 /*
460 * Remap the start sector and write pointer position of the zone
461 * to match its position in the target range.
462 */
463 zone->start += sector_diff;
464 if (zone->type != BLK_ZONE_TYPE_CONVENTIONAL) {
465 if (zone->cond == BLK_ZONE_COND_FULL)
466 zone->wp = zone->start + zone->len;
467 else if (zone->cond == BLK_ZONE_COND_EMPTY)
468 zone->wp = zone->start;
469 else
470 zone->wp += sector_diff;
471 }
472
473 args->next_sector = zone->start + zone->len;
474 return args->orig_cb(zone, args->zone_idx++, args->orig_data);
475}
476EXPORT_SYMBOL_GPL(dm_report_zones_cb);
477
478static int dm_blk_report_zones(struct gendisk *disk, sector_t sector,
479 unsigned int nr_zones, report_zones_cb cb, void *data)
480{
481 struct mapped_device *md = disk->private_data;
482 struct dm_table *map;
483 int srcu_idx, ret;
484 struct dm_report_zones_args args = {
485 .next_sector = sector,
486 .orig_data = data,
487 .orig_cb = cb,
488 };
489
490 if (dm_suspended_md(md))
491 return -EAGAIN;
492
493 map = dm_get_live_table(md, &srcu_idx);
494 if (!map)
495 return -EIO;
496
497 do {
498 struct dm_target *tgt;
499
500 tgt = dm_table_find_target(map, args.next_sector);
501 if (WARN_ON_ONCE(!tgt->type->report_zones)) {
502 ret = -EIO;
503 goto out;
504 }
505
506 args.tgt = tgt;
507 ret = tgt->type->report_zones(tgt, &args,
508 nr_zones - args.zone_idx);
509 if (ret < 0)
510 goto out;
511 } while (args.zone_idx < nr_zones &&
512 args.next_sector < get_capacity(disk));
513
514 ret = args.zone_idx;
515out:
516 dm_put_live_table(md, srcu_idx);
517 return ret;
518}
519#else
520#define dm_blk_report_zones NULL
521#endif /* CONFIG_BLK_DEV_ZONED */
522
523static int dm_prepare_ioctl(struct mapped_device *md, int *srcu_idx,
524 struct block_device **bdev)
525 __acquires(md->io_barrier)
526{
527 struct dm_target *tgt;
528 struct dm_table *map;
529 int r;
530
531retry:
532 r = -ENOTTY;
533 map = dm_get_live_table(md, srcu_idx);
534 if (!map || !dm_table_get_size(map))
535 return r;
536
537 /* We only support devices that have a single target */
538 if (dm_table_get_num_targets(map) != 1)
539 return r;
540
541 tgt = dm_table_get_target(map, 0);
542 if (!tgt->type->prepare_ioctl)
543 return r;
544
545 if (dm_suspended_md(md))
546 return -EAGAIN;
547
548 r = tgt->type->prepare_ioctl(tgt, bdev);
549 if (r == -ENOTCONN && !fatal_signal_pending(current)) {
550 dm_put_live_table(md, *srcu_idx);
551 msleep(10);
552 goto retry;
553 }
554
555 return r;
556}
557
558static void dm_unprepare_ioctl(struct mapped_device *md, int srcu_idx)
559 __releases(md->io_barrier)
560{
561 dm_put_live_table(md, srcu_idx);
562}
563
564static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
565 unsigned int cmd, unsigned long arg)
566{
567 struct mapped_device *md = bdev->bd_disk->private_data;
568 int r, srcu_idx;
569
570 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
571 if (r < 0)
572 goto out;
573
574 if (r > 0) {
575 /*
576 * Target determined this ioctl is being issued against a
577 * subset of the parent bdev; require extra privileges.
578 */
579 if (!capable(CAP_SYS_RAWIO)) {
580 DMWARN_LIMIT(
581 "%s: sending ioctl %x to DM device without required privilege.",
582 current->comm, cmd);
583 r = -ENOIOCTLCMD;
584 goto out;
585 }
586 }
587
588 r = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
589out:
590 dm_unprepare_ioctl(md, srcu_idx);
591 return r;
592}
593
594static void start_io_acct(struct dm_io *io);
595
596static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio)
597{
598 struct dm_io *io;
599 struct dm_target_io *tio;
600 struct bio *clone;
601
602 clone = bio_alloc_bioset(GFP_NOIO, 0, &md->io_bs);
603 if (!clone)
604 return NULL;
605
606 tio = container_of(clone, struct dm_target_io, clone);
607 tio->inside_dm_io = true;
608 tio->io = NULL;
609
610 io = container_of(tio, struct dm_io, tio);
611 io->magic = DM_IO_MAGIC;
612 io->status = 0;
613 atomic_set(&io->io_count, 1);
614 io->orig_bio = bio;
615 io->md = md;
616 spin_lock_init(&io->endio_lock);
617
618 start_io_acct(io);
619
620 return io;
621}
622
623static void free_io(struct mapped_device *md, struct dm_io *io)
624{
625 bio_put(&io->tio.clone);
626}
627
628static struct dm_target_io *alloc_tio(struct clone_info *ci, struct dm_target *ti,
629 unsigned target_bio_nr, gfp_t gfp_mask)
630{
631 struct dm_target_io *tio;
632
633 if (!ci->io->tio.io) {
634 /* the dm_target_io embedded in ci->io is available */
635 tio = &ci->io->tio;
636 } else {
637 struct bio *clone = bio_alloc_bioset(gfp_mask, 0, &ci->io->md->bs);
638 if (!clone)
639 return NULL;
640
641 tio = container_of(clone, struct dm_target_io, clone);
642 tio->inside_dm_io = false;
643 }
644
645 tio->magic = DM_TIO_MAGIC;
646 tio->io = ci->io;
647 tio->ti = ti;
648 tio->target_bio_nr = target_bio_nr;
649
650 return tio;
651}
652
653static void free_tio(struct dm_target_io *tio)
654{
655 if (tio->inside_dm_io)
656 return;
657 bio_put(&tio->clone);
658}
659
660u64 dm_start_time_ns_from_clone(struct bio *bio)
661{
662 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
663 struct dm_io *io = tio->io;
664
665 return jiffies_to_nsecs(io->start_time);
666}
667EXPORT_SYMBOL_GPL(dm_start_time_ns_from_clone);
668
669static void start_io_acct(struct dm_io *io)
670{
671 struct mapped_device *md = io->md;
672 struct bio *bio = io->orig_bio;
673
674 io->start_time = bio_start_io_acct(bio);
675 if (unlikely(dm_stats_used(&md->stats)))
676 dm_stats_account_io(&md->stats, bio_data_dir(bio),
677 bio->bi_iter.bi_sector, bio_sectors(bio),
678 false, 0, &io->stats_aux);
679}
680
681static void end_io_acct(struct dm_io *io)
682{
683 struct mapped_device *md = io->md;
684 struct bio *bio = io->orig_bio;
685 unsigned long duration = jiffies - io->start_time;
686
687 bio_end_io_acct(bio, io->start_time);
688
689 if (unlikely(dm_stats_used(&md->stats)))
690 dm_stats_account_io(&md->stats, bio_data_dir(bio),
691 bio->bi_iter.bi_sector, bio_sectors(bio),
692 true, duration, &io->stats_aux);
693
694 /* nudge anyone waiting on suspend queue */
695 if (unlikely(wq_has_sleeper(&md->wait)))
696 wake_up(&md->wait);
697}
698
699/*
700 * Add the bio to the list of deferred io.
701 */
702static void queue_io(struct mapped_device *md, struct bio *bio)
703{
704 unsigned long flags;
705
706 spin_lock_irqsave(&md->deferred_lock, flags);
707 bio_list_add(&md->deferred, bio);
708 spin_unlock_irqrestore(&md->deferred_lock, flags);
709 queue_work(md->wq, &md->work);
710}
711
712/*
713 * Everyone (including functions in this file), should use this
714 * function to access the md->map field, and make sure they call
715 * dm_put_live_table() when finished.
716 */
717struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
718{
719 *srcu_idx = srcu_read_lock(&md->io_barrier);
720
721 return srcu_dereference(md->map, &md->io_barrier);
722}
723
724void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
725{
726 srcu_read_unlock(&md->io_barrier, srcu_idx);
727}
728
729void dm_sync_table(struct mapped_device *md)
730{
731 synchronize_srcu(&md->io_barrier);
732 synchronize_rcu_expedited();
733}
734
735/*
736 * A fast alternative to dm_get_live_table/dm_put_live_table.
737 * The caller must not block between these two functions.
738 */
739static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
740{
741 rcu_read_lock();
742 return rcu_dereference(md->map);
743}
744
745static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
746{
747 rcu_read_unlock();
748}
749
750static char *_dm_claim_ptr = "I belong to device-mapper";
751
752/*
753 * Open a table device so we can use it as a map destination.
754 */
755static int open_table_device(struct table_device *td, dev_t dev,
756 struct mapped_device *md)
757{
758 struct block_device *bdev;
759
760 int r;
761
762 BUG_ON(td->dm_dev.bdev);
763
764 bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _dm_claim_ptr);
765 if (IS_ERR(bdev))
766 return PTR_ERR(bdev);
767
768 r = bd_link_disk_holder(bdev, dm_disk(md));
769 if (r) {
770 blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
771 return r;
772 }
773
774 td->dm_dev.bdev = bdev;
775 td->dm_dev.dax_dev = dax_get_by_host(bdev->bd_disk->disk_name);
776 return 0;
777}
778
779/*
780 * Close a table device that we've been using.
781 */
782static void close_table_device(struct table_device *td, struct mapped_device *md)
783{
784 if (!td->dm_dev.bdev)
785 return;
786
787 bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
788 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
789 put_dax(td->dm_dev.dax_dev);
790 td->dm_dev.bdev = NULL;
791 td->dm_dev.dax_dev = NULL;
792}
793
794static struct table_device *find_table_device(struct list_head *l, dev_t dev,
795 fmode_t mode)
796{
797 struct table_device *td;
798
799 list_for_each_entry(td, l, list)
800 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
801 return td;
802
803 return NULL;
804}
805
806int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
807 struct dm_dev **result)
808{
809 int r;
810 struct table_device *td;
811
812 mutex_lock(&md->table_devices_lock);
813 td = find_table_device(&md->table_devices, dev, mode);
814 if (!td) {
815 td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
816 if (!td) {
817 mutex_unlock(&md->table_devices_lock);
818 return -ENOMEM;
819 }
820
821 td->dm_dev.mode = mode;
822 td->dm_dev.bdev = NULL;
823
824 if ((r = open_table_device(td, dev, md))) {
825 mutex_unlock(&md->table_devices_lock);
826 kfree(td);
827 return r;
828 }
829
830 format_dev_t(td->dm_dev.name, dev);
831
832 refcount_set(&td->count, 1);
833 list_add(&td->list, &md->table_devices);
834 } else {
835 refcount_inc(&td->count);
836 }
837 mutex_unlock(&md->table_devices_lock);
838
839 *result = &td->dm_dev;
840 return 0;
841}
842EXPORT_SYMBOL_GPL(dm_get_table_device);
843
844void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
845{
846 struct table_device *td = container_of(d, struct table_device, dm_dev);
847
848 mutex_lock(&md->table_devices_lock);
849 if (refcount_dec_and_test(&td->count)) {
850 close_table_device(td, md);
851 list_del(&td->list);
852 kfree(td);
853 }
854 mutex_unlock(&md->table_devices_lock);
855}
856EXPORT_SYMBOL(dm_put_table_device);
857
858static void free_table_devices(struct list_head *devices)
859{
860 struct list_head *tmp, *next;
861
862 list_for_each_safe(tmp, next, devices) {
863 struct table_device *td = list_entry(tmp, struct table_device, list);
864
865 DMWARN("dm_destroy: %s still exists with %d references",
866 td->dm_dev.name, refcount_read(&td->count));
867 kfree(td);
868 }
869}
870
871/*
872 * Get the geometry associated with a dm device
873 */
874int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
875{
876 *geo = md->geometry;
877
878 return 0;
879}
880
881/*
882 * Set the geometry of a device.
883 */
884int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
885{
886 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
887
888 if (geo->start > sz) {
889 DMWARN("Start sector is beyond the geometry limits.");
890 return -EINVAL;
891 }
892
893 md->geometry = *geo;
894
895 return 0;
896}
897
898static int __noflush_suspending(struct mapped_device *md)
899{
900 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
901}
902
903/*
904 * Decrements the number of outstanding ios that a bio has been
905 * cloned into, completing the original io if necc.
906 */
907static void dec_pending(struct dm_io *io, blk_status_t error)
908{
909 unsigned long flags;
910 blk_status_t io_error;
911 struct bio *bio;
912 struct mapped_device *md = io->md;
913
914 /* Push-back supersedes any I/O errors */
915 if (unlikely(error)) {
916 spin_lock_irqsave(&io->endio_lock, flags);
917 if (!(io->status == BLK_STS_DM_REQUEUE && __noflush_suspending(md)))
918 io->status = error;
919 spin_unlock_irqrestore(&io->endio_lock, flags);
920 }
921
922 if (atomic_dec_and_test(&io->io_count)) {
923 if (io->status == BLK_STS_DM_REQUEUE) {
924 /*
925 * Target requested pushing back the I/O.
926 */
927 spin_lock_irqsave(&md->deferred_lock, flags);
928 if (__noflush_suspending(md))
929 /* NOTE early return due to BLK_STS_DM_REQUEUE below */
930 bio_list_add_head(&md->deferred, io->orig_bio);
931 else
932 /* noflush suspend was interrupted. */
933 io->status = BLK_STS_IOERR;
934 spin_unlock_irqrestore(&md->deferred_lock, flags);
935 }
936
937 io_error = io->status;
938 bio = io->orig_bio;
939 end_io_acct(io);
940 free_io(md, io);
941
942 if (io_error == BLK_STS_DM_REQUEUE)
943 return;
944
945 if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) {
946 /*
947 * Preflush done for flush with data, reissue
948 * without REQ_PREFLUSH.
949 */
950 bio->bi_opf &= ~REQ_PREFLUSH;
951 queue_io(md, bio);
952 } else {
953 /* done with normal IO or empty flush */
954 if (io_error)
955 bio->bi_status = io_error;
956 bio_endio(bio);
957 }
958 }
959}
960
961void disable_discard(struct mapped_device *md)
962{
963 struct queue_limits *limits = dm_get_queue_limits(md);
964
965 /* device doesn't really support DISCARD, disable it */
966 limits->max_discard_sectors = 0;
967 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, md->queue);
968}
969
970void disable_write_same(struct mapped_device *md)
971{
972 struct queue_limits *limits = dm_get_queue_limits(md);
973
974 /* device doesn't really support WRITE SAME, disable it */
975 limits->max_write_same_sectors = 0;
976}
977
978void disable_write_zeroes(struct mapped_device *md)
979{
980 struct queue_limits *limits = dm_get_queue_limits(md);
981
982 /* device doesn't really support WRITE ZEROES, disable it */
983 limits->max_write_zeroes_sectors = 0;
984}
985
986static void clone_endio(struct bio *bio)
987{
988 blk_status_t error = bio->bi_status;
989 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
990 struct dm_io *io = tio->io;
991 struct mapped_device *md = tio->io->md;
992 dm_endio_fn endio = tio->ti->type->end_io;
993 struct bio *orig_bio = io->orig_bio;
994
995 if (unlikely(error == BLK_STS_TARGET) && md->type != DM_TYPE_NVME_BIO_BASED) {
996 if (bio_op(bio) == REQ_OP_DISCARD &&
997 !bio->bi_disk->queue->limits.max_discard_sectors)
998 disable_discard(md);
999 else if (bio_op(bio) == REQ_OP_WRITE_SAME &&
1000 !bio->bi_disk->queue->limits.max_write_same_sectors)
1001 disable_write_same(md);
1002 else if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
1003 !bio->bi_disk->queue->limits.max_write_zeroes_sectors)
1004 disable_write_zeroes(md);
1005 }
1006
1007 /*
1008 * For zone-append bios get offset in zone of the written
1009 * sector and add that to the original bio sector pos.
1010 */
1011 if (bio_op(orig_bio) == REQ_OP_ZONE_APPEND) {
1012 sector_t written_sector = bio->bi_iter.bi_sector;
1013 struct request_queue *q = orig_bio->bi_disk->queue;
1014 u64 mask = (u64)blk_queue_zone_sectors(q) - 1;
1015
1016 orig_bio->bi_iter.bi_sector += written_sector & mask;
1017 }
1018
1019 if (endio) {
1020 int r = endio(tio->ti, bio, &error);
1021 switch (r) {
1022 case DM_ENDIO_REQUEUE:
1023 error = BLK_STS_DM_REQUEUE;
1024 fallthrough;
1025 case DM_ENDIO_DONE:
1026 break;
1027 case DM_ENDIO_INCOMPLETE:
1028 /* The target will handle the io */
1029 return;
1030 default:
1031 DMWARN("unimplemented target endio return value: %d", r);
1032 BUG();
1033 }
1034 }
1035
1036 free_tio(tio);
1037 dec_pending(io, error);
1038}
1039
1040/*
1041 * Return maximum size of I/O possible at the supplied sector up to the current
1042 * target boundary.
1043 */
1044static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
1045{
1046 sector_t target_offset = dm_target_offset(ti, sector);
1047
1048 return ti->len - target_offset;
1049}
1050
1051static sector_t max_io_len(sector_t sector, struct dm_target *ti)
1052{
1053 sector_t len = max_io_len_target_boundary(sector, ti);
1054 sector_t offset, max_len;
1055
1056 /*
1057 * Does the target need to split even further?
1058 */
1059 if (ti->max_io_len) {
1060 offset = dm_target_offset(ti, sector);
1061 if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
1062 max_len = sector_div(offset, ti->max_io_len);
1063 else
1064 max_len = offset & (ti->max_io_len - 1);
1065 max_len = ti->max_io_len - max_len;
1066
1067 if (len > max_len)
1068 len = max_len;
1069 }
1070
1071 return len;
1072}
1073
1074int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1075{
1076 if (len > UINT_MAX) {
1077 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1078 (unsigned long long)len, UINT_MAX);
1079 ti->error = "Maximum size of target IO is too large";
1080 return -EINVAL;
1081 }
1082
1083 ti->max_io_len = (uint32_t) len;
1084
1085 return 0;
1086}
1087EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1088
1089static struct dm_target *dm_dax_get_live_target(struct mapped_device *md,
1090 sector_t sector, int *srcu_idx)
1091 __acquires(md->io_barrier)
1092{
1093 struct dm_table *map;
1094 struct dm_target *ti;
1095
1096 map = dm_get_live_table(md, srcu_idx);
1097 if (!map)
1098 return NULL;
1099
1100 ti = dm_table_find_target(map, sector);
1101 if (!ti)
1102 return NULL;
1103
1104 return ti;
1105}
1106
1107static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
1108 long nr_pages, void **kaddr, pfn_t *pfn)
1109{
1110 struct mapped_device *md = dax_get_private(dax_dev);
1111 sector_t sector = pgoff * PAGE_SECTORS;
1112 struct dm_target *ti;
1113 long len, ret = -EIO;
1114 int srcu_idx;
1115
1116 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1117
1118 if (!ti)
1119 goto out;
1120 if (!ti->type->direct_access)
1121 goto out;
1122 len = max_io_len(sector, ti) / PAGE_SECTORS;
1123 if (len < 1)
1124 goto out;
1125 nr_pages = min(len, nr_pages);
1126 ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn);
1127
1128 out:
1129 dm_put_live_table(md, srcu_idx);
1130
1131 return ret;
1132}
1133
1134static bool dm_dax_supported(struct dax_device *dax_dev, struct block_device *bdev,
1135 int blocksize, sector_t start, sector_t len)
1136{
1137 struct mapped_device *md = dax_get_private(dax_dev);
1138 struct dm_table *map;
1139 bool ret = false;
1140 int srcu_idx;
1141
1142 map = dm_get_live_table(md, &srcu_idx);
1143 if (!map)
1144 goto out;
1145
1146 ret = dm_table_supports_dax(map, device_supports_dax, &blocksize);
1147
1148out:
1149 dm_put_live_table(md, srcu_idx);
1150
1151 return ret;
1152}
1153
1154static size_t dm_dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1155 void *addr, size_t bytes, struct iov_iter *i)
1156{
1157 struct mapped_device *md = dax_get_private(dax_dev);
1158 sector_t sector = pgoff * PAGE_SECTORS;
1159 struct dm_target *ti;
1160 long ret = 0;
1161 int srcu_idx;
1162
1163 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1164
1165 if (!ti)
1166 goto out;
1167 if (!ti->type->dax_copy_from_iter) {
1168 ret = copy_from_iter(addr, bytes, i);
1169 goto out;
1170 }
1171 ret = ti->type->dax_copy_from_iter(ti, pgoff, addr, bytes, i);
1172 out:
1173 dm_put_live_table(md, srcu_idx);
1174
1175 return ret;
1176}
1177
1178static size_t dm_dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1179 void *addr, size_t bytes, struct iov_iter *i)
1180{
1181 struct mapped_device *md = dax_get_private(dax_dev);
1182 sector_t sector = pgoff * PAGE_SECTORS;
1183 struct dm_target *ti;
1184 long ret = 0;
1185 int srcu_idx;
1186
1187 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1188
1189 if (!ti)
1190 goto out;
1191 if (!ti->type->dax_copy_to_iter) {
1192 ret = copy_to_iter(addr, bytes, i);
1193 goto out;
1194 }
1195 ret = ti->type->dax_copy_to_iter(ti, pgoff, addr, bytes, i);
1196 out:
1197 dm_put_live_table(md, srcu_idx);
1198
1199 return ret;
1200}
1201
1202static int dm_dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff,
1203 size_t nr_pages)
1204{
1205 struct mapped_device *md = dax_get_private(dax_dev);
1206 sector_t sector = pgoff * PAGE_SECTORS;
1207 struct dm_target *ti;
1208 int ret = -EIO;
1209 int srcu_idx;
1210
1211 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1212
1213 if (!ti)
1214 goto out;
1215 if (WARN_ON(!ti->type->dax_zero_page_range)) {
1216 /*
1217 * ->zero_page_range() is mandatory dax operation. If we are
1218 * here, something is wrong.
1219 */
1220 dm_put_live_table(md, srcu_idx);
1221 goto out;
1222 }
1223 ret = ti->type->dax_zero_page_range(ti, pgoff, nr_pages);
1224
1225 out:
1226 dm_put_live_table(md, srcu_idx);
1227
1228 return ret;
1229}
1230
1231/*
1232 * A target may call dm_accept_partial_bio only from the map routine. It is
1233 * allowed for all bio types except REQ_PREFLUSH, REQ_OP_ZONE_RESET,
1234 * REQ_OP_ZONE_OPEN, REQ_OP_ZONE_CLOSE and REQ_OP_ZONE_FINISH.
1235 *
1236 * dm_accept_partial_bio informs the dm that the target only wants to process
1237 * additional n_sectors sectors of the bio and the rest of the data should be
1238 * sent in a next bio.
1239 *
1240 * A diagram that explains the arithmetics:
1241 * +--------------------+---------------+-------+
1242 * | 1 | 2 | 3 |
1243 * +--------------------+---------------+-------+
1244 *
1245 * <-------------- *tio->len_ptr --------------->
1246 * <------- bi_size ------->
1247 * <-- n_sectors -->
1248 *
1249 * Region 1 was already iterated over with bio_advance or similar function.
1250 * (it may be empty if the target doesn't use bio_advance)
1251 * Region 2 is the remaining bio size that the target wants to process.
1252 * (it may be empty if region 1 is non-empty, although there is no reason
1253 * to make it empty)
1254 * The target requires that region 3 is to be sent in the next bio.
1255 *
1256 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1257 * the partially processed part (the sum of regions 1+2) must be the same for all
1258 * copies of the bio.
1259 */
1260void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1261{
1262 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1263 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1264 BUG_ON(bio->bi_opf & REQ_PREFLUSH);
1265 BUG_ON(bi_size > *tio->len_ptr);
1266 BUG_ON(n_sectors > bi_size);
1267 *tio->len_ptr -= bi_size - n_sectors;
1268 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1269}
1270EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1271
1272static blk_qc_t __map_bio(struct dm_target_io *tio)
1273{
1274 int r;
1275 sector_t sector;
1276 struct bio *clone = &tio->clone;
1277 struct dm_io *io = tio->io;
1278 struct dm_target *ti = tio->ti;
1279 blk_qc_t ret = BLK_QC_T_NONE;
1280
1281 clone->bi_end_io = clone_endio;
1282
1283 /*
1284 * Map the clone. If r == 0 we don't need to do
1285 * anything, the target has assumed ownership of
1286 * this io.
1287 */
1288 atomic_inc(&io->io_count);
1289 sector = clone->bi_iter.bi_sector;
1290
1291 r = ti->type->map(ti, clone);
1292 switch (r) {
1293 case DM_MAPIO_SUBMITTED:
1294 break;
1295 case DM_MAPIO_REMAPPED:
1296 /* the bio has been remapped so dispatch it */
1297 trace_block_bio_remap(clone->bi_disk->queue, clone,
1298 bio_dev(io->orig_bio), sector);
1299 ret = submit_bio_noacct(clone);
1300 break;
1301 case DM_MAPIO_KILL:
1302 free_tio(tio);
1303 dec_pending(io, BLK_STS_IOERR);
1304 break;
1305 case DM_MAPIO_REQUEUE:
1306 free_tio(tio);
1307 dec_pending(io, BLK_STS_DM_REQUEUE);
1308 break;
1309 default:
1310 DMWARN("unimplemented target map return value: %d", r);
1311 BUG();
1312 }
1313
1314 return ret;
1315}
1316
1317static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1318{
1319 bio->bi_iter.bi_sector = sector;
1320 bio->bi_iter.bi_size = to_bytes(len);
1321}
1322
1323/*
1324 * Creates a bio that consists of range of complete bvecs.
1325 */
1326static int clone_bio(struct dm_target_io *tio, struct bio *bio,
1327 sector_t sector, unsigned len)
1328{
1329 struct bio *clone = &tio->clone;
1330
1331 __bio_clone_fast(clone, bio);
1332
1333 bio_crypt_clone(clone, bio, GFP_NOIO);
1334
1335 if (bio_integrity(bio)) {
1336 int r;
1337
1338 if (unlikely(!dm_target_has_integrity(tio->ti->type) &&
1339 !dm_target_passes_integrity(tio->ti->type))) {
1340 DMWARN("%s: the target %s doesn't support integrity data.",
1341 dm_device_name(tio->io->md),
1342 tio->ti->type->name);
1343 return -EIO;
1344 }
1345
1346 r = bio_integrity_clone(clone, bio, GFP_NOIO);
1347 if (r < 0)
1348 return r;
1349 }
1350
1351 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1352 clone->bi_iter.bi_size = to_bytes(len);
1353
1354 if (bio_integrity(bio))
1355 bio_integrity_trim(clone);
1356
1357 return 0;
1358}
1359
1360static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci,
1361 struct dm_target *ti, unsigned num_bios)
1362{
1363 struct dm_target_io *tio;
1364 int try;
1365
1366 if (!num_bios)
1367 return;
1368
1369 if (num_bios == 1) {
1370 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1371 bio_list_add(blist, &tio->clone);
1372 return;
1373 }
1374
1375 for (try = 0; try < 2; try++) {
1376 int bio_nr;
1377 struct bio *bio;
1378
1379 if (try)
1380 mutex_lock(&ci->io->md->table_devices_lock);
1381 for (bio_nr = 0; bio_nr < num_bios; bio_nr++) {
1382 tio = alloc_tio(ci, ti, bio_nr, try ? GFP_NOIO : GFP_NOWAIT);
1383 if (!tio)
1384 break;
1385
1386 bio_list_add(blist, &tio->clone);
1387 }
1388 if (try)
1389 mutex_unlock(&ci->io->md->table_devices_lock);
1390 if (bio_nr == num_bios)
1391 return;
1392
1393 while ((bio = bio_list_pop(blist))) {
1394 tio = container_of(bio, struct dm_target_io, clone);
1395 free_tio(tio);
1396 }
1397 }
1398}
1399
1400static blk_qc_t __clone_and_map_simple_bio(struct clone_info *ci,
1401 struct dm_target_io *tio, unsigned *len)
1402{
1403 struct bio *clone = &tio->clone;
1404
1405 tio->len_ptr = len;
1406
1407 __bio_clone_fast(clone, ci->bio);
1408 if (len)
1409 bio_setup_sector(clone, ci->sector, *len);
1410
1411 return __map_bio(tio);
1412}
1413
1414static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1415 unsigned num_bios, unsigned *len)
1416{
1417 struct bio_list blist = BIO_EMPTY_LIST;
1418 struct bio *bio;
1419 struct dm_target_io *tio;
1420
1421 alloc_multiple_bios(&blist, ci, ti, num_bios);
1422
1423 while ((bio = bio_list_pop(&blist))) {
1424 tio = container_of(bio, struct dm_target_io, clone);
1425 (void) __clone_and_map_simple_bio(ci, tio, len);
1426 }
1427}
1428
1429static int __send_empty_flush(struct clone_info *ci)
1430{
1431 unsigned target_nr = 0;
1432 struct dm_target *ti;
1433
1434 /*
1435 * Empty flush uses a statically initialized bio, as the base for
1436 * cloning. However, blkg association requires that a bdev is
1437 * associated with a gendisk, which doesn't happen until the bdev is
1438 * opened. So, blkg association is done at issue time of the flush
1439 * rather than when the device is created in alloc_dev().
1440 */
1441 bio_set_dev(ci->bio, ci->io->md->bdev);
1442
1443 BUG_ON(bio_has_data(ci->bio));
1444 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1445 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1446 return 0;
1447}
1448
1449static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1450 sector_t sector, unsigned *len)
1451{
1452 struct bio *bio = ci->bio;
1453 struct dm_target_io *tio;
1454 int r;
1455
1456 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1457 tio->len_ptr = len;
1458 r = clone_bio(tio, bio, sector, *len);
1459 if (r < 0) {
1460 free_tio(tio);
1461 return r;
1462 }
1463 (void) __map_bio(tio);
1464
1465 return 0;
1466}
1467
1468typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1469
1470static unsigned get_num_discard_bios(struct dm_target *ti)
1471{
1472 return ti->num_discard_bios;
1473}
1474
1475static unsigned get_num_secure_erase_bios(struct dm_target *ti)
1476{
1477 return ti->num_secure_erase_bios;
1478}
1479
1480static unsigned get_num_write_same_bios(struct dm_target *ti)
1481{
1482 return ti->num_write_same_bios;
1483}
1484
1485static unsigned get_num_write_zeroes_bios(struct dm_target *ti)
1486{
1487 return ti->num_write_zeroes_bios;
1488}
1489
1490static int __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti,
1491 unsigned num_bios)
1492{
1493 unsigned len;
1494
1495 /*
1496 * Even though the device advertised support for this type of
1497 * request, that does not mean every target supports it, and
1498 * reconfiguration might also have changed that since the
1499 * check was performed.
1500 */
1501 if (!num_bios)
1502 return -EOPNOTSUPP;
1503
1504 len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1505
1506 __send_duplicate_bios(ci, ti, num_bios, &len);
1507
1508 ci->sector += len;
1509 ci->sector_count -= len;
1510
1511 return 0;
1512}
1513
1514static int __send_discard(struct clone_info *ci, struct dm_target *ti)
1515{
1516 return __send_changing_extent_only(ci, ti, get_num_discard_bios(ti));
1517}
1518
1519static int __send_secure_erase(struct clone_info *ci, struct dm_target *ti)
1520{
1521 return __send_changing_extent_only(ci, ti, get_num_secure_erase_bios(ti));
1522}
1523
1524static int __send_write_same(struct clone_info *ci, struct dm_target *ti)
1525{
1526 return __send_changing_extent_only(ci, ti, get_num_write_same_bios(ti));
1527}
1528
1529static int __send_write_zeroes(struct clone_info *ci, struct dm_target *ti)
1530{
1531 return __send_changing_extent_only(ci, ti, get_num_write_zeroes_bios(ti));
1532}
1533
1534static bool is_abnormal_io(struct bio *bio)
1535{
1536 bool r = false;
1537
1538 switch (bio_op(bio)) {
1539 case REQ_OP_DISCARD:
1540 case REQ_OP_SECURE_ERASE:
1541 case REQ_OP_WRITE_SAME:
1542 case REQ_OP_WRITE_ZEROES:
1543 r = true;
1544 break;
1545 }
1546
1547 return r;
1548}
1549
1550static bool __process_abnormal_io(struct clone_info *ci, struct dm_target *ti,
1551 int *result)
1552{
1553 struct bio *bio = ci->bio;
1554
1555 if (bio_op(bio) == REQ_OP_DISCARD)
1556 *result = __send_discard(ci, ti);
1557 else if (bio_op(bio) == REQ_OP_SECURE_ERASE)
1558 *result = __send_secure_erase(ci, ti);
1559 else if (bio_op(bio) == REQ_OP_WRITE_SAME)
1560 *result = __send_write_same(ci, ti);
1561 else if (bio_op(bio) == REQ_OP_WRITE_ZEROES)
1562 *result = __send_write_zeroes(ci, ti);
1563 else
1564 return false;
1565
1566 return true;
1567}
1568
1569/*
1570 * Select the correct strategy for processing a non-flush bio.
1571 */
1572static int __split_and_process_non_flush(struct clone_info *ci)
1573{
1574 struct dm_target *ti;
1575 unsigned len;
1576 int r;
1577
1578 ti = dm_table_find_target(ci->map, ci->sector);
1579 if (!ti)
1580 return -EIO;
1581
1582 if (__process_abnormal_io(ci, ti, &r))
1583 return r;
1584
1585 len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count);
1586
1587 r = __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1588 if (r < 0)
1589 return r;
1590
1591 ci->sector += len;
1592 ci->sector_count -= len;
1593
1594 return 0;
1595}
1596
1597static void init_clone_info(struct clone_info *ci, struct mapped_device *md,
1598 struct dm_table *map, struct bio *bio)
1599{
1600 ci->map = map;
1601 ci->io = alloc_io(md, bio);
1602 ci->sector = bio->bi_iter.bi_sector;
1603}
1604
1605#define __dm_part_stat_sub(part, field, subnd) \
1606 (part_stat_get(part, field) -= (subnd))
1607
1608/*
1609 * Entry point to split a bio into clones and submit them to the targets.
1610 */
1611static blk_qc_t __split_and_process_bio(struct mapped_device *md,
1612 struct dm_table *map, struct bio *bio)
1613{
1614 struct clone_info ci;
1615 blk_qc_t ret = BLK_QC_T_NONE;
1616 int error = 0;
1617
1618 init_clone_info(&ci, md, map, bio);
1619
1620 if (bio->bi_opf & REQ_PREFLUSH) {
1621 struct bio flush_bio;
1622
1623 /*
1624 * Use an on-stack bio for this, it's safe since we don't
1625 * need to reference it after submit. It's just used as
1626 * the basis for the clone(s).
1627 */
1628 bio_init(&flush_bio, NULL, 0);
1629 flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC;
1630 ci.bio = &flush_bio;
1631 ci.sector_count = 0;
1632 error = __send_empty_flush(&ci);
1633 bio_uninit(ci.bio);
1634 /* dec_pending submits any data associated with flush */
1635 } else if (op_is_zone_mgmt(bio_op(bio))) {
1636 ci.bio = bio;
1637 ci.sector_count = 0;
1638 error = __split_and_process_non_flush(&ci);
1639 } else {
1640 ci.bio = bio;
1641 ci.sector_count = bio_sectors(bio);
1642 while (ci.sector_count && !error) {
1643 error = __split_and_process_non_flush(&ci);
1644 if (current->bio_list && ci.sector_count && !error) {
1645 /*
1646 * Remainder must be passed to submit_bio_noacct()
1647 * so that it gets handled *after* bios already submitted
1648 * have been completely processed.
1649 * We take a clone of the original to store in
1650 * ci.io->orig_bio to be used by end_io_acct() and
1651 * for dec_pending to use for completion handling.
1652 */
1653 struct bio *b = bio_split(bio, bio_sectors(bio) - ci.sector_count,
1654 GFP_NOIO, &md->queue->bio_split);
1655 ci.io->orig_bio = b;
1656
1657 /*
1658 * Adjust IO stats for each split, otherwise upon queue
1659 * reentry there will be redundant IO accounting.
1660 * NOTE: this is a stop-gap fix, a proper fix involves
1661 * significant refactoring of DM core's bio splitting
1662 * (by eliminating DM's splitting and just using bio_split)
1663 */
1664 part_stat_lock();
1665 __dm_part_stat_sub(&dm_disk(md)->part0,
1666 sectors[op_stat_group(bio_op(bio))], ci.sector_count);
1667 part_stat_unlock();
1668
1669 bio_chain(b, bio);
1670 trace_block_split(md->queue, b, bio->bi_iter.bi_sector);
1671 ret = submit_bio_noacct(bio);
1672 break;
1673 }
1674 }
1675 }
1676
1677 /* drop the extra reference count */
1678 dec_pending(ci.io, errno_to_blk_status(error));
1679 return ret;
1680}
1681
1682/*
1683 * Optimized variant of __split_and_process_bio that leverages the
1684 * fact that targets that use it do _not_ have a need to split bios.
1685 */
1686static blk_qc_t __process_bio(struct mapped_device *md, struct dm_table *map,
1687 struct bio *bio, struct dm_target *ti)
1688{
1689 struct clone_info ci;
1690 blk_qc_t ret = BLK_QC_T_NONE;
1691 int error = 0;
1692
1693 init_clone_info(&ci, md, map, bio);
1694
1695 if (bio->bi_opf & REQ_PREFLUSH) {
1696 struct bio flush_bio;
1697
1698 /*
1699 * Use an on-stack bio for this, it's safe since we don't
1700 * need to reference it after submit. It's just used as
1701 * the basis for the clone(s).
1702 */
1703 bio_init(&flush_bio, NULL, 0);
1704 flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC;
1705 ci.bio = &flush_bio;
1706 ci.sector_count = 0;
1707 error = __send_empty_flush(&ci);
1708 bio_uninit(ci.bio);
1709 /* dec_pending submits any data associated with flush */
1710 } else {
1711 struct dm_target_io *tio;
1712
1713 ci.bio = bio;
1714 ci.sector_count = bio_sectors(bio);
1715 if (__process_abnormal_io(&ci, ti, &error))
1716 goto out;
1717
1718 tio = alloc_tio(&ci, ti, 0, GFP_NOIO);
1719 ret = __clone_and_map_simple_bio(&ci, tio, NULL);
1720 }
1721out:
1722 /* drop the extra reference count */
1723 dec_pending(ci.io, errno_to_blk_status(error));
1724 return ret;
1725}
1726
1727static blk_qc_t dm_process_bio(struct mapped_device *md,
1728 struct dm_table *map, struct bio *bio)
1729{
1730 blk_qc_t ret = BLK_QC_T_NONE;
1731 struct dm_target *ti = md->immutable_target;
1732
1733 if (unlikely(!map)) {
1734 bio_io_error(bio);
1735 return ret;
1736 }
1737
1738 if (!ti) {
1739 ti = dm_table_find_target(map, bio->bi_iter.bi_sector);
1740 if (unlikely(!ti)) {
1741 bio_io_error(bio);
1742 return ret;
1743 }
1744 }
1745
1746 /*
1747 * If in ->submit_bio we need to use blk_queue_split(), otherwise
1748 * queue_limits for abnormal requests (e.g. discard, writesame, etc)
1749 * won't be imposed.
1750 * If called from dm_wq_work() for deferred bio processing, bio
1751 * was already handled by following code with previous ->submit_bio.
1752 */
1753 if (current->bio_list) {
1754 if (is_abnormal_io(bio))
1755 blk_queue_split(&bio);
1756 /* regular IO is split by __split_and_process_bio */
1757 }
1758
1759 if (dm_get_md_type(md) == DM_TYPE_NVME_BIO_BASED)
1760 return __process_bio(md, map, bio, ti);
1761 return __split_and_process_bio(md, map, bio);
1762}
1763
1764static blk_qc_t dm_submit_bio(struct bio *bio)
1765{
1766 struct mapped_device *md = bio->bi_disk->private_data;
1767 blk_qc_t ret = BLK_QC_T_NONE;
1768 int srcu_idx;
1769 struct dm_table *map;
1770
1771 if (dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) {
1772 /*
1773 * We are called with a live reference on q_usage_counter, but
1774 * that one will be released as soon as we return. Grab an
1775 * extra one as blk_mq_submit_bio expects to be able to consume
1776 * a reference (which lives until the request is freed in case a
1777 * request is allocated).
1778 */
1779 percpu_ref_get(&bio->bi_disk->queue->q_usage_counter);
1780 return blk_mq_submit_bio(bio);
1781 }
1782
1783 map = dm_get_live_table(md, &srcu_idx);
1784
1785 /* if we're suspended, we have to queue this io for later */
1786 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1787 dm_put_live_table(md, srcu_idx);
1788
1789 if (!(bio->bi_opf & REQ_RAHEAD))
1790 queue_io(md, bio);
1791 else
1792 bio_io_error(bio);
1793 return ret;
1794 }
1795
1796 ret = dm_process_bio(md, map, bio);
1797
1798 dm_put_live_table(md, srcu_idx);
1799 return ret;
1800}
1801
1802/*-----------------------------------------------------------------
1803 * An IDR is used to keep track of allocated minor numbers.
1804 *---------------------------------------------------------------*/
1805static void free_minor(int minor)
1806{
1807 spin_lock(&_minor_lock);
1808 idr_remove(&_minor_idr, minor);
1809 spin_unlock(&_minor_lock);
1810}
1811
1812/*
1813 * See if the device with a specific minor # is free.
1814 */
1815static int specific_minor(int minor)
1816{
1817 int r;
1818
1819 if (minor >= (1 << MINORBITS))
1820 return -EINVAL;
1821
1822 idr_preload(GFP_KERNEL);
1823 spin_lock(&_minor_lock);
1824
1825 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1826
1827 spin_unlock(&_minor_lock);
1828 idr_preload_end();
1829 if (r < 0)
1830 return r == -ENOSPC ? -EBUSY : r;
1831 return 0;
1832}
1833
1834static int next_free_minor(int *minor)
1835{
1836 int r;
1837
1838 idr_preload(GFP_KERNEL);
1839 spin_lock(&_minor_lock);
1840
1841 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1842
1843 spin_unlock(&_minor_lock);
1844 idr_preload_end();
1845 if (r < 0)
1846 return r;
1847 *minor = r;
1848 return 0;
1849}
1850
1851static const struct block_device_operations dm_blk_dops;
1852static const struct dax_operations dm_dax_ops;
1853
1854static void dm_wq_work(struct work_struct *work);
1855
1856static void cleanup_mapped_device(struct mapped_device *md)
1857{
1858 if (md->wq)
1859 destroy_workqueue(md->wq);
1860 bioset_exit(&md->bs);
1861 bioset_exit(&md->io_bs);
1862
1863 if (md->dax_dev) {
1864 kill_dax(md->dax_dev);
1865 put_dax(md->dax_dev);
1866 md->dax_dev = NULL;
1867 }
1868
1869 if (md->disk) {
1870 spin_lock(&_minor_lock);
1871 md->disk->private_data = NULL;
1872 spin_unlock(&_minor_lock);
1873 del_gendisk(md->disk);
1874 put_disk(md->disk);
1875 }
1876
1877 if (md->queue)
1878 blk_cleanup_queue(md->queue);
1879
1880 cleanup_srcu_struct(&md->io_barrier);
1881
1882 if (md->bdev) {
1883 bdput(md->bdev);
1884 md->bdev = NULL;
1885 }
1886
1887 mutex_destroy(&md->suspend_lock);
1888 mutex_destroy(&md->type_lock);
1889 mutex_destroy(&md->table_devices_lock);
1890
1891 dm_mq_cleanup_mapped_device(md);
1892}
1893
1894/*
1895 * Allocate and initialise a blank device with a given minor.
1896 */
1897static struct mapped_device *alloc_dev(int minor)
1898{
1899 int r, numa_node_id = dm_get_numa_node();
1900 struct mapped_device *md;
1901 void *old_md;
1902
1903 md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
1904 if (!md) {
1905 DMWARN("unable to allocate device, out of memory.");
1906 return NULL;
1907 }
1908
1909 if (!try_module_get(THIS_MODULE))
1910 goto bad_module_get;
1911
1912 /* get a minor number for the dev */
1913 if (minor == DM_ANY_MINOR)
1914 r = next_free_minor(&minor);
1915 else
1916 r = specific_minor(minor);
1917 if (r < 0)
1918 goto bad_minor;
1919
1920 r = init_srcu_struct(&md->io_barrier);
1921 if (r < 0)
1922 goto bad_io_barrier;
1923
1924 md->numa_node_id = numa_node_id;
1925 md->init_tio_pdu = false;
1926 md->type = DM_TYPE_NONE;
1927 mutex_init(&md->suspend_lock);
1928 mutex_init(&md->type_lock);
1929 mutex_init(&md->table_devices_lock);
1930 spin_lock_init(&md->deferred_lock);
1931 atomic_set(&md->holders, 1);
1932 atomic_set(&md->open_count, 0);
1933 atomic_set(&md->event_nr, 0);
1934 atomic_set(&md->uevent_seq, 0);
1935 INIT_LIST_HEAD(&md->uevent_list);
1936 INIT_LIST_HEAD(&md->table_devices);
1937 spin_lock_init(&md->uevent_lock);
1938
1939 /*
1940 * default to bio-based until DM table is loaded and md->type
1941 * established. If request-based table is loaded: blk-mq will
1942 * override accordingly.
1943 */
1944 md->queue = blk_alloc_queue(numa_node_id);
1945 if (!md->queue)
1946 goto bad;
1947
1948 md->disk = alloc_disk_node(1, md->numa_node_id);
1949 if (!md->disk)
1950 goto bad;
1951
1952 init_waitqueue_head(&md->wait);
1953 INIT_WORK(&md->work, dm_wq_work);
1954 init_waitqueue_head(&md->eventq);
1955 init_completion(&md->kobj_holder.completion);
1956
1957 md->disk->major = _major;
1958 md->disk->first_minor = minor;
1959 md->disk->fops = &dm_blk_dops;
1960 md->disk->queue = md->queue;
1961 md->disk->private_data = md;
1962 sprintf(md->disk->disk_name, "dm-%d", minor);
1963
1964 if (IS_ENABLED(CONFIG_DAX_DRIVER)) {
1965 md->dax_dev = alloc_dax(md, md->disk->disk_name,
1966 &dm_dax_ops, 0);
1967 if (IS_ERR(md->dax_dev))
1968 goto bad;
1969 }
1970
1971 add_disk_no_queue_reg(md->disk);
1972 format_dev_t(md->name, MKDEV(_major, minor));
1973
1974 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
1975 if (!md->wq)
1976 goto bad;
1977
1978 md->bdev = bdget_disk(md->disk, 0);
1979 if (!md->bdev)
1980 goto bad;
1981
1982 dm_stats_init(&md->stats);
1983
1984 /* Populate the mapping, nobody knows we exist yet */
1985 spin_lock(&_minor_lock);
1986 old_md = idr_replace(&_minor_idr, md, minor);
1987 spin_unlock(&_minor_lock);
1988
1989 BUG_ON(old_md != MINOR_ALLOCED);
1990
1991 return md;
1992
1993bad:
1994 cleanup_mapped_device(md);
1995bad_io_barrier:
1996 free_minor(minor);
1997bad_minor:
1998 module_put(THIS_MODULE);
1999bad_module_get:
2000 kvfree(md);
2001 return NULL;
2002}
2003
2004static void unlock_fs(struct mapped_device *md);
2005
2006static void free_dev(struct mapped_device *md)
2007{
2008 int minor = MINOR(disk_devt(md->disk));
2009
2010 unlock_fs(md);
2011
2012 cleanup_mapped_device(md);
2013
2014 free_table_devices(&md->table_devices);
2015 dm_stats_cleanup(&md->stats);
2016 free_minor(minor);
2017
2018 module_put(THIS_MODULE);
2019 kvfree(md);
2020}
2021
2022static int __bind_mempools(struct mapped_device *md, struct dm_table *t)
2023{
2024 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
2025 int ret = 0;
2026
2027 if (dm_table_bio_based(t)) {
2028 /*
2029 * The md may already have mempools that need changing.
2030 * If so, reload bioset because front_pad may have changed
2031 * because a different table was loaded.
2032 */
2033 bioset_exit(&md->bs);
2034 bioset_exit(&md->io_bs);
2035
2036 } else if (bioset_initialized(&md->bs)) {
2037 /*
2038 * There's no need to reload with request-based dm
2039 * because the size of front_pad doesn't change.
2040 * Note for future: If you are to reload bioset,
2041 * prep-ed requests in the queue may refer
2042 * to bio from the old bioset, so you must walk
2043 * through the queue to unprep.
2044 */
2045 goto out;
2046 }
2047
2048 BUG_ON(!p ||
2049 bioset_initialized(&md->bs) ||
2050 bioset_initialized(&md->io_bs));
2051
2052 ret = bioset_init_from_src(&md->bs, &p->bs);
2053 if (ret)
2054 goto out;
2055 ret = bioset_init_from_src(&md->io_bs, &p->io_bs);
2056 if (ret)
2057 bioset_exit(&md->bs);
2058out:
2059 /* mempool bind completed, no longer need any mempools in the table */
2060 dm_table_free_md_mempools(t);
2061 return ret;
2062}
2063
2064/*
2065 * Bind a table to the device.
2066 */
2067static void event_callback(void *context)
2068{
2069 unsigned long flags;
2070 LIST_HEAD(uevents);
2071 struct mapped_device *md = (struct mapped_device *) context;
2072
2073 spin_lock_irqsave(&md->uevent_lock, flags);
2074 list_splice_init(&md->uevent_list, &uevents);
2075 spin_unlock_irqrestore(&md->uevent_lock, flags);
2076
2077 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2078
2079 atomic_inc(&md->event_nr);
2080 wake_up(&md->eventq);
2081 dm_issue_global_event();
2082}
2083
2084/*
2085 * Protected by md->suspend_lock obtained by dm_swap_table().
2086 */
2087static void __set_size(struct mapped_device *md, sector_t size)
2088{
2089 lockdep_assert_held(&md->suspend_lock);
2090
2091 set_capacity(md->disk, size);
2092
2093 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2094}
2095
2096/*
2097 * Returns old map, which caller must destroy.
2098 */
2099static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2100 struct queue_limits *limits)
2101{
2102 struct dm_table *old_map;
2103 struct request_queue *q = md->queue;
2104 bool request_based = dm_table_request_based(t);
2105 sector_t size;
2106 int ret;
2107
2108 lockdep_assert_held(&md->suspend_lock);
2109
2110 size = dm_table_get_size(t);
2111
2112 /*
2113 * Wipe any geometry if the size of the table changed.
2114 */
2115 if (size != dm_get_size(md))
2116 memset(&md->geometry, 0, sizeof(md->geometry));
2117
2118 __set_size(md, size);
2119
2120 dm_table_event_callback(t, event_callback, md);
2121
2122 /*
2123 * The queue hasn't been stopped yet, if the old table type wasn't
2124 * for request-based during suspension. So stop it to prevent
2125 * I/O mapping before resume.
2126 * This must be done before setting the queue restrictions,
2127 * because request-based dm may be run just after the setting.
2128 */
2129 if (request_based)
2130 dm_stop_queue(q);
2131
2132 if (request_based || md->type == DM_TYPE_NVME_BIO_BASED) {
2133 /*
2134 * Leverage the fact that request-based DM targets and
2135 * NVMe bio based targets are immutable singletons
2136 * - used to optimize both dm_request_fn and dm_mq_queue_rq;
2137 * and __process_bio.
2138 */
2139 md->immutable_target = dm_table_get_immutable_target(t);
2140 }
2141
2142 ret = __bind_mempools(md, t);
2143 if (ret) {
2144 old_map = ERR_PTR(ret);
2145 goto out;
2146 }
2147
2148 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2149 rcu_assign_pointer(md->map, (void *)t);
2150 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2151
2152 dm_table_set_restrictions(t, q, limits);
2153 if (old_map)
2154 dm_sync_table(md);
2155
2156out:
2157 return old_map;
2158}
2159
2160/*
2161 * Returns unbound table for the caller to free.
2162 */
2163static struct dm_table *__unbind(struct mapped_device *md)
2164{
2165 struct dm_table *map = rcu_dereference_protected(md->map, 1);
2166
2167 if (!map)
2168 return NULL;
2169
2170 dm_table_event_callback(map, NULL, NULL);
2171 RCU_INIT_POINTER(md->map, NULL);
2172 dm_sync_table(md);
2173
2174 return map;
2175}
2176
2177/*
2178 * Constructor for a new device.
2179 */
2180int dm_create(int minor, struct mapped_device **result)
2181{
2182 int r;
2183 struct mapped_device *md;
2184
2185 md = alloc_dev(minor);
2186 if (!md)
2187 return -ENXIO;
2188
2189 r = dm_sysfs_init(md);
2190 if (r) {
2191 free_dev(md);
2192 return r;
2193 }
2194
2195 *result = md;
2196 return 0;
2197}
2198
2199/*
2200 * Functions to manage md->type.
2201 * All are required to hold md->type_lock.
2202 */
2203void dm_lock_md_type(struct mapped_device *md)
2204{
2205 mutex_lock(&md->type_lock);
2206}
2207
2208void dm_unlock_md_type(struct mapped_device *md)
2209{
2210 mutex_unlock(&md->type_lock);
2211}
2212
2213void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
2214{
2215 BUG_ON(!mutex_is_locked(&md->type_lock));
2216 md->type = type;
2217}
2218
2219enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
2220{
2221 return md->type;
2222}
2223
2224struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2225{
2226 return md->immutable_target_type;
2227}
2228
2229/*
2230 * The queue_limits are only valid as long as you have a reference
2231 * count on 'md'.
2232 */
2233struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2234{
2235 BUG_ON(!atomic_read(&md->holders));
2236 return &md->queue->limits;
2237}
2238EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2239
2240/*
2241 * Setup the DM device's queue based on md's type
2242 */
2243int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
2244{
2245 int r;
2246 struct queue_limits limits;
2247 enum dm_queue_mode type = dm_get_md_type(md);
2248
2249 switch (type) {
2250 case DM_TYPE_REQUEST_BASED:
2251 r = dm_mq_init_request_queue(md, t);
2252 if (r) {
2253 DMERR("Cannot initialize queue for request-based dm-mq mapped device");
2254 return r;
2255 }
2256 break;
2257 case DM_TYPE_BIO_BASED:
2258 case DM_TYPE_DAX_BIO_BASED:
2259 case DM_TYPE_NVME_BIO_BASED:
2260 break;
2261 case DM_TYPE_NONE:
2262 WARN_ON_ONCE(true);
2263 break;
2264 }
2265
2266 r = dm_calculate_queue_limits(t, &limits);
2267 if (r) {
2268 DMERR("Cannot calculate initial queue limits");
2269 return r;
2270 }
2271 dm_table_set_restrictions(t, md->queue, &limits);
2272 blk_register_queue(md->disk);
2273
2274 return 0;
2275}
2276
2277struct mapped_device *dm_get_md(dev_t dev)
2278{
2279 struct mapped_device *md;
2280 unsigned minor = MINOR(dev);
2281
2282 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2283 return NULL;
2284
2285 spin_lock(&_minor_lock);
2286
2287 md = idr_find(&_minor_idr, minor);
2288 if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) ||
2289 test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2290 md = NULL;
2291 goto out;
2292 }
2293 dm_get(md);
2294out:
2295 spin_unlock(&_minor_lock);
2296
2297 return md;
2298}
2299EXPORT_SYMBOL_GPL(dm_get_md);
2300
2301void *dm_get_mdptr(struct mapped_device *md)
2302{
2303 return md->interface_ptr;
2304}
2305
2306void dm_set_mdptr(struct mapped_device *md, void *ptr)
2307{
2308 md->interface_ptr = ptr;
2309}
2310
2311void dm_get(struct mapped_device *md)
2312{
2313 atomic_inc(&md->holders);
2314 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2315}
2316
2317int dm_hold(struct mapped_device *md)
2318{
2319 spin_lock(&_minor_lock);
2320 if (test_bit(DMF_FREEING, &md->flags)) {
2321 spin_unlock(&_minor_lock);
2322 return -EBUSY;
2323 }
2324 dm_get(md);
2325 spin_unlock(&_minor_lock);
2326 return 0;
2327}
2328EXPORT_SYMBOL_GPL(dm_hold);
2329
2330const char *dm_device_name(struct mapped_device *md)
2331{
2332 return md->name;
2333}
2334EXPORT_SYMBOL_GPL(dm_device_name);
2335
2336static void __dm_destroy(struct mapped_device *md, bool wait)
2337{
2338 struct dm_table *map;
2339 int srcu_idx;
2340
2341 might_sleep();
2342
2343 spin_lock(&_minor_lock);
2344 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2345 set_bit(DMF_FREEING, &md->flags);
2346 spin_unlock(&_minor_lock);
2347
2348 blk_set_queue_dying(md->queue);
2349
2350 /*
2351 * Take suspend_lock so that presuspend and postsuspend methods
2352 * do not race with internal suspend.
2353 */
2354 mutex_lock(&md->suspend_lock);
2355 map = dm_get_live_table(md, &srcu_idx);
2356 if (!dm_suspended_md(md)) {
2357 dm_table_presuspend_targets(map);
2358 set_bit(DMF_SUSPENDED, &md->flags);
2359 set_bit(DMF_POST_SUSPENDING, &md->flags);
2360 dm_table_postsuspend_targets(map);
2361 }
2362 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2363 dm_put_live_table(md, srcu_idx);
2364 mutex_unlock(&md->suspend_lock);
2365
2366 /*
2367 * Rare, but there may be I/O requests still going to complete,
2368 * for example. Wait for all references to disappear.
2369 * No one should increment the reference count of the mapped_device,
2370 * after the mapped_device state becomes DMF_FREEING.
2371 */
2372 if (wait)
2373 while (atomic_read(&md->holders))
2374 msleep(1);
2375 else if (atomic_read(&md->holders))
2376 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2377 dm_device_name(md), atomic_read(&md->holders));
2378
2379 dm_sysfs_exit(md);
2380 dm_table_destroy(__unbind(md));
2381 free_dev(md);
2382}
2383
2384void dm_destroy(struct mapped_device *md)
2385{
2386 __dm_destroy(md, true);
2387}
2388
2389void dm_destroy_immediate(struct mapped_device *md)
2390{
2391 __dm_destroy(md, false);
2392}
2393
2394void dm_put(struct mapped_device *md)
2395{
2396 atomic_dec(&md->holders);
2397}
2398EXPORT_SYMBOL_GPL(dm_put);
2399
2400static bool md_in_flight_bios(struct mapped_device *md)
2401{
2402 int cpu;
2403 struct hd_struct *part = &dm_disk(md)->part0;
2404 long sum = 0;
2405
2406 for_each_possible_cpu(cpu) {
2407 sum += part_stat_local_read_cpu(part, in_flight[0], cpu);
2408 sum += part_stat_local_read_cpu(part, in_flight[1], cpu);
2409 }
2410
2411 return sum != 0;
2412}
2413
2414static int dm_wait_for_bios_completion(struct mapped_device *md, long task_state)
2415{
2416 int r = 0;
2417 DEFINE_WAIT(wait);
2418
2419 while (true) {
2420 prepare_to_wait(&md->wait, &wait, task_state);
2421
2422 if (!md_in_flight_bios(md))
2423 break;
2424
2425 if (signal_pending_state(task_state, current)) {
2426 r = -EINTR;
2427 break;
2428 }
2429
2430 io_schedule();
2431 }
2432 finish_wait(&md->wait, &wait);
2433
2434 return r;
2435}
2436
2437static int dm_wait_for_completion(struct mapped_device *md, long task_state)
2438{
2439 int r = 0;
2440
2441 if (!queue_is_mq(md->queue))
2442 return dm_wait_for_bios_completion(md, task_state);
2443
2444 while (true) {
2445 if (!blk_mq_queue_inflight(md->queue))
2446 break;
2447
2448 if (signal_pending_state(task_state, current)) {
2449 r = -EINTR;
2450 break;
2451 }
2452
2453 msleep(5);
2454 }
2455
2456 return r;
2457}
2458
2459/*
2460 * Process the deferred bios
2461 */
2462static void dm_wq_work(struct work_struct *work)
2463{
2464 struct mapped_device *md = container_of(work, struct mapped_device,
2465 work);
2466 struct bio *c;
2467 int srcu_idx;
2468 struct dm_table *map;
2469
2470 map = dm_get_live_table(md, &srcu_idx);
2471
2472 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2473 spin_lock_irq(&md->deferred_lock);
2474 c = bio_list_pop(&md->deferred);
2475 spin_unlock_irq(&md->deferred_lock);
2476
2477 if (!c)
2478 break;
2479
2480 if (dm_request_based(md))
2481 (void) submit_bio_noacct(c);
2482 else
2483 (void) dm_process_bio(md, map, c);
2484 }
2485
2486 dm_put_live_table(md, srcu_idx);
2487}
2488
2489static void dm_queue_flush(struct mapped_device *md)
2490{
2491 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2492 smp_mb__after_atomic();
2493 queue_work(md->wq, &md->work);
2494}
2495
2496/*
2497 * Swap in a new table, returning the old one for the caller to destroy.
2498 */
2499struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2500{
2501 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2502 struct queue_limits limits;
2503 int r;
2504
2505 mutex_lock(&md->suspend_lock);
2506
2507 /* device must be suspended */
2508 if (!dm_suspended_md(md))
2509 goto out;
2510
2511 /*
2512 * If the new table has no data devices, retain the existing limits.
2513 * This helps multipath with queue_if_no_path if all paths disappear,
2514 * then new I/O is queued based on these limits, and then some paths
2515 * reappear.
2516 */
2517 if (dm_table_has_no_data_devices(table)) {
2518 live_map = dm_get_live_table_fast(md);
2519 if (live_map)
2520 limits = md->queue->limits;
2521 dm_put_live_table_fast(md);
2522 }
2523
2524 if (!live_map) {
2525 r = dm_calculate_queue_limits(table, &limits);
2526 if (r) {
2527 map = ERR_PTR(r);
2528 goto out;
2529 }
2530 }
2531
2532 map = __bind(md, table, &limits);
2533 dm_issue_global_event();
2534
2535out:
2536 mutex_unlock(&md->suspend_lock);
2537 return map;
2538}
2539
2540/*
2541 * Functions to lock and unlock any filesystem running on the
2542 * device.
2543 */
2544static int lock_fs(struct mapped_device *md)
2545{
2546 int r;
2547
2548 WARN_ON(md->frozen_sb);
2549
2550 md->frozen_sb = freeze_bdev(md->bdev);
2551 if (IS_ERR(md->frozen_sb)) {
2552 r = PTR_ERR(md->frozen_sb);
2553 md->frozen_sb = NULL;
2554 return r;
2555 }
2556
2557 set_bit(DMF_FROZEN, &md->flags);
2558
2559 return 0;
2560}
2561
2562static void unlock_fs(struct mapped_device *md)
2563{
2564 if (!test_bit(DMF_FROZEN, &md->flags))
2565 return;
2566
2567 thaw_bdev(md->bdev, md->frozen_sb);
2568 md->frozen_sb = NULL;
2569 clear_bit(DMF_FROZEN, &md->flags);
2570}
2571
2572/*
2573 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2574 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2575 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2576 *
2577 * If __dm_suspend returns 0, the device is completely quiescent
2578 * now. There is no request-processing activity. All new requests
2579 * are being added to md->deferred list.
2580 */
2581static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2582 unsigned suspend_flags, long task_state,
2583 int dmf_suspended_flag)
2584{
2585 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2586 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2587 int r;
2588
2589 lockdep_assert_held(&md->suspend_lock);
2590
2591 /*
2592 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2593 * This flag is cleared before dm_suspend returns.
2594 */
2595 if (noflush)
2596 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2597 else
2598 DMDEBUG("%s: suspending with flush", dm_device_name(md));
2599
2600 /*
2601 * This gets reverted if there's an error later and the targets
2602 * provide the .presuspend_undo hook.
2603 */
2604 dm_table_presuspend_targets(map);
2605
2606 /*
2607 * Flush I/O to the device.
2608 * Any I/O submitted after lock_fs() may not be flushed.
2609 * noflush takes precedence over do_lockfs.
2610 * (lock_fs() flushes I/Os and waits for them to complete.)
2611 */
2612 if (!noflush && do_lockfs) {
2613 r = lock_fs(md);
2614 if (r) {
2615 dm_table_presuspend_undo_targets(map);
2616 return r;
2617 }
2618 }
2619
2620 /*
2621 * Here we must make sure that no processes are submitting requests
2622 * to target drivers i.e. no one may be executing
2623 * __split_and_process_bio. This is called from dm_request and
2624 * dm_wq_work.
2625 *
2626 * To get all processes out of __split_and_process_bio in dm_request,
2627 * we take the write lock. To prevent any process from reentering
2628 * __split_and_process_bio from dm_request and quiesce the thread
2629 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2630 * flush_workqueue(md->wq).
2631 */
2632 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2633 if (map)
2634 synchronize_srcu(&md->io_barrier);
2635
2636 /*
2637 * Stop md->queue before flushing md->wq in case request-based
2638 * dm defers requests to md->wq from md->queue.
2639 */
2640 if (dm_request_based(md))
2641 dm_stop_queue(md->queue);
2642
2643 flush_workqueue(md->wq);
2644
2645 /*
2646 * At this point no more requests are entering target request routines.
2647 * We call dm_wait_for_completion to wait for all existing requests
2648 * to finish.
2649 */
2650 r = dm_wait_for_completion(md, task_state);
2651 if (!r)
2652 set_bit(dmf_suspended_flag, &md->flags);
2653
2654 if (noflush)
2655 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2656 if (map)
2657 synchronize_srcu(&md->io_barrier);
2658
2659 /* were we interrupted ? */
2660 if (r < 0) {
2661 dm_queue_flush(md);
2662
2663 if (dm_request_based(md))
2664 dm_start_queue(md->queue);
2665
2666 unlock_fs(md);
2667 dm_table_presuspend_undo_targets(map);
2668 /* pushback list is already flushed, so skip flush */
2669 }
2670
2671 return r;
2672}
2673
2674/*
2675 * We need to be able to change a mapping table under a mounted
2676 * filesystem. For example we might want to move some data in
2677 * the background. Before the table can be swapped with
2678 * dm_bind_table, dm_suspend must be called to flush any in
2679 * flight bios and ensure that any further io gets deferred.
2680 */
2681/*
2682 * Suspend mechanism in request-based dm.
2683 *
2684 * 1. Flush all I/Os by lock_fs() if needed.
2685 * 2. Stop dispatching any I/O by stopping the request_queue.
2686 * 3. Wait for all in-flight I/Os to be completed or requeued.
2687 *
2688 * To abort suspend, start the request_queue.
2689 */
2690int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2691{
2692 struct dm_table *map = NULL;
2693 int r = 0;
2694
2695retry:
2696 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2697
2698 if (dm_suspended_md(md)) {
2699 r = -EINVAL;
2700 goto out_unlock;
2701 }
2702
2703 if (dm_suspended_internally_md(md)) {
2704 /* already internally suspended, wait for internal resume */
2705 mutex_unlock(&md->suspend_lock);
2706 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2707 if (r)
2708 return r;
2709 goto retry;
2710 }
2711
2712 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2713
2714 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2715 if (r)
2716 goto out_unlock;
2717
2718 set_bit(DMF_POST_SUSPENDING, &md->flags);
2719 dm_table_postsuspend_targets(map);
2720 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2721
2722out_unlock:
2723 mutex_unlock(&md->suspend_lock);
2724 return r;
2725}
2726
2727static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2728{
2729 if (map) {
2730 int r = dm_table_resume_targets(map);
2731 if (r)
2732 return r;
2733 }
2734
2735 dm_queue_flush(md);
2736
2737 /*
2738 * Flushing deferred I/Os must be done after targets are resumed
2739 * so that mapping of targets can work correctly.
2740 * Request-based dm is queueing the deferred I/Os in its request_queue.
2741 */
2742 if (dm_request_based(md))
2743 dm_start_queue(md->queue);
2744
2745 unlock_fs(md);
2746
2747 return 0;
2748}
2749
2750int dm_resume(struct mapped_device *md)
2751{
2752 int r;
2753 struct dm_table *map = NULL;
2754
2755retry:
2756 r = -EINVAL;
2757 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2758
2759 if (!dm_suspended_md(md))
2760 goto out;
2761
2762 if (dm_suspended_internally_md(md)) {
2763 /* already internally suspended, wait for internal resume */
2764 mutex_unlock(&md->suspend_lock);
2765 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2766 if (r)
2767 return r;
2768 goto retry;
2769 }
2770
2771 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2772 if (!map || !dm_table_get_size(map))
2773 goto out;
2774
2775 r = __dm_resume(md, map);
2776 if (r)
2777 goto out;
2778
2779 clear_bit(DMF_SUSPENDED, &md->flags);
2780out:
2781 mutex_unlock(&md->suspend_lock);
2782
2783 return r;
2784}
2785
2786/*
2787 * Internal suspend/resume works like userspace-driven suspend. It waits
2788 * until all bios finish and prevents issuing new bios to the target drivers.
2789 * It may be used only from the kernel.
2790 */
2791
2792static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
2793{
2794 struct dm_table *map = NULL;
2795
2796 lockdep_assert_held(&md->suspend_lock);
2797
2798 if (md->internal_suspend_count++)
2799 return; /* nested internal suspend */
2800
2801 if (dm_suspended_md(md)) {
2802 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2803 return; /* nest suspend */
2804 }
2805
2806 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2807
2808 /*
2809 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2810 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2811 * would require changing .presuspend to return an error -- avoid this
2812 * until there is a need for more elaborate variants of internal suspend.
2813 */
2814 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2815 DMF_SUSPENDED_INTERNALLY);
2816
2817 set_bit(DMF_POST_SUSPENDING, &md->flags);
2818 dm_table_postsuspend_targets(map);
2819 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2820}
2821
2822static void __dm_internal_resume(struct mapped_device *md)
2823{
2824 BUG_ON(!md->internal_suspend_count);
2825
2826 if (--md->internal_suspend_count)
2827 return; /* resume from nested internal suspend */
2828
2829 if (dm_suspended_md(md))
2830 goto done; /* resume from nested suspend */
2831
2832 /*
2833 * NOTE: existing callers don't need to call dm_table_resume_targets
2834 * (which may fail -- so best to avoid it for now by passing NULL map)
2835 */
2836 (void) __dm_resume(md, NULL);
2837
2838done:
2839 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2840 smp_mb__after_atomic();
2841 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2842}
2843
2844void dm_internal_suspend_noflush(struct mapped_device *md)
2845{
2846 mutex_lock(&md->suspend_lock);
2847 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2848 mutex_unlock(&md->suspend_lock);
2849}
2850EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2851
2852void dm_internal_resume(struct mapped_device *md)
2853{
2854 mutex_lock(&md->suspend_lock);
2855 __dm_internal_resume(md);
2856 mutex_unlock(&md->suspend_lock);
2857}
2858EXPORT_SYMBOL_GPL(dm_internal_resume);
2859
2860/*
2861 * Fast variants of internal suspend/resume hold md->suspend_lock,
2862 * which prevents interaction with userspace-driven suspend.
2863 */
2864
2865void dm_internal_suspend_fast(struct mapped_device *md)
2866{
2867 mutex_lock(&md->suspend_lock);
2868 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2869 return;
2870
2871 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2872 synchronize_srcu(&md->io_barrier);
2873 flush_workqueue(md->wq);
2874 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2875}
2876EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2877
2878void dm_internal_resume_fast(struct mapped_device *md)
2879{
2880 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2881 goto done;
2882
2883 dm_queue_flush(md);
2884
2885done:
2886 mutex_unlock(&md->suspend_lock);
2887}
2888EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2889
2890/*-----------------------------------------------------------------
2891 * Event notification.
2892 *---------------------------------------------------------------*/
2893int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2894 unsigned cookie)
2895{
2896 int r;
2897 unsigned noio_flag;
2898 char udev_cookie[DM_COOKIE_LENGTH];
2899 char *envp[] = { udev_cookie, NULL };
2900
2901 noio_flag = memalloc_noio_save();
2902
2903 if (!cookie)
2904 r = kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2905 else {
2906 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2907 DM_COOKIE_ENV_VAR_NAME, cookie);
2908 r = kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2909 action, envp);
2910 }
2911
2912 memalloc_noio_restore(noio_flag);
2913
2914 return r;
2915}
2916
2917uint32_t dm_next_uevent_seq(struct mapped_device *md)
2918{
2919 return atomic_add_return(1, &md->uevent_seq);
2920}
2921
2922uint32_t dm_get_event_nr(struct mapped_device *md)
2923{
2924 return atomic_read(&md->event_nr);
2925}
2926
2927int dm_wait_event(struct mapped_device *md, int event_nr)
2928{
2929 return wait_event_interruptible(md->eventq,
2930 (event_nr != atomic_read(&md->event_nr)));
2931}
2932
2933void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2934{
2935 unsigned long flags;
2936
2937 spin_lock_irqsave(&md->uevent_lock, flags);
2938 list_add(elist, &md->uevent_list);
2939 spin_unlock_irqrestore(&md->uevent_lock, flags);
2940}
2941
2942/*
2943 * The gendisk is only valid as long as you have a reference
2944 * count on 'md'.
2945 */
2946struct gendisk *dm_disk(struct mapped_device *md)
2947{
2948 return md->disk;
2949}
2950EXPORT_SYMBOL_GPL(dm_disk);
2951
2952struct kobject *dm_kobject(struct mapped_device *md)
2953{
2954 return &md->kobj_holder.kobj;
2955}
2956
2957struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2958{
2959 struct mapped_device *md;
2960
2961 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2962
2963 spin_lock(&_minor_lock);
2964 if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2965 md = NULL;
2966 goto out;
2967 }
2968 dm_get(md);
2969out:
2970 spin_unlock(&_minor_lock);
2971
2972 return md;
2973}
2974
2975int dm_suspended_md(struct mapped_device *md)
2976{
2977 return test_bit(DMF_SUSPENDED, &md->flags);
2978}
2979
2980static int dm_post_suspending_md(struct mapped_device *md)
2981{
2982 return test_bit(DMF_POST_SUSPENDING, &md->flags);
2983}
2984
2985int dm_suspended_internally_md(struct mapped_device *md)
2986{
2987 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2988}
2989
2990int dm_test_deferred_remove_flag(struct mapped_device *md)
2991{
2992 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2993}
2994
2995int dm_suspended(struct dm_target *ti)
2996{
2997 return dm_suspended_md(dm_table_get_md(ti->table));
2998}
2999EXPORT_SYMBOL_GPL(dm_suspended);
3000
3001int dm_post_suspending(struct dm_target *ti)
3002{
3003 return dm_post_suspending_md(dm_table_get_md(ti->table));
3004}
3005EXPORT_SYMBOL_GPL(dm_post_suspending);
3006
3007int dm_noflush_suspending(struct dm_target *ti)
3008{
3009 return __noflush_suspending(dm_table_get_md(ti->table));
3010}
3011EXPORT_SYMBOL_GPL(dm_noflush_suspending);
3012
3013struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type,
3014 unsigned integrity, unsigned per_io_data_size,
3015 unsigned min_pool_size)
3016{
3017 struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
3018 unsigned int pool_size = 0;
3019 unsigned int front_pad, io_front_pad;
3020 int ret;
3021
3022 if (!pools)
3023 return NULL;
3024
3025 switch (type) {
3026 case DM_TYPE_BIO_BASED:
3027 case DM_TYPE_DAX_BIO_BASED:
3028 case DM_TYPE_NVME_BIO_BASED:
3029 pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size);
3030 front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
3031 io_front_pad = roundup(front_pad, __alignof__(struct dm_io)) + offsetof(struct dm_io, tio);
3032 ret = bioset_init(&pools->io_bs, pool_size, io_front_pad, 0);
3033 if (ret)
3034 goto out;
3035 if (integrity && bioset_integrity_create(&pools->io_bs, pool_size))
3036 goto out;
3037 break;
3038 case DM_TYPE_REQUEST_BASED:
3039 pool_size = max(dm_get_reserved_rq_based_ios(), min_pool_size);
3040 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
3041 /* per_io_data_size is used for blk-mq pdu at queue allocation */
3042 break;
3043 default:
3044 BUG();
3045 }
3046
3047 ret = bioset_init(&pools->bs, pool_size, front_pad, 0);
3048 if (ret)
3049 goto out;
3050
3051 if (integrity && bioset_integrity_create(&pools->bs, pool_size))
3052 goto out;
3053
3054 return pools;
3055
3056out:
3057 dm_free_md_mempools(pools);
3058
3059 return NULL;
3060}
3061
3062void dm_free_md_mempools(struct dm_md_mempools *pools)
3063{
3064 if (!pools)
3065 return;
3066
3067 bioset_exit(&pools->bs);
3068 bioset_exit(&pools->io_bs);
3069
3070 kfree(pools);
3071}
3072
3073struct dm_pr {
3074 u64 old_key;
3075 u64 new_key;
3076 u32 flags;
3077 bool fail_early;
3078};
3079
3080static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
3081 void *data)
3082{
3083 struct mapped_device *md = bdev->bd_disk->private_data;
3084 struct dm_table *table;
3085 struct dm_target *ti;
3086 int ret = -ENOTTY, srcu_idx;
3087
3088 table = dm_get_live_table(md, &srcu_idx);
3089 if (!table || !dm_table_get_size(table))
3090 goto out;
3091
3092 /* We only support devices that have a single target */
3093 if (dm_table_get_num_targets(table) != 1)
3094 goto out;
3095 ti = dm_table_get_target(table, 0);
3096
3097 ret = -EINVAL;
3098 if (!ti->type->iterate_devices)
3099 goto out;
3100
3101 ret = ti->type->iterate_devices(ti, fn, data);
3102out:
3103 dm_put_live_table(md, srcu_idx);
3104 return ret;
3105}
3106
3107/*
3108 * For register / unregister we need to manually call out to every path.
3109 */
3110static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
3111 sector_t start, sector_t len, void *data)
3112{
3113 struct dm_pr *pr = data;
3114 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3115
3116 if (!ops || !ops->pr_register)
3117 return -EOPNOTSUPP;
3118 return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
3119}
3120
3121static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
3122 u32 flags)
3123{
3124 struct dm_pr pr = {
3125 .old_key = old_key,
3126 .new_key = new_key,
3127 .flags = flags,
3128 .fail_early = true,
3129 };
3130 int ret;
3131
3132 ret = dm_call_pr(bdev, __dm_pr_register, &pr);
3133 if (ret && new_key) {
3134 /* unregister all paths if we failed to register any path */
3135 pr.old_key = new_key;
3136 pr.new_key = 0;
3137 pr.flags = 0;
3138 pr.fail_early = false;
3139 dm_call_pr(bdev, __dm_pr_register, &pr);
3140 }
3141
3142 return ret;
3143}
3144
3145static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
3146 u32 flags)
3147{
3148 struct mapped_device *md = bdev->bd_disk->private_data;
3149 const struct pr_ops *ops;
3150 int r, srcu_idx;
3151
3152 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3153 if (r < 0)
3154 goto out;
3155
3156 ops = bdev->bd_disk->fops->pr_ops;
3157 if (ops && ops->pr_reserve)
3158 r = ops->pr_reserve(bdev, key, type, flags);
3159 else
3160 r = -EOPNOTSUPP;
3161out:
3162 dm_unprepare_ioctl(md, srcu_idx);
3163 return r;
3164}
3165
3166static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
3167{
3168 struct mapped_device *md = bdev->bd_disk->private_data;
3169 const struct pr_ops *ops;
3170 int r, srcu_idx;
3171
3172 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3173 if (r < 0)
3174 goto out;
3175
3176 ops = bdev->bd_disk->fops->pr_ops;
3177 if (ops && ops->pr_release)
3178 r = ops->pr_release(bdev, key, type);
3179 else
3180 r = -EOPNOTSUPP;
3181out:
3182 dm_unprepare_ioctl(md, srcu_idx);
3183 return r;
3184}
3185
3186static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
3187 enum pr_type type, bool abort)
3188{
3189 struct mapped_device *md = bdev->bd_disk->private_data;
3190 const struct pr_ops *ops;
3191 int r, srcu_idx;
3192
3193 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3194 if (r < 0)
3195 goto out;
3196
3197 ops = bdev->bd_disk->fops->pr_ops;
3198 if (ops && ops->pr_preempt)
3199 r = ops->pr_preempt(bdev, old_key, new_key, type, abort);
3200 else
3201 r = -EOPNOTSUPP;
3202out:
3203 dm_unprepare_ioctl(md, srcu_idx);
3204 return r;
3205}
3206
3207static int dm_pr_clear(struct block_device *bdev, u64 key)
3208{
3209 struct mapped_device *md = bdev->bd_disk->private_data;
3210 const struct pr_ops *ops;
3211 int r, srcu_idx;
3212
3213 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3214 if (r < 0)
3215 goto out;
3216
3217 ops = bdev->bd_disk->fops->pr_ops;
3218 if (ops && ops->pr_clear)
3219 r = ops->pr_clear(bdev, key);
3220 else
3221 r = -EOPNOTSUPP;
3222out:
3223 dm_unprepare_ioctl(md, srcu_idx);
3224 return r;
3225}
3226
3227static const struct pr_ops dm_pr_ops = {
3228 .pr_register = dm_pr_register,
3229 .pr_reserve = dm_pr_reserve,
3230 .pr_release = dm_pr_release,
3231 .pr_preempt = dm_pr_preempt,
3232 .pr_clear = dm_pr_clear,
3233};
3234
3235static const struct block_device_operations dm_blk_dops = {
3236 .submit_bio = dm_submit_bio,
3237 .open = dm_blk_open,
3238 .release = dm_blk_close,
3239 .ioctl = dm_blk_ioctl,
3240 .getgeo = dm_blk_getgeo,
3241 .report_zones = dm_blk_report_zones,
3242 .pr_ops = &dm_pr_ops,
3243 .owner = THIS_MODULE
3244};
3245
3246static const struct dax_operations dm_dax_ops = {
3247 .direct_access = dm_dax_direct_access,
3248 .dax_supported = dm_dax_supported,
3249 .copy_from_iter = dm_dax_copy_from_iter,
3250 .copy_to_iter = dm_dax_copy_to_iter,
3251 .zero_page_range = dm_dax_zero_page_range,
3252};
3253
3254/*
3255 * module hooks
3256 */
3257module_init(dm_init);
3258module_exit(dm_exit);
3259
3260module_param(major, uint, 0);
3261MODULE_PARM_DESC(major, "The major number of the device mapper");
3262
3263module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3264MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3265
3266module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
3267MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
3268
3269MODULE_DESCRIPTION(DM_NAME " driver");
3270MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3271MODULE_LICENSE("GPL");
1/*
2 * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
4 *
5 * This file is released under the GPL.
6 */
7
8#include "dm.h"
9#include "dm-uevent.h"
10
11#include <linux/init.h>
12#include <linux/module.h>
13#include <linux/mutex.h>
14#include <linux/moduleparam.h>
15#include <linux/blkpg.h>
16#include <linux/bio.h>
17#include <linux/mempool.h>
18#include <linux/slab.h>
19#include <linux/idr.h>
20#include <linux/hdreg.h>
21#include <linux/delay.h>
22
23#include <trace/events/block.h>
24
25#define DM_MSG_PREFIX "core"
26
27#ifdef CONFIG_PRINTK
28/*
29 * ratelimit state to be used in DMXXX_LIMIT().
30 */
31DEFINE_RATELIMIT_STATE(dm_ratelimit_state,
32 DEFAULT_RATELIMIT_INTERVAL,
33 DEFAULT_RATELIMIT_BURST);
34EXPORT_SYMBOL(dm_ratelimit_state);
35#endif
36
37/*
38 * Cookies are numeric values sent with CHANGE and REMOVE
39 * uevents while resuming, removing or renaming the device.
40 */
41#define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
42#define DM_COOKIE_LENGTH 24
43
44static const char *_name = DM_NAME;
45
46static unsigned int major = 0;
47static unsigned int _major = 0;
48
49static DEFINE_IDR(_minor_idr);
50
51static DEFINE_SPINLOCK(_minor_lock);
52/*
53 * For bio-based dm.
54 * One of these is allocated per bio.
55 */
56struct dm_io {
57 struct mapped_device *md;
58 int error;
59 atomic_t io_count;
60 struct bio *bio;
61 unsigned long start_time;
62 spinlock_t endio_lock;
63};
64
65/*
66 * For bio-based dm.
67 * One of these is allocated per target within a bio. Hopefully
68 * this will be simplified out one day.
69 */
70struct dm_target_io {
71 struct dm_io *io;
72 struct dm_target *ti;
73 union map_info info;
74};
75
76/*
77 * For request-based dm.
78 * One of these is allocated per request.
79 */
80struct dm_rq_target_io {
81 struct mapped_device *md;
82 struct dm_target *ti;
83 struct request *orig, clone;
84 int error;
85 union map_info info;
86};
87
88/*
89 * For request-based dm.
90 * One of these is allocated per bio.
91 */
92struct dm_rq_clone_bio_info {
93 struct bio *orig;
94 struct dm_rq_target_io *tio;
95};
96
97union map_info *dm_get_mapinfo(struct bio *bio)
98{
99 if (bio && bio->bi_private)
100 return &((struct dm_target_io *)bio->bi_private)->info;
101 return NULL;
102}
103
104union map_info *dm_get_rq_mapinfo(struct request *rq)
105{
106 if (rq && rq->end_io_data)
107 return &((struct dm_rq_target_io *)rq->end_io_data)->info;
108 return NULL;
109}
110EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo);
111
112#define MINOR_ALLOCED ((void *)-1)
113
114/*
115 * Bits for the md->flags field.
116 */
117#define DMF_BLOCK_IO_FOR_SUSPEND 0
118#define DMF_SUSPENDED 1
119#define DMF_FROZEN 2
120#define DMF_FREEING 3
121#define DMF_DELETING 4
122#define DMF_NOFLUSH_SUSPENDING 5
123#define DMF_MERGE_IS_OPTIONAL 6
124
125/*
126 * Work processed by per-device workqueue.
127 */
128struct mapped_device {
129 struct rw_semaphore io_lock;
130 struct mutex suspend_lock;
131 rwlock_t map_lock;
132 atomic_t holders;
133 atomic_t open_count;
134
135 unsigned long flags;
136
137 struct request_queue *queue;
138 unsigned type;
139 /* Protect queue and type against concurrent access. */
140 struct mutex type_lock;
141
142 struct target_type *immutable_target_type;
143
144 struct gendisk *disk;
145 char name[16];
146
147 void *interface_ptr;
148
149 /*
150 * A list of ios that arrived while we were suspended.
151 */
152 atomic_t pending[2];
153 wait_queue_head_t wait;
154 struct work_struct work;
155 struct bio_list deferred;
156 spinlock_t deferred_lock;
157
158 /*
159 * Processing queue (flush)
160 */
161 struct workqueue_struct *wq;
162
163 /*
164 * The current mapping.
165 */
166 struct dm_table *map;
167
168 /*
169 * io objects are allocated from here.
170 */
171 mempool_t *io_pool;
172 mempool_t *tio_pool;
173
174 struct bio_set *bs;
175
176 /*
177 * Event handling.
178 */
179 atomic_t event_nr;
180 wait_queue_head_t eventq;
181 atomic_t uevent_seq;
182 struct list_head uevent_list;
183 spinlock_t uevent_lock; /* Protect access to uevent_list */
184
185 /*
186 * freeze/thaw support require holding onto a super block
187 */
188 struct super_block *frozen_sb;
189 struct block_device *bdev;
190
191 /* forced geometry settings */
192 struct hd_geometry geometry;
193
194 /* sysfs handle */
195 struct kobject kobj;
196
197 /* zero-length flush that will be cloned and submitted to targets */
198 struct bio flush_bio;
199};
200
201/*
202 * For mempools pre-allocation at the table loading time.
203 */
204struct dm_md_mempools {
205 mempool_t *io_pool;
206 mempool_t *tio_pool;
207 struct bio_set *bs;
208};
209
210#define MIN_IOS 256
211static struct kmem_cache *_io_cache;
212static struct kmem_cache *_tio_cache;
213static struct kmem_cache *_rq_tio_cache;
214static struct kmem_cache *_rq_bio_info_cache;
215
216static int __init local_init(void)
217{
218 int r = -ENOMEM;
219
220 /* allocate a slab for the dm_ios */
221 _io_cache = KMEM_CACHE(dm_io, 0);
222 if (!_io_cache)
223 return r;
224
225 /* allocate a slab for the target ios */
226 _tio_cache = KMEM_CACHE(dm_target_io, 0);
227 if (!_tio_cache)
228 goto out_free_io_cache;
229
230 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
231 if (!_rq_tio_cache)
232 goto out_free_tio_cache;
233
234 _rq_bio_info_cache = KMEM_CACHE(dm_rq_clone_bio_info, 0);
235 if (!_rq_bio_info_cache)
236 goto out_free_rq_tio_cache;
237
238 r = dm_uevent_init();
239 if (r)
240 goto out_free_rq_bio_info_cache;
241
242 _major = major;
243 r = register_blkdev(_major, _name);
244 if (r < 0)
245 goto out_uevent_exit;
246
247 if (!_major)
248 _major = r;
249
250 return 0;
251
252out_uevent_exit:
253 dm_uevent_exit();
254out_free_rq_bio_info_cache:
255 kmem_cache_destroy(_rq_bio_info_cache);
256out_free_rq_tio_cache:
257 kmem_cache_destroy(_rq_tio_cache);
258out_free_tio_cache:
259 kmem_cache_destroy(_tio_cache);
260out_free_io_cache:
261 kmem_cache_destroy(_io_cache);
262
263 return r;
264}
265
266static void local_exit(void)
267{
268 kmem_cache_destroy(_rq_bio_info_cache);
269 kmem_cache_destroy(_rq_tio_cache);
270 kmem_cache_destroy(_tio_cache);
271 kmem_cache_destroy(_io_cache);
272 unregister_blkdev(_major, _name);
273 dm_uevent_exit();
274
275 _major = 0;
276
277 DMINFO("cleaned up");
278}
279
280static int (*_inits[])(void) __initdata = {
281 local_init,
282 dm_target_init,
283 dm_linear_init,
284 dm_stripe_init,
285 dm_io_init,
286 dm_kcopyd_init,
287 dm_interface_init,
288};
289
290static void (*_exits[])(void) = {
291 local_exit,
292 dm_target_exit,
293 dm_linear_exit,
294 dm_stripe_exit,
295 dm_io_exit,
296 dm_kcopyd_exit,
297 dm_interface_exit,
298};
299
300static int __init dm_init(void)
301{
302 const int count = ARRAY_SIZE(_inits);
303
304 int r, i;
305
306 for (i = 0; i < count; i++) {
307 r = _inits[i]();
308 if (r)
309 goto bad;
310 }
311
312 return 0;
313
314 bad:
315 while (i--)
316 _exits[i]();
317
318 return r;
319}
320
321static void __exit dm_exit(void)
322{
323 int i = ARRAY_SIZE(_exits);
324
325 while (i--)
326 _exits[i]();
327
328 /*
329 * Should be empty by this point.
330 */
331 idr_remove_all(&_minor_idr);
332 idr_destroy(&_minor_idr);
333}
334
335/*
336 * Block device functions
337 */
338int dm_deleting_md(struct mapped_device *md)
339{
340 return test_bit(DMF_DELETING, &md->flags);
341}
342
343static int dm_blk_open(struct block_device *bdev, fmode_t mode)
344{
345 struct mapped_device *md;
346
347 spin_lock(&_minor_lock);
348
349 md = bdev->bd_disk->private_data;
350 if (!md)
351 goto out;
352
353 if (test_bit(DMF_FREEING, &md->flags) ||
354 dm_deleting_md(md)) {
355 md = NULL;
356 goto out;
357 }
358
359 dm_get(md);
360 atomic_inc(&md->open_count);
361
362out:
363 spin_unlock(&_minor_lock);
364
365 return md ? 0 : -ENXIO;
366}
367
368static int dm_blk_close(struct gendisk *disk, fmode_t mode)
369{
370 struct mapped_device *md = disk->private_data;
371
372 spin_lock(&_minor_lock);
373
374 atomic_dec(&md->open_count);
375 dm_put(md);
376
377 spin_unlock(&_minor_lock);
378
379 return 0;
380}
381
382int dm_open_count(struct mapped_device *md)
383{
384 return atomic_read(&md->open_count);
385}
386
387/*
388 * Guarantees nothing is using the device before it's deleted.
389 */
390int dm_lock_for_deletion(struct mapped_device *md)
391{
392 int r = 0;
393
394 spin_lock(&_minor_lock);
395
396 if (dm_open_count(md))
397 r = -EBUSY;
398 else
399 set_bit(DMF_DELETING, &md->flags);
400
401 spin_unlock(&_minor_lock);
402
403 return r;
404}
405
406static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
407{
408 struct mapped_device *md = bdev->bd_disk->private_data;
409
410 return dm_get_geometry(md, geo);
411}
412
413static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
414 unsigned int cmd, unsigned long arg)
415{
416 struct mapped_device *md = bdev->bd_disk->private_data;
417 struct dm_table *map = dm_get_live_table(md);
418 struct dm_target *tgt;
419 int r = -ENOTTY;
420
421 if (!map || !dm_table_get_size(map))
422 goto out;
423
424 /* We only support devices that have a single target */
425 if (dm_table_get_num_targets(map) != 1)
426 goto out;
427
428 tgt = dm_table_get_target(map, 0);
429
430 if (dm_suspended_md(md)) {
431 r = -EAGAIN;
432 goto out;
433 }
434
435 if (tgt->type->ioctl)
436 r = tgt->type->ioctl(tgt, cmd, arg);
437
438out:
439 dm_table_put(map);
440
441 return r;
442}
443
444static struct dm_io *alloc_io(struct mapped_device *md)
445{
446 return mempool_alloc(md->io_pool, GFP_NOIO);
447}
448
449static void free_io(struct mapped_device *md, struct dm_io *io)
450{
451 mempool_free(io, md->io_pool);
452}
453
454static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
455{
456 mempool_free(tio, md->tio_pool);
457}
458
459static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md,
460 gfp_t gfp_mask)
461{
462 return mempool_alloc(md->tio_pool, gfp_mask);
463}
464
465static void free_rq_tio(struct dm_rq_target_io *tio)
466{
467 mempool_free(tio, tio->md->tio_pool);
468}
469
470static struct dm_rq_clone_bio_info *alloc_bio_info(struct mapped_device *md)
471{
472 return mempool_alloc(md->io_pool, GFP_ATOMIC);
473}
474
475static void free_bio_info(struct dm_rq_clone_bio_info *info)
476{
477 mempool_free(info, info->tio->md->io_pool);
478}
479
480static int md_in_flight(struct mapped_device *md)
481{
482 return atomic_read(&md->pending[READ]) +
483 atomic_read(&md->pending[WRITE]);
484}
485
486static void start_io_acct(struct dm_io *io)
487{
488 struct mapped_device *md = io->md;
489 int cpu;
490 int rw = bio_data_dir(io->bio);
491
492 io->start_time = jiffies;
493
494 cpu = part_stat_lock();
495 part_round_stats(cpu, &dm_disk(md)->part0);
496 part_stat_unlock();
497 atomic_set(&dm_disk(md)->part0.in_flight[rw],
498 atomic_inc_return(&md->pending[rw]));
499}
500
501static void end_io_acct(struct dm_io *io)
502{
503 struct mapped_device *md = io->md;
504 struct bio *bio = io->bio;
505 unsigned long duration = jiffies - io->start_time;
506 int pending, cpu;
507 int rw = bio_data_dir(bio);
508
509 cpu = part_stat_lock();
510 part_round_stats(cpu, &dm_disk(md)->part0);
511 part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration);
512 part_stat_unlock();
513
514 /*
515 * After this is decremented the bio must not be touched if it is
516 * a flush.
517 */
518 pending = atomic_dec_return(&md->pending[rw]);
519 atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
520 pending += atomic_read(&md->pending[rw^0x1]);
521
522 /* nudge anyone waiting on suspend queue */
523 if (!pending)
524 wake_up(&md->wait);
525}
526
527/*
528 * Add the bio to the list of deferred io.
529 */
530static void queue_io(struct mapped_device *md, struct bio *bio)
531{
532 unsigned long flags;
533
534 spin_lock_irqsave(&md->deferred_lock, flags);
535 bio_list_add(&md->deferred, bio);
536 spin_unlock_irqrestore(&md->deferred_lock, flags);
537 queue_work(md->wq, &md->work);
538}
539
540/*
541 * Everyone (including functions in this file), should use this
542 * function to access the md->map field, and make sure they call
543 * dm_table_put() when finished.
544 */
545struct dm_table *dm_get_live_table(struct mapped_device *md)
546{
547 struct dm_table *t;
548 unsigned long flags;
549
550 read_lock_irqsave(&md->map_lock, flags);
551 t = md->map;
552 if (t)
553 dm_table_get(t);
554 read_unlock_irqrestore(&md->map_lock, flags);
555
556 return t;
557}
558
559/*
560 * Get the geometry associated with a dm device
561 */
562int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
563{
564 *geo = md->geometry;
565
566 return 0;
567}
568
569/*
570 * Set the geometry of a device.
571 */
572int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
573{
574 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
575
576 if (geo->start > sz) {
577 DMWARN("Start sector is beyond the geometry limits.");
578 return -EINVAL;
579 }
580
581 md->geometry = *geo;
582
583 return 0;
584}
585
586/*-----------------------------------------------------------------
587 * CRUD START:
588 * A more elegant soln is in the works that uses the queue
589 * merge fn, unfortunately there are a couple of changes to
590 * the block layer that I want to make for this. So in the
591 * interests of getting something for people to use I give
592 * you this clearly demarcated crap.
593 *---------------------------------------------------------------*/
594
595static int __noflush_suspending(struct mapped_device *md)
596{
597 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
598}
599
600/*
601 * Decrements the number of outstanding ios that a bio has been
602 * cloned into, completing the original io if necc.
603 */
604static void dec_pending(struct dm_io *io, int error)
605{
606 unsigned long flags;
607 int io_error;
608 struct bio *bio;
609 struct mapped_device *md = io->md;
610
611 /* Push-back supersedes any I/O errors */
612 if (unlikely(error)) {
613 spin_lock_irqsave(&io->endio_lock, flags);
614 if (!(io->error > 0 && __noflush_suspending(md)))
615 io->error = error;
616 spin_unlock_irqrestore(&io->endio_lock, flags);
617 }
618
619 if (atomic_dec_and_test(&io->io_count)) {
620 if (io->error == DM_ENDIO_REQUEUE) {
621 /*
622 * Target requested pushing back the I/O.
623 */
624 spin_lock_irqsave(&md->deferred_lock, flags);
625 if (__noflush_suspending(md))
626 bio_list_add_head(&md->deferred, io->bio);
627 else
628 /* noflush suspend was interrupted. */
629 io->error = -EIO;
630 spin_unlock_irqrestore(&md->deferred_lock, flags);
631 }
632
633 io_error = io->error;
634 bio = io->bio;
635 end_io_acct(io);
636 free_io(md, io);
637
638 if (io_error == DM_ENDIO_REQUEUE)
639 return;
640
641 if ((bio->bi_rw & REQ_FLUSH) && bio->bi_size) {
642 /*
643 * Preflush done for flush with data, reissue
644 * without REQ_FLUSH.
645 */
646 bio->bi_rw &= ~REQ_FLUSH;
647 queue_io(md, bio);
648 } else {
649 /* done with normal IO or empty flush */
650 trace_block_bio_complete(md->queue, bio, io_error);
651 bio_endio(bio, io_error);
652 }
653 }
654}
655
656static void clone_endio(struct bio *bio, int error)
657{
658 int r = 0;
659 struct dm_target_io *tio = bio->bi_private;
660 struct dm_io *io = tio->io;
661 struct mapped_device *md = tio->io->md;
662 dm_endio_fn endio = tio->ti->type->end_io;
663
664 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
665 error = -EIO;
666
667 if (endio) {
668 r = endio(tio->ti, bio, error, &tio->info);
669 if (r < 0 || r == DM_ENDIO_REQUEUE)
670 /*
671 * error and requeue request are handled
672 * in dec_pending().
673 */
674 error = r;
675 else if (r == DM_ENDIO_INCOMPLETE)
676 /* The target will handle the io */
677 return;
678 else if (r) {
679 DMWARN("unimplemented target endio return value: %d", r);
680 BUG();
681 }
682 }
683
684 /*
685 * Store md for cleanup instead of tio which is about to get freed.
686 */
687 bio->bi_private = md->bs;
688
689 free_tio(md, tio);
690 bio_put(bio);
691 dec_pending(io, error);
692}
693
694/*
695 * Partial completion handling for request-based dm
696 */
697static void end_clone_bio(struct bio *clone, int error)
698{
699 struct dm_rq_clone_bio_info *info = clone->bi_private;
700 struct dm_rq_target_io *tio = info->tio;
701 struct bio *bio = info->orig;
702 unsigned int nr_bytes = info->orig->bi_size;
703
704 bio_put(clone);
705
706 if (tio->error)
707 /*
708 * An error has already been detected on the request.
709 * Once error occurred, just let clone->end_io() handle
710 * the remainder.
711 */
712 return;
713 else if (error) {
714 /*
715 * Don't notice the error to the upper layer yet.
716 * The error handling decision is made by the target driver,
717 * when the request is completed.
718 */
719 tio->error = error;
720 return;
721 }
722
723 /*
724 * I/O for the bio successfully completed.
725 * Notice the data completion to the upper layer.
726 */
727
728 /*
729 * bios are processed from the head of the list.
730 * So the completing bio should always be rq->bio.
731 * If it's not, something wrong is happening.
732 */
733 if (tio->orig->bio != bio)
734 DMERR("bio completion is going in the middle of the request");
735
736 /*
737 * Update the original request.
738 * Do not use blk_end_request() here, because it may complete
739 * the original request before the clone, and break the ordering.
740 */
741 blk_update_request(tio->orig, 0, nr_bytes);
742}
743
744/*
745 * Don't touch any member of the md after calling this function because
746 * the md may be freed in dm_put() at the end of this function.
747 * Or do dm_get() before calling this function and dm_put() later.
748 */
749static void rq_completed(struct mapped_device *md, int rw, int run_queue)
750{
751 atomic_dec(&md->pending[rw]);
752
753 /* nudge anyone waiting on suspend queue */
754 if (!md_in_flight(md))
755 wake_up(&md->wait);
756
757 if (run_queue)
758 blk_run_queue(md->queue);
759
760 /*
761 * dm_put() must be at the end of this function. See the comment above
762 */
763 dm_put(md);
764}
765
766static void free_rq_clone(struct request *clone)
767{
768 struct dm_rq_target_io *tio = clone->end_io_data;
769
770 blk_rq_unprep_clone(clone);
771 free_rq_tio(tio);
772}
773
774/*
775 * Complete the clone and the original request.
776 * Must be called without queue lock.
777 */
778static void dm_end_request(struct request *clone, int error)
779{
780 int rw = rq_data_dir(clone);
781 struct dm_rq_target_io *tio = clone->end_io_data;
782 struct mapped_device *md = tio->md;
783 struct request *rq = tio->orig;
784
785 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
786 rq->errors = clone->errors;
787 rq->resid_len = clone->resid_len;
788
789 if (rq->sense)
790 /*
791 * We are using the sense buffer of the original
792 * request.
793 * So setting the length of the sense data is enough.
794 */
795 rq->sense_len = clone->sense_len;
796 }
797
798 free_rq_clone(clone);
799 blk_end_request_all(rq, error);
800 rq_completed(md, rw, true);
801}
802
803static void dm_unprep_request(struct request *rq)
804{
805 struct request *clone = rq->special;
806
807 rq->special = NULL;
808 rq->cmd_flags &= ~REQ_DONTPREP;
809
810 free_rq_clone(clone);
811}
812
813/*
814 * Requeue the original request of a clone.
815 */
816void dm_requeue_unmapped_request(struct request *clone)
817{
818 int rw = rq_data_dir(clone);
819 struct dm_rq_target_io *tio = clone->end_io_data;
820 struct mapped_device *md = tio->md;
821 struct request *rq = tio->orig;
822 struct request_queue *q = rq->q;
823 unsigned long flags;
824
825 dm_unprep_request(rq);
826
827 spin_lock_irqsave(q->queue_lock, flags);
828 blk_requeue_request(q, rq);
829 spin_unlock_irqrestore(q->queue_lock, flags);
830
831 rq_completed(md, rw, 0);
832}
833EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request);
834
835static void __stop_queue(struct request_queue *q)
836{
837 blk_stop_queue(q);
838}
839
840static void stop_queue(struct request_queue *q)
841{
842 unsigned long flags;
843
844 spin_lock_irqsave(q->queue_lock, flags);
845 __stop_queue(q);
846 spin_unlock_irqrestore(q->queue_lock, flags);
847}
848
849static void __start_queue(struct request_queue *q)
850{
851 if (blk_queue_stopped(q))
852 blk_start_queue(q);
853}
854
855static void start_queue(struct request_queue *q)
856{
857 unsigned long flags;
858
859 spin_lock_irqsave(q->queue_lock, flags);
860 __start_queue(q);
861 spin_unlock_irqrestore(q->queue_lock, flags);
862}
863
864static void dm_done(struct request *clone, int error, bool mapped)
865{
866 int r = error;
867 struct dm_rq_target_io *tio = clone->end_io_data;
868 dm_request_endio_fn rq_end_io = NULL;
869
870 if (tio->ti) {
871 rq_end_io = tio->ti->type->rq_end_io;
872
873 if (mapped && rq_end_io)
874 r = rq_end_io(tio->ti, clone, error, &tio->info);
875 }
876
877 if (r <= 0)
878 /* The target wants to complete the I/O */
879 dm_end_request(clone, r);
880 else if (r == DM_ENDIO_INCOMPLETE)
881 /* The target will handle the I/O */
882 return;
883 else if (r == DM_ENDIO_REQUEUE)
884 /* The target wants to requeue the I/O */
885 dm_requeue_unmapped_request(clone);
886 else {
887 DMWARN("unimplemented target endio return value: %d", r);
888 BUG();
889 }
890}
891
892/*
893 * Request completion handler for request-based dm
894 */
895static void dm_softirq_done(struct request *rq)
896{
897 bool mapped = true;
898 struct request *clone = rq->completion_data;
899 struct dm_rq_target_io *tio = clone->end_io_data;
900
901 if (rq->cmd_flags & REQ_FAILED)
902 mapped = false;
903
904 dm_done(clone, tio->error, mapped);
905}
906
907/*
908 * Complete the clone and the original request with the error status
909 * through softirq context.
910 */
911static void dm_complete_request(struct request *clone, int error)
912{
913 struct dm_rq_target_io *tio = clone->end_io_data;
914 struct request *rq = tio->orig;
915
916 tio->error = error;
917 rq->completion_data = clone;
918 blk_complete_request(rq);
919}
920
921/*
922 * Complete the not-mapped clone and the original request with the error status
923 * through softirq context.
924 * Target's rq_end_io() function isn't called.
925 * This may be used when the target's map_rq() function fails.
926 */
927void dm_kill_unmapped_request(struct request *clone, int error)
928{
929 struct dm_rq_target_io *tio = clone->end_io_data;
930 struct request *rq = tio->orig;
931
932 rq->cmd_flags |= REQ_FAILED;
933 dm_complete_request(clone, error);
934}
935EXPORT_SYMBOL_GPL(dm_kill_unmapped_request);
936
937/*
938 * Called with the queue lock held
939 */
940static void end_clone_request(struct request *clone, int error)
941{
942 /*
943 * For just cleaning up the information of the queue in which
944 * the clone was dispatched.
945 * The clone is *NOT* freed actually here because it is alloced from
946 * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags.
947 */
948 __blk_put_request(clone->q, clone);
949
950 /*
951 * Actual request completion is done in a softirq context which doesn't
952 * hold the queue lock. Otherwise, deadlock could occur because:
953 * - another request may be submitted by the upper level driver
954 * of the stacking during the completion
955 * - the submission which requires queue lock may be done
956 * against this queue
957 */
958 dm_complete_request(clone, error);
959}
960
961/*
962 * Return maximum size of I/O possible at the supplied sector up to the current
963 * target boundary.
964 */
965static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
966{
967 sector_t target_offset = dm_target_offset(ti, sector);
968
969 return ti->len - target_offset;
970}
971
972static sector_t max_io_len(sector_t sector, struct dm_target *ti)
973{
974 sector_t len = max_io_len_target_boundary(sector, ti);
975
976 /*
977 * Does the target need to split even further ?
978 */
979 if (ti->split_io) {
980 sector_t boundary;
981 sector_t offset = dm_target_offset(ti, sector);
982 boundary = ((offset + ti->split_io) & ~(ti->split_io - 1))
983 - offset;
984 if (len > boundary)
985 len = boundary;
986 }
987
988 return len;
989}
990
991static void __map_bio(struct dm_target *ti, struct bio *clone,
992 struct dm_target_io *tio)
993{
994 int r;
995 sector_t sector;
996 struct mapped_device *md;
997
998 clone->bi_end_io = clone_endio;
999 clone->bi_private = tio;
1000
1001 /*
1002 * Map the clone. If r == 0 we don't need to do
1003 * anything, the target has assumed ownership of
1004 * this io.
1005 */
1006 atomic_inc(&tio->io->io_count);
1007 sector = clone->bi_sector;
1008 r = ti->type->map(ti, clone, &tio->info);
1009 if (r == DM_MAPIO_REMAPPED) {
1010 /* the bio has been remapped so dispatch it */
1011
1012 trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone,
1013 tio->io->bio->bi_bdev->bd_dev, sector);
1014
1015 generic_make_request(clone);
1016 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
1017 /* error the io and bail out, or requeue it if needed */
1018 md = tio->io->md;
1019 dec_pending(tio->io, r);
1020 /*
1021 * Store bio_set for cleanup.
1022 */
1023 clone->bi_end_io = NULL;
1024 clone->bi_private = md->bs;
1025 bio_put(clone);
1026 free_tio(md, tio);
1027 } else if (r) {
1028 DMWARN("unimplemented target map return value: %d", r);
1029 BUG();
1030 }
1031}
1032
1033struct clone_info {
1034 struct mapped_device *md;
1035 struct dm_table *map;
1036 struct bio *bio;
1037 struct dm_io *io;
1038 sector_t sector;
1039 sector_t sector_count;
1040 unsigned short idx;
1041};
1042
1043static void dm_bio_destructor(struct bio *bio)
1044{
1045 struct bio_set *bs = bio->bi_private;
1046
1047 bio_free(bio, bs);
1048}
1049
1050/*
1051 * Creates a little bio that just does part of a bvec.
1052 */
1053static struct bio *split_bvec(struct bio *bio, sector_t sector,
1054 unsigned short idx, unsigned int offset,
1055 unsigned int len, struct bio_set *bs)
1056{
1057 struct bio *clone;
1058 struct bio_vec *bv = bio->bi_io_vec + idx;
1059
1060 clone = bio_alloc_bioset(GFP_NOIO, 1, bs);
1061 clone->bi_destructor = dm_bio_destructor;
1062 *clone->bi_io_vec = *bv;
1063
1064 clone->bi_sector = sector;
1065 clone->bi_bdev = bio->bi_bdev;
1066 clone->bi_rw = bio->bi_rw;
1067 clone->bi_vcnt = 1;
1068 clone->bi_size = to_bytes(len);
1069 clone->bi_io_vec->bv_offset = offset;
1070 clone->bi_io_vec->bv_len = clone->bi_size;
1071 clone->bi_flags |= 1 << BIO_CLONED;
1072
1073 if (bio_integrity(bio)) {
1074 bio_integrity_clone(clone, bio, GFP_NOIO, bs);
1075 bio_integrity_trim(clone,
1076 bio_sector_offset(bio, idx, offset), len);
1077 }
1078
1079 return clone;
1080}
1081
1082/*
1083 * Creates a bio that consists of range of complete bvecs.
1084 */
1085static struct bio *clone_bio(struct bio *bio, sector_t sector,
1086 unsigned short idx, unsigned short bv_count,
1087 unsigned int len, struct bio_set *bs)
1088{
1089 struct bio *clone;
1090
1091 clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs);
1092 __bio_clone(clone, bio);
1093 clone->bi_destructor = dm_bio_destructor;
1094 clone->bi_sector = sector;
1095 clone->bi_idx = idx;
1096 clone->bi_vcnt = idx + bv_count;
1097 clone->bi_size = to_bytes(len);
1098 clone->bi_flags &= ~(1 << BIO_SEG_VALID);
1099
1100 if (bio_integrity(bio)) {
1101 bio_integrity_clone(clone, bio, GFP_NOIO, bs);
1102
1103 if (idx != bio->bi_idx || clone->bi_size < bio->bi_size)
1104 bio_integrity_trim(clone,
1105 bio_sector_offset(bio, idx, 0), len);
1106 }
1107
1108 return clone;
1109}
1110
1111static struct dm_target_io *alloc_tio(struct clone_info *ci,
1112 struct dm_target *ti)
1113{
1114 struct dm_target_io *tio = mempool_alloc(ci->md->tio_pool, GFP_NOIO);
1115
1116 tio->io = ci->io;
1117 tio->ti = ti;
1118 memset(&tio->info, 0, sizeof(tio->info));
1119
1120 return tio;
1121}
1122
1123static void __issue_target_request(struct clone_info *ci, struct dm_target *ti,
1124 unsigned request_nr, sector_t len)
1125{
1126 struct dm_target_io *tio = alloc_tio(ci, ti);
1127 struct bio *clone;
1128
1129 tio->info.target_request_nr = request_nr;
1130
1131 /*
1132 * Discard requests require the bio's inline iovecs be initialized.
1133 * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush
1134 * and discard, so no need for concern about wasted bvec allocations.
1135 */
1136 clone = bio_alloc_bioset(GFP_NOIO, ci->bio->bi_max_vecs, ci->md->bs);
1137 __bio_clone(clone, ci->bio);
1138 clone->bi_destructor = dm_bio_destructor;
1139 if (len) {
1140 clone->bi_sector = ci->sector;
1141 clone->bi_size = to_bytes(len);
1142 }
1143
1144 __map_bio(ti, clone, tio);
1145}
1146
1147static void __issue_target_requests(struct clone_info *ci, struct dm_target *ti,
1148 unsigned num_requests, sector_t len)
1149{
1150 unsigned request_nr;
1151
1152 for (request_nr = 0; request_nr < num_requests; request_nr++)
1153 __issue_target_request(ci, ti, request_nr, len);
1154}
1155
1156static int __clone_and_map_empty_flush(struct clone_info *ci)
1157{
1158 unsigned target_nr = 0;
1159 struct dm_target *ti;
1160
1161 BUG_ON(bio_has_data(ci->bio));
1162 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1163 __issue_target_requests(ci, ti, ti->num_flush_requests, 0);
1164
1165 return 0;
1166}
1167
1168/*
1169 * Perform all io with a single clone.
1170 */
1171static void __clone_and_map_simple(struct clone_info *ci, struct dm_target *ti)
1172{
1173 struct bio *clone, *bio = ci->bio;
1174 struct dm_target_io *tio;
1175
1176 tio = alloc_tio(ci, ti);
1177 clone = clone_bio(bio, ci->sector, ci->idx,
1178 bio->bi_vcnt - ci->idx, ci->sector_count,
1179 ci->md->bs);
1180 __map_bio(ti, clone, tio);
1181 ci->sector_count = 0;
1182}
1183
1184static int __clone_and_map_discard(struct clone_info *ci)
1185{
1186 struct dm_target *ti;
1187 sector_t len;
1188
1189 do {
1190 ti = dm_table_find_target(ci->map, ci->sector);
1191 if (!dm_target_is_valid(ti))
1192 return -EIO;
1193
1194 /*
1195 * Even though the device advertised discard support,
1196 * that does not mean every target supports it, and
1197 * reconfiguration might also have changed that since the
1198 * check was performed.
1199 */
1200 if (!ti->num_discard_requests)
1201 return -EOPNOTSUPP;
1202
1203 len = min(ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1204
1205 __issue_target_requests(ci, ti, ti->num_discard_requests, len);
1206
1207 ci->sector += len;
1208 } while (ci->sector_count -= len);
1209
1210 return 0;
1211}
1212
1213static int __clone_and_map(struct clone_info *ci)
1214{
1215 struct bio *clone, *bio = ci->bio;
1216 struct dm_target *ti;
1217 sector_t len = 0, max;
1218 struct dm_target_io *tio;
1219
1220 if (unlikely(bio->bi_rw & REQ_DISCARD))
1221 return __clone_and_map_discard(ci);
1222
1223 ti = dm_table_find_target(ci->map, ci->sector);
1224 if (!dm_target_is_valid(ti))
1225 return -EIO;
1226
1227 max = max_io_len(ci->sector, ti);
1228
1229 if (ci->sector_count <= max) {
1230 /*
1231 * Optimise for the simple case where we can do all of
1232 * the remaining io with a single clone.
1233 */
1234 __clone_and_map_simple(ci, ti);
1235
1236 } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
1237 /*
1238 * There are some bvecs that don't span targets.
1239 * Do as many of these as possible.
1240 */
1241 int i;
1242 sector_t remaining = max;
1243 sector_t bv_len;
1244
1245 for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
1246 bv_len = to_sector(bio->bi_io_vec[i].bv_len);
1247
1248 if (bv_len > remaining)
1249 break;
1250
1251 remaining -= bv_len;
1252 len += bv_len;
1253 }
1254
1255 tio = alloc_tio(ci, ti);
1256 clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len,
1257 ci->md->bs);
1258 __map_bio(ti, clone, tio);
1259
1260 ci->sector += len;
1261 ci->sector_count -= len;
1262 ci->idx = i;
1263
1264 } else {
1265 /*
1266 * Handle a bvec that must be split between two or more targets.
1267 */
1268 struct bio_vec *bv = bio->bi_io_vec + ci->idx;
1269 sector_t remaining = to_sector(bv->bv_len);
1270 unsigned int offset = 0;
1271
1272 do {
1273 if (offset) {
1274 ti = dm_table_find_target(ci->map, ci->sector);
1275 if (!dm_target_is_valid(ti))
1276 return -EIO;
1277
1278 max = max_io_len(ci->sector, ti);
1279 }
1280
1281 len = min(remaining, max);
1282
1283 tio = alloc_tio(ci, ti);
1284 clone = split_bvec(bio, ci->sector, ci->idx,
1285 bv->bv_offset + offset, len,
1286 ci->md->bs);
1287
1288 __map_bio(ti, clone, tio);
1289
1290 ci->sector += len;
1291 ci->sector_count -= len;
1292 offset += to_bytes(len);
1293 } while (remaining -= len);
1294
1295 ci->idx++;
1296 }
1297
1298 return 0;
1299}
1300
1301/*
1302 * Split the bio into several clones and submit it to targets.
1303 */
1304static void __split_and_process_bio(struct mapped_device *md, struct bio *bio)
1305{
1306 struct clone_info ci;
1307 int error = 0;
1308
1309 ci.map = dm_get_live_table(md);
1310 if (unlikely(!ci.map)) {
1311 bio_io_error(bio);
1312 return;
1313 }
1314
1315 ci.md = md;
1316 ci.io = alloc_io(md);
1317 ci.io->error = 0;
1318 atomic_set(&ci.io->io_count, 1);
1319 ci.io->bio = bio;
1320 ci.io->md = md;
1321 spin_lock_init(&ci.io->endio_lock);
1322 ci.sector = bio->bi_sector;
1323 ci.idx = bio->bi_idx;
1324
1325 start_io_acct(ci.io);
1326 if (bio->bi_rw & REQ_FLUSH) {
1327 ci.bio = &ci.md->flush_bio;
1328 ci.sector_count = 0;
1329 error = __clone_and_map_empty_flush(&ci);
1330 /* dec_pending submits any data associated with flush */
1331 } else {
1332 ci.bio = bio;
1333 ci.sector_count = bio_sectors(bio);
1334 while (ci.sector_count && !error)
1335 error = __clone_and_map(&ci);
1336 }
1337
1338 /* drop the extra reference count */
1339 dec_pending(ci.io, error);
1340 dm_table_put(ci.map);
1341}
1342/*-----------------------------------------------------------------
1343 * CRUD END
1344 *---------------------------------------------------------------*/
1345
1346static int dm_merge_bvec(struct request_queue *q,
1347 struct bvec_merge_data *bvm,
1348 struct bio_vec *biovec)
1349{
1350 struct mapped_device *md = q->queuedata;
1351 struct dm_table *map = dm_get_live_table(md);
1352 struct dm_target *ti;
1353 sector_t max_sectors;
1354 int max_size = 0;
1355
1356 if (unlikely(!map))
1357 goto out;
1358
1359 ti = dm_table_find_target(map, bvm->bi_sector);
1360 if (!dm_target_is_valid(ti))
1361 goto out_table;
1362
1363 /*
1364 * Find maximum amount of I/O that won't need splitting
1365 */
1366 max_sectors = min(max_io_len(bvm->bi_sector, ti),
1367 (sector_t) BIO_MAX_SECTORS);
1368 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
1369 if (max_size < 0)
1370 max_size = 0;
1371
1372 /*
1373 * merge_bvec_fn() returns number of bytes
1374 * it can accept at this offset
1375 * max is precomputed maximal io size
1376 */
1377 if (max_size && ti->type->merge)
1378 max_size = ti->type->merge(ti, bvm, biovec, max_size);
1379 /*
1380 * If the target doesn't support merge method and some of the devices
1381 * provided their merge_bvec method (we know this by looking at
1382 * queue_max_hw_sectors), then we can't allow bios with multiple vector
1383 * entries. So always set max_size to 0, and the code below allows
1384 * just one page.
1385 */
1386 else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
1387
1388 max_size = 0;
1389
1390out_table:
1391 dm_table_put(map);
1392
1393out:
1394 /*
1395 * Always allow an entire first page
1396 */
1397 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
1398 max_size = biovec->bv_len;
1399
1400 return max_size;
1401}
1402
1403/*
1404 * The request function that just remaps the bio built up by
1405 * dm_merge_bvec.
1406 */
1407static void _dm_request(struct request_queue *q, struct bio *bio)
1408{
1409 int rw = bio_data_dir(bio);
1410 struct mapped_device *md = q->queuedata;
1411 int cpu;
1412
1413 down_read(&md->io_lock);
1414
1415 cpu = part_stat_lock();
1416 part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
1417 part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
1418 part_stat_unlock();
1419
1420 /* if we're suspended, we have to queue this io for later */
1421 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1422 up_read(&md->io_lock);
1423
1424 if (bio_rw(bio) != READA)
1425 queue_io(md, bio);
1426 else
1427 bio_io_error(bio);
1428 return;
1429 }
1430
1431 __split_and_process_bio(md, bio);
1432 up_read(&md->io_lock);
1433 return;
1434}
1435
1436static int dm_request_based(struct mapped_device *md)
1437{
1438 return blk_queue_stackable(md->queue);
1439}
1440
1441static void dm_request(struct request_queue *q, struct bio *bio)
1442{
1443 struct mapped_device *md = q->queuedata;
1444
1445 if (dm_request_based(md))
1446 blk_queue_bio(q, bio);
1447 else
1448 _dm_request(q, bio);
1449}
1450
1451void dm_dispatch_request(struct request *rq)
1452{
1453 int r;
1454
1455 if (blk_queue_io_stat(rq->q))
1456 rq->cmd_flags |= REQ_IO_STAT;
1457
1458 rq->start_time = jiffies;
1459 r = blk_insert_cloned_request(rq->q, rq);
1460 if (r)
1461 dm_complete_request(rq, r);
1462}
1463EXPORT_SYMBOL_GPL(dm_dispatch_request);
1464
1465static void dm_rq_bio_destructor(struct bio *bio)
1466{
1467 struct dm_rq_clone_bio_info *info = bio->bi_private;
1468 struct mapped_device *md = info->tio->md;
1469
1470 free_bio_info(info);
1471 bio_free(bio, md->bs);
1472}
1473
1474static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
1475 void *data)
1476{
1477 struct dm_rq_target_io *tio = data;
1478 struct mapped_device *md = tio->md;
1479 struct dm_rq_clone_bio_info *info = alloc_bio_info(md);
1480
1481 if (!info)
1482 return -ENOMEM;
1483
1484 info->orig = bio_orig;
1485 info->tio = tio;
1486 bio->bi_end_io = end_clone_bio;
1487 bio->bi_private = info;
1488 bio->bi_destructor = dm_rq_bio_destructor;
1489
1490 return 0;
1491}
1492
1493static int setup_clone(struct request *clone, struct request *rq,
1494 struct dm_rq_target_io *tio)
1495{
1496 int r;
1497
1498 r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC,
1499 dm_rq_bio_constructor, tio);
1500 if (r)
1501 return r;
1502
1503 clone->cmd = rq->cmd;
1504 clone->cmd_len = rq->cmd_len;
1505 clone->sense = rq->sense;
1506 clone->buffer = rq->buffer;
1507 clone->end_io = end_clone_request;
1508 clone->end_io_data = tio;
1509
1510 return 0;
1511}
1512
1513static struct request *clone_rq(struct request *rq, struct mapped_device *md,
1514 gfp_t gfp_mask)
1515{
1516 struct request *clone;
1517 struct dm_rq_target_io *tio;
1518
1519 tio = alloc_rq_tio(md, gfp_mask);
1520 if (!tio)
1521 return NULL;
1522
1523 tio->md = md;
1524 tio->ti = NULL;
1525 tio->orig = rq;
1526 tio->error = 0;
1527 memset(&tio->info, 0, sizeof(tio->info));
1528
1529 clone = &tio->clone;
1530 if (setup_clone(clone, rq, tio)) {
1531 /* -ENOMEM */
1532 free_rq_tio(tio);
1533 return NULL;
1534 }
1535
1536 return clone;
1537}
1538
1539/*
1540 * Called with the queue lock held.
1541 */
1542static int dm_prep_fn(struct request_queue *q, struct request *rq)
1543{
1544 struct mapped_device *md = q->queuedata;
1545 struct request *clone;
1546
1547 if (unlikely(rq->special)) {
1548 DMWARN("Already has something in rq->special.");
1549 return BLKPREP_KILL;
1550 }
1551
1552 clone = clone_rq(rq, md, GFP_ATOMIC);
1553 if (!clone)
1554 return BLKPREP_DEFER;
1555
1556 rq->special = clone;
1557 rq->cmd_flags |= REQ_DONTPREP;
1558
1559 return BLKPREP_OK;
1560}
1561
1562/*
1563 * Returns:
1564 * 0 : the request has been processed (not requeued)
1565 * !0 : the request has been requeued
1566 */
1567static int map_request(struct dm_target *ti, struct request *clone,
1568 struct mapped_device *md)
1569{
1570 int r, requeued = 0;
1571 struct dm_rq_target_io *tio = clone->end_io_data;
1572
1573 tio->ti = ti;
1574 r = ti->type->map_rq(ti, clone, &tio->info);
1575 switch (r) {
1576 case DM_MAPIO_SUBMITTED:
1577 /* The target has taken the I/O to submit by itself later */
1578 break;
1579 case DM_MAPIO_REMAPPED:
1580 /* The target has remapped the I/O so dispatch it */
1581 trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)),
1582 blk_rq_pos(tio->orig));
1583 dm_dispatch_request(clone);
1584 break;
1585 case DM_MAPIO_REQUEUE:
1586 /* The target wants to requeue the I/O */
1587 dm_requeue_unmapped_request(clone);
1588 requeued = 1;
1589 break;
1590 default:
1591 if (r > 0) {
1592 DMWARN("unimplemented target map return value: %d", r);
1593 BUG();
1594 }
1595
1596 /* The target wants to complete the I/O */
1597 dm_kill_unmapped_request(clone, r);
1598 break;
1599 }
1600
1601 return requeued;
1602}
1603
1604static struct request *dm_start_request(struct mapped_device *md, struct request *orig)
1605{
1606 struct request *clone;
1607
1608 blk_start_request(orig);
1609 clone = orig->special;
1610 atomic_inc(&md->pending[rq_data_dir(clone)]);
1611
1612 /*
1613 * Hold the md reference here for the in-flight I/O.
1614 * We can't rely on the reference count by device opener,
1615 * because the device may be closed during the request completion
1616 * when all bios are completed.
1617 * See the comment in rq_completed() too.
1618 */
1619 dm_get(md);
1620
1621 return clone;
1622}
1623
1624/*
1625 * q->request_fn for request-based dm.
1626 * Called with the queue lock held.
1627 */
1628static void dm_request_fn(struct request_queue *q)
1629{
1630 struct mapped_device *md = q->queuedata;
1631 struct dm_table *map = dm_get_live_table(md);
1632 struct dm_target *ti;
1633 struct request *rq, *clone;
1634 sector_t pos;
1635
1636 /*
1637 * For suspend, check blk_queue_stopped() and increment
1638 * ->pending within a single queue_lock not to increment the
1639 * number of in-flight I/Os after the queue is stopped in
1640 * dm_suspend().
1641 */
1642 while (!blk_queue_stopped(q)) {
1643 rq = blk_peek_request(q);
1644 if (!rq)
1645 goto delay_and_out;
1646
1647 /* always use block 0 to find the target for flushes for now */
1648 pos = 0;
1649 if (!(rq->cmd_flags & REQ_FLUSH))
1650 pos = blk_rq_pos(rq);
1651
1652 ti = dm_table_find_target(map, pos);
1653 if (!dm_target_is_valid(ti)) {
1654 /*
1655 * Must perform setup, that dm_done() requires,
1656 * before calling dm_kill_unmapped_request
1657 */
1658 DMERR_LIMIT("request attempted access beyond the end of device");
1659 clone = dm_start_request(md, rq);
1660 dm_kill_unmapped_request(clone, -EIO);
1661 continue;
1662 }
1663
1664 if (ti->type->busy && ti->type->busy(ti))
1665 goto delay_and_out;
1666
1667 clone = dm_start_request(md, rq);
1668
1669 spin_unlock(q->queue_lock);
1670 if (map_request(ti, clone, md))
1671 goto requeued;
1672
1673 BUG_ON(!irqs_disabled());
1674 spin_lock(q->queue_lock);
1675 }
1676
1677 goto out;
1678
1679requeued:
1680 BUG_ON(!irqs_disabled());
1681 spin_lock(q->queue_lock);
1682
1683delay_and_out:
1684 blk_delay_queue(q, HZ / 10);
1685out:
1686 dm_table_put(map);
1687}
1688
1689int dm_underlying_device_busy(struct request_queue *q)
1690{
1691 return blk_lld_busy(q);
1692}
1693EXPORT_SYMBOL_GPL(dm_underlying_device_busy);
1694
1695static int dm_lld_busy(struct request_queue *q)
1696{
1697 int r;
1698 struct mapped_device *md = q->queuedata;
1699 struct dm_table *map = dm_get_live_table(md);
1700
1701 if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))
1702 r = 1;
1703 else
1704 r = dm_table_any_busy_target(map);
1705
1706 dm_table_put(map);
1707
1708 return r;
1709}
1710
1711static int dm_any_congested(void *congested_data, int bdi_bits)
1712{
1713 int r = bdi_bits;
1714 struct mapped_device *md = congested_data;
1715 struct dm_table *map;
1716
1717 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1718 map = dm_get_live_table(md);
1719 if (map) {
1720 /*
1721 * Request-based dm cares about only own queue for
1722 * the query about congestion status of request_queue
1723 */
1724 if (dm_request_based(md))
1725 r = md->queue->backing_dev_info.state &
1726 bdi_bits;
1727 else
1728 r = dm_table_any_congested(map, bdi_bits);
1729
1730 dm_table_put(map);
1731 }
1732 }
1733
1734 return r;
1735}
1736
1737/*-----------------------------------------------------------------
1738 * An IDR is used to keep track of allocated minor numbers.
1739 *---------------------------------------------------------------*/
1740static void free_minor(int minor)
1741{
1742 spin_lock(&_minor_lock);
1743 idr_remove(&_minor_idr, minor);
1744 spin_unlock(&_minor_lock);
1745}
1746
1747/*
1748 * See if the device with a specific minor # is free.
1749 */
1750static int specific_minor(int minor)
1751{
1752 int r, m;
1753
1754 if (minor >= (1 << MINORBITS))
1755 return -EINVAL;
1756
1757 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1758 if (!r)
1759 return -ENOMEM;
1760
1761 spin_lock(&_minor_lock);
1762
1763 if (idr_find(&_minor_idr, minor)) {
1764 r = -EBUSY;
1765 goto out;
1766 }
1767
1768 r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m);
1769 if (r)
1770 goto out;
1771
1772 if (m != minor) {
1773 idr_remove(&_minor_idr, m);
1774 r = -EBUSY;
1775 goto out;
1776 }
1777
1778out:
1779 spin_unlock(&_minor_lock);
1780 return r;
1781}
1782
1783static int next_free_minor(int *minor)
1784{
1785 int r, m;
1786
1787 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1788 if (!r)
1789 return -ENOMEM;
1790
1791 spin_lock(&_minor_lock);
1792
1793 r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m);
1794 if (r)
1795 goto out;
1796
1797 if (m >= (1 << MINORBITS)) {
1798 idr_remove(&_minor_idr, m);
1799 r = -ENOSPC;
1800 goto out;
1801 }
1802
1803 *minor = m;
1804
1805out:
1806 spin_unlock(&_minor_lock);
1807 return r;
1808}
1809
1810static const struct block_device_operations dm_blk_dops;
1811
1812static void dm_wq_work(struct work_struct *work);
1813
1814static void dm_init_md_queue(struct mapped_device *md)
1815{
1816 /*
1817 * Request-based dm devices cannot be stacked on top of bio-based dm
1818 * devices. The type of this dm device has not been decided yet.
1819 * The type is decided at the first table loading time.
1820 * To prevent problematic device stacking, clear the queue flag
1821 * for request stacking support until then.
1822 *
1823 * This queue is new, so no concurrency on the queue_flags.
1824 */
1825 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
1826
1827 md->queue->queuedata = md;
1828 md->queue->backing_dev_info.congested_fn = dm_any_congested;
1829 md->queue->backing_dev_info.congested_data = md;
1830 blk_queue_make_request(md->queue, dm_request);
1831 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1832 blk_queue_merge_bvec(md->queue, dm_merge_bvec);
1833}
1834
1835/*
1836 * Allocate and initialise a blank device with a given minor.
1837 */
1838static struct mapped_device *alloc_dev(int minor)
1839{
1840 int r;
1841 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
1842 void *old_md;
1843
1844 if (!md) {
1845 DMWARN("unable to allocate device, out of memory.");
1846 return NULL;
1847 }
1848
1849 if (!try_module_get(THIS_MODULE))
1850 goto bad_module_get;
1851
1852 /* get a minor number for the dev */
1853 if (minor == DM_ANY_MINOR)
1854 r = next_free_minor(&minor);
1855 else
1856 r = specific_minor(minor);
1857 if (r < 0)
1858 goto bad_minor;
1859
1860 md->type = DM_TYPE_NONE;
1861 init_rwsem(&md->io_lock);
1862 mutex_init(&md->suspend_lock);
1863 mutex_init(&md->type_lock);
1864 spin_lock_init(&md->deferred_lock);
1865 rwlock_init(&md->map_lock);
1866 atomic_set(&md->holders, 1);
1867 atomic_set(&md->open_count, 0);
1868 atomic_set(&md->event_nr, 0);
1869 atomic_set(&md->uevent_seq, 0);
1870 INIT_LIST_HEAD(&md->uevent_list);
1871 spin_lock_init(&md->uevent_lock);
1872
1873 md->queue = blk_alloc_queue(GFP_KERNEL);
1874 if (!md->queue)
1875 goto bad_queue;
1876
1877 dm_init_md_queue(md);
1878
1879 md->disk = alloc_disk(1);
1880 if (!md->disk)
1881 goto bad_disk;
1882
1883 atomic_set(&md->pending[0], 0);
1884 atomic_set(&md->pending[1], 0);
1885 init_waitqueue_head(&md->wait);
1886 INIT_WORK(&md->work, dm_wq_work);
1887 init_waitqueue_head(&md->eventq);
1888
1889 md->disk->major = _major;
1890 md->disk->first_minor = minor;
1891 md->disk->fops = &dm_blk_dops;
1892 md->disk->queue = md->queue;
1893 md->disk->private_data = md;
1894 sprintf(md->disk->disk_name, "dm-%d", minor);
1895 add_disk(md->disk);
1896 format_dev_t(md->name, MKDEV(_major, minor));
1897
1898 md->wq = alloc_workqueue("kdmflush",
1899 WQ_NON_REENTRANT | WQ_MEM_RECLAIM, 0);
1900 if (!md->wq)
1901 goto bad_thread;
1902
1903 md->bdev = bdget_disk(md->disk, 0);
1904 if (!md->bdev)
1905 goto bad_bdev;
1906
1907 bio_init(&md->flush_bio);
1908 md->flush_bio.bi_bdev = md->bdev;
1909 md->flush_bio.bi_rw = WRITE_FLUSH;
1910
1911 /* Populate the mapping, nobody knows we exist yet */
1912 spin_lock(&_minor_lock);
1913 old_md = idr_replace(&_minor_idr, md, minor);
1914 spin_unlock(&_minor_lock);
1915
1916 BUG_ON(old_md != MINOR_ALLOCED);
1917
1918 return md;
1919
1920bad_bdev:
1921 destroy_workqueue(md->wq);
1922bad_thread:
1923 del_gendisk(md->disk);
1924 put_disk(md->disk);
1925bad_disk:
1926 blk_cleanup_queue(md->queue);
1927bad_queue:
1928 free_minor(minor);
1929bad_minor:
1930 module_put(THIS_MODULE);
1931bad_module_get:
1932 kfree(md);
1933 return NULL;
1934}
1935
1936static void unlock_fs(struct mapped_device *md);
1937
1938static void free_dev(struct mapped_device *md)
1939{
1940 int minor = MINOR(disk_devt(md->disk));
1941
1942 unlock_fs(md);
1943 bdput(md->bdev);
1944 destroy_workqueue(md->wq);
1945 if (md->tio_pool)
1946 mempool_destroy(md->tio_pool);
1947 if (md->io_pool)
1948 mempool_destroy(md->io_pool);
1949 if (md->bs)
1950 bioset_free(md->bs);
1951 blk_integrity_unregister(md->disk);
1952 del_gendisk(md->disk);
1953 free_minor(minor);
1954
1955 spin_lock(&_minor_lock);
1956 md->disk->private_data = NULL;
1957 spin_unlock(&_minor_lock);
1958
1959 put_disk(md->disk);
1960 blk_cleanup_queue(md->queue);
1961 module_put(THIS_MODULE);
1962 kfree(md);
1963}
1964
1965static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
1966{
1967 struct dm_md_mempools *p;
1968
1969 if (md->io_pool && md->tio_pool && md->bs)
1970 /* the md already has necessary mempools */
1971 goto out;
1972
1973 p = dm_table_get_md_mempools(t);
1974 BUG_ON(!p || md->io_pool || md->tio_pool || md->bs);
1975
1976 md->io_pool = p->io_pool;
1977 p->io_pool = NULL;
1978 md->tio_pool = p->tio_pool;
1979 p->tio_pool = NULL;
1980 md->bs = p->bs;
1981 p->bs = NULL;
1982
1983out:
1984 /* mempool bind completed, now no need any mempools in the table */
1985 dm_table_free_md_mempools(t);
1986}
1987
1988/*
1989 * Bind a table to the device.
1990 */
1991static void event_callback(void *context)
1992{
1993 unsigned long flags;
1994 LIST_HEAD(uevents);
1995 struct mapped_device *md = (struct mapped_device *) context;
1996
1997 spin_lock_irqsave(&md->uevent_lock, flags);
1998 list_splice_init(&md->uevent_list, &uevents);
1999 spin_unlock_irqrestore(&md->uevent_lock, flags);
2000
2001 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2002
2003 atomic_inc(&md->event_nr);
2004 wake_up(&md->eventq);
2005}
2006
2007/*
2008 * Protected by md->suspend_lock obtained by dm_swap_table().
2009 */
2010static void __set_size(struct mapped_device *md, sector_t size)
2011{
2012 set_capacity(md->disk, size);
2013
2014 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2015}
2016
2017/*
2018 * Return 1 if the queue has a compulsory merge_bvec_fn function.
2019 *
2020 * If this function returns 0, then the device is either a non-dm
2021 * device without a merge_bvec_fn, or it is a dm device that is
2022 * able to split any bios it receives that are too big.
2023 */
2024int dm_queue_merge_is_compulsory(struct request_queue *q)
2025{
2026 struct mapped_device *dev_md;
2027
2028 if (!q->merge_bvec_fn)
2029 return 0;
2030
2031 if (q->make_request_fn == dm_request) {
2032 dev_md = q->queuedata;
2033 if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags))
2034 return 0;
2035 }
2036
2037 return 1;
2038}
2039
2040static int dm_device_merge_is_compulsory(struct dm_target *ti,
2041 struct dm_dev *dev, sector_t start,
2042 sector_t len, void *data)
2043{
2044 struct block_device *bdev = dev->bdev;
2045 struct request_queue *q = bdev_get_queue(bdev);
2046
2047 return dm_queue_merge_is_compulsory(q);
2048}
2049
2050/*
2051 * Return 1 if it is acceptable to ignore merge_bvec_fn based
2052 * on the properties of the underlying devices.
2053 */
2054static int dm_table_merge_is_optional(struct dm_table *table)
2055{
2056 unsigned i = 0;
2057 struct dm_target *ti;
2058
2059 while (i < dm_table_get_num_targets(table)) {
2060 ti = dm_table_get_target(table, i++);
2061
2062 if (ti->type->iterate_devices &&
2063 ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL))
2064 return 0;
2065 }
2066
2067 return 1;
2068}
2069
2070/*
2071 * Returns old map, which caller must destroy.
2072 */
2073static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2074 struct queue_limits *limits)
2075{
2076 struct dm_table *old_map;
2077 struct request_queue *q = md->queue;
2078 sector_t size;
2079 unsigned long flags;
2080 int merge_is_optional;
2081
2082 size = dm_table_get_size(t);
2083
2084 /*
2085 * Wipe any geometry if the size of the table changed.
2086 */
2087 if (size != get_capacity(md->disk))
2088 memset(&md->geometry, 0, sizeof(md->geometry));
2089
2090 __set_size(md, size);
2091
2092 dm_table_event_callback(t, event_callback, md);
2093
2094 /*
2095 * The queue hasn't been stopped yet, if the old table type wasn't
2096 * for request-based during suspension. So stop it to prevent
2097 * I/O mapping before resume.
2098 * This must be done before setting the queue restrictions,
2099 * because request-based dm may be run just after the setting.
2100 */
2101 if (dm_table_request_based(t) && !blk_queue_stopped(q))
2102 stop_queue(q);
2103
2104 __bind_mempools(md, t);
2105
2106 merge_is_optional = dm_table_merge_is_optional(t);
2107
2108 write_lock_irqsave(&md->map_lock, flags);
2109 old_map = md->map;
2110 md->map = t;
2111 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2112
2113 dm_table_set_restrictions(t, q, limits);
2114 if (merge_is_optional)
2115 set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2116 else
2117 clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2118 write_unlock_irqrestore(&md->map_lock, flags);
2119
2120 return old_map;
2121}
2122
2123/*
2124 * Returns unbound table for the caller to free.
2125 */
2126static struct dm_table *__unbind(struct mapped_device *md)
2127{
2128 struct dm_table *map = md->map;
2129 unsigned long flags;
2130
2131 if (!map)
2132 return NULL;
2133
2134 dm_table_event_callback(map, NULL, NULL);
2135 write_lock_irqsave(&md->map_lock, flags);
2136 md->map = NULL;
2137 write_unlock_irqrestore(&md->map_lock, flags);
2138
2139 return map;
2140}
2141
2142/*
2143 * Constructor for a new device.
2144 */
2145int dm_create(int minor, struct mapped_device **result)
2146{
2147 struct mapped_device *md;
2148
2149 md = alloc_dev(minor);
2150 if (!md)
2151 return -ENXIO;
2152
2153 dm_sysfs_init(md);
2154
2155 *result = md;
2156 return 0;
2157}
2158
2159/*
2160 * Functions to manage md->type.
2161 * All are required to hold md->type_lock.
2162 */
2163void dm_lock_md_type(struct mapped_device *md)
2164{
2165 mutex_lock(&md->type_lock);
2166}
2167
2168void dm_unlock_md_type(struct mapped_device *md)
2169{
2170 mutex_unlock(&md->type_lock);
2171}
2172
2173void dm_set_md_type(struct mapped_device *md, unsigned type)
2174{
2175 md->type = type;
2176}
2177
2178unsigned dm_get_md_type(struct mapped_device *md)
2179{
2180 return md->type;
2181}
2182
2183struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2184{
2185 return md->immutable_target_type;
2186}
2187
2188/*
2189 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2190 */
2191static int dm_init_request_based_queue(struct mapped_device *md)
2192{
2193 struct request_queue *q = NULL;
2194
2195 if (md->queue->elevator)
2196 return 1;
2197
2198 /* Fully initialize the queue */
2199 q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL);
2200 if (!q)
2201 return 0;
2202
2203 md->queue = q;
2204 dm_init_md_queue(md);
2205 blk_queue_softirq_done(md->queue, dm_softirq_done);
2206 blk_queue_prep_rq(md->queue, dm_prep_fn);
2207 blk_queue_lld_busy(md->queue, dm_lld_busy);
2208
2209 elv_register_queue(md->queue);
2210
2211 return 1;
2212}
2213
2214/*
2215 * Setup the DM device's queue based on md's type
2216 */
2217int dm_setup_md_queue(struct mapped_device *md)
2218{
2219 if ((dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) &&
2220 !dm_init_request_based_queue(md)) {
2221 DMWARN("Cannot initialize queue for request-based mapped device");
2222 return -EINVAL;
2223 }
2224
2225 return 0;
2226}
2227
2228static struct mapped_device *dm_find_md(dev_t dev)
2229{
2230 struct mapped_device *md;
2231 unsigned minor = MINOR(dev);
2232
2233 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2234 return NULL;
2235
2236 spin_lock(&_minor_lock);
2237
2238 md = idr_find(&_minor_idr, minor);
2239 if (md && (md == MINOR_ALLOCED ||
2240 (MINOR(disk_devt(dm_disk(md))) != minor) ||
2241 dm_deleting_md(md) ||
2242 test_bit(DMF_FREEING, &md->flags))) {
2243 md = NULL;
2244 goto out;
2245 }
2246
2247out:
2248 spin_unlock(&_minor_lock);
2249
2250 return md;
2251}
2252
2253struct mapped_device *dm_get_md(dev_t dev)
2254{
2255 struct mapped_device *md = dm_find_md(dev);
2256
2257 if (md)
2258 dm_get(md);
2259
2260 return md;
2261}
2262EXPORT_SYMBOL_GPL(dm_get_md);
2263
2264void *dm_get_mdptr(struct mapped_device *md)
2265{
2266 return md->interface_ptr;
2267}
2268
2269void dm_set_mdptr(struct mapped_device *md, void *ptr)
2270{
2271 md->interface_ptr = ptr;
2272}
2273
2274void dm_get(struct mapped_device *md)
2275{
2276 atomic_inc(&md->holders);
2277 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2278}
2279
2280const char *dm_device_name(struct mapped_device *md)
2281{
2282 return md->name;
2283}
2284EXPORT_SYMBOL_GPL(dm_device_name);
2285
2286static void __dm_destroy(struct mapped_device *md, bool wait)
2287{
2288 struct dm_table *map;
2289
2290 might_sleep();
2291
2292 spin_lock(&_minor_lock);
2293 map = dm_get_live_table(md);
2294 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2295 set_bit(DMF_FREEING, &md->flags);
2296 spin_unlock(&_minor_lock);
2297
2298 if (!dm_suspended_md(md)) {
2299 dm_table_presuspend_targets(map);
2300 dm_table_postsuspend_targets(map);
2301 }
2302
2303 /*
2304 * Rare, but there may be I/O requests still going to complete,
2305 * for example. Wait for all references to disappear.
2306 * No one should increment the reference count of the mapped_device,
2307 * after the mapped_device state becomes DMF_FREEING.
2308 */
2309 if (wait)
2310 while (atomic_read(&md->holders))
2311 msleep(1);
2312 else if (atomic_read(&md->holders))
2313 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2314 dm_device_name(md), atomic_read(&md->holders));
2315
2316 dm_sysfs_exit(md);
2317 dm_table_put(map);
2318 dm_table_destroy(__unbind(md));
2319 free_dev(md);
2320}
2321
2322void dm_destroy(struct mapped_device *md)
2323{
2324 __dm_destroy(md, true);
2325}
2326
2327void dm_destroy_immediate(struct mapped_device *md)
2328{
2329 __dm_destroy(md, false);
2330}
2331
2332void dm_put(struct mapped_device *md)
2333{
2334 atomic_dec(&md->holders);
2335}
2336EXPORT_SYMBOL_GPL(dm_put);
2337
2338static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
2339{
2340 int r = 0;
2341 DECLARE_WAITQUEUE(wait, current);
2342
2343 add_wait_queue(&md->wait, &wait);
2344
2345 while (1) {
2346 set_current_state(interruptible);
2347
2348 if (!md_in_flight(md))
2349 break;
2350
2351 if (interruptible == TASK_INTERRUPTIBLE &&
2352 signal_pending(current)) {
2353 r = -EINTR;
2354 break;
2355 }
2356
2357 io_schedule();
2358 }
2359 set_current_state(TASK_RUNNING);
2360
2361 remove_wait_queue(&md->wait, &wait);
2362
2363 return r;
2364}
2365
2366/*
2367 * Process the deferred bios
2368 */
2369static void dm_wq_work(struct work_struct *work)
2370{
2371 struct mapped_device *md = container_of(work, struct mapped_device,
2372 work);
2373 struct bio *c;
2374
2375 down_read(&md->io_lock);
2376
2377 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2378 spin_lock_irq(&md->deferred_lock);
2379 c = bio_list_pop(&md->deferred);
2380 spin_unlock_irq(&md->deferred_lock);
2381
2382 if (!c)
2383 break;
2384
2385 up_read(&md->io_lock);
2386
2387 if (dm_request_based(md))
2388 generic_make_request(c);
2389 else
2390 __split_and_process_bio(md, c);
2391
2392 down_read(&md->io_lock);
2393 }
2394
2395 up_read(&md->io_lock);
2396}
2397
2398static void dm_queue_flush(struct mapped_device *md)
2399{
2400 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2401 smp_mb__after_clear_bit();
2402 queue_work(md->wq, &md->work);
2403}
2404
2405/*
2406 * Swap in a new table, returning the old one for the caller to destroy.
2407 */
2408struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2409{
2410 struct dm_table *map = ERR_PTR(-EINVAL);
2411 struct queue_limits limits;
2412 int r;
2413
2414 mutex_lock(&md->suspend_lock);
2415
2416 /* device must be suspended */
2417 if (!dm_suspended_md(md))
2418 goto out;
2419
2420 r = dm_calculate_queue_limits(table, &limits);
2421 if (r) {
2422 map = ERR_PTR(r);
2423 goto out;
2424 }
2425
2426 map = __bind(md, table, &limits);
2427
2428out:
2429 mutex_unlock(&md->suspend_lock);
2430 return map;
2431}
2432
2433/*
2434 * Functions to lock and unlock any filesystem running on the
2435 * device.
2436 */
2437static int lock_fs(struct mapped_device *md)
2438{
2439 int r;
2440
2441 WARN_ON(md->frozen_sb);
2442
2443 md->frozen_sb = freeze_bdev(md->bdev);
2444 if (IS_ERR(md->frozen_sb)) {
2445 r = PTR_ERR(md->frozen_sb);
2446 md->frozen_sb = NULL;
2447 return r;
2448 }
2449
2450 set_bit(DMF_FROZEN, &md->flags);
2451
2452 return 0;
2453}
2454
2455static void unlock_fs(struct mapped_device *md)
2456{
2457 if (!test_bit(DMF_FROZEN, &md->flags))
2458 return;
2459
2460 thaw_bdev(md->bdev, md->frozen_sb);
2461 md->frozen_sb = NULL;
2462 clear_bit(DMF_FROZEN, &md->flags);
2463}
2464
2465/*
2466 * We need to be able to change a mapping table under a mounted
2467 * filesystem. For example we might want to move some data in
2468 * the background. Before the table can be swapped with
2469 * dm_bind_table, dm_suspend must be called to flush any in
2470 * flight bios and ensure that any further io gets deferred.
2471 */
2472/*
2473 * Suspend mechanism in request-based dm.
2474 *
2475 * 1. Flush all I/Os by lock_fs() if needed.
2476 * 2. Stop dispatching any I/O by stopping the request_queue.
2477 * 3. Wait for all in-flight I/Os to be completed or requeued.
2478 *
2479 * To abort suspend, start the request_queue.
2480 */
2481int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2482{
2483 struct dm_table *map = NULL;
2484 int r = 0;
2485 int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
2486 int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
2487
2488 mutex_lock(&md->suspend_lock);
2489
2490 if (dm_suspended_md(md)) {
2491 r = -EINVAL;
2492 goto out_unlock;
2493 }
2494
2495 map = dm_get_live_table(md);
2496
2497 /*
2498 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2499 * This flag is cleared before dm_suspend returns.
2500 */
2501 if (noflush)
2502 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2503
2504 /* This does not get reverted if there's an error later. */
2505 dm_table_presuspend_targets(map);
2506
2507 /*
2508 * Flush I/O to the device.
2509 * Any I/O submitted after lock_fs() may not be flushed.
2510 * noflush takes precedence over do_lockfs.
2511 * (lock_fs() flushes I/Os and waits for them to complete.)
2512 */
2513 if (!noflush && do_lockfs) {
2514 r = lock_fs(md);
2515 if (r)
2516 goto out;
2517 }
2518
2519 /*
2520 * Here we must make sure that no processes are submitting requests
2521 * to target drivers i.e. no one may be executing
2522 * __split_and_process_bio. This is called from dm_request and
2523 * dm_wq_work.
2524 *
2525 * To get all processes out of __split_and_process_bio in dm_request,
2526 * we take the write lock. To prevent any process from reentering
2527 * __split_and_process_bio from dm_request and quiesce the thread
2528 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2529 * flush_workqueue(md->wq).
2530 */
2531 down_write(&md->io_lock);
2532 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2533 up_write(&md->io_lock);
2534
2535 /*
2536 * Stop md->queue before flushing md->wq in case request-based
2537 * dm defers requests to md->wq from md->queue.
2538 */
2539 if (dm_request_based(md))
2540 stop_queue(md->queue);
2541
2542 flush_workqueue(md->wq);
2543
2544 /*
2545 * At this point no more requests are entering target request routines.
2546 * We call dm_wait_for_completion to wait for all existing requests
2547 * to finish.
2548 */
2549 r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE);
2550
2551 down_write(&md->io_lock);
2552 if (noflush)
2553 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2554 up_write(&md->io_lock);
2555
2556 /* were we interrupted ? */
2557 if (r < 0) {
2558 dm_queue_flush(md);
2559
2560 if (dm_request_based(md))
2561 start_queue(md->queue);
2562
2563 unlock_fs(md);
2564 goto out; /* pushback list is already flushed, so skip flush */
2565 }
2566
2567 /*
2568 * If dm_wait_for_completion returned 0, the device is completely
2569 * quiescent now. There is no request-processing activity. All new
2570 * requests are being added to md->deferred list.
2571 */
2572
2573 set_bit(DMF_SUSPENDED, &md->flags);
2574
2575 dm_table_postsuspend_targets(map);
2576
2577out:
2578 dm_table_put(map);
2579
2580out_unlock:
2581 mutex_unlock(&md->suspend_lock);
2582 return r;
2583}
2584
2585int dm_resume(struct mapped_device *md)
2586{
2587 int r = -EINVAL;
2588 struct dm_table *map = NULL;
2589
2590 mutex_lock(&md->suspend_lock);
2591 if (!dm_suspended_md(md))
2592 goto out;
2593
2594 map = dm_get_live_table(md);
2595 if (!map || !dm_table_get_size(map))
2596 goto out;
2597
2598 r = dm_table_resume_targets(map);
2599 if (r)
2600 goto out;
2601
2602 dm_queue_flush(md);
2603
2604 /*
2605 * Flushing deferred I/Os must be done after targets are resumed
2606 * so that mapping of targets can work correctly.
2607 * Request-based dm is queueing the deferred I/Os in its request_queue.
2608 */
2609 if (dm_request_based(md))
2610 start_queue(md->queue);
2611
2612 unlock_fs(md);
2613
2614 clear_bit(DMF_SUSPENDED, &md->flags);
2615
2616 r = 0;
2617out:
2618 dm_table_put(map);
2619 mutex_unlock(&md->suspend_lock);
2620
2621 return r;
2622}
2623
2624/*-----------------------------------------------------------------
2625 * Event notification.
2626 *---------------------------------------------------------------*/
2627int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2628 unsigned cookie)
2629{
2630 char udev_cookie[DM_COOKIE_LENGTH];
2631 char *envp[] = { udev_cookie, NULL };
2632
2633 if (!cookie)
2634 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2635 else {
2636 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2637 DM_COOKIE_ENV_VAR_NAME, cookie);
2638 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2639 action, envp);
2640 }
2641}
2642
2643uint32_t dm_next_uevent_seq(struct mapped_device *md)
2644{
2645 return atomic_add_return(1, &md->uevent_seq);
2646}
2647
2648uint32_t dm_get_event_nr(struct mapped_device *md)
2649{
2650 return atomic_read(&md->event_nr);
2651}
2652
2653int dm_wait_event(struct mapped_device *md, int event_nr)
2654{
2655 return wait_event_interruptible(md->eventq,
2656 (event_nr != atomic_read(&md->event_nr)));
2657}
2658
2659void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2660{
2661 unsigned long flags;
2662
2663 spin_lock_irqsave(&md->uevent_lock, flags);
2664 list_add(elist, &md->uevent_list);
2665 spin_unlock_irqrestore(&md->uevent_lock, flags);
2666}
2667
2668/*
2669 * The gendisk is only valid as long as you have a reference
2670 * count on 'md'.
2671 */
2672struct gendisk *dm_disk(struct mapped_device *md)
2673{
2674 return md->disk;
2675}
2676
2677struct kobject *dm_kobject(struct mapped_device *md)
2678{
2679 return &md->kobj;
2680}
2681
2682/*
2683 * struct mapped_device should not be exported outside of dm.c
2684 * so use this check to verify that kobj is part of md structure
2685 */
2686struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2687{
2688 struct mapped_device *md;
2689
2690 md = container_of(kobj, struct mapped_device, kobj);
2691 if (&md->kobj != kobj)
2692 return NULL;
2693
2694 if (test_bit(DMF_FREEING, &md->flags) ||
2695 dm_deleting_md(md))
2696 return NULL;
2697
2698 dm_get(md);
2699 return md;
2700}
2701
2702int dm_suspended_md(struct mapped_device *md)
2703{
2704 return test_bit(DMF_SUSPENDED, &md->flags);
2705}
2706
2707int dm_suspended(struct dm_target *ti)
2708{
2709 return dm_suspended_md(dm_table_get_md(ti->table));
2710}
2711EXPORT_SYMBOL_GPL(dm_suspended);
2712
2713int dm_noflush_suspending(struct dm_target *ti)
2714{
2715 return __noflush_suspending(dm_table_get_md(ti->table));
2716}
2717EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2718
2719struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity)
2720{
2721 struct dm_md_mempools *pools = kmalloc(sizeof(*pools), GFP_KERNEL);
2722 unsigned int pool_size = (type == DM_TYPE_BIO_BASED) ? 16 : MIN_IOS;
2723
2724 if (!pools)
2725 return NULL;
2726
2727 pools->io_pool = (type == DM_TYPE_BIO_BASED) ?
2728 mempool_create_slab_pool(MIN_IOS, _io_cache) :
2729 mempool_create_slab_pool(MIN_IOS, _rq_bio_info_cache);
2730 if (!pools->io_pool)
2731 goto free_pools_and_out;
2732
2733 pools->tio_pool = (type == DM_TYPE_BIO_BASED) ?
2734 mempool_create_slab_pool(MIN_IOS, _tio_cache) :
2735 mempool_create_slab_pool(MIN_IOS, _rq_tio_cache);
2736 if (!pools->tio_pool)
2737 goto free_io_pool_and_out;
2738
2739 pools->bs = bioset_create(pool_size, 0);
2740 if (!pools->bs)
2741 goto free_tio_pool_and_out;
2742
2743 if (integrity && bioset_integrity_create(pools->bs, pool_size))
2744 goto free_bioset_and_out;
2745
2746 return pools;
2747
2748free_bioset_and_out:
2749 bioset_free(pools->bs);
2750
2751free_tio_pool_and_out:
2752 mempool_destroy(pools->tio_pool);
2753
2754free_io_pool_and_out:
2755 mempool_destroy(pools->io_pool);
2756
2757free_pools_and_out:
2758 kfree(pools);
2759
2760 return NULL;
2761}
2762
2763void dm_free_md_mempools(struct dm_md_mempools *pools)
2764{
2765 if (!pools)
2766 return;
2767
2768 if (pools->io_pool)
2769 mempool_destroy(pools->io_pool);
2770
2771 if (pools->tio_pool)
2772 mempool_destroy(pools->tio_pool);
2773
2774 if (pools->bs)
2775 bioset_free(pools->bs);
2776
2777 kfree(pools);
2778}
2779
2780static const struct block_device_operations dm_blk_dops = {
2781 .open = dm_blk_open,
2782 .release = dm_blk_close,
2783 .ioctl = dm_blk_ioctl,
2784 .getgeo = dm_blk_getgeo,
2785 .owner = THIS_MODULE
2786};
2787
2788EXPORT_SYMBOL(dm_get_mapinfo);
2789
2790/*
2791 * module hooks
2792 */
2793module_init(dm_init);
2794module_exit(dm_exit);
2795
2796module_param(major, uint, 0);
2797MODULE_PARM_DESC(major, "The major number of the device mapper");
2798MODULE_DESCRIPTION(DM_NAME " driver");
2799MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2800MODULE_LICENSE("GPL");